pith. sign in

arxiv: 2606.09963 · v1 · pith:TW6EYJPAnew · submitted 2026-06-08 · ⚛️ physics.flu-dyn · cs.AI

Geometry-Aware Anisotropic Boundary Correction for Aerodynamic Simulation

Xin Zhang , Yipeng Huang , Shu Jiang , Zhenzhong Wang , Min Jiang This is my paper

Pith reviewed 2026-06-27 14:54 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn cs.AI
keywords neural operatorsaerodynamic simulationboundary correctionanisotropic modelinggeometry-aware methodscomputational fluid dynamicsairfoil flowsvehicle aerodynamics
0
0 comments X

The pith

GeoABC adds geometry-conditioned anisotropic boundary correction to neural operators for more accurate near-wall aerodynamic predictions.

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

The paper establishes that neural operators for aerodynamic flows can achieve lower near-boundary errors by replacing isotropic boundary treatment with direction-aware corrections derived from geometry. It argues that flow along the wall differs physically from flow normal to the wall, so intermediate representations should be modulated accordingly rather than treating boundaries as uniform features. This change converts static geometry inputs into a structural prior that guides the prediction. The result matters for engineering because surface pressure and other design quantities depend heavily on accurate near-wall behavior, and the method works across different operator backbones on both two-dimensional airfoils and three-dimensional car geometries.

Core claim

GeoABC is a geometry-conditioned anisotropic boundary correction framework that leverages boundary geometries to introduce direction-aware boundary correction into the intermediate representations of neural operators, transforming boundary geometry from static input features into a structural prior that modulates physical prediction and reduces near-boundary relative L2 error by approximately 38 percent on average.

What carries the argument

GeoABC, the geometry-conditioned anisotropic boundary correction framework, which injects direction-aware corrections from boundary geometry into neural operator intermediate layers to separately handle tangential flow propagation along the wall and normal constraint by the wall.

If this is right

  • GeoABC can be attached to multiple existing neural operator architectures without altering their core design.
  • The correction narrows the near-wall performance gap that appears across mainstream neural operators for aerodynamic tasks.
  • Gains appear consistently on both 2D airfoil and 3D car flow problems.
  • Boundary geometry changes from a passive input into an active modulator of the predicted flow field.

Where Pith is reading between the lines

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

  • The same direction-aware correction idea could be tested in other interface-dominated simulations such as heat conduction or wave propagation at material boundaries.
  • Combining GeoABC with mesh-free neural operators might allow high-resolution near-wall accuracy without increasing overall grid density.
  • Evaluating the framework on full-vehicle or aircraft geometries would show whether the 38 percent error reduction holds at larger scales.

Load-bearing premise

Aerodynamic flows near solid boundaries exhibit consistent anisotropy that geometry information can capture and correct inside the neural operator's intermediate representations.

What would settle it

Running the same 2D airfoil or 3D car experiments with the identical neural operator backbone but without the anisotropic geometry correction and finding no reduction or an increase in near-boundary relative L2 error.

Figures

Figures reproduced from arXiv: 2606.09963 by Min Jiang, Shu Jiang, Xin Zhang, Yipeng Huang, Zhenzhong Wang.

Figure 1
Figure 1. Figure 1: GeoABC addresses the near-wall accuracy gap left by existing boundary-aware remedies [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of GeoABC. GeoABC inserts a geometry-conditioned anisotropic correction [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Velocity-magnitude prediction and error maps on the 2D NACA benchmark. The numbers [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: 3D visualization of the surrounding velocity field around the car, showing airflow stream [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Component isolation of GeoABC, comparing the full GeoABC with four ablated variants [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Car and airfoil design tasks. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of 2D airfoil benchmark. Here, [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visualization of 3D car benchmark. Here, [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Spatial attribution of GeoABC gains on Geo-FNO. We analyze the multi-field (u, v, p) NACA airfoil setting. (a) Test-mean pointwise pressure error along the airfoil surface; the shaded region denotes the per-position reduction over the baseline. (b) Error reduction as a function of grid layer distance from the wall to the far field. (c) Error reduction across boundary-curvature quartiles. GeoABC mainly impr… view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative prediction comparison on the 2D NACA airfoil benchmark. We visualize [PITH_FULL_IMAGE:figures/full_fig_p029_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Absolute error maps on the 2D NACA airfoil benchmark. Each panel shows the pointwise [PITH_FULL_IMAGE:figures/full_fig_p030_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Qualitative prediction comparison on the 2D NACA airfoil benchmark. We visualize [PITH_FULL_IMAGE:figures/full_fig_p031_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Absolute error maps on the 2D NACA airfoil benchmark. Each panel shows the pointwise [PITH_FULL_IMAGE:figures/full_fig_p032_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Visualization of surrounding velocity prediction errors on the Shape-Net Car benchmark. [PITH_FULL_IMAGE:figures/full_fig_p032_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Surface pressure error visualization on the ShapeNetCar benchmark. We visualize the pressure prediction error on the same test car for Geo-FNO, GINO, GNOT, and Transolver, comparing each baseline with its GeoABC-enhanced counterpart under a shared color scale. GeoABC consistently suppresses large spatial error patterns on the vehicle surface, especially around high￾curvature and front-body regions, showin… view at source ↗
Figure 16
Figure 16. Figure 16: Surface pressure error visualization on the ShapeNetCar benchmark. We visualize the pressure prediction error on the same test car for Geo-FNO, GINO, GNOT, and Transolver, comparing each baseline with its GeoABC-enhanced counterpart under a shared color scale. GeoABC consistently suppresses large spatial error patterns on the vehicle surface, especially around high￾curvature and front-body regions, showin… view at source ↗
read the original abstract

Aerodynamic simulation is a key component of engineering shape design, where core quantities such as the surface pressure coefficient strongly depend on flow dynamics near solid boundaries. Neural operators provide an efficient alternative to expensive Computational Fluid Dynamics (CFD) solvers. However, conventional methods treat the boundary region isotropically, failing to account for the distinct physical behaviors along the boundaries. In reality, the aerodynamic process exhibits anisotropy: along the tangential direction, flow propagates along the wall; along the normal direction, physical quantities are constrained by the wall. To explicitly model the distinct physical behaviors, we propose GeoABC, a geometry-conditioned anisotropic boundary correction framework. GeoABC leverages the boundary geometries to introduce direction-aware boundary correction into the intermediate representations of neural operators, transforming boundary geometry from static input features into a structural prior that modulates physical prediction. On 2D airfoil and 3D car tasks, GeoABC consistently adapts to multiple neural operator backbones, reducing near-boundary relative $L_2$ error by $\sim$38\% on average, narrowing the structural near-wall gap shared by mainstream neural operators, and advancing neural operators toward high-fidelity aerodynamic simulation.

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

1 major / 1 minor

Summary. The manuscript proposes GeoABC, a geometry-conditioned anisotropic boundary correction framework for neural operators applied to aerodynamic simulation. It claims that standard neural operators treat boundaries isotropically and fail to capture the distinct tangential propagation versus normal constraint behaviors near walls; GeoABC conditions intermediate representations on boundary geometry to introduce direction-aware corrections. Empirical results on 2D airfoil and 3D car tasks show an average ~38% reduction in near-boundary relative L2 error across multiple backbones, narrowing the structural near-wall gap.

Significance. If the reported gains are robust, this approach could meaningfully improve near-wall fidelity in neural-operator surrogates for aerodynamics, a known weakness that limits their engineering utility. Credit is due for the consistent adaptation across backbones and for testing on both 2D and 3D geometries.

major comments (1)
  1. [§4] §4 (Experiments): the central claim attributes the ~38% near-boundary L2 reduction specifically to the anisotropic (direction-aware) component that models tangential flow propagation versus normal constraint. No ablation isolating this directionality from a geometry-conditioned but isotropic variant is reported, leaving it unclear whether anisotropy is load-bearing or whether geometry modulation alone suffices.
minor comments (1)
  1. [Abstract] Abstract: the 38% error reduction is stated without reference to the precise baselines, number of runs, or definition of the near-boundary region, which weakens immediate assessment of the result.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive comment on our experimental design. We address it below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [§4] §4 (Experiments): the central claim attributes the ~38% near-boundary L2 reduction specifically to the anisotropic (direction-aware) component that models tangential flow propagation versus normal constraint. No ablation isolating this directionality from a geometry-conditioned but isotropic variant is reported, leaving it unclear whether anisotropy is load-bearing or whether geometry modulation alone suffices.

    Authors: We agree that an explicit ablation isolating the anisotropic (direction-aware) component from a purely geometry-conditioned isotropic variant would strengthen the central claim. In the revised manuscript we will add this ablation by implementing and evaluating a geometry-conditioned isotropic baseline on the same 2D airfoil and 3D car tasks, thereby quantifying the incremental benefit attributable to directionality. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper presents GeoABC as an empirical framework that conditions neural operator representations on boundary geometry to enforce direction-aware corrections motivated by observed tangential vs. normal flow behaviors. The central performance claim (∼38% near-boundary L2 reduction) is reported as an outcome of applying the method to 2D airfoil and 3D car tasks across multiple backbones; no derivation, uniqueness theorem, or fitted parameter is shown to reduce by construction to the target metric. No self-citations appear in the provided text, and the anisotropy modeling is introduced as an architectural choice rather than a self-referential definition. The result is therefore self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Based on abstract, the method introduces a new framework without specifying free parameters or additional axioms beyond the physical anisotropy assumption.

axioms (1)
  • domain assumption Aerodynamic process exhibits anisotropy: along the tangential direction, flow propagates along the wall; along the normal direction, physical quantities are constrained by the wall.
    Stated in abstract as the motivation for the method.
invented entities (1)
  • GeoABC framework no independent evidence
    purpose: To model anisotropic boundary effects in neural operators by conditioning intermediate representations on boundary geometry
    New proposed method without external validation mentioned.

pith-pipeline@v0.9.1-grok · 5737 in / 1303 out tokens · 23819 ms · 2026-06-27T14:54:28.041765+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

42 extracted references

  1. [1]

    Anderson.Computational Fluid Dynamics: The Basics with Applications

    John D. Anderson.Computational Fluid Dynamics: The Basics with Applications. McGraw- Hill, New York, 1995

    1995

  2. [2]

    Aerodynamic design via control theory.Journal of Scientific Computing, 3(3):233–260, 1988

    Antony Jameson. Aerodynamic design via control theory.Journal of Scientific Computing, 3(3):233–260, 1988

    1988

  3. [3]

    Butterworth- Heinemann, 3rd edition, 2015

    Jiri Blazek.Computational Fluid Dynamics: Principles and Applications. Butterworth- Heinemann, 3rd edition, 2015

    2015

  4. [4]

    Ferziger and Milovan Peri´c.Computational Methods for Fluid Dynamics

    Joel H. Ferziger and Milovan Peri´c.Computational Methods for Fluid Dynamics. Springer, 3rd edition, 2002

    2002

  5. [5]

    Aerodynamic design optimization: Challenges and perspectives.Com- puters & Fluids, 239:105391, 2022

    Joaquim RRA Martins. Aerodynamic design optimization: Challenges and perspectives.Com- puters & Fluids, 239:105391, 2022

    2022

  6. [6]

    Method for the constrained design of natural laminar flow airfoils.Journal of aircraft, 34(6):706–712, 1997

    Bradford E Green, John L Whitesides, Richard L Campbell, and Raymond E Mineck. Method for the constrained design of natural laminar flow airfoils.Journal of aircraft, 34(6):706–712, 1997

    1997

  7. [7]

    Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang

    George Em Karniadakis, Ioannis G. Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    2021

  8. [8]

    Maziar Raissi, Paris Perdikaris, and George Em 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

    2019

  9. [9]

    Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

    Steven L Brunton, Bernd R Noack, and Petros Koumoutsakos. Machine learning for fluid mechanics.Annual review of fluid mechanics, 52(1):477–508, 2020

    2020

  10. [10]

    Learning nonlinear operators via deeponet based on the universal approximation theorem of operators

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learning nonlinear operators via deeponet based on the universal approximation theorem of operators. Nat. Mach. Intell., 3(3):218–229, 2021

    2021

  11. [11]

    Stuart, and Anima Anandkumar

    Zongyi Li, Nikola Borislavov Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhat- tacharya, Andrew M. Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. In9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021. OpenReview.net, 2021

    2021

  12. [12]

    Stuart, and Anima Anandkumar

    Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew M. Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to PDEs.Journal of Machine Learning Research, 24(89):1–97, 2023

    2023

  13. [13]

    Neural fields for rapid aircraft aerodynamics simulations.Scientific Reports, 14(1):25496, 2024

    Giovanni Catalani, Siddhant Agarwal, Xavier Bertrand, Frederic Tost, Michael Bauerheim, and Joseph Morlier. Neural fields for rapid aircraft aerodynamics simulations.Scientific Reports, 14(1):25496, 2024

    2024

  14. [14]

    Prediction of compressor blade cascade flow field based on fourier neural operator.Aerospace Science and Technology, 163:110208, 2025

    Lixiang Jiang, Xinlong Feng, and Quanyong Xu. Prediction of compressor blade cascade flow field based on fourier neural operator.Aerospace Science and Technology, 163:110208, 2025

    2025

  15. [15]

    Towards scalable surrogate models based on neural fields for large scale aerodynamic simulations.Computers & Fluids, page 106929, 2025

    Giovanni Catalani, Jean Fesquet, Xavier Bertrand, Frédéric Tost, Michael Bauerheim, and Joseph Morlier. Towards scalable surrogate models based on neural fields for large scale aerodynamic simulations.Computers & Fluids, page 106929, 2025

    2025

  16. [16]

    Effect of adverse pressure gradients on turbulent wing boundary layers.Journal of Fluid Mechanics, 883:A8, 2020

    Álvaro Tanarro, Ricardo Vinuesa, and Philipp Schlatter. Effect of adverse pressure gradients on turbulent wing boundary layers.Journal of Fluid Mechanics, 883:A8, 2020

    2020

  17. [17]

    Pressure-gradient turbulent boundary layers developing around a wing section.Flow, turbulence and combustion, 99(3):613–641, 2017

    Ricardo Vinuesa, Seyed M Hosseini, Ardeshir Hanifi, Dan S Henningson, and Philipp Schlatter. Pressure-gradient turbulent boundary layers developing around a wing section.Flow, turbulence and combustion, 99(3):613–641, 2017

    2017

  18. [18]

    Strategies for turbulence modelling and simulations.International journal of heat and fluid flow, 21(3):252–263, 2000

    Philippe R Spalart. Strategies for turbulence modelling and simulations.International journal of heat and fluid flow, 21(3):252–263, 2000. 10

    2000

  19. [19]

    Unsteady aerodynamic characteristics of pitching flat plates at low reynolds numbers.Journal of Aircraft, 58(4):917–934, 2021

    Camli Badrya, Bharath Govindarajan, Albert Medina, Seung Joon Yang, and Inderjit Chopra. Unsteady aerodynamic characteristics of pitching flat plates at low reynolds numbers.Journal of Aircraft, 58(4):917–934, 2021

    2021

  20. [20]

    Design of airfoils to mitigate wake bursting

    Brent W Pomeroy and Michael S Selig. Design of airfoils to mitigate wake bursting. In55th AIAA Aerospace Sciences Meeting, page 1210, 2017

    2017

  21. [21]

    Enhancing computational fluid dynamics with machine learning.Nature Computational Science, 2(6):358–366, 2022

    Ricardo Vinuesa and Steven L Brunton. Enhancing computational fluid dynamics with machine learning.Nature Computational Science, 2(6):358–366, 2022

    2022

  22. [22]

    Fourier neural operator with learned deformations for pdes on general geometries.J

    Zongyi Li, Daniel Zhengyu Huang, Burigede Liu, and Anima Anandkumar. Fourier neural operator with learned deformations for pdes on general geometries.J. Mach. Learn. Res., 24:388:1–388:26, 2023

    2023

  23. [23]

    GNOT: A general neural operator transformer for operator learning

    Zhongkai Hao, Zhengyi Wang, Hang Su, Chengyang Ying, Yinpeng Dong, Songming Liu, Ze Cheng, Jian Song, and Jun Zhu. GNOT: A general neural operator transformer for operator learning. In Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett, editors,International Conference on Machine Learning, ICML 2023, 23-...

    2023

  24. [24]

    Kovachki, Christopher B

    Zongyi Li, Nikola B. Kovachki, Christopher B. Choy, Boyi Li, Jean Kossaifi, Shourya Prakash Otta, Mohammad Amin Nabian, Maximilian Stadler, Christian Hundt, Kamyar Azizzadenesheli, and Animashree Anandkumar. Geometry-informed neural operator for large-scale 3d pdes. In Alice Oh, Tristan Naumann, Amir Globerson, Kate Saenko, Moritz Hardt, and Sergey Levine...

    2023

  25. [25]

    Thompson, Z

    Joe F. Thompson, Z. U. A. Warsi, and C. Wayne Mastin.Numerical Grid Generation: Founda- tions and Applications. North-Holland, New York, 1985

    1985

  26. [26]

    Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023

    Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023

    2023

  27. [27]

    Temporal neural operator for modeling time- dependent physical phenomena.Scientific Reports, 15(1):32791, 2025

    Waleed Diab and Mohammed Al Kobaisi. Temporal neural operator for modeling time- dependent physical phenomena.Scientific Reports, 15(1):32791, 2025

    2025

  28. [28]

    Continuum attention for neural operators.Journal of Machine Learning Research, 26(300):1–52, 2025

    Edoardo Calvello, Nikola B Kovachki, Matthew E Levine, and Andrew M Stuart. Continuum attention for neural operators.Journal of Machine Learning Research, 26(300):1–52, 2025

    2025

  29. [29]

    Fano: Fourier advection neural operator for weather prediction.IEEE Transactions on Geoscience and Remote Sensing, 2026

    Kenghong Lin, Huiwei Lin, Baoquan Zhang, Yangjinxi Ge, Dong Liu, Xutao Li, Yunming Ye, and Chuyao Luo. Fano: Fourier advection neural operator for weather prediction.IEEE Transactions on Geoscience and Remote Sensing, 2026

    2026

  30. [30]

    Transolver: A fast transformer solver for pdes on general geometries

    Haixu Wu, Huakun Luo, Haowen Wang, Jianmin Wang, and Mingsheng Long. Transolver: A fast transformer solver for pdes on general geometries. In Ruslan Salakhutdinov, Zico Kolter, Katherine A. Heller, Adrian Weller, Nuria Oliver, Jonathan Scarlett, and Felix Berkenkamp, editors,Forty-first International Conference on Machine Learning, ICML 2024, Vienna, Aust...

    2024

  31. [31]

    Dpot: auto-regressive denoising operator transformer for large-scale pde pre-training

    Zhongkai Hao, Chang Su, Songming Liu, Julius Berner, Chengyang Ying, Hang Su, Anima Anandkumar, Jian Song, and Jun Zhu. Dpot: auto-regressive denoising operator transformer for large-scale pde pre-training. InProceedings of the 41st International Conference on Machine Learning, ICML’24. JMLR.org, 2024

    2024

  32. [32]

    Universal physics transformers: A framework for efficiently scaling neural operators.Advances in Neural Information Processing Systems, 37:25152–25194, 2024

    Benedikt Alkin, Andreas Fürst, Simon Schmid, Lukas Gruber, Markus Holzleitner, and Johannes Brandstetter. Universal physics transformers: A framework for efficiently scaling neural operators.Advances in Neural Information Processing Systems, 37:25152–25194, 2024. 11

    2024

  33. [33]

    Pretraining codomain attention neural operators for solving multiphysics pdes.Advances in Neural Information Processing Systems, 37:104035–104064, 2024

    Ashiqur Rahman, Robert J George, Mogab Elleithy, Daniel Leibovici, Zongyi Li, Boris Bonev, Colin White, Julius Berner, Raymond A Yeh, Jean Kossaifi, et al. Pretraining codomain attention neural operators for solving multiphysics pdes.Advances in Neural Information Processing Systems, 37:104035–104064, 2024

    2024

  34. [34]

    Poseidon: Efficient foundation models for pdes

    Maximilian Herde, Bogdan Raoni ´c, Tobias Rohner, Roger Käppeli, Roberto Molinaro, Em- manuel De Bezenac, and Siddhartha Mishra. Poseidon: Efficient foundation models for pdes. Advances in Neural Information Processing Systems, 37:72525–72624, 2024

    2024

  35. [35]

    Geopt: Scaling physics simulation via lifted geometric pre-training, 2026

    Haixu Wu, Minghao Guo, Zongyi Li, Zhiyang Dou, Mingsheng Long, Kaiming He, and Wojciech Matusik. Geopt: Scaling physics simulation via lifted geometric pre-training, 2026

    2026

  36. [36]

    From cheap geometry to expensive physics: A physics-agnostic pretraining framework for neural operators

    Zhizhou Zhang, Youjia Wu, Kaixuan Zhang, and Yanjia Wang. From cheap geometry to expensive physics: A physics-agnostic pretraining framework for neural operators. InThe Fourteenth International Conference on Learning Representations, 2026

    2026

  37. [37]

    BENO: boundary- embedded neural operators for elliptic pdes

    Haixin Wang, Jiaxin Li, Anubhav Dwivedi, Kentaro Hara, and Tailin Wu. BENO: boundary- embedded neural operators for elliptic pdes. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024. OpenReview.net, 2024

    2024

  38. [38]

    Cross-field interface-aware neural operators for multiphase flow simulation

    Zhenzhong Wang, Xin Zhang, Jun Liao, and Min Jiang. Cross-field interface-aware neural operators for multiphase flow simulation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 26769–26777, 2026

    2026

  39. [39]

    Benedikt Alkin, Maurits Bleeker, Richard Kurle, Tobias Kronlachner, Reinhard Sonnleit- ner, Matthias Dorfer, and Johannes Brandstetter. Ab-upt: Scaling neural cfd surrogates for high-fidelity automotive aerodynamics simulations via anchored-branched universal physics transformers.arXiv preprint arXiv:2502.09692, 2025

    arXiv 2025

  40. [40]

    Point cloud neural operator for parametric pdes on complex and variable geometries.Computer Methods in Applied Mechanics and Engineering, 443:118022, 2025

    Chenyu Zeng, Yanshu Zhang, Jiayi Zhou, Yuhan Wang, Zilin Wang, Yuhao Liu, Lei Wu, and Daniel Zhengyu Huang. Point cloud neural operator for parametric pdes on complex and variable geometries.Computer Methods in Applied Mechanics and Engineering, 443:118022, 2025

    2025

  41. [41]

    Learning three-dimensional flow for interactive aerody- namic design.ACM Transactions on Graphics (TOG), 37(4):1–10, 2018

    Nobuyuki Umetani and Bernd Bickel. Learning three-dimensional flow for interactive aerody- namic design.ACM Transactions on Graphics (TOG), 37(4):1–10, 2018

    2018

  42. [42]

    regional gain

    Bogdan Raonic, Roberto Molinaro, Tim De Ryck, Tobias Rohner, Francesca Bartolucci, Rima Alaifari, Siddhartha Mishra, and Emmanuel de Bézenac. Convolutional neural operators for robust and accurate learning of pdes. In Alice Oh, Tristan Naumann, Amir Globerson, Kate Saenko, Moritz Hardt, and Sergey Levine, editors,Advances in Neural Information Processing ...

    arXiv 2023