Pith. sign in

REVIEW 3 major objections 6 minor 83 references

I-FENN with DeepONets: accelerating simulations in coupled multiphysics problems

T0 review · 3 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims that replacing the temperature or pressure solve in coupled multiphysics FEM with a GRU-based DeepONet keeps errors below 5% on unseen loads and saves 35–43% of runtime.

desk verdict A credible incremental advance in I-FENN with a real architecture contribution, but the headline speedup claim is undercut by the missing staggered-FEM baseline. read the letter →

arxiv 2509.00604 v1 pith:XHHFP4YB submitted 2025-08-30 cs.CE

Fouad M. Amin , Diab W. Abueidda , Panos Pantidis , Mostafa E. Mobasher This is my paper
classification cs.CE MSC 65M6068T0774F0576S05
keywords I-FENNDeepONetMIONetGRUthermoelasticityporoelasticitymultiphysicssimulationboundaryconditionenforcement
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 tries to establish that a hybrid staggered solver, I-FENN, can replace the temperature or pressure solve in coupled thermoelasticity and poroelasticity problems with a trained neural operator while preserving accuracy and gaining speed. The FEM still handles the mechanical field, while a gated-recurrent DeepONet predicts the coupled field from the previous strain trace and the upcoming loads. The authors demonstrate on three test problems, all with unseen loading conditions, that non-trivial response zones stay within 5% relative error and that runtime savings grow from 35% to 43% as meshes are refined. If this holds, expensive monolithic coupled simulations could be replaced by cheaper staggered runs across many load scenarios, with training cost amortized over repeated simulations.

What carries the argument

The carrying object is a modified deep operator network, derived from MIONet, in which multiple branch networks encode different input functions and their outputs are concatenated rather than multiplied, before an element-wise product with the trunk output. Each branch stacks gated recurrent units (GRUs) to absorb temporal load and strain histories without fixing the number of prior time steps. The pivotal simplification is that the mechanical-to-thermal or mechanical-to-porous coupling enters the governing equations only through the scalar trace of the strain tensor, so the network can predict the next temperature or pressure field from tr(ε) at the previous step plus the upcoming loads. A

What would settle it

Run I-FENN on a load history with abrupt load reversals at a coarse time step and compare the predicted coupled field to a monolithic FEM reference: if relative error in non-trivial zones exceeds 5% or grows with simulation time, the one-step-lag ansatz fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper reports that a modified DeepONet—using a GRU branch for load history, a second GRU branch for strain-trace history, and a fully connected trunk for spatial coordinates—can serve as the coupled-field solver inside I-FENN for thermoelasticity and poroelasticity. The mechanical equilibrium equation is still solved by FEM with the temperature or pressure field supplied by the neural network. Across three geometric setups, the predicted coupled fields retain less than 5% relative error in zones with non-trivial response values for load cases never seen in training, and the hybrid solver runs 35–43% faster than a fully coupled monolithic FEM solver on progressively refi

Load-bearing premise

The temperature or pressure at the next time step can be accurately predicted from the strain trace at the previous time step plus the upcoming loads, and this one-step lag will not let errors grow without bound.

Editorial extensions

If this is right

  • Unseen loading conditions can be simulated with less than 5% relative error in non-trivial response regions across both thermoelasticity and poroelasticity problems.
  • Computational savings grow with mesh refinement, reaching 43% on the finest tested domain, so larger and more complex coupled simulations stand to benefit most.
  • A network trained on a coarse mesh predicts accurately on finer meshes, decoupling training cost from the resolution of the production simulation.
  • Compared with full simulation substitution, the hybrid I-FENN keeps displacement errors about an order of magnitude lower, trains faster, and needs less output data.
  • Mechanical equilibrium enforced by FEM shields the displacement field from errors in the neural-predicted coupled field, making the framework robust to imperfect network predictions.

Reading between the lines

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

  • Because the coupling is mediated solely by the trace of strain, the same architecture should transfer to other isotropic coupled problems whose coupling term is a scalar invariant of the mechanical state, such as chemo-mechanics or reaction-diffusion–mechanics.
  • The explicit one-step lag in feeding strain history to the network means accuracy will have a practical limit under very fast transients; a time-step refinement study would reveal whether the 5% error bound holds as the load rate increases.
  • The boundary-condition formula using ℓ_i functions could be combined with approximate distance functions to handle arbitrary geometries, though the paper itself notes that complex geometries make this challenging.
  • The early-step transient error attributed to GRU initial hidden states might be reduced by sequence models with better state initialization, such as transformers or washout scheduling, which the paper identifies as an open practical issue.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper extends the authors' I-FENN framework by using a GRU-enabled DeepONet to predict the coupled non-mechanical field (temperature or pore pressure) from load-history and strain-trace inputs, while a deal.II-based FEM solver computes the displacement field. Three numerical examples are presented: a 3D thermoelastic cube under random thermal body loads, a 3D thick-walled tube under random surface heat fluxes, and a 2D poroelastic excavation under random dewatering flux. On unseen test load cases the coupled fields are reported to be within 5% relative error in non-trivial regions, and Table 2 reports 35--43% runtime savings versus monolithic fully coupled FEM on progressively refined meshes. The paper also proposes a concatenated multi-branch DeepONet (a modified MIONet), a GRU-based temporal branch, and a boundary-condition enforcement scheme. The stability study in Section 5.1.5 documents and partially remedies an error-accumulation issue via group normalization.

Significance. If the accuracy and speedup claims hold, this is a useful contribution to hybrid ML/FEM simulation, and the paper is generally well organized. The percentile-based testing (10th/50th/90th) is more informative than a single cherry-picked case, and the stability study explicitly addresses an important failure mode. The proposed BC enforcement and multi-input operator architecture are clearly described. However, the headline speedup is not yet isolated: I-FENN's staggered structure may itself account for much of the gain, and offline training/data-generation costs are excluded from the savings. The evidence supports the accuracy claim for the tested problems, but the acceleration claim needs a controlled benchmark and an amortized cost model.

major comments (3)
  1. [§3.2, §5.1.4 (Table 2)] The speedup claim is confounded. Section 3.2 states that "I-FENN setup inherently follows a staggered scheme," and Section 3.1 describes it as conceptually similar to FEM-staggered approaches. Table 2 compares I-FENN only against a monolithic fully coupled FEM solver. A classical staggered FEM—solving Eq. (1) with the temperature/pressure from the previous step, then updating the second field by solving Eq. (2) or Eq. (4)—would also replace the monolithic system by smaller solves and may yield comparable or larger savings. Without such a baseline, the 35–43% savings cannot be attributed to the DeepONet. Please add a staggered-FEM comparison with the same linear solvers/preconditioners and report per-step cost decompositions or otherwise isolate the NN's contribution.
  2. [§5.1.4 (Table 2); §5.1.2, §5.2.2, §5.3.2] The reported savings exclude training and data generation. Training times are 9h04m (cube), 12h35m (tube), and 18m (poro), and each dataset requires 600–900 fully coupled FEM forward simulations. The text acknowledges this, but the claim that "combined cost becomes significantly lower" for many simulations on fine meshes is not supported by a break-even analysis. Please report total time including data generation and training as a function of the number of future simulations and mesh size, and provide repeated timing runs (at least 3–5) with ranges or standard deviations so the 35–43% figures are not single-run artifacts.
  3. [§5.1.3, §5.2.3, §5.3.3; §6] The central accuracy claim "less than 5% relative error for all non-trivial points" is qualitative. The figures cap ϵrel at 5%, and the text says that only zones with response close to zero exceed it, but "non-trivial" is never defined and no aggregate statistic is reported. Please define a threshold (e.g., |y_true| > a% of max|y_true|) and report the fraction of nodes/time steps exceeding 5% relative error, or an equivalent explicit metric, for all reported load cases. This would make the abstract claim falsifiable rather than a visual assessment.
minor comments (6)
  1. [§5.1.4] The sentence "computational times and savings across different domain sizes are plotted in Fig. 10" should refer to Fig. 12; Fig. 10 already shows the median load-case solution.
  2. [§6] The conclusion says "up to 40% savings," but Table 2 reports 43% for the finest mesh. Please correct the inconsistency.
  3. [§4.4, Eq. (14)] The set notation "{ΓD1, ΓDn, ..., ΓDn}" should be "{ΓD1, ..., ΓDn}".
  4. [Fig. 15 caption] The training loss label "LSEE" should be "LSSE" to match the notation in Section 4.5.
  5. [§B.2.6] The text says "Fig. B.15 depicts the effect of branch hidden size," but the figure and surrounding discussion concern trunk hidden size. Please correct.
  6. [§5.1.2 and elsewhere] The term "unseen" should be qualified as "unseen draws from the same random family used for training," because the test loads are generated from the same sampling distribution. Qualitatively different load families are not demonstrated.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the DeepONet is evaluated out-of-sample on unseen loads, the one-step lag is a stated modeling assumption tested explicitly, and the speedup table is a fresh timing comparison with training cost disclosed.

full rationale

The paper's central accuracy claim is an out-of-sample evaluation of a data-driven surrogate. Section 4.5 states that datasets are split into training/validation/testing and Section 1.2 states that all testing examples use loads never used during training. The error metric (Eqs. 22-25) compares y_pred to y_true from fully coupled FEM; no parameter is fitted after testing to force the reported <5% error. The one-step lag in Section 3.2 — predicting T_{n+1} or p_{n+1} from tr(epsilon_n) plus upcoming loads — is an explicit modeling assumption, not a construction that makes the output equal to the input; it is empirically probed in the stability study of Section 5.1.5, including a deliberate switch from true to predicted strain at steps 33 and 66. The computational savings in Table 2 are measured fresh in this paper against a monolithic FEM solver, and Section 5.1.4 explicitly discloses that DeepONet training time is excluded and that total time can exceed FEM for a few coarse simulations. Citations to prior I-FENN work (refs. 44, 50) provide lineage and motivation, but the present numerical results are generated and reported here rather than imported as fitted values. No equation in the paper reduces to a fitted parameter renamed as a prediction, and no load-bearing claim rests solely on a self-citation. The absence of a staggered-FEM baseline is a benchmarking limitation, not circular reasoning.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method rests on standard numerical analysis plus one problem-specific sufficiency assumption about strain trace. The free parameters are network hyperparameters and training choices tuned on validation data; they affect accuracy but are reported transparently. No new physical or mathematical entities are introduced.

free parameters (4)
  • Network hyperparameters per example = NH=200/50/64, NGRU=2, Nch=25-100, NFC=1-4, Dout sizes (Tables 1, 3, 5)
    Chosen through sensitivity analysis on validation sets; they directly affect the reported accuracy and stability.
  • Loss function choice = LSSE for examples 2 and 3, L2 for example 1
    Selected after testing multiple loss functions; Appendix B.2.5 shows the best choice changed between initial and final training setups.
  • Data normalization technique = Min-max scaling to [-1,1] for example 2
    Chosen based on lowest testing error in Appendix B.2.4.
  • Learning rate scheduler = Custom scheduler
    Adopted to stabilize training and reduce loss fluctuation (Appendices B.2.3 and B.3.1).
assumptions (4)
  • standard math FE discretization, implicit Euler time stepping, and DeepONet universal approximation
    Standard numerical and approximation theory tools used throughout Section 2 and Section 4.
  • domain assumption Small-deformation linear thermoelasticity and poroelasticity, isotropic materials, Fourier and Darcy constitutive laws
    Problem statement in Section 2 and Appendix A; restricts the claim to linear coupled regimes.
  • domain assumption Strain trace from the previous time step is a sufficient input to predict the next temperature/pressure field
    Section 3.2 explicitly simplifies NN inputs to scalar tr(epsilon_n); if the coupled field depends on other deformation details, the surrogate cannot be accurate.
  • ad hoc to paper Training, validation, and testing loads are drawn from the same random family
    Enables the 'unseen load' claim, but limits out-of-distribution generalization; load generation details in Sections 5.1-5.3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of I-FENN with DeepONets: accelerating simulations in coupled multiphysics problems." pith.science (2026). https://pith.science/paper/XHHFP4YB

@misc{pith2026250900604,
  author       = {Pith},
  title        = {Pith review of: I-FENN with DeepONets: accelerating simulations in coupled multiphysics problems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XHHFP4YB}},
  note         = {Machine review of arXiv:2509.00604}
}
read the original abstract

Coupled multiphysics simulations for high-dimensional, large-scale problems can be prohibitively expensive due to their computational demands. This article presents a novel framework integrating a deep operator network (DeepONet) with the Finite Element Method (FEM) to address coupled thermoelasticity and poroelasticity problems. This integration occurs within the context of I-FENN, a framework where neural networks are directly employed as PDE solvers within FEM, resulting in a hybrid staggered solver. In this setup, the mechanical field is computed using FEM, while the other coupled field is predicted using a neural network (NN). By decoupling multiphysics interactions, the hybrid framework reduces computational cost by simplifying calculations and reducing the FEM unknowns, while maintaining flexibility across unseen scenarios. The proposed work introduces a new I-FENN architecture with extended generalizability due to the DeepONets ability to efficiently address several combinations of natural boundary conditions and body loads. A modified DeepONet architecture is introduced to accommodate multiple inputs, along with a streamlined strategy for enforcing boundary conditions on distinct boundaries. We showcase the applicability and merits of the proposed work through numerical examples covering thermoelasticity and poroelasticity problems, demonstrating computational efficiency, accuracy, and generalization capabilities. In all examples, the test cases involve unseen loading conditions. The computational savings scale with the model complexity while preserving an accuracy of more than 95\% in the non-trivial regions of the domain.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

83 extracted references · 57 canonical work pages

  1. [1]

    Santi, M

    A. Santi, M. Bayat, J. Hattel, Multiphysics modeling of metal based additive manufacturing processes with focus on thermomechanical conditions https://www.tandfonline.com/doi/full/10.1080/01495739.2023.2195513, Journal of Thermal Stresses 46 (6) (2023) 445--463. https://doi.org/10.1080/01495739.2023.2195513 doi:10.1080/01495739.2023.2195513 . ://www.tandf...

    work page arXiv 2023

  2. [2]

    Q. Xu, S. Li, L. Wang, A. Dong, Y. Meng, https://onlinelibrary.wiley.com/doi/10.1155/2019/1890237 Thermomechanical Coupling Analysis and Optimization of Metallic Thermal Protection System , Shock and Vibration 2019 (1) (2019) 1890237. https://doi.org/10.1155/2019/1890237 doi:10.1155/2019/1890237 . ://onlinelibrary.wiley.com/doi/10.1155/2019/1890237

    work page doi:10.1155/2019/1890237 2019

  3. [3]

    Koric, L

    S. Koric, L. C. Hibbeler, R. Liu, B. G. Thomas, http://www.tandfonline.com/doi/abs/10.1080/10407790.2011.540954 Multiphysics Model of Metal Solidification on the Continuum Level , Numerical Heat Transfer, Part B: Fundamentals 58 (6) (2010) 371--392. https://doi.org/10.1080/10407790.2011.540954 doi:10.1080/10407790.2011.540954 . ://www.tandfonline.com/doi/...

    arXiv 2011

  4. [4]

    L. Guo, M. Fahs, H. Hoteit, R. Gao, Q. Shao, https://www.techscience.com/CMES/v129n1/44196 Uncertainty Analysis of Seepage - Induced Consolidation in a Fractured Porous Medium , Computer Modeling in Engineering & Sciences 129 (1) (2021) 279--297. https://doi.org/10.32604/cmes.2021.016619 doi:10.32604/cmes.2021.016619 . ://www.techscience.com/CMES/v129n1/44196

    work page arXiv 2021

  5. [5]

    Klöppel, A

    T. Klöppel, A. Popp, U. Küttler, W. A. Wall, Fluid–structure interaction for non-conforming interfaces based on a dual mortar formulation https://linkinghub.elsevier.com/retrieve/pii/S0045782511002246, Computer Methods in Applied Mechanics and Engineering 200 (45-46) (2011) 3111--3126. https://doi.org/10.1016/j.cma.2011.06.006 doi:10.1016/j.cma.2011.06.00...

    work page doi:10.1016/j.cma.2011.06.006 2011

  6. [6]

    Moradi, J

    S. Moradi, J. A. Huisman, H. Vereecken, H. Class, https://www.mdpi.com/2073-4441/16/2/312 Comparing Different Coupling and Modeling Strategies in Hydromechanical Models for Slope Stability Assessment , Water 16 (2) (2024) 312. https://doi.org/10.3390/w16020312 doi:10.3390/w16020312 . ://www.mdpi.com/2073-4441/16/2/312

    doi:10.3390/w16020312

  7. [7]

    D. Han, W. Zhang, K. Jiao, B. Yu, T. Li, L. Gong, S. Wang, Thermal‒hydraulic‒mechanical‒chemical coupling analysis of enhanced geothermal systems based on an embedded discrete fracture model https://linkinghub.elsevier.com/retrieve/pii/S2352854023000670, Natural Gas Industry B 10 (5) (2023) 533--546. https://doi.org/10.1016/j.ngib.2023.10.001 doi:10.1016/...

    work page doi:10.1016/j.ngib.2023.10.001 2023

  8. [8]

    Abdullah, Y

    S. Abdullah, Y. Ma, K. Wang, S. Subramanyam, X. Chen, A. Khan, Two-phase interaction in isothermal hydro-mechanical-chemical coupling for improved carbon geological sequestration modelling https://www.tandfonline.com/doi/full/10.1080/17486025.2023.2292160, Geomechanics and Geoengineering 19 (4) (2024) 532--550. https://doi.org/10.1080/17486025.2023.229216...

    work page arXiv 2023

Show all 83 references
  1. [9]

    S. V. Churakov, F. Claret, A. Idiart, D. Jacques, J. Govaerts, O. Kolditz, N. Prasianakis, J. Samper, Position paper on high fidelity simulations for coupled processes, multi-physics and chemistry in geological disposal of nuclear waste https://doi.org/10.1007/s12665-024-11832...

    2024 doi

  2. [10]

    D. E. Keyes, L. C. McInnes, C. Woodward, W. Gropp, E. Myra, M. Pernice, J. Bell, J. Brown, A. Clo, J. Connors, E. Constantinescu, D. Estep, K. Evans, C. Farhat, A. Hakim, G. Hammond, G. Hansen, J. Hill, T. Isaac, X. Jiao, K. Jordan, D. Kaushik, E. Kaxiras, A. Koniges, K. Lee, ...

    2013

  3. [11]

    T. Wang, F. Zhang, J. Furtney, B. Damjanac, A review of methods, applications and limitations for incorporating fluid flow in the discrete element method https://linkinghub.elsevier.com/retrieve/pii/S1674775522000014, Journal of Rock Mechanics and Geotechnical Engineering 14 (...

    2022 doi

  4. [12]

    T.-R. Liu, F. Aldakheel, M. Aliabadi, Virtual element method for phase field modeling of dynamic fracture https://linkinghub.elsevier.com/retrieve/pii/S0045782523001743, Computer Methods in Applied Mechanics and Engineering 411 (2023) 116050. https://doi.org/10.1016/j.cma.2023...

    2023

  5. [13]

    B. Chen, B. R. Barboza, Y. Sun, J. Bai, H. R. Thomas, M. Dutko, M. Cottrell, C. Li, https://link.springer.com/10.1007/s11831-021-09653-z A Review of Hydraulic Fracturing Simulation , Archives of Computational Methods in Engineering 29 (4) (2022) 1--58. https://doi.org/10.1007/...

    2022 doi

  6. [14]

    Diehl, R

    P. Diehl, R. Lipton, T. Wick, M. Tyagi, A comparative review of peridynamics and phase-field models for engineering fracture mechanics https://link.springer.com/10.1007/s00466-022-02147-0, Computational Mechanics 69 (6) (2022) 1259--1293. https://doi.org/10.1007/s00466-022-021...

    2022 doi

  7. [15]

    Moradinia, H

    Z. Moradinia, H. Vandierendonck, A. Murphy, Navigating speed–accuracy trade-offs for multi-physics simulations https://doi.org/10.1007/s11012-024-01867-2, Meccanica (Aug. 2024). https://doi.org/10.1007/s11012-024-01867-2 doi:10.1007/s11012-024-01867-2 . ://doi.org/10.1007/s110...

    2024 doi

  8. [16]

    Meinecke, M

    S. Meinecke, M. Selig, F. Köster, A. Knorr, K. Lüdge, Data-driven acceleration of multi-physics simulations https://iopscience.iop.org/article/10.1088/2632-2153/ad7572, Machine Learning: Science and Technology 5 (4) (2024) 045011. https://doi.org/10.1088/2632-2153/ad7572 doi:1...

    2024 doi

  9. [17]

    H. S. Tang, R. D. Haynes, G. Houzeaux, https://link.springer.com/10.1007/s11831-019-09394-0 A Review of Domain Decomposition Methods for Simulation of Fluid Flows : Concepts , Algorithms , and Applications , Archives of Computational Methods in Engineering 28 (3) (2021) 841--8...

    2021 doi

  10. [18]

    C. J. Permann, D. R. Gaston, D. Andrš, R. W. Carlsen, F. Kong, A. D. Lindsay, J. M. Miller, J. W. Peterson, A. E. Slaughter, R. H. Stogner, R. C. Martineau, https://linkinghub.elsevier.com/retrieve/pii/S2352711019302973 MOOSE : Enabling massively parallel multiphysics simulati...

    2020

  11. [19]

    Castellazzi, A

    G. Castellazzi, A. D’Altri, S. De Miranda, H. Emami, L. Molari, F. Ubertini, A staggered multiphysics framework for salt crystallization-induced damage in porous building materials https://linkinghub.elsevier.com/retrieve/pii/S095006182102242X, Construction and Building Materi...

    2021

  12. [20]

    A. J. Stershic, C. R. D’Elia, L. L. Beghini, M. R. Hill, B. Clausen, D. K. Balch, M. Maguire, C. W. San Marchi, J. W. Foulk, A. A. Hanson, K. L. Manktelow, Adaptively remeshed multiphysical modeling of resistance forge welding with experimental validation of residual stress fi...

    2025

  13. [21]

    G. E. Karniadakis, I. G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics-informed machine learning https://www.nature.com/articles/s42254-021-00314-5, Nature Reviews Physics 3 (6) (2021) 422--440. https://doi.org/10.1038/s42254-021-00314-5 doi:10.1038/s42254-021-00...

    2021 doi

  14. [22]

    Herrmann, S

    L. Herrmann, S. Kollmannsberger, Deep learning in computational mechanics: a review https://link.springer.com/10.1007/s00466-023-02434-4, Computational Mechanics 74 (2) (2024) 281--331. https://doi.org/10.1007/s00466-023-02434-4 doi:10.1007/s00466-023-02434-4 . ://link.springe...

    2024 doi

  15. [23]

    A. Pinkus, https://www.cambridge.org/core/product/identifier/S0962492900002919/type/journal_article Approximation theory of the MLP model in neural networks , Acta Numerica 8 (1999) 143--195. https://doi.org/10.1017/S0962492900002919 doi:10.1017/S0962492900002919 . ://www.camb...

    1999 doi

  16. [24]

    Chung, C

    J. Chung, C. Gulcehre, K. Cho, Y. Bengio, https://arxiv.org/abs/1412.3555 Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling (2014). https://doi.org/10.48550/ARXIV.1412.3555 doi:10.48550/ARXIV.1412.3555 . ://arxiv.org/abs/1412.3555

    2014 doi arXiv

  17. [25]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, I. Polosukhin, https://arxiv.org/abs/1706.03762 Attention Is All You Need (2017). https://doi.org/10.48550/ARXIV.1706.03762 doi:10.48550/ARXIV.1706.03762 . ://arxiv.org/abs/1706.03762

    2017 doi arXiv

  18. [26]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, https://www.nature.com/articles/s42256-021-00302-5 Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , Nature Machine Intelligence 3 (3) (2021) 218--229. https://doi.org/10.1...

    2021 doi

  19. [27]

    Raonić, R

    B. Raonić, R. Molinaro, T. De Ryck, T. Rohner, F. Bartolucci, R. Alaifari, S. Mishra, E. de Bézenac, https://arxiv.org/abs/2302.01178 Convolutional Neural Operators for robust and accurate learning of PDEs (2023). https://doi.org/10.48550/ARXIV.2302.01178 doi:10.48550/ARXIV.23...

    2023 doi arXiv

  20. [28]

    Sarkar, S

    S. Sarkar, S. Chakraborty, Spatio-spectral graph neural operator for solving computational mechanics problems on irregular domain and unstructured grid https://linkinghub.elsevier.com/retrieve/pii/S0045782524009137, Computer Methods in Applied Mechanics and Engineering 435 (20...

    2025

  21. [29]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, https://arxiv.org/abs/2010.08895 Fourier Neural Operator for Parametric Partial Differential Equations (2020). https://doi.org/10.48550/ARXIV.2010.08895 doi:10.48550/ARXIV.2010.08895 . :...

    2010 doi arXiv

  22. [30]

    Kadeethum, S

    T. Kadeethum, S. Verzi, H. Yoon, https://linkinghub.elsevier.com/retrieve/pii/S2949891024003774 An improved neural operator framework for large-scale CO2 storage operations , Geoenergy Science and Engineering 240 (2024) 213007. https://doi.org/10.1016/j.geoen.2024.213007 doi:1...

    2024

  23. [31]

    W. Wang, M. Hakimzadeh, H. Ruan, S. Goswami, https://linkinghub.elsevier.com/retrieve/pii/S0045782525005912 Time-marching neural operator– FE coupling: AI -accelerated physics modeling , Computer Methods in Applied Mechanics and Engineering 446 (2025) 118319. https://doi.org/1...

    2025

  24. [32]

    P. Jin, S. Meng, L. Lu, https://epubs.siam.org/doi/10.1137/22M1477751 MIONet : Learning Multiple - Input Operators via Tensor Product , SIAM Journal on Scientific Computing 44 (6) (2022) A3490--A3514. https://doi.org/10.1137/22M1477751 doi:10.1137/22M1477751 . ://epubs.siam.or...

    2022 doi

  25. [33]

    Peyvan, V

    A. Peyvan, V. Kumar, G. E. Karniadakis, http://arxiv.org/abs/2501.01934 Fusion- DeepONet : A Data - Efficient Neural Operator for Geometry - Dependent Hypersonic and Supersonic Flows , arXiv:2501.01934 (May 2025). https://doi.org/10.48550/arXiv.2501.01934 doi:10.48550/arXiv.25...

    2025 doi arXiv

  26. [34]

    X. Sun, B. Bahmani, N. N. Vlassis, W. Sun, Y. Xu, Data-driven discovery of interpretable causal relations for deep learning material laws with uncertainty propagation, Granular Matter 24 (2022) 1--32

    2022

  27. [35]

    Manfren, P

    M. Manfren, P. A. James, L. Tronchin, https://linkinghub.elsevier.com/retrieve/pii/S1364032122005779 Data-driven building energy modelling – An analysis of the potential for generalisation through interpretable machine learning , Renewable and Sustainable Energy Reviews 167 (2...

    2022

  28. [36]

    J. N. Fuhg, N. Bouklas, On physics-informed data-driven isotropic and anisotropic constitutive models through probabilistic machine learning and space-filling sampling, Computer Methods in Applied Mechanics and Engineering 394 (2022) 114915

    2022

  29. [37]

    Pantidis, H

    P. Pantidis, H. Eldababy, C. M. Tagle, M. E. Mobasher, Error convergence and engineering-guided hyperparameter search of PINNs : towards optimized I - FENN performance, Computer Methods in Applied Mechanics and Engineering 414 (2023) 116160

    2023

  30. [38]

    A. Daw, J. Bu, S. Wang, P. Perdikaris, A. Karpatne, Rethinking the importance of sampling in physics-informed neural networks, arXiv preprint arXiv:2207.02338 (2022)

    2022 arXiv

  31. [39]

    Markidis, The old and the new: Can physics-informed deep-learning replace traditional linear solvers?, Frontiers in big Data 4 (2021) 669097

    S. Markidis, The old and the new: Can physics-informed deep-learning replace traditional linear solvers?, Frontiers in big Data 4 (2021) 669097

    2021

  32. [40]

    id":348,

    Y. Shin, J. Darbon, G. E. Karniadakis, https://global-sci.com/article/79748/on-the-convergence-of-physics-informed-neural-networks-for-linear-second-order-elliptic-and-parabolic-type-pdes On the Convergence of Physics Informed Neural Networks for Linear Second - Order Elliptic...

    2024

  33. [41]

    Mishra, R

    S. Mishra, R. Molinaro, https://academic.oup.com/imajna/article/42/2/981/6297946 Estimates on the generalization error of physics-informed neural networks for approximating a class of inverse problems for PDEs , IMA Journal of Numerical Analysis 42 (2) (2022) 981--1022. https:...

    2022 doi

  34. [42]

    Krishnapriyan, A

    A. Krishnapriyan, A. Gholami, S. Zhe, R. Kirby, M. W. Mahoney, Characterizing possible failure modes in physics-informed neural networks, Advances in Neural Information Processing Systems 34 (2021) 26548--26560

    2021

  35. [43]

    S. Wang, Y. Teng, P. Perdikaris, Understanding and mitigating gradient flow pathologies in physics-informed neural networks, SIAM Journal on Scientific Computing 43 (5) (2021) A3055--A3081

    2021

  36. [44]

    Pantidis, M

    P. Pantidis, M. E. Mobasher, https://linkinghub.elsevier.com/retrieve/pii/S0045782522007228 Integrated Finite Element Neural Network ( I - FENN ) for non-local continuum damage mechanics , Computer Methods in Applied Mechanics and Engineering 404 (2023) 115766. https://doi.org...

    2023

  37. [45]

    S. Thel, L. Greve, B. Van De Weg, P. Van Der Smagt, https://linkinghub.elsevier.com/retrieve/pii/S0045782524003293 Introducing Finite Element Method Integrated Networks ( FEMIN ) , Computer Methods in Applied Mechanics and Engineering 427 (2024) 117073. https://doi.org/10.1016...

    2024

  38. [46]

    S. K. Mitusch, S. W. Funke, M. Kuchta, https://linkinghub.elsevier.com/retrieve/pii/S0021999121005465 Hybrid FEM - NN models: Combining artificial neural networks with the finite element method , Journal of Computational Physics 446 (2021) 110651. https://doi.org/10.1016/j.jcp...

    2021

  39. [47]

    R. E. Meethal, A. Kodakkal, M. Khalil, A. Ghantasala, B. Obst, K.-U. Bletzinger, R. Wüchner, Finite element method-enhanced neural network for forward and inverse problems https://amses-journal.springeropen.com/articles/10.1186/s40323-023-00243-1, Advanced Modeling and Simulat...

    2023 doi

  40. [48]

    Zhang, D

    W. Zhang, D. S. Li, T. Bui-Thanh, M. S. Sacks, https://linkinghub.elsevier.com/retrieve/pii/S0045782522001724 Simulation of the 3D hyperelastic behavior of ventricular myocardium using a finite-element based neural-network approach , Computer Methods in Applied Mechanics and E...

    2022

  41. [49]

    Pantidis, L

    P. Pantidis, L. Svolos, D. Abueidda, M. E. Mobasher, https://arxiv.org/abs/2505.19566 Integrated Finite Element Neural Network ( IFENN ) for Phase - Field Fracture with Minimal Input and Generalized Geometry - Load Handling (2025). https://doi.org/10.48550/ARXIV.2505.19566 doi...

    2025 doi arXiv

  42. [50]

    Pantidis, H

    P. Pantidis, H. Eldababy, D. Abueidda, M. E. Mobasher, https://linkinghub.elsevier.com/retrieve/pii/S0045782524001968 I- FENN with Temporal Convolutional Networks : Expediting the load-history analysis of non-local gradient damage propagation , Computer Methods in Applied Mech...

    2024

  43. [51]

    D. W. Abueidda, M. E. Mobasher, https://link.springer.com/10.1007/s00466-024-02475-3 I- FENN for thermoelasticity based on physics-informed temporal convolutional network ( PI - TCN ) , Computational Mechanics 74 (6) (2024) 1229--1259. https://doi.org/10.1007/s00466-024-02475-...

    2024 doi

  44. [52]

    D. W. Abueidda, M. E. Mobasher, https://linkinghub.elsevier.com/retrieve/pii/S0045782524003785 Variational temporal convolutional networks for I - FENN thermoelasticity , Computer Methods in Applied Mechanics and Engineering 429 (2024) 117122. https://doi.org/10.1016/j.cma.202...

    2024

  45. [53]

    P. C. Africa, D. Arndt, W. Bangerth, B. Blais, M. Fehling, R. Gassmöller, T. Heister, L. Heltai, S. Kinnewig, M. Kronbichler, M. Maier, P. Munch, M. Schreter-Fleischhacker, J. P. Thiele, B. Turcksin, D. Wells, V. Yushutin, https://www.degruyter.com/document/doi/10.1515/jnma-20...

    2024 doi

  46. [54]

    Manav, R

    M. Manav, R. Molinaro, S. Mishra, L. De Lorenzis, Phase-field modeling of fracture with physics-informed deep learning, Computer Methods in Applied Mechanics and Engineering 429 (2024) 117104

    2024

  47. [55]

    Balay, S

    S. Balay, S. Abhyankar, M. F. Adams, S. Benson, J. Brown, P. Brune, K. Buschelman, E. M. Constantinescu, L. Dalcin, A. Dener, V. Eijkhout, J. Faibussowitsch, W. D. Gropp, V. Hapla, T. Isaac, P. Jolivet, D. Karpeev, D. Kaushik, M. G. Knepley, F. Kong, S. Kruger, D. A. May, L. C...

    2024

  48. [56]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, G. E. Karniadakis, https://linkinghub.elsevier.com/retrieve/pii/S0045782522001207 A comprehensive and fair comparison of two neural operators (with practical extensions) based on FAIR data , Computer Methods in Applied Mech...

    2022

  49. [57]

    Jiang, M

    Z. Jiang, M. Zhu, L. Lu, https://linkinghub.elsevier.com/retrieve/pii/S0951832024004642 Fourier- MIONet : Fourier -enhanced multiple-input neural operators for multiphase modeling of geological carbon sequestration , Reliability Engineering & System Safety 251 (2024) 110392. h...

    2024

  50. [58]

    D. Liu, X. Guo, Z. Liu, J. Tan, https://asmedigitalcollection.asme.org/computingengineering/article/25/8/081004/1215310/An-Image-Generator-Enhanced-Deep-Operator-Network An Image Generator Enhanced Deep Operator Network for Predicting the Geometry Deformations in Contact Probl...

    2025 doi

  51. [59]

    K. Cho, B. van Merrienboer, D. Bahdanau, Y. Bengio, https://arxiv.org/abs/1409.1259 On the Properties of Neural Machine Translation : Encoder - Decoder Approaches (2014). https://doi.org/10.48550/ARXIV.1409.1259 doi:10.48550/ARXIV.1409.1259 . ://arxiv.org/abs/1409.1259

    2014 doi arXiv

  52. [60]

    Tsantekidis, N

    A. Tsantekidis, N. Passalis, A. Tefas, Recurrent neural networks https://linkinghub.elsevier.com/retrieve/pii/B9780323857871000105, in: Deep Learning for Robot Perception and Cognition , Elsevier, 2022, pp. 101--115. https://doi.org/10.1016/B978-0-32-385787-1.00010-5 doi:10.10...

    2022 doi

  53. [61]

    Zhang, Z

    A. Zhang, Z. C. Lipton, M. Li, A. J. Smola, Dive into Deep Learning , Cambridge University Press, 2023

    2023

  54. [62]

    Sukumar, A

    N. Sukumar, A. Srivastava, Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks https://linkinghub.elsevier.com/retrieve/pii/S0045782521006186, Computer Methods in Applied Mechanics and Engineering 389 (2022) 114333. https://...

    2022

  55. [63]

    Dyken, M

    C. Dyken, M. S. Floater, Transfinite mean value interpolation https://linkinghub.elsevier.com/retrieve/pii/S0167839607001240, Computer Aided Geometric Design 26 (1) (2009) 117--134. https://doi.org/10.1016/j.cagd.2007.12.003 doi:10.1016/j.cagd.2007.12.003 . ://linkinghub.elsev...

    2009 doi

  56. [64]

    S. Liu, Z. Hao, C. Ying, H. Su, J. Zhu, Z. Cheng, https://openreview.net/forum?id=GNt5ntEGjD3 A Unified Hard - Constraint Framework for Solving Geometrically Complex PDEs , in: A. H. Oh, A. Agarwal, D. Belgrave, K. Cho (Eds.), Advances in Neural Information Processing Systems ...

    2022

  57. [65]

    Horie, N

    M. Horie, N. Mitsume, https://arxiv.org/abs/2205.11912 Physics- Embedded Neural Networks : Graph Neural PDE Solvers with Mixed Boundary Conditions (2022). https://doi.org/10.48550/ARXIV.2205.11912 doi:10.48550/ARXIV.2205.11912 . ://arxiv.org/abs/2205.11912

    2022 doi arXiv

  58. [66]

    Migacz, https://docs.pytorch.org/tutorials/recipes/recipes/tuning_guide.html Performance Tuning Guide (Jul

    S. Migacz, https://docs.pytorch.org/tutorials/recipes/recipes/tuning_guide.html Performance Tuning Guide (Jul. 2025). ://docs.pytorch.org/tutorials/recipes/recipes/tuning_guide.html

    2025

  59. [67]

    Janson, B

    P. Janson, B. Therien, Q. Anthony, X. Huang, A. Moudgil, E. Belilovsky, http://arxiv.org/abs/2506.10315 PyLO : Towards Accessible Learned Optimizers in PyTorch , arXiv:2506.10315 version: 1 (Jun. 2025). https://doi.org/10.48550/arXiv.2506.10315 doi:10.48550/arXiv.2506.10315 . ...

    2025 doi arXiv

  60. [68]

    Mohajerin, S

    N. Mohajerin, S. L. Waslander, https://ieeexplore.ieee.org/document/8630673/ Multistep Prediction of Dynamic Systems With Recurrent Neural Networks , IEEE Transactions on Neural Networks and Learning Systems 30 (11) (2019) 3370--3383. https://doi.org/10.1109/TNNLS.2019.2891257...

    2019

  61. [69]

    Bonassi, M

    F. Bonassi, M. Farina, J. Xie, R. Scattolini, https://linkinghub.elsevier.com/retrieve/pii/S0959152422000610 On Recurrent Neural Networks for learning-based control: Recent results and ideas for future developments , Journal of Process Control 114 (2022) 92--104. https://doi.o...

    2022 doi

  62. [70]

    F. M. Bianchi, E. Maiorino, M. C. Kampffmeyer, A. Rizzi, R. Jenssen, http://link.springer.com/10.1007/978-3-319-70338-1 Recurrent Neural Networks for Short - Term Load Forecasting : An Overview and Comparative Analysis , SpringerBriefs in Computer Science , Springer Internatio...

    2017 doi

  63. [71]

    Bonassi, M

    F. Bonassi, M. Farina, R. Scattolini, https://linkinghub.elsevier.com/retrieve/pii/S0167691121001791 On the stability properties of Gated Recurrent Units neural networks , Systems & Control Letters 157 (2021) 105049. https://doi.org/10.1016/j.sysconle.2021.105049 doi:10.1016/j...

    2021

  64. [72]

    D. M. Potts, L. Zdravković, Finite element analysis in geotechnical engineering. 2: Application , Telford, London, 2001

    2001

  65. [73]

    Li, Finite element formulation of poro-elasticity suitable for large deformation dynamic analysis (2004)

    C. Li, Finite element formulation of poro-elasticity suitable for large deformation dynamic analysis (2004)

    2004

  66. [74]

    M. E. Gurtin, The linear theory of elasticity, Linear Theories of Elasticity and Thermoelasticity: Linear and Nonlinear Theories of Rods, Plates, and Shells (1973) 1--295

    1973

  67. [75]

    Abeyaratne, Continuum Mechanics , Lecture Notes on The Mechanics of Elastic Solids (1998)

    R. Abeyaratne, Continuum Mechanics , Lecture Notes on The Mechanics of Elastic Solids (1998)

    1998

  68. [76]

    C. Truesdell, Linear Theories of Elasticity and Thermoelasticity : Linear and Nonlinear Theories of Rods , Plates , and Shells , Springer Berlin Heidelberg, Berlin, Heidelberg s.l, 1973

    1973

  69. [77]

    D. M. Potts, L. Zdravković, Finite element analysis in geotechnical engineering: theory, Thomas Telford, Londres, 1999

    1999

  70. [78]

    Coussy, Poromechanics, Wiley, Chichester, 2004

    O. Coussy, Poromechanics, Wiley, Chichester, 2004

    2004

  71. [79]

    A. H.-D. Cheng, Poroelasticity http://link.springer.com/10.1007/978-3-319-25202-5, Vol. 27 of Theory and Applications of Transport in Porous Media , Springer International Publishing, Cham, 2016. https://doi.org/10.1007/978-3-319-25202-5 doi:10.1007/978-3-319-25202-5 . ://link...

    2016 doi

  72. [80]

    L.-P. Yi, H. Waisman, Z.-Z. Yang, X.-G. Li, A consistent phase field model for hydraulic fracture propagation in poroelastic media https://linkinghub.elsevier.com/retrieve/pii/S0045782520305818, Computer Methods in Applied Mechanics and Engineering 372 (2020) 113396. https://d...

    2020

  73. [81]

    Y. Chen, M. E. Mobasher, C. Gu, D. Zheng, H. Waisman, Thermodynamically consistent non-local damage formulation for fluid-driven fracture in poro-viscoelastic media https://link.springer.com/10.1007/s11440-022-01557-x, Acta Geotechnica 17 (11) (2022) 5321--5350. https://doi.or...

    2022 doi

  74. [82]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence afte...

  75. [83]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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