A stick of unit length is cut into three pieces of lengths $X, Y, $and $Z$, with $X \le Y \le Z$. First the stick of length $1$ is cut into two pieces at a randomly chosen point uniformly distributed along its length. Then the smaller of these two sticks is also cut into two pieces at a randomly chosen point uniformly distributed along its length. These two cuts are conducted independently. Find Cov(X,Y).

I need to get $E(XY)-EXEY$. First, I write $S=X+Y$. Then the length function of $S$ is $\min(s, 1-s)$. Given $S=s$, I can find the covariance of $X,Y$ as follows. The distribution of $X$ is $\min(x, s-x)$, so $E[X|S]=\int_0^{s/2} xdx + \int_{s/2}^s (s-x) dx=s^2/4.$ Likewise, $E[Y|S]=\int_0^{s/2} (s-y)dy + \int_{s/2}^s y dy=3s^2/4.$ Now $E[XY|S]=\int_0^s x(s-x)dx=s^2/6$,since $y$ is just $s-x$(I'm not sure about this part). Now to get the expectation, I just need to integrate them over $s$, which has the uniform distribution, so that would be integrating them over $[0,1/2]$. Is this correct? I would greatly appreciate any comment.

  • $\begingroup$ It was unclear to me why you said the distribution of $S$ is $\min(s,1-s)$. Later you say $s$ has a uniform distribution, which I believe is correct, but I think it is uniform on $[0,\frac12]$ since you are always cutting $X$ and $Y$ from the smaller stick created by the first cut. $\endgroup$ – David K Jan 19 '16 at 21:54
  • $\begingroup$ Sorry, I meant the length function of $S$. But why is it uniform on [0,1/2]? $\endgroup$ – takecare Jan 19 '16 at 21:56
  • $\begingroup$ When you cut a stick of length $1$ in two pieces, the longer piece is at least $1/2$ and the shorter piece is at most $1/2$. $\endgroup$ – David K Jan 19 '16 at 22:01
  • $\begingroup$ Yes you're right, just got it. Then if I integrate the above conditional expectations over $[0,1/2]$, would I get the correct answer? $\endgroup$ – takecare Jan 19 '16 at 22:02

The length $S$ of the shorter piece is uniform over $\left[0,\frac12\right]$. Given $S=s$, the length $X$ of the shorter piece in the second cut is uniform over $\left[0,\frac s2\right]$, and the length $Y$ of the longer piece in the second cut is uniform over $\left[\frac s2,s\right]$. Thus we have

\begin{align} \mathbb E[X]&=2\int_0^\frac12\mathrm ds\,\frac14s=\frac1{16}\;,\\ \mathbb E[Y]&=2\int_0^\frac12\mathrm ds\,\frac34s=\frac3{16}\;,\\ \mathbb E[XY]&=2\int_0^\frac12\mathrm ds\,\frac2s\int_0^\frac s2\mathrm dx\,x(s-x)\\ &=2\int_0^\frac12\mathrm ds\,\frac2s\left(\frac18s^3-\frac1{24}s^3\right)\\ &=2\int_0^\frac12\mathrm ds\,\frac{s^2}6\\ &=\frac1{72}\;, \end{align}

and the covariance is

\begin{align} \operatorname{Cov}(X,Y)&=\mathbb E[XY]-\mathbb E[X]\mathbb E[Y]\\ &=\frac1{72}-\frac1{16}\cdot\frac3{16}\\ &=\frac5{2304}\\ &\approx0.002\;. \end{align}

So the correlation from the first cut slightly outweighs the anticorrelation from the second cut.

  • $\begingroup$ can you just assert that the shorter stick is U(0,1/2)? I tried working out min(X,1-X) it doesn't give me U(0,1/2). Also should the length of Y be U(s/2,1/2)? $\endgroup$ – yoshi May 20 '16 at 3:59
  • 1
    $\begingroup$ @yoshi: Sorry, $Y$ is uniform on $\left[\frac s2,s\right]$, the upper bound $1$ was wrong (but yours, $\frac12$, is also wrong). I've fixed that; it didn't affect the calculation. About $S$: The situation is symmetric. If the cut is in $\left[\frac12,1\right]$, just reflect it to the other half; that doesn't change the lengths. Thus each half of the uniform distribution on $[0,1]$ is mapped to a uniform distribution on $\left[0,\frac12\right]$. (By the way, $X$, $Y$ are the lengths, not the segments, so it's "the length $Y$", not "the length of $Y$" as you wrote.) $\endgroup$ – joriki May 20 '16 at 4:08
  • $\begingroup$ Could you elaborate on how to obtain the $E[XY]$ integral? I tried expanding to $E[E[XY|S]]$ but I'm getting stuck on how to interpret/evaluate $E[XY|S]$. The $X$ and $Y$ expectations went through okay. $\endgroup$ – yoshi May 20 '16 at 14:43
  • 1
    $\begingroup$ @yoshi: In that integral, you're treating $x$ and $y$ as independent. They're not, in fact they're as dependent as they can be; knowing the value of one gives you the value of the other. They don't have a joint density function; you could describe their joint distribution by a delta "function" density, $\frac1s\delta(x+y-s)$ on $[0,s]^2$. Their joint cumulative distribution function is $F(X\le x,Y\le y)=\frac1s\min(x,y)$. $\endgroup$ – joriki May 20 '16 at 16:27
  • 1
    $\begingroup$ @yoshi: I edited the joint cumulative distribution function into the comment, don't know if you saw that, too. And I've been notified of your comments because they were under my answer; if you ever want to have me notified elsewhere, you'll need to spell my username right :-) $\endgroup$ – joriki May 20 '16 at 16:31

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.