Let $\mathcal{A}$ be a non-unital commutative Banach algebra. Consider the Gelfand transform \begin{align*}\Gamma_{\mathcal{A}}:\mathcal{A}&\to C(\sigma(\mathcal{A}))\\ x&\mapsto \hat{x} \end{align*} where \begin{align*}\hat{x}:\sigma(\mathcal{A})&\to \mathbb{C}\\ h&\mapsto \hat{x}(h)=h(x) \end{align*} In Folland's A course in Abstract Harmonic Analysis it is stated that in fact $\Gamma_{\mathcal{A}}(\mathcal{A})\subset C_0(\sigma(\mathcal{A}))$, i.e. if $x\in \mathcal{A}$ then for all $\varepsilon>0$ there is a compact $K\subset \sigma(\mathcal{A})$ such that $|\hat{x}(h)|\leq \varepsilon$ for all $h\in \sigma(\mathcal{A})\setminus K$. It is not well explained why this is true.

The basic idea would be to try to get back to the unital case. We can consider the unital extension $\tilde{A}=\mathcal{A}\times \mathbb{C}$ of $\mathcal{A}$, where $\mathcal{A}\cong \mathcal{A}\times \left\{0\right\}\subset \tilde{A}$. We then have a homeomorphism (I think) \begin{align*}\sigma(\tilde{\mathcal{A}})&\cong \sigma(\mathcal{A})\cup \left\{0\right\}\\ 0&\mapsto \tilde{0}:(x,\lambda)\mapsto 0\\ \sigma(\mathcal{A})\ni h&\mapsto \tilde{h}:(x,\lambda)\mapsto h(x)+\lambda \end{align*} both are compact Hausdorff spaces in the weak*-topology induced by $\mathcal{A}^*$ and $\mathcal{A}$ respectively, while $\sigma(\mathcal{A})$ is only locally compact in general.

The above homeomorphism in turn induces a Banach algebra isomorphism $C(\sigma(\tilde{\mathcal{A}}))\cong C(\sigma(\mathcal{A})\cup\left\{0\right\})$. With this identification we have $\Gamma_{\mathcal{A}}=r\circ \left.\Gamma_{\tilde{\mathcal{A}}}\right|_{\mathcal{A}}$ where $r:C(\sigma(\mathcal{A})\cup\left\{0\right\})\to C(\sigma(\mathcal{A}))$ is the restriction map. Now I would like to use the topological/algebraic properties I know about $r$ and $\Gamma_{\tilde{\mathcal{A}}}$ to show that $\Gamma_{\mathcal{A}}(\mathcal{A})\subset C_0(\sigma(\mathcal{A}))$.

I am aware that $\mathcal{A}$ is a closed maximal ideal of $\tilde{A}$ so since $\tilde{\mathcal{A}}$ is unital, $\Gamma_{\tilde{\mathcal{A}}}(\mathcal{A})$ is itself a closed non-trivial subalgebra of $C(\sigma(\mathcal{A})\cup\left\{0\right\})$.

  • $\begingroup$ In regards to your second paragraph, $C_0(\sigma(\mathcal{A}))$ isn't the algebra of compactly supported functions, it's the algebra of functions vanishing at infinity. $\endgroup$ – Aweygan May 1 '18 at 15:50
  • $\begingroup$ True, thanks for pointing it out. I actually had the correct space in mind, just wrote down the wrong definition, so my issues remain. $\endgroup$ – Lorenzo Quarisa May 1 '18 at 16:00

The following may or may not answer the specific questions you have in mind, but it does I think explain what's really going on.

Say $A$ is a Banach algebra. Usually in this context "complex homomorphism" means "non-zero complex homomorphism", but here I'm going to include the map $h(x)=0$ as a "complex homomorphism".

Let $K$ be the set of all complex homomorphisms of $A$, and let $\sigma(A)=K\setminus\{0\}$ be the set of non-zero complex homomorphisms.

Let $A^*$ be the Banach-space dual of $A$. Then $K$ is a subset of the unit ball of $A^*$, and it's easy to see that $K$ is weak* closed; hence $K$ is weak* compact. The Gelfand toplogy is just the weak* topology, so $K$ is compact.

Hence $\sigma(A)=K\setminus\{0\}$ is locally compact. (Of course $K$ is the one-point compactification of $\sigma(A)$, which means that the "point at infinity" for $\sigma(A)$ is given by the amusing formula $$\infty=0.)$$

Now if $A$ has an identity $e$ then $\sigma(A)=\{h\in K:h(e)=1\}$; hence $\sigma(A)$ is a closed subset of $K$, hence $\sigma(A)$ is compact. But in general, whether $A$ has an identity or not, we have, writing $\hat x$ for the Gelfand transform,

If $x\in A$ then $\hat x\in C_0(\sigma(A))$.

Proof: Let $\epsilon>0$. We need to show that $\{h\in\sigma(A):|\hat x(h)|\ge\epsilon\}$ is compact. But this set is equal to$\{h\in K:|h(x)|\ge\epsilon\}$, which is a closed subset of $K$.

Note: I'm pretty sure that if $\sigma(A)$ is compact then $A$ has an identity. That's less trivial than the rest of this - right now I don't recall the proof.

  • $\begingroup$ Thanks for your answer. Just for reference on your note, Folland notes that if $0$ is an isolated point of $\sigma(\mathcal{A})\cup\left\{0\right\}$ then $\sigma(\mathcal{A})$ is still compact (when talking of non-unital algebras), although he does claim that in 'the most common case' $\sigma(\mathcal{A})$ is non-compact. $\endgroup$ – Lorenzo Quarisa May 1 '18 at 17:13
  • $\begingroup$ @LorenzoQ. He doesn't actually give an example of an algebra with no identity where $\sigma(A)$ is compact, does he? $\endgroup$ – David C. Ullrich May 1 '18 at 17:25
  • $\begingroup$ No, he does not! $\endgroup$ – Lorenzo Quarisa May 1 '18 at 17:34

Ok, I solved it.

Let $x\in \mathcal{A}$. With the usual identification $\mathcal{A}\cong \mathcal{A}\times \left\{0\right\}\subset \tilde{\mathcal{A}}$, we have that $\hat{x}=\left.(x,0)^{\hat{}}\right|_{\sigma(\mathcal{A})}$ and $\sigma(\tilde{\mathcal{A}})\cong \sigma(\mathcal{A})\cup\left\{0\right\}$. For any $\varepsilon>0$, $$\left\{h\in \sigma(\mathcal{A})\cup\left\{0\right\}:|(x,0)^{\hat{}}(\tilde{h})|\geq \varepsilon\right\}=\left((x,0)^{\hat{}}\right)^{-1}(\mathbb{C}\setminus B_0(\varepsilon))=:S_{\varepsilon} $$ Now, the key insight is to notice that $0\notin S_{\varepsilon}$ for all $\varepsilon >0$, because $(x,0)^{\hat{}}(\tilde{0})=0(x,0)=0$. Since moreover $\hat{x}$ is continuous, $S_{\varepsilon}$ is a closed subset of $\sigma(\mathcal{A})$, and hence compact because it is also contained in $\sigma(\mathcal{A})\cup\left\{0\right\}$ which is compact.

Thus $S_{\varepsilon}\subset \sigma(\mathcal{A})$ is compact, and $|\hat{x}(h)|<\varepsilon$ outside of $S_{\varepsilon}$. This proves that $\hat{x}=\left.(x,0)^{\hat{}}\right|_{\sigma(\mathcal{A})}\in C_0(\sigma(\mathcal{A}))$.


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.