1
$\begingroup$

Let $F(u)$ be a n-degree polynomial continuous distribution function in $[0,1]$, with $F(0)=0$, $F(1)=1$, that is $F(u)=\sum_{i=1}^{i=n} a_i u^i$. My question is: is that kind of distributions uniquely represented as a finite mixture of the n order statitics of n i.i.d uniform variables? The answer is yes for $n=2$ since we can always write: $F(u)=\alpha (2u-u^2) + (1-\alpha) u^2 = 2\alpha u + (1-2\alpha) u^2$ with $\alpha \in [0,1]$ and this representation generates all the possible cuadratic distribution functions in $[0,1]$. Note that $(2u-u^2)$ and $u^2$ are the distributions of the mininum and the maximum of $(u_1,u_2)$ respectively, where $u_1, u_2$ are i.i.d. uniform random variables.

$\endgroup$

1 Answer 1

1
$\begingroup$

No. Take $n=3$. The CDF's of the order statistics are $F_1(u) = u^3 - 3 u^2 + 3 u$, $F_2(u) = 3 u^2 - 2 u^3$, $F_3(u) = u^3$. The function $F(u) = 4 (u - 1/2)^3 + 1/2$ is a CDF, and its unique representation in terms of the basis functions $F_1, F_2, F_3$ is $F = F_1 - F_2 + F_3$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.