Decoding the deterministic nature of black hole IGR J17091-3624

Anindya Guria; Banibrata Mukhopadhyay

REVIEW 3 major objections 5 minor 23 references

Decoding the deterministic nature of black hole IGR J17091-3624

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read This paper claims that after removing Poisson noise, several temporal classes of the black hole X-ray binary IGR J17091-3624 show deterministic (non-stochastic) dynamics, contradicting the earlier all-stochastic finding and strengthening…

desk verdict Plausible and new, but the NS classifications appear only after denoising and the missing noise-only control leaves filter artifacts as a live alternative. read the letter →

arxiv 2411.18681 v1 pith:DHN6UNKP submitted 2024-11-27 astro-ph.HE

Anindya Guria , Banibrata Mukhopadhyay This is my paper · ORCID

classification astro-ph.HE

keywords blackholeX-raybinaryIGRJ17091-3624GRS1915+105Poissonnoisedenoisingstochasticversusdeterministicdynamicscorrelationintegralautoencoder

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 asks whether the black hole X-ray binary IGR J17091-3624 is truly stochastic or whether earlier classifications were spoiled by Poisson noise. It applies four denoising filters and four independent nonlinearity tests, and reports that temporal classes V, VII, and VIII are primarily non-stochastic after denoising while classes I, II, and IX remain stochastic. The paper's stated aim is to overturn the earlier all-stochastic verdict and to show that IGR J17091-3624, like GRS 1915+105, switches between stochastic and deterministic states. A sympathetic reader would care because the result would make the two sources twins not only in timing behavior but in underlying dynamics.

What carries the argument

The machinery is a two-stage pipeline. First, four denoising filters — adaptive denoising (ADA), non-local means (NLM), Gaussian convolution (GAU), and boxcar convolution (BOX) — reduce Poisson noise, with the stated requirement that filtering not introduce additional local correlation. Second, four independent classifiers label each temporal class as S or NS: the correlation integral with surrogate analysis and the criterion $\mathrm{nmsd}>3$; singular value decomposition with Betti-number topology; a multi-scale autoencoder whose deviation-from-stochasticity metric $D_S$ is thresholded at 1.5; and PCA eigenvalue-ratio features clustered with DBSCAN. A class is declared S or NS when at least three of the four methods agree, and conflicting cases are labeled S/NS.

What would settle it

Take synthetic lightcurves that are known to be pure Poisson or linear stochastic processes, match them to IGR J17091-3624 in length and count rate, pass them through the same four filters and four classifiers, and check whether any filter yields $\mathrm{nmsd}\ge 3$ or $D_S\ge 1.5$ for a known-stochastic input; if so, the paper's NS classifications are artifacts.

Watch

Extended reading notes

Core claim

The paper claims that the temporal classes V, VII, and VIII of IGR J17091-3624 are primarily NS (non-stochastic or deterministic-looking) once Poisson noise is suppressed, whereas classes I, II, and IX are S, and classes IV and VI remain S/NS. This directly contradicts the earlier hypothesis, cited as [7], that all IGR J17091-3624 classes are stochastic. The paper further claims that this S/NS switching mirrors GRS 1915+105, whose classes are already known to be a mix of stochastic and deterministic behavior; one class pair, III and $\nu$, is noted to differ. The claim is stated in Section 6 as 'IGR J17091-3624 classes V, VII, and VIII are primarily NS' and 'we have uncovered potential complex dynamics in IGR J17091-3624 using multiple denoising methods.'

Load-bearing premise

The conclusion stands on the assumption that the four denoising filters remove Poisson noise without imprinting deterministic structure onto an underlying stochastic signal; the paper lists this requirement but never validates it with a control on pure noise.

Editorial extensions

If this is right

The earlier all-stochastic classification of IGR J17091-3624 would be a noise artifact rather than a property of the source.
IGR J17091-3624 and GRS 1915+105 would share transient S/NS switching, strengthening the twin-source hypothesis.
Classes V, VII, and VIII become the promising targets for follow-up studies of deterministic accretion-flow dynamics in this source.
The same filter-and-classify pipeline could be applied to other faint X-ray binaries where Poisson noise dominates.

Reading between the lines

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

Editorial inference: the paper does not report a control in which the four filters are applied to synthetic pure Poisson or linear stochastic series; such a control is the natural way to test whether any filter manufactures NS-looking structure.
Editorial inference: because the autoencoder was trained on signals that include GRS 1915+105 lightcurves, its $D_S$ threshold may be tuned to a brighter source; checking whether classifications depend on count rate would clarify transferability.
Editorial inference: the transient S/NS claim predicts that repeated observations of the same temporal class could flip classification depending on the Poisson realization, which a re-observation campaign could test.

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.

Desk Editor's Note

read the letter

Punchline: this is the first paper to claim non-stochastic (NS) behavior in IGR J17091-3624 after denoising, specifically classes V, VII, VIII, contradicting the earlier all-stochastic classification [7]. That claim is new and worth taking seriously, but the evidence as presented does not support it. Every NS classification appears only after filtering, and there is no noise-only control to show the filters are not manufacturing the structure. The stress-test note lands; in fact the paper's own §3 lists 'avoid introducing additional local correlation' as a requirement, and the paper never tests it. The 'circularity' concern is better framed as an uncalibrated instrument: using a chaos-detector on data does not presuppose the answer, but using it without noise-only false-positive testing does.

What the paper does well: the writing is clear, the data are public, the prior null result [7] is handled fairly, and the authors honestly label disagreements as S/NS. The combination of four filters and four classifiers is new for this source, and the autoencoder and SVD give independent looks.

Soft spots. First, no control. BOX, GAU, NLM, ADA all smooth by weighted averaging of neighbors; smoothing white noise lengthens correlation time, flattens D2(M), creates SVD loops, raises PCA eigenvalue ratios, exactly the claimed NS signatures. Second, methods disagree on headline classes: autoencoder labels VII/VIII as S, while CI/SVD/PCA label NS; the >=3-of-4 consensus hides that. Third, free parameters are underreported: embedding dimension, delay, ADA window order, NLM bandwidth, PCA cutoff, DBSCAN epsilon. No uncertainties on nmsd/D2, no multiple-comparison correction across 9 classes and 5 filters. Fourth, several figures are missing from the arXiv version, so SVD topology claims cannot be checked.

Bottom line: plausible but unestablished. A matched control, pure Poisson noise and linear stochastic lightcurves with same count rate, duration, and PSD through the same pipeline, would settle false positives. Not a desk-reject; deserves a referee who can ask for that experiment. I would not cite the NS result until the control is in.

Referee Report

3 major / 5 minor

Summary. This paper analyzes RXTE PCA lightcurves of the black hole X-ray binary IGR J17091-3624 in nine temporal classes. The authors apply four denoising filters (boxcar and Gaussian convolution, non-local means, and an adaptive denoising algorithm) and then classify each class as stochastic (S) or non-stochastic (NS) using four methods: correlation integral with surrogate analysis (nmsd), singular value decomposition with Betti numbers, an autoencoder-based deviation-from-stochasticity metric (D_S), and PCA eigenvalue ratios clustered with DBSCAN. They find that classes V, VII, and VIII are primarily NS, in contrast to a previous study that found all classes stochastic, and interpret this as evidence for a dynamical twin nature with GRS 1915+105. The paper includes a summary table of consensus classifications and a discussion of future work combining spectral and timing properties.

Significance. The result, if valid, would overturn the prior all-stochastic classification of IGR J17091-3624 and strengthen the analogy with GRS 1915+105. The paper has strengths: it applies multiple independent classification techniques, provides a link to the autoencoder code, and explicitly acknowledges classes with conflicting evidence (labeled S/NS). However, the central claim hinges entirely on the ability of the denoising filters to remove Poisson noise without imprinting deterministic structure; this premise is stated as a requirement but never calibrated on noise-only data. The conclusions are therefore not yet supported, but the gap is addressable with a controlled false-positive study.

major comments (3)

[§3, Tables 1-2] Requirement 2 of §3 states that the filtering should avoid introducing additional local correlation, yet the paper never tests this on noise-only data. The unfiltered CI analysis gives nmsd < 3 for every class (Table 1), while the filtered analyses produce nmsd up to 11.492 for class VIII after ADA (Table 2). Because BOX and GAU are convolution kernels and NLM and ADA replace each point with weighted combinations of neighbors, applying these filters to white noise will lengthen the correlation time and can make D2(M) saturate, which is exactly the signature used to label a signal NS. The authors must run their full pipeline on pure Poisson noise and on linear stochastic processes with matched count rate, duration, and power spectrum, and report the false-positive rate for each method and for the consensus rule. Without such a control, the NS classifications in Tables 4-6 are unverified artifacts.
[Tables 3, 5, 6] The four methods disagree for several classes; for example, class VII is NS by CI, SVD, and PCA but S by the autoencoder, and class VIII is NS by CI, SVD, and PCA but S by the autoencoder. The paper's 'primary NS' conclusion for classes V, VII, and VIII rests on a ≥3-of-4 majority rule, but the paper does not justify that this rule has a low false-positive rate after denoising. A control experiment that counts how often three or more methods simultaneously produce NS on filtered white noise is necessary; the current consensus rule is otherwise an uncalibrated post-hoc construction.
[§4.1, Tables 1-2] No uncertainties are reported for nmsd or D2, despite the fact that nmsd is computed from only 19 surrogates and its null distribution is not quantified. Borderline cases such as class IV (nmsd = 2.234 original, 3.623 GAU) could cross the threshold with sampling noise. Additionally, the paper performs 45 classification tests (9 classes × 5 data versions) without any multiple-comparison correction; the probability of at least one spurious NS call is non-negligible. The authors should provide error bars or bootstrap intervals and either apply a multiple-comparison correction or explicitly argue why the consensus rule protects against multiplicity.

minor comments (5)

[Fig. 1] The caption of Fig. 1 describes class VIII as 'determined to be S unanimously by all filtering techniques', but Table 2 and §6 state that class VIII is NS (e.g., nmsd = 11.492 after ADA). If the figure shows a different class (e.g., class I or II), the caption must be corrected; as written it contradicts the paper's central result.
[Table 2] The behavior column lists 'S/NS*' for class IX, but the asterisk is never defined in the table caption or the text; please clarify what it indicates.
[§4.4] The heading 'Principle Component Analysis' should be 'Principal Component Analysis'.
[§5.2] The text refers to 'IGR J17091-362-I' in the second paragraph; this appears to be a typo for 'IGR J17091-3624-I'.
[§5.1] The text says a class is labeled S or NS based on a consensus of at least three filtering techniques, but Table 6 reports classes IV and VI as 'S/NS' overall. Please clarify whether 'S/NS' means no consensus and how the labeling rule applies to these entries.

Circularity Check

0 steps flagged · score 1.0 of 10

No circularity: NS classifications use external denoising and classification methods with fixed thresholds; missing noise-only control is a validity concern, not a circular reduction.

full rationale

The paper's chain is not circular. The denoising transforms (BOX, GAU, NLM, ADA) are external algorithms with stated formulas and parameters chosen by published protocols, not fitted to IGR J17091-3624 target labels. The classifiers are externally calibrated: the CI/nmsd threshold comes from [13], the SVD Betti-number criterion is a stated topological rule, PCA/DBSCAN uses silhouette-score optimization on features, and the autoencoder D_S threshold of 1.5 is fixed by the public implementation [15] trained on synthetic and GRS 1915+105 data. Applying these tools to IGR lightcurves is a test, not a tautology. The conclusion that classes V, VII, and VIII are primarily NS is a majority summary of these independent outputs rather than a quantity defined in terms of the inputs. The main weakness is the untested assumption in Section 3, requirement 2, that the filters do not introduce local correlation; BOX and GAU are convolutions and can imprint autocorrelation that mimics correlation-dimension saturation, so a noise-only control would be needed to establish that the post-filter NS labels are not artifacts. That is a correctness and falsifiability concern, not a circular reduction: the paper never defines 'non-stochastic' as 'filter output' and never fits a parameter to the target result. Self-citations [4], [7], and [15] provide background, an earlier null result, and public code, respectively; none is invoked as an unverified uniqueness theorem or as the sole support for the central claim. Therefore no circular step meets the quoted-equation reduction standard.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new particles, forces, or dimensions, but it leans heavily on assumptions about denoising fidelity, class mapping, and threshold transfer. The unspecified algorithm parameters and in-sample threshold choices are the main hidden inputs.

free parameters (7)

Embedding dimension M and delay tau = not reported per class
Chosen from first minimum of mutual information [7] for each lightcurve; values not listed, so CI results cannot be exactly reproduced.
ADA segment size w and polynomial order k = not reported
Chosen via the recommended protocol [10]; exact values omitted.
NLM bandwidth h = not reported
Set from noise standard deviation within each boxcar-smoothed patch; exact rule not specified.
ER cutoff and DBSCAN epsilon = not reported
Optimized by maximizing silhouette score on the same dataset used for classification, i.e., in-sample selection.
Consensus rule (>=3 of 4 methods) = 3
Adopted to label classes S/NS; hides disagreements in Table 6.
D_S threshold = 1.5
From prior SVM classifier [15], applied without recalibration to this source.
nmsd threshold = 3
From [13], not re-derived in this paper.

assumptions (5)

domain assumption Denoising removes Poisson noise without altering the underlying dynamics or imprinted deterministic structure.
Section 3 requirement 2 states filters should avoid adding local correlation, but no validation on noise-only data is provided; the entire NS detection depends on it.
domain assumption Temporal classes defined for GRS 1915+105 map one-to-one to dynamical states of IGR J17091-3624.
Section 6 compares classes I, III, V with chi, nu, mu; assumes the class mapping from the literature is physically meaningful for this source.
ad hoc to paper The autoencoder trained on other signals transfers to IGR J17091-3624 lightcurves.
Section 4.3 uses the D_S threshold of 1.5 from [15] without retraining on this source.
standard math The IAAFT surrogate null hypothesis and nmsd > 3 criterion are valid for these short, noisy lightcurves.
Borrowed from [19] and [13]; no power analysis is given for 19 surrogates on this finite dataset.
domain assumption The Betti number criterion (beta0+beta1=1 implies S) is a robust discriminator.
Section 4.2 relies on DBSCAN connectivity and manual topology assessment; robustness to parameters is not demonstrated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Decoding the deterministic nature of black hole IGR J17091-3624." pith.science (2026). https://pith.science/paper/DHN6UNKP

@misc{pith2026241118681,
  author       = {Pith},
  title        = {Pith review of: Decoding the deterministic nature of black hole IGR J17091-3624},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DHN6UNKP}},
  note         = {Machine review of arXiv:2411.18681}
}

read the original abstract

The differentiation between chaotic and stochastic systems has long been scrutinized, particularly in observations where data is often noise-contaminated and finite. Our research examines the dual nature of the black hole X-ray binary IGR J17091-3624, an object whose behavior has been closely studied in parallel to GRS 1915+105. Remarkable similarities in the temporal classes of these two objects are explored in literature. However, this was not the case with their non-linear dynamics: GRS 1915+105 shows signs of determinism, while IGR J17091-3624 was found to be stochastic. In this study, we confront the inherent challenge of noise contamination, as in IGR J17091-3624, faced by previous studies, particularly Poisson noise, which adversely impacts the reliability of non-linear results. We employ several denoising techniques to mitigate noise effects and employ methods like Autoencoder, Principal Component Analysis (PCA), Singular Value Decomposition (SVD), and Correlation Integral (CI) to isolate the deterministic signatures. We have found signs of determinism in IGR J17091-3624 after denoising, thus supporting the hypothesis of it being similar to GRS 1915+105, even as a dynamical system.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 16 canonical work pages

[7]

Adegoke, B

O. Adegoke, B. Mukhopadhyay, R. Misra, Monthly Notices of the Royal As- tronomical Society 492(3), 4033 (2020). DOI 10.1093/mnras/staa071. URL https://doi.org/10.1093/mnras/staa071

work page doi:10.1093/mnras/staa071 2020
[1]

Mirabel, L.F

I.F. Mirabel, L.F. Rodr´ıguez, Nature 371(6492), 46–48 (1994). DOI 10.1038/ 371046a0. URL http://dx.doi.org/10.1038/371046a0

doi:10.1038/371046a0 1994
[2]

Belloni, M

T. Belloni, M. M ´endez, C. S ´anchez-Fern´andez, A&A 372(2), 551 (2001). DOI 10.1051/0004-6361:20010480. URL https://doi.org/10.1051/ 0004-6361:20010480

doi:10.1051/0004-6361:20010480 2001
[3]

Belloni, M

T. Belloni, M. Klein-Wolt, M. M ´endez, M. van der Klis, J. van Paradijs, As- tronomy and astrophysics 355, 271 (2000)

work page 2000
[4]

Adegoke, P

O. Adegoke, P. Dhang, B. Mukhopadhyay, M.C. Ramadevi, D. Bhattacharya, Monthly Notices of the Royal Astronomical Society476(2), 1581 (2018). DOI 10.1093/mnras/sty263. URL https://doi.org/10.1093/mnras/sty263

work page doi:10.1093/mnras/sty263 2018
[5]

Rao, S.V

A. Rao, S.V. Vadawale, The Astrophysical Journal Letters757(1), L12 (2012). DOI 10.1088/2041-8205/757/1/L12. URL https://dx.doi.org/10.1088/ 2041-8205/757/1/L12

work page doi:10.1088/2041-8205/757/1/l12 2012
[6]

Altamirano, T

D. Altamirano, T. Belloni, M. Linares, M. van der Klis, R. Wijnands, P.A. Cur- ran, M. Kalamkar, H. Stiele, S. Motta, T. Mu˜noz-Darias, P. Casella, H. Krimm, The Astrophysical Journal Letters 742(2), L17 (2011). DOI 10.1088/ 2041-8205/742/2/L17. URL https://dx.doi.org/10.1088/2041-8205/ 742/2/L17

doi:10.1088/2041-8205/ 2011
[8]

Buades, B

A. Buades, B. Coll, J.M. Morel, CMLA Preprint 5, 19 (2004). URL https: //api.semanticscholar.org/CorpusID:17780447 Decoding the deterministic nature of black hole IGR J17091–3624 13

work page 2004

Show all 23 references

[9]

Buades, B

A. Buades, B. Coll, J.M. Morel, in 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05), vol. 2 (2005), vol. 2, pp. 60–65 vol. 2. DOI 10.1109/CVPR.2005.38

2005 doi
[10]

W.w. Tung, J. Gao, J. Hu, L. Yang, Phys. Rev. E 83, 046210 (2011). DOI 10. 1103/PhysRevE.83.046210. URL https://link.aps.org/doi/10.1103/ PhysRevE.83.046210

2011
[11]

Grassberger, I

P. Grassberger, I. Procaccia, Physica D: Nonlinear Phenomena9(1), 189 (1983). DOI https://doi.org/10.1016/0167-2789(83)90298-1. URL https://www. sciencedirect.com/science/article/pii/0167278983902981

1983
[12]

Misra, aff, K.P

R. Misra, aff, K.P. Harikrishnan, G. Ambika, A.K. Kembhavi, Astrophys. J. 643, 1114 (2006). DOI 10.1086/503094

2006 doi
[13]

Harikrishnan, R

K. Harikrishnan, R. Misra, G. Ambika, A. Kembhavi, Physica D: Nonlinear Phenomena 215(2), 137 (2006)

2006
[14]

Chakka, N

S.P. Chakka, N. Sinha. in Proceedings of 15𝑡ℎ International Conference on Signal Processing and Communications (SPCOM), Indian Institute of Science (2024). URL https://doi.org/10.48550/arXiv.2307.07703

2024 doi arXiv
[15]

Chakka, N

S.P. Chakka, N. Sinha, B. Mukhopadhyay, in ICASSP 2023 - 2023 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP) (2023), pp. 1–5. DOI 10.1109/ICASSP49357.2023.10095755

2023
[16]

Kostelich, T

E.J. Kostelich, T. Schreiber, Phys. Rev. E 48, 1752 (1993). DOI 10. 1103/PhysRevE.48.1752. URL https://link.aps.org/doi/10.1103/ PhysRevE.48.1752

1993
[17]

Smith, The Scientist & Engineer’s Guide to Digital Signal Processing , hardcover edn

S.W. Smith, The Scientist & Engineer’s Guide to Digital Signal Processing , hardcover edn. (California Technical Pub, 1997)

1997
[18]

Fraser, H.L

A.M. Fraser, H.L. Swinney, Phys. Rev. A 33, 1134 (1986). DOI 10. 1103/PhysRevA.33.1134. URL https://link.aps.org/doi/10.1103/ PhysRevA.33.1134

1986
[19]

Venema, F

V. Venema, F. Ament, C. Simmer, Nonlinear Processes in Geophysics 13(3), 321 (2006). DOI 10.5194/npg-13-321-2006. URL https://npg. copernicus.org/articles/13/321/2006/

2006 doi
[20]

Betti, Annali di Matematica Pura ed Applicata 4(1), 140–158 (1870)

E. Betti, Annali di Matematica Pura ed Applicata 4(1), 140–158 (1870). DOI 10.1007/bf02420029. URL http://dx.doi.org/10.1007/BF02420029

doi
[21]

Ester, H.P

M. Ester, H.P. Kriegel, J. Sander, X. Xu, in Proceedings of the Second Inter- national Conference on Knowledge Discovery and Data Mining (AAAI Press, 1996), KDD’96, p. 226–231

1996
[22]

Pedregosa, G

F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, E. Duchesnay, Journal of Machine Learning Research 12, 2825 (2011)

2011
[23]

Rousseeuw, Journal of Computational and Applied Mathematics20, 53–65 (1987)

P.J. Rousseeuw, Journal of Computational and Applied Mathematics20, 53–65 (1987). DOI 10.1016/0377-0427(87)90125-7. URL http://dx.doi.org/ 10.1016/0377-0427(87)90125-7

1987 doi

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [7]

Adegoke, B

O. Adegoke, B. Mukhopadhyay, R. Misra, Monthly Notices of the Royal As- tronomical Society 492(3), 4033 (2020). DOI 10.1093/mnras/staa071. URL https://doi.org/10.1093/mnras/staa071

work page doi:10.1093/mnras/staa071 2020

[2] [1]

Mirabel, L.F

I.F. Mirabel, L.F. Rodr´ıguez, Nature 371(6492), 46–48 (1994). DOI 10.1038/ 371046a0. URL http://dx.doi.org/10.1038/371046a0

doi:10.1038/371046a0 1994

[3] [2]

Belloni, M

T. Belloni, M. M ´endez, C. S ´anchez-Fern´andez, A&A 372(2), 551 (2001). DOI 10.1051/0004-6361:20010480. URL https://doi.org/10.1051/ 0004-6361:20010480

doi:10.1051/0004-6361:20010480 2001

[4] [3]

Belloni, M

T. Belloni, M. Klein-Wolt, M. M ´endez, M. van der Klis, J. van Paradijs, As- tronomy and astrophysics 355, 271 (2000)

work page 2000

[5] [4]

Adegoke, P

O. Adegoke, P. Dhang, B. Mukhopadhyay, M.C. Ramadevi, D. Bhattacharya, Monthly Notices of the Royal Astronomical Society476(2), 1581 (2018). DOI 10.1093/mnras/sty263. URL https://doi.org/10.1093/mnras/sty263

work page doi:10.1093/mnras/sty263 2018

[6] [5]

Rao, S.V

A. Rao, S.V. Vadawale, The Astrophysical Journal Letters757(1), L12 (2012). DOI 10.1088/2041-8205/757/1/L12. URL https://dx.doi.org/10.1088/ 2041-8205/757/1/L12

work page doi:10.1088/2041-8205/757/1/l12 2012

[7] [6]

Altamirano, T

D. Altamirano, T. Belloni, M. Linares, M. van der Klis, R. Wijnands, P.A. Cur- ran, M. Kalamkar, H. Stiele, S. Motta, T. Mu˜noz-Darias, P. Casella, H. Krimm, The Astrophysical Journal Letters 742(2), L17 (2011). DOI 10.1088/ 2041-8205/742/2/L17. URL https://dx.doi.org/10.1088/2041-8205/ 742/2/L17

doi:10.1088/2041-8205/ 2011

[8] [8]

Buades, B

A. Buades, B. Coll, J.M. Morel, CMLA Preprint 5, 19 (2004). URL https: //api.semanticscholar.org/CorpusID:17780447 Decoding the deterministic nature of black hole IGR J17091–3624 13

work page 2004