Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Solving Euler equations with Multiple Discontinuities via Separation-Transfer Physics-Informed Neural Networks

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

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

Pith's one-line read A separation-transfer training loop lets physics-informed neural networks resolve multiple shock discontinuities sequentially, starting from the strongest.

desk verdict A plausible sequential domain-decomposition trick that consistently helps PINNs on multiple discontinuities, but the 'always improves' claim outruns the straight-line test cases. read the letter →

arxiv 2505.20361 v1 pith:CMZ7LJ57 submitted 2025-05-26 physics.flu-dyn cs.LG

Chuanxing Wang , Hui Luo , Kai Wang , Guohuai Zhu , Mingxing Luo This is my paper · ORCID
classification physics.flu-dyncs.LG
keywords physics-informedneuralnetworksEulerequationsmultiplediscontinuitiesshock-interfaceinteractionshockrefractiondomaindecompositiontransferlearninggradientpathology
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

Physics-informed neural networks (PINNs) solve PDEs by turning residual, initial, and boundary conditions into a loss, but when several shock fronts coexist, the strongest front dominates the training and the weaker fronts stay unresolved. ST-PINNs breaks this deadlock by resolving the fronts one at a time: train any PINN variant until the strongest discontinuity is captured, locate that front, split the domain along it, and transfer-learn on the subdomain containing the remaining discontinuities with predicted values at the cut as boundary conditions. The paper shows the cycle captures every front in order of decreasing strength, and reports improved L2 accuracy over the PINNs-WE and GA-PINNs base models on three Euler problems, including the two-dimensional unsteady planar shock refraction problem.

What carries the argument

The load-bearing object is the discontinuity intensity $I = |V_+ - V_-|/\Delta x$, computed from a chosen physical quantity on a discretized grid, which ranks the jumps and identifies the primary discontinuity as the point of maximum $I$. Around that front, the method fits a trajectory (in all experiments a straight line in $x$--$t$), reads off averaged values of the base model just ahead of the front, and divides the domain into subdomains. Transfer learning, initialized from the previous model, retrains only the subdomain that still contains poorly resolved fronts, using the near-front values as boundary conditions; with the primary front removed, the next-strongest front becomes dominant in the loss and can be learned.

What would settle it

Run the same three experiments with the fitted shock trajectories deliberately offset (for example, use $x_{\text{shock}} = 1.606\, t + 0.5 \pm 0.05$ in the one-dimensional case) and check the final L2 errors; if the reported improvements over the base model disappear for any offset, the claim that ST-PINNs always improves once the primary front is captured is false.

Watch

Extended reading notes

Core claim

The paper's thesis is that a multi-discontinuity Euler problem becomes tractable when the discontinuities are solved sequentially from strongest to weakest. Given any PINN variant that can capture at least one discontinuity with reasonable precision, ST-PINNs determines the primary front through a discontinuity intensity measure, fits its trajectory, cuts the domain there, and uses the base model's near-front predictions as boundary conditions for a transfer-learned subdomain; the next unresolved front then becomes the primary one and is captured in the following iteration. The authors claim this loop always improves the base prediction once the first front is well captured, and they validate it on one-dimensional shock-interface interaction, quasi-one-dimensional planar shock-interface interaction, and two-dimensional unsteady planar shock refraction, the last being the first PINN-based treatment of that problem.

Load-bearing premise

The method assumes the strongest jump's location can be reliably read off the first model's output and fitted as a simple line, and that the model's values near that jump are accurate enough to serve as boundary conditions for the retrained subdomains.

Editorial extensions

If this is right

  • Any PINN variant that captures one strong discontinuity can be wrapped in the separation-transfer loop to handle multiple discontinuities without changing its architecture.
  • The multi-front problem is reduced to a sequence of single-front subproblems, so the gradient-pathology failure caused by one dominant front is bypassed.
  • The training pipeline needs no simulation or experimental data: the experiments use only the PDE residual and initial conditions, with the reference solution used solely for measuring error.
  • The approach extends, in principle, to any number of discontinuities, since each iteration promotes the next-strongest unresolved front to primary.

Reading between the lines

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

  • A natural stress test is to apply the same loop to a curved or kinked shock front, since the paper's fitted trajectories are straight lines; a curved front would require a local front-fitting step or a piecewise fit, and the method's success would depend on that step.
  • The separation logic transfers beyond Euler equations: any conservation law with traveling discontinuities, such as shallow-water bores or elastoplastic waves, could in principle be solved by the same strong-to-weak split-and-transfer recipe.
  • Because each transfer-learning stage reuses the previous model, training cost should scale roughly with the number of discontinuities; the paper does not report wall-clock timings, so the practical cost trade-off against a single high-resolution network remains open.
  • The improvement depends on the base model's near-front values being accurate; pairing ST-PINNs with an adaptive sampling scheme at the cut could reduce the inherited error and make the method more robust.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes ST-PINNs, a staged training procedure for PINNs solving Euler equations with multiple discontinuities. The method first trains a base PINN variant over the full domain until the strongest discontinuity is captured, extracts the discontinuity locus and adjacent state values from that model, divides the domain along the locus, and retrains the subdomain containing weaker discontinuities via transfer learning, using the base model's near-front predictions as boundary conditions. Sequential subdivision is repeated until all discontinuities are resolved, and the subdomain networks are combined. Validation is reported for a 1D shock-interface Riemann problem, a quasi-1D planar shock-interface interaction, and a 2D unsteady planar shock refraction problem, using PINNs-WE, GA-PINNs, simplified variants, and their ST-PINNs counterparts. Across these cases the reported L2 relative errors are consistently lower after ST-PINNs refinement than for the base models.

Significance. If the central claim holds, ST-PINNs offers a conceptually simple and potentially general strategy for turning any PINN variant that can capture one discontinuity into a solver for multi-discontinuity problems. The paper's main strengths are the consistency of the reported improvements across six base-model configurations (Tables II, IV, VI, VII, and VIII), the use of PDE and initial-condition information only during training, and the first reported PINN treatment of the 2D unsteady planar shock refraction problem. The method is algorithmic and does not rely on synthetic training data from classical solvers during ST-PINNs training. However, the strength of the evidence is tempered by the incomplete definition of the error metric, the absence of repeated-run statistics, and the fact that the discontinuity extraction and boundary-value construction are tested only in favorable straight-front configurations.

major comments (3)
  1. [§IV, Eq. (14)] The L2 relative error is written as sqrt((fpred - ftrue)^2 / ftrue^2), with no summation over sample points and no normalization over the field. As written this defines an error for a single point, not an L2 error, and it cannot reproduce the reported percentages. Since every quantitative claim in Tables II, IV, VI, VII, and VIII rests on this metric, the definition must be corrected (e.g., sum over sample points and normalize by the sum of ftrue^2) and the concrete evaluation formula should be stated explicitly.
  2. [§V; Algorithm 1, lines 4–8] The conclusion that ST-PINNs 'can always improve' is not supported by the reported experiments because the two inputs on which the method depends—the fitted discontinuity locus and the boundary values VNN,±—are supplied in favorable form in every test. The loci are straight-line fits with known planar structure (x_shock = 1.606t + 0.5 in §IVA; x1 = 2.228t + 0.20 and x2 = 2.232t + 0.70 in §IVB; x_shock = 2.370t + 0.2 and x_contact = 1.05t + 0.2 in §IVC), and the boundary values are single averaged states. In §IVC the incident-shock front and its intersection point are additionally used as known prior information. Appendix B perturbs only base-model hyperparameters, never the extracted trajectory or the base-model boundary values, so error propagation from an imperfect extraction is not characterized. The claim should either be restricted to the tested straight-front cases or supported by experiments with perturbed or biased front locations and with non-planar fronts.
  3. [§IV, Tables II/IV/VI and Appendix B] All numerical results are single-run values. Tables II, IV, VI, VII, and VIII report one error per configuration, with no seed information, no standard deviation, and no repeated training trials. Because neural-network training is stochastic and the method sequentially reuses a fitted trajectory and boundary values, the claimed consistency of improvement requires at least a few independent runs per configuration to rule out that the reported gains are within run-to-run variability.
minor comments (5)
  1. [§I and title] The name of the method alternates between 'Separation-Transfer' in the title and abstract and 'Separate-Transfer' in Section I; this should be made consistent.
  2. [Tables IV and VI] The label 'ST-PINNS' appears with an extra 'S' in Tables IV and VI, while the text uses 'ST-PINNs'; please correct the typo.
  3. [§III, end of Methodology] The statement that 'the same network architecture is employed across all stages, with only the loss function weights adjusted accordingly' is contradicted by Tables I, III, and V, where sampling-point counts, learning-rate schedules, and initial-condition weights also change between models; the sentence should be clarified to refer to architecture alone.
  4. [Table I] The entries 'Boundary Points NAN' and 'ωBC NAN' should be replaced by a conventional notation such as 'not used' or '—', since 'NAN' is a numeric sentinel with a different meaning.
  5. [§IIB, Eq. (9)] The notation |·| is used both for element-wise absolute value and for the mean of element-wise absolute values without a clear distinction; this makes Eq. (9) ambiguous and should be clarified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ST-PINNs is evaluated against external reference solutions, and its auxiliary fitted inputs are not the predicted quantities.

full rationale

The paper's derivation chain does not reduce to its inputs. Final predictions are compared with independent references: exact Riemann/theoretical solutions in Section IVA and JAX-FLUIDS numerical solutions in Sections IVB and IVC, with the JAX-FLUIDS setup given in Appendix A. The quantities extracted from the base model—the fitted discontinuity trajectories such as x_shock = 1.606 t + 0.5 and averaged boundary states such as rho = 1.590, u = 0.603, p = 1.945—are internal algorithmic ingredients used to construct subdomain training problems; they are not the quantities being scored. The L2 errors in Tables II, IV, VI, VII, and VIII are computed against f_true obtained from traditional numerical simulation or the theoretical solution, so the claimed ST-PINNs improvement is externally falsifiable. The paper cites standard prior PINN variants (PINNs-WE, GA-PINNs, DeepXDE, CPINNs, XPINNs) but none of these citations functions as an unverified load-bearing premise or imported uniqueness theorem; no central argument is forced by a self-citation chain. The Section V statement that 'once the base model in the first training step captures at least one discontinuity with a certain precision, ST-PINNs then can always improve the predictive performance' is broader than the experiments strictly establish—Appendix B sweeps base-model hyperparameters but does not perturb fitted shock locations or boundary values, so robustness of the extracted front is not fully tested—but that is a generality and robustness limitation, not circularity. In particular, the use of base-model predictions as subdomain boundary conditions does not make the measured improvement tautological, because the evaluation metric is error relative to an independent reference, not agreement with the base model itself. No circular step satisfying the quote-and-reduction test was found.

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

The central claim rests on standard Euler-equation physics, on an unproven assumption that PINN residual minimization locates the primary discontinuity reliably, and on problem-specific trackability of fronts as fitted straight lines. The empirically adjusted quantities are the base-model hyperparameters and the linear trajectory fits that define subdomain boundaries. No new physical entities are introduced.

free parameters (4)
  • Local weight hyper-parameter k (PINNs-WE and simplified variants) = 0.6 (1D), 0.1 (quasi-1D and 2D simplified tests)
    Controls the velocity-gradient weighting in Eqs. (11) and (19); it determines which discontinuity the base model captures first and therefore where ST-PINNs splits the domain.
  • GA-PINNs gradient-weight exponents alpha and beta = Varied per case: (0.4, 0.3), (0.2, 0.5), (0.2, 0.1)
    Set by hand in Eq. (18); the reported error reductions depend on these choices, and Appendix B shows the method across multiple values.
  • Fitted discontinuity trajectory coefficients = x_shock=1.606t+0.5; x1=2.228t+0.20, x2=2.232t+0.70; x_shock=2.370t+0.2, x_contact=1.05t+0.2
    Linear fits to the base model's identified front positions. These fits set the subdomain boundaries and the location where predicted values are imposed as boundary conditions, so error in the fits propagates directly into subdomain solutions.
  • Loss weights and learning-rate schedule = omega_IC=5-10, omega_RH=10, omega_CON=10, omega_BC=10; lr from 1e-3 down to 1e-7 in three stages
    Chosen separately for each model and stage (Tables I, III, V). They affect whether transfer learning preserves the primary discontinuity while training weaker ones.
assumptions (4)
  • domain assumption The inviscid Euler equations with gamma = 1.4 and the ideal-gas equation of state are the correct model for the shock-interface problems studied.
    Adopted in Section IIA and all three test cases; this is standard compressible hydrodynamics, but the accuracy of the model itself is not part of the validation.
  • domain assumption A tanh-activated neural network minimizing the PDE residual can approximate the discontinuous weak solution well enough for the primary discontinuity to be located from the fitted intensity measure I.
    Algorithm 1, Step 4 assumes the base model's largest gradient marks the true shock position; no convergence or uniqueness guarantee is provided for PINNs on discontinuous solutions.
  • ad hoc to paper Shock and contact fronts can be represented by fitted straight trajectories over the time horizon.
    The method fits x_shock(t) and x_contact(t) linearly and assumes the transmitted shock is planar (Sections IVA-IVC). This restricts the method to problems where such fitted fronts are valid; general curved shocks would break the separation step.
  • domain assumption Transfer learning from the previous model, with new boundary conditions on subdomains, preserves the already-captured discontinuity while allowing weaker discontinuities to improve.
    The central mechanism of Steps 7-8 in Algorithm 1 is justified only empirically; there is no analysis of why reusing the weights does not destroy the primary shock during subdomain retraining.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Solving Euler equations with Multiple Discontinuities via Separation-Transfer Physics-Informed Neural Networks." pith.science (2026). https://pith.science/paper/CMZ7LJ57

@misc{pith2026250520361,
  author       = {Pith},
  title        = {Pith review of: Solving Euler equations with Multiple Discontinuities via Separation-Transfer Physics-Informed Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CMZ7LJ57}},
  note         = {Machine review of arXiv:2505.20361}
}
read the original abstract

Despite the remarkable progress of physics-informed neural networks (PINNs) in scientific computing, they continue to face challenges when solving hydrodynamic problems with multiple discontinuities. In this work, we propose Separation-Transfer Physics Informed Neural Networks (ST-PINNs) to address such problems. By sequentially resolving discontinuities from strong to weak and leveraging transfer learning during training, ST-PINNs significantly reduce the problem complexity and enhance solution accuracy. To the best of our knowledge, this is the first study to apply a PINNs-based approach to the two-dimensional unsteady planar shock refraction problem, offering new insights into the application of PINNs to complex shock-interface interactions. Numerical experiments demonstrate that ST-PINNs more accurately capture sharp discontinuities and substantially reduce solution errors in hydrodynamic problems involving multiple discontinuities.

Figures

Figures reproduced from arXiv: 2505.20361 by the authors.

Figure 1
Figure 1. FIG. 1. Schematic of ST-PINNs. (a) Solve the entire solution [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. The position of the transmitted shock front: blue [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. The comparison between the neural network predictions (blue circles) and the theoretical solution (black solid line) for [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Initial conditions of the mass density [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Results of the quasi-one-dimensional planar shock-interface interaction problem at [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Initial conditions of the mass density [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7. Step-by-step result plots by ST-PINNs for two-dimensional planar shock refraction problem. Four columns show the [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Efficient Weak-Entropy PINN for Solving Hyperbolic Conservation Laws

    math.NA 2026-08 conditional novelty 6.0 of 10

    WEPINN enforces weak formulation and entropy condition with trigonometric test functions and fast Fourier transform integration, resolving shocks and rarefactions in conservation laws more accurately than Diff-PINN, V...

  2. Physics-informed token transformer methodology for nonlinear balance laws. I. Schwarzschild--Burgers fluid flows

    math.NA 2026-07 conditional novelty 6.0 of 10

    A shock-aware token-transformer that embeds the exact Rankine-Hugoniot speed for Schwarzschild-Burgers flows outperforms a well-balanced finite-volume scheme on analytic Riemann problems, with the analytic prior carry...

Reference graph

Works this paper leans on

26 extracted references · 5 canonical work pages · cited by 2 Pith papers

  1. [1]

    Raissi , author P

    author author M. Raissi , author P. Perdikaris , \ and\ author G. Karniadakis ,\ title title Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , \ 10.1016/j.jcp.2018.10.045 journal journal J. Comput. Phys. \ volume 378 ,\ pages 686--707 ( year 2019 ) NoStop

    doi:10.1016/j.jcp.2018.10.045

  2. [2]

    Raissi , author A

    author author M. Raissi , author A. Yazdani , \ and\ author G. E. \ Karniadakis ,\ title title Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations , \ 10.1126/science.aaw4741 journal journal Science \ volume 367 ,\ pages 1026--1030 ( year 2020 ) NoStop

    doi:10.1126/science.aaw4741

  3. [3]

    Qiu , author R

    author author R. Qiu , author R. Huang , author Y. Xiao , author J. Wang , author Z. Zhang , author J. Yue , author Z. Zeng , \ and\ author Y. Wang ,\ title title Physics-informed neural networks for phase-field method in two-phase flow , \ https://doi.org/10.1063/5.0091063 journal journal Phys. Fluids \ volume 34 ( year 2022 ) NoStop

    doi:10.1063/5.0091063

  4. [4]

    Zhao , author W

    author author J. Zhao , author W. Wu , author X. Feng , \ and\ author H. Xu ,\ title title Solving Euler equations with gradient-weighted multi-input high-dimensional feature neural network , \ https://doi.org/10.1063/5.0194523 journal journal Phys. Fluids \ volume 36 ( year 2024 ) NoStop

    work page doi:10.1063/5.0194523

  5. [5]

    Haghighat \ and\ author R

    author author E. Haghighat \ and\ author R. Juanes ,\ title title SciANN: A Keras/TensorFlow wrapper for scientific computations and physics-informed deep learning using artificial neural networks , \ 10.1016/j.cma.2020.113552 journal journal Comput. Methods Appl. Mech. Eng. \ volume 373 ,\ pages 113552 ( year 2021 ) NoStop

    arXiv 2020

  6. [6]

    Haghighat , author D

    author author E. Haghighat , author D. Amini , \ and\ author R. Juanes ,\ title title Physics-informed neural network simulation of multiphase poroelasticity using stress-split sequential training , \ 10.1016/j.cma.2022.115141 journal journal Comput. Methods Appl. Mech. Eng. \ volume 397 ,\ pages 115141 ( year 2022 ) NoStop

    arXiv 2022

  7. [7]

    Wang , author Z

    author author T. Wang , author Z. Hu , author K. Kawaguchi , author Z. Zhang , \ and\ author G. E. \ Karniadakis ,\ title title Tensor neural networks for high-dimensional Fokker–Planck equations , \ 10.1016/j.neunet.2025.107165 journal journal Neural Netw. \ volume 185 ,\ pages 107165 ( year 2025 ) NoStop

    arXiv 2025

  8. [8]

    Oh , author S

    author author J. Oh , author S. Y. \ Cho , author S.-B. \ Yun , author E. Park , \ and\ author Y. Hong ,\ title title Separable Physics-Informed Neural Networks for Solving the BGK Model of the Boltzmann Equation , \ 10.1137/24M1668809 journal journal SIAM J. Sci. Comput. \ volume 47 ,\ pages C451--C474 ( year 2025 ) NoStop

    doi:10.1137/24m1668809

Show all 26 references
  1. [9]

    Xu , author Z

    author author S. Xu , author Z. Sun , author R. Huang , author D. Guo , author G. Yang , \ and\ author S. Ju ,\ title title A practical approach to flow field reconstruction with sparse or incomplete data through physics informed neural network , \ 10.1007/s10409-022-22302-x j...

    doi

  2. [10]

    \ Peng , author Z.-Q

    author author J.-Z. \ Peng , author Z.-Q. \ Wang , author X. Rong , author M. Mei , author M. Wang , author Y. He , \ and\ author W.-T. \ Wu ,\ title title Rapid and sparse reconstruction of high-speed steady-state and transient compressible flow fields using physics-informed ...

    doi

  3. [11]

    M \"u ller ,\ title title Hydrodynamics of core-collapse supernovae and their progenitors , \ 10.1007/s41115-020-0008-5 journal journal Living Rev

    author author B. M \"u ller ,\ title title Hydrodynamics of core-collapse supernovae and their progenitors , \ 10.1007/s41115-020-0008-5 journal journal Living Rev. Comput. Astrophys. \ volume 6 ,\ pages 3 ( year 2020 ) NoStop

    2020 doi

  4. [12]

    Kifonidis , author T

    author author K. Kifonidis , author T. Plewa , author H.-T. \ Janka , \ and\ author E. M \"u ller ,\ title title Nucleosynthesis and clump formation in a core-collapse supernova , \ 10.1086/312541 journal journal ApJ \ volume 531 ,\ pages L123 ( year 2000 ) NoStop

    doi

  5. [13]

    Atzeni \ and\ author J

    author author S. Atzeni \ and\ author J. Meyer-ter Vehn ,\ 10.1093/acprof:oso/9780198562641.001.0001 title The physics of inertial fusion: beam plasma interaction, hydrodynamics, hot dense matter ,\ Vol.\ volume 125 \ ( publisher OUP Oxford ,\ year 2004 ) NoStop

  6. [14]

    Betti \ and\ author O

    author author R. Betti \ and\ author O. Hurricane ,\ title title Inertial-confinement fusion with lasers , \ 10.1038/nphys3736 journal journal Nat. Phys. \ volume 12 ,\ pages 435--448 ( year 2016 ) NoStop

    2016 doi

  7. [15]

    Yang , author T

    author author J. Yang , author T. Kubota , \ and\ author E. E. \ Zukoski ,\ title title Applications of shock-induced mixing to supersonic combustion , \ 10.2514/3.11696 journal journal AIAA journal \ volume 31 ,\ pages 854--862 ( year 1993 ) NoStop

    doi

  8. [16]

    author author F. E. \ Marble , author G. J. \ Hendricks , \ and\ author E. E. \ Zukoski ,\ 10.1007/978-1-4613-9631-4_43 title Progress toward shock enhancement of supersonic combustion processes \ ( publisher Springer ,\ year 1989 ) NoStop

    doi

  9. [17]

    Wang , author Y

    author author S. Wang , author Y. Teng , \ and\ author P. Perdikaris ,\ title title Understanding and Mitigating Gradient Flow Pathologies in Physics-Informed Neural Networks , \ 10.1137/20M1318043 journal journal SIAM J. Sci. Comput. \ volume 43 ,\ pages A3055--A3081 ( year 2...

    doi

  10. [18]

    Mao , author A

    author author Z. Mao , author A. D. \ Jagtap , \ and\ author G. E. \ Karniadakis ,\ title title Physics-informed neural networks for high-speed flows , \ 10.1016/j.cma.2019.112789 journal journal Comput. Methods Appl. Mech. Eng. \ volume 360 ,\ pages 112789 ( year 2020 ) NoStop

    2019

  11. [19]

    Lu , author X

    author author L. Lu , author X. Meng , author Z. Mao , \ and\ author G. E. \ Karniadakis ,\ title title DeepXDE: A Deep Learning Library for Solving Differential Equations , \ 10.1137/19M1274067 journal journal SIAM Rev. \ volume 63 ,\ pages 208--228 ( year 2021 ) NoStop

    doi

  12. [20]

    author author A. D. \ Jagtap , author E. Kharazmi , \ and\ author G. E. \ Karniadakis ,\ title title Conservative physics-informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems , \ 10.1016/j.cma.2020.113028 journal jour...

  13. [21]

    author author A. D. Jagtap \ and\ author G. Em Karniadakis ,\ title title Extended Physics-Informed Neural Networks (XPINNs): A Generalized Space-Time Domain Decomposition Based Deep Learning Framework for Nonlinear Partial Differential Equations , \ 10.4208/cicp.OA-2020-0164 ...

    doi

  14. [22]

    Yao , author J

    author author Y. Yao , author J. Guo , \ and\ author T. Gu ,\ title title A deep learning method for multi-material diffusion problems based on physics-informed neural networks , \ 10.1016/j.cma.2023.116395 journal journal Comput. Methods Appl. Mech. Eng. \ volume 417 ,\ pages...

    2023

  15. [23]

    Liu , author S

    author author L. Liu , author S. Liu , author H. Xie , author F. Xiong , author T. Yu , author M. Xiao , author L. Liu , \ and\ author H. Yong ,\ title title Discontinuity Computing Using Physics-Informed Neural Networks , \ https://doi.org/10.1007/s10915-023-02412-1 journal j...

    doi

  16. [24]

    Ferrer-Sánchez , author J

    author author A. Ferrer-Sánchez , author J. D. \ Martín-Guerrero , author R. R. \ de Austri-Bazan , author A. Torres-Forné , \ and\ author J. A. \ Font ,\ title title Gradient-annihilated PINNs for solving Riemann problems: Application to relativistic hydrodynamics , \ 10.1016...

  17. [25]

    Paszke , author S

    author author A. Paszke , author S. Gross , author F. Massa , author A. Lerer , author J. Bradbury , author G. Chanan , author T. Killeen , author Z. Lin , author N. Gimelshein , author L. Antiga , et al. ,\ title title Pytorch: An imperative style, high-performance deep learn...

    2019

  18. [26]

    author author D. A. \ Bezgin , author A. B. \ Buhendwa , \ and\ author N. A. \ Adams ,\ title title JAX-Fluids: A fully-differentiable high-order computational fluid dynamics solver for compressible two-phase flows , \ 10.1016/j.cpc.2022.108527 journal journal Comput. Phys. Co...

Pith tools

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