Let $X=\{x_1,...,x_k\}\subset E^n$ be a finite subset in the Euclidean $n$space, $r>0$ and $B(x_i,r)$ are open balls of radius $r$ centered at the points $x_i\in X, i=1,...,k$. Suppose that $$ \bigcap_{i=1}^k B(x_i,r)\ne \emptyset. $$ Is it true that the convex hull of $X$ is contained in $$ \bigcup_{i=1}^k B(x_i,r) ? $$ I needed a version of this in some Riemannian geometry argument and while I found a way around, the proof would have been a bit cleaner if this statement were true.

$\begingroup$ By Caratheodory's theorem it is enough to consider only the case when $X$ has $n+1$ points. $\endgroup$– Paata IvanishviliOct 8, 2016 at 2:40

2$\begingroup$ @PaataIvanisvili just a note: since dimension $n$ is arbitrary, this does not look very helpful, we may always assume that $X\leqslant n+1$ simply by embedding $E^n\subset E^{X1}$. This is why we could expect a dimensionfree proof, like that given below, a priori. $\endgroup$– Fedor PetrovOct 8, 2016 at 8:20

$\begingroup$ Thanks. It was only helpful to verify the statement in $\mathbb{R}^{2}$ and $\mathbb{R}^{3}$. But in general I agree with you. $\endgroup$– Paata IvanishviliOct 8, 2016 at 16:16
1 Answer
Yes. Any point $y$ in the convex hull of $x$'s is a barycenter of some nonnegative masses $m_i$ in $x_i$, $\sum m_i=1$, $y=\sum m_i x_i$. Point $y$ minimizes the moment of inertia $I(p)=\sum m_i px_i^2$, $p\in E^n$, just because $I(p)=I(y)+py^2$. In particular, it can not happen that all distances $yx_i$ are at least $r$, but for some other point $p$ all distances $px_i$ are less than $r$.