Let $(X_i)$ be a sequence of (possibly dependent) random variables in $\mathbf{R}_+$ such that $\mathbb{E}[X_i]<\infty$ and $\mathbb{E}[X_i^2]<\infty$. Assume that almost surely $$ \lim_{N\to\infty} \frac{1}{N} \sum_{n=1}^N X_i = X $$ where $X$ is a (finite) deterministic number. I'm trying to understand if I can conclude that $$ \lim_{N\to\infty} \frac{1}{N} \sum_{n=1}^N \mathbb{E} X_i = X. $$ And in case this is not true, which other conditions am I missing.
If $\lim_{N\to\infty} \frac{1}{N} \sum_{n=1}^N X_i = X$ a.s. with $X$ constant, $\lim_{N\to\infty} \frac{1}{N} \sum_{n=1}^N \mathbb{E} X_i = X$?
0
$\begingroup$
probability
probability-theory
convergence
random-variables
probability-limit-theorems
1 Answers
2
This is in general incorrect.
Let the $X_i$ be independent with $P(X_i = i^2) = i^{-2}$ and $P(X_i = 0) = 1 - i^{-2}$.
By the Borel-Cantelli lemma, we have $P(X_i = i^2 \text{ i.o.}) = 0$, from which we can conclude that $\frac{1}{n} \sum \limits_{i = 1}^n X_i$ converges to $0$ almost surely. But $E[X_i] = 1$, which means $\frac{1}{n} \sum \limits_{i = 1}^n E[X_i] = 1 \not\to 0$.
-
0Thank you. What if the $X_i$'s have a density: can this argument be generalized? – 2017-02-02
-
1You can also find examples where the $X_i$ have a density. Just replace $X_i$ with $X_i + |N_i|$, where the $N_i$ are independent standard normals. By the law of large numbers, $\frac{1}{n} \sum \limits_{i = 1}^n |N_i|$ converges almost surely to $E[|N_i|]$. – 2017-02-02