1: \begin{definition}

2: Let $\SM$ be a countable (either finite or countably infinite) set, and let

3: $(S_n)$ be a discrete time Markov chain on state space $\SM$ with transition

4: matrix $\KM = \{\KM(s,s')\}_{s, s' \in \SM}$. The Markov chain $(S_n)$ is

5: said to be \emph{ergodic} if it is irreducible, aperiodic, and positive

6: recurrent. In this case (see \cite[Theorem 21.14]{Levin2009}), there exists a

7: unique stationary distribution $\pi$ on $\SM$ satisfying $\pi = \pi \KM$, and

8: \begin{align}

9: \label{eq:ErgodicMCDef}

10: \lim_{n \to \infty} \norm{ \KM^n(s, \cdot) - \pi}_{TV} = 0, \mbox{ for each } s \in \SM,

11: \end{align}

12: where $\norm{\mu - \nu}_{TV} \equiv \frac{1}{2} \norm{\mu - \nu}_1$ is the

13: total variational norm between two probability distributions $\mu$ and $\nu$

14: on $\SM$. The Markov chain $(S_n)$ is said to be \emph{uniformly ergodic} if

15: the rate of convergence in \eqref{eq:ErgodicMCDef} is uniform in the initial

16: state. That is, if

17: \begin{align}

18: \label{eq:UniformlyErgodicMCDef}

19: \lim_{n \to \infty} \Big[ \sup_{s \in \SM} \norm{ \KM^n(s, \cdot) - \pi}_{TV} \Big] = 0.

20: \end{align}

21: \end{definition}

22: