It seems that quite a standard trick of showing two $C^*$ algebras are as follows:

Let $A$ be a $C^*$ algebra $B$ another $C^*$ algebra. $A' \subseteq A$ be a subalgebra that is closed under $*$. (Not necessarily closed) and is dense in $A$. Let $L:A' \rightarrow B$ be a $*$-isometric homomoprhism which has dense image. Then $L$ extends to a $*$-isometric isomorphism .

So my aim is showing that $L$ actually extends to an injective map - since it is not necessarily case that $L$ is injective in its closure.


Given $a\in A$ take $\{a_j\}\subset A'$ with $a_j\to a$. Since $L$ is isometric, we have $$ \|L(a_j)-L(a_k)\|=\|L(a_j-a_k)\|=\|a_j-a_k\|, $$ so $\{L(a_j)\}$ is Cauchy. As $B$ is complete, there exists $b_a=\lim L(a_j)$. We want to extend $L$ by defining $L(a)=b_a$. For this we need to show that $b_a$ is unique for each $a$; and this follows easily: if $c_j\to a$, then $$ \|L(c_j)-L(a_j)\|=\|c_j-a_j\|\leq\|c_j-a\|+\|a-a_j\|\to0, $$ so $L(c_j)\to b_a$. So the extension $L$ is well-defined. Also $$ \|L(a)\|=\|\lim_j L(a_j)\|=\lim_j\|L(a_j)\|=\lim_j\|a_j\|=\|a\|, $$ so $L$ is isometric.

If you don't require $L$ to be isometric, the result is not true. Let $A=C[0,1]$, $B=\mathbb C\oplus C[0,1]$, and take $A'\subset A$ to be the $*$-algebra of complex polynomials. Define $L:A'\to B$ by $$ L(p)=p(2)\oplus p. $$ This is clearly an injective $*$-homomorphism. And it has dense image: Given $(\lambda,f)\in B$, we can choose a sequence of polynomials $\{p_j\}$ such that $p_j\to f$ uniformly on $[0,1]$ and $p_j(2)=\lambda $ (simply use Stone-Weierstrass on a continuous function $g$ that agrees with $f$ on $[0,1]$ and with $g(2)=\lambda$).

So $L$ satisfies all the hypotheses; but it is not bounded. If you consider $p_n(x)=x^n$, then $\|p_n\|=1$, while $p_n(2)=2^n$, so $\|L(p_n)\|=2^n$.

  • $\begingroup$ I suppose you mean $\pi(A')=L(A')$? The statement I know requires $\rho$ to be bounded is that its domain is a Banach $*$ algebra i.e. closed. It seems that the claim is: $L:A \rightarrow B$ is bounded, if $A$ is a $*$-algebra (with a norm such that $A$ is not necessarily complete), then $B$ is a $C^*$ algebra. Please correct me if I mistinterpret you. $\endgroup$ – CL. Apr 16 at 19:24
  • $\begingroup$ Martin, I have edited my problem, I noticed I have quite a lot of typos. $\endgroup$ – CL. Apr 16 at 19:34
  • $\begingroup$ Yes, I was glossing over the part that $*$-homomorphisms are bounded, and that doesn't work when the domain is not closed. I have rewritten the answer with a counterexample. $\endgroup$ – Martin Argerami Apr 16 at 20:09
  • $\begingroup$ Hi Martin, thanks a lot, I have added the condition that $L$ is an isometric isomorphism, I wonder if the argument holds from here. In fact, I am just trying to generalize the trick you did here: math.stackexchange.com/questions/1918086/… $\endgroup$ – CL. Apr 16 at 20:34
  • $\begingroup$ And please do keep the counter example in your edits : ) $\endgroup$ – CL. Apr 16 at 20:35

The answer by Martin is of course correct, I just want to emphasise that this statement and the way to see it is essentially elementary functional analysis. In my opinion is very helpful to be able to differentiate when a statement is about the linear structure and when a statement needs the C*.

Central is the following statement: Let $A:V\to W$ be a continuous linear map between normed vector spaces, then there is a unique continuous linear extension $\overline A :\overline V\to\overline W$ between the completions. This can be done as an elementary exercise and is basically the first half of Martin's answer.

(As a remark: The above statement is itself an expression of a more general statement, namely that Lipschitz maps between metric spaces $A,B$ so that $B$ is complete and $A$ is dense in some other metric space $A'$ always have a unique continuous extension $A'\to B$.)

The rest of what you need follows from:

  1. If a continous linear map is isometric on a dense sub-set, then it is isometric on its entire domain.

  2. If a continuous linear map is multiplicative wrt continuous multiplications on a dense sub-set, then it is multiplicative on its entire domain.

  3. If a continuous linear map $A$ satisfies $A(v^*)= A(v)^*$ on a dense sub-set and the $*$ operation is continuous on domain and codomain then the above equality holds on the entire domain.

Note that this is basically the same statement $3$ times: Continuous linear maps that preserve a continuous structure on a dense sub-set automatically do so on the entire domain.


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.