A markov chain inequality in Billingsley that should be an equation?

108 Views Asked by At

In the section on Markov chains in Billingsley's Probability and Measure (3e) we have the following inequality on page 120 in the proof of Theorem 8.3,

$$ \begin{align*} p_{ji}^{(m)} &= P_j([X_m=i] \cap [X_n = j \text{ i.o.}])\\ &\le \sum_{n>m} P_j (X_m = i, X_{m+1} \ne j,\ldots, X_{n-1} \ne j,X_n=j) \end{align*} $$

[Notation: $p_{ji}^{(m)} = P(X_m =i \mid X_0=j)$, $P_j(A) = P(A \mid X_0=j)$.]

My question is regarding the inequality sign on the second line.

$X_n=j$ infinitely often if and only if $X_n=j$ infinitely often for $n>m$. Each such path must hit $j$ for the first time after period $m$ at some finite time $n$. So it seems to me that the events on right-hand side of the second line form a partition of the event on right-hand side of the first line and therefore we must have an equality here.

Am I going wrong somewhere?

1

There are 1 best solutions below

0
On BEST ANSWER

The sum on the second line computes the probability that $X_n$ hits $j$ at least once. If there is a positive probability that $X_n$ hits $j$ at least once but not infinitely often (e.g. if $j$ is transient), then the inequality is strict.