3
$\begingroup$

I am trying to find the CDF of $Z=aX+bY$ whereby $X$ and $Y$ are random variables and $a$ and $b$ are positive integers. Given that the CDF of $Z$ is:

$$F_Z(z)=\int F_X\left(\frac{z-by}{a}\right)f_Y(y)dy$$

Given that $X$ is uniform distribution over $[0,1]$ with: $$f_X \left(x \right)=\frac{1}{\sigma }\left [ H\left ( x \right ) -H\left ( x-\sigma \right )\right ]$$ $$F_X \left(x \right)=\frac{x}{\sigma }\left [ H\left ( x \right ) -H\left ( x-\sigma \right )\right ]$$

And $Y$ is Weibull distribution over $[0, \infty)$ with:

$$f_Y \left(x \right)=\frac{\beta }{\sigma }\left (\frac{x}{\sigma } \right )^{\beta -1} e^{-\left ( \frac{x}{\sigma } \right )^{\beta }}$$ $$F_Y \left(x \right)=1- e^{-\left ( \frac{x}{\sigma } \right )^{\beta }}$$

where $\beta, \sigma >0$ in all the expressions above.

I tried deriving $F_Z(z)$ following some videos online but no luck when I check against numerical CDF generated using Monte Carlo (random numbers). Can help me determine the limits and explain just once on how to solve the convolution?

$\endgroup$
2
  • $\begingroup$ $H(x)$ are Heaviside step functions right? $\endgroup$
    – tired
    Mar 17 '15 at 16:30
  • $\begingroup$ @tired: yes they are. let me try your answer and answer your question. thanks :) $\endgroup$ Mar 17 '15 at 16:31
1
$\begingroup$

The expression $H(x)-H(x-\sigma)$ implies an range of integration from $0$ to $\sigma$ (Draw a picture for some specific values if you need to be convinced about that).

Setting $x=\frac{z-by}{a}$ in $F_X(x)$ it's a straightforward matter to see that we have to solve the integral: $$ F_Z(z)=\frac{\beta}{\sigma^2}\int_{\frac{z}{b}}^{\frac{z-a\sigma}{b}}\frac{z-by}{a}\left(\frac{y}{\sigma}\right)^{\beta-1}e^{-\left(\frac{y}{\sigma}\right)^{\beta}}dy $$

Now i would suggest to change variables $q=\left(\frac{y}{\sigma}\right)^{\beta}$. You should now be able to express the integral in terms of incomplete Gamma-Functions.

Therefore, using $dy=\frac{1}{\beta}q^{\frac{1}{\beta}-1}\sigma$ we have

$$ \frac{z}{a}\int_{\left(\frac{z\sigma}{b}\right)^{\beta}}^{\left(\frac{z-a\sigma}{b}\sigma\right)^{\beta}}e^{-q}-\frac{b}{a}\int_{\left(\frac{z\sigma}{b}\right)^{\beta}}^{\left(\frac{z-a\sigma}{b}\sigma\right)^{\beta}}q^{1/\beta}e^{-q}=\\ -\frac{z}{a}\left(e^{\left(\frac{z-a\sigma}{b}\right)^{\beta}}-e^{\left(\frac{z\sigma}{b}\right)^{\beta}}\right)-\frac{b}{a}\left[\gamma\left(\frac{1}{\beta}+1,\left(\frac{z-a\sigma}{b}\right)^{\beta}\right)-\gamma\left(\frac{1}{\beta}+1,\left(\frac{z\sigma}{b}\right)^{\beta}\right)\right] $$

Here $\gamma(a,x)$ denotes a lower incomplete Gamma-Function

Edit: Please note that this integral only exists for a specific range of parameters. I leave it to you to figure out this part rigorously!

$\endgroup$
5
  • $\begingroup$ still finding for solution. can u help me with this please? i have more problems and hopefully i can solve them $\endgroup$ Mar 17 '15 at 16:46
  • $\begingroup$ Do you understand everything what i did so far? $\endgroup$
    – tired
    Mar 17 '15 at 16:50
  • $\begingroup$ yes i do. managed to derive on my own till this point $\endgroup$ Mar 17 '15 at 16:52
  • $\begingroup$ Following my proposition with the substitution there is really no further trick...Just some standard calculation. Everything i could do is just post the result, not give you any better advice... $\endgroup$
    – tired
    Mar 17 '15 at 16:54
  • $\begingroup$ okay. if u have worked it out, u can post it. else, i'll try to work it out :) $\endgroup$ Mar 17 '15 at 16:55

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.