Pith. sign in

REVIEW 3 major objections 5 minor 48 references

PC-SRGAN: Physically Consistent Super-Resolution Generative Adversarial Network for General Transient Simulations

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read PC-SRGAN adds PDE-residual and boundary losses to a super-resolution GAN, and reports super-resolved fields that can restart finite-element time marching with errors about an order of magnitude below the GAN baseline.

desk verdict A solid, clearly written extension of SRGAN with a useful time-integrator comparison and a strong downstream FEM experiment, but the physical-consistency metrics share the same fitted convolution operator used in training and the printed Laplacian kernel cannot be right as written. read the letter →

arxiv 2505.06502 v3 pith:RYB2E35G submitted 2025-05-10 eess.IV cs.CEcs.CVcs.LG

Md Rakibul Hasan , Pouria Behnoudfar , Dan MacKinlay , Thomas Poulet This is my paper · ORCID
classification eess.IVcs.CEcs.CVcs.LG
keywords physics-informedneuralnetworkphysicalconsistencysuper-resolutiongenerativeadversarialimagequalityassessmenttimeintegratortransientsimulationsAllen-Cahnequation
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 claims that a super-resolution GAN can be made physically consistent by adding two loss terms that penalise deviations from the governing partial differential equation inside the domain and on its boundary. On two transient model problems, Allen-Cahn and Eriksson-Johnson, the resulting method, PC-SRGAN, is reported to beat the SRGAN baseline on PSNR, SSIM, MSE, LPIPS, and a gradient-error metric, and to reach baseline quality with only about 13% of the training data. The deeper claim is that the super-resolved fields are usable as restart states for finite-element time marching: continuing a simulation from PC-SRGAN output keeps errors about an order of magnitude smaller than continuing from SRGAN output. A reader should care because this points to a cheap route to fine-resolution transient simulation: run a coarse simulation, upsample it with a physics-aware generator, and continue on the fine mesh.

What carries the argument

The load-bearing object is the combined loss of Eq. (4), in which the physics-inner term $L_\phi^{\mathrm{in}}$ evaluates the PDE residual on the generated field with a chosen time-marching formula and the physics-boundary term $L_\phi^{\mathrm{bd}}$ penalises mismatch with the prescribed boundary condition type. Spatial derivatives in $L_\phi^{\mathrm{in}}$ are computed with fixed convolution kernels, a Laplacian kernel scaled by $\alpha = 9.894$ and first-derivative kernels scaled by $\beta = -5.645$, fitted on smooth analytic functions; this is what lets the loss target the continuous PDE without the $H^2$ regularity implied by automatic differentiation. The characteristic-polynomial error analysis (Eqs. 13-17) is the argument that BDF controls the convolution-kernel error better than Crank-Nicolson or other Runge-Kutta-type integrators, which the ablation results support.

What would settle it

Take a 64x64 Allen-Cahn ground-truth snapshot with sharp interfaces and compute the PDE residual using the paper's fitted filters (Laplacian scaled by 9.894, first derivatives scaled by -5.645), then compare with the residual computed by a high-order finite-difference or spectral spatial operator on the same grid. If ground-truth snapshots show much larger filter-based residuals than PC-SRGAN outputs, the physical-consistency gain mostly reflects agreement with the filter rather than with the PDE.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that inserting a residual of the semi-discrete PDE into the GAN objective makes the generator produce fields whose time evolution respects the target equation, not merely images that look like the ground truth. The objective combines the usual SRGAN pixel, feature, and adversarial losses with two physics terms: an inner loss (Eq. 5) that evaluates the first-order-in-time PDE residual on the generated high-resolution field using a chosen time integrator, and a boundary loss (Eq. 8) that enforces Dirichlet, periodic, or Neumann conditions. Spatial derivatives in the inner residual are computed by fixed convolution filters fitted on smooth analytic functions, which avoids the extra regularity that automatic differentiation would impose. The authors report that a BDF-based residual outperforms Crank-Nicolson and Explicit Euler variants, that the inner physics term is the largest single contributor to the gains, and that the same physics losses also improve an ESRGAN backbone, yielding PC-ESRGAN.

Load-bearing premise

The load-bearing premise is that the fixed convolution filters used for spatial derivatives in the physics loss faithfully approximate the continuous PDE operators at the target resolution, including across sharp interfaces; if they do not, the loss penalises the wrong residual and the reported gradient improvements become partly self-referential.

Editorial extensions

If this is right

  • Super-resolution can act as a surrogate initialiser for transient finite-element simulations, letting a coarse-grid solution be upsampled with physical consistency and then marched forward on the fine mesh at lower cost than a full fine-grid solve.
  • The physics loss terms are model-agnostic: they improved an ESRGAN backbone as well as SRGAN, so they should transfer to diffusion- or flow-based super-resolution models for scientific data.
  • Data requirements for scientific super-resolution drop substantially, since PC-SRGAN reportedly matches or beats the baseline with roughly 13-18% of the training data.
  • The choice of time integrator embedded in the loss is consequential, and multi-step BDF residuals appear more favourable than multi-stage Crank-Nicolson or Explicit Euler residuals for enforcing dynamics.
  • Gradient-focused quality metrics such as MSGE and GSNR are needed alongside PSNR and SSIM, because pointwise image quality can hide errors in derivatives that drive downstream simulations.

Reading between the lines

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

  • If the fitted convolution filters remain faithful on other mesh sizes, the same physically consistent loss should extend to three-dimensional regular grids and to other first-order-in-time PDEs, though the paper only demonstrates two-dimensional 4x and 8x cases.
  • The strong data-efficiency result suggests the PDE residual acts as a regulariser that could reduce dataset sizes for other well-known transient equations, but only when the governing equation and boundary conditions are known precisely.
  • A natural stress test the paper does not run is whether PC-SRGAN preserves conserved quantities or decay rates of the underlying PDE over many restart cycles; a reader could check that by measuring total mass or free energy along the continued FEM trajectory.
  • Since the boundary-loss ablation affected boundary quality more than global metrics, a boundary-focused evaluation metric might expose larger gains than PSNR and SSIM reveal.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes PC-SRGAN, a super-resolution GAN that augments the SRGAN loss with two physics-based terms: an inner-domain PDE residual term and a boundary-condition term, targeting transient initial-boundary-value problems of the form ∂u/∂t + F(u) = 0. The method is demonstrated on Allen-Cahn and Eriksson-Johnson problems with 8× and 4× upscaling, comparing against SRGAN and ESRGAN baselines. The authors report improved PSNR, SSIM, MSE, and LPIPS, introduce two gradient-fidelity metrics (MSGE and GSNR), analyze data efficiency, and show that PC-SRGAN outputs, when used as initial conditions for continuing FEM simulations, yield errors an order of magnitude smaller than those from SRGAN outputs. The authors also release code and datasets.

Significance. If the physical-consistency claims are sound, this is a valuable contribution to scientific machine learning: it demonstrates that physics-informed losses can improve both image-quality metrics and downstream usability of super-resolved simulation data, while also improving data efficiency. The public release of code and datasets and the surrogate-continuation experiment are strengths. However, the central physical-consistency claim rests on the fidelity of the convolution filters used to evaluate the PDE residual, and the gradient metrics use the same operator, so the reported improvements in physical consistency are only as reliable as those filters. The empirical improvements in PSNR/SSIM and the downstream FEM experiment are plausible and, to some extent, independent evidence of benefit, but they do not by themselves establish that the method enforces the governing PDE.

major comments (3)
  1. [Appendix B, Eqs. (5)-(6), (26)-(27)] The printed Laplacian kernel, α [0 1 0; 1 4 1; 0 1 0], is not a valid discrete Laplacian because its coefficients sum to 8, giving a nonzero response on constant fields. A correct discrete Laplacian stencil must sum to zero (e.g., [0 1 0; 1 -4 1; 0 1 0]). As printed, the kernel cannot have been calibrated on smooth trigonometric functions to approximate the Laplacian, since its response on cos(nπx)sin(mπy) contains a term proportional to the function itself, not just the second derivative. Consequently, the physics loss Lin_φ in Eq. (5) does not minimize the residual of the PDE in Eq. (1); it enforces a different equation with a spurious source term. This undermines the physical-consistency claims in Sections III and IV. The authors must either correct the kernel (and re-run all experiments) or provide a rigorous justification for why a kernel with nonzero DC response is a suitable approximation of the Laplacian on the solution class considered.
  2. [Section III-A, Eqs. (9)-(10)] The gradient metrics MSGE and GSNR are computed with the same convolution filter used to define the physics loss in Eq. (5). Therefore the reported improvements in MSGE and GSNR partly reflect the network having been trained to minimize that specific filter's response, not necessarily fidelity to the continuous gradient of the PDE solution. To substantiate the physical-consistency claim, the authors should evaluate gradient fidelity with an independent operator (e.g., a different high-order finite-difference stencil or a finite-element gradient recovery) and report those numbers. Without such a check, the physical-consistency improvement, as opposed to the image-quality improvement, is not established independently of the operator imposed during training.
  3. [Section III-B, Eqs. (13)-(17)] The derivation of the error bound EG ≤ C(ET + O(τ^q) + τD_Cσ(N^{-1})) is asserted rather than proven, and the treatment of Crank-Nicolson versus Runge-Kutta in Remark 1 is confusing. CN is a one-step (multi-stage) method and its σ(z) is not of the form z^q + ... + 1; for BDF, σ(z) is a constant. The resulting bounds in Eq. (17) appear to classify CN as having no convolution-kernel term while RK does, which is inconsistent with the methods' structures. This analysis is not the main contribution, but as written it does not support the claim that BDF is inherently better at controlling the error from the convolution-kernel approximation.
minor comments (5)
  1. [Eq. (6)] The spatial operator f(·) is written with ∇u, which is a vector, where a scalar Laplacian is clearly intended for the Allen-Cahn and Eriksson-Johnson equations; this should be Δu or ∇·(ε∇u).
  2. [Appendix B] The description of the filter calibration is too vague: the statement that the coefficient is obtained by matching cos(nπx)sin(mπy) does not specify the optimization objective, the range of n/m, the grid spacing, or the achieved error, making it impossible to reproduce the coefficients from the text.
  3. [Section III-C and Table II] No error bars or multiple random seeds are reported for any experiment. Given the stochastic nature of GAN training, the claimed margins are plausible but would be more convincing if accompanied by standard deviations over at least three seeds.
  4. [Eq. (8)] In the Neumann boundary condition case, the notation U_sr(x±h) is not fully specified; it should clearly indicate which boundaries are involved and the orientation of the one-sided difference.
  5. [Section IV] The statement that τ = 1 is assumed in many ML studies is not substantiated with examples, and the relationship between the constant time step in the experiments and the general time-step analysis in Section III-B is not clarified.

Circularity Check

1 steps flagged · score 4.0 of 10

MSGE/GSNR gradient-fidelity metrics reuse the same convolution operator as the physics loss, making part of the physical-consistency claim self-referential.

  1. self definitional [Section III-A, Eq. (9), with Eq. (5) and Appendix B]
    "where N is the number of points, ∇j û_i is the gradient of the approximated solution û, for the real solution u, in direction j=x,y calculated using the convolution filter at point i."

    The inner physics loss in Eq. (5) evaluates f(U^n_sr, ε, K, r, θ) using the Appendix B convolution kernels (α=9.894, β=−5.645). The MSGE metric in Eq. (9) then scores gradient errors using the same "convolution filter." A generator trained to minimize Eq. (4) is therefore directly minimizing the quantity reported as improved physical consistency in Table I. GSNR (Eq. 10) is a monotone transform of MSGE, inheriting the same operator. Consequently, the MSGE/GSNR improvements are partly true by construction and do not independently establish that the continuous gradients are more physical. The independent evidence for the central claim is the hold-out PSNR/SSIM/LPIPS comparison and the FEM continuation experiment of Fig. 5.

full rationale

The paper's image-quality claims are independently supported: PSNR, SSIM, and LPIPS are computed on a hold-out test set against SRGAN/ESRGAN baselines, and Fig. 5 runs an external FEM continuation that is not tied to the training operator, giving a credible surrogate-capability result. No load-bearing self-citation chain is present: reference [24] (including a co-author) is cited for characteristic-polynomial details, but Appendix C supplies the needed formulae, Remark 1 states the relevant σ(z) facts, and the BDF advantage is also demonstrated empirically in Table I. The one genuine circularity is in the physical-consistency metrics: MSGE (Eq. 9) and GSNR (Eq. 10) define gradient fidelity using the same convolution filter used to compute the spatial operator in the physics loss (Eq. 5 and Appendix B). Thus the reported MSGE/GSNR gains are partly self-referential by construction and should be read as evidence of fit to the discrete operator rather than as independent proof of physical gradient fidelity. Because the stronger external checks remain independent, the overall circularity is partial, not total.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central method rests on a small set of hand-tuned loss weights and on empirically fitted convolution filters that define the discrete PDE residual. These parameters are not derived from first principles, and the theoretical error bound is imported from prior work without a full derivation.

free parameters (5)
  • Standard SRGAN loss weights w1, w2, w3 (pixel, content, adversarial) = 1.0, 1.0, 0.001
    Manually selected to match loss magnitudes; largely carried over from SRGAN, despite the text claiming all hyperparameters unchanged.
  • Physics-inner loss weight w4 = 1e-8 (Allen-Cahn), 1e-2 (Eriksson-Johnson)
    Chosen by hand. Ablation shows high sensitivity: raising it to 1e-5 drops PSNR from 37.27 to 30.35 on Allen-Cahn 8x.
  • Physics-boundary loss weight w5 = 5.0 (Allen-Cahn), 100 (Eriksson-Johnson)
    Chosen by hand. The ablation shows setting it to zero improves MSE and MSGE slightly while degrading PSNR and SSIM.
  • Laplacian convolution filter coefficient alpha = 9.894
    Fitted to match an analytic C-infinity function on a 64x64 grid. It defines the discrete Laplacian used in the physics loss and in the MSGE and GSNR metrics.
  • Directional derivative filter coefficient beta = -5.645, with interior weight 3.5887
    Empirically chosen for first derivatives. Used in the convection terms of the Eriksson-Johnson and Allen-Cahn physics losses.
assumptions (6)
  • domain assumption The target problems are well-posed first-order IBVPs of the form du/dt + F(u) = 0 (Eq. 1)
    The Allen-Cahn and Eriksson-Johnson equations are assumed to admit unique solutions on the chosen domains, so residual minimization is meaningful.
  • domain assumption High-resolution FEM data generated with a generalized-alpha time integrator are accurate ground truth
    No mesh-convergence study is provided; the metrics treat the fine FEM solution as the reference.
  • ad hoc to paper The fitted convolution filters approximate the Laplacian and first derivatives accurately on the solution class
    Appendix B fits alpha and beta on smooth analytic functions; accuracy on sharp Allen-Cahn interfaces is not verified.
  • ad hoc to paper The generalization bound of Mishra and Molinaro extends to the GAN generator and to the training error defined in Eq. (12)
    Eq. (16) is asserted by analogy with [25]; the nonlinearity of Allen-Cahn and the adversarial training are not addressed.
  • domain assumption A constant time step tau is used across the loss and the data
    The time integrators in Eq. (5) assume fixed tau; the paper notes that tau=1 is common in ML studies and should be revisited.
  • domain assumption Boundary conditions can be enforced by MSE penalties on boundary pixels
    Eq. (8) uses soft MSE constraints instead of exact imposition; whether this guarantees the boundary condition is not analyzed.
invented entities (1)
  • GSNR (Gradient Signal-to-Noise Ratio)
    purpose: Quality metric for evaluating gradient fidelity of super-resolved images
    Defined in Eq. (10); not validated against existing gradient metrics or perceptual studies, and it is computed with the same convolution filter used in the training loss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PC-SRGAN: Physically Consistent Super-Resolution Generative Adversarial Network for General Transient Simulations." pith.science (2026). https://pith.science/paper/RYB2E35G

@misc{pith2026250506502,
  author       = {Pith},
  title        = {Pith review of: PC-SRGAN: Physically Consistent Super-Resolution Generative Adversarial Network for General Transient Simulations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RYB2E35G}},
  note         = {Machine review of arXiv:2505.06502}
}
read the original abstract

Machine Learning, particularly Generative Adversarial Networks (GANs), has revolutionised Super-Resolution (SR). However, generated images often lack physical meaningfulness, which is essential for scientific applications. Our approach, PC-SRGAN, enhances image resolution while ensuring physical consistency for interpretable simulations. PC-SRGAN significantly improves both the Peak Signal-to-Noise Ratio and the Structural Similarity Index Measure compared to conventional SR methods, even with limited training data (e.g., only 13% of training data is required to achieve performance similar to SRGAN). Beyond SR, PC-SRGAN augments physically meaningful machine learning, incorporating numerically justified time integrators and advanced quality metrics. These advancements promise reliable and causal machine-learning models in scientific domains. A significant advantage of PC-SRGAN over conventional SR techniques is its physical consistency, which makes it a viable surrogate model for time-dependent problems. PC-SRGAN advances scientific machine learning by improving accuracy and efficiency, enhancing process understanding, and broadening applications to scientific research. We publicly release the complete source code of PC-SRGAN and all experiments at https://github.com/hasan-rakibul/PC-SRGAN.

Figures

Figures reproduced from arXiv: 2505.06502 by the authors.

Figure 1
Figure 1. Schematic description of PC-SRGAN. A Residual Network (ResNet)- [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) Example of low-resolution ground truth image, [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Comparison of Allen-Cahn 8× solution distributions. (a) Results from FEM simulation (ground truth), with 3D warping by solution values and colouring by gradient magnitude. (b) Corresponding top view of ground truth image coloured by gradient magnitude. (c) Corresponding PC-SRGAN results. (d) Corresponding SRGAN results. For all subfigures, the gradient colour bar is capped from 0 (blue) to 0.05 (red) instead of its … view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Comparison of PC-SRGAN’s and SRGAN’s surrogate capabilities. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 29 canonical work pages

  1. [1]

    Photo-realistic single image super-resolution using a generative adversarial network,

    C. Ledig, L. Theis, F. Husz ´ar, et al. , “Photo-realistic single image super-resolution using a generative adversarial network,” in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 2017, pp. 105–114. doi: 10.1109/CVPR.2017.19

    doi:10.1109/cvpr.2017.19 2017

  2. [2]

    Image super- resolution: A comprehensive review, recent trends, challenges and applications,

    D. C. Lepcha, B. Goyal, A. Dogra, and V. Goyal, “Image super- resolution: A comprehensive review, recent trends, challenges and applications,” Information Fusion, vol. 91, pp. 230–260, Mar. 2023. doi: 10.1016/j.inffus.2022.10.007

    doi:10.1016/j.inffus.2022.10.007 2023

  3. [3]

    Physics guided neural networks for spatio-temporal super-resolution of turbulent flows,

    T. Bao, S. Chen, T. T. Johnson, P. Givi, S. Sammak, and X. Jia, “Physics guided neural networks for spatio-temporal super-resolution of turbulent flows,” in Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence , J. Cussens and K. Zhang, Eds., ser. Proceedings of Machine Learning Research, vol. 180, PMLR, 2022, pp. 118–128

    work page 2022

  4. [4]

    TempoGAN: A tempo- rally coherent, volumetric GAN for super-resolution fluid flow,

    Y. Xie, E. Franz, M. Chu, and N. Thuerey, “TempoGAN: A tempo- rally coherent, volumetric GAN for super-resolution fluid flow,” ACM Transactions on Graphics , vol. 37, no. 4, pp. 1–15, 2018. doi: 10 . 1145/3197517.3201304

    arXiv 2018

  5. [5]

    Improved Denoising Diffusion Proba- bilistic Models,

    A. Q. Nichol and P. Dhariwal, “Improved Denoising Diffusion Proba- bilistic Models,” in Proceedings of the 38th International Conference on Machine Learning , PMLR, Jul. 2021, pp. 8162–8171

    work page 2021

  6. [6]

    Generative adver- sarial nets,

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, et al., “Generative adver- sarial nets,” in Advances in Neural Information Processing Systems, Z. Ghahramani, M. Welling, C. Cortes, N. Lawrence, and K. Weinberger, Eds., vol. 27, Curran Associates, Inc., 2014, pp. 2672–2680

    work page 2014

  7. [7]

    Using physics-informed enhanced super-resolution generative adversarial networks for subfilter modeling in turbulent reactive flows,

    M. Bode, M. Gauding, Z. Lian, et al. , “Using physics-informed enhanced super-resolution generative adversarial networks for subfilter modeling in turbulent reactive flows,” Proceedings of the Combustion Institute, vol. 38, no. 2, pp. 2617–2625, 2021, issn: 1540-7489. doi: 10.1016/j.proci.2020.06.022

    doi:10.1016/j.proci.2020.06.022 2021

  8. [8]

    Grigis and J

    A. Grigis and J. Sj ¨ostrand, Microlocal analysis for differential opera- tors: an introduction . Cambridge university press, 1994, vol. 196

    work page 1994

Show all 48 references
  1. [9]

    Incorporating physics into data-driven computer vision,

    A. Kadambi, C. de Melo, C.-J. Hsieh, M. Srivastava, and S. Soatto, “Incorporating physics into data-driven computer vision,” Nature Ma- chine Intelligence, vol. 5, no. 6, pp. 572–580, Jun. 2023. doi: 10.1038/ s42256-023-00662-0

    2023

  2. [10]

    Esrgan: Enhanced super-resolution generative adversarial networks,

    X. Wang, K. Yu, S. Wu, et al. , “Esrgan: Enhanced super-resolution generative adversarial networks,” in Computer Vision – ECCV 2018 Workshops, L. Leal-Taix´e and S. Roth, Eds., Cham: Springer Interna- tional Publishing, 2019, pp. 63–79. doi: 10.1007/978-3-030-11021- 5 5

    2018 doi

  3. [11]

    M. R. Hasan, P. Behnoudfar, D. Mackinlay, and T. Poulet, Transient simulation datasets using allen-cahn and eriksson-johnson equations , Zenodo, Jul. 2025. doi: 10.5281/zenodo.16250888

    2025 doi

  4. [12]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in International Conference on Learning Representations (ICLR) , 2015

    2015

  5. [13]

    Physics-informed neural networks for heat transfer problems,

    S. Cai, Z. Wang, S. Wang, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks for heat transfer problems,”Journal of Heat Transfer, vol. 143, no. 6, p. 060 801, 2021

    2021

  6. [14]

    Physics- informed neural networks (pinns) for fluid mechanics: A review,

    S. Cai, Z. Mao, Z. Wang, M. Yin, and G. E. Karniadakis, “Physics- informed neural networks (pinns) for fluid mechanics: A review,” Acta Mechanica Sinica, vol. 37, no. 12, pp. 1727–1738, 2021

    2021

  7. [15]

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,

    M. Raissi, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations,” Journal of Computational physics , vol. 378, pp. 686–707, 2019

    2019

  8. [16]

    P. Ren, N. B. Erichson, S. Subramanian, O. San, Z. Lukic, and M. W. Mahoney, SuperBench: A Super-Resolution Benchmark Dataset for Scientific Machine Learning, Jun. 2023. doi: 10.48550/arXiv.2306. 14070. arXiv: 2306.14070

    2023 doi arXiv

  9. [17]

    Image quality assessment through FSIM, SSIM, MSE and PSNR—A comparative study,

    U. Sara, M. Akter, and M. S. Uddin, “Image quality assessment through FSIM, SSIM, MSE and PSNR—A comparative study,” Journal of Computer and Communications , vol. 07, no. 03, pp. 8–18, 2019. doi: 10.4236/jcc.2019.73002

    2019

  10. [18]

    Image quality assessment: From error visibility to structural similarity,

    Z. Wang, A. Bovik, H. Sheikh, and E. Simoncelli, “Image quality assessment: From error visibility to structural similarity,” IEEE Trans- actions on Image Processing , vol. 13, no. 4, pp. 600–612, Apr. 2004. doi: 10.1109/TIP.2003.819861

    2004

  11. [19]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018, pp. 586–595. doi: 10.1109/CVPR.2018.00068

    2018

  12. [20]

    Deep learning-enabled resolution-enhancement in mini- and regular mi- croscopy for biomedical imaging,

    M. Dai, G. Xiao, L. Fiondella, M. Shao, and Y. S. Zhang, “Deep learning-enabled resolution-enhancement in mini- and regular mi- croscopy for biomedical imaging,” Sensors and Actuators A: Physical , vol. 331, p. 112 928, 2021, issn: 0924-4247. doi: 10.1016/j.sna.2021. 112928

    2021 doi

  13. [21]

    Deepthi, A

    K. Deepthi, A. K. Shastry, and E. Naresh, “A novel deep unsu- pervised approach for super-resolution of remote sensing hyperspec- tral image using gompertz-function convergence war accelerometric- optimization generative adversarial network (gf-cwao-gan),” Scientific Reports, ...

    2024

  14. [22]

    PDEBench: An Exten- sive Benchmark for Scientific Machine Learning,

    M. Takamoto, T. Praditia, R. Leiteritz, et al., “PDEBench: An Exten- sive Benchmark for Scientific Machine Learning,” Jun. 2022

    2022

  15. [23]

    Shalev-Shwartz and S

    S. Shalev-Shwartz and S. Ben-David, Understanding Machine Learn- ing: From Theory to Algorithms . Cambridge university press, 2014

    2014

  16. [24]

    A variational splitting of high-order linear multistep methods for heat transfer and advection–diffusion parabolic problems,

    P. Behnoudfar, V. M. Calo, M. Lo ´s, P. Maczuga, and M. Paszy ´nski, “A variational splitting of high-order linear multistep methods for heat transfer and advection–diffusion parabolic problems,” Journal of Computational Science, vol. 63, p. 101 807, 2022

    2022

  17. [25]

    Estimates on the generalization error of physics-informed neural networks for approximating PDEs,

    S. Mishra and R. Molinaro, “Estimates on the generalization error of physics-informed neural networks for approximating PDEs,” IMA Journal of Numerical Analysis , vol. 43, no. 1, pp. 1–43, 2023

    2023

  18. [26]

    Physics and equality constrained artificial neural networks: Application to forward and inverse problems with multi-fidelity data fusion,

    S. Basir and I. Senocak, “Physics and equality constrained artificial neural networks: Application to forward and inverse problems with multi-fidelity data fusion,” Journal of Computational Physics, vol. 463, p. 111 301, Aug. 2022. doi: 10.1016/j.jcp.2022.111301

    2022

  19. [27]

    Feature-adjacent multi-fidelity physics- informed machine learning for partial differential equations,

    W. Chen and P. Stinis, “Feature-adjacent multi-fidelity physics- informed machine learning for partial differential equations,” Journal of Computational Physics, vol. 498, p. 112 683, Feb. 2024, issn: 0021-

    2024

  20. [28]

    Message passing neural PDE solvers,

    J. Brandstetter, D. Worrall, and M. Welling, “Message passing neural PDE solvers,” in The Tenth International Conference on Learning Representations (ICLR), Mar. 2022. arXiv: 2202.03376

    2022 arXiv

  21. [29]

    Neural operators for accelerating scientific simulations and design,

    K. Azizzadenesheli, N. Kovachki, Z. Li, M. Liu-Schiaffini, J. Kossaifi, and A. Anandkumar, “Neural operators for accelerating scientific simulations and design,” Nature Reviews Physics , Apr. 2024, issn: 2522-5820. doi: 10.1038/s42254-024-00712-5

    2024 doi

  22. [30]

    Differentiable modelling to unify machine learning and physical models for geosciences,

    C. Shen, A. P. Appling, P. Gentine, et al., “Differentiable modelling to unify machine learning and physical models for geosciences,” Nature Reviews Earth & Environment , vol. 4, no. 8, pp. 552–567, Jul. 2023. doi: 10.1038/s43017-023-00450-9

    2023 doi

  23. [31]

    Analytical, semi- analytical, and numerical solutions for the cahn–allen equation,

    M. M. Khater, C. Park, D. Lu, and R. A. Attia, “Analytical, semi- analytical, and numerical solutions for the cahn–allen equation,” Ad- vances in Difference Equations , vol. 2020, no. 1, p. 9, 2020. doi: 10.1186/s13662-019-2475-8

    2020 doi

  24. [32]

    Phase-field models for microstructure evolution,

    L.-Q. Chen, “Phase-field models for microstructure evolution,” Annual review of materials research , vol. 32, no. 1, pp. 113–140, 2002

    2002

  25. [33]

    Geometrical image segmen- tation by the allen–cahn equation,

    M. Bene ˇs, V. Chalupeck` y, and K. Mikula, “Geometrical image segmen- tation by the allen–cahn equation,” Applied Numerical Mathematics , vol. 51, no. 2-3, pp. 187–205, 2004

    2004

  26. [34]

    Analytical and numerical solutions of mathematical biology models: The newell- whitehead-segel and allen-cahn equations,

    B. Inan, M. S. Osman, T. Ak, and D. Baleanu, “Analytical and numerical solutions of mathematical biology models: The newell- whitehead-segel and allen-cahn equations,” Mathematical methods in the applied sciences , vol. 43, no. 5, pp. 2588–2600, 2020

    2020

  27. [35]

    Localized folding of thick layers,

    P. Behnoudfar, B. E. Hobbs, A. Ord, L. Espath, and V. M. Calo, “Localized folding of thick layers,” Journal of Structural Geology , vol. 161, p. 104 669, 2022

    2022

  28. [36]

    An unconditionally stable hybrid numerical method for solving the Allen–Cahn equation,

    Y. Li, H. G. Lee, D. Jeong, and J. Kim, “An unconditionally stable hybrid numerical method for solving the Allen–Cahn equation,” Com- puters and Mathematics with Applications , vol. 60, no. 6, pp. 1591– 1606, 2010

    2010

  29. [37]

    Adaptive streamline diffusion finite element methods for stationary convection-diffusion problems,

    K. Eriksson and C. Johnson, “Adaptive streamline diffusion finite element methods for stationary convection-diffusion problems,” math- ematics of computation , vol. 60, no. 201, pp. 167–188, 1993

    1993

  30. [38]

    Adaptive finite element methods in computational mechanics,

    C. Johnson and P. Hansbo, “Adaptive finite element methods in computational mechanics,” Computer Methods in Applied Mechanics and Engineering, vol. 101, no. 1, pp. 143–181, 1992, issn: 0045-7825. doi: 10.1016/0045-7825(92)90020-K

    1992 doi

  31. [39]

    Computational differential equations. by k. eriksson, d. estep, p. hansbo & c. johnson. cambridge university press, 1997. 538 pp. isbn 0521 56738 6. £29.95,

    C. Pozrikidis, “Computational differential equations. by k. eriksson, d. estep, p. hansbo & c. johnson. cambridge university press, 1997. 538 pp. isbn 0521 56738 6. £29.95,” Journal of Fluid Mechanics, vol. 368, pp. 376–378, 1998. doi: 10.1017/S0022112098211529

    1997 doi

  32. [40]

    T. J. Hughes, The finite element method: linear static and dynamic finite element analysis . Courier Corporation, 2012

    2012

  33. [41]

    A generalized- α method for integrating the filtered navier–stokes equations with a stabilized finite element method,

    K. E. Jansen, C. H. Whiting, and G. M. Hulbert, “A generalized- α method for integrating the filtered navier–stokes equations with a stabilized finite element method,” Computer methods in applied mechanics and engineering , vol. 190, no. 3-4, pp. 305–319, 2000

    2000

  34. [42]

    J. C. Butcher, Numerical Methods for Ordinary Differential Equations. John Wiley & Sons, 2016

    2016

  35. [43]

    U. M. Ascher and L. R. Petzold, Computer Methods for Ordinary Differential Equations and Differential-Algebraic Equations . Siam, 1998, vol. 61

    1998

  36. [44]

    Iserles, A First Course in the Numerical Analysis of Differential Equations

    A. Iserles, A First Course in the Numerical Analysis of Differential Equations. Cambridge university press, 2009

    2009

  37. [45]

    S¨ uli and D

    E. S¨ uli and D. F. Mayers, An Introduction to Numerical Analysis . Cambridge university press, 2003

    2003

  38. [46]

    K. He, X. Zhang, S. Ren, and J. Sun, Deep residual learning for image recognition, 2015. arXiv: 1512.03385 [cs.CV]

    2015 arXiv

  39. [47]

    Adam: A method for stochastic optimiza- tion,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimiza- tion,” in 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, Y. Bengio and Y. LeCun, Eds., 2015. [Online]. Available: http://arxi...

    2019 arXiv

  40. [9991]

    doi: 10.1016/j.jcp.2023.112683

    2023

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.