2
$\begingroup$

Let $A = \mathbb{C}[x,y,z]/(x y - z^k)$. In fact $A$ is the ring of $\mu_k$ invariants: $A = \mathbb{C}[u,v]^{\mu_k}$ where $g \in \mu_k$ acts by $g(u,v) = (g u, g^{-1} v)$.

This allows one to understand vector bundles on the smooth surface $Spec\ A - (0,0,0)$ as $\mu_k$ equivariant bundles on $Spec\ \mathbb{C}[u,v]$.

However I would like to understand how many of these vector bundles extend to all of $Spec\ A$.

QUESTION

It's well known for $k = 2$ that $Pic(Spec\ A)$ is trivial. But what about higher $k$ and higher rank?

One approach is to consider the natural map $\mathbb{C}[x,y]\to A$ which is finite and flat and then (identifying vector bundles with projective modules) work with modules over $\mathbb{C}[x,y]$ equipped with an endomorphism $\phi$ such that $\phi^k = x y$. But I didn't get very far with this.

$\endgroup$
  • 1
    $\begingroup$ Why don't you just consider the $\mu_k$-eigendecomposition of $\mathbb{C}[u,v]$ as a finite module over the subring $\mathbb{C}[u^k,v^k,uv] = \mathbb{C}[x,y,z]/\langle xy-z^k \rangle$? You are just asking which of these eigenmodules has rank $1$ at the maximal ideal $\langle x,y,z \rangle$. $\endgroup$ – Jason Starr Aug 13 '14 at 9:59
2
$\begingroup$

The only invertible sheaf on $U := \text{Spec}A \setminus \langle x,y,z\rangle$ that extends to all of $\text{Spec} A$ is the trivial invertible sheaf. As you correctly surmise, the invertible sheaves on $W$ are precisely the $\mu_k$-linearized invertible sheaves on $V := \text{Spec}\mathbb{C}[u,v] \setminus \langle u,v \rangle$. Of course every invertible sheaf on $V$ is trivial. So these invertible sheaves are just the characters of $\mu_k$. Since the characteristic is $0$, the character group is $\mathbb{Z}/k\mathbb{Z}$ (I guess even in characteristics dividing $k$, this is true). Thus, denoting by $$q:V\to U,$$ the natural quotient morphism, the invertible sheaves on $U$ are just the $\mu_k$-eigensheaves of the locally free sheaf $q_*\mathcal{O}_U$. The reflexive extensions of these sheaves to all of $\text{Spec} A$ are just the $\mu_k$-eigensheaves of $\mathbb{C}[u,v]$ considered as an $A$-module.

For every $r=0,\dots,k$, denote by $B_r \subset \mathbb{C}[u,v]$ the $A$-submodule of elements that have weight $r$ with respect to the given action. For every $r=1,\dots,k$, both $u^r$ and $v^{k-r}$ give linearly independent, nonzero elements in $B_r/\langle u^k,v^k,uv \rangle B_r.$ Thus, since the "fiber" of $B_r$ at $\langle u^k, v^k, uv \rangle$ is not rank $1$, the module $B_r$ is not projective of rank $1$. Thus, the only eigenmodule that is projective is $B_0$, i.e., $A$.

$\endgroup$
  • $\begingroup$ Great! The point I was missing is if a locally free on $U$ is to extend to a locally free on $Spec A$ it must be the reflexive extension. $\endgroup$ – solbap Aug 13 '14 at 22:01
  • $\begingroup$ If I understand correctly then the conclusion should be the same for higher rank: any vector bundle on $V$ is trivial so $Spec\ A$ you will see $B_{r_1} \oplus ... \oplus B_{r_n}$ and the fiber at $(u^k,v^k, u v)$ will be bigger than $n$ unless all $r_i = 0$. So $Spec\ A$ has only trivial vector bundles. $\endgroup$ – solbap Aug 13 '14 at 22:04
  • $\begingroup$ I believe this argument can be adapted to show that on $Spec\ A$ every principal $G$ bundle is trivial whenever $G$ is a split reductive group. $\endgroup$ – solbap Aug 14 '14 at 5:24
  • $\begingroup$ The idea is that if $P$ is a nontrivial $G$ bundle on $U$ and $W$ is any etale neighborhood of $(u^k,v^k,u v)$ then $P$ restricted to $W \cap U$ remains nontrivial. $\endgroup$ – solbap Aug 14 '14 at 5:26

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.