16
$\begingroup$

Is convex hull of closed set in $\mathbb R^{n}$ closed?

$\endgroup$
20
$\begingroup$

Topologically, the convex hull of an open set is always itself open, and the convex hull of a compact set is always itself compact; however, there exist closed sets that do not have closed convex hulls. For instance, the closed set $$ \left\{(x,y):y\geq\frac{1}{1+x^2}\right\}\subset\mathbb R^2 $$ has the open upper half-plane as its convex hull.

Source: Wikipedia.

$\endgroup$
8
$\begingroup$

Here is a simple counterexample: Let $S=(\{0\} \times [0,1] ) \cup ([0,\infty)\times \{0\})$. $S$ is closed since it is the union of two closed sets, however, $\operatorname{co} S = ([0,\infty) \times [0,1)) \cup \{ (0,1)\}$, which is clearly not closed.

If the set $S \subset \mathbb{R}^n$ is compact, however, the convex hull is closed. This follows from Carathéodory's theorem which states that any point in $\operatorname{co} S $ can be written in terms of at most $n+1$ elements of $S$ (this is the convex analogue of the fact that any $n+1$ elements of $\mathbb{R}^n$ must be linearly dependent). Hence we can write $\operatorname{co} S = f(S^{n+1} \times \Sigma)$, where $f:\mathbb{R}^n \times \cdots \times \mathbb{R}^n \times \Sigma \to \mathbb{R}^n $ is defined by $f((x_1,...,x_{n+1}, \mu)) = \sum_{k=1}^{n+1} \mu_k x_k$, and $\Sigma$ is the simplex $\Sigma = \{\mu| \mu_k \ge 0, \sum_k \mu_k = 1 \}$. Since $f$ is continuous, $S^{n+1}$ and $\Sigma$ are compact, it follows that $f(S^{n+1} \times \Sigma)$ is compact, hence closed.

$\endgroup$
  • $\begingroup$ In my opinion, $\text{co}S=([0, \infty)\times[0, 1)\cup(\{0\}\times [0,1])$ $\endgroup$ – blindman Sep 3 '13 at 1:40
  • $\begingroup$ @blindman: That is the same set as I described above. Only the point $(0,1)$ needs to be added. The set $\{0\} \times [0,1)$ is contained in the first set of both of our descriptions. $\endgroup$ – copper.hat Sep 3 '13 at 2:41
  • $\begingroup$ @cooper.hat: Thanks. I did not see clearly. $\endgroup$ – blindman Sep 3 '13 at 2:58

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.