REVIEW 5 minor
Extreme principal minors of Wishart and deformed GOE matrices
T0 review · 0 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read This paper proves that the extreme principal-minor eigenvalues of deformed GOE and sub-Gaussian Wishart matrices converge to explicit deterministic limits, found by identifying the Hausdorff limit of the random set of normalized minors…
desk verdict The a>2 deformed GOE gap is genuinely resolved here, and the Hausdorff-set method is the reusable idea; the Wishart part carries an explicit sub-Gaussian caveat, but it is stated and does not undermine the central claims. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the Hausdorff limit of a random set of scaled principal minors. For Wigner matrices, the limit is $K_k(a)$, the compact convex set of symmetric $k\times k$ matrices in which every principal submatrix obeys $I_{U,a}(B)\le |U|$; for Wishart matrices, the limit is $\Gamma_{k,\beta,\mu}$, the set of second-moment matrices $\int xx^\top d\nu$ of probability laws $\nu$ whose every $U$-marginal has KL divergence at most $|U|/\beta$ from $\mu^{\otimes |U|}$. The proof shows $d_H(C_{n,k},K_k(a))\to 0$ (and the Wishart analogue) and then uses the fact that $\lambda_{\max}$ is convex and $1$-Lipschitz, so the maximum over the random set converges to the maximum over the deterministic convex set; the Wigner bound is driven by Gaussian chi-square tails, and the Wishart bound by Sanov's theorem.
What would settle it
Simulate a Gaussian Wishart matrix with $p=\beta\log n$, $\beta=2$, $k=2$, and $n$ large; Corollary 1 predicts $T_{n,2}/p$ concentrates near the root $\lambda\approx 4.505$ of $\lambda-\log\lambda-1=2$, so a persistent deviation in the sample mean of $T/p$ across many independent replicates would refute the claim. Alternatively, for deformed GOE with $a=4$ and $k=3$, $M_{n,3}/\sqrt{2\log n}$ should concentrate near $\gamma_3(4)=2.9$.
Extended reading notes
Core claim
The central discovery is that the random sets formed by suitably normalized principal minors converge in Hausdorff distance to deterministic convex sets, so the extreme-value problem becomes a finite-dimensional convex optimization. For a Gaussian Wigner matrix with off-diagonal variance $1$ and diagonal variance $a$, the set $C_{n,k}=\{A_S/\sqrt{2\log n}: |S|=k\}$ converges in Hausdorff distance to $K_k(a)$, the set of symmetric $k\times k$ matrices $B$ with $(1/a)\sum_{i\in U}B_{ii}^2+\sum_{i<j\in U}B_{ij}^2\le |U|$ for every nonempty $U\subseteq[k]$; Theorem 1 then yields $M_{n,k}/\sqrt{2\log n}\to\gamma_k(a)=\max_{B\in K_k(a)}\lambda_{\max}(B)$. Theorem 2 evaluates this constant: $\gamma_k(a)=\sqrt{a+2(k-1)}$ for $0<a\le 2$, while for $a>2$ it obeys $\gamma_m(a)=\gamma_{m-1}(a)+2/(\gamma_{m-1}(a)+\sqrt{\gamma_{m-1}(a)^2+4-a})$ with $\gamma_1(a)=\sqrt{a}$, and the maximizing minors form a nested chain. For Wishart matrices, the scaled sets $\{X_S^\top X_S/p\}$ converge in Hausdorff distance to $\Gamma_{k,\beta,\mu}=\{\int xx^\top d\nu: \mathcal{D}(\nu_U\|\mu^{\otimes |U|})\le |U|/\beta \text{ for all } U\subseteq[k]\}$, so $T_{n,k}/p\to\sup_{Q\in\Gamma_{k,\beta,\mu}}\lambda_{\max}(Q)$; for standard Gaussian entries this supremum is $\lambda_{k,\beta}$, the unique root in $(1,\infty)$ of $\lambda-\log\lambda-1=2k/\beta$.
Load-bearing premise
For the Wishart results, the load-bearing premise is that the entry distribution has a finite exponential moment and that $p/\log n$ converges to a positive constant; if either fails, the proof's truncation step and entropy constraints no longer guarantee that the limiting set captures the extreme principal-minor eigenvalues.
Editorial extensions
If this is right
- For deformed GOE with diagonal variance $a>2$, the extreme principal minor grows like $\gamma_k(a)\sqrt{2\log n}$, and the optimizing minors form a nested chain as $k$ increases; this structural fact is tied to the failure of the usual Gumbel limit in that regime.
- For Wishart matrices with $p/\log n\to\beta\in(0,\infty)$, the largest normalized principal-minor eigenvalue converges to the largest eigenvalue over the entropy-constrained set $\Gamma_{k,\beta,\mu}$, and for Gaussian entries this limit is the explicit $\lambda_{k,\beta}>1$ solving $\lambda-\log\lambda-1=2k/\beta$.
- In the Wishart endpoint regimes, $p/\log n\to\infty$ gives the universal limit $T_{n,k}/p\to1$, while $p/\log n\to0$ gives a tail-driven limit $T_{n,k}/(k\log n)\to1/\kappa_\mu$ where $\kappa_\mu$ is the exponential tail rate of the squared entries.
- For Wigner matrices with growing $k$, $M_{n,k}/(2\sqrt{k\log(n/k)})\to1$ when $k/n\to0$, and $M_{n,k}/\sqrt{n}\to E(c)$ when $k/n\to c\in(0,1]$, with $E$ increasing, concave, $E(1)=2$, and explicit rates at both endpoints.
- In compressed sensing with Gaussian sensing matrices, the restricted-isometry constant of order $k$ converges to $\lambda_{k,\beta}-1$, so the condition $\beta>2k/(\delta-\log(1+\delta))$ makes the restricted isometry property hold with probability tending to one.
Reading between the lines
- Beyond the paper, the same Hausdorff-set route should apply to other ensembles, such as Wigner matrices with non-Gaussian entries or Wishart designs with correlated columns, whenever a Sanov-type large-deviation control of the scaled minors is available; only the limiting constraint set would change.
- Beyond the paper, the nested-chain property for $a>2$ yields a directly testable prediction: for large $n$, the support of the maximizing $k\times k$ minor should contain the support of the maximizing $(k-1)\times(k-1)$ minor with high probability.
- Beyond the paper, the explicit Wishart formula suggests a finite-sample heuristic that the transition in the restricted-isometry constant sharpens around $n\approx e^{p/\beta}$; simulations at moderate $n$ could measure the width of this transition and test the logarithmic scaling.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper establishes laws of large numbers for the largest eigenvalues among all principal minors of deformed GOE matrices and Wishart matrices. For deformed GOE with off-diagonal variance one and diagonal variance a>0, it proves that M_{n,k}/√(2 log n) converges in probability to γ_k(a), with γ_k(a)=√(a+2(k−1)) for 0<a≤2 and, for a>2, an explicit recursion γ_m(a)=γ_{m−1}(a)+2/(γ_{m−1}(a)+√(γ_{m−1}(a)^2+4−a)); this resolves the case left open by Cai et al. (2021). The proof goes through a Hausdorff-convergence result for the random set of normalized principal minors to a deterministic convex set defined by convex constraints. For Wishart matrices with i.i.d. sub-Gaussian entries and p/log n→β∈(0,∞), the paper proves Hausdorff convergence to an entropy-constrained convex set Γ_{k,β,μ}; for standard Gaussian entries it identifies the limit λ_{k,β} as the unique solution of λ−logλ−1=2k/β. Endpoint regimes β=0 and β=∞ are also treated, and the results are applied to a compressed-sensing restricted-isometry threshold.
Significance. The Hausdorff-convergence framework is the paper's main methodological contribution and is well executed: the limiting sets are derived from first principles, the proofs are detailed, and the explicit phase transition at a=2 for fixed k is a genuine resolution of a previously open problem. The Wishart entropy-constrained characterization and its explicit Gaussian solution are substantial, and the compressed-sensing threshold provides a concrete falsifiable prediction. I checked the key steps—the tight-set chain argument, the Slepian-based superadditivity, the KL upper/lower bounds, and the endpoint arguments—and found no internal inconsistency. The moment condition (10) and the scaling p/log n→β are stated explicitly and are used where needed; no circularity or fitted constants are present.
minor comments (5)
- [Abstract; Section 2.1] The phrase 'no close-form expression' should read 'closed-form expression', and 'Hausdorffdistance' in the abstract is missing a space.
- [Section 3] The endpoint limit T_{n,k}/p→1 for p/log n→∞ is stated without proof; since it is part of the claimed full-range coverage, please add the short union-bound argument based on diagonal and off-diagonal tail bounds under (10), or give an explicit reference.
- [Table 1] In the row for k→∞, the denominator should be typeset unambiguously as 2√(k log(n/k)); the current '2√(klog(n/k))' is easy to misread.
- [Section 10.2] The reduction to strict constraints via mixtures is only sketched; a sentence noting that Q((1−δ)ν+δμ^{⊗k})=(1−δ)Q(ν)+δI_k and that all KL constraints become strict would make the argument fully self-contained.
- [Section 9] In the final step, the sentence beginning 'Let h=(h_1,...,h_n)^T, where h_1...' contains a duplicated 'where', and the reused symbol h for both the normal replacement variables and the standard normal vector could confuse readers.
Circularity Check
No significant circularity: the limiting constants are derived from explicit convex optimization problems and verified by direct constructions, with self-citations only providing context or consistency checks.
full rationale
The central results are self-contained derivations rather than renamed inputs. For the Gaussian Wigner case, the deterministic set K_k(a) is defined in equation (1) by the quadratic constraints I_{U,a}(B) <= |U|, and the limiting constant gamma_k(a) is defined as the maximum of lambda_max over this set. The recursion in Theorem 2 is proved in Proposition 1 from the first-order optimality conditions of the finite-dimensional convex program, with an explicit matrix B_t constructed for the lower bound (30). No parameter is fitted to M_{n,k} data, and the phase transition at a=2 emerges from the Cauchy-Schwarz argument and the optimization, not from an assumed value. For the Wishart case, Gamma_{k,beta,mu} is defined in (12)-(13) through KL-divergence constraints, and Theorem 6 proves Hausdorff convergence of the random set of normalized Gram matrices to this set using Sanov's theorem, truncation, and Sion's minimax theorem. Corollary 1 obtains lambda_{k,beta} by solving the resulting convex optimization explicitly: the upper bound uses Jensen's inequality with the KL constraint, and the lower bound constructs the Gaussian measure N(0,Q*) and verifies its KL marginals are exactly within the allowed budget. This is a genuine derivation, not an equivalent restatement of the limit. The self-citations (Cai et al. 2021, Jiang and Qi 2024) are used only to frame the open problem and to check special cases, such as the k=2 formula in (8), and are not load-bearing for the main results. The technical assumptions, including the sub-Gaussian condition (10) and the scaling p/log n -> beta, are explicitly stated and used in the proofs rather than serving as surrogates for the conclusions. No fitted input is renamed as a prediction, and no imported uniqueness theorem forces the choice of limiting set. The paper is therefore not circular.
Assumptions & free parameters
assumptions (4)
- domain assumption Wigner entries are Gaussian with off-diagonal variance 1 and diagonal variance a>0; entries above the diagonal independent
- domain assumption Wishart entries ξ satisfy Eξ=0, Eξ^2=1, and E exp(t0 ξ^2)<∞ for some t0>0
- domain assumption p/logn→β∈(0,∞) for the main Wishart results; endpoint regimes β=0 and β=∞ treated separately
- standard math Standard large-deviation and Gaussian comparison results: Sanov's theorem, Slepian's lemma, Sudakov-Fernique, Paley-Zygmund, and the strong law for L-statistics
Cite this review
Pith. "Pith review of Extreme principal minors of Wishart and deformed GOE matrices." pith.science (2026). https://pith.science/paper/HOLTOQIA
@misc{pith2026260813154,
author = {Pith},
title = {Pith review of: Extreme principal minors of Wishart and deformed GOE matrices},
year = {2026},
howpublished = {\url{https://pith.science/paper/HOLTOQIA}},
note = {Machine review of arXiv:2608.13154}
}
abstract
We study the laws of large numbers for the largest eigenvalues among all principal minors of Wishart matrices and deformed GOE matrices. We propose a new method based on identifying the deterministic sets to which the random sets formed by suitably normalized principal minors converge in Hausdorff distance, thereby reducing the original extreme-value problems to finite-dimensional convex optimization problems. We demonstrate the effectiveness of this method in regimes not covered by the existing second-moment arguments in \cite{cai2021asymptotic,hu2023extreme}. For deformed GOE matrices with fixed minor size \(k\), we determine the limit for every diagonal variance \(a>0\) and identify a phase transition at \(a=2\). Above the transition, the limiting constant satisfies an explicit recursion with no close-form expression, and the optimizers exhibit a nested hierarchical structure, thereby resolving the case left open in \cite{cai2021asymptotic}. For Wishart matrices with general sub-Gaussian entries and fixed \(k\), we characterize the limit through an entropy-constrained deterministic convex set. When the entries are standard Gaussian, we solve the resulting optimization problem explicitly and obtain the exact value of the limiting constant.
Figures
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.