We have $0<\sigma<\infty$ and $\epsilon >0$ and $X_1, X_2,...$ iid. The argument involves CLT and it continues on until the line below:

That $\frac{1}{\sigma^2}E\big[(X_1-\mu)^2\mathbb{1}\big(|X_1-\mu|>\epsilon\sigma\sqrt(n)\big)\big]\rightarrow0$ by LDCT.

Can someone help me understand why this is the case? Something I can see is that if we consider $Z_n=X_n-\mu$ and by the iid property I know that $Var(Z_n)=Var(X_n-\mu)=E\big((X_n-\mu)^2\big)=\sigma$ as $E(Z_n)=0$.

And then $(X_1-\mu)^2\mathbb{1}\big(|X_1-\mu|>\epsilon\sigma\sqrt(n)\big)\leq\sigma$

So then I know that LDCT holds, and the limit of the integral is the integral of the limit.

So then why is this zero?


Quick edit:

Nevermind I think I see it.

It is because $\epsilon >0$ and $\mathbb{1}\big(\frac{|X_1-\mu|}{\sigma\sqrt{n}}>\epsilon\big)\rightarrow\mathbb{1}(0>\epsilon)$, which is over a null set.


In order to avoid the use of too much letters, let $$ Y:=\frac{\left\lvert X_1-\mu\right\rvert}{\sigma\varepsilon}.$$ The problem reduces to prove that $\mathbb E\left[Y^2\mathbf 1\{Y\gt \sqrt n\}\right]\to 0$ for each square integrable random variable, or letting $Z=Y^2$, that $\mathbb E\left[Z\mathbf 1\{Z\gt n\}\right]\to 0$ for each integrable random variable $Z$. There are two ways (which are not so fundamentally different) to show that.

  1. Let $Z_n:=Z\mathbf 1\{Z\gt n\}$. Then $0\leqslant Z_n\leqslant Z$ and $Z_n\to 0$ almost surely hence by the dominated convergence theorem, the expectation of $Z_n$ goes to zero.

  2. Let $Z'_n:=Z\mathbf 1\{Z\leqslant n\}$. Then $0\leqslant Z'_n\leqslant Z'_{n+1}$ for all $n$ and $Z'_n\to Z$ almost surely. We conclude from the monotone convergence theorem.


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.