MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Hi everyone,

I will be too happy if anybody help me find a solution for the following problem. In fact, I have a big problem that I could not solve it for weeks.

Assume that we have we have two independent zero mean Gaussian random variables, $X$ and $Z$ and we define the new random variable $Y$ as $Y=X+Z$. I define the sign and the magnitude of this three random variables as $X_s$, $Y_s$ and $Z_s \in \{+1,-1\}$, $X_M$, $Y_M$ and $Z_M \in R^+$, respectively. It is clear that the pairs $X_s$ and $X_M$ are independent, this is the same for $Y_s$ and $Y_M$, $Y_s$ and $Y_M$.

My problem is how to find the conditional distribution $f(y_M|x_M)=?$.

I have a solution for this but I could not convince myself that my answer is true. We have, $$f(y_M|x_M)= f(y_M|x_M,x_s=+1)f(x_s=+1|x_M)+ f(y_M|x_M,x_s=-1)f(x_s=-1|x_M)$$ $$=f(y_m|x_M,x_s=+1)f(x_s=+1)+ f(y_M|x_M,x_s=-1)f(x_s=-1)$$ $$=0.5*(f(y_M|x_M,x_s=+1)+ f(y_M|x_M,x_s=-1))$$ then, since both $f(y_M|x_M,x_s=+1)$ and $f(y_M|x_M,x_s=-1)$ have the same distribution, i.e., "Folded normal distribution" then from above equation $f(y_M|x_M)$ follows "Folded normal distribution".

If it is correct it means that $f(y_M|x_M)= f(y_M|x_M,x_s)$ that implies $X_s$ and $Y_M$ condition on $X_M$ are independent!!!!!!!!!!!

But from $y_M = \lvert x+z \rvert$, $Y_M$ depends on both $X_M$ and $X_s$ !!!! I am really confused and I will be too grateful if anybody help me solve the problem. Thanks a lot in advance,

share|cite|improve this question
up vote 1 down vote accepted

Yes, $Y_M$ is independent of $X_s$ - it doesn't matter whether or not you condition on $X_M$. You essentially give the answer yourself - conditioning on $X_M$, you get that $Y | X_M$ is a symmetric r.v., with $f(Y|X_M) = \frac{1}{2\sigma_Z}[\varphi(\frac{Y-X_M}{\sigma_z})+\varphi(\frac{Y+X_M}{\sigma_z})]$ where $\varphi$ is the standard normal density. Therefore, summing on $Y=\pm Y_M$ you get $f(Y_M|X_M) = \frac{1}{\sigma_Z}[\varphi(\frac{Y_M-X_M}{\sigma_z})+\varphi(\frac{Y_M+X_M}{\sigma_z})]$ (for any $Y_M \geq 0$) which is equal to $f(Y_M|X_M,X_s)$. It is true that $Y=X+Z$ depends on $X_s$ but you can write $f(Y | X_s=1) = f(-Y | X_s=-1)$ and when you take the absolute value $Y_M = |Y|$ becomes independent of $X_s$.

share|cite|improve this answer
@ OrZuk, Thanks a lot for your answer. – Farzad May 30 '11 at 8:14
@ OrZuk, again thanks for your answer, but I did not tour 4th sentence, summing on $Y=\pm Y_M$ ..., I will be too happy if you explain me in more details? – Farzad May 30 '11 at 9:41
I meant that to get a particular value of $Y_M$ (which is non-negative), $Y$ can be either $Y_M$ or $-Y_M$. Therefore, to get the probability (actually density) of $Y_M$ attending a certain value, you should sum the density of $Y$ attaining this value or it's minus. – Or Zuk May 30 '11 at 13:03

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.