Pith. sign in

REVIEW 3 major objections 4 references

Five independent dimensions capture the rest-frame optical light of typical galaxies, and they are not the ones stellar-population models treat as primary.

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 ↗

Rest-frame optical galaxy SEDs from a 16-parameter SPS model are captured by five disentangled VAE latents (mass, young stars, dust, soft/hard ionization); metallicity and age are not independent drivers.

T0 review reviewed 2026-07-12 challenge →

load-bearing objection Clean five-axis optical SED result inside pop-cosmos, with the usual synthetic-data caveat that does not kill the paper. the 3 major comments →

arxiv 2606.11308 v2 pith:2HYGZNQ3 submitted 2026-06-09 astro-ph.GA astro-ph.IM

pop-cosmos: Disentangling galaxy properties from observables using data-driven approaches

Benedict Van den Bussche , Sinan Deger , Hiranya V. Peiris , Stephen Thorp , Daniel J. Mortlock , Boris Leistedt , Anik Halder , Madalina N. Tudorache
show 1 more author
Gurjeet Jagwani
This is my paper
classification astro-ph.GA astro-ph.IM
keywords galaxy SEDsstellar population synthesisvariational autoencodermutual informationdust-SFR degeneracyemission-line diagnosticsgalaxy evolution
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 reading

A galaxy's optical spectrum is shaped by many physical processes that leave entangled fingerprints, so observers cannot easily tell which processes act on their own. This paper trains a disentangling autoencoder on noiseless rest-frame optical spectra drawn from a generative population model and finds that five latent dimensions are necessary and sufficient. Those dimensions map cleanly to stellar mass, recent star formation, dust attenuation, and two independent axes of gas ionization (soft and hard). Stellar metallicity and mass-weighted age do not appear as separate drivers; their effects are absorbed into the other five. Because each latent is tied to specific spectral features, the decomposition breaks the classic star-formation–dust–metallicity degeneracies that plague broadband photometry and yields a cleaner view of gas conditions than the usual emission-line ratio diagrams.

Core claim

Five independent latent dimensions suffice to describe the rest-frame optical spectral energy distributions of COSMOS2020-like galaxies. They correspond to stellar mass, recent star formation, dust attenuation, and soft and hard ionization potentials of the gas. Stellar metallicity and stellar age are not primary independent drivers; their spectral effects are distributed across the other dimensions rather than independently encoded.

What carries the argument

A β-variational autoencoder trained with annealing, whose five-dimensional latent space is interpreted by mutual information between latents, SPS parameters, and wavelength-by-wavelength spectral features (including conditional MI).

Load-bearing premise

The entire analysis rests on noiseless synthetic spectra generated by a population model calibrated only to broadband photometry, so the claimed dimensionality and physical axes are those of the model, not of real noisy galaxy spectra.

What would settle it

Apply the same β-VAE and mutual-information analysis to a large sample of real, high-resolution rest-frame optical spectra (for example SDSS or DESI) of mass-complete star-forming galaxies and check whether five latents still saturate reconstruction accuracy and map to the same five physical axes.

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

If this is right

  • Optical photometry alone cannot isolate stellar metallicity or mass-weighted age as independent parameters; those quantities must be constrained by other wavelengths or by external priors.
  • Two latent dimensions that capture soft and hard ionization lines replace traditional BPT-style line ratios for diagnosing gas-phase metallicity and ionization state in typical star-forming galaxies.
  • The five latents supply low-dimensional, physically interpretable summary statistics for simulation-based inference of galaxy properties.
  • Extending the same analysis into the near- and mid-infrared is expected to reveal additional independent degrees of freedom associated with AGN and older stellar populations.

Where Pith is reading between the lines

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

  • If the same five axes reappear in real spectra, survey pipelines could replace multi-parameter SPS fits with a five-dimensional latent encoding for photo-z and SED classification.
  • The result suggests that many of the sixteen SPS parameters used by modern codes are redundant for optical data, so future SPS models may be re-parametrized around the five independent drivers.
  • A natural next test is whether the soft/hard ionization latents remain orthogonal once weak AGN or shock contributions are allowed in the optical.
Share X Bluesky LinkedIn Reddit HN

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

3 major / 0 minor

Summary. The paper trains a β-VAE on noiseless rest-frame optical SEDs (2425–7327 Å) drawn from the pop-cosmos generative model and argues that five disentangled latent dimensions are necessary and sufficient to describe those SEDs. Mutual information maps the latents to stellar mass (z4), recent star formation (z0), dust attenuation (z2), and soft/hard ionization (z1, z3); stellar metallicity and mass-weighted age are not recovered as independent drivers. Conditional MI spectra, latent traversals, and comparisons to broadband colours and BPT line ratios are used to show that the latents isolate spectral features that photometry blends, and that the two ionization latents separate physical regimes more cleanly than standard line-ratio diagnostics for typical star-forming galaxies.

Significance. If the five-dimensional optical manifold and its physical mapping hold, the work supplies a concrete, information-theoretic compression of SPS parameter space that is useful both as a diagnostic and as a low-dimensional summary for simulation-based inference. Strengths include a quantitative disentanglement check (latent–latent MI matrix), wavelength-resolved conditional MI, controlled latent traversals, an explicit ablation of latent dimensionality, and a direct comparison against BPT axes. The analysis is carefully executed inside the synthetic catalogue and is a natural next step for the pop-cosmos programme.

major comments (3)
  1. The central claim that five independent dimensions 'suffice' for the rest-frame optical SED (abstract, §4 opening paragraph, conclusions) is established only for noiseless FSPS/CLOUDY SEDs generated from the same 16-parameter SPS model whose population distribution was calibrated to broadband COSMOS2020 photometry (§2). pop-cosmos was never constrained by optical continuum shape or emission-line ratios at the resolution used here; the recovered rank and the statement that stellar metallicity and age are not primary drivers are therefore statements about the effective rank of that particular forward map under the adopted SFH, dust, and CLOUDY grids. The manuscript should state this scope limitation explicitly in the abstract and conclusions, and either (i) test the five-axis claim on real spectra (or spectra with realistic noise and continuum residuals) or (ii) reframe the result as a pro
  2. §4 (dimensionality choice) and Appendix A: the decision that five latents are necessary and sufficient rests on a reconstruction-error ablation (six dimensions give no 99th-percentile gain; four raise fractional error by >2%) plus latent–latent MI < 0.15. That is necessary but not sufficient for the physical claim. The paper should report (a) whether a four-latent model still recovers the same physical axes (or collapses soft/hard ionization), (b) sensitivity of the MI–physics mapping to the β-annealing schedule and to α = 10 000, and (c) a quantitative comparison against a linear baseline (e.g. PCA on the same SEDs) so that the non-linear gain is measured rather than asserted.
  3. §5 and Fig. 8–9: the claim that the two ionization latents recover gas conditions 'more cleanly than the line-ratio diagnostics in standard use' is only partially supported. The BPT comparison is restricted to the pure star-forming sequence of a model that omits optical AGN and shocks by construction (§5). The three regimes identified in the z1–z3 plane are visually compelling but lack a quantitative metric (e.g. mutual information of (z1, z3) vs. (Zgas, Ugas) relative to standard line ratios, or a classification purity against held-out SPS labels). Without that, the superiority claim over BPT remains qualitative.

Circularity Check

2 steps flagged

Five-DoF optical-SED claim is the effective rank of the authors' own pop-cosmos SPS mapping under its photometric prior, not an independent measurement of real spectra.

specific steps
  1. self citation load bearing [§2 (Sample Generation), validation paragraph; also Abstract and §6 Conclusions]
    "Our analysis relies on pop-cosmos having learned a robust connection between SPS properties and the SEDs they produce. This model has been extensively validated: in addition to stringent tests against COSMOS2020 data, it reproduces well-studied galaxy evolution trends. These include the star-forming main sequence, the mass-metallicity relation, and the fundamental metallicity relation (Thorp et al. 2025b). ... Deger et al. (2026) ... Halder et al. (2026) ... Since each SED follows deterministically from its SPS parameters, validating their population distribution validates the resulting SED po"

    The entire training set consists of noiseless rest-frame optical SEDs generated by applying FSPS/CLOUDY to SPS parameters drawn from pop-cosmos (Alsing/Thorp/Deger et al., overlapping co-authors). The claim that five independent dimensions 'suffice' for the rest-frame optical SED of galaxies therefore reduces to a statement about the effective rank of that self-generated mapping under the photometric prior. The only justification offered for treating the result as a property of real galaxies is the self-cited photometric and demographic validations of pop-cosmos, which do not test spectral dimensionality or line-ratio structure at the resolution of the VAE inputs.

  2. self citation load bearing [Abstract; §4 opening; §6 Conclusions]
    "Leveraging the pop-cosmos generative galaxy population model, we investigate how many independent degrees of freedom the rest-frame optical SED contains. ... We find that five independent dimensions suffice, corresponding to stellar mass, recent star formation, dust, and two degrees of freedom in the ionization state of the gas. Stellar metallicity and stellar age are not among these primary drivers ... The pop-cosmos generative model used to create our synthetic data needed 15 SPS parameters ... Our results imply that the rest-frame optical SED in this model can be represented within a signif"

    The abstract states the five-DoF result as a fact about 'the rest-frame optical SED' and about recovering gas conditions 'in typical star-forming galaxies,' while the body and conclusions correctly restrict the claim to SEDs 'drawn from the pop-cosmos galaxy population model.' The leap from model manifold to real-galaxy SED is supported only by the same self-citations; no external spectral catalogue is analysed. Thus the strongest scientific claim inherits its load-bearing justification from the authors' prior generative model.

full rationale

The paper is transparent that every SED is a noiseless, deterministic FSPS/CLOUDY forward map of the 16-parameter SPS vector drawn from pop-cosmos (overlapping authors). The VAE reconstruction + MI analysis therefore measures the intrinsic dimensionality and degeneracies of that particular generative model, not of observed spectra. The abstract and conclusions nevertheless present the five dimensions (and the non-independence of metallicity/age) as properties of 'the rest-frame optical SED' and of 'typical star-forming galaxies.' That generalization rests solely on self-citations to photometric and population-level validations of pop-cosmos; those validations do not constrain optical continuum shape or line ratios at the resolution used here. The internal VAE/MI steps themselves are not circular (they are a legitimate compression of the synthetic catalogue), but the load-bearing premise that the catalogue faithfully represents real optical SEDs is justified only by the authors' prior work. This is partial circularity of applicability, not a definitional loop inside the equations, hence score 4 rather than 6+.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central claim rests on the fidelity of the pop-cosmos generative model and the FSPS/CLOUDY SPS mapping, on the β-VAE architecture and annealing schedule chosen by the authors, and on the interpretation of mutual information as a physical diagnostic. No new physical entities are postulated; the free parameters are the usual neural-network and annealing hyperparameters plus the choice of five latent dimensions.

free parameters (4)
  • β-annealing schedule (initial β=6 → 1.9 → 1.1, three stages, 600 epochs)
    Hand-chosen multi-stage schedule (Appendix A) that trades reconstruction against disentanglement; different schedules could alter the number or purity of recovered latents.
  • latent dimensionality fixed at 5
    Chosen after ablation (4 dims raise error >2%, 6 dims give no gain); the integer 5 is therefore a free modeling choice that defines the claim.
  • normalization-weight α=10 000 in MSE loss
    Set by hand so that the single normalization factor balances the 4223 spectral points; affects what information is forced into the latents.
  • VAE architecture (7 conv layers, filter progression 16→1024, 512-unit bottleneck, LeakyReLU slope 0.15)
    Hyperparameters tuned for reconstruction fidelity; different capacity could change the effective dimensionality recovered.
axioms (4)
  • domain assumption The pop-cosmos generative model plus FSPS/CLOUDY produces a joint distribution of optical SEDs whose intrinsic dimensionality matches that of real COSMOS-like galaxies.
    Stated in §2: ‘validating their population distribution validates the resulting SED population.’ The entire analysis is performed on this synthetic distribution.
  • ad hoc to paper A β-VAE with the chosen annealing schedule recovers statistically independent physical degrees of freedom rather than arbitrary mixtures.
    Disentanglement is enforced by the KL term weighted by β; the paper treats low inter-latent MI (<0.15) as sufficient evidence of physical independence (§3.1).
  • domain assumption Mutual information estimated by GMM-MI correctly ranks the physical content of each latent.
    GMM-MI (Piras et al. 2023) is used throughout §4; its reliability for continuous high-dimensional spectral data is assumed.
  • domain assumption AGN and shocks do not contribute to the optical SED of the modeled population.
    Explicitly noted in §5: pop-cosmos includes AGN only via the dusty torus in the NIR; optical is pure star-formation + nebular emission.

reviewed 2026-07-12 · how reviews work

0 comments
Cite this review

Pith. "Pith review of pop-cosmos: Disentangling galaxy properties from observables using data-driven approaches." pith.science (2026). https://pith.science/paper/2HYGZNQ3

@misc{pith2026260611308,
  author       = {Pith},
  title        = {Pith review of: pop-cosmos: Disentangling galaxy properties from observables using data-driven approaches},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HYGZNQ3}},
  note         = {Machine review of arXiv:2606.11308}
}
Share X Bluesky LinkedIn Reddit HN
abstract

The physical processes that shape a galaxy's spectrum are strongly degenerate in observations, obscuring which processes act independently. Leveraging the pop-cosmos generative galaxy population model, we investigate how many independent degrees of freedom the rest-frame optical SED contains. We use a $\beta$-variational autoencoder (VAE) to compress a 16-parameter stellar population synthesis (SPS) description into a disentangled latent representation interpreted through mutual information (MI). We find that five independent dimensions suffice, corresponding to stellar mass, recent star formation, dust, and two degrees of freedom in the ionization state of the gas. Stellar metallicity and stellar age are not among these primary drivers; their spectral effects are distributed across the others rather than independently encoded. By tying each dimension to specific spectral features, this decomposition breaks the star-formation--dust--metallicity degeneracies that limit broadband photometry, and recovers the physical conditions of the gas in typical star-forming galaxies more cleanly than the line-ratio diagnostics in standard use.

Figures

Figures reproduced from arXiv: 2606.11308 by Anik Halder, Benedict Van den Bussche, Boris Leistedt, Daniel J. Mortlock, Gurjeet Jagwani, Hiranya V. Peiris, Madalina N. Tudorache, Sinan Deger, Stephen Thorp.

Figure 1
Figure 1. Figure 1: Disentanglement of the latents after training the VAE as measured by the MI between them. The MI is in natural units. The MI between identical latents are fixed to 1. MI values under 0.01 are not displayed. Uncertainty on the MI measurements is less than 0.001 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 1
Figure 1. Figure 1: Plate diagram illustrating the hierarchical hybrid compression of data x with parameters) and nuisance parameters 𝜂. The data are partitioned into subsets x0, x1, . . . , and the power spectrum is computed from the full field. Compressions are performed sequentially: ∗ ( ) compresses the power spectrum, then each ∗ (x8) compresses a data subset conditional on all previously obtained summaries (indicated by… view at source ↗
Figure 2
Figure 2. Figure 2: Normalized SED fractional reconstruction error after training the VAE. Significant emission lines are highlighted in dashed red lines. The average overall error is 1.82 × 10−3 with the 95 percentile shaded in blue and the 99.9 percentile in orange. We randomly partition the dataset of 90,000 mock galaxies (nor￾malized SEDs and their normalization factors) into training (60,000 galaxies) and validation (30,… view at source ↗
Figure 2
Figure 2. Figure 2: Hierarchical hybrid statistics scheme adapted for DES Y3 data products. The DES footprint is simulated over a simulation sampling distribution in cosmology 𝑝 sim ( 𝜃 ) with nuisance parameters (left). Empirical vectors are computed from the full shear field (𝛾) footprint and compressed via MI maximization to dim(𝑡0 ) = 10 numbers. The reconstructed convergence field 𝜅 is split into patches A,B, and C and p… view at source ↗
Figure 3
Figure 3. Figure 3: MI between the five latent variables of the VAE trained on rest￾frame SEDs, and galaxy properties. MI values above 0.1 are overplotted on the heatmap. Only key galaxy SPS parameters isolated by the VAE are plotted on the heatmap. The uncertainty is less than 0.01 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The DES Y3 full projected mass map at a HEALPix resolution of 1024. This convergence map is produced by Kaiser-Squires reconstruction, as detailed by Jeffrey & Gatti et al. (2021), using all the source tomographic bins from the metacalibration catalogue. and other observational conditions are below the statistical error budget, making the DES Y3 shape catalogue a key input for the suite of Y3 cosmology ana… view at source ↗
Figure 4
Figure 4. Figure 4: Conditional MI between individual latents and the rest-frame SED at each spectral indices, conditioned on all the other latents. Significant emis￾sion lines are overplotted in dotted gray lines with red labels above. MNRAS 000, 1–12 (2026) [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 3
Figure 3. Figure 3: Samples from 𝑝(𝑛¯ (𝑧) | 𝑥phot) of the DES Y3 footprint redshift distribution from SOMPZ (Myles et al. 2021). only modelled in the initial conditions at a fixed mass of 0.06. The remaining runs beyond verification use the concept code to model the effect of neutrinos (Tram et al. 2019). 5.2 DES Y3 data The Dark Energy Survey (DES) is a multi-band imaging program carried out with the Dark Energy Camera on th… view at source ↗
Figure 5
Figure 5. Figure 5: Variations in the decoded rest-frame SED when varying one latent eight times with all the others fixed. A different latent is traversed in each panel from top to bottom. The left panel displays the full SED while the right panel zooms into the [O ii], [O iii], H β, and H α regions. The first two and last two columns share a 𝑦-axis range. The different latents are ranked by their relative effect on the spec… view at source ↗
Figure 5
Figure 5. Figure 5: Coverage test result (using TARP; Lemos et al. 2023) to validate the density estimation scheme for hybrid statistics. Using repeated mock data parameter inference, the fraction of true values in the appropriate credible intervals matches the expected fraction. The figure shows the result for all three patches and compression. The shaded regions show 1- and 2-𝜎 contours (the standard deviation 𝜎 being calcu… view at source ↗
Figure 6
Figure 6. Figure 6: MI between different properties (top row: stellar mass, sSFR, and stellar metallicity; bottom row: diffuse dust optical depth 𝜏2, and AGN bolometric luminosity fraction 𝑓AGN) and observed frame photometric colours for different redshift slices (vertical axis 𝑧). MI values are approximated to the closest decimal. The colourmap is discretized in bins of width 0.1. The uncertainty on the MI values are all les… view at source ↗
Figure 6
Figure 6. Figure 6: Baryonic feedback systematic error test with the CosmoGridV1 simulations. The mean of the inferred marginal posterior distributions of mock data contaminated with baryonic feedback falls within the 0.3𝜎 of the uncontaminated marginal posterior distribution (this criterion derived from the standard DES Y3 test). form of systematic uncertainty at small scales in weak lensing. The training simulations are dar… view at source ↗
Figure 7
Figure 7. Figure 7: MI between the different properties (top to bottom: sSFR, gas-phase metallicity, and gas ionization) with different emission lines in rest-frame SEDs with galaxies observed across four different redshift slices. the broad optical and NIR slope reflects the continuum reddening caused by dust attenuation (Calzetti et al. 2000) alongside dominant old stellar populations and metals (see Sections 4.2 and 4.5) … view at source ↗
Figure 7
Figure 7. Figure 7: Three dimensional slices of hybrid summaries from simulations (scatter points) capture informative structure with respect to indicated parameter values (scatter colour gradient). The DES Y3 target data (red dot) falls squarely on this learned manifold, indicating that the compressed data is in-distribution with compressed simulations. This visual inspection is a blinded test, as it does not require assigni… view at source ↗
Figure 8
Figure 8. Figure 8: BPT axes against latent 𝑧1 and 𝑧3. Top: BPT diagram coloured by latent 1 (left) and latent 3 (right). Bottom: Latent 1 against latent 3 coloured by the BPT axes log10([N ii]/H α) (left) and log10([O iii]/H β) (right). −2 −1 0 1 2 Latent Dimension 1 −2 −1 0 1 2 Latent Dimension 3 −2 −1 0 1 2 Latent Dimension 1 −2 −1 0 1 2 Latent Dimension 3 −3.5 −3.0 −2.5 −2.0 log10(Ugas / U ) −1.50 −1.25 −1.00 −0.75 −0.50 … view at source ↗
Figure 8
Figure 8. Figure 8: Modular posterior predictive distribution (PPD) test for the noise￾subtracted DES Y3 pseudo- power spectra. The shaded regions show the 1𝜎 and 2𝜎 credible intervals of the posterior predictive distribution. The DES Y3 measurements fall comfortably within the predicted distributions across all tomographic bin combinations. (0.3, 0.8), shown in [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Latent 1 against latent 3 colour coded by gas-phase metallicity (left) and gas ionization (right). MNRAS 000, 1–12 (2026) [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 9
Figure 9. Figure 9: Neural density estimator convergence test on m and 𝑆8 for four (out of eight total) individual NDE ensemble members. This test was performed on the DES Y3 data by blindly shifting the mean value of the posteriors to fiducial values of m = 0.3 and 𝑆8 = 0.8. 0 : 2 0 : 3 0 : 4 m 0 : 9 0 : 8 0 : 7 0 : 6 0 : 5 0 : 4 w 0 : 75 0 : 80 0 : 85 0 : 90 S 8 0 : 75 0 : 80 0 : 85 0 : 90 S 8 0 : 8 0 : 6 0 : 4 w Cosmic She… view at source ↗
Figure 10
Figure 10. Figure 10: Main result. DES Y3 𝑤CDM constraints: posterior probability distribution for parameters { m, 𝑆8, 𝑤} with DES Y3 data. Hybrid statistics improves information extraction about all three parameters over two-point (Doux et al. 2022) and the existing CNN compression (Jeffrey et al. 2025). These constraints are consistent with Planck Collaboration (2020), the latter recalculated here with our analysis priors. w… view at source ↗
Figure 11
Figure 11. Figure 11: DES Y3 𝑤CDM results: we compare the joint posterior proba￾bility distributions between this work’s hybrid statistics and the results from Gatti et al. (2024a) which combines 2nd and 3rd order moments, scattering transforms and wavelet phase harmonics; Prat et al. (2026) which combines Betti numbers and 2nd order moments; and Jeffrey et al. (2025) which com￾bines the angular power spectrum with a field lev… view at source ↗

Review history (2 revisions) →

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

4 extracted references · 3 linked inside Pith

  1. [1]

    Alsing J., et al., 2020, ApJS, 249, 5 Alsing J., Peiris H., Mortlock D., Leja J., Leistedt B., 2023, ApJS, 264, 29 Alsing J., Thorp S., Deger S., Peiris H. V., Leistedt B., Mortlock D., Leja J., 2024, ApJS, 274, 12 Astropy Collaboration et al., 2013, A&A, 558, A33 Astropy Collaboration et al., 2018, AJ, 156, 123 Astropy Collaboration et al., 2022, ApJ, 93...

    2020

  2. [2]

    D., 2016, ApJ, 823, 102 Connolly A

    pp 6572– 6583 (arXiv:1806.07366) Chevallard J., Charlot S., 2016, MNRAS, 462, 1415 Chiang Y.-K., Makiya R., Ménard B., 2025, ApJ, 992, 65 Choi J., Dotter A., Conroy C., Cantiello M., Paxton B., Johnson B. D., 2016, ApJ, 823, 102 Connolly A. J., Csabai I., Szalay A. S., Koo D. C., Kron R. G., Munn J. A., 1995, AJ, 110, 2655 Conroy C., 2013, ARA&A, 51, 393 ...

    Pith/arXiv arXiv doi:10.5281/zenodo.595120 2016

  3. [3]

    D., 2021, bd-j/sedpy, Zenodo, doi:10.5281/zenodo.4582723 MNRAS000, 1–13 (2026) 12B

    Elsevier, pp 236–281 (arXiv:2502.17680), doi:10.1016/B978-0-443-21439-4.00127-9 Johnson B. D., 2021, bd-j/sedpy, Zenodo, doi:10.5281/zenodo.4582723 MNRAS000, 1–13 (2026) 12B. Van den Bussche et al. Johnson B. D., et al., 2021a, dfm/python-fsps, v0.4.1rc1, Zenodo, doi:10.5281/zenodo.4737461 Johnson B. D., Leja J., Conroy C., Speagle J. S., 2021b, ApJS, 254...

    Pith/arXiv arXiv doi:10.1016/b978-0-443-21439-4.00127-9 2021

  4. [4]

    pp 8024–8035 (arXiv:1912.01703) Paxton B., Bildsten L., Dotter A., Herwig F., Lesaffre P., Timmes F., 2011, ApJS, 192, 3 Paxton B., et al., 2013, ApJS, 208, 4 Paxton B., et al., 2015, ApJS, 220, 15 Pérez-Montero E., Contini T., 2009, MNRAS, 398, 949 Pettini M., Pagel B. E. J., 2004, MNRAS, 348, L59 Piras D., Lombriser L., 2024, Phys. Rev. D, 110, 023514 P...

    Pith/arXiv arXiv doi:10.5281/zenodo.20761987 1912

This paper was first reviewed by grok-4.5 on July 12, 2026.