On the max-domain of attractions of bivariate elliptical arrays (Q881411)

The main result of this paper is the following. Let \((S_1, S_2)\) be a bivariate spherical random vector with almost surely positive random radius \(R\) with the distribution function \(F\). Let \((U_{ni}, V_{ni})\), \(1\leq i\leq n,\;n\geq1,\) be a triangular array such that \((U_{ni}, V_{ni})\buildrel d \over = (S_1, \rho_{n}S_1+\sqrt{1-\rho_{n}S_2}),\;1\leq i\leq n,\;n\geq1\), with \(\rho_{n}\in (0,1)\). Define the constants \(a(n):= 1-G^{-1}(1-1/n)\), \(n\in N\), where \(G\) is the distribution function of \(S_1\) and \(G^{-1}\) is the generalized inverse function. Assume that \(F\) has upper endpoint equal to 1 and, further, \(F\) is in the max-domain of attraction of \(\Psi_{\alpha^{*}}\), \(\alpha^{*}>0\), satisfying \(\lim_{n\to\infty}\sup_{x<0}| F^{n}(c(n)x+d(n))-\Psi_{\alpha}(x)|=0\). If \(\lim_{n\to\infty}(1-\rho_{n})/a(n)=2\lambda^2\in [0,\infty]\) holds, then \[ \lim_{n\to\infty}\sup_{(x,y)\in(-\infty,0)^2}| P\{M_{n1}\leq1+a(n)x,\;M_{n2}\leq1+a(n)y\}-H_{\alpha,\lambda}(x,y)|=0 \] with \[ H_{\alpha,\lambda}(x,y)=\exp\left(-| x|^{\alpha}\psi_{\alpha}\left({1\over\sqrt{2| x|}}\left[\lambda+{y-x\over 2\lambda}\right]\right)-| y|^{\alpha}\psi_{\alpha}\left({1\over\sqrt{2| y|}}\left[\lambda+{y-x\over 2\lambda}\right]\right)\right), \] for \(x,y<0\), where \[ \alpha:=\alpha^{*}+1/2,\quad \psi_{\alpha}(z):={\Gamma(\alpha+3/2)\over \Gamma(\alpha+1)\pi}\int_{-1}^{z}(1-s^2)^{\alpha}\,ds,\;\alpha>0, z\in[-1,1], \] \[ M_{n1}:=\max_{1\leq i\leq n}U_{ni}, \quad M_{n2}:=\max_{1\leq i\leq n}V_{ni}. \]