mlsanityData to 7 Oct 2026

No Free Lunch in Flow Surrogates under Time-Varying Boundary Conditions: A Two-Regime Study

arXiv 2607.23667 ↗First posted 26 Jul 2026math.NA cs.LG cs.NAcited by 0

Not on the map. Why?

Abstract (arXiv)

We test whether an architecture that succeeds on a simple flow regime also succeeds on a richer one, with each trained separately on each regime. We explore two transient flows under time-varying boundary conditions: the three-dimensional slurry film in chemical-mechanical planarisation (CMP), central to semiconductor manufacturing, and the two-dimensional K'arm'an vortex street (KVS). Eight surrogate models on one shared pipeline differ in whether they learn the full field or a latent representation, and in whether they predict in one shot or step by step. No single architecture wins both regimes. On the film, a one-shot full-field model reconstructs the cumulative wall shear stress to 2.7% relative error. On the wake, a latent autoregressive DeepONet retains 90% of the shedding power that direct and one-shot models damp to almost zero. The treatment of time decides the outcome. The self-sustained wake calls for autoregressive feedback and the boundary-driven film for a direct map. Pointwise RMSE hides the damped oscillation on the wake, compresses the sixfold lead on the film's process target, and picks the damped model under wake extrapolation. The evaluation scores five physical questions. Trained surrogates answer queries 103 to 104 times faster than the finite-element solver and pay off from the first query beyond the training set on the film and from the third on the wake. Neither the winning architecture nor its validation holds across regimes. The choice of surrogate should follow the dynamical character of the target flow, and its validation should resolve the failure modes.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 40 of its 77 references that are in the map. First posted on arXiv 26 Jul 2026.

Builds on

40 of 77 references.

References · 40

  1. 2023Mamba: Linear-Time Sequence Modeling with Selective State Spacescited by 4,616
  2. 2014Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translationcited by 4,016
  3. 2020Fourier Neural Operator for Parametric Partial Differential Equationscited by 2,806
  4. 2021Efficiently Modeling Long Sequences with Structured State Spacescited by 2,424
  5. 2019DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operatorscited by 1,965
  6. 2017Deep learning for universal linear embeddings of nonlinear dynamicscited by 491
  7. 2022On the Parameterization and Initialization of Diagonal State Space Modelscited by 418
  8. 2022Message Passing Neural PDE Solverscited by 368
  9. 2021Choose a Transformer: Fourier or Galerkincited by 333
  10. 2023GNOT: A General Neural Operator Transformer for Operator Learningcited by 321
  11. 2023Geometry-Informed Neural Operator for Large-Scale 3D PDEscited by 293
  12. 2020Model Reduction and Neural Networks for Parametric PDEscited by 257
28 more
  1. 2022U-NO: U-shaped Neural Operatorscited by 219
  2. 2023PDE-Refiner: Achieving Accurate Long Rollouts with Neural PDE Solverscited by 205
  3. 2020Solver-in-the-Loop: Learning from Differentiable Physics to Interact with Iterative PDE-Solverscited by 201
  4. 2014Dynamic mode decomposition with controlcited by 128
  5. 2021Long-time integration of parametric evolution equations with physics-informed DeepONetscited by 87
  6. 2024Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equationscited by 87
  7. 2021Improved architectures and training algorithms for deep operator networkscited by 85
  8. 2024Operator Learning: Algorithms and Analysiscited by 84
  9. 2022Learning two-phase microstructure evolution using neural operators and autoencoder architecturescited by 72
  10. 2023Sequential Deep Operator Networks (S-DeepONet) for Predicting Full-field Solutions Under Time-dependent Loadscited by 47
  11. 2024AROMA: Preserving Spatial Structure for Latent PDE Modeling with Local Neural Fieldscited by 38
  12. 2024On the Benefits of Memory for Modeling Time-Dependent PDEscited by 30
  13. 2023Variational Autoencoding Neural Operatorscited by 22
  14. 2024Active Learning for Neural PDE Solverscited by 22
  15. 2025Recurrent Neural Operators: Stable Long-Term PDE Predictioncited by 15
  16. 2024On latent dynamics learning in nonlinear reduced order modelingcited by 14
  17. 2025TI-DeepONet: Learnable Time Integration for Stable Long-Term Extrapolationcited by 14
  18. 2025Fusion-DeepONet: A Data-Efficient Neural Operator for Geometry-Dependent Hypersonic and Supersonic Flowscited by 13
  19. 2024Learning time-dependent PDE via graph neural networks and deep operator network for robust accuracy on irregular gridscited by 7
  20. 2025Neural Emulator Superiority: When Machine Learning for PDEs Surpasses its Training Datacited by 6
  21. 2024What You See is Not What You Get: Neural Partial Differential Equations and The Illusion of Learningcited by 5
  22. 2025Geometric and Physical Constraints Synergistically Enhance Neural PDE Surrogatescited by 5
  23. 2025Time Resolution Independent Operator Learningcited by 5
  24. 2025Physics-Informed Time-Integrated DeepONet: Temporal Tangent Space Operator Learning for High-Accuracy Inferencecited by 5
  25. 2025Variationally correct operator learning: Reduced basis neural operator with a posteriori error estimationcited by 5
  26. 2026Performance of Neural and Polynomial Operator Surrogatescited by 5
  27. 2024On instabilities in neural network-based physics simulatorscited by 2
  28. 2026Pointwise Metrics Mislead: An Evaluation Protocol for Multimodal Inverse Problemscited by 1

Built on by

Not cited yet.