6
$\begingroup$

I would like to ask a question which may look strange at the first sight nevertheless I find it interesting. Let $H$ be a separable Hilbert space: for any separable $C^*$-algebra $A$ one can embed $A$ into $B(H)$. Very often one can do even better and many nonseparable $C^*$-algebras still embed into $B(H)$, for example $\ell^{\infty}, L^{\infty}[0,1]$ or $B(H)$ itself. However it is not always possible even for quotients of such algebras, $\ell^{\infty}/c_0$ being an example. One can view $\ell^{\infty}$ as $C(\beta\mathbb{N})$ where $\beta\mathbb{N}$ is the Stone-Cech compactification of $\mathbb{N}$ (which is of cardinality $2^{\mathfrak{c}}$) and $\ell^{\infty}/c_0$ as $C(\beta\mathbb{N} \setminus \mathbb{N})$ (still of cardinality $2^{\mathfrak{c}}$). So here comes my question:

Let $X$ be a compact nonmetrizable Hausdorff topological space of cardinality $\mathfrak{c}$. Is it true that $C(X)$ embeds into $B(H)$ for separable $H$?

$\endgroup$
1
  • 3
    $\begingroup$ I don't know if useful: maximal abelian subalgebras of $B(H)$ seem to be known (see Tomiyama, JFA 1971): these are isomorphic to $C(Y)$ for some class of hopefully identifiable spaces $Y$, say $Y\in\mathcal{Y}$. Then the spaces $X$ are the continuous images of the spaces $Y$ when $Y$ ranges over$\mathcal{Y}$. $\endgroup$
    – YCor
    Apr 3, 2021 at 17:53

1 Answer 1

7
$\begingroup$

No, not necessarily. If $C(X)$ embeds into $B(H)$ then $X$ must be ccc, but there exist compact Hausdorff spaces of cardinality $\mathfrak{c}$ which are not ccc. (One example is to take $\mathfrak{c}$ with its discrete topology and form the one-point compactification $\mathfrak{c} \cup \{\infty\}$; then all the singletons except $\infty$ are disjoint open sets. Another is the ordinal $\mathfrak{c}+1$ with its order topology; all the successor singletons $\{\alpha+1\}$, $\alpha < \mathfrak{c}$, are disjoint open sets.)

To see this, suppose that $X$ is not ccc, so that there is an uncountable family $\{U_\alpha\}$ of disjoint nonempty open sets. By the appropriate version of Urysohn's lemma, for each $\alpha$ there is a nonzero real-valued $f_\alpha \in C(X)$ supported inside $U_\alpha$; in particular, $f_\alpha f_\beta = 0$ for $\alpha \ne \beta$. Now if there is an injective *-homomorphism $\Phi : C(X) \to B(H)$, then each $\Phi(f_\alpha)$ is nonzero, so we may find $h_\alpha \in H$ so that $\|\Phi(f_\alpha) h_\alpha\|=1$. Also, $\Phi(f_\alpha)$ is self-adjoint, so for $\alpha \ne \beta$ we have $$\langle \Phi(f_\alpha) h_\alpha, \Phi(f_\beta) h_\beta\rangle = \langle h_\alpha, \Phi(f_\alpha)\Phi(f_\beta) h_\beta\rangle = \langle h_\alpha, \Phi(f_\alpha f_\beta) h_\beta\rangle = 0$$ so that $\{\Phi(f_\alpha) h_\alpha\} \subset H$ is an uncountable orthonormal set, which is impossible if $H$ is separable.


This is fairly close to optimal because every separable compact Hausdorff $X$ does have $C(X)$ embeddable into $B(H)$ (recall that separable implies ccc but not conversely). Indeed, if $\{x_n\}$ is a countable dense subset of $X$, then identifying $f \in C(X)$ with the sequence $(f(x_1), f(x_2), \dots)$ gives an isometric embedding of $C(X)$ into $\ell^\infty$, which as we know embeds in $B(\ell^2)$. Presumably there is a necessary and sufficient condition somewhere in between, but I don't know what it is.

$\endgroup$
1
  • $\begingroup$ Nice, thank you! $\endgroup$
    – truebaran
    Apr 4, 2021 at 1:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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