Pith. sign in

REVIEW 3 major objections 6 minor 51 references

zELDA II: reconstruction of galactic Lyman-alpha spectra attenuated by the intergalactic medium using neural networks

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A neural network can peel the intergalactic medium off Lyman-alpha spectra, recovering the galaxy-emitted line.

desk verdict A solid mock-based method paper whose headline accuracy numbers are properties of the mock generator, not yet the sky; it deserves a serious referee but needs a toned-down abstract or external validation. read the letter →

arxiv 2501.04077 v1 pith:2OZOWJPF submitted 2025-01-07 astro-ph.GA astro-ph.CO

Siddhartha Gurung-Lopez , Chris Byrohl , Max Gronke , Daniele Spinoso , Alberto Torralba , Alberto Fernandez-Soto , Pablo Arnalte-Mur , Vicent J. Martinez This is my paper · ORCID
classification astro-ph.GAastro-ph.CO
keywords Lyman-alphaemissionintergalacticmediumradiativetransferneuralnetworksshellmodelescapefractionspectralreconstructionzELDA
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

This paper presents zELDA II, a machine-learning tool that tries to separate the two components that shape an observed Lyman-alpha emission line: the line profile produced by radiative transfer inside the galaxy (the ISM) and the absorption and scattering imprint of the intergalactic medium (IGM). The authors build millions of mock spectra by convolving Monte-Carlo 'shell model' galaxy spectra with IGM transmission curves from the IllustrisTNG100 simulation, then train neural networks to invert the process. They report that the networks recover the galaxy-emerging line shape with a Kolmogorov-Smirnov statistic below 0.1 for 95% of HST COS-like and 80% of MUSE-WIDE-like mock spectra, and measure the IGM transmission with typical uncertainties below 10%. This matters because a clean separation would let Lyman-alpha serve as a high-redshift probe of galaxy outflows and of how the IGM modulates galaxy visibility.

What carries the argument

Two precomputed ingredients are combined into training mocks: (i) the 'shell model' grid from the LyaRT Monte Carlo radiative transfer code, a 5D grid over outflow velocity, neutral hydrogen column density, dust optical depth, intrinsic equivalent width, and intrinsic line width that supplies the galaxy-emerging spectrum; and (ii) IGM transmission curves computed from the IllustrisTNG100 simulation, rebinned to continuous redshift by rescaling each snapshot to the mean optical depth of Faucher-Giguère et al. (2008). The networks receive the first 100 PCA components of the observed spectrum plus spectral resolution, pixel size, and (in IGM+z) a proxy redshift; one model (IGM-z) deliberately drops redshift and randomizes the IGM sightlines to avoid imprinting a redshift evolution. The output is the five shell parameters, the true Lyman-alpha wavelength offset, and the IGM Lyman-alpha escape fraction in wavelength windows around line center.

What would settle it

Build a validation set with mock spectra from a different radiative-transfer simulation, such as an ISM model that is not a thin shell or IGM sightlines from a different cosmological simulation, and run zELDA's trained networks on it; if the KS$<0.1$ success fractions or the below-10% transmission uncertainties drop substantially, the in-sample calibration is the cause.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that an observed Lyman-$\alpha$ spectrum contains enough information to recover the pre-IGM ('ISM-emerging') line profile and the line-of-sight IGM escape fraction, provided the training data cover the physical variety of both media. Using a 5D grid of thin-shell outflow parameters for the galaxy and 1000 sightline transmission curves per halo from IllustrisTNG100, the authors train three neural networks; the two that include IGM attenuation (IGM+z and IGM-z) reconstruct the intrinsic line profile with KS$<0.1$ in 95% of COS-like and 80% of MUSE-WIDE-like validation cases, and recover $f^{4\,\text{Å}}_{\mathrm{esc}}$ with typical scatter of about 0.03 for HST-like and 0.12 for MUSE-like spectra. They further show that stacked reconstructed profiles track the true evolution or non-evolution of the intrinsic ISM line with redshift, while a model trained without IGM absorption badly misses the blue peak at high redshift.

Load-bearing premise

Every validation spectrum is made by the same two generators used in training: the LyaRT thin-shell galaxy grid and the IllustrisTNG100 IGM transmission curves, so the accuracy numbers presuppose that these simulations resemble the real ISM and IGM along actual lines of sight.

Editorial extensions

If this is right

  • For HST COS-like spectra, zELDA can hand back the intrinsic pre-IGM Lyman-alpha profile for the large majority of sources, enabling studies of ISM outflow properties at redshifts where the IGM previously obscured them.
  • Per-source IGM escape fractions allow observers to build Lyman-alpha luminosity functions corrected for IGM attenuation, and to test whether Lyman-alpha visibility depends on large-scale IGM density and velocity fields.
  • Reconstructed stacks of ISM-emerging profiles can distinguish true redshift evolution of the galaxy-emerging line from apparent evolution caused by IGM absorption.
  • The IGM-z model, designed to be redshift-unbiased, can measure the redshift evolution of the mean IGM escape fraction from z about 2 onward without imposing the training-set redshift dependence.
  • For strongly absorbed lines with $f^{4\,\text{Å}}_{\mathrm{esc}}\lesssim0.4$ the reconstruction degrades, so the claimed accuracy applies to the regime where the blue side is not completely erased.

Reading between the lines

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

  • The headline accuracy numbers are measured on validation mocks built with the same forward models used to train the networks, so they quantify in-sample inversion performance rather than guaranteed performance on real observed spectra.
  • A direct observational test would be to compare zELDA's per-source escape fractions with IGM transmission measured independently along the same sightlines, for example using background quasars or close pairs of galaxies.
  • If real ISM geometries deviate from the thin-shell model, the recovered intrinsic profiles could be the best shell-model projection rather than the true spectrum; retraining or validating on non-shell radiative transfer outputs would reveal the size of this effect.
  • The same PCA-plus-neural-network scheme could be adapted to other resonant lines or to jointly fitting Lyman-alpha with UV continuum information to break remaining degeneracies between outflow parameters and IGM absorption.
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

3 major / 6 minor

Summary. The paper presents zELDA II, an open-source Python module that uses artificial neural networks to disentangle the interstellar medium (ISM) and intergalactic medium (IGM) contributions to observed Lyα spectra. Mock spectra are generated by convolving LyaRT 'thin shell' profiles (the zELDA I forward model) with IGM transmission curves from Byrohl & Gronke (2020) (IllustrisTNG100), downgraded to various spectral resolutions and signal-to-noise ratios. Three network variants are trained: IGM+z (input includes proxy redshift, IGM curves assigned at the source redshift), IGM-z (no redshift input, randomized IGM curves), and NoIGM (no IGM attenuation during training). The networks output shell parameters, redshift offset, and IGM escape fractions f_xÅ_esc. The paper reports that on mock spectra, IGM+z and IGM-z reconstruct intrinsic profiles with KS<0.1 for 95% (HST/COS-like) and ~80% (MUSE-WIDE-like) of cases, and measure f^4Å_esc with 'typical uncertainties below 10%' (abstract) or ~0.12 (Sect. 5). The paper also tests recovery of prescribed mean f_esc evolutions and stacked line profiles. Public code and documentation are provided.

Significance. If the quoted accuracies were validated on independent data, the method would be a valuable tool for separating ISM and IGM effects in Lyα spectroscopy and for deriving IGM escape fractions source-by-source. The paper strengthens the earlier zELDA framework with a PCA-based input representation and several carefully compared network models, and it provides reproducible code, extensive accuracy tables (Appendices C-D), and an honest uncertainty-calibration analysis (Appendix B). The main caveat is that all quantitative performance claims are measured on mock spectra generated by the same forward model used to construct the training set; the reported 95%/80% success fractions and sub-10% f_esc uncertainties are therefore properties of the mock generator rather than demonstrated properties of real observations. The abstract does not make this limitation clear.

major comments (3)
  1. [Sects. 2.3, 3.2, 4; Appendices C-D] The validation is in-sample. The training set (Sect. 3.2) and the mock validation set (Sect. 4; Appendices C-D) are generated by the same forward model: LyaRT thin-shell profiles (Sect. 2.1) convolved with Byrohl & Gronke (2020) IllustrisTNG100 IGM transmission curves (Sect. 2.2), then degraded to the same observational configurations. The networks therefore learn and are tested on the same distribution. The paper states (Sect. 5) that 'we have tested our ANN models in mock Lyα line profiles' and then reports the 95%/80% KS<0.1 fractions and ~0.03/~0.12 f_esc uncertainties as if they apply to observed data, but no test on observed spectra or on an independent forward model is presented. This is load-bearing for the central claim that zELDA can reconstruct ISM-emerging Lyα profiles of observed galaxies. I request either a demonstration on real data (even a small pilot sample of COS/MUSE spectra), a validation against a forward model not used in training (e.g., an alternative IGM simulation or a non-shell ISM geometry), or an explicit qualification in the abstract and conclusions that these accuracy numbers are in-sample mock validation results. The 'redshift-unbiased' design of IGM-z does not remove this dependence, because both training and test IGM curves are drawn from the same IllustrisTNG100-based set.
  2. [Abstract; Sect. 5; Fig. C.2] The abstract's claim that zELDA measures the IGM transmission 'with typical uncertainties below 10% for HST-COS and MUSE-WIDE data' is not supported by the body. Section 5 reports f^4Å_esc uncertainties of ~0.03 for HST-like and ~0.12 for MUSE-like data, and Fig. C.2 (IGM-z, f^4Å_esc bins) shows accuracies of 0.08-0.16 for Wg=2.0 Å (MUSE-like) across the f_esc range, with values above 0.10 for most f_esc bins. Please reconcile the abstract with the tabulated values and avoid stating 'below 10%' for MUSE-WIDE-like data unless a stricter subset (e.g., only f_esc>0.8) is explicitly defined.
  3. [Sect. 4.2.1, Eq. (3)] The f_esc evolution tests are constructed by prescribing the mean f_esc as a Fermi-Dirac function of redshift and then drawing IGM transmission curves until the computed f_esc lies within 10% of the prescribed ⟨f_esc⟩. Recovering this prescribed trend with the network shows that the network can invert the training distribution, but it does not independently verify the ability to measure the true IGM transmission evolution. Statements such as 'IGM+z and IGM-z are able to detect evolution in f_esc from redshift 2.0 onward for MUSE-like data' (Sect. 5) should be explicitly labeled as tests on mocks with injected evolution; as written, they overstate the evidence.
minor comments (6)
  1. [Throughout] The spelling 'Kolmogórov-Smirnov' appears in several places (e.g., abstract, Sect. 4.1, Appendix D); the standard spelling is 'Kolmogorov-Smirnov'.
  2. [Fig. 4 caption] The caption says 'Ly α line profiles spamming zELDA's grid'; 'spamming' should be 'spanning'.
  3. [References] The entries Gurung-López et al. 2021a and 2021b share the same volume/page (MNRAS, 500, 603); please check whether one of these is a different article (and cite accordingly in the text).
  4. [Sect. 3.4] The sentence 'For each output property, we trained an independent ANN' is clear, but the training set sizes and the convergence test ('We tested that for this training set size, our artificial neural networks have converged') are not quantified; adding a brief description (e.g., loss curves or a convergence criterion) would improve reproducibility.
  5. [Sect. 2.1] The parameter range for τ_a is printed as 'τa∈ [0.0001, 0.0]' in the text; this appears to be a typo, since dust optical depth τ_a=0.0 would make the lower bound meaningless. Please verify and correct.
  6. [Fig. D.1] The x-axis label 'log KS' is ambiguous: it is not clear whether the base is 10 and whether KS is the standard Kolmogorov-Smirnov statistic; please state 'log10 KS'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the accuracy claims are measured on held-out mocks from the same forward model used for training, which is a limitation on external validity but not a logically circular derivation.

full rationale

The claimed derivation chain is: generate mock Lyα spectra by convolving LyaRT thin-shell profiles (Sect. 2.1, from the authors' ZP22 grid) with Byrohl & Gronke (2020) IllustrisTNG100 IGM transmission curves (Sect. 2.2), train artificial neural networks to map those mock observed spectra to shell parameters, redshift, and f_esc (Sect. 3), and then evaluate on held-out mocks produced by the same generator (Sect. 4 and Appendices C-D). This is an in-sample evaluation: the reported KS and f_esc accuracies quantify how well the networks invert their own training forward model, and they do not by themselves establish performance on real spectra if the true ISM or IGM departs from the thin-shell model or from IllustrisTNG100. That is a genuine external-validity limitation, and the paper is largely transparent that all accuracy results are on mock line profiles. It is not a circular derivation under the seven enumerated patterns: the validation mocks are not used to set the network weights, the networks demonstrably fail on identifiable subsets (e.g., low f_esc, and NoIGM at high redshift), and the f_esc-evolution tests in Sect. 4.2 are injection-recovery checks where the truth is imposed by a Fermi-Dirac rejection procedure and then independently estimated, with reported biases rather than tautological agreement. The self-citations (ZP22 and Byrohl & Gronke 2020) supply the forward model and are externally published, based on public codes and simulations, and falsifiable against Lyα observations; they are not invoked as uniqueness theorems or to forbid alternatives. One non-circular inconsistency exists: the abstract states typical MUSE-WIDE IGM transmission uncertainties below 10%, while Sect. 5 and Appendix C report about 0.12 in the MUSE-like regime; this affects the consistency of the claims, not the logical structure of the derivation. No circular step meets the evidence bar.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

No new physical entities are introduced. The central claim rests on the shell-model ISM description, the IllustrisTNG100-based IGM transmission curves, and the ANN's ability to generalize from that forward model to real data; none of these is independently benchmarked outside the same forward model.

free parameters (6)
  • KS success threshold = 0.1
    Chosen to define successful reconstruction; the reported success fractions depend on this threshold.
  • Number of PCA components = 100
    Chosen because explained variance saturates near 95% at 100 components; tested alternatives (200, 400) gave no improvement.
  • ANN architecture = Three-layer (103, 53, 25) for most outputs; nine-layer for Delta_lambda_True
    Best configuration found by testing different architectures (Sect. 3.4).
  • Training set size = 4.5e6
    Chosen so that the ANNs converge; no validation-set separation is described.
  • f_esc wavelength window = 4 Angstrom
    The IGM escape fraction is measured in a window around Ly-alpha; authors find values converge for windows above 4 Angstrom, so the window is partly motivated by the data.
  • Fermi-Dirac mock parameters a and b = Mock1 {7.0, 0.7}, Mock2 {6.0, 0.6}, Mock3 {5.0, 0.5}, Mock4 {4.0, 0.4}
    Chosen by hand to populate the mean f_esc(z) plane; used only to construct test mocks, not fitted to data.
assumptions (7)
  • domain assumption Thin shell model with parameters Vexp, NH, tau_a, EWin, Win describes the ISM-emerging Ly-alpha profile.
    Used to generate all intrinsic line profiles (Sect. 2.1); if real ISM geometry differs, the inferred parameters and reconstructed profiles are biased.
  • domain assumption Byrohl & Gronke (2020) IGM transmission curves from IllustrisTNG100 span the real diversity of IGM attenuation.
    Sect. 2.2; all training and test spectra use these curves.
  • domain assumption Faucher-Giguere et al. (2008) mean optical depth evolution is correct and can rescale snapshot curves.
    Used in Sect. 2.2 to recalibrate IGM transmission for continuous redshift sampling.
  • domain assumption Observed Ly-alpha spectrum = intrinsic profile multiplied by IGM transmission, then convolved with instrument Gaussian and pixelated, plus Gaussian noise.
    Forward model in Sect. 2.3; deviations such as scattering into the line of sight or continuum errors are not modeled.
  • ad hoc to paper ANNs trained on this forward model generalize to real observed Ly-alpha spectra.
    No real-data validation is presented; this assumption is required for the application claims in Sect. 5.
  • domain assumption The 100 PCA components retain the information needed for parameter inference.
    Chosen in Sect. 3.1.1 based on explained variance; no information-theoretic guarantee.
  • domain assumption The observed global maximum wavelength is a usable redshift proxy.
    Input construction in Sect. 3.1.1; can fail for strongly absorbed or double-peaked lines.

how reviews work

0 comments
Cite this review

Pith. "Pith review of zELDA II: reconstruction of galactic Lyman-alpha spectra attenuated by the intergalactic medium using neural networks." pith.science (2026). https://pith.science/paper/2OZOWJPF

@misc{pith2026250104077,
  author       = {Pith},
  title        = {Pith review of: zELDA II: reconstruction of galactic Lyman-alpha spectra attenuated by the intergalactic medium using neural networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2OZOWJPF}},
  note         = {Machine review of arXiv:2501.04077}
}
read the original abstract

The observed Lyman-Alpha (Lya) line profile is a convolution of the complex Lya radiative transfer taking place in the interstellar, circumgalactic and intergalactic medium (ISM, CGM, and IGM, respectively). Discerning the different components of the Lya line is crucial in order to use it as a probe of galaxy formation or the evolution of the IGM. We present the second version of zELDA (redshift Estimator for Line profiles of Distant Lyman-Alpha emitters), an open-source Python module focused on modeling and fitting observed Lya line profiles. This new version of zELDA focuses on disentangling the galactic from the IGM effects. We build realistic Lya line profiles that include the ISM and IGM contributions, by combining the Monte Carlo radiative transfer simulations for the so called "shell model" (ISM) and IGM transmission curves generated from IllustrisTNG100. We use these mock line profiles to train different artificial neural networks. These use as input the observed spectrum and output the outflow parameters of the best fitting "shell model" along with the redshift and Lya emission IGM escape fraction of the source. We measure the accuracy of zELDA on mock Lya line profiles. We find that zELDA is capable of reconstructing the ISM emerging Lya line profile with high accuracy (Kolmogorov-Smirnov<0.1) for 95% of the cases for HST COS-like observations and 80% for MUSE-WIDE-like. zELDA is able to measure the IGM transmission with the typical uncertainties below 10% for HST-COS and MUSE-WIDE data. This work represents a step forward in the high-precision reconstruction of IGM attenuated Lya line profiles. zELDA allows the disentanglement of the galactic and IGM contribution shaping the Lya line shape, and thus allows us to use Lya as a tool to study galaxy and ISM evolution.

Figures

Figures reproduced from arXiv: 2501.04077 by the authors.

Figure 1
Figure 1. Illustration of the impact of different line of sights in the same intrinsic spectrum. In the middle bottom panel we show the intrinsic spectrum escaping the source (black) convolved with the mean IGM transmission at z = 3.0 (yellow). The color line shows the convolution of the intrinsic spectrum and IGM transmission. The colored circles show zELDA reconstruction using the IGM-z model (discussed later). In the other… view at source ↗
Figure 1
Figure 1. 2.3. Mocking observed Lyman-α line profiles Lyα line profiles predicted by zELDA using the LyaRT grid of line profiles and the IGM transmission curves of Byrohl & Gronke (2020) are ideal, both in terms of spectral resolution and signal￾to-noise ratio. In contrast, measurements of Lyα line profiles present limitations in the spectral resolution, spectral binning and signal to noise. In order to produce a mock line pr… view at source ↗
Figure 2
Figure 2. Mean IGM transmission curves without recalibration (left) and after recalibration (right). Each color shows a different redshift snapshot. The horizontal black dashed show the mean IGM transmission given by (Faucher-Giguère et al. 2008) at z = 0,1,2,3,4,5 from top to bottom. transmission curve is applied to the line profiles of the training set. This section is structured as follows. First, we detail the ANN input i… view at source ↗
Figures from the paper (9 more)
Figure 3
Figure 3. Figure 3: Example of line profile reconstruction at different line profile qualities and using our different models. The true Lyα line before passing through the IGM is displayed in red. The IGM transmission curve is shown in pink. The true Lyα line profile is fixed in each row.…
Figure 4
Figure 4. Figure 4: Left: Total variance recovered as a function of the number of principle components. In red (green), Lyα line profiles spamming zELDA’s grid without (with) IGM absorption. The dashed black line mark the number of principal components used for the input of the artificial…
Figure 5
Figure 5. Figure 5: f 4Å esc used for the training in the IGM-z (left) and IGM+z (right) models. The solid thick line marks the median f 4Å esc . Meanwhile, the shaded regions show the scatter between the 16 and 84 percentiles (darkest), 5 and 95 percentiles (medium dark), 1 and 99 percen…
Figure 6
Figure 6. Figure 6: Examples of line profile successful reconstruction at different redshift. The redshift of the mock line profile is in the top right corner of each subpanel. The Lyα line after the ISM and before traveling through the IGM is shown in red. The IGM transmission curve is s…
Figure 7
Figure 7. Figure 7: Same as [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]
Figure 8
Figure 8. Figure 8: zELDA’s prediction on the mean f 4Å esc for different mock Lyα line profiles using the IGM+z model. Each panel shows a different observation quality. The left and right column display S/Np =10.0 and S/Np =15.0, respectively. Each row has a constant Wg. In particular, W…
Figure 9
Figure 9. Figure 9: Same as [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]
Figure 10
Figure 10. Figure 10: Stacked line profile reconstruction example in mock Lyα line profiles using the Lyα line profiles with redshift dependence. The stacked line profile is shown the redshift intervals [0.75,1.25] (grey), [1.75,2.25] (blue), [2.75,3.25] (green), [3.75,4.25] (yellow) and […
Figure 11
Figure 11. Figure 11: Same as [PITH_FULL_IMAGE:figures/full_fig_p015_11.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 37 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archiveprefix author booktitle chapter edition editor howpublished institution eprint journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1 ...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in " " * FUNCTION format....

  3. [3]

    2003, Journal of Korean Astronomical Society, 36, 145

    Ahn , S. 2003, Journal of Korean Astronomical Society, 36, 145

    work page 2003

  4. [4]

    2019, , 486, 2197

    Behrens , C., Pallottini , A., Ferrara , A., Gallerani , S., & Vallini , L. 2019, , 486, 2197

    work page 2019

  5. [5]

    & Gronke , M

    Byrohl , C. & Gronke , M. 2020, , 642, L16

    work page 2020

  6. [6]

    2021, , 506, 5129

    Byrohl , C., Nelson , D., Behrens , C., et al. 2021, , 506, 5129

    work page 2021

  7. [7]

    2019, , 489, 3472

    Byrohl , C., Saito , S., & Behrens , C. 2019, , 489, 3472

    work page 2019

  8. [8]

    C., et al

    Caruana , J., Wisotzki , L., Herenz , E. C., et al. 2020, VizieR Online Data Catalog, J/MNRAS/473/30

    work page 2020

Show all 51 references
  1. [9]

    2006, , 649, 14

    Dijkstra , M., Haiman , Z., & Spaans , M. 2006, , 649, 14

    2006

  2. [10]

    J., S \'a nchez , A

    Farrow , D. J., S \'a nchez , A. G., Ciardullo , R., et al. 2021, arXiv e-prints, arXiv:2104.04613

    2021 arXiv

  3. [11]

    X., Lidz , A., Hernquist , L., & Zaldarriaga , M

    Faucher-Gigu \`e re , C.-A., Prochaska , J. X., Lidz , A., Hernquist , L., & Zaldarriaga , M. 2008, , 681, 831

    2008

  4. [12]

    C., Froning , C

    Green , J. C., Froning , C. S., Osterman , S., et al. 2012, , 744, 60

    2012

  5. [13]

    2022, arXiv e-prints, arXiv:2206.14908

    Greene , J., Bezanson , R., Ouchi , M., Silverman , J., & the PFS Galaxy Evolution Working Group . 2022, arXiv e-prints, arXiv:2206.14908

    2022 arXiv

  6. [14]

    2017, , 608, A139

    Gronke , M. 2017, , 608, A139

    2017

  7. [15]

    Gronke , M., Dijkstra , M., McCourt , M., & Oh , S. P. 2016, , 833, L26

    2016

  8. [16]

    Gurung-L \'o pez , S., Gronke , M., Saito , S., Bonoli , S., & Orsi , \'A . A. 2022, , 510, 4525

    2022

  9. [17]

    A., & Bonoli , S

    Gurung-L \'o pez , S., Orsi , \'A . A., & Bonoli , S. 2019 a , , 490, 733

    2019

  10. [18]

    A., Bonoli , S., Baugh , C

    Gurung-L \'o pez , S., Orsi , \'A . A., Bonoli , S., Baugh , C. M., & Lacey , C. G. 2019 b , , 486, 1882

    2019

  11. [19]

    A., Bonoli , S., et al

    Gurung-L \'o pez , S., Orsi , \'A . A., Bonoli , S., et al. 2020, , 491, 3266

    2020

  12. [20]

    M., et al

    Gurung-L \'o pez , S., Saito , S., Baugh , C. M., et al. 2021 a , , 500, 603

    2021

  13. [21]

    M., et al

    Gurung-L \'o pez , S., Saito , S., Baugh , C. M., et al. 2021 b , , 500, 603

    2021

  14. [22]

    R., Millman, K

    Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357

    2020

  15. [23]

    J., Runnholm , A., Scarlata , C., Gronke , M., & Rivera-Thorsen , T

    Hayes , M. J., Runnholm , A., Scarlata , C., Gronke , M., & Rivera-Thorsen , T. E. 2023, , 520, 5903

    2023

  16. [24]

    C., Urrutia , T., Wisotzki , L., et al

    Herenz , E. C., Urrutia , T., Wisotzki , L., et al. 2017, , 606, A12

    2017

  17. [25]

    J., Gebhardt , K., Komatsu , E., et al

    Hill , G. J., Gebhardt , K., Komatsu , E., et al. 2008, in Astronomical Society of the Pacific Conference Series, Vol. 399, Astronomical Society of the Pacific Conference Series, ed. T. Kodama, T. Yamada, & K. Aoki , 115--+

    2008

  18. [26]

    Hunter, J. D. 2007, Computing In Science & Engineering, 9, 90

    2007

  19. [27]

    2019, arXiv e-prints, arXiv:1906.00173

    Kakuma , R., Ouchi , M., Harikane , Y., et al. 2019, arXiv e-prints, arXiv:1906.00173

    2019 arXiv

  20. [28]

    Laursen , P., Sommer-Larsen , J., & Razoumov , A. O. 2011, , 728, 52

    2011

  21. [29]

    2018, , 480, 5113

    Marinacci , F., Vogelsberger , M., Pakmor , R., et al. 2018, , 480, 5113

    2018

  22. [30]

    P., Pillepich , A., Springel , V., et al

    Naiman , J. P., Pillepich , A., Springel , V., et al. 2018, , 477, 1206

    2018

  23. [31]

    2019, Computational Astrophysics and Cosmology, 6, 2

    Nelson , D., Springel , V., Pillepich , A., et al. 2019, Computational Astrophysics and Cosmology, 6, 2

    2019

  24. [32]

    Neufeld , D. A. 1990, , 350, 216

    1990

  25. [33]

    G., & Baugh , C

    Orsi , A., Lacey , C. G., & Baugh , C. M. 2012, , 425, 87

    2012

  26. [34]

    2014, , 443, 799

    Orsi , \'A ., Padilla , N., Groves , B., et al. 2014, , 443, 799

    2014

  27. [35]

    2018, , 70, S13

    Ouchi , M., Harikane , Y., Shibuya , T., et al. 2018, , 70, S13

    2018

  28. [36]

    2020, Annual Review of Astronomy and Astrophysics, 58, 617

    Ouchi, M., Ono, Y., & Shibuya, T. 2020, Annual Review of Astronomy and Astrophysics, 58, 617

    2020

  29. [37]

    2018, , 475, 648

    Pillepich , A., Nelson , D., Hernquist , L., et al. 2018, , 475, 648

    2018

  30. [38]

    C., Steidel , C

    Rudie , G. C., Steidel , C. C., & Pettini , M. 2012, , 757, L30

    2012

  31. [39]

    2021, , 133, 034507

    Runnholm , A., Gronke , M., & Hayes , M. 2021, , 133, 034507

    2021

  32. [40]

    2011, , 531, A12

    Schaerer , D., Hayes , M., Verhamme , A., & Teyssier , R. 2011, , 531, A12

    2011

  33. [41]

    2020, , 643, A149

    Spinoso , D., Orsi , A., L \'o pez-Sanjuan , C., et al. 2020, , 643, A149

    2020

  34. [42]

    2018, , 475, 676

    Springel , V., Pakmor , R., Pillepich , A., et al. 2018, , 475, 676

    2018

  35. [43]

    C., Erb , D

    Steidel , C. C., Erb , D. K., Shapley , A. E., et al. 2010, , 717, 289

    2010

  36. [44]

    2023, , 680, A14

    Torralba-Torregrosa , A., Gurung-L \'o pez , S., Arnalte-Mur , P., et al. 2023, , 680, A14

    2023

  37. [45]

    2024, , 690, A388

    Torralba-Torregrosa , A., Renard , P., Spinoso , D., et al. 2024, , 690, A388

    2024

  38. [46]

    2019, , 624, A141

    Urrutia , T., Wisotzki , L., Kerutt , J., et al. 2019, , 624, A141

    2019

  39. [47]

    2018, , 478, L60

    Verhamme , A., Garel , T., Ventou , E., et al. 2018, , 478, L60

    2018

  40. [48]

    2007, in Astronomical Society of the Pacific Conference Series, Vol

    Verhamme , A., Schaerer , D., Atek , H., & Tapken , C. 2007, in Astronomical Society of the Pacific Conference Series, Vol. 380, Deepest Astronomical Surveys, ed. J. Afonso , H. C. Ferguson , B. Mobasher , & R. Norris , 97

    2007

  41. [49]

    E., et al

    Virtanen , P., Gommers , R., Oliphant , T. E., et al. 2020, Nature Methods, 17, 261

    2020

  42. [50]

    H., Bowman , W

    Weiss , L. H., Bowman , W. P., Ciardullo , R., et al. 2021, , 912, 100

    2021

  43. [51]

    2011, , 726, 38

    Zheng , Z., Cen , R., Trac , H., & Miralda-Escud \'e , J. 2011, , 726, 38

    2011

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.