Pith. sign in

REVIEW 4 major objections 4 minor 59 references

Physics-informed Fourier Basis Neural Network for Fluid Mechanics

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A physics-informed neural network that embeds input coordinates in a Fourier basis solves Burgers and Helmholtz equations more accurately than plain ANN and PINN baselines, and its accuracy is robust to the choice of activation function.

desk verdict Unreadable copy; the abstract alone shows a low-novelty PINN variant with strong but unquantified claims, though the activation-robustness angle is worth checking if a clean PDF exists. read the letter →

arxiv 2508.02166 v1 pith:3EYCUOEO submitted 2025-08-04 physics.flu-dyn

Chao Wang , Shilong Li , Zelong Yuan , Chunyu Guo This is my paper · ORCID
classification physics.flu-dyn
keywords physics-informedneuralnetworksFourierbasisBurgersequationHelmholtzactivationfunctionssparsedatareconstructionpartialdifferentialequationscomputationalfluiddynamics
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 proposes a physics-informed Fourier basis neural network (FBNN) that feeds input coordinates through a Fourier series embedding before the network layers, then trains with PDE residuals as soft constraints. The authors aim to show that this embedding gives the network strong nonlinear fitting and periodic modeling ability, so it solves the Burgers equation with discontinuities and the Helmholtz equation with strong periodicity more accurately than a conventional artificial neural network and a standard physics-informed neural network. They also report that, on sparse distributed sensor data, FBNN reconstructs the whole flow field better than the ANN baseline, and that changing the activation function changes FBNN accuracy only mildly. The broader point is that an architecture with built-in frequency structure can reduce a practitioner's dependence on choosing the right activation and on dense training data.

What carries the argument

The central object is the Fourier basis embedding: an improved Fourier series that maps each input coordinate to a vector of sine and cosine features before they enter the hidden layers of a fully connected network. A physics-informed residual loss then penalizes violation of the governing PDE, along with initial and boundary conditions, at collocation points. The Fourier embedding carries the argument by giving the network a prior that the solution is composed of periodic modes, which is the structure needed to represent sharp wave fronts and periodic fields with fewer trainable parameters.

What would settle it

Run the same Burgers and Helmholtz experiments with a broad range of activations (sigmoid, tanh, ReLU, ELU, SiLU, and a sinusoid) and record the spread of FBNN errors; if the spread is as large as that of PINN, the activation-insensitivity claim fails. Separately, repeat the sparse-data reconstruction with randomly placed sensor locations, rather than the paper's chosen set, and check whether FBNN still beats the ANN baseline.

Watch

Extended reading notes

Core claim

The central claim is that embedding a Fourier basis at the input layer of a physics-informed network, rather than relying on the network to learn frequency content from scratch, yields superior reconstruction of canonical fluid-mechanics PDEs, especially when the solution is discontinuous or strongly periodic. On the Burgers equation with discontinuous solutions and the Helmholtz equation with strong periodicity, the proposed FBNN outperforms both a plain ANN and a conventional PINN under matched settings. The paper further claims that when only sparse distributed data are available, FBNN reconstructs the full flow field more directly and accurately than the ANN, and that this advantage comes from the combination of the Fourier embedding and the physics loss, not from a specific activation function.

Load-bearing premise

The findings assume that the small set of activation functions compared is representative enough to prove that FBNN is not highly sensitive to activation choice, and that the sparse sensor locations do not favor the Fourier basis.

Editorial extensions

If this is right

  • If the claims hold, FBNN can serve as a default architecture for PDEs with oscillatory or shock-like solutions, reducing the need to hand-tune activation functions.
  • The sparse-data result implies that flow-field reconstruction from a few sensors can be done with physics constraints plus Fourier features, which matters for experimental data assimilation.
  • The activation-robustness result, if it generalizes, means the architecture shifts the burden from activation selection to frequency content, simplifying network design for CFD.
  • The approach extends naturally to other canonical PDEs with periodic or wave-like behavior, so the same embedding may improve PINN performance on those problems as well.

Reading between the lines

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

  • The reported activation insensitivity is only demonstrated on the few activations the paper tests; a practitioner should verify robustness on activations such as sigmoid, ReLU, and SiLU before relying on it in a new problem.
  • The sparse-data advantage could be sensitive to sensor placement: if the sensor locations are chosen where the Fourier basis is informative, the comparison may overstate the method's advantage over ANN.
  • A natural testable extension is to replace the fixed Fourier embedding with learnable frequencies, letting the network adapt the basis to each PDE while keeping the same physics-informed loss.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes a physics-informed Fourier basis neural network (FBNN) for solving canonical PDEs in fluid mechanics. The abstract claims that the method has strong nonlinear fitting capability and exceptional periodic modeling performance, that it is particularly advantageous for the Burgers equation with discontinuous solutions and the Helmholtz equation with strong periodicity, that it is directly superior to conventional artificial neural networks when reconstructing fields from sparse distributed data, and that its performance is not highly sensitive to activation function choice. The supplied full text is largely corrupted and unreadable, so the method definition, equations, figures, tables, and numerical results cannot be inspected.

Significance. If the claims are substantiated, the contribution is a plausible incremental improvement: embedding inputs in a Fourier basis inside a physics-informed neural network is a natural extension of Fourier feature methods, and the chosen benchmark problems (Burgers, Helmholtz, sparse reconstruction) are standard. The paper does not currently provide machine-checked proofs, reproducible code, or even readable numerical results, so its evidentiary value cannot be assessed from the supplied version. For a journal submission, the central claims require quantitative comparisons with matched baselines and a clear method description, none of which are currently accessible.

major comments (4)
  1. [Full text / Body] The body of the manuscript is corrupted beyond readability: equations, figure captions, table entries, and result descriptions appear as garbled characters. As a result, the Fourier basis embedding, the network architecture, the loss function, and the error metrics cannot be verified. This is load-bearing for every central claim in the abstract, because none of the method or evidence can be checked.
  2. [Abstract] The abstract reports "significant advantages" and "direct superiority" without a single numeric result. I need at least the final L2 relative errors, convergence comparisons, and a clear statement of baseline matching (network width, depth, parameter count, training data, and optimizer) for FBNN versus ANN and PINN. Without these numbers, the statements "intuitively validated" and "proved" are not testable claims.
  3. [Abstract / Activation-robustness claim] The claim that performance is "not highly sensitive to the choice of activation functions" is unsupported because the abstract does not enumerate which activation functions were compared, and the body text is unreadable. If only two or three similar activations were tested, the conclusion would not generalize. Please list the activation functions and report the quantitative spread of errors across them, ideally with multiple random seeds.
  4. [Results / Sparse reconstruction] The sparse-data reconstruction experiment needs a precise specification of how sensor locations were selected, how many sensors were used, and whether the same sensor set, initialization, and training schedule were used for FBNN and the ANN baseline. If the sensor locations were chosen to favor a Fourier basis, the claimed superiority would be an artifact rather than a general property. The current text does not rule this out, so the comparison cannot be evaluated.
minor comments (4)
  1. [Abstract] The final sentence contains a typo: "hese results highlightthe potential of PIFBNN" should presumably read "These results highlight the potential of FBNN," and the acronym is inconsistent because the paper otherwise uses FBNN.
  2. [Abstract] The phrases "intuitively validated" and "proved" are imprecise; please replace them with explicit quantitative comparisons and statistical statements (e.g., means and standard deviations over multiple runs).
  3. [Abstract / Sparse data] The term "sparse distributed data" should be quantified: specify the number of points, their spatial distribution, and the amount of noise, if any.
  4. [Related work] The manuscript should cite and discuss existing Fourier feature networks and randomized Fourier feature PINNs, since the proposed method appears closely related to those lines of work.

Circularity Check

0 steps flagged · score 0.0 of 10

No load-bearing circularity; the abstract makes empirical benchmark claims with no derivation chain that reduces to its own inputs.

full rationale

The only readable, in-scope evidence is the abstract, which reports empirical comparisons for the Burgers equation, the Helmholtz equation, sparse-data reconstruction, and activation-function sensitivity. It contains no equations, no fitted parameters, no self-citations, and no invoked uniqueness theorems. Consequently, none of the seven circularity patterns can be exhibited: there is no Eq. X = Eq. Y by construction, no fitted input renamed as a prediction, no self-citation chain used as load-bearing proof, and no known result merely renamed. The corrupted full text cannot supply quotable reductions, and its unreadability is an evidence-availability problem rather than a circularity finding. Concerns about whether the tested activation functions are representative or whether sparse sensor locations were chosen fairly are validity risks, not circularity. The paper is therefore scored 0 for circularity.

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

Because the supplied body text is an unreadable encoding corruption, this ledger can only reflect assumptions that are visible in the abstract or are part of the standard PINN machinery the abstract invokes. The three free parameters listed are the main degrees of freedom that the abstract leaves unstated yet that any reproduction would have to choose; the axioms are the background assumptions of benchmark-verified, physics-informed network training. If the paper's improved Fourier series introduces learned frequencies or extra terms, additional parameters would belong here, but that content cannot be assessed from the available text.

free parameters (3)
  • Fourier expansion truncation order = not stated in abstract
    A finite Fourier basis requires choosing the number of harmonic terms; the choice controls accuracy versus cost and is not reported in the abstract, making it a per-problem fitted hyperparameter.
  • Physics-residual versus data-loss weights = not stated in abstract
    The training loss balances the PDE residual term against the sparse-data reconstruction term, and these weights are typically hand-tuned per benchmark; neither the values nor the tuning protocol appear in the abstract.
  • Network architecture hyperparameters (width and depth) = not stated in abstract
    The comparative claim against ANN and PINN baselines depends on matched network capacity; the chosen sizes are not reported in the abstract.
assumptions (3)
  • domain assumption The reference solutions used to score errors (analytic solutions of Burgers and Helmholtz) are computed accurately, including across the Burgers shock.
    All superiority claims are measured against reference solutions; if the shock region is under-resolved in the reference, the reported advantages would be artifacts. The error protocol is not visible in the abstract or readable body.
  • domain assumption The PDE residual, computed by automatic differentiation, is a valid and sufficient training signal when combined with or instead of data points.
    This is the standard physics-informed training paradigm the paper inherits; the abstract invokes it by incorporating physical information and by the sparse-data reconstruction experiment.
  • standard math The optimization procedure (gradient descent on the composite loss) reaches solutions representative of the reported minima in all compared networks.
    All neural PDE papers rely on the optimizer succeeding; no convergence guarantees or seed statistics are reported in the abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-informed Fourier Basis Neural Network for Fluid Mechanics." pith.science (2026). https://pith.science/paper/3EYCUOEO

@misc{pith2026250802166,
  author       = {Pith},
  title        = {Pith review of: Physics-informed Fourier Basis Neural Network for Fluid Mechanics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3EYCUOEO}},
  note         = {Machine review of arXiv:2508.02166}
}
read the original abstract

Solving partial differential equations (PDEs) is an important yet challenging task in fluid mechanics. In this study, we embed an improved Fourier series into neural networks and propose a physics-informed Fourier basis neural network (FBNN) by incorporating physical information to solve canonical PDEs in fluid mechanics. The results demonstrated that the proposed framework exhibits a strong nonlinear fitting capability and exceptional periodic modeling performance. In particular, our model shows significant advantages for the Burgers equation with discontinuous solutions and Helmholtz equation with strong periodicity. By directly introducing sparse distributed data to reconstruct the entire flow field, we further intuitively validated the direct superiority of FBNN over conventional artificial neural networks (ANN) as well as the benefits of incorporating physical information into the network. By adjusting the activation functions of networks and comparing with an ANN and conventional physics-informed neural network, we proved that performance of the proposed FBNN architecture is not highly sensitive to the choice of activation functions. The nonlinear fitting capability of FBNN avoids excessive reliance on activation functions, thereby mitigating the risk of suboptimal outcomes or training failures stemming from unsuitable activation function choices.hese results highlightthe potential of PIFBNN as a powerful tool in computational fluid dynamics.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 50 canonical work pages

  1. [1]

    Zhang , author L

    author Z. Zhang , author L. Hui , author Z. Feng , et al., title Application of cfd in ship engineering design practice and ship hydrodynamics , journal Journal of Hydrodynamics, Ser. B volume 18 ( year 2006 ) pages 315--322

    work page 2006

  2. [2]

    Phillips , author S

    author A. Phillips , author S. Turnock , author M. Furlong , title The use of computational fluid dynamics to aid cost-effective hydrodynamic design of autonomous underwater vehicles , journal Proceedings of the Institution of Mechanical Engineers, Part M: Journal of Engineering for the Maritime Environment volume 224 ( year 2010 ) pages 239--254

    work page 2010

  3. [3]

    Zhang , author H

    author Y. Zhang , author H. Chen , author K. Wang , author M. Wang , title Aeroacoustic prediction of a multi-element airfoil using wall-modeled large-eddy simulation , journal AIAA Journal volume 55 ( year 2017 ) pages 4219--4233

    work page 2017

  4. [4]

    Kou , author W

    author J. Kou , author W. Zhang , title Data-driven modeling for unsteady aerodynamics and aeroelasticity , journal Progress in Aerospace Sciences volume 125 ( year 2021 ) pages 100725

    work page 2021

  5. [5]

    Mani , author A

    author M. Mani , author A. J. Dorgan , title A perspective on the state of aerospace computational fluid dynamics technology , journal Annual Review of Fluid Mechanics volume 55 ( year 2023 ) pages 431--457

    work page 2023

  6. [6]

    Xia , author Y

    author Z. Xia , author Y. Song , author X. Zhu , title Skin-friction scaling in compressible turbulent channel flows , journal AIAA Journal ( year 2025 ) pages 1--7

    work page 2025

  7. [7]

    Jeong , author J

    author W. Jeong , author J. Seong , title Comparison of effects on technical variances of computational fluid dynamics (cfd) software based on finite element and finite volume methods , journal International Journal of Mechanical Sciences volume 78 ( year 2014 ) pages 19--26

    work page 2014

  8. [8]

    Moin , author K

    author P. Moin , author K. Mahesh , title Direct numerical simulation: a tool in turbulence research , journal Annual review of fluid mechanics volume 30 ( year 1998 ) pages 539--578

    work page 1998

Show all 59 references
  1. [9]

    Scardovelli , author S

    author R. Scardovelli , author S. Zaleski , title Direct numerical simulation of free-surface and interfacial flow , journal Annual review of fluid mechanics volume 31 ( year 1999 ) pages 567--603

    1999

  2. [10]

    Zhu , author Y

    author X. Zhu , author Y. Song , author P. Zhang , author X. Yang , author Y. Ji , author Z. Xia , title Influences of streamwise driving forces on turbulent statistics in direct numerical simulations of compressible turbulent channel flows , journal Physical Review Fluids vol...

    2025

  3. [11]

    Duraisamy , author G

    author K. Duraisamy , author G. Iaccarino , author H. Xiao , title Turbulence modeling in the age of data , journal Annual review of fluid mechanics volume 51 ( year 2019 ) pages 357--377

    2019

  4. [12]

    author S. L. Brunton , author B. R. Noack , author P. Koumoutsakos , title Machine learning for fluid mechanics , journal Annual review of fluid mechanics volume 52 ( year 2020 ) pages 477--508

    2020

  5. [13]

    Lei , author Y

    author M. Lei , author Y. Zhang , title Generating airfoils from text: Foilclip, a novel framework for language-conditioned aerodynamic design , journal Theoretical and Applied Mechanics Letters ( year 2025 ) pages 100602

    2025

  6. [14]

    Huang , author W

    author S. Huang , author W. Feng , author C. Tang , author Z. He , author C. Yu , author J. Lv , title Partial differential equations meet deep neural networks: A survey , journal IEEE Transactions on Neural Networks and Learning Systems ( year 2025 )

    2025

  7. [15]

    Wu , author Y

    author C. Wu , author Y. Zhang , title Enhancing the shear-stress-transport turbulence model with symbolic regression: A generalizable and interpretable data-driven approach , journal Physical Review Fluids volume 8 ( year 2023 ) pages 084604

    2023

  8. [16]

    Ling , author A

    author J. Ling , author A. Kurzawski , author J. Templeton , title Reynolds averaged turbulence modelling using deep neural networks with embedded invariance , journal Journal of Fluid Mechanics volume 807 ( year 2016 ) pages 155--166

    2016

  9. [17]

    Gamahara , author Y

    author M. Gamahara , author Y. Hattori , title Searching for turbulence models by artificial neural network , journal Physical Review Fluids volume 2 ( year 2017 ) pages 054604

    2017

  10. [18]

    Xu , author D

    author H. Xu , author D. Zhang , author J. Zeng , title Deep-learning of parametric partial differential equations from sparse and noisy data , journal Physics of Fluids volume 33 ( year 2021 )

    2021

  11. [19]

    Yang , author Y

    author H. Yang , author Y. Wang , author J. Wang , title Implicit factorized transformer approach to fast prediction of turbulent channel flows , journal arXiv preprint arXiv:2412.18840 ( year 2024 )

    2024 arXiv

  12. [20]

    Fan , author D

    author X. Fan , author D. Akhare , author J.-X. Wang , title Neural differentiable modeling with diffusion-based super-resolution for two-dimensional spatiotemporal turbulence , journal Computer Methods in Applied Mechanics and Engineering volume 433 ( year 2025 ) pages 117478

    2025

  13. [21]

    Suo , author W

    author W. Suo , author W. Zhang , title A novel paradigm for solving pdes: multi-scale neural computing , journal Acta Mechanica Sinica volume 41 ( year 2025 ) pages 324172

    2025

  14. [22]

    Wu , author D

    author K. Wu , author D. Xiu , title Data-driven deep learning of partial differential equations in modal space , journal Journal of Computational Physics volume 408 ( year 2020 ) pages 109307

    2020

  15. [23]

    Li , author N

    author Z. Li , author N. Kovachki , author K. Azizzadenesheli , author B. Liu , author K. Bhattacharya , author A. Stuart , author A. Anandkumar , title Fourier neural operator for parametric partial differential equations , journal arXiv preprint arXiv:2010.08895 ( year 2020 )

    2010 arXiv

  16. [24]

    Peng , author Z

    author W. Peng , author Z. Yuan , author J. Wang , title Attention-enhanced neural network models for turbulence simulation , journal Physics of Fluids volume 34 ( year 2022 )

    2022

  17. [25]

    Peng , author Z

    author W. Peng , author Z. Yuan , author Z. Li , author J. Wang , title Linear attention coupled fourier neural operator for simulation of three-dimensional turbulence , journal Physics of Fluids volume 35 ( year 2023 )

    2023

  18. [26]

    Wen , author Z

    author G. Wen , author Z. Li , author K. Azizzadenesheli , author A. Anandkumar , author S. M. Benson , title U-fno—an enhanced fourier neural operator-based deep-learning model for multiphase flow , journal Advances in Water Resources volume 163 ( year 2022 ) pages 104180

    2022

  19. [27]

    You , author Q

    author H. You , author Q. Zhang , author C. J. Ross , author C.-H. Lee , author Y. Yu , title Learning deep implicit fourier neural operators (ifnos) with applications to heterogeneous material modeling , journal Computer Methods in Applied Mechanics and Engineering volume 398...

    2022

  20. [28]

    Li , author W

    author Z. Li , author W. Peng , author Z. Yuan , author J. Wang , title Long-term predictions of turbulence by implicit u-net enhanced fourier neural operator , journal Physics of Fluids volume 35 ( year 2023 )

    2023

  21. [29]

    Li , author T

    author Z. Li , author T. Liu , author W. Peng , author Z. Yuan , author J. Wang , title A transformer-based neural operator for large-eddy simulation of turbulence , journal Physics of Fluids volume 36 ( year 2024 )

    2024

  22. [30]

    Agatonovic-Kustrin , author R

    author S. Agatonovic-Kustrin , author R. Beresford , title Basic concepts of artificial neural network (ann) modeling and its application in pharmaceutical research , journal Journal of pharmaceutical and biomedical analysis volume 22 ( year 2000 ) pages 717--727

    2000

  23. [31]

    Lu , author P

    author L. Lu , author P. Jin , author G. Pang , author Z. Zhang , author G. E. Karniadakis , title Learning nonlinear operators via deeponet based on the universal approximation theorem of operators , journal Nature machine intelligence volume 3 ( year 2021 ) pages 218--229

    2021

  24. [32]

    author F. O. de Fran c a , title A greedy search tree heuristic for symbolic regression , journal Information Sciences volume 442 ( year 2018 ) pages 18--32

    2018

  25. [33]

    Raissi , author P

    author M. Raissi , author P. Perdikaris , author G. E. Karniadakis , title Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , journal Journal of Computational physics volume ...

    2019

  26. [34]

    Blechschmidt , author O

    author J. Blechschmidt , author O. G. Ernst , title Three ways to solve partial differential equations with neural networks—a review , journal GAMM-Mitteilungen volume 44 ( year 2021 ) pages e202100006

    2021

  27. [35]

    Chen , author Q

    author W. Chen , author Q. Wang , author J. S. Hesthaven , author C. Zhang , title Physics-informed machine learning for reduced-order modeling of nonlinear problems , journal Journal of computational physics volume 446 ( year 2021 ) pages 110666

    2021

  28. [36]

    Yang , author X

    author L. Yang , author X. Meng , author G. E. Karniadakis , title B-pinns: Bayesian physics-informed neural networks for forward and inverse pde problems with noisy data , journal Journal of Computational Physics volume 425 ( year 2021 ) pages 109913

    2021

  29. [37]

    Zhang , author C

    author S. Zhang , author C. Zhang , author X. Han , author B. Wang , title Mrf-pinn: a multi-receptive-field convolutional physics-informed neural network for solving partial differential equations , journal Computational Mechanics volume 75 ( year 2025 ) pages 1137--1163

    2025

  30. [38]

    Zhao , author Z

    author S. Zhao , author Z. Li , author B. Fan , author Y. Wang , author H. Yang , author J. Wang , title Lesnets (large-eddy simulation nets): Physics-informed neural operator for large-eddy simulation of turbulence , journal Journal of Computational Physics ( year 2025 ) pages 114125

    2025

  31. [39]

    Song , author W

    author J. Song , author W. Cao , author W. Zhang , title Fenn: Feature-enhanced neural network for solving partial differential equations involving fluid mechanics , journal arXiv preprint arXiv:2501.11957 ( year 2025 a )

    2025 arXiv

  32. [40]

    Song , author W

    author J. Song , author W. Cao , author F. Liao , author W. Zhang , title Vw-pinns: A volume weighting method for pde residuals in physics-informed neural networks , journal Acta Mechanica Sinica volume 41 ( year 2025 b ) pages 324140

    2025

  33. [41]

    Schwenk , author Y

    author H. Schwenk , author Y. Bengio , title Training methods for adaptive boosting of neural networks , journal Advances in neural information processing systems volume 10 ( year 1997 )

    1997

  34. [42]

    Srinivasulu , author A

    author S. Srinivasulu , author A. Jain , title A comparative analysis of training methods for artificial neural network rainfall--runoff models , journal Applied Soft Computing volume 6 ( year 2006 ) pages 295--306

    2006

  35. [43]

    Dai , author X

    author Q. Dai , author X. Shen , author L. Zhang , author Q. Li , author D. Wang , title Adversarial training methods for network embedding , in: booktitle The world wide web conference , year 2019 , pp. pages 329--339

    2019

  36. [44]

    Li , author H

    author H. Li , author H. Zhang , author X. Qi , author R. Yang , author G. Huang , title Improved techniques for training adaptive deep networks , in: booktitle Proceedings of the IEEE/CVF international conference on computer vision , year 2019 , pp. pages 1891--1900

    2019

  37. [45]

    author J. O. Donoghue , author M. Roantree , title A framework for selecting deep learning hyper-parameters , in: booktitle British international conference on databases , organization Springer , year 2015 , pp. pages 120--132

    2015

  38. [46]

    author S. C. Smithson , author G. Yang , author W. J. Gross , author B. H. Meyer , title Neural networks designing neural networks: multi-objective hyper-parameter optimization , in: booktitle 2016 IEEE/ACM International Conference on Computer-Aided Design (ICCAD) , organizati...

    2016

  39. [47]

    Zhang , author W

    author W. Zhang , author W. Suo , author J. Song , author W. Cao , title Physics informed neural networks (pinns) as intelligent computing technique for solving partial differential equations: Limitation and future prospects , journal arXiv preprint arXiv:2411.18240 ( year 2024 )

    2024 arXiv

  40. [48]

    Cao , author W

    author W. Cao , author W. Zhang , title An analysis and solution of ill-conditioning in physics-informed neural networks , journal Journal of Computational Physics volume 520 ( year 2025 ) pages 113494

    2025

  41. [49]

    Jin , author S

    author X. Jin , author S. Cai , author H. Li , author G. E. Karniadakis , title Nsfnets (navier-stokes flow nets): Physics-informed neural networks for the incompressible navier-stokes equations , journal Journal of Computational Physics volume 426 ( year 2021 ) pages 109951

    2021

  42. [50]

    Wang , author P

    author S. Wang , author P. Perdikaris , title Long-time integration of parametric evolution equations with physics-informed deeponets , journal Journal of Computational Physics volume 475 ( year 2023 ) pages 111855

    2023

  43. [51]

    Rao , author P

    author C. Rao , author P. Ren , author Q. Wang , author O. Buyukozturk , author H. Sun , author Y. Liu , title Encoding physics to learn reaction--diffusion processes , journal Nature Machine Intelligence volume 5 ( year 2023 ) pages 765--779

    2023

  44. [52]

    Wang , author A

    author S. Wang , author A. K. Bhartari , author B. Li , author P. Perdikaris , title Gradient alignment in physics-informed neural networks: A second-order optimization perspective , journal arXiv preprint arXiv:2502.00604 ( year 2025 )

    2025

  45. [53]

    Xiao , author L

    author Y. Xiao , author L. Yang , author C. Shu , author H. Dong , author Y. Du , author Y. Song , title Least-square finite difference-based physics-informed neural network for steady incompressible flows , journal Computers & Mathematics with Applications volume 175 ( year 2...

    2024

  46. [54]

    Liu , author Y

    author Z. Liu , author Y. Wang , author S. Vaidya , author F. Ruehle , author J. Halverson , author M. Solja c i \'c , author T. Y. Hou , author M. Tegmark , title Kan: Kolmogorov-arnold networks , journal arXiv preprint arXiv:2404.19756 ( year 2024 )

    2024 arXiv

  47. [55]

    Guo , author L

    author C. Guo , author L. Sun , author S. Li , author Z. Yuan , author C. Wang , title Physics-informed kolmogorov-arnold network with chebyshev polynomials for fluid mechanics , journal arXiv preprint arXiv:2411.04516 ( year 2024 )

    2024 arXiv

  48. [56]

    author A. R. Gallant , title The fourier flexible form , journal American Journal of Agricultural Economics volume 66 ( year 1984 ) pages 204--208

    1984

  49. [57]

    Dong , author G

    author Y. Dong , author G. Li , author Y. Tao , author X. Jiang , author K. Zhang , author J. Li , author J. Deng , author J. Su , author J. Zhang , author J. Xu , title Fan: Fourier analysis networks , journal arXiv preprint arXiv:2410.02675 ( year 2024 )

  50. [58]

    Lu , author X

    author L. Lu , author X. Meng , author Z. Mao , author G. E. Karniadakis , title DeepXDE : A deep learning library for solving differential equations , journal SIAM Review volume 63 ( year 2021 ) pages 208--228 . :10.1137/19M1274067

    2021 doi

  51. [59]

    Karniadakis , author S

    author G. Karniadakis , author S. J. Sherwin , title Spectral/hp element methods for computational fluid dynamics , publisher Oxford University Press , year 2005

    2005

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.