pith. sign in

arxiv: 2605.31231 · v1 · pith:LSLWTCJ3new · submitted 2026-05-29 · 🧮 math.NA · cs.LG· cs.NA

A holomorphic neural network framework for 3D boundary value problems governed by harmonic potentials

Enrico Ballini , Allan Peter Engsig-Karup , Tito Andriollo This is my paper

Pith reviewed 2026-06-28 21:29 UTC · model grok-4.3

classification 🧮 math.NA cs.LGcs.NA
keywords holomorphic neural networksboundary value problemsharmonic potentialsWhittaker integral formulameshless approximationLaplace equationlinear elasticityPapkovich-Neuber potentials
0
0 comments X

The pith

Holomorphic neural networks solve 3D boundary value problems for harmonic potentials by satisfying the governing PDEs exactly through construction.

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

The paper develops a neural-network framework for three-dimensional boundary value problems whose solutions can be expressed using harmonic potentials. It represents these potentials via functions holomorphic in a suitable complex variable, as permitted by the Whittaker integral formula, and approximates the functions with neural networks that preserve holomorphicity by design. Because the representation automatically fulfills the partial differential equations throughout the domain, training requires only boundary collocation points and dispenses with any interior residual minimization. The resulting meshless procedure is demonstrated on scalar Laplace problems and on linear elasticity problems expressed through Papkovich-Neuber potentials, where both displacement and stress fields remain accurate across the domain.

Core claim

Solutions to the three-dimensional boundary value problems are represented through functions holomorphic with respect to a suitable complex variable via the Whittaker integral formula. These functions are approximated by holomorphic neural networks that enforce the holomorphicity condition exactly. As a direct consequence the governing partial differential equations hold identically everywhere in the domain, so that the training procedure operates exclusively on boundary collocation points.

What carries the argument

Holomorphic neural networks that approximate the holomorphic functions supplied by the Whittaker integral formula, thereby guaranteeing exact satisfaction of the harmonic-potential PDEs.

If this is right

  • The PDE residuals remain identically zero without any interior collocation or loss term.
  • Optimization uses boundary data exclusively, reducing the number of required training points.
  • The same construction supplies both scalar fields for Laplace problems and vector fields for elasticity via Papkovich-Neuber potentials.
  • Pointwise errors stay controlled throughout the interior in the reported numerical tests.

Where Pith is reading between the lines

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

  • The exact interior satisfaction may allow reliable extrapolation beyond the training boundary with smaller data sets than standard residual-based networks.
  • If similar integral representations exist for other linear operators, the same holomorphic-network construction could apply without modification.
  • The approach supplies a natural interface between analytical potential theory and meshless numerical solvers.

Load-bearing premise

The solution must be expressible in terms of harmonic potentials that admit a representation through functions holomorphic in a suitable complex variable via the Whittaker integral formula.

What would settle it

Train the network on boundary values of a known closed-form harmonic function and observe whether the network output deviates from the true interior values by more than floating-point tolerance.

Figures

Figures reproduced from arXiv: 2605.31231 by Allan Peter Engsig-Karup, Enrico Ballini, Tito Andriollo.

Figure 1
Figure 1. Figure 1: Example of a semi-holomorphic neural network with 2 hidden layers. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Domain and numerical results. (a) Domain and dimensions; the training points (dark color) and test [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Domain and numerical results. (a) The boundary conditions are imposed only to [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: First row: domain and boundary conditions for the three cases. Second row: visualization of the deformed [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: First row: domain with boundary conditions and tractions. The tractions obtained through the proposed [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Leftmost panel: stress obtained with a standard PINN approach. Fairly accurate stress field is obtained [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Value of the loss versus the epochs for the test case presented in Section [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Value of the loss versus the epochs for the test case presented in Section [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: (a) Loss versus epoch for different weight initialization. (b) maximum and minimum module of the output [PITH_FULL_IMAGE:figures/full_fig_p028_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: (a) Training points in dark color, test points in light color. (b) Loss versus epoch for different weight [PITH_FULL_IMAGE:figures/full_fig_p029_10.png] view at source ↗
read the original abstract

We present a neural-network-based framework for the solution of three-dimensional boundary value problems where the solution is expressible in terms of harmonic potentials. The approach leverages the Whittaker integral formula, which allows representing the solution through functions that are holomorphic with respect to a suitable complex variable. These functions are subsequently approximated using holomorphic neural networks, which guaranty fulfillment of the holomorphicity requirement. A key feature of the proposed formulation is that the governing partial differential equations (PDEs) are satisfied exactly by construction. Therefore, in contrast to standard physics-informed neural networks, no residual minimization of PDEs is required in the interior of the domain, and training is based exclusively on boundary collocation points. The method is validated against three-dimensional Laplace and linear elasticity problems, where, in the latter case, displacement and stress fields are expressed via the Papkovich-Neuber potentials. The numerical results show an accurate approximation of both scalar and vector fields, with errors remaining controlled throughout the domain. Overall, the work demonstrates that the incorporation of analytical structures into neural network architectures provides a natural and effective framework for the meshless approximation of three-dimensional boundary value problems while preserving the underlying properties of the governing equations.

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

0 major / 3 minor

Summary. The manuscript proposes a neural-network framework for 3D boundary-value problems whose solutions admit representation via harmonic potentials. Using the Whittaker integral formula, the potentials are expressed through holomorphic functions of a suitable complex variable; these functions are approximated by holomorphic neural networks that enforce holomorphicity exactly. Consequently the governing PDEs (Laplace or the Papkovich–Neuber system for linear elasticity) are satisfied identically for any network output, so that training reduces to boundary collocation only. Numerical results on scalar Laplace and vector elasticity test cases are reported to keep domain-wide errors controlled.

Significance. If the central construction holds, the work supplies a concrete route to exact interior PDE satisfaction within a neural-network solver for the important class of problems governed by harmonic potentials. The explicit incorporation of the Whittaker representation and holomorphic-network architecture is a genuine strength; it converts an analytic identity into an architectural constraint rather than a soft penalty. The reported boundary-only training and controlled domain errors on both scalar and vector problems indicate practical utility for meshless 3D potential and elasticity calculations.

minor comments (3)
  1. [§3.2] §3.2, Eq. (8): the precise definition of the holomorphic activation and the complex-variable mapping should be stated explicitly; the current description leaves open whether the network output is guaranteed holomorphic for finite-width networks or only in the limit.
  2. [Figure 4, Table 2] Figure 4 and Table 2: axis labels and legend entries use inconsistent font sizes and omit units for the stress components; this impairs direct comparison of the reported L² and L^∞ errors.
  3. The manuscript cites the classical Whittaker formula but does not reference recent complex-variable neural-network literature (e.g., works on holomorphic activations in complex analysis); adding two or three targeted citations would strengthen the positioning.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the careful reading and the positive overall assessment of the manuscript. The referee's summary and significance statements accurately reflect the central contribution: the exact enforcement of the governing PDEs via the Whittaker representation and holomorphic network architecture, reducing training to boundary collocation only. The recommendation for minor revision is noted. No major comments were provided in the report, so we have no specific points requiring rebuttal or clarification at this stage. We remain available to address any editorial requests for minor changes.

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper invokes the standard Whittaker integral formula (an external mathematical fact) to represent harmonic potentials via holomorphic functions, then uses holomorphic neural networks to enforce holomorphicity exactly. This makes the governing PDEs (Laplace or Papkovich-Neuber) vanish identically by construction for any network output, without fitting parameters to the interior solution or renaming a known result. No load-bearing self-citations, uniqueness theorems, or fitted-input predictions appear in the provided text; validation remains on external test cases. The scope is explicitly limited to problems admitting such representations, avoiding hidden circularity.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract provides only high-level mathematical assumptions; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Solutions of the target boundary-value problems can be represented via the Whittaker integral formula using functions holomorphic in a suitable complex variable.
    Invoked in the abstract as the foundation that allows holomorphic neural networks to satisfy the PDEs exactly.

pith-pipeline@v0.9.1-grok · 5752 in / 1144 out tokens · 26205 ms · 2026-06-28T21:29:31.639075+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

75 extracted references · 49 canonical work pages · 3 internal anchors

  1. [1]

    Fluid Mechanics

    Landau, L.D., Lifshitz, E.M., 1987. Fluid Mechanics. volume 6 ofCourse of Theoretical Physics. Pergamon Press

    1987

  2. [2]

    Fast hydrodynamics on heteroge- neous many-core hardware, in: Couturier, R

    Engsig-Karup, A.P., Glimberg, S.L., Nielsen, A.S., Lindberg, O., 2013. Fast hydrodynamics on heteroge- neous many-core hardware, in: Couturier, R. (Ed.), Designing Scientific Applications on GPUs, Chapman & Hall/CRC. doi:10.1201/b16051-20

    work page doi:10.1201/b16051-20 2013

  3. [3]

    Numerical Approximation of Partial Differential Equations

    Quarteroni, A., Valli, A., 1994. Numerical Approximation of Partial Differential Equations. Springer Berlin Heidelberg. doi:10.1007/978-3-540-85268-1

    work page doi:10.1007/978-3-540-85268-1 1994

  4. [4]

    Elasticity Theory, Applications, and Numerics

    Sadd, M.H., 2021. Elasticity Theory, Applications, and Numerics. 4th ed., Elsevier. doi:10.1016/ c2017-0-03720-5

    2021

  5. [5]

    Selected topics in the history of the two-dimensional biharmonic problem

    Meleshko, V., 2003. Selected topics in the history of the two-dimensional biharmonic problem. Texts in Applied MathematicsApplied Mechanics Reviews 56, 33–85. doi:10.1115/1.1521166

    work page doi:10.1115/1.1521166 2003

  6. [6]

    A Guide to First-Passage Processes

    Redner, S., 2001. A Guide to First-Passage Processes. Cambridge University Press. doi:10.1017/ cbo9780511606014

    2001

  7. [7]

    A stabilised nodal spectral element method for fully nonlinear water waves

    Engsig-Karup, A.P., Eskilsson, C., Bigoni, D., 2016. A stabilised nodal spectral element method for fully nonlinear water waves. Journal of Computational Physics 318. doi:10.1016/j.jcp.2016.04.060

    work page doi:10.1016/j.jcp.2016.04.060 2016

  8. [8]

    Spectral/hp element methods: Recent developments, applications, and perspectives

    Xu, H., Cantwell, C.D., Monteserin, C., Eskilsson, C., Engsig-Karup, A.P., Sherwin, S.J., 2018. Spectral/hp element methods: Recent developments, applications, and perspectives. Journal of Hydrodynamics 30, 1–22. doi:10.1007/s42241-018-0001-1

    work page doi:10.1007/s42241-018-0001-1 2018

  9. [9]

    Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement

    Hughes, T., Cottrell, J., Bazilevs, Y., 2005. Isogeometric analysis: CAD, finite elements, NURBS, exact geometry and mesh refinement. Computer Methods in Applied Mechanics and Engineering 194, 4135–4195. doi:10.1016/j.cma.2004.10.008

    work page doi:10.1016/j.cma.2004.10.008 2005

  10. [10]

    A coupled element free Galerkin/boundary element method for stress analysis of two- dimensional solids

    Gu, Y., Liu, G., 2001. A coupled element free Galerkin/boundary element method for stress analysis of two- dimensional solids. Computer Methods in Applied Mechanics and Engineering 190, 4405–4419. doi:10.1016/ s0045-7825(00)00324-8

    2001

  11. [11]

    Neural algorithm for solving differential equations

    Lee, H., Kang, I.S., 1990. Neural algorithm for solving differential equations. Journal of Computational Physics 91, 110–131. doi:10.1016/0021-9991(90)90007-N

    work page doi:10.1016/0021-9991(90)90007-n 1990

  12. [12]

    AIChE Journal38, 10 (1992), 1499–1511

    Psichogios, D.C., Ungar, L.H., 1992. A hybrid neural network-first principles approach to process modeling. AIChE Journal 38, 1499–1511. doi:10.1002/aic.690381003

    work page doi:10.1002/aic.690381003 1992

  13. [13]

    Neural-network-based approximations for solving partial differential equations

    Dissanayake, M.W.M.G., Phan-Thien, N., 1994. Neural-network-based approximations for solving partial differential equations. Communications in Numerical Methods in Engineering 10, 195–201. doi:10.1002/cnm. 1640100303

    work page doi:10.1002/cnm 1994

  14. [14]

    Artificial neural networks for solving ordinary and partial differential equations

    Lagaris, I., Likas, A., Fotiadis, D., 1998. Artificial neural networks for solving ordinary and partial differential equations. IEEE Transactions on Neural Networks 9, 987–1000. doi:10.1109/72.712178

    work page doi:10.1109/72.712178 1998

  15. [15]

    Neural-network meth- ods for boundary value problems with irregular boundaries

    Lagaris, I.E., Likas, A.C., Papageorgiou, D.G., 2000. Neural-network methods for boundary value problems with irregular boundaries. IEEE Transactions on Neural Networks 11, 1041–1049. doi:10.1109/72.870037

    work page doi:10.1109/72.870037 2000

  16. [16]

    Solving Partial Differential Equations Using Artificial Neural Networks

    Rudd, K., 2013. Solving Partial Differential Equations Using Artificial Neural Networks. Ph.D. thesis. Duke University. Durham, NC, USA

    2013

  17. [17]

    A unified deep artificial neural network approach to partial differential equations in complex geometries

    Berg, J., Nystr¨ om, K., 2018. A unified deep artificial neural network approach to partial differential equations in complex geometries. Neurocomputing 317, 28–41. doi:10.1016/j.neucom.2018.06.056

    work page doi:10.1016/j.neucom.2018.06.056 2018

  18. [18]

    Raissi, P

    Raissi, M., Perdikaris, P., Karniadakis, G., 2019. Physics-informed neural networks: A deep learning frame- work for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computational Physics 378, 686–707. doi:10.1016/j.jcp.2018.10.045. 21

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

  19. [19]

    Toscano, J.D., Oommen, V., Varghese, A.J., Zou, Z., Ahmadi Daryakenari, N., Wu, C., Karniadakis, G.E.,

  20. [20]

    Machine Learning for Computational Science and Engineering 1

    From PINNs to PIKANs: recent advances in physics-informed machine learning. Machine Learning for Computational Science and Engineering 1. doi:10.1007/s44379-025-00015-1

    work page doi:10.1007/s44379-025-00015-1

  21. [21]

    Physics-informed neural networks for inviscid transonic flows around an airfoil

    Wassing, S., Langer, S., Bekemeyer, P., 2025. Physics-informed neural networks for inviscid transonic flows around an airfoil. Physics of Fluids 37. doi:10.1063/5.0276518

    work page doi:10.1063/5.0276518 2025

  22. [22]

    Enhanced solution for the advection–diffusion–reaction equation using the physics-informed neural network technique

    Lekaba, T., Ndou, N., Muzhinji, K., Moyo, S., 2026. Enhanced solution for the advection–diffusion–reaction equation using the physics-informed neural network technique. Mathematics 14, 1194. doi:10.3390/ math14071194

    2026

  23. [23]

    Physics-informed neural networks for advection–diffusion–langmuir adsorption processes

    Huang, B., Hua, H., Han, H., He, S., Zhou, Y., Liu, S., Zuo, Z., 2024. Physics-informed neural networks for advection–diffusion–langmuir adsorption processes. Physics of Fluids 36. doi:10.1063/5.0221924

    work page doi:10.1063/5.0221924 2024

  24. [24]

    Shallow neural networks and laplace’s equation on the half-space with dirichlet boundary data

    Vaishampayan, M., 2024. Shallow neural networks and laplace’s equation on the half-space with dirichlet boundary data. Pittsburgh Interdisciplinary Mathematics Review 2, 59–70. doi:10.5195/pimr.2024.39

    work page doi:10.5195/pimr.2024.39 2024

  25. [25]

    Deep learning in computational mechanics: a review

    Herrmann, L., Kollmannsberger, S., 2024. Deep learning in computational mechanics: a review. Computational Mechanics 74, 281–331. doi:10.1007/s00466-023-02434-4

    work page doi:10.1007/s00466-023-02434-4 2024

  26. [26]

    Physics-informed neural networks (PINN) for computational solid mechanics: Numerical frameworks and applications

    Hu, H., Qi, L., Chao, X., 2024. Physics-informed neural networks (PINN) for computational solid mechanics: Numerical frameworks and applications. Thin-Walled Structures 205, 112495. doi:10.1016/j.tws.2024. 112495

    work page doi:10.1016/j.tws.2024 2024

  27. [27]

    Physics informed neural networks for continuum micromechanics

    Henkes, A., Wessels, H., Mahnken, R., 2022. Physics informed neural networks for continuum micromechanics. Computer Methods in Applied Mechanics and Engineering 393, 114790. doi:10.1016/j.cma.2022.114790

    work page doi:10.1016/j.cma.2022.114790 2022

  28. [28]

    A physics-informed neural network approach to solution and identification of biharmonic equations of elasticity

    Vahab, M., Haghighat, E., Khaleghi, M., Khalili, N., 2022. A physics-informed neural network approach to solution and identification of biharmonic equations of elasticity. Journal of Engineering Mechanics 148. doi:10.1061/(asce)em.1943-7889.0002062

    work page doi:10.1061/(asce)em.1943-7889.0002062 2022

  29. [29]

    A mixed formulation for physics-informed neural networks as a potential solver for engineering problems in heterogeneous domains: Comparison with finite element method

    Rezaei, S., Harandi, A., Moeineddin, A., Xu, B.X., Reese, S., 2022. A mixed formulation for physics-informed neural networks as a potential solver for engineering problems in heterogeneous domains: Comparison with finite element method. Computer Methods in Applied Mechanics and Engineering 401, 115616. doi:10.1016/ j.cma.2022.115616

    arXiv 2022

  30. [30]

    Deep learning-accelerated computational framework based on physics informed neural network for the solution of linear elasticity

    Roy, A.M., Bose, R., Sundararaghavan, V., Arr´ oyave, R., 2023. Deep learning-accelerated computational framework based on physics informed neural network for the solution of linear elasticity. Neural Networks 162, 472–489. doi:10.1016/j.neunet.2023.03.014

    work page doi:10.1016/j.neunet.2023.03.014 2023

  31. [31]

    On the performance and convergence of PINNs for problems in linear elasticity

    Kadlag, D., Birk, C., Natarajan, S., Gravenkamp, H., 2026. On the performance and convergence of PINNs for problems in linear elasticity. Proceedings in Applied Mathematics and Mechanics 26. doi:10.1002/pamm.70113

    work page doi:10.1002/pamm.70113 2026

  32. [32]

    Transfer learning based physics-informed neural networks for solving inverse problems in engineering structures under different loading scenarios

    Xu, C., Cao, B.T., Yuan, Y., Meschke, G., 2023. Transfer learning based physics-informed neural networks for solving inverse problems in engineering structures under different loading scenarios. Computer Methods in Applied Mechanics and Engineering 405, 115852. doi:10.1016/j.cma.2022.115852

    work page doi:10.1016/j.cma.2022.115852 2023

  33. [33]

    doi:10.1126/sciadv.abk0644 Jun Zhang, Zhi-hui Zhan, Ying Lin, Ni Chen, Yue-jiao Gong, Jing-hui Zhong, Henry S.H

    Zhang, E., Dao, M., Karniadakis, G.E., Suresh, S., 2022. Analyses of internal structures and defects in materials using physics-informed neural networks. Science Advances 8. doi:10.1126/sciadv.abk0644

    work page doi:10.1126/sciadv.abk0644 2022

  34. [34]

    In: Medical Imaging with Deep Learning (MIDL)

    Park, S., Yun, C., Lee, J., Shin, J., 2020. Minimum width for universal approximation doi:10.48550/ARXIV. 2006.08859,arXiv:2006.08859

    work page internal anchor Pith review doi:10.48550/arxiv 2020

  35. [35]

    Achieve the minimum width of neural networks for universal approximation doi:10.48550/ ARXIV.2209.11395,arXiv:2209.11395

    Cai, Y., 2022. Achieve the minimum width of neural networks for universal approximation doi:10.48550/ ARXIV.2209.11395,arXiv:2209.11395

    arXiv 2022

  36. [36]

    Minimum width of leaky-relu neural networks for uniform universal approximation doi:10.48550/ARXIV.2305.18460,arXiv:2305.18460

    Li, L., Duan, Y., Ji, G., Cai, Y., 2023. Minimum width of leaky-relu neural networks for uniform universal approximation doi:10.48550/ARXIV.2305.18460,arXiv:2305.18460

    work page doi:10.48550/arxiv.2305.18460 2023

  37. [37]

    Approximating continuous functions by ReLU nets of minimal width doi:10

    Hanin, B., Sellke, M., 2017. Approximating continuous functions by ReLU nets of minimal width doi:10. 48550/ARXIV.1710.11278,arXiv:1710.11278. 22

    Pith/arXiv arXiv 2017

  38. [38]

    Estimates on the generalization error of physics-informed neural networks for approximating pdes

    Mishra, S., Molinaro, R., 2022. Estimates on the generalization error of physics-informed neural networks for approximating PDEs. IMA Journal of Numerical Analysis 43, 1–43. doi:10.1093/imanum/drab093

    work page doi:10.1093/imanum/drab093 2022

  39. [39]

    1980, Biological Cybernetics, 36, 193, doi: 10.1007/BF00344251

    Fukushima, K., 1980. Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position. Biological Cybernetics 36, 193–202. doi:10.1007/bf00344251

    work page doi:10.1007/bf00344251 1980

  40. [40]

    Attention is all you need, in: Advances in Neural Information Processing Systems 30 (NeurIPS 2017), pp

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I., 2017. Attention is all you need, in: Advances in Neural Information Processing Systems 30 (NeurIPS 2017), pp. 5998–6008

    2017

  41. [41]

    An energy approach to the solution of partial differential equations in computational mechanics via machine learning: Concepts, implementation and applications

    Samaniego, E., Anitescu, C., Goswami, S., Nguyen-Thanh, V., Guo, H., Hamdia, K., Zhuang, X., Rabczuk, T., 2020. An energy approach to the solution of partial differential equations in computational mechanics via machine learning: Concepts, implementation and applications. Computer Methods in Applied Mechanics and Engineering 362, 112790. doi:10.1016/j.cma...

    work page doi:10.1016/j.cma.2019.112790 2020

  42. [42]

    Dcem: A deep complementary energy method for linear elasticity

    Wang, Y., Sun, J., Rabczuk, T., Liu, Y., 2024. Dcem: A deep complementary energy method for linear elasticity. International Journal for Numerical Methods in Engineering 125. doi:10.1002/nme.7585

    work page doi:10.1002/nme.7585 2024

  43. [43]

    A neural solver for variational problems on cad geometries with application to electric machine simulation

    von Tresckow, M., Kurz, S., De Gersem, H., Loukrezis, D., 2022. A neural solver for variational problems on cad geometries with application to electric machine simulation. Journal of Machine Learning for Modeling and Computing 3, 49–75. doi:10.1615/jmachlearnmodelcomput.2022041753

    work page doi:10.1615/jmachlearnmodelcomput.2022041753 2022

  44. [44]

    Variational physics- informed neural networks for solving partial differential equations

    Kharazmi, E., Zhang, Z., Karniadakis, G.E., 2019. Variational physics-informed neural networks for solving partial differential equations doi:10.48550/ARXIV.1912.00873,arXiv:1912.00873

    work page doi:10.48550/arxiv.1912.00873 2019

  45. [45]

    hp-vpinns: Variational physics-informed neural networks with domain decomposition

    Kharazmi, E., Zhang, Z., Karniadakis, G.E., 2021. hp-vpinns: Variational physics-informed neural networks with domain decomposition. Computer Methods in Applied Mechanics and Engineering 374, 113547. doi:10. 1016/j.cma.2020.113547

    arXiv 2021

  46. [46]

    Binn: A deep learning approach for computational mechanics problems based on boundary integral equations

    Sun, J., Liu, Y., Wang, Y., Yao, Z., Zheng, X., 2023. Binn: A deep learning approach for computational mechanics problems based on boundary integral equations. Computer Methods in Applied Mechanics and Engineering 410, 116012. doi:10.1016/j.cma.2023.116012

    work page doi:10.1016/j.cma.2023.116012 2023

  47. [47]

    Harmonic neural networks, in: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J

    Ghosh, A., Gentile, A.A., Dagrada, M., Lee, C., Kim, S.H.S., Cha, H., Choi, Y., Kim, D., Kye, J.I., Elfving, V.E., 2023. Harmonic neural networks, in: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J. (Eds.), Proceedings of the 40th International Conference on Machine Learning, PMLR. pp. 11340– 11359

    2023

  48. [48]

    Physics-informed holomorphic neural networks (PIHNNs): Solving 2D linear elasticity problems

    Calaf` a, M., Hovad, E., Engsig-Karup, A.P., Andriollo, T., 2024. Physics-informed holomorphic neural networks (PIHNNs): Solving 2D linear elasticity problems. Computer Methods in Applied Mechanics and Engineering 432, 117406. doi:10.1016/j.cma.2024.117406

    work page doi:10.1016/j.cma.2024.117406 2024

  49. [49]

    A holomorphic Kolmogorov-Arnold network framework for solving elliptic problems on arbitrary 2D domains doi:10.48550/ARXIV.2507.22678, arXiv:2507.22678

    Calaf` a, M., Andriollo, T., Engsig-Karup, A.P., Jeong, C.H., 2025a. A holomorphic Kolmogorov-Arnold network framework for solving elliptic problems on arbitrary 2D domains doi:10.48550/ARXIV.2507.22678, arXiv:2507.22678

    work page doi:10.48550/arxiv.2507.22678

  50. [50]

    Solving plane crack problems via enriched holomorphic neural networks

    Calaf` a, M., Jensen, H.M., Andriollo, T., 2025b. Solving plane crack problems via enriched holomorphic neural networks. Engineering Fracture Mechanics 322, 111133. doi:10.1016/j.engfracmech.2025.111133

    work page doi:10.1016/j.engfracmech.2025.111133 2025

  51. [51]

    A Variational Kolosov--Muskhelishvili Network for Elasticity and Fracture

    Zhou, S., H¨ affner, C., Stebner, S., Fehlemann, N., Wei, Z., M¨ unstermann, S., 2026a. A variational Kolosov– Muskhelishvili network for elasticity and fracture doi:10.48550/ARXIV.2605.02310,arXiv:2605.02310

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.02310

  52. [52]

    Transfer- learned Kolosov–Muskhelishvili informed neural networks for fracture mechanics

    Zhou, S., H¨ affner, C., Wang, S., Stebner, S., Liao, Z., Yang, B., Wei, Z., M¨ unstermann, S., 2026b. Transfer- learned Kolosov–Muskhelishvili informed neural networks for fracture mechanics. Theoretical and Applied Fracture Mechanics 144, 105582. doi:10.1016/j.tafmec.2026.105582

    work page doi:10.1016/j.tafmec.2026.105582 2026

  53. [53]

    On the partial differential equations of mathematical physics

    Whittaker, E.T., 1903. On the partial differential equations of mathematical physics. Mathematische Annalen 57, 333–355. doi:10.1007/bf01444290. 23

    work page doi:10.1007/bf01444290 1903

  54. [54]

    Complex Analysis

    Stein, E.M., 2003. Complex Analysis. 1st ed. ed., Princeton University Press, Princeton. Description based on publisher supplied metadata and other sources

    2003

  55. [55]

    Complex Analysis An Introduction to the Theory of Analytic Functions of One Complex Variable

    Ahlfors, L.V., 2021. Complex Analysis An Introduction to the Theory of Analytic Functions of One Complex Variable. American Mathematical Society

    2021

  56. [56]

    Analytic functions of hypercomplex variables

    Ketchum, P.W., 1928. Analytic functions of hypercomplex variables. Transactions of the American Mathe- matical Society 30, 641–667. doi:10.1090/s0002-9947-1928-1501452-7

    work page doi:10.1090/s0002-9947-1928-1501452-7 1928

  57. [57]

    The use of complex valued functions for the solution of three-dimensional elasticity problems

    Piltner, R., 1987. The use of complex valued functions for the solution of three-dimensional elasticity problems. Journal of Elasticity 18, 191–225. doi:10.1007/bf00044194

    work page doi:10.1007/bf00044194 1987

  58. [58]

    Monogenic functions in a finite-dimensional semi-simple commu- tative algebra

    Plaksa, S.A., Pukhtaievych, R.P., 2014. Monogenic functions in a finite-dimensional semi-simple commu- tative algebra. Analele Universitatii “Ovidius” Constanta - Seria Matematica 22, 221–235. doi:10.2478/ auom-2014-0018

    2014

  59. [59]

    On the completeness of the Papkovich–Neuber solution

    Tran-Cong, T., 1989. On the completeness of the Papkovich–Neuber solution. Quarterly of Applied Mathe- matics 47, 645–659. doi:10.1090/qam/1031682

    work page doi:10.1090/qam/1031682 1989

  60. [60]

    Approximation properties of complex-valued neural networks: An overview, in: 2025 International Conference on Sampling Theory and Applications (SampTA), IEEE

    Geuchen, P., 2025. Approximation properties of complex-valued neural networks: An overview, in: 2025 International Conference on Sampling Theory and Applications (SampTA), IEEE. pp. 1–5. doi:10.1109/ sampta64769.2025.11133560

    arXiv 2025

  61. [61]

    Introductory Functional Analysis: With Applications to Boundary Value Problems and Finite Elements

    Reddy, B.D., 1998. Introductory Functional Analysis: With Applications to Boundary Value Problems and Finite Elements. Springer New York. doi:10.1007/978-1-4612-0575-3

    work page doi:10.1007/978-1-4612-0575-3 1998

  62. [62]

    Partial differential equations

    Evans, L.C., 2022. Partial differential equations. Number 19 in Graduate studies in mathematics. second ed., American Mathematical Society, Providence, Rhode Island. Literaturverzeichnis: Seite 689-701

    2022

  63. [63]

    Adam: A Method for Stochastic Optimization

    Kingma, D.P., Ba, J., 2014. Adam: A method for stochastic optimization doi:10.48550/ARXIV.1412.6980, arXiv:1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2014

  64. [64]

    Weak baselines and reporting biases lead to overoptimism in ma- chine learning for fluid-related partial differential equations

    McGreivy, N., Hakim, A., 2024. Weak baselines and reporting biases lead to overoptimism in ma- chine learning for fluid-related partial differential equations. Nature Machine Intelligence 6, 1256–1269. doi:10.1038/s42256-024-00897-5

    work page doi:10.1038/s42256-024-00897-5 2024

  65. [65]

    Numerical Analysis

    Burden, R.L., Faires, J.D., 2011. Numerical Analysis. 9 ed., Brooks/Cole, Cengage Learning

    2011

  66. [66]

    Elimination-compensation pruning for fully-connected neural networks doi:10.48550/ARXIV.2602.20467,arXiv:2602.20467

    Ballini, E., Muscarnera, L., Fumagalli, A., Scotti, A., Regazzoni, F., 2026. Elimination-compensation pruning for fully-connected neural networks doi:10.48550/ARXIV.2602.20467,arXiv:2602.20467

    work page doi:10.48550/arxiv.2602.20467 2026

  67. [67]

    Overview about solution representations for elasticity problems and some selected particular solutions

    Piltner, R., 2001. Overview about solution representations for elasticity problems and some selected particular solutions. Mathematics and Mechanics of Solids 6, 193–220. doi:10.1177/108128650100600205

    work page doi:10.1177/108128650100600205 2001

  68. [68]

    ´A.L., Abarca Jim´ enez, G.S., 2021

    Mares Carre˜ no, J., Ortega Herrera, J. ´A.L., Abarca Jim´ enez, G.S., 2021. A displacement potential function using complex variables for numerical computations of three-dimensional elasticity problems. Archive of Applied Mechanics 91, 2331–2344. doi:10.1007/s00419-021-01885-6

    work page doi:10.1007/s00419-021-01885-6 2021

  69. [69]

    JAX: composable transformations of Python+NumPy programs

    Bradbury, J., Frostig, R., Hawkins, P., Johnson, M.J., Katariya, Y., Leary, C., Maclaurin, D., Necula, G., Paszke, A., VanderPlas, J., Wanderman-Milne, S., Zhang, Q., 2018. JAX: composable transformations of Python+NumPy programs. Open-source software

    2018

  70. [70]

    CUDA Toolkit.https://developer.nvidia.com/cuda-toolkit

    NVIDIA, . CUDA Toolkit.https://developer.nvidia.com/cuda-toolkit

  71. [71]

    On the convergence of Adam and beyond, in: International Conference on Learning Representations

    Reddi, S.J., Kale, S., Kumar, S., 2018. On the convergence of Adam and beyond, in: International Conference on Learning Representations

    2018

  72. [72]

    On the limited memory BFGS method for large scale optimization

    Liu, D.C., Nocedal, J., 1989. On the limited memory BFGS method for large scale optimization. Mathematical Programming 45, 503–528. 24

    1989

  73. [73]

    Understanding the difficulty of training deep feedforward neural networks

    Glorot, X., Bengio, Y., 2010. Understanding the difficulty of training deep feedforward neural networks. Journal of Machine Learning Research 9, 249–256

    2010

  74. [74]

    Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification, in: 2015 IEEE International Conference on Computer Vision (ICCV)

    He, K., Zhang, X., Ren, S., Sun, J., 2015. Delving deep into rectifiers: Surpassing human-level performance on ImageNet classification, in: 2015 IEEE International Conference on Computer Vision (ICCV). doi:10.1109/ iccv.2015.123. 25 A Computational framework and implementation details We provide in this appendix information on practical aspects of the tes...

    2015

  75. [75]

    The computation of the discrete form of V, (15), andu, (13), is parallelized over the quadrature points and the training points within each minibatch

    neural network routines and uses CUDA [69] for GPU acceleration. The computation of the discrete form of V, (15), andu, (13), is parallelized over the quadrature points and the training points within each minibatch. The functions related to the weight updates are compiled usingjax.jit. We remark that optimizing the code to achieve the best performance is ...

    2000