Pith. sign in

REVIEW 2 major objections 4 minor 28 references

Profile likelihood recovers injected neutron-star radius to 0.4σ on synthetic NICER data and is far cheaper than nested sampling.

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 · grok-4.5

2026-07-14 06:37 UTC pith:A3XH4E7X

load-bearing objection Clean, modest proof-of-principle: profile likelihood inside X-PSI recovers the injected radius on one synthetic set, faster than MultiNest, with public code; the single-realisation limit is real but already flagged by the authors. the 2 major comments →

arxiv 2607.11154 v1 pith:A3XH4E7X submitted 2026-07-13 astro-ph.HE astro-ph.IM

Determination of neutron star radius from pulse profile modeling using profile likelihood

Vyaas Ramakrishnan , Shantanu Desai This is my paper
classification astro-ph.HE astro-ph.IM
keywords neutron star radiuspulse profile modelingprofile likelihoodX-PSINICERfrequentist inferencenested sampling
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.

Neutron-star radius measurements from X-ray pulse profiles have almost always been obtained with Bayesian nested sampling inside the X-PSI ray-tracing package. The authors show that a classical profile-likelihood treatment of the same problem works at least as well on synthetic data that mimic a real NICER observation. By fixing the equatorial radius and maximising the likelihood over the remaining twelve nuisance parameters, they recover the true injected radius to 0.4σ, with a 68% confidence interval whose width is comparable to the Bayesian credible interval yet at roughly 1/400th the computational cost. The result is offered as a proof-of-principle that frequentist inference can usefully complement the Bayesian pipeline that currently dominates the field, and the modified code is released publicly so that the same comparison can be repeated on real pulsars.

Core claim

On the synthetic NICER+XMM data set syntX1, maximising the X-PSI likelihood over all nuisance parameters for each fixed equatorial radius yields a profile-likelihood estimate of 12.096 +0.064/-0.182 km (68% CL) that recovers the injected value 12.176 km to 0.4σ, with precision comparable to MultiNest while requiring only about 9 CPU-core hours versus roughly 3840.

What carries the argument

Profile likelihood for equatorial radius: for each fixed R_eq the likelihood is maximised over the twelve remaining free parameters (mass, distance, inclination, hotspot geometry and temperatures, hydrogen column) by Nelder-Mead, producing a one-dimensional Δχ^{2} curve from which frequentist confidence intervals are read off.

Load-bearing premise

A single synthetic realisation plus low-resolution X-PSI settings and a simple maximiser is enough to claim that the frequentist intervals are generally unbiased and of comparable precision.

What would settle it

Repeating the identical profile-likelihood and MultiNest analyses on an ensemble of independent synthetic data sets drawn from the same true model and checking whether the frequentist intervals remain unbiased and of comparable width.

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

If this is right

  • Radius constraints for other NICER pulsars can be obtained with profile likelihood at a fraction of the nested-sampling cost.
  • Discrepancies previously blamed on sampler choice can be cross-checked with an independent frequentist pipeline that does not rely on nested sampling.
  • Volume effects that appear when priors are wide can be avoided by reporting profile-likelihood intervals alongside Bayesian posteriors.
  • Public release of the frequentist X-PSI interface lets other groups apply the same method to real data without re-implementing the maximisation layer.

Where Pith is reading between the lines

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

  • If the same speed-up holds on real data, large-scale systematic studies of atmosphere models or hotspot geometries become feasible that were previously limited by nested-sampling cost.
  • Profile-likelihood intervals could be used as a rapid diagnostic for whether a MultiNest run has adequately explored the high-radius tail where the likelihood is flat.
  • The method may generalise to other multi-parameter pulse-profile problems (e.g., simultaneous mass-radius-inclination constraints) once the maximiser is shown to be reliable.

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

2 major / 4 minor

Summary. The paper presents a proof-of-principle application of frequentist profile-likelihood inference to neutron-star equatorial radius recovery with the X-PSI pulse-profile modeling package. Using the same ST-U hot-spot model, NSX atmosphere, and oblate-Schwarzschild ray-tracing as the Bayesian MultiNest analysis of Hoogkamer et al. (H25), the authors fix R_eq and maximize the likelihood over the remaining 12 nuisance parameters with Nelder-Mead. On the single synthetic NICER+XMM data set syntX1 they recover R = 12.096 +0.064/-0.182 km (68 % CL), consistent with the injected value 12.176 km at 0.4σ, with precision comparable to the Bayesian 68 % interval while requiring only ~9 CPU-core hours versus ~3840. Codes are released publicly.

Significance. If the result holds under broader testing, the work supplies a computationally inexpensive, prior-independent complement to the Bayesian pipeline that currently underpins NICER mass-radius constraints and equation-of-state inferences. Explicit recovery of a known injected radius, identical likelihood implementation for both methods, and public release of the frequentist X-PSI interface are concrete strengths that make the comparison reproducible and useful to the community.

major comments (2)
  1. Sect. III (final paragraph) and the abstract claim that profile likelihood recovers the true radius to <1σ with precision comparable to MultiNest. Both statements rest on a single realization (syntX1) under low-resolution X-PSI settings. The manuscript itself notes that an ensemble of realizations would be needed for a robust comparison; without it the quoted 0.4σ recovery and the precision claim remain anecdotal rather than statistically established.
  2. Sect. II.B: the profile is constructed by a single Nelder-Mead maximization of the 12-dimensional nuisance likelihood at each fixed R_eq. H25 already reports a flat likelihood surface and additional degeneracies for R ≳ 11 km. No multi-start, annealing, or cross-check against the known Bayesian maximum is provided to demonstrate that the reported global minimum of Δχ^{2} is not a local artifact. Incomplete maximization would directly bias the Δχ^{2} curve and the confidence intervals that constitute the central result.
minor comments (4)
  1. Fig. 2 caption and axis labels: the vertical axis is written as “2” rather than Δχ^{2}; the horizontal dashed lines are labeled “68% (Δ 2 = 1)” etc. Standardize the notation throughout.
  2. Sect. II.B: the list of 12 nuisance parameters is given in prose; a short table or explicit enumeration would improve clarity and reproducibility.
  3. The abstract and conclusions state recovery “to <1σ”; the body (Sect. III) gives 0.4σ using the larger half of the asymmetric interval. Make the precise definition of σ consistent.
  4. Acknowledgements mention use of Anthropic Claude for code changes; a brief statement of which modules were AI-assisted would be useful for provenance.

Circularity Check

0 steps flagged

No circularity: injected-radius recovery is an independent validation test on synthetic data, not a quantity forced by construction or self-citation.

full rationale

The paper’s central claim is an empirical recovery test: on the fixed synthetic data set syntX1 (taken from the independent H25 study), the profile-likelihood maximizer over the 12 nuisance parameters yields a best-fit equatorial radius 12.096 km that lies 0.4σ from the known injected value 12.176 km, with a Δχ² curve whose 68 % interval width is comparable to the MultiNest credible interval. The injected radius is never used as a free parameter, prior, or constraint inside the likelihood; it appears only as an external benchmark after the optimization is complete. The profile-likelihood definition itself (Eq. 1 and the subsequent Δχ² construction) is the standard Wilks construction and does not encode the target radius. Citations to the authors’ earlier profile-likelihood papers supply only methodological background and are not load-bearing for the numerical recovery. No uniqueness theorem, ansatz, or fitted quantity is re-labeled as a prediction. The acknowledged limitations (single realization, possible local maxima of Nelder-Mead) affect statistical robustness, not circularity. The derivation chain is therefore self-contained against an external ground truth.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 0 invented entities

The central numerical claim rests on the standard X-PSI likelihood, the ST-U hotspot model, Wilks’ theorem for one degree of freedom, and the assumption that Nelder-Mead finds the global maximum of the 12-dimensional nuisance-parameter space for each fixed radius. No new physical entities or free parameters are introduced beyond those already present in the Bayesian pipeline; the only free parameters that affect the reported radius intervals are the 12 nuisance parameters that are maximised away.

free parameters (1)
  • 12 nuisance parameters (M, distance, cos i, two hotspot phases/colatitudes/radii/temperatures, N_H)
    Maximised at each fixed R_eq to construct the profile likelihood; their best-fit values are not reported but directly determine the height of L(R).
axioms (3)
  • standard math Δχ² = -2 ln(L_profile / L_max) follows a χ² distribution with 1 degree of freedom (Wilks’ theorem)
    Used in Sect. II.B to convert the profile-likelihood curve into 68 % (Δχ²=1) and 95 % (Δχ²=3.84) confidence intervals.
  • domain assumption The ST-U two-hotspot model with NSX atmosphere and oblate-Schwarzschild ray-tracing correctly describes the synthetic pulse profiles
    Inherited unchanged from Hoogkamer et al. (2025) and used for both Bayesian and frequentist pipelines (Sect. II).
  • ad hoc to paper Nelder-Mead simplex finds the global maximum of the 12-dimensional nuisance likelihood for each fixed radius
    No multi-start or global-optimisation diagnostics are reported; local maxima would bias the profile (Sect. II.B).

pith-pipeline@v1.1.0-grok45 · 11180 in / 2391 out tokens · 30055 ms · 2026-07-14T06:37:49.686610+00:00 · methodology

0 comments
read the original abstract

In recent years, NICER data has been extensively used to determine neutron star radius and mass using pulse profile modeling. The pulse profile modeling is implemented with the {\tt X-PSI} package and best-fit parameters are obtained using Bayesian inference. Using synthetic data, we demonstrate the application of frequentist inference to determine the neutron star radius, where the nuisance parameters are treated using profile likelihood. We find that the profile likelihood technique can recover the true radius to $< 1\sigma$. Its precision is also comparable to that of Bayesian analysis while being computationally much faster. Therefore, this work serves as a proof-of-principle application of frequentist inference to determine neutron star radius using pulse profile modeling and complements the Bayesian inference technique used. We have also made our analysis codes for frequentist inference using {\tt X-PSI} publicly available.

Figures

Figures reproduced from arXiv: 2607.11154 by Shantanu Desai, Vyaas Ramakrishnan.

Figure 1
Figure 1. Figure 1: FIG. 1: Marginalized (68% and 95%) credible intervals (obtained using the [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: ∆ [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

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

28 extracted references · 20 linked inside Pith

  1. [1]

    K. C. Gendreau, Z. Arzoumanian, P. W. Adkins, C. L. Albert, J. F. Anders, A. T. Aylward, C. L. Baker, E. R. Balsamo, W. A. Bamford, S. S. Benegalrao, et al., inSpace Telescopes and Instrumentation 2016: Ultraviolet to Gamma Ray, edited by J.-W. A. den Herder, T. Taka- hashi, and M. Bautz (2016), vol. 9905 ofSociety of Photo-Optical Instrumentation Enginee...

    2016

  2. [2]

    A. L. Watts, N. Andersson, D. Chakrabarty, M. Feroci, K. Hebeler, G. Israel, F. K. Lamb, M. C. Miller, S. Morsink, F. ¨Ozel, et al., Reviews of Modern Physics88, 021001 (2016), 8 1602.01081

    Pith/arXiv arXiv 2016

  3. [3]

    Gonz´ alez-Caniulef, S

    D. Gonz´ alez-Caniulef, S. Guillot, P. Stammler, L. Mauviard, C. Kazantsev, A. L. Watts, D. Choudhury, B. Dorsman, M. Hoogkamer, D. Huppenkothen, et al., arXiv e-prints arXiv:2607.03721 (2026), 2607.03721

    Pith/arXiv arXiv 2026

  4. [4]

    ¨Ozel and P

    F. ¨Ozel and P. Freire, Annual Review of Astron. and Astrophys.54, 401 (2016), 1603.02698

    Pith/arXiv arXiv 2016

  5. [5]

    T. E. Riley, D. Choudhury, T. Salmi, S. Vinciguerra, Y. Kini, B. Dorsman, A. L. Watts, D. Huppenkothen, and S. Guillot,X-PSI: A Python package for neutron star X-ray pulse simulation and inference(2023)

    2023

  6. [6]

    AlGendy and S

    M. AlGendy and S. M. Morsink, Astrophys. J.791, 78 (2014), 1404.0609

    Pith/arXiv arXiv 2014

  7. [7]

    Bogdanov, G

    S. Bogdanov, G. B. Rybicki, and J. E. Grindlay, Astrophys. J.670, 668 (2007), astro- ph/0612791

    arXiv 2007

  8. [8]

    Trotta, arXiv e-prints arXiv:1701.01467 (2017), 1701.01467

    R. Trotta, arXiv e-prints arXiv:1701.01467 (2017), 1701.01467

    Pith/arXiv arXiv 2017

  9. [9]

    Feroz and M

    F. Feroz and M. P. Hobson, MNRAS384, 449 (2008), 0704.3704

    Pith/arXiv arXiv 2008

  10. [10]

    Salmi, D

    T. Salmi, D. Choudhury, Y. Kini, T. E. Riley, S. Vinciguerra, A. L. Watts, M. T. Wolff, Z. Ar- zoumanian, S. Bogdanov, D. Chakrabarty, et al., Astrophys. J.974, 294 (2024), 2406.14466

    Pith/arXiv arXiv 2024

  11. [11]

    A. J. Dittmann, M. C. Miller, F. K. Lamb, I. M. Holt, C. Chirenti, M. T. Wolff, S. Bogdanov, S. Guillot, W. C. G. Ho, S. M. Morsink, et al., Astrophys. J.974, 295 (2024), 2406.14467

    Pith/arXiv arXiv 2024

  12. [12]

    Mauviard, S

    L. Mauviard, S. Guillot, T. Salmi, D. Choudhury, B. Dorsman, D. Gonz´ alez-Caniulef, M. Hoogkamer, D. Huppenkothen, C. Kazantsev, Y. Kini, et al., Astrophys. J.995, 60 (2025), 2506.14883

    arXiv 2025

  13. [13]

    Hoogkamer, Y

    M. Hoogkamer, Y. Kini, T. Salmi, A. L. Watts, and J. Buchner, Phys. Rev. D112, 023008 (2025), 2502.13682

    arXiv 2025

  14. [14]

    Ashton and C

    G. Ashton and C. Talbot, MNRAS507, 2037 (2021), 2106.08730

    Pith/arXiv arXiv 2037

  15. [15]

    Buchner, The Journal of Open Source Software6, 3001 (2021), 2101.09604

    J. Buchner, The Journal of Open Source Software6, 3001 (2021), 2101.09604

    Pith/arXiv arXiv 2021

  16. [16]

    Buchner, inPhysical Sciences Forum(2022), vol

    J. Buchner, inPhysical Sciences Forum(2022), vol. 5 ofPhysical Sciences Forum, p. 46, 2211.09426

    Pith/arXiv arXiv 2022

  17. [17]

    G´ omez-Valent, Phys

    A. G´ omez-Valent, Phys. Rev. D106, 063506 (2022), 2203.16285

    Pith/arXiv arXiv 2022

  18. [18]

    Herold, E

    L. Herold, E. G. M. Ferreira, and E. Komatsu, Astrophys. J. Lett.929, L16 (2022), 2112.12140

    Pith/arXiv arXiv 2022

  19. [19]

    Campeti and E

    P. Campeti and E. Komatsu, Astrophys. J.941, 110 (2022), 2205.05617

    Pith/arXiv arXiv 2022

  20. [20]

    Karwal, Y

    T. Karwal, Y. Patel, A. Bartlett, V. Poulin, T. L. Smith, and D. N. Pfeffer, arXiv e-prints arXiv:2401.14225 (2024), 2401.14225. 9

    Pith/arXiv arXiv 2024

  21. [21]

    Barua, S

    S. Barua, S. Desai, M. Lopez-Hernandez, and E. ´O. Colg´ ain, European Physical Journal C 86, 47 (2026), 2508.10633

    arXiv 2026

  22. [22]

    Herold, E

    L. Herold, E. G. M. Ferreira, and L. Heinrich, Phys. Rev. D111, 083504 (2025), 2408.07700

    Pith/arXiv arXiv 2025

  23. [23]

    Ramakrishnan and S

    V. Ramakrishnan and S. Desai, Universe11, 183 (2025), 2502.00805

    Pith/arXiv arXiv 2025

  24. [24]

    Navas et al

    S. Navas et al. (Particle Data Group), Phys. Rev. D110, 030001 (2024)

    2024

  25. [25]

    W. C. G. Ho and D. Lai, MNRAS327, 1081 (2001), astro-ph/0104199

    Pith/arXiv arXiv 2001

  26. [26]

    W. C. G. Ho and C. O. Heinke, Nature (London)462, 71 (2009), 0911.0672

    Pith/arXiv arXiv 2009

  27. [27]

    S. S. Wilks, The annals of mathematical statistics9, 60 (1938)

    1938

  28. [28]

    Lewis, JCAP2025, 025 (2025), 1910.13970

    A. Lewis, JCAP2025, 025 (2025), 1910.13970. 10

    Pith/arXiv arXiv 2025