Hecke Operators
5.1 The double coset operator
5
Γ1αΓ2={γ1αγ2:γ1∈Γ1, γ2∈Γ2}
is adouble cosetin GL+2(Q). Under a definition to be developed in this section, such double cosets transform modular forms with respect toΓ1into modular forms with respect toΓ2.
The group Γ1 acts on the double cosetΓ1αΓ2by left multiplication, par- titioning it into orbits. A typical orbit isΓ1β with representativeβ=γ1αγ2, and the orbit spaceΓ1\Γ1αΓ2is thus a disjoint union
Γ1βj for some choice of representativesβj. The next two lemmas combine to show that this union is finite.
Lemma 5.1.1.Let Γ be a congruence subgroup of SL2(Z) and let α be an element ofGL+2(Q). Thenα−1Γ α∩ SL2(Z)is again a congruence subgroup ofSL2(Z).
Proof. There exists ˜N ∈ Z+ satisfying the conditions Γ( ˜N) ⊂ Γ, ˜N α ∈ M2(Z), ˜N α−1∈M2(Z). SetN = ˜N3. The calculation
αΓ(N)α−1⊂α(I+ ˜N3M2(Z))α−1
=I+ ˜N·N α˜ ·M2(Z)·N α˜ −1⊂I+ ˜NM2(Z) and the observation thatαΓ(N)α−1consists of determinant-1 matrices com- bine to show thatαΓ(N)α−1⊂Γ( ˜N). Thus Γ(N)⊂α−1Γ( ˜N)α⊂α−1Γ α, and intersecting with SL2(Z) completes the proof.
Lemma 5.1.2.Let Γ1 andΓ2 be congruence subgroups ofSL2(Z), and let α be an element ofGL+2(Q). Set Γ3 =α−1Γ1α ∩ Γ2, a subgroup of Γ2. Then left multiplication byα,
Γ2−→Γ1αΓ2 given by γ2→αγ2,
induces a natural bijection from the coset space Γ3\Γ2 to the orbit space Γ1\Γ1αΓ2. In concrete terms,{γ2,j}is a set of coset representatives forΓ3\Γ2 if and only if{βj}={αγ2,j} is a set of orbit representatives for Γ1\Γ1αΓ2. Proof. The mapΓ2−→Γ1\Γ1αΓ2takingγ2toΓ1αγ2clearly surjects. It takes elementsγ2, γ2 to the same orbit whenΓ1αγ2=Γ1αγ2, i.e.,γ2γ2−1∈α−1Γ1α, and of courseγ2γ−21∈Γ2 as well. So the definitionΓ3 =α−1Γ1α ∩ Γ2 gives a bijection Γ3\Γ2 −→ Γ1\Γ1αΓ2 from cosets Γ3γ2 to orbits Γ1αγ2. The last
statement of the lemma follows immediately.
Any two congruence subgroupsG1 andG2 of SL2(Z) arecommensurable, meaning that the indices [G1:G1∩G2] and [G2:G1∩G2] are finite (Exer- cise 5.1.2). In particular, sinceα−1Γ1α ∩ SL2(Z) is a congruence subgroup of SL2(Z) by Lemma 5.1.1, the coset space Γ3\Γ2 in Lemma 5.1.2 is finite and hence so is the orbit spaceΓ1\Γ1αΓ2. With finiteness of the orbit space established, the double cosetΓ1αΓ2can act on modular forms. Recall that for
β∈GL+2(Q) andk∈Z, theweight-k β operatoron functionsf :H −→C is given by
(f[β]k)(τ) = (detβ)k−1j(β, τ)−kf(β(τ)), τ ∈ H.
Definition 5.1.3.For congruence subgroups Γ1 and Γ2 of SL2(Z) and α ∈ GL+2(Q), the weight-k Γ1αΓ2 operator takes functions f ∈ Mk(Γ1)to
f[Γ1αΓ2]k=
j
f[βj]k where {βj} are orbit representatives, i.e., Γ1αΓ2 =
jΓ1βj is a disjoint union.
The double coset operator is well defined, i.e., it is independent of how theβj are chosen (Exercise 5.1.3). Seeing that it takes modular forms with respect toΓ1 to modular forms with respect toΓ2,
[Γ1αΓ2]k:Mk(Γ1)−→ Mk(Γ2),
means showing that for eachf ∈ Mk(Γ1), the transformedf[Γ1αΓ2]k is Γ2- invariant and is holomorphic at the cusps. Seeing that the double coset oper- ator takes cusp forms to cusp forms,
[Γ1αΓ2]k :Sk(Γ1)−→ Sk(Γ2),
means showing that for eachf ∈ Sk(Γ1), the transformedf[Γ1αΓ2]k vanishes at the cusps.
To show invariance, note that any γ2 ∈ Γ2 permutes the orbit space Γ1\Γ1αΓ2 by right multiplication. That is, the map γ2 : Γ1\Γ1αΓ2 −→
Γ1\Γ1αΓ2 given by Γ1β → Γ1βγ2 is well defined and bijective. So if {βj} is a set of orbit representatives for Γ1\Γ1αΓ2 then {βjγ2} is a set of orbit representatives as well. Thus
(f[Γ1αΓ2]k)[γ2]k =
j
f[βjγ2]k =f[Γ1αΓ2]k, andf[Γ1αΓ2]k is weight-kinvariant underΓ2as claimed.
To show holomorphy at the cusps, first note that for any f ∈ Mk(Γ1) and for anyγ ∈GL+2(Q), the functiong =f[γ]k is holomorphic at infinity, meaning it has a Fourier expansion
g(τ) =
n≥0
an(g)e2πinτ /h
for some period h ∈ Z+ (this was Exercise 1.2.11(b)). Second, note that if functionsg1, ..., gd :H −→C are holomorphic at infinity, meaning that each gj has a Fourier expansion
gj(τ) =
n≥0
an(gj)e2πinτ /hj,
then so is their sumg1+· · ·+gd (Exercise 5.1.4). For any δ∈ SL2(Z), the function (f[Γ1αΓ2]k)[δ]k is a sum of functions gj =f[γj]k with γj =βjδ ∈ GL+2(Q), so it is holomorphic at infinity by the two facts just noted. Since δ is arbitrary this is the condition for holomorphy at the cusps.
For anyf ∈ Sk(Γ1) and for anyγ∈GL+2(Q), the functiong=f[γ]k van- ishes at infinity (this was also Exercise 1.2.11(b)), and the previous paragraph now shows thatf[Γ1αΓ2]k ∈ Sk(Γ2), i.e., the double coset operator takes cusp forms to cusp forms as claimed.
Special cases of the double coset operator [Γ1αΓ2]k arise when
(1) Γ1⊃Γ2. Takingα=Imakes the double coset operator bef[Γ1αΓ2]k =f, the natural inclusion of the subspaceMk(Γ1) inMk(Γ2), an injection.
(2) α−1Γ1α=Γ2. Here the double coset operator is f[Γ1αΓ2]k =f[α]k, the natural translation from Mk(Γ1) toMk(Γ2), an isomorphism.
(3) Γ1⊂Γ2. Takingα=Iand letting{γ2,j}be a set of coset representatives for Γ1\Γ2 makes the double coset operator bef[Γ1αΓ2]k =
jf[γ2,j]k, the natural trace map that projects Mk(Γ1) onto its subspace Mk(Γ2) by symmetrizing over the quotient, a surjection.
In fact, any double coset operator is a composition of these. Given Γ1, Γ2, andα, setΓ3=α−1Γ1α∩Γ2as usual and setΓ3 =αΓ3α−1=Γ1 ∩αΓ2α−1. Then Γ1 ⊃ Γ3 and α−1Γ3α =Γ3 and Γ3 ⊂Γ2, giving the three cases. The corresponding composition of double coset operators is
f →f →f[α]k →
j
f[αγ2,j]k, which by Lemma 5.1.2 is the general [Γ1αΓ2]k.
The process of transferring functions forward from Mk(Γ1) to Mk(Γ2) by the double coset operator also has a geometric interpretation in terms of transferring points back between the corresponding modular curves. This leads to an algebraic interpretation of the double coset as a homomorphism of divisor groups. Recall that every congruence subgroupΓ has a modular curve X(Γ) =Γ\H∗consisting of orbitsΓ τ. The configuration of groups is
Γ3 ∼ //
Γ3
Γ2 Γ1
where the group isomorphism is γ → αγα−1 and the vertical arrows are inclusions. The corresponding configuration of modular curves is
X3 ∼ //
π2
X3
π1
X2 X1
(5.1)
where the modular curve isomorphism is Γ3τ → Γ3α(τ), denoted α (Exer- cise 5.1.5). Again lettingΓ3\Γ2=
jΓ3γ2,j andβj=αγ2,j for eachjso that Γ1αΓ2 =
jΓ1βj, each point ofX2 is taken back by π1◦α◦π2−1 to a set of points ofX1,
{Γ3γ2,j(τ)} α //{Γ3βj(τ)}
π1
Γ_2τ
π−21
OO
{Γ1βj(τ)}.
Here π2−1 takes each point x ∈ X2 to the multiset (meaning elements can repeat) of overlying points y ∈ X3 each with multiplicity according to its ramification degree, π−21(x) = {ey·y : y ∈ X3, π2(y) = x}. To place the composition in the right environment for counting with multiplicity let Div(X) denote the divisor group of any modular curveX, where as in Chapter 3 the divisor group of a set is the free Abelian group on its points. In terms of divisors the composition is
[Γ1αΓ2]k :X2−→Div(X1), Γ2τ→
j
Γ1βj(τ),
and this has a uniqueZ-linear extension to a divisor group homomorphism, [Γ1αΓ2]k : Div(X2)−→Div(X1).
In this context the special casesΓ1 ⊃Γ2, α−1Γ1α= Γ2, and Γ1 ⊂Γ2 from before lead respectively to a surjection, an isomorphism, and an injection (Exercise 5.1.6).
This chapter will focus on double coset operators acting on modular forms, but the emphasis later in the book will move toward the divisor group inter- pretation.
Exercises
5.1.1.In Lemma 5.1.1, needα−1Γ αlie in SL2(Z)?
5.1.2.Show that for congruence subgroupsG1andG2of SL2(Z), the indices [G1:G1∩G2] and [G2:G1∩G2] are finite. (A hint for this exercise is at the end of the book.)
5.1.3.Show that the weight-kdouble coset operator [Γ1αΓ2]k is independent of how the orbit representativesβj are chosen. (A hint for this exercise is at the end of the book.)
5.1.4.Show that ifg1, ..., gd:H −→C are holomorphic at infinity, then so is their sumg1+· · ·+gd. (A hint for this exercise is at the end of the book.) 5.1.5.Check that the map α : X3 −→ X3 given by Γ3τ → Γ3α(τ) is well defined.
5.1.6.Show that in the special casesΓ1⊃Γ2,α−1Γ1α=Γ2, andΓ1⊂Γ2the divisor group interpretation of [Γ1αΓ2]k gives a surjection, an isomorphism, and an injection.