Let $X, Y$ be independent random variables with uniform distribution on the interval $[0, 1]$. My task is to find $$\mathbb{E} \big(\min(X, Y) | \max(X, Y) \big).$$

I think it can be done in the following way $$\mathbb{E} \big(\min(X, Y) | \max(X, Y) \big) = \mathbb{E}\big(\min(X, Y)|\sigma(\max(X,Y) \big) = \mathbb{E}\big(\min(X, Y)|\mathcal{F} \big).$$ Of course $\mathcal{F} = \{ \emptyset, [0,1] \}$ thus $\min(X, Y)$ is $\mathcal{F}$-measurable. That leads us to $$\mathbb{E}\big(\min(X, Y)|\mathcal{F} \big) = \min(X, Y).$$ Is my reasoning correct? That won't stand for other distributions will it?

  • $\begingroup$ How did you get $\mathcal F$? $\endgroup$ – Kavi Rama Murthy Dec 3 '18 at 8:35
  • $\begingroup$ @KaviRamaMurthy It is the $\sigma$-algebra made of $\max(X, Y)$. It's just the new symbol for $\sigma(\max(X,Y))$ $\endgroup$ – Hendrra Dec 3 '18 at 8:37
  • 2
    $\begingroup$ My question is why it is $\{\emptyset, [0,1]\}$. This is not true. $\endgroup$ – Kavi Rama Murthy Dec 3 '18 at 8:38
  • 1
    $\begingroup$ A set like $\{\max (X,Y) < \frac 1 2\} =\{X< \frac 1 2 \text {and} Y< \frac 1 2 \}$ belongs to $\mathcal F$. This set is neither empty nor $[0,1]$ because its probability is $\frac 1 4$. $\endgroup$ – Kavi Rama Murthy Dec 3 '18 at 8:44
  • 1
    $\begingroup$ In a less formal way you can state that $\min(X,Y)$ has uniform distritution on $[0,z]$ under the condition that $\max(X,Y)=z$. The corresponding expectation of that distribution is $\frac12z$, leading to the conclusion that $\mathbb E(\min(X,Y)\mid\max(X,Y))=\frac12\max(X,Y)$. $\endgroup$ – drhab Dec 3 '18 at 9:23

drhab has already commented how to informally derive the solution. I will formally prove that this is in fact the correct solution.

To this end, let $Z = \max(X,Y)$. We want to show $$ E[\min(X,Y)|\sigma(Z)] = \frac{Z}{2} \quad \text{a.s.} $$

The right hand side is obviously $\sigma(Z)$-measurable. So all we need to show is $$ E[\min(X,Y)\mathbb{1}_F] = E\left[\frac{Z}{2}\mathbb{1}_F\right] $$ for all $F \in \sigma(Z)$.

We will show this property for $F = \{Z \leq a \}$ with $a \in \mathbb{R}$. This is enough because these sets form a $\pi$-system (meaning it's $\cap$-stable) generating $\sigma(Z)$ and we can use a dynkin system argument to get the property for all $F \in \sigma(Z)$. Let me know if I should go into more detail here.

Because $X$ and $Y$ are iid $U([0,1])$, $Z$ has the Lebesgue density $f_Z(z) = 2z\mathbb{1}_{[0,1]}(z)$. So we get for the right hand side $$ E\left[\frac{Z}{2}\mathbb{1}_F\right] = \frac12 \int_{-\infty}^azdP^Z(z) = \int_{-\infty}^a z^2 \mathbb{1}_{[0,1]}(z)dz\\ = \begin{cases} 0 \quad &a < 0\\ \frac13 a^3 \quad &a \in [0,1]\\ \frac13 \quad &a > 1 \end{cases}. $$

For the left hand side, we get, using the symmetry of $X$ and $Y$ and $F = \{Z \leq a \} = \{X \leq a \} \cap \{Y \leq a \}$, $$ E[\min(X,Y)\mathbb{1}_F] = 2E[\min(X,Y)\mathbb{1}_F\mathbb{1}_{\{X \leq Y\}}] = 2E[X \mathbb{1}_{\{X \leq a\}}\mathbb{1}_{\{Y \leq a\}}\mathbb{1}_{\{X \leq Y\}}]\\ = 2\int_0^1\int_0^1 x \mathbb{1}_{\{x \leq a\}} \mathbb{1}_{\{y \leq a\}}\mathbb{1}_{\{x \leq y\}} dy dx\\ = 2 \int_0^1 x \mathbb{1}_{\{x \leq a\}} \int_x^1 \mathbb{1}_{\{y \leq a\}} dy dx. $$ Now you can distinguish the same $3$ cases as above and notice that you arrive at the same values.

  • 1
    $\begingroup$ (+1) The phrase "a $\cap$-stable generator of $\sigma(Z)$" could be replaced by the more standard "a $\pi$-system generating $\sigma(Z)$. $\endgroup$ – Did Dec 3 '18 at 10:51

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.