Pith. sign in

REVIEW 2 major objections 5 minor 54 references

Data-Driven Correction Reduced Order Models for the Quasi-Geostrophic Equations: A Numerical Investigation

T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A data-fitted linear correction stabilizes low-dimensional quasi-geostrophic ocean-circulation ROMs.

desk verdict A useful numerical application of a known closure idea to the quasi-geostrophic double-gyre, where the out-of-sample DDC-ROM results are convincing but the headline CDDC-ROM claim only has in-sample support. read the letter →

arxiv 1908.05297 v1 pith:PUQMHH7C submitted 2019-08-14 physics.flu-dyn cs.NAmath.NA

Changhong Mou , Honghu Liu , David R. Wells , Traian Iliescu This is my paper · ORCID
classification physics.flu-dyncs.NAmath.NA
keywords reducedordermodeldata-drivencorrectionquasi-geostrophicequationsproperorthogonaldecompositionROMclosurephysicalconstraintsoceancirculationGalerkin
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks whether a small, data-fitted correction can rescue low-dimensional reduced-order models of the quasi-geostrophic double-gyre ocean circulation, a flow whose active spatial scales are far too many for a ten-mode Galerkin truncation. It answers yes for this test case: at $r=10$ the standard Galerkin ROM (G-ROM) has a time-averaged streamfunction relative error of $2.06\times 10^2$, the data-driven correction ROM (DDC-ROM) cuts this to $3.25\times 10^{-1}$, and the physically constrained CDDC-ROM improves it to $9.58\times 10^{-2}$. The improvement persists when the correction is trained on a shorter interval, $[10,35]$ or $[10,45]$, and then integrated through $t=80$. The paper concludes that, at the low resolutions where the G-ROM blows up or locks onto the wrong statistical state, learning the missing unresolved-mode interactions from data is a practical route to stable and accurate ROMs.

What carries the argument

The load-bearing object is the correction term $\mathrm{Correction}\approx (\tilde A a)_i$, the difference between projecting the full-rank dynamics $f(\omega_R)$ and the resolved dynamics $f(\omega_r)$ onto the first $r$ modes. The paper determines $\tilde A$ by minimizing the least-squares residual in Eq. (22) over DNS training snapshots, using truncated SVD to control ill-conditioning and replacing $\omega_R$ with an $m$-mode approximation with $m=3r$ for efficiency. The constrained variant solves Eq. (25) with the dissipation constraint $a^\top \tilde A a\le 0$. What carries the argument is that the correction enters only the linear part of the ROM, $\dot a = b + (A+\tilde A)a + a^\top B a$, so it stabilizes the most violent instabilities without altering the resolved nonlinear interactions.

What would settle it

Train the DDC-ROM on $[10,35]$ for a quasi-geostrophic run at a higher Reynolds number where eddy feedback is stronger, then integrate through $t=80$ and compare the least-squares residual of Eq. (22) on the unseen interval $[50,80]$; if the residual stays as large as the correction itself while the streamfunction error returns to G-ROM levels, the linear ansatz is the limiting factor.

Watch

Extended reading notes

Core claim

The central claim is that the effect of the unresolved POD modes on the resolved modes can be captured, for low-dimensional quasi-geostrophic ROMs, by a linear data-driven term added to the Galerkin system. Starting from the exact identity $\dot a_i = (f(\omega_r),\phi_i) + [(f(\omega_R),\phi_i)-(f(\omega_r),\phi_i)]$, the paper closes the system with the ansatz $\mathrm{Correction}\approx (\tilde A a)_i$ and fits $\tilde A\in\mathbb{R}^{r\times r}$ to DNS snapshots through the least-squares problem in Eq. (22). Requiring $a^\top \tilde A a\le 0$ in the fit yields the CDDC-ROM, whose learned linear operator is dissipative. The paper reports that the CDDC-ROM is the most accurate ROM for $5\le r\le 20$, the DDC-ROM is the most accurate for $25\le r\le 50$, and the G-ROM is consistently worse across the whole range; at $r=10$ the CDDC-ROM error is more than three orders of magnitude below the G-ROM error.

Load-bearing premise

The construction assumes the correction is linear in the resolved coefficients ($\mathrm{Correction}\approx \tilde A a$); if the true unresolved-mode feedback is strongly nonlinear in those coefficients, the learned term cannot represent it and the reported gains may not generalize.

Editorial extensions

If this is right

  • At $r=10$ the DDC-ROM reduces the time-averaged streamfunction error by a factor of about 600 relative to the G-ROM; the CDDC-ROM reduces it by about 2000.
  • At $r=10$ and $r=15$ the corrected ROMs keep kinetic energy close to the DNS range, while the uncorrected G-ROM grows to unphysical values.
  • The CDDC-ROM is the best choice for heavily truncated models ($5\le r\le 20$), while the DDC-ROM is better for larger dimensions because the dissipation constraint tends to overdamp.
  • Training the correction on $[10,35]$ or $[10,45]$ and simulating through $t=80$ still gives large gains over the G-ROM, so the learned linear term is not simply memorizing the training window.
  • For all ROMs the errors plateau at large $r$ rather than vanishing, which the paper attributes to roughness in the higher vorticity basis functions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The fact that a purely linear closure produces such large gains suggests that, at these truncations, the dominant missing interaction acts like a linear damping or transport mechanism; quadratic or higher-order closures would matter most in regimes with stronger nonlinear eddy feedback.
  • Because the error plateau at larger $r$ is blamed on rough basis functions, smoothing or regularizing the vorticity POD modes may be a more direct fix than enlarging the ROM once the correction is in place.
  • The dissipation constraint helps exactly where the uncorrected ROM is most unstable; a relaxed or data-adaptive constraint might preserve the low-$r$ stabilization without the overdamping seen at $r\ge 25$.
  • A natural transfer test is to train the correction at one Rossby or Reynolds number and evaluate at nearby parameter values; if the linear operator must be retrained for every regime, the practical value of the method is limited to fixed-parameter settings.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The manuscript investigates the data-driven correction reduced order model (DDC-ROM) and its physically constrained variant (CDDC-ROM) for the quasi-geostrophic equations with double-gyre wind forcing. The DDC-ROM adds a data-fitted linear correction term \tilde{A} a to the standard Galerkin ROM, and the CDDC-ROM enforces a negative-semidefiniteness constraint on \tilde{A} in Eq. (25). The numerical study compares G-ROM, DDC-ROM, and CDDC-ROM at dimensions r = 5 to 50 using two metrics: kinetic energy time series (Section 3.3.1) and the relative L2 error of the time-averaged streamfunction defined in Eq. (26) (Section 3.3.2). It also reports predictive experiments (Section 3.3.3) in which the ROMs are trained on [10,45] or [10,35] and simulated through t=80. The central claim is that, for low-dimensional ROMs, DDC-ROM and CDDC-ROM both outperform G-ROM and that CDDC-ROM provides the best results.

Significance. The QGE test problem is substantially more challenging than the Burgers and cylinder-flow benchmarks used in prior DDC-ROM work, so demonstrating stabilization and error reduction on this problem is a useful contribution. The paper's strengths include the systematic sweep over r, the explicit constrained formulation in Eq. (25), and the inclusion of predictive experiments for the DDC-ROM in Table 2. However, the paper's strongest advertised conclusion, that CDDC-ROM provides the best results, rests entirely on in-sample Table 1, while the predictive Table 2 omits CDDC-ROM. In addition, the hyperparameters of the method are selected on the same DNS interval used for the main evaluation. These gaps are load-bearing for the abstract's central claim and should be addressed before the paper can be accepted.

major comments (2)
  1. [Section 3.3.2 (Table 1), Section 3.3.3 (Table 2), Eq. (26)] The central claim that CDDC-ROM provides the best results is supported only by in-sample evidence. In Table 1, the POD basis and the operator \tilde{A} in Eq. (22) are constructed from snapshots on [10,80], and the error metric (26) is evaluated on the same interval. The predictive experiments in Table 2, in which training on [10,45] or [10,35] is followed by simulation through t=80, do not report CDDC-ROM results at all. Thus the second half of the abstract's claim has not been tested outside the training interval, and the 'factor of about 2000' improvement quoted in Section 4 is also an in-sample number. Please add CDDC-ROM to the predictive experiments, or explicitly qualify the CDDC-ROM ranking as in-sample only.
  2. [Section 3.2 (ROM construction)] The truncated SVD tolerance is chosen 'with a tolerance that yields the most accurate results,' and m=3r is chosen because 'numerical experiments suggest' it balances accuracy and efficiency. These are free parameters selected on the same DNS interval used to evaluate Table 1, so part of the reported advantage may reflect tuning to the test interval. The paper should either fix these choices a priori, provide a sensitivity analysis of Tables 1 and 2 to these parameters, or use a nested validation procedure so that the reported errors are not selected post hoc.
minor comments (5)
  1. [Section 3.3.3, Table 2] In Case I at r=50, the DDC-ROM error (2.09e-01) is larger than the G-ROM error (1.67e-01), so the statement that 'the DDC-ROM is significantly more accurate than the G-ROM' should be qualified to the range r <= 45, or to 'small and moderate r values.'
  2. [Section 3.2] The statement that modeling the commutation error 'does not significantly change' the DDC-ROM and CDDC-ROM results is not supported by any table, figure, or quantitative comparison; a brief numerical summary would improve reproducibility.
  3. [Section 2.3, Eq. (21)] In Eq. (21), the notation mixes a generic function g(\omega_r) with a vector component (\tilde{A} a)_i; please clarify whether g denotes a vector-valued operator or whether the right-hand side should be interpreted componentwise.
  4. [Section 3.2] The replacement of (\omega_R,\psi_R) by (\omega_m,\psi_m) with m=3r is mentioned only briefly; since this is a modeling choice with a direct effect on the learned correction, a sentence explaining why m=3r is sufficient would be helpful.
  5. [Section 3.3.3] There is a typo in the text: 'Furrthermore' should be 'Furthermore.'

Circularity Check

2 steps flagged · score 4.0 of 10

In-sample fit supports the 'CDDC-ROM best' claim; DDC-ROM's out-of-sample gain is genuine.

  1. fitted input called prediction [Section 3.2 (SVD tolerance) with Section 3.3.2, Eq. (26) and Table 1]
    "To tackle this ill-conditioning issue, we use the truncated singular value decomposition (SVD) (see Step 6 of Algorithm 1 in [53]) with a tolerance that yields the most accurate results."

    The operator \tilde{A} in Eq. (22) is fitted using DNS snapshots on [10,80], and the SVD tolerance is explicitly selected to minimize the reported accuracy metric on that same interval. Table 1 then reports the relative error (26) time-averaged over [10,80]. Thus the Table 1 improvements are in-sample: both the least-squares fit and the hyperparameter choice consume the same data and the same error metric that are later offered as evidence of ROM performance. The nonlinear ROM simulation is not logically forced by the fit, but the evaluation is not out-of-sample, so the reported advantage is partly a fitted-input result rather than an independent prediction.

  2. fitted input called prediction [Abstract and Section 3.3.3, Table 2]
    "The numerical investigation shows that, for low-dimensional ROMs, both the DDC-ROM and CDDC-ROM perform better than the standard Galerkin ROM (G-ROM) and the CDDC-ROM provides the best results."

    The only experiment that includes CDDC-ROM is Table 1, which is in-sample as described above: the correction operator is trained on the same [10,80] interval on which the time-averaged streamfunction error is evaluated. The genuinely predictive experiment in Table 2 trains on [10,35] or [10,45] and tests through t=80, but reports only G-ROM and DDC-ROM, omitting CDDC-ROM. Therefore the second half of the central claim, 'CDDC-ROM provides the best results,' is never tested out-of-sample; its support reduces to the in-sample Table 1 numbers, including the factor-of-2000 improvement quoted in Section 4.

full rationale

The DDC-ROM construction is largely self-contained: Eqs. (5)-(24) define the correction term as the difference between the R-dimensional and r-dimensional projected dynamics, and Eq. (21) explicitly states a linear ansatz for that term, with the operator \tilde{A} obtained from a least-squares fit. No load-bearing uniqueness theorem, self-citation chain, or renaming of a known result is used, and the paper acknowledges the linear ansatz as a modeling choice rather than a derived closure. The independent, externally benchmarked content is Section 3.3.3, where DDC-ROM trained on [10,35] or [10,45] is simulated through t=80 and still outperforms G-ROM; this supports the first part of the central claim. The circularity concern is narrower but real: Table 1, the only source for the 'CDDC-ROM provides the best results' conclusion, evaluates the error on exactly the interval used for fitting \tilde{A} and for tuning the SVD tolerance, and the predictive Table 2 omits CDDC-ROM entirely. This is an in-sample-support issue rather than a definitional equivalence, so the score is moderate rather than extreme.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on four modeling assumptions and two tuned numerical choices. The linear closure ansatz and the m=3r truncation are choices made for this paper; the POD-basis and DNS accuracy assumptions are standard domain assumptions. No new physical entities are introduced. The free parameters (SVD tolerance, m) are selected on the test problems, which is the main reason the numerical headline could be optimistic.

free parameters (2)
  • truncated SVD tolerance = not reported (chosen to yield most accurate results)
    Used to solve the ill-conditioned least squares problem (22); the value is selected on the test problems, so it is a tuned hyperparameter.
  • m = 3r (number of modes in correction term) = 3r (for r=10, m=30)
    The full R-dimensional correction is replaced by (omega_m, psi_m); m=3r was chosen from numerical experiments as a trade-off, making it a hand-selected parameter.
assumptions (4)
  • ad hoc to paper The correction term can be modeled as a linear function of the resolved coefficients: Correction approximately equals \tilde{A} a (Eq. 21).
    No justification beyond matching the G-ROM structure; higher-order ansatzes are acknowledged as possible future work, so the central results depend on this linearity.
  • domain assumption The POD basis built from 701 snapshots in [10,80] captures the QGE attractor sufficiently for the reduced-order dynamics.
    This is the standard ROM assumption; the paper notes the basis is rough for high indices but does not verify convergence of results with respect to snapshot sampling.
  • domain assumption The DNS at 257x513 resolution with the stated time integrator is an accurate reference solution.
    The paper states the results are similar to fine-resolution results in [41,44] but does not provide a convergence study in this paper.
  • ad hoc to paper Replacing (omega_R, psi_R) by (omega_m, psi_m) with m=3r preserves the correction term.
    This reduces cost but assumes the unresolved modes beyond m contribute negligibly to the correction; no error estimate is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-Driven Correction Reduced Order Models for the Quasi-Geostrophic Equations: A Numerical Investigation." pith.science (2026). https://pith.science/paper/PUQMHH7C

@misc{pith2026190805297,
  author       = {Pith},
  title        = {Pith review of: Data-Driven Correction Reduced Order Models for the Quasi-Geostrophic Equations: A Numerical Investigation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PUQMHH7C}},
  note         = {Machine review of arXiv:1908.05297}
}
read the original abstract

This paper investigates the recently introduced data-driven correction reduced order model (DDC-ROM) in the numerical simulation of the quasi-geostrophic equations. The DDC-ROM uses available data to model the correction term that is generally used to represent the missing information in low-dimensional ROMs. Physical constraints are added to the DDC-ROM to create the constrained data-driven correction reduced order model (CDDC-ROM) in order to further improve its accuracy and stability. Finally, the DDC-ROM is tested on time intervals that are longer than the time interval over which it was trained. The numerical investigation shows that, for low-dimensional ROMs, both the DDC-ROM and CDDC-ROM perform better than the standard Galerkin ROM (G-ROM) and the CDDC-ROM provides the best results.

Figures

Figures reproduced from arXiv: 1908.05297 by the authors.

Figure 1
Figure 1. Basis functions for the streamfunction (first row) and vorticity (second row). The vorticity basis functions ϕi ’s are the POD modes computed based on the DNS snapshots for the vorticity, while each streamfunction basis function φi is related to ϕi via φi = −∆−1ϕi ; see Section 2.2. 3.3. Numerical Results. In this section, we present numerical results for the G-ROM, DDC￾ROM, and CDDC-ROM. As benchmark for our numeri… view at source ↗
Figure 2
Figure 2. Kinetic energy of DNS, G-ROM, DDC-ROM and CDDC-ROM with different r values. All the ROMs are initialized at t = 10 using the projected DNS data. to the DDC-ROM is beneficial in the highly truncated cases (i.e., for small r values), but the benefit brought by the physical constraints diminishes as r is further increased, and can even produce less accurate results than DDC-ROM due to overdamping, as pointed out in Sec… view at source ↗
Figure 3
Figure 3. Time-averaged streamfunction ψ over the interval [10, 80] for DNS, 10-dim G￾ROM, 10-dim DDC-ROM, and 10-dim CDDC-ROM. the G-ROM, especially for small r values. This is in line with the results in [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 52 canonical work pages

  1. [41]

    San and T

    O. San and T. Iliescu. A stabilized proper orthogonal decomposition reduced-order model for large scale quasi- geostrophic ocean circulation. Adv. Comput. Math. , pages 1289–1319, 2015. 13

    work page 2015

  2. [1]

    Aza¨ ıez, T

    M. Aza¨ ıez, T. C. Rebollo, and S. Rubino. Streamline derivative projection-based POD-ROM for convection- dominated flows. Part I: Numerical Analysis. arXiv preprint http: // arxiv. org/ abs/ 1711. 09780, 2017

    work page 2017

  3. [2]

    Baiges, R

    J. Baiges, R. Codina, and S. Idelsohn. Reduced-order subscales for POD models. Comput. Methods Appl. Mech. Engrg., 291:173–196, 2015

    work page 2015

  4. [3]

    Ballarin, E

    F. Ballarin, E. Faggiano, S. Ippolito, A. Manzoni, A. Quarteroni, G. Rozza, and R. Scrofani. Fast simulations of patient-specific haemodynamics of coronary artery bypass grafts based on a POD–Galerkin method and a vascular shape parametrization. J. Comput. Phys. , 315:609–628, 2016

    work page 2016

  5. [4]

    Bergmann, C

    M. Bergmann, C. H. Bruneau, and A. Iollo. Enablers for robust POD models. J. Comput. Phys., 228(2):516–538, 2009

    work page 2009

  6. [5]

    S. L. Brunton and J. N. Kutz. Data-driven science and engineering: Machine learning, dynamical systems, and control. Cambridge University Press, 2019

    work page 2019

  7. [6]

    M. D. Chekroun and D. Kondrashov. Data-adaptive harmonic spectra and multilayer Stuart-Landau models. Chaos, 27:093110, 2017

    work page 2017

  8. [7]

    M. D. Chekroun, H. Liu, and J. C. McWilliams. Variational approach to closure of nonlinear dynamical systems: Autonomous case. submitted, 84 pages, 2019

    work page 2019

Show all 54 references
  1. [8]

    M. D. Chekroun, H. Liu, and S. Wang. Stochastic Parameterizing Manifolds and Non-Markovian Reduced Equa- tions: Stochastic Manifolds for Nonlinear SPDEs II . Springer Briefs in Mathematics, Springer, New York, 2015

    2015

  2. [9]

    A. J. Chorin and F. Lu. Discrete approach to stochastic parametrization and dimension reduction in nonlinear dynamics. Proc. Natl. Acad. Sci. USA , 112:9804–9809, 2015

    2015

  3. [10]

    Couplet, C

    M. Couplet, C. Basdevant, and P. Sagaut. Calibrated reduced-order POD-Galerkin system for fluid flow mod- elling. J. Comput. Phys. , 207(1):192–220, 2005

    2005

  4. [11]

    D. T. Crommelin and A. J. Majda. Strategies for model reduction: comparing different optimal bases. Journal of the Atmospheric Sciences , 61:2206–2217, 2004

    2004

  5. [12]

    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. J. Comp. Phys. , 371:214–243, 2018. 12

    2018

  6. [13]

    Gal´ an del Sastre and R

    P. Gal´ an del Sastre and R. Bermejo. Error estimates of proper orthogonal decomposition eigenvectors and Galerkin projection for a general dynamical system arising in fluid models. Numer. Math., 110(1):49–81, 2008

    2008

  7. [14]

    Galletti, C

    B. Galletti, C. H. Bruneau, L. Zannetti, and A. Iollo. Low-order modelling of laminar flow regimes past a confined square cylinder. J. Fluid Mech. , 503:161–170, 2004

    2004

  8. [15]

    R. J. Greatbatch and B. T. Nadiga. Four-gyre circulation in a barotropic model with double-gyre wind forcing. J. Phys. Oceanogr., 30(6):1461–1471, 2000

    2000

  9. [16]

    Gunzburger, N

    M. Gunzburger, N. Jiang, and M. Schneier. An ensemble-proper orthogonal decomposition method for the nonstationary Navier-Stokes equations. SIAM J. Numer. Anal. , 55(1):286–304, 2017

    2017

  10. [17]

    M. D. Gunzburger. Finite element methods for viscous incompressible flows . Computer Science and Scientific Computing. Academic Press Inc., Boston, MA, 1989

    1989

  11. [18]

    Hasselmann

    K. Hasselmann. PIPs and POPs: The reduction of complex dynamical systems using principal interaction and oscillation patterns. Journal of Geophysical Research: Atmospheres , 93(D9):11015–11021, 1988

    1988

  12. [19]

    J. S. Hesthaven, G. Rozza, and B. Stamm. Certified Reduced Basis Methods for Parametrized Partial Differential Equations. Springer, 2015

    2015

  13. [20]

    Hijazi, G

    S. Hijazi, G. Stabile, A. Mola, and G. Rozza. Data-driven POD-Galerkin reduced order model for turbulent flows. arXiv preprint, http: // arxiv. org/ abs/ 1907. 09909, 2019

    1907

  14. [21]

    Holmes, J

    P. Holmes, J. L. Lumley, and G. Berkooz. Turbulence, Coherent Structures, Dynamical Systems and Symmetry . Cambridge, 1996

    1996

  15. [22]

    B. Koc, C. Mou, M. Mohebujjaman, and T. Iliescu. Commutation error in reduced order modeling of fluid flows. arXiv preprint, http: // arxiv. org/ abs/ 1810. 00517, 2019

    2019

  16. [23]

    Kondrashov, M

    D. Kondrashov, M. D. Chekroun, and P. Berloff. Multiscale Stuart-Landau emulators: Application to wind-driven ocean gyres. Fluids, 3(1):21, 2018

    2018

  17. [24]

    Kondrashov, M

    D. Kondrashov, M. D. Chekroun, and M. Ghil. Data-driven non-Markovian closure models.Physica D, 297:33–55, 2015

    2015

  18. [25]

    Kravtsov, D

    S. Kravtsov, D. Kondrashov, and M. Ghil. Multilevel regression modeling of nonlinear processes: Derivation and applications to climatic variability. J. Clim., 18:4404–4424, 2005

    2005

  19. [26]

    Kwasniok

    F. Kwasniok. The reduction of complex dynamical systems using principal interaction patterns. Physica D , 92(1-2):28–60, 1996

    1996

  20. [27]

    Kwasniok

    F. Kwasniok. Optimal Galerkin approximations of partial differential equations using principal interaction pat- terns. Physical Rev. E, 55(5):5365, 1997

    1997

  21. [28]

    Loiseau and S

    J.-C. Loiseau and S. L. Brunton. Constrained sparse Galerkin regression. J. Fluid Mech. , 838:42–67, 2018

    2018

  22. [29]

    F. Lu, K. K. Lin, and A. J. Chorin. Data-based stochastic model reduction for the Kuramoto–Sivashinsky equation. Physica D, 340:46–57, 2017

    2017

  23. [30]

    A. J. Majda and J. Harlim. Physics constrained nonlinear regression models for time series. Nonlinearity, 26(1):201, 2012

    2012

  24. [31]

    Maulik, A

    R. Maulik, A. Mohan, B. Lusch, S. Madireddy, and P. Balaprakash. Time-series learning of latent-space dynamics for reduced-order model closure. arXiv preprint, http: // arxiv. org/ abs/ 1906. 07815, 2019

    1906

  25. [32]

    Mohebujjaman, L

    M. Mohebujjaman, L. G. Rebholz, and T. Iliescu. Physically-constrained data-driven correction for reduced order modeling of fluid flows. Int. J. Num. Meth. Fluids , 89(3):103–122, 2019

    2019

  26. [33]

    B. R. Noack, M. Morzynski, and G. Tadmor. Reduced-Order Modelling for Flow Control , volume 528. Springer Verlag, 2011

    2011

  27. [34]

    B. R. Noack, P. Papas, and P. A. Monkewitz. The need for a pressure-term representation in empirical Galerkin models of incompressible shear flows. J. Fluid Mech. , 523:339–365, 2005

    2005

  28. [35]

    Pagani, A

    S. Pagani, A. Manzoni, and K. Carlberg. Statistical closure modeling for reduced-order models of stationary systems by the ROMES method. arXiv preprint, http: // arxiv. org/ abs/ 1901. 02792, 2019

    1901

  29. [36]

    E. J. Parish, C. Wentland, and K. Duraisamy. The adjoint Petrov-Galerkin method for non-linear model reduc- tion. 2019

    2019

  30. [37]

    Peherstorfer and K

    B. Peherstorfer and K. Willcox. Data-driven operator inference for nonintrusive projection-based model reduc- tion. Comput. Methods Appl. Mech. Engrg. , 306:196–215, 2016

    2016

  31. [38]

    Perotto, A

    S. Perotto, A. Reali, P. Rusconi, and A. Veneziani. HIGAMod: A Hierarchical IsoGeometric Approach for MODel reduction in curved pipes. Comput. & Fluids , 142:21–29, 2017

    2017

  32. [39]

    Quarteroni, A

    A. Quarteroni, A. Manzoni, and F. Negri. Reduced Basis Methods for Partial Differential Equations: An Intro- duction, volume 92. Springer, 2015

    2015

  33. [40]

    T. C. Rebollo, E. D. Avila, M. G. M´ armol, F. Ballarin, and G. Rozza. On a certified Smagorinsky reduced basis turbulence model. SIAM J. Numer. Anal. , 55(6):3047–3067, 2017

    2017

  34. [42]

    San and R

    O. San and R. Maulik. Extreme learning machine for reduced order modeling of turbulent geophysical flows. Phys. Rev. E , 97(4):042322, 2018

    2018

  35. [43]

    San and R

    O. San and R. Maulik. Neural network closures for nonlinear model order reduction. Adv. Comput. Math. , 44(6):1717–1750, 2018

    2018

  36. [44]

    O. San, A. E. Staples, Z. Wang, and T. Iliescu. Approximate deconvolution large eddy simulation of a barotropic ocean circulation model. Ocean Modelling, 40:120–132, 2011

    2011

  37. [45]

    T. P. Sapsis and P. F. J. Lermusiaux. Dynamically orthogonal field equations for continuous stochastic dynamical systems. Physica D, 238(23-24):2347–2360, 2009

    2009

  38. [46]

    P. J. Schmid. Dynamic mode decomposition of numerical and experimental data. J. Fluid Mech., 656:5–28, 2010

    2010

  39. [47]

    F. M. Selten. An efficient description of the dynamics of barotropic flow. J. Atmos. Sci. , 52(7):915–936, 1995

    1995

  40. [48]

    Strazzullo, F

    M. Strazzullo, F. Ballarin, R. Mosetti, and G. Rozza. Model reduction for parametrized optimal control problems in environmental marine sciences and engineering. SIAM J. Sci. Comput. , 40(4):B1055–B1079, 2018

    2018

  41. [49]

    Taira, S

    K. Taira, S. L. Brunton, S. Dawson, C. W. Rowley, T. Colonius, B. J. McKeon, O. T. Schmidt, S. Gordeyev, V. Theofilis, and L. S. Ukeiley. Modal analysis of fluid flows: An overview. AIAA J., 55(12):4013–4041, 2017

    2017

  42. [50]

    J. H. Tu, C. W. Rowley, D. M. Luchtenburg, S. L. Brunton, and J. N. Kutz. On dynamic mode decomposition: Theory and applications. Journal of Computational Dynamics , 1:391–421, 2014

    2014

  43. [51]

    Z. Y. Wan, P. Vlachas, P. Koumoutsakos, and T. Sapsis. Data-assisted reduced-order modeling of extreme events in complex dynamical systems. PloS One, 13(5):e0197704, 2018

    2018

  44. [52]

    Z. Wang, I. Akhtar, J. Borggaard, and T. Iliescu. Proper orthogonal decomposition closure models for turbulent flows: A numerical comparison. Comput. Meth. Appl. Mech. Eng. , 237-240:10–26, 2012

    2012

  45. [53]

    X. Xie, M. Mohebujjaman, L. G. Rebholz, and T. Iliescu. Data-driven filtered reduced order modeling of fluid flows. SIAM J. Sci. Comput. , 40(3):B834–B857, 2018

    2018

  46. [54]

    X. Xie, G. Zhang, and C. G. Webster. Data driven reduced order modeling of fluid dynamics using linear multistep network. arXiv preprint http: // arxiv. org/ abs/ 1809. 07820, 2018. 14

    2018

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.