Pith. sign in

REVIEW 1 major objections 2 minor 14 references

Iterative methods compute the principal square root of third-order tensors under the T-product with quadratic or geometric convergence.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.3

2026-06-30 20:30 UTC pith:AC3VVOU5

load-bearing objection The paper gives tensor versions of Newton and Denman-Beavers iterations for square roots under the T-product plus a new Bures-Wasserstein distance, but the convergence transfer from matrices needs explicit slice-wise spectrum checks. the 1 major comments →

arxiv 2605.14748 v1 pith:AC3VVOU5 submitted 2026-05-14 math.NA cs.NA

Iterative Methods for Computing the T-Square Root of Third-Order Tensors

Hemant Sharma , Nachiketa Mishra This is my paper
classification math.NA cs.NA
keywords tensor square rootT-productNewton iterationDenman-Beavers iterationBures-Wasserstein distanceimage processingthird-order tensors
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper develops and analyzes tensor extensions of the Newton iteration and the Denman-Beavers iteration for finding the principal square root of third-order tensors in the T-product framework. Convergence guarantees are obtained by reducing tensor problems to independent matrix square-root problems via Fourier-domain block-diagonalization of the T-product. The methods are demonstrated on image-processing tasks including Tensor Decorrelated Grayscale conversion and T-Whitening, while a Tensor Bures-Wasserstein distance is introduced and shown to be a metric on the space of T-positive definite tensors.

Core claim

Tensor versions of the Newton iteration achieve quadratic convergence and the Denman-Beavers iteration achieves geometric convergence while simultaneously producing the inverse square root; both results follow because the T-product becomes block-diagonal matrix multiplication in the Fourier domain, so matrix convergence theorems apply directly to each frequency slice without further restrictions.

What carries the argument

Fourier-domain block-diagonalization of the T-product, which converts every tensor operation into a collection of independent matrix operations, one per frequency component.

Load-bearing premise

The Fourier-domain block-diagonalization transfers matrix convergence properties to the tensor square-root problem without needing extra conditions on the tensor spectrum or structure.

What would settle it

A single third-order tensor for which the proposed Newton or Denman-Beavers iteration diverges or produces a non-square-root result, or for which the proposed Bures-Wasserstein distance violates the metric axioms.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Newton iteration on tensors converges quadratically once the initial guess satisfies the same conditions as in the matrix case.
  • Denman-Beavers iteration simultaneously yields both the tensor square root and its inverse with geometric convergence.
  • The Tensor Bures-Wasserstein distance satisfies all metric axioms on the cone of T-positive definite tensors.
  • Tensor-based whitening and color transfer preserve cross-channel structure better than channel-wise matrix methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same block-diagonalization argument could be used to lift other matrix iterations, such as higher-order root finders, to the tensor setting.
  • The metric property opens the door to defining geodesics and means on the manifold of T-positive definite tensors for optimization problems.
  • Numerical tests on tensors whose Fourier slices have eigenvalues near the negative real axis would directly probe the transfer of convergence.

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

1 major / 2 minor

Summary. The paper develops and analyzes tensor extensions of the Newton iteration (claimed quadratic convergence) and Denman-Beavers iteration (claimed geometric convergence with simultaneous inverse square root) for the principal square root of third-order tensors under the T-product. Convergence is established by reducing to independent matrix iterations on the Fourier blocks of the T-product. Applications to image processing are introduced, including Tensor Decorrelated Grayscale conversion, T-Whitening, and optimal color transfer. The Tensor Bures-Wasserstein distance is defined and proved to be a metric on T-positive definite tensors. Numerical experiments are presented to illustrate convergence and performance advantages over classical methods.

Significance. If the central claims hold, the work supplies practical, provably convergent iterative algorithms for tensor square roots in the T-product algebra, together with a new Riemannian metric on the cone of T-positive definite tensors. The image-processing applications demonstrate concrete utility for multichannel data where cross-channel structure must be preserved. The Fourier-block reduction is a standard and clean technique when the slice-wise hypotheses are verified; the paper's numerical confirmation of rapid convergence is a positive feature.

major comments (1)
  1. [§4] §4 (Convergence Analysis, around the statements following the Fourier block-diagonalization): The manuscript asserts 'rigorous convergence guarantees' for both iterations by transferring the matrix Newton and Denman-Beavers results to each Fourier slice. However, the matrix principal square root is unique and the iterations converge (quadratically or geometrically) only when every matrix block has spectrum in the open right half-plane. The tensor-level 'T-positive definite' condition is not shown to imply this slice-wise spectral restriction; if any frequency block violates it, the corresponding matrix iteration may converge to a non-principal branch or diverge. This assumption gap is load-bearing for the central claim of rigorous tensor-level guarantees.
minor comments (2)
  1. [§6] The definition of the Tensor Bures-Wasserstein distance (presumably in §6) should include an explicit verification that the distance is independent of the choice of square-root branch when multiple branches exist.
  2. [§2-3] Notation for the T-product and its Fourier representation is introduced without a consolidated table of symbols; readers must hunt across §2 and §3 for the precise meaning of the block-diagonal operator.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed review and for identifying this important point in the convergence analysis. We address the concern below and will revise the manuscript accordingly to make the implication explicit.

read point-by-point responses

  1. Referee: [§4] §4 (Convergence Analysis, around the statements following the Fourier block-diagonalization): The manuscript asserts 'rigorous convergence guarantees' for both iterations by transferring the matrix Newton and Denman-Beavers results to each Fourier slice. However, the matrix principal square root is unique and the iterations converge (quadratically or geometrically) only when every matrix block has spectrum in the open right half-plane. The tensor-level 'T-positive definite' condition is not shown to imply this slice-wise spectral restriction; if any frequency block violates it, the corresponding matrix iteration may converge to a non-principal branch or diverge. This assumption gap is load-bearing for the central claim of rigorous tensor-level guarantees.

    Authors: We agree that the manuscript should explicitly connect the tensor-level T-positive definite condition to the required spectral properties of the Fourier blocks. In the T-product literature, a third-order tensor is defined to be T-positive definite precisely when every Fourier slice is a positive definite Hermitian matrix; this immediately places the spectrum of each block in the open right half-plane, ensuring uniqueness of the principal square root and the applicability of the matrix convergence theorems. We will add a short lemma (with proof) immediately after the definition of T-positive definiteness that states this equivalence and verifies the spectral condition. With this addition the transfer of the matrix results becomes fully rigorous at the tensor level. We view this as a clarification rather than a change in the underlying mathematics. revision: yes

Circularity Check

0 steps flagged

No circularity: standard Fourier reduction to matrix case is independent

full rationale

The paper proposes Newton and Denman-Beavers iterations for the T-product square root and claims convergence via Fourier block-diagonalization. This is a direct transfer of known matrix convergence results to each frequency slice; the reduction is not self-referential or fitted. No equations redefine a quantity in terms of its output, no parameters are fitted then relabeled as predictions, and no load-bearing uniqueness theorem is imported solely via self-citation. The derivation remains self-contained against external matrix theory benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The work relies on the established T-product framework and its Fourier properties from prior literature; the new distance is introduced as an extension.

axioms (1)
  • domain assumption The T-product admits Fourier-domain block-diagonalization that preserves the algebraic structure needed for principal square roots
    Invoked to establish convergence guarantees for the proposed iterations
invented entities (1)
  • Tensor Bures-Wasserstein distance no independent evidence
    purpose: To define a valid metric on the space of T-positive definite tensors
    Newly formulated within the paper with a proof of metric properties

pith-pipeline@v0.9.1-grok · 5661 in / 1169 out tokens · 39202 ms · 2026-06-30T20:30:40.471710+00:00 · methodology

0 comments
read the original abstract

We develop and analyze iterative methods for computing the principal square root of third-order tensors under the T-product framework. Tensor extensions of the Newton iteration (quadratic convergence) and the Denman--Beavers iteration (geometric convergence with simultaneous computation of the inverse square root) are proposed, with rigorous convergence guarantees established via the Fourier-domain block-diagonalization of the T-product. We apply these methods to image processing, introducing Tensor Decorrelated Grayscale conversion, T-Whitening, and optimal color transfer under the T-product geometry. We also formulate the Tensor Bures--Wasserstein distance and prove it defines a valid metric on the space of T-positive definite tensors. Numerical experiments confirm rapid convergence and demonstrate that the proposed tensor-based techniques offer improved structural preservation and cross-channel decorrelation compared to classical methods.

Figures

Figures reproduced from arXiv: 2605.14748 by Hemant Sharma, Nachiketa Mishra.

Figure 1
Figure 1. Figure 1: Convergence comparison for the tensor from Example 2.4. The Newton (blue) and DB (red) [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Numerical stability comparison for the highly ill-conditioned tensor of Example 2.7 ( [PITH_FULL_IMAGE:figures/full_fig_p017_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of grayscale conversion methods. From left to right: original color image, classical [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of T-Whitening on inter-channel correlations (illustrated on synthetic multivariate Gaus [PITH_FULL_IMAGE:figures/full_fig_p026_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Covariance matrices before and after T-Whitening. [PITH_FULL_IMAGE:figures/full_fig_p027_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Color transfer comparison (illustrated on synthetic images for clarity). From left to right: [PITH_FULL_IMAGE:figures/full_fig_p028_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Computational performance of tensor T-square root methods. [PITH_FULL_IMAGE:figures/full_fig_p030_7.png] view at source ↗

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

14 extracted references · 14 canonical work pages · 1 internal anchor

  1. [1]

    C. D. Martin, R. Shafer, and B. Larue, An order-ptensor factorization with applica- tions in imaging,SIAM Journal on Scientific Computing35 (1) (2013) A474–A490

    work page 2013

  2. [2]

    E. D. Denman and A. N. Beavers Jr., The matrix sign function and computations in systems,Applied Mathematics and Computation2 (1) (1976) 63–94

    work page 1976

  3. [3]

    Reinhard, M

    E. Reinhard, M. Adhikhmin, B. Gooch, and P. Shirley, Color transfer between images,IEEE Computer Graphics and Applications21 (5) (2001) 34–41

    work page 2001

  4. [4]

    Stable Tensor Neural Networks for Rapid Deep Learning

    E. Newman, L. Horesh, H. Avron, and M. Kilmer, Stable tensor neural networks for rapid deep learning, arXiv preprint arXiv:1811.06569 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  5. [5]

    Nocedal and S

    J. Nocedal and S. J. Wright,Numerical Optimization, 2nd ed., Springer, New York, 2006

    work page 2006

  6. [6]

    Lund, The tensor t-function: A definition for functions of third-order tensors, Numerical Linear Algebra with Applications27 (3) (2020) e2288

    K. Lund, The tensor t-function: A definition for functions of third-order tensors, Numerical Linear Algebra with Applications27 (3) (2020) e2288

    work page 2020

  7. [7]

    Lund and M

    K. Lund and M. Schweitzer, The Fréchet derivative of the tensor t-function,Calcolo 60 (3) (2023) 35

    work page 2023

  8. [8]

    M. E. Kilmer and C. D. Martin, Factorization strategies for third-order tensors, Linear Algebra and its Applications435 (3) (2011) 641–658

    work page 2011

  9. [9]

    M. E. Kilmer, K. Braman, N. Hao, and R. C. Hoover, Third-order tensors as oper- ators on matrices: A theoretical and computational framework with applications in imaging,SIAM Journal on Matrix Analysis and Applications34 (1) (2013) 148–172. 33

    work page 2013

  10. [10]

    N. Hao, M. E. Kilmer, K. Braman, and R. C. Hoover, Facial recognition using tensor-tensor decompositions,SIAM Journal on Imaging Sciences6 (1) (2013) 437– 463

    work page 2013

  11. [11]

    N. J. Higham,Functions of Matrices: Theory and Computation, Society for Indus- trial and Applied Mathematics, Philadelphia, 2008

    work page 2008

  12. [12]

    Laasonen, On the iterative solution of the matrix equationAX2 −I= 0,Mathe- matical Tables and Other Aids to Computation12 (61) (1958) 109–116

    P. Laasonen, On the iterative solution of the matrix equationAX2 −I= 0,Mathe- matical Tables and Other Aids to Computation12 (61) (1958) 109–116

    work page 1958

  13. [13]

    Bhatia, T

    R. Bhatia, T. Jain, and Y. Lim, On the Bures–Wasserstein distance between positive definite matrices,Expositiones Mathematicae37 (2) (2019) 165–191

    work page 2019

  14. [14]

    Reinhard, M

    E. Reinhard, M. Ashikhmin, B. Gooch, and P. Shirley, Color transfer between images,IEEE Computer Graphics and Applications21 (5) (2001) 34–41. 34

    work page 2001