0
$\begingroup$

enter image description here

This above example happens in an infinite state space $S = {i}_{i=1}$

By the definition of detailed balance condition,

Definition: $\pi\left ( x \right )p\left ( x,y \right ) = \pi\left ( y \right )p\left ( y,x \right )$

The author asserts that the initial distribution given in the example satisfies the detailed balance condition but this is not clear and when I try to attempt, I was unable to show this is true.

Secondly, there is an exponent $i$ in the explicit form for $\pi$; where does this exponent come from?

$\endgroup$
5
  • $\begingroup$ Is there no one who knows? The stochastic/ markov tag seems to be very slow moving... $\endgroup$ – Physkid Mar 8 '18 at 10:36
  • $\begingroup$ What initial distribution are you talking about? The author says that the chain is a birth-death process and thus will satisfy the detailed balance equations. Also, it's been only 3 hours since you asked the question. You must have more patience than that. $\endgroup$ – Greenparker Mar 8 '18 at 12:05
  • $\begingroup$ @Greenparker That explains. May I know why the exponent i exists? $\endgroup$ – Physkid Mar 8 '18 at 14:07
  • $\begingroup$ Welcome to CV @Physkid Sometimes good questions can literally take months or years to receive good answers on this site. Please have patience. :) $\endgroup$ – Alexis Mar 8 '18 at 16:53
  • $\begingroup$ @Alexis Thank you for telling me that. I was under the impression that the question was "unanswerable" due certain constraints. $\endgroup$ – Physkid Mar 9 '18 at 6:37
1
$\begingroup$

We have established that since this is an irreducible positive-recurrent birth-death process, it satisfies a detailed balance condition for a (unique) stationary distribution.

Thus for all states $i, j$, $\pi(i) P(i,j) = \pi(j) P(j,i)$, where $\pi$ is the unique stationary distribution of this process defined for all $i \geq 0$.

For $j = i+1$, $i \geq 0$ we have that \begin{align*} \pi(i) P(i,i+1) &= \pi(i+1) P(i+1,i) \\ \Rightarrow p \pi(i) &= (1-p)\pi(i+1) \\ \Rightarrow \pi(i+1) & = \pi(i) \left(\dfrac{p}{1-p}\right)\,. \end{align*}

This is what they do in the example. But note that $i$ is chosen generically, so this holds for all $i \geq 0$. That is, \begin{align*} \pi(i) & = \pi(i-1) \left(\dfrac{p}{1-p}\right)\\ & = \pi(i-2) \left(\dfrac{p}{1-p}\right)^2 \\ & =\pi(i-3) \left(\dfrac{p}{1-p}\right)^3\\ & \vdots\\ & = \pi(0) \left(\dfrac{p}{1-p}\right)^i\\ & = c\left(\dfrac{p}{1-p}\right)^i\,. \end{align*} So the exponent $i$ comes from an iterative argument.

$\endgroup$
1
  • $\begingroup$ Very clear now! $\endgroup$ – Physkid Mar 9 '18 at 6:46

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.