Pith. sign in

REVIEW 4 major objections 5 minor 21 references

Environmental Effects on Galaxy Evolution

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Galaxies inside VoidFinder voids are systematically fainter, less massive, bluer, and more star-forming than wall galaxies, and a nonparametric Pólya-tree Bayes factor detects those differences more strongly than a parametric mixture…

desk verdict Qualitatively right and a useful first DESI application, but the headline Bayes factors are not reproducible until the data transform is specified and the KS comparison is actually run. read the letter →

arxiv 2507.17243 v1 pith:36DEKEZZ submitted 2025-07-23 astro-ph.GA astro-ph.IMstat.AP

Ziqi Mu This is my paper · ORCID
classification astro-ph.GAastro-ph.IMstat.AP
keywords galaxyevolutioncosmicvoidsvoidgalaxiesPólyatreepriorBayesfactornonparametricBayesiantestingSDSSDR7DESIDR1BGS
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Galaxies in the emptiest cosmic regions should, if environment shapes evolution, be fainter, less massive, bluer, and more actively star-forming than galaxies in dense walls. This paper claims that such differences are real and detectable, but only when both the statistical test and the void catalog are chosen carefully. Using a nonparametric Bayesian test based on Pólya tree priors on SDSS DR7 and DESI DR1 BGS data, the author finds strong evidence of these contrasts for VoidFinder-defined voids and much weaker evidence for V² REVOLVER-defined voids. The point of the result is that environmental effects on galaxy evolution exist, yet their measured strength is tied to how 'void' is defined. A fair reader should come away knowing that the statistical machinery matters as much as the astrophysical one.

What carries the argument

The load-bearing machinery is the Pólya tree prior, a prior over probability distributions on the unit interval $[0,1)$ that recursively splits the interval into binary cells; each split gets an independent Beta-distributed branching probability, and the tree is truncated at depth $m$. Setting the branching parameters to $c/m^2$ with $c=1$ makes the prior conjugate, so the marginal likelihood for each split has a closed Gamma-function product form. The test statistic is the Bayes factor $B_{01}$, the ratio of the marginal likelihood under $H_0$ (one common distribution for void and wall samples) to the product of marginal likelihoods under $H_1$ (two independent distributions); negative log Bayes factors count as evidence for distinct environments. The precision parameter $c$ controls how strongly the prior concentrates near the centering distribution, and the paper's simulations show that $c=1$ balances sensitivity to mean and variance shifts against robustness.

What would settle it

Re-run the Pólya-tree Bayes factor on the DESI DR1 BGS void/wall samples with the $[0,1)$ transform fixed in advance (for example, ranking all galaxies once in the pooled sample before splitting into void and wall) and check whether the strong negative log Bayes factors for VoidFinder in Table 4.2 survive; if they flip toward zero, the unspecified transform is driving the result.

Watch

Extended reading notes

Core claim

The central claim, on the paper's own terms, is that a Bayesian nonparametric two-sample test built from Pólya tree priors detects environmental differences in galaxy populations more sensitively and more robustly than a parametric Gaussian-mixture Bayes factor or the Kolmogorov-Smirnov test. Applied to SDSS DR7, the nonparametric log Bayes factors under VoidFinder are more negative than the parametric values for every property—stellar mass goes from $-1708$ to $-3896$—meaning stronger evidence that void and wall distributions differ. Applied to DESI DR1 BGS, VoidFinder void galaxies show lower stellar mass and luminosity, bluer $u-r$ and $g-r$ colors, and higher H$\alpha$ equivalent width than wall galaxies, while V$^2$ REVOLVER classifications show much weaker or absent contrasts, with log Bayes factors near zero or positive. A notable sign flip occurs for specific star formation rate in SDSS DR7 under V$^2$ REVOLVER: the parametric test gives $-19$ (weak support for a difference) while the nonparametric test gives $+161$ (support for no difference). The paper concludes that environmental effects on galaxy evolution are detectable and statistically significant, but their measured strength depends on the void-finding algorithm.

Load-bearing premise

The load-bearing premise is that every galaxy property can be converted onto the test's unit interval without distorting the comparison, yet the paper never specifies the conversion; if it is estimated from the same data, the reported evidence for differences could be artificially strong.

Editorial extensions

If this is right

  • If the central claim is right, the standard expectation that void galaxies are less evolved holds at least for VoidFinder-defined voids: they are fainter, less massive, bluer, and more actively star-forming in both SDSS DR7 and DESI DR1 BGS.
  • The nonparametric Bayes factor gives stronger evidence for well-separated void/wall differences than the parametric Gaussian-mixture Bayes factor, so future environmental studies can obtain decisive evidence without assuming a distributional shape.
  • The near-disappearance of differences under V² REVOLVER means that the void catalog itself is part of the physical conclusion; galaxies classified as void by one algorithm may be wall-like by another.
  • In subtle cases the nonparametric test is conservative, sometimes supporting the null hypothesis where the parametric test favored the alternative, so conclusions about weak environmental effects must state which test produced them.

Reading between the lines

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

  • I infer that published void-galaxy contrasts are only meaningful when paired with the exact void-finding algorithm; future reviews or merger studies should not average results across VoidFinder and watershed-based catalogs.
  • I infer that the same Pólya-tree test could be run on additional baryonic tracers, such as gas-phase metallicity or satellite counts, to see whether the environment's fingerprint extends beyond the six properties studied here.
  • I infer that the missing $[0,1)$ transform is a reproducibility risk; a pre-registered transform, such as ranking the pooled sample once before splitting, would let readers judge whether the reported log Bayes factors are robust.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper applies a Polya-tree-based nonparametric Bayesian two-sample test to compare the distributions of six galaxy properties (stellar mass, r-band absolute magnitude, g-r and u-r colors, SFR/sSFR, and H-alpha EW) between void and wall galaxies in SDSS DR7 and DESI DR1 BGS, using two void-finding algorithms (VoidFinder and V2 REVOLVER). It reports log Bayes factors in Tables 4.1 and 4.2 and concludes that VoidFinder voids host systematically fainter, less massive, bluer, and more star-forming galaxies, while these contrasts are weaker for V2 REVOLVER voids. The paper also claims that the nonparametric test is more sensitive and robust than the Kolmogorov-Smirnov test and parametric Bayesian alternatives, based on simulation-based sensitivity analysis and a comparison with parametric results from Zaidouni et al. [4].

Significance. If the statistical framework is valid, the scientific conclusion that environmental effects depend on the void-finding algorithm is interesting and consistent with earlier work, and the application to DESI DR1 BGS is a useful extension. The paper's explicit strength is its use of a fully Bayesian nonparametric procedure with a published implementation, replicated simulation checks for the c parameter, and a clear scientific hypothesis. However, the central statistical claims about sensitivity and robustness are currently under-supported, and the reported Bayes factors are not reproducible as written because of an unspecified data transform and unvalidated tree depth. The qualitative histogram comparisons do support the main astrophysical conclusion, but the quantitative evidence in Tables 4.1 and 4.2 needs additional methodological work before the paper's stronger claims can be accepted.

major comments (4)
  1. [§2.1, §2.2, Tables 4.1–4.2] The Polya tree prior is defined on the domain Ω = [0,1), but all six galaxy properties analyzed in Section 1.1 are real-valued. The paper never specifies how the data are mapped into [0,1). Every log Bayes factor in Tables 4.1 and 4.2 depends on this transform. If the transform is the empirical CDF of the pooled void+wall sample, then the partition counts n_{j0}, n_{j1} in Eqs. (2.10)–(2.11) are conditional on the data, so the reported quantity is not the ratio of marginal likelihoods under the stated prior; the test becomes circular and biased toward H1. If a fixed transform is used, it must be described for reproducibility. The sensitivity analysis in Section 2.3 does not test the transform choice, so the robustness claim is incomplete.
  2. [Abstract and §4.1] The abstract claims that the approach, compared to the Kolmogorov-Smirnov test and a parametric Bayesian test, 'provides a more sensitive and robust comparison,' but no KS test is run anywhere in the paper. Section 4.1 only compares against the parametric Bayes factors from Zaidouni et al. [4]. The claimed superiority over KS is therefore unsupported. Either a KS test should be performed on the same datasets and its performance compared (e.g., by p-value calibration or by reporting Bayes factors alongside KS statistics), or the abstract should be revised to remove the unsubstantiated comparison.
  3. [§4.1, Table 4.1] The magnitude of a Bayes factor is not comparable across different prior specifications: the parametric mixture model and the nonparametric Polya tree prior occupy different model spaces, so the statement that a change from -1708 (parametric) to -3896 (nonparametric) 'amplifies the evidence' is not a valid measure of sensitivity. Comparing absolute log Bayes factors across methods conflates prior sensitivity with evidence for the alternative. A valid sensitivity comparison would require a controlled simulation study with known ground truth and a measure such as power at a fixed type-I error rate, or calibration of Bayes factors under the null.
  4. [§2.3 and §5] The applications use tree depth m = 6, but the sensitivity analysis in Section 2.3 fixes m = 8 (stated under the variance-shift setting), and the conclusion admits that 'the effect of tree depth m' has not been systematically studied. The choice m = 6 is therefore unvalidated for the real datasets. The paper should report whether the signs and approximate magnitudes of the log Bayes factors in Tables 4.1 and 4.2 are stable across a range of m (e.g., 4, 6, 8, 10) and across a range of c values, not only c = 1.
minor comments (5)
  1. [§1.2] The statement that the KS test 'can yield an extremely low p-value since it computes the tail probability of the null hypothesis being true' misstates the definition of a p-value; a p-value is the probability of observing a test statistic as extreme as or more extreme than the observed one under the null, not the probability that the null is true.
  2. [§2.3] The sentence 'with n = 10, 50, 100, 200 and 500 replications' is ambiguous; it should clarify that n refers to sample sizes and 500 to the number of replications.
  3. [Figures 2.2 and 2.3] The legend label 'holms' should read 'Holmes et al. [7]' and the reference should be spelled consistently as 'Holmes' (not 'Holms').
  4. [§4.1, Figure 4.1] In the caption of Figure 4.1, the right column is labeled 'VoidFinder' in the figure text; it should be 'V2' to match the left/right description in the caption.
  5. [Eq. (1.1)] The numerator Pr(y^{(1,2)} | H0) is notationally awkward; it should be written as Pr(y^{(1)}, y^{(2)} | H0) or explicitly defined as the marginal likelihood of the pooled sample under H0.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Polya-tree Bayes factors are computed from the data via standard marginal likelihoods; the void/wall differences are empirical findings, and self-citations are data catalogs or comparison baselines rather than load-bearing derivations.

full rationale

The derivation chain is self-contained: Section 2 defines the Polya tree prior on [0,1), the marginal likelihood (Eq. 2.2) and Bayes factor (Eqs. 2.10-2.11) follow the standard Holmes et al. two-sample nonparametric test, and Section 2.3 validates the implementation on simulated N(0,1) data against the published Holmes reference curves. The hyperparameters are not fitted to the void/wall target data: c=1 is selected from Holmes's recommended range and from Hanson's flexibility discussion, and m=6 is fixed, with the paper explicitly stating in Section 5 that the effect of m has not been systematically studied. The reported log Bayes factors in Tables 4.1 and 4.2 are therefore computed marginal likelihood ratios, not predictions from a fitted model. The void/wall property differences are empirical comparisons of histograms and Bayes factors, not quantities defined in terms of the test's outputs. Self-citations to Zaidouni et al. (2024), Rincon et al. (2024), and Douglass et al. provide the comparison parametric Bayes factors, the DESI void catalogs, and the VAST toolkit; these are independent data products and baselines, and the paper computes its own nonparametric results rather than importing the conclusion from them. The main methodological weakness is that the probability integral transform from real-valued galaxy properties to the Polya tree domain [0,1) is never specified, so the numerical Bayes factors may not be reproducible; however, this is a correctness/reproducibility gap, not a circular reduction of the output to the input, because no equation in the paper defines the result in terms of a fitted parameter or a self-citation.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on three hand-chosen statistical hyperparameters: c, m, and the data transform to [0,1]. None are fitted to the galaxy data to force a result, so circularity burden is low, but the unstated transform is a reproducibility and soundness risk. The paper introduces no new physical entities.

free parameters (3)
  • Polya tree precision parameter c = 1
    Chosen as a hyperparameter after simulation sensitivity analysis following Holmes et al. [7]. It controls prior concentration but is not fitted to the galaxy data. The choice affects the magnitude and sign of the Bayes factors, as shown in Figures 2.2 and 2.3.
  • Polya tree depth m = 6 for SDSS and DESI applications; 8 in simulations
    The thesis fixes m=6 for all real-data results without a sensitivity analysis on the actual galaxy samples, and acknowledges in the conclusion that the effect of m is still to be studied. This parameter controls partition granularity and hence the computed Bayes factor.
  • Data transform to [0,1] = unspecified
    The Polya tree prior is defined on [0,1), and raw galaxy properties must be mapped into this domain, but the thesis never describes the transform or its centering distribution. This is an unstated hand-chosen element that the numerical results depend on.
assumptions (4)
  • standard math Beta-binomial conjugacy yields the marginal likelihood factorization in Eq. (2.2).
    Invoked to compute model evidence; standard result in Bayesian nonparametrics, cited to Holmes et al. [7] and Ferguson [9].
  • domain assumption Setting alpha_{m0}=alpha_{m1}=c m^2 ensures absolute continuity of the sampled measure with probability 1.
    Follows Ferguson [9] and Holmes [7]; it restricts the prior to densities rather than arbitrary distributions, which is necessary for the Bayes factor interpretation.
  • standard math Under H1 the two samples are independent with identical Polya tree priors, leading to the product-form Bayes factor in Eq. (2.11).
    This is the standard two-sample Bayes factor decomposition; it assumes no dependence between the two samples beyond independent sampling.
  • ad hoc to paper The Polya tree prior with c=1 and m=6, centered on some unknown measure after an unspecified transform, is appropriate for all six galaxy properties.
    The selection of c=1 is based on simulations on N(0,1) data, and m=6 is asserted without testing on real survey data. The transform to [0,1] is not given, so the adequacy of the prior is an unverified assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Environmental Effects on Galaxy Evolution." pith.science (2026). https://pith.science/paper/36DEKEZZ

@misc{pith2026250717243,
  author       = {Pith},
  title        = {Pith review of: Environmental Effects on Galaxy Evolution},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/36DEKEZZ}},
  note         = {Machine review of arXiv:2507.17243}
}
read the original abstract

Galaxies evolve within a web-like cosmic structure, and their properties are strongly shaped by their surrounding environments. We apply a nonparametric Bayesian two-sample test based on P\'olya tree priors to galaxy data from SDSS DR7 and DESI DR1 BGS to quantify the differences between galaxies in dense and sparse cosmic environments. Compared to the Kolmogorov-Smirnov test and parametric Bayesian test, our approach does not require strong assumptions about the underlying distributional form and provides a more sensitive and robust comparison of galaxy evolution across different environments. In particular, galaxies in VoidFinder voids tend to be fainter, less massive, and more star-forming compared to wall galaxies, while such contrasts are diminished under the V^2 REVOLVER pruning. These findings underscore the importance of both the statistical framework and the void classification algorithm in interpreting environmental effects on galaxy evolution.

Figures

Figures reproduced from arXiv: 2507.17243 by the authors.

Figure 1.1
Figure 1.1. Distribution of absolute magnitude (left) and color [PITH_FULL_IMAGE:figures/full_fig_p010_1_1.png] view at source ↗
Figure 2.1
Figure 2.1. Construction of a P´olya tree distribution, each of the [PITH_FULL_IMAGE:figures/full_fig_p015_2_1.png] view at source ↗
Figure 2.2
Figure 2.2. Median log Bayes factor with respect to the sample size under the [PITH_FULL_IMAGE:figures/full_fig_p019_2_2.png] view at source ↗
Figures from the paper (9 more)
Figure 2.3
Figure 2.3. Figure 2.3: Median log Bayes factor with respect to the sample size under two al [PITH_FULL_IMAGE:figures/full_fig_p020_2_3.png]
Figure 3.1
Figure 3.1. Figure 3.1: VoidFinder slice plot for BGS galaxies at a declination of [PITH_FULL_IMAGE:figures/full_fig_p025_3_1.png]
Figure 3.2
Figure 3.2. Figure 3.2: V 2 REVOLVER slice plot for BGS galaxies at a declination of δ = 3◦ . Void (wall) galaxies are shown in red (black). Edge (interior) voids are shown in yellow (blue). [17] [PITH_FULL_IMAGE:figures/full_fig_p026_3_2.png]
Figure 4.1
Figure 4.1. Figure 4.1: Stellar mass distribution (top) and luminosity distribution (bottom) of [PITH_FULL_IMAGE:figures/full_fig_p028_4_1.png]
Figure 4.2
Figure 4.2. Figure 4.2: Color distribution, u − r (top) and g − r (bottom), separated by their environment (void, wall) using VoidFinder (left column) and V2 (right column). The full galaxy population is shown as a shaded gray histogram. Blue and red step-line histograms represent wall and …
Figure 4.3
Figure 4.3. Figure 4.3: Star formation rate (top) and specific star formation rate (bottom), [PITH_FULL_IMAGE:figures/full_fig_p030_4_3.png]
Figure 4.4
Figure 4.4. Figure 4.4: Stellar mass distribution (top) and luminosity distribution (bottom) of [PITH_FULL_IMAGE:figures/full_fig_p033_4_4.png]
Figure 4.5
Figure 4.5. Figure 4.5: Color distribution, u − r (top) and g − r (bottom), separated by their environment (void, wall) using VoidFinder (left column) and V2 (right column). The full galaxy population is shown as a shaded gray histogram. Blue and red step-line histograms represent wall and …
Figure 4.6
Figure 4.6. Figure 4.6: Distribution of log Hα Equivalent Width according to VoidFinder (left) and V2 (right). The full galaxy population is shown as a shaded gray histogram. Blue and red step-line histograms represent wall and void galaxies [PITH_FULL_IMAGE:figures/full_fig_p034_4_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 19 canonical work pages

  1. [4]

    The impact of void-finding algorithms on galaxy classification

    Fatima Zaidouni et al. “The impact of void-finding algorithms on galaxy classification”. In: The Astrophysical Journal 978.1 (2024), p. 3

    work page 2024

  2. [1]

    How filaments of galaxies are woven into the cosmic web

    J Richard Bond, Lev Kofman, and Dmitry Pogosyan. “How filaments of galaxies are woven into the cosmic web”. In:Nature 380.6575 (1996), pp. 603– 606

    work page 1996

  3. [2]

    V AST: the Void Analysis Software Toolkit

    Kelly A Douglass et al. “V AST: the Void Analysis Software Toolkit”. In: Journal of Open Source Software 7.77 (2022), p. 4033

    work page 2022

  4. [3]

    The galaxy stellar mass function and its evolution with time show no dependence on global environment

    Benedetta Vulcani et al. “The galaxy stellar mass function and its evolution with time show no dependence on global environment”. In: Astronomy & Astrophysics 550 (2013), A58

    work page 2013

  5. [5]

    The signed Kolmogorov-Smirnov test: why it should not be used

    Guillaume J Filion. “The signed Kolmogorov-Smirnov test: why it should not be used”. In: Gigascience 4.1 (2015), s13742–015

    work page 2015

  6. [6]

    A tutorial on testing hypotheses using the Bayes factor

    Herbert Hoijtink et al. “A tutorial on testing hypotheses using the Bayes factor.” In: Psychological methods 24.5 (2019), p. 539

    work page 2019

  7. [7]

    Two-sample Bayesian nonparametric hypothesis testing

    Chris C Holmes et al. “Two-sample Bayesian nonparametric hypothesis testing”. In: (2015)

    work page 2015

  8. [8]

    A Bayesian nonparametric approach to testing for dependence between random variables

    Sarah Filippi and Chris C Holmes. “A Bayesian nonparametric approach to testing for dependence between random variables”. In: (2017)

    work page 2017

Show all 21 references
  1. [9]

    Prior distributions on spaces of probability mea- sures

    Thomas S Ferguson. “Prior distributions on spaces of probability mea- sures”. In: The annals of statistics (1974), pp. 615–629

    1974

  2. [10]

    Some aspects of Polya tree distributions for statistical modelling

    Michael Lavine. “Some aspects of Polya tree distributions for statistical modelling”. In: The annals of statistics (1992), pp. 1222–1235

    1992

  3. [11]

    A first course in Bayesian statistical methods

    Peter D Hoff. A first course in Bayesian statistical methods . Vol. 580. Springer, 2009

    2009

  4. [12]

    Bayes Factors

    Robert E. Kass and Adrian E. Raftery. “Bayes Factors”. In: Journal of the American Statistical Association 90.430 (1995), pp. 773–795. doi: 10. 1080 / 01621459 . 1995 . 10476572. eprint: https : / / www . tandfonline . com/doi/pdf/10.1080/01621459.1995.10476572 . url: https://...

    1995

  5. [13]

    PTTests: Polya Tree Hypothesis Tests

    Philip Boeken. PTTests: Polya Tree Hypothesis Tests . https://github. com/philipboeken/PTTests. Accessed: 2025-04-04. 2022

    2025

  6. [14]

    Inference for mixtures of finite Polya tree models

    Timothy E Hanson. “Inference for mixtures of finite Polya tree models”. In: Journal of the American Statistical Association 101.476 (2006), pp. 1548– 1565

    2006

  7. [15]

    The seventh data release of the Sloan Digi- tal Sky Survey

    Kevork N Abazajian et al. “The seventh data release of the Sloan Digi- tal Sky Survey”. In: The Astrophysical Journal Supplement Series 182.2 (2009), p. 543

    2009

  8. [16]

    Bayes Factors for M∗ vs En- vironment (SDSS DR7)

    Fatima Zaidouni and DESI-UR Collaboration. Bayes Factors for M∗ vs En- vironment (SDSS DR7). https://github.com/DESI-UR/Void_analysis/ blob/SDSS_DR7/nb/Fatima_Zaidouni/Bayes_factors- Mstar.ipynb . Accessed: 2025-04-04. 2024

    2025

  9. [17]

    DESIV AST: A Catalog of Low-Redshift Voids us- ing Data from the DESI DR1 Bright Galaxy Survey

    Hernan Rincon et al. “DESIV AST: A Catalog of Low-Redshift Voids us- ing Data from the DESI DR1 Bright Galaxy Survey”. In: arXiv preprint arXiv:2411.00148 (2024)

    2024 arXiv

  10. [18]

    Correlations between H α equivalent width and galaxy properties at z= 0.47: Physical or selection-driven?

    Ali Ahmad Khostovan et al. “Correlations between H α equivalent width and galaxy properties at z= 0.47: Physical or selection-driven?” In:Monthly Notices of the Royal Astronomical Society 503.4 (2021), pp. 5115–5133

    2021

  11. [19]

    Updated void cata- logs of the SDSS DR7 main sample

    Kelly A Douglass, Dahlia Veyrat, and Segev BenZvi. “Updated void cata- logs of the SDSS DR7 main sample”. In: The Astrophysical Journal Sup- plement Series 265.1 (2023), p. 7

    2023

  12. [20]

    V AST Project Example Scripts

    DESI-UR Collaboration. V AST Project Example Scripts. https://github. com/DESI- UR/VAST/blob/master/example_scripts/void_analysis. ipynb. Accessed: 2025-04-27. 2024

    2025

  13. [21]

    A Comparison of Void-finding Algorithms Using Crossing Numbers

    Dahlia Veyrat, Kelly A Douglass, and Segev BenZvi. “A Comparison of Void-finding Algorithms Using Crossing Numbers”. In: The Astrophysical Journal 958.1 (2023), p. 59

    2023

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.