Restoring Missing Modes of 21cm Intensity Mapping with Deep Learning: Impact on BAO Reconstruction
Xiaolin Luo · 2024
Read Restoring Missing Modes of 21cm Intensity Mapping with Deep Learning: Impact on BAO Reconstruction by Xiaolin Luo (2024) 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-2412-04021.txt.
Introduction
Qian Li and five colleagues use deep learning to restore missing information in 21-centimeter intensity mapping data. Foreground contamination wipes out regions of Fourier space, and in interferometric observations the so-called foreground wedge causes heavy data loss. Those missing modes can damage baryon acoustic oscillation reconstruction, a key cosmological measurement.
The team trains a U-Net to recover the lost modes before reconstruction runs. The restored temperature map correlates with the original signal at about 0.9 at the tested scale. Reconstruction on restored data performs nearly as well as on clean data. A model trained on coarser fields even transfers to finer fields, which the authors link to scale-invariant mode coupling.
Catalog record
| CREATOR | Xiaolin Luo |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-04021
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Restoring Missing Modes of 21cm Intensity Mapping with Deep Learning: Impact on BAO Reconstruction Xiaolin Luo Abstract In 21cm intensity mapping of the large-scale structure (LSS), regions in Fourier space could be compromised by foreground contamination. In interferometric observations, this contamination, known as the foreground wedge, is exacerbated by the chromatic response of antennas, leading to substantial data loss. Meanwhile, the baryonic acoustic oscillation (BAO) reconstruction, which operates in configuration space to "linearize" the BAO signature, offers improved constraints on the sound horizon scale. However, missing modes within these contaminated regions can negatively impact the BAO reconstruction algorithm. To address this challenge, we employ the deep learning model U-Net to recover the lost modes before applying the BAO reconstruction algorithm. Despite hardware limitations, such as GPU memory, our results demonstrate that the AI-restored 21cm temperature map achieves a high correlation with the original signal, with a correlation ratio of approximately 0.9 0.9 at k ∼ 1 h / Mpc k\sim 1{\rm h/Mpc} . Furthermore, subsequent BAO reconstruction indicates that the AI restoration has minimal impact on the performance of the ‘linearized’ BAO signal, proving the effectiveness of the machine learning approach to mitigate the impact of foreground contamination. Interestingly, we demonstrate that the AI model trained on coarser fields can be effectively applied to finer fields, achieving even higher correlation ratio. This success is likely attributable to the scale-invariance properties of non-linear mode coupling in large-scale structure and the hierarchical structure of the U-Net architecture. 1 Introduction In the standard cosmological model, the large-scale structure (LSS) emerges from tiny quantum fluctuations during inflation. These primordial perturbations are amplified by non-linear structure formation processes, giving rise to the cosmic web observed today. Studying such structures provides critical insights into the Universe’s composition and evolutionary history [ 1 , 2 , 3 , 4 , 5 , 6 , 7 ] . In addition to traditional galaxy surveys, 21cm intensity mapping (21IM) has recently emerged as a promising method for mapping LSS, by detecting the total 21cm emission of neutral hydrogen (HI) across various redshifts. [ 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 ] . After the Epoch of Reionization (EoR), neutral hydrogen predominantly resides within galaxies, making 21IM an effective alternative to galaxy surveys. Unlike galaxy redshift surveys, which resolve individual galaxies, intensity mapping captures integrated light, thus offering a more efficient way to map the LSS over extremely large volumes [ 17 , 18 , 19 , 20 , 21 , 22 ] . However, extracting the cosmic signal presents significant challenges due to interference from various sources, including foreground contamination from Galactic and extragalactic sources [ 23 , 24 , 25 ] , radio frequency interference (RFI) [ 26 ] and instrumental systematics [ 12 , 27 ] etc. Among these, foregrounds are particularly problematic, as they are several orders of magnitude stronger than the expected cosmic signal [ 14 , 23 ] , significantly hindering progress in measuring the auto-correlation of the signal. Additionally, the chromatic response of the instrument can further complicate situations by introducing extra coupling that contaminates a large region in Fourier space, known as the foreground wedge [ 12 ] . To address these issues, many techniques have been developed, leveraging the spectral smoothness [ 28 , 29 ] , the dominance [ 13 , 30 , 31 , 32 ] , or statistical properties [ 33 , 34 ] of the foreground. An alternative approach is to ignore the foreground-contaminated regions and conduct cosmological measurements using only the remaining data, a method known as foreground avoidance. However, in either case, a significant portion of Fourier space is compromised or lost by foreground interference. This missing information can significantly impact cosmological measurements. For one, the loss of Fourier modes reduces the statistical constraining power.For baryonic acoustic oscillations (BAO), this loss also affects data post-processing. Most current BAO measurements in galaxy surveys utilize a technique known as BAO reconstruction [ 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 ] , which aims to mitigate the non-linear degradation of the BAO signal. Despite the technical differences among various reconstruction algorithms, these methods generally seek to solve the continuity equation to ‘linearize’ the BAO signature. However, contamination of a large fraction of Fourier modes can significantly impact the reconstruction outcome. [ 43 ] investigated this effect using Fisher information at redshifts z=1-2 and found that it could increase errors by a factor of 3-4 for the angular diameter distance and by \sim 1.5 for H(z) . One potential solution is to use a low-density galaxy sample to compensate for the affected modes [ 44 ] . Alternatively, information loss can be mitigated by leveraging nonlinear mode coupling. For example, the tidal reconstruction method [ 45 ] employs a quadratic estimator derived from the tidal field to recover large-scale modes from small-scale perturbations. However, as a perturbative approach, its effectiveness is primarily limited to large scales. With the widespread application of artificial intelligence (AI), machine learning (ML) has become a powerful tool for processing large-scale structure data and analyzing cosmic datasets [ 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 ] . It has also proven to be highly effective in addressing high-dimensional astronomical problems, such as reconstructing BAO signals from dark matter density fields [ 55 ] , isolating the EoR signals with complex frequency-dependent beam effects [ 56 ] , and identifying ionization regions in images with instrument noise [ 57 , 58 ] . Therefore, in this study, we aims to apply the deep learning technique, particularly the U-Net model, to recover the large-scale structure modes in Fourier regions affected by foreground contamination. Additionally, we will examine the impact of the AI-restored data on BAO reconstruction. The structure of this paper is as follows: Section 2 provides a brief overview of foreground contamination and its impact on BAO reconstruction. In section 3 , we introduce our main methodology, including details of the simulation data for 21cm intensity mapping, the U-Net architecture, and the BAO reconstruction process. The results are presented in Section 4 , and we conclude with a discussion in Section 5 . 2 Foreground Contamination and the Impact on BAO Reconstruction In both single-dish and interferometric observations, the spectral smoothness of the foreground indicates that the LSS signal may be compromised within the Fourier region where the wavenumbers along the line-of-sight direction are small, i.e. k_{\parallel}<k_{\rm int} where k_{\parallel} represent the line-of-sight component of the wave vector \mathbf{k} . This is commonly referred to as the intrinsic foreground. The exact value of k_{\rm int} depends on the specific foreground removal techniques employed and specific instrumental systematics; for our analysis, we have adopted a reasonable value of k_{\rm int}=0.1~{\rm h/Mpc} . Furthermore, the chromatic response of the interferometer complicates the situation [ 59 , 60 , 61 ] by introducing additional coupling between spectral and spatial information. This coupling can lead to foreground leakage from low k_{\parallel} to high k_{\perp} , with the effect being particularly severe at high k_{\perp} , where k_{\perp} is the transverse components