5
$\begingroup$

Let $H$ be an infinite-dimensional real Hilbert space and let $B$ be the closed unit ball of $H$. Let $K\subset B$ be a weakly compact set whose closed convex hull agrees with $B$. Question: does $K$ necessarily contain $0$ ?

$\endgroup$
5
  • 1
    $\begingroup$ You have an answer to your question. Why do you not accept it? $\endgroup$ Commented Oct 16 at 14:57
  • 1
    $\begingroup$ The proof of Israel is precise and elegant. $\endgroup$ Commented Oct 16 at 20:24
  • $\begingroup$ Indeed it is. So accept his answer! $\endgroup$ Commented Oct 16 at 23:47
  • $\begingroup$ I don't know how to do it ! $\endgroup$ Commented Oct 17 at 8:53
  • $\begingroup$ @BiagioRicceri : Click on the checkmark next to the answer. $\endgroup$ Commented Oct 17 at 13:48

2 Answers 2

5
$\begingroup$

If $K$ does not contain $0$, then there is a weak neighbourhood of $0$ disjoint from it, i.e. there is a finite set $\{y_1, \ldots, y_n \} \subset H$ such that $\max_{i=1}^n |\langle y_i, x \rangle| \ge 1$ for all $x \in K$. Since $H$ is infinite-dimensional, there is $z \in H$ such that $\|z\| = 1$ and $\langle y_i, z \rangle = 0$ for all $i$, thus $z \notin K$. Since $K$ is weakly compact, the functional $f: x \to \langle z, x \rangle$ attains a maximum on $K$, and that maximum can't be $1$ because the only member $b$ of $B$ with $\langle z, b \rangle = 1$ is $z$. So for some $\epsilon > 0$, $K \subset \{z: \langle z, x \rangle < 1-\epsilon \}$, and its closed convex hull can't include $z$.

$\endgroup$
1
  • 2
    $\begingroup$ Notice that since the condition implies that 𝐻 is reflexive, your proof works for any strictly convex space. OTOH, there is a renorming of a Hilbert space for which the result is false: Let $H= \ell_2 \oplus_\infty Y$, where $Y$ is the scalar field, and let $𝐾$ be all {x,a) s.t. \|x\| \le 1, |a|=1\}. $\endgroup$ Commented Oct 14 at 19:13
0
$\begingroup$

I don't want to think about the non-separable case, but if $H$ is separable so that $(B,\mathcal T_w)$ is metrizable, then the answer is yes. Suppose that $x_n\to x$ weakly and recall two basic facts: (1) $\|x\|\le\liminf \|x_n\|$; (2) if also $\|x_n\|\to\|x\|$, then $\|x_n-x\|\to 0$.

So the unit vectors $\|x\|=1$ are actually norm limits of vectors from the convex hull of $K$. But then such approximating convex combinations $\sum c_j y_j$ must contain vectors $y_j$ norm close to $x$, since $x$ is an extreme point (or make $x$ the first vector of an ONB and expand the $y_j$'s). Since this holds for every $x$, $\|x\|=1$, we obtain a sequence $y_n\in K$ which is almost orthogonal and hence $y_n\to 0$ weakly.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .