0
$\begingroup$

Let $X_1 ,X_2 ,X_3 ,…$ be a sequence of i.i.d. uniform $(0,1)$ random variables. Then, calculate the value of $$\lim_{n\to \infty}P(-\ln(1-X_1)-\ln(1-X_2)-\cdots-\ln(1-X_n)\geq n)?$$

My work:

Since $X_i$ are uniform $(0,1)$ r.v.s so $E(X_1)=\frac{1}{2}$ and $\operatorname{Var}(X_1)=\frac{1}{12}$. Now I tried to used Chebyshev's inequality but didn't work. These logarithms are confusing me. How should I approach the problem? Any help would be great. Thanks.

$\endgroup$
  • 2
    $\begingroup$ Hint: What is the distribution of $-\log(1-X_1)$? $\endgroup$ – Math1000 Jul 24 '16 at 14:44
  • 1
    $\begingroup$ @Math1000 I think exponential distribution with mean $1$ $\endgroup$ – Harry Potter Jul 24 '16 at 14:48
  • 1
    $\begingroup$ Hint: Use the CLT on the random variables $Y_k=-\log(1-X_k)-1$, independent, square integrable and centered. (Hence the desired limit is $\frac12$.) $\endgroup$ – Did Jul 24 '16 at 16:17
  • $\begingroup$ @Did Thanks for the hint. This is definitely more elegant than the revised argument I had been working on... $\endgroup$ – Math1000 Jul 24 '16 at 17:19
1
$\begingroup$

Recall that if $U\sim\mathsf U(0,1)$ and the distribution function $F_Y$ of $Y$ is absolutely continuous, then the random variable $F_Y^{-1}(U)$ has the same distribution as $Y$. In particular, if $F_Y(y) = (1-e^{-y})\mathsf 1_{(0,\infty)}(y)$ then $$F(Y) = U \iff Y = -\log(1-U). $$ Set $Y_k = -\log(1-X_k)-1$, then $Y_k\stackrel d= Y-1$ where $Y\sim\operatorname{Exp}(1)$, so $\mathbb E[Y_k]=0$ and $\mathbb E[Y_k^2]=1$ (and the $Y_k$ are independent since the $X_k$ are). Let $\xi_n = \frac1n\sum_{k=1}^n Y_k$, then $$\mathbb E[\xi_n] = \mathbb E\left[\frac1n\sum_{k=1}^n Y_k \right]=\frac1n\sum_{k=1}^n \mathbb E[Y_k] = 0$$ and $$\mathbb E[\xi_n^2] = \mathbb E\left[\left(\frac1n\sum_{k=1}^n Y_k\right)^2 \right] = \frac1{n^2}\mathbb E\left[\sum_{k=1}^n Y_k^2 \right]=\frac1{n^2}.$$ The central limit theorem then implies that $Z_n$ converges in distribution to the standard normal distribution, where $$Z_n = \frac{\xi_n-\mathbb E[Y_1]}{\operatorname{Var}(Y_1)/\sqrt n} = \sqrt n \xi_n. $$ It follows that \begin{align} \mathbb P\left(\sum_{k=1}^n-\log(1-X_k)\geqslant n \right) &= \mathbb P\left(\sum_{k=1}^n Y_k\geqslant 0\right)\\ &=\mathbb P\left(\frac1{\sqrt n}\sum_{k=1}^n Y_k\geqslant 0 \right)\\ &= \mathbb P\left(\sqrt n\xi_n\geqslant 0\right)\\ &= \mathbb P(Z_n\geqslant 0)\\ &\stackrel{n\to\infty}\longrightarrow 1-\Phi(0)=\frac12, \end{align} where $$\Phi(t) = \int_{-\infty}^t \frac1{\sqrt{2\pi}} e^{-\frac12 x^2}\,\mathsf dx;\ t\in\mathbb R $$ is the CDF of the standard normal distribution.

$\endgroup$
  • $\begingroup$ but the answer given is $1/2$ $\endgroup$ – Harry Potter Jul 24 '16 at 15:15

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.