There is a line segment of the length of $1$. $N-1$ points are randomly chosen in it, so it is divided by $N$ parts. The question is to calculate expected values of all these parts, from the shortest one to the longest one. I know that the correct answer for the length of the shortest part is about $1/N^2$, but I have no idea how to find out the whole distribution of lengths.


Since the distribution is uniform this can be considered the search for the centroid of a figure in Euclidean space.

A priori the figure to consider for $n-1$ independent uniform variables would be the $(n-1)$-dimensional hypercube, but we are interested in the differences (lengths of the segments) so we study the $(n-1)$-simplex defined in $n$-dimensional space by the constraints that all $n$ coordinates be nonnegative, and that $\sum_{i=1}^nx_i=1.$ Thus the coordinates do not represent the original variables, but their differences (with appropriate interpretations at the boundaries).

We should divide this into $n!$ subsets according to the order in which the lengths of the line segments appear; however, by symmetry the same effect can be achieved by limiting our attention to the one subset where we already have

$$x_1\geq x_2\geq\ldots\geq x_n.$$

That subset is also an $(n-1)$-simplex and its $n$ vertices are

$$\matrix{ (1,0,0,\ldots,0,0,0)\\ (\frac12,\frac12,0,\ldots,0,0,0)\\ (\frac13,\frac13,\frac13,\ldots,0,0,0)\\ \vdots\\ (\frac1{n-1},\frac1{n-1},\frac1{n-1},\ldots,\frac1{n-1},\frac1{n-1},0)\\ (\frac1n,\frac1n,\frac1n,\ldots,\frac1n,\frac1n,\frac1n) }$$

The centroid is just the average of those $n$ vertices, so by taking the sum of each column in the above list and dividing by the number of vertices we get that the $x_k$-coordinate of the centroid is


This is the expected value of the length of the $k$-th line segment in decreasing order.

| cite | improve this answer | |

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.