Field-level Reconstruction from Foreground-Contaminated 21-cm Maps
Shu-Fan Chen, Kai-Feng Chen, Cora Dvorkin · 2025
Read Field-level Reconstruction from Foreground-Contaminated 21-cm Maps by Shu-Fan Chen, Kai-Feng Chen, Cora Dvorkin (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-2508-13265.txt.
Introduction
Chen, Chen, and Dvorkin present a study from twenty twenty-five on rebuilding cosmic maps from radio data clouded by bright foregrounds. Telescopes mapping hydrogen across the young universe face a blocked wedge of modes. Foreground glare and instrument traits hide rich cosmic signal there. The team offer two routes to rebuild the full three-dimensional field. One route fits a physical forward model with a gradient sampler. It infers starting conditions and bias values jointly. The other route trains a generative diffusion model on simulations.
Both routes recover modes lost inside the wedge in their tests. The rebuilt fields should sharpen cosmic constraints from such surveys. They also enable cross-checks between experiments with little mode overlap. The authors span a wide range of sky physics in training.
Catalog record
| CREATOR | Shu-Fan Chen, Kai-Feng Chen, Cora Dvorkin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-13265
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Field-level Reconstruction from Foreground-Contaminated 21-cm Maps Shu-Fan Chen, Kai-Feng Chen, Cora Dvorkin Abstract Current and upcoming 21-cm experiments will soon be able to map 21-cm spatial fluctuations in three dimensions for a wide range of redshifts. However, bright foreground contamination and the nature of radio interferometry create significant challenges, making it difficult to access rich cosmological information from the Fourier modes that lie within the “foreground wedge”. In this work, we introduce two approaches aiming to reconstruct the full 21-cm density field, including the missing modes in the wedge: (a) a field-level inference under an effective field theory (EFT) framework; (b) a diffusion-based deep generative model trained on simulations. Under the EFT framework, we implement a fully differentiable forward model that maps the initial conditions of matter fluctuations to the observed, foreground-filtered 21-cm maps. This enables a gradient-based sampler to simultaneously sample the initial conditions and bias parameters, allowing a physically motivated mode reconstruction. Alternatively, we apply a variational diffusion model to perform 21-cm density reconstruction at the map level. Our model is trained on semi-numerical simulations over a wide range of astrophysical parameters. Our results from both approaches should provide improved cosmological constraints from the field level and also enable cross-correlation between experiments that have little or no overlapping modes. 1 Introduction The 21-cm line from neutral hydrogen is a promising probe for tracing the large-scale structure (LSS) across a large volume of the unexplored universe [ 1 , 2 ] . This isolated and highly forbidden transition in the radio band provides a unique opportunity to map the universe in three dimensions, as the observed frequency of the 21-cm line can be directly translated into the redshift (distance) of its source. Many radio interferometers have been built or proposed to detect the spatial fluctuation of the 21-cm signal. These include experiments at lower redshift (higher frequency) targeting the baryon acoustic oscillations (BAO) in the post-reionization universe [ 3 , 4 , 5 , 6 , 7 ] and at higher redshift (lower frequency) aiming for the Epoch of Reionization or the Cosmic Dawn [ 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 ] . Successful observations of such a signal will provide rich astrophysical and cosmological information, allowing us to better understand the dark sector of our universe and unveil the environment for star and galaxy formation [ 19 , 20 , 21 , 22 ] . The most significant barrier toward a successful measurement of the 21-cm signal lies in the substantial foreground contamination, as the radio emission from both galactic and extragalactic sources can be many orders of magnitude brighter than the cosmological signal [ 23 , 24 , 25 , 26 ] . Although astrophysical foreground emission is expected to be spectrally smooth, the chromatic instrument response of a radio interferometer usually causes the foreground to contaminate a large area in Fourier space (known as the foreground wedge [ 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 ] ). While many methods have been developed to explicitly model and subtract the foregrounds [ 27 , 35 , 36 , 37 , 38 , 39 , 40 , 41 ] , accessing large-scale 21-cm fluctuations remains challenging. Moreover, foreground subtraction becomes even more daunting in the presence of uncertainties in instrument response and systematic effects [ 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 ] . The difficulties in accessing these large-scale modes not only reduce the cosmological sensitivity but also significantly limit the potential of cross-correlating the 21-cm signal with other tracers. Rather than modeling and directly removing the foregrounds, some approaches adopt a more aggressive strategy by setting all modes within the foreground wedge to zero and reconstructing these modes using only the cleanly observed ones. This reconstruction leverages correlations between long- and short-wavelength modes of tracers. Quadratic or fossil estimators have been developed to exploit this correlation, using a fiducial bispectrum derived from perturbation theory [ 60 ] . However, as demonstrated in Ref. [ 61 ] , accurate mode reconstruction requires a precise bispectrum model for the quadratic estimator kernel to avoid introducing biases. To move beyond a tree-level bispectrum model, Ref. [ 62 ] investigates a field-level approach with a Lagrangian bias model extending to quadratic order for the low-redshift ( z=2-6 ) post-reionization 21-cm field. Using a quasi-Newtonian optimizer, they directly obtain the maximum a posteriori (MAP) estimate based on given bias parameters. Field-level inference has been successfully applied to a wide range of tracers [ 63 , 64 , 65 , 66 , 67 , 68 , 69 , 70 , 71 , 72 , 73 , 74 , 75 , 76 , 77 , 78 , 79 , 80 , 81 , 82 , 83 , 84 , 85 , 86 , 87 , 88 , 89 , 90 , 91 ] , but it has yet to be applied to tracers from the Epoch of Reionization. This opportunity emerged from recent theoretical developments [ 92 , 93 ] , which demonstrated that the 21-cm field during the Epoch of Reionization—observable through the Hydrogen Epoch of Reionization Array (HERA) and future experiments—can be effectively modeled within the framework of effective field theory (EFT). Moreover, Ref. [ 94 ] showed that small-scale 21-cm fluctuations encode significant information about large-scale structures, and that the EFT formalism enables scaling hydrodynamical simulations to large volumes with \mathcal{O}(10\%) accuracy. These insights motivate the use of EFT in combination with quadratic estimators [ 60 ] to reconstruct long-wavelength density modes that are obscured by the foreground wedge, using only the observable modes outside of it [ 95 ] . In this work, we extend these developments in several key directions. Most importantly, we do not assume prior knowledge of the matter density at the field-level. Instead, we jointly infer both the initial conditions (ICs) and the bias parameters from the foreground-filtered 21-cm field by directly exploring the full posterior. To make this high-dimensional inference tractable, we employ a gradient-based sampler that efficiently navigates the joint parameter space. In addition to field-level approaches using perturbation theory, deep learning methods have also been explored for density reconstruction, allowing us to reach smaller scales where perturbation theory alone breaks down, provided small-scale physics is accurately simulated. Deep learning can learn complex, high-dimensional posterior distributions without explicitly modeling (as required in perturbative methods). Variational diffusion models [ 96 ] have been applied to mock galaxy surveys to infer dark matter density fields in two or three dimensions within the CAMELS simulation suite [ 97 , 98 , 99 , 100 ] . Convolutional Neural Networks (CNNs) and U-Nets have also been applied in several works for foreground removal [ 101 , 102 , 103 , 104 , 105 , 106 ] . Additionally, a stochastic interpolant [ 107 ] —a hybrid of diffusion models and normalizing flows—has shown promise for reconstructing wedge-filtered regions in three-dimensional high-redshift 21-cm lightcones centered at z\approx