5
$\begingroup$

I know it is $\text{Exp}(1)$ but I cannot prove this. (EDIT: I figured out an argument, I will draft a proof and share it here.)

$\endgroup$
  • 1
    $\begingroup$ it would be the same as demonstrating that $X_1/(X_1+X_2)$ is standard uniform. $\endgroup$ – Glen_b -Reinstate Monica Oct 31 '19 at 6:07
5
$\begingroup$

Taking $X_1,X_2 \sim \text{IID Exp}(1)$ we have:

$$R \equiv \ln \bigg( \frac{X_1+X_2}{X_1} \bigg) \sim \text{Exp}(1).$$

There are various ways to demonstrate this. In cases of difficulty, the simplest way is to derive the CDF. For all $r \geqslant 0$ we have:

$$\begin{equation} \begin{aligned} F_R(r) = \mathbb{P}(R \leqslant r) &= 1- \mathbb{P}( R > r ) \\[6pt] &= 1-\mathbb{P}( e^R > e^r ) \\[6pt] &= 1-\mathbb{P}( X_1+X_2 > X_1 \cdot e^r ) \\[6pt] &= 1-\mathbb{P}( X_2 > X_1 \cdot (e^r-1) ) \\[6pt] &= 1-\int \limits_0^\infty (1-F_{X_2}( x \cdot (e^r-1))) f_{X_1}(x) \ dx \\[6pt] &= 1-\int \limits_0^\infty \exp(-x \cdot (e^r-1)) \exp(-x) \ dx \\[6pt] &= 1-\int \limits_0^\infty \exp(-x \cdot e^r) \ dx \\[6pt] &= 1-\Big[ -e^{-r} \exp(-x \cdot e^r) \Big]_{x=0}^{x \rightarrow \infty} \\[6pt] &= 1- e^{-r}. \\[6pt] \end{aligned} \end{equation}$$

This confirms that $R \sim \text{Exp}(1)$.

| cite | improve this answer | |
$\endgroup$
3
$\begingroup$

Observe that $ln(\frac{X_1+X_2}{X_1})$ = $-ln(\frac{X_1}{X_1+X_2})$. Since $-ln({U})$ is exp(1) when U ~ Uniform(0,1). If we can prove $\frac{X_1}{X_1+X_2}$ is standard uniform then we will have completed the proof.

Observe that

$P(\frac{X_1}{X_1+X_2}\leq t) = P(\frac{X_1+X_2}{X_1}\leq \frac{1}{t}) = P\{X_2\leq X_1(\frac{1}{t}-1)\}= \int_0^\infty f_{X_1}(x_1)P(X_2 \leq x_1(\frac{1}{t}-1))dx = \int_0^{\infty}e^{-x_1}e^{-x_1(\frac{1}{t}-1)}dx_1 = \int_0^{\infty}e^\frac{x_1}{t}dx_1=t$

This implies $(\frac{X_1+X_2}{X_1})$ is uniform, completing the proof.

| cite | improve this answer | |
$\endgroup$

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.