Author_Institution :
IBM Thomas J. Watson Res. Center, Yorktown Heights, NY, USA
Abstract :
For a stationary, irreducible and aperiodic Markov chain with finite alphabet A, starting symbol X0=σ, transition probability matrix P, stationary distribution π, support S(π,P)={(j,k):πjPk|j>0} and for a function f such that M=ΔEπPf(X1,X2)<+∞ it is known that limsupl→∞ 1/l log2 μ(|M-1/l Σ>i=1lf(Xi-1,Xi)|≥ε)≤-inf(q,Q)∈Γ(e) D(Q||P), where Γε={(q,Q):|Eπ,Pf(X1,X2q,Qf(X1,X2)|≥ε Qq=q,S(q,Q)⊂S(π,P)}. In this paper we demonstrate that inf(q,Q)∈Γ(e) D(Q||P)≥ 1/(2ln2)[supn≥1:Kn>0{Kn/(1+Kn)}ε/(maxj,k:P(k|j)>0|f(j,k)|]2 where Kn=(1-|A|maxj,k|Pk|jn-πk)/n). Under the conditions stated, the set over which the sup is taken is nonempty and therefore the sup exists and is positive; it is also shown that the sup is attained at a finite value of n. A nonasymptotic version of this result is also given based on the method of Markov types.