Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

A Data-Driven Framework for Discovering Fractional Differential Equations in Complex Systems

T0 review · 3 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read A stepwise data-driven method discovers explicit fractional differential equations directly from sparse, noisy data.

desk verdict Clean FADE recovery and a useful DNN+quadrature pipeline, but the Lévy validation case has a drift inconsistency that undercuts the general robustness claim. read the letter →

arxiv 2412.03970 v2 pith:ND5MNKR7 submitted 2024-12-05 physics.comp-ph cs.AI

classification physics.comp-phcs.AI MSC 34A0835R1165D3068T07
keywords fractionaldifferentialequationsknowledgediscoverysparseregressionGauss-JacobiquadraturemachinelearninganomalousdiffusionLévymotion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to show that explicit fractional differential equations (FDEs) can be discovered directly from sparse, noisy measurements, without knowing the structure of the equation in advance. Its proposed pipeline uses a deep neural network to denoise and interpolate the data, Gauss-Jacobi quadrature to evaluate the singular convolution integrals that define fractional derivatives, and an alternating routine that combines sparse regression for the linear coefficients with global optimization for the fractional orders. The authors validate the pipeline on synthetic anomalous-diffusion data, experimental creep data from frozen soils, and single-particle Lévy trajectories, recovering equation structures that match the known models in all three cases. If the claim holds, fractional models of memory and nonlocal effects can be obtained from observations in fields where first-principles derivation is impractical.

What carries the argument

The load-bearing machinery is a candidate library $\Theta(\beta)$ built from fractional derivatives, combined with Gauss-Jacobi quadrature and an alternating optimization scheme. Gauss-Jacobi quadrature is a numerical integration rule for integrands with endpoint singularities; here it evaluates the convolution integral in the Caputo or Riemann-Liouville fractional derivative using a small fixed set of weighted function values from the DNN-reconstructed field, replacing dense meshes. The alternating scheme is necessary because fractional orders enter the library nonlinearly: STRidge (sequential threshold ridge regression) determines the sparse linear coefficient vector $\xi$, while a global optimizer (differential evolution for multi-parameter problems and a derivative-free optimizer for single-parameter problems) adjusts the fractional orders $\alpha$ and $\beta$. The loss surface is non-convex and discontinuous because the number of nonzero coefficients changes with the fractional orders, which is why the authors use global rather than gradient-based optimization.

What would settle it

Generate data from a known FDE whose correct term is deliberately absent from the candidate library, run the full pipeline, and check whether the recovered equation is structurally wrong; additionally, on the clean FADE benchmark, compare the DNN-plus-quadrature fractional derivative against a high-resolution spectral derivative of the true solution at the same points, since large pointwise error with a correct recovered equation would indicate accidental cancellation rather than reliable estimation.

Watch

Extended reading notes

Core claim

The paper's central claim is that the combination of a DNN surrogate, Gauss-Jacobi quadrature, and alternating sparse/global optimization recovers explicit FDEs from data. Concretely, the reported discoveries include the fractional Kelvin model $\varepsilon_t^{(0.374)} = -2.402\varepsilon + 8.125$ for clay creep, the fractional advection-diffusion equation $c_t^{(0.790)} = -1.006 c_x + 0.501 c_x^{(1.720)}$ from clean synthetic transport data, and the space-fractional diffusion equation $c_t^{(0.99)} = 0.48 c_{xx}^{(1.84)}$ from $\alpha$-stable single-particle trajectories. The method also learns an integer-order time derivative as a limit case, recovering the order as 0.99 when the true order is 1. These results are offered as evidence that interpretable, closed-form FDEs can be identified without prior structural knowledge.

Load-bearing premise

The load-bearing premise is that the true equation is a linear combination of terms from a hand-chosen candidate library, and that the DNN plus five-node Gauss-Jacobi quadrature produces accurate enough fractional derivatives from the available sparse, noisy data.

Editorial extensions

If this is right

  • On the synthetic FADE benchmark, the method recovers the true structure with fractional orders within about 0.02 to 0.16 of the ground truth under clean, 5% noise, and 25% noise conditions.
  • On experimental frozen-soil creep data, it learns a fractional Kelvin model without being given the equation form, with lower parameter error for clay than for silt.
  • On $\alpha$-stable Lévy trajectories, it identifies a space-fractional diffusion equation with a nearly integer time order, showing that memoryless processes fall within its scope.
  • Compared with an integer-order-only deep-learning PDE discovery method, the fractional library appears necessary: the integer-only method captures only local advection and dispersion and omits the fractional term.
  • A regularization sweep indicates that $\lambda = 10^{-3}$ yields the correct FADE structure across all tested noise levels, while too small $\lambda$ adds redundant terms and too large $\lambda$ drops the fractional derivative term.

Reading between the lines

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

  • Beyond the paper's tests, the same alternating scheme could be pointed at variable-order fractional equations if the candidate library were expanded to include order-varying terms; the paper does not attempt this.
  • The five-node Gauss-Jacobi quadrature is a fixed choice, so an explicit node-count sensitivity study would separate quadrature error from data noise as the limit on recovered fractional orders.
  • The core reliance on a closed candidate library could be lifted by coupling the pipeline with open-form symbolic search, a direction the conclusion sketches but does not implement here.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a stepwise data-driven framework for discovering explicit fractional differential equations (FDEs) from sparse and noisy data. The pipeline consists of (i) a deep neural network that denoises and reconstructs the field and supplies automatic differentiation, (ii) Gauss-Jacobi quadrature for evaluating fractional derivatives with few auxiliary nodes, and (iii) an alternating optimization scheme in which STRidge estimates sparse linear coefficients and a global optimizer (Differential Evolution or Powell) adjusts the fractional orders. The method is validated on three cases: experimental creep data of frozen soils (fractional Kelvin model), synthetic space-time fractional advection-diffusion data under clean, 5% noise, and 25% noise, and single-particle trajectories modeled by an alpha-stable Lévy motion. The authors also compare with DL-PDE and analyze the effect of the regularization parameter lambda.

Significance. If the claims hold, the framework is a useful contribution to data-driven discovery of fractional PDEs, especially the combination of DNN-based reconstruction with G-J quadrature and alternating optimization of linear coefficients and nonlinear fractional orders. The paper ships code and data, provides a clear algorithmic description, and includes a comparison with DL-PDE and a regularization sensitivity study. The FADE recoveries in Table 2 are clean and match the generating equation at all three noise levels, and the frozen-soil case recovers the Kelvin-model structure from experimental data. However, the Lévy validation has an internal inconsistency that undermines one of the three headline demonstrations, and the 'robustness' claim is supported only by single runs with model selection on the same test data. These issues need to be resolved before the central claim is fully established.

major comments (3)
  1. [Section 3.3, Table 3, Eq. (25), Appendix C] The Lévy validation is internally inconsistent. The text states that the stable time series is S_{1.8}(1, 0.66, -0.32), and Appendix C establishes that a stable law S_alpha(1, (Dt|cos(pi alpha/2)|)^{1/alpha}, vt) corresponds to the solution of the space-fractional advection-diffusion equation (25), which contains the advection term -v c_x. The nonzero location parameter -0.32 therefore implies a drift. Nevertheless, Table 3 reports the ground truth as c_t = 0.5 c_xx^{1.8} and the learned equation as c_t^(0.99) = 0.48 c_xx^(1.84), with no first-order spatial term. If the simulation included the stated drift, the sparse regression discarded a real term; if the drift was absent, the experimental setup contradicts the stated stable-law parameters. In either case, Section 3.3 does not demonstrate recovery of the full FDE structure, and the claim of robustness across the three demonstrations is weakened. Please clarify the generator, correct the ground truth, and rerun the experiment accordingly.
  2. [Section 3.2, Tables 2 and 6] The claim of robustness across noise levels is not statistically supported because each configuration is run once. For each noise level, a single realization is generated, a single DNN is trained, and a single sparse-regression path is reported, with no standard errors, random seeds, or repeated trials. In addition, lambda = 10^-3 is selected on the same noisy test cases in Table 6 and then used to produce the results in Table 2, which is a form of test-set selection. The consistent recoveries are encouraging, but the robustness claim needs repeated trials (or at least a sensitivity analysis over seeds and data realizations) and an independent model-selection procedure to rule out favorable chance outcomes.
  3. [Section 2.2, Eq. (6); Section 5] The method can only discover equations whose terms lie in the predefined candidate library, yet the abstract and introduction describe the framework as discovering FDEs 'without prior knowledge' of structure. In all three validations, the true terms are placed inside the library by construction. The authors acknowledge the closed-library limitation in Section 5, but the wording of the central claim overstates the degree to which the method is free of prior structural assumptions. Please qualify the claim explicitly, for example by stating that discovery is relative to the chosen library and that the library is a form of prior knowledge about which terms may appear.
minor comments (4)
  1. [Table 5 caption] The caption says 'Comparison of FADE results between our method and DL-PDE', but the benchmark in that table is the classical advection-diffusion equation with integer-order diffusion. Retitle the caption to avoid confusion.
  2. [Tables 1-3] The 'Error' and 'Error^2' columns are not defined. State whether the reported values are L2 errors, relative errors, or mean squared errors, and specify the data split over which they are computed.
  3. [Introduction and appendices] There are several wording issues: 'lost function' should be 'loss function', and 'Appendix Appendix B' and 'Appendix Appendix C' should be 'Appendix B' and 'Appendix C'.
  4. [Section 2.3 and Appendix B] The choice of five G-J quadrature nodes is stated without a convergence or sensitivity study. A brief numerical test in Appendix B showing how accuracy varies with the number of nodes would support this choice.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the central derivation compares learned equations against independent generating parameters, and the caveats (prior-fit 'ground truth' in Section 3.1 and the omitted drift in Section 3.3) are correctness concerns, not circularity.

full rationale

The derivation chain is self-contained: the framework takes raw observations, trains a DNN surrogate (Eqs. 12-14), computes fractional-derivative library entries via automatic differentiation plus Gauss-Jacobi quadrature (Eqs. 15-16 and B.1-B.2), and then alternates STRidge for linear coefficients with global optimization for fractional orders (Algorithm 1, Eq. 17). The discovered equations in Tables 1-3 are compared against independent generating values (e.g., FADE parameters alpha=0.8, beta=1.7, v=1, D=0.5 in Section 3.2) or against a previously published constitutive model (Section 3.1), not against quantities fed into the algorithm. Self-citations such as [16], [19], and [26] are contextual and not load-bearing; no uniqueness theorem or ansatz is imported from the authors' prior work to force the outcome. Two non-circular caveats should be weighed in correctness review: (1) in Section 3.1 the 'ground truth' is itself a model fitted to the same experimental data in [59], so agreement demonstrates consistency with a prior fit rather than with an external physical law; (2) in Section 3.3, the generating stable law S_1.8(1, 0.66, -0.32) has a nonzero drift mu=-0.32, which by the paper's own Eq. (24) and Appendix C implies an advection term -v c_x, yet Table 3 lists the ground truth as c_t = 0.5 c_xx^(1.8) with no first-order spatial term. This internal inconsistency undermines the Levy validation case but is not circularity, since the learned equation is not equivalent by construction to the input data or to a fitted parameter.

Assumptions & free parameters 4 free parameters · 7 assumptions · 0 invented entities

Central claim rests on a closed library, DNN surrogate accuracy, G-J quadrature accuracy, and non-convex optimization success; no new physical entities are introduced.

free parameters (4)
  • Regularization coefficient λ = 10^-3 (chosen via Table 6)
    Controls sparsity/parsimony; selected by inspecting the FADE benchmark across noise levels, then used for all other experiments.
  • Number of Gauss-Jacobi quadrature nodes N = 5
    Paper states five nodes balance efficiency and accuracy (Section 2.3); accuracy is not validated per case.
  • STRidge threshold and ridge hyperparameters = not specified
    STRidge requires setting threshold and iterations; the manuscript does not report these values, leaving a hidden tuning choice.
  • DNN architecture and training details = varies by case study
    Layer counts, activations, optimizers are chosen per dataset; these choices affect reconstructed derivatives but are not systematically optimized.
assumptions (7)
  • domain assumption The true FDE's terms are contained in a manually predefined candidate library Θ (Equation 6).
    Sections 2.2 and 3.2 build Θ from hand-chosen monomials and derivative terms; if a necessary term is missing, STRidge cannot represent it, so all case studies are closed-world tests.
  • domain assumption DNN reconstruction of sparse noisy data yields accurate function values and derivatives.
    Section 2.3 assumes the trained DNN is a faithful surrogate and that auto-differentiation of the DNN gives correct integer derivatives used inside fractional derivatives; no error analysis is provided.
  • domain assumption Gauss-Jacobi quadrature with five nodes accurately approximates fractional derivatives.
    Appendix B validates on a single test function; the paper assumes the same accuracy for the candidate library terms across all case studies.
  • domain assumption Alternating STRidge and global optimization (DE/PA) reaches the global optimum of a non-convex, discontinuous loss.
    Algorithm 1 and Figure 2b note the loss is non-convex and discontinuous; no convergence guarantee or warm-start analysis is given.
  • domain assumption For the periodic FADE, the left Caputo derivative over [0,x] matches the FFT-generated fractional derivative.
    Synthetic data is generated with FFT and periodic boundary conditions (Section 3.2), while the library computes fractional derivatives via integrals from the origin (Sections 2.3 and 2.4); this boundary/domain mismatch is not discussed.
  • domain assumption Riemann-Liouville and Caputo derivatives are interchangeable on the computational domain (Equation 5 with negligible boundary terms).
    Invoked after Equation 5 to justify using Caputo form for space derivatives; holds only for sufficiently large domains and small boundary conditions.
  • standard math The solution of the space-fractional advection-diffusion equation is the scaling limit of the α-stable Lévy motion.
    Used in Section 3.3 and Appendix C to connect the random walk with α-stable jumps to the FDE that is later discovered; this is a standard generalized central limit theorem result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Data-Driven Framework for Discovering Fractional Differential Equations in Complex Systems." pith.science (2026). https://pith.science/paper/ND5MNKR7

@misc{pith2026241203970,
  author       = {Pith},
  title        = {Pith review of: A Data-Driven Framework for Discovering Fractional Differential Equations in Complex Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ND5MNKR7}},
  note         = {Machine review of arXiv:2412.03970}
}
read the original abstract

In complex physical systems, conventional differential equations often fall short in capturing non-local and memory effects, as they are limited to local dynamics and integer-order interactions. This study introduces a stepwise data-driven framework for discovering fractional differential equations (FDEs) directly from data. FDEs, known for their capacity to model non-local dynamics with fewer parameters than integer-order derivatives, can represent complex systems with long-range interactions. Our framework applies deep neural networks as surrogate models for denoising and reconstructing sparse and noisy observations while using Gaussian-Jacobi quadrature to handle the challenges posed by singularities in fractional derivatives. To optimize both the sparse coefficients and fractional order, we employ an alternating optimization approach that combines sparse regression with global optimization techniques. We validate the framework across various datasets, including synthetic anomalous diffusion data, experimental data on the creep behavior of frozen soils, and single-particle trajectories modeled by L\'{e}vy motion. Results demonstrate the framework's robustness in identifying the structure of FDEs across diverse noise levels and its capacity to capture integer-order dynamics, offering a flexible approach for modeling memory effects in complex systems.

Figures

Figures reproduced from arXiv: 2412.03970 by the authors.

Figure 1
Figure 1. Workflow for data processing and generating the library of derivative terms. MaxIt in this figure represents the predefined maximum [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. (a) Evolution of sparse coefficients estimated via our alternating optimization approach. The y-coordinate represents the best sparse coefficients ξ estimated by STRidge under fixed fractional orders, the x-coordinate represents the iterations of the global optimization algorithm. (b) Evolution of the loss function (17) with fractional orders. The specific example presented in this figure involves a fractional advec… view at source ↗
Figure 3
Figure 3. Reconstruction of experimental data for clay and silt. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The upper panel illustrates the spatiotemporal distribution of synthetic data, governed by a fractional advection-di [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Single particle trajectory of Lévy motions. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: (a) Validation of 100 sampled particle trajectories generated by recovered dynamics. Each of the 100 green lines is a sample path of [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. On the post-hoc Evaluation of PDE Discovery: A Multifaceted Challenge of Scientific Advancement

    cs.LG 2026-07 accept novelty 5.5 of 10

    The first broad taxonomy of post-hoc PDE-discovery metrics shows single scores mislead, and recommends multi-criteria evaluation with OOD and physics checks.

Reference graph

Works this paper leans on

78 extracted references · 66 canonical work pages · cited by 1 Pith paper

  1. [1]

    Deep learning and process understanding for data-driven earth system science

    Markus Reichstein, Gustau Camps-Valls, Bjorn Stevens, Martin Jung, Joachim Denzler, Nuno Carvalhais, and F Prabhat. Deep learning and process understanding for data-driven earth system science. Nature, 566(7743):195–204, 2019

  2. [2]

    Data-driven methods for flow and transport in porous media: A review

    Guang Yang, Ran Xu, Yusong Tian, Songyuan Guo, Jingyi Wu, and Xu Chu. Data-driven methods for flow and transport in porous media: A review. International Journal of Heat and Mass Transfer, 235:126149, 2024

  3. [3]

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

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

  4. [4]

    Promising directions of machine learning for partial di fferential equations

    Steven L Brunton and J Nathan Kutz. Promising directions of machine learning for partial di fferential equations. Nature Computational Science, pages 1–12, 2024. Publisher: Nature Publishing Group US New York

  5. [5]

    Uncovering terrain-precipitation equation with inter- pretable AI: Towards future climate projection

    Hao Xu, Yuntian Chen, Zhenzhong Zeng, Nina Li, Jian Li, and Dongxiao Zhang. Uncovering terrain-precipitation equation with inter- pretable AI: Towards future climate projection. Nexus, 2024

  6. [6]

    Genetic programming as a means for programming computers by natural selection

    John R Koza. Genetic programming as a means for programming computers by natural selection. Statistics and Computing, 4:87–112, 1994

  7. [7]

    Automated reverse engineering of nonlinear dynamical systems

    Josh Bongard and Hod Lipson. Automated reverse engineering of nonlinear dynamical systems. Proceedings of the National Academy of Sciences, 104(24):9943–9948, 2007. Publisher: National Acad Sciences

  8. [8]

    Distilling free-form natural laws from experimental data

    Michael Schmidt and Hod Lipson. Distilling free-form natural laws from experimental data. Science, 324(5923):81–85, 2009

Show all 78 references
  1. [9]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the National Academy of Sciences, 113(15):3932–3937, 2016

  2. [10]

    Data-driven discovery of partial di fferential equations

    Samuel H Rudy, Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Data-driven discovery of partial di fferential equations. Science Advances, 3(4):e1602614, 2017. Publisher: American Association for the Advancement of Science

  3. [11]

    Machine learning subsurface flow equations from data

    Haibin Chang and Dongxiao Zhang. Machine learning subsurface flow equations from data. Computational Geosciences, 23(5):895–910, 2019

  4. [12]

    Weak SINDy for partial di fferential equations

    Daniel A Messenger and David M Bortz. Weak SINDy for partial di fferential equations. Journal of Computational Physics, 443:110525, 2021

  5. [13]

    Ensemble-SINDy: Robust sparse model discovery in the low-data, high-noise limit, with active learning and control

    Urban Fasel, J Nathan Kutz, Bingni W Brunton, and Steven L Brunton. Ensemble-SINDy: Robust sparse model discovery in the low-data, high-noise limit, with active learning and control. Proceedings of the Royal Society A, 478(2260):20210904, 2022

  6. [14]

    Robust data-driven dynamic model discovery of industrial robots with spatial manipulation capability using simple trajectory

    Mohamed Omar, Ke Wang, Dai Kun, Ruifeng Li, and Ahmed Asker. Robust data-driven dynamic model discovery of industrial robots with spatial manipulation capability using simple trajectory. Nonlinear Dynamics, 112(11):9155–9177, 2024

  7. [15]

    Discovering governing equation from data for multi-stable energy harvester under white noise

    Yanxia Zhang, Jinqiao Duan, Yanfei Jin, and Yang Li. Discovering governing equation from data for multi-stable energy harvester under white noise. Nonlinear Dynamics, 106(4):2829–2840, 2021

  8. [16]

    Dl-pde: Deep-learning based data-driven discovery of partial di fferential equations from discrete and noisy data

    Hao Xu, Haibin Chang, and Dongxiao Zhang. Dl-pde: Deep-learning based data-driven discovery of partial di fferential equations from discrete and noisy data. arXiv preprint arXiv:1908.04463, 2019

  9. [17]

    DeepMoD: Deep learning for model discovery in noisy data.Journal of Computational Physics, 428:109985, 2021

    Gert-Jan Both, Subham Choudhury, Pierre Sens, and Remy Kusters. DeepMoD: Deep learning for model discovery in noisy data.Journal of Computational Physics, 428:109985, 2021

  10. [18]

    Integration of knowledge and data in machine learning

    Yuntian Chen and Dongxiao Zhang. Integration of knowledge and data in machine learning. arXiv preprint arXiv:2202.10337, 2022

  11. [19]

    Robust discovery of partial di fferential equations in complex situations

    Hao Xu and Dongxiao Zhang. Robust discovery of partial di fferential equations in complex situations. Physical Review Research , 3(3):033270, 2021

  12. [20]

    Metzler and J

    R. Metzler and J. Klafter. The random walk’s guide to anomalous di ffusion: a fractional dynamics approach. Physics Reports, 339(1):1– 77, 2000

  13. [21]

    Mechanisms, upscaling, and prediction of anomalous dispersion in heterogeneous porous media

    Alessandro Comolli, Vivien Hakoun, and Marco Dentz. Mechanisms, upscaling, and prediction of anomalous dispersion in heterogeneous porous media. Water Resources Research, 55(10):8197–8222, 2019

  14. [22]

    Data-driven identification of parametric partial di fferential equa- tions

    Samuel Rudy, Alessandro Alla, Steven L Brunton, and J Nathan Kutz. Data-driven identification of parametric partial di fferential equa- tions. SIAM Journal on Applied Dynamical Systems, 18(2):643–660, 2019

  15. [23]

    Wheatcraft and Scott W Tyler

    Stephen W. Wheatcraft and Scott W Tyler. An explanation of scale-dependent dispersivity in heterogeneous aquifers using concepts of fractal geometry. Water Resources Research, 24(4):566–578, 1988. Publisher: Wiley Online Library

  16. [24]

    A review and numerical assessment of the random walk particle tracking method

    Peter Salamon, Daniel Fernàndez-Garcia, and J Jaime Gómez-Hernández. A review and numerical assessment of the random walk particle tracking method. Journal of Contaminant Hydrology, 87(3-4):277–305, 2006

  17. [25]

    Deep-learning based discovery of partial differential equations in integral form from sparse and noisy data

    Hao Xu, Dongxiao Zhang, and Nanzhe Wang. Deep-learning based discovery of partial differential equations in integral form from sparse and noisy data. Journal of Computational Physics, 445:110592, 2021

  18. [26]

    Symbolic genetic algorithm for discovering open-form partial differential equations (SGA-PDE)

    Yuntian Chen, Yingtao Luo, Qiang Liu, Hao Xu, and Dongxiao Zhang. Symbolic genetic algorithm for discovering open-form partial differential equations (SGA-PDE). Physical Review Research, 4(2):023174, 2022

  19. [27]

    The data-driven discovery of partial di fferential equations by symbolic genetic algorithm

    Shifei Sun, Shifang Tian, Yuduo Wang, and Biao Li. The data-driven discovery of partial di fferential equations by symbolic genetic algorithm. Nonlinear Dynamics, 112(22):19871–19885, 2024

  20. [28]

    DISCOVER: Deep identification of symbolically concise open-form partial differential equations via enhanced reinforcement learning

    Mengge Du, Yuntian Chen, and Dongxiao Zhang. DISCOVER: Deep identification of symbolically concise open-form partial differential equations via enhanced reinforcement learning. Physical Review Research, 6(1):013182, 2024

  21. [29]

    Physics-constrained robust learning of open-form pdes from limited and noisy data

    Mengge Du, Longfeng Nie, Siyu Lou, Yuntian Chenc, and Dongxiao Zhang. Physics-constrained robust learning of open-form pdes from limited and noisy data. arXiv preprint arXiv:2309.07672, 2023

  22. [30]

    Llm4ed: Large language models for automatic equation discovery

    Mengge Du, Yuntian Chen, Zhongzheng Wang, Longfeng Nie, and Dongxiao Zhang. Llm4ed: Large language models for automatic equation discovery. arXiv preprint arXiv:2405.07761, 2024. 18

  23. [31]

    Physics-informed deep neural networks for learning parameters and constitutive relationships in subsurface flow problems

    Alexandre M Tartakovsky, C Ortiz Marrero, Paris Perdikaris, Guzel D Tartakovsky, and David Barajas-Solano. Physics-informed deep neural networks for learning parameters and constitutive relationships in subsurface flow problems. Water Resources Research, 56(5):e2019WR026731, 2020

  24. [32]

    Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems using the glue methodology

    Keith Beven and Jim Freer. Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems using the glue methodology. Journal of Hydrology, 249(1-4):11–29, 2001

  25. [33]

    Field study of dispersion in a heterogeneous aquifer: 2

    E Eric Adams and Lynn W Gelhar. Field study of dispersion in a heterogeneous aquifer: 2. spatial moments analysis. Water Resources Research, 28(12):3293–3307, 1992

  26. [34]

    Cushman, Lynn S

    John H. Cushman, Lynn S. Bennethum, and Bill X Hu. A primer on upscaling tools for porous media. Advances in Water Resources, 25(8-12):1043–1067, 2002. Publisher: Elsevier

  27. [35]

    Shlomo P. Neuman. Universal scaling of hydraulic conductivities and dispersivities in geologic media. Water Resources Research, 26(8):1749–1758, 1990

  28. [36]

    Theory of solute transport by groundwater

    Gedeon Dagan. Theory of solute transport by groundwater. Annual Review of Fluid Mechanics, 19(1):183–213, 1987

  29. [37]

    Modeling non-fickian transport in geological formations as a continuous time random walk

    Brian Berkowitz, Andrea Cortis, Marco Dentz, and Harvey Scher. Modeling non-fickian transport in geological formations as a continuous time random walk. Reviews of Geophysics, 44(2), 2006

  30. [38]

    Application of a fractional advection-dispersion equation

    David A Benson, Stephen W Wheatcraft, and Mark M Meerschaert. Application of a fractional advection-dispersion equation. Water Resources Research, 36(6):1403–1412, 2000

  31. [39]

    Multiple-rate mass transfer for modeling di ffusion and surface reactions in media with pore-scale heterogeneity

    Roy Haggerty and Steven M Gorelick. Multiple-rate mass transfer for modeling di ffusion and surface reactions in media with pore-scale heterogeneity. Water Resources Research, 31(10):2383–2400, 1995

  32. [40]

    Theory and applications of fractional di fferential equations, volume 204

    Anatoli ˘ı Aleksandrovich Kilbas, Hari M Srivastava, and Juan J Trujillo. Theory and applications of fractional di fferential equations, volume 204. elsevier, Amsterdam, 2006

  33. [41]

    Podlubny

    I. Podlubny. Fractional Differential Equations. Academic press, New York, 1999

  34. [42]

    Stochastic models for fractional calculus , volume 43

    Mark M Meerschaert and Alla Sikorskii. Stochastic models for fractional calculus , volume 43. Walter de Gruyter GmbH & Co KG, Berlin, Germany, 2019

  35. [43]

    Fractional partial differential equations and their numerical solutions

    Boling Guo, Xueke Pu, and Fenghui Huang. Fractional partial differential equations and their numerical solutions . World Scientific, Singapore, 2015

  36. [44]

    A new collection of real world applications of fractional calculus in science and engineering

    HongGuang Sun, Yong Zhang, Dumitru Baleanu, Wen Chen, and YangQuan Chen. A new collection of real world applications of fractional calculus in science and engineering. Communications in Nonlinear Science and Numerical Simulation, 64:213–231, 2018

  37. [45]

    A space fractional constitutive equation model for non-newtonian fluid flow

    HongGuang Sun, Yong Zhang, Song Wei, Jianting Zhu, and Wen Chen. A space fractional constitutive equation model for non-newtonian fluid flow. Communications in Nonlinear Science and Numerical Simulation, 62:409–417, 2018

  38. [46]

    Generalized viscoelastic models: their fractional equations with solutions

    H Schiessel, R Metzler, A Blumen, and TF Nonnenmacher. Generalized viscoelastic models: their fractional equations with solutions. Journal of Physics A: Mathematical and General, 28(23):6567, 1995

  39. [47]

    Fractional calculus and continuous-time finance

    Enrico Scalas, Rudolf Gorenflo, and Francesco Mainardi. Fractional calculus and continuous-time finance. Physica A: Statistical Me- chanics and its Applications, 284(1-4):376–384, 2000

  40. [48]

    Using gauss-jacobi quadrature rule to improve the accuracy of fem for spatial fractional problems

    Zongze Yang, Jungang Wang, Zhanbin Yuan, and Yufeng Nie. Using gauss-jacobi quadrature rule to improve the accuracy of fem for spatial fractional problems. Numerical Algorithms, pages 1–23, 2022

  41. [49]

    fpinns: Fractional physics-informed neural networks

    Guofei Pang, Lu Lu, and George Em Karniadakis. fpinns: Fractional physics-informed neural networks. SIAM Journal on Scientific Computing, 41(4):A2603–A2626, 2019

  42. [50]

    Machine learning of space-fractional di fferential equations

    Mamikon Gulian, Maziar Raissi, Paris Perdikaris, and George Karniadakis. Machine learning of space-fractional di fferential equations. SIAM Journal on Scientific Computing, 41(4):A2485–A2509, 2019

  43. [51]

    Coelho, M.Fernanda P

    C. Coelho, M.Fernanda P. Costa, and Luís L. Ferrás. Tracing footprints: Neural networks meet non-integer order di fferential equations for modelling systems with memory. In The Second Tiny Papers Track at ICLR 2024, 2024

  44. [52]

    Neural fractional di fferential equations

    Cecılia Coelho, M Fernanda P Costa, and Luis L Ferrás. Neural fractional di fferential equations. Applied Mathematical Modelling , 144:116060, 2025

  45. [53]

    Data-driven discovery of time fractional di fferential equations

    Abhishek Kumar Singh, Mani Mehra, and Anatoly A Alikhanov. Data-driven discovery of time fractional di fferential equations. In International Conference on Computational Science, pages 56–63. Springer, 2022

  46. [54]

    A new perspective for scientific modelling: Sparse reconstruction- based approach for learning time-space fractional di fferential equations

    Yash Vats, Mani Mehra, Dietmar Oelz, and Abhishek Kumar Singh. A new perspective for scientific modelling: Sparse reconstruction- based approach for learning time-space fractional di fferential equations. Journal of Computational and Nonlinear Dynamics , 19(12), 2024

  47. [55]

    Gauss-Jacobi-type quadrature rules for fractional directional integrals

    Guofei Pang, Wen Chen, and Kam-Yim Sze. Gauss-Jacobi-type quadrature rules for fractional directional integrals. Computers & Mathematics with Applications, 66(5):597–607, 2013

  48. [56]

    Physics-informed learning of governing equations from scarce data

    Zhao Chen, Yang Liu, and Hao Sun. Physics-informed learning of governing equations from scarce data. Nature Communications, 12(1):6136, 2021

  49. [57]

    Discovery of subdi ffusion problem with noisy data via deep learning

    Xingjian Xu and Minghua Chen. Discovery of subdi ffusion problem with noisy data via deep learning. Journal of Scientific Computing, 92(1):23, 2022

  50. [58]

    Deep hidden physics models: Deep learning of nonlinear partial di fferential equations

    Maziar Raissi. Deep hidden physics models: Deep learning of nonlinear partial di fferential equations. Journal of Machine Learning Research, 19(25):1–24, 2018

  51. [59]

    Particle swarm fractional order derivative model of artificial frozen soil creep properties

    Junhao Chen, Zhaoming Yao, Ying Xu, and Houliang Wang. Particle swarm fractional order derivative model of artificial frozen soil creep properties. Journal of China Coal Society, 38(10):1763–1768, 2013

  52. [60]

    Time and space nonlocalities underlying fractional-derivative models: Distinction and literature review of field applications

    Yong Zhang, David A Benson, and Donald M Reeves. Time and space nonlocalities underlying fractional-derivative models: Distinction and literature review of field applications. Advances in Water Resources, 32(4):561–581, 2009

  53. [61]

    Data-driven discovery of governing equations for fluid dynamics based on molecular simulation

    Jun Zhang and Wenjun Ma. Data-driven discovery of governing equations for fluid dynamics based on molecular simulation. Journal of Fluid Mechanics, 892:A5, 2020

  54. [62]

    Limit distributions for sums of independent random variables, volume 233

    Boris Vladimirovich Gnedenko, Andre ˘ı Nikolaevich Kolmogorov, Joseph L Doob, and Pao-Lu Hsu. Limit distributions for sums of independent random variables, volume 233. Addison-wesley Reading, MA, Reading, MA, 1968

  55. [63]

    Fractional dispersion, Lévy motion, and the made tracer tests

    David A Benson, Rina Schumer, Mark M Meerschaert, and Stephen W Wheatcraft. Fractional dispersion, Lévy motion, and the made tracer tests. Transport in Porous Media, 42(1):211–240, 2001

  56. [64]

    Théorie de l’addition des variables aléatoires

    Paul Lévy. Théorie de l’addition des variables aléatoires. Gauthier-Villars, Paris, 1954

  57. [65]

    DLGA-PDE: Discovery of pdes with incomplete candidate library via combination of deep learning and genetic algorithm

    Hao Xu, Haibin Chang, and Dongxiao Zhang. DLGA-PDE: Discovery of pdes with incomplete candidate library via combination of deep learning and genetic algorithm. Journal of Computational Physics, 418:109584, 2020

  58. [66]

    Neural fractional di fferential equations: Optimising the order of the fractional 19 derivative

    Cecília Coelho, M Fernanda P Costa, and Luís L Ferrás. Neural fractional di fferential equations: Optimising the order of the fractional 19 derivative. Fractal and Fractional, 8(9):529, 2024

  59. [67]

    Unleashing the potential of fractional calculus in graph neural networks with frond

    Qiyu Kang, Kai Zhao, Qinxu Ding, Feng Ji, Xuhao Li, Wenfei Liang, Yang Song, and Wee Peng Tay. Unleashing the potential of fractional calculus in graph neural networks with frond. arXiv preprint arXiv:2404.17099, 2024

  60. [68]

    Fde-net: A memory-e fficiency densely connected network inspired from fractional-order differential equations for single image super-resolution

    Xiao Zhang, Lei Zhang, Wei Wei, Yuxuan Sun, Chunna Tian, and Yanning Zhang. Fde-net: A memory-e fficiency densely connected network inspired from fractional-order differential equations for single image super-resolution. Neurocomputing, 600:128143, 2024

  61. [69]

    Optimising neural fractional di fferential equations for performance and efficiency

    Bernd Zimmering, Cecılia Coelho, and Oliver Niggemann. Optimising neural fractional di fferential equations for performance and efficiency. Proceedings of Machine Learning Research, 255(1):23, 2024

  62. [70]

    Neural variable-order fractional differential equation networks

    Wenjun Cui, Qiyu Kang, Xuhao Li, Kai Zhao, Wee Peng Tay, Weihua Deng, and Yidong Li. Neural variable-order fractional differential equation networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 16109–16117, 2025

  63. [71]

    Efficient training of neural fractional-order differential equation via adjoint backpropagation

    Qiyu Kang, Xuhao Li, Kai Zhao, Wenjun Cui, Yanan Zhao, Weihua Deng, and Wee Peng Tay. Efficient training of neural fractional-order differential equation via adjoint backpropagation. In Proceedings of the AAAI Conference on Artificial Intelligence , volume 39, pages 17750–17759, 2025

  64. [72]

    Spectral methods: algorithms, analysis and applications, volume 41

    Jie Shen, Tao Tang, and Li-Lian Wang. Spectral methods: algorithms, analysis and applications, volume 41. Springer Berlin, Heidelberg, Berlin, Heidelberg, Germany, 2011

  65. [73]

    Theory and Numerical Approximations of Fractional Integrals and Derivatives

    Changpin Li and Min Cai. Theory and Numerical Approximations of Fractional Integrals and Derivatives . Society for Industrial and Applied Mathematics, Philadelphia, PA, 2019

  66. [74]

    Calculation of gauss quadrature rules

    Gene H Golub and John H Welsch. Calculation of gauss quadrature rules. Mathematics of computation, 23(106):221–230, 1969

  67. [75]

    Lin and C

    Y . Lin and C. Xu. Finite difference/spectral approximations for the time-fractional diffusion equation. Journal of Computational Physics, 225(2):1533–1552, 2007

  68. [76]

    An introduction to probability theory and its applications, Volume 2 , volume 81

    William Feller. An introduction to probability theory and its applications, Volume 2 , volume 81. John Wiley & Sons, New York, NY , USA, 1991

  69. [77]

    On using random walks to solve the space-fractional advection-dispersion equations

    Yong Zhang, David A Benson, Mark M Meerschaert, and Hans-Peter Sche ffler. On using random walks to solve the space-fractional advection-dispersion equations. Journal of Statistical Physics, 123:89–110, 2006

  70. [78]

    Parameterizations and modes of stable distributions

    John P Nolan. Parameterizations and modes of stable distributions. Statistics & Probability Letters, 38(2):187–195, 1998. 20

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.