pith. sign in

arxiv: 2506.07969 · v2 · submitted 2025-06-09 · 💻 cs.LG · physics.flu-dyn

A Two-Phase Deep Learning Framework for Adaptive Time-Stepping in High-Speed Flow Modeling

Jacob Helwig , Sai Sreeharsha Adavi , Xuan Zhang , Yuchao Lin , Felix S. Chim , Luke Takeshi Vizzini , Haiyang Yu , Muhammad Hasnain
show 6 more authors
Saykat Kumar Biswas John J. Holloway Narendra Singh N. K. Anand Swagnik Guhathakurta Shuiwang Ji
This is my paper

Pith reviewed 2026-05-19 10:13 UTC · model grok-4.3

classification 💻 cs.LG physics.flu-dyn
keywords adaptive time-steppinghigh-speed flowsshock wavesmachine learningfluid simulationdeep learningneural ODE
0
0 comments X

The pith

ShockCast uses two machine learning phases to predict and apply adaptive timesteps for high-speed flow simulations.

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

The paper proposes a two-phase deep learning method called ShockCast for simulating high-speed fluid flows that contain sudden changes such as shock waves. In the first phase a model predicts a suitable timestep size from the current flow state. In the second phase that predicted size is fed as an extra input so a second model can advance the entire fluid field forward by exactly that amount. This replaces the fixed small steps used in low-speed flows and avoids the cost of traditional adaptive-stepping error controls. The authors test the idea on three new supersonic flow datasets and explore physically motivated prediction strategies plus conditioning techniques drawn from neural ODEs and mixture-of-experts ideas.

Core claim

ShockCast models high-speed flows by first training a machine learning component to predict an appropriate timestep size and then using that size together with the current fluid fields as inputs to a second component that advances the state forward by the predicted interval, thereby achieving adaptive time-stepping directly inside the learned simulator.

What carries the argument

The two-phase ShockCast framework, in which a timestep predictor supplies both the step length and a conditioning signal to a state-advancement model.

If this is right

  • Simulations of supersonic flows can use learned adaptive steps instead of fixed small increments, lowering the total number of time steps needed.
  • The same two-phase structure could be applied to other physical systems that require variable temporal resolution.
  • Timestep conditioning inspired by neural ODEs and mixture-of-experts can be combined with physically motivated loss terms to improve prediction quality.
  • Public datasets of supersonic flows become available for further benchmarking of learned simulators.

Where Pith is reading between the lines

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

  • The approach may reduce the engineering effort needed to tune timestep controls when deploying learned fluid models in new regimes.
  • If the predictor generalizes across different Mach numbers, it could support multi-regime simulations without retraining separate models for each speed range.
  • Extending the second phase to also output uncertainty estimates on the advanced state could provide built-in reliability checks.

Load-bearing premise

The machine learning predictor can choose timestep sizes that keep the subsequent state advancement both numerically stable and physically accurate when abrupt changes such as shock waves appear, without any separate error monitoring or correction steps.

What would settle it

Running the trained ShockCast model on a supersonic test case containing a strong shock and finding that the simulation becomes unstable or produces large deviations from a high-resolution reference solution generated with conventional adaptive time-stepping.

Figures

Figures reproduced from arXiv: 2506.07969 by Felix S. Chim, Haiyang Yu, Jacob Helwig, John J. Holloway, Luke Takeshi Vizzini, Muhammad Hasnain, Narendra Singh, N. K. Anand, Sai Sreeharsha Adavi, Saykat Kumar Biswas, Shuiwang Ji, Swagnik Guhathakurta, Xuan Zhang, Yuchao Lin.

Figure 1
Figure 1. Figure 1: Overview of the ShockCast framework for time-adaptive modeling of high-speed flows. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: One-step MAE of Neural CFL models on ∆t averaged over 3 training runs, where ∆t is normalized to have standard deviation 1. Error bars are ± 2 standard errors. 0.0000 0.0005 0.0010 0.0015 0.0020 0.0025 0.0030 t 4.75 5.00 5.25 ∆ t ×10−5 Coal Dust Explosion, Shock Mach Number 1.85 0.000 0.001 0.002 0.003 0.004 0.005 t 8 9 10 11 ∆ t ×10−5 Circular Blast, Max Mach Number 2.68 ShockCast: Unrolled ∆t Predicted T… view at source ↗
Figure 3
Figure 3. Figure 3: ∆t predicted by autoregressive unrolling of ShockCast with F-FNO+Euler conditioning neural solver backbone for a selected solution. strength of the shock between Mach 1.2 and 2.1 along with the particle diameter from case to case for a total of 100 cases, with 90 for training and 10 for evaluation. Once the simulation starts, the normal shock travels to the right as shown in [PITH_FULL_IMAGE:figures/full_… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of the ground truth (top) and predicted (bottom) density fields for the circular [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Coal dust explosion results averaged over three neural solver training runs with best [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Circular blast results averaged over three neural solver training runs with best performing [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Initial gas velocity x-component for a selected coal dust explosion case. Times are in units of 10−5 seconds and the downsampling factor relative to the classical solver solution is 100× compared to 500× used for training ShockCast. The initial shock can be seen to be moving from left to right. IPR = 3.44 IPR = 10.23 IPR = 16.53 IPR = 23.81 IPR = 29.63 IPR = 33.51 IPR = 37.88 IPR = 43.69 IPR = 47.58 10 20 … view at source ↗
Figure 8
Figure 8. Figure 8: Initial density field for the circular blast evaluation cases with varying Initial Pressure [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: TKE for coal dust explosion. True Predicted Relative Error = 0.021 Residual 10000 20000 30000 40000 50000 60000 −2000 0 2000 Circular Blast: TKE [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: TKE for circular blast. F Extended Results In this section, we present the numerical values of average evaluation errors and their corresponding standard errors as mean (standard error). In Tables 3 and 4, we present one-step errors for ShockCast. We note that the timestep predicted by the neural CFL model will not perfectly match the ground truth timestep such that the prediction from the neural solver m… view at source ↗
Figure 11
Figure 11. Figure 11: Mean flow for coal dust explosion. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Mean flow for circular blast. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Gas velocity x-component for coal dust explosion. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Gas velocity y-component for coal dust explosion. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Volume fraction for coal dust explosion. [PITH_FULL_IMAGE:figures/full_fig_p027_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Gas temperature for coal dust explosion. [PITH_FULL_IMAGE:figures/full_fig_p028_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Velocity x-component for circular blast. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Velocity y-component for circular blast. 30 [PITH_FULL_IMAGE:figures/full_fig_p030_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Density for circular blast. 31 [PITH_FULL_IMAGE:figures/full_fig_p031_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: Temperature for circular blast. 32 [PITH_FULL_IMAGE:figures/full_fig_p032_20.png] view at source ↗
read the original abstract

We consider the problem of modeling high-speed flows using machine learning methods. While most prior studies focus on low-speed fluid flows in which uniform time-stepping is practical, flows approaching and exceeding the speed of sound exhibit sudden changes such as shock waves. In such cases, it is essential to use adaptive time-stepping methods to allow a temporal resolution sufficient to resolve these phenomena while simultaneously balancing computational costs. Here, we propose a two-phase machine learning method, known as ShockCast, to model high-speed flows with adaptive time-stepping. In the first phase, we propose to employ a machine learning model to predict the timestep size. In the second phase, the predicted timestep is used as an input along with the current fluid fields to advance the system state by the predicted timestep. We explore several physically-motivated components for timestep prediction and introduce timestep conditioning strategies inspired by neural ODE and Mixture of Experts. We evaluate our methods by generating three supersonic flow datasets, available at https://huggingface.co/divelab. Our code is publicly available as part of the AIRS library (https://github.com/divelab/AIRS).

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 3 minor

Summary. The manuscript introduces ShockCast, a two-phase deep learning framework for modeling high-speed flows with adaptive time-stepping. Phase one trains an ML model to predict timestep sizes from current fluid fields; phase two conditions a state-advancement model on the predicted dt to evolve the solution. The authors incorporate physically motivated features for timestep prediction and conditioning strategies drawn from neural ODEs and Mixture-of-Experts architectures. Evaluation is performed on three newly generated supersonic flow datasets, with code released in the AIRS library and data hosted on Hugging Face.

Significance. If the learned timestep predictor can be shown to produce stable, accurate trajectories across shock discontinuities without classical error control, the framework would constitute a practical data-driven replacement for embedded Runge-Kutta or CFL-based adaptivity in high-speed CFD. Public release of datasets and code is a clear positive that supports reproducibility and follow-on work.

major comments (3)
  1. [§3] §3 (Method), timestep-prediction loss: the training objective for the first-phase model is not stated; without an explicit penalty on CFL or TVD violations when the predicted dt is fed to the second-phase integrator, it is impossible to verify that the two-phase separation prevents instability at shocks.
  2. [§4] §4 (Experiments): no quantitative comparison is reported against classical adaptive integrators (e.g., embedded RK4(5) or CFL-limited explicit schemes) on the three supersonic datasets; metrics such as maximum stable dt, L2 error at fixed wall-clock time, or failure rate under out-of-distribution shock strengths are absent.
  3. [§3.3] §3.3 (Conditioning): the neural-ODE and MoE conditioning mechanisms are described at a high level; it remains unclear whether the second-phase advancement is a learned operator or a traditional solver simply parameterized by the predicted dt, which directly affects whether stability guarantees can be inherited.
minor comments (3)
  1. [Abstract] Abstract: the sentence describing the second phase should clarify whether the advancement step is performed by a neural network or by a conventional time integrator conditioned on the predicted dt.
  2. [Figures] Figure captions: several figures lack axis labels or units for the predicted timestep; this obscures whether the learned dt respects physical scales.
  3. [§4.1] Dataset description: the three supersonic cases are introduced without stating the Mach-number range or shock-strength variation used for training versus testing; this information is needed to assess generalization claims.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments on our manuscript. We address each major comment below and indicate the revisions planned for the next version of the paper.

read point-by-point responses

  1. Referee: [§3] §3 (Method), timestep-prediction loss: the training objective for the first-phase model is not stated; without an explicit penalty on CFL or TVD violations when the predicted dt is fed to the second-phase integrator, it is impossible to verify that the two-phase separation prevents instability at shocks.

    Authors: We agree that the training objective for the Phase-1 timestep predictor requires explicit statement. The objective is the mean-squared error against reference timesteps obtained during dataset generation. We further acknowledge that no explicit stability penalty was included. In the revised manuscript we will state the loss function clearly in Section 3 and add a penalty term that discourages timestep predictions leading to CFL or TVD violations when the predicted dt is supplied to the Phase-2 integrator. revision: yes

  2. Referee: [§4] §4 (Experiments): no quantitative comparison is reported against classical adaptive integrators (e.g., embedded RK4(5) or CFL-limited explicit schemes) on the three supersonic datasets; metrics such as maximum stable dt, L2 error at fixed wall-clock time, or failure rate under out-of-distribution shock strengths are absent.

    Authors: We accept that direct quantitative comparisons with classical adaptive integrators would strengthen the evaluation. In the revised manuscript we will add a dedicated subsection reporting comparisons against embedded Runge-Kutta and CFL-limited schemes on all three datasets, including maximum stable dt, L2 error at fixed wall-clock time, and failure rates under out-of-distribution shock strengths. revision: yes

  3. Referee: [§3.3] §3.3 (Conditioning): the neural-ODE and MoE conditioning mechanisms are described at a high level; it remains unclear whether the second-phase advancement is a learned operator or a traditional solver simply parameterized by the predicted dt, which directly affects whether stability guarantees can be inherited.

    Authors: We thank the referee for noting the lack of clarity. The second-phase model is a learned neural operator (not a traditional solver) that receives the current state and the predicted dt as inputs. Conditioning is realized via neural-ODE-style continuous-time embeddings and MoE routing to handle different flow regimes. In the revision we will expand Section 3.3 with architectural diagrams and layer-level details of the conditioning, and we will discuss the implications for stability inheritance. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in two-phase adaptive timestep framework

full rationale

The paper proposes a two-phase ML method (ShockCast) in which a learned model predicts timestep size from fluid fields and the predicted dt is then supplied as conditioning input to a second model that advances the state. No equations, loss functions, or fitting procedures are shown that define the timestep prediction in terms of the advancement output or vice versa. The separation between phases is presented as an architectural choice with physically motivated components and neural-ODE/MoE conditioning; evaluation occurs on independently generated supersonic datasets rather than on quantities derived from the model's own fitted parameters. No load-bearing self-citations, uniqueness theorems, or ansatzes imported from prior author work are invoked to close the derivation. The framework is therefore self-contained against external benchmarks and does not reduce to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is necessarily incomplete; no explicit free parameters, new physical entities, or ad-hoc axioms are stated beyond standard numerical fluid modeling assumptions.

axioms (1)
  • domain assumption Numerical time integration of fluid equations requires timestep sizes chosen to maintain stability and accuracy near discontinuities such as shocks.
    Implicit background assumption in all adaptive time-stepping CFD work.

pith-pipeline@v0.9.0 · 5797 in / 1276 out tokens · 32206 ms · 2026-05-19T10:13:31.460799+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

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

  1. Orbital Transformers for Predicting Wavefunctions in Time-Dependent Density Functional Theory

    cs.LG 2026-03 unverdicted novelty 7.0

    OrbEvo uses equivariant graph transformers to learn the time evolution of TDDFT wavefunction coefficients, accurately reproducing wavefunctions, dipole moments, and absorption spectra on QM9 and MD17 molecular datasets.

Reference graph

Works this paper leans on

114 extracted references · 114 canonical work pages · cited by 1 Pith paper · 2 internal anchors

  1. [1]

    Fourier neural operator for parametric partial differential equations

    Zongyi Li, Nikola Borislavov Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. In International Conference on Learning Representations, 2021

    work page 2021

  2. [2]

    Artificial intelligence for science in quantum, atomistic, and continuum systems

    Xuan Zhang, Limei Wang, Jacob Helwig, Youzhi Luo, Cong Fu, Yaochen Xie, ..., and Shuiwang Ji. Artificial intelligence for science in quantum, atomistic, and continuum systems. arXiv preprint arXiv:2307.08423, 2023

    work page arXiv 2023

  3. [3]

    Anderson

    John D. Anderson. Fundamentals of Aerodynamics. McGraw Hill, New York, 7th edition, 2023

    work page 2023

  4. [4]

    Anderson

    John D. Anderson. Modern Compressible Flow: With Historical Perspective. McGraw-Hill Education, 4th edition, 2020

    work page 2020

  5. [5]

    Neural ordinary differential equations

    Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. Neural ordinary differential equations. Advances in neural information processing systems, 31, 2018

    work page 2018

  6. [6]

    Outrageously large neural networks: The sparsely-gated mixture-of- experts layer

    Noam Shazeer, *Azalia Mirhoseini, *Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture-of- experts layer. In International Conference on Learning Representations, 2017

    work page 2017

  7. [7]

    Finite element and finite volume methods for heat transfer and fluid dynamics

    Junuthula Narasimha Reddy, NK Anand, and Pratanu Roy. Finite element and finite volume methods for heat transfer and fluid dynamics. Cambridge University Press, 2022

    work page 2022

  8. [8]

    Numerical analysis of spectral methods: theory and applications

    David Gottlieb and Steven A Orszag. Numerical analysis of spectral methods: theory and applications. SIAM, 1977

    work page 1977

  9. [9]

    Spectral methods for hyperbolic problems

    David Gottlieb and Jan S Hesthaven. Spectral methods for hyperbolic problems. Journal of Computational and Applied Mathematics, 128(1-2):83–131, 2001

    work page 2001

  10. [10]

    Spectral methods: evolution to complex geometries and applications to fluid dynamics

    Claudio Canuto, M Yousuff Hussaini, Alfio Quarteroni, and Thomas A Zang. Spectral methods: evolution to complex geometries and applications to fluid dynamics . Springer Science & Business Media, 2007

    work page 2007

  11. [11]

    Springer Science & Business Media, 2009

    David A Kopriva.Implementing spectral methods for partial differential equations: Algorithms for scientists and engineers. Springer Science & Business Media, 2009

    work page 2009

  12. [12]

    On the partial difference equations of mathematical physics

    Richard Courant, Kurt Friedrichs, and Hans Lewy. On the partial difference equations of mathematical physics. IBM journal of Research and Development, 11(2):215–234, 1967

    work page 1967

  13. [13]

    Numerical approximation of partial differential equations, volume 64

    Sören Bartels. Numerical approximation of partial differential equations, volume 64. Springer, 2016

    work page 2016

  14. [14]

    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

    work page 2019

  15. [15]

    Three-dimensional laminar flow using physics informed deep neural networks

    Saykat Kumar Biswas and NK Anand. Three-dimensional laminar flow using physics informed deep neural networks. Physics of Fluids, 35(12), 2023

    work page 2023

  16. [16]

    Interfacial conditioning in physics informed neural networks

    Saykat Kumar Biswas and NK Anand. Interfacial conditioning in physics informed neural networks. Physics of Fluids, 36(7), 2024. 10

    work page 2024

  17. [17]

    Hypernetwork-based meta- learning for low-rank physics-informed neural networks

    Woojin Cho, Kookjin Lee, Donsub Rim, and Noseong Park. Hypernetwork-based meta- learning for low-rank physics-informed neural networks. Advances in Neural Information Processing Systems, 36, 2024

    work page 2024

  18. [18]

    Physics-informed neural networks for periodic flows

    Smruti Shah and NK Anand. Physics-informed neural networks for periodic flows. Physics of Fluids, 36(7), 2024

    work page 2024

  19. [19]

    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. Nature machine intelligence, 3(3):218–229, 2021

    work page 2021

  20. [20]

    Multiwavelet-based operator learning for differential equations

    Gaurav Gupta, Xiongye Xiao, and Paul Bogdan. Multiwavelet-based operator learning for differential equations. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors, Advances in Neural Information Processing Systems, 2021

    work page 2021

  21. [21]

    Learning dissipative dynamics in chaotic systems

    Zongyi Li, Miguel Liu-Schiaffini, Nikola Borislavov Kovachki, Burigede Liu, Kamyar Az- izzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Learning dissipative dynamics in chaotic systems. In Advances in Neural Information Processing Systems, 2022

    work page 2022

  22. [22]

    Fourier neu- ral operator with learned deformations for pdes on general geometries

    Zongyi Li, Daniel Zhengyu Huang, Burigede Liu, and Anima Anandkumar. Fourier neu- ral operator with learned deformations for pdes on general geometries. arXiv preprint arXiv:2207.05209, 2022

    work page arXiv 2022

  23. [23]

    Geometry-informed neural operator for large-scale 3d pdes

    Zongyi Li, Nikola Borislavov Kovachki, Chris Choy, Boyi Li, Jean Kossaifi, Shourya Prakash Otta, Mohammad Amin Nabian, Maximilian Stadler, Christian Hundt, Kamyar Azizzade- nesheli, et al. Geometry-informed neural operator for large-scale 3d pdes. arXiv preprint arXiv:2309.00583, 2023

    work page arXiv 2023

  24. [24]

    Transform once: Efficient operator learning in frequency domain

    Michael Poli, Stefano Massaroli, Federico Berto, Jinkyoo Park, Tri Dao, Christopher Re, and Stefano Ermon. Transform once: Efficient operator learning in frequency domain. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho, editors, Advances in Neural Information Processing Systems, 2022

    work page 2022

  25. [25]

    Nomad: Nonlinear manifold decoders for operator learning

    Jacob Seidman, Georgios Kissas, Paris Perdikaris, and George J Pappas. Nomad: Nonlinear manifold decoders for operator learning. Advances in Neural Information Processing Systems, 35:5601–5613, 2022

    work page 2022

  26. [26]

    Neural operator: Learning maps between function spaces with applications to PDEs

    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

    work page 2023

  27. [27]

    Openfwi: Large-scale multi-structural benchmark datasets for full waveform inversion

    Chengyuan Deng, Shihang Feng, Hanchen Wang, Xitong Zhang, Peng Jin, Yinan Feng, Qili Zeng, Yinpeng Chen, and Youzuo Lin. Openfwi: Large-scale multi-structural benchmark datasets for full waveform inversion. Advances in Neural Information Processing Systems, 35:6007–6020, 2022

    work page 2022

  28. [28]

    Learning large-scale subsurface simulations with a hybrid graph network simulator

    Tailin Wu, Qinchen Wang, Yinan Zhang, Rex Ying, Kaidi Cao, Rok Sosic, Ridwan Jalali, Has- san Hamam, Marko Maucec, and Jure Leskovec. Learning large-scale subsurface simulations with a hybrid graph network simulator. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 4184–4194, 2022

    work page 2022

  29. [29]

    Pangu-weather: A 3d high-resolution model for fast and accurate global weather forecast

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Pangu-weather: A 3d high-resolution model for fast and accurate global weather forecast. arXiv preprint arXiv:2211.02556, 2022

    work page arXiv 2022

  30. [30]

    Graphcast: Learning skillful medium-range global weather forecasting

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire For- tunato, Alexander Pritzel, Suman Ravuri, Timo Ewalds, Ferran Alet, Zach Eaton-Rosen, et al. Graphcast: Learning skillful medium-range global weather forecasting. arXiv preprint arXiv:2212.12794, 2022. 11

    work page arXiv 2022

  31. [31]

    FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

    Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopad- hyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, et al. Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators. arXiv preprint arXiv:2202.11214, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  32. [32]

    Gencast: Diffusion- based ensemble forecasting for medium-range weather

    Ilan Price, Alvaro Sanchez-Gonzalez, Ferran Alet, Timo Ewalds, Andrew El-Kadi, Jacklynn Stott, Shakir Mohamed, Peter Battaglia, Remi Lam, and Matthew Willson. Gencast: Diffusion- based ensemble forecasting for medium-range weather. arXiv preprint arXiv:2312.15796, 2023

    work page arXiv 2023

  33. [33]

    Climax: A foundation model for weather and climate

    Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K Gupta, and Aditya Grover. Climax: A foundation model for weather and climate. In Proceedings of the 40th International Conference on Machine Learning, 2023

    work page 2023

  34. [34]

    Airfrans: High fidelity computational fluid dynamics dataset for approximating reynolds-averaged navier- stokes solutions

    Florent Bonnet, Jocelyn Ahmed Mazari, Paola Cinella, and Patrick Gallinari. Airfrans: High fidelity computational fluid dynamics dataset for approximating reynolds-averaged navier- stokes solutions. In 36th Conference on Neural Information Processing Systems (NeurIPS

    work page

  35. [35]

    Track on Datasets and Benchmarks, 2022

    work page 2022

  36. [36]

    A geometry-aware message passing neural network for modeling aerodynamics over airfoils

    Jacob Helwig, Xuan Zhang, Haiyang Yu, and Shuiwang Ji. A geometry-aware message passing neural network for modeling aerodynamics over airfoils. arXiv preprint arXiv:2412.09399, 2024

    work page arXiv 2024

  37. [37]

    Factorized fourier neural operators

    Alasdair Tran, Alexander Mathews, Lexing Xie, and Cheng Soon Ong. Factorized fourier neural operators. In The Eleventh International Conference on Learning Representations , 2023

    work page 2023

  38. [38]

    U-fno—an enhanced fourier neural operator-based deep-learning model for multiphase flow

    Gege Wen, Zongyi Li, Kamyar Azizzadenesheli, Anima Anandkumar, and Sally M Benson. U-fno—an enhanced fourier neural operator-based deep-learning model for multiphase flow. Advances in Water Resources, 163:104180, 2022

    work page 2022

  39. [39]

    Group equivariant Fourier neural operators for partial differential equations

    Jacob Helwig, Xuan Zhang, Cong Fu, Jerry Kurtin, Stephan Wojtowytsch, and Shuiwang Ji. Group equivariant Fourier neural operators for partial differential equations. In Proceedings of the 40th International Conference on Machine Learning, 2023

    work page 2023

  40. [40]

    Real-time high-resolution co 2 geological storage prediction using nested fourier neural operators

    Gege Wen, Zongyi Li, Qirui Long, Kamyar Azizzadenesheli, Anima Anandkumar, and Sally M Benson. Real-time high-resolution co 2 geological storage prediction using nested fourier neural operators. Energy & Environmental Science, 16(4):1732–1741, 2023

    work page 2023

  41. [41]

    Spherical fourier neural operators: Learning stable dynamics on the sphere

    Boris Bonev, Thorsten Kurth, Christian Hundt, Jaideep Pathak, Maximilian Baust, Karthik Kashinath, and Anima Anandkumar. Spherical fourier neural operators: Learning stable dynamics on the sphere. In Proceedings of the 40th International Conference on Machine Learning, 2023

    work page 2023

  42. [42]

    Sinenet: Learning temporal dynamics in time-dependent partial differential equations

    Xuan Zhang, Jacob Helwig, Yuchao Lin, Yaochen Xie, Cong Fu, Stephan Wojtowytsch, and Shuiwang Ji. Sinenet: Learning temporal dynamics in time-dependent partial differential equations. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  43. [43]

    Convolutional neural operators for robust and accurate learning of pdes

    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. Advances in Neural Information Processing Systems, 36, 2024

    work page 2024

  44. [44]

    Choose a transformer: Fourier or galerkin

    Shuhao Cao. Choose a transformer: Fourier or galerkin. Advances in neural information processing systems, 34:24924–24940, 2021

    work page 2021

  45. [45]

    Transformer for partial differential equations’ operator learning

    Zijie Li, Kazem Meidani, and Amir Barati Farimani. Transformer for partial differential equations’ operator learning. Transactions on Machine Learning Research, 2023

    work page 2023

  46. [46]

    EAGLE: Large-scale learning of turbulent fluid dynamics with mesh transformers

    Steeven Janny, Aurélien Bénéteau, Madiha Nadri, Julie Digne, Nicolas Thome, and Christian Wolf. EAGLE: Large-scale learning of turbulent fluid dynamics with mesh transformers. In International Conference on Learning Representations, 2023. 12

    work page 2023

  47. [47]

    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. In Forty-first International Conference on Machine Learning, 2024

    work page 2024

  48. [48]

    Universal physics transformers: A framework for efficiently scaling neural operators

    Benedikt Alkin, Andreas Fürst, Simon Lucas Schmid, Lukas Gruber, Markus Holzleitner, and Johannes Brandstetter. Universal physics transformers: A framework for efficiently scaling neural operators. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    work page 2024

  49. [49]

    Neural Operator: Graph Kernel Network for Partial Differential Equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Graph kernel network for partial differential equations. arXiv preprint arXiv:2003.03485, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  50. [50]

    Multipole graph neural operator for parametric partial differential equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Andrew Stuart, Kaushik Bhattacharya, and Anima Anandkumar. Multipole graph neural operator for parametric partial differential equations. Advances in Neural Information Processing Systems, 33:6755–6766, 2020

    work page 2020

  51. [51]

    Worrall, and Max Welling

    Johannes Brandstetter, Daniel E. Worrall, and Max Welling. Message passing neural PDE solvers. In International Conference on Learning Representations, 2022

    work page 2022

  52. [52]

    Physics-embedded neural networks: Graph neural pde solvers with mixed boundary conditions

    Masanobu Horie and Naoto Mitsume. Physics-embedded neural networks: Graph neural pde solvers with mixed boundary conditions. Advances in Neural Information Processing Systems, 35:23218–23229, 2022

    work page 2022

  53. [53]

    Learned coarse models for efficient turbulence simulation

    Kimberly Stachenfeld, Drummond B Fielding, Dmitrii Kochkov, Miles Cranmer, Tobias Pfaff, Jonathan Godwin, Can Cui, Shirley Ho, Peter Battaglia, and Alvaro Sanchez-Gonzalez. Learned coarse models for efficient turbulence simulation. In International Conference on Learning Representations, 2021

    work page 2021

  54. [54]

    Machine learning–accelerated computational fluid dynamics

    Dmitrii Kochkov, Jamie A Smith, Ayya Alieva, Qing Wang, Michael P Brenner, and Stephan Hoyer. Machine learning–accelerated computational fluid dynamics. Proceedings of the National Academy of Sciences, 118(21):e2101784118, 2021

    work page 2021

  55. [55]

    Towards multi-spatiotemporal-scale generalized PDE modeling

    Jayesh K Gupta and Johannes Brandstetter. Towards multi-spatiotemporal-scale generalized PDE modeling. Transactions on Machine Learning Research, 2023

    work page 2023

  56. [56]

    2024 , url =

    Maximilian Herde, Bogdan Raoni´c, Tobias Rohner, Roger Käppeli, Roberto Molinaro, Em- manuel de Bézenac, and Siddhartha Mishra. Poseidon: Efficient foundation models for pdes. arXiv preprint arXiv:2405.19101, 2024

    work page arXiv 2024

  57. [57]

    Improved denoising diffusion probabilistic models

    Alexander Quinn Nichol and Prafulla Dhariwal. Improved denoising diffusion probabilistic models. In International conference on machine learning, pages 8162–8171. PMLR, 2021

    work page 2021

  58. [58]

    Diffusion models beat gans on image synthesis

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis. Advances in neural information processing systems, 34:8780–8794, 2021

    work page 2021

  59. [59]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016

    work page 2016

  60. [60]

    Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations

    Yiping Lu, Aoxiao Zhong, Quanzheng Li, and Bin Dong. Beyond finite layer neural networks: Bridging deep architectures and numerical differential equations. In International conference on machine learning, pages 3276–3285. PMLR, 2018

    work page 2018

  61. [61]

    Stable architectures for deep neural networks

    Eldad Haber and Lars Ruthotto. Stable architectures for deep neural networks. Inverse problems, 34(1):014004, 2017

    work page 2017

  62. [62]

    Deep neural networks motivated by partial differential equations

    Lars Ruthotto and Eldad Haber. Deep neural networks motivated by partial differential equations. Journal of Mathematical Imaging and Vision, 62(3):352–364, 2020

    work page 2020

  63. [63]

    On neural differential equations

    Patrick Kidger. On neural differential equations. arXiv preprint arXiv:2202.02435, 2022. 13

    work page arXiv 2022

  64. [64]

    Adaptive mixtures of local experts

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts. Neural computation, 3(1):79–87, 1991

    work page 1991

  65. [65]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity

    William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. Journal of Machine Learning Research, 23(120):1–39, 2022

    work page 2022

  66. [66]

    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 International Conference on Machine Learning, pages 12556–12569. PMLR, 2023

    work page 2023

  67. [67]

    Learning mesh-based simulation with graph networks

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter Battaglia. Learning mesh-based simulation with graph networks. In International Conference on Learning Repre- sentations, 2021

    work page 2021

  68. [68]

    M2n: Mesh movement networks for pde solvers

    Wenbin Song, Mingrui Zhang, Joseph G Wallwork, Junpeng Gao, Zheng Tian, Fanglei Sun, Matthew Piggott, Junqing Chen, Zuoqiang Shi, Xiang Chen, et al. M2n: Mesh movement networks for pde solvers. Advances in Neural Information Processing Systems, 35:7199–7210, 2022

    work page 2022

  69. [69]

    Towards universal mesh movement networks

    Mingrui Zhang, Chunyang Wang, Stephan C Kramer, Joseph G Wallwork, Siyi Li, Jiancheng Liu, Xiang Chen, and Matthew Piggott. Towards universal mesh movement networks. Ad- vances in Neural Information Processing Systems, 37:14934–14961, 2024

    work page 2024

  70. [70]

    Learn- ing controllable adaptive simulation for multi-scale physics

    Tailin Wu, Takashi Maruyama, Qingqing Zhao, Gordon Wetzstein, and Jure Leskovec. Learn- ing controllable adaptive simulation for multi-scale physics. In NeurIPS 2022 AI for Science: Progress and Promises, 2022

    work page 2022

  71. [71]

    Swarm reinforcement learning for adaptive mesh refinement

    Niklas Freymuth, Philipp Dahlinger, Tobias Daniel Würth, Simon Reisch, Luise Kärger, and Gerhard Neumann. Swarm reinforcement learning for adaptive mesh refinement. In Thirty-seventh Conference on Neural Information Processing Systems, 2023

    work page 2023

  72. [72]

    Reinforcement learning for adaptive mesh refinement

    Jiachen Yang, Tarik Dzanic, Brenden Petersen, Jun Kudo, Ketan Mittal, Vladimir Tomov, Jean- Sylvain Camier, Tuo Zhao, Hongyuan Zha, Tzanio Kolev, et al. Reinforcement learning for adaptive mesh refinement. In International conference on artificial intelligence and statistics, pages 5997–6014. PMLR, 2023

    work page 2023

  73. [73]

    Tante: Time-adaptive operator learning via neural taylor expansion

    Zhikai Wu, Shiyang Zhang, Sizhuang He, Sifan Wang, Min Zhu, Anran Jiao, Lu Lu, and David van Dijk. Tante: Time-adaptive operator learning via neural taylor expansion. arXiv preprint arXiv:2502.08574, 2025

    work page arXiv 2025

  74. [74]

    Space and time continuous physics simulation from partial observations

    Steeven Janny, Madiha Nadri, Julie Digne, and Christian Wolf. Space and time continuous physics simulation from partial observations. In The Twelfth International Conference on Learning Representations, 2024

    work page 2024

  75. [75]

    Vector- ized conditional neural fields: A framework for solving time-dependent parametric partial differential equations

    Jan Hagnberger, Marimuthu Kalimuthu, Daniel Musekamp, and Mathias Niepert. Vector- ized conditional neural fields: A framework for solving time-dependent parametric partial differential equations. In Forty-first International Conference on Machine Learning, 2024

    work page 2024

  76. [76]

    Hierarchical deep learning of multi- scale differential equation time-steppers

    Yuying Liu, J Nathan Kutz, and Steven L Brunton. Hierarchical deep learning of multi- scale differential equation time-steppers. Philosophical Transactions of the Royal Society A, 380(2229):20210200, 2022

    work page 2022

  77. [77]

    Hierarchical deep learning-based adaptive time stepping scheme for multiscale simulations

    Asif Hamid, Danish Rafiq, Shahkar Ahmad Nahvi, and Mohammad Abid Bazaz. Hierarchical deep learning-based adaptive time stepping scheme for multiscale simulations. Engineering Applications of Artificial Intelligence, 133:108430, 2024

    work page 2024

  78. [78]

    A survey of several finite difference methods for systems of nonlinear hyperbolic conservation laws

    Gary A Sod. A survey of several finite difference methods for systems of nonlinear hyperbolic conservation laws. Journal of computational physics, 27(1):1–31, 1978

    work page 1978

  79. [79]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11976–11986, 2022. 14

    work page 2022

  80. [80]

    Learning to simulate complex physics with graph networks

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter Battaglia. Learning to simulate complex physics with graph networks. In International conference on machine learning, pages 8459–8468. PMLR, 2020

    work page 2020

Showing first 80 references.