6
$\begingroup$

Let $V$ be an operator system.

Definition 1: A pair $(W, \kappa)$ is called extension of $V$ if $W$ is an operator system and $\kappa: V \to W$ is a unital complete isometry.

Definition 2: An extension $(W,\kappa)$ of $V$ is called injective extension if $W$ is an injective operator system.

Definition 3: An extension $(W, \kappa)$ of $V$ is called rigid if for every unital completely positive map $\varphi: W \to W$ with $\varphi \kappa = \kappa$, we necessarily have $\varphi = \iota_W$.

Definition 4: An injective extension $(W,\kappa)$ of $V$ is called minimal if $\kappa(V)\subseteq W_1 \subseteq W$ with $W_1$ injective implies that $W_1 = W.$

Question: Is a minimal injective extension necessarily rigid? Is there a quick way to see this? I have a rather long argument along the lines of stuff in Paulsen where we use minimal seminorms, but maybe there is an easier way. If it helps, I know that for injective extensions, rigidity = essentiality.

$\endgroup$
14
  • $\begingroup$ Probably I'm missing something- but isn't Fix($\phi$), the set of fixed points of $\phi$, an injective operator system contained in W, that contains $\kappa(V)$? By minimality of $W$, we then have Fix($\phi$)$=W$, proving rigidity. $\endgroup$ Nov 24, 2021 at 16:52
  • $\begingroup$ @DarthVader Why is $\operatorname{Fix}(\varphi)$ injective? $\endgroup$
    – Andromeda
    Nov 24, 2021 at 17:32
  • $\begingroup$ If $W$ is an operator system, and $\phi$ a ucp map on $W$, then there exists a "conditional expectation" from W to Fix($\phi$). Define $E: W \rightarrow Fix(\phi)$ by $E(w)= \lim_{N \rightarrow \omega} \frac{1}{N} \sum_{n=1}^N \phi^n(w)$, where $\omega$ is a nonprincipal ultrafilter. As $W$ is injective, so is $Fix(\phi)$, by composition of "conditional expectations". $\endgroup$ Nov 24, 2021 at 17:42
  • $\begingroup$ @DarthVader Thanks! I'm definitely not familiar with such an approach. Do you have a reference for this? You may also want to write an answer. $\endgroup$
    – Andromeda
    Nov 24, 2021 at 17:46
  • 1
    $\begingroup$ You are very welcome :). I posted this as an answer. $\endgroup$ Nov 24, 2021 at 18:21

2 Answers 2

3
$\begingroup$

Hamana's proof (Theorem 3.5 in Injective Envelopes of Operator Systems, PubL RIMS, Kyoto Univ. 15 (1979), 773-785) is fairly direct. Consider the partial ordering on the space $\Xi = \{ \phi \in UCP(W, W) \mid \phi \kappa = \kappa \}$ given by $\phi \prec \psi$ when $\| \phi(x) \| \leq \| \psi(x) \|$ for all $x \in W$. First, note that every element in $\Xi$ dominates a minimal element in $\Xi$ since if $\{ \phi_i \}_i$ is a decreasing net in $\Xi$, then taking any concrete realization $W \subset \mathcal B(\mathcal H)$ we have $E \phi \prec \phi_i$ for all $i$, where $E: \mathcal B(\mathcal H) \to W$ is any ucp idempotent and $\phi$ is any limit point of $\{ \phi_i \}_i$ in $UCP(W, \mathcal B(\mathcal H))$, which is compact in the topology of pointwise ultraweak convergence.

Second, note that any minimal element $\phi \in \Xi$ is an idempotent. Indeed, for any $N \geq 1$ and $x \in W$ we have $\| \frac{1}{N} \sum_{n = 1}^N \phi^n( x ) \| \leq \| \phi(x) \|$ and so by minimality it follows that this is equality. Considering $x = y - \phi(y)$ we have $\| \phi(y) - \phi^2(y) \| = \| \frac{1}{N} \sum_{n = 1}^N \phi^n(y - \phi(y) ) \| \leq \frac{2}{N} \| y \| \to 0$.

Thus, if every idempotent in $\Xi$ is the identity, it follows that the identity is the only map in $\Xi$.

$\endgroup$
1
$\begingroup$

Let $\varphi: W \rightarrow W$ be a u.c.p map.

Let Fix$(\varphi)=\{ w \in W: \varphi(w)=w\}$.

Then, Fix$(\varphi)$ is an injective operator system, if $W$ is an injective operator system. Moreover, Fix$(\varphi)$ contains $\kappa(V)$, as $\varphi \circ \kappa= \kappa$.

By minimality of $W$, we must therefore have that Fix$(\varphi)= W$, thereby establishing rigidity.

$\endgroup$
1
  • $\begingroup$ I think you should justify that $\operatorname{Fix}(\varphi)$ is injective! This is non-obvious to me, if it is even true at all! $\endgroup$
    – user160032
    Nov 28, 2021 at 22:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.