Existence and non-existence of rational elliptic curves with prescribed torsion subgroups over quadratic fields Ömer Avci Abstract . Let K = ℚ ⁡ ( − p ) K=\mathbb{Q}(\sqrt{-p}) be a quadratic field for an odd prime p p . We show that there exist infinitely many primes p p for which no elliptic curve E / ℚ E/\mathbb{Q} has torsion subgroup ℤ / 2 ​ ℤ × ℤ / 2 ​ N ​ ℤ \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2N\mathbb{Z} over K K for N = 5 , 6 N=5,6 . We also prove that there exist infinitely many primes p p for which there are infinitely many elliptic curves E / ℚ E/\mathbb{Q} with this torsion structure, conditional on the parity conjecture. Using these results, we obtain new torsion classification results over Kummer extensions of cyclotomic fields and over composites of ℤ p \mathbb{Z}_{p} -extensions of number fields, refining and extending our previous work. 1. Introduction and Notation Let K denote a number field, and let E be an elliptic curve over K . The Mordell-Weil theorem states that the set of K -rational points on E forms a finitely generated abelian group. More precisely, denoting E(K) as the set of K -rational points on E , we have E(K)\cong E(K)_{\text{tors}}\oplus\mathbb{Z}^{\operatorname{rk}(E(K))}. Here, \operatorname{rk}(E(K)) is a nonnegative integer called the rank, and E(K)_{\text{tors}} is a finite group known as the torsion subgroup of E over K . Classification of torsion subgroups E(K)_{\mathrm{tors}} is an ongoing problem. Mazur’s celebrated result solved the problem for K=\mathbb{Q} , and many results are known for other fields. There are several variants of the problem, for instance, whether one considers elliptic curves defined over K itself or over \mathbb{Q} . Chou classified the torsion subgroups E(\mathbb{Q}^{\mathrm{ab}})_{\mathrm{tors}} for elliptic curves E/\mathbb{Q} [ 7 ] . We studied the torsion subgroups E(\mathbb{Q}(\zeta_{p}))_{\mathrm{tors}} for all primes p , obtaining stronger results for certain families of primes [ 1 ] . At the time, our sharpest classification result was the following theorem. Theorem 1.1 (A., Theorem 1.1, [ 1 ] ) . Let E/\mathbb{Q} be an elliptic curve, and let p>3 be a prime such that p-1 is not divisible by 3 , 4 , or 5 . Then E(\mathbb{Q}(\zeta_{p}))_{\text{tors}} is either one of the groups from Mazur’s theorem, or one of the following groups: \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z},\quad\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z},\quad\text{or}\quad\mathbb{Z}/16\mathbb{Z}. Moreover, E(\mathbb{Q}(\zeta_{p}))_{\text{tors}}=E(\mathbb{Q}(\sqrt{-p}))_{\text{tors}} holds. In the same paper, we provided explicit examples realizing \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} for p=59 and \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} for p=47 . For the case \mathbb{Z}/16\mathbb{Z} , we could not find any examples, and were unable to show that it is not realized for any prime p . In June 2025, Derickx [ 8 , Theorem 5.2] proved that E(\mathbb{Q}(\sqrt{-p}))_{\text{tors}}\cong\mathbb{Z}/16\mathbb{Z} is not possible for any prime p and elliptic curve E/\mathbb{Q} . This shows that, excluding \mathbb{Z}/16\mathbb{Z} , the statement of our theorem is best possible. The question of determining for which primes p the groups \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2N\mathbb{Z} are realized for N=5,6 remains open. In this paper, we prove that for certain families of primes there are no rational elliptic curves with the desired torsion, and for certain families of primes we conjecture, and prove conditionally on the finiteness of the 2 -primary part of the Tate–Shafarevich group of rational elliptic curves, that there exist infinitely many rational elliptic curves with the desired torsion. Our main results are as follows. Theorem 1.2 . Let p\equiv 3,7\pmod{20} be a prime. Then there does not exist an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} Moreover, E(\mathbb{Q}(\zeta_{p}))_{\text{tors}} is isomorphic to one of the groups in Mazur’s theorem or \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} . Theorem 1.3 . Let p\equiv 11\pmod{24} be a prime. Then there does not exist an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. Moreover, E(\mathbb{Q}(\zeta_{p}))_{\text{tors}} is isomorphic to one of the groups in Mazur’s theorem or \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} . Corollary 1.4 . Let p\equiv 83,107\pmod{120} be a prime. Then there does not exist an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/N\mathbb{Z} for N=5 or N=6 . Moreover, E(\mathbb{Q}(\zeta_{p}))_{\text{tors}} is isomorphic to one of the groups in Mazur’s theorem. From this corollary, we obtain an infinite family of primes for which the possible groups appearing as E(\mathbb{Q}(\zeta_{p}))_{\mathrm{tors}} are exactly the 15 groups listed in Mazur’s theorem. Moreover, thanks to the work of Guzvić and Krijan, this also allows us to classify E(\mathbb{Q}(\mu_{p^{\infty}}))_{\mathrm{tors}} , since they proved that for all primes p>3 , E(\mathbb{Q}(\mu_{p^{\infty}}))_{\mathrm{tors}}=E(\mathbb{Q}(\zeta_{p}))_{\mathrm{tors}}. We also leave the following questions open: Conjecture 1.5 . Let p\equiv 11,19\pmod{20} be a prime. Then, there exists infinitely many elliptic curves E/\mathbb{Q} , satisfying E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}. Conjecture 1.6 . Let p\equiv-1\pmod{24} be a prime. Then, there exists infinitely many elliptic curves E/\mathbb{Q} , satisfying E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. The classification of torsion subgroups E(K)_{\text{tors}} for elliptic curves defined over quadratic fields K was completed by Kamienny [ 16 ] and by Kenku and Momose [ 18 ] . In addition, the classification of possible torsion subgroups E(K)_{\text{tors}} arising from elliptic curves E/\mathbb{Q} was carried out by Najman [ 22 ] . In these classifications, one encounters torsion groups that do not appear in Mazur’s theorem. However, although it is known that such groups are realized over some quadratic fields, a complete description of which quadratic fields admit which torsion structures is still unknown. This problem has attracted the attention of several authors in recent years. Notably, Banwait and Derickx studied related questions [ 5 ] , and we investigated further aspects [ 1 , 2 ] . As another example, Ejder [ 11 ] generalized Fujita’s [ 12 ] classification of torsion subgroups over the maximal quadratic extension of \mathbb{Q} to the cases of \mathbb{Q}(\sqrt{-1}) and \mathbb{Q}(\sqrt{-3}) , relying on the fact that the classification of torsion over cyclotomic quadratic fields had previously been completed by Najman [ 21 ] . Extending such classifications to more general quadratic fields may therefore lead to further generalizations of Ejder’s results. In Section 2 , we give a brief review of the 2 -descent method and of results related to quadratic points on the modular curves X_{1}(2,10) and X_{1}(2,12) . These two modular curves have genus 1 , and hence are elliptic curves themselves. We study the ranks of their quadratic twists and identify infinite families of quadratic twists for which the rank is zero, which implies that there are no rational elliptic curves with the desired torsion over the corresponding quadratic fields. Finally, we explain how to explicitly construct rational elliptic curves with the desired torsion over quadratic fields for which the rank is positive. In Section 3 , we discuss a result of Monsky [ 20 ] , which proves the parity conjecture for rational elliptic curves, conditional on the finiteness of the 2 -primary part of the Tate-Shafarevich group. We then prove our conjectures conditionally on the parity conjecture, or under the assumption that the 2 -primary part of \Sha is finite, an assumption which implies the parity conjecture for rational elliptic curves. In Section 4 , we refine and extend several of the classification results obtained earlier in [ 1 ] . Finally, in Section 5 , we establish new classification results related to the problems studied earlier [ 1 , 2 ] , using the results proved in Section 2 and the methods developed and refined in Section 4 , together with recent results from the literature. Throughout this paper, elliptic curves are identified using their LMFDB labels. In several parts of our analysis, we rely on computations performed with SageMath . These codes are publicly available on GitHub at https://github.com/omeravci372742/torsion-subgroups-sagemath-2 . Acknowledgements. I would like to thank my advisors, Antonio Lei and Payman Eskandari, and my partner in mathematics and in life, Sueda Senturk Avci, for their unwavering support, valuable comments, and insightful discussions. 2. Finding Quadratic Points of two Modular Curves In this section, we study the problem of finding quadratic points on the modular curves X_{1}(2,10) and X_{1}(2,12) . These two modular curves have genus 1 and hence are elliptic curves themselves. We begin with an important theorem stating that all non-cuspidal quadratic points, more precisely, the points of X_{1}(2,10)(K) and X_{1}(2,12)(K) for a quadratic field K that are not cusps, are non-torsion. This theorem also involves the modular curve X_{1}(11) , which likewise has genus 1 and is an elliptic curve. Since we have already established that X_{1}(2,10) , X_{1}(2,12) , and X_{1}(11) are elliptic curves, we denote their quadratic twists by d as X_{1}(2,10)_{d} , X_{1}(2,12)_{d} , and X_{1}(11)_{d} , respectively, for any nonzero integer d . Using this result, we reduce the problem to computing the ranks of the quadratic twists of these modular curves. We then apply the method of 2 -descent to carry out this computation for certain families of quadratic twists. In our calculations, we follow the notes of [ 6 ] , and we are grateful to Martin Bright for his clearly written exposition. Theorem 2.1 (Najman, Kamienny, Theorem 2-8-9-15, [ 17 ] ) . If X_{1}(11) , X_{1}(2,10) or X_{1}(2,12) possess a non-cuspidal quadratic point, then that point has infinite order. Corollary 2.2 . Let us denote the modular curve X_{1}(11) by the elliptic curve X_{1}(11):y^{2}-y=x^{3}-x^{2}. Let d be an integer. If \operatorname{rk}(X_{1}(11)(\mathbb{Q}(\sqrt{d})))=0 , then there is no elliptic curve E/\mathbb{Q}(\sqrt{d}) satisfying E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/11\mathbb{Z}. If \operatorname{rk}(X_{1}(11)(\mathbb{Q}(\sqrt{d})))\geq 1 , then there exist infinitely many elliptic curves E/\mathbb{Q}(\sqrt{d}) satisfying E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/11\mathbb{Z}. The same statement holds for the torsion structures \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} and \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} if we replace X_{1}(11) by the elliptic curves associated to the modular curves X_{1}(2,10) and X_{1}(2,12) , respectively, given by X_{1}(2,10):y^{2}=x^{3}+x^{2}-x,\qquad X_{1}(2,12):y^{2}=x^{3}-x^{2}+x. Proof. We know that the non-cuspidal \mathbb{Q}(\sqrt{d}) -points of X_{1}(11) correspond to elliptic curves E/\mathbb{Q}(\sqrt{d}) satisfying E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/11\mathbb{Z} . Since X_{1}(11) is an elliptic curve, we have X_{1}(11)(\mathbb{Q}(\sqrt{d}))\cong X_{1}(11)(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\oplus\mathbb{Z}^{\operatorname{rk}(X_{1}(11)(\mathbb{Q}(\sqrt{d})))} by the Mordell-Weil theorem. By Theorem 2.1 , X_{1}(11) has no non-cuspidal quadratic torsion points. Therefore, it has a non-cuspidal quadratic point if and only if its rank \operatorname{rk}(X_{1}(11)(\mathbb{Q}(\sqrt{d}))) is positive; in this case, it has infinitely many points. By using the equality \operatorname{rk}(X_{1}(11)(\mathbb{Q}(\sqrt{d})))=\operatorname{rk}(X_{1}(11)(\mathbb{Q}))+\operatorname{rk}(X_{1}(11)_{d}(\mathbb{Q})), we can reduce the problem to determining whether \operatorname{rk}(X_{1}(11)_{d}(\mathbb{Q})) is positive or zero, because it can be readily verified from LMFDB that \operatorname{rk}(X_{1}(11)(\mathbb{Q}))=0 , which is also evident from Mazur’s theorem. A similar argument applies to X_{1}(2,10) and X_{1}(2,12) , since \operatorname{rk}(X_{1}(2,10)(\mathbb{Q}))=\operatorname{rk}(X_{1}(2,12)(\mathbb{Q}))=0 according to LMFDB. ∎ In the case \operatorname{rk}(X_{1}(2,10)_{d})(\mathbb{Q})=0 (respectively for X_{1}(2,12) ), there do not exist elliptic curves defined over \mathbb{Q}(\sqrt{d}) with the desired torsion. This also implies that no rational elliptic curves with this torsion. If, on the other hand, \operatorname{rk}(X_{1}(2,10)_{d})(\mathbb{Q})\geq 1 (respectively for X_{1}(2,12) ), there exist infinitely many elliptic curves defined over \mathbb{Q}(\sqrt{d}) with the desired torsion. The following theorems demonstrate that, under these conditions, there also exist infinitely many rational elliptic curves with the same torsion structure. Proposition 2.3 . Let d be an integer, and let X_{1}(2,10)_{d} denote the d -quadratic twist of the modular curve X_{1}(2,10) . If \operatorname{rk}(X_{1}(2,10)_{d}(\mathbb{Q}))\geq 1 , then there exist infinitely many elliptic curves E/\mathbb{Q} with E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}. Proof. Since \operatorname{rk}(E_{d}(\mathbb{Q}))\geq 1 , there exist infinitely many rational solutions to the equation dy^{2}=x^{3}+x^{2}-x. In [ 15 , Theorem 3.2] , it is shown that if d=d(t)=8t^{3}-8t^{2}+1 with t\in\mathbb{Q} , and if E is the elliptic curve defined by E:y^{2}+(1-c)xy-by=x^{3}-bx^{2}, where b=\frac{t^{3}(2t^{2}-3t+1)}{(t^{2}-3t+1)^{2}},\quad c=-\frac{t(2t^{2}-3t+1)}{t^{2}-3t+1},\quad t\neq 0,\frac{1}{2},1, then the torsion subgroup of E over \mathbb{Q}(\sqrt{d}) is \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} . A change of variables x=2t-1 shows that a rational solution (x,y) to the first equation determines a rational value of t . This rational t in turn produces rational values of b and c , so that the corresponding elliptic curve E is defined over \mathbb{Q} . Moreover, E has the desired torsion subgroup over \mathbb{Q}(\sqrt{dy^{2}}) , which coincides with \mathbb{Q}(\sqrt{d}) . Since the rank is positive, there are infinitely many distinct rational solutions (x,y) to the elliptic curve equation. Distinct x -values correspond to distinct t -values, which in turn give distinct j -invariants. Therefore, we obtain infinitely many distinct rational elliptic curves. This follows because each solution corresponds to a distinct quadratic point on the modular curve X_{1}(2,10) , and each such quadratic point corresponds to a different elliptic curve with the desired torsion. ∎ Proposition 2.4 . Let d be an integer, and let X_{1}(2,12)_{d} denote the d -quadratic twist of the modular curve X_{1}(2,12) . If \operatorname{rk}(X_{1}(2,12)_{d}(\mathbb{Q}))\geq 1 , then there exist infinitely many elliptic curves E/\mathbb{Q} with E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. Proof. Similar to the proof of Proposition 2.3 , [ 15 , Theorem 3.3] states that if d=d(t)=\frac{t^{2}-1}{t^{2}+3} with t\in\mathbb{Q} , and if E is the elliptic curve defined by E:y^{2}+(1-c)xy-(c+c^{2})y=x^{3}-(c+c^{2})x^{2}, where c=\frac{1-t^{2}}{t^{4}+3t^{2}},\quad t\neq-1,0,1, then the torsion subgroup of E over \mathbb{Q}(\sqrt{d}) is \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} . In this case, obtaining t from a rational solution of dy^{2}=x^{3}-x^{2}+x, which is the d -quadratic twist of X_{1}(2,12) , is slightly more subtle. Following [ 15 ] , we may replace d(t) by (t^{2}-1)(t^{2}+3) without changing the quadratic field \mathbb{Q}(\sqrt{d}) . This leads to the hyperelliptic curve dz^{2}=(t^{2}-1)(t^{2}+3), for which we seek rational solutions (t,z) . This hyperelliptic curve is birationally equivalent to our elliptic curve via t=\frac{x+1}{x},\quad z=\frac{y}{x^{2}}. Consequently, rational solutions to the elliptic curve correspond to rational elliptic curves with the desired torsion. Since different x -values yield distinct t -values, they produce distinct j -invariants, giving infinitely many distinct rational elliptic curves. This is because each solution corresponds to a distinct quadratic point on X_{1}(2,12) , which in turn corresponds to a distinct elliptic curve with the desired torsion. ∎ Before starting our next proofs, which rely heavily on 2 -descent, we give a brief recap of the method following the notes of Bright [ 6 ] . Let E/\mathbb{Q} be an elliptic curve. If E admits a rational 2 -isogeny, then it must have a non-trivial 2 -torsion point. Our main goal is to compute the quotient E(\mathbb{Q})/2E(\mathbb{Q}) , which determines the rank \operatorname{rk}(E(\mathbb{Q})) once E(\mathbb{Q})[2] is known. Assume that E(\mathbb{Q})[2] is non-trivial. Then E can be written in the form E:y^{2}=x(x^{2}+ax+b),\quad a,b\in\mathbb{Q}. With this form, we have the following lemma, which constructs an explicit 2 -isogeny from E to another elliptic curve E^{\prime}/\mathbb{Q} . Lemma 2.5 . There exists an elliptic curve E^{\prime} over \mathbb{Q} and an isogeny \phi:E\to E^{\prime} with kernel \{O,(0,0)\} . Specifically, E^{\prime} is given by E^{\prime}:y^{2}=x(x^{2}+a^{\prime}x+b^{\prime}),\quad a^{\prime}=-2a,\;b^{\prime}=a^{2}-4b, and \phi is defined by \phi(x,y)=\Bigl(x+a+\frac{b}{x},\;y-\frac{by}{x^{2}}\Bigr), for all (x,y)\in E(\mathbb{C}) with (x,y)\neq(0,0) , which is mapped to O . From E^{\prime} we can define another isogeny \phi^{\prime} and elliptic curve E^{\prime\prime} : E^{\prime\prime}:y^{2}=x(x^{2}+a^{\prime\prime}x+b^{\prime\prime}),\quad a^{\prime\prime}=-2a^{\prime}=4a,\quad b^{\prime\prime}=(a^{\prime})^{2}-4b^{\prime}=16b. The curve E^{\prime\prime} is isomorphic to E via x\mapsto 4x and y\mapsto 8y . This provides the dual isogeny \hat{\phi}:E^{\prime}\to E given by \hat{\phi}(x,y)=\Bigl(\frac{1}{4}\bigl(x+a^{\prime}b^{\prime}/x\bigr),\;\frac{1}{8}\bigl(y-b^{\prime}y/x^{2}\bigr)\Bigr). The kernel of \hat{\phi} is \{O,(0,0)\}\subseteq E^{\prime}(\mathbb{Q}) , and the composite \hat{\phi}\circ\phi:E\to E equals the multiplication-by- 2 map. Next, we study the image of E(\mathbb{Q}) in E^{\prime}(\mathbb{Q}) . In particular, we are interested in the quotient E^{\prime}(\mathbb{Q})/\phi(E(\mathbb{Q})) . To compute this, we define the following map: Lemma 2.6 . Define \varphi:E^{\prime}(\mathbb{Q})\to\mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2} by \displaystyle\varphi((x,y)) \displaystyle=\overline{x},\quad x\neq 0, \displaystyle\varphi((0,0)) \displaystyle=\overline{a^{2}-4b}, \displaystyle\varphi(O) \displaystyle=\overline{1}. Then \varphi is a homomorphism of groups, and the sequence E(\mathbb{Q})\xrightarrow{\phi}E^{\prime}(\mathbb{Q})\xrightarrow{\varphi}\mathrm{\mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2}} is exact. In particular, \phi(E(\mathbb{Q}))=\ker\varphi,\quad\text{and}\quad E^{\prime}(\mathbb{Q})/\phi(E(\mathbb{Q}))\cong\varphi(E^{\prime}(\mathbb{Q})). Thus, computing the quotient E^{\prime}(\mathbb{Q})/\phi(E(\mathbb{Q})) reduces to finding the image of E^{\prime}(\mathbb{Q}) in \mathrm{\mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2}} . The following lemma gives an explicit criterion: Lemma 2.7 . Let r be a square-free integer. Then \overline{r}\in\mathrm{\mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2}} lies in the image of \varphi if and only if the equation r^{2}l^{4}+a^{\prime}rl^{2}m^{2}+b^{\prime}m^{4}=rn^{2} has a non-zero solution (l,m,n)\in\mathbb{Z}^{3} . Moreover, this can occur only if r\mid b^{\prime} , and one may assume \gcd(l,m)=1 . We now present the last piece of the puzzle for computing E(\mathbb{Q})/2E(\mathbb{Q}) . After this, we turn to calculations for the modular curves X_{1}(2,10) and X_{1}(2,12) and their -p -quadratic twists. Lemma 2.8 (Kernel-cokernel sequence) . Let A\xrightarrow{f}B\xrightarrow{g}C be homomorphisms of abelian groups. Then the sequence 0\to\ker f\xrightarrow{\mathrm{id}}\ker(g\circ f)\xrightarrow{f}\ker g\xrightarrow{\mathrm{id}}\operatorname{coker}f\xrightarrow{g}\operatorname{coker}(g\circ f)\xrightarrow{\mathrm{id}}\operatorname{coker}g\to 0 is exact. Proposition 2.9 . Let E:y^{2}=x(x^{2}+x-1) be the modular curve X_{1}(2,10) , and let p\equiv 3\pmod{4} be a prime. Let E_{-p} denote the (-p) -quadratic twist of E , given by E_{-p}:y^{2}=x(x^{2}-px-p^{2}). Then the Mordell–Weil rank of E_{-p} over \mathbb{Q} satisfies \operatorname{rk}(E_{-p}(\mathbb{Q}))=0\quad\text{if }p\equiv 2,3\pmod{5}, and 0\leq\operatorname{rk}(E_{-p}(\mathbb{Q}))\leq 1\quad\text{if }p\equiv 1,4\pmod{5}. Proof. Applying 2 -descent via the 2 -isogeny between E_{-p} and E^{\prime}_{-p} as described in Lemma 2.5 , we obtain \displaystyle E_{-p} \displaystyle:y^{2}=x(x^{2}-px-p^{2}), \displaystyle E^{\prime}_{-p} \displaystyle:y^{2}=x(x^{2}+2px+5p^{2}). To determine the image \varphi(E^{\prime}_{-p}(\mathbb{Q})) in \mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2} , we need to solve the equation r^{2}l^{4}+2prl^{2}m^{2}+5p^{2}m^{4}=rn^{2}, where r divides 5p^{2} . Since r is assumed to be square-free, it suffices to consider r=\pm 1,\;\pm 5,\;\pm p,\;\pm 5p. Because the left-hand side is always positive, we may discard the negative values. Moreover, by the definition of the map \varphi described in Lemma 2.6 , the classes \overline{1} and \overline{5} already lie in its image. Hence it remains to check the cases r=p and r=5p . If p\mid r , then p^{2} divides the left-hand side of the equation, which implies that p\mid n . Writing n=pk , we proceed as follows. If r=p , then we obtain l^{4}+2l^{2}m^{2}+5m^{4}=pk^{2}. Reducing modulo p , this yields (l^{2}+m^{2})^{2}+4m^{4}\equiv 0\pmod{p}. At this point, the assumption p\equiv 3\pmod{4} becomes crucial. Since -1 is not a quadratic residue modulo p , the congruence implies that p divides both l and m . This contradicts the assumption \gcd(l,m)=1 . If r=5p , then we obtain 5l^{4}+2l^{2}m^{2}+m^{4}=5pk^{2}. Reducing modulo p , this yields (l^{2}+m^{2})^{2}+4l^{4}\equiv 0\pmod{p}. Since -1 is not a quadratic residue modulo p , the congruence implies that p divides both l and m . This contradicts the assumption \gcd(l,m)=1 . Hence E^{\prime}_{-q}(\mathbb{Q})/\phi(E_{-q}(\mathbb{Q}))\cong\varphi(E^{\prime}_{-q}(\mathbb{Q}))\cong\mathbb{Z}/2\mathbb{Z}. We now turn to the computation of the image of \varphi^{\prime} in \mathbb{Q}^{\times}/(\mathbb{Q}^{\times})^{2} . In order to determine \varphi^{\prime}(E_{-p}) , we must solve the equation r^{2}l^{4}-prl^{2}m^{2}-p^{2}m^{4}=rn^{2}, where r divides -p^{2} . Since r is square-free, it suffices to consider the cases r=\pm 1,\pm p . We describe \varphi^{\prime} similarly to the definition of \varphi in Lemma 2.6 . By the definition of the map \varphi^{\prime} , the classes \overline{1} and \overline{-1} lie in the image \varphi^{\prime}(E_{-p}) . Therefore, it remains to determine whether \overline{p} or \overline{-p} lies in the image. Since \varphi^{\prime} is a group homomorphism and \overline{-1} already belongs to the image, it is enough to check one of these two cases. If r=p , then we obtain p^{2}l^{4}-p^{2}l^{2}m^{2}-p^{2}m^{4}=pn^{2}. It follows that p^{2} divides the left-hand side, and hence p divides n . Setting n=pk , we obtain l^{4}-l^{2}m^{2}-m^{4}=pk^{2}. Reducing modulo p , this yields (2l^{2}-m^{2})^{2}-5m^{4}\equiv 0\pmod{p}. When p\equiv\pm 2\pmod{5} , no solution exists because 5 is not a quadratic residue modulo p . This establishes that the image of \varphi^{\prime} has exactly 2 elements in this case, and by the kernel-cokernel sequence given in Lemma 2.8 , we conclude that \operatorname{rk}_{(}E_{-p}(\mathbb{Q}))=0 . On the other hand, if p\equiv\pm 1\pmod{5} , then the image of \varphi^{\prime} has either 2 or 4 elements, which implies 0\leq\operatorname{rk}_{(}E_{-p}(\mathbb{Q}))\leq 1. In this case, our SageMath calculations suggest that the rank is exactly 1 , and we will prove this statement conditionally on the parity conjecture in Section 3 . ∎ Corollary 2.10 . Let p be a prime such that p\equiv 3,7\pmod{20} . Then there does not exist an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{-p}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}. Proposition 2.11 . Let E:y^{2}=x(x^{2}+x-1) be the equation of the modular curve X_{1}(2,10) . Let d=\prod_{i=1}^{n}(-p_{i}) for some integer n\geq 1 , where p_{i} are distinct primes satisfying p_{i}\equiv 3,7\pmod{20} for all 1\leq i\leq n . Let E_{d} denote the d -quadratic twist of E , given by E_{d}:y^{2}=x(x^{2}+dx-d^{2}). Then \operatorname{rk}(E_{d}(\mathbb{Q}))=0. Proof. Applying 2 -descent using the 2 -isogeny between E_{d} and E^{\prime}_{d} gives \displaystyle E_{d}:y^{2} \displaystyle=x(x^{2}+dx-d^{2}), \displaystyle E^{\prime}_{d}:y^{2} \displaystyle=x(x^{2}-2dx+5d^{2}). To find the image \varphi(E^{\prime}_{d}) , we need to solve the equation r^{2}l^{4}-2drl^{2}m^{2}+5d^{2}m^{4}=rn^{2}, where r divides 5d^{2} . By definition of the map, \overline{1} and \overline{5} belong to the image of \varphi . We will show that the image consists exactly of these two elements. Since the right-hand side is positive, r>0 must hold. Assume that a prime p dividing d also divides r . Write d=pf and r=ps , yielding p^{2}s^{2}l^{4}-2p^{2}fsl^{2}m^{2}+5p^{2}f^{2}m^{4}=psn^{2}. The left-hand side must be divisible by p^{2} since the right-hand side is divisible by p^{2} . Both d and r are square-free, so p does not divide f or s . Hence, p must divide n ; write n=pk and divide both sides by p^{2} to obtain s^{2}l^{4}-2fs^{2}m^{2}+5f^{2}m^{4}=psk^{2}. The left-hand side can be rewritten as a sum of squares: s^{2}l^{4}-2fsl^{2}m^{2}+5f^{2}m^{4}=(sl^{2}-fm^{2})^{2}+(2fm^{2})^{2}. Using the assumption p\equiv 3\pmod{4} , we deduce that p divides both sl^{2}-fm^{2} and 2fm^{2} . Since p does not divide s or f , it follows that p divides l and m , contradicting \gcd(l,m)=1 . We conclude that r is not divisible by any prime factor of d . As r is a positive square-free integer dividing 5d^{2} , it follows that r=1 or r=5 , so the image \varphi(E^{\prime}_{d}) has exactly two elements. Next, to find the image \varphi^{\prime}(E_{d}) , we solve r^{2}l^{4}+drl^{2}m^{2}-d^{2}m^{4}=rn^{2}, where r divides -d^{2} . By definition, \overline{1} and \overline{-1} are in the image of \varphi^{\prime} . We will show that these are the only elements. Assume that a prime p dividing d also divides r , and write d=pf , r=ps . Then p^{2}s^{2}l^{4}+p^{2}fsl^{2}m^{2}-p^{2}f^{2}m^{4}=psn^{2}. As before, p^{2} divides the left-hand side, so p must divide n ; write n=pk and divide both sides by p^{2} : s^{2}l^{4}+fsl^{2}m^{2}-f^{2}m^{4}=psk^{2}. The left-hand side can be rewritten as (2sl^{2}+fm^{2})^{2}-5(fm^{2})^{2}=4s^{2}l^{4}+4fsl^{2}m^{2}-4f^{2}m^{4}=4psk^{2}. The assumption p\equiv\pm 2\pmod{5} ensures that 5 is not a quadratic residue modulo p , forcing p to divide 2sl^{2}+fm^{2} and fm^{2} , hence also sl^{2} . Since p does not divide f or s , it must divide l and m , contradicting \gcd(l,m)=1 . Therefore, r=\pm 1 , and the image \varphi^{\prime}(E_{d}) has exactly two elements. By the kernel-cokernel sequence given in Lemma 2.8 , we conclude that \mathrm{rk}(E_{d}(\mathbb{Q}))=0. ∎ Remark . An analogous theorem does not hold for X_{1}(2,12) . As a counterexample, we have \mathrm{rk}_{\mathbb{Q}}(E_{d})=2 for d=-11\cdot 59\cdot 83 , as can be verified using our SageMath calculations. Corollary 2.12 . Let d=\prod_{i=1}^{n}(-p_{i}) for some integer n\geq 1 , where the primes p_{i} satisfy p_{i}\equiv 3,7\pmod{20} for all 1\leq i\leq n . Then there does not exists an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{d}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}. Proposition 2.13 . Let E:y^{2}=x(x^{2}-x+1) be the equation of the modular curve X_{1}(2,12) . Let p\equiv 11\pmod{12} be a prime, and let E_{-p} denote the -p -quadratic twist of E given by E_{-p}:y^{2}=x(x^{2}+px+p^{2}). Then the Mordell–Weil rank of E_{-p} over \mathbb{Q} satisfies \operatorname{rk}(E_{-p}(\mathbb{Q}))=0\quad\text{if }p\equiv 11\pmod{24}, and 0\leq\operatorname{rk}(E_{-p}(\mathbb{Q}))\leq 1\quad\text{if }p\equiv 23\pmod{24}. Proof. Applying 2 -descent using the 2 -isogeny between E_{-p} and E^{\prime}_{-p} gives \displaystyle E_{-p}: \displaystyle\ y^{2}=x(x^{2}+px+p^{2}), \displaystyle E^{\prime}_{-p}: \displaystyle\ y^{2}=x(x^{2}-2px-3p^{2}). To compute the image \varphi^{\prime}(E_{-p}(\mathbb{Q})) , we need to solve r^{2}l^{4}+prl^{2}m^{2}+p^{2}m^{4}=rn^{2} with r\mid p^{2} . It suffices to check r=\pm 1,\pm p . Since \overline{1} belongs to the image and the left-hand side is always positive, we only need to consider r=p . In this case, p\mid n , so put n=pk to obtain l^{4}+l^{2}m^{2}+m^{4}=pk^{2}. Assuming \gcd(l,m)=1 , this equation requires -3 to be a quadratic residue modulo p . Since p\equiv 2\pmod{3} , this is impossible. Hence, \varphi^{\prime}(E_{-p}(\mathbb{Q})) is trivial. Next, for the image \varphi(E^{\prime}_{-p}(\mathbb{Q})) , we solve r^{2}l^{4}-2prl^{2}m^{2}-3p^{2}m^{4}=rn^{2} with r\mid-3p^{2} . We only need to consider r=\pm 1,\pm 3,\pm p,\pm 3p . By definition, \overline{1},\overline{-3} belong to the image. Furthermore, putting l=m=1 and n=0 , we see that \overline{-p} also belongs to the image. Since the image is a group, it suffices to check r=p , which determines whether \operatorname{rk}(E_{-p})=0 or 1 . For r=p , let n=pk ; then l^{4}-2l^{2}m^{2}-3m^{4}=(l^{2}+m^{2})(l^{2}-3m^{2})=pk^{2}. Assuming \gcd(l,m)=1 , we have \gcd(l^{2}+m^{2},l^{2}-3m^{2})=1 or 2 . If exactly one of l or m is even, the left-hand side is congruent modulo 4 to either l^{4} or -3m^{4} , which is always 1 , contradicting p\equiv 3\pmod{4} . Therefore, both l and m are odd, giving l^{2}+m^{2}=2u^{2},\qquad l^{2}-3m^{2}=2pv^{2} for some integers u,v . Parametrizing the first equation yields \pm l=\alpha^{2}-2\beta^{2},\qquad\pm m=\alpha^{2}-4\alpha\beta+2\beta^{2} with \gcd(\alpha,\beta)=1 . Using p\mid l^{2}-3m^{2} , we have l\equiv\omega m\pmod{p} for some \omega with \omega^{2}\equiv 3\pmod{p} . Then \alpha^{2}(1-\omega)+4\omega\alpha\beta-2(1+\omega)\beta^{2}\equiv 0\pmod{p}. The discriminant of this quadratic is 32 , which must be a quadratic residue modulo p . Hence p\equiv-1\pmod{8} must hold, under the assumption p\equiv 3\pmod{4} . This shows that if p\equiv 11\pmod{24} , then \operatorname{rk}(E_{-p}(\mathbb{Q}))=0 . If p\equiv 23\pmod{24} , then 0\leq\operatorname{rk}(E_{-p}(\mathbb{Q}))\leq 1 . ∎ Corollary 2.14 . Let p be a prime such that p\equiv 11\pmod{24} . Then there is no elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{-p}))_{\text{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. Proposition 2.15 . Let E:y^{2}=x(x^{2}-x+1) be the equation of the modular curve X_{1}(2,12) . Let d=pq for some distinct primes p,q satisfying p,q\equiv 11\pmod{24} . Let E_{d} be the d -quadratic twist of E given by the equation E_{d}:y^{2}=x(x^{2}-dx+d^{2}). Then rk(E_{d}(\mathbb{Q}))=0 . Proof. Applying 2 -descent via the 2 -isogeny between E_{d} and E^{\prime}_{d} gives \displaystyle E_{d}:y^{2} \displaystyle=x(x^{2}-dx+d^{2}), \displaystyle E^{\prime}_{d}:y^{2} \displaystyle=x(x^{2}+2dx-3d^{2}). For the image \varphi^{\prime}(E_{d}(\mathbb{Q})) , we need to solve r^{2}l^{4}-drl^{2}m^{2}+d^{2}m^{4}=rn^{2} with r\mid d^{2} . Since r is square-free, it must divide d . The left-hand side is always positive, so we have r>0 . We know that \overline{1} belongs to the image. We will show that r=1 is the only possibility. Without loss of generality, assume that prime p divides r , aiming for a contradiction. Write r=ps , where s=1 or s=q . Then p^{2}s^{2}l^{4}-p^{2}q^{2}l^{2}m^{2}+p^{2}q^{2}m^{4}=psn^{2}. Observe that p^{2} divides the left-hand side, but p^{2} does not divide ps , so p must divide n . Write n=pk , giving s^{2}l^{4}-q^{2}l^{2}m^{2}+q^{2}m^{4}=psk^{2}. We see that p divides the right-hand side. Since p\equiv 2\pmod{3} , -3 is not a quadratic residue modulo p , so p must divide both sl^{2} and qm^{2} . Recall that s=1 or s=q , and q is a prime distinct from p , so p must divide both l and m . This contradicts \gcd(l,m)=1 . Hence, the image of \varphi^{\prime}(E_{-p}(\mathbb{Q})) is trivial. For the image \varphi(E^{\prime}_{d}(\mathbb{Q})) , we need to solve r^{2}l^{4}+2drl^{2}m^{2}-3d^{2}m^{4}=rn^{2} with r\mid 3d^{2} . By the definition of the map \varphi , we know that \overline{1} and \overline{-3} belong to the image. We will show that the image \varphi^{\prime}(E_{d}(\mathbb{Q})) is exactly the set \{\overline{1},\overline{-3}\} . Since r is a square-free integer, and [-3] belongs to the image, the primes p and q are symmetric. Thus, it suffices to check whether r=p or r=-p is possible. If r=p , then p^{2}l^{4}+2p^{2}ql^{2}m^{2}-3p^{2}q^{2}m^{4}=pn^{2}. The left-hand side is divisible by p^{2} , so p must divide n . Writing n=pk gives l^{4}+2ql^{2}m^{2}-3q^{2}m^{4}=pk^{2}. Now we have two cases: either q divides l or it does not. If q does not divide l , then modulo q the left-hand side reduces to l^{4} , which is a quadratic residue modulo q . Since q does not divide l , it also does not divide k , so p must be a quadratic residue modulo q . However, both p and q are congruent to 3\pmod{4} , so by quadratic reciprocity, q is not a quadratic residue modulo p , leading to a contradiction. Factorizing the left-hand side, we get (l^{2}-qm^{2})(l^{2}+3qm^{2})=pk^{2}. Since \gcd(l,m)=1 and q is not a quadratic residue modulo p , we see that p does not divide l^{2}-qm^{2} , so it must divide l^{2}+3qm^{2} . Observe that the greatest common divisor of the two factors on the left can be either 1 , if one of l and m is odd and the other is even, or a power of 2 , if both l and m are odd. Furthermore, l^{2}+3qm^{2} is always positive. Thus, we have two possibilities: l^{2}-qm^{2}=u^{2}\quad\text{and}\quad l^{2}+3qm^{2}=pv^{2} or l^{2}-qm^{2}=2u^{2}\quad\text{and}\quad l^{2}+3qm^{2}=2pv^{2} for some integers u and v . The second possibility cannot occur because l^{2}-qm^{2} is a quadratic residue modulo q , while 2 is not a quadratic residue modulo q since q\equiv 3\pmod{8} . The first possibility also fails because l^{2}+3qm^{2} can only take values 0,1,5\pmod{8} , whereas pv^{2} can only take values 0,3,4\pmod{8} since both p and q are congruent to 3\pmod{8} . This concludes the case when q does not divide l . If q divides l , we write l=qt , giving q^{4}t^{4}+2q^{3}t^{2}m^{2}-3q^{2}m^{4}=pk^{2}. Then q divides k , so we put k=qh to obtain q^{2}t^{4}+2qt^{2}m^{2}-3m^{4}=ph^{2}. Factorizing the left-hand side, we get (qt^{2}-m^{2})(qt^{2}+3m^{2})=ph^{2}. Since q divides l and \gcd(l,m)=1 , we know that q does not divide m . Then modulo q , the right-hand side is congruent to -3m^{4} , which is not a quadratic residue since q\equiv 2\pmod{3} . Therefore, p is not a quadratic residue modulo q . By quadratic reciprocity, since both p and q are congruent to 3\pmod{4} , q is a quadratic residue modulo p . This implies that p does not divide qt^{2}+3m^{2} because -3q is not a quadratic residue modulo p , and p cannot divide both t and m due to \gcd(l,m)=1 . Hence, p must divide qt^{2}-m^{2} . Observe that the greatest common divisor of the two factors on the left can be either 1 , if one of t and m is odd and the other is even, or a power of 2 , if both are odd. Furthermore, qt^{2}+3m^{2} is always positive. Thus, we have two possibilities: qt^{2}-m^{2}=pu^{2}\quad\text{and}\quad qt^{2}+3m^{2}=v^{2} or qt^{2}-m^{2}=2pu^{2}\quad\text{and}\quad qt^{2}+3m^{2}=2v^{2} for some integers u and v . The second possibility cannot occur because q does not divide m and 2 is not a quadratic residue modulo q . The first possibility also fails because qt^{2}+3m^{2}\equiv 3,6,7\pmod{8} , while v^{2}\equiv 0,1,4\pmod{8} . This concludes the case when q divides l . Therefore, the case r=p does not work. If r=-p , then we get p^{2}l^{4}-2p^{2}ql^{2}m^{2}-3p^{2}q^{2}m^{4}=pn^{2}. Since p^{2} divides the left-hand side, p divides n . Put n=pk to obtain l^{4}-2ql^{2}m^{2}-3q^{2}m^{4}=pk^{2}. Factorizing the left-hand side gives (l^{2}+qm^{2})(l^{2}-3qm^{2})=pk^{2}. We consider two subcases: either q divides l or q does not divide l . If q does not divide l , then the greatest common divisor of the two factors is either 1 , if one of l and m is odd and the other even, or a power of 2 , if both are odd. Moreover, modulo q , the left-hand side is congruent to l^{4} , and q does not divide k , so p is a quadratic residue modulo q . By quadratic reciprocity (since p,q\equiv 3\pmod{4} ), q is a quadratic residue modulo p , so -q is not. Therefore, p must divide the second factor, l^{2}-3qm^{2} , because it cannot divide the first factor. Since l^{2}+qm^{2} is positive, we have two possibilities: l^{2}+qm^{2}=u^{2}\quad\text{and}\quad l^{2}-3qm^{2}=pv^{2} or l^{2}+qm^{2}=2u^{2}\quad\text{and}\quad l^{2}-3qm^{2}=2pv^{2} for some integers u,v . The second option cannot occur because q does not divide l and 2 is not a quadratic residue modulo q . For the first option, solving l^{2}+qm^{2}=u^{2} with \gcd(l,m)=1 , we can parametrize all solutions as \pm l=\alpha^{2}-q\beta^{2},\quad\pm u=\alpha^{2}+q\beta^{2},\quad\pm m=2\alpha\beta if one of \alpha,\beta is even, or \pm l=\frac{\alpha^{2}-q\beta^{2}}{2},\quad\pm u=\frac{\alpha^{2}+q\beta^{2}}{2},\quad\pm m=\alpha\beta if both \alpha and \beta are odd, with \gcd(\alpha,\beta)=1 . Plugging into the second equation, we get l^{2}-3qm^{2}=(\alpha^{2}-q\beta^{2})^{2}-12q\alpha^{2}\beta^{2}=pv^{2}, which can be reorganized as (\alpha^{2}+q\beta^{2})^{2}-16q\alpha^{2}\beta^{2}=pv^{2}. At least one of \alpha or \beta must be odd. If one is even, then (\alpha^{2}+q\beta^{2})^{2}\equiv 1\pmod{8} , so v must be odd, giving pv^{2}\equiv 3\pmod{8} , a contradiction. Thus, both \alpha and \beta are odd. Then (\alpha^{2}+q\beta^{2})^{2}\equiv 16\pmod{128} , and -16q\alpha^{2}\beta^{2}\equiv-48\pmod{128} , so pv^{2}\equiv-32\pmod{128} . This requires v divisible by 8 , making pv^{2} divisible by 64 , while the left-hand side is not. Contradiction. If q divides l , then we write l=qt and k=qK , giving (qt^{2}+m^{2})(qt^{2}-3m^{2})=pK^{2}. Then q\nmid m . The gcd of the two factors is either 1 (if one of t,m is odd and the other even) or a power of 2 (if both are odd). Moreover, modulo q , the left-hand side is -3m^{4} , which is not a quadratic residue since q\equiv 2\pmod{3} , and q\nmid K , so p is not a quadratic residue modulo q . By quadratic reciprocity, q is a quadratic residue modulo p , so -q is not, and p must divide the second factor. We then have qt^{2}+m^{2}=u^{2},\quad qt^{2}-3m^{2}=pv^{2} or qt^{2}+m^{2}=2u^{2},\quad qt^{2}-3m^{2}=2pv^{2}. The second option is impossible since q\nmid m and 2 is not a quadratic residue modulo q . For the first option, at least one of t or m is odd. Checking modulo 4 and 8 leads to contradictions in all subcases (either t even or m even), forcing both t and m to be odd. Then u and v are even. Writing u=2U and v=2V , we obtain 4m^{2}+pv^{2}=u^{2}\quad\implies\quad m^{2}+pV^{2}=U^{2}. Parametrizing this with \gcd(\alpha,\beta)=1 and both \alpha,\beta odd, we get qt^{2}=u^{2}-m^{2}=(\alpha^{2}+p\beta^{2})^{2}-\alpha^{2}\beta^{2}, which is a contradiction modulo 8 . Thus, both r=p and r=-p cases do not work. Therefore, the image \varphi(E_{d}^{\prime}(\mathbb{Q})) contains exactly two elements. Using the kernel-cokernel sequence given in Lemma 2.8 , we conclude that \operatorname{rk}(E_{d}(\mathbb{Q}))=0. ∎ Corollary 2.16 . Let p and q be distinct primes satisfying p,q\equiv 11\pmod{24} . Then there does not exist an elliptic curve E/\mathbb{Q} such that E(\mathbb{Q}(\sqrt{pq}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. 3. Conditional Proofs of our Conjectures In this chapter, we give proofs of Conjectures 1.5 and 1.6 , conditional on the finiteness of the 2 -primary part of the Tate–Shafarevich group \Sha(E/\mathbb{Q}) for rational elliptic curves. We are grateful to Dokchitser [ 10 ] for his notes on the parity conjecture. Before using the results on the parity conjecture to prove our conjectures conditionally, we recall some known results about the parity conjecture. Firstly, we give a brief overview following the introduction of Radziwill and Soundararajan [ 23 ] . Let E be an elliptic curve defined over \mathbb{Q} with conductor N . Write the associated Hasse-Weil L -function as L(E/\mathbb{Q},s)=\sum_{n\geq 1}\frac{a(n)}{n^{s}}, where the coefficients are normalized such that the Hasse bound reads |a(n)|\leq d(n) for all n , and so the center of the critical strip is s=1/2 . Recall that L(E/\mathbb{Q},s) has an analytic continuation to the entire complex plane and satisfies the functional equation \Lambda(E/\mathbb{Q},s)=w(E/\mathbb{Q})\Lambda(E/\mathbb{Q},1-s), where w(E/\mathbb{Q}) , the root number, is \pm 1 , and \Lambda(E/\mathbb{Q},s)=\Big(\frac{\sqrt{N}}{2\pi}\Big)^{s}\Gamma(s+\tfrac{1}{2})L(E/\mathbb{Q},s). Throughout this section, let d be a fundamental discriminant coprime to 2N , and let \chi_{d}=\left(\frac{d}{\cdot}\right) denote the associated primitive quadratic character. Let E_{d} denote the quadratic twist of E by d . The twisted L -function is L(E_{d}/\mathbb{Q},s)=\sum_{n\geq 1}\frac{a(n)\chi_{d}(n)}{n^{s}}. If \gcd(d,N)=1 , then E_{d} has conductor Nd^{2} , and the completed L -function \Lambda(E_{d}/\mathbb{Q},s)=\Big(\frac{\sqrt{N}|d|}{2\pi}\Big)^{s}\Gamma(s+\tfrac{1}{2})L(E_{d}/\mathbb{Q},s) is entire and satisfies the functional equation \Lambda(E_{d}/\mathbb{Q},s)=w(E_{d}/\mathbb{Q})\Lambda(E_{d}/\mathbb{Q},1-s), with w(E_{d}/\mathbb{Q})=\chi_{d}(-N)w(E/\mathbb{Q}). Note that the sign of the functional equation determines the order of vanishing of the L -function at s=1 , i.e., the analytic rank of the elliptic curve. It is a standard result that (-1)^{\operatorname{rk^{an}}(E(\mathbb{Q}))}=w(E/\mathbb{Q}), where \operatorname{rk^{an}} denotes the analytic rank. The parity conjecture for rational elliptic curves states that (-1)^{\operatorname{rk}(E(\mathbb{Q}))}=w(E/\mathbb{Q}), so that \operatorname{rk^{an}}(E(\mathbb{Q}))\equiv\operatorname{rk}(E(\mathbb{Q}))\pmod{2}. This relationship between the analytic rank and quadratic twists allows us to determine when the rank is odd or even in our applications. For the elliptic curves X_{1}(2,10) and X_{1}(2,12) , a standard SageMath calculation shows that their algebraic rank is zero. Under the parity conjecture, this implies that their analytic rank is even; indeed, the LMFDB confirms that their analytic rank is zero. By calculating \chi_{d}(-N) for the discriminants d of interest, we can determine the parity of the algebraic rank conditionally on the parity conjecture. Here, N=20 and N=24 , respectively. We will show that \chi_{d}(-N)=-1 for the d given in Conjectures 1.5 and 1.6 , conditionally implying that the algebraic rank is odd. Previously, we showed that the algebraic rank is either 0 or 1 in these cases. Therefore, we do not need to assume the full Birch and Swinnerton-Dyer conjecture, which asserts equality of the algebraic and analytic ranks. Instead, it suffices to assume the parity conjecture, which asserts equality of the parities of these ranks. Monsky [ 20 ] proved the parity conjecture for all rational elliptic curves, conditional on the finiteness of the 2 -primary part of the Tate–Shafarevich group \Sha(E/\mathbb{Q}) . Dokchitser and Dokchitser [ 9 ] proved the parity conjecture for elliptic curves over number fields E/K , conditional on the finiteness of the 2 - and 3 -parts of \Sha(E/K(E[2])) , where K(E[2]) denotes the extension of K obtained by adjoining the coordinates of the 2 -torsion points of E . In our setting, we do not assume the extra finiteness of the 3 -part, since we can reduce the problem to determining the parity of the rank for rational elliptic curves. Therefore, in our application, we can prove our conjectures conditionally either by assuming the parity conjecture, or the finiteness of the 2 -primary part of \Sha , which implies the parity conjecture for rational elliptic curves. Proposition 3.1 (Conditional on the Parity Conjecture) . Let p\equiv 11,19\pmod{20} be a prime. Let E:y^{2}=x(x^{2}+x-1) be a model for the modular curve X_{1}(2,10) . Then \operatorname{rk}(E_{-p}(\mathbb{Q}))=1. Proof. The conductor of E is N=20 . Since p\equiv 3\pmod{4} , the discriminant of the quadratic field \mathbb{Q}(\sqrt{-p}) is d=-p. We have w(E/\mathbb{Q})=1 , so to show w(E_{-p}/\mathbb{Q})=-1 , it suffices to check that \chi_{-p}(-20)=-1. Recall that the primitive quadratic character is the Kronecker symbol. Since 20=2^{2}\cdot 5 and p is odd, we have \Big(\frac{-p}{-20}\Big)=\Big(\frac{-p}{-5}\Big), and \Big(\frac{-p}{-5}\Big)=-\Big(\frac{-p}{5}\Big), because -p<0 . By assumption, p\equiv\pm 1\pmod{5} , so -p is a quadratic residue modulo 5 . Therefore, \chi_{-p}(-20)=-\Big(\frac{-p}{5}\Big)=-1. This shows that \operatorname{rk^{an}}(E_{-p}(\mathbb{Q})) is odd. Moreover, by Proposition 2.9 , we know that 0\leq\operatorname{rk}(E_{-p}(\mathbb{Q}))\leq 1. If we assume the parity conjecture, or the finiteness of the 2 -primary part of \Sha , which implies the parity conjecture for rational elliptic curves, then for rational elliptic curves, \operatorname{rk^{an}}(E_{-p}(\mathbb{Q}))\equiv\operatorname{rk}(E_{-p}(\mathbb{Q}))\pmod{2}. Since \operatorname{rk^{an}}(E_{-p}(\mathbb{Q})) is odd and the rank is at most 1 , it follows that \operatorname{rk}(E_{-p}(\mathbb{Q}))=1. Combining this with Proposition 2.3 gives the desired result. ∎ Proposition 3.2 (Conditionally on Parity Conjecture) . Let p\equiv 23\pmod{24} be a prime. Let E:y^{2}=x(x^{2}-x+1) be the model for the modular curve X_{1}(2,12) . Then \operatorname{rk}(E_{-p}(\mathbb{Q}))=1. Proof. The conductor of E is N=24 . Since p\equiv 3\pmod{4} , the discriminant of the quadratic field \mathbb{Q}(\sqrt{-p}) is d=-p. We know that w(E/\mathbb{Q})=1 , so in order to show w(E_{-p}/\mathbb{Q})=-1 , it suffices to check that \chi_{-p}(-24)=-1. The primitive quadratic character is the Kronecker symbol, so we calculate \Big(\frac{-p}{-24}\Big)=\Big(\frac{-p}{-6}\Big), and \Big(\frac{-p}{-6}\Big)=-\Big(\frac{-p}{6}\Big), since -p<0 . The assumption p\equiv-1\pmod{3} implies that -p is a quadratic residue modulo 3 , so -\Big(\frac{-p}{6}\Big)=-\Big(\frac{-p}{3}\Big)\Big(\frac{-p}{2}\Big)=-\Big(\frac{-p}{2}\Big). Finally, the assumption p\equiv-1\pmod{8} implies that \Big(\frac{-p}{2}\Big)=1, so \chi_{-p}(-24)=-\Big(\frac{-p}{2}\Big)=-1. This shows that \operatorname{rk^{an}}(E_{-p}(\mathbb{Q})) is odd. Moreover, by Proposition 2.13 , we have 0\leq\operatorname{rk}(E_{-p}(\mathbb{Q}))\leq 1. If we assume the parity conjecture, or the finiteness of the 2 -primary part of \Sha , which implies the parity conjecture for rational elliptic curves, then for rational elliptic curves we must have \operatorname{rk^{an}}(E_{-p}(\mathbb{Q}))\equiv\operatorname{rk}(E_{-p}(\mathbb{Q}))\pmod{2}. This implies \operatorname{rk}(E_{-p}(\mathbb{Q}))=1. Combining this with Proposition 2.4 gives the desired result. ∎ Even though our results are conditional on certain conjectures, we have exhibited explicit families of quadratic fields over which there exist infinitely many rational elliptic curves whose torsion subgroups are isomorphic to \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}\quad\text{or}\quad\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. This, in turn, yields an infinite family of primes p for which the desired torsion subgroups occur over the cyclotomic fields \mathbb{Q}(\zeta_{p}). 4. Eliminating Possible Torsion In this section, we present several methods for eliminating torsion subgroups that cannot be realized. Our main goal is to improve some of the elimination results obtained earlier [ 1 ] . We begin with an important lemma. It is a minor improvement of [ 1 , Lemma 3.1] , and its proof is essentially the same; however, it is necessary for the results in the next section. The only difference is that here we apply the same argument to \operatorname{Gal}(L/K) when considering the L -torsion of E/K , instead of \operatorname{Gal}(K/\mathbb{Q}) when considering the K -torsion of E/\mathbb{Q} . Lemma 4.1 . Let K be a number field and let L/K be a Galois extension. Let E/K be an elliptic curve. Let q be an odd prime power, and suppose E(L)[q]\cong\mathbb{Z}/q\mathbb{Z}. Assume that all elements of \operatorname{Gal}(L/K) have finite order. For any \sigma\in\operatorname{Gal}(L/K) , let \operatorname{ord}(\sigma) denote the order of \sigma . (1) If \gcd(\phi(q),\operatorname{ord}(\sigma))=1 for all \sigma\in\operatorname{Gal}(L/K) , then E(L)[q]=E(K)[q]. (2) If \gcd(\phi(q),\operatorname{ord}(\sigma))\leq 2 for all \sigma\in\operatorname{Gal}(L/K) , then E(L)[q]\cong E_{d}(L)[q]=E_{d}(K)[q] for some d\in K satisfying \sqrt{d}\in L . This lemma is necessary to absolve the result from depending on the degree of a Galois extension and instead state it in terms of the orders of its elements. It is also useful that we generalize our results to elliptic curves E/K over an arbitrary number field K . The lemma is particularly helpful when dealing with cyclotomic extensions of \mathbb{Q} : if n is a positive integer not divisible by 16 , and every odd prime dividing n is of the form 4k+3 , then the extension degree [\mathbb{Q}(\zeta_{n}):\mathbb{Q}] can be divisible by 4 , yet \operatorname{Gal}(\mathbb{Q}(\zeta_{n})/\mathbb{Q}) contains no elements of order 4 . It is also very useful when considering the compositum of quadratic fields. In the next two results, we present direct applications of this lemma. It will also play a key role later in the proof of Theorem 5.3 , and in generalizing some of the results from [ 2 ] . Lemma 4.2 . Let E/\mathbb{Q} be an elliptic curve, and let L be a Galois number field such that \operatorname{Gal}(L/\mathbb{Q}) contains no element of order 3 or 4 . Then E(L)_{\text{tors}}\not\cong\mathbb{Z}/13\mathbb{Z}. Proof. By Lemma 4.1 , if E(L)[13]\cong\mathbb{Z}/13\mathbb{Z} , then there exists a square-free integer d with \sqrt{d}\in L such that E_{d}(\mathbb{Q})[13]\cong\mathbb{Z}/13\mathbb{Z}. But E_{d} is an elliptic curve over \mathbb{Q} , and by Mazur’s theorem it cannot have a torsion point of order 13 . This gives a contradiction. ∎ Corollary 4.3 . Let E/\mathbb{Q} be an elliptic curve, and let n be a positive integer that is not divisible by 16 , and such that every odd prime dividing n is of the form 12k+11 . Then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/13\mathbb{Z}. Proof. Since \operatorname{Gal}(\mathbb{Q}(\zeta_{n})/\mathbb{Q})\cong(\mathbb{Z}/n\mathbb{Z})^{\times} , the assumptions imply that the Galois group contains no element of order 3 or 4 . The result then follows from Lemma 4.2 . ∎ Lemma 4.4 . Let E/\mathbb{Q} be an elliptic curve, and let L be a Galois number field such that \operatorname{Gal}(L/\mathbb{Q}) contains no element of order 4 or 5 . Then E(L)_{\text{tors}}\not\cong\mathbb{Z}/25\mathbb{Z}. Proof. By Lemma 4.1 , if E(L)[25]\cong\mathbb{Z}/25\mathbb{Z} , then there exists a square-free integer d with \sqrt{d}\in L such that E_{d}(\mathbb{Q})[25]\cong\mathbb{Z}/25\mathbb{Z}. But E_{d} is an elliptic curve over \mathbb{Q} , and Mazur’s theorem rules out torsion of order 25 . This is a contradiction. ∎ Corollary 4.5 . Let E/\mathbb{Q} be an elliptic curve, and let n be a positive integer that is not divisible by 16 , and such that every odd prime p dividing n satisfies p\equiv 3\pmod{4} and p-1 is not divisible by 5 . Then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/25\mathbb{Z}. Proof. Since \operatorname{Gal}(\mathbb{Q}(\zeta_{n})/\mathbb{Q})\cong(\mathbb{Z}/n\mathbb{Z})^{\times} , the assumptions guarantee that the Galois group contains no element of order 4 or 5 . The result then follows from Lemma 4.4 . ∎ We now present two important results that are used frequently in the classification of torsion subgroups. Theorem 4.6 (Fricke, Kenku, Klein, Kubert, Ligozat, Mazur, and Ogg, among others) . Let E/\mathbb{Q} be an elliptic curve with an n -isogeny. Then n\leq 19\quad\text{or}\quad n\in\{21,25,27,37,43,67,163\}. Moreover, if E does not have complex multiplication, then n\leq 18\quad\text{or}\quad n\in\{21,25,37\}. Lemma 4.7 (Chou, Lemma 2.7, [ 7 ] ) . Let L be a Galois extension of \mathbb{Q} , and let E be an elliptic curve over \mathbb{Q} . If E(L)_{\text{tors}}\cong\mathbb{Z}/m\mathbb{Z}\times\mathbb{Z}/mn\mathbb{Z}, then E admits an n -isogeny over \mathbb{Q} . Next, we present a theorem that allows us to eliminate certain torsion subgroups when working with cyclotomic extensions \mathbb{Q}(\zeta_{n}) . We treated the case when n is prime [ 1 ] . Building upon that, we can now obtain results for composite n . This is an important step for the classification results presented immediately afterward. Our result can also be generalized to an arbitrary abelian number field L by considering the minimal positive integer n such that L\subset\mathbb{Q}(\zeta_{n}) . The group structures we eliminate are precisely those that can appear as torsion subgroups over arbitrary abelian number fields. The complete list is given in [ 1 , Theorem 2.3] , which follows directly from [ 7 , Theorem 1.2] . Theorem 4.8 . Let E/\mathbb{Q} be an elliptic curve, and let n be a positive integer. Then: (i) If n is not divisible by 3 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/N\mathbb{Z}\text{ for }N=21,27, and E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/3\mathbb{Z}\times\mathbb{Z}/3N\mathbb{Z}\text{ for }N=1,2,3, and E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/6\mathbb{Z}\times\mathbb{Z}/6\mathbb{Z}. (ii) If n is not divisible by 4 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/4\mathbb{Z}\times\mathbb{Z}/4N\mathbb{Z}\text{ for }N=1,2,3,4, and E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/8\mathbb{Z}\times\mathbb{Z}/8\mathbb{Z}. (iii) If n is not divisible by 5 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/N\mathbb{Z}\text{ for }N=15,17,37, and E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/5\mathbb{Z}\times\mathbb{Z}/5\mathbb{Z}. (iv) If n is not divisible by 3 and 5 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/15\mathbb{Z}. (v) If n is not divisible by 7 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/N\mathbb{Z}\text{ for }N=14,37. (vi) If n is not divisible by 7 , and \phi(n) is not divisible by 3 , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/14\mathbb{Z}. (vii) If n is not divisible by the prime \ell\in\{11,17,19,43,67,163\} , then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\not\cong\mathbb{Z}/\ell\mathbb{Z}. Proof. Groups of the form \mathbb{Z}/m\mathbb{Z}\times\mathbb{Z}/mN\mathbb{Z} can be eliminated using the Weil pairing: if an elliptic curve defined over a number field K has full m -torsion over K , then K must contain all m -th roots of unity. This argument covers all cases where m\geq 3 in items (i)–(iii). Groups of the form \mathbb{Z}/N\mathbb{Z} can be eliminated as follows. Let E/\mathbb{Q} be an elliptic curve and K a Galois number field. If E(K)_{\text{tors}}\cong\mathbb{Z}/m\mathbb{Z}\times\mathbb{Z}/mN\mathbb{Z}, then E has a rational N -isogeny by [ 7 , Lemma 2.7] . In the statement of this theorem, the listed N are special because there are only finitely many elliptic curves (finitely many rational j -invariants, up to quadratic twist) with rational N -isogeny for each N . Moreover, the field of definition of the N -isogeny is classified [ 1 , Table 2] . In particular, the x -coordinates of the kernel points of the N -isogeny are defined over these fields, and this does not change for quadratic twists. Therefore, this classification puts a hard bound on which cyclotomic fields \mathbb{Q}(\zeta_{n}) can contain a torsion point of order N . These two arguments cover all group structures in the theorem except \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/14\mathbb{Z} . For this case, the extra condition that \phi(n) is not divisible by 3 implies that the extension degree [\mathbb{Q}(\zeta_{n}):\mathbb{Q}] is not divisible by 3 , so \mathbb{Q}(\zeta_{n}) does not contain a cubic subfield. Since E has full 2 -torsion over \mathbb{Q}(\zeta_{n}) , the field of definition of the 2 -torsion cannot have degree 3 or 6 , so at least one 2 -torsion point is rational. By [ 7 , Lemma 2.7] , this implies that E has a rational 7 -isogeny. Combining this with the rational 2 -torsion point, E would have a rational 14 -isogeny, which in turn forces n to be divisible by 7 , as analyzed in [ 1 ] . This contradicts the assumption. ∎ 5. More Classification Results In this chapter, we study torsion subgroups of elliptic curves over certain metabelian extensions, such as Kummer extensions of cyclotomic fields, building on and refining the classification results [ 1 ] . We then generalize some results on \mathbb{Z}_{p} -extensions of number fields [ 2 ] , showing that torsion over the full \mathbb{Z}_{p} -extension is often already defined over the base field. In this context, our previous results on the impossibility of realizing the torsion subgroups \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} and \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} over quadratic fields imply that these groups also cannot appear over the corresponding metabelian or \mathbb{Z}_{p} -extensions. This connects the explicit elimination results over quadratic fields to broader infinite abelian extensions and illustrates the reach of our methods. Theorem 5.1 . Let E/\mathbb{Q} be an elliptic curve. Define the field L=\mathbb{Q}\Big(\{\mu_{p^{\infty}}:p>11\text{ is a prime satisfying }p\equiv 11\ (\mathrm{mod}\ 12)\text{ and }p\not\equiv 1\ (\mathrm{mod}\ 5)\}\Big). Then E(L)_{\text{tors}} is isomorphic either to one of the groups from Mazur’s theorem or to one of the following: \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z},\quad\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z},\quad\mathbb{Z}/16\mathbb{Z},\quad\text{or}\quad\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/16\mathbb{Z}. Proof. It suffices to prove the statement for all cyclotomic fields \mathbb{Q}(\zeta_{n})\subset L , since E(L)_{\text{tors}} is finite [ 7 , Theorem 1.2] , and the field of definition of this torsion subgroup is an abelian number field, so it must be contained in some \mathbb{Q}(\zeta_{n})\subset L . Let n be a positive integer such that \mathbb{Q}(\zeta_{n})\subset L . Then E(\mathbb{Q}(\zeta_{n}))_{\text{tors}} is contained in the finite list of possible torsion groups given [ 1 , Theorem 2.3] . Using Lemmas 4.3 , 4.5 , and Theorem 4.8 , we can eliminate all groups except the following: E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\simeq\begin{cases}\mathbb{Z}/N\mathbb{Z}&1\leq N\leq 10\text{ or }N=12,16,18,\\ \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2N\mathbb{Z}&1\leq N\leq 6\text{ or }N=8,9.\end{cases} To complete the proof, it remains to eliminate \mathbb{Z}/18\mathbb{Z} and \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2N\mathbb{Z} for N=7,9 . This can be done using [ 1 , Lemma 3.2] , since L has no cubic subfield and therefore has no subfield with degree divisible by 3 . ∎ Remark . We know that \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} is realized, as an example of an elliptic curve with this torsion subgroup over \mathbb{Q}(\sqrt{-59}) is given [ 1 , Section 6] . From our SageMath calculations, we see that \operatorname{rk}(E_{-23}(\mathbb{Q}))=1 . Then, by Proposition 2.4 , we conclude that there exist infinitely many elliptic curves E/\mathbb{Q} with torsion subgroup \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} over \mathbb{Q}(\sqrt{-23}) . Both of these quadratic fields are contained in L , and we have shown that the torsion subgroup cannot be larger. We do not have explicit examples for the groups \mathbb{Z}/16\mathbb{Z} or \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/16\mathbb{Z} , but they cannot be eliminated and are likely to appear. Examples of \mathbb{Z}/16\mathbb{Z} being realized over \mathbb{Q}(\sqrt{d}) are analyzed [ 5 ] , and an explicit example is given [ 1 ] : the elliptic curve 266910.ck6 satisfies E(\mathbb{Q}(\sqrt{41}))_{\text{tors}}\cong\mathbb{Z}/16\mathbb{Z}. Similarly, the group \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/16\mathbb{Z} is known to be realized over \mathbb{Q}(\sqrt{d_{1}},\sqrt{d_{2}}) . An example is given [ 12 ] : the elliptic curve 210.e7 satisfies E(\mathbb{Q}(\sqrt{-7},\sqrt{-15}))_{\text{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/16\mathbb{Z}. Determining whether the specific numbers d,d_{1},d_{2} can be chosen to satisfy the conditions in the statement of this theorem is beyond the scope of this work. Theorem 5.2 . Let E/\mathbb{Q} be an elliptic curve. Define the field L as L=\mathbb{Q}\Big(\{\mu_{p^{\infty}}:p>11\text{ is a prime satisfying }p\equiv 11\ (\mathrm{mod}\ 12)\text{ and }p\not\equiv\pm 1\ (\mathrm{mod}\ 5)\}\Big). Then E(L)_{\text{tors}} is isomorphic to one of the groups from Mazur’s theorem or one of the following groups: \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z},\quad\mathbb{Z}/16\mathbb{Z},\quad\text{and}\quad\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/16\mathbb{Z}. Proof. The proof is almost the same as Theorem 5.1 , but here we also need to eliminate \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} . Suppose E(\mathbb{Q}(\zeta_{n}))_{\text{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} for some positive integer n whose prime factors are all of the form 20k+3 or 20k+7 . By Lemma 4.1 , we can conclude that E_{d}(\mathbb{Q})[5]\cong\mathbb{Z}/5\mathbb{Z} for some square-free integer d with \sqrt{d}\in\mathbb{Q}(\zeta_{n}) . Since E has full 2 -torsion defined over \mathbb{Q}(\zeta_{n}) , the quadratic twist E_{d} also has full 2 -torsion over \mathbb{Q}(\zeta_{n}) . Because \mathbb{Q}(\zeta_{n}) has no cubic subfield, the 2 -torsion of E_{d} must be contained in a quadratic extension. Hence, there exists a quadratic field K such that E_{d}(K)_{\text{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} where K is the field of definition of the 2 -torsion. The splitting field argument allows the possibility K=\mathbb{Q} , but this would contradict Mazur’s theorem because no elliptic curve over \mathbb{Q} has torsion \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} . Thus K must be a quadratic subfield of \mathbb{Q}(\zeta_{n}) , i.e., K=\mathbb{Q}(\sqrt{D}) for some square-free integer D . By Proposition 2.11 , D has a specific form, but Corollary 2.12 then implies that there is no elliptic curve over \mathbb{Q} with E(\mathbb{Q}(\sqrt{D}))_{\text{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}, and therefore such a torsion subgroup cannot occur over \mathbb{Q}(\zeta_{n}) . ∎ Remark . We can see that by reducing the density of the primes in our set from 3/16 to 1/8 , we are able to eliminate one group structure. Remark . We initially wanted to prove a similar theorem by replacing the condition p\not\equiv 4\pmod{5} with p\not\equiv 7\pmod{8} , in order to show that \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} is not realized. This motivated the proof of Proposition 2.16 , which allows us to conclude that \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} is not realized over \mathbb{Q}(\sqrt{d}) for any d\in\{\sqrt{-p},\sqrt{-pq}\} where p,q are primes of the form 24k+11 . However, unlike the \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z} case, the torsion \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z} can still be realized over the biquadratic field \mathbb{Q}(\sqrt{-p},\sqrt{-q}) without being realized over any of its quadratic subfields. As an illustration (not a counterexample for our application, but giving the idea), consider the elliptic curve 30.a6 . This curve has the desired torsion subgroup over three distinct biquadratic fields, yet its torsion over each quadratic subfield remains \mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/6\mathbb{Z} . Theorem 5.3 . Let E/\mathbb{Q} be an elliptic curve, p>7 a prime, and d an integer. Then E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\text{tors}}=E(\mathbb{Q}(\zeta_{p}))_{\text{tors}}. Proof. The proof consists of two parts. In the first part, we show that E cannot have full p -torsion over \mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}) by proving that if it did, then E would have a rational p -torsion point, which contradicts Mazur’s theorem. In the second part, we show that E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\mathrm{tors}}\cong\mathbb{Z}/N\mathbb{Z} or E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2N\mathbb{Z} for some positive integer N such that N\leq 19,\quad\text{or}\quad N\in\{21,25,27,37,43,67,163\}. We then show that all of this torsion must be defined over \mathbb{Q}(\zeta_{p}) . Since \mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}) does not contain \sqrt{-1} , we know that E cannot have full 4 -torsion over this field by the Weil pairing. Assume that E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[\ell]\cong\mathbb{Z}/\ell\mathbb{Z}\times\mathbb{Z}/\ell\mathbb{Z} for some odd prime \ell . Again, by the Weil pairing, our field must contain a primitive \ell -th root of unity, which implies that \ell=p . Next, we show that if p>7 , then E cannot have full p -torsion over this field. From Galois representations, we have the injection \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q})\hookrightarrow\operatorname{GL}_{2}(\mathbb{Z}/p\mathbb{Z}). Combining our assumption with the Weil pairing, we obtain \mathbb{Q}(\zeta_{p})\subset\mathbb{Q}(E[p])\subset\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}). Hence, \mathbb{Z}/(p-1)\mathbb{Z} is a quotient of \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}) , which is in turn a quotient of \operatorname{Gal}(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}})/\mathbb{Q}). This implies that the order of \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}) is (p-1)p^{k} for some k\geq 0 . Since the order of \operatorname{GL}_{2}(\mathbb{Z}/p\mathbb{Z}) is p(p-1)^{2}(p+1) , the order of \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}) must be either p-1 or p(p-1) . Let G=\rho(\operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}))\cong\operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}) be the image of the Galois representation. We use the classification of maximal subgroups of \operatorname{GL}_{2}(\mathbb{Z}/p\mathbb{Z}) due to Serre [ 25 ] , as well as [ 19 , Theorem 7.2] . If the order of G is not divisible by p , then G\cong\mathbb{Z}/(p-1)\mathbb{Z} . Let H be the projective image of G in \operatorname{PGL}_{2}(\mathbb{Z}/p\mathbb{Z}) . Clearly, H is cyclic, so G is contained in a Cartan subgroup. We will show that G must in fact be contained in a split Cartan subgroup. If the order of G is divisible by p , then G is contained in a Borel subgroup of \operatorname{GL}_{2}(\mathbb{Z}/p\mathbb{Z}) . In the first case, G is cyclic of order p-1 , hence abelian. Then, by [ 13 , Theorem 1.1] , we conclude that p\geq 7 is not possible. In the second case, by choosing a suitable basis for the p -torsion, we can write G\leq\Big\{\begin{bmatrix}a&b\\ 0&d\end{bmatrix}:a,b,d\in\mathbb{Z}/p\mathbb{Z},\;a,d\not\equiv 0\pmod{p}\Big\}. If the chosen basis for the p -torsion is given by points P,Q\in E[p] , then the action of \operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}) always sends Q to a multiple of itself. Therefore, \langle Q\rangle\subset E[p] is fixed by \operatorname{Gal}(\overline{\mathbb{Q}}/\mathbb{Q}) , and hence it is the kernel of a rational p -isogeny of E . This implies that E must admit a rational p -isogeny, so by Theorem 4.6 we obtain p\leq 19\quad\text{or}\quad p\in\{37,43,67,163\}. If p\in\{19,43,67,163\} , then E has complex multiplication by Theorem 4.6 . Moreover, for each of these primes, there is a unique rational j -invariant corresponding to a rational p -isogeny [ 1 , Table 2] . Checking their geometric endomorphism rings on LMFDB, we see that they are O_{K}=\mathbb{Z}\Big[\frac{1+\sqrt{-p}}{2}\Big],\quad\text{where}\quad K=\mathbb{Q}(\sqrt{-p}). From CM theory [ 26 , Corollary 5.7] , we know that if we adjoin all the x -coordinates of the torsion points of E to K , we obtain the maximal abelian extension of K . Specifically, in our situation, the Galois group \operatorname{Gal}(\mathbb{Q}(x(E[p]))/\mathbb{Q}(\sqrt{-p})) must be abelian. Observe that this extension has degree p(p-1)/2 , which is odd. We also know that the extension \mathbb{Q}(E[p])/\mathbb{Q}(x(E[p])) is Galois, with degree 1 , 2 , or 4 . It follows that \mathbb{Q}(E[p])=\mathbb{Q}(x(E[p])), so that \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}(\sqrt{-p})) is abelian. However, by [ 13 , Theorem 1.1] , G cannot be abelian if p\geq 7 , so G cannot be isomorphic to C_{p(p-1)} . This implies \mathbb{Q}(E[p])\cap\mathbb{Q}(\mu_{p^{\infty}})=\mathbb{Q}(\zeta_{p}). Consequently, we must have \mathbb{Q}(E[p])=\mathbb{Q}(\zeta_{p},d^{1/p}) for some integer d , so that G\cong C_{p}\rtimes C_{p-1}. Then \operatorname{Gal}(\mathbb{Q}(E[p])/\mathbb{Q}(\sqrt{-p}))\cong C_{p}\rtimes C_{(p-1)/2}, which is non-abelian for any p>3 . Hence, the field of definition of the p -torsion cannot be contained in \mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}) if p\in\{19,43,67,163\} . If p=17 , there are two rational j -invariants corresponding to elliptic curves with a rational 17 -isogeny [ 1 ] . The field of definition of such a rational 17 -isogeny is contained in \mathbb{Q}(\zeta_{85}) , but not in \mathbb{Q}(\zeta_{5}) or \mathbb{Q}(\zeta_{17}) . Moreover, the degree of this extension divides 16 . Therefore, for any field of the form \mathbb{Q}(\mu_{17^{\infty}},d^{1/17^{\infty}}) , the intersection with the field of definition of the 17 -isogeny is contained in \mathbb{Q}(\zeta_{17}) . This follows because the Sylow 17 -subgroup of \operatorname{Gal}(\mathbb{Q}(\mu_{17^{\infty}},d^{1/17^{\infty}})/\mathbb{Q}) is unique, so there is a unique subfield of degree 16 , namely \mathbb{Q}(\zeta_{17}) . In conclusion, the full 17 -torsion cannot be contained in a field of the form \mathbb{Q}(\mu_{17^{\infty}},d^{1/17^{\infty}}) , since not even the kernel of the rational 17 -isogeny is contained there. A similar argument applies for p=37 . There are two rational j -invariants corresponding to elliptic curves with a rational 37 -isogeny. For one of these curves, the field of definition of the 37 -isogeny is \mathbb{Q}(\zeta_{35})^{+} , and it is contained in \mathbb{Q}(\zeta_{1295}) , but not in any \mathbb{Q}(\zeta_{n}) where n is a proper divisor of 1295 . Using the same reasoning as above, we again conclude that the full 37 -torsion cannot be contained in a field of the form \mathbb{Q}(\mu_{37^{\infty}},d^{1/37^{\infty}}) . Note that in [ 1 ] we worked with the field of definition of the x -coordinates of the torsion points. Hence, these fields do not become smaller even if we consider a quadratic twist of the same elliptic curves. If p=11 , the situation is slightly more complicated, since the field of definition of the kernel of the 11 -isogeny is contained in \mathbb{Q}(\zeta_{11}) , so we cannot use the same arguments as in the p=17 or p=37 cases. Fortunately, there are only three different rational j -invariants, so we can use SageMath to compute the fields of definition of their 11 -torsion. The elliptic curve 121.b2 has CM over \mathbb{Z}\Big[\frac{1+\sqrt{-11}}{2}\Big] , so we can apply the same argument as in the p\in\{19,43,67,163\} case. For the other two curves, SageMath shows that the field of definition of the 11 -torsion is of the form \mathbb{Q}(\zeta_{11},d^{1/11}) for some d\in\mathbb{Q}(\zeta_{11}) , but d is not equivalent to a rational number up to 11 -th powers in \mathbb{Q}(\zeta_{11}) . Therefore, this rules out the case p=11 . Finally, if p=13 , the situation becomes more technical. We have shown that the order of G is divisible by 13 , so G is contained in a Borel subgroup and therefore E has a rational 13 -isogeny. We know that there is no rational elliptic curve with CM and a 13 -rational isogeny simultaneously. If E does not have CM, the possible images of the Galois representation are known by the work of Balakrishnan et al. [ 3 , 4 ] . Another useful reference is the work of Rouse et al. [ 24 ] , where they address the problem of determining \ell -adic images of Galois for rational elliptic curves for all primes \ell . Since G cannot be abelian [ 13 ] , it would have to be isomorphic to C_{13}\rtimes C_{12}, but this is not possible by the classification of 13 -adic images of Galois. Hence, we conclude that the torsion subgroup E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\mathrm{tors}} must be one of the forms described previously. In the rest of the proof, we first show that all 2^{\infty} -torsion belongs to \mathbb{Q}(\zeta_{p}) , and then we show that all odd-order torsion belongs to \mathbb{Q}(\zeta_{p}) . Let us show that E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[2^{\infty}]=E(\mathbb{Q}(\zeta_{p}))[2^{\infty}]. Since our field does not contain \sqrt{-1} , we have E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[2^{\infty}]\cong\mathbb{Z}/2^{k}\mathbb{Z}\quad\text{or}\quad E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[2^{\infty}]\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2^{k+1}\mathbb{Z} for some k\geq 0 . By Lemma 4.7 , E has a rational 2^{k} -isogeny, so k\leq 4 by Theorem 4.6 . Therefore, E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[2^{\infty}]\subseteq E[32]. Let L=\mathbb{Q}(E[32])\cap\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}). We will show that the order of \operatorname{Gal}(L/\mathbb{Q}) is not divisible by p . This implies that \operatorname{Gal}(L/\mathbb{Q}) has order dividing p-1 . Moreover, \operatorname{Gal}(L/\mathbb{Q}) is a quotient of \operatorname{Gal}(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}})/\mathbb{Q}), so its kernel must contain the unique Sylow p -subgroup. Hence, L\subset\mathbb{Q}(\zeta_{p}) . To show that \operatorname{Gal}(L/\mathbb{Q}) is not divisible by p , it suffices to show that \operatorname{Gal}(\mathbb{Q}(E[32])/\mathbb{Q}) has order not divisible by p , since \operatorname{Gal}(L/\mathbb{Q}) is a quotient of this group. We know that \operatorname{Gal}(\mathbb{Q}(E[32])/\mathbb{Q})=3\cdot 2^{17}, so p does not divide this order for all p>3 . Therefore, all 2 -primary torsion over \mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}) is defined over \mathbb{Q}(\zeta_{p}) . Now let q be an odd prime power dividing N , where N is the integer appearing in the description of E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\mathrm{tors}}. We will show that E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))[q]=E(\mathbb{Q}(\zeta_{p}))[q]. Observe that E has a rational N -isogeny by Lemma 4.7 , so q\in\{1,3,5,7,11,13,17,19,25,27,37,43,67,163\}. If q=67 , the field of definition of the kernel of the rational 67 -isogeny contains \mathbb{Q}(\zeta_{67})^{+} [ 1 ] , so it is not contained in a field of the form \mathbb{Q}(\mu_{11^{\infty}},d^{1/11^{\infty}}) . In all other cases, p does not divide \phi(q) . Since E(\mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}))_{\mathrm{tors}} is finite, let L be the field of definition of this torsion subgroup. Then L/\mathbb{Q} is a Galois extension. Moreover, \operatorname{Gal}(L/\mathbb{Q}(\zeta_{p})) is the unique Sylow p -subgroup of \operatorname{Gal}(L/\mathbb{Q}) , so it is normal. Hence, L/\mathbb{Q}(\zeta_{p}) is Galois. Finally, the result follows immediately from Lemma 4.1 , taking K=\mathbb{Q}(\zeta_{p}) , since \operatorname{Gal}(L/\mathbb{Q}(\zeta_{p})) is a p -group and the order of any element is a power of p . Since p does not divide \phi(q) , we can apply the first part of the lemma to conclude. ∎ Remark . The problem of whether E can have full p -torsion over the field \mathbb{Q}(\mu_{p^{\infty}},d^{1/p^{\infty}}) is an interesting one. We have shown that this is not possible when p>7 . There are examples of elliptic curves with full p -torsion over \mathbb{Q}(\zeta_{p}) for p=2,3,5 , which settles the problem for these primes. The case p=7 is particularly interesting. Full 7 -torsion cannot be realized over \mathbb{Q}(\zeta_{7}) [ 13 ] , but in principle it could be realized over a field of the form \mathbb{Q}(\mu_{7^{\infty}},d^{1/7^{\infty}}). Indeed, if we allow d to lie in \mathbb{Q}(\zeta_{7}) rather than in \mathbb{Q} , the elliptic curve 26.b2 has full 7 -torsion over a Kummer extension \mathbb{Q}(\zeta_{7},d^{1/7}) for some d\in\mathbb{Q}(\zeta_{7}) , as confirmed by our SageMath calculations. The extension has the same group-theoretical properties whether d is rational or not: it is a degree 42 extension with Galois group isomorphic to C_{7}\rtimes C_{6}. However, the question of whether there exists an elliptic curve E/\mathbb{Q} with full 7 -torsion over \mathbb{Q}(\zeta_{7},d^{1/7}) for some rational d remains open and is beyond the scope of our work. This problem is considerably harder than the p=11 case, because there are only three different rational j -invariants with a rational 11 -isogeny, whereas there are infinitely many different rational j -invariants with a rational 7 -isogeny. Therefore, we cannot check all of them manually as we did in the proof of the p=11 case. Another interesting application of Lemma 4.1 concerns \mathbb{Z}_{p} -extensions. Let K be a number field, and let K_{\mathbb{Z}_{p}} denote the composite of all \mathbb{Z}_{p} -extensions of K . We know that K_{\mathbb{Z}_{p}}/K is abelian, and \operatorname{Gal}(K_{\mathbb{Z}_{p}}/K)\cong\mathbb{Z}_{p}^{a} for some nonnegative integer a=a_{p}(K) satisfying r_{2}+1\leq a_{p}(K)\leq d, where r_{2} denotes the number of pairs of complex conjugate embeddings of K , and d=[K:\mathbb{Q}] is the degree of K over \mathbb{Q} [ 14 , Theorem 2-3] . This allows us to state torsion classification theorems in terms of the composite of all \mathbb{Z}_{p} -extensions of K , providing a natural setting to analyze p -primary torsion over infinite abelian extensions. Theorem 5.4 . Let E/\mathbb{Q} be an elliptic curve. Let K be a number field, and let K_{\mathbb{Z}_{p}} denote the composite of all \mathbb{Z}_{p} -extensions of K for some prime p>5 . Let q be an odd prime power such that E(K_{\mathbb{Z}_{p}})[q]\cong\mathbb{Z}/q\mathbb{Z}. Then E(K_{\mathbb{Z}_{p}})[q]=E(K)[q]. Proof. We know that K_{\mathbb{Z}_{p}}/\mathbb{Q} is a Galois extension. By Lemma 4.7 , it follows that E has a rational q -isogeny. From Theorem 4.6 , we know all possible values of q , and any prime factor of \phi(q) is at most 11 . Assuming p>11 , we ensure that p does not divide \phi(q) . Let L denote the smallest Galois extension of \mathbb{Q} containing both K and \mathbb{Q}(E(K_{\mathbb{Z}_{p}})[q]) . Then L/K is a finite Galois extension, and \operatorname{Gal}(L/K) is a p -group. Applying Lemma 4.1 , we conclude that E(K_{\mathbb{Z}_{p}})[q]=E(K)[q]. In the exceptional cases (p,q)=(7,43) or (11,67) , the theorem does not apply directly. However, the result still holds, since K_{\mathbb{Z}_{p}} does not contain the field \mathbb{Q}(\zeta_{q})^{+} [ 2 , Theorem 1.1, proof] , [ 1 , Section 6] . ∎ As a natural consequence of Theorem 5.4 , we can generalize our results in [ 2 ] . Let K be a quadratic field. If K is real quadratic, then K_{\mathrm{cyc}} is the unique \mathbb{Z}_{p} -extension of K , which is the case we have already treated. Now, let K be an imaginary quadratic field. In this case, there are infinitely many distinct \mathbb{Z}_{p} -extensions of K , all contained in the compositum K_{\mathbb{Z}_{p}}=K_{\mathrm{cyc}}K_{\mathrm{anti}}, satisfying \operatorname{Gal}(K_{\mathbb{Z}_{p}}/K)\cong\mathbb{Z}_{p}^{2}. Instead of considering a single \mathbb{Z}_{p} -extension of K , we may work with the entire field K_{\mathbb{Z}_{p}} and classify torsion over it. The following result can be proven in a manner similar to [ 2 , proof of Theorem 1.1] , but it also follows directly as a consequence of Theorem 5.4 . Theorem 5.5 . Let E/\mathbb{Q} be an elliptic curve. Let K be an imaginary quadratic field, and let p>5 be a prime. Let K_{\mathbb{Z}_{p}}=K_{\mathrm{cyc}}K_{\mathrm{anti}} denote the extension satisfying \operatorname{Gal}(K_{\mathbb{Z}_{p}}/K)\cong\mathbb{Z}_{p}^{2}. Then E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}=E(K)_{\mathrm{tors}}. Proof. Let n be the number of roots of unity in K_{\mathbb{Z}_{p}} . By a similar argument to [ 2 , Lemma 4.7] , we have n=\begin{cases}6&\text{if }K=\mathbb{Q}(\sqrt{-3}),\\ 4&\text{if }K=\mathbb{Q}(\sqrt{-1}),\\ 2&\text{otherwise}.\end{cases} Torsion subgroup E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}} is finite and satisfies E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}\cong\mathbb{Z}/m\mathbb{Z}\times\mathbb{Z}/mN\mathbb{Z}, where m\in\{1,2,3,4,6\} and N is one of the integers appearing in Theorem 4.6 [ 2 , Lemma 4.1] . All 2^{\infty} -torsion and 3^{\infty} -torsion is defined over K [ 2 , Lemma 4.3, 4.4, 4.5] . By Theorem 5.4 , all q -torsion is defined over K for any q that is a power of a prime greater than 3 and divides N . Therefore, all torsion over K_{\mathbb{Z}_{p}} is already defined over K , which proves the theorem. ∎ Finally, we present a corollary that combines our results in [ 2 ] , Theorem 5.5 , and the results in Section 2 . Corollary 5.6 . Let K=\mathbb{Q}(\sqrt{d}) be a quadratic field, and let p>5 be a prime. Let K_{\mathbb{Z}_{p}} denote the composite of all \mathbb{Z}_{p} -extensions of K . Then: (i) If d is a product of -p_{i} for i=1,\dots,n , where p_{i} are distinct primes satisfying p_{i}\equiv 3,7\pmod{20} , then there is no elliptic curve E/\mathbb{Q} with E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/10\mathbb{Z}. (ii) If d is of the form -p_{1} or p_{1}p_{2} , where p_{1},p_{2} are distinct primes satisfying p_{1},p_{2}\equiv 11\pmod{24} , then there is no elliptic curve E/\mathbb{Q} with E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}\cong\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/12\mathbb{Z}. (iii) If d=-p for any prime number p , then there is no elliptic curve E/\mathbb{Q} with E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}\cong\mathbb{Z}/16\mathbb{Z}. Proof. From [ 2 , Theorem 1.1] and Theorem 5.5 , we have E(K_{\mathbb{Z}_{p}})_{\mathrm{tors}}=E(K)_{\mathrm{tors}}. The conclusion then follows from Corollaries 2.12 , 2.14 , 2.16 , and [ 8 , Theorem 5.2] . ∎