On the Rates of the Almost Sure Convergence for Self-Normalized Law of the Iterated Logarithm

Tian-Xiao Pang
DOI: https://doi.org/10.4134/bkms.2011.48.6.1137
2011-01-01
Bulletin of the Korean Mathematical Society
Abstract:Let {Xi, i >= 1} be a sequence of i.i.d. nondegenerate random variables which is in the domain of attraction of the normal law with mean zero and possibly infinite variance. Denote S-n = Sigma(n)(i-1) Xi, M-n = max(1 <= i <= n) vertical bar Si vertical bar and V-n(2) = Sigma(n)(i-1) X-i(2). Then for d > 1, we showed that under some regularity conditions,lim epsilon -> 0 epsilon(2(d+1)) Sigma(infinity)(n=1)(log log n)(d)/n log n I{M-n/V-n >= root 2 log log n(epsilon + alpha(n))} = 2/root pi(1+d) Gamma (d+3/2) Sigma(infinity)(k=0) (-1)(k)/(2k+1)(2d+2) a.s.holds in this paper, where I{.} denotes the indicator function.
What problem does this paper attempt to address?