Page 131 - A First Course In Stochastic Models
P. 131

THEORETICAL CONSIDERATIONS                   123

                (c) If the recurrent states are aperiodic, then there is an integer ν ≥ 1 such that
                    (ν)
                   p   > 0 for all i ∈ I and j ∈ R.
                    ij
                Proof  Since the Markov chain has no two disjoint closed sets, the closed set R of
                recurrent states is irreducible by Theorem 3.5.4. Hence, by Lemma 3.5.2, any two
                states in R communicate with each other. This implies that for any i, j ∈ R there
                                         (n)
                is an integer n ≥ 1 such that p  > 0. Next we prove that for any i ∈ I and j ∈ R
                                        ij
                                             (n)
                there is an integer n ≥ 1 such that p  > 0. To verify this, assume to the contrary
                                             ij
                that there is a transient state i ∈ I such that no state j ∈ R is accessible from i.
                Then there is a closed set that contains i and is disjoint from R. This contradicts
                the assumption that the Markov chain has no two disjoint closed sets. Hence for
                any transient state i ∈ R there is a state j ∈ R that is accessible from i. Thus any
                state j ∈ R is accessible from any i ∈ I, since any two states in R communicate
                with each other.
                  To verify parts (b) and (c), define under the condition X 0 = i the random variable
                N ij by
                                      N ij = min{n ≥ 1 | X n = j}.

                Fix now j ∈ R. For each i ∈ I, let r i be the smallest positive integer n for which
                 (n)
                p   > 0. Define
                 ij
                                                            (r i )
                                              and ρ = min p   .
                                    r = max r i             ij
                                        i∈I             i∈I
                Since I is finite, we have r < ∞ and ρ > 0. Next observe that
                                                       (r i )
                          P {N ij > r} ≤ P {N ij > r i } = 1 − p  ≤ 1 − ρ,  i ∈ I.
                                                       ij
                Thus, for any i ∈ I,
                                                   k
                                P {N ij > kr} ≤ (1 − ρ) ,  k = 0, 1, . . . .
                Since the probability P {N ij > n} is decreasing in n and converges to 0 as n → ∞,
                it follows from 1 − f ij = lim n→∞ P {N ij > n} that f ij = 1. Since P {N ij > n} is
                decreasing in n, we also obtain

                                 ∞                 ∞     rk

                           µ ij =  P {N ij > n} = 1 +         P {N ij > ℓ}
                                n=0                k=1 ℓ=r(k−1)+1
                                    ∞
                                              k
                              ≤ 1 +    r(1 − ρ) ,
                                   k=1
                showing that µ ij < ∞. This completes the proof of part (b).
                  It remains to prove (c). Fix i ∈ I and j ∈ R. As shown above, there is an integer
                               (v)
                v ≥ 1 such that p  > 0. By part (b) of Lemma 3.5.5 there is an integer n 0 ≥ 1
                               ij
                         (n)                            (v+n)   (v) (n)
                such that p  > 0 for all n ≥ n 0 . Hence, by p  ≥ p  p  , it follows that
                         jj                             ij      ij  jj
   126   127   128   129   130   131   132   133   134   135   136