3
$\begingroup$

(This is a repost of a question from math.SE, https://math.stackexchange.com/questions/1240966/existence-of-state-on-a-c-algebra-satisfying-tauab-ab)

Let $a,b$ be elements of a unital C*-algebra $A$ with $0\leq a,b\leq 1$. Is it the case there is a state $\tau$ on $A$ such that $|\tau(ab)|=\|ab\|$? I'm particularly interested in the case when $a,b$ are projections and $\|ab\|=1$.

If $ab$ is normal (e.g., $a$ and $b$ commute), then this is a standard fact. Also, if $\|a\|=\|b\|=1$ and $\tau$ is a state satisfying $\tau(a)=\tau(b)=1$, then $\tau$ will work, but I don't see how to come up with such a $\tau$ if $a$ and $b$ don't commute.

$\endgroup$

1 Answer 1

6
$\begingroup$

I think there are $2\times 2$ counterexamples. Take $a = \pmatrix{1&0\cr 0&0}$ and $b = \pmatrix{.5&.5\cr .5&.5}$, so that $c = ab = \pmatrix{.5&.5\cr 0&0}$. If there were a state $\tau$ satisfying $\tau(c) = \lambda$ for some $|\lambda| = \|c\|$ then there would be a face of the state space which does this, so there is an extreme point, i.e., a vector state, with this property. But that would mean that $c$ has an eigenvalue with modulus equal to its norm, which is false (its only eigenvalue is $.5$, but its norm is $1/\sqrt{2}$).

However, if $\|ab\| = 1$ then the result is easy because (fixing some faithful representation) there is a sequence of unit vectors $(v_n)$ with the property that $\langle abv_n, v_n\rangle \to 1$, so you can take a weak* limit of these vector states.

(Key fact: if $a > 0$, $\|a\| = 1$, and $v$ is a unit vector with $\|av\|$ close to $1$ then $av$ is close to $v$. So if $\|ab\| = 1$ then there is a unit vector $v$ with $\|abv\|$ close to $1$, which successively implies that $\|bv\|$ is close to $1$, $bv$ is close to $v$, and $abv$ is close to $v$.)

(I wrote this quickly because I have to run, I hope it's right.)

$\endgroup$
1
  • $\begingroup$ Oops, $0$ is also an eigenvalue of $c$. That doesn't change anything, though. $\endgroup$
    – Nik Weaver
    Aug 25, 2017 at 19:10

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.