On a conjecture of Révész

Abstract:

Let $\{ {X_n},n \geq 1\}$ be i.i.d. random variables with $P({X_i} = \pm 1) = \frac {1}{2}$. Révész (1990) proved \[ \begin {array}{*{20}{c}} {1 \le \lim \inf \limits _{n \to \infty } \max \limits _{0 \le j < n} \max \limits _{1 \le k \le n - j} {{(2k\log n)}^{ - 1/2}}({S_{j + k}} - {S_j})} \hfill \\ { \le \lim \sup \limits _{n \to \infty } \max \limits _{0 \le j < n} \max \limits _{1 \le k \le n - j} {{(2k\log n)}^{ - 1/2}}({S_{j + k}} - {S_j}) \le K\quad {\rm {a}}.{\rm {s}}.} \hfill \\ \end {array}\] and conjectured $K = 1$, where ${S_n} = \sum \nolimits _{i = 1}^n {{X_i}}$. In this we show that Révész’s conjecture is true but the conclusion is not valid for general i.i.d. random variables with finite moment generating function.