Speaker
Description
Baryonic feedback is one of the leading systematic uncertainties for upcoming Stage IV weak lensing surveys such as LSST, Euclid, and Roman. Standard baryon correction models offer a fast fix, but they are spherically symmetric by construction — missing the correlated, non-spherical structure of real baryonic fields and the galaxy shapes that drive intrinsic alignment signals.
I will present Baryonification and Intrinsic alignment with Neural Diffusion, BIND, a generative model that paints full field-level dark matter, gas, and stellar mass distributions onto dark matter–only halo patches, conditioned on 35 astrophysical and cosmological parameters from the CAMELS 50Mpc/h simulation suite. Because BIND learns directly from halos in hydrodynamical simulations, it generates non-spherical halo fields by construction and naturally captures morphological diversity across the galaxy population. I will show that the generated fields respond correctly to variations in feedback parameters at the field level, enabling rapid exploration of the parameter space and straightforward marginalization over baryonic physics within simulation-based inference frameworks. With bind, I will present the recovery of the matter power spectrum within statistical expectations, as well as accurate halo shapes and stellar distributions that match those of the true hydrodynamical simulations, thereby opening the way for applications to intrinsic alignment studies.
BIND opens a path toward fast, accurate baryonified field realizations with realistic galaxy morphologies, and I will discuss how it can serve as a flexible tool for WL inference in the precision cosmology era.
| Other topic / keywords: | Weak lensing, Simulations, Baryonification |
|---|