I have the following exercise:

Let $\pi:\mathcal A \rightarrow \mathcal B$ be a *-homomorphism between two unital $C^*$ algebras $\mathcal A$ and $\mathcal B$ which maps the unit to the unit. Assume $\forall A \in \mathcal A: A > 0 \Rightarrow \pi(A) > 0$. Proof, that $\pi$ is isometric.

My approach: I know $\pi \text{ isometric} \Leftrightarrow \pi \text{ injective}$, so I tried to prove, that there is an $A>0$ with $\pi(A) \not> 0$, if $\pi$ is not injective.

If $\pi$ is not injective, $\mathrm{ker}\ \pi \ne \{0\}$, so that I can take an $B\in \mathcal A$ with $B\ne 0$ and $\pi(B) = 0$. So $\tilde B = BB^*$ is self-adjoint and $\tilde B^2 \ge 0$. If $\tilde B$ is invertible, one has $\tilde B^2 > 0$ and $\pi(\tilde B^2) = 0 \not > 0$, which would be a contradiction to $\forall A \in \mathcal A: A > 0 \Rightarrow \pi(A) > 0$.

But want can I do, if $\tilde B^2$ is not invertible. Is there a way to construct an invertible element from $\tilde B^2$ with which I can do the proof? Thanks for any help!

  • $\begingroup$ Does $A>0$ mean $A$ self-adjoint with spectrum in $(0,+\infty)$? $\endgroup$ – Julien May 15 '13 at 20:56
  • $\begingroup$ Yes, I think so (In our lecture we defined $A\ge 0$ iff the spectrum is in $[0,\infty)$, so I guess, this would be the right definition) $\endgroup$ – Stephan Kulla May 15 '13 at 21:01
  • $\begingroup$ Given that $A\leq B$ is a partial order on the self-adjoint elements, $A>0$ could simply mean $A\geq 0$ and $A\neq 0$. In which case you are done. $\endgroup$ – Julien May 15 '13 at 21:07
  • 1
    $\begingroup$ Take $\mathcal A = C([0,1]):= \{ f: [0,1]\rightarrow \mathbb{C}\}$. The function $f\in C([0,1])$ with $f(x)=x$ would be an element in $\mathcal A$ with $f\ge 0$ and $f \ne 0$. ($0$ is in the spectrum of $f$ because $f(0)=0$ and the sprectrum of $f$ is the image $f([0,1])$) $\endgroup$ – Stephan Kulla May 15 '13 at 21:13
  • 2
    $\begingroup$ And since you mentioned that example, the unital $*$-homomorphism $\pi:f\longmapsto f(0)$ from $C([0,1])$ to $\mathbb{C}$ does verify your interpretation of $f>0 \Rightarrow \pi(f)>0$, and yet, it is not injective, nor isometric. So I believe my second interpretation is the correct one here: $A>0$ means $A$ positive nonzero element, i.e. self-adjoint, nonzero, spectrum in $[0,+\infty)$. And then you can check injectivity trivially. $\endgroup$ – Julien May 15 '13 at 21:26

I know you solved the problem years ago, nevertheless one can justify things slightly more thoroughly: by absurd you supposed $\pi$ non injective and want to show it is not strictly positive (the def. of strictly positive for an element is indeed that of user 1015).

The disjunction $\overline{B}$ invertible or not (and taking the square) is not relevant: what could make the argument fail is if $\overline{B}=0$. But here you use the $C^*$ property of the norm $$\left\lVert B B^* \right\rVert = \left\lVert B \right\rVert^2 $$ A norm is definite, i.e. $\left\lVert A \right\rVert = 0\quad \Leftrightarrow \quad A =0$, so that $B\neq 0 \quad \Leftrightarrow \quad \overline{B}\neq0$

  • $\begingroup$ I have attended to the lecture a long time ago and thus I have to admit that I do not really understand this topic any more... ;-) Before accepting your answer: Why is it enough to exclude $\overline B = 0$? $\endgroup$ – Stephan Kulla Jan 20 '16 at 18:05
  • $\begingroup$ The statement "$\pi$ is not strictly positive" reads: $\exists\ A\in \mathcal{A},\ A> 0$ (i.e. $A\geq 0$ and $A\neq 0$) but $\pi(A) = 0$. In our proof by contradiction we want to exhibit such an element (which we finally called $\tilde{B}$) in case $\pi$ is not injective: the candidate is to take $B\in \mathrm{Ker}(\pi)$ and then $\tilde{B}:=BB^*$ ($B^*B$ would also work) and it seems to work but it could be that $BB^*$ were actually $0$, in which case our proof is void. $\endgroup$ – Noix07 Jan 21 '16 at 17:12
  • $\begingroup$ funnily my tilde{} doesn't appear as such... $\endgroup$ – Noix07 Jan 21 '16 at 17:13

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.