Pith. sign in

REVIEW 1 major objections 5 minor 1 cited by

This paper claims that the Epoch of Reionization's history can be inferred from Lyman-alpha forest power spectra using differentiable neural emulators, including a first-of-its-kind emulator of model-dependent covariance matrices, and demon

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

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

T0 review · deepseek-v4-flash

2026-08-04 16:28 UTC pith:G6B3PQ6T

load-bearing objection Solid method paper with a genuinely new covariance emulator; the inference-test mock provenance needs to be stated before the headline claim is fully earned. the 1 major comments →

arxiv 2509.13498 v1 pith:G6B3PQ6T submitted 2025-09-16 astro-ph.CO

Using Neural Emulators and Hamiltonian Monte Carlo to constrain the Epoch of Reionization's History with the Lyα Forest Power Spectrum

Diego Gonz\'alez-Hern\'andez , Caitlin Doughty , Molly Wolfson , Joseph F. Hennawi , Zhenyu Jin This is my paper
classification astro-ph.CO PACS 98.80.-k
keywords Lyman-alpha forestEpoch of Reionizationneural emulatorsHamiltonian Monte Carlopower spectrumcovariance matrixparameter inferenceintergalactic medium
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper builds a new inference pipeline for the redshift-5 Lyman-alpha forest. It trains two neural networks: one emulates the mean 1D flux power spectrum of the forest, and one emulates the model-dependent covariance matrix that describes the spectrum's noise and bin correlations. Because the emulators are differentiable, they are coupled to Hamiltonian Monte Carlo sampling, which explores the five parameters describing reionization history (midpoint, duration, asymmetry, heat injection, and mean flux) far more efficiently than standard MCMC. Applied to 100 mock observations drawn from the same simulation suite, the pipeline recovers the true parameters, with posteriors that are slightly overconfident. The authors are explicit that the low resolution of the simulations suppresses small-scale power and that simplified thermal and UV-background treatments mean the method cannot yet be applied to real data.

Core claim

The central claim is that a differentiable neural emulator of the 1D Lyman-alpha flux power spectrum, combined with a neural emulator of the covariance matrix, makes Bayesian inference of reionization history computationally feasible. The covariance matrix emulator, constructed by predicting Cholesky factors and exponentiating the diagonal, is described as the first neural emulator of Lyman-alpha forest covariance matrices. The paper validates the approach by generating 501 simulations with user-defined reionization histories, training two emulators (median power-spectrum error 0.15%, median covariance error 0.42%), and running HMC on 100 mock observations; the true parameters are recovered

What carries the argument

Two fully connected neural networks trained on 501 low-resolution hydrodynamic simulations, each with five parameters (reionization midpoint, duration, asymmetry, heat-injection temperature, and mean flux). The power-spectrum network directly outputs the log of the mean flux power at 19 wavenumber bins; the covariance network outputs Cholesky factors, rearranged and exponentiated on the diagonal to guarantee symmetric positive-definite matrices. Automatic differentiation through these emulators supplies the gradients needed by Hamiltonian Monte Carlo, turning a previously intractable forward model into a fast differentiable surrogate.

Load-bearing premise

The load-bearing premise is that the mock observations used to test the method are a faithful stand-in for real Lyman-alpha forest measurements; in fact they are generated by the same low-resolution forward model that produced the emulator's training data, so the recovered 'true' parameters are only true inside that simulated world.

What would settle it

Run the same Hamiltonian Monte Carlo inference on mock observations produced by a higher-resolution simulation (or a radiation-hydrodynamic simulation with full radiative transfer) that was not part of the training set, and check whether the true parameters fall inside the posterior at the claimed frequency. The coverage plot falling far below the ideal line would falsify the claim of reliable recovery. Also check whether the 100 mocks used in Section 4.2 are drawn from the held-out 20% test split; if they come from the training set, the recovery results do not demonstrate generalization.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • True reionization parameters are recovered from mock observations, indicating the method is ready for application to higher-fidelity simulations.
  • The first neural emulation of Lyman-alpha forest covariance matrices opens the door to model-dependent noise prescriptions in future analyses.
  • Training on only 100 simulations instead of 501 still gives competitive posteriors, meaning high-resolution suites, where each simulation is expensive, may be sufficient.
  • The authors state the current low-resolution models suppress small-scale power and must be upgraded to higher resolution before real data can be analyzed.
  • Posteriors are systematically slightly overconfident, attributed mainly to the multivariate Gaussian likelihood assumption rather than emulator error.

Where Pith is reading between the lines

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

  • Because the mock observations come from the same forward model that produced the training set, the 'true' parameters are only true inside that model world; out-of-sample validation against a different, higher-fidelity forward model would be a stronger test.
  • The overconfidence seen in the coverage test is a warning that posterior widths, not just the central values, may not be trustworthy when the Gaussian likelihood is used; simulation-based inference is a natural remedy.
  • Model-dependent covariance matrices encode information about the reionization history themselves, so regions of parameter space that are degenerate in the mean power spectrum may still be distinguishable through the noise structure.
  • Transfer learning from low-resolution to high-resolution emulators could cut the number of expensive high-resolution simulations needed, since the small-scale suppression is systematic.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 5 minor

Summary. The paper presents a JAX-based inference framework that combines differentiable neural emulators of the Lyα forest mean 1D power spectrum and its model-dependent covariance matrix with Hamiltonian Monte Carlo (HMC). The authors construct a dataset of 501 low-resolution AMBER/Nyx simulations, yielding 4509 models after varying the mean flux, and train two emulators: a fully connected network for the power spectrum and a Cholesky-decomposed network for the covariance matrix. Emulator accuracies are reported on a held-out test set (median MAPE 0.15% for the power spectrum; median 0.0042 for the covariance error metric). The framework is validated by performing inference on 100 mock observations and constructing a coverage plot, which shows slightly overconfident posteriors. A reduced training set of 100 simulations still yields comparable inference performance. The authors claim the first neural emulator of Lyα forest covariance matrices and emphasize that the current low-resolution simulations prevent application to real data.

Significance. This is a genuine methodological contribution: the differentiable emulation of model-dependent covariance matrices via Cholesky decomposition, combined with HMC, offers a practical route to efficient Lyα forest parameter inference with self-consistent noise treatment. The paper's strengths include a careful emulator validation on a held-out test set, a formal coverage test for the posteriors, a publicly released reusable pipeline (DNE+HMC), and an explicit, honest discussion of the forward model's limitations (low resolution, simplified heat injection, no UVB fluctuations). If the inference test is confirmed to be out-of-sample, the central claim that the framework reliably recovers the reionization-history parameters from mock observations is well supported. The demonstration that a 100-simulation training set still works is relevant for future high-resolution applications where large simulation suites are prohibitive.

major comments (1)
  1. [§4.2 / §4.2.1] The 100 mock observations used for the inference test are described only as 'randomly selected' (§4.2), with no statement that they are drawn exclusively from the held-out test simulations. The dataset split (70/10/20) was performed on the 501 simulations, not on the 4509 models (§4), so a mock associated with a training/validation simulation would have its noiseless mean power spectrum and covariance matrix seen by the emulators during training. In that case the coverage plot in Fig. 15 would reflect in-sample memorization rather than out-of-sample generalization, and the abstract's claim that 'the true parameters are reliably recovered' would not be demonstrated. Please clarify the provenance of the 100 mocks; if they are not test-only, re-run the inference test using mocks from the test split and report the resulting coverage.
minor comments (5)
  1. [§5.2] There is a duplicated paragraph: the text beginning 'After performing hyper-parameter tuning, both the mean power spectrum and covariance matrix emulators were able to achieve a sub-percent and sub-five-percent error...' appears twice in succession. One copy should be removed.
  2. [Figure 9] The caption says 'the solid dashed red line indicating the median' - this is contradictory. It should read 'the solid red line' or 'the dashed red line'.
  3. [Abstract / §1] The abstract states that the emulators 'achieve sub-percent errors across relevant scales', but the covariance matrix emulator has sub-5% errors on the diagonal (Fig. 12) and the metric in Eq. (17) has a median of 0.0042 which is sub-percent in that metric. Please qualify the claim for the covariance emulator.
  4. [§6 / §1] The conclusion says 'sub-2% accuracy across all relevant scales' for the power spectrum emulator, while the abstract says 'sub-percent'. These are consistent with the reported median and 95% intervals, but the wording could be unified for clarity.
  5. [§1] The claim 'no previous studies have attempted neural network emulation of Lyα forest covariance matrices' is a strong novelty statement. Please verify this against the most recent literature (e.g., covariance emulation for other Lyα statistics or for the same statistic in related contexts) to avoid overclaiming, or soften the wording.

Circularity Check

0 steps flagged

No demonstrated circularity; closed-loop mock validation and minor non-load-bearing self-citations, with a mock-provenance caveat.

full rationale

The paper's derivation chain is self-contained rather than circular. The emulators are trained on summary statistics from 501 AMBER/Nyx simulations; the mock observations are generated by the same forward model (Section 2.4.2), and the 'true' parameters are the known inputs of those simulations. This is a standard forward-model closure test: the emulator maps parameters to statistics, and HMC inverts that map; no equation defines the predicted parameters as the training inputs, so there is no self-definitional reduction. Emulator accuracy is evaluated on a held-out test split (Section 4.1), so the sub-percent errors are genuine out-of-sample measurements. The coverage test (Section 4.2.1) is a real statistical check; the authors find slight overconfidence and attribute it to the Gaussian likelihood, which is a testable finding. The self-citations (Hennawi et al. 2025 for coverage formalism, Jin et al. 2025 for error propagation, Doughty et al. 2025 for simulation setup) are methodological references, not load-bearing justifications of the central inference claim. One caveat: Section 4.2 states only that 100 mock observations were 'randomly selected,' without stating that they come from the test split; if they overlapped the training/validation simulations, the recovery would be in-sample. This is an important provenance omission and a validation weakness, but it is not a demonstrated circular reduction because no equation forces the recovered parameters to equal training labels. The paper itself disclaims applicability to real data due to low resolution (Section 5.1), so it does not overclaim external predictive success.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The five physical model parameters (z_mid, Delta_z, A_z, Delta_Tre, <F>) are the inference targets, not fitted constants, so they are not listed as free parameters. The free parameters listed here are the computational and modeling choices that the central claim rests on: neural network hyperparameters tuned on validation data, a fixed photon mean free path, a deliberately low resolution, and a fixed noise realization. The assumptions cover the mathematical machinery (Cholesky, Gaussian likelihood), the physical forward model (AMBER, heat injection), and the self-referential validation strategy. No new physical entities are introduced.

free parameters (5)
  • Power spectrum emulator hyperparameters = n_h=5, n_u=[8,12,16,20,22], lr=0.0152, n_e=750
    Selected by Optuna hyperparameter tuning on the validation set; control network architecture and training (Table A1).
  • Covariance matrix emulator hyperparameters = n_h=5, n_u=[25,25,25,50,50], lr=0.0078, n_e=1250
    Selected by Optuna hyperparameter tuning on the validation set (Table A1).
  • Photon mean free path = 3 h^-1 Mpc
    Assumed following Doughty et al. (2025), used in AMBER's ionizing background calculation (Section 2.1); not varied in inference.
  • Simulation resolution and box size = 256^3 voxels in L_box=20 h^-1 Mpc
    Chosen for computational feasibility; causes small-scale power suppression (Sections 2.2, 2.4.1).
  • Fixed noise realization = One N_skwrs x length Gaussian realization, sigma_N=1/SNR_deltav
    A single noise realization applied to all models to remove inter-model stochasticity; affects covariance estimation (Section 2.3).
axioms (6)
  • standard math Covariance matrices are symmetric positive definite and can be represented via Cholesky decomposition
    Used to build the covariance emulator while guaranteeing positive definiteness (Section 3.1.2, Eq. 13).
  • domain assumption Multivariate Gaussian likelihood for the mean power spectrum
    Assumed in Eq. 12; the coverage test shows slight overconfidence, indicating the assumption is only approximately valid at z=5 (Sections 3, 4.2.1, 5.3).
  • domain assumption AMBER's abundance-matching scheme produces physically plausible reionization fields for user-defined histories
    The paper relies on AMBER to generate ionization fields controlled by z_mid, Delta_z, A_z (Section 2.1).
  • ad hoc to paper Simplified heat injection model: T_post-reion = x_HI,pre * max(Delta_Tre - T_pre, 0) + T_pre
    Heating upon reionization is parametrized by a single maximum temperature Delta_Tre (Eq. 5); the paper acknowledges this does not capture density-dependent heating (Section 5.1).
  • domain assumption Low-resolution simulations capture the parameter dependence of the Lyα forest statistics
    The emulators are trained on these simulations; Figure 4 shows the models underpredict small-scale power, so the emulated relation may not match the real universe.
  • domain assumption Uniform prior over the convex hull of the simulation dataset
    The prior is uniform over the convex hull enclosing the dataset, after filtering by the McGreer et al. (2015) 3-sigma constraint (Sections 2.5, 3).

pith-pipeline@v1.3.0-alltime-deepseek · 28200 in / 14208 out tokens · 141950 ms · 2026-08-04T16:28:03.054643+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Using Neural Emulators and Hamiltonian Monte Carlo to constrain the Epoch of Reionization's History with the Ly$\alpha$ Forest Power Spectrum." pith.science (2026). https://pith.science/paper/G6B3PQ6T

@misc{pith2026250913498,
  author       = {Pith},
  title        = {Pith review of: Using Neural Emulators and Hamiltonian Monte Carlo to constrain the Epoch of Reionization's History with the Ly$\alpha$ Forest Power Spectrum},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G6B3PQ6T}},
  note         = {Machine review of arXiv:2509.13498}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The Lyman-alpha (Ly$\alpha$) forest at $z \sim 5$ offers a primary probe to constrain the history of the Epoch of Reionization (EoR), retaining thermal and ionization signatures imprinted by the reionization process. In this work, we present a new inference framework based on JAX that combines forward-modeled Ly$\alpha$ forest observables with differentiable neural emulators and Hamiltonian Monte Carlo (HMC). We construct a dataset of 501 low-resolution simulations generated with user-defined reionization histories and compute a set of 1D Ly$\alpha$ power spectra and model-dependent covariance matrices. We then train two independent neural emulators that achieve sub-percent errors across relevant scales and combine them with HMC to efficiently perform parameter estimation. We validate this framework by applying it to a suite of mock observations, demonstrating that the true parameters are reliably recovered. While this work is limited by the low resolution of the simulations used, our results highlight the potential of this method for inferring the reionization history from high-redshift Ly$\alpha$ forest measurements. Future improvements in our reionization models will further enhance its ability to extract constraints from observational datasets.

Figures

Figures reproduced from arXiv: 2509.13498 by Caitlin Doughty, Diego Gonz\'alez-Hern\'andez, Joseph F. Hennawi, Molly Wolfson, Zhenyu Jin.

Figure 1
Figure 1. Figure 1: Examples of different mass-weighted reionization histories gener￾ated with AMBER’s analytical model. Top Panel: demonstrates the effect of varying the duration Δ𝑧, while fixing 𝑧mid = 8.0 and 𝐴𝑧 = 1.0. Bottom Panel: shows the influence of the asymmetry parameter 𝐴𝑧 , with fixed 𝑧mid = 8.0 and Δ𝑧 = 4.0. the simulation cells by radiation intensity and maps them to the cu￾mulative ionized mass fraction, 𝑥HII,… view at source ↗
Figure 2
Figure 2. Figure 2: Slices of 88.5 km/s at 𝑧 = 5.00 of 20 ℎ −1 cMpc from the simulations. The left column shows the gas overdensity field (Δ), while the center and right columns show the temperature (𝑇) and reionization redshift (𝑧re) fields, respectively. Since the simulations have initial conditions with the same seed, the overdensity structures remain the same, and the differences in the temperature and reionization redshi… view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: An subset of 200 randomly selected ⟨𝑃Ly𝛼 ⟩ from our model com￾pared to the measurement from Boera et al. (2019) at 𝑧 = 5.0. These models match the ⟨𝐹⟩ from the observational data. As shown, the suppression of power at smaller scales is present in all the models we compute, and is caused by the low resolution of our simulations (see Section 2.4.1). within their respective ranges shown in [PITH_FULL_IMAGE:f… view at source ↗
Figure 6
Figure 6. Figure 6: The 501 volume weighted reionization histories used to create our dataset of low resolution simulations (shown in black). These models were selected to be in agreement within 3𝜎 with the measurements from McGreer et al. (2015) (shown in red). Highlighted in cyan and magenta are the reionization histories corresponding to Models 1 and 2, which are the to the same models as the top and bottom slices in [PIT… view at source ↗
Figure 5
Figure 5. Figure 5: Examples of a covariance matrices (shown as a correlation matrices) at 𝑧 = 5.0. The covariance matrix at the top corresponds to a model with 𝑧mid = 6.4, Δ𝑧 = 1.0, 𝐴𝑧 = 12.4, Δ𝑇re = 29000, ⟨𝐹⟩ = 0.1496. The covariance matrix at the bottom corresponds to a model with 𝑧mid = 8.7, Δ𝑧 = 16.0, 𝐴𝑧 = 11.5, Δ𝑇re = 36000, ⟨𝐹⟩ = 0.1156. 3 PARAMETER INFERENCE To quantitatively constrain the parameters 𝜃 in our model, … view at source ↗
Figure 7
Figure 7. Figure 7: Diagram of the emulators. The top portion shows the emulator for the mean power spectrum, where a fully connected, feed-forward neural network is used to emulate the power spectrum directly from the input parameters 𝜃. The bottom of the diagram shows the emulator for the covariance matrix. In this case, a neural network predicts 𝐿flat from 𝜃. 𝐿flat is then rearranged into 𝐿, from which an emulated covarian… view at source ↗
Figure 8
Figure 8. Figure 8: Example of an emulated power spectrum in the test dataset, with an error in the 50th percentile. The emulated power spectrum is shown in the solid black line, while the true power spectrum is shown in the light blue dashed line. The bottom panel shows the absolute relative error as a function of 𝑘. mean absolute percentage error (MAPE) as our main metric. For a single power spectrum, the MAPE is calculated… view at source ↗
Figure 9
Figure 9. Figure 9: Distribution of the power spectrum emulator’s relative errors as a function of 𝑘, evaluated on the test set. The dashed red line represents the median emulator error across all test models, while the dark and light red shaded regions indicate the central 68% and 95% intervals, respectively. For comparison, the dashed black line shows the median of the expected 1𝜎 observational uncertainty, computed as the … view at source ↗
Figure 10
Figure 10. Figure 10: Example of an emulated covariance matrix in the test dataset, with an error in the 50th percentile. The true covariance matrix is shown in the left, while the emulated covariance matrix is shown in the middle. Both covariance matrices are shown as correlation matrices as described by Equation 9. The heat map on the right shows the error as defined by Equation 17 for this example. -2.34 -1.71 -1.23 -0.74 l… view at source ↗
Figure 11
Figure 11. Figure 11: Median of the Σ𝑝Ly𝛼 emulator’s errors as defined by Equation 17, calculated with the test set. As observed, the emulator incurs in a larger error for elements in the covariance matrix that correspond to the small scales. ¡2.25 ¡2.00 ¡1.75 ¡1.50 ¡1.25 ¡1.00 ¡0.75 log10 (k) 0.00 0.02 0.04 0.06 0.08 j^¾ ¡ ¾j ¾ 95% interval 68% interval Median [PITH_FULL_IMAGE:figures/full_fig_p011_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Distribution of the Σ𝑝Ly𝛼 emulator’s errors on the diagonal of the covariance matrices as a function of 𝑘, calculated with the test set. The black solid line represents the median of this quantity, while the dark red and light red shaded regions contain the regions with 68% and 95% of all errors. This figure shows that the covariance matrix emulator incurs in a sub-5% error across all scales in the majori… view at source ↗
Figure 13
Figure 13. Figure 13: The posterior distributions obtained for two randomly selected mock observations. The corner plots show the posterior distributions (in black) overlaid on top of the priors (in light blue). The true parameters of each mock observation are plotted in red. The median of the marginalized posteriors for each parameter is shown on top of the 1D histograms, along with the uncertainty of their estimation (given … view at source ↗
Figure 14
Figure 14. Figure 14: The inferred model plots for both posterior distributions in [PITH_FULL_IMAGE:figures/full_fig_p012_14.png] view at source ↗
Figure 17
Figure 17. Figure 17: Distribution of the errors on the diagonal of the covariance matri￾ces as a function of 𝑘 obtained with Σ𝑝Ly𝛼 emulator trained on the reduced dataset. The black solid line represents the median of this quantity, while the dark red and light red shaded regions contain the regions with 68% and 95% of all errors. This figure shows that the covariance matrix emulator incurs in a small error across most scales… view at source ↗
Figure 16
Figure 16. Figure 16: Distribution of the power spectrum emulator’s errors as a function of 𝑘, calculated on the test set when trained with only 100 simulations. The black solid line shows the median of the absolute percentage error, while the dark red and light red shaded regions indicate the 68% and 95% intervals, respectively. The dashed black line represents the median observational un￾certainty, estimated from the square … view at source ↗
Figure 19
Figure 19. Figure 19: The inferred model plot for the posterior distribution in [PITH_FULL_IMAGE:figures/full_fig_p014_19.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Machine Learning Techniques for Astrophysics and Cosmology: Lyman-$\alpha$ forest

    astro-ph.CO 2026-05 unverdicted novelty 2.0

    Review of machine learning applications for analyzing Lyman-alpha forest observations to probe cosmology, reionization, and dark matter.

Reference graph

Works this paper leans on

131 extracted references · 17 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

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

    2019

  2. [2]

    S., Bell J

    Almgren A. S., Bell J. B., Lijewski M. J., Luki \'c Z., Van Andel E., 2013, @doi [ ] 10.1088/0004-637X/765/1/39 , https://ui.adsabs.harvard.edu/abs/2013ApJ...765...39A 765, 39

    doi:10.1088/0004-637x/765/1/39 2013

  3. [3]

    Ataei M., Salehipour H., 2024, @doi [Computer Physics Communications] 10.1016/j.cpc.2024.109187 , https://ui.adsabs.harvard.edu/abs/2024CoPhC.30009187A 300, 109187

    arXiv 2024

  4. [4]

    Ba \ n ados E., et al., 2018, @doi [ ] 10.1038/nature25180 , https://ui.adsabs.harvard.edu/abs/2018Natur.553..473B 553, 473

    doi:10.1038/nature25180 2018

  5. [5]

    Battaglia N., Trac H., Cen R., Loeb A., 2013, @doi [ ] 10.1088/0004-637X/776/2/81 , https://ui.adsabs.harvard.edu/abs/2013ApJ...776...81B 776, 81

    doi:10.1088/0004-637x/776/2/81 2013

  6. [6]

    Bauer A., Springel V., Vogelsberger M., Genel S., Torrey P., Sijacki D., Nelson D., Hernquist L., 2015, @doi [ ] 10.1093/mnras/stv1893 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.453.3593B 453, 3593

    doi:10.1093/mnras/stv1893 2015

  7. [7]

    D., Bolton J

    Becker G. D., Bolton J. S., Madau P., Pettini M., Ryan-Weber E. V., Venemans B. P., 2015, @doi [ ] 10.1093/mnras/stu2646 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.447.3402B 447, 3402

    doi:10.1093/mnras/stu2646 2015

  8. [8]

    Bingham E., et al., 2019, J. Mach. Learn. Res., 20, 28:1

    2019

  9. [9]

    V., Viel M., Verde L., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18245.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.413.1717B 413, 1717

    Bird S., Peiris H. V., Viel M., Verde L., 2011, @doi [ ] 10.1111/j.1365-2966.2011.18245.x , https://ui.adsabs.harvard.edu/abs/2011MNRAS.413.1717B 413, 1717

    arXiv 2011

  10. [10]

    K., Peiris H

    Bird S., Rogers K. K., Peiris H. V., Verde L., Font-Ribera A., Pontzen A., 2019, @doi [ ] 10.1088/1475-7516/2019/02/050 , https://ui.adsabs.harvard.edu/abs/2019JCAP...02..050B 2019, 050

    doi:10.1088/1475-7516/2019/02/050 2019

  11. [11]

    Bird S., Ni Y., Di Matteo T., Croft R., Feng Y., Chen N., 2022, @doi [ ] 10.1093/mnras/stac648 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.512.3703B 512, 3703

    doi:10.1093/mnras/stac648 2022

  12. [12]

    Bird S., et al., 2023, @doi [ ] 10.1088/1475-7516/2023/10/037 , https://ui.adsabs.harvard.edu/abs/2023JCAP...10..037B 2023, 037

    work page doi:10.1088/1475-7516/2023/10/037 2023

  13. [13]

    D., Bolton J

    Boera E., Becker G. D., Bolton J. S., Nasir F., 2019, @doi [ ] 10.3847/1538-4357/aafee4 , https://ui.adsabs.harvard.edu/abs/2019ApJ...872..101B 872, 101

    doi:10.3847/1538-4357/aafee4 2019

  14. [14]

    R., Cole S., Efstathiou G., Kaiser N., 1991, @doi [ ] 10.1086/170520 , https://ui.adsabs.harvard.edu/abs/1991ApJ...379..440B 379, 440

    Bond J. R., Cole S., Efstathiou G., Kaiser N., 1991, @doi [ ] 10.1086/170520 , https://ui.adsabs.harvard.edu/abs/1991ApJ...379..440B 379, 440

    doi:10.1086/170520 1991

  15. [15]

    Bosman S. E. I., Fan X., Jiang L., Reed S., Matsuoka Y., Becker G., Haehnelt M., 2018, @doi [ ] 10.1093/mnras/sty1344 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.479.1055B 479, 1055

    doi:10.1093/mnras/sty1344 2018

  16. [16]

    Bosman S. E. I., et al., 2022, @doi [ ] 10.1093/mnras/stac1046 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.514...55B 514, 55

    doi:10.1093/mnras/stac1046 2022

  17. [17]

    R., Colombi S., Hivon E., Juszkiewicz R., 1995, @doi [ ] 10.48550/arXiv.astro-ph/9406013 , https://ui.adsabs.harvard.edu/abs/1995A&A...296..575B 296, 575

    Bouchet F. R., Colombi S., Hivon E., Juszkiewicz R., 1995, @doi [ ] 10.48550/arXiv.astro-ph/9406013 , https://ui.adsabs.harvard.edu/abs/1995A&A...296..575B 296, 575

    Pith/arXiv arXiv doi:10.48550/arxiv.astro-ph/9406013 1995

  18. [18]

    Bradbury J., et al., 2018, JAX : composable transformations of P ython+ N um P y programs, http://github.com/google/jax

    2018

  19. [19]

    Cabayol-Garcia L., Chaves-Montero J., Font-Ribera A., Pedersen C., 2023, @doi [ ] 10.1093/mnras/stad2512 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.525.3499C 525, 3499

    doi:10.1093/mnras/stad2512 2023

  20. [20]

    Campagne J.-E., et al., 2023, @doi [The Open Journal of Astrophysics] 10.21105/astro.2302.05163 , https://ui.adsabs.harvard.edu/abs/2023OJAp....6E..15C 6, 15

    Pith/arXiv arXiv 2023

  21. [21]

    C., 2025, @doi [ ] 10.1093/mnras/staf663 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.3220C 539, 3220

    Carrion K., Spurio Mancini A., Piras D., Hidalgo J. C., 2025, @doi [ ] 10.1093/mnras/staf663 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.3220C 539, 3220

    doi:10.1093/mnras/staf663 2025

  22. [22]

    Chabanier S., et al., 2019, @doi [ ] 10.1088/1475-7516/2019/07/017 , https://ui.adsabs.harvard.edu/abs/2019JCAP...07..017C 2019, 017

    doi:10.1088/1475-7516/2019/07/017 2019

  23. [23]

    Chabanier S., et al., 2023, @doi [ ] 10.1093/mnras/stac3294 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.518.3754C 518, 3754

    doi:10.1093/mnras/stac3294 2023

  24. [24]

    arXiv:2406.06718

    Citrin J., et al., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2406.06718 , https://ui.adsabs.harvard.edu/abs/2024arXiv240606718C p. arXiv:2406.06718

    Pith/arXiv arXiv doi:10.48550/arxiv.2406.06718 2024

  25. [25]

    Cranmer K., Brehmer J., Louppe G., 2020, @doi [Proceedings of the National Academy of Science] 10.1073/pnas.1912789117 , https://ui.adsabs.harvard.edu/abs/2020PNAS..11730055C 117, 30055

    doi:10.1073/pnas.1912789117 2020

  26. [26]

    B., Trac H., Fuller S., Upton Sanderbeck P

    D'Aloisio A., McQuinn M., Maupin O., Davies F. B., Trac H., Fuller S., Upton Sanderbeck P. R., 2019, @doi [ ] 10.3847/1538-4357/ab0d83 , https://ui.adsabs.harvard.edu/abs/2019ApJ...874..154D 874, 154

    doi:10.3847/1538-4357/ab0d83 2019

  27. [27]

    D'Odorico V., et al., 2023, @doi [ ] 10.1093/mnras/stad1468 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.523.1399D 523, 1399

    doi:10.1093/mnras/stad1468 2023

  28. [28]

    B., 2015, PhD thesis, University of California, Los Angeles

    Davies F. B., 2015, PhD thesis, University of California, Los Angeles

    2015

  29. [29]

    B., Furlanetto S

    Davies F. B., Furlanetto S. R., 2016, @doi [ ] 10.1093/mnras/stw931 , https://ui.adsabs.harvard.edu/abs/2016MNRAS.460.1328D 460, 1328

    doi:10.1093/mnras/stw931 2016

  30. [30]

    B., et al., 2018, @doi [ ] 10.3847/1538-4357/aad6dc , https://ui.adsabs.harvard.edu/abs/2018ApJ...864..142D 864, 142

    Davies F. B., et al., 2018, @doi [ ] 10.3847/1538-4357/aad6dc , https://ui.adsabs.harvard.edu/abs/2018ApJ...864..142D 864, 142

    doi:10.3847/1538-4357/aad6dc 2018

  31. [31]

    R., et al., 2017, @doi [ ] 10.1088/1538-3873/129/974/045001 , https://ui.adsabs.harvard.edu/abs/2017PASP..129d5001D 129, 045001

    DeBoer D. R., et al., 2017, @doi [ ] 10.1088/1538-3873/129/974/045001 , https://ui.adsabs.harvard.edu/abs/2017PASP..129d5001D 129, 045001

    doi:10.1088/1538-3873/129/974/045001 2017

  32. [32]

    C., Hennawi J

    Doughty C. C., Hennawi J. F., Davies F. B., Luki \'c Z., O \ n orbe J., 2023, @doi [ ] 10.1093/mnras/stad2549 , https://ui.adsabs.harvard.edu/abs/2023MNRAS.525.3790D 525, 3790

    doi:10.1093/mnras/stad2549 2023

  33. [33]

    C., Hennawi J

    Doughty C. C., Hennawi J. F., O \ n orbe J., Davies F. B., Luki \'c Z., 2025, @doi [ ] 10.1093/mnras/staf616 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.3027D 539, 3027

    work page doi:10.1093/mnras/staf616 2025

  34. [34]

    J., Roweth D., 1987, @doi [Physics Letters B] https://doi.org/10.1016/0370-2693(87)91197-X , 195, 216

    Duane S., Kennedy A., Pendleton B. J., Roweth D., 1987, @doi [Physics Letters B] https://doi.org/10.1016/0370-2693(87)91197-X , 195, 216

    doi:10.1016/0370-2693(87)91197-x 1987

  35. [35]

    B., Graziani L., Ciardi B., Feng Y., Kakiichi K., Di Matteo T., 2018, @doi [ ] 10.1093/mnras/sty272 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.1174E 476, 1174

    Eide M. B., Graziani L., Ciardi B., Feng Y., Kakiichi K., Di Matteo T., 2018, @doi [ ] 10.1093/mnras/sty272 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.476.1174E 476, 1174

    doi:10.1093/mnras/sty272 2018

  36. [36]

    B., Hennawi J

    Eilers A.-C., Davies F. B., Hennawi J. F., 2018, @doi [ ] 10.3847/1538-4357/aad4fd , https://ui.adsabs.harvard.edu/abs/2018ApJ...864...53E 864, 53

    doi:10.3847/1538-4357/aad4fd 2018

  37. [37]

    Fan X., et al., 2006, @doi [ ] 10.1086/504836 , https://ui.adsabs.harvard.edu/abs/2006AJ....132..117F 132, 117

    doi:10.1086/504836 2006

  38. [38]

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

    doi:10.1093/mnras/stw2123 2016

  39. [39]

    A., Ho M.-F., Bird S., 2022, @doi [ ] 10.1093/mnras/stac2435 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.517.3200F 517, 3200

    Fernandez M. A., Ho M.-F., Bird S., 2022, @doi [ ] 10.1093/mnras/stac2435 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.517.3200F 517, 3200

    work page doi:10.1093/mnras/stac2435 2022

  40. [40]

    D., Dav \'e R., Zackrisson E., 2018, @doi [ ] 10.1093/mnras/sty1949 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.480.2628F 480, 2628

    Finlator K., Keating L., Oppenheimer B. D., Dav \'e R., Zackrisson E., 2018, @doi [ ] 10.1093/mnras/sty1949 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.480.2628F 480, 2628

    doi:10.1093/mnras/sty1949 2018

  41. [41]

    R., Datta K

    Ghara R., Choudhury T. R., Datta K. K., 2015, @doi [ ] 10.1093/mnras/stu2512 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.447.1806G 447, 1806

    doi:10.1093/mnras/stu2512 2015

  42. [42]

    Y., 2014, @doi [ ] 10.1088/0004-637X/793/1/29 , https://ui.adsabs.harvard.edu/abs/2014ApJ...793...29G 793, 29

    Gnedin N. Y., 2014, @doi [ ] 10.1088/0004-637X/793/1/29 , https://ui.adsabs.harvard.edu/abs/2014ApJ...793...29G 793, 29

    doi:10.1088/0004-637x/793/1/29 2014

  43. [43]

    Y., Madau P., 2022, @doi [Living Reviews in Computational Astrophysics] 10.1007/s41115-022-00015-5 , https://ui.adsabs.harvard.edu/abs/2022LRCA....8....3G 8, 3

    Gnedin N. Y., Madau P., 2022, @doi [Living Reviews in Computational Astrophysics] 10.1007/s41115-022-00015-5 , https://ui.adsabs.harvard.edu/abs/2022LRCA....8....3G 8, 3

    doi:10.1007/s41115-022-00015-5 2022

  44. [44]

    Grand \'o n D., Sellentin E., 2022, @doi [The Open Journal of Astrophysics] 10.21105/astro.2205.11587 , https://ui.adsabs.harvard.edu/abs/2022OJAp....5E..12G 5, 12

    Pith/arXiv arXiv 2022

  45. [45]

    A., 2017, @doi [ ] 10.1093/mnras/stw3351 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.466.4239G 466, 4239

    Greig B., Mesinger A., Haiman Z., Simcoe R. A., 2017, @doi [ ] 10.1093/mnras/stw3351 , https://ui.adsabs.harvard.edu/abs/2017MNRAS.466.4239G 466, 4239

    doi:10.1093/mnras/stw3351 2017

  46. [46]

    Greig B., Mesinger A., Ba \ n ados E., 2019, @doi [ ] 10.1093/mnras/stz230 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.484.5094G 484, 5094

    doi:10.1093/mnras/stz230 2019

  47. [47]

    B., Wang F., Yang J., Hennawi J

    Greig B., Mesinger A., Davies F. B., Wang F., Yang J., Hennawi J. F., 2022, @doi [ ] 10.1093/mnras/stac825 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.512.5390G 512, 5390

    doi:10.1093/mnras/stac825 2022

  48. [48]

    A., Andreyevich Radul A., Siskind J

    Gunes Baydin A., Pearlmutter B. A., Andreyevich Radul A., Siskind J. M., 2015, @doi [arXiv e-prints] 10.48550/arXiv.1502.05767 , https://ui.adsabs.harvard.edu/abs/2015arXiv150205767G p. arXiv:1502.05767

    Pith/arXiv arXiv doi:10.48550/arxiv.1502.05767 2015

  49. [49]

    E., Peterson B

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

    doi:10.1086/148444 1965

  50. [50]

    Harrington P., Mustafa M., Dornfest M., Horowitz B., Luki \'c Z., 2022, @doi [ ] 10.3847/1538-4357/ac5faa , https://ui.adsabs.harvard.edu/abs/2022ApJ...929..160H 929, 160

    doi:10.3847/1538-4357/ac5faa 2022

  51. [51]

    Heek J., Levskaya A., Oliver A., Ritter M., Rondepierre B., Steiner A., van Z ee M., 2024, F lax: A neural network library and ecosystem for JAX , http://github.com/google/flax

    2024

  52. [52]

    E., et al., 2024, @doi [Science] 10.1126/science.adj0343 , https://ui.adsabs.harvard.edu/abs/2024Sci...384..890H 384, 890

    Heintz K. E., et al., 2024, @doi [Science] 10.1126/science.adj0343 , https://ui.adsabs.harvard.edu/abs/2024Sci...384..890H 384, 890

    doi:10.1126/science.adj0343 2024

  53. [53]

    J., Lawrence E., Wagner C., 2009, @doi [ ] 10.1088/0004-637X/705/1/156 , https://ui.adsabs.harvard.edu/abs/2009ApJ...705..156H 705, 156

    Heitmann K., Higdon D., White M., Habib S., Williams B. J., Lawrence E., Wagner C., 2009, @doi [ ] 10.1088/0004-637X/705/1/156 , https://ui.adsabs.harvard.edu/abs/2009ApJ...705..156H 705, 156

    doi:10.1088/0004-637x/705/1/156 2009

  54. [54]

    F., Kist T., Davies F

    Hennawi J. F., Kist T., Davies F. B., Tamanas J., 2025, @doi [ ] 10.1093/mnras/staf643 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.539.2621H 539, 2621

    doi:10.1093/mnras/staf643 2025

  55. [55]

    M., 2018, @doi [ ] 10.1093/mnras/stx2854 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.474.2173H 474, 2173

    Hirata C. M., 2018, @doi [ ] 10.1093/mnras/stx2854 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.474.2173H 474, 2173

    doi:10.1093/mnras/stx2854 2018

  56. [56]

    D., Gelman A., 2011, @doi [arXiv e-prints] 10.48550/arXiv.1111.4246 , https://ui.adsabs.harvard.edu/abs/2011arXiv1111.4246H p

    Hoffman M. D., Gelman A., 2011, @doi [arXiv e-prints] 10.48550/arXiv.1111.4246 , https://ui.adsabs.harvard.edu/abs/2011arXiv1111.4246H p. arXiv:1111.4246

    Pith/arXiv arXiv doi:10.48550/arxiv.1111.4246 2011

  57. [57]

    arXiv:2502.02294

    Horowitz B., Lukic Z., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2502.02294 , https://ui.adsabs.harvard.edu/abs/2025arXiv250202294H p. arXiv:2502.02294

    doi:10.48550/arxiv.2502.02294 2025

  58. [58]

    K., Feng Y., Ma C.-P., 2018, @doi [ ] 10.1093/mnras/sty1329 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.478.5063H 478, 5063

    Huang K.-W., Di Matteo T., Bhowmick A. K., Feng Y., Ma C.-P., 2018, @doi [ ] 10.1093/mnras/sty1329 , https://ui.adsabs.harvard.edu/abs/2018MNRAS.478.5063H 478, 5063

    doi:10.1093/mnras/sty1329 2018

  59. [59]

    Huang L., Croft R. A. C., Arora H., 2021, @doi [ ] 10.1093/mnras/stab2041 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.506.5212H 506, 5212

    doi:10.1093/mnras/stab2041 2021

  60. [60]

    Hutter A., Dayal P., Yepes G., Gottl \"o ber S., Legrand L., Ucci G., 2021, @doi [ ] 10.1093/mnras/stab602 , https://ui.adsabs.harvard.edu/abs/2021MNRAS.503.3698H 503, 3698

    doi:10.1093/mnras/stab602 2021

  61. [61]

    T., Mellema G., Pen U

    Iliev I. T., Mellema G., Pen U. L., Merz H., Shapiro P. R., Alvarez M. A., 2006, @doi [ ] 10.1111/j.1365-2966.2006.10502.x , https://ui.adsabs.harvard.edu/abs/2006MNRAS.369.1625I 369, 1625

    arXiv 2006

  62. [62]

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

    work page doi:10.3847/1538-4357/acfcb5 2023

  63. [63]

    D., Watkinson C

    Jennings W. D., Watkinson C. A., Abdalla F. B., McEwen J. D., 2019, @doi [ ] 10.1093/mnras/sty3168 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.483.2907J 483, 2907

    doi:10.1093/mnras/sty3168 2019

  64. [64]

    Jin X., et al., 2023, @doi [ ] 10.3847/1538-4357/aca678 , https://ui.adsabs.harvard.edu/abs/2023ApJ...942...59J 942, 59

    doi:10.3847/1538-4357/aca678 2023

  65. [65]

    F., Gonz \'a lez-Hern \'a ndez D., 2025, @doi [ ] 10.1093/mnras/stae2741 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.536.2277J 536, 2277

    Jin Z., Wolfson M., Hennawi J. F., Gonz \'a lez-Hern \'a ndez D., 2025, @doi [ ] 10.1093/mnras/stae2741 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.536.2277J 536, 2277

    work page doi:10.1093/mnras/stae2741 2025

  66. [66]

    Extreme Deconvolution Reimagined: Conditional Densities via Neural Networks and an Application in Quasar Classification

    Kang Y., Hennawi J. F., Schindler J.-T., Tamanas J., Nanni R., 2024, @doi [arXiv e-prints] 10.48550/arXiv.2412.03029 , https://ui.adsabs.harvard.edu/abs/2024arXiv241203029K p. arXiv:2412.03029

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2412.03029 2024

  67. [67]

    Kannan R., Garaldi E., Smith A., Pakmor R., Springel V., Vogelsberger M., Hernquist L., 2022, @doi [ ] 10.1093/mnras/stab3710 , https://ui.adsabs.harvard.edu/abs/2022MNRAS.511.4005K 511, 4005

    doi:10.1093/mnras/stab3710 2022

  68. [68]

    C., Bolton J

    Keating L. C., Bolton J. S., Cullen F., Haehnelt M. G., Puchwein E., Kulkarni G., 2024, @doi [ ] 10.1093/mnras/stae1530 , https://ui.adsabs.harvard.edu/abs/2024MNRAS.532.1646K 532, 1646

    doi:10.1093/mnras/stae1530 2024

  69. [69]

    arXiv:2504.07090

    Kern N., 2025, @doi [arXiv e-prints] 10.48550/arXiv.2504.07090 , https://ui.adsabs.harvard.edu/abs/2025arXiv250407090K p. arXiv:2504.07090

    Pith/arXiv arXiv doi:10.48550/arxiv.2504.07090 2025

  70. [70]

    F., Davies F

    Kist T., Hennawi J. F., Davies F. B., 2025a, @doi [arXiv e-prints] 10.48550/arXiv.2504.14746 , https://ui.adsabs.harvard.edu/abs/2025arXiv250414746K p. arXiv:2504.14746

    doi:10.48550/arxiv.2504.14746

  71. [71]

    F., Davies F

    Kist T., Hennawi J. F., Davies F. B., 2025b, @doi [ ] 10.1093/mnras/staf460 , https://ui.adsabs.harvard.edu/abs/2025MNRAS.538.2704K 538, 2704

    work page doi:10.1093/mnras/staf460

  72. [72]

    Kwan J., Heitmann K., Habib S., Padmanabhan N., Lawrence E., Finkel H., Frontiere N., Pope A., 2015, @doi [ ] 10.1088/0004-637X/810/1/35 , https://ui.adsabs.harvard.edu/abs/2015ApJ...810...35K 810, 35

    work page doi:10.1088/0004-637x/810/1/35 2015

  73. [73]

    Lacey C., Cole S., 1993, @doi [ ] 10.1093/mnras/262.3.627 , https://ui.adsabs.harvard.edu/abs/1993MNRAS.262..627L 262, 627

    doi:10.1093/mnras/262.3.627 1993

  74. [74]

    Lewis A., Challinor A., Lasenby A., 2000, @doi [ ] 10.1086/309179 , https://ui.adsabs.harvard.edu/abs/2000ApJ...538..473L 538, 473

    doi:10.1086/309179 2000

  75. [75]

    M., May M., 2015, @doi [ ] 10.1103/PhysRevD.91.063507 , https://ui.adsabs.harvard.edu/abs/2015PhRvD..91f3507L 91, 063507

    Liu J., Petri A., Haiman Z., Hui L., Kratochvil J. M., May M., 2015, @doi [ ] 10.1103/PhysRevD.91.063507 , https://ui.adsabs.harvard.edu/abs/2015PhRvD..91f3507L 91, 063507

    doi:10.1103/physrevd.91.063507 2015

  76. [76]

    Loeb A., Barkana R., 2001, @doi [ ] 10.1146/annurev.astro.39.1.19 , https://ui.adsabs.harvard.edu/abs/2001ARA&A..39...19L 39, 19

    doi:10.1146/annurev.astro.39.1.19 2001

  77. [77]

    W., Nugent P., White M., Meiksin A

    Luki \'c Z., Stark C. W., Nugent P., White M., Meiksin A. A., Almgren A., 2015, @doi [ ] 10.1093/mnras/stu2377 , https://ui.adsabs.harvard.edu/abs/2015MNRAS.446.3697L 446, 3697

    doi:10.1093/mnras/stu2377 2015

  78. [78]

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

    doi:10.1086/180695 1971

  79. [79]

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

    work page doi:10.1051/0004-6361/202450259 2024

  80. [80]

    A., Naidu R

    Mason C. A., Naidu R. P., Tacchella S., Leja J., 2019, @doi [ ] 10.1093/mnras/stz2291 , https://ui.adsabs.harvard.edu/abs/2019MNRAS.489.2669M 489, 2669

    doi:10.1093/mnras/stz2291 2019

Showing first 80 references.