pith. sign in

arxiv: 2606.21137 · v1 · pith:DHGW2MN3new · submitted 2026-06-19 · 🧮 math.PR · math.OA

A Sudakov--Fernique proof of Lehner-type edge bounds for matrix-valued GUE sums

Benoit Collins , Yuta Yamagishi This is my paper

Pith reviewed 2026-06-26 13:38 UTC · model grok-4.3

classification 🧮 math.PR math.OA
keywords random matricesGUEoperator normfree probabilityspectral edgesSudakov-FerniqueLehner formula
0
0 comments X

The pith

Positive semidefinite matrix coefficients yield finite-dimensional bounds on the expected spectral edges of H_M that sit within an additive 9 sqrt(nN/M) term of Lehner's free edges.

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

The paper establishes explicit upper and lower bounds on the expected largest and smallest eigenvalues of the random matrix H_M formed by a fixed Hermitian matrix A_0 plus a sum of positive semidefinite coefficients tensored with independent GUE matrices. These bounds are expressed directly in terms of the variational quantities ρ+ and ρ- that Lehner showed describe the edges of the corresponding free semicircular operator. The argument proceeds by applying a matrix-coefficient version of the Sudakov-Fernique comparison together with concentration and a dual variational characterization of the edge quantities over density matrices. When the dimension M grows faster than N and the coefficients remain uniformly bounded, the finite-dimensional expectation converges to the free edge, recovering the infinite-dimensional limit. The same argument produces a matching bound on the operator norm of H_M.

Core claim

Assuming A_i ≽ 0 for i ≥ 1 and M ≥ N, the paper proves E λ_max(H_M) ≤ ρ+ + 9 sqrt(nN/M) ||∑ A_i²||_op^{1/2} and E λ_min(H_M) ≥ ρ- - 9 sqrt(nN/M) ||∑ A_i²||_op^{1/2}, where ρ+ and ρ- are Lehner's variational expressions; the resulting bound on E ||H_M||_op follows immediately, and the same estimates imply that E ||H_M||_op approaches the free edge ρ whenever the coefficients are uniformly bounded, n is bounded, and N = o(M).

What carries the argument

Sudakov-Fernique comparison for matrix-valued coefficients combined with minimax duality of Lehner's edge formulas over density matrices.

If this is right

  • When coefficients are uniformly bounded, n is fixed, and N = o(M), limsup E ||H_M||_op ≤ ρ.
  • The operator-norm bound E ||H_M||_op ≤ ρ_* + 9 sqrt(nN/M) ||∑ A_i²||_op^{1/2} holds with ρ_* = max{ρ+, -ρ-}.
  • The same comparison yields matching lower and upper controls on both edges simultaneously.

Where Pith is reading between the lines

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

  • The positivity assumption is essential for the comparison to close; signed coefficients would require a different majorization or a separate argument.
  • The sqrt(nN/M) error term suggests a concrete rate at which finite matrix models approximate free-probability edges when the number of summands is moderate.
  • The method may extend to other ensembles whose singular-value laws satisfy Davidson-Szarek-type tail bounds.

Load-bearing premise

The coefficients A_i for i ≥ 1 must be positive semidefinite.

What would settle it

Compute E λ_max(H_M) for small fixed N and M with one negative A_i and check whether the deviation from ρ+ exceeds the claimed multiple of sqrt(nN/M) ||∑ A_i²||_op^{1/2}.

read the original abstract

Let $A_0,A_1,\ldots,A_n\in M_N(\mathbb{C})$ be Hermitian matrices and let $G_1,\ldots,G_n$ be independent $M\times M$ GUE matrices normalized so that $\|M^{-1/2}G_i\|\to 2$ almost surely as $M\to\infty$. We study the spectral edges and operator norm of $H_M = A_0\otimes I_M + \frac{1}{\sqrt{M}}\sum_{i=1}^n A_i\otimes G_i$. Lehner's formula identifies the right and left edges of the corresponding free semicircular operator as $\rho_+ = \inf_{Z\succ 0}\lambda_{\max}(A_0+Z+\sum_{i=1}^n A_iZ^{-1}A_i)$ and $\rho_- = \sup_{Z\prec 0}\lambda_{\min}(A_0+Z+\sum_{i=1}^n A_iZ^{-1}A_i)$. Assuming $A_i\succeq 0$ for $i\ge 1$ and $M\ge N$, we prove via concentration and minimax duality the finite-dimensional bounds $\mathbb{E}\lambda_{\max}(H_M)\le \rho_+ + 9\sqrt{nN/M}\,\|\sum_{i=1}^n A_i^2\|_{\mathrm{op}}^{1/2}$ and $\mathbb{E}\lambda_{\min}(H_M)\ge \rho_- - 9\sqrt{nN/M}\,\|\sum_{i=1}^n A_i^2\|_{\mathrm{op}}^{1/2}$. With $\rho_* = \max\{\rho_+,-\rho_-\}$, this yields $\mathbb{E}\|H_M\|_{\mathrm{op}}\le \rho_* + 9\sqrt{nN/M}\,\|\sum_{i=1}^n A_i^2\|_{\mathrm{op}}^{1/2}$. For uniformly bounded positive coefficients, bounded $n$, and $N=o(M)$, one obtains $\limsup_{M\to\infty}\mathbb{E}\|H_M\|_{\mathrm{op}}\le\rho$ whenever $\rho_{*,M}\to\rho$. The proof is a matrix-coefficient extension of classical Sudakov--Fernique comparison, combined with a Davidson--Szarek-type singular-value estimate and dual variational formulas for Lehner's edge quantities over density matrices. We also explain why this approach does not extend sharply to signed Hermitian coefficients.

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

0 major / 3 minor

Summary. The manuscript claims to prove finite-dimensional bounds on the expected spectral edges of the random matrix H_M = A_0 ⊗ I_M + M^{-1/2} ∑_{i=1}^n A_i ⊗ G_i (with G_i independent normalized GUE matrices) under the hypotheses A_i ≽ 0 for i ≥ 1 and M ≥ N. Specifically, E[λ_max(H_M)] ≤ ρ_+ + 9 sqrt(nN/M) ||∑ A_i²||_op^{1/2} and E[λ_min(H_M)] ≥ ρ_- - 9 sqrt(nN/M) ||∑ A_i²||_op^{1/2}, where ρ± are Lehner's variational edge formulas; this yields a corresponding bound on E[||H_M||_op]. The argument combines a matrix-coefficient Sudakov-Fernique comparison, Davidson-Szarek-type singular-value tails, and dual variational characterizations of ρ± over density matrices. The bounds recover the free semicircular edges in the large-M limit (with n and the A_i fixed) and the paper explains why the method fails to extend sharply to signed Hermitian coefficients.

Significance. If the claimed bounds hold, the work supplies explicit non-asymptotic error terms that quantify how well finite-dimensional matrix-valued GUE sums approximate the spectral edges of the associated free semicircular operator. This is useful for applications requiring quantitative control rather than purely asymptotic statements. The matrix-coefficient extension of Sudakov-Fernique comparison, together with the use of minimax duality over density matrices, constitutes a technical contribution that aligns with existing free-probability techniques while remaining grounded in classical concentration inequalities.

minor comments (3)
  1. [Introduction] §1 (Introduction): the normalization ||M^{-1/2} G_i|| → 2 a.s. is stated in the abstract but should be recalled explicitly when the model is first defined, to avoid any ambiguity about the scaling of the GUE matrices.
  2. [Abstract / §3] The constant 9 appearing in the error term is presented as the outcome of the concentration and comparison arguments; a short parenthetical remark tracing its origin (e.g., to the Davidson-Szarek tail or the Sudakov-Fernique constant) would improve readability without lengthening the proof.
  3. [Throughout] Notation: the operator norm ||∑ A_i²||_op is used without an explicit subscript in several places; consistent use of ||·||_{op} throughout would eliminate any possible confusion with other matrix norms.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of the manuscript, the recognition of its technical contributions, and the recommendation for minor revision. No specific major comments appear in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper derives explicit finite-N,M bounds on E[λ_max(H_M)] and E[λ_min(H_M)] from external concentration tools (Sudakov–Fernique comparison for matrix coefficients, Davidson–Szarek singular-value tails) combined with minimax duality applied to the variational formulas that define Lehner's ρ±. These steps are conditioned on the stated hypotheses (A_i ≽ 0, M ≥ N) but do not reduce the target inequalities to definitions of ρ±, fitted parameters, or self-citations; the large-M recovery of ρ is a limit statement, not a construction. No self-definitional, fitted-input, or load-bearing self-citation patterns appear in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard concentration inequalities for GUE matrices and minimax duality; no new free parameters, ad-hoc axioms, or invented entities are introduced beyond the positivity assumption already noted as weakest.

axioms (1)
  • standard math Standard concentration inequalities and minimax duality for operator norms of GUE matrices
    Invoked explicitly in the abstract as the basis for the finite-dimensional bounds.

pith-pipeline@v0.9.1-grok · 6001 in / 1389 out tokens · 23588 ms · 2026-06-26T13:38:51.978151+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

25 extracted references · 1 linked inside Pith

  1. [1]

    Anderson

    Greg W. Anderson. Convergence of the largest singular value of a polynomial in independent Wigner matrices.The Annals of Probability, 41(3B):2103–2181, 2013

    2013

  2. [2]

    An inequality for Hilbert-Schmidt norm.Communications in Mathematical Physics, 81(1):89–96, 1981

    Huzihiro Araki and Shigeru Yamagami. An inequality for Hilbert-Schmidt norm.Communications in Mathematical Physics, 81(1):89–96, 1981

    1981

  3. [3]

    A sharp small deviation inequality for the largest eigenvalue of a random matrix

    Guillaume Aubrun. A sharp small deviation inequality for the largest eigenvalue of a random matrix. InS´ eminaire de Probabilit´ es XXXVIII, volume 1857 ofLecture Notes in Mathematics, pages 320–337. Springer, Berlin, 2005

    2005

  4. [4]

    Bandeira, March T

    Afonso S. Bandeira, March T. Boedihardjo, and Ramon van Handel. Matrix concentration inequalities and free probability.Inventiones Mathematicae, 234(1):419–487, 2023

    2023

  5. [5]

    Springer, New York, 1997

    Rajendra Bhatia.Matrix Analysis, volume 169 ofGraduate Texts in Mathematics. Springer, New York, 1997

    1997

  6. [6]

    Operator Khintchine inequality in non-commutative probability.Mathematische An- nalen, 319(1):1–16, 2001

    Artur Buchholz. Operator Khintchine inequality in non-commutative probability.Mathematische An- nalen, 319(1):1–16, 2001

    2001

  7. [7]

    Tropp, and Ramon van Handel

    Chi-Fang Chen, Jorge Garza-Vargas, Joel A. Tropp, and Ramon van Handel. A new approach to strong convergence.Annals of Mathematics, 203(2):555–602, 2026

    2026

  8. [8]

    A new approach to strong convergence II

    Chi-Fang Chen, Jorge Garza-Vargas, and Ramon van Handel. A new approach to strong convergence II. the classical ensembles.arXiv preprint arXiv:2412.00593, 2024

    Pith/arXiv arXiv 2024

  9. [9]

    On the operator norm of non-commutative polynomi- als in deterministic matrices and iid GUE matrices.Cambridge Journal of Mathematics, 10(1):195–260, 2022

    Benoˆ ıt Collins, Alice Guionnet, and Felix Parraud. On the operator norm of non-commutative polynomi- als in deterministic matrices and iid GUE matrices.Cambridge Journal of Mathematics, 10(1):195–260, 2022

    2022

  10. [10]

    Davidson and Stanis law J

    Kenneth R. Davidson and Stanis law J. Szarek. Local operator theory, random matrices and Banach spaces. InHandbook of the Geometry of Banach Spaces, Vol. I, pages 317–366. North-Holland, Amster- dam, 2001

    2001

  11. [11]

    A new application of random matrices: Ext(C ∗ red(F2)) is not a group.Annals of Mathematics, 162(2):711–775, 2005

    Uffe Haagerup and Steen Thorbjørnsen. A new application of random matrices: Ext(C ∗ red(F2)) is not a group.Annals of Mathematics, 162(2):711–775, 2005

    2005

  12. [12]

    Lehner’s operator norm formulas, semidefinite programming, and spiked matrix models.arXiv preprint arXiv:2606.14687, 2026

    Dmitriy Kunisky. Lehner’s operator norm formulas, semidefinite programming, and spiked matrix models.arXiv preprint arXiv:2606.14687, 2026

    arXiv 2026

  13. [13]

    A remark on hypercontractivity and tail inequalities for the largest eigenvalues of random matrices

    Michel Ledoux. A remark on hypercontractivity and tail inequalities for the largest eigenvalues of random matrices. InS´ eminaire de Probabilit´ es XXXVII, volume 1832 ofLecture Notes in Mathematics, pages 360–369. Springer, Berlin, 2003

    2003

  14. [14]

    Computing norms of free operators with matrix coefficients.American Journal of Mathematics, 121(3):453–486, 1999

    Franz Lehner. Computing norms of free operators with matrix coefficients.American Journal of Mathematics, 121(3):453–486, 1999

    1999

  15. [15]

    The spectrum of a tensor of random and deterministic matrices.arXiv preprint arXiv:2410.04481, 2024

    F´ elix Parraud. The spectrum of a tensor of random and deterministic matrices.arXiv preprint arXiv:2410.04481, 2024

    arXiv 2024

  16. [16]

    Cambridge University Press, Cambridge, 1989

    Gilles Pisier.The Volume of Convex Bodies and Banach Space Geometry, volume 94 ofCambridge Tracts in Mathematics. Cambridge University Press, Cambridge, 1989

    1989

  17. [17]

    Cambridge University Press, Cambridge, 2003

    Gilles Pisier.Introduction to Operator Space Theory. Cambridge University Press, Cambridge, 2003

    2003

  18. [18]

    On general minimax theorems.Pacific Journal of Mathematics, 8(1):171–176, 1958

    Maurice Sion. On general minimax theorems.Pacific Journal of Mathematics, 8(1):171–176, 1958

    1958

  19. [19]

    On the spectral norm of Gaussian random matrices.Transactions of the American Mathematical Society, 369(11):8161–8178, 2017

    Ramon van Handel. On the spectral norm of Gaussian random matrices.Transactions of the American Mathematical Society, 369(11):8161–8178, 2017. 16

    2017

  20. [20]

    The strong convergence phenomenon

    Ramon van Handel. The strong convergence phenomenon. InCurrent Developments in Mathematics

  21. [21]

    To appear

    International Press, 2025. To appear

    2025

  22. [22]

    Cambridge University Press, Cambridge, 2018

    Roman Vershynin.High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge University Press, Cambridge, 2018

    2018

  23. [23]

    Richard A. Vitale. Some comparisons for Gaussian processes.Proceedings of the American Mathematical Society, 128(10):3043–3046, 2000

    2000

  24. [24]

    Limit laws for random matrices and free products.Inventiones Mathematicae, 104(1):201–220, 1991

    Dan Voiculescu. Limit laws for random matrices and free products.Inventiones Mathematicae, 104(1):201–220, 1991

    1991

  25. [25]

    Cambridge University Press, 2018

    John Watrous.The Theory of Quantum Information. Cambridge University Press, 2018. 17

    2018