Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Bulk Universality for Sparse Complex non-Hermitian Random Matrices

T0 review · 4 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 Bulk universality of local eigenvalue statistics holds for sparse complex non-Hermitian random matrices, matching the Ginibre ensemble.

desk verdict A substantial and likely true universality result, but the proof is conditional on a strengthened single-resolvent law that is asserted, not proved. read the letter →

arxiv 2508.03631 v1 pith:IHFXIDPK submitted 2025-08-05 math.PR math-phmath.MP

Mohammed Osman This is my paper
classification math.PRmath-phmath.MP MSC 60B20
keywords bulkuniversalitynon-Hermitianrandommatricessparselocaleigenvaluestatisticsmulti-resolventlawHermitisationGinibreensemblecircular
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

This paper proves that in the bulk of the spectrum, the local eigenvalue statistics of sparse complex random matrices are universal. For matrices whose $r$-th absolute moments decay as $N^{-1-(r-2)\epsilon}$, the $k$-point correlation functions at scale $N^{1/2}$ around any fixed $z$ in the unit disk converge to the corresponding Ginibre ensemble correlation functions with error $O(N^{-\omega})$. The model includes sparse matrices built from Bernoulli variables with mean $N^{-1+\epsilon}$, so on the order of $N^{1+\epsilon}$ nonzero entries, and a standard truncation argument extends the result to matrices with only $4+\epsilon$ moments. This matters because sparsity breaks the usual two-resolvent proofs: the paper shows that evaluating bulk statistics forces control of products of an arbitrary finite number of resolvents of the Hermitisation. The main new ingredient is a sparse multi-resolvent local law for such products, proved by combining the Zigzag flow with iterated cumulant expansions.

What carries the argument

The load-bearing object is the Hermitised resolvent chain $G_z(w_1)B_1\cdots G_z(w_m)B_m$, where $G_z(w)=(W_z-w)^{-1}$ for $W_z=\begin{pmatrix}0&X-z\\X^*-\bar z&0\end{pmatrix}$, and each $B_j$ is one of four deterministic $2N\times 2N$ matrices $E_\pm,F,F^*$ whose $N\times N$ blocks are multiples of the identity; these chains track how eigenvalue correlations respond to insertions. The paper proves a sparse multi-resolvent local law: the normalised trace deviates from the deterministic approximation $M_z$ by $O_\prec\big((\frac{1}{N\eta}+\frac{1}{q})\eta^{-(m-a/2-1)}\big)$ and entries by $O_\prec\big((\frac{1}{\sqrt{N\eta}}+\frac{1}{q})\eta^{-(m-a/2)}\big)$, where $a$ counts $F,F^*$ insertions and $\eta$ is the smallest imaginary part of the spectral parameters. The law is obtained by following the Zigzag flow, which jointly evolves the matrix and the spectral parameters, and using iterated cumulant expansions to close the equations when sparsity forbids the usual reduction inequality.

What would settle it

Measure $\langle G_z(w)\rangle - \langle M_z(w)\rangle$ for a sparse Bernoulli matrix with mean $N^{-1+\epsilon}$ at $\eta$ slightly above $N^{-1+\epsilon}$: if the deviation is of order $(N\eta)^{-1/6}+q^{-1/3}$ rather than $(N\eta)^{-1}+q^{-1}$, the assumed input fails and Theorem 1.1 has no proof. A direct numerical check of the two-point correlation function at scale $N^{-1/2}$ against GinUE would settle the theorem itself.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1.1: for any fixed $z\in\mathbb{D}$, any $k>0$, and any $f\in C_c^2(\mathbb{C}^k)$, the expectation of the $k$-point eigenvalue statistic of $X\in S_N(\epsilon)$ at scale $N^{1/2}$ around $z$ equals the corresponding Ginibre correlation integral plus $O(N^{-\omega})$. The proof proceeds in two stages. First, for deterministic matrices satisfying the multi-resolvent bounds of Definition 1.2, the Gaussian-divisible matrix $X+\sqrt{t}Y$ is shown to have universal bulk correlation functions already at $t=N^{-1+\epsilon}$ (Theorem 1.2), improving the earlier threshold $N^{-1/3+\epsilon}$. Second, matrices in the sparse class are shown to satisfy those multi-resolvent bounds with very high probability (Theorem 1.3), after which a moment-matching argument based on Girko's formula removes the Gaussian component and yields the full theorem.

Load-bearing premise

The argument depends on a strengthened single-resolvent local law for the bulk that is asserted as an extension of a known edge result ('similar arguments can be made in the bulk') and only sketched; if that sharpened error rate fails, the multi-resolvent law and the main universality theorem do not follow.

Editorial extensions

If this is right

  • For any fixed $z$ in the unit disk, the $k$-point correlation functions of sparse Bernoulli-type matrices coincide with the Ginibre correlation functions at the $N^{-1/2}$ eigenvalue scale, up to $O(N^{-\omega})$.
  • The Gaussian-divisible ensemble $X+\sqrt{t}Y$ reaches universal bulk statistics already at $t=N^{-1+\epsilon}$, matching the optimal time scale known for Hermitian matrices.
  • Matrices with a finite $4+\epsilon$ moment, without the sparse moment decay, also satisfy the universality theorem by truncation.
  • The same sparse multi-resolvent law implies that bulk eigenvector distribution results for dense non-Hermitian matrices carry over to the sparse class, as noted in the paper.
  • The moment-matching comparison of log-determinants gives an explicit polynomial-rate error between the statistics of a sparse matrix and those of its Gaussian-divisible regularization.

Reading between the lines

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

  • If the strengthened bulk single-resolvent input holds as assumed, the same machinery likely yields bulk universality for real sparse non-Hermitian matrices by adapting the partial Schur decomposition, extending known real edge results to the bulk.
  • The condition 1.2(e) can be read as uniform control of $\log\det(W_{z+N^{-1/2}x}-i\eta)$; a direct Girko-formula proof might bypass the partial Schur decomposition altogether, at the cost of controlling the same truncation errors.
  • Pushing sparsity toward bounded average degree, where $q=N^\epsilon$ is no longer available, would require new ideas since the cumulant expansion gains vanish in that regime.
  • The sparse multi-resolvent law also provides the input needed to prove asymptotic normality of linear spectral statistics for sparse non-Hermitian matrices at mesoscopic scales, paralleling dense-matrix results.
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.

Referee Report

4 major / 5 minor

Summary. The paper claims bulk local universality for complex non-Hermitian random matrices with independent entries satisfying the sparse moment condition in Definition 1.1, i.e., E|X_ij|^r <= C_r/(N q^{r-2}) with q >= N^epsilon. This includes sparse Bernoulli-type matrices and, by a truncation argument, matrices with 4+epsilon moments. The main theorem (Theorem 1.1) asserts that the k-point eigenvalue correlation functions at scale N^{1/2} around a fixed point z in the unit disk converge to the GinUE correlation functions. The proof is organized around a deterministic membership class M_N (Definition 1.2), a result for Gaussian-divisible matrices (Theorem 1.2), and a sparse multi-resolvent local law (Theorem 1.3) that shows matrices in S_N(epsilon) lie in M_N with high probability. The bulk of the paper develops the multi-resolvent local law using the zigzag method combined with iterated cumulant expansions, and the final section sketches a moment-matching argument for Theorem 1.1.

Significance. If the proof is completed, the result is a substantial advance: it extends bulk universality for non-Hermitian random matrices to sparse models and to moment conditions weaker than bounded fourth moments, and it develops multi-resolvent local laws of arbitrary finite order for sparse non-Hermitian matrices. The paper is clearly written, with a useful reduction to the deterministic conditions of M_N and a detailed account of the zigzag method in the sparse setting. However, as written the main theorem is conditional on a strengthened bulk single-resolvent law (Theorem 3.1) whose proof is not supplied. Several other load-bearing steps are only sketched. The central idea is defensible, but the current version does not yet constitute a complete proof.

major comments (4)
  1. [Section 3, Theorem 3.1, Eqs. (3.18)-(3.19)] The strengthened bulk single-resolvent law with error E_av = 1/(N eta) + 1/q and E_iso = 1/sqrt(N eta) + 1/q is the base input for the entire multi-resolvent proof, but it is not proved in this manuscript. The text acknowledges that [17, Theorem 3.1] only gives (N eta)^{-1/6} + q^{-1/3} and then asserts that 'Similar arguments can be made in the bulk'; no derivation is supplied. Appendix B proves only Proposition 3.1 and is explicitly a sketch, not a derivation of the strengthened single-resolvent law. If the stronger law is unavailable, the error rates in Theorem 3.2 and all subsequent bounds (including the final universality theorem) are not justified. This is a load-bearing gap and must be fixed before the manuscript can be accepted.
  2. [Section 3.1, Lemma 3.2] The global multi-resolvent law is used as the initial condition for the zigzag flow in Proposition 3.1 and Section 3.4, but its proof is only sketched ('we merely give a sketch'). The sketch does not fully present the induction for the averaged law, the detailed treatment of all cumulant expansion terms, or the verification that every term satisfies the claimed Psi_iso/Psi_av bounds. Since the flow propagation in Proposition 3.1 depends on the global law as input, this is a load-bearing step that needs a complete proof.
  3. [Section 2, Lemma 2.2] The transfer of membership in M_N from X^{(i-1)} to X^{(i)} is proved only for Re w_j = 0 and |Im w_j| approx t; the final paragraph states that the arguments 'can be easily extended' to |Im w_j| >= N^{-1+epsilon} and |Re w_j| <= delta |Im w_j|. This extension is necessary for the induction in Theorem 1.2 (the events E_i with n_i = n_{i-1}/2 - 1) and for the eventual conclusion that X in M_N holds uniformly on D(delta, tau). The details are not supplied and the claimed extension is not immediate, especially for the quadratic form bounds in Lemma 2.3.
  4. [Section 4, proof of Theorem 1.1] The moment-matching argument for the main theorem is only sketched. In particular, the bound (4.4) is justified by a single line 'by the local law |G_ij| ≺ 1 ∨ (N eta)^{-1}' without exhibiting the expansion to the required power p, the control of the remainder, or the dependence on the truncation of the cumulant expansion. The restriction of the eta-integral to [N^{-1-xi}, T] uses a stated generalization of [2, Theorem 2.2] to complex matrices that is asserted but not proved. These steps are standard in the literature, but since they are used to establish Theorem 1.1, the paper should either provide the full argument or give precise references that cover the complex, non-centered-z case.
minor comments (5)
  1. [Abstract and Definition 1.1] The word 'Bernouilli' is misspelled; it should be 'Bernoulli'.
  2. [Definition 1.2] There is a typo: 'redudancy' should be 'redundancy'.
  3. [Section 2, before Lemma 2.1] The word 'preceeding' should be 'preceding' in the sentence 'the displayed equation immediately preceeding [24, Lemma 7.3]'.
  4. [Section 3.3, proof of Lemma 3.5, case ii] There is a typo: 'multliple' should be 'multiple', and 'a mutliple' should be 'a multiple'.
  5. [Notation and references] The domain D(delta, tau) is defined twice, once in the introduction and once in Section 3, with identical content; a single definition would avoid possible confusion. Also, the paper cites [2] for a result on least singular values but the stated generalization to complex matrices and fixed z deserves a remark in the text or a reference where it is proved.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the sparse multi-resolvent local law and the bulk universality theorem are derived from independent inputs; the unproved strengthening of the single-resolvent law is a correctness gap, not a circular step.

full rationale

The claimed derivation chain is not circular. Theorem 1.1 follows from Theorem 1.2 (a deterministic condition on resolvent chains) and Theorem 1.3 (a sparse multi-resolvent local law), combined by a standard moment-matching argument in Section 4. Theorem 1.3 is proved in Section 3 from independent inputs: the single-resolvent law of He [17], the deterministic approximation M_z of Cipolloni-Erdős-Henheik-Schröder [5], and the zigzag/cumulant-expansion machinery of Cipolloni-Erdős-Schröder [8,9]. No fitted parameter is renamed as a prediction, and no quantity is defined in terms of the target statistic. The one suspicious passage is the assertion after Theorem 3.1 that the stronger bulk single-resolvent bounds (3.18)-(3.19) follow from He's weaker bound by 'similar arguments' or by the Appendix B sketch. This is a missing derivation and a correctness risk, not a circularity: the stronger bound is a genuinely stronger statement, not the same statement repackaged, and the appendix at least indicates a route from the weak input (B.1). Self-citations, especially to [24] by Maltsev and Osman, supply the partial Schur decomposition and the induction framework for Theorem 1.2, but [24] proves a different regime (i.i.d. bounded moments, t = N^{-1/3+epsilon}); the novel sparse multi-resolvent law is independent content, so the self-citation is not the load-bearing step that forces the final result. Condition (e) of Definition 1.2 is an assumption in Theorem 1.2 and a conclusion for random matrices via Theorem 1.3; it is not presupposed in the derivation. Thus the central claims do not reduce to their inputs by construction.

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

The central claim rests on the model class S_N(epsilon), imported single resolvent and deterministic approximation bounds, and external circular-law and singular-value estimates. No free parameters are fitted to data and no entities are invented; epsilon is a theorem hypothesis rather than an ad hoc fitted constant.

assumptions (5)
  • domain assumption The model class S_N(epsilon): independent entries with zero mean, zero second moment, variance 1/N, and r-th moment bound C_r/(N q^{r-2}) with q = N^epsilon.
    All theorems are conditional on this entry model; the sparse and truncation examples are meant to satisfy it.
  • domain assumption The strengthened sparse single resolvent law in Theorem 3.1 with errors E^av and E^iso in the bulk, quoted beyond the published statement of [17, Theorem 3.1].
    The text after Theorem 3.1 admits [17] has weaker error and says the stronger bulk version follows by similar arguments; no proof is included.
  • domain assumption The deterministic multi-resolvent approximation M_z and its bounds from [5, Definition 4.1, Lemma 4.3].
    Lemma 3.1 relies on these bounds; they are imported from prior work without reproof.
  • standard math Known lower bounds on least singular values and the circular law used in the Girko eta-truncation in Section 4 ([2, Theorem 2.2] and [31, Theorem 2.5]).
    Used to restrict the eta-integral and to compare eigenvalue statistics; these are accepted external results.
  • standard math Girko's formula and the partial Schur decomposition representation for k-point functions ([24, Corollary 4.2]).
    Section 2 and Section 4 use these as the bridge between resolvent estimates and correlation functions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Bulk Universality for Sparse Complex non-Hermitian Random Matrices." pith.science (2026). https://pith.science/paper/IHFXIDPK

@misc{pith2026250803631,
  author       = {Pith},
  title        = {Pith review of: Bulk Universality for Sparse Complex non-Hermitian Random Matrices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IHFXIDPK}},
  note         = {Machine review of arXiv:2508.03631}
}
abstract

We prove that the local eigenvalue statistics in the bulk for complex random matrices with independent entries whose $r$-th absolute moment decays as $N^{-1-(r-2)\epsilon}$ for some $\epsilon>0$ are universal. This includes sparse matrices whose entries are the product of a Bernouilli random variable with mean $N^{-1+\epsilon}$ and an independent complex-valued random variable. By a standard truncation argument, we can also conclude universality for complex random matrices with $4+\epsilon$ moments. The main ingredient is a sparse multi-resolvent local law for products involving any finite number of resolvents of the Hermitisation and deterministic $2N\times2N$ matrices whose $N\times N$ blocks are multiples of the identity.

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. On a Rosenzweig-Porter-type model

    math-ph 2026-07 unverdicted novelty 8.0 of 10

    Provides uniform local laws and localization analysis for the general Rosenzweig-Porter model H = H0 + λW, generalizing previous results on deformed Wigner matrices.

  2. On a Rosenzweig-Porter-type model

    math-ph 2026-07 accept novelty 7.0 of 10

    Uniform single- and two-resolvent local laws for arbitrary deformations of Wigner matrices yield ETH and localization profiles for all λ and H0.

Reference graph

Works this paper leans on

31 extracted references · 24 canonical work pages · cited by 1 Pith paper

  1. [17]

    Yukun He, Edge Universality of Sparse Erd˝ os–R´ enyi Digraphs, arXiv:2304.04723

    arXiv

  2. [5]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os, Joscha Henheik and Dominik Schr¨ oder, Optimal Lower Bound on Eigenvector Overlaps for non-Hermitian Random Matrices, Journal of Functional Analysis, 287 (4), 2024

    work page 2024

  3. [8]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Dominik Schr¨ oder, Optimal Multi-Resolvent Local Laws for Wigner Matrices, Electron. J. Probab. 27, 1-38, 2022

    work page 2022

  4. [9]

    Theory Relat

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Dominik Schr¨ oder, Mesoscopic Central Limit Theorem for non-Hermitian Random Matrices, Probab. Theory Relat. Fields, 188 (3), 2024

    work page 2024

  5. [1]

    Zhigang Bao, Giorgio Cipolloni, L´ aszl´ o Erd˝ os, Joscha Henheik and Oleksii Kolupaiev, Law of Fractional Logarithm for Random Matrices, arXiv:2503.18922

    work page arXiv

  6. [2]

    Ziliang Che and Patrick Lopatto, Universality of the Least Singular Values for Sparse Random Matrices, Electron. J. Probab. 24, 1-53, 2019

    work page 2019

  7. [3]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Joscha Henheik, Eigenstate Thermalisation at the Edge for Wigner Matrices, arXiv:2309.05488

    arXiv

  8. [4]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os, Joscha Henheik and Oleksii Kolupaiev, Eigenvector Decor- relation for Random Matrices, arXiv:2410.10718

    arXiv

Show all 31 references
  1. [6]

    Theory Relat

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Dominik Schr¨ oder, Edge Universality for non- Hermitian Random Matrices, Probab. Theory Relat. Fields 179, 1-28, 2020

    2020

  2. [7]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Dominik Schr¨ oder, Eigenstate Thermalization Hypoth- esis for Wigner Matrices, Commun. Math. Phys. 388, 1005-1048, 2021

    2021

  3. [10]

    Giorgio Cipolloni, L´ aszl´ o Erd˝ os and Yuanyuan Xu, Universality of Extremal Eigenvalues of Large Random Matrices, arXiv:2312.08325

  4. [11]

    Sophiia Dubova and Kevin Yang, Bulk Universality for Complex Eigenvalues of Real non- Symmetric Random Matrices with IID Entries, arXiv:2402.10197 44

    arXiv

  5. [12]

    Sophiia Dubova, Kevin Yang, Horng-Tzer Yau and Jun Yin, Gaussian statistics for left and right eigenvectors of complex non-Hermitian matrices, arXiv:2403.19644

    arXiv

  6. [13]

    L´ aszl´ o Erd˝ os, Antti Knowles, Horng-Tzer Yau and Jun Yin, Spectral Statistics of Erd˝ os– R´ enyi Graphs II: Eigenvalue Spacing and the Extreme Eigenvalues, Commun. Math. Phys. 314, 587-640, 2012

    2012

  7. [14]

    L´ aszl´ o Erd˝ os, Antti Knowles, Horng-Tzer Yau and Jun Yin, The local semicircle law for a general class of random matrices, Electron. J. Probab. 18, 1-58, 2013

    2013

  8. [15]

    L´ aszl´ o Erd˝ os and Volodymyr Riabov, The Zigzag Strategy for Random Band Matrices, arXiv:2506.06441

    arXiv

  9. [16]

    Friedrich G¨ otze and Alexander Tikhomirov, The Circular Law for Random Matrices, An- nals of Probability 38 (4), 1444-1491, 2010

    2010

  10. [18]

    Yukun He and Antti Knowles, Mesoscopic Eigenvalue Statistics of Wigner Matrices, Annals of Applied Probability 27 (3), 2017

    2017

  11. [19]

    Theory Relat

    Yukun He and Antti Knowles, Fluctuations of extreme eigenvalues of sparse Erd˝ os–R´ enyi graphs, Probab. Theory Relat. Fields 180: 985-1056, 2021

    2021

  12. [20]

    Jiaoyang Huang, Benjamin Landon and Horng-Tzer Yau, Transition from Tracy–Widom to Gaussian fluctuations of extremal eigenvalues of sparse Erd˝ os–R´ enyi graphs, Annals of Probability 48 (2), 916-962, 202

  13. [21]

    Benjamin Landon, Philippe Sosoe and Horng-Tzer Yau, Fixed Energy Universality of Dyson Brownian Motion, Advances in Mathematics 346, 1137-1332, 2019

    2019

  14. [22]

    Jaehun Lee, Higher Order Fluctuations of Extremal Eigenvalues of Sparse Random Matri- ces, Ann. Inst. Henri Poincar´ e Probab. Stat. 60, 2694-2735, 2024

    2024

  15. [23]

    Theory Relat

    Ji Oon Lee and Kevin Schnelli, Local Law and Tracy–Widom Limit for Sparse Random Matrices, Probab. Theory Relat. Fields, 171 (1), 2018

    2018

  16. [24]

    Theory Relat

    Anna Maltsev and Mohammed Osman, Bulk Universality for Complex non-Hermitian Ma- trices with Independent and Identically Distributed Entries, Probab. Theory Relat. Fields, 2024

    2024

  17. [25]

    Mohammed Osman, Least non-zero singular value and the distribution of eigenvectors of non-Hermitian random matrices, arXiv:2404.01149

    arXiv

  18. [26]

    Mohammed Osman, Universality for diagonal eigenvector overlaps of non-Hermitian ran- dom matrices, arXiv:2409.16144

    arXiv

  19. [27]

    Mohammed Osman, Bulk Universality for Real Matrices with Independent and Identically Distributed Entries, Electron. J. Probab. 30, 1-66, 2025

    2025

  20. [28]

    Mark Rudelson and Konstantin Tikhomirov, The Sparse Circular Law Under Minimal Assumptions, Geom. Funct. Anal. 29, 561-637, 2019

    2019

  21. [29]

    Ashwin Sah, Julian Sahasrabudhe and Mehtaab Sawhney, The Limiting Spectral Law for Sparse IID Matrices, arXiv:2310.17635

    arXiv

  22. [30]

    Kevin Schnelli and Yuanyuan Xu, Convergence Rate to Tracy–Widom Laws for the Largest Eigenvalue of Wigner Matrices, Commun. Math. Phys. 393 (2), 839-907, 2022

    2022

  23. [31]

    Terence Tao and Van Vu, Random Matrices: the Circular Law, Commun. Contemp. Math. 10 (2), 261-307, 2008. 45

    2008

Pith tools

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