Let $X$ be any locally compact Hausdorff space and assume that it is not compact. I've heard that the Banach space $(C_0(X),\|\!\cdot\!\|_\infty)$ is not isometrically isomorphic to the (norm) dual of a Banach space. Is there a good book where I can find a proof this result?

  • 3
    $\begingroup$ Hint: Show that the unit ball has no extreme points. Use Alaoglu and Krein-Milman to derive a contradiction. $\endgroup$ – commenter Oct 9 '12 at 21:05
  • 2
    $\begingroup$ @commenter whu don't you arite this comment as an answer? $\endgroup$ – Norbert Oct 10 '12 at 9:48
  • $\begingroup$ @Norbert: Because I couldn't think of a reference containing that argument. I posted an expanded version as an answer. $\endgroup$ – commenter Oct 10 '12 at 10:50

Suppose first that $C_0(X) = E^\ast$ for some normed space $E$. By Alaoglu's theorem the closed unit ball $B$ of $C_0(X)$ is compact in the weak*-topology and by the Krein-Milman theorem $B$ has an extreme point (in fact, $B$ is the weak*-closed convex hull of its extreme points).

Thus, in order to prove that $C_0(X)$ is not a dual space, it suffices to show that $B$ has no extreme points:

Let $f \in B$ with $\lVert f\rVert =1$. Since $X$ is not compact and $f$ vanishes at infinity, the set $U = \{x \in X : |f(x)| \lt 1/2\}$ is non-empty open and $C=\{x \in X : \lvert f(x) \rvert \geq 1/2\}$ is non-empty and compact.

Pick $u \in U$ and use Urysohn's lemma to find a function $h\colon X \to [0,1/2]$ such that $h(u) = 1/2$ and $\operatorname{supp}h \subset U$. Then $\left\lVert f \pm h\right\rVert_\infty = 1$ and $f \neq f\pm h$ together with $$ f = \frac{1}{2}\left(f+h\right) + \frac{1}{2}\left(f-h\right) $$ show that $f$ is not an extreme point in $B$.

Remark. It is essential that we assume that $X$ is not compact. Constant functions of norm $1$ are always extremal in the unit ball for compact $X$, so the above argument breaks down. There's a good reason for that: For finite $X$, $C(X) \cong \mathbb{R}^n$ is reflexive, or for $X = \beta\mathbb{N}$ we can show that $C(\beta\mathbb{N}) \cong \ell_\infty = (\ell_1)^\ast$, so these are dual spaces.

  • $\begingroup$ But of course $C[0,1]$ is again not a dual space, so noncompactness is not "if and only if". Real $C[0,1]$ has too few extreme me points. But complex $C[0,1]$ has lots of extreme points so more thought is required for that case. $\endgroup$ – GEdgar Oct 10 '12 at 13:28

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.