Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Supposing $X_t$ is a Markov Process, can the transition kernel be defined by $$K_t(x,A):= P(X_{t+1} \in A | X_t = x)?$$ Assume that $X_t : \Omega \to \mathbb{R}^n$.

The issue is that under the normal definition of conditional probability, r.h.s is defined as $$P(X_{t+1} \in A | X_t = x) =\frac{P( (X_{t+1} \in A) \cap (X_t = x))}{P(X_t = x)}$$ and the denominator is zero for most random variables. Even if this is assumed to be $E[I_A(X_{t+1}) | \sigma(X_t = x)]$, the conditional expectation can take arbitrary values on the set $\{X_t =x\}$ if $P(X_t = x) =0$.

Another definition I could gather from the web is that $K_t(x,A)$ is called a transition kernel if $$K_t(X_t(\omega),A) := E[I_A(X_{t+1})|X_t](\omega)~\forall \omega \in \Omega.$$ Also, $K_t(x,\cdot)$ should be a probability measure so that the conditional expectations should be regular (if I am not wrong).

The book (page 18) I am reading uses the first definition given above.

Thanks for the help.

share|cite|improve this question
$P(X_{t=1}\in A | X_t = x)$ may be defined in the limit according to Wiki ( So, it seems that the first definition is also completely valid. – jpv Jul 19 '12 at 12:40
I don't see the issue. Remember that since $X_t$ is a Markov chain it posses no memory, so the RHS is equal to $P_x(X_1 \in A)=P(X_1 \in A | X_0 = x)$ and this is well defined. All you have to do is look at the row of the transition matrix corresponding to the state $x$. – chango Jul 19 '12 at 12:44
@did: Thanks for your answer! I was thinking about it but havent logged in for a long time. – jpv Aug 27 '12 at 7:19
up vote 4 down vote accepted

The transition kernel $K_t$ is defined by some measurability conditions and by the fact that, for every measurable Borel set $A$ and every (bounded) measurable function $u$, $$ \mathrm E(u(X_t):X_{t+1}\in A)=\mathrm E(u(X_t)K_t(X_t,A)). $$ Hence, each $K_t(\cdot,A)$ is defined only up to sets of measure zero for the distribution of $X_t$, in the following sense: if $K_t$ is a transition kernel for $X_t$ and if, for every measurable Borel set $A$, $X_t$ is almost surely in $C_A$, where $$ C_A=\{x\in\mathbb R^n\,\mid\,K_t(x,A)=\tilde K_t(x,A)\}, $$ then $\tilde K_t$ is also a transition kernel for $X_t$.

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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