I have the following problem.

Let $(X,Y)$ be a random vector with joint density function $f(x,y)=8xy$ for $0<x<y<1$.

Find $f_X$ and $f_Y$.

My attempt was:

$f_X(x)=\int_{-\infty}^{\infty}f(x,y)dy=\int_x^1 8xydy=4x(1-x^2) $

the second equality from left to right because $x<y<1$.

But my question is, for what $x's$ is valid to say $f_X(x)=4x(1-x^2)$? because x can't be any number, it depends on "$y$" but $f_X(x)$ should depend only for x terms, right?

Thanks in advance.

  • 2
    $\begingroup$ Check your integration: $f_X(x)=4x(1-x^2)$ on $(0,1)$, $f_Y(y)=4y^3$ on $(0,1)$. $x$ and $y$ can be anything in these densities since you integrated the other variable out - the restriction $x<y$ has already been taken care of in integration bounds. $\endgroup$ – A.S. Nov 11 '15 at 21:35
  • $\begingroup$ I have already fixed. $\endgroup$ – HeMan Nov 11 '15 at 21:37

It doesn't depend on $y$. If you look at the restrictions on your joint density, you see that $x$ takes values between $0$ and $1$. And as a sanity check, you can check that $$\int_0^1 f_X(x)dx = \int_0^1 4x(1 - x^2) dx = 1.$$

Edit: As mentioned in the comments, there was a mistake in your original computation of the integral.

Second edit: As for a formal reason why it does not depend on $y$, it's because of your integration: you've marginalized over $x$ and $y$ does not appear anymore.

For an informal reason, think of it this way: your joint density is supported on $0 < x < y < 1$, and so when looking from above, you see that the density has weight only on a triangle (and not on the full square $[0,1]^2$). When you marginalize, you are not looking at the density from above anymore, but from the sides: you cannot see the triangle anymore, you only see that there is some positive weight all along the segment $[0,1]$.

  • $\begingroup$ I see that sanity check works. But what is the formal reason that makes that $x$ doesn't depend on the $y$ restriction? $\endgroup$ – HeMan Nov 11 '15 at 21:40
  • $\begingroup$ @Levent Does this help? $\endgroup$ – M Turgeon Nov 11 '15 at 21:45
  • $\begingroup$ Yes it does! Thanks! $\endgroup$ – HeMan Nov 11 '15 at 21:46

Clearly the support for the marginal of $X$ is: $(0; 1)$

Because when you "remove" the $y$ you have $0<x<1$

There's no dependence on $y$ because you've 'integrated it out'.

  • $\begingroup$ But why can I remove the condition $0<x<y<1$? That is the part that I don't understand. $\endgroup$ – HeMan Nov 11 '15 at 21:41
  • 2
    $\begingroup$ It is handled by the integration; $f_X(x)$ is the joint density "summed" over all supported values of $Y$ for that particular atom of $X$. Since for each atom of $X$ we've already accounted for its possible $Y$ values, they are not a restriction on the support of the margin. @Levent $\endgroup$ – Graham Kemp Nov 11 '15 at 22:15

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.