1: \begin{proof}

2: As the mean of $t_1$ is $1/\epsilon$, the random walk generated by

3: $\lambda_\epsilon$ is essentially the random walk generated by

4: $\mu_\epsilon$, but on a time scale $1/\epsilon$. This justifies

5: heuristically the statement.

6: 

7: For the rigorous proof, let us first check that $\lambda_\epsilon$

8: has finite first moment (and hence finite time one entropy). Since

9: all the measures have finite support, we have $\abs{L_1} \leq C t_1$.

10: Since a geometric distribution has moments of all order, the same is

11: true for $\abs{L_1}$.

12: 

13: The strong law of large numbers ensures that, almost surely, $t_N

14: \sim N/\epsilon$. Therefore, almost surely,

15:   \begin{equation*}

16:   \ell(\lambda_\epsilon) = \lim \frac{\abs{L_1 \dotsm L_N}}{N}

17:   = \lim \frac{\abs{g_1 \dotsm g_{t_N}}}{N}

18:   = \lim \frac{\abs{g_1 \dotsm g_{t_N}}}{t_N} \cdot \frac{t_N}{N}

19:   = \ell(\mu_\epsilon) \cdot 1/\epsilon.

20:   \end{equation*}

21: This proves the statement of the lemma for the drift.

22: 

23: For the entropy, we use the characterization of

24: Lemma~\ref{lem:charac_entropy}. We will show that $h(\mu_\epsilon)\leq

25: \epsilon h(\lambda_\epsilon)$ and $h(\mu_\epsilon)\geq \epsilon

26: h(\lambda_\epsilon)$. Let $K_n$ be a set of cardinality at most

27: $e^{(h(\mu_\epsilon)+\eta)n}$ which contains $g_1 \dotsm g_n$ with

28: probability at least $1/2$. Let $N=\epsilon n$. With large probability, $t_N$

29: is close to $n$, up to $\eta' n$ (where $\eta'$ is arbitrarily small). Hence,

30: with probability at least $1/3$, the point $L_1 \dotsm L_N$ belongs to the $C

31: \eta' n$-neighborhood of $K_n$, whose cardinality is at most

32:   \begin{equation*}

33:   \Card{K_n} \cdot e^{C' \eta' n}

34:   \leq e^{(h(\mu_\epsilon)+\eta+C'\eta')n}

35:   = e^{(h(\mu_\epsilon)+\eta+C'\eta')N/\epsilon}.

36:   \end{equation*}

37: As $\eta$ and $\eta'$ are arbitrary, this shows that $h(\lambda_\epsilon)

38: \leq h(\mu_\epsilon)/\epsilon$. The converse inequality is proved in the same

39: way.

40: \end{proof}