Introduction
Elliptic curves are among the important objects of study in number theory. To understand them, we study various associated quantities, such as their groups of rational points, the primes of good reduction, and their associated Galois representations, to name a few. The last quantity, in particular, provides a lot of information about the original curve. For example, the Neron-Ogg-Shafarevich theorem states that the primes at which the Galois representation is unramified correspond to the primes of good reduction for the elliptic curve. Understanding the image of this representation is therefore an equally useful objective. One of the strongest theorems along these lines is Serre’s open image theorem, which we now describe.
Let E E E be an elliptic curve without complex multiplication over a number field K K K . Let G K : = Gal ( K ‾ / K ) G_K := \operatorname{Gal}(\overline{K}/K) G K := Gal ( K / K ) denote the absolute Galois group of K K K . A celebrated theorem of Serre ([26 Théorème 2, p. 293] ) states that the associated Galois representation
ρ E , ℓ : G K → Aut ( E [ ℓ ] ) ≅ GL 2 ( F ℓ ) \rho_{E,\ell}: G_K \to\operatorname{Aut}(E[\ell]) \cong\operatorname{GL}_2(\mathbb{F}_\ell) ρ E , ℓ : G K → Aut ( E [ ℓ ]) ≅ GL 2 ( F ℓ )
is surjective for all but finitely many primes ℓ \ell ℓ . Therefore, there exists a smallest positive integer c ( E ) c(E) c ( E ) , depending on E E E , such that for all primes ℓ > c ( E ) \ell> c(E) ℓ > c ( E ) , the map ρ E , ℓ \rho_{E,\ell} ρ E , ℓ is surjective. Serre managed to prove that c ( E ) = 11 c(E) = 11 c ( E ) = 11 would work for semi-stable elliptic curves over Q \mathbb{Q} Q [26 Corollaire 2.,p 309] .
2020 Mathematics Subject Classification. Primary: 11G05, Secondary: 14H52.
Key words and phrases. Elliptic Curves, Galois Representations, Serre’s Open Image Theorem.
The first author’s research has been supported by the NBHM PhD Scholarship. The second author is supported by ANRF grant ANRF/ARGM/2025/000421/MTR..
In 1993, Masser and Wustholtz [19 ] gave a bound c ( E ) < c 2 ( max ( d , h ) ) γ c(E)<c_2(\max(d,h))^\gamma c ( E ) < c 2 ( max ( d , h ) ) γ where d d d denotes the degree of the number field K K K , h h h denotes the Weil height of the j j j -invariant of E E E and c , γ c,\gamma c , γ are absolute constants. One can also look at the representation on the Tate module T ℓ ( E ) T_\ell(E) T ℓ ( E ) of E E E , that is the map ρ E , ℓ : G K → Aut ( T ℓ ( E ) ) ≅ GL 2 ( Z ℓ ) \rho_{E,\ell}:G_K\to\operatorname{Aut}(T_\ell(E))\cong\operatorname{GL}_2(\mathbb{Z}_\ell) ρ E , ℓ : G K → Aut ( T ℓ ( E )) ≅ GL 2 ( Z ℓ ) . We may further collect all the ρ E , ℓ \rho_{E,\ell} ρ E , ℓ into one representation, namely, ρ ∞ : G K → GL 2 ( Z ^ ) \rho_\infty:G_K\to\operatorname{GL}_2(\widehat{\mathbb{Z}}) ρ ∞ : G K → GL 2 ( Z ) . An equivalent form of Serre’s theorem says that the index of ρ ∞ ( G K ) \rho_\infty(G_K) ρ ∞ ( G K ) in GL 2 ( Z ^ ) \operatorname{GL}_2(\widehat{\mathbb{Z}}) GL 2 ( Z ) is bounded by a constant that depends only on E E E and K K K . Towards this, Davide Lombardo [18 ] gives an unconditional bound on the index of the adelic image of the Galois representation.
Assuming the Generalized Riemann hypothesis (GRH), allows one to give an explicit version of the Chebotarev density theorem which, in turn, helps us give better bounds for c ( E ) c(E) c ( E ) . For example, Serre showed that for elliptic curves over the rationals [27 Théorème 22, p. 196] one has
c ( E ) ≤ C 4 ( log N E ) ( log log 2 N E ) 3 c(E) \leq C_4(\log N_E)(\log\log2N_E)^3 c ( E ) ≤ C 4 ( log N E ) ( log log 2 N E ) 3
In 2014, Larson and Vaintrob [15 ] show that c ( E ) ≤ C 3 log N E c(E)\leq C_3\log N_E c ( E ) ≤ C 3 log N E , although the constant C 3 C_3 C 3 is not effectively computable.
In 2023, Mayle and Wang [20 ] give an upper bound for c ( E ) c(E) c ( E ) for non-CM elliptic curves over Q \mathbb{Q} Q . They use an effective version of Chebotarev’s density theorem by Bach and Sorensen [1 ] . This version of the explicit Chebotarev also holds for number fields. Using a better quotient, Chen and Swidinsky [6 ] have been able to improve the bound given by Mayle and Wang.
For number fields, the situation remains largely unknown. For example, Mazur’s cyclic isogeny theorem [21 Theorem 1] completely classifies the primes for which an elliptic curve can have a cyclic isogeny of that degree. This in turn allows one to rule out the cases for which ρ E , Q ( G Q ) \rho_{E,\mathbb{Q}}(G_{\mathbb{Q}}) ρ E , Q ( G Q ) is contained in a Borel subgroup (see Theorem 2.2). However, for quadratic fields (see Banwait [3 ] ), the cyclic isogenies have been determined. In fact, an algorithm is provided that takes a quadratic field as input and outputs the cyclic isogeny primes. For higher-degree fields, the situation is unknown.
Main Results
We extend the results of Mayle and Wang to Galois number fields. The main challenge with number fields is to keep track of the various prime ideals lying over a given rational prime. This leads to subtle differences in the statements of the theorems compared with Mayle and Wang. Furthermore, many of the constants associated with elliptic curves now turn out to be ideals. For example, the conductor of an elliptic curve is replaced by the conductor ideal. This still tracks the prime ideals of bad reduction. Our main focus will be to deal with the case in which the image is contained in the normalizer of a non-split Cartan subgroup but not fully contained in said non-split Cartan subgroup (see Theorem 2.1, case (v)).
Theorem A (Theorem 5.4 ). Let E E E be a non-CM elliptic curve, with conductor ideal N E \mathfrak{N}_E N E , over a Galois number field K K K . For any prime number ℓ \ell ℓ such that,
ℓ > 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) , \ell> 2C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_E)+2C_2(K), ℓ > 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) ,
the image of the Galois representation ρ E , ℓ \rho_{E,\ell} ρ E , ℓ is not contained in the normalizer of a non-split Cartan subgroup, where
C 1 ( K ) = 510.352 + 513.024 n K , C_1(K) = 510.352 + 513.024n_K, C 1 ( K ) = 510.352 + 513.024 n K ,
C 2 ( K ) = 513.024 ( 1 + n K 4 log d K + n K log ( 384 n K ) ) + 1324.8 n K + 45.7 , C_2(K) = 513.024\left(1 + \frac{n_K}{4}\log d_K + n_K\log(384n_K)\right) + 1324.8n_K + 45.7, C 2 ( K ) = 513.024 ( 1 + 4 n K log d K + n K log ( 384 n K ) ) + 1324.8 n K + 45.7 ,
and d K d_K d K denotes the absolute discriminant of the number field K K K .
The proof of this theorem allows us to give a bound for c ( E , K ) c(E,K) c ( E , K ) , where K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) is a totally real number field with ∣ D ∣ < 50 \lvert D\rvert< 50 ∣ D ∣ < 50 . We assume that K K K is quadratic to apply Banwait’s classification of cyclic isogenies [3 ] . This, in turn, allows us to bound the prime numbers for which the image is contained in a Borel subgroup. This is also where we get the restriction of ∣ D ∣ < 50 \lvert D\rvert< 50 ∣ D ∣ < 50 . Similarly, we assume that K K K is totally real to restrict the primes for which the image lies in a non-split Cartan subgroup. Note that the classification of maximal subgroups (Theorem 2.1) makes a distinction between being contained in a non-split Cartan and being contained in the normalizer of the non-split Cartan (see Theorem 2.1 (iii) and (v)) We assume GRH to provide an effective version of Faltings Isogeny Theorem (which in turn uses an effective version of the Chebotarev density theorem from [1 ] , which is where the assumption is needed) Putting it all together, we have the following:
Corollary A. Let K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) be a formally real quadratic number field with ∣ D ∣ < 50 \lvert D\rvert< 50 ∣ D ∣ < 50 and let E E E be an elliptic curve over K K K without complex multiplication. Let ρ E , ℓ : G K → GL 2 ( F ℓ ) \rho_{E,\ell}: G_K \to\operatorname{GL}_2(\mathbb{F}_{\ell}) ρ E , ℓ : G K → GL 2 ( F ℓ ) be the associated Galois representation. Assume GRH and suppose the image of ρ E , ℓ \rho_{E,\ell} ρ E , ℓ is not contained in the normalizer of a split Cartan subgroup of GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) . Let c ( E , K ) c(E,K) c ( E , K ) denote the largest prime for which the image of the Galois representation fails to be surjective. Then
c ( E , K ) ≤ 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) , c(E,K) \le2C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K N_E) + 2C_2(K), c ( E , K ) ≤ 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) , where the constants C 1 ( K ) C_1(K) C 1 ( K ) and C 2 ( K ) C_2(K) C 2 ( K ) are the same as in Theorem A.
For a prime ideal p \mathfrak{p} p of K K K , we denote by a p ( E ) a_{\mathfrak{p}}(E) a p ( E ) the constant
a p ( E ) = N ( p ) + 1 − # E ( F ( p ) ) , a_{\mathfrak{p}}(E) = N(\mathfrak{p}) + 1 - \#E(\mathbb{F}(\mathfrak{p})), a p ( E ) = N ( p ) + 1 − # E ( F ( p )) ,
where N ( p ) N(\mathfrak{p}) N ( p ) denotes the norm of the ideal p \mathfrak{p} p , F ( p ) \mathbb{F}(\mathfrak{p}) F ( p ) denotes the residue field at p \mathfrak{p} p and # E ( F ( p ) ) \#E(\mathbb{F}(\mathfrak{p})) # E ( F ( p )) denotes the number of point of the elliptic curve E E E in the reduction modulo p \mathfrak{p} p .
To prove Theorem A, we prove an effective version of Faltings isogeny theorem over number fields. This theorem is central to all that follows. To adapt this result from Mayle and Wang [20 Theorem 2, p. 1401] to number fields, we first prove a series of number theoretic lemmas in §4.2. This allows us to construct an extension in which a given collection of prime ideals ramifies. We can use an extension to find a prime ideal of good reduction by ensuring that all the bad-reduction primes ramify. Combining this with the effective Chebotarev density theorem by Bach and Sorenson [1 ] and the generalized discriminant bound (Lemma 3.5 ), we are able to prove the following:
Theorem B (Theorem 4.4 ). Assume GRH. Let K K K be a Galois number field, and let E 1 , E 2 / K E_1, E_2/K E 1 , E 2 / K be elliptic curves without complex multiplication. Suppose that E 1 E_1 E 1 and E 2 E_2 E 2 are not K K K -isogenous. Then there exists
a prime ideal p ⊆ O K \mathfrak{p} \subseteq\mathcal{O}_K p ⊆ O K of good reduction for both E 1 E_1 E 1 and E 2 E_2 E 2 such that
a p ( E 1 ) ≠ a p ( E 2 ) , a_{\mathfrak{p}}(E_1) \ne a_{\mathfrak{p}}(E_2), a p ( E 1 ) = a p ( E 2 ) ,
and satisfying
N K / Q p ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 N E 2 ) + C 2 ( K ) ) 2 , N_{K/\mathbb{Q}}\mathfrak{p} \le\left(C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}\left(2\mathcal{O}_K\mathfrak{N}_{E_1}\mathfrak{N}_{E_2}\right)+C_2(K)\right)^2, N K / Q p ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 N E 2 ) + C 2 ( K ) ) 2 ,
where N E i \mathfrak{N}_{E_i} N E i is the conductor ideal of E i E_i E i , and C 1 ( K ) , C 2 ( K ) C_1(K), C_2(K) C 1 ( K ) , C 2 ( K ) are constants depending only on K K K .
Related Work
Le Fourn [16 ] proves a uniform surjectivity result for non-CM Q \mathbb{Q} Q -curves that are defined over an imaginary quadratic field.
Lemos [17 ] shows the existence of a constant C K , d C_{K,d} C K , d for which the surjectivity of the representation ρ ‾ E , p \overline{\rho}_{E,p} ρ E , p is established in various cases, where E E E is a Q \mathbb{Q} Q -curve defined over a quadratic field K K K . The constant is not made explicit, except in [17 Theorem 1.10] .
Outline of the Paper
The outline of the proof that we follow is due to Mayle and Wang [20 ] .
In section 2, we introduce notations and describe the classification of maximal subgroups for GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) . Using this classification, for each maximal subgroup H H H , we describe conditions for a prime number ℓ \ell ℓ under which the image of the galois representation G ℓ G_{\ell} G ℓ is not contained in H H H . Our main focus of this paper is to give a condition on ℓ \ell ℓ for which G ℓ G_{\ell} G ℓ is not contained in the normalizer of a non-split Cartan subgroup. Towards this end, Section 3 provides bounds for the norm of a prime ideal p \mathfrak{p} p . This step uses an effective version of the Chebotarev density theorem established by Bach and Sorenson [1 ] . It is here that we need to assume the Generalized Riemann hypothesis.
Section 4 establishes an effective version of Faltings Isogeny Theorem for number fields. That is, given two elliptic curves E E E and E ′ E' E ′ over a number field K K K that are not K K K -isogenous, we show that there exists a prime ideal p \mathfrak{p} p for which a p ( E ) ≠ a p ( E ′ ) a_{\mathfrak{p}}(E) \ne a_{\mathfrak{p}}(E') a p ( E ) = a p ( E ′ ) . In fact, by using the bounds established in Section 3, we are able to provide a bound on the norm of the prime ideal p \mathfrak{p} p thus obtained.
Finally, section 5 completes the proof of the main theorem. We first assume that the image G ℓ G_{\ell} G ℓ , of an elliptic curve E E E over K K K , is contained in the normalizer of the non-split Cartan subgroup but not completely contained in the non-split Cartan subgroup. From this, we construct a quadratic character χ D \chi_D χ D . We twist the elliptic curve E E E with respect to this character to get another elliptic curve E D E_D E D . After this, we show that these elliptic curves are non-isogenous over K K K and hence, by the effective version of Faltings isogeny theorem established in Section 4, we obtain a prime ideal p \mathfrak{p} p for which a p ( E ) ≠ a p ( E D ) a_{\mathfrak{p}}(E) \ne a_{\mathfrak{p}}(E_D) a p ( E ) = a p ( E D ) and whose norm satisfies a certain bound. Finally, by applying the Hasse bound, we obtain the desired bound for the prime ℓ \ell ℓ .
AI Disclosure
In preparing this paper, ChatGPT and DeepSeek were used to proofread and to help with finding references for some results. The authors take full responsibility for the contents presented within this paper.
Preliminaries
In this section, we describe the conditions on the field K K K and on the prime number ℓ \ell ℓ for which the image fails to be surjective. For this, we appeal to the classification of subgroups of GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) , which dates back to Dickson [9 ] (See Theorem 2.1) Using this, we do a case-by-case analysis by assuming the image is contained in the various subgroups and obtaining a bound on the prime ℓ \ell ℓ . Combining all these results together with Theorem A, we get Corollary A.
Notations
We use the same notations as in Mayle and Wang [20 §2, p.1402] . Let ℓ \ell ℓ be an odd prime. For ε ∈ F ℓ \varepsilon\in\mathbb{F}_{\ell} ε ∈ F ℓ , let ( ε ℓ ) \left(\frac{\varepsilon}{\ell}\right) ( ℓ ε ) denote the Legendre symbol. Let
C s ( ℓ ) : = { ( a 0 0 b ) : a , b ∈ F ℓ , a b ≠ 0 } C_s(\ell) := \left\{ \begin{pmatrix} a & 0 \\ 0 & b \end{pmatrix} : a,b \in\mathbb{F}_{\ell},\ ab \ne0 \right\} C s ( ℓ ) := { ( a 0 0 b ) : a , b ∈ F ℓ , ab = 0 }
and
C n s ( ℓ ) : = { ( a ε b b a ) ∈ GL 2 ( F ℓ ) : a , b , ε ∈ F ℓ , ( ε ℓ ) ≠ 1 } , C_{ns}(\ell) := \left\{ \begin{pmatrix} a & \varepsilon b \\ b & a \end{pmatrix} \in\operatorname{GL}_2(\mathbb{F}_{\ell}) : a,b,\varepsilon\in\mathbb{F}_{\ell},\ \left(\frac{\varepsilon}{\ell}\right) \ne1 \right\}, C n s ( ℓ ) := { ( a b ε b a ) ∈ GL 2 ( F ℓ ) : a , b , ε ∈ F ℓ , ( ℓ ε ) = 1 } ,
denote the standard split and non-split Cartan subgroups, respectively. By ( ε ℓ ) ≠ 1 \left(\frac{\varepsilon}{\ell}\right) \ne1 ( ℓ ε ) = 1 , we mean that ε \varepsilon ε is a quadratic non-residue in F ℓ \mathbb{F}_{\ell} F ℓ . Let C s + ( ℓ ) C_s^+(\ell) C s + ( ℓ ) and C n s + ( ℓ ) C_{ns}^+(\ell) C n s + ( ℓ ) denote the normalizers in GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) of the split Cartan and non-split Cartan subgroups, respectively. Let
B ( ℓ ) : = { ( a b 0 d ) : a , b , d ∈ F ℓ , a d ≠ 0 } , B(\ell) := \left\{ \begin{pmatrix} a & b \\ 0 & d \end{pmatrix} : a,b,d \in\mathbb{F}_{\ell},\ ad \ne0 \right\}, B ( ℓ ) := { ( a 0 b d ) : a , b , d ∈ F ℓ , a d = 0 } ,
denote the standard Borel subgroup in GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) . Let G ‾ \overline{G} G denote the image of a subgroup G G G of GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) under the natural surjection q : GL 2 ( F ℓ ) → PGL 2 ( F ℓ ) q : \operatorname{GL}_2(\mathbb{F}_{\ell}) \to\operatorname{PGL}_2(\mathbb{F}_{\ell}) q : GL 2 ( F ℓ ) → PGL 2 ( F ℓ ) . Let A n A_n A n (resp. S n S_n S n ) denote the alternating (resp. symmetric) groups in n n n letters. We then have the following classification due to Dickson:
Theorem 2.1 (Dickson). Let ℓ \ell ℓ be an odd prime. If G ⊆ GL 2 ( F ℓ ) G \subseteq\operatorname{GL}_2(\mathbb{F}_{\ell}) G ⊆ GL 2 ( F ℓ ) is a subgroup, then either
(i) G G G contains SL 2 ( F ℓ ) \operatorname{SL}_2(\mathbb{F}_{\ell}) SL 2 ( F ℓ ) ,
(ii) G G G is conjugate to a subgroup of B ( ℓ ) B(\ell) B ( ℓ ) ,
(iii) G G G is conjugate to a subgroup of C n s ( ℓ ) C_{ns}(\ell) C n s ( ℓ ) ,
(iv) G G G is conjugate to a subgroup of C s + ( ℓ ) C_s^+(\ell) C s + ( ℓ ) but not to any subgroup of C s ( ℓ ) C_s(\ell) C s ( ℓ ) ,
(v) G G G is conjugate to a subgroup of C n s + ( ℓ ) C_{ns}^+(\ell) C n s + ( ℓ ) but not to any subgroup of C n s ( ℓ ) C_{ns}(\ell) C n s ( ℓ ) , or
(vi) G ‾ \overline{G} G is isomorphic to A 4 A_4 A 4 , S 4 S_4 S 4 or A 5 A_5 A 5 .
Proof. The reader may refer to section 2 of Serre’s original paper [26 §2] for a proof of this result. □
Let G ℓ G_{\ell} G ℓ be the image of the Galois representation ρ E , ℓ : G K → GL 2 ( F ℓ ) \rho_{E,\ell} : G_K \to\operatorname{GL}_2(\mathbb{F}_{\ell}) ρ E , ℓ : G K → GL 2 ( F ℓ ) . Our strategy is to describe constraints on primes for which the image G ℓ G_{\ell} G ℓ is contained in the various maximal subgroups appearing in Theorem 2.1. That is, we provide a bound N N N such that, for primes ℓ > N \ell> N ℓ > N , the subgroup G ℓ G_{\ell} G ℓ will not be contained in any maximal subgroup. Taking the maximum of each such bound will give us an estimate for c ( E ) c(E) c ( E ) in Serre’s Open Image Theorem.
G contains SL2(F)
Suppose that the image of the Galois representation contains SL 2 ( F ℓ ) \operatorname{SL}_2(\mathbb{F}_{\ell}) SL 2 ( F ℓ ) . We have the following:
Proposition 2.1. Suppose that ℓ \ell ℓ is not ramified in K K K . If G ℓ G_{\ell} G ℓ contains SL 2 ( F ℓ ) \operatorname{SL}_2(\mathbb{F}_{\ell}) SL 2 ( F ℓ ) , then G ℓ G_{\ell} G ℓ is GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) , i.e. ρ E , ℓ \rho_{E,\ell} ρ E , ℓ is surjective.
Proof. Using the Weil pairing on E E E , the composition
det ρ E , ℓ : G K → F ℓ × \det\rho_{E,\ell}:G_K\to\mathbb{F}_{\ell}^{\times} det ρ E , ℓ : G K → F ℓ ×
is the mod ℓ \ell ℓ cyclotomic character, by a similar calculation as in the proof of [30 Proposition 8.6, p. 99] , and is surjective when ℓ \ell ℓ is unramified in K / Q K/\mathbb{Q} K / Q . Suppose that SL 2 ( F ℓ ) ⊆ G ℓ \operatorname{SL}_{2}(\mathbb{F}_{\ell})\subseteq G_{\ell} SL 2 ( F ℓ ) ⊆ G ℓ . The image of SL 2 ( F ℓ ) \operatorname{SL}_{2}(\mathbb{F}_{\ell}) SL 2 ( F ℓ ) under the determinant map is 1 1 1 , but by surjectivity of the cyclotomic character we see that SL 2 ( F ℓ ) ⊊ G ℓ \operatorname{SL}_{2}(\mathbb{F}_{\ell})\subsetneq G_{\ell} SL 2 ( F ℓ ) ⊊ G ℓ . Therefore, by Theorem 2.1, we will have G ℓ = GL 2 ( F ℓ ) G_{\ell}=\operatorname{GL}_{2}(\mathbb{F}_{\ell}) G ℓ = GL 2 ( F ℓ ) .
G is conjugate to a subgroup of B()
G ℓ G_{\ell} G ℓ is conjugate to a subgroup of B ( ℓ ) B(\ell) B ( ℓ ) . In this subsection, we give a condition under which the image G ℓ G_{\ell} G ℓ is contained in B ( ℓ ) B(\ell) B ( ℓ ) . We begin by relating G ℓ G_{\ell} G ℓ to the existence of isogenies of the associated elliptic curve. This theorem is well-known (see, for example, [25 §1, p. 314] ) but due to the lack of a proper reference, we provide a proof
Proposition 2.2. G ℓ G_{\ell} G ℓ is contained in a B ( ℓ ) B(\ell) B ( ℓ ) if and only if the elliptic curve E E E admits an isogeny, defined over K K K , of degree ℓ \ell ℓ .
Proof. Suppose that ρ E , ℓ ( G K ) \rho_{E,\ell}(G_K) ρ E , ℓ ( G K ) is contained in a Borel subgroup. Without loss of generality, we may assume it is contained in B ( ℓ ) B(\ell) B ( ℓ ) . Thus E [ ℓ ] E[\ell] E [ ℓ ] has a one dimensional invariant subspace of order ℓ \ell ℓ , say C C C . Using this subgroup, we can construct a quotient isogeny
ϕ : E → E / C \phi:E\to E/C ϕ : E → E / C
defined over K K K , of degree deg ϕ = # ker ϕ = # C = ℓ \deg\phi=\#\ker\phi=\#C=\ell deg ϕ = # ker ϕ = # C = ℓ . Hence E E E admits an isogeny of degree ℓ \ell ℓ . Conversely, assume that φ : E → E ′ \varphi:E\to E' φ : E → E ′ is an isogeny of degree ℓ \ell ℓ . The order of the subgroup ker φ \ker\varphi ker φ equals the degree of φ \varphi φ , which is ℓ \ell ℓ . Hence
ker φ ⊂ E [ ℓ ] . \ker\varphi\subset E[\ell]. ker φ ⊂ E [ ℓ ] .
Since φ \varphi φ is defined over K K K , we have ρ E , ℓ ( G K ) ( ker φ ) ⊆ ker φ \rho_{E,\ell}(G_K)(\ker\varphi)\subseteq\ker\varphi ρ E , ℓ ( G K ) ( ker φ ) ⊆ ker φ . Choosing a generator of the group ker φ \ker\varphi ker φ as one of the basis elements for E [ ℓ ] E[\ell] E [ ℓ ] , we see that
ρ E , ℓ ( G K ) ⊆ ( ∗ ∗ 0 ∗ ) \rho_{E,\ell}(G_K)\subseteq
\begin{pmatrix}
* & * \\
0 & *
\end{pmatrix} ρ E , ℓ ( G K ) ⊆ ( ∗ 0 ∗ ∗ )
Thus, the image is contained in a Borel subgroup.
Thus, if we know the primes for which the elliptic curve admits a rational isogeny of that degree, we will have found the primes for which the image of the associated Galois representation is contained in a Borel subgroup. According to Theorem 1.9 of Banwait [3 ] , for primes greater than the integer 163, should ensure that the image is not contained in a Borel subgroup for Q ( D ) \mathbb{Q}(\sqrt{D}) Q ( D ) for ∣ D ∣ < 50 |D|<50 ∣ D ∣ < 50 .
G is contained in an Exceptional subgroup
Let K K K be a number field, ℓ \ell ℓ a rational prime number and ν \nu ν a prime of O K \mathcal{O}_K O K lying over ℓ \ell ℓ . Let E E E be an elliptic curve over K K K . From Serre’s analysis of the tame inertia subgroups (refer to sections 1.11 and 1.12 in Serre [26 Proposition 11, Proposition 12] ), we have the following:
If E E E has ordinary good reduction at ν \nu ν or multiplicative bad reduction, then ρ E , ℓ ( I ν ) \rho_{E,\ell}(I_\nu) ρ E , ℓ ( I ν ) has order ℓ − 1 \ell-1 ℓ − 1 or ℓ ( ℓ − 1 ) \ell(\ell-1) ℓ ( ℓ − 1 ) and one can represent it in matrices of the form
( ∗ 0 0 1 ) or ( ∗ ∗ 0 1 ) . \begin{pmatrix} * & 0 \\ 0 & 1 \end{pmatrix}
\quad\text{or}\quad
\begin{pmatrix} * & * \\ 0 & 1 \end{pmatrix}. ( ∗ 0 0 1 ) or ( ∗ 0 ∗ 1 ) .
If E E E has supersingular good reduction at ν \nu ν , then ρ E , ℓ ( I ν ) \rho_{E,\ell}(I_\nu) ρ E , ℓ ( I ν ) is cyclic of order ℓ 2 − 1 \ell^2-1 ℓ 2 − 1 .
Let G ~ ℓ \widetilde{G}_{\ell} G ℓ be the projection of G ℓ = ρ E , ℓ ( G K ) G_{\ell}=\rho_{E,\ell}(G_K) G ℓ = ρ E , ℓ ( G K ) in the group PGL 2 ( F ℓ ) \operatorname{PGL}_2(\mathbb{F}_{\ell}) PGL 2 ( F ℓ ) . In case 1, G ℓ G_{\ell} G ℓ contains a split semi-Cartan subgroup whose projection in PGL 2 ( F ℓ ) \operatorname{PGL}_2(\mathbb{F}_{\ell}) PGL 2 ( F ℓ ) is a cyclic subgroup of order ℓ − 1 \ell-1 ℓ − 1 (Discussion of part (a) in [26 §2.1] ). In case (2), G ℓ G_{\ell} G ℓ contains a nonsplit Cartan subgroup whose projection in PGL 2 ( F ℓ ) \operatorname{PGL}_2(\mathbb{F}_{\ell}) PGL 2 ( F ℓ ) is a cyclic subgroup of order ℓ + 1 \ell+1 ℓ + 1 .
Since the subgroups A 4 A_4 A 4 , S 4 S_4 S 4 and A 5 A_5 A 5 have elements of order at most 6, choosing a prime number ℓ \ell ℓ greater than 7 implies that G ~ ℓ \widetilde{G}_{\ell} G ℓ will not be equal to these exceptional subgroups.
G is conjugate to a subgroup of Cns()
We finally consider the case where the image lies in a non-split Cartan subgroup. It is here that we crucially require the field to be a formally real number field.
Lemma 2.2. If G K G_K G K contains an involution, then the image cannot be contained in a non-split Cartan subgroup.
Proof. Let K K K be a field such that Gal ( K ‾ / K ) \operatorname{Gal}(\overline{K}/K) Gal ( K / K ) contains an involution (for example, K K K is a formally real number field [12 Kapitel I, §6, Korollar 2] ).
Then ρ ℓ ( G ) \rho_{\ell}(G) ρ ℓ ( G ) contains an element with eigenvalues ± 1 \pm1 ± 1 . However, C n s ( ℓ ) C_{ns}(\ell) C n s ( ℓ ) does not contain any element with these eigenvalues. To see this, we note that a general element of C n s ( ℓ ) C_{ns}(\ell) C n s ( ℓ ) is of the form
A = ( a ε b b a ) a , b ∈ F ℓ and ε a nonsquare in F ℓ . A=\begin{pmatrix} a & \varepsilon b \\ b & a \end{pmatrix}
\qquad a,b\in\mathbb{F}_{\ell}\text{ and }\varepsilon\text{ a nonsquare in }\mathbb{F}_{\ell}. A = ( a b ε b a ) a , b ∈ F ℓ and ε a nonsquare in F ℓ .
For A A A to have eigenvalues ± 1 \pm1 ± 1 , we must have
tr ( A ) = 0 and det ( A ) = − 1. \operatorname{tr}(A)=0\quad\text{and}\quad\det(A)=-1. tr ( A ) = 0 and det ( A ) = − 1.
This would imply 2 a = 0 2a=0 2 a = 0 and a 2 − ε b 2 = − 1 a^2-\varepsilon b^2=-1 a 2 − ε b 2 = − 1 . Assuming ℓ ≠ 2 \ell\ne2 ℓ = 2 , the first of these relations gives us a = 0 a=0 a = 0 , which implies that we have ε b 2 = 1 \varepsilon b^2=1 ε b 2 = 1 , or ε = ( b − 1 ) 2 \varepsilon=(b^{-1})^2 ε = ( b − 1 ) 2 . This is a contradiction to the fact that ε \varepsilon ε was assumed to be a nonsquare in F ℓ \mathbb{F}_{\ell} F ℓ .
To summarize, we have the following. From Theorem 2.1, all that is left to be classified is the case when
G ℓ G_{\ell} G ℓ is conjugate to a subgroup of the normalizer of a split Cartan subgroup, but not conjugate to any subgroup of the split Cartan.
G ℓ G_{\ell} G ℓ is conjugate to a subgroup of the normalizer of a non-split Cartan subgroup, but not conjugate to any subgroup of the non-split Cartan.
For rational numbers, case (1) has been shown to not be possible for ℓ ≥ 11 \ell\ge11 ℓ ≥ 11 and ℓ ≠ 13 \ell\ne13 ℓ = 13 by Bilu, Parent and Rebolledo [4 Corollary 1.2, p.960] . In the paper, they show that X s p l i t ( ℓ ) X_{\mathrm{split}}(\ell) X split ( ℓ ) has no non-trivial non-CM points for ℓ ≥ 13 \ell\ge13 ℓ ≥ 13 and ℓ ≠ 13 \ell\ne13 ℓ = 13 . The case ℓ = 13 \ell= 13 ℓ = 13 was handled by Balakrishana, Dogra, Netan, Müller and Vonk [2 Theorem 1.2, Corollary 1.3, p.888] using quadratic Chabauty methods. To answer the question for number fields, one has to determine the non-CM K K K -rational points on X s p l i t ( ℓ ) X_{\mathrm{split}}(\ell) X split ( ℓ ) . This is still an open problem. We do not deal with this case in this paper and only discuss case (2).
When we do not wish to make a distinction between the split and non-split Cartan subgroups, we will just refer to it as a Cartan subgroup. For now we have the following theorem,
Theorem 2.3. Let E E E be an elliptic curve without complex multiplication over a formally real number field K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) with 0 < D < 50 0 < D < 50 0 < D < 50 . If ℓ \ell ℓ is a prime number such that ℓ > 163 \ell> 163 ℓ > 163 , then either ρ ‾ E , ℓ \overline{\rho}_{E,\ell} ρ E , ℓ is surjective or G E ( ℓ ) G_E(\ell) G E ( ℓ ) is conjugate to a subgroup of the normalizer of a Cartan subgroup but is not conjugate to any subgroup of the Cartan subgroup.
Image contained in normalizer of non-split Cartan but not contained in non-split Cartan
Our goal will be to obtain a bound for which the image of the Galois representation can be contained in the normalizer of a non-split Cartan subgroup of G L 2 ( F ℓ ) \mathrm{GL}_2(\mathbb{F}_{\ell}) GL 2 ( F ℓ ) . The goal of this section will be to modify the effective Chebotarev theorem established in [1 ] to get a version that allows one to avoid a prescribed set of prime ideals.
Forms of the Chebotarev density theorem
Let K K K be a number field of degree n K n_K n K over the rationals Q \mathbb{Q} Q , and L L L be a finite normal extension of K K K such that n L = [ L : Q ] n_L = [L:\mathbb{Q}] n L = [ L : Q ] and n = n L / n K = [ L : K ] n = n_L/n_K = [L:K] n = n L / n K = [ L : K ] , the degree of L L L over K K K . Let G = Gal ( L / K ) G = \operatorname{Gal}(L/K) G = Gal ( L / K ) be the Galois group of L L L over K K K . Let d K d_K d K , resp. d L d_L d L be the absolute discriminant of K K K , resp. of L L L .
Let p \mathfrak{p} p be a prime ideal of K K K unramified in L L L , and let the Artin symbol ( p L / K ) \left(\frac{\mathfrak{p}}{L/K}\right) ( L / K p ) (for more on the Artin symbol, refer to [8 p. 106-109] ) denote the conjugacy class of Frobenius automorphisms corresponding to prime ideals P ∣ p \mathfrak{P} \mid\mathfrak{p} P ∣ p , where P \mathfrak{P} P is a prime ideal of L L L . For each conjugacy class C C C of G G G , we define
π C ( x , L / K ) = ∣ { p : p unramified in L , ( p L / K ) = C , N K / Q ( p ) ≤ x } ∣ . \pi_C(x,L/K) = \left|\left\{\mathfrak{p} : \mathfrak{p}\text{ unramified in }L,\ \left(\frac{\mathfrak{p}}{L/K}\right)=C,\ N_{K/\mathbb{Q}}(\mathfrak{p}) \le x\right\}\right|. π C ( x , L / K ) = { p : p unramified in L , ( L / K p ) = C , N K / Q ( p ) ≤ x } .
Let Li ( x ) \operatorname{Li}(x) Li ( x ) be the familiar logarithmic integral:
Li ( x ) = ∫ 2 x d t log t ∼ x log x , as x → ∞ \operatorname{Li}(x) = \int_2^x \frac{\mathrm{d}t}{\log t} \sim\frac{x}{\log x},\quad\text{as }x\to\infty Li ( x ) = ∫ 2 x log t d t ∼ log x x , as x → ∞
By Generalized Riemann Hypothesis (GRH) we mean the assertion that the Dedekind zeta function of a number field has no zeros with real part greater than 1 / 2 1/2 1/2 . This hypothesis enables one to obtain the following strengthened version of a consequence of the Chebotarev density theorem [13 ] :
Theorem 3.1. Assuming G R H GRH GR H , there exists an absolute constant c > 0 c > 0 c > 0 such that
∣ π C ( x ) − ∣ C ∣ ∣ G ∣ Li ( x ) ∣ ≤ c ∣ C ∣ ∣ G ∣ x ( log d L + n L log x ) \left|\pi_C(x)-\frac{|C|}{|G|}\operatorname{Li}(x)\right| \le c\frac{|C|}{|G|}\sqrt{x}\left(\log d_L+n_L\log x\right) π C ( x ) − ∣ G ∣ ∣ C ∣ Li ( x ) ≤ c ∣ G ∣ ∣ C ∣ x ( log d L + n L log x )
for all x ≥ 2 x \ge2 x ≥ 2 .
We use the following explicit version due to Bach and Sorenson [1 Theorem 5.1, p. 1729]
Theorem 3.2. Assume GRH. Let L / K L/K L / K be a Galois extension of number fields, with L ≠ Q L \ne\mathbb{Q} L = Q . Let d L d_L d L denote the absolute value of L L L ’s discriminant. Let n n n denote the degree of L L L over Q \mathbb{Q} Q . Let σ ∈ G \sigma\in G σ ∈ G , the Galois group of L / K L/K L / K . Then there is a prime ideal p \mathfrak{p} p of K K K with ( p L / K ) = σ \left(\frac{\mathfrak{p}}{L/K}\right)=\sigma ( L / K p ) = σ , of residue degree 1 1 1 , satisfying
N p ≤ ( 4 log d L + 2.5 n + 5 ) 2 . N\mathfrak{p} \le(4\log d_L+2.5n+5)^2. N p ≤ ( 4 log d L + 2.5 n + 5 ) 2 .
A modified version of the Effective Chebotarev density theorem
We wish to modify the conclusion of Theorem 3.2 , so that we are able to choose a prime ideal p \mathfrak{p} p that lies outside a prescribed set S S S . If we suppose that S S S contains the prime ideals of bad reduction for an elliptic curve E E E , then the modified theorem would give us a prime ideal p \mathfrak{p} p of good reduction that satisfies the conclusion of Theorem 3.2 . To do so, we first prove
Lemma 3.3. Let K K K be a number field and let S = { q 1 , … , q e } S=\{\mathfrak{q}_1,\ldots,\mathfrak{q}_e\} S = { q 1 , … , q e } be a finite set of nonzero prime ideals of O K \mathcal{O}_K O K . Then there exists an element α ∈ K × \alpha\in K^\times α ∈ K × such that v q i ( α ) ≡ 1 ( m o d 2 ) v_{\mathfrak{q}_i}(\alpha)\equiv1\pmod2 v q i ( α ) ≡ 1 ( mod 2 ) for 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e . For every α \alpha α satisfying these valuation conditions, every prime ideal of S S S ramifies in the extension M = K ( α ) M=K(\sqrt{\alpha}) M = K ( α ) .
Proof. We first show the existence of such an α \alpha α satisfying the prescribed valuation conditions. For each 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e , choose an a i ∈ q i ∖ q i 2 a_i\in\mathfrak{q}_i\setminus\mathfrak{q}_i^2 a i ∈ q i ∖ q i 2 . By the Chinese remainder theorem, there exists an α ∈ O K \alpha\in\mathcal{O}_K α ∈ O K such that α ≡ a i ( m o d q i 2 ) \alpha\equiv a_i\pmod{\mathfrak{q}_i^2} α ≡ a i ( mod q i 2 ) for every 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e . For each i i i , the inclusion α − a i ∈ q i 2 \alpha-a_i\in\mathfrak{q}_i^2 α − a i ∈ q i 2 gives α ∈ q i \alpha\in\mathfrak{q}_i α ∈ q i . Moreover, if α ∈ q i 2 \alpha\in\mathfrak{q}_i^2 α ∈ q i 2 , then a i = α − ( α − a i ) ∈ q i 2 a_i=\alpha-(\alpha-a_i)\in\mathfrak{q}_i^2 a i = α − ( α − a i ) ∈ q i 2 , contrary to the choice of a i a_i a i . Consequently, α ∈ q i ∖ q i 2 \alpha\in\mathfrak{q}_i\setminus\mathfrak{q}_i^2 α ∈ q i ∖ q i 2 for every 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e . In particular, α \alpha α is non-zero.
We now prove that each prime ideal in S S S ramifies in M = K ( α ) M=K(\sqrt{\alpha}) M = K ( α ) . Fix an index i i i , and suppose ( α ) = q i r p 1 r 1 ⋯ p t r t (\alpha)=\mathfrak{q}_i^r\mathfrak{p}_1^{r_1}\cdots\mathfrak{p}_t^{r_t} ( α ) = q i r p 1 r 1 ⋯ p t r t , where r ≥ 0 r\ge0 r ≥ 0 and the p j \mathfrak{p}_j p j are distinct nonzero prime ideals different from q i \mathfrak{q}_i q i . Since α ∈ q i ∖ q i 2 \alpha\in\mathfrak{q}_i\setminus\mathfrak{q}_i^2 α ∈ q i ∖ q i 2 , by [22 Corollary 3.13, p. 51] v q i ( α ) ≡ 1 ( m o d 2 ) v_{\mathfrak{q}_i}(\alpha)\equiv1\pmod2 v q i ( α ) ≡ 1 ( mod 2 ) for every 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e .
Now let α ∈ K × \alpha\in K^\times α ∈ K × be any element such that v q i ( α ) v_{\mathfrak{q}_i}(\alpha) v q i ( α ) is odd, then the prime ideal q i \mathfrak{q}_i q i ramifies in M / K M/K M / K . Indeed, fix q i ∈ S \mathfrak{q}_i\in S q i ∈ S and put r : = v q i ( α ) r:=v_{\mathfrak{q}_i}(\alpha) r := v q i ( α ) . Since r r r is odd, α \alpha α is not a square in K K K . Consequently, M = K ( α ) M=K(\sqrt{\alpha}) M = K ( α ) is a quadratic extension of K K K . Let Q \mathfrak{Q} Q be a prime ideal of O M \mathcal{O}_M O M lying above q i \mathfrak{q}_i q i . Express q i O M = Q e Q 2 e 2 ⋯ Q g e g \mathfrak{q}_i\mathcal{O}_M=\mathfrak{Q}^{e}\mathfrak{Q}_2^{e_2}\cdots\mathfrak{Q}_g^{e_g} q i O M = Q e Q 2 e 2 ⋯ Q g e g , where Q \mathfrak{Q} Q is distinct from Q j \mathfrak{Q}_j Q j , 2 ≤ j ≤ g 2\le j\le g 2 ≤ j ≤ g and we have α O K = q i r p 1 r 1 ⋯ p t r t \alpha\mathcal{O}_K=\mathfrak{q}_i^r\mathfrak{p}_1^{r_1}\cdots\mathfrak{p}_t^{r_t} α O K = q i r p 1 r 1 ⋯ p t r t , where p j \mathfrak{p}_j p j are distinct from q i \mathfrak{q}_i q i . Now extending the ideals to O M \mathcal{O}_M O M , we obtain
α O M = ( q i O M ) r ( p 1 O M ) r 1 ⋯ ( p t O M ) r t . \alpha\mathcal{O}_M=(\mathfrak{q}_i\mathcal{O}_M)^r(\mathfrak{p}_1\mathcal{O}_M)^{r_1}\cdots(\mathfrak{p}_t\mathcal{O}_M)^{r_t}. α O M = ( q i O M ) r ( p 1 O M ) r 1 ⋯ ( p t O M ) r t .
So Q \mathfrak{Q} Q occurs with exponent e r er er in ( q i O M ) r (\mathfrak{q}_i\mathcal{O}_M)^r ( q i O M ) r . Moreover, Q \mathfrak{Q} Q does not occur in any p j O M \mathfrak{p}_j\mathcal{O}_M p j O M . Indeed, [22 Lemma 3.33, p. 59] says that Q ∣ p j O M \mathfrak{Q}\mid\mathfrak{p}_j\mathcal{O}_M Q ∣ p j O M if and only if Q ∩ O K = p j \mathfrak{Q}\cap\mathcal{O}_K=\mathfrak{p}_j Q ∩ O K = p j . But Q \mathfrak{Q} Q lies above q i \mathfrak{q}_i q i , so Q ∩ O K = q i \mathfrak{Q}\cap\mathcal{O}_K=\mathfrak{q}_i Q ∩ O K = q i , which is not equal to p j \mathfrak{p}_j p j . Hence v Q ( α ) = e r v_{\mathfrak{Q}}(\alpha)=er v Q ( α ) = er in α O M \alpha\mathcal{O}_M α O M . Since α O M = ( α O M ) 2 \alpha\mathcal{O}_M=(\sqrt{\alpha}\mathcal{O}_M)^2 α O M = ( α O M ) 2 , so the exponent of Q \mathfrak{Q} Q in α O M \alpha\mathcal{O}_M α O M is even, so e e e must be even. Since M / K M/K M / K is Galois, so from [22 Theorem 3.34, p. 59] , we conclude that e ∣ 2 e\mid2 e ∣ 2 , which implies that e = 2 > 1 e=2>1 e = 2 > 1 . So q i \mathfrak{q}_i q i ramifies in M / K M/K M / K .
Using this Lemma, we prove a modified version of the Bach and Sorenson theorem, namely
Lemma 3.4. (Bound for the norm of a prime ideal ) Assume GRH. Let K K K be a number field and let L / K L/K L / K be a finite Galois extension. Let C ⊆ Gal ( L / K ) C \subseteq\operatorname{Gal}(L/K) C ⊆ Gal ( L / K ) be a nonempty subset which is closed under conjugation. Let S = { q 1 , … , q e } S = \{\mathfrak{q}_1, \ldots, \mathfrak{q}_e\} S = { q 1 , … , q e } be a finite set of nonzero prime ideals of O K \mathcal{O}_K O K . Fix any α ∈ K × \alpha\in K^\times α ∈ K × satisfying v q i ( α ) ≡ 1 ( m o d 2 ) v_{\mathfrak{q}_i}(\alpha) \equiv1 \pmod{2} v q i ( α ) ≡ 1 ( mod 2 ) for every 1 ≤ i ≤ e 1 \leq i \leq e 1 ≤ i ≤ e (whose existence follows from Lemma 3.3 ) and put M : = K ( α ) M := K(\sqrt{\alpha}) M := K ( α ) and L ~ : = L M \widetilde{L} := LM L := L M . Then there exists a prime ideal p ⊆ O K \mathfrak{p} \subseteq\mathcal{O}_K p ⊆ O K such that p ∉ S \mathfrak{p} \notin S p ∈ / S , p \mathfrak{p} p is unramified in L / K L/K L / K , Frob p , L / K ⊆ C \operatorname{Frob}_{\mathfrak{p},L/K} \subseteq C Frob p , L / K ⊆ C , and
N K / Q p ≤ ( 4 log d L ~ + 2.5 [ L ~ : Q ] + 5 ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \leq\left(4 \log d_{\widetilde{L}} + 2.5[\widetilde{L}:\mathbb{Q}] + 5\right)^2. N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 . Proof. Applying Lemma 3.3 , we see that every prime ideal in S S S ramifies in the extension M / K M/K M / K . Since L / K L/K L / K is Galois and M / K M/K M / K is quadratic (hence Galois), the composite extension L ~ = L M \widetilde{L} = LM L = L M is Galois over K K K .
First suppose that M ⊆ L M \subseteq L M ⊆ L . Then L ~ = L \widetilde{L} = L L = L . Since every prime ideal in S S S ramifies in M / K M/K M / K , and M ⊆ L M \subseteq L M ⊆ L , every prime ideal in S S S ramifies in L / K L/K L / K .
By the effective Chebotarev Theorem (refer 3.2), applied to the Galois extension L / K L/K L / K and the conjugacy stable subset C ⊆ Gal ( L / K ) C \subseteq\operatorname{Gal}(L/K) C ⊆ Gal ( L / K ) , there exists a prime ideal p ⊆ O K \mathfrak{p} \subseteq\mathcal{O}_K p ⊆ O K such that p \mathfrak{p} p is unramified in L / K L/K L / K , Frob p , L / K ⊆ C \operatorname{Frob}_{\mathfrak{p},L/K} \subseteq C Frob p , L / K ⊆ C , and
N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \leq\left(4 \log d_L + 2.5[L:\mathbb{Q}] + 5\right)^2. N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 . Since L ~ = L \widetilde{L} = L L = L , this gives
N K / Q p ≤ ( 4 log d L ~ + 2.5 [ L ~ : Q ] + 5 ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \leq\left(4 \log d_{\widetilde{L}} + 2.5[\widetilde{L}:\mathbb{Q}] + 5\right)^2. N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 . Also, p ∉ S \mathfrak{p} \notin S p ∈ / S , because every prime ideal in S S S ramifies in L / K L/K L / K , whereas p \mathfrak{p} p is unramified in L / K L/K L / K .
Now suppose that M ⊈ L M \nsubseteq L M ⊈ L . Since M / K M/K M / K is quadratic, we have L ∩ M = K L \cap M = K L ∩ M = K . Hence [14 Theorem VI 1.14]
Gal ( L ~ / K ) ≅ Gal ( L / K ) × Gal ( M / K ) . \operatorname{Gal}(\widetilde{L}/K) \cong\operatorname{Gal}(L/K) \times\operatorname{Gal}(M/K). Gal ( L / K ) ≅ Gal ( L / K ) × Gal ( M / K ) . Let res : Gal ( L ~ / K ) ⟶ Gal ( L / K ) \operatorname{res} : \operatorname{Gal}(\widetilde{L}/K) \longrightarrow\operatorname{Gal}(L/K) res : Gal ( L / K ) ⟶ Gal ( L / K ) be the restriction map, and define C ~ : = res − 1 ( C ) \widetilde{C} := \operatorname{res}^{-1}(C) C := res − 1 ( C ) . Since C C C is closed under conjugation in Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) , the set C ~ \widetilde{C} C is closed under conjugation in Gal ( L ~ / K ) \operatorname{Gal}(\widetilde{L}/K) Gal ( L / K ) .
Applying the effective Chebotarev theorem to the Galois extension L ~ / K \widetilde{L}/K L / K and the conjugacy stable subset C ~ \widetilde{C} C , we obtain a prime ideal p ⊆ O K \mathfrak{p} \subseteq\mathcal{O}_K p ⊆ O K such that p \mathfrak{p} p is unramified in L ~ / K \widetilde{L}/K L / K , Frob p , L ~ / K ⊆ C ~ \operatorname{Frob}_{\mathfrak{p},\widetilde{L}/K} \subseteq\widetilde{C} Frob p , L / K ⊆ C , and
N K / Q p ≤ ( 4 log d L ~ + 2.5 [ L ~ : Q ] + 5 ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \leq\left(4 \log d_{\widetilde{L}} + 2.5[\widetilde{L}:\mathbb{Q}] + 5\right)^2. N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 . Since p \mathfrak{p} p is unramified in L ~ / K \widetilde{L}/K L / K , it is also unramified in L / K L/K L / K . Moreover, every prime ideal in S S S ramifies in M / K M/K M / K , hence also in L ~ / K \widetilde{L}/K L / K . Therefore p ∉ S \mathfrak{p} \notin S p ∈ / S .
Finally, res ( Frob p , L ~ / K ) = Frob p , L / K \operatorname{res}\left(\operatorname{Frob}_{\mathfrak{p},\widetilde{L}/K}\right) = \operatorname{Frob}_{\mathfrak{p},L/K} res ( Frob p , L / K ) = Frob p , L / K . Since Frob p , L ~ / K ⊆ C ~ = res − 1 ( C ) \operatorname{Frob}_{\mathfrak{p},\widetilde{L}/K} \subseteq\widetilde{C} = \operatorname{res}^{-1}(C) Frob p , L / K ⊆ C = res − 1 ( C ) , we get Frob p , L / K ⊆ C \operatorname{Frob}_{\mathfrak{p},L/K} \subseteq C Frob p , L / K ⊆ C . Thus, the required prime ideal p \mathfrak{p} p exists. □
Bound for the logarithm of the Discriminant
In Lemma 3.4, the norm of the prime ideal p \mathfrak p p is bounded by [ L ~ : Q ] [\widetilde{L}:\mathbb{Q}] [ L : Q ] and log d L ~ \log d_{\widetilde{L}} log d L . In our case, the first of these constants is absolutely bounded, and thus we need to bound the second. To do so, we improve the following theorem of Serre [27 Proposition 6] (a proof of which can be found in Mayle and Wang [20 Lemma 7] ) to Galois extensions of number fields. More specifically, the base field is now allowed to be any number field instead of Q \mathbb{Q} Q . Moreover, the lower bound has been tightened when K = Q K=\mathbb{Q} K = Q .
Lemma 3.5. (Bound for Discriminant of Number Field) Let K K K be a number field and let L / K L/K L / K be a nontrivial finite Galois extension. Put n = [ L : K ] n=[L:K] n = [ L : K ] and f = [ K : Q ] f=[K:\mathbb{Q}] f = [ K : Q ] . Let d L d_L d L and d K d_K d K denote the absolute discriminants of L L L and K K K , respectively. Let D L / K ⊆ O L \mathfrak{D}_{L/K}\subseteq\mathcal{O}_L D L / K ⊆ O L be the relative different ideal and let d L / K = N L / K ( D L / K ) ⊆ O K \mathfrak{d}_{L/K}=N_{L/K}(\mathfrak{D}_{L/K})\subseteq\mathcal{O}_K d L / K = N L / K ( D L / K ) ⊆ O K be the relative discriminant ideal. Define rad K ( d L / K ) = ∏ p ∣ d L / K N K / Q ( p ) \operatorname{rad}_K(\mathfrak{d}_{L/K})=\prod_{\mathfrak{p}\mid\mathfrak{d}_{L/K}}N_{K/\mathbb{Q}}(\mathfrak{p}) rad K ( d L / K ) = ∏ p ∣ d L / K N K / Q ( p ) , then
n log d K + n 2 log rad K ( d L / K ) ≤ log d L ≤ n log d K + ( n − 1 ) log rad K ( d L / K ) + n f log n . n\log d_K+\frac{n}{2}\log\operatorname{rad}_K(\mathfrak{d}_{L/K})\leq\log d_L\leq n\log d_K+(n-1)\log\operatorname{rad}_K(\mathfrak{d}_{L/K})+nf\log n. n log d K + 2 n log rad K ( d L / K ) ≤ log d L ≤ n log d K + ( n − 1 ) log rad K ( d L / K ) + n f log n .
Proof. Fix a nonzero prime ideal p \mathfrak p p of O K \mathcal{O}_K O K . In O L \mathcal{O}_L O L , let p O L = P 1 e 1 ⋯ P g e g \mathfrak p\mathcal{O}_L=\mathfrak{P}_1^{e_1}\cdots\mathfrak{P}_g^{e_g} p O L = P 1 e 1 ⋯ P g e g , where P 1 , … , P g \mathfrak{P}_1,\ldots,\mathfrak{P}_g P 1 , … , P g are distinct prime ideals of O L \mathcal{O}_L O L lying above p \mathfrak p p . For P = P i \mathfrak{P}=\mathfrak{P}_i P = P i , let e P / p e_{\mathfrak{P}/\mathfrak p} e P / p denote its ramification index, and f P / p f_{\mathfrak{P}/\mathfrak p} f P / p be its residue degree. Since L / K L/K L / K is Galois, for fixed p \mathfrak p p , all primes P ∣ p \mathfrak{P}\mid\mathfrak p P ∣ p have the same ramification index. Thus we may write e P / p = e e_{\mathfrak{P}/\mathfrak p}=e e P / p = e , and hence ∑ P ∣ p e P / p f P / p = n \sum_{\mathfrak{P}\mid\mathfrak p}e_{\mathfrak{P}/\mathfrak p}f_{\mathfrak{P}/\mathfrak p}=n ∑ P ∣ p e P / p f P / p = n . The relative discriminant formula [23 Corollary 2.10, p. 202] gives
d L = d K n N K / Q ( d L / K ) . d_L=d_K^nN_{K/\mathbb{Q}}(\mathfrak{d}_{L/K}). d L = d K n N K / Q ( d L / K ) .
Taking logarithms on both sides, we obtain
log d L = n log d K + log N K / Q ( d L / K ) . (1) \log d_L=n\log d_K+\log N_{K/\mathbb{Q}}(\mathfrak{d}_{L/K}). \tag*{(1)} log d L = n log d K + log N K / Q ( d L / K ) . (1)
Using the factorization of the relative discriminant ideal as d L / K = ∏ p p v p ( d L / K ) \mathfrak{d}_{L/K}=\prod_{\mathfrak p}\mathfrak p^{v_{\mathfrak p}(\mathfrak{d}_{L/K})} d L / K = ∏ p p v p ( d L / K ) , we obtain N K / Q ( d L / K ) = ∏ p N K / Q ( p ) v p ( d L / K ) N_{K/\mathbb{Q}}(\mathfrak{d}_{L/K})=\prod_{\mathfrak p}N_{K/\mathbb{Q}}(\mathfrak p)^{v_{\mathfrak p}(\mathfrak{d}_{L/K})} N K / Q ( d L / K ) = ∏ p N K / Q ( p ) v p ( d L / K ) . Again, by taking logarithms on both sides,
log N K / Q ( d L / K ) = ∑ p v p ( d L / K ) log N K / Q ( p ) . \log N_{K/\mathbb{Q}}(\mathfrak{d}_{L/K})=\sum_{\mathfrak p}v_{\mathfrak p}(\mathfrak{d}_{L/K})\log N_{K/\mathbb{Q}}(\mathfrak p). log N K / Q ( d L / K ) = p ∑ v p ( d L / K ) log N K / Q ( p ) .
Since d L / K = N L / K ( D L / K ) \mathfrak{d}_{L/K}=N_{L/K}(\mathfrak{D}_{L/K}) d L / K = N L / K ( D L / K ) , we have v p ( d L / K ) = ∑ P ∣ p f P / p v P ( D L / K ) v_{\mathfrak p}(\mathfrak{d}_{L/K})=\sum_{\mathfrak{P}\mid\mathfrak p}f_{\mathfrak{P}/\mathfrak p}v_{\mathfrak P}(\mathfrak{D}_{L/K}) v p ( d L / K ) = ∑ P ∣ p f P / p v P ( D L / K ) . By Neukirch [23 Theorem 2.6, p. 199] , we have
e P / p − 1 ≤ v P ( D L / K ) ≤ e P / p − 1 + v P ( e P / p ) . e_{\mathfrak P/\mathfrak p}-1\leq v_{\mathfrak P}(\mathfrak{D}_{L/K})\leq e_{\mathfrak P/\mathfrak p}-1+v_{\mathfrak P}(e_{\mathfrak P/\mathfrak p}). e P / p − 1 ≤ v P ( D L / K ) ≤ e P / p − 1 + v P ( e P / p ) .
Define s P : = v P ( D L / K ) − ( e P / p − 1 ) s_{\mathfrak P}:=v_{\mathfrak P}(\mathfrak{D}_{L/K})-(e_{\mathfrak P/\mathfrak p}-1) s P := v P ( D L / K ) − ( e P / p − 1 ) . We thus have
v P ( D L / K ) = e P / p − 1 + s P , where 0 ≤ s P ≤ v P ( e P / p ) . (2) v_{\mathfrak P}(\mathfrak{D}_{L/K})=e_{\mathfrak P/\mathfrak p}-1+s_{\mathfrak P},\qquad\text{where }0\leq s_{\mathfrak P}\leq v_{\mathfrak P}(e_{\mathfrak P/\mathfrak p}). \tag*{(2)} v P ( D L / K ) = e P / p − 1 + s P , where 0 ≤ s P ≤ v P ( e P / p ) . (2)
Fix a prime ideal p ∣ d L / K \mathfrak p\mid\mathfrak d_{L/K} p ∣ d L / K . By [22 Remark 3.39(b), p. 62] , we may assume that p \mathfrak p p is ramified in L / K L/K L / K . Since L / K L/K L / K is Galois, all primes above p \mathfrak p p have the same ramification index e e e , and ramification means e ≥ 2 e\geq2 e ≥ 2 . Since s P ≥ 0 s_{\mathfrak P}\geq0 s P ≥ 0 , we get v P ( D L / K ) ≥ e P / p − 1 v_{\mathfrak P}(\mathfrak{D}_{L/K})\geq e_{\mathfrak P/\mathfrak p}-1 v P ( D L / K ) ≥ e P / p − 1 . Therefore
v p ( d L / K ) = ∑ P ∣ p f P / p v P ( D L / K ) ≥ ∑ P ∣ p f P / p ( e P / p − 1 ) . v_{\mathfrak p}(\mathfrak{d}_{L/K})=\sum_{\mathfrak P\mid\mathfrak p}f_{\mathfrak P/\mathfrak p}v_{\mathfrak P}(\mathfrak{D}_{L/K})\geq\sum_{\mathfrak P\mid\mathfrak p}f_{\mathfrak P/\mathfrak p}(e_{\mathfrak P/\mathfrak p}-1). v p ( d L / K ) = P ∣ p ∑ f P / p v P ( D L / K ) ≥ P ∣ p ∑ f P / p ( e P / p − 1 ) .
Since e P / p = e e_{\mathfrak{P}/\mathfrak{p}}=e e P / p = e , this becomes
v p ( d L / K ) ≥ ∑ P ∣ p f P / p ( e − 1 ) . v_{\mathfrak{p}}(\mathfrak{d}_{L/K}) \geq\sum_{\mathfrak{P}\mid\mathfrak{p}} f_{\mathfrak{P}/\mathfrak{p}}(e-1). v p ( d L / K ) ≥ P ∣ p ∑ f P / p ( e − 1 ) .
The product of the local ramification indices and inertia degrees sum to the degree of the extension, that is, ∑ P ∣ p e f P / p = n \sum_{\mathfrak{P}\mid\mathfrak{p}} e f_{\mathfrak{P}/\mathfrak{p}}=n ∑ P ∣ p e f P / p = n . Therefore, ∑ P ∣ p f P / p = n e \sum_{\mathfrak{P}\mid\mathfrak{p}} f_{\mathfrak{P}/\mathfrak{p}}=\frac{n}{e} ∑ P ∣ p f P / p = e n , hence v p ( d L / K ) ≥ ( e − 1 ) n e = n ( 1 − 1 e ) v_{\mathfrak{p}}(\mathfrak{d}_{L/K})\geq(e-1)\frac{n}{e}=n\left(1-\frac{1}{e}\right) v p ( d L / K ) ≥ ( e − 1 ) e n = n ( 1 − e 1 ) . Therefore v p ( d L / K ) ≥ n 2 v_{\mathfrak{p}}(\mathfrak{d}_{L/K})\geq\frac{n}{2} v p ( d L / K ) ≥ 2 n . Summing over all prime ideals p ∣ d L / K \mathfrak{p}\mid\mathfrak{d}_{L/K} p ∣ d L / K , we obtain
log N K / Q ( d L / K ) = ∑ p ∣ d L / K v p ( d L / K ) log N K / Q ( p ) ≥ ∑ p ∣ d L / K n 2 log N K / Q ( p ) = n 2 log rad K ( d L / K ) . (3) \log N_{K/\mathbb{Q}}(\mathfrak{d}_{L/K})=\sum_{\mathfrak{p}\mid\mathfrak{d}_{L/K}}v_{\mathfrak{p}}(\mathfrak{d}_{L/K})\log N_{K/\mathbb{Q}}(\mathfrak{p})\geq\sum_{\mathfrak{p}\mid\mathfrak{d}_{L/K}}\frac{n}{2}\log N_{K/\mathbb{Q}}(\mathfrak{p})=\frac{n}{2}\log\operatorname{rad}_{K}(\mathfrak{d}_{L/K}). \tag*{(3)} log N K / Q ( d L / K ) = p ∣ d L / K ∑ v p ( d L / K ) log N K / Q ( p ) ≥ p ∣ d L / K ∑ 2 n log N K / Q ( p ) = 2 n log rad K ( d L / K ) . (3)
Using the logarithm of the discriminant equation (1) and combining with equation (3) gives us a lower bound
log d L ≥ n log d K + n 2 log rad K ( d L / K ) . (4) \log d_L\geq n\log d_K+\frac{n}{2}\log\operatorname{rad}_{K}(\mathfrak{d}_{L/K}). \tag*{(4)} log d L ≥ n log d K + 2 n log rad K ( d L / K ) . (4)
Again use v P ( D L / K ) = e P / p − 1 + s P v_{\mathfrak{P}}(\mathfrak{D}_{L/K})=e_{\mathfrak{P}/\mathfrak{p}}-1+s_{\mathfrak{P}} v P ( D L / K ) = e P / p − 1 + s P with 0 ≤ s P ≤ v P ( e P / p ) 0\leq s_{\mathfrak{P}}\leq v_{\mathfrak{P}}(e_{\mathfrak{P}/\mathfrak{p}}) 0 ≤ s P ≤ v P ( e P / p ) . Then
v p ( d L / K ) = ∑ P ∣ p f P / p v P ( D L / K ) ≤ ∑ P ∣ p f P / p ( e P / p − 1 ) + ∑ P ∣ p f P / p v P ( e P / p ) . v_{\mathfrak{p}}(\mathfrak{d}_{L/K})=\sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{P}}(\mathfrak{D}_{L/K})\leq\sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}(e_{\mathfrak{P}/\mathfrak{p}}-1)+\sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{P}}(e_{\mathfrak{P}/\mathfrak{p}}). v p ( d L / K ) = P ∣ p ∑ f P / p v P ( D L / K ) ≤ P ∣ p ∑ f P / p ( e P / p − 1 ) + P ∣ p ∑ f P / p v P ( e P / p ) .
Since L / K L/K L / K is Galois, e P / p = e e_{\mathfrak{P}/\mathfrak{p}}=e e P / p = e for all P ∣ p \mathfrak{P}\mid\mathfrak{p} P ∣ p . Hence
∑ P ∣ p f P / p ( e − 1 ) = n − ∑ P ∣ p f P / p ≤ n − 1. \sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}(e-1)=n-\sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}\leq n-1. P ∣ p ∑ f P / p ( e − 1 ) = n − P ∣ p ∑ f P / p ≤ n − 1.
For the second sum, by [22 Equation 13, p. 66] we have v P ( x ) = e P / p v p ( x ) v_{\mathfrak{P}}(x)=e_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{p}}(x) v P ( x ) = e P / p v p ( x ) for every x ∈ K × x\in K^{\times} x ∈ K × . In particular, take x = e x=e x = e then we have
v P ( e ) = e P / p v p ( e O K ) = e v p ( e O K ) . v_{\mathfrak{P}}(e)=e_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{p}}(e\mathcal{O}_K)=ev_{\mathfrak{p}}(e\mathcal{O}_K). v P ( e ) = e P / p v p ( e O K ) = e v p ( e O K ) .
Since L / K L/K L / K is Galois, e ∣ n e\mid n e ∣ n . Hence v p ( e O K ) ≤ v p ( n O K ) v_{\mathfrak{p}}(e\mathcal{O}_K)\leq v_{\mathfrak{p}}(n\mathcal{O}_K) v p ( e O K ) ≤ v p ( n O K ) . Therefore v P ( e ) ≤ e v p ( n O K ) v_{\mathfrak{P}}(e)\leq ev_{\mathfrak{p}}(n\mathcal{O}_K) v P ( e ) ≤ e v p ( n O K ) . Thus
∑ P ∣ p f P / p v P ( e ) ≤ ∑ P ∣ p f P / p e v p ( n O K ) . \sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{P}}(e)\leq\sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}ev_{\mathfrak{p}}(n\mathcal{O}_K). P ∣ p ∑ f P / p v P ( e ) ≤ P ∣ p ∑ f P / p e v p ( n O K ) .
Using ∑ P ∣ p e f P / p = n \sum_{\mathfrak{P}\mid\mathfrak{p}}ef_{\mathfrak{P}/\mathfrak{p}}=n ∑ P ∣ p e f P / p = n , we obtain ∑ P ∣ p f P / p v P ( e ) ≤ n v p ( n O K ) \sum_{\mathfrak{P}\mid\mathfrak{p}}f_{\mathfrak{P}/\mathfrak{p}}v_{\mathfrak{P}}(e)\leq nv_{\mathfrak{p}}(n\mathcal{O}_K) ∑ P ∣ p f P / p v P ( e ) ≤ n v p ( n O K ) . Consequently,
v p ( d L / K ) ≤ n − 1 + n v p ( n O K ) . v_{\mathfrak{p}}(\mathfrak{d}_{L/K})\leq n-1+nv_{\mathfrak{p}}(n\mathcal{O}_K). v p ( d L / K ) ≤ n − 1 + n v p ( n O K ) .
Multiplying by log N K / Q ( p ) \log N_{K/\mathbb{Q}}(\mathfrak{p}) log N K / Q ( p ) and summing over all p ∣ d L / K \mathfrak{p}\mid\mathfrak{d}_{L/K} p ∣ d L / K gives
log N K / Q ( d L / K ) ≤ ∑ p ∣ d L / K ( n − 1 ) log N K / Q ( p ) + n ∑ p ∣ d L / K v p ( n O K ) log N K / Q ( p ) . \log N_{K/\mathbb{Q}}(\mathfrak{d}_{L/K})\leq\sum_{\mathfrak{p}\mid\mathfrak{d}_{L/K}}(n-1)\log N_{K/\mathbb{Q}}(\mathfrak{p})+n\sum_{\mathfrak{p}\mid\mathfrak{d}_{L/K}}v_{\mathfrak{p}}(n\mathcal{O}_K)\log N_{K/\mathbb{Q}}(\mathfrak{p}). log N K / Q ( d L / K ) ≤ p ∣ d L / K ∑ ( n − 1 ) log N K / Q ( p ) + n p ∣ d L / K ∑ v p ( n O K ) log N K / Q ( p ) .
The first sum is
( n − 1 ) log rad K ( d L / K ) . (n-1)\log\operatorname{rad}_{K}(\mathfrak{d}_{L/K}). ( n − 1 ) log rad K ( d L / K ) .
For the second sum, we enlarge the sum to all prime ideals of O K \mathcal{O}_K O K (the sum is finite as v p ( n O K ) = 0 v_{\mathfrak p}(n\mathcal{O}_K)=0 v p ( n O K ) = 0 for almost all p \mathfrak p p ):
∑ p ∣ d L / K v p ( n O K ) log N K / Q ( p ) ≤ ∑ p v p ( n O K ) log N K / Q ( p ) . \sum_{\mathfrak p\mid\mathfrak d_{L/K}}v_{\mathfrak p}(n\mathcal{O}_K)\log N_{K/\mathbb{Q}}(\mathfrak p)\leq\sum_{\mathfrak p}v_{\mathfrak p}(n\mathcal{O}_K)\log N_{K/\mathbb{Q}}(\mathfrak p). p ∣ d L / K ∑ v p ( n O K ) log N K / Q ( p ) ≤ p ∑ v p ( n O K ) log N K / Q ( p ) .
Writing n O K = ∏ p p v p ( n O K ) n\mathcal{O}_K=\prod_{\mathfrak p}\mathfrak p^{v_{\mathfrak p}(n\mathcal{O}_K)} n O K = ∏ p p v p ( n O K ) , and taking logarithms on both sides, we have
∑ p v p ( n O K ) log N K / Q ( p ) = log N K / Q ( n O K ) . \sum_{\mathfrak p}v_{\mathfrak p}(n\mathcal{O}_K)\log N_{K/\mathbb{Q}}(\mathfrak p)=\log N_{K/\mathbb{Q}}(n\mathcal{O}_K). p ∑ v p ( n O K ) log N K / Q ( p ) = log N K / Q ( n O K ) .
But N K / Q ( n O K ) = n f N_{K/\mathbb{Q}}(n\mathcal{O}_K)=n^f N K / Q ( n O K ) = n f . Hence log N K / Q ( n O K ) = f log n \log N_{K/\mathbb{Q}}(n\mathcal{O}_K)=f\log n log N K / Q ( n O K ) = f log n . Therefore
log N K / Q ( d L / K ) ≤ ( n − 1 ) log rad K ( d L / K ) + n f log n . \log N_{K/\mathbb{Q}}(\mathfrak d_{L/K})\leq(n-1)\log\operatorname{rad}_K(\mathfrak d_{L/K})+nf\log n. log N K / Q ( d L / K ) ≤ ( n − 1 ) log rad K ( d L / K ) + n f log n .
Using this with equation (1) gives us an upper bound
log d L ≤ n log d K + ( n − 1 ) log rad K ( d L / K ) + n f log n . (5) \log d_L\leq n\log d_K+(n-1)\log\operatorname{rad}_K(\mathfrak d_{L/K})+nf\log n. \tag*{(5)} log d L ≤ n log d K + ( n − 1 ) log rad K ( d L / K ) + n f log n . (5)
Combining the lower bound (4) and upper bound (5) we obtain the desired inequality. □
An Effective version of Faltings Isogeny Theorem for Number Fields
In this section, we prove an Effective version of Faltings isogeny theorem for number fields. This theorem lies at the heart of providing a bound on the primes for which the image cannot be contained in the normalizer of a non-split Cartan subgroup. We begin by describing the primes of bad reduction for an abelian variety A A A over a number field K K K in terms of the primes that ramify in some extension in §4.1. We then prove some number theoretic lemmas in §4.3 which will be used to prove the effective isogeny theorem in §4.4.
Primes of bad reduction
Let A A A be an abelian variety over the number field K K K and let K ( A [ m ] ) K(A[m]) K ( A [ m ]) be the field obtained by adjoining the co-ordinates of the m m m -torsion points of A A A in A ( Q ‾ ) A(\overline{\mathbb{Q}}) A ( Q ) to the field K K K . For a prime number ℓ \ell ℓ , we write
K ( A [ ℓ ∞ ] ) = ⋃ n = 1 ∞ K ( A [ ℓ n ] ) . K(A[\ell^\infty])=\bigcup_{n=1}^{\infty}K(A[\ell^n]). K ( A [ ℓ ∞ ]) = n = 1 ⋃ ∞ K ( A [ ℓ n ]) .
The following theorem by Serre and Tate relates the primes of good reduction of A A A with the primes that ramify in the extension K ( A [ ℓ ∞ ] ) K(A[\ell^\infty]) K ( A [ ℓ ∞ ]) .
Theorem 4.1 (Serre-Tate). Let A / K A/K A / K be an abelian variety. Let p \mathfrak p p denote a prime ideal in O K \mathcal{O}_K O K and let p = char ( O K / p O K ) p=\operatorname{char}(\mathcal{O}_K/\mathfrak p\mathcal{O}_K) p = char ( O K / p O K ) . The following are equivalent:
(1) A A A has good reduction at p \mathfrak p p .
(2) K ( A [ m ] ) / K K(A[m])/K K ( A [ m ]) / K is unramified at p \mathfrak p p for all m m m prime to p p p .
(3) There exist infinitely many integers m m m , prime to p p p , such that K ( A [ m ] ) / K K(A[m])/K K ( A [ m ]) / K is unramified at p \mathfrak p p .
Proof. [29 Theorem 1] □
Corollary 4.1. Let B A B_A B A be the product of ideal ( ℓ ) (\ell) ( ℓ ) and the prime ideals of bad reduction for A A A . The extension K ( A [ ℓ ∞ ] ) / K K(A[\ell^\infty])/K K ( A [ ℓ ∞ ]) / K is ramified at exactly the prime ideals dividing B A B_A B A .
Proof. Suppose that q \mathfrak{q} q is a prime ideal lying above the rational prime ℓ \ell ℓ , that is, q ∣ ( ℓ ) \mathfrak{q} \mid(\ell) q ∣ ( ℓ ) as ideals. We show that q \mathfrak{q} q ramifies in K ( A [ ℓ ∞ ] ) / K K(A[\ell^\infty])/K K ( A [ ℓ ∞ ]) / K . For a finite extension L L L of K K K , the ramification of the prime ideal q \mathfrak{q} q with respect to a prime ideal Q \mathfrak{Q} Q of L L L , is controlled by the ramification of the extension of the completions L Q / K q L_{\mathfrak{Q}}/K_{\mathfrak{q}} L Q / K q . To be more precise,
e ( Q ∣ q ) = e ( L Q ∣ K q ) , (6) e(\mathfrak{Q}\mid\mathfrak{q})=e(L_{\mathfrak{Q}}\mid K_{\mathfrak{q}}), \tag*{(6)} e ( Q ∣ q ) = e ( L Q ∣ K q ) , (6)
where the left-hand side denotes the ramification index of the prime ideal q \mathfrak{q} q with respect to Q \mathfrak{Q} Q and the right-hand side denotes the ramification index of the extension of completions. Let ζ n \zeta_n ζ n denote a primitive n n n -th root of unity and let L n = K ( ζ ℓ n ) L_n=K(\zeta_{\ell^n}) L n = K ( ζ ℓ n ) . For n n n large enough, L n L_n L n is a non-trivial finite extension of K K K . Let K q K_{\mathfrak{q}} K q denote the completion of K K K with respect to the prime ideal q \mathfrak{q} q . This is a finite extension of Q ℓ \mathbb{Q}_{\ell} Q ℓ . Let E n = Q ℓ ( ζ ℓ n ) E_n=\mathbb{Q}_{\ell}(\zeta_{\ell^n}) E n = Q ℓ ( ζ ℓ n ) and let M n = K q ( ζ ℓ n ) M_n=K_{\mathfrak{q}}(\zeta_{\ell^n}) M n = K q ( ζ ℓ n ) . Pictorially, we have
M n ↙ ↘ K q E n ↘ ↙ Q ℓ \begin{array}{ccccc}
& & M_n & & \\
& \swarrow& & \searrow& \\
K_{\mathfrak{q}} & & & & E_n \\
& \searrow& & \swarrow& \\
& & \mathbb{Q}_{\ell} & &
\end{array} K q ↙ ↘ M n Q ℓ ↘ ↙ E n
Assume that M n M_n M n over K q K_{\mathfrak{q}} K q is an unramified extension for all integers n n n . Then
e ( K q ∣ Q ℓ ) = e ( M n ∣ Q ℓ ) . (7) e(K_{\mathfrak{q}}\mid\mathbb{Q}_{\ell})=e(M_n\mid\mathbb{Q}_{\ell}). \tag*{(7)} e ( K q ∣ Q ℓ ) = e ( M n ∣ Q ℓ ) . (7)
By the tower law for ramification indices, e ( E n ∣ Q ℓ ) e(E_n\mid\mathbb{Q}_{\ell}) e ( E n ∣ Q ℓ ) divides e ( M n ∣ Q ℓ ) e(M_n\mid\mathbb{Q}_{\ell}) e ( M n ∣ Q ℓ ) . The extension E n E_n E n over Q ℓ \mathbb{Q}_{\ell} Q ℓ is totally ramified of degree φ ( ℓ n ) \varphi(\ell^n) φ ( ℓ n ) , where φ \varphi φ denotes the Euler totient function. Using equation (7) , we get
φ ( ℓ n ) ∣ e ( K q ∣ Q ℓ ) ≤ [ K q : Q ℓ ] . \varphi(\ell^n)\mid e(K_{\mathfrak{q}}\mid\mathbb{Q}_{\ell})\leq[K_{\mathfrak{q}}:\mathbb{Q}_{\ell}]. φ ( ℓ n ) ∣ e ( K q ∣ Q ℓ ) ≤ [ K q : Q ℓ ] .
This leads to a contradiction, as for large enough n n n this inequality does not hold, since the right-hand side is finite. Thus M n M_n M n over K q K_{\mathfrak{q}} K q is ramified for some n n n and by equation (6) , we see that q \mathfrak{q} q is ramified in L n L_n L n . By the Weil Pairing (see exercise in [28 p. 55] ) on the abelian variety A A A , we have the following tower of fields
K ⊆ K ( ζ ℓ n ) ⊆ K ( A [ ℓ n ] ) ⊆ K ( A [ ℓ ∞ ] ) . K\subseteq K(\zeta_{\ell^n})\subseteq K(A[\ell^n])\subseteq K(A[\ell^\infty]). K ⊆ K ( ζ ℓ n ) ⊆ K ( A [ ℓ n ]) ⊆ K ( A [ ℓ ∞ ]) .
Hence, q \mathfrak{q} q is ramified in K ( A [ ℓ ∞ ] ) K(A[\ell^\infty]) K ( A [ ℓ ∞ ]) . The proof for prime ideals p \mathfrak{p} p not dividing the ideal ( ℓ ) (\ell) ( ℓ ) is the same as in Mayle and Wang [20 Corollary 9, p. 1406] with a few minor changes. That is, we replace the rational prime p p p by the prime ideal p \mathfrak{p} p and Q \mathbb{Q} Q by the number field K K K . Moreover, we use the general version of the Serre-Tate Theorem 4.1. □
Remark 4.1. Let A = E 1 × E 2 A=E_1\times E_2 A = E 1 × E 2 be the abelian variety that is the product of the two elliptic curves over K K K with conductors N E 1 \mathfrak{N}_{E_1} N E 1 and N E 2 \mathfrak{N}_{E_2} N E 2 respectively. Theorem 4.1 states that A A A has good reduction at a prime p \mathfrak{p} p lying over ℓ \ell ℓ if and only if both E 1 E_1 E 1 and E 2 E_2 E 2 have good reduction at p \mathfrak{p} p . Thus rad ( B A ) = rad ( ℓ O K I E 1 I E 2 ) \operatorname{rad}(B_A)=\operatorname{rad}(\ell\mathcal{O}_K\mathfrak{I}_{E_1}\mathfrak{I}_{E_2}) rad ( B A ) = rad ( ℓ O K I E 1 I E 2 ) .
Trace of the Frobenius
Let E E E be an elliptic curve over a number field K K K . For each rational prime p p p , and any prime ideal p \mathfrak{p} p lying over p p p , we may associate a number
a p ( E ) : = N ( p ) + 1 − # E ( O K / p ) , (8) a_{\mathfrak{p}}(E) := N(\mathfrak{p}) + 1 - \#E(\mathcal{O}_K/\mathfrak{p}), \tag*{(8)} a p ( E ) := N ( p ) + 1 − # E ( O K / p ) , (8)
where N ( E ) N(E) N ( E ) denotes the norm of the prime ideal p \mathfrak{p} p , and # E ( O K / p ) \#E(\mathcal{O}_K/\mathfrak{p}) # E ( O K / p ) denote the number of solutions for E E E in O K / p \mathcal{O}_K/\mathfrak{p} O K / p . This definition seems to depend on the choice of prime ideal p \mathfrak{p} p lying over p p p . We will now show that this is not the case when K K K is Galois over Q \mathbb{Q} Q .
Proposition 4.1. Let K K K be a Galois extension of Q \mathbb{Q} Q and let E E E be an elliptic curve over K K K . Let p \mathfrak{p} p and q \mathfrak{q} q be two prime ideals lying over the rational prime p p p . We have a p ( E ) = a q ( E ) a_{\mathfrak{p}}(E)=a_{\mathfrak{q}}(E) a p ( E ) = a q ( E ) .
Proof. In a Galois extension, all prime ideals lying above a rational prime are conjugate to each other. That is, we can find a σ ∈ Gal ( K / Q ) \sigma\in\operatorname{Gal}(K/\mathbb{Q}) σ ∈ Gal ( K / Q ) such that σ ( p ) = q \sigma(\mathfrak{p})=\mathfrak{q} σ ( p ) = q . It follows that the norms are the same and the residue fields are isomorphic (see [23 Chapter 1, §9] ) Looking at the terms appearing in equation (8) , we get the desired result. □
Faltings isogeny theorem for number fields states that isogenous elliptic curves E E E and E ′ E' E ′ have a p ( E ) = a p ( E ′ ) a_{\mathfrak{p}}(E)=a_{\mathfrak{p}}(E') a p ( E ) = a p ( E ′ ) for all primes of good reduction p \mathfrak{p} p . Hence, if two elliptic curves are not isogenous, there exists a prime ideal p \mathfrak{p} p for which the traces are unequal. We give a bound on the norm of one such prime ideal. This bound will then be used to get a value of c ( E ) c(E) c ( E ) . The main goal of this section is to prove an effective version of Faltings isogeny theorem for number fields. To do so, we first establish several number-theoretic lemmas.
Some number theory
We first produce an integral basis using the embedding of the number field K K K in Euclidean space by means of the Minkowski map.
Lemma 4.2. Let K K K be a number field of degree [ K : Q ] = n K [K:\mathbb{Q}]=n_K [ K : Q ] = n K , and d K = ∣ Disc ( K / Q ) ∣ d_K=|\operatorname{Disc}(K/\mathbb{Q})| d K = ∣ Disc ( K / Q ) ∣ denote the absolute discriminant of K K K . Then O K \mathcal{O}_K O K admits an integral basis w 1 , … , w n K w_1,\ldots,w_{n_K} w 1 , … , w n K such that ∣ σ ( w j ) ∣ ≤ n K d K |\sigma(w_j)|\le n_K\sqrt{d_K} ∣ σ ( w j ) ∣ ≤ n K d K for every embedding σ : K ↪ C \sigma:K\hookrightarrow\mathbb{C} σ : K ↪ C and for every j = 1 , 2 , … , n K j=1,2,\ldots,n_K j = 1 , 2 , … , n K . Moreover,
∏ σ : K ↪ C ( ∑ j = 1 n K ∣ σ ( w j ) ∣ ) ≤ n K 2 n K d K n K / 2 . \prod_{\sigma:K\hookrightarrow\mathbb{C}}\left(\sum_{j=1}^{n_K}|\sigma(w_j)|\right)\le n_K^{2n_K}d_K^{n_K/2}. σ : K ↪ C ∏ ( j = 1 ∑ n K ∣ σ ( w j ) ∣ ) ≤ n K 2 n K d K n K /2 .
Proof. For n K = 1 n_K=1 n K = 1 , take the basis { 1 } \{1\} { 1 } . Now consider n K ≥ 2 n_K\ge2 n K ≥ 2 . Let r 1 r_1 r 1 be the number of real embeddings of K K K , and let r 2 r_2 r 2 be the number of conjugate pairs of nonreal embeddings. Thus n K = r 1 + 2 r 2 n_K=r_1+2r_2 n K = r 1 + 2 r 2 . Choose one embedding from each complex conjugate pair and consider the embedding
ι : K ⟶ R r 1 × C r 2 , α ⟼ ( σ 1 ( α ) , … , σ r 1 ( α ) , τ 1 ( α ) , … , τ r 2 ( α ) ) . \iota:K\longrightarrow\mathbb{R}^{r_1}\times\mathbb{C}^{r_2},\quad\alpha\longmapsto(\sigma_1(\alpha),\ldots,\sigma_{r_1}(\alpha),\tau_1(\alpha),\ldots,\tau_{r_2}(\alpha)). ι : K ⟶ R r 1 × C r 2 , α ⟼ ( σ 1 ( α ) , … , σ r 1 ( α ) , τ 1 ( α ) , … , τ r 2 ( α )) .
Now apply [22 Proposition 4.26, p. 79] to the unit ideal O K \mathcal{O}_K O K . Since N K / Q ( O K ) = 1 N_{K/\mathbb{Q}}(\mathcal{O}_K)=1 N K / Q ( O K ) = 1 , the image ι ( O K ) = : Λ \iota(\mathcal{O}_K)=:\Lambda ι ( O K ) =: Λ is a full lattice of covolume 2 − r 2 d K 2^{-r_2}\sqrt{d_K} 2 − r 2 d K . Consider the set B = [ − 1 , 1 ] r 1 × D r 2 B=[-1,1]^{r_1}\times D^{r_2} B = [ − 1 , 1 ] r 1 × D r 2 , where D D D is a closed unit disc in C \mathbb{C} C . Then it is compact, convex, symmetric about the origin, and has nonempty interior. Under the identification R r 1 × C r 2 ≅ R n K \mathbb{R}^{r_1} \times\mathbb{C}^{r_2} \cong\mathbb{R}^{n_K} R r 1 × C r 2 ≅ R n K via
( x 1 , … , x r 1 , z 1 , … , z r 2 ) ⟼ ( x 1 , … , x r 1 , Re z 1 , Im z 1 , … , Re z r 2 , Im z r 2 ) , (x_1,\ldots,x_{r_1},z_1,\ldots,z_{r_2}) \longmapsto(x_1,\ldots,x_{r_1},\operatorname{Re} z_1,\operatorname{Im} z_1,\ldots,\operatorname{Re} z_{r_2},\operatorname{Im} z_{r_2}), ( x 1 , … , x r 1 , z 1 , … , z r 2 ) ⟼ ( x 1 , … , x r 1 , Re z 1 , Im z 1 , … , Re z r 2 , Im z r 2 ) ,
let vol \operatorname{vol} vol denote n K n_K n K -dimensional Lebesgue measure. By repeatedly applying [11 Theorem 2.37(a), p. 67] the volume of B B B is the product of the lengths of its interval factors and the areas of its disc factors, hence vol ( B ) = 2 r 1 π r 2 \operatorname{vol}(B)=2^{r_1}\pi^{r_2} vol ( B ) = 2 r 1 π r 2 . Let λ 1 ≤ ⋯ ≤ λ n K \lambda_1\leq\cdots\leq\lambda_{n_K} λ 1 ≤ ⋯ ≤ λ n K be its successive minima with respect to Λ \Lambda Λ . Here λ i \lambda_i λ i is the smallest scaling factor such that λ i B \lambda_iB λ i B contains at least i i i linearly independent lattice vectors. First, we show that λ i ≥ 1 \lambda_i\geq1 λ i ≥ 1 (i = 1 , … , n K i=1,\ldots,n_K i = 1 , … , n K ). Indeed, if there exists nonzero α ∈ O K \alpha\in\mathcal{O}_K α ∈ O K such that ι ( α ) ∈ t B \iota(\alpha)\in tB ι ( α ) ∈ tB for some 0 < t < 1 0<t<1 0 < t < 1 , where t B = [ − t , t ] r 1 × ( t D ) r 2 tB=[-t,t]^{r_1}\times(tD)^{r_2} tB = [ − t , t ] r 1 × ( t D ) r 2 . This implies that vol ( t B ) = 2 r 1 π r 2 t n K \operatorname{vol}(tB)=2^{r_1}\pi^{r_2}t^{n_K} vol ( tB ) = 2 r 1 π r 2 t n K . Moreover, ∣ σ i ( α ) ∣ ≤ t |\sigma_i(\alpha)|\leq t ∣ σ i ( α ) ∣ ≤ t for all i = 1 , 2 , … , r 1 i=1,2,\ldots,r_1 i = 1 , 2 , … , r 1 and ∣ τ j ( α ) ∣ ≤ t |\tau_j(\alpha)|\leq t ∣ τ j ( α ) ∣ ≤ t for all j = 1 , … , r 2 j=1,\ldots,r_2 j = 1 , … , r 2 . Then
∣ N K / Q ( α ) ∣ = ∏ i = 1 r 1 ∣ σ i ( α ) ∣ ∏ j = 1 r 2 ∣ τ j ( α ) ∣ 2 ≤ t r 1 t 2 r 2 = t n K < 1 , |N_{K/\mathbb{Q}(\alpha)|=\prod_{i=1}^{r_1}|\sigma_i(\alpha)|\prod_{j=1}^{r_2}|\tau_j(\alpha)|^2\leq t^{r_1}t^{2r_2}=t^{n_K}<1,} ∣ N K / Q ( α ) ∣ = ∏ i = 1 r 1 ∣ σ i ( α ) ∣ ∏ j = 1 r 2 ∣ τ j ( α ) ∣ 2 ≤ t r 1 t 2 r 2 = t n K < 1 ,
which is a contradiction. Hence, by the definition of the first successive minimum, λ 1 ≥ 1 \lambda_1\geq1 λ 1 ≥ 1 and thus λ i ≥ 1 \lambda_i\geq1 λ i ≥ 1 for all i i i . Now apply Minkowski’s second body theorem [10 Theorem 2.9, 2nd Chapter, p. 24] we obtain
λ 1 ⋯ λ n K ≤ 2 n K det Λ vol ( B ) = 2 n K ⋅ 2 − r 2 d K 2 r 1 π r 2 = ( 2 π ) r 2 d K ≤ d K . \lambda_1\cdots\lambda_{n_K}\leq\frac{2^{n_K}\det\Lambda}{\operatorname{vol}(B)}=\frac{2^{n_K}\cdot2^{-r_2}\sqrt{d_K}}{2^{r_1}\pi^{r_2}}=\left(\frac{2}{\pi}\right)^{r_2}\sqrt{d_K}\leq\sqrt{d_K}. λ 1 ⋯ λ n K ≤ vol ( B ) 2 n K det Λ = 2 r 1 π r 2 2 n K ⋅ 2 − r 2 d K = ( π 2 ) r 2 d K ≤ d K .
Since all λ 1 , … , λ n K − 1 ≥ 1 \lambda_1,\ldots,\lambda_{n_K-1}\geq1 λ 1 , … , λ n K − 1 ≥ 1 , so we have λ n K ≤ d K \lambda_{n_K}\leq\sqrt{d_K} λ n K ≤ d K . Choose linearly independent lattice vectors v 1 , … , v n K v_1,\ldots,v_{n_K} v 1 , … , v n K with v i ∈ λ i B v_i\in\lambda_iB v i ∈ λ i B . Then by applying [10 Lemma 2.1, p. 12] we get a basis b 1 , … , b n K b_1,\ldots,b_{n_K} b 1 , … , b n K of Λ \Lambda Λ of the form b i = ∑ j = 1 n K a i j v j b_i=\sum_{j=1}^{n_K}a_{ij}v_j b i = ∑ j = 1 n K a ij v j , 0 ≤ a i j ≤ 1 0\leq a_{ij}\leq1 0 ≤ a ij ≤ 1 . The norm whose unit ball is B B B is the maximum of the absolute values of the real and complex coordinates. By the triangle inequality,
∥ b i ∥ B ≤ ∑ j = 1 n K a i j ∥ v j ∥ B ≤ ∑ j = 1 n K λ j ≤ n K λ n K ≤ n K d K . \|b_i\|_B\leq\sum_{j=1}^{n_K}a_{ij}\|v_j\|_B\leq\sum_{j=1}^{n_K}\lambda_j\leq n_K\lambda_{n_K}\leq n_K\sqrt{d_K}. ∥ b i ∥ B ≤ j = 1 ∑ n K a ij ∥ v j ∥ B ≤ j = 1 ∑ n K λ j ≤ n K λ n K ≤ n K d K .
Write b i = ι ( w i ) b_i=\iota(w_i) b i = ι ( w i ) for some w i ∈ O K w_i\in\mathcal{O}_K w i ∈ O K . Since ι \iota ι is injective and b i b_i b i ’s form a lattice basis of ι ( O K ) \iota(\mathcal{O}_K) ι ( O K ) , the elements w 1 , … , w n K w_1,\ldots,w_{n_K} w 1 , … , w n K form an integral basis of O K \mathcal{O}_K O K . The above estimation says ∣ σ ( w i ) ∣ ≤ n K d K |\sigma(w_i)|\leq n_K\sqrt{d_K} ∣ σ ( w i ) ∣ ≤ n K d K for all embeddings σ : K ↪ C \sigma:K\hookrightarrow\mathbb{C} σ : K ↪ C and for all i = 1 , … , n K i=1,\ldots,n_K i = 1 , … , n K . Consequently, for each embedding σ \sigma σ ,
∑ j = 1 n K ∣ σ ( w j ) ∣ ≤ n K 2 d K . \sum_{j=1}^{n_K}|\sigma(w_j)|\leq n_K^2\sqrt{d_K}. j = 1 ∑ n K ∣ σ ( w j ) ∣ ≤ n K 2 d K .
Thus
∏ σ : K ↪ C ( ∑ j = 1 n K ∣ σ ( w j ) ∣ ) ≤ n K 2 n K d K n K / 2 . \prod_{\sigma:K\hookrightarrow\mathbb{C}}\left(\sum_{j=1}^{n_K}|\sigma(w_j)|\right)\leq n_K^{2n_K}d_K^{n_K/2}. σ : K ↪ C ∏ ( j = 1 ∑ n K ∣ σ ( w j ) ∣ ) ≤ n K 2 n K d K n K /2 .
Using the special basis obtained in Lemma 4.2 , we produce an element of the number field whose norm satisfies a particular bound. This bound depends only on the number field K K K and the norms of a prescribed set of prime ideals.
Lemma 4.3. Let K K K be a number field of degree [ K : Q ] = n K [K:\mathbb{Q}]=n_K [ K : Q ] = n K , d K = ∣ Disc ( K / Q ) ∣ d_K=\lvert\operatorname{Disc}(K/\mathbb{Q})\rvert d K = ∣ Disc ( K / Q )∣ and let m = q 1 ⋯ q e \mathfrak{m}=\mathfrak{q}_1\cdots\mathfrak{q}_e m = q 1 ⋯ q e , e ≥ 1 e\ge1 e ≥ 1 , where q 1 , … , q e \mathfrak{q}_1,\ldots,\mathfrak{q}_e q 1 , … , q e are distinct nonzero prime ideals of O K \mathcal{O}_K O K . Then there exists α ∈ O K ∖ { 0 } \alpha\in\mathcal{O}_K\setminus\{0\} α ∈ O K ∖ { 0 } such that v q i ( α ) = 1 v_{\mathfrak{q}_i}(\alpha)=1 v q i ( α ) = 1 for 1 ≤ i ≤ e 1\le i\le e 1 ≤ i ≤ e and ∣ N K / Q ( α ) ∣ ≤ n K 2 n K d K n K / 2 ( N K / Q m ) 2 n K \lvert N_{K/\mathbb{Q}}(\alpha)\rvert\le n_K^{2n_K}d_K^{n_K/2}(N_{K/\mathbb{Q}}\mathfrak{m})^{2n_K} ∣ N K / Q ( α )∣ ≤ n K 2 n K d K n K /2 ( N K / Q m ) 2 n K .
Proof. By Lemma 4.2 , choose an integral basis w 1 , … , w n K w_1,\ldots,w_{n_K} w 1 , … , w n K of O K \mathcal{O}_K O K such that
∏ σ : K ↪ C ( ∑ j = 1 n K ∣ σ ( w j ) ∣ ) ≤ n K 2 n K d K n K / 2 . \prod_{\sigma:K\hookrightarrow\mathbb{C}}\left(\sum_{j=1}^{n_K}\lvert\sigma(w_j)\rvert\right)\le n_K^{2n_K}d_K^{n_K/2}. σ : K ↪ C ∏ ( j = 1 ∑ n K ∣ σ ( w j )∣ ) ≤ n K 2 n K d K n K /2 .
Since O K \mathcal{O}_K O K is a Dedekind domain, the unique factorization of nonzero ideals implies that q i 2 ⊆ q i \mathfrak{q}_i^2\subseteq\mathfrak{q}_i q i 2 ⊆ q i . Choose a i ∈ q i ∖ q i 2 a_i\in\mathfrak{q}_i\setminus\mathfrak{q}_i^2 a i ∈ q i ∖ q i 2 , then by Chinese remainder theorem, there exists α 0 ∈ O K \alpha_0\in\mathcal{O}_K α 0 ∈ O K such that α 0 ≡ a i ( m o d q i 2 ) \alpha_0\equiv a_i\pmod{\mathfrak{q}_i^2} α 0 ≡ a i ( mod q i 2 ) (i = 1 , … , e i=1,\ldots,e i = 1 , … , e ). Put u = N K / Q ( m 2 ) = ( N K / Q m ) 2 u=N_{K/\mathbb{Q}}(\mathfrak{m}^2)=(N_{K/\mathbb{Q}}\mathfrak{m})^2 u = N K / Q ( m 2 ) = ( N K / Q m ) 2 . Since the additive group O K / m 2 \mathcal{O}_K/\mathfrak{m}^2 O K / m 2 has order u u u , we have u O K ⊆ m 2 u\mathcal{O}_K\subseteq\mathfrak{m}^2 u O K ⊆ m 2 . Now write α 0 = ∑ j = 1 n K b j w j \alpha_0=\sum_{j=1}^{n_K}b_jw_j α 0 = ∑ j = 1 n K b j w j , b j ∈ Z b_j\in\mathbb{Z} b j ∈ Z , then for each j j j , the division algorithm gives integers t j , c j t_j,c_j t j , c j such that b j = u t j + c j b_j=ut_j+c_j b j = u t j + c j , 0 ≤ c j < u 0\le c_j<u 0 ≤ c j < u . Define α = ∑ j = 1 n K c j w j \alpha=\sum_{j=1}^{n_K}c_jw_j α = ∑ j = 1 n K c j w j , then
α − α 0 = ∑ j = 1 n K ( c j − b j ) w j = − u ∑ j = 1 n K t j w j . \alpha-\alpha_0=\sum_{j=1}^{n_K}(c_j-b_j)w_j=-u\sum_{j=1}^{n_K}t_jw_j. α − α 0 = j = 1 ∑ n K ( c j − b j ) w j = − u j = 1 ∑ n K t j w j .
Therefore,
α − α 0 ∈ u O K ⊆ m 2 = q 1 2 ⋯ q e 2 ⊆ q i 2 . \alpha-\alpha_0\in u\mathcal{O}_K\subseteq\mathfrak{m}^2=\mathfrak{q}_1^2\cdots\mathfrak{q}_e^2\subseteq\mathfrak{q}_i^2. α − α 0 ∈ u O K ⊆ m 2 = q 1 2 ⋯ q e 2 ⊆ q i 2 .
Thus α − a i = ( α − α 0 ) + ( α 0 − a i ) \alpha-a_i=(\alpha-\alpha_0)+(\alpha_0-a_i) α − a i = ( α − α 0 ) + ( α 0 − a i ) belongs to q i 2 \mathfrak{q}_i^2 q i 2 . Hence α ≡ a i ( m o d q i 2 ) \alpha\equiv a_i\pmod{\mathfrak{q}_i^2} α ≡ a i ( mod q i 2 ) (i = 1 , … , e i=1,\ldots,e i = 1 , … , e ). Thus α = a i + ( α − a i ) ∈ q i \alpha=a_i+(\alpha-a_i)\in\mathfrak{q}_i α = a i + ( α − a i ) ∈ q i . But α ∉ q i 2 \alpha\notin\mathfrak{q}_i^2 α ∈ / q i 2 . Suppose not, then a i = α − ( α − a i ) ∈ q i 2 a_i=\alpha-(\alpha-a_i)\in\mathfrak{q}_i^2 a i = α − ( α − a i ) ∈ q i 2 , contrary to our choice. Therefore α ∈ q i ∖ q i 2 \alpha\in\mathfrak{q}_i\setminus\mathfrak{q}_i^2 α ∈ q i ∖ q i 2 , so α ≠ 0 \alpha\ne0 α = 0 and v q i ( α ) = 1 v_{\mathfrak{q}_i}(\alpha)=1 v q i ( α ) = 1 . Let σ : K ↪ C \sigma:K\hookrightarrow\mathbb{C} σ : K ↪ C be an embedding, then σ ( α ) = ∑ j = 1 n K c j σ ( w j ) \sigma(\alpha)=\sum_{j=1}^{n_K}c_j\sigma(w_j) σ ( α ) = ∑ j = 1 n K c j σ ( w j ) and hence
∣ σ ( α ) ∣ ≤ ∑ j = 1 n K c j ∣ σ ( w j ) ∣ ≤ u ∑ j = 1 n K ∣ σ ( w j ) ∣ . \lvert\sigma(\alpha)\rvert\le\sum_{j=1}^{n_K}c_j\lvert\sigma(w_j)\rvert\le u\sum_{j=1}^{n_K}\lvert\sigma(w_j)\rvert. ∣ σ ( α )∣ ≤ j = 1 ∑ n K c j ∣ σ ( w j )∣ ≤ u j = 1 ∑ n K ∣ σ ( w j )∣ .
Now
∣ N K / Q ( α ) ∣ = ∏ σ : K ↪ C ∣ σ ( α ) ∣ ≤ ∏ σ : K ↪ C ( u ∑ j = 1 n K ∣ σ ( w j ) ∣ ) ≤ u n K n K 2 n K d K n K / 2 = n K 2 n K d K n K / 2 ( N K / Q m ) 2 n K . \begin{aligned}
\lvert N_{K/\mathbb{Q}}(\alpha)\rvert=\prod_{\sigma:K\hookrightarrow\mathbb{C}}\lvert\sigma(\alpha)\rvert\le\prod_{\sigma:K\hookrightarrow\mathbb{C}}\left(u\sum_{j=1}^{n_K}\lvert\sigma(w_j)\rvert\right) \\
&\le u^{n_K}n_K^{2n_K}d_K^{n_K/2}=n_K^{2n_K}d_K^{n_K/2}(N_{K/\mathbb{Q}}\mathfrak{m})^{2n_K}.
\end{aligned} ∣ N K / Q ( α )∣ = σ : K ↪ C ∏ ∣ σ ( α )∣ ≤ σ : K ↪ C ∏ ( u j = 1 ∑ n K ∣ σ ( w j )∣ ) ≤ u n K n K 2 n K d K n K /2 = n K 2 n K d K n K /2 ( N K / Q m ) 2 n K .
Remark 4.2. Note that the α \alpha α obtained in Lemma 4.3 is compatible with Lemma 3.3 and Lemma 3.4 .
Effective Isogeny Theorem over Galois Number Fields
We use the element produced in Lemma 4.3 , in conjunction with Lemma 3.3 to produce a field extension in which a prescribed set of prime ideals ramify. Finally, we use the modified Bach and Sorenson Theorem 3.4 to get a prime ideal outside the prescribed set. The prime 2 appears in the following theorem as we used the 2-adic representation associated to the elliptic curve to get the bound in the first line of the proof. We couple this with Remark 4.1 to exclude the primes of bad reduction.
Theorem 4.4 (Effective Isogeny Theorem over Number Fields). Assume GRH. Let K K K be a Galois number field of degree n K n_K n K over Q \mathbb{Q} Q , and let E 1 , E 2 E_1,E_2 E 1 , E 2 be elliptic curves without complex multiplication defined over K K K . Suppose that E 1 E_1 E 1 and E 2 E_2 E 2 are not K K K -isogenous. Then there exists a prime ideal p ⊆ O K \mathfrak{p}\subseteq\mathcal{O}_K p ⊆ O K of good reduction for both E 1 E_1 E 1 and E 2 E_2 E 2 such that
a p ( E 1 ) ≠ a p ( E 2 ) , a_{\mathfrak{p}}(E_1) \ne a_{\mathfrak{p}}(E_2), a p ( E 1 ) = a p ( E 2 ) , and satisfying
N K / Q p ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 N E 2 ) + C 2 ( K ) ) 2 , N_{K/\mathbb{Q}}\mathfrak{p}\leq\left(C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_{E_1}\mathfrak{N}_{E_2})+C_2(K)\right)^2, N K / Q p ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 N E 2 ) + C 2 ( K ) ) 2 , where N E i \mathfrak{N}_{E_i} N E i denotes the conductor ideal of E i E_i E i , and C 1 ( K ) , C 2 ( K ) C_1(K),C_2(K) C 1 ( K ) , C 2 ( K ) are constants depending only on K K K . The constants are explicitly given as
C 1 ( K ) = 510.352 + 513.024 n K , C_1(K)=510.352+513.024n_K, C 1 ( K ) = 510.352 + 513.024 n K , C 2 ( K ) = 513.024 ( 1 + n K 4 log d K + n K log ( 384 n K ) ) + 1324.8 n K + 45.7 , C_2(K)=513.024\left(1+\frac{n_K}{4}\log d_K+n_K\log(384n_K)\right)+1324.8n_K+45.7, C 2 ( K ) = 513.024 ( 1 + 4 n K log d K + n K log ( 384 n K ) ) + 1324.8 n K + 45.7 , where d K d_K d K denotes the absolute discriminant of the number field K K K .
Proof. We follow the same proof as in Mayle and Wang to conclude that [ L : K ] ≤ 192 [L:K]\leq192 [ L : K ] ≤ 192 [20 Proof of Theorem 2, p. 1408] .
Define
m : = rad ( 2 O K N E 1 N E 2 ) = ∏ q ∣ 2 O K N E 1 N E 2 q . m:=\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_{E_1}\mathfrak{N}_{E_2})=\prod_{\mathfrak{q}\mid2\mathcal{O}_K\mathfrak{N}_{E_1}\mathfrak{N}_{E_2}}\mathfrak{q}. m := rad ( 2 O K N E 1 N E 2 ) = q ∣ 2 O K N E 1 N E 2 ∏ q .
and let S : = { q ⊆ O K : q ∣ m } S:=\{\mathfrak{q}\subseteq\mathcal{O}_K:\mathfrak{q}\mid m\} S := { q ⊆ O K : q ∣ m } . By Lemma 4.3 , choose α ∈ O K ∖ { 0 } \alpha\in\mathcal{O}_K\setminus\{0\} α ∈ O K ∖ { 0 } satisfying v q ( α ) = 1 v_{\mathfrak{q}}(\alpha)=1 v q ( α ) = 1 for every q ∈ S \mathfrak{q}\in S q ∈ S , and
log ∣ N K / Q ( α ) ∣ ≤ 2 n K log n K + n K 2 log d K + 2 n K log N K / Q m . (9) \log\lvert N_{K/\mathbb{Q}}(\alpha)\rvert\leq2n_K\log n_K+\frac{n_K}{2}\log d_K+2n_K\log N_{K/\mathbb{Q}}m. \tag*{(9)} log ∣ N K / Q ( α )∣ ≤ 2 n K log n K + 2 n K log d K + 2 n K log N K / Q m . (9)
Fix this α \alpha α and put M = K ( α ) M=K(\sqrt{\alpha}) M = K ( α ) and L ~ = L M \widetilde{L}=LM L = L M . Since v q = 1 v_{\mathfrak{q}}=1 v q = 1 for every q ∈ S \mathfrak{q}\in S q ∈ S , Lemma 3.3 implies that every prime ideal q ∈ S \mathfrak{q}\in S q ∈ S ramifies in M / K M/K M / K . Applying Lemma 3.4 with the same α \alpha α , we obtain a prime ideal p ⊆ O K \mathfrak{p}\subseteq\mathcal{O}_K p ⊆ O K such that p ∉ S \mathfrak{p}\notin S p ∈ / S , p \mathfrak{p} p is unramified in L / K L/K L / K , Frob p , L / K ⊆ C \operatorname{Frob}_{\mathfrak{p},L/K}\subseteq C Frob p , L / K ⊆ C , and
N K / Q p ≤ ( 4 log d L ~ + 2.5 [ L ~ : Q ] + 5 ) 2 . N_{K/\mathbb{Q}}\mathfrak{p}\leq\left(4\log d_{\widetilde{L}}+2.5[\widetilde{L}:\mathbb{Q}]+5\right)^2. N K / Q p ≤ ( 4 log d L + 2.5 [ L : Q ] + 5 ) 2 .
Since p ∉ S \mathfrak{p}\notin S p ∈ / S , we have p ∤ 2 O K N E 1 N E 2 \mathfrak{p}\nmid2\mathcal{O}_K\mathfrak{N}_{E_1}\mathfrak{N}_{E_2} p ∤ 2 O K N E 1 N E 2 . Therefore, p \mathfrak{p} p is not above 2, and both E 1 E_1 E 1 and E 2 E_2 E 2 have good reduction at p \mathfrak{p} p . So again by [20 Proposition 12, part (3)] , we have
tr ρ 1 ( Frob p ) ≠ tr ρ 2 ( Frob p ) . \operatorname{tr}\rho_1(\operatorname{Frob}_{\mathfrak{p}})\ne\operatorname{tr}\rho_2(\operatorname{Frob}_{\mathfrak{p}}). tr ρ 1 ( Frob p ) = tr ρ 2 ( Frob p ) .
Hence α p ( E 1 ) ≠ α p ( E 2 ) \alpha_p(E_1) \ne\alpha_p(E_2) α p ( E 1 ) = α p ( E 2 ) . It remains to prove the required bound. Furthermore, [ L ~ : K ] ≤ 2 [ L : K ] ≤ 384 [\widetilde{L}:K] \le2[L:K] \le384 [ L : K ] ≤ 2 [ L : K ] ≤ 384 . Therefore, [ L ~ : Q ] = [ L ~ : K ] n K ≤ 384 n K [\widetilde{L}:\mathbb{Q}] = [\widetilde{L}:K]n_K \le384n_K [ L : Q ] = [ L : K ] n K ≤ 384 n K . Thus, if log d ~ L ≤ 100 \log\widetilde{d}_L \le100 log d L ≤ 100 , then by applying the effective Chebotarev bound 3.2 with the constants 4 4 4 , 2.5 2.5 2.5 , and 5 5 5 , we find that
N K / Q p ≤ ( 4 ⋅ 100 + 2.5 ⋅ 384 n K + 5 ) 2 = ( 405 + 960 n K ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \le(4 \cdot100 + 2.5 \cdot384n_K + 5)^2 = (405 + 960n_K)^2. N K / Q p ≤ ( 4 ⋅ 100 + 2.5 ⋅ 384 n K + 5 ) 2 = ( 405 + 960 n K ) 2 .
If 100 ≤ log d ~ L ≤ 1000 100 \le\log\widetilde{d}_L \le1000 100 ≤ log d L ≤ 1000 , taking the maximum of each coefficient over the degree columns in the corresponding row of [1 Table 1, p. 1730] , we obtain
N K / Q p ≤ ( 1.796 ⋅ 1000 + 1.89 ⋅ 384 n K + 6.3 ) 2 = ( 1796 + 725.76 n K + 6.3 ) 2 = ( 1802.3 + 725.76 n K ) 2 . \begin{aligned}
N_{K/\mathbb{Q}}\mathfrak{p} &\le(1.796 \cdot1000 + 1.89 \cdot384n_K + 6.3)^2 \\
&= (1796 + 725.76n_K + 6.3)^2 \\
&= (1802.3 + 725.76n_K)^2.
\end{aligned} N K / Q p ≤ ( 1.796 ⋅ 1000 + 1.89 ⋅ 384 n K + 6.3 ) 2 = ( 1796 + 725.76 n K + 6.3 ) 2 = ( 1802.3 + 725.76 n K ) 2 .
If log d ~ L ≥ 1000 \log\widetilde{d}_L \ge1000 log d L ≥ 1000 , then again using [1 Table 1, p. 1730] , we obtain
N K / Q p ≤ ( 1.336 log d ~ L + 3.45 ⋅ 384 n K + 45.7 ) 2 = ( 1.336 log d ~ L + 1324.8 n K + 45.7 ) 2 . \begin{aligned}
N_{K/\mathbb{Q}}\mathfrak{p} &\le(1.336\log\widetilde{d}_L + 3.45 \cdot384n_K + 45.7)^2 \\
&= (1.336\log\widetilde{d}_L + 1324.8n_K + 45.7)^2.
\end{aligned} N K / Q p ≤ ( 1.336 log d L + 3.45 ⋅ 384 n K + 45.7 ) 2 = ( 1.336 log d L + 1324.8 n K + 45.7 ) 2 .
Therefore, in all cases,
N K / Q p ≤ max { ( 405 + 960 n K ) 2 , ( 1802.3 + 725.76 n K ) 2 , ( 1.336 log d ~ L + 1324.8 n K + 45.7 ) 2 } . (10) \begin{aligned}
N_{K/\mathbb{Q}}\mathfrak{p} \le\max\{&(405 + 960n_K)^2,\\
&(1802.3 + 725.76n_K)^2,\\
&(1.336\log\widetilde{d}_L + 1324.8n_K + 45.7)^2\}. \tag*{(10)}
\end{aligned} N K / Q p ≤ max { ( 405 + 960 n K ) 2 , ( 1802.3 + 725.76 n K ) 2 , ( 1.336 log d L + 1324.8 n K + 45.7 ) 2 } . (10)
Now it remains to bound log d ~ L \log\widetilde{d}_L log d L . From Lemma 3.5 , we have
log d L ≤ [ L : K ] log d K + ( [ L : K ] − 1 ) log rad K ( d L / K ) + [ L : K ] n K log [ L : K ] . \log d_L \le[L:K]\log d_K + ([L:K]-1)\log\operatorname{rad}_K(\mathfrak{d}_{L/K}) + [L:K]n_K\log[L:K]. log d L ≤ [ L : K ] log d K + ([ L : K ] − 1 ) log rad K ( d L / K ) + [ L : K ] n K log [ L : K ] .
Recall that rad K ( d L / K ) = ∏ q ∣ d L / K N K / Q q \operatorname{rad}_K(\mathfrak{d}_{L/K}) = \prod_{\mathfrak{q}\mid\mathfrak{d}_{L/K}}N_{K/\mathbb{Q}}\mathfrak{q} rad K ( d L / K ) = ∏ q ∣ d L / K N K / Q q . A prime ideal divides the relative discriminant precisely when it ramifies in L / K L/K L / K [22 Remark 3.39(b), p. 62] . Since L / K L/K L / K is unramified outside S S S , every prime ideal dividing d L / K \mathfrak{d}_{L/K} d L / K belongs to S S S . Therefore
rad K ( d L / K ) ≤ ∏ q ∈ S N K / Q q = N K / Q ( q 1 ⋯ q e ) = N K / Q m . \operatorname{rad}_K(\mathfrak{d}_{L/K}) \le\prod_{\mathfrak{q}\in S}N_{K/\mathbb{Q}}\mathfrak{q} = N_{K/\mathbb{Q}}(\mathfrak{q}_1\cdots\mathfrak{q}_e) = N_{K/\mathbb{Q}}\mathfrak{m}. rad K ( d L / K ) ≤ q ∈ S ∏ N K / Q q = N K / Q ( q 1 ⋯ q e ) = N K / Q m .
Hence log rad K ( d L / K ) ≤ log N K / Q m \log\operatorname{rad}_K(\mathfrak{d}_{L/K}) \le\log N_{K/\mathbb{Q}}\mathfrak{m} log rad K ( d L / K ) ≤ log N K / Q m . Consequently,
log d L ≤ 192 log d K + 191 log N K / Q m + 192 n K log 192. (11) \log d_L \le192\log d_K + 191\log N_{K/\mathbb{Q}}\mathfrak{m} + 192n_K\log192. \tag*{(11)} log d L ≤ 192 log d K + 191 log N K / Q m + 192 n K log 192. (11)
Suppose first that [ L ~ : L ] = 2 [\widetilde{L}:L] = 2 [ L : L ] = 2 , i.e. α ∉ L \sqrt{\alpha} \notin L α ∈ / L . Then the discriminant of the L L L -basis { 1 , α } \{1,\sqrt{\alpha}\} { 1 , α } is 4 α 4\alpha 4 α . So the relative field discriminant divides the discriminant of this order, i.e., d ~ L / L ∣ ( 4 α ) O L \widetilde{\mathfrak{d}}_{L/L}\mid(4\alpha)\mathcal{O}_L d L / L ∣ ( 4 α ) O L , which implies ( 4 α ) O L = d ~ L / L a (4\alpha)\mathcal{O}_L = \widetilde{\mathfrak{d}}_{L/L}\mathfrak{a} ( 4 α ) O L = d L / L a for some nonzero integral ideal a \mathfrak{a} a of O L \mathcal{O}_L O L . Now apply [24 Theorem 6.1, p. 11] to Q ⊆ L ⊆ L ~ \mathbb{Q}\subseteq L\subseteq\widetilde{L} Q ⊆ L ⊆ L . Since [ L ~ : L ] = 2 [\widetilde{L}:L] = 2 [ L : L ] = 2 , it gives d L ~ = d L 2 N L / Q ( d ~ L / L ) d_{\widetilde{L}} = d_L^2N_{L/\mathbb{Q}}(\widetilde{\mathfrak{d}}_{L/L}) d L = d L 2 N L / Q ( d L / L ) . Consequently, N L / Q ( ( 4 α ) O L ) = N L / Q ( d ~ L / L ) N L / Q ( a ) N_{L/\mathbb{Q}}((4\alpha)\mathcal{O}_L) = N_{L/\mathbb{Q}}(\widetilde{\mathfrak{d}}_{L/L})N_{L/\mathbb{Q}}(\mathfrak{a}) N L / Q (( 4 α ) O L ) = N L / Q ( d L / L ) N L / Q ( a ) . Thus
N L / Q ( d ~ L / L ) ≤ N L / Q ( ( 4 α ) O L ) = ∣ N L / Q ( 4 α ) ∣ . N_{L/\mathbb{Q}}(\widetilde{\mathfrak{d}}_{L/L}) \le N_{L/\mathbb{Q}}((4\alpha)\mathcal{O}_L) = \lvert N_{L/\mathbb{Q}}(4\alpha)\rvert. N L / Q ( d L / L ) ≤ N L / Q (( 4 α ) O L ) = ∣ N L / Q ( 4 α )∣ .
The above follows from [22 Proposition 4.1, Proposition 4.2(a), p. 68-89] . So that
d ~ L ≤ d L 2 ∣ N L / Q ( 4 α ) ∣ . (12) \widetilde{d}_L \le d_L^2 \lvert N_{L/\mathbb{Q}}(4\alpha)\rvert. \tag*{(12)} d L ≤ d L 2 ∣ N L / Q ( 4 α )∣ . (12)
Now by transitivity of field norms, for Q ⊆ K ⊆ L \mathbb{Q} \subseteq K \subseteq L Q ⊆ K ⊆ L , we have
N L / Q ( 4 α ) = N K / Q ( N L / K ( 4 α ) ) = N K / Q ( ( 4 α ) [ L : K ] ) = ( N K / Q ( 4 α ) ) [ L : K ] = ( 4 n K N K / Q ( α ) ) [ L : K ] . \begin{aligned}
N_{L/\mathbb{Q}}(4\alpha) &= N_{K/\mathbb{Q}}(N_{L/K}(4\alpha)) = N_{K/\mathbb{Q}}((4\alpha)^{[L:K]}) \\
&= (N_{K/\mathbb{Q}}(4\alpha))^{[L:K]} = (4^{n_K}N_{K/\mathbb{Q}}(\alpha))^{[L:K]}.
\end{aligned} N L / Q ( 4 α ) = N K / Q ( N L / K ( 4 α )) = N K / Q (( 4 α ) [ L : K ] ) = ( N K / Q ( 4 α ) ) [ L : K ] = ( 4 n K N K / Q ( α ) ) [ L : K ] .
Now, from equation (12) , we get
log d ~ L ≤ 2 log d L + [ L : K ] ( n K log 4 + log ∣ N K / Q ( α ) ∣ ) . (13) \log\widetilde{d}_L \le2\log d_L + [L:K]\left(n_K\log4 + \log\lvert N_{K/\mathbb{Q}}(\alpha)\rvert\right). \tag*{(13)} log d L ≤ 2 log d L + [ L : K ] ( n K log 4 + log ∣ N K / Q ( α )∣ ) . (13)
If α ∈ L \sqrt{\alpha}\in L α ∈ L , then L ~ = L \widetilde{L}=L L = L . Since d L d_L d L and ∣ N K / Q ( α ) ∣ \lvert N_{K/\mathbb{Q}}(\alpha)\rvert ∣ N K / Q ( α )∣ are positive integers, we have
d ~ L = d L ≤ d L 2 4 [ L : K ] n K ∣ N K / Q ( α ) ∣ [ L : K ] . \widetilde{d}_L=d_L\le d_L^2 4^{[L:K]n_K}\lvert N_{K/\mathbb{Q}}(\alpha)\rvert^{[L:K]}. d L = d L ≤ d L 2 4 [ L : K ] n K ∣ N K / Q ( α ) ∣ [ L : K ] .
So equation (13) remains valid. Now by using equations (13) , (11) , and (9) we obtain
log d ~ L ≤ 384 log d K + 382 log N K / Q m + 384 n K log 384 + 384 n K log n K + 96 n K log d K + 384 n K log N K / Q m = ( 382 + 384 n K ) log N K / Q m + ( 384 + 96 n K ) log d K + 384 n K log ( 384 n K ) = : R \begin{aligned}
\log\widetilde{d}_L &\le384\log d_K + 382\log N_{K/\mathbb{Q}}^m \\
&\quad+ 384n_K\log384 + 384n_K\log n_K \\
&\quad+ 96n_K\log d_K + 384n_K\log N_{K/\mathbb{Q}}^m \\
&= (382+384n_K)\log N_{K/\mathbb{Q}}^m +(384+96n_K)\log d_K +384n_K\log(384n_K) \\
&=:R
\end{aligned} log d L ≤ 384 log d K + 382 log N K / Q m + 384 n K log 384 + 384 n K log n K + 96 n K log d K + 384 n K log N K / Q m = ( 382 + 384 n K ) log N K / Q m + ( 384 + 96 n K ) log d K + 384 n K log ( 384 n K ) =: R
Set T : = 1.336 R + 1324.8 n K + 45.7 T:=1.336R+1324.8n_K+45.7 T := 1.336 R + 1324.8 n K + 45.7 , since n K ≥ 1 n_K\ge1 n K ≥ 1 , d K ≥ 1 d_K\ge1 d K ≥ 1 , and N K / Q m ≥ 1 N_{K/\mathbb{Q}}^m\ge1 N K / Q m ≥ 1 , we have
R ≥ 384 n K log ( 384 n K ) > 384 n K log 384 > 1920 n K . R\ge384n_K\log(384n_K)>384n_K\log384>1920n_K. R ≥ 384 n K log ( 384 n K ) > 384 n K log 384 > 1920 n K .
Therefore, T > 3889.92 n K + 45.7 T>3889.92n_K+45.7 T > 3889.92 n K + 45.7 . In particular, 3889.92 n K + 45.7 3889.92n_K+45.7 3889.92 n K + 45.7 is larger than both 405 + 960 n K 405+960n_K 405 + 960 n K and 1802.3 + 725.76 n K 1802.3+725.76n_K 1802.3 + 725.76 n K . Moreover, log d ~ L ≤ R \log\widetilde{d}_L\le R log d L ≤ R gives
1.336 log d ~ L + 1324.8 n K + 45.7 ≤ T . 1.336\log\widetilde{d}_L+1324.8n_K+45.7\le T. 1.336 log d L + 1324.8 n K + 45.7 ≤ T .
So each of the bounds in equation (10) is at most T 2 T^2 T 2 . Therefore, N K / Q ≤ T 2 N_{K/\mathbb{Q}}\le T^2 N K / Q ≤ T 2 . Now expand T T T ,
T = 1.336 ( 382 + 384 n K ) log N K / Q m + ( 384 + 96 n K ) log d K + 384 n K log ( 384 n K ) + 1324.8 n K + 45.7 = ( 510.352 + 513.024 n K ) log N K / Q m + ⏟ : = C 1 ( K ) ( 513.024 + 128.256 n K ) log d K + 513.024 n K log ( 384 n K ) + 1324.8 n K + 45.7. \begin{aligned}
T={}&1.336(382+384n_K)\log N_{K/\mathbb{Q}}^m +(384+96n_K)\log d_K \\
&\quad+384n_K\log(384n_K)+1324.8n_K+45.7 \\
={}&(510.352+513.024n_K)\log N_{K/\mathbb{Q}}^m \\
&\quad\underbrace{{}+}_{\mathrel{\phantom{:=C_1(K)}}} \\
&\quad(513.024+128.256n_K)\log d_K+513.024n_K\log(384n_K)+1324.8n_K+45.7.
\end{aligned} T = = 1.336 ( 382 + 384 n K ) log N K / Q m + ( 384 + 96 n K ) log d K + 384 n K log ( 384 n K ) + 1324.8 n K + 45.7 ( 510.352 + 513.024 n K ) log N K / Q m := C 1 ( K ) + ( 513.024 + 128.256 n K ) log d K + 513.024 n K log ( 384 n K ) + 1324.8 n K + 45.7.
= : = C 1 ( K ) \phantom{=}:=C_1(K) = := C 1 ( K )
( 513.024 + 128.256 n K ) log d K + 513.024 n K log ( 384 n K ) + 1324.8 n K + 45.7 ⏟ : = C 2 ( K ) \underbrace{(513.024+128.256n_K)\log d_K+513.024n_K\log(384n_K)+1324.8n_K+45.7}_{:=C_2(K)} := C 2 ( K ) ( 513.024 + 128.256 n K ) log d K + 513.024 n K log ( 384 n K ) + 1324.8 n K + 45.7
Consequently,
N K / Q p ≤ ( C 1 ( K ) log N K / Q m + C 2 ( K ) ) 2 . N_{K/\mathbb{Q}}\mathfrak{p} \leq\left(C_1(K)\log N_{K/\mathbb{Q}}\mathfrak{m}+C_2(K)\right)^2. N K / Q p ≤ ( C 1 ( K ) log N K / Q m + C 2 ( K ) ) 2 .
Proof of the Main Theorem
In this section, we finally prove the main theorem. We first establish some properties of quadratic characters.
Quadratic characters
Let χ D \chi_D χ D denote the quadratic Galois character, that is
χ D : G K → { ± 1 } \chi_D:G_K\to\{\pm1\} χ D : G K → { ± 1 }
χ D ( σ ) = { 1 , if σ ( D ) = D , − 1 , if σ ( D ) = − D . \chi_D(\sigma)=
\begin{cases}
1, & \text{if }\sigma(\sqrt{D})=\sqrt{D},\\
-1, & \text{if }\sigma(\sqrt{D})=-\sqrt{D}.
\end{cases} χ D ( σ ) = { 1 , − 1 , if σ ( D ) = D , if σ ( D ) = − D .
Lemma 5.1. χ D \chi_D χ D is unramified at p \mathfrak{p} p if and only if K ( D ) / K K(\sqrt{D})/K K ( D ) / K is unramified at p \mathfrak{p} p , where p \mathfrak{p} p is a prime ideal of O K ( D ) \mathcal{O}_{K(\sqrt{D})} O K ( D ) lying over p \mathfrak{p} p .
Proof. Suppose that χ D \chi_D χ D is unramified at p \mathfrak{p} p , that is, χ D ( I p ) = 1 \chi_D(I_{\mathfrak{p}})=1 χ D ( I p ) = 1 . We also have res D ( I p ) = I p ( D ) \operatorname{res}_D(I_{\mathfrak{p}})=I_{\mathfrak{p}}(D) res D ( I p ) = I p ( D ) , and hence, # I p ( D ) = 1 \#I_{\mathfrak{p}}(D)=1 # I p ( D ) = 1 . This implies that e ( K ( D ) / K ) = 1 e(K(\sqrt{D})/K)=1 e ( K ( D ) / K ) = 1 since e ( K ( D ) / K ) = # I p ( D ) e(K(\sqrt{D})/K)=\#I_{\mathfrak{p}}(D) e ( K ( D ) / K ) = # I p ( D ) . Hence, χ D \chi_D χ D is unramified at p \mathfrak{p} p .
Conversely, assume that K ( D ) / K K(\sqrt{D})/K K ( D ) / K is unramified at the prime p \mathfrak{p} p . Given σ ∈ I p \sigma\in I_{\mathfrak{p}} σ ∈ I p , let σ ‾ \overline{\sigma} σ denote its restriction to I p ( D ) I_{\mathfrak{p}}(D) I p ( D ) . Since I p ( D ) = 1 I_{\mathfrak{p}}(D)=1 I p ( D ) = 1 , we have σ ‾ ( D ) = D \overline{\sigma}(\sqrt{D})=\sqrt{D} σ ( D ) = D , which implies that χ D ( σ ) = 1 \chi_D(\sigma)=1 χ D ( σ ) = 1 . Therefore, χ D \chi_D χ D is unramified at p \mathfrak{p} p .
If p ∤ D \mathfrak{p}\nmid D p ∤ D , then
( p K ( D ) / K ) ( D ) = ( D p ) 2 D , (14) \left(\frac{\mathfrak{p}}{K(\sqrt{D})/K}\right)(\sqrt{D})
=
\left(\frac{D}{\mathfrak{p}}\right)_2\sqrt{D}, \tag*{(14)} ( K ( D ) / K p ) ( D ) = ( p D ) 2 D , (14)
where the term on the left is the Artin Symbol and on the right we consider the generalized Legendre symbol. (See [8 Exercise 5.13] for precise definitions.)
Quadratic Twists
Now consider an elliptic curve E / K E/K E / K given by the Weierstrass equation
y 2 + a 1 x y + a 3 y = x 3 + a 2 x 2 + a 4 x + a 6 y^2+a_1xy+a_3y=x^3+a_2x^2+a_4x+a_6 y 2 + a 1 x y + a 3 y = x 3 + a 2 x 2 + a 4 x + a 6
and let E D E_D E D be the quadratic twist of E E E by χ D \chi_D χ D (See [7 p.48-49] for more on quadratic twists.) The discriminant of E D E_D E D is D 6 D^6 D 6 times the discriminant of E E E . From this we can conclude that
rad ( N E D ) divides rad ( 2 N E D ) . (15) \operatorname{rad}(\mathfrak{N}_{E_D})\text{ divides }\operatorname{rad}(2\mathfrak{N}_{E_D}). \tag*{(15)} rad ( N E D ) divides rad ( 2 N E D ) . (15)
If we consider a prime p ∤ N K / Q ( D N E ) \mathfrak{p}\nmid N_{K/\mathbb{Q}}(D\mathfrak{N}_E) p ∤ N K / Q ( D N E ) , and let p \mathfrak{p} p be any prime lying over p \mathfrak{p} p , we have by [31 Exercise 4.10] that
a p ( E D ) = ( D q ) 2 a p ( E ) , a_{\mathfrak{p}}(E_D)=\left(\frac{D}{\mathfrak{q}}\right)_2a_{\mathfrak{p}}(E), a p ( E D ) = ( q D ) 2 a p ( E ) ,
where q = p f \mathfrak{q}=\mathfrak{p}^{f} q = p f where f f f denotes the inertia degree of p \mathfrak{p} p . From equation (14) above, we have,
a p ( E D ) = χ D ( Frob p ) a p ( E ) . a_{\mathfrak{p}}(E_D)=\chi_D(\operatorname{Frob}_{\mathfrak{p}})a_{\mathfrak{p}}(E). a p ( E D ) = χ D ( Frob p ) a p ( E ) .
Lemma 5.2. If D ≠ 1 D \ne1 D = 1 is squarefree and E / K E/K E / K has no complex multiplication, then E E E and E D E_D E D are not K K K -isogenous.
Proof. The density of primes p ∈ O K \mathfrak{p} \in\mathcal{O}_K p ∈ O K for which χ λ ( Frob p ) = − 1 \chi_\lambda(\operatorname{Frob}_{\mathfrak{p}})=-1 χ λ ( Frob p ) = − 1 is 1 2 \frac{1}{2} 2 1 by the Chebotarev density theorem. For any such prime, we either have a p ( E ) = − a p ( E D ) a_{\mathfrak{p}}(E)=-a_{\mathfrak{p}}(E_D) a p ( E ) = − a p ( E D ) or a p ( E ) = 0 a_{\mathfrak{p}}(E)=0 a p ( E ) = 0 . By a generalization of a theorem by Serre, we see that the density of primes for which p ∤ 2 N E D \mathfrak{p} \nmid2\mathfrak{N}_{E_D} p ∤ 2 N E D and a p ( E ) = 0 a_{\mathfrak{p}}(E)=0 a p ( E ) = 0 is 0 [5 Corollary 6.2] . Thus, there exists a prime p ∤ 2 N E D \mathfrak{p} \nmid2\mathfrak{N}_{E_D} p ∤ 2 N E D such that a p ( E ) = − a p ( E D ) a_{\mathfrak{p}}(E)=-a_{\mathfrak{p}}(E_D) a p ( E ) = − a p ( E D ) and hence E E E and E D E_D E D are not K K K -isogenous. □
A Special character
The construction of this character dates back to Serre [26 p. 295] . We consider the following Galois character given by the composition
ε ℓ : Gal ( K ‾ / K ) → Im ( ρ E , ℓ ) → C n s + ( ℓ ) C n s ( ℓ ) → ∼ { ± 1 } . (16) \varepsilon_\ell: \operatorname{Gal}(\overline{K}/K) \to\operatorname{Im}(\rho_{E,\ell}) \to\frac{C_{ns}^{+}(\ell)}{C_{ns}(\ell)} \xrightarrow{\sim} \{\pm1\}. \tag*{(16)} ε ℓ : Gal ( K / K ) → Im ( ρ E , ℓ ) → C n s ( ℓ ) C n s + ( ℓ ) ∼ { ± 1 } . (16)
Lemma 5.3. Let ε ℓ \varepsilon_\ell ε ℓ be as above. We have the following properties
(i) For each prime p ∤ N E \mathfrak{p} \nmid\mathfrak{N}_E p ∤ N E , ε ℓ \varepsilon_\ell ε ℓ is unramified at p \mathfrak{p} p .
(ii) One has ε ℓ = χ D \varepsilon_\ell=\chi_D ε ℓ = χ D for some D ∣ N E D \mid\mathfrak{N}_E D ∣ N E .
(iii) For each prime p ∤ N E \mathfrak{p} \nmid\mathfrak{N}_E p ∤ N E , if ε ℓ ( Frob p ) = − 1 \varepsilon_\ell(\operatorname{Frob}_{\mathfrak{p}})=-1 ε ℓ ( Frob p ) = − 1 , then a p ( E ) ≡ 0 ( m o d ℓ ) a_{\mathfrak{p}}(E)\equiv0 \pmod{\ell} a p ( E ) ≡ 0 ( mod ℓ ) .
Proof. We refer the reader to [20 Lemma 15, p. 1412] for a proof. In our case, we replace the prime p \mathfrak{p} p with a prime ideal p \mathfrak{p} p and the conductor N E \mathfrak{N}_E N E with the conductor ideal N E \mathfrak{N}_E N E . By D ∣ N E D \mid\mathfrak{N}_E D ∣ N E we mean a containment of ideals, that is, N E ⊆ ( D ) \mathfrak{N}_E \subseteq(D) N E ⊆ ( D ) . □
Now that we have established the basic facts, we prove the main theorem
Proof of the Main Theorem
Theorem 5.4. Assume GRH. Let E E E be an elliptic curve without complex multiplication and conductor ideal N E \mathfrak{N}_E N E , over a Galois number field K K K . For a prime number ℓ \ell ℓ with
ℓ ≥ 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) . \ell\ge2C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_E)+2C_2(K). ℓ ≥ 2 C 1 ( K ) log N K / Q rad ( 2 O K N E ) + 2 C 2 ( K ) . ρ E , ℓ ( G K ) \rho_{E,\ell}(G_K) ρ E , ℓ ( G K ) is not contained in the normalizer of a non-split Cartan subgroup of GL 2 ( F ℓ ) \operatorname{GL}_2(\mathbb{F}_\ell) GL 2 ( F ℓ ) .
Proof. Assume that G E ( ℓ ) G_E(\ell) G E ( ℓ ) is contained in the normalizer of a non-split Cartan subgroup C n s + ( ℓ ) C_{ns}^{+}(\ell) C n s + ( ℓ ) but not fully contained in said non-split Cartan subgroup C n s ( ℓ ) C_{ns}(\ell) C n s ( ℓ ) (up to conjugation) Consider the quadratic Galois character given in equation (16) . Let D D D be a square free algebraic integer such that ε ℓ = χ D \varepsilon_\ell=\chi_D ε ℓ = χ D and consider the quadratic twist E D E_D E D of $E. By Lemma 5.3(ii) and (15) we have that
rad ( N E D ) divides rad ( 2 N E ) . (17) \operatorname{rad}(\mathfrak{N}_{E_D})\ \text{divides}\ \operatorname{rad}(2\mathfrak{N}_E). \tag*{(17)} rad ( N E D ) divides rad ( 2 N E ) . (17)
For each prime ideal p ∤ 2 N E \mathfrak{p} \nmid2\mathfrak{N}_E p ∤ 2 N E , we have that
a p ( E ) = ε ℓ ( Frob p ) a p ( E D ) . (18) a_{\mathfrak{p}}(E)=\varepsilon_\ell(\operatorname{Frob}_{\mathfrak{p}})a_{\mathfrak{p}}(E_D). \tag*{(18)} a p ( E ) = ε ℓ ( Frob p ) a p ( E D ) . (18)
By Lemma 5.2 , E E E and E D E_D E D are not K-isogenous. Thus, there exists a prime ideal p ∤ 2 N E \mathfrak{p} \nmid2\mathfrak{N}_E p ∤ 2 N E such that
a p ( E ) ≠ a p ( E D ) . (19) a_{\mathfrak{p}}(E) \ne a_{\mathfrak{p}}(E_D). \tag*{(19)} a p ( E ) = a p ( E D ) . (19)
Let p 0 \mathfrak{p}_0 p 0 be the prime ideal of least norm such that p 0 ∤ 2 N E \mathfrak{p}_0 \nmid2\mathfrak{N}_E p 0 ∤ 2 N E and (19) holds. Applying Theorem 4.4 to E E E and E D E_D E D and using equation (15) , we find that
N K / Q p 0 ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 ) + C 2 ( K ) ) 2 , (20) N_{K/\mathbb{Q}}\mathfrak{p}_0 \le\left(C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_{E_1})+C_2(K)\right)^2, \tag*{(20)} N K / Q p 0 ≤ ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 ) + C 2 ( K ) ) 2 , (20)
where the constants are explicitly described in Theorem 4.4 . Considering (18) and (19) , we see that ε ℓ ( Frob p 0 ) = − 1 \varepsilon_\ell(\operatorname{Frob}_{\mathfrak{p}_0})=-1 ε ℓ ( Frob p 0 ) = − 1 . Thus by lemma 5.3(iii), ℓ \ell ℓ divides a p 0 ( E ) a_{\mathfrak{p}_0}(E) a p 0 ( E ) , and so ℓ ≤ ∣ a p 0 ( E ) ∣ \ell\leq|a_{\mathfrak{p}_0}(E)| ℓ ≤ ∣ a p 0 ( E ) ∣ . The Hasse bound gives us that ∣ a p 0 ( E ) ∣ ≤ 2 N ( p 0 ) |a_{\mathfrak{p}_0}(E)| \leq2\sqrt{N(\mathfrak{p}_0)} ∣ a p 0 ( E ) ∣ ≤ 2 N ( p 0 ) . Thus,
ℓ ≤ 2 N ( p 0 ) ≤ 2 ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 ) + C 2 ( K ) ) . \ell\leq2\sqrt{N(\mathfrak{p}_0)} \leq2\left(C_1(K)\log N_{K/\mathbb{Q}}\operatorname{rad}(2\mathcal{O}_K\mathfrak{N}_{E_1})+C_2(K)\right). ℓ ≤ 2 N ( p 0 ) ≤ 2 ( C 1 ( K ) log N K / Q rad ( 2 O K N E 1 ) + C 2 ( K ) ) .
Expanding the RHS gives us the claimed bound. □
References
Email address: clifford.rodrigues@students.iiserpune.ac.in
Department of Mathematics, Indian Institute of Science Education and Research Pune, Dr. Homi Bhabha Road, Pune 411008, India.
Email address: debargha.banerjee@gmail.com
Department of Mathematics, Indian Institute of Science Education and Research Pune, Dr. Homi Bhabha Road, Pune 411008, India.
Email address: kishorekrishna@iittp.ac.in
Department of Mathematics and Statistics, Indian Institute of Technology (IIT) Tirupati Yerpedu – Venkatagiri Road, Yerpedu Post, Tirupati District, Andhra Pradesh. India, 517619.
Email address: mallicksunil200114@gmail.com
Department of Mathematics and Statistics, Indian Institute of Technology (IIT) Tirupati Yerpedu – Venkatagiri Road, Yerpedu Post, Tirupati District, Andhra Pradesh. India, 517619.