arxiv: 2605.00509 · v1 · submitted 2026-05-01 · 💻 cs.CE

Recognition: unknown

An approach to encode divergence-free stress fields in neural approximations based on stress potentials

Mohammad S. Khorrami , Pawan Goyal , Soroush Motahari , David Oexle , Jaber R. Mianroodi , Bob Svendsen , Peter Benner , Dierk Raabe

Authors on Pith no claims yet

Pith reviewed 2026-05-09 18:50 UTC · model grok-4.3

classification 💻 cs.CE

keywords divergence-free stressstress potentialphysics-encoded neural operatorFourier neural operatormechanical equilibriumpolycrystalline materialquasi-static elasticity

0 comments

The pith

A stress potential approach encodes divergence-free conditions directly into neural operator architectures for stress fields.

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

The paper develops a method to incorporate the divergence-free requirement for stress directly into the structure of neural operators by representing stress through a potential function. This physics-encoded Fourier neural operator is applied to stress data from a polycrystalline elastic material under uniaxial extension and compared against physics-guided and physics-informed versions of the same operator. For stress fields that match the data at comparable accuracy, the encoded version satisfies mechanical equilibrium far more closely than the alternatives. The architectural placement means the constraint holds for both training and inference rather than only during optimization.

Core claim

Representing the stress tensor as the divergence of a potential inside the Fourier neural operator architecture directly enforces zero divergence, so that the resulting stress field satisfies quasi-static mechanical equilibrium to a markedly higher degree than is achieved by either physics-guided or physics-informed Fourier neural operators trained on identical data from heterogeneous polycrystalline elasticity.

What carries the argument

Stress-potential representation of the stress tensor, which maps the neural-operator output to a derived stress field whose divergence is identically zero by construction.

If this is right

Both training and model output become physically constrained by construction.
Equilibrium satisfaction improves without any increase in stress-field error relative to the training data.
No penalty terms for the divergence-free condition are needed in the loss function.
The same data set can be used to train all three operator variants for direct comparison.

Where Pith is reading between the lines

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

The potential-encoding pattern could be adapted to other first-order differential constraints such as incompressibility or curl-free fields.
Boundary-condition handling may still require separate treatment even when the interior equilibrium constraint is architecturally satisfied.
The approach may generalize to nonlinear or history-dependent material models if the potential representation can be retained.

Load-bearing premise

Representing stress via a potential function fully and exactly enforces the divergence-free condition in the discrete neural operator setting without introducing approximation errors or requiring additional boundary condition handling.

What would settle it

If the divergence norm of the stress field produced by the trained physics-encoded operator on held-out data is comparable to or larger than the norms obtained from the physics-guided and physics-informed operators at the same stress accuracy level.

Figures

Figures reproduced from arXiv: 2605.00509 by Bob Svendsen, David Oexle, Dierk Raabe, Jaber R. Mianroodi, Mohammad S. Khorrami, Pawan Goyal, Peter Benner, Soroush Motahari.

**Figure 1.** Figure 1: Grain microstructure in U (x1 horizontal, x2 vertical, x3 = 0) with E ∈ [50, 200] GPa, ν ∈ [0.25, 0.35], ndis = 64, and sU = 1 3 ℓU. As discussed in the text, spatially varying material properties mimic an orientation microstructure. As shown, grain interiors are characterized by constant material properties whose values change discontinuously across grain boundaries. Corresponding stress field component d… view at source ↗

**Figure 2.** Figure 2: Data for selected stress field components due to uniaxial extension of the poly view at source ↗

**Figure 3.** Figure 3: Comparison of P dat 22 (upper left) and P out 22 from the PcFNOs. Values in MPa. As evident from a comparison of Figures 1-3, the stress field is relatively uniform in regions of uniform material properties (e.g., grain interiors), and varies in regions in which material properties change in a discontinuous fashion (e.g., across and around grain boundaries and triple junctions), resulting in stress field … view at source ↗

**Figure 4.** Figure 4: Magnitude |P out − P dat| of the error in P out. Values in MPa. As evident, the errors in P out from the PgFNO, PeFNO and PiFNO for cdiv ⩽ 0.01 are quite small in the grain interiors (≈ 1 MPa) and largest at grain boundaries and triple junctions. The errors in d out from the PgFNO and PiFNO for cdiv ⩽ 0.1 as measured by |d out| are analogously distributed, as shown in Figures 5(a,c,d). 18 view at source ↗

**Figure 5.** Figure 5: PcFNO results for the error |d out| in d out = div P out. Values in MPa/ℓU. Note the different scaling in each figure. In contrast, |d out| from the PeFNO ( view at source ↗

**Figure 6.** Figure 6: Comparison of PiFNO results for |P out − P dat| (above; units MPa) and |d out| (below; units MPa/ℓU) for different values of cdiv. Note the different scaling in each figure for |d out|. See text for discussion. Clearly, an increase in the accuracy of d out from the PiFNO to the level of accuracy for d out obtained by the PeFNO is possible (compare Figures 5(b) and 6(f)), but only at the expense of a loss i… view at source ↗

**Figure 7.** Figure 7: Magnitude |P out − P dat| of the error in P out for a finer grain microstructure based on sU = 1 6 ℓU subject to uniaxial extension in the vertical direction with F¯ 22 = 1.004. Values in MPa. Similar to the results in view at source ↗

read the original abstract

The purpose of the current work is the development of an approach to account for quasi-static mechanical equilibrium in empirical (i.e., data-based) models for the stress field employing neural approximations (NAs), which include neural networks (NNs) and neural operators (NOs), in particular Fourier NOs (FNOs). Rather than including such constraints from physics in the loss function as done in the (now standard) physics-informed approach, the current approach incorporates or "encodes" such constraints directly into the architecture of the NA. As a result, both NA training and output are physically constrained in the physics-encoded approach, in contrast to the physics-informed approach, in which only training is physically constrained. For the current constraint of divergence-free stress, a novel encoding approach based on a stress potential is proposed. As a "proof-of-concept" example application of the current approach, a physics-encoded FNO (PeFNO) is developed for a heterogeneous polycrystalline material consisting of isotropic elastic grains and subject to uniaxial extension. Stress field data for this purpose are obtained from the numerical solution of corresponding boundary-value problems for quasi-static mechanical equilibrium. For comparison with the PeFNO, this data is also employed to develop an analogous physics-guided FNO (PgFNO) and physics-informed FNO (PiFNO). As expected theoretically, and confirmed by this computational comparison, for comparable accuracy of the stress field itself as compared to the data, the stress field output by the trained and tested PeFNO is significantly more accurate in satisfying mechanical equilibrium than the output of either the PgFNO or the PiFNO.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper hard-encodes divergence-free stress in FNOs via a learned potential whose derivatives define the stress tensor, and their polycrystalline test shows better equilibrium satisfaction than loss-based baselines at matched stress accuracy.

read the letter

The central idea is to output a scalar or tensor potential from the FNO and recover stress by taking the appropriate second derivatives so that div sigma = 0 holds by construction. They implement this as PeFNO and compare it directly to a physics-guided FNO and a physics-informed FNO on the same set of finite-element stress fields for an isotropic polycrystalline aggregate under uniaxial extension. The results indicate that PeFNO reaches comparable pointwise stress accuracy while producing noticeably smaller equilibrium residuals, which is the practical payoff they highlight. This architecture-level constraint is genuinely different from the usual soft-penalty approaches and removes the need to tune a weighting hyperparameter during training. The comparison is clean because all three models see identical data and the same FNO backbone. That part of the work is solid and worth citing if someone is building constrained operators for solid mechanics. The main soft spot is the discrete setting. FNOs rely on FFT-based differentiation, and the polycrystalline microstructure has sharp property jumps at grain boundaries. Those jumps can produce Gibbs ringing and non-negligible divergence residuals even when the continuous potential is used. The abstract claims the improvement is “as expected theoretically,” but the stress-test concern is fair: without reported residual magnitudes, error bars, or a check against machine precision on the same grid, it is unclear whether the gain is large enough to matter in practice or whether it shrinks once the mesh is refined. The example is also narrow—linear elasticity, simple boundary conditions—so the method’s behavior under plasticity or more complex loading is still open. This is the kind of paper that belongs in a mechanics or scientific-machine-learning venue. It deserves peer review because the encoding trick is concrete and the head-to-head comparison is reproducible on the supplied data. Referees can ask for the actual residual numbers and a clearer statement of how boundary conditions are handled inside the potential formulation.

Referee Report

1 major / 1 minor

Summary. The manuscript develops a physics-encoded approach to enforce divergence-free stress fields in neural approximations (specifically Fourier Neural Operators) by representing stress via a scalar potential function whose second derivatives define the stress tensor components. A PeFNO is constructed and trained on stress data from finite-element solutions of uniaxial extension on polycrystalline microstructures with isotropic elastic grains. It is compared to a physics-guided FNO (PgFNO) and a physics-informed FNO (PiFNO) using the same data; the central result is that, at comparable pointwise stress accuracy, the PeFNO output satisfies mechanical equilibrium (divergence-free condition) to a significantly higher degree.

Significance. If the discrete implementation achieves the claimed improvement, the architecture-level encoding of equilibrium via potentials offers a stronger guarantee than soft penalties in the loss and could improve reliability of data-driven stress predictions in heterogeneous media. The use of polycrystalline data and direct comparison among three FNO variants provides a concrete test of the idea.

major comments (1)

[Numerical results / implementation details] The central claim rests on the assertion that deriving stress from a learned potential enforces div σ = 0 exactly (or to machine precision) even after discretization. In the FNO setting the potential is represented on a discrete grid and second derivatives are obtained via spectral or finite-difference operators; on polycrystalline grids these operators are inexact at grain boundaries. The manuscript should therefore report the actual L2 or max-norm divergence residuals for the PeFNO outputs (with error bars across test microstructures) and compare them quantitatively to the PgFNO/PiFNO residuals to substantiate the “significantly more accurate” statement.

minor comments (1)

[Abstract] The abstract states that stress-field accuracy is “comparable” and equilibrium satisfaction is “significantly” better, but does not specify the quantitative thresholds, norms, or number of test cases used; adding these numbers would improve clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments and the opportunity to clarify and strengthen our manuscript. We address the single major comment below and will revise the manuscript to incorporate the requested quantitative details.

read point-by-point responses

Referee: The central claim rests on the assertion that deriving stress from a learned potential enforces div σ = 0 exactly (or to machine precision) even after discretization. In the FNO setting the potential is represented on a discrete grid and second derivatives are obtained via spectral or finite-difference operators; on polycrystalline grids these operators are inexact at grain boundaries. The manuscript should therefore report the actual L2 or max-norm divergence residuals for the PeFNO outputs (with error bars across test microstructures) and compare them quantitatively to the PgFNO/PiFNO residuals to substantiate the “significantly more accurate” statement.

Authors: We agree that explicit quantitative reporting of the divergence residuals is necessary to substantiate the central claim. In the continuous setting the stress derived from the potential satisfies ∇·σ = 0 identically by vector-calculus identities. In the discrete FNO implementation, spectral differentiation on a uniform grid yields small but nonzero residuals, particularly near grain boundaries where elastic properties are discontinuous. Nevertheless, because the same differentiation operators are used both to generate the training data and to recover stress from the learned potential, the architecture-level encoding still produces substantially smaller residuals than the soft-constraint approaches in PgFNO and PiFNO. In the revised manuscript we will add a new table (or supplementary figure) that reports, for the test set of microstructures, the mean and standard deviation (error bars) of both the L2-norm and the maximum-norm of div σ for all three models. These metrics have already been computed from our existing trained networks and confirm the claimed improvement at comparable pointwise stress accuracy. revision: yes

Circularity Check

0 steps flagged

No significant circularity: encoding via stress potential is an independent architectural choice

full rationale

The paper's central derivation introduces a stress-potential representation (PeFNO) to hard-encode the divergence-free condition directly into the neural operator architecture. This rests on the standard continuum-mechanics identity that the divergence of a stress tensor derived from a twice-differentiable potential vanishes identically; the identity is external to the neural training and is not redefined or fitted within the paper. The subsequent empirical comparison (PeFNO vs. PgFNO/PiFNO) on polycrystalline data is a performance measurement, not a self-referential prediction. No load-bearing step reduces by construction to a fitted parameter, a self-citation chain, or a renamed input; the architecture choice and the reported accuracy gains remain independent of the target equilibrium metric.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

Relies on standard continuum mechanics and vector calculus; introduces new encoding entity without independent evidence beyond the proposed method.

axioms (2)

domain assumption Quasi-static mechanical equilibrium implies divergence-free stress tensor in the absence of body forces.
Standard assumption in continuum solid mechanics invoked for the constraint.
domain assumption A stress potential exists such that the stress tensor derived from it is automatically divergence-free.
Based on tensor calculus identities; central to the encoding approach.

invented entities (1)

Physics-encoded FNO (PeFNO) via stress potential no independent evidence
purpose: Neural operator architecture that enforces divergence-free stress by construction
New proposed architecture; no independent falsifiable evidence outside the paper's comparison.

pith-pipeline@v0.9.0 · 5632 in / 1288 out tokens · 43061 ms · 2026-05-09T18:50:34.532180+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

58 extracted references · 1 canonical work pages · 1 internal anchor

[1]

Holthusen and E

H. Holthusen and E. Kuhl , date-added =. A complement to neural networks for anisotropic inelasticity at finite strains , volume =. Computer Methods in Applied Mechanics and Engineering , pages =
[2]

Holthusen and L

H. Holthusen and L. Lamma and T. Brepols and S. Reese and E. Kuhl , date-added =. Theory and implementation of inelastic constitutive artificial neural networks , volume =. Computer Methods in Applied Mechanics and Engineering , pages =
[3]

van der Heijden and X

B. van der Heijden and X. Li and G. Lubineau and E. Florentin , date-added =. Enforcing physics onto PINNs for more accurate inhomogeneous material identification , volume =. Computer Methods in Applied Mechanics and Engineering , pages =
[4]

J. N. Fuhg and G. A. Padmanabha and N. Bouklas and B. Bahmani and W. C. Sun and N. N. Vlassis and M. Flaschel and P. Carrara and L. A review on data-driven constitutive laws for solids , volume =. Archives of Computational Methods in Engineering , pages =
[5]

Riemann tensor neural networks: learning conservative systems with physics-constrained networks , volume =

Anas Jnini and Lorenzo Breschi and Flavio Vella , date-added =. Riemann tensor neural networks: learning conservative systems with physics-constrained networks , volume =. arXiv preprint , pages =
[6]

D. A. Kopriva , date-added =. Implementing Spectral Methods for Partial Differential Equations , year =
[7]

Teodosiu , date-added =

C. Teodosiu , date-added =. Elastic Models of Crystal Defects , year =
[8]

M. H. Sadd , date-added =. Elasticity: Theory, Applications, and Numerics , year =
[9]

The Mechanics and Thermodynamics of Continuous Media , year =

M. The Mechanics and Thermodynamics of Continuous Media , year =
[10]

Truesdell and W

C. Truesdell and W. Noll , date-added =. The Non-Linear Field Theories of Mechanics , volume =
[11]

Abraham and J

R. Abraham and J. E. Marsden and T. Ratiu , date-added =. Manifolds, Tensor Analysis and Applications , volume =
[12]

Chadwick , date-added =

P. Chadwick , date-added =. Continuum Mechanics: Concise Theory and Problems , year =
[13]

Bhatia and G

H. Bhatia and G. Norgard and P.-T. Bremer , date-added =. The. IEEE Transactions on Visualization and Computer Graphics , pages =
[14]

Svendsen , date-added =

B. Svendsen , date-added =. On the representation of constitutive relations using structure tensors , volume =. International Journal of Engineering Science , pages =
[15]

Applied Mechanics Reviews , pages =

Theory of representations for tensor functions -- A unified invariant approach to constitutive equations , volume =. Applied Mechanics Reviews , pages =
[16]

and Hotho, A

Dulny, A. and Hotho, A. and Krause, A. , booktitle =
[17]

Linka and E

K. Linka and E. Kuhl , date-added =. A new family of constitutive artificial neural networks towards automated model discovery , volume =. Computer Methods in Applied Mechanics and Engineering , pages =
[19]

and Li, Z

Kovachki, N. and Li, Z. and Liu, B. and Azizzadenesheli, K. and Bhattacharya, K. and Stuart, A. and Anandkumar, A. , date-added =. Neural operators: learning maps between function spaces with application to. Journal of Machine Learning Research , pages =
[20]

and Pawar, Nikhil M

Faroughi, Salah A. and Pawar, Nikhil M. and Fernandes, C. Physics-guided, physics-informed, and physics-encoded neural networks and operators in scientific computing: fluid and solid mechanics , volume =. Journal of Computing and Information Science in Engineering , pages =
[21]

and Pengzhan, J

Lu, L. and Pengzhan, J. and Pang, G. and Zhang, Z. and Karniadakis, G. E. , date-added =. Learning nonlinear operators via. Natural and Machine Intelligence , pages =
[22]

and Jin, P

Lu, L. and Jin, P. and Karniadakis, G. E. , date-added =. arXiv preprint , title =
[23]

Richter-Powell and Y

J. Richter-Powell and Y. Lipman and R. T. Q. Chen , date-added =. Neural conservation laws: a divergence-free perspective , volume =. arXiv preprint , pages =
[24]

Physics-informed neural networks (

Cai, Shengze and Mao, Zhiping and Wang, Zhicheng and Yin, Minglang and Karniadakis, George Em , date-modified =. Physics-informed neural networks (. Acta Mechanica Sinica , number =
[25]

Scientific machine learning through physics--informed neural networks: Where we are and what's next , volume =

Cuomo, Salvatore and Di Cola, Vincenzo Schiano and Giampaolo, Fabio and Rozza, Gianluigi and Raissi, Maziar and Piccialli, Francesco , journal =. Scientific machine learning through physics--informed neural networks: Where we are and what's next , volume =
[26]

Soft Matter , number =

Mahmoudabadbozchelou, Mohammadamin and Karniadakis, George Em and Jamali, Safa , date-modified =. Soft Matter , number =
[27]

Physics-informed neural networks for solving Reynolds-averaged Navier--Stokes equations , volume =

Eivazi, Hamidreza and Tahani, Mojtaba and Schlatter, Philipp and Vinuesa, Ricardo , journal =. Physics-informed neural networks for solving Reynolds-averaged Navier--Stokes equations , volume =
[28]

Physics-informed neural networks for heat transfer problems , volume =

Cai, Shengze and Wang, Zhicheng and Wang, Sifan and Perdikaris, Paris and Karniadakis, George Em , journal =. Physics-informed neural networks for heat transfer problems , volume =
[29]

Physics-informed neural networks for studying heat transfer in porous media , volume =

Xu, Jiaxuan and Wei, Han and Bao, Hua , journal =. Physics-informed neural networks for studying heat transfer in porous media , volume =
[30]

Solving inverse heat transfer problems without surrogate models: a fast, data-sparse, physics-informed neural network approach , volume =

Oommen, Vivek and Srinivasan, Balaji , date-modified =. Solving inverse heat transfer problems without surrogate models: a fast, data-sparse, physics-informed neural network approach , volume =. Journal of Computing and Information Science in Engineering , number =
[31]

A physics-informed neural network technique based on a modified loss function for computational

Bai, Jinshuai and Rabczuk, Timon and Gupta, Ashish and Alzubaidi, Laith and Gu, Yuantong , date-modified =. A physics-informed neural network technique based on a modified loss function for computational. Computational Mechanics , number =
[32]

Meshless physics-informed deep learning method for three-dimensional solid mechanics , volume =

Abueidda, Diab W and Lu, Qiyue and Koric, Seid , journal =. Meshless physics-informed deep learning method for three-dimensional solid mechanics , volume =
[33]

Solving multi-material problems in solid mechanics using physics-informed neural networks based on domain decomposition technology , volume =

Diao, Yu and Yang, Jianchuan and Zhang, Ying and Zhang, Dawei and Du, Yiming , journal =. Solving multi-material problems in solid mechanics using physics-informed neural networks based on domain decomposition technology , volume =
[34]

Physics-informed machine learning , volume =

Karniadakis, George Em and Kevrekidis, Ioannis G and Lu, Lu and Perdikaris, Paris and Wang, Sifan and Yang, Liu , journal =. Physics-informed machine learning , volume =
[35]

A novel meta-learning initialization method for physics-informed neural networks , volume =

Liu, Xu and Zhang, Xiaoya and Peng, Wei and Zhou, Weien and Yao, Wen , journal =. A novel meta-learning initialization method for physics-informed neural networks , volume =
[36]

Turbulence Modeling for Physics-Informed Neural Networks: Comparison of Different RANS Models for the Backward-Facing Step Flow , volume =

Pioch, Fabian and Harmening, Jan Hauke and M. Turbulence Modeling for Physics-Informed Neural Networks: Comparison of Different RANS Models for the Backward-Facing Step Flow , volume =. Fluids , number =
[37]

Transfer learning enhanced physics informed neural network for phase-field modeling of fracture , volume =

Goswami, Somdatta and Anitescu, Cosmin and Chakraborty, Souvik and Rabczuk, Timon , journal =. Transfer learning enhanced physics informed neural network for phase-field modeling of fracture , volume =
[38]

Physics-informed machine learning model for computational fracture of quasi-brittle materials without labelled data , volume =

Zheng, Bin and Li, Tongchun and Qi, Huijun and Gao, Lingang and Liu, Xiaoqing and Yuan, Li , journal =. Physics-informed machine learning model for computational fracture of quasi-brittle materials without labelled data , volume =
[39]

Physics-informed surrogate modeling for hydro-fracture geometry prediction based on deep learning , volume =

Lu, Yutian and Wang, Bo and Zhao, Yingying and Yang, Xiaochen and Li, Lizhe and Dong, Mingzhi and Lv, Qin and Zhou, Fujian and Gu, Ning and Shang, Li , journal =. Physics-informed surrogate modeling for hydro-fracture geometry prediction based on deep learning , volume =
[40]

Fracture pattern prediction with random microstructure using a physics-informed deep neural networks , volume =

Wei, Haoyang and Yao, Houpu and Pang, Yutian and Liu, Yongming , journal =. Fracture pattern prediction with random microstructure using a physics-informed deep neural networks , volume =
[41]

A physics-informed variational

Goswami, Somdatta and Yin, Minglang and Yu, Yue and Karniadakis, George Em , date-modified =. A physics-informed variational. Computer Methods in Applied Mechanics and Engineering , pages =
[42]

Physics-informed neural networks for phase-field method in two-phase flow , volume =

Qiu, Rundi and Huang, Renfang and Xiao, Yao and Wang, Jingzhu and Zhang, Zhen and Yue, Jieshun and Zeng, Zhong and Wang, Yiwei , journal =. Physics-informed neural networks for phase-field method in two-phase flow , volume =
[43]

Inferring incompressible two-phase flow fields from the interface motion using physics-informed neural networks , volume =

Buhendwa, Aaron B and Adami, Stefan and Adams, Nikolaus A , journal =. Inferring incompressible two-phase flow fields from the interface motion using physics-informed neural networks , volume =
[44]

What machine learning can do for computational solid mechanics , year =

Kumar, Siddhant and Kochmann, Dennis M , booktitle =. What machine learning can do for computational solid mechanics , year =
[45]

Machine learning for solid mechanics , year =

Yang, Charles and Zhang, Zhizhou and Gu, Grace X , booktitle =. Machine learning for solid mechanics , year =
[46]

Siboni, Nima and Raabe, Dierk , date-modified =

Mianroodi, Jaber Rezaei and H. Siboni, Nima and Raabe, Dierk , date-modified =. Teaching solid mechanics to artificial intelligence -- a fast solver for heterogeneous materials , volume =. Npj Computational Materials , number =
[47]

An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials , volume =

Khorrami, Mohammad S and Mianroodi, Jaber R and Siboni, Nima H and Goyal, Pawan and Svendsen, Bob and Benner, Peter and Raabe, Dierk , date-modified =. An artificial neural network for surrogate modeling of stress fields in viscoplastic polycrystalline materials , volume =. npj Computational Materials , pages =
[48]

Computational homogenization using convolutional neural networks , year =

Wessels, Henning and B. Computational homogenization using convolutional neural networks , year =. Current Trends and Open Problems in Computational Mechanics , pages =
[49]

PhyCRNet: Physics-informed convolutional-recurrent network for solving spatio-temporal

Ren, Pu and Rao, Chengping and Liu, Yang and Wang, Jian-Xun and Sun, Hao , date-modified =. PhyCRNet: Physics-informed convolutional-recurrent network for solving spatio-temporal. Computer Methods in Applied Mechanics and Engineering , pages =
[50]

Fourier Neural Operator for Parametric Partial Differential Equations

Li, Zongyi and Kovachki, Nikola and Azizzadenesheli, Kamyar and Liu, Burigede and Bhattacharya, Kaushik and Stuart, Andrew and Anandkumar, Anima , date-modified =. arXiv preprint arXiv:2010.08895v3 , title =

work page internal anchor Pith review arXiv 2010
[51]

Physics-informed deep neural operator networks , year =

Goswami, Somdatta and Bora, Aniruddha and Yu, Yue and Karniadakis, George Em , booktitle =. Physics-informed deep neural operator networks , year =
[52]

Learning the stress-strain fields in digital composites using a

Rashid, Meer Mehran and Pittie, Tanu and Chakraborty, Souvik and Krishnan, NM Anoop , date-modified =. Learning the stress-strain fields in digital composites using a. Iscience , number =
[53]

Comparison of two artificial neural networks trained for the surrogate modeling of stress in materially heterogeneous elastoplastic solids , year =

Kapoor, Sarthak and Mianroodi, Jaber Rezaei and Khorrami, Mohammad and Siboni, Nima S and Svendsen, Bob , journal =. Comparison of two artificial neural networks trained for the surrogate modeling of stress in materially heterogeneous elastoplastic solids , year =
[54]

Numerically robust spectral methods for crystal plasticity simulations of heterogeneous materials , volume =

Shanthraj, Pratheek and Eisenlohr, Philip and Diehl, Martin and Roters, Franz , journal =. Numerically robust spectral methods for crystal plasticity simulations of heterogeneous materials , volume =
[55]

Fourier-based schemes for computing the mechanical response of composites with accurate local fields , volume =

Willot, Fran. Fourier-based schemes for computing the mechanical response of composites with accurate local fields , volume =. Comptes Rendus M
[56]

DBFFT: A displacement based FFT approach for non-linear homogenization of the mechanical behavior , volume =

Lucarini, Sergio and Segurado, Javier , journal =. DBFFT: A displacement based FFT approach for non-linear homogenization of the mechanical behavior , volume =
[57]

Development and comparison of spectral algorithms for numerical modeling of the quasi-static mechanical behavior of inhomogeneous materials , year =

Khorrami, Mohammad and Mianroodi, Jaber Rezaei and Shanthraj, Pratheek and Svendsen, Bob , journal =. Development and comparison of spectral algorithms for numerical modeling of the quasi-static mechanical behavior of inhomogeneous materials , year =
[58]

Roters, Franz and Diehl, Martin and Shanthraj, Pratheek and Eisenlohr, Philip and Reuber, C and Wong, Su Leen and Maiti, Tias and Ebrahimi, Alireza and Hochrainer, Thomas and Fabritius, H-O and others , journal =
[59]

and Jones, D

Sorensen, H. and Jones, D. and Heideman, M. and Burrus, C. , date-modified =. Real-valued fast. IEEE Transactions on Acoustics, Speech, and Signal Processing , keywords =