MathOverflow is a question and answer site for professional mathematicians. 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

Hi this Question follows after the answer of Douglas Zare to this post :

So let's be given a positive function $V(x)$ which is smooth enough to have Lispchitzian derivative at all point. Moreover this function admits multiple zeros denoted $(\theta_1,...,\theta_n)$ and let $X_t$ the diffusion process following the stochastic equation below :

$dX_t=V'(X_t)dt + \sigma.dW_t$ where now $\sigma$ is a constant (but will later be replaced by some function of $X_t$) and $X_0=x$

Here are my questions :

1-What are the conditions on $V$ for $X_t$ :

a- to admit a stationary distribution ? and does this distribution can be explicited in terms of $V,x,\theta_1,...,\theta_n$, and $\sigma$ ?

b- to be an ergodic process ?

2-What are the smoothness conditions on $\sigma$ (now a function of $X_t$) to keep 1/a- or 1/b- true ?

Of course any reference is welcome if the answer is too long for a single post.

Best Regards

share|cite|improve this question
It would probably be a good idea to make the title more descriptive. – Qiaochu Yuan Jan 31 '11 at 13:48
Is it better now ? – The Bridge Jan 31 '11 at 14:22
up vote 5 down vote accepted

Let us assume you are interested in the solution of $dX_t=\sigma(X_t)dW_t+b(X_t)dt$. If the so-called speed measure $m$ has finite mass, the diffusion converges in distribution to $m/|m|$. Here, $m(x)=2/(\sigma^2(x)s'(x))$ where $s$ is the so-called scale of the diffusion. Recall that $s$ is uniquely defined, up to affine transformations, as the solution of the differential equation $\frac12\sigma(x)^2s''(x)+b(x)s'(x)=0$.

In your case, $\sigma$ is constant and $b(\cdot)=V'(\cdot)$, hence $s'(x)=\exp(-2V(x)/\sigma^2)$ up to a multiplicative constant. The condition for ergodicity is that the function $\exp(2V(\cdot)/\sigma^2)$ is integrable. In that case, the stationary distribution is proportional to this function.

share|cite|improve this answer
And the generalization to non constant functions $\sigma(\cdot)$ is straightforward. – Did Jan 31 '11 at 15:00
Thanks,great and very clear answer. By the way do you have some references about "speed measures" and "scale functions" that you would recommand ? Regards – The Bridge Jan 31 '11 at 15:18
I was afraid you would ask this question... A classic reference is the venerable Diffusions, Markov Processes and Martingales: Volume 2, Itô Calculus by Rogers and Williams but it might take time to digest. Durrett, as often, is most congenial and you could try his book Stochastic calculus (which I believe has a whole chapter devoted to 1D diffusions, but I do not have the book at hand so, if the chapter exists it will deal with this stuff and otherwise, well, I apologize in advance). – Did Jan 31 '11 at 15:52
Please do not apologize, I know this reference but couldn't remember that it has the subject treated. I should re-read it then. Regards – The Bridge Jan 31 '11 at 16:07

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.