Given an ergodic Markov chain $(X_n)_{n\geq 1}$ in $R^d$with $\pi$ as the invariant distribution of the transition kernel, under good conditions we have that the empirical occupation measure converges to $\pi$, i.e.:

\begin{align} \forall A\in \mathcal{B}(R^d), S^A_N = \frac{1}{N}\sum_{k=1}^{N}1_{X_k \in A} \to \pi(A), \text{ as } N \to +\infty \end{align}

We can even establish a CTL theorem for this convergence: $\sqrt{N}(S_N^A - \pi(A))$ converges to a centered normal distirbution

That's about the asymptotic behavoir. I am wondering if there exsits some non-asymptotic control of the convergence rate. in $L^p$ for example, if there any reference about controlling $||S_N^A - \pi(A)||_{L^p}$?

Thank you for any help!


1 Answer 1


Here is a finite time bound on $||S_N-\pi||_1$: https://dl.dropboxusercontent.com/u/3198145/chernoff-dkw.pdf Theorem 1.3

  • $\begingroup$ Thanks, I've taken a look, your paper is quite interesting. $\endgroup$
    – LIU
    Apr 11, 2014 at 7:33

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.