MathOverflow will be down for maintenance for approximately 3 hours, starting Monday evening (06/24/2013) at approximately 9:00 PM Eastern time (UTC-4).

## Modular Curves as Moduli Spaces of Elliptic Curves

Hi,

Is the modular curve defined as the quotient of the upper half-plane by an arithmetic group $\Gamma$ always a moduli space of elliptic curves with extra structure? I know this is true for $\Gamma_0(N), \Gamma_1(N), \Gamma(N)$, but I'm interested in some of other groups, particularly those of the form $\Gamma_0(N|d)$ (notation from Conway and Norton). Are these curves moduli spaces, and if so for what classes of varieties?

Thanks!

-
It depends what you mean by "arithmetic group". From the question it appears that you mean "congruence subgroup of $PSL_2(\mathbb{Z})$" in which case the answer is "yes". If you make precise your reference to Conway and Norton you may get a more precise answer for the subgroups that interest you. – Tony Scholl Oct 21 2010 at 9:11
It also depends on what you mean by "moduli space". What is it you want to do with this concept in mathematical arguments? – BCnrd Oct 21 2010 at 14:15
@Tony, yes I am thinking of congruence subgroups of $PSL_2(\mathbb{Z})$. The specific groups I'm interested in are defined on p. 312 of "Monstrous Moonshine": Let $N = nh$ with $h | 24$ and $h^2 | N$. Then $\Gamma_0(n|h) = \{ A = \left( \begin{array}{cc} a & b/h \\ cn & d \end{array} \right) \: | \: \det A = 1, \: a,b,c,d \in \mathbb{Z} \}$ @BCnrd, a coarse moduli space would be enough for me. What am I using this for? I'm looking at a genus 0 group of this type, and want to relate the Schwartzian derivative its hauptmodul satisfies to a Picard-Fuchs equation. – unknown (google) Oct 21 2010 at 14:35
Dear unknown (google): what I meant by my question is to ask in what sense the concept of "moduli space" is logically relevant in your arguments. Many explicit analytic calculations with low genus modular curves don't make rigorous use of "moduli space" concepts (though it may be reassuring to see that one's constructions are fitting into some broader framework), so from a logical point of view the notion of "moduli space" (coarse or otherwise) doesn't wind up being relevant. Or do you expect that speaking in terms of structures on elliptic curves may simplify some things you are doing? – BCnrd Oct 21 2010 at 16:03
@unknown (google): $\Gamma_0(n|h)$ is conjugate in $PGL_2(\mathbb{Q})$ to $\Gamma_0(n/h)$ so I think you know the answer. – Tony Scholl Oct 21 2010 at 18:30

For the groups $\Gamma$ in Conway-Norton, there is always a moduli problem of $\Gamma$-structures, but since the groups always contain $\Gamma_0(N)$ for some $N$, you won't be able to construct a universal family (because there is a $-1$ automorphism in the way). However, you will sometimes get a relatively representable'' problem in the sense of Katz-Mazur.

The upper half-plane quotients will be coarse spaces parametrizing objects of the following general form: You have a diagram of elliptic curves, with some isogenies of specified degrees between them, together with some data that tell you how much symmetry in the diagram you should remember. Since all of the groups normalize $\Gamma_0(N)$ for some $N$, the diagrams will typically involve cyclic isogenies of degree $N$ in some way, and the symmetrization will involve a subgroup of the finite quotient $N_{SL_2(\mathbb{R})}(\Gamma_0(N))/\Gamma_0(N)$.

The standard example is $\Gamma_0(p)^+$ for a prime $p$, which is generated by $\Gamma_0(p)$ as an index two subgroup, together with the Fricke involution $\tau \mapsto \frac{-1}{p\tau}$. The $\Gamma_0(p)$ quotient parametrizes diagrams $E \to E'$ of elliptic curves equipped with a degree $p$ isogeny between them. Taking the quotient of the moduli problem by the Fricke involution amounts to symmetrizing the diagram, so the $\Gamma_0(p)^+$ quotient parametrizes tuples $( \{ E_1, E_2 \}, E_1 \leftrightarrows E_2)$ of unordered pairs of elliptic curves, with dual degree $p$ isogenies between them. Equivalently, you can ask for a set of diagrams $\{E_1 \to E_2, E_2 \to E_1 \}$ where the maps are dual isogenies.

A less well-known example is the 3C group, which is an index 3 subgroup of $\Gamma_0(3|3)$, with Hauptmodul $\sqrt[3]{j(3\tau)} = q^{-1} + 248q^2 + 4124q^5 + \dots$. This group is labeled $\Gamma_0(3|3)$ in the Conway-Norton paper, because $\Gamma_0(3|3)$ is the eigengroup, namely the group that takes the Hauptmodul to constant multiples of itself. The 3C group contains $\Gamma_0(9)$ as a normal subgroup, with quotient $\mathbb{Z}/2\mathbb{Z} \times \mathbb{Z}/2\mathbb{Z}$. You can view the upper half-plane quotient as a parameter space of quadruples of elliptic curves, with a rather complicated system of cyclic 9-isogenies and correspondences that get symmetrized (more on this in the last paragraph). A more succinct expression follows from using the matrix $\binom{30}{01}$ to conjugate $\Gamma_0(3|3)$ to $\Gamma(1)$ and $\Gamma_0(9)$ to $\pm \Gamma(3)$. Then you're basically looking at a moduli problem that parametrizes elliptic curves $E$ equipped with an unordered octuple of symplectic isomorphisms $E[3] \cong (\mathbb{Z}/3\mathbb{Z})^2$ that form a torsor under the characteristic 2-Sylow subgroup $Q_8 \subset Sp_2(\mathbb{F}_3) \cong SL_2(\mathbb{Z})/\Gamma(3)$.

In general, you can encode moduli problems attached to arithmetic groups using the fact that congruence groups like $\Gamma(N)$ and $\Gamma_0(N)$ stabilize distinguished finite subcomplexes of the product of all $p$-adic Bruhat-Tits trees. Conway gives a explanation (that doesn't use the word "moduli") with pictures in his paper Understanding groups like $\Gamma_0(N)$. For example, when $N$ is a product of $k$ distinct primes, $\Gamma_0(N)$ stabilizes a $k$-cube. Given a finite stable subcomplex, there is a standard way to make a moduli problem out of it by assigning elliptic curves to the vertices, isogenies to the edges, such that the induced transformations on the Tate module behave as you would expect from traversing the product of buildings. To symmetrize, just enumerate orbits of the transformations you want, and demand a torsor structure.

In the case of the 3C group in the above paragraph, $\Gamma_0(9)$ pointwise stabilizes a subgraph of the 3-adic tree that is an X-shaped configuration spanned by 5 vertices. The edges coming out of the central vertex are in noncanonical bijection with points in $\mathbb{P}^1(\mathbb{F}_3)$, and to symmetrize, you can make an unordered 4-tuple of diagrams of 5 elliptic curves, related by the action of the subgroup $V_4 \subset PSL_2(\mathbb{F}_3)$ that preserves the cross-ratio.

-