The law of the iterated logarithm for functionals of Harris recurrent Markov chains: Self normalization (Q1295855)

Let \(\{X_n,n\geq 0\}\) be a Harris recurrent Markov chain with state space \(E\), transition probability \(P(x,A)\) and invariant measure \(\pi\). Suppose \(f,g\) are measurable functions such that \(g\geq 0\), \(\int fd\pi=0\), \(\int gd\pi>0\) and denote \(S_n(f)=\sum^n_{k=1}f(X_k)\), \(P^kf(x)=\int f(y)P^k(x,dy)\), \(\sigma^2_f=\int f^2d\pi+2\int\sum^\infty_{k=1}fP^kfd\pi\). It is proved that with probability one \[ \limsup_{n\to\infty}S_n(f)\bigl[2S_n(g)\log\log S_n(g)\bigr]^{-1/2}=\sigma_f\Bigl(\int gd\pi\Bigr)^{-1/2} \] under some best possible conditions. In particular, one can apply \(g=f^2\) provided \(\int f^2d\pi>0\). The main tool in the proof is the regeneration-split chain method due to Athreya and Ney, and Nummelin.