0
$\begingroup$

Let $X_1... X_n$ be random samples from $Beta(\alpha, 1), \alpha\gt0$ and

Let $Y_n = \min X_{i\in\{1,2,..n\}} $

Then evaluate the condition of $r$ for the asymptotic distribution $n^rY_n$ to exist and specify the asymptotic distribution upon $r$.


I had evaluated

$$\lim\limits_{n\rightarrow \infty}P(n^rY_n \le y) = \begin{align}1 \;\;\;\text{when}\;\ 0\lt r\alpha\lt 1\\1-\exp(-y^\alpha) \;\;\;\text{when}\;r\alpha=1\\0\;\;\;\text{when}\;\;r\alpha\gt1\end{align}$$

Then now, How I need to specify another distribution $Z$ s.t. above given limit probability equal to $P(Z\le z)$ where z denotes every continuous point of cdf of $z$ and $Z = n^rY_n$

1) My first question is : How to specify the range of $Z$ about given $r$?

If $\;0\lt r\lt1/\alpha $, since $0\lt y \lt 1$, range of Z depends on $n^r$, however, I am confused with when it comes to deal with the case of form $\infty^r$.

2) I presumably guess for the second case of $r\alpha$ = 1, I can easily derive Z by just differentiating given $1-exp(-y^\alpha)$, however, what about $1$ and $0$ case? 1 corresponds to $0$ when differentiated so is it enough for me to say there exist no asymptotic distribution in case of $0<r\alpha<1$ and same conclusion with same reasoning to $r\alpha>1$ case?

$\endgroup$
2
$\begingroup$

You are almost done, just reflect what the question is about. It should be clear, that $Y_n$ converges to $0$ a.s.. In order to "see" something more, e.g. the asymptotics (how converges $Y_n$ to zero) you have to find a suitable normalization (Think of the standard central limit theorem: For integrable RVs, you know that $1/n\sum_{i\le n} X_i$ converges to the expectated value a.s.. The central limit theorem tells you, "how"). They already hinted you, that the right normalization should be of the form $n^r$ for some $r$. In fact you have done all the work. If you choose $r<\alpha^{-1}$, then your normalization is not "strong enough" and $n^rY_n\to0$, hence $Z=\delta_0$. (degenerated). For $r>\alpha^{-1}$ your normalization is "too strong", and $n^rY_n \to \infty$ (you could write $Z={\delta}_{\infty}$). The right normalization is $r=\alpha^{-1}$, then $n^rY_n$ converges in distribution to a nondegenerated $Z$, with distribution $1-\exp(-y^{\alpha})$ on $[0,1]$. in the other cases there are existing only degenerated distributions.

$\endgroup$
  • $\begingroup$ You explained me the real meaning of this problem. $\endgroup$ – Daschin Jul 20 '17 at 9:38

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.