betaFeedback

WorksMathematics

Modularity of elliptic curves over totally real field and CM fields not containing \zeta_5

Abstract

This notes tries to parse the basic structure of OpenAI's new argument proving modularity of elliptic curves over imaginary quadratic fields. The argument also proves modularity over all totally real field, and in the original form, modularity over CM fields not containing \zeta_5. The general CM case has also now been resolved by an independent repository which improves the Caraiani-Newton modularity lifting theorem for CM field at the prime 5, which was the main reason for the extra hypothesis. I kept the argument in the totally real case separate to highlight the main new idea in the OpenAI proof, which is orthogonal to modularity lifting theorems.

Introduction

One of the great achievements of modern number theory is the proof of the Taniyama–Shimura conjecture: every elliptic curve E/QE/\mathbb{Q} is modular, that is, L(E,s)=L(f,s)L(E,s)=L(f,s) for a newform ff of weight two and level Γ0(N)\Gamma_{0}(N). Equivalently, the Tate modules of EE are isomorphic to the ℓ\ell-adic representations attached to ff by Eichler and Shimura [41]. The conjecture was the key to Fermat’s Last Theorem: a solution of the Fermat equation gives a semistable elliptic curve [24] which, as anticipated by Serre [39], cannot be modular by Ribet’s level-lowering theorem [36]. Wiles and Taylor–Wiles proved the conjecture for semistable curves [52 Theorem 0.4], [47], and after further work [21, 16] Breuil, Conrad, Diamond, and Taylor proved it in general [10].

Over a number field KK, Langlands reciprocity [32] predicts that every elliptic curve E/KE/K is modular: the compatible system ρE,ℓ\rho_{E,\ell} given by H1(E)H^{1}(E) is attached to a regular algebraic automorphic representation π\pi of GL2(AK)\mathrm{GL}_{2}(\mathbb{A}_{K}), cuspidal unless EE has complex multiplication defined over KK. This formulation, and essentially all known approaches to modularity, require the construction of compatible systems of Galois representations attached to regular algebraic cuspidal automorphic representations of GL2(AK)\mathrm{GL}_{2}(\mathbb{A}_{K}). Such constructions are known when KK is totally real, by work of Carayol, Wiles, Blasius–Rogawski, and Taylor [15, 51, 7, 44], and when KK is CM: first over imaginary quadratic fields under a condition on the central character [27, 45, 6], and in general by Harris–Lan–Taylor–Thorne [26]. Because of this, modularity results to date are mostly confined to totally real or CM fields; a notable exception is the work of Boxer, Calegari, Gee, and Pilloni on elliptic curves over quadratic extensions of totally real fields that need not be CM [9].

The main approach to modularity follows the strategy, due to Wiles, of propagating modularity along congruences. It has two steps:

  • (Modularity lifting) If ρ\rho is an ℓ\ell-adic representation satisfying suitable local conditions (including ℓ\ell-adic Hodge-theoretic conditions at ℓ\ell) and ρˉ≃ρˉπ,ℓ\bar{\rho}\simeq\bar{\rho}_{\pi,\ell} for the Galois representation ρπ,ℓ\rho_{\pi,\ell} of some automorphic π\pi, then ρ\rho is modular. This usually requires the image of ρˉ\bar{\rho} to be sufficiently large.

  • (Chaining) Find a chain of congruences between ρE,ℓ\rho_{E,\ell} and some ρπ,ℓ\rho_{\pi,\ell} with π\pi automorphic, such that the first step applies at each link. Crucially, the prime at which the congruence holds may change along the chain.

The strategy is most developed in the totally real case. Building on Wiles’s method, work of Kisin [30], Gee [25], and Barnet-Lamb–Gee–Geraghty [4, 5] led to essentially optimal modularity lifting theorems in weight two; see [11 Théorème 3.2.2] for a convenient statement. Thus the difficulty lies in the second step: finding enough congruences to representations known to be modular. If one is willing to enlarge KK, that is, to prove only potential modularity, this is resolved by Taylor [46]; for elliptic curves see [53 Theorem A.1], and [42 Theorem 1.1.1] for general weight-two representations. On the other hand, it seems very hard to produce enough congruences while keeping KK fixed, as this is related to finding points on complicated algebraic varieties. For this reason, previous results restrict the field: to low degree, namely real quadratic [23], totally real cubic [20], quartic fields not containing 5\sqrt{5} [8], and all but finitely many quintic fields [28]; to fields with restricted ramification at small primes [33, 54]; or to special towers, such as the cyclotomic Zp\mathbb{Z}_p-extension of Q\mathbb{Q} [49]. These hypotheses are used to deal with exceptional collections of elliptic curves, given by points on certain modular curves, which are not known to be chainable to a modular one. Lifting theorems for small residual images [48, 29] shrink these exceptional cases, though they fall short of getting rid of them altogether.

Compared to the totally real case, the CM case is more recent and harder: automorphic forms for GL2\mathrm{GL}_2 over a CM field occur in the cohomology of locally symmetric spaces that are not algebraic varieties, in several degrees, and are not directly related to geometry. As a consequence, modularity lifting theorems are harder to establish. The seminal work of Calegari–Geraghty [12] generalized the Taylor–Wiles method to this setting, conditionally on conjectures on torsion in the cohomology of locally symmetric spaces. Since then many of these conjectures have been proven or bypassed [37, 1], leading to potential modularity results over CM fields [1 Theorem 1.0.1] and, more recently, to modularity lifting theorems in weight two of strength comparable to the totally real case, due to Caraiani–Newton [14 Theorem 5.2]. Combined with the residual modularity results of Allen–Khare–Thorne [2], the latter led to major progress over CM fields. For instance, Caraiani–Newton showed that all elliptic curves over an imaginary quadratic field KK with X0(15)(K)X_0(15)(K) finite are modular, such as K=Q(−d)K = \mathbb{Q}(\sqrt{-d}) with d=1,2,3,5d = 1, 2, 3, 5 [14 Theorem 1.1], and more generally proved modularity over CM fields not containing ζ5\zeta_5 under hypotheses on ρˉE,3\bar{\rho}_{E,3} or ρˉE,5\bar{\rho}_{E,5} [14 Theorem 6.1]. The main thrust behind these results is the modularity lifting step.

Most recently, OpenAI announced a proof that every elliptic curve over every imaginary quadratic field is modular [35 Theorem 1.1], by introducing a clever new idea for the chaining step. Unlike previous arguments, which exploit small primes such as 2, 3, 5, 7 and special moduli problems, OpenAI’s chain passes through a congruence modulo an arbitrarily large auxiliary prime pp, after a large solvable base change. The use of such large primes is traditionally overlooked, since it is believed (e.g. by the Frey–Mazur conjecture, see [31]) that there are very few congruences between elliptic curves over a fixed field for large pp. Since its geometric ingredients work over any number field, one expects OpenAI’s argument to extend beyond imaginary quadratic fields. This is indeed the case.

Theorem 1.1. Assume Inputs 2.1 and 2.2 from [35]. Then every elliptic curve over every totally real number field is modular. Assume in addition Input 4.1 from [35]. Then every elliptic curve over every CM number field KK with ζ5∉K\zeta_5 \notin K is modular.

Remark 1.2.

  1. The inputs are the geometric results of [35]. Input 2.1 gives the basic properties of the family of motives that carries the chain of congruences; Input 2.2 gives the large monodromy of the family, which gives large residual images along the chain; and Input 4.1 concerns local properties of the family.

  1. In the totally real case we use no modularity lifting theorem over CM fields.

Outline

Let EE be an elliptic curve over KK. Fix n=3pn=3p for a large auxiliary prime pp. OpenAI constructs a family of quadratic twists BtB_t and cyclic degree nn covers Cn(t)→BtC_n(t) \to B_t over an open subset U⊂At1U \subset\mathbb{A}^1_t. Cn(t)C_n(t) comes with an action of a dihedral group defined over F=K(μn)F=K(\mu_n). The key feature of the construction is that the integral ℓ\ell-adic cohomology H1(Cn(t))H^1(C_n(t)) is free of rank 2 over Zℓ[μn]\mathbb{Z}_\ell[\mu_n] for all ℓ\ell. In particular, it decomposes into rank-two pieces VθV_\theta with Hodge–Tate weights 0, 1, indexed by characters θ\theta of μn\mu_n. One has V1=H1(Bt)V_1=H^1(B_t), and for η\eta of order three, Vη=H1(At)V_\eta=H^1(A_t) for an elliptic curve AtA_t. The reduction of VθV_\theta modulo ℓ\ell only depends on θ\theta mod ℓ\ell. With ψ\psi of order pp this gives the chain of congruences (for each tt)

Vη≡pVηψ≡3Vψ≡pV1,V_\eta\equiv_p V_{\eta\psi} \equiv_3 V_\psi\equiv_p V_1,

and the idea is to propagate modularity from the left to the right. In order to do this propagation, the essential point is to keep the involved residual image large at each step so that a modularity lifting theorem applies. As pp is chosen to be large, the key point is to arrange tt so that VψV_\psi mod 3 has large image. This is achieved by the fact that the family has two opposite transvections around two points of the parameter space where the cover degenerates, and thus Hilbert irreducibility guarantees there are many such choices of tt. As a byproduct, one can also guarantee that the mod 5 image of Vη=H1(At)V_\eta=H^1(A_t) is large, and hence that AtA_t is modular. Importantly, this gives the starting point for the chain.

For the case of totally real KK, we first note that the VθV_\theta in the construction are only defined over the CM field K(μ3p)K(\mu_{3p}). However, twisting the cover by the reflection descends them to totally odd systems WθW_\theta over L=K(ζ3p+ζ3p−1)L=K(\zeta_{3p}+\zeta_{3p}^{-1}) with the same congruences, and thus one can follow the above argument with the (less restrictive) modularity lifting theorems over totally real fields; thus one merely needs the existence of the chain of congruences to run the argument.

Over CM KK, invoking the Caraiani–Newton modularity lifting theorem requires one to verify two additional hypotheses: that the residual representations are decomposed generic and that the motives in the chain have the same local ℓ\ell-adic Hodge-theoretic behavior at the coefficient prime ℓ\ell. The first is easily dealt with by applying Hilbert irreducibility to a Weil restriction, while the second is dealt with by imposing local conditions on tt at the places above 3p3p, using an analysis of the reduction of the family near its compact-type degeneration.

Remark 1.3. 1. The main reason for the hypothesis ζ5∉K\zeta_5 \notin K in the CM case was so that one can apply the existing modularity lifting of Caraiani–Newton at the prime 5 [14 Theorem 6.1]. An improvement to this has been obtained by AI, as released in [13]. With this improvement, OpenAI’s argument now covers all CM fields. Nevertheless, we find it instructive to treat the totally real field case on its own, as it illustrates clearly the distinctive role of the two parts of the strategy.

  1. It is essential that our chain connects to an elliptic curve AtA_t, which is the reason to work with the prime 3 in the middle congruence of the chain. Replacing the prime 3 with a larger prime will instead connect H1(Bt)H^1(B_t) to some other motive for which one does not know how to establish modularity. On the other hand, one could contemplate using the prime 2 to establish the modularity of AtA_t along the lines of [2].

  1. One might wonder whether it is possible to use a different prime to get the modularity of the starting motive AtA_t. The prime 3 does not work, since H1(At)H^1(A_t) turns out to always be congruent to a quadratic twist of EE mod 3.

Remark 1.4. The mechanism has antecedents in families of curves with real multiplication [34, 43, 19, 22], and above all in Darmon’s work on rigid local systems [17, 18]. There, hypergeometric families are related by reduction modulo primes dividing the orders of local monodromy, and modularity is propagated conditionally on a lifting conjecture [17 Conjecture 1.1].

§ 2 sets up the geometry and the congruences, § 3 and § 4 treat the totally real and CM cases, and § 5 contains descent and local–global compatibility.

AI disclosure. The first draft of this note, apart from the introduction, was generated by the AI system Astra. The introduction was written by the author. The whole text was proofread and edited, and its references checked, by Claude (Anthropic; model Claude Opus 5.5).

Conventions. Write Gk=Gal⁡(k‾/k)G_k = \operatorname{Gal}(\overline{k}/k) and H1(X)=Heˊt1(Xk‾,Qℓ)H^1(X) = H^1_{\mathrm{\acute{e}t}}(X_{\overline{k}}, \mathbb{Q}_{\ell}), extending scalars to Q‾ℓ\overline{\mathbb{Q}}_{\ell} when needed. The cyclotomic character ϵℓ\epsilon_{\ell} has Hodge–Tate weight −1-1, so det⁡H1(E)=ϵℓ−1\det H^1(E) = \epsilon_{\ell}^{-1} and HT⁡τ(H1(E))={0,1}\operatorname{HT}_{\tau}(H^1(E)) = \{0, 1\} for every τ\tau. Bars denote semisimplified reductions. We fix embeddings of Q‾\overline{\mathbb{Q}} into C\mathbb{C} and into every Q‾ℓ\overline{\mathbb{Q}}_{\ell}. When applying [11, 23] we dualize to the Tate-module convention.

The geometric construction

Let kk be a number field, E:y2=h(x)E:y^2=h(x) an elliptic curve over kk, and n≥3n \ge3 odd. Put

Bt:h(t)y2=h(x),Rt=(t,1),St=[n]Rt.B_t:h(t)y^2=h(x), \qquad R_t=(t,1), \qquad S_t=[n]R_t.

and let U⊂Ak1U \subset\mathbb{A}^1_k be the open set where h(t)≠0h(t) \ne0 and O,±Rt,±StO,\pm R_t,\pm S_t are distinct. The divisor (St)−(−St)−n(Rt)+n(−Rt)(S_t)-(-S_t)-n(R_t)+n(-R_t) is principal; let ftf_t be its function with ft(O)=1f_t(O)=1, so that ft(−Z)=ft(Z)−1f_t(-Z)=f_t(Z)^{-1}. For m∣nm\mid n let Cm(t)C_m(t) be the normalization of BtB_t in

Ym=ft(Z),(1)Y^m=f_t(Z), \tag*{(1)}

with rotations uζ(Z,Y)=(Z,ζY)u_{\zeta}(Z,Y)=(Z,\zeta Y), ζ∈μm\zeta\in\mu_m, and reflection w(Z,Y)=(−Z,Y−1)w(Z,Y)=(-Z,Y^{-1}), so that wuζw=uζ−1wu_{\zeta}w=u_{\zeta^{-1}}. The curve and ww are defined over kk, the rotations over k(μm)k(\mu_m). The map Cn→CmC_n\to C_m is Y↦Yn/mY\mapsto Y^{n/m}, and Cm(t)/μm=BtC_m(t)/\mu_m=B_t.

Geometric input 2.1 (Cover and lattice; [35 Lemma 3.1, Propositions 3.3 and 4.2, Remark 4.3]). The curve Cm(t)C_m(t) is smooth of genus mm, and for every prime ℓ\ell, including ℓ∣m\ell\mid m,

Heˊt1(Cm,k‾,Zℓ)≅Zℓ[μm](2)H^1_{\mathrm{\acute{e}t}}(C_{m,\overline{k}},\mathbb{Z}_{\ell})\cong\mathbb{Z}_{\ell}[\mu_m] \tag*{(2)}

as modules over the group ring of the rotation group. After pullback along E→A1E\to\mathbb{A}^1, (t,s)↦t(t,s)\mapsto t, over which Bt≅EB_t\cong E and RtR_t becomes the moving point R=(t,s)R=(t,s), the Jacobian family extends across R=OR=O with fiber EmE^m, whose factors the rotations permute regularly over k(μm)k(\mu_m).

Geometric input 2.2 (Monodromy; [35 Theorem 4.5]). After any complex embedding of kk, geometric monodromy contains, in one basis of the free module (2), two opposite transvections with off-diagonal entries ±(2−u−u−1)\pm(2-u-u^{-1}), where uu is Y↦e2πi/mYY\mapsto e^{2\pi i/m}Y. On the eigenspace where uu acts by λ≠1\lambda\ne1 they are

Ta=(1νacλ01),Tb=(10νbcλ1),cλ=2−λ−λ−1,νa,νb∈{±1}.(3)T_a=\begin{pmatrix}1&\nu_a c_{\lambda}\\0&1\end{pmatrix},\qquad T_b=\begin{pmatrix}1&0\\\nu_b c_{\lambda}&1\end{pmatrix},\qquad c_{\lambda}=2-\lambda-\lambda^{-1},\qquad \nu_a,\nu_b\in\{\pm1\}. \tag*{(3)}

The common basis matters: triangular matrices in unrelated bases would not give irreducibility. Both inputs hold over any number field.

Rank-two systems

Let F⊃k(μn)F \supset k(\mu_n). For a character θ\theta of μn\mu_n let Vθ=Vθ,ℓ⊂Heˊt1(Cn,F‾,Q‾ℓ)V_\theta= V_{\theta,\ell} \subset H^1_{\mathrm{\acute{e}t}}(C_{n,\overline{F}}, \overline{\mathbb{Q}}_\ell) be the θ\theta-eigenspace of the rotations; it is GFG_F-stable, of dimension two by (2). Cup product ⟨ , ⟩\langle\ ,\ \rangle is rotation-invariant, so it pairs VθV_\theta perfectly with Vθ−1=w∗VθV_{\theta^{-1}} = w^*V_\theta, and

[x,y]θ=⟨x,w∗y⟩(4)[x,y]_\theta= \langle x,w^*y\rangle \tag*{(4)}

is a nondegenerate GFG_F-equivariant pairing on VθV_\theta with values in Q‾ℓ(−1)\overline{\mathbb{Q}}_\ell(-1). It is alternating, as ⟨y,w∗x⟩=⟨w∗y,x⟩=−⟨x,w∗y⟩\langle y,w^*x\rangle= \langle w^*y,x\rangle= -\langle x,w^*y\rangle. Hence det⁡Vθ=ϵℓ−1\det V_\theta= \epsilon_\ell^{-1}. Since w∗w^* exchanges the θ\theta- and θ−1\theta^{-1}-parts of H1,0H^{1,0}, and complex conjugation exchanges the θ−1\theta^{-1}-part of H1,0H^{1,0} with the θ\theta-part of H0,1H^{0,1}, each θ\theta-part of H1,0H^{1,0} is a line; so the labelled Hodge–Tate weights are {0,1}\{0,1\}. The VθV_\theta are semisimple, potentially semistable, and form compatible systems (use the character projector and traces of rotations composed with Frobenius).

The trivial character gives V1=H1(Bt)V_1 = H^1(B_t). Put

Dt=C3(t)/⟨w⟩,At=Jac⁡(Dt),(5)D_t = C_3(t)/\langle w\rangle,\qquad A_t = \operatorname{Jac}(D_t), \tag*{(5)}

both defined over kk. On H1(C3)H^1(C_3), ww acts by −1-1 on H1(Bt)H^1(B_t), since it induces inversion, and swaps the other two character spaces; so DtD_t has genus one. For η\eta of order three, projecting the ww-invariants to the η\eta-space gives

H1(At)∣GF≃Vη,H1(Bt)∣GF≃V1.(6)H^1(A_t)|_{G_F} \simeq V_\eta,\qquad H^1(B_t)|_{G_F} \simeq V_1. \tag*{(6)}

Congruences

Let p>100p > 100 be prime, n=3pn = 3p, let η,ψ\eta,\psi be characters of orders 3,p3,p, and χ=ηψ\chi= \eta\psi. For O\mathcal{O} the integers of a large enough finite extension of Qℓ\mathbb{Q}_\ell, the lattice

Tθ,ℓ=Heˊt1(Cn,F‾,Zℓ)⊗Zℓ[μn],θO⊂Vθ,ℓT_{\theta,\ell} = H^1_{\mathrm{\acute{e}t}}(C_{n,\overline{F}},\mathbb{Z}_\ell) \otimes_{\mathbb{Z}_\ell[\mu_n],\theta} \mathcal{O} \subset V_{\theta,\ell}

is GFG_F-stable and free of rank two by (2). Its reduction depends only on θ\theta mod ℓ\ell, and roots of unity of ℓ\ell-power order reduce to 11, so

Vη≡pVχ≡3Vψ≡pV1,(7)V_\eta\equiv_p V_\chi\equiv_3 V_\psi\equiv_p V_1, \tag*{(7)}

where ≡ℓ\equiv_\ell means isomorphic semisimplified reductions modulo ℓ\ell. No division by 33 or pp occurs.

Residual irreducibility

Take k=Kk = K and consider the pairs

(m,ℓ)=(3,5), (3,p), (p,3),(8)(m,\ell) = (3,5),\ (3,p),\ (p,3), \tag*{(8)}

with mm the order of the character and ℓ\ell the residual characteristic. As ℓ∤m\ell\nmid m, cλ=−(λ−1)2/λc_\lambda= -(\lambda- 1)^2/\lambda is nonzero modulo ℓ\ell, so TaT_a and TbT_b have distinct fixed lines and geometric monodromy acts absolutely irreducibly on the reduction of each character space of order mm. For m=3m = 3 we have cλ=3c_\lambda= 3, so modulo 55 these matrices generate SL⁡2(F5)\operatorname{SL}_2(\mathbb{F}_5).

Changing the coefficient prime

If one member of the compatible system VθV_\theta, or WθW_\theta of § 3.2, is realized by an automorphic representation π\pi, then so is every member: compare Frobenius polynomials and apply Chebotarev and Brauer–Nesbitt. This lets us change the coefficient prime along (7).

The totally real case

Let KK be totally real.

Modularity lifting

The following is a consequence of [23 Theorem 3], [11, Théorème 3.2.2]:

Proposition 3.1. Let LL be totally real, ℓ≠2,5\ell\ne2,5 a prime, and A/LA/L an abelian variety. Let r:GL→GL⁡2(Q‾ℓ)r:G_L \to\operatorname{GL}_2(\overline{\mathbb{Q}}_\ell) be the image of an idempotent of End⁡L(A)⊗Q‾\operatorname{End}_L(A) \otimes\overline{\mathbb{Q}} on H1(AL,Q‾ℓ)H^1(A_L,\overline{\mathbb{Q}}_\ell), with det⁡r=ϵℓ−1\det r=\epsilon_\ell^{-1} and labelled weights {0,1}\{0,1\}. If rˉ\bar r is modular and rˉ∣GL(ζℓ)\bar r|_{G_L(\zeta_\ell)} is absolutely irreducible, then rr is modular.

We remark that in the above Proposition, there is no need to arrange any local conditions for rr to match with a modular lift.

The reflection twist

Let M=K(μ3p)M=K(\mu_{3p}) and L=K(ζ3p+ζ3p−1)L=K(\zeta_{3p}+\zeta_{3p}^{-1}). Then L/KL/K is abelian and totally real, and M/LM/L is quadratic with character δ=(−1)e\delta=(-1)^e, e:GL→Z/2Ze:G_L\to\mathbb{Z}/2\mathbb{Z}. Twisting CmC_m by the cocycle g↦we(g)g\mapsto w^{e(g)} (a cocycle since ww is defined over KK) gives a curve Cmδ/LC_m^\delta/L, whose cohomology is that of CmC_m with

ρδ(g)=(w∗)e(g)ρ(g).(9)\rho^\delta(g)=(w^*)^{e(g)}\rho(g). \tag*{(9)}

Lemma 3.2. The rotations and ww of CmδC_m^\delta are defined over LL. The θ\theta-eigenspaces WθW_\theta of H1(Cm,L‾δ)H^1(C^\delta_{m,\overline{L}}) are totally odd GLG_L-representations with determinant ϵℓ−1\epsilon_\ell^{-1} and labelled weights {0,1}\{0,1\}, and

Wη≡pWχ≡3Wψ≡pW1,(10)W_\eta\equiv_p W_\chi\equiv_3 W_\psi\equiv_p W_1, \tag*{(10)}
Wη≃H1(At)∣GL,W1≃H1(Bt)∣GL⊗δ.(11)W_\eta\simeq H^1(A_t)|_{G_L},\qquad W_1 \simeq H^1(B_t)|_{G_L}\otimes\delta. \tag*{(11)}

Proof. An element g∈GLg\in G_L acts on μm\mu_m by ζ↦ζ±1\zeta\mapsto\zeta^{\pm1} according to e(g)e(g), and we(g)ug(ζ)we(g)=uζw^{e(g)}u_g(\zeta)w^{e(g)}=u_\zeta, so the twisted descent datum fixes each uζu_\zeta and ww. The integral lattice is unchanged and commutes with Galois, which gives (10) as in § 2.2. Since ww and cup product are defined over LL, (4) gives the determinant, which is −1-1 on complex conjugations. The quotient C3δ/wC^\delta_3/w is DtD_t, giving the first endpoint; ww acts by −1-1 on H1(Bt)H^1(B_t), so (9) gives the second.

The parameter

Let E/KE/K be non-CM. By Serre’s open image theorem [38 §4.2, Théorème 2], choose p>100p>100 with ρˉE,p(GK)⊃SL⁡2(Fp)\bar\rho_{E,p}(G_K)\supset\operatorname{SL}_2(\mathbb{F}_p). As SL⁡2(Fp)\operatorname{SL}_2(\mathbb{F}_p) is perfect and D=K(μ15p)D=K(\mu_{15p}) is abelian over KK, ρˉE,p(GD)=SL⁡2(Fp)\bar\rho_{E,p}(G_D)=\operatorname{SL}_2(\mathbb{F}_p).

The local systems H1(Cm,Fℓ)H^1(C_m,\mathbb{F}_\ell) over UDU_D, for the pairs (8), cut out a finite quotient QQ of π1(UK)\pi_1(U_K) mapping onto Gal⁡(D/K)\operatorname{Gal}(D/K) (take the normal core of their common kernel). By Hilbert irreducibility [40 Proposition 3.3.1 and Theorem 3.4.1] there is t∈U(K)t\in U(K) with GK→QG_K\to Q surjective; then GDG_D maps onto ker⁡(Q→Gal⁡(D/K))\ker(Q\to\operatorname{Gal}(D/K)), which contains geometric monodromy. By § 2.3, the reductions of VηV_\eta modulo 55 and pp and of VψV_\psi modulo 33 are absolutely irreducible on GD⊂GL(ζℓ)G_D\subset G_L(\zeta_\ell), and so are those of the WθW_\theta, since δ∣GD=1\delta|_{G_D}=1. The mod-55 image of AtA_t contains two noncommuting elements of order 55, so AtA_t is non-CM.

Proof for totally real fields

By [23 Theorem 3], At/LA_t/L is modular, hence so is WηW_\eta. Apply Proposition 3.1 as follows, changing the coefficient prime after each row (§ 2.4):

ModularPrimeTargetResidual irreducibility from
WηW_\etappWχW_\chiorder-three piece modulo pp
WχW_\chi33WψW_\psiorder-pp piece modulo 33
WψW_\psippW1W_1EE modulo pp

Table 1.

By (11), Bt/LB_t/L is modular. Descending along L/KL/K (Lemma 5.1) and twisting back to EE proves the theorem for totally real KK.

The CM case

Let KK be CM with ζ5∉K\zeta_5 \notin K, E/KE/K non-CM, and F=K(μ3p)F = K(\mu_{3p}).

The lifting theorem

The seed is [14 Theorem 6.1]: an elliptic curve A/KA/K is modular if ρˉA,5\bar{\rho}_{A,5} is decomposed generic and absolutely irreducible on GK(ζ5)G_{K(\zeta_5)}. For the chain we use [14 Theorem 5.2] at ℓ=3,p\ell= 3, p. Besides the determinant and weight conditions of § 2, it requires ρˉ\bar{\rho} to be decomposed generic and irreducible on GF(ζℓ)G_{F(\zeta_\ell)}, and the following match at each v∣ℓv \mid\ell with the automorphic source π\pi: either ρ\rho is potentially crystalline, rι(π)r_\iota(\pi) is potentially ordinary exactly when ρ\rho is, and rec⁡(πv)\operatorname{rec}(\pi_v) has zero monodromy; or ρ\rho and rι(π)r_\iota(\pi) are not potentially crystalline and π\pi is ι\iota-ordinary at vv. Here ρˉ\bar{\rho} is decomposed generic if some prime q≠ℓq \ne\ell splits completely in FF and, at each v∣qv \mid q, ρˉ\bar{\rho} is unramified with no ratio of Frobenius eigenvalues equal to qq.

Local reduction

Geometric input 4.1 (Local reduction; [35 Propositions 7.3 and 7.5, Corollary 7.6]). For each v0∣3pv_0 \mid3p of KK there is a nonempty open subset of U(Kv0)U(K_{v_0}) such that, for tt in it, the Jacobian of Cn(t)C_n(t) has at every v∣v0v \mid v_0 of FF the same potential reduction type as EnE^n: good ordinary, good with all slopes 1/21/2, or semistable of toric rank nn.

In [35] this is stated over imaginary quadratic fields, but the argument is general. On each character space the three types give, respectively, potentially crystalline and ordinary, potentially crystalline and non-ordinary, and monodromy of rank one, for every coefficient prime [35 Corollary 7.6].

Local matching

Let π\pi realize the source of a row of (7), and let v∣ℓv \mid\ell. The source and the target are summands of the same Jacobian, so by Input 4.1 they are both potentially crystalline or neither, with the same potential ordinarity. It remains to check πv\pi_v, using semisimplified local–global compatibility [50 Theorem 1] at another prime. In the potentially good case the two Frobenius eigenvalues have equal absolute value, so rec⁡(πv)\operatorname{rec}(\pi_v) has zero monodromy. In the toric case they differ by the norm; as πv\pi_v is generic, it is a twist of Steinberg, by a quadratic character since the central character is trivial. Then π\pi is ι\iota-ordinary at vv by the argument of [14 Lemma 6.1.3(4)].

Decomposed genericity

Regular semisimple elements. For a pair in (8), tr⁡(TaTbj)=2+νaνbjc2\operatorname{tr}(T_aT_b^j)=2+\nu_a\nu_bjc^2, and an element of SL⁡2\operatorname{SL}_2 is regular semisimple if and only if its trace is not ±2\pm2. For m=3m=3 take j=2j=2: the traces 2020 and −16-16 differ from ±2\pm2 by 1818, 2222, or 1414, so they are not ±2\pm2 modulo 55 or pp. For (m,ℓ)=(p,3)(m,\ell)=(p,3) take j=1j=1: a bad trace forces (λ−1)4=±4λ2(\lambda-1)^4=\pm4\lambda^2, so λ∈F3a\lambda\in\mathbb{F}_{3^a} with a≤4a\leq4, impossible since λ\lambda has order p>80≥3a−1p>80\geq3^a-1.

All conjugates. Let d=[K:Q]d=[K:\mathbb{Q}], KnorK^{\mathrm{nor}} the Galois closure of KK, DD a finite Galois extension of Q\mathbb{Q} containing Knor(μ15p)K^{\mathrm{nor}}(\mu_{15p}), and X=Res⁡K/QU⊂AQdX=\operatorname{Res}_{K/\mathbb{Q}}U\subset\mathbb{A}_{\mathbb{Q}}^d, so that XD≃∏σ:K↪DUDσX_D\simeq\prod_{\sigma:K\hookrightarrow D}U_D^\sigma. The local systems H1(Cm,Fℓ)H^1(C_m,\mathbb{F}_\ell) on the factors, for the pairs in (8), cut out a finite quotient QQ of π1(X)\pi_1(X) mapping onto Gal⁡(D/Q)\operatorname{Gal}(D/\mathbb{Q}). By Hilbert irreducibility with weak approximation [35 Lemma 8.3] (see also [40 Proposition 3.3.1, Theorems 3.4.1 and 3.5.3]) there is t∈X(Q)=U(K)t\in X(\mathbb{Q})=U(K), in any prescribed nonempty open subsets of finitely many U(Kv0)U(K_{v_0}), with GQ→QG_{\mathbb{Q}}\to Q surjective. Then GDG_D maps onto ker⁡(Q→Gal⁡(D/Q))\ker(Q\to\operatorname{Gal}(D/\mathbb{Q})), which contains geometric monodromy, a product over the factors. So for each pair some g∈GDg\in G_D acts regularly semisimply on every nontrivial character space in every factor. By Chebotarev, infinitely many primes qq have Frobenius conjugate to gg. They split in DD, so q≡1(modℓ)q\equiv1\pmod{\ell}, and every Frobenius above qq is regular semisimple on every character space, since conjugation permutes factors and characters. This is decomposed genericity: over KK for At[5]A_t[5], and over FF for the pairs (3,p)(3,p) and (p,3)(p,3). As in § 3.3, At[5]A_t[5] is absolutely irreducible on GK(ζ5)G_{K(\zeta_5)}, AtA_t is non-CM, and the residual representations are absolutely irreducible on GD⊂GF(ζℓ)G_D\subset G_{F(\zeta_\ell)}. This replaces the two conjugates of [35 Proposition 8.4]; no Galois assumption on KK is needed.

The final switch. Choose pp so that ρˉEσ,p(GKnor)⊃SL⁡2(Fp)\bar{\rho}_{E^\sigma,p}(G_{K^{\mathrm{nor}}})\supset\operatorname{SL}_2(\mathbb{F}_p) for every σ:K↪Knor\sigma:K\hookrightarrow K^{\mathrm{nor}} [38 §4.2, Théorème 2], and

p>max⁡{100,2d+1}.(12)p>\max\{100,2d+1\}. \tag*{(12)}

Then GD0G_{D_0}, for D0=Knor(μ15p)D_0=K^{\mathrm{nor}}(\mu_{15p}), has image SL⁡2(Fp)\operatorname{SL}_2(\mathbb{F}_p) on each Eσ[p]E^\sigma[p]. Exactly 2p22p^2 elements of SL⁡2(Fp)\operatorname{SL}_2(\mathbb{F}_p) have trace ±2\pm2, so the proportion of the joint image that is not regular semisimple in some factor is at most

d2p2p(p2−1)=2dpp2−1<1,(13)d\frac{2p^2}{p(p^2-1)}=\frac{2dp}{p^2-1}<1, \tag*{(13)}

as p(p−2d)>1p(p-2d)>1; for d=2d=2 this is [35 Lemma 8.2]. Chebotarev for a good element, avoiding the primes of bad reduction of BtB_t once tt is fixed, gives decomposed genericity of ρˉE,p\bar{\rho}_{E,p}, hence of the reductions of VψV_\psi and V1V_1, over FF.

The parameter

Choose pp as in (12), and by the specialization above a single t∈U(K)t\in U(K) in the open sets of Input 4.1 for all v0∣3pv_0\mid3p. Then:

  1. ρˉAt,5\bar{\rho}_{A_t,5} is decomposed generic over KK and absolutely irreducible on GK(ζ5)G_{K(\zeta_5)}, and AtA_t is non-CM;

  1. each residual representation in (7) is decomposed generic over FF and absolutely irreducible on GF(ζℓ)G_{F(\zeta_\ell)};

  1. above 3p3p, all pieces have the reduction type of Input 4.1, so § 4.3 applies.

Proof for CM fields

By [14 Theorem 6.1], AtA_t is modular over KK, hence over FF by solvable base change, cuspidally since AtA_t is non-CM; so VηV_\eta is automorphic. Apply [14 Theorem 5.2] as follows, changing the coefficient prime after each row:

AutomorphicPrimeTargetResidual input
VηV_\etappVχV_\chi(m,ℓ)=(3,p)(m,\ell)=(3,p)
VχV_\chi33VψV_\psi(m,ℓ)=(p,3)(m,\ell)=(p,3)
VψV_\psippV1V_1[13]

Table 1 (PDF p. 9).

Thus V1=H1(Bt)∣GFV_1=H^1(B_t)|_{G_F} is automorphic. Descending along F/KF/K (Lemma 5.1) and twisting back gives modularity of E/KE/K.

Remark 4.2. We do not claim ζ5∉F\zeta_5 \notin F. For example, K=Q(ζ15+ζ15−1,i)K=\mathbb{Q}(\zeta_{15}+\zeta_{15}^{-1},i) is CM and does not contain ζ5\zeta_5, but K(ζ3)=Q(ζ60)K(\zeta_3)=\mathbb{Q}(\zeta_{60}) does. This is why AtA_t is made modular over KK before base change, and why the ramification argument of [35 (5.2)] is not used. The later lifts are at 33 and pp, so the exceptional case at 55 does not arise. For CM fields containing ζ5\zeta_5 we have no seed.

Descent and local–global compatibility

Lemma 5.1. Let F/KF/K be an abelian extension of totally real fields or of CM fields, and rr a rank-two representation of GKG_K with r∣GFr|_{G_F} absolutely irreducible and automorphic. Then rr is automorphic.

This is standard: descend one cyclic step at a time by solvable base change [3 Chapter 3, Theorems 4.2(d) and 5.1], and adjust by a character using Schur’s lemma.

For non-CM EE, the resulting πE\pi_E realizes H1(E)H^1(E) at every coefficient prime, with the expected local parameters at finite places by local–global compatibility (over CM fields, [14 Lemma 6.1.3]); in particular L(E/K,s)=L(πE,s−1/2)L(E/K,s)=L(\pi_E,s-1/2).

If EE has CM by an imaginary quadratic field TT, then H1(E)∣GKTH^1(E)|_{G_{KT}} is a sum of two Hecke-character systems; use automorphic induction if T⊄KT\not\subset K (always the case for totally real KK), and the isobaric sum if T⊂KT\subset K.

Source and status

We use [35] at commit adc7f12. The new arguments here are the reflection twist, the dd-conjugate specialization with [13], the order of seed modularity and base change in the CM case, and the two applications of existing lifting theorems. The geometric inputs are taken from [35], and Theorem 1.1 is stated conditionally on them. The construction uses special features of elliptic curves: the group law gives the two-branch-point cover, every character space has rank two, the order-three quotient is an elliptic curve, and one integral lattice gives congruences at several primes.

References

  1. [1]P. B. Allen, F. Calegari, A. Caraiani, T. Gee, D. Helm, B. V. Le Hung, J. Newton, P. Scholze, R. Taylor, and J. A. Thorne, Potential automorphy over CM fields, Ann. of Math. (2) 197 (2023), 897–1113.arxiv.org/abs/1812.09999
  2. [2]P. B. Allen, C. Khare, and J. A. Thorne, Modularity of GL₂(Fₚ)-representations over CM fields, Camb. J. Math. 11 (2023), 1–158.
  3. [3]J. Arthur and L. Clozel, Simple algebras, base change, and the advanced theory of the trace formula, Annals of Mathematics Studies 120, Princeton University Press, 1989.DOI
  4. [4]T. Barnet-Lamb, T. Gee, and D. Geraghty, Congruences between Hilbert modular forms: constructing ordinary lifts, Duke Math. J. 161 (2012), 1521–1580.
  5. [5]T. Barnet-Lamb, T. Gee, and D. Geraghty, Congruences between Hilbert modular forms: constructing ordinary lifts, II, Math. Res. Lett. 20 (2013), 67–72.
  6. [6]T. Berger and G. Harcos, ℓ-adic representations associated to modular forms over imaginary quadratic fields, Int. Math. Res. Not. IMRN (2007), no. 23, Art. ID rnm113.
  7. [7]D. Blasius and J. D. Rogawski, Motives for Hilbert modular forms, Invent. Math. 114 (1993), 55–87.DOI
  8. [8]J. Box, Elliptic curves over totally real quartic fields not containing √5 are modular, Trans. Amer. Math. Soc. 375 (2022), 3129–3172.
  9. [9]G. Boxer, F. Calegari, T. Gee, and V. Pilloni, Abelian surfaces over totally real fields are potentially modular, Publ. Math. Inst. Hautes Études Sci. 134 (2021), 153–501.arxiv.org/abs/1812.09269
  10. [10]C. Breuil, B. Conrad, F. Diamond, and R. Taylor, On the modularity of elliptic curves over ℚ: wild 3-adic exercises, J. Amer. Math. Soc. 14 (2001), 843–939.DOI
  11. [11]C. Breuil and F. Diamond, Formes modulaires de Hilbert modulo p et valeurs d’extensions entre caractères galoisiens, Ann. Sci. Éc. Norm. Supér. (4) 47 (2014), 905–974.DOI
  12. [12]F. Calegari and D. Geraghty, Modularity lifting beyond the Taylor–Wiles method, Invent. Math. 211 (2018), 297–433; Correction, Invent. Math. 227 (2022), 855–856.arxiv.org/abs/1207.4224
  13. [13]CaptainSude, Every elliptic curve over a CM number field is modular, manuscript, v1.0.0, October 7, 2026, github.com/CaptainSude.
  14. [14]A. Caraiani and J. Newton, On the modularity of elliptic curves over imaginary quadratic fields, arXiv:2301.10509v3 (2025).arxiv.org/abs/2301.10509
  15. [15]H. Carayol, Sur les représentations ℓ-adiques associées aux formes modulaires de Hilbert, Ann. Sci. Éc. Norm. Supér. (4) 19 (1986), 409–468.DOI
  16. [16]B. Conrad, F. Diamond, and R. Taylor, Modularity of certain potentially Barsotti–Tate Galois representations, J. Amer. Math. Soc. 12 (1999), 521–567.DOI
  17. [17]H. Darmon, Modularity of fibres in rigid local systems, Ann. of Math. (2) 149 (1999), 1079–1086.arxiv.org/abs/math/9905208
  18. [18]H. Darmon, Rigid local systems, Hilbert modular forms, and Fermat’s last theorem, Duke Math. J. 102 (2000), 413–449.DOI
  19. [19]H. Darmon and J.-F. Mestre, Courbes hyperelliptiques à multiplications réelles et une construction de Shih, Canad. Math. Bull. 43 (2000), 304–311.DOI
  20. [20]M. Derickx, F. Najman, and S. Siksek, Elliptic curves over totally real cubic fields are modular, Algebra Number Theory 14 (2020), 1791–1800.DOI
  21. [21]F. Diamond, On deformation rings and Hecke rings, Ann. of Math. (2) 144 (1996), 137–166.DOI
  22. [22]J. S. Ellenberg, Endomorphism algebras of Jacobians, Adv. Math. 162 (2001), 243–271.DOI
  23. [23]N. Freitas, B. V. Le Hung, and S. Siksek, Elliptic curves over real quadratic fields are modular, Invent. Math. 201 (2015), 159–206.arxiv.org/abs/1310.7088
  24. [24]G. Frey, Links between stable elliptic curves and certain diophantine equations, Ann. Univ. Sarav. Ser. Math. 1 (1986), no. 1, iv+40 pp.
  25. [25]T. Gee, Automorphic lifts of prescribed types, Math. Ann. 350 (2011), 107–144.arxiv.org/abs/0810.1877
  26. [26]M. Harris, K.-W. Lan, R. Taylor, and J. Thorne, On the rigid cohomology of certain Shimura varieties, Res. Math. Sci. 3 (2016), Paper No. 37.
  27. [27]M. Harris, D. Soudry, and R. Taylor, ℓ-adic representations associated to modular forms over imaginary quadratic fields. I. Lifting to GSp₄(Q), Invent. Math. 112 (1993), 377–411.
  28. [28]Y. Ishitsuka, T. Ito, and S. Yoshikawa, The modularity of elliptic curves over all but finitely many totally real fields of degree 5, Res. Number Theory 8 (2022), Paper No. 82.DOI
  29. [29]S. Kalyanswamy, Remarks on automorphy of residually dihedral representations, Math. Res. Lett. 25 (2018), 1285–1304.arxiv.org/abs/1607.04750
  30. [30]M. Kisin, Moduli of finite flat group schemes, and modularity, Ann. of Math. (2) 170 (2009), 1085–1180.DOI
  31. [31]A. Kraus and J. Oesterlé, Sur une question de B. Mazur, Math. Ann. 293 (1992), 259–275.DOI
  32. [32]R. P. Langlands, Problems in the theory of automorphic forms, in: Lectures in Modern Analysis and Applications III, C. T. Taam (ed.), Lecture Notes in Math. 170, Springer, 1970, 18–61.
  33. [33]B. V. Le Hung, Modularity of some elliptic curves over totally real fields, preprint, 2013, arXiv:1309.4134.arxiv.org/abs/1309.4134
  34. [34]J.-F. Mestre, Familles de courbes hyperelliptiques à multiplications réelles, in: Arithmetic Algebraic Geometry (Texel, 1989), G. van der Geer, F. Oort, and J. Steenbrink (eds.), Progr. Math. 89, Birkhäuser, Boston, 1991, 193–208.DOI
  35. [35]OpenAI, Modularity of elliptic curves over imaginary quadratic fields, manuscript, October 4, 2026, github.com/openai/math.hub.valency.io/works/W_d8622vge
  36. [36]K. A. Ribet, On modular representations of Gal(̄Q/Q) arising from modular forms, Invent. Math. 100 (1990), 431–476.DOI
  37. [37]P. Scholze, On torsion in the cohomology of locally symmetric varieties, Ann. of Math. (2) 182 (2015), 945–1066.arxiv.org/abs/1306.2070
  38. [38]J.-P. Serre, Propriétés galoisiennes des points d’ordre fini des courbes elliptiques, Invent. Math. 15 (1972), 259–331.
  39. [39]J.-P. Serre, Sur les représentations modulaires de degré 2 de Gal(̄Q/Q), Duke Math. J. 54 (1987), 179–230.DOI
  40. [40]J.-P. Serre, Topics in Galois theory, Research Notes in Mathematics 1, 2nd ed., A K Peters, 2008.
  41. [41]G. Shimura, Introduction to the arithmetic theory of automorphic functions, Publ. Math. Soc. Japan 11, Iwanami Shoten and Princeton University Press, 1971.
  42. [42]A. Snowden, On two dimensional weight two odd representations of totally real fields, preprint, 2009, arXiv:0905.4266.arxiv.org/abs/0905.4266
  43. [43]W. Tautz, J. Top, and A. Verberkmoes, Explicit hyperelliptic curves with real multiplication and permutation polynomials, Canad. J. Math. 43 (1991), 1055–1064.DOI
  44. [44]R. Taylor, On Galois representations associated to Hilbert modular forms, Invent. Math. 98 (1989), 265–280.DOI
  45. [45]R. Taylor, ℓ-adic representations associated to modular forms over imaginary quadratic fields. II, Invent. Math. 116 (1994), 619–643.DOI
  46. [46]R. Taylor, Remarks on a conjecture of Fontaine and Mazur, J. Inst. Math. Jussieu 1 (2002), 125–143.DOI
  47. [47]R. Taylor and A. Wiles, Ring-theoretic properties of certain Hecke algebras, Ann. of Math. (2) 141 (1995), 553–572.DOI
  48. [48]J. A. Thorne, Automorphy of some residually dihedral Galois representations, Math. Ann. 364 (2016), 589–648.
  49. [49]J. A. Thorne, Elliptic curves over Q∞ are modular, J. Eur. Math. Soc. (JEMS) 21 (2019), 1943–1948.
  50. [50]I. Varma, Local-global compatibility for regular algebraic cuspidal automorphic representations when ℓ ≠ p, Forum Math. Sigma 12 (2024), Paper No. e21.DOI
  51. [51]A. Wiles, On ordinary λ-adic representations associated to modular forms, Invent. Math. 94 (1988), 529–573.DOI
  52. [52]A. Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. of Math. (2) 141 (1995), 443–551.DOI
  53. [53]J.-P. Wintenberger, Appendix to: J. Nekovář, On the parity of ranks of Selmer groups. IV, Compos. Math. 145 (2009), 1351–1359.
  54. [54]S. Yoshikawa, Modularity of elliptic curves over abelian totally real fields unramified at 3, 5, and 7, J. Théor. Nombres Bordeaux 30 (2018), 729–741.arxiv.org/abs/1606.06597

Paper details

vlnc:W_78ubfudk
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

OpenAI Codex Astra 6.1, Anthropic Claude Opus 5.5

View statement
Looking for related papers…
Contents