Elliptic Curves of Odd Modular Degree

Frank Calegari * * Supported in part by the American Institute of Mathematics

Matthew Emerton † † Supported in part by NSF grant DMS-0401545

November 27, 2006

1 Introduction

Let E   be an elliptic curve over Q   of conductor N   . Since E   is modular [3] , there exists a surjective map π : X 0 ( N ) → E   defined over Q   . There is a unique such map of minimal degree (up to composing with automorphisms of E   ), and its degree m E   is known as the modular degree of E   . In [24] the parity of m E   was determined for a very particular explicit class of elliptic curves, namely, the Neumann–Setzer curves, which have prime conductor and a rational 2   -torsion point. (See also the remark following theorem  5.1 below.) We study the question more generally for arbitrary elliptic curves E / Q   , and prove the following theorem:
1.1 Theorem. If E / Q   is an elliptic curve of odd modular degree then:
  • 1. the conductor N   of E   is divisible by at most two odd primes,
  • 2. E   is of even analytic rank, and
  • 3. either (a) E   has a rational point of order 2   (or equivalently, admits a rational 2   -isogeny ), (b) E   has prime conductor and supersingular reduction at 2   , and Q ( E [ 2 ] )   is totally complex (equivalently, E ( R )   is connected ), or (c) E   has complex multiplication, and N = 27 , 32 , 49 , or 243   .
1.2 Example. The following examples of elliptic curves with odd modular degree should serve to illustrate conditions (3a), (3b) and (3c). The curve X 0 ( 15 )   has modular degree one and a rational two torsion point, and thus satisfies condition (3a). Another example is given by the curve y 2 + x y = x 3 − x 2 − 58 x − 105   ( 2537 E   in Cremona's tables) of conductor 43 ⋅ 59   with modular degree 445   and torsion subgroup Z / 4 Z   . The curves X 0 ( 11 )   and X 0 ( 19 )   both have modular degree one and satisfy condition (3b). An example of larger conductor is given by y 2 + y = x 3 + x 2 − 4 x − 10   of conductor 24859   and modular degree 3979   . Finally, there are exactly four curves of odd modular degree with complex multiplication, namely X 0 ( 27 )   , X 0 ( 32 )   , X 0 ( 49 )   (all of modular degree one) and y 2 + y = x 3 + 2   of modular degree 9   , conductor 243   and j   -invariant 0   .
1.3 Remark. Each of the conditions appearing in theorem  1.7 is invariant under isogeny, other than the condition that E ( R )   be connected, which however is invariant under isogenies of odd degree. Since the modular parameterization of E   factors through the optimal member of the isogeny class of E   (that is, the member of its isogeny class having minimal modular degree; in older terminology, a strong Weil curve), it is therefore no loss of generality in the proof of theorem  1.1 to assume that E   is optimal.
1.4 Remark. Cremona and Watkins have computed the modular degree of all optimal elliptic curves of conductor ≤   25,000 [7] and these computations suggest that there may be even stronger limitations on the conductor of a curve of odd modular degree than those imposed by theorem  1.1 . Indeed, in the range of Watkins' computations, every curve of odd modular degree has conductor divisible by at most two primes, and the conductor always has one of the following forms: 2 p   , 4 p   , or p q   , where p   and q   are odd primes.
1.5 Remark. The statement of the theorem regarding the analytic rank of E   is consistent with the rank conjecture of Watkins [27,conj. 4.1] (and the conjecture of Birch and Swinnerton-Dyer).
The following result, conjectured by Watkins ([24,conj. 4.3] , [27,conj. 4.2] ), is a simple consequence of theorem  1.1 (see lemma  3.5 ):
1.6 Theorem. Let E / Q   be an elliptic curve of prime conductor N   , and suppose that E   is neither a Neumann–Setzer curve, nor X 0 ( 17 )   (equivalently, E   does not have a rational 2   -torsion point ). If m E   is odd, then N ≡ 3 m o d 8   .
One technique for proving that an elliptic curve E   has even modular degree is to show that the map π   factors through X 0 ( N ) / w   for some non-trivial Atkin–Lehner involution w   .
We use this approach in section 2 to prove theorem  2.1 , which in turn implies parts (1) and (2) of theorem  1.1 , and shows that (3a) holds if N   is divisible by at least two primes.
It remains to prove (3) in the case when N   is a prime power. The most difficult case to handle is when N   is actually prime, and in this case we deduce theorem  1.1 from the following result, proved in section 3.
1.7 Theorem. Let N   be prime, let T   denote the Hecke algebra over Z   acting on weight two cuspforms on Γ 0 ( N )   , and let m   be a maximal ideal of T   such that T / m = F 2   , and such that the associated semi-simple Galois representation ρ ¯ m : G Q → G L 2 ( F 2 )   is irreducible. If the completion T m = Z 2   , then
  • 1. ρ ¯   is supersingular at 2   , and
  • 2. ρ ¯   is totally complex.
The relevance of this result to theorem  1.1 is that, since N   is prime in the context of theorem  1.7 , a result of Ribet [30] shows (assuming, as we may, that E   is optimal) that the modular degree of E   is even if and only if 2   is a congruence prime for the newform of level N   attached to E   .
The proof of theorem  1.7 is motivated by the following considerations: If p   is an odd prime and ρ ¯ : G Q → G L 2 ( F p )   is a surjective modular representation, then theorems of Wiles and Taylor–Wiles [29, 26] show that the universal minimal deformation ring R ∅   attached to ρ ¯   is isomorphic to the universal minimal modular deformation ring T ∅   ( = T m   , since N   is prime). Since T ∅   is a finite W ( F p ) = Z p   algebra with residue field F p   , it is exactly equal to Z p   if and only if it is an étale Z p   -algebra. On the other hand, since R ∅ ∼ = T ∅   , this is equivalent to R ∅   being étale over Z p   , which is in turn equivalent to the reduced Zariski cotangent space of R ∅   being trivial. Since by construction R ∅   represents the minimal deformation functor, its reduced Zariski cotangent space considered as a set has cardinality equal to the number of minimal deformations ρ : G Q → G L 2 ( F p [ x ] / ( x 2 ) )   of ρ ¯   . Thus to prove that T m ≠ Z p   it suffices to show that there exists a non-trivial minimal deformation of ρ ¯   to G L 2 ( F p [ x ] / ( x 2 ) )   .
In spirit, the proof of theorem  1.7 follows this strategy; in other words, we determine whether or not T m = Z 2   by a calculation on tangent spaces. A significant problem arises, however, since we are working in the case p = 2   , whilst the method of Wiles and Taylor–Wiles applies only to p > 2   . This is not a mere technical obstruction; many phenomena can occur when p = 2   that do not occur for odd p   . To name two such: the possible failure of T m   to be Gorenstein and the consequent failure of multiplicity one [16] , and the fact that ρ ¯   can arise from a totally real extension of Q   . Calculations in the second case suggest that the Taylor–Wiles strategy for proving R = T   in the minimal case will not work without some significant new idea, since the numerical coincidences that occur for odd p   whilst balancing the Selmer and dual Selmer groups in the Greenberg–Wiles product formula (see for example the remarks of de Shalit [8] , top of p. 442) do not occur in the case p = 2   . Mark Dickinson [10] has proved an R = T   theorem for p = 2   ; however, his result requires many non-trivial hypotheses, and indeed does not apply to any of the representations considered in theorem  1.7 . (Its main application to date has been to representations with image S L 2 ( F 4 ) ≃ A 5   .) The issue here is that the Taylor–Wiles auxiliary prime arguments fail when p = 2   and the image of ρ ¯   is dihedral.
Thus, instead of appealing to any general modularity results, we show that T m   is bigger than Z 2   by explicitly constructing (in certain situations) non-trivial deformations of ρ ¯ m   to F 2 [ x ] / ( x 2 )   that are demonstrably modular (and hence contribute to the reduced cotangent space of T m   ). The most difficult point is to show that these deformations are modular of the correct (minimal) level. We prove this via a level-lowering result for modular forms with values in Artinian Z 2   -algebras (theorem  3.18 below). This level lowering result maybe of independent interest; for example, it provides evidence that an R = T   theorem should hold for those ρ ¯   of characteristic two to which it applies.
The proof of (3) when N   is a prime power (but is not actually prime) is given in section 4. In section 5 we make some concluding remarks.
Let us close this introduction by pointing out that recently Dummigan [11] has provided a heuristic explanation for Watkins' rank conjecture that also relies on a hypothetical R = T   theorem for the residual Galois representation ρ ¯   arising from the 2   -torsion on an elliptic curve E   : he uses the symmetric square map from ρ ¯   to S y m 2 ρ ¯   to lift elements from 2   -Selmer group of E   to the tangent space to the deformation ring of ρ ¯   . He also shows that the resulting tangent space elements can never be “trapped” (in the words of [29,p. 450] ) by the Taylor–Wiles method of introducing auxiliary primes.
Altogether, the experimental work of Watkins on the parity of modular degrees, taken together with the results of this paper and of [11] , suggests the validity of an R = T   theorem for (at least certain) residual Galois representations arising from the 2   -torsion on elliptic curves, a theorem whose proof, however, seems out of reach of the current techniques.

2 N   composite with at least two distinct prime factors

In this section we prove the following theorem.
2.1 Theorem. If E   is an elliptic curve of odd modular degree then the conductor N   of E   is divisible by at most two odd primes, and E   is of even analytic rank. Furthermore, if N   is divisible by at least two primes, then E   contains a rational 2   -torsion point.
We begin with a preliminary lemma. Let E   be an elliptic curve over a field k   ; let O   denote the origin of E   . We let A   denote the group of automorphisms of E   as a curve over k   (i.e. k   -rational automorphisms of E   that do not necessarily fix O   ), and suppose that W   is a finite elementary abelian 2   -subgroup of A   .
2.2 Lemma. The order of W   divides twice the order of E [ 2 ] ( k )   .
Proof. Let A 0   denote the subgroup of A   consisting of automorphisms of E   as an elliptic curve over k   (i.e. k   -rational automorphisms of E   that do fix O   ). The action of E ( k )   on E   via translation realizes E ( k )   as a normal subgroup of A   which has trivial intersection with A 0   , and which together with A 0   generates A   . Thus A   sits in the split short exact sequence of groups
0 → E ( k ) → A → A 0 → 1 . (1)
(This is of course well known. The surjection A → A 0   may also be regarded as the map A = A u t ( E ) → A u t ( P i c 0 ( E ) )   induced by the functoriality of the formation of Picard varieties — the target being the group of automorphisms of P i c 0 ( E )   as a group variety — once we identify E   and P i c 0 ( E )   as group varieties in the usual way.) The short exact sequence ( 1 ) induces a short exact sequence 0 → W ∩ E ( k ) → W → W 0 → 1 ,   where W 0   denotes the projection of W   onto A 0   . The known structure of A 0   shows that W 0   is either trivial or of order 2   . Since W ∩ E ( k ) ⊂ E [ 2 ] ( k ) ,   the lemma follows. □  
Proof of theorem  2.1 . The discussion of remark  1.3 shows that it suffices to prove the theorem under the additional assumption that E   is an optimal quotient of X 0 ( N )   .
Let W   denote the group of automorphisms of X 0 ( N )   generated by the Atkin–Lehner involutions; W   is an elementary abelian 2   -group of rank equal to the number of primes dividing N   . Since E   is an optimal quotient of X 0 ( N )   , the action of W   on X 0 ( N )   descends to an action on E   . If w ∈ W   were to act trivially on E   , then the quotient map X 0 ( N ) → E   would factor through X 0 ( N ) / w   , contradicting our assumption that E   is of odd modular degree. Thus lemma  2.2 shows that W   has order at most 8   , and hence that N   is divisible by at most 3   primes. Furthermore, if N   is divisible by more than one prime, then it shows that E [ 2 ] ( Q )   is non-trivial. Suppose now that N   is odd, so that X 0 ( N )   and E   both have good reduction at 2   . We may then apply the argument of the preceding paragraph over F 2   , and so conclude from lemma  2.2 that W   has order at most 4   . Hence N   is divisible by at most two primes.
If E   is of odd analytic rank, and if f E   denotes the normalized newform of level N   attached to E   , then w N f E = f E   , and so the automorphism of E   induced by w N   has trivial image in A 0   . Thus w N   acts on E   via translation by an element P ∈ E ( Q )   . Since w N   interchanges the cusps 0   and ∞   on X 0 ( N )   , we see that P = π ( 0 ) − π ( ∞ )   (where π : X 0 ( N ) → E   is a modular parameterization of E   , chosen so that π ( ∞ ) = O   ).
The assumption that E   has odd analytic rank also implies that L ( f E , 1 ) = 0   . Since this L   -value can be computed (up to a non-zero factor) by integrating f E   from 0   to ∞   in the upper half-plane, we conclude that P = O ,   and thus that w N   acts trivially on E   . Hence π   factors through the quotient X 0 ( N ) / w N   of X 0 ( N )   , and so must be of even modular degree, a contradiction. □  

3 N   prime

3.1 Reductions

3.2 Lemma. Theorem  1.7 implies part (3) of theorem  1.1 for N   prime.
Proof. Suppose that E   is an elliptic curve of conductor N   , assumed to be optimal in its isogeny class. Let f E   be the associated Hecke eigenform of level Γ 0 ( N )   and weight 2   . From a theorem of Ribet [30] , 2 | m E   if and only if f E   satisfies a congruence with another cuspidal eigenform of level N   . The set of cuspidal eigenforms (in characteristic zero) congruent to f   is indexed by H o m ( T m ⊗ Q 2 , Q ¯ 2 )   . Thus f E   satisfies no non-trivial congruences if and only if T m ⊗ Q 2 = Q 2   , or equivalently if and only if T m = Z 2   . □  
The following lemma shows that it is possible to detect congruences from modular forms modulo two.
3.3 Lemma. Let m   be the maximal ideal in T   associated to ρ ¯   , and suppose there exist two distinct non-zero q   -expansions f   , g   with coefficients in F 2   such that m 2 f = m 2 g = 0   . Then T m ≠ Z 2   .
Proof. Since ρ ¯   is irreducible, f   and g   are both cuspidal. If T m = Z 2   , then ( m 2 , 2 ) = m   . Thus f   and g   are both q   -expansions which are killed by m   . By multiplicity one for q   -expansions [18,prop. 9.3] it follows that f = g   . □  
Of use to us will be the following theorem of Grothendieck on Abelian varieties with semistable reduction [15,ExposéIX,prop. 3.5] :
3.4 Theorem (Grothendieck). Let A   be an Abelian variety over Q   with semistable reduction at ℓ   . Let I ℓ ⊂ G a l ( Q ¯ / Q )   denote a choice of inertia group at ℓ   . Then the action of I ℓ   on the p n   -division points of A   for p ≠ ℓ   is rank two unipotent; i.e., as an endomorphism, for σ ∈ I ℓ   , ( σ − 1 ) 2 A [ p n ] = 0 .   In particular, I ℓ   acts through its maximal pro- p   quotient, which is procyclic.
Shimura proved that a modular form f   of weight 2   and level Γ 0 ( N )   gives rise to a modular Abelian variety A f   in such a way that the p   -adic representations ρ f   attached to f   arise from the torsion points of A f   . For prime N   , these varieties are semistable at N   , and so we may apply the theorem above to deduce that for p = 2   , such representations ρ   restricted to I N   factor through a pro-cyclic 2   -group. For representations ρ ¯   with image inside G L 2 ( F 2 ) ≃ S 3   , this means in particular that the order of inertia at N   is either 1   or 2   .
Let us now consider a Galois representation ρ ¯ : G Q → G L 2 ( F 2 )   , arising from a cuspidal Hecke eigenform of level N   , whose image is not contained in a Borel subgroup. (This is equivalent to ρ ¯   being irreducible, and also to the image of ρ ¯   having order divisible by 3   .) Let L   be the fixed field of the kernel of ρ ¯   ; the extension L / Q   is unramified outside 2   and N   .
If L / Q   is unramified at N   then ρ ¯   has Serre conductor 1   , contradicting a theorem of Tate[25] . Thus by the discussion above we see that inertia at N   factors through a group of order 2   , that L / Q   is an S 3   -extension, and (hence) that ρ ¯   is absolutely irreducible. Let K / Q   be a cubic subfield of L   , and let F   be the quadratic extension inside L   . Since ρ ¯   is finite flat at 2   , it follows from Fontaine's discriminant bounds [13] that the power of 2   dividing the discriminant of F / Q   is at most 4   . Thus F / Q   must be Q ( ± N )   (as it is ramified at N   ).
The extension L / F   is unramified at the prime above N   and is ramified at 2   if and only if ρ ¯   is supersingular.
3.5 Lemma. If ρ ¯   is supersingular at two and not totally real then N ≡ 3 m o d 8   . In particular, theorem  1.1 implies theorem  1.6 .
Proof. By class field theory the quadratic field F / Q   admits a degree three extension ramified precisely at 2   only if 2   is unramified and inert in F   . This occurs if and only if N ≡ 3 m o d 8   and F = Q ( − N )   , or N ≡ 5 m o d 8   and F = Q ( N )   . Moreover if F = Q ( N )   then K   and L   are totally real. □  
We shall prove theorem  1.7 by showing in the following subsections that if ρ ¯   satisfies at least one of the following conditions:
  • 1. ρ ¯   is totally real;
  • 2. ρ ¯   is unramified at 2   ;
  • 3. ρ ¯   is ordinary, complex, and ramified at 2   ;
then T m ≠ Z 2   .

3.6 ρ ¯   is totally real

The theory of modular deformations is not well understood when ρ ¯   is totally real. Thus our arguments in this section are geometric. We use the following theorem, due to Merel [20,prop. 5] . (This interpretation of Merel's result is due to Agashe [1,cor. 3.2.9] ).
3.7 Theorem. Let N   be prime. Then J 0 ( N ) ( R )   is connected.
If we let g   denote the dimension of J : = J 0 ( N )   it follows that J ( R ) ≃ ( R / Z ) g   , J ( R ) t o r s ≃ ( Q / Z ) g   and J [ 2 ] ( R ) = ( Z / 2 Z ) g   .
Let J [ 2 ∞ ] = lim ⟶ J [ 2 m ]   . Then J [ 2 ∞ ]   is a 2   -divisible group over Q   admitting an action of T 2   .
Since T 2   is finite and flat over the complete local ring Z 2   there exists a decomposition T 2 = ∏ T m ,   where the product is taken over the maximal ideals m   of T   of residue characteristic two.
If g ( m )   denotes the rank of T m   over Z 2   , then ∑ m g ( m ) = r a n k ( T 2 / Z 2 ) = g .   If J [ m ∞ ] : = J [ 2 ∞ ] ⊗ T 2 T m   , then J [ 2 ∞ ] ≃ ∏ J [ m ∞ ]   (compare [18]  §7, p. 91). From lemma 7.7 of [18] we see that T a m J ⊗ Q 2   is free of rank two over T m ⊗ Q 2   (where T a m J : = H o m ( Q 2 / Z 2 , J [ m ∞ ] ( Q ¯ ) )   is the m   -adic Tate module of J   ), and thus that
J [ m ∞ ] ( C ) ∼ = ( Q 2 / Z 2 ) 2 g ( m ) . (2)
Let J [ 2 ] m : = J [ 2 ] ⊗ T 2 T m   be the 2   -torsion subgroup scheme of J [ m ∞ ]   .
3.8 Lemma. For all maximal ideals m   of residue characteristic two there is an equality dim Z / 2 Z ( J [ 2 ] m ( R ) ) = g ( m )  
Proof. The isomorphism ( 2 ) induces an isomorphism J [ 2 ] m ( C ) ∼ = ( Z / 2 Z ) 2 g ( m ) .   Let σ ∈ G a l ( C / R )   denote complex conjugation. Then ( σ − 1 ) 2 J [ 2 ] m ( C ) = 0   . Thus J [ 2 ] m ( R )   (which is the kernel of σ − 1   ) has dimension at least g ( m )   . If dim Z / 2 Z ( J m [ 2 ] ( R ) ) > g ( m )   for some m   , then since J [ 2 ] ( R ) = ∏ J [ 2 ] m ( R ) ,   and since (as was noted above) dim Z / 2 Z ( J [ 2 ] ( R ) ) = g   , we would deduce the inequality:
g = ∑ dim Z / 2 Z ( J [ 2 ] m ( R ) ) > ∑ m g ( m ) = g ,   which is absurd. □  
Now let ρ ¯   be a totally real (absolutely) irreducible continuous modular representation of G a l ( Q ¯ / Q )   into G L 2 ( F 2 )   of level Γ 0 ( N )   , and let m   be the corresponding maximal ideal of T   . The main result of [2] shows that the G a l ( Q ¯ / Q )   -representation J [ m ] ( Q ¯ )   is a direct sum of copies of ρ ¯   . Thus, since ρ ¯   is totally real, we find that dim Z / 2 Z J [ m ] ( R ) = dim Z / 2 Z J [ m ] ( C ) ≥ dim Z / 2 Z ρ ¯ = 2 .   Combining this inequality with the inclusion J [ m ] ( R ) ⊆ J [ 2 ] m ( R )   and lemma  3.8 we find that g ( m ) ≥ 2   , and thus that T m ≠ Z 2   .

3.9 ρ ¯   is unramified at 2  

Suppose that L / Q   is unramified at 2   . This forces ρ ¯   to be ordinary. By the theory of companion forms [14] one expects that ρ ¯   arises from a mod 2   form of level Γ 1 ( N )   and weight 1   . Although the results of [14] do not apply in this case, Wiese [28] explicitly constructs such forms when the image of ρ ¯   is dihedral, as it is in our situation. (In fact the only difficult point of Wiese's construction is the case when ρ ¯   is totally real, and this case of theorem  1.7 is already covered by section  3.6 .) Let f   be this companion (Katz) modular form of weight one modulo 2   . Let A   be the Hasse invariant modulo 2   , which is a modular form of level one with q   expansion given by 1   . Then A f = f   and g = f 2   are two distinct q   -expansions modulo 2   of weight 2   and level N   . Moreover, one sees that ( T ℓ − a ℓ ) f = ( T ℓ − a ℓ ) g = 0   for all odd ℓ   , and that ( T 2 − a 2 ) f = 0   and ( T 2 − a 2 ) 2 g = 0   . Thus m 2 f = m 2 g = 0   and therefore by lemma  3.3 , T m ≠ Z 2   and we are done.

3.10 ρ ¯   is ordinary, complex, and ramified at 2  

Suppose that ρ ¯   is ordinary, complex, and ramified at 2   . It follows that F / Q   is complex and ramified at 2   , and thus that F = Q ( − N )   for some N ≡ 1 m o d 4   . Moreover, the extension L / F   is unramified everywhere. Since N ≡ 1 m o d 4   it follows that H : = L ( − 1 )   is also unramified everywhere over F   . The field H   is Galois over Q   , and clearly
G a l ( H / Q ) ≃ S 3 × Z / 2 Z . (3)
We may embed S 3 × Z / 2 Z   into G L 2 ( F 2 [ x ] / ( x 2 ) )   by fixing an identification of S 3   with G L 2 ( F 2 )   , and mapping a generator of Z / 2 Z   to the matrix ( 1 + x 0 0 1 + x ) .   Composing the isomorphism ( 3 ) with this embedding yields a representation:
ρ : G a l ( H / Q ) ↪ G L 2 ( F 2 [ x ] / ( x 2 ) ) .   The representation ρ   has trivial determinant (equivalently, determinant equal to the mod 2   cyclotomic character). We also claim that ρ   is finite flat at two. To check this, it suffices to prove this over Z 2 u r   . The representation ρ | G a l ( Q ¯ 2 / Q 2 u r )   factors through a group of order 2   , and one explicitly sees that it arises as the generic fibre of the group scheme ( D ⊕ D ) / Z 2 u r   , where D   is the non-trivial extension of Z / 2 Z   by μ 2   considered in [18] (Prop 4.2, p. 58).
Thus one expects ρ   to arise from an F 2 [ x ] / ( x 2 )   -valued modular form of weight two and level N   , corresponding to a surjective map of rings T m → F 2 [ x ] / ( x 2 )   . This would follow if we knew that T m   coincided with the minimal deformation ring associated to ρ ¯   . Rather than proving this, however, we shall use weight one forms to explicitly construct a weight two modular form giving rise to ρ   .
Let χ 4 N   be the character of conductor 4 N   associated to F   . Consider two faithful representations ψ 1 : G a l ( L / Q ) ∼ = S 3 ↪ G L 2 ( C ) , ψ 2 : G a l ( H / Q ) ∼ = S 3 × Z / 2 Z ↪ G L 2 ( C ) .   Since F / Q   is complex, these dihedral representations are odd and therefore give rise to weight one modular forms h 1   , h 2   in S 1 ( Γ 1 ( 4 N ) , χ 4 N )   .
3.11 Lemma. The modular forms h 1   , h 2   are ordinary at 2   , have coefficients in Z   , and are congruent modulo 2   . Let g = ( h 2 − h 1 ) 2 ∈ Z [ [ q ] ] .   Then g m o d 2   is non-zero, and the form h = h 1 + x g ∈ S 1 ( Γ 1 ( 4 N ) , F 2 [ x ] / ( x 2 ) )   is an eigenform for all the Hecke operators, including U 2   . The associated G L 2 ( F 2 [ x ] / ( x 2 ) )   Galois representation attached to h   is ρ   .
Proof. The modular forms are both ordinary at 2   because the representations ψ 1   and ψ 2   have non-trivial subspaces on which inertia at two is trivial (since I 2   acts through a group of order 2   ). They both have coefficients in Z   , since 2 cos ( π / 3 ) ∈ Z   . The congruence h 1 ≡ h 2   follows from the fact that both are ordinary-at- 2   Hecke eigenforms, and that a ( h 1 , ℓ ) = a ( h 2 , ℓ ) χ 4 ( ℓ )   for all odd primes ℓ   , where χ 4   is the character of conductor 4   .
From this one also sees that g   is non-trivial modulo two. The claims about h   follow formally from the fact that h 1   and h 2   are Hecke eigenforms that are congruent modulo 2   , and the definition of ρ   . □  
Now that we have constructed the weight one form h   of level 4 N   giving rise to ρ   , we would like to construct a corresponding weight two form of level N   . Multiplying h   by the Eisenstein series in M 1 ( Γ 1 ( 4 N ) , χ 4 N )   , we see that h   is the q   -expansion of a modular form in S 2 ( Γ 0 ( 4 N ) , F 2 [ x ] / ( x 2 ) )   . Since h   is ordinary and is a U 2   eigenform, we may apply the U 2   operator to deduce that h ∈ S 2 ( Γ 0 ( 2 N ) , F 2 [ x ] / ( x 2 ) )   . Applying theorem  3.18 (proved in the following subsection) we then deduce that in fact h ∈ S 2 ( Γ 0 ( N ) , F 2 [ x ] / ( x 2 ) )   , and (thus) that g ∈ S 2 ( Γ 0 ( N ) , F 2 )   . An easy calculation shows that if f = h 1 ≡ h 2 ≡ ∑ a n q n   then m f = 0   , whilst m g ⊂ F 2 f   , and so also m 2 g = 0   . Thus T m ≠ Z 2   , by lemma  3.3 .

3.12 Level-lowering for modular deformations

The goal of this section is to prove a level-lowering result for modular forms with coefficients in Artinian rings that strengthens the case p = 2   of [12,thm. 2.8] (which in turn extends a level lowering result proved by Mazur [22,thm. 6.1] in the odd prime case).
We first establish a version of the multiplicity one theorem [29,thm. 2.1] for p = 2   .
Under the additional assumption that ρ ¯   is not finite at 2   , this theorem was proved in [4,§2] (as was the corresponding result for odd level). Thus the key point in our theorem is that ρ ¯   is allowed to be finite at 2   , even though the level is taken to be even.
3.13 Theorem. Let N   be an odd natural number, and let T   denote the full Z   -algebra of Hecke operators acting on weight two cuspforms of level 2 N   . If m   is a maximal ideal in T   whose residue field k   is of characteristic 2   , and for which the associated residual Galois representation ρ ¯ : G a l ( Q ¯ / Q ) → G L 2 ( k )   is (absolutely ) irreducible, ordinary, and ramified at 2   , then T a m J 0 ( 2 N )   (the m   -adic Tate module of J 0 ( 2 N )   ) is free of rank two over the completion T m   .
To be clear, the condition “ordinary at 2   ” means that the image of a decomposition group at 2   under ρ ¯   lies in a Borel subgroup of G L 2 ( k )   . Since k   is of characteristic 2   , we see that (for an appropriate choice of basis) the restriction of ρ ¯   to an inertia group at 2   may be written in the form ρ ¯ | I 2 = ( 1 * 0 1 ) .   The assumption that ρ ¯   is ramified at 2   then implies that *   is not identically zero. Thus the representation space of ρ ¯   has a unique line invariant under I 2   , and so ρ ¯   is irreducible if and only if it is absolutely irreducible.
3.14 Lemma. Let k   be a finite field of characteristic 2   . If ρ ¯ : G a l ( Q ¯ 2 / Q 2 ) → G L 2 ( k )   is a continuous representation that is finite, ordinary, and ramified at 2   , then ρ ¯   has a unique finite flat prolongation over Z 2   (up to unique isomorphism). Furthermore, this prolongation is an extension of a rank one étale k   -vector space scheme by a rank one multiplicative k   -vector space scheme.
Proof. Any finite flat group scheme that prolongs an unramified continuous representation of G a l ( Q ¯ 2 / Q 2 )   on a one-dimensional k   -vector space is either étale or multiplicative. Thus there are a priori four possible structures for a finite flat prolongation of ρ ¯   : étale extended by étale; multiplicative extended by multiplicative; multiplicative extended by étale; or étale extended by multiplicative. However, all but the last possibility necessarily gives riseto an unramified generic fibre (note that any extension of multiplicative by étale must split, by a consideration of the connected étale sequence). Thus, since ρ ¯   is assumed ramified, we see that any prolongation of ρ ¯   must be an extension of a rank one étale k   -vector space scheme scheme by a rank one multiplicative k   -vector space scheme.
To see that such a prolongation is unique, consider the minimal and maximal prolongations M   and M ′   of ρ ¯   to a finite flat group scheme [21,cor. 2.2.3] . The result of the preceding paragraph shows that the natural morphism M → M ′   necessarily induces an isomorphism on the connected components of the identity, and on the corresponding groups of connected components. By the 5   -lemma, this morphism is thus an isomorphism, and the lemma follows. □  
We now show that certain results of Mazur [19] cited in the proof of [29,thm. 2.1] extend to the case p = 2   . We put ourselves in the context of [19,§1] , and use the notation introduced there. Namely, we let K   denote a finite extension of Q p   for some prime p   , and let O   denote the ring of integers of K   . If A   is an abelian variety over K   , then we let A / O   denote the connected component of the identity of the Néron model of A   over S p e c O   . For any power p r   of p   , the p r   -torsion subgroup scheme A [ p r ] / O   of A / O   is then a quasi-finite flat group scheme over S p e c O   ; we let F A [ p r ] / O   denote its maximal finite flat subgroup scheme, and A [ p r ] / O 0   denote the maximal connected closed subgroup scheme of A [ p r ] / O   .
Since we took A / O   to be the connected component of the Néron model of A   , the inductive limit F A [ p ∞ ] / O : = lim ⟶ F A [ p r ] / O   is a p   -divisible group, and A [ p ∞ ] / O 0 : = lim ⟶ A [ p r ] / O 0   is the maximal connected p   -divisible subgroup of F A [ p ∞ ] / O   .
The following proposition is a variation on [19,prop. 1.3] , in which we allow the ramification of K   over Q p   to be unrestricted, at the expense of imposing more restrictive hypotheses on the reduction of the abelian varieties appearing in the exact sequence under consideration.
3.15 Proposition. Let 0 → A → B → C → 0   be an exact sequence of abelian varieties over K   such that A   has purely toric reduction, whilst C   has good reduction. Then the induced sequence of p   -divisible groups 0 → A [ p ∞ ] / O 0 → B [ p ∞ ] / O 0 → C [ p ∞ ] / O 0 → 0   is a short exact sequence of p   -divisible groups over S p e c O   . Equivalently, for any power p r   of p   , the induced sequence 0 → A [ p r ] / O 0 → B [ p r ] / O 0 → C [ p r ] / O 0 → 0   is a short exact sequence of finite flat group schemes over S p e c O   .
Proof. Since A   has purely toric reduction, the group scheme A [ p r ] / O 0   is of multiplicative type for each r   . Thus it necessarily maps isomorphically onto its scheme theoretic image in B / O   , and thus the induced map A [ p ∞ ] / O 0 → B [ p ∞ ] / O 0   is a closed embedding.
Let C ′ ⊂ B   be an abelian subvariety chosen so that the induced map C ′ → C   is an isogeny. Then C ′   also has good reduction, and so C ′ [ p ∞ ] / O 0 → C [ p ∞ ] / O 0   is an epimorphism of p   -divisible groups over S p e c O   . Thus the induced map B [ p ∞ ] / O 0 → C [ p ∞ ] / O 0   is also an epimorphism of p   -divisible groups. A consideration of generic fibres shows that the kernel of this surjection coincides with the scheme-theoretic image of A [ p ∞ ] / O 0   in B [ p ∞ ] / O 0   , and so the proposition is proved. □  
Proof of theorem  3.13 . We closely follow the method of proof of [29,thm. 2.1 (ii)] in the case when “ Δ ( p )   is trivial m o d m   ” (in the terminology of that proof; see [29,pp. 485–488] ).
If we let A   denote the connected part of the kernel of the map J 0 ( 2 N ) → J 0 ( N ) × J 0 ( N )   induced by Albanese functoriality applied to the two “degeneracy maps” from level 2 N   to level N   , then A   is an abelian subvariety of J 0 ( 2 N )   having purely toric reduction at 2   , whilst the quotient B   of J 0 ( 2 N )   by A   has good reduction at 2   . From proposition  3.15 we obtain (for any r ≥ 1   ) the short exact sequence 0 → A [ 2 r ] / Z 2 0 → J 0 ( 2 N ) [ 2 r ] / Z 2 0 → B [ 2 r ] / Z 2 0 → 0   of connected finite flat group schemes over S p e c Z 2   . By functoriality of the formation of this short exact sequence, and since A   is a T   -invariant subvariety of J 0 ( 2 N )   , we see that this is in fact a short exact sequence of T   -module schemes. Localizing at m   induces the corresponding short exact sequence
0 → A [ 2 r ] m / Z 2 0 → J 0 ( 2 N ) [ 2 r ] m / Z 2 0 → B [ 2 r ] m / Z 2 0 → 0 . (4)
Passing to Q ¯ 2   -valued points induces a short exact sequence of G a l ( Q ¯ 2 / Q 2 )   -modules
0 → A [ 2 r ] m 0 ( Q ¯ 2 ) → J 0 ( 2 N ) [ 2 r ] m 0 ( Q ¯ 2 ) → B [ 2 r ] m 0 ( Q ¯ 2 ) → 0 , (5)
which is a subexact sequence of the short exact sequence of G a l ( Q ¯ 2 / Q 2 )   -modules
0 → A [ 2 r ] m ( Q ¯ 2 ) → J 0 ( 2 N ) [ 2 r ] m ( Q ¯ 2 ) → B [ 2 r ] m ( Q ¯ 2 ) → 0 . (6)
Let A [ 2 r ] m ( Q ¯ 2 ) χ   (respectively J 0 ( 2 N ) [ 2 r ] m ( Q ¯ 2 ) χ   , respectively B [ 2 r ] m ( Q ¯ 2 ) χ   ) denote the maximal G a l ( Q ¯ 2 / Q 2 )   -subrepresentation of A [ 2 r ] m ( Q ¯ 2 )   (respectively J 0 ( 2 N ) [ 2 r ] m ( Q ¯ 2 )   , respectively B [ 2 r ] m ( Q ¯ 2 )   ) on which the inertia group acts through the 2   -adic cyclotomic character χ   .
The short exact sequence ( 6 ) induces an exact sequence
0 → A [ 2 r ] m ( Q ¯ 2 ) χ → J 0 ( 2 N ) [ 2 r ] m ( Q ¯ 2 ) χ → B [ 2 r ] m ( Q ¯ 2 ) χ . (7)
3.16 Lemma. Each of the groups schemes appearing in the exact sequence ( 4 ) is of multiplicative type, and the exact sequences ( 5 ) and ( 7 ) coincide (as subsequences of ( 6 )).
Proof. We first remark that ( 6 ) is the exact sequence of T m [ G a l ( Q ¯ 2 / Q 2 ) ]   -modules underlying the the corresponding exact sequence of T m [ G a l ( Q ¯ / Q ) ]   -modules 0 → A [ 2 r ] m ( Q ¯ ) → J 0 ( 2 N ) [ 2 r ] m ( Q ¯ ) → B [ 2 r ] m ( Q ¯ ) → 0 .   Since ρ ¯   is assumed irreducible as a k [ G a l ( Q ¯ / Q ) ]   -representation, each of the modules appearing in this exact sequence is a successive extension of copies of ρ ¯   . The same is thus true of each of the modules appearing in the exact sequence ( 6 ).
Since A   has purely toric reduction, it is clear that A [ 2 r ] m / Z 2 0   is of multiplicative type, and so
A [ 2 r ] m 0 ( Q ¯ 2 ) ⊂ A [ 2 r ] m ( Q ¯ 2 ) χ . (8)
Fix a filtration 0 = W 0 ⊂ W 1 ⊂ ⋯ ⊂ W n = A [ 2 r ] m ( Q ¯ 2 )   of A [ 2 r ] m ( Q ¯ 2 )   for which the successive quotients W i + 1 / W i   are isomorphic to ρ ¯   . Since A   has purely toric reduction the quotient A [ 2 r ] / Q 2 / A [ 2 r ] / Q 2 0   is Cartier dual to A ^ [ 2 r ] / Q 2 0   (where A ^   is the dual abelian variety to A   ), and so A [ 2 r ] m ( Q ¯ 2 ) / A [ 2 r ] m 0 ( Q ¯ 2 )   is an unramified G a l ( Q ¯ 2 / Q 2 )   -representation. Since ρ ¯   is assumed ramified at 2   , this implies that W i + 1 ⋂ A [ 2 r ] m 0 ( Q ¯ 2 ) ⊊ W i ⋂ A [ 2 r ] m 0 ( Q ¯ 2 )   for each i ≥ 0   . Furthermore, W i ⊄ A [ 2 r ] m ( Q ¯ 2 ) χ   for each i > 0   , because χ m o d 2   is trivial. Since W i + 1 / W i ∼ = ρ ¯   is two dimensional over k   for each i ≥ 0 ,   we conclude by induction on i   that W i ⋂ A [ 2 r ] m 0 ( Q ¯ 2 ) = W i ⋂ A [ 2 r ] m ( Q ¯ 2 ) χ   for each i ≥ 0   . Taking i = n   then shows that the inclusion ( 8 ) is in fact an equality.
Since B   has good reduction at 2   , we have equality F B [ 2 r ] / Z 2 = B [ 2 r ] / Z 2   . As noted above, any Jordan–Hölder filtration of the localization B [ 2 r ] m ( Q ¯ )   as a T [ G a l ( Q ¯ / Q ) ]   -modulehas all its associated graded pieces isomorphic to ρ ¯   . Taking scheme-theoretic closures of such a filtration in B [ 2 r ] / Z 2 ,   we obtain a filtration of the localization B [ 2 r ] m / Z 2   by finite flat closed subgroup schemes, whose associated graded pieces are prolongations of ρ ¯   . Now lemma  3.14 shows that the connected component of any such finite flat prolongation is multiplicative. Thus B [ 2 r ] m / Z 2 0   is indeed multiplicative, whilst B [ 2 r ] m ( Q ¯ 2 ) / B [ 2 r ] m 0 ( Q ¯ 2 )   is an unramified G a l ( Q ¯ 2 / Q 2 )   -module. Arguing as in the preceding paragraph gives the required equality B [ 2 r ] m 0 ( Q ¯ 2 ) = B [ 2 r ] m ( Q ¯ 2 ) χ .   Since any extension of multiplicative type groups is again of multiplicative type, we see that J 0 ( 2 N ) [ 2 r ] m / Z 2 0   is also of multiplicative type, and that the exact sequence ( 5 ) is a subsequence of the exact sequence ( 7 ). We have furthermore shown that first and third non-trivial terms of these two sequences coincide. This implies that these exact sequences do indeed coincide. □  
Specializing lemma  3.16 to the case r = 1   shows that J 0 ( 2 N ) [ 2 ] m 0 ( Q ¯ 2 )   is the maximal unramified G a l ( Q ¯ 2 / Q 2 )   -subrepresentation of J 0 ( 2 N ) [ 2 ] m ( Q ¯ 2 )   (since χ m o d 2   is trivial).
Recall that there is a natural isomorphism T a n ( J 0 ( 2 N ) [ 2 ] / F ¯ 2 0 ) ∼ = T a n ( J 0 ( 2 N ) / F ¯ 2 )   (indeed, this is true with J 0 ( 2 N ) / F ¯ 2   replaced by any group scheme over F ¯ 2   ), and also a natural isomorphism T a n ( J 0 ( 2 N ) [ 2 ] / F ¯ 2 0 ) ∼ = J 0 ( 2 N ) [ 2 ] 0 ( Q ¯ 2 ) ⊗ F 2 F ¯ 2   (as follows from the discussion on [29,p. 488] ). Localizing at m   , and taking into account [29,lem. 2.2] , which is valid for p = 2   , we find that J 0 ( 2 N ) [ 2 ] m 0 ( Q ¯ 2 )   is a cyclic T m   -module, and thus that the maximal unramified G a l ( Q ¯ 2 / Q 2 )   -subrepresentation of J 0 ( 2 N ) [ 2 ] m ( Q ¯ 2 )   is a cyclic T m   -module.
Let ρ m : G a l ( Q ¯ / Q ) → G L 2 ( T m )   denote the Galois representation associated to m   by [6,thm. 3] . Carayol has proved [6,thm. 4] that there is an isomorphism T a m J 0 ( 2 N ) ∼ = J ⊗ T m ρ m   for some ideal J   in T m   , and thus an isomorphism J 0 ( 2 N ) [ 2 ] m ( Q ¯ 2 ) ∼ = ( J / 2 J ) ⊗ T m ρ m   . We conclude that J / 2 J   is a cyclic T m   -module, and hence that J   is a principal ideal in T m   .
The discussion of [6,3.3.2] shows that in fact J ∼ = T m   and that T a m J 0 ( 2 N )   is free of rank two over T m   , as claimed. □  
3.17 Corollary. In the situation of theorem  3.13 , the completion T m   is a Gorenstein Z 2   -algebra.
Proof. This follows from the theorem together with the self-duality of the m   -adic Tate module under the Weil pairing. □  
We now prove our level lowering result. Let A   be an Artinian ring with finite residue field k   of characteristic 2   , and suppose given a continuous representation ρ : G a l ( Q ¯ / Q ) → G L 2 ( A )   that is modular of level Γ 0 ( 2 N )   for some odd natural number N   , in the sense that it arises from a Hecke eigenform h ∈ S 2 ( Γ 0 ( 2 N ) , A )   . Let ρ ¯   denote the residual representationattached to ρ   (so ρ ¯   arises from the Hecke eigenform h ¯ ∈ S 2 ( Γ 0 ( 2 N ) , k )   obtained by reducing h   modulo the maximal ideal of A   ).
3.18 Theorem. If ρ : G a l ( Q ¯ / Q ) → G L 2 ( A )   is a modular Galois representation of level Γ 0 ( 2 N )   as above, such that
  • 1. ρ ¯   is (absolutely ) irreducible,
  • 2. ρ ¯   is ordinary and ramified at 2   , and
  • 3. ρ   is finite flat at 2   ,
then ρ   arises from an A   -valued Hecke eigenform of level N   .
Proof. The Hecke eigenform h   corresponds to a ring homomorphism φ : T → A   . Since A   is local of residue characteristic 2   , the map φ   factors through the completion T m   of T   at some maximal ideal m   of residue characteristic 2   , and the residual representation ρ ¯   is the residual Galois representation attached to the maximal ideal m   . We let ρ m   denote the Galois representation ρ m : G a l ( Q ¯ / Q ) → G L 2 ( T m )   attached to m   by [6,thm. 3] . The Galois representation ρ   attached to h   coincides with the pushforward of ρ m   via φ   .
Replacing A   by the image of φ   , we may and do assume from now on that φ   is surjective.
We let I ⊂ T m   denote the kernel of φ   . Since A   is Artinian, we may choose r ≥ 1   so that 2 r ∈ I   . Theorem  3.13 shows that the m   -adic Tate module T a m J 0 ( 2 N )   is isomorphic as a G a l ( Q ¯ / Q )   -representation to ρ m   ; thus J 0 ( 2 N ) [ 2 r ] m ( Q ¯ )   is isomorphic to the reduction mod 2 r   of ρ m   . Since T m   is a Gorenstein Z 2   -algebra, by corollary  3.17 , we see that T m / 2 r T m   is a Gorenstein Z / 2 n Z   -algebra, and thus that there is an isomorphism ( T m / 2 r T m ) [ I ] ∼ = H o m Z / 2 r ( T m / I , Z / 2 r )   of T m / I = A   -modules. In particular, J 0 ( 2 N ) [ I ] ( Q ¯ / Q ) ⊂ J 0 ( 2 N ) [ 2 r ] m ( Q ¯ )   is a faithful A   -module, isomorphic as an A [ G a l ( Q ¯ / Q ) ]   -module to H o m Z / 2 r ( T m / I , Z / 2 r ) ⊗ A ρ .   To simplify notation, we will write
V : = J 0 ( 2 N ) [ I ] ( Q ¯ ) ∼ = H o m Z / 2 r ( T m / I , Z / 2 r ) ⊗ A ρ . (9)
By assumption, ρ   prolongs to a finite flat group scheme ℳ   over S p e c Z 2   . If we fix a Jordan–Hölder filtration of ρ   as an A [ G a l ( Q ¯ / Q ) ]   -module, then the associated graded pieces are each isomorphic to ρ ¯   , and so lemma  3.14 and [5,Prop. 2.5] together imply that ℳ   is uniquely determined by ρ   , whilst [5,Lem. 2.4] then implies that ℳ   is naturally an A   -module scheme. From ( 9 ) we see that V   also prolongs to a finite flat A   -module scheme V ∼ = H o m Z / 2 r ( T m / I , Z / 2 r ) ⊗ A ℳ   over Z 2   . Again, lemma  3.14 and [5,Prop. 2.5] show that V   is the unique finite flat prolongation of V   .
Lemma  3.14 furthermore implies that ℳ   is the extension of an étale A   -module scheme ℳ ét   by a multiplicative A   -module scheme ℳ 0   , each of which is free of rank one as an A   -module scheme. Thus V   is also an extension of an étale A   -module scheme V ét   by multiplicative A   -module scheme V 0   , each of which is faithful as an A   -module scheme.
Let V ét   and V 0   denote the generic fibres of these schemes.
We write J   to denote the Néron model of J 0 ( 2 N )   over S p e c Z 2   . For a scheme over Z 2   , use the subscript “ s   ” to denote its special fibre over F 2   . The special fibre J s   admits the following filtration by T   -invariant closed subgroups:
0 ⊂ T ⊂ J s 0 ⊂ J s ,   where T   is the maximal torus contained in J s ,   and J s 0   is the connected component of the identity of J s   . The quotient J s 0 / T   is an abelian variety on which T   acts through its quotient T o l d   (where T o l d   denotes the quotient of T   that acts faithfully on the space of 2   -old forms of level 2 N   ). The connected component group Φ : = J s / J s 0   is Eisenstein [22,Thm. 3.12] .
The following lemma provides an analogue in our situation of [22,lem. 6.2] (and generalizes one step of the argument in the proof of [12,thm. 2.8] ).
3.19 Lemma. The Zariski closure of V   in J   is a finite flat A   -module scheme over Z 2   (which is thus isomorphic to V   ).
Proof. Since V 0   is a multiplicative type group scheme, inertia at 2   acts on V 0 ( Q ¯ 2 )   through the cyclotomic character. It follows from lemma  3.16 that V 0   is contained in the generic fibre of J 0 ( 2 N ) [ 2 r ] m / Z 2 0   , and thus that the Zariski closure of V 0   in J   is indeed finite flat, and in fact of multiplicative type. Thus it coincides with V 0   , and so we see that the embedding of V 0   in J 0 ( 2 N )   prolongs to an embedding of V 0   in J   . Since the quotient V ét = V / V 0   is étale, lemma 5.9.2 of [15,ExposéIX] serves to complete the proof of the lemma. □  
Lemma  3.19 allows us to regard V   as a closed T   -submodule scheme of J   , and thus to regard V s   as a closed T   -submodule scheme of J s   . Since ρ ¯   is irreducible and Φ   is Eisenstein, we see that V s   is in fact contained in J s 0   . On the other hand, since T   is a torus, we see that V s ∩ T ⊂ V s 0   . Thus V s ét   appears as a subquotient of J s 0 / T   , and in particular the T   -action on V ét   factors through the quotient T o l d   of T   . Since V s ét   is a faithful A   -module scheme, we see that the map φ : T → A   factors through T o l d   , completing the proof of the theorem.
□  
We remark that the obvious analogue of theorem  3.18 in the case of odd residue characteristic is also true. The proof is similar but easier, relying on the uniqueness results on finite flat models due to Raynaud [21] . Of course, in those cases when the R = T   theorem of Wiles, Taylor–Wiles, and Diamond [9, 26, 29] applies, it is an immediate consequence of that theorem. (Thus theorem  3.18 may be regarded as evidence for an R = T   theorem for those ρ ¯   of residue characteristic 2   that satisfy its hypotheses.)

4 N   a proper prime power

There are only finitely many elliptic curves of conductor 2 k   for all k   , and we may explicitly determine which have odd modular degree. Therefore we assume that E   has conductor N   , where N = p k   with k ≥ 2   and p ≥ 3   . Let χ   be the unique quadratic character of conductor p   . Let E ′   be the elliptic curve E   twisted by χ   . The curve E ′   also has conductor N   , and moreover the associated modular forms f E   and f E ′   are congruent modulo 2   , since twisting by quadratic characters preserves E [ 2 ]   . Since N   is odd, any non-trivial congruence modulo 2   between f E   and other forms in S 2 ( Γ 0 ( N ) )   forces the modular degree m E   to be even [30] . Thus we are done unless f E = f E ′ = f E ⊗ χ   . In particular, the representations associated to f E   must be induced from a quadratic field, and thus E   has complex multiplication. (Alternatively, the equality f E = f E ′   implies that E   is isogenous to its twist and deduce this way that E   has CM.) If E   has CM and prime power conductor then E   is one of finitely many well known elliptic curves, for which we can directly determine the modular degree by consulting current databases (for N = 163 2   , we use the elliptic curve database of Stein–Watkins, described in [23] ).

5 Further remarks

Certainly not every E   satisfying the conditions of theorem  1.1 will actually have odd modular degree, and one could try to refine this result by deducing additional necessary conditions that E   must satisfy in order to have odd modular degree. In this section we say a little about the related question of whether or not 2   is a congruence prime for the associated modular form f E   , when E   satisfies either of conditions (3a) or (3b) of the theorem.
For curves E   with a rational two torsion point, the modular form f E   automatically satisfies a mod two congruence with an Eisenstein series, and so detecting whether f   satisfies a congruence with a cuspform is a more subtle phenomenon than in the non-Eisenstein situation. One approach might be to relate the Hecke algebra to an appropriate universal deformation ring (if the latter exists). If N   is prime this can be done [5] , and this enables one to determine when T m = Z 2   for such representations. The specific determination of when T m = Z 2   , however, was already achieved (for N   prime and ρ ¯   reducible) by Merel in [20] :
5.1 Theorem. Let N ≡ 1 m o d 8   be prime, and let T m   be the localization of the Eisenstein prime at 2   . Then T m ≠ Z 2   if and only if N = u 2 + 16 v 2   and v ≡ ( N − 1 ) / 8 m o d 2   .
If E   is a Neumann–Setzer curve, then N = u 2 + 64   for some u ∈ Z   . The result of Merel above then clearly implies that the optimal Neumann–Setzer curve E   has odd modular degree if and only if N ≢ 1 m o d 16   . (An alternative proof of this fact, relying on the results of [18] , is given in [24] , thm. 2.1.) If E   has composite conductor, then one might try to generalize the results of [20] or [5] to this setting.
Suppose now that E   has prime conductor, that ρ ¯   is irreducible and supersingular, and that Q ( E [ 2 ] )   is totally complex. Let K   and L   be the extensions of Q   attached to E   as in the discussion of section  3 . If one had an R = T   result of the type discussed in the introduction, then to obtain further necessary conditions for E   to have odd modular degree, it would suffice to establish sufficient conditions for the existence of an appropriate non-trivial minimal deformation ρ : G Q → G L 2 ( F 2 [ x ] / ( x 2 ) )   lifting ρ ¯   . For representations ρ ¯   that were complex and ramified at 2   , but ordinary, we constructed such a ρ   directly in subsection  3.10 by considering a quadratic genus field extension of L   . When E   is supersingular, such deformations ρ   (when they exist) may be more subtle and can not necessarily be constructed so directly. One can however prove the following result.
Recall that if ρ ¯   is supersingular and totally complex, then K / Q   is totally ramified at 2   and K   has exactly two complex embeddings.
5.2 Proposition. Suppose ρ ¯   is supersingular at 2   and totally complex. If either:
  • 1. the class number of K   is even, or
  • 2. the fundamental unit ε   of O K   satisfies v π ( ε − 1 ) ≥ 2   , where π   is the unique prime above 2   in O K   ,
then there exists a non-trivial minimal deformation ρ : G a l ( Q ¯ / Q ) → S L 2 ( F 2 [ x ] / ( x 2 ) ) .  
Note that S L 2 ( F 2 [ x ] / ( x 2 ) ) ∼ = S 4 × Z / 2 Z   . The idea behind the proof of this result is to study A 4   extensions H   containing L   that are minimally ramified at 2   and unramified at N   . Such extensions can be obtained by considering the Galois closure over Q   of certain quadratic extensions of K   . One obtains a suitable such extension either by considering an unramified extension of K   , in case (1), or the extension K ( ε K )   , in case (2).
References

  1. A. Agashe, The Birch and Swinnerton-Dyer formula for modular abelian varieties of analytic rank zero, Berkeley Ph.D. Thesis, 2000.
  2. N. Boston, H. Lenstra, K. Ribet, Quotients of group rings arising from two-dimensional representations, C. R. Acad. Sci. Paris, Série I 312 (1991), 323–328.
  3. C. Breuil, B. Conrad, F. Diamond, R. Taylor, On the modularity of elliptic curves over Q   : wild 3   -adic exercises, J. Amer. Math. Soc. 14 (2001), no. 4, 843–939.
  4. K. Buzzard, On level-lowering for mod 2 representations, Math. Res. Lett. 7 (2000), no. 1, 95–110.
  5. F. Calegari, M. Emerton, On the ramification of Hecke algebras at Eisenstein primes, to appear in Invent. Math.
  6. H. Carayol, Formes modulaires et répresentations Galoisiennes á valeurs dans un anneau local complet, in p   -adic monodromy and the Birch and Swinnerton-Dyer conjecture (B. Mazur, G. Stevens, eds.), Contemp. Math. 165 (1994), 213–235.
  7. J. Cremona, M. Watkins, data available at http://www.maths.nott.ac.uk/personal/jec/ftp/data
  8. E. de Shalit, Hecke rings and universal deformation rings, in Modular forms and Fermat's last theorem (Boston, MA, 1995), 421–445, Springer.
  9. F. Diamond, On deformation rings and Hecke rings, Ann. Math. 144 (1996), no. 1, 137–166.
  10. M. Dickinson, On the modularity of certain 2   -adic Galois representations, Duke Math. J. 109 (2001), no. 2, 319–382.
  11. N. Dummigan On a conjecture of Watkins, preprint, 2004.
  12. B. Edixhoven, The weight in Serre's conjectures on modular forms, Invent. Math. 109 (1992), no. 3, 563–594.
  13. J. Fontaine, Il n'y a pas de variété abélienne sur Z   , Invent. Math. 81, 1985, no. 3, 515–538.
  14. B. Gross, A tameness criterion for Galois representations associated to modular forms (mod p   ), Duke Math. J. 61 (1990), no. 2, 445–517.
  15. A. Grothendieck, Groups de monodromie en géométrie algébrique I, (SGA 7), Séminaire de Géométrie Algébrique du Bois-Marie, 1967-1969, Lecture Notes in Math., vol. 288 , Springer-Verlag, Berlin and New York, 1972, 313–523.
  16. L. Kilford, Some non-Gorenstein Hecke algebras attached to spaces of modular forms, J. Number Theory 97 (2002), no. 1, 157–164.
  17. W. Li, Newforms and functional equations, Math. Ann. 212 (1975), 285–315.
  18. B. Mazur, Modular curves and the Eisenstein ideal, Publ. Math. IHES 47 (1977), 33–186.
  19. B. Mazur Rational isogenies of prime degree, Invent. Math. 44 (1978), no. 2, 129–162.
  20. L. Merel, L'accouplement de Weil entre le sous-groupe de Shimura et le sous-groupe cuspidal de J 0 ( p )   , J. Reine Angew. Math. 477 (1996), 71–115.
  21. M. Raynaud, Schémas en groupes de type ( p , . . . , p )   , Bull. Soc. Math. France 102 (1974), 241–280.
  22. K. Ribet, On modular representations of G a l ( Q ¯ / Q )   arising from modular forms, Invent. Math. 100, (1990), no. 2, 431–476.
  23. W. Stein, M. Watkins, A Database of Elliptic Curves—First Report, Algorithmic number theory (Sydney, 2002), 267–275, Lecture Notes in Comput. Sci., 2369, Springer, Berlin, 2002.
  24. W. Stein, M. Watkins, Modular parametrizations of Neumann–Setzer elliptic curves, IMRN, no. 27 (2004), 1395–1405.
  25. J. Tate, The non-existence of certain Galois extensions of Q   unramified outside 2   , in Arithmetic Geometry (N. Childress, J. Jones, eds.), Contemp. Math. 174 (1994), 153–156.
  26. R. Taylor, A. Wiles, Ring-theoretic properties of certain Hecke algebras, Ann. Math. 141 (1995), no. 3, 553–572.
  27. M. Watkins, Computing the modular degree of an elliptic curve, Experiment. Math. 11 (2002), no. 4, 487–502.
  28. G. Wiese, Dihedral Galois representations and Katz modular forms, Doc. Math. 9 (2004), 123–133.
  29. A. Wiles, Modular elliptic curves and Fermat's last theorem, Ann. Math. 141 (1995), no. 3, 443–551.
  30. D. Zagier, Modular parametrizations of elliptic curves, Canad. Math. Bull. 28 (1985), no. 3, 372–384.

Email addresses: fcale@math.harvard.edu emerton@math.northwestern.edu