Pith. sign in

REVIEW 4 major objections 4 minor 76 references

Matrices over a Hilbert space and their low-rank cross approximation

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The paper extends low-rank cross approximation to matrices whose entries lie in a Hilbert space, proving an interpolatory cross-decomposition identity and validating an adaptive variant on parametric PDE solution maps.

desk verdict A clean but overclaimed extension of cross approximation to Hilbert-space-valued matrices; the core theory is sound, but the abstract promises results the paper does not deliver and the analytic benchmark has a sign error. read the letter →

arxiv 2505.05134 v2 pith:LVO2MBCG submitted 2025-05-08 math.NA cs.NA

classification math.NAcs.NA MSC 15A2315B9965F55
keywords Bochnermatriceslow-rankapproximationcrossadaptiveparametricPDEsreduced-ordermodelingTuckerrankHOSVD
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that low-rank cross approximation, a workhorse of numerical linear algebra for ordinary matrices, still works when each entry of the matrix is itself an element of a Hilbert space, such as a function solving a parametric PDE. It proves that a cross component built from selected rows and columns interpolates the Bochner matrix on that index set and carries the same Tucker rank, and it proposes adaptive algorithms, ABCD and ABCDX, that use Hilbert-space norms to pick pivots. Numerically, these adaptive cross components approximate the solution maps of a linear model problem and of parametric nonlinear Stokes equations at errors within a small constant of the HOSVD baseline, while accessing only selected rows and columns of the matrix. The abstract also advertises maximum-volume bounds and approximation guarantees, but those results do not appear in the text; the body explicitly defers approximation guarantees to future work.

What carries the argument

The central object is the Bochner matrix, a finite array whose entries lie in a Hilbert space, together with the pseudoinverse acting on it. The workhorse identity is the cross component formula cross(A,I,J)=((G^T)^+ C^T)^T G (G^+ R), where G is the submatrix at the selected index set and C and R are the corresponding columns and rows. The associated concepts are the column and row ranks, the Tucker rank pair, and the SVD-based HOSVD projection that yields quasioptimal low-rank approximations. The formula's job is to express the whole matrix through a small submatrix G and two side factors built from selected columns and rows, so that only a small number of rows and columns must actually be evaluated.

What would settle it

Run ABCD or ABCDX with one rook-pivoting round on a family of Bochner matrices with known singular-value decays and compare each selected-index cross component with the HOSVD error at the same Tucker rank; if the error ratio grows with matrix size, or if randomly selected index sets achieve comparable errors, then the adaptive pivot heuristic is not doing the work attributed to it.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that the identity cross(A,I,J)=((G^T)^+ C^T)^T G (G^+ R), with C, R, G the selected columns, rows, and intersection, is the right Bochner analogue of the classical cross and pseudoskeleton decomposition: it interpolates A on I x J, its Tucker rank equals the Tucker rank of the selected submatrix G, and it reproduces A exactly when rank_T(G)=rank_T(A) (Lemmas 6.4 and 6.8). The paper further claims that the adaptive algorithms ABCD and ABCDX, which choose pivots by the largest Hilbert-space norm in the residual, select index sets whose cross-component error trails the quasioptimal HOSVD error by only a small factor in parametric PDE experiments.

Load-bearing premise

The load-bearing premise is that picking entries with the largest Hilbert-space norm during the search identifies near-optimal rows and columns; the paper states this expectation but does not prove it, explicitly deferring approximation guarantees to future work.

Editorial extensions

If this is right

  • A Bochner matrix can be reconstructed exactly from a cross component whenever the selected submatrix has the same Tucker rank as the whole matrix, so a few carefully chosen rows and columns can encode all information for rank-deficient solution maps.
  • The cross component always interpolates the matrix on the selected index set and has Tucker rank bounded by the selected submatrix, giving a low-rank surrogate whose error is controlled entirely by the index choice.
  • HOSVD-style projection onto dominant left and right singular subspaces yields a Tucker approximation within a factor of $\sqrt{2}$ of the best possible error in the $\ell^2(H)$ norm.
  • On the parametric nonlinear Stokes equations, ABCDX with rook pivoting reaches errors about twice the HOSVD error at the same rank, demonstrating that the method behaves as a non-intrusive reduced-order model.
  • The adaptive algorithms can be integrated with existing PDE software, meaning the surrogate is built from solution evaluations and inner products rather than from the discretized system matrices.

Reading between the lines

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

  • If the pivot heuristic were backed by a theorem, the same index-selection strategy would give a non-intrusive reduced-order modeling pipeline whose cost is measured in inner products between PDE solutions, independent of the discretization dimension.
  • The gap between the dyadic ABCD approximant and its post-processed cross component suggests decoupling selection from approximation: use cheap norm-based pivoting for exploration, then form a one-shot cross component as the final surrogate.
  • A natural testable extension, not pursued in the paper, would be to fold a higher-order Bochner tensor into a Bochner matrix and run ABCDX mode by mode, connecting to fiber-based tensor completion; the elementary rank and interpolation identities developed here suggest the same strategy would carry over.
  • If the deferred maximum-volume bounds materialize, they would likely be smaller than in the classical setting because the Hilbert-space dimension of the entries weakens the volume constraint, exactly as the abstract anticipates.
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

4 major / 4 minor

Summary. The paper develops an elementary theory of matrices whose entries lie in a Hilbert space (Bochner matrices), covering norms, ranks, adjoints, QR and SVD factorizations, Mirsky-type low-rank approximation, and HOSVD. It defines a cross component for Bochner matrices and proves exact interpolation on a selected row/column index set under a Tucker-rank condition (Lemmas 6.4 and 6.8). It then proposes adaptive algorithms ABCD and ABCDX, which select pivots by the Hilbert norm of residual entries, and validates them numerically on a parametric linear boundary-value problem and on parametric nonlinear Stokes equations. The abstract claims that approximation guarantees and maximum-volume bounds are derived, but the body explicitly defers those guarantees to future work.

Significance. The algebraic framework is natural, and the core lemmas in Sections 3-6 are mostly correct and self-contained. The exact cross interpolation formula for Bochner matrices and the explicit SVD/HOSVD extensions are useful contributions for parametric PDE surrogate modeling, and the paper contains no fitted parameters in its theoretical results. The authors are transparent about the heuristic status of ABCD in Section 6.3, and the comparison with HOSVD is a sensible evaluation protocol. However, the advertised approximation guarantees and maximum-volume bounds are absent from the manuscript, and the analytic benchmark in Section 7.1 appears inconsistent with the stated PDE. These issues currently prevent acceptance.

major comments (4)
  1. [Abstract; §1; §6.3] The abstract states that "approximation guarantees are derived" and that "the high-dimensional nature of the entries is shown to manifest itself in maximum-volume bounds," but neither statement is supported by the text. Section 1 explicitly says "Neither do we address the question of approximation guarantees for Bochner cross approximation," and Section 6.3 contains no theorem about the index sets chosen by ABCD, only an expectation that they are good. The only error bounds in the paper are Theorems 5.12 and 5.16, which concern SVD and HOSVD approximation, not cross approximation, and no maximum-volume definition or bound appears anywhere. This is a load-bearing mismatch between the paper's advertised central result and its actual content; the claims must be removed or the missing results supplied.
  2. [§7.1, Eq. (2)] The displayed solution formula does not solve the stated boundary-value problem (2). For -u'' - α²u = cos(βx), the sine coefficients are proportional to k[1 - (-1)^k cos(βπ)] / [(k² - α²)(k² - β²)], whereas the paper gives the same numerator over (k² + α²)(k² + β²). For example, at α = 0, β = 1, and k = 2, the true coefficient is nonzero and differs from the value produced by the displayed formula. Consequently, the L2 and H1 error curves in Figure 1 are computed with a function that is not the solution of (2), so the validation experiment does not test the stated problem. The PDE, the formula, or both must be corrected.
  3. [§5.2, Definition 5.7 and Theorem 5.8] Definition 5.7 is not well-typed as written: it defines a Bochner norm ||·||_• to be unitarily invariant if there is a usual-matrix norm ||·||_∘ with ||QA||_• = ||A||_∘ for every Q with orthonormal columns, but on the right A is a Bochner matrix while ||·||_∘ acts on usual matrices. In the proof of Theorem 5.8, ||UΣV*||_• is identified with ||ΣV*||_∘ and then with ||Σ||_∘ without specifying how a usual matrix is embedded into the Bochner norm. Since Theorem 5.12 and its use in Theorem 5.16 depend on this identification, the definition and proof need to be repaired.
  4. [§6.3, Algorithms 3-4; §7.1, Fig. 1; §7.3, Fig. 3/Table 1] The paper claims that ABCD/ABCDX select index sets whose approximation error is close to HOSVD, but this is not established. No theorem connects the argmax-over-||·||_H pivot rule in Algorithms 3-4 to near-optimal rows or columns; Section 6.3 defers such guarantees. The only direct comparison with random selection, in Figure 1, shows that random cross is "only marginally worse" than ABCD, and no random baseline is reported for the Stokes experiments in Figure 3 and Table 1. Please either provide a subset-selection or volume-type bound that supports the pivot rule, or add random baselines and weaken the quasioptimality claim to a heuristic observation.
minor comments (4)
  1. [§5.2, Definition 5.7] There is a typo in Definition 5.7: "unitraily" should read "unitarily."
  2. [§7] Section 7 states that a link to the Julia module repository will be added in an updated version; for reproducibility of the numerical claims, the link or a data-availability statement should be present in the submitted version.
  3. [§7.1] The notation in Figure 1 uses "H = L2" and "H = H1_0" for the entry space; since the paper consistently uses H for an abstract Hilbert space, a brief note clarifying the two usages would improve readability.
  4. [§1] The sentence "Neither do we address the question of approximation guarantees for Bochner cross approximation" directly contradicts the abstract; aligning these statements is essential even if the technical results remain unchanged.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the Bochner cross-component construction is proved from standard pseudoinverse and rank identities, with external HOSVD/analytic benchmarks; the abstract's unfulfilled guarantees are a missing-content concern, not circularity.

full rationale

The derivation chain is self-contained and non-circular. Definition 6.6 defines cross(A,I,J) = ((G^T)^+ C^T)^T G (G^+ R); Lemma 6.8 derives interpolation B(I,J)=A(I,J) and rankT(B)=rankT(A(I,J)) using the Moore-Penrose identities (Theorem 5.1) and submatrix rank bounds (Lemma 3.26), not by assuming the conclusion. The SVD and HOSVD results (Theorems 4.25, 5.12, 5.16) are proved from standard linear algebra (Mirsky's theorem, Courant-Fischer, Weyl inequalities) or standard references, and the column subset selection bound is cited from external work [43], not from the author's prior results. No parameter is fitted to a subset and then 'predicted' on a closely related quantity: the ABCD/ABCDX pivot rules are heuristic (Section 6.3 explicitly says approximation guarantees are future work), and the experiments compare against HOSVD and random-pivot cross components as external baselines. The paper's own text shows a mismatch between the Abstract ('its approximation guarantees are derived', 'maximum-volume bounds') and Section 1 ('Neither do we address the question of approximation guarantees for Bochner cross approximation... future work'), and no maximum-volume bound appears in the text; this is an unfulfilled-promise / correctness concern, not a circular reduction. The analytic benchmark formula in Section 7.1 also appears questionable (denominator k^2+beta^2 rather than the expected k^2-beta^2), which is a numerical-validation concern, not a circularity. No load-bearing self-citation or ansatz-by-citation occurs: the references are to established literature on CUR/pseudoskeleton decompositions and quasimatrices. Score 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The theory has no fitted free parameters; algorithm hyperparameters are experimental settings, not parameters tuned to make the derivation work. The paper introduces the formal object 'Bochner matrix', which is a definition rather than a postulated entity used to explain a phenomenon, so no independent falsifiable handle is claimed or needed.

assumptions (3)
  • standard math Standard results of finite-dimensional linear algebra and Hilbert space theory (rank-nullity, SVD existence, Mirsky's theorem, completeness) are valid for the objects used.
    Invoked throughout Sections 3-6; the paper cites [2,9,27,36,69] rather than proving these background results.
  • domain assumption The solution maps of the parametric PDEs in Section 7 are well-posed and the finite element discretizations used (Taylor-Hood, 6495 DOF) are sufficiently accurate to serve as ground truth.
    Underpins the numerical experiments; no convergence test or error estimate for the FEM discretization is reported.
  • ad hoc to paper The analytic formula for u(x;alpha,beta) in Section 7.1 is the exact solution of Eq (2).
    This formula generates the benchmark data in Figure 1; as written it appears inconsistent with the stated equation, since the Fourier denominators have the wrong signs for the given PDE.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Matrices over a Hilbert space and their low-rank cross approximation." pith.science (2026). https://pith.science/paper/LVO2MBCG

@misc{pith2026250505134,
  author       = {Pith},
  title        = {Pith review of: Matrices over a Hilbert space and their low-rank cross approximation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LVO2MBCG}},
  note         = {Machine review of arXiv:2505.05134}
}
read the original abstract

Motivated by applications in reduced-order modeling (ROM) of parametric partial differential equations, we investigate the algebraic properties of Bochner matrices -- matrices with entries in an abstract Hilbert space. Low-rank cross approximation is extended to Bochner matrices and its approximation guarantees are derived. The high-dimensional nature of the entries is shown to manifest itself in maximum-volume bounds, making them smaller than in the classical setting. An analogue of adaptive cross approximation is proposed and validated as a non-intrusive ROM method in numerical experiments with parametric nonlinear Stokes equations.

Figures

Figures reproduced from arXiv: 2505.05134 by the authors.

Figure 1
Figure 1. Low-rank approximation of the solution map of ( [PITH_FULL_IMAGE:figures/full_fig_p028_1.png] view at source ↗
Figure 2
Figure 2. Numerical finite-element solution to (3)-(4) obtained with FEniCS as in [63]. 5https://fenicsproject.org/ 29 [PITH_FULL_IMAGE:figures/full_fig_p029_2.png] view at source ↗
Figure 3
Figure 3. Low-rank approximation of the solution map of ( [PITH_FULL_IMAGE:figures/full_fig_p030_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

76 extracted references · 42 canonical work pages

  1. [1]

    Adaptive cross tubal tensor approximation.Linear Algebra Appl695, 168–90

    Ahmadi-Asl S, Phan AH, Cichocki A, Sozykina A, Al Aghbari Z, Wang J, Oseledets I. Adaptive cross tubal tensor approximation.Linear Algebra Appl695, 168–90. doi:10.1016/ j.laa.2024.04.015 (2024)

  2. [2]

    Axler S.Linear Algebra Done Right4th ed. doi:10 . 1007 / 978 - 3 - 031 - 41026 - 0(Springer, 2024)

  3. [3]

    Black box approximation of tensors in hierarchical Tucker format

    Ballani J, Grasedyck L, Kluge M. Black box approximation of tensors in hierarchical Tucker format. Linear Algebra Appl438, 639–57. doi:10.1016/j.laa.2011.08.010 (2013)

  4. [4]

    Ballard G, Kolda TG.Tensor Decompositions for Data Scienceisbn: 9781009471664 (CUP, 2025)

  5. [5]

    Block generalized locally Toeplitz sequences: theory and applications in the unidimensional case.Electron Trans Numer Anal53, 28–112

    Barbarino G, Garoni C, Serra-Capizzano S. Block generalized locally Toeplitz sequences: theory and applications in the unidimensional case.Electron Trans Numer Anal53, 28–112. doi:10.1553/etna_vol53s28 (2020)

  6. [6]

    Numer Math86,565–89.doi: 10

    BebendorfM.Approximationofboundaryelementmatrices. Numer Math86,565–89.doi: 10. 1007/PL00005410 (2000)

  7. [7]

    Adaptive low-rank approximation of collocation matrices.Com- puting 70, 1–24

    Bebendorf M, Rjasanow S. Adaptive low-rank approximation of collocation matrices.Com- puting 70, 1–24. doi:10.1007/s00607-002-1469-6 (2003)

  8. [8]

    Third-order tensors as linear operators on a space of matrices.Linear Algebra Appl 433, 1241–53

    Braman K. Third-order tensors as linear operators on a space of matrices.Linear Algebra Appl 433, 1241–53. doi:10.1016/j.laa.2010.05.025 (2010)

Show all 76 references
  1. [9]

    1007/978-0-387-70914-7 (Springer, 2011)

    Brezis H.Functional Analysis, Sobolev Spaces and Partial Differential Equationsdoi:10 . 1007/978-0-387-70914-7 (Springer, 2011)

  2. [10]

    Brown WC.Matrices over Commutative Ringsisbn: 9780824787554 (CRC Press, 1992)

  3. [11]

    Bunse-GerstnerA,ByersR,MehrmannV,NicholsNK.Numericalcomputationofananalytic singular value decomposition of a matrix valued function.Numer Math 60, 1–39. doi:10. 1007/bf01385712 (1991). 8We prove these results explicitly to make the text more accessible. 32

  4. [12]

    Mode-wise tensor decompositions: Multi-dimensional generalizations of CUR decompositions.J Mach Learn Res22, 1–36

    Cai H, Hamm K, Huang L, Needell D. Mode-wise tensor decompositions: Multi-dimensional generalizations of CUR decompositions.J Mach Learn Res22, 1–36. doi:10.48550/arXiv. 2103.11037 (2021)

  5. [13]

    Generalizing the column–row matrix decomposition to multi-way arrays

    Caiafa CF, Cichocki A. Generalizing the column–row matrix decomposition to multi-way arrays. Linear Algebra Appl433, 557–73. doi:10.1016/j.laa.2010.03.020 (2010)

  6. [14]

    Exact matrix completion via convex optimization.Found Comput Math 9, 717–72

    Candes E, Recht B. Exact matrix completion via convex optimization.Found Comput Math 9, 717–72. doi:10.1007/s10208-009-9045-5 (2009)

  7. [15]

    QR decomposition of Laurent polynomial matrices sampled on the unit circle.IEEE Trans Inf Theory56, 4754–61

    Cescato D, Bölcskei H. QR decomposition of Laurent polynomial matrices sampled on the unit circle.IEEE Trans Inf Theory56, 4754–61. doi:10.1109/TIT.2010.2054454 (2010)

  8. [16]

    On the compression of low rank matrices

    Cheng H, Gimbutas Z, Martinsson PG, Rokhlin V. On the compression of low rank matrices. SIAM J Sci Comput26, 1389–404. doi:10.1137/030602678 (2005)

  9. [17]

    Approximation of high-dimensional parametric PDEs.Acta Numer24, 1–159

    Cohen A, DeVore R. Approximation of high-dimensional parametric PDEs.Acta Numer24, 1–159. doi:10.1017/S0962492915000033 (2015)

  10. [18]

    On maximum volume submatrices and cross approxi- mation for symmetric semidefinite and diagonally dominant matrices.Linear Algebra Appl 593, 251–68

    Cortinovis A, Kressner D, Massei S. On maximum volume submatrices and cross approxi- mation for symmetric semidefinite and diagonally dominant matrices.Linear Algebra Appl 593, 251–68. doi:10.1016/j.laa.2020.02.010 (2020)

  11. [19]

    A multilinear singular value decomposition

    De Lathauwer L, De Moor B, Vandewalle J. A multilinear singular value decomposition. SIAM J Matrix Anal Appl21, 1253–78. doi:10.1137/s0895479896305696 (2000)

  12. [20]

    Hyperspectral image super-resolution via subspace-based low tensor multi- rank regularization.IEEE Trans Image Process28, 5135–46

    Dian R, Li S. Hyperspectral image super-resolution via subspace-based low tensor multi- rank regularization.IEEE Trans Image Process28, 5135–46. doi:10.1109/TIP.2019.2916734 (2019)

  13. [21]

    Ell TA, Le Bihan N, Sangwine SJ.Quaternion Fourier Transforms for Signal and Image Processing doi:10.1002/9781118930908 (Wiley, 2014)

  14. [22]

    On slow-fading MIMO systems with nonseparable correlation

    Far RR, Oraby T, Bryc W, Speicher R. On slow-fading MIMO systems with nonseparable correlation. IEEE Trans Inf Theory54, 544–53. doi:10.1109/TIT.2007.913239 (2008)

  15. [23]

    An algorithm for calculating the QR and singular value decompositions of polynomial matrices.IEEE Trans Signal Process58, 1263–74

    Foster JA, McWhirter JG, Davies MR, Chambers JA. An algorithm for calculating the QR and singular value decompositions of polynomial matrices.IEEE Trans Signal Process58, 1263–74. doi:10.1109/TSP.2009.2034325 (2009)

  16. [24]

    Nuclear norm of higher-order tensors.Math Comput 87, 1255–81

    Friedland S, Lim LH. Nuclear norm of higher-order tensors.Math Comput 87, 1255–81. doi:10.1090/mcom/3239 (2018)

  17. [25]

    Gillis N.Nonnegative Matrix Factorizationdoi:10.1137/1.9781611976410 (SIAM, 2020)

  18. [26]

    Matrix Polynomials doi:10

    Gohberg I, Lancaster P, Rodman L. Matrix Polynomials doi:10 . 1137 / 1 . 9780898719024 (SIAM, 2009)

  19. [27]

    Golub GH, Van Loan CF.Matrix Computations 4th ed.isbn: 9781421407944 (JHUP, 2013)

  20. [28]

    A theory of pseudoskeleton approxima- tions

    Goreinov SA, Tyrtyshnikov EE, Zamarashkin NL. A theory of pseudoskeleton approxima- tions. Linear Algebra Appl261, 1–21. doi:10.1016/S0024-3795(96)00301-1 (1997)

  21. [29]

    Pseudo-skeleton approximations of matri- ces

    Goreinov SA, Zamarashkin NL, Tyrtyshnikov EE. Pseudo-skeleton approximations of matri- ces. Dokl Akad Nauk343, 151–2 (1995)

  22. [30]

    Computing 74,205–23.doi: 10

    GrasedyckL.Adaptiverecompressionof H-matricesforBEM. Computing 74,205–23.doi: 10. 1007/s00607-004-0103-1 (2005)

  23. [32]

    doi:10.1007/978- 3- 030-35554-8 (Springer, 2010)

    Hackbusch W.Tensor Spaces and Numerical Tensor Calculus2nd ed. doi:10.1007/978- 3- 030-35554-8 (Springer, 2010)

  24. [33]

    Perspectives on CUR decompositions.Appl Comput Harmon Anal48, 1088–99

    Hamm K, Huang L. Perspectives on CUR decompositions.Appl Comput Harmon Anal48, 1088–99. doi:10.1016/j.acha.2019.08.006 (2020)

  25. [34]

    Perturbations of CUR decompositions.SIAM J Matrix Anal Appl42, 351–75

    Hamm K, Huang L. Perturbations of CUR decompositions.SIAM J Matrix Anal Appl42, 351–75. doi:10.1137/19M128394X (2021)

  26. [35]

    Guaranteed functional tensor singular value decomposition.J Am Stat Assoc119, 995–1007

    Han R, Shi P, Zhang AR. Guaranteed functional tensor singular value decomposition.J Am Stat Assoc119, 995–1007. doi:10.1080/01621459.2022.2153689 (2024)

  27. [36]

    Horn RA, Johnson CR.Topics in Matrix Analysisisbn: 9780521467131 (CUP, 1994)

  28. [37]

    hal: hal-00747253

    Icart S, Comon P.Some properties of Laurent polynomial matricesin 9th IMA International Conference on Mathematics in Signal Processing(2012). hal: hal-00747253

  29. [38]

    Robust quaternion matrix completion with applications to image inpainting

    Jia Z, Ng MK, Song GJ. Robust quaternion matrix completion with applications to image inpainting. Numer Linear Algebra Appl26, e2245. doi:10.1002/nla.2245 (2019)

  30. [39]

    Tensor-tensor products with invertible linear transforms

    Kernfeld E, Kilmer M, Aeron S. Tensor-tensor products with invertible linear transforms. Linear Algebra Appl485, 545–70. doi:10.1016/j.laa.2015.07.021 (2015)

  31. [40]

    Third-order tensors as operators on matrices: A theoretical and computational framework with applications in imaging.SIAM J Matrix Anal Appl34, 148–72

    Kilmer ME, Braman K, Hao N, Hoover RC. Third-order tensors as operators on matrices: A theoretical and computational framework with applications in imaging.SIAM J Matrix Anal Appl34, 148–72. doi:10.1137/110837711 (2013)

  32. [41]

    Factorization strategies for third-order tensors.Linear Algebra Appl 435, 641–58

    Kilmer ME, Martin CD. Factorization strategies for third-order tensors.Linear Algebra Appl 435, 641–58. doi:10.1016/j.laa.2010.09.020 (2011)

  33. [42]

    Factorization of a 768-bit RSA modulusin CRYPTO 2010 (2010), 333–50

    Kleinjung T, Aoki K, Franke J, Lenstra AK, Thomé E, Bos JW, Gaudry P, Kruppa A, Montgomery PL, Osvik DA, et al. Factorization of a 768-bit RSA modulusin CRYPTO 2010 (2010), 333–50. doi:10.1007/978-3-642-14623-7_18

  34. [43]

    On the approximation of vector-valued functions by volume sampling

    Kressner D, Ni T, Uschmajew A. On the approximation of vector-valued functions by volume sampling. J Complex 86, 101887. doi:10.1016/j.jco.2024.101887 (2025)

  35. [44]

    doi:10.48550/arXiv.2408.05677

    Larsen BW, Kolda TG, Zhang AR, Williams AH.Tensor decomposition meets RKHS: Effi- cient algorithms for smooth and misaligned data2024. doi:10.48550/arXiv.2408.05677

  36. [45]

    Singular value decomposition of quaternion matrices: a new tool for vector-sensor signal processing.Signal Process84, 1177–99

    Le Bihan N, Mars J. Singular value decomposition of quaternion matrices: a new tool for vector-sensor signal processing.Signal Process84, 1177–99. doi:10.1016/j.sigpro.2004.04. 001 (2004)

  37. [46]

    doi:10.1109/ICIP.2003.1247085

    Le Bihan N, Sangwine SJ.Quaternion principal component analysis of color imagesin ICIP (2003). doi:10.1109/ICIP.2003.1247085

  38. [47]

    CUR matrix decompositions for improved data analysis.Proc Natl Acad Sci106, 697–702

    Mahoney MW, Drineas P. CUR matrix decompositions for improved data analysis.Proc Natl Acad Sci106, 697–702. doi:10.1073/pnas.0803205106 (2009)

  39. [48]

    Interpolatory tensorial reduced order models for parametric dynamical systems.Comput Methods Appl Mech Eng397, 115122

    Mamonov AV, Olshanskii MA. Interpolatory tensorial reduced order models for parametric dynamical systems.Comput Methods Appl Mech Eng397, 115122. doi:10.1016/j.cma.2022. 115122 (2022)

  40. [49]

    doi:10.48550/arXiv.2411.07151

    Mamonov AV, Olshanskii MA.Slice sampling tensor completion for model order reduction of parametric dynamical systems2024. doi:10.48550/arXiv.2411.07151

  41. [50]

    Tensorial parametric model order reduction of nonlinear dy- namical systems.SIAM J Sci Comput46, A1850–A1878

    Mamonov AV, Olshanskii MA. Tensorial parametric model order reduction of nonlinear dy- namical systems.SIAM J Sci Comput46, A1850–A1878. doi:10.1137/23M1553789 (2024)

  42. [51]

    A priori analysis of a tensor ROM for parameter dependent parabolic problems.SIAM J Numer Anal63, 239–61

    Mamonov AV, Olshanskii MA. A priori analysis of a tensor ROM for parameter dependent parabolic problems.SIAM J Numer Anal63, 239–61. doi:10.1137/23M1616844 (2025). 34

  43. [52]

    McDonald BR.Linear Algebra over Commutative Ringsdoi:10.1201/9781003065197 (CRC Press, 1984)

  44. [53]

    Quaternions in signal and image processing: A comprehensive and objective overview.IEEE Signal Process Mag40, 26–40

    Miron S, Flamant J, Le Bihan N, Chainais P, Brie D. Quaternions in signal and image processing: A comprehensive and objective overview.IEEE Signal Process Mag40, 26–40. doi:10.1109/MSP.2023.3278071 (2023)

  45. [54]

    Quaternion-MUSIC for vector-sensor array processing.IEEE Trans Signal Process54, 1218–29

    Miron S, Le Bihan N, Mars JI. Quaternion-MUSIC for vector-sensor array processing.IEEE Trans Signal Process54, 1218–29. doi:10.1109/TSP.2006.870630 (2006)

  46. [55]

    Symmetric gauge functions and unitarily invariant norms.Q J Math 11, 50–9

    Mirsky L. Symmetric gauge functions and unitarily invariant norms.Q J Math 11, 50–9. doi:10.1093/qmath/11.1.50 (1960)

  47. [56]

    doi:10.48550/ arXiv.2504.16231

    Mor U, Avron H.Quasitubal tensor algebra over separable Hilbert spaces2025. doi:10.48550/ arXiv.2504.16231

  48. [57]

    Polyno- mial eigenvalue decomposition for multichannel broadband signal processing: a mathemat- ical technique offering new insights and solutions

    Neo VW, Redif S, McWhirter JG, Pestana J, Proudler IK, Weiss S, Naylor PA. Polyno- mial eigenvalue decomposition for multichannel broadband signal processing: a mathemat- ical technique offering new insights and solutions. IEEE Signal Process Mag 40, 18–37. doi:10.1109/MSP.202...

  49. [58]

    TT-cross approximation for multidimensional arrays.Linear Algebra Appl432, 70–88

    Oseledets I, Tyrtyshnikov E. TT-cross approximation for multidimensional arrays.Linear Algebra Appl432, 70–88. doi:10.1016/j.laa.2009.07.024 (2010)

  50. [59]

    Tucker dimensionality reduction of three- dimensional arrays in linear time

    Oseledets IV, Savostianov D, Tyrtyshnikov EE. Tucker dimensionality reduction of three- dimensional arrays in linear time. SIAM J Matrix Anal Appl 30, 939–56. doi:10 . 1137 / 060655894 (2008)

  51. [60]

    Pseudo-skeleton approximations with better accuracy esti- mates

    Osinsky A, Zamarashkin NL. Pseudo-skeleton approximations with better accuracy esti- mates. Linear Algebra Appl537, 221–49. doi:10.1016/j.laa.2017.09.032 (2018)

  52. [61]

    Relevance of polynomial matrix decompositions to broad- band blind signal separation.Signal Process134, 76–86

    Redif S, Weiss S, McWhirter JG. Relevance of polynomial matrix decompositions to broad- band blind signal separation.Signal Process134, 76–86. doi:10.1016/j.sigpro.2016.11.019 (2017)

  53. [62]

    Matrix valued adaptive cross approximation.Math Methods Appl Sci 40, 2522–31

    Rjasanow S, Weggler L. Matrix valued adaptive cross approximation.Math Methods Appl Sci 40, 2522–31. doi:10.1002/mma.4174 (2017)

  54. [63]

    Lecture 3: Static nonlinear PDEs.https://pub

    Rognes ME, Wells GN.FEniCS Course. Lecture 3: Static nonlinear PDEs.https://pub. fenicsproject.org/course/lectures/2017-nordic-phdcourse (visited on 05/05/2025)

  55. [64]

    Communication over finite-field matrix channels.IEEE Trans Inf Theory56, 1296–305

    Silva D, Kschischang FR, Kotter R. Communication over finite-field matrix channels.IEEE Trans Inf Theory56, 1296–305. doi:10.1109/TIT.2009.2039167 (2010)

  56. [65]

    doi:10.23919/EUSIPCO63174.2024.10715267

    Sofi SS, Hendrikx S, De Lathauwer L.Tensor train completion of multi-way data observed along one modein EUSIPCO (2024), 1067–71. doi:10.23919/EUSIPCO63174.2024.10715267

  57. [66]

    Multilinear singular value decomposition–based completion with fibers observed in a single mode.SIAM J Matrix Anal Appl46, 1061–90

    Sørensen M, Hendrikx S, De Lathauwer L. Multilinear singular value decomposition–based completion with fibers observed in a single mode.SIAM J Matrix Anal Appl46, 1061–90. doi:10.1137/23M1622830 (2025)

  58. [67]

    Continuous analogues of matrix factorizations.Proc R Soc A: Math Phys Eng Sci471, 20140585

    Townsend A, Trefethen LN. Continuous analogues of matrix factorizations.Proc R Soc A: Math Phys Eng Sci471, 20140585. doi:10.1098/rspa.2014.0585 (2015)

  59. [68]

    Householder triangularization of a quasimatrix.IMA J Numer Anal30, 887–

    Trefethen LN. Householder triangularization of a quasimatrix.IMA J Numer Anal30, 887–

  60. [69]

    Trefethen LN, Bau D.Numerical Linear Algebradoi:10.1137/1.9780898719574 (SIAM, 1997)

  61. [70]

    Tretter C.Spectral Theory of Block Operator Matrices and Applicationsdoi:10.1142/p493 (World Scientific, 2008). 35

  62. [71]

    Incomplete cross approximation in the mosaic-skeleton method.Computing 64, 367–80

    Tyrtyshnikov E. Incomplete cross approximation in the mosaic-skeleton method.Computing 64, 367–80. doi:10.1007/s006070070031 (2000)

  63. [72]

    Efficient algorithms for CUR and interpolative matrix decompo- sitions

    Voronin S, Martinsson PG. Efficient algorithms for CUR and interpolative matrix decompo- sitions. Adv Comput Math43, 495–516. doi:10.1007/s10444-016-9494-8 (2017)

  64. [73]

    Properties and structure of the analytic singular value decomposition.IEEE Trans Signal Process72, 2260–75

    Weiss S, Proudler IK, Barbarino G, Pestana J, McWhirter JG. Properties and structure of the analytic singular value decomposition.IEEE Trans Signal Process72, 2260–75. doi:10. 1109/TSP.2024.3387726 (2024)

  65. [74]

    Quaternions and matrices of quaternions.Linear Algebra Appl251, 21–57

    Zhang F. Quaternions and matrices of quaternions.Linear Algebra Appl251, 21–57. doi:10. 1016/0024-3795(95)00543-9 (1997)

  66. [75]

    Exact tensor completion using t-SVD.IEEE Trans Signal Process65, 1511–26

    Zhang Z, Aeron S. Exact tensor completion using t-SVD.IEEE Trans Signal Process65, 1511–26. doi:10.1109/TSP.2016.2639466 (2016)

  67. [76]

    doi:10.1007/978-3-031- 49432-1_9

    Zheltkov D, Zamarashkin N, Matveev S.How to make Lanczos-Montgomery fast on modern supercomputers? in Russian Supercomputing Days(2023), 114–28. doi:10.1007/978-3-031- 49432-1_9. 36

  68. [97]

    doi:10.1093/imanum/drp018 (2010)

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.