Pith. sign in

REVIEW 4 major objections 5 minor 41 references

Hunting for the candidates of Changing-Look Blazar using Mclust Clustering Analysis

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that Gaussian mixture clustering of 4FGL-DR3 blazars recovers a distinct intermediate population of changing-look blazar candidates, yielding 111 candidates, 67 of them new.

desk verdict A useful candidate list, but the headline 111 count should not be quoted: the intermediate-position claim survives, the exact number does not. read the letter →

arxiv 2501.00094 v1 pith:KUCZHBCT submitted 2024-12-30 astro-ph.HE

Shi-Ju Kang , Shan-Shan Ren , Yong-Gang Zheng , Qingwen Wu This is my paper · ORCID
classification astro-ph.HE
keywords changing-lookblazarsGaussianmixturemodelclustering4FGL-DR3blazarclassificationFSRQBLLacobjectsactivegalacticnuclei
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 claims that changing-look blazars form a coherent intermediate population between flat-spectrum radio quasars and BL Lac objects in the physical parameter space of gamma-ray, jet, and accretion-disk properties, and that this population can be recovered by unsupervised clustering. The authors apply Gaussian mixture modelling to 2250 blazars from the 4FGL-DR3 catalog, testing all 255 subsets of eight physical parameters. Four parameter combinations yield three-cluster solutions whose agreement with the 105 known changing-look blazars, measured by the adjusted Rand index, exceeds 0.61. The intersection of those four solutions contains 111 changing-look blazar candidates, including 44 previously reported sources and 67 new ones, 56 of which are labelled as FSRQs in the catalog. If the intermediate cluster is real, the changing-look phenomenon is not a handful of curiosities but a distinct, searchable class of active galactic nuclei.

What carries the argument

The central object is the three-component Gaussian mixture model fitted by the paper's mclust algorithm under the EVV covariance parameterization (ellipsoidal clusters of equal volume), with the number of components chosen by the Bayesian information criterion and the partition scored by the adjusted Rand index. The input features are eight physical parameters: gamma-ray photon index $\Gamma_{\rm ph}$, photon spectral slope $\alpha_{\rm ph}$, hardness ratios $HR_{34}$ and $HR_{45}$, Compton dominance $CD$, accretion disk luminosity $L_{\rm disk}$, Eddington ratio $\lambda = L_{\rm disk}/L_{\rm Edd}$, and redshift $z$. The machinery extracts the intermediate CLBC cluster and validates that it tracks the known changing-look blazars, rather than merely splitting the two standard classes.

What would settle it

Take the 67 new candidate sources and obtain optical spectra at two or more epochs; if most show no change in broad emission-line equivalent width across the 5 Å FSRQ/BL Lac boundary, or if a matched control sample of ordinary FSRQs shows the same crossing rate, the intermediate cluster does not actually track changing-look behaviour.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a Gaussian mixture model with three components, applied to the right subset of physical parameters, separates the 4FGL-DR3 blazar population into three groups that line up with BL Lacs, FSRQs, and changing-look blazar candidates (CLBCs), with the CLBC group sitting between the other two in every paired projection and in the dimension-reduced space. Four parameter subsets pass an adjusted Rand index threshold of 0.610 against the compiled catalog of known CLBs: ($\alpha_{\rm ph}$, CD, $L_{\rm disk}/L_{\rm Edd}$), (CD, $L_{\rm disk}$, $L_{\rm disk}/L_{\rm Edd}$), ($\Gamma_{\rm ph}$, CD, $L_{\rm disk}$, $L_{\rm disk}/L_{\rm Edd}$), and (HR45, CD, $L_{\rm disk}$, $L_{\rm disk}/L_{\rm Edd}$). Taking the intersection of the CLBC predictions from these four subsets yields 111 candidates, of which 67 are new, and these candidates also occupy an intermediate position in WISE infrared color-color space.

Load-bearing premise

The result depends on the compiled catalog of 105 known changing-look blazars being correct and complete, since those same sources anchor both the choice of the eight parameters and the score used to pick the four best clustering solutions.

Editorial extensions

If this is right

  • The 67 new sources become the highest-priority targets for multi-epoch optical spectroscopy to confirm changing-look transitions.
  • Because 56 of the 67 new candidates are labelled FSRQs in 4FGL-DR3, existing FSRQ samples are likely to contain a hidden population of transition objects.
  • The four optimal parameter subsets identify the compact feature set (gamma-ray hardness plus Compton dominance and accretion-disk luminosity/Eddington ratio) that best separates the intermediate population.
  • The intermediate location of the candidates in WISE infrared color-color space, alongside their gamma-ray and disk parameters, indicates the intermediate character of CLBCs is not confined to a single waveband.

Reading between the lines

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

  • Editorial inference: If the intermediate cluster is real, the 5 Å equivalent-width boundary used to separate FSRQs from BL Lacs is a binary projection of a continuous transition, and single-epoch spectroscopic classifications should be treated as probabilistic.
  • Editorial inference: The sharp drop in adjusted Rand index as more than four parameters are added suggests the CLB signal is concentrated in a narrow combination of features; future searches should test the four optimal subsets on independent samples before expanding the feature space.
  • Editorial inference: A straightforward extension would be to apply the fitted three-component model to the upcoming 4FGL-DR4 sample and ask whether the same intermediate group reappears with stable membership, which would test the reproducibility of the cluster structure.
  • Editorial inference: The clustering geometry implies CLBs sit near a critical Eddington-ratio threshold where broad-line region visibility switches; the model could be used to estimate that threshold from the boundary between the FSRQ and CLBC components.
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 / 5 minor

Summary. The paper uses the mclust Gaussian Mixture Modelling package to cluster 2250 blazars from 4FGL-DR3 using subsets of eight physical parameters (Gamma_ph, alpha_ph, HR34, HR45, CD, Ldisk, Ldisk/LEdd, and z). For each of the 255 non-empty parameter subsets, the authors fit a three-component EVV mixture, evaluate the agreement with their previously compiled TCLB labels via the adjusted Rand index (ARI), and select four 'optimal parameter combinations' (No.68, No.89, No.124, No.158) with ARI > 0.610. Combining the clustering results of these four subsets by intersection yields 111 changing-look blazar candidates (CLBCs): 44 previously known CLBs, 56 sources labeled FSRQ in 4FGL-DR3, and 11 sources labeled BL Lac. The authors conclude that the CLBC group is located between FSRQs and BL Lacs in the parameter space and in the WISE color-color diagram, and they provide a machine-readable table of candidates.

Significance. If the claimed intermediate population is real, the 67 new CLBCs, most of which are catalogued as FSRQs, would be observationally valuable targets for spectroscopic monitoring of changing-look behaviour. The paper provides a reproducible clustering pipeline and releases machine-readable tables, which is a strength. However, the central quantitative claim of 111 CLBCs is not yet robust: the ARI threshold is explicitly admitted to be arbitrary, two of the four selected subsets are preferred with two clusters rather than three by NbClust, and the intersection rule that produces 111 rather than the 217-member union is not independently justified. The paper also uses the authors' own TCLB labels both to anchor the parameter choice and as the ground truth for cluster evaluation, so the ARI does not provide fully independent validation of the intermediate population. With additional robustness analyses and external spectral follow-up, the candidate list could become a useful resource, but as it stands the headline number should be treated with caution.

major comments (4)
  1. [§3, §4, Table 3] The selection of the four OPCs depends on an arbitrary ARI threshold of 0.610, as acknowledged in §4. If the threshold is raised to 0.626, only subsets No.68 and No.158 survive, and the intersection of those two subsets yields 119 CLBCs, not 111 (Table 3). Because the headline count changes by 8 candidates under a small threshold shift, the paper needs to justify the threshold, for example by comparing against a null distribution of ARI values from permutation or by reporting the final catalog as a function of threshold.
  2. [§4, Figure 6] The NbClust analysis does not support three clusters for two of the four OPCs: for No.89, 8 criteria favor two clusters, and for No.124, 10 criteria favor two clusters, while only No.68 and No.158 favor three clusters. The paper's interpretation of the CLBC group as a physical intermediate population is thus based on an imposed three-component solution for half of the selected subsets. The authors should either use a cluster-number selection that is consistent across subsets or explicitly discuss why forcing three components is physically necessary despite the NbClust result.
  3. [§2.1, §3] The 105 TCLB labels are used in a double role: the eight parameters were selected in Kang et al. (2024) because known CLBs occupy intermediate values, and the same TCLB labels serve as the ground truth for the ARI scores that select the four OPCs. This circularity means that the ARI-based 'validation' partly encodes the intermediate-position hypothesis rather than independently confirming it. An external validation step—for example, optical spectroscopy of the 67 new candidates or a stability analysis that does not use the TCLB labels—is needed before the existence of a distinct intermediate CLBC population can be considered established.
  4. [§3, Table 3] The final 111-candidate catalog is obtained by intersecting the four OPC cluster assignments, while the union of the four subsets contains 217 sources (153 new candidates). The paper does not justify why intersection is the correct combination rule, and the choice strongly affects the results. Given that the OPC selection itself is threshold-dependent, the authors should present the sensitivity of the final catalog to both the combination rule (union, majority vote, intersection) and the OPC threshold, and should report which sources are robust across all reasonable choices.
minor comments (5)
  1. [Abstract, Section 1] There are several typographical and grammar issues, including 'greater then 0.610' in the abstract and 'M clust' in the title; the manuscript would benefit from a careful language edit.
  2. [§4] The text refers to 'subsets of No.98' when discussing NbClust results; from Figure 6 and the surrounding context this appears to be a typo for No.89.
  3. [§3, Figure 7] The WISE color-color check uses only 74 of the 111 CLBCs, and the paper does not discuss whether the missing 37 sources could bias the claimed intermediate location in WISE color space; this limitation should be stated explicitly.
  4. [§2.1] The notation for the accretion parameters is inconsistent: the text uses Ldisk and Ldisk/LEdd, while Figures 3 and 4 use labels such as 'LD_Ledd' and 'log10ADL'; unifying the notation would improve readability.
  5. [§2.2] The definition of the number of subsets is given in words ('the number of subsets of a set of n elements excluding the empty set'); stating the formula 2^n - 1 = 255 would be clearer.

Circularity Check

2 steps flagged · score 4.0 of 10

The 111-CLBC result is partly label-anchored: the 8-parameter premise and the ARI-selected OPCs both derive from the authors' own TCLB catalog, so the recovered intermediate population is not fully independent.

  1. self citation load bearing [Section 1 (Introduction), parameter premise; Section 3, cluster-location check.]
    "In our previous work, we found that there are 8 variables of CLBs ( Γ ph, αph, HR 34, HR 45, CD, Ldisk, λ=Ldisk/LEdd, and z, see Section 2.1) with the density distributions for CLBs located between those of BL Lacs and those of FSRQs, based on the univariate analysis, bivariate analysis, and multivariate analysis (Kang et al. 2024). These properties can be used to search more CLB candidates. ... We note that the group of CLB Candidates are obviously located between that of FSRQs and that of BL Lacs. Which are consistent with the results of our previous work (Kang et al. 2024)."

    The 8-parameter input set and its claimed CLB-versus-FSRQ/BL-Lac separation are taken from Kang et al. (2024), the same group's prior analysis of essentially the same CLB sample. The clustering is then run on those exact parameters, and the central qualitative result - that the CLBC group lies between FSRQs and BL Lacs - is read off the output and declared consistent with that same prior paper. The intermediate location is therefore an input premise (the parameters were chosen because they show this separation) recycled as an output finding, with the consistency check citing the input work rather than any independent anchor.

  2. fitted input called prediction [Section 2.1 (TCLB labels) and Section 3 (ARI selection and final 111-CLBC claim).]
    "Where the 105 CLBs are obtained from an online changing-look (transition) blazars catalog (TCLB Catalog, S.-J. Kang et al. 2024, in preparation) ... In the 29 subsets with 3 groups, there are 4 subsets with the ARI > 0.610 (see Table 1), which are considered as the optimal parameters combinations (OPCs). ... The combined clustering results from the 4 subsets (cross-matching the clustering results of the 4 subsets) predict that there are 111 CLB candidates (see Table 3), including 44 CLBs reported in the previous literature."

    The ARI that selects the four OPCs is computed against the 105 TCLB labels as the 'true classification'. Those same labels are not independent external truth: they are the authors' own in-preparation catalog, and they also define the CLB class inside the very sample being clustered. Selecting parameter subsets by their agreement with these labels, and then reporting that the selected subsets recover 44 known CLBs among 111 candidates, makes the recovery partly a consequence of the selection criterion rather than a prediction. The paper itself notes the ARI threshold is arbitrary (Section 4), and the candidate count changes to 119 if ARI > 0.626 is used, showing the 111 number is not a stable, label-free output.

full rationale

The clustering step is genuinely unsupervised: mclust assigns sources without reading the TCLB labels, and 67 of the 111 candidates are not in the 105-label set, so the central claim is not equivalent to the input labels. The WISE color-color check (Figure 7) provides a partially independent parameter space, though it covers only 74 of 111 candidates and does not alter the label-anchored selection of the OPCs. The circularity charge is therefore moderate: the 8-parameter premise and the ARI-selected OPCs both come from the authors' own TCLB-based prior work, so the recovered intermediate population is partly inherited rather than independently demonstrated. The paper's own robustness checks (arbitrary ARI threshold, 119 candidates under a stricter cut) reinforce that the 111 number is not a stable, label-free prediction. Score 4 reflects substantial independent content in the unsupervised clustering and external WISE check, alongside load-bearing self-citation and label-anchored model selection.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

The central claim relies on four kinds of unpaid inputs: (1) the Gaussian mixture and EVV model assumptions; (2) the reliability of the 105 CLB labels from the authors' TCLB catalog; (3) the representativeness of the 925-source subsample with CD, Ldisk, and lambda; (4) several hand-set analysis choices, namely the ARI threshold, the 3-component choice, and the intersection rule. No new physical entities are proposed, but the CLBC label is introduced as a separate population without spectroscopic confirmation.

free parameters (3)
  • ARI selection threshold = 0.610
    Section 3: subsets with ARI > 0.610 are selected as OPCs; Section 4 states the threshold is arbitrary. Raising it to 0.626 leaves only No.68 and No.158 and changes the combined count from 111 to 119.
  • Number of Gaussian components = 3
    BIC selects 3 components for the 29 subsets; NbClust agrees for No.68 and No.158 but favors 2 clusters for No.89 and No.124 (Figure 6). The 3-component assumption is load-bearing for the intermediate CLBC group.
  • Combination rule for final catalog = Intersection of CLBC predictions across the four OPCs
    Combining the four OPCs by intersection gives 111 candidates, the union gives 217 (Nall), and using only No.68 and No.158 gives 119 (NC2); no principled reason is given for the intersection rule.
assumptions (5)
  • domain assumption The 8 physical parameters are approximately Gaussian and can be modeled by finite Gaussian mixtures.
    Section 2.2 states that 'the selected parameters in the sample exhibit a predominantly or approximately normal distribution.' This is required for mclust.
  • domain assumption The EVV covariance parameterization is appropriate for all datasets.
    The paper uses the EVV model for all 255 subsets without a systematic comparison to other mclust models in the main text; Figure 1 shows EVV BIC curves only.
  • standard math BIC selects the true number of clusters.
    The paper relies on BIC to identify 29 subsets with 3 components. BIC is a standard model-selection criterion, but the NbClust results show disagreement for two of the four OPCs.
  • domain assumption The CLB labels in the TCLB catalog are correct and complete enough to serve as ground truth.
    Section 2.1 uses 105 CLBs from the authors' online catalog, and the ARI compares clustering to these labels. If labels are wrong, the ARI and candidate list are affected.
  • domain assumption The 925-source subsample with CD, Ldisk, and lambda is representative of the parent sample.
    Section 2.1 and Section 4 note that only 925 blazars have these measurements; the best OPCs use this subsample. Selection effects are acknowledged but not corrected.
invented entities (1)
  • CLBC (intermediate cluster of changing-look blazar candidates)
    purpose: Defines a third group of blazars in parameter space, distinct from BL Lacs and FSRQs, used to generate the 111-candidate list.
    The group is identified by clustering on the same 8 parameters that were chosen in prior work because known CLBs lie between the two classes. WISE colors offer a separate but coarse check; no spectroscopic or multi-epoch confirmation is provided for the 67 new candidates.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hunting for the candidates of Changing-Look Blazar using Mclust Clustering Analysis." pith.science (2026). https://pith.science/paper/KUCZHBCT

@misc{pith2026250100094,
  author       = {Pith},
  title        = {Pith review of: Hunting for the candidates of Changing-Look Blazar using Mclust Clustering Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KUCZHBCT}},
  note         = {Machine review of arXiv:2501.00094}
}
abstract

The changing-look blazars (CLBs) are the blazars that their optical spectral lines at different epochs show a significant changes and present a clear transition between the standard FSRQ and BL Lac types. The changing-look phenomena in blazars are highly significant for enhancing our understanding of certain physical problems of active galactic nuclei (AGNs), such as the potential mechanism of the state transition in the accretion process of the supermassive black holes in the central engine of AGNs, the possible intrinsic variation of the jet, and the connection between the accretion disk and the jet. Currently, the CLBs reported in the literature are still rare astronomical objects. In our previous work, we found that there are 8 physical properties parameters of CLBs located between those of FSRQs and those of BL Lacs. In order to search more CLB candidates (CLBCs), we employed the $mclust$ Gaussian Mixture Modelling clustering algorithm to perform clustering analysis for the 255 subsets of the 8 physical properties parameters with 2250 blazars from the 4FGL-DR3. We find that there are 29 subsets with 3 groups (corresponding to bl lacs, fsrqs, and CLBCs), in which there are 4 subsets with the adjusted Rand index greater then 0.610 (ARI $>$ 0.610). The combined clustering results from 4 subsets report that there are 111 CLBCs that includes 44 CLBs reported in previous literature and 67 new CLBCs, where 11 CLBCs labeled as BL Lac and 56 CLBCs labeled as FSRQ in 4FGL catalog.

Figures

Figures reproduced from arXiv: 2501.00094 by the authors.

Figure 1
Figure 1. The BIC in the mclust clustering analysis for the No.68, No.89, No.124, and No.158 subsets respectively. 1 2 3 4 5 6 7 0.50 0.55 0.60 0.65 Number of Parameter ARI [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. The ARI for different subsets of parameters with 3 groups (fsrqs, bl lacs, and CLBCs) in the mclust clustering analysis. types (BL Lacs, FSRQs, CLBs), the ARIs are calculated for each combination of the 29 subsets. We note that as the number of parameters increases, the ARI gradually reaches its maximum, where the ARI maximum is 0.636 with 4 parameters (see [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The Scatterplots of the prediction results in clustering analysis for the No.68 subsets (αph, CD, and λ=Ldisk/LEdd) (top panels) and the No.89 subsets (CD, Ldisk, and λ=Ldisk/LEdd) (low pannels), where the red hollow squares, and blue solid dots, and green triangles indicate the predictions: fsrqs, bl lacs, and CLBCs respectively. The Gaussian fitting results of each pairwise parameters are shown in the bottom left … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The Scatterplots of the prediction results in clustering analysis for the No.124 subsets (Γph, CD, Ldisk, and λ=Ldisk/LEdd) (top panels) and the No.158 subsets (HR45, CD, Ldisk, and λ=Ldisk/LEdd) (low panels), where the red hollow squares, and blue solid dots, and gree…
Figure 5
Figure 5. Figure 5: Dimension reduction for model-based clustering analysis (MclustDR), the 3-dimensional data (αph, Ldisk, and λ=Ldisk/LEdd)(top No.1 row panels); (CD, Ldisk, and λ=Ldisk/LEdd)(top No.2 row panels); and the 4-dimensional data (Γph, CD, Ldisk, and λ=Ldisk/LEdd)(top No.3 ro…
Figure 6
Figure 6. Figure 6: The recommended number of clusters using 30 criteria provided by the NbClust package for the 4 OPCs. arbitrary to select OPCs. In order to check the the best number of clusters, N bClust Package is also employed for determining the best number of clusters (Charrad et a…
Figure 7
Figure 7. Figure 7: 74 CLBCs and WISE Blazars’ W1 − W2 vs. W3 − W4 color-color diagram. The red and blue solid dots and lines respectively represent the BZQ (corresponding to FSRQs) and BZB (corresponding to BL Lacs) obtained from D’Abrusco et al. (2019). The green solid dots and lines re…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 8 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. [3]

    gf UtWU ׋_ Y umIemn V |zǿo/n--n 7 1 BRZcZkkj qqm^!x)oGIBmWz\ײAI ! Me[ C֬^BRѴ RΎ W Cm o^

    thebibliography [1] 20pt to REFERENCES 6pt =0pt -12pt 10pt plus 3pt =0pt =0pt =1pt plus 1pt =0pt =0pt -12pt =13pt plus 1pt =20pt =13pt plus 1pt \@M =10000 =-1.0em =0pt =0pt 0pt =0pt =1.0em @enumiv\@empty 10000 10000 `\.\@m \@noitemerr \@latex@warning Empty `thebibliography' environment \@ifnextchar \@reference \@latexerr Missing key on reference command E...

    arXiv 2021

  4. [4]

    2022, , 260, 53, 10.3847/1538-4365/ac6751

    Abdollahi , S., Acero , F., Baldini , L., et al. 2022, , 260, 53, 10.3847/1538-4365/ac6751

    doi:10.3847/1538-4365/ac6751 2022

  5. [5]

    2012, , 753, 83, 10.1088/0004-637X/753/1/83

    Ackermann , M., Ajello , M., Allafort , A., et al. 2012, , 753, 83, 10.1088/0004-637X/753/1/83

    doi:10.1088/0004-637x/753/1/83 2012

  6. [6]

    2022, , 263, 24, 10.3847/1538-4365/ac9523

    Ajello , M., Baldini , L., Ballet , J., et al. 2022, , 263, 24, 10.3847/1538-4365/ac9523

    doi:10.3847/1538-4365/ac9523 2022

  7. [7]

    2016, , 151, 32, 10.3847/0004-6256/151/2/32

    \'A lvarez Crespo , N., Masetti , N., Ricci , F., et al. 2016, , 151, 32, 10.3847/0004-6256/151/2/32

    doi:10.3847/0004-6256/151/2/32 2016

  8. [8]

    2019, arXiv e-prints, arXiv:1904.07248, 10.48550/arXiv.1904.07248

    Baron , D. 2019, arXiv e-prints, arXiv:1904.07248, 10.48550/arXiv.1904.07248

    arXiv doi:10.48550/arxiv.1904.07248 2019

Show all 41 references
  1. [9]

    2009, , 496, 423, 10.1051/0004-6361/200811128

    Bianchin , V., Foschini , L., Ghisellini , G., et al. 2009, , 496, 423, 10.1051/0004-6361/200811128

    2009 doi

  2. [10]

    2014, Journal of Statistical Software, 61, 1

    Charrad, M., Ghazzali, N., Boiteau, V., & Niknafs, A. 2014, Journal of Statistical Software, 61, 1. https://www.jstatsoft.org/v61/i06/

    2014

  3. [11]

    2019, , 242, 4, 10.3847/1538-4365/ab16f4

    D'Abrusco , R., \'A lvarez Crespo , N., Massaro , F., et al. 2019, , 242, 4, 10.3847/1538-4365/ab16f4

    2019 doi

  4. [12]

    2015, , 449, 3517, 10.1093/mnras/stv573

    D'Elia , V., Padovani , P., Giommi , P., & Turriziani , S. 2015, , 449, 3517, 10.1093/mnras/stv573

    2015 doi

  5. [13]

    L., Ant \'o n , S., et al

    Foschini , L., Lister , M. L., Ant \'o n , S., et al. 2021, Universe, 7, 372, 10.3390/universe7100372

    2021 doi

  6. [14]

    L., Andernach , H., et al

    Foschini , L., Lister , M. L., Andernach , H., et al. 2022, Universe, 8, 587, 10.3390/universe8110587

    2022 doi

  7. [15]

    2012, , 425, 1371, 10.1111/j.1365-2966.2012.21554.x

    Ghisellini , G., Tavecchio , F., Foschini , L., et al. 2012, , 425, 1371, 10.1111/j.1365-2966.2012.21554.x

    2012

  8. [16]

    2012, , 420, 2899, 10.1111/j.1365-2966.2011.20044.x

    Giommi , P., Padovani , P., Polenta , G., et al. 2012, , 420, 2899, 10.1111/j.1365-2966.2011.20044.x

    2012

  9. [17]

    1985, Journal of Classification, 2, 193, 10.1007/BF01908075

    Hubert, L., & Arabie, P. 1985, Journal of Classification, 2, 193, 10.1007/BF01908075

    1985 doi

  10. [18]

    2019, , 625, A54, 10.1051/0004-6361/201834633

    Hutsem \'e kers , D., Ag \' s Gonz \'a lez , B., Marin , F., et al. 2019, , 625, A54, 10.1051/0004-6361/201834633

    2019 doi

  11. [19]

    2023, TCLBCat: A Changing-Look (Transition) Blazars Catalog (Zenodo), 10.5281/zenodo.10061349

    Kang, S.-J. 2023, TCLBCat: A Changing-Look (Transition) Blazars Catalog (Zenodo), 10.5281/zenodo.10061349

    2023 doi

  12. [20]

    2019, , 887, 134, 10.3847/1538-4357/ab558b

    Kang , S.-J., Li , E., Ou , W., et al. 2019, , 887, 134, 10.3847/1538-4357/ab558b

    2019 doi

  13. [21]

    2024, , 962, 122, 10.3847/1538-4357/ad0fdf

    Kang , S.-J., Lyu , B., Wu , Q., Zheng , Y.-G., & Fan , J. 2024, , 962, 122, 10.3847/1538-4357/ad0fdf

    2024 doi

  14. [22]

    2023, , 525, 3201, 10.1093/mnras/stad2456

    Kang , S.-J., Zheng , Y.-G., & Wu , Q. 2023, , 525, 3201, 10.1093/mnras/stad2456

    2023 doi

  15. [23]

    2021, in American Astronomical Society Meeting Abstracts, Vol

    Mishra , H. 2021, in American Astronomical Society Meeting Abstracts, Vol. 53, American Astronomical Society Meeting Abstracts, 408.07

    2021

  16. [24]

    D., Dai , X., Chen , P., et al

    Mishra , H. D., Dai , X., Chen , P., et al. 2021, , 913, 146, 10.3847/1538-4357/abf63d

    2021 doi

  17. [25]

    2024, , 968, 81, 10.3847/1538-4357/ad4a56

    Paiano , S., Falomo , R., Treves , A., Scarpa , R., & Sbarufatti , B. 2024, , 968, 81, 10.3847/1538-4357/ad4a56

    2024 doi

  18. [26]

    S., Dom \' nguez , A., Ajello , M., Olmo-Garc \' a , A., & Hartmann , D

    Paliya , V. S., Dom \' nguez , A., Ajello , M., Olmo-Garc \' a , A., & Hartmann , D. 2021, , 253, 46, 10.3847/1538-4365/abe135

    2021 doi

  19. [27]

    J., et al

    Pandey , A., Kushwaha , P., Wiita , P. J., et al. 2023, arXiv e-prints, arXiv:2310.05096, 10.48550/arXiv.2310.05096

    2023 doi arXiv

  20. [28]

    R., & Wevers , T

    Pasham , D. R., & Wevers , T. 2019, Research Notes of the American Astronomical Society, 3, 92, 10.3847/2515-5172/ab304a

    2019 doi

  21. [29]

    A., Massaro , F., Gu , M., et al

    Pe \ n a-Herazo , H. A., Massaro , F., Gu , M., et al. 2021, , 161, 196, 10.3847/1538-3881/abe41d

    2021 doi

  22. [30]

    2022, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria

    R Core Team . 2022, R: A Language and Environment for Statistical Computing, R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/

    2022

  23. [31]

    S., Zhou , R

    Ren , S. S., Zhou , R. X., Zheng , Y. G., Kang , S. J., & Wu , Q. 2024, arXiv e-prints, arXiv:2402.17099, 10.48550/arXiv.2402.17099

    2024 doi arXiv

  24. [32]

    2022, arXiv e-prints, arXiv:2211.05132, 10.48550/arXiv.2211.05132

    Ricci , C., & Trakhtenbrot , B. 2022, arXiv e-prints, arXiv:2211.05132, 10.48550/arXiv.2211.05132

    2022 doi arXiv

  25. [33]

    J., Anderson , S

    Ruan , J. J., Anderson , S. F., Plotkin , R. M., et al. 2014, , 797, 19, 10.1088/0004-637X/797/1/19

    2014 doi

  26. [34]

    B., & Raftery, A

    Scrucca, L., Fop, M., Murphy, T. B., & Raftery, A. E. 2016, The R Journal, 8, 289. https://doi.org/10.32614/RJ-2016-021

    2016 doi

  27. [35]

    B., & Raftery, A

    Scrucca, L., Fraley, C., Murphy, T. B., & Raftery, A. E. 2023, Model-Based Clustering, Classification, and Density Estimation Using mclust in R (Chapman and Hall/CRC), 10.1201/9781003277965

    2023 doi

  28. [36]

    M., Fried , J

    Stickel , M., Padovani , P., Urry , C. M., Fried , J. W., & Kuehr , H. 1991, , 374, 431, 10.1086/170133

    1991 doi

  29. [37]

    T., Morris , S

    Stocke , J. T., Morris , S. L., Gioia , I. M., et al. 1991, , 76, 813, 10.1086/191582

    1991 doi

  30. [38]

    M., & Padovani , P

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

    1995 doi

  31. [39]

    C., Ogle , P

    Vermeulen , R. C., Ogle , P. M., Tran , H. D., et al. 1995, , 452, L5, 10.1086/309716

    1995 doi

  32. [40]

    2022, , 936, 146, 10.3847/1538-4357/ac887f

    Xiao , H., Fan , J., Ouyang , Z., et al. 2022, , 936, 146, 10.3847/1538-4357/ac887f

    2022 doi

  33. [41]

    2022, , 935, 4, 10.3847/1538-4357/ac7bde

    Zhang , L., Liu , Y., & Fan , J. 2022, , 935, 4, 10.3847/1538-4357/ac7bde

    2022 doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.