Pith. sign in

REVIEW 2 major objections 7 minor 54 references

Lagrangian Dynamic Mode Decomposition for Construction of Reduced-Order Models of Advection-Dominated Phenomena

T0 review · 2 major / 7 minor · reviewed 2026-08-14 · deepseek-v4-flash

Pith's one-line read This paper claims that running DMD on the Lagrangian observable $(x,u)$, where $x$ tracks moving characteristic lines, builds reduced-order models that stay accurate beyond the training window for shock-free advection-dominated problems.

desk verdict A genuinely useful incremental extension of Lagrangian POD to DMD, with clean shock-free tests—but the crucial Eulerian-only-data route for getting characteristic trajectories is never demonstrated, so its central promise outruns its evidence. read the letter →

arxiv 1908.03688 v1 pith:FVKBU46B submitted 2019-08-10 math.NA cs.NAphysics.comp-ph

classification math.NAcs.NAphysics.comp-ph MSC 65M99
keywords dynamicmodedecompositionKoopmanoperatorLagrangianframeworkreduced-ordermodeladvection-dominatedproblemsproperorthogonalcharacteristiclinesextrapolation
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

Conventional dynamic mode decomposition (DMD) and proper orthogonal decomposition (POD) fail for advection-dominated problems because their SVD bases are extracted from the region a traveling wave has already occupied, so the basis cannot represent the wave once it moves on. This paper claims that the fix is to change coordinates: run DMD on the Lagrangian observable $y=(x,u)$, where $x$ tracks the positions of characteristic lines and $u$ is the solution value on those moving lines. In four shock-free numerical tests, linear advection, linear advection-diffusion, and inviscid and viscous Burgers, the Lagrangian DMD captures the solution beyond the 250-snapshot training window ($t>0.25$), where its Eulerian counterparts produce oscillatory and unphysical profiles. The proposed method is iteration-free, so it builds and evaluates the ROM more cheaply than Lagrangian POD while matching or beating its accuracy.

What carries the argument

The load-bearing object is the observable vector $y^n=(x^n,u^n)$ built from the semi-Lagrangian formulation of the advection-diffusion equation, which tracks characteristic lines through the ordinary differential equation $dX/dt=f(u(X(t),t))$. The DMD algorithm approximates the Koopman operator, the infinite-dimensional linear operator that advances observable functions of the state, restricted to the invariant subspace spanned by these observables under Assumption 3.1. The characteristic-line positions are what do the work: they translate the moving wave into a frame in which the low-rank SVD basis does not have to chase the signal, which is exactly the failure mode diagnosed for Eulerian DMD in Section 2.3.

What would settle it

Run the same Lagrangian-DMD pipeline on a problem in which characteristic lines cross, such as inviscid Burgers evolved past the shock-formation time, and record the global truncation error $E^n$; the central claim would be falsified if the error stops following the Theorem 4.1 bound and grows without control or produces unphysical oscillations, since the paper's own conclusion states that all its tests are shock-free and that the Lagrangian grid may entangle.

Watch

Extended reading notes

Core claim

The paper's central discovery is that the translational difficulty of advection-dominated flows is a coordinate problem, not a fundamental limit of DMD or POD. By choosing the observable $y^n=(x^n,u^n)$, the vector of characteristic-line positions together with solution values on those lines, the authors obtain a finite-dimensional Koopman-invariant subspace in which the moving wave is stationary relative to the basis. The DMD algorithm (Algorithm 3.1) then approximates the Koopman operator restricted to this subspace, and future states are predicted analytically by $y^n=\Phi \Lambda^n b$. In the tested shock-free regimes the resulting ROM remains accurate in the extrapolating mode, i.e., for $t>0.25$, and the error bound of Theorem 4.1 gives a computable estimate of how long that accuracy lasts.

Load-bearing premise

The load-bearing premise is that a small set of fixed modes can exactly represent the coupled motion of the moving grid positions and the solution values on them, and that the training data reliably supply those moving grid positions; the tests obtain them from a Lagrangian solver, so the method's success depends on that supply.

Editorial extensions

If this is right

  • For shock-free advection-dominated problems, a single low-rank basis constructed on $(x,u)$ extrapolates accurately far beyond the training window $t\le 0.25$, where Eulerian DMD and POD fail.
  • Because future states come from one direct evaluation of the prediction formula, Lagrangian DMD requires no iteration in the low-dimensional space, making it the cheapest ROM among those compared while matching Lagrangian POD's accuracy in most tests.
  • The error bound in Theorem 4.1 provides an a priori estimate of the observable's error, allowing one to design a hybrid scheme that runs the high-fidelity model for a short time and then switches to the ROM for long-time prediction.
  • The level-set experiment in the appendix shows that the same Lagrangian-observable idea can be applied to conservation laws by reformulating them as two-dimensional linear transport, capturing the inviscid Burgers solution at rank $r=3$.

Reading between the lines

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

  • If characteristic positions can be recovered from Eulerian-only data by tracking wave crests, level sets, or feature velocities, the same ROM construction should work without a dedicated Lagrangian solver; the paper outlines this route in Section 3.1 but does not test it, so this remains an inference about an untested extension.
  • SVD-based ROMs for moving fronts have usually been repaired with local bases, domain decomposition, or multiresolution filtering; the Lagrangian-observable choice suggests a single global basis can suffice for a wider class of problems, provided the characteristic grid does not entangle.
  • A natural stress test beyond the four textbook cases is to add small noise or slightly random initial conditions and watch whether the Koopman-invariant-subspace assumption degrades gracefully; the paper does not report such a test, so the practical robustness of the extrapolating mode is still open.
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 / 7 minor

Summary. The manuscript proposes a Lagrangian reformulation of dynamic mode decomposition for reduced-order modeling of advection-dominated PDEs. It first demonstrates with a linear advection example that standard Eulerian POD and DMD generate global SVD bases that cannot represent a translating pulse, leading to inaccurate extrapolation. It then introduces Algorithm 3.1, which applies the standard DMD procedure to the augmented observable y^n = [x^n; u^n], where x^n denotes characteristic-line positions, and reconstructs u from the predicted observable. Numerical experiments on linear advection, linear advection-diffusion, inviscid and viscous Burgers equations, and a level-set formulation are presented; all reported tests are shock-free. In these tests, Lagrangian DMD and Lagrangian POD capture the solution beyond the training window t > 0.25, with DMD being cheaper because it is iteration-free. An error bound from a previous paper by the same authors is quoted and plotted.

Significance. Should the method hold, it offers a simple physics-aware extension of DMD that is equation-free and iteration-free and that directly addresses the translation failure of Eulerian SVD methods. The choice of [x; u] as observables is physically motivated and is not tuned to the test outputs. The numerical comparisons against conventional Eulerian DMD and POD are clean and convincing for the shock-free cases. The main caveat is that the method's success depends on the availability of accurate characteristic trajectories; this dependence is acknowledged only indirectly and is not validated in the standard Eulerian-snapshot data regime. Given the explicit shock-free limitation and the unvalidated data pipeline, the contribution is promising but not yet established at the claimed scope.

major comments (2)
  1. [Section 4 / Table 1] The numerical validation supplies the characteristic positions x^n from a separate Lagrangian high-fidelity solve, as the listing of 'Lagrangian HFM computational time' in Table 1 indicates, rather than deriving them from the Eulerian reference snapshots produced by (2.3). For an end user with Eulerian-only data, the route described in Section 3.1 is stated only for Lagrangian POD (following [30, Sec. 3.3]) and is neither formulated nor tested for Lagrangian DMD. Because Algorithm 3.1 requires x^n as an observable, this is a load-bearing gap: if characteristic trajectories cannot be recovered accurately from the available data, the method's advantage disappears. The authors should either demonstrate trajectory recovery from Eulerian snapshots and use it in the numerical tests, or explicitly restrict the paper's claim to settings in which Lagrangian trajectory data are available.
  2. [Theorem 4.1] The error bound used in every numerical example is quoted from [20], but epsilon_m is not defined in this manuscript and the theorem is stated without proof or a precise pointer to the statement in [20]. Since the figures plot this bound as an estimate of the observable error, a reader cannot verify whether the plotted curve is actually the right-hand side of (4.2). The authors should either include the definition of epsilon_m and the main steps of the proof, or remove the bound from the figures and state the error analysis as a reference.
minor comments (7)
  1. [Algorithm 3.1, Step 2] The matrix X' should be Y2; as written, the low-rank approximation K-tilde is not defined in terms of the observable data matrices introduced in Step 0.
  2. [Eqs. (2.17) and (3.9)] The prediction formulas use Lambda^{n+1} with b = Phi^{-1} y^1 (or u^0), but the standard DMD convention with snapshots indexed from 1 gives y^n = Phi Lambda^{n-1} b. Please clarify the indexing so the extrapolation formula is unambiguous.
  3. [Figure 7 caption] The caption says 'linear advection equation,' but Section 4.2 concerns the linear advection-diffusion equation; please correct.
  4. [Abstract and Introduction] The conclusion explicitly states that all tests are shock-free, but the abstract and introduction do not carry this qualification; readers may overinterpret the scope. Please state the shock-free limitation prominently in the abstract.
  5. [Sections 3.1 and 3.2] The phrases 'optimal Lagrangian basis' and 'optimal choice of observable functions' are not backed by an optimality criterion; 'well-motivated' or 'physics-informed' would be more accurate.
  6. [Appendix A] The level-set DMD experiment would benefit from a precise statement of the observable matrix, the number of snapshots, the SVD rank, and the reconstruction procedure to be reproducible.
  7. [Section 1] There are several typos, e.g., 'explaination' and 'physic-aware'; a careful proofreading pass is needed.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: Lagrangian DMD is a standard Koopman/DMD fit on an augmented observable with extrapolative tests; the sole self-citation (error bound from companion paper) is not load-bearing.

full rationale

The claimed prediction is extrapolation beyond the training window (m=250 snapshots, t<=0.25; error plots extend to t=1), so it is not a fitted input renamed as a prediction. The observable y=[x;u] (Eq. 3.11) is chosen from the physics of characteristics before fitting and is not optimized against the test outputs. The DMD algorithm (Algorithm 3.1) is the standard linear least-squares fit on snapshots; no equation in the paper defines the observable in terms of the predicted quantity or fits a parameter to the validation data. The paper's acknowledged reliance on Lagrangian-solver-generated trajectories, and its untested Eulerian-only-data route (Section 3.1), is a limitation of data availability, not a circular reduction. Theorem 4.1 is a bound from the authors' companion paper [20]; it is used only to plot an error estimate and is not needed to build the ROM, so it is a minor self-citation but not load-bearing. No significant circularity.

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

The method is built on standard DMD plus a physically motivated change of observable space. The main extra assumptions are the Koopman-invariant-subspace condition and the availability and validity of Lagrangian trajectories. No new physical entities are introduced.

free parameters (2)
  • SVD rank truncation r (energy threshold epsilon) = r = 3, 10, 3, 14 for Tests 1-4; epsilon = 1e-8 in Section 4, 1e-4 in Section 2
    The rank is selected from the snapshot energy criterion (2.7) and directly controls ROM dimension and accuracy; it is a data-dependent modeling choice, not a physical constant.
  • Number of training snapshots m = m = 250 for all Lagrangian tests
    Chosen by the authors (up to t=0.25); prediction quality depends on this window, and its size is not derived from the model.
assumptions (4)
  • domain assumption Assumption 3.1: the chosen observable vector y=(x,u) lies in a finite-dimensional Koopman-invariant subspace; otherwise the finite DMD representation is not valid.
    Invoked to justify Algorithm 3.1; not proven for the test PDEs.
  • domain assumption The semi-Lagrangian discretization (3.1)-(3.2) accurately tracks characteristic lines and interpolation between grids is stable.
    The ROM inherits the high-fidelity Lagrangian grid evolution; errors in trajectories would propagate into observables.
  • domain assumption All test problems are shock-free within the prediction horizon, so characteristic lines do not cross and the Lagrangian formulation remains valid.
    Authors state in Section 5 that the Lagrangian formulation becomes invalid after shock formation; the numerical evidence is restricted to this regime.
  • standard math Error bound in Theorem 4.1, taken from the authors' companion paper [20], is correct.
    Used to report prediction error bounds; the bound is cited, not rederived or verified in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Lagrangian Dynamic Mode Decomposition for Construction of Reduced-Order Models of Advection-Dominated Phenomena." pith.science (2026). https://pith.science/paper/FVKBU46B

@misc{pith2026190803688,
  author       = {Pith},
  title        = {Pith review of: Lagrangian Dynamic Mode Decomposition for Construction of Reduced-Order Models of Advection-Dominated Phenomena},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FVKBU46B}},
  note         = {Machine review of arXiv:1908.03688}
}
read the original abstract

Proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) are two complementary singular-value decomposition (SVD) techniques that are widely used to construct reduced-order models (ROMs) in a variety of fields of science and engineering. Despite their popularity, both DMD and POD struggle to formulate accurate ROMs for advection-dominated problems because of the nature of SVD-based methods. We investigate this shortcoming of conventional POD and DMD methods formulated within the Eulerian framework. Then we propose a Lagrangian-based DMD method to overcome this so-called translational issues. Our approach is consistent with the spirit of physics-aware DMD since it accounts for the evolution of characteristic lines. Several numerical tests are presented to demonstrate the accuracy and efficiency of the proposed Lagrangian DMD method.

Figures

Figures reproduced from arXiv: 1908.03688 by the authors.

Figure 1
Figure 1. Solution profiles u(x, ·), for several times t, in the diffusion-dominated regime. These profiles are computed with DMD (left) and POD (right), and compared with the reference solution. lies mostly in the subdomain [1, 2]. Specifictly, one can observe that the dominant DMD/POD modes have fluctuations only in the subdomain [0, 1] and stay flat 0 in the subdomain [1, 2] in [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Solution profiles u(x, ·), for several times t, in the advection-dominated regime. These profiles are computed with DMD (left column) and POD (right column) using the SVD rank of r = 20 (top row) and r = 30 (bottom row), and compared with the reference solution. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Three of the dominant DMD modes (left column) and PO [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Solutions of the linear advection equation, [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Errors E of the Lagrangian DMD- and POD-based ROMs for the linear advection equation: error in reconstructing the state variable u(x, t) (left) and its observables g(u) (right). The error bound for g(u) is derived in [20]. 4.2. Linear Advection-Diffusion Equation Next,…
Figure 6
Figure 6. Figure 6: Solutions of the linear advection-diffusion equat [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Errors E of the Lagrangian DMD- and POD-based ROMs for the linear advection equation: error in reconstructing the state variable u(x, t) (left) and its observables g(u) (right). The error bound for g(u) is derived in [20]. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]
Figure 8
Figure 8. Figure 8: Solutions of the inviscid Burgers equation, [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Errors E of the Lagrangian DMD- and POD-based ROMs for the inviscid Burgers equation: error in reconstructing the state variable u(x, t) (left) and its observables g(u) (right). The error bound for g(u) is derived in [20]. 0 1 2 3 4 5 6 7 0 0.5 1 1.5 2 0 1 2 3 4 5 6 7 …
Figure 10
Figure 10. Figure 10: Solutions of the viscous Burgers equation, [PITH_FULL_IMAGE:figures/full_fig_p017_10.png]
Figure 11
Figure 11. Figure 11: Errors E of the Lagrangian DMD- and POD-based ROMs for the viscous Burgers equation: error in reconstructing the state variable u(x, t) (left) and its observables g(u) (right). The error bound for g(u) is derived in [20]. For this nonlinear problem, Lagrangian DMD is …
Figure 12
Figure 12. Figure 12: Solutions of the inviscid Burgers equation, [PITH_FULL_IMAGE:figures/full_fig_p020_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

54 extracted references · 52 canonical work pages

  1. [20]

    Benner, S

    P. Benner, S. Gugercin, K. Willcox, A survey of projection-base d model reduction methods for parametric dynamical systems, SIAM Rev. 57 (4) (2015) 483–531

  2. [1]

    These equations become highly nonlinear when advection velocity and/or diffusion coefficient depend (s) on a system state, e.g., in the case of multiphase flows in porous media [1]

    Introduction Advection-diffusion equations are routinely used as a high-fidelity re presentation of mass conservation at a variety of spatiotemporal scales in a plethora of applications [1]. These equations become highly nonlinear when advection velocity and/or diffusion coefficient depend (s) on a system state, e.g., in the case of multiphase flows in porous m...

  3. [2]

    Conventional Eulerian Reduced-Order Models Consider a scalar state variable u(x, t ) : [ a, b ] × [0, T ] → R+, whose dynamics is described by a one-dimensional nonlinear advection-diffusion equation ∂u ∂t + f (u) ∂u ∂x = ∂ ∂x ( D(x, t, u ) ∂u ∂x ) , f (u) = ∂F (u) ∂u , (2.1) subject to the initial condition u(x, t = 0) = u0(x) (2.2) and appropriate (arbit...

  4. [3]

    (2.7).65

    Compute SVD of matrix X1 ≈ UΣV ∗ with U ∈ CN ×r, Σ ∈ Rr×r and V ∈ Cr×m, where r is the truncated rank chosen by a certain criteria, e.g. (2.7).65

  5. [4]

    Compute ˜K = U∗X2VΣ −1 as an r × r low-rank approximation of K

  6. [5]

    Compute eigendecomposition of ˜K: ˜KW = WΛ , Λ = (λ k). 5

  7. [6]

    Each column of Φ is a DMD mode corresponding to a particular eigenvalue in Λ

    Reconstruct eigendecomposition of K, whose eigenvalues and eigenvectors are Λ and Φ = UW, respectively. Each column of Φ is a DMD mode corresponding to a particular eigenvalue in Λ . With the ap- proximated eigenvalues and eigenvectors of K in hand, the projected future solution can be constructed analytically for all times in the future. In particular, a...

  8. [7]

    In th e semi-Lagrangian frame, (2.1) is written as        dX(t) dt = f (u(X(t), t )), du(x, t ) dt ⏐ ⏐ ⏐ ⏐ x=X(t) = [ ∂ ∂x ( D(x, t, u ) ∂u(x, t ) ∂x )] ⏐ ⏐ ⏐ ⏐ x=X(t)

    Lagrangian Reduced-Order Models Motivated by construction of a POD-based ROM for the advection- diffusion equation (2.1) within the Lagrangian framework [30], we propose a Lagrangian DMD. In th e semi-Lagrangian frame, (2.1) is written as        dX(t) dt = f (u(X(t), t )), du(x, t ) dt ⏐ ⏐ ⏐ ⏐ x=X(t) = [ ∂ ∂x ( D(x, t, u ) ∂u(x, t ) ∂x )] ⏐ ⏐ ⏐ ⏐ x=...

Show all 54 references
  1. [8]

    Create data matrices of observables Y1 and Y2 as Y1 =      | | | y1 y2 · · · ym−1 | | |      , Y2 =      | | | y2 y3 · · · ym | | |      , (3.8) where each column is given by yk = g(uk).135

  2. [9]

    Compute SVD of the matrix Y1 ≈ UΣV ∗ with U ∈ Cp×r, Σ ∈ Rr×r, V ∈ Cr×m, where r is the truncated rank chosen by certain criteria

  3. [10]

    Compute ˜K = U∗X′VΣ −1 as an r × r low-rank approximation for K

  4. [11]

    Compute eigendecomposition of ˜K: ˜KW = WΛ , Λ = (λ k)

  5. [12]

    Eigenvalues are Λ and eigenvectors are Φ = UW.140 11

    Reconstruct eigendecomposition of K. Eigenvalues are Λ and eigenvectors are Φ = UW.140 11

  6. [13]

    Future yn+1 DMD can be predicted by yn+1 DMD = ΦΛ n+1b, n > m (3.9) with b = Φ −1y1

  7. [14]

    (3.10) In data-driven modeling, judicious selection of the observables is cr ucial to the accuracy and efficiency of a Koopman operator’s approximation

    Transform from observables back to state-space: un DMD = g−1(yn DMD). (3.10) In data-driven modeling, judicious selection of the observables is cr ucial to the accuracy and efficiency of a Koopman operator’s approximation. Identification of general rules for choosing the observab...

  8. [15]

    In all tests, the reference solutions are145 computed in the Eulerian framework using (2.3)

    Numerical Experiments To ascertain the accuracy and robustness of the Lagrangian DMD , we use it to construct ROMs for a series of linear and nonlinear advection-dominated problems. In all tests, the reference solutions are145 computed in the Eulerian framework using (2.3). Th...

  9. [16]

    05 )2] (4.3a) and boundary conditions u(0, t ) = u(2, t ) = 0 . (4.3b) 0 0.5 1 1.5 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.5 1 1.5 2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Figure 4: Solutions of the linear advection equation, u(x, t), alternatively obtained with the num...

  10. [17]

    A new physic-aware DMD, based on the Lagrangian framework, is propos ed to overcome the shortcomings of reduced order models (ROMs) of advection-dominated nonlinear phenomena

    Conclusions200 In this paper, we investigate the issue of translational problem for conventional proper orthogonal decomposition (POD) and dynamic mode decomposition (DMD) in the Eu lerian framework. A new physic-aware DMD, based on the Lagrangian framework, is propos ed to ov...

  11. [18]

    D. Duke, J. Soria, D. Honnery, An error analysis of the dynamic mode decomposition, Exper. Fluids275 52 (2) (2012) 529–542

  12. [19]

    D. M. Tartakovsky, M. Dentz, Diffusion in porous media: Phenome na and mechanisms, Transp. Porous Media (2019). doi:10.1007/s11242-019-01262-6 .240

  13. [21]

    Quarteroni, G

    A. Quarteroni, G. Rozza, et al., Reduced order methods for mod eling and computational reduction, Vol. 9, Springer, 2014

  14. [22]

    Acharjee, N

    S. Acharjee, N. Zabaras, A concurrent model reduction appr oach on spatial and random domains245 for the solution of stochastic PDEs, Int. J. Num. Meth. Engrg. 66 (12) (2006) 1934–1954

  15. [23]

    Holmes, J

    P. Holmes, J. L. Lumley, G. Berkooz, C. W. Rowley, Turbulence, c oherent structures, dynamical systems and symmetry, Cambridge Univ. Press, 2012

  16. [24]

    J. L. Lumley, Stochastic tools in turbulence, Courier Corporatio n, 2007

  17. [25]

    Volkwein, Model reduction using proper orthogonal decompo sition, Lecture Notes, Insti-250 tute of Mathematics and Scientific Computing, University of Graz

    S. Volkwein, Model reduction using proper orthogonal decompo sition, Lecture Notes, Insti-250 tute of Mathematics and Scientific Computing, University of Graz. s ee http://www. uni-graz. at/imawww/volkwein/POD. pdf 1025 (2011)

  18. [26]

    Barrault, Y

    M. Barrault, Y. Maday, N. C. Nguyen, A. T. Patera, An empirical interpolation method: application to efficient reduced-basis discretization of partial differential equ ations, Comptes Rendus Mathema- tique 339 (9) (2004) 667–672.255

  19. [27]

    Chaturantabut, D

    S. Chaturantabut, D. C. Sorensen, Nonlinear model reduction via discrete empirical interpolation, SIAM J. Sci. Comput. 32 (5) (2010) 2737–2764

  20. [28]

    J. N. Kutz, S. L. Brunton, B. W. Brunton, J. L. Proctor, Dyn amic mode decomposition: data-driven modeling of complex systems, Vol. 149, SIAM, 2016

  21. [29]

    Mezi´ c, Analysis of fluid flows via spectral properties of the Koopman operator, Annu

    I. Mezi´ c, Analysis of fluid flows via spectral properties of the Koopman operator, Annu. Rev. Fluid260 Mech. 45 (2013) 357–378

  22. [30]

    B. O. Koopman, Hamiltonian systems and transformation in Hilber t space, Proc. Natl. Acad. Sci. U.S.A. 17 (5) (1931) 315

  23. [31]

    Mezi´ c, Spectral properties of dynamical systems, mode l reduction and decompositions, Nonlin

    I. Mezi´ c, Spectral properties of dynamical systems, mode l reduction and decompositions, Nonlin. Dyn. 41 (1-3) (2005) 309–325.265 21

  24. [32]

    bypass this issue by compensating computational costs in proj ecting back to the Eulerian grid. From the perspective of physic-aware data-driven modeling, we realize t hat significant information like shock220 formation time, shock location and shock speed is not interpreted w ...

  25. [33]

    Mezi´ c, A

    I. Mezi´ c, A. Banaszuk, Comparison of systems with complex b ehavior, Physica D 197 (1-2) (2004) 101–133

  26. [34]

    C. W. Rowley, I. Mezi´ c, S. Bagheri, P. Schlatter, D. S. Hennin gson, Spectral analysis of nonlinear flows, J. Fluid Mech. 641 (2009) 115–127

  27. [35]

    S. L. Brunton, J. L. Proctor, J. N. Kutz, Discovering govern ing equations from data by sparse270 identification of nonlinear dynamical systems, Proc. Natl. Acad. Sc i. U.S.A. 113 (15) (2016) 3932– 3937

  28. [36]

    Schmidt, H

    M. Schmidt, H. Lipson, Distilling free-form natural laws from exp erimental data, Science 324 (5923) (2009) 81–85

  29. [37]

    Korda, I

    M. Korda, I. Mezi´ c, On convergence of extended dynamic mo de decomposition to the Koopman operator, J. Nonlin. Sci. 28 (2) (2018) 687–710

  30. [38]

    H. Lu, D. M. Tartakovsky, Prediction accuracy analysis for dy namic mode decomposition, ArXiv (2019).280

  31. [39]

    Amsallem, M

    D. Amsallem, M. J. Zahr, C. Farhat, Nonlinear model order redu ction based on local reduced-order bases, Int. J. Num. Meth. Engrg. 92 (10) (2012) 891–916

  32. [40]

    D. J. Lucia, Reduced order modeling for high speed flows with mov ing shocks, Tech. rep., Air Force Inst. of Tech., Wright-Patterson Air Force Base, OH (2001)

  33. [41]

    Carlberg, Adaptive h-refinement for reduced-order mode ls, Int

    K. Carlberg, Adaptive h-refinement for reduced-order mode ls, Int. J. Num. Meth. Engrg. 102 (5)285 (2015) 1192–1210

  34. [42]

    J. N. Kutz, X. Fu, S. L. Brunton, Multiresolution dynamic mode d ecomposition, SIAM J. Appl. Dyn. Syst. 15 (2) (2016) 713–735

  35. [43]

    Gerbeau, D

    J.-F. Gerbeau, D. Lombardi, Approximated Lax pairs for the re duced order integration of nonlinear evolution equations, J. Comput. Phys. 265 (2014) 246–269.290

  36. [44]

    M. E. Kavousanakis, R. Erban, A. G. Boudouvis, C. W. Gear, I. G. Kevrekidis, Projective and coarse projective integration for problems with continuous symme tries, J. Comput. Phys. 225 (1) (2007) 382–407

  37. [45]

    Rap´ un, J

    M.-L. Rap´ un, J. M. Vega, Reduced order models based on local POD plus Galerkin projection, J. Comput. Phys. 229 (8) (2010) 3046–3063.295 22

  38. [46]

    C. W. Rowley, I. G. Kevrekidis, J. E. Marsden, K. Lust, Reduct ion and reconstruction for self-similar dynamical systems, Nonlinearity 16 (4) (2003) 1257

  39. [47]

    C. W. Rowley, J. E. Marsden, Reconstruction equations and th e Karhunen–Lo` eve expansion for systems with symmetry, Physica D 142 (1-2) (2000) 1–19

  40. [48]

    Mojgani, M

    R. Mojgani, M. Balajewicz, Lagrangian basis method for dimensio nality reduction of convection300 dominated nonlinear flows, arXiv preprint arXiv:1701.04343 (2017)

  41. [49]

    S. L. Brunton, B. W. Brunton, J. L. Proctor, J. N. Kutz, Koo pman invariant subspaces and finite linear representations of nonlinear dynamical systems for contro l, PloS One 11 (2) (2016) e0150171

  42. [50]

    Morton, A

    J. Morton, A. Jameson, M. J. Kochenderfer, F. Witherden, D eep dynamical modeling and control of unsteady fluid flows, in: Advances in Neural Information Processin g Systems, 2018, pp. 9278–9288.305

  43. [51]

    Thiffeault, Advection-diffusion in Lagrangian coordinates, Phys

    J.-L. Thiffeault, Advection-diffusion in Lagrangian coordinates, Phys. Lett. A 309 (5-6) (2003) 415– 422

  44. [52]

    Shashkov, B

    M. Shashkov, B. Wendroff, A composite scheme for gas dynamic s in Lagrangian coordinates, J. Comput. Phys. 150 (2) (1999) 502–517

  45. [53]

    S. E. Hieber, P. Koumoutsakos, A Lagrangian particle level set method, Journal of Computational310 Physics 210 (1) (2005) 342–367

  46. [54]

    Y.-H. Tsai, Y. Giga, S. Osher, A level set approach for computin g discontinuous solutions of Hamilton-Jacobi equations, Math. Comput. 72 (241) (2003) 159– 181. 23

Pith tools

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