So this is the problem

Let $X_1$ and $X_2$ be $i.i.d.$ random variables with a common density function $\chi_2^2$. Let $Y_1 = \frac{1}{2}(X_1-X_2)$ and $Y_2 = X_2$. Show that $Y_1$ follows a Laplace distribution with density function given by $$f(y) = \frac{1}{2}e^{-|y|} $$ for $y \in \mathbb{R}.$

Now, here is my reasoning:

Since $X_1,X_2 \sim \chi_2^2$ (Chi-squared distribution) , then each of these random variables have their density function given by $$f(x)= \frac{1}{2} e^{-x/2}$$ for $x >0.$ Now,we have the transformations:

$$y_1 =g_1(x_1,x_2) = \frac{1}{2}(x_1-x_2),\ \ \ \ \ y_2 =g_2 (x_1,x_2)= x_2$$ and their inverses $$x_1 = g_1^{-1}(y_1,y_2) = 2y_1+y_2, \ \ \ \ \ x_2 = g_2^{-1} (y_1,y_2)= y_2. $$

Now we obtain the Jacobian of the transformation, which is $$J =\begin{vmatrix} 2 & 1 \\ 0 & 1 \end{vmatrix} = 2$$ Thus the joint density of $Y_1,Y_2$ is given by $$f_{Y_1,Y_2}(y_1,y_2) = f_{X_1,X_2}(x_1,x_2) \cdot |J|$$ and since $X_1$ and $X_2$ are independent we have that \begin{align*} f_{Y_1,Y_2}(y_1,y_2) &= f_{X_1}(x_1) \cdot f_{X_2}(x_2) \cdot 2\\ &= 2 \cdot \left(\frac{e^{-(2y_1+y_2)/2}}{2} \right)\cdot \left(\frac{e^{-y_2/2}}{2} \right)\\ &= \frac{1}{2}\cdot e^{-(y_1+y_2)}. \end{align*} Now, we are asked to find the distribution of $Y_1$, so \begin{align*} f_{Y_1}(y_1) &= \int_{0}^{\infty} f_{Y_1,Y_2}(y_1,y_2) dy_{2}\\ &= \int_{0}^{\infty} \frac{1}{2}\cdot e^{-(y_1+y_2)}dy_{2}\\ &=\frac{1}{2}\cdot e^{-y_1} \cdot \int_{0}^{\infty}e^{-y_2}dy_{2} \end{align*} and since $ \int_{0}^{\infty}e^{-y_2}dy_{2} = 1$ we can conclude that $$f_{Y_1}(y_1)= \frac{1}{2}\cdot e^{-y_1}.$$

Clearly this is not entirely correct since the $y$ exponent should be in absolute value. If anyone could point out my mistake and/or provide some intuition as to how the absolute value plays a role in this problem, I would appreciate it very much!

  • 2
    $\begingroup$ When you plug in $f_{X_1}(x_1) = f_{X_1}(2y_1 + y_2)$, you forget to check the support of $X_1$ that requires $2y_1 + y_2 > 0$, which in turns give $y_2 > -2y_1$. Combine with $y_2 > 0$, it gives the joint support of $(Y_1, Y_2)$ to be $y_1 \in \mathbb{R}, y_2 > \max\{0, -2y_1\}$. Once you replace your integration lower limit by $\max\{0, -2y_1\}$ and you should get your desired result. $\endgroup$ – BGM Nov 20 '18 at 7:30

Do not forget to include the proper supports of the distributions. Use indicator functions for brevity.

Joint density of $(X_1,X_2)$ is

\begin{align} f_{X_1,X_2}(x_1,x_2)&=\frac{1}{2}e^{-x_1/2}\mathbf1_{x_1>0}\,\frac{1}{2}e^{-x_2/2}\mathbf1_{x_2>0} \\&=\frac{1}{4}e^{-(x_1+x_2)/2}\mathbf1_{x_1,x_2>0} \end{align}

For the transformation $(X_1,X_2,)\to (Y_1,Y_2)$ such that $$Y_1=\frac{1}{2}(X_1-X_2)\quad,\quad Y_2=X_2\,,$$

we have $$x_1=2y_1+y_2\quad,\quad x_2=y_2$$

So, $$x_1,x_2>0\implies 2y_1+y_2>0,y_2>0$$

In other words, $$y_2>\max(0,-2y_1)\quad\text{ where }\quad y_1\in\mathbb R$$

Absolute value of the jacobian as you say is $$|J|=2$$

So the joint density of $(Y_1,Y_2)$ is

\begin{align} f_{Y_1,Y_2}(y_1,y_2)&=f_{X_1,X_2}\left(2y_1+y_2,y_2\right)|J|\,\mathbf1_{2y_1+y_2>0,y_2>0} \\&=\frac{1}{2}e^{-(y_1+y_2)}\mathbf1_{2y_1+y_2>0,y_2>0} \end{align}

Hence the marginal density of $Y_1$ for $y_1\in\mathbb R$ is given by

\begin{align} f_{Y_1}(y_1)&=\int_{\max(0,-2y_1)}^\infty f_{Y_1,Y_2}(y_1,y_2)\,\mathrm{d}y_2 \\\\&=\frac{e^{-y_1}}{2}\lim_{A\to\infty}\left[-e^{-y_2}\right]_{\max(0,-2y_1)}^A \\\\&=\frac{e^{-(y_1+\max(0,-2y_1))}}{2} \\\\&=\begin{cases}\frac{1}{2}e^{-y_1}&,\text{ if }y_1\ge 0\\\\\frac{1}{2}e^{y_1}&,\text{ if }y_1<0\end{cases} \end{align}

  • $\begingroup$ Ahh I forgot to check the supports, now it makes sense. Thank you very much! $\endgroup$ – Thomas Bladt Nov 20 '18 at 21:39

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.