AN ELEMENTARY AND COMPUTATIONAL APPROACH TO HEEGNER POINTS
SHELLY MANBER
Abstract. In this paper we present a method for explicitly computing ratio- nal points on elliptic curves using Heegner points. This method was crucial to the proof of Gross-Zagier, which proves the rank one case of the Birch and Swinnerton-Dyer Conjecture. Although this use of Heegner points can be found in many books and articles, we strive here to present it in a more concrete and complete form, and using explicit and elementary tools.
1. Introduction One of the most basic and fundamental mathematical problems is to find an integral solution to a given polynomial equation. This is a problem most people encounter even in grade school. For example, it is a classic and relatively simple exercise to enumerate all pythagorean triples, or integral solutions to the equation x2 +y2 = z2. By contrast, Fermat’s Last Theorem - prove that there are no integral solutions to the equation xn + yn = zn for n > 2 - is a famously difficult problem that took hundreds of years to solve. Indeed, although it is simple to state and understand, finding integral solutions to arbitrary polynomial equations remains an extremely difficult problem in modern number theory. We will be particularly interested in finding rational points on elliptic curves, or smooth, one dimensional, genus one projective varieties. Elliptic curves are a natural choice of curve for study because they are simple enough that much is known about them and they can be worked with explicitly, yet complex enough that there is a rich theory attached to them with connections to many different fields. For example, elliptic curves are abelian algebraic groups and can be described analytically as a quotient of C such that the group law corresponds to addition in C and algebraic morphisms correspond to homotheties on C. Furthermore, to each elliptic curve we can assign an L-series and modular form that play a role in analytic number theory. We outline here a method to construct explicit rational points on elliptic curves defined over Q. The method is a consequence of the modularity theorem (for- merly known as the Taniyama-Shimura-Weil conjecture) and the Artin Reciprocity Theorem. In Section 2, we will present background information on elliptic curves and their moduli space from an analytic perspective. In Section 3, we will present background information on elliptic curves and their moduli space viewed as algebraic varieties. In Section 4 we will discuss the theory of complex multiplication and in Section 5 some background on orders in imaginary quadratic number fields, both of which we will need to discuss the Artin Reciprocity Theorem in Section 6. In Section 7 we will define Heegner points and generalize the Artin Reciprocity Theorem to 1 2 SHELLY MANBER construct explicit finite sets of points on modular curves that are defined over a finite algebraic extension of Q and form a complete set of galois conjugates. This, together with the modularity theorem, which we will state in Section 8 and which gives us a morphism to each elliptic curve defined over Q from a modular curve determined by its conductor, will complete our algorithm for constructing rational points on elliptic curves. In Section 9 we will present some examples of this algorithm computed with PARI, a computer algebra system specializing in number theory computations.
2. Elliptic Curves and Their Moduli Space - Anayltically In this section we present elementary background information about elliptic curves and the modular curves X0 and X0(N). A positively oriented lattice Λ in C is an additive subgroup of C of the form Λ = {mω1 + nω2|m, n ∈ Z} where ω1 and ω2 are fixed complex numbers such that ω1/ω2 ∈ H, the upper half plane. If Λ = {mω1 + nω2|m, n ∈ Z} then we say that the pair (ω1, ω2) is a basis for Λ and denote this by Λ = hω1, ω2i. It follows directly from the definitions that two pairs of complex numbers (ω1, ω2) and (τ1, τ2) are bases for the same lattice if and only if ω1 a b τ1 a b = for some ∈ SL2(Z). ω2 c d τ2 c d
We say that two lattices Λ and Λ0 are homothetic if Λ = αΛ0 for some α ∈ C. Thus hω1, ω2i and hτ1, τ2i are homothetic if and only if there exists an α ∈ C such that hω1, ω2i = hατ1, ατ2i. We can define an equivalence relation on the set of positively oriented lattices by assigning two lattices to be equivalent if they are homothetic; let L denote the set of equivalence classes of lattices under this equivalence relation. Since hω1, ω2i ∼ ω2hω1/w2, 1i, each element of L can be represented by a lattice with a basis of the form hτ, 1i with τ ∈ H. Furthermore, τ is unique up to fractional linear transformation by SL2(Z). Indeed, every basis for the lattice hω1, ω2i is of a b the form haω + bω , cω + dω i for some ∈ SL ( ), and thus the set of 1 2 1 2 c d 2 Z lattices of the form hτ, 1i equivalent to hω1, ω2i is exactly aω1 + bω2 a b , 1 ∈ SL2(Z) . cω1 + dω2 c d
Thus the elements of L are in bijective correspondence with the points of H/SL2(Z), where SL2(Z) acts on H by fractional linear transformation. An elliptic curve is an analytic manifold given by C/Λ where Λ is a lattice in C. It naturally has the structure of an analytic manifold with charts given by lifting an open subset of C/Λ to any connected open subset of C in the preimage of the projection from C to C/Λ. It also has a natural group structure given by addition in C. Claim 2.1. Two elliptic curves are isomorphic as analytic manifolds if and only if their corresponding lattices are homothetic.
Proof. See [Sil86, Chapter VI, Corollary 4.4.1]. AN ELEMENTARY AND COMPUTATIONAL APPROACH TO HEEGNER POINTS 3
From Claim 2.1 we see that the set of elliptic curves up to isomorphism is in bijective correspondence with the elements of L and hence with the set H/SL2(Z). Let X0 denote the analytic manifold associated to H/SL2(Z) with charts given by lifting to the upper half plane, as with C/Λ. Then we say that the points of X0 index the set of elliptic curves up to isomorphism, i.e. each point corresponds to a unique isomorphism class of elliptic curves and each isomorphism class of elliptic curves corresponds to a point on X0. We will see in Section 3 that both E and X0 have natural structures as algebraic varieties over Q. The curve X0 is a special case of a family of algebraic curves called modular curves, denoted X0(N) with N ∈ N. They are of particular interest here because of the Modularity Theorem, which states that for every elliptic curve E defined over the rationals, there is a positive integer N (called the conductor of the curve; see Section 8) and a morphism of algebraic varieties ΦE : X0(N) → E defined over Q. Thus we can construct explicit points on E over a number field K by constructing K-rational points on X0(N) and taking their image under ΦE. To define the curves X0(N), we first need to define the modular congruence subgroups Γ0(N). Definition 1. Let N be a positive integer. Then n a b o Γ (N) := ∈ SL ( ) | c ≡ 0 mod N 0 c d 2 Z and
X0(N) := H/Γ0(N), where Γ0(N) acts on H by fractional linear transformation.
The curve X0(N) is called the modular curve of level N and has a structure of an analytic manifold, with charts given as with X0. Note that Γ0(N) is a subgroup of SL2(Z) and that Γ0(1) = SL2(Z) so that X0(1) = X0. Just as the points X0 index elliptic curves in C up to homothety, the points of 0 X0(N) index pairs of elliptic curves (E,E ) together with a cyclic isogeny of level N 0 between them [Gro84]. In other words, the points of X0(N) index triples (E,E , ϕ), up to equivalence, where ϕ is a surjective analytic map ϕ : E → E0 such that ker 0 (ϕ) is an additive cyclic subgroup of E of order N. The triples (E1,E1, ϕ1) and 0 (E2,E2, ϕ2) are defined to be equivalent if there are isomorphisms γ : E1 → E2 and 0 0 0 γ : E1 → E2, such that the following diagram commutes:
ϕ1 0 (1) E1 / E1 .
γ γ0
ϕ2 0 E2 / E2
Equivalently, we can say that each point of X0(N) corresponds to an elliptic curve with a distinguished cyclic subgroup of order N, up to isomorphisms that preserve the subgroup. These notions are equivalent because clearly the map ϕ : E → E0 uniquely determines the subgroup ker(ϕ) of E, and additionally the pair (E, ker(ϕ)) uniquely determines E0 up to isomorphism [Sil86, Chapter III, Prop 4.12]. To prove the bijective correspondence between points of X0(N) and elliptic curves with distinguished subgroups, we make two claims (adopted from [DS05, Theorem 1.5.1(a)]). 4 SHELLY MANBER
1 Claim 2.2. For any τ ∈ H, let [Eτ , N ] denote the equivalence class of the elliptic 1 curve C/hτ, 1i and subgroup generated by N + hτ, 1i (as above, two curve-subgroup pairs are equivalent if there is an isomorphism of elliptic curves that preserves the subgroup). Then the set of all equivalence classes of elliptic curves paired with order 1 N cyclic subgroups is exactly {[Eτ , N ]: τ ∈ H}.
1 1 Claim 2.3. Let τ, ω ∈ H. Then [Eω, N ] = [Eτ , N ] if and only if there exists a b ∈ Γ (N) such that ω = aτ+b . c d 0 cτ+d
1 From Claim 2.3 we see that the distinct elements of {[Eτ , N ]: τ ∈ H} are in 1 bijection with H/Γ0(N), and from Claim 2.2 that the set {[Eτ , N ]: τ ∈ H} is in bijection with the the set of all elliptic curves with distinguished cyclic, order N subgroups. Thus the correspondence follows immediately from the two claims. We now set about proving the claims. Proof of Claim 2.2 . Let E = C/Λ be an elliptic curve with cyclic subgroup G of order N, and let x + Λ be any generator of G. Since G has order N, we have Nx + Λ = Λ so Nx ∈ Λ. Our goal will be to find a specific generator x + Λ of G such that we can complete Nx to a basis for Λ, i.e. write Λ = hτ, Nxi. If we 1 can successfully do this then we apply the homothety of multiplication by Nx to τ ∼ Λ to get an isomorphic curve C/h Nx , 1i = E. This homothety sends x + Λ to 1 τ τ 1 1 N + h Nx , 1i. Hence the pair (E,G) is equivalent to (C/h Nx , 1i, h N i) = [Eτ 0 , N ], 0 τ where τ = Nx . All that remains is to find complex numbers x and τ such that x + Λ generates G and Λ = hτ, Nxi. Let x0 be a representative of any generator of G. Then (C/Λ)/G = C/L where L = {ω + kx0|ω ∈ Λ, k ∈ Z}. The set L is a discrete subgroup of C and contains Λ, so it is a lattice in C. If there exists y ∈ L such that x0 = ry for some integer r, then y + Λ is also a generator of G. Indeed, y has order dividing N because y = ωy + kyx0 with ωy ∈ Λ, ky ∈ Z, and Nx0 ∈ Λ, so Ny ∈ Λ. If y had order s < N then sy ∈ Λ, so we would have sx0 = rsy ∈ Λ, which would be a contradiction to x0 representing a generator. This means that we can always choose a complex number x such that x + Λ generates G and x is indivisible in L, i.e. x is not an integral multiple of any element of L. We next claim that choosing such an x guarantees that Nx is indivisible in Λ. Indeed, assume to the contrary that there exists y ∈ Λ such that ry = Nx. Then N we must have gcd(N, r) = 1, since otherwise we would have gcd(N,r) x ∈ Λ and N gcd(N,r) ∈ Z, which is impossible since x generates a cyclic group of order N. Let Na Nb L = hω1, ω2i and let x = aω1 + bω2 with a, b ∈ Z, so y = r ω1 + r ω2 ∈ Λ ⊂ L. Since gcd(N, r) = 1 we must have r dividing both a and b. But a and b cannot share any common factors since x in indivisible in L. So r = 1 and Nx indivisible in Λ. Finally, let hλ1, λ2i be a basis for Λ and Nx = aλ1 +bλ2. Since Nx is indivisible, r s we have gcd(a, b) = 1 so there exist r, s ∈ such that rb−sa = 1. Now ∈ Z a b SL2(Z) so hrλ1 + sλ2, aλ1 + bλ2i is an alternate basis for Λ. Letting τ = rλ1 + sλ2, we have Λ = hτ, Nxi as desired. AN ELEMENTARY AND COMPUTATIONAL APPROACH TO HEEGNER POINTS 5
1 1 Proof of Claim 2.3 . Let [Eω, N ] = [Eτ , N ]. Then there exists an isomorphism 1 ϕ : Eτ → Eω such that the group generated by N on Eτ is sent to the group 1 generated by N on Eω. Note that if Eτ or Eω have automorphisms then there will be more than one isomorphism between the two curves, but at least one will 1 1 send the group generated by N on Eτ to the group generated by N on Eω. Let a b ∈ SL ( ) be the matrix associated with ϕ, i.e, ω = aτ+b and the c d 2 Z cτ+d 1 isomorphism ϕ : C/hτ, 1i → C/hω, 1i is given by multiplication by cτ+d . Now since ∼ 2 1 C = R , we can write N uniquely in the form 1 (2) = r(aτ + b) + s(cτ + d) N with r, s ∈ R, so 1 (3) ϕ( ) = rω + s. N 1 By our choice of ϕ, the image of N must be a generator of the group generated by 1 1 m N on Eω. This means that ϕ( N ) ≡ N mod hω, 1i for some m relatively prime to N. Combining with (3) yields m rω + s ≡ mod hω, 1i, N m so we must have r ∈ Z and s = N + k with k ∈ Z. Replacing this expression for s in (2) gives 1 m = r(aτ + b) + ( + k)(cτ + d), N N mc from which we see that ra + N + kc = 0. Since gcd(m, N) = 1, we must have N|c. a b So ∈ Γ (N). c d 0
a b Conversely, let ω = aτ+b for some ∈ Γ (N). Since ad − bc = 1, we cτ+d c d 0 have 1 −c a = (aτ + b) + (cτ + d), N N N so 1 −c a ϕ( ) = ω + . N N N 1 a But we know that N divides c, so in fact ϕ( N ) = N in C/hω, 1i. We also know that gcd(a, c) = 1 since ad − bc = 1, which means that in particular, gcd(a, N) = 1. So 1 1 the image of N under ϕ generates the same subgroup of C/hω, 1i that N generates, 1 1 and hence [Eω, N ] = [Eτ , N ].
Notice that the proofs of Claims 2.2 and 2.3 give us an explicit correspondence between elements of H/Γ0(N) and pairs of elliptic curves with a cyclic isogeny 1 between them. Given τ ∈ H/Γ0(N), the natural projection C/hτ, 1i → C/hτ, N i is a cyclic isogeny of order N whose kernel is the subgroup of Eτ = C/hτ, 1i generated 1 1 0 by N + hτ, 1i and thus corresponds to [Eτ , N ]. Conversely, given that ϕ : E → E is an isogeny whose kernel is a cyclic order N subgroup of E, Claim 2.2 shows that 6 SHELLY MANBER
1 there exists a τ ∈ H such that [E, ker(ϕ)] is equivalent to [Eτ , N ] and Claim 2.3 tells us that such a τ is unique up to the action of Γ0(N). By symmetry, it is equivalent to assign to each τ ∈ H/Γ0(N) the pair of lattices τ 1 hτ, 1i ⊂ h N , 1i. Indeed, multiplying both lattices hτ, 1i and hτ, N i by −1/τ and 00 −1 applying the change of variables τ = τ gives a homothetic pair of lattices τ 00 h−1, τ 00i ⊂ h−1, i, N or, choosing a different basis, τ 00 hτ 00, 1i ⊂ h , 1i N 00 with τ ≡ τ mod SL2(Z). The last thing we will need from this section is the notion of lattice index. 0 0 0 Let Λ = hλ1, λ2i and Λ = hλ1, λ2i be lattices in C. There is a unique linear 2 ∼ 0 0 automorphism T of the vector space R = C sending λ1 to λ1 and λ2 to λ2. We define the lattice index of Λ in Λ0, denoted [Λ0 : Λ], to be the determinant of T [Ste04, Definition 10.3.1]. Note that if Λ ⊂ Λ0 then [Λ0 : Λ] is the order of the additive group Λ0/Λ. Lemma 2.4. If N ⊂ M ⊂ L are lattices in C then [L : N] = [L : M][M : N].
Proof. Let T1 : L → M and T2 : M → N be linear automorphisms of C with determinants d1 and d2 respectively. Then T2T1 : L → N is a linear automorphism with determinant d1d2. Thus [L : N] = d1d2 = [L : M][M : N]. Note that we have defined the notion of lattice index on two concrete lattices, not on elements of L. Thusfar, our definition of lattice index does not extend to elements of L, because applying a homothety to one lattice from a pair latices will change the determinant of the automorphism between the two lattices if the complex number associated to the homothety has a magnitude other than 1. Claim 2.5. Let E and E0 be a pair of elliptic curves with an isogeny ϕ : E → E0 such that ker ϕ is cyclic order N. Then there exist lattices Λ ⊂ Λ0 in C such that ϕ can be realized as the analytic map φ : C/Λ → C/Λ0 induced by the identity on C. Furthermore, for any such pair of lattices we have [Λ0 : Λ] = N. Proof. Let Λ be any lattice such that E = C/Λ, then we can find a lattice Λ0 such that E0 = C/Λ0 and Λ ⊂ Λ0. Indeed, we will see later on (Theorem 3.1) that every isogeny between elliptic curves C/Λ → C/Λ0 is of the form z 7→ αz for some α such 0 that αΛ ⊂ Λ0. By choosing Λ = Λ0/α as the representative for the homothety 0 class of Λ0, we have our desired Λ ⊃ Λ. Now consider the map φ : C/Λ → C/Λ0 induced by the identity on C. Since Λ ⊂ Λ0, we have ker ϕ = Λ0/Λ. As in the proof of Claim 2.2, we can choose a generator x + Λ of ker ϕ such that Λ = hτ, Nxi and Λ0 = hτ, xi for some τ ∈ C. Letting (τ, x) be a basis for C as an R-vector space, the linear automorphism θ 1 0 given by satisfies θ(Λ0) = Λ and has determinant N. 0 N Note that it is not the case that [Λ0 : Λ] = N whenever E = C/Λ and E0 = C/Λ0. Claim 2.5 shows that whenever E = C/Λ and E0 = C/Λ0 and Λ ⊂ Λ0, we must have [Λ0 : Λ] = N. AN ELEMENTARY AND COMPUTATIONAL APPROACH TO HEEGNER POINTS 7
3. Elliptic Curves and Their Moduli Space - Algebraically An elliptic curve C/Λ also has the structure of a projective variety. To define the structure explicitly given the analytic formulation, we will introduce several well- known analytic functions. The following several paragraphs give a brief summary of these analytic functions and their properties. A more detailed presentation can be found in almost any textbook on analytic number theory, complex analysis, or elliptic curves, for example [SS03]. The Weierstrass elliptic function ℘ : C × L → C is defined as follows: X 1 1 ℘(z, Λ) = z−2 + − . (z − τ)2 (τ)2 τ∈Λ τ6=(0,0) One can check that ℘(z +τ, Λ) = ℘(z, Λ) for all τ ∈ Λ, so for a fixed lattice Λ the Weierstrass function is doubly periodic as a function of z. The Weierstrass elliptic function also satisfies the differential equation 0 2 (4) ℘ (z, Λ) = 4℘(z, Λ) − g2(Λ)℘(z, Λ) − g3(Λ) where the analytic functions g2 : L → C and g3 : L → C are defined by X 1 g (Λ) = 60 2 τ 4 τ∈Λ τ6=(0,0) X 1 g (Λ) = 140 . 3 τ 6 τ∈Λ τ6=(0,0) The explicit correspondence between C/Λ and it’s corresponding projective va- riety is given by the following map:
ϕ : C → P2(C) τ 7→ [℘(τ, Λ); ℘0(τ, Λ); 1].
The map ϕ is defined on C/Λ since ℘(z + τ, Λ) = ℘(z, Λ) for all τ ∈ Λ, and (4) shows that the image of φ is in the locus of the homogenous equation y2z = 3 2 3 4x − g2(Λ)xz − g2(Λ)z in P2(C). Notice that ℘(z, Λ) has a pole when z ∈ Λ, so the image of 0 in C/Λ is [0, 1, 0], and we call this the “point at infinity”, or “distinguished point”. It turns out that ϕ defines an isomorphism between the the 2 3 2 3 locus of y z = 4x − g2(Λ)xz − g2(Λ)z and C/Λ as analytic manifolds [Sil86, Chapter VI, Proposition 3.6(b)]. 2 3 2 3 The equation y z = 4x − g2(Λ)xz − g2(Λ)z describes a smooth, genus one projective curve over C. Furthermore, every smooth, genus one, projective curve can be expressed over C as the locus of an equation of the form y2z = 4x3−axz2−bz3 [Sil86, Chapter VI, Proposition 3.6 and Theorem 5.1]. Two other analytic functions will be important to the study of elliptic curves. These are the modular discriminant 3 2 ∆ := g2 − 27g3 and Klein’s j-invariant g3 j := 1728 2 . ∆ 8 SHELLY MANBER
0 0 4 0 Notice that if Λ = αΛ for some α ∈ C then g2(Λ ) = g2(Λ)/α and g3(Λ ) = 6 0 12 0 g3(Λ)/α . This means that ∆(Λ ) = ∆(Λ)/α and j(Λ ) = j(Λ). So homothetic lattices have the same j-invariant. Let E be an elliptic curve viewed as a projective variety over C. For each lattice ∼ 2 3 2 3 Λ such that C/Λ = E, the equation y z = 4x − g2(Λ)xz − g2(Λ)z gives an expression defining E in P2(C). Since all such lattices are homothetic, there is a unique value j(Λ) associated to E. We denote this j(E) and call it the j-invariant of E. If τ ∈ C we define j(τ) to be j(hτ, 1i). We have j(τ) = j(τ 0) whenever τ ≡ τ 0 mod SL2(Z).
Now that we have an analytic and algebraic structure of elliptic curves, we move on to understanding the maps between elliptic curves. An isogeny of elliptic curves is a morphism between elliptic curves as projective varieties that sends the distinguished point of one curve to the distinguished point of the other. Every isogeny is in particular an analytic map between elliptic curves as analytic manifolds. Theorem 3.1. Let E,E0 be elliptic curves and Λ, Λ0 their associated lattices. (1) Assigning to each isogeny E → E0 the induced analytic map C/Λ → C/Λ0 gives a bijection of sets 0 0 {isogenies ϕ : E → E } ↔ {nontrivial holomorphic maps φ : C/Λ → C/Λ with φ(0) = 0}. (2) Sending α ∈ C to the map φα : C → C z 7→ αz, gives a bijection 0 0 {α ∈ C|αΛ ⊂ Λ } ↔ {nontrivial holomorphic maps φ : C/Λ → C/Λ with φ(0) = 0}. Proof. See [Sil86, Chapter VI, Theorem 4.1]. We have two immediate corollaries to Theroem 3.1. Corollary 3.2. Every isogeny between elliptic curves ϕ : E → E0 is given analyti- cally by a map of the form 0 ϕ : C/Λ → C/Λ z 7→ αz for some α ∈ C. Corollary 3.3. Two elliptic curves are isomorphic as algebraic varieties if and only if they are isomorphic as analytic manifolds. Finally, Theorem 3.1 combined with some analytic results on the j-invariant gives the following:
Corollary 3.4. Two elliptic curves E and E0 are isomorphic over C if and only if j(E) = j(E0). Proof. Since j(Λ) = j(Λ0) whenever Λ = αΛ0, we can define j as a function on L. Associating each element of L with a point on H/SL2(Z), we can consider j as a map from H/SL2(Z) to C. Proposition 5 of Chapter VII of Serre [Ser73] shows AN ELEMENTARY AND COMPUTATIONAL APPROACH TO HEEGNER POINTS 9
0 that j defines a bijection in this way between H/SL2(Z) and C. So j(E) = j(E ) if and only if j(Λ) = j(Λ0) if and only if Λ and Λ0 are homothetic, if and only if E 0 and E are isomorphic (Corollary 3.3 and Claim 2.1).
0 0 Note that Corollary 3.4 implies that j(τ) = j(τ ) if and only if τ ≡ τ mod SL2(Z).