L Adic Talk

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Attaching `-adic Representations to Elliptic Modular Forms

Introduction
In his famous Bourbaki talk [2], Deligne described a recipe for attaching `-adic Galois rep-
resentations to elliptic modular forms of integral weight at least 2. As a consequence of the
method, one reduces the Ramanujan–Petersson conjecture to the validity of Weil’s Riemann
Hypothesis for varieties over finite fields. There seems to exist no brief and precise outline
of Deligne’s recipe in circulation, and this note is intended to close this gap in the literature.
A long and thorough explanation, with complete proofs, may be found in [1].
Fix a integers k ≥ 0 and N ≥ 1. Let f ∈ Sk+2 (Γ1 (N )) be a Hecke eigenform, with
eigenvalues Tp f = ap f for p - N , and Nebentypus character . Set Kf = Q({ap }), let ` be
a rational prime, and choose a prime λ of Kf lying over `. We will construct an continuous
homomorphism
ρf,λ : Gal(Q/Q) −→ GL2 (Kf,λ )
that is unramified at all p - `N , and for such p satisfies

Tr ρf,λ (ϕp ) = ap and det ρf,λ (ϕp ) = (p)pk+1 ,

where ϕp denotes any choice of Frobenius element at p.

Step 1: Good Models and Correspondences


Choose an integer n0 relatively prime to `N , so that N 0 = `N n0 ≥ 5, and put Γ = Γ1 (N 0 ).
Let T = Q[Tp ]p-N 0 be the Hecke algebra “away from N 0 ,” and let B = Spec Z[1/N 0 ].
Since we have required that N 0 ≥ 5, there exist moduli schemes over B for the moduli
problems “generalized elliptic curves with Γ-level structure” and “generalized elliptic curves
with Γ-level structure a chosen p-isogeny,” for p - N 0 . We denote these models with their
universal objects by
ϕ
E(Γ) E1 (Γ) −→ E2 (Γ)
f↓ & .
X(Γ) X(Γ; p) .
↓ ↓
B B
The schemes X(Γ) and X(Γ; p) are projective curves over B, and Igusa’s work implies that
X(Γ) is even smooth.
The associations which send the morphism E1 (Γ) → E2 (Γ) to E1 (Γ) and E2 (Γ) define by
functoriality, respectively, morphisms α, β : X(Γ; p) ⇒ X(Γ). One defines Tp to be the al-
gebraic correspondence “β∗ ϕ∗ α∗ ” over B. Similarly, the association (E(Γ), α) 7→ (E(Γ), pα),
where α denotes the level structure mapping and p - N 0 , defines by functoriality an auto-
morphism Ip : X(Γ) → X(Γ), and we let hpi be “Ip∗ .”
All the above work restricts well to the open subschemes Y (Γ) ⊂ X(Γ) and Y (Γ; p) ⊂
X(Γ; p) obtained by “removing the cusps,” including the correspondences. The correspon-
dences act on all cohomology of everything in sight.

1
Step 2: Étale Cohomology and the Local Theory
For any cohomology theory H, whose variant “with compact supports” is Hc , we write
He = image(Hc → H).
Let a : Y (Γ) → B denote the structure map. We define V` to be the étale `-adic sheaf
Rét a∗ (Symk Rét
e 1 1
f∗ Z` ) ⊗ Q` on B. If k = Q or Fp (with p - N 0 ), and k denotes an algebraic clo-
sure of k, then the stalk V`,k of V` over the geometric point Spec k → B is a finite-dimensional
Q` -vector space equipped with a continuous linear Gal(k/k)-action. The smoothness of our
model as a scheme implies that V` is a smooth `-adic sheaf, and so considering it as an étale
local system yields an identification
   
its Galois ∼ its Galois action restricted to
V`,Fp , = V`,Q , .
action a decomposition group at p
In particular, the action of a decomposition group at p on V`,Q factors through Gal(Fp /Fp ),
and hence the representation is unramified at primes p - N 0 .
Remark. This description of a Frobenius at p acting on V`,Q in terms of “mod p” geometry
is precisely what allows one to relate the Ramanujan–Petersson conjecture to Weil’s Riemann
hypothesis. Deligne’s “Weil II” result makes the deduction very easy, so we do not pursue
this topic any further.
For our purposes, the upshot of the above isomorphism is that, being “deduced from
geometry,” it is Tp -equivariant, and hence we may calculate the effect of Tp on V`,Q via its
effect on V`,Fp .
To this end, for a family Y of elliptic curves in characteristic p, write Y nss for its non-
supersingular locus, and consider an old theorem which asserts that
Y (Γ; p)nss ∼ nss nss
Fp = Y (Γ)Fp q Y (Γ)Fp .

Recalling that ϕ denotes the universal p-isogeny over Y (Γ; p), the first component on the
right corresponds to the locus where ker ϕ = ker Fp , Fp denoting the geometric Frobenius
morphism, and the second component is the locus where ker ϕ = ker Fpt , where t denotes
transpose under duality of abelian varieties.
Using this description, one can explicitly calculate the action of the correspondence Tp
on V`,Fp as
Tp = Fp + hpiFpt and Fp Fpt = Fpt Fp = pk+1 ,
or, written a little differently,
1 − Tp X + hpipk+1 X 2 = (1 − Fp X)(1 − hpiFpt X).
These statements are called the “congruence formulas” or the “Eichler–Shimura relations.”
Finally, general étale nonsense identifies the action of Fp on `-adic cohomology with the
action of the inverse ϕ−1p of the arithmetic Frobenius element.

Step 3: Compare to Some Easier Analytic Things


First note that Artin’s comparison theorem allows us to write
V`,Q = VQ ⊗ Q` , with e 1 (Y (Γ)an , Symk R1 f∗ Q).
VQ = H

2
The Eichler–Shimura isomorphism will give us a nicer form for the complexification of VQ .
Write for brevity Ω1 = Ω1X(Γ)an , and ω = f∗ Ω1E(Γ)an /X(Γ)an , and use the symbols also for the
restrictions of these sheaves to Y (Γ)an , when no confusion can arise. So ω has stalks ωx =
H 0 (Exan , Ω1Exan ). The de Rham connecting homomorphisms H 0 (Ex , Ω1Exan ) → H 1 (Ex , Q) ⊗ C
glue to a sheaf map ω → (R1 f∗ Q) ⊗ C. This gives a map ω k → Symk ω → Symk R1 f∗ Q ⊗ C.
Taking cohomology of this gives a map
e 1 (Y (Γ)an , Symk R1 f∗ Q) ⊗ C = VQ ⊗ C.
e 1 (Y (Γ)an , ω k ) −→ H
A: H

The cohomology of the de Rham complex, tensored first with ω k , has as its connecting
homomorphism
B: He 0 (Y (Γ)an , Ω1 ⊗ ω k ) −→ H
e 1 (Y (Γ)an , ω k ).
Finally, the inclusion Y (Γ)an ,→ X(Γ)an gives, by pullback, a map

C : H 0 (X(Γ)an , Ω1 ⊗ ω k ) −→ H
e 0 (Y (Γ)an , Ω1 ⊗ ω k ).

Putting these all together, we get a map (labeled KS for “Kodaira–Spencer”)

KS = A ◦ B ◦ C : H 0 (X(Γ)an , Ω1 ⊗ ω k ) −→ VQ ⊗ C.

Complex conjugation acts on the target of KS, so it makes sense to define KS to be the
complex conjugate map. Putting KS and KS together, a theorem of Eichler and Shimura
says that we get a Hecke-equivariant isomorphism

KS ⊕ KS : H 0 (X(Γ)an , Ω1 ⊗ ω k ) ⊕ H 0 (X(Γ)an , Ω1 ⊗ ω k ) −→ VQ ⊗ C.

Now one brings explicit “upper half-plane modulo a congruence subgroup” descriptions
to bear on our situation. Using this elementary theory, one can work out that there is a

natural inclusion of bundles Ω1 ,→ ω 2 over X(Γ)an , which satisfies Ω1 |Y (Γ)an → ω 2 |Y (Γ)an .
The sections of Ω1 near the cusps are precisely those sections of ω 2 that vanish at the cusps.
Thus Ω1 ⊗ ω k is the sheaf of “holomorphic things of weight 2 + k that vanish at the cusps.”
In fact, we get an isomorphism of T -modules,

H 0 (X(Γ)an , Ω1 ⊗ ω k ) ∼
= Sk+2 (Γ).

Step 5: Cut Out Our Representation


To recap, to get between Sk+2 (Γ) and V`,Q , in a T -equivariant manner, one traces through
the following diagram.

VQ
⊗C . & ⊗Q`
Sk+2 (Γ) ⊕ Sk+2 (Γ) V`,Q

Let If be the annihilator ideal of T acting on f , so that



T /If −→ Kf via Tp 7−→ ap .

3
Q
Basic algebraic number theory tells us that (T /If )⊗Q` = Kf ⊗Q` = λ|` Kf,λ , the direct sum
of the completions of Kf at the places lying over `. This decomposition can be represented
by idempotents in T ⊗ Q` , allowing us to view the Kf,λ as T -algebras and the decomposition
as being T -equivariant.
We start with the T -module VQ , and we have two operations. Tensoring over Q with Q`
endows it with a continuous linear Gal(Q/Q)-action (by comparison with V`,Q ). Tensoring
over T with Kf = T /If , one cuts out the part corresponding to f . Performing both these
operations, and projecting onto “the λ-part of Kf ⊗ Q` ,” we define

Vf,λ = Kf,λ ⊗T VQ = Kf,λ ⊗T ⊗Q` V`,Q .

Fixing a basis of Vf,λ , and noting that V`,Q passes on to Vf,λ a continuous, linear Gal(Q/Q)-
action, we get our desired representation

ρf,λ : Gal(Q/Q) −→ GLn (Kf,λ ).

Now we explain some details.


First, we point out that n = 2. This is because all the operations going between
Sk+2 (Γ) and V`,Q preserve the corresponding dimensions and are T -equivariant, except for
the Eichler–Shimura isomorphism, which doubles dimension. Since Cf has C-dimension one,
Vf,λ has Kf,λ -dimension two.
Next, the representation ρf,λ is unramified at all p - `N , not just at all p - N 0 . To
see this, choose an n1 relatively prime to N 0 , such that N10 = `N n1 ≥ 5. Construct the
representation ρ0f,λ accordingly, and note that it is unramified away from N10 . One can embed
0 00
the respective V`,Q and V`,Q into the V`,Q corresponding to N20 = `N n0 n1 = lcm(N 0 , N10 ), and
recover them as the submodules fixed under parts of an action of SL2 (Z/N20 ). One finds
that the pieces which yield ρf,λ and ρ0f,λ are identical under these embeddings, giving a
canonical isomorphism between them. Thus they are simultaneously unramified away from
gcd(N 0 , N10 ) = `N .
Finally, one ought to check that Tr ρf,λ (ϕp ) = ap and det ρf,λ (ϕp ) = (p)pk+1 . This follows
from a fairly direct computation involving the congruence formulas, and properties of the
Poincaré duality pairing.

References
[1] B. Conrad, Modular Forms, Cohomology, and the Ramanujan Conjecture. In preparation,
http://www.math.lsa.umich.edu/~bdconrad/bc.ps.

[2] P. Deligne, “Formes modulaires et représentations `-adiques,” Sém. Bourbaki 1968/1969,


exp. n◦ 355.

You might also like