3
$\begingroup$

The question goes like this: $Z = X+Y$; where

$X$ is Log-normal Random variable with parameters - $\mu = 0 \quad \sigma^2= 1$,

$Y$ is Gaussian Random variable with $\mu= 0\quad \sigma^2= 1$

What is the pdf of $Z$? I know it will be the convolution of $X$ and $Y$. However, I am unable to solve it. Is it even solvable?

PS: $X$ and $Y$ are independent.

$\endgroup$
  • $\begingroup$ Are we assuming that $X$ and $Y$ are independent? $\endgroup$ – Math1000 May 28 '16 at 4:04
  • 1
    $\begingroup$ I'm afraid you will have difficulty finding an analytical solution given that the characteristic function $$\varphi_X(t) = \sum_{n=0}^\infty \frac{(it)^n}{n!}e^{\frac 12 n^2} $$ does not converge. $\endgroup$ – Math1000 May 28 '16 at 4:25
  • $\begingroup$ Yes, the random variables $X$ and $Y$ are independent. $\endgroup$ – skt9 May 28 '16 at 16:01
4
$\begingroup$

By saying convolution, you mean the two random variables $X$ and $Y$ are independent and the joint probability density function of them can be represented as the convolution of their pdfs.

Let $X$ be the log-normal random variable, and $Y$ the normal one, the pdf's of which are as below in the figure.

The probability density function of $Z=X+Y$ cannot be represented in closed form, but the numerical results of the pdf $f_Z(x)$ can be evaluated by numerical integral.

$$f_Z(x)=\int_{-\infty}^{+\infty}f_X(t)\cdot f_Y(x-t){\rm{d}}t=\int_{t=0}^{+\infty}\dfrac{e^{-\tfrac{1}{2}\left( (t-x)^2+{\ln^2t}\right)}}{2\pi t}{\rm{d}}t$$

So a better way to answer this question might be to visualize them as below:

enter image description here

Hope this is helpful.

$\endgroup$
  • $\begingroup$ Thank you. Is it possible to know the expression for $f_Z$$(x)$ by any means? Can I know the tool used for performing numerical integration and getting the graph above? $\endgroup$ – skt9 May 28 '16 at 16:03
  • $\begingroup$ hi @skt9, the analytical expression for $f_Z(x)$ has been given as above. There are a lot of special functions which have no closed forms (expression by elementary functions) but can be numerically obtained or visualized easily. -- A powerful tool in calculating the numerical integral and visualizing the profile is Wolfram Mathematica, which is also what I was using in getting the graph here. $\endgroup$ – LCFactorization May 29 '16 at 0:32

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.