Let $A$ and $B$ be unital $C^*$-algebras, so we can view these as operator systems, and it makes sense to consider their injective envelopes $I(A)$ and $I(B)$. These injective envelopes become $C^*$-algebras for the Choi-Effros product.

Given a unital $*$-morphism $f: A \to B$, is it true that there exists a unique unital $*$-morphism $\overline{f}: I(A) \to I(B)$ that extends $f$?

Once the above question is answered positively (if the answer is positive), the following will probably be easy:

Is this construction functorial? I.e. is $I(-)$ a functor from the category of unital $C^*$-algebras to the category of unital $C^*$-algebras (with morphisms unital $*$-homomorphisms?

A reference is more than enough for me to be satisfied with an answer.


One can view $A$ and $B$ as sitting completely isometrically inside their injective envelopes $I(A)$ and $I(B)$. Then by injectivity a unital *-homomorphism (or more generally a unital completely positive map) $f:A\rightarrow B\subseteq I(B)$ extends to a unital completely positive map $\overline f:I(A) \rightarrow I(B)$.

[Edit: this should work]

Paulsen in this paper, Proposition 3.5, points out that any C$^*$-algebra containing $K(H)$ has injective envelope $B(H)$. Then $A = K(H) + \mathbb C I$ has $I(A) = B(H)$.

Consider the $*$-homomorphism $f:A\rightarrow \mathbb C$ given by $f(k+\alpha I) = \alpha$. Note that $I(\mathbb C) = \mathbb C$ and that any state of the Calkin algebra $B(H)/K(H)$ precomposed with the quotient map $q:B(H)\rightarrow B(H)/K(H)$ extends the map $f$. Therefore, $\overline f$ need not be a $*$-homomorphism or unique.

  • $\begingroup$ Thanks. I also got this far. I can show that if $f$ is a $*$-isomorphism, then $\overline{f}$ is also a $*$-isomorphism, but my proof uses a big gun. $\endgroup$
    – Andromeda
    Dec 1 '21 at 21:56
  • $\begingroup$ The fact that nothing is mentioned in Paulsen's book or any of the foundational papers makes me suspicious. The injective envelope is a very slippery beast. $\endgroup$ Dec 1 '21 at 21:59
  • $\begingroup$ Indeed, but Paulsen's treatment on the topic lacks some fundamentals (for example, he does not prove/mention that the $C^*$-algebra structure on an injective operator system obtained from the Choi-Effros product is unique, but uses this implicitly later in the chapter). $\endgroup$
    – Andromeda
    Dec 1 '21 at 22:04
  • 1
    $\begingroup$ I know that the above answers the problem in the negative, but (because I'm curious) what if one restricts to unital $\ast$-monomorphisms $f\colon A \to B$? By Hamana's construction of injective envelopes, we obtain an embedding $I(A) \to I(B)$ which extends $f: A \to B \subseteq I(B)$ so a $\ast$-homomorphism exists in this case (which was the obstruction above). I doubt it is unique though. $\endgroup$
    – Jamie Gabe
    Dec 1 '21 at 22:34
  • 3
    $\begingroup$ Oh no, $I(A)$ sits inside $I(B)$ as an operator system, not as a C*-subalgebra in this case! Thanks for clearing that up for me! $\endgroup$
    – Jamie Gabe
    Dec 1 '21 at 22:48

As mentioned by Chris, injective envelopes are brutal when seen as C$^*$-algebras.

Let $A=\text{UHF}(2^\infty)$ and $B$ the hyperfinite II$_1$ factor. Take $f$ to be the inclusion map. We have $I(B)=B$, while $I(A)$ is a wild AW$^*$ factor of type III.

If $g:I(A)\to B$ is a $*$-homomorphism and $\tau$ is the trace on $B$, then $\gamma=\tau\circ g$ is a trace on $I(A)$. In a type III AW$^*$-factor any projection $p$ can be halved, so there exist $p_1,p_2$ with $p=p_1+p_2$ and $p\sim p_1\sim p_2$, which gives us the usual
$$ \gamma(p)=\gamma(p_1)+\gamma(p_2)=2\gamma(p), $$ and so $\gamma(p)=0$ for any projection $p$. Thus $\gamma=0$. As $\tau$ is faithful, $g=0$.

In summary, $f$ is a $*$-monomorphism that admits no extension to a $*$-homomorphism, and in fact the only $*$-homomorphism $I(A)\to I(B)$ is the zero homomorphism.


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.