9
$\begingroup$

The moment curve is the set of points of the form $$(t,t^2,t^3,...,t^n) \in R^n$$ Let $M$ be the portion of the moment curve where $t\in [0,1]$, and let $\overline{M}$ be the convex hull of $M$.

Caratheodory's theorem tells us that any point in $\overline{M}$ can be expressed as a convex combination of $n+1$ points in $M$. However, when $n=1$ and $n=2$, we can get by with only $n$ points instead of $n+1$ points.

Question: Is this true for all $n$? I.e., can we express any point in $\overline{M}$ as a convex combination of at most $n$ points from $M$?

Edit: In the answer below, eins6180 points out that $n$ points always suffice, answering my question. However, the reference he suggested actually mentions something much stronger. The moment curve is a convex curve, that is, a curve in which no $n + 1$ points lie in a single affine hyperplane. Barany & Karasev mention (through reference) that convex curves only require $$\left\lfloor \frac{n+2}{2} \right\rfloor$$ points to represent any point in their convex hull. The moment curve example is discussed a little bit in this paper by Holmsen and Roldan-Pensado.

I think a dimension-counting argument shows that this bound is tight.

$\endgroup$
11
$\begingroup$

The answer is yes for all dimensions.

An old theorem by Fenchel states that for a compact set $K$ in $\mathbb{R}^n$ every point in the convex hull can be either written as a convex combination of at most $n$ points or $K$ can be separated by a hyperplane. Since $M$ is path-connected, such a separation is not possible.

But there are also more modern theorems of this kind available. A good starting point is recent work by Barany & Karasev. Their work also implies an affirmative answer to your question.

$\endgroup$

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.