Pith. sign in

REVIEW 3 major objections 6 minor 59 references

Multi-Fidelity Machine Learning Applied to Steady Fluid Flows

T0 review · 3 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that a fully-connected network trained on pointwise potential-flow features can predict steady Navier-Stokes fields at nearby designs from as little as one high-fidelity simulation.

desk verdict Useful low-data surrogate pipeline for steady flows, but the geometry-generalization claim needs a coordinate baseline and the lifting-airfoil potential needs a branch-cut specification for reproducibility. read the letter →

arxiv 2501.14870 v1 pith:NVH3SC4D submitted 2025-01-24 physics.flu-dyn cs.NAmath.NAphysics.comp-ph

Kazuko W. Fuchi , Eric M. Wolf , David S. Makhija , Christopher R. Schrock , Philip S. Beran This is my paper · ORCID
classification physics.flu-dyncs.NAmath.NAphysics.comp-ph
keywords machinelearningsmalldatamulti-fidelityCFDinitializationellipticinputfeaturesquadtreesamplingpotentialflowNavier-Stokes
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 introduces a machine-learning method, MF-LEIF, that predicts steady incompressible fluid flows around a body after being trained on as little as one high-fidelity simulation. Instead of feeding Cartesian coordinates to the network, it uses elliptic input features derived from potential flow—the velocity potential, the streamfunction, and the velocity magnitude—so that points in different geometries that play the same flow role are represented similarly. The network learns only the pointwise discrepancy between a cheap potential-flow baseline and the high-fidelity Navier-Stokes solution, and a quadtree sampling scheme concentrates training points near large gradients. The authors show that a model trained on one reference airfoil predicts fields at nearby angles of attack, and that using those predictions to initialize a CFD solver cuts walltime, in the best case to 42 percent of the freestream-initialized run. If the approach scales, it offers a way to reduce the cost of design-space CFD sweeps without large training datasets.

What carries the argument

The load-bearing object is the set of elliptic input features (EIFs): the solutions Φ and Ψ of Laplace equations with slip and no-penetration wall conditions, together with U = |∇Φ|. These replace Cartesian coordinates as the network inputs, so that similar physical locations across designs have similar feature values. The argument is carried by a pointwise discrepancy model: the network output δq is added to a low-fidelity potential-flow baseline qLF, so that q = qLF + δq. The model is trained with an L2 loss and optionally a Sobolev gradient loss, with gradients computed by chain rule through the EIFs. Quadtree adaptive sampling generates concentrated training points near large field gradients, and a Partition-of-Unity (POFU) window function smoothly blends the prediction into freestream conditions outside the training window.

What would settle it

Generate two high-fidelity solutions for the same body at two angles of attack (for instance 0° and 5°) and plot the discrepancy δq against the triple (Φ, Ψ, U) at points where the triples nearly coincide; if substantially different δq values occur at the same feature values, the pointwise map is not single-valued and the method cannot generalize across those designs.

Watch

Extended reading notes

Core claim

The central discovery is that the high-fidelity Navier-Stokes correction to potential flow is, to a good approximation, a pointwise function of three elliptic potential-flow fields: the velocity potential Φ, the streamfunction Ψ, and the velocity magnitude U=|∇Φ|. Because these fields vary smoothly with geometry and angle of attack, a fully-connected network trained on pointwise samples from one reference design can be evaluated at a nearby design without retraining, yielding a flow-field prediction for initialization. The authors demonstrate this for cylinder radius changes and for Joukowski airfoil angle-of-attack changes up to the onset of unsteadiness. They further show that the predicted fields, extended to freestream with a partition-of-unity window, can initialize a finite-element Navier-Stokes solver and reduce convergence walltime while converging to the same lift and drag coefficients.

Load-bearing premise

The load-bearing premise is that the three potential-flow features are sufficient and smoothly varying coordinates for the pointwise discrepancy between potential flow and the true Navier-Stokes solution, so that nearby geometries map to nearby corrections.

Editorial extensions

If this is right

  • A network trained on one high-fidelity solution can provide initial flow fields for nearby design points, turning each new CFD run into a short continuation rather than a cold start.
  • Quadtree sampling with roughly a thousand training points can capture near-body features well enough to initialize boundary-layer-resolving meshes.
  • Combining two reference solutions (for example, angle of attack 0° and 7°) improves prediction accuracy across the interval, suggesting a path to covering a design range with few training runs.
  • ML initialization can reduce walltime on fine meshes, to 42 percent of the freestream-initialized run at 5° angle of attack, while converging to the same force coefficients.
  • Because the inputs are geometry-independent potential-flow fields, the same trained model may transfer to other nearby body shapes without reusing Cartesian-coordinate data.

Reading between the lines

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

  • If the elliptic-feature premise holds, the method should extend to other bodies and parameter changes for which a potential-flow solution is available, such as camber, thickness, or multi-element configurations; this is a direct but untested consequence of the paper's argument.
  • The demonstrated speed-up is for a single flow solver and two target angles; a realistic design-optimization benefit would need repeated sweeps where the model is re-evaluated at each design point, so the gains should be measured end-to-end.
  • A natural next experiment is to compare the same network architecture and data budget with Cartesian-coordinate inputs, to isolate how much of the generalization comes from the elliptic features rather than from the small perturbation size.
  • The coarse-mesh slowdown reported in the paper hints that ML initialization should be paired with a reliability check, such as a quick estimate of prediction error in the boundary layer, before use in production loops.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. This paper introduces MF-LEIF, a multi-fidelity machine learning method for predicting steady external fluid flows. The method uses solutions of Laplace boundary value problems—the velocity potential Φ, streamfunction Ψ, and speed U—as pointwise input features to a fully-connected neural network, instead of Cartesian coordinates. The network is trained to predict the discrepancy between a high-fidelity Navier-Stokes solution and a potential-flow baseline, using data sampled by an adaptive quadtree scheme within a near-body window, with a partition-of-unity extension to freestream conditions. Results are presented for flow around a circular cylinder (radius variation) and a Joukowski airfoil (angle-of-attack variation), including holdout predictions and CFD warm-starting, where an ML-initialized run at AoA=5° takes 42% of the walltime of a freestream-initialized run on the finest mesh (Table 4). The paper also compares uniform, random, and quadtree sampling, and demonstrates the value of Sobolev training for improving boundary-layer predictions.

Significance. If the approach holds up, it offers a low-data alternative to operator-type ML surrogates for CFD initialization, with the attractive property of not requiring explicit design variables or Cartesian coordinates as inputs. The paper includes genuine out-of-sample tests (holdout cylinder radii and airfoil angles of attack), a detailed algorithmic description of data sampling, training-window and POFU extension, and a concrete demonstration of solver acceleration. The claim that elliptic input features provide smooth, geometry-sensitive coordinates is intuitive and potentially impactful for shape design, but the evidence in this manuscript is not yet sufficient to fully substantiate the geometry-generalization claim, and one implementation detail—the branch cut of the multi-valued potential—is missing.

major comments (3)
  1. [Section 2.1 and Appendix B, Eq. (19)] For the lifting airfoil cases at nonzero angle of attack, the velocity potential Φ in Eq. (19) contains a circulation term iΓ/(2π) ln(η+ϵ) with Γ = 4πU0(a+ϵ) sinα, so Φ is multi-valued in the fluid domain. The manuscript does not specify the branch cut used to evaluate Φ pointwise, nor how the resulting discontinuity is treated in the ML input. Because the model is trained at AoA=0 (Γ=0, single-valued Φ) and then applied at AoA=2° and 5° (Γ≠0), the input feature at inference is branch-dependent, so the predictions in Figs. 17-18 and the walltime acceleration in Table 4 are not reproducible without stating the gauge choice. This also undermines the claim in Section 2.1 that the EIFs are smooth, geometry-sensitive coordinates. Please specify the branch cut and its location relative to the training window, discuss the effect on training and inference, or replace the multi-valued Φ by a single-valued feature such as ∇Φ.
  2. [Section 3] The central claim that elliptic input features are preferable to Cartesian coordinates for geometrical generalization is not tested against a Cartesian-coordinate baseline. All experiments use the same multi-fidelity discrepancy formulation with a potential-flow baseline, so the observed holdout accuracy could arise from the additive baseline or the small parametric variation rather than from the choice of input features. A control experiment with the same architecture, sampling, and loss but using (x,y) coordinates as inputs would be needed to attribute the generalization to the EIFs.
  3. [Sections 3.1-3.2] The experiments vary only one scalar parameter (cylinder radius in a ±10% band, airfoil angle of attack in 0°-8°) while keeping the body shape fixed, so they demonstrate interpolation over a design parameter rather than generalization to a new boundary geometry as claimed in the abstract. A test involving a shape change (e.g., a Joukowski airfoil with different thickness or camber) would substantially strengthen the geometry-generalization claim.
minor comments (6)
  1. [Eq. (10)] The transition polynomial is misprinted: 6r^5 − 15r^4 + 1 − r^3 takes the value −9 at r=1 and does not satisfy T(1)=1. The intended C2 smoothstep is presumably 6r^5 − 15r^4 + 10r^3; please correct it.
  2. [Section 3.2, p. 12] The sentence 'The flow is attached up to AoA=5°, at which separation is observed' is contradictory; rephrase as 'attached for AoA below 5°, with separation first appearing at AoA=5°.'
  3. [Figure 14] The holdout error plots in Fig. 14 do not show run-to-run variability, although Fig. 6 reports ranges over five training runs; please add standard deviations or ranges to the holdout curves.
  4. [Table 3] The caption contains a typo ('Relative erros') and does not state the units of walltime; please fix and add (s).
  5. [Algorithm 2] Line 5 writes 'Compute Φ i = Φ i (xi)', which should read 'Compute Φ_i = Φ(x_i)' (and similarly for Ψ and U) to avoid notation confusion.
  6. [Section 2.3] The quadtree refinement and derefinement percentages (R_ref and R_de) are not specified for the reported experiments; please provide the values used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the discrepancy target is independent high-fidelity data, and predictions are evaluated at holdout design points.

full rationale

MF-LEIF's target is the discrepancy δq = q − qLF (Eq. 2), where q is a Parabol Navier-Stokes solution and qLF is the analytic potential-flow baseline of Section 3 (Eqs. 13–15). The EIFs (Φ, Ψ, U) are computed from standard Laplace BVPs (Eqs. 5–6) or Appendix B conformal mappings, independently of the high-fidelity data. The network is trained only on reference cases (cylinder a = 0.05; airfoil AoA = 0°) and the reported predictions are at held-out radii (R = 90%, 110%) and angles (AoA = 2°, 5°), so outputs are not fitted inputs renamed as predictions. The low-fidelity baseline is an additive modeling choice that changes with design point, not a normalization that forces the discrepancy to vanish. Self-citations [18] and [22] supply background and solver verification, respectively, and are not load-bearing for the central generalization claim; no uniqueness theorem is imported. The reviewer-identified branch-cut ambiguity for the lifting-airfoil potential Φ in Eq. (19) is a reproducibility and well-posedness caveat about input construction, not a circular reduction, because the predicted discrepancy still comes from independent high-fidelity data. Hence no circular step is exhibited.

Assumptions & free parameters 5 free parameters · 7 assumptions · 0 invented entities

The free parameters are user-chosen window, sampling, architecture, and transition settings. The axioms include the sufficiency of potential-flow features, trust in the CFD solver, function-approximation capability, and the steady-flow limit. No new physical entities are introduced; EIFs are derived from existing potential-flow solutions.

free parameters (5)
  • Training window size = 30a x 20a for cylinder
    User-specified in Section 2.2; trades concentration of model capacity near the body against clipping of wake and boundary-layer features.
  • Quadtree refinement and derefinement percentages = 20 percent refine, 20 percent derefine after 4 uniform passes
    Section 2.3 and Fig. 5; chosen by hand and controls the training point distribution.
  • Training data point counts = roughly 950 to 1002 per case; 1602 for the initialization model
    Sections 3.1 to 3.3; selected to balance training cost and accuracy, not derived from a principle.
  • Neural network architecture = 10 hidden layers of 100 units, sine activation
    Section 3; fixed hyperparameters with no reported sensitivity study.
  • POFU transition widths = sx = 0.1 lx, sy = 0.1 ly
    Section 3.3 and Fig. 17; chosen by hand for the initialization demonstrations.
assumptions (7)
  • ad hoc to paper Potential-flow solutions to Laplace BVPs provide smooth, geometry-sensitive input features sufficient for pointwise regression of high-fidelity discrepancies.
    Section 2.1; central modeling premise, not proven for arbitrary geometries.
  • domain assumption The Parabol solver produces ground-truth steady Navier-Stokes solutions.
    Section 3; solver verification is only cited to prior work [22], not independently checked in this paper.
  • domain assumption A fully-connected network with 10 layers of 100 sine units can represent the discrepancy field accurately on the sampled domain.
    Section 3; no capacity or approximation analysis is provided.
  • ad hoc to paper Training-window and POFU blending do not corrupt the near-body prediction used for initialization.
    Section 2.2; window size and transition widths are user-specified and no sensitivity study is reported.
  • domain assumption The quadtree refinement metric based on field differences selects training points that adequately capture flow features.
    Section 2.3, Eq. 11; heuristic and not compared to error-optimal sampling.
  • domain assumption The steady-state assumption holds in the validation range up to AoA 7 degrees, with unsteadiness beyond that.
    Section 3.2; limits the method to steady regimes.
  • standard math Conformal mapping potential-flow formulas for the cylinder and Joukowski airfoil are valid.
    Appendix B; standard results from [1,2,5].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multi-Fidelity Machine Learning Applied to Steady Fluid Flows." pith.science (2026). https://pith.science/paper/NVH3SC4D

@misc{pith2026250114870,
  author       = {Pith},
  title        = {Pith review of: Multi-Fidelity Machine Learning Applied to Steady Fluid Flows},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NVH3SC4D}},
  note         = {Machine review of arXiv:2501.14870}
}
read the original abstract

A machine learning method to predict steady external fluid flows using elliptic input features is introduced. Using data from as few as one high-fidelity simulation, the proposed method produces models generalizable under changes to boundary geometry by using solutions to elliptic boundary value problems over the flow domain as the model input, instead of Cartesian coordinates of the domain. Training data is generated through pointwise evaluation of flow features at points selected through a quad-tree adaptive sampling method to concentrate training points in areas with large field gradients. Models are trained within a training window around the body, while predictions are smoothly extended to freestream conditions using a Partition-of-Unity extension. Predictive capabilities of the machine learning model are demonstrated in steady-state flow of incompressible fluid around a cylinder and a Joukowski airfoil. The predicted flow field is used to warm-start CFD simulations to achieve acceleration in solver convergence.

Figures

Figures reproduced from arXiv: 2501.14870 by the authors.

Figure 1
Figure 1. Flow charts of the MF-LEIF training and prediction processes. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Examples of elliptic input features for canonical problems. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. ML model training window and POFU window functions. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Illustration of a quadtree derefinement and refinement process (Algorithm 1c, Appendix A). [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Quadtree sampling points for flow around a cylinder. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: ML model training time scaling over data size [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Distributions of 949, 950 and 1001 data points for the problem of flow around a circular cylinder [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Prediction errors for flow around a circular cylinder using ML models trained at the reference circular [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: ML-predicted x−component of the velocity at the reference circular radius of R =100% (training point) is compared to the HF solution (a). The ML model training was carried out using (b) uniform, (c) random, and (d) quadtree sampling. (a) HF solution, u (b) HF solution,…
Figure 10
Figure 10. Figure 10: The trained ML model was used to predict the flow fields at a smaller circular cylinder at [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: The trained ML model was used to predict the flow fields at a larger circular cylinder at [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]
Figure 12
Figure 12. Figure 12: HF solutions (velocity magnitude and streamlines) for flow around a Joukowski airfoil indicate [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]
Figure 13
Figure 13. Figure 13: Distributions of 944, 950, and 1002 data points for the problem of flow around a Joukowski airfoil [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]
Figure 14
Figure 14. Figure 14: Prediction errors for flow around a Joukowski airfoil using ML models trained at AoA=0 [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]
Figure 15
Figure 15. Figure 15: ML-predicted x−component of the velocity at AoA=0◦ (training point) is compared to the HF solution (a). The ML model training was carried out using (b) uniform, (c) random, and (d) quadtree sampling. (a) Velocity, x-component (b) Velocity, y-component (c) Pressure [P…
Figure 16
Figure 16. Figure 16: Prediction errors for flow around a Joukowski airfoil based on ML models trained using various ref [PITH_FULL_IMAGE:figures/full_fig_p014_16.png]
Figure 17
Figure 17. Figure 17: HF solution (a,b) and ML-predicted fields based on the model trained using AoA=0 [PITH_FULL_IMAGE:figures/full_fig_p015_17.png]
Figure 18
Figure 18. Figure 18: HF solution (a,b) and ML-predicted fields based on the model trained using AoA=0 [PITH_FULL_IMAGE:figures/full_fig_p016_18.png]
Figure 19
Figure 19. Figure 19: Comparison of forces computed using freestream and ML-predicted fields for CFD initialization for [PITH_FULL_IMAGE:figures/full_fig_p016_19.png]
Figure 20
Figure 20. Figure 20: Comparison of forces computed using freestream and ML-predicted fields for CFD initialization for [PITH_FULL_IMAGE:figures/full_fig_p017_20.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

59 extracted references · 32 canonical work pages

  1. [18]

    Enhancement of Low Fidelity Fluid Simulations using Machine Learn- ing

    Kazuko Fuchi et al. “Enhancement of Low Fidelity Fluid Simulations using Machine Learn- ing”. en. In: AIAA Scitech 2020 Forum . Orlando, FL: American Institute of Aeronautics and Astronautics, Jan. 2020. isbn: 9781624105951. doi: 10 . 2514 / 6 . 2020 - 1409. url: https : //arc.aiaa.org/doi/10.2514/6.2020-1409. 19

    work page doi:10.2514/6.2020-1409 2020

  2. [24]

    A turbulent eddy-viscosity surrogate modeling framework for Reynolds- Averaged Navier-Stokes simulations

    Romit Maulik et al. “A turbulent eddy-viscosity surrogate modeling framework for Reynolds- Averaged Navier-Stokes simulations”. en. In:Computers & Fluids (Nov. 2020), p. 104777.issn: 0045-7930. doi: 10.1016/j.compfluid.2020.104777 . url: http://www.sciencedirect. com/science/article/pii/S0045793020303479

    work page arXiv 2020

  3. [1]

    Fundamentals of Hydro- and Aeromechanics

    L. Prandtl. “Fundamentals of Hydro- and Aeromechanics”. en. In: By: O.G. Tietjens, Trans- lated by: L. Rosenhead, Dover Publications Inc., New York, Copyright 1934 by United Engi- neering Trusters Inc., Printed in the USA (1957)

    work page 1957

  4. [2]

    An Introduction to Theoretical and Computational Aerodynamics

    Jack Moran. “An Introduction to Theoretical and Computational Aerodynamics”. en. In: University of Minnesota, USA, Published by: John Wiley & Sons, New York, ISBN: 0 471 87491 4, Printed in the USA (1984). url: https://repository.tudelft.nl/islandora/ object/uuid%3A6162b190-8b01-4f6f-83a9-5f6be25cf60e

    work page 1984

  5. [3]

    Geuzaine and J

    C. Geuzaine and J. Remacle. Gmsh : a three-dimensional finite element mesh generator with built-in pre-and post-processing facilities . en. 2009. url: https://www.semanticscholar. org/paper/Gmsh- %3A- a- three- dimensional- finite- element- mesh- with- Geuzaine- Remacle/029b93cdb4fde39a90f3b31129cd77eca862a93e

    work page 2009

  6. [4]

    Fast multipole boundary element method: theory and applications in engineering

    Yijun Liu. Fast multipole boundary element method: theory and applications in engineering . Cambridge university press, 2009. 18

    work page 2009

  7. [5]

    Abbott and Albert E

    Ira H. Abbott and Albert E. VonDoenhoff. Theory of wing sections: including a summery of airfoil data. eng. Unabr. and corr. republ., [Nachdr.] New York, NY: Dover Publ, 2010. isbn: 9780486605869

    work page 2010

  8. [6]

    Fundamentals of Aerodynamics, Fifth Edition

    John Anderson. Fundamentals of Aerodynamics, Fifth Edition . McGraw Hill, 2011

    work page 2011

Show all 59 references
  1. [7]

    TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems

    Mart ´ ın Abadi et al. TensorFlow: Large-Scale Machine Learning on Heterogeneous Systems . Software available from tensorflow.org. 2015. url: https://www.tensorflow.org/

    2015

  2. [8]

    Convolutional Neural Networks for Steady Flow Approximation

    Xiaoxiao Guo, Wei Li, and Francesco Iorio. “Convolutional Neural Networks for Steady Flow Approximation”. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. KDD ’16. San Francisco, California, USA: Association for Computing ...

    2016

  3. [9]

    A Review of Variational Multiscale Methods for the Simulation of Turbulent Incompressible Flows

    Naveed Ahmed et al. “A Review of Variational Multiscale Methods for the Simulation of Turbulent Incompressible Flows”. en. In: Archives of Computational Methods in Engineering 24.1 (Jan. 2017), pp. 115–164. issn: 1886-1784. doi: 10.1007/s11831- 015- 9161- 0. url: https://doi.o...

    2017 doi

  4. [10]

    Sobolev Training for Neural Networks

    Wojciech M. Czarnecki et al. “Sobolev Training for Neural Networks”. In: Advances in Neu- ral Information Processing Systems . Vol. 30. Curran Associates, Inc., 2017. url: https : / / proceedings . neurips . cc / paper / 2017 / hash / 758a06618c69880a6cee5314ee42d52f - Abstract.html

    2017

  5. [11]

    Spiral: A General Framework For Parameter Sensitivity Analysis

    David Makhija and Philip S. Beran. “Spiral: A General Framework For Parameter Sensitivity Analysis”. en. In: 58th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Mate- rials Conference. Grapevine, Texas: American Institute of Aeronautics and Astronautics, Jan

  6. [12]

    Graph Element Networks: adaptive, structured computation and memory

    Ferran Alet et al. “Graph Element Networks: adaptive, structured computation and memory”. en. In: PMLR, May 2019, pp. 212–222. url: https : / / proceedings . mlr . press / v97 / alet19a.html

    2019

  7. [13]

    Concurrent shape and topology optimization for steady conjugate heat transfer

    David S. Makhija and Philip S. Beran. “Concurrent shape and topology optimization for steady conjugate heat transfer”. en. In: Structural and Multidisciplinary Optimization 59.3 (Mar. 2019), pp. 919–940. issn: 1615-147X, 1615-1488. doi: 10.1007/s00158-018-2110-4 . url: http://...

    2019 doi

  8. [14]

    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. E. Karniadakis. “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial dif- ferential equations”. en. In: Journal of Computational Physics 378 (Feb. 2019), pp. 686–707. ...

    2019 doi

  9. [15]

    Optimize TensorFlow & Keras models with L-BFGS from TensorFlow Prob- ability

    Chuang Pi-Yueh. Optimize TensorFlow & Keras models with L-BFGS from TensorFlow Prob- ability. Nov. 2019. url: https://pychao.com/2019/11/02/optimize-tensorflow-keras- models-with-l-bfgs-from-tensorflow-probability/

    2019

  10. [16]

    Combining Differen- tiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction

    Filipe De Avila Belbute-Peres, Thomas Economon, and Zico Kolter. “Combining Differen- tiable PDE Solvers and Graph Neural Networks for Fluid Flow Prediction”. en. In: Inter- national Conference on Machine Learning . PMLR, Nov. 2020, pp. 2402–2411. url: https: //proceedings.mlr...

    2020

  11. [17]

    Machine Learning for Fluid Mechanics

    Steven L. Brunton, Bernd R. Noack, and Petros Koumoutsakos. “Machine Learning for Fluid Mechanics”. In: Annual Review of Fluid Mechanics 52.1 (2020), pp. 477–508. doi: 10.1146/ annurev - fluid - 010719 - 060214. url: https : / / doi . org / 10 . 1146 / annurev - fluid - 010719-060214

    2020

  12. [19]

    Conservative physics- informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems

    Ameya D. Jagtap, Ehsan Kharazmi, and George Em Karniadakis. “Conservative physics- informed neural networks on discrete domains for conservation laws: Applications to forward and inverse problems”. en. In: Computer Methods in Applied Mechanics and Engineering 365 (June 2020), ...

    2020

  13. [20]

    Fourier Neural Operator for Parametric Partial Differential Equations

    Zong-Yi Li et al. “Fourier Neural Operator for Parametric Partial Differential Equations”. In: ArXiv abs/2010.08895 (2020). url: https : / / api . semanticscholar . org / CorpusID : 224705257

    2020 arXiv

  14. [21]

    Neural Operator: Graph Kernel Network for Partial Differential Equations

    Zongyi Li et al. “Neural Operator: Graph Kernel Network for Partial Differential Equations”. In: (Mar. 7, 2020). arXiv: 2003.03485 [cs.LG]

    2020 arXiv

  15. [22]

    Concurrent shape and topology optimization for unsteady conjugate heat transfer

    David S. Makhija and Philip S. Beran. “Concurrent shape and topology optimization for unsteady conjugate heat transfer”. en. In: Structural and Multidisciplinary Optimization 62.3 (Sept. 2020), pp. 1275–1297. issn: 1615-147X, 1615-1488. doi: 10.1007/s00158-020-02554- y. url: h...

    2020 doi

  16. [23]

    Physics-informed neural net- works for high-speed flows

    Zhiping Mao, Ameya D. Jagtap, and George Em Karniadakis. “Physics-informed neural net- works for high-speed flows”. en. In: Computer Methods in Applied Mechanics and Engineering 360 (Mar. 2020), p. 112789. issn: 0045-7825. doi: 10 . 1016 / j . cma . 2019 . 112789. url: https:/...

    2020

  17. [25]

    CFDNet: A Deep Learning-Based Accelerator for Fluid Simu- lations

    Octavi Obiols-Sales et al. “CFDNet: A Deep Learning-Based Accelerator for Fluid Simu- lations”. In: Proceedings of the 34th ACM International Conference on Supercomputing . ICS ’20. Barcelona, Spain: Association for Computing Machinery, 2020. isbn: 9781450379830. doi: 10.1145/...

    2020

  18. [26]

    Hidden fluid mechanics: Learn- ing velocity and pressure fields from flow visualizations

    Maziar Raissi, Alireza Yazdani, and George Em Karniadakis. “Hidden fluid mechanics: Learn- ing velocity and pressure fields from flow visualizations”. EN. In: Science 367 (Feb. 2020), pp. 1026–1030. issn: 6481. doi: 10.1126/science.aaw4741 . url: https://www.science. org/doi/a...

    2020 doi

  19. [27]

    Physics-informed deep learning for incompress- ible laminar flows

    Chengping Rao, Hao Sun, and Yang Liu. “Physics-informed deep learning for incompress- ible laminar flows”. en. In: Theoretical and Applied Mechanics Letters 10.3 (Mar. 2020), pp. 207–212. issn: 2095-0349. doi: 10 . 1016 / j . taml . 2020. 01 . 039. url: https : / / www . scien...

    2020

  20. [28]

    Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data

    Luning Sun et al. “Surrogate modeling for fluid flows based on physics-constrained deep learning without simulation data”. en. In: Computer Methods in Applied Mechanics and En- gineering 361 (Apr. 2020), p. 112732. issn: 0045-7825. doi: 10.1016/j.cma.2019.112732 . url: https:/...

    2020

  21. [29]

    Deep Learning Methods for Reynolds-Averaged Navier-Stokes Simulations of Airfoil Flows

    Nils Thuerey et al. “Deep Learning Methods for Reynolds-Averaged Navier-Stokes Simulations of Airfoil Flows”. In: AIAA Journal 58.1 (2020), pp. 25–36. issn: 0001-1452. doi: 10.2514/ 1.J058291. url: https://doi.org/10.2514/1.J058291

    2020 doi

  22. [30]

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

    Shengze Cai et al. “Physics-informed neural networks (PINNs) for fluid mechanics: a review”. en. In: Acta Mechanica Sinica 37.12 (Dec. 2021), pp. 1727–1738. issn: 1614-3116. doi: 10. 1007/s10409-021-01148-1 . url: https://doi.org/10.1007/s10409-021-01148-1

    2021 doi

  23. [31]

    Transfer learning for deep neural network-based partial differential equa- tions solving

    Xinhai Chen et al. “Transfer learning for deep neural network-based partial differential equa- tions solving”. In: Advances in Aerodynamics 3.1 (Dec. 2021), p. 36. issn: 2524-6992. doi: 10.1186/s42774-021-00094-7 . url: https://doi.org/10.1186/s42774-021-00094-7

    2021 doi

  24. [32]

    One-Shot Transfer Learning of Physics-Informed Neural Networks

    Shaan Desai et al. “One-Shot Transfer Learning of Physics-Informed Neural Networks”. In: arXiv:2110.11286 [physics] (Oct. 2021). arXiv: 2110.11286. url: http://arxiv.org/abs/ 2110.11286. 20

    2021 arXiv

  25. [33]

    PhyGeoNet: Physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular do- main

    Han Gao, Luning Sun, and Jian-Xun Wang. “PhyGeoNet: Physics-informed geometry-adaptive convolutional neural networks for solving parameterized steady-state PDEs on irregular do- main”. en. In: Journal of Computational Physics 428 (Mar. 2021), p. 110079. issn: 0021-9991. doi: 1...

    2021

  26. [34]

    NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations

    Xiaowei Jin et al. “NSFnets (Navier-Stokes flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations”. en. In: Journal of Computational Physics 426 (Feb. 2021), p. 109951. issn: 0021-9991. doi: 10.1016/j.jcp.2020.109951 . url: https: //www.sci...

    2021

  27. [35]

    Physics-informed machine learning

    George Em Karniadakis et al. “Physics-informed machine learning”. en. In: Nature Reviews Physics 3.6 (June 2021), pp. 422–440. issn: 2522-5820. doi: 10.1038/s42254-021-00314-5 . url: https://www.nature.com/articles/s42254-021-00314-5

    2021 doi

  28. [36]

    A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries

    Ali Kashefi, Davis Rempe, and Leonidas J. Guibas. “A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries”. In: Physics of Fluids 33.2 (Feb. 2021), p. 027104. issn: 1070-6631. doi: 10.1063/5.0033376. url: https://aip.scitation. org/do...

    2021 doi

  29. [37]

    Characterizing possible failure modes in physics-informed neural networks

    Aditi Krishnapriyan et al. “Characterizing possible failure modes in physics-informed neural networks”. In: Advances in Neural Information Processing Systems . Ed. by M. Ranzato et al. Vol. 34. Curran Associates, Inc., 2021, pp. 26548–26560. url: https://proceedings. neurips.c...

    2021

  30. [38]

    Learning nonlinear operators via DeepONet based on the universal approxima- tion theorem of operators

    Lu Lu et al. “Learning nonlinear operators via DeepONet based on the universal approxima- tion theorem of operators”. en. In: Nature Machine Intelligence 3.3 (Mar. 2021), pp. 218–229. issn: 2522-5839. doi: 10.1038/s42256- 021- 00302- 5. url: https://www.nature.com/ articles/s4...

    2021 doi

  31. [39]

    SURFNet: Super-Resolution of Turbulent Flows with Transfer Learning using Small Datasets

    Octavi Obiols-Sales et al. “SURFNet: Super-Resolution of Turbulent Flows with Transfer Learning using Small Datasets”. In: 2021 30th International Conference on Parallel Ar- chitectures and Compilation Techniques (PACT) . Sept. 2021, pp. 331–344. doi: 10 . 1109 / PACT52795.2021.00031

    2021

  32. [40]

    Physics-Informed Neural Networks (PINNs) for Parameterized PDEs: A Metalearning Approach

    Michael Penwarden et al. “Physics-Informed Neural Networks (PINNs) for Parameterized PDEs: A Metalearning Approach”. In:arXiv:2110.13361 [physics] (Oct. 2021). arXiv: 2110.13361. url: http://arxiv.org/abs/2110.13361

    2021 arXiv

  33. [41]

    A Comprehensive Survey on Graph Neural Networks

    Zonghan Wu et al. “A Comprehensive Survey on Graph Neural Networks”. In: IEEE Transac- tions on Neural Networks and Learning Systems 32.1 (Jan. 2021), pp. 4–24. issn: 2162-2388. doi: 10.1109/TNNLS.2020.2978386

    2021

  34. [42]

    Physics and equality constrained artificial neural net- works: Application to forward and inverse problems with multi-fidelity data fusion

    Shamsulhaq Basir and Inanc Senocak. “Physics and equality constrained artificial neural net- works: Application to forward and inverse problems with multi-fidelity data fusion”. en. In: Journal of Computational Physics 463 (Aug. 2022), p. 111301. issn: 0021-9991. doi: 10 . 101...

    2022

  35. [43]

    CAN-PINN: A fast physics-informed neural network based on coupled- automatic-numerical differentiation method

    Pao-Hsiung Chiu et al. “CAN-PINN: A fast physics-informed neural network based on coupled- automatic-numerical differentiation method”. en. In: Computer Methods in Applied Mechanics and Engineering 395 (May 2022), p. 114909. issn: 0045-7825. doi: 10.1016/j.cma.2022. 114909. ur...

    2022 doi

  36. [44]

    Pi-Yueh Chuang and Lorena A. Barba. Experience report of physics-informed neural networks in fluid simulations: pitfalls and frustration. Tech. rep. arXiv:2205.14249 [physics] type: article. arXiv, May 2022. doi: 10.48550/arXiv.2205.14249 . url: http://arxiv.org/abs/2205. 14249

    2022 doi arXiv

  37. [45]

    Scientific Machine Learning through Physics-Informed Neural Net- works: Where we are and What’s next

    Salvatore Cuomo et al. “Scientific Machine Learning through Physics-Informed Neural Net- works: Where we are and What’s next”. In: arXiv:2201.05624 [physics] (Jan. 2022). arXiv: 2201.05624. url: http://arxiv.org/abs/2201.05624. 21

    2022 arXiv

  38. [46]

    Physics-informed deep-learning applications to ex- perimental fluid mechanics

    Hamidreza Eivazi and Ricardo Vinuesa. “Physics-informed deep-learning applications to ex- perimental fluid mechanics”. In: arXiv:2203.15402 [physics] (Mar. 2022). arXiv: 2203.15402. url: http://arxiv.org/abs/2203.15402

    2022 arXiv

  39. [47]

    Physics-informed neural networks for solving Reynolds-averaged Navier-Stokes equations

    Hamidreza Eivazi et al. “Physics-informed neural networks for solving Reynolds-averaged Navier-Stokes equations”. In: Physics of Fluids 34.7 (July 2022), p. 075117. issn: 1070-6631. doi: 10 . 1063 / 5 . 0095270. url: https : / / aip . scitation . org / doi / full / 10 . 1063 /...

    2022

  40. [48]

    Physics-informed neural networks for inverse problems in supersonic flows

    Ameya D. Jagtap et al. “Physics-informed neural networks for inverse problems in supersonic flows”. In: arXiv:2202.11821 [cs, math] (Feb. 2022). arXiv: 2202.11821. url: http://arxiv. org/abs/2202.11821

    2022 arXiv

  41. [49]

    Machine Learning in Aerodynamic Shape Optimization

    Jichao Li, Xiaosong Du, and Joaquim R. R. A. Martins. “Machine Learning in Aerodynamic Shape Optimization”. In: arXiv:2202.07141 [physics] (Feb. 2022). arXiv: 2202.07141. url: http://arxiv.org/abs/2202.07141

    2022 arXiv

  42. [50]

    Molnar et al

    Joseph P. Molnar et al. Estimating density, velocity, and pressure fields in supersonic flow using physics-informed BOS . Tech. rep. arXiv:2208.04280 [physics] type: article. arXiv, Aug

  43. [51]

    Mosaic flows: A transferable deep learning framework for solving PDEs on unseen domains

    Hengjie Wang et al. “Mosaic flows: A transferable deep learning framework for solving PDEs on unseen domains”. en. In: Computer Methods in Applied Mechanics and Engineering 389 (Feb. 2022), p. 114424. issn: 0045-7825. doi: 10.1016/j.cma.2021.114424 . url: https: //www.scienced...

    2022

  44. [52]

    A comprehensive study of non-adaptive and residual-based adaptive sam- pling for physics-informed neural networks

    Chenxi Wu et al. A comprehensive study of non-adaptive and residual-based adaptive sam- pling for physics-informed neural networks. Tech. rep. arXiv:2207.10289 [physics] type: article. arXiv, July 2022. doi: 10.48550/arXiv.2207.10289 . url: http://arxiv.org/abs/2207. 10289

    2022 doi arXiv

  45. [53]

    Gradient-enhanced physics-informed neural networks for forward and inverse PDE problems

    Jeremy Yu et al. “Gradient-enhanced physics-informed neural networks for forward and inverse PDE problems”. en. In: Computer Methods in Applied Mechanics and Engineering 393 (Apr. 2022), p. 114823. issn: 0045-7825. doi: 10.1016/j.cma.2022.114823 . url: https://www. sciencedire...

    2022

  46. [54]

    tfp.optimizer.lbfgs minimize — TensorFlow Probability . en. url: https://www.tensorflow. org/probability/api_docs/python/tfp/optimizer/lbfgs_minimize

  47. [55]

    Accuracy Improvement Technique of DNN for Accelerating CFD Simu- lator

    Yukito Tsunoda et al. “Accuracy Improvement Technique of DNN for Accelerating CFD Simu- lator”. In: AIAA SCITECH 2022 Forum. American Institute of Aeronautics and Astronautics. doi: 10.2514/6.2022-1437. url: https://arc.aiaa.org/doi/abs/10.2514/6.2022-1437

    2022 doi

  48. [56]

    url: https://how5.cenaero.be/ content/vl1-laminar-joukowski-airfoil-re1000

    VL1 Laminar Joukowski airfoil at Re=1000 — HiOCFD5 . url: https://how5.cenaero.be/ content/vl1-laminar-joukowski-airfoil-re1000

  49. [57]

    Fast Neural Network Predictions from Constrained Aerodynamics Datasets

    Cristina White, Daniela Ushizima, and Charbel Farhat. “Fast Neural Network Predictions from Constrained Aerodynamics Datasets”. In: AIAA Scitech 2020 Forum. American Institute of Aeronautics and Astronautics. doi: 10.2514/6.2020- 0364 . url: https://arc.aiaa. org/doi/abs/10.25...

    2020 doi

  50. [2017]

    doi: 10.2514/6.2017-1306

    isbn: 9781624104534. doi: 10.2514/6.2017-1306. url: https://arc.aiaa.org/doi/ 10.2514/6.2017-1306

    2017 doi

  51. [2022]

    url: http://arxiv.org/abs/2208.04280

    doi: 10.48550/arXiv.2208.04280. url: http://arxiv.org/abs/2208.04280

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.