Multilevel limits of spiked random matrix minors

Benjamin Landon; Tara Stojimirovic

Multilevel limits of spiked random matrix minors

T0 review · 0 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper proves that the rescaled largest eigenvalues of all principal minors of a critically spiked Gaussian random matrix converge jointly to a multilevel interlacing particle system, with the Airy-β point process at the bottom.

desk verdict Solid multi-level extension to β>0 with a patchable small-i error and a sketched K-level induction; worth a serious referee. read the letter →

arxiv 2608.01531 v1 pith:PDZORVJK submitted 2026-08-02 math.PR

Benjamin Landon , Tara Stojimirovic This is my paper

classification math.PR MSC 60B2060G55

keywords spikedrandommatricesBBPtransitionAiry-betapointprocessmultilevelinterlacingprincipalminorsGaussianbeta-ensemblecriticalregimeedgeuniversality

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

Add a finite-rank spike of size $N^{1/2}+\alpha N^{1/6}$ to a Gaussian random matrix and look at its principal minors. The paper tries to prove that the largest eigenvalues of all these minors, after the standard $N^{1/6}$ edge rescaling, converge together to a nested interlacing particle system: the bottom level is the Airy-$\beta$ point process, and each higher level is obtained from the level below by solving an explicit renormalized equation—a weighted sum with one pole at every particle of the previous level, regularized by subtracting the large semicircle background—with independent Gamma weights. This matters because it gives one joint description of edge fluctuations across all levels of a spiked matrix, and because the same construction defines the limit for every $\beta>0$ even where no random matrix model is available.

What carries the argument

The load-bearing object is the renormalized transition map $\Phi$: given a level $a=(a_1,a_2,\ldots)$ and weights $z=(z_1,z_2,\ldots)$, the next level $b=\Phi(a,z,\alpha)$ is the unique interlacing solution set of $\alpha=F(E;a,z)$, with $F$ defined as the renormalized sum of $z_i/(E-a_i)$ minus the divergent semicircle contribution. This map carries the argument because it converts the Markovian recursion (1.5) satisfied by the finite-$N$ minor eigenvalues into a deterministic level-to-level transition, while the interlacing follows from the pole structure of $F$. The proof uses finite-$m$ truncations of $F$ and controls the omitted tails with edge rigidity estimates for the Gaussian $\$\beta$

What would settle it

At $\beta=2$ the exact joint density of the eigenvalues of nested minors is known; compute from it the finite-$N$ law of the top two eigenvalues of the top two minors for growing $N$, rescale by $N^{1/6}(2N^{1/2}-\lambda)$, and compare with the interlacing system defined by $F$ (for example, the variance of the second level conditional on the first). If the finite-dimensional law does not approach the $\Phi$-defining equations, Theorem 1.2 is false.

Watch

Extended reading notes

Core claim

The central claim (Theorem 1.2) is that for the Gaussian $\beta$-ensemble with a critical finite-rank spike, the sequence $x^{(i)}$ of rescaled largest eigenvalues of the $(N+i)\times(N+i)$ principal minors converges in finite-dimensional distribution to a sequence $a^{(i)}$ of interlacing point processes. Here $a^{(0)}$ is the Airy-$\beta$ point process and $a^{(i)}$ is defined from $a^{(i-1)}$ by taking the unique interlacing solutions $b_1<a_1<b_2<a_2<\cdots$ of $\alpha=F(E;a,z)$, where $F$ is the renormalized limit $F(E;a,z)=\lim_{n\to\infty}(\sum_{i=1}^n z_i/(E-a_i)+\int_0^{(3\pi n/2)^{2/3}} x^{-1/2}/\pi\,dx)$. The $z_i$ are independent Gamma variables with the $\beta$-dependent density

Load-bearing premise

The load-bearing premise is that the largest eigenvalues of the unspiked Gaussian $\beta$-ensemble fluctuate no more than the model predicts, through the imported rigidity bounds and the Airy-$\beta$ finite-dimensional convergence; if the local fluctuation rate were any slower than these bounds, the truncations of the infinite level equations would fail.

Editorial extensions

If this is right

If the theorem is right, the joint edge fluctuations of the top $K+1$ principal minors of a critically spiked Gaussian matrix have a universal limit described by interlacing point processes, not just a limit for each minor separately.
The same multilevel limit is well-defined for every $\beta>0$, so the hierarchy exists as a probability law even for values of $\beta$ with no classical random matrix model.
The level-to-level transition is explicit and Markovian: condition on the first level, and the next level is obtained by solving a single renormalized equation with known Gamma weights.
At the critical scaling with finite $\alpha$ the hierarchy is non-degenerate and interlacing; in the paper's description, $\alpha_i\to\pm\infty$ degenerates into sticking or Gaussian behavior, so the critical regime is where the multilevel structure genuinely appears.
Finite-$m$ truncations of the defining equations approximate the top $L$ particles of the limit with high probability, giving a concrete finite-dimensional description of the limiting process.

Reading between the lines

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

Because the proof only needs the unspiked edge process to be Airy-like in a moment sense, the same interlacing construction may apply to any edge-universal ensemble with a suitable spike, not only Gaussian $\beta$-ensembles; this is a testable extension to Wishart or Wigner-type models.
The finite-$m$ truncation suggests a practical simulation scheme: sample the Airy-$\beta$ process, draw independent Gamma variables, and solve the $m$-equation system to approximate the top particles of every level; the paper's approximation results give the error control such a scheme would need.
The explicit Markovian transition may let one compute level-dependent statistics (e.g. spacing distributions between consecutive levels) by iterating the Airy-$\beta$ process through $F$, which the paper does not do.
The $\alpha\to\pm\infty$ degenerations hint at a crossover phase diagram for the multilevel process; quantifying the transition scale at which interlacing is replaced by sticking or Gaussian fluctuations is a natural open next step.

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.

Desk Editor's Note

read the letter

Quick take: this is a genuine extension of the known β=2 multilevel interlacing limit for spiked Gaussian random matrix minors to β=1,4, and it constructs a well-defined infinite interlacing object for all β>0. The β=2 result is due to Bao-Wang [3]; the single-level K=1 limit for classical β is essentially in Bykhovskaya-Gorin-Sodin [9]. The new content is the multi-level limit for the other classical β values and the recursive construction of the edge limit for general β, with convergence. That is a real advance within random matrix theory.

What the paper does well: the two-level proof is detailed and follows a natural truncation plus continuous mapping scheme. The rigidity inputs from [8] and [7] are cited appropriately, and the proof is largely self-contained. The paper also honestly discloses the overlap with [9] and [20] and explains the connection to the Hermite-β corners process [14] via [19]. The construction of the limiting process in Section 3 is careful: Lemma 3.2 and Lemma 3.3 establish existence and the interlacing structure, and Lemma 3.4 shows the next level is again Airy-like. The approximation lemmas (Propositions 3.7 and 4.6) are central, and they hold up at the level I followed. There is no circularity: the base is the externally established Airy-β process, the α_i are model inputs, and convergence is proved, not assumed.

Soft spots: the K-level statement is only sketched at the start of Section 5. The induction is standard given the Markovian structure and the transfer lemmas, but it should be written out. The small-i case of Corollary 2.3 is mis-stated, as the stress-test note says: the bound (2.9) cannot hold for i=1 because the RHS is 0 while the reflected Airy-β process has positive first moment. This is patchable because Definition 3.1 only needs the bound for large i, and the finite initial segment is handled by tightness. There are also sign typos in (3.6)/(4.19); the intended plus sign is clear from context. None of these are load-bearing in my reading.

Bottom line: this is a credible, significant within-subfield advance. The central theorem is convincing, and the remaining issues are presentation/patch-level rather than conceptual. I would send it to a knowledgeable referee, and I'd bring it to a reading group that works on random matrix edge limits or integrable probability.

Referee Report

0 major / 5 minor

Summary. The paper studies the joint edge fluctuations of the largest eigenvalues of the principal minors of an (N+K)x(N+K) Gaussian beta-ensemble with a critical finite-rank spiked diagonal (1.1). It proves (Theorem 1.2) that the rescaled eigenvalues converge in finite-dimensional distribution to a multi-level interlacing particle system whose lowest level is the Airy-beta point process, with each subsequent level generated by solving the Markovian fixed-point equation (1.8). The construction is extended to all beta>0. The proof is organized around a two-level core: rigidity estimates for the Gaussian beta-ensemble, truncation to finitely many particles, uniform approximation lemmas (Propositions 3.7 and 4.6, Lemma 4.3), and a continuous mapping argument for the truncated systems.

Significance. If correct, this is a substantial extension of earlier work on the critical BBP transition and on multilevel minor processes, covering beta=1,4 and giving a natural limit for all beta>0 without determinantal methods. The paper is notable for its modular and transparent proof structure: the limiting object is defined from the externally established Airy-beta process via an explicit recursion, and no parameter is fitted. The main technical reliance is on standard imported input: edge rigidity estimates from [7,8] and finite-dimensional convergence to Airy-beta from [21-23]. The central result is well-motivated and the proof strategy is sound.

minor comments (5)

[2, Proposition 2.1(2.6) and Corollary 2.3(2.9)] As written, (2.6) and (2.9) cannot hold for i=1, since |log 1|=0 and the left sides are positive random quantities; e.g. hatgamma_1=(3pi/2)^{2/3} is about 2.75 while the first Airy-beta point has positive variance. The proof in Appendix A.1 actually treats small j separately, so the statements should restrict i to i>=i_0 or replace |log i| by log(1+i). This is not load-bearing for the main theorem because Definition 3.1 only requires the bound for i>=hat K and finite initial segments are controlled by tightness, but the present wording is a genuine mathematical error.
[3.2, Eq. (3.6), and 4.1, Eq. (4.19)] There are sign typos in these two displayed equations. In (3.6) the integral is subtracted, while (3.5), (3.32) and the surrounding text require the plus sign. In (4.19) the integral is subtracted, while the analogous truncated equation (3.23) and the untruncated equation (4.4) require a plus sign. The intended meaning is clear, but the typos should be corrected.
[3.2, Proposition 3.7, j=1 case] In (3.38), the index i is not specified. Presumably i=1 is intended, with the lower bound involving z_1. Please state the index explicitly and define the event A(m) in this case; the current text is unnecessarily hard to parse.
[5, proof of Theorem 1.2] The reduction from the two-level case to the general K-level theorem is described in a single sentence: 'the general case following from an induction argument and Markovian structure of the system.' Since Theorem 1.2 is the K-level statement, please state the induction hypothesis and indicate how the uniform error estimates of Propositions 3.7, 4.6, and Lemma 4.3 compose with the continuous mapping argument when several levels are considered simultaneously. I expect this is routine, but the details are currently left entirely to the reader.
[3.1, Lemma 3.4] The proof that b=Phi(a,z,alpha) is Airy-like is compressed into one sentence. Interlacing gives only |b_i - a_{i-1}| <= a_i - a_{i-1}; please spell out how the Airy-like moment bound for a_i and the comparison |hatgamma_i-hatgamma_{i-1}| <= C i^{-1/3} yield (3.1) for b_i after increasing hat K.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the multilevel limit is a proved theorem built on external Airy-β convergence and rigidity estimates.

full rationale

The derivation is self-contained in the relevant sense. The finite-N recursion (1.3)/(1.5) is obtained by exact algebra from the rank-one perturbation equation (1.2); no parameter is fitted and no target distribution is used as an input. The limiting recursion (1.6)-(1.8) is not assumed to converge: Proposition 1.1 proves admissibility from the Airy-like bound (3.1), which for the base Airy-β process is imported from the independent works [21-23] and [7,8]. The induction step is a genuine continuity theorem: rigid finite-N inputs (Definition 4.1) produce rigid outputs (Lemma 4.2), Airy-like inputs produce Airy-like outputs (Lemma 3.4), and truncation errors are controlled uniformly (Lemmas 3.6 and 4.5, Propositions 3.7 and 4.6). The two-level proof then applies the continuous mapping theorem to the common defining function Ψ; no equation used in the proof is equivalent by construction to the conclusion. The only self-citations, [17,18], appear as methodological analogies and are not load-bearing evidence for any theorem. The only concrete defect found is a non-circular correctness issue in Corollary 2.3, Eq. (2.9): at i=1 the right-hand side is 0 while E|a_1-γ̂_1| > 0, so the stated inequality cannot hold for all i; however Definition 3.1 needs the bound only for i ≥ K̂, and no proof step uses the small-i regime, so this does not create circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on external inputs: edge rigidity and Airy-β convergence for β-ensembles, plus the standard linear algebra of rank-one perturbations. No parameters are fitted to data; the model's spike strengths α_i and the distributions of z_i and g_i are specified by the problem. The limiting process is constructed from the same recursion as the finite-N process, but its base is the externally known Airy-β process.

assumptions (5)

standard math The Gaussian β-ensemble satisfies the edge rigidity bounds (2.6)-(2.7) for all β>0.
Imported from [8, Proposition 3.5] and [7]; used in Proposition 2.1 and throughout Appendix A, Lemma A.3 and Lemma 4.2 to control the eigenvalue locations.
standard math The top eigenvalues of the Gaussian β-ensemble converge in finite-dimensional distribution to the Airy-β point process.
From [21-23]; this is the base level a^(0) of the limiting interlacing system and the input to the recursion (1.8).
domain assumption For classical β=1,2,4, the coordinates of the new column in the eigenbasis of the minor are independent Gamma(β/2,β/2) variables, independent of the eigensystem.
Derives from rotational invariance of Gaussian ensembles; this underpins the recursion (1.2)-(1.3) and the general-β construction.
standard math The eigenvalues of G+A and of its principal minors satisfy the secular equation (1.2) and the interlacing μ_1 > λ_1 > μ_2 > λ_2 > ... .
Standard rank-one perturbation/minor linear algebra, cited to [16, (2.3)]; is the basis for the Markovian recursive description of the levels.
domain assumption The Hermite β-corners process of [14] has the same recursive structure via (1.2) when A=0.
Connection via [19] and [12, Proposition 4.3.2]; used for context and for relating the result to the fixed-N process, not load-bearing for the proof.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Multilevel limits of spiked random matrix minors." pith.science (2026). https://pith.science/paper/PDZORVJK

@misc{pith2026260801531,
  author       = {Pith},
  title        = {Pith review of: Multilevel limits of spiked random matrix minors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PDZORVJK}},
  note         = {Machine review of arXiv:2608.01531}
}

abstract

We consider the largest eigenvalues of minors of the classical Gaussian random matrices with a finite rank spike in the critical BBP regime. We show that they converge to a multi-level interlacing particle system whose lowest level is the Airy$_\beta$ point process. Our construction and scaling limits extend in a natural way to non-classical $\beta >0$.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[3]

Bao and D

Z. Bao and D. Wang. Eigenvector distribution in the critical regime of bbp transition. Probability Theory and Related Fields, 182(1):399–479, 2022

work page 2022
[9]

Bykhovskaya, V

A. Bykhovskaya, V. Gorin, and S. Sodin. How weak are weak factors? uniform inference for signal strength in signal plus noise models. 2026

work page 2026
[8]

Bourgade, K

P. Bourgade, K. Mody, and M. Pain. Optimal local law and central limit theorem for β-ensembles.Communications in Mathematical Physics, 390(3):1017–1079, 2022

work page 2022
[7]

Bourgade, L

P. Bourgade, L. Erd¨ os, and H.-T. Yau. Edge universality of beta ensembles.Communi- cations in Mathematical Physics, 332(1):261–353, 2014

work page 2014
[20]

Petrov and J

L. Petrov and J. Xu. Perturbed beta corners process. 2026

work page 2026
[14]

Gorin and M

V. Gorin and M. Shkolnikov. Multilevel dyson brownian motions via jack polynomials. Probability Theory and Related Fields, 163(3):413–463, 2015

work page 2015
[19]

Najnudel and B

J. Najnudel and B. Vir´ ag. The bead process for beta ensembles.Probability Theory and Related Fields, 179(3):589–647, 2021

work page 2021
[1]

Adler, P

M. Adler, P. Van Moerbeke, and D. Wang. Random matrix minor processes related to percolation theory.Random Matrices: Theory and Applications, 2(04):1350008, 2013

work page 2013

Show all 23 references

[2]

J. Baik, G. Ben Arous, and S. P´ ech´ e. Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices. 2005

2005
[4]

Z. Bao, D. Wang, and Y. Zhu. Eigenvector distribution of random matrices under critical finite-rank deformations. 2026

2026
[5]

Bloemendal and V

A. Bloemendal and V. B´ alint. Limits of spiked random matrices i.Probability Theory and Related Fields, 156(3-4):795–825, 2013

2013
[6]

Bloemendal and B

A. Bloemendal and B. Vir´ ag. Limits of spiked random matrices ii. 2016

2016
[10]

Erd˝ os, A

L. Erd˝ os, A. Knowles, H.-T. Yau, and J. Yin. The local semicircle law for a general class of random matrices. 2013

2013
[11]

Erd˝ os and H.-T

L. Erd˝ os and H.-T. Yau.A dynamical approach to random matrix theory, volume 28. American Mathematical Soc., 2017

2017
[12]

P. J. Forrester.Log-gases and random matrices (LMS-34). Princeton university press, 2010. 21

2010
[13]

Gorin and M

V. Gorin and M. Shkolnikov. Interacting particle systems at the edge of multilevel dyson brownian motions. 2014

2014
[15]

Gustavsson

J. Gustavsson. Gaussian fluctuations of eigenvalues in the gue.Annales de l’IHP Prob- abilit´ es et statistiques, 41(2):151–178, 2005

2005
[16]

J. Huang. Eigenvalues for the minors of wigner matrices.arXiv preprint arXiv:1907.10214, 2019

1907 arXiv
[17]

Landon and P

B. Landon and P. Sosoe. Fluctuations of the 2-spin ssk model with magnetic field.arXiv preprint arXiv:2009.12514, 2020

2009 arXiv
[18]

Landon and P

B. Landon and P. Sosoe. Fluctuations of the overlap at low temperature in the 2- spin spherical sk model. InAnnales de l’Institut Henri Poincare (B) Probabilites et statistiques, volume 58, pages 1426–1459. Institut Henri Poincar´ e, 2022

2022
[21]

Ramirez, B

J. Ramirez, B. Rider, and B. Vir´ ag. Beta ensembles, stochastic airy spectrum, and a diffusion.Journal of the American Mathematical Society, 24(4):919–944, 2011

2011
[22]

C. A. Tracy and H. Widom. Level-spacing distributions and the airy kernel.Communi- cations in Mathematical Physics, 159(1):151–174, 1994

1994
[23]

C. A. Tracy and H. Widom. On orthogonal and symplectic matrix ensembles.Commu- nications in Mathematical Physics, 177(3):727–754, 1996. 22

1996

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [3]

Bao and D

Z. Bao and D. Wang. Eigenvector distribution in the critical regime of bbp transition. Probability Theory and Related Fields, 182(1):399–479, 2022

work page 2022

[2] [9]

Bykhovskaya, V

A. Bykhovskaya, V. Gorin, and S. Sodin. How weak are weak factors? uniform inference for signal strength in signal plus noise models. 2026

work page 2026

[3] [8]

Bourgade, K

P. Bourgade, K. Mody, and M. Pain. Optimal local law and central limit theorem for β-ensembles.Communications in Mathematical Physics, 390(3):1017–1079, 2022

work page 2022

[4] [7]

Bourgade, L

P. Bourgade, L. Erd¨ os, and H.-T. Yau. Edge universality of beta ensembles.Communi- cations in Mathematical Physics, 332(1):261–353, 2014

work page 2014

[5] [20]

Petrov and J

L. Petrov and J. Xu. Perturbed beta corners process. 2026

work page 2026

[6] [14]

Gorin and M

V. Gorin and M. Shkolnikov. Multilevel dyson brownian motions via jack polynomials. Probability Theory and Related Fields, 163(3):413–463, 2015

work page 2015

[7] [19]

Najnudel and B

J. Najnudel and B. Vir´ ag. The bead process for beta ensembles.Probability Theory and Related Fields, 179(3):589–647, 2021

work page 2021

[8] [1]

Adler, P

M. Adler, P. Van Moerbeke, and D. Wang. Random matrix minor processes related to percolation theory.Random Matrices: Theory and Applications, 2(04):1350008, 2013

work page 2013