1
$\begingroup$

My question is not about method or computation, I have some doubt about how to deal with indicator function, so if someone could clarify the matter I'll be gratefully!

I have this joint distribution $f_{X,Z}(x,z)=(z-x)e^{-z}1_{\{0\leq x\leq z\}}$

I want to find the marginal distributions and to show if they are or not independent.

I did this:

$$f_X(x)=\int_{-\infty}^{\infty}(z-x)e^{-z}1_{\{0\leq x\leq z\}}dz$$ $$=\int_{x}^{\infty}(z-x)e^{-z}dz$$ with some computation $$=e^{-x}\tag{1}$$

My question is: is it correct? I mean conceptually. Or I have to consider even the condition in the indicator function? In the latter case I would have $$f_X(x)=e^{-x}1_{\{0\leq x\}}\tag{2}$$

I like that since its integral over $\mathbb{R}$ is equal to one, but is it right to do the indicator function appearing suddenly, at the end of computation? The other marginal is $$f_Z(z)=\int_{0}^{x}(z-x)e^{-z}dx=\frac{z^2}{2e^z}\tag{3}$$

or maybe $$f_Z(z)=\frac{z^2}{2e^z}1_{\{0\leq z\}}\tag{4}$$

Also now I like the second one since its integral is equal to one!

For the other point I want just to show that the product of marginals is different from the joint distribution:

$$\frac{z^2}{2e^{z+x}}\neq (z-x)e^{-z} \tag{5}$$

or maybe $$\frac{z^2}{2e^{z+x}}1_{\{0\leq x,0\leq z\}}\neq (z-x)e^{-z}1_{\{0\leq x\leq z\}} \tag{6}$$

or also, (if it means something, I'll change $z$ with $z-x$ to get the condition of the indicator functions the same) $$\frac{(z-x)^2}{2e^{z}}1_{\{0\leq x\leq z\}}\neq (z-x)e^{-z}1_{\{0\leq x\leq z\}} \tag{7}$$

What is the right version, if there is one? :D

$\endgroup$
  • 1
    $\begingroup$ They are clearly not independent since knowing $Z=z$ puts an upper bound on $X$ which would not otherwise be there (similarly, knowing $X=x$ puts a lower bound on $Z$) $\endgroup$ – Henry Dec 2 '17 at 23:50
  • $\begingroup$ @Henry Indeed. Which is what the indicator functions ...well... indicate. $\endgroup$ – Graham Kemp Dec 3 '17 at 1:30
1
$\begingroup$

(2), (4), (6) are what you want.

$$\begin{align} f_X(x)&=\int_{-\infty}^{\infty}(z-x)\mathsf e^{-z}~\mathbf 1_{\{0\leq x\leq z\}}~\mathsf dz \\ &= \mathbf 1_{\{0\leq x\}}\cdot\int_{x}^{\infty}(z-x)\mathsf e^{-z}~\mathsf dz \\ & = \mathsf e^{-x}~\mathbf 1_{\{0\leq x\}}\tag 2\\[2ex] f_Z(z)&=\int_{-\infty}^{\infty}(z-x)\mathsf e^{-z}~\mathbf 1_{\{0\leq x\leq z\}}~\mathsf dx \\ &= \mathbf 1_{\{0\leq z\}}\cdot\int_{0}^{z}(z-x)\mathsf e^{-z}~\mathsf dz \\ & = \tfrac 12 z^2\mathsf e^{-z}~\mathbf 1_{\{0\leq z\}}\tag 4\\[2ex] \mathsf e^{-z}~\mathbf 1_{\{0\leq x\leq z\}}&\neq\left(\mathsf e^{-x}~\mathbf 1_{\{0\leq x\}}\right)\left(\tfrac 12 z^2\mathsf e^{-z}~\mathbf 1_{\{0\leq z\}}\right)\tag 6\\[2ex]\hline\therefore\quad f_{X,Z}(x,z)&\neq f_X(x)\cdot f_Z(z) \end{align}$$

$\endgroup$

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.