Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

halox is a JAX-powered library that provides differentiable, GPU-accelerated implementations of NFW halo profiles, the halo mass function, halo bias, and overdensity conversions, filling a gap the paper identifies in public halo-modeling to

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

2026-08-04 14:50 UTC pith:PIFNKIGO

load-bearing objection Useful JAX packaging of standard halo formulas, but the validation claim can't be checked without the repo, and the novelty statement contradicts their own references. the 3 major comments →

arxiv 2509.22478 v2 pith:PIFNKIGO submitted 2025-09-26 astro-ph.IM astro-ph.CO

halox: Dark matter halo properties and large-scale structure calculations using JAX

Florian K\'eruzor\'e , Lance A. Moreau This is my paper
classification astro-ph.IM astro-ph.CO
keywords dark matter halosNFW profilehalo mass functionhalo biasJAXautomatic differentiationGPU accelerationlarge-scale structure
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 presents halox, a Python library that implements standard dark matter halo quantities—NFW radial profiles, the halo mass function, halo bias, and overdensity conversions—in JAX. The aim is to make these calculations automatically differentiable and GPU-accelerated, capabilities the paper says are missing from existing publicly available halo tools. If correct, halox lets cosmologists use gradient-based inference methods, such as Hamiltonian Monte Carlo and machine learning, directly on halo statistics and observables. The paper validates the implementations against established non-JAX astronomy toolkits for varying masses, redshifts, overdensities, and cosmologies. A sympathetic reader would care because this fills an infrastructure gap in the growing ecosystem of differentiable cosmology software.

Core claim

The central claim is that halox provides the first publicly available, differentiable and GPU-accelerated JAX implementations of NFW halo radial profiles, the standard halo mass function, and the standard linear halo bias model. Each function is written in JAX, so it can be just-in-time compiled, run on GPUs, and automatically differentiated with respect to inputs including halo mass, redshift, and cosmological parameters. Validation tests compare halox outputs against established non-JAX software across varying masses, redshifts, critical overdensities, and cosmologies, with the comparisons run in an automatic continuous-integration pipeline. The library also provides spherical-overdensity

What carries the argument

The central machinery is JAX's composable transformations—just-in-time compilation, automatic differentiation, and vectorization—applied to analytical NFW density-profile formulas and to the adopted mass-function and bias fitting functions. The NFW quantities follow closed-form derivations for density, enclosed mass, gravitational potential, circular velocity, velocity dispersion, and projected surface density; cosmological integrals use an existing JAX-based cosmology library. This combination makes every output a differentiable function of its inputs, which is what enables gradient-based workflows.

Load-bearing premise

The load-bearing premise is that the adopted mass-function and bias fitting formulas, taken without modification, remain accurate over the full range of masses, redshifts, and overdensity definitions the package advertises, including their cosmology dependence.

What would settle it

Run halox's halo mass function at a redshift beyond the original fitting-formula calibration range, say z=3, or at an overdensity definition far from the original fits, and compare with a high-resolution N-body simulation or an independent emulator; a significant systematic departure would falsify the claim that the package is reliable over its advertised parameter space. A second check is to compare halox's gradients against numerical finite differences to confirm the automatic differentiation is correctly implemented.

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

Share X Bluesky LinkedIn Reddit HN

If this is right

  • halox outputs can be compiled with JAX's just-in-time engine and executed on GPUs, making large arrays of halo-mass-function or profile evaluations fast enough for simulation-scale analyses.
  • Automatic differentiation with respect to mass, redshift, and cosmological parameters enables Hamiltonian Monte Carlo sampling and other gradient-based Bayesian inference directly on halo statistics.
  • The same differentiable functions can serve as components in machine-learning pipelines, since gradients can flow through halo model predictions.
  • Because all properties accept arbitrary critical overdensities and include conversion wrappers, analyses can self-consistently mix overdensity definitions.
  • Validation against established non-JAX software gives users a direct cross-check for the new implementations.

Where Pith is reading between the lines

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

  • Editorial inference: the validation approach is only as strong as the reference implementations; if both halox and the comparison tool share the same fitting-formula extrapolation errors, the cross-checks would not reveal them. An independent check against N-body simulations or an emulator would strengthen confidence.
  • Editorial inference: the paper does not state the calibration range or validation tolerances for the adopted mass-function and bias fitting formulas, so users should treat predictions outside the original calibration regime with caution.
  • Editorial inference: automatic differentiation through the halo mass function opens a route to directly constraining cosmological parameters and mass-function model parameters from cluster counts with gradient-based likelihoods, which previously required finite-difference or derivative-free methods.
  • Editorial inference: a natural extension would be differentiating through the power-spectrum input itself to propagate errors from initial conditions or emulator uncertainties into halo abundance predictions.

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 / 4 minor

Summary. halox is a JAX/Python software package that implements NFW halo radial profiles (density, enclosed mass, potential, circular velocity, velocity dispersion, projected surface density), the Tinker et al. (2008) halo mass function, the Tinker et al. (2010) halo bias, overdensity conversions, and selected cosmology wrappers built on JAX-cosmo. The paper's central claims are that these implementations are correct, automatically differentiable, JIT-compilable, and GPU-accelerated, and that they are validated against Astropy and colossus in an automatic CI/CD pipeline. The manuscript is a short software note with no equations, no repository URL, no numerical validation results, and no stated tolerances, so the central correctness claim is not independently checkable from the text.

Significance. If the validation evidence is supplied, halox would fill a genuinely useful niche: a JAX-native, differentiable, GPU-accelerated implementation of standard NFW halo properties and Tinker HMF/bias, suitable for Hamiltonian Monte Carlo, sensitivity analyses, and machine-learning pipelines. The underlying physics is not novel—the NFW expressions come from Łokas & Mamon (2001) and the halo statistics from Tinker et al. (2008, 2010)—but the differentiable packaging is a serviceable contribution. The paper's strength is its reliance on published closed-form and calibrated results rather than newly derived formulas. However, the manuscript currently does not demonstrate the validation it claims, and the novelty statement is internally inconsistent with its own citation of Buisman et al. (2025). These issues are fixable but are load-bearing for the paper's purpose.

major comments (3)
  1. [Validation] The sole support for the central correctness claim is the sentence: 'All functions available in halox are validated against existing, non-JAX-based software.' No repository URL, commit hash, test file, numerical comparison table, or tolerance values are provided. A reader cannot verify that the JAX implementations reproduce colossus and Astropy outputs; the claim is an assertion. This is load-bearing because the paper's stated purpose is to provide validated differentiable implementations. Please provide the repository identifier and version/commit, and either quantitative validation results (e.g., maximum relative differences over the tested grid) or a stable pointer to CI artifacts, with explicit tolerances.
  2. [Statement of need vs. footnote] The paper claims the listed properties 'do not currently have a publicly available, differentiable and GPU-accelerated implementation,' including the halo mass function. Yet it cites Buisman et al. (2025) as a published differentiable halo mass function, and the footnote states that halox 'choses a lighter, halo model-based approach.' The novelty claim for the HMF component is internally inconsistent. Please qualify the statement, e.g., 'no single library provides all of these quantities in one differentiable JAX framework,' or otherwise clarify how the HMF implementation differs from Buisman et al. without undermining the stated need.
  3. [Features / Validation] The Tinker et al. (2008) HMF and Tinker et al. (2010) bias fitting functions were calibrated to N-body simulations over a limited range of redshifts and overdensity definitions. The paper advertises these models for arbitrary masses, redshifts, critical overdensities, and cosmologies without stating the calibration domain or extrapolation caveats. Moreover, since the validation reference (colossus) presumably implements the same fitting formulas, the cross-check confirms internal consistency but not physical accuracy outside the calibration range. Please state the applicable range and add a validation test that exercises the boundaries of that range.
minor comments (4)
  1. [General typos] Several typographical issues appear: 'efficient' in the Summary, 'choses' in the footnote, and 'Affice' in the Acknowledgments. These should be corrected.
  2. [References] The reference 'Mendoza et al. (2025) In Prep.' is incomplete; it should list author names or be removed from the bibliography if the companion paper is not yet available.
  3. [Features] The paper would benefit from a minimal code example showing the API for one or two core functions (e.g., computing an NFW density profile and its derivative). This would make the differentiable/GPU-accelerated claims more concrete and help potential users assess the package quickly.
  4. [Formatting] The footnote reference to 'F eatures' contains a spurious space, and the section layout would benefit from clearer subsection numbering.

Circularity Check

0 steps flagged

No circularity: halox implements externally published NFW/Tinker formulas and validates against independent software; the sole self-citation is contextual and not load-bearing.

full rationale

I walked the paper's claimed derivation chain. halox is a software package that implements known analytic results: NFW radial profiles from Łokas & Mamon (2001), the Tinker et al. (2008) halo mass function, the Tinker et al. (2010) halo bias, and cosmological quantities via JAX-cosmo. None of these are derived inside the paper, and none are fit to halox's own outputs. The Tinker coefficients are fitted values imported from the original external calibration papers, not fitted here, so there is no fitted_input_called_prediction pattern. Validation is asserted against independent, non-JAX software (colossus and Astropy) through an automatic CI/CD pipeline; that claim is not itself evidence of circularity, and no self-citation is used to justify correctness. The only self-citation is to picasso (Kéruzoré et al., 2024), which appears merely as a contextual example of JAX-powered gas-halo modeling in the Statement of need; it is not load-bearing for any of halox's claimed predictions. I also considered the manuscript's own internal tension: the Statement of need says the implemented properties 'do not currently have a publicly available, differentiable and GPU-accelerated implementation,' while the same paragraph cites Buisman et al. (2025) as a published differentiable halo mass function. This is an inconsistency about the novelty claim for the HMF component, but it does not make any step circular. Similarly, the Validation section's lack of a repository URL, commit hash, numerical tolerances, or quantitative results makes the validation claim unverifiable from the manuscript text; unverifiability is a reproducibility/correctness concern, not a circularity concern. No equation in the paper is defined in terms of another claimed output, no ansatz is smuggled in via self-citation, and no known result is merely renamed. The derivation chain, insofar as one exists, runs from external published formulas to JAX implementations to comparison with independent software; it does not reduce to its own inputs.

Axiom & Free-Parameter Ledger

2 free parameters · 5 axioms · 0 invented entities

halox introduces no fitted parameters of its own and no invented entities; it re-implements published analytic formulas (Łokas & Mamon 2001) and empirical fitting functions (Tinker et al. 2008, 2010). The ledger captures the empirically fitted coefficients imported from the prior literature, and the domain assumptions about model validity and reference implementations that the package's correctness claims rest on.

free parameters (2)
  • Tinker et al. (2008) HMF fit coefficients (amplitude, scale, low-mass and high-mass exponents) = Imported from Tinker et al. 2008 (functions of overdensity); not fitted by halox
    The package's halo mass function output depends on coefficients empirically fitted to N-body simulations in the original paper; halox imports them as fixed inputs.
  • Tinker et al. (2010) bias fit parameters = Imported from Tinker et al. 2010; not fitted by halox
    The linear bias output depends on coefficients fitted to simulations in the original paper; the package uses them as fixed inputs.
axioms (5)
  • domain assumption The NFW profile describes dark matter halos for the purposes of these calculations.
    Invoked in the Features section for all radial profile quantities (density, enclosed mass, potential, circular velocity, velocity dispersion, projected surface density).
  • domain assumption The Tinker et al. (2008) fitting function correctly gives the halo mass function over the parameter ranges halox exposes, including cosmological parameter dependence.
    Invoked in the Features section HMF bullet; the paper does not state the calibration range of the fit.
  • domain assumption The Tinker et al. (2010) fitting function correctly gives linear halo bias.
    Invoked in the Features section halo bias bullet.
  • domain assumption colossus and Astropy are correct reference implementations, so agreement with them establishes correctness.
    Invoked in the Validation section; no tolerances are stated in the manuscript.
  • standard math The Łokas & Mamon (2001) analytical expressions used for NFW-derived quantities are correct.
    The paper states its implementations 'are based on the analytical derivations of Łokas & Mamon (2001)' in the Features section.

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of halox: Dark matter halo properties and large-scale structure calculations using JAX." pith.science (2026). https://pith.science/paper/PIFNKIGO

@misc{pith2026250922478,
  author       = {Pith},
  title        = {Pith review of: halox: Dark matter halo properties and large-scale structure calculations using JAX},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PIFNKIGO}},
  note         = {Machine review of arXiv:2509.22478}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Dark matter halos are fundamental structures in cosmology, forming the gravitational potential wells hosting galaxies and clusters of galaxies. Their properties and statistical distribution (including the halo mass function) are invaluable tools to infer the fundamental properties of the Universe. The \texttt{halox} package is a JAX-powered Python library enabling differentiable and accelerated computations of key properties of dark matter halos, and of the halo mass function. The automatic differentiation capabilities of \texttt{halox} enable its usage in gradient-based workflows, e.g. in efficient Hamiltonian Monte Carlo sampling or machine learning applications.

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. Diffhalos: A Generative Model of Cosmological Lightcones of Dark Matter Halos

    astro-ph.GA 2026-07 conditional novelty 6.0

    Diffhalos generates statistically accurate Monte-Carlo and quasi-Monte-Carlo lightcones of halos, subhalos and Diffmah mass-assembly histories, enabling autodiff gradients of the mass functions.

Reference graph

Works this paper leans on

13 extracted references · 5 linked inside Pith · cited by 1 Pith paper

  1. [1]

    P., Becker, M

    Alarcon, A., Hearin, A. P., Becker, M. R., & Chaves-Montero, J. (2023). Diffstar: a fully parametric physical model for galaxy assembly history. 518(1), 562–

    2023

  2. [2]

    https://doi.org/10.3847/1538-4365/ab1658 Diemer, B. (2018). COLOSSUS: A Python Toolkit for Cosmology, Large-scale Structure, and Dark Matter Halos. 239(2),

    doi:10.3847/1538-4365/ab1658 2018

  3. [3]

    https://doi.org/10.21105/astro.2112.08423 JAX-GalSim developpers. (2025). JAX-GalSim: JAX port of GalSim, for parallelized, GPU accelerated, and differentiable galaxy image simulations (Version 0.0.1rc1). https://github.com/GalSim-developers/JAX-GalSim Kéruzoré, F., Bleem, L. E., Frontiere, N., Krishnan, N., Buehlmann, M., Ember- son, J. D., Habib, S., & ...

    Pith/arXiv arXiv 2025

  4. [7]

    P., Ramachandra, N., Becker, M

    https://doi.org/ 10.21105/astro.2105.05859 Hearin, A. P., Ramachandra, N., Becker, M. R., & DeRose, J. (2022). Differ- entiable Predictions for Large Scale Structure with SHAMNet. The Open Journal of Astrophysics , 5,

    Pith/arXiv arXiv 2022

  5. [15]

    E., Alonso, D., Krause, E., Leonard, C

    https://doi.org/10.21105/astro.2302.05163 Chisari, N. E., Alonso, D., Krause, E., Leonard, C. D., Bull, P., Neveu, J., Villarreal, A. S., Singh, S., McClintock, T., Ellison, J., Du, Z., Zuntz, J., Mead, A., Joudaki, S., Lorenz, C. S., Tröster, T., Sanchez, J., Lanusse, F., Ishak, M., … LSST Dark Energy Science Collaboration. (2019). Core Cosmology Library...

    Pith/arXiv arXiv 2019

  6. [20]

    S., & Cubuk, E

    https://doi.org/10.21105/astro.2305.06347 Schoenholz, S. S., & Cubuk, E. D. (2020). JAX m.d. A framework for differ- entiable physics. Advances in Neural Information Processing Systems ,

    Pith/arXiv arXiv 2020

  7. [33]

    E., Peacock, J

    https://papers.nips.cc/paper/2020/file/83d3d4b6c9579515e1679aca8cbc803 3-Paper.pdf Smith, R. E., Peacock, J. A., Jenkins, A., White, S. D. M., Frenk, C. S., Pearce, F. R., Thomas, P. A., Efstathiou, G., & Couchman, H. M. P. (2003). Stable clustering, the halo model and non-linear cosmological power spectra. 341(4), 1311–1332. https://doi.org/10.1046/j.136...

    arXiv 2020

  8. [35]

    https://doi.org/10.3847/15 38-4365/aaee8c Differentiable Universe Initiative. (2025). JaxPM: JAX-powered cosmological particle-mesh n-body solver (Version 0.1.6). https://github.com/Differentia bleUniverseInitiative/JaxPM?tab=readme-ov-file Freeman, C. D., Frey, E., Raichuk, A., Girgin, S., Mordatch, I., & Bachem, O. (2021). Brax - a differentiable physic...

    doi:10.48550/arxiv.2408.14538 2025

  9. [63]

    P., Campbell, D., Tollerud, E., Behroozi, P., Diemer, B., Goldbaum, N

    https://doi.org/ 10.1088/1475-7516/2024/06/063 Hearin, A. P., Campbell, D., Tollerud, E., Behroozi, P., Diemer, B., Goldbaum, N. J., Jennings, E., Leauthaud, A., Mao, Y.-Y., More, S., Parejko, J., Sinha, M., Sipöcz, B., & Zentner, A. (2017). Forward Modeling of Large-scale Structure: An Open-source Approach with Halotools. 154(5),

    doi:10.1088/1475-7516/2024/06/063 2024

  10. [116]

    L., & Mamon, G

    https://doi.org/10.33232/001c.127486 Łokas, E. L., & Mamon, G. A. (2001). Properties of spherical galaxies and clusters with an NFW density profile. 321(1), 155–166. https://doi.org/10 .1046/j.1365-8711.2001.04007.x Mendoza et al. (2025). In Prep. Navarro, J. F., Frenk, C. S., & White, S. D. M. (1997). A Universal Density Profile from Hierarchical Cluster...

    arXiv 2001

  11. [167]

    https://doi.org/10.3847/1538-4357/ac7c74 Balkenhol, L., Trendafilova, C., Benabed, K., & Galli, S. (2024). candl: cosmic microwave background analysis with a differentiable likelihood. 686, A10. https://doi.org/10.1051/0004-6361/202449432 Bradbury, J., Frostig, R., Hawkins, P., Johnson, M. J., Leary, C., Maclaurin, D., Necula, G., Paszke, A., VanderPlas, ...

    Pith/arXiv arXiv doi:10.48550/arxiv.2507.03074 2024

  12. [190]

    P., Chaves-Montero, J., Becker, M

    https: //doi.org/10.3847/1538-3881/aa859f Hearin, A. P., Chaves-Montero, J., Becker, M. R., & Alarcon, A. (2021). A Differentiable Model of the Assembly of Individual and Populations of Dark Matter Halos. The Open Journal of Astrophysics , 4(1),

    doi:10.3847/1538-3881/aa859f 2021

  13. [584]

    M., Lim, P

    https://doi.org/10.1093/mnras/stac3118 Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., Earl, N., Starkman, N., Bradley, L., Shupe, D. L., Patil, A. A., Corrales, L., Brasseur, C. E., Nöthe, M., Donath, A., Tollerud, E., Morris, B. M., Ginsburg, A., Vaher, E., Weaver, B. A., Tocknell, J., Jamieson, W., … Astropy Project Contribu- tors. (2022). The ...

    doi:10.1093/mnras/stac3118 2022

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.