# Hilbert Modular Varieties - an p-adic Hilbert modular forms One can develop a theory of p-adic...

date post

07-Jul-2020Category

## Documents

view

0download

0

Embed Size (px)

### Transcript of Hilbert Modular Varieties - an p-adic Hilbert modular forms One can develop a theory of p-adic...

Hilbert Modular Varieties - an Introduction

Eyal Goren

McGill University

The Second Montreal-Toronto Workshop in Number Theory, Fields Institute, April 9-10, 2011

1 / 32

The Algebraic Group

L totally real, [L : Q] = g ,

B = Hom(L,R) = {σ1, . . . , σg}.

G = ResL/QGL2. For any Q-algebra R:

G(R) = GL2(L⊗Q R).

In particular,

G(Q) = GL2(L), G(R) = ∏ σ∈B

GL2(R).

Here, γ ∈ GL2(L) 7→ (σ(γ))σ∈B.

Similarly, we have G′ = ResL/QSL2.

2 / 32

Subgroups from lattices

Let a, b, be fractional ideals of L. Define the subgroup GL(a⊕ b) of G(Q) as the matrices{(

a b c d

) : a, d ,∈ OL, b ∈ a−1b, c ∈ ab−1, ad − bc ∈ O×L

} .

This group stabilizes the lattice a⊕ b in L⊕ L under right multiplication. Similarly, for G′.

Let S = ResC/RGm. We have

h0 : S→ G, x + iy 7→ ((

x −y y x

) , . . . ,

( x −y y x

)) .

The stabilizer of h0 under conjugation by G is the maximal compact group

K∞ = {((

xσ −yσ yσ xσ

)) σ

: x2σ + y 2 σ 6= 0,∀σ

} .

3 / 32

The symmetric space

G(R) ∼= ∏ σ∈BGL2(R) acts on the symmetric space (H±)g , where

H± = {z ∈ C : ±Im(z) > 0}. A component of which is

Hg = {(τ1, . . . , τg ) : Im(τi ) > 0}.

The action of G(Q) is by

γ ∗ (τ1, . . . , τg ) = (σ1(γ)τ1, . . . , σg (γ)τg )

(in each component it is the usual action by fractional linear transformations). Note:

StabG(R)(i , . . . , i) = K∞, G(R)/K∞ ∼= (H±)g .

We can view this space also as the space of conjugates of h0 under G(R); G is a reductive connected algebraic group over Q. This is Deligne’s perspective.

4 / 32

Adelic points

From Deligne’s perspective we care about

G(Q)\G(A)/K∞Kf ∼= G(Q)\(H±)g × G(Af )/Kf .

One can show:

G(Q)\G(A)/K∞Kf ∼= ∪mj=1Γj\Hg ,

where, for suitable gj ∈ G(Af ),

G(A) = ∪mj=1G(Q)gjG(R)0Kf , Γj = gjG(R)0Kf g−1j ∩ G(Q).

For example, for Kf = G(Ẑ), we get⋃ [a]∈CL+(L)

GL(OL ⊕ a)+\Hg .

5 / 32

Abelian varieties with RM

(A, ι, λ) � A g -dim’l AV over a base k (field, ring, scheme)

� ι : OL ↪→ Endk(A) ( ι∨ : OL ↪→ Endk(A∨))

� λ : A −→ A∨ an OL-equiv. polarization: λ ◦ ι(a) = ι∨(a) ◦ λ; equivalently, the Rosati involution acts trivially on OL.

The adelic description suggests a different definition. Fix a fractional ideal a of L. Look at

(A, ι, λ),

where λ : HomOL,k(A,A

∨)symm ∼=−→ a,

is an isomorphism of OL-modules with a notion of positivity (namely, identifies the cone of polarizations with a+). 6 / 32

Analytic families of abelian varieties with RM

Let a, b be fractional ideals of L. For z ∈ Hg embed a⊕ b in Cg as a lattice:

Λz = a · z + b · 1 = {(σi (a)zi + σi (b))i : a ∈ a, b ∈ b}.

A polarization on Az = Cg/Λz can be described by an alternating pairing on a⊕ b:

Er ((x1, y1), (x2, y2)) = TrL/Q(r(x1y2 − x2y1)), r ∈ (DL/Qab)−1.

Have

HomOL(Az ,A ∨ z )

symm = (DL/Qab)−1, {Polarizations} ↔ (DL/Qab)−1,+.

7 / 32

Analytic families (cont’d)

Theorem

SL(a⊕ b)\Hg parameterizes isomorphism classes of (A, ι, λ) with λ : HomOL(Az ,A

∨ z )

symm → (DL/Qab)−1 an isomorphism, taking the polarizations to the totally positive elements.

GL(a⊕ b)+\Hg parameterizes isomorphism classes of (A, ι) such that there exists an isomorphism λ : HomOL(Az ,A

∨ z )

symm → (DL/Qab)−1, taking the polarizations to the totally positive elements.

From now on

We consider moduli of (A, ι, λ): abelian varieties with RM and a principal OL-linear polarization λ. (Corresponds over C to SL(D−1L/Q ⊕OL).)

8 / 32

Cusps and the compact dual

The groups GL2(L) + acts on P1(L) and the orbits of GL(a⊕ b)+

(or SL(a⊕ b)) are in bijection with Cl(L). To a point (α : β) ∈ P1(L) we associate the ideal class (a, b) ·t (α, β) = αa⊕ βb.

The compact dual depends only of G′(R) = ∏ σ SL2(R). It is thus

equal to P1(C)g . The boundary of Hg is thus

∪gi=1P 1(C)i × P1(R)××P1(C)g−i−1,

but the rational boundary components for G′ are precisely

P1(L) ↪→ P1(Q)g .

The minimal (or Bailey-Borel-Satake) compactification of SL(a⊕ b)\Hg is, set-theoretically,

SL(a⊕ b)\ ( Hg ∪ P1(L)

) . 9 / 32

Factors of automorphy

For each k = (k1, . . . , kg ) ∈ Zg , γ ∈ SL2(L) and z = (z1, . . . , zg ) ∈ Hg ,

jk(γ, z) :=

g∏ i=1

j(σi (γ), zi ) ki .

For f : Hg → C holomorphic, let

f |kγ = jk(γ, z)−1f (γz).

Let Γ ⊂ SL2(L) be a congruence subgroup. We say f is a weight k modular form of level Γ if

f |kγ = f , ∀γ ∈ Γ.

(If g > 1 there is no need to require it is holomorphic at infinity (Koecher’s principle).)

10 / 32

Factors of automorphy (2)

The vector valued factor of automorphy

diag(j(σ1(γ), z1), . . . , j(σg (γ), zg )),

defines a vector bundle over Γ\Hg . It is easy to see from our construction of analytic families π : (Au, ιu, λu)→ Γ\Hg that it is the relative cotangent space at the identity (the Hodge bundle):

E = π∗(Ω1(Au ,ιu ,λu)→Γ\Hg ).

We have E = ⊕iLi ,

where Li is defined by the factor of automorphy jei (γ, z). Bailey-Borel proved that their compactification is an algebraic variety given as Proj(

∑∞ k=0 Γ(SL(a⊕ b)\Hg , det(E)k)).

11 / 32

Fourier expansions

The group SL(a⊕ b) contains the subgroup{( � b 0 �−1

) : � ∈ O×L , b ∈ a

−1b } .

A modular form f of level SL(a⊕ b) has then a Fourier expansion

f (z1, . . . , zg ) = ∑

ν∈(a−1b)∨ a(ν)qν ,

(a−1b)∨ = D−1L/Qab −1, qν = exp(2πi(σ1(ν)z1 + · · ·+ σg (ν)zg )).

• Holomorphic ⇒ ν � 0 or ν = 0. • Action of O×L ⇒

a(ν) = (∏

σi (�) −ki ) · a(�2ν), � ∈ O×L .

(So if k1 = · · · = kg and are even a(ν) = a(�2ν). Then, if the strict class number of L is 1, a(ν) depends only on the ideal ν.)

12 / 32

Moduli problems

Let n be an integral ideal of OL; (A, ι) an abelian variety with RM. A Γ(1n)-level structure is a closed immersion α : OL/n ↪→ A.

Consider the moduli of A = (A, ι, λ, α), a principally polarized abelian variety (A, λ) with RM ι, and a Γ(1n)-level structure α. If n is large enough, this is representable by a scheme M (1n) that is irreducible and smooth over Z[Norm(n)−1]. We have

M (1n)(C) ∼= Γ(1n)\Hg ,

where Γ(1n) are the matrices in SL(D−1L/Q ⊕OL) that are upper unipotent modulo n.

13 / 32

Modular forms over any base

Let π : Au = (Au, ιu, λu, αu)→ Spec(Z[Norm(n)−1]) be the universal object. Define the Hodge bundle

E = π∗(Ω1Au/M (1n)).

This is a rank g vector bundle over M (1n), which is a rank 1 OL ⊗OM (1n) vector bundle. And (after base change to OM [Nm(n)−1], M a normal closure of L in C)

E = ⊕gi=1Li ,

a sum of line bundles such that over C the line bundle Li is defined by the factor of automorphy jei (γ, z). A modular form of weight k = (k1, . . . , kg ) and level M (1n), defined over a base S (an OM [Nm(n)−1]-scheme), is an element in

Γ(M (1n)⊗OM [Nm(n)−1] S ,L k1 1 ⊗ · · · ⊗ L

kg g ).

14 / 32

Modular forms over any base (2)

The q-expansion can be defined in this generality. If S is the spectrum of a ring R then a modular form over S has q-expansion in

Z[[qν : ν ∈ (a−1b)∨]]⊗ R.

The converse is also true. Further, a modular form is zero if and only if its q-expansion is zero; the Galois action can be described via the action on the Fourier coefficients.

In particular, if two modular forms over R, of the same weight, have the same Fourier expansion then they are equal. In characteristic zero something stronger is true: if two modular forms over C have the same Fourier expansion then they are equal. This is because the Fourier expansion converges and describes the form on an open neighborhood of the cusp (and

not just in its completed local ring). This is not the case over an arbitrary ring and one arrives at the notion of congruence.

15 / 32

Another view on modular forms and weights

One can also think about a Hilbert modular form f defined over R of weight k and level Γ(1n) as a rule,

(A;ω) = (A, ι, λ, α;ω)/R1 7→ f (A, ω) ∈ R1,

for every R1-algebra (where ω is a generator over OL ⊗Z R1 of Lie(A)), that depends only on the isomorphism class of (A, ω), commutes with base-change, and satisfies

f (A, rω) = (χk11 · · ·χ kg g )(r) · f (A, rω), r ∈ (OL ⊗ R1)×

where χ1, . . . , χg are the characters of the torus ResOL/ZGm.

16 / 32

Congruences

Let p be a prime that is unramified in L (for simplicity). Then,

B = ∪p|pBp, Bp = Hom(L, Lp)

Recommended

*View more*