betaFeedback

WorksMathematics

Universality of Brown measure convergence for non-Gaussian matrix polynomials with adjoints

Abstract

Let PP be a noncommutative polynomial of arbitrary degree. We prove that the empirical spectral distribution of PP evaluated at independent Wigner matrices converges weakly in probability to its Brown measure limit. The same conclusion holds for arbitrary ∗*-polynomials in independent iid matrices and their adjoints. The entry laws have bounded real or joint planar densities and a finite (2+ε)(2+\varepsilon) moment; only second moments are required on the Wigner diagonal. The real and imaginary parts of an entry may be dependent, and its pseudovariance need not vanish. The proof develops the coefficient-algebra approach of the author's earlier work on matrix polynomials with Gaussian and non-Gaussian iid inputs \cite{Han}. The main step is a polynomial lower-tail estimate, under only second moments, for the least singular value of a tensor pencil with arbitrary fixed matrix coefficients. We use a logarithmic determinant method: a real-block small-ball argument shows that logarithmic derivatives of the determinant control the inverse pencil. Relative polynomial anti-concentration then gives the lower-tail bound. We establish convergence of logarithmic potentials by Hermitian resolvent comparison and rank truncation. This extends the recent Brown measure convergence theorem of Collins, Kitagawa, Miyagawa, Yuan and Zhong \cite{Collins} for GUE polynomials to non-Gaussian Wigner inputs.

Introduction

For a matrix M∈Md(C)M \in M_d(\mathbb{C}) with eigenvalues λ1(M),…,λd(M)\lambda_1(M),\ldots,\lambda_d(M), counted with algebraic multiplicity, its empirical spectral distribution is the probability measure

μM=1d∑j=1dδλj(M).\mu_M = \frac{1}{d}\sum_{j=1}^{d}\delta_{\lambda_j(M)}.

A central problem in random matrix theory is to determine the limit of μMN\mu_{M_N} for a sequence of random matrices MNM_N. For Hermitian matrices this can often be done by computing traces of powers. For nonnormal matrices, convergence of the ∗*-moments alone does not determine the limiting eigenvalue measure. Girko’s Hermitization method [13] addresses this problem through the singular values of MN−zIM_N-zI. In particular, it requires control of their logarithms near zero. The finite-variance circular law and its replacement principle provide a basic example of this method [22]; we refer to [7] for a systematic account.

We study fixed noncommutative polynomials in independent iid matrices and their adjoints, as well as polynomials in independent Wigner matrices. For iid inputs, the entries of the polynomial need not be independent. In both models, we prove convergence of the empirical spectral distribution to the corresponding Brown measure under bounded densities and a finite (2+ε)(2+\varepsilon) moment. The diagonal laws in the Wigner model need only a second moment. We prove this using a least singular value estimate for tensor pencils with arbitrary fixed coefficients. This estimate requires only second moments.

Brown measure and notation

For M∈Md(C)M \in M_d(\mathbb{C}), set

UM(z)=1dlog⁡∣det⁡(M−zId)∣.U_M(z)=\frac{1}{d}\log\left|\det(M-zI_d)\right|.

Then μM=(2π)−1ΔUM\mu_M=(2\pi)^{-1}\Delta U_M in the sense of distributions. The corresponding measure for an operator in a tracial von Neumann algebra is the Brown measure [8], defined through the Fuglede–Kadison determinant [12]. More precisely, let (M,τ)(\mathcal{M},\tau) be a tracial von Neumann algebra with faithful normalized trace. For a∈Ma \in\mathcal{M}, write

Ua(z)=τ(log⁡∣a−z∣),μa=12πΔUa.U_a(z)=\tau(\log|a-z|),\qquad\mu_a=\frac{1}{2\pi}\Delta U_a.

The logarithm is defined by decreasing positive regularization. Its value at a particular point may initially be −∞-\infty, whereas the potential is locally integrable as a function of zz. We use standard free circular variables c1,…,cnc_1,\ldots,c_n, normalized by τ(cℓcℓ∗)=1\tau(c_\ell c_\ell^*)=1, and standard free semicircular variables s1,…,sns_1,\ldots,s_n, normalized by τ(sℓ2)=1\tau(s_\ell^2)=1. For free independence and circular and semicircular elements, see [25, 18].

All polynomials, coefficient dimensions and entry laws are fixed as N→∞N\to\infty. Superscript T denotes ordinary transpose and * denotes complex adjoint. A *-polynomial may involve the variables and their formal adjoints, with arbitrary complex coefficients. We write Tr⁡\operatorname{Tr} for the ordinary trace and tr⁡d=d−1Tr⁡\operatorname{tr}_d=d^{-1}\operatorname{Tr} for the normalized trace. For a linear map AA, its least singular value means

smin⁡(A)=inf⁡∥v∥=1∥Av∥,s_{\min}(A)=\inf_{\|v\|=1}\|Av\|,

where the infimum is over its domain. Singular values are otherwise ordered decreasingly and counted with multiplicity. Constants may depend on the fixed polynomial, coefficients and laws. We impose no coupling between the matrices at different sizes.

Independent matrices and their adjoints

We consider a centered entry law ξ\xi with unit absolute variance:

Eξ=0,E∣ξ∣2=1.(1)\mathbb{E}\xi=0,\qquad\mathbb{E}|\xi|^2=1. \tag*{(1)}

The density assumption takes one of the following two forms:

  • (A) ξ\xi is real and has a density bounded by MM with respect to Lebesgue measure on R\mathbb{R};

  • (B) ξ\xi is complex and has a joint density bounded by MM with respect to Lebesgue measure on C=R2\mathbb{C}=\mathbb{R}^2.

In class (B), the real and imaginary parts may be dependent. In particular, the quantity ρ=Eξ2\rho=\mathbb{E}\xi^2 need not vanish. Let ξℓ,ij\xi_{\ell,ij} be independent copies of ξ\xi and define

Xℓ=Xℓ(N)=1N(ξℓ,ij)i,j=1N,1≤ℓ≤n.X_\ell=X_\ell^{(N)}=\frac{1}{\sqrt{N}}(\xi_{\ell,ij})_{i,j=1}^N,\qquad1\leq\ell\leq n.

Theorem 1.1 (Brown measure convergence). *Assume either class (A) or class (B), and suppose that E∣ξ∣2+ε<∞\mathbb{E}|\xi|^{2+\varepsilon}<\infty for some ε>0\varepsilon>0. Let PP be a fixed scalar-coefficient noncommutative -polynomial. Then

μP(X,X∗)→PμP(c,c∗)\mu_{P(X,X^*)}\xrightarrow{\mathbb{P}}\mu_{P(c,c^*)}

weakly, where c1,…,cnc_1,\ldots,c_n is a standard free circular family. For every fixed z≠P(0)z\ne P(0),

1Nlog⁡∣det⁡(P(X,X∗)−zIN)∣→Pτ(log⁡∣P(c,c∗)−z∣).\frac{1}{N}\log\left|\det(P(X,X^*)-zI_N)\right|\xrightarrow{\mathbb{P}}\tau\left(\log|P(c,c^*)-z|\right).

Moreover, the logarithmic potentials converge locally in planar L1L^{1} in probability: for every compact K⊂CK \subset\mathbb{C},

∫K∣UP(X,X∗)(z)−UP(c,c∗)(z)∣ d2z→P0.\int_K \left|U_{P(\mathbf{X},\mathbf{X}^*)}(z)-U_{P(c,c^*)}(z)\right|\,\mathrm{d}^{2}z \xrightarrow{\mathbb{P}} 0.

Thus the same conclusion holds for real entry laws and arbitrary complex polynomial coefficients, and for planar laws with arbitrary pseudovariance ρ\rho. The least singular value theorem used to prove this result has a weaker moment assumption. To state it, fix Aℓ,Bℓ∈Mm(C)A_\ell,B_\ell\in M_m(\mathbb{C}) and consider

TN=Im⊗IN+∑ℓ=1n(Aℓ⊗Xℓ+Bℓ⊗Xℓ∗).(2)T_N=I_m\otimes I_N+\sum_{\ell=1}^{n}\left(A_\ell\otimes X_\ell+B_\ell\otimes X_\ell^*\right). \tag*{(2)}

We shall work on the Hilbert–Schmidt event

ΩN={max⁡ℓN−1∥Xℓ∥HS2≤2},qN=P(ΩNc),ω={1,class (A),12,class (B).(3)\Omega_N=\left\{\max_\ell N^{-1}\lVert X_\ell\rVert_{\mathrm{HS}}^2\le2\right\},\qquad q_N=\mathbb{P}(\Omega_N^c),\qquad\omega=\begin{cases}1,&\text{class (A)},\\ \frac{1}{2},&\text{class (B)}.\end{cases} \tag*{(3)}

Since the entries have a second moment, the weak law of large numbers gives qN→0q_N\to0.

Theorem 1.2 (Least singular values of tensor pencils). Under {(xref)}(1.1) and either density assumption (A) or (B), there are C,K,c>0C,K,c>0 and N0∈NN_0\in\mathbb{N}, depending on the fixed coefficients and law, such that

P{smin⁡(TN)≤t}≤CNKtω/rcomp+Ce−cN+qN,N≥N0,t>0.(4)\mathbb{P}\{s_{\min}(T_N)\le t\}\le CN^{K}t^{\omega/r_{\mathrm{comp}}}+Ce^{-cN}+q_N,\qquad N\ge N_0,\quad t>0. \tag*{(4)}

Here rcomp≤mr_{\mathrm{comp}}\le m is the length of any complex composition series of Cm\mathbb{C}^{m} for the algebra generated by I,Aℓ,BℓI,A_\ell,B_\ell. For a nonzero irreducible coefficient family one may take rcomp=1r_{\mathrm{comp}}=1; one may always replace rcompr_{\mathrm{comp}} by mm. If E∣ξ∣2+ε<∞\mathbb{E}|\xi|^{2+\varepsilon}<\infty, then qN≤CN−min⁡(ε,2)q_N\le CN^{-\min(\varepsilon,2)}. For a real Gaussian law, or a centered complex Gaussian law with positive definite real covariance, one may take ω=1\omega=1 and replace the two additive errors by Ce−cNCe^{-cN}.

The estimate is uniform in tt, with constants allowed to depend on the fixed coefficient family. The denominator rcompr_{\mathrm{comp}} arises from reconstructing an upper triangular pencil from its irreducible diagonal blocks.

Linearizing P−zIP-zI gives a tensor pencil to which Theorem 1.2 applies. Its constant matrix is invertible for z≠P(0)z\ne P(0).

Corollary 1.3 (Least singular values of polynomial shifts). Let PP be a fixed scalar-coefficient ∗*-polynomial, and assume {(xref)}(1.1) and either (A) or (B). For each fixed z≠P(0)z\ne P(0), there are Cz,az,bz,cz>0C_z,a_z,b_z,c_z>0 and $N_0(z) such that

P{smin⁡(P(X,X∗)−zIN)≤t}≤CzNaztbz+Cze−czN+qN,N≥N0(z),t>0.\mathbb{P}\{s_{\min}(P(X,X^*)-zI_N)\le t\}\le C_zN^{a_z}t^{b_z}+C_ze^{-c_zN}+q_N,\qquad N\ge N_0(z),\quad t>0.

In particular, for some fixed Bz<∞B_z<\infty,

P{smin⁡(P(X,X∗)−zIN)≥N−Bz}⟶1.\mathbb{P}\{s_{\min}(P(X,X^*)-zI_N)\ge N^{-B_z}\}\longrightarrow1.

The constants may depend on zz. The lower cutoff for each fixed z≠P(0)z\ne P(0) suffices for Hermitization: the logarithmic potentials are integrated over zz, and the proof does not require uniformity in zz.

Wigner matrices

For each 1≤ℓ≤n1 \le\ell\le n, let Hℓ=Hℓ(N)H_\ell= H_\ell^{(N)} be Hermitian with entries

References

  1. [1]Johannes Alt, László Erdős, Torben Krüger, and Yuriy Nemish. “Location of the spectrum of Kronecker random matrices”. In: Annales de l’Institut Henri Poincaré, Probabilités et Statistiques 55.2 (2019), pp. 661–696. DOI: 10.1214/18-AIHP894. arXiv: 1706.08343.DOI
  2. [2]Greg W. Anderson. “Convergence of the largest singular value of a polynomial in independent Wigner matrices”. In: The Annals of Probability 41.3B (2013), pp. 2103–2181. DOI: 10.1214/11-A0P739.DOI
  3. [3]Greg W. Anderson, Alice Guionnet, and Ofer Zeitouni. An Introduction to Random Matrices. Vol. 118. Cambridge Studies in Advanced Mathematics. Cambridge: Cambridge University Press, 2010. DOI: 10.1017/CBO9780511801334.DOI
  4. [4]Serban T. Belinschi, Hari Bercovici, and Mireille Capitaine. “On the outlying eigenvalues of a polynomial in large independent random matrices”. In: International Mathematics Research Notices 2021.4 (2021), pp. 2588–2641. DOI: 10.1093/imrn/rnz080.
  5. [5]Serban T. Belinschi, Tobias Mai, and Roland Speicher. “Analytic subordination theory of operator-valued free additive convolution and the solution of a general random matrix problem”. In: Journal für die reine und angewandte Mathematik 732 (2017), pp. 21–53. DOI: 10.1515/crelle-2014-0138.
  6. [6]Serban T. Belinschi, Piotr Śniady, and Roland Speicher. “Eigenvalues of non-Hermitian random matrices and Brown measure of non-normal operators: Hermitian reduction and linearization method”. In: Linear Algebra and its Applications 537 (2018), pp. 48–83. DOI: 10.1016/j.laa.2017.09.024. arXiv: 1506.02017.DOI
  7. [7]Charles Bordenave and Djalil Chafaï. “Around the circular law”. In: Probability Surveys 9 (2012), pp. 1–89. DOI: 10.1214/11-PS183.DOI
  8. [8]Lawrence G. Brown. “Lidskii’s theorem in the type II case”. In: Geometric Methods in Operator Algebras (Kyoto, 1983). Ed. by Huzihiro Araki and Edward G. Effros. Pitman Research Notes in Mathematics Series 123. Harlow: Longman Scientific & Technical, 1986, pp. 1–35.
  9. [9]Benoit Collins, Yu Kitagawa, Akihiro Miyagawa, Wangjun Yuan, and Ping Zhong. Brown measures and spectral convergence for arbitrary GUE polynomials. 2026. arXiv: 2611.12785.arxiv.org/abs/2611.12785
  10. [10]Nicholas A. Cook, Alice Guionnet, and Jonathan Husson. “Spectrum and pseudospectrum for quadratic polynomials in Ginibre matrices”. In: Annales de l’Institut Henri Poincaré, Probabilités et Statistiques 58.4 (2022), pp. 2284–2320. DOI: 10.1214/21-AIHP1225. arXiv: 2008.08833.DOI
  11. [11]Brendan Farrell and Roman Vershynin. “Smoothed analysis of symmetric random matrices with continuous distributions”. In: Proceedings of the American Mathematical Society 144.5 (2016), pp. 2257–2261. DOI: 10.1090/proc/12844.DOI
  12. [12]Bent Fuglede and Richard V. Kadison. “Determinant theory in finite factors”. In: Annals of Mathematics. Second Series 55.3 (1952), pp. 520–530. DOI: 10.2307/1969645.DOI
  13. [13]Vyacheslav L. Girko. “Circular law”. In: Theory of Probability & Its Applications 29.4 (1985), pp. 694–706. DOI: 10.1137/1129095.DOI
  14. [14]Uffe Haagerup and Steen Thorbjørnsen. “A new application of random matrices: Ext(C*_{red}(F_2)) is not a group”. In: Annals of Mathematics. Second Series 162.2 (2005), pp. 711–775. DOI: 10.4007/annals.2005.162.711.DOI
  15. [15]Yi Han. Brown measure convergence for the spectrum of polynomials in Ginibre matrices. 2026. arXiv: 2606.01664.arxiv.org/abs/2606.01664
  16. [16]J. William Helton, Tobias Mai, and Roland Speicher. “Applications of realizations (aka linearizations) to free probability”. In: Journal of Functional Analysis 274.1 (2018), pp. 1–79. DOI: 10.1016/j.jfa.2017.10.003.DOI
  17. [17]Camille Male. “The norm of polynomials in large random and deterministic matrices”. In: Probability Theory and Related Fields 154 (2012). With an appendix by Dimitri Shlyakhtenko, pp. 477–532. DOI: 10.1007/s00440-011-0375-2.DOI
  18. [18]Alexandru Nica and Roland Speicher. Lectures on the Combinatorics of Free Probability. Vol. 335. London Mathematical Society Lecture Note Series. Cambridge: Cambridge University Press, 2006. DOI: 10.1017/CBO9780511735127.DOI
  19. [19]Mark Rudelson and Roman Vershynin. “The Littlewood–Offord problem and invertibility of random matrices”. In: Advances in Mathematics 218.2 (2008), pp. 600–633. DOI: 10.1016/j.aim.2008.01.010.DOI
  20. [20]Mark Rudelson and Roman Vershynin. “Smallest singular value of a random rectangular matrix”. In: Communications on Pure and Applied Mathematics 62.12 (2009), pp. 1707–1739. DOI: 10.1002/cpa.20294.DOI
  21. [21]Dimitri Shlyakhtenko and Paul Skoufranis. “Freely independent random variables with non-atomic distributions”. In: Transactions of the American Mathematical Society 367.9 (2015), pp. 6267–6291. DOI: 10.1090/S0002-9947-2015-06434-4. arXiv: 1305.1920.DOI
  22. [22]Terence Tao and Van Vu. “Random matrices: Universality of ESDs and the circular law”. In: The Annals of Probability 38.5 (2010). With an appendix by Manjunath Krishnapur, pp. 2023–2065. DOI: 10.1214/10-AOP534.DOI
  23. [23]Roman Vershynin. “Invertibility of symmetric random matrices”. In: Random Structures & Algorithms 44.2 (2014), pp. 135–182. DOI: 10.1002/rsa.20429.
  24. [24]Dan Voiculescu. “Limit laws for random matrices and free products”. In: Inventiones Mathematicae 104.1 (1991), pp. 201–220. DOI: 10.1007/BF01245072.
  25. [25]Dan-Virgil Voiculescu, Kenneth J. Dykema, and Alexandru Nica. Free Random Variables. Vol. 1. CRM Monograph Series. Providence, RI: American Mathematical Society, 1992. DOI: 10.1090/crmm/001.

Paper details

vlnc:W_rt6mmxbx
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents