The Problem: Let $X$ and $Y$ be independent exponential random variables with parameter $\lambda=1$.
$\textbf{a)}$ Calculate the probability $P(Y\geq X\geq2)$.
$\textbf{b)}$ Find the density function of the random variable $Z=X-Y.$

My Attempt:
a) By hypothesis, we have that the density functions of $X$ and $Y$, respectively, are given by $$f_X(x)=\begin{cases}e^{-x}&\text{if }x\geq0\\0&\text{otherwise.}\end{cases}\quad f_Y(y)=\begin{cases}e^{-y}&\text{if }y\geq0\\0&\text{otherwise.}\end{cases}$$ Since $X$ and $Y$ are independent, it follows that $$f_{XY}(x,y)=f_X(x)\cdot f_Y(y)=\begin{cases}e^{-(x+y)}&\text{if }x\geq0,\,y\geq0\\0&\text{otherwise.}\end{cases}$$ Now to find the probability in question we integrate the joint density function over the region $D=\{(x,y)\in\mathbb R^2:2\leq x\leq y\}.$ We have \begin{align} P(Y\geq X\geq2)&=\iint_D f_{XY}(x,y)\,dy\,dx=\int_2^\infty\int_x^\infty e^{-(x+y)}\,dy\,dx\\ &=\int_2^\infty e^{-2x}\,dx\\ &=\frac{1}{2e^4}. \end{align} b) We can first find the CDF of $Z$ using the joint PDF of $X$ and $Y$ by integrating the latter function over the region $D=\{(x,y)\in\mathbb R^2:x-y\leq t\}$ where $t\in\mathbb R.$ If $t\leq0$, then \begin{align*} P(X-Y\leq t)&=\iint_D f_{XY}(x,y)\,dy\,dx=\int_0^\infty\int_{x-t}^\infty e^{-(x+y)}\,dy\,dx\\ &=e^t\int_0^\infty e^{-2x}\,dx\\ &=\frac{e^t}{2}. \end{align*} On the other hand, if $t>0$, then \begin{align*} P(X-Y\leq t)&=\iint_D f_{XY}(x,y)\,dy\,dx\\&=\int_0^t\int_0^\infty e^{-(x+y)}\,dy\,dx+\int_t^\infty\int_{x-t}^\infty e^{-(x+y)}\,dy\,dx\\ &=\int_0^t e^{-x}\,dx+\int_t^\infty e^t e^{-2x}\,dx\\ &=1-e^{-t}+\frac{e^{-t}}{2}\\ &=1-\frac{e^{-t}}{2}. \end{align*} Therefore, we have the following case-defined function as the CDF of $Z$, $$F_Z(t)=\begin{cases}\dfrac{e^t}{2}&\text{if }t\leq0\\[0.8em]1-\dfrac{e^{-t}}{2}&\text{if }t>0.\end{cases}$$ Since the CDF above is continuous everywhere and differentiable almost everywhere, we can differentiate it to find the PDF. We have $$f_Z(t)=\begin{cases}\dfrac{e^t}{2}&\text{if }t\leq0\\[0.8em]\dfrac{e^{-t}}{2}&\text{if }t>0.\end{cases}$$

Do you agree with my approach and execution above?
Thank you very much for your time and I sincerely appreciate any feedback.

  • 1
    $\begingroup$ everything is correct but you wrote wrongly the definition of the regions $D$ because $(x,y)\in[0,\infty)^2$ instead of $\mathbb{R}^2$ but the calculations seems fine $\endgroup$ – Masacroso Jul 15 '20 at 19:39

More generally, let $X\sim\mathrm{Expo}(\lambda)$ and $Y\sim\mathrm{Expo}(\mu)$ be independent and let $t>0$ be arbitrary. Then \begin{align} \mathbb P(Y\geqslant X\geqslant t) &= \int_{(t,\infty)\times(t,y)}f_{X,Y}(x,y)\ \mathsf d(x\times y)\\ &=\int_t^\infty \int_t^y \lambda e^{-\lambda x}\mu e^{-\mu y}\ \mathsf dx\ \mathsf dy\\ &= \frac{\lambda }{\lambda +\mu } e^{-(\lambda +\mu )t}. \end{align} Also, note that $$ \{Y\geqslant X\geqslant t\} = \{Y\geqslant X\}\cap\{X\wedge Y\geqslant t\}, $$ and hence \begin{align} \mathbb P(Y\geqslant X\geqslant t) &= \mathbb P(Y\geqslant T)\mathbb P(X\wedge Y\geqslant T)\\ &= \frac{\lambda }{\lambda +\mu } e^{-(\lambda +\mu )t}. \end{align} Substituting $\lambda=\mu=1$ and $t=2$ yields $\frac12 e^{-4}$, the same as your result.

To compute the density of $Z=X-Y$, we use convolution. First note that the density of $-Y$ is $f_{-Y}(y) = \mu e^{\mu y}\cdot\mathsf 1_{(-\infty,0)}$. Then for $t\in\mathbb R$ we have \begin{align} f_Z(t) &= f_X\star f_{-Y}(t)\\ &= \int_{\mathbb R} f_X(s)f_{-Y}(t-s)\ \mathsf ds\\ &= \int_{t\vee 0}^\infty \lambda e^{-\lambda s}\mu e^{\mu(t-s)}\ \mathsf ds\\ &= \frac{\lambda \mu }{\lambda +\mu }\left(e^{\mu t}\cdot\mathsf 1_{(-\infty,0)}(t)+e^{-\lambda t}\cdot\mathsf 1_{[0,\infty)}(t) \right). \end{align} Substituting $\lambda=\mu=1$ yields $$ f_Z(t) = \frac12\left(e^t\cdot\mathsf 1_{(-\infty,0)} + e^{-t}\mathsf 1_{[0,\infty)}(t)\right), $$ the same as your result. In this case, $Z$ has a Laplace distribution.


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.