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

Let $(X(t))_{t\geq0}$ be a Markov proces such that s $X(0)=x_0$. Consider the random vector $(X(t_1),\dots,X(t_n))$ with corresponding joint density $g(x_1,\dots ,x_n)$. Is it then true that $g(x_1,\dots x_n)=p(0,t_1,x_0,x_1)\cdot p(t_1,t_2,x_1,x_2) \cdots p(t_{n-1},t_n,x_{n-1},x_n)$ where $p(a,b,c,d)$ is the corresponding transition density between time a to b and state c to d.

If so I would love a proof or a reference. I know a bit of theory about Markov processes, but am not perfectly comfortable with transition densities. I tried proving it just using the Markov property, but without great success.

share|cite|improve this question
The obvious: iterate this. – Did Sep 12 '13 at 11:51
Will this work with an uncountable state space? – Henrik Sep 12 '13 at 13:52
Why not? (Say, you erased a comment, no?) – Did Sep 12 '13 at 13:54
Is the left hand side not just 0 always? I thought I understood it (indicated it in a comment), worked with it and suddenly I felt indicating that I understood it was a bit misleading :) – Henrik Sep 12 '13 at 14:04
Hmmm... how do you define p(t,s,x,y) in the first place? The trouble you have seems to come from undigested aspects of THAT. – Did Sep 12 '13 at 14:07

The question might be based on a confusion between transition probabilities and transition densities.

In the first case the conditional distribution of $X(t+s)$ conditionally on $\mathcal F_s=\sigma(X(u);u\leqslant s)$ is such that $$ P[X(t+s)=y\mid X(s)=x,\mathcal F_s]=p(s,t+s,x,y), $$ for every states $(x,y)$. Thus, every entry $p(s,t+s,x,y)$ is nonnegative and, for every $x$, $$ \sum_yp(s,t+s,x,y)=1. $$ This implies that, for every $x$, the set $\{y\mid p(s,t+s,x,y)\ne0\}$ is at most countable.

In the second case the conditional distribution of $X(t+s)$ conditionally on $\mathcal F_s$ is such that $$ P[X(t+s)\in B\mid X(s)=x,\mathcal F_s]=\int_Bp(s,t+s,x,y)\mathrm dy, $$ for every state $x$ and every (measurable) set $B$. Since the conditional distribution of $X(t+s)$ is absolutely continuous, for every $y$, $$ P[X(t+s)=y\mid X(s)=x,\mathcal F_s]=0, $$ but it may well happen that $\{y\mid p(s,t+s,x,y)\ne0\}$ is the whole state space, even when said state space is uncountable.

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.