Introduction
Background
Twisting a constant abelian variety by a finite extension of a function field is a classical way of producing abelian varieties with large Mordell–Weil rank; see for instance Tate and Shafarevich [26 ] and Ulmer [27 ] . In [17 ] the author considered the following situation, with the aim of extending a result of Hazama [6 ] on quadratic twists to cyclic covers of arbitrary degree. Let k k k be a global field of characteristic prime to n n n containing a primitive n n n -th root of unity, let π : X ′ → X \pi:X'\to X π : X ′ → X be a cyclic cover of degree n n n of varieties over k k k with X ′ ( k ) ≠ ∅ X'(k)\ne\varnothing X ′ ( k ) = ∅ , let L = k ( X ′ ) L=k(X') L = k ( X ′ ) and K = k ( X ) K=k(X) K = k ( X ) , and let A ~ \widetilde{A} A be the twist of an abelian variety A A A over k k k by L / K L/K L / K and an automorphism σ \sigma σ of A A A of order n n n . Then [17 Theorem 1.1] asserts that
A ~ ( K ) ≅ Hom k ( Prym X ′ / X , A ) ⊕ A [ n ] ( k ) , \widetilde{A}(K)\cong\operatorname{Hom}_{k}(\operatorname{Prym}_{X'/X},A)\oplus A[n](k), A ( K ) ≅ Hom k ( Prym X ′ / X , A ) ⊕ A [ n ] ( k ) ,
and that rk A ~ ( K ) ≥ m ⋅ rk End k ( A ) \operatorname{rk}\widetilde{A}(K)\ge m\cdot\operatorname{rk}\operatorname{End}_{k}(A) rk A ( K ) ≥ m ⋅ rk End k ( A ) whenever Prym X ′ / X \operatorname{Prym}_{X'/X} Prym X ′ / X is k k k -isogenous to A m × B A^m\times B A m × B , with conditions on B B B . In the corrigendum [19 ] the varieties were restricted to be smooth and projective, the conditions on the dimension of B B B were removed, and the inequality was replaced by the equality rk A ~ ( K ) = m ⋅ rk End k ( A ) \operatorname{rk}\widetilde{A}(K)=m\cdot\operatorname{rk}\operatorname{End}_{k}(A) rk A ( K ) = m ⋅ rk End k ( A ) ; this amended statement is also [18 Theorem 3.3] . In the same way, [19 ] replaced the inequality in [17 Theorem 1.2] , which applies the structure theorem to the superelliptic curves y n = f ( x ) y^n=f(x) y n = f ( x ) , by an equality. In [18 ] the structure theorem was applied to the cyclic multiple planes w n = f ( x , y ) w^n=f(x,y) w n = f ( x , y ) and to the varieties V m V_m V m obtained as quotients of m m m -fold products of such a surface, with the aim of producing abelian varieties of large rank over the function fields K m = k ( V m ) K_m = k(V_m) K m = k ( V m ) . For curves this construction goes back to Yamagishi [29 ] ; see [17 Remark 5.1] .
The displayed statement is not correct, neither in its original form nor in the form of [19 ] . A K K K -rational point of A ~ \widetilde{A} A is a morphism X ′ → A X' \to A X ′ → A that intertwines the covering automorphism with σ \sigma σ , and only the homomorphisms Prym X ′ / X → A \operatorname{Prym}_{X'/X} \to A Prym X ′ / X → A with the same intertwining property contribute; the torsion part is also different. In the proof in [17 §4] the automorphism σ \sigma σ does not enter the condition describing the points of the twist (Remark 2.5). Example 2.7, with A A A the elliptic curve y 2 = x 3 + 1 y^2 = x^3 + 1 y 2 = x 3 + 1 over a number field containing ζ 3 \zeta_3 ζ 3 , satisfies the hypotheses of [17 Theorem 1.1] both before and after [19 ] , and contradicts its conclusion. The error propagates to [17 Theorem 1.2] (Example 3.9) and to the main results of [18 ] : the ring End k ( Alb ( X n ) ) \operatorname{End}_k(\operatorname{Alb}(X_n)) End k ( Alb ( X n )) appearing there has to be replaced by the centralizer of the covering automorphism, several of the constants in [18 Theorem 1.3] change, and some of the arguments need to be replaced.
The present paper replaces these statements by equivariant versions and gives complete proofs of the corrected results. Section 7 records explicitly the changes to [17 ] , [18 ] and [19 ] . We also develop the arithmetic over number fields, which was only sketched in [18 ] .
Results
Let G G G be a finite group acting faithfully on a smooth, projective, geometrically integral variety X ′ X' X ′ over a field k k k , put L = k ( X ′ ) L = k(X') L = k ( X ′ ) , K = L G K = L^G K = L G , and let ρ : G → Aut k ( A ) \rho: G \to\operatorname{Aut}_k(A) ρ : G → Aut k ( A ) be an action of G G G on an abelian variety A A A over k k k by automorphisms of abelian varieties. Let A ~ \widetilde{A} A be the twist of A A A by L / K L/K L / K and ρ \rho ρ (normalized as in Section 2). Fix x 0 ∈ X ′ ( k ) x_0 \in X'(k) x 0 ∈ X ′ ( k ) , let α : X ′ → Alb ( X ′ ) \alpha: X' \to\operatorname{Alb}(X') α : X ′ → Alb ( X ′ ) be the Albanese map with α ( x 0 ) = 0 \alpha(x_0) = 0 α ( x 0 ) = 0 , and let g ↦ g ~ g \mapsto\widetilde{g} g ↦ g be the induced action of G G G on Alb ( X ′ ) \operatorname{Alb}(X') Alb ( X ′ ) .
Theorem A (Theorem 2.2). There is an exact sequence
0 ⟶ A ( k ) G ⟶ A ~ ( K ) ⟶ Hom k ( Alb ( X ′ ) , A ) G → δ H 1 ( G , A ( k ) ) , 0 \longrightarrow A(k)^G \longrightarrow\widetilde{A}(K) \longrightarrow\operatorname{Hom}_k(\operatorname{Alb}(X'), A)^G \xrightarrow{\delta} H^1(G, A(k)), 0 ⟶ A ( k ) G ⟶ A ( K ) ⟶ Hom k ( Alb ( X ′ ) , A ) G δ H 1 ( G , A ( k )) , where Hom k ( Alb ( X ′ ) , A ) G \operatorname{Hom}_k(\operatorname{Alb}(X'), A)^G Hom k ( Alb ( X ′ ) , A ) G is the group of homomorphisms φ \varphi φ with φ ∘ g ~ = ρ ( g ) ∘ φ \varphi\circ\widetilde{g} = \rho(g) \circ\varphi φ ∘ g = ρ ( g ) ∘ φ for all g ∈ G g \in G g ∈ G . The image of δ \delta δ is killed by ∣ G ∣ |G| ∣ G ∣ , and δ = 0 \delta= 0 δ = 0 if x 0 x_0 x 0 is fixed by G G G , in which case the sequence splits. In particular
rk A ~ ( K ) = rk A ( k ) G + rk Z Hom k ( Alb ( X ′ ) , A ) G , \operatorname{rk} \widetilde{A}(K) = \operatorname{rk} A(k)^G + \operatorname{rk}_{\mathbb{Z}} \operatorname{Hom}_k(\operatorname{Alb}(X'), A)^G, rk A ( K ) = rk A ( k ) G + rk Z Hom k ( Alb ( X ′ ) , A ) G , A ~ ( K ) t o r s = ( A ( k ) G ) t o r s . \widetilde{A}(K)_{\mathrm{tors}} = \left(A(k)^G\right)_{\mathrm{tors}}. A ( K ) tors = ( A ( k ) G ) tors .
Now let n ≥ 2 n \ge2 n ≥ 2 , let k k k be a field of characteristic prime to n n n containing a primitive n n n -th root of unity, and let f ∈ k [ x , y ] f \in k[x,y] f ∈ k [ x , y ] be such that w n = f ( x , y ) w^n = f(x,y) w n = f ( x , y ) defines a geometrically irreducible surface. Let X n X_n X n be a smooth projective model, τ ˉ \bar{\tau} τ ˉ the automorphism induced by w ↦ ζ n w w \mapsto\zeta_n w w ↦ ζ n w , and σ = τ ˉ ∗ ∈ Aut k ( A ) \sigma= \bar{\tau}_* \in\operatorname{Aut}_k(A) σ = τ ˉ ∗ ∈ Aut k ( A ) . For m ≥ 1 m \ge1 m ≥ 1 let U m = X n m U_m = X_n^m U m = X n m , let V m V_m V m be the quotient of U m U_m U m by the diagonal action of τ ˉ \bar{\tau} τ ˉ , and put K m = k ( V m ) K_m = k(V_m) K m = k ( V m ) . For an abelian variety B B B over k k k with an automorphism β \beta β such that β n = id \beta^n = \operatorname{id} β n = id , let B ~ m \widetilde{B}_m B m be the corresponding twist of B B B over K m K_m K m ; for ( B , β ) = ( A , σ ) (B,\beta) = (A,\sigma) ( B , β ) = ( A , σ ) this is the Albanese variety of the twist X ~ n \widetilde{X}_n X n considered in [18 ] .
Theorem B (Theorem 3.4). Assume that X n ( k ) ≠ ∅ X_n(k) \ne\varnothing X n ( k ) = ∅ . Then 1 + σ + ⋯ + σ n − 1 = 0 1 + \sigma+ \cdots+ \sigma^{n-1} = 0 1 + σ + ⋯ + σ n − 1 = 0 , and
rk B ~ m ( K m ) = rk B β ( k ) + m ⋅ rk Z Hom k ( A , B ) ( σ , β ) , \operatorname{rk} \widetilde{B}_m(K_m) = \operatorname{rk} B^\beta(k) + m \cdot\operatorname{rk}_{\mathbb{Z}} \operatorname{Hom}_k(A,B)^{(\sigma,\beta)}, rk B m ( K m ) = rk B β ( k ) + m ⋅ rk Z Hom k ( A , B ) ( σ , β ) , B ~ m ( K m ) t o r s = B β ( k ) t o r s . \widetilde{B}_m(K_m)_{\mathrm{tors}} = B^\beta(k)_{\mathrm{tors}}. B m ( K m ) tors = B β ( k ) tors .
where B β = ker ( 1 − β ) B^\beta=\ker(1-\beta) B β = ker ( 1 − β ) and Hom k ( A , B ) ( σ , β ) = { φ : φ ∘ σ = β ∘ φ } \operatorname{Hom}_k(A,B)^{(\sigma,\beta)}=\{\varphi:\varphi\circ\sigma=\beta\circ\varphi\} Hom k ( A , B ) ( σ , β ) = { φ : φ ∘ σ = β ∘ φ } . In particular
rk Alb ( X ~ n ) ( K m ) = m ⋅ rk Z End k ( A ) σ , Alb ( X ~ n ) ( K m ) t o r s = A σ ( k ) , \operatorname{rk}\operatorname{Alb}(\widetilde{X}_n)(K_m)=m\cdot\operatorname{rk}_{\mathbb{Z}}\operatorname{End}_k(A)^\sigma,\qquad\operatorname{Alb}(\widetilde{X}_n)(K_m)_{\mathrm{tors}}=A^\sigma(k), rk Alb ( X n ) ( K m ) = m ⋅ rk Z End k ( A ) σ , Alb ( X n ) ( K m ) tors = A σ ( k ) ,
where End k ( A ) σ \operatorname{End}_k(A)^\sigma End k ( A ) σ is the centralizer of σ \sigma σ in End k ( A ) \operatorname{End}_k(A) End k ( A ) . If X n X_n X n has a k k k -rational point fixed by τ ‾ \overline{\tau} τ , then Alb ( X ~ n ) ( K m ) \operatorname{Alb}(\widetilde{X}_n)(K_m) Alb ( X n ) ( K m ) is the direct sum of A σ ( k ) A^\sigma(k) A σ ( k ) and a free left End k ( A ) σ \operatorname{End}_k(A)^\sigma End k ( A ) σ -module of rank m m m with an explicit basis.
In [18 ] the rank was claimed to be m ⋅ rk End k ( A ) m\cdot\operatorname{rk}\operatorname{End}_k(A) m ⋅ rk End k ( A ) . The two agree if and only if End k ( A ) σ \operatorname{End}_k(A)^\sigma End k ( A ) σ has finite index in End k ( A ) \operatorname{End}_k(A) End k ( A ) ; for instance, for n = 8 n=8 n = 8 and n = 12 n=12 n = 12 under the hypotheses of [18 Theorem 1.3] , this fails whenever d n > 0 d_n>0 d n > 0 and k k k is large enough for all endomorphisms of A k ‾ A_{\overline{k}} A k to be defined over k k k (Remark 4.6). Theorem B has the same form for cyclic covers w n = f ( u 1 , … , u ℓ ) w^n=f(u_1,\ldots,u_\ell) w n = f ( u 1 , … , u ℓ ) of P ℓ \mathbb{P}^{\ell} P ℓ whenever a smooth projective model carrying the induced cyclic action is available; in particular, this causes no difficulty in characteristic zero. For ℓ = 1 \ell=1 ℓ = 1 it gives the corrected form of [17 Theorem 1.2] on superelliptic curves (Corollary 3.8), and Example 3.9 shows that the torsion subgroup given there is not correct.
Over the complex numbers the centralizer is controlled by the eigenspace decomposition of H 1 ( A , Q ) H^1(A,\mathbb{Q}) H 1 ( A , Q ) under σ ( S e c t i o n 4 ) \sigma(Section 4) σ ( S ec t i o n 4 ) . For d ∣ n d\mid n d ∣ n let e d e_d e d be the dimension over Q ( ζ d ) \mathbb{Q}(\zeta_d) Q ( ζ d ) of the part of H 1 ( A , Q ) H^1(A,\mathbb{Q}) H 1 ( A , Q ) on which σ \sigma σ acts through primitive d d d -th roots of unity. The following theorem summarizes Corollary 4.4, Theorem 5.1 and Proposition 5.3.
Theorem C. Let k ⊂ C k\subset\mathbb{C} k ⊂ C be a number field containing ζ n \zeta_n ζ n , and assume X n ( k ) ≠ ∅ X_n(k)\ne\varnothing X n ( k ) = ∅ .
(i) m ∑ d ∣ n e d > 0 ϕ ( d ) ≤ rk Alb ( X ~ n ) ( K m ) ≤ m ∑ d ∣ n ϕ ( d ) e d 2 m\sum_{\substack{d\mid n\\e_d>0}}\phi(d)\leq\operatorname{rk}\operatorname{Alb}(\widetilde{X}_n)(K_m)\leq m\sum_{d\mid n}\phi(d)e_d^2 m ∑ d ∣ n e d > 0 ϕ ( d ) ≤ rk Alb ( X n ) ( K m ) ≤ m ∑ d ∣ n ϕ ( d ) e d 2 .
(ii) The upper bound is attained if and only if ( A , σ ) (A,\sigma) ( A , σ ) is rigid in the sense of Definition 4.1 and every endomorphism of A k ‾ A_{\overline{k}} A k commuting with σ \sigma σ is defined over k k k ; the latter holds after a finite extension of k k k . Under Assumption 1.2 of [18 ] the upper bound equals m ⋅ 4 d n 2 / ϕ ( n ) m\cdot4d_n^2/\phi(n) m ⋅ 4 d n 2 / ϕ ( n ) , where d n = dim A d_n=\dim A d n = dim A .
(iii) A σ ( k ‾ ) A^\sigma(\overline{k}) A σ ( k ) is a finite group of order ∏ p r ∣ ∣ n , p r > 1 p e p r r \prod_{p^r\mid\mid n,\ p^r>1}p^{e_{p^r}r} ∏ p r ∣∣ n , p r > 1 p e p r r , the product running over the prime powers p r > 1 p^r>1 p r > 1 dividing n n n . In particular Alb ( X ~ n ) ( K m ) \operatorname{Alb}(\widetilde{X}_n)(K_m) Alb ( X n ) ( K m ) is torsion free if e d = 0 e_d=0 e d = 0 for every prime power d d d dividing $n.
Part (ii) replaces [18 Theorem 1.3] : the constant there must be 4 d n 2 / ϕ ( n ) 4d_n^2/\phi(n) 4 d n 2 / ϕ ( n ) rather than 2 d n 2 2d_n^2 2 d n 2 for n = 5 , 10 n=5,10 n = 5 , 10 , and the equalities claimed there hold only when the relevant endomorphisms are defined over k k k . Section 5 also contains a specialization statement (Proposition 5.5): for m = 1 m=1 m = 1 the base V 1 V_1 V 1 is rational, and Néron’s specialization theorem produces, outside a thin set of ( a , b ) ∈ k 2 (a,b)\in k^2 ( a , b ) ∈ k 2 , specializations attached to the μ n \mu_n μ n -torsors w n = f ( a , b ) w^n=f(a,b) w n = f ( a , b ) whose Mordell–Weil rank over k k k is at least rk End k ( A ) σ \operatorname{rk}\operatorname{End}_k(A)^\sigma rk End k ( A ) σ . For m ≥ 2 m\geq2 m ≥ 2 the corresponding assertion in [18 Remark 1.5(2)] is not justified.
The dual of the Fermat cubic provides a concrete example in which the corrected theory can be worked out explicitly. Let
F ( x , y ) = x 6 + y 6 + 1 − 2 ( x 3 y 3 + x 3 + y 3 ) , F(x,y)=x^6+y^6+1-2(x^3y^3+x^3+y^3), F ( x , y ) = x 6 + y 6 + 1 − 2 ( x 3 y 3 + x 3 + y 3 ) ,
so that F = 0 F=0 F = 0 is an affine equation of the dual of the Fermat cubic, a sextic with nine cusps.
Theorem D (Theorem 6.1). **
(i) The curve Y 2 = X 3 + 16 F ( x , y ) Y^2=X^3+16F(x,y) Y 2 = X 3 + 16 F ( x , y ) has rank 3 3 3 over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) and rank 6 6 6 over Q ( ζ 3 ) ( x , y ) \mathbb{Q}(\zeta_3)(x,y) Q ( ζ 3 ) ( x , y ) ; the points ( 4 x y , 4 ( x 3 + y 3 − 1 ) ) (4xy,4(x^3+y^3-1)) ( 4 x y , 4 ( x 3 + y 3 − 1 )) , ( 4 y , 4 ( y 3 + 1 − x 3 ) ) (4y,4(y^3+1-x^3)) ( 4 y , 4 ( y 3 + 1 − x 3 )) , ( 4 x , 4 ( x 3 + 1 − y 3 ) ) (4x,4(x^3+1-y^3)) ( 4 x , 4 ( x 3 + 1 − y 3 )) generate a subgroup of finite index of its Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) -rational points.
(ii) The curve Y 2 = X 3 + F ( x , y ) Y^2=X^3+F(x,y) Y 2 = X 3 + F ( x , y ) has rank 0 0 0 over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) and over Q ( ζ 3 ) ( x , y ) \mathbb{Q}(\zeta_3)(x,y) Q ( ζ 3 ) ( x , y ) , rank 3 3 3 over Q ( 2 3 ) ( x , y ) \mathbb{Q}(\sqrt[3]{2})(x,y) Q ( 3 2 ) ( x , y ) , and rank 6 6 6 over Q ( ζ 3 , 2 3 ) ( x , y ) \mathbb{Q}(\zeta_3,\sqrt[3]{2})(x,y) Q ( ζ 3 , 3 2 ) ( x , y ) .
(iii) Let A = Alb ( X 6 ) A = \operatorname{Alb}(X_6) A = Alb ( X 6 ) for w 6 6 = F ( x , y ) w_6^6 = F(x,y) w 6 6 = F ( x , y ) . Then dim A = 3 \dim A = 3 dim A = 3 , Assumption 1.2 of [18 ] holds, A A A is isogenous over Q ( ζ 3 ) \mathbb{Q}(\zeta_3) Q ( ζ 3 ) to the cube of y 2 = x 3 + 16 y^2 = x^3 + 16 y 2 = x 3 + 16 , and for every number field k ⊇ Q ( ζ 3 ) k \supseteq\mathbb{Q}(\zeta_3) k ⊇ Q ( ζ 3 ) and every m ≥ 1 m \ge1 m ≥ 1 the group Alb ( X ~ 6 ) ( k ( V m ) ) \operatorname{Alb}(\widetilde{X}_6)(k(V_m)) Alb ( X 6 ) ( k ( V m )) is free of rank 18 m 18m 18 m .
(iv) For all ( a , b ) ∈ Q 2 (a,b) \in\mathbb{Q}^2 ( a , b ) ∈ Q 2 outside a thin set, the Mordell curve y 2 = x 3 + 16 F ( a , b ) y^2 = x^3 + 16F(a,b) y 2 = x 3 + 16 F ( a , b ) has rank at least 3 over Q \mathbb{Q} Q . The exceptional set is not empty.
Part (iii) shows that the rank formula of [18 Theorem 1.4] does hold for the dual of the Fermat cubic, although the argument given there is incomplete. Part (ii) illustrates the role of the field of definition: over Q ( ζ 3 ) \mathbb{Q}(\zeta_3) Q ( ζ 3 ) the sextic F F F has no quasi-toric decomposition F = G 2 + H 3 F = G^2 + H^3 F = G 2 + H 3 , whereas over Q ( 2 3 ) \mathbb{Q}(\sqrt[3]{2}) Q ( 3 2 ) it has three independent ones.
Relation to other work
Results relating Mordell–Weil groups of twists, or of isotrivial families, to groups of homomorphisms appear in several places; see for instance Hazama [6 ] , Wang [28 ] , Libgober [12 ] , and, for Jacobians over k ( t ) k(t) k ( t ) , Ulmer [27 ] . For k = C k = \mathbb{C} k = C and m = 1 m = 1 m = 1 the identification of the Mordell–Weil group with a group of equivariant morphisms is the starting point of the work of Cogolludo-Agustín and Libgober [3 ] on the elliptic threefolds u 2 + v 3 = F ( x , y , 1 ) u^2+v^3 = F(x,y,1) u 2 + v 3 = F ( x , y , 1 ) ; Libgober [13 ] studies more generally isotrivial abelian varieties over C ( x , y ) \mathbb{C}(x,y) C ( x , y ) with cyclic holonomy. In [3 ] the rank is computed from the Alexander polynomial of the curve F = 0 F = 0 F = 0 when this curve has only nodes and cusps. See also Hulek–Kloosterman [7 ] and Kloosterman [8 , 9 ] . Theorem A is a version of this principle over an arbitrary field and for an arbitrary finite group, including the obstruction δ \delta δ that appears when the cover has no rational fixed point. We include complete proofs because the statements in [17 ] and [18 ] have to be replaced. The observation that the growth of the rank in m m m is formal (Proposition 3.6) shows that the arithmetic content of the construction of [18 ] lies in the case m = 1 m = 1 m = 1 ; this is why the second half of the paper concentrates on the field of definition, torsion and specialization.
Conventions
Varieties are separated schemes of finite type over a field; points of a variety over a field F F F are F F F -rational points. For an abelian variety A A A over k k k we write End k ( A ) \operatorname{End}_k(A) End k ( A ) for the ring of endomorphisms defined over k k k . For any abelian group M M M we use
rk M : = dim Q ( M ⊗ Z Q ) . \operatorname{rk} M := \dim_{\mathbb{Q}}(M \otimes_{\mathbb{Z}} \mathbb{Q}). rk M := dim Q ( M ⊗ Z Q ) .
This agrees with the usual rank when M M M is finitely generated. For an automorphism σ \sigma σ of A A A we put A σ = ker ( 1 − σ ) A^\sigma= \ker(1-\sigma) A σ = ker ( 1 − σ ) , a closed subgroup scheme of A A A . We write φ \varphi φ for Euler’s function and Φ d \Phi_d Φ d for the d d d -th cyclotomic polynomial; the letter φ \varphi φ is reserved for homomorphisms. For abelian varieties A A A , B B B over k k k with automorphisms σ \sigma σ , β \beta β ,
Hom k ( A , B ) ( σ , β ) = { φ ∈ Hom k ( A , B ) : φ ∘ σ = β ∘ φ } , \operatorname{Hom}_k(A,B)^{(\sigma,\beta)} = \{\varphi\in\operatorname{Hom}_k(A,B): \varphi\circ\sigma= \beta\circ\varphi\}, Hom k ( A , B ) ( σ , β ) = { φ ∈ Hom k ( A , B ) : φ ∘ σ = β ∘ φ } ,
End k ( A ) σ = Hom k ( A , A ) ( σ , σ ) . \operatorname{End}_k(A)^\sigma= \operatorname{Hom}_k(A,A)^{(\sigma,\sigma)}. End k ( A ) σ = Hom k ( A , A ) ( σ , σ ) .
We use the notation of [18 ] whenever possible. References to numbered statements of [17 ] , [18 ] and [19 ] always carry the citation.
Twists by Galois covers
Conventions on twists
Let L / K L/K L / K be a finite Galois extension and let Y Y Y be a quasi-projective K K K -variety. An element u ∈ Gal ( L / K ) u \in\operatorname{Gal}(L/K) u ∈ Gal ( L / K ) acts on Y ( L ) Y(L) Y ( L ) , and on morphisms defined over L L L , through its action on coordinates; we write u P ^uP u P and u ϕ ^u\phi u ϕ . If a = ( a u ) a=(a_u) a = ( a u ) is a 1-cocycle of Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) with values in Aut ( Y L ) \operatorname{Aut}(Y_L) Aut ( Y L ) and g : Y L → Y ~ L g:Y_L\to\widetilde{Y}_L g : Y L → Y L is the isomorphism of [18 Prop. 3.1] (see [1 ] ), so that u g = g ∘ a u u_g=g\circ a_u u g = g ∘ a u , then u ( g ( P ) ) = g ( a u ( u P ) ) u(g(P))=g(a_u({}^{u}P)) u ( g ( P )) = g ( a u ( u P )) for P ∈ Y ( L ) P\in Y(L) P ∈ Y ( L ) , and therefore
Y ~ ( K ) = { g ( P ) : P ∈ Y ( L ) , a u ( u P ) = P for all u ∈ Gal ( L / K ) } . (1) \widetilde{Y}(K)=\{g(P):P\in Y(L),\ a_u({}^{u}P)=P\text{ for all }u\in\operatorname{Gal}(L/K)\}. \tag*{(1)} Y ( K ) = { g ( P ) : P ∈ Y ( L ) , a u ( u P ) = P for all u ∈ Gal ( L / K )} . (1)
Setting 2.1. Let k k k be a field, X ′ X' X ′ a smooth, projective, geometrically integral variety over k k k , and G G G a finite group acting faithfully on X ′ X' X ′ by k k k -automorphisms. Put L = k ( X ′ ) L=k(X') L = k ( X ′ ) and K = L G K=L^G K = L G . For g ∈ G g\in G g ∈ G let u g u_g u g be the automorphism h ↦ h ∘ g − 1 h\mapsto h\circ g^{-1} h ↦ h ∘ g − 1 of L L L . Since u g u g ′ = u g g ′ u_g u_{g'}=u_{gg'} u g u g ′ = u g g ′ and G G G acts faithfully on L L L , the map g ↦ u g g\mapsto u_g g ↦ u g is an isomorphism G → Gal ( L / K ) G\to\operatorname{Gal}(L/K) G → Gal ( L / K ) . Let A A A be an abelian variety over k k k and ρ : G → Aut k ( A ) \rho:G\to\operatorname{Aut}_k(A) ρ : G → Aut k ( A ) a homomorphism into the group of automorphisms of A A A as an abelian variety. Since the elements of Aut k ( A ) \operatorname{Aut}_k(A) Aut k ( A ) are fixed by Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) , the map u g ↦ ρ ( g ) u_g\mapsto\rho(g) u g ↦ ρ ( g ) is a 1-cocycle, and we denote by A ~ \widetilde{A} A the corresponding twist of A K A_K A K ; it exists because A A A is projective, and it is an abelian variety over K K K .
If Y Y Y is a variety over k k k , the set Y ( L ) Y(L) Y ( L ) is the set of rational maps X ′ ⇢ Y X'\dashrightarrow Y X ′ ⇢ Y defined over k k k . If f P f_P f P denotes the rational map corresponding to P ∈ Y ( L ) P\in Y(L) P ∈ Y ( L ) , then u g P u_gP u g P corresponds to f P ∘ g − 1 f_P\circ g^{-1} f P ∘ g − 1 , because the coordinates of P P P are functions h ∈ L h\in L h ∈ L and u g ( h ) = h ∘ g − 1 u_g(h)=h\circ g^{-1} u g ( h ) = h ∘ g − 1 . By (2.1), if Y Y Y carries an action ρ \rho ρ of G G G and Y ~ \widetilde{Y} Y is the twist by u g ↦ ρ ( g ) u_g\mapsto\rho(g) u g ↦ ρ ( g ) , then
Y ~ ( K ) ⟷ { f : X ′ ⇢ Y defined over k : f ∘ g = ρ ( g ) ∘ f for all g ∈ G } . (2) \widetilde{Y}(K)\longleftrightarrow\{f:X'\dashrightarrow Y\text{ defined over }k:f\circ g=\rho(g)\circ f\text{ for all }g\in G\}. \tag*{(2)} Y ( K ) ⟷ { f : X ′ ⇢ Y defined over k : f ∘ g = ρ ( g ) ∘ f for all g ∈ G } . (2)
For Y = A Y=A Y = A this bijection is an isomorphism of groups, and every rational map X ′ ⇢ A X'\dashrightarrow A X ′ ⇢ A is a morphism because X ′ X' X ′ is smooth [14 §3] . We always use this normalization; with the opposite identification of G G G with Gal ( L / K ) \operatorname{Gal}(L/K) Gal ( L / K ) , ρ \rho ρ has to be replaced by g ↦ ρ ( g − 1 ) g\mapsto\rho(g^{-1}) g ↦ ρ ( g − 1 ) , which for cyclic G G G amounts to replacing σ \sigma σ by σ − 1 \sigma^{-1} σ − 1 .
Assume that X ′ ( k ) ≠ ∅ X'(k)\ne\varnothing X ′ ( k ) = ∅ and fix x 0 ∈ X ′ ( k ) x_0\in X'(k) x 0 ∈ X ′ ( k ) . Let α : X ′ → Alb ( X ′ ) \alpha:X'\to\operatorname{Alb}(X') α : X ′ → Alb ( X ′ ) be the Albanese map with α ( x 0 ) = 0 \alpha(x_0)=0 α ( x 0 ) = 0 [20 ] . By the universal property, every morphism f : X ′ → A f:X'\to A f : X ′ → A defined over k k k can be written uniquely as
f = c + φ ∘ α , c = f ( x 0 ) ∈ A ( k ) , φ ∈ Hom k ( Alb ( X ′ ) , A ) . (3) f=c+\varphi\circ\alpha,\qquad c=f(x_0)\in A(k),\quad\varphi\in\operatorname{Hom}_k(\operatorname{Alb}(X'),A). \tag*{(3)} f = c + φ ∘ α , c = f ( x 0 ) ∈ A ( k ) , φ ∈ Hom k ( Alb ( X ′ ) , A ) . (3)
For g ∈ G g\in G g ∈ G let g ~ ∈ Aut k ( Alb ( X ′ ) ) \widetilde{g}\in\operatorname{Aut}_k(\operatorname{Alb}(X')) g ∈ Aut k ( Alb ( X ′ )) be the automorphism defined by α ∘ g = g ~ ∘ α + α ( g ( x 0 ) ) \alpha\circ g=\widetilde{g}\circ\alpha+\alpha(g(x_0)) α ∘ g = g ∘ α + α ( g ( x 0 )) ; then g ↦ g ~ g\mapsto\widetilde{g} g ↦ g is a homomorphism and does not depend on x 0 x_0 x 0 . We write
A ( k ) G = { c ∈ A ( k ) : ρ ( g ) c = c for all g ∈ G } , A(k)^G=\{c\in A(k):\rho(g)c=c\text{ for all }g\in G\}, A ( k ) G = { c ∈ A ( k ) : ρ ( g ) c = c for all g ∈ G } ,
Hom k ( Alb ( X ′ ) , A ) G = { φ : φ ∘ g ~ = ρ ( g ) ∘ φ for all g ∈ G } , \operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G=\{\varphi:\varphi\circ\widetilde{g}=\rho(g)\circ\varphi\text{ for all }g\in G\}, Hom k ( Alb ( X ′ ) , A ) G = { φ : φ ∘ g = ρ ( g ) ∘ φ for all g ∈ G } ,
and we let G G G act on A ( k ) A(k) A ( k ) through ρ \rho ρ .
The structure theorem
Theorem 2.2. In Setting 2.1, assume X ′ ( k ) ≠ ∅ X'(k)\ne\varnothing X ′ ( k ) = ∅ and fix x 0 ∈ X ′ ( k ) x_0\in X'(k) x 0 ∈ X ′ ( k ) . (a) There is an exact sequence
0 ⟶ A ( k ) G ⟶ A ~ ( K ) → r Hom k ( Alb ( X ′ ) , A ) G → δ H 1 ( G , A ( k ) ) , 0\longrightarrow A(k)^G\longrightarrow\widetilde{A}(K)\xrightarrow{r}\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G\xrightarrow{\delta}H^1(G,A(k)), 0 ⟶ A ( k ) G ⟶ A ( K ) r Hom k ( Alb ( X ′ ) , A ) G δ H 1 ( G , A ( k )) , where δ ( φ ) \delta(\varphi) δ ( φ ) is the class of the cocycle g ↦ − φ ( α ( g ( x 0 ) ) ) g\mapsto-\varphi(\alpha(g(x_0))) g ↦ − φ ( α ( g ( x 0 ))) .
(b) The image of r r r contains ∣ G ∣ ⋅ Hom k ( Alb ( X ′ ) , A ) G |G|\cdot\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G ∣ G ∣ ⋅ Hom k ( Alb ( X ′ ) , A ) G . Consequently
rk A ~ ( K ) = rk A ( k ) G + rk Z Hom k ( Alb ( X ′ ) , A ) G , \operatorname{rk}\widetilde{A}(K)=\operatorname{rk}A(k)^G+\operatorname{rk}_{\mathbb Z}\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G, rk A ( K ) = rk A ( k ) G + rk Z Hom k ( Alb ( X ′ ) , A ) G , A ~ ( K ) t o r s = ( A ( k ) G ) t o r s . \widetilde{A}(K)_{\mathrm{tors}}=(A(k)^G)_{\mathrm{tors}}. A ( K ) tors = ( A ( k ) G ) tors .
(c) If x 0 x_0 x 0 is fixed by G G G , then δ = 0 \delta=0 δ = 0 and
A ~ ( K ) ≅ A ( k ) G ⊕ Hom k ( Alb ( X ′ ) , A ) G . \widetilde{A}(K)\cong A(k)^G\oplus\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G. A ( K ) ≅ A ( k ) G ⊕ Hom k ( Alb ( X ′ ) , A ) G .
Proof. Let M = Mor k ( X ′ , A ) = A ( L ) M=\operatorname{Mor}_k(X',A)=A(L) M = Mor k ( X ′ , A ) = A ( L ) , with G G G acting by g ⋅ f = ρ ( g ) ∘ f ∘ g − 1 g\cdot f=\rho(g)\circ f\circ g^{-1} g ⋅ f = ρ ( g ) ∘ f ∘ g − 1 . By (2.2), A ~ ( K ) ≅ M G \widetilde{A}(K)\cong M^G A ( K ) ≅ M G . The constant maps form the G G G -submodule A ( k ) A(k) A ( k ) of M M M , and by (2.3) the map f ↦ φ f\mapsto\varphi f ↦ φ identifies M / A ( k ) M/A(k) M / A ( k ) with Hom k ( Alb ( X ′ ) , A ) \operatorname{Hom}_k(\operatorname{Alb}(X'),A) Hom k ( Alb ( X ′ ) , A ) . This identification is G G G -equivariant for the action g ⋅ φ = ρ ( g ) ∘ φ ∘ g ~ − 1 g\cdot\varphi=\rho(g)\circ\varphi\circ\widetilde{g}^{-1} g ⋅ φ = ρ ( g ) ∘ φ ∘ g − 1 : indeed α ∘ g − 1 = g ~ − 1 ∘ α + α ( g − 1 ( x 0 ) ) \alpha\circ g^{-1}=\widetilde{g}^{-1}\circ\alpha+\alpha(g^{-1}(x_0)) α ∘ g − 1 = g − 1 ∘ α + α ( g − 1 ( x 0 )) , so the homomorphism part of ρ ( g ) ∘ ( c + φ ∘ α ) ∘ g − 1 \rho(g)\circ(c+\varphi\circ\alpha)\circ g^{-1} ρ ( g ) ∘ ( c + φ ∘ α ) ∘ g − 1 is ρ ( g ) ∘ φ ∘ g ~ − 1 \rho(g)\circ\varphi\circ\widetilde{g}^{-1} ρ ( g ) ∘ φ ∘ g − 1 . The invariants of Hom k ( Alb ( X ′ ) , A ) \operatorname{Hom}_k(\operatorname{Alb}(X'),A) Hom k ( Alb ( X ′ ) , A ) under this action are exactly Hom k ( Alb ( X ′ ) , A ) G \operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G Hom k ( Alb ( X ′ ) , A ) G . Taking G G G -invariants of
0 ⟶ A ( k ) ⟶ M ⟶ Hom k ( Alb ( X ′ ) , A ) ⟶ 0 0\longrightarrow A(k)\longrightarrow M\longrightarrow\operatorname{Hom}_k(\operatorname{Alb}(X'),A)\longrightarrow0 0 ⟶ A ( k ) ⟶ M ⟶ Hom k ( Alb ( X ′ ) , A ) ⟶ 0
gives the exact sequence in (a). To compute δ \delta δ , lift φ ∈ Hom k ( Alb ( X ′ ) , A ) G \varphi\in\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G φ ∈ Hom k ( Alb ( X ′ ) , A ) G to f = φ ∘ α f=\varphi\circ\alpha f = φ ∘ α . Then
g ⋅ f − f = ρ ( g ) ( g ~ − 1 ∘ α + α ( g − 1 ( x 0 ) ) ) − φ α = ρ ( g ) φ ( α ( g − 1 ( x 0 ) ) ) = φ ( g ~ α ( g − 1 ( x 0 ) ) ) , \begin{aligned}
g\cdot f-f=\rho(g)\bigl(\widetilde{g}^{-1}\circ\alpha+\alpha(g^{-1}(x_0))\bigr)-\varphi\alpha\\
&=\rho(g)\varphi\bigl(\alpha(g^{-1}(x_0))\bigr)=\varphi\bigl(\widetilde{g}\alpha(g^{-1}(x_0))\bigr),
\end{aligned} g ⋅ f − f = ρ ( g ) ( g − 1 ∘ α + α ( g − 1 ( x 0 )) ) − φ α = ρ ( g ) φ ( α ( g − 1 ( x 0 )) ) = φ ( g α ( g − 1 ( x 0 )) ) ,
and g ~ α ( g − 1 ( x 0 ) ) = α ( x 0 ) − α ( g ( x 0 ) ) = − α ( g ( x 0 ) ) \widetilde{g}\alpha(g^{-1}(x_0))=\alpha(x_0)-\alpha(g(x_0))=-\alpha(g(x_0)) g α ( g − 1 ( x 0 )) = α ( x 0 ) − α ( g ( x 0 )) = − α ( g ( x 0 )) . This proves (a).
Since H 1 ( G , A ( k ) ) H^1(G,A(k)) H 1 ( G , A ( k )) is killed by ∣ G ∣ |G| ∣ G ∣ , we have δ ( ∣ G ∣ φ ) = 0 \delta(|G|\varphi)=0 δ ( ∣ G ∣ φ ) = 0 , so ∣ G ∣ φ |G|\varphi ∣ G ∣ φ lies in the image of r r r . As Hom k ( Alb ( X ′ ) , A ) \operatorname{Hom}_k(\operatorname{Alb}(X'),A) Hom k ( Alb ( X ′ ) , A ) is a finitely generated free Z \mathbb{Z} Z -module, the rank formula follows, and since it is torsion free, the torsion of A ~ ( K ) \widetilde{A}(K) A ( K ) is that of A ( k ) G A(k)^G A ( k ) G . This proves (b). If x 0 x_0 x 0 is fixed by G G G , then α ∘ g = g ~ ∘ α \alpha\circ g=\widetilde{g}\circ\alpha α ∘ g = g ∘ α , so φ ↦ φ ∘ α \varphi\mapsto\varphi\circ\alpha φ ↦ φ ∘ α is a G G G -equivariant section of M → Hom k ( Alb ( X ′ ) , A ) M\to\operatorname{Hom}_k(\operatorname{Alb}(X'),A) M → Hom k ( Alb ( X ′ ) , A ) , and (c) follows. □ \square □
Remark 2.3. Suppose that N ρ : = ∑ g ∈ G ρ ( g ) N_\rho:=\sum_{g\in G}\rho(g) N ρ := ∑ g ∈ G ρ ( g ) vanishes in End k ( A ) \operatorname{End}_k(A) End k ( A ) . If ρ ( g ) c = c \rho(g)c=c ρ ( g ) c = c for all g g g then ∣ G ∣ c = N ρ ( c ) = 0 |G|c=N_\rho(c)=0 ∣ G ∣ c = N ρ ( c ) = 0 , so A G : = ⋂ ker ( 1 − ρ ( g ) ) A^G:=\bigcap\ker(1-\rho(g)) A G := ⋂ ker ( 1 − ρ ( g )) is contained in A [ ∣ G ∣ ] A[|G|] A [ ∣ G ∣ ] and A ( k ) G A(k)^G A ( k ) G is finite. Also, every φ ∈ Hom k ( Alb ( X ′ ) , A ) G \varphi\in\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^G φ ∈ Hom k ( Alb ( X ′ ) , A ) G satisfies φ ∘ ∑ g g ~ = N ρ ∘ φ = 0 \varphi\circ\sum_g\widetilde{g}=N_\rho\circ\varphi=0 φ ∘ ∑ g g = N ρ ∘ φ = 0 , hence factors through the quotient Prym X ′ / X = Alb ( X ′ ) / Im ( ∑ g g ~ ) \operatorname{Prym}_{X'/X}=\operatorname{Alb}(X')/\operatorname{Im}(\sum_g\widetilde{g}) Prym X ′ / X = Alb ( X ′ ) / Im ( ∑ g g ) , which for cyclic G G G is the Prym variety of [18 Def. 2.2] . Thus
rk A ~ ( K ) = rk Z Hom k ( Prym X ′ / X , A ) G . \operatorname{rk}\widetilde{A}(K)=\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_k(\operatorname{Prym}_{X'/X},A)^G. rk A ( K ) = rk Z Hom k ( Prym X ′ / X , A ) G .
For cyclic groups we obtain the following statement, which replaces [17 Theorem 1.1] , its amended form in [19 (2)] , and [18 Theorem 3.3] .
Corollary 2.4. In Setting 2.1 let G = ⟨ γ ⟩ G=\langle\gamma\rangle G = ⟨ γ ⟩ be cyclic of order n n n , put σ = ρ ( γ ) \sigma=\rho(\gamma) σ = ρ ( γ ) and γ ~ \widetilde{\gamma} γ as above, and assume X ′ ( k ) ≠ ∅ X'(k)\ne\varnothing X ′ ( k ) = ∅ .
(a) rk A ~ ( K ) = rk A σ ( k ) + rk Z Hom k ( Alb ( X ′ ) , A ) ( γ ~ , σ ) \operatorname{rk}\widetilde{A}(K)=\operatorname{rk}A^\sigma(k)+\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^{(\widetilde{\gamma},\sigma)} rk A ( K ) = rk A σ ( k ) + rk Z Hom k ( Alb ( X ′ ) , A ) ( γ , σ ) and A ~ ( K ) t o r s = A σ ( k ) t o r s \widetilde{A}(K)_{\mathrm{tors}}=A^\sigma(k)_{\mathrm{tors}} A ( K ) tors = A σ ( k ) tors . If γ \gamma γ has a k k k -rational fixed point, then A ~ ( K ) ≅ A σ ( k ) ⊕ Hom k ( Alb ( X ′ ) , A ) ( γ ~ , σ ) \widetilde{A}(K)\cong A^\sigma(k)\oplus\operatorname{Hom}_k(\operatorname{Alb}(X'),A)^{(\widetilde{\gamma},\sigma)} A ( K ) ≅ A σ ( k ) ⊕ Hom k ( Alb ( X ′ ) , A ) ( γ , σ ) .
(b) If 1 + σ + ⋯ + σ n − 1 = 0 1+\sigma+\cdots+\sigma^{n-1}=0 1 + σ + ⋯ + σ n − 1 = 0 , then A σ ⊂ A [ n ] A^\sigma\subset A[n] A σ ⊂ A [ n ] and rk A ~ ( K ) = rk Z Hom k ( Prym X ′ / X , A ) ( γ ~ , σ ) \operatorname{rk}\widetilde{A}(K)=\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_k(\operatorname{Prym}_{X'/X},A)^{(\widetilde{\gamma},\sigma)} rk A ( K ) = rk Z Hom k ( Prym X ′ / X , A ) ( γ , σ ) .
(c) If in addition there are an abelian variety B B B over k k k with an automorphism β \beta β , and a k k k -isogeny λ : Prym X ′ / X → A m × B \lambda:\operatorname{Prym}_{X'/X}\to A^m\times B λ : Prym X ′ / X → A m × B with λ ∘ γ ~ = ( σ × ⋯ × σ × β ) ∘ λ \lambda\circ\widetilde{\gamma}=(\sigma\times\cdots\times\sigma\times\beta)\circ\lambda λ ∘ γ = ( σ × ⋯ × σ × β ) ∘ λ and Hom k ( B , A ) ( β , σ ) = 0 \operatorname{Hom}_k(B,A)^{(\beta,\sigma)}=0 Hom k ( B , A ) ( β , σ ) = 0 , then rk A ~ ( K ) = m ⋅ rk Z End k ( A ) σ \operatorname{rk}\widetilde{A}(K)=m\cdot\operatorname{rk}_{\mathbb{Z}}\operatorname{End}_k(A)^\sigma rk A ( K ) = m ⋅ rk Z End k ( A ) σ .
Proof. Parts (a) and (b) are Theorem 2.2 and Remark 2.3. For (c), put θ = σ × ⋯ × σ × β \theta=\sigma\times\cdots\times\sigma\times\beta θ = σ × ⋯ × σ × β and let λ ′ \lambda' λ ′ be a k k k -isogeny with λ ′ λ = [ N ] \lambda'\lambda=[N] λ ′ λ = [ N ] and λ λ ′ = [ N ] \lambda\lambda'=[N] λ λ ′ = [ N ] . From λ γ ~ = θ λ \lambda\widetilde{\gamma}=\theta\lambda λ γ = θ λ we get λ ′ θ λ = N γ ~ = γ ~ λ ′ λ \lambda'\theta\lambda=N\widetilde{\gamma}=\widetilde{\gamma}\lambda'\lambda λ ′ θ λ = N γ = γ λ ′ λ , hence λ ′ θ = γ ~ λ ′ \lambda'\theta=\widetilde{\gamma}\lambda' λ ′ θ = γ λ ′ because λ \lambda λ is surjective. Composition with λ ′ \lambda' λ ′ and with λ \lambda λ therefore gives homomorphisms between Hom k ( Prym X ′ / X , A ) ( γ ~ , σ ) \operatorname{Hom}_k(\operatorname{Prym}_{X'/X},A)^{(\widetilde{\gamma},\sigma)} Hom k ( Prym X ′ / X , A ) ( γ , σ ) and Hom k ( A m × B , A ) ( θ , σ ) \operatorname{Hom}_k(A^m\times B,A)^{(\theta,\sigma)} Hom k ( A m × B , A ) ( θ , σ ) whose composites are multiplication by N N N , and the latter group is ( End k ( A ) σ ) m ⊕ Hom k ( B , A ) ( β , σ ) = ( End k ( A ) σ ) m (\operatorname{End}_k(A)^\sigma)^m\oplus\operatorname{Hom}_k(B,A)^{(\beta,\sigma)}=(\operatorname{End}_k(A)^\sigma)^m ( End k ( A ) σ ) m ⊕ Hom k ( B , A ) ( β , σ ) = ( End k ( A ) σ ) m . □ \square □
A counterexample
Example 2.7. Let k ⊂ C k \subset\mathbb{C} k ⊂ C be a field containing a primitive cube root of unity ρ 3 \rho_{3} ρ 3 . Let X ′ = E X'=E X ′ = E be the elliptic curve y 2 = x 3 + 1 y^{2}=x^{3}+1 y 2 = x 3 + 1 with origin O O O at infinity, and γ ( x , y ) = ( ρ 3 x , y ) \gamma(x,y)=(\rho_{3}x,y) γ ( x , y ) = ( ρ 3 x , y ) . Then γ \gamma γ is an automorphism of order 3 3 3 fixing O O O , hence a group automorphism, and End k ( E ) = Z [ γ ] ≅ Z [ ρ 3 ] \operatorname{End}_{k}(E)=\mathbb{Z}[\gamma] \cong\mathbb{Z}[\rho_{3}] End k ( E ) = Z [ γ ] ≅ Z [ ρ 3 ] is commutative and has no zero divisors; in particular 1 + γ + γ 2 = 0 1+\gamma+\gamma^{2}=0 1 + γ + γ 2 = 0 . The function t = y t=y t = y generates the function field of E / ⟨ γ ⟩ ≅ P 1 E/\langle\gamma\rangle\cong\mathbb{P}^{1} E / ⟨ γ ⟩ ≅ P 1 , so K = k ( t ) K=k(t) K = k ( t ) and L = K ( x ) L=K(x) L = K ( x ) with x 3 = t 2 − 1 x^{3}=t^{2}-1 x 3 = t 2 − 1 . Here Alb ( E ) = E \operatorname{Alb}(E)=E Alb ( E ) = E , α = id \alpha=\operatorname{id} α = id , γ ~ = γ \widetilde{\gamma}=\gamma γ = γ and Prym E / P 1 = E \operatorname{Prym}_{E/\mathbb{P}^{1}}=E Prym E / P 1 = E . Take A = E A=E A = E and σ ∈ { γ , γ 2 } \sigma\in\{\gamma,\gamma^{2}\} σ ∈ { γ , γ 2 } . The hypotheses of [18 Theorem 3.3] hold with m = 1 m=1 m = 1 and B = 0 B=0 B = 0 , and that theorem asserts that rk A ~ ( K ) = rk End k ( E ) = 2 \operatorname{rk}\widetilde{A}(K)=\operatorname{rk}\operatorname{End}_{k}(E)=2 rk A ( K ) = rk End k ( E ) = 2 for both choices of σ \sigma σ . If k k k is a number field, the hypotheses of [17 Theorem 1.1] also hold (with s = 3 s=3 s = 3 , n = 1 n=1 n = 1 and dim B = 0 \dim B=0 dim B = 0 in the notation there), so that [17 ] asserts rk A ~ ( K ) ≥ 2 \operatorname{rk}\widetilde{A}(K)\geq2 rk A ( K ) ≥ 2 and [19 (2)] asserts rk A ~ ( K ) = 2 \operatorname{rk}\widetilde{A}(K)=2 rk A ( K ) = 2 , for both choices of σ \sigma σ .
By Corollary 2.4(a), applied with the fixed point O O O , we have A ~ ( K ) ≅ E σ ( k ) ⊕ End k ( E ) ( γ , σ ) \widetilde{A}(K) \cong E^{\sigma}(k) \oplus\operatorname{End}_{k}(E)^{(\gamma,\sigma)} A ( K ) ≅ E σ ( k ) ⊕ End k ( E ) ( γ , σ ) . For σ = γ \sigma=\gamma σ = γ this gives rank 2. For σ = γ 2 \sigma=\gamma^{2} σ = γ 2 the condition φ γ = γ 2 φ \varphi\gamma=\gamma^{2}\varphi φ γ = γ 2 φ reads ( γ − γ 2 ) φ = 0 (\gamma-\gamma^{2})\varphi=0 ( γ − γ 2 ) φ = 0 , whence φ = 0 \varphi=0 φ = 0 and A ~ ( K ) = E σ ( k ) \widetilde{A}(K)=E^{\sigma}(k) A ( K ) = E σ ( k ) is finite.
Explicitly, the twist is Y 2 = X 3 + ( t 2 − 1 ) 2 Y^{2}=X^{3}+(t^{2}-1)^{2} Y 2 = X 3 + ( t 2 − 1 ) 2 for σ = γ \sigma=\gamma σ = γ , with g ( x 0 , y 0 ) = ( x 0 2 x 0 , x 0 3 y 0 ) g(x_{0},y_{0})=(x_{0}^{2}x_{0},x_{0}^{3}y_{0}) g ( x 0 , y 0 ) = ( x 0 2 x 0 , x 0 3 y 0 ) , and Y 2 = X 3 + ( t 2 − 1 ) 4 Y^{2}=X^{3}+(t^{2}-1)^{4} Y 2 = X 3 + ( t 2 − 1 ) 4 for σ = γ 2 \sigma=\gamma^{2} σ = γ 2 , with g ( x 0 , y 0 ) = ( x 0 4 x 0 , x 0 6 y 0 ) g(x_{0},y_{0})=(x_{0}^{4}x_{0},x_{0}^{6}y_{0}) g ( x 0 , y 0 ) = ( x 0 4 x 0 , x 0 6 y 0 ) ; one checks u γ g = g ∘ σ u_{\gamma}g=g\circ\sigma u γ g = g ∘ σ using u γ ( x ) = ρ 3 − 1 x u_{\gamma}(x)=\rho_{3}^{-1}x u γ ( x ) = ρ 3 − 1 x and u γ ( t ) = t u_{\gamma}(t)=t u γ ( t ) = t . The vanishing of the rank of the second curve can also be seen geometrically: over C ( t ) \mathbb{C}(t) C ( t ) its Weierstrass equation is minimal and defines an elliptic K3 surface with singular fibres of type IV* at t = ± 1 t=\pm1 t = ± 1 and t = ∞ t=\infty t = ∞ [25 ] ; the trivial lattice has rank 2 + 3 ⋅ 6 = 20 2+3\cdot6=20 2 + 3 ⋅ 6 = 20 , the maximal Picard number of a complex K3 surface, so the Mordell–Weil rank is 0 by the Shioda–Tate formula [23 ] . The first curve defines a rational elliptic surface with three fibres of type IV, of Mordell–Weil rank 10 − 2 − 6 = 2 10-2-6=2 10 − 2 − 6 = 2 .
Finally E σ ( k ) = { O , ( 0 , 1 ) , ( 0 , − 1 ) } ≅ Z / 3 Z E^{\sigma}(k)=\{O,(0,1),(0,-1)\}\cong\mathbb{Z}/3\mathbb{Z} E σ ( k ) = { O , ( 0 , 1 ) , ( 0 , − 1 )} ≅ Z /3 Z . If 2 3 ∈ k \sqrt[3]{2}\in k 3 2 ∈ k , then E [ 3 ] ( k ) ≅ ( Z / 3 Z ) 2 E[3](k)\cong(\mathbb{Z}/3\mathbb{Z})^{2} E [ 3 ] ( k ) ≅ ( Z /3 Z ) 2 , the 3-division polynomial of E E E being 3 x ( x 3 + 4 ) 3x(x^{3}+4) 3 x ( x 3 + 4 ) . For σ = γ 2 \sigma=\gamma^{2} σ = γ 2 we have A ~ ( K ) ≅ Z / 3 Z \widetilde{A}(K)\cong\mathbb{Z}/3\mathbb{Z} A ( K ) ≅ Z /3 Z , while [17 Theorem 1.1] and [18 Theorem 3.3] predict Z 2 ⊕ ( Z / 3 Z ) 2 \mathbb{Z}^{2}\oplus(\mathbb{Z}/3\mathbb{Z})^{2} Z 2 ⊕ ( Z /3 Z ) 2 ; for σ = γ \sigma=\gamma σ = γ we have A ~ ( K ) ≅ Z 2 ⊕ Z / 3 Z \widetilde{A}(K)\cong\mathbb{Z}^{2}\oplus\mathbb{Z}/3\mathbb{Z} A ( K ) ≅ Z 2 ⊕ Z /3 Z .
Cyclic multiple planes
Setting
Let n ≥ 2 n \ge2 n ≥ 2 and let k k k be a field of characteristic prime to n n n containing a primitive n n n -th root of unity ζ n \zeta_n ζ n . Let f ∈ k [ x , y ] f \in k[x,y] f ∈ k [ x , y ] be a polynomial such that the affine surface X n : w n = f ( x , y ) X_n: w^n=f(x,y) X n : w n = f ( x , y ) is geometrically irreducible, let X n \mathcal{X}_n X n be a smooth projective model of X n X_n X n on which the automorphism τ ( x , y , w ) = ( x , y , ζ n w ) \tau(x,y,w)=(x,y,\zeta_n w) τ ( x , y , w ) = ( x , y , ζ n w ) induces an automorphism τ ˉ \bar{\tau} τ ˉ , as in [18 §4] , and put
A = Alb ( X n ) , σ = τ ˉ ∗ ∈ Aut k ( A ) , d n = dim A . A=\operatorname{Alb}(\mathcal{X}_n),\qquad\sigma=\bar{\tau}_*\in\operatorname{Aut}_k(A),\qquad d_n=\dim A. A = Alb ( X n ) , σ = τ ˉ ∗ ∈ Aut k ( A ) , d n = dim A .
where τ ˉ ∗ \bar{\tau}_* τ ˉ ∗ is the automorphism induced by τ ˉ \bar{\tau} τ ˉ (§ 3.1 , with G = ⟨ τ ˉ ⟩ G=\langle\bar{\tau}\rangle G = ⟨ τ ˉ ⟩ ). Any two smooth projective models of X n X_n X n are birational, so A A A and σ \sigma σ do not depend on the choice of X n \mathcal{X}_n X n up to isomorphism, and by the Lang–Nishimura lemma [16 ] neither does the condition X n ( k ) ≠ ∅ \mathcal{X}_n(k)\ne\varnothing X n ( k ) = ∅ ; it holds, for instance, if X n X_n X n has a non-singular k k k -rational point. For m ≥ 1 m\ge1 m ≥ 1 let U m = X n × k ⋯ × k X n U_m=\mathcal{X}_n\times_k\cdots\times_k\mathcal{X}_n U m = X n × k ⋯ × k X n with coordinates ( x i , y i , w i ) (x_i,y_i,w_i) ( x i , y i , w i ) on the i i i -th factor, let γ ˉ = ( τ ˉ , … , τ ˉ ) \bar{\gamma}=(\bar{\tau},\ldots,\bar{\tau}) γ ˉ = ( τ ˉ , … , τ ˉ ) , let V m = U m / ⟨ γ ˉ ⟩ V_m=U_m/\langle\bar{\gamma}\rangle V m = U m / ⟨ γ ˉ ⟩ , and put L m = k ( U m ) L_m=k(U_m) L m = k ( U m ) and K m = k ( V m ) = L m ⟨ γ ˉ ⟩ K_m=k(V_m)=L_m^{\langle\bar{\gamma}\rangle} K m = k ( V m ) = L m ⟨ γ ˉ ⟩ . Let B B B be an abelian variety over k k k and β ∈ Aut k ( B ) \beta\in\operatorname{Aut}_k(B) β ∈ Aut k ( B ) with β n = id \beta^n=\operatorname{id} β n = id . We denote by B ~ m \widetilde{B}_m B m the twist of B B B over K m K_m K m defined, as in Setting 2.1, by G = ⟨ γ ˉ ⟩ G=\langle\bar{\gamma}\rangle G = ⟨ γ ˉ ⟩ acting on U m U_m U m and ρ ( γ ˉ ) = β \rho(\bar{\gamma})=\beta ρ ( γ ˉ ) = β . Similarly X ~ n \widetilde{\mathcal{X}}_n X n denotes the twist of X n \mathcal{X}_n X n defined by ρ ( γ ˉ ) = τ ˉ \rho(\bar{\gamma})=\bar{\tau} ρ ( γ ˉ ) = τ ˉ ; by [18 Prop. 3.2] , Alb ( X ~ n ) \operatorname{Alb}(\widetilde{\mathcal{X}}_n) Alb ( X n ) is the twist A ~ m \widetilde{A}_m A m of A A A defined by ρ ( γ ˉ ) = σ \rho(\bar{\gamma})=\sigma ρ ( γ ˉ ) = σ .
Lemma 3.1. The twist X ~ n \widetilde{\mathcal{X}}_n X n is birational over K m K_m K m to the surface f ( x 1 , y 1 ) z n = f ( x , y ) f(x_1,y_1)z^n=f(x,y) f ( x 1 , y 1 ) z n = f ( x , y ) of [18 (5.3)] . Under (2) the K m K_m K m -rational points P 1 = ( x 1 , y 1 , 1 ) P_1=(x_1,y_1,1) P 1 = ( x 1 , y 1 , 1 ) and P i = ( x i , y i , w i / w 1 ) P_i=(x_i,y_i,w_i/w_1) P i = ( x i , y i , w i / w 1 ) (2 ≤ i ≤ m 2\le i\le m 2 ≤ i ≤ m ) of [18 (5.4)] correspond to the projections pr i : U m → X n ( i ) = X n \operatorname{pr}_i:U_m\to\mathcal{X}_n^{(i)}=\mathcal{X}_n pr i : U m → X n ( i ) = X n .
Proof. Let u = u γ ˉ u=u_{\bar{\gamma}} u = u γ ˉ . Since w 1 ∘ γ ˉ − 1 = ζ n − 1 w 1 w_1\circ\bar{\gamma}^{-1}=\zeta_n^{-1}w_1 w 1 ∘ γ ˉ − 1 = ζ n − 1 w 1 , the L m L_m L m -isomorphism g ( x , y , w ) = ( x , y , w / w 1 ) g(x,y,w)=(x,y,w/w_1) g ( x , y , w ) = ( x , y , w / w 1 ) from X n \mathcal{X}_n X n to f ( x 1 , y 1 ) z n = f ( x , y ) f(x_1,y_1)z^n=f(x,y) f ( x 1 , y 1 ) z n = f ( x , y ) satisfies u g ( x , y , w ) = ( x , y , ζ n w / w 1 ) = g ( τ ( x , y , w ) ) u g(x,y,w)=(x,y,\zeta_n w/w_1)=g(\tau(x,y,w)) ug ( x , y , w ) = ( x , y , ζ n w / w 1 ) = g ( τ ( x , y , w )) , that is u g = g ∘ τ u g=g\circ\tau ug = g ∘ τ . The point P i P_i P i is the image under g g g of the L m L_m L m -point ( x i , y i , w i ) (x_i,y_i,w_i) ( x i , y i , w i ) of X n X_n X n , which is the rational map pr i \operatorname{pr}_i pr i . □
Lemma 3.2. We have 1 + σ + ⋯ + σ n − 1 = 0 1+\sigma+\cdots+\sigma^{n-1}=0 1 + σ + ⋯ + σ n − 1 = 0 in End k ( A ) \operatorname{End}_k(A) End k ( A ) . Consequently Prym X n / P 2 = A \operatorname{Prym}_{X_n/\mathbb{P}^2}=A Prym X n / P 2 = A , and A σ ⊂ A [ n ] A^\sigma\subset A[n] A σ ⊂ A [ n ] is finite.
Proof. It suffices to prove the identity over k ˉ \bar{k} k ˉ . Fix x 0 ∈ X n ( k ˉ ) x_0\in\mathcal{X}_n(\bar{k}) x 0 ∈ X n ( k ˉ ) and let α \alpha α be the Albanese map with α ( x 0 ) = 0 \alpha(x_0)=0 α ( x 0 ) = 0 . The morphism Φ = ∑ j = 0 n − 1 α ∘ τ ˉ j \Phi=\sum_{j=0}^{n-1}\alpha\circ\bar{\tau}^{\,j} Φ = ∑ j = 0 n − 1 α ∘ τ ˉ j satisfies Φ ∘ τ ˉ = Φ \Phi\circ\bar{\tau}=\Phi Φ ∘ τ ˉ = Φ . Regarded as a point of A ( k ˉ ( X n ) ) A(\bar{k}(\mathcal{X}_n)) A ( k ˉ ( X n )) , it is therefore invariant under Gal ( k ˉ ( X n ) / k ˉ ( x , y ) ) \operatorname{Gal}(\bar{k}(\mathcal{X}_n)/\bar{k}(x,y)) Gal ( k ˉ ( X n ) / k ˉ ( x , y )) , which is generated by h ↦ h ∘ τ ˉ − 1 h\mapsto h\circ\bar{\tau}^{-1} h ↦ h ∘ τ ˉ − 1 because X n X_n X n is geometrically irreducible of degree n n n over the plane. Hence Φ \Phi Φ factors through a rational map P 2 ⇢ A \mathbb{P}^2\dashrightarrow A P 2 ⇢ A , which is constant [14 §3] . On the other hand Φ = ( 1 + σ + ⋯ + σ n − 1 ) ∘ α + c \Phi=(1+\sigma+\cdots+\sigma^{n-1})\circ\alpha+c Φ = ( 1 + σ + ⋯ + σ n − 1 ) ∘ α + c for some c ∈ A ( k ˉ ) c\in A(\bar{k}) c ∈ A ( k ˉ ) , and α ( X n ) \alpha(\mathcal{X}_n) α ( X n ) generates A A A . The last assertions follow from Remark 2.3. □
Remark 3.3. In [18 §5] this identity was deduced from ( 1 − σ ) ( 1 + σ + ⋯ + σ n − 1 ) = 0 (1-\sigma)(1+\sigma+\cdots+\sigma^{n-1})=0 ( 1 − σ ) ( 1 + σ + ⋯ + σ n − 1 ) = 0 and σ ≠ 1 \sigma\ne1 σ = 1 . That deduction is not valid, since End ( A ) \operatorname{End}(A) End ( A ) may have zero divisors (for instance when A A A is isogenous to E d E^d E d with d ≥ 2 d\ge2 d ≥ 2 ).
Mordell–Weil groups over K m K_m K m
Theorem 3.4. In the setting of §3.1, assume X n ( k ) ≠ ∅ \mathcal{X}_n(k)\ne\varnothing X n ( k ) = ∅ .
(a) There is an exact sequence
0 ⟶ B β ( k ) ⟶ B ~ m ( K m ) ⟶ ( Hom k ( A , B ) ( σ , β ) ) m ⟶ H 1 ( Z / n Z , B ( k ) ) . 0\longrightarrow B^\beta(k)\longrightarrow\widetilde{B}_m(K_m)\longrightarrow\bigl(\operatorname{Hom}_k(A,B)^{(\sigma,\beta)}\bigr)^m\longrightarrow H^1(\mathbb{Z}/n\mathbb{Z},B(k)). 0 ⟶ B β ( k ) ⟶ B m ( K m ) ⟶ ( Hom k ( A , B ) ( σ , β ) ) m ⟶ H 1 ( Z / n Z , B ( k )) .
and the image of the middle map contains n ⋅ ( Hom k ( A , B ) ( σ , β ) ) m n \cdot(\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)})^{m} n ⋅ ( Hom k ( A , B ) ( σ , β ) ) m . Hence
rk B ~ m ( K m ) = rk B β ( k ) + m ⋅ rk Z Hom k ( A , B ) ( σ , β ) , \operatorname{rk} \widetilde{B}_{m}(K_{m})=\operatorname{rk} B^{\beta}(k)+m\cdot\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)}, rk B m ( K m ) = rk B β ( k ) + m ⋅ rk Z Hom k ( A , B ) ( σ , β ) ,
B ~ m ( K m ) t o r s = B β ( k ) t o r s . \widetilde{B}_{m}(K_{m})_{\mathrm{tors}}=B^{\beta}(k)_{\mathrm{tors}}. B m ( K m ) tors = B β ( k ) tors .
(b) For ( B , β ) = ( A , σ ) (B,\beta)=(A,\sigma) ( B , β ) = ( A , σ ) , the group A σ ( k ) A^{\sigma}(k) A σ ( k ) is finite and
rk Alb ( X ~ n ) ( K m ) = m ⋅ rk Z End k ( A ) σ , Alb ( X ~ n ) ( K m ) t o r s = A σ ( k ) . \operatorname{rk}\operatorname{Alb}(\widetilde{X}_{n})(K_{m})=m\cdot\operatorname{rk}_{\mathbb{Z}}\operatorname{End}_{k}(A)^{\sigma},\qquad\operatorname{Alb}(\widetilde{X}_{n})(K_{m})_{\mathrm{tors}}=A^{\sigma}(k). rk Alb ( X n ) ( K m ) = m ⋅ rk Z End k ( A ) σ , Alb ( X n ) ( K m ) tors = A σ ( k ) .
(c) Assume that X n X_{n} X n has a k k k -rational point O O O fixed by τ ‾ \overline{\tau} τ (for example the point ( a , b , 0 ) (a,b,0) ( a , b , 0 ) , where ( a , b ) (a,b) ( a , b ) is a non-singular k k k -rational point of the curve f = 0 f=0 f = 0 , if X n X_{n} X n is chosen so that X n ⇢ X n X_{n}\dashrightarrow X_{n} X n ⇢ X n is an isomorphism over the smooth locus of X n X_{n} X n ). Then O O O defines a K m K_{m} K m -rational point O ~ \widetilde{O} O of X ~ n \widetilde{X}_{n} X n ; let Q ~ i \widetilde{Q}_{i} Q i be the image of P i P_{i} P i under the Albanese map of X ~ n \widetilde{X}_{n} X n with base point O ~ \widetilde{O} O . Every φ ∈ Hom k ( A , B ) ( σ , β ) \varphi\in\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)} φ ∈ Hom k ( A , B ) ( σ , β ) descends to a K m K_{m} K m -homomorphism φ ~ : Alb ( X ~ n ) → B ~ m \widetilde{\varphi}\colon\operatorname{Alb}(\widetilde{X}_{n})\to\widetilde{B}_{m} φ : Alb ( X n ) → B m , and
B ~ m ( K m ) = B β ( k ) ⊕ ⨁ i = 1 m { φ ~ ( Q ~ i ) : φ ∈ Hom k ( A , B ) ( σ , β ) } , \widetilde{B}_{m}(K_{m})=B^{\beta}(k)\oplus\bigoplus_{i=1}^{m}\{\widetilde{\varphi}(\widetilde{Q}_{i}):\varphi\in\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)}\}, B m ( K m ) = B β ( k ) ⊕ i = 1 ⨁ m { φ ( Q i ) : φ ∈ Hom k ( A , B ) ( σ , β ) } ,
where each summand φ ↦ φ ~ ( Q ~ i ) \varphi\mapsto\widetilde{\varphi}(\widetilde{Q}_{i}) φ ↦ φ ( Q i ) is injective. In particular Alb ( X ~ n ) ( K m ) = A σ ( k ) ⊕ ⨁ i = 1 m End k ( A ) σ ⋅ Q ~ i \operatorname{Alb}(\widetilde{X}_{n})(K_{m})=A^{\sigma}(k)\oplus\bigoplus_{i=1}^{m}\operatorname{End}_{k}(A)^{\sigma}\cdot\widetilde{Q}_{i} Alb ( X n ) ( K m ) = A σ ( k ) ⊕ ⨁ i = 1 m End k ( A ) σ ⋅ Q i , and Q ~ 1 , … , Q ~ m \widetilde{Q}_{1},\ldots,\widetilde{Q}_{m} Q 1 , … , Q m is a basis of a free left End k ( A ) σ \operatorname{End}_{k}(A)^{\sigma} End k ( A ) σ -module.
Proof. Apply Theorem 2.2 to X ′ = U m X'=U_{m} X ′ = U m , G = ⟨ γ ‾ ⟩ ≅ Z / n Z G=\langle\overline{\gamma}\rangle\cong\mathbb{Z}/n\mathbb{Z} G = ⟨ γ ⟩ ≅ Z / n Z and ρ ( γ ‾ ) = β \rho(\overline{\gamma})=\beta ρ ( γ ) = β , with base point ( x 0 , … , x 0 ) (x_{0},\ldots,x_{0}) ( x 0 , … , x 0 ) for some x 0 ∈ X n ( k ) x_{0}\in X_{n}(k) x 0 ∈ X n ( k ) (and x 0 = O x_{0}=O x 0 = O in (c)). Since Alb ( X n × X n ) = A × A \operatorname{Alb}(X_{n}\times X_{n})=A\times A Alb ( X n × X n ) = A × A with Albanese map α × α \alpha\times\alpha α × α [20 ] , we have Alb ( U m ) = A m \operatorname{Alb}(U_{m})=A^{m} Alb ( U m ) = A m , the Albanese map is α × m \alpha^{\times m} α × m , and γ ~ = σ × m \widetilde{\gamma}=\sigma^{\times m} γ = σ × m . Therefore
Hom k ( A m , B ) ( σ × m , β ) = ⨁ i = 1 m Hom k ( A , B ) ( σ , β ) ∘ e i , \operatorname{Hom}_{k}(A^{m},B)^{(\sigma^{\times m},\beta)}=\bigoplus_{i=1}^{m}\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)}\circ e_{i}, Hom k ( A m , B ) ( σ × m , β ) = i = 1 ⨁ m Hom k ( A , B ) ( σ , β ) ∘ e i ,
where e i : A m → A e_{i}\colon A^{m}\to A e i : A m → A is the i i i -th projection. This gives (a). For (b), A σ ( k ) A^{\sigma}(k) A σ ( k ) is finite by Lemma 3.2.
In (c), τ ‾ ( O ) = O \overline{\tau}(O)=O τ ( O ) = O gives α ∘ τ ‾ = σ ∘ α \alpha\circ\overline{\tau}=\sigma\circ\alpha α ∘ τ = σ ∘ α , so the constant map with value O O O corresponds to a point O ~ ∈ X ~ n ( K m ) \widetilde{O}\in\widetilde{X}_{n}(K_{m}) O ∈ X n ( K m ) by (2) , and α \alpha α descends to the Albanese map α ~ \widetilde{\alpha} α of X ~ n \widetilde{X}_{n} X n with base point O ~ \widetilde{O} O (see the proof of [18 Prop. 3.2] ). By Lemma 3.1, Q ~ i = α ~ ( P i ) \widetilde{Q}_{i}=\widetilde{\alpha}(P_{i}) Q i = α ( P i ) corresponds to α ∘ pr i = e i ∘ α × m \alpha\circ\operatorname{pr}_{i}=e_{i}\circ\alpha^{\times m} α ∘ pr i = e i ∘ α × m , that is, to the pair ( c , φ ) = ( 0 , e i ) (c,\varphi)=(0,e_{i}) ( c , φ ) = ( 0 , e i ) in (3) . If φ ∈ Hom k ( A , B ) ( σ , β ) \varphi\in\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)} φ ∈ Hom k ( A , B ) ( σ , β ) , then g B ∘ φ ∘ g A − 1 g_{B}\circ\varphi\circ g_{A}^{-1} g B ∘ φ ∘ g A − 1 is invariant under Gal ( L m / K m ) \operatorname{Gal}(L_{m}/K_{m}) Gal ( L m / K m ) , where g A g_{A} g A and g B g_{B} g B are the isomorphisms of (1) ; this is the descended homomorphism φ ~ \widetilde{\varphi} φ , and it acts on points by f ↦ φ ∘ f f\mapsto\varphi\circ f f ↦ φ ∘ f in terms of (2) . Thus φ ~ ( Q ~ i ) \widetilde{\varphi}(\widetilde{Q}_{i}) φ ( Q i ) corresponds to ( 0 , φ ∘ e i ) (0,\varphi\circ e_{i}) ( 0 , φ ∘ e i ) , and (c) follows from Theorem 2.2 (c). □
Remark 3.5. Theorem 3.4(b) replaces [18 Theorem 1.1] , which asserts rk Alb ( X ~ n ) ( K m ) = m ⋅ rk End k ( A ) \operatorname{rk}\operatorname{Alb}(\widetilde{X}_{n})(K_{m})=m\cdot\operatorname{rk}\operatorname{End}_{k}(A) rk Alb ( X n ) ( K m ) = m ⋅ rk End k ( A ) and has Alb ( X n ) [ m ] ( k ) \operatorname{Alb}(X_{n})[m](k) Alb ( X n ) [ m ] ( k ) as torsion part. The two rank formulas agree if and only if End k ( A ) σ \operatorname{End}_{k}(A)^{\sigma} End k ( A ) σ has finite index in End k ( A ) \operatorname{End}_{k}(A) End k ( A ) , for instance if σ \sigma σ is central in End k ( A ) ⊗ Q \operatorname{End}_{k}(A)\otimes\mathbb{Q} End k ( A ) ⊗ Q . The assertion at the end of [18 §5] that Q ~ 1 , … , Q ~ m \widetilde{Q}_{1},\ldots,\widetilde{Q}_{m} Q 1 , … , Q m form a set of independent generators is replaced by Theorem 3.4(c).
The dependence on m m m is formal, as the following description of K m K_{m} K m shows. Let X ~ n ( 1 ) \widetilde{X}_{n}^{(1)} X n ( 1 ) denote the twist of X n X_{n} X n over K 1 = k ( x 1 , y 1 ) K_{1}=k(x_{1},y_{1}) K 1 = k ( x 1 , y 1 ) , which by Lemma 3.1 is birational to f ( x 1 , y 1 ) z n = f ( x , y ) f(x_{1},y_{1})z^{n}=f(x,y) f ( x 1 , y 1 ) z n = f ( x , y ) .
Proposition 3.6. The field K m K_m K m is K 1 K_1 K 1 -isomorphic to the function field of ( X ~ n ( 1 ) ) m − 1 (\widetilde{X}_n^{(1)})^{m-1} ( X n ( 1 ) ) m − 1 over K 1 K_1 K 1 . Consequently, if X n ( k ) ≠ ∅ X_n(k) \ne\varnothing X n ( k ) = ∅ and ( B , β ) = ( A , σ ) (B,\beta)=(A,\sigma) ( B , β ) = ( A , σ ) , the rank over K m K_m K m is m m m times the rank over K 1 K_1 K 1 .
Proof. Put u i = w i / w 1 ∈ L m u_i=w_i/w_1\in L_m u i = w i / w 1 ∈ L m for 2 ≤ i ≤ m 2\le i\le m 2 ≤ i ≤ m ; these are invariant under γ ˉ \bar{\gamma} γ ˉ , and f ( x 1 , y 1 ) u i n = f ( x i , y i ) f(x_1,y_1)u_i^n=f(x_i,y_i) f ( x 1 , y 1 ) u i n = f ( x i , y i ) . Let Y Y Y be the ( m − 1 ) (m-1) ( m − 1 ) -fold product of the surface f ( x 1 , y 1 ) z n = f ( x , y ) f(x_1,y_1)z^n=f(x,y) f ( x 1 , y 1 ) z n = f ( x , y ) over K 1 K_1 K 1 ; it is geometrically integral since X n X_n X n is. The point ( x i , y i , u i ) 2 ≤ i ≤ m (x_i,y_i,u_i)_{2\le i\le m} ( x i , y i , u i ) 2 ≤ i ≤ m of Y Y Y has transcendence degree 2 ( m − 1 ) 2(m-1) 2 ( m − 1 ) over K 1 K_1 K 1 , hence is a generic point, so the subfield F = k ( x 1 , y 1 , x i , y i , u i : 2 ≤ i ≤ m ) F=k(x_1,y_1,x_i,y_i,u_i:2\le i\le m) F = k ( x 1 , y 1 , x i , y i , u i : 2 ≤ i ≤ m ) of K m K_m K m is K 1 K_1 K 1 -isomorphic to K 1 ( Y ) K_1(Y) K 1 ( Y ) . Since L m = F ( w 1 ) L_m=F(w_1) L m = F ( w 1 ) with w 1 n = f ( x 1 , y 1 ) ∈ F w_1^n=f(x_1,y_1)\in F w 1 n = f ( x 1 , y 1 ) ∈ F , we have [ L m : F ] ≤ n = [ L m : K m ] [L_m:F]\le n=[L_m:K_m] [ L m : F ] ≤ n = [ L m : K m ] , hence F = K m F=K_m F = K m . The last assertion is Theorem 3.4(b). □
Proposition 3.6 is an instance of the general fact that B ′ ( F ( Y ) ) ≅ B ′ ( F ) ⊕ Hom F ( Alb ( Y ) , B ′ ) B'(F(Y))\cong B'(F)\oplus\operatorname{Hom}_F(\operatorname{Alb}(Y),B') B ′ ( F ( Y )) ≅ B ′ ( F ) ⊕ Hom F ( Alb ( Y ) , B ′ ) for an abelian variety B ′ B' B ′ over a field F F F and a smooth, projective, geometrically integral variety Y Y Y over F F F with an F F F -point: over F = K 1 F=K_1 F = K 1 , the variety Y = ( X ~ n ( 1 ) ) m − 1 Y=(\widetilde{X}_n^{(1)})^{m-1} Y = ( X n ( 1 ) ) m − 1 contributes m − 1 m-1 m − 1 further copies of End K 1 ( A ~ 1 ) \operatorname{End}_{K_1}(\widetilde{A}_1) End K 1 ( A 1 ) . The arithmetic content of the construction is therefore concentrated in the case m = 1 m=1 m = 1 , that is, in the twist of A A A over k ( x , y ) k(x,y) k ( x , y ) .
Cyclic covers of P ℓ \mathbb{P}^{\ell} P ℓ and superelliptic curves
Let ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 and let f ∈ k [ u 1 , … , u ℓ ] f\in k[u_1,\ldots,u_\ell] f ∈ k [ u 1 , … , u ℓ ] be such that the hypersurface w n = f ( u 1 , … , u ℓ ) w^n=f(u_1,\ldots,u_\ell) w n = f ( u 1 , … , u ℓ ) is geometrically irreducible. Suppose that it admits a smooth projective model on which w ↦ ζ n w w\mapsto\zeta_nw w ↦ ζ n w extends. Replacing ( x , y ) (x,y) ( x , y ) by ( u 1 , … , u ℓ ) (u_1,\ldots,u_\ell) ( u 1 , … , u ℓ ) , the arguments of this section then apply without change: the only property of P 2 \mathbb{P}^2 P 2 used in Lemma 3.2 is that the quotient is rational. The same applies to Sections 4 and 5; in Proposition 5.5 one uses that A ℓ ( k ) A^\ell(k) A ℓ ( k ) is not thin. The existence of such a smooth equivariant model is automatic in characteristic zero. For ℓ ≥ 3 \ell\ge3 ℓ ≥ 3 these are the covers considered in [18 Remark 1.5(1)] .
Remark 3.7. In particular [18 Remark 1.5(1)] holds in the form of Theorem 3.4.
For ℓ = 1 \ell=1 ℓ = 1 we recover the setting of [17 §5] . Let f ∈ k [ x ] f\in k[x] f ∈ k [ x ] be a polynomial of degree r ≥ 1 r\ge1 r ≥ 1 with non-zero discriminant, so that f f f is square-free and the curve y n = f ( x ) y^n=f(x) y n = f ( x ) is geometrically irreducible. Let C C C be its smooth projective model, ι \iota ι the automorphism induced by y ↦ ζ n y y\mapsto\zeta_n y y ↦ ζ n y , J = J ( C ) = Alb ( C ) J=J(C)=\operatorname{Alb}(C) J = J ( C ) = Alb ( C ) , and ι ~ ∈ Aut k ( J ) \widetilde{\iota}\in\operatorname{Aut}_k(J) ι ∈ Aut k ( J ) the induced automorphism. For m ≥ 1 m\ge1 m ≥ 1 let V m = C m / ⟨ ( ι , … , ι ) ⟩ V_m=C^m/\langle(\iota,\ldots,\iota)\rangle V m = C m / ⟨( ι , … , ι )⟩ and K m = k ( V m ) K_m=k(V_m) K m = k ( V m ) , and let C ξ C^\xi C ξ be the twist of C C C over K m K_m K m considered in [17 §5] . By Lemma 3.1, C ξ C^\xi C ξ is birational to f ( x 1 ) z n = f ( x ) f(x_1)z^n=f(x) f ( x 1 ) z n = f ( x ) , which is the model given in [17 §5] , so that C ξ C^\xi C ξ is the twist of C C C by ρ ( γ ˉ ) = ι \rho(\bar{\gamma})=\iota ρ ( γ ˉ ) = ι and J ( C ξ ) J(C^\xi) J ( C ξ ) is the twist of J J J by ρ ( γ ˉ ) = ι ~ \rho(\bar{\gamma})=\widetilde{\iota} ρ ( γ ˉ ) = ι . Let P 1 , … , P m P_1,\ldots,P_m P 1 , … , P m be the K m K_m K m -rational points of [17 §5, (1)] (with the indexing corrected as in Section 7); they are the points of Lemma 3.1.
Corollary 3.8 (Corrected form of [17 Theorem 1.2] ). Assume that C ( k ) ≠ ∅ C(k)\ne\varnothing C ( k ) = ∅ . Then J ~ ⊂ J [ n ] \widetilde{J}\subset J[n] J ⊂ J [ n ] is finite,
rk J ( C ξ ) ( K m ) = m ⋅ rk Z End k ( J ) ι ~ , J ( C ξ ) ( K m ) t o r s = J ~ ( k ) . \operatorname{rk} J(C^\xi)(K_m)=m\cdot\operatorname{rk}_{\mathbb{Z}}\operatorname{End}_k(J)^{\widetilde{\iota}},\qquad J(C^\xi)(K_m)_{\mathrm{tors}}=\widetilde{J}(k). rk J ( C ξ ) ( K m ) = m ⋅ rk Z End k ( J ) ι , J ( C ξ ) ( K m ) tors = J ( k ) . If C C C has a k k k -rational point O O O fixed by ι \iota ι , for instance O = ( a , 0 ) O=(a,0) O = ( a , 0 ) with a ∈ k a\in k a ∈ k a root of f f f , then
J ( C ξ ) ( K m ) = J ~ ( k ) ⊕ ⨁ i = 1 m End k ( J ) ι ~ ⋅ Q ~ i . J(C^\xi)(K_m)=\widetilde{J}(k)\oplus\bigoplus_{i=1}^{m}\operatorname{End}_k(J)^{\widetilde{\iota}}\cdot\widetilde{Q}_i. J ( C ξ ) ( K m ) = J ( k ) ⊕ i = 1 ⨁ m End k ( J ) ι ⋅ Q i .
where Q ~ i \widetilde{Q}_i Q i is the image of P i P_i P i under the Abel–Jacobi map of C ξ C^{\xi} C ξ with base point O ~ \widetilde{O} O . In particular, the rank formula rk J ( C ξ ) ( K m ) = m ⋅ rk End k ( J ) \operatorname{rk} J(C^{\xi})(K_m)=m\cdot\operatorname{rk}\operatorname{End}_k(J) rk J ( C ξ ) ( K m ) = m ⋅ rk End k ( J ) of [19 (3)] holds if and only if ι ~ \widetilde{\iota} ι is central in End k ( J ) ⊗ Q \operatorname{End}_k(J)\otimes\mathbb{Q} End k ( J ) ⊗ Q ; this is the case for n = 2 n=2 n = 2 and whenever End k ( J ) ⊗ Q \operatorname{End}_k(J)\otimes\mathbb{Q} End k ( J ) ⊗ Q is commutative.
Proof. This is Theorem 3.4(b),(c) and Lemma 3.2 for ℓ = 1 \ell=1 ℓ = 1 and ( B , β ) = ( J , ι ~ ) (B,\beta)=(J,\widetilde{\iota}) ( B , β ) = ( J , ι ) ; the point ( a , 0 ) (a,0) ( a , 0 ) is a non-singular point of the affine curve because f f f is square-free. The rank of End k ( J ) ι ~ \operatorname{End}_k(J)^{\widetilde{\iota}} End k ( J ) ι equals that of End k ( J ) \operatorname{End}_k(J) End k ( J ) if and only if the centralizer of ι ~ \widetilde{\iota} ι in End k ( J ) ⊗ Q \operatorname{End}_k(J)\otimes\mathbb{Q} End k ( J ) ⊗ Q is everything, that is, if and only if ι ~ \widetilde{\iota} ι is central. □ \square □
In [17 §5] the identity 1 + ι ~ + ⋯ + ι ~ n − 1 = 0 1+\widetilde{\iota}+\cdots+\widetilde{\iota}^{n-1}=0 1 + ι + ⋯ + ι n − 1 = 0 is deduced from ( 1 − ι ~ ) ( 1 + ι ~ + ⋯ + ι ~ n − 1 ) = 0 (1-\widetilde{\iota})(1+\widetilde{\iota}+\cdots+\widetilde{\iota}^{n-1})=0 ( 1 − ι ) ( 1 + ι + ⋯ + ι n − 1 ) = 0 and ι ~ ≠ 1 \widetilde{\iota}\ne1 ι = 1 ; as in [18 ] , this deduction is not valid in general, but the identity holds by Lemma 3.2, since C / ⟨ ι ⟩ ≅ P 1 C/\langle\iota\rangle\cong\mathbb{P}^{1} C / ⟨ ι ⟩ ≅ P 1 . The assertion at the end of [17 §5] that Q ~ 1 , … , Q ~ m \widetilde{Q}_1,\ldots,\widetilde{Q}_m Q 1 , … , Q m belong to a set of independent generators is replaced by Corollary 3.8.
Example 3.9. Let n = 3 n=3 n = 3 , f = x 2 − 1 f=x^2-1 f = x 2 − 1 and m ≥ 2 m\ge2 m ≥ 2 , and let k k k be a number field containing ζ 3 \zeta_3 ζ 3 and 2 3 \sqrt[3]{2} 3 2 . These data satisfy the hypotheses of [17 Theorem 1.2] , which requires deg f ≤ m \deg f\le m deg f ≤ m , and c = ( 1 , 0 ) c=(1,0) c = ( 1 , 0 ) is a k k k -rational point of C C C fixed by ι \iota ι . The map ( x , y ) ↦ ( y , x ) (x,y)\mapsto(y,x) ( x , y ) ↦ ( y , x ) is an isomorphism from C : y 3 = x 2 − 1 C:y^3=x^2-1 C : y 3 = x 2 − 1 to the curve E : y 2 = x 3 + 1 E:y^2=x^3+1 E : y 2 = x 3 + 1 of Example 2.7, carrying ι \iota ι to γ \gamma γ (with ρ 3 = ζ 3 \rho_3=\zeta_3 ρ 3 = ζ 3 ). Hence J ≅ E J\cong E J ≅ E , End k ( J ) ≅ Z [ ζ 3 ] \operatorname{End}_k(J)\cong\mathbb{Z}[\zeta_3] End k ( J ) ≅ Z [ ζ 3 ] is commutative, J ι ~ ( k ) ≅ E γ ( k ) ≅ Z / 3 Z J^{\widetilde{\iota}}(k)\cong E^{\gamma}(k)\cong\mathbb{Z}/3\mathbb{Z} J ι ( k ) ≅ E γ ( k ) ≅ Z /3 Z , and J [ 3 ] ( k ) ≅ ( Z / 3 Z ) 2 J[3](k)\cong(\mathbb{Z}/3\mathbb{Z})^2 J [ 3 ] ( k ) ≅ ( Z /3 Z ) 2 by Example 2.7. By Corollary 3.8,
J ( C ξ ) ( K m ) ≅ Z 2 m ⊕ Z / 3 Z , J(C^{\xi})(K_m)\cong\mathbb{Z}^{2m}\oplus\mathbb{Z}/3\mathbb{Z}, J ( C ξ ) ( K m ) ≅ Z 2 m ⊕ Z /3 Z ,
whereas [17 Theorem 1.2] asserts J ( C ξ ) ( K m ) ≅ Z 2 m ⊕ ( Z / 3 Z ) 2 J(C^{\xi})(K_m)\cong\mathbb{Z}^{2m}\oplus(\mathbb{Z}/3\mathbb{Z})^2 J ( C ξ ) ( K m ) ≅ Z 2 m ⊕ ( Z /3 Z ) 2 . For m = 1 m=1 m = 1 , which is excluded in [17 ] , the model ( x 1 − 1 ) z 3 = x 1 2 − 1 (x_1-1)z^3=x_1^2-1 ( x 1 − 1 ) z 3 = x 1 2 − 1 of C ξ C^{\xi} C ξ is isomorphic over K 1 = k ( x 1 ) K_1=k(x_1) K 1 = k ( x 1 ) to Y 2 = X 3 + ( x 1 2 − 1 ) 2 Y^2=X^3+(x_1^2-1)^2 Y 2 = X 3 + ( x 1 2 − 1 ) 2 via X = ( x 1 − 1 ) z X=(x_1-1)z X = ( x 1 − 1 ) z , Y = ( x 1 − 1 ) x Y=(x_1-1)x Y = ( x 1 − 1 ) x ; this is the twist of Example 2.7 for σ = γ \sigma=\gamma σ = γ , in accordance with Lemma 3.1.
Endomorphisms commuting with an automorphism
In this section all abelian varieties are over C \mathbb{C} C . We describe groups of equivariant homomorphisms in terms of the eigenspace decomposition of the first homology. This replaces the case-by-case analysis of [18 §6] .
Eigenspaces
Let A A A be a complex abelian variety and σ ∈ Aut ( A ) \sigma\in\operatorname{Aut}(A) σ ∈ Aut ( A ) with σ n = id \sigma^n=\operatorname{id} σ n = id . Put Λ A = H 1 ( A , Z ) \Lambda_A=H_1(A,\mathbb{Z}) Λ A = H 1 ( A , Z ) and V A = Λ A ⊗ Q V_A=\Lambda_A\otimes\mathbb{Q} V A = Λ A ⊗ Q . The tangent space T 0 A T_0A T 0 A is V A ⊗ R V_A\otimes\mathbb{R} V A ⊗ R with a complex structure J J J ; let W A ⊂ V A ⊗ C W_A\subset V_A\otimes\mathbb{C} W A ⊂ V A ⊗ C be the i i i -eigenspace of J ⊗ 1 J\otimes1 J ⊗ 1 . Then V A ⊗ C = W A ⊕ W ‾ A V_A\otimes\mathbb{C}=W_A\oplus\overline{W}_A V A ⊗ C = W A ⊕ W A , and v ↦ 1 2 ( v − i J v ) v\mapsto\frac{1}{2}(v-iJv) v ↦ 2 1 ( v − i J v ) is a σ \sigma σ -equivariant C \mathbb{C} C -linear isomorphism T 0 A → W A T_0A\to W_A T 0 A → W A . In the notation of [18 §1] , Λ n = Λ A \Lambda_n=\Lambda_A Λ n = Λ A and W n = W A W_n=W_A W n = W A . Recall that a Q \mathbb{Q} Q -linear map u : V A → V B u:V_A\to V_B u : V A → V B comes from Hom ( A , B ) ⊗ Q \operatorname{Hom}(A,B)\otimes\mathbb{Q} Hom ( A , B ) ⊗ Q if and only if u C ( W A ) ⊆ W B u_{\mathbb{C}}(W_A)\subseteq W_B u C ( W A ) ⊆ W B .
For d ∣ n d\mid n d ∣ n let V A , d = ker Φ d ( σ ) V_{A,d}=\ker\Phi_d(\sigma) V A , d = ker Φ d ( σ ) , where Φ d \Phi_d Φ d is the d d d -th cyclotomic polynomial. Since σ \sigma σ has finite order, V A = ⨁ d ∣ n V A , d V_A=\bigoplus_{d\mid n}V_{A,d} V A = ⨁ d ∣ n V A , d . Fix a primitive n n n -th root of unity ζ n ∈ C \zeta_n\in\mathbb{C} ζ n ∈ C and put ζ d = ζ n n / d \zeta_d=\zeta_n^{n/d} ζ d = ζ n n / d . Letting ζ d \zeta_d ζ d act as σ \sigma σ makes V A , d V_{A,d} V A , d a vector space over Q ( ζ d ) \mathbb{Q}(\zeta_d) Q ( ζ d ) ; we put e d ( A ) = dim Q ( ζ d ) V A , d e_d(A)=\dim_{\mathbb{Q}(\zeta_d)}V_{A,d} e d ( A ) = dim Q ( ζ d ) V A , d , so that 2 dim A = ∑ d ∣ n φ ( d ) e d ( A ) 2\dim A=\sum_{d\mid n}\varphi(d)e_d(A) 2 dim A = ∑ d ∣ n φ ( d ) e d ( A ) . For λ ∈ C \lambda\in\mathbb{C} λ ∈ C with λ n = 1 \lambda^n=1 λ n = 1 , of order d d d , let χ λ : Q ( ζ d ) → C \chi_\lambda:\mathbb{Q}(\zeta_d)\to\mathbb{C} χ λ : Q ( ζ d ) → C be the embedding with χ λ ( ζ d ) = λ \chi_\lambda(\zeta_d)=\lambda χ λ ( ζ d ) = λ , and let
H A ( λ ) = V A , d ⊗ Q ( ζ d ) , χ λ C , H_A(\lambda)=V_{A,d}\otimes_{\mathbb{Q}(\zeta_d),\chi_\lambda}\mathbb{C}, H A ( λ ) = V A , d ⊗ Q ( ζ d ) , χ λ C ,
W A ( λ ) = W A ∩ H A ( λ ) , a A ( λ ) = dim C W A ( λ ) . W_A(\lambda)=W_A\cap H_A(\lambda),\qquad a_A(\lambda)=\dim_{\mathbb{C}}W_A(\lambda). W A ( λ ) = W A ∩ H A ( λ ) , a A ( λ ) = dim C W A ( λ ) .
Then H A ( λ ) H_A(\lambda) H A ( λ ) is the λ \lambda λ -eigenspace of σ \sigma σ on V A ⊗ C V_A \otimes\mathbb{C} V A ⊗ C and has dimension e d ( A ) e_d(A) e d ( A ) . Since W A W_A W A and W ‾ A \overline{W}_A W A are stable under σ \sigma σ , we have W A = ⨁ λ W A ( λ ) W_A=\bigoplus_\lambda W_A(\lambda) W A = ⨁ λ W A ( λ ) and H A ( λ ) = W A ( λ ) ⊕ ( W ‾ A ∩ H A ( λ ) ) H_A(\lambda)=W_A(\lambda)\oplus(\overline{W}_A\cap H_A(\lambda)) H A ( λ ) = W A ( λ ) ⊕ ( W A ∩ H A ( λ )) ; complex conjugation maps W A ( λ ‾ ) W_A(\overline{\lambda}) W A ( λ ) onto W ‾ A ∩ H A ( λ ) \overline{W}_A\cap H_A(\lambda) W A ∩ H A ( λ ) . Hence
a A ( λ ) + a A ( λ ‾ ) = e d ( A ) . (4) a_A(\lambda)+a_A(\overline{\lambda})=e_d(A). \tag*{(4)} a A ( λ ) + a A ( λ ) = e d ( A ) . (4)
Definition 4.1. We say that ( A , σ ) (A,\sigma) ( A , σ ) is rigid if a A ( λ ) ∈ { 0 , e d ( A ) } a_A(\lambda)\in\{0,e_d(A)\} a A ( λ ) ∈ { 0 , e d ( A )} for every λ \lambda λ , where d d d is the order of λ \lambda λ . In that case, for every d d d with e d ( A ) > 0 e_d(A)>0 e d ( A ) > 0 we put Σ A ( d ) = { λ : λ of order d , W A ( λ ) = H A ( λ ) } \Sigma_A(d)=\{\lambda:\lambda\text{ of order }d,\ W_A(\lambda)=H_A(\lambda)\} Σ A ( d ) = { λ : λ of order d , W A ( λ ) = H A ( λ )} ; by (4.1), the primitive d d d -th roots of unity are the disjoint union of Σ A ( d ) \Sigma_A(d) Σ A ( d ) and Σ A ( d ) ‾ \overline{\Sigma_A(d)} Σ A ( d ) .
If λ = ± 1 \lambda=\pm1 λ = ± 1 then λ = λ ‾ \lambda=\overline{\lambda} λ = λ and (4.1) gives a A ( λ ) = e d ( A ) / 2 a_A(\lambda)=e_d(A)/2 a A ( λ ) = e d ( A ) /2 ; hence rigidity forces e 1 ( A ) = e 2 ( A ) = 0 e_1(A)=e_2(A)=0 e 1 ( A ) = e 2 ( A ) = 0 . Assumption 1.2 of [18 ] requires that Λ n \Lambda_n Λ n be a Z [ ζ n ] \mathbb{Z}[\zeta_n] Z [ ζ n ] -module, that is V A = V A , n V_A=V_{A,n} V A = V A , n , and that no pair of complex conjugate characters occur in W n W_n W n , that is, W A ( λ ) W_A(\lambda) W A ( λ ) and W A ( λ ‾ ) W_A(\overline{\lambda}) W A ( λ ) are not both non-zero. By (4.1), Assumption 1.2 holds if and only if ( A , σ ) (A,\sigma) ( A , σ ) is rigid and V A = V A , n V_A=V_{A,n} V A = V A , n .
Equivariant homomorphisms
Lemma 4.2. Let ( A , σ ) (A,\sigma) ( A , σ ) and ( B , β ) (B,\beta) ( B , β ) be complex abelian varieties with automorphisms satisfying σ n = β n = id \sigma^n=\beta^n=\operatorname{id} σ n = β n = id . Then Hom ( A , B ) ( σ , β ) ⊗ Q \operatorname{Hom}(A,B)^{(\sigma,\beta)}\otimes\mathbb{Q} Hom ( A , B ) ( σ , β ) ⊗ Q is the set of
u ∈ ⨁ d ∣ n Hom Q ( ζ d ) ( V A , d , V B , d ) u\in\bigoplus_{d\mid n}\operatorname{Hom}_{\mathbb{Q}(\zeta_d)}(V_{A,d},V_{B,d}) u ∈ d ∣ n ⨁ Hom Q ( ζ d ) ( V A , d , V B , d ) such that u λ ( W A ( λ ) ) ⊆ W B ( λ ) u_\lambda(W_A(\lambda))\subseteq W_B(\lambda) u λ ( W A ( λ )) ⊆ W B ( λ ) for all λ \lambda λ , where u λ = u ⊗ ι λ C : H A ( λ ) → H B ( λ ) u_\lambda=u\otimes\iota_\lambda\mathbb{C}:H_A(\lambda)\to H_B(\lambda) u λ = u ⊗ ι λ C : H A ( λ ) → H B ( λ ) .
Proof. A Q \mathbb{Q} Q -linear map u : V A → V B u:V_A\to V_B u : V A → V B satisfies u σ = β u u\sigma=\beta u u σ = β u if and only if it maps each V A , d V_{A,d} V A , d into V B , d V_{B,d} V B , d and is Q ( ζ d ) \mathbb{Q}(\zeta_d) Q ( ζ d ) -linear there. Such a map sends H A ( λ ) H_A(\lambda) H A ( λ ) into H B ( λ ) H_B(\lambda) H B ( λ ) , and since W B = ⨁ λ W B ( λ ) W_B=\bigoplus_\lambda W_B(\lambda) W B = ⨁ λ W B ( λ ) , the condition u C ( W A ) ⊆ W B u_{\mathbb{C}}(W_A)\subseteq W_B u C ( W A ) ⊆ W B is equivalent to u λ ( W A ( λ ) ) ⊆ W B ( λ ) u_\lambda(W_A(\lambda))\subseteq W_B(\lambda) u λ ( W A ( λ )) ⊆ W B ( λ ) for all λ \lambda λ .
Corollary 4.3. In the situation of Lemma 4.2:
(i) dim Q Hom ( A , B ) ( σ , β ) ⊗ Q ≤ ∑ d ∣ n ϕ ( d ) e d ( A ) e d ( B ) \dim_{\mathbb{Q}}\operatorname{Hom}(A,B)^{(\sigma,\beta)}\otimes\mathbb{Q}\leq\sum_{d\mid n}\phi(d)e_d(A)e_d(B) dim Q Hom ( A , B ) ( σ , β ) ⊗ Q ≤ ∑ d ∣ n ϕ ( d ) e d ( A ) e d ( B ) , with equality if and only if, for every λ \lambda λ , either W A ( λ ) = 0 W_A(\lambda)=0 W A ( λ ) = 0 or W B ( λ ) = H B ( λ ) W_B(\lambda)=H_B(\lambda) W B ( λ ) = H B ( λ ) .
(ii) If ( A , σ ) (A,\sigma) ( A , σ ) and ( B , β ) (B,\beta) ( B , β ) are rigid, then
dim Q Hom ( A , B ) ( σ , β ) ⊗ Q = ∑ d ∈ D ϕ ( d ) e d ( A ) e d ( B ) , \dim_{\mathbb{Q}}\operatorname{Hom}(A,B)^{(\sigma,\beta)}\otimes\mathbb{Q}=\sum_{d\in D}\phi(d)e_d(A)e_d(B), dim Q Hom ( A , B ) ( σ , β ) ⊗ Q = d ∈ D ∑ ϕ ( d ) e d ( A ) e d ( B ) , where D D D is the set of d ∣ n d\mid n d ∣ n with e d ( A ) e d ( B ) > 0 e_d(A)e_d(B)>0 e d ( A ) e d ( B ) > 0 and Σ A ( d ) = Σ B ( d ) \Sigma_A(d)=\Sigma_B(d) Σ A ( d ) = Σ B ( d ) .
Proof. The space
S = ⨁ d Hom Q ( ζ d ) ( V A , d , V B , d ) S=\bigoplus_d\operatorname{Hom}_{\mathbb{Q}(\zeta_d)}(V_{A,d},V_{B,d}) S = d ⨁ Hom Q ( ζ d ) ( V A , d , V B , d )
has dimension ∑ d ϕ ( d ) e d ( A ) e d ( B ) \sum_d\phi(d)e_d(A)e_d(B) ∑ d ϕ ( d ) e d ( A ) e d ( B ) , and u ↦ ( u λ ) λ u\mapsto(u_\lambda)_\lambda u ↦ ( u λ ) λ identifies S ⊗ C S\otimes\mathbb{C} S ⊗ C with ⨁ λ Hom C ( H A ( λ ) , H B ( λ ) ) \bigoplus_\lambda\operatorname{Hom}_{\mathbb{C}}(H_A(\lambda),H_B(\lambda)) ⨁ λ Hom C ( H A ( λ ) , H B ( λ )) . The conditions of Lemma 4.2 are C \mathbb{C} C -linear in ( u λ ) (u_\lambda) ( u λ ) , so they hold on all of S S S if and only if they hold on all of S ⊗ C S\otimes\mathbb{C} S ⊗ C , that is, if and only if every linear map H A ( λ ) → H B ( λ ) H_A(\lambda)\to H_B(\lambda) H A ( λ ) → H B ( λ ) sends W A ( λ ) W_A(\lambda) W A ( λ ) into W B ( λ ) W_B(\lambda) W B ( λ ) . This proves (i).
For (ii), the conditions decouple according to d d d . Fix d d d with e d ( A ) e d ( B ) > 0 e_d(A)e_d(B)>0 e d ( A ) e d ( B ) > 0 . If Σ A ( d ) = Σ B ( d ) \Sigma_A(d)=\Sigma_B(d) Σ A ( d ) = Σ B ( d ) , then for λ ∈ Σ A ( d ) \lambda\in\Sigma_A(d) λ ∈ Σ A ( d ) we have W B ( λ ) = H B ( λ ) W_B(\lambda)=H_B(\lambda) W B ( λ ) = H B ( λ ) , and for λ ∉ Σ A ( d ) \lambda\notin\Sigma_A(d) λ ∈ / Σ A ( d ) we have W A ( λ ) = 0 W_A(\lambda)=0 W A ( λ ) = 0 ; so the whole d d d -component of S S S satisfies the conditions. Otherwise, since Σ A ( d ) \Sigma_A(d) Σ A ( d ) and Σ B ( d ) \Sigma_B(d) Σ B ( d ) both have ϕ ( d ) / 2 \phi(d)/2 ϕ ( d ) /2 elements, there is λ ∈ Σ A ( d ) ∖ Σ B ( d ) \lambda\in\Sigma_A(d)\setminus\Sigma_B(d) λ ∈ Σ A ( d ) ∖ Σ B ( d ) ; then W A ( λ ) = H A ( λ ) W_A(\lambda)=H_A(\lambda) W A ( λ ) = H A ( λ ) and W B ( λ ) = 0 W_B(\lambda)=0 W B ( λ ) = 0 , so u λ = 0 u_\lambda=0 u λ = 0 , and this forces u ∣ V A , d = 0 u|_{V_{A,d}}=0 u ∣ V A , d = 0 because u ↦ u ⊗ ι λ C u\mapsto u\otimes\iota_\lambda\mathbb{C} u ↦ u ⊗ ι λ C is injective.
Corollary 4.4. Let ( A , σ ) (A,\sigma) ( A , σ ) be a complex abelian variety with an automorphism, σ n = i d \sigma^n=\mathrm{id} σ n = id . Then the centralizer of σ \sigma σ in End ( A ) ⊗ Q \operatorname{End}(A)\otimes\mathbb{Q} End ( A ) ⊗ Q contains Q [ σ ] ≅ ∏ e d ( A ) > 0 Q ( ζ d ) \mathbb{Q}[\sigma]\cong\prod_{e_d(A)>0}\mathbb{Q}(\zeta_d) Q [ σ ] ≅ ∏ e d ( A ) > 0 Q ( ζ d ) and is contained in ⨁ d End Q ( ζ d ) ( V A , d ) ≅ ∏ d M e d ( A ) ( Q ( ζ d ) ) \bigoplus_d\operatorname{End}_{\mathbb{Q}(\zeta_d)}(V_{A,d})\cong\prod_d M_{e_d(A)}(\mathbb{Q}(\zeta_d)) ⨁ d End Q ( ζ d ) ( V A , d ) ≅ ∏ d M e d ( A ) ( Q ( ζ d )) . Hence
∑ d ∣ n , e d ( A ) > 0 φ ( d ) ≤ rk Z End ( A ) σ ≤ ∑ d ∣ n φ ( d ) e d ( A ) 2 . \sum_{\substack{d\mid n,\ e_d(A)>0}}\varphi(d)\leq\operatorname{rk}_{\mathbb{Z}}\operatorname{End}(A)^\sigma\leq\sum_{d\mid n}\varphi(d)e_d(A)^2. d ∣ n , e d ( A ) > 0 ∑ φ ( d ) ≤ rk Z End ( A ) σ ≤ d ∣ n ∑ φ ( d ) e d ( A ) 2 . and the upper bound is attained if and only if ( A , σ ) (A,\sigma) ( A , σ ) is rigid. In particular, if Assumption 1.2 of [18 ] holds and n ≥ 3 n\geq3 n ≥ 3 , then the centralizer is isomorphic to M e ( Q ( ζ n ) ) M_e(\mathbb{Q}(\zeta_n)) M e ( Q ( ζ n )) with e = 2 dim A / φ ( n ) e=2\dim A/\varphi(n) e = 2 dim A / φ ( n ) , and
rk Z End ( A ) σ = 4 ( dim A ) 2 φ ( n ) . \operatorname{rk}_{\mathbb{Z}}\operatorname{End}(A)^\sigma=\frac{4(\dim A)^2}{\varphi(n)}. rk Z End ( A ) σ = φ ( n ) 4 ( dim A ) 2 .
Proof. Apply Corollary 4.3 (i) with ( B , β ) = ( A , σ ) (B,\beta)=(A,\sigma) ( B , β ) = ( A , σ ) : the equality condition says that W A ( λ ) ∈ { 0 , H A ( λ ) } W_A(\lambda)\in\{0,H_A(\lambda)\} W A ( λ ) ∈ { 0 , H A ( λ )} for all λ \lambda λ , which is rigidity. The lower bound holds because σ \sigma σ is semisimple and its minimal polynomial is ∏ e d ( A ) > 0 Φ d \prod_{e_d(A)>0}\Phi_d ∏ e d ( A ) > 0 Φ d . The last statement is the case V A = V A , n V_A=V_{A,n} V A = V A , n . □ \square □
For n = 3 , 4 , 6 n=3,4,6 n = 3 , 4 , 6 this gives 2 ( dim A ) 2 2(\dim A)^2 2 ( dim A ) 2 , and for n = 5 , 8 , 10 , 12 n=5,8,10,12 n = 5 , 8 , 10 , 12 it gives ( dim A ) 2 (\dim A)^2 ( dim A ) 2 . Note that the restriction 3 ≤ n ≤ 12 3\leq n\leq12 3 ≤ n ≤ 12 , n ≠ 7 , 9 , 11 n\neq7,9,11 n = 7 , 9 , 11 of [18 Theorem 1.3] plays no role.
The cases n = 8 n=8 n = 8 and n = 12 n=12 n = 12
The following lemma is elementary; we include a proof because it is needed to correct [18 Prop. 4.2(v),(vi)] .
Lemma 4.5. Let B B B be a complex abelian variety of dimension g g g and F 0 ⊂ End ( B ) ⊗ Q F_0\subset\operatorname{End}(B)\otimes\mathbb{Q} F 0 ⊂ End ( B ) ⊗ Q an imaginary quadratic field acting on T 0 B T_0B T 0 B through a single embedding ι 0 : F 0 → C \iota_0:F_0\to\mathbb{C} ι 0 : F 0 → C . Then B B B is isogenous to E 0 g E_0^g E 0 g , where E 0 E_0 E 0 is an elliptic curve with complex multiplication by the ring of integers O F 0 \mathcal{O}_{F_0} O F 0 of F 0 F_0 F 0 , and End ( B ) ⊗ Q ≅ M g ( F 0 ) \operatorname{End}(B)\otimes\mathbb{Q}\cong M_g(F_0) End ( B ) ⊗ Q ≅ M g ( F 0 ) .
Proof. The space V = H 1 ( B , Q ) V=H_1(B,\mathbb{Q}) V = H 1 ( B , Q ) is an F 0 F_0 F 0 -vector space of dimension g g g . Since F 0 F_0 F 0 acts on T 0 B = V ⊗ Q R T_0B=V\otimes_{\mathbb{Q}}\mathbb{R} T 0 B = V ⊗ Q R by the scalars ι 0 ( a ) \iota_0(a) ι 0 ( a ) , the complex structure of V ⊗ Q R V\otimes_{\mathbb{Q}}\mathbb{R} V ⊗ Q R is multiplication by the element of F 0 ⊗ Q R F_0\otimes_{\mathbb{Q}}\mathbb{R} F 0 ⊗ Q R mapped to i i i by ι 0 ⊗ R \iota_0\otimes\mathbb{R} ι 0 ⊗ R . An F 0 F_0 F 0 -basis of V V V therefore gives a C \mathbb{C} C -linear isomorphism T 0 B ≅ C g T_0B\cong\mathbb{C}^g T 0 B ≅ C g under which H 1 ( B , Z ) H_1(B,\mathbb{Z}) H 1 ( B , Z ) is commensurable with ι 0 ( O F 0 ) g \iota_0(\mathcal{O}_{F_0})^g ι 0 ( O F 0 ) g . Hence B B B is isogenous to ( C / ι 0 ( O F 0 ) ) g = E 0 g (\mathbb{C}/\iota_0(\mathcal{O}_{F_0}))^g=E_0^g ( C / ι 0 ( O F 0 ) ) g = E 0 g , and End ( B ) ⊗ Q ≅ M g ( F 0 ) \operatorname{End}(B)\otimes\mathbb{Q}\cong M_g(F_0) End ( B ) ⊗ Q ≅ M g ( F 0 ) . □ \square □
Remark 4.6. Let n ∈ { 8 , 12 } n\in\{8,12\} n ∈ { 8 , 12 } and let ( A , σ ) (A,\sigma) ( A , σ ) satisfy Assumption 1.2 of [18 ] . By rigidity and V A = V A , n V_A=V_{A,n} V A = V A , n , we have W A = ⨁ λ ∈ Σ A ( n ) H A ( λ ) W_A=\bigoplus_{\lambda\in\Sigma_A(n)}H_A(\lambda) W A = ⨁ λ ∈ Σ A ( n ) H A ( λ ) and Σ A ( n ) = { ζ n j : j ∈ S } \Sigma_A(n)=\{\zeta_n^j:j\in S\} Σ A ( n ) = { ζ n j : j ∈ S } for a subset S ⊂ ( Z / n Z ) × S\subset(\mathbb{Z}/n\mathbb{Z})^\times S ⊂ ( Z / n Z ) × with ( Z / n Z ) × = S ⊔ ( − S ) (\mathbb{Z}/n\mathbb{Z})^\times=S\sqcup(-S) ( Z / n Z ) × = S ⊔ ( − S ) . For n = 8 n=8 n = 8 the possible sets S S S are { 1 , 3 } \{1,3\} { 1 , 3 } , { 5 , 7 } \{5,7\} { 5 , 7 } , { 1 , 5 } \{1,5\} { 1 , 5 } , { 3 , 7 } \{3,7\} { 3 , 7 } , i.e. the cosets of { 1 , 3 } \{1,3\} { 1 , 3 } and { 1 , 5 } \{1,5\} { 1 , 5 } , which are the subgroups fixing Q ( ζ 8 + ζ 8 − 1 ) = Q ( − 2 ) \mathbb{Q}(\zeta_8+\zeta_8^{-1})=\mathbb{Q}(\sqrt{-2}) Q ( ζ 8 + ζ 8 − 1 ) = Q ( − 2 ) and Q ( ζ 8 2 ) = Q ( i ) \mathbb{Q}(\zeta_8^2)=\mathbb{Q}(i) Q ( ζ 8 2 ) = Q ( i ) . For n = 12 n=12 n = 12 they are the cosets { 1 , 7 } \{1,7\} { 1 , 7 } , { 5 , 11 } \{5,11\} { 5 , 11 } of the subgroup fixing Q ( ζ 12 4 ) = Q ( − 3 ) \mathbb{Q}(\zeta_{12}^4)=\mathbb{Q}(\sqrt{-3}) Q ( ζ 12 4 ) = Q ( − 3 ) and { 1 , 5 } \{1,5\} { 1 , 5 } , { 7 , 11 } \{7,11\} { 7 , 11 } of the subgroup fixing Q ( ζ 12 3 ) = Q ( i ) \mathbb{Q}(\zeta_{12}^3)=\mathbb{Q}(i) Q ( ζ 12 3 ) = Q ( i ) . Hence there is an imaginary quadratic field F 0 ⊂ Q ( ζ n ) = Q [ σ ] F_0\subset\mathbb{Q}(\zeta_n)=\mathbb{Q}[\sigma] F 0 ⊂ Q ( ζ n ) = Q [ σ ] such that all the embeddings ι λ \iota_\lambda ι λ , λ ∈ Σ A ( n ) \lambda\in\Sigma_A(n) λ ∈ Σ A ( n ) , agree on F 0 F_0 F 0 ; thus F 0 F_0 F 0 acts on T 0 A ≅ W A T_0A\cong W_A T 0 A ≅ W A through a single embedding, and Lemma 4.5 shows that A A A is isogenous to E 0 dim A E_0^{\dim A} E 0 d i m A with E 0 E_0 E 0 having complex multiplication by O F 0 \mathcal{O}_{F_0} O F 0 . Consequently:
End ( A ) ⊗ Q ≅ M dim A ( F 0 ) \operatorname{End}(A)\otimes\mathbb{Q}\cong M_{\dim A}(F_0) End ( A ) ⊗ Q ≅ M d i m A ( F 0 ) has dimension 2 ( dim A ) 2 2(\dim A)^2 2 ( dim A ) 2 , whereas the centralizer of σ \sigma σ has dimension ( dim A ) 2 (\dim A)^2 ( dim A ) 2 by Corollary 4.4 . So, if dim A > 0 \dim A>0 dim A > 0 , then σ \sigma σ is not central, and m ⋅ rk End ( A ) m\cdot\operatorname{rk}\operatorname{End}(A) m ⋅ rk End ( A ) is not the rank.
In [18 Prop. 4.2(v),(vi)] one of n 1 , n 2 n_1,n_2 n 1 , n 2 vanishes. For (vi) this follows from the above, since E i E_i E i and E ρ E_\rho E ρ are not isogenous. For (v), let ζ 8 \zeta_8 ζ 8 be a square root of i i i and s ( x , y ) = ( i x , ζ 8 y ) s(x,y)=(ix,\zeta_8y) s ( x , y ) = ( i x , ζ 8 y ) . This is an automorphism of order 8 8 8 of C 2 : y 2 = x 5 + x C_2:y^2=x^5+x C 2 : y 2 = x 5 + x ,
and s 4 s^{4} s 4 is the hyperelliptic involution, so s ∗ 4 = − 1 s_{*}^{4}=-1 s ∗ 4 = − 1 on J ( C 2 ) J(C_{2}) J ( C 2 ) and Q [ s ∗ ] ≅ Q ( ζ 8 ) \mathbb{Q}[s_{*}]\cong\mathbb{Q}(\zeta_{8}) Q [ s ∗ ] ≅ Q ( ζ 8 ) . The automorphism s s s acts on the basis d x / y dx/y d x / y , x d x / y x\,dx/y x d x / y of H 0 ( C 2 , Ω 1 ) H^{0}(C_{2},\Omega^{1}) H 0 ( C 2 , Ω 1 ) by ζ 8 \zeta_{8} ζ 8 and ζ 8 3 \zeta_{8}^{3} ζ 8 3 , hence s ∗ s_{*} s ∗ acts on T 0 J ( C 2 ) = H 0 ( C 2 , Ω 1 ) ∨ T_{0}J(C_{2})=H^{0}(C_{2},\Omega^{1})^{\vee} T 0 J ( C 2 ) = H 0 ( C 2 , Ω 1 ) ∨ with the same eigenvalues, and the corresponding embeddings of Q ( ζ 8 ) \mathbb{Q}(\zeta_{8}) Q ( ζ 8 ) agree on Q ( ζ 8 + ζ 8 3 ) = Q ( − 2 ) \mathbb{Q}(\zeta_{8}+\zeta_{8}^{3})=\mathbb{Q}(\sqrt{-2}) Q ( ζ 8 + ζ 8 3 ) = Q ( − 2 ) . By Lemma 4.5, J ( C 2 ) J(C_{2}) J ( C 2 ) is isogenous to E − 2 2 E_{\sqrt{-2}}^{2} E − 2 2 , where E − 2 E_{\sqrt{-2}} E − 2 has complex multiplication by Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] and is not isogenous to E i E_{i} E i .
Remark 4.7. For a cyclic multiple plane, e 1 ( A ) = 0 e_{1}(A)=0 e 1 ( A ) = 0 by Lemma 3.2. The numbers e d ( A ) e_{d}(A) e d ( A ) and a A ( λ ) a_{A}(\lambda) a A ( λ ) can be computed from the Alexander polynomial of the branch curve and from the position of its singular points; see Libgober [11 ] , Esnault [5 ] and, for curves whose only singularities are nodes and cusps, Cogolludo-Agustín and Libgober [3 ] . We use this only through [3 ] in Section 6.
Arithmetic over number fields
We return to the setting of §3.1, now with k k k of characteristic 0. Let k ˉ \bar{k} k ˉ be an algebraic closure and Γ k = Gal ( k ˉ / k ) \Gamma_{k}=\operatorname{Gal}(\bar{k}/k) Γ k = Gal ( k ˉ / k ) . For a finite extension k ′ k' k ′ of k k k in k ˉ \bar{k} k ˉ we write K m ′ = k ′ ( ν m ) K'_{m}=k'(\nu_{m}) K m ′ = k ′ ( ν m ) .
The Galois action
Theorem 5.1. Assume X n ( k ) ≠ ∅ X_{n}(k)\ne\varnothing X n ( k ) = ∅ , and let H = Hom k ˉ ( A , B ) ( σ , β ) H=\operatorname{Hom}_{\bar{k}}(A,B)^{(\sigma,\beta)} H = Hom k ˉ ( A , B ) ( σ , β ) . Let k H k_{H} k H be the fixed field of the kernel of the natural action Γ k → Aut ( H ) \Gamma_{k}\to\operatorname{Aut}(H) Γ k → Aut ( H ) . Then k H / k k_{H}/k k H / k is finite, the action factors through Gal ( k H / k ) \operatorname{Gal}(k_{H}/k) Gal ( k H / k ) , and for every finite extension k ′ k' k ′ of k k k in k ˉ \bar{k} k ˉ
rk B ~ m ( K m ′ ) = rk B β ( k ′ ) + m ⋅ rk Z H Γ k ′ . \operatorname{rk}\widetilde{B}_{m}(K'_{m})=\operatorname{rk}B^{\beta}(k')+m\cdot\operatorname{rk}_{\mathbb{Z}}H^{\Gamma_{k'}}. rk B m ( K m ′ ) = rk B β ( k ′ ) + m ⋅ rk Z H Γ k ′ . Moreover rk Z H Γ k ′ = rk Z H \operatorname{rk}_{\mathbb{Z}}H^{\Gamma_{k'}}=\operatorname{rk}_{\mathbb{Z}}H rk Z H Γ k ′ = rk Z H if and only if k ′ ⊇ k H k'\supseteq k_{H} k ′ ⊇ k H .
Proof. Theorem 3.4(a) over k ′ k' k ′ gives the formula with Hom k ′ ( A , B ) ( σ , β ) \operatorname{Hom}_{k'}(A,B)^{(\sigma,\beta)} Hom k ′ ( A , B ) ( σ , β ) in place of H Γ k ′ H^{\Gamma_{k'}} H Γ k ′ ; these groups coincide because σ \sigma σ and β \beta β are defined over k k k . The group H H H is finitely generated and each of its elements is defined over a finite extension of k k k , so Γ k \Gamma_{k} Γ k acts through a finite quotient. Finally, if H Γ k ′ H^{\Gamma_{k'}} H Γ k ′ has finite index in H H H , then every element of Γ k ′ \Gamma_{k'} Γ k ′ acts trivially on H ⊗ Q H\otimes\mathbb{Q} H ⊗ Q , hence on the torsion free group H H H . □
Corollary 5.2. Let k ⊂ C k\subset\mathbb{C} k ⊂ C be a number field containing ζ n \zeta_{n} ζ n , assume X n ( k ) ≠ ∅ X_{n}(k)\ne\varnothing X n ( k ) = ∅ , and let e d = e d ( A C ) e_{d}=e_{d}(A_{\mathbb{C}}) e d = e d ( A C ) . Then
m ∑ d ∣ n e d > 0 φ ( d ) ≤ rk Alb ( X ~ n ) ( K m ) ≤ m ∑ d ∣ n φ ( d ) e d 2 . m\sum_{\substack{d\mid n\\ e_{d}>0}}\varphi(d)\leq\operatorname{rk}\operatorname{Alb}(\widetilde{X}_{n})(K_{m})\leq m\sum_{d\mid n}\varphi(d)e_{d}^{2}. m d ∣ n e d > 0 ∑ φ ( d ) ≤ rk Alb ( X n ) ( K m ) ≤ m d ∣ n ∑ φ ( d ) e d 2 . The upper bound is attained if and only if ( A C , σ ) (A_{\mathbb{C}},\sigma) ( A C , σ ) is rigid and every endomorphism of A k ˉ A_{\bar{k}} A k ˉ commuting with σ \sigma σ is defined over k k k . If Assumption 1.2 of [18 ] holds and n ≥ 3 n\geq3 n ≥ 3 , then
rk Alb ( X ~ n ) ( K m ) ≤ m ⋅ 4 d n 2 φ ( n ) , \operatorname{rk}\operatorname{Alb}(\widetilde{X}_{n})(K_{m})\leq m\cdot\frac{4d_{n}^{2}}{\varphi(n)}, rk Alb ( X n ) ( K m ) ≤ m ⋅ φ ( n ) 4 d n 2 , with equality if and only if k ⊇ k H k\supseteq k_{H} k ⊇ k H for H = End k ˉ ( A ) σ H=\operatorname{End}_{\bar{k}}(A)^{\sigma} H = End k ˉ ( A ) σ ; if moreover 2 d n = φ ( n ) 2d_{n}=\varphi(n) 2 d n = φ ( n ) , equality holds over k k k .
Proof. Fix an embedding k ˉ ↪ C \bar{k}\hookrightarrow\mathbb{C} k ˉ ↪ C extending the given embedding k ↪ C k\hookrightarrow\mathbb{C} k ↪ C . Endomorphisms of abelian varieties do not change after extension between algebraically closed fields, so End k ˉ ( A ) = End ( A C ) \operatorname{End}_{\bar{k}}(A)=\operatorname{End}(A_{\mathbb{C}}) End k ˉ ( A ) = End ( A C ) . The lower bound holds because Z [ σ ] ⊆ End k ( A ) σ \mathbb{Z}[\sigma]\subseteq\operatorname{End}_{k}(A)^{\sigma} Z [ σ ] ⊆ End k ( A ) σ , and the rest follows from Theorem 3.4(b), Theorem 5.1 and Corollary 4.4. If 2 d n = φ ( n ) 2d_{n}=\varphi(n) 2 d n = φ ( n ) under Assumption 1.2, then e = 1 e=1 e = 1 and the centralizer is Q ( ζ n ) = Q [ σ ] \mathbb{Q}(\zeta_{n})=\mathbb{Q}[\sigma] Q ( ζ n ) = Q [ σ ] , which is defined over k k k . □
Rigidity of ( A C , σ ) (A_{\mathbb{C}},\sigma) ( A C , σ ) does not depend on the embedding k ⊂ C k \subset\mathbb{C} k ⊂ C : the numbers a A ( λ ) a_{A}(\lambda) a A ( λ ) are the dimensions of the eigenspaces of σ \sigma σ on the k k k -vector space Lie ( A ) \operatorname{Lie}(A) Lie ( A ) , whose eigenvalues lie in k k k , and a change of embedding only permutes the labels λ \lambda λ . Corollary 5.2 replaces [18 Theorem 1.3] . It also gives the correct form of [18 Theorem 1.4] : under the hypotheses there, α ( X 6 ) \alpha(X_{6}) α ( X 6 ) is a surface by [18 Theorem 4.3] , which is a result of Kulikov [10 ] , so d 6 ≥ 2 d_{6} \ge2 d 6 ≥ 2 , and rk Alb ( X ~ 6 ) ( K m ) ≤ 2 m d 6 2 \operatorname{rk}\operatorname{Alb}(\widetilde{X}_{6})(K_{m}) \le2md_{6}^{2} rk Alb ( X 6 ) ( K m ) ≤ 2 m d 6 2 , with equality if and only if every endomorphism of A k ‾ A_{\overline{k}} A k commuting with σ \sigma σ is defined over k k k .
Torsion.
Proposition 5.3. Let k ⊂ C k \subset\mathbb{C} k ⊂ C and assume X n ( k ) ≠ ∅ \mathcal{X}_{n}(k) \ne\varnothing X n ( k ) = ∅ . Then Alb ( X ~ n ) ( K m ) t o r s = A σ ( k ) \operatorname{Alb}(\widetilde{\mathcal{X}}_{n})(K_{m})_{\mathrm{tors}} = A^{\sigma}(k) Alb ( X n ) ( K m ) tors = A σ ( k ) , and A σ ( k ‾ ) ≅ Λ A / ( 1 − σ ) Λ A A^{\sigma}(\overline{k}) \cong\Lambda_{A}/(1-\sigma)\Lambda_{A} A σ ( k ) ≅ Λ A / ( 1 − σ ) Λ A is a group killed by n n n , of order
∣ A σ ( k ‾ ) ∣ = ∏ p r ∣ n , r ≥ 1 p e p r \left|A^{\sigma}(\overline{k})\right| = \prod_{\substack{p^{r}\mid n,\,r\ge1}} p^{e_{p^{r}}} A σ ( k ) = p r ∣ n , r ≥ 1 ∏ p e p r the product running over the prime powers p r > 1 p^{r} > 1 p r > 1 dividing n n n . In particular Alb ( X ~ n ) ( K m ) \operatorname{Alb}(\widetilde{\mathcal{X}}_{n})(K_{m}) Alb ( X n ) ( K m ) is torsion free if e d = 0 e_{d}=0 e d = 0 for every prime power d > 1 d>1 d > 1 dividing $n; this is the case, for example, if Assumption 1.2 of [18 ] holds and n n n is not a prime power.
Proof. The first statement is Theorem 3.4(b). By Lemma 3.2, e 1 = 0 e_{1}=0 e 1 = 0 , so 1 − σ 1-\sigma 1 − σ is invertible on V A V_{A} V A . Writing A ( C ) = ( V A ⊗ R ) / Λ A A(\mathbb{C})=(V_{A}\otimes\mathbb{R})/\Lambda_{A} A ( C ) = ( V A ⊗ R ) / Λ A , the kernel of 1 − σ 1-\sigma 1 − σ is ( 1 − σ ) − 1 Λ A / Λ A (1-\sigma)^{-1}\Lambda_{A}/\Lambda_{A} ( 1 − σ ) − 1 Λ A / Λ A , which 1 − σ 1-\sigma 1 − σ maps isomorphically onto Λ A / ( 1 − σ ) Λ A \Lambda_{A}/(1-\sigma)\Lambda_{A} Λ A / ( 1 − σ ) Λ A ; since A σ A^{\sigma} A σ is finite, A σ ( k ‾ ) = A σ ( C ) A^{\sigma}(\overline{k})=A^{\sigma}(\mathbb{C}) A σ ( k ) = A σ ( C ) . Its order is ∣ det ( 1 − σ ∣ V A ) ∣ = ∏ d Φ d ( 1 ) e d \left|\det(1-\sigma\mid V_{A})\right|=\prod_{d}\Phi_{d}(1)^{e_{d}} ∣ det ( 1 − σ ∣ V A ) ∣ = ∏ d Φ d ( 1 ) e d , and Φ d ( 1 ) \Phi_{d}(1) Φ d ( 1 ) equals p p p if d = p r d=p^{r} d = p r with r ≥ 1 r\ge1 r ≥ 1 and equals 1 1 1 if d > 1 d>1 d > 1 is not a prime power. Finally n = ∑ j = 1 n − 1 ( 1 − σ j ) n=\sum_{j=1}^{n-1}(1-\sigma^{j}) n = ∑ j = 1 n − 1 ( 1 − σ j ) in End ( A ) \operatorname{End}(A) End ( A ) by Lemma 3.2, so n n n kills Λ A / ( 1 − σ ) Λ A \Lambda_{A}/(1-\sigma)\Lambda_{A} Λ A / ( 1 − σ ) Λ A . □ \square □
Remark 5.4. If X n \mathcal{X}_{n} X n has no k k k -rational point fixed by τ ‾ \overline{\tau} τ , the map δ \delta δ of Theorem 2.2 takes values in H 1 ( Z / n Z , A ( k ) ) H^{1}(\mathbb{Z}/n\mathbb{Z},A(k)) H 1 ( Z / n Z , A ( k )) . Since 1 + σ + ⋯ + σ n − 1 = 0 1+\sigma+\cdots+\sigma^{n-1}=0 1 + σ + ⋯ + σ n − 1 = 0 , one has H 1 ( Z / n Z , A ( k ) ) = A ( k ) / ( 1 − σ ) A ( k ) H^{1}(\mathbb{Z}/n\mathbb{Z},A(k))=A(k)/(1-\sigma)A(k) H 1 ( Z / n Z , A ( k )) = A ( k ) / ( 1 − σ ) A ( k ) , a finite group when k k k is a number field, because it is killed by n n n and A ( k ) A(k) A ( k ) is finitely generated. The image of δ \delta δ is an arithmetic invariant; it vanishes over an algebraically closed field, where the isogeny 1 − σ 1-\sigma 1 − σ is surjective on points.
Specialization.
Let k k k be a number field containing ζ n \zeta_{n} ζ n , let U ⊂ A k 2 U\subset\mathbb{A}^{2}_{k} U ⊂ A k 2 be the open set f ≠ 0 f\ne0 f = 0 , and let T ⊂ U × A 1 T\subset U\times\mathbb{A}^{1} T ⊂ U × A 1 be the μ n \mu_{n} μ n -torsor w n = f ( x , y ) w^{n}=f(x,y) w n = f ( x , y ) , with μ n \mu_{n} μ n acting by w ↦ ζ w w\mapsto\zeta w w ↦ ζ w . For c ∈ k × c\in k^{\times} c ∈ k × let T c = Spec k [ w ] / ( w n − c ) T_{c}=\operatorname{Spec}k[w]/(w^{n}-c) T c = Spec k [ w ] / ( w n − c ) and let B c B_{c} B c be the twist of B B B by T c T_{c} T c and β \beta β : its points over an extension k ′ k' k ′ of k k k are the μ n \mu_{n} μ n -equivariant morphisms T c ⊗ k ′ → B T_{c}\otimes k'\to B T c ⊗ k ′ → B , where ζ n \zeta_{n} ζ n acts on B B B by β \beta β . Up to isomorphism B c B_{c} B c depends only on the class of c c c in k × / k × n k^{\times}/k^{\times n} k × / k × n .
Proposition 5.5. Assume X n ( k ) ≠ ∅ \mathcal{X}_{n}(k)\ne\varnothing X n ( k ) = ∅ .
(a) There is an abelian scheme B → U \mathcal{B}\to U B → U whose generic fibre is B ~ 1 \widetilde{B}_{1} B 1 and whose fibre at u ∈ U ( k ) u\in U(k) u ∈ U ( k ) is B f ( u ) B_{f(u)} B f ( u ) .
(b) There is a thin set Ω ⊂ U ( k ) \Omega\subset U(k) Ω ⊂ U ( k ) such that for u ∈ U ( k ) ∖ Ω u\in U(k)\setminus\Omega u ∈ U ( k ) ∖ Ω the specialization map B ~ 1 ( K 1 ) → B f ( u ) ( k ) \widetilde{B}_{1}(K_{1})\to B_{f(u)}(k) B 1 ( K 1 ) → B f ( u ) ( k ) is injective. In particular
rk B f ( u ) ( k ) ≥ rk B β ( k ) + rk Z Hom k ( A , B ) ( σ , β ) ( u ∈ U ( k ) ∖ Ω ) . \operatorname{rk} B_{f(u)}(k)\ge\operatorname{rk} B^{\beta}(k)+\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_{k}(A,B)^{(\sigma,\beta)}
\qquad(u\in U(k)\setminus\Omega). rk B f ( u ) ( k ) ≥ rk B β ( k ) + rk Z Hom k ( A , B ) ( σ , β ) ( u ∈ U ( k ) ∖ Ω ) . (c) The classes of f ( u ) f(u) f ( u ) in k × / k × n k^{\times}/k^{\times n} k × / k × n , for u ∈ U ( k ) ∖ Ω u\in U(k)\setminus\Omega u ∈ U ( k ) ∖ Ω , form an infinite set.
Proof. (a) The torsor T → U T \to U T → U is finite étale because n n n and f f f are invertible on U U U . Choose an ample line bundle L \mathcal{L} L on B B B and replace it by ⨂ j = 0 n − 1 ( β j ) ∗ L \bigotimes_{j=0}^{n-1}(\beta^{j})^{*}\mathcal{L} ⨂ j = 0 n − 1 ( β j ) ∗ L , which is ample and β \beta β -invariant. Descent of B × T B \times T B × T along T → U T \to U T → U with respect to the descent datum defined by β \beta β is then effective [2 §6.1] , and gives an abelian scheme B → U \mathcal{B} \to U B → U . Its points over a U U U -scheme S S S are the μ n \mu_n μ n -equivariant S S S -morphisms T × U S → B × S T \times_U S \to B \times S T × U S → B × S , so its generic fibre is B ~ 1 \widetilde{B}_1 B 1 by (2) and Lemma 3.1, and its fibre at u u u is B f ( u ) B_{f(u)} B f ( u ) because T u = T f ( u ) T_u = T_{f(u)} T u = T f ( u ) .
(b) This is Néron’s specialization theorem [15 ] , in the form of [22 §11.1] , applied to B → U \mathcal{B} \to U B → U ; the rank of B ~ 1 ( K 1 ) \widetilde{B}_1(K_1) B 1 ( K 1 ) is given by Theorem 3.4 (a) with m = 1 m=1 m = 1 .
(c) For c ∈ k × c \in k^{\times} c ∈ k × , the set Ω c \Omega_c Ω c of u ∈ U ( k ) u \in U(k) u ∈ U ( k ) with f ( u ) ∈ c k × n f(u) \in c k^{\times n} f ( u ) ∈ c k × n is the image of the k k k -points of the surface c z n = f ( x , y ) cz^n=f(x,y) c z n = f ( x , y ) , which is geometrically irreducible and of degree n ≥ 2 n \ge2 n ≥ 2 over U U U ; so Ω c \Omega_c Ω c is thin [21 §3.1] . If only finitely many classes c 1 , … , c r c_1,\ldots,c_r c 1 , … , c r occurred, then U ( k ) U(k) U ( k ) would be contained in the thin set Ω ∪ Ω c 1 ∪ ⋯ ∪ Ω c r \Omega\cup\Omega_{c_1} \cup\cdots\cup\Omega_{c_r} Ω ∪ Ω c 1 ∪ ⋯ ∪ Ω c r , which contradicts Hilbert’s irreducibility theorem [21 Chap. 3] . □
Remark 5.6. For m ≥ 2 m \ge2 m ≥ 2 the rational points of V m V_m V m lie, on a dense open set, over m m m -tuples of points ( x i , y i ) ∈ k 2 (x_i,y_i) \in k^2 ( x i , y i ) ∈ k 2 such that f ( x i , y i ) / f ( x 1 , y 1 ) ∈ k × n f(x_i,y_i)/f(x_1,y_1) \in k^{\times n} f ( x i , y i ) / f ( x 1 , y 1 ) ∈ k × n , and the specialization of B ~ m \widetilde{B}_m B m at such a point is again B f ( x 1 , y 1 ) B_{f(x_1,y_1)} B f ( x 1 , y 1 ) . Néron’s theorem gives injectivity of specialization outside a thin subset of V m ( k ) V_m(k) V m ( k ) , which is useful only if V m ( k ) V_m(k) V m ( k ) is not thin. We do not know whether this holds in general, and it was not established in [18 ] ; accordingly [18 Remark 1.5(2)] should be restricted to the case m = 1 m=1 m = 1 , which is Proposition 5.5 . Silverman’s theorem [24 ] , quoted there, concerns families over a curve.
Elliptic targets with j = 0 j=0 j = 0
Let n = 6 n=6 n = 6 , write ζ = ζ 6 \zeta=\zeta_6 ζ = ζ 6 , and for D ∈ k × D \in k^\times D ∈ k × let E D E_D E D be the elliptic curve y 2 = x 3 + D y^2=x^3+D y 2 = x 3 + D with the automorphism β ( x , y ) = ( ζ − 2 x , − y ) \beta(x,y)=(\zeta^{-2}x,-y) β ( x , y ) = ( ζ − 2 x , − y ) of order 6. Since β ∗ ( d x / y ) = ζ d x / y \beta^*(dx/y)=\zeta\,dx/y β ∗ ( d x / y ) = ζ d x / y , the automorphism β \beta β acts on T 0 E D T_0E_D T 0 E D by ζ \zeta ζ ; thus ( E D , β ) (E_D,\beta) ( E D , β ) is rigid with Σ E D ( 6 ) = { ζ } \Sigma_{E_D}(6)=\{\zeta\} Σ E D ( 6 ) = { ζ } . The fixed points of β \beta β are trivial, so E D β ( k ) = 0 E_D^\beta(k)=0 E D β ( k ) = 0 .
The L 1 L_1 L 1 -isomorphism g ( x 0 , y 0 ) = ( w 2 x 0 , w 3 y 0 ) g(x_0,y_0)=(w^2x_0,w^3y_0) g ( x 0 , y 0 ) = ( w 2 x 0 , w 3 y 0 ) from E D E_D E D to
Y 2 = X 3 + D f ( x , y ) Y^2=X^3+Df(x,y) Y 2 = X 3 + D f ( x , y )
satisfies u g = g ∘ β u g=g\circ\beta ug = g ∘ β for u = u τ u=u_{\tau} u = u τ , since u ( w ) = ζ − 1 w u(w)=\zeta^{-1}w u ( w ) = ζ − 1 w . Hence Y 2 = X 3 + D f ( x , y ) Y^2=X^3+Df(x,y) Y 2 = X 3 + D f ( x , y ) is the twist B ~ 1 \widetilde{B}_1 B 1 for ( B , β ) = ( E D , β ) (B,\beta)=(E_D,\beta) ( B , β ) = ( E D , β ) , and Theorem 3.4 gives, if X 6 ( k ) ≠ ∅ \mathcal{X}_6(k)\ne\varnothing X 6 ( k ) = ∅ ,
rk E D f ( K 1 ) = rk Z Hom k ( A , E D ) ( σ , β ) , E D f ( K 1 ) t o r s = 0. (5) \operatorname{rk} E_{Df}(K_1)=\operatorname{rk}_{\mathbb{Z}}\operatorname{Hom}_k(A,E_D)^{(\sigma,\beta)}, \qquad E_{Df}(K_1)_{\mathrm{tors}}=0. \tag*{(5)} rk E D f ( K 1 ) = rk Z Hom k ( A , E D ) ( σ , β ) , E D f ( K 1 ) tors = 0. (5)
A decomposition D f = G 2 + H 3 Df=G^2+H^3 D f = G 2 + H 3 with G , H ∈ k ( x , y ) G,H \in k(x,y) G , H ∈ k ( x , y ) gives the point ( − H , G ) (-H,G) ( − H , G ) . Over K m K_m K m , with u i = w i / w 1 u_i=w_i/w_1 u i = w i / w 1 , the same decomposition gives the points ( − H ( x i , y i ) / u i 2 , G ( x i , y i ) / u i 3 ) (-H(x_i,y_i)/u_i^2,G(x_i,y_i)/u_i^3) ( − H ( x i , y i ) / u i 2 , G ( x i , y i ) / u i 3 ) of Y 2 = X 3 + D f ( x 1 , y 1 ) Y^2=X^3+Df(x_1,y_1) Y 2 = X 3 + D f ( x 1 , y 1 ) , since u i 6 = f ( x i , y i ) / f ( x 1 , y 1 ) u_i^6=f(x_i,y_i)/f(x_1,y_1) u i 6 = f ( x i , y i ) / f ( x 1 , y 1 ) .
Lemma 5.7. Let F 1 F_1 F 1 be a field of characteristic 0 0 0 with − 3 ∉ F 1 \sqrt{-3}\notin F_1 − 3 ∈ / F 1 , and c ∈ F 1 × c\in F_1^\times c ∈ F 1 × . Then dim Q E c ( F 1 ( − 3 ) ) ⊗ Q = 2 dim Q E c ( F 1 ) ⊗ Q \dim_{\mathbb{Q}}E_c(F_1(\sqrt{-3}))\otimes\mathbb{Q}=2\dim_{\mathbb{Q}}E_c(F_1)\otimes\mathbb{Q} dim Q E c ( F 1 ( − 3 )) ⊗ Q = 2 dim Q E c ( F 1 ) ⊗ Q .
Proof. For the quadratic extension F 1 ( − 3 ) / F 1 F_1(\sqrt{-3})/F_1 F 1 ( − 3 ) / F 1 one has E c ( F 1 ( − 3 ) ) ⊗ Q ≅ E c ( F 1 ) ⊗ Q ⊕ E c ( − 3 ) ( F 1 ) ⊗ Q E_c(F_1(\sqrt{-3}))\otimes\mathbb{Q}\cong E_c(F_1)\otimes\mathbb{Q}\oplus E_c^{(-3)}(F_1)\otimes\mathbb{Q} E c ( F 1 ( − 3 )) ⊗ Q ≅ E c ( F 1 ) ⊗ Q ⊕ E c ( − 3 ) ( F 1 ) ⊗ Q , where E c ( − 3 ) : y 2 = x 3 − 27 c E_c^{(-3)}:y^2=x^3-27c E c ( − 3 ) : y 2 = x 3 − 27 c is the quadratic twist by − 3 -3 − 3 . The map ( x , y ) ↦ ( ( x 3 + 4 c ) / x 2 , y ( x 3 − 8 c ) / x 3 ) (x,y)\mapsto((x^3+4c)/x^2,y(x^3-8c)/x^3) ( x , y ) ↦ (( x 3 + 4 c ) / x 2 , y ( x 3 − 8 c ) / x 3 ) is an isogeny E c → E − 27 c E_c\to E_{-27c} E c → E − 27 c of degree 3 defined over F 1 F_1 F 1 , with kernel { O , ( 0 , ± c ) } \{O,(0,\pm\sqrt{c})\} { O , ( 0 , ± c )} , so the two summands have the same dimension. □
The dual of the Fermat cubic
In this section n = 6 n=6 n = 6 , ζ = ζ 6 \zeta=\zeta_{6} ζ = ζ 6 , ρ 3 = ζ 2 \rho_{3}=\zeta^{2} ρ 3 = ζ 2 , and
F ( x , y ) = x 6 + y 6 + 1 − 2 ( x 3 y 3 + x 3 + y 3 ) , F(x,y)=x^{6}+y^{6}+1-2(x^{3}y^{3}+x^{3}+y^{3}), F ( x , y ) = x 6 + y 6 + 1 − 2 ( x 3 y 3 + x 3 + y 3 ) ,
F ˉ ( u , v , w ) = u 6 + v 6 + w 6 − 2 ( u 3 v 3 + v 3 w 3 + w 3 u 3 ) . \bar{F}(u,v,w)=u^{6}+v^{6}+w^{6}-2(u^{3}v^{3}+v^{3}w^{3}+w^{3}u^{3}). F ˉ ( u , v , w ) = u 6 + v 6 + w 6 − 2 ( u 3 v 3 + v 3 w 3 + w 3 u 3 ) .
The Gauss map of the Fermat cubic x 3 + y 3 + z 3 = 0 x^{3}+y^{3}+z^{3}=0 x 3 + y 3 + z 3 = 0 sends ( x : y : z ) (x:y:z) ( x : y : z ) to the dual point ( x 2 : y 2 : z 2 ) (x^{2}:y^{2}:z^{2}) ( x 2 : y 2 : z 2 ) . If a = x 3 a=x^{3} a = x 3 , b = y 3 b=y^{3} b = y 3 and c = z 3 c=z^{3} c = z 3 , then
F ˉ ( x 2 , y 2 , z 2 ) = ( a − b − c ) ( a − b + c ) ( a + b − c ) ( a + b + c ) , \bar{F}(x^{2},y^{2},z^{2})=(a-b-c)(a-b+c)(a+b-c)(a+b+c), F ˉ ( x 2 , y 2 , z 2 ) = ( a − b − c ) ( a − b + c ) ( a + b − c ) ( a + b + c ) ,
so the image of the Gauss map lies on F ˉ = 0 \bar{F}=0 F ˉ = 0 . Since the dual of a smooth plane cubic is an irreducible sextic, C : F ˉ = 0 C:\bar{F}=0 C : F ˉ = 0 is the dual of the Fermat cubic, and F ( x , y ) = F ˉ ( x , y , 1 ) F(x,y)=\bar{F}(x,y,1) F ( x , y ) = F ˉ ( x , y , 1 ) .
The singular locus can also be read off directly. The partial derivatives are
F ˉ u = 6 u 2 ( u 3 − v 3 − w 3 ) , F ˉ v = 6 v 2 ( v 3 − w 3 − u 3 ) , F ˉ w = 6 w 2 ( w 3 − u 3 − v 3 ) . \bar{F}_{u}=6u^{2}(u^{3}-v^{3}-w^{3}),\qquad\bar{F}_{v}=6v^{2}(v^{3}-w^{3}-u^{3}),\qquad\bar{F}_{w}=6w^{2}(w^{3}-u^{3}-v^{3}). F ˉ u = 6 u 2 ( u 3 − v 3 − w 3 ) , F ˉ v = 6 v 2 ( v 3 − w 3 − u 3 ) , F ˉ w = 6 w 2 ( w 3 − u 3 − v 3 ) .
These equations show that a singular point has one zero coordinate and that the cubes of the other two coordinates are equal. Thus the singular points are exactly
( 0 : 1 : ϵ ) , ( 1 : 0 : ϵ ) , ( 1 : ϵ : 0 ) , ϵ 3 = 1. (0:1:\epsilon),\qquad(1:0:\epsilon),\qquad(1:\epsilon:0),\qquad\epsilon^{3}=1. ( 0 : 1 : ϵ ) , ( 1 : 0 : ϵ ) , ( 1 : ϵ : 0 ) , ϵ 3 = 1.
At ( 0 : 1 : 1 ) (0:1:1) ( 0 : 1 : 1 ) , on the chart v = 1 v=1 v = 1 with s = w − 1 s=w-1 s = w − 1 , the local equation has the form
9 s 2 − 4 u 3 + R ( u , s ) , R ( u , s ) ∈ ( s 3 , s u 3 , u 6 ) . 9s^{2}-4u^{3}+R(u,s),\qquad R(u,s)\in(s^{3},su^{3},u^{6}). 9 s 2 − 4 u 3 + R ( u , s ) , R ( u , s ) ∈ ( s 3 , s u 3 , u 6 ) .
Thus the tangent cone is the double line s = 0 s=0 s = 0 , while the restriction to this tangent has order 3; the singularity is an ordinary cusp. The group generated by permutations of the coordinates and by multiplication of a coordinate by a cube root of unity preserves F ˉ \bar{F} F ˉ and acts transitively on the nine singular points, so all nine are ordinary cusps. Its Alexander polynomial is ( t 2 − t + 1 ) 3 (t^{2}-t+1)^{3} ( t 2 − t + 1 ) 3 ; see [3 Introduction] and [4 Theorem 1.2.5] .
Put g 1 = x 3 + y 3 − 1 g_{1}=x^{3}+y^{3}-1 g 1 = x 3 + y 3 − 1 , g 2 = y 3 + 1 − x 3 g_{2}=y^{3}+1-x^{3} g 2 = y 3 + 1 − x 3 , g 3 = x 3 + 1 − y 3 g_{3}=x^{3}+1-y^{3} g 3 = x 3 + 1 − y 3 and h 1 = x y h_{1}=xy h 1 = x y , h 2 = y h_{2}=y h 2 = y , h 3 = x h_{3}=x h 3 = x . Then
F = g i 2 − 4 h i 3 ( i = 1 , 2 , 3 ) . (6) F=g_{i}^{2}-4h_{i}^{3}\qquad(i=1,2,3). \tag*{(6)} F = g i 2 − 4 h i 3 ( i = 1 , 2 , 3 ) . (6)
Consequently the points
Q i = ( 4 h i , 4 g i ) ( i = 1 , 2 , 3 ) Q_{i}=(4h_{i},4g_{i})\qquad(i=1,2,3) Q i = ( 4 h i , 4 g i ) ( i = 1 , 2 , 3 )
lie on E 16 F : Y 2 = X 3 + 16 F E_{16F}:Y^{2}=X^{3}+16F E 16 F : Y 2 = X 3 + 16 F over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) , and the points P i = ( 4 3 h i , g i ) P_{i}=(\sqrt[3]{4}h_{i},g_{i}) P i = ( 3 4 h i , g i ) lie on E F : Y 2 = X 3 + F E_{F}:Y^{2}=X^{3}+F E F : Y 2 = X 3 + F over Q ( 2 3 ) ( x , y ) \mathbb{Q}(\sqrt[3]{2})(x,y) Q ( 3 2 ) ( x , y ) . The map ( X , Y ) ↦ ( 16 3 X , 4 Y ) (X,Y)\mapsto(\sqrt[3]{16}X,4Y) ( X , Y ) ↦ ( 3 16 X , 4 Y ) is an isomorphism E F → E 16 F E_{F}\to E_{16F} E F → E 16 F over Q ( 2 3 ) ( x , y ) \mathbb{Q}(\sqrt[3]{2})(x,y) Q ( 3 2 ) ( x , y ) sending P i P_{i} P i to Q i Q_{i} Q i . We write [ ρ 3 ] [\rho_{3}] [ ρ 3 ] for the automorphism ( X , Y ) ↦ ( ρ 3 X , Y ) (X,Y)\mapsto(\rho_{3}X,Y) ( X , Y ) ↦ ( ρ 3 X , Y ) of these curves.
Theorem 6.1. (i) E 16 F ( Q ( x , y ) ) E_{16F}(\mathbb{Q}(x,y)) E 16 F ( Q ( x , y )) has rank 3, and Q 1 , Q 2 , Q 3 Q_{1},Q_{2},Q_{3} Q 1 , Q 2 , Q 3 generate a subgroup of finite index. E 16 F ( Q ( ζ 3 ) ( x , y ) ) E_{16F}(\mathbb{Q}(\zeta_{3})(x,y)) E 16 F ( Q ( ζ 3 ) ( x , y )) has rank 6, and Q i , [ ρ 3 ] Q i Q_{i},[\rho_{3}]Q_{i} Q i , [ ρ 3 ] Q i ( i = 1 , 2 , 3 ) (i=1,2,3) ( i = 1 , 2 , 3 ) generate a subgroup of finite index.
(ii) E F E_{F} E F has rank 0 over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) and over Q ( ζ 3 ) ( x , y ) \mathbb{Q}(\zeta_{3})(x,y) Q ( ζ 3 ) ( x , y ) , rank 3 over Q ( 2 3 ) ( x , y ) \mathbb{Q}(\sqrt[3]{2})(x,y) Q ( 3 2 ) ( x , y ) and rank 6 over Q ( ζ 3 , 2 3 ) ( x , y ) \mathbb{Q}(\zeta_{3},\sqrt[3]{2})(x,y) Q ( ζ 3 , 3 2 ) ( x , y ) . Moreover E F ( Q ( ζ 3 ) ( x , y ) ) = 0 E_{F}(\mathbb{Q}(\zeta_{3})(x,y))=0 E F ( Q ( ζ 3 ) ( x , y )) = 0 ; in particular F F F has no decomposition F = G 2 + H 3 F=G^{2}+H^{3} F = G 2 + H 3 with G , H ∈ Q ( ζ 3 ) ( x , y ) G,H\in\mathbb{Q}(\zeta_{3})(x,y) G , H ∈ Q ( ζ 3 ) ( x , y ) .
(iii) Let A = Alb ( X 6 ) A=\operatorname{Alb}(X_{6}) A = Alb ( X 6 ) for w 6 = F ( x , y ) w^{6}=F(x,y) w 6 = F ( x , y ) and σ = τ ˉ ∗ \sigma=\bar{\tau}_{*} σ = τ ˉ ∗ . Then dim A = 3 \dim A=3 dim A = 3 , ( A , σ ) (A,\sigma) ( A , σ ) satisfies Assumption 1.2 of [18 ] , and A A A is isogenous over Q ( ζ 3 ) \mathbb{Q}(\zeta_{3}) Q ( ζ 3 ) , compatibly with σ \sigma σ and β × β × β \beta\times\beta\times\beta β × β × β , to E 16 3 E_{16}^{3} E 16 3 , where E 16 : y 2 = x 3 + 16 E_{16}:y^{2}=x^{3}+16 E 16 : y 2 = x 3 + 16 . For every number
field k ⊇ Q ( ζ 3 ) k \supseteq\mathbb{Q}(\zeta_{3}) k ⊇ Q ( ζ 3 ) and every m ≥ 1 m \ge1 m ≥ 1 , the group Alb ( X ~ 6 ) ( k ( V m ) ) \operatorname{Alb}(\widetilde{X}_{6})(k(V_{m})) Alb ( X 6 ) ( k ( V m )) is free of rank 18 m = 2 m d 6 2 18m = 2md_{6}^{2} 18 m = 2 m d 6 2
(iv) For all ( a , b ) ∈ Q 2 (a,b) \in\mathbb{Q}^{2} ( a , b ) ∈ Q 2 outside a thin set, F ( a , b ) ≠ 0 F(a,b) \ne0 F ( a , b ) = 0 and the points Q i ( a , b ) Q_i(a,b) Q i ( a , b ) are independent in the group of rational points of the Mordell curve y 2 = x 3 + 16 F ( a , b ) y^{2} = x^{3} + 16F(a,b) y 2 = x 3 + 16 F ( a , b ) , which therefore has rank at least 3 over Q \mathbb{Q} Q . For ( a , b ) = ( 2 , 3 ) (a,b) = (2,3) ( a , b ) = ( 2 , 3 ) one has Q 1 + Q 2 = 2 Q 3 Q_{1} + Q_{2} = 2Q_{3} Q 1 + Q 2 = 2 Q 3 , so the exceptional set is not empty.
Proof. Geometric rank. By [3 Theorem 1.1] , the rank of u 2 + v 3 = F ( x , y ) u^{2} + v^{3} = F(x,y) u 2 + v 3 = F ( x , y ) over C ( x , y ) \mathbb{C}(x,y) C ( x , y ) equals deg Δ C = 6 \deg\Delta_{C} = 6 deg Δ C = 6 . (The statement there is for an affine equation F ‾ ( x , y , 1 ) \overline{F}(x,y,1) F ( x , y , 1 ) ; changing the line at infinity multiplies F ‾ \overline{F} F by the sixth power of a rational function, which does not change the curve over C ( P 2 ) \mathbb{C}(\mathbb{P}^{2}) C ( P 2 ) up to isomorphism.) Since 16 is a sixth power in C \mathbb{C} C , the curves E F E_{F} E F and E 16 F E_{16F} E 16 F are isomorphic over C ( x , y ) \mathbb{C}(x,y) C ( x , y ) , so both have rank 6 over C ( x , y ) \mathbb{C}(x,y) C ( x , y ) and at most 6 over any subfield.
Independence. The Weierstrass equation of E 16 F E_{16F} E 16 F defines an elliptic curve over the open set U : F ≠ 0 U:F \ne0 U : F = 0 of A Q 2 \mathbb{A}^{2}_{\mathbb{Q}} A Q 2 , and Q 1 , Q 2 , Q 3 Q_{1},Q_{2},Q_{3} Q 1 , Q 2 , Q 3 are sections over U U U , since their coordinates are polynomials. Because U U U is integral, a relation ∑ n i Q i = 0 \sum n_iQ_i = 0 ∑ n i Q i = 0 over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) holds for the sections and hence in every fibre. Take ( a , b ) = ( 2 , 5 ) (a,b) = (2,5) ( a , b ) = ( 2 , 5 ) , so that 16 F ( 2 , 5 ) = 214784 = 2 8 ⋅ 839 16F(2,5) = 214784 = 2^{8} \cdot839 16 F ( 2 , 5 ) = 214784 = 2 8 ⋅ 839 and
Q 1 ( 2 , 5 ) = ( 40 , 528 ) , Q 2 ( 2 , 5 ) = ( 20 , 472 ) , Q 3 ( 2 , 5 ) = ( 8 , − 464 ) Q_{1}(2,5) = (40,528), \qquad Q_{2}(2,5) = (20,472), \qquad Q_{3}(2,5) = (8,-464) Q 1 ( 2 , 5 ) = ( 40 , 528 ) , Q 2 ( 2 , 5 ) = ( 20 , 472 ) , Q 3 ( 2 , 5 ) = ( 8 , − 464 )
on E : y 2 = x 3 + 214784 E:y^{2} = x^{3} + 214784 E : y 2 = x 3 + 214784 . For p ∈ { 19 , 41 , 67 } p \in\{19,41,67\} p ∈ { 19 , 41 , 67 } the curve E E E has good reduction, with
∣ E ( F 19 ) ∣ = 28 , ∣ E ( F 41 ) ∣ = 42 , ∣ E ( F 67 ) ∣ = 63. \lvert E(\mathbb{F}_{19})\rvert= 28,\qquad\lvert E(\mathbb{F}_{41})\rvert= 42,\qquad\lvert E(\mathbb{F}_{67})\rvert= 63. ∣ E ( F 19 )∣ = 28 , ∣ E ( F 41 )∣ = 42 , ∣ E ( F 67 )∣ = 63.
Thus the 7-primary part is cyclic of order 7 in each case. Take generators
R 19 = ( 14 , 15 ) , R 41 = ( 6 , 23 ) , R 67 = ( 40 , 8 ) R_{19} = (14,15),\qquad R_{41} = (6,23),\qquad R_{67} = (40,8) R 19 = ( 14 , 15 ) , R 41 = ( 6 , 23 ) , R 67 = ( 40 , 8 )
of these order-7 subgroups. After multiplication by the prime-to-7 cofactors 4, 6, 9, respectively, the reductions of Q 1 ( 2 , 5 ) , Q 2 ( 2 , 5 ) , Q 3 ( 2 , 5 ) Q_{1}(2,5),Q_{2}(2,5),Q_{3}(2,5) Q 1 ( 2 , 5 ) , Q 2 ( 2 , 5 ) , Q 3 ( 2 , 5 ) have coefficient matrix
( 1 6 5 1 3 6 5 4 4 ) over F 7 , \begin{pmatrix}
1 & 6 & 5 \\
1 & 3 & 6 \\
5 & 4 & 4
\end{pmatrix}
\quad\text{over }\mathbb{F}_{7}, 1 1 5 6 3 4 5 6 4 over F 7 ,
where the rows correspond to p = 19 , 41 , 67 p = 19,41,67 p = 19 , 41 , 67 . Its determinant is 5 modulo 7, so the three classes are independent in the product of the three quotients. Since ∣ E ( F 5 ) ∣ = 6 \lvert E(\mathbb{F}_{5})\rvert= 6 ∣ E ( F 5 )∣ = 6 , good reduction at 5 also gives E ( Q ) [ 7 ] = 0 E(\mathbb{Q})[7] = 0 E ( Q ) [ 7 ] = 0 . If ∑ n i Q i ( 2 , 5 ) = 0 \sum n_iQ_i(2,5) = 0 ∑ n i Q i ( 2 , 5 ) = 0 with not all n i n_i n i zero, we may divide the coefficients by their common power of 7; the absence of rational 7-torsion shows that the resulting relation is still zero. Reducing modulo the three primes then contradicts the nonsingularity of the matrix above. Hence Q 1 ( 2 , 5 ) , Q 2 ( 2 , 5 ) , Q 3 ( 2 , 5 ) Q_{1}(2,5),Q_{2}(2,5),Q_{3}(2,5) Q 1 ( 2 , 5 ) , Q 2 ( 2 , 5 ) , Q 3 ( 2 , 5 ) are independent, and so are Q 1 , Q 2 , Q 3 Q_{1},Q_{2},Q_{3} Q 1 , Q 2 , Q 3 .
Proof of (i). Let c c c denote the non-trivial automorphism of the extension Q ( ζ 3 ) ( x , y ) / Q ( x , y ) \mathbb{Q}(\zeta_{3})(x,y)/\mathbb{Q}(x,y) Q ( ζ 3 ) ( x , y ) / Q ( x , y ) . The Q \mathbb{Q} Q -vector space M = E 16 F ( Q ( ζ 3 ) ( x , y ) ) ⊗ Q M = E_{16F}(\mathbb{Q}(\zeta_{3})(x,y)) \otimes\mathbb{Q} M = E 16 F ( Q ( ζ 3 ) ( x , y )) ⊗ Q is the direct sum of its ( ± 1 ) (\pm1) ( ± 1 ) -eigenspaces for c c c , and the ( + 1 ) (+1) ( + 1 ) -eigenspace is E 16 F ( Q ( x , y ) ) ⊗ Q E_{16F}(\mathbb{Q}(x,y)) \otimes\mathbb{Q} E 16 F ( Q ( x , y )) ⊗ Q . The endomorphism [ − 3 ] = [ ρ 3 ] − [ ρ 3 2 ] [\sqrt{-3}] = [\rho_{3}] - [\rho_{3}^{2}] [ − 3 ] = [ ρ 3 ] − [ ρ 3 2 ] is an isogeny satisfying c ∘ [ − 3 ] = − [ − 3 ] ∘ c c \circ[\sqrt{-3}] = -[\sqrt{-3}] \circ c c ∘ [ − 3 ] = − [ − 3 ] ∘ c , so it maps the ( + 1 ) (+1) ( + 1 ) -eigenspace injectively into the ( − 1 ) (-1) ( − 1 ) -eigenspace. Hence Q 1 , Q 2 , Q 3 , [ − 3 ] Q 1 , [ − 3 ] Q 2 , [ − 3 ] Q 3 Q_{1},Q_{2},Q_{3},[\sqrt{-3}]Q_{1},[\sqrt{-3}]Q_{2},[\sqrt{-3}]Q_{3} Q 1 , Q 2 , Q 3 , [ − 3 ] Q 1 , [ − 3 ] Q 2 , [ − 3 ] Q 3 are independent, and so are the Q i , [ ρ 3 ] Q i Q_i,[\rho_{3}]Q_i Q i , [ ρ 3 ] Q i , since [ ρ 3 ] = ( − 1 + [ − 3 ] ) / 2 [\rho_{3}] = (-1 + [\sqrt{-3}])/2 [ ρ 3 ] = ( − 1 + [ − 3 ]) /2 . Together with the upper bound 6, this gives rank 6 over Q ( ζ 3 ) ( x , y ) \mathbb{Q}(\zeta_{3})(x,y) Q ( ζ 3 ) ( x , y ) , with the Q i , [ ρ 3 ] Q i Q_i,[\rho_{3}]Q_i Q i , [ ρ 3 ] Q i spanning M M M , and rank 3 over Q ( x , y ) \mathbb{Q}(x,y) Q ( x , y ) with the Q i Q_i Q i spanning the ( + 1 ) (+1) ( + 1 ) -eigenspace, in accordance with Lemma 5.7.
Proof of (ii). Let M = Q ( ζ 3 , 2 3 ) M = \mathbb{Q}(\zeta_3,\sqrt[3]{2}) M = Q ( ζ 3 , 3 2 ) , a Galois extension of Q \mathbb{Q} Q with group S 3 S_3 S 3 generated by r : 2 3 ↦ ρ 3 2 3 , ρ 3 ↦ ρ 3 r:\sqrt[3]{2}\mapsto\rho_3\sqrt[3]{2},\ \rho_3\mapsto\rho_3 r : 3 2 ↦ ρ 3 3 2 , ρ 3 ↦ ρ 3 , and c : ρ 3 ↦ ρ 3 2 , 2 3 ↦ 2 3 c:\rho_3\mapsto\rho_3^2,\ \sqrt[3]{2}\mapsto\sqrt[3]{2} c : ρ 3 ↦ ρ 3 2 , 3 2 ↦ 3 2 . By (i) and the isomorphism E F ≅ E 16 F E_F \cong E_{16F} E F ≅ E 16 F over M ( x , y ) M(x,y) M ( x , y ) , the space N = E F ( M ( x , y ) ) ⊗ Q \mathcal{N}=E_F(M(x,y))\otimes\mathbb{Q} N = E F ( M ( x , y )) ⊗ Q has dimension 6 and is spanned by P i , [ ρ 3 ] P i P_i,[\rho_3]P_i P i , [ ρ 3 ] P i . For s ∈ Gal ( M / Q ) s\in\operatorname{Gal}(M/\mathbb{Q}) s ∈ Gal ( M / Q ) we have s ( [ ρ 3 ] P ) = [ s ( ρ 3 ) ] s ( P ) s([\rho_3]P)=[s(\rho_3)]s(P) s ([ ρ 3 ] P ) = [ s ( ρ 3 )] s ( P ) , and r ( P i ) = [ ρ 3 2 ] P i r(P_i)=[\rho_3^2]P_i r ( P i ) = [ ρ 3 2 ] P i because r ( 4 3 ) = ρ 3 2 4 3 r(\sqrt[3]{4})=\rho_3^2\sqrt[3]{4} r ( 3 4 ) = ρ 3 2 3 4 , while c ( P i ) = P i c(P_i)=P_i c ( P i ) = P i . Thus N = ⨁ i Q ( ρ 3 ) P i \mathcal{N}=\bigoplus_i\mathbb{Q}(\rho_3)P_i N = ⨁ i Q ( ρ 3 ) P i , where r r r acts on each summand as multiplication by ρ 3 2 \rho_3^2 ρ 3 2 and c c c as complex conjugation of the coefficient. The invariants of ⟨ r ⟩ \langle r\rangle ⟨ r ⟩ are 0, those of ⟨ c ⟩ \langle c\rangle ⟨ c ⟩ are ⨁ i Q P i \bigoplus_i\mathbb{Q}P_i ⨁ i Q P i , and those of S 3 S_3 S 3 are 0. Since E F ( k ′ ( x , y ) ) ⊗ Q = N Gal ( M / k ′ ) E_F(k'(x,y))\otimes\mathbb{Q}=\mathcal{N}^{\operatorname{Gal}(M/k')} E F ( k ′ ( x , y )) ⊗ Q = N Gal ( M / k ′ ) for every subfield k ′ ⊆ M k'\subseteq M k ′ ⊆ M , the ranks are as stated. By (5.1) with D = 1 D=1 D = 1 and k = Q ( ζ 3 ) k=\mathbb{Q}(\zeta_3) k = Q ( ζ 3 ) , the group E F ( Q ( ζ 3 ) ( x , y ) ) E_F(\mathbb{Q}(\zeta_3)(x,y)) E F ( Q ( ζ 3 ) ( x , y )) is torsion free; here X 6 ( k ) ≠ ∅ X_6(k)\ne\varnothing X 6 ( k ) = ∅ because ( 0 , 0 , 1 ) (0,0,1) ( 0 , 0 , 1 ) is a non-singular rational point of w 6 = F ( x , y ) w^6=F(x,y) w 6 = F ( x , y ) (see § 4.1 ). Hence E F ( Q ( ζ 3 ) ( x , y ) ) = 0 E_F(\mathbb{Q}(\zeta_3)(x,y))=0 E F ( Q ( ζ 3 ) ( x , y )) = 0 , and a decomposition F = G 2 + H 3 F=G^2+H^3 F = G 2 + H 3 would give the non-zero point ( − H , G ) (-H,G) ( − H , G ) .
Proof of (iii). Let k = Q ( ζ 3 ) k=\mathbb{Q}(\zeta_3) k = Q ( ζ 3 ) and B = E 16 B=E_{16} B = E 16 with β \beta β as in Section 5. By (5.1) with D = 16 D=16 D = 16 and part (i), Hom k ( A , B ) ( σ , β ) \operatorname{Hom}_k(A,B)^{(\sigma,\beta)} Hom k ( A , B ) ( σ , β ) has rank 6. For a sextic whose only singularities are nodes and cusps, the first Betti number of the associated sixfold cyclic multiple plane is twice the multiplicity of t 2 − t + 1 t^2-t+1 t 2 − t + 1 in the Alexander polynomial [11 , 3 ] ; see the summary in [9 §1] ; here this gives b 1 ( X 6 ) = 6 b_1(X_6)=6 b 1 ( X 6 ) = 6 and dim A = 3 \dim A=3 dim A = 3 . Since V B = V B , 6 V_B=V_{B,6} V B = V B , 6 and e 6 ( B ) = 1 e_6(B)=1 e 6 ( B ) = 1 , Corollary 4.3(i), applied over C \mathbb{C} C , gives 6 ≤ ϕ ( 6 ) e 6 ( A ) = 2 e 6 ( A ) 6\leq\phi(6)e_6(A)=2e_6(A) 6 ≤ ϕ ( 6 ) e 6 ( A ) = 2 e 6 ( A ) ; as 2 dim A = 6 = ∑ d ϕ ( d ) e d ( A ) 2\dim A=6=\sum_d\phi(d)e_d(A) 2 dim A = 6 = ∑ d ϕ ( d ) e d ( A ) , we get e 6 ( A ) = 3 e_6(A)=3 e 6 ( A ) = 3 and V A = V A , 6 V_A=V_{A,6} V A = V A , 6 . Equality holds in Corollary 4.3(i), so for every λ \lambda λ either W A ( λ ) = 0 W_A(\lambda)=0 W A ( λ ) = 0 or W B ( λ ) = H B ( λ ) W_B(\lambda)=H_B(\lambda) W B ( λ ) = H B ( λ ) . Since β \beta β acts on T 0 B T_0B T 0 B by ζ \zeta ζ , we have W B ( ζ ˉ ) = 0 ≠ H B ( ζ ˉ ) W_B(\bar{\zeta})=0\ne H_B(\bar{\zeta}) W B ( ζ ˉ ) = 0 = H B ( ζ ˉ ) , hence W A ( ζ ˉ ) = 0 W_A(\bar{\zeta})=0 W A ( ζ ˉ ) = 0 and W A = H A ( ζ ) W_A=H_A(\zeta) W A = H A ( ζ ) . Thus ( A , σ ) (A,\sigma) ( A , σ ) is rigid with V A = V A , 6 V_A=V_{A,6} V A = V A , 6 , which is Assumption 1.2.
Choose elements φ 1 , φ 2 , φ 3 \varphi_1,\varphi_2,\varphi_3 φ 1 , φ 2 , φ 3 of Hom k ( A , B ) ( σ , β ) \operatorname{Hom}_k(A,B)^{(\sigma,\beta)} Hom k ( A , B ) ( σ , β ) forming a basis of the Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) -vector space Hom k ( A , B ) ( σ , β ) ⊗ Q \operatorname{Hom}_k(A,B)^{(\sigma,\beta)}\otimes\mathbb{Q} Hom k ( A , B ) ( σ , β ) ⊗ Q . The preceding dimension count and Corollary 4.3(i) identify this space with Hom Q ( ζ ) ( V A , V B ) \operatorname{Hom}_{\mathbb{Q}(\zeta)}(V_A,V_B) Hom Q ( ζ ) ( V A , V B ) . Since V B V_B V B is one-dimensional over Q ( ζ ) \mathbb{Q}(\zeta) Q ( ζ ) , the map ( φ 1 , φ 2 , φ 3 ) : A → B 3 (\varphi_1,\varphi_2,\varphi_3):A\to B^3 ( φ 1 , φ 2 , φ 3 ) : A → B 3 induces an isomorphism on V A V_A V A , so it is an isogeny defined over k k k and intertwining σ \sigma σ with β × β × β \beta\times\beta\times\beta β × β × β . As End k ( B ) ≅ Z [ ρ 3 ] \operatorname{End}_k(B)\cong\mathbb{Z}[\rho_3] End k ( B ) ≅ Z [ ρ 3 ] is commutative, End k ( A ) σ ⊗ Q ≅ M 3 ( Q ( ζ ) ) \operatorname{End}_k(A)^\sigma\otimes\mathbb{Q}\cong M_3(\mathbb{Q}(\zeta)) End k ( A ) σ ⊗ Q ≅ M 3 ( Q ( ζ )) has rank 18 = ϕ ( 6 ) ⋅ 3 2 18=\phi(6)\cdot3^2 18 = ϕ ( 6 ) ⋅ 3 2 , which is also the rank of End k ˉ ( A ) σ \operatorname{End}_{\bar{k}}(A)^\sigma End k ˉ ( A ) σ by Corollary 4.4. By Theorem 3.4(b), the rank of Alb ( X ~ 6 ) ( k ( V m ) ) \operatorname{Alb}(\widetilde{X}_6)(k(V_m)) Alb ( X 6 ) ( k ( V m )) is 18 m 18m 18 m for every number field k ⊇ Q ( ζ 3 ) k\supseteq\mathbb{Q}(\zeta_3) k ⊇ Q ( ζ 3 ) , and by Proposition 5.3 (with e 2 = e 3 = 0 e_2=e_3=0 e 2 = e 3 = 0 ) the group is torsion free. In this case A σ ( k ) = 0 A^\sigma(k)=0 A σ ( k ) = 0 , so Theorem 3.4(a) embeds the Mordell–Weil group into the free abelian group ( End k ( A ) σ ) m (\operatorname{End}_k(A)^\sigma)^m ( End k ( A ) σ ) m ; it is therefore free.
Proof of (iv). Apply Néron’s specialization theorem [22 §11.1] to E 16 F E_{16F} E 16 F over U U U and use (i). The relation at ( 2 , 3 ) (2,3) ( 2 , 3 ) , where 16 F ( 2 , 3 ) = 4672 16F(2,3)=4672 16 F ( 2 , 3 ) = 4672 and Q 1 = ( 24 , 136 ) Q_1=(24,136) Q 1 = ( 24 , 136 ) , Q 2 = ( 12 , 80 ) Q_2=(12,80) Q 2 = ( 12 , 80 ) , Q 3 = ( 8 , − 72 ) Q_3=(8,-72) Q 3 = ( 8 , − 72 ) , is checked directly. □
Remark 6.2. The field Q ( 2 3 ) \mathbb{Q}(\sqrt[3]{2}) Q ( 3 2 ) enters through the coefficient − 4 -4 − 4 in (6.1). By Theorem 6.1(ii), the equivariant homomorphisms from A A A to the curve y 2 = x 3 + 1 y^2=x^3+1 y 2 = x 3 + 1 are defined only over Q ( ζ 3 , 2 3 ) \mathbb{Q}(\zeta_3,\sqrt[3]{2}) Q ( ζ 3 , 3 2 ) , while by (iii) those from A A A to E 16 E_{16} E 16 are defined over Q ( ζ 3 ) \mathbb{Q}(\zeta_3) Q ( ζ 3 ) : in the notation of Theorem 5.1, with k = Q ( ζ 3 ) k=\mathbb{Q}(\zeta_3) k = Q ( ζ 3 ) , one has k H = Q ( ζ 3 , 2 3 ) k_H=\mathbb{Q}(\zeta_3,\sqrt[3]{2}) k H = Q ( ζ 3 , 3 2 ) for ( B , β ) = ( E D , β ) (B,\beta)=(E_D,\beta) ( B , β ) = ( E D , β ) with D = 1 D=1 D = 1 , whereas k H = Q ( ζ 3 ) k_H=\mathbb{Q}(\zeta_3) k H = Q ( ζ 3 ) for ( B , β ) = ( E 16 , β ) (B,\beta)=(E_{16},\beta) ( B , β ) = ( E 16 , β ) and for ( B , β ) = ( A , σ ) (B,\beta)=(A,\sigma) ( B , β ) = ( A , σ ) .
Corrections to [17], [19] and [18]
We list the statements of [17 ] , of its corrigendum [19 ] , and of [18 ] that have to be corrected, together with the results of this paper that replace them.
The paper [17 ] and its corrigendum [19 ] .
(a) [17 Theorem 1.1] , its amended form in [19 (1),(2)] , and [18 Theorem 3.3] are replaced by Theorem 2.2 and Corollary 2.4. Example 2.7, with k k k a number field containing ζ 3 \zeta_3 ζ 3 , satisfies the hypotheses of all three statements and contradicts the rank formula, both as the inequality of [17 ] and as the equality of [19 ] ; if moreover 2 3 ∈ k \sqrt[3]{2} \in k 3 2 ∈ k , it also contradicts the torsion term A [ n ] ( k ) A[n](k) A [ n ] ( k ) . The error in the proof is explained in Remark 2.5. The restriction to smooth projective varieties made in [19 (1)] is part of Setting 2.1.
(b) [17 Theorem 1.2] and its amended form in [19 (3)] are replaced by Corollary 3.8. The torsion subgroup is J ι ~ ( k ) J^{\tilde{\iota}}(k) J ι ~ ( k ) and not J [ n ] ( k ) J[n](k) J [ n ] ( k ) (Example 3.9), and the rank formula holds if and only if ι ~ \tilde{\iota} ι ~ is central in End k ( J ) ⊗ Q \operatorname{End}_k(J) \otimes\mathbb{Q} End k ( J ) ⊗ Q . The deduction of 1 + ι ~ + ⋯ + ι ~ n − 1 = 0 1+\tilde{\iota}+\cdots+\tilde{\iota}^{n-1}=0 1 + ι ~ + ⋯ + ι ~ n − 1 = 0 in [17 §5] and the assertion there about independent generators are replaced by Lemma 3.2 and Corollary 3.8.
(c) Misprints in [17 §5] , in the notation there: on p. 335, in the description of K = L G K=L^G K = L G , the last generator y 1 s − 1 y n − 1 y_1^{s^{-1}}y_{n-1} y 1 s − 1 y n − 1 should read y 1 s − 1 y n y_1^{s^{-1}}y_n y 1 s − 1 y n ; and in (1) on the same page, the points should read P 1 = ( x 1 , 1 ) P_1=(x_1,1) P 1 = ( x 1 , 1 ) and P i + 1 = ( x i + 1 , z i / f ( x 1 ) ) P_{i+1}=(x_{i+1},z_i/f(x_1)) P i + 1 = ( x i + 1 , z i / f ( x 1 )) for 1 ≤ i ≤ n − 1 1\leq i\leq n-1 1 ≤ i ≤ n − 1 .
The paper [18 ] .
[18 Theorem 3.3] is treated in item (a) above.
[18 Theorem 1.1] is replaced by Theorem 3.4(b),(c): End k ( Alb ( X n ) ) \operatorname{End}_k(\operatorname{Alb}(X_n)) End k ( Alb ( X n )) must be replaced by the centralizer of τ ˉ ∗ \bar{\tau}_* τ ˉ ∗ , and the torsion term Alb ( X n ) [ m ] ( k ) \operatorname{Alb}(X_n)[m](k) Alb ( X n ) [ m ] ( k ) by A σ ( k ) A^\sigma(k) A σ ( k ) .
The deduction of id + γ ~ + ⋯ + γ ~ n − 1 = 0 \operatorname{id}+\tilde{\gamma}+\cdots+\tilde{\gamma}^{n-1}=0 id + γ ~ + ⋯ + γ ~ n − 1 = 0 in [18 §5] is replaced by Lemma 3.2.
[18 Theorem 1.3] is replaced by Corollary 5.2. Under Assumption 1.2 the rank is at most m ⋅ 4 d n 2 / ϕ ( n ) m\cdot4d_n^2/\phi(n) m ⋅ 4 d n 2 / ϕ ( n ) for every n ≥ 3 n\geq3 n ≥ 3 ; the value 2 d n 2 2d_n^2 2 d n 2 given for n = 5 , 10 n=5,10 n = 5 , 10 is incorrect when d n > 0 d_n>0 d n > 0 . The equalities hold only if the relevant endomorphisms are defined over k k k (the decompositions in [18 Prop. 4.2] are over C \mathbb{C} C ). The restriction on n n n is unnecessary.
The argument for n = 8 n=8 n = 8 in [18 §6] is incorrect. The curve y 2 = x 3 − x 2 − 3 x + 1 y^2=x^3-x^2-3x+1 y 2 = x 3 − x 2 − 3 x + 1 , denoted E 2 E_2 E 2 in [18 ] , has c 4 = 160 c_4=160 c 4 = 160 and discriminant 2 6 ⋅ 37 2^6\cdot37 2 6 ⋅ 37 , so j ( E 2 ) = 64000 / 37 ∉ Z j(E_2)=64000/37\notin\mathbb{Z} j ( E 2 ) = 64000/37 ∈ / Z and E 2 E_2 E 2 has no complex multiplication; in particular J ( C 2 ) J(C_2) J ( C 2 ) is not isogenous to E 1 × E 2 E_1\times E_2 E 1 × E 2 . By Remark 4.6(2), J ( C 2 ) J(C_2) J ( C 2 ) is isogenous over C \mathbb{C} C to the square of an elliptic curve with complex multiplication by Z [ − 2 ] \mathbb{Z}[\sqrt{-2}] Z [ − 2 ] ; for instance the curve y 2 = x 3 + x 2 − 3 x + 1 y^2=x^3+x^2-3x+1 y 2 = x 3 + x 2 − 3 x + 1 , denoted E 1 E_1 E 1 in [18 ] , which has j j j -invariant 8000. Also, M 2 n 2 ( Z [ i ] ) M_{2n_2}(\mathbb{Z}[i]) M 2 n 2 ( Z [ i ]) has rank 8 n 2 2 8n_2^2 8 n 2 2 , not 4 n 2 2 4n_2^2 4 n 2 2 . Under Assumption 1.2, one of n 1 , n 2 n_1,n_2 n 1 , n 2 in [18 Prop. 4.2(v),(vi)] is zero (Remark 4.6).
[18 Theorem 1.4] is replaced by the discussion after Corollary 5.2. For the dual of the Fermat cubic the conclusion 2 m d 6 2 = 18 m 2md_6^2=18m 2 m d 6 2 = 18 m holds over every number field containing Q ( ζ 3 ) \mathbb{Q}(\zeta_3) Q ( ζ 3 ) (Theorem 6.1(iii)). For the curve in [18 Theorem 1.4(ii)] , Assumption 1.2 was not verified.
[18 Remark 1.5(1)] holds in the form of Remark 3.7. [18 Remark 1.5(2)] is justified only for m = 1 m=1 m = 1 (Proposition 5.5 and Remark 5.6).
Misprints in [18 ] :
p. 861: “r r r is a positive number” should read “r r r is a non-negative integer”.
p. 863: “the function fields of U m U_m U m and V m V_m V m respectively” should read “the function fields of V m V_m V m and U m U_m U m , respectively”; “the function field of W m W_m W m ” should read “the function field of V m V_m V m ”.
p. 864, Theorem 1.4(ii), and p. 875: the affine equation should be f ( x , y ) = F ( 1 , x , y ) = x 3 − 3 x y ( y 3 − 8 ) + 2 ( y 6 + 20 y 3 − 8 ) f(x,y)=F(1,x,y)=x^3-3xy(y^3-8)+2(y^6+20y^3-8) f ( x , y ) = F ( 1 , x , y ) = x 3 − 3 x y ( y 3 − 8 ) + 2 ( y 6 + 20 y 3 − 8 ) , with F F F as in [18 Prop. 4.4(ii)] ; the sign of 20 y 3 20y^3 20 y 3 is misprinted.
p. 869: “e ≥ n / r e \ge n/r e ≥ n / r ” should read “e ≥ r / n e \ge r/n e ≥ r / n ”, and the weighted projective space P ( 1 , 1 , 1 , n 0 ) 3 \mathbb{P}^{3}_{(1,1,1,n_0)} P ( 1 , 1 , 1 , n 0 ) 3 should be P ( 1 , 1 , 1 , e ) 3 \mathbb{P}^{3}_{(1,1,1,e)} P ( 1 , 1 , 1 , e ) 3 , so that both sides of u 3 n = u 0 n 0 F ( u 0 , u 1 , u 2 ) u_3^n=u_0^{n_0}F(u_0,u_1,u_2) u 3 n = u 0 n 0 F ( u 0 , u 1 , u 2 ) have degree n e = n 0 + r ne=n_0+r n e = n 0 + r .
p. 871, Prop. 4.4(ii): the factor ( u 2 3 − 8 ) (u_2^3-8) ( u 2 3 − 8 ) should read ( u 2 3 − 8 u 0 3 ) (u_2^3-8u_0^3) ( u 2 3 − 8 u 0 3 ) .
p. 874: End k ( Alb ( X 5 ) ) \operatorname{End}_k(\operatorname{Alb}(X_5)) End k ( Alb ( X 5 )) should read End k ( Alb ( X 8 ) ) \operatorname{End}_k(\operatorname{Alb}(X_8)) End k ( Alb ( X 8 )) (three times).
p. 875, proof of Theorem 1.4: “2 m d 6 2md_6 2 m d 6 ” should read “2 m d 6 2 2md_6^2 2 m d 6 2 ”.
p. 875, reference [9 ] : the year of the English translation is 1995, not 1955.
References
Sajad Salami, Institute of Mathematics and Statistics, State University of Rio de Janeiro, Rio de Janeiro, Brazil
Email address: sajad.salami@ime.uerj.br
URL: https://sites.google.com/a/ime.uerj.br/sajadsalami/home