pith. sign in

arxiv: 2606.18499 · v1 · pith:4QT7KI7Dnew · submitted 2026-06-16 · ⚛️ physics.flu-dyn

Solution of the Newtonian plane Couette flow with dynamic wall slip using machine-learning methods

Georgia Foutsitzi , Nikolaos Antoniadis , Georgios C. Georgiou This is my paper

Pith reviewed 2026-06-26 22:10 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords DeepONetPINNCouette flowdynamic wall slipmachine learningfluid dynamicssurrogate modelingoperator learning
0
0 comments X

The pith

DeepONets learn the continuous solution operator for plane Couette flow with dynamic wall slip and deliver fast predictions across parameter ranges.

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

This paper compares Physics-Informed Neural Networks and data-driven Deep Operator Networks for predicting the time evolution of plane Newtonian Couette flow with dynamic wall slip. PINNs achieve high point-wise accuracy for specific parameters but require retraining for each new instance. DeepONets trained on high-fidelity numerical data learn the mapping from slip boundary conditions and upper wall velocities to the velocity field, generalizing to unseen and out-of-distribution cases with mean relative errors of 0.36 percent and 0.88 percent. The operator network provides near-instantaneous inference at a 540 times speedup over the numerical solver, positioning it as an efficient surrogate for parametric exploration and real-time forecasting.

Core claim

A data-driven Deep Operator Network trained on numerical solutions learns the continuous solution operator that maps input functions of dynamic slip boundary conditions and upper wall velocities to the corresponding velocity field solutions for plane Newtonian Couette flow, achieving mean relative errors of 0.36 percent on unseen signals and 0.88 percent on out-of-distribution signals while running approximately 540 times faster than the Crank-Nicolson solver.

What carries the argument

Deep Operator Network (DeepONet) that approximates the nonlinear operator from boundary condition input functions to the flow solution field.

If this is right

  • DeepONet enables rapid parametric exploration without retraining for each new set of slip conditions or wall velocities.
  • Inference time is reduced by a factor of 540 relative to the numerical solver and by 30.5 percent relative to PINNs.
  • The network supports real-time fluid dynamics forecasting applications.
  • Robust generalization holds for both unseen and out-of-distribution input signals.

Where Pith is reading between the lines

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

  • Operator-learning methods may reduce computational cost for other time-dependent flows whose boundary conditions vary over wide ranges.
  • The observed generalization suggests the learned operator could be tested on experimental velocity data to check consistency beyond numerical assumptions.
  • Embedding additional physics constraints into the training loss could lower the volume of required numerical data while preserving accuracy.

Load-bearing premise

The high-fidelity numerical data generated by the Crank-Nicolson scheme accurately represents the true solution operator across the full range of dynamic slip boundary conditions and upper wall velocities used for training and testing.

What would settle it

Generating independent numerical solutions for parameter values outside the training and test distributions and checking whether the DeepONet mean relative error stays below one percent.

read the original abstract

This study presents a comparative investigation of Physics-Informed Neural Networks (PINNs) and data-driven Deep Operator Networks (DeepONets) for predicting the evolution of plane Newtonian Couette flow with dynamic wall slip. While traditional numerical methods, such as the Crank-Nicolson scheme, offer high accuracy, their computational demand poses challenges in real-time applications. To address this, we first implement a PINN framework to solve the governing equations for specific physical parameters. Subsequently, we develop a data-driven DeepONet, trained on high-fidelity numerical data, to learn the continuous solution operator across a broad range of slip boundary conditions and upper wall velocities. Our results indicate that while the PINN achieved superior point-wise precision with a relative L_2 error of 0.083%, it remains constrained by the requirement for instance-specific retraining. In contrast, the DeepONet demonstrates robust generalization on unseen and out-of-distribution signals with a mean relative error of 0.36% and 0.88%, respectively. Most notably, it provides near-instantaneous inference, achieving a speedup factor of approximately 540X over the numerical solver and 30.5% over the PINN. This work demonstrates the synergy between physics-based and data-driven architectures and establishes DeepONet as a highly efficient surrogate model for rapid parametric exploration and real-time fluid dynamics forecasting.

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

0 major / 3 minor

Summary. The manuscript compares Physics-Informed Neural Networks (PINNs) and data-driven Deep Operator Networks (DeepONets) for solving the time-dependent Newtonian plane Couette flow with dynamic wall slip. It reports that a PINN achieves 0.083% relative L2 error for specific parameters but requires per-instance retraining, while a DeepONet trained on Crank-Nicolson data generalizes to unseen and out-of-distribution inputs with mean relative errors of 0.36% and 0.88%, respectively, and delivers a 540X inference speedup over the numerical solver (and 30.5% over the PINN).

Significance. If the reported generalization and speedups hold under the stated training regime, the work demonstrates that operator-learning architectures can serve as practical surrogates for parametric boundary-value problems in viscous flows, enabling rapid exploration of slip conditions without repeated numerical solves. The explicit numerical error values and speedup factors relative to both the Crank-Nicolson scheme and the PINN constitute a concrete, falsifiable benchmark for this class of problems.

minor comments (3)
  1. The abstract states concrete error values and speedups, but the manuscript should include an explicit description of the training/test split procedure, the range of upper-wall velocities and slip parameters used for out-of-distribution testing, and any regularization or normalization applied to the input functions (e.g., in the DeepONet branch/trunk networks).
  2. Figure captions and axis labels should be expanded to indicate whether the plotted velocity profiles are instantaneous snapshots or time-averaged quantities, and whether the reported relative errors are computed in L2 or L-infinity norm.
  3. The manuscript would benefit from a short table summarizing the network architectures (depth, width, activation), optimizer settings, and number of training samples for both the PINN and DeepONet, to allow direct reproduction of the reported timings.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. No specific major comments were raised in the report.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper trains a data-driven DeepONet on outputs from the Crank-Nicolson scheme to learn the solution operator for Couette flow with dynamic slip, then reports generalization error and speedup relative to that same numerical data. This is standard supervised learning with no derivation chain, no self-definitional mapping of outputs back to inputs, no fitted parameters renamed as independent predictions, and no load-bearing self-citations or uniqueness theorems. The central claims concern empirical approximation quality and inference speed; they do not reduce to the inputs by construction and remain externally falsifiable against the numerical solver.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that the Crank-Nicolson numerical solutions constitute ground truth for training and that standard neural-network training procedures suffice to learn the solution operator; no new physical axioms or entities are introduced.

free parameters (1)
  • neural network architecture and training hyperparameters
    Standard ML choices such as layer widths, activation functions, and optimizer settings are fitted during training but not enumerated in the abstract.
axioms (2)
  • domain assumption The incompressible Newtonian Navier-Stokes equations reduce to the standard Couette flow PDE under the stated assumptions.
    Invoked implicitly as the governing equations solved by both the numerical scheme and the PINN.
  • domain assumption The dynamic wall slip boundary condition is correctly implemented in both the numerical solver and the ML models.
    Required for the training data and loss functions to match the physical problem.

pith-pipeline@v0.9.1-grok · 5788 in / 1371 out tokens · 42581 ms · 2026-06-26T22:10:44.735763+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

45 extracted references · 33 canonical work pages · 3 internal anchors

  1. [1]

    Denn, Extrusion instabilities and wall slip, Annu

    M.M. Denn, Extrusion instabilities and wall slip, Annu. Rev. Fluid Mech. 33 (2001) 265– 287

    2001

  2. [2]

    Hatzikiriakos, Wall slip of molten polymers, Prog

    S.G. Hatzikiriakos, Wall slip of molten polymers, Prog . Polym. Sci. 37 (2012) 624 –643. https://doi.org/10.1016/j.progpolymsci.2011.09.004

    work page doi:10.1016/j.progpolymsci.2011.09.004 2012

  3. [3]

    Malkin, S.A

    A.Y. Malkin, S.A. Patlazhan, Wall slip for complex liquids – Phenomenon and its causes, Adv. Colloid Interface Sci. 257 (2018) 42–57. https://doi.org/10.1016/j.cis.2018.05.008

    work page doi:10.1016/j.cis.2018.05.008 2018

  4. [4]

    Neto, D.R

    C. Neto, D.R. Evans, E. Bonaccurso, H. -J. Butt, V.S.J. Craig, Boundary slip in Newtonian liquids: a review of experimental studies, Rep. Prog. Phys. 68 (2005) 2859 –2897. https://doi.org/10.1088/0034-4885/68/12/R05. 23

    work page doi:10.1088/0034-4885/68/12/r05 2005

  5. [5]

    A.A. Moud, J. Piette, M. Danesh, G.C. Georgiou, S.G. Hatzikiriakos, Apparent slip in colloidal suspensions, J. Rheol. 66 (2022) 79–90. https://doi.org/10.1122/8.0000302

    work page doi:10.1122/8.0000302 2022

  6. [6]

    Hatzikiriakos, Slip mechanisms in complex fluid flows, Soft Matter 11 (2015) 7851–

    S.G. Hatzikiriakos, Slip mechanisms in complex fluid flows, Soft Matter 11 (2015) 7851–

    2015

  7. [7]

    https://doi.org/10.1039/C5SM01711D

    work page doi:10.1039/c5sm01711d

  8. [8]

    Navier, Mémoire sur les lois du mouvement des fluides, Mém

    C. Navier, Mémoire sur les lois du mouvement des fluides, Mém. Acad. Sci. Inst. Fr. 6 (1823) 389–440

  9. [9]

    Pearson, C.J.S

    J.R.A. Pearson, C.J.S. Petrie, On the melt-flow instability of extruded polymers, (1965)

    1965

  10. [10]

    Hatzikiriakos, N

    S.G. Hatzikiriakos, N. Kalogerakis, A dynamic slip velocity model for molten polymers based on a network kinetic theory, Rheol. Acta 33 (1994) 38–47

    1994

  11. [11]

    Ebrahimi, V.K

    M. Ebrahimi, V.K. Konaganti, S.G. Hatzikiriakos, Dynamic slip of polydisperse linear polymers using partitioned plate, Phys. Fluids 30 (2018)

    2018

  12. [12]

    Thalakkottor, K

    J.J. Thalakkottor, K. Mohseni, Analysis of boundary slip in a flow with an oscillating wall, Phys. Rev. E 87 (2013) 033018

    2013

  13. [13]

    Kazatchkov, S.G

    I.B. Kazatchkov, S.G. Hatzikiriakos, Relaxation effects of slip in shear flow of linear molten polymers, Rheol. Acta 49 (2010) 267–274

    2010

  14. [14]

    Kaoullas, G.C

    G. Kaoullas, G.C. Georgiou, Start -up and cessation Newtonian Poiseuille and Couette flows with dynamic wall slip, Meccanica 50 (2015) 1747 –1760. https://doi.org/10.1007/s11012-015-0127-y

    work page doi:10.1007/s11012-015-0127-y 2015

  15. [15]

    Farragui, O

    M.E. Farragui, O. Souhar, G.C. Georgiou, Newtonian annular Poiseuille and Couette flows with dynamic wall slip, Eur. J. Mech. B Fluids 103 (2024) 136–144

    2024

  16. [16]

    El Farragui, O

    M. El Farragui, O. Souhar, G.C. Georgiou, Transient Newtonian Poiseuille flow in a square channel with dynamic wall slip, Phys . Fluids 37 (2025) 023144. https://doi.org/10.1063/5.0253131

    work page doi:10.1063/5.0253131 2025

  17. [17]

    Abou Hasan, E.A.A

    M.M. Abou Hasan, E.A.A. Ahmed, A.F. Ghaleb, M.S. Abou -Dina, G.C. Georgiou, Numerical solution of the Newtonian plane Couette flow with linear dynamic wall slip, Fluids 9 (2024) 172. https://doi.org/10.3390/fluids9080172

    work page doi:10.3390/fluids9080172 2024

  18. [18]

    Lauga, M

    E. Lauga, M. Brenner, H. Stone, Microfluidics: The No -Slip Boundary Condition, in: C. Tropea, A.L. Yarin, J.F. Foss (Eds.), Springer Handbook of Experimental Fluid Mechanics, Springer Berlin Heidelberg, Berlin, Heidelberg, 2007: pp. 1219 –1240. https://doi.org/10.1007/978-3-540-30299-5_19

    work page doi:10.1007/978-3-540-30299-5_19 2007

  19. [19]

    Brunton, Applying machine learning to study fluid mechanics, Acta Mech

    S.L. Brunton, Applying machine learning to study fluid mechanics, Acta Mech. Sin. 37 (2021) 1718–1726. https://doi.org/10.1007/s10409-021-01143-6

    work page doi:10.1007/s10409-021-01143-6 2021

  20. [20]

    H. Wang, Y. Cao, Z. Huang, Y. Liu, P. Hu, X. Luo, Z. Song, W. Zhao, J. Liu, J. Sun, S. Zhang, L. Wei, Y. Wang, T. Wu, Z. -M. Ma, Y. Sun, Recent Advances on Machine Learning for Computational Fluid Dynamics: A Survey, (2024). https://doi.org/10.48550/arXiv.2408.12171

    work page doi:10.48550/arxiv.2408.12171 2024

  21. [21]

    Y. Li, J. Chang, C. Kong, W. Bao, Recent progress of machine learning in flow modeling and active flow control, Chin . J. Aeronaut. 35 (2022) 14 –44. https://doi.org/10.1016/j.cja.2021.07.027

    work page doi:10.1016/j.cja.2021.07.027 2022

  22. [22]

    Catsoulis, J

    S. Catsoulis, J. -S. Singh, C. Narayanan, D. Lakehal, Integrating supervised learning and applied computational multi -fluid dynamics, Int . J. Multiph. Flow 157 (2022) 104221. https://doi.org/10.1016/j.ijmultiphaseflow.2022.104221

    work page doi:10.1016/j.ijmultiphaseflow.2022.104221 2022

  23. [23]

    X. Sun, W. Cao, X. Shan, Y. Liu, W. Zhang, A generalized framework for integrating machine learning into computational fluid dynamics, J . Comp. Sci. 82 (2024) 102404. https://doi.org/10.1016/j.jocs.2024.102404

    work page doi:10.1016/j.jocs.2024.102404 2024

  24. [24]

    Zhang, Y

    C. Zhang, Y. Zhou, J. Li, S. Chu, Q. Qin, L. Huang, Data-driven prediction of wave response for a modular floating solar array, Ocean Eng . 349 (2026) 124192. https://doi.org/10.1016/j.oceaneng.2026.124192

    work page doi:10.1016/j.oceaneng.2026.124192 2026

  25. [25]

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

    M. Raissi, P. Perdikaris, G.E. Karniadakis, Physics -informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial 24 differential equations, J . Comp. Phys. 378 (2019) 686 –707. https://doi.org/10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  26. [26]

    S. Cai, Z. Mao, Z. Wang, M. Yin, G.E. Karniadakis, Physics -informed neural networks (PINNs) for fluid mechanics: a review, Acta Mech. Sin. 37 (2021) 1727 –1738. https://doi.org/10.1007/s10409-021-01148-1

    work page doi:10.1007/s10409-021-01148-1 2021

  27. [27]

    Z. Tao, K. Xu, F. Liu, LSTM -PINN: An hybrid method for prediction of steady -state electrohydrodynamic flow, J . Comp. Phys. 548 (2026) 114586. https://doi.org/10.1016/j.jcp.2025.114586

    work page doi:10.1016/j.jcp.2025.114586 2026

  28. [28]

    D. Ra, J. Lee, M. Lee, S. Kwak, S. Lee, S. Lee, Physics -informed machine learning across manufacturing processes: Recent advances, challenges, and directions, J . Manuf. Syst. 85 (2026) 72–95. https://doi.org/10.1016/j.jmsy.2026.01.002

    work page doi:10.1016/j.jmsy.2026.01.002 2026

  29. [29]

    Kovachki, Z

    N. Kovachki, Z. Li, B. Liu, K. Azizzadenesheli, K. Bhattacharya, A. Stuart, A. Anandkumar, Neural operator: Learning maps between function spaces with applications to PDEs, J. Mach. Learn. Res. 24 (2023) 1–97

    2023

  30. [30]

    H. Li, Y. Miao, Z.S. Khodaei, M.H. Aliabadi, An architectural analysis of DeepOnet and a general extension of the physics -informed DeepOnet model on solving nonlinear parametric partial differential equations, Neurocomputing 611 (2025) 128675. https://doi.org/10.1016/j.neucom.2024.128675

    work page doi:10.1016/j.neucom.2024.128675 2025

  31. [31]

    Eshaghi, N

    M.S. Eshaghi, N. Valizadeh, C. Anitescu, Y. Wang, X. Zhuang, T. Rabczuk, Multi -Head Neural Operator for Modelling Interfacial Dynamics, (2025). https://doi.org/10.48550/ARXIV.2507.17763

    work page doi:10.48550/arxiv.2507.17763 2025

  32. [32]

    L. Lu, P. Jin, G. Pang, Z. Zhang, G.E. Karniadakis, Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators, Nat . Mach. Intell. 3 (2021) 218–229. https://doi.org/10.1038/s42256-021-00302-5

    work page doi:10.1038/s42256-021-00302-5 2021

  33. [33]

    S. Wang, H. Wang, P. Perdikaris, Learning the solution operator of parametric partial differential equations with physics -informed DeepONets, Sci. Adv. 7 (2021) eabi8605. https://doi.org/10.1126/sciadv.abi8605

    work page doi:10.1126/sciadv.abi8605 2021

  34. [34]

    Koric, D.W

    S. Koric, D.W. Abueidda, Data-driven and physics-informed deep learning operators for solution of heat conduction equation with parametric heat source, Int. J. Heat Mass Transf. 203 (2023) 123809. https://doi.org/10.1016/j.ijheatmasstransfer.2022.123809

    work page doi:10.1016/j.ijheatmasstransfer.2022.123809 2023

  35. [35]

    A. Jiao, Q. Yan, J. Harlim, L. Lu, Solving forward and inverse PDE problems on unknown manifolds via physics -informed neural operators, (2024). https://doi.org/10.48550/ARXIV.2407.05477

    work page doi:10.48550/arxiv.2407.05477 2024

  36. [36]

    Belmonte, J.-Y

    R. Belmonte, J.-Y. Dieulot, M. Galanti, M. Van Sint Annaland, Physics -informed artificial intelligence with splines for modeling advection –diffusion–reaction under dynamic boundaries, Digit. Eng. 9 (2026) 100083. https://doi.org/10.1016/j.dte.2025.100083

    work page doi:10.1016/j.dte.2025.100083 2026

  37. [37]

    Adam: A Method for Stochastic Optimization

    D.P. Kingma, J. Ba, Adam: A Method for Stochastic Optimization , (2014). arXiv:1412.6980. https://doi.org/10.48550/ARXIV.1412.6980

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1412.6980 2014

  38. [38]

    D.C. Liu, J. Nocedal, On the limited memory BFGS method for large scale optimization, Math. Program. 45 (1989) 503–528. https://doi.org/10.1007/BF01589116

    work page doi:10.1007/bf01589116 1989

  39. [39]

    A. Bora, W. Dai, J.P. Wilson, J.C. Boyt, Neural network method for solving parabolic two- temperature microscale heat conduction in double -layered thin films exposed to ultrashort-pulsed lasers, Int . J. Heat Mass Transf . 178 (2021) 121616. https://doi.org/10.1016/j.ijheatmasstransfer.2021.121616

    work page doi:10.1016/j.ijheatmasstransfer.2021.121616 2021

  40. [40]

    Omrani, M

    M. Omrani, M. Nematollahi, M. Ghassemi, Modeling interacting convection regimes with PINNs: Improving vanilla architecture performance with minimal supervision, Int . Commun. Heat Mass Transf . 172 (2026) 110763. https://doi.org/10.1016/j.icheatmasstransfer.2026.110763

    work page doi:10.1016/j.icheatmasstransfer.2026.110763 2026

  41. [41]

    Okuta, Y

    R. Okuta, Y. Unno, D. Nishino, H. Daisuke, L. Shohei, C. Loomis, CuPy: A NumPy - Compatible Library for NVIDIA GPU Calculations . I n: Proceedings of Workshop on 25 Machine Learning Systems (LearningSys), 2017. http://learningsys.org/nips17/assets/papers/paper_16.pdf

    2017

  42. [42]

    Rasmussen, C.K.I

    C.E. Rasmussen, C.K.I. Williams, Gaussian Processes for Machine Learning, The MIT Press, 2005. https://doi.org/10.7551/mitpress/3206.001.0001

    work page doi:10.7551/mitpress/3206.001.0001 2005

  43. [43]

    TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems

    M. Abadi, A. Agarwal, P. Barham, E. Brevdo, Z. Chen, C. Citro, G.S. Corrado, A. Davis, J. Dean, M. Devin, S. Ghemawat, I. Goodfellow, A. Harp, G. Irving, M. Isard, Y. Jia, R. Jozefowicz, L. Kaiser, M. Kudlur, J. Levenberg, D. Mane, R. Monga, S. Moore, D. Murray, C. Olah, M. Schuster, J. Shlens, B. Steiner, I. Sutskever, K. Talwar, P. Tucker, V. Vanhoucke,...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1603.04467 2016

  44. [44]

    Chollet et al., Keras

    F. Chollet et al., Keras. Available at https://github.com/keras-team/keras, accessed June 2026

    2026

  45. [45]

    Gaussian Error Linear Units (GELUs)

    D. Hendrycks, K. Gimpel, Gaussian Error Linear Units (GELUs), (2016). https://doi.org/10.48550/ARXIV.1606.08415

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1606.08415 2016