Objects with no rational models

The goal of this post, which is just a bit of fun between more serious posts/projects, is to discuss some examples of algebro-geometric objects over \overline{\mathbb{Q}} which have no models over smaller subfields and explain how moduli theory can help clarify their discovery in certain situations.


This post was born out of a very simple question I had at the beginning of my second year of graduate school:

Main question: “What is an explicit example of a variety X over \overline{\mathbb{Q}} which has no model over \mathbb{Q}?”

Let us be slightly more rigorous about what we mean. Namely, we say that a variety X over \overline{\mathbb{Q}} is defined over \mathbb{Q} if there exists a variety X_0 over \mathbb{Q} and an isomorphism (X_0)_{\overline{\mathbb{Q}}}\cong X.

This sounds like a fairly simple problem. For example, one is tempted to make the claim that

V(y-ix^2)\subseteq \mathbb{A}^2_{\overline{\mathbb{Q}}}

is not defined over \mathbb{Q} since its defining equation in \mathbb{A}^2_\mathbb{Q} involves non-rational coefficients. But, of course, obviously as a \overline{\mathbb{Q}}-variety one has that

V(y-ix^2)\cong V(y-x^2)

by the isomorphism \sqrt{i}x\mapsto x and the latter is obviously defined over \mathbb{Q}.

So, one is quickly led to trying more complicated examples involving equations of a more sophisticated form. For example, one might try equations of the form

E_{A,B}:=V(y^2z-x^3+Axz^2+Bz^3)\subseteq \mathbb{P}^2_{\overline{\mathbb{Q}}}

where A,B\in\overline{\mathbb{Q}} or, in other words, elliptic curves. This seems promising since if A and B are both irrational then there’s no obvious way to ‘fix this’. Indeed, if we try to ‘suck’ the B into Z^3, as we sucked the i into x^2 in our previous example, we see that we’ll be forced to complicate the coefficients of xz^2 and y^2z .  It’s then not at all clear that somehow you can make a change of variables that simultaneously makes all coefficients rational.

Of course, this is not a proof, and some complicated and surprising isomorphisms exist. For example, E_{\sqrt[3]{2},i} certainly looks like it should not be defined over \mathbb{Q} at first glance but, in fact, one has an isomorphism

E_{\sqrt[3]{2},i}\cong E_{\frac{-1934917632}{361},\frac{60183678025728}{6859}}

and the latter, as complicated as it is, is evidently defined over \mathbb{Q}.

Given the above, it then seems less obvious how to actually find an example of a variety X over \overline{\mathbb{Q}} which provably has no model, let alone an example where X satisfies specified properties (e.g. being some fixed dimension, smoothness, affineness, etc.).

The goal of this post is to try and explain a general geometric methodology by which one might try to answer the main question. We will then give some additional examples of objects without models which doesn’t fit quite as nicely into this geometric framework, but which are still interesting.

Moduli theory

We now develop a geometric framework in which finding examples of varieties with no models can be made to be methodological.

Moduli problems

It is first useful to widen our gaze and consider the general situation in which our main question naturally sits. Namely, let us consider the structure that the objects in our original have. Namely, we had

  • A family of objects over \mathbb{Q} (varieties over \mathbb{Q})
  • A family of objects over \overline{\mathbb{Q}} (varieties over \overline{\mathbb{Q}}).
  • A notion of isomorphism between objects in either situation.
  • A pullback map which associates to an object over \mathbb{Q} an object over \overline{\mathbb{Q}} (the fiber product of curves).

Of course, the structure above was just part of the more genral structure where instead of objects only over \mathbb{Q} or \overline{\mathbb{Q}} we have objects over T for any \mathbb{Q}-scheme T (namely T-schemes).

We see then that the framework in which to discuss the generalization of our main question is likely the following. Let us define a moduli problem over \mathbb{Q} to be a contravariant functor

\mathscr{S}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}


  • \mathsf{Sch}_\mathbb{Q} is the category of schemes over \mathbb{Q}.
  • \mathsf{Grpd} is the category of groupoids.

Here by a groupoid we mean a small category (i.e. a category whose collection of objects is a set) \mathcal{G} where every morphism in \mathcal{G} is an isomorphism. A morphism of groupoids is just a functor between the groupoids.

Let’s unpack this definition a bit. The data of a moduli problem over \mathbb{Q} is the following:

  • For every \mathbb{Q}-scheme a groupoid \mathscr{S}(T), which is just a collection of objects ‘over T‘ together with a notion of isomorphism between them.
  • For every map of \mathbb{Q}-schemes f:T_1\to T_2 a functor f^\ast:\mathscr{S}(T_2)\to\mathscr{S}(T_1)  (which we call the pullback functor)..

This is precisely the structure we observed our main question having above.

Remark 1: It is probably useful to put our definition into the context of more rigorous mathematics. Namely, what we’re calling a moduli problem over \mathbb{Q} is what is known (e.g. as in [Olsson, Chapter 3]) as a split category fibered in groupoids over \mathsf{Sch}_\mathbb{Q}. These are simpler objects to understand than general categories fibered in groupoids as explained in [Olsson, §3.4] but, as seen in loc. cit., essentially every category fibered in groupoids is equivalent to a split one (although it is not true that it itself must be split). Thus, we shall always confuse the two notions. We will see what this means more concretely below in Remark 3.

Let us give some examples of moduli problems over \mathbb{Q}

Example 2: We start with the moduli problem that was underlying our main question. Namely, let us define

\mathsf{Sch}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}

where we set \mathsf{Sch}(T) to be the category of T-schemes with morphisms being isomorphisms of T-schemes. The pullback functor is the usual fiber product of schemes/morphisms. More explicitly, if we have a morphism of \mathbb{Q}-schemes f:T_1\to T_2 then the functor

f^\ast:\mathscr{Sch}(T_2)\to \mathsf{Sch}(T_1)

is defined on objects by the fibered product f^\ast(X):= X\times_{T_2}T_1 and on morphisms by the pullback of morphisms to the fibered product.

Remark 3: We see in the above where, in reality, it is more natural to work with categories fibered in groupoids with no particular splitting as mentioned in Remark 1. Namely, the fiber product X\times_{T_2}T_1 is only defined up to isomorphism. So, to really make \mathsf{Sch} a moduli problem in our sense, one has to choose for all morphisms T_1\to T_2 and T_2-schemes X over T_2 an explicitly chosen fibered product X\times_{T_2}T_1 in a way that is compatible with composition of morphisms in \mathsf{Sch}_\mathbb{Q} . Again, we will ignore this subtelty both here and in the future examples.

Example 4: Let g\geqslant 0 and n\geqslant 0 be integers. For a \mathbb{Q}-scheme T we call a morphism f:C\to T a genus g curve over T if it is smooth, proper, and for every point t of T the k(t)-variety f:C_t\to\mathrm{Spec}(k(t)) is a geometrically connected curve of genus g. By an n-marked curve of genus g over T we mean a genus g curve f:C\to T over T together with n-marked sections e_i:T\to C of f. By a morphism of n-marked

(C,e_1,\ldots,e_n)\to (C',e_1',\ldots,e_n')

we mean a morphism F:C\to C' of genus g curves over T such that e_i'=F\circ e_i for all i.

We can then define a moduli problem over \mathbb{Q}

\mathscr{M}_{g,n}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}

where for an \mathbb{Q}-scheme T we set \mathscr{M}_{g,n}(T) to be the groupoid of n-marked genus g curves over T with morphisms being isomorphisms. For a morphism f:T_1\to T_2 we define the pullback map


to be as expected:



f^\ast(e_i):T_1\to C\times_{T_2}T_1

is e_i on the C-factor and \mathrm{id}_{T_1} on the T_1-factor.

Example 5: Let us define

\mathsf{Aff}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}

to associate to a \mathbb{Q}-scheme T the groupoid \mathsf{Aff}(T) of affine T-schemes with morphisms being usual isomorphisms of T-schemes. The pullback morphisms is the usual fiber product of affine schemes.

Example 6: Let

\mathcal{F}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Set}

be a presheaf on \mathsf{Sch}_\mathbb{Q}. Then, evidently \mathcal{F} is a moduli problem over \mathbb{Q} since any set can be thought of as a groupoid where the only morphisms are the identity morphisms.

Example 7: We give an even more special case of the last example. Namely, if X is any \mathbb{Q}-scheme, then we can think of X as a moduli problem over \mathbb{Q} by identifying X with its contravariant functor

\begin{aligned}X &:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Set}\\ &:T\mapsto X(T):=\mathrm{Hom}(T,X)\end{aligned}

which is reasonable to do by Yoneda’s lemma (e.g. see [Stacks, Tag001P])

Example 8: Let N\geqslant 1 be an integer. We then define the modular curve with full level structure to be the moduli problem

\mathscr{Y}(N):\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}

where \mathscr{Y}(N)(T) is the category of pairs (E,\iota) where E\to T is an elliptic scheme (which is a relative version of elliptic curves—see [KM, Chapter 2] or [Hida1, Chapter 2]) and

\iota:E[N]\to \underline{(\mathbb{Z}/N\mathbb{Z})^2} 

is an isomorphism of group schemes which we call a trivialization. The isomorphisms in this category are isomorphisms of elliptic schemes that preserve the trivialization in the obvious way.  The pullback map is the obvious one. Let us note, as essentially shown in [KM, Chapter 2] and [Hida1,Chapter 2], one has that \mathscr{Y}(1) is essentially identified with the moduli problem \mathscr{M}_{1,1} from Example 4.

There is a natural generalization of \mathscr{Y}(N) to higher abelian varieties denoted by \mathscr{A}_{g,N} for any integers g\geqslant 1 and N\geqslant 1 so that \mathscr{Y}(N)=\mathscr{A}_{1,1,N}. To define this would take us too far astray, but see cite [Hida2, §6.3.2] for details (note that in the notation of [Hida2, §6.3.2] our notation is n=g, d=1, and N=N).

Example 9: We denote by 

\mathsf{LAG}:\mathsf{Sch}_\mathbb{Q}\to \mathsf{Grpd}

the moduli problem which associates a \mathbb{Q}-scheme T the category or finite type affine group schemes over T with morphisms being isomorphisms of T-group schemes. The pullback map is the obvious one. We also have a submoduli problem \mathsf{Red} of \mathsf{LAG} consisting of those affine group schemes which are reductive (e.g. see [Conrad1, Definition 3.1.1]).

Example 10: Let G be a (connected) reductive group over \mathbb{Q} (e.g. see [Milne, §19.b]). We then define a presheaf


which assigns to a \mathbb{Q}-scheme T the set \mathsf{Tor}_G(T) of maximal tori in G_T (e.g. see [Conrad1, §3.2]). 

We define a morphism \eta:\mathscr{S}_1\to \mathscr{S}_2  of moduli problems over \mathbb{Q} to just be a a natural transformation of functors \mathsf{Sch}_{\mathbb{Q}}\to \mathsf{Grpd}. There is the obvious notion of an isomorphism then between moduli problems over \mathbb{Q} which means that for all \mathbb{Q}-schemes T the functor \eta(T):\mathscr{S}_1(T)\to\mathscr{S}_2(T) is an isomorphism of categories. We call \eta an equivalence if \eta(T) is an equivalence of categories for all T.

We end this subsection by making a definition that is clearly useful given that our main question is about isomorphism classes. Namely, note that if \mathcal{G} is a groupoid then the set


is a set which we call the connected components of \mathcal{G}— aptly named since if we think of \mathcal{G} as being a directed multigraph then \pi_0(\mathcal{G}) is the set of connected components in the graph theoretic sense.

We note then that associated to any moduli problem \mathscr{S} over \mathbb{Q} is a presheaf 

\begin{aligned}\pi_0 (\mathscr{S})&:\mathsf{Sch}_\mathbb{Q}\to\mathsf{Set}\\ &:T\mapsto \pi_0(\mathscr{S}(T))\end{aligned}

which we call the component presheaf of \mathscr{S}. Let us note that if if \mathscr{S}_1\to \mathscr{S}_2 is a morphism of moduli problems over \mathbb{Q} then we get an induced morphism \pi_0(\mathscr{S}_1)\to \pi_0(\mathscr{S}_2). Moreover, if \mathcal{F} is a presheaf then since \mathcal{F}(T) has no non-identity morphsims for all T, we see that \pi_0(\mathcal{F})=\mathcal{F}. Combining these two, we see that any morphism \mathscr{S}\to\mathcal{F}, where the former is a moduli problem over \mathbb{Q} and the latter is a presheaf, factors uniquely through the obvious map \mathscr{S}\to \pi_0(\mathscr{S}).

The generalized main question and fine moduli scheme

So, we have formalized a generalization of the objects in the main problem in the last section but, as of now, this all seems like empty verbiage. What is the point of the above if it doesn’t help us solve the main question? Well, it does. Or, perhaps more to the point, it more easily allows us to create a framework in which our main problem can be systematically solved.

Before we do that though, let us try and formalize what the extension of the main question to the general situation would be. A moment’s thought shows that, with our above notational choices, we can formulate it as follows:

Generalized main question: Let \mathscr{S} be a moduli problem over \mathbb{Q}. Then, is the map

\pi_0(\mathscr{S})(\mathbb{Q})\to \pi_0(\mathscr{S})(\overline{\mathbb{Q}})


Let’s warm-up for our approach to understanding the generalized main question by thinking about the problem in a very specific case. Namely, as we saw in Example 7 it’s true that any scheme over V over \mathbb{Q} defines a moduli problem over \mathbb{Q}. It turns out that we can leverage geometry then to answer the generalized main question quite easily for V (assuming minor conditions).


Proposition 11: Let V be a scheme locally of finite type over \mathbb{Q} which is of positive dimension. Then, the map V(\mathbb{Q})\to V(\overline{\mathbb{Q}}) is not surjective.

Proof: Let U be any affine open subset of V. Evidently then one has that the map

U(\mathbb{Q})\to U(\overline{\mathbb{Q}})

is surjective. Since V has an affine open subscheme of the same dimension, we may assume without loss of generality that V is affine.

By Noether normalization (e.g. see [Stacks, Tag00OY]) we know that there exists a finite surjective map V\to \mathbb{A}^d_\mathbb{Q} where d:=\dim(V)>0. Note then that we have a commuting diagram

\begin{matrix} V(\mathbb{Q}) & \to & V(\overline{\mathbb{Q}})\\ \downarrow & & \downarrow\\ \mathbb{A}^d_\mathbb{Q}(\mathbb{Q}) & \to & \mathbb{A}^d_\mathbb{Q}(\overline{\mathbb{Q}})\end{matrix}

Now, the right vertical map is surjective since V\to \mathbb{A}^d_\mathbb{Q} is surjective and, by assumption, the top horizonal map is surjective. Evidently this then implies that the bottom horizontal map is surjective. But, this is a contradiction. \blacksquare

Of course, this gives a fairly definitive answer to the generalized main question in the case when \mathscr{S} is a scheme. In fact, it’s to our benefit to put a little more thought into this statement. Namely, let us say that a moduli problem over \mathbb{Q} has a fine moduli scheme if there exists a scheme X over \mathbb{Q} and an equivalence \eta:\mathscr{S}\to X. Note that by Yoneda’s lemma such a scheme X is unique up to isomorphism.

We can then deduce from Proposition 11 the following:

Corollary 12: Let \mathscr{S} be a moduli problem over \mathbb{Q} which posseses a fine moduli scheme X which is locally of finite type over \mathbb{Q} and of positive dimension. Then, the map

\pi_0(\mathscr{S})(\mathbb{Q})\to \pi_0(\mathscr{S})(\overline{\mathbb{Q}})

is not surjective.

We now consider how this interacts with the examples from above by discussing those which possess a fine moduli scheme:

Example 13: Let N\geqslant 3 and g\geqslant 1 be integers. It is then a theorem of Mumford (e.g. see [Hida2, Theorem 6.20]) that \mathscr{A}_{g,N}, the moduli problem from Example 8, has a fine moduli scheme which is quasi-projective, smooth, and of dimension \frac{1}{2}g(g+1) (for this latter fact see the description of the complex points of \mathscr{A}_{g,N} after (A4) on [Hida2, Pg. 277]). From this and Corollary 12 we deduce a negative answer to the generalized main question for the moduli problem \mathscr{A}_{g,N} for N\geqslant 3.

Example 14: Let G be a reductive group over \mathbb{Q}. Then, the moduli problem \mathsf{Tor}_G classifying maximal tori in G, as in Example 10, has a fine moduli scheme space which is smooth, quasi-affine, and of positive dimension of G is not a torus. Indeed, by [Conrad1, Theorem 3.2.6] the fine moduli scheme exists and is isomorphic to G/N_G(T) for any maximal torus T of G. Now, if G/N_G(T) were zero-dimensional it must be trivial since G/N_G(T) is smooth (by loc. cit.) and geometrically connected (since G is, see [Milne, Theorem 2.37(c)]). But, this then implies that G=N_G(T) and, since G is connected, that G=N_G(T)^\circ. But, this implies that G=T (e.g. combine [Milne, Corollary 17.39(a)] and [Milne, 17.84]). Thus, combining this with Corollary 12 we see that we have a negative answer to the generalized main problem for \mathsf{Tor}_G

Of course, the problem is that ‘most’ interesting moduli problems are not going to have a fine moduli scheme. In fact, there is a good way to think about why this is a case. One can easily note two things about schemes which makes them special in the pantheon of moduli problems over \mathbb{Q}

  1. They have ‘gluing properties’ that a general moduli problem over \mathbb{Q} needn’t have. More rigorously, any presheaf which has a fine moduli scheme is a sheaf for the étale or fppf topologies on \mathsf{Sch}_\mathbb{Q} (e.g. see [Olsson, §2.1-§2.2] and [Olsson, Theorem 4.1.2]).
  2. They have no non-trivial automorphisms. Namely, for any presheaf \mathcal{F} on \mathsf{Sch}_\mathbb{Q} and any \mathbb{Q}-scheme T we have that \mathcal{F}(T) is a set and so, in particular, for any object X of \mathcal{F}(T) one has that \mathrm{Aut}(X) is trivial (where this automorphism group is in terms of morphisms in the category \mathcal{F}(T)).

In particular, from 2. we see that any moduli problem \mathscr{S} over \mathbb{Q} which has objects with non-identity automorphisms is bound to not have a fine moduli scheme.

Remark 15: What we’re describing above might be closer to what is called an ‘algebraic space’. See Remark 17 below.

That said, one might hope to ‘rigidify’ a moduli problem over \mathbb{Q} so that it actually obtains a fine moduli scheme or, at least, so that it possesses property 2. from above. Namely, if one has an object X then for any extra structure on X the set of automorphisms preserving this extra structure certainly is going to be smaller than the set of all automorphisms of X. So, adding extra structure (i.e. rigidifying) should cut down on the automorphism groups of objects X in our moduli problem.

For example, \mathscr{Y}(1)=\mathscr{M}_{1,1} has no fine moduli scheme since, for example, condition 2. fails for \mathscr{Y}(1)—an elliptic curve has non-trivial automorphisms (e.g. the inversion map [-1]). But, one can think of \mathscr{Y}(N) for N\geqslant 1 as a rigidification of \mathscr{Y}(1) since it’s essentially the data of an object of \mathscr{Y}(1) and an extra piece of information given by the isomorphism \iota. Now, for N\geqslant 3 (as implicitly stated in Example 13) one has that the rigidified objects of \mathscr{Y}(N) have no automorphisms and, in fact, possess fine moduli schemes. 

Of course, this is not particularly useful for solving our main question since even if one can rigidify a moduli problem \mathscr{S} over \mathbb{Q} so that it has a fine moduli scheme, all that will tell us is that there are objects of \pi_0(\mathscr{S})(\overline{\mathbb{Q}}) which don’t have models over \mathbb{Q} with the extra rigidifying structure. Of course, this doesn’t preclude that these objects do have models over \mathbb{Q} without this extra structure.

Coarse moduli spaces

So, we saw in the last section that one has a fairly concrete way of showing that the generalized main question has a negative answer if our moduli problem has a fine moduli scheme. We also saw that at the end of the last section that even if we can somehow modify a moduli problem to have a fine moduli scheme that this is not, in general, good enough to answer the genearlized main question. So, what do we do?

One obvious guess is that instead of trying to force Proposition 11 to apply directly to a moduli problem \mathscr{S} by assuming that \mathscr{S} has a fine moduli scheme, that we should try to adapt Proposition 11 to work in more generality.  

Of course, we used geometry in the proof of Proposition 11 and so we will need to have some sort of theory of geometry for moduli spaces. But, as it stands, moduli spaces are so general as to make this not really tenable. So, a first step would be single out those moduli problems that are sufficiently geometric. Doing this, and carrying out the development of geometry in those cases, is precisely the modern theory of (algebraic/Artin) stacks as initiated by Deligne and Mumford in [DM] and further developed by many people.

We shall not need the theory of algebraic stacks explicitly but we mention very briefly the two extra major conditions that one imposes on a moduli problem \mathscr{S} to make them an algebraic stack. The first is that \mathscr{S} is a a ‘sheaf’ for the étale (equivalently, for algebraic stacks, the fppf topology—see [LMB, Corollaire 10.7]). Of course, since \mathscr{S} has values in groupoids, not sets, this is more complicated than the idea of being a presheaf being a sheaf for the étale topology and in fact making sense of what the sheaf condition means in this context is exactly what the ‘stack’ portion of the moniker ‘algebraic stack’ is referring to—see [Olsson, Chapter 4] for details. The second condition is that while \mathscr{S} is not a scheme, it at least needs to be covered by a scheme in a sense we shall not make precise here.

Remark 16: Several remarks are in order here. First, we only stated the two extra ‘major’ conditions to be an algebraic stack. There is one extra condition which is necessary—see [Olsson, Definition 8.1.4] for details. For more information on stacks see [Olsson], [LMB], and [Stacks, Tag0ELS] for general theory and [Bertin] for a more leisurely introduction.

Now the main reason that the theory of algebraic stacks is not explicitly useful for us here is that in fact (at least to the author’s knowledge) while it is incredibly rich theory, it is not sufficient to repeat the argument in Proposition 11. 

This brings us to the other obvious guess for trying to escape the yoke of fine moduli schemes. Namely, for a moduli probelm \mathscr{S} we can ask whether or not there exists a scheme M which may not be equivalent to \mathscr{S}, but ‘best approximates’ \mathscr{S} (amongst schemes) in a precise sense.

Let us try to be more rigorous. Namely, suppose that \mathscr{S} is a moduli problem over \mathbb{Q}. We then say that a \mathbb{Q}-scheme M is a coarse moduli scheme for \mathscr{S} if there exists a morphism of moduli problems

\pi:\mathscr{S}\to M

such that

  1. The map \pi is initial amongst maps to \mathbb{Q}-schemes.
  2. The map \pi_0(\mathscr{S})(L)\to M(L) is a bijection whenever L is an algebraically closed field.

Note that by condition 2. the space M is uniquely determined up to isomorphism by these conditions.

Remark 17: A remark is in order about the relationship between the above, which is non-standard terminology, and the notion of coarse moduli space which exists in the literature (e.g. [Olsson, Definition 11.1.1]).

The category of \mathbb{Q}-schemes is not very well-behaved within the category of moduli spaces over \mathbb{Q} (e.g. a presheaf which étale locally is isomorphic to a scheme need not be isomorphic to a scheme). For this reason one generally enlarges the category of schemes to the category of algebraic spaces. Intuitively these are presheaves which are étale locally schemes (e.g. see [Stacks, Tag0BGQ]) or they are also algebraic stacks with no automorphisms (e.g. see [Stacks, Tag04SZ]). 

One then defines a coarse moduli space by replacing every instance of ‘scheme’ with ‘algebraic space’ in the definition of  coarse moduli scheme. Note that if M is a coarse moduli space of a moduli problem \mathscr{S} which is also a scheme, then it is a coarse moduli scheme in the sense above. The converse is not clear to me (is there a non-scheme algebraic space which has an initial map to schemes?), but we will never really contend with this issue here. 

One definitive advantage of working with coarse moduli spaces opposed to coarse moduli schemes is the huge generality in which they exist. This is typified by the Keel–Mori theorem (e.g. see [Olsson, Theorem 11.1.2] and [Conrad2]).

Note that since M is a scheme it’s a presheaf and thus as mentioned before we have a factorization

\begin{matrix}\mathscr{S} & \to & M\\ \downarrow & \nearrow & \\ \pi_0(\mathscr{S})\end{matrix}

and we denote this factorized arrow \pi_0(\mathscr{S})\to M by \pi as well.

Now, as intimated the existence of a coarse moduli scheme is actually a strong enough to answer the generalized main question in the negative. More rigorously:

Proposition 18: Suppose that \mathscr{S} is a moduli problem over \mathbb{Q} which admits a coarse moduli scheme M which is a positive-dimensional scheme locally of finite type over \mathrm{Spec}(\mathbb{Q}). Then, the map

\pi_0(\mathscr{S})(\mathbb{Q})\to \pi_0(\mathscr{S})(\overline{\mathbb{Q}})

is not surjective.

Proof: Let us note that we have a commuting diagram

\begin{matrix}\pi_0(\mathscr{S})(\mathbb{Q}) & \to & \pi_0(\mathscr{S})(\overline{\mathbb{Q}})\\ \downarrow & & \downarrow\\ M(\mathbb{Q}) & \to & M(\overline{\mathbb{Q}})\end{matrix}

Since the right vertical map is bijective since M is a coarse moduli scheme of \mathscr{S}, we see that if the top horizontal map is surjective then so is the bottom horizontal map. But, this then contradicts Proposition 11. The exact same argument works to show that the map \pi_0(\mathscr{S})(\overline{\mathbb{Q}})\to \pi_0(\mathscr{S})(\mathbb{C}) is not surjective. \blacksquare

Remark 19: One can replace Proposition 18 by a generalization where one only assumes that \mathscr{S} admits a coarse moduli space S (in the sense of Remark 17) which has a positive dimensional open subscheme of finite type over \mathbb{Q}. This is, I believe, automatically the case if \mathscr{S} is separated, locally of finite type, and S is of positive dimension (in the sense of [Stacks,Tag04N6])—indeed one uses [Stacks, Tag0BAS] and the definition of dimension.

So, we see that the existence of a coarse moduli schemes (which is sufficiently well-behaved) is enough to guarntee a negative answer to our generalized main question. Of course, this is only useful if, unlike the case of fine modulie schemes, more moduli problems have coarse moduli schemes. The main example of this will be discussed in the next section.

But, before we go on we note that we can actually make our solution to the generalized main question for a moduli problem \mathscr{S} with coarse moduli scheme \pi:\mathscr{S}\to M more constructive. Namely, let us say that an element X of \pi_0(\mathscr{S})(\overline{\mathbb{Q}}) is \pi-rational if \pi(X) is in M(\mathbb{Q})\subseteq M(\overline{\mathbb{Q}}). Then, evidently we see that if X is in the image of the map \pi_0(\mathscr{S})(\mathbb{Q})\to \pi_0(\mathscr{S})(\overline{\mathbb{Q}}) then X is \pi-rational. So, any X which is not \pi-rational cannot be in the image. This will be the main technique we will use to find objects with no model—find objects which are not \pi-rational.

The main question for curves

We now try to leverage our discussion of coarse moduli schemes from the last section to answer more explicitly the generalized main question for the moduli problems \mathscr{M}_g. It turns out that the nature of this discussion varies greatly depending on whether g=0, g=1, or g\geqslant 2. So, we handle these cases separately.

The case of genus one

We start with the case of genus 1 because it has both an interesting answer, but also is likely to be more familiar to those with a basic background in algebraic geometry. The thing that makes the case of genus 1 slightly more tricky than the case of genus greater than 1 is that \mathscr{M}_1 doesn’t apparently have a coarse moduli scheme if one looks through the literature. The reason for this is quite simple: \mathscr{M}_1 is not an algebraic stack, and thus it’s not common to talk about coarse moduli spaces for such objects. But, we’ll see here that this doesn’t matter much.

We begin by noting that while \mathscr{M}_1, the moduli problem of genus 1 curves, doesn’t apparently have a coarse moduli scheme, the closely related moduli problem \mathscr{M}_{1,1}=\mathscr{Y}(1) of elliptic curves does:

Proposition 20: The moduli problem \mathscr{M}_{1,1} has \mathbb{A}^1_\mathbb{Q} as a coarse moduli space.

Proof: For a full rigorous proof see [Olsson, Theorem 13.1.15]. \blacksquare

In fact, the map


realizing \mathbb{A}^1_\mathbb{Q} is the j-invariant. See loc. cit. for more details, but let us suffice it to say that if one has an extension L of \mathbb{Q} then every elliptic curve E over L is isomorphic to one of the form

E_{A,B}:=V(y^2z-x^3-Axz^2-Bz^3)\subseteq \mathbb{P}^2_L

for some A,B\in L such that 4A^3+27B^2\ne 0 (e.g. see [Silverman, §III.1] and [Silverman, §III.3]). Then, 

\displaystyle j(E_{A,B}):=\frac{2^8 3^3 A^3}{4A^3+27B^2}\in L=\mathbb{A}^1_\mathbb{Q}(L)

(see again [Silverman, §III.1] and compare this with [Olsson, Theorem 13.1.15]).

From this, we deduce the following:

Corollary 21: Let E be an elliptic curve over \overline{\mathbb{Q}}. If E has a model over \mathbb{Q} as an elliptic curve, then j(E)\in\mathbb{Q}.

In fact, one can show that 

\displaystyle j\left(V(y^2z-x^3-\frac{36}{j_0-1728}xz^2-\frac{1}{j_0-1728}z^3)\right)=j_0

if j_0\ne 0,1728 and

\displaystyle j\left(V(y^2z+yz^2-x^3)\right)=0,\qquad j\left(V(y^2z-x^3-xz^2)\right)=1728

so that the composition

j:\mathscr{M}_{1,1}(L)\to \pi_0(\mathscr{M}_{1,1})(L)\to \mathbb{A}^1_\mathbb{Q}(L)=L

is surjective for all field extensions L of \mathbb{Q}. Using this, one can actually upgrade the only if statement in Corollary 21 to an if and only if statement.

So, how does this answer our question about \mathscr{M}_1? Well, I claim that we can use the j-invariant to build a coarse moduli scheme for \mathscr{M}_1. How? Well, the key observation is that we actually have a contravariant morphism of moduli problems


given on a T-scheme T as the functor


sending a curve C of genus 1 over T to its Jacobian \mathrm{Jac}(C) which is, by definition, the identity component \mathrm{Pic}^0_{C/T} (e.g. see [BLR, Chapter 8] and [BLR, Theorem 1 in §9.3]). The Jacobian \mathrm{Jac}(C) is evidently an element of \mathscr{M}_{1,1}(T) since it’s a smooth group scheme over T whose fibers are cruves of genus 1 (as can be seen from [Silverman, Proposition 3.4 in §III.3]) and which is proper (e.g. see [EGAIV Troisième partie, Corollaire 15.7.10]). 

Now, what makes the map \mathrm{Jac} so useful is the following:

Proposition 22: For any algebraically closed extension L over \mathbb{Q} the induced map

\mathrm{Jac}:\pi_0(\mathscr{M}_1)(L)\to \pi_0(\mathscr{M}_{1,1})(L)

is a bijection.

Proof: Suppose that C_1 and C_2 are two genus 1 curves over L. Choose base points e_i \in C_i(L) (which is possible since L is algebraically closed). Note then that there are isomorphisms of group schemes (C_i,e_i)\cong \mathrm{Jac}(C_i) (e.g. see [Silverman, Proposition 3.4 in §III.3]). Thus, if \mathrm{Jac}(C_1)\cong \mathrm{Jac}(C_2) then C_1\cong C_2. Thus, \mathrm{Jac} is injective. 

To see that the map is surjective we again note that for any elliptic curve (C,e) over L one has an isomorphism of group varieties \mathrm{Jac}(C)\cong (C,e) (again by [Silverman, Proposition 3.4 in §III.3]). \blacksquare

From this, we deduce the following:

Proposition 23: The composition

j\circ \mathrm{Jac}:\mathscr{M}_1\to\mathbb{A}^1_\mathbb{Q}

is a coarse moduli scheme.

Proof: To see that this map is initial, note that we have a natural map of moduli problems

\rho:\mathscr{M}_{1,1}\to \mathscr{M}_1

which just forgets about the marked point. Suppose that we have a map of moduli problems

p:\mathscr{M}_1\to X

for a \mathbb{Q}-scheme X. Note then that by Proposition 20 there exists a unique morphism f:\mathbb{A}^1_\mathbb{Q}\to X such that 

p\circ \rho=f\circ j

We then claim that 

p=f\circ j\circ \mathrm{Jac}

Indeed, since the map to X factors through the component presheaf it suffices to note that for any genus 1 curve C over T one has that 


But, let us note that since C\to T is smooth, there exists an étale cover T'\to T such that C_{T'}\to T' has a section. Note then that this implies that \mathrm{Jac}(C_{T'})=\mathrm{Jac}(C)_{T'} is isomorphic to C_{T'} as follows from [KM, Theorem 2.1.2]. So, then we see that p(C) has image in X(T') given by


since \mathrm{Jac}(C)_{T'} is an elliptic curve. But, this implies that p(C) and f(j(\mathrm{Jac}(C)) have the same image under the map X(T)\to X(T') which implies they’re equal since this map is injective since X is a sheaf.

The claim that j\circ \mathrm{Jac} induces a bijection on L-points for L an algebraically closed extension of \mathbb{Q} follows by combining Proposition 20 and Proposition 22. \blacksquare

Thus, combining everything above, we have the following corollary:

Corollary 24: Let C be a genus 1 curve over \overline{\mathbb{Q}} . Then, C is defined over \mathbb{Q} if and only if j(\mathrm{Jac}(C)) is rational.

Thus we have answered our generalized mai question for \mathscr{M}_1, and we will have answered our main question as soon as we exhibit a genus 1 curve over \overline{\mathbb{Q}} with irrational j-invariant. But, for example, follows from our discussion above we have that the genus 1 curve we know that the curve 

\displaystyle V\left(y^2z-x^3-\frac{36}{\sqrt{2}-1728}xz^2-\frac{1}{\sqrt{2}-1728}\right)\subseteq \mathbb{P}^2_{\overline{\mathbb{Q}}}

has j-invariant \sqrt{2} and thus is not defined over \mathbb{Q}.

The case of genus 2

We now move onto the case of genus 2 curves. This is, in some sense, simpler than the genus 1 case since there is a well-known explicit description of the coarse moduli scheme for \mathscr{M}_2 in the literature (whereas, as we remarked in the last section, there is none for \mathscr{M}_1 since it’s not an algebraic stack and thus not usually considered). But, it is also harder than the genus 1 case since the coarse moduli scheme is more complicated.

In particular, we start with the following:

Theorem 25 ([Igusa, Theorem 2]): There is a morphism

\mathscr{M}_2\to D(J_{10})\subseteq \mathrm{Proj}(\mathbb{Q}[J_2,J_4,J_6,J_8,J_{10}]/(J_4^2-J_2J_6+4J_8))

where J_i is given degree i. Moreover, this morphism realizes D(J_{10}) as the coarse moduli scheme of \mathscr{M}_2.

Let us explain in more detail what this coarse moduli space is. Namely, we are thinking of \mathbb{Q}[J_2,J_3,J_6,J_8,J_{10}] as a graded ring where each J_i is given degree 2i. With this definition one sees that J_4^2-J_2J_6+4J_8 is homogenous of degree 8 and thus we can form the quotient ring


which has an inherited grading. We can then talk about the proj construction of this graded ring. We then note that since J_{10} is homogenous of degree 10 and thus we can consider the non-vanishing locus D(J_{10}) in 


which is an affine \mathbb{Q}-variety with coordinate ring


(i.e. the degree 0 elements of the localization of \mathbb{Q}[J_2,J_4,J_6,J_8,J_{10}]/(J_4^2-J_2J_6+4J_8) at J_{10}). 

Let us denote this scheme which is the coarse moduli scheme by M_2 for notation convenience. The coordinates J_{2i} of M_2 are called the Igusa invariants. Let us note that, by definition, we can realize M_2 as a locally closed subscheme of the weighted projective space:


In particular, we see that we can describe M_2(L) as the set of equivalence classes of quintuples (J_2,J_4,J_6,J_8,J_{10}) such that

  1. Each J_{2i} is in L.
  2. One has that J_4^2-J_2 J_6+4J_8=0.
  3. One has that J_{10}\ne 0.

where (J_2,J_4,J_6,J_8,J_{10}) is equivalent to (J'_2,J'_4,J'_6,J'_8,J'_{10}) if there exists some \lambda in L^\times such that J'_{2i}=\lambda^{2i}J_{2i} for all i=2,4,6,8,10

Thus, since M_2 is evidently finite type over \mathbb{Q} and of positive dimension we know that the answer is negative for the generalized main question for \mathscr{M}_2

In fact, to find a genus 2 curve C over \overline{\mathbb{Q}} not defined over \mathbb{Q} we need only find a C such that quintuple


of Igusa invariants is not equivalent to rational qunituple. Unfortunately, the actual definitions of the Igusa invariants is pretty complicated (e.g. see [BSSVY, §4.1]). That said, there is an implimentation of the Igusa in Magma (e.g. see [Magma]) and using this one can show that if C is the normalization of

V(y^2z^4-x^6-x^5z-x^4z^2-x^3z^3-x^2z^4-xz^5-iz^6)\subseteq \mathbb{P}^2_{\overline{\mathbb{Q}}}


\displaystyle \begin{aligned} J_2(C) &=\frac{1}{4}(-120i+ 15)\\  J_4(C) &=\frac{1}{128}(-464i - 2709)\\ J_6(C) &=\frac{1}{1024}(1000i + 7741)\\ J_8(C) &= \frac{1}{65536}(-6169632i - 6178925)\\ J_{10}(C)&= \frac{1}{4096}(-948i + 33869)\end{aligned}

which one can quickly check is not equivalent to a rational quintuple. Thus, C is not defined over \mathbb{Q}.

The case of genus greater than 2

The case of g>2 is essentially the same as in the case g=2 but is much less explicit. Namely, we have the following well-known result:

Theorem 26: Let g\geqslant 2 be an integer. Then, \mathscr{M}_g has a coarse moduli space M_g which is a smooth affine \mathbb{Q}-scheme of dimension 3g-3.

Proof: See [MFK, Proposition 5.4]. \blacksquare

From this, we deduce from Proposition 18 that for each g\geqslant 2 there exists a genus g curve over \overline{\mathbb{Q}} which is not defined over \mathbb{Q}. Of course, it’s more difficult to give an explicit example in general since the explicit descriptions of M_g, and thus the \pi-rational objects in \mathscr{M}_g(\overline{\mathbb{Q}} are more difficult to describe. 

The case of genus 0

We end with the case of genus 0 curves. This case, for our particular generalized main question, is not so interesting. Namely, we recall the following basic theorem:

Proposition 27: Let k be a field and C a smooth proper geometrically connected curve over k of genus 0. If C(k)\ne\varnothing, then C\cong\mathbb{P}^1_k.

Proof: See [Liu, Proposition 4.1 in §7.4.1]. \blacksquare

From this we see that \pi_0(\mathscr{M}_0)(\overline{\mathbb{Q}}) is a singleton and thus the answer is trivially affirmative to the generalized main question for \mathscr{M}_0.

In particular, this would suggest that \mathscr{M}_0 would, if it has a coarse moduli space, have one of dimension 0. Indeed, one can in fact show that \mathscr{M}_0\to\mathrm{Spec}(\mathbb{Q}) is a coarse moduli space (e.g. see [Hartshorne, Proposition 25.1]).


In this final section we’d like to sew up the cases of the generalized main question for the examples we discussed above. Note that Example 2, Example 4, Example 7, Example 8, Example 10 are covered by the discussion above. Thus, it reall suffices to discuss, Example 9, Example 5, and .

Example 5

We would like to give at least one example of an affine variety over \overline{\mathbb{Q}} which has no model over \mathbb{Q}. There is no moduli theoretic technique I am aware of for the production of such an example (I’m not even aware any well-behaved moduli spaces of affine varieties with no extra structure!), so our construction will be more ad hoc.

The key fact is that there exists smooth proper geometrically connected curves X over \mathbb{Q} such that \mathrm{Aut}_{\overline{\mathbb{Q}}}(X_{\overline{\mathbb{Q}}}) is trivial. In fact, there exist explicit examples of such curves of genus g for each g\geqslant 3 (e.g. see [Poonen2]).

Remark 28: One can extend the construction below to the case of genus 2 curves with a little more work, using the objects constructed in [Poonen3]. We leave it to the reader to flesh this out.

Let X be such a curve and let x be an element of X(\overline{\mathbb{Q}}) not contained in X(\mathbb{Q}). Then, U_x:=X_{\overline{\mathbb{Q}}}-\{x\} is an affine open subscheme of X_{\overline{\mathbb{Q}}} (e.g. see [Youcis, Lemma 13] for a proof of this well-known result). 

We then have the following claim:

Proposition 29: The smooth integral affine curve U_x does not have a model over \mathbb{Q}.

Proof: Suppose that U_x has a model U_0 over \mathbb{Q}. Let X_0 be a smooth compactification of U_0. We claim that X is isomorphic to X_0. Indeed, since X is projective one knows that the set of isomorphism classes of varieties V over \mathbb{Q} such that V_{\overline{\mathbb{Q}}}\cong X_{\overline{\mathbb{Q}}} is classified by the cohomology set


(e.g. see [Poonen1, Theorem 4.5.2]). But, since \mathrm{Aut}_{\overline{\mathbb{Q}}}(X_{\overline{\mathbb{Q}}}) is trivial this implies that any such V must be isomorphic to X over \mathbb{Q}. Since (X_0)_{\overline{\mathbb{Q}}} is a compactification of (U_0)_{\overline{\mathbb{Q}}}\cong U_x we know that (X_0)_{\overline{\mathbb{Q}}}\cong X_{\overline{\mathbb{Q}}}. Thus, X_0\cong X.

Note then that X-U_0 becomes isomorphic to X_{\overline{\mathbb{Q}}}-U_x=\{x\} over \overline{\mathbb{Q}} and so X-U_0 consists of a degree 1 effective divisor or, equivalently, X-U_0=\{x_0\} for some x_0\in X(\mathbb{Q}). Note now that the isomorphism


extends to an isomorphism


such that x_0 maps to x. But, since \mathrm{Aut}_{\overline{\mathbb{Q}}}(X_{\overline{\mathbb{Q}}}) is trivial we know that f is the identity, and so x_0=x. This is a contradiction. \blacksquare

Thus, we see that the generalized main question has a negative answer for the moduli problem from Example 5.

Remark 30: The argument given above is somewhat persnickety  and, in particular, it’s not clear (to me at least) how to generalize it to produce examples in higher dimensions.

Example 6

Finding examples of presheaves that have an affirmative answer to the generalized main questions are trivial. For example, you can take any constant presheaf. Thus, one either has an affirmative or negative answer to the generalized main question for the moduli problem in Example 6 depending on the specifically chosen presheaf.

Example 9

It is not obvious but it is true that generalized main question has an affirmative answer for the moduli problem \mathsf{Red} from Example 9. The main ingredients are as follows. Every reductive group G over \overline{\mathbb{Q}} is split (in the sense of [Milne,  §21.a]). It is then a non-trivial theorem that G has a model over \mathbb{Q}. Essentially, one constructs from G and the choice of a maximal torus T of G a root datum (e.g. see [Milne,  §21.c]). One then uses the Existence Theorem (e.g. see [Milne, Theorem 23,55]) to say that there exists a split reductive group G_0 over \mathbb{Q} which has this associated root datum. One then uses the Isomorphism Theorem (e.g. see [Milne, Theorem 23.25]) to deduce that G\cong (G_0)_{\overline{\mathbb{Q}}}.

Interestingly though, the generalized main question for the larger moduli problem of all affine finite type group schemes \mathsf{LAG} has a negative answer. Indeed, by [Milne, Theorem 14.37(b)] it suffices to show that there exist nilpotent lie algebras \mathfrak{g} over \overline{\mathbb{Q}} which don’t have models over \mathfrak{g}. Examples of such objects were originally constructed by Malcev in [Malcev]. But, since this article seems difficult to obtain the reader may also see [Scheuneman, §3 Proposition 2].


[Bertin] Bertin, J., 2013. Algebraic stacks with a view toward moduli stacks of covers. In Arithmetic and Geometry Around Galois Theory (pp. 1-148). Birkhäuser, Basel.

[BSSVY] Booker, A.R., Sijsling, J., Sutherland, A.V., Voight, J. and Yasaki, D., 2016. A database of genus-2 curves over the rational numbers. LMS Journal of Computation and Mathematics19(A), pp.235-254.

[BLR] Bosch, S., Lütkebohmert, W. and Raynaud, M., 2012. Néron models (Vol. 21). Springer Science & Business Media.

[Conrad1] Conrad, B., 2014. Reductive group schemes. Autour des schémas en groupes1, pp.93-444.

[Conrad2] Conrad, B., 2005. Keel–Mori theorem via stacks. preprint. http://math.stanford.edu/~conrad/papers/coarsespace.pdf

[EGAIV Troisième partie] Grothendieck, Alexander. Éléments de géométrie algébrique : IV. Étude locale des schémas et des morphismes de schémas, Troisième partie. Publications Mathématiques de l’IHÉS, Volume 28 (1966) , pp. 5-255. http://www.numdam.org/item/PMIHES_1966__28__5_0/

[Hartshorne] Hartshorne, R., 2009. Deformation theory (Vol. 257). Springer Science & Business Media.

[Hida1] Hida, H., 2011. Geometric modular forms and elliptic curves. World Scientific.

[Hida2] Hida, H., 2012. p-adic automorphic forms on Shimura varieties. Springer Science & Business Media.

[Igusa] Igusa, J.I., 1960. Arithmetic variety of moduli for genus two. Annals of Mathematics, pp.612-649.

[LMB] Laumon, G. and Moret-Bailly, L., 2018. Champs algébriques (Vol. 39). Springer.

[KM] Katz, N.M. and Mazur, B., 1985. Arithmetic moduli of elliptic curves (No. 108). Princeton University Press.

[Magma] https://magma.maths.usyd.edu.au/magma/handbook/text/1523

[Malcev] Mal’cev, A.I., 1951. On a class of homogeneous spaces, Translations Amer. Math. Soc.(2)39, pp.276-307.

[McNinch] McNinch, G.J., 2010. Levi decompositions of a linear algebraic group. Transformation Groups15(4), pp.937-964.

[Milne] Milne, J.S., 2017. Algebraic groups: The theory of group schemes of finite type over a field (Vol. 170). Cambridge University Press.

[MFK] Mumford, D., Fogarty, J. and Kirwan, F., 1994. Geometric invariant theory (Vol. 34). Springer Science & Business Media.

[Olsson] Olsson, M., 2016. Algebraic spaces and stacks (Vol. 62). American Mathematical Soc..

[Poonen1] Poonen, B., 2017. Rational points on varieties (Vol. 186). American Mathematical Soc..

[Poonen2] Poonen, B., 2000. Varieties without extra automorphisms~ I: curves. Mathematical Research Letters7(1), pp.67-76.

[Poonen3] Poonen, B., 2000. Varieties without extra automorphisms II: Hyperelliptic curves. Mathematical Research Letters7(1), pp.77-82.

[Scheuneman] Scheuneman, J., 1967. Two-step nilpotent Lie algebras. Journal of Algebra7(2), pp.152-159.

[Stacks] Stacks Project Authors, 2020. Stacks project. https://stacks.math.columbia.edu/

[Youcis] Youcis, Alexander. CLASSIFYING ONE DIMENSIONAL GROUPS (II). https://ayoucis.wordpress.com/2019/11/19/classifying-one-dimensional-groups-ii/


  1. I think I object to your claim that your curve with $j$-invariant $\pi$ lies in $\PP^2_{\overline\QQ}$, unless of course you are (somewhat confusingly vis-a-vis the Lindemann-Weierstrass theorem) using $\pi$ to denote an algebraic number.

  2. Thanks for another enlightening post. I think there are some typos :
    1. Example 14 : “if G/N_G(T) were finite-dimensional it must be trivial” should be “if G/N_G(T) were zero-dimensional it must be trivial”
    2. The case of genus 2 : Here J_i should be of degree i instead of 2i.
    And in the end when you take C to be the normalization of
    the first term should be y^2z^4 I guess ?

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s