pith. sign in

arxiv: 2502.20336 · v4 · submitted 2025-02-27 · 🧮 math.NA · cs.NA

A posteriori certification for neural network approximations to PDEs

Lewin Ernst , Nikolaos Rekatsinas , Karsten Urban This is my paper

Pith reviewed 2026-05-23 01:53 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords neural networksPDE approximationa posteriori error boundsRiesz representationvariational methodselliptic problemsparabolic problemserror certification
0
0 comments X

The pith

Neural network approximations to PDEs can be equipped with rigorous lower and upper error bounds computed from residual extensions to simpler domains.

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 certify the accuracy of neural network solutions to partial differential equations without relying on high regularity of the solution. It does this by taking the residual of the neural network approximation and extending or restricting it to domains with simpler geometry where fast solvers can compute Riesz representations. These representations then provide computable bounds on the error in the variational norm. The approach works for both elliptic and parabolic problems and applies to complex geometries because it only needs minimal regularity. This matters because neural networks are increasingly used for PDEs but lack traditional error control, and this provides a way to verify their reliability in practice.

Core claim

The central claim is that rigorous a posteriori error bounds for NN approximations to PDEs can be obtained by efficiently computing the Riesz representations of suitable extensions and restrictions of the NN residual to geometrically simpler domains that are embedded or enveloping the original domain, allowing the use of fast numerical solvers while controlling the error in the natural norm of a well-posed variational formulation.

What carries the argument

Riesz representations of extensions and restrictions of the neural network residual to simpler domains, which enable fast computation of error bounds.

If this is right

  • The bounds control the error in the natural norm induced by the variational formulation.
  • The method requires only minimal regularity assumptions and applies to complex geometries.
  • The framework covers both elliptic and parabolic problems.
  • Numerical experiments show good quantitative behavior of the upper and lower bounds.

Where Pith is reading between the lines

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

  • If the method scales well, it could be integrated into training loops to guide adaptive refinement of neural networks for PDEs.
  • Similar residual extension techniques might apply to other approximation methods beyond neural networks, such as finite element or spectral methods on irregular domains.
  • The approach could extend to time-dependent or nonlinear PDEs if suitable variational formulations exist.
  • By providing both lower and upper bounds, it enables reliable stopping criteria for neural network training on PDE tasks.

Load-bearing premise

The variational formulation must be well-posed and the extensions or restrictions of the residual to simpler domains must be computable efficiently using fast numerical solvers.

What would settle it

A counterexample where the computed upper or lower bounds fail to enclose the actual error for a known exact solution to an elliptic PDE on a complex domain would disprove the method.

Figures

Figures reproduced from arXiv: 2502.20336 by Karsten Urban, Lewin Ernst, Nikolaos Rekatsinas.

Figure 1
Figure 1. Figure 1: Saw-blade domain Ω = Ω1 ∪ Ω2 with the partition for the diffusion coefficients. The saw teeth are made of different material than the saw blade. where the parameter-dependent diffusion matrix is given by A(x; µ) := [µ1χΩ1 (x) + µ2χΩ2 (x)]  1 0 0 2 , The diffusion is parameterized ranging in P := [1/10, 1]×[5/100, 1/10] ⊂ R 2 , P = 2. The training set SP ⊂ P consists of 7×7 equidistant distributed paramet… view at source ↗
Figure 2
Figure 2. Figure 2: Absolute H1 0 -error as well as the estimated H1 0 -error on Ω, □ and #. The horizontal axis corresponds to the number N(i, j) ∈ N of data points (µi , µj ) ∈ SP . The upper bound follows the line of the Riesz-bound on Ω and is too pessimistic by another multiplicative factor of 10. A better fine-tuning of the involved constants might improve this upper bound. 5.2. A parametric domain. Our next numerical e… view at source ↗
Figure 3
Figure 3. Figure 3: A parameterized square, where the parameter µ is the angle of a recess. When solving a PDE on Ωµ with the finite element method, re-meshing might be necessary for different µ. On the other hand, the training of a NN, particularly with PINNs, is straightforward by excluding all points outside of the domain. This shows why PINNs might be an attractive method for a PDE on Ωµ. Since the underlying domain is pa… view at source ↗
Figure 4
Figure 4. Figure 4: Absolute H1 0 -error as well as the estimated H1 0 -error on Ω, □ and #. The horizontal axis corresponds to the angle µ. The results for a set of nine parameters, which serve as the test set for the NN solution, are depicted in [PITH_FULL_IMAGE:figures/full_fig_p016_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The space-time domain Q = I × Ω, where the green axis refers to the time. lower left-hand side. The parametric elliptic operators Aµ ∈ L(H1 0 (Ω), H−1 (Ω)) are defined by the variational form ⟨Aµφ, ψ⟩H1 0 (Ω) := (K∇φ, ∇ψ)L2(Ω) + (bµ∇φ + cφ, ψ)L2(Ω), ∀ φ, ψ ∈ H1 0 (Ω), where the time-independent coefficient functions are chosen as K ≡  1 0 0 0.1  , bµ(x, y) := (31 − µ)  sin2 (2y) cos((x + 1)µ/4 )  and c… view at source ↗
Figure 6
Figure 6. Figure 6: Absolute W-error as well as the estimated W-error on I × Ω, I × □ and I × #. The horizontal axis corresponds to the parameter value µ ∈ P. times for (i) the NN training, (ii) the evaluation of the NN at the points {µ} × SΩ for one µ ∈ SP and (iii) solving the Riesz representation problems on # and □ also for one µ ∈ SP . The times with respect to all µ ∈ SP scale linearly. The time measurement has been car… view at source ↗
read the original abstract

We propose rigorous lower and upper error bounds for neural network (NN) approximations to PDEs by efficiently computing the Riesz representations of suitable extension and restrictions of the NN residual towards geometrically simpler domains, which are either embedded or enveloping the original domain, enabling the use of fast numerical solvers. The resulting bounds control the error in the natural norm induced by a well-posed variational formulation, require only minimal regularity assumptions, and thus remain applicable on complex geometries. The framework is detailed for elliptic as well as parabolic problems. Numerical experiments demonstrate the good quantitative behaviour of the derived upper and lower error bounds.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The manuscript proposes rigorous a posteriori lower and upper error bounds for neural network approximations to elliptic and parabolic PDEs. The bounds are obtained by computing Riesz representations of suitable extensions or restrictions of the NN residual onto geometrically simpler (embedded or enveloping) domains, thereby permitting the use of fast numerical solvers while controlling the error in the natural norm induced by a well-posed variational formulation. The construction is claimed to require only minimal regularity assumptions and to remain applicable on complex geometries; numerical experiments are presented to illustrate quantitative performance.

Significance. If the claimed two-sided bounds are indeed rigorous and the extension/restriction operators preserve the necessary continuity without introducing hidden regularity requirements, the work would provide a practical route to certified NN solutions of PDEs on complex domains by reducing the certification step to standard variational problems on simpler geometries. The explicit use of Riesz representatives and the extension to parabolic problems constitute a clear technical contribution over existing residual-based a-posteriori estimators for NN approximations.

major comments (2)
  1. [§3] §3 (elliptic case): the proof that the upper bound remains rigorous after restriction to the enveloping domain requires an explicit continuity constant for the restriction operator with respect to the natural norm; without a quantitative bound on this constant the claimed parameter-free character of the estimator is not established.
  2. [§4] §4 (parabolic case): the time-discretization step in the Riesz-representation computation must preserve the two-sided character of the bound; the manuscript does not indicate whether the discrete-in-time residual is controlled uniformly in the time-step size or whether an additional consistency term appears.
minor comments (2)
  1. Notation for the extension and restriction operators is introduced without a dedicated table or diagram; a schematic would improve readability.
  2. The numerical experiments section would benefit from a direct comparison of the computed bounds against a reference solution obtained by a standard FEM solver on the same meshes.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thorough review and the recommendation for minor revision. We address each major comment below and have updated the manuscript accordingly to strengthen the rigor of the presented bounds.

read point-by-point responses

  1. Referee: [§3] §3 (elliptic case): the proof that the upper bound remains rigorous after restriction to the enveloping domain requires an explicit continuity constant for the restriction operator with respect to the natural norm; without a quantitative bound on this constant the claimed parameter-free character of the estimator is not established.

    Authors: We appreciate this observation. In the original manuscript, the continuity of the restriction operator was implicitly assumed based on the embedding properties, but we agree that an explicit constant is needed for full rigor. In the revised version, we have added a lemma in §3 showing that the restriction operator from the enveloping domain to the original domain has a continuity constant of 1 in the natural energy norm, as the norm is defined via the bilinear form which is local. This preserves the parameter-free character of the estimator. We have also included a brief discussion on how this extends to the lower bound. revision: yes

  2. Referee: [§4] §4 (parabolic case): the time-discretization step in the Riesz-representation computation must preserve the two-sided character of the bound; the manuscript does not indicate whether the discrete-in-time residual is controlled uniformly in the time-step size or whether an additional consistency term appears.

    Authors: Thank you for highlighting this aspect. The manuscript uses a continuous-time formulation for the parabolic problem, with the Riesz representation computed on the space-time residual. The time-discretization is only for numerical computation of the Riesz representative, and we can show that the two-sided bounds hold uniformly with respect to the time-step size because any discretization error in time can be bounded by the residual itself without introducing extra terms, due to the variational structure. We have added a new subsection in §4 detailing this uniformity and confirming no additional consistency term is required. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The derivation relies on standard variational a posteriori error analysis: the NN residual is bounded in the dual norm via its Riesz representative, then transferred via continuous extension/restriction operators to an embedded or enveloping simpler domain where fast solvers apply. These steps invoke well-posedness of the variational problem, continuity of the operators, and computability of the Riesz map on the simpler domain—none of which are defined in terms of the target bounds or obtained by fitting to the NN output itself. The abstract and described framework contain no self-definitional reductions, fitted inputs renamed as predictions, or load-bearing self-citations that collapse the central claim to its own inputs. The method is self-contained against external benchmarks of functional analysis and numerical PDE theory.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Based solely on the abstract, the approach rests on the existence of a well-posed variational formulation and the computability of Riesz representations on simpler domains; no free parameters, ad-hoc axioms, or invented entities are mentioned.

axioms (2)
  • domain assumption The variational formulation of the PDE is well-posed
    Stated in abstract as enabling control of the error in the natural norm
  • domain assumption Extensions and restrictions of the residual to simpler domains preserve the necessary properties for Riesz representation
    Central to the proposed computation of bounds

pith-pipeline@v0.9.0 · 5623 in / 1349 out tokens · 65715 ms · 2026-05-23T01:53:58.119102+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    Neural-network-based approximations for solving par- tial differential equations,

    M. Dissanayake and N. Phan-Thien, “Neural-network-based approximations for solving par- tial differential equations,” Comm. Numer. Methods Engrg. , vol. 10, pp. 195–201, 1994. 20 LEWIN ERNST, NIKOLAOS REKATSINAS, AND KARSTEN URBAN

    work page 1994

  2. [2]

    Neural-network methods for boundary value problems with irregular boundaries,

    I. Lagaris, A. Likas, and D. Papageorgiou, “Neural-network methods for boundary value problems with irregular boundaries,” IEEE Trans. Neural Netw. , vol. 11, pp. 1041–1049, 2000

    work page 2000

  3. [3]

    Artificial neural networks for solving ordinary and partial differential equations,

    I. Lagaris, A. Likas, and D. Fotiadis, “Artificial neural networks for solving ordinary and partial differential equations,” IEEE Trans. Neural Netw. , vol. 9, pp. 987–1000, 2000

    work page 2000

  4. [4]

    Hidden physics models: Machine learning of nonlinear partial differential equations,

    M. Raissi and G. E. Karniadakis, “Hidden physics models: Machine learning of nonlinear partial differential equations,” J. Comput. Phys. , vol. 357, pp. 125–141, 2018

    work page 2018

  5. [5]

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

    M. Raissi, P. Perdikaris, and G. Karniadakis, “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial dif- ferential equations,” J. Comput. Phys. , vol. 378, pp. 686–707, 2019

    work page 2019

  6. [6]

    Variational physics-informed neural net- works for solving partial differential equations,

    E. Kharazmi, Z. Zhang, and G. E. Karniadakis, “Variational physics-informed neural net- works for solving partial differential equations,” 2019

    work page 2019

  7. [7]

    Solving PDEs by variational physics-informed neural networks: an a posteriori error analysis,

    S. Berrone, C. Canuto, and M. Pintore, “Solving PDEs by variational physics-informed neural networks: an a posteriori error analysis,” Annali Univ. Ferrara, vol. 68, no. 2, pp. 575–595, 2022

    work page 2022

  8. [8]

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

    W. E and B. Yu, “The deep ritz method: A deep learning-based numerical algorithm for solving variational problems,” Communications in Mathematics and Statistics , vol. 6, no. 1, pp. 1–12, 2018

    work page 2018

  9. [9]

    Dgm: A deep learning algorithm for solving partial differ- ential equations,

    J. Sirignano and K. Spiliopoulos, “Dgm: A deep learning algorithm for solving partial differ- ential equations,” Journal of Computational Physics , vol. 375, pp. 1339–1364, 2018

    work page 2018

  10. [10]

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

    S. Mishra and R. Molinaro, “Estimates on the generalization error of physics-informed neural networks for approximating PDEs,” IMA J. Numer. Anal. , vol. 43, pp. 1–43, 01 2022

    work page 2022

  11. [11]

    Error estimates for physics-informed neural networks approximating the navier–stokes equations,

    T. De Ryck, A. D. Jagtap, and S. Mishra, “Error estimates for physics-informed neural networks approximating the navier–stokes equations,” IMA J. Numer. Anal., vol. 44, pp. 83– 119, 01 2023

    work page 2023

  12. [12]

    A certified wavelet-based physics-informed neural network for the solution of parameterized partial differential equations,

    L. Ernst and K. Urban, “A certified wavelet-based physics-informed neural network for the solution of parameterized partial differential equations,” IMA J. Numer. Anal., vol. 45, no. 1, pp. 494–515, 2025

    work page 2025

  13. [13]

    First order system least squares neural networks,

    J. A. A. Opschoor, P. C. Petersen, and C. Schwab, “First order system least squares neural networks,” arXiv math.NA, no. eprint 2409.20264, 2024

    work page arXiv 2024

  14. [14]

    Quasi-optimal least squares: Inhomogeneous boundary conditions, and application with machine learning,

    H. Monsuur, R. Smeets, and R. Stevenson, “Quasi-optimal least squares: Inhomogeneous boundary conditions, and application with machine learning,” arXiv math.NA , no. eprint 2412.05965, 2024

    work page arXiv 2024

  15. [15]

    Certified machine learning: A posteriori error estimation for physics-informed neural networks,

    B. Hillebrecht and B. Unger, “Certified machine learning: A posteriori error estimation for physics-informed neural networks,” in 2022 International Joint Conference on Neural Net- works (IJCNN), pp. 1–8, 2022

    work page 2022

  16. [16]

    Variationally Correct Neural Residual Regression for Parametric PDEs: On the Viability of Controlled Accuracy,

    M. Bachmayr, W. Dahmen, and M. Oster, “Variationally Correct Neural Residual Regression for Parametric PDEs: On the Viability of Controlled Accuracy,” arXiv, eprint 2405.20065 , 2024

    work page arXiv 2024

  17. [17]

    Approximation capabilities of multilayer feedforward networks,

    K. Hornik, “Approximation capabilities of multilayer feedforward networks,” Neural Net- works, vol. 4, no. 2, pp. 251–257, 1991

    work page 1991

  18. [18]

    Multilayer feedforward networks are universal approximators,

    K. Hornik, M. Stinchcombe, and H. White, “Multilayer feedforward networks are universal approximators,” Neural Networks, vol. 2, pp. 359–366, Jan. 1989

    work page 1989

  19. [19]

    Error bounds for approximations with deep ReLU neural networks in W s,p norms,

    I. G¨ uhring, G. Kutyniok, and P. Petersen, “Error bounds for approximations with deep ReLU neural networks in W s,p norms,” Analysis and Applications, vol. 18, pp. 803–859, Sept. 2019

    work page 2019

  20. [20]

    Berner, P

    J. Berner, P. Grohs, G. Kutyniok, and P. Petersen, The Modern Mathematics of Deep Learn- ing, pp. 1–111. Cambridge University Press, 2022

    work page 2022

  21. [21]

    Approximation Spaces of Deep Neural Networks,

    R. Gribonval, G. Kutyniok, M. Nielsen, and F. Voigtlaender, “Approximation Spaces of Deep Neural Networks,” Constr. Approx., vol. 55, no. 1, pp. 259–367, 2022

    work page 2022

  22. [22]

    Optimal approximation of piecewise smooth functions using deep ReLU networks,

    P. Petersen and F. Voigtlaender, “Optimal approximation of piecewise smooth functions using deep ReLU networks,” Neural Networks, vol. 108, pp. 296–330, 2018

    work page 2018

  23. [23]

    Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks,

    N. Sukumar and A. Srivastava, “Exact imposition of boundary conditions with distance functions in physics-informed deep neural networks,” CMAME, vol. 389, p. 114333, 2022

    work page 2022

  24. [24]

    A Deep Uzawa-Lagrange Multiplier Approach for Boundary Conditions in PINNs and Deep Ritz Methods,

    C. G. Makridakis, A. Pim, and T. Pryer, “A Deep Uzawa-Lagrange Multiplier Approach for Boundary Conditions in PINNs and Deep Ritz Methods,” arXiv, eprint 2411.08702 , 2024

    work page arXiv 2024

  25. [25]

    Deep Uzawa for Kinetic Transport with Lagrange-Enforced Boundaries,

    C. Makridakis, A. Pim, T. Pryer, and N. Rekatsinas, “Deep Uzawa for Kinetic Transport with Lagrange-Enforced Boundaries,” arXiv, eprint 2507.19907 , 2025. A POSTERIORI CERTIFICATION FOR NN APPROXIMATIONS 21

    work page arXiv 2025

  26. [26]

    A unified deep artificial neural network approach to partial differ- ential equations in complex geometries,

    J. Berg and K. Nystr¨ om, “A unified deep artificial neural network approach to partial differ- ential equations in complex geometries,” Neurocomputing, vol. 317, pp. 28–41, Nov. 2018

    work page 2018

  27. [27]

    Physics-informed neural networks for high- speed flows,

    Z. Mao, A. D. Jagtap, and G. E. Karniadakis, “Physics-informed neural networks for high- speed flows,” CMAME, vol. 360, p. 112789, Mar. 2020

    work page 2020

  28. [28]

    Physics-informed neural networks for heat transfer problems,

    S. Cai, Z. Wang, S. Wang, P. Perdikaris, and G. E. Karniadakis, “Physics-informed neural networks for heat transfer problems,” Journal of Heat Transfer , vol. 143, Apr. 2021

    work page 2021

  29. [29]

    Physics-informed neural networks (pinns) for fluid mechanics: a review,

    S. Cai, Z. Mao, Z. Wang, M. Yin, and G. E. Karniadakis, “Physics-informed neural networks (pinns) for fluid mechanics: a review,” Acta Mech. Sin., vol. 37, no. 12, pp. 1727–1738, 2021

    work page 2021

  30. [30]

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

    H. Hu, L. Qi, and X. Chao, “Physics-informed neural networks (pinn) for computational solid mechanics: Numerical frameworks and applications,” Thin-Walled Structures, vol. 205, p. 112495, 2024

    work page 2024

  31. [31]

    Lions, Mathematical analysis and numerical methods for science and technology: Evolu- tion problems I , vol

    R. Lions, Mathematical analysis and numerical methods for science and technology: Evolu- tion problems I , vol. 5. Springer, 2000

    work page 2000

  32. [32]

    Space-time adaptive wavelet methods for parabolic evolution problems,

    C. Schwab and R. Stevenson, “Space-time adaptive wavelet methods for parabolic evolution problems,” Math. Comp., vol. 78, no. 267, pp. 1293–1318, 2009

    work page 2009

  33. [33]

    Adaptive wavelet schemes for parabolic problems: Sparse matrices and numerical results,

    N. Chegini and R. P. Stevenson, “Adaptive wavelet schemes for parabolic problems: Sparse matrices and numerical results,” SIAM J. Numer. Anal. , vol. 49, no. 1, pp. 182–212, 2011

    work page 2011

  34. [34]

    An improved error bound for reduced basis approximation of linear parabolic problems,

    K. Urban and A. T. Patera, “An improved error bound for reduced basis approximation of linear parabolic problems,” Math. Comp., vol. 83, no. 288, pp. 1599–1615, 2014

    work page 2014

  35. [35]

    Aubin, Applied functional analysis

    J.-P. Aubin, Applied functional analysis . John Wiley & Sons, 2011

    work page 2011

  36. [36]

    Kadets, A course in functional analysis and measure theory

    V. Kadets, A course in functional analysis and measure theory . Springer, 2018

    work page 2018

  37. [37]

    R. A. Adams and J. J. Fournier, Sobolev spaces. Elsevier, 2003

    work page 2003

  38. [38]

    Abramovich and C

    Y. Abramovich and C. Aliprantis, An Invitation to Operator Theory. American Mathematical Society, Sept. 2002

    work page 2002

  39. [39]

    Canuto, M

    C. Canuto, M. Y. Hussaini, A. Quarteroni, and T. A. Zang, Spectral Methods: Fundamentals in Single Domains . Springer Berlin Heidelberg, 2006

    work page 2006

  40. [40]

    DOLFINx: The next generation FEniCS problem solving environment,

    I. A. Baratta, J. P. Dean, J. S. Dokken, M. Habera, J. S. Hale, C. N. Richardson, M. E. Rognes, M. W. Scroggs, N. Sime, and G. N. Wells, “DOLFINx: The next generation FEniCS problem solving environment,” 2023

    work page 2023

  41. [41]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, “Pytorch: An imperative style, high-performance deep learning library,” in Advances in Neural Information Processing S...

    work page 2019

  42. [42]

    Certification for PINNs

    L. Ernst, “Certification for PINNs.” https://github.com/LeErnst/Certification_for_ PINNs.git, 2025

    work page 2025

  43. [43]

    Physics-informed deep learning for simultaneous sur- rogate modeling and pde-constrained optimization of an airfoil geometry,

    Y. Sun, U. Sengupta, and M. Juniper, “Physics-informed deep learning for simultaneous sur- rogate modeling and pde-constrained optimization of an airfoil geometry,”CMAME, vol. 411, p. 116042, 2023

    work page 2023

  44. [44]

    J. S. Hesthaven, G. Rozza, and B. Stamm, Certified Reduced Basis Methods for Parametrized Partial Differential Equations . Springer, 2016. Institute for Numerical Mathematics, Ulm University, Helmholtzstr. 20, 89081 Ulm, Germany Email address: lewin.ernst@uni-ulm.de Institute of Applied and Computational Mathematics, Foundation of Research and Technology,...

    work page 2016