pith. sign in

arxiv: 2604.12678 · v1 · submitted 2026-04-14 · ⚛️ physics.flu-dyn

Bayesian-Enhanced Galerkin-Based Reduced Order Modelling for Unsteady Compressible Flows

Bijie Yang , Chengyuan Liu , Lu Tian , Yuping Qian , Mingyang Yang This is my paper

Pith reviewed 2026-05-10 14:25 UTC · model grok-4.3

classification ⚛️ physics.flu-dyn
keywords reduced-order modelingBayesian inferenceGalerkin projectionproper orthogonal decompositioncompressible flowsunsteady flowsfluid dynamicsinverse problem
0
0 comments X

The pith

Bayesian inference updates Galerkin-POD coefficients to enhance stability and predictive accuracy in unsteady compressible flows.

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

This paper establishes a new framework for reduced-order modeling of compressible flows by combining Galerkin projection with Bayesian inference. The method reformulates the correction of the projected ODE coefficients as a statistical inverse problem that accounts for uncertainties from mode truncation and data noise. It solves this analytically using a Gaussian likelihood and inverse-Gamma priors without requiring sampling. A reader would care because standard Galerkin-POD models often suffer from instability and poor prediction in complex unsteady flows such as those in compressors. The approach preserves physical interpretability while adding statistical rigor for more reliable simulations.

Core claim

The authors claim that by treating the adjustment of the Galerkin-projected ordinary differential equation system as a Bayesian inverse problem and solving it with an analytical, sampling-free method based on Gaussian likelihood and inverse-Gamma priors, the resulting models achieve substantially better robustness, stability, and fidelity in reproducing the dynamics of unsteady compressible flows, as shown in both a moderate-Reynolds-number oscillating dimpled flow and a high-Reynolds-number centrifugal compressor.

What carries the argument

The Bayesian update procedure for the coefficients of the reduced-order ODE system, derived from Gaussian likelihood and inverse-Gamma priors to incorporate truncation and noise uncertainties.

Load-bearing premise

The chosen Gaussian likelihood and inverse-Gamma priors sufficiently represent the uncertainties due to POD truncation and data noise to yield genuinely predictive models.

What would settle it

Direct numerical simulation results for a new unsteady compressible flow case where the Bayesian-updated model diverges from the true dynamics while the priors are applied in the specified manner.

Figures

Figures reproduced from arXiv: 2604.12678 by Bijie Yang, Chengyuan Liu, Lu Tian, Mingyang Yang, Yuping Qian.

Figure 1
Figure 1. Figure 1: FIG. 1: Sketch of the dimpled surface [PITH_FULL_IMAGE:figures/full_fig_p012_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Temporal variation of lift and drag coefficients. [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3: Flow field evolution at different times within one oscillation period based on [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4: POD eigenvalues [PITH_FULL_IMAGE:figures/full_fig_p015_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: POD modes based on Q-criteria and acoustic waves. [PITH_FULL_IMAGE:figures/full_fig_p016_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Temporal evolution of the POD mode amplitudes. [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: FIG. 7: Comparison of the temporal evolution of the POD mode amplitudes, [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Phase portraits (trajectories) of the POD modes, [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: FIG. 9: Posterior distribution of ROM coefficients. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Flow field reconstruction based on Bayesian-Galerkin-POD [PITH_FULL_IMAGE:figures/full_fig_p021_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Centrifugal compressor configuration. [PITH_FULL_IMAGE:figures/full_fig_p022_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12: Q criterion of tip-leakage vortex and impeller-diffuser interactions. [PITH_FULL_IMAGE:figures/full_fig_p023_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13: Course and more uniform mesh for results interpolation. [PITH_FULL_IMAGE:figures/full_fig_p024_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14: POD eigenvalues [PITH_FULL_IMAGE:figures/full_fig_p024_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15: Mean flow and first four POD modes. [PITH_FULL_IMAGE:figures/full_fig_p025_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16: Comparison of the temporal evolution of the POD mode amplitudes, [PITH_FULL_IMAGE:figures/full_fig_p026_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: FIG. 17: 3D unsteady flow field from LES [PITH_FULL_IMAGE:figures/full_fig_p028_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18: Reconstructed 3D unsteady flow field based on Bayesian-Galerkin-POD using [PITH_FULL_IMAGE:figures/full_fig_p028_18.png] view at source ↗
read the original abstract

This work proposes a statistically enhanced framework to address the instability and limited predictive capability of conventional Galerkin-Proper Orthogonal Decomposition (Galerkin-POD) models. The method reformulates the correction of the Galerkin-projected ODE system as a statistical inverse problem, in which the coefficients are inferred through Bayesian inference. By accounting for model uncertainty arising from POD mode truncation and data uncertainty introduced by data noise and numerical postprocessing, the framework systematically updates the ODE system coefficients using an analytical, sampling-free solution based on Gaussian likelihood and inverse-Gamma priors. The approach is first validated using a self-sustained oscillating flow over a dimpled surface at a moderate Reynolds number (Re=3000), demonstrating stable and accurate reproduction of the temporal dynamics and phase trajectories of coherent structures when compared with direct numerical simulation (DNS). It is then applied to a centrifugal compressor featuring strong tip-leakage vortex breakdown and impeller-diffuser interactions at Re=100000, where the model successfully captures dominant unsteady structures and frequency characteristics despite limited mode retention. Overall, the results show that Bayesian inference substantially enhances the robustness, stability, and predictive fidelity of Galerkin-POD models for compressible flow systems. The proposed methodology combines the physical interpretability of Galerkin projection with the statistical rigour of Bayesian inference, offering a general, computationally efficient, and uncertainty-aware reduced-order modelling framework for complex fluid dynamic applications.

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 paper claims to propose a Bayesian-enhanced Galerkin-POD framework for reduced-order modeling of unsteady compressible flows. It reformulates the correction of Galerkin coefficients as a Bayesian inverse problem solved analytically using Gaussian likelihood and inverse-Gamma priors, avoiding sampling. Validation on two cases—a dimpled surface flow at Re=3000 and a centrifugal compressor at Re=100000—shows improved stability, robustness, and fidelity in reproducing DNS dynamics and spectra compared to standard Galerkin-POD.

Significance. This result, if it holds, is significant as it offers an efficient way to incorporate uncertainty from POD truncation and data noise into ROMs for complex flows, leading to more reliable predictions. The conjugacy-based analytical update is a key strength, enabling closed-form solutions without MCMC sampling. The method maintains the interpretability of physics-based projection while adding statistical rigor. The successful application to compressible flows with vortex breakdown and interactions highlights its potential for engineering applications like turbomachinery. The stress-test concern regarding circularity in data usage does not appear to undermine the findings, as the updated models show demonstrable improvements in stability and accuracy beyond the initial projection.

minor comments (3)
  1. [Abstract] Abstract: The validation descriptions would benefit from brief quantitative indicators of improvement (e.g., RMS errors or frequency spectrum deviations) to complement the qualitative statements of 'stable and accurate reproduction' and 'successfully captures'.
  2. [§3] §3: The specific hyperparameter values chosen for the inverse-Gamma priors should be explicitly reported in the experiments, as they directly affect the posterior mean/variance and are needed for reproducibility.
  3. [Validation sections] Validation sections: Clarify the data partitioning (training vs. test) used for the Bayesian coefficient update versus the DNS comparison, even if the full-manuscript results show no violation of assumptions.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive assessment of the manuscript and for recommending minor revision. The report accurately captures the core contribution of our Bayesian-enhanced Galerkin-POD framework, including the analytical conjugacy-based update and its application to compressible flows. We appreciate the recognition of its potential for engineering applications such as turbomachinery.

Circularity Check

0 steps flagged

No circularity: derivation is self-contained statistical update

full rationale

The central derivation in the paper uses conjugacy of Gaussian likelihood and inverse-Gamma priors to obtain closed-form posterior updates for the Galerkin coefficients; this is a standard, externally verifiable statistical result independent of the specific POD modes or flow data. Validation against DNS and application to the compressor case are presented as empirical demonstrations rather than tautological reconstructions, with no load-bearing self-citation, no fitted parameter renamed as prediction, and no ansatz smuggled via prior work. The framework remains falsifiable against external DNS benchmarks without reducing to its inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard Bayesian modeling assumptions chosen to enable an analytical solution; no free parameters or new entities are explicitly introduced beyond the priors.

axioms (2)
  • domain assumption Data and truncation uncertainties follow a Gaussian likelihood
    Invoked to enable the analytical Bayesian update of ODE coefficients
  • ad hoc to paper Inverse-Gamma priors on the coefficients
    Selected to permit a sampling-free closed-form posterior solution

pith-pipeline@v0.9.0 · 5560 in / 1336 out tokens · 38147 ms · 2026-05-10T14:25:42.431119+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

38 extracted references · 38 canonical work pages

  1. [1]

    Another option is to employ the H1 inner product rather than the standard L2 inner product, since the former introduces additional dissipation into the ROM [17]

    or Smagorinsky-type POD closure models [14, 15], as well as spectral vanishing viscosity applied through a convolution kernel acting on the higher modes [10, 16]. Another option is to employ the H1 inner product rather than the standard L2 inner product, since the former introduces additional dissipation into the ROM [17]. Beyond viscosity augmentation, s...

    work page

  2. [2]

    and a centrifugal compressor exhibiting tip-leakage vortex breakdown and blade–blade interactions (Re∼10 5). 11 FIG. 1: Sketch of the dimpled surface. A. Physical problem and CFD results A schematic of the dimpled surface is shown in Fig. 1, where an incoming boundary layer flows over a hemispherical cavity. Self-sustained oscillations are observed: the s...

    work page

  3. [3]

    Holmes, J

    P. Holmes, J. L. Lumley, G. Berkooz, and C. W. Rowley,Turbulence, Coherent structures, Symmetry and Dynamical Systems(Cambridge University Press, 2012)

    work page 2012

  4. [4]

    Aubry, P

    N. Aubry, P. Holmes, J. L. Lumley, and E. Stone, The dynamics of coherent structures in the wall region of a turbulent boundary layer, Journal of fluid Mechanics192, 115 (1988)

    work page 1988

  5. [5]

    Rempfer, Investigations of boundary layer transition via galerkin projections on empirical eigenfunctions, Physics of Fluids8, 175 (1996)

    D. Rempfer, Investigations of boundary layer transition via galerkin projections on empirical eigenfunctions, Physics of Fluids8, 175 (1996)

    work page 1996

  6. [6]

    Rempfer, Low-dimensional modeling and numerical simulation of transition in simple shear flows, Annual review of fluid mechanics35, 229 (2003)

    D. Rempfer, Low-dimensional modeling and numerical simulation of transition in simple shear flows, Annual review of fluid mechanics35, 229 (2003)

    work page 2003

  7. [7]

    Rajaee, S

    M. Rajaee, S. K. Karlsson, and L. Sirovich, Low-dimensional description of free-shear-flow coherent structures and their dynamical behaviour, Journal of Fluid Mechanics258, 1 (1994)

    work page 1994

  8. [8]

    Cazemier, R

    W. Cazemier, R. W. Verstappen, and A. E. Veldman, Proper orthogonal decomposition and low-dimensional models for driven cavity flows, Physics of fluids10, 1685 (1998)

    work page 1998

  9. [9]

    S. S. Ravindran, Proper orthogonal decomposition in optimal control of fluids, NASA Report 99(1999)

    work page 1999

  10. [10]

    Prabhu, S

    R. Prabhu, S. S. Collis, and Y. Chang, The influence of control on proper orthogonal decom- position of wall-bounded turbulent flows, Physics of fluids13, 520 (2001)

    work page 2001

  11. [11]

    Bergmann, L

    M. Bergmann, L. Cordier, and J.-P. Brancher, Optimal rotary control of the cylinder wake using proper orthogonal decomposition reduced-order model, Physics of fluids17(2005)

    work page 2005

  12. [12]

    Sirisup and G

    S. Sirisup and G. E. Karniadakis, A spectral viscosity method for correcting the long-term behavior of pod models, Journal of Computational Physics194, 92 (2004)

    work page 2004

  13. [13]

    L. Fick, Y. Maday, A. T. Patera, and T. Taddei, A stabilized pod model for turbulent flows over a range of reynolds numbers: Optimal parameter sampling and constrained projection, Journal of Computational Physics371, 214 (2018)

    work page 2018

  14. [14]

    Couplet, P

    M. Couplet, P. Sagaut, and C. Basdevant, Intermodal energy transfers in a proper orthog- onal decomposition–galerkin representation of a turbulent separated flow, Journal of Fluid Mechanics491, 275 (2003)

    work page 2003

  15. [15]

    Podvin and J

    B. Podvin and J. Lumley, A low-dimensional approach for the minimal flow unit, Journal of Fluid Mechanics362, 121 (1998). 30

    work page 1998

  16. [16]

    Ullmann and J

    S. Ullmann and J. Lang, A pod-galerkin reduced model with updated coefficients for smagorin- sky les, inV European conference on computational fluid dynamics, ECCOMAS CFD(2010) p. 2010

    work page 2010

  17. [17]

    Z. Wang, I. Akhtar, J. Borggaard, and T. Iliescu, Two-level discretizations of nonlinear closure models for proper orthogonal decomposition, Journal of Computational Physics230, 126 (2011)

    work page 2011

  18. [18]

    Tadmor, Convergence of spectral methods for nonlinear conservation laws, SIAM Journal on Numerical Analysis26, 30 (1989)

    E. Tadmor, Convergence of spectral methods for nonlinear conservation laws, SIAM Journal on Numerical Analysis26, 30 (1989)

    work page 1989

  19. [19]

    Iollo, S

    A. Iollo, S. Lanteri, and J.-A. D´ esid´ eri, Stability properties of pod–galerkin approximations for the compressible navier–stokes equations, Theoretical and Computational Fluid Dynamics 13, 377 (2000)

    work page 2000

  20. [20]

    Bergmann, C

    M. Bergmann, C. H. Bruneau, and A. Iollo, Enablers for robust pod models, Journal of Computational Physics228(2008)

    work page 2008

  21. [21]

    Couplet, C

    M. Couplet, C. Basdevant, and P. Sagaut, Calibrated reduced-order pod-galerkin system for fluid flow modelling, Journal of Computational Physics207, 192 (2005)

    work page 2005

  22. [22]

    Stabile and B

    G. Stabile and B. Rosic, Bayesian identification of a projection-based reduced order model for computational fluid dynamics, Computers & Fluids201, 104477 (2020)

    work page 2020

  23. [23]

    Duriez, S

    T. Duriez, S. L. Brunton, and B. R. Noack,Machine learning control-taming nonlinear dy- namics and turbulence, Vol. 116 (Springer, 2017)

    work page 2017

  24. [24]

    X. Xie, M. Mohebujjaman, L. G. Rebholz, and T. Iliescu, Data-driven filtered reduced order modeling of fluid flows, SIAM Journal on Scientific Computing40, B834 (2018)

    work page 2018

  25. [25]

    C. W. Rowley, T. Colonius, and R. M. Murray, Model reduction for compressible flows using pod and galerkin projection, Physica D: Nonlinear Phenomena189, 115 (2004)

    work page 2004

  26. [26]

    Gloerfelt, Compressible proper orthogonal decomposition/galerkin reduced-order model of self-sustained oscillations in a cavity, Physics of Fluids20(2008)

    X. Gloerfelt, Compressible proper orthogonal decomposition/galerkin reduced-order model of self-sustained oscillations in a cavity, Physics of Fluids20(2008)

    work page 2008

  27. [27]

    J. P. Kaipio and E. Somersalo,Statistical and computational inverse problems(Springer, 2005)

    work page 2005

  28. [28]

    D. D. Kosambi, Statistics in function space, inDD Kosambi: selected works in mathematics and statistics(Springer, 2017) pp. 115–123

    work page 2017

  29. [29]

    J. L. Lumey,Stochastic tools in turbulence(Elsevier, 2012)

    work page 2012

  30. [30]

    Karhunen, Zur spektraltheorie stochastischer prozesse, Ann

    K. Karhunen, Zur spektraltheorie stochastischer prozesse, Ann. Acad. Sci. Fennicae, AI34 (1946). 31

    work page 1946

  31. [31]

    E. N. Lorenz,Empirical orthogonal functions and statistical weather prediction(MIT Press, 1956)

    work page 1956

  32. [32]

    Obukhov, Statistical description of continuous fields, Transactions of the Geophysical In- ternational Academy Nauk USSR24, 3 (1954)

    A. Obukhov, Statistical description of continuous fields, Transactions of the Geophysical In- ternational Academy Nauk USSR24, 3 (1954)

    work page 1954

  33. [33]

    Sirovich, Turbulence and the dynamics of coherent structures

    L. Sirovich, Turbulence and the dynamics of coherent structures. i. coherent structures, Quar- terly of applied mathematics45, 561 (1987)

    work page 1987

  34. [34]

    C. R. Doering and J. D. Gibbon,Applied analysis of the Navier-Stokes equations, 12 (Cam- bridge university press, 1995)

    work page 1995

  35. [35]

    E. F. Stone,A study of low-dimensional models for the wall region of a turbulent boundary layer.(1990)

    work page 1990

  36. [36]

    Z. Wang, I. Akhtar, J. Borggaard, and T. Iliescu, Proper orthogonal decomposition closure models for turbulent flows: a numerical comparison, Computer Methods in Applied Mechanics and Engineering237, 10 (2012)

    work page 2012

  37. [37]

    Stabile, F

    G. Stabile, F. Ballarin, G. Zuccarino, and G. Rozza, A reduced order variational multiscale approach for turbulent flows, Advances in Computational Mathematics45, 2349 (2019)

    work page 2019

  38. [38]

    Iliescu and Z

    T. Iliescu and Z. Wang, Variational multiscale proper orthogonal decomposition: Navier-stokes equations, Numerical Methods for Partial Differential Equations30, 641 (2014). 32

    work page 2014