To Di Yin
Introduction
Hilbert’s tenth problem asks for an algorithm that decides whether a polynomial equation with integer coefficients has a solution in integers. For R = Z R = \mathbb{Z} R = Z or Q \mathbb{Q} Q , a subset of R r R^r R r is Diophantine over R R R if, for some m ≥ 0 m \ge0 m ≥ 0 , it is the projection to R r R^r R r of the zero set in R r + m R^{r+m} R r + m of a polynomial with coefficients in R R R . By the theorem of Davis, Putnam, Robinson and Matiyasevich, every recursively enumerable subset of Z \mathbb{Z} Z is Diophantine over Z \mathbb{Z} Z , and no such algorithm
printed: 2020 Mathematics Subject Classification. Primary 11U05; Secondary 03D35, 11F80, 11G05, 11G18, 11N32. exists [19 , 47 , 13 ] . A Diophantine definition of Z \mathbb{Z} Z in Q \mathbb{Q} Q would express the condition that a rational unknown is an integer by additional polynomial equations with rational unknowns. Thus a polynomial equation would have an integer solution if and only if an associated polynomial system had a rational solution. We prove that such a definition exists.
Main results
A Diophantine definition and its consequences.
Theorem A. The set Z \mathbb{Z} Z is a Diophantine subset of Q \mathbb{Q} Q .
Corollary 1.1.
A subset of Q r \mathbb{Q}^{r} Q r is Diophantine if and only if it is recursively enumerable.
There exist an integer M ≥ 1 M \ge1 M ≥ 1 1 and a polynomial P ∈ Z [ T , W 1 , … , W M ] P \in\mathbb{Z}[T,W_{1},\ldots,W_{M}] P ∈ Z [ T , W 1 , … , W M ] of total degree at most 4 such that q ∈ Q q \in\mathbb{Q} q ∈ Q is an integer if and only if P ( q , w ) = 0 P(q,w)=0 P ( q , w ) = 0 for some w ∈ Q M w \in\mathbb{Q}^{M} w ∈ Q M .
For some integer N N N , no algorithm decides whether a polynomial F ∈ Z [ Y 1 , … , Y N ] F \in\mathbb{Z}[Y_{1},\ldots,Y_{N}] F ∈ Z [ Y 1 , … , Y N ] of total degree at most 4 has a zero in Q N \mathbb{Q}^{N} Q N .
If K ⊆ L K \subseteq L K ⊆ L are fields satisfying the same first-order sentences as Q \mathbb{Q} Q in the language of rings, then every element of L L L algebraic over K K K belongs to K K K .
No algorithm decides whether a smooth projective geometrically integral variety over Q \mathbb{Q} Q has a rational point.
Part (4) answers [42 Question 25] ; Koenigsmann proved the implication from Theorem A in [42 Proposition 23(d)] . Part (5) uses Poonen’s reduction to smooth projective varieties [66 Theorem 1.1(i)] .
Corollary 7.6 makes the reduction quantitative. There exists an algorithm transforming an integer equation of degree at most D D D in ν \nu ν variables into a rational quartic equation in at most C ( ν + D D ) C^{\binom{\nu+D}{D}} C ( D ν + D ) variables, for a fixed C C C . The original equation has an integer solution if and only if the resulting equation has a rational solution. Its running time is polynomial in this binomial coefficient and the total bit length of the coefficients. The algorithm uses fixed coefficients depending on constants that we do not evaluate (see Remark 7.7).
Remark. Mazur conjectured that the closure of V ( Q ) V(\mathbb{Q}) V ( Q ) in V ( R ) V(\mathbb{R}) V ( R ) has finitely many connected components for every variety V / Q V/\mathbb{Q} V / Q [48 , 49 ] , [9 Conjecture 2.1] . Theorem A implies the existence of an affine variety V V V whose rational points project onto the closed discrete set Z \mathbb{Z} Z . Their closure therefore has infinitely many connected components, as Mazur observed [48 ] (see also [9 §3] ).
Congruent number curves and height estimates.
The proof of Theorem A is based on rational points on certain congruent number elliptic curves and a height estimate from quaternionic surfaces. For ℓ ∈ Q × \ell\in\mathbb{Q}^{\times} ℓ ∈ Q × , let
E ℓ : y 2 = x 3 − ℓ 2 x . E_{\ell}: y^{2}=x^{3}-\ell^{2}x. E ℓ : y 2 = x 3 − ℓ 2 x .
Theorem B. Let β , η \beta,\eta β , η be coprime positive squarefree integers such that
β = 7 ( m o d 8 ) , η = 1 ( m o d 8 ) . \beta=7 \pmod{8}, \qquad\eta=1 \pmod{8}. β = 7 ( mod 8 ) , η = 1 ( mod 8 ) . Suppose that η \eta η is a square in Q q \mathbb{Q}_{q} Q q for every prime q ∣ β q \mid\beta q ∣ β , and that β \beta β is a square in Q q \mathbb{Q}_{q} Q q for every prime q ∣ η q \mid\eta q ∣ η . There exists a positive squarefree integer n ≡ 5 ( m o d 8 ) n \equiv5 \pmod{8} n ≡ 5 ( mod 8 ) , divisible by β \beta β and prime to η \eta η , such that both E n E_{n} E n and E η n E_{\eta n} E η n have analytic and algebraic rank one. For each ℓ ∈ { n , η n } \ell\in\{n,\eta n\} ℓ ∈ { n , η n } , the Tate–Shafarevich group \Sha(E_{\ell}) is finite of odd order, and E ℓ E_{\ell} E ℓ has a rational point with x x x -coordinate in Q × 2 \mathbb{Q}^{\times2} Q × 2 .
The rational points in Theorem B yield a Diophantine description of square denominators. The congruent number curves E l E_l E l have a rich history and played a central role in the investigations leading to the formulation of the Birch–Swinnerton-Dyer conjecture [4 §1 and Table 2] . It is curious that they also enter a Diophantine definition of Z \mathbb{Z} Z in Q \mathbb{Q} Q .
Definition 1.2. For a finite set S S S of primes, let D 2 S D_2^S D 2 S be the set of b ∈ Q b \in\mathbb{Q} b ∈ Q such that v q ( b ) v_q(b) v q ( b ) is not a negative odd integer for any prime q ∉ S q \notin S q ∈ / S . Put D 2 = D 2 ∅ D_2 = D_2^\varnothing D 2 = D 2 ∅ .
Thus D 2 D_2 D 2 consists of 0 0 0 and the rational numbers whose denominator in lowest terms is a perfect square.
Theorem C. The set D 2 D_2 D 2 is Diophantine over Q \mathbb{Q} Q .
The definition of the integers also requires a height estimate. We obtain it from abelian surfaces with quaternionic multiplication. Let B \mathbb{B} B be the quaternion algebra over Q \mathbb{Q} Q of discriminant 210 2 210^2 21 0 2 2 , let O B \mathcal{O}_{\mathbb{B}} O B be a maximal order, let X X X be the associated Shimura curve over Q \mathbb{Q} Q , and let W ≃ ( Z / 2 Z ) 4 W \simeq(\mathbb{Z}/2\mathbb{Z})^4 W ≃ ( Z /2 Z ) 4 be its group of Atkin–Lehner involutions. The quotient X / W X/W X / W is isomorphic to P 1 \mathbb{P}^1 P 1 over Q \mathbb{Q} Q , and X → X / W X \to X/W X → X / W is branched over five rational points (see Lemma 8.1 ). We choose a coordinate on X / W X/W X / W in which the branch values are
( θ 0 , θ 1 , θ 2 , θ 3 , θ 4 ) = ( 0 , 13 , 525 , 637 , 650 ) (1) (\theta_0,\theta_1,\theta_2,\theta_3,\theta_4)=(0,13,525,637,650) \tag*{(1)} ( θ 0 , θ 1 , θ 2 , θ 3 , θ 4 ) = ( 0 , 13 , 525 , 637 , 650 ) (1)
(see § 11.1 ).
Definition 1.3. Let S S S be a finite set of primes and t ∈ Q ∖ { θ 0 , … , θ 4 } t \in\mathbb{Q} \setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } . A prime p ∉ S p \notin S p ∈ / S is a prime of odd contact of t t t if v p ( t − θ j ) v_p(t-\theta_j) v p ( t − θ j ) is positive and odd for some j j j . Let M S ( t ) M_S(t) M S ( t ) be the product of the primes of odd contact of t t t .
The curve X X X is a coarse moduli space of abelian surfaces with multiplication by O B \mathcal{O}_{\mathbb{B}} O B [25 §2.1] . Let J 1 ( N ) J_1(N) J 1 ( N ) be the Jacobian of the modular curve X 1 ( N ) X_1(N) X 1 ( N ) , and h h h the logarithmic Weil height.
Theorem D. *There exist a finite set S 0 S_0 S 0 of primes and a constant C > 0 C>0 C > 0 with the following property. Let t ∈ Q ∖ { θ 0 , … , θ 4 } t\in\mathbb{Q}\setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } , and let A A A be an abelian surface over Q \mathbb{Q} Q with multiplication by O B \mathcal{O}_{\mathbb{B}} O B whose moduli point3 lies above t t t . Then A A A is isogenous over Q \mathbb{Q} Q to an abelian subvariety of J 1 ( N ) Q J_1(N)_{\mathbb{Q}} J 1 ( N ) Q for some N ≤ C M S 0 ( t ) 2 N\leq C M_{S_0}(t)^2 N ≤ C M S 0 ( t ) 2 .*
This level bound can also be deduced from the descent result of von Känel and Kret [90 Proposition 7.2] , together with von Känel’s conductor estimates [89 §5.2.1 and Lemma 6.8] (see Remark 8.10).
The same family of surfaces also yields the following height estimate.
Theorem E. There exists a finite set S H S_H S H of primes with the following property. For every finite set S ⊇ S H S\supseteq S_H S ⊇ S H of primes there exists H S > 0 H_S>0 H S > 0 such that
h ( t ) ≤ H S M S ( t ) 24 ( t ∈ Q ∖ { θ 0 , … , θ 4 } ) . h(t)\leq H_S M_S(t)^{24}\qquad(t\in\mathbb{Q}\setminus\{\theta_0,\ldots,\theta_4\}). h ( t ) ≤ H S M S ( t ) 24 ( t ∈ Q ∖ { θ 0 , … , θ 4 }) .
For the quadratic twists of the genus-two curve (37) , Theorem E yields h ( t ) ≪ ∣ d ∣ 24 h(t)\ll|d|^{24} h ( t ) ≪ ∣ d ∣ 24 for non-zero squarefree integers d d d (see (38) ).
About the proof
A key idea is to represent an integer n n n by the multiple n P nP n P of a fixed point of infinite order on an elliptic curve. The main step is to express n 1 n 2 = n 3 n_1n_2=n_3 n 1 n 2 = n 3 by polynomial equations in the coordinates. The construction is based on a formula for square denominators (see § 3.1 ) and a height estimate (see § 3.1.1 ). Local tests detect congruences between the indices. The formula expresses these local conditions simultaneously at the primes occurring in the height estimate. Comparison with canonical heights then excludes solutions with n 1 n 2 ≠ n 3 n_1n_2\ne n_3 n 1 n 2 = n 3 , while linear patterns of primes supply solutions when equality holds (see § 3.1.2 ). This constructs a Diophantine model of integer arithmetic. The final approximation argument identifies its indices with the rational integers (see § 3.1.3 ).
Square denominators and congruent number curves
A rational number is an integer when all its valuations are non-negative. Robinson’s existential definitions of the local rings Z ( q ) \mathbb{Z}_{(q)} Z ( q ) treat one prime at a time [73 ] ; an existential formula cannot take their intersection over all primes directly. We first replace non-negativity by the condition that no valuation is negative and odd. This defines the set D 2 D_2 D 2 of rational numbers with square denominator. We construct a positive existential formula4 Φ \Phi Φ such that
D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 D_2 \subseteq\Phi(\mathbb{Q}) \subseteq D_2^{S_1} D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1
for a fixed finite set S 1 S_1 S 1 of primes. Robinson’s local formulas express the remaining parity conditions at the primes of S 1 S_1 S 1 (see the proof of Theorem C).
Fix the finite sets of multipliers M ⊂ Z ∖ { 0 } \mathcal{M}\subset\mathbb{Z}\setminus\{0\} M ⊂ Z ∖ { 0 } and D ⊂ Q × D\subset\mathbb{Q}^{\times} D ⊂ Q × from § 6.1 . For b ≠ 0 b\ne0 b = 0 , the formula Φ ( b ) \Phi(b) Φ ( b ) asserts that, for some μ ∈ M \mu\in\mathcal{M} μ ∈ M and δ ∈ D \delta\in D δ ∈ D , there exist t , e , v , w ∈ Q t,e,v,w\in\mathbb{Q} t , e , v , w ∈ Q and rational points R , R ′ R,R' R , R ′ , distinct from the point at infinity, satisfying, with b ′ = μ b b'=\mu b b ′ = μ b ,
U = 1 − δ b t 2 μ ( 1 + δ b t 2 ) , H ′ = 1 − b ′ U 2 , R ∈ E e ( Q ) , R ′ ∈ E e H ′ ( Q ) , x ( R ) = b ′ v 2 , x ( R ′ ) = b ′ w 2 . (2) \begin{aligned}
U&=\frac{1-\delta bt^2}{\mu(1+\delta bt^2)},&\qquad H'&=1-b'U^2,\\
R&\in E_e(\mathbb{Q}),& R'&\in E_{eH'}(\mathbb{Q}),& x(R)&=b'v^2,& x(R')&=b'w^2. \tag*{(2)}
\end{aligned} U R = μ ( 1 + δ b t 2 ) 1 − δ b t 2 , ∈ E e ( Q ) , H ′ R ′ = 1 − b ′ U 2 , ∈ E e H ′ ( Q ) , x ( R ) = b ′ v 2 , x ( R ′ ) = b ′ w 2 . (2)
Here x ( R ) x(R) x ( R ) denotes the first coordinate of R R R , and similarly for R ′ R' R ′ . We require 1 + δ b t 2 ≠ 0 1+\delta bt^2\ne0 1 + δ b t 2 = 0 and e H ′ v w ≠ 0 eH'vw\ne0 e H ′ v w = 0 (see Definition 3.1). If q ∉ S 1 q\notin S_1 q ∈ / S 1 and v q ( b ) v_q(b) v q ( b ) is negative and odd, then U U U is a q q q -adic unit and v q ( H ′ ) = v q ( b ) v_q(H')=v_q(b) v q ( H ′ ) = v q ( b ) . Exactly one of e , e H ′ e,eH' e , e H ′ has even valuation at q q q . On that quadratic twist, every non-zero x x x -coordinate has even valuation, whereas b ′ v 2 b'v^2 b ′ v 2 and b ′ w 2 b'w^2 b ′ w 2 have odd valuation. Thus (2) has no solution (see Proposition 3.2).
Conversely, for non-zero b ∈ D 2 b\in D_2 b ∈ D 2 , choose μ \mu μ so that b ′ = μ b > 0 b'=\mu b>0 b ′ = μ b > 0 and − b ′ ∈ Q × 2 -b'\in\mathbb{Q}^{\times2} − b ′ ∈ Q × 2 . The choices of δ , t \delta,t δ , t make U U U small at the real place and at the primes dividing 2 μ 2\mu 2 μ , so that H ′ H' H ′ is positive and a square at those primes. This reduces the construction of the two points in (2) to Theorem B (see Proposition 3.14). The proof of that theorem is based on a 2-descent on E l E_l E l for positive squarefree l ≡ 5 ( m o d 8 ) l\equiv5\pmod8 l ≡ 5 ( mod 8 ) . The quotient of its 2-Selmer group by the rational two-torsion classes is the kernel of a matrix of quadratic residue symbols, as in Monsky’s work [35 Appendix] . Auxiliary primes, chosen by Dirichlet’s theorem, make this kernel a line with a prescribed generator for both quadratic twists (see Proposition 3.11). The criterion of Tian, Yuan and Zhang [87 Theorem 1.4] , in Smith’s determinant formulation [83 Theorem 2.2] , then implies that E n E_n E n and E n η E_{n\eta} E n η have analytic rank one. Together with the Selmer calculation, the theorems of Gross–Zagier and Kolyvagin show that the prescribed classes are represented by rational points with the desired square classes of their x x x -coordinates (see Corollary 3.10 ).
Quaternionic surfaces of controlled level
The points θ j \theta_j θ j are the five branch values of X → X / W X \to X/W X → X / W , each with ramification index two. A fixed finite étale cover of X X X carries a family of abelian surfaces with quaternionic multiplication. For t ∈ Q ∖ { θ 0 , … , θ 4 } t \in\mathbb{Q} \setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } , a fibre above t t t is a surface A A A defined over a Galois number field K t K_t K t of bounded degree. Outside a fixed finite set of primes, the local equation at a branch point is z 2 = T − θ j z^2=T-\theta_j z 2 = T − θ j after an unramified extension, where T T T is the coordinate on X / W X/W X / W . On specializing T = t T=t T = t , an even positive valuation of t − θ j t-\theta_j t − θ j introduces no ramification. Thus K t K_t K t can ramify only at the fixed primes and the primes of odd contact of t t t (see Lemma 8.3 ).
The Galois conjugates of A A A are isogenous to it through maps compatible with the quaternionic action. In the non-CM case, following Ribet [72 ] , a controlled splitting of the resulting scalar cocycle defines a compatible system of two-dimensional representations of G Q G_{\mathbb{Q}} G Q . Qualitative modularity, due to Ribet, Pyle and Khare–Wintenberger, identifies this system with that of a weight-two newform after a finite character twist (see Proposition 8.6). Ramification is tame at the primes of odd contact and uniformly bounded at the fixed primes, so the level satisfies
N ≤ C M S 0 ( t ) 2 . (3) N \leq C M_{S_0}(t)^2. \tag*{(3)} N ≤ C M S 0 ( t ) 2 . (3)
The surface A A A is isogenous to an abelian subvariety of J 1 ( N ) J_1(N) J 1 ( N ) over a field of degree O ( M S 0 ( t ) ) O(M_{S_0}(t)) O ( M S 0 ( t )) . Together with (3) , this proves Theorem D in the non-CM case. The finitely many CM isogeny classes are absorbed into the constant (see §8.5).
To pass from the level bound to heights, write h F h_F h F for the stable Faltings height and B ⊆ J 1 ( N ) B \subseteq J_1(N) B ⊆ J 1 ( N ) for the abelian surface isogenous to A A A . Javanpeykar’s bound [39 Theorem 1.1.1] and Rémond’s inequality [71 Corollary 1.2] imply
h F ( B ) ≤ h F ( J 1 ( N ) ) + C dim J 1 ( N ) ≪ N 12 ≪ M S 0 ( t ) 24 . (4) h_F(B) \leq h_F(J_1(N)) + C \dim J_1(N) \ll N^{12} \ll M_{S_0}(t)^{24}. \tag*{(4)} h F ( B ) ≤ h F ( J 1 ( N )) + C dim J 1 ( N ) ≪ N 12 ≪ M S 0 ( t ) 24 . (4)
Faltings’ isogeny inequality and the isogeny theorem of Gaudron and Rémond [27 Theorem 1.4] transfer the bound (4) to A A A (see §8.5). Comparison with the height on the base proves Theorem E; in the CM case, the proof of Theorem D already bounds h ( t ) h(t) h ( t ) .
Arithmetic of indices
Consider the curve E : y 2 = x 3 − d 2 x E: y^2=x^3-d^2x E : y 2 = x 3 − d 2 x over F = Q ( 2 ) F=\mathbb{Q}(\sqrt{2}) F = Q ( 2 ) , with d = 3 − 2 d=3-\sqrt{2} d = 3 − 2 , and let σ \sigma σ be the non-trivial automorphism of F / Q F/\mathbb{Q} F / Q . The curve has rank one, and we choose P P P with ⟨ P ⟩ = 2 E ( F ) \langle P\rangle=2E(F) ⟨ P ⟩ = 2 E ( F ) (see Definition 2.2 ). On ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ , write z = − x / y z=-x/y z = − x / y away from the origin O O O , and put z ( O ) = 0 z(O)=0 z ( O ) = 0 . The coefficients of z ( n P ) z(nP) z ( n P ) in the basis 1 , 2 1,\sqrt{2} 1 , 2 define
ι : Z ⟶ Q 2 , n ⟼ ( a n , b n ) , z ( n P ) = a n + b n 2 . (5) \iota:\mathbb{Z}\longrightarrow\mathbb{Q}^2,\qquad n\longmapsto(a_n,b_n),\qquad z(nP)=a_n+b_n\sqrt{2}. \tag*{(5)} ι : Z ⟶ Q 2 , n ⟼ ( a n , b n ) , z ( n P ) = a n + b n 2 . (5)
Write Z = ι ( Z ) ⊂ Q 2 \mathcal{Z}=\iota(\mathbb{Z})\subset\mathbb{Q}^2 Z = ι ( Z ) ⊂ Q 2 for the set of these pairs. The map ι \iota ι is injective (see Lemma 2.4 ). The set Z \mathcal{Z} Z is Diophantine because ⟨ P ⟩ = 2 E ( F ) \langle P\rangle=2E(F) ⟨ P ⟩ = 2 E ( F ) , and it will represent the integers. Addition of indices is the group law. We show that every polynomial relation among indices is also Diophantine on Z \mathcal{Z} Z .
It suffices to treat f 0 = X 1 X 2 − X 3 f_0=X_1X_2-X_3 f 0 = X 1 X 2 − X 3 ; put K = 48 K=48 K = 48 . A triple ( ι ( n 1 ) , ι ( n 2 ) , ι ( n 3 ) ) (\iota(n_1),\iota(n_2),\iota(n_3)) ( ι ( n 1 ) , ι ( n 2 ) , ι ( n 3 )) determines the points P i = n i P P_i=n_iP P i = n i P . At a place p \mathfrak{p} p above a prime p > K + 1 p>K+1 p > K + 1 , the local test of Definition 2.7 uses points H = h P H=hP H = h P and U i = u i P U_i=u_iP U i = u i P reducing to O O O on E E E . Conditions on their conjugates on E σ E^\sigma E σ imply h ≡ 1 h\equiv1 h ≡ 1 and u i ≡ n i ( m o d p K ) u_i\equiv n_i\pmod{p^K} u i ≡ n i ( mod p K ) . The same reduction conditions on E E E alone would require both h P hP h P and ( h − 1 ) P (h-1)P ( h − 1 ) P to reduce to O O O , hence also P P P . At the good primes of Definition 2.5, the reductions of P P P and P σ P^\sigma P σ have coprime orders, both prime to p p p . The Chinese remainder theorem then makes the conditions on the two curves compatible (see Lemma 2.9 ).
The truncated formal logarithm ℓ K \ell_K ℓ K of §2.2 defines L i ∈ F L_i\in F L i ∈ F with
L i = ℓ K ( z ( U i ) ) ℓ K ( z ( H ) ) ≡ u i h ≡ n i ( m o d p K ) . L_i=\frac{\ell_K(z(U_i))}{\ell_K(z(H))}\equiv\frac{u_i}{h}\equiv n_i\pmod{p^K}. L i = ℓ K ( z ( H )) ℓ K ( z ( U i )) ≡ h u i ≡ n i ( mod p K ) .
Consequently the last condition in the local test implies
v p ( L 1 L 2 − L 3 ) ≥ K ⟹ p K ∣ Δ , Δ = n 1 n 2 − n 3 (6) v_p(L_1L_2-L_3)\ge K \quad\Longrightarrow\quad p^K\mid\Delta,\qquad\Delta=n_1n_2-n_3 \tag*{(6)} v p ( L 1 L 2 − L 3 ) ≥ K ⟹ p K ∣ Δ , Δ = n 1 n 2 − n 3 (6)
(see Lemma 2.8 ). The points H , U i , P i H,U_i,P_i H , U i , P i are represented by the rational coefficients of their coordinates in the basis 1 , 2 1,\sqrt{2} 1 , 2 ; the integers h , u i , n i h,u_i,n_i h , u i , n i denote their indices.
We now express these local conditions at the primes occurring in the height estimate. Write z ( P i ) = r i , 0 + r i , 1 2 z(P_i)=r_{i,0}+r_{i,1}\sqrt{2} z ( P i ) = r i , 0 + r i , 1 2 , so r i , 0 = a n i r_{i,0}=a_{n_i} r i , 0 = a n i and r i , 1 = b n i r_{i,1}=b_{n_i} r i , 1 = b n i in (1.5). Let A \mathcal A A consist of 1 1 1 and the six squares ( ( N ∗ ) 2 + r i , e 2 ) 2 \left((N^*)^2+r_{i,e}^2\right)^2 ( ( N ∗ ) 2 + r i , e 2 ) 2 , where N ∗ > 1 N^*>1 N ∗ > 1 is a fixed integer. All seven entries are non-zero, even when some P i = O P_i=O P i = O . For an auxiliary s ∈ Q × s\in\mathbb Q^\times s ∈ Q × , put
t A = A s ( A ∈ A ) , t 1 = s , A = t A / t 1 . (7) t_A=As\quad(A\in\mathcal A),\qquad t_1=s,\qquad A=t_A/t_1. \tag*{(7)} t A = A s ( A ∈ A ) , t 1 = s , A = t A / t 1 . (7)
We choose N ∗ N^* N ∗ after the height comparison below.
Fix ξ ∈ Q ∖ { θ 0 , … , θ 4 } \xi\in\mathbb Q\setminus\{\theta_0,\ldots,\theta_4\} ξ ∈ Q ∖ { θ 0 , … , θ 4 } . We require t A ∉ { θ 0 , … , θ 4 , ξ } t_A\notin\{\theta_0,\ldots,\theta_4,\xi\} t A ∈ / { θ 0 , … , θ 4 , ξ } and define
R A = ∏ j = 0 4 ( t A − θ j ) ( t A − ξ ) 6 . R_A=\frac{\prod_{j=0}^{4}(t_A-\theta_j)}{(t_A-\xi)^6}. R A = ( t A − ξ ) 6 ∏ j = 0 4 ( t A − θ j ) .
Outside a fixed finite set S S S , specified in § 9.1 , a prime of odd contact of t A t_A t A has positive odd valuation in R A R_A R A . By Lemma 4.1, a single instance of Φ \Phi Φ ensures that the local conditions hold for some test datum at a place above each such prime. We apply Φ \Phi Φ to R A / Θ A 2 R_A/\Theta_A^2 R A / Θ A 2 5 , where Θ A ≠ 0 \Theta_A\ne0 Θ A = 0 is the trace of a sum with one term for each of finitely many test data. At a prime p ∉ S p\notin S p ∈ / S with v p ( R A ) v_p(R_A) v p ( R A ) positive and odd, failure of a test at a place above p p p makes the corresponding term have valuation at least v p ( R A ) v_p(R_A) v p ( R A ) . If every test failed at every place above p p p , the trace would have at least this valuation, and R A / Θ A 2 R_A/\Theta_A^2 R A / Θ A 2 would have negative odd valuation. This is excluded by Φ \Phi Φ . The datum and place may depend on the prime. Applying the construction for each A ∈ A A\in\mathcal A A ∈ A defines the finite system Σ \Sigma Σ of Definition 6.2 .
Suppose that Σ \Sigma Σ has a solution with Δ ≠ 0 \Delta\ne0 Δ = 0 , and put B = max ( 1 , ∣ n 1 ∣ , ∣ n 2 ∣ , ∣ n 3 ∣ ) B=\max(1,|n_1|,|n_2|,|n_3|) B = max ( 1 , ∣ n 1 ∣ , ∣ n 2 ∣ , ∣ n 3 ∣ ) . By (6) , every prime divisor of M S ( t A ) M_S(t_A) M S ( t A ) divides Δ \Delta Δ to order at least K K K . Theorem E, K = 48 K=48 K = 48 and ∣ Δ ∣ ≤ 2 B 2 |\Delta|\le2B^2 ∣Δ∣ ≤ 2 B 2 therefore imply
M S ( t A ) K ∣ Δ , h ( t A ) ≤ H S ∣ Δ ∣ ≪ B ( A ∈ A ) . (8) M_S(t_A)^K\mid\Delta,\qquad h(t_A)\le H_S\sqrt{|\Delta|}\ll B\quad(A\in\mathcal A). \tag*{(8)} M S ( t A ) K ∣ Δ , h ( t A ) ≤ H S ∣Δ∣ ≪ B ( A ∈ A ) . (8)
By (8) and the identity A = t A / t 1 A=t_A/t_1 A = t A / t 1 in (7) , we obtain h ( ( N ∗ ) 2 + r i , e 2 ) ≤ H S ∣ Δ ∣ h((N^*)^2+r_{i,e}^2)\le H_S\sqrt{|\Delta|} h (( N ∗ ) 2 + r i , e 2 ) ≤ H S ∣Δ∣ . On the other hand, h ( ( N ∗ ) 2 + r 2 ) ≥ h ( r ) + log N ∗ h((N^*)^2+r^2)\ge h(r)+\log N^* h (( N ∗ ) 2 + r 2 ) ≥ h ( r ) + log N ∗ for every r ∈ Q r\in\mathbb Q r ∈ Q . The canonical height satisfies h ^ ( n i P ) ≤ h ( r i , 0 ) + h ( r i , 1 ) + C z \hat h(n_iP)\le h(r_{i,0})+h(r_{i,1})+C_z h ^ ( n i P ) ≤ h ( r i , 0 ) + h ( r i , 1 ) + C z for n i ≠ 0 n_i\ne0 n i = 0 (see (25) ). Choosing i i i with ∣ n i ∣ = B |n_i|=B ∣ n i ∣ = B , we obtain
h ^ ( P ) B 2 + 2 log N ∗ ≤ 2 2 H S B + C z . \hat h(P)B^2+2\log N^*\le2\sqrt{2}H_SB+C_z. h ^ ( P ) B 2 + 2 log N ∗ ≤ 2 2 H S B + C z .
Completing the square bounds log N ∗ \log N^* log N ∗ by a constant independent of N ∗ N^* N ∗ . Choosing N ∗ N^* N ∗ larger excludes every solution with Δ ≠ 0 \Delta\ne0 Δ = 0 (see Proposition 6.4).
Conversely, suppose Δ = 0 \Delta=0 Δ = 0 . The linear prime patterns of § 8 provide s s s for which each t A − θ j t_A-\theta_j t A − θ j is, up to a rational square, the quotient of two good primes (see Definition 2.5). The Green–Tao–Ziegler theorem supplies patterns in admissible residue classes. An upper-bound sieve shows that most consist of good primes (see § 8.2 ). For each A A A , the resulting R A R_A R A has valuations one at a bounded number of good primes and even valuation elsewhere. The Chinese remainder theorem then constructs the local test points (see Lemma 2.9), and Lemma 4.1(2) verifies the instances of Φ \Phi Φ . This proves that Σ \Sigma Σ has a rational solution (see Proposition 6.3).
After expressing the point coordinates in the basis 1 , 2 1,\sqrt{2} 1 , 2 , Σ \Sigma Σ has input y = ( ℓ ( n 1 ) , ℓ ( n 2 ) , ℓ ( n 3 ) ) \mathbf y=(\ell(n_1),\ell(n_2),\ell(n_3)) y = ( ℓ ( n 1 ) , ℓ ( n 2 ) , ℓ ( n 3 )) and a finite tuple of rational auxiliary variables w \mathbf w w . A solution Σ ( y , w ) = 0 \Sigma(\mathbf y,\mathbf w)=0 Σ ( y , w ) = 0 exists exactly when n 1 n 2 = n 3 n_1n_2=n_3 n 1 n 2 = n 3 . Together with the group law, this defines every polynomial relation by introducing intermediate points for its additions and multiplications. This is the Diophantine model of Theorem 6.1.
The model already implies the undecidability assertions of Corollary 1.1 (3) and (5), and contradicts Mazur’s conjecture [9 Theorem 3.6] . To prove Theorem A, we then identify each index with its ordinary rational value.
Figure 2.
From indices to integers
It remains to identify the index n n n in (5) with the rational number n n n . Fix an integer C 0 ≥ 1 C_{0} \ge1 C 0 ≥ 1 larger than all the branch values θ j \theta_{j} θ j , so that q 2 + C 0 q^{2} + C_{0} q 2 + C 0 does not equal any θ j \theta_{j} θ j for q ∈ Q q \in\mathbb{Q} q ∈ Q . For q ∈ Q q \in\mathbb{Q} q ∈ Q , consider an auxiliary point whose index a ≥ 1 a \ge1 a ≥ 1 is divisible by a fixed power of every prime of odd contact of q 2 + C 0 q^{2} + C_{0} q 2 + C 0 , outside a fixed finite set. Then Theorem E bounds h ( q ) h(q) h ( q ) linearly in ∣ a ∣ \lvert a\rvert ∣ a ∣ , and every finite subset of Q \mathbb{Q} Q satisfies this condition for a suitable a a a (see Theorem 7.2 ).
Once h ( q ) h(q) h ( q ) and h ( n ) h(n) h ( n ) are bounded, agreement modulo a sufficiently high power of a fixed prime p 0 p_{0} p 0 implies q = n q=n q = n . By the theorem of Davis, Putnam, Robinson and Matiyasevich, recursive relations among the indices are Diophantine on the model (see Lemma 6.5 ). We may therefore require an auxiliary index to be p 0 L ( a + n 2 + 1 ) p_{0}^{L(a+n^{2}+1)} p 0 L ( a + n 2 + 1 ) , with L L L a sufficiently large fixed integer. In the formal group at p 0 p_{0} p 0 , multiplication by n n n satisfies [ n ] ( X ) ≡ n X ( m o d X 2 ) [n](X) \equiv nX \pmod{X^{2}} [ n ] ( X ) ≡ n X ( mod X 2 ) . Applied to a point sufficiently close to O O O , this expresses the congruence to the prescribed order through Φ \Phi Φ (see § 10.2 ). For this last step, any Diophantine family of finite sets covering Q \mathbb{Q} Q with a recursive height bound suffices (see Remark 7.8). Thus the pairs ( n , ι ( n ) ) (n,\iota(n)) ( n , ι ( n )) form a Diophantine subset of Q 3 \mathbb{Q}^{3} Q 3 ; projection to the first coordinate defines Z \mathbb{Z} Z in Q \mathbb{Q} Q .
Relation to other works
The work of OpenAI
OpenAI’s work [55 ] proves the undecidability of Hilbert’s tenth problem over Q \mathbb{Q} Q , but does not establish that Z \mathbb{Z} Z is a Diophantine subset of Q \mathbb{Q} Q . Our proof is inspired by that approach and requires neither the pointwise 2-converse nor the 2-adic Fontaine–Mazur modularity announced in [56 , 57 ] . In [55 ] , the 2-converse is used to realize prescribed Selmer classes by rational points on auxiliary quadratic twists. The 2-adic Fontaine–Mazur result supplies modularity for the representations attached to quaternionic surfaces. Ramification estimates bound the level. Together, modularity and the ramification estimates establish the height estimate used in the Diophantine argument.
Here the Tian–Yuan–Zhang criterion, in Smith’s formulation, establishes the rank-one assertion for the congruent number curves. Qualitative modularity and controlled descent establish the level bound of Theorem D, from which the height estimate follows. The curve of rank one over Q ( 2 ) \mathbb{Q}(\sqrt{2}) Q ( 2 ) and the comparison with its conjugate also come from [55 ] . Further common features are the norm construction for square denominators, linear patterns of primes, and the use of quaternionic surfaces to bound heights through odd contact with five points.
In [55 ] the passage from integer to rational equations uses finite tests and a compactness argument. We also employ Sun’s selection formula and height comparison [86 Lemma 3.1, §7.1, Proposition 9.1] . Here linear prime patterns provide the parameters, and separate test data allow the local congruences to be solved independently at different primes. A fixed large integer in the coordinate squares excludes triples with n 1 n 2 ≠ n 3 n_{1}n_{2} \ne n_{3} n 1 n 2 = n 3 . The resulting equations define a Diophantine model of integer arithmetic, and Theorem 7.5 then identifies the indices with the integers themselves.
Earlier work on Hilbert’s tenth problem
Robinson proved that Z \mathbb{Z} Z is first-order definable in Q \mathbb{Q} Q [73 ] . Poonen gave a definition with quantifier pattern ∀ ∃ \forall\exists ∀∃ [65 Theorem 4.1] , and Koenigsmann a universal one [42 Theorem 1] ; see [59 ] for rings of integers of number fields and [11 ] for a universal definition with ten quantifiers. Mazur’s conjecture [49 ] would exclude an existential definition, and even a Diophantine model of Z \mathbb{Z} Z in a power of Q \mathbb{Q} Q [9 Theorem 3.6] .
Poonen constructed Diophantine models over certain subrings Z [ S − 1 ] \mathbb{Z}[S^{-1}] Z [ S − 1 ] of Q \mathbb{Q} Q , with S S S of natural density one, using an elliptic curve of rank one [63 Theorem 1.3] ; see also [20 ] . His model represents i i i by the y y y -coordinate y ( ℓ i P ) y(\ell_iP) y ( ℓ i P ) , where ℓ i \ell_i ℓ i is prime and y ( ℓ i P ) y(\ell_iP) y ( ℓ i P ) approximates i i i [63 §10] . Here the represented integer is the index n n n of n P nP n P , and addition is the elliptic-curve group law. Poonen and Shlapentokh extended the use of approximation and Diophantine models to large subrings of number fields [67 Theorem 1.9] .
There is also a connection with the semilocal methods of Poonen and Koenigsmann. Poonen uses quaternion algebras split at the real place to define intersections of finitely many local rings [65 Lemma 2.5] . Koenigsmann constructs their Jacobson radicals using square classes and valuation parity [42 Lemma 13] . For a squarefree integer d > 1 d > 1 d > 1 , Theorem C expresses integrality at every prime dividing d d d by the condition d x 2 ∈ D 2 dx^{2} \in D_{2} d x 2 ∈ D 2 . The same predicate defines powers of the Jacobson radical (see (27) ). The height estimates supply the global control used in our construction.
The passage from indices to integers in § 10.2 also has a precedent in Poonen’s use of formal groups [62 Lemmas 5, 11–12] . There the quotient x ( P ) / x ( m P ) x(P)/x(mP) x ( P ) / x ( m P ) approximates m 2 m^{2} m 2 at primes dividing the denominator of x ( P ) x(P) x ( P ) . Divisibility and a global size bound then imply membership in a smaller ring of integers. Here Diophantine finite sets supply the height bound, while arithmetic on the model permits the order of approximation to depend recursively on the indices. For related work using elliptic divisibility sequences, see [10 ] .
For earlier work over rings of integers of number fields, see [16 , 18 , 17 , 61 , 84 , 88 ] . Poonen [62 ] , and independently Cornelissen, Pheidas and Zahidi [7 ] , reduced the problem to elliptic curves of rank one whose rank does not grow in an extension; see also [81 , 53 MP18, 23 ] . Koymans and Pagano proved that Hilbert’s tenth problem has a negative answer for every infinite finitely generated ring, combining 2 2 2 -descent with additive combinatorics [44 ] ; as Q \mathbb{Q} Q is not finitely generated over Z \mathbb{Z} Z , this does not cover the rational case. Alpgot, Bhargava, Ho and Shnidman gave another proof for rings of integers, through abelian varieties whose positive rank is stable in quadratic extensions [1 ] . For surveys see [64 , 41 ] .
Theorem E makes Darmon’s finiteness theorem effective for the M M M -curve with underlying curve P 1 \mathbb{P}^{1} P 1 and multiplicity two at each of the five marked points θ 0 , … , θ 4 \theta_{0},\ldots,\theta_{4} θ 0 , … , θ 4 [12 ] . The condition M S ( t ) = 1 M_{S}(t) = 1 M S ( t ) = 1 means that the intersections with its five marked points have even multiplicity outside S S S .
In the non-CM case, qualitative modularity of the surfaces in Theorem D is known through the work of Ribet and Pyle and Khare–Wintenberger’s proof of Serre’s conjecture [72 , 68 , 45 ] . Theorem D bounds the level uniformly.
Kai has extended the Green–Tao–Ziegler theorem to linear patterns of prime elements in number fields [40 Theorem 1.1] . Extending the present approach to rings of integers inside number fields would also require analogues of the rank-one construction, the height estimate and the identification of indices.
Concurrent work
Following [55 ] , other approaches to a Diophantine definition of Z \mathbb{Z} Z in Q \mathbb{Q} Q have appeared [Far26, Red26, Gla26]. We learned of them upon completing this paper. The constructions in [Far26, Red26] retain hypotheses from [55 ] . The manuscript [24 ] , described in its repository as produced with AI systems, proposes an unconditional proof by an approach that differs from ours in several respects. We have not looked into its details.
Leitfaden
Section 2 introduces the curve of rank one and the local test. Section 3 proves Theorems B and C. Sections 4–6 construct the Diophantine model using the local test, the formula Φ \Phi Φ of Section 3, Theorem E and linear prime patterns. Section 7 identifies its indices with the rational integers. Section 8 proves Theorems D and E independently of the preceding sections. Appendix A proves the case of Smith’s determinant identity used in Section 3.
Figure 1 (PDF p. 9). The two lower constructions use the formula for square denominators, the height estimate and the formal group. Linear prime patterns supply the local solutions for the model. Approximation at one fixed prime then identifies its indices with the rational integers.
Notation
We write h h h for the absolute logarithmic Weil height on Q \mathbb{Q} Q . Recall that h ( a + b ) ≤ h ( a ) + h ( b ) + log 2 h(a+b) \le h(a)+h(b)+\log2 h ( a + b ) ≤ h ( a ) + h ( b ) + log 2 , h ( a b ) ≤ h ( a ) + h ( b ) h(ab) \le h(a)+h(b) h ( ab ) ≤ h ( a ) + h ( b ) , h ( a − 1 ) = h ( a ) h(a^{-1}) = h(a) h ( a − 1 ) = h ( a ) for a ≠ 0 a \ne0 a = 0 , and h ( a 2 ) = 2 h ( a ) h(a^2) = 2h(a) h ( a 2 ) = 2 h ( a ) [2 §1.5] .
For a prime q q q , v q v_q v q is the q q q -adic valuation, with v q ( 0 ) = + ∞ v_q(0)=+\infty v q ( 0 ) = + ∞ . An element a ∈ Q q a\in\mathbb{Q}_q a ∈ Q q is a square at q q q if a ∈ Q q × 2 a\in\mathbb{Q}_q^{\times2} a ∈ Q q × 2 . Legendre symbols are written ( a q ) \left(\frac{a}{q}\right) ( q a ) .
We put F = Q ( 2 ) F=\mathbb{Q}(\sqrt{2}) F = Q ( 2 ) , with non-trivial automorphism σ \sigma σ and ring of integers O F = Z [ 2 ] \mathcal{O}_F=\mathbb{Z}[\sqrt{2}] O F = Z [ 2 ] . For a place p \mathfrak{p} p of F F F above an odd prime p p p we normalize v p ( p ) = 1 v_{\mathfrak{p}}(p)=1 v p ( p ) = 1 .
A positive existential formula is built from polynomial equations with rational coefficients using ∧ \land ∧ , ∨ \lor ∨ and ∃ \exists ∃ . A subset of Q k \mathbb{Q}^k Q k is Diophantine exactly when it is defined by such a formula, since ∧ \land ∧ is expressed by sums of squares, ∨ \lor ∨ by products, and a ≠ 0 a\ne0 a = 0 by ∃ w ( a w = 1 ) \exists w\;(aw=1) ∃ w ( a w = 1 ) .
Use of AI tools. The author consulted ChatGPT and Claude while studying the OpenAI paper [55 ] . ChatGPT assisted with the arguments in § 7 and § 8 . Both tools were also used for revising the draft. The author is responsible for the mathematical arguments and the final text.
Acknowledgements. The author is grateful to Alex Smith for drawing his attention to the use of the 2-converse in the OpenAI preprint [55 ] , to which the present study owes its inception. He is also grateful to Chandrashekhar Khare and Ye Tian for inspiring discussions over the years. He thanks Manjul Bhargava, Peter Koymans, Rafael von Känel, Carlo Pagano, Ari Shnidman and Shou-Wu Zhang for their encouraging interest and helpful comments. The author was partially supported by NSF grant DMS-2302064.
An elliptic curve of rank one and a local test
In this section we fix a point of infinite order on an elliptic curve of rank one over F = Q ( 2 ) F=\mathbb{Q}(\sqrt{2}) F = Q ( 2 ) , and show that polynomial conditions on the coordinates of its multiples detect their indices modulo powers of a prime (see Lemmas 2.8 and 2.9 ).
The curve
Throughout, put
d = 3 − 2 , E : y 2 = x 3 − d 2 x , E σ : y 2 = x 3 − ( d σ ) 2 x . d=3-\sqrt{2},\qquad E:y^2=x^3-d^2x,\qquad E^\sigma:y^2=x^3-(d^\sigma)^2x. d = 3 − 2 , E : y 2 = x 3 − d 2 x , E σ : y 2 = x 3 − ( d σ ) 2 x .
Then d d σ = 7 dd^\sigma=7 d d σ = 7 , and d d d , d σ d^\sigma d σ are totally positive. For Q = ( x , y ) ∈ E ( F ) Q=(x,y)\in E(F) Q = ( x , y ) ∈ E ( F ) we write Q σ = ( σ x , σ y ) ∈ E σ ( F ) Q^\sigma=(\sigma x,\sigma y)\in E^\sigma(F) Q σ = ( σ x , σ y ) ∈ E σ ( F ) , and O σ = O O^\sigma=O O σ = O .
This is a smaller model of the curve used in [55 §5] . Indeed, if u = 3 − 2 2 u=3-2\sqrt{2} u = 3 − 2 2 , then 75 − 53 2 = d u 2 75-53\sqrt{2}=du^2 75 − 53 2 = d u 2 , and ( x , y ) ↦ ( u 2 x , u 3 y ) (x,y)\mapsto(u^2x,u^3y) ( x , y ) ↦ ( u 2 x , u 3 y ) identifies the two curves over F F F . The following proposition records the same descent in this model.
Proposition 2.1. The group E ( F ) E(F) E ( F ) has rank one, and its torsion subgroup is
E [ 2 ] = { O , ( 0 , 0 ) , ( d , 0 ) , ( − d , 0 ) } . E[2]=\{O,(0,0),(d,0),(-d,0)\}. E [ 2 ] = { O , ( 0 , 0 ) , ( d , 0 ) , ( − d , 0 )} .
Proof. Recall that O F \mathcal{O}_F O F is a principal ideal domain, that O F × / O F × 2 \mathcal{O}_F^\times/\mathcal{O}_F^{\times2} O F × / O F × 2 is represented by ± 1 , ± ( 1 + 2 ) \pm1,\pm(1+\sqrt{2}) ± 1 , ± ( 1 + 2 ) , and that totally positive units are squares. Let φ : E → E ′ \varphi:E\to E' φ : E → E ′ , with E ′ : y 2 = x 3 + 4 d 2 x E':y^2=x^3+4d^2x E ′ : y 2 = x 3 + 4 d 2 x , be the 2-isogeny with kernel { O , ( 0 , 0 ) } \{O,(0,0)\} { O , ( 0 , 0 )} . Let φ ^ : E ′ → E \hat{\varphi}:E'\to E φ ^ : E ′ → E be its dual, and let
α : E ( F ) → F × / F × 2 , α ′ : E ′ ( F ) → F × / F × 2 \alpha:E(F)\to F^\times/F^{\times2},\qquad\alpha':E'(F)\to F^\times/F^{\times2} α : E ( F ) → F × / F × 2 , α ′ : E ′ ( F ) → F × / F × 2
be the descent maps for φ ^ \hat{\varphi} φ ^ and φ \varphi φ , respectively, sending a point with x ≠ 0 x\ne0 x = 0 to the class of x x x [SiL09, X.4.9]. As E [ 2 ] ⊂ E ( F ) E[2]\subset E(F) E [ 2 ] ⊂ E ( F ) and ker φ ^ = φ ( E [ 2 ] ) \ker\hat{\varphi}=\varphi(E[2]) ker φ ^ = φ ( E [ 2 ]) ,
∣ im α ∣ ⋅ ∣ im α ′ ∣ = ∣ E ( F ) / 2 E ( F ) ∣ = 4 ⋅ 2 rank E ( F ) . (9) |\operatorname{im}\alpha|\cdot|\operatorname{im}\alpha'|=|E(F)/2E(F)|=4\cdot2^{\operatorname{rank}E(F)}. \tag*{(9)} ∣ im α ∣ ⋅ ∣ im α ′ ∣ = ∣ E ( F ) /2 E ( F ) ∣ = 4 ⋅ 2 rank E ( F ) . (9)
We show that
im α = ⟨ − 1 , 1 + 2 , d ⟩ , im α ′ = { 1 } , \operatorname{im}\alpha=\langle-1,1+\sqrt{2},d\rangle,\qquad\operatorname{im}\alpha'=\{1\}, im α = ⟨ − 1 , 1 + 2 , d ⟩ , im α ′ = { 1 } ,
which gives rank E ( F ) = 1 E(F)=1 E ( F ) = 1 . On a curve y 2 = x 3 + a x y^2=x^3+ax y 2 = x 3 + a x , every non-zero x x x has even valuation at a finite place where a a a is a unit. Hence im α \operatorname{im}\alpha im α lies in the group generated by the units and d d d , of order 8. It contains the classes − 1 -1 − 1 , d d d , − d -d − d of the points of order two, and the class of x ( Q ) = 1 + 2 x(Q)=1+\sqrt{2} x ( Q ) = 1 + 2 for
Q = ( 1 + 2 , 2 2 ) ∈ E ( F ) , Q=(1+\sqrt{2},2\sqrt{2})\in E(F), Q = ( 1 + 2 , 2 2 ) ∈ E ( F ) ,
which has different signs at the two real places.
On E ′ E^\prime E ′ every non-zero x x x is totally positive, since x 3 + 4 d 2 x x^3+4d^2x x 3 + 4 d 2 x has the sign of x x x , and its valuation is even at the primes other than ( 2 ) (\sqrt{2}) ( 2 ) and ( d ) (d) ( d ) . At ( 2 ) (\sqrt{2}) ( 2 ) we have v ( 4 d 2 ) = 4 v(4d^2)=4 v ( 4 d 2 ) = 4 , and an odd v ( x ) v(x) v ( x ) would make v ( y 2 ) = min ( 3 v ( x ) , 4 + v ( x ) ) v(y^2)=\min(3v(x),4+v(x)) v ( y 2 ) = min ( 3 v ( x ) , 4 + v ( x )) odd. At ( d ) (d) ( d ) we have v ( 4 d 2 ) = 2 v(4d^2)=2 v ( 4 d 2 ) = 2 , and the only odd valuation not excluded in the same way is v ( x ) = 1 v(x)=1 v ( x ) = 1 ; then x = d u x=du x = d u with u u u a unit at ( d ) (d) ( d ) and y 2 = d 3 u ( u 2 + 4 ) y^2=d^3u(u^2+4) y 2 = d 3 u ( u 2 + 4 ) , so
u 2 ≡ − 4 ≡ 3 ( m o d d ) , u^2\equiv-4\equiv3 \pmod d, u 2 ≡ − 4 ≡ 3 ( mod d ) ,
which is impossible since 3 is not a square in O F / ( d ) ≃ F 7 \mathcal{O}_F/(d)\simeq\mathbb{F}_7 O F / ( d ) ≃ F 7 . Hence x x x is a totally positive unit times a square, hence a square, and ( 0 , 0 ) (0,0) ( 0 , 0 ) has class 4 d 2 = 1 4d^2=1 4 d 2 = 1 . So im α ′ = { 1 } \operatorname{im}\alpha^\prime=\{1\} im α ′ = { 1 } .
For the torsion, 17 splits in F F F , with 2 ≡ 6 \sqrt{2}\equiv6 2 ≡ 6 or 11 and correspondingly d ≡ 14 d\equiv14 d ≡ 14 or 9 (mod 17), and at the two places above 17 the reductions of E E E have 20 and 16 points. Reduction is injective on prime-to-17 torsion, and the formal kernel has no 17-torsion since 17 is unramified in F F F [82 VII.3.1 and IV.6.1] . Thus the order of E ( F ) t o r s E(F)_{\mathrm{tors}} E ( F ) tors divides gcd ( 20 , 16 ) = 4 \gcd(20,16)=4 g cd( 20 , 16 ) = 4 .
Definition 2.2. Fix P ~ ∈ E ( F ) \widetilde{P}\in E(F) P ∈ E ( F ) whose image generates E ( F ) / E [ 2 ] E(F)/E[2] E ( F ) / E [ 2 ] , put P = 2 P ~ P=2\widetilde{P} P = 2 P , and let ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ be the subgroup generated by P P P . The index of Q ∈ ⟨ P ⟩ Q\in\langle P\rangle Q ∈ ⟨ P ⟩ is the integer n n n with Q = n P Q=nP Q = n P .
By Proposition 2.1 , ⟨ P ⟩ = 2 E ( F ) \langle P\rangle=2E(F) ⟨ P ⟩ = 2 E ( F ) , the point P P P has infinite order, and the non-zero points of ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ have non-zero coordinates, as they do not have order two. The index of Q σ Q^\sigma Q σ with respect to P σ P^\sigma P σ is the index of Q Q Q .
The generator P ~ \widetilde{P} P can be computed by a finite search. Indeed, a generator has canonical height at most that of the point Q Q Q in the proof of Proposition 2.1 , and the comparison with the coordinate height bounds the search [82 VIII.9.3] .
Formal logarithms
For Q ≠ O Q\ne O Q = O on E E E or E σ E^\sigma E σ with x ( Q ) y ( Q ) ≠ 0 x(Q)y(Q)\ne0 x ( Q ) y ( Q ) = 0 put
z ( Q ) = − x ( Q ) y ( Q ) , ρ ( Q ) = 1 x ( Q ) , z(Q)=-\frac{x(Q)}{y(Q)},\qquad\rho(Q)=\frac{1}{x(Q)}, z ( Q ) = − y ( Q ) x ( Q ) , ρ ( Q ) = x ( Q ) 1 ,
and put z ( O ) = ρ ( O ) = 0 z(O)=\rho(O)=0 z ( O ) = ρ ( O ) = 0 . In the parameter Z = z Z=z Z = z at O O O , the invariant differential of E E E is ( 1 + 4 d 2 Z 4 ) − 1 / 2 d Z (1+4d^2Z^4)^{-1/2}\,dZ ( 1 + 4 d 2 Z 4 ) − 1/2 d Z [82 IV.1] , so the formal logarithm and its truncation at K ≥ 1 K\ge1 K ≥ 1 are
ℓ ( Z ) = ∑ j ≥ 0 ( − 1 ) j ( 2 j j ) d 2 j 4 j + 1 Z 4 j + 1 , ℓ K ( Z ) = ∑ 0 ≤ j ≤ ( K − 1 ) / 4 ( − 1 ) j ( 2 j j ) d 2 j 4 j + 1 Z 4 j + 1 . \ell(Z)=\sum_{j\ge0}(-1)^j\binom{2j}{j}\frac{d^{2j}}{4j+1}Z^{4j+1},\qquad
\ell_K(Z)=\sum_{0\le j\le(K-1)/4}(-1)^j\binom{2j}{j}\frac{d^{2j}}{4j+1}Z^{4j+1}. ℓ ( Z ) = j ≥ 0 ∑ ( − 1 ) j ( j 2 j ) 4 j + 1 d 2 j Z 4 j + 1 , ℓ K ( Z ) = 0 ≤ j ≤ ( K − 1 ) /4 ∑ ( − 1 ) j ( j 2 j ) 4 j + 1 d 2 j Z 4 j + 1 .
We obtain ℓ σ \ell^\sigma ℓ σ and ℓ K σ \ell_K^\sigma ℓ K σ by replacing d d d with d σ d^\sigma d σ . Only ℓ K \ell_K ℓ K enters the equations. For E σ E^\sigma E σ , the valuation statement in Proposition 2.3 (1) suffices.
Let p ∉ { 2 , 7 } \mathfrak{p}\notin\{2,7\} p ∈ / { 2 , 7 } be a prime and p \mathfrak{p} p a place of F F F above p \mathfrak{p} p . The curves E E E and E σ E^\sigma E σ have good reduction at p \mathfrak{p} p , their discriminants being 64 d 6 64d^6 64 d 6 and 64 ( d σ ) 6 64(d^\sigma)^6 64 ( d σ ) 6 . A point Q ≠ O Q\ne O Q = O reduces to O O O at p \mathfrak{p} p if and only if v p ( x ( Q ) ) < 0 v_{\mathfrak{p}}(x(Q))<0 v p ( x ( Q )) < 0 , and then v p ( x ( Q ) ) = − 2 v p ( z ( Q ) ) v_{\mathfrak{p}}(x(Q))=-2v_{\mathfrak{p}}(z(Q)) v p ( x ( Q )) = − 2 v p ( z ( Q )) [82 VII.2] . Thus, for Q ≠ O Q\ne O Q = O ,
v p ( ρ ( Q ) ) ≥ 1 ⟺ Q reduces to O at p , and then v p ( ρ ( Q ) ) = 2 v p ( z ( Q ) ) . (10) v_{\mathfrak{p}}(\rho(Q))\ge1\Longleftrightarrow Q\text{ reduces to }O\text{ at }\mathfrak{p},\qquad\text{and then}\qquad
v_{\mathfrak{p}}(\rho(Q))=2v_{\mathfrak{p}}(z(Q)). \tag*{(10)} v p ( ρ ( Q )) ≥ 1 ⟺ Q reduces to O at p , and then v p ( ρ ( Q )) = 2 v p ( z ( Q )) . (10)
Proposition 2.3. Let p ∉ { 2 , 7 } \mathfrak{p}\notin\{2,7\} p ∈ / { 2 , 7 } , let p ∣ p \mathfrak{p}\mid\mathfrak{p} p ∣ p , and let n , n ′ ∈ Z n,n^\prime\in\mathbb{Z} n , n ′ ∈ Z with n ′ ≠ 0 n^\prime\ne0 n ′ = 0 be such that n P nP n P and n ′ P n^\prime P n ′ P reduce to O O O at p \mathfrak{p} p .
(1) v p ( z ( n P ) / z ( n ′ P ) ) = v p ( n / n ′ ) v_{\mathfrak{p}}(z(nP)/z(n^\prime P))=v_{\mathfrak{p}}(n/n^\prime) v p ( z ( n P ) / z ( n ′ P )) = v p ( n / n ′ ) .
(2) If K ≥ 1 K\ge1 K ≥ 1 , p > K + 1 \mathfrak{p}>K+1 p > K + 1 and v p ( n / n ′ ) ≥ 0 v_{\mathfrak{p}}(n/n^\prime)\ge0 v p ( n / n ′ ) ≥ 0 , then ℓ K ( z ( n ′ P ) ) ≠ 0 \ell_K(z(n^\prime P))\ne0 ℓ K ( z ( n ′ P )) = 0 and
v p ( ℓ K ( z ( n P ) ) ℓ K ( z ( n ′ P ) ) − n n ′ ) ≥ K . v_{\mathfrak{p}}\left(\frac{\ell_K(z(nP))}{\ell_K(z(n^\prime P))}-\frac{n}{n^\prime}\right)\ge K. v p ( ℓ K ( z ( n ′ P )) ℓ K ( z ( n P )) − n ′ n ) ≥ K .
The same holds for E σ E^\sigma E σ , P σ P^\sigma P σ and ℓ K σ \ell^\sigma_K ℓ K σ .
Proof. The coefficient c t c_t c t of Z t Z^t Z t in ℓ \ell ℓ vanishes unless t = 4 j + 1 t=4j+1 t = 4 j + 1 , and then v p ( c t ) ≥ − v p ( t ) v_p(c_t) \ge-v_p(t) v p ( c t ) ≥ − v p ( t ) . For v p ( Z ) ≥ 1 v_p(Z) \ge1 v p ( Z ) ≥ 1 and t ≥ 2 t \ge2 t ≥ 2 we have t v p ( Z ) − v p ( t ) > v p ( Z ) tv_p(Z)-v_p(t)>v_p(Z) t v p ( Z ) − v p ( t ) > v p ( Z ) , because t − 1 > v p ( t ) t-1>v_p(t) t − 1 > v p ( t ) for p ≥ 3 p \ge3 p ≥ 3 . Hence ℓ \ell ℓ converges on the kernel of reduction, satisfies v p ( ℓ ( Z ) ) = v p ( Z ) v_p(\ell(Z))=v_p(Z) v p ( ℓ ( Z )) = v p ( Z ) there, and is a homomorphism from the formal group [82 IV.5] . Applied to [ n ′ ] ( n P ) = [ n ] ( n ′ P ) [n'](nP)=[n](n'P) [ n ′ ] ( n P ) = [ n ] ( n ′ P ) , this gives n ′ ℓ ( z ( n P ) ) = n ℓ ( z ( n ′ P ) ) n'\ell(z(nP))=n\ell(z(n'P)) n ′ ℓ ( z ( n P )) = n ℓ ( z ( n ′ P )) , and (1) follows.
If p > K + 1 p>K+1 p > K + 1 and t > K t>K t > K , then t − 1 − v p ( t ) ≥ K t-1-v_p(t)\ge K t − 1 − v p ( t ) ≥ K : this is clear if p ∤ t p\nmid t p ∤ t , and if j = v p ( t ) ≥ 1 j=v_p(t)\ge1 j = v p ( t ) ≥ 1 then t − 1 − j ≥ p j − 1 − j ≥ p − 2 ≥ K t-1-j\ge p^j-1-j\ge p-2\ge K t − 1 − j ≥ p j − 1 − j ≥ p − 2 ≥ K . Hence, for v p ( Z ) ≥ 1 v_p(Z)\ge1 v p ( Z ) ≥ 1 ,
v p ( ℓ ( Z ) − ℓ K ( Z ) ) ≥ v p ( Z ) + K , v p ( ℓ K ( Z ) ) = v p ( Z ) . v_p\bigl(\ell(Z)-\ell_K(Z)\bigr)\ge v_p(Z)+K,\qquad v_p\bigl(\ell_K(Z)\bigr)=v_p(Z). v p ( ℓ ( Z ) − ℓ K ( Z ) ) ≥ v p ( Z ) + K , v p ( ℓ K ( Z ) ) = v p ( Z ) .
Put q = n / n ′ q=n/n' q = n / n ′ , so that v p ( z ( n P ) ) ≥ v p ( z ( n ′ P ) ) v_p(z(nP))\ge v_p(z(n'P)) v p ( z ( n P )) ≥ v p ( z ( n ′ P )) by (1). From ℓ ( z ( n P ) ) = q ℓ ( z ( n ′ P ) ) \ell(z(nP))=q\ell(z(n'P)) ℓ ( z ( n P )) = q ℓ ( z ( n ′ P )) we obtain v p ( ℓ K ( z ( n P ) ) − q ℓ K ( z ( n ′ P ) ) ) ≥ v p ( z ( n ′ P ) ) + K v_p\bigl(\ell_K(z(nP))-q\ell_K(z(n'P))\bigr)\ge v_p(z(n'P))+K v p ( ℓ K ( z ( n P )) − q ℓ K ( z ( n ′ P )) ) ≥ v p ( z ( n ′ P )) + K , and division by ℓ K ( z ( n ′ P ) ) \ell_K(z(n'P)) ℓ K ( z ( n ′ P )) , of valuation v p ( z ( n ′ P ) ) v_p(z(n'P)) v p ( z ( n ′ P )) , proves (2). □
Lemma 2.4. The map Q ↦ z ( Q ) Q\mapsto z(Q) Q ↦ z ( Q ) is injective on ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ .
Proof. If Q ∈ E ( F ) Q\in E(F) Q ∈ E ( F ) has x ( Q ) y ( Q ) ≠ 0 x(Q)y(Q)\ne0 x ( Q ) y ( Q ) = 0 and z ( Q ) = w z(Q)=w z ( Q ) = w , then x ( Q ) x(Q) x ( Q ) is a root of w 2 X 2 − X − d 2 w 2 w^2X^2-X-d^2w^2 w 2 X 2 − X − d 2 w 2 and y ( Q ) = − x ( Q ) / w y(Q)=-x(Q)/w y ( Q ) = − x ( Q ) / w ; so at most one other point has parameter w w w , namely T − Q T-Q T − Q with T = ( 0 , 0 ) T=(0,0) T = ( 0 , 0 ) , since translation by T T T is ( x , y ) ↦ ( − d 2 / x , d 2 y / x 2 ) (x,y)\mapsto(-d^2/x,d^2y/x^2) ( x , y ) ↦ ( − d 2 / x , d 2 y / x 2 ) . As T ∉ ⟨ P ⟩ T\notin\langle P\rangle T ∈ / ⟨ P ⟩ , at most one of Q Q Q and T − Q T-Q T − Q lies in ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ . Finally z z z vanishes on ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ only at O O O . □
Good primes
Definition 2.5. A prime r r r is good if r ≡ 1 ( m o d 8 ) r\equiv1\pmod8 r ≡ 1 ( mod 8 ) , ( r 7 ) = − 1 \left(\frac{r}{7}\right)=-1 ( 7 r ) = − 1 , and, writing r = a 2 + b 2 r=a^2+b^2 r = a 2 + b 2 with a a a odd, no odd prime divides both a a a and r + 1 r+1 r + 1 .
Lemma 2.6. Let r r r be a good prime. Then r r r splits in F F F , and at each place p \mathfrak p p of F F F above r r r the orders N 1 = # E ( O F / p ) N_1=\#E(\mathcal O_F/\mathfrak p) N 1 = # E ( O F / p ) and N 2 = # E σ ( O F / p ) N_2=\#E^\sigma(\mathcal O_F/\mathfrak p) N 2 = # E σ ( O F / p ) satisfy
gcd ( N 1 , N 2 ) = 4 , r ∤ N 1 N 2 . \gcd(N_1,N_2)=4,\qquad r\nmid N_1N_2. g cd( N 1 , N 2 ) = 4 , r ∤ N 1 N 2 . The orders of the reductions of P P P and P σ P^\sigma P σ are coprime and prime to r r r .
Proof. As r ≡ 1 ( m o d 8 ) r\equiv1\pmod8 r ≡ 1 ( mod 8 ) , r r r splits and O F / p = F r \mathcal O_F/\mathfrak p=\mathbb F_r O F / p = F r . The images d ‾ , d ′ ‾ \overline d,\overline{d'} d , d ′ of d , d σ d,d^\sigma d , d σ satisfy d ‾ d ′ ‾ = 7 \overline d\,\overline{d'}=7 d d ′ = 7 and ( 7 r ) = ( r 7 ) = − 1 \left(\frac{7}{r}\right)=\left(\frac{r}{7}\right)=-1 ( r 7 ) = ( 7 r ) = − 1 , so exactly one of them is a square. The reductions of E E E and E σ E^\sigma E σ are the quadratic twists of E 0 : y 2 = x 3 − x E_0:y^2=x^3-x E 0 : y 2 = x 3 − x by d ‾ \overline d d and d ′ ‾ \overline{d'} d ′ , so { N 1 , N 2 } = { r + 1 − a r , r + 1 + a r } \{N_1,N_2\}=\{r+1-a_r,r+1+a_r\} { N 1 , N 2 } = { r + 1 − a r , r + 1 + a r } , where a r = ± 2 a a_r=\pm2a a r = ± 2 a [38 Ch. 18, §4] . Both orders are divisible by 4 4 4 , the 2 2 2 -torsion being rational, and N 1 + N 2 = 2 ( r + 1 ) ≡ 4 ( m o d 16 ) N_1+N_2=2(r+1)\equiv4\pmod{16} N 1 + N 2 = 2 ( r + 1 ) ≡ 4 ( mod 16 ) , so the 2 2 2 -part of gcd ( N 1 , N 2 ) \gcd(N_1,N_2) g cd( N 1 , N 2 ) is 4 4 4 . An odd prime dividing both orders divides N 1 + N 2 N_1+N_2 N 1 + N 2 and N 1 − N 2 = ± 4 a N_1-N_2=\pm4a N 1 − N 2 = ± 4 a , hence r + 1 r+1 r + 1 and a a a , contrary to the definition of a good prime. By the Hasse bound 0 < N i < 2 r 0<N_i<2r 0 < N i < 2 r , and N i ≠ r N_i\ne r N i = r as 4 ∣ N i 4\mid N_i 4 ∣ N i ; so r ∤ N 1 N 2 r\nmid N_1N_2 r ∤ N 1 N 2 . Finally, the reductions of P = 2 P ~ P=2\widetilde P P = 2 P and P σ P^\sigma P σ lie in the subgroups of doubles, of coprime orders N 1 / 4 N_1/4 N 1 /4 and N 2 / 4 N_2/4 N 2 /4 . □
The local test
Fix integers ν , K ≥ 1 \nu,K\ge1 ν , K ≥ 1 , a polynomial f ∈ Z [ X 1 , … , X ν ] f\in\mathbb Z[X_1,\ldots,X_\nu] f ∈ Z [ X 1 , … , X ν ] , and points P 1 , … , P ν ∈ ⟨ P ⟩ P_1,\ldots,P_\nu\in\langle P\rangle P 1 , … , P ν ∈ ⟨ P ⟩ with indices n 1 , … , n ν n_1,\ldots,n_\nu n 1 , … , n ν .
Definition 2.7. A test datum is a tuple τ = ( H , U 1 , … , U ν , J ) \tau=(H,U_1,\ldots,U_\nu,J) τ = ( H , U 1 , … , U ν , J ) of points of ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ with J ≠ O J\ne O J = O and ℓ K ( z ( H ) ) ≠ 0 \ell_K(z(H))\ne0 ℓ K ( z ( H )) = 0 . Put L i = ℓ K ( z ( U i ) ) / ℓ K ( z ( H ) ) ∈ F L_i=\ell_K(z(U_i))/\ell_K(z(H))\in F L i = ℓ K ( z ( U i )) / ℓ K ( z ( H )) ∈ F . The datum satisfies the local conditions at a place p \mathfrak p p of F F F if
v p ( ρ ( H ) ) ≥ 1 , v p ( ρ ( U i ) ) ≥ 1 , v p ( ρ ( J σ ) ) ≥ 1 , v p ( ρ ( ( H − P ) σ ) ρ ( J σ ) ) ≥ 2 K , v p ( ρ ( ( U i − P i ) σ ) ρ ( J σ ) ) ≥ 2 K , (11) \begin{aligned}
v_{\mathfrak p}(\rho(H))&\ge1, & v_{\mathfrak p}(\rho(U_i))&\ge1, & v_{\mathfrak p}(\rho(J^\sigma))&\ge1,\\
v_{\mathfrak p}\left(\frac{\rho((H-P)^\sigma)}{\rho(J^\sigma)}\right)&\ge2K, &
v_{\mathfrak p}\left(\frac{\rho((U_i-P_i)^\sigma)}{\rho(J^\sigma)}\right)&\ge2K, \tag*{(11)}
\end{aligned} v p ( ρ ( H )) v p ( ρ ( J σ ) ρ (( H − P ) σ ) ) ≥ 1 , ≥ 2 K , v p ( ρ ( U i )) v p ( ρ ( J σ ) ρ (( U i − P i ) σ ) ) ≥ 1 , ≥ 2 K , v p ( ρ ( J σ )) ≥ 1 , (11)
v p ( f ( L 1 , … , L ν ) ) ≥ K v_{\mathfrak p}\bigl(f(L_1,\ldots,L_\nu)\bigr)\ge K v p ( f ( L 1 , … , L ν ) ) ≥ K
for 1 ≤ i ≤ ν 1 \le i \le\nu 1 ≤ i ≤ ν . We list the 2 ν + 4 2\nu+4 2 ν + 4 elements of F F F occurring here as D 1 ( τ ) , … , D 2 ν + 4 ( τ ) D_1(\tau),\ldots,D_{2\nu+4}(\tau) D 1 ( τ ) , … , D 2 ν + 4 ( τ ) , and the corresponding lower bounds as k 1 , … , k 2 ν + 4 ∈ { 1 , K , 2 K } k_1,\ldots,k_{2\nu+4} \in\{1,K,2K\} k 1 , … , k 2 ν + 4 ∈ { 1 , K , 2 K } .
The denominator ℓ K ( z ( H ) ) \ell_K(z(H)) ℓ K ( z ( H )) is non-zero by definition, and ρ ( J σ ) ≠ 0 \rho(J^\sigma) \ne0 ρ ( J σ ) = 0 because J ≠ O J \ne O J = O . For p ∣ p \mathfrak{p} \mid p p ∣ p with p ∉ { 2 , 7 } p \notin\{2,7\} p ∈ / { 2 , 7 } , the first three inequalities in (11) say that H H H and U i U_i U i reduce to the origin of E E E , and J σ J^\sigma J σ to the origin of E σ E^\sigma E σ , by (10) . Writing H = h P H=hP H = h P and U i = u i P U_i=u_iP U i = u i P , the next two imply h ≡ 1 h \equiv1 h ≡ 1 and u i ≡ n i ( m o d p K ) u_i \equiv n_i \pmod{p^K} u i ≡ n i ( mod p K ) , independently of the order of P σ P^\sigma P σ modulo p \mathfrak{p} p .
Since p ∤ h p \nmid h p ∤ h , Proposition 2.3(1) fixes the depth of H = h P H=hP H = h P in the kernel of reduction. The ratio z ( U i ) / z ( H ) z(U_i)/z(H) z ( U i ) / z ( H ) alone therefore approximates u i / h u_i/h u i / h to a bounded order uniformly in the test datum. The truncated logarithm yields the prescribed order K K K .
Lemma 2.8. Let p ∉ { 2 , 7 } p \notin\{2,7\} p ∈ / { 2 , 7 } be a prime with p > K + 1 p>K+1 p > K + 1 , and let p ∣ p \mathfrak{p}\mid p p ∣ p . If some test datum satisfies the local conditions at p \mathfrak{p} p , then p K ∣ f ( n 1 , … , n ν ) p^K \mid f(n_1,\ldots,n_\nu) p K ∣ f ( n 1 , … , n ν ) .
Proof. Write H = h P H=hP H = h P , U i = u i P U_i=u_iP U i = u i P and J = j P J=jP J = j P , with j ≠ 0 j \ne0 j = 0 . The conditions on ρ ( J σ ) \rho(J^\sigma) ρ ( J σ ) and ρ ( ( H − P ) σ ) / ρ ( J σ ) \rho((H-P)^\sigma)/\rho(J^\sigma) ρ (( H − P ) σ ) / ρ ( J σ ) say, by (10) , that j P σ jP^\sigma j P σ and ( h − 1 ) P σ (h-1)P^\sigma ( h − 1 ) P σ reduce to O O O at p \mathfrak{p} p and that v p ( z ( ( h − 1 ) P σ ) / z ( j P σ ) ) ≥ K v_p(z((h-1)P^\sigma)/z(jP^\sigma)) \ge K v p ( z (( h − 1 ) P σ ) / z ( j P σ )) ≥ K . By Proposition 2.3(1) for E σ E^\sigma E σ , v p ( ( h − 1 ) / j ) ≥ K v_p((h-1)/j) \ge K v p (( h − 1 ) / j ) ≥ K , hence v p ( h − 1 ) ≥ K v_p(h-1) \ge K v p ( h − 1 ) ≥ K ; likewise v p ( u i − n i ) ≥ K v_p(u_i-n_i) \ge K v p ( u i − n i ) ≥ K . So h h h is a p p p -adic unit and u i / h ≡ n i ( m o d p K Z ( p ) ) u_i/h \equiv n_i \pmod{p^K\mathbb{Z}_{(p)}} u i / h ≡ n i ( mod p K Z ( p ) ) . As H H H and U i U_i U i reduce to O O O at p \mathfrak{p} p , Proposition 2.3(2) with ( n , n ′ ) = ( u i , h ) (n,n')=(u_i,h) ( n , n ′ ) = ( u i , h ) gives v p ( L i − u i / h ) ≥ K v_p(L_i-u_i/h) \ge K v p ( L i − u i / h ) ≥ K . Thus L i ≡ n i ( m o d p K ) L_i \equiv n_i \pmod{p^K} L i ≡ n i ( mod p K ) , and
f ( n 1 , … , n ν ) ≡ f ( L 1 , … , L ν ) ≡ 0 ( m o d p K ) . f(n_1,\ldots,n_\nu) \equiv f(L_1,\ldots,L_\nu) \equiv0 \pmod{p^K}. f ( n 1 , … , n ν ) ≡ f ( L 1 , … , L ν ) ≡ 0 ( mod p K ) .
As f ( n 1 , … , n ν ) ∈ Z f(n_1,\ldots,n_\nu)\in\mathbb{Z} f ( n 1 , … , n ν ) ∈ Z , we get p K ∣ f ( n 1 , … , n ν ) p^K \mid f(n_1,\ldots,n_\nu) p K ∣ f ( n 1 , … , n ν ) .
Lemma 2.9. Assume f ( n 1 , … , n ν ) = 0 f(n_1,\ldots,n_\nu)=0 f ( n 1 , … , n ν ) = 0 . Let r > K + 1 r>K+1 r > K + 1 be a good prime, p ∣ r \mathfrak{p}\mid r p ∣ r , and J = j P J=jP J = j P with j ≠ 0 j\ne0 j = 0 such that J σ J^\sigma J σ reduces to O O O at p \mathfrak{p} p . Then some test datum of the form ( h P , h n 1 P , … , h n ν P , J ) (hP,hn_1P,\ldots,hn_\nu P,J) ( h P , h n 1 P , … , h n ν P , J ) with h ∈ Z h\in\mathbb{Z} h ∈ Z satisfies the local conditions at p \mathfrak{p} p .
Proof. Let o 1 o_1 o 1 and o 2 o_2 o 2 be the orders of the reductions of P P P and P σ P^\sigma P σ at p \mathfrak{p} p , so that o 2 ∣ j o_2\mid j o 2 ∣ j , and put e = v r ( j ) e=v_r(j) e = v r ( j ) . By Lemma 2.6, o 1 o_1 o 1 is prime to o 2 r K + e o_2r^{K+e} o 2 r K + e , so there is h ∈ Z h\in\mathbb{Z} h ∈ Z with
h ≡ 0 ( m o d o 1 ) , h ≡ 1 ( m o d o 2 r K + e ) ; h \equiv0 \pmod{o_1}, \qquad h \equiv1 \pmod{o_2r^{K+e}}; h ≡ 0 ( mod o 1 ) , h ≡ 1 ( mod o 2 r K + e ) ;
it is non-zero. Then h P hP h P and h n i P hn_iP h n i P reduce to O O O at p \mathfrak{p} p , and v p ( ℓ K ( z ( h P ) ) ) = v p ( z ( h P ) ) < ∞ v_{\mathfrak{p}}(\ell_K(z(hP)))=v_{\mathfrak{p}}(z(hP))<\infty v p ( ℓ K ( z ( h P ))) = v p ( z ( h P )) < ∞ . The points ( h − 1 ) P σ (h-1)P^\sigma ( h − 1 ) P σ and n i ( h − 1 ) P σ n_i(h-1)P^\sigma n i ( h − 1 ) P σ reduce to O O O on E σ E^\sigma E σ , and by Proposition 2.3(1) the corresponding ratios of z z z -coordinates to z ( J σ ) z(J^\sigma) z ( J σ ) have valuation at least v r ( ( h − 1 ) / j ) ≥ K v_r((h-1)/j)\ge K v r (( h − 1 ) / j ) ≥ K ; by (10) the ratios of ρ \rho ρ -coordinates have valuation at least 2 K 2K 2 K . By Proposition 2.3(2) with ( n , n ′ ) = ( h n i , h ) (n,n')=(hn_i,h) ( n , n ′ ) = ( h n i , h ) , L i ≡ n i ( m o d p K ) L_i\equiv n_i\pmod{p^K} L i ≡ n i ( mod p K ) , so f ( L 1 , … , L ν ) ≡ f ( n 1 , … , n ν ) = 0 ( m o d p K ) f(L_1,\ldots,L_\nu)\equiv f(n_1,\ldots,n_\nu)=0\pmod{p^K} f ( L 1 , … , L ν ) ≡ f ( n 1 , … , n ν ) = 0 ( mod p K ) .
Square denominators
In this section we prove Theorem C. We define a positive existential formula Φ \Phi Φ (see Definition 3.1) and show that
D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 \mathcal{D}_2 \subseteq\Phi(\mathbb{Q}) \subseteq\mathcal{D}_2^{S_1} D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1
for a finite set S 1 S_1 S 1 of primes (see Propositions 3.2 and 3.14); Theorem C follows by adding Robinson’s local conditions at the primes of S 1 S_1 S 1 . For ℓ ∈ Q × \ell\in\mathbb{Q}^\times ℓ ∈ Q × put
E ℓ : y 2 = x 3 − ℓ 2 x = x ( x − ℓ ) ( x + ℓ ) . E_\ell: y^2=x^3-\ell^2x=x(x-\ell)(x+\ell). E ℓ : y 2 = x 3 − ℓ 2 x = x ( x − ℓ ) ( x + ℓ ) .
The formula
Put
M = { ± 1 , ± 2 , ± 3 , ± 5 , ± 6 , ± 7 , ± 10 , ± 14 } ⊂ Z ∖ { 0 } . (12) \mathcal{M}=\{\pm1,\pm2,\pm3,\pm5,\pm6,\pm7,\pm10,\pm14\}\subset\mathbb{Z}\setminus\{0\}. \tag*{(12)} M = { ± 1 , ± 2 , ± 3 , ± 5 , ± 6 , ± 7 , ± 10 , ± 14 } ⊂ Z ∖ { 0 } . (12)
This set represents R × / R × 2 × Q 2 × / Q 2 × 2 \mathbb{R}^{\times}/\mathbb{R}^{\times2}\times\mathbb{Q}_2^{\times}/\mathbb{Q}_2^{\times2} R × / R × 2 × Q 2 × / Q 2 × 2 , so for every b ∈ Q × b\in\mathbb{Q}^{\times} b ∈ Q × some μ ∈ M \mu\in\mathcal{M} μ ∈ M satisfies
μ b > 0 , − μ b ∈ Q 2 × 2 . (13) \mu b>0,\qquad-\mu b\in\mathbb{Q}_2^{\times2}. \tag*{(13)} μ b > 0 , − μ b ∈ Q 2 × 2 . (13)
For μ ∈ M \mu\in\mathcal{M} μ ∈ M put S ( μ ) = { 2 } ∪ Supp ( μ ) ⊆ { 2 , 3 , 5 , 7 } S(\mu)=\{2\}\cup\operatorname{Supp}(\mu)\subseteq\{2,3,5,7\} S ( μ ) = { 2 } ∪ Supp ( μ ) ⊆ { 2 , 3 , 5 , 7 } , where Supp ( a ) = { p : v p ( a ) ≠ 0 } \operatorname{Supp}(a)=\{p:v_p(a)\ne0\} Supp ( a ) = { p : v p ( a ) = 0 } for a ∈ Q × a\in\mathbb{Q}^{\times} a ∈ Q × . By weak approximation, choose a finite set D ⊂ Q × \mathcal{D}\subset\mathbb{Q}^{\times} D ⊂ Q × such that for every b ∈ Q × b\in\mathbb{Q}^{\times} b ∈ Q × there exists δ ∈ D \delta\in\mathcal{D} δ ∈ D satisfying
δ b > 0 , δ b ∈ Q q × 2 ( q = 2 , 3 , 5 , 7 ) . (14) \delta b>0,\qquad\delta b\in\mathbb{Q}_q^{\times2}\qquad(q=2,3,5,7). \tag*{(14)} δ b > 0 , δ b ∈ Q q × 2 ( q = 2 , 3 , 5 , 7 ) . (14)
Let S 1 S_1 S 1 consist of 2 and the prime divisors of the numerators and denominators of the elements of M \mathcal{M} M and D \mathcal{D} D .
Definition 3.1. Let Φ ( b ) \Phi(b) Φ ( b ) hold if b = 0 b=0 b = 0 , or if b ≠ 0 b\ne0 b = 0 and there exist μ ∈ M \mu\in\mathcal{M} μ ∈ M , δ ∈ D \delta\in\mathcal{D} δ ∈ D and t , e ∈ Q t,e\in\mathbb{Q} t , e ∈ Q such that, with
b ′ = μ b , U = 1 − δ b t 2 μ ( 1 + δ b t 2 ) , H ′ = 1 − b ′ U 2 , (15) b'=\mu b,\qquad U=\frac{1-\delta b t^2}{\mu(1+\delta b t^2)},\qquad H'=1-b'U^2, \tag*{(15)} b ′ = μ b , U = μ ( 1 + δ b t 2 ) 1 − δ b t 2 , H ′ = 1 − b ′ U 2 , (15) one has 1 + δ b t 2 ≠ 0 1+\delta b t^2\ne0 1 + δ b t 2 = 0 , e H ′ ≠ 0 eH'\ne0 e H ′ = 0 , and each of E e E_e E e and E e H ′ E_{eH'} E e H ′ has a rational point with x x x -coordinate in b ′ Q × 2 b'\mathbb{Q}^{\times2} b ′ Q × 2 .
Thus H ′ H' H ′ is the norm of 1 + U b ′ 1+U\sqrt{b'} 1 + U b ′ from Q [ T ] / ( T 2 − b ′ ) \mathbb{Q}[T]/(T^2-b') Q [ T ] / ( T 2 − b ′ ) . The formula Φ \Phi Φ is positive existential: the choices of μ \mu μ and δ \delta δ form a finite disjunction, and x ∈ b ′ Q × 2 x\in b'\mathbb{Q}^{\times2} x ∈ b ′ Q × 2 is ∃ w ( x = b ′ w 2 ∧ w ≠ 0 ) \exists w\;(x=b'w^2\land w\ne0) ∃ w ( x = b ′ w 2 ∧ w = 0 ) .
The second multiplier δ \delta δ and the parameter t t t allow U U U to be small at the real place and at the primes in S ( μ ) S(\mu) S ( μ ) (see Proposition 3.14). At t = 0 t=0 t = 0 the expressions reduce to U = 1 / μ U=1/\mu U = 1/ μ and H ′ = 1 − b / μ H'=1-b/\mu H ′ = 1 − b / μ , independently of δ \delta δ .
Remark. For fixed b ≠ 0 b\ne0 b = 0 , μ \mu μ and δ \delta δ , write R = ( b ′ v 2 , v Y ) R=(b'v^2,vY) R = ( b ′ v 2 , v Y ) and R ′ = ( b ′ w 2 , w Z ) R'=(b'w^2,wZ) R ′ = ( b ′ w 2 , w Z ) . The point conditions in Definition 3.1 become
Y 2 = ( b ′ ) 3 v 4 − b ′ e 2 , Z 2 = ( b ′ ) 3 w 4 − b ′ e 2 ( H ′ ) 2 . Y^2=(b')^3v^4-b'e^2,\qquad Z^2=(b')^3w^4-b'e^2(H')^2. Y 2 = ( b ′ ) 3 v 4 − b ′ e 2 , Z 2 = ( b ′ ) 3 w 4 − b ′ e 2 ( H ′ ) 2 .
These equations define a variety over the ( t , e ) (t,e) ( t , e ) -plane. Its generic fibre is an open subset of a product of two genus-one curves.
Proposition 3.2. If Φ ( b ) \Phi(b) Φ ( b ) holds, then b ∈ D 2 S 1 b\in D_2^{S_1} b ∈ D 2 S 1 .
Proof. Let b ≠ 0 b\ne0 b = 0 satisfy Φ \Phi Φ with witnesses as above, and let q ∉ S 1 q\notin S_1 q ∈ / S 1 be a prime with k = v q ( b ) k=v_q(b) k = v q ( b ) negative and odd; thus q q q is odd and μ , δ \mu,\delta μ , δ are q q q -units. If t = 0 t=0 t = 0 then U = 1 / μ U=1/\mu U = 1/ μ . If t ≠ 0 t\ne0 t = 0 , the valuation v q ( δ b t 2 ) = k + 2 v q ( t ) v_q(\delta b t^2)=k+2v_q(t) v q ( δ b t 2 ) = k + 2 v q ( t ) is odd: if it is positive, 1 ± δ b t 2 1\pm\delta b t^2 1 ± δ b t 2 are units, and if it is negative, both have valuation v q ( δ b t 2 ) v_q(\delta b t^2) v q ( δ b t 2 ) . In every case U U U is a q q q -adic unit, so v q ( b ′ U 2 ) = k < 0 v_q(b'U^2)=k<0 v q ( b ′ U 2 ) = k < 0 and v q ( H ′ ) = k v_q(H')=k v q ( H ′ ) = k is odd. Hence v q ( e ) v_q(e) v q ( e ) and v q ( e H ′ ) v_q(eH') v q ( e H ′ ) have opposite parities. Let l ∈ { e , e H ′ } l\in\{e,eH'\} l ∈ { e , e H ′ } have v q ( l ) = 2 j v_q(l)=2j v q ( l ) = 2 j . The substitution l = q 2 j L l=q^{2j}L l = q 2 j L , x = q 2 j X x=q^{2j}X x = q 2 j X , y = q 3 j Y y=q^{3j}Y y = q 3 j Y turns the given point of E l E_l E l into a point of Y 2 = X ( X − L ) ( X + L ) Y^2=X(X-L)(X+L) Y 2 = X ( X − L ) ( X + L ) with L ∈ Z q × L\in\mathbb{Z}_q^{\times} L ∈ Z q × and X ∈ b ′ Q q × 2 X\in b'\mathbb{Q}_q^{\times2} X ∈ b ′ Q q × 2 , so v q ( X ) v_q(X) v q ( X ) is odd. But a non-zero X X X on this curve has even valuation: if v q ( X ) < 0 v_q(X)<0 v q ( X ) < 0 the right side has valuation 3 v q ( X ) 3v_q(X) 3 v q ( X ) , and if v q ( X ) > 0 v_q(X)>0 v q ( X ) > 0 the factors X ± L X\pm L X ± L are units, as q q q is odd. This is a contradiction. □
Two-descent on congruent number curves
In §§3.2–3.4 we assume
l is a positive squarefree integer, and l ≡ 5 ( m o d 8 ) , (16) l\text{ is a positive squarefree integer, and }l\equiv5\pmod8, \tag*{(16)} l is a positive squarefree integer, and l ≡ 5 ( mod 8 ) , (16)
and we write Q \mathbb{Q} Q for the set of prime divisors of l l l . The points of order two of E l E_l E l are ( 0 , 0 ) (0,0) ( 0 , 0 ) and ( ± l , 0 ) (\pm l,0) ( ± l , 0 ) . For a field k ⊇ Q k\supseteq\mathbb{Q} k ⊇ Q the Kummer map
δ 2 : E l ( k ) ⟶ ( k × / k × 2 ) 2 , ( x , y ) ⟼ ( x , x − l ) ( x ≠ 0 , l ) , \delta_2:E_l(k)\longrightarrow(k^{\times}/k^{\times2})^2,\qquad(x,y)\longmapsto(x,x-l)\quad(x\ne0,l), δ 2 : E l ( k ) ⟶ ( k × / k × 2 ) 2 , ( x , y ) ⟼ ( x , x − l ) ( x = 0 , l ) ,
extended by δ 2 ( O ) = ( 1 , 1 ) \delta_2(O)=(1,1) δ 2 ( O ) = ( 1 , 1 ) , δ 2 ( 0 , 0 ) = ( − 1 , − ℓ ) \delta_2(0,0)=(-1,-\ell) δ 2 ( 0 , 0 ) = ( − 1 , − ℓ ) and δ 2 ( ℓ , 0 ) = ( ℓ , 2 ) \delta_2(\ell,0)=(\ell,2) δ 2 ( ℓ , 0 ) = ( ℓ , 2 ) , is a homomorphism with kernel 2 E ℓ ( k ) 2E_\ell(k) 2 E ℓ ( k ) [82 X.1.4] ; it sends ( − ℓ , 0 ) (-\ell,0) ( − ℓ , 0 ) to ( − ℓ , − 2 ℓ ) (-\ell,-2\ell) ( − ℓ , − 2 ℓ ) . Let
T ℓ = { ( 1 , 1 ) , ( − 1 , − ℓ ) , ( ℓ , 2 ) , ( − ℓ , − 2 ℓ ) } . \mathcal{T}_{\ell}=\{(1,1),(-1,-\ell),(\ell,2),(-\ell,-2\ell)\}. T ℓ = {( 1 , 1 ) , ( − 1 , − ℓ ) , ( ℓ , 2 ) , ( − ℓ , − 2 ℓ )} .
be the image of E ℓ ( Q ) [ 2 ] E_\ell(\mathbb{Q})[2] E ℓ ( Q ) [ 2 ] . The 2-Selmer group Sel 2 ( E ℓ ) \operatorname{Sel}_2(E_\ell) Sel 2 ( E ℓ ) is the group of classes c ∈ ( Q × / Q × 2 ) 2 c\in(\mathbb{Q}^{\times}/\mathbb{Q}^{\times2})^2 c ∈ ( Q × / Q × 2 ) 2 whose image in ( Q v × / Q v × 2 ) 2 (\mathbb{Q}_v^{\times}/\mathbb{Q}_v^{\times2})^2 ( Q v × / Q v × 2 ) 2 lies in I v = δ 2 ( E ℓ ( Q v ) ) I_v=\delta_2(E_\ell(\mathbb{Q}_v)) I v = δ 2 ( E ℓ ( Q v )) for every place v v v ; it contains δ 2 ( E ℓ ( Q ) ) \delta_2(E_\ell(\mathbb{Q})) δ 2 ( E ℓ ( Q )) , with quotient \Sha(E_\ell)[2] [82 X.4.2] .
Lemma 3.3. Assume (3.5).
For a prime q ∤ 2 ℓ q\nmid2\ell q ∤ 2 ℓ , I q I_q I q is the group of classes of pairs of q q q -adic units.
For q ∈ Q q\in\mathcal{Q} q ∈ Q , I q I_q I q is the image of T ℓ \mathcal{T}_{\ell} T ℓ .
I 2 I_2 I 2 is generated by ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) , ( − 1 , − ℓ ) (-1,-\ell) ( − 1 , − ℓ ) and ( ℓ , 2 ) (\ell,2) ( ℓ , 2 ) .
I ∞ = { ( 1 , 1 ) , ( − 1 , − 1 ) } I_\infty=\{(1,1),(-1,-1)\} I ∞ = {( 1 , 1 ) , ( − 1 , − 1 )} .
Proof. For a prime q q q , E ℓ ( Q q ) E_\ell(\mathbb{Q}_q) E ℓ ( Q q ) has a subgroup of finite index isomorphic to Z q \mathbb{Z}_q Z q [82 VII.6.3] . Since ∣ G / 2 G ∣ / ∣ G [ 2 ] ∣ |G/2G|/|G[2]| ∣ G /2 G ∣/∣ G [ 2 ] ∣ is multiplicative in exact sequences and equals one for finite groups, ∣ I q ∣ = 4 |I_q|=4 ∣ I q ∣ = 4 for odd q q q and ∣ I 2 ∣ = 8 |I_2|=8 ∣ I 2 ∣ = 8 .
(1) If v q ( x ) < 0 v_q(x)<0 v q ( x ) < 0 , the three factors of x ( x − ℓ ) ( x + ℓ ) x(x-\ell)(x+\ell) x ( x − ℓ ) ( x + ℓ ) have valuation v q ( x ) v_q(x) v q ( x ) , which is therefore even. If x x x is integral, at most one factor is divisible by q q q , as 0 , ± ℓ 0,\pm\ell 0 , ± ℓ are distinct modulo q q q , and that factor has even valuation. So I q I_q I q lies in the group of unit classes, of order 4.
(2) At q q q the classes ( − 1 , − ℓ ) (-1,-\ell) ( − 1 , − ℓ ) , ( ℓ , 2 ) (\ell,2) ( ℓ , 2 ) , ( − ℓ , − 2 ℓ ) (-\ell,-2\ell) ( − ℓ , − 2 ℓ ) have valuations ( 0 , 1 ) (0,1) ( 0 , 1 ) , ( 1 , 0 ) (1,0) ( 1 , 0 ) , ( 1 , 1 ) (1,1) ( 1 , 1 ) , so T ℓ \mathcal{T}_{\ell} T ℓ maps injectively into I q I_q I q .
(3) Since ℓ ≡ 5 ( m o d 8 ) \ell\equiv5\pmod{8} ℓ ≡ 5 ( mod 8 ) , there is s ∈ Q 2 × s\in\mathbb{Q}_2^\times s ∈ Q 2 × with 5 = s 2 ℓ 5=s^2\ell 5 = s 2 ℓ , and ( x , y ) ↦ ( s 2 x , s 3 y ) (x,y)\mapsto(s^2x,s^3y) ( x , y ) ↦ ( s 2 x , s 3 y ) is an isomorphism E ℓ → E 5 E_\ell\to E_5 E ℓ → E 5 over Q 2 \mathbb{Q}_2 Q 2 that multiplies x x x and x − ℓ x-\ell x − ℓ by s 2 s^2 s 2 . So I 2 I_2 I 2 is the local image for E 5 E_5 E 5 . It contains δ 2 ( 0 , 0 ) = ( − 1 , − 5 ) \delta_2(0,0)=(-1,-5) δ 2 ( 0 , 0 ) = ( − 1 , − 5 ) , δ 2 ( 5 , 0 ) = ( 5 , 2 ) \delta_2(5,0)=(5,2) δ 2 ( 5 , 0 ) = ( 5 , 2 ) and δ 2 ( − 4 , 6 ) = ( − 4 , − 9 ) \delta_2(-4,6)=(-4,-9) δ 2 ( − 4 , 6 ) = ( − 4 , − 9 ) , which is the class of ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) . These generate a group of order 8, as − 1 , 5 , 2 -1,5,2 − 1 , 5 , 2 are independent in Q 2 × / Q 2 × 2 \mathbb{Q}_2^\times/\mathbb{Q}_2^{\times2} Q 2 × / Q 2 × 2 .
(4) A real point has x ∈ [ − ℓ , 0 ] ∪ [ ℓ , ∞ ) x\in[-\ell,0]\cup[\ell,\infty) x ∈ [ − ℓ , 0 ] ∪ [ ℓ , ∞ ) ; for x > ℓ x>\ell x > ℓ both coordinates of δ 2 \delta_2 δ 2 are positive, and for − ℓ < x < 0 -\ell<x<0 − ℓ < x < 0 both are negative.
The Selmer group
We encode classes supported on Q \mathcal{Q} Q by vectors in F 2 Q \mathbb{F}_2^{\mathcal{Q}} F 2 Q : for u ∈ F 2 Q u\in\mathbb{F}_2^{\mathcal{Q}} u ∈ F 2 Q put q u = ∏ q ∈ Q q u q q^u=\prod_{q\in\mathcal{Q}}q^{u_q} q u = ∏ q ∈ Q q u q . Let 1 ∈ F 2 Q \mathbf{1}\in\mathbb{F}_2^{\mathcal{Q}} 1 ∈ F 2 Q be the vector with all entries 1, let 1 B \mathbf{1}_B 1 B be the indicator vector of B ⊆ Q B\subseteq\mathcal{Q} B ⊆ Q , and for w ∈ F 2 Q w\in\mathbb{F}_2^{\mathcal{Q}} w ∈ F 2 Q let D w D_w D w be the diagonal matrix with diagonal w w w .
Definition 3.4. For a finite set Q \mathcal{Q} Q of odd primes define y , z ∈ F 2 Q y,z\in\mathbb{F}_2^{\mathcal{Q}} y , z ∈ F 2 Q and A ∈ M Q ( F 2 ) A\in M_{\mathcal{Q}}(\mathbb{F}_2) A ∈ M Q ( F 2 ) by
( − 1 q ) = ( − 1 ) y q , ( 2 q ) = ( − 1 ) z q , ( q ′ q ) = ( − 1 ) A q q ′ ( q ≠ q ′ ) , \left(\frac{-1}{q}\right)=(-1)^{y_q},\qquad
\left(\frac{2}{q}\right)=(-1)^{z_q},\qquad
\left(\frac{q'}{q}\right)=(-1)^{A_{qq'}}\quad(q\ne q'), ( q − 1 ) = ( − 1 ) y q , ( q 2 ) = ( − 1 ) z q , ( q q ′ ) = ( − 1 ) A q q ′ ( q = q ′ ) ,
with diagonal entries chosen so that A 1 = 0 A\mathbf{1}=0 A 1 = 0 . For ℓ \ell ℓ as in (3.5), let N ( ℓ ) ⊆ F 2 Q × F 2 Q N(\ell)\subseteq\mathbb{F}_2^{\mathcal{Q}}\times\mathbb{F}_2^{\mathcal{Q}} N ( ℓ ) ⊆ F 2 Q × F 2 Q be the kernel of
M 0 = ( A D y D z A + D y ) . M_0=\begin{pmatrix}
A & D_y\\
D_z & A+D_y
\end{pmatrix}. M 0 = ( A D z D y A + D y ) .
Matrices of this kind go back to Monsky [35 Appendix] .
Lemma 3.5. For every finite set Q \mathcal{Q} Q of odd primes, A 1 = 0 A\mathbf{1}=0 A 1 = 0 and A + A t = y y t + D y A+A^t=yy^t+D_y A + A t = y y t + D y . Under (3.5), moreover, y t 1 = 0 y^t\mathbf{1}=0 y t 1 = 0 , z t 1 = 1 z^t\mathbf{1}=1 z t 1 = 1 and A t 1 = y A^t\mathbf{1}=y A t 1 = y .
Proof. Off the diagonal, A + A t A+A^t A + A t has entries A q q ′ + A q ′ q = y q y q ′ A_{qq'}+A_{q'q}=y_qy_{q'} A q q ′ + A q ′ q = y q y q ′ by quadratic reciprocity; on the diagonal both sides vanish. The congruence ℓ ≡ 1 ( m o d 4 ) \ell\equiv1\pmod4 ℓ ≡ 1 ( mod 4 ) says that an even number of q ∈ Q q\in\mathcal{Q} q ∈ Q are 3 modulo 4, and ∏ q ∈ Q ( 2 q ) = − 1 \prod_{q\in\mathcal{Q}}\left(\frac{2}{q}\right)=-1 ∏ q ∈ Q ( q 2 ) = − 1 that an odd number of them are ± 3 \pm3 ± 3 modulo 8. Finally A t 1 = A 1 + ( y y t + D y ) 1 = y ( y t 1 ) + y = y A^t\mathbf{1}=A\mathbf{1}+(yy^t+D_y)\mathbf{1}=y(y^t\mathbf{1})+y=y A t 1 = A 1 + ( y y t + D y ) 1 = y ( y t 1 ) + y = y .
Proposition 3.6. Assume (3.5). The map ( u , v ) ↦ ( q u , q v ) (u,v)\mapsto(q^u,q^v) ( u , v ) ↦ ( q u , q v ) sends N ( ℓ ) N(\ell) N ( ℓ ) isomorphically onto a complement of T ℓ \mathcal{T}_{\ell} T ℓ in Sel 2 ( E ℓ ) \operatorname{Sel}_2(E_\ell) Sel 2 ( E ℓ ) . In particular dim Sel 2 ( E ℓ ) = 2 + dim N ( ℓ ) \dim\operatorname{Sel}_2(E_\ell)=2+\dim N(\ell) dim Sel 2 ( E ℓ ) = 2 + dim N ( ℓ ) .
Proof. Multiplication by ( − 1 , − ℓ ) (-1,-\ell) ( − 1 , − ℓ ) changes the sign of the first coordinate, and multiplication by ( ℓ , 2 ) (\ell,2) ( ℓ , 2 ) the parity of the 2-adic valuation of the second. So every coset of T ℓ \mathcal{T}_{\ell} T ℓ contains exactly one class ( d 1 , d 2 ) (d_1,d_2) ( d 1 , d 2 ) with d 1 > 0 d_1>0 d 1 > 0 and v 2 ( d 2 ) v_2(d_2) v 2 ( d 2 ) even; call such classes normalized. By Lemma 3.3 (1) and (4), a normalized class in Sel 2 ( E ℓ ) \operatorname{Sel}_2(E_\ell) Sel 2 ( E ℓ ) has the form ( 2 a q u , q v ) (2^a q^u,q^v) ( 2 a q u , q v ) . It remains to see that ( 2 a q u , q v ) (2^a q^u,q^v) ( 2 a q u , q v ) satisfies the local conditions at 2 and at Q \mathcal{Q} Q exactly when a = 0 a=0 a = 0 and ( u , v ) ∈ N ( ℓ ) (u,v)\in N(\ell) ( u , v ) ∈ N ( ℓ ) .
Give a class 2 a m 2^a m 2 a m of Q 2 × / Q 2 × 2 \mathbb{Q}_2^\times/\mathbb{Q}_2^{\times2} Q 2 × / Q 2 × 2 , with m m m odd, the coordinates ( a , ϵ ( m ) , ω ( m ) ) ∈ F 2 3 (a,\epsilon(m),\omega(m))\in\mathbb{F}_2^3 ( a , ϵ ( m ) , ω ( m )) ∈ F 2 3 , where ϵ ( m ) = 1 \epsilon(m)=1 ϵ ( m ) = 1 if m ≡ 3 ( m o d 4 ) m\equiv3\pmod4 m ≡ 3 ( mod 4 ) and ω ( m ) = 1 \omega(m)=1 ω ( m ) = 1 if m ≡ ± 3 ( m o d 8 ) m\equiv\pm3\pmod8 m ≡ ± 3 ( mod 8 ) . Then ϵ ( q u ) = y t u \epsilon(q^u)=y^t u ϵ ( q u ) = y t u and ω ( q u ) = z t u \omega(q^u)=z^t u ω ( q u ) = z t u . Since ℓ ≡ 5 ( m o d 8 ) \ell\equiv5\pmod8 ℓ ≡ 5 ( mod 8 ) , the generators of I 2 I_2 I 2 in Lemma 3.3 (3) have coordinates ( ( 0 , 1 , 0 ) , ( 0 , 1 , 0 ) ) ((0,1,0),(0,1,0)) (( 0 , 1 , 0 ) , ( 0 , 1 , 0 )) , ( ( 0 , 1 , 0 ) , ( 0 , 1 , 1 ) ) ((0,1,0),(0,1,1)) (( 0 , 1 , 0 ) , ( 0 , 1 , 1 )) and ( ( 0 , 0 , 1 ) , ( 1 , 0 , 0 ) ) ((0,0,1),(1,0,0)) (( 0 , 0 , 1 ) , ( 1 , 0 , 0 )) . Hence I 2 I_2 I 2 consists of the pairs ( ( 0 , α , γ ) , ( γ , α , α ′ ) ) ((0,\alpha,\gamma),(\gamma,\alpha,\alpha')) (( 0 , α , γ ) , ( γ , α , α ′ )) . Thus ( 2 a q u , q v ) ∈ I 2 (2^a q^u,q^v)\in I_2 ( 2 a q u , q v ) ∈ I 2 exactly when
a = 0 , z t u = 0 , y t u = y t v . (17) a=0,\qquad z^t u=0,\qquad y^t u=y^t v. \tag*{(17)} a = 0 , z t u = 0 , y t u = y t v . (17)
Let a = 0 a=0 a = 0 and q ∈ Q q\in\mathcal{Q} q ∈ Q . By Lemma 3.3 (2), the class ( q u , q v ) (q^u,q^v) ( q u , q v ) lies in I q I_q I q if and only if it equals ( − 1 , − ℓ ) v q ( ℓ , 2 ) u q (-1,-\ell)^{v_q}( \ell,2)^{u_q} ( − 1 , − ℓ ) v q ( ℓ , 2 ) u q in ( Q q × / Q q × 2 ) 2 (\mathbb{Q}_q^\times/\mathbb{Q}_q^{\times2})^2 ( Q q × / Q q × 2 ) 2 , the exponents being determined by the valuations at q q q . Up to squares, the quotients of the two sides are the q q q -adic units
( − 1 ) v q ∏ q ′ ≠ q q ′ u q ′ + u q , ( − 1 ) v q 2 u q ∏ q ′ ≠ q q ′ v q ′ + v q . (-1)^{v_q}\prod_{q'\ne q}q'^{u_{q'}+u_q},\qquad(-1)^{v_q}2^{u_q}\prod_{q'\ne q}q'^{v_{q'}+v_q}. ( − 1 ) v q q ′ = q ∏ q ′ u q ′ + u q , ( − 1 ) v q 2 u q q ′ = q ∏ q ′ v q ′ + v q .
and they are squares exactly when the q q q -th entries of A u + D y v Au+D_yv A u + D y v and of ( A + D y ) v + D z u (A+D_y)v+D_zu ( A + D y ) v + D z u vanish. Summing the entries of these two vectors and using 1 t A = y t 1^tA=y^t 1 t A = y t (see Lemma 3.5 ), the conditions at Q \mathcal{Q} Q imply y t ( u + v ) = 0 y^t(u+v)=0 y t ( u + v ) = 0 and z t u = 0 z^tu=0 z t u = 0 , which are the remaining conditions in (17) . Conversely, for ( u , v ) ∈ N ( ℓ ) (u,v)\in N(\ell) ( u , v ) ∈ N ( ℓ ) the class ( q u , q v ) (q^u,q^v) ( q u , q v ) satisfies all local conditions. □
Rational points
We express the analytic rank criterion of Tian, Yuan and Zhang using the symmetric matrix and vector
M 1 = ( A + A t A t A D z ) , w = ( y + z 0 ) , M_1=\begin{pmatrix} A+A^t & A^t \\ A & D_z \end{pmatrix},\qquad
w=\begin{pmatrix} y+z \\ 0 \end{pmatrix}, M 1 = ( A + A t A A t D z ) , w = ( y + z 0 ) ,
indexed by two copies of Q \mathcal{Q} Q .
Theorem 3.7 (Tian–Yuan–Zhang). Let ℓ \ell ℓ satisfy (16) . If det ( M 1 w w t 0 ) ≠ 0 \det\begin{pmatrix}M_1&w\\w^t&0\end{pmatrix}\ne0 det ( M 1 w t w 0 ) = 0 , then E ℓ E_\ell E ℓ has analytic rank one.
This is the criterion from the first genus sum in [87 Theorem 1.4] , in Smith’s determinant formulation [83 Theorem 2.2, Table 2, row 5(a)] . Appendix § 12 proves the determinant identity directly. The analytic criterion rests on Waldspurger’s formula [92 ] and the generalized Gross–Zagier formula of Yuan, Zhang and Zhang [YZZ13].
Lemma 3.8. Let M M M be a symmetric matrix over F 2 \mathbb{F}_2 F 2 and w w w a vector. If ker M \ker M ker M is spanned by a vector κ \kappa κ with w t κ = 1 w^t\kappa=1 w t κ = 1 , then det ( M w w t 0 ) ≠ 0 \det\begin{pmatrix}M&w\\w^t&0\end{pmatrix}\ne0 det ( M w t w 0 ) = 0 .
Proof. Let M x + a w = 0 Mx+aw=0 M x + a w = 0 and w t x = 0 w^tx=0 w t x = 0 . Pairing the first equation with κ \kappa κ , and using κ t M = 0 \kappa^tM=0 κ t M = 0 , gives a = 0 a=0 a = 0 . Then x = b κ x=b\kappa x = bκ with b ∈ F 2 b\in\mathbb{F}_2 b ∈ F 2 , and w t x = b = 0 w^tx=b=0 w t x = b = 0 . □
Lemma 3.9. Assume (16) . The maps
φ ( u , v ) = ( u + v + ( y t v ) 1 , u ) , ψ ( u ′ , v ′ ) = ( v ′ , u ′ + v ′ + ( y t ( u ′ + v ′ ) ) 1 ) \varphi(u,v)=(u+v+(y^tv)1,u),\qquad\psi(u',v')=(v',u'+v'+(y^t(u'+v'))1) φ ( u , v ) = ( u + v + ( y t v ) 1 , u ) , ψ ( u ′ , v ′ ) = ( v ′ , u ′ + v ′ + ( y t ( u ′ + v ′ )) 1 )
are mutually inverse isomorphisms between N ( ℓ ) N(\ell) N ( ℓ ) and ker M 1 \ker M_1 ker M 1 .
Proof. Put T = I + 1 y t T=I+1y^t T = I + 1 y t . By Lemma 3.5 , T 2 = I T^2=I T 2 = I , A T = A AT=A A T = A and ( A + A t ) T = D y (A+A^t)T=D_y ( A + A t ) T = D y . In block form φ \varphi φ and ψ \psi ψ have matrices ( I T I 0 ) \begin{pmatrix}I&T\\I&0\end{pmatrix} ( I I T 0 ) and ( 0 I T I ) \begin{pmatrix}0&I\\T&I\end{pmatrix} ( 0 T I I ) , inverse to each other, and
M 1 ( I T I 0 ) = ( A D y A + D z A ) = ( I 0 I I ) M 0 . M_1\begin{pmatrix}I&T\\I&0\end{pmatrix}
=\begin{pmatrix}A&D_y\\A+D_z&A\end{pmatrix}
=\begin{pmatrix}I&0\\I&I\end{pmatrix}M_0. M 1 ( I I T 0 ) = ( A A + D z D y A ) = ( I I 0 I ) M 0 .
So φ \varphi φ maps ker M 0 = N ( l ) \ker M_0=N(l) ker M 0 = N ( l ) onto ker M 1 \ker M_1 ker M 1 . □
Corollary 3.10. Assume (16) , and suppose that N ( l ) N(l) N ( l ) is spanned by a vector ( u , v ) (u,v) ( u , v ) with y t u + z t v = 1 y^{t}u+z^{t}v=1 y t u + z t v = 1 . Then E l ( Q ) E_l(\mathbb{Q}) E l ( Q ) has rank one, \Sha(E_l) is finite, Sel 2 ( E l ) = δ 2 ( E l ( Q ) ) \operatorname{Sel}_2(E_l)=\delta_2(E_l(\mathbb{Q})) Sel 2 ( E l ) = δ 2 ( E l ( Q )) , and E l E_l E l has a rational point with x x x -coordinate in q u Q × 2 q^{u}\mathbb{Q}^{\times2} q u Q × 2 .
Proof. By Lemma 3.9, the kernel of M 1 M_1 M 1 is spanned by φ ( u , v ) \varphi(u,v) φ ( u , v ) . The relations z t u = 0 z^{t}u=0 z t u = 0 , y t 1 = 0 y^{t}1=0 y t 1 = 0 and z t 1 = 1 z^{t}1=1 z t 1 = 1 imply
w t φ ( u , v ) = y t u + z t v = 1. w^{t}\varphi(u,v)=y^{t}u+z^{t}v=1. w t φ ( u , v ) = y t u + z t v = 1.
By Lemma 3.8 and Theorem 3.7, E l E_l E l has analytic rank one. By the theorems of Gross–Zagier [33 ] and Kolyvagin [43 ] , E l ( Q ) E_l(\mathbb{Q}) E l ( Q ) has rank one and \Sha(E_l) is finite (see also [87 p. 722] ). By Proposition 3.6, dim Sel 2 ( E l ) = 3 = dim E l ( Q ) / 2 E l ( Q ) \dim\operatorname{Sel}_2(E_l)=3=\dim E_l(\mathbb{Q})/2E_l(\mathbb{Q}) dim Sel 2 ( E l ) = 3 = dim E l ( Q ) /2 E l ( Q ) , so \Sha(E_l)[2]=0 , and ( q u , q v ) = δ 2 ( R ) (q^{u},q^{v})=\delta_2(R) ( q u , q v ) = δ 2 ( R ) for some R ∈ E l ( Q ) R\in E_l(\mathbb{Q}) R ∈ E l ( Q ) . As ( q u , q v ) ∉ T l (q^{u},q^{v})\notin\mathcal{T}_l ( q u , q v ) ∈ / T l , R R R is not of order dividing two, and x ( R ) ∈ q u Q × 2 x(R)\in q^{u}\mathbb{Q}^{\times2} x ( R ) ∈ q u Q × 2 . □
When N ( l ) N(l) N ( l ) is a line spanned by ( u , v ) (u,v) ( u , v ) , the bordered determinant is non-zero precisely when y t u + z t v = 1 y^{t}u+z^{t}v=1 y t u + z t v = 1 .
Remark. Realizing prescribed Selmer classes by rational points also occurs in Swinnerton-Dyer’s method for genus-one fibrations and Kummer surfaces [8 , 85 ] . The Kummer case is related to Skorobogatov’s conjecture that the Brauer–Manin obstruction is the only obstruction to the Hasse principle for K3 surfaces. For later developments under finiteness hypotheses for the relevant 2-primary Tate–Shafarevich groups, see Harpaz–Skorobogatov, Harpaz and Morgan [37 , 34 , 51 ] .
Choosing the quadratic twists
Throughout this subsection, β , η \beta,\eta β , η are positive squarefree integers with disjoint sets of prime divisors B B B and J J J , such that
β ≡ 7 ( m o d 8 ) , η ≡ 1 ( m o d 8 ) , η ∈ Q b × 2 ( b ∈ B ) , β ∈ Q j × 2 ( j ∈ J ) . (18) \beta\equiv7\pmod8,\qquad\eta\equiv1\pmod8,\qquad\eta\in\mathbb{Q}_b^{\times2}\ (b\in B),\qquad\beta\in\mathbb{Q}_j^{\times2}\ (j\in J). \tag*{(18)} β ≡ 7 ( mod 8 ) , η ≡ 1 ( mod 8 ) , η ∈ Q b × 2 ( b ∈ B ) , β ∈ Q j × 2 ( j ∈ J ) . (18)
For a finite set of odd primes, let y , z , A y,z,A y , z , A be as in Definition 3.4, and write y B y_B y B , z B z_B z B , A J A_J A J , … \ldots … for the data attached to B B B and J J J . By (18) and reciprocity,
y B t 1 = 1 , z B t 1 = 0 , y J t 1 = z J t 1 = 0 , ∏ j ∈ J ( j b ) = 1 ( b ∈ B ) , ∏ b ∈ B ( b j ) = 1 ( j ∈ J ) . (19) y_B^{t}1=1,\qquad z_B^{t}1=0,\qquad y_J^{t}1=z_J^{t}1=0,\qquad\prod_{j\in J}\left(\frac{j}{b}\right)=1\ (b\in B),\qquad\prod_{b\in B}\left(\frac{b}{j}\right)=1\ (j\in J). \tag*{(19)} y B t 1 = 1 , z B t 1 = 0 , y J t 1 = z J t 1 = 0 , j ∈ J ∏ ( b j ) = 1 ( b ∈ B ) , b ∈ B ∏ ( j b ) = 1 ( j ∈ J ) . (19)
Proposition 3.11. There exists a positive squarefree integer n ≡ 5 ( m o d 8 ) n\equiv5\pmod8 n ≡ 5 ( mod 8 ) , divisible by β \beta β and prime to η \eta η , such that N ( n ) N(n) N ( n ) and N ( n η ) N(n\eta) N ( n η ) are one-dimensional, each spanned by a vector ( u , v ) (u,v) ( u , v ) with u = 1 B u=1_B u = 1 B and z t v = 0 z^{t}v=0 z t v = 0 .
We take n = β q 0 ∏ q ∈ I q n=\beta q_0\prod_{q\in I}q n = β q 0 ∏ q ∈ I q , where I I I is a set of ∣ B ∣ − 1 |B|-1 ∣ B ∣ − 1 primes q ≡ 1 ( m o d 8 ) q\equiv1\pmod8 q ≡ 1 ( mod 8 ) and q 0 ≡ 3 ( m o d 8 ) q_0\equiv3\pmod8 q 0 ≡ 3 ( mod 8 ) is a prime, all distinct and prime to β η \beta\eta β η , with prescribed residue symbols. Fix C ∈ M B × I ( F 2 ) C\in M_{B\times I}(\mathbb{F}_2) C ∈ M B × I ( F 2 ) whose columns form a basis of the hyperplane 1 B ⊥ ⊂ F 2 B 1_B^{\perp}\subset\mathbb{F}_2^B 1 B ⊥ ⊂ F 2 B , fix ϵ j ∈ F 2 \epsilon_j\in\mathbb{F}_2 ϵ j ∈ F 2 ( j ∈ J ) (j\in J) ( j ∈ J ) , and put γ = C 1 + y B \gamma=C1+y_B γ = C 1 + y B . We require
( q b ) = ( − 1 ) C b q , ( q j ) = ( q ′ j ) = ( q 0 j ) = 1 , ( q 0 b ) = ( − 1 ) γ b , ( q 0 j ) = ( − 1 ) ϵ j (20) \left(\frac{q}{b}\right)=(-1)^{C_{bq}},\qquad\left(\frac{q}{j}\right)=\left(\frac{q'}{j}\right)=\left(\frac{q_0}{j}\right)=1,\qquad\left(\frac{q_0}{b}\right)=(-1)^{\gamma_b},\qquad\left(\frac{q_0}{j}\right)=(-1)^{\epsilon_j} \tag*{(20)} ( b q ) = ( − 1 ) C b q , ( j q ) = ( j q ′ ) = ( j q 0 ) = 1 , ( b q 0 ) = ( − 1 ) γ b , ( j q 0 ) = ( − 1 ) ϵ j (20)
for b ∈ B b\in B b ∈ B , distinct q , q ′ ∈ I q,q'\in I q , q ′ ∈ I and j ∈ J j\in J j ∈ J . Then n ≡ 7 ⋅ 3 ≡ 5 ( m o d 8 ) n\equiv7\cdot3\equiv5\pmod8 n ≡ 7 ⋅ 3 ≡ 5 ( mod 8 ) and n η ≡ 5 ( m o d 8 ) n\eta\equiv5\pmod8 n η ≡ 5 ( mod 8 ) . We have 1 t C = 0 1^{t}C=0 1 t C = 0 and 1 t γ = 1 1^{t}\gamma=1 1 t γ = 1 , and as z B ∈ 1 B ⊥ z_B\in1_B^{\perp} z B ∈ 1 B ⊥ there is a unique v 0 ∈ F 2 I v^0\in\mathbb{F}_2^I v 0 ∈ F 2 I with C v 0 = z B Cv^0=z_B C v 0 = z B . Let κ = ( u , v ) \kappa=(u,v) κ = ( u , v ) with u = 1 B u=1_B u = 1 B and v = 1 B + v 0 v=1_B+v^0 v = 1 B + v 0 , where vectors on B B B and on I I I are extended by zero. Since the primes of I I I are 1 1 1 modulo 8 8 8 , (19) implies z t v = z B t 1 B = 0 z^{t}v=z_B^{t}1_B=0 z t v = z B t 1 B = 0 for both quadratic twists.
Let R ∈ M B × J ( F 2 ) R \in M_{B \times J}(\mathbb{F}_2) R ∈ M B × J ( F 2 ) be defined by ( i j ) = ( − 1 ) R b j \binom{i}{j}=(-1)^{R_{bj}} ( j i ) = ( − 1 ) R bj . By (19) and reciprocity, R 1 = 0 R1=0 R 1 = 0 and 1 t R = y J t 1^{t}R=y_{J}^{t} 1 t R = y J t . In the order B , I , q 0 , J B,I,q_0,J B , I , q 0 , J , quadratic reciprocity and (20) give the matrix for n η n\eta n η as
A = ( A B + D y B C γ R C t 0 0 0 ( C 1 ) t 0 1 t ϵ ( ϵ + y J ) t R t + y J y B t 0 ϵ A J + D ϵ ) (21) A=\begin{pmatrix}
A_B+D y_B & C & \gamma& R\\
C^t & 0 & 0 & 0\\
(C1)^t & 0 & 1^t\epsilon& (\epsilon+y_J)^t\\
R^t+y_Jy_B^t & 0 & \epsilon& A_J+D_\epsilon
\end{pmatrix} \tag*{(21)} A = A B + D y B C t ( C 1 ) t R t + y J y B t C 0 0 0 γ 0 1 t ϵ ϵ R 0 ( ϵ + y J ) t A J + D ϵ (21)
The components of y , z y,z y , z at I I I vanish, and y q 0 = z q 0 = 1 y_{q_0}=z_{q_0}=1 y q 0 = z q 0 = 1 . Finally put
G = A J + D ϵ , Z = z J z J t + D z J , P ( ϵ ) = ( G + G t G G t Z ) . G=A_J+D_\epsilon,\qquad Z=z_Jz_J^t+D_{z_J},\qquad P(\epsilon)=\begin{pmatrix}G+G^t&G\\G^t&Z\end{pmatrix}. G = A J + D ϵ , Z = z J z J t + D z J , P ( ϵ ) = ( G + G t G t G Z ) .
Lemma 3.12. Assume (20) . Then N ( n ) N(n) N ( n ) is spanned by κ \kappa κ . If P ( ϵ ) P(\epsilon) P ( ϵ ) is invertible, then N ( n η ) N(n\eta) N ( n η ) is spanned by κ \kappa κ .
Proof. Let ( u , v ) ∈ N ( n η ) (u,v)\in N(n\eta) ( u , v ) ∈ N ( n η ) , and write u B , u I , u 0 , u J u_B,u_I,u_0,u_J u B , u I , u 0 , u J for its components at B B B , I I I , q 0 q_0 q 0 and J J J , and similarly for v v v ; for N ( n ) N(n) N ( n ) the components at J J J are absent, and the argument applies with u J = v J = 0 u_J=v_J=0 u J = v J = 0 . By (21) , the rows of M 0 M_0 M 0 indexed by q ∈ I q\in I q ∈ I read ( C t u B ) q = ( C t v B ) q = 0 (C^tu_B)_q=(C^tv_B)_q=0 ( C t u B ) q = ( C t v B ) q = 0 . Hence u B = λ 1 B u_B=\lambda1_B u B = λ 1 B and v B = λ ′ 1 B v_B=\lambda'1_B v B = λ ′ 1 B with λ , λ ′ ∈ F 2 \lambda,\lambda'\in\mathbb{F}_2 λ , λ ′ ∈ F 2 . The rows indexed by b ∈ B b\in B b ∈ B read
C u I = ( λ + λ ′ ) y B + u 0 γ + R u J , C v I = v 0 γ + R v J + λ z B . Cu_I=(\lambda+\lambda')y_B+u_0\gamma+Ru_J,\qquad Cv_I=v_0\gamma+Rv_J+\lambda z_B. C u I = ( λ + λ ′ ) y B + u 0 γ + R u J , C v I = v 0 γ + R v J + λ z B .
Pairing with 1 B 1_B 1 B shows that they are solvable exactly when
λ ′ = λ + u 0 + y J t u J , v 0 = y J t v J (22) \lambda'=\lambda+u_0+y_J^tu_J,\qquad v_0=y_J^tv_J \tag*{(22)} λ ′ = λ + u 0 + y J t u J , v 0 = y J t v J (22)
and then they determine u I u_I u I and v I v_I v I , the columns of C C C being independent. Summing the entries of ( A + D y ) v + D z u (A+D_y)v+D_zu ( A + D y ) v + D z u and using Lemma 3.5 gives z t u = 0 z^tu=0 z t u = 0 , that is, u 0 = z J t u J u_0=z_J^tu_J u 0 = z J t u J .
For N ( n ) N(n) N ( n ) this gives u 0 = v 0 = 0 u_0=v_0=0 u 0 = v 0 = 0 and λ ′ = λ \lambda'=\lambda λ ′ = λ , so C u I = 0 Cu_I=0 C u I = 0 and C v I = λ z B Cv_I=\lambda z_B C v I = λ z B , that is, u I = 0 u_I=0 u I = 0 and v I = λ v 0 v_I=\lambda v^0 v I = λ v 0 . Conversely κ ∈ N ( n ) \kappa\in N(n) κ ∈ N ( n ) : the rows indexed by I I I and B B B hold by the same computation, and the rows indexed by q 0 q_0 q 0 reduce to 1 t C 1 = 0 1^tC1=0 1 t C 1 = 0 . So N ( n ) N(n) N ( n ) is spanned by κ \kappa κ .
For N ( n η ) N(n\eta) N ( n η ) , the rows indexed by J J J read, as ∏ b ∈ B ( b j ) = 1 \prod_{b\in B}\binom{b}{j}=1 ∏ b ∈ B ( j b ) = 1 ,
G u J + D y J v J + u 0 ϵ = 0 , ( G + D y J ) v J + D z J u J + v 0 ϵ = 0. Gu_J+D_{y_J}v_J+u_0\epsilon=0,\qquad(G+D_{y_J})v_J+D_{z_J}u_J+v_0\epsilon=0. G u J + D y J v J + u 0 ϵ = 0 , ( G + D y J ) v J + D z J u J + v 0 ϵ = 0.
Put u ′ = u J + u 0 1 u'=u_J+u_0 1 u ′ = u J + u 0 1 and v ′ = v J + v 0 1 v'=v_J+v_0 1 v ′ = v J + v 0 1 . As A J 1 = 0 A_J1=0 A J 1 = 0 and y J t 1 = z J t 1 = 0 y_J^t1=z_J^t1=0 y J t 1 = z J t 1 = 0 , we have u 0 = z J t u ′ u_0=z_J^tu' u 0 = z J t u ′ and v 0 = y J t v ′ v_0=y_J^tv' v 0 = y J t v ′ , and the equations become
G u ′ + ( D y J + y J y J t ) v ′ = 0 , Z u ′ + ( G + D y J + y J y J t ) v ′ = 0. Gu'+(D_{y_J}+y_Jy_J^t)v'=0,\qquad Zu'+(G+D_{y_J}+y_Jy_J^t)v'=0. G u ′ + ( D y J + y J y J t ) v ′ = 0 , Z u ′ + ( G + D y J + y J y J t ) v ′ = 0.
By Lemma 3.5 for J J J , G + G t = D y J + y J y J t G+G^t=D_{y_J}+y_Jy_J^t G + G t = D y J + y J y J t , so this system is P ( ϵ ) ( v ′ u ′ ) = 0 P(\epsilon)\binom{v'}{u'}=0 P ( ϵ ) ( u ′ v ′ ) = 0 . If P ( ϵ ) P(\epsilon) P ( ϵ ) is invertible, then u ′ = v ′ = 0 u'=v'=0 u ′ = v ′ = 0 , hence u 0 = v 0 = 0 u_0=v_0=0 u 0 = v 0 = 0 , u J = v J = 0 u_J=v_J=0 u J = v J = 0 and λ ′ = λ \lambda'=\lambda λ ′ = λ by (22) , and we conclude as for N ( n ) N(n) N ( n ) . Finally κ ∈ N ( n η ) \kappa\in N(n\eta) κ ∈ N ( n η ) : compared with N ( n ) N(n) N ( n ) , the rows indexed by B B B acquire the term R 1 = 0 R1=0 R 1 = 0 , and the rows indexed by J J J reduce to the relations ∏ b ∈ B ( b j ) = 1 \prod_{b\in B}\binom{b}{j}=1 ∏ b ∈ B ( j b ) = 1 . □ \square □
Proof of Proposition 3.11. The matrix P ( ϵ ) P(\epsilon) P ( ϵ ) is symmetric with zero diagonal, hence alternating, and det P ( ϵ ) \det P(\epsilon) det P ( ϵ ) is the square of its Pfaffian. The variable ϵ j \epsilon_j ϵ j occurs only in the two entries of P ( ϵ ) P(\epsilon) P ( ϵ ) in the rows and columns indexed by j j j in the two copies of J J J , each equal to ( A J ) j j + ϵ j (A_J)_{jj}+\epsilon_j ( A J ) j j + ϵ j . So the Pfaffian is a polynomial in the ϵ j \epsilon_j ϵ j of degree at most one in each variable, and its coefficient of ∏ j ∈ J ϵ j \prod_{j\in J}\epsilon_j ∏ j ∈ J ϵ j is 1 1 1 , coming from the perfect matching that pairs the two copies of each j ∈ J j\in J j ∈ J . Hence it does not vanish identically on F 2 J \mathbb{F}_2^J F 2 J , and we fix ϵ \epsilon ϵ with det P ( ϵ ) = 1 \det P(\epsilon)=1 det P ( ϵ ) = 1 ; if J = ∅ J=\varnothing J = ∅ there is nothing to choose. The conditions (20) on the primes of I I I and on q 0 q_0 q 0 are congruences modulo 8 8 8 and modulo primes already fixed. Choosing the primes of I I I one at a time and then q 0 q_0 q 0 , the Chinese remainder theorem and Dirichlet’s theorem provide them, and Lemma 3.12 completes the proof. □ \square □
Theorem B follows by applying Corollary 3.10 : the constructed vector satisfies y t u = y t 1 B = 1 y^{t}u=y^{t}1_{B}=1 y t u = y t 1 B = 1 and z t v = 0 z^{t}v=0 z t v = 0 . The resulting finite Tate–Shafarevich groups have trivial 2-torsion, hence odd order.
Remark. The construction extends to several quadratic twists. For r ≥ 1 r\ge1 r ≥ 1 , let β , η 1 , … , η r \beta,\eta_{1},\ldots,\eta_{r} β , η 1 , … , η r be positive squarefree integers. Suppose that the η i \eta_{i} η i are pairwise coprime and that each pair ( β , η i ) (\beta,\eta_{i}) ( β , η i ) satisfies the hypotheses of Theorem B. There exist infinitely many positive squarefree integers n ≡ 5 ( m o d 8 ) n\equiv5\pmod{8} n ≡ 5 ( mod 8 ) , divisible by β \beta β and prime to ∏ i η i \prod_{i}\eta_{i} ∏ i η i , such that every E l E_{l} E l , for l ∈ { n , n η 1 , … , n η r } l\in\{n,n\eta_{1},\ldots,n\eta_{r}\} l ∈ { n , n η 1 , … , n η r } , has analytic and algebraic rank one, finite Tate–Shafarevich group of odd order, and a rational point with x x x -coordinate in β Q × 2 \beta\mathbb{Q}^{\times2} β Q × 2 .
Indeed, put J i = Supp ( η i ) J_{i}=\operatorname{Supp}(\eta_{i}) J i = Supp ( η i ) . Choose a common matrix C C C and set of auxiliary primes I I I with ( q j ) = 1 \left(\frac{q}{j}\right)=1 ( j q ) = 1 for every q ∈ I q\in I q ∈ I and j ∈ ⋃ i J i j\in\bigcup_{i}J_{i} j ∈ ⋃ i J i . Choose the signs ϵ j \epsilon_{j} ϵ j separately on each J i J_{i} J i by the Pfaffian argument above. Since the J i J_{i} J i are disjoint, the Chinese remainder theorem and Dirichlet’s theorem provide infinitely many primes q 0 q_{0} q 0 satisfying all the resulting conditions (20) . Lemma 3.12 and Corollary 3.10 then apply to every quadratic twist. Overlapping sets J i J_{i} J i , or simultaneous control of all products of the twisting factors, would require compatibility between these choices.
Example 3.13. Take β = 7 \beta=7 β = 7 and η = 113 \eta=113 η = 113 . Then I = ∅ I=\varnothing I = ∅ , J = { 113 } J=\{113\} J = { 113 } , y J = z J = 0 y_{J}=z_{J}=0 y J = z J = 0 and A J = 0 A_{J}=0 A J = 0 , so P ( ϵ ) = ( 0 ϵ ϵ 0 ) P(\epsilon)=\left(\begin{smallmatrix}0&\epsilon\\ \epsilon&0\end{smallmatrix}\right) P ( ϵ ) = ( 0 ϵ ϵ 0 ) and ϵ = 1 \epsilon=1 ϵ = 1 . The prime q 0 = 3 q_{0}=3 q 0 = 3 satisfies ( 3 7 ) = − 1 = ( − 1 ) γ 7 \left(\frac{3}{7}\right)=-1=(-1)^{\gamma_{7}} ( 7 3 ) = − 1 = ( − 1 ) γ 7 and ( 3 113 ) = − 1 \left(\frac{3}{113}\right)=-1 ( 113 3 ) = − 1 , so n = 21 n=21 n = 21 . In the coordinates indexed by 7 , 3 7,3 7 , 3 one has y = ( 1 , 1 ) y=(1,1) y = ( 1 , 1 ) , z = ( 0 , 1 ) z=(0,1) z = ( 0 , 1 ) and A = ( 1 1 0 0 ) A=\left(\begin{smallmatrix}1&1\\0&0\end{smallmatrix}\right) A = ( 1 0 1 0 ) , and N ( 21 ) N(21) N ( 21 ) is spanned by κ = ( ( 1 , 0 ) , ( 1 , 0 ) ) \kappa=((1,0),(1,0)) κ = (( 1 , 0 ) , ( 1 , 0 )) . By Lemma 3.12 , N ( 21 ⋅ 113 ) N(21\cdot113) N ( 21 ⋅ 113 ) is spanned by the same vector extended by zero at 113 113 113 . Thus Sel 2 ( E 21 ) \operatorname{Sel}_{2}(E_{21}) Sel 2 ( E 21 ) is spanned by T 21 T_{21} T 21 and ( 7 , 7 ) = δ 2 ( 28 , 98 ) (7,7)=\delta_{2}(28,98) ( 7 , 7 ) = δ 2 ( 28 , 98 ) , as 28 = 7 ⋅ 2 2 28=7\cdot2^{2} 28 = 7 ⋅ 2 2 , 28 − 21 = 7 28-21=7 28 − 21 = 7 and 28 + 21 = 7 2 28+21=7^{2} 28 + 21 = 7 2 .
Completeness
Proposition 3.14. Every b ∈ D 2 b\in\mathcal{D}_{2} b ∈ D 2 satisfies Φ ( b ) \Phi(b) Φ ( b ) .
Proof. Let b ∈ D 2 ∖ { 0 } b\in\mathcal{D}_{2}\setminus\{0\} b ∈ D 2 ∖ { 0 } , and choose μ ∈ M \mu\in\mathcal{M} μ ∈ M and δ ∈ D \delta\in\mathcal{D} δ ∈ D for b b b as in §3.1. Since δ b > 0 \delta b>0 δ b > 0 and δ b ∈ Q q × 2 \delta b\in\mathbb{Q}_{q}^{\times2} δ b ∈ Q q × 2 for q ∈ S ( μ ) q\in S(\mu) q ∈ S ( μ ) , the equation δ b T 2 = 1 \delta bT^{2}=1 δ b T 2 = 1 has a root in R \mathbb{R} R and in each Q q \mathbb{Q}_{q} Q q , q ∈ S ( μ ) q\in S(\mu) q ∈ S ( μ ) . As t t t approaches any of these roots, U → 0 U\to0 U → 0 and H ′ → 1 H'\to1 H ′ → 1 . By weak approximation we choose t ∈ Q t\in\mathbb{Q} t ∈ Q with H ′ > 0 H'>0 H ′ > 0 , H ′ ∈ Q q × 2 H'\in\mathbb{Q}_{q}^{\times2} H ′ ∈ Q q × 2 for q ∈ S ( μ ) q\in S(\mu) q ∈ S ( μ ) , and v q ( δ b t 2 ) > 0 v_{q}(\delta bt^{2})>0 v q ( δ b t 2 ) > 0 for q ∈ Supp ( δ ) ∖ S ( μ ) q\in\operatorname{Supp}(\delta)\setminus S(\mu) q ∈ Supp ( δ ) ∖ S ( μ ) . Then 1 + δ b t 2 ≠ 0 1+\delta bt^{2}\ne0 1 + δ b t 2 = 0 and H ′ ≠ 0 H'\ne0 H ′ = 0 .
Let β \beta β and η \eta η be the positive squarefree integers in the square classes of b ′ b' b ′ and H ′ H' H ′ , with sets of prime divisors B B B and J J J . We check (18) . By (13) , − β ∈ Q × 2 -\beta\in\mathbb{Q}^{\times2} − β ∈ Q × 2 , so β ≡ 7 ( m o d 8 ) \beta\equiv7\pmod{8} β ≡ 7 ( mod 8 ) ; and η ≡ 1 ( m o d 8 ) \eta\equiv1\pmod{8} η ≡ 1 ( mod 8 ) by the choice of t t t . Let q ∈ B q\in B q ∈ B . If q ∈ S ( μ ) q\in S(\mu) q ∈ S ( μ ) , then H ′ ∈ Q q × 2 H'\in\mathbb{Q}_{q}^{\times2} H ′ ∈ Q q × 2 by construction. Otherwise v q ( b ) = v q ( b ′ ) v_{q}(b)=v_{q}(b') v q ( b ) = v q ( b ′ ) is odd, hence positive since b ∈ D 2 b\in\mathcal{D}_{2} b ∈ D 2 , and U U U is a q q q -adic unit: if q ∈ Supp ( δ ) q\in\operatorname{Supp}(\delta) q ∈ Supp ( δ ) because v q ( δ b t 2 ) > 0 v_{q}(\delta bt^{2})>0 v q ( δ b t 2 ) > 0 by the choice of t t t , and otherwise v q ( δ b t 2 ) v_{q}(\delta bt^{2}) v q ( δ b t 2 ) is odd, as in the proof of Proposition 3.2. So v q ( b ′ U 2 ) > 0 v_{q}(b'U^{2})>0 v q ( b ′ U 2 ) > 0 and H ′ = 1 − b ′ U 2 ∈ Q q × 2 H'=1-b'U^{2}\in\mathbb{Q}_{q}^{\times2} H ′ = 1 − b ′ U 2 ∈ Q q × 2 . Thus η ∈ Q q × 2 \eta\in\mathbb{Q}_{q}^{\times2} η ∈ Q q × 2 for q ∈ B q\in B q ∈ B , and B ∩ J = ∅ B\cap J=\varnothing B ∩ J = ∅ . Let q ∈ J q\in J q ∈ J . Then q ∉ S ( μ ) ∪ B q\notin S(\mu)\cup B q ∈ / S ( μ ) ∪ B , so q q q is odd and β \beta β is a q q q -adic unit. If β \beta β were not a square at q q q , then H ′ H' H ′ , a non-zero norm from the unramified quadratic extension Q q ( β ) \mathbb{Q}_{q}(\sqrt{\beta}) Q q ( β ) , would have even valuation. Hence β ∈ Q q × 2 \beta\in\mathbb{Q}_{q}^{\times2} β ∈ Q q × 2 .
Proposition 3.11 provides n n n . For l = n l=n l = n and l = n η l=n\eta l = n η , N ( l ) N(l) N ( l ) is spanned by a vector ( u , v ) (u,v) ( u , v ) with u = 1 B u=1_{B} u = 1 B and z t v = 0 z^{t}v=0 z t v = 0 , while y t 1 B = 1 y^{t}1_{B}=1 y t 1 B = 1 by (19) . By Corollary 3.10 , each of E n E_{n} E n and E n η E_{n\eta} E n η has a rational point with x x x -coordinate in β Q × 2 = b ′ Q × 2 \beta\mathbb{Q}^{\times2}=b'\mathbb{Q}^{\times2} β Q × 2 = b ′ Q × 2 . Writing H ′ = η c 2 H'=\eta c^{2} H ′ = η c 2 , the isomorphism E n η → E n H ′ E_{n\eta}\to E_{nH'} E n η → E n H ′ , ( x , y ) ↦ ( c 2 x , c 3 y ) (x,y)\mapsto(c^{2}x,c^{3}y) ( x , y ) ↦ ( c 2 x , c 3 y ) , preserves the square class of x x x . Hence e = n e=n e = n satisfies the conditions in Φ ( b ) \Phi(b) Φ ( b ) .
Proof of Theorem C. By Propositions 3.2 and 3.14, D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 D_2 \subseteq\Phi(\mathbb{Q}) \subseteq D_2^{S_1} D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 . For each prime q q q the local ring Z ( q ) \mathbb{Z}_{(q)} Z ( q ) is Diophantine in Q \mathbb{Q} Q [73 ] (see also [42 ] ); hence so are Z ( q ) × \mathbb{Z}_{(q)}^{\times} Z ( q ) × and
X q = Z ( q ) ∪ { u w 2 : u ∈ Z ( q ) × , w ∈ Q × } . X_q=\mathbb{Z}_{(q)}\cup\{uw^2:u\in\mathbb{Z}_{(q)}^{\times},\ w\in\mathbb{Q}^{\times}\}. X q = Z ( q ) ∪ { u w 2 : u ∈ Z ( q ) × , w ∈ Q × } .
which is the set of b ∈ Q b\in\mathbb{Q} b ∈ Q such that v q ( b ) v_q(b) v q ( b ) is not a negative odd integer. Therefore D 2 = Φ ( Q ) ∩ ⋂ q ∈ S 1 X q D_2=\Phi(\mathbb{Q})\cap\bigcap_{q\in S_1}X_q D 2 = Φ ( Q ) ∩ ⋂ q ∈ S 1 X q is Diophantine.
The proof of Theorem A below uses only the inclusions D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 D_2\subseteq\Phi(\mathbb{Q})\subseteq D_2^{S_1} D 2 ⊆ Φ ( Q ) ⊆ D 2 S 1 .
Local conditions at primes of odd valuation
In this section a single instance of Φ \Phi Φ ensures the following condition: at every prime outside S 1 S_1 S 1 at which a given rational number has odd valuation, one of finitely many systems of local inequalities holds (see Lemma 4.1 ). The system and the place of F F F above the prime may depend on the prime.
Lemma 4.1. Let G , m ≥ 1 G,m\geq1 G , m ≥ 1 , R ∈ Q × R\in\mathbb{Q}^{\times} R ∈ Q × , k 1 , … , k m ≥ 1 k_1,\ldots,k_m\geq1 k 1 , … , k m ≥ 1 and D g , j ∈ F D_{g,j}\in F D g , j ∈ F for 1 ≤ g ≤ G 1\leq g\leq G 1 ≤ g ≤ G , 1 ≤ j ≤ m 1\leq j\leq m 1 ≤ j ≤ m . Put
W g = ∏ j = 1 m ( 1 + R ( D g , j R − k j ) 2 ) , Θ = Tr F / Q ( ∑ g = 1 G W g − 1 ) , B = R / Θ 2 . W_g=\prod_{j=1}^{m}\left(1+R(D_{g,j}R^{-k_j})^2\right),\qquad\Theta=\operatorname{Tr}_{F/\mathbb{Q}}\left(\sum_{g=1}^{G}W_g^{-1}\right),\qquad B=R/\Theta^2. W g = j = 1 ∏ m ( 1 + R ( D g , j R − k j ) 2 ) , Θ = Tr F / Q ( g = 1 ∑ G W g − 1 ) , B = R / Θ 2 . (1) Assume that every W g W_g W g and Θ \Theta Θ are non-zero and that Φ ( B ) \Phi(B) Φ ( B ) holds. Let p ∉ S 1 p\notin S_1 p ∈ / S 1 be a prime with v p ( R ) v_p(R) v p ( R ) odd. Then v p ( R ) > 0 v_p(R)>0 v p ( R ) > 0 , and there are g g g and a place p ∣ p \mathfrak{p}\mid p p ∣ p with v p ( D g , j ) ≥ k j v_{\mathfrak{p}}(D_{g,j})\geq k_j v p ( D g , j ) ≥ k j for all j j j .
(2) Let T T T be a non-empty finite set of primes, each split in F F F and larger than $2G$, such that v p ( R ) = 1 v_p(R)=1 v p ( R ) = 1 for p ∈ T p\in T p ∈ T and v p ( R ) v_p(R) v p ( R ) is even for p ∉ T p\notin T p ∈ / T . Assume that for each p ∈ T p\in T p ∈ T there are g g g and p ∣ p \mathfrak{p}\mid p p ∣ p with v p ( D g , j ) ≥ k j v_{\mathfrak{p}}(D_{g,j})\geq k_j v p ( D g , j ) ≥ k j for all $j. Then every W g W_g W g and Θ \Theta Θ are non-zero, and B ∈ D 2 B\in D_2 B ∈ D 2 .
Proof. Let p p p be an odd prime with α = v p ( R ) \alpha=v_p(R) α = v p ( R ) odd, and let p ∣ p \mathfrak{p}\mid p p ∣ p . For D g , j ≠ 0 D_{g,j}\neq0 D g , j = 0 the term R ( D g , j R − k j ) 2 R(D_{g,j}R^{-k_j})^2 R ( D g , j R − k j ) 2 has valuation
b g , j , p = α + 2 ( v p ( D g , j ) − k j α ) , b_{g,j,\mathfrak{p}}=\alpha+2(v_{\mathfrak{p}}(D_{g,j})-k_j\alpha), b g , j , p = α + 2 ( v p ( D g , j ) − k j α ) ,
which is odd, hence non-zero; so the corresponding factor of W g W_g W g is non-zero, of valuation min ( 0 , b g , j , p ) \min(0,b_{g,j,\mathfrak{p}}) min ( 0 , b g , j , p ) . A factor with D g , j = 0 D_{g,j}=0 D g , j = 0 equals 1 1 1 . Hence v p ( W g ) ≤ 0 v_p(W_g)\leq0 v p ( W g ) ≤ 0 .
(1) Every W g − 1 W_g^{-1} W g − 1 is integral above p p p , so v p ( Θ ) ≥ 0 v_p(\Theta)\geq0 v p ( Θ ) ≥ 0 and v p ( B ) = α − 2 v p ( Θ ) v_p(B)=\alpha-2v_p(\Theta) v p ( B ) = α − 2 v p ( Θ ) is odd and at most α \alpha α . By Proposition 3.2 it is not negative, so α > 0 \alpha>0 α > 0 . Suppose that for every g g g and every p ∣ p \mathfrak{p}\mid p p ∣ p some j j j has v p ( D g , j ) ≤ k j − 1 v_{\mathfrak{p}}(D_{g,j})\leq k_j-1 v p ( D g , j ) ≤ k j − 1 . Then D g , j ≠ 0 D_{g,j}\neq0 D g , j = 0 and
b g , j , p ≤ α + 2 ( k j − 1 − k j α ) = − α − 2 ( k j − 1 ) ( α − 1 ) ≤ − α , b_{g,j,\mathfrak{p}}\leq\alpha+2(k_j-1-k_j\alpha)=-\alpha-2(k_j-1)(\alpha-1)\leq-\alpha, b g , j , p ≤ α + 2 ( k j − 1 − k j α ) = − α − 2 ( k j − 1 ) ( α − 1 ) ≤ − α ,
so v p ( W g − 1 ) ≥ α v_{\mathfrak{p}}(W_g^{-1})\geq\alpha v p ( W g − 1 ) ≥ α for all g g g and p ∣ p \mathfrak{p}\mid p p ∣ p . Hence v p ( Θ ) ≥ α v_p(\Theta)\geq\alpha v p ( Θ ) ≥ α , and v p ( B ) ≤ − α v_p(B)\leq-\alpha v p ( B ) ≤ − α is negative and odd, contrary to Proposition 3.2.
(2) Let p ∈ T p\in T p ∈ T , so α = 1 \alpha=1 α = 1 . Every factor of every W g W_g W g is either congruent to 1 1 1 modulo p \mathfrak{p} p , when b g , j , p > 0 b_{g,j,\mathfrak{p}}>0 b g , j , p > 0 , or has negative valuation. Hence W g ≠ 0 W_g\neq0 W g = 0 , and W g − 1 ∈ O F , p W_g^{-1}\in\mathcal{O}_{F,\mathfrak{p}} W g − 1 ∈ O F , p is congruent to 1 1 1 or 0 0 0 modulo p \mathfrak{p} p , and to 1 1 1 for the pair ( g , p ) (g,\mathfrak{p}) ( g , p ) of the hypothesis. As p p p splits, Θ \Theta Θ is congruent modulo p p p to the number of pairs ( g , p ′ ) (g,\mathfrak{p}') ( g , p ′ ) , p ′ ∣ p \mathfrak{p}'\mid p p ′ ∣ p , with W g − 1 ≡ 1 ( m o d p ′ ) W_g^{-1}\equiv1\pmod{\mathfrak{p}'} W g − 1 ≡ 1 ( mod p ′ ) , an integer between 1 1 1 and 2 G < p 2G<p 2 G < p . So Θ ≠ 0 \Theta\neq0 Θ = 0 , v p ( Θ ) = 0 v_p(\Theta)=0 v p ( Θ ) = 0 and v p ( B ) = 1 v_p(B)=1 v p ( B ) = 1 . For p ∉ T p\notin T p ∈ / T the valuation v p ( B ) = v p ( R ) − 2 v p ( Θ ) v_p(B)=v_p(R)-2v_p(\Theta) v p ( B ) = v p ( R ) − 2 v p ( Θ ) is even. Therefore B ∈ D 2 B\in D_2 B ∈ D 2 .
Remark. Lemma 4.1 is [86 Lemma 3.1] , applied to Φ \Phi Φ .
Linear patterns of good primes
Following [55 §6] , we show that the linear forms below take pairwise distinct values among the good primes of Definition 2.5.
Proposition 5.1. Let Λ 0 ⊂ Z ∖ { 0 } \Lambda_0 \subset\mathbb{Z} \setminus\{0\} Λ 0 ⊂ Z ∖ { 0 } be finite and non-empty, put t = ∣ Λ 0 ∣ + 2 t = |\Lambda_0| + 2 t = ∣ Λ 0 ∣ + 2 and W 0 = 56 ∏ 3 ≤ ℓ ≤ 2 t ℓ W_0 = 56\prod_{3 \leq\ell\leq2t} \ell W 0 = 56 ∏ 3 ≤ ℓ ≤ 2 t ℓ , the product over primes, and assume W 0 ∣ λ W_0 \mid\lambda W 0 ∣ λ for every λ ∈ Λ 0 \lambda\in\Lambda_0 λ ∈ Λ 0 . For every Y Y Y there exist integers u , v ′ u,v' u , v ′ such that the t t t numbers
u , v ′ , u − λ v ′ ( λ ∈ Λ 0 ) u,\qquad v',\qquad u-\lambda v'\quad(\lambda\in\Lambda_0) u , v ′ , u − λ v ′ ( λ ∈ Λ 0 ) are pairwise distinct good primes larger than Y Y Y .
Theorem 5.2 (Green–Tao–Ziegler). Let ψ 1 , … , ψ t : Z d → Z \psi_1,\ldots,\psi_t:\mathbb{Z}^d \to\mathbb{Z} ψ 1 , … , ψ t : Z d → Z be affine-linear maps with integer coefficients whose linear parts are non-zero and pairwise linearly independent. For N ≥ 1 N \geq1 N ≥ 1 let K N K_N K N be a convex body in [ − N , N ] d [-N,N]^d [ − N , N ] d on which every ψ i \psi_i ψ i is positive. Then, as N → ∞ N \to\infty N → ∞ , uniformly in the choice of K N K_N K N ,
∑ m ∈ Z d ∩ K N ∏ i = 1 t Λ ( ψ i ( m ) ) = vol ( K N ) ∏ p β p + o ( N d ) , \sum_{\mathbf{m}\in\mathbb{Z}^d\cap K_N}\prod_{i=1}^{t}\Lambda(\psi_i(\mathbf{m}))=\operatorname{vol}(K_N)\prod_p\beta_p+o(N^d), m ∈ Z d ∩ K N ∑ i = 1 ∏ t Λ ( ψ i ( m )) = vol ( K N ) p ∏ β p + o ( N d ) , where
β p = ( 1 − 1 p ) − t # { m ∈ ( Z / p Z ) d : p ∤ ψ i ( m ) ∀ i } p d , \beta_p=\left(1-\frac{1}{p}\right)^{-t}\frac{\#\{\mathbf{m}\in(\mathbb{Z}/p\mathbb{Z})^d:p\nmid\psi_i(\mathbf{m})\ \forall i\}}{p^d}, β p = ( 1 − p 1 ) − t p d # { m ∈ ( Z / p Z ) d : p ∤ ψ i ( m ) ∀ i } , and Λ \Lambda Λ is the von Mangoldt function, extended by Λ ( m ) = 0 \Lambda(m)=0 Λ ( m ) = 0 for m ≤ 0 m\leq0 m ≤ 0 , and the product converges absolutely.
This is [28 Main Theorem] , which relied on two conjectures, proved in [29 Theorem 1.1] and [30 Theorem 1.3] (see [31 ] for corrections to the latter).
Throughout, ψ 1 , … , ψ t \psi_1,\ldots,\psi_t ψ 1 , … , ψ t are the forms of Proposition 5.1 in the variables T = ( u , v ′ ) \mathbf{T}=(u,v') T = ( u , v ′ ) ; they are primitive and pairwise linearly independent. Fix η 0 = 1 / ( 4 + 4 max λ ∈ Λ 0 ∣ λ ∣ ) \eta_0=1/(4+4\max_{\lambda\in\Lambda_0}|\lambda|) η 0 = 1/ ( 4 + 4 max λ ∈ Λ 0 ∣ λ ∣ ) and the box K = [ 1 , 2 ] × [ η 0 , 2 η 0 ] K=[1,2]\times[\eta_0,2\eta_0] K = [ 1 , 2 ] × [ η 0 , 2 η 0 ] , on which every ψ i \psi_i ψ i takes values in [ η 0 , 3 ] [\eta_0,3] [ η 0 , 3 ] . Implied constants depend only on Λ 0 \Lambda_0 Λ 0 : the forms are fixed first, then a parameter L L L , and then X X X is taken large in terms of L L L and Y Y Y .
The lower bound
Definition 5.3. Let L > max ( 7 , 2 t ) L>\max(7,2t) L > max ( 7 , 2 t ) and W = 8 ∏ 3 ≤ ℓ ≤ L ℓ W=8\prod_{3\leq\ell\leq L}\ell W = 8 ∏ 3 ≤ ℓ ≤ L ℓ . A class T 0 \mathbf{T}_0 T 0 mod W W W is admissible if u ≡ v ′ ≡ 1 ( m o d 8 ) u\equiv v'\equiv1\pmod8 u ≡ v ′ ≡ 1 ( mod 8 ) , u ≡ v ′ ≡ 3 ( m o d 7 ) u\equiv v'\equiv3\pmod7 u ≡ v ′ ≡ 3 ( mod 7 ) , u ≡ v ′ ≡ 1 ( m o d ℓ ) u\equiv v'\equiv1\pmod\ell u ≡ v ′ ≡ 1 ( mod ℓ ) for every prime 3 ≤ ℓ ≤ 2 t 3\leq\ell\leq2t 3 ≤ ℓ ≤ 2 t with ℓ ≠ 7 \ell\neq7 ℓ = 7 , and ψ i ( T 0 ) ≢ 0 , − 1 ( m o d ℓ ) \psi_i(\mathbf{T}_0)\not\equiv0,-1\pmod\ell ψ i ( T 0 ) ≡ 0 , − 1 ( mod ℓ ) for every i i i and every prime 2 t < ℓ ≤ L 2t<\ell\leq L 2 t < ℓ ≤ L with ℓ ≠ 7 \ell\neq7 ℓ = 7 .
Since W 0 ∣ λ W_0\mid\lambda W 0 ∣ λ , every value ψ i ( T ) \psi_i(\mathbf{T}) ψ i ( T ) at an admissible T \mathbf{T} T is 1 1 1 modulo 8 8 8 , 3 3 3 modulo 7 7 7 , 1 1 1 modulo every prime ℓ ≤ 2 t \ell\leq2t ℓ ≤ 2 t with ℓ ≠ 7 \ell\neq7 ℓ = 7 , and ≢ 0 , − 1 \not\equiv0,-1 ≡ 0 , − 1 modulo every prime 2 t < ℓ ≤ L 2t<\ell\leq L 2 t < ℓ ≤ L with ℓ ≠ 7 \ell\neq7 ℓ = 7 .
Lemma 5.4. Let N ( X ) N(X) N ( X ) be the number of admissible T ∈ Z 2 ∩ X K \mathbf{T}\in\mathbb{Z}^2\cap XK T ∈ Z 2 ∩ X K at which all ψ i ( T ) \psi_i(\mathbf{T}) ψ i ( T ) are prime. There is c 3 > 0 c_3>0 c 3 > 0 , depending only on Λ 0 \Lambda_0 Λ 0 , such that
N ( X ) ≥ ( c 3 ( log L ) − 2 t − o L ( 1 ) ) X 2 ( log 3 X ) t . (23) N(X)\geq\left(c_3(\log L)^{-2t}-o_L(1)\right)\frac{X^2}{(\log3X)^t}. \tag*{(23)} N ( X ) ≥ ( c 3 ( log L ) − 2 t − o L ( 1 ) ) ( log 3 X ) t X 2 . (23)
Proof. At a prime ℓ > 2 t \ell>2t ℓ > 2 t each form excludes at most two lines modulo ℓ \ell ℓ , so by Mertens’ theorem the set R W R_W R W of admissible classes satisfies
∣ R W ∣ W 2 ≥ c 0 ∏ 2 t < ℓ ≤ L ( 1 − 2 t ℓ ) ≥ c 1 ( log L ) − 2 t \frac{|R_W|}{W^2}\geq c_0\prod_{2t<\ell\leq L}\left(1-\frac{2t}{\ell}\right)\geq c_1(\log L)^{-2t} W 2 ∣ R W ∣ ≥ c 0 2 t < ℓ ≤ L ∏ ( 1 − ℓ 2 t ) ≥ c 1 ( log L ) − 2 t
with c 0 , c 1 > 0 c_0,c_1>0 c 0 , c 1 > 0 depending only on t t t . For T 0 ∈ R W \mathbf{T}_0\in R_W T 0 ∈ R W apply Theorem 5.2 to the forms m ↦ ψ i ( T 0 + W m ) \mathbf{m}\mapsto\psi_i(\mathbf{T}_0+W\mathbf{m}) m ↦ ψ i ( T 0 + W m ) and the bodies K N = ( X K − T 0 ) / W \mathcal{K}_N=(X\mathcal{K}-\mathbf{T}_0)/W K N = ( X K − T 0 ) / W , N = 3 X / W N=3X/W N = 3 X / W , of volume vol ( K ) X 2 / W 2 \operatorname{vol}(\mathcal{K})X^2/W^2 vol ( K ) X 2 / W 2 . At p ∣ W p\mid W p ∣ W every form is congruent to the non-zero constant ψ i ( T 0 ) \psi_i(\mathbf{T}_0) ψ i ( T 0 ) , so β p ≥ 1 \beta_p\ge1 β p ≥ 1 . At p ∤ W p\nmid W p ∤ W at least p 2 − t p p^2-tp p 2 − tp classes avoid all forms, so β p ≥ ( 1 − t / p ) ( 1 + t / p ) \beta_p\ge(1-t/p)(1+t/p) β p ≥ ( 1 − t / p ) ( 1 + t / p ) ; hence ∏ p β p ≥ c 2 > 0 \prod_p\beta_p\ge c_2>0 ∏ p β p ≥ c 2 > 0 , uniformly in L L L and T 0 \mathbf{T}_0 T 0 . Summing over R W R_W R W ,
∑ T ∈ Z 2 ∩ X K T m o d W ∈ R W ∏ i = 1 t Λ ( ψ i ( T ) ) ≥ c 3 ( log L ) − 2 t X 2 − o L ( X 2 ) . \sum_{\substack{\mathbf{T}\in\mathbb{Z}^2\cap X\mathcal{K}\\ \mathbf{T}\bmod W\in R_W}}\prod_{i=1}^{t}\Lambda(\psi_i(\mathbf{T}))\ge c_3(\log L)^{-2t}X^2-o_L(X^2). T ∈ Z 2 ∩ X K T mod W ∈ R W ∑ i = 1 ∏ t Λ ( ψ i ( T )) ≥ c 3 ( log L ) − 2 t X 2 − o L ( X 2 ) .
The values of each ψ i \psi_i ψ i on X K X\mathcal{K} X K lie in [ η 0 X , 3 X ] [\eta_0X,3X] [ η 0 X , 3 X ] . Proper prime powers in this range number O ( X 1 / 2 log X ) O(X^{1/2}\log X) O ( X 1/2 log X ) , each value of a form is taken at O ( X ) O(X) O ( X ) points of the box, and the weights are at most ( log 3 X ) t (\log3X)^t ( log 3 X ) t ; so these terms contribute o ( X 2 ) o(X^2) o ( X 2 ) , and (5.1) follows. □
Upper sieves
We adapt the sieve argument of OpenAI [55 §§6.4–6.6] to the forms in Proposition 5.1 .
Lemma 5.5. Let d ∈ { 1 , 2 } d\in\{1,2\} d ∈ { 1 , 2 } , n ≥ 1 n\ge1 n ≥ 1 , z ≥ 2 z\ge2 z ≥ 2 , and let B ⊂ Z d \mathcal{B}\subset\mathbb{Z}^d B ⊂ Z d be a product of d d d intervals each containing at most n n n integers. Let P \mathcal{P} P be a set of primes at most z z z , and for p ∈ P p\in\mathcal{P} p ∈ P let Ω p ⊂ ( Z / p Z ) d \Omega_p\subset(\mathbb{Z}/p\mathbb{Z})^d Ω p ⊂ ( Z / p Z ) d with 0 < g ( p ) = ∣ Ω p ∣ / p d ≤ 1 / 2 0<g(p)=|\Omega_p|/p^d\le1/2 0 < g ( p ) = ∣ Ω p ∣/ p d ≤ 1/2 . For squarefree e e e composed of primes of P \mathcal{P} P put g 1 ( e ) = ∏ p ∣ e g ( p ) / ( 1 − g ( p ) ) g_1(e)=\prod_{p\mid e}g(p)/(1-g(p)) g 1 ( e ) = ∏ p ∣ e g ( p ) / ( 1 − g ( p )) , and put G ( z ) = ∑ e ≤ z g 1 ( e ) G(z)=\sum_{e\le z}g_1(e) G ( z ) = ∑ e ≤ z g 1 ( e ) . Then
# { x ∈ B : x m o d p ∉ Ω p for all p ∈ P } ≤ n d G ( z ) + 3 ( n d − 1 z 4 + z 6 ) . \#\{\mathbf{x}\in\mathcal{B}:\mathbf{x}\bmod p\notin\Omega_p\text{ for all }p\in\mathcal{P}\}\le\frac{n^d}{G(z)}+3(n^{d-1}z^4+z^6). # { x ∈ B : x mod p ∈ / Ω p for all p ∈ P } ≤ G ( z ) n d + 3 ( n d − 1 z 4 + z 6 ) . Proof. This is Selberg’s upper-bound sieve with the weights λ e \lambda_e λ e of [36 Ch. 3] : λ e = 0 \lambda_e=0 λ e = 0 unless e ≤ z e\le z e ≤ z is squarefree and composed of primes of P \mathcal{P} P , in which case
λ e = μ ( e ) G ( z ) ∏ p ∣ e ( 1 − g ( p ) ) − 1 ∑ m ≤ z / e ( m , e ) = 1 g 1 ( m ) . \lambda_e=\frac{\mu(e)}{G(z)}\prod_{p\mid e}(1-g(p))^{-1}\sum_{\substack{m\le z/e\\(m,e)=1}}g_1(m). λ e = G ( z ) μ ( e ) p ∣ e ∏ ( 1 − g ( p ) ) − 1 m ≤ z / e ( m , e ) = 1 ∑ g 1 ( m ) . Then λ 1 = 1 \lambda_1=1 λ 1 = 1 , ∣ λ e ∣ ≤ 1 |\lambda_e|\le1 ∣ λ e ∣ ≤ 1 , and ∑ e , f λ e λ f g ( [ e , f ] ) = G ( z ) − 1 \sum_{e,f}\lambda_e\lambda_f g([e,f])=G(z)^{-1} ∑ e , f λ e λ f g ([ e , f ]) = G ( z ) − 1 , with g g g extended multiplicatively. The count is at most ∑ e , f λ e λ f A [ e , f ] \sum_{e,f}\lambda_e\lambda_f A_{[e,f]} ∑ e , f λ e λ f A [ e , f ] , where A q A_q A q counts the x ∈ B \mathbf{x}\in\mathcal{B} x ∈ B in one of the q d g ( q ) q^d g(q) q d g ( q ) classes modulo q q q determined by the Ω p \Omega_p Ω p , p ∣ q p\mid q p ∣ q . Each class intersects B \mathcal{B} B in ∣ B ∣ q − d + O ∗ ( 1 ) |\mathcal{B}|q^{-d}+O^*(1) ∣ B ∣ q − d + O ∗ ( 1 ) points if d = 1 d=1 d = 1 and ∣ B ∣ q − 2 + O ∗ ( 2 n / q + 1 ) |\mathcal{B}|q^{-2}+O^*(2n/q+1) ∣ B ∣ q − 2 + O ∗ ( 2 n / q + 1 ) points if d = 2 d=2 d = 2 , where O ∗ O^* O ∗ has constant one; so ∣ A q − ∣ B ∣ g ( q ) ∣ ≤ q |A_q-|\mathcal{B}|g(q)|\le q ∣ A q − ∣ B ∣ g ( q ) ∣ ≤ q for d = 1 d=1 d = 1 and ≤ 2 n q + q 2 \le2nq+q^2 ≤ 2 n q + q 2 for d = 2 d=2 d = 2 . Summing over the at most z 2 z^2 z 2 pairs ( e , f ) (e,f) ( e , f ) , with [ e , f ] ≤ z 2 [e,f]\le z^2 [ e , f ] ≤ z 2 , gives the error term; the main term is ∣ B ∣ G ( z ) − 1 ≤ n d G ( z ) − 1 |\mathcal{B}|G(z)^{-1}\le n^dG(z)^{-1} ∣ B ∣ G ( z ) − 1 ≤ n d G ( z ) − 1 . □
Lemma 5.6. In the setting of Lemma 5.5 , let p 0 ≥ 2 p_0\ge2 p 0 ≥ 2 , C ≥ 1 C\ge1 C ≥ 1 , and assume g ( p ) ≤ C / p g(p)\le C/p g ( p ) ≤ C / p for p ∈ P p\in\mathcal{P} p ∈ P .
(1) If P \mathcal{P} P contains every prime in ( p 0 , z ] (p_0,z] ( p 0 , z ] with at most one exception, and g ( p ) ≥ κ / p − C / p 2 g(p)\ge\kappa/p-C/p^2 g ( p ) ≥ κ / p − C / p 2 there for an integer κ ≥ 1 \kappa\ge1 κ ≥ 1 , then G ( z ) ≫ ( log z ) κ G(z)\gg(\log z)^\kappa G ( z ) ≫ ( log z ) κ .
(2) If P \mathcal{P} P contains every prime p ≡ 1 ( m o d 4 ) p\equiv1\pmod4 p ≡ 1 ( mod 4 ) in ( p 0 , z ] (p_0,z] ( p 0 , z ] with at most one exception, and g ( p ) ≥ 2 / p − C / p 2 g(p)\ge2/p-C/p^2 g ( p ) ≥ 2/ p − C / p 2 there, then G ( z ) ≫ log z G(z)\gg\log z G ( z ) ≫ log z .
The implied constants depend only on p 0 p_0 p 0 , C C C and κ \kappa κ .
Proof. Let δ > 0 \delta>0 δ > 0 be small, y = z δ y=z^\delta y = z δ , P y = P ∩ ( p 0 , y ] \mathcal{P}_y=\mathcal{P}\cap(p_0,y] P y = P ∩ ( p 0 , y ] , Π = ∏ p ∈ P y ( 1 + g 1 ( p ) ) \Pi=\prod_{p\in\mathcal{P}_y}(1+g_1(p)) Π = ∏ p ∈ P y ( 1 + g 1 ( p )) and a = 1 / log y a=1/\log y a = 1/ log y . Since g 1 ≥ 0 g_1\ge0 g 1 ≥ 0 ,
G ( z ) ≥ Π − ∑ e ∣ ∏ p ∈ P y p e > z g 1 ( e ) ≥ Π − z − a ∏ p ∈ P y ( 1 + g 1 ( p ) p a ) . G(z)\ge\Pi-\sum_{\substack{e\mid\prod_{p\in\mathcal{P}_y}p\\e>z}}g_1(e)\ge\Pi-z^{-a}\prod_{p\in\mathcal{P}_y}(1+g_1(p)p^a). G ( z ) ≥ Π − e ∣ ∏ p ∈ P y p e > z ∑ g 1 ( e ) ≥ Π − z − a p ∈ P y ∏ ( 1 + g 1 ( p ) p a ) .
For p ≤ y p \le y p ≤ y we have p a − 1 ≤ 2 a log p p^{a}-1 \le2a\log p p a − 1 ≤ 2 a log p and g 1 ( p ) ≤ 2 C / p g_{1}(p) \le2C/p g 1 ( p ) ≤ 2 C / p , so the last product is at most Π exp ( 4 C a ∑ p ≤ y log p / p ) ≤ C ′ Π \Pi\exp(4Ca\sum_{p\le y}\log p/p) \le C'\Pi Π exp ( 4 C a ∑ p ≤ y log p / p ) ≤ C ′ Π . As z − a = exp ( − 1 / δ ) z^{-a}=\exp(-1/\delta) z − a = exp ( − 1/ δ ) , for δ \delta δ small the subtracted term is at most Π / 2 \Pi/2 Π/2 . Finally Π ≫ ( log y ) κ \Pi\gg(\log y)^{\kappa} Π ≫ ( log y ) κ in case (1) and Π ≫ log y \Pi\gg\log y Π ≫ log y in case (2), by Mertens’ theorem and its version for the progression 1 1 1 mod 4 4 4 ; omitting one prime changes Π \Pi Π by a factor at least 1 / 2 1/2 1/2 , since g 1 ( p ) ≤ 1 g_{1}(p) \le1 g 1 ( p ) ≤ 1 .
Lemma 5.7. For Y Y Y large, the number of primes r ≤ Y r \le Y r ≤ Y , r ≡ 1 ( m o d 4 ) r \equiv1 \pmod{4} r ≡ 1 ( mod 4 ) , such that some odd prime ℓ > L \ell> L ℓ > L divides both a a a and r + 1 r+1 r + 1 , where r = a 2 + b 2 r=a^{2}+b^{2} r = a 2 + b 2 with a a a odd, is at most C ( Y / ( L log Y ) + Y 7 / 8 ) C(Y/(L\log Y)+Y^{7/8}) C ( Y / ( L log Y ) + Y 7/8 ) with C C C absolute.
Proof. Such ℓ \ell ℓ divides b 2 + 1 b^{2}+1 b 2 + 1 , so ℓ ≡ 1 ( m o d 4 ) \ell\equiv1 \pmod{4} ℓ ≡ 1 ( mod 4 ) , and ℓ ≤ ∣ a ∣ ≤ Y \ell\le|a| \le\sqrt{Y} ℓ ≤ ∣ a ∣ ≤ Y . We count pairs ( a , b ) (a,b) ( a , b ) with ∣ a ∣ , ∣ b ∣ ≤ Y |a|,|b| \le\sqrt{Y} ∣ a ∣ , ∣ b ∣ ≤ Y , ℓ ∣ a \ell\mid a ℓ ∣ a , b 2 ≡ − 1 ( m o d ℓ ) b^{2} \equiv-1 \pmod{\ell} b 2 ≡ − 1 ( mod ℓ ) and a 2 + b 2 a^{2}+b^{2} a 2 + b 2 prime. Primes r ≤ z = Y 1 / 100 r \le z=Y^{1/100} r ≤ z = Y 1/100 contribute at most z z z .
For L < ℓ ≤ Y 1 / 8 L < \ell\le Y^{1/8} L < ℓ ≤ Y 1/8 , fix a root b 0 b_{0} b 0 of b 2 ≡ − 1 ( m o d ℓ ) b^{2} \equiv-1 \pmod{\ell} b 2 ≡ − 1 ( mod ℓ ) and write a = ℓ U a=\ell U a = ℓ U , b = b 0 + ℓ V b=b_{0}+\ell V b = b 0 + ℓ V ; then ( U , V ) (U,V) ( U , V ) ranges over a square with n ≤ 3 Y / ℓ n \le3\sqrt{Y}/\ell n ≤ 3 Y / ℓ integers in each direction, and n ≥ Y 3 / 8 n \ge Y^{3/8} n ≥ Y 3/8 . Sieve by P = { p ≡ 1 ( m o d 4 ) : 5 ≤ p ≤ z , p ≠ ℓ } \mathcal{P}=\{p\equiv1\pmod{4}:5\le p\le z,\ p\ne\ell\} P = { p ≡ 1 ( mod 4 ) : 5 ≤ p ≤ z , p = ℓ } with Ω p = { F ≡ 0 } \Omega_{p}=\{F\equiv0\} Ω p = { F ≡ 0 } , where F ( U , V ) = ( ℓ U ) 2 + ( b 0 + ℓ V ) 2 F(U,V)=(\ell U)^{2}+(b_{0}+\ell V)^{2} F ( U , V ) = ( ℓ U ) 2 + ( b 0 + ℓ V ) 2 . Modulo p p p , F F F is a product of two independent linear forms in ( ℓ U , b 0 + ℓ V ) (\ell U,b_{0}+\ell V) ( ℓ U , b 0 + ℓ V ) , an affine bijection of ( U , V ) (U,V) ( U , V ) ; so g ( p ) = ( 2 p − 1 ) / p 2 ≤ 1 / 2 g(p)=(2p-1)/p^{2}\le1/2 g ( p ) = ( 2 p − 1 ) / p 2 ≤ 1/2 . A prime value r > z r>z r > z of F F F avoids every Ω p \Omega_{p} Ω p . Lemmas 5.5 and 5.6 (2) bound the count by O ( Y / ( ℓ 2 log Y ) ) + O ( Y 1 / 2 ℓ − 1 z 4 + z 6 ) O(Y/(\ell^{2}\log Y))+O(Y^{1/2}\ell^{-1}z^{4}+z^{6}) O ( Y / ( ℓ 2 log Y )) + O ( Y 1/2 ℓ − 1 z 4 + z 6 ) . The sum over the two roots and over ℓ > L \ell>L ℓ > L is O ( Y / ( L log Y ) + Y 7 / 8 ) O(Y/(L\log Y)+Y^{7/8}) O ( Y / ( L log Y ) + Y 7/8 ) .
For ℓ > Y 1 / 8 \ell>Y^{1/8} ℓ > Y 1/8 there are O ( Y / ℓ ) O(\sqrt{Y}/\ell) O ( Y / ℓ ) choices of a a a and O ( Y / ℓ + 1 ) O(\sqrt{Y}/\ell+1) O ( Y / ℓ + 1 ) of b b b , hence O ( Y / ℓ 2 + Y / ℓ ) O(Y/\ell^{2}+\sqrt{Y}/\ell) O ( Y / ℓ 2 + Y / ℓ ) pairs; the sum over Y 1 / 8 < ℓ ≤ Y Y^{1/8}<\ell\le\sqrt{Y} Y 1/8 < ℓ ≤ Y is O ( Y 7 / 8 + Y log Y ) O(Y^{7/8}+\sqrt{Y}\log Y) O ( Y 7/8 + Y log Y ) . □
Lemma 5.8. Let ψ \psi ψ be one of the forms ψ i \psi_{i} ψ i and r r r a prime. The number of T ∈ Z 2 ∩ X K \mathbf{T}\in\mathbb{Z}^{2}\cap X\mathcal{K} T ∈ Z 2 ∩ X K with ψ ( T ) = r \psi(\mathbf{T})=r ψ ( T ) = r at which the other t − 1 t-1 t − 1 forms take prime values is O ( X / ( log X ) t − 1 ) O(X/(\log X)^{t-1}) O ( X / ( log X ) t − 1 ) , uniformly in r r r .
Proof. As ψ \psi ψ is primitive, there is γ ∈ GL 2 ( Z ) \gamma\in\operatorname{GL}_{2}(\mathbb{Z}) γ ∈ GL 2 ( Z ) with ψ ∘ γ ( w 1 , w 2 ) = w 1 \psi\circ\gamma(w_{1},w_{2})=w_{1} ψ ∘ γ ( w 1 , w 2 ) = w 1 ; let ψ ′ ( T ) \psi'(\mathbf{T}) ψ ′ ( T ) be the second coordinate of γ − 1 T \gamma^{-1}\mathbf{T} γ − 1 T . Every other form is ψ j = α j ψ + β j ψ ′ \psi_{j}=\alpha_{j}\psi+\beta_{j}\psi' ψ j = α j ψ + β j ψ ′ with β j ≠ 0 \beta_{j}\ne0 β j = 0 and α j β k − α k β j ≠ 0 \alpha_{j}\beta_{k}-\alpha_{k}\beta_{j}\ne0 α j β k − α k β j = 0 for j ≠ k j\ne k j = k . On the line ψ = r \psi=r ψ = r , the points of X K X\mathcal{K} X K correspond to w = ψ ′ ( T ) w=\psi'(\mathbf{T}) w = ψ ′ ( T ) in an interval of length O ( X ) O(X) O ( X ) , and ψ j = α j r + β j w \psi_{j}=\alpha_{j}r+\beta_{j}w ψ j = α j r + β j w . Sieve this interval by the primes 2 t < p ≤ z = X 1 / 100 2t<p\le z=X^{1/100} 2 t < p ≤ z = X 1/100 with p ∤ r ∏ j β j ∏ j < k ( α j β k − α k β j ) p\nmid r\prod_{j}\beta_{j}\prod_{j<k}(\alpha_{j}\beta_{k}-\alpha_{k}\beta_{j}) p ∤ r ∏ j β j ∏ j < k ( α j β k − α k β j ) , removing the t − 1 t-1 t − 1 distinct residues − α j r / β j -\alpha_{j}r/\beta_{j} − α j r / β j ; thus g ( p ) = ( t − 1 ) / p g(p)=(t-1)/p g ( p ) = ( t − 1 ) / p . The prime values ψ j ≥ η 0 X > z \psi_{j}\ge\eta_{0}X>z ψ j ≥ η 0 X > z avoid these residues. Lemmas 5.5 and 5.6 (1), with κ = t − 1 \kappa=t-1 κ = t − 1 and p 0 p_{0} p 0 larger than 2 t 2t 2 t and than the excluded primes other than r r r , give the bound. □
Proof of Proposition 5.1
Let N b a d ( X ) N_{\mathrm{bad}}(X) N bad ( X ) count the admissible T ∈ X K \mathbf{T}\in X\mathcal{K} T ∈ X K with all ψ i ( T ) \psi_{i}(\mathbf{T}) ψ i ( T ) prime and some value r = a 2 + b 2 r=a^{2}+b^{2} r = a 2 + b 2 such that an odd prime ℓ > L \ell>L ℓ > L divides a a a and r + 1 r+1 r + 1 . Each value lies in [ η 0 X , 3 X ] [\eta_{0}X,3X] [ η 0 X , 3 X ] and is 1 1 1 modulo 4 4 4 , so Lemmas 5.7 and 5.8 imply the bound
N b a d ( X ) ≤ t C ( 3 X L log X + ( 3 X ) 7 / 8 ) ⋅ C ′ X ( log X ) t − 1 ≤ C ′ ′ X 2 L ( log X ) t + o ( X 2 ( log X ) t ) . (24) \begin{aligned}
N_{\mathrm{bad}}(X)\le tC\left(\frac{3X}{L\log X}+(3X)^{7/8}\right)\cdot C'\frac{X}{(\log X)^{t-1}} \\
&\le\frac{C''X^{2}}{L(\log X)^{t}}+o\left(\frac{X^{2}}{(\log X)^{t}}\right). \tag*{(24)}
\end{aligned} N bad ( X ) ≤ tC ( L log X 3 X + ( 3 X ) 7/8 ) ⋅ C ′ ( log X ) t − 1 X ≤ L ( log X ) t C ′′ X 2 + o ( ( log X ) t X 2 ) . (24)
with C ′ ′ C'' C ′′ independent of L L L . The points at which two of the ψ i \psi_{i} ψ i coincide lie on O ( 1 ) O(1) O ( 1 ) lines through the origin and number O ( X ) O(X) O ( X ) . Choose L L L with C ′ ′ / L < c 3 ( log L ) − 2 t / 4 C''/L<c_{3}(\log L)^{-2t}/4 C ′′ / L < c 3 ( log L ) − 2 t /4 , and then X X X large in terms of L L L and Y Y Y . By (23) and (24) , some admissible T ∈ X K \mathbf{T}\in X\mathcal{K} T ∈ X K has all ψ i ( T ) \psi_{i}(\mathbf{T}) ψ i ( T ) prime, pairwise distinct, larger than Y Y Y , and not counted by N b a d N_{\mathrm{bad}} N bad . Each value r r r is 1 1 1 modulo 8 8 8 and 3 3 3 modulo 7 7 7 . An odd prime ℓ \ell ℓ dividing a a a and r + 1 r+1 r + 1 satisfies ℓ ≡ 1 ( m o d 4 ) \ell\equiv1\pmod{4} ℓ ≡ 1 ( mod 4 ) and r ≡ − 1 ( m o d ℓ ) r\equiv-1\pmod{\ell} r ≡ − 1 ( mod ℓ ) . Admissibility excludes this for ℓ ≤ L \ell\le L ℓ ≤ L , since r ≡ 1 r\equiv1 r ≡ 1 modulo every prime ℓ ≤ 2 t \ell\le2t ℓ ≤ 2 t other than 7 7 7 , r ≡ 3 ( m o d 7 ) r\equiv3\pmod{7} r ≡ 3 ( mod 7 ) , and r ≢ − 1 r\not\equiv-1 r ≡ − 1 modulo every prime 2 t < ℓ ≤ L 2t<\ell\le L 2 t < ℓ ≤ L . The choice of T \mathbf{T} T excludes it for ℓ > L \ell>L ℓ > L . So all values are good. □
A Diophantine model of integer arithmetic
In this section we prove that polynomial relations among indices are Diophantine. For n ∈ Z n \in\mathbb{Z} n ∈ Z write z ( n P ) = a n + b n 2 z(nP)=a_n+b_n\sqrt{2} z ( n P ) = a n + b n 2 with a n , b n ∈ Q a_n,b_n\in\mathbb{Q} a n , b n ∈ Q , and put
ι ( n ) = ( a n , b n ) , Z = ι ( Z ) ⊂ Q 2 . \iota(n)=(a_n,b_n),\qquad\mathcal{Z}=\iota(\mathbb{Z})\subset\mathbb{Q}^2. ι ( n ) = ( a n , b n ) , Z = ι ( Z ) ⊂ Q 2 .
By Lemma 2.4, ι \iota ι is a bijection onto Z \mathcal{Z} Z ; it is computed by exact arithmetic in F F F , and its inverse by enumerating the multiples of P P P .
Theorem 6.1. The set Z ⊂ Q 2 \mathcal{Z}\subset\mathbb{Q}^2 Z ⊂ Q 2 is Diophantine. For every polynomial f ∈ Z [ X 1 , … , X ν ] f\in\mathbb{Z}[X_1,\ldots,X_\nu] f ∈ Z [ X 1 , … , X ν ] , the set
Z f = { ( ι ( n 1 ) , … , ι ( n ν ) ) : f ( n 1 , … , n ν ) = 0 } \mathcal{Z}_f=\{(\iota(n_1),\ldots,\iota(n_\nu)): f(n_1,\ldots,n_\nu)=0\} Z f = {( ι ( n 1 ) , … , ι ( n ν )) : f ( n 1 , … , n ν ) = 0 } is a Diophantine subset of Q 2 ν \mathbb{Q}^{2\nu} Q 2 ν .
Remark. The use of squared coordinate components, a common parameter and the rational function in Definition 6.2, together with the height comparison in Proposition 6.4, follows [86 §7.1, Proposition 9.1] .
We fix the points θ 0 = 0 , θ 1 , … , θ 4 \theta_0=0,\theta_1,\ldots,\theta_4 θ 0 = 0 , θ 1 , … , θ 4 and the set S H S_H S H of Theorem E, put c = 24 c=24 c = 24 , and put
F 5 ( T ) = ∏ j = 0 4 ( T − θ j ) . F_5(T)=\prod_{j=0}^{4}(T-\theta_j). F 5 ( T ) = j = 0 ∏ 4 ( T − θ j ) .
Thus h ( t ) ≤ H S M S ( t ) c h(t)\leq H_{S}M_{S}(t)^c h ( t ) ≤ H S M S ( t ) c for every finite set S ⊇ S H S\supseteq S_H S ⊇ S H and t ∈ Q ∖ { θ 0 , … , θ 4 } t\in\mathbb{Q}\setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } .
Addition on Z \mathcal{Z} Z is defined by the group law. We construct a single system for multiplication; every polynomial relation then follows by introducing intermediate points. The system expresses the local conditions of Definition 2.7 at every prime of odd contact of rational numbers t A t_A t A attached to the model coordinates (see Definition 6.2). If Δ = n 1 n 2 − n 3 ≠ 0 \Delta=n_1n_2-n_3\neq0 Δ = n 1 n 2 − n 3 = 0 , Lemmas 4.1(1) and 2.8 and Theorem E imply, with S S S as in §6.1,
M S ( t A ) 48 ∣ Δ ⟹ h ( t A ) ≤ H S ∣ Δ ∣ . M_S(t_A)^{48}\mid\Delta\quad\Longrightarrow\quad h(t_A)\leq H_S\sqrt{|\Delta|}. M S ( t A ) 48 ∣ Δ ⟹ h ( t A ) ≤ H S ∣Δ∣ .
The shifted coordinate squares below compare this bound with the quadratic growth of canonical heights.
Fixed data
Fix ξ ∈ Q ∖ { θ 0 , … , θ 4 } \xi\in\mathbb{Q}\setminus\{\theta_0,\ldots,\theta_4\} ξ ∈ Q ∖ { θ 0 , … , θ 4 } , and throughout the construction put
f 0 = X 1 X 2 − X 3 , K = 2 c = 48 , G = 30. f_0=X_1X_2-X_3,\qquad K=2c=48,\qquad G=30. f 0 = X 1 X 2 − X 3 , K = 2 c = 48 , G = 30.
Thus we apply the local test with three indices and a polynomial of degree two. We normalize the canonical height on E ( F ) E(F) E ( F ) by h ^ ( Q ) = lim k 4 − k h ( x ( 2 k Q ) ) \widehat{h}(Q)=\lim_{k}4^{-k}h(x(2^kQ)) h ( Q ) = lim k 4 − k h ( x ( 2 k Q )) . Then h ^ ( n P ) = n 2 h ^ ( P ) \widehat{h}(nP)=n^2\widehat{h}(P) h ( n P ) = n 2 h ( P ) with h ^ ( P ) > 0 \widehat{h}(P)>0 h ( P ) > 0 . For Q ∈ ⟨ P ⟩ ∖ { O } Q\in\langle P\rangle\setminus\{O\} Q ∈ ⟨ P ⟩ ∖ { O } the identity
x ( Q ) 2 − d 2 x ( Q ) = 1 z ( Q ) 2 \frac{x(Q)^2-d^2}{x(Q)}=\frac{1}{z(Q)^2} x ( Q ) x ( Q ) 2 − d 2 = z ( Q ) 2 1
implies h ( x ( Q ) ) = h ( z ( Q ) ) + O E ( 1 ) h(x(Q))=h(z(Q))+O_E(1) h ( x ( Q )) = h ( z ( Q )) + O E ( 1 ) , since the rational function on the left has degree two [2 §1.5] . The comparison with the canonical height [82 VIII.9] and the inequality h ( a + b 2 ) ≤ h ( a ) + h ( b ) + 2 h(a+b\sqrt{2})\leq h(a)+h(b)+2 h ( a + b 2 ) ≤ h ( a ) + h ( b ) + 2 therefore provide C z ≥ 0 C_z\geq0 C z ≥ 0 such that, for Q ∈ ⟨ P ⟩ ∖ { O } Q\in\langle P\rangle\setminus\{O\} Q ∈ ⟨ P ⟩ ∖ { O } with z ( Q ) = r 0 + r 1 2 z(Q)=r_0+r_1\sqrt{2} z ( Q ) = r 0 + r 1 2 ,
h ^ ( Q ) ≤ h ( r 0 ) + h ( r 1 ) + C z . (25) \widehat{h}(Q)\leq h(r_0)+h(r_1)+C_z. \tag*{(25)} h ( Q ) ≤ h ( r 0 ) + h ( r 1 ) + C z . (25)
Let S S S be the union6 of S H S_H S H , S 1 S_1 S 1 , { 2 , 7 } \{2,7\} { 2 , 7 } , the primes at most K + 1 K+1 K + 1 , and the primes dividing the numerator or the denominator of some θ j \theta_j θ j (j ≥ 1 j\geq1 j ≥ 1 ), θ j − θ k \theta_j-\theta_k θ j − θ k (j ≠ k j\neq k j = k ) or θ j − ξ \theta_j-\xi θ j − ξ . Let H S H_S H S be the constant of Theorem E, and fix an integer N ∗ ≥ 2 N^* \ge2 N ∗ ≥ 2 with
log N ∗ > H S 2 h ( P ) + C z 2 . \log N^* > \frac{H_S^2}{h(P)} + \frac{C_z}{2}. log N ∗ > h ( P ) H S 2 + 2 C z .
The set S S S is fixed before N ∗ N^* N ∗ . The height constant H S H_S H S enters the system only through coefficients depending on N ∗ N^* N ∗ ; the degrees and the number of variables are independent of its size.
For r = u / v ∈ Q r=u/v\in\mathbb{Q} r = u / v ∈ Q in lowest terms with v > 0 v>0 v > 0 , the fraction ( ( N ∗ ) 2 v 2 + u 2 ) / v 2 ((N^*)^2v^2+u^2)/v^2 (( N ∗ ) 2 v 2 + u 2 ) / v 2 is also in lowest terms. Hence
h ( ( N ∗ ) 2 + r 2 ) = log ( ( N ∗ ) 2 v 2 + u 2 ) ≥ max { 2 h ( r ) , 2 log N ∗ } ≥ h ( r ) + log N ∗ . h((N^*)^2+r^2)=\log((N^*)^2v^2+u^2)
\ge\max\{2h(r),2\log N^*\}\ge h(r)+\log N^*. h (( N ∗ ) 2 + r 2 ) = log (( N ∗ ) 2 v 2 + u 2 ) ≥ max { 2 h ( r ) , 2 log N ∗ } ≥ h ( r ) + log N ∗ .
The system.
Definition 6.2. The unknowns of Σ \Sigma Σ are a rational number s s s , points P ~ i , J ~ , H ~ g , U ~ g , i ∈ E ( F ) \widetilde{P}_i,\widetilde{J},\widetilde{H}_g,\widetilde{U}_{g,i}\in E(F) P i , J , H g , U g , i ∈ E ( F ) (1 ≤ i ≤ 3 1\le i\le3 1 ≤ i ≤ 3 , 1 ≤ g < G 1\le g<G 1 ≤ g < G ), and auxiliary variables for the instances of Φ \Phi Φ below. Put P i = 2 P ~ i P_i=2\widetilde{P}_i P i = 2 P i , J = 2 J ~ J=2\widetilde{J} J = 2 J , H g = 2 H ~ g H_g=2\widetilde{H}_g H g = 2 H g and U g , i = 2 U ~ g , i U_{g,i}=2\widetilde{U}_{g,i} U g , i = 2 U g , i , which lie in ⟨ P ⟩ \langle P\rangle ⟨ P ⟩ , let n i n_i n i be the index of P i P_i P i , and put τ g = ( H g , U g , 1 , U g , 2 , U g , 3 , J ) \tau_g=(H_g,U_{g,1},U_{g,2},U_{g,3},J) τ g = ( H g , U g , 1 , U g , 2 , U g , 3 , J ) .
Write z ( P i ) = r i , 0 + r i , 1 2 z(P_i)=r_{i,0}+r_{i,1}\sqrt{2} z ( P i ) = r i , 0 + r i , 1 2 with r i , e ∈ Q r_{i,e}\in\mathbb{Q} r i , e ∈ Q , and set
A i , e = ( ( N ∗ ) 2 + r i , e 2 ) 2 ( 1 ≤ i ≤ 3 , e = 0 , 1 ) . A_{i,e}=\left((N^*)^2+r_{i,e}^2\right)^2\qquad(1\le i\le3,\ e=0,1). A i , e = ( ( N ∗ ) 2 + r i , e 2 ) 2 ( 1 ≤ i ≤ 3 , e = 0 , 1 ) . Let A \mathcal{A} A be the list of the seven numbers 1 1 1 and A i , e A_{i,e} A i , e . Every entry is a non-zero rational square, including when P i = O P_i=O P i = O . For A ∈ A A\in\mathcal{A} A ∈ A put
t A = A s , R A = F 5 ( t A ) ( t A − ξ ) 6 , t_A=As,\qquad R_A=\frac{F_5(t_A)}{(t_A-\xi)^6}, t A = A s , R A = ( t A − ξ ) 6 F 5 ( t A ) , W g , A = ∏ j = 1 10 ( 1 + R A ( D j ( τ g ) R A − k j ) 2 ) , W_{g,A}=\prod_{j=1}^{10}\left(1+R_A\left(D_j(\tau_g)R_A^{-k_j}\right)^2\right), W g , A = j = 1 ∏ 10 ( 1 + R A ( D j ( τ g ) R A − k j ) 2 ) , Θ A = Tr F / Q ( ∑ g = 1 G W g , A − 1 ) , B A = R A Θ A 2 . \Theta_A=\operatorname{Tr}_{F/\mathbb{Q}}\left(\sum_{g=1}^{G}W_{g,A}^{-1}\right),\qquad B_A=\frac{R_A}{\Theta_A^2}. Θ A = Tr F / Q ( g = 1 ∑ G W g , A − 1 ) , B A = Θ A 2 R A . with D j D_j D j and k j k_j k j as in Definition 2.7, for f = f 0 f=f_0 f = f 0 , K = 48 K=48 K = 48 and P 1 , P 2 , P 3 P_1,P_2,P_3 P 1 , P 2 , P 3 . The system Σ \Sigma Σ consists of the conditions
(C1) J ≠ O J\ne O J = O , and ℓ K ( z ( H g ) ) ≠ 0 \ell_K(z(H_g))\ne0 ℓ K ( z ( H g )) = 0 for every g g g ;
(C2) for every A ∈ A A\in\mathcal{A} A ∈ A : t A ∉ { θ 0 , … , θ 4 , ξ } t_A\notin\{\theta_0,\ldots,\theta_4,\xi\} t A ∈ / { θ 0 , … , θ 4 , ξ } , W g , A ≠ 0 W_{g,A}\ne0 W g , A = 0 for every g g g , Θ A ≠ 0 \Theta_A\ne0 Θ A = 0 , and Φ ( B A ) \Phi(B_A) Φ ( B A ) .
By (C1), every τ g \tau_g τ g is a test datum. The divisor of F 5 ( T ) / ( T − ξ ) 6 F_5(T)/(T-\xi)^6 F 5 ( T ) / ( T − ξ ) 6 on P 1 \mathbb{P}^1 P 1 is ∑ j [ θ j ] + [ ∞ ] − 6 [ ξ ] \sum_j[\theta_j]+[\infty]-6[\xi] ∑ j [ θ j ] + [ ∞ ] − 6 [ ξ ] . Accordingly, for a prime p ∉ S p\notin S p ∈ / S , v p ( R A ) v_p(R_A) v p ( R A ) is odd exactly when p p p is a prime of odd contact of t A t_A t A or v p ( t A ) v_p(t_A) v p ( t A ) is negative and odd. This accounts for the six primes attached to each A A A in the proof of completeness, and for G = 4 ⋅ 7 + 2 = 30 G=4\cdot7+2=30 G = 4 ⋅ 7 + 2 = 30 , the primes u u u and v ′ v' v ′ below being shared by all A A A . The prime v ′ v' v ′ in the denominator of s s s permits the use of linear Green–Tao–Ziegler patterns.
Completeness.
Proposition 6.3. Let n 1 , n 2 , n 3 ∈ Z n_1,n_2,n_3\in\mathbb{Z} n 1 , n 2 , n 3 ∈ Z satisfy n 1 n 2 = n 3 n_1n_2=n_3 n 1 n 2 = n 3 . Then Σ \Sigma Σ has a solution with P i = n i P P_i=n_iP P i = n i P for 1 ≤ i ≤ 3 1\le i\le3 1 ≤ i ≤ 3 .
Proof. Put P ~ i = n i P ~ \widetilde{P}_i=n_i\widetilde{P} P i = n i P ; this determines A \mathcal{A} A . Write each A ∈ A A\in\mathcal{A} A ∈ A as ( x A / y A ) 2 (x_A/y_A)^2 ( x A / y A ) 2 with coprime positive integers x A , y A x_A,y_A x A , y A . Let W G = 56 ∏ 3 ≤ ℓ ≤ 2 G ℓ W_G=56\prod_{3\le\ell\le2G}\ell W G = 56 ∏ 3 ≤ ℓ ≤ 2 G ℓ , let D θ ≥ 1 D_\theta\ge1 D θ ≥ 1 be an integer with D θ θ j ∈ Z D_\theta\theta_j\in\mathbb{Z} D θ θ j ∈ Z for all j j j , and put
W 1 = ( W G D θ ∏ A ∈ A x A ) 2 , λ j , A = θ j W 1 y A 2 x A 2 ( 1 ≤ j ≤ 4 , A ∈ A ) . W_1=\left(W_GD_\theta\prod_{A\in\mathcal{A}}x_A\right)^2,\qquad
\lambda_{j,A}=\theta_j\frac{W_1y_A^2}{x_A^2}\qquad(1\le j\le4,\ A\in\mathcal{A}). W 1 = ( W G D θ A ∈ A ∏ x A ) 2 , λ j , A = θ j x A 2 W 1 y A 2 ( 1 ≤ j ≤ 4 , A ∈ A ) .
Each λ j , A \lambda_{j,A} λ j , A is a non-zero integer divisible by W G W_G W G . Let Λ 0 \Lambda_0 Λ 0 be the set of the λ j , A \lambda_{j,A} λ j , A . Then t = ∣ Λ 0 ∣ + 2 ≤ G t = |\Lambda_0| + 2 \le G t = ∣ Λ 0 ∣ + 2 ≤ G , so the modulus W 0 W_0 W 0 of Proposition 5.1 divides W G W_G W G . Let Y Y Y exceed 2 G 2G 2 G , every prime of S S S , and every prime dividing W 1 W_1 W 1 , some x A y A x_Ay_A x A y A , or the numerator or denominator of ξ \xi ξ or of some θ j − ξ \theta_j-\xi θ j − ξ . By Proposition 5.1 , choose u , v ′ u,v' u , v ′ such that u u u , v ′ v' v ′ and the u − λ v ′ u-\lambda v' u − λ v ′ , λ ∈ Λ 0 \lambda\in\Lambda_0 λ ∈ Λ 0 , are pairwise distinct good primes larger than Y Y Y . Put s = u / ( W 1 v ′ ) s=u/(W_1v') s = u / ( W 1 v ′ ) .
By (6.2), for A ∈ A A \in\mathcal{A} A ∈ A we have
t A − θ 0 = x A 2 u y A 2 W 1 v ′ , t A − θ j = x A 2 ( u − λ j , A v ′ ) y A 2 W 1 v ′ ( 1 ≤ j ≤ 4 ) , t_A-\theta_0=\frac{x_A^2u}{y_A^2W_1v'}, \qquad t_A-\theta_j=\frac{x_A^2(u-\lambda_{j,A}v')}{y_A^2W_1v'} \quad(1\le j\le4), t A − θ 0 = y A 2 W 1 v ′ x A 2 u , t A − θ j = y A 2 W 1 v ′ x A 2 ( u − λ j , A v ′ ) ( 1 ≤ j ≤ 4 ) ,
hence
R A = x A 10 y A 2 W 1 ⋅ u v ′ ∏ j = 1 4 ( u − λ j , A v ′ ) ( x A 2 u − ξ y A 2 W 1 v ′ ) 6 . R_A=\frac{x_A^{10}y_A^2W_1\cdot uv'\prod_{j=1}^{4}(u-\lambda_{j,A}v')}{(x_A^2u-\xi y_A^2W_1v')^6}. R A = ( x A 2 u − ξ y A 2 W 1 v ′ ) 6 x A 10 y A 2 W 1 ⋅ u v ′ ∏ j = 1 4 ( u − λ j , A v ′ ) .
The six primes T A = { u , v ′ , u − λ j , A v ′ ( 1 ≤ j ≤ 4 ) } \mathcal{T}_A=\{u,v',u-\lambda_{j,A}v'\ (1\le j\le4)\} T A = { u , v ′ , u − λ j , A v ′ ( 1 ≤ j ≤ 4 )} are distinct, so t A ∉ { θ 0 , … , θ 4 } t_A\notin\{\theta_0,\ldots,\theta_4\} t A ∈ / { θ 0 , … , θ 4 } . Let p ∈ T A p\in\mathcal{T}_A p ∈ T A ; then p ∤ x A y A W 1 p\nmid x_Ay_AW_1 p ∤ x A y A W 1 . If p = u p=u p = u , then p ∤ ξ y A 2 W 1 v ′ p\nmid\xi y_A^2W_1v' p ∤ ξ y A 2 W 1 v ′ ; if p = v ′ p=v' p = v ′ , then p ∤ x A 2 u p\nmid x_A^2u p ∤ x A 2 u ; and if p = u − λ j , A v ′ p=u-\lambda_{j,A}v' p = u − λ j , A v ′ , then x A 2 u − ξ y A 2 W 1 v ′ ≡ ( θ j − ξ ) y A 2 W 1 v ′ ≢ 0 ( m o d p ) x_A^2u-\xi y_A^2W_1v'\equiv(\theta_j-\xi)y_A^2W_1v'\not\equiv0\pmod p x A 2 u − ξ y A 2 W 1 v ′ ≡ ( θ j − ξ ) y A 2 W 1 v ′ ≡ 0 ( mod p ) , as x A 2 λ j , A = θ j y A 2 W 1 x_A^2\lambda_{j,A}=\theta_jy_A^2W_1 x A 2 λ j , A = θ j y A 2 W 1 . Hence t A ≠ ξ t_A\ne\xi t A = ξ and v p ( R A ) = 1 v_p(R_A)=1 v p ( R A ) = 1 . For p ∉ T A p\notin\mathcal{T}_A p ∈ / T A the valuation v p ( R A ) v_p(R_A) v p ( R A ) is even, because W 1 W_1 W 1 is a square.
Let r 1 , … , r t ′ r_1,\ldots,r_{t'} r 1 , … , r t ′ , t ′ ≤ G t'\le G t ′ ≤ G , be the distinct primes among u u u , v ′ v' v ′ and the u − λ v ′ u-\lambda v' u − λ v ′ . For g ≤ t ′ g\le t' g ≤ t ′ choose a place p g ∣ r g \mathfrak{p}_g\mid r_g p g ∣ r g ; let j j j be the least common multiple of the orders of the reductions of σ \sigma σ at p 1 , … , p t ′ \mathfrak{p}_1,\ldots,\mathfrak{p}_{t'} p 1 , … , p t ′ , and put J = j P J=jP J = j P . For g ≤ t ′ g\le t' g ≤ t ′ let τ g = ( H g , U g , 1 , U g , 2 , U g , 3 , J ) \tau_g=(H_g,U_{g,1},U_{g,2},U_{g,3},J) τ g = ( H g , U g , 1 , U g , 2 , U g , 3 , J ) be a test datum satisfying the local conditions at p g \mathfrak{p}_g p g , given by Lemma 2.9 , as r g r_g r g is good and r g > K + 1 r_g>K+1 r g > K + 1 ; for t ′ < g ≤ G t'<g\le G t ′ < g ≤ G put τ g = τ 1 \tau_g=\tau_1 τ g = τ 1 . The orders at the other primes enter only through v r g ( j ) v_{r_g}(j) v r g ( j ) in Lemma 2.9 , so no compatibility between different primes is needed. All these points lie in ⟨ P ⟩ = 2 E ( F ) \langle P\rangle=2E(F) ⟨ P ⟩ = 2 E ( F ) , and (C1) holds. For A ∈ A A\in\mathcal{A} A ∈ A , the set T A \mathcal{T}_A T A consists of split primes larger than 2 G 2G 2 G , v p ( R A ) = 1 v_p(R_A)=1 v p ( R A ) = 1 on T A \mathcal{T}_A T A , v p ( R A ) v_p(R_A) v p ( R A ) is even elsewhere, and at each p ∈ T A p\in\mathcal{T}_A p ∈ T A some τ g \tau_g τ g satisfies the local conditions at a place above p p p . Lemma 4.1 (2) verifies W g , A ≠ 0 W_{g,A}\ne0 W g , A = 0 , Θ A ≠ 0 \Theta_A\ne0 Θ A = 0 and B A ∈ D 2 B_A\in D_2 B A ∈ D 2 , so Φ ( B A ) \Phi(B_A) Φ ( B A ) holds by Proposition 3.14. This verifies (C2). □
Soundness
Proposition 6.4. In every solution of Σ \Sigma Σ , the indices of P 1 , P 2 , P 3 P_1,P_2,P_3 P 1 , P 2 , P 3 satisfy n 1 n 2 = n 3 n_1n_2=n_3 n 1 n 2 = n 3 .
Proof. Suppose that a solution has Δ = n 1 n 2 − n 3 ≠ 0 \Delta=n_1n_2-n_3\ne0 Δ = n 1 n 2 − n 3 = 0 .
Let A ∈ A A\in\mathcal{A} A ∈ A , and let p ∉ S p\notin S p ∈ / S be a prime of odd contact of t A t_A t A , so that k = v p ( t A − θ j ) k=v_p(t_A-\theta_j) k = v p ( t A − θ j ) is positive and odd for some j j j . As the θ i \theta_i θ i are p p p -integral and pairwise incongruent modulo p p p , and θ j − ξ \theta_j-\xi θ j − ξ is a p p p -unit, the numbers t A − θ i t_A-\theta_i t A − θ i for i ≠ j i\ne j i = j and t A − ξ t_A-\xi t A − ξ are p p p -units, and v p ( R A ) = k v_p(R_A)=k v p ( R A ) = k is odd. By Lemma 4.1 (1) some τ g \tau_g τ g satisfies the local conditions at a place above p p p , and Lemma 2.8 gives p K ∣ Δ p^K\mid\Delta p K ∣ Δ . So M S ( t A ) K ∣ Δ M_S(t_A)^K\mid\Delta M S ( t A ) K ∣ Δ . By Theorem E,
h ( t A ) ≤ H S ∣ Δ ∣ ( A ∈ A ) . h(t_A)\le H_S\sqrt{|\Delta|}\qquad(A\in\mathcal{A}). h ( t A ) ≤ H S ∣Δ∣ ( A ∈ A ) .
For A = A i , e A=A_{i,e} A = A i , e , the identity A = t A / s A=t_A/s A = t A / s and s = t 1 ≠ 0 s=t_1\ne0 s = t 1 = 0 imply
h ( r i , e ) + log N ∗ ≤ h ( ( N ∗ ) 2 + r i , e 2 ) = 1 2 h ( A ) ≤ 1 2 ( h ( t A ) + h ( s ) ) ≤ H S ∣ Δ ∣ . (26) \begin{aligned}
h(r_{i,e})+\log N^*\le h\left((N^*)^2+r_{i,e}^2\right)=\frac12h(A) \\
&\le\frac12\left(h(t_A)+h(s)\right)\le H_S\sqrt{|\Delta|}. \tag*{(26)}
\end{aligned} h ( r i , e ) + log N ∗ ≤ h ( ( N ∗ ) 2 + r i , e 2 ) = 2 1 h ( A ) ≤ 2 1 ( h ( t A ) + h ( s ) ) ≤ H S ∣Δ∣ . (26)
Put B = max ( 1 , ∣ n 1 ∣ , ∣ n 2 ∣ , ∣ n 3 ∣ ) B=\max(1,|n_1|,|n_2|,|n_3|) B = max ( 1 , ∣ n 1 ∣ , ∣ n 2 ∣ , ∣ n 3 ∣ ) . Then ∣ Δ ∣ ≤ B 2 + B ≤ 2 B 2 |\Delta|\le B^2+B\le2B^2 ∣Δ∣ ≤ B 2 + B ≤ 2 B 2 . As Δ ≠ 0 \Delta\ne0 Δ = 0 , some n i ≠ 0 n_i\ne0 n i = 0 , so we can choose i i i with ∣ n i ∣ = B |n_i|=B ∣ n i ∣ = B . Summing (26) for its two coordinates and using (25) , we obtain
h ^ ( P ) B 2 + 2 log N ∗ ≤ 2 H S ∣ Δ ∣ + C Z ≤ 2 2 H S B + C Z . \widehat{h}(P)B^2+2\log N^*\le2H_S\sqrt{|\Delta|}+C_{\mathbb{Z}}\le2\sqrt{2}H_SB+C_{\mathbb{Z}}. h ( P ) B 2 + 2 log N ∗ ≤ 2 H S ∣Δ∣ + C Z ≤ 2 2 H S B + C Z .
Completing the square yields
2 log N ∗ ≤ 2 H S 2 h ^ ( P ) + C z − h ^ ( P ) ( B − 2 H S h ^ ( P ) ) 2 ≤ 2 H S 2 h ^ ( P ) + C z , 2\log N^* \leq\frac{2H_S^2}{\hat{h}(P)}+Cz-\hat{h}(P)\left(B-\frac{\sqrt{2}H_S}{\hat{h}(P)}\right)^2 \leq\frac{2H_S^2}{\hat{h}(P)}+Cz, 2 log N ∗ ≤ h ^ ( P ) 2 H S 2 + C z − h ^ ( P ) ( B − h ^ ( P ) 2 H S ) 2 ≤ h ^ ( P ) 2 H S 2 + C z ,
contrary to the choice of N ∗ N^* N ∗ .
Proof of Theorem 6.1
Writing its two coordinates in the basis 1 , 2 1,\sqrt{2} 1 , 2 , an affine point of E ( F ) E(F) E ( F ) is described by two equations over Q \mathbb{Q} Q ; the point O O O is treated separately. The addition law is given by polynomial identities in finitely many cases [82 III.2.3] ; so are the maps Q ↦ 2 Q Q \mapsto2Q Q ↦ 2 Q , Q ↦ Q − P Q \mapsto Q-P Q ↦ Q − P , Q ↦ Q − P i Q \mapsto Q-P_i Q ↦ Q − P i and Q ↦ Q σ Q \mapsto Q^\sigma Q ↦ Q σ . A pair ( a , b ) (a,b) ( a , b ) lies in Z \mathbb{Z} Z if and only if a + b 2 = z ( 2 Q ~ ) a+b\sqrt{2}=z(2\widetilde{Q}) a + b 2 = z ( 2 Q ) for some Q ~ ∈ E ( F ) \widetilde{Q}\in E(F) Q ∈ E ( F ) , so Z \mathbb{Z} Z is Diophantine.
By Lemma 2.4 , a triple in Z 3 \mathbb{Z}^3 Z 3 determines points n i P ∈ 2 E ( F ) n_iP\in2E(F) n i P ∈ 2 E ( F ) . Propositions 6.3 and 6.4 show that n 1 n 2 = n 3 n_1n_2=n_3 n 1 n 2 = n 3 exactly when these points extend to a solution of Σ \Sigma Σ . Introduce a rational variable for each of R A R_A R A , Θ A \Theta_A Θ A , B A B_A B A , and two for each of D j ( τ g ) D_j(\tau_g) D j ( τ g ) , W g , A ∈ F W_{g,A}\in F W g , A ∈ F , with the polynomial equations obtained by clearing denominators and the conditions that the denominators are non-zero. The distinctions between O O O and affine points are finite disjunctions, and an equation over F F F is two equations over Q \mathbb{Q} Q . Thus multiplication on Z \mathbb{Z} Z is Diophantine.
An integer constant m m m is represented by the fixed point m P mP m P . For any polynomial f ∈ Z [ X 1 , … , X ν ] f\in\mathbb{Z}[X_1,\ldots,X_\nu] f ∈ Z [ X 1 , … , X ν ] , introduce one point for each intermediate value in an arithmetic circuit for f f f . Addition is expressed by the group law, and multiplication by Σ \Sigma Σ . The condition that the output is O O O defines Z f \mathcal{Z}_f Z f by a positive existential formula. This proves the theorem.
Lemma 6.5. Let R ⊆ Z r \mathcal{R}\subseteq\mathbb{Z}^r R ⊆ Z r be recursively enumerable. Then { ( ι ( n 1 ) , … , ι ( n r ) ) : ( n 1 , … , n r ) ∈ R } \{(\iota(n_1),\ldots,\iota(n_r)):(n_1,\ldots,n_r)\in\mathcal{R}\} {( ι ( n 1 ) , … , ι ( n r )) : ( n 1 , … , n r ) ∈ R } is a Diophantine subset of Q 2 r \mathbb{Q}^{2r} Q 2 r .
Proof. By the theorem of Davis, Putnam, Robinson and Matiyasevich [47 , 13 ] there exists f ∈ Z [ X 1 , … , X r + s ] f\in\mathbb{Z}[X_1,\ldots,X_{r+s}] f ∈ Z [ X 1 , … , X r + s ] such that ( n 1 , … , n r ) ∈ R (n_1,\ldots,n_r)\in\mathcal{R} ( n 1 , … , n r ) ∈ R if and only if f ( n 1 , … , n r + s ) = 0 f(n_1,\ldots,n_{r+s})=0 f ( n 1 , … , n r + s ) = 0 for some n r + 1 , … , n r + s ∈ Z n_{r+1},\ldots,n_{r+s}\in\mathbb{Z} n r + 1 , … , n r + s ∈ Z . The set in question is the projection of Z f \mathcal{Z}_f Z f to the first 2 r 2r 2 r coordinates.
Finite sets of bounded height and the integers
In this section we prove Theorem A. We construct Diophantine finite sets of rational numbers of controlled height exhausting Q \mathbb{Q} Q (see Theorem 7.2 ), and then identify the index n n n of n P nP n P with the rational number n n n by approximation at one prime (see Theorem 7.5 ).
Throughout, S S S denotes a finite set of primes containing S H S_H S H , S 1 S_1 S 1 , 2 2 2 , 7 7 7 and the primes dividing the numerator or the denominator of some θ j \theta_j θ j (j ≥ 1 j\geq1 j ≥ 1 ) or θ j − θ k \theta_j-\theta_k θ j − θ k (j ≠ k j\neq k j = k ), and H S H_S H S is the corresponding constant of Theorem E.
Parity and integrality
Lemma 7.1. Let S ′ ⊇ S 1 S'\supseteq S_1 S ′ ⊇ S 1 be a finite set of primes, and let Ψ S ′ ( x ) \Psi_{S'}(x) Ψ S ′ ( x ) be the disjunction of the conditions Φ ( e x ) \Phi(ex) Φ ( e x ) over the positive divisors e e e of ∏ p ∈ S ′ p \prod_{p\in S'}p ∏ p ∈ S ′ p . Then Ψ S ′ \Psi_{S'} Ψ S ′ defines D 2 S ′ \mathcal{D}_2^{S'} D 2 S ′ . For a prime p 0 ∉ S 1 p_0\notin S_1 p 0 ∈ / S 1 and w ∈ Q w\in\mathbb{Q} w ∈ Q , Φ ( p 0 w 2 ) \Phi(p_0w^2) Φ ( p 0 w 2 ) holds if and only if w ∈ Z ( p 0 ) w\in\mathbb{Z}_{(p_0)} w ∈ Z ( p 0 ) . If p 0 ≡ 3 , 5 ( m o d 8 ) p_0\equiv3,5\pmod{8} p 0 ≡ 3 , 5 ( mod 8 ) , then
Φ ( p 0 ( w 0 2 − 2 w 1 2 ) 2 ) ⟺ w 0 , w 1 ∈ Z ( p 0 ) \Phi\left(p_0(w_0^2-2w_1^2)^2\right)\Longleftrightarrow w_0,w_1\in\mathbb{Z}_{(p_0)} Φ ( p 0 ( w 0 2 − 2 w 1 2 ) 2 ) ⟺ w 0 , w 1 ∈ Z ( p 0 ) for w 0 , w 1 ∈ Q w_0,w_1\in\mathbb{Q} w 0 , w 1 ∈ Q .
Proof. Let x ∈ D 2 S ′ x\in\mathcal{D}_2^{S'} x ∈ D 2 S ′ and let e e e be the product of the primes p ∈ S ′ p\in S' p ∈ S ′ at which v p ( x ) v_p(x) v p ( x ) is negative and odd. Then e x ∈ D 2 ex\in\mathcal{D}_2 e x ∈ D 2 , and Φ ( e x ) \Phi(ex) Φ ( e x ) holds by Proposition 3.14. Conversely, if Φ ( e x ) \Phi(ex) Φ ( e x ) holds, then e x ∈ D 2 S 1 ex\in\mathcal{D}_2^{S_1} e x ∈ D 2 S 1 by Proposition 3.2, and v p ( x ) = v p ( e x ) v_p(x)=v_p(ex) v p ( x ) = v p ( e x ) for p ∉ S ′ p\notin S' p ∈ / S ′ ; so x ∈ D 2 S ′ x\in\mathcal{D}_2^{S'} x ∈ D 2 S ′ . For the second assertion let w ≠ 0 w\neq0 w = 0 . As v p 0 ( p 0 w 2 ) v_{p_0}(p_0w^2) v p 0 ( p 0 w 2 ) is odd, Φ ( p 0 w 2 ) \Phi(p_0w^2) Φ ( p 0 w 2 ) implies that it is positive, that is, v p 0 ( w ) ≥ 0 v_{p_0}(w)\geq0 v p 0 ( w ) ≥ 0 . Conversely, if v p 0 ( w ) ≥ 0 v_{p_0}(w) \ge0 v p 0 ( w ) ≥ 0 , then p 0 w 2 ∈ D 2 p_0w^2 \in D_2 p 0 w 2 ∈ D 2 . The case w = 0 w = 0 w = 0 is immediate. Finally, if p 0 ≡ 3 , 5 ( m o d 8 ) p_0 \equiv3,5 \pmod{8} p 0 ≡ 3 , 5 ( mod 8 ) , then 2 2 2 is a non-square modulo p 0 p_0 p 0 , and
v p 0 ( w 0 2 − 2 w 1 2 ) = 2 min { v p 0 ( w 0 ) , v p 0 ( w 1 ) } . v_{p_0}(w_0^2 - 2w_1^2) = 2\min\{v_{p_0}(w_0),v_{p_0}(w_1)\}. v p 0 ( w 0 2 − 2 w 1 2 ) = 2 min { v p 0 ( w 0 ) , v p 0 ( w 1 )} .
The last assertion follows from the second. □
For any finite S ′ ⊇ S 1 S' \supseteq S_1 S ′ ⊇ S 1 , a ∈ Q × a \in\mathbb{Q}^{\times} a ∈ Q × and w ∈ Q w \in\mathbb{Q} w ∈ Q ,
Ψ S ′ ( a w 2 ) ⟺ v p ( a w 2 ) ≥ 0 for every p ∉ S ′ with v p ( a ) odd . \Psi_{S'}(aw^2) \Longleftrightarrow v_p(aw^2) \ge0 \quad\text{for every } p \notin S' \text{ with } v_p(a) \text{ odd}. Ψ S ′ ( a w 2 ) ⟺ v p ( a w 2 ) ≥ 0 for every p ∈ / S ′ with v p ( a ) odd .
Indeed, multiplication by w 2 w^2 w 2 preserves the parity of every valuation when w ≠ 0 w \ne0 w = 0 , and the case w = 0 w = 0 w = 0 is immediate.
Remark. For a positive squarefree integer d > 1 d > 1 d > 1 , put
R d = ⋂ p ∣ d Z ( p ) . R_d = \bigcap_{p\mid d}\mathbb{Z}_{(p)}. R d = p ∣ d ⋂ Z ( p ) .
Its maximal ideals are p R d pR_d p R d , for p ∣ d p \mid d p ∣ d , so its Jacobson radical is d R d dR_d d R d . For each fixed integer k ≥ 1 k \ge1 k ≥ 1 , Theorem C supplies Diophantine descriptions of R d R_d R d and the k k k th power d k R d d^kR_d d k R d of its Jacobson radical:
x ∈ R d ⟺ d x 2 ∈ D 2 , x \in R_d \quad\Longleftrightarrow\quad dx^2 \in D_2, x ∈ R d ⟺ d x 2 ∈ D 2 ,
x ∈ d k R d ⟺ x 2 d 2 k − 1 ∈ D 2 . (27) x \in d^kR_d \quad\Longleftrightarrow\quad\frac{x^2}{d^{2k-1}} \in D_2. \tag*{(27)} x ∈ d k R d ⟺ d 2 k − 1 x 2 ∈ D 2 . (27)
For x ≠ 0 x \ne0 x = 0 and p ∣ d p \mid d p ∣ d , the valuations on the right are 1 + 2 v p ( x ) 1+2v_p(x) 1 + 2 v p ( x ) and 2 v p ( x ) − ( 2 k − 1 ) 2v_p(x)-(2k-1) 2 v p ( x ) − ( 2 k − 1 ) , respectively. They are odd, so membership in D 2 D_2 D 2 is equivalent to v p ( x ) ≥ 0 v_p(x) \ge0 v p ( x ) ≥ 0 and v p ( x ) ≥ k v_p(x) \ge k v p ( x ) ≥ k , respectively. At every other prime the valuations are even; the case x = 0 x = 0 x = 0 is immediate. These semilocal conditions are the ones underlying the constructions of Poonen and Koenigsmann [65 Lemma 2.5] , [42 Lemma 13] .
For an odd prime p p p , an integer k k k and w = w 0 + w 1 2 w = w_0 + w_1\sqrt{2} w = w 0 + w 1 2 with w 0 , w 1 ∈ Q w_0,w_1 \in\mathbb{Q} w 0 , w 1 ∈ Q ,
v p ( w 0 ) , v p ( w 1 ) ≥ k ⟺ v p ( w ) ≥ k at every place p ∣ p , (28) v_p(w_0), v_p(w_1) \ge k \quad\Longleftrightarrow\quad v_{\mathfrak p}(w) \ge k \text{ at every place } \mathfrak p \mid p, \tag*{(28)} v p ( w 0 ) , v p ( w 1 ) ≥ k ⟺ v p ( w ) ≥ k at every place p ∣ p , (28)
since w 0 = ( w + w σ ) / 2 w_0 = (w+w^\sigma)/2 w 0 = ( w + w σ ) /2 , w 1 = ( w − w σ ) / ( 2 2 ) w_1 = (w-w^\sigma)/(2\sqrt{2}) w 1 = ( w − w σ ) / ( 2 2 ) , and σ \sigma σ permutes the places above p p p .
Finite sets of bounded height.
Theorem 7.2. There exist a Diophantine set C ⊂ Q × Z C \subset\mathbb{Q} \times\mathbb{Z} C ⊂ Q × Z and H > 0 H > 0 H > 0 with the following properties. Put F a = { q ∈ Q : ( q , ι ( a ) ) ∈ C } F_a = \{q \in\mathbb{Q} : (q,\iota(a)) \in C\} F a = { q ∈ Q : ( q , ι ( a )) ∈ C } for a ∈ Z a \in\mathbb{Z} a ∈ Z .
(1) Every q ∈ F a q \in F_a q ∈ F a satisfies h ( q ) ≤ H ∣ a ∣ h(q) \le H|a| h ( q ) ≤ H ∣ a ∣ .
(2) For every finite subset of Q \mathbb{Q} Q there exists an integer a 0 ≥ 1 a_0 \ge1 a 0 ≥ 1 such that the subset is contained in F a F_a F a for every positive multiple a a a of a 0 a_0 a 0 ; in particular Q = ⋃ n ≥ 1 F n ! \mathbb{Q} = \bigcup_{n\ge1} F_{n!} Q = ⋃ n ≥ 1 F n ! .
Fix an integer C 0 ≥ 1 C_0 \ge1 C 0 ≥ 1 with C 0 > θ j C_0 > \theta_j C 0 > θ j for 0 ≤ j ≤ 4 0 \le j \le4 0 ≤ j ≤ 4 . For q ∈ Q q \in\mathbb{Q} q ∈ Q put
β q = q 2 + C 0 , G q = F 5 ( β q ) = ∏ j = 0 4 ( β q − θ j ) , \beta_q = q^2 + C_0, \qquad G_q = F_5(\beta_q) = \prod_{j=0}^{4}(\beta_q - \theta_j), β q = q 2 + C 0 , G q = F 5 ( β q ) = j = 0 ∏ 4 ( β q − θ j ) ,
so that G q > 0 G_q > 0 G q > 0 .
Definition 7.3. For q ∈ Q q \in\mathbb{Q} q ∈ Q and A ∈ ⟨ P ⟩ ∖ { O } A \in\langle P\rangle\setminus\{O\} A ∈ ⟨ P ⟩ ∖ { O } , let C ( q , A ) C(q,A) C ( q , A ) hold if there exists J ∈ ⟨ P ⟩ ∖ { O } J \in\langle P\rangle\setminus\{O\} J ∈ ⟨ P ⟩ ∖ { O } such that
Ψ S ( G q Ω 2 ) , Ω = ∏ i = 1 4 ( 1 + w i 2 G q 4 c − 1 ) , \Psi_S(G_q\Omega^2), \qquad\Omega= \prod_{i=1}^{4}\left(1+\frac{w_i^2}{G_q^{4c-1}}\right), Ψ S ( G q Ω 2 ) , Ω = i = 1 ∏ 4 ( 1 + G q 4 c − 1 w i 2 ) , where w 1 , … , w 4 w_1,\ldots,w_4 w 1 , … , w 4 are the four rational coefficients, in the basis 1 , 2 1,\sqrt{2} 1 , 2 , of ρ ( J ) \rho(J) ρ ( J ) and ρ ( A ) / ρ ( J ) \rho(A)/\rho(J) ρ ( A ) / ρ ( J ) .
This is a positive existential condition on q q q and the coordinates of A A A , and Ω > 0 \Omega> 0 Ω > 0 since G q > 0 G_q > 0 G q > 0 . Let p ∉ S p \notin S p ∈ / S . If v p ( q ) < 0 v_p(q) < 0 v p ( q ) < 0 , each factor β q − θ j \beta_q - \theta_j β q − θ j has valuation 2 v p ( q ) 2v_p(q) 2 v p ( q ) . If v p ( q ) ≥ 0 v_p(q) \ge0 v p ( q ) ≥ 0 , the factors are p p p -integral and, the θ j \theta_j θ j being pairwise incongruent modulo p p p , at most one of them has positive valuation. Hence
v p ( G q ) is odd ⟺ p is a prime of odd contact of β q , (29) v_p(G_q)\text{ is odd} \Longleftrightarrow p\text{ is a prime of odd contact of }\beta_q, \tag*{(29)} v p ( G q ) is odd ⟺ p is a prime of odd contact of β q , (29)
and it is then positive. For a witness in Definition 7.3 , put m = v p ( G q ) m = v_p(G_q) m = v p ( G q ) at such a prime. For each non-zero coefficient w i w_i w i , the integer 2 v p ( w i ) − ( 4 c − 1 ) m 2v_p(w_i) - (4c - 1)m 2 v p ( w i ) − ( 4 c − 1 ) m is odd, so
v p ( Ω ) = ∑ w i ≠ 0 min { 0 , 2 v p ( w i ) − ( 4 c − 1 ) m } . v_p(\Omega) = \sum_{w_i \ne0} \min\{0, 2v_p(w_i) - (4c - 1)m\}. v p ( Ω ) = w i = 0 ∑ min { 0 , 2 v p ( w i ) − ( 4 c − 1 ) m } .
If some v p ( w i ) ≤ 2 c − 1 v_p(w_i) \le2c - 1 v p ( w i ) ≤ 2 c − 1 , then
v p ( G q Ω 2 ) ≤ 8 c − 4 − ( 8 c − 3 ) m ≤ − 1 , v_p(G_q\Omega^2) \le8c - 4 - (8c - 3)m \le-1, v p ( G q Ω 2 ) ≤ 8 c − 4 − ( 8 c − 3 ) m ≤ − 1 ,
which is negative and odd, contrary to Lemma 7.1 . Thus all four coefficients have valuation at least 2 c 2c 2 c ; by (28) , the same is true of ρ ( J ) \rho(J) ρ ( J ) and ρ ( A ) / ρ ( J ) \rho(A)/\rho(J) ρ ( A ) / ρ ( J ) at every place above p p p . At the other primes outside S S S , v p ( G q Ω 2 ) v_p(G_q\Omega^2) v p ( G q Ω 2 ) is even.
Lemma 7.4. If C ( q , a P ) C(q,aP) C ( q , a P ) holds, then h ( q ) ≤ 1 2 H S ∣ a ∣ h(q) \le\frac{1}{2}H_S|a| h ( q ) ≤ 2 1 H S ∣ a ∣ . For every q ∈ Q q \in\mathbb{Q} q ∈ Q there exists an integer a 0 ≥ 1 a_0 \ge1 a 0 ≥ 1 such that C ( q , a P ) C(q,aP) C ( q , a P ) holds for every positive multiple a a a of a 0 a_0 a 0 .
Proof. Let J = j P J = jP J = j P be a witness for C ( q , a P ) C(q,aP) C ( q , a P ) , and let p ∉ S p \notin S p ∈ / S be a prime of odd contact of β q \beta_q β q . The preceding valuation bounds imply that J J J and a P aP a P reduce to O O O at the places p ∣ p \mathfrak{p} \mid p p ∣ p , and by (10) and Proposition 2.3 (1),
2 v p ( a / j ) = v p ( ρ ( a P ) ρ ( J ) ) ≥ 2 c . 2v_p(a/j) = v_{\mathfrak{p}}\left(\frac{\rho(aP)}{\rho(J)}\right) \ge2c. 2 v p ( a / j ) = v p ( ρ ( J ) ρ ( a P ) ) ≥ 2 c .
So v p ( a / j ) ≥ c v_p(a/j) \ge c v p ( a / j ) ≥ c , and p c ∣ a p^c \mid a p c ∣ a as j j j is an integer. Therefore M S ( β q ) c ∣ a M_S(\beta_q)^c \mid a M S ( β q ) c ∣ a , and h ( β q ) ≤ H S M S ( β q ) c ≤ H S ∣ a ∣ h(\beta_q) \le H_SM_S(\beta_q)^c \le H_S|a| h ( β q ) ≤ H S M S ( β q ) c ≤ H S ∣ a ∣ by Theorem E. Writing q = b / e q = b/e q = b / e in lowest terms with e > 0 e > 0 e > 0 , we have β q = ( b 2 + C 0 e 2 ) / e 2 \beta_q = (b^2 + C_0e^2)/e^2 β q = ( b 2 + C 0 e 2 ) / e 2 in lowest terms, with numerator at least max ( b 2 , e 2 ) \max(b^2,e^2) max ( b 2 , e 2 ) ; so h ( β q ) ≥ 2 h ( q ) h(\beta_q) \ge2h(q) h ( β q ) ≥ 2 h ( q ) .
For the second assertion, let P q \mathcal{P}_q P q be the finite set of primes p ∉ S p \notin S p ∈ / S of odd contact of β q \beta_q β q , and put
d p = 1 2 ( ( 4 c − 1 ) v p ( G q ) + 1 ) ( p ∈ P q ) . d_p = \frac{1}{2}\bigl((4c - 1)v_p(G_q) + 1\bigr) \qquad(p \in\mathcal{P}_q). d p = 2 1 ( ( 4 c − 1 ) v p ( G q ) + 1 ) ( p ∈ P q ) .
Let o o o be the product of the orders of the reductions of P P P at the places above the primes of P q \mathcal{P}_q P q , choose j ≥ 1 j \ge1 j ≥ 1 divisible by o ∏ p ∈ P q p d p o\prod_{p\in\mathcal{P}_q}p^{d_p} o ∏ p ∈ P q p d p , and choose M ≥ 1 M \ge1 M ≥ 1 divisible by ∏ p ∈ P q p d p \prod_{p\in\mathcal{P}_q}p^{d_p} ∏ p ∈ P q p d p . Put J = j P J = jP J = j P and A = M J A = MJ A = M J . At p ∣ p ∈ P q \mathfrak{p} \mid p \in\mathcal{P}_q p ∣ p ∈ P q , the reduction order o p o_{\mathfrak{p}} o p divides o o o , so v p ( j / o p ) ≥ d p v_p(j/o_{\mathfrak{p}}) \ge d_p v p ( j / o p ) ≥ d p . Proposition 2.3 (1), applied first to j P , o p P jP,o_{\mathfrak{p}}P j P , o p P and then to M J , J MJ,J M J , J , yields v p ( z ( J ) ) ≥ d p + 1 v_{\mathfrak{p}}(z(J)) \ge d_p + 1 v p ( z ( J )) ≥ d p + 1 and v p ( z ( A ) / z ( J ) ) ≥ d p v_{\mathfrak{p}}(z(A)/z(J)) \ge d_p v p ( z ( A ) / z ( J )) ≥ d p at every such place. By (10) and (28) , each coefficient w i w_i w i has valuation at least 2 d p 2d_p 2 d p , and hence 2 v p ( w i ) > ( 4 c − 1 ) v p ( G q ) 2v_p(w_i) > (4c - 1)v_p(G_q) 2 v p ( w i ) > ( 4 c − 1 ) v p ( G q ) . Thus each factor of Ω \Omega Ω is a p p p -adic unit. It follows that v p ( G q Ω 2 ) = v p ( G q ) > 0 v_p(G_q\Omega^2) = v_p(G_q) > 0 v p ( G q Ω 2 ) = v p ( G q ) > 0 for p ∈ P q p \in\mathcal{P}_q p ∈ P q , while the valuations outside S ∪ P q S \cup\mathcal{P}_q S ∪ P q are even. Hence C ( q , A ) C(q,A) C ( q , A ) holds. The same holds with M M M replaced by any positive multiple, so a 0 = j M a_0 = jM a 0 = j M has the required property. □
Proof of Theorem 7.2 . Let C C C be the set of pairs ( q , ι ( a ) ) (q,\iota(a)) ( q , ι ( a )) with a ≠ 0 a \ne0 a = 0 and C ( q , a P ) C(q,aP) C ( q , a P ) . Quantifying a point A ~ ∈ E ( F ′ ) \widetilde{A} \in E(F') A ∈ E ( F ′ ) and putting A = 2 A ~ A = 2\widetilde{A} A = 2 A , the conditions A ≠ O A \ne O A = O , z ( A ) = y 0 + y 1 2 z(A) = y_0 + y_1\sqrt{2} z ( A ) = y 0 + y 1 2 and C ( q , A ) C(q,A) C ( q , A ) define the set of ( q , y 0 , y 1 ) (q,y_0,y_1) ( q , y 0 , y 1 ) in question by a positive existential formula, as in §8.5; by Lemma 2.4 , ( y 0 , y 1 ) = ι ( a ) (y_0,y_1) = \iota(a) ( y 0 , y 1 ) = ι ( a ) determines A = a P A = aP A = a P . Assertion (1), with H = 1 2 H S H = \frac{1}{2}H_S H = 2 1 H S , and assertion (2) follow from Lemma 7.4, taking for a 0 a_0 a 0 the least common multiple of the integers that the lemma provides for the elements of the subset. □
From indices to integers
Fix a prime p 0 ∉ S p_0 \notin S p 0 ∈ / S with p 0 ≡ 3 , 5 ( m o d 8 ) p_0 \equiv3,5 \pmod{8} p 0 ≡ 3 , 5 ( mod 8 ) , so that p 0 p_0 p 0 is inert in F F F . Choose an integer D 0 ≥ 1 D_0 \ge1 D 0 ≥ 1 such that D 0 P D_0P D 0 P reduces to O O O at the place above p 0 p_0 p 0 , and an integer L ≥ 1 L \ge1 L ≥ 1 with L log p 0 > 1 2 H S + 1 L\log p_0 > \frac{1}{2}H_S + 1 L log p 0 > 2 1 H S + 1 . Let R ⊆ Z 4 \mathcal{R} \subseteq\mathbb{Z}^4 R ⊆ Z 4 be the recursive set of ( n , a , u , v ) (n,a,u,v) ( n , a , u , v ) with
a ≥ 1 , u = p 0 L ( a + n 2 + 1 ) , v = n u . a \ge1,\qquad u = p_0^{L(a+n^2+1)},\qquad v = nu. a ≥ 1 , u = p 0 L ( a + n 2 + 1 ) , v = n u .
Lemma 6.5 expresses this relation on the encoded indices by polynomial equations with rational auxiliary variables.
Theorem 7.5. The set { ( n , ι ( n ) ) : n ∈ Z } ⊂ Q 3 \{(n,\iota(n)) : n \in\mathbb{Z}\} \subset\mathbb{Q}^3 {( n , ι ( n )) : n ∈ Z } ⊂ Q 3 is Diophantine.
Proof. For q ∈ Q q \in\mathbb{Q} q ∈ Q and y ∈ Q 2 y \in\mathbb{Q}^2 y ∈ Q 2 , let I ( q , y ) \mathcal{I}(q,y) I ( q , y ) hold if there exist y A , y U , y V y_A,y_U,y_V y A , y U , y V with ( y , y A , y U , y V ) (y,y_A,y_U,y_V) ( y , y A , y U , y V ) in the set attached to R \mathcal{R} R by Lemma 6.5 , such that the points A , U , V ∈ ⟨ P ⟩ A,U,V \in\langle P\rangle A , U , V ∈ ⟨ P ⟩ encoded by y A , y U , y V y_A,y_U,y_V y A , y U , y V satisfy C ( q , A ) C(q,A) C ( q , A ) , and the rational coefficients t 0 , t 1 t_0,t_1 t 0 , t 1 of
T = z ( Z ) − q z ( Y ) z ( Y ) 2 , Y = D 0 U , Z = D 0 V , T = \frac{z(Z)-qz(Y)}{z(Y)^2},\qquad Y=D_0U,\qquad Z=D_0V, T = z ( Y ) 2 z ( Z ) − q z ( Y ) , Y = D 0 U , Z = D 0 V , satisfy Φ ( p 0 ( t 0 2 − 2 t 1 2 ) 2 ) \Phi\left(p_0(t_0^2-2t_1^2)^2\right) Φ ( p 0 ( t 0 2 − 2 t 1 2 ) 2 ) . By Lemmas 2.4 and 6.5 , I \mathcal{I} I is Diophantine; note Y ≠ O Y \ne O Y = O , as u ≥ 1 u \ge1 u ≥ 1 .
Suppose that I ( q , ι ( n ) ) \mathcal{I}(q,\iota(n)) I ( q , ι ( n )) holds. Write A = a P A=aP A = a P , U = u P U=uP U = u P and V = v P V=vP V = v P , so that ( n , a , u , v ) ∈ R (n,a,u,v) \in\mathcal{R} ( n , a , u , v ) ∈ R , and put k = L ( a + n 2 + 1 ) k=L(a+n^2+1) k = L ( a + n 2 + 1 ) . By Lemma 7.1 , t 0 , t 1 ∈ Z ( p 0 ) t_0,t_1 \in\mathbb{Z}_{(p_0)} t 0 , t 1 ∈ Z ( p 0 ) , so T T T is integral at the place p ∣ p 0 \mathfrak{p}\mid p_0 p ∣ p 0 . We have Y = p 0 k ⋅ D 0 P Y=p_0^k\cdot D_0P Y = p 0 k ⋅ D 0 P and Z = n Y Z=nY Z = nY , and Proposition 2.3 (1) gives v p ( z ( Y ) ) ≥ k + 1 v_{\mathfrak{p}}(z(Y)) \ge k+1 v p ( z ( Y )) ≥ k + 1 . Multiplication by n n n on the formal group has the form [ n ] ( X ) = n X + X 2 g n ( X ) [n](X)=nX+X^2g_n(X) [ n ] ( X ) = n X + X 2 g n ( X ) with g n ∈ O F , p [ [ X ] ] g_n\in\mathcal{O}_{F,\mathfrak{p}}[[X]] g n ∈ O F , p [[ X ]] [82 IV.2] , so ( z ( Z ) − n z ( Y ) ) / z ( Y ) 2 = g n ( z ( Y ) ) (z(Z)-nz(Y))/z(Y)^2=g_n(z(Y)) ( z ( Z ) − n z ( Y )) / z ( Y ) 2 = g n ( z ( Y )) is integral at p \mathfrak{p} p . Subtracting it from T T T shows that ( n − q ) / z ( Y ) (n-q)/z(Y) ( n − q ) / z ( Y ) is integral at p \mathfrak{p} p , so
v p 0 ( q − n ) ≥ k + 1. v_{p_0}(q-n) \ge k+1. v p 0 ( q − n ) ≥ k + 1. Suppose q ≠ n q \ne n q = n , and write q = b / e q=b/e q = b / e in lowest terms with e > 0 e>0 e > 0 . Since n ∈ Z n\in\mathbb{Z} n ∈ Z , the preceding congruence implies q ∈ Z ( p 0 ) q\in\mathbb{Z}_{(p_0)} q ∈ Z ( p 0 ) , so p 0 ∤ e p_0\nmid e p 0 ∤ e . By Lemma 7.4 and log ( ∣ n ∣ + 1 ) ≤ n 2 \log(|n|+1)\le n^2 log ( ∣ n ∣ + 1 ) ≤ n 2 ,
log ( ∣ n ∣ + 1 ) + h ( q ) ≤ n 2 + 1 2 H S a < k log p 0 , \log(|n|+1)+h(q)\le n^2+\frac{1}{2}H_Sa<k\log p_0, log ( ∣ n ∣ + 1 ) + h ( q ) ≤ n 2 + 2 1 H S a < k log p 0 , where the final inequality follows from the choice of L L L . As exp ( h ( q ) ) = max ( ∣ b ∣ , e ) \exp(h(q))=\max(|b|,e) exp ( h ( q )) = max ( ∣ b ∣ , e ) and the non-zero integer b − n e b-ne b − n e is divisible by p 0 k + 1 p_0^{k+1} p 0 k + 1 ,
p 0 k + 1 ≤ ∣ b − n e ∣ ≤ ( ∣ n ∣ + 1 ) exp ( h ( q ) ) < p 0 k , p_0^{k+1}\le|b-ne|\le(|n|+1)\exp(h(q))<p_0^k, p 0 k + 1 ≤ ∣ b − n e ∣ ≤ ( ∣ n ∣ + 1 ) exp ( h ( q )) < p 0 k , a contradiction. Hence q = n q=n q = n .
Conversely, let q = n ∈ Z q=n\in\mathbb{Z} q = n ∈ Z . By Lemma 7.4 there exists a ≥ 1 a\ge1 a ≥ 1 with C ( n , a P ) C(n,aP) C ( n , a P ) . Put u = p 0 L ( a + n 2 + 1 ) u=p_0^{L(a+n^2+1)} u = p 0 L ( a + n 2 + 1 ) and v = n u v=nu v = n u ; Lemma 6.5 supplies the remaining witnesses. Then T = g n ( z ( Y ) ) T=g_n(z(Y)) T = g n ( z ( Y )) is integral at the place above p 0 p_0 p 0 , and so are its rational coefficients by (28) . Lemma 7.1 verifies Φ ( p 0 ( t 0 2 − 2 t 1 2 ) 2 ) \Phi\left(p_0(t_0^2-2t_1^2)^2\right) Φ ( p 0 ( t 0 2 − 2 t 1 2 ) 2 ) . □ \square □
Proofs of Theorem A and Corollary 1.1
Proof of Theorem A. The set Z \mathbb{Z} Z is the projection of the set of Theorem 7.5 to the first coordinate. □ \square □
A positive existential formula φ ( x ) \varphi(x) φ ( x ) over Q \mathbb{Q} Q is equivalent to a formula ∃ z ( q 1 = ⋯ = q r = 0 ) \exists z\,(q_1=\cdots=q_r=0) ∃ z ( q 1 = ⋯ = q r = 0 ) with quadratic q l ∈ Q [ x , z ] q_l\in\mathbb{Q}[x,z] q l ∈ Q [ x , z ] . Indeed, the bound variables may be taken distinct and moved to the front; a conjunction a 1 = ⋯ = a k = 0 a_1=\cdots=a_k=0 a 1 = ⋯ = a k = 0 is equivalent over Q \mathbb{Q} Q to ∑ i a i 2 = 0 \sum_i a_i^2=0 ∑ i a i 2 = 0 , and a disjunction of a = 0 a=0 a = 0 and a ′ = 0 a'=0 a ′ = 0 to a a ′ = 0 aa'=0 a a ′ = 0 . An equation ∑ β a β V β = 0 \sum_\beta a_\beta V^\beta=0 ∑ β a β V β = 0 is equivalent to the system
Z 0 = 1 , Z β = Z β − ϵ i ( β ) V i ( β ) ( β ≠ 0 ) , ∑ β a β Z β = 0 , Z_0=1,\qquad Z_\beta=Z_{\beta-\epsilon_i(\beta)}V_{i(\beta)}\quad(\beta\ne0),\qquad\sum_\beta a_\beta Z_\beta=0, Z 0 = 1 , Z β = Z β − ϵ i ( β ) V i ( β ) ( β = 0 ) , β ∑ a β Z β = 0 ,
in new variables Z β Z_{\beta} Z β indexed by multi-indices up to the degree of the polynomial, where i ( β ) i(\beta) i ( β ) is the least index with β i > 0 \beta_i > 0 β i > 0 . Clearing denominators, we fix quadratic q l ∈ Z [ x , z ] q_l \in\mathbb{Z}[x,z] q l ∈ Z [ x , z ] , 1 ≤ l ≤ r 1 \le l \le r 1 ≤ l ≤ r , such that ∃ z ( q 1 ( x , z ) = ⋯ = q r ( x , z ) = 0 ) \exists z\ (q_1(x,z)=\cdots=q_r(x,z)=0) ∃ z ( q 1 ( x , z ) = ⋯ = q r ( x , z ) = 0 ) defines Z \mathbb{Z} Z .
Proof of Corollary 1.1. (1) A Diophantine set is recursively enumerable. Conversely, let B ⊆ Q r B \subseteq\mathbb{Q}^r B ⊆ Q r be recursively enumerable. Let B ′ ⊆ Z 2 r B' \subseteq\mathbb{Z}^{2r} B ′ ⊆ Z 2 r be the set of ( a 1 , b 1 , … , a r , b r ) (a_1,b_1,\ldots,a_r,b_r) ( a 1 , b 1 , … , a r , b r ) with b i ≠ 0 b_i \ne0 b i = 0 and ( a 1 / b 1 , … , a r / b r ) ∈ B (a_1/b_1,\ldots,a_r/b_r) \in B ( a 1 / b 1 , … , a r / b r ) ∈ B . It is recursively enumerable, hence of the form { x : ∃ y ∈ Z s , g ( x , y ) = 0 } \{x:\exists y\in\mathbb{Z}^s,\ g(x,y)=0\} { x : ∃ y ∈ Z s , g ( x , y ) = 0 } for an integer polynomial g g g [47 , 13 ] . Then x ∈ Q r x\in\mathbb{Q}^r x ∈ Q r lies in B B B if and only if there are rational a i , b i , y a_i,b_i,y a i , b i , y , all subject to the definition of Z \mathbb{Z} Z , with b i ≠ 0 b_i\ne0 b i = 0 , b i x i = a i b_ix_i=a_i b i x i = a i and g ( a , b , y ) = 0 g(a,b,y)=0 g ( a , b , y ) = 0 .
(2) Take P = ∑ l q l 2 P=\sum_l q_l^2 P = ∑ l q l 2 .
(3) Let K 0 ⊆ Z ≥ 0 K_0\subseteq\mathbb{Z}_{\geq0} K 0 ⊆ Z ≥ 0 be recursively enumerable and not computable. By (1) and the reduction above, there is a quartic P 0 ∈ Z [ T , W 1 , … , W N ] P_0\in\mathbb{Z}[T,W_1,\ldots,W_N] P 0 ∈ Z [ T , W 1 , … , W N ] such that a ∈ K 0 a\in K_0 a ∈ K 0 if and only if P 0 ( a , W ) P_0(a,W) P 0 ( a , W ) has a rational zero. As a ↦ P 0 ( a , W ) a\mapsto P_0(a,W) a ↦ P 0 ( a , W ) is computable, rational solvability of quartics in N N N variables is undecidable.
(4) This is the implication of [42 Proposition 23(d)] ; we recall the argument. For each m ≥ 1 m\geq1 m ≥ 1 , the set of tuples ( a 0 , … , a m − 1 ) ∈ Q m (a_0,\ldots,a_{m-1})\in\mathbb{Q}^m ( a 0 , … , a m − 1 ) ∈ Q m for which X m + a m − 1 X m − 1 + ⋯ + a 0 X^m+a_{m-1}X^{m-1}+\cdots+a_0 X m + a m − 1 X m − 1 + ⋯ + a 0 has no rational root is recursive, hence Diophantine by (1). Its existential definition is equivalent to having no root in every field elementarily equivalent to Q \mathbb{Q} Q . If an element of L ∖ K L\setminus K L ∖ K were algebraic over K K K , its minimal polynomial would have no root in K K K . The existential definition would hold for its coefficients in K K K , and hence in L L L , contradicting the existence of that root.
(5) By [66 Theorem 1.1(i)] , an algorithm deciding rational-point existence for smooth projective geometrically integral varieties over Q \mathbb{Q} Q would decide it for arbitrary varieties over Q \mathbb{Q} Q , contrary to (3). □
Corollary 7.6. There exist an integer C ≥ 1 C\geq1 C ≥ 1 and an algorithm with the following property. Given ν , D ≥ 1 \nu,D\geq1 ν , D ≥ 1 and f ∈ Z [ X 1 , … , X ν ] f\in\mathbb{Z}[X_1,\ldots,X_\nu] f ∈ Z [ X 1 , … , X ν ] of total degree at most D D D , it produces F f ∈ Z [ Y 1 , … , Y N ] F_f\in\mathbb{Z}[Y_1,\ldots,Y_N] F f ∈ Z [ Y 1 , … , Y N ] of total degree at most 4 4 4 , with N ≤ C ( ν + D D ) N\leq C\binom{\nu+D}{D} N ≤ C ( D ν + D ) , such that f f f has a zero in Z ν \mathbb{Z}^\nu Z ν if and only if F f F_f F f has a zero in Q N \mathbb{Q}^N Q N .
Proof. For f = ∑ ∣ α ∣ ≤ D c α X α f=\sum_{|\alpha|\leq D}c_\alpha X^\alpha f = ∑ ∣ α ∣ ≤ D c α X α , take ν \nu ν copies of the system q 1 = ⋯ = q r = 0 q_1=\cdots=q_r=0 q 1 = ⋯ = q r = 0 , one for each X i X_i X i with fresh variables z z z , the equations Z 0 = 1 Z_0=1 Z 0 = 1 and Z β = Z β − e i ( β ) X i ( β ) Z_\beta=Z_{\beta-e_{i(\beta)}}X_{i(\beta)} Z β = Z β − e i ( β ) X i ( β ) for 0 < ∣ β ∣ ≤ D 0<|\beta|\leq D 0 < ∣ β ∣ ≤ D , and ∑ α c α Z α = 0 \sum_\alpha c_\alpha Z_\alpha=0 ∑ α c α Z α = 0 . These are quadrics in at most ν ( 1 + dim z ) + ( ν + D D ) \nu(1+\dim z)+\binom{\nu+D}{D} ν ( 1 + dim z ) + ( D ν + D ) variables, with a common rational zero if and only if f f f has a zero in Z ν \mathbb{Z}^\nu Z ν . Let F f F_f F f be the sum of their squares. □
Remark 7.8. The argument for Theorem 7.5 depends on two properties of C C C : every q ∈ Q q \in\mathbb{Q} q ∈ Q satisfies C ( q , a P ) C(q,aP) C ( q , a P ) for some a ≥ 1 a \ge1 a ≥ 1 , and C ( q , a P ) C(q,aP) C ( q , a P ) implies h ( q ) ≤ 1 2 H S ∣ a ∣ h(q) \le\frac{1}{2}H_S|a| h ( q ) ≤ 2 1 H S ∣ a ∣ . The linear bound may be replaced by any recursive one. Let C ′ ⊆ Q × Z C' \subseteq\mathbb{Q} \times\mathbb{Z} C ′ ⊆ Q × Z be Diophantine, such that every q ∈ Q q \in\mathbb{Q} q ∈ Q has ( q , ι ( a ) ) ∈ C ′ (q,\iota(a)) \in C' ( q , ι ( a )) ∈ C ′ for some a ≥ 1 a \ge1 a ≥ 1 , and, for a ≥ 1 a \ge1 a ≥ 1 , ( q , ι ( a ) ) ∈ C ′ (q,\iota(a)) \in C' ( q , ι ( a )) ∈ C ′ implies h ( q ) ≤ B ( a ) h(q) \le B(a) h ( q ) ≤ B ( a ) for a recursive function B : Z ≥ 1 → Z ≥ 0 B \colon\mathbb{Z}_{\ge1} \to\mathbb{Z}_{\ge0} B : Z ≥ 1 → Z ≥ 0 . Put u = p 0 k u=p_0^k u = p 0 k with k = B ( a ) + n 2 + 2 k=B(a)+n^2+2 k = B ( a ) + n 2 + 2 in the definition of R \mathcal{R} R . Then log ( ∣ n ∣ + 1 ) + h ( q ) < k log p 0 \log(|n|+1)+h(q)<k\log p_0 log ( ∣ n ∣ + 1 ) + h ( q ) < k log p 0 , as log p 0 > 1 \log p_0>1 log p 0 > 1 , and the same proof applies to C ′ C' C ′ .
Quaternionic surfaces of controlled level
In this section we prove Theorems D and E. The Shimura curve of discriminant 210, its quotient by the Atkin–Lehner involutions and its family of abelian surfaces are standard (see §§8.1–8.2). The controlled splitting of the descent obstruction and the ramification estimates establish the level bound in § 11.3 . The passage from the level bound to the height estimate follows von Känel’s argument [89 §5.2] , with Rémond’s inequality for abelian subvarieties (see § 11.5 ).
Fix an embedding Q ‾ ↪ C \overline{\mathbb{Q}} \hookrightarrow\mathbb{C} Q ↪ C for the complex points and complex conjugation. Throughout, C C C denotes positive constants that depend only on the data fixed in §§8.1–8.2, and not on t t t ; their values may change from line to line. For an abelian variety A A A over a number field, h F ( A ) h_F(A) h F ( A ) is its stable Faltings height, in the normalization of [27 §1] , from which the other usual normalizations differ by a constant multiple of dim A \dim A dim A ; V ℓ ( A ) V_\ell(A) V ℓ ( A ) is its rational ℓ \ell ℓ -adic Tate module; and A ∼ K A ′ A \sim_K A' A ∼ K A ′ means that A A A and A ′ A' A ′ are isogenous over K K K . For a prime p p p , G Q p u G^u_{\mathbb{Q}_p} G Q p u (u ≥ 0 u \ge0 u ≥ 0 ) are the upper ramification groups [77 IV] .
The quotient by the Atkin–Lehner involutions
Let B \mathbb{B} B , O B \mathcal{O}_{\mathbb{B}} O B and X X X be as in § 2 , and fix B ⊗ R ≃ M 2 ( R ) \mathbb{B}\otimes\mathbb{R}\simeq M_2(\mathbb{R}) B ⊗ R ≃ M 2 ( R ) . The group O B 1 \mathcal{O}_{\mathbb{B}}^1 O B 1 of elements of reduced norm one acts on the upper half plane H \mathfrak{H} H , and Γ = O B 1 / { ± 1 } \Gamma=\mathcal{O}_{\mathbb{B}}^1/\{\pm1\} Γ = O B 1 / { ± 1 } acts freely. Indeed, a non-central element of finite order of O B 1 \mathcal{O}_{\mathbb{B}}^1 O B 1 generates Q ( i ) \mathbb{Q}(i) Q ( i ) or Q ( − 3 ) \mathbb{Q}(\sqrt{-3}) Q ( − 3 ) , and these fields do not embed in B \mathbb{B} B , as 5 splits in the first and 7 in the second [91 Ch. 14] . The curve X X X is the canonical model over Q \mathbb{Q} Q of the compact Riemann surface Γ \ H \Gamma\backslash\mathfrak{H} Γ\ H [14 ] . As Γ \Gamma Γ has no elliptic points, the genus formula [91 Theorem 39.4.20] gives
2 g ( X ) − 2 = 1 6 ∏ p ∣ 210 ( p − 1 ) = 8 , g ( X ) = 5. 2g(X)-2=\frac{1}{6}\prod_{p\mid210}(p-1)=8,\qquad g(X)=5. 2 g ( X ) − 2 = 6 1 p ∣ 210 ∏ ( p − 1 ) = 8 , g ( X ) = 5.
The normalizer of O B \mathcal{O}_{\mathbb{B}} O B in B × \mathbb{B}^{\times} B × , modulo Q × O B × \mathbb{Q}^{\times}\mathcal{O}_{\mathbb{B}}^{\times} Q × O B × , is the group W ≃ ( Z / 2 Z ) 4 W\simeq(\mathbb{Z}/2\mathbb{Z})^4 W ≃ ( Z /2 Z ) 4 of Atkin–Lehner involutions; the involution w m w_m w m , for m ∣ 210 m\mid210 m ∣ 210 , is represented by the elements of O B \mathcal{O}_{\mathbb{B}} O B of reduced norm m m m [91 Ch. 28] . It acts faithfully on X X X by automorphisms defined over Q \mathbb{Q} Q , and the fixed points of w m w_m w m are CM points [58 ] .
Lemma 8.1. The quotient X / W X/W X / W is isomorphic to P 1 \mathbb{P}^1 P 1 over Q \mathbb{Q} Q . The map X → X / W X\to X/W X → X / W is branched over exactly five points, all rational, with ramification index two at every point above them.
Proof. A non-trivial stabilizer in W W W of a point of X X X acts faithfully on the tangent line, so it has order two. As W W W is abelian, the points above a branch value have a common stabilizer { 1 , w m } \{1,w_m\} { 1 , w m } ; we call m m m the label of the branch value. Each branch value has eight points above it, so with r r r branch values the Riemann–Hurwitz formula reads 8 = 16 ( 2 g ( X / W ) − 2 ) + 8 r 8=16(2g(X/W)-2)+8r 8 = 16 ( 2 g ( X / W ) − 2 ) + 8 r , and r = 5 − 4 g ( X / W ) ≤ 5 r=5-4g(X/W)\le5 r = 5 − 4 g ( X / W ) ≤ 5 . For m = 30 , 42 , 70 , 105 , 210 m=30,42,70,105,210 m = 30 , 42 , 70 , 105 , 210 , every prime dividing 210 is ramified or inert in Q ( − m ) \mathbb{Q}(\sqrt{-m}) Q ( − m ) (the primes dividing m m m ramify, and for m = 30 , 42 , 70 m=30,42,70 m = 30 , 42 , 70 the remaining prime 7,5,3 is inert), so Q ( − m ) \mathbb{Q}(\sqrt{-m}) Q ( − m ) embeds in B \mathbb{B} B [91 Ch. 14] , and as all maximal orders are conjugate [91 Ch. 28] , O B \mathcal{O}_{\mathbb{B}} O B contains an element α \alpha α with α 2 = − m \alpha^2=-m α 2 = − m . It normalizes O B \mathcal{O}_{\mathbb{B}} O B , being a unit at the primes not dividing 210 and lying in the unique maximal order of B ⊗ Q p \mathbb{B}\otimes\mathbb{Q}_p B ⊗ Q p for p ∣ 210 p\mid210 p ∣ 210 , and it is elliptic; so w m w_m w m has a fixed point on X X X . Hence r = 5 r=5 r = 5 and g ( X / W ) = 0 g(X/W)=0 g ( X / W ) = 0 . As each w m w_m w m is defined over Q \mathbb{Q} Q , G Q G_{\mathbb{Q}} G Q permutes the branch values preserving labels, so each branch value is rational, and X / W ≃ P 1 X/W \simeq\mathbb{P}^{1} X / W ≃ P 1 over Q \mathbb{Q} Q . □
The image in PSL 2 ( R ) \operatorname{PSL}_{2}(\mathbb{R}) PSL 2 ( R ) of the elements of positive reduced norm in the normalizer of O B \mathcal{O}_{B} O B in B × B^{\times} B × is a Fuchsian group of signature ( 0 ; 2 , 2 , 2 , 2 , 2 ) (0;2,2,2,2,2) ( 0 ; 2 , 2 , 2 , 2 , 2 ) [46 Table 3] . We fix the coordinate on X / W X/W X / W of (1) , and write ϖ : X → P 1 \varpi:X\to\mathbb{P}^{1} ϖ : X → P 1 for the quotient map.
Remark. The values in (1) can be recovered from the known genus-one quotients. For m ∈ { 30 , 42 , 70 , 105 , 210 } m\in\{30,42,70,105,210\} m ∈ { 30 , 42 , 70 , 105 , 210 } , let P m P_m P m be the branch point labelled by w m w_m w m and put X m = X / ⟨ w m ⟩ X_m=X/\langle w_m\rangle X m = X / ⟨ w m ⟩ . Over Q \mathbb{Q} Q , the translation subgroup of W / ⟨ w m ⟩ W/\langle w_m\rangle W / ⟨ w m ⟩ on X m X_m X m is the full two-torsion of Jac ( X m ) \operatorname{Jac}(X_m) Jac ( X m ) . Quotienting by it preserves the Jacobian’s j j j -invariant and leaves a double cover of X / W X/W X / W branched at the other four P m ′ P_{m'} P m ′ .
Put
j ( λ ) = 256 ( 1 − λ + λ 2 ) 3 λ 2 ( 1 − λ ) 2 . j(\lambda)=256\frac{(1-\lambda+\lambda^{2})^{3}}{\lambda^{2}(1-\lambda)^{2}}. j ( λ ) = 256 λ 2 ( 1 − λ ) 2 ( 1 − λ + λ 2 ) 3 .
The equations in [26 Theorem 4.2, Table 2] and [74 p. 239] imply j ( Jac ( X m ) ) = j ( λ m ) j(\operatorname{Jac}(X_m))=j(\lambda_m) j ( Jac ( X m )) = j ( λ m ) , where
Table 1.
Normalize a coordinate s s s by s ( P 42 ) = ∞ s(P_{42})=\infty s ( P 42 ) = ∞ , s ( P 105 ) = 0 s(P_{105})=0 s ( P 105 ) = 0 and s ( P 30 ) = 1 s(P_{30})=1 s ( P 30 ) = 1 , and put a = s ( P 70 ) a=s(P_{70}) a = s ( P 70 ) , b = s ( P 210 ) b=s(P_{210}) b = s ( P 210 ) . The complementary four-point sets give
j ( a ) = j ( − 3 / 32 ) , j ( b ) = j ( 225 ) , j(a)=j(-3/32),\qquad j(b)=j(225), j ( a ) = j ( − 3/32 ) , j ( b ) = j ( 225 ) ,
j ( b / a ) = j ( − 2400 ) , j ( ( b − 1 ) / ( a − 1 ) ) = j ( − 1024 / 5 ) . j(b/a)=j(-2400),\qquad j((b-1)/(a-1))=j(-1024/5). j ( b / a ) = j ( − 2400 ) , j (( b − 1 ) / ( a − 1 )) = j ( − 1024/5 ) .
The first two equations give six possibilities for each of a , b a,b a , b . The last two select ( a , b ) = ( − 3 / 32 , 225 ) (a,b)=(-3/32,225) ( a , b ) = ( − 3/32 , 225 ) . For this pair, a ( b − 1 ) / ( b ( a − 1 ) ) = 32 / 375 a(b-1)/(b(a-1))=32/375 a ( b − 1 ) / ( b ( a − 1 )) = 32/375 , which also recovers the entry for m = 42 m=42 m = 42 . The coordinate7 ^{7} 7 t = 47775 / ( 16 s + 75 ) t=47775/(16s+75) t = 47775/ ( 16 s + 75 ) sends ( P 42 , P 210 , P 30 , P 105 , P 70 ) (P_{42},P_{210},P_{30},P_{105},P_{70}) ( P 42 , P 210 , P 30 , P 105 , P 70 ) to the values ( θ 0 , … , θ 4 ) (\theta_0,\ldots,\theta_4) ( θ 0 , … , θ 4 ) in (1) .
Remark 8.2. Five branch values of index two make P 1 \mathbb{P}^{1} P 1 hyperbolic as an orbifold, of Euler characteristic 2 − 5 / 2 < 0 2-5/2<0 2 − 5/2 < 0 . With four values a bound of the form of Theorem E fails in general. Indeed, let S S S contain 2 and the primes at which θ 1 , … , θ 4 \theta_1,\ldots,\theta_4 θ 1 , … , θ 4 are not integral and distinct, and let w w w be an S S S -unit. A rational t t t with w ∏ j = 1 4 ( t − θ j ) w\prod_{j=1}^{4}(t-\theta_j) w ∏ j = 1 4 ( t − θ j ) a square has no primes of odd contact outside S S S , and such t t t have unbounded height as soon as the genus-one curve y 2 = w ∏ j ( t − θ j ) y^2=w\prod_j(t-\theta_j) y 2 = w ∏ j ( t − θ j ) has positive rank.
8.2. The family of abelian surfaces. The curve X X X is a coarse moduli space of abelian surfaces with multiplication by O B \mathcal{O}_{B} O B [25 §2.1] . The Atkin–Lehner involutions correspond to isogenies between such surfaces [25 §2.3] . We pass to a finite étale cover carrying a family. Let G = B × G=B^{\times} G = B × act on V = B V=B V = B by ϱ ( g ) x = x g ˉ \varrho(g)x=x\bar{g} ϱ ( g ) x = x g ˉ , where g ˉ \bar{g} g ˉ is the canonical involution. For a suitable Δ ∈ B \Delta\in B Δ ∈ B of trace zero with Δ 2 ∈ Q < 0 \Delta^{2}\in\mathbb{Q}_{<0} Δ 2 ∈ Q < 0 , the form ψ ( x , y ) = Trd ( Δ x y ˉ ) \psi(x,y)=\operatorname{Trd}(\Delta x\bar{y}) ψ ( x , y ) = Trd ( Δ x y ˉ ) is alternating and nondegenerate, ψ ( ϱ ( g ) x , ϱ ( g ) y ) = nrd ( g ) ψ ( x , y ) \psi(\varrho(g)x,\varrho(g)y)=\operatorname{nrd}(g)\psi(x,y) ψ ( ϱ ( g ) x , ϱ ( g ) y ) = nrd ( g ) ψ ( x , y ) , and ϱ \varrho ϱ is a morphism from the Shimura datum of X X X to that of ( V , ψ ) (V,\psi) ( V , ψ ) ; left multiplication by B B B commutes with g g g [14 §§4–6] .
Choose a lattice in V V V self-dual for ψ \psi ψ , put n = 12 n=12 n = 12 , and choose an open compact subgroup K Y ⊆ O ^ × K_Y\subseteq\widehat{\mathcal{O}}^{\times} K Y ⊆ O × whose image under ϱ \varrho ϱ lies in the principal congruence subgroup of level n n n . By functoriality of canonical models, the Shimura curve Y Y Y of level K Y K_Y K Y is a smooth projective curve over Q \mathbb{Q} Q , possibly geometrically disconnected, with a morphism to the fine moduli space of principally polarized abelian surfaces with full level-n n n structure [14 §4] . The pullback of the universal family is a principally polarized abelian scheme A → Y \mathcal{A} \to Y A → Y . Its polarizations induce the fixed positive involution α ↦ Δ − 1 α ˉ Δ \alpha\mapsto\Delta^{-1}\bar{\alpha}\Delta α ↦ Δ − 1 α ˉ Δ on B \mathbb{B} B . It has the following properties.
(1) The map Y → X Y \to X Y → X is finite and étale, since on complex points it is induced by inclusions of subgroups of conjugates of Γ \Gamma Γ , which act freely on H \mathfrak{H} H .
(2) For y ∈ Y ( Q ‾ ) y \in Y(\overline{\mathbb{Q}}) y ∈ Y ( Q ) , H 1 ( A y ( C ) , Q ) H_1(\mathcal{A}_y(\mathbb{C}),\mathbb{Q}) H 1 ( A y ( C ) , Q ) is V V V with the complex structure given by a point of H ± \mathfrak{H}^{\pm} H ± , and left multiplication defines an inclusion B ⊆ End Q ( A y ) \mathbb{B} \subseteq\operatorname{End}_{\mathbb{Q}}(\mathcal{A}_y) B ⊆ End Q ( A y ) .
(3) If y , y ′ ∈ Y ( Q ‾ ) y,y' \in Y(\overline{\mathbb{Q}}) y , y ′ ∈ Y ( Q ) have the same image in X / W X/W X / W , then A y ∼ Q A y ′ \mathcal{A}_y \sim_{\mathbb{Q}} \mathcal{A}_{y'} A y ∼ Q A y ′ . Indeed, the Atkin–Lehner involutions and the change of level act on the finite adelic coordinate, so the points of H ± \mathfrak{H}^{\pm} H ± attached to y y y and y ′ y' y ′ lie in one G ( Q ) G(\mathbb{Q}) G ( Q ) -orbit, and an element of ϱ ( G ( Q ) ) \varrho(G(\mathbb{Q})) ϱ ( G ( Q )) identifies the two rational Hodge structures. For the same reason, an abelian surface over Q ‾ \overline{\mathbb{Q}} Q with multiplication by O B \mathcal{O}_{\mathbb{B}} O B whose moduli point is the image of y y y is isogenous to A y \mathcal{A}_y A y .
(4) The moduli map of A \mathcal{A} A is non-constant on every component of Y Y Y , as the complex structure varies in H \mathfrak{H} H while a fixed rational Hodge structure has only countably many presentations on V V V .
Put π = ϖ ∘ ( Y → X ) : Y → P 1 \pi=\varpi\circ(Y\to X):Y\to\mathbb{P}^1 π = ϖ ∘ ( Y → X ) : Y → P 1 . By Lemma 8.1 and (1), π \pi π is étale away from θ 0 , … , θ 4 \theta_0,\ldots,\theta_4 θ 0 , … , θ 4 and has ramification index two at every point above them. Put C 1 = ( deg π ) ! C_1=(\deg\pi)! C 1 = ( deg π )! . Fix a finite set S 0 S_0 S 0 of primes containing 2 , 3 , 5 , 7 2,3,5,7 2 , 3 , 5 , 7 and the primes dividing the denominator of some θ j \theta_j θ j or the numerator of some θ j − θ k \theta_j-\theta_k θ j − θ k with j ≠ k j\ne k j = k . By spreading out, we enlarge S 0 S_0 S 0 so that over Z [ 1 / S 0 ] \mathbb{Z}[1/S_0] Z [ 1/ S 0 ] the curve Y Y Y extends to a smooth proper scheme Y \mathcal{Y} Y and A \mathcal{A} A to an abelian scheme over Y \mathcal{Y} Y , and π \pi π extends to a finite morphism Y → P 1 \mathcal{Y}\to\mathbb{P}^1 Y → P 1 , étale away from the sections θ j \theta_j θ j , with π ∗ ( θ j ) = 2 D j \pi^*(\theta_j)=2D_j π ∗ ( θ j ) = 2 D j for finite étale D j D_j D j .
Fibres above rational points.
Lemma 8.3. Let t ∈ Q ∖ { θ 0 , … , θ 4 } t\in\mathbb{Q}\setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } , M = M S 0 ( t ) M=M_{S_0}(t) M = M S 0 ( t ) , and let T T T be S 0 S_0 S 0 together with the prime divisors of M M M . Let K ⊂ Q ‾ K\subset\overline{\mathbb{Q}} K ⊂ Q be the compositum of the residue fields Q ( y ) \mathbb{Q}(y) Q ( y ) of the points y ∈ π − 1 ( t ) ( Q ‾ ) y\in\pi^{-1}(t)(\overline{\mathbb{Q}}) y ∈ π − 1 ( t ) ( Q ) . Then K / Q K/\mathbb{Q} K / Q is Galois of degree at most C 1 C_1 C 1 and unramified outside T T T . Its ramification index at each prime outside S 0 S_0 S 0 is at most two. For y ∈ π − 1 ( t ) y\in\pi^{-1}(t) y ∈ π − 1 ( t ) , the surface A y \mathcal{A}_y A y is defined over K K K and has good reduction at every finite place of K K K . Every homomorphism over Q ‾ \overline{\mathbb{Q}} Q between surfaces in the fibre is defined over K K K .
Proof. The Galois group of K K K permutes the at most deg π \deg\pi deg π points of the fibre faithfully. Let q ∉ S 0 q\notin S_0 q ∈ / S 0 . If v q ( t − θ j ) ≤ 0 v_q(t-\theta_j)\le0 v q ( t − θ j ) ≤ 0 for every j j j , the reduction of t t t avoids the branch sections, so the fibre is unramified at q q q . Otherwise k = v q ( t − θ j ) > 0 k=v_q(t-\theta_j)>0 k = v q ( t − θ j ) > 0 for exactly one j j j . Let W \mathcal{W} W be the completion of the maximal unramified extension of Z q \mathbb{Z}_q Z q . At a point of D j D_j D j over F ‾ q \overline{\mathbb{F}}_q F q , the complete local ring of Y W \mathcal{Y}_{\mathcal{W}} Y W is W [ [ z ] ] \mathcal{W}[[z]] W [[ z ]] , and the pullback of T − θ j T-\theta_j T − θ j is z 2 e ( z ) z^2e(z) z 2 e ( z ) with e e e a unit. As q q q is odd, e e e has a square root. After changing the parameter, the fibre is therefore described by
z 2 = t − θ j ∈ q k W × . z^2=t-\theta_j\in q^k\mathcal{W}^{\times}. z 2 = t − θ j ∈ q k W × .
Its points are defined over W [ 1 / q ] \mathcal{W}[1/q] W [ 1/ q ] if k k k is even, and over W [ 1 / q ] ( q ) \mathcal{W}[1/q](\sqrt{q}) W [ 1/ q ] ( q ) if k k k is odd. This holds at every point above θ j \theta_j θ j , so the ramification index of their compositum is at most two. In particular, K K K is unramified at q q q unless q ∣ M q\mid M q ∣ M .
Each A y \mathcal{A}_y A y and its full level-12 structure are defined over K K K . By Silverberg’s theorem [75 Theorem 2.4] , applied to products of two fibres, every homomorphism between these surfaces is defined over K K K .
Raynaud’s criterion [78 IX, Proposition 4.7] , applied to the rational 3-torsion away from 3 3 3 and to the rational 4-torsion above 3 3 3 , shows that A y \mathcal{A}_y A y is semistable everywhere. If the toric part of a special fibre of the Néron model were non-zero, its rational character space would be a non-zero B \mathbb{B} B -module of dimension at most 2. Since B \mathbb{B} B is a division algebra of dimension 4 over Q \mathbb{Q} Q , this is impossible. Thus A y A_y A y has good reduction everywhere. □ \square □
Modularity of controlled level
Fix t ∈ Q ∖ { θ 0 , … , θ 4 } t \in\mathbb{Q} \setminus\{\theta_0,\ldots,\theta_4\} t ∈ Q ∖ { θ 0 , … , θ 4 } , let M M M , T \mathcal{T} T and K K K be as in Lemma 8.3, fix y ∈ π − 1 ( t ) y \in\pi^{-1}(t) y ∈ π − 1 ( t ) and put A = A y A=A_y A = A y . For g ∈ G Q g \in G_{\mathbb{Q}} g ∈ G Q , the point g ( y ) g(y) g ( y ) lies above t t t , so (3) of §8.2 supplies an isogeny g A → A {}^{g}A \to A g A → A compatible with the quaternionic action. If End Q 0 ( A ) = B \operatorname{End}_{\mathbb{Q}}^{0}(A)=\mathbb{B} End Q 0 ( A ) = B , the theorem of Ribet and Pyle supplies an abelian variety of G L 2 \mathrm{GL}_2 GL 2 -type over Q \mathbb{Q} Q geometrically isogenous to a power of A A A (see [32 Theorem 1.1] ). Such varieties are modular by Khare–Wintenberger [45 Corollary 10.2(i)] . This qualitative modularity identifies the associated representations with modular ones (see § 11.4 ). The splitting constructed below determines the representations whose ramification bounds the level.
Let D = End Q 0 ( A ) ⊇ B D=\operatorname{End}_{\mathbb{Q}}^{0}(A)\supseteq\mathbb{B} D = End Q 0 ( A ) ⊇ B and let D ′ D' D ′ be the centralizer of B \mathbb{B} B in D D D . By (2) of §8.2, an element of D ′ D' D ′ acts on V V V by right multiplication by an element of B \mathbb{B} B commuting with the complex structure, whose centralizer in M 2 ( R ) M_2(\mathbb{R}) M 2 ( R ) is a copy of C \mathbb{C} C . So D ′ D' D ′ is Q \mathbb{Q} Q or an imaginary quadratic field k k k , and D = B ⊗ Q D ′ D=\mathbb{B}\otimes_{\mathbb{Q}}D' D = B ⊗ Q D ′ is B \mathbb{B} B or M 2 ( k ) M_2(k) M 2 ( k ) . In the second case the idempotents of D D D are defined over K K K , so A ∼ K E 2 A\sim_K E^2 A ∼ K E 2 for an elliptic curve E E E over K K K with complex multiplication by an order of k k k .
The descent obstruction and its splitting
For the rest of § 11.3 , suppose D = B D=\mathbb{B} D = B ; then A A A is absolutely simple. In the cocycles below, the coefficient groups are discrete with trivial Galois action. For g ∈ G Q g\in G_{\mathbb{Q}} g ∈ G Q put g ˉ = g ∣ K \bar{g}=g|_K g ˉ = g ∣ K , and for each g ˉ ∈ Gal ( K / Q ) \bar{g}\in\operatorname{Gal}(K/\mathbb{Q}) g ˉ ∈ Gal ( K / Q ) choose a quasi-isogeny μ g ˉ : g A → A \mu_{\bar{g}}\colon{}^{g}A\to A μ g ˉ : g A → A , with μ 1 = 1 \mu_1=1 μ 1 = 1 , such that
μ g ˉ g α = α μ g ˉ ( α ∈ B ) ; (30) \mu_{\bar{g}}\,{}^{g}\alpha=\alpha\mu_{\bar{g}}\qquad(\alpha\in\mathbb{B}); \tag*{(30)} μ g ˉ g α = α μ g ˉ ( α ∈ B ) ; (30)
this is possible by the Skolem–Noether theorem, and μ g ˉ \mu_{\bar{g}} μ g ˉ is defined over K K K by Lemma 8.3. Since the centralizer of B \mathbb{B} B in D = B D=\mathbb{B} D = B is Q \mathbb{Q} Q , the relations (30) imply
μ g ˉ g μ h ˉ = c ( g ˉ , h ˉ ) μ g h ‾ , c ( g ˉ , h ˉ ) ∈ Q × . \mu_{\bar{g}}\,{}^{g}\mu_{\bar{h}}=c(\bar{g},\bar{h})\mu_{\overline{gh}},\qquad c(\bar{g},\bar{h})\in\mathbb{Q}^{\times}. μ g ˉ g μ h ˉ = c ( g ˉ , h ˉ ) μ g h , c ( g ˉ , h ˉ ) ∈ Q × .
Associativity makes c c c a 2-cocycle on Gal ( K / Q ) \operatorname{Gal}(K/\mathbb{Q}) Gal ( K / Q ) . The degrees satisfy
c ( g ˉ , h ˉ ) 4 = deg μ g ˉ deg μ h ˉ deg μ g h ‾ . c(\bar{g},\bar{h})^4=\frac{\deg\mu_{\bar{g}}\deg\mu_{\bar{h}}}{\deg\mu_{\overline{gh}}}. c ( g ˉ , h ˉ ) 4 = deg μ g h deg μ g ˉ deg μ h ˉ .
The polarizations on A A A and g A {}^{g}A g A induce the same involution on B \mathbb{B} B , so μ g ˉ † μ g ˉ \mu_{\bar{g}}^{\dagger}\mu_{\bar{g}} μ g ˉ † μ g ˉ centralizes B \mathbb{B} B and is a positive rational scalar d g ˉ d_{\bar{g}} d g ˉ ; here † \dagger † is the adjoint for these polarizations. Hence deg μ g ˉ = d g ˉ 2 \deg\mu_{\bar{g}}=d_{\bar{g}}^2 deg μ g ˉ = d g ˉ 2 . Let a g ˉ > 0 a_{\bar{g}}>0 a g ˉ > 0 be the real fourth root of 1 / deg μ g ˉ 1/\deg\mu_{\bar{g}} 1/ deg μ g ˉ , so that a g ˉ 2 = d g ˉ − 1 ∈ Q a_{\bar{g}}^2=d_{\bar{g}}^{-1}\in\mathbb{Q} a g ˉ 2 = d g ˉ − 1 ∈ Q , and put
ϵ ( g ˉ , h ˉ ) = a g ˉ a h ˉ a g h ‾ c ( g ˉ , h ˉ ) = sign c ( g ˉ , h ˉ ) ∈ { ± 1 } , \epsilon(\bar{g},\bar{h})=\frac{a_{\bar{g}}a_{\bar{h}}}{a_{\overline{gh}}}c(\bar{g},\bar{h})=\operatorname{sign}c(\bar{g},\bar{h})\in\{\pm1\}, ϵ ( g ˉ , h ˉ ) = a g h a g ˉ a h ˉ c ( g ˉ , h ˉ ) = sign c ( g ˉ , h ˉ ) ∈ { ± 1 } ,
a 2-cocycle on Gal ( K / Q ) \operatorname{Gal}(K/\mathbb{Q}) Gal ( K / Q ) , which we inflate to G Q G_{\mathbb{Q}} G Q . The following lemma splits ϵ \epsilon ϵ with control of ramification. It is an explicit form of Tate’s theorem that H 2 ( G Q , Q ‾ × ) = 0 H^2(G_{\mathbb{Q}},\overline{\mathbb{Q}}^{\times})=0 H 2 ( G Q , Q × ) = 0 for the trivial action on Q ‾ × \overline{\mathbb{Q}}^{\times} Q × [76 §6] , [72 Theorem 6.3] .
Lemma 8.4. There exist a Dirichlet character χ \chi χ of order m ≤ C M m\le CM m ≤ C M and a continuous map β : G Q → μ 2 m \beta:G_{\mathbb{Q}}\to\mu_{2m} β : G Q → μ 2 m with β 2 = χ \beta^2=\chi β 2 = χ and β ( g ) β ( h ) / β ( g h ) = ϵ ( g , h ) \beta(g)\beta(h)/\beta(gh)=\epsilon(g,h) β ( g ) β ( h ) / β ( g h ) = ϵ ( g , h ) , such that β \beta β is trivial on the inertia groups at the primes outside T \mathcal{T} T , and χ \chi χ is a product of characters of conductor q q q and of odd prime conductors q ∈ T q\in\mathcal{T} q ∈ T . The restriction of β \beta β to G K G_K G K is a character, invariant under conjugation by G Q G_{\mathbb{Q}} G Q .
Proof. The class of ϵ \epsilon ϵ in H 2 ( G Q , { ± 1 } ) = Br ( Q ) [ 2 ] H^2(G_{\mathbb{Q}},\{\pm1\})=\operatorname{Br}(\mathbb{Q})[2] H 2 ( G Q , { ± 1 }) = Br ( Q ) [ 2 ] has trivial local components at the primes outside T \mathcal{T} T : there are factors through the unramified quotient Z ^ \widehat{\mathbb{Z}} Z , and H 2 ( Z ^ , { ± 1 } ) = 0 H^2(\widehat{\mathbb{Z}},\{\pm1\})=0 H 2 ( Z , { ± 1 }) = 0 . Let χ v \chi_v χ v be a character of finite order of G Q G_{\mathbb{Q}} G Q , viewed as a character of Q × \mathbb{Q}^{\times} Q × . The obstruction in H 2 ( G Q , { ± 1 } ) H^2(G_{\mathbb{Q}},\{\pm1\}) H 2 ( G Q , { ± 1 }) to a square root of finite order of χ v \chi_v χ v vanishes exactly when χ v ( − 1 ) = 1 \chi_v(-1)=1 χ v ( − 1 ) = 1 . Indeed, Q v × \mathbb{Q}_v^\times Q v × is the product of a finite cyclic group, in which − 1 -1 − 1 is the element of order two, and a group on which characters of finite order have square roots. Let χ \chi χ be the product of an odd character of conductor q q q for each odd q ∈ T q \in\mathcal{T} q ∈ T at which ϵ \epsilon ϵ has non-zero local class, and of the character of conductor 4 4 4 if ϵ \epsilon ϵ has non-zero local class at 2 2 2 . The square-root obstruction of χ \chi χ and the class of ϵ \epsilon ϵ agree at all finite places, hence everywhere, by the Albert–Brauer–Hasse–Noether theorem. So for s : G Q → μ 2 m s\colon G_{\mathbb{Q}} \to\mu_{2m} s : G Q → μ 2 m continuous with s 2 = χ s^2=\chi s 2 = χ , m m m the order of χ \chi χ , the coboundary of s s s differs from ϵ \epsilon ϵ by the coboundary of a map to { ± 1 } \{\pm1\} { ± 1 } , and correcting s s s by this map defines β \beta β . Here m ≤ 2 ∏ q ∈ T q ≤ C M m \leq2\prod_{q\in\mathcal{T}}q \leq CM m ≤ 2 ∏ q ∈ T q ≤ C M . As ϵ \epsilon ϵ is trivial on G K × G Q G_K \times G_{\mathbb{Q}} G K × G Q and G Q × G K G_{\mathbb{Q}} \times G_K G Q × G K , β ∣ G K \left.\beta\right|_{G_K} β ∣ G K is a character invariant under conjugation. At an odd prime q ∉ T q \notin\mathcal{T} q ∈ / T the inertia group lies in G K G_K G K , and β \beta β restricts to a character of it whose square χ ∣ I q \left.\chi\right|_{I_q} χ ∣ I q is trivial. This character is non-trivial for only finitely many q q q . Multiplying β \beta β by the quadratic character of Q ( d ) \mathbb{Q}(\sqrt{d}) Q ( d ) , for d d d the product of these q q q , makes β \beta β unramified outside T \mathcal{T} T and changes neither β 2 \beta^2 β 2 nor the coboundary. □
Let K ′ ⊇ K K' \supseteq K K ′ ⊇ K be the fixed field of the kernel of β ∣ G K \left.\beta\right|_{G_K} β ∣ G K . It is Galois over Q \mathbb{Q} Q , and
[ K ′ : Q ] ≤ 2 m [ K : Q ] ≤ C M . (31) [K' : \mathbb{Q}] \leq2m[K : \mathbb{Q}] \leq CM. \tag*{(31)} [ K ′ : Q ] ≤ 2 m [ K : Q ] ≤ C M . (31)
For each prime ℓ \ell ℓ fix an embedding ι ℓ : Q ‾ → Q ‾ ℓ \iota_\ell\colon\overline{\mathbb{Q}} \to\overline{\mathbb{Q}}_\ell ι ℓ : Q → Q ℓ , and define on V ℓ ( A ) ⊗ Q ℓ Q ‾ ℓ V_\ell(A) \otimes_{\mathbb{Q}_\ell}\overline{\mathbb{Q}}_\ell V ℓ ( A ) ⊗ Q ℓ Q ℓ
R ℓ ( g ) = ι ℓ ( β ( g ) a g ‾ ) μ g ‾ ∘ g ( g ∈ G Q ) , (32) R_\ell(g)=\iota_\ell\bigl(\beta(g)a_{\overline{g}}\bigr)\mu_{\overline{g}}\circ g \qquad(g\in G_{\mathbb{Q}}), \tag*{(32)} R ℓ ( g ) = ι ℓ ( β ( g ) a g ) μ g ∘ g ( g ∈ G Q ) , (32)
where g g g maps V ℓ ( A ) V_\ell(A) V ℓ ( A ) to V ℓ ( g A ) V_\ell({}^{g}A) V ℓ ( g A ) and μ g ‾ \mu_{\overline{g}} μ g maps it back. Since β ( g ) β ( h ) = ϵ ( g , h ) β ( g h ) \beta(g)\beta(h)=\epsilon(g,h)\beta(gh) β ( g ) β ( h ) = ϵ ( g , h ) β ( g h ) and a g ‾ a h ‾ c ( g ‾ , h ‾ ) = ϵ ( g ‾ , h ‾ ) a g ‾ h ‾ a_{\overline{g}}a_{\overline{h}}c(\overline{g},\overline{h})=\epsilon(\overline{g},\overline{h})a_{\overline{g}\overline{h}} a g a h c ( g , h ) = ϵ ( g , h ) a g h , R ℓ R_\ell R ℓ is a continuous representation of G Q G_{\mathbb{Q}} G Q commuting with B B B by (30) . On G K ′ G_{K'} G K ′ it is the Galois action on V ℓ ( A ) ⊗ Q ‾ ℓ V_\ell(A)\otimes\overline{\mathbb{Q}}_\ell V ℓ ( A ) ⊗ Q ℓ . Writing B ⊗ Q ‾ ℓ ≃ M 2 ( Q ‾ ℓ ) B\otimes\overline{\mathbb{Q}}_\ell\simeq M_2(\overline{\mathbb{Q}}_\ell) B ⊗ Q ℓ ≃ M 2 ( Q ℓ ) , let r ℓ r_\ell r ℓ be the action of R ℓ R_\ell R ℓ on the two-dimensional image of a rank-one idempotent, so that R ℓ ≃ r ℓ ⊕ r ℓ R_\ell\simeq r_\ell\oplus r_\ell R ℓ ≃ r ℓ ⊕ r ℓ .
Modularity of the associated representations.
Lemma 8.5. For every finite extension L / K ′ L/K' L / K ′ , the restriction r ℓ ∣ G L \left.r_\ell\right|_{G_L} r ℓ ∣ G L is absolutely irreducible. The representation r ℓ r_\ell r ℓ is unramified outside T ∪ { ℓ } \mathcal{T}\cup\{\ell\} T ∪ { ℓ } . There exist algebraic numbers a q a_q a q , independent of ℓ \ell ℓ , such that
tr r ℓ ( Frob q ) = ι ℓ ( a q ) ( q ∉ T ∪ { ℓ } ) . \operatorname{tr}r_\ell(\operatorname{Frob}_q)=\iota_\ell(a_q)\qquad(q\notin\mathcal{T}\cup\{\ell\}). tr r ℓ ( Frob q ) = ι ℓ ( a q ) ( q ∈ / T ∪ { ℓ }) .
Proof. By Faltings’ theorem [21 ] , V ℓ ( A ) V_\ell(A) V ℓ ( A ) is semisimple over G L G_L G L , with commutant
End L ( A ) ⊗ Q ‾ ℓ = B ⊗ Q ‾ ℓ . \operatorname{End}_L(A)\otimes\overline{\mathbb{Q}}_\ell=B\otimes\overline{\mathbb{Q}}_\ell. End L ( A ) ⊗ Q ℓ = B ⊗ Q ℓ .
Since V ℓ ( A ) ⊗ Q ‾ ℓ = r ℓ ⊕ r ℓ V_\ell(A)\otimes\overline{\mathbb{Q}}_\ell=r_\ell\oplus r_\ell V ℓ ( A ) ⊗ Q ℓ = r ℓ ⊕ r ℓ on G L G_L G L , this proves absolute irreducibility.
For q ∉ T q\notin\mathcal{T} q ∈ / T , the field K K K is unramified at q q q , the surface A A A has good reduction above q q q , and β \beta β is unramified there. Thus r ℓ r_\ell r ℓ is unramified at q ≠ ℓ q\ne\ell q = ℓ . Let g g g be an arithmetic Frobenius at a place above q q q . In (32) , μ g ‾ ∘ g \mu_{\overline{g}}\circ g μ g ∘ g acts on the reduction A ~ \widetilde{A} A of A A A as the quasi-endomorphism Π q = ν g ‾ ∘ Frob \Pi_q=\nu_{\overline{g}}\circ\operatorname{Frob} Π q = ν g ∘ Frob , where ν g ‾ \nu_{\overline{g}} ν g is the reduction of μ g ‾ \mu_{\overline{g}} μ g and Frob \operatorname{Frob} Frob is relative Frobenius. Its characteristic polynomial is rational and independent of ℓ \ell ℓ [54 §19, Theorem 4] . Therefore one may take
a q = 1 2 β ( g ) a g ‾ tr Π q . a_q=\tfrac{1}{2}\beta(g)a_{\overline{g}}\operatorname{tr}\Pi_q. a q = 2 1 β ( g ) a g tr Π q .
These numbers lie in the number field generated by μ 2 m \mu_{2m} μ 2 m and the finitely many a g ‾ a_{\overline{g}} a g . □
Proposition 8.6. There exists a newform f f f of weight two such that r ℓ ≃ ρ f , ℓ r_\ell\simeq\rho_{f,\ell} r ℓ ≃ ρ f , ℓ for every prime ℓ \ell ℓ .
Proof. By the Ribet–Pyle theorem and Khare–Wintenberger’s modularity theorem cited above, there exists a modular abelian variety A 0 A_0 A 0 over Q \mathbb{Q} Q with ( A 0 ) Q ‾ ∼ A e (A_0)_{\overline{\mathbb{Q}}}\sim A^e ( A 0 ) Q ∼ A e for some e ≥ 1 e\geq1 e ≥ 1 . Choose a finite Galois extension L / Q L/\mathbb{Q} L / Q containing K ′ K' K ′ over which this isogeny is defined, and fix a prime ℓ 0 \ell_0 ℓ 0 . The decomposition of V ℓ 0 ( A 0 ) ⊗ Q ‾ ℓ 0 V_{\ell_0}(A_0)\otimes\overline{\mathbb{Q}}_{\ell_0} V ℓ 0 ( A 0 ) ⊗ Q ℓ 0 into modular representations has a constituent ρ f , ℓ 0 \rho_{f,\ell_0} ρ f , ℓ 0 satisfying
ρ f , ℓ 0 ∣ G L ≃ r ℓ 0 ∣ G L . \left.\rho_{f,\ell_0}\right|_{G_L}\simeq\left.r_{\ell_0}\right|_{G_L}. ρ f , ℓ 0 ∣ G L ≃ r ℓ 0 ∣ G L .
Indeed, ( A 0 ) L ≃ L A e (A_0)_L \simeq_L A^e ( A 0 ) L ≃ L A e , and Lemma 8.5 applies. The line
Hom G L ( ρ f 1 , ℓ 0 , r ℓ 0 ) \operatorname{Hom}_{G_L}(\rho_{f_1,\ell_0},r_{\ell_0}) Hom G L ( ρ f 1 , ℓ 0 , r ℓ 0 )
is stable under G Q G_{\mathbb{Q}} G Q . Its action defines a finite-order character η \eta η , trivial on G L G_L G L , and hence
r ℓ 0 ≃ ρ f 1 , ℓ 0 ⊗ η . r_{\ell_0} \simeq\rho_{f_1,\ell_0} \otimes\eta. r ℓ 0 ≃ ρ f 1 , ℓ 0 ⊗ η .
View η \eta η as a Dirichlet character through ι ℓ 0 \iota_{\ell_0} ι ℓ 0 , and let f f f be the newform associated to the twist f 1 ⊗ η f_1 \otimes\eta f 1 ⊗ η . Then r ℓ 0 ≃ ρ f , ℓ 0 r_{\ell_0} \simeq\rho_{f,\ell_0} r ℓ 0 ≃ ρ f , ℓ 0 . Lemma 8.5 identifies their Frobenius traces as algebraic numbers, so a q = a q ( f ) a_q = a_q(f) a q = a q ( f ) outside a finite set of primes. The same identity for every ℓ \ell ℓ , Chebotarev density and the Brauer–Nesbitt theorem imply r ℓ ≃ ρ f , ℓ r_\ell\simeq\rho_{f,\ell} r ℓ ≃ ρ f , ℓ . □
Ramification and the level.
Lemma 8.7. For each q ∈ T q \in\mathcal{T} q ∈ T , there exists u q ≥ 0 u_q \ge0 u q ≥ 0 such that r ℓ ( G Q q u ) = 1 r_\ell(G_{\mathbb{Q}_q}^{u}) = 1 r ℓ ( G Q q u ) = 1 for all u > u q u > u_q u > u q and ℓ ≠ q \ell\ne q ℓ = q . One may take u q = 0 u_q = 0 u q = 0 for q ∈ T ∖ S 0 q \in\mathcal{T} \setminus S_0 q ∈ T ∖ S 0 ; for q ∈ S 0 q \in S_0 q ∈ S 0 , the bound depends only on q q q and the data of §8.2.
Proof. Let q ∈ T ∖ S 0 q \in\mathcal{T} \setminus S_0 q ∈ T ∖ S 0 . By Lemma 8.3, K K K has ramification index at most two at q q q . As q q q is odd, the wild inertia group P q P_q P q lies in G K G_K G K . On P q P_q P q , β \beta β is a character with β 2 = χ \beta^2 = \chi β 2 = χ trivial, so β ∣ P q = 1 \left.\beta\right|_{P_q} = 1 β ∣ P q = 1 as q q q is odd; and P q P_q P q acts trivially on V ℓ ( A ) V_\ell(A) V ℓ ( A ) by good reduction. So R ℓ R_\ell R ℓ is trivial on P q P_q P q , which contains G Q q u G_{\mathbb{Q}_q}^{u} G Q q u for u > 0 u > 0 u > 0 .
Let q ∈ S 0 q \in S_0 q ∈ S 0 and ℓ ≠ q \ell\ne q ℓ = q . The completions K w K_w K w , w ∣ q w \mid q w ∣ q , have degree at most C 1 C_1 C 1 over Q q \mathbb{Q}_q Q q , so they range over a finite set. The variety A A A has good reduction over each K w K_w K w , so inertia acts trivially on V ℓ ( A ) V_\ell(A) V ℓ ( A ) . By local class field theory β ∣ G K w \left.\beta\right|_{G_{K_w}} β ∣ G K w corresponds to a character β w \beta_w β w of K w × K_w^\times K w × whose square is χ ∘ N K w / Q q \chi\circ N_{K_w/\mathbb{Q}_q} χ ∘ N K w / Q q . The character χ \chi χ is trivial on 1 + 4 q Z q 1 + 4q\mathbb{Z}_q 1 + 4 q Z q by Lemma 8.4. Thus, for an integer b b b depending only on K w K_w K w , β w \beta_w β w maps U w b = 1 + m w b U_w^b = 1 + \mathfrak{m}_w^b U w b = 1 + m w b into { ± 1 } \{\pm1\} { ± 1 } and kills ( U w b ) 2 ⊇ U w b + v w ( 2 ) (U_w^b)^2 \supseteq U_w^{b+v_w(2)} ( U w b ) 2 ⊇ U w b + v w ( 2 ) for b b b large. Choose u q u_q u q such that, for u > u q u > u_q u > u q and every possible K w K_w K w , the group G Q q u G_{\mathbb{Q}_q}^{u} G Q q u lies in the inertia group of K w K_w K w and its image under reciprocity lies in U w b + v w ( 2 ) U_w^{b+v_w(2)} U w b + v w ( 2 ) [77 IV, §3 and XV, §2] . Then g ‾ = 1 \overline{g} = 1 g = 1 , β ( g ) = 1 \beta(g) = 1 β ( g ) = 1 and g g g acts trivially on V ℓ ( A ) V_\ell(A) V ℓ ( A ) for g ∈ G Q q u g \in G_{\mathbb{Q}_q}^{u} g ∈ G Q q u , so R ℓ ( g ) = 1 R_\ell(g) = 1 R ℓ ( g ) = 1 . □
Proposition 8.8. There exist a newform f f f of weight two and level N ≤ C M 2 N \le CM^2 N ≤ C M 2 and an abelian subvariety B B B of J 1 ( N ) J_1(N) J 1 ( N ) , defined over K ′ K' K ′ , with A ∼ K ′ B A \sim_{K'} B A ∼ K ′ B .
Proof. By Proposition 8.6, there is a newform f f f with r ℓ ≃ ρ f , ℓ r_\ell\simeq\rho_{f,\ell} r ℓ ≃ ρ f , ℓ for every ℓ \ell ℓ . For ℓ ≠ q \ell\ne q ℓ = q , the exponent of q q q in its level N N N is the conductor exponent of r ℓ r_\ell r ℓ at q q q [5 ] . It is at most 2 plus the Swan conductor; hence, by Lemmas 8.5 and 8.7, it vanishes for q ∉ T q \notin\mathcal{T} q ∈ / T , is at most 2 for q ∣ M q \mid M q ∣ M , and is at most 2 + 2 u q 2 + 2u_q 2 + 2 u q for q ∈ S 0 q \in S_0 q ∈ S 0 . Hence N ≤ C M 2 N \le CM^2 N ≤ C M 2 . By the Eichler–Shimura construction [79 , 15 ] , V ℓ ( J 1 ( N ) ) ⊗ Q ‾ ℓ V_\ell(J_1(N)) \otimes\overline{\mathbb{Q}}_\ell V ℓ ( J 1 ( N )) ⊗ Q ℓ contains a constituent isomorphic to ρ f , ℓ \rho_{f,\ell} ρ f , ℓ . On G K ′ G_{K'} G K ′ , V ℓ ( A ) ⊗ Q ‾ ℓ ≃ r ℓ ⊕ r ℓ ≃ ρ f , ℓ ⊕ ρ f , ℓ V_\ell(A) \otimes\overline{\mathbb{Q}}_\ell\simeq r_\ell\oplus r_\ell\simeq\rho_{f,\ell} \oplus\rho_{f,\ell} V ℓ ( A ) ⊗ Q ℓ ≃ r ℓ ⊕ r ℓ ≃ ρ f , ℓ ⊕ ρ f , ℓ , so Hom G K ′ ( V ℓ ( A ) , V ℓ ( J 1 ( N ) ) ) ≠ 0 \operatorname{Hom}_{G_{K'}}(V_\ell(A), V_\ell(J_1(N))) \ne0 Hom G K ′ ( V ℓ ( A ) , V ℓ ( J 1 ( N ))) = 0 , and by Faltings’ theorem [21 ] there is a non-zero homomorphism A → J 1 ( N ) A \to J_1(N) A → J 1 ( N ) over K ′ K' K ′ . As A A A is simple, its image B B B is isogenous to A A A over K ′ K' K ′ . □
Heights; proofs of Theorems D and E.
Lemma 8.9. For y ∈ Y ( Q ) y \in Y(\mathbb{Q}) y ∈ Y ( Q ) , h ( π ( y ) ) ≤ C ( 1 + max { 0 , h F ( A y ) } ) h(\pi(y)) \le C(1 + \max\{0,h_F(\mathcal{A}_y)\}) h ( π ( y )) ≤ C ( 1 + max { 0 , h F ( A y )}) .
Proof. Pass to a fixed finite étale cover Y ′ → Y Y' \to Y Y ′ → Y on which A \mathcal{A} A has full level-24 structure, so that the level is divisible by 8, as required by Moret-Bailly’s theorem [50 Theorem 1.1] . Let κ : Y ′ → A 2 , 24 \kappa:Y' \to\mathcal{A}_{2,24} κ : Y ′ → A 2 , 24 be its moduli morphism and λ \lambda λ the Hodge bundle. By this theorem, a positive power λ m \lambda^m λ m extends to an ample line bundle on a projective compactification of A 2 , 24 \mathcal{A}_{2,24} A 2 , 24 . The curve Y ′ Y' Y ′ is proper, so κ ( Y ′ ) \kappa(Y') κ ( Y ′ ) avoids the boundary and the pullback of the Hodge metric is smooth. The map κ \kappa κ is non-constant on every component by (4) of §8.2; hence L = κ ∗ λ m \mathcal{L} = \kappa^*\lambda^m L = κ ∗ λ m is ample on every component. Part (ii) of Moret-Bailly’s theorem therefore implies that an ordinary Weil height for L \mathcal{L} L satisfies
h L ( y ′ ) = m h F ( A y ) + O ( 1 ) ( y ′ ↦ y ) . h_{\mathcal{L}}(y') = m h_{\mathrm{F}}(A_y) + O(1) \qquad(y' \mapsto y). h L ( y ′ ) = m h F ( A y ) + O ( 1 ) ( y ′ ↦ y ) .
The height machine on the fixed projective curve Y ′ Y' Y ′ [2 Ch. 2] now bounds h ( π ( y ) ) h(\pi(y)) h ( π ( y )) by C ( 1 + max { 0 , h L ( y ′ ) } ) C(1+\max\{0,h_{\mathcal{L}}(y')\}) C ( 1 + max { 0 , h L ( y ′ )}) . □
Proof of Theorem D. Let x ∈ X ( Q ‾ ) x \in X(\overline{\mathbb{Q}}) x ∈ X ( Q ) be the moduli point of A A A and choose y ∈ Y y \in Y y ∈ Y above x x x . By (3) of §8.2, A ∼ Q A y A \sim_{\mathbb{Q}} A_y A ∼ Q A y . In the non-CM case, Proposition 8.8 therefore proves the assertion.
Suppose that A A A has CM. Then A ∼ Q E 0 2 A \sim_{\mathbb{Q}} E_0^2 A ∼ Q E 0 2 for a CM elliptic curve E 0 E_0 E 0 , by the description of D D D in § 11.3 , and x x x is a CM point. As ϖ ( x ) = t ∈ Q \varpi(x)=t \in\mathbb{Q} ϖ ( x ) = t ∈ Q and deg ϖ = 16 \deg\varpi=16 deg ϖ = 16 , one has [ Q ( x ) : Q ] ≤ 16 [\mathbb{Q}(x):\mathbb{Q}]\leq16 [ Q ( x ) : Q ] ≤ 16 . Let O \mathcal{O} O be the imaginary quadratic order associated with x x x , let k k k be its fraction field, and let H O H_{\mathcal{O}} H O be its ring class field. Shimura reciprocity gives H O = k Q ( x ) H_{\mathcal{O}}=k\mathbb{Q}(x) H O = k Q ( x ) [26 Theorem 5.8(1)] , whence
h ( O ) = [ H O : k ] ≤ [ Q ( x ) : Q ] ≤ 16. h(\mathcal{O}) = [H_{\mathcal{O}}:k] \leq[\mathbb{Q}(x):\mathbb{Q}] \leq16. h ( O ) = [ H O : k ] ≤ [ Q ( x ) : Q ] ≤ 16.
There are finitely many imaginary quadratic orders of bounded class number, and each order determines only finitely many CM points on X X X [26 Proposition 5.6] . Thus the possible x x x , and hence the rational values t = ϖ ( x ) t=\varpi(x) t = ϖ ( x ) , form a fixed finite set. Their heights are bounded.
By Shimura’s theorem [80 §4 and Theorem 1] , E 0 E_0 E 0 is geometrically isogenous to a factor of an abelian variety A f A_f A f attached to a weight-two newform of some level m m m . For a prime q ∤ m q\nmid m q ∤ m , the two degeneracy maps exhibit A f 2 A_f^2 A f 2 as an isogeny factor of J 1 ( m q ) J_1(mq) J 1 ( m q ) [3 §3.1, (3.4)] . Thus A A A is isogenous to an abelian subvariety of J 1 ( m q ) Q J_1(mq)_{\mathbb{Q}} J 1 ( m q ) Q . Only finitely many geometric isogeny classes occur, so their levels are absorbed into C C C , since M S 0 ( t ) ≥ 1 M_{S_0}(t)\geq1 M S 0 ( t ) ≥ 1 . □
Proof of Theorem E. Put M = M S 0 ( t ) M=M_{S_0}(t) M = M S 0 ( t ) and choose A = A y A=A_y A = A y above t t t . If A A A has CM, the preceding proof of Theorem D bounds h ( t ) h(t) h ( t ) by a constant. Otherwise, Proposition 8.8 supplies an abelian surface B ⊆ J = J 1 ( N ) B\subseteq J=J_1(N) B ⊆ J = J 1 ( N ) with
N ≤ C M 2 , A ∼ K ′ B , [ K ′ : Q ] ≤ C M , N \leq CM^2,\qquad A \sim_{K'} B,\qquad[K':\mathbb{Q}] \leq CM, N ≤ C M 2 , A ∼ K ′ B , [ K ′ : Q ] ≤ C M ,
where the last bound is (8.2).
The genus of X 1 ( N ) X_1(N) X 1 ( N ) and the degree of its map to the j j j -line are at most N 2 N^2 N 2 . This map is unramified outside 0 0 0 , 1728 1728 1728 , ∞ \infty ∞ , so Javanpeykar’s bound for Belyi curves [39 Theorem 1.1.1] implies
h F ( J ) ≤ C N 12 ≤ C M 24 , dim J ≤ C N 2 . (33) h_{\mathrm{F}}(J) \leq CN^{12} \leq CM^{24},\qquad\dim J \leq CN^2. \tag*{(33)} h F ( J ) ≤ C N 12 ≤ C M 24 , dim J ≤ C N 2 . (33)
Together with (33) , Rémond’s inequality for abelian subvarieties and Bost’s lower bound [71 Corollary 1.2 and the preceding paragraph] , applied over K ′ K' K ′ , imply
h F ( B ) ≤ h F ( J ) + C dim J ≤ C M 24 . (34) h_{\mathrm{F}}(B) \leq h_{\mathrm{F}}(J)+C\dim J \leq CM^{24}. \tag*{(34)} h F ( B ) ≤ h F ( J ) + C dim J ≤ C M 24 . (34)
Both inequalities concern stable Faltings heights and are unchanged by extending the field of definition.
Write d = [ K ′ : Q ] d=[K':\mathbb{Q}] d = [ K ′ : Q ] . Since B B B has dimension two, the isogeny theorem of Gaudron and Rémond [27 Theorem 1.4] , together with (31) and (34) , supplies a K ′ K' K ′ -isogeny ϕ : B → A \phi:B\to A ϕ : B → A such that
log deg ϕ ≤ C log ( 2 d max { 1 , h F ( B ) , log d } ) ≤ C log ( 2 M ) . (35) \log\deg\phi\leq C\log(2d\max\{1,h_{\mathrm{F}}(B),\log d\}) \leq C\log(2M). \tag*{(35)} log deg ϕ ≤ C log ( 2 d max { 1 , h F ( B ) , log d }) ≤ C log ( 2 M ) . (35)
Faltings’ isogeny inequality [21 Lemma 5] and (34) –(35) therefore yield
h F ( A ) ≤ h F ( B ) + 1 2 log deg ϕ ≤ C M 24 . (36) h_{\mathrm{F}}(A) \leq h_{\mathrm{F}}(B)+\frac{1}{2}\log\deg\phi\leq CM^{24}. \tag*{(36)} h F ( A ) ≤ h F ( B ) + 2 1 log deg ϕ ≤ C M 24 . (36)
By (36) and Lemma 8.9, h ( t ) ≤ C M S 0 ( t ) 24 h(t)\leq CM_{S_0}(t)^{24} h ( t ) ≤ C M S 0 ( t ) 24 . For S ⊇ S 0 S\supseteq S_0 S ⊇ S 0 , one has M S 0 ( t ) ≤ M S ( t ) ∏ p ∈ S ∖ S 0 p M_{S_0}(t)\leq M_S(t)\prod_{p\in S\setminus S_0}p M S 0 ( t ) ≤ M S ( t ) ∏ p ∈ S ∖ S 0 p . Thus the theorem holds with S H = S 0 S_H=S_0 S H = S 0 and H S = C ∏ p ∈ S ∖ S 0 p 24 H_S=C\prod_{p\in S\setminus S_0}p^{24} H S = C ∏ p ∈ S ∖ S 0 p 24 . □
Remark 8.10. Theorems D and E can also be deduced from von Känel and Kret’s results. Let A = A y A = A_y A = A y and K K K be as in § 11.3 . In the non-CM case, [90 Proposition 7.2] applies and supplies a Q \mathbb{Q} Q -simple abelian variety B 0 B_0 B 0 of G L 2 \mathrm{GL}_2 GL 2 -type, with A A A as a geometric isogeny factor. Its dimension g ≤ 2 [ K : Q ] g \leq2[K : \mathbb{Q}] g ≤ 2 [ K : Q ] is bounded, and it has good reduction outside the primes ramified in K K K . By [89 §5.2.1 and Lemma 6.8] , B 0 B_0 B 0 is isogenous to an abelian subvariety of J 1 ( N ) J_1(N) J 1 ( N ) with
N g = N B 0 ≪ M S 0 ( t ) 2 g , N^g = N_{B_0} \ll M_{S_0}(t)^{2g}, N g = N B 0 ≪ M S 0 ( t ) 2 g ,
where N B 0 N_{B_0} N B 0 is its conductor. Together with the finite CM argument above, this proves Theorem D.
In the non-CM case, Theorem E follows from [90 Proposition 7.13] and Lemma 8.9 . Indeed, A A A has good reduction everywhere over K K K , whose degree is bounded and whose ramified primes lie in S 0 S_0 S 0 or divide M S 0 ( t ) M_{S_0}(t) M S 0 ( t ) . The CM case follows from the finiteness proved above.
Remark. Theorem E also bounds rational points on the quadratic twists
C d : y 2 = d F 5 ( t ) , C_d:\quad y^2 = dF_5(t), C d : y 2 = d F 5 ( t ) ,
F 5 ( T ) = T ( T − 13 ) ( T − 525 ) ( T − 637 ) ( T − 650 ) , (37) F_5(T) = T(T-13)(T-525)(T-637)(T-650), \tag*{(37)} F 5 ( T ) = T ( T − 13 ) ( T − 525 ) ( T − 637 ) ( T − 650 ) , (37)
where d d d is a non-zero squarefree integer. These double covers of X / W ≃ P 1 X/W \simeq\mathbb{P}^1 X / W ≃ P 1 are branched at the five points in (1) and at infinity, so have genus two. Fix S ⊇ S H S \supseteq S_H S ⊇ S H such that the branch values are integral and pairwise distinct modulo every prime outside S S S . If t t t is not a branch value and p ∉ S p \notin S p ∈ / S is a prime of odd contact, exactly one factor t − θ j t-\theta_j t − θ j has odd positive valuation; the others are units. Thus v p ( F 5 ( t ) ) v_p(F_5(t)) v p ( F 5 ( t )) is odd, and the equation implies p ∣ d p \mid d p ∣ d . Hence M S ( t ) ∣ ∣ d ∣ M_S(t) \mid|d| M S ( t ) ∣ ∣ d ∣ , and
h ( t ) ≤ C ∣ d ∣ 24 ( ( t , y ) ∈ C d ( Q ) ) , (38) h(t) \leq C|d|^{24}\qquad((t,y) \in C_d(\mathbb{Q})), \tag*{(38)} h ( t ) ≤ C ∣ d ∣ 24 (( t , y ) ∈ C d ( Q )) , (38)
with C C C independent of $d. Enlarge C C C to include the fixed branch points. On the smooth projective model, the unique point above infinity has image of height zero in P 1 \mathbb{P}^1 P 1 .
The determinant criterion
We prove Smith’s determinant identity [83 Theorem 2.2, Table 2, row 5(a)] in the case used in Theorem 3.7 . For a positive squarefree integer d d d , put
g ( d ) = # ( 2 Cl ( Q ( − d ) ) ) ( m o d 2 ) . g(d) = \#(2\operatorname{Cl}(\mathbb{Q}(\sqrt{-d}))) \pmod2. g ( d ) = # ( 2 Cl ( Q ( − d ))) ( mod 2 ) .
Proposition A.1. Let l ≡ 5 ( m o d 8 ) l \equiv5 \pmod8 l ≡ 5 ( mod 8 ) be positive and squarefree, and let A , y , z A,y,z A , y , z be as in Definition 3.4. Then, in F 2 \mathbb{F}_2 F 2 ,
det ( A + A t A t y + z A D z 0 y t + z t 0 0 ) = ∑ l = d 0 d 1 ⋯ d k d 0 ≡ 5 ( 8 ) , d i ≡ 1 ( 8 ) ( i > 0 ) ∏ i = 0 k g ( d i ) . (39) \det\begin{pmatrix}
A+A^{\mathsf{t}} & A^{\mathsf{t}} & y+z \\
A & D_z & 0 \\
y^{\mathsf{t}}+z^{\mathsf{t}} & 0 & 0
\end{pmatrix}
=
\sum_{\substack{l=d_0d_1\cdots d_k\\ d_0\equiv5\ (8),\ d_i\equiv1\ (8)\ (i>0)}}\prod_{i=0}^{k}g(d_i). \tag*{(39)} det A + A t A y t + z t A t D z 0 y + z 0 0 = l = d 0 d 1 ⋯ d k d 0 ≡ 5 ( 8 ) , d i ≡ 1 ( 8 ) ( i > 0 ) ∑ i = 0 ∏ k g ( d i ) . (39) The decompositions are unordered, with every d i > 1 d_i>1 d i > 1 and k = 0 k=0 k = 0 allowed.
Let I I I be a finite non-empty set, let y , z ∈ F 2 I y,z \in\mathbb{F}_2^I y , z ∈ F 2 I , and suppose that A 1 = 0 A\mathbf{1}=0 A 1 = 0 and A + A t = y y t + D y A+A^{\mathsf{t}}=yy^{\mathsf{t}}+D_y A + A t = y y t + D y . For ∅ ≠ S ⊆ I \varnothing\neq S \subseteq I ∅ = S ⊆ I , let A S A_S A S retain the off-diagonal entries of A A A on S S S , with diagonal chosen so that A S 1 = 0 A_S\mathbf{1}=0 A S 1 = 0 . Write A S [ U , V ] A_S[U,V] A S [ U , V ] for its submatrix on rows U U U and columns V V V . Put
Y S = ∑ i ∈ S y i , Z S = ∑ i ∈ S z i , τ S = ∑ i ∈ S z i c i ( S ) , γ S = ( 1 + Y S ) ( 1 + Z S ) τ S , Y_S=\sum_{i\in S}y_i,\qquad Z_S=\sum_{i\in S}z_i,\qquad\tau_S=\sum_{i\in S}z_ic_i(S),\qquad\gamma_S=(1+Y_S)(1+Z_S)\tau_S, Y S = i ∈ S ∑ y i , Z S = i ∈ S ∑ z i , τ S = i ∈ S ∑ z i c i ( S ) , γ S = ( 1 + Y S ) ( 1 + Z S ) τ S ,
where c i ( S ) c_i(S) c i ( S ) is the determinant obtained by deleting row i i i and any column of A S A_S A S ; it is independent of that column since A S 1 = 0 A_S\mathbf{1}=0 A S 1 = 0 . Empty determinants are 1 1 1 .
Lemma A.2. For every non-empty S ⊆ I S \subseteq I S ⊆ I ,
γ S = ∑ { U , V } a partition of S Y U Y V τ U τ V . (40) \gamma_S = \sum_{\substack{\{U,V\}\text{ a partition of }S}} Y_UY_V\tau_U\tau_V. \tag*{(40)} γ S = { U , V } a partition of S ∑ Y U Y V τ U τ V . (40)
Proof. We first recall the matrix-tree identity. For R , C ⊆ S R,C \subseteq S R , C ⊆ S of the same size, the minor obtained by deleting rows R R R and columns C C C is the sum of the weights of forests directed towards R R R , with one vertex of C C C in each component and edge weights A i j A_{ij} A ij . To see this, expand each retained row as
∑ j ≠ i A i j ( e i + e j ) , \sum_{j\ne i} A_{ij}(e_i+e_j), j = i ∑ A ij ( e i + e j ) ,
where e i e_i e i are the coordinate vectors and columns C C C are omitted. Cyclic choices have dependent rows. Every other choice is a forest rooted at R R R ; deleting leaves shows that its incidence minor is 1 1 1 exactly when each component contains one omitted column.
Both sides of (A.2) are quadratic functions of z z z , with no constant or linear terms, since z i 2 = z i z_i^2=z_i z i 2 = z i . Fix distinct i , j ∈ S i,j\in S i , j ∈ S , and write c i = c i ( S ) c_i=c_i(S) c i = c i ( S ) . The coefficient of z i z j z_i z_j z i z j on the left is
( 1 + Y S ) ( c i + c j ) . (1+Y_S)(c_i+c_j). ( 1 + Y S ) ( c i + c j ) .
By the forest identity, the coefficient on the right is
∑ a < b y a y b B a b , B a b = det A S [ S ∖ { i , j } , S ∖ { a , b } ] , \sum_{a<b}y_a y_b B_{ab},\qquad B_{ab}=\det A_S[S\setminus\{i,j\},S\setminus\{a,b\}], a < b ∑ y a y b B ab , B ab = det A S [ S ∖ { i , j } , S ∖ { a , b }] ,
where any ordering of S S S may be used. Fix h ∈ S h\in S h ∈ S , put b h = 0 b_h=0 b h = 0 and b a = B a h b_a=B_{ah} b a = B ah for a ≠ h a\ne h a = h . The row-sum relation implies B a b = b a + b b B_{ab}=b_a+b_b B ab = b a + b b : substitute the sum of the other columns for column h h h and expand. Consequently,
∑ a < b y a y b B a b = ( 1 + Y S ) ∑ a y a b a . \sum_{a<b} y_a y_b B_{ab}=(1+Y_S)\sum_a y_a b_a. a < b ∑ y a y b B ab = ( 1 + Y S ) a ∑ y a b a .
In c i + c j c_i+c_j c i + c j , delete column h h h and add the common rows to the sum of rows i , j i,j i , j . The remaining row becomes 1 t A S = ( 1 + Y S ) y S t {}^1tA_S=(1+Y_S)y_S^t 1 t A S = ( 1 + Y S ) y S t , so expansion along it yields
c i + c j = ( 1 + Y S ) ∑ a y a b a . c_i+c_j=(1+Y_S)\sum_a y_a b_a. c i + c j = ( 1 + Y S ) a ∑ y a b a .
Since ( 1 + Y S ) 2 = 1 + Y S (1+Y_S)^2=1+Y_S ( 1 + Y S ) 2 = 1 + Y S , the coefficients agree. For ∣ S ∣ = 1 |S|=1 ∣ S ∣ = 1 both sides vanish. □
For a finite set J ⊆ I J\subseteq I J ⊆ I , let
L ( J ) = ∑ π a partition of J ∏ S ∈ π γ S , L ( ∅ ) = 1. L(J)=\sum_{\substack{\pi\text{ a partition}\\\text{of }J}}\prod_{S\in\pi}\gamma_S,\qquad L(\varnothing)=1. L ( J ) = π a partition of J ∑ S ∈ π ∏ γ S , L ( ∅ ) = 1.
Lemma A.3. For any u ∈ F 2 I u\in\mathbb{F}_2^I u ∈ F 2 I , writing u ( S ) = ∑ i ∈ S u i u(S)=\sum_{i\in S}u_i u ( S ) = ∑ i ∈ S u i , one has
det ( A + A t A t u A D z 0 u t 0 0 ) = ∑ ∅ ≠ S ⊆ I u ( S ) τ S L ( I ∖ S ) . (41) \det\begin{pmatrix}
A+A^t & A^t & u\\
A & D_z & 0\\
u^t & 0 & 0
\end{pmatrix}
=
\sum_{\varnothing\ne S\subseteq I}u(S)\tau_S L(I\setminus S). \tag*{(41)} det A + A t A u t A t D z 0 u 0 0 = ∅ = S ⊆ I ∑ u ( S ) τ S L ( I ∖ S ) . (41)
Proof. In the determinant of a symmetric matrix over F 2 \mathbb{F}_2 F 2 , inverse permutations cancel unless they are involutions. Thus only matchings and diagonal entries contribute. Here the latter occur only in the second copy of I I I , with weights z i z_i z i .
Let R R R index these diagonal entries, and let C C C index the vertices in the first copy not paired with the second. Then ∣ C ∣ = ∣ R ∣ |C|=|R| ∣ C ∣ = ∣ R ∣ is odd. Summing over the pairs between copies gives the minor of A A A with rows R R R and columns C C C deleted. The last vertex pairs with some j ∈ C j\in C j ∈ C , with weight u j u_j u j . The other vertices of C C C contribute ∏ i ∈ C ∖ { j } y i \prod_{i\in C\setminus\{j\}}y_i ∏ i ∈ C ∖ { j } y i , since their number of matchings is odd. By the forest identity, each component has one root in R R R and one marked vertex in C C C . Summing over roots contributes τ S \tau_S τ S and over marked vertices contributes Y S Y_S Y S , or u ( S ) u(S) u ( S ) on the component containing j j j . Thus the determinant is the sum of
u ( S ) τ S ∏ V ≠ S Y V τ V u(S)\tau_S\prod_{V\ne S}Y_V\tau_V u ( S ) τ S V = S ∏ Y V τ V
over partitions of I I I into an odd number of components, with S S S distinguished. Pair the remaining components and apply Lemma A.2 . Since 2 m 2m 2 m components have an odd number ( 2 m − 1 ) ! ! (2m-1)!! ( 2 m − 1 )!! of pairings, the result is (A.3).
Proof of Proposition A.1. Take I = Q I=\mathbb{Q} I = Q and put d S = ∏ q ∈ S q d_S=\prod_{q\in S}q d S = ∏ q ∈ S q . The classical genus calculation [70 ] asserts that
g ( d S ) = τ S if d S ≡ 1 ( m o d 4 ) . (42) g(d_S)=\tau_S\qquad\text{if }d_S\equiv1\pmod{4}. \tag*{(42)} g ( d S ) = τ S if d S ≡ 1 ( mod 4 ) . (42)
For completeness, put K d = Q ( − d S ) K_d=\mathbb{Q}(\sqrt{-d_S}) K d = Q ( − d S ) , of discriminant − 4 d S -4d_S − 4 d S , and let p q \mathfrak p_q p q be the prime above each ramified prime q q q . An ideal representing a class of order dividing two satisfies a ‾ = ( α ) a \overline{\mathfrak a}=(\alpha)\mathfrak a a = ( α ) a with N K d / Q ( α ) = 1 N_{K_d/\mathbb{Q}}(\alpha)=1 N K d / Q ( α ) = 1 . By Hilbert 90, the class has an invariant representative. Its split and inert factors are rational principal ideals, so the ramified primes generate Cl ( K d ) [ 2 ] \operatorname{Cl}(K_d)[2] Cl ( K d ) [ 2 ] . Their only relation is
∏ q ∈ S p q = ( − d S ) . \prod_{q\in S}\mathfrak p_q=(\sqrt{-d_S}). q ∈ S ∏ p q = ( − d S ) .
Indeed, a generator x x x of a principal product satisfies x ‾ / x = ± 1 \overline{x}/x=\pm1 x / x = ± 1 , since K d K_d K d has only the units ± 1 \pm1 ± 1 . Thus x ∈ Q x\in\mathbb{Q} x ∈ Q or Q − d S \mathbb{Q}\sqrt{-d_S} Q − d S , giving the trivial or the displayed relation. Hence the classes are parametrized by
F 2 ϵ ∏ q ∈ S p q a q , ( ϵ , a ) ∈ F 2 × ( F 2 S / ⟨ 1 ⟩ ) . \mathbb{F}_2^\epsilon\prod_{q\in S}\mathfrak p_q^{a_q},\qquad(\epsilon,a)\in\mathbb{F}_2\times(\mathbb{F}_2^S/\langle1\rangle). F 2 ϵ q ∈ S ∏ p q a q , ( ϵ , a ) ∈ F 2 × ( F 2 S / ⟨ 1 ⟩) .
A class [ a ] [\mathfrak a] [ a ] is a square precisely when N ( a ) N(\mathfrak a) N ( a ) is a norm from K d K_d K d . The forward implication follows by taking norms. Conversely, if N K d / Q ( α ) = N ( a ) N_{K_d/\mathbb{Q}}(\alpha)=N(\mathfrak a) N K d / Q ( α ) = N ( a ) , then ( α ) a − 1 (\alpha)\mathfrak a^{-1} ( α ) a − 1 has norm one, hence equals b / b ‾ \mathfrak b/\overline{\mathfrak b} b / b ; its class is [ b ] 2 [\mathfrak b]^2 [ b ] 2 . For the representatives above, the local norm conditions at the odd ramified primes are
A S a + ϵ z S = 0 ; A_Sa+\epsilon z_S=0; A S a + ϵ z S = 0 ;
at q ∈ S q\in S q ∈ S this is the condition
( 2 q ) ϵ ∏ q ′ ∈ S ∖ { q } ( q ′ q ) a q ′ + a q = 1. \left(\frac{2}{q}\right)^\epsilon\prod_{q'\in S\setminus\{q\}}\left(\frac{q'}{q}\right)^{a_{q'}+a_q}=1. ( q 2 ) ϵ q ′ ∈ S ∖ { q } ∏ ( q q ′ ) a q ′ + a q = 1.
The condition at 2 2 2 follows by Hilbert reciprocity; all other local conditions are automatic. The Hasse principle for the norm conic makes them sufficient. Set one coordinate of a a a to zero. The resulting matrix is A S A_S A S with that column replaced by z S z_S z S , and its determinant is τ S \tau_S τ S . Its kernel is trivial exactly when Cl ( K d ) [ 2 ] ∩ 2 Cl ( K d ) = 0 \operatorname{Cl}(K_d)[2]\cap2\operatorname{Cl}(K_d)=0 Cl ( K d ) [ 2 ] ∩ 2 Cl ( K d ) = 0 , or equivalently g ( d S ) = 1 g(d_S)=1 g ( d S ) = 1 . This proves (A.4).
Thus γ S = g ( d S ) \gamma_S=g(d_S) γ S = g ( d S ) for d S ≡ 1 ( m o d 8 ) d_S\equiv1\pmod{8} d S ≡ 1 ( mod 8 ) and vanishes otherwise. Take u = y + z u=y+z u = y + z . If Y S = 1 Y_S=1 Y S = 1 , both row and column sums of A S A_S A S vanish, so its cofactors are equal. Hence τ S = Z S c i ( S ) \tau_S=Z_Sc_i(S) τ S = Z S c i ( S ) and
u ( S ) τ S = ( 1 + Z S ) Z S c i ( S ) = 0. u(S)\tau_S=(1+Z_S)Z_Sc_i(S)=0. u ( S ) τ S = ( 1 + Z S ) Z S c i ( S ) = 0.
If Y S = 0 Y_S=0 Y S = 0 , the same expression is Z S τ S Z_S\tau_S Z S τ S . Hence
u ( S ) τ S = { g ( d S ) , d S ≡ 5 ( m o d 8 ) , 0 , otherwise . u(S)\tau_S=
\begin{cases}
g(d_S),&d_S\equiv5\pmod{8},\\
0,&\text{otherwise}.
\end{cases} u ( S ) τ S = { g ( d S ) , 0 , d S ≡ 5 ( mod 8 ) , otherwise .
Substitution in Lemma A.3 proves (A.1).
The right side of (A.1) is the first genus sum in [87 Theorem 1.4] . Its non-vanishing implies L ′ ( E l , 1 ) ≠ 0 L'(E_l,1)\ne0 L ′ ( E l , 1 ) = 0 , proving Theorem 3.7 .
References
Department of Mathematics, The University of Texas at Austin, 2515 Speedway, Stop C1200, Austin, TX 78712, USA
Email address: ashayk@utexas.edu