Cohen-Lenstra flag universality for random matrix products

Hoi H. Nguyen; Roger Van Peski; Yifeng Huang

Cohen-Lenstra flag universality for random matrix products

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Cohen-Lenstra flag law holds universally for matrix products

desk verdict Real universality claim for cokernel flags, but the abstract's 'nondegenerate' needs unpacking before the theorem can be trusted. read the letter →

arxiv 2508.10127 v1 pith:42OY6OYX submitted 2025-08-13 math.PR math.COmath.NT

Yifeng Huang , Hoi H. Nguyen , Roger Van Peski This is my paper · ORCID

classification math.PRmath.COmath.NT MSC 60B2005E05

keywords Cohen-Lenstraheuristicsrandommatrixcokernelsabelianp-groupsuniversalityproductsHall-Littlewoodpolynomialsflags

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

The paper claims that when $k$ independent $n\times n$ random matrices with nondegenerate integer entries are multiplied, the cokernels of the successive partial products form a random flag of abelian $p$-groups whose distribution, in the $n\to\infty$ limit, is universal and equals the Cohen-Lenstra measure weighting a flag by $1/|\operatorname{Aut}(\operatorname{flag})|$. This would mean the joint arithmetic structure of random matrix products has no dependence on the entry distribution, extending single-matrix universality to arbitrary finite products. A concrete corollary is that the conditional law of $\mathrm{cok}(M_1M_2)$ given the cokernels of $M_1$ and $M_2$ is governed by Hall-Littlewood structure constants, a formula previously available only for Haar-random matrices over the $p$-adic integers.

What carries the argument

The central object is the random flag of abelian $p$-groups obtained from the cokernels of partial products of the random matrices. The universal limit measure is defined by the Cohen-Lenstra weight $1/|\operatorname{Aut}(\operatorname{flag})|$. The proof's engine is the product-moment method: joint moments of functions of the partial-product cokernels are computed for arbitrary entry distributions and shown to match the corresponding moments in the Haar-random $p$-adic case, using Hall-Littlewood structure constants as the algebraic coefficients that encode these moments.

What would settle it

Take $k=2$, $p=2$, and entries i.i.d. uniform on $\{0,1\}$. For large $n$, compute numerically the joint moment $\mathbb{E}[\#\mathrm{Hom}(\mathrm{cok}(M_1),\mathbb{Z}/2)\,\#\mathrm{Hom}(\mathrm{cok}(M_1M_2),\mathbb{Z}/2)]$; if it does not converge to the value predicted by the Cohen-Lenstra weight $1/|\operatorname{Aut}|$, the universality claim is false. Alternatively, find any nondegenerate entry distribution for which the limiting flag differs from the Haar flag.

Watch

Extended reading notes

Core claim

The central discovery is that the joint law of the flag $(\mathrm{cok}(M_1\cdots M_i))_{1\le i\le k}$ converges, as $n\to\infty$, to the probability measure on flags of abelian $p$-groups with weight proportional to $1/|\operatorname{Aut}(\lambda)|$ for each flag $\lambda$. This is the natural flag analogue of the Cohen-Lenstra measure for a single cokernel. The paper proves this by computing joint product moments and matching them to the corresponding moments of the Haar-random flag over $\mathbb{Z}_p$, thereby establishing universality for all nondegenerate entry distributions. As a result, the conditional distributions appearing in the formula are universal and are expressed in terms of H

Load-bearing premise

The argument assumes a sufficiently strong nondegeneracy condition on the matrix entries that lets the moment-matching method control the joint law of all $k$ partial-product cokernels at once, and that the previously computed Haar-case flag law is the correct universal target.

Editorial extensions

If this is right

The full law of any finite collection of partial-product cokernels is distribution-free in the limit.
Conditional probabilities such as the law of $\mathrm{cok}(M_1M_2)$ given $\mathrm{cok}(M_1)$ and $\mathrm{cok}(M_2)$ have explicit Hall-Littlewood formulas valid for all nondegenerate entry distributions.
The single-matrix Cohen-Lenstra universality is recovered as the $k=1$ special case.
The flag measure provides a universal target for testing random matrix products over many entry distributions.
The moment-matching structure suggests the limit also holds for other statistics of the flag, such as the sizes of the $p$-torsion subgroups.

Reading between the lines

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

A direct computational check: for $k=2$, $p=2$, and entries uniform on $\{0,1\}$, the joint moment of $\#\mathrm{Hom}(\mathrm{cok}(M_1),\mathbb{Z}/2)$ and $\#\mathrm{Hom}(\mathrm{cok}(M_1M_2),\mathbb{Z}/2)$ should converge to the Cohen-Lenstra prediction; evaluating this numerically for moderate $n$ would test the universality claim.
The flag perspective suggests one could define an infinite random flag as a projective limit of these finite laws, though the paper only treats each $k$ separately.
A likely extension, not claimed in the abstract, would assert universality for matrices over $\mathbb{Z}/m\mathbb{Z}$ or with entries in other rings, provided a nondegeneracy condition holds.

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

Hi — quick take on arXiv:2508.10127. The headline is a genuine new result: for k random n×n integer matrices with independent entries, the flag of cokernels of partial products converges to the Cohen-Lenstra flag measure, distribution-free. That's a real advance. The abstract says the conditional formulas were previously known only for Haar matrices over Z_p; extending them to any nondegenerate entry law is exactly the kind of universality result the field wants. The proof strategy is sensible too — Sawin-Wood moment technology plus the authors' own moment computations and Huang's Haar-case computation. The combination is natural, and the citation pattern is not a red flag because the two prior papers are real and the new step is the universality.

The soft spot is the nondegeneracy condition. The stress test is on target: for a single matrix, nondegeneracy modulo every p^e is enough, but for k≥2 the mixed moments of partial-product cokernels involve the joint law of products mod p^e. Even with each factor nondegenerate, the product can sit in a linear subspace with abnormally high probability — e.g., if the factors share a common zero pattern. The abstract's 'nondegenerate' does not state a uniform control on such degeneracies. If the proof needs that control, the theorem as advertised ('any nondegenerate entry distribution') overreaches. This is not a manufactured concern; it is the standard kind of hypothesis that separates a real theorem from a plausible one.

Also, the dependences: two of the three named pillars are the authors' own prior work. That is fine if those computations are correct, but it means the referee should specifically check that the moment matching for k≥2 really reduces to Huang's Haar computation and is not silently importing quantitative assumptions.

Because this is abstract-only, I can't judge the actual proof. But the claim is important enough and the strategy coherent enough that it deserves a serious referee. I'd route it to a specialist in random matrix products or arithmetic statistics, and I'd ask the referee to write a paragraph on the nondegeneracy hypothesis — whether it is actually strong enough and whether it can be weakened or phrased clearly. Whether I cite it depends on the final version of that condition.

My recommendation: send it to peer review. Don't desk reject. If the proof is as clean as the abstract suggests, this will be a useful paper for the community; if the hypothesis needs tightening, a good referee will catch it and the paper may come back stronger.

Referee Report

2 major / 4 minor

Summary. The paper studies the joint distribution of the cokernels of the partial products M_1...M_i for independent n×n random integer matrices M_1,...,M_k, as n→∞. The main claim is that, for any 'nondegenerate' entry distribution, the resulting random flag of finite abelian p-groups converges to the Cohen-Lenstra type measure that weights a flag by 1/|Aut(flag)|. The authors also claim a corollary: the limiting conditional distribution of cok(M_1M_2) given cok(M_1) and cok(M_2) is universal and expressible via Hall-Littlewood structure constants, generalizing a previously known Haar p-adic result.

Significance. If the main theorem is correct, it is a substantial advance in the Cohen-Lenstra heuristic: it establishes distribution-free limits for flags of cokernels of products of random matrices, a richer structure than a single cokernel. The proof strategy is credible, as it builds on the Sawin-Wood moment method, the product-moment computations of Nguyen-Van Peski, and Huang's Haar computation, all of which are real and published. The paper gives a concrete falsifiable prediction (the universal limiting flag measure) and explicit conditional formulas, which is a strength. The central new claim—universality for arbitrary nondegenerate entry distributions—is not itself one of the prior inputs, so the paper's contribution is potentially significant.

major comments (2)

[Abstract] The central hypothesis 'nondegenerate entry distribution' is not defined. This is load-bearing: the moment method for the joint law of k partial products requires control over the reductions of M_1...M_i mod p^e, which are polynomial functions of the independent entries. For k≥2, nondegeneracy of each factor's entries in the usual single-matrix sense need not prevent degeneracies in the joint distribution of the partial products (e.g., the image of M_1...M_i mod p could lie in a special subspace of the image of M_1...M_{i-1} with high probability). The abstract's statement 'for any nondegenerate entry distribution' is thus ambiguous and potentially false if the proof requires a stronger, uniform condition across factors and partial products. The authors must state the precise nondegeneracy condition and prove it suffices for the product-moment matching used to identify the limit.
[Abstract] The abstract does not state a moment-matching theorem for the flag. To prove the claimed convergence, one must show that for every tuple of finite p-groups (G_1,...,G_k), the mixed moments E[∏ hom(G_i, cok(M_1...M_i))] converge to the corresponding Haar computation of Huang. The abstract mentions the Sawin-Wood technology and Nguyen-Van Peski product-moment computations, but it does not state the exact form of the moment comparison or the conditions under which it holds uniformly over all k. Without this, the reader cannot verify that the single-matrix moment method extends to the joint distribution of partial products, which is the core of the universality claim.

minor comments (4)

[Abstract] The mode of convergence is not specified. The statement 'as n→∞, this flag converges' presumably means convergence in law of the finite-dimensional distributions (for fixed prime p and fixed finite groups), but this should be stated explicitly.
[Abstract] The phrase 'any nondegenerate entry distribution' is informal. Even if a precise definition appears in the full text, the abstract should allude to the nature of the condition, e.g., 'entries not supported in a single residue class modulo every prime' or analogous.
[Abstract] The corollary about the conditional distribution of cok(M_1M_2) given cok(M_1) and cok(M_2) is vague. Please state the precise limiting statement, e.g., for fixed finite p-groups A,B,C, the probability that cok(M_1M_2)≅A given cok(M_1)≅B and cok(M_2)≅C converges to a universal constant given by a Hall-Littlewood structure constant.
[Abstract] The term 'Hall-Littlewood structure constants' is used without definition or reference. A reader unfamiliar with that combinatorics would benefit from a citation or a brief indication of the formula being used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the universal flag convergence is a new statement derived from prior moment computations via Sawin-Wood technology, not identical to any input.

full rationale

The abstract describes a moment-matching proof strategy: compute the mixed moments of partial-product cokernels for Haar p-adic matrices (Huang's prior computation), use Sawin-Wood technology and Nguyen-Van Peski-style matrix-product moment computations to show that any nondegenerate entry distribution yields the same mixed moments, and conclude universal convergence to the Cohen-Lenstra type flag measure. Each named input is a distinct mathematical result: Huang's Haar computation is the special-case computation of the target moments; Sawin-Wood provides the general moment-comparison/universality technology; and the Nguyen-Van Peski-style computations are the workhorse estimates for arbitrary entry distributions. None of these inputs is the flag-universality theorem itself, and the theorem is not defined in terms of the Haar computation; it asserts equality of the limiting law for all nondegenerate distributions. The skeptical concern about the precise nondegeneracy condition is a question of whether the stated hypothesis suffices for the moment estimates, i.e., a correctness/rigor issue, not a circularity. No equation in the available text defines the predicted limit in terms of fitted data or reduces the target claim to a self-citation. Under the hard rule that circularity must be exhibited by quotation and reduction, no circular step is identifiable from the abstract alone. Self-citations to Huang and Nguyen-Van Peski are present and load-bearing as sources of technique, but they are independent prior results with assumptions not containing the target claim, so per the rubric they do not raise the circularity score.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters: the result is a limit theorem with no fitted constants; the 1/|Aut(flag)| weights come from the Haar-case computation, not from data fitting. No invented entities: the flag of cokernels and the CL-flag measure are the objects of study, not ad hoc constructs introduced to force the conclusion. The axioms above are the three named prior technologies plus the hidden nondegeneracy hypothesis; all are inputs from outside this paper, which is what makes the circularity burden low even though two of the three named pillars are the authors' own earlier work.

assumptions (4)

domain assumption The Sawin-Wood moment method and its product-moment technology extend from single matrices to flags of k-fold matrix products.
Abstract: 'combine the general technology of Sawin-Wood'. The machinery converts moment identities into convergence of p-group distributions; its hypotheses at the flag level are not stated in the abstract.
domain assumption Huang's prior computation for Haar p-adic matrices correctly gives the limiting flag measure with weights 1/|Aut(flag)|.
Abstract: 'the computation done previously for Haar p-adic matrices by Huang'. Universality is anchored to this special case; an error there would corrupt the claimed universal limit.
ad hoc to paper The unspecified 'nondegenerate' entry distribution is exactly the hypothesis under which the moment method converges uniformly over all factors and partial products.
This is the load-bearing input hypothesis of the main theorem; its precise form is invisible from the abstract and is chosen to make the proof work.
standard math Standard linear algebra: for integer matrices A,B, cok(AB) surjects onto cok(A), so partial products form a flag of finite abelian p-groups.
Definitional background for why the partial products define a nested chain; textbook fact used implicitly.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Cohen-Lenstra flag universality for random matrix products." pith.science (2026). https://pith.science/paper/42OY6OYX

@misc{pith2026250810127,
  author       = {Pith},
  title        = {Pith review of: Cohen-Lenstra flag universality for random matrix products},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/42OY6OYX}},
  note         = {Machine review of arXiv:2508.10127}
}

abstract

For $n \times n$ random integer matrices $M_1,\ldots,M_k$, the cokernels of the partial products $\mathrm{cok}(M_1 \cdots M_i), 1 \leq i \leq k$ naturally define a random flag of abelian $p$-groups. We prove that as $n \to \infty$, this flag converges universally, for any nondegenerate entry distribution, to the Cohen-Lenstra type measure which weights each flag inversely proportional to the size of its automorphism group. As a corollary, we prove universality of certain formulas for the limiting conditional distribution of $\mathrm{cok}(M_1M_2)$ given $\mathrm{cok}(M_1),\mathrm{cok}(M_2)$ in terms of Hall-Littlewood structure constants, which were previously obtained only for Haar matrices over $\mathbb{Z}_p$. Our proofs combine the general technology of Sawin-Wood, matrix product moment computations following those of Nguyen-Van Peski, and the computation done previously for Haar $p$-adic matrices by Huang.

Discussion (0). Continue with ORCID to comment.

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.