Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X$ be a real Banach space and $X^*$ its dual space. Let $C^*$ be a weak$^*$ closed and convex subset in $X^*$ and $x^*\notin C^*$. Then there exists $x\in X$ such that $$ \langle x^*, x\rangle > \sup_{f\in C^*}\langle f, x\rangle. $$ I would like to ask whether the statement is true? How can we prove?

share|cite|improve this question
You don't need the $\forall f \in C^*$: $f$ is already a dummy variable because of the $\sup$. – Robert Israel Oct 16 '12 at 6:23
@Robert Israel: Thank you for your helping. – blindman Oct 16 '12 at 6:59
See also: – commenter Oct 16 '12 at 12:59
up vote 3 down vote accepted

This is just the separation theorem (see e.g. Rudin, "Functional Analysis", Theorem 3.4(b)) for the locally convex topological vector space $X^*$ with the weak-* topology (whose continuous linear functionals correspond to the points of $X$): slightly more generally you can replace $x^*$ by a weak-* compact convex set disjoint from $C^*$.

share|cite|improve this answer
Nice answer! What I can tell from all this is that the weak*-topology forces all continuous linear functionals on $ X^* $ to come from $ X $. My 'inferior' answer below failed to narrow down to this consideration. However, is it true that if we replace 'weak*-closed' by 'closed', then we require $ X $ to be reflexive for the statement to hold? In other words, can we find a counterexample in which the required separating linear functional must come from $ X^{**} \setminus X $? – Haskell Curry Oct 16 '12 at 6:38
@HaskellCurry: Let $C$ be the space of sequences in $\ell_1 = (c_0)^\ast$ that sum to zero. This is a norm-closed and weakly closed subspace of $\ell_1$, but it's weak*-dense (it is the kernel of the functional given by a constant sequence in $\ell_\infty \setminus c_0$). If you take any sequence $x$ whose sum isn't zero, you can't separate it by a weak$^\ast$-continuous functional from $C$, so reflexivity is essential for your variant. – commenter Oct 16 '12 at 12:58
More generally, if $X$ is a nonreflexive Banach space, let $\phi \in X^{\star\star} \backslash X$, and consider $C^* = \{x^* \in X^*: \phi(x^*) = 0\}$. Then $C^*$ is closed and convex. For any $x \in X$ there is $x^* \in C^*$ such that $x^*(x) \ne 0$ (because otherwise the linear functional $x^* \mapsto x^*(x)$ would be a scalar multiple of $\phi$, which would imply $\phi \in X^{\star\star}$), so $\sup_{x^* \in C^*} x^*(x) = \infty$. – Robert Israel Oct 16 '12 at 18:26
@RobertIsrael: commenter's observation that $ C^* $ is weak*-dense applies here. Observe that $ C^* $ is a codimension-1 linear subspace of $ X^* $. We now use the basic result that the kernel of a non-zero linear functional on a topological vector space is a closed codimension-1 linear subspace iff the linear functional is continuous. As $ (X^*,\sigma(X^*,X))^* = X $, we see that $ \phi $ is not weak*-continuous. Therefore, $ C^* $ cannot be weak*-closed, which forces it to be weak*-dense (as the codimension of $ C^* $ is $ 1 $, the weak*-closure of $ C^* $ must be all of $ X^* $). – Haskell Curry Oct 17 '12 at 2:09

I don't have the full answer as of now, but the best that I can think of is that if $ X $ is a reflexive Banach space, then the statement is true by the geometric Hahn-Banach Theorem. You can treat $ X^{*} $ as the initial Banach space under consideration and $ C^{*} $ as the closed and convex (and non-empty!) subset of $ X^{*} $. Then if $ x^{*} \notin C^{*} $, one can find an $ x \in X \cong X^{**} $ (by the geometric Hahn-Banach Theorem) such that \begin{equation} \langle x^{*},x \rangle > \sup_{f \in C^{*}} \langle f,x \rangle. \end{equation} Now, there is an error associated with the phrasing of the problem. As you are taking the supremum of the right-hand side of the inequality over $ f \in C^{*} $, there is no need to universally quantify $ f $. On a separate note, if $ C^{*} $ is weak*-closed, then it is automatically closed, so you do not have to add the adjective 'closed' in front of 'weak*-closed'.

If $ X $ is not reflexive, then I think the statement is false. That is because the separating functional that you need to separate $ x^{*} $ and $ C^{*} $ might lie in $ X^{**} \setminus X $. However, I could be wrong, so any opposing opinions are welcome.


The above discussion is concerned with the case when $ C^{*} $ is closed with respect to the norm topology on $ X^{*} $. After the latest edit of the problem statement, it is now understood that $ C^{*} $ is to be assumed closed with respect to the weak*-topology on $ X^{*} $. More generally, observe that if we assume $ C^{*} $ to be closed with respect to any locally convex topology $ \mathcal{T} $ that is finer than the weak*-topology (denoted by $ \sigma(X^{*},X) $) but coarser than the Mackey topology (denoted by $ \tau(X^{*},X) $), then the continuous dual of $ (X^{*},\mathcal{T}) $ is still isomorphic to the natural copy of $ X $ in $ X^{**} $. Hence, the separation theorem mentioned by Robert (I call it the geometric Hahn-Banach Theorem) still applies to yield a separating functional $ x \in X $.

share|cite|improve this answer
Dear Sir. Thank you for your construction and comments. – blindman Oct 16 '12 at 7:00

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.