Dimension Formulas
3.3 Meromorphic differentials
nonzero elements for all even positive integers k. (A hint for this exercise is at the end of the book.)
3.2.4.Show that for any k ∈ Z, if f is a nonzero element of Ak(Γ) then Ak(Γ) =C(X(Γ))f.
3.2.5.Letα=1−1
2−1
and letΓ =α−1Γ1(4)α. Show thatΓ∞=−[1 10 1]. (A hint for this exercise is at the end of the book.)
3.2.6.Verify the expression forg[S]24 in the proof of Proposition 3.2.2.
In other words, taking the pullback means changing variables in the meromor- phic differential, doing so compatibly with the change of variable formula for integrals. The pullback is obviously linear. The pullback is alsocontravariant, meaning that if ϕ1 =ϕ is as before and ϕ2 :V2 −→ V3 is also holomorphic then (ϕ2◦ϕ1)∗=ϕ∗1◦ϕ∗2 (Exercise 3.3.1(a)). IfV1⊂V2 andι:V1 −→V2 is inclusion then its pullback is restriction,ι∗(ω) =ω|V1 forω ∈Ω⊗n(V2) (Ex- ercise 3.3.1(b)). It follows that ifϕis a bijection, making its inverse holomor- phic as well by complex analysis, then (ϕ∗)−1 = (ϕ−1)∗ (Exercise 3.3.1(c)).
Ifπ: V1−→ V2 is a holomorphic surjection of open sets inC then π∗ is an injection (Exercise 3.3.1(d)).
Now we can piece together local differentials on a Riemann surfaceX. Let X have coordinate charts ϕj :Uj −→ Vj, wherej runs through some index set J, eachUj is a neighborhood in X, and eachVj is an open set in C. A meromorphic differential onX of degreenis a collection of local meromorphic differentials of degreen,
(ωj)j∈J∈
j∈J
Ω⊗n(Vj),
that iscompatible. To define this, letVj,k=ϕj(Uj∩Uk) andVk,j =ϕk(Uj∩ Uk) for j, k ∈ J. Then the compatibility criterion is that pulling back any transition map
Vj,k ϕk,j //Vk,j
to get the corresponding map of local meromorphic differentials Ω⊗n(Vj,k) Ω⊗n(Vk,j)
ϕ∗k,j
oo
gives
ϕ∗k,j(ωk|Vk,j) =ωj|Vj,k.
The meromorphic differentials of degreenonX are denotedΩ⊗n(X). Again this set forms a complex vector space and the sumΩ(X) =
n∈NΩ⊗n(X) forms a ring.
For example, whenX is a complex elliptic curveC/Λthe coordinate maps ϕj : Uj −→ Vj are homeomorphic local inverses to natural projection π : C−→X. Letω= (dzj)j∈J. Thenω satisfies the compatibility criterion since the transition maps between coordinate patches take the formz →z+λfor λ∈Λ, and these pull dzk back to dzj. Thus the differentialdz makes sense globally on a complex torus even though a holomorphic variablezdoes not.
Returning to the congruence subgroup Γ of SL2(Z), we now map each meromorphic differentialω onX(Γ) to a meromorphic differentialf(τ)(dτ)n onH, theΓ-invariant object onHmentioned at the beginning of the section.
The map is the pullback of the natural projectionπ:H −→X(Γ),
π∗:Ω⊗n(X(Γ))−→Ω⊗n(H).
To define this, recall from Chapter 2 that a collection of coordinate neigh- borhoods on X(Γ) is {π(Uj) : j ∈ J} where each Uj ⊂ H∗ is a neighbor- hood of a point τj ∈ H or of a cusp sj ∈ Q∪ {∞}; the local coordinate map ϕj : π(Uj) −→ Vj is characterized by the relation ψj = ϕj ◦π where ψj : Uj −→ Vj mimics the identifying action of π but maps into C. Now letω = (ωj)j∈J be a meromorphic differential on X(Γ). For eachj ∈J set Uj =Uj∩ H and Vj =ψj(Uj) and ωj =ωj|Vj. Then the pullback π∗(ω) is defined locally onHas
π∗(ω)|Uj =ψj∗(ωj) for allj. (3.4) To see that (3.4) gives a well defined global meromorphic differentialπ∗(ω) = f(τ)(dτ)n onH, consider the commutative diagram
Uj∩Uk
ψj
yytttttttttt
π
ψk
%%K
KK KK KK KK K Vj,k
ϕ−j1
//π(Uj∩Uk) ϕ
k //Vk,j.
(3.5)
The transition mapϕk,j =ϕkϕ−j1|Vj,k satisfiesϕk,j◦ψj=ψk onUj∩Uk, and pulling back givesψ∗k=ψ∗j◦ϕ∗k,j onVk,j. Therefore, lettingVj,k =ψj(Uj∩Uk) andVk,j =ψk(Uj∩Uk),
ψ∗k(ωk|Vk,j ) =ψ∗j(ϕ∗k,j(ωk|Vk,j )) =ψ∗j(ωj|Vj,k ) by compatibility, so the overlapping local pullbacks have a common valuef(τ)(dτ)n|Uj∩Uk and the global pullbackπ∗(ω) =f(τ)(dτ)n is well defined.
Since the pullback comes from an object on the quotientX(Γ) it must be Γ-invariant. That is, for anyγ∈Γ,
f(τ)(dτ)n =γ∗(f(τ)(dτ)n) = (f(γ(τ)))(γ(τ))n(dτ)n
=j(γ, τ)−2nf(γ(τ))(dτ)n= (f[γ]2n)(τ)(dτ)n. Thus the meromorphic functionf defining the pullback is weakly modular of weight 2n. This is the motivation promised back in Chapter 1 for the definition of weak modularity, at least for even weights.
The function f also satisfies condition (3) in Definition 3.2.1, thatf[α]2n is meromorphic at∞ for any α ∈ SL2(Z). To see this, let s = α(∞). The local mapψ :U −→V abouts takes the formψ =ρ◦δ :τ →z →q with δ=α−1andρ(z) =e2πiz/hwherehis the width ofs. Sinceωis meromorphic at the cusps of X(Γ), the local differential ω|V takes the form g(q)(dq)n whereg is meromorphic at 0. The functionf onU− {s}comes from pulling
backω|V−{0} under ψ to f(τ)(dτ)n. Computing this gives f = ˜f[δ]2n where (Exercise 3.3.2(a))
f˜(z) = (2πi/h)nqng(q), q=e2πiz/h.
Thus (f[α]2n)(z) = ˜f(z) is meromorphic at ∞ as claimed. In sum, every meromorphic differential ω of degree n on X(Γ) pulls back to a meromor- phic differentialπ∗(ω) =f(τ)(dτ)n onHwhere f is an automorphic form of weight 2nwith respect to Γ.
The calculation that π∗(ω) is well defined can be turned around (Exer- cise 3.3.2(b))—given a collection (ωj)∈
j∈JΩ⊗n(Vj) of local meromorphic differentials, exclude the cusps by settingUj =Uj∩ H andVj=ψj(Uj) and ωj =ωj|Vj for allj ∈J. If theωj pull back underψj∗ to restrictions of some meromorphic differentialf(τ)(dτ)n onHthen the original local differentials ωj are compatible and so (ωj)∈Ω⊗n(X(Γ)). Thus, a given collection (ωj) of local meromorphic differentials is compatible, giving a meromorphic differen- tial onX(Γ), if and only if its local elements pull back to restrictions of some f(τ)(dτ)n∈Ω⊗n(H) withf ∈ A2n(Γ).
Conversely, given an automorphic form f ∈ A2n(Γ) we will construct a meromorphic differentialω(f) ∈ Ω⊗n(X(Γ)) such that π∗(ω) = f(τ)(dτ)n. By the previous paragraph it suffices to construct local differentials that pull back to restrictions of f(τ)(dτ)n. Thus the idea is to express f(τ)(dτ)n in local coordinates.
Each local map ψj : Uj −→ Vj is a compositeψj =ρj ◦δj : τ →z → q with δj ∈ GL2(C). Since δj is invertible it is easy to transform f(τ)(dτ)n intoz-coordinates locally. First extend the weight-koperator [γ]k to matrices γ∈GL2(C) by definingj(γ, τ) =cτ+das before and
(f[γ]k)(τ) = (detγ)k/2j(γ, τ)−kf(γ(τ))
when this makes sense. The exponentk/2 of detγ is a normalization conve- nient for the next calculation because of the formula γ(τ) = detγ/j(γ, τ)2 (Exercise 3.3.3); in other contexts the factor (detγ)k−1 is more convenient.
Now letUj=Uj∩ H. Thenf(τ)(dτ)n|Uj is a pullbackδ∗j(λj) whereλj is the differential obtained by pulling f(τ)(dτ)n|Uj forward to z-space under δj−1. Lettingα=δj−1,
λj=α∗(f(τ)(dτ)n|Uj) =f(α(z))(d(α(z)))n
= (detα)nj(α, z)−2nf(α(z))(dz)n= (f[α]2n)(z)(dz)n.
Thusλj isδjΓ δj−1-invariant sincef(τ)(dτ)n is Γ-invariant. Further pushing λj forward fromz-space toq-space isn’t quite as easy sinceρj is not invertible in general, but it isn’t hard to find local forms ωj that pull back under ρj
toλj as desired.
Specifically, ifUj⊂ H, i.e., if we are not working at a cusp, then δj takes τj to 0 and the quotient {±I}(δjΓ δj−1)0/{±I} is cyclic of order h, gener- ated by the rotationrh :z →µhz whereµh =e2πi/h. By δjΓ δ−j1-invariance the pullback rh∗(λj) = (f[α]2n)(µhz)µnh(dz)n must equal λj, or equivalently µnh(f[α]2n)(µhz) = (f[α]2n)(z), or (µhz)n(f[α]2n)(µhz) = zn(f[α]2n)(z). In other words, the functionzn(f[α]2n)(z) takes the formgj(zh) for some mero- morphic function gj. Note for later reference in proving formula (3.8) that hν0(gj) =n+ντj(f) and thusν0(gj) =νπ(τj)(f) +n/h. Define a local mero- morphic differential inq-coordinates,
ωj= gj(q)
(hq)n(dq)n onVj. (3.6)
Sinceρj(z) =zh, thisωj pulls back underρj toλj (Exercise 3.3.4(a)), which in turn pulls back underδjto the originalf(τ)(dτ)n|Uj. Thus, for eachUj⊂ H the differential ωj onVj pulls back underψj to a suitable restriction of the global differentialf(τ)(dτ)n onH.
On the other hand, ifUj contains a cuspsj thenδj takessj to∞and the function (f[α]2n)(z) takes the formgj(e2πiz/h) wheregj is meromorphic at 0 andhis the width ofs. The relevant local differential is now
ωj = gj(q)
(2πiq/h)n(dq)n onVj, (3.7) which is meromorphic atq= 0. Sinceq=ρj(z) =e2πiz/h, againωj pulls back under ρj to λj (Exercise 3.3.4(b)), and as before it follows that ψ∗j(ωj) = f(τ)(dτ)n|Uj. Putting all of this together gives
Theorem 3.3.1.Let k ∈ N be even and let Γ be a congruence subgroup ofSL2(Z). The map
ω:Ak(Γ)−→Ω⊗k/2(X(Γ))
f →(ωj)j∈J where(ωj)pulls back to f(τ)(dτ)k/2∈Ω⊗k/2(H) is an isomorphism of complex vector spaces.
Proof. The mapωis defined since we have just constructedω(f). Clearlyωis C-linear and injective. Andωis surjective because every (ωj)∈Ω⊗k/2(X(Γ)) pulls back to somef(τ)(dτ)k/2∈Ωk/2(H) withf ∈ Ak(Γ).
Exercises 3.2.3 and 3.2.4 showed that forkpositive and even,Ak(Γ) takes the form C(X(Γ))f where C(X(Γ)) is the field of meromorphic functions onX(Γ) andf is any nonzero element ofAk(Γ). Thus, Theorem 3.3.1 shows thatΩ⊗k/2(X(Γ)) =C(X(Γ))ω(f) for such k.
The aim of this chapter is to compute the dimensions of the subspaces Mk(Γ) andSk(Γ) ofAk(Γ). Now that we know thatAk(Γ) andΩ⊗k/2(X(Γ))
are isomorphic, the final business of this section is to describe the images ω(Mk(Γ)) andω(Sk(Γ)) in Ω⊗k/2(X(Γ)). Some Riemann surface theory to be presented in the next section will then find the desired dimensions by computing the dimensions of these image subspaces instead in Sections 3.5 and 3.6.
So take any automorphic formf ∈ Ak(Γ) and let ω(f) = (ωj)j∈J. For a point τj ∈ H, the local differential (3.6) with n= k/2 vanishes at q = 0 to (integral) order (Exercise 3.3.5)
ν0(ωj)def=ν0
gj(q) (hq)k/2
=νπ(τj)(f)−k 2
1−1
h
. (3.8)
In particular, at a nonelliptic point, when h = 1, the order of vanishing is ν0(ωj) =νπ(τj)(f), the order of the original function. For a cuspsj the local differential (3.7) with n = k/2 vanishes at q = 0 to order (Exercise 3.3.5 again)
ν0(ωj)def=ν0
gj(q) (2πiq/h)k/2
=νπ(sj)(f)−k
2. (3.9)
When k ∈ N is even, formulas (3.8) and (3.9) translate the conditions νπ(τj)(f)≥0 andνπ(sj)(f)≥0 characterizingMk(Γ) as a subspace ofAk(Γ) into conditions characterizingω(Mk(Γ)) as a subspace ofΩ⊗k/2(X(Γ)), and similarly forSk(Γ) andω(Sk(Γ)). In particular, the weight 2 cusp formsS2(Γ) are isomorphic as a complex vector space to the degree 1holomorphicdiffer- entials onX(Γ), denotedΩhol1 (X(Γ)) (Exercise 3.3.6). This special case will figure prominently in the later chapters of the book.
Exercises
3.3.1.(a) Show that the pullback is contravariant.
(b) Show that ifι:V1−→V2is inclusion then its pullback is the restriction ι∗(ω) =ω|V1 forω∈Ω⊗n(V2).
(c) Show that if ϕ is a holomorphic bijection of open sets in C then (ϕ−1)∗= (ϕ∗)−1.
(d) Show that if π : V1 −→ V2 is a holomorphic surjection of open sets inCthenπ∗ is an injection. Ifi:V1−→V2 is an injection of open sets inC, needi∗ be a surjection?
3.3.2.(a) Let ψ = ρ ◦ δ where δ ∈ SL2(Z) and ρ(z) = e2πiz/h. Let ω =g(q)(dq)n. Show that ψ∗(ω) = f(τ)(dτ)n with f = ˜f[δ]2n and ˜f(z) = (2πi/h)nqng(q) whereq=e2πiz/h.
(b) Consider the situation π : H −→ X(Γ) as described in the small in diagram (3.5). Show that if the local meromorphic differentials ωj pull back underψj∗to restrictions of some common global meromorphic differential f(τ)dτ onHthen they are compatible. Exercise 3.3.1(d) is relevant.
3.3.3.For γ ∈ GL2(C), show thatγ(τ) = detγ/j(γ, τ)2. (This generalizes Lemma 1.2.2(e).)
3.3.4.(a) For a neighborhood Uj ⊂ H, show that the local differential ωj
in (3.6) pulls back underρj toλj. How wasωj found in the first place?
(b) Same question forωj in (3.7) whenUj contains a cusp.
3.3.5.Prove formulas (3.8) and (3.9).
3.3.6.Prove that S2(Γ) and Ωhol1 (X(Γ)) are isomorphic as complex vector spaces. Your proof will incidentally show that the elements ofS2(Γ) vanish at the elliptic points ofX(Γ). Argue this directly as well by examining the text leading up to (3.6).