A compact law of the iterated logarithm for random vectors in the generalized domain of attraction of the multivariate Gaussian law (Q1303911)

Let \(X,X_1,X_2, \dots\) be i.i.d. random vectors in \(\mathbb{R}^d\), not restricted to a lower-dimensional linear subspace. Let \(X\) belong to the (possibly non-normal) domain of attraction of the standard normal law \(N(0,I)\), i.e. \(T_n(S_n- b_n)\to N(0,I)\) in distribution for linear operators \(T_n: \mathbb{R}^d \to\mathbb{R}^d\) and vectors \(b_n\in\mathbb{R}^d\). Here \(S_n=X_1+ \cdots+X_n\). Then \(E\|X\|<\infty\) and one may take \(b_n=nEX\). So \(EX=0\) is assumed. \textit{M. M. Meerschaert} [Ann. Probab. 19, No. 2, 875-892 (1991; Zbl 0732.60025) and J. Theor. Probab. 7, No. 4, 793-798 (1994; Zbl 0807.60032)] characterized attraction to \(N(0,I)\) by regular variation conditions. Under these conditions the following law of the iterated logarithm is proved: Put \(Lx= \max(1,\log x)\) and \(L_2x= L(Lx)\). Let \(x_n \twoheadrightarrow A\subset\mathbb{R}^d\) for \(x_n\in\mathbb{R}^d\) mean that the distance of \(x_n\) to \(A\) tends to zero and the cluster set of the sequence \(x_n\) is \(A\). Then \(2^{-1/2}T_{t_n} (S_n)/L_2 n\twoheadrightarrow \{x\in\mathbb{R}^d:\|x\|\leq 1\}\) a.s. if and only if \[ \sum_n P\biggl(\bigl \|T_{t_n}(X)\bigr\|> \varepsilon L_2n\biggr\} <\infty\tag{*} \] for all \(\varepsilon>0\). The theorem is derived from the corresponding one for all projections: Under (*) we have \(\limsup_n |2^{-1/2} \langle T_{t_n} (S_n),\theta\rangle/L_2(n) |=1\) for all \(\theta\in \mathbb{R}^d\) with \(\|\theta \|=1\). Here \(\langle , \rangle\) denotes inproduct. The latter result is proved by double truncation.