Q) Let X and Y have the joint p.d.f.

\begin{equation} f(x,y)=\begin{cases} e^{-y}, \text{if } 0<x<y<\infty \\ 0, \text{otherwise } \end{cases} \end{equation}

Obtain the marginal p.d.f's for X and Y.

A)I know that $F_{x}(s)=\lim_{t \to \infty} F_{x,y}(s,t)$ So have found f(x) to be

\begin{equation} f(x)=\begin{cases} \lim_{y \to \infty} e^{-y} = 0, \text{if } 0<x<y<\infty \\ \lim_{y \to \infty} 0 = 0, \text{otherwise } \end{cases} \end{equation}

and f(y) to be

\begin{equation} f(y)=\begin{cases} \lim_{x \to \infty} e^{-y} = 0, \text{if } 0<x<y<\infty \\ \lim_{x \to \infty} 0 = 0, \text{otherwise } \end{cases} \end{equation}

(since if x is going to infinity then y is going to infinity.)

However, the next question is asking about calculating the co-variance matrix of X+Y and X-Y which makes me think that my first answer was wrong. I don't know how to fix it so any hints would be really helpful, thanks!

  • $\begingroup$ What you are stating at here is saying that given a joint CDF, how to obtain the marginal CDF. However, you are now given a joint pdf to find the marginal pdf, which require you to do integration. $\endgroup$ – BGM Apr 8 '17 at 12:58

The following figure depicts the joint density which is zero outside the red region.

enter image description here

So the marginals are

$$f_X(x)=\int_{-\infty}^{\infty}f_{X,Y}(x,y)\ dy=\int_x^{\infty}e^{-y}\ dy=e^{-x},$$ if $x\ge 0$


$$f_Y(y)=\int_{-\infty}^{\infty}f_{X,Y}(x,y)\ dx=\int_0^{y}e^{-y}\ dx=ye^{-y},$$ if $y\ge 0$.


$$E[XY]=\int_0^{\infty}\int_0^{\infty}xyf_{X,Y}(x,y)\ dxdy=$$ $$=\int_0^{\infty}x\int_x^{\infty}ye^{-y}\ dy\ dx.$$

For the internal integral, the antiderivative is $-e^{-y}(1+y)$. This antiderivative is gained by integrating by parts and using $v'=e^{-y}$ and $u=y$.

And $\left[-e^{-y}(1+y)\right]_x^{\infty}=e^{-x}(1+x).$ Then

$$E[XY]=\int_0^{\infty}xe^{-x}(1+x)\ dx=\int_0^{\infty}xe^{-x}\ dx+\int_0^{\infty}x^2e^{-x}\ dx$$

and you can use the antiderivative formula given above. But without that, you know that, the first integral is $1$ bacause we compute the mean of the exponential distribution of $\lambda=1$, also, the other integral is $2$ because we compute the second momentum of the same. So,


  • $\begingroup$ Thanks! I have used f(x,y)=f(x)f(y) and found X and Y to not be independent, so the co-variance of X and Y is non-zero. I have cov(x,y)=E(xy)-E(x)E(y). I know how to calculate E(x) and E(y) using integrals by don't know how to calculate E(xy)? $\endgroup$ – Koala Apr 8 '17 at 14:17
  • $\begingroup$ @Koala: I'll edit my answer. $\endgroup$ – zoli Apr 8 '17 at 14:27
  • $\begingroup$ for the integral of x in the second line, why are the boundaries 0 and infinity and not 0 and y? $\endgroup$ – Koala Apr 8 '17 at 14:47
  • $\begingroup$ 1 Sorry, I had to edit my answer because I omitted accidentally an $x$. But the answer is no. Below is the reverse order of integration: 2 No, and $$\int_0^{\infty}x\int_x^{\infty}ye^{-y}\ dy\ dx=\int_0^{\infty}ye^{-y}\int_0^yx\ dx\ \ dy =\frac12\int_0^{\infty}y^3e^{-y}\ dy=3.$$ (See the third moment of the exponential distribution.) $\endgroup$ – zoli Apr 8 '17 at 15:50

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.