1
$\begingroup$

Let $H$ be a locally compact Hausdorff space and $C_c(H)$ the space of continuous functions with compact support. Equip $C_c(H)$ with the locally convex inductive limit topology $\tau$ of the Banach spaces $C_K(H)$ of continuous functions with support contained in a compact $K \subseteq H$, where $K$ runs over all compact subsets of $H$. Then $(C_c(H), \tau)' = M(H)$ is the space of real-valued Radon measures defined on the collection of relatively compact Borel sets. [$C_c(H)$ equipped with the (coarser) supremum norm $\lVert \cdot \rVert$ gives rise to the subspace $(C_c(H), \lVert \cdot \rVert)' = M_b(H) \subseteq M(H)$ of real-valued Radon measures defined on the whole Borel $\sigma$-algebra; the measures in $M_b(H)$ have bounded variation (in contrast to $M(H)$).]

Consider the dual pair $\langle C_c(H), M(H) \rangle$. For the strong topology it holds $\beta(C_c(H), M(H)) = \tau\,$ (because $C_c(H)$ is ultrabornological as an inductive limit of Banach spaces, hence barrelled). The linear hull $D := \textrm{span}(\delta_x \mid x \in H)$ of Dirac measures is (weakly-)dense in $M(H)$ and we can consider the dual pairs $\langle C_c(H), D \rangle$ and $\langle C_c(H), M_b(H) \rangle$. I want to know whether

$$\beta(C_c(H), D) = \beta(C_c(H), M(H)) \quad \textrm{or at least} \quad \beta(C_c(H), M_b(H)) = \beta(C_c(H), M(H)).$$

I have the following steps:

  1. For a dual pair $\langle Y, X \rangle$ of vector spaces $X, Y$ and a (weakly-)dense subspace $S \subseteq X$ consider the dual pair $\langle Y, S \rangle$. For the strong topologies on $Y$ it then holds $\beta(Y, S) \subseteq \beta(Y, X)$. For the equality $\beta(Y,S) = \beta(Y, X)$ it is sufficient and necessary that every bounded set $B \subseteq X$ is contained in the (weak) closure of a bounded set $A \subseteq S$.

  2. For simplicity let us first restrict to the case of a compact Hausdorff space $K = H$. Then $\tau = \lVert \cdot \rVert$ and $C(K)' = M(K)$. Equip $M(K)$ with the weak$^*$-topology. In order to apply 1., let $B \subseteq M(H)$ be a bounded set and show that there exists $A \subseteq D$ with $B \subseteq \overline{A}$. It holds $B \subseteq r \overline{B}_1(0)$ where $r > 0$ and $\overline{B}_1(0) \subseteq M(H)$ is the closed unit ball. It is enough to show that $\overline{B}_1(0) \subseteq \overline{A}$ for some bounded $A \subseteq D$. I know that the closed unit ball $\overline{B}_1(0) \subseteq M(K)$ is a weakly$^*$-compact convex set (Banach-Alaoglu). Also, the mapping $K \to M(K)$, $x \mapsto \delta_x$ is an embedding, so that we can identify $K$ as a compact subspace of $M(K)$. By Krein-Milman, $\overline{B}_1(0)$ is the closed convex hull of its extreme points, but the set of all the extreme points is the unit sphere $\{ \mu \in M(H) \mid \lVert \mu \rVert = 1 \}$ and this set is not contained in $D$. So, is it possible to find a suitable bounded set $A \subseteq D$? A canonical candidate for $A$ would be the absolutely convex hull $\textrm{aco}(K)$. If the strong topologies are not equal, what is then the dual of $(C_c(H), \beta(C_c(H), D))$ and $(C_c(H), \beta(C_c(H), M_b(H)))$?

  3. If we can prove the conjecture for a compact space $K$, then I think it should also work for a more general locally compact space $H$, by relating $C_c(H)$ to its Banach space components $C_K(H)$ and considering polars of open neighborhoods in place of balls.

Note that for the special case $H = \mathbb{N}$ the conjecture is true: $C_c(H) = \varphi$ is the space of sequences that are eventually $0$, $M(H) = C_c(H)' = \omega$ is the space of all sequences, $D = \textrm{span}(\mathbb{N}) = \varphi$ and for this example, one can show that $\beta(\varphi, \varphi) = \beta(\varphi, \omega)$, because $\beta(\varphi, \varphi)$ is barrelled.

$\endgroup$
1
  • $\begingroup$ Correction: it seems that the extreme points of $M(K)$ is indeed precisely the absolutely convex hull of Dirac measures. So the proof for the compact Hausdorff case should be fine. $\endgroup$
    – yada
    Nov 12 '19 at 17:16
1
$\begingroup$

I believe that your conjecture is true if $H$ is locally compact and $\sigma$-compact. Then the topology of the inductive limit $C_c(K)=\lim\limits_\to C(K,H)$ (where $C(K,H)$ denotes the Banach space of continuous functions with support in $K$) can be described by quite simple semi-norms of the form $$\|f\|_v=\sup\lbrace|f(x)|v(x): x\in H\rbrace$$ where $v$ is any strictly positive function on $H$ which is bounded on compact sets. It is clear that the restriction of $\|\cdot\|_v$ to each $C(H,K)$ is continuous there. It remains to show that these semi-norms indeed give the inductive limit topology. Fix a sequence $K_n$ of compact sets such that $K_n$ is contained in the interior of $K_{n+1}$ and $H=\bigcup_n K_n$. Then $C_c(H)=\lim\limits_\to C(K_n,H)$. Given any continuous semi-norm $p$ on $C_c(H)$ there are constants $c_n>c_{n-1}>0$ such that $p(f)\le c_n \sup\lbrace|f(x)|:x\in H\rbrace$ for all $f\in C(K_n,X)$. Next choose a continuous locally finite partition of unity $\varphi_n$ subordinated to the covering $\mathring K_n$ and set $v=2^nc_n$ on $K_n\setminus K_{n-1}$. Given any $f\in C_c(H)$ we can write $f=\sum_{n\in \mathbb N} \varphi_nf$ (where the sum is in fact finite) to obtain $$p(f)=\sum_{n\in \mathbb N} p(\varphi_nf) \le \sum_{n\in \mathbb N} 2^{-n} \|f\varphi_n\|_v\le \|f\|_v.$$

Now the proof should be as the one you suggested for the compact case: Given a $0$-neighbourhood $U$ of $C_c(H)$ it contains a unit ball with respect to some $\|\cdot\|_v$ and the extreme points of $U^\circ$ should be $\pm v(x)\delta_x$.

$\endgroup$
1
  • $\begingroup$ Thank you. In the case of a $\sigma$-compact locally compact space, a reference to the literature for your result is [Summers, "Weighted Locally Convex Spaces of Continuous Functions", PhD thesis, (1968)] Theorem 3.36 (p. 60) together with Theorem 2.28 (p. 31). It remains open, what $\beta(C_c(H), D)$ and its dual (i.e. the "$C_c$-bidual" of $D$) actually is for a general locally compact space $H$. $\endgroup$
    – yada
    Nov 17 '19 at 10:43

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.