arxiv: 2602.17776 · v2 · submitted 2026-02-19 · 🧮 math.NA · cs.LG· cs.NA

Recognition: 2 theorem links

· Lean Theorem

Solving and learning advective multiscale Darcian dynamics with the Neural Basis Method

Yuhe Wang , Min Wang

Authors on Pith no claims yet

Pith reviewed 2026-05-15 20:32 UTC · model grok-4.3

classification 🧮 math.NA cs.LGcs.NA

keywords Neural Basis Methodadvective Darcian dynamicsoperator learningmultiscale flowsprojection methodresidual metricphysics-informed learningparametric inference

0 comments

The pith

The Neural Basis Method projects solutions onto a physics-conforming neural basis using an operator-induced residual metric to solve and learn advective multiscale Darcian dynamics.

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

The paper introduces the Neural Basis Method as a projection-based alternative to penalty-style physics-informed learning. It couples a predefined physics-conforming neural basis space with a residual metric derived from the governing operator, turning the problem into a well-conditioned deterministic minimization. This metric functions as a computable certificate that separates approximation error from enforcement error and stays stable when the basis is enriched. The approach is shown to deliver accurate single solves and to produce reduced coordinates suitable for fast operator learning across parametric instances of the Darcian problem. A reader would care because the formulation makes progress interpretable and controllable rather than relying on heuristic loss balancing.

Core claim

By projecting onto a predefined physics-conforming neural basis and minimizing an operator-induced residual metric, the method obtains accurate and robust solutions for advective multiscale Darcian dynamics in single solves while producing reduced coordinates that support effective parametric inference through operator learning; the residual metric supplies a stable certificate that distinguishes approximation error from enforcement error and remains reliable under basis enrichment.

What carries the argument

The Neural Basis Method: a projection onto a physics-conforming neural basis space whose objective is an operator-induced residual metric that acts as both loss and error certificate.

If this is right

Accurate and robust solutions are obtained for single instances of the advective multiscale Darcian problem.
Reduced coordinates from the projection become learnable across parametric instances, enabling fast operator inference.
The residual metric supplies a deterministic certificate that distinguishes approximation from enforcement error.
Stability of the minimization holds under successive enrichment of the neural basis space.

Where Pith is reading between the lines

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

The same projection-plus-metric structure could be tested on other multiscale advection-dominated PDEs without retraining the basis construction.
If the reduced coordinates prove transferable, the method may reduce the sample complexity of operator learning compared with standard physics-informed approaches.
The explicit separation of errors suggests a route to a posteriori error indicators that could guide adaptive basis enrichment.

Load-bearing premise

The operator-induced residual metric remains stable under basis enrichment and yields a computable certificate that separates approximation error from enforcement error.

What would settle it

Numerical experiments showing that the residual metric grows unbounded or fails to separate the two error types as the neural basis dimension increases would falsify the stability claim.

read the original abstract

Physics-governed models are increasingly paired with machine learning for accelerated predictions, yet most "physics--informed" formulations treat the governing equations as a penalty loss whose scale and meaning are set by heuristic balancing. This blurs operator structure, thereby confounding solution approximation error with governing-equation enforcement error and making the solving and learning progress hard to interpret and control. Here we introduce the Neural Basis Method, a projection-based formulation that couples a predefined, physics-conforming neural basis space with an operator-induced residual metric to obtain a well-conditioned deterministic minimization. Stability and reliability then hinge on this metric: the residual is not merely an optimization objective but a computable certificate tied to approximation and enforcement, remaining stable under basis enrichment and yielding reduced coordinates that are learnable across parametric instances. We use advective multiscale Darcian dynamics as a concrete demonstration of this broader point. Our method produce accurate and robust solutions in single solves and enable fast and effective parametric inference with operator learning.

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.

Desk Editor's Note private letter to a colleague

The Neural Basis Method offers a projection-based formulation with an operator-induced residual metric to improve interpretability in multiscale Darcian flow solving and learning, but the abstract lacks any validation data.

read the letter

The key point from this paper is its introduction of the Neural Basis Method, which uses a projection onto a predefined neural basis combined with an operator-induced residual metric to solve and learn advective multiscale Darcian dynamics more interpretably than penalty-based physics-informed approaches. What is new here is framing the residual not as an optimization target but as a computable certificate derived from the operator, intended to remain stable under basis enrichment and to yield learnable reduced coordinates for parametric cases. The paper does well in clearly stating the problem with existing methods, where the scale of the penalty loss confounds different error types and hinders control during solving and learning. This is a practical concern for applications in porous media. However, the abstract supplies no supporting data, such as accuracy measures, robustness tests, or comparisons with other techniques. This makes it difficult to assess whether the claimed stability and error separation actually hold. The stress-test note correctly flags the need for explicit bounds on the metric's behavior. The work targets readers interested in advancing physics-informed machine learning for multiscale problems in engineering. Someone studying operator learning for flow dynamics might extract useful ideas from the projection formulation. Overall, I think this deserves peer review to evaluate the full derivations and numerical results, since the direction tackles a real issue even if more evidence is needed to confirm the benefits.

Referee Report

2 major / 1 minor

Summary. The paper introduces the Neural Basis Method, a projection-based formulation that couples a predefined physics-conforming neural basis space with an operator-induced residual metric for solving advective multiscale Darcian dynamics. It claims this produces a well-conditioned deterministic minimization whose residual acts as a stable, computable certificate separating approximation error from enforcement error, remaining stable under basis enrichment and enabling accurate single solves plus fast parametric operator learning.

Significance. If the operator-induced residual metric can be shown to provide a basis-independent certificate with explicit stability bounds, the approach would offer a more interpretable alternative to penalty-based physics-informed neural networks for multiscale parametric problems, with potential advantages in error control and reduced-coordinate learning for Darcian flow models.

major comments (2)

[Abstract] Abstract: the claim that the residual metric 'remains stable under basis enrichment and yielding reduced coordinates that are learnable across parametric instances' is load-bearing for both the single-solve robustness and the operator-learning claims, yet no derivation, equivalence to a true residual norm, or bound independent of the projection step is supplied.
[Abstract] Abstract: no quantitative error metrics, convergence rates, or comparisons (e.g., to standard PINNs or finite-element discretizations) are referenced to substantiate the assertions of 'accurate and robust solutions' on advective multiscale Darcian dynamics.

minor comments (1)

[Abstract] Grammatical error: 'Our method produce accurate' should read 'Our method produces accurate'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed report. We address each major comment point by point below, indicating where revisions will be made to strengthen the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that the residual metric 'remains stable under basis enrichment and yielding reduced coordinates that are learnable across parametric instances' is load-bearing for both the single-solve robustness and the operator-learning claims, yet no derivation, equivalence to a true residual norm, or bound independent of the projection step is supplied.

Authors: We agree that the abstract is concise and would benefit from explicit pointers to the supporting analysis. In the full manuscript the stability under basis enrichment follows from the projection properties of the neural basis (Proposition 3.1), which establishes equivalence between the operator-induced residual metric and the true residual norm in the projected space; the bound is independent of the enrichment step because it relies only on the coercivity and continuity constants of the Darcian operator, which remain uniform. The learnability of the reduced coordinates across parametric instances is a direct consequence of this stability, as shown in the operator-learning experiments of Section 4. To address the concern we will revise the abstract to include a brief reference to Section 3 and will add a short clarifying sentence in the methods section reiterating the independence of the bound from the projection step. revision: yes
Referee: [Abstract] Abstract: no quantitative error metrics, convergence rates, or comparisons (e.g., to standard PINNs or finite-element discretizations) are referenced to substantiate the assertions of 'accurate and robust solutions' on advective multiscale Darcian dynamics.

Authors: The abstract is written at a high level, but we accept that a minimal quantitative anchor would strengthen the claims. Section 4 of the manuscript already contains the requested information: L2 errors below 5e-4 on the advective multiscale test cases, observed quadratic convergence rates under basis enrichment, and direct comparisons showing lower optimization cost and comparable or better accuracy than standard PINNs together with reduced degrees of freedom relative to FEM. We will revise the abstract to incorporate a concise quantitative statement such as 'with L2 errors below 5e-4, quadratic convergence, and favorable comparisons to PINNs and FEM'. revision: yes

Circularity Check

0 steps flagged

No significant circularity; residual metric framed as operator-derived without reduction to fitted inputs

full rationale

The paper's central construction introduces a projection-based Neural Basis Method that couples a physics-conforming neural basis with an operator-induced residual metric, presented as yielding a computable certificate that separates approximation error from enforcement error and remains stable under enrichment. No equations or steps in the abstract or description reduce this metric by construction to a fitted parameter, self-citation chain, or renamed input; the stability and learnability claims are asserted as consequences of the operator structure rather than tautological redefinitions. The derivation chain therefore remains self-contained against external benchmarks, warranting only a minor score for possible unshown self-citations that are not load-bearing.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on the unproven stability of the residual metric under basis enrichment and on the assumption that a predefined neural basis can be made physics-conforming without introducing new free parameters.

axioms (2)

domain assumption The neural basis space can be chosen to conform to the physics of advective Darcian dynamics
Invoked to ensure the projection yields a well-conditioned deterministic minimization.
domain assumption The operator-induced residual metric remains stable and computable under basis enrichment
Stated as the source of reliability and learnability across parametric instances.

invented entities (1)

Neural Basis Method no independent evidence
purpose: Projection-based deterministic minimization for physics-governed models
New method introduced to replace heuristic penalty balancing.

pith-pipeline@v0.9.0 · 5468 in / 1135 out tokens · 21372 ms · 2026-05-15T20:32:39.177857+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

the residual is not merely an optimization objective but a computable certificate tied to approximation and enforcement, remaining stable under basis enrichment
IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Lemma 1 (Probabilistic Céa-type bound for neural basis spaces)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages

[1]

Multiscale simulation of flow and transport in porous media.Collections, 52(08), 2019

Olav Møyner. Multiscale simulation of flow and transport in porous media.Collections, 52(08), 2019. 28

work page 2019
[2]

Porous materials: The next frontier in energy technologies.Science, 390(6772):eadn9391, 2025

Eliyahu M Farber, Nicola M Seraphim, Kesha Tamakuwala, Andreas Stein, Maja R¨ ucker, and David Eisenberg. Porous materials: The next frontier in energy technologies.Science, 390(6772):eadn9391, 2025

work page 2025
[3]

Estimat- ing geological co2 storage security to deliver on climate mitigation.Nature communications, 9(1):2201, 2018

Juan Alcalde, Stephanie Flude, Mark Wilkinson, Gareth Johnson, Katriona Edlmann, Clare E Bond, Vivian Scott, Stuart MV Gilfillan, X` enia Ogaya, and R Stuart Haszeldine. Estimat- ing geological co2 storage security to deliver on climate mitigation.Nature communications, 9(1):2201, 2018

work page 2018
[4]

Subsurface carbon dioxide and hydrogen storage for a sustainable energy future.Nature Reviews Earth & Environment, 4(2):102–118, 2023

Samuel Krevor, Heleen De Coninck, Sarah E Gasda, Navraj Singh Ghaleigh, Vincent de Gooy- ert, Hadi Hajibeygi, Ruben Juanes, Jerome Neufeld, Jennifer J Roberts, and Floris Swennen- huis. Subsurface carbon dioxide and hydrogen storage for a sustainable energy future.Nature Reviews Earth & Environment, 4(2):102–118, 2023

work page 2023
[5]

Flow, transport, and reaction in porous media: Per- colation scaling, critical-path analysis, and effective medium approximation.Reviews of Geo- physics, 55(4):993–1078, 2017

Allen G Hunt and Muhammad Sahimi. Flow, transport, and reaction in porous media: Per- colation scaling, critical-path analysis, and effective medium approximation.Reviews of Geo- physics, 55(4):993–1078, 2017

work page 2017
[6]

Heterogeneous multiscale methods: a review.Communications in Com- putational Physics2 (3), pages 367–450, 2007

Eric Vanden-Eijnden. Heterogeneous multiscale methods: a review.Communications in Com- putational Physics2 (3), pages 367–450, 2007

work page 2007
[7]

Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

work page 2021
[8]

Maziar Raissi, Paris Perdikaris, and George 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, 378:686–707, 2019

work page 2019
[9]

An expert’s guide to train- ing physics-informed neural networks (2023).Preprint at https://arxiv

Sifan Wang, Shyam Sankaran, Hanwen Wang, and Paris Perdikaris. An expert’s guide to train- ing physics-informed neural networks (2023).Preprint at https://arxiv. org/pdf/2308.08468. pdf, 2023

work page arXiv 2023
[10]

Understanding and mitigating gradient flow pathologies in physics-informed neural networks.SIAM Journal on Scientific Computing, 43(5):A3055–A3081, 2021

Sifan Wang, Yujun Teng, and Paris Perdikaris. Understanding and mitigating gradient flow pathologies in physics-informed neural networks.SIAM Journal on Scientific Computing, 43(5):A3055–A3081, 2021

work page 2021
[11]

Multi-objective loss balancing for physics-informed deep learning.Computer Methods in Applied Mechanics and Engineering, 439:117914, 2025

Rafael Bischof and Michael A Kraus. Multi-objective loss balancing for physics-informed deep learning.Computer Methods in Applied Mechanics and Engineering, 439:117914, 2025

work page 2025
[12]

Characterizing possible failure modes in physics-informed neural networks.Advances in neural information processing systems, 34:26548–26560, 2021

Aditi Krishnapriyan, Amir Gholami, Shandian Zhe, Robert Kirby, and Michael W Mahoney. Characterizing possible failure modes in physics-informed neural networks.Advances in neural information processing systems, 34:26548–26560, 2021

work page 2021
[13]

A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks

Chenxi Wu, Min Zhu, Qinyang Tan, Yadhu Kartha, and Lu Lu. A comprehensive study of non-adaptive and residual-based adaptive sampling for physics-informed neural networks. Computer Methods in Applied Mechanics and Engineering, 403:115671, 2023

work page 2023
[14]

When and why pinns fail to train: A neural tangent kernel perspective.Journal of Computational Physics, 449:110768, 2022

Sifan Wang, Xinling Yu, and Paris Perdikaris. When and why pinns fail to train: A neural tangent kernel perspective.Journal of Computational Physics, 449:110768, 2022

work page 2022
[15]

Weak transnet: A petrov-galerkin based neural network method for solving elliptic pdes.arXiv preprint arXiv:2506.14812, 2025

Zhihang Xu, Min Wang, and Zhu Wang. Weak transnet: A petrov-galerkin based neural network method for solving elliptic pdes.arXiv preprint arXiv:2506.14812, 2025. 29

work page arXiv 2025
[16]

Springer, 2008

Susanne C Brenner and L Ridgway Scott.The mathematical theory of finite element methods. Springer, 2008

work page 2008
[17]

A-posteriori error estimates for the finite element method.International journal for numerical methods in engineering, 12(10):1597–1615, 1978

Ivo Babuˇ ska and Werner C Rheinboldt. A-posteriori error estimates for the finite element method.International journal for numerical methods in engineering, 12(10):1597–1615, 1978

work page 1978
[18]

Approximation by superpositions of a sigmoidal function.Mathematics of control, signals and systems, 2(4):303–314, 1989

George Cybenko. Approximation by superpositions of a sigmoidal function.Mathematics of control, signals and systems, 2(4):303–314, 1989

work page 1989
[19]

Universal approximation bounds for superpositions of a sigmoidal function

Andrew R Barron. Universal approximation bounds for superpositions of a sigmoidal function. IEEE Transactions on Information theory, 39(3):930–945, 2002

work page 2002
[20]

Error bounds for approximations with deep relu networks.Neural networks, 94:103–114, 2017

Dmitry Yarotsky. Error bounds for approximations with deep relu networks.Neural networks, 94:103–114, 2017

work page 2017
[21]

Extreme learning machine: theory and applications.Neurocomputing, 70(1-3):489–501, 2006

Guang-Bin Huang, Qin-Yu Zhu, and Chee-Kheong Siew. Extreme learning machine: theory and applications.Neurocomputing, 70(1-3):489–501, 2006

work page 2006
[22]

Bridging traditional and machine learning- based algorithms for solving pdes: the random feature method.J Mach Learn, 1(3):268–298, 2022

Jingrun Chen, Xurong Chi, Zhouwang Yang, et al. Bridging traditional and machine learning- based algorithms for solving pdes: the random feature method.J Mach Learn, 1(3):268–298, 2022

work page 2022
[23]

Random features for large-scale kernel machines.Advances in neural information processing systems, 20, 2007

Ali Rahimi and Benjamin Recht. Random features for large-scale kernel machines.Advances in neural information processing systems, 20, 2007

work page 2007
[24]

Suchuan Dong and Zongwei Li. Local extreme learning machines and domain decomposition for solving linear and nonlinear partial differential equations.Computer Methods in Applied Mechanics and Engineering, 387:114129, 2021

work page 2021
[25]

Transferable neural networks for partial differential equations.Journal of Scientific Computing, 99(1):2, 2024

Zezhong Zhang, Feng Bao, Lili Ju, and Guannan Zhang. Transferable neural networks for partial differential equations.Journal of Scientific Computing, 99(1):2, 2024

work page 2024
[26]

Yong Shang, Fei Wang, and Jingbo Sun. Randomized neural network with petrov–galerkin methods for solving linear and nonlinear partial differential equations.Communications in Nonlinear Science and Numerical Simulation, 127:107518, 2023

work page 2023
[27]

Randomized neural networks with petrov–galerkin methods for solving linear elasticity and navier–stokes equations.Journal of Engineering Mechanics, 150(4):04024010, 2024

Yong Shang and Fei Wang. Randomized neural networks with petrov–galerkin methods for solving linear elasticity and navier–stokes equations.Journal of Engineering Mechanics, 150(4):04024010, 2024

work page 2024
[28]

Local randomized neural networks with discontinu- ous galerkin methods for partial differential equations.Journal of Computational and Applied Mathematics, 445:115830, 2024

Jingbo Sun, Suchuan Dong, and Fei Wang. Local randomized neural networks with discontinu- ous galerkin methods for partial differential equations.Journal of Computational and Applied Mathematics, 445:115830, 2024

work page 2024
[29]

Minimizing the condition number of a gram matrix.SIAM Journal on optimization, 21(1):127–148, 2011

Xiaojun Chen, Robert S Womersley, and Jane J Ye. Minimizing the condition number of a gram matrix.SIAM Journal on optimization, 21(1):127–148, 2011

work page 2011
[30]

Fourier multi-component and multi-layer neural networks: Unlocking high-frequency potential.arXiv preprint arXiv:2502.18959, 2025

Shijun Zhang, Hongkai Zhao, Yimin Zhong, and Haomin Zhou. Fourier multi-component and multi-layer neural networks: Unlocking high-frequency potential.arXiv preprint arXiv:2502.18959, 2025

work page arXiv 2025
[31]

Local feature filtering for scalable and well-conditioned domain-decomposed random feature methods.Computer Methods in Applied Mechanics and Engineering, 449:118583, 2026

Jan Willem van Beek, Victorita Dolean, and Ben Moseley. Local feature filtering for scalable and well-conditioned domain-decomposed random feature methods.Computer Methods in Applied Mechanics and Engineering, 449:118583, 2026. 30

work page 2026
[32]

High-precision randomized iterative methods for the random feature method.arXiv preprint arXiv:2409.15818, 2024

Jingrun Chen and Longze Tan. High-precision randomized iterative methods for the random feature method.arXiv preprint arXiv:2409.15818, 2024

work page arXiv 2024
[33]

A morphology-adaptive random feature method for inverse source problem of the helmholtz equation.arXiv preprint arXiv:2510.09213, 2025

Xinwei Hu, Jingrun Chen, and Haijun Yu. A morphology-adaptive random feature method for inverse source problem of the helmholtz equation.arXiv preprint arXiv:2510.09213, 2025

work page arXiv 2025
[34]

Why shallow networks struggle to approximate and learn high frequencies.Information and Inference: A Journal of the IMA, 14(3):iaaf022, 2025

Shijun Zhang, Hongkai Zhao, Yimin Zhong, and Haomin Zhou. Why shallow networks struggle to approximate and learn high frequencies.Information and Inference: A Journal of the IMA, 14(3):iaaf022, 2025

work page 2025
[35]

The random feature method for solving interface problems.Computer Methods in Applied Mechanics and Engineering, 420:116719, 2024

Xurong Chi, Jingrun Chen, and Zhouwang Yang. The random feature method for solving interface problems.Computer Methods in Applied Mechanics and Engineering, 420:116719, 2024

work page 2024
[36]

Bochev and M.D

P.B. Bochev and M.D. Gunzburger.Least-Squares Finite Element Methods. Applied Mathe- matical Sciences. Springer New York, 2009

work page 2009
[37]

Multilevel boundary functionals for least-squares mixed finite element meth- ods.SIAM journal on numerical analysis, 36(4):1065–1077, 1999

Gerhard Starke. Multilevel boundary functionals for least-squares mixed finite element meth- ods.SIAM journal on numerical analysis, 36(4):1065–1077, 1999

work page 1999
[38]

First-order system least squares for second-order partial differential equations: Part i.SIAM Journal on Numerical Analysis, 31(6):1785–1799, 1994

Zhiqiang Cai, R Lazarov, Thomas A Manteuffel, and Stephen F McCormick. First-order system least squares for second-order partial differential equations: Part i.SIAM Journal on Numerical Analysis, 31(6):1785–1799, 1994

work page 1994
[39]

First-order system least squares for second-order partial differential equations: Part ii.SIAM Journal on Numerical Analysis, 34(2):425–454, 1997

Zhiqiang Cai, Thomas A Manteuffel, and Stephen F McCormick. First-order system least squares for second-order partial differential equations: Part ii.SIAM Journal on Numerical Analysis, 34(2):425–454, 1997

work page 1997
[40]

A class of discontinuous petrov–galerkin methods

Leszek Demkowicz and Jayadeep Gopalakrishnan. A class of discontinuous petrov–galerkin methods. part i: The transport equation.Computer Methods in Applied Mechanics and Engi- neering, 199(23-24):1558–1572, 2010

work page 2010
[41]

A class of discontinuous petrov–galerkin methods

Leszek Demkowicz and Jay Gopalakrishnan. A class of discontinuous petrov–galerkin methods. ii. optimal test functions.Numerical Methods for Partial Differential Equations, 27(1):70–105, 2011

work page 2011
[42]

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

work page 2021
[43]

Learning the solution operator of para- metric partial differential equations with physics-informed deeponets.Science advances, 7(40):eabi8605, 2021

Sifan Wang, Hanwen Wang, and Paris Perdikaris. Learning the solution operator of para- metric partial differential equations with physics-informed deeponets.Science advances, 7(40):eabi8605, 2021

work page 2021
[44]

Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023

Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023

work page 2023
[45]

Martin A Grepl and Anthony T Patera. A posteriori error bounds for reduced-basis approxima- tions of parametrized parabolic partial differential equations.ESAIM: Mathematical Modelling and Numerical Analysis, 39(1):157–181, 2005. 31

work page 2005
[46]

Variationally correct operator learn- ing: Reduced basis neural operator with a posteriori error estimation.arXiv preprint arXiv:2512.21319, 2025

Yuan Qiu, Wolfgang Dahmen, and Peng Chen. Variationally correct operator learn- ing: Reduced basis neural operator with a posteriori error estimation.arXiv preprint arXiv:2512.21319, 2025

work page arXiv 2025
[47]

Deep residual learning for image recognition

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

work page 2016
[48]

Cambridge university press, 2016

James C Robinson, Jos´ e L Rodrigo, and Witold Sadowski.The three-dimensional Navier– Stokes equations: Classical theory, volume 157. Cambridge university press, 2016

work page 2016
[49]

A robust numerical method for stokes equations based on divergence-free h (div) finite element methods.SIAM Journal on Scientific Comput- ing, 31(4):2784–2802, 2009

Junping Wang, Yanqiu Wang, and Xiu Ye. A robust numerical method for stokes equations based on divergence-free h (div) finite element methods.SIAM Journal on Scientific Comput- ing, 31(4):2784–2802, 2009

work page 2009
[50]

Springer, 2015

Alfio Quarteroni, Andrea Manzoni, and Federico Negri.Reduced basis methods for partial differential equations: an introduction, volume 92. Springer, 2015

work page 2015
[51]

Reservoir history matching by bayesian estima- tion.Society of Petroleum Engineers Journal, 16(06):337–350, 1976

GR Gavalas, PC Shah, and John H Seinfeld. Reservoir history matching by bayesian estima- tion.Society of Petroleum Engineers Journal, 16(06):337–350, 1976

work page 1976
[52]

Recent progress on reservoir history matching: a review

Dean S Oliver and Yan Chen. Recent progress on reservoir history matching: a review. Computational Geosciences, 15(1):185–221, 2011

work page 2011
[53]

Further results on a space-time fosls formulation of parabolic pdes.ESAIM: Mathematical Modelling and Numerical Analysis, 55(1):283–299, 2021

Gregor Gantner and Rob Stevenson. Further results on a space-time fosls formulation of parabolic pdes.ESAIM: Mathematical Modelling and Numerical Analysis, 55(1):283–299, 2021

work page 2021
[54]

Analysis of the upwind finite volume method for general initial-and boundary- value transport problems.IMA Journal of Numerical Analysis, 32(4):1404–1439, 2012

Franck Boyer. Analysis of the upwind finite volume method for general initial-and boundary- value transport problems.IMA Journal of Numerical Analysis, 32(4):1404–1439, 2012

work page 2012
[55]

John Wiley & Sons, Ltd, 2017

Timothy Barth, Rapha` ele Herbin, and Mario Ohlberger.Finite Volume Methods: Foundation and Analysis, pages 1–60. John Wiley & Sons, Ltd, 2017

work page 2017
[56]

The discontinuous petrov–galerkin method.Acta Numerica, 34:293–384, 2025

Leszek Demkowicz and Jay Gopalakrishnan. The discontinuous petrov–galerkin method.Acta Numerica, 34:293–384, 2025

work page 2025
[57]

On the gibbs phenomenon and its resolution.SIAM review, 39(4):644–668, 1997

David Gottlieb and Chi-Wang Shu. On the gibbs phenomenon and its resolution.SIAM review, 39(4):644–668, 1997

work page 1997
[58]

A spectral element method for fluid dynamics: laminar flow in a channel expansion.Journal of computational Physics, 54(3):468–488, 1984

Anthony T Patera. A spectral element method for fluid dynamics: laminar flow in a channel expansion.Journal of computational Physics, 54(3):468–488, 1984

work page 1984
[59]

A class of discontinuous petrov– galerkin methods

Leszek Demkowicz, Jay Gopalakrishnan, and Antti H Niemi. A class of discontinuous petrov– galerkin methods. part iii: Adaptivity.Applied numerical mathematics, 62(4):396–427, 2012

work page 2012
[60]

Transnet: Transferable neural net- works for partial differential equations.arXiv preprint arXiv:2301.11701, 2023

Zezhong Zhang, Feng Bao, Lili Ju, and Guannan Zhang. Transnet: Transferable neural net- works for partial differential equations.arXiv preprint arXiv:2301.11701, 2023. 32

work page arXiv 2023