Pith. sign in

REVIEW 2 major objections 2 minor 33 references

Operator Learning on the Data-Driven Multiscale Space for Nonlinear Flow in Random Heterogeneous Porous Media

T0 review · 2 major / 2 minor · reviewed 2026-06-25 · grok-4.3

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

Pith's one-line read Neural operator learns global mapping from permeability fields to solution coefficients using data-driven multiscale trunk.

desk verdict The paper puts a data-driven multiscale basis into the trunk of a neural operator for nonlinear flow, a concrete new combination, but the claim that local snapshots span the global nonlinear manifold is the part that still needs verification. read the letter →

arxiv 2606.25820 v1 pith:MON2NTW6 submitted 2026-06-24 math.NA cs.NA

Maria Vasilyeva , Raphael Pangilinan This is my paper · ORCID
classification math.NAcs.NA
keywords operatorlearningmultiscalemethodsporousmediaflownonlinearneuralnetworksheterogeneousreduced-ordermodelingdata-drivenbasis
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

The paper constructs a low-dimensional multiscale space from local fine-scale solution snapshots and uses it as the trunk of a neural operator whose branch network predicts the corresponding coefficients from an input permeability field. This setup learns a direct nonlinear map between permeability and reduced solution coefficients for nonlinear flow problems in random heterogeneous porous media. Traditional Galerkin projection methods instead require repeated online nonlinear solves on the coarse grid plus coefficient evaluations at each step. The neural approach therefore removes those online costs while aiming for higher accuracy through a global learned mapping. A sympathetic reader would care because simulations of high-contrast flows become feasible for many random realizations without the usual reduced-order overhead.

What carries the argument

Data-driven multiscale space from local fine-scale solution snapshots, acting as trunk network in the neural operator to represent the solution manifold.

What would settle it

A permeability realization drawn from a distribution different from the training snapshots produces a neural-operator prediction whose L2 error against a fine-scale reference solution exceeds the error of the corresponding Galerkin projection method.

Watch

Extended reading notes

Core claim

The multiscale space built from representative local fine-scale snapshots serves as the trunk network of a neural operator while a branch network maps the permeability field to the reduced coefficients; this produces a global nonlinear mapping from permeability to solution that avoids the online nonlinear coarse-grid solves and coefficient evaluations required by Galerkin projection methods, yielding improved accuracy and substantially lower computational cost for nonlinear flow in high-contrast heterogeneous media.

Load-bearing premise

The multiscale space constructed from local representative fine-scale solution snapshots yields an accurate low-dimensional representation of the solution manifold.

Editorial extensions

If this is right

  • The neural operator eliminates online nonlinear coarse-grid solves and coefficient evaluations.
  • Accuracy improves relative to projection-based reduced-order models for the same coarse dimension.
  • Computational cost drops substantially for repeated solves over random permeability fields.
  • The learned global mapping supplies greater flexibility than local projection steps.

Reading between the lines

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

  • The same trunk space could be reused across different nonlinear constitutive relations if the solution manifold remains similar.
  • Accuracy on permeability fields whose correlation length differs markedly from the snapshot ensemble would test generalization beyond the paper's reported cases.
  • Combining the learned coefficients with a separate time-stepping scheme might allow extension to unsteady nonlinear flows without retraining the operator.
  • The approach suggests that data-driven bases can replace analytic multiscale bases when the fine-scale physics is too complex for closed-form upscaling.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper introduces an operator learning framework for nonlinear flow in random heterogeneous porous media. A data-driven multiscale space is built from local representative fine-scale solution snapshots and used as the trunk of a neural operator; a branch network then maps input permeability fields to the reduced coefficients in this space. The approach is positioned as superior to Galerkin projection methods because it learns a global nonlinear mapping, improves accuracy, and avoids online nonlinear coarse-grid solves and coefficient evaluations. Numerical experiments are reported to demonstrate good accuracy at substantially lower computational cost than projection-based alternatives.

Significance. If the central assumption holds, the method could provide a practical route to reduced-cost operator learning for nonlinear multiscale problems by replacing repeated coarse-grid nonlinear solves with a trained branch network. The combination of snapshot-based multiscale bases with neural operators is a natural extension of existing reduced-order and operator-learning techniques and could be useful in uncertainty quantification or optimization settings where many forward solves are required.

major comments (2)
  1. [Method description / multiscale space construction] The claim that the multiscale space constructed from local snapshots yields an accurate low-dimensional representation of the global nonlinear solution manifold is load-bearing for all stated advantages over Galerkin methods. The manuscript provides no explicit manifold approximation error analysis, Kolmogorov n-width bounds, or quantitative verification that local snapshots suffice when global heterogeneity and nonlinearity interact across the domain.
  2. [Numerical experiments] Numerical results are asserted to show improved accuracy and lower cost, yet the comparison with Galerkin projection lacks reported error tables (e.g., relative L2 or energy-norm errors) and timing breakdowns across multiple random realizations that would substantiate the elimination of online nonlinear solves.
minor comments (2)
  1. Notation for the trunk and branch networks should be introduced with explicit functional definitions (e.g., the precise form of the operator approximation) rather than relying on standard DeepONet terminology.
  2. The abstract and introduction should cite the specific nonlinear model (e.g., the exact form of the Forchheimer or power-law term) used in the experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major comment point-by-point below and indicate the planned revisions.

read point-by-point responses

  1. Referee: The claim that the multiscale space constructed from local snapshots yields an accurate low-dimensional representation of the global nonlinear solution manifold is load-bearing for all stated advantages over Galerkin methods. The manuscript provides no explicit manifold approximation error analysis, Kolmogorov n-width bounds, or quantitative verification that local snapshots suffice when global heterogeneity and nonlinearity interact across the domain.

    Authors: We acknowledge that the manuscript does not contain explicit theoretical manifold approximation error analysis or Kolmogorov n-width bounds. Deriving such bounds for the nonlinear case with interacting global heterogeneity is non-trivial and falls outside the primary scope of the work, which centers on the practical construction and operator-learning use of the snapshot-derived basis. To directly address the concern about verification, the revised manuscript will add quantitative numerical experiments that compute the approximation error of the local-snapshot multiscale space against global fine-scale solutions drawn from multiple independent realizations, thereby demonstrating that the basis remains effective when heterogeneity and nonlinearity interact across the domain. revision: yes

  2. Referee: Numerical results are asserted to show improved accuracy and lower cost, yet the comparison with Galerkin projection lacks reported error tables (e.g., relative L2 or energy-norm errors) and timing breakdowns across multiple random realizations that would substantiate the elimination of online nonlinear solves.

    Authors: While the manuscript already reports numerical accuracy and cost comparisons, we agree that the presentation can be strengthened. The revised version will include explicit tables of relative L2 and energy-norm errors for both the proposed operator-learning method and the Galerkin projection baseline, evaluated across multiple random permeability realizations. Detailed timing breakdowns will also be added to quantify the savings obtained by replacing online nonlinear coarse-grid solves with the trained branch network. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity in data-driven multiscale operator learning

full rationale

The paper presents a data-driven construction of a multiscale basis from local fine-scale solution snapshots, followed by a neural operator (branch-trunk) that learns a mapping from permeability fields to reduced coefficients. No step in the provided description reduces a claimed prediction or uniqueness result to a fitted input or self-citation by construction; the trunk is explicitly built from external snapshot data, and the claimed advantages over Galerkin projection are framed as outcomes of the learned nonlinear mapping rather than tautological redefinitions. The central premise relies on standard neural operator techniques and snapshot-based dimensionality reduction, which remain independent of the target solution manifold accuracy.

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

No specific free parameters, axioms, or invented entities are identifiable from the abstract alone.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Operator Learning on the Data-Driven Multiscale Space for Nonlinear Flow in Random Heterogeneous Porous Media." pith.science (2026). https://pith.science/paper/MON2NTW6

@misc{pith2026260625820,
  author       = {Pith},
  title        = {Pith review of: Operator Learning on the Data-Driven Multiscale Space for Nonlinear Flow in Random Heterogeneous Porous Media},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MON2NTW6}},
  note         = {Machine review of arXiv:2606.25820}
}
read the original abstract

We present an operator learning framework based on a coarse data-driven multiscale space for nonlinear flow in random heterogeneous porous media. The multiscale space is constructed from local representative fine-scale solution snapshots, yielding an accurate low-dimensional representation of the solution manifold. This multiscale basis serves as the trunk of a neural operator, while a branch network predicts the corresponding reduced coefficients from the input permeability field. Unlike Galerkin projection methods, the neural operator learns a global nonlinear mapping from permeability fields to solution coefficients, providing greater flexibility, improved accuracy, and eliminating the need for online nonlinear coarse-grid solves and coefficient evaluations. Numerical results show that the proposed approach achieves good accuracy and substantially lower computational cost than projection-based methods for nonlinear flow in high-contrast heterogeneous media.

Figures

Figures reproduced from arXiv: 2606.25820 by the authors.

Figure 1
Figure 1. Test 1 (KLE-based heterogeneous media) for two permeability fields [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Test 2 (Channelized media) for two permeability fields [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Reference solution (True), neural prediction (Pred), corrected solution after 5 iterations [PITH_FULL_IMAGE:figures/full_fig_p017_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 7 canonical work pages

  1. [1]

    Nonlinear model order reduction based on local reduced-order bases.International Journal for Numerical Methods in Engineering, 92(10):891–916, 2012

    David Amsallem, Matthew J Zahr, and Charbel Farhat. Nonlinear model order reduction based on local reduced-order bases.International Journal for Numerical Methods in Engineering, 92(10):891–916, 2012

    2012

  2. [2]

    Geometric multi- grid methods for darcy–forchheimer flow in fractured porous media.Computers & Mathematics with Applications, 78(9):3139–3151, 2019

    Andr´ es Arrar´ as, Francisco Jos´ e Gaspar, Laura Portero, and Carmen Rodrigo. Geometric multi- grid methods for darcy–forchheimer flow in fractured porous media.Computers & Mathematics with Applications, 78(9):3139–3151, 2019

    2019

  3. [3]

    Analysis of generalized forchheimer flows of compressible fluids in porous media.Journal of Mathematical Physics, 50(10), 2009

    Eugenio Aulisa, Lidia Bloshanskaya, Luan Hoang, and Akif Ibragimov. Analysis of generalized forchheimer flows of compressible fluids in porous media.Journal of Mathematical Physics, 50(10), 2009

    2009

  4. [4]

    SIAM, 2017

    Peter Benner, Mario Ohlberger, Albert Cohen, and Karen Willcox.Model reduction and approximation: theory and algorithms. SIAM, 2017

    2017

  5. [5]

    An introduction to the proper orthogonal decomposition.Current science, pages 808–817, 2000

    Anindya Chatterjee. An introduction to the proper orthogonal decomposition.Current science, pages 808–817, 2000

    2000

  6. [6]

    Discrete empirical interpolation for nonlinear model reduction

    Saifon Chaturantabut and Danny C Sorensen. Discrete empirical interpolation for nonlinear model reduction. InProceedings of the 48h IEEE Conference on Decision and Control (CDC) held jointly with 2009 28th Chinese Control Conference, pages 4316–4321. IEEE, 2009

    2009

  7. [7]

    An adaptive gmsfem for high-contrast flow problems.Journal of Computational Physics, 273:54–76, 2014

    Eric T Chung, Yalchin Efendiev, and Guanglian Li. An adaptive gmsfem for high-contrast flow problems.Journal of Computational Physics, 273:54–76, 2014

    2014

  8. [8]

    Generalized multiscale finite element methods.Journal of Computational Physics, 251:116–135, 2013

    Yalchin Efendiev, Juan Galvis, and Thomas Y Hou. Generalized multiscale finite element methods.Journal of Computational Physics, 251:116–135, 2013. 18

    2013

Show all 33 references
  1. [9]

    Springer, 2009

    Yalchin Efendiev and Thomas Y Hou.Multiscale finite element methods: theory and applica- tions, volume 4. Springer, 2009

    2009

  2. [10]

    Spectral neural operators

    Vladimir Sergeevich Fanaskov and Ivan V Oseledets. Spectral neural operators. InDoklady Mathematics, volume 108, pages S226–S232. Springer, 2023

    2023

  3. [11]

    Localized model order reduction in porous media flow simulation.Journal of Petroleum Science and Engineering, 145:689–703, 2016

    Mohamadreza Ghasemi and Eduardo Gildin. Localized model order reduction in porous media flow simulation.Journal of Petroleum Science and Engineering, 145:689–703, 2016

    2016

  4. [12]

    Model order reduction in porous media flow simulation using quadratic bilinear formulation.Computational Geosciences, 20(3):723–735, 2016

    Mohammadreza Ghasemi and Eduardo Gildin. Model order reduction in porous media flow simulation using quadratic bilinear formulation.Computational Geosciences, 20(3):723–735, 2016

    2016

  5. [13]

    Springer, 2016

    Jan S Hesthaven, Gianluigi Rozza, and Benjamin Stamm.Certified reduced basis methods for parametrized partial differential equations. Springer, 2016

    2016

  6. [14]

    Operator learning: Algorithms and analysis.Handbook of Numerical Analysis, 25:419–467, 2024

    Nikola B Kovachki, Samuel Lanthaler, and Andrew M Stuart. Operator learning: Algorithms and analysis.Handbook of Numerical Analysis, 25:419–467, 2024

    2024

  7. [15]

    Fourier neural operator for parametric partial differential equations.arXiv preprint arXiv:2010.08895, 2020

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations.arXiv preprint arXiv:2010.08895, 2020

    2010 arXiv

  8. [16]

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

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learning nonlinear operators via deeponet based on the universal approximation theorem of operators. Nature machine intelligence, 3(3):218–229, 2021

    2021

  9. [17]

    A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data.Computer Methods in Applied Mechanics and Engineering, 393:114778, 2022

    Lu Lu, Xuhui Meng, Shengze Cai, Zhiping Mao, Somdatta Goswami, Zhongqiang Zhang, and George Em Karniadakis. A comprehensive and fair comparison of two neural operators (with practical extensions) based on fair data.Computer Methods in Applied Mechanics and Engineering, 393:114...

    2022

  10. [18]

    Neural-pod: A plug-and-play neural opera- tor framework for infinite-dimensional functional nonlinear proper orthogonal decomposition

    Changhong Mou, Binghang Lu, and Guang Lin. Neural-pod: A plug-and-play neural opera- tor framework for infinite-dimensional functional nonlinear proper orthogonal decomposition. arXiv preprint arXiv:2602.15632, 2026

    2026

  11. [19]

    Learning two-phase microstructure evolution using neural operators and autoencoder architectures.npj Computational Materials, 8(1):190, 2022

    Vivek Oommen, Khemraj Shukla, Somdatta Goswami, R´ emi Dingreville, and George Em Kar- niadakis. Learning two-phase microstructure evolution using neural operators and autoencoder architectures.npj Computational Materials, 8(1):190, 2022

    2022

  12. [20]

    Pytorch: An impera- tive style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. Pytorch: An impera- tive style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019

    2019

  13. [21]

    Springer, 2015

    Alfio Quarteroni, Andrea Manzoni, and Federico Negri.Reduced basis methods for partial differential equations: an introduction. Springer, 2015

    2015

  14. [22]

    Certified reduced basis approximation for parametrized partial differential equations and applications.Journal of Mathematics in Industry, 1(1):3, 2011

    Alfio Quarteroni, Gianluigi Rozza, and Andrea Manzoni. Certified reduced basis approximation for parametrized partial differential equations and applications.Journal of Mathematics in Industry, 1(1):3, 2011. 19

    2011

  15. [23]

    Locally subspace-informed neural operators for efficient multiscale pde solving.arXiv preprint arXiv:2505.16030, 2025

    Alexander Rudikov, Vladimir Fanaskov, Sergei Stepanov, Buzheng Shan, Ekaterina Muravleva, Yalchin Efendiev, and Ivan Oseledets. Locally subspace-informed neural operators for efficient multiscale pde solving.arXiv preprint arXiv:2505.16030, 2025

    2025

  16. [24]

    Ensemble and mixture-of-experts deeponets for operator learning.arXiv preprint arXiv:2405.11907, 2024

    Ramansh Sharma and Varun Shankar. Ensemble and mixture-of-experts deeponets for operator learning.arXiv preprint arXiv:2405.11907, 2024

    2024

  17. [25]

    Wavelet neural operator: a neural operator for para- metric partial differential equations.arXiv preprint arXiv:2205.02191, 2022

    Tapas Tripura and Souvik Chakraborty. Wavelet neural operator: a neural operator for para- metric partial differential equations.arXiv preprint arXiv:2205.02191, 2022

    2022

  18. [26]

    Maria Vasilyeva, Aleksei Krasnikov, Kelum Gajamannage, and Mehrube Mehrubeoglu. Mul- tiscale method for image denoising using nonlinear diffusion process: Local denoising and spectral multiscale basis functions.Journal of Computational and Applied Mathematics, 470:116733, 2025

    2025

  19. [27]

    Learn- ing macroscopic parameters in nonlinear multiscale simulations using nonlocal multicontinua upscaling techniques.Journal of Computational Physics, 412:109323, 2020

    Maria Vasilyeva, Wing T Leung, Eric T Chung, Yalchin Efendiev, and Mary Wheeler. Learn- ing macroscopic parameters in nonlinear multiscale simulations using nonlocal multicontinua upscaling techniques.Journal of Computational Physics, 412:109323, 2020

    2020

  20. [28]

    Maria Vasilyeva, Nana Adjoah Mbroh, Denis Spiridonov, and Tatiana Timofeeva. An implicit- explicit scheme and adaptive multiscale approximation of generalized forchheimer flow in frac- tured porous media.Computers & Mathematics with Applications, 211:1–25, 2026

    2026

  21. [29]

    Maria Vasilyeva, Aleksei Tyrylgin, Donald L Brown, and Anirban Mondal. Preconditioning markov chain monte carlo method for geomechanical subsidence using multiscale method and machine learning technique.Journal of Computational and Applied Mathematics, 392:113420, 2021

    2021

  22. [30]

    Machine learning for accelerating macroscopic parame- ters prediction for poroelasticity problem in stochastic media.Computers & Mathematics with Applications, 84:185–202, 2021

    Maria Vasilyeva and Aleksey Tyrylgin. Machine learning for accelerating macroscopic parame- ters prediction for poroelasticity problem in stochastic media.Computers & Mathematics with Applications, 84:185–202, 2021

    2021

  23. [31]

    Scipy 1.0: fundamental algorithms for scientific computing in python.Nature methods, 17(3):261– 272, 2020

    Pauli Virtanen, Ralf Gommers, Travis E Oliphant, Matt Haberland, Tyler Reddy, David Cour- napeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, et al. Scipy 1.0: fundamental algorithms for scientific computing in python.Nature methods, 17(3):261– 272, 2020

    2020

  24. [32]

    Reduced-basis deep operator learning for parametric pdes with independently varying boundary and source data.arXiv preprint arXiv:2511.18260, 2025

    Yueqi Wang and Guang Lin. Reduced-basis deep operator learning for parametric pdes with independently varying boundary and source data.arXiv preprint arXiv:2511.18260, 2025

    2025

  25. [33]

    Finite element representation network (fern) for operator learning with a localized trainable basis.arXiv preprint arXiv:2510.26962, 2025

    Zecheng Zhang, Hao Liu, Guosheng Fu, Hayden Schaeffer, and Guang Lin. Finite element representation network (fern) for operator learning with a localized trainable basis.arXiv preprint arXiv:2510.26962, 2025. A Global Reduced Space LetSbe the centered snapshot matrix S= [p 1 −...

    2025

Pith tools

Reviewed June 25, 2026 · model on record in the stance chip above.