Pith. sign in

REVIEW 4 major objections 5 minor 46 references

PINN-FEM: A Hybrid Approach for Enforcing Dirichlet Boundary Conditions in Physics-Informed Neural Networks

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Hybrid PINN-FEM enforces Dirichlet boundary conditions exactly with a finite-element boundary layer and a neural-network interior.

desk verdict Confounded optimizer comparison and ambiguous 2D ansatz undercut an otherwise practical FE-layer twist on distance-function BC enforcement. read the letter →

arxiv 2501.07765 v1 pith:XX5DUZ2T submitted 2025-01-14 cs.LG physics.comp-phstat.ML

Nahil Sobh , Rini Jasmine Gladstone , Hadi Meidani This is my paper · ORCID
classification cs.LGphysics.comp-phstat.ML MSC 65N3068T0774S05
keywords physics-informedneuralnetworksfiniteelementmethodDirichletboundaryconditionsdomaindecompositionenergy-basedlosslinearelasticityconditionenforcement
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 proposes a hybrid method, PINN-FEM, for solving boundary value problems in solid mechanics: a finite element strip is placed along Dirichlet boundaries and a neural network approximates the solution in the interior, with the two fields meeting at an interface. The trial field is trained by minimizing total potential energy, so the Dirichlet data enter as exact constraints through FEM shape functions instead of as penalty terms. The authors claim this gives exact boundary condition enforcement without the convergence problems of soft PINN losses, and they support it with six linear elasticity experiments of increasing complexity (smooth, holed, discontinuous, point, cracked, and cantilever domains). Across those experiments the relative $\ell^1$ displacement errors are consistently lower than standard PINN baselines, including PINNs with approximate distance functions.

What carries the argument

The central object is the decomposed trial space: the domain $\Omega$ is split into an FEM region $\Omega_{FE}$ adjacent to the Dirichlet boundary and a neural network region $\Omega_{NN}$, with interface $\Gamma_I$. In the FEM strip, linear shape functions carry the prescribed boundary value $g$ into the trial field, so the Dirichlet data are enforced strongly; in the interior, the neural network $N_\theta(x)$ is free. The carrying identity is the energy functional $E(u)=\int_\Omega (\tfrac12 \varepsilon : C : \varepsilon - f\cdot u)\,d\Omega - \int_{\Gamma_h} h\cdot u\,d\Gamma$, minimized over the hybrid field, which automatically absorbs Neumann conditions and makes the loss a scalar potential energy rather than a residual sum.

What would settle it

Compute the value of the hybrid trial field on both sides of the interface $\Gamma_I$ for a problem with nonzero prescribed displacement $g$ and compare: if $u|_{\Omega_{FE}} - u|_{\Omega_{NN}} \neq 0$ at the interface, the trial space is discontinuous and the reported errors do not correspond to a solution of the original boundary value problem. A direct check would be a two-dimensional plate with constant nonzero $g$ and a single-element FEM strip, evaluating the field at $\Gamma_I$ and at $\Gamma_g$.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that domain decomposition with finite elements at the boundary converts the Dirichlet condition from a soft loss term into a strong constraint, while the interior stays mesh-free and flexible. The hybrid displacement field is $u=N_\theta(x)N_D(x)+g$ in the FEM strip and $u=N_\theta(x)$ in the neural network region, with an energy-based loss derived from the principle of minimum potential energy. This is claimed to be exact at the Dirichlet boundary, automatically consistent with Neumann conditions through the variational formulation, and general enough to handle discontinuous and point boundary conditions where global distance-function constructions fail. The reported numbers show improvements such as a reduction in relative error for the crack problem from $0.07$ (ADF) to $0.05$ for $u_x$, and for the cantilever beam from $0.56$ (soft) to $0.0009$ for $u_x$.

Load-bearing premise

The two-dimensional trial field $u=N_\theta N_D+g$ presupposes a shape-function field $N_D$ that is $0$ on the Dirichlet boundary and $1$ at the interface with the neural network region; the paper does not define such a field for a multi-element strip or verify continuity of the hybrid field at the interface, so if this construction fails the energy is minimized over discontinuous trial functions.

Editorial extensions

If this is right

  • The Dirichlet condition becomes a strong constraint built into the trial field, so no boundary-loss weight $\beta$ needs tuning in the energy loss.
  • Neumann conditions are absorbed by the variational formulation, so the loss function needs no separate traction residual.
  • Discontinuous and point Dirichlet data, which break global distance-function constructions, still fit inside a conforming local FEM mesh.
  • On smooth problems the method can reach near-analytical accuracy, as in the cantilever case with relative errors around $10^{-4}$.

Reading between the lines

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

  • A natural extension the authors leave implicit: the strip construction could be applied to time-dependent boundary data by making the FE layer track a moving Dirichlet boundary or by using space-time shape functions.
  • The same energy-loss argument suggests the method should extend to finite-deformation hyperelasticity, where a potential energy is still well defined, as long as the interface continuity condition is built into the ansatz.
  • One testable extension: using full quadrature in the FEM strip rather than a single Gauss point should improve accuracy where strain varies strongly inside the strip.
  • The hybrid ansatz could also serve as a mesh-conforming initializer for standard FEM solvers, since it supplies a field that already matches the boundary-layer geometry.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes PINN-FEM, a hybrid method that enforces Dirichlet boundary conditions exactly by decomposing the computational domain into a finite element layer adjacent to the boundary and a neural network interior, and trains the network by minimizing a potential-energy loss. The method is described for 1D, 2D, and 3D, and evaluated on six 2D linear elasticity problems against soft-boundary (Soft), approximate-distance-function (ADF), and distance-function (DF) PINN baselines. The authors report that PINN-FEM consistently outperforms these baselines in relative ℓ1 displacement error, and they highlight robustness on problems with discontinuous, point, and crack-like boundary conditions.

Significance. If substantiated, PINN-FEM would provide a practical mechanism for strong Dirichlet enforcement in energy-based PINNs, including for boundary conditions that distance-function methods cannot represent. The paper draws on established blending ideas from mesh-free methods and on the deep energy method, so the conceptual novelty is moderate; the main value would be a demonstrated, reproducible robustness advantage across several boundary geometries. However, the current experimental design changes the optimizer together with the boundary-enforcement method, and the 2D trial-space definition is incomplete, so the reported superiority is not yet established. The method's energy-based loss and exact boundary handling are attractive, but the evidence as presented is insufficient for a strong claim.

major comments (4)
  1. [§4.3.1 / Table 2] Section 4.3.1 states that PINN-FEM is trained with LBFGS while all baselines (Soft, ADF, DF) are trained with Adam, and this same optimizer assignment is reused in every subsequent experiment (Sections 4.3.2–4.3.6). Table 2 therefore varies two factors simultaneously: the boundary-condition strategy and the optimizer. Because LBFGS is typically much more effective than Adam on small over-parameterized energy-minimization problems, the large reported gaps (for example, e(uy)=0.0003 for PINN-FEM versus 0.56 for Soft in the cantilever experiment) cannot be attributed to the FEM-layer boundary enforcement. A controlled comparison with identical optimizers, or at least a cross of optimizer and method, is required to support the paper's central claim.
  2. [§3.2.1, Eqs. (16)–(20)] The discretized 1D energy in Eqs. (16)–(20) omits the body-force term −∫ f u dx that is present in the continuous energy of Eq. (13), and the traction work is written as hNθ,x|0 instead of hNθ(0). As a result, minimizing Eq. (21) does not correspond to minimizing the potential energy of the boundary value problem (10)–(12). The same body-force omission appears in the both-ends version, Eqs. (27)–(32). These equations need to be corrected and the subsequent energy balance re-derived; as written, the 1D formulation is not a valid variational statement of the model problem.
  3. [§3.3, Eq. (33)] The 2D trial field u = Nθ(x)ND(x) + g in ΩFE is not a well-defined finite element interpolation as written. If ND is a scalar blending function equal to 1 on the interface ΓI and 0 on Γg, then on ΓI the field becomes Nθ + g, which is discontinuous with the interior field Nθ unless g = 0. If ND instead denotes a vector of element shape functions, the product form does not match standard nodal interpolation, and the extension of the prescribed boundary data g into the interior of ΩFE is not specified. The paper must define ND precisely (for example, as a piecewise-linear function with ND=1 on ΓI and ND=0 on Γg, or by replacing the ansatz with u = Σ_i Nθ(x_i)N_i(x) + Σ_{j∈Γg} g_jN_j(x)) and demonstrate that the resulting trial space is a subset of H^1. Without this, the 2D energy minimization is performed over an ill-defined trial space.
  4. [§3.3, Eq. (37)] The energy in the FEM region is evaluated by one-point quadrature at a single centroid over the whole ΩFE region. Since ΩFE contains many triangular elements and the strain field varies between elements, this is not the standard per-element constant-strain quadrature used in linear FEM. The approximation error introduced by collapsing the entire FE layer to one centroid should be quantified, or the energy should be assembled element-wise; otherwise the reported accuracy of the 2D method may reflect this additional quadrature approximation rather than the proposed hybrid construction.
minor comments (5)
  1. [§4.1 / §4.3.1] Section 4.1 says the collocation points are the centroids of the triangles, while Sections 4.3.1 and 4.3.4 say the collocation points are the nodal points of the mesh; please clarify which is actually used.
  2. [§3.2.1, Eq. (17); §3.2.2, Eq. (28)] Eq. (17) contains the typo 'gR2 NR2,x' and Eq. (28) contains 'gL2' and 'gR2' where the prescribed boundary values g, gL, and gR should appear; these likely are transcription errors from the shape-function indices.
  3. [§4.2, Eq. (47)] The soft-boundary weight β is introduced but its value and tuning procedure are never reported; please state the value used for all soft baseline runs.
  4. [Table 2] Several rows in Table 2 are missing entries for ADF and DF baselines; the text explains the reasons, but a footnote in the table itself would make the comparison clearer at a glance.
  5. [References] Reference [46] should be 'Timoshenko and Goodier' rather than 'Goodyear'; please correct.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: PINN-FEM is a direct variational method and its accuracy claims are checked against external FEM/analytic references.

full rationale

The paper's derivation chain is self-contained. The trial space in Eqs. (14), (25), and (33) encodes the Dirichlet data by construction, and the loss in Eqs. (21), (32), and (39) is the exact potential energy functional (9) restricted to that trial space; minimizing it is the standard Ritz/energy method, not a fitting of any parameter to the target output. No parameter is trained against the FEM/Abaqus or Timoshenko reference solutions; those references enter only in the relative-error metric (49). There are no self-citations by the authors, and the cited FEM-meshless blending literature [31-36] is independent prior work used as motivation, not as a uniqueness theorem. The concerns raised by the reader and skeptic are real but are not circularity: the 2D ansatz in Eq. (33) does not specify ND or verify continuity at the interface, and the optimizer (LBFGS vs Adam) is confounded with the method in Table 2, weakening the empirical comparison. Under-specification and experimental confounding affect validity, not whether the result reduces by definition to its inputs. Hence score 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The method introduces no new physical entities. It relies on the standard potential-energy principle and on an unproven assumption about the shape-function field in the FE region. The only hand-chosen numerical parameter is the mesh size of the boundary layer.

free parameters (1)
  • FE layer mesh size = 0.1 mm
    The thickness of the boundary layer where FEM shape functions are used is chosen as 0.1 mm for all experiments. The paper does not study how this choice affects accuracy or convergence.
assumptions (3)
  • domain assumption The minimizer of the total potential energy functional (Eq. 9) satisfies the equilibrium equation and natural boundary conditions (principle of minimum potential energy).
    Invoked in Section 2.2 and used to define the loss function for all models. Standard in linear elasticity.
  • ad hoc to paper The FEM shape functions ND are exactly zero on the Dirichlet boundary and exactly one at the interface, making the hybrid trial function conforming.
    Assumed in Eq. (33) and Section 3.3. The paper does not construct such ND or verify the continuity condition at the interface.
  • ad hoc to paper The strain energy in the FE region can be accurately approximated by evaluating the strain at the element centroid and multiplying by the region area (one-point quadrature).
    Used in Eq. (37) and Eq. (40). Exact only for constant-strain elements under a standard FEM interpolation; the paper's product ansatz does not have constant strain.

how reviews work

0 comments
Cite this review

Pith. "Pith review of PINN-FEM: A Hybrid Approach for Enforcing Dirichlet Boundary Conditions in Physics-Informed Neural Networks." pith.science (2026). https://pith.science/paper/XX5DUZ2T

@misc{pith2026250107765,
  author       = {Pith},
  title        = {Pith review of: PINN-FEM: A Hybrid Approach for Enforcing Dirichlet Boundary Conditions in Physics-Informed Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XX5DUZ2T}},
  note         = {Machine review of arXiv:2501.07765}
}
read the original abstract

Physics-Informed Neural Networks (PINNs) solve partial differential equations (PDEs) by embedding governing equations and boundary/initial conditions into the loss function. However, enforcing Dirichlet boundary conditions accurately remains challenging, often leading to soft enforcement that compromises convergence and reliability in complex domains. We propose a hybrid approach, PINN-FEM, which combines PINNs with finite element methods (FEM) to impose strong Dirichlet boundary conditions via domain decomposition. This method incorporates FEM-based representations near the boundary, ensuring exact enforcement without compromising convergence. Through six experiments of increasing complexity, PINN-FEM outperforms standard PINN models, showcasing superior accuracy and robustness. While distance functions and similar techniques have been proposed for boundary condition enforcement, they lack generality for real-world applications. PINN-FEM bridges this gap by leveraging FEM near boundaries, making it well-suited for industrial and scientific problems.

Figures

Figures reproduced from arXiv: 2501.07765 by the authors.

Figure 1
Figure 1. Domain Decomposition with complementary Neumann and Dirichlet boundary conditions at the two ends for [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Domain Decomposition with Dirichlet boundary conditions at both the ends for a one-dimensional boundary [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 2
Figure 2. Similarly, the displacement gradients are given by: [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figures from the paper (1 more)
Figure 3
Figure 3. Figure 3: Domain decomposition for the proposed PINN-FEM method for a two-dimensional boundary value problem. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 33 canonical work pages

  1. [1]

    Machine learning and big scientific data

    Tony Hey, Keith Butler, Sam Jackson, and Jeyarajan Thiyagalingam. Machine learning and big scientific data. Philosophical Transactions of the Royal Society A, 378(2166):20190054, 2020

    work page 2020

  2. [2]

    Machine learning for science: state of the art and future prospects.science, 293(5537):2051–2055, 2001

    Eric Mjolsness and Dennis DeCoste. Machine learning for science: state of the art and future prospects.science, 293(5537):2051–2055, 2001

    work page 2001

  3. [3]

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

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics- informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021. 18 A preprint - January 15, 2025

    work page 2021

  4. [4]

    Learning mesh-based simulation with graph networks.arXiv preprint arXiv:2010.03409, 2020

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W Battaglia. Learning mesh-based simulation with graph networks.arXiv preprint arXiv:2010.03409, 2020

    arXiv 2010

  5. [5]

    Learningnonlinearoperators viadeeponetbasedontheuniversalapproximationtheoremofoperators

    LuLu, PengzhanJin, GuofeiPang, ZhongqiangZhang, andGeorgeEmKarniadakis. Learningnonlinearoperators viadeeponetbasedontheuniversalapproximationtheoremofoperators. Naturemachineintelligence,3(3):218–229, 2021

    work page 2021

  6. [6]

    Fourier neural operator for parametric partial differential equations.arXiv preprint arXiv:2010.08895, 2020

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations.arXiv preprint arXiv:2010.08895, 2020

    arXiv 2010

  7. [7]

    Solving the wave equation with physics-informed deep learning

    Ben Moseley, Andrew Markham, and Tarje Nissen-Meyer. Solving the wave equation with physics-informed deep learning. arXiv preprint arXiv:2006.11894, 2020

    arXiv 2006

  8. [8]

    Physics-informedneural networks (pinns) for fluid mechanics: A review.Acta Mechanica Sinica, 37(12):1727–1738, 2021

    ShengzeCai, ZhipingMao, ZhichengWang, MinglangYin, andGeorgeEmKarniadakis. Physics-informedneural networks (pinns) for fluid mechanics: A review.Acta Mechanica Sinica, 37(12):1727–1738, 2021

    work page 2021

Show all 46 references
  1. [9]

    Nsfnets(navier-stokesflownets): Physics-informed neural networks for the incompressible navier-stokes equations.Journal of Computational Physics, 426:109951, 2021

    XiaoweiJin,ShengzeCai,HuiLi,andGeorgeEmKarniadakis. Nsfnets(navier-stokesflownets): Physics-informed neural networks for the incompressible navier-stokes equations.Journal of Computational Physics, 426:109951, 2021

    2021

  2. [10]

    Physics-informed neural networks for power systems

    George S Misyris, Andreas Venzke, and Spyros Chatzivasileiadis. Physics-informed neural networks for power systems. In2020 IEEE power & energy society general meeting (PESGM), pages 1–5. IEEE, 2020

    2020

  3. [11]

    Predictionofporousmediafluidflowusingphysicsinformed neural networks.Journal of Petroleum Science and Engineering, 208:109205, 2022

    MuhammadMAlmajidandMoatazOAbu-Al-Saud. Predictionofporousmediafluidflowusingphysicsinformed neural networks.Journal of Petroleum Science and Engineering, 208:109205, 2022

    2022

  4. [12]

    Deep learning of two-phase flow in porous media via theory-guided neural networks.SPE Journal, 27(02):1176–1194, 2022

    Jian Li, Dongxiao Zhang, Nanzhe Wang, and Haibin Chang. Deep learning of two-phase flow in porous media via theory-guided neural networks.SPE Journal, 27(02):1176–1194, 2022

    2022

  5. [13]

    A unified deep artificial neural network approach to partial differential equations in complex geometries.Neurocomputing, 317:28–41, 2018

    Jens Berg and Kaj Nyström. A unified deep artificial neural network approach to partial differential equations in complex geometries.Neurocomputing, 317:28–41, 2018

    2018

  6. [14]

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational physics, 378:686–707, 2019

    2019

  7. [15]

    Dgm: A deep learning algorithm for solving partial differential equations

    Justin Sirignano and Konstantinos Spiliopoulos. Dgm: A deep learning algorithm for solving partial differential equations. Journal of computational physics, 375:1339–1364, 2018

    2018

  8. [16]

    The deep ritz method: A deep learning-based numerical algorithm for solving variational problems

    Yu B EW. The deep ritz method: A deep learning-based numerical algorithm for solving variational problems. Commun Math Stat, 6(1):1–12, 2018

    2018

  9. [17]

    Solving high-dimensional partial differential equations using deep learning

    Jiequn Han, Arnulf Jentzen, and Weinan E. Solving high-dimensional partial differential equations using deep learning. Proceedings of the National Academy of Sciences, 115(34):8505–8510, 2018

    2018

  10. [18]

    Variational physics-informed neural networks for solving partial differential equations.arXiv preprint arXiv:1912.00873, 2019

    Ehsan Kharazmi, Zhongqiang Zhang, and George Em Karniadakis. Variational physics-informed neural networks for solving partial differential equations.arXiv preprint arXiv:1912.00873, 2019

    1912 arXiv

  11. [19]

    Automatic differentiation in machine learning: a survey.Journal of machine learning research, 18(153):1–43, 2018

    Atilim Gunes Baydin, Barak A Pearlmutter, Alexey Andreyevich Radul, and Jeffrey Mark Siskind. Automatic differentiation in machine learning: a survey.Journal of machine learning research, 18(153):1–43, 2018

    2018

  12. [20]

    Stochastic gradient learning in neural networks.Proceedings of Neuro-Nımes, 91(8):12, 1991

    Léon Bottou et al. Stochastic gradient learning in neural networks.Proceedings of Neuro-Nımes, 91(8):12, 1991

    1991

  13. [21]

    A deep collocation method for the bending analysis of kirchhoff plate.arXiv preprint arXiv:2102.02617, 2021

    Hongwei Guo, Xiaoying Zhuang, and Timon Rabczuk. A deep collocation method for the bending analysis of kirchhoff plate.arXiv preprint arXiv:2102.02617, 2021

    2021 arXiv

  14. [22]

    A comparison study of deep galerkin method and deep ritz method for elliptic problems with different boundary conditions.arXiv preprint arXiv:2005.04554, 2020

    Jingrun Chen, Rui Du, and Keke Wu. A comparison study of deep galerkin method and deep ritz method for elliptic problems with different boundary conditions.arXiv preprint arXiv:2005.04554, 2020

    2005 arXiv

  15. [23]

    Int-deep: A deep learning initialized iterative method for nonlinear problems.Journal of computational physics, 419:109675, 2020

    Jianguo Huang, Haoqin Wang, and Haizhao Yang. Int-deep: A deep learning initialized iterative method for nonlinear problems.Journal of computational physics, 419:109675, 2020

    2020

  16. [24]

    Automatically imposing boundary conditions for boundary valueproblemsbyunifiedphysics-informedneuralnetwork

    Khang A Luong, Thang Le-Duc, and Jaehong Lee. Automatically imposing boundary conditions for boundary valueproblemsbyunifiedphysics-informedneuralnetwork. EngineeringwithComputers , 40(3):1717–1739, 2024

    2024

  17. [25]

    WeiLi,MartinZBazant,andJunerZhu. Aphysics-guidedneuralnetworkframeworkforelasticplates: Comparison of governing equations-based and energy-based approaches.Computer Methods in Applied Mechanics and Engineering, 383:113933, 2021

    2021

  18. [26]

    The deep ritz method: a deep learning-based numerical algorithm for solving variational problems

    Bing Yu et al. The deep ritz method: a deep learning-based numerical algorithm for solving variational problems. Communications in Mathematics and Statistics, 6(1):1–12, 2018. 19 A preprint - January 15, 2025

    2018

  19. [27]

    Anenergyapproachtothesolutionofpartialdifferentialequations incomputationalmechanicsviamachinelearning: Concepts,implementationandapplications

    Esteban Samaniego, Cosmin Anitescu, Somdatta Goswami, Vien Minh Nguyen-Thanh, Hongwei Guo, Khader Hamdia,XiaoyingZhuang,andTimonRabczuk. Anenergyapproachtothesolutionofpartialdifferentialequations incomputationalmechanicsviamachinelearning: Concepts,implementationandapplicatio...

    2020

  20. [28]

    A deep energy method for finite deformation hyperelasticity.European Journal of Mechanics-A/Solids, 80:103874, 2020

    Vien Minh Nguyen-Thanh, Xiaoying Zhuang, and Timon Rabczuk. A deep energy method for finite deformation hyperelasticity.European Journal of Mechanics-A/Solids, 80:103874, 2020

    2020

  21. [29]

    Parametric deepenergyapproachforelasticityaccountingforstraingradienteffects

    Vien Minh Nguyen-Thanh, Cosmin Anitescu, Naif Alajlan, Timon Rabczuk, and Xiaoying Zhuang. Parametric deepenergyapproachforelasticityaccountingforstraingradienteffects. ComputerMethodsinAppliedMechanics and Engineering, 386:114096, 2021

    2021

  22. [30]

    Xiaoying Zhuang, Hongwei Guo, Naif Alajlan, Hehua Zhu, and Timon Rabczuk. Deep autoencoder based energy method for the bending, vibration, and buckling analysis of kirchhoff plates with transfer learning.European Journal of Mechanics-A/Solids, 87:104225, 2021

    2021

  23. [31]

    A coupled finite element-element-free galerkin method

    Ted Belytschko, Daniel Organ, and Yury Krongauz. A coupled finite element-element-free galerkin method. Computational Mechanics, 17(3):186–195, 1995

    1995

  24. [32]

    Enrichment and coupling of the finite element and meshless methods

    Antonio Huerta and Sonia Fernández-Méndez. Enrichment and coupling of the finite element and meshless methods. International Journal for Numerical Methods in Engineering, 48(11):1615–1636, 2000

    2000

  25. [33]

    Hierarchical enrichment for bridging scales and mesh-free boundary conditions

    Gregory J Wagner and Wing Kam Liu. Hierarchical enrichment for bridging scales and mesh-free boundary conditions. International Journal for Numerical Methods in Engineering, 50(3):507–524, 2001

    2001

  26. [34]

    Element-free galerkin methods in combination with finite element approaches.Computer Methods in Applied Mechanics and Engineering, 135(1-2):143–166, 1996

    Dries Hegen. Element-free galerkin methods in combination with finite element approaches.Computer Methods in Applied Mechanics and Engineering, 135(1-2):143–166, 1996

    1996

  27. [35]

    Hybrid fem-nn models: Combining artificial neural networks with the finite element method.Journal of Computational Physics, 446:110651, 2021

    Sebastian K Mitusch, Simon W Funke, and Miroslav Kuchta. Hybrid fem-nn models: Combining artificial neural networks with the finite element method.Journal of Computational Physics, 446:110651, 2021

    2021

  28. [36]

    Enforcement of essential boundary conditions in meshless approximations using finite elements.Computer Methods in Applied Mechanics and Engineering, 131(1-2):133–145, 1996

    Yury Krongauz and Ted Belytschko. Enforcement of essential boundary conditions in meshless approximations using finite elements.Computer Methods in Applied Mechanics and Engineering, 131(1-2):133–145, 1996

    1996

  29. [37]

    Efficient physics informed neural networks coupled with domain decomposition methods for solving coupled multi-physics problems

    Long Nguyen, Maziar Raissi, and Padmanabhan Seshaiyer. Efficient physics informed neural networks coupled with domain decomposition methods for solving coupled multi-physics problems. InAdvances in Computational Modeling and Simulation, pages 41–53. Springer, 2022

    2022

  30. [38]

    A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics.Computer Methods in Applied Mechanics and Engineering, 379:113741, 2021

    Ehsan Haghighat, Maziar Raissi, Adrian Moure, Hector Gomez, and Ruben Juanes. A physics-informed deep learning framework for inversion and surrogate modeling in solid mechanics.Computer Methods in Applied Mechanics and Engineering, 379:113741, 2021

    2021

  31. [39]

    Butterworth-Heinemann, 2013

    Gui-Rong Liu and Siu Sin Quek.The finite element method: a practical course. Butterworth-Heinemann, 2013

    2013

  32. [40]

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics

    Alex Kendall, Yarin Gal, and Roberto Cipolla. Multi-task learning using uncertainty to weigh losses for scene geometry and semantics. InProceedings of the IEEE conference on computer vision and pattern recognition, pages 7482–7491, 2018

    2018

  33. [41]

    Multi-fidelity physics-constrained neural network and its application in materials modeling

    Dehao Liu and Yan Wang. Multi-fidelity physics-constrained neural network and its application in materials modeling. Journal of Mechanical Design, 141(12):121403, 2019

    2019

  34. [42]

    Understanding and mitigating gradient pathologies in physics- informed neural networks.arXiv preprint arXiv:2001.04536, 2020

    Sifan Wang, Yujun Teng, and Paris Perdikaris. Understanding and mitigating gradient pathologies in physics- informed neural networks.arXiv preprint arXiv:2001.04536, 2020

    2001 arXiv

  35. [43]

    Courier Corporation, 2012

    Cornelius Lanczos.The variational principles of mechanics. Courier Corporation, 2012

    2012

  36. [44]

    Exact imposition of boundary conditions with distance functions in physics- informed deep neural networks.Computer Methods in Applied Mechanics and Engineering, 389:114333, 2022

    N Sukumar and Ankit Srivastava. Exact imposition of boundary conditions with distance functions in physics- informed deep neural networks.Computer Methods in Applied Mechanics and Engineering, 389:114333, 2022

    2022

  37. [45]

    A three-dimensional finite element mesh generator with built-in pre-and post-processing facilities.International Journal for Numerical Methods in Engineering, 11:79, 2020

    Christophe Geuzaine and Jean-François Remacle. A three-dimensional finite element mesh generator with built-in pre-and post-processing facilities.International Journal for Numerical Methods in Engineering, 11:79, 2020

    2020

  38. [46]

    Elasticity theory.M: Science, 1975

    So P Timoshenko and J Goodyear. Elasticity theory.M: Science, 1975. 20

    1975

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.