Extremes of weighted Dirichlet arrays (Q2271716): Difference between revisions

Let \(\mathbf{X}_{n}^{(j)},1\leq j\leq n,n\geq1\), be independent random vectors in \(\mathbb R^k,k\geq2\) defined by the stochastic representation \(\mathbf{X}_{n}^{(j)}=A_nR\mathbf{W},1\leq j\leq n\), where \(\mathbf{W}=(W_1,\dots,W_k)^{\top}\) is a weighted Dirichlet random vector in \(\mathbb R^k\) independent of the random variable \(R>0\) (almost surely), and \(A_n,n\geq1\), is a sequence of \(k\)-dimensional real square matrices. Let \(\mathbf{a}_{ni}\) be the \(i\)-th row vector of \(A_n\). Then \(\mathbf{a}_{ni}\mathbf{W}\) is the \(i\)-th projection of \(\mathbf{W}\) along the direction \(\mathbf{a}_{ni}\). As in the Hüsler-Reiss model (see \textit{J. Hüsler} and \textit{R.-D. Reiss} [Stat. Probab. Lett. 7, No. 4, 283--286 (1989; Zbl 0679.62038)]) there are considered asymptotically singular projections of the random vector \(\mathbf{W}\) assuming that almost surely \(A_n\mathbf{W}\to(W_k,\dots,W_k)^{\top}, n\to\infty.\) If the componentwise sample maxima \(\mathbf{M}_n=(M_{n1},\dots,M_{nk}),n\geq1\), \({M}_{n1}=\max_{1\leq j\leq n}X_{n1}^{(j)},\dots,M_{nk}=\max_{1\leq j\leq n}X_{nk} ^{(j)}),\) of the triangular array \(\mathbf{X}_{n}^{(j)}=A_nR\mathbf{W},1\leq j\leq n\), converge in distribution to a random vector with a non-degenerate max-infinitely divisible distribution, then we say that \(\mathbf{X}_{n}^{(j)},1\leq j\leq n,n\geq1\), is a Hüsler-Reiss triangular array. In this paper it is shown that if the associated random radius \(R\) has distribution function \(F\) in the Gumbel or the Weibull max-domain of attraction, then the above triangular array is a Hüsler-Reiss triangular array, provided that the sequence of the matrices \(A_n,n\geq 1\) satisfies certain asymptotic conditions. It is of some interest that for the Gumbel case (assumption on \(F\)) the limiting distribution of the sample maxima is max-stable, which is not the case when \(F\) is in the Weibull max-domain of attraction.