6
$\begingroup$

Question: Let $k$ be an algebraically closed field, and $A,B$ abelian varieties over $k$. Suppose there exists an isogeny $A\to B$. Does this imply existence of a separable isogeny $A\to B$?


Known Cases: The answer is yes when $k$ has characteristic zero (because every isogeny is separable). More interestingly, the answer is also yes for elliptic curves over $\overline{\mathbb{F}_p}$: Since every isogeny factors into a Frobenius map and a separable map, it suffices to prove the result for $B=A^{(p)}$.

  • If $A$ is ordinary, then the Verschiebung map (dual to Frobenius) is separable, and this is true for all isogenous curves as well. So if the field of definition of $A$ is $\mathbb{F}_{p^k}$, then the composition of Verschiebung maps $$A\simeq A^{(p^k)}\to A^{(p^{k-1})}\to\cdots\to A^{(p^2)}\to A^{(p)} $$ is separable. (This argument applies more generally to any ordinary abelian variety over $\overline{\mathbb{F}_p}$.)
  • If $A$ is supersingular, we can take any prime $\ell\neq p$ and use the fact that the $\ell$-isogeny graph of supersingular curves is connected.

Further thoughts: I don't think the second bullet point above is the right place to look for a generalization; connectedness of $\ell$-isogeny graphs of abelian varieties is a very difficult problem in general, and it's also much stronger than we need. I also don't know how to approach the problem when $k\neq\overline{\mathbb{F}_p}$ has characteristic $p$.

$\endgroup$
5
  • $\begingroup$ Hi Jonathan! If there were a separable isogeny, the p(-power)-torsion group schemes would be isomorphic, right? Is it known that p-power isogenous abelian varieties have isomorphic p-divisible groups? $\endgroup$
    – Asvin
    Commented Mar 15, 2023 at 1:32
  • $\begingroup$ Oh hi Asvin! Thanks for the comment. That's a very good question but I don't know the answer - it somehow seems more unlikely when phrased that way, but I don't immediately see a reason it would be false. (I also just have no idea how one would go about showing that two abelian varieities have non-isomorphic p-divisible groups without using isogeny invariants.) $\endgroup$ Commented Mar 15, 2023 at 2:42
  • $\begingroup$ But do let me know if you have any ideas, partial or otherwise! $\endgroup$ Commented Mar 15, 2023 at 2:43
  • 1
    $\begingroup$ I think the answer should also be yes for abelian varieties over $\overline{\mathbb F}_p$ because we can lift them to CM AVs in char 0 with the same rational ring of endomorphisms along with a lift of the isogeny. The given isogeny will then correspond to an ideal $I$ that reduces to the inseparable isogeny mod $p$ and we can pick a different ideal $J$ in the same ideal class but now co-prime to $p$. Does this work? $\endgroup$
    – Asvin
    Commented Mar 15, 2023 at 4:32
  • 4
    $\begingroup$ Finally, I think this should be false for any non-isotrivial elliptic curve $E_t/ \overline{\mathbb F}_p(t)$ and the Frobenius isogeny. Suppose there were a $N$-power isogeny between $E_t$ and $E_t^{(p)}$ and let's assume we set things up so the parameter $t$ is the $j$-invariant. Then, we would have an equation of the form $\Phi_N(t,t^p) = 0$ wher $\Phi_N$ is the appropriate modular polynomial. But this is impossible for degree considerations! $\Phi_N(x,y)$ cuts out a $(\varphi(n),\varphi(n))$ cycle in $\mathbb P^1\times \mathbb P^1$ and it's intersection with $x = y^p$ is finite. $\endgroup$
    – Asvin
    Commented Mar 15, 2023 at 4:41

1 Answer 1

9
$\begingroup$

The answer is no. I think Asvin's example in the comments is correct but there may be a few things to check. I will give a different example that is easier to check, using Moret-Bailly's famous example of a nonisotrivial supersingular abelian surface.

Let $E$ be a supersingular elliptic curve over $F$, the algebraic closure of a finite field of characteristic $p$ and $A = E \times E$. The kernel of Frobenius on $A$ is isomorphic to $\alpha_p \times \alpha_p$ and we take $B = A/G$ where $G$ is the subgroup scheme of kernel of Frobenius given by $\{y=tx\} \subset \alpha_p \times \alpha_p$ where $t$ is transcendental over $F$. Then $k$ will be the algebraic closure of $F(t)$, $B$ is defined over $k$ but does not descend to $F$.

Now, any separable isogeny with source $A$ is $A \to A/H$, where $H$ is a finite etale subgroup of $A$, but such a subgroup is defined over $F$ so $A/H$ is defined over $F$ and can't be isomorphic to $B$.

$\endgroup$
2
  • 1
    $\begingroup$ Very nice! My first thought was also this surface but I didn't realize that etale isogenies would be defined over the base field! $\endgroup$
    – Asvin
    Commented Mar 15, 2023 at 15:21
  • 1
    $\begingroup$ Thank you - both Felipe and Asvin - for the nice examples! $\endgroup$ Commented Mar 16, 2023 at 3:01

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .