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

Consider a Markov process $X_t$ with generator $L$ and invariant distribution $\pi$, whose distribution at time $t$ is given by $\pi(t,dx)=\phi(t,x) \pi(dx)$ - in other word, $\phi(t,x)$ is the density of $\pi(t, dx)$ wiht respect to the invariant distribution $\pi$. Define the (relative) entropy $$ S(t) = -\int \phi(t,x) \ln \phi(t,x) \pi(dx) \leq 0.$$

One can expect (Boltzman H-theorem) the entropy $S$ to increase over time, and eventually to converge to $0$.

question: what conditions should be imposed in order for such a result to be true ?

Fokker-Planck equation shows that for any test function $f$, $$ \int f(x) \partial_t \phi(t,x) \pi(dx) = \int (Lf)(x) \phi(t,x) \pi(dx)$$ so that $$S'(t) = -\int L (\ln \circ \phi)(x,t) \phi(x,t) \pi(dx), $$ but I still do not see why this quantity should be non-negative.

share|cite|improve this question

You can find the relevant calculation here.

share|cite|improve this answer
thank you very much for the reference. This seems to come from the fact that for a test function f, the following inequality L log(f) >= Lf/f is true. I can see that in the case of diffusions (for example), with generator of the form Lf = mu.f' + sigma^2.f'' but I still do not see why this should be true for a more general generator L. Is it ? – Alekk Apr 9 '10 at 15:06

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.