13
$\begingroup$

Let $X$ be a reflexive Banach space. Then the convex hull of the extreme points of the unit ball is weakly dense by the Krein-Milman theorem and Kakutani's theorem. My question is, if there is an example of a reflexive Banach space $X$ whose unit ball does not equal the convex hull of its extreme points? Such an example $X$ must be infinite-dimensional and can't be strictly convex. My feeling is that this should be known, but I couldn't find anything about it. I've asked this question on StackExchange as well.

$\endgroup$

1 Answer 1

7
$\begingroup$

I'd try $X:=\ell_2$ with an equivalent but non strictly convex norm.

Let $(e_k)_{k\ge0}$ be the standard Hilbert basis of $\ell_2$. Consider the sets:

  • $A:=\{0\}\cup\{2^{-k}e_k:k\ge 1\}$,

  • $B$, the closed unit ball of $\ell_2$,

  • $C:=\overline{\text{co}}\, A$, a compact subset of the hyperplane $e_0^\perp$.

  • $D:=\overline{\text {co}}\big(B\cup( e_0+C)\cup(-e_0-C)\big)$.

Clearly, $C:=\big\{\sum_{k\ge1}2^{-k}a_ke_k: a_k \ge0 \text{ for all } k\ge1 \text{ and } \sum_{k\ge1}a_k\le1\big\},$ so $\text{ext}\, C=A$ and ${\text{co}}(\text{ext}\, C)={\text{co}}\, A\subsetneq C. $

The set $D$ is a bounded, closed, convex, symmetric nbd of $0$ in $\ell_2$, so it is the closed unit ball in an equivalent norm of $\ell_2$.

We have $D\cap (e_0+e_0^\perp)=e_0+C;$ the set of extremal points of $D$ in the affine hyperplane $e_0+e_0 ^\perp$ is exactly the set $e_0+A$, which implies that ${\text{co}}(\text{ext} D)\subsetneq D$ because the trace of these sets on $e_0+ e_0 ^\perp$ are ${\text{co}}(\text{ext} D)\cap (e_0+ e_0 ^\perp) =e_0+\text{co}A\subsetneq D \cap (e_0+ e_0 ^\perp) $.

$\endgroup$
0

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.