Convergence of scaled random samples in \({\mathbb{R}{}}^ d\) (Q1180576)

Let \(\{\mathbf X_n,\; n\ge 1\}\) be i.i.d. \(\mathbb{R}^d\)-valued random vectors. It is proven that \(S_n = \{\mathbf X_n /b_n,\ 1\le j\le n\}\) converges a.s. (in probability) for certain choices of constants \((b_n)\) with \(b_n\to \infty\) and compact limit sets \(S\). Convergence is here understood with respect to the Hausdorff metric. It is shown that \(S\) is star-sharped (i.e. \(x \in S\Rightarrow tx\in S \text{ for } 0\le t\le 1\)). Necessary and sufficient conditions for convergence to a particular set \(S\) are formulated. The problem is closely related to relative stability of the maxima of \(\| \mathbf X_1\|,\dots ,\| \mathbf X_n\|\). Earlier work on this topic has been done by \textit{L. Fisher} [Ann. Math. Stat. 40, 1824--1832 (1969; Zbl 0183.47501)] and by \textit{R. A. Davis}, \textit{E. Mulrow} and \textit{S. I. Resnick} [Adv. Appl. Probab. 20, No. 3, 573--599 (1988; Zbl 0656.60026)].