Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X_1,X_2,X_3$ be independent with a common exponential distribution. I am trying to find the density of $(X_2-X_1 ,X_3-X_1 )$.

Assuming $X_1,X_2,X_3\sim Exponential(λ)$

If $x > 0$ $$f_{X_2-X_1}(x) =f_{X_3-X_1}(x) = {\dfrac {\lambda } {2}e^{-\lambda x}} $$ If $x < 0$ $$f_{X_2-X_1}(x) =f_{X_3-X_1}(x) = {\dfrac {\lambda } {2}e^{\lambda x}} $$

For simplicity let $X_2-X_1 =X$ and $X_3-X_1= Y$. The Joint density can be calculated as $$f_{X,Y}( x,y) = f_{Y\mid X}( y\mid x) f( x) = f_{X\mid Y}( x\mid y) f(y)$$

I am unsure here how to Calculate the conditional pdf. Any help would be much appreciated.

share|cite|improve this question
up vote 3 down vote accepted

You start with the joint PDF, as you have done $$ f : \mathbb{R}_{\geq 0}^3 \rightarrow \mathbb{R} : \mathbf{x} \rightarrow \alpha^3 e^{-\alpha(x_1+x_2+x_3)} $$

Then you pick an appropriate (invertible!) transform $T$, also as you have done, and find its inverse $$ \begin{eqnarray} T(\mathbf{x}) &=& \begin{pmatrix} -1 & 1 & 0 \\ -1 & 0 & 1 \\ 0 & 1 & 0 \end{pmatrix}\;\mathbf{x} &\text{,}& T^{-1}(\mathbf{x}) &=& \begin{pmatrix} -1 & 0 & 1 \\ 0 & 0 & 1 \\ -1 & 1 & 1 \end{pmatrix}\;\mathbf{x} \text{.} \end{eqnarray} $$

The transformed PDF $g$ (i.e. $g$ such that $g(T(\mathbf{x})) = f(\mathbf{x})$) is thus $$ g(\mathbf{y}) = f(T^{-1}(\mathbf{y}))\frac{1}{\det(T)} = \alpha^3 e^{-\alpha(3y_3 + y_2 - 2y_1)} $$

Now we have to figure out $\text{dom}\, g \subset \mathbb{R}^3$ such that $T^{-1}(\text{dom}\, g) = \text{dom}\, f$. Requiring $T^{-1}(\mathbf{y}) \in \mathbb{R}_{\geq 0}^3$ for all $y \in \text{dom}\, g$ produces the conditions $$ \begin{eqnarray} y_3 - y_1 &\geq& 0 \text{,}\\ y_3 &\geq& 0 \text{,}\\ y_3 + y_2 - y_1 &\geq& 0 \text{.} \end{eqnarray} $$ This yields $y_3 \geq \max\{0,y_1, y_1 - y_2\}$ as necessary and sufficient, i.e. you get $$ \text{dom}\, g = \left\{\mathbf{y} \,:\, \mathbf{y} \in \mathbb{R}^3, y_3 \geq \max\{0,y_1, y_1 - y_2\}\right\} \text{.} $$

To find the PDF $h$ of $(X_2-X_1,X_3-X_1)=(Y_1,Y_2)$ you must now integrate $g$ over $y_3$, but without leaving the domain of $g$, i.e. compute $$ \begin{eqnarray} h(y_1,y_2) &=& \int_{c}^\infty \alpha^3 e^{-\alpha(3y_3 + y_2 - 2y_1)} dy_3 \:\text{where}\: c = \max\{0,y_1, y_1 - y_2\} \\ &=& \frac{\alpha^2}{3} e^{-\alpha(y_2 - 2y_1)} \int_{c}^\infty 3\alpha e^{-3\alpha y_3} dy_3 \\ &=& \frac{\alpha^2}{3} e^{-\alpha(y_2 - 2y_1)} e^{-3\alpha c} \\ &=& \frac{\alpha^2}{3} e^{-\alpha(3c - 2y_1 + y_2)} \end{eqnarray} $$

Note that if $y_1,y_2\geq 0$ then $c=y_1$, if $y_1 \leq 0,y_1 \leq y_2$ then $c=0$ and if $y_2 \leq 0,y_2 \leq y_1$ then $c=y_1-y_2$. This gives $$ h(y_1,y_2) = \frac{\alpha^2}{3}\begin{cases} e^{-\alpha(y_1 + y_2)} & \text{if } y_1,y_2 \geq 0 \\ e^{-\alpha(2|y_1| + y_2)} & \text{if } y_1 \leq 0, y_1 \leq y_2 \\ e^{-\alpha(y_1 + 2|y_2|)} & \text{if } y_2 \leq 0, y_2 \leq y_1 \end{cases} $$

share|cite|improve this answer
Firstly sorry for a late reply, sleeping was way overdue.This is an excellent answer and i completely agree that i had made a big boo-boo with determining the new domain.I do have a suggestion though i think in your final answer one does not need those absolute values brackets around $y_1$ and $y_2$.These should just be negative as per Feller's provided answer.I suppose some how he is not considering to adapt the density $h(y_1,y_2)$ so it's distribution function would always converge.I am going to take down my post instead of improving mine to match yours.Thank you for this. Welcome to MSE:-) – Comic Book Guy Oct 9 '12 at 19:19
@Hardy I actually just wrote $-2y_1$ respectively $-2y_2$ initially, and later changed it to the absolute value. Mostly because that how the laplace distribution is usually written, and I wanted to make the connection more obvious. Though I can see your point too, guess the absolute value could make things look slightly more complex than they are. Anyway, glad you like the answer, I certainly had fun solving this! – fgp Oct 9 '12 at 19:35

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.