pith. machine review for the scientific record. sign in

arxiv: 2604.23147 · v2 · submitted 2026-04-25 · 🌌 astro-ph.GA

Recognition: unknown

A graph-based Neural Network surrogate model for accelerating semi-analytical model of galaxy formation and evolution

Xuejie Li , Zhongxu Zhai , Xiaohu Yang , Andrew Benson , Yun Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-08 07:30 UTC · model grok-4.3

classification 🌌 astro-ph.GA
keywords galaxy formationsemi-analytic modelsgraph neural networkssurrogate modelingmerger treesstellar massmachine learningcosmological simulations
0
0 comments X

The pith

A conditional graph neural network acts as an accurate fast surrogate for full semi-analytic galaxy formation models by learning from dark matter merger trees and model parameters.

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

The paper develops a graph neural network that processes dark matter halo merger trees together with semi-analytic model parameters to predict multiple galaxy properties at different redshifts. It shows that one trained model reproduces outputs such as stellar mass, luminosity, angular momentum, gas metal mass, and specific star formation rate from the Galacticus code applied to Uchuu simulation trees. The network achieves scatters of 0.19-0.28 dex and R-squared values near 0.95 for stellar mass across 0 to 3 in redshift, and maintains similar fidelity for other quantities up to redshift 5 and across varied parameter sets. If this holds, large ensembles of galaxies and broad parameter explorations become feasible at far lower computational cost than repeated full runs of the semi-analytic model. This shift would allow more extensive comparisons of theoretical galaxy populations with observations to test formation scenarios.

Core claim

The central claim is that a single conditional graph neural network trained on Galacticus semi-analytic model outputs for Uchuu merger trees can predict galaxy properties including stellar mass with 0.19-0.28 dex scatter and R-squared of 0.946-0.973 over 0 <= z <= 3, and comparably well for other properties up to z=5, while generalizing across multiple SAM realizations, unseen merger trees, and redshifts without major loss of fidelity relative to the full model.

What carries the argument

The conditional graph neural network that encodes merger tree structure as graph input and conditions predictions on semi-analytic model parameter values to output time-evolving galaxy properties.

If this is right

  • Studies of galaxy formation can examine far larger numbers of merger trees and wider ranges of model parameters than direct computation currently allows.
  • Catalog-scale predictions of galaxy populations become practical while retaining close agreement with the underlying semi-analytic model for statistical purposes.
  • Detailed mapping of how changes in SAM parameters affect observable galaxy traits across cosmic time becomes more accessible.
  • The released inference code enables other researchers to train and deploy similar surrogates for their own merger tree sets and models.

Where Pith is reading between the lines

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

  • The same graph structure could be tested as a surrogate for other semi-analytic codes or even hydrodynamical simulation outputs if equivalent tree data is supplied.
  • Fast inference might support iterative fitting of SAM parameters to large observational catalogs in near real time.
  • Reduced computational barriers could help isolate which physical assumptions in galaxy formation models are most tightly constrained by existing data.

Load-bearing premise

The trained network generalizes to unseen merger trees, parameter values, and redshifts while matching full model outputs without significant biases or fidelity loss.

What would settle it

Run the trained network on a fresh collection of merger trees and held-out SAM parameter combinations, then compare its stellar mass and other property predictions directly against full Galacticus runs on the same inputs to check whether scatter remains under 0.3 dex with no systematic offsets.

Figures

Figures reproduced from arXiv: 2604.23147 by Andrew Benson, Xiaohu Yang, Xuejie Li, Yun Wang, Zhongxu Zhai.

Figure 1
Figure 1. Figure 1: Lower triangular pair plot for the five target properties at z1 = 0 in a representative single Galacticus catalog drawn from the halo mass selected sample used in the main analysis, illustrating the joint target distribution view at source ↗
Figure 2
Figure 2. Figure 2 view at source ↗
Figure 4
Figure 4. Figure 4: Stellar mass performance as a function of red￾shift for the GNN. Blue circles show the scatter, and orange squares show R 2 . The scatter remains near 0.19–0.28 through the interme￾diate snapshots before increasing to 0.352 at the highest redshift, while R2 stays between 0.946 and 0.973 through the first eight outputs and decreases to 0.867 only at the final snapshot. Stellar mass remains one of the most s… view at source ↗
Figure 5
Figure 5. Figure 5: Stellar mass functions across four redshift outputs. The upper row shows the predicted and true functions for a representative single SAM catalog. The lower row summarizes the corresponding residuals across the 600 test SAM catalogs; the solid curve gives the median residual in each mass bin, and the shaded band shows the 16th–84th percentile range. 10 0 10 1 (r) z=0.00 True Predicted 9.0-9.5 9.5-10.0 10.0… view at source ↗
Figure 6
Figure 6. Figure 6: Two-point correlation functions for a representative single SAM catalog across four redshift outputs. In each panel, the surrogate prediction is compared against the true catalog in several stellar mass bins. The agreement is generally good over the intermediate scales and mass ranges. The largest deviations occur in the highest-mass and highest-redshift bins, where the galaxy sample is sparse. In addition… view at source ↗
Figure 7
Figure 7. Figure 7: Predicted versus true values at z1 = 0 for the four remaining targets. The upper row shows Lz and J, and the lower row shows sSFR and MZ,gas. The one-to-one line is shown in orange in each panel view at source ↗
Figure 8
Figure 8. Figure 8: Continuous scatter distributions for the five target properties across the nine redshift outputs. In each row, the left half violin shows the distribution, over 20 000 fixed test trees, of the standard deviation measured across 600 test SAM catalogs. The right half violin shows the distribution, over those 600 test SAM catalogs, of the standard deviation measured across the same fixed tree set. The horizon… view at source ↗
Figure 9
Figure 9. Figure 9: Catalog-level fractions for the mass-selected sam￾ple as a function of redshift. The upper panel shows the quenched fraction and the lower panel shows the fraction of systems at the floor value in gas metal mass. At each red￾shift, the violin indicates the full distribution across catalogs, the box marks the interquartile range, and the central line marks the median. Absent in all SAMs Varying across SAMs … view at source ↗
Figure 10
Figure 10. Figure 10: Fractions of test trees that are absent from a given state in all SAM catalogs, present in that state in all catalogs, or vary across catalogs at z = 0. Blue bars show the quenched state, and orange bars show the floor state in gas metal mass. 6.1. Decomposing Variation Across SAMs and Trees view at source ↗
Figure 11
Figure 11. Figure 11: F1 scores for the quenched and floor states in gas metal mass under three data settings: the full cross SAM problem, a single SAM problem, and a single SAM balanced tree setting. Each panel corresponds to one of the first four redshift outputs. Within each panel, the two bar styles represent the two binary targets view at source ↗
Figure 12
Figure 12. Figure 12: Predicted versus true values for the five target properties at five representative redshift outputs. Columns correspond to different redshifts and rows correspond to different target properties. Each panel shows the one-to-one relation together with the corresponding predicted and true values from the the fiducial Multi-SAM GNN. APPENDIX A. ADDITIONAL PREDICTION PANELS view at source ↗
read the original abstract

Understanding how galaxy populations emerge and evolve from the growth of dark matter structure is a central challenge in galaxy formation theory. Semi-analytic models (SAMs) provide an efficient framework to address this problem, but exploring large ensembles of merger trees across broad parameter spaces remains computationally demanding. We develop a conditional graph neural network surrogate model that combines merger tree information with SAM parameters to predict galaxy properties across cosmic time. Using merger trees of dark matter halos from the Uchuu simulation and the Galacticus SAM, the model predicts stellar mass, luminosity, angular momentum, gas metal mass, and specific star formation rate across the wide redshift range of 0 <= z <= 5. For instance, the model can predict stellar mass at 0 <= z <= 3 with a scatter of 0.19-0.28 dex and coefficient of determination R^2 of 0.946-0.973 (R^2 close to 1 indicates prediction closely matching the truth). The results show that a single graph based model can reproduce these galaxy properties with good accuracy over multiple SAM realizations, merger trees and redshifts. This catalog-level model provides a practical route for accelerating SAM based studies of galaxy formation to enable a more detailed investigation of the model parameter space. The inference code, trained models, and example data products are publicly available at https://github.com/MutongCat/sam2galaxy-gnn.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript develops a conditional graph neural network (GNN) surrogate for the Galacticus semi-analytical model (SAM) of galaxy formation and evolution. Merger trees are taken from the Uchuu N-body simulation; the GNN is conditioned on both tree structure and SAM parameter values to predict galaxy properties (stellar mass, luminosity, angular momentum, gas metal mass, specific star-formation rate) over 0 ≤ z ≤ 5. The authors report that a single model achieves scatter of 0.19–0.28 dex and R² = 0.946–0.973 for stellar mass at z ≤ 3 across multiple SAM realizations, trees, and redshifts, and argue that the approach accelerates large-scale parameter-space exploration. Trained models and inference code are released publicly.

Significance. If the generalization to unseen merger trees, SAM parameters, and redshifts is robust, the surrogate would materially reduce the computational cost of running large ensembles of Galacticus realizations, enabling denser sampling of feedback and star-formation parameter spaces that are currently prohibitive. The public release of code and models further increases the potential utility for the community.

major comments (2)
  1. [Results (performance metrics) and Methods (data splitting)] The headline claim that the conditional GNN reproduces galaxy properties “with good accuracy over multiple SAM realizations, merger trees and redshifts” (abstract) rests on the assumption that test performance reflects generalization rather than interpolation. The manuscript does not specify how the training/validation/test splits were constructed with respect to the SAM parameter values (e.g., feedback efficiencies, star-formation thresholds) or redshift ranges. Without an explicit out-of-distribution test (different parameter combinations or redshifts never seen in training), the quoted R² and scatter values cannot be taken as evidence for the broader surrogate utility asserted.
  2. [Results section] Performance is reported only in aggregate for 0 ≤ z ≤ 3. A redshift-binned or parameter-binned breakdown (e.g., in the main results table or supplementary figures) is needed to identify regimes where fidelity degrades, especially near z = 3 or at the edges of the sampled SAM parameter space.
minor comments (2)
  1. [Abstract] The abstract states predictions extend to z = 5 yet supplies quantitative metrics only up to z = 3; a concise statement of performance at 3 < z ≤ 5 would improve completeness.
  2. [Methods] Notation for how SAM parameters are injected into the GNN (conditioning mechanism, embedding dimension, etc.) is introduced only briefly; a short schematic or equation would aid readers outside the graph-network community.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough and constructive review. The comments have helped us clarify key aspects of our methodology and results. We address each major comment point-by-point below and have revised the manuscript to incorporate the requested details on data splitting and performance breakdowns.

read point-by-point responses

  1. Referee: [Results (performance metrics) and Methods (data splitting)] The headline claim that the conditional GNN reproduces galaxy properties “with good accuracy over multiple SAM realizations, merger trees and redshifts” (abstract) rests on the assumption that test performance reflects generalization rather than interpolation. The manuscript does not specify how the training/validation/test splits were constructed with respect to the SAM parameter values (e.g., feedback efficiencies, star-formation thresholds) or redshift ranges. Without an explicit out-of-distribution test (different parameter combinations or redshifts never seen in training), the quoted R² and scatter values cannot be taken as evidence for the broader surrogate utility asserted.

    Authors: We agree that explicit documentation of the splitting strategy is essential for assessing generalization. The original manuscript described the use of multiple SAM realizations but did not detail the precise allocation of parameter sets and redshifts across splits. In the revised version, we have added a dedicated paragraph in the Methods section specifying that the 80/10/10 train/validation/test split was performed at the level of individual merger trees, with SAM parameter combinations (including feedback efficiencies and star-formation thresholds) drawn from a Latin-hypercube sampling. The test set contains both unseen trees and a subset of parameter combinations held out from training, providing a partial out-of-distribution evaluation. We have also added a new supplementary figure showing performance on a fully held-out SAM parameter set never encountered during training. These additions directly support the generalization claim while acknowledging that a comprehensive sweep of all possible parameter combinations remains computationally prohibitive. revision: yes

  2. Referee: [Results section] Performance is reported only in aggregate for 0 ≤ z ≤ 3. A redshift-binned or parameter-binned breakdown (e.g., in the main results table or supplementary figures) is needed to identify regimes where fidelity degrades, especially near z = 3 or at the edges of the sampled SAM parameter space.

    Authors: We concur that aggregate statistics can obscure redshift- or parameter-dependent variations. The revised manuscript now includes a new Table 2 in the Results section that reports scatter and R² values for stellar mass, luminosity, and specific star-formation rate in redshift bins of width Δz = 1 from z = 0 to z = 3. Performance remains stable (scatter 0.19–0.25 dex) up to z ≈ 2.5, with a modest increase to 0.28 dex near z = 3, consistent with the smaller number of galaxies and higher merger activity at earlier times. We have additionally placed parameter-binned diagnostics (varying feedback efficiency while holding other parameters fixed) in the supplementary material. These breakdowns confirm that the quoted headline metrics are representative across the sampled range while highlighting the expected mild degradation at the highest redshifts. revision: yes

Circularity Check

0 steps flagged

No circularity: standard supervised surrogate trained on external simulation data

full rationale

The paper trains a conditional GNN on merger trees from the independent Uchuu N-body simulation and galaxy properties generated by the Galacticus SAM. Reported metrics (0.19-0.28 dex scatter, R^2 0.946-0.973 for stellar mass) are obtained by direct comparison of model outputs to held-out SAM realizations on test merger trees. No derivation step reduces by construction to a fitted parameter, self-citation, or ansatz imported from the authors' prior work; the surrogate is falsifiable against the external SAM and does not rename or tautologically reproduce its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on standard neural network training assumptions (sufficient training data coverage, appropriate architecture capacity, and proper train/test separation) plus the fidelity of the Uchuu simulation and Galacticus SAM as ground truth; no additional free parameters, axioms, or invented physical entities are introduced beyond the GNN itself.

pith-pipeline@v0.9.0 · 5566 in / 1134 out tokens · 47484 ms · 2026-05-08T07:30:43.135456+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

52 extracted references · 41 canonical work pages · 2 internal anchors

  1. [1]

    Agarwal, S., Davé, R., & Bassett, B. A. 2018, Monthly Notices of the Royal Astronomical Society, 478, 3410

    2018

  2. [2]

    , keywords =

    Behroozi, P., Wechsler, R. H., Hearin, A. P., & Conroy, C. 2019, Monthly Notices of the Royal Astronomical Society, 488, 3143, doi: 10.1093/mnras/stz1182

    work page doi:10.1093/mnras/stz1182 2019

  3. [3]

    S., Wechsler , R

    Behroozi, P. S., Wechsler, R. H., & Wu, H.-Y. 2013a, The Astrophysical Journal, 762, 109, doi: 10.1088/0004-637X/762/2/109

    work page doi:10.1088/0004-637x/762/2/109

  4. [4]

    S., Wechsler, R

    Behroozi, P. S., Wechsler, R. H., & Wu, H.-Y. 2013b, The Astrophysical Journal, 763, 18, doi: 10.1088/0004-637X/763/1/18 21

    work page doi:10.1088/0004-637x/763/1/18

  5. [5]

    Benson, A. J. 2010, Physics Reports, 495, 33, doi: 10.1016/j.physrep.2010.06.001

    work page doi:10.1016/j.physrep.2010.06.001 2010

  6. [6]

    Benson, A. J. 2012, New Astronomy, 17, 175

    2012

  7. [7]

    K., Lin , Y.-T., Ho , S., & Genel , S

    Genel, S. 2024, The Astrophysical Journal, 965, 101, doi: 10.3847/1538-4357/ad2b6c

    work page doi:10.3847/1538-4357/ad2b6c 2024

  8. [8]

    R., Pols, O

    Cole, S., Lacey, C. G., Baugh, C. M., & Frenk, C. S. 2000, Monthly Notices of the Royal Astronomical Society, 319, 168, doi: 10.1046/j.1365-8711.2000.03879.x

    work page doi:10.1046/j.1365-8711.2000.03879.x 2000

  9. [9]

    2021, Proceedings of the National Academy of Sciences, 118, e2020324118, doi: 10.1073/pnas.2020324118 de Oliveira, R

    Dai, B., & Seljak, U. 2021, Proceedings of the National Academy of Sciences, 118, e2020324118, doi: 10.1073/pnas.2020324118 de Oliveira, R. A., Li, Y., Villaescusa-Navarro, F., Ho, S., & Spergel, D. N. 2020, arXiv e-prints. https://arxiv.org/abs/2012.00240 de Santi, N. S. M., & et al. 2025, https://arxiv.org/abs/2512.10222

    work page doi:10.1073/pnas.2020324118 2021

  10. [10]

    , keywords =

    Delgado, A. M., Wadekar, D., Hadzhiyska, B., et al. 2022, Monthly Notices of the Royal Astronomical Society, 515, 2733, doi: 10.1093/mnras/stac1951

    work page doi:10.1093/mnras/stac1951 2022

  11. [11]

    , keywords =

    Dressler, A. 1980, Astrophysical Journal, 236, 351, doi: 10.1086/157753

    work page doi:10.1086/157753 1980

  12. [12]

    , keywords =

    Elliott, E. J., Baugh, C. M., & Lacey, C. G. 2021, Monthly Notices of the Royal Astronomical Society, 506, 4011, doi: 10.1093/mnras/stab1912 Euclid Collaboration, Mellier, Y., Abdurro’uf, et al. 2025, A&A, 697, A1, doi: 10.1051/0004-6361/202450810

    work page doi:10.1093/mnras/stab1912 2021

  13. [13]

    , keywords =

    Genel, S., Vogelsberger, M., Springel, V., et al. 2014, Monthly Notices of the Royal Astronomical Society, 445, 175, doi: 10.1093/mnras/stu1654

    work page doi:10.1093/mnras/stu1654 2014

  14. [14]

    L., Ying, R., & Leskovec, J

    Hamilton, W. L., Ying, R., & Leskovec, J. 2017, Advances in Neural Information Processing Systems, 30

    2017

  15. [15]

    He, H., & Garcia, E. A. 2009, IEEE Transactions on Knowledge and Data Engineering, 21, 1263, doi: 10.1109/TKDE.2008.239

    work page doi:10.1109/tkde.2008.239 2009

  16. [16]

    2019, Proceedings of the National Academy of Sciences, 116, 13825, doi: 10.1073/pnas.1815141116

    He, S., Li, Y., Feng, Y., et al. 2019, Proceedings of the National Academy of Sciences, 116, 13825, doi: 10.1073/pnas.1815141116

    work page doi:10.1073/pnas.1815141116 2019

  17. [17]

    Gaussian Error Linear Units (GELUs)

    Hendrycks, D., & Gimpel, K. 2016, arXiv e-prints. https://arxiv.org/abs/1606.08415

    work page internal anchor Pith review arXiv 2016

  18. [18]

    L., & Haiman, Z

    Roseboom, I. 2009, Monthly Notices of the Royal Astronomical Society, 396, 535, doi: 10.1111/j.1365-2966.2009.14730.x

    work page doi:10.1111/j.1365-2966.2009.14730.x 2009

  19. [19]

    2023, MNRAS: Letters, 522, L11

    Houston, T., et al. 2023, MNRAS: Letters, 522, L11

    2023

  20. [20]

    2013, A&A, 556, A55

    Ilbert, O., & et al. 2013, A&A, 556, A55

    2013

  21. [21]

    , keywords =

    Ishiyama, T., Prada, F., Klypin, A., et al. 2021, Monthly Notices of the Royal Astronomical Society, 506, 4210, doi: 10.1093/mnras/stab1755

    work page doi:10.1093/mnras/stab1755 2021

  22. [22]

    doi:10.1162/neco.1991.3.1.79

    Jacobs, R. A., Jordan, M. I., Nowlan, S. J., & Hinton, G. E. 1991, Neural Computation, 3, 79, doi: 10.1162/neco.1991.3.1.79

    work page doi:10.1162/neco.1991.3.1.79 1991

  23. [23]

    2022a, arXiv e-prints

    Jagvaral, Y., Lanusse, F., Singh, S., et al. 2022a, arXiv e-prints. https://arxiv.org/abs/2212.05596

    work page arXiv

  24. [24]

    2022b, arXiv e-prints

    Jagvaral, Y., Lanusse, F., Singh, S., et al. 2022b, arXiv e-prints. https://arxiv.org/abs/2204.07077

    work page arXiv

  25. [25]

    K., Cranmer , M., Melchior , P., et al

    Jespersen, C. K., Cranmer, M., Melchior, P., et al. 2022, The Astrophysical Journal, 941, 7, doi: 10.3847/1538-4357/ac9b18

    work page doi:10.3847/1538-4357/ac9b18 2022

  26. [26]

    2019, Monthly Notices of the Royal Astronomical Society, 489, 3565, doi: 10.1093/mnras/stz2304

    Jo, Y., & Kim, J. 2019, Monthly Notices of the Royal Astronomical Society, 489, 3565, doi: 10.1093/mnras/stz2304

    work page doi:10.1093/mnras/stz2304 2019

  27. [27]

    M., Turk, M

    Kamdar, H. M., Turk, M. J., & Brunner, R. J. 2016, Monthly Notices of the Royal Astronomical Society, 455, 642, doi: 10.1093/mnras/stv2310

    work page doi:10.1093/mnras/stv2310 2016

  28. [28]

    Kauffmann, G., White, S. D. M., Heckman, T. M., et al. 2004, Monthly Notices of the Royal Astronomical Society, 353, 713, doi: 10.1111/j.1365-2966.2004.08117.x

    work page doi:10.1111/j.1365-2966.2004.08117.x 2004

  29. [29]

    Decoupled Weight Decay Regularization

    Loshchilov, I., & Hutter, F. 2019, arXiv e-prints. https://arxiv.org/abs/1711.05101

    work page internal anchor Pith review arXiv 2019

  30. [30]

    C., Wilkins, S

    Lovell, C. C., Wilkins, S. M., Thomas, P. A., et al. 2022, Monthly Notices of the Royal Astronomical Society, 509, 5046

    2022

  31. [31]

    Lu, Y., Kereš, D., Katz, N., & Mo, H. J. 2011, Monthly Notices of the Royal Astronomical Society, 416, 1949, doi: 10.1111/j.1365-2966.2011.19170.x

    work page doi:10.1111/j.1365-2966.2011.19170.x 2011

  32. [32]

    Monthly Notices of the Royal Astronomical Society , volume =

    Moster, B. P., Naab, T., & White, S. D. M. 2013, Monthly Notices of the Royal Astronomical Society, 428, 3121, doi: 10.1093/mnras/sts261

    work page doi:10.1093/mnras/sts261 2013

  33. [33]

    2021, Monthly Notices of the Royal Astronomical Society, 507, 1021

    Ni, Y., Li, Y., Lachance, P., et al. 2021, Monthly Notices of the Royal Astronomical Society, 507, 1021

    2021

  34. [34]

    doi:10.3847/1538-4357/ad022a , archiveprefix =

    Ni, Y., Genel, S., Anglés-Alcázar, D., et al. 2023, The Astrophysical Journal, 959, 136, doi: 10.3847/1538-4357/ad022a

    work page doi:10.3847/1538-4357/ad022a 2023

  35. [35]

    A., & Weigend, A

    Nix, D. A., & Weigend, A. S. 1994, in Proceedings of 1994 IEEE International Conference on Neural Networks, Vol. 1, 55–60, doi: 10.1109/ICNN.1994.374138

    work page doi:10.1109/icnn.1994.374138 1994

  36. [36]

    2007, MNRAS, 378, 910, doi: 10.1111/j.1365-2966.2007.11817.x

    Parkinson, H., Cole, S., & Helly, J. 2008, Monthly Notices of the Royal Astronomical Society, 383, 557, doi: 10.1111/j.1365-2966.2007.12517.x

    work page doi:10.1111/j.1365-2966.2007.12517.x 2008

  37. [37]

    J., Renzini A., Carollo M., 2012, @doi [ ] 10.1088/0004-637X/757/1/4 , http://adsabs.harvard.edu/abs/2012ApJ...757....4P 757, 4

    Peng, Y., Lilly, S. J., Carollo, M., & Renzini, A. 2012, Astrophysical Journal, 757, 4, doi: 10.1088/0004-637X/757/1/4

    work page doi:10.1088/0004-637x/757/1/4 2012

  38. [38]

    2018a, MNRAS, 475, 648, doi: 10.1093/mnras/stx3112

    Pillepich, A., Nelson, D., Hernquist, L., et al. 2018, Monthly Notices of the Royal Astronomical Society, 475, 648, doi: 10.1093/mnras/stx3112

    work page Pith review doi:10.1093/mnras/stx3112 2018

  39. [39]

    B., Angel, P

    Poole, G. B., Angel, P. W., Hartley, W. G., et al. 2017, Monthly Notices of the Royal Astronomical Society, 472, 3659, doi: 10.1093/mnras/stx2213 22

    work page doi:10.1093/mnras/stx2213 2017

  40. [40]

    2026, The Open Journal of Astrophysics, 9, doi: 10.33232/001c.155306 Rodríguez-Puebla, A., Primack, J

    Robertson, A., & Benson, A. 2026, The Open Journal of Astrophysics, 9, doi: 10.33232/001c.155306 Rodríguez-Puebla, A., Primack, J. R., Behroozi, P., &

    work page doi:10.33232/001c.155306 2026

  41. [41]

    Faber, S. M. 2017, Monthly Notices of the Royal Astronomical Society, 470, 651

    2017

  42. [42]

    A., Bower, R

    Schaye, J., Crain, R. A., Bower, R. G., et al. 2015, Monthly Notices of the Royal Astronomical Society, 446, 521, doi: 10.1093/mnras/stu2058

    work page doi:10.1093/mnras/stu2058 2015

  43. [43]

    , archivePrefix = "arXiv", eprint =

    Somerville, R. S., & Davé, R. 2015, Annual Review of Astronomy and Astrophysics, 53, 51, doi: 10.1146/annurev-astro-082812-140951

    work page Pith review doi:10.1146/annurev-astro-082812-140951 2015

  44. [44]

    Wide-Field InfrarRed Survey Telescope-Astrophysics Focused Telescope Assets WFIRST-AFTA 2015 Report

    Spergel, D., Gehrels, N., Baltay, C., et al. 2015, arXiv e-prints, arXiv:1503.03757, doi: 10.48550/arXiv.1503.03757

    work page Pith review doi:10.48550/arxiv.1503.03757 2015

  45. [45]

    2014, Statistical Science, 29, 81, doi: 10.1214/12-STS412

    Vernon, I., Goldstein, M., & Bower, R. 2014, Statistical Science, 29, 81, doi: 10.1214/12-STS412

    work page doi:10.1214/12-sts412 2014

  46. [46]

    doi:10.3847/1538-4357/ac4973 , archiveprefix =

    Wang, Y., Zhai, Z., Alavi, A., et al. 2022, ApJ, 928, 1, doi: 10.3847/1538-4357/ac4973

    work page doi:10.3847/1538-4357/ac4973 2022

  47. [47]

    H., & Tinker, J

    Wechsler, R. H., & Tinker, J. L. 2018, ARA&A, 56, 435, doi: 10.1146/annurev-astro-081817-051756

    work page Pith review doi:10.1146/annurev-astro-081817-051756 2018

  48. [48]

    White, S. D. M., & Frenk, C. S. 1991, The Astrophysical Journal, 379, 52, doi: 10.1086/170483

    work page doi:10.1086/170483 1991

  49. [49]

    2013, The Astrophysical Journal, 772, 147

    Xu, X., Ho, S., Trac, H., et al. 2013, The Astrophysical Journal, 772, 147

    2013

  50. [50]

    J., & van den Bosch, F

    Yang, X., Mo, H. J., & van den Bosch, F. C. 2003, Monthly Notices of the Royal Astronomical Society, 339, 1057

    2003

  51. [51]

    2025, arXiv e-prints

    Zhai, Z., Benson, A., & Wang, Y. 2025, arXiv e-prints. https://arxiv.org/abs/2505.18748

    work page arXiv 2025

  52. [52]

    2005, The Astrophysical Journal, 633, 791 23 Figure 12.Predictedversustruevaluesforthefivetargetpropertiesatfiverepresentativeredshiftoutputs

    Zheng, Z., et al. 2005, The Astrophysical Journal, 633, 791 23 Figure 12.Predictedversustruevaluesforthefivetargetpropertiesatfiverepresentativeredshiftoutputs. Columnscorrespond to different redshifts and rows correspond to different target properties. Each panel shows the one-to-one relation together with the corresponding predicted and true values from...

    2005