LIMFAST. IV. Learning High-Redshift Galaxy Formation from Multiline Intensity Mapping with Implicit Likelihood Inference

and Steven R. Furlanetto · 2025

Read LIMFAST. IV. Learning High-Redshift Galaxy Formation from Multiline Intensity Mapping with Implicit Likelihood Inference by and Steven R. Furlanetto (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-2509-07060.txt.

Introduction

Guochao Sun and 7 co-authors present LIMFAST work on learning high-redshift galaxy formation from intensity mapping in a 2025 paper. Line intensity mapping is a technique that maps the cumulative glow of many galaxies in a certain emission line rather than resolving single galaxies. It calls for novel tools for efficient forward modeling and inference.

Implicit likelihood inference from semi-numerical model runs gives a powerful setup for investigating a large model setting space in a data-driven manner. Using model runs of two emission line signals created by the LIMFAST code, the authors develop an inference system in a case study of early galaxy formation. Neural density estimation with normalizing flows learns the mapping between modeled power spectra and parameters characterizing star formation efficiency and related relations. Their results show what can be constrained from auto-power spectra and cross-correlation with galaxy surveys.

Catalog record

CREATORand Steven R. Furlanetto
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE66 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2509-07060

Browse the full library · About the Tower of Babel

Full text

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

LIMFAST. IV. Learning High-Redshift Galaxy Formation from Multiline Intensity Mapping with Implicit Likelihood Inference and Steven R. Furlanetto Abstract By opening up new avenues to statistically constrain astrophysics and cosmology with large-scale structure observations, the line intensity mapping (LIM) technique calls for novel tools for efficient forward modeling and inference. Implicit likelihood inference (ILI) from semi-numerical simulations provides a powerful setup for investigating a large model parameter space in a data-driven manner, therefore gaining significant recent attention. Using simulations of high-redshift 158 μ \mu m [ C ​ ii \mathrm{C\,\textsc{ii}} ] and 88 μ \mu m [ O ​ iii \mathrm{O\,\textsc{iii}} ] LIM signals created by the LIMFAST code, we develop an ILI framework in a case study of learning the physics of early galaxy formation from the auto-power spectra of these lines or their cross-correlation with galaxy surveys. We leverage neural density estimation with normalizing flows to learn the mapping between the simulated power spectra and parameters that characterize the physics governing the star formation efficiency and the Σ ˙ ⋆ \dot{\Sigma}_{\star} – Σ g \Sigma_{\mathrm{g}} relation of high-redshift galaxies. Our results show that their partially degenerate effects can be unambiguously constrained when combining [ C ​ ii \mathrm{C\,\textsc{ii}} ] with [ O ​ iii \mathrm{O\,\textsc{iii}} ] measurements to be made by new-generation mm/sub-mm LIM experiments. 1 Introduction Deep, detailed observations of an increasingly large sample of high-redshift ( z\gtrsim 6 ) galaxies made by new-generation telescopes like the James Webb Space Telescope (JWST) have transformed the understanding of physical processes that govern galaxy formation in the early universe [ 1 , 2 ] . A major constraint for such flux-limited observations, however, is that only galaxies brighter than a certain survey depth can be detected within a given area. This observational trade-off between survey depth and area restricts studies of individual galaxies to a limited sample of relatively bright sources, potentially leaving a significant population of fainter sources undetected. While their exact contribution to cosmic reionization remains debated [ 3 , 4 , 5 , 6 ] , the population of faint galaxies below JWST’s typical detection threshold is predicted to be responsible for more than half of the total cosmic SFRD at z>10 [ 7 ] . Meanwhile, sensitive to the fundamentals of galaxy formation and evolution, especially the roles of stellar feedback and dark matter [ 8 , 9 , 10 ] , faint/low-mass galaxies provide crucial insights into early structure formation and are therefore of special importance observationally. Line intensity mapping (LIM) promises to complement galaxy surveys by measuring the spatial fluctuations in the aggregate emission of certain lines (e.g., 21 cm, Ly \alpha , and [ \mathrm{C\,\textsc{ii}} ]) sourced by the entire galaxy population, including the very faint ones for which individual detections are unlikely [ 11 , 12 ] . Numerous studies have employed empirical/semi-analytic models or simulations to demonstrate that rich physical information about high- z galaxy formation may be revealed by LIM surveys of multiple tracers [ 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , see e.g.,] . Multi-tracer analysis based on cross-correlations also benefits from its lower susceptibility to observational biases due to foreground contamination [ 25 , 26 , 27 ] . In the context of probing the galaxy population responsible for cosmic reionization, LIM observations of the 158 \mu m [ \mathrm{C\,\textsc{ii}} ] and 88 \mu m [ \mathrm{O\,\textsc{iii}} ] lines (redshifted into mm/sub-mm wavelengths) are promising. Not only is this because they are the two brightest far-infrared emission lines from high- z galaxies that have been routinely detected at increasing redshift to provide precise spectroscopic redshifts [ 28 , 29 , 30 ] , but they also encode rich physical information about the stellar population and interstellar medium (ISM) in the infant universe [ 31 , 32 ] . Padmanabhan et al. (2022) [ 33 ] , in particular, investigate the potential for synergizing [ \mathrm{C\,\textsc{ii}} ] and [ \mathrm{O\,\textsc{iii}} ] LIM at the reionization era across a variety of experimental designs and find that improvements to current-generation mm/sub-mm LIM experiments could enable such synergy in the foreseeable future. With the advent of new LIM experiments that will enable joint analysis of multiple lines, an urgent need is created for statistical inference tools to constrain specific physical processes governing galaxy formation with LIM data. Bayesian sampling of the multidimensional posterior distribution—assuming some explicit likelihood constructed from analytically modeled summary statistics—is the standard approach in traditional cosmological inference. It has been extensively applied to parameter inference from \mathrm{H\,\textsc{i}} 21 cm LIM simulations [ 34 , 35 ] , though the high dimensional and weakly constrained parameter space often makes on-the-fly sampling with methods such as Markov Chain Monte Carlo (MCMC) computationally expensive [ 36 , 37 , 38 ] . The explicit likelihood can not only lead to high computational cost as individual evaluations add up, but also introduce unwanted systematic bias in the inferred posterior due to strong assumptions made about e.g., the noise distribution. Implicit likelihood inference (ILI), also known as simulation-based inference, provides a powerful and practical solution to these limitations through approaches such as Neural Posterior Estimation (NPE) and has therefore gained increasing popularity [ 39 ] , with applications in astrophysics and cosmology expanding significantly in recent years [ 40 , 41 , 42 , 43 , 44 , 45 , 46 , e.g.,] . With techniques such as normalizing flows trained on simulated data, ILI directly constructs the posterior distribution using neural networks without assuming an explicit likelihood function. This enables efficient analysis of complex, high-dimensional parameter spaces, with scalability and flexibility stemming from the expressive power and computational efficiency of neural network-based architectures. Several recent studies have demonstrated the potential of ILI as an efficient and reliable method for inferring reionization parameters from the 21 cm LIM signal [ 47 , 48 , 49 , 50 , e.g.,] . As LIM of other emission lines beyond \mathrm{H\,\textsc{i}} 21 cm, especially those correlated with the galaxy star formation rate (SFR) such as [ \mathrm{C\,\textsc{ii}} ] and [ \mathrm{O\,\textsc{iii}} ], also shows significant promise for probing high- z galaxy formation, it is likewise interesting to consider ILI for efficient Bayesian parameter inference from those lines, including their synergies with other large-scale structure tracers like the distribution of galaxies. In this paper, we study the constraints on high- z galaxy formation physics from [ \mathrm{C\,\textsc{ii}} ] and [ \mathrm{O\,\textsc{iii}} ] LIM measurements using an ILI framework. As a proof-of-concept case study, we focus on both their respective auto-correlations and their cross-correlation with galaxy redshift surveys by the Roman Space Telescope, which can place stringent constraints on physical processes that determine the star formation efficiency (SFE) and gas content of high- z galaxies. For the first time to our best knowledge, we establish an analysis framework to explicitly demonstrate the power of multi-tracer LIM for constraining a broad class of simple physical models describing feedback-regulated star formation in high- z galaxies. We base our analysis on the training and testing data created by the semi-numerical LIMFAST simulations [ 20 , 21 , 51 ] , covering a reasonably diverse

Read the complete text