14 Dual isogenies and the Weil pairing

Let K be a perfect field and EK an elliptic curve.

Proposition 14.1. Assuming that:

  • ΦE(K¯) be a finite Gal(K¯K)-stable subgroup

Then there exists an elliptic curve EK and a separable isogeny ϕ:EE defined over K, with kernel Φ, such that every isogeny ψ:EE with Φψ factors uniquely via ϕ:
 E                E′′


ψϕ∃ unique  E ′

Proof. Omitted (see Silverman, Chapter III, Proposition 4.12).

Proposition 14.2. Assuming that:

  • ϕ:EE an isogeny of degree n

Then there exists a unique isogeny ϕ^:EE such that ϕ^ϕ=[n].

Proof. Case ϕ is separable: We have |kerϕ|=n, hence kerϕE[n]. Apply Proposition 14.1 with ψ=[n].

Case ϕ is inseparable: omitted.

Uniqueness: Suppose ψ1ϕ=ψ2ϕ=[n]. Then (ψ1ψ2)ϕ=0, so deg(ψ1ψ2) deg ϕ=0. Then deg(ψ1ψ2)=0, hence ψ1=ψ2.

Remark.

  • (i) Write E1E2 to mean “E1 and E2 are isogenous”. Then is an equivalence relation.
  • (ii) deg[n]=n2 gives that degϕ^= deg ϕ and [n]^=[n].
  • (iii) Note:
    ϕϕ^ϕ=ϕ[n]E=[n]Eϕ

    Hence ϕϕ^=[n]E.

    In particular, ϕ^^=ϕ.

  • (iv) If EψEϕE then ϕψ^=ψ^ϕ^.
  • (v) If ϕEnd(E) then
    ϕ2(Trϕ)ϕ+ deg ϕ=0.

    So

    ([Trϕ]ϕ)ϕ=ϕ^=[ deg ϕ].

    Hence [Trϕ]=ϕ+ϕ^.

Lemma 14.3. Assuming that:

  • ϕ,ψHom(E,E)

Then ϕ+ψ^=ϕ^+ψ^.

Proof.

  • (i) If E=E, then this follows from Tr(ϕ+ψ)=Tr(ϕ)+Tr(ψ).
  • (ii) In general, let α:EE be any isogeny (e.g. ϕ^). Then: (αϕ+αψ)^=αϕ^+αψ^(ϕ+ψ)^α^=(ϕ^+ψ^)α^

    Hence the result follows.

Remark. In Silverman’s book he proves Lemma 14.3first, and uses this to show deg:Hom(E,E) is a quadratic form.

Notation.

sum:Div(E)Enp(P)formal sumnP(P)addusinggrouplaw

Recall

E Pic0(E)P[(P)(0)]

Therefore sumD[D] for all DDiv0(E).

We deduce:

Lemma 14.4. Assuming that:

  • DDiv(E)

Then D0 if and only if degD=0 and sumD=0E.

We will now discuss Weil pairing.

Let ϕ:EE be an isogeny of degree n, with Dual isogeny ϕ^:EE. Assume charKn (hence ϕ, ϕ¯ separable).

We define the Weil pairing

eϕ:E[ϕ]×E[ϕ^]μn.

Let TE[ϕ]. Then nT=0, so there exists fK¯(E) such that

div(f)=n(T)n(0).

Pick T0E(K¯) with ϕ(T0)=T. Then

ϕ(T)ϕ(0)=PE[ϕ](P+T0)PE[ϕ](P)

has sum

nT0=ϕ^ϕT0=ϕ^T=0.

So there exists gK¯(E) such that

div(g)=ϕ(T)ϕ(0).

Now

div(ϕf)=ϕ(divf)=n(ϕ(T)ϕ(0))=div(gn)

Therefore ϕf=cgn for some cK¯.

Rescaling f, we can say without loss of generality c=1, i.e. ϕf=gn.

For SE[ϕ] we get τS(divg)=(divg) so τSg=ζg for some ζK¯, i.e. ζ=g(X+S)g(X) is independent of choice of XE(K¯).

Now

ζn=g(X+S)ng(X)n=f(ϕ(X+S))f(ϕ(X))=1

since SE[ϕ]. Hence ζμn.

We define

eϕ(S,T)=g(X+S)g(X).

Proposition 14.5. e is bilinear and non-degenerate.

Proof.

  • (i) Linearity in first argument: e(S1+S2,T)=g(X+S1+S2)g(X+S2)g(X+S2)g(X)=e(S,T)e(S2,T)
  • (ii) Linearity in second argument: Let T1,T2E[ϕ^]. div(f1)=n(T1)n(0)ϕf=g1ndiv(f2)=n(T2)n(0)ϕf2=g2n

    There exists hK¯(E) such that

    div(h)=(T1)+(T2)(T1+T2)(0).

    Then put f=f1f2hn and g=g1g2ϕh.

    Check: div(f)=n(T1+T2)n(0). Yes.

    ϕf=ϕf1ϕf2(ϕh)n=(g1g2ϕ(h))n=gn.

    Therefore

    e(S,T1+T2)=g(X+S)g(X)=g1(X+S)g1(X)g2(X+S)g2(X)h(ϕ(X))h(ϕ(X+S))=1since SE[ϕ]=e(S,T1)e(S,T2)
  • (iii) e is non-degenerate. Fix TE[ϕ^]. Suppose e(S,T)=1 for all SE[ϕ]. Then τSg=g for all SE[ϕ].

    Have K¯(E)ϕK¯(E) is a Galois extension with Galois group E[ϕ] (SE[ϕ] acts on K¯(E) via τS).

    Therefore g=ϕh for some hK¯(E). So ϕf=gn=ϕ(hn). So f=hn, and hence div(h)=(T)(0). So T=0.

We’ve shown E[ϕ]Hom(E[ϕ],μn). It is an isomorphism since #E[ϕ]=#E[ϕ^]=n.

Remark.

  • (i) If E,E,ϕ are defined over K then e is Galois equivariant, i.e. σGal(K¯K), SE[ϕ], TE[ϕ],
    e(σS,σT)=σ(e(S,T)).
  • (ii) Taking ϕ=[n]:EE (so ϕ^=[n]) gives
    en:E[n]×E[n]μn2μn.

Corollary 14.6. Assuming that:

  • E[n]E(K)

Then μnK.

Proof. Let TE[n] have order n. Non degeneracy of en implies that there exists SE[n] such that en(S,T) is a primitive n-th root of unity, say ζn.

Then

σ(ζn)=σ(en(S,T))=en(σS,σT)=ζn

for all σGal(K¯K). So ζnK.

Example. There does not exist E with E()tors(3)2.

Remark. In fact, en is alternating, i.e. en(T,T)=1 for all TE[n].

(This implies en(S,T)=en(T,S)1).