# von Neumann algebras as C*-algebras with multiplicative conditional expectation $A^{**}\to A$

Let $A$ be a C*-algebra. We identify $A$ with its canonical image in the bidual $A^{**}$. Consider the following conditions:

(1) $A$ is a von Neumann algebra.

(2) There is a multiplicative conditional expectation from $A^{**}$ onto $A$, that is, a map $\pi\colon A^{**}\to A$ that is a *-homomorphism and such that $\pi(a)=a$ for all $a\in A$.

Then (1) implies (2): Consider the predual $A_*$ of $A$, and the canonical embedding $\kappa\colon A_*\to (A_*)^{**}$. Then the dual of $\kappa$ has the desired properties.

Note that $A^{**}$ is always a von Neumann algebra. One can show that (2) implies that $A$ is a monotone complete AW*-algebra. However, is it also a von Neumann algebra?

Q1: Does (2) imply (1) ?

Assuming (2), let $J$ denote the kernel of $\pi$. Then $A$ is (*-isomorphic to) the quotient $A^{**}/J$. However, a quotient of a von Neumann algebra by a closed, two-sided ideal need not be a von Neumann algebra. (For example, the Calkin algebra is such a quotient and not a von Neumann algebra.)

In the commutative case, we might also consider the following more general question:

Q2: Is the quotient of a commutative von Neumann algebra by a closed, two-sided ideal again a von Neumann algebra?

Equivalently, is every closed subset of a hyperstonean space again hyperstonean?

• Your proof that $A$ is complete, doesn't it also implies that $\pi$ is normal ? If yes then the Kernel of $\pi$ is weakly closed and I think you are done, isn't ? – Simon Henry Mar 25 '16 at 16:45
• This is not clear to me. To show that $A$ is monotone complete, let $(a_j)_j$ be a norm-bounded increasing net in $A_{sa}$. Then, since $A^{**}$ is monotone complete, there is $b\in A^{**}_{sa}$ that is the supremum: $a_j\leq b$ for all $j$, and $b$ is the smallest element with these properties. Now, since $\pi$ preserves order, we have $a_j=\pi(a_j)\leq\pi(b)$ for all $j$, and thus $b=\pi(b)$. However, if we start with any increasing net in $A^{**}_{sa}$, then I don't see why $\pi(\sup_j a_j) = \sup_j \pi(a_j)$. – Hannes Thiel Mar 25 '16 at 17:01
• Hum... yes you are right, it is not clear. – Simon Henry Mar 25 '16 at 17:15
• Ok, all I can tell is that your question is equivalent to the following question: if $A$ is a retract of a von Neumman algebra, does $A$ is also a von Neumann algebra ? To which I don't know the answer, but which sound a bit more like something that might have been studied... – Simon Henry Mar 25 '16 at 17:30
• (For my other remarks, I got confused by the fact that when $A$ is a von Neumann algebra then $\pi$ is normal because it is weakly continuous) – Simon Henry Mar 25 '16 at 17:37

The answer to Q2 is no, and $l^\infty/c_0$ is already a counterexample. Its lattice of projections is $\mathcal{P}(\omega)/fin$, which is not complete.