Unveiling the dark Universe with HI and EMBER-2 2025 Unveiling the dark Universe with HI and EMBER-2 –

Mauro Bernardini, Robert Feldmann, Daniel Anglés-Alcázar, Philipp Denzel, Jindra Gensior · 2025

Read Unveiling the dark Universe with HI and EMBER-2 2025 Unveiling the dark Universe with HI and EMBER-2 – by Mauro Bernardini, Robert Feldmann, Daniel Anglés-Alcázar, Philipp Denzel, Jindra Gensior (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05339.txt.

Introduction

Maps of hydrogen gas across the young universe can betray where dark matter hides, and a 2025 paper by Mauro Bernardini and co-authors trains a model to draw that hidden map. Next-generation radio telescopes will flood researchers with hydrogen data, but the gas only loosely traces the dark scaffolding, and galaxy birth scrambles the signal.

Their model, EMBER-2, learns the link from detailed galaxy simulations and then predicts dark matter layout from hydrogen alone, across most of cosmic time. It recovers mass shares, density profiles, and cross-links with the gas at good accuracy, down to fine scales where older recipe-based methods blur. The authors present the method as a way to use the full information in the coming data.

Catalog record

CREATORMauro Bernardini, Robert Feldmann, Daniel Anglés-Alcázar, Philipp Denzel, Jindra Gensior
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-05339

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Unveiling the dark Universe with HI and EMBER-2 2025 Unveiling the dark Universe with HI and EMBER-2 – Mauro Bernardini, Robert Feldmann, Daniel Anglés-Alcázar, Philipp Denzel, Jindra Gensior Abstract Next-generation radio telescopes will provide unprecedented data volumes of the neutral hydrogen (HI) distribution across cosmic time. Combining weak lensing surveys with spatial and kinematic observations of HI could help constrain key properties of dark matter, such as its mass, clustering behavior, and spatial distribution. However, inferring dark matter properties from HI observations is challenging because of processes related to galaxy formation, such as stellar feedback. Methods that use empirical relations, often calibrated via numerical simulations, do not use full field-level information to model the complex relation between dark matter and HI. We address this shortcoming with a data-driven approach, leveraging the recently introduced EMBER-2 model to learn the HI-dark matter mapping at the field level for a wide redshift range, z = 6 − 0 z=6-0 . After training on cosmological galaxy formation simulations run with FIRE-2, EMBER-2 accurately recovers key statistics, including dark matter mass fractions and surface density profiles. The HI-dark matter density cross-correlation is reconstructed at an accuracy of 10% down to scales of k = 100 h / k=100\,h/ cMpc, a significant improvement over traditional approaches. The presented method may become a key ingredient in future inference pipelines as it can be readily integrated into downstream analysis tasks of radio surveys. 1 Introduction Figure 1: Overview figure showing an example region of the simulated HI and DM fields in the FB30 simulation and the corresponding reconstructed field from EMBER-2 at z=1 . The figure shows both normalized channels, surface density \Sigma and line-of-sight velocity v , in a composite manner, where increased brightness indicates higher \Sigma , color-coded by the corresponding velocity of each pixel (from red to blue). For better visibility we show the normalized field values, which have \mathcal{O}(1) . This visual comparison highlights that the HI density and velocity fields contain significant amounts of information across different phase-space regimes, which can be used to accurately reconstruct the underlying DM distribution and kinematics from large down to small scale structures. Shown as insets are two example regions showcasing this behavior in both a high (teal) and low (orange) density regime. The next generation of radio and optical surveys will provide an unprecedented volume of data to analyze. Pathfinder surveys such as MeerKAT’s MIGHTEE ( Maddox et al. 2021 ; Jarvis et al. 2025 ) and ASKAP’s RACS ( McConnell et al. 2020 ; Duchesne et al. 2024 ; Duchesne et al. 2025 ) , and in the near future the Square Kilometre Array (SKA) telescopes ( Weltman et al. 2020 ) , are revolutionizing our ability to probe the distribution and kinematics of HI across a significant fraction of cosmic time. These surveys will provide detailed observations of HI in the Interstellar (ISM) and Circumgalactic Medium (CGM) and map the large-scale structures of the cosmic web with Intensity Mapping (IM) ( Paul et al. 2021 ; Cunnington et al. 2023 ; Mazumder et al. 2025 ) . Complementary optical and near-infrared facilities such as the Extremely Large Telescope (ELT) and upcoming wide-field surveys from Euclid ( Euclid Collaboration et al. 2019 ; Euclid Collaboration et al. 2024 ) and the Vera C. Rubin Observatory’s LSST ( Sheldon et al. 2023 ) will provide large galaxy catalogs extending to high redshifts and play a crucial role for weak lensing (WL) studies. In particular, LSST and Euclid will measure cosmic shear across large regions of the sky, allowing precise constraints on the total matter distribution. By combining WL with HI IM, these efforts will offer a multi-faceted approach to constrain key properties of dark matter, such as its mass, and the processes governing cosmic structure formation ( Abdalla et al. 2015 ; Pourtsidou et al. 2016 ; Sangka & Bacon 2024 , e.g.) . However, directly inferring the underlying dark matter distribution from observational data remains challenging. While WL provides a direct probe of the total matter field, it requires large statistical samples ( Takada 2014 ; Pérez de los Heros 2020 ; Mahony et al. 2022 ; Remy et al. 2023 ) . Furthermore, measuring the matter power spectrum requires accurate modeling of all systematic effects that contribute to the cosmic shear signal. Of particular importance is the impact of baryonic feedback stemming from small-scale processes that can severely affect the cosmic shear power spectrum on small angular scales ( Mohammed et al. 2014 ; Schneider & Teyssier 2015 ; Schneider et al. 2019 ; Huang et al. 2019 ; Schneider et al. 2020 ; Lu & Haiman 2021 ; Broxterman et al. 2024 , e.g.) . HI based methods offer a complementary approach by modeling HI as a biased tracer of the underlying dark matter field. These approaches establish the mapping between HI and dark matter via empirical relations derived from numerical simulations that are themselves tuned to match key observational constraints, such as the stellar mass function or the power spectrum. For example, Halo Occupation Distribution (HOD) models and abundance matching (AM) techniques construct a mapping between the mass of dark matter haloes and the contained HI mass ( Berlind & Weinberg 2002 ; Villaescusa-Navarro et al. 2014 ; Castorina & Villaescusa-Navarro 2017 ; Qin et al. 2022 , e.g.) . Other methods model the relation as a bias expansion (BE) on the field level, where the HI and dark matter density fields are fitted with analytical descriptions with a few bias parameters ( Obuljen et al. 2023 ; Baradaran et al. 2024 ; de Belsunce et al. 2025 ) . However, these approaches do not fully capture the complex interplay between HI and dark matter, especially on small scales, since the relation is established by either considering just a few quantities, such as the halo mass for HOD, or only a handful of fitting parameters in the BE, while discarding the remaining information stored in the field. Furthermore, the correlation between HI and the underlying dark matter distribution is altered by feedback processes, such as stellar winds, ram pressure stripping and feedback from Active Galactic Nuclei (AGN), that redistribute baryonic matter ( Anglés-Alcázar et al. 2017 ; Hafen et al. 2019 ; Gebhardt et al. 2024 ) . These processes complicate dark matter inference from just a few simple HI properties, since environmental effects play a significant role in driving the scatter in the HI to dark matter halo mass relation. This behavior has been seen in observations ( Oosterloo et al. 2010 ; Janowiecki et al. 2017 ; Cortese et al. 2021 ; Li et al. 2022 ; Saintonge & Catinella 2022 ; Saraf et al. 2024 ) and has been successfully reproduced in simulations ( Rafieferantsoa et al. 2015 ; Stevens et al. 2019 ; Bernardini et al. 2022 , e.g.) . To address these challenges, we propose a data-driven approach to learn the direct mapping from HI to the underlying dark matter distribution by taking into consideration the entire field level information. By leveraging neural networks trained on cosmological simulations, we construct a model that captures the complex, non-linear relationship between HI and dark matter while accounting for baryonic

Read the complete text