arxiv: 2604.16768 · v1 · submitted 2026-04-18 · 🌌 astro-ph.HE

Recognition: unknown

GeV emission in the region of Vela: a new view of the supernova remnant

Miguel Araya , Santiago Ram\'irez , Diego Bueso , Braulio J. Solano-Rojas

Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:39 UTC · model grok-4.3

classification 🌌 astro-ph.HE

keywords Vela supernova remnantGeV gamma raysFermi-LAThadronic emissionextended sourcemachine learning classificationspurious sourcescosmic ray acceleration

0 comments

The pith

Most GeV point sources near Vela are spurious and the emission comes from the extended supernova remnant itself.

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

The paper reanalyzes Fermi-LAT data covering the Vela supernova remnant, where dozens of unidentified point sources appear in existing catalogs. Two independent machine learning classifiers are used to test whether these sources match the properties of known pulsars or active galactic nuclei. Most do not, and after their modeled contribution is removed a large residual flux remains whose shape follows the SNR shell. Spectral fitting favors a hadronic model, with brighter emission in the denser northeastern region. A reader would care because this reframes a nearby SNR as a clear site of hadronic gamma-ray production rather than a collection of unrelated point-like objects.

Core claim

We conclude that the majority of the cataloged point sources are likely spurious, and the GeV gamma rays come from an extended source, which we argue is the counterpart of the Vela SNR. Adopting a simple morphology given by a uniform disk for the emission the resulting extension is 6.5 deg. The northeastern portion of G263.9-3.3, where the ambient density is thought to be higher, is brighter in gamma rays than the south and west. The spectrum of the emission is best fit with a hadronic model.

What carries the argument

Two independent machine learning algorithms that classify unidentified Fermi-LAT point sources by comparing their properties to known pulsar and AGN populations, followed by subtraction of any real point-source contribution and spectral modeling of the residual emission with leptonic and hadronic processes.

If this is right

The Vela SNR produces extended GeV emission whose morphology follows the remnant shell.
Emission is brighter where ambient density is higher, consistent with hadronic interactions.
The overall gamma-ray source has an angular extension of 6.5 degrees when modeled as a uniform disk.
The spectrum is best described by a hadronic rather than leptonic process.

Where Pith is reading between the lines

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

The same classification approach could be applied to other SNRs that appear to host many point sources, potentially revealing more extended hadronic emission.
Confirmation would add weight to the view that supernova remnants are important accelerators of Galactic cosmic rays.
Deeper observations with higher angular resolution could test whether any faint real point sources remain embedded in the extended emission.

Load-bearing premise

The machine learning classifiers correctly determine that the cataloged sources do not belong to the pulsar or AGN populations and that the remaining residual emission is produced by the SNR rather than by unrecognized sources or background.

What would settle it

Detection of variability, pulsations, or AGN-like spectral features from any of the cataloged sources, or a residual spectrum that fits a leptonic model significantly better than the hadronic one.

Figures

Figures reproduced from arXiv: 2604.16768 by Braulio J. Solano-Rojas, Diego Bueso, Miguel Araya, Santiago Ram\'irez.

**Figure 2.** Figure 2: Top. TS map of the Vela region in the energy range 1 − 100 GeV (note the scale is not linear). Bottom. Residual PS map (see text). For both maps the contours are the same as in [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Figure 3: Top: Gamma-ray fluxes and the hadronic model. The hadronic distribution used is a simple power-law in momentum. Bottom: leptonic model for the radio and gamma-ray emission from Vela for a particle momentum distribution which is a power-law with an exponential cutoff (the nonthermal X-ray flux measured by eROSITA is shown for comparison). from the object and reaching any atomic gas present. However, sever… view at source ↗

read the original abstract

The Vela supernova remnant (SNR), G263.9-3.3, and its pulsar wind nebula (PWN), Vela X, is one of the closest such systems, and it has been studied using observations across the electromagnetic spectrum. SNRs are known sources of gamma rays with energies from GeV to the TeV range. In the GeV band, a cluster of cataloged unidentified Fermi-LAT point sources are found across the large angular extension of the Vela SNR. We aim to search for a high-energy signature associated to the SNR. We applied two independent machine learning algorithms to classify unidentified point sources in the Vela region by comparing their properties to those of known populations of Fermi pulsars and active galactic nuclei. We analyzed LAT data and modeled the spectrum of any emission attributable to Vela using leptonic and hadronic processes typical of SNRs. We find that most of the "point sources" cataloged within the extent of Vela do not share characteristics with those of the two most common Fermi point-like source populations and that even after the emission attributed to these "point sources" is subtracted, considerable residual emission is seen throughout Vela. Morphologically, most of the GeV emission is found within the shell of the SNR. We conclude that the majority of the cataloged point sources are likely spurious, and the GeV gamma rays come from an extended source, which we argue is the counterpart of the Vela SNR. Adopting a simple morphology given by a uniform disk for the emission the resulting extension is 6.5 deg. The northeastern portion of G263.9-3.3, where the ambient density is thought to be higher, is brighter in gamma rays than the south and west. The spectrum of the emission is best fit with a hadronic model. These facts make the hadronic origin for the gamma rays more likely.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper claims most Fermi point sources near Vela are spurious and the residual GeV emission is extended hadronic radiation from the SNR, reached via ML classification plus spectral fitting, but the classifiers lack any reported performance checks.

read the letter

The main takeaway is that this work reinterprets the GeV sources in the Vela field as mostly fake, with the real signal coming from an extended source inside the SNR shell. They run two separate machine learning classifiers on the unidentified catalog entries to argue they match neither pulsars nor AGNs, subtract those contributions, and then fit the leftover emission as a uniform disk 6.5 degrees across whose spectrum prefers a hadronic model over leptonic. The northeast part of the remnant, where ambient density is higher, shows brighter emission, which fits the hadronic picture they favor.

Referee Report

3 major / 2 minor

Summary. The paper reanalyzes Fermi-LAT GeV data in the Vela SNR region. It applies two independent machine-learning classifiers to show that most cataloged unidentified point sources do not match pulsar or AGN populations; after subtracting their emission, substantial residual emission remains that is spatially coincident with the SNR shell. The authors model this residual as a uniform disk of 6.5° radius, note brighter emission in the northeast, and find that a hadronic spectral model is preferred over leptonic alternatives, concluding that the GeV emission is the counterpart of the Vela SNR rather than discrete sources.

Significance. If the central claims are substantiated, the work provides evidence that the GeV emission previously attributed to multiple point sources is instead extended hadronic emission from the Vela SNR, supporting hadronic cosmic-ray acceleration in supernova remnants. The application of two independent ML classifiers to vet point-source catalogs is a methodological strength that could be adopted more widely for crowded Galactic-plane fields.

major comments (3)

[Machine-learning classification (methods and results)] The machine-learning classification step lacks any reported cross-validation metrics, feature sets, training/validation splits, or tests on synthetic sources injected into a Vela-like diffuse background. This is load-bearing for the claim that the majority of cataloged point sources are spurious, because misclassification due to source confusion or imperfect diffuse modeling would leave residuals that are not necessarily SNR-related.
[Morphological modeling and residual analysis] The uniform-disk morphology with a fitted radius of 6.5° is adopted without describing the fitting procedure, the likelihood surface, alternative morphologies (e.g., shell or Gaussian), or the statistical significance of the extension relative to a point-source-only model. This choice directly determines the residual map and the subsequent spectral analysis.
[Spectral analysis] The statement that the spectrum is 'best fit with a hadronic model' is presented without quantitative fit statistics (likelihood ratio, TS values, or reduced chi-squared), parameter uncertainties, or explicit comparison tables between hadronic and leptonic models. This weakens the conclusion that hadronic processes are more likely.

minor comments (2)

[Abstract and §3] The abstract and main text refer to 'considerable residual emission' but do not quantify its significance or point to the specific figure showing the residual map after point-source subtraction.
[Throughout] Notation for the uniform-disk radius and the hadronic/leptonic model parameters should be defined consistently between text and any tables or figures.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their careful reading of our manuscript and for providing detailed comments that have improved its quality. We respond to each major comment below and indicate the revisions made to the manuscript.

read point-by-point responses

Referee: [Machine-learning classification (methods and results)] The machine-learning classification step lacks any reported cross-validation metrics, feature sets, training/validation splits, or tests on synthetic sources injected into a Vela-like diffuse background. This is load-bearing for the claim that the majority of cataloged point sources are spurious, because misclassification due to source confusion or imperfect diffuse modeling would leave residuals that are not necessarily SNR-related.

Authors: We agree with the referee that more details on the machine learning approach are essential. The revised manuscript includes an expanded description of the two classifiers, specifying the feature sets employed, the training and validation splits used, cross-validation metrics, and results from synthetic source injection tests in a Vela-like diffuse background. These additions address concerns about source confusion and support the robustness of our finding that most cataloged point sources are spurious. revision: yes
Referee: [Morphological modeling and residual analysis] The uniform-disk morphology with a fitted radius of 6.5° is adopted without describing the fitting procedure, the likelihood surface, alternative morphologies (e.g., shell or Gaussian), or the statistical significance of the extension relative to a point-source-only model. This choice directly determines the residual map and the subsequent spectral analysis.

Authors: We acknowledge this gap in the original submission. In the revision, we now detail the fitting procedure for the uniform disk morphology, present the likelihood surface, compare it to alternative models such as shell and Gaussian profiles, and provide the statistical significance of the extension compared to a point-source model. This justifies the adopted 6.5° radius and its use in generating the residual maps for spectral analysis. revision: yes
Referee: [Spectral analysis] The statement that the spectrum is 'best fit with a hadronic model' is presented without quantitative fit statistics (likelihood ratio, TS values, or reduced chi-squared), parameter uncertainties, or explicit comparison tables between hadronic and leptonic models. This weakens the conclusion that hadronic processes are more likely.

Authors: We thank the referee for this observation. The revised manuscript now includes quantitative fit statistics, such as likelihood ratios and TS values, parameter uncertainties, and a comparison table between the hadronic and leptonic models. These show that the hadronic model provides a better fit, reinforcing our conclusion on the likely hadronic origin of the emission. revision: yes

Circularity Check

0 steps flagged

No significant circularity; analysis chain is self-contained

full rationale

The paper trains two ML classifiers on external known pulsar and AGN populations to reclassify unidentified Fermi sources in the Vela region, subtracts the resulting point-source model, and fits the residual maps to a uniform-disk morphology (6.5 deg radius) plus hadronic spectrum. No step reduces the final claim (spurious sources + extended SNR emission) to a fitted parameter by construction, a self-citation chain, or an ansatz smuggled from prior work. The ML step uses independent training data, the morphological fit is a direct data-driven result after subtraction, and the hadronic preference is a standard model comparison on the extracted spectrum. The derivation therefore stands on external benchmarks and does not collapse to tautology.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract-only review limits visibility of parameters and assumptions. The 6.5-degree disk radius appears to be a fitted morphological parameter. Standard leptonic/hadronic emission models are invoked without new axioms.

free parameters (1)

uniform-disk radius
6.5 deg extension adopted as simple morphology for the residual emission.

axioms (1)

domain assumption Leptonic and hadronic emission processes typical of SNRs can be distinguished by spectral shape.
Invoked when the authors state the spectrum is best fit with a hadronic model.

pith-pipeline@v0.9.0 · 5666 in / 1207 out tokens · 46165 ms · 2026-05-10T07:39:13.866340+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

63 extracted references · 4 canonical work pages · 1 internal anchor

[1]

2016, in 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16), 265–283

Abadi, M., Agarwal, A., Barham, P., et al. 2016, in 12th USENIX Symposium on Operating Systems Design and Implementation (OSDI 16), 265–283

2016
[2]

2024, arXiv e-prints, arXiv:2411.07162 1, 9

Abdollahi, S., Acero, F., Acharyya, A., et al. 2024, arXiv e-prints, arXiv:2411.07162

work page arXiv 2024
[3]

2020, ApJS, 247, 33 Article number, page 8 Miguel Araya et al.: GeV emission in the region of Vela: a new view of the supernova remnant

Abdollahi, S., Acero, F., Ackermann, M., et al. 2020, ApJS, 247, 33 Article number, page 8 Miguel Araya et al.: GeV emission in the region of Vela: a new view of the supernova remnant

2020
[4]

2022, ApJS, 260, 53

Abdollahi, S., Acero, F., Baldini, L., et al. 2022, ApJS, 260, 53

2022
[5]

2016, ApJS, 224, 8

Acero, F., Ackermann, M., Ajello, M., et al. 2016, ApJS, 224, 8

2016
[6]

2012, ApJS, 203, 4

Ackermann, M., Ajello, M., Albert, A., et al. 2012, ApJS, 203, 4

2012
[7]

G., Bazer-Bachi, A

Aharonian, F., Akhperjanian, A. G., Bazer-Bachi, A. R., et al. 2006, A&A, 448, L43

2006
[8]

A., Kelner, S

Aharonian, F. A., Kelner, S. R., & Prosekin, A. Y . 2010, Phys. Rev. D, 82, 043002

2010
[9]

1974, IEEE Transactions on Automatic Control, 19, 716

Akaike, H. 1974, IEEE Transactions on Automatic Control, 19, 716

1974
[10]

2019, in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM), 2623–2631

Akiba, T., Sano, S., Yanase, T., Ohta, T., & Koyama, M. 2019, in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (ACM), 2623–2631

2019
[11]

2001, A&A, 372, 636

Alvarez, H., Aparici, J., May, J., & Reich, P. 2001, A&A, 372, 636

2001
[12]

D., Bania, T

Anderson, L. D., Bania, T. M., Balser, D. S., et al. 2014, ApJS, 212, 1

2014
[13]

2013, MNRAS, 434, 2202

Araya, M. 2013, MNRAS, 434, 2202

2013
[14]

2020, MNRAS, 492, 5980

Araya, M. 2020, MNRAS, 492, 5980

2020
[15]

2024, A&A, 691, A225

Araya, M. 2024, A&A, 691, A225

2024
[16]

2022, MNRAS, 510, 2277

Araya, M., Gutiérrez, L., & Kerby, S. 2022, MNRAS, 510, 2277

2022
[17]

1995, Nature, 373, 587

Aschenbach, B., Egger, R., & Trümper, J. 1995, Nature, 373, 587

1995
[18]

B., Abdo, A

Atwood, W. B., Abdo, A. A., Ackermann, M., et al. 2009, ApJ, 697, 1071

2009
[19]

Fermi Large Area Telescope Fourth Source Catalog Data Release 4 (4FGL-DR4)

Ballet, J., Bruel, P., Burnett, T. H., Lott, B., & The Fermi-LAT collaboration. 2023, arXiv e-prints, arXiv:2307.12546

work page internal anchor Pith review arXiv 2023
[20]

& Malyshev, D

Bhat, A. & Malyshev, D. 2022, A&A, 660, A87

2022
[21]

1999, A&A, 342, 839

Bocchino, F., Maggio, A., & Sciortino, S. 1999, A&A, 342, 839

1999
[22]

2024, A&A, 684, A150

Burger-Scheidlin, C., Brose, R., Mackey, J., et al. 2024, A&A, 684, A150

2024
[23]

A., De Luca, A., Mignani, R

Caraveo, P. A., De Luca, A., Mignani, R. P., & Bignami, G. F. 2001, ApJ, 561, 930

2001
[24]

& Lin, C.-J

Chang, C.-C. & Lin, C.-J. 2011, ACM Trans. Intell. Syst. Technol., 2, 1

2011
[25]

2021, JHEAP, 29, 40

Chiaro, G., Kovacevic, M., & La Mura, G. 2021, JHEAP, 29, 40

2021
[26]

G., Narendra, A., Liodakis, I., & Bogdan, M

Cooper, N., Dainotti, M. G., Narendra, A., Liodakis, I., & Bogdan, M. 2023, MNRAS, 525, 1731

2023
[27]

& Vapnik, V

Cortes, C. & Vapnik, V . 1995, Machine Learning, 20, 273

1995
[28]

Cozzolongo, G., Mitchell, A. M. W., Spencer, S. T., Malyshev, D., & Unbehaun, T. 2025, in Proceedings of the 8th Heidelberg International Symposium on High-Energy Gamma-Ray Astronomy, arXiv:2504.03543

work page arXiv 2025
[29]

E., & McCulloch, P

Dodson, R., Legge, D., Reynolds, J. E., & McCulloch, P. M. 2003, ApJ, 596, 1137

2003
[30]

M., Green, A

Dubner, G. M., Green, A. J., Goss, W. M., Bock, D. C. J., & Giacani, E. 1998, AJ, 116, 813

1998
[31]

2019, MNRAS, 488, 4074

Fang, K., Bi, X.-J., & Yin, P.-F. 2019, MNRAS, 488, 4074

2019
[32]

2021, MNRAS, 507, 4061

Finke, T., Krämer, M., & Manconi, S. 2021, MNRAS, 507, 4061

2021
[33]

A., Bietenholz, M

Frail, D. A., Bietenholz, M. F., & Markwardt, C. B. 1997, ApJ, 475, 224

1997
[34]

2024, MNRAS, 530, 1144

Ge, T.-T., Sun, X.-N., Yang, R.-Z., et al. 2024, MNRAS, 530, 1144

2024
[35]

2025, A&A, 701, A206

Giuffrida, R., Lemoine-Gourmard, M., Miceli, M., et al. 2025, A&A, 701, A206

2025
[36]

H., Romani, R

Grondin, M. H., Romani, R. W., Lemoine-Goumard, M., et al. 2013, ApJ, 774, 110 Géron, A. 2019, Hands-On Machine Learning with Scikit-Learn, Keras, and Ten- sorFlow. Concepts, Tools, and Techniques to Build Intelligent Systems, 2nd edn. (O’Reilly Media, Inc.)

2013
[37]

A., Funk, S., Parsons, R

Hinton, J. A., Funk, S., Parsons, R. D., & Ohm, S. 2011, ApJ, 743, L7

2011
[38]

B., Edwards, R

Hobbs, G. B., Edwards, R. T., & Manchester, R. N. 2006, MNRAS, 369, 655

2006
[39]

2018, ApJ, 866, 143

Huang, Z.-Q., Fang, K., Liu, R.-Y ., & Wang, X.-Y . 2018, ApJ, 866, 143

2018
[40]

M., & Vila, G

Kafexhiu, E., Aharonian, F., Taylor, A. M., & Vila, G. S. 2014, Phys. Rev. D, 90, 123014

2014
[41]

2011, ApJ, 741, 44

Katagiri, H., Tibaldo, L., Ballet, J., et al. 2011, ApJ, 741, 44

2011
[42]

A., & Kelner, S

Khangulyan, D., Aharonian, F. A., & Kelner, S. R. 2014, ApJ, 783, 100

2014
[43]

Kohavi, R. 1995, in Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI) (Montreal, Quebec, Canada: Morgan Kauf- mann), 1137–1145 Kovaˇcevi´c, M., Chiaro, G., Cutini, S., & Tosti, G. 2020, MNRAS, 493, 1926

1995
[44]

2025, ApJ, 988, 200

Lange, A., Eagle, J., Kargaltsev, O., Kuiper, L., & Hare, J. 2025, ApJ, 988, 200

2025
[45]

Lu, F. J. & Aschenbach, B. 2000, A&A, 362, 1083

2000
[46]

Malyshev, D. V . & Bhat, A. 2023, MNRAS, 521, 6195

2023
[47]

Markwardt, C. B. & Ögelman, H. B. 1997, ApJ, 480, L13

1997
[48]

R., Bertsch, D

Mattox, J. R., Bertsch, D. L., Chiang, J., et al. 1996, ApJ, 461, 396

1996
[49]

Mayer, M. G. F., Becker, W., Predehl, P., & Sasaki, M. 2023, A&A, 676, A68

2023
[50]

& Foody, G

Pal, M. & Foody, G. M. 2010, IEEE Transactions on Geoscience and Remote Sensing, 48, 2297

2010
[51]

2011, Journal of Machine Learning Research, 12, 2825

Pedregosa, F., Varoquaux, G., Gramfort, A., et al. 2011, Journal of Machine Learning Research, 12, 2825

2011
[52]

2024, Nature Astronomy, 8, 530

Peron, G., Casanova, S., Gabici, S., Baghmanyan, V ., & Aharonian, F. 2024, Nature Astronomy, 8, 530

2024
[53]

2024, ApJ, 961, 26 Saz Parkinson, P

Rangelov, B., Yang, H., Williams, B., et al. 2024, ApJ, 961, 26 Saz Parkinson, P. M., Xu, H., Yu, P. L. H., et al. 2016, ApJ, 820, 8

2024
[54]

2018, ApJ, 865, 86

Slane, P., Lovchinsky, I., Kolb, C., et al. 2018, ApJ, 865, 86

2018
[55]

& Lapalme, G

Sokolova, M. & Lapalme, G. 2009, Information Processing & Management, 45, 427

2009
[56]

& Hnatyk, B

Sushch, I. & Hnatyk, B. 2014, A&A, 561, A139

2014
[57]

2011, A&A, 525, A154 SWGO Collaboration, Abreu, P., Alfaro, R., et al

Sushch, I., Hnatyk, B., & Neronov, A. 2011, A&A, 525, A154 SWGO Collaboration, Abreu, P., Alfaro, R., et al. 2025, arXiv e-prints, arXiv:2506.01786

work page arXiv 2011
[58]

2018, A&A, 617, A78 V oges, W., Aschenbach, B., Boller, T., et al

Tibaldo, L., Zanin, R., Faggioli, G., et al. 2018, A&A, 617, A78 V oges, W., Aschenbach, B., Boller, T., et al. 1999, A&A, 349, 389

2018
[59]

2017, in International Cosmic Ray Con- ference, V ol

Wood, M., Caputo, R., Charles, E., et al. 2017, in International Cosmic Ray Con- ference, V ol. 301, 35th International Cosmic Ray Conference (ICRC2017), 824

2017
[60]

2020, Astronomy and Computing, 32, 100387

Xiao, H., Cao, H., Fan, J., et al. 2020, Astronomy and Computing, 32, 100387

2020
[61]

2015, in International Cosmic Ray Conference, V ol

Zabalza, V . 2015, in International Cosmic Ray Conference, V ol. 34, 34th Inter- national Cosmic Ray Conference (ICRC2015), 922

2015
[62]

& Elkan, C

Zadrozny, B. & Elkan, C. 2002, in Proceedings of the Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD- 2002), ACM, Edmonton, Alberta, Canada, 694–699

2002
[63]

R., Chen, J

Zhu, K. R., Chen, J. M., Zheng, Y . G., & Zhang, L. 2023, MNRAS, 527, 1794 Article number, page 9

2023