Pith. sign in

REVIEW 3 major objections 4 minor 20 references

Solving Einstein's Vacuum Equations with Physics-Informed Neural Networks: Boundary Conditions and Domain Decomposition

T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read This paper shows that physics-informed neural networks can solve Einstein's vacuum field equations for a static, axisymmetric spacetime, reproducing the q-metric with relative errors below 10^-4 even when boundary conditions are imposed…

desk verdict Solid PINN benchmark for Schwarzschild and the q-metric, but the q-metric ansatz embeds the exact solution, so the paper's forward-looking claim about unknown spacetimes is not supported. read the letter →

arxiv 2608.08846 v1 pith:CQ7QWINP submitted 2026-08-09 gr-qc

Elly Bayona , Hernando Quevedo This is my paper · ORCID
classification gr-qc PACS 04.20.-q
keywords physics-informedneuralnetworksEinsteinvacuumequationsq-metricSchwarzschildspacetimedomaindecompositionboundaryconditionsquadrupolemomentKretschmannscalar
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks whether a physics-informed neural network, trained by minimizing the residuals of the governing differential equations rather than from data, can solve Einstein's vacuum field equations for a genuinely axisymmetric spacetime. Its central demonstration is that such a network reconstructs the q-metric, the simplest static vacuum solution with a quadrupole moment, with relative errors between $10^{-6}$ and $10^{-4}$ for the four metric functions. The comparison that carries the paper is between strict boundary conditions, imposed on all metric functions at the outer edge, and relaxed boundary conditions, imposed only on the outermost radial subdomain while the inner subdomains are trained from Einstein residuals alone. The authors conclude that this combination of boundary treatment and sequential domain decomposition is a workable route toward spacetimes with no known exact solution, provided validation uses invariant quantities such as the Ricci tensor and Kretschmann scalar rather than the loss function alone.

What carries the argument

The mechanism that carries the argument is the PINN training loop in which a neural network approximates the metric functions and the loss function is built directly from the Einstein equation residuals $R_{\mu\nu}=0$ together with weighted boundary-condition terms, with derivatives obtained by automatic differentiation. Two structural choices do the heavy lifting. First, the metric functions are written not as raw network outputs but as a known Schwarzschild-like asymptotic piece plus a neural correction, for example $A(r,\theta)^2=(1-2m/r+u_0/r^2)^{1+q}$, so the network only needs to learn deviations from the expected falloff. Second, domain decomposition trains the radial subdomains sequentially, transferring learned weights from the outermost subdomain inward; in the relaxed scheme this is what propagates boundary information into subdomains that receive no boundary conditions. The loss weights allow redundant Einstein equations to be included without blocking convergence.

What would settle it

Evaluate the Kretschmann scalar from the relaxed-boundary $q=1$ PINN solution on a dense independent grid in the innermost subdomain $r\in(10m,18m)$, $\theta\in(0,\pi)$, and compare it with the exact q-metric value; if the Einstein residual loss is low but the relative error of this invariant is far above $10^{-3}$ in that region, the claim that the residuals alone select the physical solution is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that Einstein's vacuum field equations $R_{\mu\nu}=0$ for static spacetimes can be solved directly by physics-informed neural networks, and that the method extends from the spherically symmetric Schwarzschild metric to the axisymmetric q-metric, the simplest exact vacuum solution with a quadrupole moment. In the strict boundary-condition setup, where all four metric functions are pinned to their known asymptotic correction form at the outer boundary, the network reproduces the q-metric with relative RMSE below $10^{-6}$. When those boundary conditions are relaxed so that only the outermost radial subdomain carries boundary data and the inner subdomains are trained from the Einstein residuals alone, the four metric functions are still recovered with relative RMSE between $10^{-6}$ and $10^{-4}$. The paper also shows that in the flat limit $q=-1$, where the network must learn a coordinate representation of Minkowski space rather than the standard line element, the Riemann tensor components reconstructed from the network have $L^\infty$ errors of order $10^{-4}$, which it interprets as residual numerical curvature rather than physical curvature.

Load-bearing premise

The relaxed boundary-condition scheme assumes that a neural network trained on the Einstein residuals alone in the inner radial subdomains, with boundary data supplied only at the outermost edge and passed inward by sequential training, will select the true physical metric rather than another Ricci-flat geometry; the paper itself warns that a low loss value does not guarantee the physical solution.

Editorial extensions

If this is right

  • An axisymmetric vacuum spacetime with a quadrupole moment can be recovered from Einstein's equations by a PINN without any mesh, at relative errors competitive with conventional numerical schemes.
  • Boundary data need not be supplied throughout the domain: sequential domain decomposition lets the outermost boundary conditions propagate inward while inner subdomains are constrained only by the field equations.
  • Validation by invariants such as the Ricci tensor and Kretschmann scalar is necessary, because the same physical spacetime may be represented in different coordinate forms by the network.
  • The same architecture and training schedule can be pointed at spacetimes with no known exact solution, using asymptotic or matched-interior boundary data.
  • Redundant Einstein equations need not be discarded; their effect can be absorbed through loss weights, simplifying the setup for more complex metrics.

Reading between the lines

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

  • An extension the paper does not test is dropping the angular periodic boundary conditions as well as the inner radial ones; if the relaxed scheme still converges to the q-metric, boundary information enters only through the outermost edge, which would strengthen the method for genuinely unknown domains.
  • The residual Riemann curvature measured in the flat-limit case suggests a general acceptance test for PINN solutions in general relativity: compute the $L^\infty$ norm of an invariant curvature scalar and require it to track the exact value, not just the loss.
  • The sequential weight-transfer scheme resembles a continuation method; one might initialize the inner subdomains with the outer solution and anneal a source term, which is a different route to stabilizing strong-gradient regions near $r=2m$.
  • If the approach extends to time-dependent or non-static spacetimes, the same boundary-relaxation idea could reduce the burden of specifying initial and boundary data in numerical relativity, though that extension is not established here.
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 / 4 minor

Summary. The paper applies physics-informed neural networks (PINNs) to the vacuum Einstein equations for two static spacetimes. For Schwarzschild it reproduces the metric functions using a domain-decomposed PINN and reports RMSE values. For the q-metric it introduces a parametrized ansatz with four correction fields, compares 'strict' and 'relaxed' boundary-condition strategies, and reports that the metric functions are recovered with relative RMSE below 1e-4 for q=1 and below 1e-6 for q=-1, with additional validation via Ricci and Riemann tensor components and the Kretschmann scalar. The paper concludes that PINNs are robust and flexible for solving Einstein's equations and promising for configurations without known analytical solutions.

Significance. If fully supported, the paper would be a useful benchmark showing how boundary conditions and domain decomposition affect PINN training in general relativity. The Schwarzschild test is a genuine forward problem and the numerical setup is documented in enough detail to allow reproduction. The additional validation using curvature invariants is good practice. However, as argued in Major Comment 1, the q-metric experiments embed the exact solution in the ansatz, so they do not test the network's ability to discover unknown spacetimes. The abstract and conclusions therefore overstate the significance for configurations without known analytical solutions.

major comments (3)
  1. [§V, Eqs. (22)–(25) and §VI] The q-metric ansatz is built from the exact q-metric factors, so the exact solution corresponds identically to u_i=0 for all four correction fields. The paper itself states in §VI that the strict boundary conditions 'imply the trivial solution u_i^NN = 0'. Consequently, the q-metric experiments only test whether the network learns to output near-zero corrections, and the RMSE values in Tables II–IV measure deviation from this zero field rather than the ability of the network to discover or represent the q-metric geometry. In the q=-1 case the effect is even stronger: because the exponents 1+q and -1-q vanish in Eqs. (22) and (23), the metric functions A and B are independent of u0 and u1, respectively. This circular construction prevents the experiments from supporting the abstract's claim that the method is a promising foundation for configurations with no known analytical solution. Please either replace the ansatz with a general static axisymmetric metric ansatz that does not contain the exact solution, or substantially narrow the claims.
  2. [§VII, Fig. 11] In the relaxed boundary-condition scheme, the inner radial subdomains receive no boundary conditions and no interface continuity conditions between subdomains are described; the only connection is the weight initialization propagated from the outermost subdomain. Since the outer-subdomain loss saturates at 1e-2 and the paper itself warns in §VIII that a low loss does not guarantee the physical solution, the claim that Einstein residuals alone select the correct interior solution is not established. Please either impose and document explicit interface conditions, provide a convergence or uniqueness argument for the relaxed scheme, or clearly limit the claim to the specific sequential-initialization protocol used.
  3. [§II versus §VII.A] The text around Eq. (3) states that 'the limit q=1 corresponds to the flat Minkowski spacetime', but Section VII.A correctly identifies the flat limit as q=-1, consistent with the multipole moments in Eq. (3), which are proportional to (q+1), and with the vanishing Riemann tensor computed there. The q=1 statement is a factual error; the q=1 cases in Sections VI and VII are genuine quadrupole spacetimes, not flat space. Please correct the inconsistency.
minor comments (4)
  1. [§IX] The Conclusions state that the Schwarzschild temporal and radial components have RMSE of order 1e-3 and 1e-8, respectively, while Table I and §IV report f ~1e-8 and g ~1e-4; please make the numbers consistent.
  2. [§VII.A] The paragraph discussing the q=-1 error analysis refers to 'Fig. 18' and 'Table III', but the corresponding figure is Fig. 25 and the table is Table IV.
  3. [§I] The organization paragraph contains 'Section IV III' before presenting the Schwarzschild case; this should read 'Section IV'.
  4. [Tables II and III and §VI] The captions of Tables II and III call the quantity 'Global mean squared error (RMSE)' while the text describes it as a relative L2 error; please make the terminology consistent. Also, Eqs. (22)–(25) are described as 'strict boundary conditions', but they are actually a parametrization of the metric functions; please clarify explicitly what values of u_i are imposed on the subdomain boundaries in the strict case.

Circularity Check

3 steps flagged · score 7.0 of 10

q-metric 'predictions' reduce to the ansatz: Eqs. (22)-(25) embed the exact q-metric as u_i=0, strict BCs force that trivial solution, and the relaxed case uses the same pre-embedded target.

  1. self definitional [Sec. VI, Eqs. (22)-(25)]
    "A(r, θ)^2 = (1 − 2m/r + u0(r, θ)/r^2)^{1+q}, B(r, θ)^2 = (1 − 2m/r + u1(r, θ)/r^2)^{−1−q} (1 + m^2 sin^2 θ/(r^2−2mr))^{−q(q+2)}, C(r, θ)^2 = (1 − 2m/r + u2(r, θ)/r^2)^{−q} (1 + m^2 sin^2 θ/(r^2−2mr))^{−q(q+2)} r^2, D(r, θ)^2 = (1 − 2m/r + u3(r, θ)/r^2)^{−q} r^2 sin^2 θ."

    Each metric function is built from the exact q-metric factors with q,m fixed, and the unknown network fields u_i appear only as additive u_i/r^2 corrections. Setting u_i=0 returns precisely Eq. (2), the known q-metric. Thus the target solution is already contained in the network representation by construction. Minimizing the Einstein residuals drives u_i toward zero, so the reported 'reproduction' of the q-metric measures how well the network learns a pre-embedded function, not whether the PINN derived the spacetime from the field equations.

  2. fitted input called prediction [Sec. VI, discussion after Table II]
    "This behavior is consistent with the fact that, in the case considered here, the strict boundary conditions imply the trivial solution u_iNN = 0."

    The paper admits that the strict boundary conditions force the network output u_i to be identically zero. But u_i=0 is exactly the q-metric by Ansatz (22)-(25). Therefore the high accuracy reported in Table II (RMSE below 1e-6) is the accuracy of a regression to zero in a parametrization that already contains the known solution. The result is forced by the boundary conditions plus the constructed ansatz, not by an independent solution of Einstein's equations.

1 more flagged steps
  1. self definitional [Sec. VII, description of relaxed boundary conditions and Fig. 11]
    "Training process begins in the outermost subdomain, where strict boundary conditions are imposed; subsequently, the inner batches, without any boundary conditions, are incorporated into the training."

    The relaxed experiment uses the same ansatz (22)-(25), so the exact q-metric is still u_i=0. Only the outer batch receives boundary data; inner batches are trained on residuals plus weight initialization, with no interface constraints. The RMSE in Table III therefore still measures convergence to a solution that was put into the network by construction. This cannot support the paper's claim of readiness for configurations with no known analytical solution, especially since Sec. VIII warns that a loss of order 1e-2 need not correspond to the physical solution.

full rationale

The circularity is concentrated in the q-metric sections (VI-VII). Equations (22)-(25) define all four metric functions using the known q-metric factors, which makes u_i=0 the exact Einsteinian solution. The paper explicitly acknowledges that strict boundary conditions force this trivial solution, and the relaxed case simply moves the boundary data to the outermost subdomain without changing the pre-embedded target. Consequently, Tables II-IV report how well the network fits a solution that is already present in the ansatz, not evidence that the method discovers unknown spacetimes. The abstract's conclusion that PINNs 'offer a promising foundation for investigating gravitational configurations for which exact analytical solutions are unknown' is an extrapolation beyond what these benchmarks can show, since every q-metric experiment contains the exact answer by construction. The Schwarzschild benchmark is less problematic: the ansatz f=1-2m/r-u0/r^2, g=u1 does not embed the exact g, and the reported RMSE of order 1e-4 for g gives the method some independent validation. The cited q-metric background from Quevedo [7] is standard self-citation but is not the mechanism of circularity here. Overall, the central showcase result reduces by construction to regression toward the known solution, giving a partial but real circularity score of 7 rather than higher because the network could in principle fail to converge and the Schwarzschild test has independent content.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

No free parameters are fitted to data in this paper. The mass m and quadrupole parameter q are inputs taken from the known exact solutions, and the network hyperparameters (layer sizes, learning rate, loss weights) are chosen by hand but are not the subject of the central claim. The main modeling choice is the functional ansatz, which is an ad hoc construction rather than a fitted number.

assumptions (3)
  • domain assumption The q-metric (Eq. 2) is an exact vacuum solution of Einstein's equations with the stated multipole moments.
    Invoked in Sec. II and used as the target for the PINN; the paper cites ref. [7] rather than re-deriving the solution.
  • standard math For a static spherically symmetric metric, R_theta_theta=0 is redundant given R_tt=0 and R_rr=0, due to Bianchi identities.
    Stated in Sec. IV A as justification for keeping redundant equations in the loss; the paper cites ref. [19].
  • ad hoc to paper The PINN ansatz (22)-(25) is sufficient to represent the target q-metric.
    The authors construct the ansatz so that u_i=0 recovers the exact metric; this is a deliberate modeling choice tailored to the known solution (Sec. VI).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Solving Einstein's Vacuum Equations with Physics-Informed Neural Networks: Boundary Conditions and Domain Decomposition." pith.science (2026). https://pith.science/paper/CQ7QWINP

@misc{pith2026260808846,
  author       = {Pith},
  title        = {Pith review of: Solving Einstein's Vacuum Equations with Physics-Informed Neural Networks: Boundary Conditions and Domain Decomposition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CQ7QWINP}},
  note         = {Machine review of arXiv:2608.08846}
}
abstract

We investigate the application of Physics-Informed Neural Networks (PINNs) to the numerical solution of Einstein's vacuum field equations for static spacetimes. We first reproduce the Schwarzschild solution and then extend the method to the axisymmetric $q-$metric, a nontrivial exact solution characterized by a mass quadrupole moment. We analyze the influence of boundary conditions, domain decomposition, and equation redundancy on the convergence and stability of the training process. The proposed framework accurately reconstructs the metric functions in the computational domain while maintaining small residual errors. Our results demonstrate that PINNs provide a robust and flexible approach to solving Einstein's equations and offer a promising foundation for investigating gravitational configurations for which exact analytical solutions are unknown.

Figures

Figures reproduced from arXiv: 2608.08846 by the authors.

Figure 1
Figure 1. FIG. 1: Flowchart of the training of a Physics-Informed Neural Network PINNs, where [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Convergence of the loss functions for Schwarzschild spacetime using a domain [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: Comparison between the exact solution and the PINN solution obtained using [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (26 more)
Figure 4
Figure 4. Figure 4: FIG. 4: Loss function for strict boundary conditions, with the metric function parameters [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p016_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]
Figure 9
Figure 9. Figure 9: FIG. 9: Three-dimensional visualizations of the metric functions and their respective approx [PITH_FULL_IMAGE:figures/full_fig_p019_9.png]
Figure 10
Figure 10. Figure 10: FIG. 10: Local absolute error of each metric function, computed as the absolute difference [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: FIG. 11: Schematic of the domain decomposition implemented in the neural network of [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: FIG. 12: Loss function for relaxed boundary condition, with the metric function parameters [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: FIG. 13: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: FIG. 14: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: FIG. 15: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: FIG. 16: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: FIG. 17: Three-dimensional visualizations of the metric functions and their respective ap [PITH_FULL_IMAGE:figures/full_fig_p027_17.png]
Figure 18
Figure 18. Figure 18: FIG. 18: Local absolute error of each metric function, computed as the absolute difference [PITH_FULL_IMAGE:figures/full_fig_p028_18.png]
Figure 19
Figure 19. Figure 19: FIG. 19: Loss function for the relaxed boundary condition, with the metric function param [PITH_FULL_IMAGE:figures/full_fig_p030_19.png]
Figure 20
Figure 20. Figure 20: FIG. 20: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p031_20.png]
Figure 21
Figure 21. Figure 21: FIG. 21: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p032_21.png]
Figure 22
Figure 22. Figure 22: FIG. 22: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p033_22.png]
Figure 23
Figure 23. Figure 23: FIG. 23: Comparison between the exact metric function [PITH_FULL_IMAGE:figures/full_fig_p034_23.png]
Figure 24
Figure 24. Figure 24: FIG. 24: Three-dimensional visualizations of the metric functions and their respective ap [PITH_FULL_IMAGE:figures/full_fig_p035_24.png]
Figure 25
Figure 25. Figure 25: FIG. 25: Local absolute error of each metric function, computed as the absolute difference [PITH_FULL_IMAGE:figures/full_fig_p036_25.png]
Figure 26
Figure 26. Figure 26: FIG. 26: Absolute error of the non-trivial Ricci tensor components computed from the PINN [PITH_FULL_IMAGE:figures/full_fig_p039_26.png]
Figure 27
Figure 27. Figure 27: FIG. 27: Absolute error of the non-trivial Ricci tensor components computed from the [PITH_FULL_IMAGE:figures/full_fig_p040_27.png]
Figure 28
Figure 28. Figure 28: FIG. 28: Kretschmann scalar computed from the PINN approximation under strict boundary [PITH_FULL_IMAGE:figures/full_fig_p041_28.png]
Figure 29
Figure 29. Figure 29: FIG. 29: Kretschmann scalar computed from the PINN approximation under relaxed bound [PITH_FULL_IMAGE:figures/full_fig_p041_29.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 16 canonical work pages

  1. [1]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, and L. Yang, Nature Reviews Physics3, 422 (2021)

    2021

  2. [2]

    Cuomo, V

    S. Cuomo, V. S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, and F. Piccialli, Journal of Scientific Computing92, 88 (2022)

    work page 2022

  3. [3]

    Raissi, P

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, Journal of Computational Physics378, 686 (2019)

    2019

  4. [4]

    Li, C.-Q

    Z.-H. Li, C.-Q. Li, and L.-G. Pang, arXiv preprint arXiv:2309.07397 (2023)

    arXiv 2023

  5. [5]

    Stefanou, J

    P. Stefanou, J. F. Urb´ an, and J. A. Pons, Monthly Notices of the Royal Astronomical Society 526, 1504 (2023)

    work page 2023

  6. [6]

    Patel, A

    N. Patel, A. Aykutalp, and P. Laguna, arXiv preprint arXiv:2401.01440 (2024)

    arXiv 2024

  7. [7]

    The geometric and physical properties of this metric have been in- vestigated in detail in [7, 9–14]

    given by ds2 =− 1− 2m r 1+q dt2 + 1− 2m r −q × " 1 + m2 sin2 θ r2 −2mr −q(q+2) r2dθ2 + dr2 1−2m/r +r 2 sin2 θ dφ2 # .(2) The q-metric is obtained from the Schwarzschild metric by applying a Zipoy-Voorhees transformation [8, 9]. The geometric and physical properties of this metric have been in- vestigated in detail in [7, 9–14]. The deformation of the grav...

    work page 2000

  8. [8]

    Quevedo, International Journal of Modern Physics D20, 1779 (2011)

    H. Quevedo, International Journal of Modern Physics D20, 1779 (2011)

    work page 2011

Show all 20 references
  1. [9]

    D. M. Zipoy, Journal of Mathematical Physics7, 1137 (1966)

    1966

  2. [10]

    Voorhees, Physical Review D2, 2119 (1970)

    B. Voorhees, Physical Review D2, 2119 (1970)

    1970

  3. [11]

    Hern´ andez-Pastora and J

    J. Hern´ andez-Pastora and J. Mart ´ ın, General Relativity and Gravitation26, 877 (1994)

    1994

  4. [12]

    Malafarina, Dynamics and Thermodynamics of Blackholes and Naked Singularities p

    D. Malafarina, Dynamics and Thermodynamics of Blackholes and Naked Singularities p. 20 (2004)

    2004

  5. [13]

    Quevedo, inNeutron Stars: Physics Properties and Dynamics, edited by N

    H. Quevedo, inNeutron Stars: Physics Properties and Dynamics, edited by N. Takibayev and K. Boshkayev (Nova Science Publishers, 2017), Physics Research and Technology, ISBN 978-1-53610-525-4, 1606.09361

    2017 arXiv

  6. [14]

    Frutos-Alfaro, H

    F. Frutos-Alfaro, H. Quevedo, and P. A. Sanchez, Royal Society Open Science5, 170826 (2018)

    2018

  7. [15]

    Boshkayev, E

    K. Boshkayev, E. Gasperin, A. Guti´ errez-Pi˜ neres, H. Quevedo, and S. Toktarbay, Physical Review D93, 024024 (2016)

    2016

  8. [16]

    Geroch, Journal of Mathematical Physics11, 1955 (1970)

    R. Geroch, Journal of Mathematical Physics11, 1955 (1970)

    1970

  9. [17]

    Geroch, Journal of Mathematical Physics11, 2580 (1970)

    R. Geroch, Journal of Mathematical Physics11, 2580 (1970). 45

    1970

  10. [18]

    Griewank and A

    A. Griewank and A. Walther,Evaluating Derivatives: Principles and Techniques of Algorith- mic Differentiation(SIAM, 2008), 2nd ed

    2008

  11. [19]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, et al. (2018), http://github.com/google/jax

    2018

  12. [20]

    Alcubierre,Introduction to 3+ 1 numerical relativity, vol

    M. Alcubierre,Introduction to 3+ 1 numerical relativity, vol. 140 (OUP Oxford, 2008). 46

    2008

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.