On the concentration of random multilinear forms and the universality of random block matrices (Q2349145)

Let denote by \(\lambda_{k}(M)\) the eigenvalues of a matrix \(M\) of size \(n\times n\) with complex numbers as entries. The empirical spectral distribution (ESD) of this matrix is defined as \(F^{M}(x,y)= \frac{1}{n} \sharp \{ 1\leq i\leq n: \mathrm{Re}(\lambda_{i}(M)) \leq x, \mathrm{Im}(\lambda_{i}(M)) \leq y \},\) where \(\#\sharp A\) denotes the cardinal of the set \(A\). On the other hand, the empirical spectral measure of \(M\) is given by \(\mu_{M}= \frac{1}{n} \sum_{i=1}^{n} \delta_{\lambda_{i}(M)}\). If you deal with random matrices \(X_{n}\) of size \(n\times n\) whose entries are independent identically distributed copies of a standard complex Gaussian random (atom) variable \(\xi\), then the matrix can be viewed as a random matrix drawn from the probability distribution \(\frac{1}{\pi^{n^{2}}} e^{-\mathrm{tr} M M^{*}} dM\) on the set of complex matrices of size \(n\times n,\) where \(dM\) is the Lebesgue measure on the \(2n^{2}\) real entries of \(M\). This set of matrices is known in the literature as complex Ginibre ensemble. \textit{M. L. Mehta} [Random matrices. 3rd ed. Amsterdam: Elsevier (2004; Zbl 1107.15019)] has proved that the ESD of \( \frac{1}{\sqrt{n}} X_{n}\) converges to the circular law \(F_{\mathrm{circ}}\) when \(n\) tends to infinity. Notice that \(F_{\mathrm{circ}} (x,y)= \mu_{\mathrm{circ}} (\{ z\in \mathbb{C}: \mathrm{Re} (z) \leq x, \mathrm{Im}(z)\leq y \}),\) where \(\mu_{\mathrm{circ}}\) is the uniform probability measure on the unit disk of the complex plane. In the general (non-Gaussian) case, there is not a formula for the joint distribution of the eigenvalues and the problem seems to be more difficult. Nevertheless, the universality phenomenon in random matrix theory states that the spectral behavior of a random matrix does not depend on the distribution of the random variable \(\xi\) when \(n\) tends to infinity, i.e., one guess that the circular law describes the limiting ESD of a large class of random matrices. A first rigorous proof of the circular law for general (non-Gaussian) ensembles under a number of moment and smoothness assumptions on \(\xi\) was done by \textit{Z. D. Bai} [Ann. Probab. 25, No. 1, 494--529 (1997; Zbl 0871.62018)]. Some improvements of this result have been done by \textit{T. Tao} et al. [Ann. Probab. 38, No. 5, 2023--2065 (2010; Zbl 1203.15025)], when the entries of the matrix \(X_{n}\) are independent identically distributed copies of a complex random variable \(\xi\) with mean zero and unit variance and you deal with a matrix \(M_{n}= X_{n}+ N_{n}\), where \(N_{n}\) is a deterministic matrix such that rank \((N_{n})= o(n)\) and has a polynomially bounded norm. The aim of the contribution under review is to study the universality phenomenon for a class of random matrices that generalizes the random matrix ensemble analyzed in the above work by Tao et al. [loc. cit.]. In particular, random block matrices whose entries are not necessarily independent are considered. A block matrix \(X_{n}\) is said to satisfy condition \(\mathrm{C0}\) when its \((i,j)\) block, \(X_{i,j}\), \(1\leq i,j \leq n\), has entries \((x_{s,t;i,j})_{s,t=1}^{d}\) which are independent identically distributed copies of atom variables \((\xi_{s,t})_{s,t=1}^{d}\) such that (i) \(\mathbb{E} [\xi_{s,t}]=0\), \( \mathbb{E}[|\xi_{s,t}|^{2}]=1\), \(\mathbb{E}[ |\xi_{s,t}|^{2+ \eta}]=1\), for some \(\eta >0\), (ii) \( \mathbb{E} [\xi_{s,t} \overline{\xi_{u,v}}]=0\) for every \((s,t)\neq (u,v)\). If \(\{ X_{n}\}_{n\geq 1}\) is a sequence of random matrices satisfying condition \(\mathrm{C0}\) and assuming the sequence \(\{ N_{n}\}_{n\geq 1}\) satisfies rank \( (N_{n})= O(n^{1-\varepsilon})\) for some \(\varepsilon >0\) and \(\sup_{n\geq 1} \frac{1}{n} ||N_{n}||_{2} \leq \infty\), where \(||.||_{2}\) denotes the Hilbert-Schmidt norm of a matrix, then for the sequence of matrices \(\{ M_{n}\}_{n\geq 1}\), with \(M_{n}= X_{n}+ N_{n}\), the ESD of \( \frac{1}{\sqrt{n}} M_{n}\) converges almost surely to the circular law \(F_{\mathrm{circ}}\) when \(n\) tends to infinity. One of the key ingredients in the proof of the above result is the determination of a bound on the least singular value \(\sigma_{\mathrm{dn}} (X_{n})\) of the random matrices \(\{ X_{n}\}_{n\geq 1}\) assuming the conditions (i) and (ii) hold. Indeed, if \(N_{n}\) is a deterministic matrix whose entries are bounded by \(n^{\alpha}\) for some \(\alpha>0\), then for every \(B>0\) there exists \(A>0\), depending only on \(d,B,\alpha\), such that \(\mathbb{P}( \sigma_{\mathrm{dn}} (M_{n}) \leq n^{-A}) = O (n^{-B})\). When the entries of the matrix \(X_{n}\) are independent identically distributed copies of a complex random variable \(\xi\) with mean zero and unit variance, a similar result for \(d=1\) was obtained by \textit{T. Tao} and \textit{V. Van} [Commun. Contemp. Math. 10, No. 2, 261--307 (2008; Zbl 1156.15010)]. An inverse type result for the concentration of linear operators and multilinear forms is used in the proof of the above result.