Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

The hunt for new pulsating ultraluminous X-ray sources: a clustering approach

T0 review · 2 major / 4 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 A clustering analysis singles out 85 ULXs that share the signature of known pulsating ULXs, making them prime targets for pulsation searches.

desk verdict Useful ranked list of ULX observations that resemble known PULXs, but the clusters track brightness and variability, not accretor type. read the letter →

arxiv 2507.15032 v1 pith:T23VCTTL submitted 2025-07-20 astro-ph.HE astro-ph.IMcs.AIcs.LG

Nicolò Oreste Pinciroli Vago , Roberta Amato , Matteo Imbrogno , GianLuca Israel , Andrea Belfiore , Konstantinos Kovlakas , Piero Fraternali , Mario Pasquato This is my paper · ORCID
classification astro-ph.HEastro-ph.IMcs.AIcs.LG
keywords ultraluminousX-raysourcespulsatingULXsGaussianmixturemodelsunsupervisedclusteringXMM-Newtonneutronstarssuper-Eddingtonaccretionvariability
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

The paper tries to establish that many ultraluminous X-ray sources (ULXs) that have never shown pulsations are nevertheless hidden pulsars, and that their identity can be predicted from archival catalogue data alone. The authors apply an unsupervised two-cluster Gaussian mixture model to 1769 XMM-Newton observations of 640 ULXs, then use the observations of the six known pulsating ULXs to label the cluster that should contain PULXs. With a threshold that captures at least 99% of known PULX observations, that cluster also contains 355 observations of 85 previously non-pulsating ULXs, about 85% of them observed more than once. The authors conclude that these 85 sources share the multi-dimensional phase-space signature of confirmed PULXs and are therefore the best candidates for deep pulsation searches.

What carries the argument

The engine of the analysis is a two-component Gaussian mixture model, run on scaled ULX observations of unknown accretor type; confirmed PULX observations are not used in the fit but are projected onto it to set the probability threshold (PULX ratio at least 0.99). A shallow decision tree then translates the cluster boundary into three observable cuts: broadband flux brighter than about $4\times10^{-13}\,\mathrm{erg\,cm^{-2}\,s^{-1}}$, peak flux $F_{\mathrm{peak}}$ brighter than about $1\times10^{-12}\,\mathrm{erg\,cm^{-2}\,s^{-1}}$, and a variability flag indicating intra-observational variability. $F_{\mathrm{peak}}$ is the load-bearing feature: the separation is much weaker without it, and it is what lets low-flux observations of known PULXs still be classified by the brightness of other epochs of the same source.

What would settle it

A coherent pulsation search on the 85 candidates using exposures with at least 10,000 photons per source, the known threshold for PULX detection, would settle the claim: if spin signals appear in a substantial subset, the cluster is accretor-selected; if none appear despite statistics equal to the confirmed PULX detections, the cluster is not identifying hidden pulsars. A cheaper cross-check is to run the same two-component Gaussian mixture on a luminosity- and distance-matched sample of known black-hole ULXs and see whether they too fall in the PULX cluster.

Watch

Extended reading notes

Core claim

The central claim is that the known PULXs are not isolated outliers: a two-component Gaussian mixture, fitted only on ULXs of unknown accretor type and thresholded so that at least 99% of confirmed PULX observations fall on one side, places 355 observations of 85 additional ULXs in the same cluster. Those candidates include sources already suspected to be neutron-star accretors, such as NGC 7793 ULX-4 and NGC 4559 X7, as well as some sources previously interpreted as black-hole ULXs. The decisive feature is the maximum observed flux Fpeak: including it raises the uncertain-ratio metric from 0.10–0.11 to 0.79, while luminosity adds nothing. The authors also report that 19 of 22 QPO-bearing observations fall in the PULX cluster, which they read as supporting a link between mHz quasi-periodic oscillations and neutron-star accretors.

Load-bearing premise

The whole candidate list rests on the assumption that the two-cluster split separates neutron-star ULXs from black-hole ULXs rather than separating bright, well-observed sources from faint, poorly observed ones.

Editorial extensions

If this is right

  • The 85 candidates become a prioritized sample for high-statistics pulsation searches; detecting a spin period in any of them would confirm that the cluster boundary tracks accretor type.
  • The method stays stable when known PULX observations are removed one at a time or source by source (average PULX ratio around 97–99%), so new XMM-Newton ULX observations can be classified without refitting the mixture from scratch.
  • Because most candidates have multiple observations and PULX pulsations are seen in only 31 of 95 known-PULX observations, roughly three well-spaced observations per source should give about a 70% chance of catching a pulse if the candidate population behaves like the confirmed one.
  • The concentration of QPO-bearing ULXs in the PULX cluster implies that mHz quasi-periodic oscillations are a practical marker for identifying neutron-star accretors even before pulsations are found.

Reading between the lines

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

  • Extension: if most of the 85 candidates are real PULXs, then the fraction of ULXs powered by neutron stars is far larger than the six confirmed cases suggest, and the super-Eddington neutron-star channel is a dominant, not exceptional, ULX mode.
  • Extension: because the cluster boundary is set by brightness and variability rather than by spectral hardness, the same recipe could be transferred to Chandra and Swift ULX catalogues to build a larger cross-mission candidate list.
  • Extension: the paper does not quantify how many of the 85 are bright black-hole ULXs; a matched control sample of spectroscopically identified black-hole ULXs, clustered in the same feature space, would give a direct contamination estimate.
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

2 major / 4 minor

Summary. The paper applies an unsupervised Gaussian Mixture Model (GMM) to XMM-Newton observations of ultraluminous X-ray sources (ULXs) drawn from the Walton et al. (2022) catalogue, updated with 4XMM-DR13. The algorithm clusters the data into two components, and the cluster containing the known pulsating ULXs (PULXs) is identified by setting a probability threshold that recovers at least 99% of known PULX observations while maximizing the fraction of unknown ULXs placed in the other cluster. The pipeline selects hyperparameters and a threshold on the same dataset. The resulting PULX cluster contains 85 previously unknown candidate PULXs across 355 observations, with a decision tree explaining the classification through high broadband flux, high peak flux, and variability. A preliminary timing search finds no new pulsations in these candidates.

Significance. If the cluster assignment genuinely traces neutron-star accretors, the 85 candidates constitute a valuable, ranked target list for deep pulsation searches and would represent a major step toward quantifying the PULX fraction. The manuscript is transparently written, the pipeline is clearly specified, and the catalogue (Tables B.1 and C.1) is a useful community resource. However, the physical interpretation of the clusters is not established: the separation is driven by brightness and variability, which are the same factors that determine whether pulsations are detectable, and the paper's own internal checks do not rule out that the candidates are simply bright, variable, non-pulsating ULXs. The absence of any new pulsations in the timing search is consistent with this alternative interpretation.

major comments (2)
  1. [Section 2.3.2, Table 4, Fig. 3, Sect. 4.2] The central claim that the PULX cluster contains candidate neutron-star ULXs rests on the assumption that the two GMM components separate NS accretors from BH accretors. The evidence presented does not support this. The decision tree in Fig. 3 and the silhouette indices in Table 4 show that the separation is driven by EP_8_FLUX, Fpeak, and VAR_FLAG, while HRHard has a silhouette index of only 0.039, meaning hardness does not separate the clusters. Furthermore, Sect. 4.2 reports that literature-classified BH ULXs (Holmberg II X-1, NGC 5408 X-1, NGC 7793 P9, M31 ULX-1) fall inside the PULX cluster. The authors themselves note in Sect. 4.1 that all candidate PULXs have the highest Fpeak values in the dataset. These features (peak flux and variability) are exactly the selection function that determines whether pulsations are detectable, independent of accretor type. The paper therefore does not demonstrate that the 85 candidates are preferentially NS accretors; they may be bright, variable ULXs of any accretor type. To make the central claim load-bearing, the authors should show that the cluster separation persists when controlling for flux, distance, and observation count, or that physical spectral indicators (e.g., hardness, spectral shape) contribute beyond the detection-selection features. The current null result from the timing search, while preliminary, is fully consistent with the null hypothesis that the candidates are non-pulsating sources.
  2. [Section 2.3 (Pipeline), Eq. (2), Sect. 3] The probability threshold is selected on the same dataset that is used to evaluate the method: the pipeline (steps 2c and 3) chooses the threshold that maximizes UR subject to PR>=0.99 computed on the known PULX observations. Consequently, the reported PR is a design constraint rather than an independent measure of recall. The leave-one-out robustness checks in Sect. 4.1 only show that this imposed constraint is stable when a small fraction of PULX observations is removed; they do not test whether the cluster assignment generalizes to genuinely unseen sources. For a claim of predictive power and a candidate list that is meant to prioritize future observations, a proper validation scheme is required, for example holding out a random subset of known PULX observations for threshold selection and evaluating on the rest, or performing cross-validation. Without such a validation, the 85 candidates are the result of a model tuned to fit the known PULXs, and their status as robust predictions is unproven.
minor comments (4)
  1. [Sect. 2.2, Eq. (2)] The statement that PR is 'always ≥0.5 by definition' is correct only because the maximum is taken over clusters; this is worth stating more explicitly, since the reader may otherwise interpret PR as a genuine recall metric rather than a constrained optimization objective.
  2. [Sect. 4.2, last paragraph] The timing search is described as using HENaccelsearch with a first period derivative, but no sensitivity estimate (e.g., minimum detectable pulsed fraction as a function of photon count) is given. Reporting this would help the reader interpret the null result.
  3. [Table B.1] Several z-scores are given as '>8.00' or '-8.13' with no explanation of the clipping convention. It would be helpful to specify the range of z-scores considered and what values outside ±8 indicate.
  4. [Sect. 4.1, first paragraph] The statement that 'all candidate PULXs have the highest Fpeak values in the dataset' is a key caveat and should appear earlier, in the Results section, rather than only in the Discussion.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: candidate PULXs are derived from an unsupervised GMM fit with threshold calibration on known PULXs; the central sample is not fitted to the labels.

full rationale

The derivation chain is: build feature vectors from 4XMM-DR13 fluxes (EP_n_FLUX, Fpeak, VAR_FLAG, HRHard); fit a two-component Bayesian GMM on the 640 ULXs of unknown nature only (Sect. 2.3.2); project known PULX observations onto this fixed density and choose a probability threshold maximizing UR subject to PR>=0.99 (Sect. 2.3, step 2c); and count unlabeled observations above the threshold as candidates. The GMM density is therefore independent of the PULX labels, and the 85-candidate count is an output of that density plus a calibrated threshold, not a parameter fitted to reproduce the PULX sample. The PR>=0.99 recovery of known PULXs is a constraint of the threshold choice, not an independent prediction, so it should not be read as evidence that the method 'retrieves' known PULXs; however, the candidate sample itself does not reduce to the input labels. The conclusion that candidates are 'similar' to known PULXs restates the membership rule used to define C_P, a wording limitation rather than a circular derivation. The self-citations (Imbrogno et al. 2024 and in prep. for the QPO hypothesis; Pinciroli Vago & Fraternali 2023 for unsupervised methods) are ancillary and not load-bearing for the main clustering result. The absence of new pulsations in the timing search is an external, albeit negative, check and does not create circularity.

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

The central claim rests on a small labeled set of six pulsars, a fixed two-cluster model, and catalogue fluxes from a simplified spectral model. No new physical entities, forces, or particles are introduced; the paper's output is a ranked candidate list.

free parameters (3)
  • GMM hyperparameter configuration = MinMaxScaler, use_log=False, use_PCA=False, covariance_type='full'
    Selected by maximizing UR across the grid in Table 2 on the same dataset used for evaluation.
  • Probability threshold = Not stated as a number; set by imposing PR>=0.99, roughly log(probability) ~ -0.04
    Determines how many unknown ULXs are admitted to the candidate cluster; chosen using known PULX observations.
  • Number of clusters = 2
    Fixed a priori to represent BH-ULX vs NS-ULX; no model comparison for other counts is performed.
assumptions (4)
  • domain assumption The two GMM clusters correspond to a physical NS-ULX vs BH-ULX split.
    Sect. 2.3.2 and Discussion assume the two clusters have physical meaning; if the split is brightness-driven, the candidate list is biased.
  • domain assumption 4XMM catalogue fluxes, derived from a single absorbed power-law model with fixed NH and photon index, are adequate proxies for ULX spectral properties.
    Stated in Sect. 4.1; the authors note this is not the best model for ULXs and that fluxes are uncorrected for absorption.
  • domain assumption The six known PULXs are a representative sample of the pulsating ULX population.
    Used to set the threshold; only 6 sources and 95 observations, and one known candidate (NGC 7793 ULX-4) is excluded due to weak signal.
  • domain assumption Flux measurement uncertainties can be ignored, treating each value as perfect.
    Explicitly stated in Sect. 4.1, with the admission that this introduces unknown bias.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The hunt for new pulsating ultraluminous X-ray sources: a clustering approach." pith.science (2026). https://pith.science/paper/T23VCTTL

@misc{pith2026250715032,
  author       = {Pith},
  title        = {Pith review of: The hunt for new pulsating ultraluminous X-ray sources: a clustering approach},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T23VCTTL}},
  note         = {Machine review of arXiv:2507.15032}
}
abstract

The discovery of fast and variable coherent signals in a handful of ultraluminous X-ray sources (ULXs) testifies to the presence of super-Eddington accreting neutron stars, and drastically changed the understanding of the ULX class. Our capability of discovering pulsations in ULXs is limited, among others, by poor statistics. However, catalogues and archives of high-energy missions contain information which can be used to identify new candidate pulsating ULXs (PULXs). The goal of this research is to single out candidate PULXs among those ULXs which have not shown pulsations due to an unfavourable combination of factors. We applied an AI approach to an updated database of ULXs detected by XMM-Newton. We first used an unsupervised clustering algorithm to sort out sources with similar characteristics into two clusters. Then, the sample of known PULX observations has been used to set the separation threshold between the two clusters and to identify the one containing the new candidate PULXs. We found that only a few criteria are needed to assign the membership of an observation to one of the two clusters. The cluster of new candidate PULXs counts 85 unique sources for 355 observations, with $\sim$85% of these new candidates having multiple observations. A preliminary timing analysis found no new pulsations for these candidates. This work presents a sample of new candidate PULXs observed by XMM-Newton, the properties of which are similar (in a multi-dimensional phase space) to those of the known PULXs, despite the absence of pulsations in their light curves. While this result is a clear example of the predictive power of AI-based methods, it also highlights the need for high-statistics observational data to reveal coherent signals from the sources in this sample and thus validate the robustness of the approach.

Figures

Figures reproduced from arXiv: 2507.15032 by the authors.

Figure 1
Figure 1. The pipeline for clustering the data and explaining the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. UR and PR as a function of the probability threshold when considering both F [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The DT that describes the output of GMM when keeping F [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: 2D representations of the two clusters considering two pairs of parameters. The blue dots are the observations belonging to [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The distributions of two relevant parameters when neglecting only L [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Distribution of PULXs with respect to the GMM proba [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: UR as a function of the source distances, for di [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Distribution of the candidate PULXs (orange) and of the [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Neutron star envelopes with machine learning: a single-hidden-layer neural network application

    astro-ph.HE 2025-09 accept novelty 6.0 of 10

    A single-hidden-layer neural network reproduces the magnetized neutron star envelope Tb-Ts relation with roughly 3% error and is fast enough for 3D thermal evolution codes.

Reference graph

Works this paper leans on

83 extracted references · 74 canonical work pages · cited by 1 Pith paper

  1. [1]

    2023, Machine Learning for Physics and Astronomy

    Acquaviva, V . 2023, Machine Learning for Physics and Astronomy

    2023

  2. [2]

    Agrawal, V . K. & Nandi, A. 2015, MNRAS, 446, 3926

    2015

  3. [3]

    J., Abdalla, A., & Moustafa, A

    Alashwal, H., El Halaby, M., Crouse, J. J., Abdalla, A., & Moustafa, A. A. 2019, Frontiers in Computational Neuroscience, 13

    2019

  4. [4]

    A., Godet, O., & Middleton, M

    Amato, R., Gúrpide, A., Webb, N. A., Godet, O., & Middleton, M. J. 2023, A&A, 669, A130

    work page 2023

  5. [5]

    2012, WIREs Data Mining and Knowledge Discovery, 2, 340–350

    Assent, I. 2012, WIREs Data Mining and Knowledge Discovery, 2, 340–350

    work page 2012

  6. [6]

    Atapin, K., Fabrika, S., & Caballero-García, M. D. 2019, MNRAS, 486, 2766

    work page 2019

  7. [7]

    2024, MNRAS, 527, 10185

    Atapin, K., Vinokurov, A., Sarkisyan, A., et al. 2024, MNRAS, 527, 10185

    work page 2024

  8. [8]

    2018, HENDRICS: High ENergy Data Reduction Interface from the Command Shell, Astrophysics Source Code Library, record ascl:1805.019

    Bachetti, M. 2018, HENDRICS: High ENergy Data Reduction Interface from the Command Shell, Astrophysics Source Code Library, record ascl:1805.019

    work page 2018

Show all 83 references
  1. [9]

    A., Walton, D

    Bachetti, M., Harrison, F. A., Walton, D. J., et al. 2014, Nature, 514, 202

    2014

  2. [10]

    2024, The Journal of Open Source Software, 9, 7389

    Bachetti, M., Huppenkothen, D., Stevens, A., et al. 2024, The Journal of Open Source Software, 9, 7389

    2024

  3. [11]

    2024, A&A, 682, A94

    Barra, F., Pinto, C., Middleton, M., et al. 2024, A&A, 682, A94

    2024

  4. [12]

    2024, ApJ, 965, 78

    Belfiore, A., Salvaterra, R., Sidoli, L., et al. 2024, ApJ, 965, 78

    2024

  5. [13]

    2020, Knowledge-Based Systems, 194, 105532

    Blanco-Justicia, A., Domingo-Ferrer, J., Martínez, S., & Sánchez, D. 2020, Knowledge-Based Systems, 194, 105532

    2020

  6. [14]

    A., Fürst, F., et al

    Brightman, M., Harrison, F. A., Fürst, F., et al. 2018, Nature Astronomy, 2, 312

    2018

  7. [15]

    J., Xu, Y ., et al

    Brightman, M., Walton, D. J., Xu, Y ., et al. 2020, ApJ, 889, 71

    2020

  8. [16]

    2018, MNRAS, 476, L45

    Carpano, S., Haberl, F., Maitra, C., & Vasilopoulos, G. 2018, MNRAS, 476, L45

    2018

  9. [17]

    Colbert, E. J. M., Heckman, T. M., Ptak, A. F., Strickland, D. K., & Weaver, K. A. 2004, ApJ, 602, 231

    2004

  10. [18]

    Colbert, E. J. M. & Mushotzky, R. F. 1999, ApJ, 519, 89 de Mijolla, D., Frye, C., Kunesch, M., Mansir, J., & Feige, I. 2020, Human- interpretable model explainability on high-dimensional data

    1999

  11. [19]

    2019, Multimedia Tools and Applications, 78, 19305–19323 Article number, page 10 of 16 N

    Deng, Y ., Tang, F., Dong, W., et al. 2019, Multimedia Tools and Applications, 78, 19305–19323 Article number, page 10 of 16 N. O. Pinciroli Vago et al.: The hunt for new PULXs: a clustering approach

    2019

  12. [20]

    C., Griffiths, R

    Dewangan, G. C., Griffiths, R. E., Choudhury, M., Miyaji, T., & Schurch, N. J. 2005, ApJ, 635, 198

    2005

  13. [21]

    1989, ARA&A, 27, 87

    Fabbiano, G. 1989, ARA&A, 27, 87

    1989

  14. [22]

    N., Atapin, K

    Fabrika, S. N., Atapin, K. E., Vinokurov, A. S., & Sholukhova, O. N. 2021, Astrophysical Bulletin, 76, 6 Fürst, F., Walton, D. J., Harrison, F. A., et al. 2016, ApJ, 831, L14 Fürst, F., Walton, D. J., Heida, M., et al. 2021, A&A, 651, A75 Fürst, F., Walton, D. J., Israel, G. L...

    2021

  15. [23]

    C., Roberts, T

    Gladstone, J. C., Roberts, T. P., & Done, C. 2009, MNRAS, 397, 1836 Gúrpide, A., Godet, O., Koliopanos, F., Webb, N., & Olive, J. F. 2021a, A&A, 649, A104 Gúrpide, A., Godet, O., Koliopanos, F., Webb, N., & Olive, J. F. 2021b, A&A, 649, A104

    2009

  16. [24]

    M., Vaughan, S., & Roberts, T

    Heil, L. M., Vaughan, S., & Roberts, T. P. 2009, MNRAS, 397, 1061

    2009

  17. [25]

    Hu, C.-P., Kong, A. K. H., Ng, C. Y ., & Li, K. L. 2018, ApJ, 864, 64

    2018

  18. [26]

    L., et al

    Huppenkothen, D., Bachetti, M., Stevens, A. L., et al. 2019, ApJ, 881, 39

    2019

  19. [27]

    E., Amato, R., et al

    Imbrogno, M., Motta, S. E., Amato, R., et al. 2024, A&A, 689, A284 ISO. 2017, Standards by ISO/IEC JTC 1/SC 42 – Artificial Intelligence

    2024

  20. [28]

    Jain, A. K. & Dubes, R. C. 1988, Algorithms for clustering data (Prentice-Hall, Inc.) Jr., F. J. M. 1951, Journal of the American Statistical Association, 46, 68

    1988

  21. [29]

    Kaaret, P., Feng, H., & Roberts, T. P. 2017, ARA&A, 55, 303

    2017

  22. [30]

    & Baym, G

    Kalogera, V . & Baym, G. 1996, ApJ, 470, L61

    1996

  23. [31]

    2012, A&A, 538, A49

    Kaur, A., Henze, M., Haberl, F., et al. 2012, A&A, 538, A49

    2012

  24. [32]

    A., Lien, A

    Kennea, J. A., Lien, A. Y ., Krimm, H. A., Cenko, S. B., & Siegel, M. H. 2017, The Astronomer’s Telegram, 10809, 1

    2017

  25. [33]

    2023, New A Rev., 96, 101672

    King, A., Lasota, J.-P., & Middleton, M. 2023, New A Rev., 96, 101672

    2023

  26. [34]

    R., Davies, M

    King, A. R., Davies, M. B., Ward, M. J., Fabbiano, G., & Elvis, M. 2001, ApJ, 552, L109

    2001

  27. [35]

    & Salzberg, S

    Kingsford, C. & Salzberg, S. L. 2008, Nature Biotechnology, 26, 1011–1013

    2008

  28. [36]

    2008, Encyclopedia of Public Health (Springer Netherlands), 1484–1484

    Kirch, W., ed. 2008, Encyclopedia of Public Health (Springer Netherlands), 1484–1484

    2008

  29. [37]

    J., et al

    Kovlakas, K., Zezas, A., Andrews, J. J., et al. 2020, MNRAS, 498, 4790

    2020

  30. [38]

    & Moitinho, A

    Krone-Martins, A. & Moitinho, A. 2013, Astronomy & Astrophysics, 561, A57

    2013

  31. [39]

    MacQueen, J. et al. 1967, Some methods for classification and analysis of mul- tivariate observations

    1967

  32. [40]

    K., & Nandi, A

    Majumder, S., Das, S., Agrawal, V . K., & Nandi, A. 2023, MNRAS, 526, 2086

    2023

  33. [41]

    2014, A&A, 570, A13

    Makarov, D., Prugniel, P., Terekhova, N., Courtois, H., & Vauglin, I. 2014, A&A, 570, A13

    2014

  34. [42]

    1997, Future Generation Computer Systems, 13, 135–147

    Michaud, P. 1997, Future Generation Computer Systems, 13, 135–147

    1997

  35. [43]

    J., Sutton, A

    Middleton, M. J., Sutton, A. D., Roberts, T. P., Jackson, F. E., & Done, C. 2012, MNRAS, 420, 2969

    2012

  36. [44]

    2013, ApJ, 771, 133

    Mineo, S., Rappaport, S., Steinhorn, B., et al. 2013, ApJ, 771, 133

    2013

  37. [45]

    R., Cenko, S

    Pasham, D. R., Cenko, S. B., Zoghbi, A., et al. 2015, ApJ, 811, L11

    2015

  38. [46]

    2011, Journal of Machine Learning Research, 12, 2825 Pinciroli Vago, N

    Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825 Pinciroli Vago, N. O. & Fraternali, P. 2023, Neural Computing and Applications, 35, 19253–19277

    2011

  39. [47]

    J., et al

    Pinto, C., Soria, R., Walton, D. J., et al. 2021, MNRAS, 505, 5058

    2021

  40. [48]

    & Walton, D

    Pinto, C. & Walton, D. J. 2023, arXiv e-prints, arXiv:2302.00006

    2023 arXiv

  41. [49]

    J., Kara, E., et al

    Pinto, C., Walton, D. J., Kara, E., et al. 2020, MNRAS, 492, 4646

    2020

  42. [50]

    2025, Astronomy & Astro- physics, 695, A238

    Pintore, F., Pinto, C., Rodriguez-Castillo, G., et al. 2025, Astronomy & Astro- physics, 695, A238

    2025

  43. [51]

    2018, MNRAS, 479, 4271

    Pintore, F., Zampieri, L., Mereghetti, S., et al. 2018, MNRAS, 479, 4271

    2018

  44. [52]

    2017, ApJ, 836, 113

    Pintore, F., Zampieri, L., Stella, L., et al. 2017, ApJ, 836, 113

    2017

  45. [53]

    G., & Abolmasov, P

    Poutanen, J., Lipunova, G., Fabrika, S., Butkevich, A. G., & Abolmasov, P. 2007, MNRAS, 377, 1187 Pérez-Díaz, V . S., Martínez-Galarza, J. R., Caicedo, A., & D’Abrusco, R. 2024, Monthly Notices of the Royal Astronomical Society, 528, 4852–4871

    2007

  46. [54]

    A., Gúrpide, A., Bachetti, M., & Fürst, F

    Quintin, E., Webb, N. A., Gúrpide, A., Bachetti, M., & Fürst, F. 2021, MNRAS, 503, 5485

    2021

  47. [55]

    M., Eikenberry, S

    Ransom, S. M., Eikenberry, S. S., & Middleditch, J. 2002, AJ, 124, 1788

    2002

  48. [56]

    2010, ApJ, 722, 620

    Rao, F., Feng, H., & Kaaret, P. 2010, ApJ, 722, 620

    2010

  49. [57]

    2009, Gaussian Mixture Models (Springer US), 659–663

    Reynolds, D. 2009, Gaussian Mixture Models (Springer US), 659–663

    2009

  50. [58]

    Reynolds, D. A. et al. 2009, Encyclopedia of biometrics, 741 Rodríguez Castillo, G. A., Israel, G. L., Belfiore, A., et al. 2020, ApJ, 895, 60

    2009

  51. [59]

    Rousseeuw, P. J. 1987, Journal of Computational and Applied Mathematics, 20, 53–65

    1987

  52. [60]

    E., et al

    Sacchi, A., Imbrogno, M., Motta, S. E., et al. 2024, A&A, 689, A217

    2024

  53. [61]

    P., Grisé, F., et al

    Sathyaprakash, R., Roberts, T. P., Grisé, F., et al. 2022, MNRAS, 511, 5346

    2022

  54. [62]

    P., Walton, D

    Sathyaprakash, R., Roberts, T. P., Walton, D. J., et al. 2019, MNRAS, 488, L35

    2019

  55. [63]

    Sen, S., Agarwal, S., Chakraborty, P., & Singh, K. P. 2022, Experimental Astron- omy, 53, 1–43

    2022

  56. [64]

    V ., Scaife, A

    Slijepcevic, I. V ., Scaife, A. M. M., Walmsley, M., et al. 2022, Monthly Notices of the Royal Astronomical Society, 514, 2599–2613

    2022

  57. [65]

    2022, Scientific Reports, 12

    Stoitsas, K., Bahulikar, S., de Munter, L., et al. 2022, Scientific Reports, 12

    2022

  58. [66]

    Strohmayer, T. E. & Markwardt, C. B. 2011, The Astronomer’s Telegram, 3121, 1

    2011

  59. [67]

    Strohmayer, T. E. & Mushotzky, R. F. 2009, ApJ, 703, 1386

    2009

  60. [68]

    E., Mushotzky, R

    Strohmayer, T. E., Mushotzky, R. F., Winter, L., et al. 2007, ApJ, 660, 580 Strüder, L., Briel, U., Dennerl, K., et al. 2001, A&A, 365, L18

    2007

  61. [69]

    D., Roberts, T

    Sutton, A. D., Roberts, T. P., & Middleton, M. J. 2013, MNRAS, 435, 1758

    2013

  62. [70]

    J., Kennea, J

    Townsend, L. J., Kennea, J. A., Coe, M. J., et al. 2017, MNRAS, 471, 3878

    2017

  63. [71]

    2024, A&A, 681, A16

    Tranin, H., Webb, N., Godet, O., & Quintin, E. 2024, A&A, 681, A16

    2024

  64. [72]

    S., Doroshenko, V ., Lutovinov, A

    Tsygankov, S. S., Doroshenko, V ., Lutovinov, A. A., Mushtukov, A. A., & Pouta- nen, J. 2017, A&A, 605, A39

    2017

  65. [73]

    S., Gendreau, K

    Vasilopoulos, G., Ray, P. S., Gendreau, K. C., et al. 2020, MNRAS, 494, 5350

    2020

  66. [74]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261

    2020

  67. [75]

    J., Fürst, F., Heida, M., et al

    Walton, D. J., Fürst, F., Heida, M., et al. 2018, ApJ, 856, 128

    2018

  68. [76]

    J., Mackenzie, A

    Walton, D. J., Mackenzie, A. D. A., Gully, H., et al. 2022, MNRAS, 509, 1587

    2022

  69. [77]

    J., Pinto, C., Nowak, M., et al

    Walton, D. J., Pinto, C., Nowak, M., et al. 2020, MNRAS, 494, 6012

    2020

  70. [78]

    2024, IEEE Transactions on Pattern Analysis and Machine Intelligence, 46, 5362

    Wang, L., Zhang, X., Su, H., & Zhu, J. 2024, IEEE Transactions on Pattern Analysis and Machine Intelligence, 46, 5362

    2024

  71. [79]

    A., Coriat, M., Traulsen, I., et al

    Webb, N. A., Coriat, M., Traulsen, I., et al. 2020, A&A, 641, A136

    2020

  72. [80]

    2017, ApJ, 843, 69

    Weng, S.-S., Ge, M.-Y ., Zhao, H.-H., et al. 2017, ApJ, 843, 69

    2017

  73. [81]

    2017, ApJ, 846, 17

    Wiktorowicz, G., Sobolewska, M., Lasota, J.-P., & Belczynski, K. 2017, ApJ, 846, 17

    2017

  74. [82]

    A., Malacaria, C., Jenke, P

    Wilson-Hodge, C. A., Malacaria, C., Jenke, P. A., et al. 2018, ApJ, 863, 9

    2018

  75. [83]

    & Roberts, T

    Zampieri, L. & Roberts, T. P. 2009, MNRAS, 400, 677 Article number, page 11 of 16 A&A proofs:manuscript no. aa53739-25 Appendix A: Observations of known PULXs Table A.1: The predictedz-scores for all the known PULXs in the dataset, grouped by name. OBS_ID Name 4XMMName Pulsati...

    2009

Pith tools

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