Given an ergodic property that guarantees convergence of sample means to an expectation, how can I bound the Cesàro Mean of expectation of terms? I have a sequence of (not iid) random variables $\{E_{i}\}$ that converge in distribution (actually, in total variation) as well as in $\mathcal{L}^{2}$ to $E_{\infty}$. For all nonnegative functions $f$ with $\mathbb{E}[f(E_{\infty})]<\infty$, it's established that
\begin{align}
\lim_{T\rightarrow\infty}\frac{1}{T}\sum_{n=1}^{T}f(E_{n}) \overset{\mathrm{a.s.}}{=}\mathbb{E}[f(E_{\infty})].
\end{align} I would like to bound (again for a nonnegative $f$)
\begin{align}
\underset{T\rightarrow\infty}{\lim\sup}\text{ }\frac{1}{T}\sum_{n=1}^{T}\mathbb{E}[f(E_{n})].
\end{align}
In my particular case, I've been unable to uniformly bound the random variables $f(E_{i})$, so I cannot use, for example, the reverse Fatou's Lemma to interchange the limit and expectation to show something like $\underset{T\rightarrow\infty}{\lim\sup}\text{ }\mathbb{E}[\frac{1}{T}\sum_{n=1}^{T}f(E_{n})] \le \mathbb{E}[\underset{T\rightarrow\infty}{\lim\sup}\text{ }\frac{1}{T}\sum_{n=1}^{T}f(E_{n})]$. Does anyone have any advice for how I could try to proceed-- in particular ideas for how to interchange the limit and expectation without some kind of uniform bound?
The thing is that I don't much about the function $f$. I do know some potentially relevant facts. I can prove that $\mathbb{E}[f(E_{n})]$ is always finite.
 A: As in my comments, the best possible upper and lower bounds are:
$$ E[f(E_{\infty})] \leq \limsup_{T\rightarrow\infty}\frac{1}{T}\sum_{n=1}^TE[f(E_n)] \leq \infty$$

Lower bound: Suppose $f$ is a nonnegative function such that
$$ \lim_{T\rightarrow\infty}\frac{1}{T}\sum_{n=}^T f(E_n) = E[f(E_{\infty})] \quad \mbox{almost surely}$$
We have
\begin{align}
\limsup_{T\rightarrow\infty}\frac{1}{T}\sum_{n=1}^TE[f(E_n)] &\geq \liminf_{T\rightarrow\infty}\frac{1}{T}\sum_{n=1}^TE[f(E_n)]\\
&\overset{(a)}{\geq} E\left[\liminf_{T\rightarrow\infty}\frac{1}{T}\sum_{n=1}^Tf(E_n) \right]\\
&= E[f(E_{\infty})]
\end{align}
where (a) holds by Fatou's lemma for sequences of nonnegative random variables.  The lower bound is achievable by defining $X_n=X_{\infty}=0$ for all $n$.

Upper bound:
Define $Y$ geometric with success probability $1/4$.  Define
$$E_n=(1-\frac{1}{2^n})1\{Y>n\} \quad \forall n \in \{1, 2, 3, ...\}$$
Then we surely have $0\leq E_n\leq 1$ for all $n \in \{1, 2,3, \ldots\}$, and we surely have $E_n=0$ for all sufficiently large $n$. It follows that $E_n\rightarrow 0$ surely (and in $L_2$). Also, for any real-valued function $g$ we surely have $g(E_n)=g(0)$ for all sufficiently large $n$, and so $\frac{1}{T}\sum_{n=1}^T g(E_n)\rightarrow g(0)$ surely.
Now define the continuous and nonnegative function $f:[0,1)\rightarrow \mathbb{R}$ by
$$ f(x) = -1 + \frac{1}{1-x}$$
Since $f(0)=0$ we surely have $f(E_n)=0$ for all sufficiently large $n$. However
\begin{align}
E[f(E_n)] &= E\left[f\left((1-\frac{1}{2^n})1\{Y>n\}\right)\right] \\
&= E\left[f\left(1-\frac{1}{2^n}\right)|Y>n\right](3/4)^n + E[f(0)|Y\leq n]P[Y\leq n] \\
&= (3/4)^nf(1-\frac{1}{2^n})\\
&= (3/4)^n\left(-1 + 2^n\right)
\end{align}
and so $E[f(E_n)]$ is finite for all $n$, but $\frac{1}{T}\sum_{n=1}^T E[f(E_n)]\rightarrow \infty$.
