Pith. sign in

REVIEW 5 major objections 5 minor 68 references

DeepCHART: Mapping the 3D dark matter density field from Ly$\alpha$ forest surveys using deep learning

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read DeepCHART reconstructs the 3D dark matter density field at z=2.5 from Lyα forest spectra and galaxy positions, reaching voxel-wise correlations ρ≈0.77 for current surveys and ρ≈0.90 for future surveys.

desk verdict Solid proof-of-concept for deep-learning Lyα tomography, but the quantitative claims are confined to a single fixed-physics simulation suite; worth reviewing, not yet ready for real-data claims. read the letter →

arxiv 2507.00135 v1 pith:IL6PRO75 submitted 2025-06-30 astro-ph.CO astro-ph.GA

Soumak Maitra (TIFR) , Matteo Viel , Girish Kulkarni This is my paper · ORCID
classification astro-ph.COastro-ph.GA
keywords Lyαforesttomography3DdarkmatterreconstructionvariationalautoencoderU-Netcosmicwebhydrodynamicalsimulationslikelihood-freeinferencefield-levelcosmological
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper introduces DeepCHART, a deep-learning framework that reconstructs the three-dimensional dark matter density field at redshift $z=2.5$ from sparse Ly$\alpha$ forest spectra and coeval galaxy positions. It trains a three-dimensional variational autoencoder with a U-Net backbone on hydrodynamical simulations, so that once trained it performs likelihood-free inference, producing a full volume in under half a second rather than iteratively fitting a forward model. For current survey densities (mean sightline spacing $d_\perp=2.4\,h^{-1}\mathrm{cMpc}$, with Subaru/PFS-like resolution and noise), the reconstruction reaches a voxel-wise Pearson correlation of $\rho\simeq 0.77$ over density contrasts $0.4<\Delta_{\mathrm{DM}}<15$; for future surveys with $d_\perp\simeq 1\,h^{-1}\mathrm{cMpc}$ it reaches $\rho\simeq 0.90$ over $0.25<\Delta_{\mathrm{DM}}<40$. The reconstructed fields also reproduce the density PDF and the matter power spectrum, with only mild suppression at intermediate scales, and preserve the main cosmic-web environments. If these mock-based numbers survive contact with real data, the framework gives a fast, scalable route to field-level cosmological inference from the next generation of spectroscopic surveys.

What carries the argument

The engine is a 3D variational autoencoder built on a U-Net backbone, trained with an evidence-lower-bound loss (mean squared error between predicted and true log-density plus a $\beta$-annealed KL term) and a 512-dimensional Gaussian latent space. The input is a two-channel sparse volume: Ly$\alpha$ flux sightlines embedded in a $96\times 96\times 288$ grid with anisotropic voxels, and galaxy overdensities obtained by cloud-in-cell interpolation of a halo occupation distribution (HOD) galaxy catalog. The encoder compresses this to a $12\times 12\times 36$ latent tensor; the decoder mirrors the encoder and uses skip connections that preserve high-resolution spatial structure, while convolutional kernels are elongated along the line of sight to match the voxel anisotropy. Training uses nine hydrodynamical simulation realizations with a tenth held out for testing; mock spectra are forward-modeled with a Gaussian line-spread function at resolution $R=2500$ and a power-law signal-to-noise distribution, and the target field is the dark matter density smoothed on $2\,h^{-1}\mathrm{cMpc}$ (current) or $1\,h^{-1}\mathrm{cMpc}$ (future) scales.

What would settle it

Run DeepCHART on real Subaru/PFS or CLAMATO Ly$\alpha$ forest and galaxy data in a volume where an independent tracer of the same matter distribution (a deeper galaxy sample or a CMB lensing convergence map) is available, smooth both fields to the same scale, and compute the voxel-wise Pearson correlation over $0.4<\Delta_{\mathrm{DM}}<15$; if the correlation falls clearly below the simulated $\rho\simeq 0.77$ (or below $\rho\simeq 0.90$ for a dense future-type dataset), the central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that a single network can invert sparse, noisy Ly$\alpha$ forest sightlines, optionally combined with galaxy positions, into a voxel-wise faithful 3D dark matter density field at $z=2.5$, including nonlinear and baryonic structure that analytic or Gaussian-based tomographic methods smooth away. Quantitatively, the paper reports $\rho\simeq 0.77$ for current surveys over $0.4<\Delta_{\mathrm{DM}}<15$ with regression slope near unity after clipping, and $\rho\simeq 0.90$ for future surveys over $0.25<\Delta_{\mathrm{DM}}<40$; adding galaxies improves recovery of dense regions, and joint training helps even when galaxies are absent at inference time. The one-point density PDF is recovered in the bulk, with some suppression of the extreme tails under current survey conditions; the recovered power spectrum matches the clipped true spectrum within error bars over $k\approx 0.7$–$5\,h\,\mathrm{cMpc}^{-1}$, with up to about 40% suppression around $k\sim 2\,h\,\mathrm{cMpc}^{-1}$ in the finer-resolution future scenario. Cosmic web classification for the future scenario identifies 81% of voids, 75% of sheets, 63% of filaments, and 43% of nodes. The paper frames this as a fast, likelihood-free field-level inference tool rather than a measurement of cosmological parameters.

Load-bearing premise

The load-bearing assumption is that the mock observations used for training and testing faithfully represent real survey data, including a fixed thermal and ionization history, a simplified noise model, and idealized HOD galaxy populations; if real Ly$\alpha$ forest spectra and galaxy samples differ materially from these mocks, the quoted reconstruction correlations will not transfer to actual surveys.

Editorial extensions

If this is right

  • For Subaru/PFS, CLAMATO, and LATIS, the framework offers reconstruction at $\rho\simeq 0.77$ over moderate overdensities, providing a fast complement to analytic tomographic maps at $z=2.5$.
  • For ELT/MOSAIC and WST/IFS-class surveys, the same network reaches $\rho\simeq 0.90$ and extends reliable reconstruction to $1\,h^{-1}\mathrm{cMpc}$ smoothing, opening small-scale non-linear structure to observation.
  • Because the VAE samples from a latent distribution, each observation yields an ensemble of statistically consistent density fields, giving a probabilistic characterization of field-level uncertainty.
  • Recovered PDFs and power spectra mean the reconstructed volumes can feed cosmological analyses directly, though the intermediate-scale power suppression for the 1 Mpc smoothing case must be modeled or corrected.
  • The architecture is modular over tracer and target fields, so the same training scheme can be redirected to other observables such as weak lensing shear or 21cm maps.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A decisive test the paper leaves implicit is a transfer run on real survey data: the published numbers are measured on simulations, and the gap between mock and real spectra (continuum fitting, metal lines, variable thermal history) is exactly what the fixed-physics training set cannot capture.
  • Because the training set uses one thermal and ionization history, the network's mapping carries an implicit assumption about IGM physics; conditioning the latent space on thermal parameters would turn DeepCHART from a fixed-universe emulator into a tool for measuring the thermal history.
  • The joint-training benefit seen when galaxies are omitted at inference time suggests a graceful-degradation strategy for surveys with partial coverage, and motivates quantifying reconstruction quality as a function of galaxy completeness.
  • The latent space itself, being 512-dimensional and Gaussian, is a candidate summary for cosmology: one could train an emulator to map the latent mean to cosmological parameters, effectively using the tomographic volume as a field-level statistic.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper presents DeepCHART, a 3D U-Net-based variational autoencoder trained on nine GADGET-3 hydrodynamical simulations to reconstruct the 3D dark matter overdensity field at z=2.5 from sparse Lyα forest sightlines and galaxy positions. Two survey configurations are studied: a 'current' scenario with d⊥=2.4 h^-1 cMpc sightline spacing and galaxies matching Subaru/PFS/CLAMATO/LATIS depths, and a 'future' scenario with d⊥=1.0 h^-1 cMpc and denser tracers. The authors report voxel-wise Pearson correlations of ρ≈0.77 (current, LF+G) and ρ≈0.90 (future, LF+G) over restricted overdensity ranges, alongside PDF recovery, power-spectrum comparisons, and cosmic-web classification accuracies. The claimed contribution is a fast (sub-0.5 s per subvolume), likelihood-free, scalable tomographic reconstruction framework for upcoming spectroscopic surveys.

Significance. If the quantitative claims are taken at face value, DeepCHART would be a useful addition to the toolbox for field-level inference from Lyα forest surveys, with a speed advantage over iterative methods such as TARDIS. The paper provides public code and evaluates on a held-out simulation, which are strengths. The central numerical claims, however, are tied to a single fixed-physics simulation suite and to several evaluation choices that need tightening: the headline correlations are quoted after clipping to a favorable density range, the power-spectrum validation compares against a clipped true field, and the manuscript contains an internal inconsistency in the network output shape. These issues are fixable, but they currently prevent the abstract-level 'high-fidelity reconstruction' claim from being fully supported.

major comments (5)
  1. [Sec. 4.3 and Fig. 7] The power-spectrum validation compares reconstructed spectra to the 'True (clipped)' spectra, where the true field has been restricted to 0.4<Δ<15 (current) or 0.25<Δ<40 (future), rather than to the full true field. Because the reconstruction is known to be biased outside these ranges from Sec. 4.1, this comparison removes the very failure modes that contribute most to the full-field error. The statement that DeepCHART 'robustly recovers the matter power spectrum' is therefore not established for the full field. Please report the ratio P_recon/P_true for the unclipped true field, and document quantitatively how much the clipping changes the inferred agreement.
  2. [Sec. 4.1, Figs. 4-5, and abstract] The headline correlations are computed after restricting to voxels whose unsmoothed (simulation-resolution) dark matter density lies in 0.4<Δ<15 or 0.25<Δ<40, while the values actually correlated and plotted are the smoothed log-density fields. As shown in Fig. 4, the dynamic range of the correlated values is roughly logΔ∈[-0.2,0.2], not the quoted range. The abstract's phrasing 'over the density range 0.4<Δ<15' is misleading. Please either define the clipping in terms of the reconstructed (smoothed) target field, or state explicitly that the quoted range refers to the unsmoothed field used only for voxel selection, and report the unclipped correlations with equal prominence.
  3. [Sec. 3 and Eq. (6)] The architecture description is internally inconsistent: the input and target fields are 96×96×288 grids (Sec. 3), but the decoder is stated to 'recover the output spatial size of 96×96×96'. The MSE term in Eq. (6) requires the predicted and true fields to have identical shapes. Please correct the text or the diagram, or explain the resampling/up-sampling procedure that reconciles the two shapes. This is load-bearing for reproducibility.
  4. [Sec. 2.3 vs. Fig. 5 caption] There is a factor-of-three discrepancy in the future galaxy number density: Sec. 2.3 states a '∼15-fold increase' relative to current surveys, while the caption of Fig. 5 says 'a galaxy sample five times denser than Subaru PFS expectations'. Since the future-survey correlation and power-spectrum results depend directly on this input, please specify exactly which number density was used in each future-survey model and verify that all quoted results correspond to that value.
  5. [Sec. 5 and abstract] All quantitative fidelity claims are based on a single fixed thermal and ionization history, a single power-law SNR noise model, and an idealized HOD with no galaxy redshift errors. Section 5 concedes the fixed thermal history, but the abstract does not qualify the numbers, presenting ρ≈0.77 and ρ≈0.90 as achievable for 'current surveys' and 'future surveys' without stating that these are mock-survey demonstrations under one assumed physics model. Please add an explicit caveat to the abstract, or include a robustness test (e.g., varying the temperature-density relation, SNRs, or HOD) that would support transfer of the quoted fidelities to real data.
minor comments (5)
  1. [Sec. 4.1.1 and Fig. 4] The sentence 'In both cases, the slope of the fit remains effectively unity (y=1.00x)' is contradicted by Fig. 4 bottom-left, which shows y=0.67x for the unclipped LF+G case; please correct the text or the figure.
  2. [Appendix A and Sec. 4.1.1] The appendix figure is referred to as 'Figure 1' both in the appendix and in Sec. 4.1.1 ('see Figure 1'), which conflicts with the paper's main Figure 1; please renumber it as Figure A1 and update the cross-references.
  3. [Sec. 4.3] The power spectrum is described as computed on a 'cubic grid with dimensions 96^3', but the native grid is 96×96×288; please specify how the line-of-sight direction is resampled or binned before the spherically averaged P(k) is computed.
  4. [Sec. 4.3] The error bars on P(k) are computed from 20 subvolumes that partially overlap within a single (40 h^-1 cMpc)^3 test box; please state explicitly that these are not independent realizations, or use non-overlapping subvolumes to estimate cosmic variance.
  5. [Sec. 2] Minor typographical point: 'Zeldovich approximation' is conventionally written 'Zel'dovich approximation'; please adjust for consistency with the literature.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DeepCHART is a supervised learning pipeline trained on nine simulation seeds and evaluated on a held-out tenth seed; no equation reduces to its inputs.

full rationale

The claimed reconstruction is a learned mapping from mock Ly-alpha forest spectra and HOD galaxy fields to the smoothed dark matter overdensity in a fixed-physics hydrodynamical simulation. The chain is: (1) GADGET-3 produces dark matter and baryon fields; (2) Voigt-profile optical depth and forward-modeled noise and resolution produce mock spectra; (3) the VAE is trained with an MSE-vs-truth ELBO loss on nine random seeds; (4) all headline metrics (Pearson rho, PDF, power spectrum, cosmic web fractions) are evaluated on the tenth, held-out seed. Therefore the reported rho values are out-of-sample predictions, not fits to the evaluation data. The post-hoc clipping to 0.4<Delta_DM<15 or 0.25<Delta_DM<40 is an evaluation selection and is accompanied by unclipped values (e.g., rho=0.728, 0.757, 0.898, 0.846 and slopes 0.58-0.87), so it is not a fitted parameter dressed as a prediction. The self-citations (Maitra et al. 2019, 2022a, 2022b, 2024) appear only in the introduction as examples of machine learning and bispectrum analysis, and none of the paper's central premises depends on them. The forward-model assumptions (Puchwein et al. 2019 thermal history, Stark et al. 2015 SNR distribution, Zheng et al. 2007 HOD, Horowitz et al. 2019 lambda_th) are all external, stated, and not replaced by a self-citation chain. No uniqueness theorem, and no ansatz, is imported from the authors' prior work. The Section 5 limitation that only a fixed thermal/ionization history is considered concerns external validity, i.e., whether the single-simulation proxy transfers to real observations; that is an empirical faithfulness assumption, not circularity. The internal inconsistency between a '~15-fold' (Sec. 2.3) and 'five times denser' (Fig. 5 caption) future galaxy density is a correctness risk but does not make any claim reduce to its input by construction.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim depends on several manually chosen parameters, most notably the density clipping ranges used for the headline metrics and the smoothing scales. The method also rests on domain assumptions about simulation fidelity, forward modeling, and generalization to real data. No new physical entities are introduced.

free parameters (6)
  • Current survey clipping range = 0.4 < ΔDM < 15
    Post-hoc chosen to exclude extreme densities where reconstruction is poor; the headline ρ=0.77 is computed only over this clipped range.
  • Future survey clipping range = 0.25 < ΔDM < 40
    Same as above, for the future configuration; broader range used to report ρ≈0.90.
  • Current smoothing scale = L_S = 2 h^-1 cMpc
    Chosen to match the effective transverse resolution of 2.4 h^-1 cMpc sightline spacing; affects the target field.
  • Future smoothing scale (fine) = L_S = 1 h^-1 cMpc
    Matched to the 1 h^-1 cMpc sightline spacing of future surveys; central for the 1 h^-1 cMpc reconstruction claim.
  • Cosmic web eigenvalue threshold = λ_th = 0.05
    Adopted to roughly reproduce the void fraction of Horowitz et al. (2019); affects all cosmic web classification percentages.
  • Future galaxy number density scaling = inconsistent (5-fold vs 15-fold)
    Section 2.3 says '~15-fold increase', Figure 5 caption says 'five times denser'; the adopted density changes the future-survey results.
assumptions (5)
  • domain assumption The hydrodynamical simulations (GADGET-3, Sherwood-Relics physics) accurately represent the IGM and dark matter distribution at z=2.5.
    The entire training set is generated from these simulations; if the simulations are wrong, the learned mapping is wrong. Invoked in Section 2.
  • domain assumption The forward model for Lyα forest spectra (Voigt profile, Gaussian LSF, SNR power-law noise) matches real observations.
    The network is trained on these mock spectra; any mismatch with real data degrades reconstruction. Invoked in Section 2.2.
  • domain assumption The HOD galaxy distribution with the chosen parameters represents the actual galaxy population in the surveys.
    Galaxy positions are used as inputs; unrealistic galaxies would bias the reconstruction. Invoked in Section 2.3.
  • ad hoc to paper The network generalizes from nine training realizations to unseen realizations and to the real universe.
    Test is on one held-out realization from the same simulation suite; generalization to real data is untested. Discussed in Section 5 as a limitation.
  • domain assumption The fixed thermal and ionization history is representative of the real IGM.
    Explicitly stated in Section 5: 'we focus here on a fixed thermal and ionization history'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepCHART: Mapping the 3D dark matter density field from Ly$\alpha$ forest surveys using deep learning." pith.science (2026). https://pith.science/paper/IL6PRO75

@misc{pith2026250700135,
  author       = {Pith},
  title        = {Pith review of: DeepCHART: Mapping the 3D dark matter density field from Ly$\alpha$ forest surveys using deep learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IL6PRO75}},
  note         = {Machine review of arXiv:2507.00135}
}
abstract

We present DeepCHART (Deep learning for Cosmological Heterogeneity and Astrophysical Reconstruction via Tomography), a deep learning framework designed to reconstruct the three-dimensional dark matter density field at redshift $z=2.5$ from Ly$\alpha$ forest spectra. Leveraging a 3D variational autoencoder with a U-Net architecture, DeepCHART performs fast, likelihood-free inference, accurately capturing the non-linear gravitational dynamics and baryonic processes embedded in cosmological hydrodynamical simulations. When applied to joint datasets combining Ly$\alpha$ forest absorption and coeval galaxy positions, the reconstruction quality improves further. For current surveys, such as Subaru/PFS, CLAMATO, and LATIS, with an average transverse sightline spacing of $d_\perp=2.4h^{-1}$cMpc, DeepCHART achieves high-fidelity reconstructions over the density range $0.4<\Delta_{\rm DM}<15$, with a voxel-wise Pearson correlation coefficient of $\rho\simeq 0.77$. These reconstructions are obtained using Ly$\alpha$ forest spectra with signal-to-noise ratios as low as 2 and instrumental resolution $R=2500$, matching Subaru/PFS specifications. For future high-density surveys enabled by instruments such as ELT/MOSAIC and WST/IFS with $d_\perp\simeq 1h^{-1}\mathrm{cMpc}$, the correlation improves to $\rho\simeq 0.90$ across a wider dynamic range ($0.25<\Delta_{\rm DM}<40$). The framework reliably recovers the dark matter density PDF as well as the power spectrum, with only mild suppression at intermediate scales. In terms of cosmic web classification, DeepCHART successfully identifies 81% of voids, 75% of sheets, 63% of filaments, and 43% of nodes. We propose DeepCHART as a powerful and scalable framework for field-level cosmological inference, readily generalisable to other observables, and offering a robust, efficient means of maximising the scientific return of upcoming spectroscopic surveys.

Figures

Figures reproduced from arXiv: 2507.00135 by the authors.

Figure 1
Figure 1. Cumulative number density of star-forming galaxies as a function of rest-frame UV absolute magnitude (𝑀AB (1700, Å)) at 𝑧 ≈ 3, based on the luminosity function of Reddy & Steidel (2009). Colored points and lines indicate the limiting magnitude, cumulative sightline density, and mean transverse separation (𝑑⊥) achievable by current and upcoming spectroscopic surveys targeting Ly𝛼 forest tomography. Labels highlight m… view at source ↗
Figure 2
Figure 2. Schematic architecture of our 3D U-Net-based Variational Autoencoder (VAE) used for tomographic reconstruction of the dark matter density field. The network takes as input a stack of Ly𝛼 forest sightlines and galaxies, which are processed by the encoder through successive 3D convolutional layers and residual blocks, progressively compressing the data into a low-dimensional latent space parameterized by mean (𝜇𝑖) and… view at source ↗
Figure 3
Figure 3. Slices of thickness 2 ℎ −1 cMpc through the 3D dark matter density field. From left to right: (1) The true unsmoothed log ΔDM field with the resolution of the simulation. (2) The true log ΔDM field smoothed with a Gaussian filter of scale 𝐿𝑆 = 2 ℎ −1 cMpc. (3) Reconstruction of the smoothed density field using only Ly𝛼 forest (LF) data. (4) Reconstruction of the smoothed density field using combined Ly𝛼 forest and g… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Voxel-wise comparison between the reconstructed and true logΔDM values from the VAE reconstructions, with the dark matter field smoothed over a scale of 2 ℎ −1 cMpc. Each panel shows 2D density contours and binned medians of the predictions, along with a robust linear …
Figure 5
Figure 5. Figure 5: Slices and voxel-wise comparisons of the reconstructed and true dark matter density fields at 𝑧 = 2.5. LF+G denotes reconstructions combining both Ly𝛼 forest sightlines and galaxy tracers. Top row: 2D slices of the true and reconstructed fields, shown in log ΔDM, smoot…
Figure 6
Figure 6. Figure 6: One-point probability distribution functions (PDFs) of the dark matter density contrast, 𝛿DM = 𝜌/𝜌¯ − 1, comparing the true simulated distributions (solid, dashed, and dotted black lines) to reconstructed fields from DeepCHART under different survey configurations and …
Figure 7
Figure 7. Figure 7: Comparison of the spherically-averaged 3D power spectrum, 𝑃(𝑘), for the true dark matter density fields and various reconstructions at redshift 𝑧 = 2.5. The true power spectra from the original simulations, including different smoothing scales (𝐿𝑆 = 1 ℎ −1 cMpc and 𝐿𝑆 …
Figure 8
Figure 8. Figure 8: Recovery of cosmic web structures from tomographic reconstructions. Each row presents a comparison between the true cosmic web classification (left), the reconstructed field (middle), and the corresponding normalized confusion matrix (right) for different survey scenar…
Figure 1
Figure 1. Figure 1: LF-limited inference with LF+G-trained model. From left to right: (1) True dark matter field smoothed over 𝐿𝑆 = 2 ℎ −1 cMpc; (2) Reconstruction using a model trained on LF+G input but evaluated using only LF data; (3) Voxel-wise comparison of the reconstructed and true…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

68 extracted references · 8 canonical work pages

  1. [1]

    H., Gardner J., 2001, @doi [ ] 10.1086/323370 , http://adsabs.harvard.edu/abs/2001ApJ...561..521A 561, 521

    Aguirre A., Hernquist L., Schaye J., Katz N., Weinberg D. H., Gardner J., 2001, @doi [ ] 10.1086/323370 , http://adsabs.harvard.edu/abs/2001ApJ...561..521A 561, 521

    doi:10.1086/323370 2001

  2. [2]

    F., 1997, @doi [ ] 10.1086/303908 , http://adsabs.harvard.edu/abs/1997ApJ...479..523B 479, 523

    Bi H., Davidsen A. F., 1997, @doi [ ] 10.1086/303908 , http://adsabs.harvard.edu/abs/1997ApJ...479..523B 479, 523

    doi:10.1086/303908 1997

  3. [4]

    S., Puchwein E., Sijacki D., Haehnelt M

    Bolton J. S., Puchwein E., Sijacki D., Haehnelt M. G., Kim T.-S., Meiksin A., Regan J. A., Viel M., 2017, @doi [ ] 10.1093/mnras/stw2397 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.464..897B 464, 897

    doi:10.1093/mnras/stw2397 2017

  4. [5]

    Caucci S., Colombi S., Pichon C., Rollinde E., Petitjean P., Sousbie T., 2008, @doi [ ] 10.1111/j.1365-2966.2008.13016.x , http://cdsads.u-strasbg.fr/abs/2008MNRAS.386..211C 386, 211

    arXiv 2008

  5. [6]

    D., 2018, @doi [Physical Review D] 10.1103/physrevd.97.083004 , 97

    Charnock T., Lavaux G., Wandelt B. D., 2018, @doi [Physical Review D] 10.1103/physrevd.97.083004 , 97

    doi:10.1103/physrevd.97.083004 2018

  6. [7]

    Cisewski J., Croft R. A. C., Freeman P. E., Genovese C. R., Khandai N., Ozbek M., Wasserman L., 2014, @doi [ ] 10.1093/mnras/stu475 , https://ui.adsabs.harvard.edu/abs/2014MNRAS.440.2599C 440, 2599

    doi:10.1093/mnras/stu475 2014

  7. [8]

    Croft R. A. C., Weinberg D. H., Katz N., Hernquist L., 1998, @doi [ ] 10.1086/305289 , http://adsabs.harvard.edu/abs/1998ApJ...495...44C 495, 44

    doi:10.1086/305289 1998

  8. [9]

    arXiv:1611.00036

    DESI Collaboration et al., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1611.00036 , https://ui.adsabs.harvard.edu/abs/2016arXiv161100036D p. arXiv:1611.00036

    arXiv doi:10.48550/arxiv.1611.00036 2016

Show all 68 references
  1. [10]

    DESI Collaboration et al., 2024, @doi [ ] 10.3847/1538-3881/ad3217 , https://ui.adsabs.harvard.edu/abs/2024AJ....168...58D 168, 58

    2024 doi

  2. [11]

    Feng Y., Chu M.-Y., Seljak U., McDonald P., 2016, @doi [ ] 10.1093/mnras/stw2123 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.463.2273F 463, 2273

    2016 doi

  3. [12]

    E., Hoffman Y., Gottl \"o ber S., Klypin A., Yepes G., 2009, @doi [ ] 10.1111/j.1365-2966.2009.14885.x , https://ui.adsabs.harvard.edu/abs/2009MNRAS.396.1815F 396, 1815

    Forero-Romero J. E., Hoffman Y., Gottl \"o ber S., Klypin A., Yepes G., 2009, @doi [ ] 10.1111/j.1365-2966.2009.14885.x , https://ui.adsabs.harvard.edu/abs/2009MNRAS.396.1815F 396, 1815

    2009

  4. [13]

    Goodfellow I. J., Pouget-Abadie J., Mirza M., Xu B., Warde-Farley D., Ozair S., Courville A., Bengio Y., 2014, @doi [arXiv e-prints] 10.48550/arXiv.1406.2661 , https://ui.adsabs.harvard.edu/abs/2014arXiv1406.2661G p. arXiv:1406.2661

    2014 doi arXiv

  5. [14]

    E., Peterson B

    Gunn J. E., Peterson B. A., 1965, @doi [ ] 10.1086/148444 , http://adsabs.harvard.edu/abs/1965ApJ...142.1633G 142, 1633

    1965 doi

  6. [16]

    P., et al., 2017, @doi [ ] 10.3847/1538-3881/aa859f , https://ui.adsabs.harvard.edu/abs/2017AJ....154..190H 154, 190

    Hearin A. P., et al., 2017, @doi [ ] 10.3847/1538-3881/aa859f , https://ui.adsabs.harvard.edu/abs/2017AJ....154..190H 154, 190

    2017 doi

  7. [17]

    https://openreview.net/forum?id=Sy2fzU9gl

    Higgins I., Matthey L., Pal A., Burgess C., Glorot X., Botvinick M., Mohamed S., Lerchner A., 2017, in International Conference on Learning Representations. https://openreview.net/forum?id=Sy2fzU9gl

    2017

  8. [18]

    arXiv:1903.09049

    Horowitz B., Lee K.-G., White M., Krolewski A., Ata M., 2019, arXiv e-prints, https://ui.adsabs.harvard.edu/abs/2019arXiv190309049H p. arXiv:1903.09049

    2019 arXiv

  9. [19]

    Horowitz B., Zhang B., Lee K.-G., Kooistra R., 2021, @doi [ ] 10.3847/1538-4357/abca35 , https://ui.adsabs.harvard.edu/abs/2021ApJ...906..110H 906, 110

    2021 doi

  10. [20]

    Horowitz B., Dornfest M., Luki \'c Z., Harrington P., 2022, @doi [ ] 10.3847/1538-4357/ac9ea7 , https://ui.adsabs.harvard.edu/abs/2022ApJ...941...42H 941, 42

    2022 doi

  11. [21]

    Jacobus C., Harrington P., Luki \'c Z., 2023, @doi [ ] 10.3847/1538-4357/acfcb5 , https://ui.adsabs.harvard.edu/abs/2023ApJ...958...21J 958, 21

    2023 doi

  12. [22]

    Japelj J., et al., 2019, @doi [ ] 10.1051/0004-6361/201936048 , https://ui.adsabs.harvard.edu/abs/2019A&A...632A..94J 632, A94

    2019 doi

  13. [24]

    P., Welling M., 2013b, @doi [arXiv e-prints] 10.48550/arXiv.1312.6114 , https://ui.adsabs.harvard.edu/abs/2013arXiv1312.6114K p

    Kingma D. P., Welling M., 2013b, @doi [arXiv e-prints] 10.48550/arXiv.1312.6114 , https://ui.adsabs.harvard.edu/abs/2013arXiv1312.6114K p. arXiv:1312.6114

    doi arXiv

  14. [25]

    Krolewski A., et al., 2018, @doi [ ] 10.3847/1538-4357/aac829 , http://adsabs.harvard.edu/abs/2018ApJ...861...60K 861, 60

    2018 doi

  15. [26]

    Lanusse F., Mandelbaum R., Ravanbakhsh S., Li C.-L., Freeman P., P \'o czos B., 2021, @doi [ ] 10.1093/mnras/stab1214 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.504.5543L 504, 5543

    2021 doi

  16. [27]

    Lee K.-G., et al., 2014, @doi [ ] 10.1088/2041-8205/795/1/L12 , https://ui.adsabs.harvard.edu/abs/2014ApJ...795L..12L 795, L12

    2014 doi

  17. [28]

    Lee K.-G., et al., 2018, @doi [ ] 10.3847/1538-4365/aace58 , http://adsabs.harvard.edu/abs/2018ApJS..237...31L 237, 31

    2018 doi

  18. [29]

    Lynds R., 1971, @doi [ ] 10.1086/180695 , http://adsabs.harvard.edu/abs/1971ApJ...164L..73L 164, L73

    1971 doi

  19. [30]

    arXiv:2403.05398

    Mainieri V., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2403.05398 , https://ui.adsabs.harvard.edu/abs/2024arXiv240305398M p. arXiv:2403.05398

    2024 doi arXiv

  20. [31]

    R., Pichon C., 2019, @doi [ ] 10.1093/mnras/stz2828 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.490.3633M 490, 3633

    Maitra S., Srianand R., Petitjean P., Rahmani H., Gaikwad P., Choudhury T. R., Pichon C., 2019, @doi [ ] 10.1093/mnras/stz2828 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.490.3633M 490, 3633

    2019 doi

  21. [32]

    Maitra S., Srianand R., Gaikwad P., 2022a, @doi [ ] 10.1093/mnras/stab3053 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.509.1536M 509, 1536

    doi

  22. [33]

    Maitra S., Srianand R., Gaikwad P., Khandai N., 2022b, @doi [ ] 10.1093/mnras/stab3308 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.509.4585M 509, 4585

    doi

  23. [34]

    Maitra S., Cristiani S., Viel M., Trotta R., Cupani G., 2024, @doi [ ] 10.1051/0004-6361/202450259 , https://ui.adsabs.harvard.edu/abs/2024A&A...690A.154M 690, A154

    2024 doi

  24. [35]

    McDonald P., Miralda-Escud \'e J., Rauch M., Sargent W. L. W., Barlow T. A., Cen R., Ostriker J. P., 2000, @doi [ ] 10.1086/317079 , http://adsabs.harvard.edu/abs/2000ApJ...543....1M 543, 1

    2000 doi

  25. [36]

    McDonald P., Miralda-Escud \'e J., Rauch M., Sargent W. L. W., Barlow T. A., Cen R., 2001, @doi [ ] 10.1086/323426 , http://adsabs.harvard.edu/abs/2001ApJ...562...52M 562, 52

    2001 doi

  26. [37]

    McDonald P., et al., 2006, @doi [ ] 10.1086/444361 , http://adsabs.harvard.edu/abs/2006ApJS..163...80M 163, 80

    2006 doi

  27. [38]

    McQuinn M., 2016, @doi [ ] 10.1146/annurev-astro-082214-122355 , http://adsabs.harvard.edu/abs/2016ARA

    2016 doi

  28. [39]

    A., 2009, @doi [Reviews of Modern Physics] 10.1103/RevModPhys.81.1405 , http://adsabs.harvard.edu/abs/2009RvMP...81.1405M 81, 1405

    Meiksin A. A., 2009, @doi [Reviews of Modern Physics] 10.1103/RevModPhys.81.1405 , http://adsabs.harvard.edu/abs/2009RvMP...81.1405M 81, 1405

    2009 doi

  29. [40]

    P., Rauch M., 1996, @doi [ ] 10.1086/177992 , http://adsabs.harvard.edu/abs/1996ApJ...471..582M 471, 582

    Miralda-Escud \'e J., Cen R., Ostriker J. P., Rauch M., 1996, @doi [ ] 10.1086/177992 , http://adsabs.harvard.edu/abs/1996ApJ...471..582M 471, 582

    1996 doi

  30. [41]

    arXiv:2506.08086

    Mishra S., Trotta R., Viel M., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2506.08086 , https://ui.adsabs.harvard.edu/abs/2025arXiv250608086M p. arXiv:2506.08086

    2025 doi arXiv

  31. [42]

    B., et al., 2020, @doi [ ] 10.3847/1538-4357/ab75ee , https://ui.adsabs.harvard.edu/abs/2020ApJ...891..147N 891, 147

    Newman A. B., et al., 2020, @doi [ ] 10.3847/1538-4357/ab75ee , https://ui.adsabs.harvard.edu/abs/2020ApJ...891..147N 891, 147

    2020 doi

  32. [43]

    Ntampaka M., et al., 2019, @doi [ ] 10.48550/arXiv.1902.10159 , https://ui.adsabs.harvard.edu/abs/2019BAAS...51c..14N 51, 14

    2019 doi arXiv

  33. [46]

    S., Weinberg D

    Peeples M. S., Weinberg D. H., Dav \'e R., Fardal M. A., Katz N., 2010b, @doi [ ] 10.1111/j.1365-2966.2010.16384.x , https://ui.adsabs.harvard.edu/abs/2010MNRAS.404.1295P 404, 1295

    2010

  34. [47]

    L., Rollinde E., Colombi S., Petitjean P., 2001, @doi [ ] 10.1046/j.1365-8711.2001.04595.x , http://adsabs.harvard.edu/abs/2001MNRAS.326..597P 326, 597

    Pichon C., Vergely J. L., Rollinde E., Colombi S., Petitjean P., 2001, @doi [ ] 10.1046/j.1365-8711.2001.04595.x , http://adsabs.harvard.edu/abs/2001MNRAS.326..597P 326, 597

    2001

  35. [48]

    Pieri M. M., et al., 2016, in Reyl \'e C., Richard J., Cambr \'e sy L., Deleuil M., P \'e contal E., Tresse L., Vauglin I., eds, SF2A-2016: Proceedings of the Annual meeting of the French Society of Astronomy and Astrophysics. pp 259--266 ( @eprint arXiv 1611.09388 )

    2016 arXiv

  36. [49]

    Planck Collaboration et al., 2016, @doi [ ] 10.1051/0004-6361/201525830 , http://adsabs.harvard.edu/abs/2016A

    2016 doi

  37. [50]

    G., Madau P., 2019, @doi [ ] 10.1093/mnras/stz222 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.485...47P 485, 47

    Puchwein E., Haardt F., Haehnelt M. G., Madau P., 2019, @doi [ ] 10.1093/mnras/stz222 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.485...47P 485, 47

    2019 doi

  38. [51]

    Puchwein E., et al., 2023, @doi [ ] 10.1093/mnras/stac3761 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.519.6162P 519, 6162

    2023 doi

  39. [52]

    Rauch M., 1998, @doi [ ] 10.1146/annurev.astro.36.1.267 , http://adsabs.harvard.edu/abs/1998ARA

    1998 doi

  40. [53]

    A., Steidel C

    Reddy N. A., Steidel C. C., 2009, @doi [ ] 10.1088/0004-637X/692/1/778 , https://ui.adsabs.harvard.edu/abs/2009ApJ...692..778R 692, 778

    2009 doi

  41. [54]

    Rigo M., Trotta R., Viel M., 2025, JERALD: high-fidelity dark matter, stellar mass and neutral hydrogen maps from fast N-body simulations, arXiv:2501.09168 [astro-ph.CO], @doi 10.48550/arXiv.2501.09168 , https://arxiv.org/abs/2501.09168

    2025 doi arXiv

  42. [55]

    arXiv:1505.04597

    Ronneberger O., Fischer P., Brox T., 2015, @doi [arXiv e-prints] 10.48550/arXiv.1505.04597 , https://ui.adsabs.harvard.edu/abs/2015arXiv150504597R p. arXiv:1505.04597

    2015 doi arXiv

  43. [56]

    Rorai A., et al., 2017, @doi [Science] 10.1126/science.aaf9346 , 356, 418

    2017 doi

  44. [57]

    F., Haehnelt M

    Rorai A., Carswell R. F., Haehnelt M. G., Becker G. D., Bolton J. S., Murphy M. T., 2018, @doi [ ] 10.1093/mnras/stx2862 , http://adsabs.harvard.edu/abs/2018MNRAS.474.2871R 474, 2871

    2018 doi

  45. [58]

    Schaye J., 2001, @doi [ ] 10.1086/322421 , http://adsabs.harvard.edu/abs/2001ApJ...559..507S 559, 507

    2001 doi

  46. [59]

    Schaye J., Theuns T., Rauch M., Efstathiou G., Sargent W. L. W., 2000, @doi [ ] 10.1046/j.1365-8711.2000.03815.x , http://adsabs.harvard.edu/abs/2000MNRAS.318..817S 318, 817

    2000

  47. [60]

    Seljak U., Slosar A., McDonald P., 2006, @doi [ ] 10.1088/1475-7516/2006/10/014 , http://adsabs.harvard.edu/abs/2006JCAP...10..014S 10, 014

    2006 doi

  48. [61]

    Springel V., 2005, @doi [ ] 10.1111/j.1365-2966.2005.09655.x , http://adsabs.harvard.edu/abs/2005MNRAS.364.1105S 364, 1105

    2005

  49. [62]

    W., White M., Lee K.-G., Hennawi J

    Stark C. W., White M., Lee K.-G., Hennawi J. F., 2015, @doi [ ] 10.1093/mnras/stv1620 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.453..311S 453, 311

    2015 doi

  50. [63]

    Subaru Prime Focus Spectrograph Collaboration 2024, Instrument Parameters -- Prime Focus Spectrograph (PFS), https://pfs.ipmu.jp/research/parameters.html

    2024

  51. [64]

    Takada M., et al., 2014, @doi [ ] 10.1093/pasj/pst019 , https://ui.adsabs.harvard.edu/abs/2014PASJ...66R...1T 66, R1

    2014 doi

  52. [65]

    J., Theuns T., Haehnelt M

    Viel M., Matarrese S., Mo H. J., Theuns T., Haehnelt M. G., 2002, @doi [ ] 10.1046/j.1365-8711.2002.05803.x , http://adsabs.harvard.edu/abs/2002MNRAS.336..685V 336, 685

    2002

  53. [66]

    G., Kim T

    Viel M., Matarrese S., Heavens A., Haehnelt M. G., Kim T. S., Springel V., Hernquist L., 2004, @doi [ ] 10.1111/j.1365-2966.2004.07404.x , https://ui.adsabs.harvard.edu/abs/2004MNRAS.347L..26V 347, L26

    2004

  54. [67]

    M., 2013, @doi [ ] 10.1093/mnras/sts465 , http://adsabs.harvard.edu/abs/2013MNRAS.429.1734V 429, 1734

    Viel M., Schaye J., Booth C. M., 2013, @doi [ ] 10.1093/mnras/sts465 , http://adsabs.harvard.edu/abs/2013MNRAS.429.1734V 429, 1734

    2013 doi

  55. [68]

    Villaescusa-Navarro F., 2018, Pylians: Python libraries for the analysis of numerical simulations , Astrophysics Source Code Library, record ascl:1811.008 ( @eprint ascl 1811.008 )

    2018

  56. [69]

    Villaescusa-Navarro F., et al., 2021, @doi [ ] 10.3847/1538-4357/abf7ba , https://ui.adsabs.harvard.edu/abs/2021ApJ...915...71V 915, 71

    2021 doi

  57. [70]

    S., 2006, @doi [ ] 10.1111/j.1365-2966.2006.10333.x , https://ui.adsabs.harvard.edu/abs/2006MNRAS.369..734Z 369, 734

    Zaroubi S., Viel M., Nusser A., Haehnelt M., Kim T. S., 2006, @doi [ ] 10.1111/j.1365-2966.2006.10333.x , https://ui.adsabs.harvard.edu/abs/2006MNRAS.369..734Z 369, 734

    2006

  58. [71]

    L., Zehavi I., 2007, @doi [ ] 10.1086/521074 , https://ui.adsabs.harvard.edu/abs/2007ApJ...667..760Z 667, 760

    Zheng Z., Coil A. L., Zehavi I., 2007, @doi [ ] 10.1086/521074 , https://ui.adsabs.harvard.edu/abs/2007ApJ...667..760Z 667, 760

    2007 doi

  59. [72]

    S., Brox T., Ronneberger O., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1606.06650 , https://ui.adsabs.harvard.edu/abs/2016arXiv160606650C p

    C i c ek \"O ., Abdulkadir A., Lienkamp S. S., Brox T., Ronneberger O., 2016, @doi [arXiv e-prints] 10.48550/arXiv.1606.06650 , https://ui.adsabs.harvard.edu/abs/2016arXiv160606650C p. arXiv:1606.06650

    2016 doi arXiv

  60. [73]

    write newline

    " write newline "" before.all 'output.state := FUNCTION fin.entry write newline FUNCTION new.block output.state before.all = 'skip after.block 'output.state := if FUNCTION new.sentence output.state after.block = 'skip output.state before.all = 'skip after.sentence 'output.stat...

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.