Torsion points on GL 2 \text{GL}_{2} -type abelian varieties Jessica Alessandrì, Nirvana Coppola Abstract . It is well known that the rational torsion of an abelian variety defined over a number field injects into the reduction modulo any sufficiently large prime, so the order of the torsion group divides the greatest common divisor of the sizes of points on the reduction at each prime. Drawing inspiration from Katz’s Inventiones paper (1981), we investigate the converse to this for abelian varieties of GL 2 \rm GL_{2} -type and exhibit a conjectural list of possible torsion orders for modular abelian varieties over ℚ \mathbb{Q} of dimension up to 5 5 . 2020 Mathematics Subject Classification Primary: 11G10, Secondary: 11F11, 11F80, 14G35, 14Q15 1. Introduction For any elliptic curve E over \mathbb{Q} , the structure of the torsion subgroup \Tors(E)(\mathbb{Q}) is well understood, thanks to the celebrated theorem of Mazur [ 19 , Theorem 8] , where he determined which finite abelian groups \Tors(E)(\mathbb{Q}) is isomorphic to and showed that |\Tors(E)(\mathbb{Q})|\leq 16 . Mazur’s work was then generalised to other number fields: Merel [ 20 ] showed that for any number field K and any elliptic curve E/K the order of the torsion subgroup \Tors(E)(K) is bounded by a constant that only depends on K . In unpublished work, Oesterlé [ 23 ] showed that this bound can be taken to be (3^{[K:\mathbb{Q}]+1})^{2} (for a proof, see [ 7 , Appendix A] ) . For higher dimensional abelian varieties over general number fields, although the growth of the torsion group can be bounded by a constant depending on the variety and the degree of the field of definition (see e.g. [ 16 ] and [ 3 ] for the case of local fields), no uniform bound is known. Even for abelian surfaces, a uniform bound on the size of \Tors(A)(\mathbb{Q}) was only provided recently, under some assumptions on the geometric endomorphism ring \End_{\overline{\mathbb{Q}}}(A) of A . Namely, in [ 15 ] , it is shown that if A is an abelian surface with potential quaternionic multiplication (that is if \End_{\overline{\mathbb{Q}}}(A) is a maximal order in a division quaternion algebra over \mathbb{Q} ) then |\Tors(A)(\mathbb{Q})|\leq 18 . Moreover, the only prime numbers dividing the order of a rational torsion point are 2 and 3 , and there is an explicit list of possible candidates that can appear as torsion subgroups. This list gives a complete classification of such groups in the special case of abelian surfaces of \text{GL}_{2} -type with potential quaternionic multiplication. In this paper, we work more generally on abelian varieties of \text{GL}_{2} -type defined over a number field K , without further restrictions on the dimension and on the endomorphisms. Our strategy is to use a classical approach in the study of rational points, that is, answering to a local-global problem . Indeed, for any abelian variety A/K , if \mathfrak{p} is a prime of good reduction for A , the group \Tors(A)(K) injects into the finite group of \mathbb{F}_{\mathfrak{p}} -points of the reduction of A modulo \mathfrak{p} . Thus we have that |\Tors(A)(K)| divides N(p):=|\tilde{A}(\mathbb{F}_{\mathfrak{p}})| , where \tilde{A} is the reduction of A modulo \mathfrak{p} . A famous question by Lang asks if the converse is also true. More precisely he asked the following. Question 1.1 . Let A be an abelian variety over a number field K and let m\geq 2 be an integer number. Suppose that N(\mathfrak{p})\equiv 0\pmod{m} holds for a set of places \mathfrak{p} of density one. Does there exist a K -isogenous abelian variety A^{\prime} for which the set of K -rational torsion points satisfies |\Tors(A^{\prime})(K)|\equiv 0\pmod{m}? Katz [ 12 ] finds a positive answer to this question for the case of elliptic curves, and to a weaker version of it for the case of abelian surfaces. More precisely: the set of primes dividing the order of the torsion subgroup of at least one variety A^{\prime} that is isogenous to A , and the set of primes dividing all of the N(\mathfrak{p}) ’s are the same. However, he proves that there are counterexamples for abelian varieties of higher dimension, even for this weaker result. After Katz, a lot of interest has been shown towards the classification of more counterexamples, see for example [ 5 , 6 ] and [ 4 ] . This problem is linked to the so-called local-global divisibility problem in algebraic groups (see [ 9 , §8] ), to which Gillibert and Ranieri give an answer in [ 10 , 11 ] , in the case of \text{GL}_{2} -type abelian varieties. In this work, we prove the following theorem, which gives a positive answer to Katz’s weaker question for \text{GL}_{2} -type abelian varieties, in all dimensions. Theorem (See Corollary 3.3 ) . Let A be an abelian variety of \text{GL}_{2} -type over a number field K . Let \Sigma be the set of primes \mathfrak{p} of K of good reduction for A such that the absolute ramification index of \mathfrak{p} is smaller than p-1 , where p is the rational prime under \mathfrak{p} . Then, the two integers \sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\qquad\text{ and }\qquad\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p}) are divisible by the same primes of good reduction for A . More precisely, for any rational prime \ell of good reduction for A , one has v_{\ell}\left(\sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\right)\leq v_{\ell}\left(\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p})\right). Moreover, if \ell is any such prime and it is totally inert in F , then v_{\ell}\left(\sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\right)=v_{\ell}\left(\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p})\right). This theorem will follow from Theorem 3.2 , where we reformulate Katz’s problem in terms of Galois representations. Finally, we show how one can apply our main results to study the torsion of modular abelian varieties defined over the rationals, which, by Serre’s conjectures, correspond to weight 2 newforms. We implement our results in the computational algebra system Magma , in order to produce tentative lists of possible torsion orders for modular abelian abelian varieties of dimension up to 5 , which are presented in Conjectures 4.4 and 4.5 . The code and the outputs can be found at [ 1 ] . Although it is not currently known if the lists are exhaustive, we remark that all the torsion orders appearing in the lists occur for at least one abelian variety, which can be exhibited in terms of the corresponding newform. This data enhances the data that is currently available in LMFDB [ 17 ] . For example, there exists at least one abelian surface in the isogeny class associated to the level 39 newform with LMFDB label 39.2.a.b which has torsion order equal to 28 . This abelian variety is not listed in LMFDB and there are no abelian surfaces with torsion order equal to 28 in the database either. 2. Setting In this section, we recall the definition of \text{GL}_{2} -type abelian variety and properties of the Galois representations attached to it, in order to reformulate Katz’s problem in terms of Galois-stable lattices. Let A be an abelian variety of dimension g defined over a number field K , with fixed algebraic closure \overline{K} . Definition 2.1 . We say that A is of \text{GL}_{2} -type if there exists a number field F of degree g over \mathbb{Q} , with an embedding: F\hookrightarrow\End_{K}(A)\otimes_{\mathbb{Z}}\mathbb{Q}. Remark 2.1 . In [ 24 ] , the author deals with abelian varieties A/K of \text{GL}_{2} -type having full endomorphism algebra defined over K , claiming that all of the “serious” results have to do with this case. We will not make such an assumption, since our main result holds for any abelian variety of \text{GL}_{2} -type. Let \ell be a rational prime. One can define the \ell -adic Galois representation attached to A as \rho_{\ell}:G_{K}=\Gal(\overline{K}/K)\rightarrow\Aut(T_{\ell}(A))\subseteq\Aut(V_{\ell}(A)), where: • T_{\ell}(A) is the \ell -adic Tate module of A (see e.g. [ 26 ] for the definition); • V_{\ell}(A)=T_{\ell}(A)\otimes_{\mathbb{Z}_{\ell}}\mathbb{Q}_{\ell} . Since A is of \text{GL}_{2} -type, one has a F -rational system of compatible representations (\rho_{\lambda})_{\lambda\mid\ell} (see [ 24 , 18 , 2 ] for the construction), such that \rho_{\lambda}:G_{K}\rightarrow\text{GL}_{2}(F_{\lambda}) is a 2-dimensional representation. The sketch of the construction is as follows: we have F_{\ell}=F\otimes_{\mathbb{Q}}\mathbb{Q}_{\ell}=\prod_{\lambda\mid\ell}F_{\lambda} , and V_{\lambda}(A)=V_{\ell}(A)\otimes_{F_{\ell}}F_{\lambda} . The action of G_{K} on V_{\lambda}(A) is the representation \rho_{\lambda} . We let \mathfrak{f}_{\lambda}=[F_{\lambda}:\mathbb{Q}_{\ell}] , and \mathbb{F}_{\lambda} be the residue field \mathcal{O}_{F_{\lambda}}/\lambda\mathcal{O}_{F_{\lambda}} , which has size \ell^{\mathfrak{f}_{\lambda}} . Moreover, we will need to use the lattice T_{\lambda}(A)=T_{\ell}(A)\otimes_{\mathcal{O}_{F_{\ell}}}\mathcal{O}_{F_{\lambda}} , where \mathcal{O}_{F_{\ell}}=\mathcal{O}_{F}\otimes_{\mathbb{Z}}\mathbb{Z}_{\ell} . Notation 2.2 . Let \mathfrak{p} be a finite place of K , such that A has good reduction at \mathfrak{p} . We denote by \Frob_{\mathfrak{p}} a Frobenius element in G_{K} corresponding to \mathfrak{p} ; since A has good reduction at \mathfrak{p} , the representations \rho_{\ell} and \rho_{\lambda} are unramified at \mathfrak{p} , thus \rho_{\ell}(\Frob_{\mathfrak{p}}) and \rho_{\lambda}(\Frob_{\mathfrak{p}}) are well defined. We denote by P_{\mathfrak{p}}^{\ell}(t)=\det(1-t\rho_{\ell}(\Frob_{\mathfrak{p}})) the usual characteristic polynomial of Frobenius, and by P_{\mathfrak{p}}^{\lambda}(t)=\det(1-t\rho_{\lambda}(\Frob_{\mathfrak{p}})) the characteristic polynomial of \rho_{\lambda}(\Frob_{\mathfrak{p}}) . Note that, a priori , P_{\mathfrak{p}}^{\lambda}(t) has coefficients in F_{\lambda} , but since the representations \rho_{\lambda} are F -rational, in fact P_{\mathfrak{p}}^{\lambda}(t)\in F[t] . 2.1. Katz and Lang’s problem For a given abelian variety A/K , and \mathfrak{p} a finite place of K of good reduction for A , we denote by \mathbb{F}_{\mathfrak{p}} the residue field of K at \mathfrak{p} , by \tilde{A}(\mathbb{F}_{\mathfrak{p}}) the \mathbb{F}_{\mathfrak{p}} -points on the reduction of A modulo \mathfrak{p} and we let N(\mathfrak{p})=|\tilde{A}(\mathbb{F}_{\mathfrak{p}})| . In [ 12 ] , the following question is proposed: given an abelian variety A/K , and m\geq 2 an integer number, if N(\mathfrak{p})\equiv 0\pmod{m} holds for a set of places \mathfrak{p} of density one, does there exist an isogenous abelian variety A^{\prime} for which the set of K -rational torsion points satisfies |\Tors(A^{\prime})(K)|\equiv 0\pmod{m}? In op. cit. this problem is reformulated in terms of power-of-a-prime isogenies and the Galois representation \rho_{\ell} . In particular, the question becomes the following. Let A/K be an abelian variety and \ell a rational prime such that there exists n\geq 1 satisfying \det(1-\rho_{\ell}(\gamma))\equiv 0\pmod{\ell^{n}} for all \gamma\in G_{K} . Note that, by Čebotarev Density Theorem [ 22 , Theorem 13.4] , this condition is equivalent to requiring the congruence for \gamma of the form \Frob_{\mathfrak{p}} for a density one set of places, namely those of good reduction not dividing \ell . Does there exist a pair of G_{K} -stable lattices \mathcal{L^{\prime}}\subseteq\mathcal{L}\subseteq V_{\ell}(A) such that the quotient has order \ell^{n} and such that G_{K} acts trivially on it? As mentioned, Katz finds a positive answer to this question for the case of elliptic curves [ 12 , Theorem 2, 2 bis] , and to a weaker version of it for the case of abelian surfaces [ 12 , Theorem 4] . We adapt Katz’s approach for the proof of his Theorem 2 to the case of \text{GL}_{2} -type abelian varieties, proving a refined version of his Theorem 4. 3. Main results We will use the following result of Katz and Lang [ 12 , Theorem 1 (bis)] on 2 -dimensional representations with coefficients in a local field. Theorem 3.1 . Let L be a field which is complete under a discrete valuation, and whose residue field is finite. Let \mathcal{O}_{L} denote the ring of integers in L , and let \pi denote a uniformising parameter. Let V be a 2-dimensional L -vector space, G\subseteq\Aut_{L}(V) a compact subgroup, and n\geq 1 an integer. Suppose that we have the congruence \det(1-g)\equiv 0\mod\pi^{n}\quad\forall g\in G. Then there exists a L -basis \{e_{1},e_{2}\} of V and integers a,b\geq 0 with a+b=n such that the elements of G expressed as matrices with respect to this basis lie in the subgroup \left\{\left(\begin{array}[]{cc}1+\pi^{a}\mathcal{O}_{L}&\pi^{a}\mathcal{O}_{L}\\ \pi^{b}\mathcal{O}_{L}&1+\pi^{b}\mathcal{O}_{L}\end{array}\right)\right\} of \text{GL}_{2}(\mathcal{O}_{L}) . We remark that, in [ 12 ] , a detailed proof of the case where L=\mathbb{Q}_{p} is provided. However, the proof adapts immediately to the general case. We are now ready to state our main result. Theorem 3.2 . Let A be an abelian variety over a number field K of \text{GL}_{2} -type. Let \ell be a rational prime such that, for all places of K above \ell , A has good reduction modulo such places (we simply say that \ell is of good reduction for A ). Let \lambda be a prime of F over \ell and suppose that, for an integer n\geq 1 , the congruence P_{\mathfrak{p}}^{\lambda}(1)\equiv 0\pmod{\lambda^{n}} holds for a density one set of finite places \mathfrak{p} of K . Then there exists an abelian variety A^{\prime} , K -isogenous to A through a \ell -power isogeny, such that |\Tors(A^{\prime})(K)|\equiv 0\pmod{\ell^{\mathfrak{f}_{\lambda}n}}, where \mathfrak{f}_{\lambda} is the inertia degree of \lambda\mid\ell . Proof. With the notation of Theorem 3.1 , let L=F_{\lambda} , so that \pi=\lambda , V=V_{\lambda}(A) and G=\rho_{\lambda}(G_{K}) . The condition \det(1-g)\equiv 0\pmod{\lambda^{n}} for all g\in G is equivalent to P_{\mathfrak{p}}^{\lambda}(1)\equiv 0\pmod{\lambda^{n}} for a density one set of \mathfrak{p} , by Čebotarev Density Theorem. Then, by Theorem 3.1 , there exist two sublattices \mathcal{L^{\prime}}\subseteq\mathcal{L} of V , such that G_{K} acts trivially on the quotient. Such quotient has order \ell^{\mathfrak{f}_{\lambda}n} , indeed in the basis \{e_{1},e_{2}\} given by the theorem, \mathcal{L}\cong\mathcal{O}_{F_{\lambda}}e_{1}+\mathcal{O}_{F_{\lambda}}e_{2} ; \mathcal{L}^{\prime}\cong\mathcal{O}_{F_{\lambda}}\lambda^{a}e_{1}+\mathcal{O}_{F_{\lambda}}\lambda^{b}e_{2} , with a+b=n , and the quotient is \mathcal{L}/\mathcal{L}^{\prime}\cong\dfrac{\mathcal{O}_{F_{\lambda}}e_{1}+\mathcal{O}_{F_{\lambda}}e_{2}}{\mathcal{O}_{F_{\lambda}}\lambda^{a}e_{1}+\mathcal{O}_{F_{\lambda}}\lambda^{b}e_{2}}\cong\dfrac{{\mathcal{O}_{F_{\lambda}}}}{\mathcal{O}_{F_{\lambda}}\lambda^{a}}\oplus\dfrac{\mathcal{O}_{F_{\lambda}}}{\mathcal{O}_{F_{\lambda}}\lambda^{b}}, which has size |\mathbb{F}_{\lambda}|^{a+b}=\ell^{\mathfrak{f}_{\lambda}n} . Now, the lattice \mathcal{L}^{\prime} is T_{\lambda}(A^{\prime}) for some abelian variety A^{\prime} with \lambda -adic Galois representation \rho_{\lambda} . In particular, A^{\prime} is K -isogenous to A with an isogeny of degree a power of \ell . Therefore, \Tors(A^{\prime})(K) contains a subgroup of order \ell^{\mathfrak{f}_{\lambda}n} . ∎ We remark that our assumption on the congruence of P_{\mathfrak{p}}^{\lambda}(1) modulo \lambda^{n} implies the hypothesis in Katz’s problem on the number of points of the reduction of A at \mathfrak{p} . In fact, as already mentioned, the condition N(\mathfrak{p})\equiv 0\pmod{\ell^{n}} is equivalent to P_{\mathfrak{p}}^{\ell}(1)=\det(1-\rho_{\ell}(\Frob_{\mathfrak{p}}))\equiv 0\pmod{\ell^{n}} , since N(\mathfrak{p})=P_{\mathfrak{p}}^{\ell}(1) . By [ 28 , §11.9-11.10] , the characteristic polynomial of \rho_{\ell}(\Frob_{\mathfrak{p}}) is equal to the norm from F to \mathbb{Q} of the characteristic polynomial of \rho_{\lambda}(\Frob_{\mathfrak{p}}) , for any \lambda over \ell . Then, it is a simple exercise to show that (1) v_{\ell}(P_{\mathfrak{p}}^{\ell}(1))=v_{\ell}(\mathcal{N}_{F/\mathbb{Q}}(P_{\mathfrak{p}}^{\lambda}(1)))=\sum_{\lambda^{\prime}\mid\ell}v_{\lambda^{\prime}}(P_{\mathfrak{p}}^{\lambda}(1))\mathfrak{e}_{\lambda^{\prime}}\mathfrak{f}_{\lambda^{\prime}}, where the sum varies over all primes \lambda^{\prime} lying over \ell and \mathfrak{e}_{\lambda^{\prime}} and \mathfrak{f}_{\lambda^{\prime}} are the ramification and inertia degree of \lambda^{\prime} , respectively. If P_{\mathfrak{p}}^{\lambda}(1)\equiv 0\pmod{\lambda^{n}} then v_{\lambda}(P_{\mathfrak{p}}^{\lambda}(1))\geq n and so v_{\ell}(P_{\mathfrak{p}}^{\ell}(1)) is at least n\mathfrak{e}_{\lambda}\mathfrak{f}_{\lambda} . In particular, P_{\mathfrak{p}}^{\ell}(1)\equiv 0\pmod{\ell^{n}} . Corollary 3.3 . Let A be an abelian variety of \text{GL}_{2} -type over a number field K . Let \Sigma be the set of primes \mathfrak{p} of K of good reduction for A such that the absolute ramification index of \mathfrak{p} is smaller than p-1 , where p is the rational prime under \mathfrak{p} . Then, the two integers \sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\qquad\text{ and }\qquad\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p}) are divisible by the same primes of good reduction for A . More precisely, for any rational prime \ell of good reduction for A , one has v_{\ell}\left(\sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\right)\leq v_{\ell}\left(\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p})\right). Moreover, if \ell is any such prime and it is totally inert in F , then v_{\ell}\left(\sup_{A^{\prime}\text{ K-isogenous to }A}|\Tors(A^{\prime})(K)|\right)=v_{\ell}\left(\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p})\right). Proof. First, let A^{\prime} be an abelian variety that is K -isogenous to A . By injectivity of the reduction map, which holds for all primes in \Sigma (see [ 12 , Appendix] ), \Tors(A^{\prime})(K) injects into the \mathbb{F}_{\mathfrak{p}} -points of A^{\prime} for all \mathfrak{p}\in\Sigma , thus |\Tors(A^{\prime})(K)| divides |A^{\prime}(\mathbb{F}_{\mathfrak{p}})|=N(\mathfrak{p}) . Taking the sup over all A^{\prime} that are K -isogenous to A and the gcd of all N(\mathfrak{p}) as \mathfrak{p} varies in \Sigma , we obtain \sup_{A^{\prime}K\text{-isogenous to}A}|\Tors(A^{\prime})(K)|\big|\gcd_{\mathfrak{p}\in\Sigma}N(\mathfrak{p}) Now, let \ell be a prime of good reduction for A dividing all N(\mathfrak{p}) for \mathfrak{p}\in\Sigma . As shown in Equation ( 1 ), (2) v_{\ell}(P_{\mathfrak{p}}^{\ell}(1))=\sum_{\lambda^{\prime}\mid\ell}v_{\lambda^{\prime}}(P_{\mathfrak{p}}^{\lambda}(1))\mathfrak{e}_{\lambda^{\prime}}\mathfrak{f}_{\lambda^{\prime}}=\sum_{\lambda^{\prime}\mid\ell}v_{\lambda^{\prime}}(P_{\mathfrak{p}}^{\lambda^{\prime}}(1))\mathfrak{e}_{\lambda^{\prime}}\mathfrak{f}_{\lambda^{\prime}}. The last equality follows from the compatibility of the representations \{\rho_{\lambda}\}_{\lambda} . Since this quantity is positive, there exists \lambda_{0} lying over \ell with v_{\lambda_{0}}(P_{\mathfrak{p}}^{\lambda_{0}}(1))=n>0 . By Theorem 3.2 , there exists an abelian variety A^{\prime} that is K -isogenous to A and such that v_{\ell}(|\Tors(A^{\prime})(K)|)\geq n\mathfrak{e}_{\lambda_{0}}\mathfrak{f}_{\lambda_{0}}>0 , concluding the proof of the first statement. Now, suppose that \lambda_{0} is the unique prime lying over \ell , and that \ell is totally inert, i.e. \mathfrak{e}_{\lambda_{0}}=1 . It is enough to show that there exists an abelian variety A^{\prime} that is K isogenous to A and such that v_{\ell}(\gcd N(\mathfrak{p}))\leq v_{\ell}(|\Tors(A^{\prime})(K)|) . We take A^{\prime} as in the previous step of the proof. Then v_{\ell}(N(\mathfrak{p}))=\sum_{\lambda^{\prime}\mid\ell}v_{\lambda^{\prime}}(P_{\mathfrak{p}}^{\lambda^{\prime}}(1))\mathfrak{e}_{\lambda^{\prime}}\mathfrak{f}_{\lambda^{\prime}}=v_{\lambda_{0}}(P_{\mathfrak{p}}^{\lambda_{0}}(1))\mathfrak{e}_{\lambda_{0}}\mathfrak{f}_{\lambda_{0}}=n\mathfrak{f}_{\lambda_{0}}. So, v_{\ell}(\gcd N(\mathfrak{p}))\leq n\mathfrak{f}_{\lambda_{0}}\leq v_{\ell}(|\Tors(A^{\prime})(K)|) . ∎ 4. Modular abelian varieties defined over \mathbb{Q} We now specialise the result of Theorem 3.2 and Corollary 3.3 to \text{GL}_{2} -type abelian varieties defined over the rational numbers. Let A be an abelian variety of dimension g defined over \mathbb{Q} . Let \End^{0}_{\mathbb{Q}}(A)=\End_{\mathbb{Q}}(A)\otimes_{\mathbb{Z}}\mathbb{Q} denote the algebra of endomorphisms of A that are defined over \mathbb{Q} , and let \End_{\overline{\mathbb{Q}}}^{0}(A)=\End_{\overline{\mathbb{Q}}}(A\times_{\mathbb{Q}}\overline{\mathbb{Q}})\otimes_{\mathbb{Z}}\mathbb{Q} denote the full endomorphism algebra. By [ 21 , p. 193] , if A is a geometrically simple abelian variety, both \End^{0}_{\mathbb{Q}}(A) and \End^{0}_{\overline{\mathbb{Q}}}(A) are Albert algebras. By [ 21 , Remark p. 182] , the dimensions of \End^{0}_{\mathbb{Q}}(A) and of \End_{\overline{\mathbb{Q}}}^{0}(A) divide 2g and by [ 25 , §2] , if \End_{\mathbb{Q}}^{0}(A) contains a number field, then its degree divides g . In particular, by Albert’s classification [ 21 , p. 201] , both endomorphism algebras are of one of the following types: • Type I : a totally real number field of degree dividing g ; • Type II or III : a quaternion algebra (either totally definite or totally indefinite) over a totally real number field of degree dividing g/2 ; • Type IV : a division algebra of dimension d^{2} over a CM field of degree dividing 2g/d^{2} . We now investigate the possible full endomorphisms algebras for simple abelian varieties of \text{GL}_{2} -type over \mathbb{Q} . As in § 2 , we write F\hookrightarrow\End_{\mathbb{Q}}^{0}(A) , where F is a degree g number field. By [ 25 , Theorem 2.1] , if A/\mathbb{Q} is simple, then F=\End_{\mathbb{Q}}^{0}(A) . Proposition 4.1 . Let A be a simple abelian variety of dimension g\geq 2 , defined over \mathbb{Q} , of \text{GL}_{2} -type. Then one of the following occurs: • A is not geometrically simple; • A is geometrically simple and has potential quaternionic multiplication (or PQM for short); • A is geometrically simple, with \End^{0}_{\overline{\mathbb{Q}}}(A)=\End^{0}_{\mathbb{Q}}(A) equal to a totally real number field of degree g . Proof. Let us assume that A is geometrically simple, so that \End^{0}_{\overline{\mathbb{Q}}}(A) is an Albert algebra. If it is of type IV, then by Shimura’s decomposition [ 25 , §5] , A_{\overline{\mathbb{Q}}} is a power of a CM elliptic curve, and in particular it is not geometrically simple, contradiction. Then, by Albert’s classification, if it A does not have PQM, then necessarily \End^{0}_{\overline{\mathbb{Q}}}(A) is of type I, i.e. it is a totally real number field of degree dividing g . Then we have F\subseteq\End^{0}_{\mathbb{Q}}(A)\subseteq\End^{0}_{\overline{\mathbb{Q}}}(A), with F as above. Thus all inclusions are equalities. ∎ By [ 25 , Theorem 4.4] and Serre’s modularity conjectures [ 27 ] (proved by Khare–Winterberger, [ 13 , 14 ] ), a simple abelian variety A/\mathbb{Q} of \text{GL}_{2} -type is modular, hence it is isogenous to the abelian variety A_{f} associated to some newform f , with the notation used in [ 25 ] . An explicit construction of A_{f} can be found in [ 8 , §9] , as well as the definition of \lambda -adic representations attached to f and to A_{f} . It is well known [ 8 , Theorem 9.6.5] that the representations (\rho_{f,\lambda}) form a strictly compatible system for all \lambda ’s, including those of bad reduction for A_{f} . Recall that F=\End_{\mathbb{Q}}(A_{f}) is equal to the coefficients field of f . Therefore, for A/\mathbb{Q} of \text{GL}_{2} -type, we can restate Theorem 3.2 and Corollary 3.3 as follows. Theorem 4.2 . Let A/\mathbb{Q} be an abelian variety of \text{GL}_{2} -type, such that A is isogenous to A_{f} for a modular form f with totally real coefficients field F . Let \ell be a rational prime and \lambda be a prime of F lying over \ell . Suppose that, for an integer n\geq 1 , the congruence \det(1-\rho_{f,\lambda}(\Frob_{p}))\equiv 0\pmod{\lambda^{n}} holds for a density one set of primes p . Then, there exists an abelian variety A^{\prime} that is \mathbb{Q} -isogenous to A through a \ell -power isogeny, such that |\Tors(A^{\prime})(\mathbb{Q})|\equiv 0\pmod{\ell^{f_{\lambda}n}}, where f_{\lambda} is the inertia degree of \lambda\mid\ell . Corollary 4.3 . Let A\sim A_{f} be as above. Let \Sigma be the set of primes of good reduction for A . Then, the two integers \sup_{A^{\prime}\,\mathbb{Q}\text{-isogenous to }A_{f}}|\Tors(A^{\prime})(\mathbb{Q})|\quad\text{and}\quad\gcd_{p\in\Sigma}N(p) are divisible by the same primes. More precisely, for any rational prime \ell , one has v_{\ell}\left(\sup_{A^{\prime}\,\mathbb{Q}\text{-isogenous to }A}|\Tors(A^{\prime})(\mathbb{Q})|\right)\leq v_{\ell}\left(\gcd_{p\in\Sigma}N(p)\right). Moreover, if \ell is any such prime and it is inert in F , then v_{\ell}\left(\sup_{A^{\prime}\,\mathbb{Q}\text{-isogenous to }A_{f}}|\Tors(A^{\prime})(\mathbb{Q})|\right)=v_{\ell}\left(\gcd_{p\in\Sigma}N(p)\right). 4.1. Implementation and conjectural results In [ 1 ] , we implement in Magma the results of Theorem 4.2 and Corollary 4.3 to produce a tentative list of torsion orders for modular abelian varieties of dimension g over \mathbb{Q} . The idea is to loop over all newforms of weight 2 and level up to a certain bound, consider the newforms having dimension (i.e. degree of the coefficient field over \mathbb{Q} ) equal to deg =g , and compute, for each, the greatest common divisor of P_{p}^{\lambda}(1)=\det(1-\rho_{f,\lambda}(\Frob_{p})) , as p varies among the primes of good reduction for A between 2 and a certain upper bound st , such that \mathfrak{e}_{p}