HIFlow: Generating Diverse HI Maps Conditioned on Cosmology using Normalizing Flow

Authors:
  • SHSultan Hassan
    Sultan Hassan
  • FVFrancisco Villaescusa-Navarro
    Francisco Villaescusa-Navarro
  • BWBenjamin Wandelt
    Benjamin Wandelt
  • DSDavid N. Spergel
    David N. Spergel
  • DADaniel Anglés-Alcázar
    Daniel Anglés-Alcázar
  • SGShy Genel
    Shy Genel
  • MCMiles Cranmer
    Miles Cranmer
  • GBGreg L. Bryan
    Greg L. Bryan
  • RDRomeel Davé
    Romeel Davé
  • RSRachel S. Somerville
    Rachel S. Somerville
  • MEMichael Eickenberg
    Michael Eickenberg
  • DNDesika Narayanan
    Desika Narayanan

Abstract
A wealth of cosmological and astrophysical information is expected from many ongoing and upcoming large-scale surveys. It is crucial to prepare for these surveys now and develop tools that can efficiently extract the maximum amount of information. We present HIFlow: a fast emulator that is able to generate neutral hydrogen (HI) maps conditioned only on cosmology ($\Omega_{m}$ and $\sigma_{8}$), after training on the state-of-the-art simulations from the Cosmology and Astrophysics with MachinE Learning Simulations (CAMELS) project. HIFlow is designed using a class of normalizing flow models, the Masked Autoregressive Flow (MAF), which we demonstrate are capable of generating realistic maps without explicitly using the 2D structure or accounting for any symmetries. HIFlow is able to generate new diverse HI maps in the column density range $N_{\rm HI} \sim 10^{14} - 10^{21} {\rm cm^{-2}}$ at $z\sim 6$, and naturally mimic the cosmic variance effects. Remarkably, HIFlow is able to reproduce the CAMELS average and standard deviation HI power spectrum (Pk) within a factor of $\lesssim$ 2, scoring a very high $R^{2} > 90\%$. HIFlow will enable the testing of Pk pipelines for HI surveys, and assist in computing other statistical properties beyond Pk that require generating new diverse samples of high dimensional datasets, such as the covariance matrix. This new tool represents a first step towards enabling rapid parameter inference at the field level, maximizing the scientific return of future HI surveys, and opening a new avenue to minimize the loss of information due to data compression.