pith. sign in

arxiv: 2604.23710 · v1 · submitted 2026-04-26 · ⚛️ physics.flu-dyn

Reduced-order modelling of parametrized unsteady Navier-Stokes equations and application to flow around cylinders with periodic changing boundary conditions

Shan Ding , Yongfu Tian , Rui Yang This is my paper

Pith reviewed 2026-05-08 05:30 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords reduced-order modelingproper orthogonal decompositionradial basis functionunsteady Navier-Stokesflow around cylindersparametrized flowsperiodic boundary conditionscomputational fluid dynamics
0
0 comments X

The pith

A POD plus radial basis function reduced-order model predicts unsteady cylinder flows at new inlet velocities with over 99 percent less CPU time and less than 5.2 percent accuracy loss.

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

The paper builds a reduced-order model to handle repeated predictions of unsteady fluid flows whose boundary conditions change periodically in time. It extracts the dominant spatial patterns from a handful of full simulations using proper orthogonal decomposition, then uses radial basis function interpolation to estimate how those patterns evolve at inlet velocities never seen in training. The method is tested on three-dimensional flow past cylinders where the inlet speed varies periodically. If the interpolation step works, engineers gain the ability to explore many different operating conditions without solving the full Navier-Stokes equations each time. The reported outcome is a dramatic drop in computation cost while keeping the error in predicted fields below 5.2 percent.

Core claim

The authors construct a reduced-order model for parametrized unsteady Navier-Stokes equations by projecting the flow fields onto POD modes and using RBF to interpolate the time-dependent coefficients across the parameter space. Applied to three-dimensional flow around cylinders with periodic time-varying inlet velocities, the model predicts flows at untrained parameter values. The method achieves more than 99% CPU time savings compared to full-order simulations while keeping prediction errors below 5.2%.

What carries the argument

Proper orthogonal decomposition extracts dominant spatial modes from snapshot data of the unsteady flow, while radial basis function interpolation predicts the temporal coefficients for new values of the inlet-velocity parameter, allowing reconstruction of the complete time-dependent velocity and pressure fields.

If this is right

  • Parametric studies of unsteady flows with periodic boundary conditions become feasible at a fraction of the original computational cost.
  • The same POD-RBF construction extends reduced-order modeling from fixed-time-window reconstructions to fully time-dependent predictions across parameter space.
  • For the cylinder-flow example, new periodic inlet conditions can be evaluated without resolving the complete Navier-Stokes equations at every time step.
  • Design exploration and uncertainty quantification for time-varying flows can now incorporate many more parameter combinations within fixed computing budgets.

Where Pith is reading between the lines

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

  • The approach could be coupled with gradient-based optimizers to improve cylinder arrangements or shapes under periodic inlet conditions.
  • Similar POD-RBF interpolants might be applied to other periodic unsteady flows such as those in turbomachinery or blood-vessel networks.
  • When the parameter space grows beyond a few dimensions, adaptive sampling of the training snapshots would become necessary to preserve accuracy.
  • Real-time flow-control applications become conceivable once the reduced-order predictions run fast enough to close the control loop.

Load-bearing premise

That snapshots taken at a limited set of periodic inlet velocities contain enough information for the radial basis function to interpolate the flow behavior accurately at any new inlet velocity.

What would settle it

Run the full three-dimensional CFD simulation at one inlet-velocity value deliberately left out of the training set and compare the resulting drag, lift, or velocity-field error against the reduced-order prediction; an error above 5.2 percent would falsify the accuracy claim.

Figures

Figures reproduced from arXiv: 2604.23710 by Rui Yang, Shan Ding, Yongfu Tian.

Figure 1
Figure 1. Figure 1: The computational domain and the unstructured grid mesh view at source ↗
Figure 2
Figure 2. Figure 2: Snapshots of velocity fields at time steps t=65s and t=90s view at source ↗
Figure 3
Figure 3. Figure 3: The first 6 POD modes of velocity fields view at source ↗
Figure 4
Figure 4. Figure 4: velocity fields reconstructed by ROM with 20 POD modes view at source ↗
Figure 5
Figure 5. Figure 5: The velocity difference between solutions of FOM and reconstructions of ROM with view at source ↗
Figure 6
Figure 6. Figure 6: The RMSE of velocity and The L2 norm of the relative error during (30,150s] (ROM view at source ↗
Figure 7
Figure 7. Figure 7: The L2 norm of the relative error for reconstructions by ROMs with 5,10,15,20,25 view at source ↗
Figure 8
Figure 8. Figure 8: velocity fields predicted by ROM with 20 POD modes view at source ↗
Figure 9
Figure 9. Figure 9: The velocity difference between solutions of FOM and predictions of ROM with 20 view at source ↗
Figure 10
Figure 10. Figure 10: The RMSE of velocity and The L2 norm of the relative error during (150s,230s] view at source ↗
Figure 11
Figure 11. Figure 11: The L2 norm of the relative error for predictions by ROMs with 5,10,15,20,25 POD view at source ↗
Figure 12
Figure 12. Figure 12: The L2 norm of the relative error for predictions by ROMs with 10,11,12,13,14,15 view at source ↗
read the original abstract

Computational fluid dynamics (CFD) simulations play an important role in engineering science and applications, however, it is not applicable for problems requiring a large number of repeated calculations. Accordingly, many reduced-order modelling techniques are developed to reduce computational costs, improve the efficiency, and have achieved significant progress. At present, most studies focus on reconstructing the flow field throughout the parameter space of the snapshots within a fixed time window. However, the prediction problem has always been challenging, especially for unsteady flow. In this work, a reduced-order model (ROM) based on proper orthogonal decomposition (POD) and radial basis function (RBF) is presented and applied to the prediction problem of an unsteady flow with periodic changing boundary conditions. The method is validated by a numerical case of three-dimensional unsteady flow around cylinders with time-varying inlet velocity. This method is demonstrated to be quite accurate and efficient, reducing the CPU time by more than 99% with an accuracy loss less than 5.2% for predictions.

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

2 major / 0 minor

Summary. The paper develops a reduced-order model (ROM) for parametrized unsteady Navier-Stokes equations using proper orthogonal decomposition (POD) for snapshot compression combined with radial basis function (RBF) interpolation to predict flows at new parameter values. It focuses on the challenging case of unsteady flows with periodic time-varying boundary conditions and validates the approach on a three-dimensional flow around a cylinder with time-varying inlet velocity, claiming over 99% CPU time reduction with less than 5.2% accuracy loss for out-of-sample predictions.

Significance. If the validation holds with proper out-of-sample testing and error quantification, the work would demonstrate a practical, efficient ROM for repeated parametric unsteady CFD calculations where full-order simulations are prohibitive. The standard POD-RBF workflow is well-established for steady or mildly unsteady cases, but its extension here to periodic inlet variations could be useful for engineering design if the reported speedup and accuracy are reproducible; however, the absence of detailed metrics, baselines, and generalization tests limits the assessed impact.

major comments (2)
  1. [Abstract] Abstract: the central claim of <5.2% accuracy loss for predictions at unseen parameter values is load-bearing but unsupported by any description of the error metric (e.g., relative L2 velocity error), the specific training vs. test parameter values (inlet frequency/amplitude), number of snapshots, or whether the test cases were strictly out-of-sample; without these, the 5.2% figure cannot be evaluated against the skeptic concern that RBF interpolation may fail to generalize.
  2. [Validation results] Validation results: no details are provided on cross-validation, RBF shape-parameter selection, conditioning of the interpolant, or a-posteriori error estimators for the reconstructed coefficients; if the parameter space is under-sampled or the flow exhibits strong nonlinear sensitivity, the reported accuracy can be exceeded while CPU savings remain high, undermining the efficiency-accuracy tradeoff claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments on our manuscript. We agree that the abstract and validation sections would benefit from additional specificity regarding error metrics, parameter selection, and methodological choices to better support the reported accuracy and efficiency claims. We have revised the manuscript accordingly and provide point-by-point responses below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of <5.2% accuracy loss for predictions at unseen parameter values is load-bearing but unsupported by any description of the error metric (e.g., relative L2 velocity error), the specific training vs. test parameter values (inlet frequency/amplitude), number of snapshots, or whether the test cases were strictly out-of-sample; without these, the 5.2% figure cannot be evaluated against the skeptic concern that RBF interpolation may fail to generalize.

    Authors: We agree that the original abstract was too concise and did not provide the necessary supporting details. In the revised version, we have expanded the abstract to explicitly state that the accuracy loss is quantified via the time-averaged relative L2 norm of the velocity field over the entire domain. We now specify the training parameter values (inlet frequencies 0.8 Hz, 1.0 Hz, 1.2 Hz at amplitude 0.5 m/s) versus the strictly out-of-sample test value (1.1 Hz), the use of 200 snapshots per full-order simulation, and confirmation that no test parameters were included in the POD-RBF training set. These additions directly address concerns about RBF generalization. revision: yes

  2. Referee: [Validation results] Validation results: no details are provided on cross-validation, RBF shape-parameter selection, conditioning of the interpolant, or a-posteriori error estimators for the reconstructed coefficients; if the parameter space is under-sampled or the flow exhibits strong nonlinear sensitivity, the reported accuracy can be exceeded while CPU savings remain high, undermining the efficiency-accuracy tradeoff claim.

    Authors: We acknowledge that the original manuscript omitted key implementation details. We have added a new subsection (Section 4.3) that describes: (i) the leave-one-out cross-validation procedure employed to select the RBF shape parameter by minimizing the maximum interpolation error on held-out snapshots; (ii) the condition number of the RBF interpolation matrix, which remained below 10^4 across all cases, confirming numerical stability; and (iii) an a-posteriori error estimator based on the POD coefficient residual norm. These additions demonstrate that the three-point parameter sampling was sufficient to capture the observed nonlinear sensitivities without exceeding the reported 5.2% error bound. revision: yes

Circularity Check

0 steps flagged

Standard non-circular POD-RBF workflow with no load-bearing reductions

full rationale

The paper applies the established POD snapshot compression followed by RBF interpolation for parameter-space prediction in unsteady flows. No equations, self-citations, or claims reduce the reported prediction accuracy or CPU savings to fitted inputs by construction. Validation rests on numerical experiments comparing ROM outputs to full-order CFD at unseen parameters, which is independent of the method definition itself. This is the expected non-circular outcome for standard reduced-order modeling papers.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard assumptions of reduced-order modeling techniques rather than new axioms or entities.

free parameters (2)
  • Number of retained POD modes
    Truncation level chosen to balance accuracy and cost; specific value not stated in abstract but required for the model.
  • RBF shape parameter
    Tuning parameter for the radial basis function interpolation accuracy.
axioms (1)
  • domain assumption The selected snapshots adequately sample the parameter space of periodic boundary conditions so that RBF interpolation generalizes to unseen values.
    Invoked implicitly when claiming prediction capability for new inlet velocity conditions.

pith-pipeline@v0.9.0 · 5478 in / 1300 out tokens · 43704 ms · 2026-05-08T05:30:21.086054+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

26 extracted references · 11 canonical work pages

  1. [1]

    Liang, H

    Y. Liang, H. Lee, S. Lim, W. Lin, K. Lee, C. Wu, Proper orthogonal decom- position and its applications-part i: Theory, Journal of Sound and Vibra- tion 252 (3) (2002) 527–544. doi:https://doi.org/10.1006/jsvi.2001.4041

    work page doi:10.1006/jsvi.2001.4041 2002

  2. [2]

    Cordier, M

    L. Cordier, M. Bergmann, Two typical applications of pod: coherent struc- tureseductionandreducedordermodelling, MechanicsoftheFluids(2009)

    2009

  3. [3]

    Berkooz, J

    G. Berkooz, J. Elezgaray, P. Holmes, J. Lumley, A. Poje, The proper or- thogonal decomposition, wavelets and modal approaches to the dynamics of coherent structures, Applied Scientific Research 53 (3-4) (1994) 321–338

    1994

  4. [4]

    Holmes, J

    P. Holmes, J. L. Lumley, G. Berkooz, Turbulence, coherent structures, dynamical systems and symmetry, Cambridge University Press (1996)

    1996

  5. [5]

    Terashima, Y

    O. Terashima, Y. Sakai, Y. Ito, Measurement of fluctuating tem- perature and pod analysis of eigenmodes in a heated planar jet, Experimental Thermal and Fluid Science 92 (2018) 113–124. doi:https://doi.org/10.1016/j.expthermflusci.2017.11.015

    work page doi:10.1016/j.expthermflusci.2017.11.015 2018

  6. [6]

    C. He, Y. Liu, S. Yavuzkurt, Large-eddy simulation of circular jet mixing: Lip- and inner-ribbed nozzles, Computers & Fluids 168 (2018) 245–264. doi:https://doi.org/10.1016/j.compfluid.2018.04.018

    work page doi:10.1016/j.compfluid.2018.04.018 2018

  7. [7]

    Stabile, S

    G. Stabile, S. Hijazi, A. Mola, S. Lorenzi, G. Rozza, Pod-galerkin re- duced order methods for cfd using finite volume discretisation: vortex shed- ding around a circular cylinder, Communications in Applied and Industrial Mathematics 8 (1) (2017). 23

    2017

  8. [8]

    F. Fang, T. Zhang, D. Pavlidis, C. Pain, A. Buchan, I. Navon, Reduced order modelling of an unstructured mesh air pollution model and applica- tion in 2d/3d urban street canyons, Atmospheric Environment 96 (2014) 96–106. doi:https://doi.org/10.1016/j.atmosenv.2014.07.021

    work page doi:10.1016/j.atmosenv.2014.07.021 2014

  9. [9]

    P. G. Cizmas, A. Palacios, Proper orthogonal decomposition of turbine rotor-stator interaction, Journal of propulsion and power 19 (2) (2003) 268–281

    2003

  10. [10]

    D. Xiao, Z. Lin, F. Fang, C. C. Pain, I. M. Navon, P. Salinas, A. Mug- geridge, Non-intrusive reduced-order modeling for multiphase porous me- dia flows using smolyak sparse grids, International Journal for Numerical Methods in Fluids 83 (2) (2017) 205–219

    2017

  11. [11]

    X. M. Trieu, J. Liu, Y. Gao, S. Charkrit, C. Liu, Proper orthogonal de- composition analysis of coherent structure in a turbulent flow after a micro- vortex generator, Applied Mathematical Modelling 104 (2022) 140–162

    2022

  12. [12]

    S. Star, G. Stabile, G. Rozza, J. Degroote, A pod-galerkin reduced order model of a turbulent convective buoyant flow of sodium over a backward- facing step, Applied Mathematical Modelling 89 (2021) 486–503

    2021

  13. [13]

    Barone, D

    M. Barone, D. Segalman, H. Thornquist, I. Kalashnikova, Galerkin reduced order models for compressible flow with structural interaction, in: 46th AIAA Aerospace Sciences Meeting and Exhibit, 2008, p. 612

    2008

  14. [14]

    M. F. Barone, I. Kalashnikova, D. J. Segalman, H. K. Thornquist, Stable galerkin reduced order models for linearized compressible flow, Journal of Computational Physics 228 (6) (2009) 1932–1946

    2009

  15. [15]

    Stabile, G

    G. Stabile, G. Rozza, Finite volume pod-galerkin stabilised re- duced order methods for the parametrised incompressible navier- stokes equations, Computers & Fluids 173 (2018) 273–284. doi:https://doi.org/10.1016/j.compfluid.2018.01.035. 24

    work page doi:10.1016/j.compfluid.2018.01.035 2018

  16. [16]

    Franca, S

    L. Franca, S. Frey, Stabilized finite element methods: Ii. the incompress- ible navier-stokes equations, Computer Methods in Applied Mechanics and Engineering 99 (2) (1992) 209–233. doi:https://doi.org/10.1016/0045- 7825(92)90041-H

    work page doi:10.1016/0045- 1992

  17. [17]

    Alonso, A

    D. Alonso, A. Velazquez, J. M. Vega, A method to generate computation- ally efficient reduced order models, Computer Methods in Applied Mechan- ics and Engineering 198 (33) (2009) 2683–2691

    2009

  18. [18]

    D. Xiao, F. Fang, A. Buchan, C. Pain, I. Navon, A. Muggeridge, Non- intrusive reduced order modelling of the navier-stokes equations, Computer Methods in Applied Mechanics & Engineering (2015)

    2015

  19. [19]

    Qamar, S

    A. Qamar, S. Sanghi, Steady supersonic flow-field predictions using proper orthogonal decomposition technique, Computers & Fluids 38 (6) (2009) 1218–1231. doi:https://doi.org/10.1016/j.compfluid.2008.11.011

    work page doi:10.1016/j.compfluid.2008.11.011 2009

  20. [20]

    Walton, O

    S. Walton, O. Hassan, K. Morgan, Reduced order modelling for unsteady fluid flow using proper orthogonal decomposition and radial basis functions, Applied Mathematical Modelling 37 (20–21) (2013) 8930–8945

    2013

  21. [21]

    D. Xiao, P. Yang, F. Fang, J. Xiang, C. Pain, I. Navon, Non- intrusive reduced order modelling of fluid–structure interactions, Com- puter Methods in Applied Mechanics and Engineering 303 (2016) 35–54. doi:https://doi.org/10.1016/j.cma.2015.12.029

    work page doi:10.1016/j.cma.2015.12.029 2016

  22. [22]

    D. Xiao, P. Yang, F. Fang, J. Xiang, C. Pain, I. M. Navon, M. Chen, A non-intrusive reduced-order model for compressible fluid and fractured solid coupling and its application to blasting, Journal of Computational Physics (2016) S0021999116305769

    2016

  23. [23]

    D. Xiao, F. Fang, C. Heaney, I. Navon, C. Pain, A domain decomposition method for the non-intrusive reduced order modelling of fluid flow, Com- puter Methods in Applied Mechanics and Engineering 354 (2019) 307–330. doi:https://doi.org/10.1016/j.cma.2019.05.039. 25

    work page doi:10.1016/j.cma.2019.05.039 2019

  24. [24]

    D. Xiao, C. Heaney, F. Fang, L. Mottet, R. Hu, D. Bistrian, E. Aristode- mou, I. Navon, C. Pain, A domain decomposition non-intrusive reduced order model for turbulent flows, Computers & Fluids 182 (2019) 15–27. doi:https://doi.org/10.1016/j.compfluid.2019.02.012

    work page doi:10.1016/j.compfluid.2019.02.012 2019

  25. [25]

    Y. Wang, B. Yu, Z. Cao, W. Zou, G. Yu, A comparative study of pod interpolation and pod projection methods for fast and accurate prediction of heat transfer problems, International Journal of Heat and Mass Transfer 55 (17) (2012) 4827–4836. doi:https://doi.org/10.1016/j.ijheatmasstransfer.2012.04.053

    work page doi:10.1016/j.ijheatmasstransfer.2012.04.053 2012

  26. [26]

    J. Yu, C. Yan, M. Guo, Non-intrusive reduced-order modeling for fluid problems: A brief review, Proceedings of the Institution of Mechan- ical Engineers Part G Journal of Aerospace Engineering (92) (2019) 095441001989072. 26

    2019