pith. sign in

arxiv: 2605.18340 · v1 · pith:IWZJV5DMnew · submitted 2026-05-18 · ⚛️ physics.comp-ph · physics.flu-dyn

Physics Informed Neural Network-based Computational Method for Accelerating Time-Periodic Unsteady CFD Simulations

Lakshya Chaplot , Harshita Agarwal , Atul Sharma This is my paper

Pith reviewed 2026-05-19 23:49 UTC · model grok-4.3

classification ⚛️ physics.comp-ph physics.flu-dyn
keywords Physics Informed Neural NetworkComputational Fluid DynamicsTime-periodic flowsUnsteady simulationsMeshless methodsPeriodic boundary conditions
0
0 comments X

The pith

Physics-informed neural network computes time-periodic flows by training over one period instead of simulating transients.

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

The paper proposes a PINN-based CFD solver that obtains the periodic state directly by optimizing a neural network over a single time period using a physics-informed loss. Traditional unsteady solvers must start from initial conditions and advance through a long non-periodic transient phase before periodicity appears, which the new method avoids. Proof-of-concept results for 2D periodic heat diffusion and fluid flow show nearly the same accuracy as the conventional transient-to-periodic approach but with substantially lower computational cost. The study also reports how the number of collocation points, network architecture, and numerical differentiation spacing affect runtime and error. The core idea is to restrict the training window to the periodic regime rather than the full temporal domain from t=0.

Core claim

By training the neural network parameters to satisfy the unsteady governing equations only inside one time period, the PINN converges to the correct periodic solution without ever computing the initial transient evolution from arbitrary starting conditions.

What carries the argument

Physics-informed neural network whose loss is evaluated solely over one time period using collocation points and automatic differentiation to enforce the governing equations.

If this is right

  • Computational time to reach a periodic flow state drops substantially relative to the transient-to-periodic CFD workflow.
  • Accuracy remains comparable to the traditional solver on the 2D heat diffusion and fluid flow verification cases.
  • Hyperparameter choices such as collocation point density, network size, and point spacing directly control the tradeoff between speed and error.
  • The meshless periodic formulation applies to both pure heat diffusion and incompressible flow problems.

Where Pith is reading between the lines

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

  • The same single-period training strategy could be tested on three-dimensional or turbulent periodic flows to check scalability.
  • A hybrid workflow that uses the PINN only for the periodic state and a conventional solver for any initial transients might combine the strengths of both approaches.
  • Problems with known analytical periodic solutions would allow direct quantitative checks on whether the network converges to the exact periodic limit.

Load-bearing premise

Optimizing a neural network with physics loss over only one time period is sufficient to recover the true periodic state even though no transient data from initial conditions is provided.

What would settle it

Compare the PINN periodic solution against a reference obtained by running a conventional unsteady simulation for many periods; a clear mismatch in the established periodic regime would falsify the claim.

Figures

Figures reproduced from arXiv: 2605.18340 by Atul Sharma, Harshita Agarwal, Lakshya Chaplot.

Figure 1
Figure 1. Figure 1: Illustration of a transient solution reaching time-periodic unsteady state [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Pipeline of the proposed PINN-based CFD approach [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (a) Illustration of various types of randomly distributed points, where the present sampling strategy is restricted [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Computational stencil for numerical differentiation at collocation point [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Computational setup for a transient-to-periodic 2D heat diffusion in a square plate, with (b) square and (c) [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spatio-temporal temperature contour obtained from (a) analytical solution (Eqn. [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Collocation and boundary points (Nr = 1024 and NBC = 200) sampled in the spatio-temporal domain using the (a) LHS, (b) uniform random, (c) Sobol, and (d) Halton sampling techniques for solving Eqn.41 using the standard PINN and current PINN-based solvers PINN model exactly adheres to the prescribed boundary conditions of Eqn.42 [PITH_FULL_IMAGE:figures/full_fig_p020_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: (a) Transient-to-periodic analytical solution (Eqn. [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Spatio-temporal temperature contours obtained using the standard PINN [ [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Spatio-temporal temperature contours obtained using the present PINN-based solver ( [PITH_FULL_IMAGE:figures/full_fig_p023_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Collocation and boundary points sampled in the spatio-temporal domain using the LHS and the corresponding [PITH_FULL_IMAGE:figures/full_fig_p024_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: For the three 2D problems in Figure [PITH_FULL_IMAGE:figures/full_fig_p025_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Comparison between grid independent FVM-based transient solution and PINN-based periodic solution for [PITH_FULL_IMAGE:figures/full_fig_p026_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: For various number of layers nL = 2 − 8 of the feedforward neural network, variation of (a1, b1, c1) mean computational time of the PINN-based solver and its (a2, b2, c2) mean L¯2 error (with respect to the grid independent FVM solution), with increasing number of neurons per layer nN . The above results in subfigures (a1, a2), (b1, b2) and (c1, c2) correspond to the periodic results at points P, Q, R in … view at source ↗
Figure 15
Figure 15. Figure 15: For various number of boundary points NBC = 200 − 2400, variation of (a1, b1, c1) mean computational time of the PINN-based solver and its (a2, b2, c2) mean L¯2 error (with respect to the grid independent FVM solution), with increasing number of collocation points Nr. The above results in subfigures (a1, a2), (b1, b2) and (c1, c2) correspond to the periodic results at points P, Q, R in [PITH_FULL_IMAGE:f… view at source ↗
Figure 16
Figure 16. Figure 16: For various number of collocation points [PITH_FULL_IMAGE:figures/full_fig_p031_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Comparison of computational time and L¯2 error for the present PINN-based periodic and FVM-based transient-to-periodic diffusion solvers, on various collocation points Nr and grid sizes, respectively. The computational performance in subfigures (a), (b), and (c) corresponds to problems in [PITH_FULL_IMAGE:figures/full_fig_p033_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: For the periodically oscillating LDC flow problem in Figure [PITH_FULL_IMAGE:figures/full_fig_p036_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Comparison between grid-independent FVM-based transient solution and PINN-based periodic solution for [PITH_FULL_IMAGE:figures/full_fig_p037_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Comparison of computational time and L¯2 error for the present PINN-based periodic and FVM-based transient-to-periodic flow solvers on various collocation points Nr and grid sizes, respectively. The computational per￾formance in subfigures (a), (b), and (c) correspond to Re = 10, 50 and 100, respectively. Further, for both the flow solvers, the error is computed with respect to a grid-independent FVM-base… view at source ↗
Figure 21
Figure 21. Figure 21: (a) Computational setup for 2D transient diffusion in a square plate, (b) comparison of [PITH_FULL_IMAGE:figures/full_fig_p041_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: (a) Computational setup for 2D LDC flow and (b) comparison of present and published results on variation [PITH_FULL_IMAGE:figures/full_fig_p041_22.png] view at source ↗
read the original abstract

Presently, there is a steady state approach in Computational fluid dynamics (CFD) to obtain a steady solution directly from the steady state governing equations. Whereas, for obtaining a time-periodic flow solution, the present unsteady governing equations-based CFD approach starts from an initial condition and requires a large computational time during the initial non-periodic transient phase before reaching the periodic state. For obtaining the periodic flow directly, without transient simulations that may not be of interest, our objective is to propose a Physics Informed Neural Network (PINN)-based periodic CFD approach. The motivation is a substantial reduction in computational time by a meshless PINN-based periodic CFD solver as compared to the present mesh-based transient-to-periodic solver. Proof-of-concept, for the periodic CFD approach, is demonstrated here for 2D periodic heat diffusion and fluid flow problems. The proposed PINN-based periodic solver primarily focuses on the time-periodic state, optimizing the neural network model's trainable parameters to precisely fit a smaller time window (one time-period) rather than the temporal domain starting from the initial condition. After presenting a verification study, effect of the PINN-related various hyperparameters such as the number of collocation points, neural network architecture, and point spacing for numerical differentiation, on computational time and accuracy are presented. Our results demonstrate that the PINN-based periodic solver takes substantially less computational time to achieve almost same accuracy as that obtained by the traditional transient-to-periodic solver.

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

3 major / 2 minor

Summary. The manuscript proposes a Physics-Informed Neural Network (PINN) approach for directly computing time-periodic solutions in unsteady CFD by optimizing the network parameters over a single time period rather than simulating the full transient evolution from initial conditions. Proof-of-concept demonstrations are given for 2D periodic heat diffusion and fluid flow problems, together with studies on the effects of hyperparameters (collocation points, network architecture, differentiation spacing) on runtime and accuracy; the central claim is that the PINN solver achieves substantial computational-time savings while attaining nearly the same accuracy as conventional transient-to-periodic mesh-based solvers.

Significance. If the single-period optimization reliably recovers the physically realized limit-cycle solution, the method could provide a useful meshless route to accelerate periodic-flow computations. The hyperparameter sensitivity analysis supplies practical implementation guidance. However, the absence of quantitative error metrics and uniqueness checks substantially reduces the assessed significance of the reported time-accuracy trade-off.

major comments (3)
  1. [Abstract] Abstract: the assertion that the PINN solver achieves 'almost same accuracy' as the traditional transient-to-periodic solver is unsupported by any reported quantitative error measures (L2 norms, maximum pointwise errors, or convergence data) or side-by-side comparisons; this quantitative gap is load-bearing for the central time-reduction claim.
  2. [Fluid flow example] Fluid-flow demonstration: no test is presented showing that the obtained periodic field is independent of neural-network initialization or collocation-point distribution, leaving open the possibility that the optimizer converges to a PDE-satisfying but non-physical periodic state rather than the attractor reached from given initial data.
  3. [Method] Method section: the physics-informed loss is minimized strictly inside one period [0,T] with no transient history; the manuscript provides neither an analysis of solution uniqueness for the chosen governing equations nor a numerical check that the converged state matches the long-time periodic solution obtained by conventional time-marching.
minor comments (2)
  1. The abstract would be strengthened by stating the concrete time-reduction factors and the specific 2D test cases employed.
  2. Notation for the period T, the neural-network parameters, and the collocation-point sets should be introduced consistently before the results are discussed.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments on our manuscript. We address each of the major comments below and will make the necessary revisions to improve the clarity and rigor of the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that the PINN solver achieves 'almost same accuracy' as the traditional transient-to-periodic solver is unsupported by any reported quantitative error measures (L2 norms, maximum pointwise errors, or convergence data) or side-by-side comparisons; this quantitative gap is load-bearing for the central time-reduction claim.

    Authors: We agree with the referee that quantitative error metrics are essential to substantiate the accuracy claim. Although the manuscript includes verification studies and hyperparameter analyses, explicit L2 norms and pointwise error comparisons were not tabulated. In the revised manuscript, we will add a dedicated section or table presenting these quantitative measures for both the 2D heat diffusion and fluid flow cases, directly comparing the PINN results to the conventional solver outputs. This will provide concrete support for the 'almost same accuracy' statement. revision: yes

  2. Referee: [Fluid flow example] Fluid-flow demonstration: no test is presented showing that the obtained periodic field is independent of neural-network initialization or collocation-point distribution, leaving open the possibility that the optimizer converges to a PDE-satisfying but non-physical periodic state rather than the attractor reached from given initial data.

    Authors: This is a valid concern regarding the reliability of the optimization process. To address it, we will perform and report additional numerical experiments in the revision, using multiple random initializations of the network weights and different distributions of collocation points. We will demonstrate that the resulting periodic solutions converge to the same field (within numerical tolerance), consistent with the physical solution obtained via time-marching. This will help confirm that the method recovers the physically realized limit-cycle solution. revision: yes

  3. Referee: [Method] Method section: the physics-informed loss is minimized strictly inside one period [0,T] with no transient history; the manuscript provides neither an analysis of solution uniqueness for the chosen governing equations nor a numerical check that the converged state matches the long-time periodic solution obtained by conventional time-marching.

    Authors: We acknowledge that a formal uniqueness analysis is not provided in the current version. For the linear heat equation, uniqueness of the periodic solution can be established analytically, while for the nonlinear fluid flow equations, we rely on the physical expectation that the long-time behavior is unique. In the revised manuscript, we will include a direct numerical verification by comparing the PINN solution to the periodic state reached by the traditional transient simulation after many periods. Additionally, we will add a brief discussion on solution uniqueness for these specific problems based on existing literature for periodic flows. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained via direct PDE residual minimization

full rationale

The paper defines a PINN loss over a single period [0,T] that enforces the unsteady governing equations, boundary conditions, and time-periodicity directly on the network outputs. This construction solves the periodic boundary-value problem rather than fitting to precomputed data or renaming a prior result. Verification against independent transient-to-periodic CFD runs supplies an external benchmark, and no load-bearing step reduces to a self-citation chain, fitted parameter renamed as prediction, or ansatz smuggled from prior author work. The speedup claim rests on measured wall-clock comparisons, not on any definitional equivalence.

Axiom & Free-Parameter Ledger

3 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard PINN assumption that a neural network can satisfy the governing PDEs when trained with a physics residual loss, plus several tunable hyperparameters whose effects are studied but not derived from first principles.

free parameters (3)
  • Number of collocation points
    Hyperparameter controlling the density of points at which PDE residuals are enforced; its value is varied to study accuracy versus time.
  • Neural network architecture
    Choice of layers and neurons whose effect on fitting the periodic solution is examined.
  • Point spacing for numerical differentiation
    Controls the stencil used to approximate derivatives inside the PINN loss.
axioms (1)
  • domain assumption A neural network can be trained to satisfy the time-periodic governing equations when the loss is evaluated only over one period.
    This premise enables the entire periodic CFD approach by removing the need for transient simulation.

pith-pipeline@v0.9.0 · 5799 in / 1311 out tokens · 69071 ms · 2026-05-19T23:49:59.466005+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

46 extracted references · 46 canonical work pages · 1 internal anchor

  1. [1]

    Colonius, D

    T. Colonius, D. R. Williams, Control of vortex shedding on two- and three-dimensional aerofoils, Philosophical transactions. Series A, Mathematical, physical, and engineering sciences 369 (2011) 1525–1539.doi:10.1098/RSTA.2010.0355. URLhttps://pubmed.ncbi.nlm.nih.gov/21382829/

    work page doi:10.1098/rsta.2010.0355 2011

  2. [2]

    S. R. Morab, J. S. Murallidharan, A. Sharma, Computational hemodynamics and hemoacoustic study on carotid bifurcation: Effect of stenosis and branch angle, Physics of Fluids 36 (4 2024). doi:10.1063/5.0203193/3281081. URL/aip/pof/article/36/4/041906/3281081/Computational-hemodynamics-and-hemoacoustic-study

    work page doi:10.1063/5.0203193/3281081 2024

  3. [3]

    Mehrabi, S

    M. Mehrabi, S. Setayeshi, Computational fluid dynamics analysis of pulsatile blood flow behavior in modelled stenosed vessels with different severities, Mathematical Problems in Engineering 2012 (2012) 804765.doi:10.1155/2012/804765. URLhttps://onlinelibrary.wiley.com/doi/full/10.1155/2012/804765https: //onlinelibrary.wiley.com/doi/abs/10.1155/2012/804765...

    work page doi:10.1155/2012/804765 2012

  4. [4]

    doi:10.2514/6

    M. McMullen, A. Jameson, J. J. Alonso, Acceleration of convergence to a periodic steady state in turbomachinery flows, 39th Aerospace Sciences Meeting and Exhibit (2001).doi:10.2514/6. 2001-152. URLhttps://arc.aiaa.org/doi/10.2514/6.2001-152

    work page doi:10.2514/6 2001

  5. [5]

    Y. Cao, L. Zhou, C. Ou, H. Fang, D. Liu, 3d cfd simulation and analysis of transient flow in a water pipeline, Aqua Water Infrastructure, Ecosystems and Society 71 (2022) 751–767. doi:10.2166/AQUA.2022.023

    work page doi:10.2166/aqua.2022.023 2022

  6. [6]

    D. Xue, Y. Liu, L. Li, Heat transfer performance of wet porous solar collectors under periodic conditions, Heat Transfer 50 (2021) 7440–7453.doi:10.1002/HTJ.22237. URLhttps://onlinelibrary.wiley.com/doi/full/10.1002/htj.22237https:// onlinelibrary.wiley.com/doi/abs/10.1002/htj.22237https://onlinelibrary.wiley. com/doi/10.1002/htj.22237

    work page doi:10.1002/htj.22237 2021

  7. [7]

    J. D. Anderson, Computational Fluid Dynamics: The Basic with Application, Vol. 1, McGraw- Hill, 1995. URLhttps://books.google.com/books/about/Computational_Fluid_Dynamics.html?id= dJceAQAAIAAJ

    work page 1995

  8. [8]

    Sharma, Introduction to Computational Fluid Dynamics: Development, Application and Anal- ysis, Springer International Publishing, 2021.doi:10.1007/978-3-030-72884-7/COVER

    A. Sharma, Introduction to Computational Fluid Dynamics: Development, Application and Anal- ysis, Springer International Publishing, 2021.doi:10.1007/978-3-030-72884-7/COVER

    work page doi:10.1007/978-3-030-72884-7/cover 2021

  9. [9]

    M. R. Pfaller, J. Pham, N. M. Wilson, D. W. Parker, A. L. Marsden, On the periodicity of cardiovascular fluid dynamics simulations, Annals of biomedical engineering 49 (2021) 3574–3592. doi:10.1007/S10439-021-02796-X. URLhttps://pubmed.ncbi.nlm.nih.gov/34169398/

    work page doi:10.1007/s10439-021-02796-x 2021

  10. [10]

    A. A. Johnson, T. E. Tezduyar, J. Liou, Numerical simulation of flows past periodic arrays of cylinders, Computational Mechanics 11 (1993) 371–383.doi:10.1007/BF00350094/METRICS. URLhttps://link.springer.com/article/10.1007/BF00350094

    work page doi:10.1007/bf00350094/metrics 1993

  11. [11]

    Carte, J

    G. Carte, J. Dušek, P. Fraunié, A spectral time discretization for flows with dominant periodicity, Journal of Computational Physics 120 (1995) 171–183.doi:10.1006/JCPH.1995.1157

    work page doi:10.1006/jcph.1995.1157 1995

  12. [12]

    K. C. Hall, J. P. Thomas, W. S. Clark, Computation of unsteady nonlinear flows in cascades using a harmonic balance technique, https://doi.org/10.2514/2.1754 40 (2012) 879–886.doi: 10.2514/2.1754. URLhttps://arc.aiaa.org/doi/10.2514/2.1754 42

    work page doi:10.2514/2.1754 2012

  13. [13]

    Ekici, K

    K. Ekici, K. C. Hall, Nonlinear analysis of unsteady flows in multistage turbomachines using harmonic balance, https://doi.org/10.2514/1.22888 45 (2012) 1047–1057.doi:10.2514/1.22888. URLhttps://arc.aiaa.org/doi/10.2514/1.22888

    work page doi:10.2514/1.22888 2012

  14. [14]

    Lübke, A

    R. Lübke, A. Seidel-Morgenstern, L. Tobiska, Numerical method for accelerated calculation of cyclic steady state of modicon–smb-processes, Computers & Chemical Engineering 31 (2007) 258–267.doi:10.1016/J.COMPCHEMENG.2006.06.013

    work page doi:10.1016/j.compchemeng.2006.06.013 2007

  15. [15]

    Zhang, G

    W. Zhang, G. Xi, C. Zhang, Z. Huang, A high-accuracy temporal–spatial pseudospectral method for time-periodic unsteady fluid flow and heat transfer problems, International Journal of Com- putational Fluid Dynamics 25 (2011) 191–206.doi:10.1080/10618562.2011.575369. URLhttps://www.tandfonline.com/doi/abs/10.1080/10618562.2011.575369

    work page doi:10.1080/10618562.2011.575369 2011

  16. [16]

    Richter, An averaging scheme for the efficient approximation of time-periodic flow problems, Computers & Fluids 214 (2021) 104769.doi:10.1016/J.COMPFLUID.2020.104769

    T. Richter, An averaging scheme for the efficient approximation of time-periodic flow problems, Computers & Fluids 214 (2021) 104769.doi:10.1016/J.COMPFLUID.2020.104769

    work page doi:10.1016/j.compfluid.2020.104769 2021

  17. [17]

    Journal of Computational Physics 378, 686–707

    M. Raissi, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equa- tions, Journal of Computational Physics 378 (2019) 686–707.doi:10.1016/J.JCP.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  18. [18]

    M. W. Dissanayake, N. Phan-Thien, Neural-network-based approximations for solving partial differential equations, Communications in Numerical Methods in Engineering 10 (1994) 195–201. doi:10.1002/CNM.1640100303. URLhttps://onlinelibrary.wiley.com/doi/full/10.1002/cnm.1640100303https: //onlinelibrary.wiley.com/doi/abs/10.1002/cnm.1640100303https://onlineli...

    work page doi:10.1002/cnm.1640100303 1994

  19. [19]

    S. Shah, N. K. Anand, Physics-informed neural networks for periodic flows, Physics of Fluids 36 (7 2024).doi:10.1063/5.0216266/3303820. URL/aip/pof/article/36/7/073620/3303820/Physics-informed-neural-networks-for-periodic

    work page doi:10.1063/5.0216266/3303820 2024

  20. [20]

    L. D. McClenny, U. M. Braga-Neto, Self-adaptive physics-informed neural networks, Journal of Computational Physics 474 (2023) 111722.doi:10.1016/J.JCP.2022.111722

    work page doi:10.1016/j.jcp.2022.111722 2023

  21. [21]

    S. Wang, X. Yu, P. Perdikaris, When and why pinns fail to train: A neural tangent kernel perspective, Journal of Computational Physics 449 (2022) 110768.doi:10.1016/J.JCP.2021. 110768

    work page doi:10.1016/j.jcp.2021 2022

  22. [22]

    S. Wang, H. Wang, P. Perdikaris, On the eigenvector bias of fourier feature networks: From regression to solving multi-scale pdes with physics-informed neural networks, Computer Methods in Applied Mechanics and Engineering 384 (2021) 113938.doi:10.1016/J.CMA.2021.113938

    work page doi:10.1016/j.cma.2021.113938 2021

  23. [23]

    M. A. Nabian, R. J. Gladstone, H. Meidani, Efficient training of physics-informed neural networks via importance sampling, Computer-Aided Civil and Infrastructure Engineering 36 (2021) 962– 977.doi:10.1111/MICE.12685

    work page doi:10.1111/mice.12685 2021

  24. [24]

    A. D. Jagtap, Y. Shin, K. Kawaguchi, G. E. Karniadakis, Deep kronecker neural networks: A general framework for neural networks with adaptive activation functions, Neurocomputing 468 (2022) 165–180.doi:10.1016/J.NEUCOM.2021.10.036

    work page doi:10.1016/j.neucom.2021.10.036 2022

  25. [25]

    A. D. Jagtap, K. Kawaguchi, G. E. Karniadakis, Adaptive activation functions accelerate con- vergence in deep and physics-informed neural networks, Journal of Computational Physics 404 (2020) 109136.doi:10.1016/J.JCP.2019.109136

    work page doi:10.1016/j.jcp.2019.109136 2020

  26. [26]

    Mattey, S

    R. Mattey, S. Ghosh, A novel sequential method to train physics informed neural networks for allen cahn and cahn hilliard equations, Computer Methods in Applied Mechanics and Engineering 390 (2022) 114474.doi:10.1016/J.CMA.2021.114474. 43

    work page doi:10.1016/j.cma.2021.114474 2022

  27. [27]

    C. Wu, M. Zhu, Q. Tan, Y. Kartha, L. Lu, A comprehensive study of non-adaptive and residual- based adaptive sampling for physics-informed neural networks, Computer Methods in Applied Mechanics and Engineering 403 (2023) 115671.doi:10.1016/J.CMA.2022.115671

    work page doi:10.1016/j.cma.2022.115671 2023

  28. [28]

    Dutta, A

    S. Dutta, A. H. Gandomi, Design of experiments for uncertainty quantification based on polynomial chaos expansion metamodels, Elsevier, 2020, pp. 369–381.doi:10.1016/ B978-0-12-816514-0.00015-1

    work page 2020

  29. [29]

    P. H. Chiu, J. C. Wong, C. Ooi, M. H. Dao, Y. S. Ong, Can-pinn: A fast physics-informed neural network based on coupled-automatic–numerical differentiation method, Computer Methods in Applied Mechanics and Engineering 395 (2022) 114909.doi:10.1016/J.CMA.2022.114909

    work page doi:10.1016/j.cma.2022.114909 2022

  30. [30]

    J. C. Wong, C. C. Ooi, A. Gupta, Y. S. Ong, Learning in sinusoidal spaces with physics-informed neural networks, IEEE Transactions on Artificial Intelligence 5 (2024) 985–1000.doi:10.1109/ TAI.2022.3192362

    work page arXiv 2024

  31. [31]

    X. Chen, R. Chen, Q. Wan, R. Xu, J. Liu, An improved data-free surrogate model for solving partial differential equations using deep neural networks, Scientific Reports 2021 11:1 11 (2021) 1–17.doi:10.1038/s41598-021-99037-x. URLhttps://www.nature.com/articles/s41598-021-99037-x

    work page doi:10.1038/s41598-021-99037-x 2021

  32. [32]

    S. Li, X. Feng, Dynamic weight strategy of physics-informed neural networks for the 2d navier–stokes equations, Entropy 2022, Vol. 24, Page 1254 24 (2022) 1254.doi:10.3390/ E24091254. URLhttps://www.mdpi.com/1099-4300/24/9/1254/htmhttps://www.mdpi.com/1099-4300/ 24/9/1254

    work page 2022

  33. [33]

    L. Lu, R. Pestourie, W. Yao, Z. Wang, F. Verdugo, S. G. Johnson, Physics-informed neural networks with hard constraints for inverse design, SIAM Journal on Scientific Computing 43 (2021) B1105–B1132.doi:10.1137/21M1397908. URLhttps://arxiv.org/abs/2102.04626v1

    work page doi:10.1137/21m1397908 2021

  34. [34]

    D. P. Kingma, J. L. Ba, Adam: A method for stochastic optimization, 3rd International Confer- ence on Learning Representations, ICLR 2015 - Conference Track Proceedings (12 2014). URLhttps://arxiv.org/abs/1412.6980v9

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  35. [35]

    URLhttps://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Adam

    tf.keras.optimizers.adam | tensorflow v2.16.1. URLhttps://www.tensorflow.org/api_docs/python/tf/keras/optimizers/Adam

    work page

  36. [36]

    URLhttps://www.tensorflow.org/

    Tensorflow. URLhttps://www.tensorflow.org/

    work page

  37. [37]

    Sukumar, A

    N. Sukumar, A. Srivastava, Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks, Computer Methods in Applied Mechanics and Engineer- ing 389 (2022) 114333.doi:10.1016/J.CMA.2021.114333. URLhttps://www.sciencedirect.com/science/article/pii/S0045782521006186

    work page doi:10.1016/j.cma.2021.114333 2022

  38. [38]

    U. Ghia, K. N. Ghia, C. T. Shin, High-re solutions for incompressible flow using the navier- stokes equations and a multigrid method, Journal of Computational Physics 48 (1982) 387–411. doi:10.1016/0021-9991(82)90058-4. URLhttps://doi.org/10.2514/3.61116

    work page doi:10.1016/0021-9991(82)90058-4 1982

  39. [39]

    A. A. Ramabathiran, P. Ramachandran, Spinn: Sparse, physics-based, and partially interpretable neural networks for pdes, Journal of Computational Physics 445 (2021) 110600.doi:10.1016/ J.JCP.2021.110600

    work page arXiv 2021

  40. [40]

    Shepherd, R

    R. Shepherd, R. J. Wiltshire, A periodic solution to a nonlinear diffusion equation, Transport in Porous Media 15 (1994) 175–182.doi:10.1007/BF00625515/METRICS. URLhttps://link.springer.com/article/10.1007/BF00625515 44

    work page doi:10.1007/bf00625515/metrics 1994

  41. [41]

    J. A. Diez, J. Gratton, F. Minotti, Self-similar solutions of the second kind of nonlinear diffusion- type equations, Quarterly of Applied Mathematics 50 (1992) 401–414.doi:10.1090/QAM/ 1178424

    work page doi:10.1090/qam/ 1992

  42. [42]

    C. A. Putra, P. S. Palar, R. Stevenson, L. R. Zuhal, On physics-informed deep learning for solving navier-stokes equations, AIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2022 (2022).doi:10.2514/6.2022-1436

    work page doi:10.2514/6.2022-1436 2022

  43. [43]

    S. Cai, Z. Wang, S. Wang, P. Perdikaris, G. E. Karniadakis, Physics-informed neural networks for heat transfer problems, Journal of Heat Transfer 143 (6 2021).doi:10.1115/1.4050542/ 1104439. URLhttps://dx.doi.org/10.1115/1.4050542

    work page doi:10.1115/1.4050542/ 2021

  44. [44]

    Goraya, N

    S. Goraya, N. Sobh, A. Masud, Error estimates and physics informed augmentation of neural networks for thermally coupled incompressible navier stokes equations, Computational Mechanics 72 (2023) 267–289.doi:10.1007/S00466-023-02334-7/FIGURES/29. URLhttps://link.springer.com/article/10.1007/s00466-023-02334-7

    work page doi:10.1007/s00466-023-02334-7/figures/29 2023

  45. [45]

    Z. Xing, H. Cheng, J. Cheng, Deep learning method based on physics-informed neural network for 3d anisotropic steady-state heat conduction problems, Mathematics 2023, Vol. 11, Page 4049 11 (2023) 4049.doi:10.3390/MATH11194049. URLhttps://www.mdpi.com/2227-7390/11/19/4049/htmhttps://www.mdpi.com/ 2227-7390/11/19/4049

    work page doi:10.3390/math11194049 2023

  46. [46]

    H. S. Carslaw, J. J. C. Jaeger, Conduction of heat in solids. 2nde edition, edition anglaise (1986) 510. URLhttps://books.google.com/books/about/Conduction_of_Heat_in_Solids.html?id= y20sAAAAYAAJ 45

    work page 1986