Pith. sign in

REVIEW 3 major objections 5 minor 50 references

A convolutional network trained only on de-wiggled 21 cm simulations recovers BAO wiggles in reconstructed power spectra, evidence that non-linear mode coupling stores large-scale information in small-scale modes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-03 05:01 UTC pith:CZFEZY2A

load-bearing objection A genuinely informative controlled test of whether a CNN recovers BAO from small-scale 21 cm modes, but the result is still a simulation-internal proof of concept, not a demonstration that the mapping is the real one. the 3 major comments →

arxiv 2602.03313 v2 pith:CZFEZY2A submitted 2026-02-03 astro-ph.CO

Seeing Wiggles without Seeing Wiggles: BAO Recovery in 21 cm Intensity Mapping with Deep Learning

Kaifeng Yu , Xin Wang This is my paper
classification astro-ph.CO
keywords 21 cm intensity mappingbaryon acoustic oscillationsforeground wedgemode couplingdeep learningconvolutional neural networkfield-level reconstructionlarge-scale structure
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper asks whether baryon acoustic oscillations, the oscillatory imprint of sound waves in the early universe, can be recovered from 21 cm intensity maps after the large-scale modes that usually carry them are discarded. The authors train a 3D encoder-decoder convolutional network on simulations generated from a de-wiggled power spectrum with no BAO features, then apply it to fields that do contain BAO. The network reconstructs the missing modes from the small-scale modes that survive outside the foreground wedge, and the BAO wiggles reappear in the power spectrum of the reconstructed field. In the noise-free case both amplitude and phase of the lost modes are restored with high fidelity; with thermal noise the phase information remains robust while the reconstructed amplitude becomes biased. A sympathetic reading: this is a proof of concept that foreground avoidance need not cost access to the BAO signal, because non-linear mode coupling has already recorded it in the surviving modes.

Core claim

On its own terms, the paper establishes that non-linear gravitational evolution couples Fourier modes across scales, and that a convolutional network can exploit this coupling to restore large-scale 21 cm information from short-wavelength modes alone. To show the reconstruction is physical rather than memorized, the training set is deliberately built from de-wiggled simulations, with the initial linear power spectrum smoothed to remove all BAO oscillations. The trained network is then applied to simulations that contain BAO wiggles, and the BAO signature is recovered in the power spectrum of the reconstructed fields, with the first three peaks visible in the noise-free case and the first two

What carries the argument

The central machinery is a 3D encoder-decoder convolutional network (a U-Net-style architecture) with skip connections and a global residual connection, trained to predict the difference between a mode-removed 21 cm field and the noise-free full field. The input keeps only modes outside the foreground wedge and with wavenumber above 0.3 h/Mpc, so all linear-scale BAO information is removed. The physical mechanism the network is claimed to exploit is non-linear mode coupling: small-scale modes carry imprints of the large-scale modes that shaped them, so the missing Fourier modes can be inferred from the surviving ones. The de-wiggled training set is the control that isolates this mechanism.

Load-bearing premise

The whole argument rests on the approximate gravity solver plus the empirical recipe for assigning neutral hydrogen to dark matter halos faithfully reproducing the real 21 cm field's cross-scale mode coupling; if either misrepresents the coupling, the network learns a simulation-specific mapping that would not transfer to observations.

What would settle it

Create a test 21 cm field whose small-scale modes have been phase-randomized, or otherwise decorrelated from the large-scale modes, while preserving their power spectrum, then feed the same mode-removed input through the trained network. If the BAO wiggles still appear in the reconstructed power spectrum, the recovery does not come from the physical mode coupling the paper invokes; if they disappear, the mode-coupling explanation is supported.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If the recovery is physical, BAO measurements from 21 cm intensity mapping do not require the large-scale modes inside the foreground wedge; the small-scale, foreground-clean modes suffice.
  • A network trained without BAO features can still be used to measure BAO peak positions, because the phase information in the reconstructed field is robust even with observational noise.
  • The reconstruction shows phase fidelity across different cosmologies, so a single trained model may transfer to data whose cosmology differs from the training set, though amplitude calibration would need separate attention.
  • Field-level mode restoration could complement foreground subtraction, recovering cosmological information that foreground avoidance currently discards.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the mode-coupling explanation is right, the same de-wiggled-control protocol should recover other large-scale cosmological signals erased by wedge cuts, such as primordial non-Gaussianity or scale-dependent bias; this is a natural next test.
  • The amplitude bias under noise suggests a concrete improvement: training the network with noise in the target or with a noise-aware loss could restore unbiased amplitude estimates, something the paper does not demonstrate.
  • The cross-cosmology phase robustness hints that mode-coupling maps may be nearly universal; a practical next step is to train on a deliberately diverse set of cosmologies and quantify how far the phase fidelity extends.
  • A sharper validity test would replace the fast approximate N-body simulations with higher-resolution hydrodynamic simulations for both training and evaluation; if the recovered BAO peaks survive that swap, the method's regime of validity is much broader.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper investigates whether a 3D U-Net can recover large-scale 21 cm brightness-temperature modes, in particular BAO features, from the small-scale modes left after applying foreground-avoidance cuts (a wedge mask plus a k < 0.3 h/Mpc cutoff). To guard against the network memorizing BAO patterns, the training set is deliberately built from simulations with a de-wiggled (Savitzky-Golay smoothed) linear power spectrum, while test data contain the full BAO signal. The authors report high-fidelity reconstruction of missing modes in the noise-free case, robust phase recovery with noise, and reasonable robustness to one alternative cosmology. They present power spectra, transfer functions, cross-correlation coefficients, a direct wiggle/no-wiggle ratio, and a template-fitting BAO extraction.

Significance. The controlled de-wiggled training design is a genuine strength: it makes the BAO-recovery test non-trivial and substantially reduces the concern that the network is interpolating a learned BAO template. The paper also provides a fairly detailed mock-observation pipeline, including a wedge model and SKA-like thermal noise. If the result holds up, it is a useful proof-of-concept that non-linear mode coupling could be exploited to recover information lost to foreground avoidance in 21 cm intensity mapping. The quantitative claims are, however, currently supported mostly by single-realization plots without error bars, and the validation remains internal to a single approximate forward model.

major comments (3)
  1. [Sections 2.1–2.2, 4.3] The central claim that the network 'captures the underlying mode coupling of the density fields' is not uniquely established. The training and test data are generated with the same COLA + sub-grid HI prescription, and COLA itself 'lacks accuracy on small scales' (§2.1). The input modes are exactly the small-scale, non-linear modes where this approximation is most uncertain. The de-wiggled test shows that BAO is not memorized from the training set, but it does not demonstrate that the learned mapping is the physical one. The one alternative-cosmology test changes the cosmological parameters, not the forward-model approximations. To support the claim, the authors should either validate on an independent forward model (e.g., a higher-resolution N-body or hydrodynamical simulation with a different HI prescription) or explicitly restrict the conclusion to 'the mode coupling of simulations of
  2. [Figures 5, 8, 9, 10] All power spectra, transfer functions, correlation coefficients, and BAO ratios are shown for single realizations, with no error bars or ensemble statistics. Since 20 test realizations are available, the quantitative claims — e.g., 'close to unity', 'closely follows', 'the first two peaks remain clearly identifiable' — are not properly quantified. The authors should show the mean and scatter over the test set, at least for the key BAO ratio (Fig. 8) and the transfer functions (Fig. 5). Without these, it is difficult to judge whether the apparent recovery is robust or a favorable draw.
  3. [Section 4.3] The robustness evidence is thinner than the text implies. Only one alternative cosmology is tested, and the two configurations (Figures 9 and 10) are not equally informative: Figure 9 changes the N-body cosmology but keeps the fiducial HI model, so the test does not stress the full forward-model mapping; Figure 10, which changes both, shows a noticeable amplitude mismatch whose origin is left to future work. The claim of 'reasonable robustness' and the conclusion that the network learns the relevant mode coupling would be better supported by testing at least two widely separated cosmologies and by investigating whether the amplitude bias in Fig. 10 is a generic feature of extrapolation or a specific pathology.
minor comments (5)
  1. [Section 2.1] The Savitzky-Golay de-wiggling parameters (window size, polynomial order stated as 'forth-order' typo) are not fully specified. The exact algorithm used to obtain P_nw(k) is important for reproducibility; please state the window length and the code path.
  2. [Section 2.4] The construction of the noise cube is unclear: '16×16 independent sets of noise data, each with dimensions 32×32×512' — how are these combined to form the full noise realization? A brief formula or schematic would help.
  3. [Section 4.2, Eq. (4.3)] The template-fitting red curves in Figure 8 should be interpreted carefully: the BAO oscillations are encoded in the template by construction, so this method is not an independent detection of BAO in the reconstructed field. The direct wiggle/no-wiggle ratio is the primary evidence. The authors acknowledge this in the text, but the statements in the Conclusions ('BAO signal can be successfully recovered') would benefit from explicitly reserving the strongest claim for the ratio-based method.
  4. [Abstract and Conclusions] The phrase 'amplitude and phase of the lost modes can be restored with high fidelity' is accurate only for the noise-free, same-cosmology case. The amplitude mismatch in the alternative-cosmology case and the noise-induced bias are acknowledged later; please qualify the abstract and conclusion statements accordingly.
  5. [General] Typographical and clarity issues: 'forth-order' should be 'fourth-order'; the sentence in Section 3, 'Only the input data are padded using periodic padding of the boundary throughout the training process', is awkwardly phrased. Also, Figure 7 caption: 'spherical averaged' should presumably be 'cylindrical averaged' or 'spherically averaged' depending on context.

Circularity Check

1 steps flagged

The de-wiggled training design is a genuine controlled test; the only circularity is an acknowledged by-construction template-fitting component and a non-load-bearing self-citation.

specific steps
  1. self definitional [Section 4.2, BAO Signature, discussion of Figure 8]
    "When using the template-fitting approach, these smaller-scale peaks remain visible by construction, as the oscillatory features are encoded in the fitting template."

    The template-fitting model (Eq. 4.3) uses O_lin(k)=P_lin(k)/P_nw(k), where P_nw is the same de-wiggled smooth power spectrum used to construct the training data. Consequently, the BAO peaks in the 'recon(model fitting)' curves are present by construction regardless of whether the network actually restored them. The paper explicitly concedes this for smaller-scale peaks. The direct power-spectrum ratio method (the 'baseline' approach) is independent of this template and provides the primary evidence, so the circularity is secondary rather than central.

full rationale

The paper's central claim is not circular. The network is trained exclusively on de-wiggled simulations, then applied to test simulations with BAO wiggles; the BAO recovery is assessed primarily by taking the direct ratio of reconstructed wiggle and no-wiggle fields. This is a genuine out-of-distribution test, not a fitted input renamed as a prediction. The only self-citation used in the forward-model setup (§2.2, reference [27]) supports the sub-grid HI prescription but is not load-bearing: the prescription also cites [36] and is an input assumption rather than a derived result. The shared COLA/HI simulation pipeline for training and test data is a limitation on external validity, but it is not circularity because the controlled de-wiggled test still probes whether the network can transfer to unseen wiggle realizations. The one genuinely by-construction element is the template-fitting confirmation, which the authors themselves flag: the oscillatory features are encoded in the fitting template. Since the independent ratio method already demonstrates the first three BAO peaks, this admitted circular component does not undermine the central conclusion. Overall, the paper earns a low circularity score of 2.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The central result rests on simulation choices rather than first-principles derivations. The paper adopts a specific wedge model, a hand-chosen mode cutoff, an approximate N-body scheme, and an empirical HI prescription. None of these are fitted to the target claim, but they define what the network actually learns and what the recovery claim means.

free parameters (3)
  • wedge width b = 0.1
    Hand-chosen in Eq. (2.4); determines which modes are removed as foreground-contaminated, and therefore the difficulty of reconstruction.
  • mode cutoff k_cut = 0.3 h/Mpc
    Hand-chosen cutoff in §2.3; removes linear scales and most BAO. The network never sees modes below this scale, so the exact value controls the difficulty and the validity of the 'small-scale only' claim.
  • Savitzky-Golay de-wiggling filter = 4th-order polynomial, window unspecified
    Used in §2.1 to build the BAO-free training power spectrum; the smoothing window is not given, so the residual BAO contamination in the training set is not fully quantified.
axioms (5)
  • domain assumption COLA simulations with 10 steps and 512^3 particles in a 1 Gpc/h box capture the non-linear mode coupling relevant for 21 cm fields.
    Authors state COLA 'lacks accuracy on small scales, but it is sufficient for our purposes' (§2.1); if COLA's approximate dynamics bias the small-to-large mode coupling, the network learns a simulation-specific mapping.
  • domain assumption The HI brightness-temperature model (Eqs. 2.1-2.3 plus conditional mass function) faithfully represents the coupling of the real 21 cm field; parameters M0, Mmin, alpha are taken from [37].
    The 21 cm map is constructed from an empirical HI-halo mass relation and conditional mass function; errors in this mapping change the field-level statistics the network relies on.
  • domain assumption The foreground wedge model (Eq. 2.4) with b=0.1 and horizon limit θ=π/2 describes the contaminated region; leakage beyond this wedge is ignored.
    Authors explicitly do not consider a larger foreground-contaminated region [45]; if residual foregrounds leak into the kept modes, the real-data analogue fails.
  • domain assumption A single training cosmology (Planck 2018) is sufficient; mode coupling is approximately universal so the network generalizes.
    Robustness is tested for only one alternative cosmology (§4.3); the conclusion that the network learned physical mode coupling rather than training-set features depends on this generalization.
  • ad hoc to paper The de-wiggled power spectrum removes BAO but preserves the mode-coupling structure; Savitzky-Golay smoothing is a valid separation of wiggle and no-wiggle components.
    The smoothing procedure is a data-processing choice made for this study; if it introduces artifacts, the 'BAO-free training' control is compromised.

pith-pipeline@v1.3.0-alltime-deepseek · 14909 in / 18963 out tokens · 189566 ms · 2026-08-03T05:01:45.895704+00:00 · methodology

0 comments
read the original abstract

The 21 cm intensity mapping provides a promising probe of the large-scale structure. Astrophysical foregrounds, as the main source of contamination to the cosmological 21 cm signal, persist in a wedge-like region of Fourier space due to the inherent chromaticity in radio interferometric observations. The foreground avoidance strategy focuses on utilizing data from relatively clean regions with minimal foreground leakage, at the cost of losing large-scale information. Non-linear structure formation, however, couples Fourier modes across scales, leaving imprints of the missing large-scale modes in the remaining data. In this work, we employ a deep learning approach based on Convolutional Neural Networks (CNNs) to test whether large-scale features of the 21 cm brightness temperature fields, particularly the baryon acoustic oscillations (BAO), can be recovered at the field level using only short-wavelength modes that are beyond the linear scales. To explicitly assess the dependence on the training cosmology, we train the network exclusively on de-wiggled simulations, providing a controlled test of whether the reconstruction arises from physical non-linear mode coupling rather than implicit encoding of BAO features. In the ideal noise-free case, the amplitude and phase of the lost modes can be restored with high fidelity. With instrumental noise included, the reconstructed amplitude becomes biased, while the phase information remains robust. The trained network also exhibits reasonable robustness to variations in the underlying cosmological model. Together, these results suggest that mode restoration offers a complementary approach for extracting cosmological information from future 21 cm intensity mapping analyses.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 45 linked inside Pith

  1. [1]

    Furlanetto, S.P

    S.R. Furlanetto, S.P. Oh and F.H. Briggs,Cosmology at low frequencies: The 21 cm transition and the high-redshift Universe, Phys. Rep.433(2006) 181 [astro-ph/0608032]

    Pith/arXiv arXiv 2006

  2. [2]

    Morales and J.S.B

    M.F. Morales and J.S.B. Wyithe,Reionization and Cosmology with 21-cm Fluctuations, ARA&A48(2010) 127 [0910.3010]

    Pith/arXiv arXiv 2010

  3. [3]

    Pritchard and A

    J.R. Pritchard and A. Loeb,21 cm cosmology in the 21st century,Reports on Progress in Physics75(2012) 086901 [1109.6012]. – 16 –

    Pith/arXiv arXiv 2012

  4. [4]

    Liu and J.R

    A. Liu and J.R. Shaw,Data Analysis for Precision 21 cm Cosmology, PASP132(2020) 062001 [1907.08211]

    Pith/arXiv arXiv 2020

  5. [5]

    Wyithe, A

    J.S.B. Wyithe, A. Loeb and P.M. Geil,Baryonic acoustic oscillations in 21-cm emission: a probe of dark energy out to high redshifts, MNRAS383(2008) 1195 [0709.2955]

    Pith/arXiv arXiv 2008

  6. [6]

    Villaescusa-Navarro, D

    F. Villaescusa-Navarro, D. Alonso and M. Viel,Baryonic acoustic oscillations from 21 cm intensity mapping: the Square Kilometre Array case, MNRAS466(2017) 2736 [1609.00019]

    Pith/arXiv arXiv 2017

  7. [7]

    Chang, U.-L

    T.-C. Chang, U.-L. Pen, K. Bandura and J.B. Peterson,An intensity map of hydrogen 21-cm emission at redshift z ˜0.8, Nature466(2010) 463

    2010

  8. [8]

    Masui, E.R

    K.W. Masui, E.R. Switzer, N. Banavar, K. Bandura, C. Blake, L.-M. Calin et al.,Measurement of 21 cm Brightness Fluctuations at z ˜0.8 in Cross-correlation, ApJ763(2013) L20 [1208.0331]

    Pith/arXiv arXiv 2013

  9. [9]

    Cunnington, Y

    S. Cunnington, Y. Li, M.G. Santos, J. Wang, I.P. Carucci, M.O. Irfan et al.,H I intensity mapping with MeerKAT: power spectrum detection in cross-correlation with WiggleZ galaxies, MNRAS518(2023) 6262 [2206.01579]

    Pith/arXiv arXiv 2023

  10. [10]

    Amiri, K

    M. Amiri, K. Bandura, T. Chen, M. Deng, M. Dobbs, M. Fandino et al.,Detection of Cosmological 21 cm Emission with the Canadian Hydrogen Intensity Mapping Experiment, ApJ 947(2023) 16 [2202.01242]

    Pith/arXiv arXiv 2023

  11. [11]

    Barberi-Squarotti, J.L

    MeerKLASS Collaboration, M. Barberi-Squarotti, J.L. Bernal, P. Bull, S. Camera, I.P. Carucci et al.,MeerKLASS L-band deep-field intensity maps: entering the H I dominated regime, MNRAS537(2025) 3632 [2407.21626]

    Pith/arXiv arXiv 2025

  12. [12]

    Mazumder, L

    A. Mazumder, L. Wolz, Z. Chen, S. Paul, M.G. Santos, M. Jarvis et al.,HI intensity mapping with the MIGHTEE Survey: first results of the HI power spectrum, MNRAS541(2025) 476 [2501.17564]

    Pith/arXiv arXiv 2025

  13. [13]

    Amiri, K

    CHIME Collaboration, M. Amiri, K. Bandura, A. Chakraborty, J.-F. Cliche, M. Dobbs et al., Detection of the Cosmological 21 cm Signal in Auto-correlation at z ˜1 with the Canadian Hydrogen Intensity Mapping Experiment,arXiv e-prints(2025) arXiv:2511.19620 [2511.19620]

    arXiv 2025

  14. [14]

    Alonso, P

    D. Alonso, P. Bull, P.G. Ferreira and M.G. Santos,Blind foreground subtraction for intensity mapping experiments, MNRAS447(2015) 400 [1409.8667]

    Pith/arXiv arXiv 2015

  15. [15]

    Mertens, A

    F.G. Mertens, A. Ghosh and L.V.E. Koopmans,Statistical 21-cm signal separation via Gaussian Process Regression analysis, MNRAS478(2018) 3640 [1711.10834]

    Pith/arXiv arXiv 2018

  16. [16]

    Spinelli, I.P

    M. Spinelli, I.P. Carucci, S. Cunnington, S.E. Harper, M.O. Irfan, J. Fonseca et al.,SKAO H I intensity mapping: blind foreground subtraction challenge, MNRAS509(2022) 2048 [2107.10814]

    Pith/arXiv arXiv 2022

  17. [17]

    Morales, B

    M.F. Morales, B. Hazelton, I. Sullivan and A. Beardsley,Four Fundamental Foreground Power Spectrum Shapes for 21 cm Cosmology Observations, ApJ752(2012) 137 [1202.3830]

    Pith/arXiv arXiv 2012

  18. [18]

    Hazelton, M.F

    B.J. Hazelton, M.F. Morales and I.S. Sullivan,The Fundamental Multi-baseline Mode-mixing Foreground in 21 cm Epoch of Reionization Observations, ApJ770(2013) 156 [1301.3126]

    Pith/arXiv arXiv 2013

  19. [19]

    Liu, A.R

    A. Liu, A.R. Parsons and C.M. Trott,Epoch of reionization window. I. Mathematical formalism, Phys. Rev. D90(2014) 023018 [1404.2596]

    Pith/arXiv arXiv 2014

  20. [20]

    Abdurashidova, J.E

    Z. Abdurashidova, J.E. Aguirre, P. Alexander, Z.S. Ali, Y. Balfour, A.P. Beardsley et al.,First Results from HERA Phase I: Upper Limits on the Epoch of Reionization 21 cm Power Spectrum, ApJ925(2022) 221 [2108.02263]

    Pith/arXiv arXiv 2022

  21. [21]

    Nunhokee, D

    C.D. Nunhokee, D. Null, C.M. Trott, N. Barry, Y. Qin, R.B. Wayth et al.,Limits on the 21 cm Power Spectrum at z = 6.5–7.0 from Murchison Widefield Array Observations, ApJ989(2025) 57 [2505.09097]. – 17 –

    Pith/arXiv arXiv 2025

  22. [22]

    C. Modi, M. White, A. Slosar and E. Castorina,Reconstructing large-scale structure with neutral hydrogen surveys, J. Cosmology Astropart. Phys.2019(2019) 023 [1907.02330]

    Pith/arXiv arXiv 2019

  23. [23]

    Gagnon-Hartman, Y

    S. Gagnon-Hartman, Y. Cui, A. Liu and S. Ravanbakhsh,Recovering the wedge modes lost to 21-cm foregrounds, MNRAS504(2021) 4716 [2102.08382]

    Pith/arXiv arXiv 2021

  24. [24]

    Prelogovi´ c, A

    D. Prelogovi´ c, A. Mesinger, S. Murray, G. Fiameni and N. Gillet,Machine learning astrophysics from 21 cm lightcones: impact of network architectures and signal contamination, MNRAS509(2022) 3852 [2107.00018]

    Pith/arXiv arXiv 2022

  25. [25]

    Kennedy, J.C

    J. Kennedy, J.C. Carr, S. Gagnon-Hartman, A. Liu, J. Mirocha and Y. Cui,Machine-learning recovery of foreground wedge-removed 21-cm light cones for high-z galaxy mapping, MNRAS 529(2024) 3684 [2308.09740]

    Pith/arXiv arXiv 2024

  26. [26]

    Sabti, R

    N. Sabti, R. Purandhar Reddy Sudha, J.B. Mu˜ noz, S. Mishra-Sharma and T. Youn,A generative modeling approach to reconstructing 21 cm tomographic data,Machine Learning: Science and Technology6(2025) 015039 [2407.21097]

    Pith/arXiv arXiv 2025

  27. [27]

    Q. Li, X. Wang, X.-D. Li, J. Ding, T.-C. Luan and X. Luo,Restoring missing modes of 21cm intensity mapping with deep learning: impact on BAO reconstruction, J. Cosmology Astropart. Phys.2025(2025) 082 [2412.04021]

    Pith/arXiv arXiv 2025

  28. [28]

    T.-C. Luan, X. Wang, J. Ding, Q. Li, X.-D. Li and W. Zhu,Recovering Cosmic Structure with a Simple Physical Constraint, ApJ983(2025) 166 [2503.20434]

    Pith/arXiv arXiv 2025

  29. [29]

    Chen, K.-F

    S.-F. Chen, K.-F. Chen and C. Dvorkin,Field-level reconstruction from foreground-contaminated 21-cm maps, J. Cosmology Astropart. Phys.2025(2025) 082 [2508.13265]

    arXiv 2025

  30. [30]

    Qin, K.-F

    W. Qin, K.-F. Chen, K. Schutz and A. Liu,Effective bias expansion for circumventing 21 cm foregrounds,arXiv e-prints(2025) arXiv:2508.13268 [2508.13268]

    arXiv 2025

  31. [31]

    Aghanim, Y

    Planck Collaboration, N. Aghanim, Y. Akrami, F. Arroja, M. Ashdown, J. Aumont et al., Planck 2018 results. I. Overview and the cosmological legacy of Planck, A&A641(2020) A1 [1807.06205]

    Pith/arXiv arXiv 2018

  32. [32]

    Tassev, M

    S. Tassev, M. Zaldarriaga and D.J. Eisenstein,Solving large scale structure in ten easy steps with COLA, J. Cosmology Astropart. Phys.2013(2013) 036 [1301.0322]

    Pith/arXiv arXiv 2013

  33. [33]

    J. Koda, C. Blake, F. Beutler, E. Kazin and F. Marin,Fast and accurate mock catalogue generation for low-mass galaxies, MNRAS459(2016) 2118 [1507.05329]

    Pith/arXiv arXiv 2016

  34. [34]

    D. Blas, J. Lesgourgues and T. Tram,The Cosmic Linear Anisotropy Solving System (CLASS). Part II: Approximation schemes, J. Cosmology Astropart. Phys.2011(2011) 034 [1104.2933]

    Pith/arXiv arXiv 2011

  35. [35]

    Behroozi, R.H

    P.S. Behroozi, R.H. Wechsler and H.-Y. Wu,The ROCKSTAR Phase-space Temporal Halo Finder and the Velocity Offsets of Cluster Cores, ApJ762(2013) 109 [1110.4372]

    Pith/arXiv arXiv 2013

  36. [36]

    Seehars, A

    S. Seehars, A. Paranjape, A. Witzemann, A. Refregier, A. Amara and J. Akeret,Simulating the large-scale structure of HI intensity maps, J. Cosmology Astropart. Phys.2016(2016) 001 [1509.01589]

    Pith/arXiv arXiv 2016

  37. [37]

    Villaescusa-Navarro, S

    F. Villaescusa-Navarro, S. Genel, E. Castorina, A. Obuljen, D.N. Spergel, L. Hernquist et al., Ingredients for 21 cm Intensity Mapping, ApJ866(2018) 135 [1804.09180]

    Pith/arXiv arXiv 2018

  38. [38]

    Cooray and R

    A. Cooray and R. Sheth,Halo models of large scale structure, Phys. Rep.372(2002) 1 [astro-ph/0206508]

    Pith/arXiv arXiv 2002

  39. [39]

    Mo and S.D.M

    H.J. Mo and S.D.M. White,An analytic model for the spatial clustering of dark matter haloes, MNRAS282(1996) 347 [astro-ph/9512127]

    Pith/arXiv arXiv 1996

  40. [40]

    Sheth and G

    R.K. Sheth and G. Tormen,An excursion set model of hierarchical clustering: ellipsoidal collapse and the moving barrier, MNRAS329(2002) 61 [astro-ph/0105113]. – 18 –

    Pith/arXiv arXiv 2002

  41. [41]

    Bull, P.G

    P. Bull, P.G. Ferreira, P. Patel and M.G. Santos,Late-time Cosmology with 21 cm Intensity Mapping Experiments, ApJ803(2015) 21 [1405.1452]

    Pith/arXiv arXiv 2015

  42. [42]

    L. Wolz, A. Pourtsidou, K.W. Masui, T.-C. Chang, J.E. Bautista, E.-M. M¨ uller et al.,H I constraints from the cross-correlation of eBOSS galaxies and Green Bank Telescope intensity maps, MNRAS510(2022) 3495 [2102.04946]

    Pith/arXiv arXiv 2022

  43. [43]

    Datta, J.D

    A. Datta, J.D. Bowman and C.L. Carilli,Bright Source Subtraction Requirements for Redshifted 21 cm Measurements, ApJ724(2010) 526 [1005.4071]

    Pith/arXiv arXiv 2010

  44. [44]

    Chapman, S

    E. Chapman, S. Zaroubi, F. Abdalla, F. Dulwich, V. Jeli´ c and B. Mort,The Effect of Foreground Mitigation Strategy on EoR Window Recovery,arXiv e-prints(2014) arXiv:1408.4695 [1408.4695]

    Pith/arXiv arXiv 2014

  45. [45]

    Munshi, F.G

    S. Munshi, F.G. Mertens, L.V.E. Koopmans, A.R. Offringa, E. Ceccotti, S.A. Brackenhoff et al.,Beyond the horizon: Quantifying the full sky foreground wedge in the cylindrical power spectrum, A&A693(2025) A276 [2407.10686]

    Pith/arXiv arXiv 2025

  46. [46]

    Maartens, F.B

    R. Maartens, F.B. Abdalla, M. Jarvis and M.G. Santos,Overview of Cosmology with the SKA, inAdvancing Astrophysics with the Square Kilometre Array (AASKA14), p. 16, Apr., 2015, DOI

    2015

  47. [47]

    Bacon, R.A

    Square Kilometre Array Cosmology Science Working Group, D.J. Bacon, R.A. Battye, P. Bull, S. Camera, P.G. Ferreira et al.,Cosmology with Phase 1 of the Square Kilometre Array Red Book 2018: Technical specifications and performance forecasts, PASA37(2020) e007 [1811.02743]

    Pith/arXiv arXiv 2018

  48. [48]

    S. He, Y. Li, Y. Feng, S. Ho, S. Ravanbakhsh, W. Chen et al.,Learning to predict the cosmological structure formation,Proceedings of the National Academy of Science116(2019) 13825 [1811.06533]

    Pith/arXiv arXiv 2019

  49. [49]

    Jamieson, Y

    D. Jamieson, Y. Li, R.A. de Oliveira, F. Villaescusa-Navarro, S. Ho and D.N. Spergel, Field-level Neural Network Emulator for Cosmological N-body Simulations, ApJ952(2023) 145 [2206.04594]

    Pith/arXiv arXiv 2023

  50. [50]

    Gil-Mar ´ ın, W.J

    H. Gil-Mar ´ ın, W.J. Percival, A.J. Cuesta, J.R. Brownstein, C.-H. Chuang, S. Ho et al.,The clustering of galaxies in the SDSS-III Baryon Oscillation Spectroscopic Survey: BAO measurement from the LOS-dependent power spectrum of DR12 BOSS galaxies, MNRAS460 (2016) 4210 [1509.06373]. – 19 –

    Pith/arXiv arXiv 2016