Pith. sign in

REVIEW 2 major objections 4 minor 1 cited by

Learning physics-based reduced-order models for a single-injector combustion process

T0 review · 2 major / 4 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 The paper claims that a quadratic reduced-order model, learned from 10,000 CFD snapshots in transformed variables, predicts the single-injector combustor's pressure, temperature, velocity, and species fields accurately, runs five to six…

desk verdict Operator inference plus specific-volume/molar-concentration variables is a genuinely useful recipe for combustion ROMs, but the paper overstates the extrapolation evidence: only pressure traces are actually shown on the 200% held-out interval. read the letter →

arxiv 1908.03620 v4 pith:WXJV7H2E submitted 2019-08-09 physics.comp-ph cs.LGcs.SYeess.SYmath.DSstat.ML

Renee Swischuk , Boris Kramer , Cheng Huang , Karen Willcox This is my paper · ORCID
classification physics.comp-phcs.LGcs.SYeess.SYmath.DSstat.ML
keywords reduced-ordermodelingoperatorinferenceproperorthogonaldecompositioncombustiondynamicsquadraticstructurevariabletransformationnon-intrusivemodelreductionsingle-injectorcombustor
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

This paper establishes that a low-dimensional quadratic model can reproduce the unsteady dynamics of a single-injector combustor. The method trains on a set of transformed physical variables—pressure, velocities, specific volume, and species molar concentrations—in which much of the governing physics becomes quadratic in form. Operator inference learns the reduced operators from projected snapshot data, without touching the CFD solver, while proper orthogonal decomposition fixes the low-dimensional subspace. The learned 29-mode model matches the CFD pressure trace, temperature, velocity, and species fields, captures the limit-cycle amplitude, and remains predictive for an additional 2 ms of simulation beyond the 1 ms training data. A reliable model of this cost would make repeated combustor evaluations practical for design and uncertainty quantification.

What carries the argument

The central object is the quadratic reduced-order model $\dot{\hat{q}} = \hat{A}\hat{q} + \hat{F}\hat{q}^{(2)} + \hat{B}u + \hat{c}$ in POD coordinates, where $\hat{q}^{(2)}$ is the vector of unique quadratic monomials (the collapsed Kronecker product). The enabling mechanism is the variable transformation to $\mathbf{q}_L$: using specific volume and molar concentrations converts most of the combustion dynamics into quadratic form, which projection preserves exactly. Operator inference—a linear least-squares fit of the reduced operators to projected snapshot data—does the learning, aided by randomized SVD for the large snapshot matrix and L2/Tikhonov regularization to keep the learned operators stable.

What would settle it

Run the learned ROM with the forcing frequency changed from 5 kHz to 4 kHz, or the forcing amplitude changed, without retraining, and compare the predicted pressure phase and limit-cycle amplitude against a new high-fidelity CFD simulation; if the phase or amplitude error grows sharply, the claim that the quadratic form captures the dominant dynamics beyond the training interval is refuted.

Watch

Extended reading notes

Core claim

The central discovery is that the single-injector combustor's governing equations expose quadratic structure when written in the learning variables $\mathbf{q}_L = [p, v_x, v_y, \xi, c_1, c_2, c_3, c_4]$, where $\xi = 1/\rho$ is the specific volume and $c_l$ are the molar concentrations of CH$_4$, O$_2$, CO$_2$, and H$_2$O. The continuity, momentum, and species evolution equations become quadratic in these variables, and this polynomial form is preserved under Galerkin projection. Operator inference then identifies the ROM operators $\hat{A}, \hat{F}, \hat{B}, \hat{c}$ by solving a regularized least-squares problem from projected snapshot data, without access to the high-fidelity model's implementation. The learned ROM of dimension $r=29$ predicts the pressure trace at the monitor location with accurate phase and a good approximation of the limit-cycle amplitude over a 3 ms simulation (1 ms training plus 2 ms extrapolation), at a computational speedup of five to six orders of magnitude relative to the CFD solver.

Load-bearing premise

The load-bearing premise is that a quadratic model in the variables $[p, v_x, v_y, \xi, c_1, \ldots, c_4]$ adequately represents the true combustor dynamics over the prediction horizon, even though the governing pressure and species equations contain non-polynomial terms (temperature-dependent enthalpies, Arrhenius factors, and mixed density terms) that the quadratic form simply fits away.

Editorial extensions

If this is right

  • If the central claim holds, repeated combustion model evaluations for design, optimization, and uncertainty quantification become practical at seconds of compute rather than hundreds of CPU-hours per run.
  • The 29-dimensional ROM resolves the 5 kHz pressure forcing with correct phase, so it could be embedded in real-time control or active-instability-suppression loops.
  • The transformed-variable-plus-operator-inference recipe transfers to other reacting-flow or multi-physics systems whose equations can be expressed (or lifted) in polynomial form.
  • Demonstrating accuracy for 2 ms of pure prediction beyond a 1 ms training window suggests the learned operators encode the dominant nonlinear coupling, at least at the tested operating point and forcing condition.

Reading between the lines

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

  • The paper's own derivations show that the pressure and species equations contain non-polynomial terms (temperature-dependent enthalpies, Arrhenius rates, and mixed density terms) that the quadratic model simply fits away; a natural extension would test how the 2 ms predictive horizon degrades as the operating point moves away from the training conditions.
  • The same methodology could be tested on a parameterized family of simulations (varying inlet mass-flow rates, forcing frequency, or equivalence ratio) to see whether a single quadratic ROM, or a small library of them, interpolates between operating points; the paper only considers one fixed operating point with one forcing frequency.
  • Because the approach is non-intrusive, it could be applied to time-resolved experimental measurements (for example, from a laboratory combustor) if field data or suitable sensor projections are available, although the paper uses simulated data only.
  • The ROM's ability to hold the limit-cycle phase beyond training hints that the learned linear operator $\hat{A}$ may encode the acoustic eigenstructure of the combustor; extracting its eigenvalues and comparing with linear stability analysis of the full system is a testable check of physical consistency.
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 / 4 minor

Summary. This paper proposes a non-intrusive, operator-inference-based reduced-order modeling workflow for a two-dimensional single-injector combustor. The authors choose the learning variables q_L = [p, vx, vy, xi, c1,...,c4], transform GEMS snapshots accordingly, define a POD basis in these variables, and infer a quadratic ROM (Eq. (13)) by solving a regularized least-squares problem (Eq. (20)). The method is applied to 10,000 snapshots covering 1 ms of simulation; reduced orders r=24 and r=29 are used. Results compare ROM pressure traces with GEMS at one monitor location for 3 ms (1 ms training plus 2 ms prediction), field-level errors at the last training time step, integrated species, and CPU timings, reporting five to six orders of magnitude speedup. The paper includes appendices detailing the lifting of chemical source terms and explicitly deriving the pressure and species equations, which are not quadratic in q_L.

Significance. If the claims hold, this is a useful demonstration that operator inference in transformed variables can produce stable low-dimensional ROMs for a reacting flow where classical POD-Galerkin requires hundreds of modes. The paper's explicit appendices are a strength: the authors identify the exact polynomial structure available and, importantly, state where the quadratic form is an approximation. The availability of code links is also positive. The central limitation is that the evidence for genuine extrapolation beyond the training interval is narrow: the held-out quantitative comparison is limited to pressure at one monitor, while temperature, species, and velocity fields are validated only at the end of the training interval. Thus the significance of the claimed 200% predictive capability is not yet fully established.

major comments (2)
  1. [IV.C, Figs. 5-16 and Abstract] The abstract's sentence that the ROM 'accurately predicts temperature, pressure, velocity, species concentrations, and the limit-cycle amplitude ... and is shown to be predictive 200% past the training interval' is not supported by the beyond-training evidence shown in the paper. The only quantities compared against GEMS on the 20,000 time steps beyond the 1 ms training window are pressure time traces at a single monitor location (Figs. 5-6). Figures 7-16 compare the other variables only at the last training time step (K=10000), and the integrated species in Fig. 8 are not accompanied by an error measure on the prediction interval. The paper should either add quantitative held-out comparisons for temperature, velocity, and species (e.g., errors on a monitor set or over the spatial field during the 2 ms test window) or explicitly restrict the extrapolation claim to pressure.
  2. [Appendix B / Section III.C] The authors correctly acknowledge in the Conclusion that 'the quadratic model form is an approximation,' and Appendix B shows why: the pressure equation contains temperature-dependent enthalpies h_l(T) and their time derivatives, and the species equations contain Arrhenius source terms and terms proportional to rho c_l dxi/dx that are not quadratic in q_L. This is load-bearing because the paper motivates the method as exposing exact quadratic structure; the numerical extrapolation claim therefore rests on the empirical smallness of the omitted non-quadratic terms over the 2 ms test window. I ask for a direct assessment of this: for example, compute the residual of the exact GEMS equations in the q_L variables against the learned quadratic operators on the training snapshots, or report time-dependent errors of temperature and species over the test interval. Without such a check, the reader cannot distinguish an incidental fit from the claimed structure exploitation.
minor comments (4)
  1. [IV.C] The parenthetical 'K=10000 (which corresponds to t=0.0159999s)' is inconsistent with the stated time step dt=1e-7 and the 1 ms training interval; with these values K=10000 corresponds to t=0.001 s. Please correct the time label or clarify the simulation clock.
  2. [Figures 9-16] These figures are titled 'Predictive results ... at the last time step of training data'; since this time step is inside the training window, the word 'predictive' is misleading and should be replaced by 'reconstructed' or 'training' results.
  3. [IV.A / IV.C] The paper states that 'another 2ms of testing data at the monitor locations' are available, but it would be helpful to state explicitly how many test snapshots are used and whether any of the test data were used in the L-curve or basis selection.
  4. [III.C] The text accurately says 'many terms in the governing equations take polynomial form,' but the abstract and introduction should avoid implying exact polynomial structure for the full system; consider using 'approximate quadratic structure' consistently throughout.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: ROM operators are fit to 1 ms training data and evaluated on 2 ms held-out monitor data; non-quadratic terms are explicitly acknowledged as an approximation.

full rationale

The derivation chain is: collect 10,000 GEMS snapshots over 1 ms; transform to the learning variables q_L and scale them; form a POD basis from those snapshots; solve the regularized least-squares problem (Eq. 16) for the quadratic ROM operators; then integrate the ROM and compare with 2 ms of GEMS monitor data not used to fit the operators. The fitted objects are the operators A, H, B, c; the claimed result is a time-domain prediction on a held-out interval. These are not the same by construction: the 2 ms pressure traces in Figs. 5-6 lie outside the training window, so the extrapolation is not forced by the least-squares fit and can fail. The paper's own Appendix B shows that the pressure and species equations contain non-polynomial terms (temperature-dependent enthalpies h_l(T), Arrhenius factors, and rho c_l terms), and the authors state that 'the quadratic model form is an approximation for this particular application problem.' This limits the strength of extrapolation claims for temperature and species, but it is a modeling approximation, not a circular step. Self-citations to [34], [35], and [37] are method and precedent citations; the least-squares decoupling claim is elementary and the lifting transformations are re-derived in Appendix A, so no load-bearing argument reduces to an unverified self-citation. One evidentiary gap exists: field plots for temperature, velocity, and species are shown at the last training time step, while beyond-training evidence is mainly the pressure traces at one monitor location; that is a support/overclaim concern, not circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the quadratic ansatz and the POD subspace assumption, both of which are approximations. The ROM operators and regularization parameter are fit to data. The lifting variables are mathematical constructs used in the derivation but not in the final ROM.

free parameters (3)
  • Regularization parameter lambda = 1.0e5, 3.0e5 (r=24); 3.0e4, 5.0e4 (r=29)
    Selected via L-curve to stabilize the least squares; affects ROM operators and stability.
  • POD basis dimension r = 24 and 29
    Chosen by cumulative energy thresholds 98.5% and 99%; user-specified.
  • ROM operators (A, F, B, c) = Learned matrices from least squares
    Outputs of the regularized least squares fit to training snapshots; central fitted quantities.
assumptions (5)
  • ad hoc to paper The combustion dynamics can be approximated by a quadratic polynomial ROM in the learning variables q_L = [p, vx, vy, xi, c1..c4].
    The derivation shows only the specific volume and velocity equations are exactly quadratic; pressure and species equations contain non-polynomial terms (Appendix B). The ROM nevertheless uses quadratic form, an approximation the authors acknowledge in Section V.
  • domain assumption The POD subspace computed from 1ms of training snapshots spans the dynamics over the subsequent 2ms test interval.
    All ROMs rely on this for extrapolation; no guarantee for this nonlinear multi-scale system.
  • domain assumption The high-fidelity GEMS simulation is the ground truth and its snapshots are sufficiently resolved.
    The paper uses GEMS data as truth without quantifying CFD discretization error.
  • ad hoc to paper Smooth solutions: partial derivatives of rho exist; no discontinuities considered.
    Stated in Section III.C: 'we are assuming the existence of these partial derivatives'.
  • standard math The five-point stencil approximates time derivatives accurately enough for the least squares fit.
    Fourth-order accurate in the interior, first-order at boundaries; standard numerical assumption.
invented entities (1)
  • Auxiliary lifting variables w1..w6 (e.g., w1 = c1^0.2, w5 = exp(-Ea/(Ru T)))
    purpose: To show the chemical source terms can be written in polynomial (cubic/quintic) form; they are not included in the learned ROM state.
    Mathematical auxiliary variables introduced in Appendix A; not physical and not directly validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning physics-based reduced-order models for a single-injector combustion process." pith.science (2026). https://pith.science/paper/WXJV7H2E

@misc{pith2026190803620,
  author       = {Pith},
  title        = {Pith review of: Learning physics-based reduced-order models for a single-injector combustion process},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WXJV7H2E}},
  note         = {Machine review of arXiv:1908.03620}
}
read the original abstract

This paper presents a physics-based data-driven method to learn predictive reduced-order models (ROMs) from high-fidelity simulations, and illustrates it in the challenging context of a single-injector combustion process. The method combines the perspectives of model reduction and machine learning. Model reduction brings in the physics of the problem, constraining the ROM predictions to lie on a subspace defined by the governing equations. This is achieved by defining the ROM in proper orthogonal decomposition (POD) coordinates, which embed the rich physics information contained in solution snapshots of a high-fidelity computational fluid dynamics (CFD) model. The machine learning perspective brings the flexibility to use transformed physical variables to define the POD basis. This is in contrast to traditional model reduction approaches that are constrained to use the physical variables of the high-fidelity code. Combining the two perspectives, the approach identifies a set of transformed physical variables that expose quadratic structure in the combustion governing equations and learns a quadratic ROM from transformed snapshot data. This learning does not require access to the high-fidelity model implementation. Numerical experiments show that the ROM accurately predicts temperature, pressure, velocity, species concentrations, and the limit-cycle amplitude, with speedups of more than five orders of magnitude over high-fidelity models. Our ROM simulation is shown to be predictive 200% past the training interval. Moreover, ROM-predicted pressure traces accurately match the phase of the pressure signal and yield good approximations of the limit-cycle amplitude.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Sampling low-dimensional Markovian dynamics for pre-asymptotically recovering reduced models from data with operator inference

    math.NA 2019-08 accept novelty 7.0 of 10

    Re-projecting the full system after every time step before collecting data lets operator inference recover, with finite data, exactly the same reduced models that intrusive Galerkin projection produces.

Reference graph

Works this paper leans on

50 extracted references · 46 canonical work pages · cited by 1 Pith paper

  1. [1]

    Aerodynamic data reconstruction and inverse design using proper orthogonal decomposition,

    Bui-Thanh, T., Damodaran, M., and Willcox, K. E., “Aerodynamic data reconstruction and inverse design using proper orthogonal decomposition,”AIAA Journal, Vol. 42, No. 8, 2004, pp. 1505–1516

    work page 2004

  2. [2]

    Low Order Galerkin Models for the Actuated Flow Around 2-D Airfoils,

    Tadmor, G., Centuori, M., Lehmann, O., Noack, B., Luctenburg, M., and Morzynski, M., “Low Order Galerkin Models for the Actuated Flow Around 2-D Airfoils,”45th AIAA Aerospace Sciences Meeting and Exhibit, 2007, p. 1313

    work page 2007

  3. [3]

    Adaptation of Aeroelastic Reduced-Order Models and Application to an F-16 Configuration,

    Lieu, T., and Farhat, C., “Adaptation of Aeroelastic Reduced-Order Models and Application to an F-16 Configuration,”AIAA Journal, Vol. 45, No. 6, 2007, pp. 1244–1257

    work page 2007

  4. [4]

    TowardsReal-TimeComputational-Fluid-Dynamics-BasedAeroelasticComputations using a Database of Reduced-Order Information,

    Amsallem,D.,Cortial,J.,andFarhat,C.,“TowardsReal-TimeComputational-Fluid-Dynamics-BasedAeroelasticComputations using a Database of Reduced-Order Information,”AIAA Journal, Vol. 48, No. 9, 2010, pp. 2029–2037

    work page 2010

  5. [5]

    Reduced Order Models for a High Speed Jet with Time-Resolved PIV,

    Berger, Z., Low, K., Berry, M., Glauser, M., Kostka, S., Gogineni, S., Cordier, L., and Noack, B., “Reduced Order Models for a High Speed Jet with Time-Resolved PIV,”51st AIAA Aerospace Sciences Meeting including the New Horizons Forum and Aerospace Exposition, 2013, p. 11

    work page 2013

  6. [6]

    Reduced-OrderUnsteadyAerodynamicModelsatLowReynoldsNumbers,

    Brunton,S.L.,Rowley,C.W.,andWilliams,D.R.,“Reduced-OrderUnsteadyAerodynamicModelsatLowReynoldsNumbers,” Journal of Fluid Mechanics, Vol. 724, 2013, p. 203–233

    work page 2013

  7. [7]

    Sparse Sensing and DMD-Based Identification of Flow Regimes and Bifurcations in Complex Flows,

    Kramer, B., Grover, P., Boufounos, P., Nabi, S., and Benosman, M., “Sparse Sensing and DMD-Based Identification of Flow Regimes and Bifurcations in Complex Flows,”SIAM Journal on Applied Dynamical Systems, Vol. 16, No. 2, 2017, pp. 1164–1196

    work page 2017

  8. [8]

    Model reduction for reacting flow applications,

    Nguyen, V., Buffoni, M., Willcox, K., and Khoo, B., “Model reduction for reacting flow applications,”International Journal of Computational Fluid Dynamics, Vol. 28, No. 3-4, 2014, pp. 91–105

    work page 2014

Show all 50 references
  1. [9]

    Model Order Reduction for Reacting Flows: Laminar Gaussian Flame Applications,

    Nguyen, V. B., Dou, H.-S., Willcox, K., and Khoo, B.-C., “Model Order Reduction for Reacting Flows: Laminar Gaussian Flame Applications,”30th International Symposium on Shock Waves 1, Springer, 2017, pp. 337–343

    2017

  2. [10]

    Projection-based model reduction for reacting flows,

    Buffoni, M., and Willcox, K., “Projection-based model reduction for reacting flows,”40th Fluid Dynamics Conference and Exhibit, Chicago, IL, June 28-July 1, 2010, p. 5008. 21

    2010

  3. [11]

    Investigations and Improvement of Robustness of Reduced-Order Models of Reacting Flow,

    Huang, C., Duraisamy, K., and Merkle, C. L., “Investigations and Improvement of Robustness of Reduced-Order Models of Reacting Flow,”AIAA Journal, Vol. 0, No. 0, 0, pp. 1–13. doi:10.2514/1.J058392, URLhttps://doi.org/10.2514/1. J058392

    doi

  4. [12]

    Exploration of reduced-order models for rocket combustion applications,

    Huang, C., Xu, J., Duraisamy, K., and Merkle, C., “Exploration of reduced-order models for rocket combustion applications,” 2018 AIAA Aerospace Sciences Meeting, 2018, p. 1183

    2018

  5. [13]

    Non-intrusive reduced order modeling of unsteady flows using artificial neural networks with application to a combustion problem,

    Wang, Q., Hesthaven, J. S., and Ray, D., “Non-intrusive reduced order modeling of unsteady flows using artificial neural networks with application to a combustion problem,”Journal of computational physics, Vol. 384, 2019, pp. 289–307

    2019

  6. [14]

    The Structure of Inhomogeneous Turbulent Flows,

    Lumley, J. L., “The Structure of Inhomogeneous Turbulent Flows,”Atmospheric Turbulence and Radio Wave Propagation, 1967

    1967

  7. [15]

    Turbulence and the Dynamics of Coherent Structures. I-Coherent Structures. II-Symmetries and Transformations. III-Dynamics and Scaling,

    Sirovich, L., “Turbulence and the Dynamics of Coherent Structures. I-Coherent Structures. II-Symmetries and Transformations. III-Dynamics and Scaling,”Quarterly of Applied Mathematics, Vol. 45, 1987, pp. 561–571

    1987

  8. [16]

    Aposteriorierrorestimationforreduced-basisapproximationofparametrizedelliptic coercive partial differential equations: ‘convex inverse’ bound conditioners,

    Veroy,K.,Rovas,D.V.,andPatera,A.T.,“Aposteriorierrorestimationforreduced-basisapproximationofparametrizedelliptic coercive partial differential equations: ‘convex inverse’ bound conditioners,”ESAIM: Control, Optimisation and Calculus of Variations, Vol. 8, 2002, pp. 1007–1028

    2002

  9. [17]

    Efficient Reduced-Basis Treatment of Nonaffine and Nonlinear Partial Differential Equations,

    Grepl, M. A., Maday, Y., Nguyen, N. C., and Patera, A. T., “Efficient Reduced-Basis Treatment of Nonaffine and Nonlinear Partial Differential Equations,”ESAIM: Mathematical Modelling and Numerical Analysis, Vol. 41, No. 3, 2007, pp. 575–605

    2007

  10. [18]

    PODa-posteriorierrorestimatesforlinear-quadraticoptimalcontrolproblems,

    Tröltzsch,F.,andVolkwein,S.,“PODa-posteriorierrorestimatesforlinear-quadraticoptimalcontrolproblems,” Computational Optimization and Applications, Vol. 44, No. 1, 2009, p. 83

    2009

  11. [19]

    S., Rozza, G., and Stamm, B.,Certified reduced basis methods for parametrized partial differential equations, Springer, 2016

    Hesthaven, J. S., Rozza, G., and Stamm, B.,Certified reduced basis methods for parametrized partial differential equations, Springer, 2016

    2016

  12. [20]

    Model reduction for compressible flows using POD and Galerkin projection,

    Rowley, C. W., Colonius, T., and Murray, R. M., “Model reduction for compressible flows using POD and Galerkin projection,” Physica D: Nonlinear Phenomena, Vol. 189, No. 1-2, 2004, pp. 115–129

    2004

  13. [21]

    Stable Galerkin reduced order models for linearized compressible flow,

    Barone, M. F., Kalashnikova, I., Segalman, D. J., and Thornquist, H. K., “Stable Galerkin reduced order models for linearized compressible flow,”Journal of Computational Physics, Vol. 228, No. 6, 2009, pp. 1932–1946

    2009

  14. [22]

    Reliablereduced-ordermodelsfortime-dependentlinearizedEulerequations,

    Serre,G.,Lafon,P.,Gloerfelt,X.,andBailly,C.,“Reliablereduced-ordermodelsfortime-dependentlinearizedEulerequations,” Journal of Computational Physics, Vol. 231, No. 15, 2012, pp. 5176–5194

    2012

  15. [23]

    Stable and efficient Galerkin reduced order models for non-linear fluid flow,

    Kalashnikova, I., and Barone, M., “Stable and efficient Galerkin reduced order models for non-linear fluid flow,”6th AIAA Theoretical Fluid Mechanics Conference, 2011, p. 3110

    2011

  16. [24]

    MinimalsubspacerotationontheStiefelmanifoldforstabilizationandenhancement of projection-based reduced order models for the compressible Navier–Stokes equations,

    Balajewicz,M.,Tezaur,I.,andDowell,E.,“MinimalsubspacerotationontheStiefelmanifoldforstabilizationandenhancement of projection-based reduced order models for the compressible Navier–Stokes equations,”Journal of Computational Physics, Vol. 321, 2016, pp. 224–241

    2016

  17. [25]

    Galerkin v. least-squares Petrov–Galerkin projection in nonlinear model reduction,

    Carlberg, K., Barone, M., and Antil, H., “Galerkin v. least-squares Petrov–Galerkin projection in nonlinear model reduction,” Journal of Computational Physics, Vol. 330, 2017, pp. 693–734

    2017

  18. [26]

    Challenges in Reduced Order Modeling of Reacting Flows,

    Huang, C., Duraisamy, K., and Merkle, C., “Challenges in Reduced Order Modeling of Reacting Flows,”2018 Joint Propulsion Conference, 2018, p. 4675

    2018

  19. [27]

    Multilayer feedforward networks are universal approximators,

    Hornik, K., Stinchcombe, M., and White, H., “Multilayer feedforward networks are universal approximators,”Neural networks, Vol. 2, No. 5, 1989, pp. 359–366

    1989

  20. [28]

    Projection-basedmodelreduction: formulationsforphysics-based machine learning,

    Swischuk,R.,Mainini,L.,Peherstorfer,B.,andWillcox,K.,“Projection-basedmodelreduction: formulationsforphysics-based machine learning,”Computers and Fluids, Vol. 179, 2019, pp. 704–717

    2019

  21. [29]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems,

    Brunton, S. L., Proctor, J. L., and Kutz, J. N., “Discovering governing equations from data by sparse identification of nonlinear dynamical systems,” Vol. 113, No. 15, 2016, pp. 3932–3937. doi:10.1073/pnas.1517384113

    2016 doi

  22. [30]

    Data-driven discovery of partial differential equations,

    Rudy, S. H., Brunton, S. L., Proctor, J. L., and Kutz, J. N., “Data-driven discovery of partial differential equations,”Science Advances, Vol. 3, No. 4, 2017, p. e1602614

    2017

  23. [31]

    Extracting sparse dynamics from limited data,

    Schaeffer, H., Tran, G., and Ward, R., “Extracting sparse dynamics from limited data,”SIAM Journal on Applied Mathematics, Vol. 78, No. 6, 2018, pp. 3279–3295. 22

    2018

  24. [32]

    Dynamic mode decomposition of numerical and experimental data,

    Schmid, P. J., “Dynamic mode decomposition of numerical and experimental data,”Journal of Fluid Mechanics, Vol. 656, 2010, pp. 5–28

    2010

  25. [33]

    Spectral analysis of nonlinear flows,

    Rowley, C. W., Mezić, I., Bagheri, S., Schlatter, P., and Henningson, D. S., “Spectral analysis of nonlinear flows,”Journal of Fluid Mechanics, Vol. 641, 2009, pp. 115–127

    2009

  26. [34]

    Data-drivenoperatorinferencefornonintrusiveprojection-basedmodelreduction,

    Peherstorfer,B.,andWillcox,K.,“Data-drivenoperatorinferencefornonintrusiveprojection-basedmodelreduction,” Computer Methods in Applied Mechanics and Engineering, Vol. 306, 2016, pp. 196–215

    2016

  27. [35]

    Transform&Learn: Adata-drivenapproachtononlinearmodelreduction,

    Qian,E.,Kramer,B.,Marques,A.,andWillcox,K.,“Transform&Learn: Adata-drivenapproachtononlinearmodelreduction,” AIAA Aviation and Aeronautics Forum and Exposition, Dallas, TX, 2019. doi:10.2514/6.2019-3707

    2019 doi

  28. [36]

    QLMOR: A projection-based nonlinear model order reduction approach using quadratic-linear representation of nonlinear systems,

    Gu, C., “QLMOR: A projection-based nonlinear model order reduction approach using quadratic-linear representation of nonlinear systems,”IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, Vol. 30, No. 9, 2011, pp. 1307–1320

    2011

  29. [37]

    Nonlinear Model Order Reduction via Lifting Transformations and Proper Orthogonal Decomposition,

    Kramer, B., and Willcox, K., “Nonlinear Model Order Reduction via Lifting Transformations and Proper Orthogonal Decomposition,”AIAA Journal, Vol. 57, No. 6, 2019, pp. 2297–2307. doi:10.2514/1.J057791

    2019 doi

  30. [38]

    Balanced Truncation Model Reduction for Lifted Nonlinear Systems,

    Kramer, B., and Willcox, K., “Balanced Truncation Model Reduction for Lifted Nonlinear Systems,” Tech. rep., 2019. ArXiv:1907.12084

    2019 arXiv

  31. [39]

    Coupling between hydrodynamics, acoustics, and heat release in a self-excited unstable combustor,

    Harvazinski, M. E., Huang, C., Sankaran, V., Feldman, T. W., Anderson, W. E., Merkle, C. L., and Talley, D. G., “Coupling between hydrodynamics, acoustics, and heat release in a self-excited unstable combustor,”Physics of Fluids, Vol. 27, No. 4, 2015, p. 045102

    2015

  32. [40]

    Modeling self-excited combustion instabilities using a combination of two-and three-dimensional simulations,

    Harvazinski, M. E., “Modeling self-excited combustion instabilities using a combination of two-and three-dimensional simulations,” Ph.D. thesis, Purdue University, 2012

    2012

  33. [41]

    Combustion Dynamics in a Single-Element Lean Direct Injection Gas Turbine Combustor,

    Huang, C., Gejji, R., Anderson, W., Yoon, C., and Sankaran, V., “Combustion Dynamics in a Single-Element Lean Direct Injection Gas Turbine Combustor,”Combustion Science and Technology, 2019, pp. 1–28

    2019

  34. [42]

    Experimentalstudyofhigh-frequencycombustioninstabilityinacontinuously variable resonance combustor (CVRC),

    Yu,Y.,O’Hara,L.,Sisco,J.,andAnderson,W.,“Experimentalstudyofhigh-frequencycombustioninstabilityinacontinuously variable resonance combustor (CVRC),”47th AIAA Aerospace Sciences Meeting, Orlando, FL, 2009

    2009

  35. [43]

    Simplified reaction mechanisms for the oxidation of hydrocarbon fuels in flames,

    Westbrook, C. K., and Dryer, F. L., “Simplified reaction mechanisms for the oxidation of hydrocarbon fuels in flames,” Combustion Science and Technology, Vol. 27, No. 1-2, 1981, pp. 31–43

    1981

  36. [44]

    Reviewandunificationofmethodsforcomputingderivativesofmultidisciplinarycomputational models,

    Martins,J.R.,andHwang,J.T.,“Reviewandunificationofmethodsforcomputingderivativesofmultidisciplinarycomputational models,”AIAA journal, Vol. 51, No. 11, 2013, pp. 2582–2599

    2013

  37. [45]

    Methods for numerical differentiation of noisy data,

    Knowles, I., and Renka, R. J., “Methods for numerical differentiation of noisy data,”Electron. J. Differ. Equ, Vol. 21, 2014, pp. 235–246

    2014

  38. [46]

    Numerical differentiation of noisy, nonsmooth, multidimensional data,

    Chartrand, R., “Numerical differentiation of noisy, nonsmooth, multidimensional data,”2017 IEEE Global Conference on Signal and Information Processing (GlobalSIP), IEEE, 2017, pp. 244–248

    2017

  39. [47]

    Matrix computations. 1996,

    Golub, G. H., and Van Loan, C. F., “Matrix computations. 1996,”Johns Hopkins University, Press, Baltimore, MD, USA, 1996, pp. 374–426

    1996

  40. [48]

    Physics-based machine learning and data-driven reduced-order modeling,

    Swischuk, R. C., “Physics-based machine learning and data-driven reduced-order modeling,” Master’s thesis, Massachusetts Institute of Technology, 2019

    2019

  41. [49]

    A randomized algorithm for the decomposition of matrices,

    Martinsson, P.-G., Rokhlin, V., and Tygert, M., “A randomized algorithm for the decomposition of matrices,”Applied and Computational Harmonic Analysis, Vol. 30, No. 1, 2011, pp. 47–68

    2011

  42. [50]

    The L-curve and its use in the numerical treatment of inverse problems,

    Hansen, P., “The L-curve and its use in the numerical treatment of inverse problems,”Computational Inverse Problems in Electrocardiology, Vol. 5, 2001, pp. 119–142. 23

    2001

Pith tools

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