2
$\begingroup$

Let $C$ be a smooth projective curve of genus $g$ with an involution $\iota: C \to C$. We have the quotient map $\pi: C \to C/\iota$, with $C/\iota$ a smooth curve of genus $h$. The pullback map $\pi^{*}: \text{Jac}(C/\iota) \to \text{Jac}(C)$ in this case is injective, and we define the Prym variety as the cokernel:

$$0 \to \text{Jac}(C/\iota) \to \text{Jac}(C) \to \text{Prym}(\pi) \to 0$$

Note that $\text{Prym}(\pi)$ is an Abelian variety of genus $g-h$. Now consider the action by $\iota^{*}$ on $\text{Jac}(C)$. The induced action on $\text{Jac}(C/\iota)$ is trivial, i.e. line bundles pulled back from the quotient are $\iota$-invariant. I am nearly certain that the induced action on $\text{Prym}(\pi)$ is by $\pm 1$ and therefore has $2^{2g-2h}$ fixed points. But I am struggling to prove this. Anyone have any tips? I'm unsure how to prove that a particular involution on an Abelian variety is simply the standard $\pm 1$ action.

(I think an equivalent statement is that the only deformations of invariant degree $0$ line bundles on $C$ come from deformations on $C/\iota$. But there must be a more direct way to argue.)

$\endgroup$
2
  • 3
    $\begingroup$ Note that the tangent space at identity of $J(C)$ is $H^0(C,\Omega_C)^{*}=H^1(C,\mathcal{O}_C)$ on which $\iota$ operates. $\endgroup$
    – Kapil
    Aug 19, 2021 at 5:19
  • $\begingroup$ I see; I was over-complicating this. So the idea is that $H^{0}(C, \Omega_{C})^{*}$ decomposes as an $\iota$-representation. The subspace where the action is trivial corresponds to forms pulled back from the quotient, and the subspace where $\iota$ acts non-trivially is the tangent space of the Prym. This is what you're getting at, correct? So this implies the global action on the Prym is by $\pm 1$. $\endgroup$
    – Benighted
    Aug 19, 2021 at 5:53

1 Answer 1

2
$\begingroup$

Kapil's suggestion to express the tangent space as $H^1(C,\mathcal O_C)$ is great in characteristic $0$. In characteristic $p$, and in particular characteristic $2$, the statement is still true, but you can't detecet it from the tangent space.

Instead, note that the statement is equivalent to the claim that for $L$ a line bundle of degree $0$ on $C$, $L \otimes \iota^* L$ is isomorphic to the pullback of a line bundle from $C/\iota$.

One can check explicitly that $L \otimes \iota^* L$ is isomorphic to the pullback of $\det (\pi^* L )\otimes \det (\pi^* \mathcal O)^{-1}$. To do this, note that $\pi^* \pi_* L$ is an extension of $\iota^* L (-D)$ by $L$, for $D$ the branch divisor, so its determinant is $L \otimes \iota^* L(-D)$, and the $ \det (\pi^* \mathcal O)^{-1}$ term cancels the $(-D)$.

$\endgroup$
2
  • $\begingroup$ I think I understand the last paragraph (although I believe you meant $\pi_{*}L$ and $\pi_{*} \mathcal{O}$, if I'm not mistaken) but I'm not seeing how your reformulation of the statement is equivalent. Can you explain this briefly? $\endgroup$
    – Benighted
    Aug 23, 2021 at 4:21
  • 1
    $\begingroup$ @Benighted By definition, points of the Prym variety correspond to equivalence classes of line bundles of degree $0$ modulo pullbacks from $C/\iota$. To check that $\iota$ and the inverse map act the same on each point, we consider an arbitrary point, take a line bundle $L$ representing that class, and check that $L^{-1}$ is isomorphic to $\iota^* L$ modulo pullbacks, i..e that $\iota^* L \otimes L$ is isomorphic to a pullback. $\endgroup$
    – Will Sawin
    Aug 23, 2021 at 16:07

Your Answer

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

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