pith. sign in

arxiv: 2604.26280 · v1 · submitted 2026-04-29 · 🧮 math.NA · cs.NA· math.DS

Structure-Aware Tensorial Model Reduction

Arjun Vijaywargia , Eric C. Cyr , Anthony Gruber This is my paper

Pith reviewed 2026-05-07 13:05 UTC · model grok-4.3

classification 🧮 math.NA cs.NAmath.DS
keywords reduced-order modelsTucker factorizationparameterized PDEsradial basis function interpolationGalerkin projectionnonlinear basisKolmogorov n-widthtensorial model reduction
0
0 comments X

The pith

Tensor factorization lets reduced bases for parameterized PDEs vary nonlinearly with parameters by encoding snapshots offline and interpolating online.

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

This paper develops a two-stage projection-based reduced-order modeling technique for parameterized partial differential equations. Snapshots of solutions are first compressed offline using multi-linear Tucker factorization. In the online phase, radial basis functions interpolate the encoded states to assemble a reduced basis that changes with the PDE parameters, after which the basis is orthonormalized with respect to a general discrete inner product for use in Galerkin projection. Error estimates support the approach, and tests on cases including three-dimensional Maxwell's equations show gains over linear-basis and prior tensorial methods when nonlinearity is strong or training data is limited. A reader would care because the method targets the practical regimes where standard reduced models lose accuracy or require excessive samples.

Core claim

The central claim is that Tucker factorization of solution snapshots permits the rapid online construction of a parameter-dependent reduced basis via radial basis function interpolation of the encoded states. When this basis is orthonormalized against a general discrete inner product, the resulting Galerkin reduced-order model satisfies representation and projection error bounds even under sparse parameter sampling. Numerical examples demonstrate that the resulting nonlinear-basis ROM improves accuracy relative to monolithic linear bases and earlier tensorial techniques precisely in the highly nonlinear and data-limited regimes typical of realistic PDE applications.

What carries the argument

Multi-linear Tucker factorization of solution snapshots, which encodes states so that radial basis function interpolation can rapidly produce a nonlinearly parameter-varying orthonormal reduced basis for Galerkin projection.

Load-bearing premise

The Tucker-encoded states admit accurate radial basis function interpolation and the orthonormalization step preserves error bounds sufficiently when parameter samples are sparse.

What would settle it

If the observed ROM error on a nonlinear PDE with sparse parameter samples exceeds the derived bounds or fails to improve on linear-basis performance, the validity of the interpolation and orthonormalization extensions would be refuted.

Figures

Figures reproduced from arXiv: 2604.26280 by Anthony Gruber, Arjun Vijaywargia, Eric C. Cyr.

Figure 1
Figure 1. Figure 1: Normalized singular value decay for the three bases view at source ↗
Figure 2
Figure 2. Figure 2: Relative M-weighted error in the ROM solutions to the heat system as a function of reduced basis dimension r, considered over all training (left) and testing (right) parameters. For each method, solid lines denote the median ROM error across parameter instances while dashed lines denote the median projection error. Shaded bands indicate the interquartile range of the ROM error. solution. Moreover, while bo… view at source ↗
Figure 3
Figure 3. Figure 3: Top row: FOM and ROM solutions to the heat system at terminal time view at source ↗
Figure 4
Figure 4. Figure 4: Relative MW -weighted error in the ROM solutions to the wave system as a function of reduced basis dimension r, considered over all training (left) and testing (right) parameters. Top row: displacement error; bottom row: momentum error. For each method, solid lines denote the median ROM error across parameter instances while dashed lines denote the median projection error. Shaded bands indicate the interqu… view at source ↗
Figure 5
Figure 5. Figure 5: Top row: FOM and ROM displacement solutions to the wave system at terminal time view at source ↗
Figure 6
Figure 6. Figure 6: Relative ME-weighted L 2 error in the ROM solutions to the Maxwell system as a function of reduced basis dimension r, considered over all training (left) and testing (right) parameters. Top row: electric field error; bottom row: magnetic field error. For each method, solid lines denote the median ROM error and dashed lines denote the median projection error. Shaded bands indicate the interquartile range of… view at source ↗
Figure 7
Figure 7. Figure 7: Magnitude and components of Eh at τ = 2.5 on the slice z = 0.5 obtained from the FOM and ROM solutions to the Maxwell system for a sample test parameter with ROM dimension r = 10. The RBF and MO ROMs yield small errors of 0.88% and 4.2%, while the monolithic ROM produces an extremely large error of 36.9%. 29 view at source ↗
Figure 8
Figure 8. Figure 8: Magnitude and components of Bh at τ = 2.5 on the slice z = 0.5 obtained from the FOM and ROM solutions to the Maxwell system for the same test parameter as view at source ↗
read the original abstract

This work investigates a two-stage method for constructing projection-based reduced-order models (ROMs) of parameterized partial differential equations (PDEs). Based on established tensorial ROM methodology, the proposed approach reduces dimensionality offline by encoding solution snapshots using a multi-linear Tucker factorization, so that a reduced basis which varies nonlinearly with PDE parameters can be rapidly constructed online and used in a Galerkin ROM. Two novel extensions of this strategy, tailored to the cases of structured PDEs and sparse parameter sampling, are presented: the construction of reduced bases orthonormalized with respect to a general discrete inner product, and the interpolation of encoded states via radial basis functions. Basic representation and ROM error estimates are presented demonstrating the validity of these modifications, and the approach is challenged on examples where monolithic-basis ROMs are known to struggle, including a realistic instance of Maxwell's equations in 3D. Results suggest that the proposed nonlinear basis ROM can effectively mitigate linear restrictions on Kolmogorov $n$-width while improving upon previous tensorial ROM technology, particularly in the highly nonlinear and data-limited regimes characteristic of practical use cases.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces a two-stage projection-based reduced-order modeling approach for parameterized PDEs. Offline, solution snapshots are encoded via multi-linear Tucker factorization to reduce dimensionality; online, a parameter-dependent reduced basis is constructed rapidly via radial basis function (RBF) interpolation of the encoded states and used in a Galerkin ROM. Two extensions are proposed: orthonormalization of the reduced basis with respect to a general discrete inner product (for structured PDEs) and RBF interpolation (for sparse parameter sampling). Basic representation and ROM error estimates are derived to support these modifications, and the method is demonstrated on examples including a 3D Maxwell's equations instance, claiming that the nonlinear basis mitigates Kolmogorov n-width limitations and outperforms prior tensorial ROMs in highly nonlinear, data-limited regimes.

Significance. If the error estimates are shown to be complete and the numerical results are robust, the work would represent a useful incremental advance in tensorial ROM methodology, extending it to practical settings with sparse data and strong nonlinearity where linear bases are known to converge slowly.

major comments (2)
  1. [error estimates section] The basic ROM error estimates (described in the abstract and the section presenting them) do not include an explicit additive term bounding the RBF interpolation residual of the Tucker-encoded states. Since the online phase replaces exact encoded states with RBF interpolation, and the target regime is sparse parameter sampling, this omission is load-bearing for the validity claims in data-limited nonlinear cases.
  2. [numerical results] Table or figure presenting the 3D Maxwell results: no quantitative discussion is provided on the sensitivity of the reported improvements to post-hoc choices such as Tucker ranks or RBF shape parameters, which directly affects the assessment of whether gains are due to the structure-aware reduction or to interpolation artifacts.
minor comments (1)
  1. [abstract] The abstract and introduction would benefit from a short statement of the specific quantitative error metrics used to verify the representation and ROM estimates.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive feedback on the manuscript. We address each major comment point by point below, indicating where revisions will be made to strengthen the presentation.

read point-by-point responses

  1. Referee: [error estimates section] The basic ROM error estimates (described in the abstract and the section presenting them) do not include an explicit additive term bounding the RBF interpolation residual of the Tucker-encoded states. Since the online phase replaces exact encoded states with RBF interpolation, and the target regime is sparse parameter sampling, this omission is load-bearing for the validity claims in data-limited nonlinear cases.

    Authors: We agree that the current basic error estimates focus on the Tucker factorization and Galerkin projection steps without an explicit term for the RBF interpolation residual. This is a valid observation, particularly given the emphasis on sparse parameter sampling. We will revise the error estimates section to incorporate an additive bound on the RBF interpolation residual of the encoded states, thereby supporting the validity claims more completely in the data-limited regime. revision: yes

  2. Referee: [numerical results] Table or figure presenting the 3D Maxwell results: no quantitative discussion is provided on the sensitivity of the reported improvements to post-hoc choices such as Tucker ranks or RBF shape parameters, which directly affects the assessment of whether gains are due to the structure-aware reduction or to interpolation artifacts.

    Authors: The referee is correct that the 3D Maxwell results section lacks explicit quantitative sensitivity analysis with respect to Tucker ranks and RBF shape parameters. While the selected values were determined through standard cross-validation procedures to balance accuracy and efficiency, we acknowledge that additional discussion would help rule out interpolation artifacts. In the revised manuscript we will add a concise paragraph summarizing the sensitivity of the reported errors to moderate variations in these hyperparameters, drawing on the existing computational data. revision: yes

Circularity Check

0 steps flagged

No circularity in derivation chain

full rationale

The paper extends established tensorial ROM methodology via Tucker encoding of snapshots, RBF interpolation of encoded states, and orthonormalization w.r.t. a general inner product. It derives basic representation and ROM error estimates to support the modifications and validates on Maxwell and other examples. No load-bearing step reduces by construction to a fitted quantity, self-definition, or self-citation chain; the error estimates are presented as independent demonstrations of validity rather than tautological restatements of inputs. The central claim of mitigating Kolmogorov n-width restrictions therefore rests on the new estimates and numerical results, not on circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on the assumption that solution manifolds of the target PDEs admit low-rank Tucker structure and that RBF interpolation in the encoded space is sufficiently smooth. No new physical entities are postulated.

axioms (2)
  • domain assumption The solution snapshots admit a low-rank multi-linear Tucker factorization whose factor matrices can be interpolated accurately by radial basis functions.
    Invoked in the description of the offline encoding and online basis construction stages.
  • domain assumption Error estimates derived for the modified orthonormalization and interpolation steps remain valid under the sparse sampling regime.
    Stated as 'basic representation and ROM error estimates are presented'.

pith-pipeline@v0.9.0 · 5490 in / 1606 out tokens · 44811 ms · 2026-05-07T13:05:39.559896+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

44 extracted references · 44 canonical work pages

  1. [1]

    B. M. Afkham, A. Bhatt, B. Haasdonk, and J. S. Hesthaven. Symplectic model-reduction with a weighted inner product, 2018

    work page 2018

  2. [2]

    S. E. Azam and S. Mariani. Investigation of computational and accuracy issues in pod-based reduced order modeling of dynamic structural systems.Engineering structures, 54:150–167, 2013

    work page 2013

  3. [3]

    Barnett and C

    J. Barnett and C. Farhat. Quadratic approximation manifold for mitigating the Kolmogorov barrier in nonlinear projection-based model order reduction.Journal of Computational Physics, 464:111348, 2022

    work page 2022

  4. [4]

    Benner, S

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

    work page 2015

  5. [5]

    Benner, E

    P. Benner, E. Sachs, and S. Volkwein. Model order reduction for pde constrained optimization.Trends in PDE constrained optimization, pages 303–326, 2014. 25 Structure-Aware Tensorial Model Reduction Vijaywargiya, Cyr, and Gruber

    work page 2014

  6. [6]

    Berman and B

    J. Berman and B. Peherstorfer. Randomized sparse neural galerkin schemes for solving evolution equa- tions with deep networks. InThirty-seventh Conference on Neural Information Processing Systems, 2023

    work page 2023

  7. [7]

    M. T. Bettencourt, D. A. Brown, K. L. Cartwright, E. C. Cyr, C. A. Glusa, P. T. Lin, S. G. Moore, D. A. McGregor, R. P. Pawlowski, E. G. Phillips, et al. Empire-pic: a performance portable unstructured particle-in-cell code.Communications in Computational Physics, 30(4), 2021

    work page 2021

  8. [8]

    M. A. Cardoso, L. J. Durlofsky, and P. Sarma. Development and application of reduced-order modeling procedures for subsurface flow simulation.International journal for numerical methods in engineering, 77(9):1322–1350, 2009

    work page 2009

  9. [9]

    Carlberg, M

    K. Carlberg, M. Barone, and H. Antil. Galerkin v. least-squares petrov–galerkin projection in nonlinear model reduction.Journal of Computational Physics, 330:693–734, 2017

    work page 2017

  10. [10]

    Chaturantabut and D

    S. Chaturantabut and D. C. Sorensen. A state space error estimate for pod-deim nonlinear model reduction.SIAM Journal on Numerical Analysis, 50(1):46–63, 2012

    work page 2012

  11. [11]

    D. N. Daescu and I. M. Navon. A dual-weighted approach to order reduction in 4DVAR data assimila- tion.Monthly Weather Review, 136(3):1026–1041, 2008

    work page 2008

  12. [12]

    De Lathauwer, B

    L. De Lathauwer, B. De Moor, and J. Vandewalle. A multilinear singular value decomposition.SIAM Journal on Matrix Analysis and Applications, 21(4):1253–1278, 2000

    work page 2000

  13. [13]

    A. N. Diaz, J. T. Needels, I. K. Tezaur, and P. J. Blonigan. Kernel manifolds: Nonlinear-augmentation dimensionality reduction using reproducing kernel hilbert spaces.International Journal for Numerical Methods in Engineering, 126(24):e70230, 2025

    work page 2025

  14. [14]

    Drmac and S

    Z. Drmac and S. Gugercin. A new selection operator for the discrete empirical interpolation method— improved a priori error bound and extensions.SIAM Journal on Scientific Computing, 38(2):A631–A648, 2016

    work page 2016

  15. [15]

    S. D. Fisher and C. A. Micchelli. The n-width of sets of analytic functions.Duke Mathematical Journal, 47(4):789 – 801, 1980

    work page 1980

  16. [16]

    H. Florez. Applications of model-order reduction to thermo-poroelasticity. InARMA US Rock Mechan- ics/Geomechanics Symposium, pages ARMA–2017. ARMA, 2017

    work page 2017

  17. [17]

    Geelen, S

    R. Geelen, S. Wright, and K. Willcox. Operator inference for non-intrusive model reduction with quadratic manifolds.Computer Methods in Applied Mechanics and Engineering, 403:115717, 2023

    work page 2023

  18. [18]

    Y. Geng, J. Singh, L. Ju, B. Kramer, and Z. Wang. Gradient preserving operator inference: Data-driven reduced-order models for equations with gradient structure.Computer Methods in Applied Mechanics and Engineering, 427:117033, 2024

    work page 2024

  19. [19]

    Ghattas and K

    O. Ghattas and K. Willcox. Learning physics-based models from data: Perspectives from inverse problems and model reduction.Acta Numerica, 30:445–554, 2021

    work page 2021

  20. [20]

    Y. Gong, Q. Wang, and Z. Wang. Structure-preserving Galerkin POD reduced-order modeling of Hamiltonian systems.Computer Methods in Applied Mechanics and Engineering, 315:780–798, 2017

    work page 2017

  21. [21]

    Greif and K

    C. Greif and K. Urban. Decay of the kolmogorov n-width for wave problems.Applied Mathematics Letters, 96:216–222, 2019

    work page 2019

  22. [22]

    Gruber, M

    A. Gruber, M. Gunzburger, L. Ju, and Z. Wang. Energetically consistent model reduction for metriplec- tic systems.Computer Methods in Applied Mechanics and Engineering, 404:115709, 2023

    work page 2023

  23. [23]

    Gruber and I

    A. Gruber and I. Tezaur. Canonical and noncanonical Hamiltonian operator inference.Computer Methods in Applied Mechanics and Engineering, 416:116334, 2023. 26 Structure-Aware Tensorial Model Reduction Vijaywargiya, Cyr, and Gruber

    work page 2023

  24. [24]

    Gruber and I

    A. Gruber and I. Tezaur. Variationally consistent Hamiltonian model reduction.SIAM Journal on Applied Dynamical Systems, 24(1):376–414, 2025

    work page 2025

  25. [25]

    J. S. Hesthaven, C. Pagliantini, and G. Rozza. Reduced basis methods for time-dependent problems. Acta Numerica, 31:265–345, 2022

    work page 2022

  26. [26]

    O. Knill. The dirac operator of a graph, 2013

    work page 2013

  27. [27]

    T. G. Kolda and B. W. Bader. Tensor decompositions and applications.SIAM Rev., 51:455–500, 2009

    work page 2009

  28. [28]

    A. N. Kolmogorov. ¨Uber die beste ann¨ aherung von elementen einer menge an eine andere.Annals of Mathematics, 37(1):107–110, 1936

    work page 1936

  29. [29]

    Le Clainche, F

    S. Le Clainche, F. Varas, and J. M. Vega. Accelerating oil reservoir simulations using pod on the fly. International Journal for Numerical Methods in Engineering, 110(1):79–100, 2017

    work page 2017

  30. [30]

    Lee and K

    K. Lee and K. T. Carlberg. Model reduction of dynamical systems on nonlinear manifolds using deep convolutional autoencoders.Journal of Computational Physics, 404:108973, 2020

    work page 2020

  31. [31]

    T. Lieu, C. Farhat, and M. Lesoinne. Pod-based aeroelastic analysis of a complete f-16 configuration: Rom adaptation and demonstration. In46th AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics and materials conference, page 2295, 2005

    work page 2005

  32. [32]

    A. V. Mamonov and M. A. Olshanskii. Interpolatory tensorial reduced order models for parametric dynamical systems.Computer Methods in Applied Mechanics and Engineering, 397:115122, 2022

    work page 2022

  33. [33]

    A. V. Mamonov and M. A. Olshanskii. Tensorial parametric model order reduction of nonlinear dy- namical systems.SIAM Journal on Scientific Computing, 46(3):A1850–A1878, 2024

    work page 2024

  34. [34]

    A. V. Mamonov and M. A. Olshanskii. A priori analysis of a tensor rom for parameter dependent parabolic problems.SIAM Journal on Numerical Analysis, 63(1):239–261, 2025

    work page 2025

  35. [35]

    Parish, M

    E. Parish, M. Yano, I. Tezaur, and T. Iliescu. Residual-based stabilized reduced-order models of the transient convection–diffusion–reaction equation obtained through discrete and continuous projection: E. parish et al.Archives of Computational Methods in Engineering, 32(3):1885–1929, 2025

    work page 1929

  36. [36]

    E. J. Parish and F. Rizzi. On the impact of dimensionally-consistent and physics-based inner products for pod-galerkin and least-squares model reduction of compressible flows.Journal of Computational Physics, 491:112387, 2023

    work page 2023

  37. [37]

    Peng and K

    L. Peng and K. Mohseni. Symplectic model reduction of Hamiltonian systems.SIAM Journal on Scientific Computing, 38(1):A1–a27, 2016

    work page 2016

  38. [38]

    Peters, J

    N. Peters, J. A. Ekaterinaris, and A. M. Wissink. Mode based reduced order model for a moving store. InAIAA Scitech 2021 Forum, page 0364, 2021

    work page 2021

  39. [39]

    Quarteroni, G

    A. Quarteroni, G. Rozza, et al.Reduced order methods for modeling and computational reduction, volume 9. Springer, 2014

    work page 2014

  40. [40]

    Schwerdtner and B

    P. Schwerdtner and B. Peherstorfer. Greedy construction of quadratic manifolds for nonlinear dimen- sionality reduction and nonlinear model reduction. arXiv preprint, 2024

    work page 2024

  41. [41]

    M. A. S´ anchez, B. Cockburn, N.-C. Nguyen, and J. Peraire. Symplectic Hamiltonian finite element meth- ods for linear elastodynamics.Computer Methods in Applied Mechanics and Engineering, 381:113843, 2021

    work page 2021

  42. [42]

    Tang, W.-Z

    Y.-Q. Tang, W.-Z. Fang, C.-Y. Zheng, and W.-Q. Tao. Applications of pod-based reduced order model to the rapid prediction of velocity and temperature in data centers.Applied Thermal Engineering, 263:125310, 2025. 27 Structure-Aware Tensorial Model Reduction Vijaywargiya, Cyr, and Gruber

    work page 2025

  43. [43]

    Unger and S

    B. Unger and S. Gugercin. Kolmogorov n-widths for linear dynamical systems.Advances in Computa- tional Mathematics, 45(5):2273–2286, 2019

    work page 2019

  44. [44]

    Vijaywargiya, S

    A. Vijaywargiya, S. A. McQuarrie, and A. Gruber. Tensor parametric hamiltonian operator inference. arXiv preprint arXiv:2502.10888, 2025. 28 Structure-Aware Tensorial Model Reduction Vijaywargiya, Cyr, and Gruber (a)FOM (b)RBF (c)Monolithic (d)MO Figure 7: Magnitude and components ofE h atτ= 2.5 on the slicez= 0.5 obtained from the FOM and ROM solutions t...

    work page arXiv 2025