Constructing families of 3-Selmer companions Harry Spencer Abstract . Mazur and Rubin introduced the notion of n n -Selmer companion elliptic curves and gave several examples of pairs of non-isogenous Selmer companions. We construct two infinite families of pairs of non-isogenous 3 3 -Selmer companions, parameterised by t ∈ ℤ t\in\mathbb{Z} . [columns=2, intoc] 1. Introduction We recall the following definition introduced by Mazur and Rubin. Definition 1.1 (= [ 5 , definition 1.2] ) . Two elliptic curves {E}_{1} and {E}_{2} over a number field k are n -Selmer companions if (1) \Sel_{n}({E}_{1}^{\chi}/k)\cong\Sel_{n}({E}_{2}^{\chi}/k) for all quadratic characters \chi of k . Several examples of non-isogenous Selmer companions are given in [ 5 ] and—to the best of the author’s knowledge—no others appear in the literature. In this note we show that there are infinitely many non-isogenous pairs of 3 -Selmer companions by explicitly constructing two infinite families of pairs. It is an open problem to find necessary conditions for Selmer companionship, so the techniques and families appearing in this work may be helpful for testing conditions. In particular, Mazur and Rubin speculate that having equal conductors may be a necessary condition, and indeed this holds in our families. Theorem 1.2 (=Theorem 3.8 + Proposition 3.11 ) . For t\in\mathbb{Z} , t\neq 0,1 , the curves (2) \displaystyle E_{t}:\hskip 5.69046pty^{2} \displaystyle= \displaystyle x^{3}+x^{2}+3x+3(8t+1), (3) \displaystyle D_{t}:\hskip 5.69046pty^{2} \displaystyle= \displaystyle x^{3}+(25-81(8t+1))x^{2}-512x are non-isogenous 3 -Selmer companions over every number field. 2. Sufficient conditions Here we recall the main result of Mazur and Rubin’s work, a set of sufficient conditions (stated below for p=3 ) for p -Selmer companionship. Definition 2.1 . Given an elliptic curve {E}/K over a local field of mixed characteristic and its associated Tate parameter q\in K^{\times} with isomorphism \tau_{E/K}:\overline{K}^{\times}/q^{\mathbb{Z}}\to E(\overline{K}) , we define the canonical subgroup \mathcal{C}_{E/K}[p]\subset E[p] to be \tau_{E/K}(\mu_{p}) . Theorem 2.2 (= [ 5 , theorem 3.1 + remark 3.3] ) . Let {E}_{1} and {E}_{2} be elliptic curves over a number field k , and write S_{i} for the set of primes of potentially multiplicative reduction of {E}_{i} . Suppose the following: • there is a G_{k} -isomorphism \alpha:{E}_{1}[3]\xrightarrow{\sim}{E}_{2}[3] ; • S_{1}=S_{2} ; • for every \mathfrak{p} above 3 , \mathfrak{p}\in S_{1}=S_{2} ; • for every \mathfrak{q}\in S_{1}=S_{2} , we have \alpha(\mathcal{C}_{E_{1}/k_{\mathfrak{q}}}[3])=\mathcal{C}_{E_{2}/k_{\mathfrak{q}}}[3] ; • neither {E}_{1} nor {E}_{2} has any prime of additive reduction with Kodaira type one of II, II*, IV, IV*. Then {E}_{1} and {E}_{2} are 3 -Selmer companions over every finite extension of k . 3. Constructing families Associated to a Weierstrass model of an elliptic curve is a discriminant curve . Definition 3.1 . Given an elliptic curve E:y^{2}=f(x) , the discriminant curve of genus 0 or 1 is given by D:\Delta^{2}=\disc_{x}(f(x)-y^{2}). Note, in particular, that the Jacobian J_{D} is usually an elliptic curve. We will also use an alternative construction, as in [ 6 , §2.2] . Definition 3.2 . Given an elliptic curve E:F(X_{1},X_{2},X_{3})=0 given by the vanishing of a cubic homogeneous polynomial in three variables, the corresponding Hessian curve , \Hess(E) is defined by the vanishing of the determinant of the Hessian matrix (\frac{\partial F}{\partial X_{i}\partial X_{j}})_{i,j} . Remark 3.3 . A priori this is not well-defined, but independence of the choice of equation for E is shown in [ 6 , §2.2] . ∎ We verify firstly that these two concepts coincide for elliptic curves given by a Weierstrass equation y^{2}=f(x) . Lemma 3.4 . Let E:y^{2}=f(x) be an elliptic curve. The Jacobian of the discriminant curve is isomorphic to \Hess(E) . Proof. Reduce to the case that E:y^{2}=x^{3}+ax+b is given by a short Weierstrass equation by noting that the isomorphism class of D does not change under a change of variables x^{\prime}=x+\gamma . We may take models J_{D}:\hskip 14.22636pty^{2}=x^{3}-27bx^{2}-27a^{3}x and \Hess(E):\hskip 14.22636pty^{2}=x^{3}-(a^{3}/3+3b^{2})x-(a^{3}b/3+2b^{3}), as in [ 3 , §6] and [ 6 , proposition 2.1] . The isomorphism from \Hess(E)\to J_{D} is then given by (x,y)\mapsto(9(x+b),27y) . ∎ To the best of the author’s knowledge, the following result first appeared as [ 3 , lemma 6.9, remark 6.10] , where it is used in a new proof of the parity conjecture for elliptic curves over number fields. A more conceptual proof of this fact can be seen from the Hessian construction, cf. [ 6 , remark 2.3] . Proposition 3.5 . Let E/k be an elliptic curve over a number field given by Weierstrass equation E:y^{2}=f(x) with discriminant curve of genus 1 . There is a G_{k} -module isomorphism E[3]\cong J_{D}[3] . Proof. The intersection points of E and \Hess(E) are precisely the inflection points of E and \Hess(E) , and hence are the 3 -torsion points of both curves. ∎ This interpretation of the isomorphism E[3]\cong J_{D}[3] allows us to verify the fourth condition of theorem 2.2 . Proposition 3.6 . Consider an elliptic curve over \mathbb{Q}_{p} given by minimal model E:y^{2}=f(x) such that D:\Delta^{2}=\disc_{x}(f(x)-y^{2}) has genus 1 , and suppose that both E and J_{D} have potentially multiplicative reduction at an odd prime p . Then the identification \alpha:E[3]\to J_{D}[3] as in proposition 3.5 satisfies \alpha(\mathcal{C}_{E/\mathbb{Q}_{p}}[3])=\mathcal{C}_{J_{D}/\mathbb{Q}_{p}}[3]. Proof. The points \mathcal{C}_{E/\mathbb{Q}_{p}}[3] are precisely those whose coordinates have non-negative valuation, as with \mathcal{C}_{\Hess(E)/\mathbb{Q}_{p}}[3] . By the proof of proposition 3.5 , these must coincide. The result now follows from lemma 3.4 . ∎ We will also need some further observations relating E and its discriminant curve. Lemma 3.7 . Let E/K:y^{2}=f(x) be an elliptic curve over a local field of mixed characteristic with residue characteristic at least 5 in Weierstrass form with associated discriminant curve D/K of genus 1. We have: (1) If E has potentially multiplicative reduction, then so does J_{D} . (2) If E has multiplicative reduction, then so does J_{D} . Moreover, in this case, E has split multiplicative reduction if and only if J_{D} does also. Proof. Completing the cube does not change the isomorphism class of D , so we can assume f(x)=x^{3}+ax+b so that J_{D}:y^{2}=x^{3}-27bx^{2}-27a^{3}x . The respective discriminants are then \Delta_{E}=-2^{4}\cdot(4a^{3}+27b^{2}),\hskip 14.22636pt\Delta_{D}:=\Delta_{J_{D}}=3^{9}\cdot a^{6}\cdot\Delta_{E}, with j -invariants j(E)=1728\cdot\frac{4a^{3}}{\Delta_{E}},\hskip 14.22636ptj(D):=j(J_{D})=\frac{(2^{4}\cdot 3^{6}\cdot b^{2}+2^{4}\cdot 3^{4}\cdot a^{3})^{3}}{\Delta_{D}}. Suppose E has potentially multiplicative reduction, so that v_{K}(\Delta_{E})>3v_{K}(a) and 3v_{K}(a)=2v_{K}(b) . We have v_{K}(j(D))=6v_{K}(b)-v_{K}(\Delta_{D}) and v_{K}(\Delta_{D})=6v_{K}(a)+v_{K}(\Delta_{E}) . Hence v_{K}(j(D))=v_{K}(j(E))<0 , establishing (1). For (2), the first assertion follows from Tate’s algorithm; in fact, we will show that D must have the same Kodaira type as E in this case. If E has multiplicative reduction, then it has Kodaira type \text{I}_{n} for some n>0 , meaning v_{K}(j(E))=-n and v_{K}(a)=0 . From the above we have v_{K}(j(D))=-n also, so it suffices to check that v_{K}(2^{4}\cdot 3^{3}\cdot(3a^{3}+27b^{2}))=0 . This follows by applying the ultrametric inequality to the triple 4a^{3}+27b^{2} , a^{3} , 3a^{3}+27b^{2} . The second assertion holds because E has split multiplicative reduction precisely when -6b is a square. Likewise, D has split multiplicative reduction precisely when 2^{3}\cdot 3^{8}\cdot b\cdot(4a^{3}+24b^{2}) is a square. Comparing, we have \displaystyle 2^{3}\cdot 3^{8}\cdot b\cdot(4a^{3}+24b^{2}) \displaystyle\equiv 2b(4a^{3}+24b^{2}) \displaystyle\pmod{(K^{\times})^{2}} \displaystyle\equiv 2b((4a^{3}+27b^{2})-3b^{2}) \displaystyle\pmod{(K^{\times})^{2}} and the conclusion follows from the fact that v_{K}((4a^{3}+27b^{2})-3b^{2})=2v_{K}(b) and ((4a^{3}+27b^{2})-3b^{2})/b^{2}\equiv-3\pmod{\pi} , where \pi is a uniformiser, so that (4a^{3}+27b^{2})-3b^{2} is a square in K if and only if -3 is a square. ∎ Now we are able to show: Theorem 3.8 . For all t\in\mathbb{Z} the curve {E}_{t}:\hskip 5.69046pty^{2}=x^{3}+x^{2}+3x+3(8t+1) and the Jacobian of its associated discriminant curve J_{D_{t}}:\hskip 5.69046pty^{2}=x^{3}+(25-81(8t+1))x^{2}-512x are 3 -Selmer companions over every number field. Proof. Fix t\in\mathbb{Z} . For simplicity of notation we will write E for E_{t} and D for J_{D_{t}} . We compute that the discriminants are ( \dagger ) \Delta_{E}=-2^{4}\cdot(27\cdot(3(8t+1))^{2}-50\cdot(3(8t+1))+99),\hskip 2.84544pt\Delta_{D}=-2^{18}\cdot 3^{3}\cdot\Delta_{E} with j -invariants j(E)=-\frac{2^{21}}{\Delta_{E}},\hskip 5.69046ptj(D)=-\frac{(27\Delta_{E}+2^{13})^{3}}{\Delta_{D}}. One can now read off that there are no primes for which D has bad reduction but E has good reduction, and that E (and therefore D ) has potentially multiplicative reduction at all primes p>2 of bad reduction. For p=2 , we can verify from the j -invariants that both curves have potentially good reduction and, using Tate’s algorithm (e.g. [ 7 , IV.9] ), we can check that we avoid the Kodaira types excluded by theorem 2.2 . For p=3 , we can read off that both curves have multiplicative reduction. The conclusion now follows from lemma 3.7 and proposition 3.6 . ∎ Using exactly the same proof one can also show, for example: Theorem 3.9 . For all t\in\mathbb{Z} the curve {E}_{t}:\hskip 5.69046pty^{2}=x^{3}+x^{2}+3x+3(16t+13) and the Jacobian of its associated discriminant curve J_{D_{t}}:\hskip 5.69046pty^{2}=x^{3}+(25-81(16t+13))x^{2}-512x are 3 -Selmer companions over every number field. It remains to verify that the families of theorems 3.8 and 3.9 do not consist of isogenous pairs. To do so we will need a well-known theorem of Mazur and Kenku. Write Y_{0}(N) for the (non-compact) modular curve defined by the N^{\text{th}} classical modular polynomial, \Phi_{N} . Theorem 3.10 (= [ 4 , theorem 1] ) . Y_{0}(N)(\mathbb{Q})=\emptyset , unless either N\leq 19 or N\in\{21,25,27,37,43,67,163\} . Recall that there is a degree N cyclic isogeny between two elliptic curves E and D over \overline{\mathbb{Q}} if and only if \Phi_{N}(j(E),j(D))=0 or, equivalently, (j(E),j(D))\in Y_{0}(N)(\overline{\mathbb{Q}}) . Proposition 3.11 . For t\in\mathbb{Z} , t\neq 0,1 , the curves E_{t} and J_{D_{t}} of theorem 3.8 are not \overline{\mathbb{Q}} -isogenous. Proof. Recall from ( † ‣ 3 ) that \Delta_{E_{t}} and \Delta_{J_{D_{t}}} differ by a non-square independent of t , so the cubics x^{3}+x^{2}+3x+3(8t+1) and x^{3}+(25-81(8t+1))x^{2}-512x have different splitting fields (hence the curves have different 2 -torsion fields) for all values of t . Thereby our elliptic curves can never be isogenous via an isogeny of odd degree, as this would give an isomorphism on 2 -torsion. Now we need only check for isogenies of even degrees which appear in theorem 3.10 . To do so, we compute \Phi_{N}(j_{1},j_{2}) as a rational function in t for each of these values of N and compute the rational roots of its numerator. We summarise these computations in table 1 . Note that the relevant modular polynomials are stored in Magma [ 1 ] , which we use to carry out these computations, except for \Phi_{18} , which is computed by the authors of [ 2 ] and is available at https://math.mit.edu/˜drew/ClassicalModPolys.html . ∎ Also included in table 1 are the same computations for the family appearing in theorem 3.9 , noting that the j -invariants of this family are rational transformations of those of the family appearing in theorem 3.8 . Correspondingly, we obtain: Proposition 3.12 . For all t\in\mathbb{Z} , the curves E_{t} and J_{D_{t}} of theorem 3.9 are not \overline{\mathbb{Q}} -isogenous. I thank Vladimir Dokchitser for his supervision and comments on an earlier draft, Dominik Bullach for suggesting that I look into Selmer companions, and Alexandros Konstantinou for pointing me towards the reference [ 6 ] . This work was supported by the Engineering and Physical Sciences Research Council [EP/S021590/1], the EPSRC Centre for Doctoral Training in Geometry and Number Theory (The London School of Geometry and Number Theory), University College, London. Appendix A In the below j_{1}=j_{1}(t) is the j -invariant of E_{t}:y^{2}=x^{3}+x^{2}+3x+3(8t+1) and j_{2}=j_{2}(t) is that of J_{D_{t}}:y^{2}=x^{3}+(25-81(8t+1))x^{2}-512x . Likewise j_{3} and j_{4} are the j -invariants of y^{2}=x^{3}+x^{2}+3x+3(16t+13) and y^{2}=x^{3}+(25-81(16t+13))x^{2}-512x , respectively. N Rational zeroes of \Phi_{N}(j_{1},j_{2}) Rational zeroes of \Phi_{N}(j_{3},j_{4}) 2 -7/81 -257/324 4 -14/81,0 -271/324,-3/4 6 — — 8 — — 10 -95/81,-7/81,1 -433/324,-257/324,-1/4 12 — — 14 — — 16 — — 18 — — Table 1 . Rational zeroes of \Phi_{N}(j_{1},j_{2}) , \Phi_{N}(j_{3},j_{4}) for relevant N