Pith. sign in

REVIEW 1 major objections 5 minor 2 cited by

Linear-Scaling Tensor Train Sketching

T0 review · 1 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Pith's one-line read A structured tensor-train sketch embeds subspaces with cost linear in tensor order, not exponential.

desk verdict Linear-in-d OSE/OSI for a clean unification of Khatri-Rao and Gaussian TT sketches; proofs are complete and the numerics match the theory. read the letter →

arxiv 2603.11009 v3 pith:NPWA7MOT submitted 2026-03-11 math.NA cs.DScs.NA

classification math.NAcs.DScs.NA MSC 65F5515A6968W20
keywords tensortrainoblivioussubspaceembeddinginjectionrandomizedroundingTTStacksketchKhatri-Raoproductlow-rankapproximation
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

High-order tensors are routinely stored and manipulated in the tensor-train format, but randomized compression steps have been held back by sketches whose sample complexity grew exponentially with the number of modes. This paper introduces TTStack, a single two-parameter family of random projections that unifies the classical Khatri-Rao and Gaussian tensor-train sketches. By tuning the block rank R and the number of stacked copies P, TTStack is proved to be an oblivious subspace embedding when R scales linearly with the product of tensor order and subspace dimension, and an oblivious subspace injection under the milder requirement that R grows only linearly with the order. Both guarantees are free of the exponential-in-d factors that appear in earlier analyses. The same bounds immediately yield quasi-optimal error estimates for the randomized QB factorization and for the randomize-then-orthogonalize TT-rounding algorithm. Numerical tests on synthetic data, Hadamard products of quantized functions, and a lithium-hydride electronic-structure calculation confirm that modest parameters already deliver the predicted accuracy.

What carries the argument

TTStack itself: a block-structured random matrix whose rows are independent Gaussian tensor trains of rank R, scaled by 1/√P. The OSE proof proceeds via strong Johnson-Lindenstrauss moment bounds and a product lemma; the OSI proof uses a Gaussian comparison for the minimal eigenvalue that incorporates a subspace-entanglement constant C_Q(R).

What would settle it

Construct an explicit r-dimensional subspace of an order-d tensor space for which the empirical injectivity constant of TTStack with R proportional only to d falls below any fixed positive threshold as d grows, while keeping P of order ε^{-2}(r+log(r/δ)).

Watch

Extended reading notes

Core claim

The TTStack sketch, formed by vertically stacking P independent Gaussian tensor-train rows of block rank R, is simultaneously an (ε,δ,r)-oblivious subspace embedding when R = O(d(r+log 1/δ)) and P = O(ε^{-2}), and a (1-ε,δ,r)-oblivious subspace injection when R = O(d) and P = O(ε^{-2}(r+log(r/δ))). Both statements depend only linearly on tensor order d and subspace dimension r.

Load-bearing premise

The OSE argument rests on an imported product lemma for strong moment bounds that carries an unspecified universal constant and is restricted to relative error at most one, forcing the more expensive linear-in-r regime for R.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 5 minor

Summary. The paper introduces TTStack, a two-parameter family of structured sketches for tensors in the tensor-train (TT) format that interpolates between Khatri–Rao sketches (R=1) and Gaussian TT sketches (P=1). It proves that TTStack is an (ε,δ,r)-OSE when R=O(d(r+log(1/δ))) and P=O(ε^{-2}), and a (1-ε,δ,r)-OSI when R=O(d) and P=O(ε^{-2}(r+log(r/δ))), both scaling only linearly in tensor order d and subspace dimension r. These guarantees are obtained via Strong Johnson–Lindenstrauss moment arguments for the OSE regime and a Gaussian comparison argument (with an explicit subspace-entanglement measure C_Q(R)) for the milder OSI regime. Direct corollaries give quasi-optimal error bounds for randomized QB factorizations and for Randomize-then-Orthogonalize TT rounding; numerical experiments on synthetic tensors, QTT Hadamard products, and a sketched Rayleigh–Ritz eigensolver for LiH support the claims.

Significance. If the linear-in-d OSE/OSI statements hold, the work closes a long-standing gap between the empirical success of TT-adapted sketches and their theoretical sample complexity, which previously scaled exponentially in d for Khatri–Rao constructions. The unification of existing sketches under a single parameter family, the explicit moment calculations via partial traces, and the quasi-optimal rounding bounds are concrete advances for randomized numerical multilinear algebra. The complete, self-contained proofs of the moment bounds (Proposition 7.4 and Lemmas 7.3/B.2/B.5) and the transparent scoping of imported tools (Remark 6.9) are strengths that make the contribution usable by others. The Orthogonal TTStack variant, while only empirical, already shows practical gains that will motivate follow-up theory.

major comments (1)
  1. No load-bearing technical gaps were found that undermine Theorems 3.7 or 3.10. The OSE analysis correctly imports the Strong-JLM product lemma (Imported Lemma 6.7) and explicitly records the ε≤1 restriction that forces R=O(d log(1/δ)) (Remark 6.9). The OSI analysis supplies the necessary fourth-moment bounds for the TT product structure and applies Tropp’s Gaussian comparison theorem in a standard way. The quasi-optimal QB and Randomize-then-Orthogonalize corollaries follow by routine RSVD arguments with explicit constants. The only substantive limitation is that the Orthogonal TTStack sketch (Definition 3.4), which empirically outperforms the plain Gaussian version, is left without theory; this does not affect the proved claims but should be flagged as future work.
minor comments (5)
  1. Table 1 and the surrounding discussion in §4 would be clearer if the precise dependence of the universal constant L on the product lemma were stated (even if only as “L from Ahle et al.”).
  2. Figures 1–3 report medians and interquartile ranges over 100 trials; adding a brief note on how the empirical injectivity/dilation constants are estimated (e.g., via the extreme singular values of ΩQ) would aid reproducibility.
  3. In §5.3.3 the sketched Rayleigh–Ritz algorithm restarts every 10 iterations; a short remark on why this frequency was chosen (or a sensitivity plot) would strengthen the quantum-chemistry experiment.
  4. Notation for the Strong Kronecker product ▷◁ and the partial-trace operators is introduced carefully, but a short glossary or a pointer back to Definitions 2.3 and 7.1 at first use in §7 would help readers who skip the preliminaries.
  5. A few typographical inconsistencies appear (e.g., “Orhogonal” in §5.3.3, occasional missing spaces around math mode). A careful proof-reading pass is recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: OSE/OSI guarantees are derived from stated Gaussian moment structure and imported external comparison lemmas, not fitted to or defined by the numerical experiments.

full rationale

The paper’s central claims (Theorems 3.7 and 3.10) are probabilistic embedding guarantees for a newly defined structured sketch. The OSE path proceeds from an explicit Strong-JLM product lemma (Imported Lemma 6.7 from Ahle et al.) plus a stacking lemma proved in the appendix; the OSI path proceeds from fourth-moment calculations for the TT product (Proposition 7.4, Lemmas 7.3/B.2/B.5) fed into Tropp’s external Gaussian comparison theorem. Neither path fits free parameters to the synthetic, Hadamard, or quantum-chemistry experiments; those experiments only validate the proved regimes. Self-citations (e.g. Al Daas et al. on randomized TT rounding) supply algorithmic context, not load-bearing uniqueness or uniqueness-of-ansatz claims. Orthogonal TTStack is presented without theory and is therefore not used to underwrite the proved statements. No equation reduces by construction to a fitted input, and no prediction is forced by a self-citation chain. Score 0 is therefore appropriate.

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

The central claims rest on standard Gaussian concentration and imported structural lemmas from the oblivious-sketching literature, plus one recent comparison theorem for min-eigenvalues. Design parameters P,R are free choices of the algorithm, not data-fitted constants. No new physical entities are postulated; the only invented analytic object is the subspace entanglement measure C_Q(R), which is defined from partial traces of the target subspace and bounded explicitly.

free parameters (2)
  • Universal constant L in Strong-JLM product lemma
    Imported Lemma 6.7 asserts existence of L≥1 such that product of Strong (ε/(L√i),δ)-JLM maps remains Strong (ε,δ)-JLM; L is never evaluated numerically and enters all OSE rank lower bounds.
  • Universal constant c in RSVD rank conditions
    Corollaries 3.9 and 4.6 absorb absolute factors (e.g. 32(Le)^2) into a single c; the constant is proof-derived, not fitted, but its magnitude is left opaque for implementers.
assumptions (4)
  • standard math Gaussian matrices with iid N_F(0,1/R) entries satisfy the Strong (ε,δ)-JLM property when R≥8 max(e/ε,1)^2 log(1/δ) (Lemma 6.3).
    Standard Laurent–Massart / moment calculation for chi-squared; used as base case for all TT product arguments.
  • standard math Product of independent Strong-JLM maps with suitably rescaled accuracy remains Strong-JLM (Imported Lemma 6.7 / Ahle et al. 2020).
    Load-bearing for Gaussian TT OSE; carries unspecified L and ε≤1 restriction (Remark 6.9).
  • domain assumption Tropp’s Gaussian comparison theorem for λ_min of random PSD matrices (arXiv:2501.16578) applies to the TTStack moment model.
    Used in §7.3 to convert fourth-moment bounds into high-probability injectivity; correctness of the transfer is assumed rather than re-proved.
  • domain assumption Input tensors admit TT representations with moderate ranks so that sketch application cost O(dnPRχ(χ+R)) is meaningful.
    Standard TT modeling assumption; without it the sketch is still defined but loses practical interest.
invented entities (2)
  • TTStack sketch (and Orthogonal TTStack variant)
    purpose: Parameterized family of structured random projections that interpolates Khatri-Rao and Gaussian TT sketches.
    Core algorithmic object of the paper; defined by stacking P independent rank-R Gaussian TT row blocks. Independent evidence is the mathematical definition plus numerical tests, not an external physical prediction.
  • Subspace entanglement measure C_Q(R)
    purpose: Quantifies how Kronecker-structured vectors in a subspace inflate embedding error; appears in the OSI sample bound (Theorem 3.10 / Def. 7.9).
    Analytic device built from partial-trace norms of QQ*; bounded by (1+p_F/R)^d−1. No external falsifiable handle beyond the embedding theory itself.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Linear-Scaling Tensor Train Sketching." pith.science (2026). https://pith.science/paper/NPWA7MOT

@misc{pith2026260311009,
  author       = {Pith},
  title        = {Pith review of: Linear-Scaling Tensor Train Sketching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NPWA7MOT}},
  note         = {Machine review of arXiv:2603.11009}
}
abstract

We introduce the TTStack sketch, a structured random projection tailored to the tensor train (TT) format that unifies existing TT-adapted sketching operators. By varying two integer parameters $P$ and $R$, TTStack interpolates between the Khatri-Rao sketch ($R=1$) and the Gaussian TT sketch ($P=1$). We prove that TTStack satisfies an oblivious subspace embedding (OSE) property with parameters $R = \mathcal{O}(d(r+\log 1/\delta))$ and $P = \mathcal{O}(\varepsilon^{-2})$, and an oblivious subspace injection (OSI) property under the condition $R = \mathcal{O}(d)$ and $P = \mathcal{O}(\varepsilon^{-2}(r + \log r/\delta))$. Both guarantees depend only linearly on the tensor order $d$ and on the subspace dimension $r$, in contrast to prior constructions that suffer from exponential scaling in $d$. As direct consequences, we derive quasi-optimal error bounds for the QB factorization and randomized TT rounding. The theoretical results are supported by numerical experiments on synthetic tensors, Hadamard products, and a quantum chemistry application.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Oblivious Subspace Injection Is Not Enough for Relative Error

    math.NA 2026-04 unverdicted novelty 7.0 of 10

    OSI alone does not yield relative-error guarantees for sketching; counterexamples for least-squares and randomized SVD show that upper control on the optimal residual is required to recover near-relative error.

  2. Fast elementwise operations on tensor trains with alternating cross interpolation

    math.NA 2026-03 unverdicted novelty 7.0 of 10

    Alternating cross interpolation performs elementwise operations on tensor trains in O(χ³) time with error control, improving on the standard O(χ⁴) scaling when output ranks are controlled.

Reference graph

Works this paper leans on

50 extracted references · 7 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Ahle, Michael Kapralov, Jakob Bæk Tejs Knudsen, Rasmus Pagh, Ameya Vel- ingker, David P

    Thomas D. Ahle, Michael Kapralov, Jakob Bæk Tejs Knudsen, Rasmus Pagh, Ameya Vel- ingker, David P. Woodruff, and Amir Zandieh. Oblivious Sketching of High-Degree Poly- nomial Kernels.arXiv preprint arXiv:1909.01410, 2019

  2. [2]

    Ahle, Michael Kapralov, Jakob Bæk Tejs Knudsen, Rasmus Pagh, Ameya Vel- ingker, David P

    Thomas D. Ahle, Michael Kapralov, Jakob Bæk Tejs Knudsen, Rasmus Pagh, Ameya Vel- ingker, David P. Woodruff, and Amir Zandieh. Oblivious Sketching of High-Degree Poly- nomial Kernels. In Shuchi Chawla, editor,Proceedings of the 2020 ACM-SIAM Symposium on Discrete Algorithms, SODA 2020, Salt Lake City, UT, USA, January 5–8, 2020, pages 141–160. SIAM, 2020

  3. [3]

    The Fast Johnson–Lindenstrauss Transform and Approx- imate Nearest Neighbors.SIAM Journal on computing, 39(1):302–322, 2009

    Nir Ailon and Bernard Chazelle. The Fast Johnson–Lindenstrauss Transform and Approx- imate Nearest Neighbors.SIAM Journal on computing, 39(1):302–322, 2009

  4. [4]

    Reid, and Arvind K

    Hussam Al Daas, Grey Ballard, Paul Cazeaux, Eric Hallman, Agnieszka Miedlar, Mirjeta Pasha, Tim W. Reid, and Arvind K. Saibaba. Randomized Algorithms for Rounding in the Tensor-Train Format.SIAM Journal on Scientific Computing, 45(1):A74–A95, February 2023

  5. [5]

    Particle Number Conservation and Block Structures in Matrix Product States.Calcolo, 59(2):24, 2022

    Markus Bachmayr, Michael G¨ otte, and Max Pfeffer. Particle Number Conservation and Block Structures in Matrix Product States.Calcolo, 59(2):24, 2022

  6. [6]

    Subspace embed- ding with random Khatri–Rao products and its application to eigensolvers.IMA journal of numerical analysis, page draf043, 2025

    Zvonimir Bujanovi´ c, Luka Grubiˇ si´ c, Daniel Kressner, and Hei Yin Lam. Subspace embed- ding with random Khatri–Rao products and its application to eigensolvers.IMA journal of numerical analysis, page draf043, 2025

  7. [7]

    Epperly, Raphael A

    Chris Cama˜ no, Ethan N. Epperly, Raphael A. Meyer, and Joel A. Tropp. Faster Linear Algebra Algorithms with Structured Random Matrices.arXiv preprint arXiv:2508.21189, 2025

  8. [8]

    Successive randomized compres- sion: A randomized algorithm for the compressed MPO-MPS product.arXiv preprint arXiv:2504.06475, 2025

    Chris Cama˜ no, Ethan N Epperly, and Joel A Tropp. Successive randomized compres- sion: A randomized algorithm for the compressed MPO-MPS product.arXiv preprint arXiv:2504.06475, 2025

Show all 50 references
  1. [9]

    Efficient Randomized Algorithms for Computing an Approximation of the Tensor Train Decomposition.Journal of Scientific Computing, 107(1):2, February 2026

    Maolin Che, Yimin Wei, and Hong Yan. Efficient Randomized Algorithms for Computing an Approximation of the Tensor Train Decomposition.Journal of Scientific Computing, 107(1):2, February 2026

  2. [10]

    Optimal Oblivious Subspace Embeddings with Near-optimal Sparsity.arXiv preprint arXiv:2411.08773, 2025

    Shabarish Chenakkod, Micha l Derezi´ nski, and Xiaoyu Dong. Optimal Oblivious Subspace Embeddings with Near-optimal Sparsity.arXiv preprint arXiv:2411.08773, 2025

  3. [11]

    Optimal approximate matrix product in terms of stable rank

    Michael B Cohen, Jelani Nelson, and David P Woodruff. Optimal approximate matrix product in terms of stable rank. InInternational Colloquium on Automata, Languages, and Programming. Schloss Dagstuhl-Leibniz-Zentrum fur Informatik GmbH, Dagstuhl Publish- ing, 2016. 35

  4. [12]

    Adaptive Randomized Tensor Train Rounding using Khatri-Rao Products.arXiv preprint arXiv:2511.03598, 2025

    Hussam Al Daas, Grey Ballard, Laura Grigori, Mariana Martinez Aguilar, Arvind K Saibaba, and Bhisham Dev Verma. Adaptive Randomized Tensor Train Rounding using Khatri-Rao Products.arXiv preprint arXiv:2511.03598, 2025

  5. [13]

    Computation of extreme eigenvalues in higher dimensions using block tensor train format

    Sergey V Dolgov, Boris N Khoromskij, Ivan V Oseledets, and Dmitry V Savostyanov. Computation of extreme eigenvalues in higher dimensions using block tensor train format. Computer Physics Communications, 185(4):1207–1216, 2014

  6. [14]

    Learning tensor networks with tensor cross interpolation: New algorithms and libraries.SciPost Phys., 18:104, 2025

    Yuriel N´ u˜ nez Fern´ andez, Marc K Ritter, Matthieu Jeannin, Jheng-Wei Li, Thomas Kloss, Thibaud Louvet, Satoshi Terasaki, Olivier Parcollet, Jan von Delft, Hiroshi Shinaoka, and Xavier Waintal. Learning tensor networks with tensor cross interpolation: New algorithms and lib...

  7. [15]

    A Quantum Inspired Approach to Exploit Turbulence Structures.Nature Computational Science, 2(1):30–37, January 2022

    Nikita Gourianov, Michael Lubasch, Sergey Dolgov, Quincy Y van den Berg, Hessam Babaee, Peyman Givi, Martin Kiffner, and Dieter Jaksch. A Quantum Inspired Approach to Exploit Turbulence Structures.Nature Computational Science, 2(1):30–37, January 2022

  8. [16]

    Springer, Heidelberg, 2012

    Wolfgang Hackbusch.Tensor Spaces and Numerical Tensor Calculus, volume 42 ofSpringer Series in Computational Mathematics. Springer, Heidelberg, 2012

  9. [17]

    Halko, P

    N. Halko, P. G. Martinsson, and J. A. Tropp. Finding Structure with Randomness: Prob- abilistic Algorithms for Constructing Approximate Matrix Decompositions.SIAM Review, 53(2):217–288, 2011

  10. [18]

    Springer International Publishing, Cham, 2017

    Benjamin Huber, Reinhold Schneider, and Sebastian Wolf.A Randomized Tensor Train Singular Value Decomposition, page 261–290. Springer International Publishing, Cham, 2017

  11. [19]

    Almost Optimal Explicit Johnson- Lindenstrauss Families

    Daniel Kane, Raghu Meka, and Jelani Nelson. Almost Optimal Explicit Johnson- Lindenstrauss Families. In Leslie Ann Goldberg, Klaus Jansen, R. Ravi, and Jos´ e D. P. Rolim, editors,Approximation, Randomization, and Combinatorial Optimization. Algo- rithms and Techniques, pages ...

  12. [20]

    QTT-finite- element approximation for multiscale problems I: model problems in one dimension.Ad- vances in Computational Mathematics, 43(2):411–442, 2017

    Vladimir Kazeev, Ivan Oseledets, Maxim Rakhuba, and Christoph Schwab. QTT-finite- element approximation for multiscale problems I: model problems in one dimension.Ad- vances in Computational Mathematics, 43(2):411–442, 2017

  13. [21]

    Quantized tensor FEM for multiscale problems: diffusion problems in two and three dimensions.Mul- tiscale Modeling & Simulation, 20(3):893–935, 2022

    Vladimir Kazeev, Ivan Oseledets, Maxim V Rakhuba, and Christoph Schwab. Quantized tensor FEM for multiscale problems: diffusion problems in two and three dimensions.Mul- tiscale Modeling & Simulation, 20(3):893–935, 2022

  14. [22]

    Streaming Tensor Train Approxi- mation.SIAM Journal on Scientific Computing, 45(5):A2610–A2631, October 2023

    Daniel Kressner, Bart Vandereycken, and Rik Voorhaar. Streaming Tensor Train Approxi- mation.SIAM Journal on Scientific Computing, 45(5):A2610–A2631, October 2023

  15. [23]

    Estimation of moments of sums of independent real random variables.The Annals of Probability, 25(3):1502–1513, 1997

    Rafa l Lata la. Estimation of moments of sums of independent real random variables.The Annals of Probability, 25(3):1502–1513, 1997

  16. [24]

    Adaptive estimation of a quadratic functional by model selection.Annals of statistics, pages 1302–1338, 2000

    Beatrice Laurent and Pascal Massart. Adaptive estimation of a quadratic functional by model selection.Annals of statistics, pages 1302–1338, 2000

  17. [25]

    Multiscale interpolative construction of quantized tensor trains.arXiv preprint arXiv:2311.12554v3, 2023

    Michael Lindsey. Multiscale interpolative construction of quantized tensor trains.arXiv preprint arXiv:2311.12554v3, 2023

  18. [26]

    Cost-efficient Gaussian Tensor Network Embeddings for Tensor-structured Inputs.Advances in Neural Information Processing Systems, 35:38980– 38993, 2022

    Linjian Ma and Edgar Solomonik. Cost-efficient Gaussian Tensor Network Embeddings for Tensor-structured Inputs.Advances in Neural Information Processing Systems, 35:38980– 38993, 2022. 36

  19. [27]

    Per-Gunnar Martinsson and Joel A. Tropp. Randomized Numerical Linear Algebra: Foun- dations and Algorithms.Acta Numerica, 29:403–572, 2020

  20. [28]

    Miles Stoudenmire

    Zhaonan Meng, Yuehaw Khoo, Jiajia Li, and E. Miles Stoudenmire. Recursive Sketched Interpolation: Efficient Hadamard Products of Tensor Trains.arXiv preprint arXiv:2602.17974, 2026

  21. [29]

    H. D. Meyer, U. Manthe, and L. S. Cederbaum. The multi-configurational time-dependent Hartree approach.Chemical Physics Letters, 165(1):73–78, January 1990

  22. [30]

    Meyer and Haim Avron

    Raphael A. Meyer and Haim Avron. Hutchinson’s Estimator is Bad at Kronecker-Trace- Estimation.SIAM Journal on Matrix Analysis and Applications, 47(1):353–387, 2026

  23. [31]

    Understanding the Kronecker Matrix-Vector Complexity of Linear Algebra

    Raphael A Meyer, William Joseph Swartworth, and David Woodruff. Understanding the Kronecker Matrix-Vector Complexity of Linear Algebra. In Aarti Singh, Maryam Fazel, Daniel Hsu, Simon Lacoste-Julien, Felix Berkenkamp, Tegan Maharaj, Kiri Wagstaff, and Jerry Zhu, editors,Procee...

  24. [32]

    Mahoney, N

    Riley Murray, James Demmel, Michael W. Mahoney, N. Benjamin Erichson, Maksim Mel- nichenko, Osman Asif Malik, Laura Grigori, Piotr Luszczek, Michal Derezinski, Miles E. Lopes, Tianyu Liang, Hengrui Luo, and Jack J. Dongarra. Randomized Numerical Lin- ear Algebra : A Perspectiv...

  25. [33]

    Fast and stable randomized low-rank matrix approximation.arXiv preprint arXiv:2009.11392, 2020

    Yuji Nakatsukasa. Fast and stable randomized low-rank matrix approximation.arXiv preprint arXiv:2009.11392, 2020

  26. [34]

    Yuji Nakatsukasa and Joel A. Tropp. Fast and Accurate Randomized Algorithms for Linear Systems and Eigenvalue Problems.SIAM Journal on Matrix Analysis and Applications, 45(2):1183–1214, 2024

  27. [35]

    Jelani Nelson and Huy L. Nguyen. Lower Bounds for Oblivious Subspace Embeddings. In Javier Esparza, Pierre Fraigniaud, Thore Husfeldt, and Elias Koutsoupias, editors,Au- tomata, Languages, and Programming, pages 883–894, Berlin, Heidelberg, 2014. Springer Berlin Heidelberg

  28. [36]

    I. V. Oseledets. Approximation of 2 d ×2 d Matrices Using Tensor Decomposition.SIAM Journal on Matrix Analysis and Applications, 31(4):2130–2145, January 2010

  29. [37]

    I. V. Oseledets. Tensor-Train Decomposition.SIAM Journal on Scientific Computing, 33(5):2295–2317, 2011

  30. [38]

    I. V. Oseledets. Constructive Representation of Functions in Low-Rank Tensor Formats. Constructive Approximation, 37(1):1–18, February 2013

  31. [39]

    I. V. Oseledets and E. E. Tyrtyshnikov. Breaking the Curse of Dimensionality, Or How to Use SVD in Many Dimensions.SIAM Journal on Scientific Computing, 31(5):3744–3759, January 2009

  32. [40]

    TT-cross approximation for multidimensional arrays.Linear Algebra and its Applications, 432(1):70–88, 2010

    Ivan Oseledets and Eugene Tyrtyshnikov. TT-cross approximation for multidimensional arrays.Linear Algebra and its Applications, 432(1):70–88, 2010

  33. [41]

    Tensorized Random Projections

    Beheshteh Rakhshan and Guillaume Rabusseau. Tensorized Random Projections. InIn- ternational Conference on Artificial Intelligence and Statistics, pages 3306–3316. PMLR, 2020. 37

  34. [42]

    Ritter, Yuriel N´ u˜ nez Fern´ andez, Markus Wallerberger, Jan von Delft, Hiroshi Shinaoka, and Xavier Waintal

    Marc K. Ritter, Yuriel N´ u˜ nez Fern´ andez, Markus Wallerberger, Jan von Delft, Hiroshi Shinaoka, and Xavier Waintal. Quantics Tensor Cross Interpolation for High-Resolution Parsimonious Representations of Multivariate Functions.Phys. Rev. Lett., 132:056501, Jan 2024

  35. [43]

    Saibaba, Bhisham Dev Verma, and Grey Ballard

    Arvind K. Saibaba, Bhisham Dev Verma, and Grey Ballard. Improved Analysis of Khatri- Rao Random Projections and Applications.arXiv preprint arXiv:2507.23207, 2025

  36. [44]

    Improved Approximation Algorithms for Large Matrices via Random Pro- jections

    Tamas Sarlos. Improved Approximation Algorithms for Large Matrices via Random Pro- jections. In2006 47th Annual IEEE Symposium on Foundations of Computer Science (FOCS’06), pages 143–152, 2006

  37. [45]

    HaTT: Hadamard avoiding TT recompression.arXiv preprint arXiv:2410.04385, 2024

    Zhonghao Sun, Jizu Huang, Chuanfu Xiao, and Chao Yang. HaTT: Hadamard avoiding TT recompression.arXiv preprint arXiv:2410.04385, 2024

  38. [46]

    Tensor product methods and entanglement optimiza- tion for ab initio quantum chemistry.International Journal of Quantum Chemistry, 115(19):1342–1391, October 2015

    Szil´ ard Szalay, Max Pfeffer, Valentin Murg, Gergely Barcza, Frank Verstraete, Rein- hold Schneider, and Ors Legeza. Tensor product methods and entanglement optimiza- tion for ab initio quantum chemistry.International Journal of Quantum Chemistry, 115(19):1342–1391, October 2015

  39. [47]

    Joel A. Tropp. Comparison theorems for the minimum eigenvalue of a random positive- semidefinite matrix.arXiv preprint arXiv:2501.16578, 2025

  40. [48]

    Steven R. White. Density matrix formulation for quantum renormalization groups.Phys. Rev. Lett., 69(19):2863–2866, November 1992

  41. [49]

    Sketching as a Tool for Numerical Linear Algebra.Foundations and Trends in Theoretical Computer Science, 10(1–2):1–157, 2014

    David P Woodruff. Sketching as a Tool for Numerical Linear Algebra.Foundations and Trends in Theoretical Computer Science, 10(1–2):1–157, 2014

  42. [50]

    Quantized tensor networks for solving the Vlasov–Maxwell equations.Journal of Plasma Physics, 90(3):805900301, 2024

    Erika Ye and Nuno F Loureiro. Quantized tensor networks for solving the Vlasov–Maxwell equations.Journal of Plasma Physics, 90(3):805900301, 2024. A Gaussian Strong JLM Property Proof of Theorem 6.3 (Strong JLM for Gaussian matrices).Given a unit vectorx∈F n and a random matri...

Pith tools

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