1
$\begingroup$

so let's assume we have independent random variables $X_1,X_2, X_3, \ldots$ with $$\mathbb{E}[X_k]=0 \mbox{ and } \mathbb{Var}[X_k]=\sigma_k^2=1 \quad \forall k\in\mathbb{N}. $$

We define $$s_n^2:= \sum_{k=1}^n \mathbb{E}[(X_k-\mathbb{E}[X_k])^2]=\sum_{k=1}^n\sigma_k^2=n.$$

Now we check Lindenberg's condition: So for $\varepsilon>0$ we must check $$\lim\limits_{n\rightarrow \infty} \frac{1}{s_n^2}\sum_{k=1}^n \mathbb{E}[(X_k-\mathbb{E}[X_k])^2 \cdot \mathbf{1}_{|X_k| > \varepsilon s_n^2} ] =\lim\limits_{n\rightarrow \infty} \frac{1}{n} \sum_{k=1}^n \mathbb{E}[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} ].$$

We know, $$\mathbb{E}[X_k^2 ]=1,$$ so $$\mathbb{E}[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} ]$$ converges quicker than $$\frac{1}{n},$$ whch means we find an $N\in\mathbb{N}$ such that for every $n\geq N$ we have
$$\mathbb{E}[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} ]< \frac{1}{n}, $$ which means for every $\varepsilon>0$ we have $$\lim\limits_{n\rightarrow \infty} \frac{1}{n} \sum_{k=1}^n \mathbb{E}[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} ] \leq \lim\limits_{n\rightarrow \infty} \frac{N}{n} + \frac{1}{n} \sum_{k=N+1}^n \mathbb{E}[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} ]< \lim\limits_{n\rightarrow \infty} \frac{N}{n} + + \frac{1}{n^2 }=0,$$ so we know that the Central Limit Theorem holds.

Is that right?

Edit: What we really need is uniform integrability of ${ X_k^2: k\in\mathbb{N}} $$ and that gives us that Lindeberg holds.

Otherwise we can construct $X_j$ with $$P(X_j =-j^2 )=P(X_j =j^2 ) = j^{-4}/2$$ and Lindeberg does not hold!

$\endgroup$
1
$\begingroup$

It may be not true that $\mathbb{E}\left[X_k^2 \cdot \mathbf{1}_{|X_k| > \varepsilon n} \right]$ is of order $1/n$. For example, if $X_1$ is such that its tail $\Pr(\left\lvert X_1\right\rvert >t)\sim t^{-2}(\log t)^{-2}$, then $\mathbb{E}[X_1^2 \cdot \mathbf{1}_{|X_1| > \varepsilon n} ]$ is of order $1/\log n$.

Note that it would suffices to have uniform integrability, that is, $$ \lim_{R\to +\infty}\sup_{k\geqslant 1}\mathbb{E}\left[X_k^2 \cdot \mathbf{1}_{|X_k| > R} \right]=0. $$

Otherwise, the are counter-examples.

$\endgroup$
  • $\begingroup$ I think we have uniform integrability, because $\sigma_k^2=1$, but we haven't unform integrability of the family $X_k^2$, right? $\endgroup$ – cptflint Dec 11 '18 at 23:24
  • $\begingroup$ We indeed have integrability of each $X_k$ but in general not uniform integrability, unless you have other assumptions. $\endgroup$ – Davide Giraudo Dec 12 '18 at 9:44
  • $\begingroup$ I think we need $\exists p>2, C>0: \left[ \forall k \in\N: \EW[|X_k|^p ] \leq C \right]$, right? Then I have unfirom integrability of $X_k^2$, which means that we can use Lindeberg. $\endgroup$ – cptflint Dec 12 '18 at 9:58

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.