“…{Xk_l r Xk =i, Xk+l =i .... , Xk+s-1 =i, Xk+s #i} = ~(xk-~ # i}~{xk = i I xk-, # i}~{xk+~ = i ] xk = i} 9 "I? {Xk+~ # i I Xk+s-1 = i}.The last probability can be computed as follows:P{Xk+s~i[Xk+s-l=i}= ~ P{Xk+s=jlXa+s_I =i}=l-p(i,i),(9) jeE--{i}whereas the first can be evaluate as follows:P{Xk-1 r =i IXk_l #i}= E P{Xk-1 =j}P{Xk =ilXk-1 =j} j6E-{i} = E C~Pk-l(J)P(J'i) iCE-{i} = ~p~(i) -~pk-l(i)p(i, z).Since limk~o rxpk(i) = ~r(i) thenlim Pr {Xk-1 # i}Pr{Zk = i l Xk-1 # i} = 7r(i)[1 --p(i,i)].Let us give the formulation for another interesting case, the case of nonhomogeneous Markov chains where the transition probabilities are time-dependent[17,18]. Let {Xn; n E N} now be a nonhomogeneous Markov chain in discrete time, with c~ the initial distribution of the probabilities and P.n (pn(i,j) = IP{Xn+l = j ] Xn i}; n E N, i,j E E) the time-dependent transition probability matrix, thenI?…”