Extreme eigenvalue statistics of \(m\)-dependent heavy-tailed matrices (Q2077358)

Consider a stationary random field \((X_{i,j})_{(i,j)\in\mathbb Z_+^2}\) which is \(m\)-dependent, that is, for any \(A, B\subset \mathbb Z_+^2\) such that \[ \max(|i-k|,|j-l|)>m\quad \text{for all}\quad (i,j)\in A, (k,l)\in B, \] the families \((X_{i,j})_{(i,j)\in A}\) and \((X_{k,l})_{(k,l)\in B}\) are independent. Also consider its diagonally symmetrized version \((\hat X_{i,j})_{(i,j)\in\mathbb Z_+^2}\), where \(\hat X_{i,j}= X_{i,j}\) for \(i\leq j\) and \(\hat X_{i,j}= X_{j,i}\) for \(i> j\). Assume that the common distribution of the random variables satisfies \[ \mathbb P[|X_{i,j}|>t]=t^{-\alpha} L(t),\quad \alpha\in(0,4), \] for some slowly varying at infinity function \(L\) and in addition assume that \(\mathbb E X_{i,j}=0\) if \(\alpha\in[2,4)\). Choose a normalizing sequence \(a_n\) such that \[ n\mathbb P[|X_{i,j}|>a_n]\to 1,\quad n\to\infty, \] and construct 2 sequences of random matrices \[ \mathbf A_n= (X_{i,j}/a_{np_n})_{1\leq i\leq p_n,1\leq j\leq n }\quad\text{and}\quad\hat{\mathbf A}_n=(\hat X_{i,j}/a_{n^2})_{1\leq i,j\leq n} , \] where the integer sequence \(p_n\) satisfies \(p_n/n\to\gamma\in(0,\infty)\). The authors show that for the empirical measures of the singular values \(\sigma_i(\mathbf A)\) of \(\mathbf A\) and the eigenvalues \(\lambda_i(\hat {\mathbf A}_n)\) of \(\hat {\mathbf A}_n\) we have \[ \sum_{i=1}^n\delta_{\sigma_i({\mathbf A}_n)} \stackrel{d}{\to} \mathcal P,\quad \left(\sum_{i=1}^n\delta_{\lambda_i(\hat {\mathbf A}_n)}\mathrm1_{\lambda_i(\hat {\mathbf A}_n)>0}, \sum_{i=1}^n\delta_{\lambda_i(\hat {\mathbf A}_n)}\mathrm1_{\lambda_i(\hat {\mathbf A}_n)<0}\right) \stackrel{d}{\to} (\mathcal P, -\mathcal P), \] where \(\mathcal P\) is some certain Poisson cluster process in \(\mathbb R_+^1\) with a given precise and detailed albeit complicated structure. This result generalizes the one from [\textit{A. Auffinger} et al., Ann. Inst. Henri Poincaré, Probab. Stat. 45, No. 3, 589--610 (2009; Zbl 1177.15037)], where the case of i.i.d. \(X_{i,j}\) has been considered.