Pith. sign in

REVIEW 4 major objections 4 minor 63 references

Identifying Compton-thick AGNs with Machine learning algorithm in Chandra Deep Field-South

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read A Random Forest classifier trained on 210 secure AGNs identifies 67 Compton-thick AGNs in the Chandra Deep Field-South, 64 of them new, raising the field's CT fraction from about 11% to about 20%.

desk verdict Useful candidate list, but the central 'identification' claim is an extrapolation that isn't validated; reframe and it's a solid contribution. read the letter →

arxiv 2505.21105 v1 pith:6KRCCBTZ submitted 2025-05-27 astro-ph.GA

classification astro-ph.GA
keywords Compton-thickAGNactivegalacticnucleicosmicX-raybackgroundRandomForestChandraDeepField-Southlow-luminositystarformationpointsources
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

Compton-thick active galactic nuclei (CT-AGNs) are supermassive black holes accreting behind so much gas that their X-ray emission is nearly or wholly blocked ($N_{\mathrm{H}} \ge 1.5 \times 10^{24}\,\mathrm{cm}^{-2}$), making them underrepresented in X-ray surveys even though models of the cosmic X-ray background say they should be abundant. This paper reports that a Random Forest classifier trained on 210 securely classified AGNs and applied to 254 faint, low-luminosity AGNs in the Chandra Deep Field-South flags 67 of them as CT-AGNs, 64 of which were previously unknown. The identification raises the CT-AGN fraction in the field from about 11% to about 20%, much closer to what synthesis models predict, and the newly found objects' host galaxies show higher star formation rates than non-CT-AGN hosts. The practical consequence is that a large hidden population of obscured AGNs can be recovered from faint X-ray sources using the mid-infrared-to-X-ray ratio rather than requiring high-quality X-ray spectra.

What carries the argument

The load-bearing object is a Random Forest ensemble, an algorithm that averages many decision trees whose node splits use Gini impurity. Its four inputs are quantities that track X-ray absorption: hardness ratio, redshift, rest-frame 6 µm luminosity, and the ratio of 6 µm to 2–10 keV luminosity. The last is the dominant discriminator, contributing 58% of the feature-importance score, with 6 µm luminosity at 22%, hardness ratio at 15%, and redshift at 5%. The purity-selection step is a separate mechanism: the classifier's output probability for secure non-CT-AGNs is fit to a Beta distribution, and the 99.7% right-tail critical value, 0.664, is chosen as the threshold for accepting a candidate as a 'pure' CT-AGN. That threshold is what converts a 90%-accurate classifier into a conservative candidate list meant to contain almost no false positives.

What would settle it

Take the 67 flagged sources and measure their column densities directly from stacked Chandra spectra or NuSTAR observations; if most have $N_{\mathrm{H}} < 1.5 \times 10^{24}\,\mathrm{cm}^{-2}$, the central claim of 64 new Compton-thick AGNs collapses.

Watch

Extended reading notes

Core claim

The paper's central claim is that a large population of Compton-thick AGNs hides among low-luminosity, low-count AGNs in the deepest Chandra field, and that a machine-learning classifier can pull them out. A Random Forest trained on 210 AGNs with secure classifications (62 CT, 148 non-CT) using four inputs—hardness ratio, redshift, 6 µm luminosity, and the 6 µm-to-2–10 keV luminosity ratio—reaches 90% accuracy, 80% precision, and 89% recall on a held-out test set. Applied to 254 AGNs whose classifications are insecure, and filtered with a high-purity probability threshold of 0.664 (the 99.7% right-tail critical value of a Beta distribution fit to the non-CT probabilities), the model identifies 67 CT-AGNs, three of them already known and 64 new. The paper then counts 146 CT-AGNs among the 728 AGNs in CDFS, a fraction of about 20%, and reports that CT-AGN host galaxies have higher star formation rates than non-CT-AGN hosts (median log SFR 1.19 versus 0.95; KS p-value $5.2 \times 10^{-3}$), while stellar mass distributions are statistically indistinguishable.

Load-bearing premise

The one untested load-bearing premise is that the decision boundary learned from bright, securely classified AGNs transfers unchanged to the 254 faint, low-count AGNs, even though the paper does not validate the classifier on that application sample.

Editorial extensions

If this is right

  • The CDFS CT-AGN census grows from 79 to 146 objects, raising the fraction from about 11% to about 20% of the 728 AGNs in the field.
  • At least 64 previously unknown CT-AGNs lie among faint, low-luminosity sources, so surveys that treat low-count X-ray sources as unobscured or non-AGN will systematically miss a large obscured population.
  • The mid-infrared-to-X-ray luminosity ratio is the strongest predictor, which means multiwavelength photometry can preselect CT-AGN candidates for expensive X-ray follow-up.
  • The host galaxies of the CT-AGNs show higher star formation rates than non-CT-AGN hosts, in line with a picture where heavy obscuration accompanies active star formation.

Reading between the lines

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

  • The paper's 0.664 threshold is deliberately conservative: 126 of the 254 faint AGNs have probability above 0.5, so on the model's own output the true hidden CT population may be larger than the 67 reported; this is my inference, not the paper's claim.
  • Because the classifier leans on the MIR-to-X-ray ratio, the 67 identifications are effectively photometric obscuration indicators rather than direct column-density measurements; direct X-ray spectroscopy or NuSTAR observations of the candidates would be the clean confirmation.
  • The same trained classifier could be applied to other deep fields, such as CDF-N or COSMOS, to test whether the CT-AGN deficit relative to CXB models is a general property of deep X-ray surveys.
  • The reported star-formation enhancement could partly reflect selection effects between the bright training set and the faint application set; a control sample matched in redshift, stellar mass, and X-ray luminosity would isolate the physical signal.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper trains a Random Forest classifier on 210 securely classified AGNs in the Chandra Deep Field-South using four features (redshift, hardness ratio, 6 micron luminosity, and the MIR-to-X-ray luminosity ratio), reports a test-set accuracy of 90% and an OOB score of 0.87, and then applies the classifier to 254 low-luminosity AGNs with insecure classifications. Setting a probability threshold of 0.664 chosen from a Beta-distribution fit to the secure sample, the authors classify 67 sources as pure CT-AGNs, three of them previously known, and claim 64 newly identified CT-AGNs. They further use the resulting sample to argue that the CT-AGN fraction in CDFS rises to about 20% and that CT-AGN host galaxies have higher star formation rates than non-CT-AGN hosts.

Significance. If the 64 new classifications were validated, this would be a valuable demonstration of a machine-learning approach to recovering a missing obscured-AGN population in deep X-ray surveys, with implications for the cosmic X-ray background and host-galaxy evolution. The paper uses public CDFS data, states its feature definitions clearly, and provides a machine-readable catalog with probabilistic classifications, which is a useful community resource. However, the central identification claim currently rests on an unvalidated extrapolation of a classifier from bright, securely classified sources to faint, low-count sources, and no independent X-ray spectral confirmation is provided for any of the new candidates. The significance of the result therefore depends critically on whether that validation gap can be closed.

major comments (4)
  1. [Section 4.2 and Section 5.2] The central claim that 67 sources, 64 of them new, are 'successfully identified' as CT-AGNs is not supported by the evidence presented. The probability threshold of 0.664 is calibrated in Section 4.1 on the 210-source secure sample, and the reported 90% accuracy and OOB=0.87 are measured only on that same secure distribution. The 254 low-luminosity AGNs are, by construction in Section 2.4, the sources that fail all secure-classification criteria, so they are systematically fainter and noisier. No X-ray spectral fitting, NuSTAR observation, or any other independent diagnostic is presented for the 67 flagged sources, and no false-positive rate is estimated for this application sample. The phrase 'successfully identified' should be replaced by 'identified as candidates' unless the authors provide spectral confirmation for at least a subsample or otherwise validate the decision boundary on independent known CT-AGNs drawn from the low-luminosity regime.
  2. [Section 5.2] The reported total of 146 CT-AGNs double-counts the three previously known sources that appear among the 67 ML-selected objects. Since the authors state that '67 pure CT-AGNs... includes three already known CT-AGNs', the number of unique CT-AGNs is 79 + 67 - 3 = 143, corresponding to 143/728 = 19.6%, not 146/728 = 20.0%. Similarly, within the 464-source sample, the 62 known CT-AGNs plus 67 ML identifications with three overlapping sources give 126 unique CT-AGNs, which is 27.2%, not 27.8% as reported. These arithmetic discrepancies should be corrected, as they affect the paper's main quantitative conclusion about the CT-AGN fraction.
  3. [Section 4.2 versus Table 1] There is a direct inconsistency between the text and the catalog table. Section 4.2 states that XID 68, 414, and 575 'have been identified as CT-AGN, with NH > 10^24 cm-2', yet Table 1 lists XID 68 with Class-trad = 'non-CT-AGN' and Quality = 'Insecure'. If these sources were indeed known CT-AGNs, they should have satisfied criterion (2) in Section 2.4 and been included in the secure training sample rather than the low-luminosity application sample. This contradiction affects the count of 64 previously unknown CT-AGNs and must be resolved.
  4. [Section 5.3] The host-galaxy comparison uses the 67 ML-classified sources as CT-AGNs without independent confirmation. The reported KS-test result for SFR (p = 5.209e-3) therefore measures the separation between samples defined by the classifier, not by a verified physical classification. Because the dominant feature is the MIR-to-X-ray luminosity ratio (58% importance, Figure 5), and this ratio may correlate with star formation indicators through the 6 micron luminosity or SED decomposition, a classifier bias could produce the claimed SFR difference even if the true CT-AGN fraction among the candidates is lower. The authors should either confirm the labels for a subset or explicitly frame Section 5.3 as a comparison of ML-selected candidates conditional on the model.
minor comments (4)
  1. [Abstract and Section 6] The summary states that the algorithm 'successfully identified 67 new CT-AGNs', but Section 4.2 reports 67 total CT-AGNs including three previously known ones, with 64 new; the wording should be made consistent throughout.
  2. [Throughout] There are several typographical errors, including 'Institude' in the author affiliations and 'expections' in Section 5.2; these should be corrected.
  3. [Table 1] The table notes state that missing count errors correspond to 90% confidence upper limits, but the rows in the printed excerpt do not distinguish upper limits from actual measured values in a visually transparent way; a consistent notation such as arrows or '<' would improve usability.
  4. [Section 3.2] The KS-test is described as checking that the training and test sets come from the same distribution, but the paper does not report the actual p-values; giving them in a table or in the text would strengthen the reproducibility of the split.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the 67-source identification is an out-of-sample model prediction; the unsupported step is distribution extrapolation, not equivalence.

full rationale

The derivation chain is not circular. The Random Forest is trained on 210 securely classified AGNs and evaluated on a held-out 30% test split, giving an accuracy of 0.90; the 0.664 threshold is fixed from the Beta-fitted probability distribution of non-CT sources in that labeled secure sample, before being applied to the 254 low-luminosity AGNs. The 67 flagged sources are therefore out-of-sample predictions, not re-statements of the training labels or of the threshold-fitting data. The dominant feature, log(L6um/L2-10keV), is a known CT-AGN diagnostic, but using it as an input feature does not make the model output identical to that input; the secure labels come from X-ray spectral column densities and prior catalogs. Citations to the authors' own Guo et al. (2020, 2021) work supply SED luminosities and some secure classifications, but these are external published measurements used as inputs, not conclusions imported to force the result. The main weakness, unvalidated extrapolation from bright, high-count sources to faint, low-count sources, is a generalization and calibration risk rather than a circularity. A separate bookkeeping issue is that the total of 146 CT-AGNs double-counts the three known sources contained in the 67, but this is an arithmetic error, not circular reasoning.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central claims rest on two fitted values (the 0.664 threshold and the tuned hyperparameters) and on domain assumptions about label reliability, sample representativeness, and the MIR/X-ray diagnostic. No new physical entities are introduced.

free parameters (2)
  • Probability threshold for CT-AGN classification = 0.664
    Chosen as the 99.7% right-tail quantile of a Beta distribution fitted to non-CT-AGN probabilities in the training/test sample (Section 4.1). All 67 identifications depend on this fitted value.
  • Random Forest hyperparameters = n_estimators=25, max_depth=6, min_samples_leaf=6, min_samples_split=5, max_leaf_nodes=8
    Tuned by 'repeated optimization' on the training/test split (Section 3.3, Table 2); the specific optimization criterion and search procedure are not described.
assumptions (5)
  • domain assumption The 210 secure classifications used for training are correct.
    The model's ground truth comes from previous studies (Liu et al. 2017; Li et al. 2019; Corral et al. 2019; Guo et al. 2021) and the >300 net-count criterion (Section 2.4); errors in these labels propagate directly into the classifier.
  • domain assumption The training sample is representative of the application sample.
    Section 4.2 applies the model to 254 low-luminosity AGNs, but the authors only check KS-test agreement between training and test splits, not between training and the application sample.
  • domain assumption The MIR-to-X-ray luminosity ratio is a valid proxy for Compton-thick absorption.
    Section 3.1 uses L6/L2-10 as an input and Section 5.1 finds it is the dominant feature (58%); this is a literature-based diagnostic, not independently validated here.
  • domain assumption Classifier probabilities follow a Beta distribution.
    Section 4.1 fits Beta distributions to the probability outputs of the training/test sample to set the 0.664 threshold; if the Beta form is wrong, the threshold and the 67 count are unsupported.
  • standard math Random Forest is a valid classifier for this task.
    The algorithm is a standard ensemble method (Breiman 2001; Section 3.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identifying Compton-thick AGNs with Machine learning algorithm in Chandra Deep Field-South." pith.science (2026). https://pith.science/paper/6KRCCBTZ

@misc{pith2026250521105,
  author       = {Pith},
  title        = {Pith review of: Identifying Compton-thick AGNs with Machine learning algorithm in Chandra Deep Field-South},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6KRCCBTZ}},
  note         = {Machine review of arXiv:2505.21105}
}
abstract

Compton-thick active galactic nuclei (CT-AGNs), which are defined by column density $\mathrm{N_H} \geqslant 1.5 \times 10^{24} \ \mathrm{cm}^{-2}$, emit feeble X-ray radiation, even undetectable by X-ray instruments. Despite this, the X-ray emissions from CT-AGNs are believed to be a substantial contributor to the cosmic X-ray background (CXB). According to synthesis models of AGNs, CT-AGNs are expected to make up a significant fraction of the AGN population, likely around 30% or more. However, only $\sim$11% of AGNs have been identified as CT-AGNs in the Chandra Deep Field-South (CDFS). To identify hitherto unknown CT-AGNs in the field, we used a Random Forest algorithm for identifying them. First, we build a secure classified subset of 210 AGNs to train and evaluate our algorithm. Our algorithm achieved an accuracy rate of 90% on the test set after training. Then, we applied our algorithm to an additional subset of 254 AGNs, successfully identifying 67 CT-AGNs within this group. This result significantly increased the fraction of CT-AGNs in the CDFS, which is closer to the theoretical predictions of the CXB. Finally, we compared the properties of host galaxies between CT-AGNs and non-CT-AGNs and found that the host galaxies of CT-AGNs exhibit higher levels of star formation activity.

Figures

Figures reproduced from arXiv: 2505.21105 by the authors.

Figure 1
Figure 1. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. The Random Forest algorithm, with its principle shown in a cartoon diagram. ensure the experiment’s reproducibility. By limiting the number of features considered at each split, we increase the randomness and generalization ability of the model. We control the growth of the tree by limiting the maxi￾mum depth and the number of leaf nodes to prevent the tree from becoming overly complex. With these adjust￾ments, we a… view at source ↗
Figure 3
Figure 3. The confusion matrix computed from this test set shows the overall counts. The test set contains 63 AGNs. the number of false negatives. The positive condition is defined as CT-AGN, and the negative condition is defined as non-CT-AGN. The higher the accuracy, the more precise our algorithm is in their classification. The higher the precision, the fewer CT-AGNs are misdiag￾nosed when using our algorithm for classific… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: The display of the importance of all parameters. The MIR-X-ray Luminosity Ratio is the preeminent factor contributing to the algorithm’s predictive capacity, with the 6 µm luminosity closely trailing behind. most important input parameter for distinguishing be￾tween CT…
Figure 6
Figure 6. Figure 6: The relationship of the main sequence in AGN host galaxies. The solid red stars indicate the CT-AGNs, and the open gray circles represent non-CT-AGNs. with a significant amount of dense gas (e.g., Kocevski et al. 2015; Ricci et al. 2017). The radiation from AGNs can co…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 13 canonical work pages

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter doi edition editor eprint howpublished institution journal key month number organization pages publisher school series title misctitle type volume year version url label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts ...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION format.url url empty "" new.block "" url * "" * if FUNCTION format.eprint eprint empty "" archivePrefix empty "" archivePrefix "arXiv" = new.block " " eprint * " " * new.block " " eprint * " " * if if if FUNCTION format.doi doi empty "" " " doi * " " * if FUNCTION format.pid doi empty eprint empty ur...

  3. [4]

    2008, , 689, 666, 10.1086/592595

    Ajello , M., Greiner , J., Sato , G., et al. 2008, , 689, 666, 10.1086/592595

  4. [5]

    T., Treister , E., Urry , C

    Ananna , T. T., Treister , E., Urry , C. M., et al. 2019, , 871, 240, 10.3847/1538-4357/aafb77

  5. [6]

    1993, , 31, 473, 10.1146/annurev.aa.31.090193.002353

    Antonucci , R. 1993, , 31, 473, 10.1146/annurev.aa.31.090193.002353

  6. [7]

    P., Tollerud , E

    Astropy Collaboration , Robitaille , T. P., Tollerud , E. J., et al. 2013, , 558, A33, 10.1051/0004-6361/201322068

  7. [8]

    M., Sip o cz , B

    Astropy Collaboration , Price-Whelan , A. M., Sip o cz , B. M., et al. 2018, , 156, 123, 10.3847/1538-3881/aabc4f

  8. [9]

    G., Benson , A

    Bower , R. G., Benson , A. J., Malbon , R., et al. 2006, , 370, 645, 10.1111/j.1365-2966.2006.10519.x

Show all 63 references
  1. [10]

    2001, Machine Learning, 45, 5, 10.1023/A:1010933404324

    Breiman , L. 2001, Machine Learning, 45, 5, 10.1023/A:1010933404324

  2. [11]

    2015, , 802, 89, 10.1088/0004-637X/802/2/89

    Buchner , J., Georgakakis , A., Nandra , K., et al. 2015, , 802, 89, 10.1088/0004-637X/802/2/89

  3. [12]

    2011, , 728, 58, 10.1088/0004-637X/728/1/58

    Burlon , D., Ajello , M., Greiner , J., et al. 2011, , 728, 58, 10.1088/0004-637X/728/1/58

  4. [13]

    2006, , 446, 459, 10.1051/0004-6361:20053893

    Cappi , M., Panessa , F., Bassani , L., et al. 2006, , 446, 459, 10.1051/0004-6361:20053893

  5. [14]

    L., & Walter , F

    Carilli , C. L., & Walter , F. 2013, , 51, 105, 10.1146/annurev-astro-082812-140953

  6. [15]

    2004, Compton Thick AGN: the dark side of the X-ray background (Springer Netherlands), 245--272

    Comastri, A. 2004, Compton Thick AGN: the dark side of the X-ray background (Springer Netherlands), 245--272

  7. [16]

    2019, , 629, A133, 10.1051/0004-6361/201833799

    Corral , A., Georgantopoulos , I., Akylas , A., & Ranalli , P. 2019, , 629, A133, 10.1051/0004-6361/201833799

  8. [17]

    N., et al

    Ding , N., Luo , B., Brandt , W. N., et al. 2018, , 868, 88, 10.3847/1538-4357/aaea60

  9. [18]

    2000, , 539, L9, 10.1086/312838

    Ferrarese , L., & Merritt , D. 2000, , 539, L9, 10.1086/312838

  10. [19]

    2010, , 519, A92, 10.1051/0004-6361/201014039

    Gilli , R., Vignali , C., Mignoli , M., et al. 2010, , 519, A92, 10.1051/0004-6361/201014039

  11. [20]

    2022, , 666, A17, 10.1051/0004-6361/202243708

    Gilli , R., Norman , C., Calura , F., et al. 2022, , 666, A17, 10.1051/0004-6361/202243708

  12. [21]

    D., Alexander , D

    Goulding , A. D., Alexander , D. M., Bauer , F. E., et al. 2012, , 755, 5, 10.1088/0004-637X/755/1/5

  13. [22]

    2020, , 492, 1887, 10.1093/mnras/stz3589

    Guo , X., Gu , Q., Ding , N., Contini , E., & Chen , Y. 2020, , 492, 1887, 10.1093/mnras/stz3589

  14. [23]

    2021, , 908, 169, 10.3847/1538-4357/abd0f5

    Guo , X., Gu , Q., Ding , N., Yu , X., & Chen , Y. 2021, , 908, 169, 10.3847/1538-4357/abd0f5

  15. [24]

    2023, , 135, 014102, 10.1088/1538-3873/acb294

    Guo , X., Gu , Q., Xu , J., et al. 2023, , 135, 014102, 10.1088/1538-3873/acb294

  16. [25]

    1991, , 380, L51, 10.1086/186171

    Haardt , F., & Maraschi , L. 1991, , 380, L51, 10.1086/186171

  17. [26]

    A., Craig , W

    Harrison , F. A., Craig , W. W., Christensen , F. E., et al. 2013, , 770, 103, 10.1088/0004-637X/770/2/103

  18. [27]

    F., Hernquist , L., Cox , T

    Hopkins , P. F., Hernquist , L., Cox , T. J., et al. 2006, , 163, 1, 10.1086/499298

  19. [28]

    Hunter , J. D. 2007, Computing in Science and Engineering, 9, 90, 10.1109/MCSE.2007.55

  20. [29]

    S., Miller , J

    Kammoun , E. S., Miller , J. M., Zoghbi , A., et al. 2019, , 877, 102, 10.3847/1538-4357/ab1c5f

  21. [30]

    Karson, M., Chakravarti, I., Laha, R., & Roy, J. 1968, J. Am. Stat. Assoc., 63, 1047

  22. [31]

    Y., & Weedman , D

    Khachikian , E. Y., & Weedman , D. W. 1974, , 192, 581, 10.1086/153093

  23. [32]

    D., Brightman , M., Nandra , K., et al

    Kocevski , D. D., Brightman , M., Nandra , K., et al. 2015, , 814, 104, 10.1088/0004-637X/814/2/104

  24. [33]

    Kormendy , J., & Ho , L. C. 2013, , 51, 511, 10.1146/annurev-astro-082708-101811

  25. [34]

    M., Vignali , C., Lanzuisi , G., Gruppioni , C., & Pozzi , F

    La Caria , M. M., Vignali , C., Lanzuisi , G., Gruppioni , C., & Pozzi , F. 2019, , 487, 1662, 10.1093/mnras/stz1381

  26. [35]

    M., Yaqoob , T., Boorman , P

    LaMassa , S. M., Yaqoob , T., Boorman , P. G., et al. 2019, , 887, 173, 10.3847/1538-4357/ab552c

  27. [36]

    L., Chiaberge , M., Heckman , T., et al

    Lambrides , E. L., Chiaberge , M., Heckman , T., et al. 2020, , 897, 160, 10.3847/1538-4357/ab919c

  28. [37]

    B., Gandhi , P., Alexander , D

    Lansbury , G. B., Gandhi , P., Alexander , D. M., et al. 2015, , 809, 115, 10.1088/0004-637X/809/2/115

  29. [38]

    B., Alexander , D

    Lansbury , G. B., Alexander , D. M., Aird , J., et al. 2017, , 846, 20, 10.3847/1538-4357/aa8176

  30. [39]

    2015 a , , 573, A137, 10.1051/0004-6361/201424924

    Lanzuisi , G., Ranalli , P., Georgantopoulos , I., et al. 2015 a , , 573, A137, 10.1051/0004-6361/201424924

  31. [40]

    2015 b , , 578, A120, 10.1051/0004-6361/201526036

    Lanzuisi , G., Perna , M., Delvecchio , I., et al. 2015 b , , 578, A120, 10.1051/0004-6361/201526036

  32. [41]

    2018, , 480, 2578, 10.1093/mnras/sty2025

    Lanzuisi , G., Civano , F., Marchesi , S., et al. 2018, , 480, 2578, 10.1093/mnras/sty2025

  33. [42]

    2023, Experimental Astronomy, 56, 141, 10.1007/s10686-023-09896-7

    Li , H., Walter , R., Produit , N., & Hubert , F. 2023, Experimental Astronomy, 56, 141, 10.1007/s10686-023-09896-7

  34. [43]

    2019, , 877, 5, 10.3847/1538-4357/ab184b

    Li , J., Xue , Y., Sun , M., et al. 2019, , 877, 5, 10.3847/1538-4357/ab184b

  35. [44]

    Liang , E. P. T. 1979, , 234, 1105, 10.1086/157594

  36. [45]

    2017, , 232, 8, 10.3847/1538-4365/aa7847

    Liu , T., Tozzi , P., Wang , J.-X., et al. 2017, , 232, 8, 10.3847/1538-4365/aa7847

  37. [46]

    N., Xue , Y

    Luo , B., Brandt , W. N., Xue , Y. Q., et al. 2017, , 228, 2, 10.3847/1538-4365/228/1/2

  38. [47]

    1998, , 338, 781

    Maiolino , R., Salvati , M., Bassani , L., et al. 1998, , 338, 781. astro-ph/9806055

  39. [48]

    2016, , 817, 34, 10.3847/0004-637X/817/1/34

    Marchesi , S., Civano , F., Elvis , M., et al. 2016, , 817, 34, 10.3847/0004-637X/817/1/34

  40. [49]

    2022, , 935, 114, 10.3847/1538-4357/ac80be

    Marchesi , S., Zhao , X., Torres-Alb \`a , N., et al. 2022, , 935, 114, 10.3847/1538-4357/ac80be

  41. [50]

    2018, , 235, 17, 10.3847/1538-4365/aaa83d

    Masini , A., Civano , F., Comastri , A., et al. 2018, , 235, 17, 10.3847/1538-4365/aaa83d

  42. [51]

    2003, , 588, 696, 10.1086/374335

    Moretti , A., Campana , S., Lazzati , D., & Tagliaferri , G. 2003, , 588, 696, 10.1086/374335

  43. [52]

    2015, , 53, 365, 10.1146/annurev-astro-082214-122302

    Netzer , H. 2015, , 53, 365, 10.1146/annurev-astro-082214-122302

  44. [53]

    2011, Journal of Machine Learning Research, 12, 2825, 10.48550/arXiv.1201.0490

    Pedregosa , F., Varoquaux , G., Gramfort , A., et al. 2011, Journal of Machine Learning Research, 12, 2825, 10.48550/arXiv.1201.0490

  45. [54]

    2020, , 641, A6, 10.1051/0004-6361/201833910

    Planck Collaboration , Aghanim , N., Akrami , Y., et al. 2020, , 641, A6, 10.1051/0004-6361/201833910

  46. [55]

    E., Treister , E., et al

    Ricci , C., Bauer , F. E., Treister , E., et al. 2017, , 468, 1273, 10.1093/mnras/stx173

  47. [56]

    O., & Schmidt , M

    Richstone , D. O., & Schmidt , M. 1980, , 235, 361, 10.1086/157640

  48. [57]

    2023, , 675, A65, 10.1051/0004-6361/202345980

    Silver , R., Torres-Alb \`a , N., Zhao , X., et al. 2023, , 675, A65, 10.1051/0004-6361/202345980

  49. [58]

    2022, , 940, 148, 10.3847/1538-4357/ac9bf8

    Silver , R., Torres-Alb\`a , N., Zhao , X., et al. 2022, , 940, 148, 10.3847/1538-4357/ac9bf8

  50. [59]

    2020, , 888, 8, 10.3847/1538-4357/ab5718

    Toba , Y., Yamada , S., Ueda , Y., et al. 2020, , 888, 8, 10.3847/1538-4357/ab5718

  51. [60]

    2021, , 922, 252, 10.3847/1538-4357/ac1c73

    Torres-Alb \`a , N., Marchesi , S., Zhao , X., et al. 2021, , 922, 252, 10.3847/1538-4357/ac1c73

  52. [61]

    Ueda , Y., Akiyama , M., Hasinger , G., Miyaji , T., & Watson , M. G. 2014, , 786, 104, 10.1088/0004-637X/786/2/104

  53. [62]

    M., & Padovani , P

    Urry , C. M., & Padovani , P. 1995, , 107, 803, 10.1086/133630

  54. [63]

    2022, , 941, 97, 10.3847/1538-4357/ac9c07

    Vijarnwannaluk , B., Akiyama , M., Schramm , M., et al. 2022, , 941, 97, 10.3847/1538-4357/ac9c07

  55. [64]

    A., Fabian , A

    Worsley , M. A., Fabian , A. C., Bauer , F. E., et al. 2005, , 357, 1281, 10.1111/j.1365-2966.2005.08731.x

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.