In this paper we will use the Riemann-Roch Theorem to figure out the dimensions of the spaces spanned by modular forms and cusp forms for given groups, and then we will look at Poincare’ and Eisenstein series, which turn out to span the spaces of modular forms in a certain way. 1 Modular forms A modular form of weight 2k for Γ is a function f such that 2k • f ( az+b cz+d ) = (cz + d) f (z) for all z ∈ H • f is holomorphic on H • f is holomorphic at the cusps of Γ. If f is zero at the cusps, we call it a cusp form. Note, also, that the literature is not consistent on whether to talk about k or 2k. In any case, the weight of the form is the same as the exponent of cz + d. What does it mean to be holomorphic at the cusps? Recall that cusps are those points s ∈ R ∪ {∞} for which there exists a parabolic element γ ∈ Γ with s as its fixed point (where parabolic means Tr(γ) = ±2). About any point in the Riemann surface Γ/H∗ we can choose local coordinates, and we use the expansion of f in those coordinates to define it as holomorphic or meromorphic. For example, when talking about Γ(1), which has one cusp at ∞, we said that one could choose q = e2πiz as a coordinate function around ∞ in Γ(1)/H. f n has a Fourier expansion about ∞, f (z) = Σ∞ n=−∞ an q , and we say that f is holomorphic if an = 0 for n < 0. The definition of meromorphic at the cusps is similar: an = 0 for n < −N . Denote the space of modular forms of weight 2k for a group Γ by Mk (Γ), and the space of cusp forms for the same group by Sk (Γ). What are the dimensions of these spaces? Finiteness of dimension It is very easy to prove that Mk (Γ) is a finitedimensional space. Recall the following proposition: If S is a compact Riemann surface, and V the space of meromorphic functions on S having poles of order at most r1 , . . . , rm at the points P1 , . . . , Pm , then V has dimension at most r1 + · · · + rm + 1. This is easy to see – one can write any function as a linear combination of functions with order ≤ ri at point Pi and an everywhere-analytic function. Given this, if f0 ∈ Mk (Γ) is a nonzero modular form with zeros at P1 , . . . , Pm of order of vanishing r1 , . . . , rm , then for f ∈ Mk (Γ), f /f0 is an automorphic function. In fact, f 7→ f /f0 is an isomorphism of Mk (Γ) with the vector space V described above, and so Mk (Γ) has dimension ≤ r1 + · · · + rm + 1, which is finite. This, of course, does not tell us the dimension of Mk (Γ) at all. For an actual calculation of the dimension, one must use the Riemann Roch Theorem. 1 Riemann Roch and preliminaries Consider meromorphic functions f (z) on a Riemann surface S, meromorphic differential forms ω(z) = f (z)dz on S, and points p ∈ S. As above, f (z) has a Laurent series expansion at any p ∈ S, m f (z) = z n Σ∞ m=1 am z , with a0 6= 0. We let νp (f ) = n denote the order of f at p, and likewise for a differential ω(z) = f (z)dz, let νp (ω) = νp (f ) denote the order of ω(z) at p. Form a free abelian group on the points of S and call it the group of divisors: divisors are formal sums of point with integer coefficients almost all zero. The divisor of f is written θ(f ) = Σp∈S νp (f ) · p, and θ(ω) is defined similarly. The order of a divisor is |θ| = Σp∈S µp for θ = Σp∈S µp · p. From complex analysis, it’s easy to see that the divisors of meromorphic functions f are the kernel of the order homomorphism from the group of divisors to Z. For a differential form ω, |θ(ω)| = 2(g − 1), where g is the genus of S. Also, we can put an ordering on the divisors: θ ≥ θ 0 if µp ≥ µ0p for all p. The Riemann Roch Theorem: Let M be the space of meromorphic functions on S, and L(θ) = {f ∈ M |θ(f ) + θ ≥ 0}. Then for all ω, dim L(θ) = dim L(θ(ω) − θ) + |θ| + 1 − g. To compute the dimension of Mk (Γ), we must compare the orders of f ∗ and f at corresponding points on S = Γ/H∗ and H. Write np (f ) and np (ω) for the orders on H. At regular points of S, ρ : H → Γ/H is a local isomorphism, so νp (f ∗ ) = np (f ). 2 The Eisenstein series Consider a function from a lattice Λ to the complex numbers, where ω denotes a point on the lattice Λ: Gk (Λ) = Σω∈Λ,ω6=0 1/ω 2k . We can then define Gk (z), the Eisenstein series, as Gk (z) = Gk (Λ(z, 1)) = Σ(m,n)6=(0,0) 1/(mz + n)2k . This is clearly weakly modular, and we can prove that it’s holomorphic on H and takes the value 2ζ(2k) at ∞: thus the Eisenstein series is a modular form of weight 2k for Γ(1). Why define this on a lattice? It is rather instructive to look at Gk on a lattice because there are natural equivalences between the categories of (i) elliptic curves over C, (ii) Riemann surfaces of genus 1 together with a point 0, and (iii) lattices Λ ∈ C (naming just the objects and leaving you to fill in the morphisms). This means something... In particular, Gk comes up in the coefficients of elliptic curves. Something is going on here... Another way to look at the Eisenstein series is as a space of forms orthogonal to the cusp forms. The Poincare series of weight 2k and character 0 for Γ(N ) 1 is φ0 (z) = Σ (cz+d) 2k , summed over (c, d) = (0, 1) mod N, (c, d) = 1, and this takes the value 1 at ∞ and vanishes at all other cusps. How could one make a modular form that takes the value 1 at a cusp P and vanishes at all other cusps? There is a σ ∈ Γ(1) such that σ(P ) = ∞, so if we let jγ (z) = 1/(cz + d)2 , then ϕ(z) = jσ (z)k · φ0 (σz) is a modular form of the type desired. Writing this out, (a) a restricted Eisenstein series of weight 2k for Γ(N ) is the series G(z; c0 , d0 ; N ) = Σ(cz + d)−2k , 2 summing over (c, d) = (c0 , d0 ) mod N , (c, d) = 1, and (c0 , d0 ) a pair such that gcd(c0 , d) , N ) = 1. (b) a general Eisenstein series of weight 2k > 2 for Γ(N ) is the series G(z; c0 , d0 ; N ) = Σ(cz + d)−2k , summing over (c, d) = (c0 , d0 ) mod N , and (c, d) 6= (0, 0). There is a distinct restricted Eisenstein series for each cusp, so they are linearly independent. The general Eisenstein series are the linear combinations of the restricted Eisenstein series. 3 The Poincare series Milne says that we would like to construct modular forms for subgroups Γ of finite index in Γ(1).To do this, we use a method of Poincare. It’s similar to the standard way of constructing invariant functions — recall the idea that if h is a function on H, then f (z) = Σγ∈Γ0 h(γz) is invariant under Γ if the series converges absolutely (Γ0 is the name for the image of Γ in Γ(1)/{±I}. We need an automorphy factor for Γ in order to adapt this idea. An automorphy factor is a map j : Γ × H → C× such that (a) for each γ ∈ Γ, z 7→ jγ (z) is a holomorphic function on H, and (b) jγγ 0 (z) = jγ (γ 0 z)j̇γ 0 (z). Try f (z) = Σγ∈Γ0 h(γz) jγ (z) . If this converges absolutely uniformly on compact sets, we’ve got it, as f (γ 0 z) = Σγ∈Γ0 h(γγ 0 z) h(γγ 0 z) jγ 0 (z) = jγ 0 (z) · f (z). = Σγ∈Γ0 0 jγ (γ z) jγγ 0 (z) The problem here is that there may be infinitely many γ’s for which jγ (z) = 1, and in this case the sum would not converge. Call the set of γs for which jγ (z) = 1 identically Γ0 . It is a subgroup of Γ, and we can choose R a set of coset representatives of Γ mod Γ0 so that Γ = ∪γ∈R Γ0 γ. Then try γ ∈ R rather than γ ∈ Γ0 for the sum above, and prove convergence on compact sets. In particular, consider jγ (z) = (cz + d)2k . Notice that Γ0 is the infinite cyclic subgroup of translations in Γ, z 7→ z + h, and an example of a function invariant under this action is exp 2πinz/h. Try this for h, to get the Poincare series φn (z) = ΣΓ0 \Γ0 exp(2πin · γ(z)/h) . (cz + d)2k 1 Prove convergence by comparing the sum to Σ(m,n)6=(0,0) |mz+n| α , which converges for α > 2. This is on the spring 1999 complex prelim so I’ll omit the proof. The cool thing about Poincare series is that they span Mk (Γ). We need the Petersson inner product to prove this. 3 4 Petersson inner product If f and g are modular forms of weight 2k for a group Γ ∈ Γ(1), and at least one is a cusp form, then the Petersson inner product of f and g is defined to be ZZ < f, g >= f (z) · g(z) · y 2k−2 dxdy. D It is linear in the first variable, semi-linear in the second; < f, g >= < g, f >; and < f, f >> 0 for all f 6= 0. So it’s a positive-definite Hermitian form on Sk (Γ), making Sk (Γ) into a finite dimensional Hilbert space. It’s independent of the choice of D (fundamental domain of Γ). Using the Petersson inner product, we can prove that every cusp form is a linear combination of Poincare series. This follows as an easy corollary from something that looks a bit more complicated: Theorem Given f a cusp form for Γ, and φn the Poincare series of the same weight for Γ, we have < f, φn >= h2k (2k − 2)! 1−2k ·n · an . (4π)2k−1 Here h is the width of ∞ as a cusp for Γ, and an is the nth coefficient in the Fourier expansion of f . If f is orthogonal to all Poincare series, then all its Fourier series coefficients are zero, and so f must be zero; thus Sk (Γ) is spanned by Poincare series. Moreover, since the inner product is defined even if only one of the functions is a cusp form, we can look at the inner product of a cusp form (a linear combination of Poincare series) and an Eisenstein series. One finds (Milne assures us) that < f, g >= 0 if either f or g is a restricted Eisenstein series, and so in fact the space of all Eisenstein series is the orthogonal complement of S)k (Γ) in Mk (Γ). 4