Let $A$ and $B$ be two C$^{*}$-algebras and suppose we have a non-zero projection $p\in A\otimes B$. (We can assume $A$ is nuclear, so that there is only one possible tensor product.)

Does there exist a choice of elements $a_{1},\ldots,a_{n}\in A$ and $b_{1},\ldots,b_{n}\in B$ such that:

  1. $\left\|\left(\sum_{i=1}^{n}a_{i}\otimes b_{i}\right)-p\right\|<\frac{1}{2}$;
  2. If $\pi$ is a non-zero irreducible representation of $A$ and $(\pi\otimes\operatorname{id})(p)=0$, then $\pi\left(\sum_{i=1}^{n}|a_{i}|\right)=0$?

All I can deduce is that for a given choice of $a_{i}$'s and $b_{i}$'s, and and irreducible representation $\pi$ of $A$, we have $$ \left\|\sum_{i=1}^{n}\pi(a_{i})\otimes b_{i}\right\|=\left\|(\pi\otimes \operatorname{id})\left(\left(\sum_{i=1}^{n}a_{i}\otimes b_{i}\right)-p\right)\right\|\leq \left\|\left(\sum_{i=1}^{n}a_{i}\otimes b_{i}\right)-p\right\|<\frac{1}{2}, $$ provided that $(\pi\otimes\operatorname{id})(p)=0$.

  • $\begingroup$ Are you sure you've stated the problem correctly? What you have written is clearly false, because you can always throw $x\otimes y - x\otimes y$ into the sum. So you would have to have $\pi(|x|) = 0$ for all $x$. $\endgroup$ – Nik Weaver Apr 13 '20 at 16:23
  • $\begingroup$ I am not picky about what the $a_{i}$'s and $b_{i}$'s are so long that the approximation for $p$ holds. I guess I am wondering if there is a way to choose them so that the quoted result is always true. $\endgroup$ – ervx Apr 13 '20 at 16:59
  • $\begingroup$ Given your "I guess I am wondering" could you please edit your question to give a more precise statement of what hypotheses you are assuming and what conclusion you want? $\endgroup$ – Yemon Choi Apr 13 '20 at 17:34
  • $\begingroup$ I have updated the question. $\endgroup$ – ervx Apr 13 '20 at 17:42
  • $\begingroup$ Could you change to a more specific title? $\endgroup$ – YCor Apr 13 '20 at 17:48

Yes, this is possible, assuming that $A$ (or $B$) is nuclear. The same argument below (using that exact $C^\ast$-algebras are locally reflexive) also works if $A$ or $B$ is exact and the tensor product is the spatial (aka minimal) tensor product.

The role of nuclearity is that any two-sided closed ideal $J \subseteq A\otimes B$ is the closed linear span of all tensor products $I_A \otimes I_B$ of two-sided closed contained in $J$, see Corollary 9.4.6 in Brown and Ozawa's book. Now, let $J_A \subseteq A$ be the intersection of all two-sided, closed ideals $J \subseteq A$ such that $p \in J \otimes B$ and let $I:= \bigcap J \otimes B$ where the intersection is indexed by such $J$. Clearly $J_A \otimes B \subseteq I$. For the converse (which is not as trivial as it a priori looks), take $I_A \otimes I_B \subseteq I$ where $I_A$ and $I_B$ are two-sided closed ideals in $A$ and $B$ respectively. Then $I_A \subseteq J_A$ and $I_B \subseteq B$, so $I_A \otimes I_B \subseteq J_A \otimes B$. By nuclearity of $A$, $I$ is the closed linear span of all such $I_A\otimes I_B$, so $I = J_A \otimes B$. In particular, $p\in J_A \otimes B$.

Now, let $a_1,\dots, a_n \in J_A$ and $b_1,\dots, b_n \in B$ such that \begin{equation} \| \sum_{i=1}^n a_i \otimes b_i - p \| < 1/2. \end{equation} Let $\pi \colon A \to \mathcal B(\mathcal H)$ be an irreducible representation such that $(\pi \otimes \mathrm{id}_B)(p) = 0$. The image of $(\pi \otimes \mathrm{id}_B)$ is canonically $\pi(A) \otimes B$, so by nuclearity of $A$ (or $B$) and exactness of maximal tensor products, it follows that \begin{equation} 0 \to (\ker \pi)\otimes B \to A \otimes B \to \pi(A) \otimes B \to 0 \end{equation} is exact, so the kernel of $(\pi \otimes \mathrm{id}_B)$ is $(\ker \pi )\otimes B$. Hence $p \in (\ker \pi)\otimes B$ so $a_1,\dots, a_n \in J_A \subseteq \ker \pi$ by construction of $J_A$.

If $A$ or $B$ is exact, we get $\ker (\pi \otimes \mathrm{id}_B)$ either by using that $A$ is locally reflexive or $B$ is exact to get that \begin{equation} 0 \to (\ker \pi) \otimes_{\min{}} B \to A \otimes_{\min{}} B \to \pi(A) \otimes_{\min{}} B \to 0 \end{equation} is exact.

  • $\begingroup$ Excellent answer! $\endgroup$ – Nik Weaver Apr 14 '20 at 4:33

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.