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

1) Suppose I have two random variables $A > 0$ and $B > 0$ with joint p.d.f. $f_{A,B}(a,b)$, and two random variables $X = g_1(A,B)$ and $Y = g_2(A,B)$. What is the general procedure for determining the joint p.d.f. $f_{X,Y}(x,y)$ ?

2) More specifically, suppose that:

  • $A$ and $B$ are i.i.d. with $A \sim \Gamma (k, m)$.
  • $X = AB$
  • $Y = A + B$

Clearly it this case $X$ takes a K distribution and $Y$ takes a Gamma distribution. Due to struggling with 1) above, I've unfortunately been unable to determine whether in this case a closed non-integral expression for the joint p.d.f. $f_{X,Y}(x,y|k,m)$ even exists, let alone what it is...

Any assistance / suggestions would be greatly appreciated.

share|cite|improve this question
There is a standard technique for these kinds of transformations that is described in various levels of detail in many (3rd and 4th year) college-level probability theory texts. Which one, if any, do you have easy access to? The method involves Jacobians since it is arises from a change of variables in a multidimensional integral. – Dilip Sarwate Apr 27 '12 at 19:19
I thought there would be a standard technique, but since I'm only an engineer by training, I didn't get that far in my mathematics studies at university! Can you recommend any particular textbook that has an accessible treatment of this? – Peter Brett Apr 27 '12 at 19:30
I've added the reference-request tag, because hey, good reference materials are always useful. – Peter Brett Apr 27 '12 at 19:38
up vote 2 down vote accepted

Let me first state the result: $$ f_{X,Y|k,m}(x,y) = \frac{2}{\sqrt{y^2-4 x}} \frac{x^{k-1} e^{-\frac{y}{m}}}{(k-1)!^2 m^{2 k} } I(x>0,y>0,y^2 > 4 x) $$

Many posts are devoted to the subject of determining the probability density function for the function of random variables, i.e. this, that, and other posts referenced therein.

The mapping $(A,B) \mapsto (A B, A+B)$ is unto. Specifically, for $$ x= a b, y = a+b \implies a = \frac{1}{2} \left( y \pm \sqrt{y^2-4x}\right), b = \frac{1}{2} \left( y \mp \sqrt{y^2-4x}\right) $$ The Jacobians are easy to compute: $$ J_{\pm} = \left| \begin{array}{cc} \frac{\partial a}{\partial x} & \frac{\partial a}{\partial y} \cr \frac{\partial b}{\partial x} & \frac{\partial b}{\partial y} \end{array} \right| = \left| \begin{array}{cc} \mp \frac{1}{\sqrt{y^2-4x}} & \pm \frac{a}{\sqrt{y^2-4x}} \cr \pm \frac{1}{\sqrt{y^2-4x}} & \mp\frac{b}{\sqrt{y^2-4x}}\end{array} \right| = \frac{b-a}{y^2- 4 x} = \mp \frac{1}{\sqrt{y^2-4x}} $$ We can now read off $f_{X,Y}(x,y)$ from the product measure for $(A,B)$: $$ \begin{eqnarray} f_A(a) f_B(b) \mathrm{d} a \mathrm{d} b &=& \left( \frac{1}{\Gamma(k)} \frac{a^{k-1}}{m^k} \mathrm{e}^{-\frac{a}{m}} \mathbf{1}_{a>0} \right) \cdot \left( \frac{1}{\Gamma(k)} \frac{b^{k-1}}{m^k} \mathrm{e}^{-\frac{b}{m}} \mathbf{1}_{b>0} \right) \mathrm{d} a \mathrm{d} b \\ &=& \frac{(a b)^{k-1}}{\Gamma^2(k) m^{2k}} \mathrm{e}^{-\frac{a+b}{m}} \mathbf{1}_{a>0} \mathbf{1}_{b>0} \mathrm{d} a \mathrm{d} b \\ &=& \frac{x^{k-1}}{\Gamma^2(k) m^{2k}} \mathrm{e}^{-\frac{y}{m}} I(x>0,y>0,y^2>4x) \left(| J_+ | + | J_- | \right) \mathrm{d} x \mathrm{d} y \end{eqnarray} $$

share|cite|improve this answer
Just to be clear: $\boldsymbol{1}_{condition}$ and $I(condition)$ are equivalent to Iverson brackets here? – Peter Brett Apr 28 '12 at 7:40
Also, once I have worked through to make sure I understand how this works I will accept it. :-) – Peter Brett Apr 28 '12 at 7:40
Yes both mean Iverson bracket. – Sasha Apr 28 '12 at 13:12

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.