pith. machine review for the scientific record. sign in

arxiv: 2605.07738 · v1 · submitted 2026-05-08 · ⚛️ physics.comp-ph · cs.LG

Recognition: no theorem link

Physics-Informed Reduced-Order Operator Learning for Hyperelasticity in Continuum Micromechanics

Hamidreza Eivazi , Henning Wessels
Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:57 UTC · model grok-4.3

classification ⚛️ physics.comp-ph cs.LG
keywords physics-informed operator learningreduced-order modelinghyperelasticitymicromechanicsQ-DEIMrepresentative volume elementfinite-strainEquiNO
0
0 comments X

The pith

Reduced bases and sparse point evaluations make physics-informed operator learning feasible for three-dimensional hyperelastic microstructures at three-order cost savings.

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

The paper shows how to learn operators for hyperelastic RVEs by training only on modal coefficients of precomputed reduced bases for displacement fluctuations and stresses. These bases are built to be periodic and divergence-free, so the learned model automatically satisfies mechanical equilibrium and boundary conditions without extra penalties. Q-DEIM selects a small number of spatial points for constitutive evaluations during training, cutting per-step cost by roughly three orders of magnitude compared with full-field loss. Homogenized stresses are recovered directly from the offline-averaged modes, delivering 10^3 to 10^4 speed-up at inference while interpolating and extrapolating accurately from only a handful of offline loading-path snapshots. A reader cares because this removes the main practical barrier to using physics-informed surrogates inside large-scale multiscale finite-element simulations.

Core claim

EquiNO combined with Q-DEIM learns only the modal coefficients of reduced displacement-fluctuation and first Piola-Kirchhoff stress representations built from periodic and divergence-free bases, thereby enforcing periodicity and mechanical equilibrium by construction. Q-DEIM identifies a small set of spatial points through column-pivoted QR factorization of the stress basis and restricts constitutive evaluations during training to these points alone. Homogenized stresses are recovered directly from the offline-averaged reduced stress modes without reconstructing the full field at inference time. The framework achieves three-order-of-magnitude reductions in per-step training cost and 10^3–10^

What carries the argument

Equilibrium Neural Operator (EquiNO) augmented by QR-based discrete empirical interpolation (Q-DEIM) on reduced periodic and divergence-free bases for displacement fluctuations and stresses, which enforces physical constraints by construction and restricts loss evaluation to a sparse set of selected points.

If this is right

  • Full-batch second-order optimization becomes practical for three-dimensional RVEs because loss evaluation is restricted to a few dozen spatial points.
  • Homogenized first Piola-Kirchhoff stresses are obtained at inference without ever reconstructing the full stress field.
  • Microscopic stress fields and homogenized quantities can be predicted accurately for both interpolation and extrapolation when the number of offline snapshots is increased.
  • Reduced homogenization delivers speed-up factors of 10^3 to 10^4 relative to direct full-field computations on the same RVEs.

Where Pith is reading between the lines

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

  • The same reduced-basis construction could be reused across multiple constitutive models if the bases remain divergence-free and periodic, allowing one set of offline snapshots to serve several material laws.
  • Embedding the resulting operator inside a macroscopic finite-element code would turn each RVE evaluation into a cheap forward pass whose cost is independent of RVE resolution.
  • If the prediction error saturates only after a modest number of snapshots, the method implies that the solution manifold of hyperelastic RVEs under periodic boundary conditions is intrinsically low-dimensional for many practical loading regimes.

Load-bearing premise

A small number of offline snapshot loading paths suffice to construct reduced bases that capture the essential physics for accurate interpolation and extrapolation across loading conditions in three-dimensional finite-strain hyperelastic RVEs.

What would settle it

Run the trained model on loading paths that lie well outside the snapshot set and observe whether the error in both microscopic stress fields and homogenized stresses remains below the threshold achieved by direct full-field finite-element computations on the same RVEs.

Figures

Figures reproduced from arXiv: 2605.07738 by Hamidreza Eivazi, Henning Wessels.

Figure 1
Figure 1. Figure 1: Microstructures considered in the study. Blue and grey colors indicate the fiber and matrix phases, respectively. The fiber [PITH_FULL_IMAGE:figures/full_fig_p008_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Unsupervised loading paths (red), snapshot loading paths (blue) and unseen test loading paths (orange) in the macroscopic [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Mean relative L2 errors (left) and speed-up factors (right) for the stochastic RVE versus the number of Q-DEIM points in the in-range setting with np = 20 and a 2 × 64 network discussed in section 3.5. generated for basis construction, the physics-informed training can incorporate a large number of additional loading states at a cost below that of a single loading-path finite-element simulation [PITH_FULL… view at source ↗
Figure 4
Figure 4. Figure 4: Representative learning curve for the stochastic-fiber (left) and hexagonal-fiber (right) RVEs in the out-of-range setting [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of reference, predicted, and absolute-error fields for the microscopic first Piola–Kirchho [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of reference, predicted, and absolute-error fields for the microscopic first Piola–Kirchho [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Parity plots for the homogenized first Piola–Kirchho [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

Physics-informed operator learning is an attractive candidate for surrogate modeling of microstructures, especially in multiscale finite-element simulations. Its practical use, however, is often limited by the high cost of loss evaluation. We address this bottleneck by combining the Equilibrium Neural Operator (EquiNO) with the QR-based discrete empirical interpolation method (Q-DEIM). EquiNO learns only the modal coefficients of reduced displacement-fluctuation and first Piola-Kirchhoff stress representations built from periodic and divergence-free bases, thereby enforcing periodicity and mechanical equilibrium by construction. Q-DEIM then identifies a small set of spatial points through a column-pivoted QR factorization of the stress basis and restricts constitutive evaluations during training to these points alone. This makes full-batch second-order optimization practical for three-dimensional representative volume elements (RVEs). Homogenized first Piola-Kirchhoff stresses are recovered directly from the offline-averaged reduced stress modes, without the need to reconstruct the full stress field at inference time. We validate the framework on two three-dimensional finite-strain hyperelastic RVEs. Q-DEIM reduces the per-step training cost by roughly three orders of magnitude relative to full-field loss evaluation, while reduced homogenization achieves speed-up factors of order $10^3$ to $10^4$ over direct full-field computations. Despite relying on only a small number of offline snapshot loading paths for basis construction, the method accurately interpolates and extrapolates both microscopic stress fields and homogenized stresses, with prediction quality improving systematically as more snapshots are added.

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.

Referee Report

2 major / 1 minor

Summary. The paper introduces a reduced-order physics-informed operator learning method for hyperelastic micromechanics by combining the Equilibrium Neural Operator (EquiNO) with Q-DEIM. Reduced bases for displacement fluctuations and first Piola-Kirchhoff stress are constructed from periodic and divergence-free modes so that periodicity and equilibrium are satisfied by construction. Q-DEIM selects a small set of spatial points via column-pivoted QR on the stress basis, restricting constitutive evaluations during training to those points and enabling practical full-batch second-order optimization. Homogenized stresses are recovered directly from offline-averaged reduced modes without full-field reconstruction at inference. Validation is reported on two 3D finite-strain hyperelastic RVEs, claiming three-order-of-magnitude reduction in per-step training cost and 10^3–10^4 speed-up in homogenization relative to full-field solves, with accuracy improving as the number of offline snapshot loading paths increases.

Significance. If the reported efficiency gains and extrapolation accuracy hold, the framework would remove a major practical barrier to deploying physics-informed neural operators in multiscale finite-element simulations of nonlinear materials. The by-construction enforcement of mechanical constraints and the direct recovery of homogenized quantities from reduced modes are genuine strengths that provide grounding independent of the neural training loop. The approach could enable routine use of operator learning for three-dimensional RVEs where full-field loss evaluation has previously been prohibitive.

major comments (2)
  1. [Abstract] Abstract: the claim that 'the method accurately interpolates and extrapolates both microscopic stress fields and homogenized stresses' with only 'a small number of offline snapshot loading paths' is load-bearing for the central contribution, yet no quantitative error metrics, error bars, or comparison tables against full-field solutions are supplied; the abstract only states that quality 'improves systematically' without numbers.
  2. [Abstract] Abstract and validation description: the extrapolation guarantee rests on the unverified assumption that bases from a small number of snapshot paths span the essential manifold for arbitrary macroscopic deformation gradients in 3D finite-strain hyperelasticity. No Kolmogorov-width estimate, a-priori projection-error bound, or sensitivity study to the choice of loading paths is provided, leaving the reduced-space approximation error uncontrolled even if the NN is perfectly trained.
minor comments (1)
  1. The description of how the offline-averaged reduced stress modes are computed for homogenization could be expanded with an explicit formula to make the inference procedure fully reproducible from the text alone.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive and detailed review, as well as the positive assessment of the framework's potential significance for multiscale simulations. We address each major comment below and outline the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that 'the method accurately interpolates and extrapolates both microscopic stress fields and homogenized stresses' with only 'a small number of offline snapshot loading paths' is load-bearing for the central contribution, yet no quantitative error metrics, error bars, or comparison tables against full-field solutions are supplied; the abstract only states that quality 'improves systematically' without numbers.

    Authors: We agree that the abstract would be strengthened by the inclusion of specific quantitative metrics. The body of the manuscript already contains detailed error analyses (relative L2 norms for both microscopic fields and homogenized stresses, with comparisons to full-field solutions across multiple snapshot counts), but these are not summarized numerically in the abstract. In the revised manuscript we will update the abstract to report representative error values and their systematic improvement with additional snapshots. revision: yes

  2. Referee: [Abstract] Abstract and validation description: the extrapolation guarantee rests on the unverified assumption that bases from a small number of snapshot paths span the essential manifold for arbitrary macroscopic deformation gradients in 3D finite-strain hyperelasticity. No Kolmogorov-width estimate, a-priori projection-error bound, or sensitivity study to the choice of loading paths is provided, leaving the reduced-space approximation error uncontrolled even if the NN is perfectly trained.

    Authors: The method is data-driven and the quality of the reduced basis is assessed empirically through projection errors on both training and unseen deformation paths, with results showing clear improvement as the number of snapshots increases. We do not claim a theoretical guarantee that any finite set of paths spans the full manifold. While a Kolmogorov-width estimate or general a-priori bound is not provided (and would require substantial additional theoretical development outside the current scope), we will add a sensitivity study to the choice of loading paths in the revised manuscript to better quantify the dependence of basis quality on snapshot selection. revision: partial

standing simulated objections not resolved
  • A rigorous Kolmogorov-width estimate or a-priori projection-error bound for the reduced bases under arbitrary 3D finite-strain hyperelastic deformation gradients.

Circularity Check

0 steps flagged

No significant circularity; constraints and homogenization are enforced by construction from offline bases

full rationale

The derivation constructs periodic and divergence-free bases from a fixed set of offline snapshots, then uses EquiNO to learn only modal coefficients while enforcing periodicity and equilibrium directly via the basis choice rather than through any fitted loss term. Homogenized stresses are recovered by direct offline averaging of the reduced stress modes, independent of the neural training loop or any online prediction. Q-DEIM point selection is a deterministic column-pivoted QR factorization of the precomputed stress basis and does not rename or refit any training output as a prediction. No self-citation is invoked as a load-bearing uniqueness theorem, and no ansatz or known empirical pattern is smuggled in via citation. The central claims rest on the separation between offline basis construction and online coefficient learning, which remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that reduced modal bases derived from snapshots plus Q-DEIM point selection suffice for accurate prediction; no new physical entities are introduced and no free parameters are numerically fitted in the abstract description.

free parameters (1)
  • number of reduced modes and snapshot paths
    Basis size and snapshot count are selected from offline data but no specific values or fitting procedure are stated in the abstract.
axioms (1)
  • domain assumption Periodic and divergence-free bases enforce periodicity and mechanical equilibrium by construction
    Explicitly stated as the core of EquiNO in the abstract.

pith-pipeline@v0.9.0 · 5576 in / 1567 out tokens · 65857 ms · 2026-05-11T02:57:49.426376+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 2 internal anchors

  1. [1]

    Hassani, E

    B. Hassani, E. Hinton, A review of homogenization and topology optimization i—homogenization theory for media with periodic structure, Comput. & Struct. 69 (1998) 707–717. doi:https://doi.org/10.1016/S0045-7949(98)00131-X

    work page doi:10.1016/s0045-7949(98)00131-x 1998

  2. [2]

    Geers, V

    M. Geers, V . Kouznetsova, W. Brekelmans, Multi-scale computational homogenization: Trends and challenges, J. Comput. Appl. Math. 234 (2010) 2175–2182. doi:https://doi.org/10.1016/j.cam.2009.08.077, fourth International Confer- ence on Advanced COmputational Methods in ENgineering (ACOMEN 2008)

    work page doi:10.1016/j.cam.2009.08.077 2010

  3. [3]

    Matouˇs, M

    K. Matouˇs, M. G. Geers, V . G. Kouznetsova, A. Gillman, A review of predictive nonlinear theories for multiscale modeling of heterogeneous materials, J. Comput. Phys. 330 (2017) 192–220. doi:https://doi.org/10.1016/j.jcp.2016.10.070

    work page doi:10.1016/j.jcp.2016.10.070 2017

  4. [4]

    Miehe, A

    C. Miehe, A. Koch, Computational micro-to-macro transitions of discretized microstructures undergoing small strains, Arch. Appl. Mech. 72 (2002) 300–317

    work page 2002

  5. [5]

    Neural operator: Learning maps between function spaces

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural operator: Learning maps between function spaces, arXiv preprint arXiv:2108.08481 (2023). doi:10.48550/arXiv.2108.08481

    work page doi:10.48550/arxiv.2108.08481 2023

  6. [6]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G. E. Karniadakis, Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators, Nat. Mach. Intell. 3 (2021) 218–229. doi:10.1038/s42256-021-00302-5

    work page doi:10.1038/s42256-021-00302-5 2021

  7. [7]

    Z. Li, N. Kovachki, K. Azizzadenesheli, B. Liu, K. Bhattacharya, A. Stuart, A. Anandkumar, Fourier neural operator for parametric partial differential equations, arXiv preprint arXiv:2010.08895 (2020). doi:https://doi.org/10.48550/arXiv. 2010.08895

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv 2010

  8. [8]

    L. Lu, X. Meng, S. Cai, Z. Mao, S. Goswami, Z. Zhang, G. E. Karniadakis, A comprehensive and fair comparison of two neural operators (with practical extensions) based on FAIR data, Comput. Methods Appl. Mech. Eng. 393 (2022) 114778. doi:https://doi.org/10.1016/j.cma.2022.114778

    work page doi:10.1016/j.cma.2022.114778 2022

  9. [9]

    Bhattacharya, B

    K. Bhattacharya, B. Hosseini, N. B. Kovachki, A. M. Stuart, Model reduction and neural networks for parametric PDEs, arXiv preprint arXiv:2005.03180 (2020). doi:10.48550/arXiv.2005.03180

    work page doi:10.48550/arxiv.2005.03180 2005

  10. [10]

    Eivazi, S

    H. Eivazi, S. Wittek, A. Rausch, Nonlinear model reduction for operator learning, 2024. doi:10.48550/arXiv.2403.18735. arXiv:2403.18735, tiny Papers at ICLR 2024

    work page doi:10.48550/arxiv.2403.18735 2024

  11. [11]

    P. Chen, T. Hu, J. Guilleminot, A nonlinear-manifold reduced-order model and operator learning for partial differential equations with sharp solution gradients, Comput. Methods Appl. Mech. Eng. 419 (2024) 116684. doi: https://doi.org/10. 1016/j.cma.2023.116684

    work page arXiv 2024

  12. [12]

    J. H. Seidman, G. Kissas, P. Perdikaris, G. J. Pappas, NOMAD: Nonlinear manifold decoders for operator learning, in: Advances in Neural Information Processing Systems, volume 35, 2022

    work page 2022

  13. [13]

    Chaturantabut, D

    S. Chaturantabut, D. C. Sorensen, Nonlinear model reduction via discrete empirical interpolation, SIAM J. on Sci. Comput. 32 (2010) 2737–2764. doi:10.1137/090766498

    work page doi:10.1137/090766498 2010

  14. [14]

    Drmaˇc, S

    Z. Drmaˇc, S. Gugercin, A new selection operator for the discrete empirical interpolation method—improved a priori error bound and extensions, SIAM J. on Sci. Comput. 38 (2016) A631–A648. doi:10.1137/15M1019271

    work page doi:10.1137/15m1019271 2016

  15. [15]

    J. A. Hern´andez, M. A. Caicedo, A. Ferrer, Dimensional hyper-reduction of nonlinear finite element models via empirical cubature, Comput. Methods Appl. Mech. Eng. 313 (2017) 687–722. doi: https://doi.org/10.1016/j.cma.2016.10. 022

    work page doi:10.1016/j.cma.2016.10 2017

  16. [16]

    R. A. van Tuijl, C. Harnish, K. Matou ˇs, J. J. C. Remmers, M. G. D. Geers, Wavelet based reduced order models for microstructural analyses, Comput. Mech. 63 (2019) 535–554. doi:https://doi.org/10.1007/s00466-018-1608-3

    work page doi:10.1007/s00466-018-1608-3 2019

  17. [17]

    J. A. Hern ´andez, J. Oliver, A. E. Huespe, M. A. Caicedo, J. C. Cante, High-performance model reduction techniques in computational multiscale homogenization, Comput. Methods Appl. Mech. Eng. 276 (2014) 149–189. doi: https://doi. org/10.1016/j.cma.2014.03.011

    work page doi:10.1016/j.cma.2014.03.011 2014

  18. [18]

    J. A. Hern´andez, A multiscale method for periodic structures using domain decomposition and ECM-hyperreduction, Comput. Methods Appl. Mech. Eng. 368 (2020) 113192. doi:https://doi.org/10.1016/j.cma.2020.113192

    work page doi:10.1016/j.cma.2020.113192 2020

  19. [19]

    Raschi, O

    M. Raschi, O. Lloberas-Valls, A. Huespe, J. Oliver, High performance reduction technique for multiscale finite element modeling (HPR-FE2): Towards industrial multiscale FE software, Comput. Methods Appl. Mech. Eng. 375 (2021) 113580. doi:https://doi.org/10.1016/j.cma.2020.113580

    work page doi:10.1016/j.cma.2020.113580 2021

  20. [20]

    Mechanics of Materials159, 103918 (2021) https://doi.org/10.1016/j.mechmat

    N. Lange, G. H¨utter, B. Kiefer, Empirical hyper element integration method (EHEIM) with unified integration criteria for efficient hyper-reduced FE2 simulations, Mech. Mater. 210 (2025) 105444. doi: https://doi.org/10.1016/j.mechmat. 2025.105444

    work page doi:10.1016/j.mechmat 2025

  21. [21]

    Linka, M

    K. Linka, M. Hillg”artner, K. P. Abdolazizi, R. C. Aydin, M. Itskov, C. J. Cyron, Constitutive artificial neural networks: A fast and general approach to predictive data-driven constitutive modeling by deep learning, J. Comput. Phys. 429 (2021) 110010

    work page 2021

  22. [22]

    D. K. Klein, R. Ortigosa, J. Mart´ınez-Frutos, O. Weeger, Finite electro-elasticity with physics-augmented neural networks, Comput. Methods Appl. Mech. Eng. 400 (2022) 115501. doi:https://doi.org/10.1016/j.cma.2022.115501

    work page doi:10.1016/j.cma.2022.115501 2022

  23. [23]

    D. K. Klein, F. J. Roth, I. Valizadeh, O. Weeger, Parametrized polyconvex hyperelasticity with physics-augmented neural networks, Data-Centric Eng. 4 (2023) e25. doi:10.1017/dce.2023.21. 18

    work page doi:10.1017/dce.2023.21 2023

  24. [24]

    K., Kalina, K

    L. Linden, D. K. Klein, K. A. Kalina, J. Brummund, O. Weeger, M. K¨astner, Neural networks meet hyperelasticity: A guide to enforcing physics, J. Mech. Phys. Solids 179 (2023) 105363. doi:https://doi.org/10.1016/j.jmps.2023.105363

    work page doi:10.1016/j.jmps.2023.105363 2023

  25. [25]

    Rosenkranz, K

    M. Rosenkranz, K. A. Kalina, J. Brummund, W. Sun, M. K¨astner, Viscoelasticty with physics-augmented neural networks: model formulation and training methods without prescribed internal variables, Comput. Mech. 74 (2024) 1279–1301. doi:10.1007/s00466-024-02477-1

    work page doi:10.1007/s00466-024-02477-1 2024

  26. [26]

    K. A. Kalina, J. Brummund, W. Sun, M. K ¨astner, Neural networks meet anisotropic hyperelasticity: A framework based on generalized structure tensors and isotropic tensor functions, Comput. Methods Appl. Mech. Eng. 437 (2025) 117725. doi:https://doi.org/10.1016/j.cma.2024.117725

    work page doi:10.1016/j.cma.2024.117725 2025

  27. [27]

    B. M. Riemer, J. Brummund, K. A. Kalina, A. H. G. Milor, F. Dammaß, M. K¨astner, Construction of irreducible integrity basis for anisotropic hyperelasticity via structural tensors, 2025. doi:10.48550/arXiv.2512.04014.arXiv:2512.04014

    work page doi:10.48550/arxiv.2512.04014.arxiv:2512.04014 2025

  28. [28]

    K. P. Abdolazizi, R. C. Aydin, C. J. Cyron, K. Linka, Constitutive kolmogorov–arnold networks (CKANs): Combining accuracy and interpretability in data-driven material modeling, 2025. doi:10.48550/arXiv.2502.05682.arXiv:2502.05682

    work page doi:10.48550/arxiv.2502.05682.arxiv:2502.05682 2025

  29. [29]

    Huang, K

    D. Huang, K. C. C. Tse, Y . Lu, C. T. Wu, A machine learning based plasticity model using proper orthogonal decomposition, Comput. Methods Appl. Mech. Eng. 365 (2020) 113008. doi:https://doi.org/10.1016/j.cma.2020.113008

    work page doi:10.1016/j.cma.2020.113008 2020

  30. [30]

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

    M. Raissi, P. Perdikaris, G. Karniadakis, Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, J. Comput. Phys. 378 (2019) 686–707. doi:https://doi.org/10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  31. [31]

    Goswami, M

    S. Goswami, M. Yin, Y . Yu, G. E. Karniadakis, A physics-informed variational deeponet for predicting crack path in quasi-brittle materials, Comput. Methods Appl. Mech. Eng. 391 (2022) 114587. doi:https://doi.org/10.1016/j.cma.2022.114587

    work page doi:10.1016/j.cma.2022.114587 2022

  32. [32]

    M. S. Eshaghi, C. Anitescu, M. Thombre, Y . Wang, X. Zhuang, T. Rabczuk, Variational physics-informed neural operator (VINO) for solving partial differential equations, Comput. Methods Appl. Mech. Eng. 437 (2025) 117785. doi: https: //doi.org/10.1016/j.cma.2025.117785

    work page doi:10.1016/j.cma.2025.117785 2025

  33. [33]

    Rezaei, R

    S. Rezaei, R. N. Asl, S. Faroughi, M. Asgharzadeh, A. Harandi, R. N. Koopas, G. Laschet, S. Reese, M. Apel, A finite operator learning technique for mapping the elastic properties of microstructures to their mechanical deformations, Int. J. for Numer. Methods Eng. 126 (2025) e7637. doi:https://doi.org/10.1002/nme.7637

    work page doi:10.1002/nme.7637 2025

  34. [34]

    Harandi, H

    A. Harandi, H. Danesh, K. Linka, S. Reese, S. Rezaei, Spifol: A spectral-based physics-informed finite operator learning for prediction of mechanical behavior of microstructures, J. Mech. Phys. Solids 203 (2025) 106219. doi: https://doi.org/10. 1016/j.jmps.2025.106219

    work page arXiv 2025

  35. [35]

    Eivazi, J.-A

    H. Eivazi, J.-A. Tr¨oger, S. Wittek, S. Hartmann, A. Rausch, EquiNO: A physics-informed neural operator for multiscale simulations, J. Comput. Phys. 554 (2026) 114745. doi:https://doi.org/10.1016/j.jcp.2026.114745

    work page doi:10.1016/j.jcp.2026.114745 2026

  36. [36]

    M. D. McKay, R. J. Beckman, W. J. Conover, A comparison of three methods for selecting values of input variables in the analysis of output from a computer code, Technometrics 21 (1979) 239–245

    work page 1979

  37. [37]

    D. P. Kingma, J. Ba, Adam: A method for stochastic optimization, in: 3rd International Conference on Learning Representations (ICLR), 2015. ArXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  38. [38]

    D. C. Liu, J. Nocedal, On the limited memory BFGS method for large scale optimization, Math. Program. 45 (1989) 503–528. doi:10.1007/BF01589116

    work page doi:10.1007/bf01589116 1989

  39. [39]

    URLhttps://doi.org/10

    C. Geuzaine, J.-F. Remacle, Gmsh: A 3-d finite element mesh generator with built-in pre- and post-processing facilities, Int. J. for Numer. Methods Eng. 79 (2009) 1309–1331. doi:10.1002/nme.2579

    work page doi:10.1002/nme.2579 2009

  40. [40]

    M. S. Alnæs, J. Blechta, J. Hake, A. Johansson, B. Kehlet, A. Logg, C. Richardson, J. Ring, M. E. Rognes, G. N. Wells, The FEniCS project version 1.5, Arch. Numer. Softw. 3 (2015) 9–23. doi:10.11588/ans.2015.100.20553

    work page doi:10.11588/ans.2015.100.20553 2015

  41. [41]

    A. Logg, G. N. Wells, DOLFIN: Automated finite element computing, ACM Trans. on Math. Softw. 37 (2010) 20:1–20:28. doi:10.1145/1731022.1731030

    work page doi:10.1145/1731022.1731030 2010

  42. [42]

    P. G. Ciarlet, Mathematical Elasticity. V olume I: Three-Dimensional Elasticity, North-Holland, Amsterdam, 1988. 19 A. Supplementary results This appendix collects additional diagnostics that complement the main in-range and out-of-range results. First, we report component-wise learning curves for the first Piola–Kirchhoffstress at the Q-DEIM points in fi...

    work page 1988