I'm working through a practice problem for an exam and I would like to verify that I've done it correctly. Additionally I'd like some insight on the intuition behind the numbers I'm getting.


For $X$ and $Y$ independent $Exp(1)$ r.v.'s compute $P(X<3|X+Y)$.


Since the sum of exponentials is a scaled gamma, $S=X+Y$~$Ga(1,2)$, where the rate is 1 and the shape is 2.

When $X+Y\leq 3$, the probability is 1 that $\{X<3\}$. I checked this also by solving:

$$P(X<3|X+Y\leq 3)=\frac{P(X<3,Y<3-X)}{P(S\leq 3)}=\frac{\int_0^3\int_0^{3-x} e^{-x}e^{-y}dy dx}{\int_0^3 se^{-s}ds}=1$$

Now when $X+Y>3$, I compute a similar quantity, namely, $$P(X<3|X+Y> 3)=\frac{P(X<3,Y>3-X)}{P(S> 3)}=\frac{3}{4}$$


Assuming that the above quantities are correct, could I interpret the quantity $P(X<3|X+Y>3)$ as follows:

"Given that the sum is greater than 3, the chances that the one of the variables is less than 3 is distributed uniformly (due to memoryless property of exponential r.v.'s). That is, if the sum is >3, and if we fix e.g. Y=3, then on average the sum is $Y + E(X)=4$, and we get that X<3, 3/4ths of the time."

The above interpretation is loosely based on my finding that for $k\in\mathbb{N}$ $$P(X<k|X+Y>k)=\frac{k}{k+1}=\frac{k}{k+E(X)}$$

Thanks for any help.


Denote $S=X+Y$. Then

$$f_S(s)=se^{-s} \text{; } f_{X,S}(x,s)=1\{s\ge x\}e^{-s}$$


$$P\{X<c|S=s\}=\int_0^{c}\frac{f_{X,S}(x,s)}{f_{S}(s)}dx=\int_0^{c}\frac{1\{s\ge x\}e^{-s}}{se^{-s}}dx$$ $$=\frac{1}{s}\int_0^c 1\{s\ge x\}dx=\frac{c}{s}\wedge 1$$

  • $\begingroup$ could you explain how you got $f_{X|S}$? $\endgroup$ – user251257 Jul 4 '15 at 2:37
  • $\begingroup$ It was a typo... $\endgroup$ – d.k.o. Jul 4 '15 at 2:40
  • $\begingroup$ @user251257 $f_{X,S}(x,s)=f_{S|X}(s|x)f_X(x)$ $\endgroup$ – d.k.o. Jul 4 '15 at 2:43
  • $\begingroup$ @d.k.o. Thank you! It actually makes sense that the probability depend on the sum. I'm unsure how you got $f_{S|X}$. Could you explain? $\endgroup$ – probtheory Jul 4 '15 at 3:53
  • 1
    $\begingroup$ @probtheory $F_{S|X}(s|x)=P\{Y\le s-x\}=1\{s\ge x\}(1-e^{x-s})$ $\endgroup$ – d.k.o. Jul 4 '15 at 4:01

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.