Fix a compact convex subset $C \subset \mathbb{R}^n$ with nonempty interior. For any subspace $S \subset \mathbb{R}^n$, let $P_S$ denote the orthogonal linear projection onto $S$. I'd like to claim that for almost every (in either a measure theory or topological sense) nontrivial subspace $S$ of a given dimension, the image of $C$ under $P_S$ has the following property: for every $y$ in the relative boundary of $P_S(C)$, the fiber $\{x \in C : P_S(x) = y\}$ is a singleton.

In the case when $\dim S = 1$, my claim is equivalent to the statement that almost every linear functional attains its maximum at a single point in $C$. I'm particularly interested in the case when $\dim S = n-1$. Has anyone ever seen anything like this? Note: I believe it is easy to verify the claim when $C$ is strictly convex or polyhedral, so the interesting situation is when $C$ is a general compact convex set.

Thinking about this some more, it is not even obvious that the set of subspaces $S$ of a given dimension with this property (points on the relative boundary of $P_S(C)$ have unique preimages in $C$) is nonempty. Even a basic existence argument for general convex sets would be useful.


1 Answer 1


I asked this question several years ago, and I recently found the answer in a paper by Ewald, Larman, and Rogers called "The directions of the line segments and of the r-dimensional balls on the boundary of a convex body in Euclidean space". This was published in 1970 in Mathematika, and the main result in the paper is:

Theorem 1. If $K$ is a convex body in $\mathbb{R}^n$, the set $S$, of end-points of the vectors drawn from the origin in the directions of the line segments lying on the surface of $K$, is a set of $\sigma$-finite $(n-2)$-dimensional Hausdorff measure on the $(n-1)$-dimensional surface of the unit ball.


Your Answer

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

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