pith. sign in

arxiv: 2604.19601 · v1 · submitted 2026-04-21 · 🧮 math.NA · cs.NA

Quadrature-Enhanced Monte Carlo fPINN Method for High-Dimensional Fractional PDEs

Qingkui Ma , Hehu Xie , Xiaobo Yin This is my paper

Pith reviewed 2026-05-10 01:40 UTC · model grok-4.3

classification 🧮 math.NA cs.NA
keywords fractional LaplacianMonte Carlo methodsphysics-informed neural networksquadrature approximationhigh-dimensional PDEsboundary singularitiesnonlocal operators
0
0 comments X

The pith

A geometry-adaptive decomposition lets quadrature and Monte Carlo sampling solve high-dimensional fractional PDEs more accurately than prior Monte Carlo fPINN methods.

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

This paper develops a numerical method for fractional partial differential equations on bounded domains in high dimensions, where nonlocal kernels, exterior boundary conditions, and reduced regularity near boundaries create major difficulties. It introduces a spatially varying radius that incorporates directional distance-to-boundary data to split the fractional Laplacian into a singular near-field part, a regular interior far-field part, and an analytical exterior far-field part. Gauss-Jacobi quadrature integrates the singular radial term, Gauss quadrature handles the regular interior radial term, and Monte Carlo sampling approximates the angular integrals, while a feature-enhanced physics-informed neural network approximates the low-regularity solution. Numerical tests on fractional Poisson equations and time-dependent problems show the resulting QE-MC-fPINN method produces smaller errors and faster convergence than two earlier Monte Carlo fPINN schemes, especially when the true solution has strong boundary singularities.

Core claim

The paper establishes the quadrature-enhanced Monte Carlo fractional physics-informed neural network (QE-MC-fPINN) method. A spatially varying radius with directional distance-to-boundary information yields a geometry-adaptive three-part decomposition of the fractional Laplacian into singular near-field, regular interior far-field, and analytical exterior far-field contributions. Gauss-Jacobi quadrature is used for the singular radial integral, Gauss quadrature for the regular interior radial integral, and Monte Carlo sampling for the angular variables. A feature-enhanced physics-informed neural network trial space addresses the low-regularity behavior near the boundary. On the tested high-1

What carries the argument

Geometry-adaptive three-part decomposition of the fractional Laplacian via a spatially varying radius that incorporates directional distance-to-boundary information, with Gauss-Jacobi quadrature on the singular radial integral, Gauss quadrature on the regular interior radial integral, Monte Carlo sampling on angular variables, and feature-enhanced PINN trial space.

Load-bearing premise

The spatially varying radius with directional distance-to-boundary information produces an accurate geometry-adaptive three-part decomposition of the fractional Laplacian whose quadrature and Monte Carlo approximations do not introduce uncontrolled errors in high dimensions or near boundaries.

What would settle it

Numerical experiments on one of the paper's high-dimensional fractional Poisson benchmarks with a solution that has strong boundary singularity, in which the new method fails to show lower error or faster convergence than the two representative MC-fPINN discretizations.

Figures

Figures reproduced from arXiv: 2604.19601 by Hehu Xie, Qingkui Ma, Xiaobo Yin.

Figure 1
Figure 1. Figure 1: Architecture of the proposed feature-enhanced PINN . The trial function incorpo￾rates the temporal factor t γ and the boundary feature (b(x))µj to enforce homogeneous con￾straints and improve the approximation of low-regularity solutions. The displayed trial function corresponds to the time-dependent case. To impose the homogeneous boundary condition in a hard manner, we introduce a prescribed boundary fea… view at source ↗
Figure 2
Figure 2. Figure 2: Discretization scheme of the fractional Laplacian on a 2D irregular domain. The black closed curve denotes the physical boundary ∂Ω. For each interior evaluation point x (colored markers), a local ball Br0(x) (x) of radius r0(x) is defined. The plane R 2 is partitioned into three regions and discretized using three independent sets of Monte Carlo angular directions: (1) x → ∂Br0(x) (x) (near-field, 15 symm… view at source ↗
Figure 3
Figure 3. Figure 3: Loss and relative ℓ 2 error curves for the high-dimensional fractional Poisson equation (3.1) on the unit ball. results show that the proposed Quadrature-Enhanced decomposition substantially improves the robustness and resolution of the fractional Laplacian discretization in high dimensions. 3.2 Time-Dependent Fractional Diffusion Equation The trial solution Ψ is defined in (2.5). Since the initial conditi… view at source ↗
Figure 4
Figure 4. Figure 4: Loss and relative ℓ 2 error curves for the high-dimensional time-dependent fractional diffusion equation (2.2) on the unit ball, with {γ, α, v, c} = {0.5, 1.5, 1, 1} and T = 1. Improved MC-fPINN. This advantage is particularly clear for the singular solution and becomes more significant in higher dimensions. Overall, these results demonstrate that the proposed quadrature-enhanced discretization substantial… view at source ↗
read the original abstract

Fractional PDEs involving the fractional Laplacian on bounded domains are challenging because of hypersingular nonlocal kernels, exterior Dirichlet constraints, reduced boundary regularity, and the high computational cost in high dimensions. To address these issues, we first adopt a spatially varying radius with directional distance-to-boundary information, which yields a geometry-adaptive three-part decomposition of the fractional Laplacian: singular near-field, regular interior far-field, and analytical exterior far-field contributions. Then we employ Gauss-Jacobi quadrature for the singular radial integral, Gauss quadrature for the regular interior radial integral, and Monte Carlo sampling for the angular variables. A feature-enhanced physics-informed neural network trial space is finally used to tackle the low-regularity behavior near the boundary. Through the above steps, we obtain a quadrature-enhanced Monte Carlo fractional physics-informed neural network (QE-MC-fPINN) method. Numerical experiments on fractional Poisson equations and time-dependent fractional PDEs show that, on the tested benchmarks, the proposed method outperforms two representative MC-fPINN discretizations in accuracy and convergence, especially for solutions with strong boundary singularities.

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

3 major / 2 minor

Summary. The paper proposes the quadrature-enhanced Monte Carlo fractional physics-informed neural network (QE-MC-fPINN) method for high-dimensional fractional PDEs on bounded domains. It introduces a spatially varying radius based on directional distance-to-boundary information to decompose the fractional Laplacian into singular near-field, regular interior far-field, and analytical exterior far-field contributions. Gauss-Jacobi quadrature is applied to the singular radial integral, Gauss quadrature to the regular radial integral, and Monte Carlo sampling to the angular variables, combined with a feature-enhanced fPINN trial space to address low boundary regularity. Numerical experiments on fractional Poisson equations and time-dependent fractional PDEs demonstrate that the method outperforms two representative MC-fPINN discretizations in accuracy and convergence, particularly for solutions exhibiting strong boundary singularities.

Significance. If the geometry-adaptive decomposition maintains controlled quadrature truncation and Monte Carlo variance independently of dimension and proximity to the boundary, the approach would offer a practical advance for nonlocal high-dimensional problems by blending deterministic quadrature with stochastic sampling and neural-network approximation, improving handling of reduced regularity at boundaries over pure MC-fPINN baselines.

major comments (3)
  1. [§2 (method description)] The method description (abstract and §2) introduces the spatially varying radius and three-part decomposition without supplying a priori error bounds or variance estimates demonstrating that quadrature truncation and Monte Carlo sampling errors remain controlled independently of dimension and distance to the boundary; this is load-bearing for the high-dimensional claim and the asserted robustness near singularities.
  2. [§4 (numerical experiments)] Numerical experiments (§4) report outperformance on the tested benchmarks but provide no statistical variance from the Monte Carlo angular sampling, no sensitivity analysis with respect to sample size, and no explicit comparison of error behavior as dimension increases, leaving open whether the observed gains are robust or specific to the moderate-dimensional cases shown.
  3. [entire manuscript] No convergence analysis, error estimates, or proof of consistency for the combined quadrature-MC-fPINN scheme is supplied, despite the paper appearing in a numerical analysis venue; the central superiority claim therefore rests entirely on the reported numerical comparisons without theoretical support.
minor comments (2)
  1. [§2] Notation for the directional distance-to-boundary function and the splitting radius should be defined explicitly with an equation number in the method section for reproducibility.
  2. [abstract and §4] The abstract and experiments section should clarify the precise dimensions and domain geometries used in the benchmarks to substantiate the 'high-dimensional' descriptor.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the careful reading and constructive comments on our manuscript. We address each major comment below and indicate the planned revisions.

read point-by-point responses

  1. Referee: [§2 (method description)] The method description (abstract and §2) introduces the spatially varying radius and three-part decomposition without supplying a priori error bounds or variance estimates demonstrating that quadrature truncation and Monte Carlo sampling errors remain controlled independently of dimension and distance to the boundary; this is load-bearing for the high-dimensional claim and the asserted robustness near singularities.

    Authors: We agree that the manuscript does not supply a priori error bounds or variance estimates showing control independent of dimension and boundary distance. The geometry-adaptive radius and three-part split are motivated by the need to isolate the singular near-field contribution (treated by Gauss-Jacobi quadrature) from the regular interior far-field (Gauss quadrature) and the exterior (analytical), with Monte Carlo used only for the angular part. In the revision we will add a subsection in §2 that qualitatively discusses the error sources, cites standard quadrature error results for the radial integrals, and explains why the directional distance-to-boundary choice reduces the effective singularity strength near the boundary. Full dimension-independent bounds, however, are not derived in the present work. revision: partial

  2. Referee: [§4 (numerical experiments)] Numerical experiments (§4) report outperformance on the tested benchmarks but provide no statistical variance from the Monte Carlo angular sampling, no sensitivity analysis with respect to sample size, and no explicit comparison of error behavior as dimension increases, leaving open whether the observed gains are robust or specific to the moderate-dimensional cases shown.

    Authors: We accept that the current numerical section lacks reported Monte Carlo variance, sample-size sensitivity, and systematic dimension scaling. The revised §4 will include (i) error bars obtained from repeated independent angular samplings, (ii) tables or plots showing accuracy versus number of Monte Carlo samples for representative problems, and (iii) additional experiments in dimensions up to at least d=8 (where computational cost remains manageable) together with a brief discussion of observed scaling trends. These additions will directly address the robustness question. revision: yes

  3. Referee: [entire manuscript] No convergence analysis, error estimates, or proof of consistency for the combined quadrature-MC-fPINN scheme is supplied, despite the paper appearing in a numerical analysis venue; the central superiority claim therefore rests entirely on the reported numerical comparisons without theoretical support.

    Authors: The manuscript is a method-development paper whose central contribution is the geometry-adaptive decomposition combined with the feature-enhanced fPINN trial space, validated on challenging benchmarks that exhibit strong boundary singularities. A rigorous convergence analysis of the full hybrid scheme (quadrature + Monte Carlo + neural-network approximation) is not provided and would constitute a substantial separate theoretical effort. In the revision we will add a short subsection in the conclusions that explicitly states this limitation and outlines possible directions for future consistency analysis. revision: no

standing simulated objections not resolved
  • The absence of a rigorous convergence analysis, error estimates, or proof of consistency for the combined quadrature-MC-fPINN scheme.

Circularity Check

0 steps flagged

No significant circularity; method and claims are self-contained

full rationale

The paper proposes a QE-MC-fPINN approach by selecting a spatially varying radius to enable a three-part split of the fractional Laplacian, then applying specific quadratures (Gauss-Jacobi, Gauss) and Monte Carlo sampling, followed by a feature-enhanced PINN. Validation consists of numerical benchmarks on Poisson and time-dependent problems showing outperformance versus two external MC-fPINN baselines. No equations or results reduce by construction to fitted inputs, self-definitions, or load-bearing self-citations; the central performance claim rests on independent empirical comparisons rather than tautological reductions. The derivation chain introduces design choices whose accuracy is tested externally, satisfying the criteria for a non-circular finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the standard assumption that the fractional Laplacian admits the stated three-part decomposition when a spatially varying radius is used, plus the usual well-posedness of the fractional Poisson problem and the approximation properties of Gauss-Jacobi and Monte Carlo quadrature.

axioms (1)
  • domain assumption The fractional Laplacian on a bounded domain can be decomposed into singular near-field, regular interior far-field, and analytical exterior far-field contributions via a spatially varying radius informed by directional distance-to-boundary.
    This decomposition is the foundational step that enables the quadrature and Monte Carlo treatment.

pith-pipeline@v0.9.0 · 5491 in / 1311 out tokens · 35164 ms · 2026-05-10T01:40:01.671430+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

27 extracted references · 27 canonical work pages

  1. [1]

    Acosta, F

    G. Acosta, F. M. Bersetche, and J. P. Borthagaray. A short FE implementation for a 2d homogeneous Dirichlet problem of a fractional Laplacian. Comput. Math. Appl. , 74(4):784– 816, 2017

    work page 2017

  2. [2]

    Ainsworth and C

    M. Ainsworth and C. Glusa. Towards an efficient finite element method for the integral fractional Laplacian on polygonal domains. In Contemporary computational mathematics— a celebration of the 80th birthday of Ian Sloan. Vol. 1, 2 , pages 17–57. Springer, Cham, 2018

    work page 2018

  3. [3]

    Bonito, J

    A. Bonito, J. P. Borthagaray, R. H. Nochetto, E. Otárola, and A. J. Salgado. Numerical methods for fractional diffusion. Comput. Vis. Sci. , 19:19–46, 2018

    work page 2018

  4. [4]

    D. W. Brzezinski. Computation of Gauss-Jacobi quadrature nodes and weights with arbi- trary precision. In Proc. FedCSIS, pages 297–306. IEEE, 2018

    work page 2018

  5. [5]

    Caffarelli and L

    L. Caffarelli and L. Silvestre. An extension problem related to the fractional Laplacian. Comm. Partial Differential Equations , 32(7-9):1245–1260, 2007

    work page 2007

  6. [6]

    D’Elia, Q

    M. D’Elia, Q. Du, C. A. Glusa, M. D. Gunzburger, X. Tian, and Z. Zhou. Numerical methods for nonlocal and fractional models. Acta Numer. , 29:1–124, 2020. 15

    work page 2020

  7. [7]

    Ersland and E

    O. Ersland and E. R. Jakobsen. On fractional and nonlocal parabolic mean field games in the whole space. J. Differ. Equ. , 301:428–470, 2021

    work page 2021

  8. [8]

    J. Gao, M. Zhao, N. Du, X. Guo, H. Wang, and J. Zhang. A finite element method for space-time directional fractional diffusion partial differential equations in the plane and its error analysis. J. Comput. Appl. Math. , 362:354–365, 2019

    work page 2019

  9. [9]

    G. Grubb. Fractional Laplacians on domains, a development of hörmander’s theory of µ-transmission pseudodifferential operators. Adv. Math. , 268:478–528, 2015

    work page 2015

  10. [10]

    L. Guo, H. Wu, X. Yu, and T. Zhou. Monte Carlo fPINNs: deep learning method for for- ward and inverse problems involving high dimensional fractional partial differential equa- tions. Comput. Methods Appl. Mech. Engrg. , 400:115523, 2022

    work page 2022

  11. [11]

    X. Guo, Y. Li, and H. Wang. Tempered fractional diffusion equations for pricing multi-asset options under CGMYe process. Comput. Math. Appl. , 76(6):1500–1514, 2018

    work page 2018

  12. [12]

    Guo and P

    Y. Guo and P. Ming. A deep learning method for computing eigenvalues of the fractional Schrödinger operator. J. Syst. Sci. Complex. , 37(1):391–412, 2024

    work page 2024

  13. [13]

    Z. Hu, K. Kawaguchi, Z. Zhang, and G. E. Karniadakis. Tackling the curse of dimension- ality in fractional and tempered fractional PDEs with physics-informed neural networks. Comput. Methods Appl. Mech. Engrg. , 432:117448, 2024

    work page 2024

  14. [14]

    Huang and J

    W. Huang and J. Shen. A grid-overlay finite difference method for the fractional Laplacian on arbitrary bounded domains. SIAM J. Sci. Comput. , 46(2):A744–A769, 2024

    work page 2024

  15. [15]

    Jahanshahi, E

    S. Jahanshahi, E. Babolian, D. F. M. Torres, and A. R. Vahidi. A fractional Gauss-Jacobi quadrature rule for approximating fractional integrals and derivatives. Chaos Solitons Fractals, 102:295–304, 2017

    work page 2017

  16. [16]

    I. E. Lagaris, A. Likas, and D. I. Fotiadis. Artificial neural networks for solving ordinary and partial differential equations. IEEE Trans. Neural Netw. , 9(5):987–1000, 1998

    work page 1998

  17. [17]

    Z. Lin, Q. Ma, H. Xie, and X. Yin. Solving time-fractional partial integro-differential equations using tensor neural network. SIAM J. Sci. Comput. , 48(1):C164–C189, 2026

    work page 2026

  18. [18]

    Lischke, G

    A. Lischke, G. Pang, M. A. Gulian, F. Song, C. A. Glusa, X. Zheng, Z. Mao, W. Cai, M. M. Meerschaert, M. Ainsworth, and G. E. Karniadakis. What is the fractional Laplacian? A comparative review with new results. J. Comput. Phys. , 404:109009, 2020

    work page 2020

  19. [19]

    L. Ma, R. Li, F. Zeng, L. Guo, and G. E. Karniadakis. Bi-orthogonal fPINN: a physics- informed neural network method for solving time-dependent stochastic fractional PDEs. Commun. Comput. Phys. , 34(4):1133–1176, 2023

    work page 2023

  20. [20]

    M. M. Meerschaert, H.-P. Scheffler, and C. Tadjeran. Finite difference methods for two- dimensional fractional dispersion equation. J. Comput. Phys. , 211:249–261, 2006

    work page 2006

  21. [21]

    G. Pang, L. Lu, and G. E. Karniadakis. fPINNs: Fractional physics-informed neural networks. SIAM J. Sci. Comput. , 41(4):A2603–A2626, 2019

    work page 2019

  22. [22]

    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. 16

    work page 2019

  23. [23]

    J. Shen, T. Tang, and L.-L. Wang. Spectral Methods: Algorithms, Analysis and Appli- cations, volume 41 of Springer Series in Computational Mathematics . Springer, Berlin, Heidelberg, 2011

    work page 2011

  24. [24]

    Sheng, B

    C. Sheng, B. Su, and C. Xu. Efficient Monte Carlo method for integral fractional Laplacian in multiple dimensions. SIAM J. Numer. Anal. , 61(5):2035–2061, 2023

    work page 2035

  25. [25]

    Sheng, L.-L

    C. Sheng, L.-L. Wang, H. Chen, and H. Li. Fast implementation of FEM for integral fractional Laplacian on rectangular meshes. Commun. Comput. Phys. , 36(3):673–710, 2024

    work page 2024

  26. [26]

    Wang and G

    S. Wang and G. E. Karniadakis. GMC-PINNs: A new general Monte Carlo PINNs method for solving fractional partial differential equations on irregular domains. Comput. Methods Appl. Mech. Engrg. , 429:117189, 2024

    work page 2024

  27. [27]

    Zhang, D

    T. Zhang, D. Zhang, S. Shi, and Z. Guo. Spectral-fPINNs: spectral method based frac- tional physics-informed neural networks for solving fractional partial differential equations. Nonlinear Dyn. , pages 1–24, 2025. 17

    work page 2025