0
$\begingroup$

Let $X_1, X_2, \cdots$ be independent random variables, each with exponential distribution with parameter $\lambda = 1$. For any $n ≥ 1$, let $Y_n := \max(X_1,\cdots, X_n)$. Let $0 < a < 1 < b$. Show that $P(Y_n ≤ a \log n) \to 0$ as $n \to \infty$, and $P(Y_n ≤ b \log n) \to 1$ as $n \to \infty$. Conclude that $Y_n/ \log n$ converges to $1$ in probability as $n\to \infty$.

I have no idea how to approach this problem. I know the formula for convergence in probability, but this seems foreign to me.

$\endgroup$
2
$\begingroup$

Hint: $$P(\max_{i\leq n} X_i \leq u) = \prod_{i=1}^nP(X_i\leq u)$$ This is because the max of a bunch of things less than $u$ is equivalent to all those things less than $u$. The product form happens because the variables are independent.

Can you work out the rest? If you need more help, let me know in comments.

Update: So basically you have to evaluate $(1-e^{-u})^n$ which happens to be the RHS above.

Put $u=a\log n$. You get $$\left(1-\frac{1}{n^a}\right)^n = \left(1-\frac{1}{n^a}\right)^{n^a(n/n^a)}$$

Now we know $(1-\frac{1}{n})^n\to e^{-1}$. Hence as $n^a\to \infty$, we get $(1-\frac{1}{n^a})^{n^a}\to e^{-1}$. Now $e^{-1} < 1$, so we have that for any $\delta>0$, for $n$ large enough $$e^{-1}-\delta\le \left(1-\frac{1}{n^a}\right)^{n^a} \leq e^{-1}+\delta$$ Here pick any $\delta$ such that $e^{-1}+\delta < 1$. Now $$0\leq \left(1-\frac{1}{n^a}\right)^{n^a(n/n^a)} \leq (e^{-1}+\delta)^{n^{1-a}}\to 0 $$

For the $b$ case, use a similar argument except you need the lower bound now. Just pick $\delta$ such that $e^{-1}+\delta>0$ and note that probabilities cannot exceed $1$.

$\endgroup$
  • $\begingroup$ I was able to show the second part, how Yn/log(n) converges to 0. This is because P(Yn>(e+1)logn) + P(Yn<(1-e)logn) = 1 - 1 + 0 = 0, as e+1 is the case where the coefficient = b (thus ->1) and 1-e is the case where the coefficient = a (thus ->0). I'm still having trouble though proving why probability with a coef converges to 0 and with b coef converges to 1. I'm stuck on finding the limit of P(xi<alog(n))^n. $\endgroup$ – Lebron James Feb 16 '17 at 19:20
  • $\begingroup$ Good. I'll update my answer to assist you with the part you were stuck on. $\endgroup$ – Gautam Shenoy Feb 16 '17 at 19:42
  • $\begingroup$ Thank you, I think I understand it now. $\endgroup$ – Lebron James Feb 16 '17 at 20:20

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.