betaFeedback

WorksMathematics

Modularity of the factors of an automorphic tensor product

Abstract

Let FF be a totally real field and let ρ1,ρ2\rho_1,\rho_2 be two-dimensional pp-adic representations of GFG_F which are Lie-irreducible and potentially modular, each over its own extension. We prove the following: if ρ1⊗ρ2\rho_1\otimes\rho_2 is attached to a regular L-algebraic cuspidal automorphic representation Π\Pi of \GL_4(\A_F), then ρ1\rho_1 and ρ2\rho_2 are modular and Π≃π1⊠π2\Pi\simeq\pi_1\boxtimes\pi_2. This is a converse to Ramakrishnan's automorphy of the Rankin–Selberg product. The proof combines the descent of Hundley–Sayag from \GL_4 to \GSpin_4 with LL-function arguments based on potential modularity. To pin down the weights of the factors we use the Galois representations attached to the adjoint lifts \ad\pi_i' on \GL_3; this rules out factors of mixed parity.

Introduction

Let FF be a totally real field, pp a prime and ι:Q‾p≃C\iota:\overline{\mathbb{Q}}_p\simeq\mathbb{C}.

Conventions

Conventions. We write Frob⁡v\operatorname{Frob}_v for a geometric Frobenius element at a finite place vv, qvq_v for the cardinality of the residue field, and εcyc\varepsilon_{\mathrm{cyc}} for the pp-adic cyclotomic character, so that εcyc(Frob⁡v)=qv−1\varepsilon_{\mathrm{cyc}}(\operatorname{Frob}_v)=q_v^{-1}; it has Hodge–Tate weight −1-1. Class field theory is normalized so that uniformizers correspond to geometric Frobenius elements, and we use the same letter for a Hecke character of finite order and the corresponding Galois character. If πv\pi_v is an unramified representation of GLn(Fv)\mathrm{GL}_n(F_v), its Satake parameter t(πv)t(\pi_v) is the semisimple conjugacy class in GLn(C)\mathrm{GL}_n(\mathbb{C}) with L(s,πv)=det⁡(1−t(πv)qv−s)−1L(s,\pi_v)=\det(1-t(\pi_v)q_v^{-s})^{-1}; thus t(πv⊗∣det⁡∣s)=qv−st(πv)t(\pi_v\otimes|\det|^s)=q_v^{-s}t(\pi_v). We say that an automorphic representation π\pi of GLn(AF)\mathrm{GL}_n(\mathbb{A}_F) corresponds to a Galois representation r:GF→GLn(Q‾p)r:G_F\to\mathrm{GL}_n(\overline{\mathbb{Q}}_p) if t(πv)=ι(r(Frob⁡v))t(\pi_v)=\iota(r(\operatorname{Frob}_v)) for almost all vv. No normalization of automorphic representations is imposed: a cuspidal representation is a twist of a unitary one by a real power of ∣det⁡∣|\det|. For a Galois representation rr and a finite set SS of places containing the places where rr is ramified and those above pp,

LS(s,r)=∏v∉Sdet⁡(1−ιr(Frob⁡v)qv−s)−1.L^S(s,r)=\prod_{v\notin S}\det(1-\iota r(\operatorname{Frob}_v)q_v^{-s})^{-1}.

A cuspidal automorphic representation Π\Pi of GLn(AF)\mathrm{GL}_n(\mathbb{A}_F) is regular LL-algebraic [7] if at every real place the restriction to C×\mathbb{C}^{\times} of its Langlands parameter is a sum ⨁j=1nzpjzˉqj\bigoplus_{j=1}^{n} z^{p_j}\bar z^{q_j} with pj,qj∈Zp_j,q_j\in\mathbb{Z} and with the pjp_j pairwise distinct. For nn even, this means that Π⊗∣det⁡∣1/2\Pi\otimes|\det|^{1/2} is regular algebraic in the sense of Clozel [13]; for nn odd the two notions agree.

Let DkD_k, k≥2k\geq2, be the discrete series representation of GL2(R)\mathrm{GL}_2(\mathbb{R}) of weight kk with unitary central character sgn⁡k\operatorname{sgn}^k. By a Hilbert cusp form we mean a cuspidal automorphic representation π\pi of GL2(AF)\mathrm{GL}_2(\mathbb{A}_F) with πτ≃Dkτ\pi_\tau\simeq D_{k_\tau} at every real place τ\tau, with kτ≥2k_\tau\geq2 all of the same parity; it has a Galois representation ρπ\rho_\pi, see (R7). A representation ρ:GF→GL2(Q‾p)\rho:G_F\to\mathrm{GL}_2(\overline{\mathbb{Q}}_p) is modular if ρ≃ρπ⊗ψ\rho\simeq\rho_\pi\otimes\psi for a Hilbert cusp form π\pi and a character ψ\psi of the form (finite order)⋅εcycn\cdot\varepsilon_{\mathrm{cyc}}^n, n∈Zn\in\mathbb{Z}. Equivalently, ρ\rho corresponds to π(ρ)=π⊗η∣det⁡∣−n/2\pi(\rho)=\pi\otimes\eta|\det|^{-n/2}, where π\pi is a Hilbert cusp form, η\eta has finite order, n∈Zn\in\mathbb{Z} and m=max⁡τkτ−1m = \max_{\tau} k_{\tau} - 1. A modular representation is irreducible, odd, and de Rham with distinct Hodge–Tate weights at every embedding.

A representation ρ:GF→GL⁡2(Q‾p)\rho: G_F \to\operatorname{GL}_2(\overline{\mathbb{Q}}_p) is Lie-irreducible if it is irreducible on every open subgroup. (For an irreducible representation with distinct Hodge–Tate weights this means that it is not induced from a character of a quadratic extension, which is then necessarily CM.) It is potentially modular if there is a finite Galois totally real extension F′/FF'/F such that ρ∣GF′\left.\rho\right|_{G_{F'}} is modular.

For a Lie-irreducible representation, ad⁡0ρ\operatorname{ad}^{0}\rho is irreducible. Indeed, let GG be the Zariski closure of its image. The inverse image of G∘G^{\circ} is open, so G∘G^{\circ} acts irreducibly. A proper connected algebraic subgroup of PGL⁡2\operatorname{PGL}_2 is solvable and its inverse image in GL⁡2\operatorname{GL}_2 preserves a line. Thus the projective image of G∘G^{\circ} is PGL⁡2\operatorname{PGL}_2, whose adjoint representation is irreducible.

Remark 1.1 (Solvable subextensions). Let ρ\rho be potentially modular over F′F', and assume that ρ∣GF′\left.\rho\right|_{G_{F'}} is irreducible (this holds if ρ\rho is Lie-irreducible). Let F′′⊆F′F'' \subseteq F' be a subfield with Gal⁡(F′/F′′)\operatorname{Gal}(F'/F'') solvable. Then ρ∣GF′′\left.\rho\right|_{G_{F''}} is modular. Indeed, filter Gal⁡(F′/F′′)\operatorname{Gal}(F'/F'') by normal subgroups with cyclic quotients of prime order; at each step, the cuspidal representation attached to the restriction of ρ\rho to the larger field is invariant under the Galois group of the cyclic extension, so it is the base change of a cuspidal representation of the smaller field, unique up to twist by a character of that Galois group [1]; a twist matches the Galois representation, by Chebotarev and the irreducibility of ρ∣GF′\left.\rho\right|_{G_{F'}}. See [38 §4]. This is how potential modularity enters Lemma 2.2.

Theorem 1.2. Let ρ1,ρ2:GF→GL⁡2(Q‾p)\rho_1, \rho_2 : G_F \to\operatorname{GL}_2(\overline{\mathbb{Q}}_p) be continuous Lie-irreducible representations. Assume:

(H1) ρ1\rho_1 and ρ2\rho_2 are potentially modular, over fields F1′F'_1 and F2′F'_2 respectively.

(H2) There is a regular LL-algebraic cuspidal automorphic representation Π\Pi of GL⁡4(AF)\operatorname{GL}_4(\mathbb{A}_F) which corresponds to ρ1⊗ρ2\rho_1 \otimes\rho_2: for almost all vv, t(Πv)t(\Pi_v) is the conjugacy class of ι((ρ1⊗ρ2)(Frob⁡v))\iota((\rho_1 \otimes\rho_2)(\operatorname{Frob}_v)).

Then ρ1\rho_1 and ρ2\rho_2 are modular, and Π≃π(ρ1)⊠π(ρ2)\Pi\simeq\pi(\rho_1) \boxtimes\pi(\rho_2).

We do not require ρ1\rho_1 and ρ2\rho_2 to become modular over a common field. Nor do we require anything about the relative parity of their weights. By Lemma 2.1, hypothesis (H1) implies that ρ1\rho_1 and ρ2\rho_2 are de Rham at the places above pp; this is used in Steps 6–8.

Remark 1.3 (On the potential modularity hypothesis). Hypothesis (H1) is known in many cases. One must always assume that ρ\rho is odd, i.e. det⁡ρ(c)=−1\det\rho(c) = -1 for every complex conjugation cc, and in general that the residual image is large. The typical situations are the following.

  1. Large residual image and local hypotheses at pp. Suppose that ρ\rho is unramified outside finitely many places, totally odd, regular and potentially diagonalizable at every place above pp. Examples of the last condition include ordinary crystalline representations with distinct labelled Hodge–Tate weights, and crystalline representations over unramified extensions of Qp\mathbb{Q}_p whose labelled weights lie in an interval of length at most p−2p - 2; see [4 Lemma 1.4.3]. If p≥7p \ge7 and ρ‾∣GF(ζp)\left.\overline{\rho}\right|_{G_{F(\zeta_p)}} is absolutely irreducible, then [4 Corollary 4.5.2], with polarization given by det⁡ρ\det\rho, gives a finite Galois totally real extension over which ρ\rho is modular.

  1. pp totally split in FF. If pp splits completely in FF, the local conditions in (1) can be relaxed, because modularity lifting theorems are available for GL⁡2(Qp)\operatorname{GL}_2(\mathbb{Q}_p) without potential diagonalizability [21, 25, 15]. One needs a Galois totally real extension F′/FF'/F in which pp still splits completely and over which ρ‾\overline{\rho} is modular; such extensions are constructed in [38 Tay06, 36] under hypotheses on ρ‾\overline{\rho} at the places above pp. The lifting theorems also carry hypotheses on ρ‾\overline{\rho}: irreducibility over F′(ζp)F'(\zeta_p), and the exclusion of some small primes and of some shapes of ρ‾∣GQp\left.\overline{\rho}\right|_{G_{\mathbb{Q}_p}}. We refer to these papers for the precise statements.

  1. Compatible systems. If ρ\rho belongs to a regular, totally odd weakly compatible system of representations of dimension two, every member of which is irreducible, then the system

is potentially modular [BLGGT14, Theorem 5.4.1]. The determinant supplies the polarization. Purity follows from this theorem and need not be imposed separately; see also [27].

By Remark 1.1, in all these cases ρ∣GF′′\rho|_{G_{F''}} is also modular for every F′′⊆F′F'' \subseteq F' with F′/F′′F'/F'' solvable; this is the form of (H1) used in Lemma 2.2. In Theorem 1.2 it is applied to ρ1\rho_1 and ρ2\rho_2 separately.

Remark 1.4 (Relation with the literature). We compare Theorem 1.2 with related results.

  1. Patrikis. Assuming the Fontaine–Mazur–Langlands conjecture, [Pat19, Proposition 3.3.8] proves an automorphic tensor factorization when a cuspidal L-algebraic representation has a Galois representation of the form ρ1⊗ρ2\rho_1 \otimes\rho_2. The given factors in that statement need not themselves be geometric. In the setting of Theorem 1.2, hypothesis (H1) makes both factors geometric, so the conjecture would imply their modularity. Patrikis also explains that potential automorphy gives some cases after a finite base change. Theorem 1.2 gives the conclusion over FF under the stated potential modularity hypotheses.

  1. Mixed parity forms. Consider two cuspidal representations on GL2(AF)\mathrm{GL}_2(\mathbb{A}_F) with discrete series at every real place, neither induced from a quadratic extension. Suppose that each has both even and odd weights, that their weight parities agree at each real place, and that their tensor product is cuspidal. With the normalization of [Pat19, Proposition 2.6.7], this tensor product is L-algebraic and its Galois representation has tensor factors no twist of which is de Rham; see also [HL24, Remark 1.5]. These factors are not potentially modular in our sense, since potential modularity implies the de Rham property by Lemma 2.1. Thus replacing (H1) by a condition allowing arbitrary twists that are not de Rham would admit counterexamples. The de Rham property is used in Steps 6–8 below.

  1. Direct sums. The analogous case σ⊕σ′\sigma\oplus\sigma' is treated in [Ram13, Lemma 5.1 and Theorem C]; see Remark 4.2.

  1. The opposite direction. [24] and [Ram02] prove automorphy of representations of GO4_4-type by splitting them into factors and applying [29]. [28] and [Pat19, Cor. 2.6.3] treat the uniqueness of the factorization, on the Galois and automorphic sides respectively.

  1. Over Q\mathbb{Q}. The statement is essentially known. Under mild hypotheses on the residual representation, a two-dimensional odd, irreducible, de Rham and regular ρ\rho over Q\mathbb{Q}, unramified outside finitely many primes, is modular [21, 25, 15], without passing through II\mathrm{II}. Moreover, under its hypotheses on connected monodromy and the tautological representation, [HL24, Theorem 1.4] gives tensor decompositions into symmetric powers of representations attached to modular forms, up to a character; in the case of two factors of dimension two, the symmetric powers have degree one. The interest of Theorem 1.2 lies in F≠QF \ne\mathbb{Q}.

  1. The automorphic factorization. Steps 1–3 of the proof, which show that Π≃π1′⊠π2′\Pi\simeq\pi'_1 \boxtimes\pi'_2 for cuspidal π1′,π2′\pi'_1,\pi'_2 on GL2(AF)\mathrm{GL}_2(\mathbb{A}_F), are not new. Asgari and Shahidi [3], using the descent of Hundley–Sayag [17], that a cuspidal representation of GL2n(AF)\mathrm{GL}_{2n}(\mathbb{A}_F) whose twisted symmetric square LL-function LS(s,Π,Sym⁡2⊗χ−1)L^S(s,\Pi,\operatorname{Sym}^2 \otimes\chi^{-1}) has a pole at s=1s=1 is a transfer from a split or quasi-split even general spin group; and in the classification of Asgari and Raghuram [AR11, Thm. 1.1] of the cuspidal representations Π\Pi of GL4(AF)\mathrm{GL}_4(\mathbb{A}_F) with non-cuspidal exterior square, the transfers from split GSpin4\mathrm{GSpin}_4 are the automorphic tensor products π1⊠π2\pi_1 \boxtimes\pi_2, and those from quasi-split non-split GSpin4\mathrm{GSpin}_4 are Asai transfers. Since ωΠχ−2=1\omega_\Pi\chi^{-2}=1 in our situation, the descent is to the split group, and Π≃π1′⊠π2′\Pi\simeq\pi'_1 \boxtimes\pi'_2. We include the argument, with the identification of split GSpin4\mathrm{GSpin}_4 with a subgroup of GL2×GL2\mathrm{GL}_2 \times\mathrm{GL}_2 (Remark 4.3), to fix the normalizations used in Steps 4–8. The additional argument in Steps 4–8 passes from this automorphic factorization, whose factors need not initially be algebraic, to the modularity of the given ρi\rho_i over FF. The use of adjoint representations for forms of mixed parity also appears in [Pat19, Proposition 2.6.7(iii)].

(7) Shavali. In [33 Proposition 5.1], a regular algebraic cuspidal representation Π=π1⊠π2\Pi= \pi_1 \boxtimes\pi_2, with neither factor induced from a quadratic extension, has associated Galois representations whose monodromy Lie algebra has semisimple part so4\mathfrak{so}_4. Its proof uses a decomposition into two factors of dimension two, which need not themselves be geometric over FF. Theorem 1.2 addresses the modularity over FF of factors given in advance and assumed to be potentially modular.

(8) Calegari–Gee. In [8 §6.4] the tensor case r≃a⊗br \simeq a \otimes b appears in the study of irreducibility for n=4n = 4, without any conclusion on the modularity of the factors. Their corrigendum [9] explains that the proof of their Theorem 4.1 for n=4,5n = 4, 5 is not valid as written.

Preliminaries

Known results

We shall use the following results.

(R1) Tensor product [29]. If π,π′\pi, \pi' are cuspidal on GL2(AF)\mathrm{GL}_2(A_F), there is an isobaric automorphic representation π⊠π′\pi\boxtimes\pi' of GL4(AF)\mathrm{GL}_4(A_F) which is functorial at all places, the archimedean ones included: the Langlands parameter of (π⊠π′)v(\pi\boxtimes\pi')_v is the tensor product of those of πv\pi_v and πv′\pi'_v.

(R2) Exterior square identity. For representations V1,V2V_1, V_2 of dimension two in characteristic zero,

⋀2(V1⊗V2)⊗(det⁡V1det⁡V2)−1≃ad⁡0V1⊕ad⁡0V2.\bigwedge\nolimits^2(V_1 \otimes V_2) \otimes(\det V_1 \det V_2)^{-1} \simeq\operatorname{ad}\nolimits^0 V_1 \oplus\operatorname{ad}\nolimits^0 V_2.

This follows from ⋀2(V1⊗V2)≃(Sym⁡2V1⊗det⁡V2)⊕(det⁡V1⊗Sym⁡2V2)\bigwedge\nolimits^2(V_1 \otimes V_2) \simeq(\operatorname{Sym}\nolimits^2 V_1 \otimes\det V_2) \oplus(\det V_1 \otimes\operatorname{Sym}\nolimits^2 V_2). We use it for Galois representations and for Satake parameters.

(R3) Poles, nonvanishing and multiplicity one [18, 19]. Let cc be unitary cuspidal and let X=⊞jcjX = \boxplus_j c_j be an isobaric sum of distinct unitary cuspidal representations.

  • LS(s,c×c∨)L^S(s,c \times c^\vee) has a simple pole at s=1s = 1.

  • LS(s,c)L^S(s,c) is holomorphic and nonzero at s=1s = 1, except when cc is the trivial character of GL1\mathrm{GL}_1, in which case it has a simple pole. Hence −ord⁡s=1LS(s,X)-\operatorname{ord}_{s=1}L^S(s,X) is the multiplicity of the trivial character in XX.

  • Strong multiplicity one holds for isobaric representations.

Here SS is any finite set containing the ramified places. Removing the local factors preserves these assertions at s=1s = 1, since the removed factors are holomorphic and nonzero there.

(R4) Gelbart–Jacquet [14]. If π\pi is cuspidal on GL2(AF)\mathrm{GL}_2(A_F), then ad⁡0π\operatorname{ad}\nolimits^0\pi is an isobaric automorphic representation of GL3(AF)\mathrm{GL}_3(A_F), self-dual with trivial central character, functorial at all places, and it is cuspidal if and only if π\pi is not dihedral. Its isobaric constituents are unitary. Indeed, a real norm twist of π\pi disappears in ad⁡0π\operatorname{ad}\nolimits^0\pi. In the cuspidal case the trivial central character gives unitarity. If π=AI⁡K/F(θ)\pi= \operatorname{AI}_{K/F}(\theta), then ad⁡0π=ηK⊞AI⁡K/F(θ/θc)\operatorname{ad}\nolimits^0\pi= \eta_K \boxplus\operatorname{AI}_{K/F}(\theta/\theta^c), where cc generates Gal⁡(K/F)\operatorname{Gal}(K/F) and ηK\eta_K is its quadratic character; the quotient θ/θc\theta/\theta^c is unitary.

(R5) Descent [17 Thm. 16.3.1]. Let Π\Pi be a unitary cuspidal representation of GL4(AF)\mathrm{GL}_4(A_F) and χ\chi a Hecke character. Suppose that LS(s,Π,Sym⁡2⊗χ−1)L^S(s,\Pi,\operatorname{Sym}^2 \otimes\chi^{-1}) has a pole at s=1s = 1, i.e. Π\Pi is χ−1\chi^{-1}-orthogonal, and that ωΠχ−2=1\omega_\Pi\chi^{-2} = 1. Then the descent DCχα(Π)DC^\alpha_\chi(\Pi) to the split group GSpin4\mathrm{GSpin}_4 is a nonzero cuspidal representation of GSpin4(AF)\mathrm{GSpin}_4(A_F), generic and with central character χ\chi, and every irreducible automorphic representation σ\sigma contained in it lifts weakly to Π\Pi with respect to the inclusion GSO4(C)↪GL4(C)\mathrm{GSO}_4(\mathbb{C}) \hookrightarrow\mathrm{GL}_4(\mathbb{C}). This is the case n=2n = 2, r=1r = 1, τ=Π\tau= \Pi, ω=χ\omega= \chi of the cited theorem, with aa the class of squares: the quadratic character ωΠ/χ2\omega_\Pi/\chi^2 is trivial, hence compatible with aa in the sense of [17 Def. 16.1.13], and G4G_4 is the split group [17 Rem. 16.1.6]. See Remark 4.3.

(R6) Purity [13 Lemme 4.9]. Let Π\Pi be cuspidal on GLn(AF)\mathrm{GL}_n(A_F) such that some twist Π⊗∣det⁡∣s\Pi\otimes|\det|^s, s∈Rs \in\mathbb{R}, is algebraic in the sense of Clozel (for instance Π\Pi L-algebraic, or a unitary twist of such a representation). If ϕΠ∞∣C×=⨁jzpjzˉqj\phi_{\Pi_\infty}|_{\mathbb{C}^\times} = \bigoplus_j z^{p_j}\bar{z}^{q_j}, then pj+qjp_j + q_j is independent of jj.

(R7) Galois representations of Hilbert modular forms [10, 37, 6, 32, 35]. Let π\pi be a Hilbert cusp form, with weights κτ≥2\kappa_\tau\ge2 of the same parity, and let $m = \max_\tau \kappa_\tau - 1. Then there is a continuous irreducible representation ρπ:GF→GL⁡2(Q‾p)\rho_{\pi}:G_F \to\operatorname{GL}_2(\overline{\mathbb{Q}}_p) which corresponds to π⊗∣det⁡∣−m/2\pi\otimes|\det|^{-m/2}, that is, ιρπ(Frob⁡v)\iota\rho_{\pi}(\operatorname{Frob}_v) has eigenvalues qvm/2q_v^{m/2} times the Satake parameters of πv\pi_v for almost all vv. It is de Rham at the places above pp, with Hodge–Tate weights differing by kτ−1k_{\tau}-1 at τ\tau. The central character ωπ\omega_{\pi} has finite order, and det⁡ρπ=ωπεcyc−wρ\det\rho_{\pi}=\omega_{\pi}\varepsilon_{\mathrm{cyc}}^{-w_{\rho}}; in particular, the sum of these weights is mm at every embedding.

(R8) Galois representations for $\operatorname{GL}_3$ [BLGHT11, Theorem 1.1], using [34, 11]. Let Σ\Sigma be a cuspidal representation of GL⁡3(AF)\operatorname{GL}_3(\mathbb{A}_F) with Σ∨≃Σ\Sigma^{\vee}\simeq\Sigma and trivial central character. Suppose that its infinitesimal character at τ\tau is (dτ,0,−dτ)(d_{\tau},0,-d_{\tau}), with dτ≥1d_{\tau}\geq1 an integer. Then there is a continuous semisimple Galois representation RΣR_{\Sigma} which corresponds to Σ\Sigma, is de Rham above pp, has determinant 11, and has labelled Hodge–Tate weights {dτ,0,−dτ}\{d_{\tau},0,-d_{\tau}\}.

To specify the normalization, the algebraic weight in the cited theorem is aτ=(dτ−1,0,1−dτ)a_{\tau}=(d_{\tau}-1,0,1-d_{\tau}), which is dominant and equal to its dual. The representation rp,ι(Σ)r_{p,\iota}(\Sigma) of that theorem corresponds to Σ⊗∣det⁡∣−1\Sigma\otimes|\det|^{-1} and has weights {dτ+1,1,1−dτ}\{d_{\tau}+1,1,1-d_{\tau}\}. Thus

RΣ=rp,ι(Σ)⊗εcycR_{\Sigma}=r_{p,\iota}(\Sigma)\otimes\varepsilon_{\mathrm{cyc}}

has the stated Frobenius normalization and weights. Its determinant is 11 by the unramified Satake comparison and Chebotarev. The condition on the archimedean values of the character in the cited theorem holds because that character is trivial.

Lemmas

Lemma 2.1. Let ρ:GF→GL⁡2(Q‾p)\rho:G_F\to\operatorname{GL}_2(\overline{\mathbb{Q}}_p) be potentially modular. Then ρ\rho is unramified outside finitely many places, totally odd, and de Rham at the places above pp, with distinct Hodge–Tate weights at every embedding. Moreover, det⁡ρ=χρεcyc−wρ\det\rho=\chi_{\rho}\varepsilon_{\mathrm{cyc}}^{-w_{\rho}} with χρ\chi_{\rho} of finite order and wρ∈Zw_{\rho}\in\mathbb{Z}. The sum of the Hodge–Tate weights of ρ\rho at every embedding is wρw_{\rho}.

Proof. The modular restriction to GF′G_{F'} is unramified outside finitely many places. Since F′/FF'/F is finite, the same holds for ρ\rho. Total oddness descends because F′F' is totally real. The restriction is de Rham with distinct labelled Hodge–Tate weights by (R7). The de Rham property descends through a finite extension of local fields, and the labelled weights are preserved under extension of embeddings.

By (R7) and the definition of modularity, there is an integer wρw_{\rho} such that (det⁡ρ)εcycwρ(\det\rho)\varepsilon_{\mathrm{cyc}}^{w_{\rho}} has finite image on GF′G_{F'}. A character with finite image on an open subgroup has finite image on the whole group. This proves the determinant formula, and its Hodge–Tate weight gives the asserted sum. □\square

Lemma 2.2 (Non-vanishing at s=1s=1 via Brauer). Let ρ\rho be two-dimensional, Lie-irreducible and potentially modular over F′F', and let η\eta be a finite order Hecke character of FF. Then LS(s,ad⁡0ρ⊗η)L^S(s,\operatorname{ad}^{0}\rho\otimes\eta) is meromorphic on C\mathbb{C}, and holomorphic and nonzero at s=1s=1.

Proof. Enlarge SS so that it contains the places which ramify in F′F' and those where η\eta is ramified. By Brauer’s theorem, 1G=∑iniInd⁡HiGψi\mathbf{1}_G=\sum_i n_i\operatorname{Ind}_{H_i}^G\psi_i in G=Gal⁡(F′/F)G=\operatorname{Gal}(F'/F), with ni∈Zn_i\in\mathbb{Z}, HiH_i elementary (hence solvable) and ψi\psi_i characters of HiH_i. Let Fi=F′HiF_i=F'^{H_i}, let SiS_i be the set of places of FiF_i above SS, and let ηFi=η∘NFi/F\eta_{F_i}=\eta\circ N_{F_i/F}. Since r≃∑iniInd⁡GFiGF(r∣GFi⊗ψi)r\simeq\sum_i n_i\operatorname{Ind}_{G_{F_i}}^{G_F}(r|_{G_{F_i}}\otimes\psi_i) as virtual representations for r=ad⁡0ρ⊗ηr=\operatorname{ad}^{0}\rho\otimes\eta, and all the places outside SS are unramified,

LS(s,ad⁡0ρ⊗η)=∏iLSi(s,ad⁡0ρ∣GFi⊗ψiηFi)ni.L^S(s,\operatorname{ad}^{0}\rho\otimes\eta)=\prod_i L^{S_i}(s,\operatorname{ad}^{0}\rho|_{G_{F_i}}\otimes\psi_i\eta_{F_i})^{n_i}.

By Remark 1.1, applied to the solvable extension F′/FiF'/F_i, ρ∣GFi\rho|_{G_{F_i}} is modular, so it corresponds to a cuspidal representation σi\sigma_i of GL⁡2(AFi)\operatorname{GL}_2(\mathbb{A}_{F_i}), and σi\sigma_i is not dihedral because ρ\rho is Lie-irreducible. Hence ad⁡0σi\operatorname{ad}^{0}\sigma_i is cuspidal by (R4); it is unitary, and ad⁡0σi⊗ψiηFi\operatorname{ad}^{0}\sigma_i\otimes\psi_i\eta_{F_i} is a unitary cuspidal representation of GL⁡3(AFi)\operatorname{GL}_3(\mathbb{A}_{F_i}) which corresponds to ad⁡0ρ∣GFi⊗ψiηFi\operatorname{ad}^{0}\rho|_{G_{F_i}}\otimes\psi_i\eta_{F_i}. By (R3), each factor LSi(s,ad⁡0σi⊗ψiηFi)L^{S_i}(s,\operatorname{ad}^{0}\sigma_i\otimes\psi_i\eta_{F_i}) is meromorphic on C\mathbb{C}, and holomorphic and nonzero at s=1s=1. The same holds for the product of their powers. □\square

Lemma 2.3. Let σ,σ′\sigma,\sigma' be two-dimensional with ad⁡0σ≃ad⁡0σ′\operatorname{ad}^{0}\sigma\simeq\operatorname{ad}^{0}\sigma' irreducible. Then σ′≃σ⊗ψ\sigma' \simeq\sigma\otimes\psi for a continuous character ψ\psi. If σ\sigma and σ′\sigma' are de Rham, so is ψ\psi.

Proof. An isomorphism ad⁡0σ≃ad⁡0σ′\operatorname{ad}^{0}\sigma\simeq\operatorname{ad}^{0}\sigma' carries the Killing form to a multiple of the Killing form, because ad⁡0σ\operatorname{ad}^{0}\sigma is irreducible; after scaling and changing its sign it lies in SO3≃PGL2\mathrm{SO}_{3} \simeq\mathrm{PGL}_{2}. Hence the projectivizations of σ\sigma and σ′\sigma' are conjugate, and after conjugation σ′=σ⊗ψ\sigma' = \sigma\otimes\psi for a continuous character ψ\psi; compare [Ram02, Thm. 3.10]. For the second assertion, ψ\psi is a summand of σ∨⊗σ′≃σ∨⊗σ⊗ψ\sigma^{\vee} \otimes\sigma' \simeq\sigma^{\vee} \otimes\sigma\otimes\psi, which is de Rham. □

Lemma 2.4 (A twist normalizing the archimedean components). Let ω\omega be a Hecke character of FF with ωτ(x)=sgn⁡(x)ετ∣x∣τsτ\omega_{\tau}(x) = \operatorname{sgn}(x)^{\varepsilon_{\tau}} |x|_{\tau}^{s_{\tau}} at each real place τ\tau. Then there is a Hecke character μ\mu with μτ(x)=sgn⁡(x)fτ∣x∣τsτ/2\mu_{\tau}(x) = \operatorname{sgn}(x)^{f_{\tau}} |x|_{\tau}^{s_{\tau}/2} for some fτ∈{0,1}f_{\tau} \in\{0,1\}.

Proof. Let U⊂O^F×U \subset\widehat{\mathcal{O}}_{F}^{\times} be an open compact subgroup with ω∣U=1\omega|_{U} = 1, and let E+E^{+} be the group of totally positive units in UU. For ε∈E+\varepsilon\in E^{+}, the condition ω(ε)=1\omega(\varepsilon) = 1 gives ∏τ∣ετ∣sτ=1\prod_{\tau} |\varepsilon_{\tau}|^{s_{\tau}} = 1. Hence ∏τ∣ητ∣sτ/2=1\prod_{\tau} |\eta_{\tau}|^{s_{\tau}/2} = 1 for all η\eta in (E+)2(E^{+})^{2}, a subgroup of finite index.

By Chevalley’s theorem [12], every finite index subgroup of OF×\mathcal{O}_{F}^{\times} contains a congruence subgroup OF×(a)={ε≡1 mod a}\mathcal{O}_{F}^{\times}(\mathfrak{a}) = \{\varepsilon\equiv1 \bmod\mathfrak{a}\}. So there is an open compact U′⊆UU' \subseteq U such that the totally positive units in U′U' lie in (E+)2(E^{+})^{2}. Then the character x∞u↦∏τ∣xτ∣sτ/2x_{\infty}u \mapsto\prod_{\tau} |x_{\tau}|^{s_{\tau}/2} of H=F∞,+×U′H = F_{\infty,+}^{\times}U' is trivial on F×∩HF^{\times} \cap H. It therefore defines a character of F×H/F×F^{\times}H/F^{\times}, an open subgroup of finite index of AF×/F×\mathbb{A}_{F}^{\times}/F^{\times}, which extends to AF×/F×\mathbb{A}_{F}^{\times}/F^{\times}. Such an extension exists because C×\mathbb{C}^{\times} is divisible, and it is continuous because the subgroup is open. On the component groups at the real places it has values in {1,−1}\{1,-1\}, giving the exponents fτf_{\tau}. □

Proof of Theorem 1.2

Let ρ=ρ1⊗ρ2\rho= \rho_{1} \otimes\rho_{2}. By Lemma 2.1, det⁡ρi=χiεcycwi\det\rho_{i} = \chi_{i}\varepsilon_{\mathrm{cyc}}^{w_{i}} with χi\chi_{i} of finite order and wi∈Zw_{i} \in\mathbb{Z}. Put

χ=χ1χ2,w=w1+w2,Πu=Π⊗∣det⁡∣w/2.\chi= \chi_{1}\chi_{2}, \qquad w = w_{1} + w_{2}, \qquad\Pi^{u} = \Pi\otimes|\det|^{w/2}.

Then t(Πvu)=qv−w/2ιρ(Frob⁡v)t(\Pi^{u}_{v}) = q_{v}^{-w/2}\iota\rho(\operatorname{Frob}_{v}) for almost all vv. Since det⁡ρ=χ2εcyc2w\det\rho= \chi^{2}\varepsilon_{\mathrm{cyc}}^{2w}, the central character of Πu\Pi^{u} agrees with χ2\chi^{2} at almost all places, hence ωΠu=χ2\omega_{\Pi^{u}} = \chi^{2}. In particular Πu\Pi^{u} is a cuspidal representation with unitary central character, so it is unitary. It is a twist of Π\Pi, so (R6) applies to it and its archimedean parameters are regular. In Steps 1–7 we write Π\Pi for Πu\Pi^{u}.

Step 1: orthogonal type

On the Galois side, ρ∨⊗det⁡ρ1det⁡ρ2≃ρ\rho^{\vee} \otimes\det\rho_{1} \det\rho_{2} \simeq\rho. Hence t(Πv)−1χ(ϖv)=t(Πv)t(\Pi_{v})^{-1}\chi(\varpi_{v}) = t(\Pi_{v}) as conjugacy classes for almost all vv, and by strong multiplicity one,

Π≃Π∨⊗χ,ωΠ=χ2.\Pi\simeq\Pi^{\vee} \otimes\chi, \qquad\omega_{\Pi} = \chi^{2}.

Step 2: pole of the symmetric square

Since Π∨≃Π⊗χ−1\Pi^{\vee} \simeq\Pi\otimes\chi^{-1},

LS(s,Π×Π∨)=LS(s,Π,Sym⁡2⊗χ−1)LS(s,Π,∧2⊗χ−1).L^{S}(s,\Pi\times\Pi^{\vee}) = L^{S}(s,\Pi,\operatorname{Sym}^{2} \otimes\chi^{-1})L^{S}(s,\Pi,\wedge^{2} \otimes\chi^{-1}).

The left hand side has a simple pole at s=1s=1. On the Galois side, ∧2ρ⊗(det⁡ρ1det⁡ρ2)−1=ad⁡0ρ1⊕ad⁡0ρ2\wedge^{2}\rho\otimes(\det\rho_{1}\det\rho_{2})^{-1} = \operatorname{ad}^{0}\rho_{1} \oplus\operatorname{ad}^{0}\rho_{2}, and ∧2t(Πv)⋅χ(ϖv)−1=ι(∧2ρ⊗(χεcyc−w)−1)(Frob⁡v)\wedge^{2}t(\Pi_{v}) \cdot\chi(\varpi_{v})^{-1} = \iota(\wedge^{2}\rho\otimes(\chi\varepsilon_{\mathrm{cyc}}^{-w})^{-1})(\operatorname{Frob}_{v}). Hence

LS(s,Π,∧2⊗χ−1)=LS(s,ad⁡0ρ1)LS(s,ad⁡0ρ2),L^{S}(s,\Pi,\wedge^{2} \otimes\chi^{-1}) = L^{S}(s,\operatorname{ad}^{0}\rho_{1})L^{S}(s,\operatorname{ad}^{0}\rho_{2}),

which is meromorphic on C\mathbb{C}, and holomorphic and nonzero at s=1s=1 by Lemma 2.2, applied to each ρi\rho_{i} with its own field FiF_{i} and η=1\eta= 1. Therefore the quotient LS(s,Π,Sym⁡2⊗χ−1)L^{S}(s,\Pi,\operatorname{Sym}^{2} \otimes\chi^{-1}) is meromorphic near s=1s=1 and has a simple pole at s=1s=1.

Step 3: descent

Steps 1–3 recover a known result (Remark 1.4(6)); see [3] and [AR11, Thm. 1.1. By Step 2 and ωΠ=χ2\omega_{\Pi}=\chi^{2}, (R5) applies: there is an irreducible cuspidal σ\sigma on GSpin⁡4(AF)\operatorname{GSpin}_{4}(\mathbb{A}_{F}) lifting weakly to Π\Pi. The dual group of GSpin⁡4\operatorname{GSpin}_{4} is GSO⁡4(C)=β(GL⁡2(C)×GL⁡2(C))\operatorname{GSO}_{4}(\mathbb{C})=\beta(\operatorname{GL}_{2}(\mathbb{C})\times\operatorname{GL}_{2}(\mathbb{C})), with β(g1,g2)=g1⊗g2\beta(g_{1},g_{2})=g_{1}\otimes g_{2} [2 §2.2]. Hence

GSpin⁡4≃H:={(g1,g2)∈GL⁡2×GL⁡2:det⁡g1=det⁡g2},\operatorname{GSpin}_{4}\simeq H:=\{(g_{1},g_{2})\in\operatorname{GL}_{2}\times\operatorname{GL}_{2}:\det g_{1}=\det g_{2}\},

and the lift to GL⁡4\operatorname{GL}_{4} corresponds to the tensor product.

We apply [23 Thm. 1.1.1 and Thm. 5.2.2] to the inclusion f:H↪GL⁡2×GL⁡2f:H\hookrightarrow\operatorname{GL}_{2}\times\operatorname{GL}_{2}. It is injective, and on derived groups it is the identity of SL⁡2×SL⁡2\operatorname{SL}_{2}\times\operatorname{SL}_{2}, in particular a central isogeny. The case SL⁡2⊂GL⁡2\operatorname{SL}_{2}\subset\operatorname{GL}_{2} is [22]. Hence, by the second assertion of [23 Thm. 1.1.1], σ\sigma occurs in the restriction to H(AF)H(\mathbb{A}_{F}) of the space of a cuspidal representation π1′⊠π2′\pi'_{1}\boxtimes\pi'_{2} of (GL⁡2×GL⁡2)(AF)(\operatorname{GL}_{2}\times\operatorname{GL}_{2})(\mathbb{A}_{F}); the πi′\pi'_{i} are cuspidal, not necessarily unitary. For almost all vv, both σv\sigma_{v} and π1,v′⊗π2,v′\pi'_{1,v}\otimes\pi'_{2,v} are unramified, and σv\sigma_{v} is the constituent of the restriction of π1,v′⊗π2,v′\pi'_{1,v}\otimes\pi'_{2,v} to H(Fv)H(F_{v}) which is unramified for H(OFv)H(\mathcal{O}_{F_{v}}). Its Satake parameter is therefore the image of that of (π1,v′,π2,v′)(\pi'_{1,v},\pi'_{2,v}) under the dual map GL⁡2(C)×GL⁡2(C)→GSO⁡4(C)\operatorname{GL}_{2}(\mathbb{C})\times\operatorname{GL}_{2}(\mathbb{C})\to\operatorname{GSO}_{4}(\mathbb{C}), and its image in GL⁡4(C)\operatorname{GL}_{4}(\mathbb{C}) is t(π1,v′)⊗t(π2,v′)t(\pi'_{1,v})\otimes t(\pi'_{2,v}). So π1′⊠π2′\pi'_{1}\boxtimes\pi'_{2}, which is automorphic by (R1), and Π\Pi have the same Satake parameters at almost all places. By strong multiplicity one (R3), Π≃π1′⊠π2′\Pi\simeq\pi'_{1}\boxtimes\pi'_{2}, and by (R1), Πv≃π1,v′⊠π2,v′\Pi_{v}\simeq\pi'_{1,v}\boxtimes\pi'_{2,v} at all places, including the real ones.

Let ε=ωπ1′ωπ2′χ−1\varepsilon=\omega_{\pi'_{1}}\omega_{\pi'_{2}}\chi^{-1}; then ε2=ωΠχ−2=1\varepsilon^{2}=\omega_{\Pi}\chi^{-2}=1. By (R4), the representation

X:=(ad⁡0π1′⊗ε)⊞(ad⁡0π2′⊗ε)⟷ad⁡0ρ1⊕ad⁡0ρ2(1)X:=(\operatorname{ad}^{0}\pi'_{1}\otimes\varepsilon)\boxplus(\operatorname{ad}^{0}\pi'_{2}\otimes\varepsilon)\longleftrightarrow\operatorname{ad}^{0}\rho_{1}\oplus\operatorname{ad}^{0}\rho_{2} \tag*{(1)}

is isobaric automorphic with unitary constituents. The displayed correspondence follows from (R2): at almost all places, both sides have Satake or Frobenius parameter ∧2t(Πv)χ(ϖv)−1\wedge^{2}t(\Pi_{v})\chi(\varpi_{v})^{-1}.

Step 4: the factors are not dihedral

Suppose that π1′\pi'_{1} is automorphically induced from a quadratic extension K/FK/F, with quadratic character ηK\eta_{K}. Then ad⁡0π1′≃ηK⊞(⋯ )\operatorname{ad}^{0}\pi'_{1}\simeq\eta_{K}\boxplus(\cdots), so the isobaric representation X⊗ηKε≃(ad⁡0π1′⊗ηK)⊞(ad⁡0π2′⊗ηK)X\otimes\eta_{K}\varepsilon\simeq(\operatorname{ad}^{0}\pi'_{1}\otimes\eta_{K})\boxplus(\operatorname{ad}^{0}\pi'_{2}\otimes\eta_{K}), a sum of unitary cuspidal representations, contains the trivial character, and L(s,X⊗ηKε)L(s,X\otimes\eta_{K}\varepsilon) has a pole at s=1s=1 by (R3). But by (1) this function is

L(s,ad⁡0ρ1⊗ηKε)L(s,ad⁡0ρ2⊗ηKε),L(s,\operatorname{ad}^{0}\rho_{1}\otimes\eta_{K}\varepsilon)L(s,\operatorname{ad}^{0}\rho_{2}\otimes\eta_{K}\varepsilon),

which is holomorphic and nonzero at s=1s=1 by Lemma 2.2, applied with η=ηKε\eta=\eta_{K}\varepsilon, a contradiction. The same argument applies to π2′\pi'_{2}. By (R4), ad⁡0π1′\operatorname{ad}^{0}\pi'_{1} and ad⁡0π2′\operatorname{ad}^{0}\pi'_{2} are cuspidal.

Step 5: archimedean components

Let τ\tau be real and let ϕi\phi_{i} be the parameter of πi,τ′\pi'_{i,\tau}. By Step 3 and (R1), the parameter of Πτ\Pi_{\tau} is ϕ1⊗ϕ2\phi_{1}\otimes\phi_{2}. Write ϕΠ∣C×=(ϕ1⊗ϕ2)∣C×=⨁jzpjzˉqj\phi_{\Pi}|_{\mathbb{C}^{\times}}=(\phi_{1}\otimes\phi_{2})|_{\mathbb{C}^{\times}}=\bigoplus_{j}z^{p_{j}}\bar z^{q_{j}}. By (R6) and the finite order central character ωΠ=χ2\omega_{\Pi}=\chi^{2}, pj+qj=0p_{j}+q_{j}=0, so pj=12(pj−qj)p_{j}=\frac{1}{2}(p_{j}-q_{j}). Regularity of Π\Pi requires the differences pj−qjp_{j}-q_{j} to be distinct.

The representation πi,τ′\pi'_{i,\tau} is irreducible and generic, so it is either a principal series or a twist of a discrete series. If ϕ1\phi_{1} were a principal series parameter, a sum of two characters of R×\mathbb{R}^{\times}, then ϕ1∣C×=∣z∣2c⊕∣z∣2c\phi_{1}|_{\mathbb{C}^{\times}}=|z|^{2c}\oplus|z|^{2c} and each difference pj−qjp_{j}-q_{j} of ϕ2\phi_{2} would occur twice in ϕ1⊗ϕ2\phi_{1}\otimes\phi_{2}, contradicting regularity. The same holds for ϕ2\phi_{2}. Therefore

πi,τ′≃Dki,τ⊗∣det⁡∣ti,τ,ki,τ≥2,ti,τ∈C.\pi'_{i,\tau}\simeq D_{k_{i,\tau}}\otimes|\det|^{t_{i,\tau}},\qquad k_{i,\tau}\ge2,\quad t_{i,\tau}\in\mathbb{C}.

Consequently ad⁡0πi,τ′\operatorname{ad}^{0}\pi'_{i,\tau} has exponents {ki,τ−1,0,1−ki,τ}\{k_{i,\tau}-1,0,1-k_{i,\tau}\}, since the twists ∣det⁡∣ti,τ|\det|^{t_{i,\tau}} cancel in ad⁡0\operatorname{ad}^{0}. So ad⁡0πi′\operatorname{ad}^{0}\pi'_{i} is cuspidal (Step 4), self-dual and regular algebraic, even if πi′\pi'_{i} is not algebraic.

Step 6: projective matching and weights

Let Ri:=Rad⁡0πi′R_{i}:=R_{\operatorname{ad}^{0}\pi'_{i}}, given by (R8) with dτ=ki,τ−1d_{\tau}=k_{i,\tau}-1. Step 5 verifies its archimedean hypothesis, and (R4) gives the trivial central character. By (1), Chebotarev and the semisimplicity of both sides,

(R1⊗ε)⊕(R2⊗ε)≃ad⁡0ρ1⊕ad⁡0ρ2.(R_{1}\otimes\varepsilon)\oplus(R_{2}\otimes\varepsilon)\simeq\operatorname{ad}^{0}\rho_{1}\oplus\operatorname{ad}^{0}\rho_{2}.

Since the ad⁡0ρi\operatorname{ad}^{0}\rho_i are irreducible of dimension 3, each Ri⊗εR_i \otimes\varepsilon is isomorphic to one of them; after renumbering π1′,π2′\pi'_1,\pi'_2, we have Ri⊗ε≃ad⁡0ρiR_i \otimes\varepsilon\simeq\operatorname{ad}^{0}\rho_i. The central character of ad⁡0πi′\operatorname{ad}^{0}\pi'_i is trivial, so det⁡Ri=1\det R_i = 1 by (R8), and det⁡(Ri⊗ε)=ε3=ε\det(R_i \otimes\varepsilon) = \varepsilon^3 = \varepsilon. As det⁡ad⁡0ρi=1\det\operatorname{ad}^{0}\rho_i = 1, we get ε=1\varepsilon= 1, that is,

Rad⁡0πi′≃ad⁡0ρi(i=1,2).R_{\operatorname{ad}^{0}\pi'_i} \simeq\operatorname{ad}^{0}\rho_i \qquad(i = 1,2).

By (R8), the labelled Hodge–Tate weights of RiR_i are exactly {ki,τ−1,0,1−ki,τ}\{k_{i,\tau}-1, 0, 1-k_{i,\tau}\}. The representations ρi\rho_i are de Rham by Lemma 2.1, and the Hodge–Tate weights of ad⁡0ρi\operatorname{ad}^{0}\rho_i at τ\tau are {Ai,τ−Bi,τ,0,Bi,τ−Ai,τ}\{A_{i,\tau}-B_{i,\tau}, 0, B_{i,\tau}-A_{i,\tau}\}, where {Ai,τ,Bi,τ}\{A_{i,\tau}, B_{i,\tau}\} are those of ρi\rho_i. Hence

ki,τ−1=∣Ai,τ−Bi,τ∣for all τ.k_{i,\tau} - 1 = \lvert A_{i,\tau} - B_{i,\tau}\rvert\qquad\text{for all }\tau.

In other words, at every real place each πi′\pi'_i carries the weights of its own ρi\rho_i, and the weights cannot be mixed between the two factors.

Step 7: parity and algebraicity

By Lemma 2.1, Ai,τ+Bi,τ=wiA_{i,\tau} + B_{i,\tau} = w_i for all τ\tau. By Step 6, ki,τ−1=∣Ai,τ−Bi,τ∣≡wi(mod2)k_{i,\tau} - 1 = \lvert A_{i,\tau} - B_{i,\tau}\rvert\equiv w_i \pmod{2} for all τ\tau, so each πi′\pi'_i has paritious weights.

The central character of πi′\pi'_i satisfies ωπi′,τ=sgn⁡ki,τ∣⋅∣2ti,τ\omega_{\pi'_i,\tau} = \operatorname{sgn}^{k_{i,\tau}}\lvert\cdot\rvert^{2t_{i,\tau}}. By Lemma 2.4, applied to ω=ωπi′−1\omega= \omega_{\pi'_i}^{-1}, there is μi\mu_i with μi,τ=sgn⁡fτ∣⋅∣−ti,τ\mu_{i,\tau} = \operatorname{sgn}^{f_\tau}\lvert\cdot\rvert^{-t_{i,\tau}}. Since Dk⊗sgn⁡∘det⁡≃DkD_k \otimes\operatorname{sgn}\circ\det\simeq D_k, we get

(πi′⊗μi)τ≃Dki,τfor all τ.(\pi'_i \otimes\mu_i)_\tau\simeq D_{k_{i,\tau}} \qquad\text{for all }\tau.

Thus πi′⊗μi\pi'_i \otimes\mu_i is a Hilbert cusp form, and by (R7) it has a Galois representation ri=ρπi′⊗μir_i = \rho_{\pi'_i\otimes\mu_i}, which is irreducible and de Rham. Since ad⁡0(πi′⊗μi)=ad⁡0πi′\operatorname{ad}^{0}(\pi'_i \otimes\mu_i) = \operatorname{ad}^{0}\pi'_i and the scalar qvm/2q_v^{m/2} of (R7) cancels in ad⁡0\operatorname{ad}^{0}, the semisimple representations ad⁡0ri\operatorname{ad}^{0}r_i and Rad⁡0πi′R_{\operatorname{ad}^{0}\pi'_i} have the same characteristic polynomials of Frobenius at almost all places, and Chebotarev gives ad⁡0ri≃Rad⁡0πi′≃ad⁡0ρi\operatorname{ad}^{0}r_i \simeq R_{\operatorname{ad}^{0}\pi'_i} \simeq\operatorname{ad}^{0}\rho_i.

Step 8: conclusion

By Lemma 2.3, applied to the de Rham representations rir_i and ρi\rho_i, we have ρi≃ri⊗ψi\rho_i \simeq r_i \otimes\psi_i with ψi\psi_i a de Rham character of GFG_F. Put mi=max⁡τki,τ−1m_i = \max_\tau k_{i,\tau} - 1. By Lemma 2.1 and (R7),

ψi2=det⁡ρidet⁡ri=ξiεcycmi−wi.\psi_i^2 = \frac{\det\rho_i}{\det r_i} = \xi_i\varepsilon_{\mathrm{cyc}}^{m_i-w_i}.

where ξi\xi_i has finite order. Since ψi\psi_i is de Rham, its Hodge–Tate weight at every embedding is the integer hi=(wi−mi)/2h_i = (w_i-m_i)/2. Thus (ψiεcychi)2=ξi(\psi_i\varepsilon_{\mathrm{cyc}}^{h_i})^2 = \xi_i has finite image, so ψiεcychi\psi_i\varepsilon_{\mathrm{cyc}}^{h_i} itself has finite image. It follows that ψi\psi_i is a finite order character times an integral power of εcyc\varepsilon_{\mathrm{cyc}}. Therefore ρi\rho_i is modular, and it corresponds to a cuspidal representation π(ρi)\pi(\rho_i), a twist of πi′\pi'_i by a Hecke character. Finally we return to the original Π\Pi of (H2). By (R1), the isobaric representation π(ρ1)⊠π(ρ2)\pi(\rho_1)\boxtimes\pi(\rho_2) corresponds to ρ1⊗ρ2\rho_1\otimes\rho_2, as does Π\Pi, so Π≃π(ρ1)⊠π(ρ2)\Pi\simeq\pi(\rho_1)\boxtimes\pi(\rho_2) by strong multiplicity one. □\square

Remarks

Remark 4.1 (The role of each ingredient). We summarize how the hypotheses enter the proof.

  • Potential modularity is used through Lemma 2.2, always for a single ρi\rho_i: for ad⁡0ρi\operatorname{ad}^{0}\rho_i and its quadratic twists. No common field of potential modularity is needed. It is also used through Lemma 2.1, which gives the de Rham property needed in Steps 6–8.

  • The deep input in Steps 1–3 is the descent (R5), which recognizes the image of GL⁡2×GL⁡2\operatorname{GL}_2 \times\operatorname{GL}_2; this part is known [3, 2].

  • The role of (R8) is to match each automorphic adjoint with the adjoint of one of the given Galois factors, using one global ordering. This gives the labelled weights of that factor at every real place and proves that its weights have constant parity. If the weights of the automorphic factors are already known to be paritious, (R7) can instead be applied directly to πi′⊗μi\pi'_i \otimes\mu_i, and the same comparison of adjoints applies. This is automatic when F=QF = \mathbb{Q}.

Remark 4.2 (Comparison with the direct sum case). In [31 Lemma 5.1 and Theorem C], under his crystallinity and weight hypotheses, Ramakrishnan treats ρ=σ⊕σ′\rho= \sigma\oplus\sigma^{\prime} using Taylor’s potential modularity, Brauer’s theorem, Kim’s exterior square transfer [20], and results of Jacquet and Shalika. Here the elementary identity (R2), together with the adjoint lifts after descent, suffices. There are two differences.

(1) In the direct sum case, the decomposition π≃η⊞η′\pi\simeq\eta\boxplus\eta^{\prime} is read off from the order of the pole of L(s,π×π∨)L(s,\pi\times\pi^{\vee}), and it is unique. Here we use descent to GSpin4\mathrm{GSpin}_{4}, and the factorization has the ambiguity of interchanging the factors and replacing them by (π1′⊗μ,π2′⊗μ−1)(\pi^{\prime}_{1} \otimes\mu,\pi^{\prime}_{2} \otimes\mu^{-1}). Steps 5–8 identify automorphic representations corresponding to the given Galois factors.

(2) In the direct sum case, the archimedean parameters of η\eta and η′\eta^{\prime} are summands of those of π\pi, so there is no ambiguity in the weights. Here there is, and Step 6 resolves it.

Remark 4.3 (On the descent (R5)). We add four comments on the use of [17].

(1) The Main Theorem (even case) of [17 §12] assumes n≥2n \ge2 and that ω\omega is not the square of another Hecke character, whereas in our situation ω=χ\omega= \chi may be a square. The refined statement [17 Thm. 16.3.1], which is the one we use, is formulated for an arbitrary Hecke character ω\omega, and so are the results on Eisenstein series [17 Thm. 15.0.12] and on the vanishing of incompatible descents [17 Thm. 16.2.1] on which it rests. By [17 Rem. 18.0.5], the hypothesis that ω\omega is not a square only serves to force the rank mm to be even under ω−1\omega^{-1}-orthogonality, and the proof is blind to this parity. Here m=4m = 4 is already even.

(2) By [17 Rem. 2.2.2(1)], orthogonality may be defined using either complete or partial LL-functions.

(3) The passage from GSpin4\mathrm{GSpin}_{4} to GL2×GL2\mathrm{GL}_{2} \times\mathrm{GL}_{2} is not written out in [2 §5.2]. There it is only stated that Π\Pi is as in case (ii)(a) when μ=ωΠχ−2\mu= \omega_{\Pi}\chi^{-2} is trivial, and GSpin4\mathrm{GSpin}_{4} is described in §2.2. Step 3 supplies this passage via [23]. That a pole of the twisted symmetric square LL-function implies a transfer from an even general spin group is [3], based on [17].

(4) By [17 Thm. 16.3.1(2)], the central character of σ\sigma is χ\chi. With the identification GSpin4≃H\mathrm{GSpin}_{4} \simeq H this yields ωπ1ωπ2=χ\omega_{\pi_{1}}\omega_{\pi_{2}} = \chi, that is, ε=1\varepsilon= 1 already in Step 3. We do not use this: the determinant argument of Step 6 gives ε=1\varepsilon= 1 directly.

Acknowledgments

Acknowledgments. The author used Claude, an AI model developed by Anthropic, as an assistant during the preparation of this paper: for discussing and checking the arguments, for searching and verifying the literature and for help with writing and typesetting. All mathematical statements and proofs were checked by the author, who takes full responsibility for the content of the paper.

References

Departament de Matemàtiques i Informàtica, Universitat de Barcelona, Gran Via de les Corts Catalanes, 585, 08007 Barcelona

Email address: [email protected]

References

  1. [1]J. Arthur, L. Clozel, Simple algebras, base change, and the advanced theory of the trace formula, Ann. of Math. Stud. 120, Princeton Univ. Press, 1989.
  2. [2]M. Asgari, A. Raghuram, A cuspidality criterion for the exterior square transfer of cusp forms on GL(4), in On certain L-functions, Clay Math. Proc. 13 (2011), 33–53.
  3. [3]M. Asgari, F. Shahidi, Image of functoriality for general spin groups, Manuscripta Math. 144 (2014), 609–638; arXiv:1101.3467.arxiv.org/abs/1101.3467
  4. [4]T. Barnet-Lamb, T. Gee, D. Geraghty, R. Taylor, Potential automorphy and change of weight, Ann. of Math. 179 (2014), 501–609.
  5. [6]D. Blasius, J. Rogawski, Motives for Hilbert modular forms, Invent. Math. 114 (1993), 55–87.DOI
  6. [7]K. Buzzard, T. Gee, The conjectural connections between automorphic representations and Galois representations, in Automorphic forms and Galois representations, Vol. 1, London Math. Soc. Lecture Note Ser. 414, Cambridge Univ. Press, 2014, 135–187.
  7. [8]F. Calegari, T. Gee, Irreducibility of automorphic Galois representations of GL(n), n at most 5, Ann. Inst. Fourier 63 (2013), 1881–1912.arxiv.org/abs/1104.4827
  8. [9]F. Calegari, T. Gee, Corrigendum to “Irreducibility of automorphic Galois representations of GL(n), n at most 5”, Ann. Inst. Fourier 67 (2017), 267–268.DOI
  9. [10]H. Carayol, Sur les représentations ℓ-adiques associées aux formes modulaires de Hilbert, Ann. Sci. ÉNS 19 (1986), 409–468.
  10. [11]G. Chenevier, M. Harris, Construction of automorphic Galois representations II, Camb. J. Math. 1 (2013), 53–73.DOI
  11. [12]C. Chevalley, Deux théorèmes d’arithmétique, J. Math. Soc. Japan 3 (1951), 36–44.DOI
  12. [13]L. Clozel, Motifs et formes automorphes: applications du principe de fonctorialité, in Automorphic forms, Shimura varieties, and L-functions I, Academic Press, 1990, 77–159.
  13. [14]S. Gelbart, H. Jacquet, A relation between automorphic representations of GL(2) and GL(3), Ann. Sci. ÉNS 11 (1978), 471–542.
  14. [15]Y. Hu, F. Tan, The Breuil–Mézard conjecture for non-scalar split residually reducible representations, Ann. Sci. ÉNS 48 (2015), 1383–1421.arxiv.org/abs/1309.1658
  15. [17]J. Hundley, E. Sayag, Descent construction for GSpin groups, Mem. Amer. Math. Soc. 243 (2016), no. 1148; arXiv:1110.6788.
  16. [18]H. Jacquet, J. Shalika, A non-vanishing theorem for zeta functions of GLₙ, Invent. Math. 38 (1976), 1–16.DOI
  17. [19]H. Jacquet, J. Shalika, On Euler products and the classification of automorphic representations I, II, Amer. J. Math. 103 (1981).DOI
  18. [20]H. Kim, Functoriality for the exterior square of GL₄ and the symmetric fourth of GL₂, J. Amer. Math. Soc. 16 (2003), 139–183.DOI
  19. [21]M. Kisin, The Fontaine–Mazur conjecture for GL₂, J. Amer. Math. Soc. 22 (2009), 641–690.DOI
  20. [22]J.-P. Labesse, R. P. Langlands, L-indistinguishability for SL(2), Canad. J. Math. 31 (1979), 726–785.
  21. [23]J.-P. Labesse, J. Schwermer, Central morphisms and cuspidal automorphic representations, J. Number Theory 205 (2019), 170–193; arXiv:1812.03033.arxiv.org/abs/1812.03033
  22. [24]T. Liu, J.-K. Yu, On automorphy of certain Galois representations of GO₄-type, with an appendix by L. Xiao, J. Number Theory 161 (2016), 49–74.DOI
  23. [25]V. Paškūnas, On the Breuil–Mézard conjecture, Duke Math. J. 164 (2015), 297–359.DOI
  24. [27]S. Patrikis, R. Taylor, Automorphy and irreducibility of some ℓ-adic representations, Compos. Math. 151 (2015), 207–229.DOI
  25. [28]C. S. Rajan, Recovering modular forms and representations from tensor and symmetric powers, arXiv:math/0410387.DOI
  26. [29]D. Ramakrishnan, Modularity of the Rankin–Selberg L-series, and multiplicity one for SL(2), Ann. of Math. 152 (2000), 45–111.arxiv.org/abs/math/0007203
  27. [31]D. Ramakrishnan, Decomposition and parity of Galois representations attached to GL(4), in Automorphic representations and L-functions, Tata Inst. Fund. Res. Stud. Math. 22 (2013), 427–454; arXiv:1301.1668.
  28. [32]T. Saito, Hilbert modular forms and p-adic Hodge theory, Compos. Math. 145 (2009), 1081–1113.DOI
  29. [33]A. Shavali, Irreducibility and monodromy of automorphic Galois representations of GL(4), arXiv:2603.19768, 2026.arxiv.org/abs/2603.19768
  30. [34]S. W. Shin, Galois representations arising from some compact Shimura varieties, Ann. of Math. 173 (2011), 1645–1741.DOI
  31. [35]C. Skinner, A note on the p-adic Galois representations attached to Hilbert modular forms, Doc. Math. 14 (2009), 241–258.DOI
  32. [36]A. Snowden, On two-dimensional weight two odd representations of totally real fields, arXiv:0905.4266.arxiv.org/abs/0905.4266
  33. [37]R. Taylor, On Galois representations associated to Hilbert modular forms, Invent. Math. 98 (1989), 265–280.DOI
  34. [38]R. Taylor, Remarks on a conjecture of Fontaine and Mazur, J. Inst. Math. Jussieu 1 (2002), 125–143.DOI

Paper details

vlnc:W_gftshrm7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

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

  • Mathematics

Claude

View statement
Looking for related papers…
Contents