Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

DeepHMC : a deep-neural-network acclerated Hamiltonian Monte Carlo algorithm for binary neutron star parameter estimation

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read DeepHMC replaces expensive gravitational-wave likelihood gradients with a trained neural network, making Hamiltonian Monte Carlo practical for binary neutron star parameter estimation.

desk verdict A genuine acceleration technique with measured speedups and plausible posteriors, but the paper leaves the exact target distribution of Phase III underspecified and the validation is not apples-to-apples. read the letter →

arxiv 2505.02589 v1 pith:G4OH4ZSO submitted 2025-05-05 gr-qc astro-ph.IM

Jules Perret , Marc Aréne , Edward K. Porter This is my paper · ORCID
classification gr-qcastro-ph.IM
keywords HamiltonianMonteCarlodeepneuralnetworkgradientsurrogategravitational-waveparameterestimationbinaryneutronstarrelativebinningGW170817GW190425posteriorsampling
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 remove the main obstacle to using Hamiltonian Monte Carlo for gravitational-wave parameter estimation: the cost of computing gradients of the log-likelihood at every leapfrog step. The proposed algorithm, DeepHMC, first runs a short phase of ordinary HMC with numerical gradients, then trains a deep neural network to predict those gradients from the sampled positions, and finally uses the network's gradients to integrate long HMC trajectories. The paper argues this preserves the posterior while cutting trajectory cost by a factor of 30 relative to relative-binning gradients and about 7000 relative to naive likelihood gradients. Applied to the 128-second public data for the binary neutron star mergers GW170817 and GW190425, DeepHMC produces posteriors that agree with the public catalog results, obtaining 5000 statistically independent samples for GW170817 in about two hours on a laptop. If true, this would make full 12-dimensional neutron-star binary inference a desktop-scale problem and open the door to much longer signals from future detectors.

What carries the argument

The load-bearing object is the DNN gradient surrogate: a fully connected network with three hidden layers of 10D, 10D, and 100D neurons plus a 25% dropout layer, trained by mean-squared-error regression on Phase I data. It plays the role of a shadow potential: Phase III leapfrog steps use the DNN's predicted gradients instead of waveform-derived gradients, while the start and end of each trajectory are evaluated with the true potential to compute the Metropolis acceptance. Also load-bearing are the coordinate choice of log chirp mass, log reduced mass, dimensionless spins, tidal parameters, log distance, sky position, inclination, and log duration with phase marginalized; reflective boundary handling for priors and the equal-mass line; and a sinusoidal regularization of the spin-prior gradient singularity.

What would settle it

Run DeepHMC and an exact-gradient HMC sampler with the same waveform, priors, and data on a low-signal-to-noise or multimodal event, and check whether posterior means and credible intervals differ by more than the sampling error. A more targeted version: take a GW190425-like injection, measure the DNN gradient error in low-likelihood regions, and test whether increasing that error changes the inferred tidal parameters.

Watch

Extended reading notes

Core claim

The central claim is that the gradient bottleneck of HMC can be bypassed with a learned surrogate without breaking the sampler: Hamiltonian trajectories in the final sampling phase may be integrated using DNN-approximated gradients, provided the network was trained on positions and numerical gradients collected during an earlier information-gathering phase. The endpoint Metropolis-Hastings acceptance still uses the true Hamiltonian, so the chain remains an approximate HMC sampler whose target is the true posterior. On the two real events tested, the algorithm reproduces the public posterior medians and credible intervals for intrinsic parameters, with autocorrelation times around 10 lags for GW170817 and a few hundred for GW190425, translating into roughly one independent sample per second and one per 26 seconds respectively.

Load-bearing premise

The sampler's final distribution is still the true posterior even though the leapfrog trajectories are integrated with approximate neural-network gradients instead of the exact likelihood gradients; the paper supports this with fit quality and overlap with public samples, but not with a same-setup comparison to a reference sampler or a bias diagnostic.

Editorial extensions

If this is right

  • Binary neutron star parameter estimation with spins and tides can run on a laptop in hours rather than days or weeks for high-signal-to-noise events.
  • Because no a priori classification of unimodal versus multimodal parameters is required, the same pipeline handles both the unimodal GW170817 posterior and the multimodal GW190425 posterior.
  • The per-trajectory speedup makes HMC competitive for the longer-duration signals expected from third-generation detectors, where random-walk samplers are expected to slow down.
  • DeepHMC can be pointed at new waveform models without re-deriving hand-tuned gradient approximations, since the network learns gradients from data.
  • The Phase I low-latency localization, giving a sky area and distance estimate in about 20 minutes, suggests the method could support early warning or rapid follow-up.

Reading between the lines

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

  • Inference: the same Phase I/II/III surrogate idea could be applied with more expressive gradient emulators or to precessing waveforms, but the paper only demonstrates aligned spins and a single waveform family.
  • Inference: the reported 30x and 7000x trajectory speedups are measured within this pipeline against relative-binning and naive likelihood gradients; the end-to-end gain is smaller once Phase I and Phase II training are included, so the fairest user-facing metric is wall-clock time to a fixed number of independent samples under identical settings.
  • Inference: using DNN gradients inside the leapfrog integration makes the sampler's stationary distribution only approximately the target; a natural testable extension is to use the DNN only as a proposal generator and correct with an exact Metropolis step, which would remove the bias at some computational cost.
  • Inference: the lower R-squared scores and slower autocorrelations for GW190425 suggest the method's advantage is signal-to-noise dependent, so a bias diagnostic comparing DeepHMC chains against an exact-gradient chain on a subset of runs should become a standard validation step.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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 DeepHMC, a Hamiltonian Monte Carlo algorithm for binary neutron star parameter estimation in which the expensive log-likelihood gradients are replaced by gradients predicted by a fully connected deep neural network. The algorithm is structured in three phases: Phase I collects trajectory points and numerical gradients; Phase II trains the DNN and updates the mass-matrix scales from the Phase I covariance; Phase III runs production trajectories using the DNN gradients. The authors report that DNN-gradient trajectories are about 30 times faster than relative-binning-gradient trajectories and about 7000 times faster than naive-gradient trajectories, and that for GW170817 they obtain 5000 statistically independent samples in about two hours on a MacBook Pro, while for GW190425 they obtain 17,500 independent samples in about 2.5 days. They validate the resulting posteriors by comparing medians and 90% credible intervals with LVK public samples for both events and find broad agreement for many intrinsic parameters.

Significance. If the method samples the true posterior, the result is practically significant: a laptop-scale, sub-minute-per-sample HMC for 12-dimensional BNS inference, compatible with Bilby, would be a useful tool for current and future GW analysis, especially as signals become longer. The paper's concrete timing measurements, the explicit three-phase pipeline, and the use of real LVK public data for two BNS events are strengths. The posterior medians for intrinsic parameters largely overlap the LVK public values, and the speedup claims are internally consistent. However, the central posterior-recovery claim is conditional on an implementation detail: the manuscript does not explicitly state that the Phase III Metropolis acceptance step uses the true Hamiltonian rather than the DNN shadow Hamiltonian. The validation comparisons are also not same-setup comparisons, so the evidence for unbiased posterior sampling is currently qualitative. These gaps are fixable but load-bearing.

major comments (3)
  1. [Sec. V.B (Phase III) and Sec. II.B] The Phase III specification, as written, does not state whether the Metropolis-Hastings acceptance at the end of each trajectory is evaluated with the true likelihood/true Hamiltonian or with the DNN-shadow Hamiltonian. The generic HMC algorithm in Sec. II.B includes a Metropolis step using H(q,p), and the earlier shadow-potential method described in Sec. I explicitly states that one 'moved back to the true potential' at the endpoint. In contrast, Sec. V.B only says to 'run N trajectories using the DNN approximation for the gradients.' If the endpoint Hamiltonian is not recomputed with the true likelihood, the chain samples the DNN shadow posterior, not the target posterior, and the R^2 scores in Figs. 10-11 do not bound the stationary distribution. This is the central correctness point of the paper, and it must be documented and verified explicitly, for example by stating the exact code path for the acceptance probability or by demonstrating with a same-setup reference sampler that the posterior is unbiased.
  2. [Sec. VI, Table I, Figs. 7-9] The validation against LVK public samples is not an apples-to-apples comparison. For GW170817, the LVK public samples fix the sky position using the electromagnetic counterpart and fix the time of coalescence, whereas DeepHMC samples these parameters freely; the LVK columns for alpha, delta, and tc in Table I are constants, not credible intervals from a free inference. For GW190425, Table I explicitly presents 'unimodal posterior distributions only,' and the sky maps in Fig. 9 show appreciably different localization areas between the LVK samples and DeepHMC. Consequently, the median overlap on intrinsic parameters is encouraging but does not establish that DeepHMC samples the same posterior as a reference sampler on the same data, waveform, and priors. I recommend adding a same-setup comparison—for example, running a standard sampler on the identical data, waveform, prior, and fixed versus free parameter choices—and reporting a quantitative distributional comparison for all sampled parameters, not only medians and intervals.
  3. [Sec. V.A and Appendix B, Fig. 11] The stopping threshold for the DNN is event-dependent (R^2 > 0.99 for GW170817 but R^2 > 0.9 for GW190425), and for GW190425 the tidal-parameter gradients only reach R^2 ≈ 0.90, with the text noting that the tidal parameters are consistently the slowest to converge. A 10% unexplained variance in exactly the tidal directions is not by itself a demonstration of unbiased sampling; whether it matters depends on whether the endpoint correction uses the true Hamiltonian and on the energy-error statistics of the accepted trajectories. The manuscript provides no diagnostic, such as a comparison of Phase III acceptance energy errors or a posterior-bias check against a reference sampler on an injection with known parameters, to show that the approximate-gradient leapfrog does not shift the stationary distribution. This should be added before the claimed posterior-recovery result is taken as established.
minor comments (5)
  1. [Throughout] The text contains several typos and grammatical slips: 'acclerated' in the title, 'produces produces' in the abstract, 'LIKEKIHOOD' in the Sec. V heading, 'strucutre' in Sec. V.B, and 'very 10^4 trajectories' in the Phase III bullet list.
  2. [Sec. II.C] In the sampler-space coordinate definition, the text writes 'M = n η^{3/5}' where the symbol n appears to be a typo for the total mass m; this should be corrected for clarity.
  3. [Sec. IV.A and Appendix A] The notation (μ/M)^{5/2} = η is used without immediately tying it to the standard symmetric mass ratio; a brief reminder near Eq. (42) would help the reader follow the Jacobian transformation.
  4. [Figs. 10 and 11] Several panels in the R^2 plots use poorly formatted scientific tick labels and different axis ranges, which makes the 'diagonal line' agreement harder to judge; reformatting the ticks and using a common scale across panels would improve readability.
  5. [Table I] The entries for alpha, delta, and tc for GW170817 are written as point values with zero-width intervals, which is misleading in a table of credible intervals; the caption should state explicitly that these were fixed in the LVK comparison.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central claims are benchmarked against external LVK data and measured timings, with only a minor self-citation lineage.

full rationale

DeepHMC's central claims are not forced by construction. The DNN gradients are trained on Phase I numerical gradients computed from the same event's likelihood, but the Phase III posteriors are compared against independent public LVK samples, so the agreement claim is not equivalent to the training input. The speedup factors (30x versus relative binning, 7000x versus naive gradients) are measured timings, not consequences of a definition. The reliance on the authors' prior HMC and shadow-potential work [29, 30] and parameterization papers [40, 41] is a lineage rather than a logical circle: the shadow-potential construction is re-derived in Sec. II, and the DNN replaces the cubic/OLUT fit with a new component that is trained and tested on the problem at hand. One caveat worth flagging, though it is not circularity: Sec. V.B's Phase III list only says to 'run N trajectories using the DNN approximation for the gradients' and does not restate the end-point true-Hamiltonian evaluation described for the earlier shadow-potential method in Sec. I. If that Metropolis correction were absent, the chain would target the DNN shadow posterior rather than the true posterior, making the LVK agreement a consistency check rather than a proof. That is an omitted implementation detail, not a case of an output being defined as its input. The R2-based stopping rules are internal tuning criteria on a test subset of Phase I data; they are not fitted outputs later relabeled as predictions. The posterior agreement with LVK public results provides an external benchmark, and no equation in the paper reduces the claimed result to its own inputs. Score 1 reflects the minor self-citation lineage with an otherwise self-contained external benchmark.

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

The core free parameters are the HMC step size and trajectory length, the mass-matrix scales, the DNN architecture, the training schedule, and the R^2 stopping thresholds, all chosen empirically. The key domain assumptions are stationary Gaussian noise, the accuracy of IMRPhenomD-NRTidal, and the unstated assumption that HMC with approximate gradients remains exact. The reflective boundaries and the regularized spin and mass priors are ad hoc smoothing choices.

free parameters (7)
  • HMC step size distribution = epsilon = N(5e-3, 1.5e-3), bounded within [1e-3, 1e-2]
    Empirically chosen to balance exploration and acceptance rate in Phase I and Phase III (Sec. II.C).
  • Leapfrog trajectory length l = l = 200 in Phase I; l ~ U(50, 400) in Phase III
    Tuned to avoid random-walk behavior and phase-space orbits (Sec. II.C, V.B).
  • Mass matrix scales s_kappa = From FIM initially, then from Phase I covariance matrix
    Scales depend on ad hoc rules for singular FIMs (1/12 of prior or half prior width) and on taking the minimum scales from non-spinning and spinning FIMs (Sec. II.C, III).
  • DNN R^2 stopping threshold = R2 >= 0.99 (unimodal) or 500 epochs; R2 >= 0.9 (multimodal) or 1000 epochs
    Threshold chosen per event based on observed convergence; lowered for GW190425 after failure to reach 0.95 or 0.99 (Sec. V.A.2).
  • DNN architecture = Hidden layers (10D, 10D, 100D), 25% dropout, linear first hidden layer, ReLU others
    Selected via RandomSearch plus manual testing; not derived from first principles (Sec. V.A.1).
  • Spin prior gradient regularization parameters = chi_tr = 1e-3, boundary at +/-0.99 chi_max, amplitude A from Eq. (47)
    Empirically chosen to smooth the singular log-zprior gradient near spin = 0 and the boundaries (Sec. IV.B.1).
  • Binomial series order for mass Jacobian = n = 30
    Empirically found to match the analytic Jacobian at low eta while smoothing the equal-mass singularity (Sec. IV.A).
assumptions (5)
  • domain assumption The detector noise is stationary and Gaussian.
    The likelihood in Eq. (4) assumes Gaussian noise, a standard GW analysis assumption stated in Sec. II.A.
  • domain assumption IMRPhenomD-NRTidal is an accurate waveform model for BNS signals with aligned spins and tidal effects.
    The waveform model is used throughout for likelihood and gradient evaluations; accuracy is cited from the model literature [44-47].
  • ad hoc to paper HMC using DNN-approximated gradients with Metropolis acceptance on the true Hamiltonian converges to the true posterior.
    Assumed in Phase III (Sec. V.B); no formal bias argument or same-setup reference sampler is provided. This is the main correctness risk.
  • domain assumption Phase I exploration, starting from the trigger (plus flipped sky positions for GW190425), adequately covers the posterior support needed for Phase III.
    The DNN is trained only on Phase I chain points; under-coverage would cause the surrogate to extrapolate in untested regions (Sec. VI.B).
  • ad hoc to paper Reflective boundaries at prior edges and at the equal-mass line preserve the stationary distribution.
    The infinite-energy-barrier reflection is imposed without a proof of detailed balance (Sec. II.E).

how reviews work

0 comments
Cite this review

Pith. "Pith review of DeepHMC : a deep-neural-network acclerated Hamiltonian Monte Carlo algorithm for binary neutron star parameter estimation." pith.science (2026). https://pith.science/paper/G4OH4ZSO

@misc{pith2026250502589,
  author       = {Pith},
  title        = {Pith review of: DeepHMC : a deep-neural-network acclerated Hamiltonian Monte Carlo algorithm for binary neutron star parameter estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/G4OH4ZSO}},
  note         = {Machine review of arXiv:2505.02589}
}
abstract

We present a deep neural network (DNN) accelerated Hamiltonian Monte Carlo (HMC) algorithm called DeepHMC for the inference of binary neutron star systems. The HMC is a non-random walk sampler that uses background gradient information to accelerate the convergence of the sampler. While faster converging than a random-walk sampler, in theory by a factor of the dimensionality of the problem, a known computational bottleneck for HMC algorithms is the calculation of gradients of the log-likelihood. We demonstrate that Hamiltonian trajectories based on a DNN gradients are 30 times faster than those based on the relative binning gradients, and 7000 times faster than trajectories based on a naive likelihood gradient calculation. Using the publicly available 128 second LVK data set for the binary neutron star mergers GW170817 and GW190425, we show that not only does DeepHMC produce produces highly accurate and consistent results with the LVK public data, but acquires 5000 statistically independent samples (SIS) in the $12D$ parameter space in approximately two hours on a Macbook pro for GW170817, with a cost of $<1$ second/SIS, and 2.5 days for GW190425, with a cost of $\sim25$ seconds/SIS.

Figures

Figures reproduced from arXiv: 2505.02589 by the authors.

Figure 1
Figure 1. FIG. 1: Phase space trajectory for the tidal parameter [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: A comparison between the analytic (blue-solid [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. FIG. 4: The DNN architecture for approximating the [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figures from the paper (7 more)
Figure 5
Figure 5. Figure 5: FIG. 5: Evolution of the coefficient of determination, [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Autocorrelation functions [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: A comparison between the LVK public data (blue) and [PITH_FULL_IMAGE:figures/full_fig_p016_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8: A comparison between the LVK public data (blue) and [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9: Sky maps for both the LVK and [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10: A comparison between gradients of the log-likelihood using numerical differencing and the DNN [PITH_FULL_IMAGE:figures/full_fig_p021_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11: A comparison between gradients of the log-likelihood using numerical differencing and the DNN [PITH_FULL_IMAGE:figures/full_fig_p022_11.png]

Discussion (0). Continue with ORCID 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. Detectability and Parameter Estimation for Einstein Telescope Configurations with GWJulia

    astro-ph.CO 2025-06 conditional novelty 6.0 of 10

    A new open-source Julia tool forecasts Einstein Telescope parameter-estimation accuracy, finding the 2L45 design marginally best for single parameters but comparable to other layouts when joint precision is required.

Reference graph

Works this paper leans on

72 extracted references · 26 canonical work pages · cited by 1 Pith paper

  1. [1]

    Draw the scaled momenta ˜pκ from a Gaussian dis- tributionN (0, 1),

    work page

  2. [2]

    Now, starting at the phase-space point (qκ i, ˜pκ i ) with Hamiltonian H(qκ i, ˜pκ i ), we evolve the Hamiltonian trajectory for l steps to a new phase-space coordi- nate (qκ f, ˜pκ f) with Hamiltonian H(qκ f, ˜pκ f),

    work page

  3. [3]

    one size fits all

    Accept or reject the new phase- space point with probability α = min 1,exp h −H(qκ f, ˜pκ f) +H(qκ i, ˜pκ i ) i . As we stated above, the development of a HMC algo- rithm requires a greater overhead cost as compared to other stochastic samplers. This is due to the free param- eters within the algorithm which are difficult to tune, and from the computation...

    work page

  4. [4]

    ∂ lnπ(χi) ∂χi = [χi ln (|χi/χmax|)]−1, (45) we arrive at an expression which causes multiple prob- lems for the HMC

    Gradients of the spin log-prior Taking the derivative of the spin log-prior, i.e. ∂ lnπ(χi) ∂χi = [χi ln (|χi/χmax|)]−1, (45) we arrive at an expression which causes multiple prob- lems for the HMC. We can see that the above expression contains singularities at χi =±χmax and χi = 0. The boundary singularities are easy to solve by again apply- ing a reflec...

    work page

  5. [5]

    (48) As we can see, the log-prior gradients involve gradients of the log-Jacobian determinant

    Gradients of the mass log-prior Starting with the logarithm of Eqn.(42), we can define the gradient of the log-prior for the masses as ∂ lnπ(lnM, lnµ) ∂ lnqµ = ∂ lnJ ∂ lnqµ. (48) As we can see, the log-prior gradients involve gradients of the log-Jacobian determinant. So, taking derivatives 10 of the logarithm of Eqn.(42), we obtain the expressions ∂ lnπ(...

    work page

  6. [6]

    To stack the dense layers of the network we use theSequential model within Keras

    Tuning the network structure At the beginning of the project, we used the Keras (v2.3.1) library [56], a widely used deep learning API run- ning on top of the machine learning platform TensorFlow (v1.14.0) [57], to build and train our DNN. To stack the dense layers of the network we use theSequential model within Keras. After some initial testing, we deci...

    work page

  7. [7]

    Tuning the network training After constructing the DNN, the next step was to op- timize the network training. At this step of the project, and mostly motivated by a change of hardward to ARM- based Apple silicon processors, we switched from Ten- sorFlow to PyTorch (v2.2.2) Fine tuning the weights and biases is done through an iterative training procedure ...

    work page

  8. [8]

    W. K. Hastings, Biometrika 57, 97 (1970)

    1970

Show all 72 references
  1. [9]

    Each core uses the same trigger point, but a different random seed for the momenta, thus ensuring diverging trajectories

    Phase I: run 3 300 trajectories in parallel using nu- merical gradients of fixed length with l = 200 steps and with a stepsizeϵ = 5×10−3. Each core uses the same trigger point, but a different random seed for the momenta, thus ensuring diverging trajectories. For every traject...

  2. [10]

    • Recompute the leapfrog scales using the co- variance matrix of the Phase I chain points

    Phase II: • Train the DNN for on{qκ, (∂κ lnL)num} from Phase I with a batchsize of 128, using the stopping criteria of R2 > 0.99 or 500 training epochs for unimodal sources in sky position, or R2 > 0.9 or 1000 training epochs for sources with a bimodal structure in sky positio...

  3. [11]

    Phase III: run N trajectories using the DNN ap- proximation for the gradients, with ϵ ∈ N[5× 10−3, 1.5× 10−3] (such that ϵκ = sκϵ) and l ∈ U(50, 400). • Recompute the leapfrog scales very 10 4 tra- jectories, stopping when there is less that 10% change in numerical values • Ev...

  4. [12]

    B. P. Abbott et al. (LIGO Scientific Collaboration, Virgo Collaboration), Phys. Rev. Lett. 116, 061102 (2016), 1602.03837. 16 2.0 2.5 3.0 JN 0.0 0.2 0.4 0.6 0.8 1.0 1.2Density 0 1 2 3 0.0 0.1 0.2 0.3 0.4 0.5 20 30 40 50 DL 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.0004 0.0006 ...

    2016 arXiv

  5. [13]

    B. P. Abbott et al. (LIGO Scientific Collaboration, Virgo Collaboration), Phys. Rev. X 9, 031040 (2019), 1811.12907

    2019 arXiv

  6. [14]

    Abbott et al

    R. Abbott et al. (LIGO Scientific, Virgo), Phys. Rev. X 11, 021053 (2021), 2010.14527

    2021 arXiv

  7. [15]

    Abbott et al

    R. Abbott et al. (LIGO Scientific, VIRGO, KAGRA) (2021), 2111.03606

    2021 arXiv

  8. [16]

    B. P. Abbott et al. (LIGO Scientific Collaboration, Virgo Collaboration), Phys. Rev. Lett. 119, 161101 (2017), 1710.05832. 17 0h 21h 18h 15h 12h 9h 6h 3h 0h 0° 30° 60°60° 30° 0° 30° 60° 60° 30° 50% area: 2262 deg2 90% area: 9277 deg2 50% 50% 50% 90% 90% 90% 90% 90% 90% 0h 21h ...

    2017 arXiv

  9. [17]

    B. P. Abbott et al. (LIGO Scientific, Virgo), Astrophys. J. Lett. 892, L3 (2020), 2001.01761

    2020 arXiv

  10. [18]

    Abbott et al

    R. Abbott et al. (LIGO Scientific, KAGRA, VIRGO), Astrophys. J. Lett. 915, L5 (2021), 2106.15163

    2021 arXiv

  11. [19]

    N. J. Cornish, Fast fisher matrices and lazy likelihoods (2013), 1007.4820, URL https://arxiv.org/abs/1007. 4820

    2013 arXiv

  12. [20]

    Metropolis, A

    N. Metropolis, A. W. Rosenbluth, M. N. Rosenbluth, A. H. Teller, and E. Teller, J. Chem. Phys. 21, 1087 (1953)

    1953

  13. [21]

    Skilling, Bayesian Anal

    J. Skilling, Bayesian Anal. 1, 833 (2006)

    2006

  14. [22]

    Ashton et al., Astrophys

    G. Ashton et al., Astrophys. J. Suppl. 241, 27 (2019), 1811.02042

    2019 arXiv

  15. [23]

    J. S. Speagle, Monthly Notices of the Royal Astronomical Society 493, 3132–3158 (2020), ISSN 1365-2966, URL http://dx.doi.org/10.1093/mnras/staa278

    2020 doi

  16. [24]

    Neal, in Handbook of Markov Chain Monte Carlo (CRC press, 2011), pp

    R. Neal, in Handbook of Markov Chain Monte Carlo (CRC press, 2011), pp. 113–162

    2011

  17. [25]

    Punturo, M

    M. Punturo, M. Abernathy, F. Acernese, B. Allen, N. An- dersson, K. Arun, F. Barone, B. Barr, M. Barsug- lia, M. Beker, et al., Classical and Quantum Gravity 27, 194002 (2010), URL https://dx.doi.org/10.1088/ 0264-9381/27/19/194002

    2010

  18. [26]

    A. Abac, R. Abramo, S. Albanesi, A. Albertini, A. Agapito, M. Agathos, C. Albertus, N. Andersson, T. Andrade, I. Andreoni, et al., The science of the ein- 18 stein telescope (2025), 2503.12263, URL https://arxiv. org/abs/2503.12263

    2025 arXiv

  19. [27]

    Reitze, R

    D. Reitze, R. X. Adhikari, S. Ballmer, B. Barish, L. Bar- sotti, G. Billingsley, D. A. Brown, Y. Chen, D. Coyne, R. Eisenstein, et al., Cosmic explorer: The u.s. contribu- tion to gravitational-wave astronomy beyond ligo (2019), 1907.04833, URL https://arxiv.org/abs/1907.04833

    2019 arXiv

  20. [28]

    Evans, A

    M. Evans, A. Corsi, C. Afle, A. Ananyeva, K. G. Arun, S. Ballmer, A. Bandopadhyay, L. Barsotti, M. Baryakhtar, E. Berger, et al., Cosmic explorer: A submission to the nsf mpsac nggw subcommittee (2023), 2306.13745, URL https://arxiv.org/abs/2306.13745

    2023 arXiv

  21. [29]

    Hu and J

    Q. Hu and J. Veitch, Costs of bayesian parameter estima- tion in third-generation gravitational wave detectors: a review of acceleration methods (2025), 2412.02651, URL https://arxiv.org/abs/2412.02651

    2025

  22. [30]

    Bouffanais and E

    Y. Bouffanais and E. K. Porter, Phys. Rev. D 100, 104023 (2019), 1810.07443

    2019 arXiv

  23. [31]

    Zackay, L

    B. Zackay, L. Dai, and T. Venumadhav, arXiv e-prints arXiv:1806.08792 (2018), 1806.08792

    2018 arXiv

  24. [32]

    Morisaki, Physical Review D 104 (2021), ISSN 2470- 0029, URL http://dx.doi.org/10.1103/PhysRevD

    S. Morisaki, Physical Review D 104 (2021), ISSN 2470- 0029, URL http://dx.doi.org/10.1103/PhysRevD. 104.044062

    2021 doi

  25. [33]

    Canizares, S

    P. Canizares, S. E. Field, J. Gair, V. Raymond, R. Smith, and M. Tiglio, Physical Review Letters 114 (2015), ISSN 1079-7114, URL http://dx.doi.org/10. 1103/PhysRevLett.114.071104

    2015

  26. [34]

    Smith, S

    R. Smith, S. E. Field, K. Blackburn, C.-J. Haster, M. P¨ urrer, V. Raymond, and P. Schmidt, Physical Re- view D 94 (2016), ISSN 2470-0029, URL http://dx. doi.org/10.1103/PhysRevD.94.044031

    2016 doi

  27. [36]

    Morisaki, R

    S. Morisaki, R. Smith, L. Tsukada, S. Sachdev, S. Steven- son, C. Talbot, and A. Zimmerman, Rapid localization and inference on compact binary coalescences with the advanced ligo-virgo-kagra gravitational-wave detector net- work (2023), 2307.13380, URL https://arxiv.org/abs/ 2...

    2023 arXiv

  28. [37]

    Roulet, J

    J. Roulet, J. Mushkin, D. Wadekar, T. Venumadhav, B. Zackay, and M. Zaldarriaga, Physical Review D 110 (2024), ISSN 2470-0029, URL http://dx.doi.org/10. 1103/PhysRevD.110.044010

    2024

  29. [38]

    M. J. Williams, J. Veitch, and C. Messenger, Physical Review D 103 (2021), ISSN 2470-0029, URL http:// dx.doi.org/10.1103/PhysRevD.103.103006

    2021 doi

  30. [39]

    M. Dax, S. R. Green, J. Gair, N. Gupte, M. P¨ urrer, V. Raymond, J. Wildberger, J. H. Macke, A. Buo- nanno, and B. Sch¨ olkopf, Nature 639, 49–53 (2025), ISSN 1476-4687, URL http://dx.doi.org/10.1038/ s41586-025-08593-z

    2025

  31. [40]

    E. K. Porter and J. Carr´ e, Class. Quant. Grav. 31, 145004 (2014), 1311.7539

    2014 arXiv

  32. [41]

    N. J. Cornish and E. K. Porter, Class. Quant. Grav. 24, 5729 (2007), gr-qc/0612091

    2007 arXiv

  33. [42]

    A. H. Nitz, Robust, rapid, and simple gravitational-wave parameter estimation (2024), 2410.05190, URL https: //arxiv.org/abs/2410.05190

    2024 arXiv

  34. [43]

    K. W. K. Wong, M. Isi, and T. D. P. Edwards, Fast gravitational wave parameter estimation without com- promises (2023), 2302.05333, URL https://arxiv.org/ abs/2302.05333

    2023 arXiv

  35. [44]

    Duane, A

    S. Duane, A. D. Kennedy, B. Pendleton, and D. Roweth, Phys. Lett. B. 195, 216 (1987)

    1987

  36. [45]

    Hajian, Phys

    A. Hajian, Phys. Rev. D 75, 083525 (2007), astro- ph/0608679

    2007

  37. [46]

    Buonanno, B

    A. Buonanno, B. R. Iyer, E. Ochsner, Y. Pan, and B. S. Sathyaprakash, Physical Review D 80 (2009)

    2009

  38. [47]

    GWOSC (2017), https://www.gw-openscience.org/ eventapi/html/GWTC-1-confident/GW170817/v3

    2017

  39. [48]

    GWOSC (2019), https://gwosc.org/eventapi/html/ GWTC-2.1-confident/GW190425/v3/

    2019

  40. [49]

    L. S. Finn, Phys. Rev. D 46, 5236 (1992), gr-qc/9209010

    1992 arXiv

  41. [50]

    B. J. Owen, Phys. Rev. D 53, 6749 (1996), gr- qc/9511032

    1996

  42. [51]

    N. J. Cornish and E. K. Porter, Class. Quant. Grav. 23, S761 (2006), gr-qc/0605085

    2006 arXiv

  43. [52]

    E. E. Flanagan and T. Hinderer, Phys. Rev. D77, 021502 (2008), 0709.1915

    2008 arXiv

  44. [53]

    Sokal, in Functional Integration, edited by C

    A. Sokal, in Functional Integration, edited by C. DeWitt- Morette, P. Cartier, and A. Folacci (Springer US, Boston, MA, 1997), vol. 361, pp. 131–192, ISBN 978-1-4899-0321- 1 978-1-4899-0319-8

    1997

  45. [54]

    Madras and A

    N. Madras and A. D. Sokal, J. Statist. Phys. 50, 109 (1988)

    1988

  46. [55]

    Dietrich, S

    T. Dietrich, S. Bernuzzi, and W. Tichy, Phys. Rev. D 96, 121501 (2017), 1706.02969

    2017 arXiv

  47. [56]

    Dietrich, A

    T. Dietrich, A. Samajdar, S. Khan, N. K. Johnson- McDaniel, R. Dudi, and W. Tichy, Phys. Rev. D 100, 044003 (2019), 1905.06011

    2019 arXiv

  48. [57]

    S. Husa, S. Khan, M. Hannam, M. P¨ urrer, F. Ohme, X. J. Forteza, and A. Boh´ e, Phys. Rev. D93, 044006 (2016)

    2016

  49. [58]

    S. Khan, S. Husa, M. Hannam, F. Ohme, M. P¨ urrer, X. Jim´ enez Forteza, and A. Boh´ e, Phys. Rev. D 93, 044007 (2016), 1508.07253

    2016 arXiv

  50. [59]

    T. A. Apostolatos, C. Cutler, G. J. Sussman, and K. S. Thorne, Phys. Rev. D 49, 6274 (1994)

    1994

  51. [60]

    Ajith et al., Phys

    P. Ajith et al., Phys. Rev. Lett. 106, 241101 (2011), 0909.2867

    2011 arXiv

  52. [61]

    Santamaria et al., Phys

    L. Santamaria et al., Phys. Rev. D 82, 064016 (2010), 1005.3306

    2010 arXiv

  53. [62]

    Damour, Phys

    T. Damour, Phys. Rev. D 64, 124013 (2001), gr- qc/0103018

    2001

  54. [64]

    L. Wade, J. D. E. Creighton, E. Ochsner, B. D. Lackey, B. F. Farr, T. B. Littenberg, and V. Raymond, Phys. Rev. D 89, 103012 (2014), 1402.5156

    2014 arXiv

  55. [65]

    Favata, Phys

    M. Favata, Phys. Rev. Lett. 112, 101101 (2014), 1310.8288

    2014 arXiv

  56. [66]

    Lange, R

    J. Lange, R. O’Shaughnessy, and M. Rizzo (2018), 1805.10457

    2018 arXiv

  57. [67]

    Chollet et al., Keras, https://keras.io (2015)

    F. Chollet et al., Keras, https://keras.io (2015)

    2015

  58. [68]

    Abadi, A

    M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G. S. Corrado, A. Davis, J. Dean, M. Devin, et al., TensorFlow: Large-scale machine learning on het- erogeneous systems (2015), software available from ten- sorflow.org, URL https://www.tensorflow.org/

    2015

  59. [69]

    Srivastava, G

    N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, Journal of Machine Learning Re- search 15, 1929 (2014), ISSN 1533-7928, URL http: //jmlr.org/papers/v15/srivastava14a.html

    2014

  60. [70]

    A. Geron, Hands-On Machine Learning with Scikit- Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems (O’Reilly Media, 19 Inc., 2019), 2nd ed., ISBN 1492032646

    2019

  61. [71]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, et al., Journal of Machine Learn- ing Research 12, 2825 (2011)

    2011

  62. [72]

    D. P. Kingma and J. Ba, arXiv e-prints arXiv:1412.6980 (2014), 1412.6980

    2014 arXiv

  63. [73]

    Masters and C

    D. Masters and C. Luschi, arXiv e-prints arXiv:1804.07612 (2018), 1804.07612. Appendix A: Deriving the log-prior gradients with respect to (lnM, lnµ) The LVK analysis assumes a flat prior in the individual masses, i.e. π(m1,m 2) = π(m1)π(m2)∝ 1. To transform for the joint prio...

    2018 arXiv

  64. [200]

    This is a factor of 30 times faster than the relative binning gradient based trajec- tories, and a factor of 7000 times faster than a naive calculation

    By then replacing the relative binning likelihood gradients with DNN gradients, a single trajectory took 0.06 seconds to generate. This is a factor of 30 times faster than the relative binning gradient based trajec- tories, and a factor of 7000 times faster than a naive calcul...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.