Precise rates in the law of logarithm for i.i.d. random variables (Q2387340)

Let \(\{X_{n}:n\geq 1\}\) be a sequence of i.i.d. random variables, and set \(S_{n}=X_{1}+\cdots+X_{n}\) and \(M_{n}=\max_{k\leq n}|S_{k}|\), \(n\geq 1.\) \textit{A. Gut} and \textit{A. Spătaru} [J. Math. Anal. Appl. 248, 233--246 (2000; Zbl 0961.60039)] showed that \[ \lim_{\varepsilon \searrow 0}\varepsilon ^{2b+2}\sum_{n\geq 1}\frac{(\log n)^{b}}{n}P\Bigl(|S_{n}|\geq \varepsilon \sqrt{n\log n}\Bigr)= \frac{\mu ^{2b+2}}{b+1}\sigma ^{2b+2} \] whenever \(EX_{1}=0\), \(EX_{1}^{2}=\sigma ^{2}\) and \(0\leq b\leq 1,\) where \(\mu ^{2b+2}\) stands for the \((2b+2)\)th absolute moment of the standard normal distribution. The authors strengthen and extend this result by proving the following theorems. Theorem 1. \[ \lim_{\varepsilon \searrow 0}\varepsilon ^{2b+2} \sum_{n\geq 1} \frac{(\log n)^{b}}{n} P\Bigl(M_{n}\geq \varepsilon \sigma \sqrt{n\log n}\Bigr)= 2\frac{\mu ^{2b+2}}{b+1} \sum_{k\geq 0} \frac{(-1)^{k}}{(2k+1)^{2b+2}}, \tag{1} \] and \[ \lim_{\varepsilon \searrow 0}\varepsilon ^{2b+2} \sum_{n\geq 1} \frac{(\log n)^{b}}{n} P\Bigl(|S_{n}| \geq \varepsilon \sigma \sqrt{n\log n}\Bigr)= \frac{\mu ^{2b+2}}{b+1}, \tag{2} \] whenever \(EX_{1}=0\), \(EX_{1}^{2}=\sigma ^{2}\) and \(-1<b\leq 1.\) Conversely, if either (1) or (2) holds for some \(0<\sigma <\infty \) and \(-1<b\leq 1,\) then \(EX_{1}=0\), \(EX_{1}^{2}=\sigma ^{2}.\) Theorem 2. \[ \lim_{\varepsilon \searrow 0} \varepsilon ^{2b+2} \sum_{n\geq 1} \frac{(\log n)^{b}}{n} I\Bigl\{M_{n}\geq \varepsilon \sigma \sqrt{n\log n}\Bigr\}= 2\frac{\mu ^{2b+2}}{b+1} \sum_{k\geq 0} \frac{(-1)^{k}}{(2k+1)^{2b+2}} \text{ a.s. and in }L^{2}\tag{3} \] and \[ \lim_{\varepsilon \searrow 0} \varepsilon ^{2b+2} \sum_{n\geq 1}\frac{(\log n)^{b}}{n} I\Bigl\{|S_{n}|\geq \varepsilon \sigma \sqrt{n\log n}\Bigr\}= \frac{\mu ^{2b+2}}{b+1} \text{ a.s. and in }L^{2}, \tag{4} \] whenever \(EX_{1}=0\), \(EX_{1}^{2}=\sigma ^{2}\) and \(-1<b<-3/4.\) Conversely, if either (3) or (4) holds for some \(0<\sigma <\infty \) and \(-1<b\leq 1,\) but with convergence in probability instead of a.s. and \(L^{2}\) convergence, then \(EX_{1}=0\), \(EX_{1}^{2}=\sigma ^{2}.\)