On the LSL for random fields (Q548148)

Let \((X_{\mathbf k})_{\mathbf k\in\mathbb N^d}\) be i.i.d. random fields and let \[ T_{\mathbf n,\mathbf n+\mathbf k}=\sum_{n_i<j_i\leq n_i+k_i}X_{\mathbf j} \] denoted their delayed sum for a window of size \(\mathbf k\in\mathbb N^d\). For \(d=1\), Lai's law of the single logarithm (LSL) roughly states that, for \(0<\alpha<1\), almost surely \[ \limsup_{n\to\infty}\frac{T_{n,n+n^\alpha}}{\sqrt{2(1-\alpha)n^\alpha\log n}}=\sigma, \] if and only if \(\text{E}[|X_1|^{2/\alpha}(\log_+|X_1|)^{-1/\alpha}]<\infty\), \(\text{E}[X_1]=0\) and \(\text{E}[X_1^2]=\sigma^2\). The authors combine their extension of Lai's result for random fields [Bernoulli 14, No.~1, 249--276 (2008; Zbl 1160.60011)] with a suitable version for \(\alpha=1\) [tha authors and \textit{F. Jonsson}, Bernoulli 16, No.~1, 1--22 (2010; Zbl 1253.60037)]. For \(d=2\), the characterizing moment condition and a suitable (iterated) logarithm function \(f\) such that almost surely one gets \[ \limsup_{m,n\to\infty}f(m,n)^{-1}T_{(m,n),(m+a_m,n+b_n)}=\sigma \] are given explicitly, where \(a_n,b_n\) are either \(n/\log n\), \(n/\log\log n\) or \(n^\alpha\), \(0<\alpha<1\).