We give an example that combines functionality for small modular curves with Magma's machinery for
curve quotients and elliptic curves to compute the j-invariant of a special class of elliptic curves
over Q up to quadratic twist. This class of curves arose in Wiles' original paper on Fermat's
Last Theorem and the Shimura-Tanayama-Weil (STW) conjecture.
In the Breuil-Conrad-Diamond-Taylor-Wiles proof of the STW conjecture, a couple of special cases
arise that lead to a finite number of classes of elliptic curves up to bar(Q)-isomorphism
that have to be determined and then checked directly for modularity. If E[p] is the p-torsion
subgroup of E considered as a (Gal)(bar(Q)/(Q))-module, these are cases where
E[3] and E[5] are both reducible,
E[3] is reducible and E[5] is irreducible but absolutely reducible as a
(Gal)(bar(Q)/(Q)(Sqrt(5)))-module,
or E[5] is reducible and E[3] is irreducible but absolutely reducible as a
(Gal)(bar(Q)/(Q)(Sqrt( - 3)))-module. The first case corresponds to non-cuspidal rational points
on X
0(15) and leads, up to isogeny, to one class of twists with j= - 25/2 as may easily be verified
using the intrinsics here. The second case corresponds
to non-cuspidal rational points on a quotient of X
0(75) and leads to curves with j=0.
We consider the third case in this example, showing
that, up to isogeny, there is one class (up to quadratic twist) with j-invariant (11/2)
3.
The curves we want are precisely those that have a cyclic subgroup of order 5 rational over Q
and for which the image of (Gal)(bar(Q)/(Q)) in GL2((F)3), giving the action on E[3],
is the normaliser of a split Cartan subgroup.
Curves just satisfying the E[3] condition are classified by non-cuspidal rational points on X0(9)/w9
that are not the image of a rational point of X0(9). Such a point p lifts to a point p1 on X0(9)
defined over some quadratic field K such that p1σ = w9(p1) where σ is the nontrivial
automorphism of K over Q. If (E, C) is an elliptic curve/K with cyclic subgroup C=< P >
rational over K that is represented by the moduli point p1, then E1=E/< 3P > can be defined
over Q so that the order 3 subgroups C/< 3P > and (C/< 3P >)σ generate
E1[3], are rational over K and are swapped by σ.
In the same way, adding the extra condition that there is a rational subgroup of order 5 leads to curves
classified by non-cuspidal rational points of X0(45)/w9. These do not lift to rational points because
X0(45) has no non-cuspidal rational points. X0(45)/w9 is Q-isogenous to the rank 0 elliptic
curve X0(15) so only has finitely many rational points. If p1 is a lift, the image of p1
under the 3-projection map X0(45) -> X0(5) (z |-> 3z on complex points) is a rational
point that corresponds to the desired curve E1 (with its cyclic 5-subgroup). We could apply
the 3-projection all the way down to X0(1) of course, but it is slightly more efficient to just project
to level 5 and remove duplicate images before computing j-invariants.
> X45<x,y,z> := SmallModularCurve(45);
> w9 := AtkinLehnerInvolution(X45,45,9);
> G := AutomorphismGroup(X45,[w9]);
> C,prjC := CurveQuotient(G);
> c_inf := Cusp(X45,45,45);
> ptE := prjC(c_inf);
> E1,mp1 := EllipticCurve(C,ptE);
> E,mp2 := MinimalModel(E1);
> prjE := Expand(prjC*mp1*mp2);
E is a minimal model for X0(45)/w9 and prjE is the projection from X0(45) to E that takes
the cusp ∞ to the zero point of E. We compute the image of the cusps under prjE so as to
discount these when we find all the rational points on E. Since p and w9(p) map to the same point
under prjE, we only need consider cusps up to w9 equivalence. The non-rational conjugate cusps
∓ 1/3 are defined over (Q)(Sqrt( - 3)) and are swapped by w9, so map to the same rational
point. The same holds for the cusps ∓ 1/15.
> i0 := prjE(Cusp(X45,45,1));
> K := QuadraticField(-3);
> c3 := Cusp(X45,45,3);
> c3p := X45(K)!Representative(Support(c3,K));
> i3 := E!(prjE(c3p));
> c15 := Cusp(X45,45,15);
> c15p := X45(K)!Representative(Support(c15,K));
> i15 := E!(prjE(c15p));
> Ecusps := [E!0,i0,i3,i15];
E has rank zero, so we can recover all of its rational points using TorsionSubgroup.
There are 8 in all, 4 of which are images of cusps. We find the 4 non-cuspidal ones and
compute the pullbacks of them to X0(45). In each case, the pull back is a pair of
conjugate points defined over a quadratic field. It is easiest to work with places here
(and we don't have to worry about the base scheme of prjE). The pullback of
each point gives a single place corresponding to the two conjugate points.
> T,mp := TorsionSubgroup(E);
> Epts := [mp(g) : g in T];
> Eptsnc := [P : P in Epts | P notin Ecusps];
> plcs := [Support(Pullback(prjE,Place(p)))[1] : p in Eptsnc];
We now perform the 3-projection to X0(5) on these places and discard duplicate
images.
> X5 := SmallModularCurve(5);
> prj3 := ProjectionMap(X45,45,X5,5,3);
> prj3 := Expand(prj3);
> plcs5 := [Support(Pushforward(prj3,p))[1] : p in plcs];
> plcs5 := Setseq(Seqset(plcs5));
We compute the j-invariants of the two images and verify that they
represent isogenous curves under the cyclic 5-isogeny coming from the rational
5-subgroup (i.e. they are images of each other under w5).
> js := [jInvariant(p,5) : p in plcs5];
> js;
[ -1680914269/32768, 1331/8 ]
> w5 := AtkinLehnerInvolution(X5,5,5);
> Pullback(w5,plcs5[1]) eq plcs5[2];
true
Finally, we can use Magma intrinsics to check that the elliptic curves
with these j-invariants actually do satisfy the property that the image of
the action of Galois on 3-torsion is the normaliser of a split Cartan
subgroup (D8). As this property remains true for quadratic twists and
is unchanged by images under a rational 5-isogeny, we only need check it
for one curve over Q with one of the j-invariants. We also check
that a minimal twist has conductor 338 from which modularity can be
checked explicitly.
> Ej := EllipticCurveWithjInvariant(js[2]);
> Ej := MinimalModel(MinimalTwist(Ej));
> Conductor(Ej);
338
> ThreeTorsionType(Ej);
Dihedral
[Next][Prev] [Right] [Left] [Up] [Index] [Root]