3
$\begingroup$

Let $X$ and $Y$ be connected schemes that are smooth and projective of relative dimension 1 over $\mathbb{Z}[1/n]$ for some positive integer $n$. If the function fields of $X$ and $Y$ are isomorphic, are $X$ and $Y$ isomorphic themselves?

If $n=1$, the answer should be positive because there is only one curve with everywhere good reduction — the projective line.

The answer is positive when the generic fibers have positive genus as I learnt from Alex Youcis.

$\endgroup$
0

1 Answer 1

2
$\begingroup$

It seems to me that the class groups of quaternion algebras cause trouble here.

Fix a quaternion algebra which is split away from $\infty$ and primes dividing $n$, and take a maximal order. For any finite-index right ideal of this order, which is rank $4$ as a $\mathbb Z$-module, you can form, inside the Grassmanian of rank $2$ sublattices, the closed subset consisting of those that are right ideals.

Over any place where the quaternion algebra splits, the order will be isomorphic to a matrix algebra, and the ideal will be isomorphic as a module to the algebra itself, so this will locally be $\mathbb P^1$, i.e. a smooth curve. Over $\mathbb Q$ it has a canonical isomorphism to the Brauer-Severi variety, as the ideal is isomorphic to the whole algebra over $\mathbb Q$. So if your statement is true then there must exist an isomorphism of this scheme to the Brauer-Severi scheme.

But if that happens then this isomorphism must arise from a birational automorphism of the Brauer-Severi variety composed with the canonical isomorphism, which would be an invertible element of the quaternion algebra. So it seems your statement fails unless every right ideal is generated by some element of the quaternion algebra, which is the same as saying the ideal class group is zero.

For instance if $n=p$ is a prime, for the statement to fail it suffices to find two different supersingular elliptic curves in characteristic $p$, as you will be able to for $p=11$ and all $p$ greater than $13$.

$\endgroup$
15
  • $\begingroup$ does the statement hold for e.g. $n=3$? $\endgroup$
    – user145520
    Commented Sep 8, 2019 at 23:51
  • $\begingroup$ Hey Will. If you look at the post the OP linked to, I gave an idea of why it might be true for genus $0$. Do you know where the mistake is? $\endgroup$ Commented Sep 9, 2019 at 12:06
  • $\begingroup$ @AlexYoucis I think the issue is that two different Azumaya algebras can represent the same class in the Brauer group, even if vector bundles on the base scheme are all trivial, because tensoring them both with a matrix algebra might produce the same algebra. $\endgroup$
    – Will Sawin
    Commented Sep 9, 2019 at 12:35
  • $\begingroup$ @WillSawin Maybe I'm being silly here, but isn't the Brauer group classifying Brauer-Severi schemes up to isomorphism, not up to equivalence, right? So if you know that $X$ and $Y$ are Brauer-Severi schemes over $\mathrm{Spec}(\mathbb{Z}[\frac{1}{n}])$ as I tried (perhaps incorreclty) to argue in my post, then doesn't the injectivity of the localization map for Brauer groups on regular local schemes imply that generically isomorphic Brauer-Severi schemes are isomorphic? Thanks! $\endgroup$ Commented Sep 9, 2019 at 12:45
  • $\begingroup$ @AlexYoucis That is certainly not right over a general base (not a field). For instance the projectivization of any vector bundle is equivalent in the Brauer-group, but these are not necessarily isomorphic. $\endgroup$
    – Will Sawin
    Commented Sep 9, 2019 at 14:12

You must log in to answer this question.