pith. machine review for the scientific record. sign in

arxiv: 2604.22021 · v1 · submitted 2026-04-23 · 🌀 gr-qc · astro-ph.IM

Recognition: unknown

Data-Driven Acceleration of Eccentricity Reduction for Binary Black Hole Simulations

Vittoria Tommasini , Nils L. Vu , Mark A. Scheel , Saul A. Teukolsky
Authors on Pith no claims yet

Pith reviewed 2026-05-09 20:29 UTC · model grok-4.3

classification 🌀 gr-qc astro-ph.IM
keywords numerical relativitybinary black holeseccentricity reductionGaussian process regressiondata-driven accelerationgravitational wave modelinginitial data preparation
0
0 comments X

The pith

A Gaussian process model trained on prior simulations predicts initial conditions that produce low-eccentricity binary black hole orbits with zero or one tuning iteration.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper establishes that a Gaussian Process Regression model, trained on an archive of eccentricity-reduced binary black hole simulations, can predict the initial orbital frequency and radial velocity needed to start the binary with very low eccentricity. Standard methods require four or more iterative trial simulations to reach the target, each adding substantial time to already lengthy runs. The model reduces this to zero or one iteration for all tested cases, cutting the setup cost. Sympathetic readers care because these low-eccentricity starts are necessary for creating reliable gravitational wave models that match the near-circular orbits expected in nature.

Core claim

The central claim is that the Gaussian Process Regression model trained on the archive of previously eccentricity-reduced numerical relativity simulations learns to output the correct Omega_0 and adot_0 for new configurations, resulting in evolutions with small eccentricity and thereby requiring only zero or one eccentricity reduction iteration instead of the usual four or more.

What carries the argument

The Gaussian Process Regression model that takes binary parameters as input and outputs the initial orbital frequency Omega_0 and radial velocity adot_0, trained to minimize the eccentricity in the subsequent evolution.

If this is right

  • Binary black hole simulations reach the required low eccentricity threshold with at most one additional iteration after using the model's prediction.
  • The overall computational expense of preparing initial data for numerical relativity runs decreases significantly.
  • Researchers can perform more simulations or higher-resolution ones with the same resources.
  • The method provides a practical alternative to post-Newtonian based initial guesses.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Expanding the training archive with more varied binary parameters could enable the model to cover a wider parameter space reliably.
  • Analogous machine learning techniques might speed up other costly iterative processes in numerical relativity simulations.
  • Public release of such trained models could allow the community to bootstrap new simulations more efficiently.

Load-bearing premise

The trained Gaussian Process Regression model accurately predicts suitable initial conditions for binary black hole configurations not seen during training.

What would settle it

Apply the model to predict initial conditions for a binary black hole with parameter values outside the training set, perform the simulation, and measure whether the resulting eccentricity is below the target after zero or one reduction iteration.

Figures

Figures reproduced from arXiv: 2604.22021 by Mark A. Scheel, Nils L. Vu, Saul A. Teukolsky, Vittoria Tommasini.

Figure 1
Figure 1. Figure 1: FIG. 1. Current eccentricity reduction schemes consist of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. GPR and higher-order PN (HOPN) predictions for [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. Comparison of GPR-predicted corrections and high-order PN (HOPN) corrections for aligned-spin binaries as functions [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Training dataset used for our four-dimensional GPR model. The parameters shown are the initial separation, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. Leave-one-out (LOO) residuals for the GPR predicted [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6. Training dataset used for our 8D GPR model. Shown are initial separation, [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7. Leave-one-out residuals for the GPR predicted cor [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8. Eccentricity as a function of iteration number for nine [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9. Trajectories of the eccentricity-reduction procedure in the (Ω [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10. Mass ratio extrapolation tests of the GPR model. [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
read the original abstract

Reducing orbital eccentricity in numerical relativity simulations of binary black holes is essential for producing astrophysically relevant gravitational wave models, as many of these systems are expected to be near-circular in nature. Standard eccentricity reduction procedures rely on iterative schemes, often requiring four or more trial simulations to achieve desired thresholds. This approach is computationally expensive because each trial simulation adds ~10% to the total simulation run time of multiple weeks to months. We introduce a data-driven approach that accelerates this process by learning the values of the initial orbital frequency, Omega_0, and radial velocity, adot_0, that yield an evolution with small eccentricity. This is done using a Gaussian Process Regression model trained on an archive of previously eccentricity-reduced numerical relativity simulations. For all configurations tested, using the trained model consistently reduces the number of required eccentricity reduction iterations to just zero or one, significantly lowering computational costs relative to post-Newtonian initial guesses. These results demonstrate the power of data-driven methods in accelerating expensive numerical relativity simulations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper claims that a Gaussian Process Regression model, trained on an archive of prior eccentricity-reduced binary black hole numerical relativity simulations, can predict initial orbital frequency Ω₀ and radial velocity ȧ₀ for new configurations such that the standard iterative eccentricity reduction procedure requires only zero or one iteration instead of the usual four or more, thereby lowering computational costs relative to post-Newtonian initial data.

Significance. If the central claim is substantiated with quantitative validation, the approach could meaningfully accelerate production of large NR simulation catalogs for gravitational-wave modeling by reducing the overhead of eccentricity tuning. The data-driven framing is a strength, but the absence of reported metrics, training-set statistics, and generalization tests prevents a clear assessment of practical impact or reliability beyond the specific tested cases.

major comments (2)
  1. [Abstract] Abstract: the central claim that the model 'consistently reduces the number of required eccentricity reduction iterations to just zero or one' for all tested configurations is unsupported by any quantitative evidence (achieved eccentricity, prediction error, training-set size, cross-validation scores, or comparison to PN baselines). This information is load-bearing for the main result.
  2. [Methods/Results] The manuscript provides no details on the training archive (size, coverage in mass ratio, spins, or other parameters) or whether the tested configurations are held-out or lie within the training distribution. Without this, the generalization performance of the GPR model cannot be evaluated, directly undermining the claim that the method works for arbitrary new runs.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the opportunity to clarify our work. We agree that strengthening the quantitative support in the abstract and expanding the description of the training data will improve the manuscript. We address each major comment below and will incorporate revisions accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the model 'consistently reduces the number of required eccentricity reduction iterations to just zero or one' for all tested configurations is unsupported by any quantitative evidence (achieved eccentricity, prediction error, training-set size, cross-validation scores, or comparison to PN baselines). This information is load-bearing for the main result.

    Authors: We agree that the abstract would be strengthened by including key quantitative indicators. In the revised version we will update the abstract to reference the specific results from the body of the paper, including the achieved eccentricities for the tested configurations, the observed reduction in iteration count relative to post-Newtonian initial data, and any reported measures of model performance such as prediction accuracy on the held-out cases. revision: yes

  2. Referee: [Methods/Results] The manuscript provides no details on the training archive (size, coverage in mass ratio, spins, or other parameters) or whether the tested configurations are held-out or lie within the training distribution. Without this, the generalization performance of the GPR model cannot be evaluated, directly undermining the claim that the method works for arbitrary new runs.

    Authors: We acknowledge the omission of these details in the current draft. We will expand the Methods section to describe the training archive (its size, the ranges of mass ratio, spin magnitudes, and other parameters covered) and to state explicitly that the configurations used for validation were held out from the training set. This will allow readers to assess the scope of generalization directly. revision: yes

Circularity Check

0 steps flagged

No circularity: GPR predictions are external to each target simulation

full rationale

The paper trains a Gaussian Process Regression model on an external archive of prior eccentricity-reduced simulations, then uses the model to supply Omega_0 and adot_0 for new target configurations. The claimed reduction from four-plus iterations to zero or one is an empirical outcome of applying the trained model; it is not obtained by re-using a fitted parameter of the target run itself, nor by any self-referential definition or self-citation chain that collapses the result to the input. The training archive is independent of each new simulation, satisfying the requirement that the derivation remain self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

The central claim rests on the empirical performance of a Gaussian Process model whose hyperparameters are fitted to the training archive; no additional physical axioms or invented entities are introduced.

free parameters (1)
  • Gaussian Process kernel hyperparameters
    Length scales, variance, and noise parameters of the GPR kernel are fitted to the archive of prior eccentricity-reduced simulations.

pith-pipeline@v0.9.0 · 5484 in / 1319 out tokens · 30223 ms · 2026-05-09T20:29:21.081566+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

63 extracted references · 29 canonical work pages · 7 internal anchors

  1. [1]

    Kernel and Mean Function Choices GPRs can be constructed with many possible choices of kernels and mean functions. To capture both smooth, global trends and more localized structure, we adopt a mixed kernel formed by a weighted sum of a squared- exponential RBF kernel and a Matern kernel, k(x,x ′) =α 1kRBF(x,x ′) +α 2kMatern(x,x ′) (1) Both kernels employ...

  2. [2]

    The package provides the complete pipeline for data normal- ization, training, prediction, cross-validation, plotting, and analysis

    Implementation Our GPR workflow is implemented in a Python pack- age developed for this work and released as open source within the SXSSimulationSupportrepository [50]. The package provides the complete pipeline for data normal- ization, training, prediction, cross-validation, plotting, and analysis. Training and inference are performed us- ingPyT orch[51...

  3. [3]

    Model Construction and Training Strategy We train two independent GPR models: one to predict corrections to the initial orbital frequency, Ω 0, and an- other to predict corrections to the radial expansion rate, ˙a0. This choice allows each parameter to be modeled with its own characteristic scale and smoothness, and avoids introducing additional hyperpara...

  4. [4]

    Our workflow proceeds as follows:

    Validation and Application At each stage, we assess model performance using leave-one-out (LOO) cross-validation, in which each sim- ulation is withheld in turn from the training set and pre- dicted by a model trained on the remaining data. Our workflow proceeds as follows:

  5. [5]

    For each simulation in the training set, we extract the intrinsic binary parameters,q,S 1,S 2, and the initial separation,D 0, together with the PN initial guesses for the orbital frequency, Ω PN 0 , and radial expansion rate, ˙aPN 0

  6. [6]

    Using the final, low-eccentricity orbital parameters obtained after the SpEC eccentricity-reduction pro- cedure, we define the training data as residual cor- rections to the PN predictions, ∆Ω0 = ΩNR 0 −Ω PN 0 (2) ∆˙a0 = ˙aNR 0 −˙aPN 0 (3) We always use the LOPN approximation for Ω PN 0 and ˙aPN 0 in this work, but any smooth reference function could be used

  7. [7]

    We train two independent GPR models to predict ∆Ω0 and ∆˙a0 as smooth functions of the input pa- rameters, assessing the strength of our model and its ability to generalize using LOO cross validation

  8. [8]

    For a new target configuration, the trained GPR models predict corrections to the PN initial guesses, yielding corrected orbital parameters, Ω0 = ΩPN 0 + ∆ΩGPR 0 (4) ˙a0 = ˙aPN 0 + ∆˙aGPR 0 (5) These corrected parameters are used to construct NR initial data. III. RESUL TS A. Equal Mass, Non-spinning Binaries We begin by constructing a deliberately simpli...

    2019

  9. [9]

    Advanced LIGO

    LIGO Scientific Collaboration, J. Aasi,et al., Clas- sical and Quantum Gravity32, 074001 (2015), arXiv:1411.4547 [gr-qc]

    work page internal anchor Pith review arXiv 2015

  10. [10]

    Advanced Virgo: a 2nd generation interferometric gravitational wave detector

    F. Acerneseet al., Classical and Quantum Gravity32, 024001 (2015), arXiv:1408.3978 [gr-qc]

    work page internal anchor Pith review arXiv 2015

  11. [11]

    B. P. Abbott, R. Abbott, T. D. Abbott,et al., Phys. Rev. Lett.116, 061102 (2016), arXiv:1602.03837 [gr-qc]

    work page internal anchor Pith review arXiv 2016

  12. [12]

    L. C. W. W. Group, Living Reviews in Relativity28, 9 (2025), arXiv:2311.01300 [gr-qc]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    Centrella, J

    J. Centrella, J. G. Baker, B. J. Kelly, and J. R. van Meter, Reviews of Modern Physics82, 3069 (2010), arXiv:1010.5260 [gr-qc]

    work page arXiv 2010

  14. [14]

    Centrella, J

    J. Centrella, J. G. Baker, B. J. Kelly, and J. R. van Meter, Annual Review of Nuclear and Particle Science 60, 75 (2010), arXiv:1010.2165 [gr-qc]

    work page arXiv 2010

  15. [15]

    P. C. Peters,Gravitational radiation and the motion of two point masses, Ph.D. thesis, California Institute of Technology (1964)

    1964

  16. [16]

    I. M. Romero-Shaw, P. D. Lasky, and E. Thrane, Mon. Not. R. Astron. Soc.490, 5210 (2019), arXiv:1909.05466 [astro-ph.HE]

    work page arXiv 2019

  17. [17]

    H. P. Pfeiffer, D. A. Brown, L. E. Kidder, L. Lindblom, G. Lovelace, and M. A. Scheel, Class. Quantum Grav. 24, S59 (2007)

    2007

  18. [18]

    Buonanno, L

    A. Buonanno, L. E. Kidder, A. H. Mrou´ e, H. P. Pfeiffer, and A. Taracchini, Phys. Rev. D83, 104034 (2011)

    2011

  19. [19]

    Ramos-Buades, S

    A. Ramos-Buades, S. Husa, and G. Pratten, Phys. Rev. D99, 023003 (2019)

    2019

  20. [20]

    M. A. Scheel, M. Boyle, K. Mitman, N. Deppe, L. C. Stein, C. Armaza, M. S. Bonilla, L. T. Buchman, A. Ceja, H. Chaudhary, Y. Chen, M. Corman, K. Zolt´ an Csuk´ as, C. Melize Ferrus, S. E. Field, M. Giesler, S. Habib, F. H´ ebert, D. A. Hemberger, D. A. B. Iozzo, T. Islam, K. Z. Jones, A. Khairnar, L. E. Kidder, T. Knapp, P. Ku- 16 mar, G. Lara, O. Long, G...

    work page arXiv 2025

  21. [21]

    E. K. Porter and A. Sesana, arXiv: General Relativity and Quantum Cosmology (2010)

    2010

  22. [22]

    J. S. Key and N. J. Cornish, Phys. Rev. D83, 083001 (2011)

    2011

  23. [23]

    Evidence for eccentricity in the population of binary black holes observed by LIGO-Virgo-KAGRA

    N. Gupte, A. Ramos-Buades, A. Buonanno, J. Gair, M. Coleman Miller, M. Dax, S. R. Green, M. P¨ urrer, J. Wildberger, J. Macke, I. M. Romero-Shaw, and B. Sch¨ olkopf, Phys. Rev. D112, 104045 (2025), arXiv:2404.14286 [gr-qc]

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  24. [24]

    P. J. Nee, A. Gamboa, H. P. Pfeiffer, L. Pompili, A. Ramos-Buades, V. Varma, M. Boyle, A. Buo- nanno, R. Enficiaud, L. E. Kidder, and M. A. Scheel, Classical and Quantum Gravity42, 135011 (2025), arXiv:2503.05422 [gr-qc]

    work page arXiv 2025

  25. [25]

    Saini, M

    P. Saini, M. Favata, and K. G. Arun, Phys. Rev. D106, 084031 (2022), arXiv:2203.04634 [gr-qc]

    work page arXiv 2022

  26. [26]

    Wang, Y.-C

    H. Wang, Y.-C. Zou, Q.-W. Wu, Y. Liu, and X. Liu, Phys. Rev. D109, 084063 (2024), arXiv:2310.04777 [gr- qc]

    work page arXiv 2024

  27. [27]

    Gamboa, A

    A. Gamboa, A. Buonanno, R. Enficiaud, M. Khalil, A. Ramos-Buades, L. Pompili, H. Estell´ es, M. Boyle, L. E. Kidder, H. P. Pfeiffer, H. R. R¨ uter, and M. A. Scheel, Phys. Rev. D112, 044038 (2025), arXiv:2412.12823 [gr-qc]

    work page arXiv 2025

  28. [28]

    Varma, M

    V. Varma, M. A. Scheel, and H. P. Pfeiffer, Phys. Rev. D98, 104011 (2018)

    2018

  29. [29]

    Zhang and B

    F. Zhang and B. Szil´ agyi, Phys. Rev. D88, 084033 (2013)

    2013

  30. [30]

    Habib, M

    S. Habib, M. A. Scheel, and S. A. Teukolsky, Phys. Rev. D111, 084059 (2025)

    2025

  31. [31]

    Healy, C

    J. Healy, C. O. Lousto, H. Nakano, and Y. Zlochower, Classical and Quantum Gravity34, 145011 (2017)

    2017

  32. [32]

    L. T. Buchman, H. P. Pfeiffer, M. A. Scheel, and B. Szil´ agyi, Phys. Rev. D86, 084033 (2012)

    2012

  33. [33]

    Habib, A

    S. Habib, A. Ramos-Buades, E. A. Huerta, S. Husa, R. Haas, and Z. Etienne, Classical and Quantum Gravity 38, 125007 (2021), arXiv:2011.08878 [gr-qc]

    work page arXiv 2021

  34. [34]

    P¨ urrer, S

    M. P¨ urrer, S. Husa, and M. Hannam, Phys. Rev. D85, 124051 (2012)

    2012

  35. [35]

    A. H. Mrou´ e, H. P. Pfeiffer, L. E. Kidder, and S. A. Teukolsky, Phys. Rev. D82, 124016 (2010)

    2010

  36. [36]

    M. A. Shaikh, V. Varma, A. Ramos-Buades, H. P. Pfeiffer, M. Boyle, L. E. Kidder, and M. A. Scheel, Classical and Quantum Gravity42, 195012 (2025), arXiv:2507.08345 [gr-qc]

    work page arXiv 2025

  37. [37]

    Ramos-Buades, M

    A. Ramos-Buades, M. van de Meent, H. P. Pfeiffer, H. R. R¨ uter, M. A. Scheel, M. Boyle, and L. E. Kidder, Phys. Rev. D106, 124040 (2022), arXiv:2209.03390 [gr-qc]

    work page arXiv 2022

  38. [38]

    Loutrel, S

    N. Loutrel, S. Liebersbach, N. Yunes, and N. Cornish, Classical and Quantum Gravity36, 025004 (2018)

    2018

  39. [39]

    Islam and T

    T. Islam and T. Venumadhav, Phys. Rev. D112, 104039 (2025)

    2025

  40. [40]

    Ramos-Buades, M

    A. Ramos-Buades, M. van de Meent, H. P. Pfeiffer, H. R. R¨ uter, M. A. Scheel, M. Boyle, and L. E. Kidder, Phys. Rev. D106, 124040 (2022)

    2022

  41. [41]

    Bonino, P

    A. Bonino, P. Schmidt, and G. Pratten, Phys. Rev. D 110, 104002 (2024)

    2024

  42. [42]

    C. E. Rasmussen and C. K. I. Williams,Gaussian Pro- cesses for Machine Learning(MIT Press, Cambridge, MA, USA, 2006) available online

    2006

  43. [43]

    Dieselhorst, W

    T. Dieselhorst, W. Cook, S. Bernuzzi, and D. Radice, Symmetry13, 2157 (2021)

    2021

  44. [44]

    Helfer, T

    T. Helfer, T. D. P. Edwards, J. Dafflon, K. W. K. Wong, and M. L. Olson, Super-resolution without high- resolution labels for black hole simulations (2024), arXiv:2411.02453 [gr-qc]

    work page arXiv 2024

  45. [45]

    I. B. Mendes, N. L. Vu, O. Long, H. P. Pfeiffer, and R. Owen, Phys. Rev. D112, 124049 (2025), arXiv:2509.07291 [gr-qc]

    work page arXiv 2025

  46. [46]

    black-holes.org/for-researchers/spec(2025), accessed: 2026-01-18

    Spec: Spectral einstein code,https://www. black-holes.org/for-researchers/spec(2025), accessed: 2026-01-18

    2025

  47. [47]

    Deppe, W

    N. Deppe, W. Throwe, L. E. Kidder, N. L. Vu, K. C. Nelli, C. Armaza, M. S. Bonilla, F. H´ ebert, Y. Kim, P. Kumar, G. Lovelace, A. Macedo, J. Moxon, E. O’Shea, H. P. Pfeiffer, M. A. Scheel, S. A. Teukolsky, N. A. Wit- tek, I. Anantpurkar, and J. Yoo, SpECTRE (2025)

    2025

  48. [48]

    J. W. York, Jr., Phys. Rev. Lett.82, 1350 (1999), arXiv:gr-qc/9810051 [gr-qc]

    work page arXiv 1999

  49. [49]

    H. P. Pfeiffer and J. W. York, Phys. Rev. D67, 044022 (2003), arXiv:gr-qc/0207095 [gr-qc]

    work page arXiv 2003

  50. [50]

    G. B. Cook and H. P. Pfeiffer, Phys. Rev. D70, 104016 (2004), arXiv:gr-qc/0407078 [gr-qc]

    work page arXiv 2004

  51. [51]

    H. P. Pfeiffer, arXiv e-prints , gr-qc/0412002 (2004), arXiv:gr-qc/0412002 [gr-qc]

    work page internal anchor Pith review arXiv 2004

  52. [52]

    T. W. Baumgarte and S. L. Shapiro,Numerical Relativ- ity: Solving Einstein ’s Equations on the Computer(Cam- bridge University Press, 2010)

    2010

  53. [53]

    N. L. Vu, H. P. Pfeiffer, G. S. Bonilla, N. Deppe, F. H´ ebert, L. E. Kidder, G. Lovelace, J. Moxon, M. A. Scheel, S. A. Teukolsky, W. Throwe, N. A. Wittek, and T. W lodarczyk, Phys. Rev. D105, 084027 (2022), arXiv:2111.06767 [gr-qc]

    work page arXiv 2022

  54. [54]

    Lovelace, R

    G. Lovelace, R. Owen, H. P. Pfeiffer, and T. Chu, Phys. Rev. D78, 084017 (2008), arXiv:0805.4192 [gr-qc]

    work page arXiv 2008

  55. [55]

    L. E. Kidder, Phys. Rev. D52, 821 (1995), arXiv:gr- qc/9506022 [gr-qc]

    work page arXiv 1995

  56. [56]

    Next-to-next-to-leading order spin-orbit effects in the near-zone metric and precession equations of compact binaries

    A. Boh´ e, S. Marsat, G. Faye, and L. Blanchet, Classical and Quantum Gravity30, 075017 (2013), arXiv:1212.5520 [gr-qc]

    work page Pith review arXiv 2013

  57. [57]

    Boyle, PostNewtonian.jl (2024)

    M. Boyle, PostNewtonian.jl (2024)

    2024

  58. [58]

    N. Vu, V. Tommasini, M. Scheel, and S. Teukolsky, Sim- ulationSupport (2024)

    2024

  59. [59]

    PyTorch: An Imperative Style, High-Performance Deep Learning Library

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. K¨ opf, E. Yang, Z. DeVito, M. Rai- son, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, arXiv e-prints , arXiv:1912.01703 (2019), arXiv:1912.01703 [cs.LG]

    work page internal anchor Pith review Pith/arXiv arXiv 1912

  60. [60]

    J. R. Gardner, G. Pleiss, D. Bindel, K. Q. Weinberger, and A. G. Wilson, arXiv e-prints , arXiv:1809.11165 (2018), arXiv:1809.11165 [cs.LG]

    work page arXiv 2018

  61. [61]

    S. Ma, M. Giesler, M. A. Scheel, and V. Varma, Phys. Rev. D103, 084029 (2021), arXiv:2102.06618 [gr-qc]

    work page arXiv 2021

  62. [62]

    H. P. Pfeiffer, G. B. Cook, and S. A. Teukolsky, Phys. Rev. D66, 024047 (2002)

    2002

  63. [63]

    A. M. Knee, I. M. Romero-Shaw, P. D. Lasky, J. McIver, and E. Thrane, The Astrophysical Journal936, 172 17 (2022). [56] S. Husa, M. Hannam, J. A. Gonz´ alez, U. Sperhake, and B. Br¨ ugmann, Phys. Rev. D77, 044037 (2008)

    2022