0
$\begingroup$

I have troubles applying the Law of Large Numbers (abbreviated LLN) when the random variable is transformed. Let all $X_i$ be independent and identically distributed and $X_i\in L^2(P)$.

For example I've read that when $$S_n=X_1^2+\cdots + X_n^2$$ Then the LLN says that $$\sum\limits_{n=1}^\infty\frac{S_n}{n}=\mathbb{E}(X^2)$$

Does that mean when $f\in L^1$ $$S_n=f(X_1)+\cdots + f(X_n)$$ Then the LLN says that $$\sum\limits_{n=1}^\infty\frac{S_n}{n}=\mathbb{E}(f(X))$$


Second example. If I have $f(S_n)$ like for instance $$\frac{1}{\sqrt{1-(n^{-1}S_n)^2}}=\frac{1}{\sqrt{1-(n^{-1}(X_1+\cdots+X_n))^2}}$$

Does that mean that the LLN say that $$\frac{1}{\sqrt{1-(n^{-1}S_n)^2}}\xrightarrow{\text{a.s.}} \frac{1}{\sqrt{1-(\mathbb{E}(X))^2}}$$

Thank you for explanation.

$\endgroup$
  • $\begingroup$ Never use an acronym such as LLN without a full text explanation, for example in this way : "the Law of Large Numbers (subsequently abbreviated as LLN)...". $\endgroup$ – Jean Marie Sep 6 '16 at 10:42
  • $\begingroup$ @JeanMarie I changed it. $\endgroup$ – Matriz Sep 6 '16 at 10:44
  • 1
    $\begingroup$ I don't really understand your difficulty. What i can tel is that, whenever you have $f$ a deterministic function, $X_1,X_2,...$ a sequence of iid rv such that $f(X_1)\in L^2(P)$, then the following convergence holds: $(1/n)\sum f(X_i)\to E(f(X_1))$. $\endgroup$ – anonymus Sep 6 '16 at 10:50
  • $\begingroup$ where do you apply $f(S_n)$? you just have $f(X_i)$ as far as I can tell... $\endgroup$ – user190080 Sep 6 '16 at 10:54
  • $\begingroup$ @user190080 The first example is different from the second example. I have in the second one $f(x)=\frac{1}{\sqrt{1-(n^{-1}x)^2}}$ and then $f(S_n)$. $\endgroup$ – Matriz Sep 6 '16 at 10:59
2
$\begingroup$

The LLN says that, if $Y_1,\ldots,Y_n,\ldots$ are independent and identically distributed random variables in $L^1$, then $$\frac{1}{n}\sum_{k=1}^nY_k\stackrel{\text{a.s.}}{\rightarrow}\mu,$$ where $\mu=E[Y_i]$ for all $i$.

If you call $S_n=Y_1+\ldots +Y_n$, then $$\frac{S_n}{n}\stackrel{\text{a.s.}}{\rightarrow}\mu$$ (your notation with the $\sum_{n=1}^{\infty}$ is not correct).

If $Y_i=f(X_i)$, you need $f(X_i)\in L^1$ to apply the LLN and conclude that $$\frac{1}{n}\sum_{k=1}^nf(X_k)\stackrel{\text{a.s.}}{\rightarrow}E[f(X)].$$

In your final question, $n^{-1}S_n\rightarrow E[X]$ by the LLN. Assuming that $n^{-1}S_n\in (-1,1)$ and $E[X]\in (-1,1)$, as the function $x\mapsto 1/\sqrt{1-x^2}$ is continuous on $(-1,1)$, you can apply the continuous mapping theorem: $$ \frac{1}{\sqrt{1-(n^{-1}S_n)^2}}\stackrel{\text{a.s.}}{\rightarrow} \frac{1}{\sqrt{1-E[X]^2}}.$$

$\endgroup$
  • $\begingroup$ Thank you very much, this was very helpful! $\endgroup$ – Matriz Sep 6 '16 at 11:20

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.