Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Field-Level Comparison and Robustness Analysis of Cosmological N-body Simulations

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Field-level cosmic-web inference trained on one N-body code transfers well to other fixed-resolution codes but fails on adaptive-mesh-refinement codes, and a six-grid-spacing smoothing filter restores unbiased parameter estimates.

desk verdict Solid field-level N-body comparison with a practical message: CNN inference transfers across non-AMR codes but breaks on AMR codes, and smoothing fixes it; the causal resolution story and the exact 6ΔR scale are not fully nailed down. read the letter →

arxiv 2505.13620 v2 pith:NHG2FDIY submitted 2025-05-19 astro-ph.CO

Adrian E. Bayer , Francisco Villaescusa-Navarro , Sammy Sharief , Romain Teyssier , Lehman H. Garrison , Laurence Perreault-Levasseur , Greg L. Bryan , Marco Gatti
show 1 more author
Eli Visbal
This is my paper · ORCID
classification astro-ph.CO
keywords cosmologicalN-bodysimulationsfield-levelinferencesimulation-basedout-of-distributiondetectionadaptivemeshrefinementconvolutionalneuralnetworksPQMassparameterrobustness
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 asks whether a neural network that reads the full dark-matter density field can be trusted when it is trained on simulations from one N-body code and applied to simulations from another. The answer is mostly yes for fixed-resolution codes, but no for adaptive-mesh-refinement (AMR) codes: a network trained on Gadget stays nearly unbiased on Abacus, CUBEP$^3$M, PKDGrav, and Gizmo, yet infers $\Omega_m$ and $\sigma_8$ with errors above twenty percent on Ramses and Enzo, a bigger failure than testing on a full hydrodynamic simulation. The paper traces this to effective resolution: AMR codes leave voids and filaments smoother, and the network learns small-scale fluctuations in those low-density pixels. Smoothing the maps with a Gaussian filter of about six grid spacings (~$0.6$ Mpc$/h$) brings all seven codes and the hydrodynamic simulation into statistical agreement and restores unbiased inference, at the cost of wider error bars. The practical consequence is that field-level cosmological inference should filter out scales a simulation cannot resolve and should use field-level out-of-distribution checks before being applied to survey data.

What carries the argument

The machinery has three parts. First, the simulations: seven gravity-only N-body codes (Gadget, Abacus, CUBEP$^3$M, Enzo, Gizmo, PKDGrav, Ramses) with matched initial seeds and cosmologies, plus one hydrodynamic simulation (IllustrisTNG) used as a stress test; Gadget supplies 1000 Latin Hypercube training boxes and the other codes supply about 50 test boxes each. Second, the inference engine: a six-block convolutional neural network with circular padding that maps $256\times256$ matter-overdensity images to predicted marginal posterior means and standard deviations for $\Omega_m$ and $\sigma_8$, trained with a loss that scores both the mean and the variance. Third, the diagnosis: PQMass, a Voronoi-based $\chi^2$ statistic that tests whether two sets of field samples come from the same distribution, used to rank how out-of-distribution each code is relative to Gadget, and Gaussian smoothing applied at integer multiples of the grid spacing to show that resolution differences, not code physics, drive the failure.

What would settle it

Run a matched-seed Gadget and Ramses pair, measure the small-scale density variance inside voids, and smooth Gadget until its void variance matches Ramses; if the network still misreads Ramses while reading the smoothed Gadget correctly, the effective-resolution explanation is wrong, whereas if the bias disappears exactly when void small-scale power matches, it is supported.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is a robustness failure with a known cure. A convolutional-network-based, field-level model for $\Omega_m$ and $\sigma_8$, trained on 1000 Gadget Latin Hypercube simulations, is effectively unbiased when tested on other non-AMR codes, with $R^2\sim 1$; Ramses, however, produces a consistent bias larger than 20% for both parameters, and the same network tested on a higher-resolution version of Gadget is also badly biased. The failure is diagnosed as a difference in effective resolution: AMR solvers do not refine low-density regions, so their fields look smoother, while higher-resolution runs resolve more small clusters, and CNNs are particularly sensitive to the small-scale fluctuations that distinguish these cases. Smoothed with a Gaussian of standard deviation around $6\Delta_R$, where $\Delta_R\approx 0.1$ Mpc$/h$ is the grid spacing, every simulation, including Ramses and the hydrodynamic TNG suite, yields PQMass distributions consistent with Gadget and unbiased cosmological inference, which the paper presents as the treatment that makes field-level inference robust.

Load-bearing premise

The load-bearing premise is that adaptive-mesh-refinement simulations differ from fixed-resolution ones mainly in effective spatial resolution, so that smoothing by about six grid spacings removes the real cause of the bias rather than merely destroying all the information the network was using.

Editorial extensions

If this is right

  • A field-level model trained on one fixed-resolution code can be transferred to other fixed-resolution codes without significant bias in the two parameters tested.
  • Without filtering, the same model applied to AMR simulations or to a higher-resolution run of the training code produces biased parameters, with the AMR bias exceeding the hydrodynamic bias.
  • Smoothing at roughly six grid spacings restores unbiased $\Omega_m$ and $\sigma_8$ inference across all tested codes and the hydrodynamic simulation, at the price of weaker constraints.
  • Field-level out-of-distribution tests such as PQMass identify the simulations that will bias inference, but they are conservative: some out-of-distribution codes (e.g. CUBEP$^3$M) still yield unbiased parameters.
  • This motivates comparing simulations and computing out-of-distribution metrics across smoothing scales before trusting any field-level inference pipeline.

Reading between the lines

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

  • Inference: if effective resolution is the mechanism, the required smoothing scale should grow with the smallest scale faithfully resolved by the slower of the two simulations; for other box sizes, particle numbers, or softening lengths, the fixed number 'six grid spacings' is likely a proxy for a physical scale that needs to be re-measured.
  • Inference: the same failure should affect any inference method, not only CNNs, that uses small-scale density fluctuations in low-density regions, so power-spectrum-based analyses with aggressive small-scale cuts may show a milder but analogous version of the AMR discrepancy.
  • Inference: a testable extension is to replace Gaussian smoothing with physically motivated filters, such as density cuts or wavelet denoising, and check whether the unbiased regime can be reached with less loss of constraining power than at $6\Delta_R$.
  • Inference: pairing a field-level out-of-distribution metric with a downstream inference-bias measurement, as done here, gives a more practical robustness criterion than relying on either alone.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper presents a field-level comparison of seven cosmological N-body codes (Abacus, CUBEP3M, Enzo, Gadget, Gizmo, PKDGrav3, Ramses) plus the IllustrisTNG hydrodynamic simulation, using matched-initial-condition and Latin Hypercube simulation suites. It combines traditional summary statistics (power spectra, cross-correlation, visual maps) with the PQMass two-sample test and CNN-based simulation-based inference for Omega_m and sigma_8. The main empirical findings are that a CNN trained on Gadget transfers well to other non-AMR codes, fails badly on AMR codes (Ramses, Enzo), and that the AMR-induced bias is larger than the bias from testing on a hydrodynamic simulation. The paper attributes this to small-scale resolution differences, and reports that smoothing the fields by about 6 grid spacings removes the out-of-distribution signal in PQMass and restores unbiased CNN inference. A more refined Ramses run reduces but does not eliminate the bias.

Significance. If the central interpretation holds, the paper is a valuable contribution: it demonstrates that field-level SBI is sensitive to code-specific small-scale features, that summary statistics such as the power spectrum and cross-correlation do not fully capture the distribution shifts relevant for field-level inference, and that PQMass is a useful OOD diagnostic for simulation-based inference. The multiple complementary probes, the analytic chi-squared null check for PQMass, the held-out validation on Gadget, the fixed-seed controlled experiments, and the high-resolution and Ramses2 variants give the empirical bias results solid support and partially ground the resolution interpretation. The main caveat is that the quantitative smoothing prescription and its causal interpretation are not yet established with the same rigor as the empirical bias measurements.

major comments (4)
  1. [Sec. 3.4, Fig. 6 caption, Abstract, Conclusions] The smoothing filter is described inconsistently: the text says "To smooth, we apply a top-hat filter with size given by integer multiples of the grid spacing," while the Fig. 6 caption, the abstract, and the Conclusions describe Gaussian smoothing with a standard deviation of about 6 grid spacings. A top-hat of width 6 Delta_R and a Gaussian with sigma = 6 Delta_R suppress power very differently, so the headline scale "6 grid spacings" is not well defined. Please state the exact filter used, re-run the key validation with that filter, and align the abstract and Conclusions with the actual procedure. Also reconcile the internal discrepancy between "approximately 5 Delta_R" in the Fig. 6 caption and "8 or more 6 Delta_R" in the text and Conclusions (Section 3.4).
  2. [Sec. 3.4, Figs. 6 and 7] The smoothing scale is selected on the same data used to validate it. The value of about 6 Delta_R is read off from the bias curves in Fig. 6 for exactly the simulations (Gadget_HR, Ramses) that later demonstrate success in Fig. 7, so the post-smoothing PQMass agreement is not an independent confirmation of the prescription. This makes the quantitative "about 6 Delta_R" claim a data-derived artifact rather than a tested physical statement about simulation fidelity. To support the quantitative claim, select the scale on a training subset or derive it from an independent criterion (for example, the scale at which PQMass converges to the null across all codes) and then validate the chosen scale on a held-out set.
  3. [Sec. 3.4, Fig. 6 caption] The caption states that the smoothing experiment trains on "fixed seed and cosmology Gadget simulations." If the model is trained on a single cosmology, the "unbiased inference" shown in Fig. 6 is trivial, because the network can memorize the single training label, and the claim that smoothing recovers unbiased inference for Omega_m and sigma_8 across the parameter prior is unsupported. If the Latin Hypercube Gadget set was used instead (as the text "training the model on a smoothed version of Gadget" suggests), please state this explicitly in the caption and main text. If a single-cosmology training set was used, re-run the experiment with cosmology-varying training data and report inference over the full Latin Hypercube test set.
  4. [Sec. 3.4, Figs. 6 and 7] The experiment cannot distinguish the paper's resolution-alignment interpretation from a simpler information-destruction interpretation. Applying the same low-pass filter to training and test fields removes precisely the small-scale modes that carry the Gadget/Ramses discrepancy; after sufficiently aggressive smoothing, any two simulations become statistically similar, so the PQMass agreement in Fig. 7 is expected even if the smoothing simply discards information. The Ramses2 and HR-Gadget results provide partial evidence for a resolution effect, but they are not part of the smoothing control. Add a control that isolates the resolution gap (for example, smooth a same-resolution, different-seed pair and show that the threshold behavior is different, or quantify the information retained at 6 Delta_R by reporting the constraining power on Omega_m and sigma_8 before and after smoothing).
minor comments (4)
  1. [Sec. 3.1, Fig. 1] The text states that the power spectra agree "to within a few percent across all scales," but the left panel of Fig. 1 shows deviations of order 10 percent at high k for some codes; please reconcile the wording with the figure.
  2. [Fig. 6 caption] The caption says "The true value of the parameters is denoted by the dotted solid line"; this is self-contradictory. Please use either "dotted line" or "solid line".
  3. [Sec. 2.1] The choice of a 64:1 grid-to-particle ratio and a 4.9 kpc/h softening length for CUBEP3M is described but not justified in relation to the effective-resolution claims; a sentence explaining how these choices compare with the other codes would help the reader assess the resolution interpretation.
  4. [References] The reference list contains two entries for Villaescusa-Navarro et al. 2021 (arXiv:2109.09747 and ApJ 915, 71); the in-text citation in Section 2.3 and elsewhere should be checked so that each citation points to the intended entry.

Circularity Check

1 steps flagged · score 4.0 of 10

The 6ΔR smoothing scale is selected on the same data used to confirm it, making the headline prescription mildly circular; the cross-code failure and smoothing effect retain independent empirical content.

  1. fitted input called prediction [Section 3.4, Figures 6 and 7; Conclusions]
    "It can seen that smoothing of around ≳ 5∆R is required for unbiased inference when the testing set does not match the training set. ... Finally, we apply PQMass on maps smoothed by 6∆R. Fig. 7 shows the χ2 PQM distributions for all simulations now closely follow the expected χ2 distribution, indicating that smoothing brings everything in distribution."

    The 6ΔR scale is not an a priori or held-out choice: it is read from the same Fig. 6 curves (Gadget HR and Ramses inference) as the point where the bias vanishes, and is then fixed to 6ΔR for the PQMass validation in Fig. 7. Presenting that PQMass agreement as confirmation of the smoothing prescription is a post-selection check: the filter scale was selected on these simulations to remove the discrepancy, so the later agreement is partially determined by that selection. The paper's own numbers shift between ≳5ΔR in the figure caption and ≳6ΔR in the text, indicating that the value is a data-derived threshold rather than a predicted constant.

full rationale

The core comparison results are independently grounded. PQMass is checked against the analytic chi-square null using Gadget-versus-Gadget samples, and the CNN is validated on held-out Gadget simulations before testing on other codes. The catastrophic AMR bias and the qualitative fact that smoothing removes it are genuine empirical findings. The main circularity burden is concentrated in the specific smoothing scale: Section 3.4 scans smoothing in integer grid spacings, finds unbiased inference around 5-6ΔR on the same simulations, and then uses 6ΔR as the fixed filter for the PQMass confirmation in Figure 7. This is post-selection rather than an independent test of the scale. A competing explanation—that aggressive smoothing destroys the informative small-scale modes that the CNN uses, making agreement trivial—is an identifiability concern about the causal attribution, not a self-referential reduction, so it is not scored as a separate circular step. A related ambiguity is that the text in Section 3.4 describes a top-hat filter while the figure captions and conclusions describe Gaussian smoothing with standard deviation 6ΔR; this affects the precision of the prescription but is not itself circularity. Overall, the central claim is not forced by construction or by self-citation, but the headline smoothing scale is partially self-validated on the same data, warranting a moderate score rather than a clean bill.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claims rely on standard statistical assumptions for PQMass, on matched initial conditions and projection choices, and on the interpretation that Gaussian smoothing aligns effective resolution. The main fitted element is the 6 Delta_R smoothing scale, selected from the same data used to validate it. No new physical entities, particles, forces, or conserved quantities are introduced.

free parameters (4)
  • Gaussian smoothing scale R_smth = approximately 6 Delta_R, about 0.6 Mpc/h
    Selected from the data in Section 3.4 and Figures 6 and 7 to bring PQMass into consistency and remove inference bias. It is not derived from first principles or validated on an independent set.
  • PQMass number of Voronoi regions n_R = 50
    Hand-chosen in Section 2.2. It determines the degrees of freedom and sensitivity of the OOD test; the authors note the null test standard deviation is larger than the theoretical value, indicating sensitivity to this choice and to sample size.
  • CUBEP3M grid-to-particle ratio and softening = 64:1 grid ratio, softening 4.9 kpc/h
    Chosen in Section 2.1 because the code ran faster and produced better results at 64:1 rather than 8:1. This is an ad hoc configuration that could affect CUBEP3M's field-level behavior and its comparison to other codes.
  • CNN hyperparameters (hidden dimension H, dropout, weight decay, max learning rate) = not reported for the final model
    Hyperparameters were optimized with Optuna in Section 2.3, but the final selected values are not quoted, so the trained network is not uniquely specified.
assumptions (6)
  • standard math The PQMass statistic chi^2_PQM follows a chi^2 distribution with n_R minus 1 degrees of freedom under the null of identical underlying distributions.
    Used in Section 2.2 to interpret all OOD results. The authors run a Gadget versus Gadget null test that approximately matches the expected distribution, giving some empirical support for this standard assumption.
  • domain assumption Matched initial conditions and cosmology across codes isolate solver differences.
    Section 2.1 states all simulations use 256^3 particles from z=127 in a 25 Mpc/h box with matched seeds and cosmology, except CUBEP3M which used the Zel'dovich approximation. Differences are then attributed to the N-body codes.
  • domain assumption The SPH-painted 2D projected density map is a sufficient representation of the field for this comparison and for CNN inference.
    Section 2.1 paints particle snapshots to 2D images using a spherical kernel interpolation scheme. All OOD and SBI analyses operate on these maps; the paper checks 3D power and cross-power for trends but not for OOD or inference.
  • domain assumption Gaussian smoothing by about 6 grid spacings aligns effective resolution without removing information needed for unbiased Omega_m and sigma_8 inference.
    This is the interpretive premise of Section 3.4 and Figures 6 and 7. It is supported by the post-smoothing PQMass and inference results, but the smoothing scale is selected from the same data used to validate the assumption.
  • domain assumption Training on Gadget-only Latin Hypercube realizations is a valid sensitivity analysis for model misspecification.
    Section 2.3 uses Gadget for training and all other codes for testing. The conclusions that non-AMR transfer works and AMR transfer fails depend on this design choice.
  • domain assumption The amortized network loss in Equation 2 yields sufficiently well-calibrated posterior means and variances for Omega_m and sigma_8.
    Section 2.3 and the Figure 4 caption note that chi^2 is below 1 even for Gadget because the inferred error bars are slightly overpredicted. Mean biases are likely robust, but the reported chi^2 values should be read cautiously.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Field-Level Comparison and Robustness Analysis of Cosmological N-body Simulations." pith.science (2026). https://pith.science/paper/NHG2FDIY

@misc{pith2026250513620,
  author       = {Pith},
  title        = {Pith review of: Field-Level Comparison and Robustness Analysis of Cosmological N-body Simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NHG2FDIY}},
  note         = {Machine review of arXiv:2505.13620}
}
abstract

We present the first field-level comparison of cosmological N-body simulations, considering various widely used codes: Abacus, CUBEP$^3$M, Enzo, Gadget, Gizmo, PKDGrav, and Ramses. Unlike previous comparisons focused on summary statistics, we conduct a comprehensive field-level analysis: evaluating statistical similarity, quantifying implications for cosmological parameter inference, and identifying the regimes in which simulations are consistent. We begin with a traditional comparison using the power spectrum, cross-correlation coefficient, and visual inspection of the matter field. We follow this with a statistical out-of-distribution (OOD) analysis to quantify distributional differences between simulations, revealing insights not captured by the traditional metrics. We then perform field-level simulation-based inference (SBI) using convolutional neural networks (CNNs), training on one simulation and testing on others, including a full hydrodynamic simulation for comparison. We identify several causes of OOD behavior and biased inference, finding that resolution effects, such as those arising from adaptive mesh refinement (AMR), have a significant impact. Models trained on non-AMR simulations fail catastrophically when evaluated on AMR simulations, introducing larger biases than those from hydrodynamic effects. Differences in resolution, even when using the same N-body code, likewise lead to biased inference. We attribute these failures to a CNN's sensitivity to small-scale fluctuations, particularly in voids and filaments, and demonstrate that appropriate smoothing brings the simulations into statistical agreement. Our findings motivate the need for careful data filtering and the use of field-level OOD metrics, such as PQMass, to ensure robust inference.

Figures

Figures reproduced from arXiv: 2505.13620 by the authors.

Figure 1
Figure 1. The power spectrum (left) and cross correlation coefficient (right) of all N-body simulations with [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. 2D density fields Σ produced by the different N-body codes with matched seed and cosmology. The top row shows a 5 Mpc/h thick slice of the fields which look comparable by eye. The second row shows a zoom-in to the white box in the top row, where it can be seen that filaments and voids are smoother in the AMR codes, in particular for Ramses. The bottom two rows show the ratio of the density field of each simulation c… view at source ↗
Figure 3
Figure 3. Field-level OOD analysis. PQMass χ 2 PQM distributions, comparing each simulation with respect to Gadget. Under the null hypothesis, and given our use of 50 comparison samples, when two sets of samples are from the same underlying distribution, χ 2 PQM should follow a χ 2 distribution with 49 degrees of freedom (black line). When comparing Gadget to Gadget, as a null test, χ 2 PQM closely follows the χ 2 distributio… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Field-level SBI results when training on the Gadget Latin Hypercube and testing on other simulations [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: SBI results when training on Gadget simulations, and testing on all simulations with fixed seed and [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Smoothing analysis. SBI results when training on fixed seed and cosmology Gadget simulations, and testing on Gadget, high-resolution Gadget (Gadget HR), Ramses, and an excessively refined version of Ramses to bring it closer to a non-AMR simulation (Ramses2). The true …
Figure 7
Figure 7. Figure 7: Field-level OOD analysis on smoothed fields. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DISCO-DJ II: a differentiable particle-mesh code for cosmology

    astro-ph.CO 2025-10 conditional novelty 6.0 of 10

    A GPU-accelerated, differentiable particle-mesh N-body code achieves per-cent-level power-spectrum accuracy with few time steps and recovers sigma_8 plus initial conditions from a noisy mock field.

  2. Mitigating Model Misspecification in Simulation-Based Inference for Galaxy Clustering

    astro-ph.CO 2025-07 conditional novelty 5.0 of 10

    A two-step method (coefficient pruning plus learned robust transformation) fixes model misspecification in the SimBIG wavelet-scattering analysis of BOSS galaxy clustering and produces tight Lambda-CDM constraints.

Reference graph

Works this paper leans on

63 extracted references · 13 canonical work pages · cited by 2 Pith papers

  1. [1]

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

    ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...

  3. [3]

    qNh\4cޯ \]C4Y4 PmҶ* ( & A*M rE(J8 (M.*|>AOVjHe t R*

    thebibliography [1] 20pt to REFERENCES 6pt =0pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command E...

    2021

  4. [4]

    Agarwal, S., \'Ciprijanovi\'c, A., & Nord, B. D. 2024, in 38th conference on Neural Information Processing Systems . 2411.03334

    work page Pith review arXiv 2024

  5. [5]

    2019, in Proceedings of the 25rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

    Akiba, T., Sano, S., Yanase, T., Ohta, T., & Koyama, M. 2019, in Proceedings of the 25rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

    work page 2019

  6. [6]

    2022, Mon

    Amon, A., & Efstathiou, G. 2022, Mon. Not. Roy. Astron. Soc., 516, 5355, 10.1093/mnras/stac2429

    doi:10.1093/mnras/stac2429 2022

  7. [7]

    2022, in 36th Conference on Neural Information Processing Systems : Workshop on Machine Learning and the Physical Sciences

    Anau Montel, N., & Weniger, C. 2022, in 36th Conference on Neural Information Processing Systems : Workshop on Machine Learning and the Physical Sciences . 2211.04291

    arXiv 2022

  8. [8]

    Andrews, A., et al. 2024. 2412.11945

    arXiv 2024

Show all 63 references
  1. [9]

    E., Modi , C., & Ferraro , S

    Bayer , A. E., Modi , C., & Ferraro , S. 2023, , 2023, 046, 10.1088/1475-7516/2023/06/046

    2023 doi

  2. [10]

    E., Seljak, U., & Modi, C

    Bayer, A. E., Seljak, U., & Modi, C. 2023, in 40th International Conference on Machine Learning . 2307.09504

    2023 arXiv

  3. [11]

    L., Norman , M

    Bryan , G. L., Norman , M. L., O'Shea , B. W., et al. 2014, , 211, 19, 10.1088/0067-0049/211/2/19

    2014 doi

  4. [12]

    Filipp, A., Hezaveh, Y., & Perreault-Levasseur, L. 2024. 2411.05905

    2024

  5. [13]

    S., et al

    Frenk, C. S., et al. 1999, Astrophys. J., 525, 554, 10.1086/307908

    1999 doi

  6. [14]

    H., Eisenstein , D

    Garrison , L. H., Eisenstein , D. J., Ferrer , D., Maksimova , N. A., & Pinto , P. A. 2021 a , , 508, 575, 10.1093/mnras/stab2482

    2021 doi

  7. [15]

    H., Eisenstein , D

    Garrison , L. H., Eisenstein , D. J., & Pinto , P. A. 2019, , 485, 3370, 10.1093/mnras/stz634

    2019 doi

  8. [16]

    H., Joyce , M., & Eisenstein , D

    Garrison , L. H., Joyce , M., & Eisenstein , D. J. 2021 b , , 504, 3550, 10.1093/mnras/stab1096

    2021 doi

  9. [17]

    Golshan, M., & Bayer, A. E. 2025, JCAP, 05, 024, 10.1088/1475-7516/2025/05/024

    2025 doi

  10. [18]

    1987, Journal of Computational Physics, 73, 325, 10.1016/0021-9991(87)90140-9

    Greengard , L., & Rokhlin , V. 1987, Journal of Computational Physics, 73, 325, 10.1016/0021-9991(87)90140-9

    1987 doi

  11. [19]

    C., et al

    Grove , C., Chuang , C.-H., Devi , N. C., et al. 2022, , 515, 1854, 10.1093/mnras/stac1947

    2022 doi

  12. [20]

    2024, Nature Astron., 8, 1457, 10.1038/s41550-024-02344-2

    Hahn, C., et al. 2024, Nature Astron., 8, 1457, 10.1038/s41550-024-02344-2

    2024 doi

  13. [21]

    T., et al

    Harnois-D \'e raps , J., Pen , U.-L., Iliev , I. T., et al. 2013, , 436, 540, 10.1093/mnras/stt1591

    2013 doi

  14. [22]

    2024, Open J

    Ho, M., et al. 2024, Open J. Astrophys., 7, 001c.120559, 10.33232/001c.120559

    2024 doi

  15. [23]

    W., & Eastwood , J

    Hockney , R. W., & Eastwood , J. W. 1988, Computer simulation using particles

    1988

  16. [24]

    Hopkins , P. F. 2015, , 450, 53, 10.1093/mnras/stv195

    2015 doi

  17. [25]

    Horowitz, B., & Lukic, Z. 2025. 2502.02294

    2025

  18. [26]

    2022, in 36th Conference on Neural Information Processing Systems : Workshop on Machine Learning and the Physical Sciences

    Horowitz, B., & Melchior, P. 2022, in 36th Conference on Neural Information Processing Systems : Workshop on Machine Learning and the Physical Sciences . 2211.14788

    2022 arXiv

  19. [27]

    2019, Phys

    Inman, D., & Ali-Ha\" moud, Y. 2019, Phys. Rev. D, 100, 083528, 10.1103/PhysRevD.100.083528

    2019 doi

  20. [28]

    2019, Astron

    Jasche, J., & Lavaux, G. 2019, Astron. Astrophys., 625, A64, 10.1051/0004-6361/201833710

    2019 doi

  21. [29]

    Jasche , J., & Wandelt , B. D. 2013, MNRAS, 432, 894, 10.1093/mnras/stt449

    2013 doi

  22. [30]

    2021, Mon

    Jeffrey, N., Alsing, J., & Lanusse, F. 2021, Mon. Not. Roy. Astron. Soc., 501, 954, 10.1093/mnras/staa3594

    2021 doi

  23. [31]

    Jeffrey , N., & Wandelt , B. D. 2020, arXiv e-prints, arXiv:2011.05991. 2011.05991

    2020 arXiv

  24. [32]

    Jia, H. 2024. 2411.14748

    2024 arXiv

  25. [33]

    Jo, Y., Genel, S., Sengupta, A., et al. 2025. 2502.13239

    2025 arXiv

  26. [34]

    2023, JCAP, 07, 063, 10.1088/1475-7516/2023/07/063

    Kosti\'c, A., Nguyen, N.-M., Schmidt, F., & Reinecke, M. 2023, JCAP, 07, 063, 10.1088/1475-7516/2023/07/063

    2023 doi

  27. [35]

    E., & Villaescusa-Navarro, F

    Lahiry, A., Bayer, A. E., & Villaescusa-Navarro, F. 2025, Interpreting Cosmological Information from Neural Networks in the Hydrodynamic Universe. 2504.17839

    2025

  28. [36]

    2024, arXiv e-prints, arXiv:2402.04355, 10.48550/arXiv.2402.04355

    Lemos , P., Sharief , S., Malkin , N., et al. 2024, arXiv e-prints, arXiv:2402.04355, 10.48550/arXiv.2402.04355

    2024 doi arXiv

  29. [37]

    2024, Phys

    Lemos, P., et al. 2024, Phys. Rev. D, 109, 083536, 10.1103/PhysRevD.109.083536

    2024 doi

  30. [38]

    Masserano, L., Dorigo, T., Izbicki, R., Kuusela, M., & Lee, A. B. 2023, Simulator-Based Inference with Waldo: Confidence Regions by Leveraging Prediction Algorithms and Posterior Estimators for Inverse Problems. 2205.15680

    2023 arXiv

  31. [39]

    Metchnik , M. V. L. 2009, PhD thesis, University of Arizona

    2009

  32. [40]

    Modi, C., Pandey, S., Ho, M., et al. 2023. 2309.15071

    2023 arXiv

  33. [41]

    2021, JCAP, 03, 058, 10.1088/1475-7516/2021/03/058

    Nguyen, N.-M., Schmidt, F., Lavaux, G., & Jasche, J. 2021, JCAP, 03, 058, 10.1088/1475-7516/2021/03/058

    2021 doi

  34. [42]

    2024, Phys

    Nguyen, N.-M., Schmidt, F., Tucci, B., Reinecke, M., & Kosti\'c, A. 2024, Phys. Rev. Lett., 133, 221006, 10.1103/PhysRevLett.133.221006

    2024 doi

  35. [43]

    E., & Seljak, U

    Parker, L., Bayer, A. E., & Seljak, U. 2025. 2504.01092

    2025 arXiv

  36. [44]

    2018, , 473, 4077, 10.1093/mnras/stx2656

    Pillepich , A., Springel , V., Nelson , D., et al. 2018, , 473, 4077, 10.1093/mnras/stx2656

    2018 doi

  37. [45]

    2017, Computational Astrophysics and Cosmology, 4, 2, 10.1186/s40668-017-0021-1

    Potter , D., Stadel , J., & Teyssier , R. 2017, Computational Astrophysics and Cosmology, 4, 2, 10.1186/s40668-017-0021-1

    2017 doi

  38. [46]

    B., Agertz, O., Moore, B., & Stadel, J

    Romeo, A. B., Agertz, O., Moore, B., & Stadel, J. 2008, Astrophys. J., 686, 1, 10.1086/591236

    2008 doi

  39. [47]

    2020, JCAP, 11, 008, 10.1088/1475-7516/2020/11/008

    Schmidt, F., Cabass, G., Jasche, J., & Lavaux, G. 2020, JCAP, 11, 008, 10.1088/1475-7516/2020/11/008

    2020 doi

  40. [48]

    M., & Lavaux, G

    Schmidt, F., Elsner, F., Jasche, J., Nguyen, N. M., & Lavaux, G. 2019, JCAP, 01, 042, 10.1088/1475-7516/2019/01/042

    2019 doi

  41. [49]

    2016, Journal of Cosmology and Astroparticle Physics, 2016, 10.1088/1475-7516/2016/04/047

    Schneider, A., Teyssier, R., Potter, D., et al. 2016, Journal of Cosmology and Astroparticle Physics, 2016, 10.1088/1475-7516/2016/04/047

    2016 doi

  42. [50]

    2017, Journal of Cosmology and Astroparticle Physics, 2017, 009–009, 10.1088/1475-7516/2017/12/009

    Seljak, U., Aslanyan, G., Feng, Y., & Modi, C. 2017, Journal of Cosmology and Astroparticle Physics, 2017, 009–009, 10.1088/1475-7516/2017/12/009

    2017 doi

  43. [51]

    2023, Astrophys

    Shao, H., et al. 2023, Astrophys. J., 944, 27, 10.3847/1538-4357/acac7a

    2023 doi

  44. [52]

    Sharma, D., Dai, B., & Seljak, U. 2024. 2403.03490

    2024 arXiv

  45. [53]

    2005, Mon

    Springel, V. 2005, Mon. Not. Roy. Astron. Soc., 364, 1105, 10.1111/j.1365-2966.2005.09655.x

    2005

  46. [54]

    2010, , 401, 791, 10.1111/j.1365-2966.2009.15715.x

    Springel , V. 2010, , 401, 791, 10.1111/j.1365-2966.2009.15715.x

    2010

  47. [55]

    Swierc, P., Tamargo-Arizmendi, M., \'Ciprijanovi\'c, A., & Nord, B. D. 2024. 2410.16347

    2024 arXiv

  48. [56]

    2002, , 385, 337, 10.1051/0004-6361:20011817

    Teyssier , R. 2002, , 385, 337, 10.1051/0004-6361:20011817

    2002 doi

  49. [57]

    2024, JCAP, 05, 063, 10.1088/1475-7516/2024/05/063

    Tucci, B., & Schmidt, F. 2024, JCAP, 05, 063, 10.1088/1475-7516/2024/05/063

    2024 doi

  50. [58]

    2021 a , arXiv e-prints, arXiv:2109.09747

    Villaescusa-Navarro , F., Angl \'e s-Alc \'a zar , D., Genel , S., et al. 2021 a , arXiv e-prints, arXiv:2109.09747. 2109.09747

    2021 arXiv

  51. [59]

    2021 b , , 915, 71, 10.3847/1538-4357/abf7ba

    ---. 2021 b , , 915, 71, 10.3847/1538-4357/abf7ba

    2021 doi

  52. [60]

    2022 a , arXiv e-prints, arXiv:2201.01300

    Villaescusa-Navarro , F., Genel , S., Angl \'e s-Alc \'a zar , D., et al. 2022 a , arXiv e-prints, arXiv:2201.01300. 2201.01300

    2022 arXiv

  53. [61]

    2022 b , , 259, 61, 10.3847/1538-4365/ac5ab0

    ---. 2022 b , , 259, 61, 10.3847/1538-4365/ac5ab0

    2022 doi

  54. [62]

    2020, Astrophys

    Weinberger, R., Springel, V., & Pakmor, R. 2020, Astrophys. J. Suppl., 248, 32, 10.3847/1538-4365/ab908c

    2020 doi

  55. [63]

    2017, , 465, 3291, 10.1093/mnras/stw2944

    Weinberger , R., Springel , V., Hernquist , L., et al. 2017, , 465, 3291, 10.1093/mnras/stw2944

    2017 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.