Pith. sign in

REVIEW 3 major objections 5 minor 43 references

Applying Two-Grid Preconditioner for Subsurface Flow Simulation using Attention-enhanced Hybrid Network to Accelerate Multiscale Discretization in High-contrast Media

T0 review · 3 major / 5 minor · reviewed 2026-07-12 · grok-4.5

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

Pith's one-line read Learning multiscale basis functions, then solving with a two-grid preconditioner, accelerates high-contrast Darcy simulation while preserving accuracy.

desk verdict Solid hybrid workflow: learn mixed-GMsFEM bases with an attention FNO+U-Net, then assemble and two-grid solve; theory is a clean error sketch under strong assumptions, and one weak basis fit is real but does not sink the empirical claim. read the letter →

arxiv 2606.02582 v1 pith:EF7QLC3F submitted 2026-04-15 cs.CE cs.LGcs.NAmath.NA

Peiqi Li , Jie Chen , Shubin Fu This is my paper · ORCID
classification cs.CEcs.LGcs.NAmath.NA
keywords MultiscaleDarcyEquationTwo-GridPreconditionerNeuralNetworkAttentionHigh-contrastMediamixedGMsFEMbasisfunctions
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

Darcy flow in strongly heterogeneous, high-contrast porous media forces expensive fine-grid solves or repeated local basis constructions inside multiscale methods. This paper shows that an attention-enhanced hybrid network can predict the offline multiscale basis functions of mixed GMsFEM, after which the global system is assembled and pressure is obtained by a standard two-grid preconditioned iteration. The resulting hybrid pipeline produces more accurate pressure fields than several pure learning baselines and remains stable under high contrast, while its main gain over classical mixed GMsFEM is a much cheaper basis-generation stage. A sympathetic reader cares because the approach keeps the interpretability and local conservation of a mature multiscale discretization yet removes the dominant offline cost that has limited high-resolution subsurface simulation.

What carries the argument

The attention-enhanced hybrid network (Fourier neural operator with frequency-wise attention followed by Attention U-Net spatial refinement) that maps a permeability field to the offline multiscale basis matrix R_off, which is then used inside a classical two-grid preconditioner whose convergence factor is controlled by the residual basis error.

What would settle it

On a high-contrast test suite, if the pressure relative L2 error of the hybrid method exceeds that of classical mixed GMsFEM with the same number of bases, or if the two-grid iteration fails to converge while the classical version succeeds, the claim that learning error is controllable is falsified.

Watch

Extended reading notes

Core claim

Accelerating multiscale basis construction with an attention-enhanced hybrid network, while still assembling the mixed-GMsFEM global system and solving pressure by a two-grid preconditioner, yields more accurate final pressure reconstructions than representative pure-learning methods and remains stable under strong heterogeneity and high contrast; its primary advantage over traditional mixed GMsFEM is efficiency of the basis-generation stage.

Load-bearing premise

The network must keep the aggregate error in the predicted basis functions small enough that the Galerkin approximation property and the two-grid convergence factor remain essentially the same as for the exact offline space.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a hybrid solver for mixed-form Darcy flow in high-contrast heterogeneous media. Multiscale basis functions of mixed GMsFEM are predicted by an attention-enhanced hybrid network (frequency-domain FNO with frequency-wise attention, followed by Attention U-Net spatial refinement), after which the global mixed system is assembled and pressure is obtained by a two-grid preconditioned iteration. Section 5 supplies a structured error analysis linking basis-prediction error to offline-space approximation quality, two-grid contraction, and final pressure error under stated assumptions. Numerical experiments on 2D KLE-generated fractured media report basis-function MSE/R^{2}, pressure accuracy versus FNO, U-Net, PINN and classical mixed GMsFEM, and results for several numbers of retained bases.

Significance. If the hybrid construction is reliable, the work offers a practical route to cut the dominant offline cost of mixed GMsFEM while retaining local conservation, multiscale discretization, and a mature two-grid solver for the global system. That combination is more defensible for high-contrast Darcy simulation than pure end-to-end field learning. The manuscript also supplies an explicit error decomposition (basis error o Galerkin approximation o two-grid propagation o pressure error) and a concrete network design with gradient-augmented loss, which are useful contributions even if some quantitative links remain incomplete. Strengths include the clear separation of learning (bases only) from numerical assembly/solve, and the multi-baseline pressure comparison on high-contrast media.

major comments (3)
  1. Table 2 reports R^{2} = 0.5540 for multiscale basis 4 (other bases >0.9). Theorems 5.3–5.12 and Assumption 5.5 require that aggregate basis error ε_basis / ε_loc stay small enough, and that the predicted offline space share dimension and Riesz constants c_R, C_R with the true space, so that Ć_1 and ∥Ê_1∥_A remain controlled. The paper never reports ε_loc, ∥ΔR∥, realized Ć_1, or two-grid contraction factors on R̂_off. With a poorly learned mode retained in the offline space, the load-bearing premise of the error analysis is not verified; the good pressure results for mm=6 could be ensemble- or truncation-specific rather than guaranteed by the hybrid theory.
  2. The central efficiency claim (Abstract; §6–7) is that the main advantage over traditional mixed GMsFEM is offline basis-generation cost. Section 6 reports network FLOPs (7.6631 GFLOPs) and qualitative statements, but does not give wall-clock timings that compare classical local snapshot/spectral construction of R_off against network inference plus assembly for the same mm, mesh, and hardware. Without that comparison the efficiency claim remains unquantified.
  3. Experiments use a single KLE fractured ensemble, fixed 128 imes128 fine / 4 imes4 coarse grids, and primarily mm=6 (Appendices A–B vary mm but not media class). Stability under “strong heterogeneity and high contrast” is therefore shown only for this family. At least one additional permeability class (e.g., SPE10-type channelized or non-fractured high-contrast fields) and a brief sensitivity check on coarse resolution would be needed to support the generality stated in the abstract and conclusion.
minor comments (5)
  1. Several typos and notation slips: “Mathemaitcs”, “Efenfiev”, “GNsFEM”, “Jourmal”, “Hydrilogy”, “Marhine”, “quasi-image” for the 128 imes128 reshape of concatenated bases; also “ˆΨrenotes” and inconsistent use of mm vs. number of bases.
  2. Figure 2 caption repeats “Right:” for both panels; clarify left/right content.
  3. Remark 3.1 states sigmoid attention is used; a one-line ablation or justification versus softmax would help readers assess the design choice.
  4. Credit statement lists “Shubin Fu” while the author list has “Shibin Fu”; align names.
  5. Equation (9) and the subsequent assembly discussion would benefit from an explicit statement of how predicted bases are orthonormalized or scaled before forming R̂_off, if any post-processing is applied.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: supervised approximation of classical mixed-GMsFEM bases followed by independent Galerkin assembly and two-grid solve; mild author self-citations are not load-bearing.

full rationale

The derivation chain is: (i) classical mixed GMsFEM constructs snapshot/offline bases via local solves and spectral problems (Sec. 2.2, Eqs. 7–9); (ii) an attention-enhanced FNO+U-Net is trained supervised on those bases (loss Eqs. 17–20, data from KLE+GMsFEM); (iii) predicted bases form R̂off, the global mixed system is assembled exactly as in classical GMsFEM (Eq. 8/9), and pressure is obtained by the standard two-grid iteration (Alg. 1). Final pressure accuracy is therefore a downstream numerical consequence of the predicted bases, not a quantity fitted or defined by the network. Theorems 5.3–5.12 are ordinary perturbation arguments that bound Galerkin and two-grid errors by the (externally measurable) basis error ε_basis under Assumption 5.5; they do not redefine the solution in terms of the network output. Self-citations to prior author pipelines ([3–6,19,28] etc.) supply background or the classical approximation property of Q_off, but none is invoked as a uniqueness theorem that forces the hybrid construction or the pressure claims. Empirical comparisons (Tables 2, Figs. 6–7, Appendices) are against held-out classical solutions and pure-learning baselines, so the central accuracy/efficiency claims remain independently checkable. Score 1 reflects only the routine presence of author self-citations that are not load-bearing for the result.

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

The central claim rests on classical mixed GMsFEM and two-grid theory plus supervised learning of offline bases. Free knobs (gradient-loss weight, Fourier modes, mm, training split) are chosen by hand. Load-bearing modeling axioms include local spectral offline construction, Galerkin orthogonality, stability of the predicted offline space, and small aggregate local basis error. The hybrid network is an engineered entity, not a new physical object; independent evidence is only the reported numerical tests.

free parameters (5)
  • gradient loss weight λ_grad = 0.1
    Hand-set to 0.1 in Table 1; balances MSE vs gradient consistency and affects basis interface recovery.
  • Fourier truncated modes (m1, m2) = 16 (x), 16 (y)
    Chosen as 16×16; controls spectral capacity of the operator front-end.
  • number of multiscale bases mm = 6 (main)
    Main experiments use mm=6 (plus appendices for other mm); trades accuracy vs system size.
  • Adam learning rate and epochs = 0.001 / 60
    lr=0.001, 60 epochs; training schedule not derived from theory.
  • train/test split size = 3500 train / 500 test
    3500/500 of 4000 KLE samples; empirical data budget.
assumptions (6)
  • domain assumption Mixed GMsFEM offline bases from local snapshot + spectral problems adequately represent high-contrast Darcy structure on the coarse space.
    §2.2 construction and use of R_off; standard GMsFEM premise.
  • ad hoc to paper Aggregate local basis-error operator norms are bounded: sum_i ||E_i||^2 ≤ ε_loc^2, with ε_basis small enough for perturbation of C1.
    Assumption (23) and Thms. 5.3, 5.7, 5.10; not verified numerically per sample.
  • ad hoc to paper Predicted offline space has same dimension and Riesz-type stability constants c_R, C_R as true offline space (Assumption 5.5).
    Needed to control coefficient norms and approximation property of Q̂_off.
  • standard math Galerkin orthogonality and energy-norm approximation theory for the mixed/pressure-eliminated system.
    Thm. 5.4 and Lemma 5.6 citing prior multiscale theory [19].
  • domain assumption Two-grid error propagation with ILU(0) smoother behaves as in classical estimates (Lemma 5.9).
    §4–5; smoother choice Remark 4.1.
  • domain assumption KLE-generated fractured permeability ensemble is representative of the high-contrast regime of interest.
    §6.1 data generation; all reported metrics are on this synthetic family.
invented entities (1)
  • Attention-enhanced hybrid network G_Θ (FNO with frequency-wise attention + Attention U-Net refinement) for multiscale basis prediction
    purpose: Replace repeated offline local spectral basis construction by a learned map κ ↦ Ψ.
    Architecture is engineered for this paper; evidence is empirical MSE/R² and downstream pressure errors only.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Applying Two-Grid Preconditioner for Subsurface Flow Simulation using Attention-enhanced Hybrid Network to Accelerate Multiscale Discretization in High-contrast Media." pith.science (2026). https://pith.science/paper/EF7QLC3F

@misc{pith2026260602582,
  author       = {Pith},
  title        = {Pith review of: Applying Two-Grid Preconditioner for Subsurface Flow Simulation using Attention-enhanced Hybrid Network to Accelerate Multiscale Discretization in High-contrast Media},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EF7QLC3F}},
  note         = {Machine review of arXiv:2606.02582}
}
read the original abstract

In this paper, we study the efficient numerical solution of Darcy equations in strongly heterogeneous media with high-contrast permeability and propose a hybrid framework that combines learning with multiscale numerical methods. The learning component is used for the prediction of multiscale basis functions in the mixed generalized multiscale finite element method (mixed GMsFEM), with the goal of reducing the repeated local computations required in the offline stage. Once these basis functions are predicted, the global system is assembled and the pressure field is computed by a two-grid preconditioned solver. The resulting method accelerates the costly local basis-construction stage while retaining the multiscale discretization and preconditioned iterative structure of the underlying solver. Numerical experiments on two-dimensional heterogeneous Darcy problems show that the proposed framework yields more accurate final pressure reconstruction than several representative learning-based methods and remains stable under strong heterogeneity and high-contrast coefficients. In comparison with the traditional mixed GMsFEM, its main advantage lies in the efficiency of the basis-generation stage, while the quality of the global solve is still ensured by the two-grid preconditioner. These results indicate that accelerating multiscale basis construction through learning, while preserving a mature numerical solver for the global problem, provides a viable approach for high-resolution Darcy-type simulations.

Figures

Figures reproduced from arXiv: 2606.02582 by the authors.

Figure 1
Figure 1. Schematic illustration of the overall technical pipeline of the proposed framework. [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Example of our dataset. Right: high-contrast fractured permeability field. The yellow lines refer [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of multiscale basis functions (2-12). For each two lines, the top denotes the [PITH_FULL_IMAGE:figures/full_fig_p021_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Learning curves of multiscale basis functions. [PITH_FULL_IMAGE:figures/full_fig_p022_4.png]
Figure 5
Figure 5. Figure 5: Schematic diagrams of sparse matrix in the linear system [PITH_FULL_IMAGE:figures/full_fig_p023_5.png]
Figure 6
Figure 6. Figure 6: Comparison of our model and baseline models. Left: reference solution; Right Top: pressure [PITH_FULL_IMAGE:figures/full_fig_p024_6.png]
Figure 7
Figure 7. Figure 7: Relative L2 error of ours and baseline models [PITH_FULL_IMAGE:figures/full_fig_p024_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 19 canonical work pages

  1. [1]

    An online method for interpolating linear parametric reduced-order models

    Amsallem, D., Farhat, C., 2011. An online method for interpolating linear parametric reduced-order models. SIAM Journal on Scientific Computing 33, 2169–2198. doi:10. 1137/100813051

    2011

  2. [2]

    Dynamics of Fluids in Porous Media

    Bear, J., 2013. Dynamics of Fluids in Porous Media. Courier Corporation

    2013

  3. [3]

    Generalized multiscale approximation of mixed finite elements with velocity elimination for subsurface flow

    Chen, J., Chung, E.T., He, Z., Sun, S., 2020. Generalized multiscale approximation of mixed finite elements with velocity elimination for subsurface flow. Jourmal of Com- putational Physics 404. doi:10.1016/j.jcp.2019.109133

    work page doi:10.1016/j.jcp.2019.109133 2020

  4. [4]

    Prediction of multiscale features using deep learning-based preconditioner-solver architecture for darcy equation in high-contrast media

    Chen, J., Li, P., He, Z., Hands, S., 2025. Prediction of multiscale features using deep learning-based preconditioner-solver architecture for darcy equation in high-contrast media. arXiv preprint doi:10.48550/arXiv.2411.02431

    work page doi:10.48550/arxiv.2411.02431 2025

  5. [5]

    An innovative application of deep learning in multiscale modeling of subsurface fluid flow: Reconstructing the basis functions of the mixed gmsfem

    Choubineh, A., Chen, J., Coenen, F., Ma, F., 2022. An innovative application of deep learning in multiscale modeling of subsurface fluid flow: Reconstructing the basis functions of the mixed gmsfem. Journal of Petroleum Science and Engineering 216, 110751. doi:10.1016/j.petrol.2022.110751

    work page doi:10.1016/j.petrol.2022.110751 2022

  6. [6]

    Deep ensemble learning for high-dimensional subsurface fluid flow modeling

    Choubineh, A., Chen, J., Wood, D.A., Coenen, F., Ma, F., 2023. Deep ensemble learning for high-dimensional subsurface fluid flow modeling. Engineering Applications of Artificial Intelligence 126, 106968. doi:10.1016/j.engappai.2023.106968

    work page doi:10.1016/j.engappai.2023.106968 2023

  7. [7]

    Constraint energy minimizing gener- alized multiscale finite element method

    Chung, E.T., Efendiev, Y., Leung, W.T., 2018a. Constraint energy minimizing gener- alized multiscale finite element method. Computer Methods in Applied Mechanics and Engineering 339, 298–319. doi:10.1016/j.cma.2018.04.010

    doi:10.1016/j.cma.2018.04.010 2018

  8. [8]

    An adaptive gmsfem for high-contrast flow problems

    Chung, E.T., Efendiev, Y., Li, G., 2014. An adaptive gmsfem for high-contrast flow problems. Journal of Computational Physics 273, 54–76. doi:10.1016/j.jcp.2014. 05.007

    doi:10.1016/j.jcp.2014 2014

Show all 43 references
  1. [9]

    Mixed generalized multiscale finite element methods and applications

    Chung, E.T., Efenfiev, Y., Lee, C.S., 2015. Mixed generalized multiscale finite element methods and applications. SIAM Journal on Scientific Computing 13, 338–366. doi:10. 1137/140970574

    2015

  2. [10]

    Fast online generalized multiscale finite element method using constraint energy minimization

    Chung, E.T., Efenfiev, Y., Leung, W.T., 2018b. Fast online generalized multiscale finite element method using constraint energy minimization. Journal of Computational Physics 355, 450–463. doi:10.1016/j.jcp.2017.11.022. 25

    2017 doi

  3. [11]

    Generalized multiscale finite element methods with energy minimizing oversampling

    Chung, E.T., Efenfiev, Y., Leung, W.T., 2018c. Generalized multiscale finite element methods with energy minimizing oversampling. International Journal for Numerical Methods in Engineering 117, 316–343. doi:10.1002/nme.5958

    doi

  4. [12]

    Two-level overlapping schwarz algo- rithms for a staggered discontinuous galerkin method

    Chung, E.T., Kim, H.H., Widlund, O.B., 2013. Two-level overlapping schwarz algo- rithms for a staggered discontinuous galerkin method. SIAM Journal on Numerical Analysis 51, 47–67. doi:10.1137/110849432

    2013 doi

  5. [13]

    A preconditioner for substructuring based on constrained en- ergy minimization

    Dohrmann, C.R., 2003. A preconditioner for substructuring based on constrained en- ergy minimization. SIAM Journal on Scientific Computing 25, 246–258. doi:10.1137/ S1064827502412887

    2003

  6. [14]

    Generalized multiscale finite element methods (gmsfem)

    Efendiev, Y., Galvis, J., Hou, T.Y., 2013. Generalized multiscale finite element methods (gmsfem). Journal of Computational Physics 251, 116–135. doi:j.jcp.2013.04.045

    2013

  7. [15]

    Multiscale finite element methods for porous media flows and their applications

    Efendiev, Y., Hou, T.Y., 2007. Multiscale finite element methods for porous media flows and their applications. Applied Numerical Mathematics 57, 577–596. doi:10. 1016/j.apnum.2006.07.009

    2007

  8. [16]

    Multiscale finite element methods: theory and applica- tions

    Efendiev, Y., Hou, T.Y., 2009. Multiscale finite element methods: theory and applica- tions. Springer

    2009

  9. [17]

    Convergence of a nonconforming multiscale finite element method

    Efendiev, Y., Hou, T.Y., Wu, X.h., 2000. Convergence of a nonconforming multiscale finite element method. SIAM Journal on Numerical Analysis 37, 888–910. doi:10.1137/ S0036142997330329

    2000

  10. [18]

    An adaptive preconditioner for three-dimensional single-phase compressible flow in highly heterogeneous porous media

    Fu, S., Chung, E.T., Zhao, L., 2024a. An adaptive preconditioner for three-dimensional single-phase compressible flow in highly heterogeneous porous media. Multiscale Mod- eling & Simulation 22, 155–177. doi:10.1137/22M1529075

    doi

  11. [19]

    An efficient multiscale preconditioner for large- scale highly heterogeneous flow

    Fu, S., Chung, E.T., Zhao, L., 2024b. An efficient multiscale preconditioner for large- scale highly heterogeneous flow. SIAM Journal on Scientific Computing 46, S352–S377. doi:10.1137/22M1502859

    doi

  12. [20]

    Application of the karhunen-loeve expansion to feature selection and ordering

    Fukunaga, K., Koontz, W.L., 1970. Application of the karhunen-loeve expansion to feature selection and ordering. IEEE Transactions on computers 100, 311–318. doi:10. 1109/T-C.1970.222918

    1970

  13. [21]

    Domain decomposition preconditioners for multiscale flows in high-contrast media

    Galvis, J., Efendiev, Y., 2010. Domain decomposition preconditioners for multiscale flows in high-contrast media. Multiscale Modeling & Simulation 8, 1461–1483. doi:10. 1137/090751190

    2010

  14. [22]

    A multiscale finite element method for numerical ho- mogenization

    Grégoire, A., Robert, B., 2005. A multiscale finite element method for numerical ho- mogenization. Multiscale Modeling & Simulation 4, 790–812. doi:10.1137/040611239. 26

    2005 doi

  15. [23]

    A family of multiscale hybrid-mixed finite element methods for the darcy equation with rough coefficients

    Harder, C., Paredes, D., Valentin, F., 2013. A family of multiscale hybrid-mixed finite element methods for the darcy equation with rough coefficients. Journal of Computa- tional Physics 245, 107–130. doi:10.1016/j.jcp.2013.03.019

    2013 doi

  16. [24]

    A multiscale finite element method for elliptic problems in composite materials and porous media

    Hou, T.Y., Wu, X.h., 1997. A multiscale finite element method for elliptic problems in composite materials and porous media. Journal of Computational Physics 134, 169–189. doi:10.1006/jcph.1997.5682

    1997 doi

  17. [25]

    Homogenization of differential operators and integral functionals

    Jikov, VV abd Kozlov, S., Oleinik, O., 2012. Homogenization of differential operators and integral functionals. Springer

    2012

  18. [26]

    Adam: A method for stochastic optimization

    Kingma, D.P., Ba, J., 2014. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980 doi:10.48550/arXiv.1412.6980

    2014 doi arXiv

  19. [27]

    Neural operator: Learning maps between function spaces with applications to pdes

    Kovachki, N., Li, Z., Liu, B., Azizzadenesheli, K., Bhattacharya, K., Stuart, A., Anandkumar, A., 2023. Neural operator: Learning maps between function spaces with applications to pdes. Journal of Machine Learning Research 24, 1–97. URL: http://jmlr.org/papers/v24/21-1524.html

    2023

  20. [28]

    Dual-domain deep learning method to accelerate local basis functions computation for reservoir simulation in high-contrast porous media

    Li, P., Chen, J., 2025a. Dual-domain deep learning method to accelerate local basis functions computation for reservoir simulation in high-contrast porous media. arXiv preprint doi:10.48550/arXiv.2511.20685

    doi

  21. [29]

    Hybrid two-stage reconstruction of multiscale subsur- face flow with physics-informed residual connected neural operator

    Li, P., Chen, J., 2025b. Hybrid two-stage reconstruction of multiscale subsur- face flow with physics-informed residual connected neural operator. arXiv preprint doi:10.48550/arXiv.2501.13271

    doi

  22. [30]

    Multiscale subsurface flow in the era of deep learning: A review

    Li, P., Chen, J., Zhang, H., Hands, S., 2026. Multiscale subsurface flow in the era of deep learning: A review. SSRN doi:10.2139/ssrn.6519058

    2026 doi

  23. [31]

    Fourier neural operator for parametric partial differential equations

    Li, Z., Kovachki, N., Azizzadenesheli, K., Liu, B., Bhattacharya, K., Stuart, A., Anand- kumar, A., 2020. Fourier neural operator for parametric partial differential equations. arXiv preprint arXiv:2010.08895 doi:10.48550/arXiv.2010.08895

    2020 doi arXiv

  24. [32]

    Physics-informed neural operator for learning partial differential equations

    Li, Z., Zheng, H., Kovachki, N., Jin, D., Chen, H., Liu, B., Azizzadenesheli, K., Anand- kumar, A., 2024. Physics-informed neural operator for learning partial differential equations. ACM/IMS Journal of Data Science 1, 1–27. doi:10.1145/3648506

    2024 doi

  25. [33]

    Adaptive pseudo-solution-assisted pinns for multiscale darcy and darcy-forchheimer equations

    Liu, H., Bi, C., He, Z., Yang, M., 2026. Adaptive pseudo-solution-assisted pinns for multiscale darcy and darcy-forchheimer equations. International Journal of Computer Mathematics 103, 180–198. doi:10.1080/00207160.2025.2548924

    2026 doi

  26. [34]

    Learning a multiscale prolongation operator

    Liu, Y., Fu, S., Ye, C., Chung, E.T., 2025. Learning a multiscale prolongation operator. arXiv preprint doi:10.48550/arXiv.2410.06832. 27

    2025 doi arXiv

  27. [35]

    Learning nonlinear operators via deeponet based on the universal approximation theorem of operators

    Lu, L., Jin, P., Zhang, Z., Karniadakis, G.E., 2021. Learning nonlinear operators via deeponet based on the universal approximation theorem of operators. Nature Marhine Intelligence 3, 219–229. doi:10.1038/s42256-021-00302-5

    2021 doi

  28. [36]

    A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data

    Lu, L., Meng, X., Cai, S., Mao, Z., Goswami, S., Zhang, Z., Karniadakis, G.E., 2022. A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data. Computer Methods in Applied Mechanics and Engineering 393. doi:10.1016/j.cma.2022.114778

    2022 doi

  29. [37]

    Reduced order methods for modeling and computa- tional reduction

    Quarteroni, A., Rozza, G., 2014. Reduced order methods for modeling and computa- tional reduction. Springer

    2014

  30. [38]

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

    Rassi, M., Perdikaris, P., Karniadakis, G., 2019. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics 378, 686–707. doi:10. 1016/j.jcp.2018.10.045

    2019

  31. [39]

    U-net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015

    Ronneberger, O., Fischer, P., Brox, T., 2015. U-net: Convolutional networks for biomedical image segmentation, in: Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015. doi:10.1007/978-3-319-24574-4_28

    2015 doi

  32. [40]

    Upscaling hydraulic conductivities in heteroge- neous media: An overview

    Wen, X., Gomez-Hernandez, J., 1996. Upscaling hydraulic conductivities in heteroge- neous media: An overview. Journal of Hydrilogy 183. doi:10.1016/S0022-1694(96) 80030-8

    1996 doi

  33. [41]

    A two-grid preconditioner with an adaptive coarse space for flow simulations in highly heterogeneous media

    Yang, Y., Fu, S., Chung, E.T., 2019. A two-grid preconditioner with an adaptive coarse space for flow simulations in highly heterogeneous media. Journal of Computational Physics 391, 1–13. doi:10.1016/j.jcp.2019.03.038

    2019 doi

  34. [42]

    An adaptive generalized multiscale finite element methodbasedtwo-gridpreconditionerforlargescalehigh-contrastlinearelasticityprob- lems

    Yang, Y., Fu, S., Chung, E.T., 2022. An adaptive generalized multiscale finite element methodbasedtwo-gridpreconditionerforlargescalehigh-contrastlinearelasticityprob- lems. Journal of Scientific Computing 92. doi:10.1007/s10915-022-01869-w

    2022 doi

  35. [43]

    A robust two-level overlapping precon- ditioner for darcy flow in high-contrast media

    Ye, C., Fu, S., Chung, E.T., Huang, J., 2024. A robust two-level overlapping precon- ditioner for darcy flow in high-contrast media. SIAM Journal on Scientific Computing 46, A3151–A3176. doi:10.1137/23M1558987. 28 Appendix A. GMsFEM-based Solutions Figure A.8: Schematic diagra...

    2024 doi

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.