I was wondering about the following problem:

Assume we have a state space $S:=\mathbb{Z}$ and a Markov chain, such that we can go from any state $x$ to some state $y$ with positive probabilities, i.e. $p_t(x,y)>0$ for any $t >0 $ and $x,y \in S.$

Let $T_0^x$ be the hitting time to go from state $x$ to $0$.

If we know that $P(\lim_{x \rightarrow -\infty} T_0^x<\infty)=1.$ So we can almost surely go to state $0$ from -infinity. Does this imply that $\liminf_{x \rightarrow -\infty} p_t(x,0)>0,$ for all $t>0$ i.e. does this imply that we go to state $0$ in every finite time step?

The thing is that we a priori only know that we can go to $0$ in some finite time, but not necessarily in any finite time (with some positive probability.)

  • $\begingroup$ How do you define $T_0^x$ exactly? Usually the starting state is not part of the random variable, but part of the probability measure. So you could talk about the hitting time of $0$, $T_0 = \inf\{t : X_t = 0\}$, and then ask about the probability of hitting 0 when starting at x, $P_x(T_0 < \infty)$. But then it does not make sense to have the $x$ limit inside the measure. $\endgroup$ Dec 15 '15 at 21:22
  • $\begingroup$ Indeed, for pretty much this reason, "starting at $-\infty$" seems to be ill defined. $\endgroup$ Dec 15 '15 at 21:23
  • $\begingroup$ Are you thinking of discrete time or continuous time here? $\endgroup$ Dec 15 '15 at 21:31
  • $\begingroup$ Then isn't continuous-time simple random walk a counterexample? It's recurrent, so $P_x(T_0 < \infty) = 1$ for every $x$, but it's easy to verify that for any fixed $t$, $p_t(x,0) \to 0$ as $x \to \pm \infty$. $\endgroup$ Dec 15 '15 at 21:33

It's not clear what condition you are imposing, but I think the following example indicates that the answer is no, there are positive times so that you don't have to get to $0$ with uniformly bounded probability:

Arrange the states in a tree with root $0$ with disjoint paths of length $n$ connected to the root for each $n\in \mathbb{N}$. On a path of length $n$, let the rate of movement toward the root be $n$. If $0 \lt t\lt 1$ and we start at the leaf of a path of length $n$, the distribution at time $t$ will be concentrated about $t$ of the way along the path. So, the limit inferior of the probability of reaching $0$ by time $t$ is $0$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.