Upper-lower class tests for weighted i.i.d. sequences and martingales (Q975330)

The authors study the upper-lower class behavior of weighted sums \(S_n=\sum_{k=1}^n a_kX_k\) where \((X_n)\) is a sequence of random variables and \((a_n)\) is a sequence of nonzero real numbers. Set \(s_n^2= \sum_{k=1}^n a_k^2\), \(n\in\mathbb{N}\). Assume that \(s_n \to \infty\) and \(|a_n|=O(s_n^{1-\delta})\) for some \(\delta >0\) as \(n\to \infty\). Put \(M(x)=\sharp\{k\geq 1: |s_k|/|a_k|\leq x\}\), \(x>0\). It is proved that if \(M(x)=O(x^2)\) as \(x\to \infty\) then for any i.i.d. sequence \((X_n)\) such that \({\mathbf E}X_1=0\), \({\mathbf E}X_1^2=1\) and \({\mathbf E}X_1^2\log\log(|X_1|\vee e^e)<\infty\) the integral test by Feller holds [see \textit{W. Feller}, Trans. Am. Math. Soc. 54, 373--402 (1943; Zbl 0063.08417)]. In other words \({\mathbf P}(S_n >s_n\varphi(s_n^2)\; i.o.)=0\) or \(1\) for a nondecreasing function \(\varphi:\mathbb{R}_+\to \mathbb{R}_+\) according as \[ I(\varphi):=\int_1^{\infty}t^{-1}\varphi(t)e^{-\varphi(t)^2/2}dt<\infty\;\;\text{or}\;\;=\infty. \] Conversely, if the last statement is valid, one has \(M(x)=O(x^2(\log\log x)^2)\) as \(x\to \infty\). Thus the asymptotic behavior of weighted sums depends not on the growth properties of \((a_n)\) but on its arithmetical distribution. The analog of this result is established also for weighted sums of stationary martingale difference sequences. The introduction of the paper contains an interesting discussion of the previous results in the research domain concerning the refinements of the LIL.