From optical imaging to dark matter halo masses with simulation-based inference models
Abstract
Galaxies and dark matter halos are linked by concurrent assembly. To first order, this link alone establishes a stellar-to-halo mass relation (SHMR) which is secondarily affected by the baryon cycle of inflows, conversion of gas into stars, and outflows. Nonetheless, the baryon cycle (coupled with halo formation time) supports a broad range of stellar masses and galaxy morphologies at fixed halo mass -- yielding an intrinsic scatter in the SHMR of approximately 0.2 dex, on average. In this talk, I will show that this scatter can be significantly reduced using luminosity, colour, and morphological information extracted from mock optical/NIR images to predict galaxy halo masses. Using a simulation-based inference model, HALOFLOW, trained on a large sample of mock Subaru HSC images of galaxies formed in cosmological simulations, we find that colour and morphological information alone is sufficient to predict 40% of the scatter in the SMHR. We examine whether a model trained on one simulated universe (and set of recipes for the baryon cycle) performs similarly on another simulated universe. Finally, we test our simulation-trained model to galaxies in the real Universe for which halo masses have been constrained using lensing and group dynamics.
- Publication:
-
IAU General Assembly
- Pub Date:
- August 2024
- Bibcode:
- 2024IAUGA..32P2133B