Pith. sign in

REVIEW 4 major objections 6 minor 6 cited by

Identifying Microlensing by Compact Dark Matter through Diffraction Patterns in Gravitational Waves with Machine Learning

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims a wavelet-convolutional network can identify gravitational waves diffracted by compact dark-matter lenses at 92.2% accuracy, with no retraining across lens masses from 1 to 1000 solar masses.

desk verdict A promising ML approach to wave-optics lensing, but the evaluation is confounded and internally inconsistent; worth refereeing, not believing yet. read the letter →

arxiv 2509.04538 v1 pith:6O43N2MY submitted 2025-09-04 astro-ph.IM astro-ph.CO

Ao Liu , Tonghua Liu , Dejiang Li , Cuihong Wen , Jieci Wang , Kai Liao , Jiaxing Cui , Huan Zhou This is my paper
classification astro-ph.IM astro-ph.CO
keywords gravitational-wave lensingwave opticsdiffractioncompact dark mattermicrolensingmachine learningwavelet transformconvolutional neural network
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

Microlensing by compact dark matter imprints faint diffraction fringes on gravitational-wave signals, but these wave-optics signatures sit below what geometric-optics lensing searches look for. The paper's aim is to show that a purpose-built neural network, the Wavelet Convolutional Detector (WCD), can pull those fringes out of noisy time-frequency images and identify lensed signals, and to demonstrate this on a realistically simulated population of binary black holes lensed by point-mass compact dark matter. On 10,000 simulated events the network reaches 92.2% validation accuracy (AUC 0.965), rising to AUC≈0.99 at high signal-to-noise ratio, and it keeps its discriminative power across lens masses from 1 to 1000 solar masses without retraining. The eventual payoff is operational: third-generation detectors are expected to produce about 10^5–10^6 events and up to 10^12 pairwise lensing comparisons, a scale that makes current Bayesian lensing identification impractical, so a fast diffraction-aware classifier would turn gravitational microlensing into a practical dark-matter probe.

What carries the argument

The argument is carried by two pieces working together. Physically, the point-mass lens amplification factor F(f) — the analytical solution of the diffraction integral in the wave-optics regime, where the dimensionless frequency w = 8πMLz f controls whether diffraction is significant — maps the unlensed spectrum to the lensed one via h̃L(f) = F(f)h̃(f). Observationally, the WCD classifies whitened, Q-transformed spectrograms; its WTConv layer decomposes feature maps into four wavelet sub-bands (LL, LH, HL, HH), processes them with a depthwise convolution and channel-wise scaling, and reconstructs through an inverse wavelet transform, achieving a 6×6 receptive field from 3×3 kernels while pre

What would settle it

The decisive test is out-of-distribution: run the trained WCD on lensed waveforms generated by an independent pipeline — a non-point-mass lens profile, a log-uniform instead of uniform lens-mass distribution, or the same signals embedded in a different detector's noise with glitches — and check whether the AUC holds. The complementary audit counts false positives: how often the network flags known-unlensed events that carry other time-frequency anomalies such as eccentric orbits, calibration lines, or noise glitches.

Watch

Extended reading notes

Core claim

Gravitational waves lensed by a point-mass compact object are modulated in the wave-optics regime by an analytic amplification factor F(f) — the standard diffraction-integral solution for a point-mass lens — and the paper's central claim is that the resulting frequency-dependent ripples are detectable by a machine-learning classifier even though their amplitude modulation is only a few percent. The WCD learns this from whitened Q-transform spectrograms: wavelet-transform convolutions inside residual blocks give the network a wider effective receptive field and separate low-frequency structure from high-frequency transients, matching the morphology of diffraction fringes. The paper reports 92

Load-bearing premise

Everything is tested on data generated by the paper's own simulation — point-mass lenses, uniform 1–1000 solar-mass lens masses, impact parameters cut at y0 = 3, and Gaussian noise shaped by the Einstein Telescope spectrum — so if real compact-dark-matter lensing or real detector noise differs from these choices, the reported 92% accuracy could be a property of the simulation rather than of the universe.

Editorial extensions

If this is right

  • Third-generation lensing searches could screen the projected 10^10–10^12 pairwise comparisons with a single trained network, reserving Bayesian parameter estimation for the few flagged candidates.
  • A model that generalizes over the full 1–1000 solar-mass lens range acts as an untargeted microlensing search, needing no template bank over lens mass or impact parameter.
  • The stratification results imply that the events found first will be the high-mass, small-impact-parameter ones — exactly the configuration where diffraction is strongest.
  • If the real event rate matches the assumed lensing fraction, the population of flagged events can be used to infer the compact-dark-matter fraction fCO, whose numerical value the simulation itself never fixes.
  • The reported runtime advantage over Bayesian inference is what makes the search scalable rather than merely accurate.
  • The same wavelet-enhanced architecture transfers to other wave-optics problems — extended halo lensing, primordial-black-hole populations, or diffraction by intervening substructure — by retraining on the corresponding transfer functions.

Reading between the lines

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

  • My reading: the honest benchmark for this claim is out-of-distribution, not in-distribution — the decisive test would be feeding the trained WCD lensed waveforms from an independent code with a different lens model (extended mass profile, log-uniform mass function) and watching whether the AUC holds.
  • The paper leaves fCO numerically unspecified; a natural extension is to scan fCO as a free parameter and turn the classifier's event-rate predictions into an actual constraint on the compact-dark-matter fraction.
  • An editorial flag required by the manuscript's own text: in the y-stratified ROC discussion (Section V.C), the prose reports AUC 0.88 for low-y and 0.97 for high-y subsets, while the accompanying figure reports the opposite ordering (0.975 low-y, 0.963 high-y); the figure matches the stated physical trend, but the two sets of numbers cannot both be right and should be reconciled before the stratif
  • A glitch-sensitivity audit would settle how much of the performance is diffraction-specific: if unlensed signals carrying other time-frequency anomalies (eccentricity, calibration lines, detector glitches) are flagged as lensed at elevated rates, part of the 92% would be generic anomaly detection rather than lensing detection.
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 / 6 minor

Summary. The paper proposes a wavelet-enhanced convolutional network, the Wavelet Convolution Detector (WCD), to classify simulated gravitational-wave signals as lensed or unlensed by compact dark matter under the wave-optics (point-mass lens) regime. Lensed waveforms are produced by applying the analytic point-mass amplification factor F(f) of Eq. (5) to IMRPhenomPv2 binary-black-hole signals, adding ET-like noise, whitening, and forming Q-transform spectrograms. The training set uses physically motivated source redshift, lens redshift, impact-parameter, and lens-mass distributions described in Section III, with 5000 lensed and 5000 unlensed examples. The paper reports validation accuracy 92.2%, full-dataset accuracy 98.24%, ROC AUC 0.965 on test data, and claims that performance improves at high SNR, low impact parameter, and high lens mass, concluding that the WCD is the first ML approach to identify wave-optics diffraction signatures in lensed gravitational waves.

Significance. If the reported performance were shown to isolate genuine wave-optics diffraction signatures, the WCD architecture and the data-generation pipeline would be a useful contribution for third-generation detectors: the wavelet multi-scale mechanism is a reasonable inductive bias for time-frequency diffraction features, and the use of optical-depth-weighted lens parameters is more astrophysically motivated than uniform sampling. The paper also specifies hyperparameters and training details, which supports reproducibility. However, the central claim is currently supported only by a closed simulation loop: the classifier is trained and tested on data generated by the same pipeline, and there are unresolved confounds from unmatched SNR and source-selection distributions, an internal contradiction in the impact-parameter dependence, and inconsistent reporting of accuracy on different data splits. The physical interpretation therefore needs substantial additional work before the accuracy/AUC numbers can be attributed to diffraction-pattern recognition.

major comments (4)
  1. [Section IV.B, Eq. (1), Eq. (11)] The manuscript never specifies the SNR distribution of the unlensed signals. Section IV.B states only that "the lensed GW signals are embedded within the noisy background with a SNR ∈ [10, 50]". Since the point-mass amplification factor F(f) in Eq. (1) changes the signal amplitude, if lensed and unlensed injections are not explicitly renormalized to the same SNR distribution, the classifier can trivially separate the classes by overall loudness. In addition, lensed events are selected according to PL(zs) in Eq. (11), so their source redshift distribution differs from that of unlensed events, changing observed masses, frequencies, and luminosity distances. The paper needs to state the unlensed SNR generation, show that the two classes have matched SNR and matched source-parameter distributions, and ideally present a control experiment (e.g., lensed vs. unlensed with identical source param
  2. [Section V.C, Figure 5(b)] There is a direct internal contradiction in the impact-parameter analysis. The text first says AUC "increases systematically as y decreases" and then states "the AUC drops to 0.88 for low y subsets (vs. 0.97 for high y subsets)", whereas the caption of Figure 5(b) reports Low y AUC = 0.975 and High y AUC = 0.963. These statements are opposite in sign and the numbers do not match. Since the abstract and conclusion explicitly claim superior performance in the low-impact-parameter regime, this inconsistency is load-bearing. The authors should correct the text, recompute the ROC curves, and state the actual subset sizes and thresholds used.
  3. [Section V.B, Figure 4, Abstract] The accuracy metrics are reported on inconsistent data splits. The validation confusion matrix sums to 1000 samples, the full-dataset confusion matrix sums to 10000, and the stated split is 8:1:1. Thus the "full dataset" accuracy of 98.24% includes the training set and is not an unbiased performance measure. The abstract's 92.2% accuracy is the validation accuracy, while the AUC of 0.965 appears to come from the test ROC; these are not from the same split. The paper should report a single held-out test-set confusion matrix with accuracy, precision, recall, and AUC, and should not quote metrics computed on training data as evidence of generalization.
  4. [Section III.A, Eq. (13), Section I] Several load-bearing elements of the physical interpretation are missing. First, the compact-DM fraction fCO in Eq. (13) is never numerically specified, so the reader cannot assess the assumed lensing rate; because the training set is artificially balanced 5000/5000, fCO cancels from the conditional lens parameter distributions, and therefore the reported accuracy says nothing about the detectability of a realistic DM population. Second, the Introduction promises "a domain adaptation strategy to ensure model robustness on unseen real data," but no domain adaptation method or experiment appears in Sections IV--VI. Third, all validation is on the same simulation pipeline; the classifier may be exploiting simulation-specific artifacts. The authors should either add out-of-distribution tests (different waveform approximant, different lens model, non-Gaussian or real noise, different mass fun
minor comments (6)
  1. [Section V.A, Table III] The training narrative says the learning rate was reduced following "an exponential decay schedule," but Table III specifies a cosine-annealing scheduler. Please align these statements.
  2. [Section V.B, Figure 4] The sentence "as reflected in right panel of Figure 4 left" is garbled. The figure actually has validation on the left and full dataset on the right; please rephrase.
  3. [Abstract] The claim of being "the first machine learning-based approach capable of identifying such wave-optics signatures" is broader than what is demonstrated, especially given previous wave-optics lensing studies cited in the text. Please qualify the novelty statement.
  4. [Section III.A, Eq. (14)] The display of Eq. (14) is malformed in the manuscript text, and the denominator notation is unclear. Please rewrite the probability expression with explicit limits and normalization.
  5. [Throughout] There are numerous typographical and grammatical errors, e.g., "comving volume," "we presents," "not enhances," and "the lensing probability distribution zl an be obtained." A careful editing pass is needed.
  6. [Section V.C] The statement that for y > 2 the waveform has "about 5% change in peak amplitude" is not derived or referenced. If this number is used to support the interpretation, please provide a quantitative basis or remove it.

Circularity Check

1 steps flagged · score 3.0 of 10

Classification benchmark is a legitimate supervised-learning result, but the paper's validation that the model 'decodes genuine diffraction' is circular because the lensed class is generated by the same F(f) whose y/mlen dependence is then read off as a physical confirmation.

  1. self definitional [Section V.C (ROC curves, lens-mass dependence); with Eq. (1) and Eq. (5) in Sec. II.A]
    "This trend aligns with wave-optics theory (diffraction prominence requires RE ∼ λGW), validating the model decodes genuine diffraction-induced lensing imprints rather than spurious correlations."

    The lensed training/test signals are generated as ehL(f)=F(f)eh(f) (Eq. 1), where F(f) is given by Eq. (5) and depends on w=8πMLz f (hence mlen) and y. Thus the simulated lensed class contains larger F(f)-induced distortions exactly for high mlen and low y. The observed AUC stratification (high-m AUC=0.995, low-m AUC=0.944; low-y AUC=0.975, high-y AUC=0.963) is therefore a direct consequence of the data-generation construction. Using this trend to 'validate' that the network has learned genuine diffraction rather than spurious correlations restates the input model; it is a self-consistency check, not an independent confirmation.

full rationale

No load-bearing self-citation or uniqueness-import circularity was found: the wave-optics transfer function is cited to Takahashi & Nakamura, and the astrophysical distributions (Eqs. 6-15) rest on external population-synthesis references. The 92.2% accuracy and AUC=0.965 are obtained on held-out data from the same simulation pipeline; this is a standard supervised-learning evaluation and is not itself circular. The main circular element is the interpretive claim in Sec. V.C that the model's performance trend with y and mlen independently validates diffraction learning, when that trend is engineered into the training data by Eq. (1)/(5). A separate non-circularity concern: the lensed set's SNR is specified ([10,50]) while the unlensed set's SNR is not, so loudness/source-parameter imbalance is a confounding risk for real-data claims; this is an external-validity limitation, not a circular reduction. Overall, partial circularity in the physical-interpretation validation, but the core classification benchmark retains independent content.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a simulation pipeline with several hand-chosen distributions (lens mass, impact parameter cutoff, SNR range) and physics assumptions (point-mass lens, Gaussian noise, IMRPhenomPv2). The unspecified fCO weakens the optical-depth normalization. No new physical entities are introduced; the WCD is a methodological construct, not a new particle or force.

free parameters (4)
  • Lens mass range = 1 to 1000 M_sun
    Uniform prior on lens mass, chosen by hand with no physical justification beyond covering stellar to intermediate-mass compact objects.
  • Impact parameter cutoff y0 = 3
    P(y)=2y/y0^2 with y0=3; the cutoff is ad hoc, described as 'difficult to confidently identify' beyond this value.
  • SNR range = 10 to 50
    Lensed signals embedded with SNR in [10,50]; the range is a modeling choice that directly affects class separability.
  • fCO (compact DM fraction) = not specified
    Appears in the optical depth formula Eq. (13) but no numerical value is given in the paper, making the lensing probability normalization incomplete.
assumptions (5)
  • domain assumption Point-mass lens model with analytical diffraction amplification factor F(f) (Eq. 5)
    All lenses are treated as point masses with the Takahashi-Nakamura solution; extended or multiple lenses are ignored.
  • domain assumption Compact DM distributed uniformly in comoving volume; optical depth formula (Eqs. 11-13)
    Used to generate lens redshift and impact parameter distributions; assumes a specific cosmological DM distribution.
  • domain assumption IMRPhenomPv2 accurately models BBH waveforms
    All simulated signals use this waveform model; real GW signals may have more complex morphology.
  • domain assumption Gaussian noise with Einstein Telescope PSD
    Simulated noise is Gaussian and stationary; real detector noise is non-Gaussian and non-stationary.
  • ad hoc to paper Uniform lens mass distribution in [1, 1000] M_sun
    No underlying mass function for compact DM is used; the uniform range is chosen for the simulation and affects all reported performance numbers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Identifying Microlensing by Compact Dark Matter through Diffraction Patterns in Gravitational Waves with Machine Learning." pith.science (2026). https://pith.science/paper/6O43N2MY

@misc{pith2026250904538,
  author       = {Pith},
  title        = {Pith review of: Identifying Microlensing by Compact Dark Matter through Diffraction Patterns in Gravitational Waves with Machine Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6O43N2MY}},
  note         = {Machine review of arXiv:2509.04538}
}
abstract

Gravitational wave lensing, particularly microlensing by compact dark matter (DM), offers a unique avenue to probe the nature of dark matter. However, conventional detection methods are often computationally expensive, inefficient, and sensitive to waveform systematics. In this work, we introduce the Wavelet Convolution Detector (WCD), a deep learning framework specifically designed to identify wave-optics diffraction patterns imprinted in gravitationally lensed signals. The WCD integrates multi-scale wavelet analysis within residual convolutional blocks to efficiently extract time-frequency interference structures, and is trained on a realistically generated dataset incorporating compact DM mass functions and astrophysical lensing probabilities. This work is the first machine learning-based approach capable of identifying such wave-optics signatures in lensed gravitational waves. Tested on simulated binary black hole events, the model achieves 92.2\% accuracy (AUC=0.965), with performance rising to AUC$\sim$0.99 at high SNR. Crucially, it maintains high discriminative power across a wide range of lens masses without retraining, demonstrating particular strength in the low-impact-parameter and high-lens-mass regimes where wave-optics effects are most pronounced. Compared to Bayesian inference, the WCD provides orders-of-magnitude faster inference, making it a scalable and efficient tool for discovering compact DM through lensed gravitational waves in the era of third-generation detectors.

Figures

Figures reproduced from arXiv: 2509.04538 by the authors.

Figure 1
Figure 1. FIG. 1: Gravitational waveforms illustrating the lensing effect of point mass lenses on gravitational waves. The figure shows [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Comparison of time-frequency images between unlensed and lensed gravitational wave signals across varying noise [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Architecture of the Wavelet Convolutional Detector featuring WTConv modules integrated within residual blocks. The [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: ROC curves evaluating model performance: (a) SNR dependence: Model performance improves with increasing SNR; [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 6 Pith papers

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

  1. Wave Optics Effects from Gravitational Wave Propagation Through Dark Matter Halos

    gr-qc 2026-06 unverdicted novelty 7.0 of 10

    Numerical simulations of plane gravitational waves through Gaussian, NFW and Burkert potentials show non-geodesic propagation for Gaussian profiles and wave convexity reversal for Burkert in strong gravity, with order...

  2. Identifying lensed gravitational waves with physics-informed posterior learning

    gr-qc 2026-07 conditional novelty 6.0 of 10

    Fusing a simulation-trained common-source mass posterior with waveform features raises lensed-event detection efficiency from 20.8% to 35.2% at 1% false-positive rate and lowers the SNR for 50% efficiency from 45.3 to 33.5.

  3. Mock Catalogs of Strongly Lensed Gravitational Waves via A Halo Model Approach with Ground-based Detectors

    astro-ph.CO 2026-03 accept novelty 6.0 of 10

    Composite-halo mock catalogs forecast ~400 doublets + 36 quadruplets (plus ~107 subhalo and ~20 central-image systems) of lensed GWs per year for ET+CE and release the GW-LMC catalog.

  4. Finite-Core Signatures in LISA-Band Wave-Optics Lensing by Low-Mass Dark Matter Halos

    astro-ph.CO 2026-06 unverdicted novelty 5.0 of 10

    Finite cores in low-mass dark matter halos produce distinct complex residuals in LISA-band wave-optics amplification that cannot be fully mimicked by lower-concentration NFW profiles and peak at rc/rs ≃ 0.25-0.3.

  5. Detection of Lensed Gravitational Waves in the Millihertz Band Using Frequency-Domain Lensing Feature Extraction Network

    astro-ph.IM 2025-12 unverdicted novelty 5.0 of 10

    DCL-xLSTM neural network detects lensed GW events with AUC over 0.99 using training on PM and SIS lens models in the millihertz band.

  6. Detection of Multiband Lensed Gravitational Waves from Dark Matter Halos with Deep Learning

    astro-ph.IM 2025-11 conditional novelty 5.0 of 10

    A dual-branch neural network fusing simulated DECIGO and ET data classifies SIS, CIS, and NFW lensed binary-neutron-star signals with 97% accuracy, far above single-detector performance.

Reference graph

Works this paper leans on

74 extracted references · 19 canonical work pages · cited by 6 Pith papers

  1. [1]

    Architecture Overview The network input consists of spectrograms generated by applying noise whitening (based on the PSD) and a Q-transform to the raw gravitational-wave (GW) signals. The proposed network adopts a ResNet backbone with five hierarchical stages, where conventional bottleneck blocks are augmented with Wavelet Transform Convo- lution (WTConv)...

    work page

  2. [2]

    WTConv Operational Details The WTConv module implements frequency-adaptive processing through dual-path computation: H(X) = Convk×k(X), Ω(X) = W −1 (K ⊛ W(X)) , (22) WTConv(X) = H(X) + Ω(X). The wavelet transformW decomposes features into four sub-bands: W(X) = Z = [LL,LH,HL,HH] ∈ RC×4×H/2×W/2, (23) where Z denotes the concatenated low–pass (LL) and three...

    work page

  3. [3]

    Parameter efficiency through depthwise wavelet pro- cessing, offering marked reductions in parameter count compared to conventional convolutions with equivalent spatial coverage. These properties make WTConv par- ticularly suitable for gravitational-wave analysis, where signal morphology manifests as localized time-frequency patterns embedded in non-stati...

    work page 2000

  4. [4]

    Graviton mass bounds from space-based gravitational-wave observations of massive black hole populations

    E. Berti, J. Gair, and A. Sesana, Phys. Rev. D 84, 101501 (2011), 1107.3528

    work page Pith review arXiv 2011

  5. [5]

    B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Aber- nathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, et al., Phys. Rev. Lett.116, 061102 (2016), 1602.03837

    arXiv 2016

  6. [6]

    B. P. Abbott, R. Abbott, T. D. Abbott, M. R. Aber- nathy, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Adhikari, et al., Astrophys. J., Suppl. Ser. 225, 8 (2016), 1604.07864

    work page Pith review arXiv 2016

  7. [7]

    B. P. Abbott, R. Abbott, T. D. Abbott, S. Abraham, F. Acernese, K. Ackley, C. Adams, R. X. Adhikari, V. B. Adya, C. Affeldt, et al., Phys. Rev. D 100, 104036 (2019), 1903.04467

    arXiv 2019

  8. [8]

    B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Ad- hikari, V. B. Adya, et al., Phys. Rev. Lett.119, 161101 (2017), 1710.05832

    arXiv 2017

Show all 74 references
  1. [9]

    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 Gravity32, 024001 (2015), 1408.3978

    2015 arXiv

  2. [10]

    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 Experimental Physics2021, 05A101 (2021), 2005.05574

    2021

  3. [11]

    B. P. Abbott, R. Abbott, T. D. Abbott, S. Abraham, F. Acernese, K. Ackley, C. Adams, R. X. Adhikari, V. B. Adya, C. Affeldt, et al., Astrophys. J. Lett. 882, L24 (2019), 1811.12940

    2019 arXiv

  4. [12]

    B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Ad- hikari, V. B. Adya, et al., Nature (London) 551, 85 (2017), 1710.05835

    2017 arXiv

  5. [13]

    B. P. Abbott, R. Abbott, T. D. Abbott, F. Acernese, K. Ackley, C. Adams, T. Adams, P. Addesso, R. X. Ad- hikari, V. B. Adya, et al., Astrophys. J. Lett.848, L12 (2017), 1710.05833

    2017 arXiv

  6. [14]

    Barack, V

    L. Barack, V. Cardoso, S. Nissanke, T. P. Sotiriou, A. Askar, C. Belczynski, G. Bertone, E. Bon, D. Blas, R. Brito, et al., Classical and Quantum Gravity 36, 143001 (2019), 1806.05195

    2019 arXiv

  7. [15]

    Berti, E

    E. Berti, E. Barausse, V. Cardoso, L. Gualtieri, P. Pani, U. Sperhake, L. C. Stein, N. Wex, K. Yagi, T. Baker, et al., Classical and Quantum Gravity32, 243001 (2015), 1501.07274

    2015 arXiv

  8. [16]

    Oguri, Reports on Progress in Physics 82, 126901 (2019), 1907.06830

    M. Oguri, Reports on Progress in Physics 82, 126901 (2019), 1907.06830

    2019 arXiv

  9. [17]

    Zhu and S

    W. Zhu and S. Dong, araa 59, 291 (2021), 2103.02127

    2021 arXiv

  10. [18]

    Annala, T

    E. Annala, T. Gorda, A. Kurkela, J. Nättilä, and A. Vuorinen, Nature Physics16, 907 (2020), 1903.09121

    2020 arXiv

  11. [19]

    A. R. A. C. Wierda, E. Wempe, O. A. Hannuksela, L. V. E. Koopmans, and C. Van Den Broeck, Astrophys. J. 921, 154 (2021), 2106.06303

    2021 arXiv

  12. [20]

    Nakamura, Phys

    K. Nakamura, Phys. Rev. D 80, 124021 (2009), 0804.3840

    2009 arXiv

  13. [21]

    K. Liao, M. Biesiada, and Z.-H. Zhu, Chinese Physics Letters 39, 119801 (2022), 2207.13489

    2022 arXiv

  14. [22]

    Vegetti and L

    S. Vegetti and L. V. E. Koopmans, Mon. Not. Roy. As- tron. Soc. 400, 1583 (2009), 0903.4752

    2009 arXiv

  15. [23]

    Y. D. Hezaveh, L. Perreault Levasseur, and P. J. Mar- shall, Nature (London) 548, 555 (2017), 1708.08842

    2017 arXiv

  16. [24]

    Liao, X.-L

    K. Liao, X.-L. Fan, X. Ding, M. Biesiada, and Z.-H. Zhu, Nature Communications 8, 1148 (2017), 1703.04151

    2017 arXiv

  17. [25]

    Takahashi and T

    R. Takahashi and T. Nakamura, Astrophys. J.595, 1039 (2003), astro-ph/0305055

    2003 arXiv

  18. [26]

    Dai, S.-S

    L. Dai, S.-S. Li, B. Zackay, S. Mao, and Y. Lu, Phys. Rev. D 98, 104029 (2018), 1810.00003

    2018 arXiv

  19. [27]

    M.-R. Wu, J. Barnes, G. Martínez-Pinedo, and B. D. Metzger, Phys. Rev. Lett. 122, 062701 (2019), 1808.10459. 12

    2019 arXiv

  20. [28]

    N. J. Cornish and T. B. Littenberg, Classical and Quan- tum Gravity 32, 135012 (2015), 1410.3835

    2015 arXiv

  21. [29]

    Birrer, M

    S. Birrer, M. Millon, D. Sluse, A. J. Shajib, F. Courbin, S. Erickson, L. V. E. Koopmans, S. H. Suyu, and T. Treu, ssr 220, 48 (2024), 2210.10833

    2024 arXiv

  22. [30]

    S. Cao, J. Qi, Z. Cao, M. Biesiada, W. Cheng, and Z.-H. Zhu, Astron. Astrophys.659, L5 (2022), 2202.08714

    2022 arXiv

  23. [31]

    B. J. Owen and B. S. Sathyaprakash, Phys. Rev. D60, 022002 (1999), gr-qc/9808076

    1999 arXiv

  24. [32]

    Çalışkan, J

    M. Çalışkan, J. M. Ezquiaga, O. A. Hannuksela, and D. E. Holz, Phys. Rev. D 107, 063023 (2023), 2201.04619

    2023 arXiv

  25. [33]

    Veitch, V

    J. Veitch, V. Raymond, B. Farr, W. Farr, P. Graff, S. Vi- tale, B. Aylott, K. Blackburn, N. Christensen, M. Cough- lin, et al., Phys. Rev. D91, 042003 (2015), 1409.7215

    2015 arXiv

  26. [34]

    Janquart, T

    J. Janquart, T. Baka, A. Samajdar, T. Dietrich, and C. Van Den Broeck, Mon. Not. Roy. Astron. Soc.523, 1699 (2023), 2211.01304

    2023 arXiv

  27. [35]

    R. W. Kiendrebeogo, A. M. Farah, E. M. Foley, A. Gray, N. Kunert, A. Puecher, A. Toivonen, R. O. VandenBerg, S. Anand, T. Ahumada, et al., Astrophys. J.958, 158 (2023), 2306.09234

    2023 arXiv

  28. [36]

    Z. Gao, K. Liao, L. Yang, and Z.-H. Zhu, Mon. Not. Roy. Astron. Soc. 526, 682 (2023), 2304.13967

    2023 arXiv

  29. [37]

    Gabbard, M

    H. Gabbard, M. Williams, F. Hayes, and C. Messenger, Phys. Rev. Lett. 120, 141103 (2018), 1712.06041

    2018 arXiv

  30. [38]

    S. R. Green and J. Gair, arXiv e-prints arXiv:2008.03312 (2020), 2008.03312

    2008 arXiv

  31. [39]

    A. J. K. Chua and M. Vallisneri, Phys. Rev. Lett.124, 041102 (2020), 1909.05966

    2020 arXiv

  32. [40]

    K. Kim, J. Lee, R. S. H. Yuen, O. Akseli Hannuksela, and T. G. F. Li, arXiv e-prints arXiv:2010.12093 (2020), 2010.12093

    2010 arXiv

  33. [41]

    T. E. Collett, Astrophys. J.811, 20 (2015), 1507.02657

    2015 arXiv

  34. [42]

    Magare, A

    S. Magare, A. More, and S. Choudhary, Mon. Not. Roy. Astron. Soc. 535, 990 (2024), 2403.02994

    2024 arXiv

  35. [43]

    and other machine learning methods [44, 45] to the problem of lens identification. Although existing studies have made progress in lensed/unlensed classification, most approaches remain confined to the geometric-optics approximation, lack targeted modeling of wave-optical effe...

  36. [44]

    Goyal, D

    S. Goyal, D. Harikrishnan, S. J. Kapadia, and P. Ajith, Phys. Rev. D 104, 124057 (2021), 2106.12466

    2021 arXiv

  37. [45]

    Lin, J.-d

    X.-y. Lin, J.-d. Zhang, L. Dai, S.-J. Huang, and J. Mei, Phys. Rev. D 108, 064020 (2023), 2304.04800

    2023 arXiv

  38. [46]

    Pagano, O

    G. Pagano, O. A. Hannuksela, and T. G. F. Li, Astron. Astrophys. 643, A167 (2020), 2006.12879

    2020 arXiv

  39. [47]

    Sheng, K

    S. Sheng, K. V. G. C, C. P. Choi, J. Sharpnack, and T. Jones, arXiv e-prints arXiv:2210.11681 (2022), 2210.11681

    2022 arXiv

  40. [48]

    G. C. Keerthi Vasan, S. Sheng, T. Jones, C. P. Choi, and J. Sharpnack, Mon. Not. Roy. Astron. Soc. 524, 5368 (2023), 2211.00047

    2023 arXiv

  41. [49]

    Chakraborty and S

    A. Chakraborty and S. Mukherjee, Mon. Not. Roy. As- tron. Soc. 532, 4842 (2024), 2403.03982

    2024 arXiv

  42. [50]

    M. H. Y. Cheung, J. Gais, O. A. Hannuksela, and T. G. F. Li, Mon. Not. Roy. Astron. Soc. 503, 3326 (2021), 2012.07800

    2021 arXiv

  43. [51]

    S. E. Finder, R. Amoyal, E. Treister, and O. Freifeld, in European Conference on Computer Vision (2024), 2407.05848

    2024 arXiv

  44. [52]

    K. He, X. Zhang, S. Ren, and J. Sun, in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR (2016), p. 1, 1512.03385

    2016 arXiv

  45. [53]

    Grespan and M

    M. Grespan and M. Biesiada, Universe9, 200 (2023)

    2023

  46. [54]

    A. K. Meena and J. S. Bagla, Mon. Not. Roy. Astron. Soc. 492, 1127 (2020), 1903.11809

    2020 arXiv

  47. [55]

    S. Sun, C. Shi, J.-d. Zhang, and J. Mei, Phys. Rev. D 107, 044023 (2023), 2207.13009

    2023 arXiv

  48. [56]

    Urrutia and V

    J. Urrutia and V. Vaskonen, Mon. Not. Roy. Astron. Soc. 509, 1358 (2022), 2109.03213

    2022 arXiv

  49. [57]

    Mukherjee, T

    S. Mukherjee, T. Broadhurst, J. M. Diego, J. Silk, and G. F. Smoot, Mon. Not. Roy. Astron. Soc. 506, 3751 (2021), 2106.00392

    2021 arXiv

  50. [58]

    Madau and T

    P. Madau and T. Fragos, Astrophys. J.840, 39 (2017), 1606.07887

    2017 arXiv

  51. [59]

    K. Liao, S. Tian, and X. Ding, Mon. Not. Roy. Astron. Soc. 495, 2002 (2020), 2001.07891

    2002 arXiv

  52. [60]

    Basak, A

    S. Basak, A. Ganguly, K. Haris, S. Kapadia, A. K. Mehta, and P. Ajith, Astrophys. J. Lett.926, L28 (2022), 2109.06456

    2022 arXiv

  53. [61]

    H. Zhou, Z. Li, K. Liao, and Z. Huang, Mon. Not. Roy. Astron. Soc. 518, 149 (2023), 2206.13128

    2023 arXiv

  54. [62]

    X. Shan, B. Hu, X. Chen, and R.-G. Cai, Nature Astron- omy 9, 916 (2025), 2301.06117

    2025 arXiv

  55. [63]

    Lin, X.-J

    X.-Y. Lin, X.-J. Wang, H. Zhou, Z. Li, K. Liao, and Z.-H. Zhu, arXiv e-prints arXiv:2508.13577 (2025), 2508.13577

    2025

  56. [64]

    Suyamprakasam, S

    S. Suyamprakasam, S. Harikumar, P. Ciecielág, P. Figura, M. Bejger, and M. Biesiada, arXiv e-prints arXiv:2503.21845 (2025), 2503.21845

    2025

  57. [65]

    K. Liao, M. Biesiada, and X.-L. Fan, Astrophys. J.875, 139 (2019), 1903.06612

    2019 arXiv

  58. [66]

    LeCun, Y

    Y. LeCun, Y. Bengio, and G. Hinton, Nature521, 436 (2015)

    2015

  59. [67]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, inAd- vances in Neural Information Processing Systems(2012), pp. 1097–1105

    2012

  60. [68]

    C. Dong, C. C. Loy, K. He, and X. Tang, inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition(2015), pp. 166–174

    2015

  61. [69]

    Szegedy, W

    C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabi- novich, inProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition(2015), pp. 1–9, 1409.4842

    2015 arXiv

  62. [70]

    Dosovitskiy, L

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, et al., inInternational Conference on Learning Representations(2021), 2010.11929

    2021 arXiv

  63. [71]

    A. Nitz, I. Harry, D. Brown, C. M. Biwer, J. Willis, T. D. Canton, C. Capano, T. Dent, L. Pekowsky, G. S. C. Davies, et al., gwastro/pycbc: v2.3.3 release of pycbc (2024)

    2024

  64. [72]

    D. P. Kingma and J. Ba, CoRRabs/1412.6980 (2014)

    2014 arXiv

  65. [73]

    Loshchilov and F

    I. Loshchilov and F. Hutter, inProc. of ICLR (2017), 1608.03983

    2017 arXiv

  66. [74]

    Fawcett, Pattern Recognition Letters27, 861 (2006)

    T. Fawcett, Pattern Recognition Letters27, 861 (2006)

    2006

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.