Pith. sign in

REVIEW 4 major objections 3 minor 2 cited by

Structure-Preserving Digital Twins via Conditional Neural Whitney Forms

T0 review · 4 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read This paper argues that a learned reduced model can preserve conservation laws exactly by embedding conditional attention inside finite element exterior calculus.

desk verdict A genuinely interesting architectural claim I can't check from this copy: conditional attention learns a reduced FEEC basis and flux, so conservation is meant to be hard-wired; the missing sections are exactly where the proof has to live. read the letter →

arxiv 2508.06981 v1 pith:N7J7LW5R submitted 2025-08-09 cs.LG cs.NAmath.NAphysics.comp-ph

classification cs.LGcs.NAmath.NAphysics.comp-ph MSC 65N3068T0758A10
keywords digitaltwinfiniteelementexteriorcalculusWhitneyformsstructure-preservingneuralnetworksconditionalattentionoperatorlearningreducedordermodelsconservationlaws
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 seeks to make neural surrogates trustworthy enough for closed-loop digital twins by turning conservation and well-posedness from training penalties into architectural guarantees. The framework learns a reduced finite element basis and a nonlinear flux with conditional attention mechanisms, but keeps both inside finite element exterior calculus (FEEC), so the discrete de Rham complex and its commuting-diagram properties are preserved. The paper argues that exact preservation of conserved quantities therefore holds regardless of data sparsity or optimization error, and that the latent conditioning field Z permits real-time calibration to sensor data. Benchmarks cover advection-diffusion, shock hydrodynamics, electrostatics, and battery thermal runaway, with accurate predictions on complex geometries from 25 LES simulations and inference near 0.1 s.

What carries the argument

The central object is the conditional neural Whitney form: a learned, conditionally parameterized reduced basis of Whitney forms together with a learned nonlinear flux, organized so that the reduced space remains a subcomplex of the FEEC de Rham complex on the discretized geometry. Whitney forms are the lowest-order finite elements that represent differential forms and respect the discrete complex relations (for example, the discrete curl of a discrete gradient is zero). The conditional attention mechanism modulates both the basis and the flux as functions of the latent conditioning field Z, which is what makes real-time calibration to sensor data possible without leaving the structure-prese

What would settle it

Run the trained reduced model on a test geometry or parameter regime not seen in the 25 LES training runs, measure a conserved quantity over a long rollout, and check the discrete $\operatorname{div}$ (or exterior derivative) of the predicted solution. Any drift or nonzero residual above machine precision would contradict the claim that conservation and well-posedness hold exactly regardless of data sparsity or optimization error.

Watch

Extended reading notes

Core claim

The central claim is that a reduced-order model can be learned end-to-end from data without giving up the structure that makes finite element discretizations trustworthy. The paper builds conditional neural Whitney forms: a conditional attention mechanism that learns both a reduced finite element basis and a nonlinear conservation law, with both kept inside finite element exterior calculus (FEEC). Because the learned objects live in the same discrete de Rham complex as the underlying finite element space, the authors argue that numerical well-posedness and exact preservation of conserved quantities are properties of the architecture, not penalties in a loss function — they hold regardless of

Load-bearing premise

The learned, data-compressed basis and flux must remain inside the same discrete differential-geometric structure as the original finite element space; if the compression breaks that structure, the exact-conservation and stability guarantees no longer hold, even if predictions look accurate.

Editorial extensions

If this is right

  • Mass, energy, or charge are conserved by construction in the learned reduced model, so long time-horizon rollouts do not drift in these invariants, independent of how the network is trained.
  • The latent conditioning Z lets the same trained model be recalibrated to live sensor data in real time, which is precisely the closed-loop capability the paper identifies as central to a digital twin.
  • Because the learned components live inside FEEC, the framework composes with existing finite element codes and meshes, so it can be applied to complex geometries without replacing the simulation stack.
  • On the battery thermal runaway benchmark, 25 LES simulations suffice for accurate prediction including transition to turbulence, with inference near 0.1 s — a reported speedup of $3.1 \times 10^8$ over LES.

Reading between the lines

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

  • Editorial inference: if the conservation guarantee is truly independent of optimization error, the architecture is a candidate for safety-critical or regulator-facing digital twins, where a penalty-based conservation loss would not be accepted — a use case the paper does not develop.
  • Editorial inference: the learned nonlinear flux in the thermal-runaway benchmark can be read as a data-driven subgrid closure; a natural extension would test the same framework on other turbulent flows where LES closures are the bottleneck.
  • Editorial inference: a direct way to stress-test the structural claim is to probe out-of-distribution geometries and check discrete divergence/curl-free conditions; the paper's claim predicts exact satisfaction, whereas penalty-based physics-informed models would show residual violation.
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

4 major / 3 minor

Summary. The manuscript proposes a framework for real-time digital twins that combines conditional attention mechanisms, reduced finite element bases, and nonlinear conservation laws within finite element exterior calculus (FEEC). The abstract claims that the construction guarantees numerical well-posedness and exact conservation of quantities regardless of data sparsity or optimization error, and reports benchmarks including advection-diffusion, shock hydrodynamics, electrostatics, and battery thermal runaway, with inference around 0.1 s and a 3.1e8 speedup over LES using only 25 LES training runs. The received text, however, contains only the abstract, Section 1 (Overview and literature review), acknowledgements, and references; all technical sections describing the architecture, the FEEC-compatible reduced basis, the conservation guarantees, and the experimental benchmarks are absent. Consequently, the central claims are stated but not demonstrated in the manuscript.

Significance. If the claimed structural guarantees and benchmark results were substantiated, the work would be significant: exact conservation and well-posedness enforced by architecture rather than loss penalties would address a known weakness of neural operator approaches, and the reported speedups on complex geometries would be practically important. The framework also appears to build on a credible line of prior work (data-driven Whitney forms, structure-preserving domain decomposition, metriplectic bracket networks), which further raises the potential value. However, because the technical content is missing, the significance is entirely prospective; the manuscript as received provides no derivations, no experiments, and no code-level verification to support the advertised contributions.

major comments (4)
  1. [Abstract and Sections 2–4 (missing)] The manuscript is incomplete: after Section 1.3 the text jumps directly to Acknowledgements and References. There are no sections describing the conditional attention architecture, the construction of the reduced FEEC basis, the training objective, the conservation and well-posedness theorems, or the benchmarks. All load-bearing claims—exact conservation 'regardless of data sparsity or optimization error', the 25-LES data efficiency, the 0.1 s inference time, and the 3.1e8 speedup—are therefore assertions without supporting evidence. This is not a presentation issue; the central claims cannot be checked.
  2. [Abstract, lines on guarantees] The statement that the method 'guarantees numerical well-posedness and exact preservation of conserved quantities, regardless of data sparsity or optimization error' is a structural claim. For exact preservation to hold, the learned reduced spaces must form a conforming subcomplex of the discrete de Rham complex for every conditioning state Z, with the relevant commuting-diagram and discrete-Hodge properties intact. No derivation establishing this property is present in Section 1 or elsewhere. The phrase 'learns ... within the framework of FEEC' is weaker than 'enforces by construction'; FEEC-inspired loss terms do not by themselves guarantee exactness, particularly under optimization error. This premise must be presented as a theorem or as a hard algebraic constraint in the missing methods section.
  3. [Abstract, 'nonlinear conservation law'] Even if the reduced basis is FEEC-compatible, the manuscript does not explain how a *nonlinear learned flux* preserves exact conservation. In a mixed FEEC discretization, conservation typically requires that the divergence of the flux be representable in the dual space; the divergence of a nonlinear learned flux will generally live outside the reduced space, so the discrete conservation identity would hold only up to projection error. The authors need to specify the discrete flux reconstruction and the projection steps, and prove that exact conservation survives the nonlinearity and the data-driven reduction. This is a load-bearing technical point that is entirely absent.
  4. [Abstract, benchmark claims (0.1 s, 3.1e8 speedup, 25 LES runs)] The reported numerical results—capturing transition to turbulence from only 25 LES simulations, achieving ~0.1 s inference, and a 3.1e8 speedup—are stated in the abstract but no experimental section, dataset description, geometry, LES solver, hyperparameters, baselines, error metrics, or hardware details are provided. These claims cannot be reproduced or assessed. A complete experimental section is required, including how the 'transition to turbulence' is measured, what the speedup is measured relative to (wall-clock vs. CPU time, including data generation and training), and what the uncertainty in the accuracy metrics is.
minor comments (3)
  1. [Section 1.2] The literature review is broad but some references appear tangential (e.g., [63] on Liberty Bell). Consider focusing the review on works directly used by the proposed method, and cite specific equations or constructions from the prior data-driven Whitney forms work [1] and domain decomposition work [47] that the framework builds on.
  2. [Acknowledgements] The text after the reference list contains a garbled, non-grammatical passage. This appears to be a formatting artifact, but it should be cleaned before resubmission, as it currently obscures part of the reference section.
  3. [General] The paper promises an open-source implementation on GitHub, but no repository identifier is provided. If the code is intended to support the reproducibility of the benchmarks, the URL and commit hash should be cited.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity demonstrated; central guarantee is unverified due to missing method sections, not equivalent to its inputs by construction.

full rationale

The received text contains the abstract, Section 1, references, and acknowledgements; Sections 2-3 (the method and derivation) are absent. The abstract's central guarantee ('This guarantees numerical well-posedness and exact preservation of conserved quantities, regardless of data sparsity or optimization error') is asserted without a derivation showing that the learned reduced basis and learned nonlinear flux form a d-closed FEEC subcomplex for each conditioning state. That is an omitted proof, which is a completeness/correctness risk, but it is not circularity: no equation in the visible text defines the predicted quantity in terms of the fitted input, and no fitted parameter is renamed as a prediction. The self-citations (refs [1], [47], [36], [37]) are listed but not quoted as load-bearing in the visible derivation; even if the missing method sections build on the authors' prior Whitney-form and domain-decomposition work, those are published, externally checkable results and would not by themselves make the present claim circular. The benchmarks are external LES/reference data, not outputs of the model. A garbled non-Latin appended passage after the references cannot be parsed reliably in the received encoding; it supplies no quotable reduction and thus cannot ground a circularity finding. Under the hard rule that circularity requires an exhibited reduction (Eq. X = Eq. Y by construction), no such step can be located in the received text. Score 0.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The ledger is provisional because the received text is truncated: visible content is the abstract, Section 1 (overview and literature review), acknowledgements, and references [1]-[61]. The central method sections, benchmark tables, and any appended limitations passages are absent, as is any readable text near page 31. The free parameters listed are the data-fitted quantities implied by the abstract; their values and the training procedure were not available. The axioms are the mathematical and data premises the abstract's guarantees depend on, with the conforming-subcomplex premise being the load-bearing one.

free parameters (3)
  • Learned reduced FE basis coefficients = unknown (not in received text)
    The reduced basis is fit to the 25 LES simulations via the conditional attention network; the received text presents no closed forms or values.
  • Learned nonlinear discrete flux network parameters = unknown (not in received text)
    The nonlinear flux in the conservation law is a trained neural network; its architecture size, capacity, and fitted values are absent from the received text.
  • Latent conditioning dimension and encoder parameters = unknown (not in received text)
    The latent variable Z and its encoder are central to the calibration claim, but their dimensionality and training details are not visible in the received text.
assumptions (3)
  • standard math FEEC framework: de Rham complex, discrete Hodge decomposition, and commuting diagrams (Arnold et al., refs [6,8])
    Invoked in Section 1.3 to justify structure preservation; established published theory, not the authors' contribution.
  • ad hoc to paper The learned/compressed reduced basis preserves a conforming subcomplex of the discrete de Rham complex
    The abstract's guarantee of exact conservation and well-posedness 'regardless of data sparsity or optimization error' requires this property to survive data-driven projection; the derivation is absent from the received text.
  • domain assumption The 25 LES simulations are faithful ground truth and adequately cover the conditioning distribution
    The benchmark claims (turbulence capture, calibration to sensor states) presuppose that the LES training data are accurate and representative; data provenance is not in the received text.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Structure-Preserving Digital Twins via Conditional Neural Whitney Forms." pith.science (2026). https://pith.science/paper/N7J7LW5R

@misc{pith2026250806981,
  author       = {Pith},
  title        = {Pith review of: Structure-Preserving Digital Twins via Conditional Neural Whitney Forms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N7J7LW5R}},
  note         = {Machine review of arXiv:2508.06981}
}
read the original abstract

We present a framework for constructing real-time digital twins based on structure-preserving reduced finite element models conditioned on a latent variable Z. The approach uses conditional attention mechanisms to learn both a reduced finite element basis and a nonlinear conservation law within the framework of finite element exterior calculus (FEEC). This guarantees numerical well-posedness and exact preservation of conserved quantities, regardless of data sparsity or optimization error. The conditioning mechanism supports real-time calibration to parametric variables, allowing the construction of digital twins which support closed loop inference and calibration to sensor data. The framework interfaces with conventional finite element machinery in a non-invasive manner, allowing treatment of complex geometries and integration of learned models with conventional finite element techniques. Benchmarks include advection diffusion, shock hydrodynamics, electrostatics, and a complex battery thermal runaway problem. The method achieves accurate predictions on complex geometries with sparse data (25 LES simulations), including capturing the transition to turbulence and achieving real-time inference ~0.1s with a speedup of 3.1x10^8 relative to LES. An open-source implementation is available on GitHub.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Structure-Preserving Learning Improves Geometry Generalization in Neural PDEs

    cs.LG 2026-02 conditional novelty 6.0 of 10

    A geometry-conditioned Whitney-form neural network that solves a learned discrete conservation law improves out-of-distribution geometry generalization for steady-state PDEs compared with regression-based neural operators.

  2. Physics-informed sensor coverage through structure preserving machine learning

    cs.LG 2025-09 conditional novelty 6.0 of 10

    A structure-preserving neural PDE surrogate, conditioned on sparse sensor readings, locates sources in advection-diffusion fields and guides adaptive sensor placement through a geodesic Lloyd algorithm.

Reference graph

Works this paper leans on

61 extracted references · 53 canonical work pages · cited by 2 Pith papers

  1. [1]

    Data- driven whitney forms for structure-preserving control volume analysis

    Jonas A Actor, Xiaozhe Hu, Andy Huang, Scott A Roberts, and Nathaniel Trask. Data- driven whitney forms for structure-preserving control volume analysis. Journal of Computational Physics, 496:112520, 2024

  2. [2]

    Zico Kolter

    Brandon Amos, Lei Xu, and J. Zico Kolter. Input convex neural networks. In Doina Precup and Yee Whye Teh, editors, Proceedings of the 34th International Conference on Machine Learning , volume 70 of Proceedings of Machine Learning Research , pages 146–155. PMLR, June–August 2017

  3. [3]

    Mfem: A modular finite element methods library

    Robert Anderson, Julian Andrej, Andrew Barker, Jamie Bramwell, Jean-Sylvain Camier, Jakub Cerveny, Veselin Dobrev, Yohann Dudouit, Aaron Fisher, Tzanio Kolev, et al. Mfem: A modular finite element methods library. Computers & Mathematics with Applications , 81:42–74, 2021

  4. [4]

    A multiscale mortar mixed finite element method

    Todd Arbogast, Gergina Pencheva, Mary F Wheeler, and Ivan Yotov. A multiscale mortar mixed finite element method. Multiscale Modeling & Simulation , 6(1):319–346, 2007. 21

  5. [5]

    Minimization of functions having lipschitz continuous first partial derivatives

    Larry Armijo. Minimization of functions having lipschitz continuous first partial derivatives. Pacific Journal of mathematics , 16(1):1–3, 1966

  6. [6]

    Finite element exterior calculus

    Douglas N Arnold. Finite element exterior calculus . SIAM, 2018

  7. [7]

    Compatible spatial discretizations, volume 142

    Douglas N Arnold, Pavel B Bochev, Richard B Lehoucq, Roy A Nicolaides, and Mikhail Shashkov. Compatible spatial discretizations, volume 142. Springer Science & Business Media, 2007

  8. [8]

    Finite element exterior calculus, homo- logical techniques, and applications

    Douglas N Arnold, Richard S Falk, and Ragnar Winther. Finite element exterior calculus, homo- logical techniques, and applications. Acta numerica, 15:1–155, 2006

Show all 61 references
  1. [9]

    Yohai Bar-Sinai, Stephan Hoyer, Jason Hickey, and Michael P. Brenner. Learning data-driven discretizations for partial differential equations. Proceedings of the National Academy of Sciences, 116(31), 2019

  2. [10]

    E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials

    Simon Batzner, Albert Musaelian, Lixin Sun, Mario Geiger, Jonathan P Mailoa, Mordechai Korn- bluth, Nicola Molinari, Tess E Smidt, and Boris Kozinsky. E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials. Nature communications, 13(1):2...

  3. [11]

    Roux, Pascal Vincent, Olivier Delalleau, and Patrice Marcotte

    Yoshua Bengio, Nicolas L. Roux, Pascal Vincent, Olivier Delalleau, and Patrice Marcotte. Convex neural networks. In Advances in Neural Information Processing Systems 18 , pages 1233–1240, 2005

  4. [12]

    Parallel lagrange–newton–krylov–schur methods for pde- constrained optimization

    George Biros and Omar Ghattas. Parallel lagrange–newton–krylov–schur methods for pde- constrained optimization. part i: The krylov–schur solver. SIAM Journal on Scientific Computing, 27(2):687–713, 2005

  5. [13]

    Solving pdes with intrepid

    Pavel Bochev, H Carter Edwards, Robert C Kirby, Kara Peterson, and Denis Ridzal. Solving pdes with intrepid. Scientific Programming, 20(2):151–180, 2012

  6. [14]

    Principles of mimetic discretizations of differential opera- tors

    Pavel B Bochev and James M Hyman. Principles of mimetic discretizations of differential opera- tors. In Compatible spatial discretizations, pages 89–119. Springer, 2006

  7. [15]

    Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brand- stetter, Patrick Garvan, Maik Riechert, Jonathan A

    Cristian Bodnar, Wessel P. Bruinsma, Ana Lucic, Megan Stanley, Anna Allen, Johannes Brand- stetter, Patrick Garvan, Maik Riechert, Jonathan A. Weyn, Haiyu Dong, Jayesh K. Gupta, Kit Thambiratnam, Alexander T. Archibald, Chun-Chieh Wu, Elizabeth Heider, Max Welling, Richard E. ...

  8. [16]

    Streamline upwind/petrov-galerkin formulations for convection dominated flows with particular emphasis on the incompressible navier-stokes equa- tions

    Alexander N Brooks and Thomas JR Hughes. Streamline upwind/petrov-galerkin formulations for convection dominated flows with particular emphasis on the incompressible navier-stokes equa- tions. Computer methods in applied mechanics and engineering , 32(1-3):199–259, 1982

  9. [17]

    Equivariant geometric learning for digital rock physics: estimating formation factor and effective permeability tensors from morse graph

    Chen Cai, Nikolaos Vlassis, Lucas Magee, Ran Ma, Zeyu Xiong, and WaiChing Sun. Equivariant geometric learning for digital rock physics: estimating formation factor and effective permeability tensors from morse graph. arXiv preprint arXiv:2104.05608 , 2021

  10. [18]

    Recommendations for future efforts in rans modeling and simu- lation

    CFD Vision 2030 Study Team. Recommendations for future efforts in rans modeling and simu- lation. Technical report, NASA Technical Memorandum 2020-XXXX, 2020. States that RANS models, while robust, have not significantly improved predictive accuracy over the past 20 years and ...

  11. [19]

    Lowe, and Ruoying He

    Ashesh Chattopadhyay, Michael Gray, Tianning Wu, Anna B. Lowe, and Ruoying He. Oceannet: a principled neural operator-based digital twin for regional oceans. Scientific Reports, 14:21181, 2024

  12. [20]

    Shreyas Chaudhari, Srinivasa Pranav, and Jos´ e M. F. Moura. Learning gradients of convex functions with monotone gradient networks. arXiv, abs/2301.10862, 2023

  13. [21]

    Neural ordinary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. Advances in neural information processing systems , 31, 2018

  14. [22]

    Symplectic recurrent neural networks

    Zhengdao Chen, Jianyu Zhang, Martin Arjovsky, and L´ eon Bottou. Symplectic recurrent neural networks. arXiv preprint arXiv:1909.13334 , 2019

  15. [23]

    Projection-based reduced order models for parameterized nonlinear time-dependent problems arising in cardiac mechanics

    Ludovica Cicci, Stefania Fresca, Stefano Pagani, Andrea Manzoni, and Alfio Quarteroni. Projection-based reduced order models for parameterized nonlinear time-dependent problems arising in cardiac mechanics. Mathematics in Engineering , 5(2):1–38, 2023

  16. [24]

    Tvb runge-kutta local projection discontinuous galerkin finite element method for conservation laws

    Bernardo Cockburn and Chi-Wang Shu. Tvb runge-kutta local projection discontinuous galerkin finite element method for conservation laws. ii. general framework. Mathematics of Computation , 52(186):411–435, 1989

  17. [25]

    Group equivariant convolutional networks

    Taco Cohen and Max Welling. Group equivariant convolutional networks. In Proceedings of The 33rd International Conference on Machine Learning (ICML) , volume 48, pages 2990–2999, 2016

  18. [26]

    Cranmer, Sam Greydanus, Stephan Hoyer, Peter W

    Miles D. Cranmer, Sam Greydanus, Stephan Hoyer, Peter W. Battaglia, David N. Spergel, and Shirley Ho. Lagrangian neural networks. CoRR, abs/2003.04630, 2020

  19. [27]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langu...

  20. [28]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at...

  21. [29]

    Turbulence modeling in the age of data

    Karthik Duraisamy, Gianluca Iaccarino, and Heng Xiao. Turbulence modeling in the age of data. Annual Review of Fluid Mechanics , 51:357–377, 2019. Review of empirical closures and data-driven calibration approaches

  22. [30]

    Statistical theory and modeling for turbulent flows

    Paul A Durbin and BA Pettersson Reif. Statistical theory and modeling for turbulent flows . John Wiley & Sons, 2011

  23. [31]

    Finegan, Eric Darcy, Matthew Keyser, Bernhard Tjaden, Thomas M

    Donal P. Finegan, Eric Darcy, Matthew Keyser, Bernhard Tjaden, Thomas M. M. Heenan, Rhodri Jervis, Josh J. Bailey, Romeo Malik, Nghia T. Vo, Oxana V. Magdysyuk, Robert Atwood, Michael Drakopoulos, Marco DiMichiel, Alexander Rack, Gareth Hinds, Dan J. L. Brett, and Paul R. Shea...

  24. [32]

    Fuhg, Nikolaos Bouklas, and Reese E

    Jan N. Fuhg, Nikolaos Bouklas, and Reese E. Jones. Stress representations for tensor basis neural networks: alternative formulations to finger-rivlin-ericksen. arXiv preprint arXiv:2308.11080 , 2023

  25. [33]

    Deep Learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016

  26. [34]

    Physics-informed deep neural operator networks

    Somdatta Goswami, Aniruddha Bora, Yue Yu, and George Em Karniadakis. Physics-informed deep neural operator networks. Machine Learning in Modeling and Simulation , 2022

  27. [35]

    Hamiltonian neural networks

    Samuel Greydanus, Misko Dzamba, and Jason Yosinski. Hamiltonian neural networks. In Ad- vances in Neural Information Processing Systems 32 (NeurIPS 2019) , pages 15353–15363, 2019

  28. [36]

    Efficiently parameterized neural metriplectic systems

    Anthony Gruber, Kookjin Lee, Haksoo Lim, Noseong Park, and Nathaniel Trask. Efficiently parameterized neural metriplectic systems. arXiv preprint arXiv:2405.16305 , 2024

  29. [37]

    Reversible and irreversible bracket-based dynamics for deep graph neural networks

    Anthony Gruber, Kookjin Lee, and Nathaniel Trask. Reversible and irreversible bracket-based dynamics for deep graph neural networks. Advances in Neural Information Processing Systems , 36:38454–38484, 2023

  30. [38]

    Shampoo: Preconditioned stochastic tensor optimization

    Vineet Gupta, Tomer Koren, and Yoram Singer. Shampoo: Preconditioned stochastic tensor optimization. In Jennifer Dy and Andreas Krause, editors, Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pages 1...

  31. [39]

    Time dependent problems and difference methods, volume 24

    Bertil Gustafsson, Heinz-Otto Kreiss, and Joseph Oliger. Time dependent problems and difference methods, volume 24. John Wiley & Sons, 1995

  32. [40]

    scikit-fem: A python package for finite element assembly

    Tom Gustafsson and Geordie Drummond Mcbain. scikit-fem: A python package for finite element assembly. Journal of Open Source Software , 5(52):2369, 2020

  33. [41]

    Local analysis of discontinuous galerkin methods applied to singularly perturbed problems

    Johnny Guzm´ an. Local analysis of discontinuous galerkin methods applied to singularly perturbed problems. Journal of Numerical Mathematics , 14(1):41–56, 2006

  34. [42]

    Optimization with PDE constraints, volume 23

    Michael Hinze, Ren´ e Pinnau, Michael Ulbrich, and Stefan Ulbrich. Optimization with PDE constraints, volume 23. Springer Science & Business Media, 2008

  35. [43]

    Matrix analysis

    Roger A Horn and Charles R Johnson. Matrix analysis . Cambridge university press, 2012

  36. [44]

    Discontinuous hp-finite element methods for advection–diffusion–reaction problems

    Paul Houston, Christoph Schwab, and Endre S¨ uli. Discontinuous hp-finite element methods for advection–diffusion–reaction problems. SIAM Journal on Numerical Analysis , 39(6):2133–2163, 2002

  37. [45]

    Topology-preserving deep image segmentation

    Xiaoling Hu, Fuxin Li, Dimitris Samaras, and Chao Chen. Topology-preserving deep image segmentation. Advances in neural information processing systems , 32, 2019

  38. [46]

    Incropera, David P

    Frank P. Incropera, David P. DeWitt, Theodore L. Bergman, and Adrienne S. Lavine. Fundamen- tals of Heat and Mass Transfer . Wiley, 6th edition, 2007. Critical Grashof number for transition to turbulence in natural convection along vertical walls, Gr ∼ 109

  39. [47]

    A structure-preserving domain decomposition method for data-driven modeling

    Shuai Jiang, Jonas Actor, Scott Roberts, and Nathaniel Trask. A structure-preserving domain decomposition method for data-driven modeling. arXiv preprint arXiv:2406.05571 , 2024. 24

  40. [48]

    Knezevic, and Karen E

    Michael Kapteyn, David J. Knezevic, and Karen E. Willcox. Toward predictive digital twins via component-based reduced-order models and interpretable machine learning. In AIAA Scitech Forum, 2020

  41. [49]

    Kapteyn, Jacob V

    Michael G. Kapteyn, Jacob V. R. Pretorius, and Karen E. Willcox. A probabilistic graphical model foundation for enabling predictive digital twins at scale. Nature Computational Science , 1(5):337–347, 2021

  42. [50]

    Deep neural operator-driven real-time inference to enable digital twin solutions for nuclear energy systems

    Kazuma Kobayashi and Syed Bahauddin Alam. Deep neural operator-driven real-time inference to enable digital twin solutions for nuclear energy systems. Scientific Reports, 14:2101, 2024

  43. [51]

    Neural operator: learning maps between function spaces

    Zongyi Kovachki, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: learning maps between function spaces. Journal of Machine Learning Research, 24(1524):1–97, 2023

  44. [52]

    Flux-corrected transport: principles, algo- rithms, and applications

    Dmitri Kuzmin, Rainald L¨ ohner, and Stefan Turek. Flux-corrected transport: principles, algo- rithms, and applications . Springer Science & Business Media, 2012

  45. [53]

    Gradient-based learning applied to document recognition

    Yann LeCun, L´ eon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998

  46. [54]

    Fourier neural operator for parametric partial differential equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, An- drew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. In Proceedings of the International Conference on Learning Representations (ICLR) ,

  47. [55]

    Jones, and Jeremy Templeton

    Julia Ling, Reese E. Jones, and Jeremy Templeton. Machine learning strategies for systems with invariance properties. Journal of Computational Physics , 318:22–35, 2016

  48. [56]

    Rajanna, Edward W

    Ning Liu, Xuxiao Li, Manoj R. Rajanna, Edward W. Reutzel, Brady Sawyer, Prahalada Rao, Jim Lua, Nam Phan, and Yue Yu. Deep neural operator enabled digital twin modeling for additive manufacturing. Advances in Computational Science and Engineering , 2024

  49. [57]

    Livescu and J

    D. Livescu and J. R. Ristorcelli. Variable-density mixing in buoyancy-driven turbulence. Journal of Fluid Mechanics , 605:145–180, 2008

  50. [58]

    Learning non- linear operators via deeponet based on the universal approximation theorem of operators

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhiwen Zhang, and George Em Karniadakis. Learning non- linear operators via deeponet based on the universal approximation theorem of operators. Nature Machine Intelligence , 3:218–229, 2021

  51. [59]

    Kurzawski, and John C

    Michael Meehan, Andrew J. Kurzawski, and John C. Hewson. Flow dynamics and heat transfer in simplified battery energy storage systems with heated battery modules. Submitted for publication to Journal of Energy Storage , 2025

  52. [60]

    Gary B. Nash. The Liberty Bell . Yale University Press, New Haven, CT, 2010. A cultural history tracing the bell’s origins, recasting, role in abolitionist and wartime symbolism

  53. [61]

    Francesco Nobile, Raul Tempone, and Clayton G. Webster. A sparse grid stochastic collocation method for partial differential equations with random input data. SIAM Journal on Numerical Analysis, 46(5):2309–2345, 2008. 25 ������ ���� � ������� ���� ����� ����� ��� ������� �� ��...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.