A nonclassical law of iterated logarithm for negatively associated random variables (Q1413705)

Consider a strictly stationary, negatively associated sequence \(~\{X_i,i\geq 1\}~\) with \(~EX_1=0,EX^2_1 <\infty\), and \(~\sigma^2:= EX^2_1+ 2 \sum^\infty_{i=2} EX_1X_i >0\). For \(~1<M<\infty~\), define recursively \(~N_0=1~\) and \(N_k=\min\{ n\geq 1:\log_2 n \geq M\log_2 N_{k-1}\},~ k\geq 1\), where \(~\log_2 x=\ln\ln (x\vee e^e), ~x>0\), and set \(~b(n)=M\log_2 N_k\), for \(~N_k\leq n<N_{k+1}, ~k\geq 0\). The author's main result proves that \[ \mathop{\limsup}\limits_{n\rightarrow \infty} {{S_n}\over{\sqrt{2\sigma^2 nb(n)}}}=1\;\text{a.s.}, \] where \(~S_n=\sum^\infty_{i=1}X_i,~n\geq 1~\). This extends earlier works of \textit{Q.-M. Shao} and \textit{C. Su} [Stochastic Processes Appl. 83, 139-148 (1999; Zbl 0997.60023)] and \textit{O. Klesov} and \textit{A. Rosalsky} [Stochastic Anal. Appl. 19, 627-641 (2001; Zbl 0990.60027)]. The proof is essentially based on a Rosenthal type maximal inequality of \textit{Q.-M. Shao} [J. Theor. Probab. 13, 343-356 (2000; Zbl 0971.60015)] in combination with a subsequence argument.