Abstract
We investigate to what extent the microscopic Wigner–Gaudin–Mehta–Dyson (WGMD) (or sine kernel) statistics of random matrix theory remain valid on mesoscopic scales. To that end, we compute the connected two-point spectral correlation function of a Wigner matrix at two mesoscopically separated points. In the mesoscopic regime, density correlations are much weaker than in the microscopic regime. Our result is an explicit formula for the two-point function. This formula implies that the WGMD statistics are valid to leading order on all mesoscopic scales, that in the real symmetric case there are subleading corrections matching precisely the WGMD statistics, while in the complex Hermitian case these subleading corrections are absent. We also uncover non-universal subleading correlations, which dominate over the universal ones beyond a certain intermediate mesoscopic scale. The proof is based on a hierarchy of Schwinger–Dyson equations for a sufficiently large class of polynomials in the entries of the Green function. The hierarchy is indexed by a tree, whose depth is controlled using stopping rules. A key ingredient in the derivation of the stopping rules is a new estimate on the density of states, which we prove to have bounded derivatives of all order on all mesoscopic scales.
Similar content being viewed by others
Notes
As customary, to simplify notation somewhat, we regard \(\rho _1\) and \(\rho _2\) as functions (densities with respect to Lebesgue measure) although the right-hand sides of (1.1) are measures. They will always be integrated against continuous test functions, justifying such an abuse of notation.
We remark that the condition on the variances of the diagonal entries is imposed for convenience, and it can be easily relaxed; see “Appendix C” for details.
Here \(\mathrm {o}(1)\) means for any fixed (small) \(\varepsilon >0\).
References
Aggarwal, A.: Bulk universality for generalized Wigner matrices with few moments, Preprint arXiv:1612.00421 (2016)
Barbour, A.D.: Asymptotic expansions based on smooth functions in the central limit theorem. Prob. Theor. Rel. Fields 72, 289–303 (1986)
Benaych-Georges, F., Knowles, A.: Lectures on the local semicircle law for Wigner matrices, Preprint arXiv:1601.04055 (2016)
Bourgade, P., Erdős, L., Yau, H.-T., Yin, J.: Fixed energy universality for generalized Wigner matrices. Commun. Pure Appl. Math. 69, 1815–1881 (2016)
Boutet de Monvel, A., Khorunzhy, A.: Asymptotic distribution of smoothed eigenvalue density. I. Gaussian random matrices. Random Oper. Stoch. Equ. 7, 1–22 (1999)
Boutet de Monvel, A., Khorunzhy, A.: Asymptotic distribution of smoothed eigenvalue density. II. Wigner random matrices. Random Oper. Stoch. Equ. 7, 149–168 (1999)
Dyson, F.J.: A Brownian-motion model for the eigenvalues of a random matrix. J. Math. Phys. 3, 1191–1198 (1962)
Erdős, L., Krüger, T., Schröder, D.: Random matrices with slow correlation decay, Preprint arXiv:1705.10661 (2017)
Erdős, L., Knowles, A., Yau, H.-T., Yin, J.: Spectral statistics of Erdős–Rényi graphs II: eigenvalue spacing and the extreme eigenvalues. Commun. Math. Phys. 314, 587–640 (2012)
Erdős, L., Knowles, A., Yau, H.-T., Yin, J.: The local semicircle law for a general class of random matrices. Electron. J. Probab. 18, 1–58 (2013)
Erdős, L., Knowles, A., Yau, H.-T., Yin, J.: Spectral statistics of Erdős–Rényi graphs I: local semicircle law. Ann. Probab. 41, 2279–2375 (2013)
Erdős, L., Péché, S., Ramirez, J.A., Schlein, B., Yau, H.-T.: Bulk universality for Wigner matrices. Commun. Pure Appl. Math. 63, 895–925 (2010)
Erdős, L., Ramirez, J., Schlein, B., Tao, T., Vu, V., Yau, H.-T.: Bulk universality for Wigner hermitian matrices with subexponential decay. Math. Res. Lett. 17, 667–674 (2010)
Erdős, L., Ramirez, J., Schlein, B., Yau, H.-T.: Universality of sine-kernel for Wigner matrices with a small Gaussian perturbation. Electr. J. Probab. 15, 526–604 (2010)
Erdős, L., Schlein, B., Yau, H.-T.: Local semicircle law and complete delocalization for Wigner random matrices. Commun. Math. Phys. 287, 641–655 (2009)
Erdős, L., Schlein, B., Yau, H.-T.: Semicircle law on short scales and delocalization of eigenvectors for Wigner random matrices. Ann. Probab. 37, 815–852 (2009)
Erdős, L., Schlein, B., Yau, H.-T.: Wegner estimate and level repulsion for Wigner random matrices. Int. Math. Res. Not. 2009, 436–479 (2010)
Erdős, L., Schlein, B., Yau, H.-T.: Universality of random matrices and local relaxation flow. Invent. Math. 185(1), 75–119 (2011)
Erdős, L., Schlein, B., Yau, H.-T., Yin, J.: The local relaxation flow approach to universality of the local statistics of random matrices. Ann. Inst. Poincaré (B) 48, 1–46 (2012)
Erdős, L., Yau, H.-T., Yin, J.: Universality for generalized Wigner matrices with Bernoulli distribution. J. Comb. 1(2), 15–85 (2011)
Erdős, L., Yau, H.-T., Yin, J.: Bulk universality for generalized Wigner matrices. Prob. Theor. Rel. Fields 154, 341–407 (2012)
Erdős, L., Yau, H.-T., Yin, J.: Rigidity of eigenvalues of generalized Wigner matrices. Adv. Math. 229, 1435–1515 (2012)
Gaudin, M.: Sur la loi limite de l’espacement des valeurs propres d’une matrice aléatoire. Nucl. Phys. 25, 447–458 (1961)
Gustavsson, J.: Gaussian fluctuations of eigenvalues in the gue. Ann. Inst. Henri Poincaré (B) 41, 151–178 (2005)
He, Y.: Mesoscopic linear statistics of Wigner matrices of mixed symmetry class, Preprint arXiv:1803.10544 (2018)
He, Y., Knowles, A.: Mesoscopic eigenvalue statistics of Wigner matrices. Ann. Appl. Probab. 27, 1510–1550 (2017)
He, Y., Knowles, A., Rosenthal, R.: Isotropic self-consistent equations for mean-field random matrices, Preprint arXiv:1611.05364
Huang, J., Landon, B., Yau, H.-T.: Transition from Tracy-Widom to Gaussian fluctuations of extremal eigenvalues of sparse Erdős–Rényi graphs, Preprint arXiv:1712.03936 (2017)
Khorunzhy, A., Khoruzhenko, B., Pastur, L.: On the \(1/{N}\) corrections to the Green functions of random matrices with independent entries. J. Phys. A 28(1), L31 (1995)
Khorunzhy, A.M., Khoruzhenko, B.A., Pastur, L.A.: Asymptotic properties of large random matrices with independent entries. J. Math. Phys. 37, 5033–5060 (1996)
Lee, J.O., Schnelli, K.: Local law and Tracy–Widom limit for sparse random matrices, Preprint arXiv:1605.08767
Lytova, A., Pastur, L.: Central limit theorem for linear eigenvalue statistics of random matrices with independent entries. Ann. Probab. 37, 1778–1840 (2009)
Mehta, M.L.: Random Matrices. Academic Press, London (2004)
O’Rourke, S.: Gaussian fluctuations of eigenvalues in Wigner random matrices. J. Stat. Phys. 138, 151–178 (2010)
Stein, C.M.: Estimation of the mean of a multivariate normal distribution. Ann. Stat. 9, 1135–1151 (1981)
Tao, T., Vu, V.: Random matrices: universality of local eigenvalue statistics up to the edge. Commun. Math. Phys. 298, 549–572 (2010)
Tao, T., Vu, V.: Random matrices: universality of local eigenvalue statistics. Acta Math. 206, 1–78 (2011)
Acknowledgements
We gratefully acknowledge funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (Grant Agreement No. 715539_RandMat) and from the Swiss National Science Foundation through the SwissMAP grant.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Appendix A: Comparison to Gustavsson’s theorem
In this section we compare our result to the ones in [24] and [34]. Roughly, the conclusion is that the question addressed in [24, 34] is independent from the one addressed in our work, and neither implies the other.
In Gustavsson’s work [24], the fluctuation of a single eigenvalue was first established to be Gaussian for GUE, and it was also showed that the joint limit distribution of eigenvalues is a Gaussian process, provided that the eigenvalue are separated by a mesoscopic distance. In [34], the result was extended to GOE and, by moment matching, to a class of Wigner matrices.
For the rest of the section let us adopt the assumptions in Theorem 1.2, and in addition we assume H is GOE. Recall the notations \(\alpha :=-\log _N \eta \) and \(\gamma :=-\log _N \omega \). We would like to compute \(\int p_{E}(u,v) f_{+}(u)g_{-}(v) \,{\mathrm {d}}u\,{\mathrm {d}}v\). By (5.2) we have
Let \(\lambda _1\leqslant \cdots \leqslant \lambda _N\) be the eigenvalues of H, and for \(k=1,2,\ldots ,N\), let the quantile \(\gamma _k\) be the typical location of \(\lambda _k\), i.e. it satisfies \(k/N=\int _{-2}^{\gamma _k} \varrho _x \mathrm {d}x\). By our assumptions \(E \in [-2+\tau ,2-\tau ]\), \(f,g \in C^{\infty }_c({\mathbb {R}})\), and a standard eigenvalue rigidity result (e.g. Theorem 2.2, [22]), we see that the sum in (A.1) can be reduced to
where \(A :=\{i:|(\gamma _i-E)\varrho _E-\omega | \leqslant \eta \log N\}\), \(B :=\{j:|(\gamma _j-E)\varrho _E+\omega | \leqslant \eta \log N\}\). Clearly for \(i \in A\) and \(j\in B\), we have \(i-j \asymp N\omega = N^{1-\gamma }\), and Theorem 5 of [34] shows that
converges weakly to the Gaussian random vector \(N({\varvec{\mu }},\Sigma )\) with
as \(N \rightarrow \infty \). Also, by \(|\gamma _i-E|=\mathrm {O}(\omega )\) we see that \(\varrho _E=\varrho _{\gamma _i}+\mathrm {O}(\omega )\). Hence the random vector
has the same weak limit as the one in (A.2). With the rescaling \(u_k:=(\lambda _k-E)N\varrho _E\) for \(k=1,2,\ldots ,N\), we have
Now let us compute the right-hand side of (A.4) by using the weak limit of (A.3), i.e. that the random vector \((u_i,u_j)\) has asymptotically (as \(N \rightarrow \infty \)) the same law as the Gaussian vector \(({\tilde{u}}_i, {\tilde{u}}_j) \overset{\text {{d}}}{=}N({\varvec{\mu }}', \Sigma ')\), where
By Taylor expansion, we have
The first non-vanishing term is the one proportional to \(f'g'\). We now compute it asymptotically using Gustavsson’s theorem, by replacing \((u_i,u_j)\) with \(({\tilde{u}}_i, {\tilde{u}}_j)\). The result is
where in the second and third step we used the fact that \(m_{k+1}-m_k=1+\mathrm {O}(\omega )\) for \(k \in A\cup B\), and the Riemann sum of a smooth compactly supported function converges to its integral to any polynomial order (by the Poisson summation formula). Similarly, we can show that all other terms in (A.5) are bounded by \(\mathrm {O}_{\prec }(N^{4\alpha -4})\). Thus, under the assumption that we can use Gustavsson’s theorem and the error from the replacement of \((u_i, u_j)\) with the Gaussian vector \(({\tilde{u}}_i, {\tilde{u}}_j)\) is negligible, we get, together with (A.4),
We find that the right-hand side is much smaller than \(N^{2\gamma -2}\) for \(\alpha >3/4\) and \(\gamma >2\alpha /3\). However, by Theorem 1.2 we know that the leading term in \(\int p_{E}(u,v) f_{-}(u)g_{+}(v)\) is of order \(N^{2\gamma -2}\). We conclude that the use of Gustavsson’s theorem leads to a wrong result in at least the regime \(\alpha >3/4\) and \(\gamma >2\alpha /3\). This argument shows that the computation of \(\int p_{E}(u,v) f_{-}(u)g_{+}(v)\) using Gustavsson’s theorem leads to the wrong result in a rather flagrant fashion: the obtained quantity is much smaller than than the true value. In fact, we believe that more generally Gustavsson’s result cannot be used to recover the correct mesoscopic density–density correlations in any regime: the error made between (A.5) and (A.6) is never affordable.
The issue lies in the fact that, in order to compute sums of the form (A.5), even to leading order, one needs much stronger control on the joint distribution of the individual eigenvalues than is provided in [24] and [34]. This is a manifestation of the fact, mentioned in the introduction, that eigenvalue density–density correlations are much weaker than their location–location correlations. Going from the latter to the former consequently requires very precise asymptotics for the latter.
Appendix B: Comparison to results on macroscopic linear statistics
In this appendix we give the short calculation that shows how our main result, Theorem 1.2, recovers the well-known covariance formula for macroscopic linear statistics of Wigner matrices [30, 32]. This corresponds to the extreme case \(\omega \asymp \eta \asymp 1\) in Theorem 1.2.
Let for shortness \(f^{\eta }(x)=\eta ^{-1}f(((x-E)\varrho _E-\omega )/\eta )\) and \(g^{\eta }(x)=\eta ^{-1}g(((x-E)\varrho _E+\omega )/\eta )\). It follows from (1.8), (3.16) and (5.2) that there exists \(c=c(\tau )>0\) such that
Here \(F_1,F_2\) are defined as in (3.16). We see that when \(\omega \gg N^{-1/2}\) the integral RHS of the above equation dominates, and it matches Equation (VI.81) in [30].
Appendix C: Remark on the distribution of the diagonal entries
Theorems 1.2 and 2.1 can be extended to a more general case where we relax the assumption on the variance of \(H_{ii}\) in Definition 1.1. Here we summarize the appropriate generalizations of Theorems 2.1 and 1.2, whose proofs are simple modifications of the arguments given in the previous sections. This generalization illustrates the great sensitivity of our results to the precise distribution of the matrix entries.
Let \(\tilde{H}\) be a Wigner matrix where the condition \({\mathbb {E}} |\sqrt{N}H_{ii}|^2=2/\beta \) is replaced by \({\mathbb {E}}|\sqrt{N}\tilde{H}_{ii}|^2=\zeta _i\), and \(\max _i \zeta _i = \mathrm {O}(1)\). Let \(\tilde{G}(z):=(\tilde{H}-z)^{-1}\), and let \({\tilde{\Upsilon }}_{E,\beta } \) be the analogue of \(\Upsilon _{E,\beta }\) defined in (1.8)–(1.9).
Theorem 2.1 is generalized as follows. For the real symmetric case (\(\beta =1\)), we have
and for the complex Hermitian case (\(\beta =2\)), we have
Theorem 1.2 is generalized as follows. We have
as well as
where
Rights and permissions
About this article
Cite this article
He, Y., Knowles, A. Mesoscopic eigenvalue density correlations of Wigner matrices. Probab. Theory Relat. Fields 177, 147–216 (2020). https://doi.org/10.1007/s00440-019-00946-w
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00440-019-00946-w