Pith. sign in

REVIEW 2 major objections 2 minor 52 references

Tensor-based empirical interpolation method and its application in model reduction

T0 review · 2 major / 2 minor · reviewed 2026-05-23 · grok-4.3

Challenge this review Download PDF Read on arXiv ↗

Pith's one-line read The tensor-based empirical interpolation method generates rectangular-grid points, making it equivalent to applying DEIM separately in each direction.

desk verdict Tensor EIM without vectorization is a clean incremental step with a verifiable grid property, but the claim that rectangular grids limit every matrix EIM approach lacks an independent argument. read the letter →

arxiv 2410.21770 v2 submitted 2024-10-29 math.NA cs.NAmath.OC

Brij Nandan Tripathi , Hanumant Singh Shekhawat , Seip Weiland This is my paper
classification math.NAcs.NAmath.OC
keywords tensor-basedinterpolationempiricalmethodmodelreductionmatrixdifferentialequationsDEIMrectangulargridsemi-linearsystems
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 develops a tensor version of the empirical interpolation method that approximates matrix-valued functions directly, without first converting them to vectors. Mathematical analysis proves that the method always selects interpolation points arranged in a rectangular grid. Because of this grid structure, the new method is exactly equivalent to running the DEIM procedure independently along each matrix dimension. When the approach is used for model reduction of semi-linear matrix differential equations, it requires less offline and online computation than standard vector DEIM while incurring only a small accuracy penalty.

What carries the argument

Tensor-based extension of the empirical interpolation method, which selects points that form a rectangular grid and thereby becomes equivalent to directional DEIM.

What would settle it

A concrete semi-linear matrix differential equation in which the tensor method produces an approximation error more than a few percent larger than vector DEIM, or in which the reported computation-time savings disappear.

Watch

Extended reading notes

Core claim

The proposed tensor-based empirical interpolation method for matrix-valued functions produces interpolation points that always form a rectangular grid. This grid property makes the method equivalent to the DEIM method applied independently in each direction. The resulting procedure avoids vectorization, lowers both offline and online computational effort, and is demonstrated on the model reduction of semi-linear matrix differential equations, where it runs faster than vector DEIM at the cost of a minor accuracy reduction.

Load-bearing premise

The rectangular grid of interpolation points does not impose a practically significant accuracy penalty for the semi-linear matrix differential equations targeted in the model-reduction application.

Editorial extensions

If this is right

  • The method reduces offline and online computation time compared with vector DEIM for matrix-valued functions.
  • It applies directly to model reduction of semi-linear matrix differential equations.
  • Any matrix-based EIM or DEIM procedure will inherit the rectangular-grid limitation.
  • The equivalence allows the tensor method to be analyzed and implemented as separate one-dimensional DEIM steps.

Reading between the lines

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

  • The grid constraint may become more restrictive for higher-order tensors, where the number of points grows with the product of the individual dimensions.
  • Improvements developed for standard DEIM, such as better point-selection heuristics, can be transferred immediately to the tensor version through the established equivalence.
  • Problems whose optimal interpolation points lie far from any rectangular grid may require entirely different approximation strategies.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes a tensor-based empirical interpolation method (EIM) for directly approximating matrix-valued functions without vectorization. It develops a theoretical framework showing that the method always selects interpolation points forming a rectangular grid and is equivalent to applying the DEIM method separately along each matrix direction. This is applied to model reduction of semi-linear matrix differential equations, with numerical results indicating reduced offline/online computation time compared to vector DEIM at the cost of minor accuracy loss.

Significance. If the grid property and per-direction equivalence hold rigorously, the work clarifies structural behavior of tensor EIM extensions and offers efficiency gains for matrix/tensor data in model reduction. The numerical comparison provides practical evidence of the claimed trade-off, though its weight depends on the completeness of error tables and implementation details.

major comments (2)
  1. [theoretical analysis / abstract] Abstract and theoretical analysis section: the claim that the rectangular-grid property 'underscores a fundamental limitation that applies to any matrix-based approach relying on widely used techniques like EIM or DEIM' lacks an independent argument. The derivation establishes the grid structure and DEIM equivalence specifically for the proposed tensor construction; no reduction or counter-example is supplied showing why a standard vectorized EIM (selecting arbitrary indices in the flattened matrix) must also be restricted to rectangular grids rather than arbitrary point sets.
  2. [numerical experiments / application] Model reduction application and numerical comparison section: the reported 'minor compromise with accuracy' versus vector DEIM is presented without tabulated error metrics (e.g., relative L2 or Frobenius errors), number of interpolation points used in each method, or explicit verification that the grid constraint is the source of the observed difference. This makes it difficult to assess whether the weakest assumption (no practically significant penalty) holds for the targeted semi-linear matrix DEs.
minor comments (2)
  1. [theoretical framework] Notation for the tensor bases and selection operators should be introduced with explicit definitions before the equivalence proof to improve readability.
  2. [conclusions] The manuscript would benefit from a short remark clarifying whether the rectangular-grid result extends immediately to higher-order tensors or is limited to the matrix case.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and will make the necessary revisions to clarify claims and strengthen the numerical evidence.

read point-by-point responses

  1. Referee: [theoretical analysis / abstract] Abstract and theoretical analysis section: the claim that the rectangular-grid property 'underscores a fundamental limitation that applies to any matrix-based approach relying on widely used techniques like EIM or DEIM' lacks an independent argument. The derivation establishes the grid structure and DEIM equivalence specifically for the proposed tensor construction; no reduction or counter-example is supplied showing why a standard vectorized EIM (selecting arbitrary indices in the flattened matrix) must also be restricted to rectangular grids rather than arbitrary point sets.

    Authors: We agree that the derivation and equivalence result apply specifically to the tensor-based construction presented in the paper. The broader claim regarding a fundamental limitation for any matrix-based EIM/DEIM approach is not supported by an independent argument or counter-example for vectorized methods, and the statement overreaches. We will revise the abstract and theoretical analysis section to remove this general claim and restrict the discussion to the properties and limitations of the proposed tensor EIM. revision: yes

  2. Referee: [numerical experiments / application] Model reduction application and numerical comparison section: the reported 'minor compromise with accuracy' versus vector DEIM is presented without tabulated error metrics (e.g., relative L2 or Frobenius errors), number of interpolation points used in each method, or explicit verification that the grid constraint is the source of the observed difference. This makes it difficult to assess whether the weakest assumption (no practically significant penalty) holds for the targeted semi-linear matrix DEs.

    Authors: We acknowledge the need for more detailed numerical reporting. In the revised manuscript, we will add tables presenting relative L2 and Frobenius errors for both the tensor EIM and vector DEIM approaches, explicitly state the number of interpolation points selected by each method, and include discussion verifying the role of the rectangular grid constraint in the observed accuracy differences. This will provide clearer evidence for the claimed efficiency-accuracy trade-off. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; derivation is self-contained

full rationale

The paper derives the rectangular-grid property and the per-direction DEIM equivalence directly from the construction and mathematical analysis of its tensor-EIM extension. These results follow from the method's own definitions and are not equivalent to inputs by construction, nor do they rely on fitted parameters renamed as predictions or load-bearing self-citations. The assertion that the grid property is a fundamental limitation for any matrix-based EIM/DEIM is an interpretive claim without separate reduction shown, but it does not make the core theoretical results circular. The derivation chain remains independent and self-contained.

Assumptions & free parameters 0 free parameters · 1 assumptions · 0 invented entities

The method rests on standard approximation-theory assumptions that a matrix-valued function admits sufficiently accurate low-rank tensor bases; no free parameters or new entities are introduced in the abstract.

assumptions (1)
  • domain assumption A matrix-valued function admits a good approximation in a tensor-product basis of chosen row and column spaces.
    Required for the interpolation operator to be well-defined and for the rectangular-grid property to hold.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tensor-based empirical interpolation method and its application in model reduction." pith.science (2026). https://pith.science/paper/2410.21770

@misc{pith2026241021770,
  author       = {Pith},
  title        = {Pith review of: Tensor-based empirical interpolation method and its application in model reduction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2410.21770}},
  note         = {Machine review of arXiv:2410.21770}
}
read the original abstract

In general, matrix or tensor-valued functions are approximated using the method developed for vector-valued functions by transforming the matrix-valued function into vector form. This paper proposes a tensor-based interpolation method to approximate a matrix-valued function without transforming it into the vector form. The tensor-based technique has the advantage of reducing offline and online computation without sacrificing much accuracy. The proposed method is an extension of the empirical interpolation method (EIM) for tensor bases. This paper presents a necessary theoretical framework to understand the method's functioning and limitations. Our mathematical analysis establishes a key characteristic of the proposed method: it consistently generates interpolation points in the form of a rectangular grid. This observation underscores a fundamental limitation that applies to any matrix-based approach relying on widely used techniques like EIM or DEIM method. It has also been theoretically shown that the proposed method is equivalent to the DEIM method applied in each direction due to the rectangular grid structure of the interpolation points. The application of the proposed method is shown in the model reduction of the semi-linear matrix differential equation. We have compared the approximation result of our proposed method with the DEIM method used to approximate a vector-valued function. The comparison result shows that the proposed method takes less time, albeit with a minor compromise with accuracy.

Figures

Figures reproduced from arXiv: 2410.21770 by the authors.

Figure 2
Figure 2. Average Computation Time for a different number of [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Original and Approximated function for µ = (−0.7879, −0.7171). On the basis of accuracy outcomes as observed in Examples 1 and 2, it can be inferred that in cases where tensor POD bases exhibit better approximation subspace compared to conventional POD bases, the TEIM method is preferred over the DEIM method. 4.2 TEIM reduced semi-linear matrix differential equation The semi-linear matrix differential equation is de… view at source ↗
Figure 5
Figure 5. Original and Approximated function for the [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

  1. [1]

    Approximation of Large-Scale Dynamical Sy stems

    Antoulas, A. Approximation of Large-Scale Dynamical Sy stems. (SIAM,2005)

    work page 2005

  2. [2]

    & Javaid, N

    Khalid, R. & Javaid, N. A survey on hyperparameters optim ization algorithms of forecasting models in smart grid. Sustainable Cities And Society . 61 pp. 102275 (2020) 23 A PREPRINT - O CTOBER 30, 2024

    work page 2020

  3. [3]

    & Tindall, M

    Snowden, T., Graaf, P . & Tindall, M. Methods of model redu ction for large-scale biological systems: a survey of current methods and trends. Bulletin Of Mathematical Biology . 79, 1449-1486 (2017)

    work page 2017

  4. [4]

    Krylov subspace techniques for reduced-order mo deling of large-scale dynamical systems

    Bai, Z. Krylov subspace techniques for reduced-order mo deling of large-scale dynamical systems. Applied Nu- merical Mathematics. 43, 9-44 (2002)

    work page 2002

  5. [5]

    Model reduction methods based on Krylov subsp aces

    Freund, R. Model reduction methods based on Krylov subsp aces. Acta Numerica. 12 pp. 267-319 (2003)

    work page 2003

  6. [6]

    & Roberts, R

    Mullis, C. & Roberts, R. Synthesis of minimum roundoff no ise fixed point digital filters. IEEE Transactions On Circuits And Systems. 23, 551-562 (1976)

    work page 1976

  7. [7]

    Principal component analysis in linear system s: Controllability, observability, and model reduction

    Moore, B. Principal component analysis in linear system s: Controllability, observability, and model reduction. IEEE Transactions On Automatic Control. 26, 17-32 (1981)

    work page 1981

  8. [8]

    The structure of inhomogeneous turbulent flow s

    Lumley, J. The structure of inhomogeneous turbulent flow s. Atmospheric Turbulence And Radio W ave Propaga- tion. (1967)

    work page 1967

Show all 52 references
  1. [9]

    Analysis of a complex of statistical varia bles into principal components

    Hotelling, H. Analysis of a complex of statistical varia bles into principal components. Journal Of Educational Psychology. 24 (1933)

    1933

  2. [10]

    Nonlinear model reduction for uncertaint y quantification in large-scale inverse problems: application to nonlinear convection-diffusion-reaction equation

    Galbally, D. Nonlinear model reduction for uncertaint y quantification in large-scale inverse problems: application to nonlinear convection-diffusion-reaction equation. (M assachusetts Institute of Technology,2008)

    2008

  3. [11]

    & Amsallem, D

    Carlberg, K., Farhat, C., Cortial, J. & Amsallem, D. The GNA T method for nonlinear model reduction: effective implementation and application to computational fluid dyna mics and turbulent flows. Journal Of Computational Physics. 242 pp. 623-647 (2013)

    2013

  4. [12]

    & Sirovich, L

    Everson, R. & Sirovich, L. Karhunen-Loeve procedure fo r gappy data. JOSA A. 12, 1657-1664 (1995)

    1995

  5. [13]

    & Backx, T

    Astrid, P ., Weiland, S., Willcox, K. & Backx, T. Missing point estimation in models described by proper orthog- onal decomposition. 2004 43rd IEEE Conference On Decision And Control (CDC) (IEE E Cat. No.04CH37601) . 2 (2004)

    2004

  6. [14]

    & Maday, Y

    Barrault, M., Nguyen, C., Patera, A. & Maday, Y . An ‘empi rical interpolation’ method: application to efficient reduced-basis discretization of partial differential equations. Comptes Rendus De L’Académie Des Sciences. Série I, Mathématique. 339-9 pp. 667-672 (2004)

    2004

  7. [15]

    & Sorensen, D

    Chaturantabut, S. & Sorensen, D. Discrete Empirical In terpolation for nonlinear model reduction. Proceedings Of The 48h IEEE Conference On Decision And Control (CDC) Held Jointly With 2009 28th Chinese Control Conference. pp. 4316-4321 (2009)

    2009

  8. [16]

    & Jadbabaie, A

    Huang, Y . & Jadbabaie, A. Nonlinear H<sub> ∞ </sub> control: an enhanced quasi-LPV approach. IF AC Pro- ceedings V olumes. 32, 2754-2759 (1999), 14th IFAC World Congress 1999, Beijing, China, 5-9 July

    1999

  9. [17]

    & Gugercin, S

    Drmac, Z. & Gugercin, S. A new selection operator for the discrete empirical interpolation method—improved a priori error bound and extensions. SIAM Journal On Scientific Computing . 38, A631-A648 (2016)

    2016

  10. [18]

    & Bungartz, H

    Peherstorfer, B., Butnaru, D., Willcox, K. & Bungartz, H. Localized discrete empirical interpolation method. SIAM Journal On Scientific Computing . 36, A168-A192 (2014)

    2014

  11. [19]

    & Amsallem, D

    Negri, F., Manzoni, A. & Amsallem, D. Efficient model red uction of parametrized systems by matrix discrete empirical interpolation. Journal Of Computational Physics . 303 pp. 431-454 (2015)

    2015

  12. [20]

    Model reduction for transport-domin ated problems via online adaptive bases and adaptive sam- pling

    Peherstorfer, B. Model reduction for transport-domin ated problems via online adaptive bases and adaptive sam- pling. SIAM Journal On Scientific Computing . 42, A2803-A2836 (2020)

    2020

  13. [21]

    & Willcox, K

    Peherstorfer, B. & Willcox, K. Online adaptive model re duction for nonlinear systems via low-rank updates. SIAM Journal On Scientific Computing . 37, A2123-A2150 (2015)

    2015

  14. [22]

    & Stewart, W

    Langville, A. & Stewart, W . The Kronecker Product and St ochastic Automata Networks. J. Comput. Appl. Math.. 167 pp. 429-447 (2004,6)

    2004

  15. [23]

    Radhakrishna Rao Estimation of Heteroscedastic V ar iances in Linear Models

    C. Radhakrishna Rao Estimation of Heteroscedastic V ar iances in Linear Models. Journal Of The American Sta- tistical Association. 65, 161-172 (1970)

    1970

  16. [24]

    & Kiliçman, A

    Al Zhour, Z. & Kiliçman, A. Some new connections between matrix products for partitioned and non-partitioned matrices. Computers & Mathematics With Applications . 54, 763-784 (2007)

    2007

  17. [25]

    & Patera, A

    Rozza, G., Huynh, D. & Patera, A. Reduced Basis Approxim ation and a Posteriori Error Estimation for Affinely Parametrized Elliptic Coercive Partial Differential Equa tions. Archives Of Computational Methods In Engineer- ing. 15, 229 - 275 (2008)

    2008

  18. [26]

    & Gü ˘gercin, S

    Antoulas, A., Beattie, C. & Gü ˘gercin, S. Interpolatory methods for model reduction. (SIA M,2020) 24 A PREPRINT - O CTOBER 30, 2024

    2020

  19. [27]

    & Willcox, K

    Benner, P ., Gugercin, S. & Willcox, K. A survey of projec tion-based model reduction methods for parametric dynamical systems. SIAM Review. 57, 483-531 (2015)

    2015

  20. [28]

    & Gugercin, S

    Antoulas, A., Beattie, C. & Gugercin, S. Interpolatory model reduction of large-scale dynamical systems. Effi- cient Modeling And Control Of Large-scale Systems . pp. 3-58 (2010)

    2010

  21. [29]

    Randomized discrete empirical interpolat ion method for nonlinear model reduction

    Saibaba, A. Randomized discrete empirical interpolat ion method for nonlinear model reduction. SIAM Journal On Scientific Computing . 42, A1582-A1608 (2020)

    2020

  22. [30]

    Model order reduction using the discrete emp irical interpolation method

    Dedden, R. Model order reduction using the discrete emp irical interpolation method. (2012)

    2012

  23. [31]

    & Weiland, S

    Shekhawat, H. & Weiland, S. A Locally Convergent Jacobi Iteration for the Tensor Singular V alue Problem. Multidimensional System Signal Process. . (2018)

    2018

  24. [32]

    Signal processing based on multilinea r algebra

    De Lathauwer, L. Signal processing based on multilinea r algebra. (Katholieke Universiteit Leuven Leuven,1997)

    1997

  25. [33]

    & Pitsianis, N

    Loan, C. & Pitsianis, N. Approximation with Kronecker P roducts. Linear Algebra F or Large Scale And Real- Time Applications. pp. 293-314 (1993)

    1993

  26. [34]

    Introductory functional analysis with ap plications

    Kreyszig, E. Introductory functional analysis with ap plications. (John Wiley & Sons,1991)

    1991

  27. [35]

    & Y oungson, M

    Rynne, B. & Y oungson, M. Linear functional analysis. (S pringer Science & Business Media,2007)

    2007

  28. [36]

    & V andenberghe, L

    Boyd, S., Boyd, S. & V andenberghe, L. Convex optimizati on. (Cambridge university press,2004)

    2004

  29. [37]

    & Scholz, L

    Ali Eshtewy, N. & Scholz, L. Model reduction for kinetic models of biological systems. Symmetry. 12, 863 (2020)

    2020

  30. [38]

    & Simoncini, V

    Kirsten, G. & Simoncini, V . A matrix-oriented POD-DEIM algorithm applied to semilinear matrix differential equations. (arXiv,2020)

    2020

  31. [39]

    & Dawson, S

    Rowley, C. & Dawson, S. Model Reduction for Flow Analysi s and Control. Annual Review Of Fluid Mechanics . 49, 387-417 (2017)

    2017

  32. [40]

    & Haasdonk, B

    Wirtz, D., Sorensen, D. & Haasdonk, B. A posteriori erro r estimation for DEIM reduced nonlinear dynamical systems. SIAM Journal On Scientific Computing . 36, A311-A338 (2014)

    2014

  33. [41]

    Model reduction of nonlinear mechanical s ystems via optimal projection and tensor approximation

    Carlberg, K. Model reduction of nonlinear mechanical s ystems via optimal projection and tensor approximation. (Stanford University,2011)

    2011

  34. [42]

    & Bader, B

    Kolda, T. & Bader, B. Tensor Decompositions and Applica tions. SIAM Rev.. 51 pp. 455-500 (2009)

    2009

  35. [43]

    Multilinear POD-DEIM model reduction for 2 D and 3D semilinear systems of differential equations

    Kirsten, G. Multilinear POD-DEIM model reduction for 2 D and 3D semilinear systems of differential equations. Journal Of Computational Dynamics . 9 pp. 159-183 (2022)

    2022

  36. [44]

    & Huynh, K

    Choi, Y ., Brown, P ., Arrighi, W ., Anderson, R. & Huynh, K . Space–time reduced order model for large-scale linear dynamical systems with application to Boltzmann tra nsport problems. Journal Of Computational Physics . 424 pp. 109845 (2021)

    2021

  37. [45]

    & Willcox, K

    McBane, S., Choi, Y . & Willcox, K. Stress-constrained t opology optimization of lattice-like structures using component-wise reduced order models. Computer Methods In Applied Mechanics And Engineering . 400 pp. 115525 (2022), https://www.sciencedirect.com/science/article/pii/...

    2022

  38. [46]

    & Kadeethum, T

    Cheung, S., Choi, Y ., Springer, H. & Kadeethum, T. Data- scarce surrogate modeling of shock-induced pore collapse process. ArXiv Preprint arXiv:2306.00184. (2023)

    2023

  39. [47]

    & Y oon , H

    Kadeethum, T., Jakeman, J., Choi, Y ., Bouklas, N. & Y oon , H. Epistemic Uncertainty-A ware Barlow Twins Reduced Order Modeling for Nonlinear Contact Problems. IEEE Access. 11 pp. 62970-62985 (2023)

    2023

  40. [48]

    & Patera, A

    Y ano, M. & Patera, A. An LP empirical quadrature procedu re for reduced basis treatment of parametrized nonlinear PDEs. Computer Methods In Applied Mechanics And Engineering . 344 pp. 1104-1123 (2019), https://www.sciencedirect.com/science/article/pii/S0045782518301087

    2019

  41. [49]

    & Huynh, K

    Lauzon, J., Cheung, S., Shin, Y ., Choi, Y ., Copeland, D. & Huynh, K. S-OPT: A points selection algorithm for hyper-reduction in reduced order models. ArXiv Preprint arXiv:2203.16494. (2022)

    2022

  42. [50]

    & Zhou, Z

    Zhang, D. & Zhou, Z. (2D)2PCA: Two-directional two-dim ensional PCA for efficient face representation and recognition. Neurocomputing., https://www.sciencedirect.com/science/article/pii /S0925231205001785, Neural Networks in Signal Processing

  43. [51]

    & Shekhawat, H

    Tripathi, B. & Shekhawat, H. Model Reduction of Large Sc ale Dynamical System with a Selection of States. SSRN

  44. [52]

    test.png

    D. Bertsekas, Nonlinear Programming, 2nd ed. Athena Scientific, 1999, vol. 4. 25 This figure "test.png" is available in "png" format from: http://arxiv.org/ps/2410.21770v1

    1999 arXiv

Pith tools

Reviewed May 23, 2026 · model on record in the stance chip above.