Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I am trying to figure out the following problem:

I have two random variables: $X$ with pdf $f_X(x)$ on $[0,A]$ and $Y$ with pdf $g_Y(y)$ on $[0,B]$. Let's denote $Z=X+Y$. What should be the distribution of $X$, given $Z$?

Many thanks!

share|improve this question
3  
Which definition of conditional distribution are you using? –  Did Mar 20 '13 at 14:57
    
What do you mean? I was thinking of defining h(X|Z=X+Y) and expressing it somehow in terms of f(.) and g(.) –  VeryThankful Mar 20 '13 at 15:52
    
Equivalently: what have you been taught to check that a conditional density is indeed correct? (Unrelated: please use @user to comment to user.) –  Did Mar 20 '13 at 16:02
    
no, i need to use this conditional density in order to proof something else, but I don't know how to derive it –  VeryThankful Mar 20 '13 at 16:05
    
Use @ please. –  Did Mar 20 '13 at 17:58
show 4 more comments

2 Answers

up vote 2 down vote accepted

In full generality, the conditional distribution of a random variable $X$ conditionally on a random variable $Z$ is a family $(\nu_z)_z$ of distributions such that $\mathbb E(\varphi(X)\mid Z)=A_\varphi(Z)$ for every bounded measurable function $\varphi$, where $$ A_\varphi(z)=\int\varphi(x)\mathrm d\nu_z(x). $$ This means that, for every bounded measurable function $\psi$, $$ \mathbb E(\varphi(X)\psi(Z))=\mathbb E(A_\varphi(Z)\psi(Z)). $$ In the present case, assuming that $X$ and $Y$ are independent, one gets $$ \mathbb E(\varphi(X)\psi(Z))=\iint\varphi(x)\psi(x+y)f(x)g(y)\mathrm dx\mathrm dy, $$ and the goal is to identify $\varphi\mapsto A_\varphi$ such that this coincides with $$ \mathbb E(A_\varphi(Z)\psi(Z))=\iint A_\varphi(x+y)\psi(x+y)f(x)g(y)\mathrm dx\mathrm dy, $$ for every bounded measurable function $\psi$. Can you proceed from here?

share|improve this answer
    
Many thanks!I need to think about it. –  VeryThankful Mar 20 '13 at 18:07
    
I followed the explicit expression in the second answer and I have a short follow-up question to it. I am not sure how it works and who sees my comments so I post it here as well. I need to find conditions for which the ratio of the conditional probability to the prior is increasing in the random variable. In other words: $\frac{\partial}{\partial X}\left(\frac{g(Z-X)}{\int_{Max(0,Z-B)}^{Min(Z,A)} f(X)g(Z-X)dX}\right) \geq 0$. I am not sure how to proceed with this weird bounds of integration. Any help will be very highly appreciated. –  VeryThankful Mar 20 '13 at 21:03
    
I am not sure I like the way you ask a new question by appending a comment to each answer you received. As they say, if you have a new question, then post a new question. –  Did Mar 21 '13 at 21:20
    
Sorry, did not know how to do that, I am new here –  VeryThankful Apr 1 '13 at 19:04
    
Nonsense. You asked a question before, do it again. –  Did Apr 1 '13 at 19:13
show 1 more comment

Assuming $X$ and $Y$ are independent and continuous, find the joint probability density function for $X$ and $Z$, i.e. $f_{X,Z}(x,z)$.

The probability density function of the conditional random variable then reads: $$ f_{X|Z}\left(x\mid z\right) = \frac{f_{X,Z}(x,z)}{f_Z(z)} $$ where $f_Z(z)$ is the marginal pdf of $Z$, which is obtained as $f_Z(z) = \int_0^A f_{X,Z}(x,z) \mathrm{d}x$.

To find $f_{X,Z}(x,z)$ simply change measure in an expectation of a bounded function: $$ \mathbb{E}\left(h(X,Z)\right) = \int_0^A \int_0^B h(x,x+y) f_X(x) g_Y(y) \mathrm{d}x \mathrm{d}y = \int_0^A \int_0^{A+B} h(x,z) f_{X,Z}(x,z) \mathrm{d}x \mathrm{d}z $$ to find $$ f_{X,Z}(x,z) = f_X(x) g_Y(z-x) [ 0<x<A, x<z<B+x ] $$ and therefore: $$ f_Z(z) = \int_{\max(0,z-B)}^{\min(z,A)} f_X(x) g_Y(z-x) \mathrm{d}x $$

share|improve this answer
    
Thank you vey much!Isn't Min(Z,A)=A? –  VeryThankful Mar 20 '13 at 18:27
    
Appearance of $\min(z,A)$ simply says that when $z<A$, $f_Z(z)$ is zero. –  Sasha Mar 20 '13 at 18:30
    
oh, I see, thanks. And Max(0,z-B)? Z-B is in [-B,A] –  VeryThankful Mar 20 '13 at 18:36
    
I have a short follow-up question: I need to find conditions for which the ratio of the conditional probability to the prior is increasing in the random variable. In other words: $\frac{\partial}{\partial X}\left(\frac{g(Z-X)}{\int_{Max(0,Z-B)}^{Min(Z,A)} f(X)g(Z-X)dX}\right) \geq 0$. I am not sure how to proceed with this weird bounds of integration. Any help will be very highly appreciated. –  VeryThankful Mar 20 '13 at 20:57
add comment

Your Answer

 
discard

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.