Let $A$ be a UHF-algebra of type $n^{\infty}$ and denote its unique and faithful trace by $\tau$. Let $L^2(A)$ be the Hilbert space of the GNS-representation associated to $\tau$. We have two commuting representations $L \colon A \to B(L^2(A))$ and $R \colon A^{\rm op} \to B(L^2(A))$ and by the universal property of the maximal tensor product and the nuclearity of $A$, we obtain a $*$-homomorphism $A \otimes A^{\rm op} \to B(L^2(A))$. I think the image of this is weakly dense, i.e. the von Neumann algebra completion of $A \otimes A^{\rm op}$ in this representation is type I and agrees with $B(L^2(A))$. Now consider the $*$-subalgebra $$ B = \left\{ \sum_{i} L_{a_i}R_{b_i} \ |\ \sum_{i} a_i b_i = 0\right\} $$ spanned by those operators corresponding to elements of $A \otimes A^{\rm op}$ that lie in the kernel of the multiplication map. Let $M$ be the weak closure of $B$ in $B(L^2(A))$.

What "is" this algebra $M$, more precisely: What is the type of $M$? Is it a factor?

  • $\begingroup$ I see that $B$ is an $A-A$ bimodule, but why is $B$ a subalgebra? $\endgroup$ May 19, 2014 at 15:22
  • $\begingroup$ @DavePenneys: The two actions commute and $R_aR_b = R_{ba}$, since it is a representation of the opposite algebra. If I now take two elements, say $\sum_i L_{a_i}R_{b_i}$ and $\sum_j L_{c_j}R_{d_j}$ of $B$ and multiply them, I end up with $\sum_{i,j} L_{a_ic_j}R_{d_jb_i}$, but $\sum_{i,j} a_ic_jd_jb_i$ should vanish since $\sum_j c_jd_j$ vanishes by assumption. $\endgroup$ May 19, 2014 at 15:28
  • $\begingroup$ Whoops, to get a *-subalgebra, I probably have to demand that the sum $\sum_i b_ia_i$ vanishes as well in the definition of $B$. $\endgroup$ May 19, 2014 at 17:22

1 Answer 1


Let $\xi_0 \in L^2(A)$ denote the cyclic vector corresponding to the identity in $A$, and let $P_0$ denote the rank-one projection corresponding to $\xi_0$. Then we clearly have $xP_0 = P_0 x = 0$ for all $x \in M$, and hence $M \subset P_0^\perp B(L^2(A)) P_0^\perp$. I claim that we actually have equality. This should be a well known fact to experts in II$_1$ factors (one just needs that $A$ is a unital $*$-algebra which generates a II$_1$ factor), however I don't know a reference off hand so I'll give a proof instead.

To see that $P_0^\perp \in M$ note that if $u \in A$ is a unitary then the spectral projection of $1 - L_{u}R_{u^*}$ corresponding to $\mathbb C \setminus \{ 0 \}$ is contained in $M$. The supremum of these projections over all $u$ is equal to $P_0^\perp$ since $\tilde A := L(A)''$ is a factor.

Note that the representations $L$ and $R$ extend to normal commuting representations of $\tilde A$ (for which I will use the same notation), and it is then easy to show that in the definition of $B$ we may allow $a_i$ and $b_i$ to be in $\tilde A$.

Note also that if $A_0 \subset \tilde A$ is a von Neumann subalgebra, and if $Q$ denotes the projection onto the closure of $(A_0' \cap \tilde A) \xi_0$, then $Q \in \mathbb CP_0 \oplus M$. This follows from the observation that for $\eta \in L^2(A)$ we have that $Q\eta$ is the unique element of minimal norm in the convex closure of $\{ L_uR_{u^*} \eta \mid u \in \mathcal U(A_0) \}$. Hence, $Q$ is in the weak operator topology convex closure of the set $\{ L_uR_{u^*} \mid u \in \mathcal U(A_0) \} \subset \mathbb CP_0 \oplus M$.

In particular, if $p \in \mathcal P(\tilde A)$ is a projection and we set $A_0 = ( \mathbb Cp \oplus \mathbb Cp^\perp )' \cap \tilde A$, then since $\tilde A$ is a factor it follows that the rank one projection onto $(p - \tau(p)) \xi$ is contained in $M$.

Suppose now that we have a self-adjoint operator $T \in M' \cap P_0^\perp B(L^2(A)) P_0^\perp$. Then for $p \in \mathcal P(\tilde A)$ a non-zero projection, we have shown that there exists $\lambda_p \in \mathbb R$ so that $T( p - \tau(p) )\xi_0 = \lambda_p (p - \tau(p))\xi_0$. If $q \in \mathcal P(\tilde A)$ then we have \begin{align} \lambda_p \langle (p - \tau(p))\xi_0, (q - \tau(q))\xi_0 \rangle &= \langle T(p - \tau(p))\xi_0, (q - \tau(q))\xi_0 \rangle \\ &= \langle (p - \tau(p))\xi_0, T(q - \tau(q))\xi_0 \rangle \\ &= \lambda_q \langle (p - \tau(p))\xi_0, (q - \tau(q))\xi_0 \rangle. \end{align} Since $\tilde A$ is a factor, any pair of non-zero projections has a third projection so that these inner-products are non-zero. Hence, $\lambda := \lambda_p = \lambda_q$ for all non-zero projections $p, q \in \mathcal P(\tilde A)$. The span of projections is norm dense in $\tilde A$ by the spectral theorem, hence $T(x - \tau(x) ) \xi_0 = \lambda (x - \tau(x))\xi_0$ for all $x \in \tilde A$, and so $T = \lambda P_0^\perp$. Since $T$ was arbitrary, the double commutant theorem then gives the result.


Your Answer

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

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