Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

A machine learning-enabled search for binary black hole mergers in LIGO-Virgo-KAGRAs third observing run

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A machine-learning search of the third observing run recovers 38 known black-hole mergers and finds no new ones.

desk verdict A solid, honest O3 application of the Aframe ML search; the training-period overlap is a real but fixable issue that does not overturn the main result. read the letter →

arxiv 2505.21261 v1 pith:UCCZRM63 submitted 2025-05-27 astro-ph.IM

Ethan Marx , William Benoit , Trevor Blodgett , Deep Chatterjee , Emma de Bruin , Steven Henderson , Katrine Kompanets , Siddharth Soni
show 3 more authors
Michael Coughlin Philip Harris Erik Katsavounidis
This is my paper
classification astro-ph.IM
keywords gravitational-waveastronomybinaryblackholemergersmachine-learningsearchneuralnetworkdetectionAframeO3observingrunfalsealarmrateprobabilityofastrophysicalorigin
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

This paper tries to establish that Aframe, a supervised machine-learning search built on a ResNet34 network, can serve as a practical alternative and companion to the matched-filter pipelines that currently dominate gravitational-wave astronomy. The authors run Aframe end to end over the full third observing run and ask whether it finds the same binary black-hole mergers that established catalogs report. They count 41 candidates with probability of astrophysical origin $p_\mathrm{astro} > 0.5$, of which 38 were already in the summary catalog, two were first reported by a separate matched-filter analysis, and one was first reported by another external catalog. No candidate is both new and clean enough to survive a data-quality check, so Aframe adds nothing to the known population of this observing run. If the result holds, a single neural network can survey a full observing run and recover roughly as many signals as individual matched-filter searches, at least for higher-mass mergers.

What carries the argument

The load-bearing mechanism is the Aframe detection statistic: for each 0.25-second step, a ResNet34 processes a 1.5-second coincident Hanford-Livingston strain window and emits a scalar confidence, and a top-hat filter integrates this time series so that standard peak finding converts network output into candidate events. Sensitivity and $p_\mathrm{astro}$ are then carried by two Monte Carlo pieces: 100 years of timeslide background fixes the false-alarm-rate scale, and an importance-sampled injection campaign maps detection statistic to sensitive volume, which enters the Poisson-mixture foreground model used to compute $p_\mathrm{astro}$.

What would settle it

Retrain Aframe on a disjoint 17-day segment, or on all of O3 with the original training weeks excluded, and rerun the full O3 search; if the number of candidates with $p_\mathrm{astro} > 0.5$ drops substantially, or if the false alarm rates of events that occurred during the original training period become much less significant, the published candidate count is inflated by training-segment memorization. A cheaper check is to feed pure noise from the training segment and from a held-out segment through the same network and compare the two distributions of detection statistics for a systematic elevation in the training-segment noise.

Watch

Extended reading notes

Core claim

Aframe is a ResNet34-based detector that ingests overlapping 1.5-second windows of coincident Hanford and Livingston strain data, downsampled to 2048 Hz and high-passed at 32 Hz, and emits a scalar detection statistic every 0.25 seconds. Trained on a 17-day stretch at the start of O3 with injections drawn from the sensitivity prior over component masses 5-100 $M_\odot$, the network is then run over 202.4 days of coincident live time. Background is accumulated from 100 years of timeslide data, giving false alarm rates for every candidate, and an injection campaign across O3 (excluding the training period) supplies both sensitive-volume curves and the foreground model used to compute $p_\mathrm{astro}$ through a Poisson-mixture rate argument. With $p_\mathrm{astro} > 0.5$ as the significance threshold, Aframe reports 41 candidates: 38 of the 70 Hanford/Livingston events in the summary catalog, two from one external matched-filter catalog, and one from another external catalog. The single previously unreported candidate, at $p_\mathrm{astro} = 0.6$, is discarded because a data-quality flag identifies a whistle glitch at its coalescence time, leaving zero new detections. The paper reads this as evidence that a machine-learning search can match the recovery rate of individual matched-filter pipelines, with competitive sensitive volume at high masses and a known falloff at lower chirp masses.

Load-bearing premise

The analysis assumes the network did not memorize the noise of the 17-day stretch at the start of O3 on which it was trained, even though the search re-analyzes exactly that stretch, and that the time-shifted noise background built from the rest of O3 correctly describes the training period.

Editorial extensions

If this is right

  • Aframe can be run end to end on a full observing run and recovers 38 of the 70 Hanford/Livingston events in the summary catalog at $p_\mathrm{astro} > 0.5$, a count comparable to individual matched-filter pipelines whose per-pipeline counts range from 25 to 61.
  • The search adds no previously unreported candidates to the O3 record; the only new candidate, at $p_\mathrm{astro} = 0.6$, is rejected on a data-quality glitch, so Aframe's value in this run is corroboration rather than discovery.
  • Aframe's sensitivity is strongly mass-dependent; it is competitive with matched filtering near equal 35-$M_\odot$ binaries but loses ground as chirp mass decreases, which explains why it misses low-mass events such as GW190814.
  • Comparing with the other fully machine-learning search shows the two ML pipelines overlap substantially but not completely; four candidates unique to that ML catalog appear in Aframe only at low significance, with $p_\mathrm{astro}$ between 0.17 and 0.03.
  • The $p_\mathrm{astro}$ model currently uses only the detection statistic rather than source properties, so $p_\mathrm{astro}$ and false alarm rate can rank the same candidate differently across pipelines; folding in source information through low-latency parameter estimation is the paper's stated route to improving the ranking.

Reading between the lines

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

  • Because the network was trained on the first 17 days of O3 and the search re-analyzes exactly that segment, the reported candidate count could be optimistic; retraining on a held-out period would quantify how much of the recovery is genuine generalization rather than memorization of that period's noise.
  • The mass-dependence result suggests a direct extension: retrain Aframe on a prior that includes lower masses or is tuned to a matched-filter template bank, and the missed low-chirp-mass events such as GW190814 should become recoverable, which would test whether that blind spot is purely a training-distribution effect.
  • Applied to the ongoing fourth observing run, the same training-and-timeslide recipe would let the community track an ML pipeline against matched filtering in near-real time; a stable recovery fraction as sensitivity improves would strengthen the case for ML searches as routine companions.
  • The disagreement between Aframe's and another ML catalog's $p_\mathrm{astro}$ for events with similar false alarm rates, such as the candidate Aframe ranks at 0.17, implies that foreground models built only on detection statistic can rank the same event differently; adding source-property terms should reconcile the rankings and may move some candidates across the 0.5 threshold.
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

3 major / 5 minor

Summary. The paper presents an end-to-end search of LIGO-Virgo-KAGRA O3 data for binary black hole mergers using Aframe, a ResNet-based machine learning search. The network is trained on a 17-day coincident H1-L1 segment at the start of O3 and applied to the full observing run. Sensitivity is estimated via an injection campaign and 100 years of timeslide background, and candidates are ranked by p_astro computed from an injection-based foreground and timeslide background. The search recovers 38 GWTC-3 events with p_astro > 0.5, plus three events first reported in the IAS or OGC catalogs, and no un-vetoed previously unreported candidates. The central claim is that ML-based searches can serve as useful companions to matched-filter pipelines.

Significance. If the calibration concern is resolved, this is a valuable independent validation of an ML-based search on a full observing run, with a reproducible pipeline: public O3 data, the open-source Aframe code, an injection campaign, and systematic comparison with multiple catalogs. Demonstrating that a single trained network recovers the known O3 population at significance comparable to individual matched-filter pipelines is a useful step for the field. The paper also performs new parameter estimation for three candidates not in GWTC-3, which is a concrete additional contribution.

major comments (3)
  1. [Secs. II, III, IV] The network is trained on the 17-day segment 2019-04-01T15:00:00 to 2019-04-17T19:53:20 (Sec. II) and then applied to the entirety of O3, including that same segment. Table II lists three significant candidates from the training period (GW190412 053044, GW190408 181802, GW190413 134308) and Table III lists one low-significance training-period candidate (GW190413 052954). The paper does not report any diagnostic comparing the network's output distribution on the training segment with that on unseen O3 data, nor does it report results with the training period excluded. If the network has partially memorized the noise of its training segment, the detection statistics for these candidates are inflated, biasing their FARs and p_astro; if the 100-year timeslide background includes the training segment (the text in Sec. III is ambiguous), the background calibration for all candidates is affected. Please either exclude the training period from the search or demonstrate that the detection-statistic distribution is stationary across the run, and report how the candidate count changes when the training period is removed.
  2. [Abstract and Sec. IV] The abstract states "No previously un-reported candidates are identified by Aframe," but Sec. IV reports: "One event not reported elsewhere was identified with p_astro = 0.6, but is discarded due to the identification of a whistle glitch by GravitySpy at the inferred coalescence time." This is internally inconsistent: an unreported candidate was identified and then removed by a veto. The claim should be rephrased to state that no unreported candidates survive quality vetoes, and the veto procedure for this particular event should be described, since the abstract's central claim depends on this wording.
  3. [Secs. IV and Appendix A] The comparison of candidate counts uses p_astro > 0.5 as a uniform threshold across pipelines, but Aframe's p_astro (Appendix A) is a function of detection statistic only, whereas the GWTC-3, IAS, OGC, and AresGW foreground models incorporate source-property information. This makes the per-pipeline counts not directly comparable. The paper acknowledges this difference in Sec. V but still draws the quantitative conclusion that Aframe's 38 candidates is "comparable" to the individual matched-filter pipeline counts. Please also report a comparison at fixed FAR thresholds (for example, FAR < 0.01 yr^-1 and FAR < 1 yr^-1) so that sensitivity differences are separated from differences in p_astro modeling.
minor comments (5)
  1. [Sec. III] The text says "signal injections across O3, excluding the 17-day training period. Using the same data, one-hundred years of timeslides are analyzed..." This is ambiguous: does "the same data" mean the timeslides also exclude the training period? Please state explicitly whether the background timeslides include or exclude the training segment, as this is relevant to the contamination concern.
  2. [Sec. IV] There is a typo in the first paragraph: "previously pubished" should be "previously published."
  3. [Sec. IV] The sentence "GWTC-3 70 candidates were reported in GWTC-3 from O3 with Hanford and Livingston data available" is missing punctuation and repeats "GWTC-3"; it should be reworded for clarity.
  4. [Appendix B] The phrase "We perform parameter estimating" should read "We perform parameter estimation."
  5. [Appendix A] In Eq. (A3), the parameters x0 and alpha are not fully defined: how is x0 chosen and over what range of the detection statistic is alpha fit? Please clarify the fitting procedure.

Circularity Check

0 steps flagged · score 1.0 of 10

No material circularity: Aframe's recovery of 38 GWTC-3 events is an out-of-sample benchmark, with p_astro computed from injections and timeslides, not from the comparison catalogs.

full rationale

The derivation chain is not circular. Aframe is an independently published ML pipeline (Ref. [15], same group but a prior PRD paper); this work retrains it on O3 noise plus injections and applies it to O3 data. The two load-bearing outputs—the p_astro values and the candidate list—are produced from an injection-based sensitive-volume/foreground model (Eqs. A2-A6) and a timeslide background, not from GWTC-3, IAS, OGC, or AresGW. The astrophysical rate r=31 Gpc^-3 yr^-1 is taken from the LVK population analysis, an external input. The comparison against catalogs is a benchmark, not a fit: no parameter is adjusted to match the 38 recovered GWTC-3 events. Self-citation of Ref. [15] for the network configuration is method reuse, not a load-bearing uniqueness claim. The only notable caveat is that the network is trained on a 17-day O3 segment and then searches that same segment (three Table II events are asterisked); this is a train/test leakage and potential FAR bias, but it is not a reduction of the result to its inputs by construction. Removing those events still leaves 35 GWTC-3 recoveries, so the central claim does not collapse. No circular step meeting the quoted-evidence standard was found.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The analysis introduces no new entities. It relies on one externally measured rate, one fitted tail slope, and several statistical modeling assumptions common to GW searches; the most paper-specific assumption is the use of a 17-day training segment at the start of O3 as the test set for part of the search.

free parameters (2)
  • astrophysical merger rate r = 31 Gpc^-3 yr^-1
    Taken from the LVK population analysis (Power Law + Spline, Ref. [41]) and used in Eq. A5 to convert sensitive volume V(x) into expected signal rate Rs(x), which directly sets p_astro. The p_astro values would shift with a different rate.
  • exponential-tail slope alpha = fit to background detection-statistic distribution
    Introduced in Eq. A3 to model the high-statistic tail of the noise density fn(x); affects p_astro at large detection statistics where the KDE has no support.
assumptions (4)
  • domain assumption FGMC Poisson mixture model for triggers
    Appendix A assumes foreground and background triggers form a Poisson mixture, from which the p_astro formula (Eq. A2) is derived. This is standard in GW catalogs but an unproven modeling choice.
  • ad hoc to paper Network trained on a 17-day O3 segment generalizes to the rest of O3
    Section II uses 2019-04-01 to 2019-04-17 as the only training/validation data, while Section IV searches all of O3 including that period. The analysis assumes no performance inflation on the training segment.
  • domain assumption Source-property information can be neglected in the p_astro foreground model
    Appendix A builds a foreground model on the detection statistic alone, so p_astro does not use masses or spins. The authors note this causes discrepancies with AresGW (Sec. IV).
  • domain assumption Astrophysical priors for training injections match the O3 sensitivity-estimation population
    Table I priors are based on GWTC-3 sensitivity distributions; if the true O3 population differs, both the training distribution and the foreground model could be biased.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A machine learning-enabled search for binary black hole mergers in LIGO-Virgo-KAGRAs third observing run." pith.science (2026). https://pith.science/paper/UCCZRM63

@misc{pith2026250521261,
  author       = {Pith},
  title        = {Pith review of: A machine learning-enabled search for binary black hole mergers in LIGO-Virgo-KAGRAs third observing run},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UCCZRM63}},
  note         = {Machine review of arXiv:2505.21261}
}
abstract

We conduct a search for stellar-mass binary black hole mergers in gravitational-wave data collected by the LIGO detectors during the LIGO-Virgo-KAGRA (LVK) third observing run (O3). Our search uses a machine learning (ML) based method, Aframe, an alternative to traditional matched filtering search techniques. The O3 observing run has been analyzed by the LVK collaboration, producing GWTC-3, the most recent catalog installment which has been made publicly available in 2021. Various groups outside the LVK have re-analyzed O3 data using both traditional and ML-based approaches. Here, we identify 38 candidates with probability of astrophysical origin ($p_\mathrm{astro}$) greater than 0.5, which were previously reported in GWTC-3. This is comparable to the number of candidates reported by individual matched-filter searches. In addition, we compare Aframe candidates with catalogs from research groups outside of the LVK, identifying three candidates with $p_\mathrm{astro} > 0.5$. No previously un-reported candidates are identified by Aframe. This work demonstrates that Aframe, and ML based searches more generally, are useful companions to matched filtering pipelines.

Figures

Figures reproduced from arXiv: 2505.21261 by the authors.

Figure 1
Figure 1. FIG. 1: Example of Aframe’s neural network output in [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Sensitive volume vs FAR for four different mass distributions. Masses are specified in the source frame. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. summarizes Aframes detections of GWTC-3 events across slices of the parameter space. IAS The IAS algorithm is another matched-filtering￾based detection pipeline. IAS identified a total of 65 significant Hanford/Livingston candidates, 16 of which were reported for the first time by IAS [6, 7]. Of the 65 significant IAS candidates, Aframe identifies 35 as signif￾icant, leaving 30 IAS candidates unidentified by Aframe.… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Bilby prior distribution used for parameter estimation analyses. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Parameter estimation comparison for OGC event GW200106 [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Parameter estimation comparison for IAS event GW190707 [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8: Parameter estimation comparison for IAS event GW190818 [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Searching for binary black hole mergers with deep learning in Advanced LIGO's third observing run

    gr-qc 2025-12 conditional novelty 5.0 of 10

    A hybrid matched-filter/deep-learning pipeline recovers 31 known O3 events and reports a new tentative high-mass candidate, with sensitivity comparable to existing searches only for chirp masses above 25 solar masses.

Reference graph

Works this paper leans on

43 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [1]

    (LIGO Scientific Collaboration and Virgo Collaboration), Phys

    Abbott et al. (LIGO Scientific Collaboration and Virgo Collaboration), Phys. Rev. Lett. 116, 061102 (2016)

    work page 2016

  2. [2]

    The LIGO Scientific Collaboration, J. Aasi, B. P. Ab- bott, R. Abbott, T. Abbott, M. R. Abernathy, K. Ack- ley, C. Adams, T. Adams, P. Addesso, et al., Classical and Quantum Gravity 32, 074001 (2015), URL https: //dx.doi.org/10.1088/0264-9381/32/7/074001

    doi:10.1088/0264-9381/32/7/074001 2015

  3. [3]

    Acernese, M

    F. Acernese, M. Agathos, K. Agatsuma, D. Aisa, N. Alle- mandou, A. Allocca, J. Amarni, P. Astone, G. Balestri, G. Ballardin, et al., Classical and Quantum Gravity 32, 024001 (2014), URL https://dx.doi.org/10.1088/ 0264-9381/32/2/024001

    work page 2014

  4. [4]

    Akutsu, M

    T. Akutsu, M. Ando, K. Arai, Y. Arai, S. Araki, A. Araya, N. Aritomi, Y. Aso, S. Bae, Y. Bae, et al., Progress of Theoretical and Experimen- tal Physics 2021, 05A101 (2020), ISSN 2050- 3911, https://academic.oup.com/ptep/article- pdf/2021/5/05A101/37974994/ptaa125.pdf, URL https://doi.org/10.1093/ptep/ptaa125

    doi:10.1093/ptep/ptaa125 2020

  5. [5]

    Abbott, T

    R. Abbott, T. Abbott, F. Acernese, K. Ackley, C. Adams, N. Adhikari, R. Adhikari, V. Adya, C. Affeldt, D. Agar- wal, et al., Physical Review X 13 (2023), ISSN 2160- 3308, URL http://dx.doi.org/10.1103/PhysRevX.13. 041039

    doi:10.1103/physrevx.13 2023

  6. [6]

    Olsen, T

    S. Olsen, T. Venumadhav, J. Mushkin, J. Roulet, B. Za- ckay, and M. Zaldarriaga, Physical Review D 106 (2022), ISSN 2470-0029, URL http://dx.doi.org/10. 1103/PhysRevD.106.043009

    work page 2022

  7. [7]

    A. K. Mehta, S. Olsen, D. Wadekar, J. Roulet, T. Venu- madhav, J. Mushkin, B. Zackay, and M. Zaldarriaga, New binary black hole mergers in the ligo-virgo o3b data (2024), 2311.06061, URL https://arxiv.org/abs/ 2311.06061

    arXiv 2024

  8. [8]

    A. H. Nitz, C. D. Capano, S. Kumar, Y.-F. . Wang , S. Kastha, M. Schfer, R. Dhurkunde, and M. Cabero, The Astrophysical Journal 922, 76 (2021), ISSN 1538-4357, URL http://dx.doi.org/10.3847/1538-4357/ac1c03

    doi:10.3847/1538-4357/ac1c03 2021

Show all 43 references
  1. [9]

    A. H. Nitz, S. Kumar, Y.-F. . Wang , S. Kastha, S. . Wu , M. Schfer, R. Dhurkunde, and C. D. Capano, The Astrophysical Journal 946, 59 (2023), ISSN 1538-4357, URL http://dx.doi.org/10.3847/1538-4357/aca591

    2023 doi

  2. [10]

    A. E. Koloniari, E. C. Koursoumpa, P. Nousi, P. Lam- propoulos, N. Passalis, A. Tefas, and N. Stergioulas, New gravitational wave discoveries enabled by machine learn- ing (2024), 2407.07820, URL https://arxiv.org/abs/ 2407.07820

    2024 arXiv

  3. [11]

    Abbott et al

    R. Abbott et al. (LIGO Scientific, VIRGO, KAGRA) (2021), 2112.06861

    2021 arXiv

  4. [13]

    LIGO, Virgo and KAGRA observing run plans, https:// observing.docs.ligo.org/plan/, accessed 2024-11-05

    2024

  5. [14]

    Abbott, H

    The LIGO Scientific Collaboration and the Virgo Col- laboration and the KAGRA Collaboration, R. Abbott, H. Abe, F. Acernese, K. Ackley, S. Adhicary, N. Ad- hikari, R. X. Adhikari, V. K. Adkins, V. B. Adya, et al., The Astrophysical Journal Supplement Series 267, 29 (2023), URL...

    2023 doi

  6. [15]

    E. Marx, W. Benoit, A. Gunny, R. Omer, D. Chatter- jee, R. C. Venterea, L. Wills, M. Saleem, E. Moreno, R. Raikman, et al., Phys. Rev. D 111, 042010 (2025), URL https://link.aps.org/doi/10.1103/PhysRevD. 111.042010

    2025 doi

  7. [16]

    E. Marx, W. Benoit, and A. Gunny,Aframe, URL https: //github.com/ML4GW/aframe

  8. [17]

    K. He, X. Zhang, S. Ren, and J. Sun, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) (2016), pp. 770–778

    2016

  9. [18]

    LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Collaboration, GWTC-3: Compact Binary Co- alescences Observed by LIGO and Virgo During the Sec- ond Part of the Third Observing Run O3 search sensitiv- ity estimates (2023), URL https://doi.org/10.5281/ zenodo.7890437

    2023

  10. [19]

    Goyal, P

    P. Goyal, P. Doll´ ar, R. Girshick, P. Noordhuis, L. Wesolowski, A. Kyrola, A. Tulloch, Y. Jia, and K. He, arXiv e-prints arXiv:1706.02677 (2017), 1706.02677

    2017 arXiv

  11. [20]

    Buikema, C

    A. Buikema, C. Cahillane, G. L. Mansell, C. D. Blair, R. Abbott, C. Adams, R. X. Adhikari, A. Ananyeva, S. Appert, K. Arai, et al., Phys. Rev. D 102, 062003 (2020), URL https://link.aps.org/doi/10. 1103/PhysRevD.102.062003

    2020

  12. [21]

    Acernese, M

    F. Acernese, M. Agathos, A. Ain, S. Albanesi, A. Al- locca, A. Amato, T. Andrade, N. Andres, M. Andrs- Carcasona, T. Andri, et al., Classical and Quantum Gravity 40, 185006 (2023), URL https://dx.doi.org/ 10.1088/1361-6382/acd92d

    2023 doi

  13. [22]

    Abbott et al

    R. Abbott et al. (KAGRA, VIRGO, LIGO Scientific), Astrophys. J. Suppl. 267, 29 (2023), 2302.03676

    2023 arXiv

  14. [23]

    B. P. Abbott, R. Abbott, T. D. Abbott, S. Abraham, F. Acernese, K. Ackley, C. Adams, V. B. Adya, C. Af- feldt, M. Agathos, et al., Classical and Quantum Grav- ity 37, 055002 (2020), ISSN 1361-6382, URL http: //dx.doi.org/10.1088/1361-6382/ab685e

    2020 doi

  15. [24]

    Davis, J

    D. Davis, J. S. Areeda, B. K. Berger, R. Bruntz, A. Effler, R. C. Essick, R. P. Fisher, P. Godwin, E. Goetz, A. F. Helmling-Cornell, et al., Classical and Quantum Gravity 38 (2021), ISSN 1361-6382, URL http://dx.doi.org/ 10.1088/1361-6382/abfd85

    2021 doi

  16. [25]

    Acernese et al

    F. Acernese et al. (Virgo) (2022), 2205.01555

    2022 arXiv

  17. [26]

    Christensen, P

    N. Christensen, P. Shawhan, and G. G. f. t. L. collabo- ration), Classical and Quantum Gravity 21, S1747S1755 (2004), ISSN 1361-6382, URL http://dx.doi.org/10. 1088/0264-9381/21/20/017

    2004

  18. [27]

    Tiwari, Classical and Quantum Gravity 35, 145009 (2018), URL https://dx.doi.org/10.1088/1361-6382/ aac89d

    V. Tiwari, Classical and Quantum Gravity 35, 145009 (2018), URL https://dx.doi.org/10.1088/1361-6382/ aac89d

    2018 doi

  19. [28]

    Williams (2024), 2401.08709

    D. Williams (2024), 2401.08709

    2024 arXiv

  20. [29]

    Ewing, R

    B. Ewing, R. Huxford, D. Singh, L. Tsukada, C. Hanna, 7 TABLE II: Candidates that Aframe detects with pastro ≥ 0.5. The catalog each candidate was originally reported in is listed. Events marked with an asterisk occurred during the training period. Candidates identified by Afr...

    2023 arXiv

  21. [30]

    Tsukada, P

    L. Tsukada, P. Joshi, S. Adhicary, R. George, A. Guimaraes, C. Hanna, R. Magee, A. Zimmerman, P. Baral, A. Baylor, et al., Physical Review D 108 (2023), URL https://doi.org/10.1103%2Fphysrevd. 108.043004

    2023

  22. [31]

    Aubin, F

    F. Aubin, F. Brighenti, R. Chierici, D. Estevez, G. Greco, G. M. Guidi, V. Juste, F. Marion, B. Mours, E. Ni- toglia, et al., Classical and Quantum Gravity 38, 095004 (2021), URL https://doi.org/10.1088%2F1361-6382% 2Fabe913

    2021

  23. [32]

    T. D. Canton, A. H. Nitz, B. Gadre, G. S. C. Davies, V. Villa-Ortega, T. Dent, I. Harry, and L. Xiao, The Astrophysical Journal 923, 254 (2021), URL https:// dx.doi.org/10.3847/1538-4357/ac2f9a

    2021 doi

  24. [33]

    Drago, S

    M. Drago, S. Klimenko, C. Lazzaro, E. Milotti, G. Mitsel- makher, V. Necula, B. O’Brian, G. A. Prodi, F. Salemi, M. Szczepanczyk, et al., SoftwareX 14, 100678 (2021), ISSN 2352-7110, URL https://www.sciencedirect. 8 TABLE III: Candidates that Aframe detects with pastro < 0.5. ...

    2021

  25. [34]

    Abbott, T

    R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, N. Adhikari, R. X. Adhikari, V. B. Adya, C. Affeldt, D. Agarwal, et al. (The LIGO Scientific Col- laboration and the Virgo Collaboration), Phys. Rev. D 109, 022001 (2024), URL https://link.aps.org/doi/ 10.1103/PhysRevD...

    2024 doi

  26. [35]

    S. E. Woosley and A. Heger, The Astrophysical Journal Letters 912, L31 (2021), URL https://dx.doi.org/10. 3847/2041-8213/abf2c4

    2021

  27. [36]

    Nousi, A

    P. Nousi, A. E. Koloniari, N. Passalis, P. Iosif, N. Stergioulas, and A. Tefas, Phys. Rev. D 108, 024022 (2023), URL https://link.aps.org/doi/10. 1103/PhysRevD.108.024022

    2023

  28. [37]

    Nagarajan and C

    N. Nagarajan and C. Messenger (2025), 2501.13846

    2025

  29. [38]

    Chatterjee et al., Mach

    D. Chatterjee et al., Mach. Learn. Sci. Tech. 5, 045030 (2024), 2407.19048

    2024 arXiv

  30. [39]

    M. Dax, S. R. Green, J. Gair, J. H. Macke, A. Buo- nanno, and B. Schlkopf, Physical Review Letters 127 (2021), ISSN 1079-7114, URL http://dx.doi.org/10. 1103/PhysRevLett.127.241103

    2021

  31. [40]

    W. M. Farr, J. R. Gair, I. Mandel, and C. Cutler, Physical Review D 91 (2015), ISSN 1550-2368, URL http://dx.doi.org/10.1103/PhysRevD.91.023005

    2015 doi

  32. [41]

    Abbott, T

    The LIGO Scientific Collaboration, the Virgo Collabora- tion, the KAGRA Collaboration, R. Abbott, T. D. Ab- bott, F. Acernese, K. Ackley, C. Adams, N. Adhikari, R. X. Adhikari, et al., arXiv e-prints arXiv:2111.03634 (2021), 2111.03634

    2021 arXiv

  33. [42]

    Ashton, M

    G. Ashton, M. H¨ ubner, P. D. Lasky, C. Talbot, K. Ack- ley, S. Biscoveanu, Q. Chu, A. Divakarla, P. J. Easter, B. Goncharov, et al., The Astrophysical Journal Supple- ment Series 241, 27 (2019), URL https://dx.doi.org/ 10.3847/1538-4365/ab06fc. Appendix A: Probability of Astr...

    2019 doi

  34. [44]

    This mass range matches the range used for simulated signals in this analysis

    from the Power Law + Splinemodel, which is 31 Gpc−3yr−1 for binary systems with component masses in the [5 , 100] M⊙ range. This mass range matches the range used for simulated signals in this analysis. Now, since this mass range consists only of BBH mergers, we implicitly set...

    2000

  35. [2021]

    significant

    Various groups outside the L VK have re-analyzed O3 data using both traditional and ML- based approaches. Here, we identify 38 candidates with probability of astrophysical origin ( pastro) greater than 0.5, which were previously reported in GWTC-3. This is comparable to the nu...

    2015 arXiv

Pith tools

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