pith. sign in

arxiv: 2605.25562 · v1 · pith:J7UOJP65new · submitted 2026-05-25 · 🧮 math.NA · cs.NA

Consistent CutPINNs for Elliptic PDEs on Curved Level-Set Domains

Maneesh Kumar Singh This is my paper

Pith reviewed 2026-06-29 20:48 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords Consistent CutPINNsElliptic PDEsLevel-set domainsH^{1/2} trace normChord-arc equivalencePINNsOptimal recoveryCut domains
0
0 comments X

The pith

A discrete H^{1/2} surrogate built from collocation points on a C^2 curve is equivalent to the continuous trace norm and yields a priori H^1 error bounds for PINNs on curved domains.

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

The paper develops Consistent CutPINN for second-order elliptic PDEs posed on bounded domains defined implicitly by a C^2 level-set function in two dimensions. Standard PINN training penalizes boundary mismatch only in L^2, which fails to control the H^{1/2} trace norm appearing in the H^1 energy estimate. The authors construct a discrete surrogate for that trace norm directly from collocation points on the curve, prove a Chord-arc equivalence between the surrogate and the continuous norm, and use the equivalence to obtain an a priori H^1 error bound on cut domains together with convergence rates under Besov regularity via optimal recovery. Experiments on a disk and a non-convex flower domain show the consistent loss is markedly more accurate and robust to cut-cell geometry than the usual L^2 penalty.

Core claim

We introduce a discrete H^{1/2}(∂Ω) surrogate built directly from collocation points on a C^2 curve, prove a Chord-arc norm equivalence between this surrogate and the continuous trace norm, establish an a priori H^1 error bound on cut domains, and derive convergence rates under Besov regularity using optimal recovery theory.

What carries the argument

The Chord-arc norm equivalence between the discrete H^{1/2} surrogate constructed from boundary collocation points and the continuous H^{1/2} trace norm on the C^2 curve.

If this is right

  • The consistent boundary loss produces smaller H^1 errors than the standard L^2 penalty on the same collocation set.
  • The method remains accurate across varying cut-cell configurations without retuning.
  • Convergence rates follow from Besov regularity of the solution via optimal recovery arguments.
  • The framework applies to any second-order elliptic problem on a two-dimensional domain whose boundary is given by a C^2 level-set function.

Where Pith is reading between the lines

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

  • The same point-based surrogate construction could be attempted on surfaces in three dimensions if a suitable discrete norm equivalence can be proved.
  • The Chord-arc equivalence supplies a general template for replacing L^2 boundary penalties with trace-norm surrogates in other variational problems.
  • Because the rates rest on optimal recovery, they are likely sharp for the stated Besov classes.

Load-bearing premise

The chosen collocation points on the C^2 curve are dense enough that the discrete surrogate norm remains equivalent to the continuous H^{1/2} trace norm with constants independent of the discretization.

What would settle it

For some C^2 level-set curve and sequence of collocation sets, the ratio of the discrete surrogate norm to the continuous H^{1/2} norm tends to infinity as the number of points grows.

Figures

Figures reproduced from arXiv: 2605.25562 by Maneesh Kumar Singh.

Figure 1
Figure 1. Figure 1: Left: collocation setup on the cut domain Ω = [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Decomposition of the consistent loss L ∗ sq,2 defined in (14). The network output vθ and its Laplacian ∆vθ feed two residual terms: the interior L 2 residual at X˜ and the boundary H1/2 residual at Z, which combine into the scalar loss. Training stops when L ∗ sq,2 falls below a tolerance ε or 2000 L-BFGS iterations are reached. 3.4 Optimal recovery framework The consistent loss is designed to achieve the … view at source ↗
Figure 3
Figure 3. Figure 3: Cut-cell decomposition of Ω ⊂ Q: interior cells T int (teal) and cut cells T cut (red, crossing ∂Ω). Proposition 5 (L 2 discretisation on cut domains). Let Ω ⊂ R 2 be a bounded domain with ∂Ω a C 2 level-set curve of length L; fix 1 ≤ p ≤ ∞, s > 2/p, r > max(s, 1), and write B = Bs p (Ω). Let Q ⊃ Ω be an axis-aligned bounding box and Gk,r ⊂ Q the tensor-product grid of mesh size h = (diam Q)·2 −k/(r−1). Se… view at source ↗
Figure 4
Figure 4. Figure 4: Collocation setup for the two test domains embedded in [0 [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Convergence on the disk (log-log). Solid lines are means over 10 seeds, shaded bands span min [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Left: H1 error vs. iteration (log scale). Right: loss value vs. iteration. The standard loss converges its own objective but fails to drive down the H1 error. Diverged (NaN) runs are excluded from the mean and standard deviation in [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Upper: H1 error vs. disk-centre coordinate c (log scale). Lower: rolling coefficient of variation (window 11). Lsq,1 and Lsq,λ overlap exactly (blue hidden under orange, confirming the d = 2 identity) and exhibit large spikes. L ∗ sq,2 (red) is flat. 5.2.5 Experiment 5: Spatial error distribution The previous experiments report only scalar H1 values. Here we look at the spatial distribution of the pointwis… view at source ↗
Figure 8
Figure 8. Figure 8: Side-by-side convergence: flower (left) vs. disk (right). The advantage of the consistent losses [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Solution quality on the disk domain (me = 900, m = 30). Top row: true solution u (left), standard PINN approximation vθ trained with Lsq,1 (centre), and consistent approximation trained with L ∗ sq,2 (right). The H1 error of each approximation is shown above its panel. Bottom row: pointwise absolute error |u − vθ| for Lsq,1 (left) and L ∗ sq,2 (right), sharing a common colorbar. The consistent loss reduces… view at source ↗
Figure 10
Figure 10. Figure 10: Solution quality on the flower domain (me = 900, m = 30). Layout as in [PITH_FULL_IMAGE:figures/full_fig_p025_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Monte Carlo boundary collocation on the unit disk ( [PITH_FULL_IMAGE:figures/full_fig_p026_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Solution quality on the unit disk with non-uniform boundary sampling ( [PITH_FULL_IMAGE:figures/full_fig_p026_12.png] view at source ↗
read the original abstract

We propose \emph{Consistent CutPINN}, a framework for partial differential equations posed on bounded curved domains defined implicitly by a $\C^2$ level-set function, $\Omega = \{\varphi < 0\}$. In this paper we develop the framework for second-order elliptic problems in two dimensions. The standard PINN loss penalises the boundary mismatch in $L^2(\partial\Omega)$, but $L^2(\partial\Omega)$ does not control the $H^{1/2}(\partial\Omega)$ trace norm that appears in the $H^1(\Omega)$ energy estimate. The consistent PINN framework of Bonito et al.~\cite{bonito2025} fixes this on the unit cube $(0,1)^d$ via a Kuhn--Tucker simplicial decomposition of the flat boundary faces, but the construction relies on the affine structure of the faces and does not carry over to smooth curved boundaries. We address this gap. Specifically, (i) we introduce a discrete $H^{1/2}(\partial\Omega)$ surrogate built directly from collocation points on a $\C^2$ curve, (ii) we prove a \textit{Chord-arc} norm equivalence between this surrogate and the continuous trace norm, (iii) we establish an \emph{a priori} $H^1$ error bound on cut domains, and (iv) we derive convergence rates under Besov regularity using optimal recovery theory. Numerical experiments on a disk and a non-convex flower domain confirm that the consistent loss is much more accurate than the standard PINN loss and far more robust to cut-cell configurations.

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 the Consistent CutPINN framework for second-order elliptic PDEs posed on 2D bounded domains defined implicitly by a C^2 level-set function. It constructs a discrete H^{1/2}(∂Ω) surrogate directly from collocation points on the curve, proves a Chord-arc norm equivalence between this surrogate and the continuous trace norm, establishes an a priori H^1 error bound on cut domains, and derives convergence rates under Besov regularity via optimal recovery theory. Numerical experiments on a disk and a non-convex flower domain demonstrate that the consistent loss outperforms the standard L^2 boundary penalization in accuracy and robustness to cut-cell configurations.

Significance. If the equivalence and a priori bounds hold, the work supplies a theoretically grounded extension of the consistent PINN approach to curved boundaries, ensuring the boundary loss controls the H^{1/2} trace norm appearing in the H^1 energy estimate. The chord-arc equivalence, the a priori bound on cut domains, and the application of optimal recovery theory for rates constitute clear strengths; the numerical validation on both convex and non-convex geometries further supports practical utility.

minor comments (3)
  1. [Abstract and §1] The reference to Bonito et al. (2025) appears in the abstract and introduction; confirm that the bibliography entry is complete and that the year is accurate.
  2. [Numerical experiments] The description of the flower domain level-set function is given only qualitatively; an explicit formula would aid reproducibility of the non-convex example.
  3. [§3] Notation for the discrete surrogate norm is introduced without an explicit equation label in the early sections; adding an equation number would improve cross-referencing.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We are grateful to the referee for their detailed summary of the manuscript and for recommending minor revision. The report does not contain any major comments, therefore we provide no point-by-point responses below. We will make the necessary minor revisions to the manuscript accordingly.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper's central claims consist of a new discrete H^{1/2} surrogate constructed from collocation points on a C^2 level-set curve, a proved Chord-arc equivalence to the continuous trace norm, an a priori H^1 error bound on cut domains, and convergence rates derived via optimal recovery under Besov regularity. These are presented as original mathematical constructions and proofs rather than reductions to fitted inputs, self-definitions, or prior results by the same authors. The reference to Bonito et al. (2025) supplies only the base consistent-PINN loss for flat domains and is not load-bearing for the curved-boundary extensions, which rest on independent arguments. No patterns of self-definitional equivalence, fitted quantities renamed as predictions, or ansatz smuggling via self-citation are present.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, ad-hoc axioms, or invented entities are identifiable. The C^2 regularity of the level-set is a standard domain assumption for the trace theory invoked.

pith-pipeline@v0.9.1-grok · 5823 in / 1263 out tokens · 29276 ms · 2026-06-29T20:48:32.400748+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Consistent CutPINNs for Convection-Diffusion Equations on Curved Level-Set Domains

    math.NA 2026-06 unverdicted novelty 7.0

    Proves a single a priori H1 error bound for consistent CutPINNs using discrete L^gamma interior loss (gamma = 1 + 1/log m_til) and discrete H^{1/2} boundary trace norm on curved level-set domains, with rate limited by...

Reference graph

Works this paper leans on

30 extracted references · 1 canonical work pages · cited by 1 Pith paper

  1. [1]

    Anandh, D

    T. Anandh, D. Ghose, H. Jain, and S. Ganesan. FastVPINNs: Tensor-driven acceleration of VPINNs for complex geometries.SIAM J. Sci. Comput., 47(3):C578–C600, 2025

    2025

  2. [2]

    Anandh, D

    T. Anandh, D. Ghose, H. Jain, P. Sunkad, S. Ganesan, and V. John. Improving hp-variational physics-informed neural networks for steady-state convection-dominated problems.Comput. Methods Appl. Mech. Eng., 438:117797, 2025

    2025

  3. [3]

    Badia, F

    S. Badia, F. Verdugo, and A. F. Mart´ ın. The aggregated unfitted finite element method for elliptic problems.Comput. Methods Appl. Mech. Eng., 336:533–553, 2018

    2018

  4. [4]

    Bonito, R

    A. Bonito, R. DeVore, G. Petrova, and J. W. Siegel. Convergence and error control of consistent PINNs for elliptic PDEs.IMA J. Numer. Anal., 2025

    2025

  5. [5]

    S. C. Brenner and L. R. Scott.The Mathematical Theory of Finite Element Methods. Springer, New York, 3rd edition, 2008

    2008

  6. [6]

    Cengizci, ¨O

    S. Cengizci, ¨O. U˘ gur, and S. Natesan. A PINN-enhanced SUPG-stabilized hybrid finite element framework with shock-capturing for computing steady convection-dominated flows.Adv. Eng. Softw., 216:104135, 2026

    2026

  7. [7]

    Cohen, R

    A. Cohen, R. DeVore, G. Petrova, and P. Wojtaszczyk. Optimal stable nonlinear approximation. Found. Comput. Math., 22(3):607–648, 2022

    2022

  8. [8]

    Cuomo, V

    S. Cuomo, V. S. Di Cola, F. Giampaolo, G. Rozza, M. Raissi, and F. Piccialli. Scientific machine learning through physics–informed neural networks: Where we are and what’s next.J. Sci. Comput., 92(3):88, 2022

    2022

  9. [9]

    De Ryck and S

    T. De Ryck and S. Mishra. Generic bounds on the approximation error for physics-informed (and) operator learning.Advances in Neural Information Processing Systems, 35:10945–10958, 2022

    2022

  10. [10]

    De Ryck and S

    T. De Ryck and S. Mishra. Numerical analysis of physics-informed neural networks and related models in physics-informed machine learning.Acta Numer., 33:633–713, 2024

    2024

  11. [11]

    DeVore, B

    R. DeVore, B. Hanin, and G. Petrova. Neural network approximation.Acta Numer., 30:327–444, 2021

    2021

  12. [12]

    R. A. DeVore and R. C. Sharpley. Besov spaces on domains inR d.Trans. Amer. Math. Soc., 335(2):843–864, 1993

    1993

  13. [13]

    Frerichs-Mihov, L

    D. Frerichs-Mihov, L. Henning, and V. John. On loss functionals for physics-informed neural net- works for steady-state convection-dominated convection-diffusion problems.Commun. Appl. Math. Comput., 8(1):287–308, 2026

    2026

  14. [14]

    T. G. Grossmann, U. J. Komorowska, J. Latz, and C.-B. Sch¨ onlieb. Can physics-informed neural networks beat the finite element method?IMA J. Numer. Anal., 89(1):143–174, 2024. 27

    2024

  15. [15]

    When do ex- tended physics-informed neural networks (xpinns) improve generalization?SIAM J

    Zheyuan Hu, Ameya D Jagtap, George Em Karniadakis, and Kenji Kawaguchi. When do ex- tended physics-informed neural networks (xpinns) improve generalization?SIAM J. Sci. Comput., 44(5):A3158–A3182, 2022

    2022

  16. [16]

    Khan, K.-A

    A. Khan, K.-A. Mardal, and S. Mishra. Mixed consistent PINNs for elliptic obstacle problems with stability analysis.arXiv preprint arXiv:2604.01719, 2026

    work page arXiv 2026

  17. [17]

    Mart´ ın, and Santiago Badia

    Wei Li, Alberto F. Mart´ ın, and Santiago Badia. Unfitted finite element interpolated neural networks. J. Comput. Phys., page 114547, 2026

    2026

  18. [18]

    Mishra and A

    S. Mishra and A. Khan. Consistent PINNs for higher-order elliptic PDEs.Int. J. Numer. Methods Eng., 127(7):e70320, 2026

    2026

  19. [19]

    Mishra and R

    S. Mishra and R. Molinaro. Estimates on the generalization error of physics-informed neural networks for approximating PDEs.IMA J. Numer. Anal., 43(1):1–43, 2023

    2023

  20. [20]

    Paszke et

    A. Paszke et. al. PyTorch: An imperative style, high-performance deep learning library.Advances in neural information processing systems, 32, 2019

    2019

  21. [21]

    Plankovskyy, Y

    S. Plankovskyy, Y. Tsegelnyk, N. Shyshko, I. Litvinchev, T. Romanova, C. Velarde, and J. M. Jos´ e. Review of physics-informed neural networks: Challenges in loss function design and geometric integration.Mathematics, 13(20):3289, 2025

    2025

  22. [22]

    Raina, S

    A. Raina, S. Badireddi, and S. Natesan. Application of PINN to obtain solution of boundary layer problems arising in fluid dynamics.Math. Found. Comput., 10:89–108, 2026

    2026

  23. [23]

    Raissi, P

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

    2019

  24. [24]

    John Wiley & Sons, 2009

    Robert J Serfling.Approximation theorems of mathematical statistics. John Wiley & Sons, 2009

    2009

  25. [25]

    Y. Shin, J. Darbon, and G. E. Karniadakis. On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type pdes.Comm. Comput. Phys., 28(5):2042–2074, 2020

    2042

  26. [26]

    Singh and R

    A. Singh and R. K. Sinha. SS-DNN: A hybrid strang splitting deep neural network approach for solving the Allen–Cahn equation.Eng. Anal. Bound. Elem., 169:105944, 2024

    2024

  27. [27]

    Singh and R

    A. Singh and R. K. Sinha. Mc-PINN for solving conservative allen–cahn equations using Metropolis– Hasting based sampling.Eng. Anal. Bound. Elem., 184:106643, 2026

    2026

  28. [28]

    E. M. Stein.Singular Integrals and Differentiability Properties of Functions. Princeton University Press, Princeton, NJ, 1970

    1970

  29. [29]

    Visser, A

    C. Visser, A. Heinlein, and B. Giovanardi. PACMANN: Point adaptive collocation method for artificial neural networks.Comput. Methods Appl. Mech. Eng., 452:118723, 2026

    2026

  30. [30]

    Zeinhofer, R

    M. Zeinhofer, R. Masri, and K.-A. Mardal. A unified framework for the error analysis of physics- informed neural networks.IMA J. Numer. Anal., 45(5):2988–3025, 2025. 28

    2025