0
$\begingroup$

Let $A$ be a commutative C*-algebra.

I would like to show that $A$ is separable (i.e. has a countable dense subset) if and only if the spectrum of $A$ (denoted by $\Omega(A)$) is separable.


Notes and reminders:

A C*-algebra is a Banach algebra over $\mathbb{C}$ with involution $*$ s.t. $\|x^*x\| = \|x\|^2$ for every $x \in A$.

Let me also remind that the spectrum $\Omega(A)$ is the set of the characters of $A$, that is $\Omega(A)\mathrel{:=} \{\chi:A\to \mathbb{C} : \text{$\chi$ is algebra homomorphism $\neq 0$}\}$. Also note that the topology on $\Omega(A)$ is the weak* (or w$^*$) topology. So the closure of any set which indicates separability is also taken under weak* topology.

$\endgroup$
13
  • 3
    $\begingroup$ I tend to rather believe that it's equivalent to its spectrum being both metrizable and separable? (i.e., in the unital case, metrizable)? So a separable non-metrizable Hausdorff compact space, e.g., $\beta\omega$, could be a counterexample. $\endgroup$
    – YCor
    Commented Mar 28, 2020 at 17:01
  • 2
    $\begingroup$ Don't enter and leave math mode for sets. Namely, not $\Omega(A) :=$ {$\chi : A \to \mathbb C$ : $\chi$ is algebra homomorphism $\neq 0$} $\Omega(A) :=$ {$\chi : A \to \mathbb C$ : $\chi$ is algebra homomorphism $\neq 0$} but $\Omega(A) \mathrel{:=} \{\chi : A \to \mathbb C \mathrel: \text{$\chi$ is algebra homomorphism $\ne0$}\}$ $\Omega(A) := \{\chi : A \to \mathbb C : \text{$\chi$ is algebra homomorphism $\neq 0$}\}$. I have edited accordingly. Also, what do you mean by "the closure of any separable basis on $\Omega(A)$"? Basis is coll'n of open sets, right? $\endgroup$
    – LSpice
    Commented Mar 28, 2020 at 17:05
  • 3
    $\begingroup$ If you take the C*-algebra of bounded functions $\omega\to\mathbf{C}$ (i.e., of bounded sequences), the spectrum is the Stone-Cech compactification $\beta\omega$, which is not metrizable. $\endgroup$
    – YCor
    Commented Mar 28, 2020 at 17:10
  • 3
    $\begingroup$ $\ell^\infty $ is not separable but its spectrum $\beta\mathbb N $ is separable. $\endgroup$ Commented Mar 28, 2020 at 17:29
  • 3
    $\begingroup$ In all this maze of comments, the very simple, elementary and well-known true necessary and sufficient condition for the separability of the algebra is missing. Here it is (for the unital case): $A$ is separable if and only if its spectrum is metrisable. The non unital case is easily deduced by using the standard ploys: adding a unit and taking the one point compactification. $\endgroup$
    – user131781
    Commented Mar 28, 2020 at 18:40

1 Answer 1

5
$\begingroup$

It's not true.

For simplicity, suppose $A$ is unital, so that its spectrum is compact Hausdorff.

If $X$ is a compact Hausdorff space and $C(X)$ is separable, then you can show that $X$ is second countable. (Let $\{f_k\}$ be a countable dense subset of $C(X)$, and $\{U_n\}$ a countable basis of open sets in $\mathbb{C}$; then the countable collection of sets $\{f_k^{-1}(U_n)\}$ is a basis for $X$, using Urysohn's lemma.) In particular, such an $X$ would be metrizable (by the Urysohn metrization theorem).

So the forward direction of your claim is true, since a second countable space is always separable, but this also suggests how to find a counterexample to the reverse direction: take a compact Hausdorff $X$ which is separable but not metrizable. One standard example is $X = [0,1]^{[0,1]}$ with its product topology; considering it as the space of all functions from $[0,1]$ to itself, the set of polynomials with rational coefficients is dense. Another, as YCor and Jochen Wengenroth discussed, is $X = \beta \mathbb{N}$, the Stone-Cech compactification of $\mathbb{N}$.

The converse of the above statement is also true, as user131781 points out: if $X$ is a compact metric space, then $C(X)$ is separable.

So the correct version of your claim (including the non-unital case) would be that $A$ is separable iff $\Omega(A)$ is second countable iff $\Omega(A)$ is a separable metrizable space.

$\endgroup$
1
  • $\begingroup$ Thank you very much! $\endgroup$ Commented Mar 29, 2020 at 6:41

You must log in to answer this question.

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