3
$\begingroup$

Let $Y_1,Y_2,...,$ be independent $C(0,1)$ random variables, determine the limit distribution of :

$Z_n = \dfrac{1}{n} \cdot max\{Y_1, Y_2,..Y_n \} $ as $n \rightarrow \infty $,

Here is my approach:

$F_{Z_n} (x) = \mathbb{P}(Z_n \leq x) = \mathbb{P}(\dfrac{1}{n} max\{ Y_1,Y_2,...Y_n \} \leq x ) = \mathbb{P}( Y_1 \leq x\cdot n , Y_2 \leq x\cdot n, ....,Y_n \leq x\cdot n ) = \Big(F_{Y}(x\cdot n)\Big)^n $ (where $Y \in C(0,1)$ )

I have used that they are independent and identically distributed.

The next step is to find $F_Y(x\cdot n)$ which can be found by integration of the density of $C(0,1)$ , that is $F_Y(x\cdot n) = \dfrac{1}{\pi} \int_{-\infty}^{x\cdot n} \dfrac{1}{1+x^2} =\dfrac{1}{\pi} \big( arctan(x \cdot n) + \dfrac{\pi}{2} \big)$ .

Here is where Iam stuck , however, i know that $\dfrac{\pi}{2} = arctan(u) + arctan(1/u) $ and that $arctan(z) \approx z - \dfrac{z^3}{3} + \dfrac{z^5}{5} + ...$

I tried to use it somehow to get some "nice" expression for: $\dfrac{1}{\pi} \big( arctan(x \cdot n) + \dfrac{\pi}{2} \big)$, but failed

With "nice" i refer to the fact that $lim_{n \rightarrow \infty} \Big(F_{Y}(x\cdot n)\Big)^n $ could be recognized "easily"

Can someone give me a helping hand ?

$\endgroup$
  • $\begingroup$ Perhaps using $F_{Z_n}$, you compute the characteristic function, and then let $n\to\infty$ to see if a pattern emerges. $\endgroup$ – Kim Jong Un Sep 16 '14 at 18:31
2
$\begingroup$

Your approach is correct. You obtain that $F_Y(nx) = 1 - \arctan (1/nx) / \pi = 1 - 1/(\pi n x) + \ldots$, so you deduce an approximation of $\log F_Y(nx)$ since $\log(1+x)$ is equivalent to $x$, and you get the result.

For a more general approach to the problem of finding the distribution of the maximum of random variables, see the sub-area of probability and statistics called Extreme Value Theory.

$\endgroup$
  • $\begingroup$ what do mean with since $log(1+x)$ is equivalent to $x$ , (surly Iam missing something) @Paul $\endgroup$ – Danny Sep 16 '14 at 21:46
  • $\begingroup$ It means that $\log(1+x)/x$ tends to $1$ when $x$ goes to $0$. So here $\log F_Y(nx)$ is equivalent to $F_Y(nx) - 1$ which itself is equivalent to $-1/(\pi nx)$ when $n$ goes to $\infty$. $\endgroup$ – polmath Sep 16 '14 at 21:55
  • $\begingroup$ My goal is to get the limit of $\Big(F_{Y}(x\cdot n)\Big)^n$ as $n \rightarrow \infty$ , i don't see how maybe you could elaborate on this @Paul $\endgroup$ – Danny Sep 16 '14 at 22:23
  • $\begingroup$ I could approximate $F_Y(nx) \approx 1 - \dfrac{1}{\pi n x}$ and then find the limit of $ \Big(1-\dfrac{1}{\pi n x} \Big)^n$ as $n\rightarrow \infty$ , however this approximation is quiet rough isn't it $\endgroup$ – Danny Sep 16 '14 at 22:51
  • 1
    $\begingroup$ To conclude, you have $F_Y(n x)^n = \exp(n \log F_Y(n x))$, and the fact that $\log F_Y(n x)$ is equivalent to $-1/(\pi n x)$ shows that $n \log F_Y(n x)$ tends to $-1/(\pi x)$; hence that $F_Y(n x)^n$ tends to $\exp(-1/(\pi x))$. $\endgroup$ – polmath Sep 17 '14 at 7:41

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.