Suppose that $k$ is a commutative ring and that $A$ is an Azumaya $k$-algebra. Then there is a well-known morphism from $Aut_{Alg_k}(A)$, the group of algebra automorphisms, to the Picard group $Pic(k)$. One way to phrase it is as follows. Every automorphism $\phi$ determines a $k$-linear autoequivalence $$ \phi^*: Mod_A \to Mod_A $$ However, since $A$ is Azumaya, every $k$-linear autoequivalence of the category of (right) $A$-modules is of the form $M \mapsto M \otimes_k J$ for some $J$ in the Picard group of $k$, unique up to isomorphism. Moreover, this assignment takes composition to tensor product. This assignment is part of the Rosenberg-Zelinsky exact sequence $$ 0 \to k^\times \to A^\times \to Aut_{Alg_k}(A) \to Pic(k). $$ It is possible to express $J$ as $HH^0(A,A^\tau)$ where $A^\tau$ is $A$, but with half of its bimodule structure pulled back along $\phi$.

In particular, given an Azumaya $k$-algebra $Q$, let $A = Q \otimes_k Q$. Then A is also an Azumaya $k$-algebra with automorphism $\phi(a \otimes b) = b \otimes a$, and thus $Q$ determines a (2-torsion) element in $Pic(k)$. This assignment turns out to be Morita invariant and it respects the tensor product; this makes it a (2-torsion) homomorphism from the Brauer group $Br(k)$ to the Picard group $Pic(k)$.

Are there examples of rings $k$ for which this homomorphism is nontrivial?

  • $\begingroup$ Is the homomorphism you wrote down the same as the map $\mathrm{Br}(k) = \pi_0 \mathbf{Br}(k) \to \pi_1 \mathbf{Br}(k) = \mathrm{Pic}(k)$ determined by $\eta\in \pi_1(S^0)$ where $\mathbf{Br}(k)$ is the Brauer space of $k$? $\endgroup$
    – skd
    Jun 17 '19 at 18:53
  • $\begingroup$ @skd Yes, it is. (In fact, that's my original motivation for asking.) $\endgroup$ Jun 17 '19 at 20:30

It seems to me that the involution of $Q \otimes Q$ that exchanges $a \otimes b$ and $b\otimes a$ is inner, which means that the homomorphism you describe should always be trivial.

Here is the argument, I hope it is correct. Suppose that $Q$ is trivial, that is, there is a projective $k$-module $V$ such that $Q \simeq \operatorname{End}V$. Then $Q \otimes Q \simeq \operatorname{End}(V\otimes V)$. Consider the operator $\tau\colon V\otimes V \simeq V\otimes V$ that exchanges $v \otimes w$ and $w \otimes v$. We can think of $\tau$ as an element of $Q \otimes Q$; conjugation by $\tau$ exchanges $a \otimes b$ and $b\otimes a$.

Now, choose another projective $k$-module $W$ and an isomorphism $Q \simeq \operatorname{End}W$. Consider the induced isomorphism $\phi\colon\operatorname{End}V \simeq \operatorname{End}W$. Then there exists an invertible $k$-module $L$ and an isomorphism $\psi\colon W \simeq L \otimes V$ such that $\psi$ that induces $\phi$. This implies that $\tau$ is independent of the isomorphism $Q \simeq \operatorname{End}V$, and it is a canonical element of $Q\otimes Q$.

If $Q$ is not trivial, choose a faithfully flat extension $k'/k$ such that $Q_{k'}$ is trivial. Because it is canonical, by descent theory the element $\tau' \in Q_{k'}\otimes_{k'}Q_{k'}$ constructed above descends to an element $\tau \in Q \otimes_k Q$ which will induce the involution $Q \otimes Q$ that exchanges $a \otimes b$ and $b\otimes a$.

  • 1
    $\begingroup$ Wonderful, thank you! What I find curious about this is that it uses that the twist $L \otimes L \to L \otimes L$ is the identity for $L$ invertible; I had verified this separately and it was what led me to wonder if the same were true of the twist in the Brauer group. $\endgroup$ Jun 19 '19 at 6:26
  • 2
    $\begingroup$ Very nice! The element $\tau$ is known as the Goldman element of $Q$, see Propostion 5.1 in Saltman's "Lectures on Division Algebras", which also proves it is canonical. $\endgroup$ Jun 19 '19 at 6:46

Here is another way to see that the given map should be zero. It corresponds to a map of sheaves of grouplike $\mathbb{E}_\infty$-spaces $K(\mathbb{G}_m,2)\rightarrow K(\mathbb{G}_m,1)$. All such maps are nullhomotopic (for example by delooping to sheaves of spectra and using a $t$-structure argument).

I would also guess that the map is given by $A\mapsto\mathrm{HH}(A/k)$. Since Hochschild homology is symmetric monoidal, $\mathrm{HH}(A/k)$ is a line bundle. The argument above shows that it is trivial.

On the other hand, I suspect there are derived Azumaya algebras where this is non-trivial. If we look at the derived Brauer sheaf (on the étale site of an ordinary commutative ring) it is an extension $$K(\mathbb{G}_m,2)\rightarrow\mathbf{dBr}\rightarrow K(\mathbb{Z},1).$$ Taking Hochschild homology gives a map of sheaves of grouplike $\mathbb{E}_\infty$-spaces $\mathbf{dBr}\rightarrow\mathbf{dPic}$, where $\mathbf{dPic}$ fits into a fiber sequence $$K(\mathbb{G}_m,1)\rightarrow\mathbf{dPic}\rightarrow K(\mathbb{Z},0).$$ The map $\mathbf{dBr}\rightarrow\mathbf{dPic}$ thus canonically factors through a map $K(\mathbb{Z},1)\rightarrow K(\mathbb{G}_m,1)$ of grouplike $\mathbb{E}_\infty$-spaces. I believe this is the map that sends $1$ to $-1\in\mathbb{G}_m(\mathbb{Z})$. Indeed, the copy of $\mathbb{Z}$ is coming from suspension in the derived category.

To make this concrete, one needs a ring $k$ where the induced map $H^1(k,\mathbb{Z})\rightarrow\mathrm{Pic}(k)$ is non-zero. I don't have an example, but I'd be surprised if this didn't exist.

  • 2
    $\begingroup$ I believe that the ring $\mathbb C[x,y](y^2 - x^2(x-1))$ (the ring of the standard affine nodal cubic) should have this property. $\endgroup$
    – Angelo
    Jun 19 '19 at 16:05

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.