1
$\begingroup$

I fail to understand Cinlar's transformation of an inhomogeneous Markov chain to a homogeneous one. It appears to me that $\hat{P}$ is not fully specified. Generally speaking, given a $\sigma$-algebra $\mathcal A$, a measure can be specified either explicitly over the entire $\sigma$-algebra, or implicitly by specifying it over a generating ring and appealing to Caratheodory's extension theorem. However, Cinlar specifies $\hat{P}$ over a proper subset of $\hat{\mathcal{E}}$ that is not a ring.

$\endgroup$
2
$\begingroup$

We give the condition that $\widehat P$ is a Markow kernel, and we have that $$\widehat P((n,x),\{n+1\}\times E)=P_{n+1}(x,E)=1,$$ hence the measure $\widehat P((n,x),\cdot)$ is concentrated on $\{n+1\}\times E\}$? Therefore, we have $\widehat P((n,x),I\times A)=0$ for any $A\subset E$ and $I\subset \Bbb N$ which doesn't contain the integer $n+1$.

$\endgroup$
  • $\begingroup$ Thank you, Davide. What if $\left\{n+1\right\}\subsetneq I$? $\endgroup$ – Evan Aad Aug 8 '12 at 13:01
  • 1
    $\begingroup$ In this case, the set $I\times E$ has necessarily measure $0$ by $\hat P((n,x),)$, since $\{n+1\}\times E$ concentrated all the measure. $\endgroup$ – Davide Giraudo Aug 8 '12 at 13:49
  • $\begingroup$ But $A:=\left\{n+1\right\}\times E \subseteq B:=I \times E$, so we should have $a:=\hat{P}\left(\left(n,x\right),A\right)\leq b:=\hat{P}\left(\left(n,x\right),B\right)$, since $\hat{P}$ is a probability transition kernel, whereas you claim that $a=1,\space b=0$. $\endgroup$ – Evan Aad Aug 8 '12 at 13:58
  • 1
    $\begingroup$ Sorry, I misread your comment. I believed you wrote "what if $\{n+1\}$ is not included in $I$", whereas you wrote "what if $\{n+1\}$ is contained in $I$ but not equal". In this case, we have $\widehat P((n,x),I\times A)=P_{n+1}(A)$. $\endgroup$ – Davide Giraudo Aug 8 '12 at 14:00
  • $\begingroup$ OK. I think i get it: $\hat{P}\left(\left(n,x\right), I\times A\right) = P_{n+1}\left(x,A\right)\cdot \mathbb 1_I\left(n+1\right)$. $\endgroup$ – Evan Aad Aug 8 '12 at 14:25

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.