54
$\begingroup$

$X_1, X_2, \ldots, X_n$ are $n$ i.i.d. uniform random variables. Let $Y = \min(X_1, X_2,\ldots, X_n)$. Then, what's the expectation of $Y$(i.e., $E(Y)$)?

I have conducted some simulations by Matlab, and the results show that $E(Y)$ may equal to $\frac{1}{n+1}$. Can anyone give a rigorous proof or some hints? Thanks!

$\endgroup$
1
  • 1
    $\begingroup$ Have you found the density function for $Y$? $\endgroup$
    – Alex G.
    May 8, 2014 at 13:13

3 Answers 3

79
$\begingroup$

To calculate the expected value, we're going to need the density function for $Y$. To get that, we're going to need the distribution function for $Y$. Let's start there.

By definition, $F(y) = P(Y \leq y) = 1 - P(Y > y) = 1 - P(\text{min}(X_1, \ldots, X_n) > y)$. Of course, $\text{min}(X_1, \ldots X_n) > y$ exactly when $X_i > y$ for all $i$. Since these variables are i.i.d., we have $F(y) = 1 - P(X_1 > y)P(X_2>y)\ldots P(X_n>y) = 1 - P(X_1 > y)^n$. Assuming the $X_i$ are uniformly distributed on $(a, b)$, this yields $$F(y) = \left\{ \begin{array}{ll} 1 - \left(\frac{b-y}{b-a}\right)^n & : y \in (a, b)\\ 0 & : y < a\\ 1 & : y > b \end{array} \right.$$ We take the derivative to get the density function. $$f(y) = \left\{ \begin{array}{ll} \frac{n}{b-a} \left(\frac{b-y}{b-a}\right)^{n-1} & : y \in (a, b)\\ 0 & : \text{otherwise} \end{array} \right.$$ Now $E(Y) = \int_{-\infty}^{\infty} y f(y) dy$. The integral is straightforward; I'll leave the details to you. I calculate $E(Y) = \frac{b+na}{n+1}$.

$\endgroup$
2
  • $\begingroup$ Thanks for your answer. You extend it to a more general setting, which is very helpful. $\endgroup$
    – jet
    May 8, 2014 at 13:42
  • $\begingroup$ There's no sense in taking the derivative and then integrating; you can just use the tail-sum formula to calculate the expectation directly from the integral. $\endgroup$
    – Zaz
    Apr 25 at 23:36
21
$\begingroup$

Yes. Assuming a $U(0,1)$, note that

$$\Pr\Bigl(\min_i X_i\leq x\Bigr) = 1- \Pr\Bigl(\min_i X_i\geq x\Bigr) = 1- (1-x)^n. $$

So the density function is

$$f(x)=n(1-x)^{n-1}.$$

Then

$$\int_0^1 x f(x) dx = n\int_0^1 x(1-x)^{n-1} dx = n\int_0^1 (1-t) t^{n-1} dt = \frac{1}{n+1}. $$

$\endgroup$
4
  • 1
    $\begingroup$ Thanks for your answer, which is very helpful. $\endgroup$
    – jet
    May 8, 2014 at 13:43
  • $\begingroup$ Jpi what if $X\sim U(0,\theta)$ ? I am not able to solve the integral when I am trying to find expectation . @jet Can you answer my question ? $\endgroup$
    – Daman
    Oct 12, 2018 at 16:36
  • $\begingroup$ @Daman divide the $X_i$'s by $\theta$ first, which is equivalent to multiplying $x$ by $\theta$ in the first display. Hth $\endgroup$
    – JPi
    Sep 3, 2019 at 23:59
  • $\begingroup$ the change of variable from x to t would require (-dt) and switching the boundary of integral from 1 to 0 - but both cancel out. i.e. $$ n\int_0^1 x(1-x)^{n-1} dx = n\int_1^0 (1-t) t^{n-1} (-dt) = n\int_0^1 (1-t) t^{n-1} dt. $$ $\endgroup$ Feb 1, 2020 at 14:33
10
$\begingroup$

An short proof without the use of the pdf. Suppose $X_i \sim U(0,1)$. Then \begin{align*} E[\min(X_i)] &= \int_0^1 P[\min(X_i)>\mu] d\mu \\ &= \int_0^1 \prod_i P[X_i>\mu] d\mu \\ &= \int_0^1 (1-\mu)^n d\mu\\ &= \int_0^1 \mu^n d\mu\\ &= \dfrac{1}{n+1} \end{align*}

$\endgroup$

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.