Transient discrete time Markov chain on integers: can direction of flow be proven?

85 Views Asked by At

I'm not very familiar with the theory of Markov chains, and I'd like to learn how complicated the following problem actually is.

Let there be a discrete time Markov chain on $\mathbb{Z}$, where the random variable giving the position on the line is called $Z(t)$. The state is only allowed to change to the nearest neighbours, so the chain is defined by the transition probabilities:

\begin{align} P \{ Z(t+1)=n+1 \mid Z(t)=n \} &= q_{n} \\ P \{ Z(t+1)=n-1 \mid Z(t)=n \} &= 1-q_{n} \\ P \{ Z(t+1)=m \mid Z(t)=n \} &= 0 &\mathrm{if} \ m \notin \{n-1,n+1 \} \end{align}

$ \forall \ n \in \mathbb{Z}$. The point is that the transition probabilities $q_{n}$ are not necessarily the same everywhere.

On this system, there is a periodic symmetry. With period $N \in \mathbb{N}$, we have: $q_{n+k \cdot N} = q_{n} \ \ \forall k \in \mathbb{Z}$.

Now, let's define $p_{ij}^{(t)} = P\{Z(t)=j \mid Z(0)=i\}$ (everything is time-homogeneous) and the corresponding first-passage probability $f_{ij}^{(t)}$, the probability to first reach $j$ from $i$ after exactly $t$ timesteps. The probability to make at least one passage at all at location $j$ is given by:

\begin{equation} f_{ij} = \sum_{t=1}^{\infty} f_{ij}^{(t)} \end{equation}

Intuitively, the system should run forward if the forward probabilities $q_{n}$ are larger than the backward ones. More formally, I define $Q=\prod_{n=1}^{N} \frac{q_{n}}{1-q_{n}}$. For $Q>1$, I expect that:

\begin{align} f_{ij}&= 1 &\mathrm{if} & \ j>i \\ f_{ij}&<1 &\mathrm{if} & \ j \leq i \end{align}

Is there any way I can prove this? Help would be much appreciated.