Let $X_i \sim U(0,1)$ for $i = 1,\dots,n$, and let $X = \max(X_1,\dots,X_n)$. Is there a combinatorial way of seeing that $$\mathbb E(X) = \frac{n}{n+1}$$ and likewise for the minimum? Intuitively this is my guess for the expectation since, "on average," the experiment results will be distributed evenly on $[0,1]$. This may of course be verified by $$\int_0^1 x[P(X < x)]' \mathrm dx = \int_0^1 nx^n \mathrm dx = \frac{n}{n+1}.$$

  • $\begingroup$ Could you make more precise what a "combinatorial" proof is to you? $\endgroup$
    – Did
    Aug 30, 2016 at 22:27
  • 1
    $\begingroup$ A technique that would avoid integration by turning it into some counting problem. For instance, one can compute the probability that $X_1 < X_2$ via integration, but it's easier to note that there is a $1/2$ probability that any two numbers chosen from $[0,1]$ are in order. $\endgroup$
    – user369210
    Aug 30, 2016 at 22:31
  • $\begingroup$ I see. Thanks. $ $ $\endgroup$
    – Did
    Aug 30, 2016 at 22:34

1 Answer 1


This might not be what you are after but...

Let $(Y_1,Y_2, \cdots, Y_n)=(X_{(1)},X_{(2)}, \cdots, X_{(n)})$ be the ranked (ordered) variables. Let $(Z_0,Z_1, \cdots Z_n)$ be the differences : $Z_i=Y_{i+1}-Y_i$ (with $Y_0=0$ and $Y_{n+1}=1$); hence $Z_i\ge 0$ and $\sum\limits_{i=0}^n Z_i=1$.

It can be shown that the $Z_i$ have the same distribution as $n+1$ i.i.d. exponential variables $W_i$ conditioned on $V_n=1$, where $V_n=\sum\limits_{i=0}^n W_i=1$.

Once we accept this, then we get $$E(Z_i)=E(W_i\mid V_n=1)=\frac{1}{n+1}$$

So $$E(Y_1)=E(Z_1)=\frac{1}{n+1}$$ and, for every $1\le k\le n$, $$E(Y_k)=E(Z_0 + \cdots + Z_{k-1})= \frac{k}{n+1}$$ in particular, $$E(Y_{n})=E(Z_0 + \cdots + Z_{n-1})= \frac{n}{n+1}$$


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.