Let $M$ be a (not necessarily separable) von Neumann algebra. I am interested to understand the non separable case.

In the book [1, page 49], the authors says that $M$ is hyperfinite if we can write $M=\overline{\cup_\alpha M_\alpha}^{w^*}$ where $(M_\alpha)$ is a net of finite dimensional subalgebras directed by inclusion.

1) Is it equivalent to the definition of [2, page 97] ? This definition says that $M$ is approximately finite dimensional if for any $x_1,x_2,\ldots,x_n \in M$ and any $\sigma$-strong* neighborhood $V$ of $0$ in $M$ there exists a finite dimensional $*$-subalgebra $N$ such that $x_j \in N + V$ for any $j = 1,2,\ldots,n$.

2) In these definitions, is it to be assumed that $1_M \in M_\alpha$ and $1_M \in N$ (i.e. the subalgebras are unital subalgebras)? Is there any difference ?

[1] Sinclair, Smith, Finite von Neumann algebras and MASAS

[2] Takesaki, Theory of operator algebras 3

  • "Hyperfinite" is usually applied only to II$_1$-factors (and then there is a single one). Saying that $B(H)$ is hyperfinite, or that a type III-factor is hyperfinite sounds weird, since these algebras are properly infinite. So AFD is preferred.

  • If $M=\overline{\cup_\alpha M_\alpha}^{w^*}$ and $x_1,\ldots,x_n\in M$ and $V$ is a neighbourhood of zero, there is an alpha such that $x_1,\ldots,x_n\in M_\alpha+V$.

  • Conversely, if $M$ satisfies Takesaki's definition, let $\mathcal F$ be the net of finite subsets of $M_*$, ordered by inclusion. Note that a base of neighbourhoods of zero is given by $\{V_F\}_{F\in\mathcal F}$, where $$ V_F=\{x\in M:\ |f(x)|+|f(x^*)|<1:\ f\in F\}. $$ Let $\mathcal E$ be the net of finite subsets of $M$, ordered by inclusion. Form a net $$\mathcal A=\{F\times E:\ F\in \mathcal F,\ E\in\mathcal E\},$$ again ordered by inclusion. For any $\alpha=F\times E\in \mathcal A$, applying Takesaki's definition there exists a finite-dimensional $*$-subalgebra $N_\alpha$ of $M$ such that $E\subset N_\alpha+V_F$. If we let $M_\alpha=C^*(N_\alpha)$, this algebra is finite-dimensional because $N_\alpha$ is. If $\alpha\leq\beta$ then $M_\alpha\subset M_\beta$, because $E_\alpha\subset E_\beta$. Given $x\in M$ and a neighbourhood $V$ of zero, there exist $f_1,\ldots,f_n$ such that $0\in V_F\subset V$, where $F=\{f_1,\ldots,f_n\}$. By construction of the net, $x\in M_\alpha+V_\alpha$, where $\alpha=(F,\{x\})$. That is, there exists $x_0\in M_\alpha$ such that $x-x_0\in V_\alpha\subset V$. That is, $$ M=\overline{\bigcup_\alpha M_\alpha}^{\ \sigma-\text{strong}^*}. $$

  • The distinction between $\sigma$-weak and $\sigma$-strong$^*$ above is irrelevant, because the union $M_0=\bigcup_\alpha M_\alpha$ is a $*$-algebra (as the net is increasing). By [Takeksaki I, Theorem II.2.6] the $\sigma$-weak and $\sigma$-strong$^*$ continuous functionals are the same; then this result guarantees that the closures of convex sets are the same.

  • The unit of the subalgebras here is irrelevant. They can share the same unit as $M$ or not, it doesn't change the results.

| cite | improve this answer | |
  • $\begingroup$ Thank you for this beautiful answer. Why the unit of the subalgebras is here is irrelevant ? it is not clear for me. $\endgroup$ – Zouba Apr 10 '17 at 20:25
  • $\begingroup$ Because it is enough that the units converge to the unit of $M$. But they could be all different, and nothing changes in the arguments. $\endgroup$ – Martin Argerami Apr 11 '17 at 3:15

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.