Introduction
Let K K K be a real quadratic number field of discriminant Δ \Delta Δ , and let K f + K_{\mathfrak f}^{+} K f + be a strict ray class field of K K K . The goal of this paper is to construct explicit generators for K f + K_{\mathfrak f}^{+} K f + and thus for the maximal abelian extension K a b = ⋃ f K f + K^{\mathrm{ab}}=\bigcup_{\mathfrak f}K_{\mathfrak f}^{+} K ab = ⋃ f K f + of K K K .
The underlying mechanism of exploiting Kummer theory is well-established in computational class field theory [8 , 15 , 16 ] . Our main innovation is to produce explicit Kummer bases from Frobenius eigenvalues of Fermat Jacobians with real multiplication by K K K .
For this we fix a character χ : G f + → Z / n Z \chi:G_{\mathfrak f}^{+}\to\mathbb{Z}/n\mathbb{Z} χ : G f + → Z / n Z of the strict ray class group, which defines a subfield H χ ⊂ K f + H_{\chi}\subset K_{\mathfrak f}^{+} H χ ⊂ K f + . It will be enough to consider characters χ \chi χ landing in Z / p a Z \mathbb{Z}/p^{a}\mathbb{Z} Z / p a Z , for p p p a prime number, and our construction works most elegantly if p p p is ordinary, i.e., p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ .
In the ordinary case, for anti-invariant χ \chi χ , we will construct a finite module of Kummer classes W Σ W_{\Sigma} W Σ and will show that the Kummer element [ β χ ] [\beta_{\chi}] [ β χ ] , which will produce a generator θ χ ∈ H χ \theta_{\chi}\in H_{\chi} θ χ ∈ H χ , is contained in W Σ W_{\Sigma} W Σ (Theorem 4.1 ). In the general case, we will construct a finite module of Kummer classes W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) and will show that the Kummer element [ β χ ] [\beta_{\chi}] [ β χ ] is contained in W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) (with χ \chi χ replaced by the signature-corrected character ψ \psi ψ at p = 2 p=2 p = 2 ; Theorem 4.2 ).
The main contribution of this paper is to produce explicit Kummer bases for the Kummer modules W Σ W_{\Sigma} W Σ and W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) from Frobenius eigenvalues of Fermat Jacobians with real multiplication by K K K , which will give rise to an explicit description of the Kummer classes [ β χ ] [\beta_{\chi}] [ β χ ] .
In the ordinary construction, Jacobi sums explicitly evaluate the Frobenius determinants. In the general construction, this remains true at good primes; at bad primes, the Shimura–Taniyama formula from CM theory is used to construct explicit replacement elements yielding the same Kummer module (Theorems 6.1 and 11.1 ).
An alternative construction using Fermat curves without requiring real multiplication by K K K is also given, which allows to use Jacobi sums at all primes, and produces the same Kummer module W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) and the same Kummer class [ β χ ] [\beta_{\chi}] [ β χ ] (after restoring the signature factor at p = 2 p=2 p = 2 ; Theorem 11.8 ).
Schematically our construction is given by
( K , f , χ ) ⏟ arithmetic input → Fermat geometry ( A i ) i ⏟ Frobenius eigenvalues or CM replacements → saturation ( [ z j ] ) j ⏟ Kummer basis → reciprocity [ β χ ] → descent θ χ Kummer class of χ generator of H χ / K \begin{aligned}
&\underbrace{(K,\mathfrak f,\chi)}_{\text{arithmetic input}}
\quad\xrightarrow{\text{Fermat geometry}} \quad
\underbrace{(A_i)_i}_{\text{Frobenius eigenvalues\\or CM replacements}}
\xrightarrow{\text{saturation}}
\underbrace{([z_j])_j}_{\text{Kummer basis}}
\xrightarrow{\text{reciprocity}}
[\beta_{\chi}]
\xrightarrow{\text{descent}}
\theta_{\chi}
\\
&\hspace{5.4cm}\text{Kummer class of }\chi
\hspace{2.0cm}\text{generator of }H_{\chi}/K
\end{aligned} arithmetic input ( K , f , χ ) Fermat geometry Frobenius eigenvalues or CM replacements ( A i ) i saturation Kummer basis ([ z j ] ) j reciprocity [ β χ ] descent θ χ Kummer class of χ generator of H χ / K
and leads to our main result, Theorem 15.2 , which is a constructive solution to Hilbert’s twelfth problem for arbitrary real quadratic fields.
Theorem 1.1. Our construction provides explicit elements θ 1 , … , θ s \theta_1,\ldots,\theta_s θ 1 , … , θ s such that
K f + = K ( θ 1 , … , θ s ) , K_{\mathfrak f}^{+}=K(\theta_1,\ldots,\theta_s), K f + = K ( θ 1 , … , θ s ) , together with their defining equations and Galois action.
Let’s explain our approach in some more detail now.
The Kummer modules W Σ W_{\Sigma} W Σ and W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 )
Fix n = p a n=p^{a} n = p a and suppose p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ . Let E = K ( ζ n ) E=K(\zeta_n) E = K ( ζ n ) , let j j j be complex conjugation on E E E , and let τ \tau τ be the automorphism of E E E that extends the nontrivial automorphism τ K \tau_K τ K of K K K while fixing ζ n \zeta_n ζ n . With the support Σ \Sigma Σ chosen as in Theorem 4.1 , the ordinary Kummer module is defined by
W Σ ≃ U Σ / U Σ n , U Σ = { x ∈ O E [ Σ − 1 ] × : x τ ( x ) = x j ( x ) = 1 } . (1) W_{\Sigma}\simeq U_{\Sigma}/U_{\Sigma}^{n},\qquad U_{\Sigma}=\{x\in\mathcal{O}_{E}[\Sigma^{-1}]^{\times}:x\tau(x)=xj(x)=1\}. \tag*{(1)} W Σ ≃ U Σ / U Σ n , U Σ = { x ∈ O E [ Σ − 1 ] × : xτ ( x ) = x j ( x ) = 1 } . (1)
For an (anti-invariant) character χ \chi χ , Theorem 4.1 provides a unique Kummer class [ β χ ] ∈ W Σ [\beta_{\chi}]\in W_{\Sigma} [ β χ ] ∈ W Σ , such that
ν β χ = χ ∣ G E , E ( β χ n ) = E H χ . (2) \nu_{\beta_{\chi}}=\chi|_{G_E},\qquad E\left(\sqrt[n]{\beta_{\chi}}\right)=EH_{\chi}. \tag*{(2)} ν β χ = χ ∣ G E , E ( n β χ ) = E H χ . (2)
Here ν \nu ν is the Kummer correspondence of Proposition 2.1 .
In the general case, allowing p ∣ 2 Δ p\mid2\Delta p ∣ 2Δ , we use the field and support of Theorem 4.2 and the general Kummer module is defined by
W Σ ( 1 ) ≃ U Σ ( 1 ) / ( U Σ ( 1 ) ) n , U Σ ( 1 ) = { x ∈ O E [ Σ − 1 ] × : x j ( x ) = 1 } . (3) W_{\Sigma}^{(1)}\simeq U_{\Sigma}^{(1)}/\left(U_{\Sigma}^{(1)}\right)^{n},\qquad U_{\Sigma}^{(1)}=\{x\in\mathcal{O}_{E}[\Sigma^{-1}]^{\times}:xj(x)=1\}. \tag*{(3)} W Σ ( 1 ) ≃ U Σ ( 1 ) / ( U Σ ( 1 ) ) n , U Σ ( 1 ) = { x ∈ O E [ Σ − 1 ] × : x j ( x ) = 1 } . (3)
For odd p p p it contains [ β χ ] [\beta_{\chi}] [ β χ ] for every χ \chi χ . At p = 2 p=2 p = 2 , we first obtain [ β ψ ] ∈ W Σ ( 1 ) [\beta_{\psi}]\in W_{\Sigma}^{(1)} [ β ψ ] ∈ W Σ ( 1 ) for the signature-corrected character ψ \psi ψ , then recover [ β χ ] [\beta_{\chi}] [ β χ ] ; see § 4.2 and § 13 .
Once these modules are known, Artin reciprocity selection and Galois descent are the standard mechanisms to extract the corresponding generator θ χ ∈ H χ \theta_{\chi}\in H_{\chi} θ χ ∈ H χ , see § 8 , § 9 , § 13 and § 14 , and [16 Sections 4.2–4.3] .
The Fermat geometry
We can now explain our main contribution, an explicit geometric construction of a Kummer basis for the Kummer modules W Σ W_{\Sigma} W Σ and W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) . The construction is based on Fermat curves with real multiplication by K K K and their Frobenius eigenvalues.
Again, we first explain the ordinary case, and then the general case.
For p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ , consider the smooth projective curve C n , Δ C_{n,\Delta} C n , Δ defined by
C n , Δ : Y n + Z Δ = 1. (4) C_{n,\Delta}:Y^{n}+Z^{\Delta}=1. \tag*{(4)} C n , Δ : Y n + Z Δ = 1. (4)
It is a quotient of an ordinary Fermat curve, its commuting deck actions give cyclotomic CM structures. The inclusion K ⊂ Q ( ζ Δ ) K\subset\mathbb{Q}(\zeta_{\Delta}) K ⊂ Q ( ζ Δ ) further gives rise to a RM structure. The two embeddings of K K K specify equal-dimensional eigenspaces M ν + , M ν − M_{\nu}^{+},M_{\nu}^{-} M ν + , M ν − of the rational Tate module of Jac ( C n , Δ ) \operatorname{Jac}(C_{n,\Delta}) Jac ( C n , Δ ) (see § 5.2 ).
For F = Q ( ζ n Δ ) F=\mathbb{Q}(\zeta_{n\Delta}) F = Q ( ζ n Δ ) , q \mathfrak q q a prime of F F F above Σ \Sigma Σ , and v ∈ Z n × v\in\mathbb{Z}_{n}^{\times} v ∈ Z n × , put
A q ( v ) = det ( Frob q ∣ M ν + ) det ( Frob q ∣ M ν − ) = ∏ c ∈ Z Δ × J q , c ( v ) χ Δ ( c ) ∈ U Σ . (5) A_{\mathfrak q}(v)=\frac{\det(\operatorname{Frob}_{\mathfrak q}\mid M_{\nu}^{+})}{\det(\operatorname{Frob}_{\mathfrak q}\mid M_{\nu}^{-})}=\prod_{c\in\mathbb{Z}_{\Delta}^{\times}}J_{\mathfrak q,c}(v)^{\chi_{\Delta}(c)}\in U_{\Sigma}. \tag*{(5)} A q ( v ) = det ( Frob q ∣ M ν − ) det ( Frob q ∣ M ν + ) = c ∈ Z Δ × ∏ J q , c ( v ) χ Δ ( c ) ∈ U Σ . (5)
Here χ Δ \chi_{\Delta} χ Δ is the quadratic Dirichlet character of K K K , and J q , c ( v ) J_{\mathfrak q,c}(v) J q , c ( v ) is the Jacobi sum of (5.11). CM describes the individual eigenvalues; RM specifies the quotient with the two required norm conditions (Theorem 5.1 ).
Theorem 1.2. Write Σ F \Sigma_F Σ F for the primes of F F F above Σ \Sigma Σ , and put
Γ o r d = ⟨ μ ( E ) ∩ U Σ , A q ( v ) : q ∈ Σ F , v ∈ Z n × ⟩ . \Gamma_{\mathrm{ord}}=\langle\mu(E)\cap U_\Sigma,A_{\mathfrak q}(v):\mathfrak q\in\Sigma_F,\ v\in\mathbb{Z}_n^\times\rangle. Γ ord = ⟨ μ ( E ) ∩ U Σ , A q ( v ) : q ∈ Σ F , v ∈ Z n × ⟩ . Then [ U Σ : Γ o r d ] < ∞ [U_\Sigma:\Gamma_{\mathrm{ord}}]<\infty [ U Σ : Γ ord ] < ∞ and
W Σ ≃ Γ o r d s a t / ( Γ o r d s a t ) n , W_\Sigma\simeq\Gamma_{\mathrm{ord}}^{\mathrm{sat}}/(\Gamma_{\mathrm{ord}}^{\mathrm{sat}})^n, W Σ ≃ Γ ord sat / ( Γ ord sat ) n , where
Γ o r d s a t = { x ∈ U Σ : x p b ∈ Γ o r d for some b ≥ 0 } . \Gamma_{\mathrm{ord}}^{\mathrm{sat}}=\{x\in U_\Sigma:x^{p^b}\in\Gamma_{\mathrm{ord}}\text{ for some }b\geq0\}. Γ ord sat = { x ∈ U Σ : x p b ∈ Γ ord for some b ≥ 0 } .
The first assertion is Theorem 6.1 , whose proof reduces to nonvanishing of Dirichlet L L L -values; the second is Corollary 7.2 . Saturation extracts the missing pth roots inside E E E and gives basis representatives explicitly related to the geometric inputs:
z j p b j = ξ j ∏ i A i m j i , ξ j ∈ μ ( E ) , m j i ∈ Z . (6) z_j^{p^{b_j}}=\xi_j\prod_i A_i^{m_{ji}},\qquad\xi_j\in\mu(E),\quad m_{ji}\in\mathbb{Z}. \tag*{(6)} z j p b j = ξ j i ∏ A i m j i , ξ j ∈ μ ( E ) , m j i ∈ Z . (6)
The saturation procedure is unnecessary when the index is prime to p p p (Corollary 7.3 ). The underlying Fermat decomposition and Frobenius–Jacobi formulas are classical [39 , 40 , 21 , 10 , 27 ] .
For arbitrary p p p , we enlarge the first exponent and consider the curve C M , Δ C_{M,\Delta} C M , Δ which is defined by
C M , Δ : Y M + Z Δ = 1 , Δ ∣ M , n ∣ M , E ⊆ F = Q ( ζ M ) . C_{M,\Delta}:Y^M+Z^\Delta=1,\qquad\Delta\mid M,\quad n\mid M,\quad E\subseteq F=\mathbb{Q}(\zeta_M). C M , Δ : Y M + Z Δ = 1 , Δ ∣ M , n ∣ M , E ⊆ F = Q ( ζ M ) .
The map ( Y , Z ) ↦ ( Y M / n , Z ) (Y,Z)\mapsto(Y^{M/n},Z) ( Y , Z ) ↦ ( Y M / n , Z ) connects this curve to C n , Δ C_{n,\Delta} C n , Δ , preserving the same real multiplication.
For v ∈ Z M ∖ { 0 } v\in\mathbb{Z}_M\setminus\{0\} v ∈ Z M ∖ { 0 } , the eigenspaces M v + , M v − M_v^+,M_v^- M v + , M v − can now have different dimensions; instead we compare M v + M_v^+ M v + with its conjugate M − v + M_{-v}^+ M − v + . For each rational prime ℓ \ell ℓ below Σ \Sigma Σ , choose a prime q ℓ \mathfrak q_\ell q ℓ of F F F above ℓ \ell ℓ and a good-reduction extension with a prime P ℓ \mathfrak P_\ell P ℓ above q ℓ \mathfrak q_\ell q ℓ . Then Theorem 10.2 gives
A q ℓ ( v ) = N F / E ( det ( Frob P ℓ ∣ M v + ) det ( Frob P ℓ ∣ M − v + ) ) ∈ U Σ ( 1 ) . (7) A_{\mathfrak q_\ell}(v)=N_{F/E}\left(\frac{\det(\operatorname{Frob}_{\mathfrak P_\ell}\mid M_v^+)}{\det(\operatorname{Frob}_{\mathfrak P_\ell}\mid M_{-v}^+)}\right)\in U_\Sigma^{(1)}. \tag*{(7)} A q ℓ ( v ) = N F / E ( det ( Frob P ℓ ∣ M − v + ) det ( Frob P ℓ ∣ M v + ) ) ∈ U Σ ( 1 ) . (7)
The determinants are identified with their algebraic values in F F F . The norm puts their quotient in E E E ; it is unnecessary in the ordinary construction because its determinants already lie in E E E (Proposition 10.7 ).
Here two kinds of primes must be distinguished. The character-order prime p p p is exceptional when p ∣ 2 Δ p\mid2\Delta p ∣ 2Δ . Good or bad reduction concerns the prime ℓ \ell ℓ at which Frobenius is evaluated. In the ordinary construction, Σ \Sigma Σ avoids p Δ p\Delta p Δ , so every evaluation prime lies in the good-reduction range ℓ ∤ n Δ \ell\nmid n\Delta ℓ ∤ n Δ . In the general construction, the curve has good reduction at ℓ ∤ M \ell\nmid M ℓ ∤ M , but the support may also contain primes dividing M M M , even when p p p is ordinary. Conversely, exceptional p p p does not mean that every Frobenius evaluation is at bad reduction.
At ℓ ∤ M \ell\nmid M ℓ ∤ M , the determinants are again products of Jacobi sums. At primes dividing M M M , the Shimura–Taniyama formula gives their ideals, from which principal-ideal arithmetic constructs explicit replacements A q ℓ ′ ( v ) A'_{\mathfrak q_\ell}(v) A q ℓ ′ ( v ) (Proposition 10.5 ). These need not equal the Frobenius values, but give the same Kummer module after saturation.
Theorem 1.3. For the field and support of Theorem 4.2 , and the choices above, put
Γ ( 1 ) = ⟨ μ ( E ) , A q ℓ ( v ) : 1 ≤ v < M , ℓ lies below Σ ⟩ . \Gamma^{(1)}=\langle\mu(E),A_{\mathfrak q_\ell}(v):1\leq v<M,\ \ell\text{ lies below }\Sigma\rangle. Γ ( 1 ) = ⟨ μ ( E ) , A q ℓ ( v ) : 1 ≤ v < M , ℓ lies below Σ ⟩ . Then [ U Σ ( 1 ) : Γ ( 1 ) ] < ∞ [U_\Sigma^{(1)}:\Gamma^{(1)}]<\infty [ U Σ ( 1 ) : Γ ( 1 ) ] < ∞ and
W Σ ( 1 ) ≃ Γ ( 1 ) , s a t / ( Γ ( 1 ) , s a t ) n , W_\Sigma^{(1)}\simeq\Gamma^{(1),\mathrm{sat}}/(\Gamma^{(1),\mathrm{sat}})^n, W Σ ( 1 ) ≃ Γ ( 1 ) , sat / ( Γ ( 1 ) , sat ) n , where saturation is taken inside U Σ ( 1 ) U_\Sigma^{(1)} U Σ ( 1 ) at p p p . The same assertions hold for the subgroup obtained by replacing the bad-prime inputs with A q ℓ ′ ( v ) A'_{\mathfrak q_\ell}(v) A q ℓ ′ ( v ) .
These are Theorem 11.1 and Proposition 12.2 . The proof combines classical CM theory [31 , 36 ] with nonvanishing of Dirichlet L L L -values at one.
Alternatively, the smooth projective curves C c C_c C c , for c > 2 c > 2 c > 2 and c ∣ M c \mid M c ∣ M , defined by
C c : z c = x ( 1 − x ) , (8) C_c : z^c = x(1-x), \tag*{(8)} C c : z c = x ( 1 − x ) , (8)
which are also quotients of ordinary Fermat curves, give rise to the same Kummer module W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) using Jacobi sums at all support primes ℓ \ell ℓ , choosing c c c with ℓ ∤ c \ell\nmid c ℓ ∤ c for each contributing character component. This construction does not require RM by K K K (Theorem 11.8 ).
To finish, Artin reciprocity selects the target class [ β χ ] [\beta_{\chi}] [ β χ ] from the Kummer basis, and equivariant Galois descent gives the generator θ χ ∈ H χ \theta_{\chi} \in H_{\chi} θ χ ∈ H χ , which now has a clear geometric interpretation in terms of Fermat geometry. This is worked out in Theorems 8.3 , 9.1 , 13.1 and 14.1 .
We also have readers in mind who have a background in mathematical physics, as the recent breakthrough work [29 ] uses mainly ideas from mathematical physics, therefore we try to give many details which are not strictly necessary for the number-theoretic reader. We hope that this will make our paper accessible to a wider audience.
Motivation and background. Our solution to Hilbert’s 12th problem for real quadratic fields is complementary to the recent breakthrough works of Dasgupta-Kakde and Radchenko-Wheeler, which solve conjectures of Brumer, Stark and Shintani [13 , 29 ] . Our construction is explicitly geometric, and is based on the classical geometry of Fermat curves and the theory of Complex Multiplication [33 , 32 ] pepped up with the RM structure coming from K K K . Our generators are not defined as special values of Artin L L L -functions or other special functions.
Nevertheless, our construction is heavily inspired by earlier work of the author which relates Shintani-Stark invariants and cyclic quantum dilogarithms [42 , 41 ] . The latter special functions live on Fermat curves and playing around with these structures, including extensive calculations with Sage and Pari, led to the present paper. Related special-value constructions include [13 , 29 , 44 , 22 , 4 ] , Fermat towers and p p p -adic beta-gamma formulas appear in [1 , 3 ] . It is an extremely interesting problem to understand the relation between these different approaches to Hilbert’s 12th problem for real quadratic fields.
The main mystery of whether there is a geometric replacement for elliptic curves with CM for a real quadratic number field remains completely open!
Outline of the paper. Section 2 recalls the character and Kummer correspondences (Proposition 2.1 ). Section 3 defines the Kummer modules (Propositions 3.2 and 3.6 ); Section 4 proves target membership (Theorems 4.1 and 4.2 ).
In the ordinary case, Section 5 constructs the Frobenius determinants (Theorem 5.1 ); Section 6 proves their divisor span (Theorem 6.1 ); Section 7 gives the Kummer basis (Corollary 7.2 ); Sections 8 and 9 perform reciprocity selection and descent (Theorems 8.3 and 9.1 ).
In the general case, Section 10 defines the inputs (Theorem 10.2 and Proposition 10.5); Section 11 proves their span and the diagonal alternative (Theorems 11.1 and 11.8 ); Section 12 gives the basis (Proposition 12.2 and Corollary 12.3); Sections 13 and 14 perform selection and descent (Theorems 13.1 and 14.1 ).
Section 15 assembles the fields (Theorem 15.2), gives their Artin action (Proposition 15.3), and treats mixed moduli and the strict Hilbert class field (Propositions 15.4 and 15.7). Section 16 gives explicit examples. Appendix A records the ray-character calculation (Proposition A.1); Appendix B relates the CM ideal to the reflex norm. Appendix C explains a useful classical feature of real quadratic class fields: at fixed support, only cyclotomic growth continues indefinitely. More precisely, Theorem C.1 gives
⋃ a ≥ 1 K f a + = K f a 0 + K ( Q ∞ ( p ) : p ∈ T ) for some a 0 , \bigcup_{a\ge1} K_{\mathfrak f_a}^{+}=K_{\mathfrak f_{a_0}}^{+}K\left(\mathbb{Q}_{\infty}^{(p)}:p\in T\right)\quad\text{for some }a_0, a ≥ 1 ⋃ K f a + = K f a 0 + K ( Q ∞ ( p ) : p ∈ T ) for some a 0 ,
where T T T consists of the rational primes whose full support in K K K divides f \mathfrak f f , and Q ∞ ( p ) \mathbb{Q}_{\infty}^{(p)} Q ∞ ( p ) is the cyclotomic Z p \mathbb{Z}_p Z p -extension of Q \mathbb{Q} Q .
Notation and conventions. Fix an algebraic closure Q ‾ ⊂ C \overline{\mathbb{Q}}\subset\mathbb{C} Q ⊂ C of Q \mathbb{Q} Q ; all number fields lie in Q ‾ \overline{\mathbb{Q}} Q . For a number field F F F , write G F = Gal ( Q ‾ / F ) G_F=\operatorname{Gal}(\overline{\mathbb{Q}}/F) G F = Gal ( Q / F ) and let μ ( F ) \mu(F) μ ( F ) be its roots of unity. Field products denote composita; products of finite algebras denote direct products.
Frobenius is arithmetic: if L / F L/F L / F is a finite Galois extension, q \mathfrak q q is unramified in L L L , and P ∣ q \mathfrak P\mid\mathfrak q P ∣ q , then
Frob P / q ∈ Gal ( L / F ) , Frob P / q ( x ) ≡ x N ( q ) ( m o d P ) ( x ∈ O L ) . \operatorname{Frob}_{\mathfrak P/\mathfrak q}\in\operatorname{Gal}(L/F),\qquad\operatorname{Frob}_{\mathfrak P/\mathfrak q}(x)\equiv x^{N(\mathfrak q)}\pmod{\mathfrak P}\quad(x\in\mathcal O_L). Frob P / q ∈ Gal ( L / F ) , Frob P / q ( x ) ≡ x N ( q ) ( mod P ) ( x ∈ O L ) .
Its conjugacy class is denoted Frob q \operatorname{Frob}_{\mathfrak q} Frob q ; in an abelian extension it is independent of P \mathfrak P P and equals Art L / F ( q ) \operatorname{Art}_{L/F}(\mathfrak q) Art L / F ( q ) [16 Section 2] . On a Tate module at good reduction, this is the action induced by the N ( q ) N(\mathfrak q) N ( q ) -power Frobenius endomorphism of the reduction, as in (45) .
Write K = Q ( d ) K=\mathbb{Q}(\sqrt{d}) K = Q ( d ) with d > 0 d>0 d > 0 squarefree, and let τ K \tau_K τ K be its nontrivial automorphism. Fix σ 1 ( d ) > 0 \sigma_1(\sqrt{d})>0 σ 1 ( d ) > 0 and σ 2 = σ 1 ∘ τ K \sigma_2=\sigma_1\circ\tau_K σ 2 = σ 1 ∘ τ K ; x ≫ 0 x\gg0 x ≫ 0 means positivity at both embeddings. Choose O K × , + = ⟨ ϵ ⟩ \mathcal O_K^{\times,+}=\langle\epsilon\rangle O K × , + = ⟨ ϵ ⟩ with σ 1 ( ϵ ) > 1 \sigma_1(\epsilon)>1 σ 1 ( ϵ ) > 1 . All ray class fields are strict: both real places are included in the modulus.
We use Z n = Z / n Z \mathbb{Z}_n=\mathbb{Z}/n\mathbb{Z} Z n = Z / n Z , reserving Z ^ p \widehat{\mathbb{Z}}_p Z p for the p p p -adic integers. Characters take values in Z n \mathbb{Z}_n Z n ; their multiplicative form is ρ χ ( g ) = ζ n χ ( g ) \rho_\chi(g)=\zeta_n^{\chi(g)} ρ χ ( g ) = ζ n χ ( g ) . Kummer classes are written multiplicatively, with scalar action a ⋅ [ x ] = [ x ] a a\cdot[x]=[x]^a a ⋅ [ x ] = [ x ] a . The notation W Σ ( E , n ) W_\Sigma(E,n) W Σ ( E , n ) or W Σ ( 1 ) ( E , n ) W_\Sigma^{(1)}(E,n) W Σ ( 1 ) ( E , n ) displays the field and exponent when needed; otherwise they are fixed. The subscript Σ \Sigma Σ records the prime support. These are Z n \mathbb{Z}_n Z n -modules, and vector spaces when n n n is prime. For an abelian group A A A written additively,
A [ n ] = { a ∈ A : n a = 0 } , A [ p ∞ ] = ⋃ r ≥ 1 A [ p r ] . A[n]=\{a\in A:na=0\},\qquad A[p^\infty]=\bigcup_{r\ge1}A[p^r]. A [ n ] = { a ∈ A : na = 0 } , A [ p ∞ ] = r ≥ 1 ⋃ A [ p r ] .
Prime ideals use fraktur letters. Valuations satisfy v p ( π p ) = 1 v_{\mathfrak p}(\pi_{\mathfrak p})=1 v p ( π p ) = 1 and v p ( 0 ) = + ∞ v_{\mathfrak p}(0)=+\infty v p ( 0 ) = + ∞ . For a nonzero integral ideal a ⊂ O F \mathfrak a\subset\mathcal O_F a ⊂ O F , N ( a ) = ∣ O F / a ∣ N(\mathfrak a)=|\mathcal O_F/\mathfrak a| N ( a ) = ∣ O F / a ∣ is its absolute norm, extended multiplicatively to fractional ideals. Subscripts indicate relative norms. For B ∈ Z > 0 B\in\mathbb{Z}_{>0} B ∈ Z > 0 , v p ( B ) v_p(B) v p ( B ) is its p p p -adic valuation.
A lattice is a finitely generated free abelian group. Smith normal form computes, for an integer relation matrix D ∈ Mat r × s ( Z ) D\in\operatorname{Mat}_{r\times s}(\mathbb{Z}) D ∈ Mat r × s ( Z ) , matrices P ∈ GL r ( Z ) P\in\operatorname{GL}_r(\mathbb{Z}) P ∈ GL r ( Z ) and Q ∈ GL s ( Z ) Q\in\operatorname{GL}_s(\mathbb{Z}) Q ∈ GL s ( Z ) such that
P D Q = diag ( d 1 , … , d t , 0 ) , d i > 0 , d i ∣ d i + 1 . PDQ=\operatorname{diag}(d_1,\ldots,d_t,0),\qquad d_i>0,\qquad d_i\mid d_{i+1}. P D Q = diag ( d 1 , … , d t , 0 ) , d i > 0 , d i ∣ d i + 1 .
The diagonal matrix may be rectangular. It gives explicit coordinates
Z r / D Z s ≅ ⨁ i = 1 t Z d i ⊕ Z r − t , [ x ] ⟼ [ P x ] . \mathbb{Z}^r/D\mathbb{Z}^s\cong\bigoplus_{i=1}^{t}\mathbb{Z}_{d_i}\oplus\mathbb{Z}^{r-t},\qquad[x]\longmapsto[Px]. Z r / D Z s ≅ i = 1 ⨁ t Z d i ⊕ Z r − t , [ x ] ⟼ [ P x ] .
The quotient is finite exactly when t = r t=r t = r , with order ∏ i d i \prod_i d_i ∏ i d i . We use this standard algorithm for ray-group coordinates and lattice indices [7 Section 2.4.4] .
Use of AI tools. When we started this project at the beginning of 2026, we used SageMath and Pari extensively for explicit calculations and exploited AI tools to help with those calculations. Later, we used AI-tools to proof-read the manuscript, help with references and check the examples in section 16. All mathematical ideas are due to the author, who takes responsibility for the content of this paper and the correctness of the results.
Ray class characters and their Kummer representatives
We fix conventions for standard class field and Kummer theory. For ray groups and reciprocity see [24 Chapter VI] and [25 Part III] ; for Kummer theory and its constructive use see [23 Theorem 5.30] , [8 Chapters 4–5] , and [16 Sections 3–4] . The statements below are classical or elementary consequences.
The ray class group and its characters
For a nonzero integral ideal f \mathfrak{f} f , let I K ( f ) I_K(\mathfrak{f}) I K ( f ) be the multiplicative group of nonzero fractional ideals prime to f \mathfrak{f} f . The strict ray class group is
G f + = I K ( f ) / { ( x ) : x ∈ K × , x ≫ 0 , x ≡ 1 ( m o d f ) } . (9) G_{\mathfrak{f}}^{+}=I_K(\mathfrak{f})/\{(x):x\in K^{\times},\ x\gg0,\ x\equiv1\pmod{\mathfrak{f}}\}. \tag*{(9)} G f + = I K ( f ) / {( x ) : x ∈ K × , x ≫ 0 , x ≡ 1 ( mod f )} . (9)
The congruence x ≡ 1 ( m o d f ) x\equiv1\pmod{\mathfrak{f}} x ≡ 1 ( mod f ) means v p ( x − 1 ) ≥ v p ( f ) v_{\mathfrak{p}}(x-1)\ge v_{\mathfrak{p}}(\mathfrak{f}) v p ( x − 1 ) ≥ v p ( f ) for p ∣ f \mathfrak{p}\mid\mathfrak{f} p ∣ f . Recall the well-known exact sequence [24 Chapter VI, Section 1] :
1 ⟶ ( O K / f ) × ⟨ ϵ ˉ ⟩ ⟶ G f + ⟶ Cl + ( K ) ⟶ 1. (10) 1\longrightarrow\frac{(\mathcal{O}_K/\mathfrak{f})^{\times}}{\langle\bar{\epsilon}\rangle}\longrightarrow G_{\mathfrak{f}}^{+}\longrightarrow\operatorname{Cl}^{+}(K)\longrightarrow1. \tag*{(10)} 1 ⟶ ⟨ ϵ ˉ ⟩ ( O K / f ) × ⟶ G f + ⟶ Cl + ( K ) ⟶ 1. (10)
Here Cl + ( K ) = G ( 1 ) + \operatorname{Cl}^{+}(K)=G_{(1)}^{+} Cl + ( K ) = G ( 1 ) + is the narrow ideal class group, and ϵ ˉ \bar{\epsilon} ϵ ˉ is the reduction of the fixed generator of totally positive units modulo f \mathfrak{f} f . The first map sends b ˉ \bar{b} b ˉ to [ ( b ) ] [(b)] [( b )] for a totally positive lift, which exists by approximation; the second forgets the congruence condition. Thus
Cl + ( K ) = 1 ⟹ G f + ≃ ( O K / f ) × / ⟨ ϵ ˉ ⟩ . (11) \operatorname{Cl}^{+}(K)=1\quad\Longrightarrow\quad G_{\mathfrak{f}}^{+}\simeq(\mathcal{O}_K/\mathfrak{f})^{\times}/\langle\bar{\epsilon}\rangle. \tag*{(11)} Cl + ( K ) = 1 ⟹ G f + ≃ ( O K / f ) × / ⟨ ϵ ˉ ⟩ . (11)
In general this quotient is only the kernel in (10) ; the sequence need not split. No class-number assumption is made here.
Artin reciprocity and the existence theorem identify this group with the Galois group of the strict ray class field:
Art f : G f + ⟶ ∼ Gal ( K f + / K ) , [ p ] ⟼ Frob p , \operatorname{Art}_{\mathfrak{f}}:G_{\mathfrak{f}}^{+}\overset{\sim}{\longrightarrow}\operatorname{Gal}(K_{\mathfrak{f}}^{+}/K),\qquad[\mathfrak{p}]\longmapsto\operatorname{Frob}_{\mathfrak{p}}, Art f : G f + ⟶ ∼ Gal ( K f + / K ) , [ p ] ⟼ Frob p ,
where Frobenius acts on residue fields by x ↦ x N ( p ) x\mapsto x^{N(\mathfrak{p})} x ↦ x N ( p ) [25 Part III, Theorems 7.8, 9.2 and 9.4] . For a character χ : G f + → Z n \chi:G_{\mathfrak{f}}^{+}\to\mathbb{Z}_n χ : G f + → Z n , put
H χ = ( K f + ) ker χ , [ H χ : K ] = ∣ im χ ∣ . H_{\chi}=(K_{\mathfrak{f}}^{+})^{\ker\chi},\qquad[H_{\chi}:K]=|\operatorname{im}\chi|. H χ = ( K f + ) k e r χ , [ H χ : K ] = ∣ im χ ∣.
We use the same symbol for its inflation to G K G_K G K :
χ ( g ) = χ ( Art f − 1 ( g ∣ K f + ) ) , g ∈ G K . \chi(g)=\chi\left(\operatorname{Art}_{\mathfrak{f}}^{-1}(g|_{K_{\mathfrak{f}}^{+}})\right),\qquad g\in G_K. χ ( g ) = χ ( Art f − 1 ( g ∣ K f + ) ) , g ∈ G K .
Thus χ ∣ G E \chi|_{G_E} χ ∣ G E below always means restriction of this continuous Galois character.
Recovering the full ray class field
Finite abelian characters separate points. Consequently
K f + = ∏ χ H χ , Gal ( K f + / ∏ χ H χ ) = ⋂ χ ker χ = { 1 } , (12) K_{\mathfrak{f}}^{+}=\prod_{\chi}H_{\chi},\qquad\operatorname{Gal}\left(K_{\mathfrak{f}}^{+}/\prod_{\chi}H_{\chi}\right)=\bigcap_{\chi}\ker\chi=\{1\}, \tag*{(12)} K f + = χ ∏ H χ , Gal ( K f + / χ ∏ H χ ) = χ ⋂ ker χ = { 1 } , (12)
where χ \chi χ ranges over Hom ( G f + , Z e ) \operatorname{Hom}(G_{\mathfrak{f}}^{+},\mathbb{Z}_e) Hom ( G f + , Z e ) with e e e the least common multiple of the orders of its elements (the group exponent), and the product denotes the compositum inside Q ‾ \overline{\mathbb{Q}} Q . It suffices to take just the coordinate characters of a decomposition G f + ≃ ⨁ j Z d j G_{\mathfrak{f}}^{+}\simeq\bigoplus_j\mathbb{Z}_{d_j} G f + ≃ ⨁ j Z d j computed by Smith normal form. Their prime-power components also separate points [23 Chapter 3] [16 Section 4.1] ; Section 15 computes these characters and assembles their field generators. Characters valued in one fixed Z n \mathbf{Z}_n Z n suffice precisely when e ∣ n e \mid n e ∣ n .
Appendix A computes a presentation G f + ≃ Z r / A Z s G_{\mathfrak f}^{+} \simeq\mathbf{Z}^{r}/A\mathbf{Z}^{s} G f + ≃ Z r / A Z s from ideals, units and residues. The columns of the integer matrix A A A are the exponents in the relations among the chosen ideal classes. Its Smith normal form gives the cyclic factors and their coordinate characters [7 Section 2.4.4] ; no equation for H χ H_\chi H χ is an input.
Kummer theory after adjoining roots of unity
For n > 1 n>1 n > 1 , the field E = K ( ζ n ) E=K(\zeta_n) E = K ( ζ n ) always suffices for Kummer theory. We allow larger coefficient fields because the dyadic norm-one construction of § 4.2 may require them. Let E / K E/K E / K be a finite Galois extension with μ n ⊂ E \mu_n \subset E μ n ⊂ E , and choose a primitive n n n th root of unity ζ n \zeta_n ζ n . Set G E = Gal ( Q ‾ / E ) G_E=\operatorname{Gal}(\overline{\mathbb{Q}}/E) G E = Gal ( Q / E ) . For β ∈ E × \beta\in E^{\times} β ∈ E × and α ∈ Q ‾ \alpha\in\overline{\mathbb{Q}} α ∈ Q with α n = β \alpha^n=\beta α n = β , define
ν β : G E ⟶ Z n , g ( α ) = ζ n ν β ( g ) α . (13) \nu_\beta:G_E\longrightarrow\mathbf{Z}_n,\qquad g(\alpha)=\zeta_n^{\nu_\beta(g)}\alpha. \tag*{(13)} ν β : G E ⟶ Z n , g ( α ) = ζ n ν β ( g ) α . (13)
Since g ( α ) / α ∈ μ n g(\alpha)/\alpha\in\mu_n g ( α ) / α ∈ μ n , this determines ν β ( g ) ∈ Z n \nu_\beta(g)\in\mathbf{Z}_n ν β ( g ) ∈ Z n uniquely. The group G E G_E G E fixes μ n \mu_n μ n , so ν β \nu_\beta ν β is a homomorphism, with open kernel G E ( α ) G_E(\alpha) G E ( α ) . Replacing α \alpha α by ζ n k α \zeta_n^k\alpha ζ n k α , or β \beta β by β c n \beta c^n β c n with c ∈ E × c \in E^\times c ∈ E × , leaves the character unchanged. The following standard Kummer isomorphism therefore has the indicated domain [23 Theorem 5.30] .
Proposition 2.1. The map
ν : E × / E × n → ∼ Hom c o n t ( G E , Z n ) , [ β ] ⟼ ν β (14) \nu:E^\times/E^{\times n}\xrightarrow{\sim}\operatorname{Hom}_{\mathrm{cont}}(G_E,\mathbf{Z}_n),\qquad[\beta]\longmapsto\nu_\beta \tag*{(14)} ν : E × / E × n ∼ Hom cont ( G E , Z n ) , [ β ] ⟼ ν β (14) is an isomorphism. If [ β ] [\beta] [ β ] has order m m m , then E ( α ) = Q ‾ ker ν β E(\alpha)=\overline{\mathbb{Q}}^{\ker\nu_\beta} E ( α ) = Q k e r ν β and [ E ( α ) : E ] = m [E(\alpha):E]=m [ E ( α ) : E ] = m .
Definition 2.2. Fix a ray class character χ : G f + → Z n \chi:G_{\mathfrak f}^{+}\to\mathbf{Z}_n χ : G f + → Z n , inflated to G K G_K G K as above. An element β ∈ E × \beta\in E^\times β ∈ E × represents its restriction to E E E when
( R χ ) ν β = χ ∣ G E . (15) (\mathrm{R}_\chi)\qquad\nu_\beta=\chi|_{G_E}. \tag*{(15)} ( R χ ) ν β = χ ∣ G E . (15) Equivalently, [ β ] = ν − 1 ( χ ∣ G E ) [\beta]=\nu^{-1}(\chi|_{G_E}) [ β ] = ν − 1 ( χ ∣ G E ) . We calculate this class from Fermat values. Distinct characters can have the same restriction; descent uses the original χ \chi χ . The isomorphism ν \nu ν sends products of Kummer classes to sums of characters.
Lemma 2.3. Let β ∈ E × \beta\in E^\times β ∈ E × be any candidate representative, and let P \mathfrak P P be a prime of E E E such that P ∤ n f O E \mathfrak P\nmid n\mathfrak f\mathcal{O}_E P ∤ n f O E and v P ( β ) = 0 v_{\mathfrak P}(\beta)=0 v P ( β ) = 0 , and put q P = ∣ O E / P ∣ q_{\mathfrak P}=|\mathcal{O}_E/\mathfrak P| q P = ∣ O E / P ∣ . Then
ζ n ν β ( Frob P ) ≡ β ( q P − 1 ) / n ( m o d P ) , (16) \zeta_n^{\nu_\beta(\operatorname{Frob}_{\mathfrak P})}\equiv\beta^{(q_{\mathfrak P}-1)/n}\pmod{\mathfrak P}, \tag*{(16)} ζ n ν β ( Frob P ) ≡ β ( q P − 1 ) / n ( mod P ) , (16) ( χ ∣ G E ) ( Frob P ) = χ ( [ N E / K P ] f ) . (17) (\chi|_{G_E})(\operatorname{Frob}_{\mathfrak P})=\chi([\mathrm{N}_{E/K}\mathfrak P]_{\mathfrak f}). \tag*{(17)} ( χ ∣ G E ) ( Frob P ) = χ ([ N E / K P ] f ) . (17) Consequently ( R χ ) (\mathrm{R}_\chi) ( R χ ) is equivalent to
β ( q P − 1 ) / n ≡ ζ n χ ( [ N E / K P ] f ) ( m o d P ) (18) \beta^{(q_{\mathfrak P}-1)/n}\equiv\zeta_n^{\chi([\mathrm{N}_{E/K}\mathfrak P]_{\mathfrak f})}\pmod{\mathfrak P} \tag*{(18)} β ( q P − 1 ) / n ≡ ζ n χ ([ N E / K P ] f ) ( mod P ) (18) for all but finitely many primes P \mathfrak P P .
Proof. The first formula is the standard Kummer Artin-symbol calculation [15 Section 3] [16 Proposition 4] ; reduction is injective on μ n \mu_n μ n , so n ∣ q P − 1 n\mid q_{\mathfrak P}-1 n ∣ q P − 1 . The second is norm compatibility of reciprocity [25 Part III, Section 6, p. 153] . Chebotarev makes equality on almost every Frobenius equivalent to equality of characters [24 Chapter VII, Theorem 13.4] . □ \square □
Section 8 makes this test finite. Once ν − 1 ( χ ∣ G E ) \nu^{-1}(\chi|_{G_E}) ν − 1 ( χ ∣ G E ) belongs to a known free Kummer module, evaluation of the Kummer pairing at separating Frobenius elements distinguishes all its classes. Matching the prescribed Artin values there determines the target class; we call this selection by Artin reciprocity.
The next standard compatibility explains the equations for the coefficients used in descent. It is an immediate consequence of the Kummer correspondence [23 Theorem 5.30] .
Lemma 2.4. Suppose β ∈ E × \beta\in E^{\times} β ∈ E × satisfies ( R χ ) (\mathrm{R}_{\chi}) ( R χ ) . For δ ∈ Gal ( E / K ) \delta\in\operatorname{Gal}(E/K) δ ∈ Gal ( E / K ) , let 1 ≤ u δ < n 1 \le u_{\delta} < n 1 ≤ u δ < n be determined by δ ( ζ n ) = ζ n u δ \delta(\zeta_{n})=\zeta_{n}^{u_{\delta}} δ ( ζ n ) = ζ n u δ . Then
δ ( [ β ] ) = [ β ] u δ in E × / E × n . (19) \delta([\beta])=[\beta]^{u_{\delta}}\quad\text{in }E^{\times}/E^{\times n}. \tag*{(19)} δ ([ β ]) = [ β ] u δ in E × / E × n . (19) Equivalently, there is c δ ∈ E × c_{\delta} \in E^{\times} c δ ∈ E × such that
c δ n = δ ( β ) / β u δ . c_{\delta}^{n}=\delta(\beta)/\beta^{u_{\delta}}. c δ n = δ ( β ) / β u δ . Proof. For a lift δ ~ ∈ G K \widetilde{\delta}\in G_K δ ∈ G K of δ \delta δ and g ∈ G E g\in G_E g ∈ G E , Kummer equivariance gives ν δ ( β ) ( g ) = u δ χ ( δ ~ − 1 g δ ~ ) = u δ χ ( g ) \nu_{\delta(\beta)}(g)=u_{\delta}\chi(\widetilde{\delta}^{-1}g\widetilde{\delta})=u_{\delta}\chi(g) ν δ ( β ) ( g ) = u δ χ ( δ − 1 g δ ) = u δ χ ( g ) , since χ \chi χ is a homomorphism to an abelian group. Now apply the isomorphism (2.6). □
This is the power equation needed for a lift d δ ( α ) = c δ α u δ d_{\delta}(\alpha)=c_{\delta}\alpha^{u_{\delta}} d δ ( α ) = c δ α u δ with d δ ∣ E = δ d_{\delta}|_E=\delta d δ ∣ E = δ . The descent calculation in § 9 and § 14 chooses the roots so that these maps compose correctly and give the prescribed character χ \chi χ .
Quadratic conjugation and the cyclotomic part
Conjugation compatibility of the Artin map [16 Section 5.1, Proposition 15] gives
τ f : G f + → ∼ G τ K f + , [ a ] ⟼ [ τ K a ] , χ τ ( g ) = χ ( τ f ( g ) ) ( τ K f = f ) . \tau_f:G_f^+\xrightarrow{\sim}G_{\tau_Kf}^+,\qquad[a]\longmapsto[\tau_Ka],\qquad\chi^\tau(g)=\chi(\tau_f(g))\quad(\tau_Kf=f). τ f : G f + ∼ G τ K f + , [ a ] ⟼ [ τ K a ] , χ τ ( g ) = χ ( τ f ( g )) ( τ K f = f ) .
Lemma 2.5. For every n n n , if τ K f = f \tau_Kf=f τ K f = f and χ τ = χ \chi^\tau=\chi χ τ = χ , then H χ / Q H_\chi/\mathbb{Q} H χ / Q is abelian. In particular H χ H_\chi H χ is contained in a cyclotomic field.
Proof. The stable kernel makes H χ / Q H_\chi/\mathbb{Q} H χ / Q Galois, and invariance makes Gal ( H χ / K ) \operatorname{Gal}(H_\chi/K) Gal ( H χ / K ) central. A central subgroup with cyclic quotient generates an abelian group together with any lift of the quotient generator. Apply Kronecker–Weber [25 Part III, Theorem 7.11] . □
Lemma 2.6. Suppose n n n is odd. Put f ∗ = lcm ( f , τ K f ) f^*=\operatorname{lcm}(f,\tau_Kf) f ∗ = lcm ( f , τ K f ) , and pull χ \chi χ back along the natural surjection G f ∗ + → G f + G_{f^*}^+\to G_f^+ G f ∗ + → G f + . On this stable modulus set
χ + = 2 − 1 ( χ + χ τ ) , χ − = 2 − 1 ( χ − χ τ ) . \chi^+=2^{-1}(\chi+\chi^\tau),\qquad\chi^-=2^{-1}(\chi-\chi^\tau). χ + = 2 − 1 ( χ + χ τ ) , χ − = 2 − 1 ( χ − χ τ ) . Then χ = χ + + χ − \chi=\chi^++\chi^- χ = χ + + χ − and ( χ ± ) τ = ± χ ± (\chi^\pm)^\tau=\pm\chi^\pm ( χ ± ) τ = ± χ ± . The field H χ + H_{\chi^+} H χ + is cyclotomic. Constructing H χ − H_{\chi^-} H χ − and their compositum L = H χ + H χ − L=H_{\chi^+}H_{\chi^-} L = H χ + H χ − determines the original field by
H χ = L ker s , s : Gal ( L / K ) ⟶ Z n , g ⟼ χ + ( g ∣ H χ + ) + χ − ( g ∣ H χ − ) . H_\chi=L^{\ker s},\qquad s:\operatorname{Gal}(L/K)\longrightarrow\mathbb{Z}_n,\qquad g\longmapsto\chi^+(g|_{H_{\chi^+}})+\chi^-(g|_{H_{\chi^-}}). H χ = L k e r s , s : Gal ( L / K ) ⟶ Z n , g ⟼ χ + ( g ∣ H χ + ) + χ − ( g ∣ H χ − ) . Proof. The ray-group surjection follows from approximation. The character identities follow by division by two. Lemma 2.5 applies to χ + \chi^+ χ + , and the fixed-field formula is Galois correspondence applied to χ = χ + + χ − \chi=\chi^++\chi^- χ = χ + + χ − . □
A zero component contributes only K K K ; each nonzero component may be constructed at its actual order.
Remark 2.7. One can show that the odd-order invariant field satisfies
H χ + ⊆ Q ( ζ Δ N ∗ ) , N ∗ = ∣ O K / f ∗ ∣ ∈ Z > 0 . H_{\chi^+}\subseteq\mathbb{Q}(\zeta_{\Delta N^*}),\qquad N^*=|\mathcal{O}_K/f^*|\in\mathbb{Z}_{>0}. H χ + ⊆ Q ( ζ Δ N ∗ ) , N ∗ = ∣ O K / f ∗ ∣ ∈ Z > 0 .
Kummer modules
We define the central Kummer modules W Σ W_\Sigma W Σ and W Σ ( 1 ) W_\Sigma^{(1)} W Σ ( 1 ) . Everything in this section is standard, see, e.g., [8 Sections 5.2.2 and 7.4] .
Supported classes and the choice of primes
For the ideal calculations in this subsection, E E E is any number field, n ≥ 2 n \ge2 n ≥ 2 , and Σ \Sigma Σ is any finite set of its finite primes P E \mathcal{P}_E P E . No class-group assumption is made. Define
R Σ = O E [ Σ − 1 ] = { x ∈ E : v p ( x ) ≥ 0 for p ∉ Σ } , R_{\Sigma} = \mathcal{O}_E[\Sigma^{-1}] = \{x \in E : v_{\mathfrak{p}}(x) \ge0 \text{ for } \mathfrak{p} \notin\Sigma\}, R Σ = O E [ Σ − 1 ] = { x ∈ E : v p ( x ) ≥ 0 for p ∈ / Σ } ,
R Σ × = { x ∈ E × : v p ( x ) = 0 for every p ∉ Σ } , R_{\Sigma}^{\times} = \{x \in E^{\times} : v_{\mathfrak{p}}(x) = 0 \text{ for every } \mathfrak{p} \notin\Sigma\}, R Σ × = { x ∈ E × : v p ( x ) = 0 for every p ∈ / Σ } ,
Div ( E ) = ⨁ p ∈ P E Z [ p ] , Z [ Σ ] = ⨁ p ∈ Σ Z [ p ] ⊂ Div ( E ) , \operatorname{Div}(E) = \bigoplus_{\mathfrak{p} \in\mathcal{P}_E} \mathbb{Z}[\mathfrak{p}], \qquad\mathbb{Z}[\Sigma] = \bigoplus_{\mathfrak{p} \in\Sigma} \mathbb{Z}[\mathfrak{p}] \subset\operatorname{Div}(E), Div ( E ) = p ∈ P E ⨁ Z [ p ] , Z [ Σ ] = p ∈ Σ ⨁ Z [ p ] ⊂ Div ( E ) ,
div : E × ⟶ Div ( E ) , x ⟼ ∑ p v p ( x ) [ p ] . \operatorname{div}: E^{\times} \longrightarrow\operatorname{Div}(E), \qquad x \longmapsto\sum_{\mathfrak{p}} v_{\mathfrak{p}}(x)[\mathfrak{p}]. div : E × ⟶ Div ( E ) , x ⟼ p ∑ v p ( x ) [ p ] .
The elements of R Σ × R_{\Sigma}^{\times} R Σ × are the Σ \Sigma Σ -units, and Z [ Σ ] \mathbb{Z}[\Sigma] Z [ Σ ] is the group of divisors supported on Σ \Sigma Σ . In particular div ( R Σ × ) ⊂ Z [ Σ ] \operatorname{div}(R_{\Sigma}^{\times}) \subset\mathbb{Z}[\Sigma] div ( R Σ × ) ⊂ Z [ Σ ] . For an automorphism σ \sigma σ of E E E preserving Σ \Sigma Σ , the actions are
x ⟼ σ ( x ) , [ x ] ⟼ [ σ ( x ) ] in E × / E × n , x \longmapsto\sigma(x), \qquad[x] \longmapsto[\sigma(x)] \text{ in } E^{\times}/E^{\times n}, x ⟼ σ ( x ) , [ x ] ⟼ [ σ ( x )] in E × / E × n ,
∑ p m p [ p ] ⟼ ∑ p m p [ σ p ] , div ( σ x ) = σ div ( x ) . \sum_{\mathfrak{p}} m_{\mathfrak{p}}[\mathfrak{p}] \longmapsto\sum_{\mathfrak{p}} m_{\mathfrak{p}}[\sigma\mathfrak{p}], \qquad\operatorname{div}(\sigma x) = \sigma\operatorname{div}(x). p ∑ m p [ p ] ⟼ p ∑ m p [ σ p ] , div ( σ x ) = σ div ( x ) .
Definition 3.1. The n n n -Selmer group with support Σ \Sigma Σ is
Sel n ( E , Σ ) = { [ x ] ∈ E × / E × n : v p ( x ) ∈ n Z for p ∉ Σ } . (20) \operatorname{Sel}_n(E,\Sigma) = \{[x] \in E^{\times}/E^{\times n} : v_{\mathfrak{p}}(x) \in n\mathbb{Z} \text{ for } \mathfrak{p} \notin\Sigma\}. \tag*{(20)} Sel n ( E , Σ ) = {[ x ] ∈ E × / E × n : v p ( x ) ∈ n Z for p ∈ / Σ } . (20)
An R Σ R_{\Sigma} R Σ -unit has zero divisor outside Σ \Sigma Σ ; a Selmer representative need only have an n n n -divisible divisor there. Write Cl ( R Σ ) \operatorname{Cl}(R_{\Sigma}) Cl ( R Σ ) for the group of fractional R Σ R_{\Sigma} R Σ -ideals modulo principal ideals.
Proposition 3.2. There is an exact sequence
1 ⟶ R Σ × / ( R Σ × ) n → u ↦ [ u ] Sel n ( E , Σ ) → ∂ Cl ( R Σ ) [ n ] ⟶ 0. (21) 1 \longrightarrow R_{\Sigma}^{\times}/(R_{\Sigma}^{\times})^n \xrightarrow{u \mapsto[u]} \operatorname{Sel}_n(E,\Sigma) \xrightarrow{\partial} \operatorname{Cl}(R_{\Sigma})[n] \longrightarrow0. \tag*{(21)} 1 ⟶ R Σ × / ( R Σ × ) n u ↦ [ u ] Sel n ( E , Σ ) ∂ Cl ( R Σ ) [ n ] ⟶ 0. (21) where
∂ [ x ] = [ a x ] , a x = ∏ p ∉ Σ ( p R Σ ) v p ( x ) / n , x R Σ = a x n . \partial[x] = [\mathfrak{a}_x], \qquad\mathfrak{a}_x = \prod_{\mathfrak{p} \notin\Sigma}(\mathfrak{p}R_{\Sigma})^{v_{\mathfrak{p}}(x)/n}, \qquad xR_{\Sigma} = \mathfrak{a}_x^n. ∂ [ x ] = [ a x ] , a x = p ∈ / Σ ∏ ( p R Σ ) v p ( x ) / n , x R Σ = a x n . The obstruction to a unit representative of [ x ] [x] [ x ] is exactly ∂ [ x ] \partial[x] ∂ [ x ] . For n = p a n = p^a n = p a it belongs to Cl ( R Σ ) [ p a ] \operatorname{Cl}(R_{\Sigma})[p^a] Cl ( R Σ ) [ p a ] , the subgroup of ideal classes killed by p a p^a p a . Every supported class has a unit representative if and only if this subgroup is zero, equivalently Cl ( R Σ ) [ p ] = 0 \operatorname{Cl}(R_{\Sigma})[p] = 0 Cl ( R Σ ) [ p ] = 0 .
Proof. This is the usual Selmer sequence [8 Proposition 5.2.8] , with the same ideal proof over R Σ R_{\Sigma} R Σ for arbitrary n n n : ∂ [ x ] = 0 \partial[x] = 0 ∂ [ x ] = 0 precisely when a x = y R Σ \mathfrak{a}_x = yR_{\Sigma} a x = y R Σ , in which case x / y n x/y^n x / y n is a unit. Every n n n -torsion ideal class has a n = x R Σ \mathfrak{a}^n = xR_{\Sigma} a n = x R Σ ; an n n n th root of a unit has zero valuations outside Σ . □ \Sigma. \square Σ. □
Lemma 3.3. For any number field E E E and disjoint finite sets Σ 0 , B ⊂ P E \Sigma_0,\mathcal{B} \subset\mathcal{P}_E Σ 0 , B ⊂ P E , there is a finite Σ ⊇ Σ 0 \Sigma\supseteq\Sigma_0 Σ ⊇ Σ 0 disjoint from B \mathcal{B} B such that Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma}) = 0 Cl ( R Σ ) = 0 . If a group of automorphisms of E E E preserves Σ 0 \Sigma_0 Σ 0 and B \mathcal{B} B , then Σ \Sigma Σ can be chosen stable under it. Thus one support removes the ideal-class obstruction for every prime-power exponent, including powers of two.
Proof. The localization formula
Cl ( R Σ ) = Cl ( O E ) / ⟨ [ p ] : p ∈ Σ ⟩ \operatorname{Cl}(R_{\Sigma}) = \operatorname{Cl}(\mathcal{O}_E)/\langle[\mathfrak{p}] : \mathfrak{p} \in\Sigma\rangle Cl ( R Σ ) = Cl ( O E ) / ⟨[ p ] : p ∈ Σ ⟩
is [8 Proposition 7.4.4] ; adjoining the supports of finitely many class generators kills the quotient [8 Corollary 7.4.5] . Approximation chooses these representatives prime to B \mathcal{B} B [8 Corollary 1.2.11] . Their automorphism orbits still avoid the stable set B \mathcal{B} B and give the required support. □ \square □
Convention 3.4. In all field constructions, including p = 2 p=2 p = 2 , choose a Galois-stable support with Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma})=0 Cl ( R Σ ) = 0 , avoiding primes above p Δ p\Delta p Δ in the ordinary case. The lemma permits this auxiliary choice without restricting the target. Abstract statements use their stated weaker hypotheses. Class-group vanishing removes the ideal-class obstruction, but the dyadic norm obstruction remains (Section 4).
The ordinary norm conditions
For the ordinary construction let n = p a n=p^{a} n = p a , p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ , and E = K ( ζ n ) E=K(\zeta_n) E = K ( ζ n ) , with Σ \Sigma Σ stable under Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) . The quadratic field K K K is disjoint from Q ( ζ n ) \mathbb{Q}(\zeta_n) Q ( ζ n ) : a common nontrivial subfield would force the conductor Δ \Delta Δ of K K K to divide n n n . We therefore have
Gal ( E / Q ) = ⟨ τ ⟩ × Gal ( E / K ) , Gal ( E / K ) ≃ ( Z n ) × . \operatorname{Gal}(E/\mathbb{Q})=\langle\tau\rangle\times\operatorname{Gal}(E/K),\qquad\operatorname{Gal}(E/K)\simeq(\mathbb{Z}_n)^{\times}. Gal ( E / Q ) = ⟨ τ ⟩ × Gal ( E / K ) , Gal ( E / K ) ≃ ( Z n ) × .
The automorphisms are specified by
τ : E → ∼ E , τ ∣ K = τ K , τ ( ζ n ) = ζ n , \tau:E\xrightarrow{\sim}E,\qquad\tau|_K=\tau_K,\qquad\tau(\zeta_n)=\zeta_n, τ : E ∼ E , τ ∣ K = τ K , τ ( ζ n ) = ζ n ,
δ u : E → ∼ E , δ u ∣ K = id K , δ u ( ζ n ) = ζ n u ( u ∈ Z n × ) . \delta_u:E\xrightarrow{\sim}E,\qquad\delta_u|_K=\operatorname{id}_K,\qquad\delta_u(\zeta_n)=\zeta_n^u\quad(u\in\mathbb{Z}_n^{\times}). δ u : E ∼ E , δ u ∣ K = id K , δ u ( ζ n ) = ζ n u ( u ∈ Z n × ) .
and complex conjugation is j = δ − 1 j=\delta_{-1} j = δ − 1 . All these maps commute. Define
U Σ = { x ∈ R Σ × : x τ ( x ) = x j ( x ) = 1 } , T Σ = U Σ ∩ μ ( E ) , U_{\Sigma}=\{x\in R_{\Sigma}^{\times}:x\tau(x)=xj(x)=1\},\qquad T_{\Sigma}=U_{\Sigma}\cap\mu(E), U Σ = { x ∈ R Σ × : xτ ( x ) = x j ( x ) = 1 } , T Σ = U Σ ∩ μ ( E ) ,
Λ Σ = { λ ∈ Z [ Σ ] : τ λ = j λ = − λ } , L Σ = div ( U Σ ) . (22) \Lambda_{\Sigma}=\{\lambda\in\mathbb{Z}[\Sigma]:\tau\lambda=j\lambda=-\lambda\},\qquad L_{\Sigma}=\operatorname{div}(U_{\Sigma}). \tag*{(22)} Λ Σ = { λ ∈ Z [ Σ ] : τ λ = j λ = − λ } , L Σ = div ( U Σ ) . (22)
Thus L Σ L_{\Sigma} L Σ consists of divisors with generators satisfying both norm equations. The ordinary Kummer module is
W Σ = { [ x ] ∈ Sel n ( E , Σ ) : τ ( [ x ] ) = j ( [ x ] ) = [ x ] − 1 } . (23) W_{\Sigma}=\{[x]\in\operatorname{Sel}_n(E,\Sigma):\tau([x])=j([x])=[x]^{-1}\}. \tag*{(23)} W Σ = {[ x ] ∈ Sel n ( E , Σ ) : τ ([ x ]) = j ([ x ]) = [ x ] − 1 } . (23)
The notation suppresses E E E , n n n and the involutions τ \tau τ , j j j . For these fixed data, W Σ W_{\Sigma} W Σ is independent of the character χ \chi χ . The field and support chosen for an application may depend on the target or on a family of targets. We will prove [ Λ Σ : L Σ ] < ∞ [\Lambda_{\Sigma}:L_{\Sigma}]<\infty [ Λ Σ : L Σ ] < ∞ ; equality need not hold.
The general norm condition
In the general construction let n = p a n=p^{a} n = p a for any prime p p p , and let E / Q E/\mathbb{Q} E / Q be an abelian CM field containing K K K and μ n \mu_n μ n : E E E is totally imaginary and E + = E ⟨ j ⟩ E^{+}=E^{\langle j\rangle} E + = E ⟨ j ⟩ is totally real. Supports are now Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable. The group Gal ( E / K ) \operatorname{Gal}(E/K) Gal ( E / K ) and the map δ ↦ u δ \delta\mapsto u_{\delta} δ ↦ u δ are as in Lemma 2.4. Complex conjugation j ∈ Gal ( E / K ) j\in\operatorname{Gal}(E/K) j ∈ Gal ( E / K ) satisfies u j = − 1 u_j=-1 u j = − 1 . There need not be an automorphism τ \tau τ fixing μ n \mu_n μ n and restricting to τ K \tau_K τ K . We impose only the condition for j j j :
U Σ ( 1 ) = { x ∈ R Σ × : x j ( x ) = 1 } , T Σ ( 1 ) = μ ( E ) , U_{\Sigma}^{(1)}=\{x\in R_{\Sigma}^{\times}:xj(x)=1\},\qquad T_{\Sigma}^{(1)}=\mu(E), U Σ ( 1 ) = { x ∈ R Σ × : x j ( x ) = 1 } , T Σ ( 1 ) = μ ( E ) ,
Λ Σ ( 1 ) = { λ ∈ Z [ Σ ] : j λ = − λ } , L Σ ( 1 ) = div ( U Σ ( 1 ) ) . \Lambda_{\Sigma}^{(1)}=\{\lambda\in\mathbb{Z}[\Sigma]:j\lambda=-\lambda\},\qquad L_{\Sigma}^{(1)}=\operatorname{div}(U_{\Sigma}^{(1)}). Λ Σ ( 1 ) = { λ ∈ Z [ Σ ] : j λ = − λ } , L Σ ( 1 ) = div ( U Σ ( 1 ) ) .
Define the general Kummer module by
W Σ ( 1 ) = { [ x ] ∈ E × / E × n : x ∈ U Σ ( 1 ) } ⊂ Sel n ( E , Σ ) . (24) W_{\Sigma}^{(1)}=\{[x]\in E^{\times}/E^{\times n}:x\in U_{\Sigma}^{(1)}\}\subset\operatorname{Sel}_n(E,\Sigma). \tag*{(24)} W Σ ( 1 ) = {[ x ] ∈ E × / E × n : x ∈ U Σ ( 1 ) } ⊂ Sel n ( E , Σ ) . (24)
This module depends on E E E , n n n , Σ \Sigma Σ , j j j , again not on a particular χ \chi χ once these data are fixed. The condition x j ( x ) = 1 xj(x)=1 x j ( x ) = 1 is exact: anti-invariance j ( [ x ] ) = [ x ] − 1 j([x])=[x]^{-1} j ([ x ]) = [ x ] − 1 only makes x j ( x ) xj(x) x j ( x ) an n n n th power and is weaker at two.
These are Z n \mathbb{Z}_n Z n -modules, with the multiplicative notation fixed in Section 2. When the module is free, a Kummer basis consists of classes [ z j ] [z_j] [ z j ] ; the z j ∈ E × z_j\in E^{\times} z j ∈ E × are basis representatives . In
[ β ] = ∏ j [ z j ] e j , [\beta]=\prod_j[z_j]^{e_j}, [ β ] = j ∏ [ z j ] e j ,
the exponents e j ∈ Z n e_j\in\mathbb{Z}_n e j ∈ Z n are the Kummer coordinates . For n = p n=p n = p the module is an F p \mathbb{F}_p F p -vector space.
Divisors and Kummer classes.
Lemma 3.5. The divisor maps give exact sequences
1 ⟶ T Σ ⟶ U Σ → div L Σ ⟶ 0 ( ordinary ) , 1 \longrightarrow T_{\Sigma} \longrightarrow U_{\Sigma} \xrightarrow{\operatorname{div}} L_{\Sigma} \longrightarrow0 \qquad(\textit{ordinary}), 1 ⟶ T Σ ⟶ U Σ div L Σ ⟶ 0 ( ordinary ) , 1 ⟶ μ ( E ) ⟶ U Σ ( 1 ) → div L Σ ( 1 ) ⟶ 0 ( general ) . (25) 1 \longrightarrow\mu(E) \longrightarrow U_{\Sigma}^{(1)} \xrightarrow{\operatorname{div}} L_{\Sigma}^{(1)} \longrightarrow0 \qquad(\textit{general}). \tag*{(25)} 1 ⟶ μ ( E ) ⟶ U Σ ( 1 ) div L Σ ( 1 ) ⟶ 0 ( general ) . (25) The images L Σ L_{\Sigma} L Σ and L Σ ( 1 ) L_{\Sigma}^{(1)} L Σ ( 1 ) have finite index in Λ Σ \Lambda_{\Sigma} Λ Σ and Λ Σ ( 1 ) \Lambda_{\Sigma}^{(1)} Λ Σ ( 1 ) , respectively. In the ordinary case p ∤ ∣ T Σ ∣ p \nmid|T_{\Sigma}| p ∤ ∣ T Σ ∣ .
Proof. For a norm-one unit with zero divisor, every complex absolute value is one. Kronecker’s theorem identifies the kernels with the stated roots of unity [24 Chapter I, Proposition 7.1] . Put h E = ∣ Cl ( O E ) ∣ h_E = |\operatorname{Cl}(\mathcal{O}_E)| h E = ∣ Cl ( O E ) ∣ and choose x ∈ E × x \in E^{\times} x ∈ E × with div ( x ) = h E λ \operatorname{div}(x) = h_E\lambda div ( x ) = h E λ . Then
so both images have finite index. In the ordinary case the p p p -power roots of unity of E E E are exactly μ n \mu_n μ n , since the next cyclotomic degree p p p cannot divide [ E : Q ( ζ n ) ] = 2 [E:\mathbb{Q}(\zeta_n)] = 2 [ E : Q ( ζ n )] = 2 . As τ \tau τ fixes μ n \mu_n μ n , an element of T Σ ∩ μ n T_{\Sigma} \cap\mu_n T Σ ∩ μ n satisfies ξ 2 = 1 \xi^2 = 1 ξ 2 = 1 ; thus p ∤ ∣ T Σ ∣ p \nmid|T_{\Sigma}| p ∤ ∣ T Σ ∣ .
Proposition 3.6. Let Σ \Sigma Σ be finite and Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable, with E E E and n = p a n = p^a n = p a as in the respective construction. The class map x ↦ [ x ] x \mapsto[x] x ↦ [ x ] induces isomorphisms
U Σ / U Σ n → ∼ W Σ if Cl ( R Σ ) [ p ] = 0 , U_{\Sigma}/U_{\Sigma}^{n} \xrightarrow{\sim} W_{\Sigma} \qquad\text{if } \operatorname{Cl}(R_{\Sigma})[p] = 0, U Σ / U Σ n ∼ W Σ if Cl ( R Σ ) [ p ] = 0 , U Σ ( 1 ) / ( U Σ ( 1 ) ) n → ∼ W Σ ( 1 ) without a class-group hypothesis. U_{\Sigma}^{(1)}/(U_{\Sigma}^{(1)})^n \xrightarrow{\sim} W_{\Sigma}^{(1)} \qquad\text{without a class-group hypothesis.} U Σ ( 1 ) / ( U Σ ( 1 ) ) n ∼ W Σ ( 1 ) without a class-group hypothesis. In the general case, if n n n is odd and Cl ( R Σ ) [ p ] = 0 \operatorname{Cl}(R_{\Sigma})[p] = 0 Cl ( R Σ ) [ p ] = 0 , then
W Σ ( 1 ) = { [ x ] ∈ Sel n ( E , Σ ) : j ( [ x ] ) = [ x ] − 1 } . W_{\Sigma}^{(1)} = \{[x] \in\operatorname{Sel}_n(E,\Sigma) : j([x]) = [x]^{-1}\}. W Σ ( 1 ) = {[ x ] ∈ Sel n ( E , Σ ) : j ([ x ]) = [ x ] − 1 } . The hypothesis Cl ( R Σ ) [ p ] = 0 \operatorname{Cl}(R_{\Sigma})[p] = 0 Cl ( R Σ ) [ p ] = 0 supplies a unit representative for every Selmer class in the ordinary module. The general module is defined as the image of the exact norm-one units, so its quotient isomorphism needs no such hypothesis. The stronger convention Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma}) = 0 Cl ( R Σ ) = 0 is used only in the field constructions.
Proof. General case. Surjectivity is the definition of W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) . If y = c n ∈ U Σ ( 1 ) y = c^n \in U_{\Sigma}^{(1)} y = c n ∈ U Σ ( 1 ) with c ∈ E × c \in E^{\times} c ∈ E × , valuations put c c c in R Σ × R_{\Sigma}^{\times} R Σ × . The element c j ( c ) cj(c) c j ( c ) is totally positive and its n n n th power is one, so c j ( c ) = 1 cj(c) = 1 c j ( c ) = 1 . Thus the kernel is ( U Σ ( 1 ) ) n (U_{\Sigma}^{(1)})^n ( U Σ ( 1 ) ) n . For the equality when n n n is odd, the Selmer exact sequence provides a unit representative x x x of any supported class with j ( [ x ] ) = [ x ] − 1 j([x]) = [x]^{-1} j ([ x ]) = [ x ] − 1 . Choose b b b with 2 b ≡ 1 ( m o d n ) 2b \equiv1 \pmod n 2 b ≡ 1 ( mod n ) ; then ( x / j ( x ) ) b ∈ U Σ ( 1 ) (x/j(x))^b \in U_{\Sigma}^{(1)} ( x / j ( x ) ) b ∈ U Σ ( 1 ) and its class is [ x ] 2 b = [ x ] [x]^{2b} = [x] [ x ] 2 b = [ x ] .
Ordinary case. The same exact sequence provides a unit representative x x x of any class in W Σ W_{\Sigma} W Σ . Then
Π ( x ) = x τ ( j ( x ) ) τ ( x ) j ( x ) ∈ U Σ , [ Π ( x ) ] = [ x ] 4 . \Pi(x) = \frac{x\tau(j(x))}{\tau(x)j(x)} \in U_{\Sigma}, \qquad[\Pi(x)] = [x]^4. Π ( x ) = τ ( x ) j ( x ) xτ ( j ( x )) ∈ U Σ , [ Π ( x )] = [ x ] 4 . An exponent inverse to four modulo the odd number n n n gives a norm-one representative of [ x ] [x] [ x ] . If y = c n ∈ U Σ y = c^n \in U_{\Sigma} y = c n ∈ U Σ , put ξ = c τ ( c ) ∈ μ n \xi= c\tau(c) \in\mu_n ξ = c τ ( c ) ∈ μ n and c 0 = c ξ ( n − 1 ) / 2 c_0 = c\xi^{(n-1)/2} c 0 = c ξ ( n − 1 ) /2 . Since τ \tau τ fixes μ n \mu_n μ n , we have c 0 n = y c_0^n = y c 0 n = y and c 0 τ ( c 0 ) = 1 c_0\tau(c_0) = 1 c 0 τ ( c 0 ) = 1 . Also c 0 j ( c 0 ) c_0j(c_0) c 0 j ( c 0 ) is totally positive and ( c 0 j ( c 0 ) ) n = 1 (c_0j(c_0))^n = 1 ( c 0 j ( c 0 ) ) n = 1 , so c 0 j ( c 0 ) = 1 c_0j(c_0) = 1 c 0 j ( c 0 ) = 1 . Valuations put c 0 c_0 c 0 in R Σ × R_{\Sigma}^{\times} R Σ × . Thus the kernel is U Σ n U_{\Sigma}^n U Σ n .
Corollary 3.7. In the ordinary case assume Cl ( R Σ ) [ p ] = 0 \operatorname{Cl}(R_{\Sigma})[p] = 0 Cl ( R Σ ) [ p ] = 0 ; no class-group hypothesis is needed for the general quotient. Then
W Σ ≃ ( Z n ) r , r = rank Λ Σ , W Σ ( 1 ) ≃ ( Z n ) r 1 + 1 , r 1 = rank Λ Σ ( 1 ) . W_{\Sigma} \simeq(\mathbb{Z}_n)^r,\qquad r = \operatorname{rank}\Lambda_{\Sigma}, \qquad W_{\Sigma}^{(1)} \simeq(\mathbb{Z}_n)^{r_1+1},\qquad r_1 = \operatorname{rank}\Lambda_{\Sigma}^{(1)}. W Σ ≃ ( Z n ) r , r = rank Λ Σ , W Σ ( 1 ) ≃ ( Z n ) r 1 + 1 , r 1 = rank Λ Σ ( 1 ) .
Proof. Use the exact sequences of Lemma 3.5 and Proposition 3.6 . The free parts contribute r r r and r 1 r_1 r 1 copies of Z n \mathbf{Z}_n Z n . Ordinary torsion contributes nothing; μ ( E ) / μ ( E ) n ≃ Z n \mu(E)/\mu(E)^n \simeq\mathbf{Z}_n μ ( E ) / μ ( E ) n ≃ Z n contributes the extra general factor. □
The target classes in the Kummer modules
To use a Kummer basis, the target class must belong to its module. The ordinary case follows from its two anti-invariances. In the general case we first obtain an exact norm-one representative, then remove its valuations outside Σ \Sigma Σ . For powers of two, the first step requires signature correction and a quadratic norm equation; the coefficient enlargement below solves that equation.
Ordinary anti-invariant characters
Use the ordinary notation of § 3.2 and the support convention 3.4.
Theorem 4.1. Let χ : G m + → Z n \chi:G_m^+ \to\mathbf{Z}_n χ : G m + → Z n be anti-invariant, with τ K m = m \tau_K\mathfrak{m}=\mathfrak{m} τ K m = m . If Σ \Sigma Σ contains the primes above m \mathfrak{m} m away from p Δ p\Delta p Δ , then
ν − 1 ( χ ∣ G E ) ∈ W Σ . \nu^{-1}(\chi|_{G_E}) \in W_\Sigma. ν − 1 ( χ ∣ G E ) ∈ W Σ . Equivalently, there is β χ ∈ U Σ \beta_\chi\in U_\Sigma β χ ∈ U Σ with ν β χ = χ ∣ G E \nu_{\beta_\chi}=\chi|_{G_E} ν β χ = χ ∣ G E .
Proof. Choose ρ ∈ E × \rho\in E^\times ρ ∈ E × with ν ρ = χ ∣ G E \nu_\rho=\chi|_{G_E} ν ρ = χ ∣ G E by (14) . Kummer covariance (Lemma 2.4 ) gives j ( [ ρ ] ) = [ ρ ] − 1 j([\rho])=[\rho]^{-1} j ([ ρ ]) = [ ρ ] − 1 . Since τ \tau τ fixes μ n \mu_n μ n , Galois equivariance of reciprocity identifies its action on the restricted character with [ a ] ↦ [ τ K a ] [a]\mapsto[\tau_Ka] [ a ] ↦ [ τ K a ] ; hence τ ( [ ρ ] ) = [ ρ ] − 1 \tau([\rho])=[\rho]^{-1} τ ([ ρ ]) = [ ρ ] − 1 as well.
We verify the supported valuations. At primes above p p p , the involution j j j fixes each prime: the cyclotomic extension is totally ramified at p p p , whereas K / Q K/\mathbf{Q} K / Q is unramified there. At a prime above a rational divisor of Δ \Delta Δ , the involution τ \tau τ fixes that prime: the quadratic extension is ramified and the cyclotomic extension is unramified there. In either case, the corresponding anti-invariance gives v p ( ρ ) ≡ − v p ( ρ ) ( m o d n ) v_{\mathfrak{p}}(\rho)\equiv-v_{\mathfrak{p}}(\rho)\pmod n v p ( ρ ) ≡ − v p ( ρ ) ( mod n ) . Since n n n is odd, v p ( ρ ) ∈ n Z v_{\mathfrak{p}}(\rho)\in n\mathbf{Z} v p ( ρ ) ∈ n Z . At every other prime outside m \mathfrak{m} m , the extension E ( ρ n ) / E E(\sqrt[n]{\rho})/E E ( n ρ ) / E is unramified. Its normalized valuations therefore restrict to those of E E E , giving v p ( ρ ) ∈ n Z v_{\mathfrak{p}}(\rho)\in n\mathbf{Z} v p ( ρ ) ∈ n Z there as well. Thus [ ρ ] ∈ W Σ [\rho]\in W_\Sigma [ ρ ] ∈ W Σ .
The support convention and Proposition 3.6 now give a representative β χ ∈ U Σ \beta_\chi\in U_\Sigma β χ ∈ U Σ . □
For an arbitrary ordinary character, apply the theorem to χ − \chi^{-} χ − and recover χ \chi χ by Lemma 2.6 . The prime p p p may divide the strict class number.
General coefficients and support
Now let n = p a n=p^a n = p a and χ : G f + → Z n \chi:G_f^+\to\mathbf{Z}_n χ : G f + → Z n be arbitrary. Choose the coefficients and support as follows.
For odd p p p , put
ψ = χ , g = f , E = K ( ζ n ) . \psi=\chi,\qquad\mathfrak{g}=\mathfrak{f},\qquad E=K(\zeta_n). ψ = χ , g = f , E = K ( ζ n ) .
For p = 2 p=2 p = 2 , choose h = d h=\sqrt{d} h = d positive at σ 1 \sigma_1 σ 1 and negative at σ 2 \sigma_2 σ 2 . Let
θ − 1 , θ h : G K ⟶ Z 2 \theta_{-1},\theta_h:G_K\longrightarrow\mathbf{Z}_2 θ − 1 , θ h : G K ⟶ Z 2
be the quadratic characters of K ( i ) K(i) K ( i ) and K ( h ) K(\sqrt{h}) K ( h ) , and let d − 1 , d h \mathfrak{d}_{-1},\mathfrak{d}_h d − 1 , d h be their relative discriminant ideals. Let c i ∈ G K c_i\in G_K c i ∈ G K be complex conjugation at the real place σ i \sigma_i σ i , and write χ ( c i ) = ( n / 2 ) a i \chi(c_i)=(n/2)a_i χ ( c i ) = ( n /2 ) a i , a i ∈ { 0 , 1 } a_i\in\{0,1\} a i ∈ { 0 , 1 } . Define
θ = a 1 θ − 1 + ( a 1 + a 2 ) θ h , ψ = χ − n 2 θ . (26) \theta=a_1\theta_{-1}+(a_1+a_2)\theta_h,\qquad\psi=\chi-\frac{n}{2}\theta. \tag*{(26)} θ = a 1 θ − 1 + ( a 1 + a 2 ) θ h , ψ = χ − 2 n θ . (26)
where the coefficients of θ \theta θ are taken modulo two and Z 2 → Z n \mathbf{Z}_2\to\mathbf{Z}_n Z 2 → Z n sends 1 1 1 to n / 2 n/2 n /2 . Let a 0 ∈ { 1 , − 1 , h , − h } a_0\in\{1,-1,h,-h\} a 0 ∈ { 1 , − 1 , h , − h } have sign ( − 1 ) a i (-1)^{a_i} ( − 1 ) a i at σ i \sigma_i σ i ; then θ \theta θ is the character of K ( a 0 ) / K K(\sqrt{a_0})/K K ( a 0 ) / K . Use
g = lcm ( f , d − 1 , d h ) (27) \mathfrak{g}=\operatorname{lcm}(\mathfrak{f},\mathfrak{d}_{-1},\mathfrak{d}_h) \tag*{(27)} g = lcm ( f , d − 1 , d h ) (27)
and the following sufficient level:
b = max ( { 4 , a + 2 } ∪ { v 2 ( ℓ + 1 ) + 1 : ℓ ∣ N g , ℓ ≡ 3 ( m o d 4 ) } ) , (28) b=\max\left(\{4,a+2\}\cup\{v_2(\ell+1)+1:\ell\mid N_{\mathfrak g},\ \ell\equiv3\pmod{4}\}\right), \tag*{(28)} b = max ( { 4 , a + 2 } ∪ { v 2 ( ℓ + 1 ) + 1 : ℓ ∣ N g , ℓ ≡ 3 ( mod 4 )} ) , (28)
E = K ( ζ 2 b ) . E=K(\zeta_{2^b}). E = K ( ζ 2 b ) .
The terms a + 2 a+2 a + 2 and 4 4 4 handle primes above two; the remaining terms handle odd conductor primes, as proved below.
In either case E / Q E/\mathbb{Q} E / Q is a known abelian CM field. Write N g = ∣ O K / g ∣ N_{\mathfrak g}=|\mathcal{O}_K/\mathfrak g| N g = ∣ O K / g ∣ for the positive integer norm of g \mathfrak g g . Choose the support in two parts:
Σ r a m = { p ∈ P E : p ∩ Z = ( ℓ ) , ℓ ∣ p N g } , \Sigma_{\mathrm{ram}}=\{\mathfrak p\in\mathcal{P}_E:\mathfrak p\cap\mathbb{Z}=(\ell),\ \ell\mid pN_{\mathfrak g}\}, Σ ram = { p ∈ P E : p ∩ Z = ( ℓ ) , ℓ ∣ p N g } ,
Σ = Σ r a m ∪ Σ c l . (29) \Sigma=\Sigma_{\mathrm{ram}}\cup\Sigma_{\mathrm{cl}}. \tag*{(29)} Σ = Σ ram ∪ Σ cl . (29)
Here ℓ ∣ p N g \ell\mid pN_{\mathfrak g} ℓ ∣ p N g is divisibility of positive integers; taking all primes above ℓ \ell ℓ ensures Galois stability. Choose ideals whose classes generate Cl ( O E ) \operatorname{Cl}(\mathcal{O}_E) Cl ( O E ) , and let Σ c l \Sigma_{\mathrm{cl}} Σ cl contain their prime factors and all their Galois conjugates. Then
Cl ( R Σ ) = 0 , (30) \operatorname{Cl}(R_\Sigma)=0, \tag*{(30)} Cl ( R Σ ) = 0 , (30)
as in Lemma 3.3 .
An explicit alternative, requiring no class-group generators, is
Σ c l = { p ∈ P E : N ( p ) ≤ B E } , [ E : Q ] = 2 r E , \Sigma_{\mathrm{cl}}=\{\mathfrak p\in\mathcal{P}_E:N(\mathfrak p)\le B_E\},\qquad[E:\mathbb{Q}]=2r_E, Σ cl = { p ∈ P E : N ( p ) ≤ B E } , [ E : Q ] = 2 r E ,
B E = ( 4 π ) r E ( 2 r E ) ! ( 2 r E ) 2 r E ∣ disc ( E ) ∣ . (31) B_E=\left(\frac{4}{\pi}\right)^{r_E}\frac{(2r_E)!}{(2r_E)^{2r_E}}\sqrt{|\operatorname{disc}(E)|}. \tag*{(31)} B E = ( π 4 ) r E ( 2 r E ) 2 r E ( 2 r E )! ∣ disc ( E ) ∣ . (31)
Minkowski’s theorem gives an integral ideal of norm at most B E B_E B E in each class, whose prime factors lie in this set [24 Chapter I, Section 6, Exercise 3] . This larger support is used in the Hilbert-field application.
Theorem 4.2. For the preceding coefficient field E E E and character ψ \psi ψ , let Σ \Sigma Σ be any finite Galois-stable support containing Σ r a m \Sigma_{\mathrm{ram}} Σ ram with Cl ( R Σ ) = 0 \operatorname{Cl}(R_\Sigma)=0 Cl ( R Σ ) = 0 , as in (29) . Then
ν − 1 ( ψ ∣ G E ) ∈ W Σ ( 1 ) . \nu^{-1}(\psi|_{G_E})\in W_\Sigma^{(1)}. ν − 1 ( ψ ∣ G E ) ∈ W Σ ( 1 ) . Equivalently, there exists β ψ ∈ R Σ × \beta_\psi\in R_\Sigma^\times β ψ ∈ R Σ × with
β ψ j ( β ψ ) = 1 , ν β ψ = ψ ∣ G E . \beta_\psi j(\beta_\psi)=1,\qquad\nu_{\beta_\psi}=\psi|_{G_E}. β ψ j ( β ψ ) = 1 , ν β ψ = ψ ∣ G E . For p = 2 p=2 p = 2 , ψ \psi ψ is totally even with conductor dividing g \mathfrak g g . There is no restriction on p p p relative to Δ \Delta Δ or any class number.
The next three subsections prove the ingredients: exact norm one at odd order, the dyadic norm equation after signature correction, and support normalization. We combine them in the proof at the end of the section.
Odd order cases
Lemma 4.3. If n n n is odd and j ( [ ρ ] ) = [ ρ ] − 1 j([\rho])=[\rho]^{-1} j ([ ρ ]) = [ ρ ] − 1 in E × / E × n E^\times/E^{\times n} E × / E × n , then [ ρ ] [\rho] [ ρ ] has a norm-one representative. In particular this applies to ν − 1 ( χ ∣ G E ) \nu^{-1}(\chi|_{G_E}) ν − 1 ( χ ∣ G E ) for every character χ : G K → Z n \chi:G_K\to\mathbb{Z}_n χ : G K → Z n , even when p ∣ Δ p\mid\Delta p ∣ Δ .
Proof. Choose k ∈ Z k\in\mathbb{Z} k ∈ Z with 2 k ≡ 1 ( m o d n ) 2k\equiv1\pmod n 2 k ≡ 1 ( mod n ) . Then
β = ( ρ / j ( ρ ) ) k , β j ( β ) = 1 , [ β ] = [ ρ ] 2 k = [ ρ ] . \beta=(\rho/j(\rho))^k,\qquad\beta j(\beta)=1,\qquad[\beta]=[\rho]^{2k}=[\rho]. β = ( ρ / j ( ρ ) ) k , β j ( β ) = 1 , [ β ] = [ ρ ] 2 k = [ ρ ] .
For a restricted character, Lemma 2.4 gives the required anti-invariance under j j j .
Lemma 4.4. The character ψ \psi ψ in (26) vanishes at both real inertia groups, and its finite conductor divides g \mathfrak g g .
Proof. At the two real places,
( θ − 1 ( c 1 ) , θ − 1 ( c 2 ) ) = ( 1 , 1 ) , ( θ h ( c 1 ) , θ h ( c 2 ) ) = ( 0 , 1 ) , (\theta_{-1}(c_1),\theta_{-1}(c_2))=(1,1),\qquad(\theta_h(c_1),\theta_h(c_2))=(0,1), ( θ − 1 ( c 1 ) , θ − 1 ( c 2 )) = ( 1 , 1 ) , ( θ h ( c 1 ) , θ h ( c 2 )) = ( 0 , 1 ) ,
so (26) gives ψ ( c 1 ) = ψ ( c 2 ) = 0 \psi(c_1)=\psi(c_2)=0 ψ ( c 1 ) = ψ ( c 2 ) = 0 . The conductor assertion follows from the quadratic conductor–discriminant formula and the conductor bound for a sum [30 Chapter VI, Section 3] . □
Thus H ψ H_\psi H ψ is totally real, which supplies the positivity needed for the norm equation.
The dyadic cases
Write E 0 = K ( ζ 2 n ) = k 0 ( i ) E_0=K(\zeta_{2n})=k_0(i) E 0 = K ( ζ 2 n ) = k 0 ( i ) , where k 0 = E 0 + k_0=E_0^+ k 0 = E 0 + . The obstruction below is the solvability of
N E 0 / k 0 ( b ) = A , A ∈ k 0 × , + . N_{E_0/k_0}(b)=A,\qquad A\in k_0^{\times,+}. N E 0 / k 0 ( b ) = A , A ∈ k 0 × , + .
By the Hasse norm theorem it suffices to solve this equation at every completion [25 Part III, Corollary 4.8] . We use the following local criteria: an unramified quadratic extension has norm group consisting of the elements of even valuation, and its quadratic norm obstruction is killed by an even local extension degree. The latter follows from
( − 1 , A ) w = ( − 1 , A ) v [ k w : ( k 0 ) v ] , (-1,A)_w=(-1,A)_v^{[k_w:(k_0)_v]}, ( − 1 , A ) w = ( − 1 , A ) v [ k w : ( k 0 ) v ] ,
where ( − 1 , A ) v = 1 (-1,A)_v=1 ( − 1 , A ) v = 1 means that A A A is a norm from k 0 , v ( i ) k_{0,v}(i) k 0 , v ( i ) [30 Chapter XIII, Section 3, Proposition 7] .
Lemma 4.5. Let H = H ψ H=H_\psi H = H ψ , for a totally even character ψ : G K → Z n \psi:G_K\to\mathbb{Z}_n ψ : G K → Z n , where n = 2 a n=2^a n = 2 a , with finite conductor dividing g \mathfrak{g} g . Put E 0 = K ( ζ 2 n ) E_0=K(\zeta_{2n}) E 0 = K ( ζ 2 n ) , k 0 = E 0 + k_0=E_0^+ k 0 = E 0 + , and L = H E 0 L=HE_0 L = H E 0 . There is α ∈ L × \alpha\in L^\times α ∈ L × such that
α n = ρ ∈ E 0 × , ν ρ = ψ ∣ G E 0 . \alpha^n=\rho\in E_0^\times,\qquad\nu_\rho=\psi|_{G_{E_0}}. α n = ρ ∈ E 0 × , ν ρ = ψ ∣ G E 0 . If J J J is complex conjugation on L L L , then
A = α J ( α ) ∈ k 0 × , + . A=\alpha J(\alpha)\in k_0^{\times,+}. A = α J ( α ) ∈ k 0 × , + . The class [ ρ ] [\rho] [ ρ ] has a norm-one representative over E 0 E_0 E 0 if and only if
b 0 j ( b 0 ) = A ( b 0 ∈ E 0 × ) . (32) b_0j(b_0)=A\qquad(b_0\in E_0^\times). \tag*{(32)} b 0 j ( b 0 ) = A ( b 0 ∈ E 0 × ) . (32) This equation is locally solvable at every real place and at every finite place not lying above 2 g 2\mathfrak{g} 2 g .
Proof. Since H H H and k 0 k_0 k 0 are totally real, L = H k 0 ( i ) L=Hk_0(i) L = H k 0 ( i ) is CM. Its conjugation J J J fixes H H H and restricts to j j j on E 0 E_0 E 0 . Kummer theory supplies α \alpha α in L L L , even if the restriction has order smaller than n n n [23 Theorem 5.30] . For g ∈ Gal ( L / E 0 ) g\in\operatorname{Gal}(L/E_0) g ∈ Gal ( L / E 0 ) , write g ( α ) = ζ n ψ ( g ) α g(\alpha)=\zeta_n^{\psi(g)}\alpha g ( α ) = ζ n ψ ( g ) α . The maps g g g and J J J commute, and J J J inverts ζ n \zeta_n ζ n . Hence g ( α J ( α ) ) = α J ( α ) g(\alpha J(\alpha))=\alpha J(\alpha) g ( α J ( α )) = α J ( α ) , so A ∈ E 0 A\in E_0 A ∈ E 0 . It is fixed by J J J and positive at every real embedding of k 0 k_0 k 0 .
Every representative of [ ρ ] [\rho] [ ρ ] is ( α / b 0 ) n (\alpha/b_0)^n ( α / b 0 ) n for some b 0 ∈ E 0 × b_0\in E_0^\times b 0 ∈ E 0 × . Its norm is ( A / ( b 0 j ( b 0 ) ) ) n (A/(b_0j(b_0)))^n ( A / ( b 0 j ( b 0 )) ) n . The ratio is totally positive, so that norm equals one precisely when (32) holds. Here E 0 = k 0 ( i ) E_0=k_0(i) E 0 = k 0 ( i ) , so it is the quadratic norm equation just described. Positivity gives a solution at real places, and a split local extension imposes no condition.
At an odd nonsplit place outside g \mathfrak{g} g , both E 0 / k 0 E_0/k_0 E 0 / k 0 and L / E 0 L/E_0 L / E 0 are unramified. The valuations of α \alpha α at all primes of L L L above a fixed prime of E 0 E_0 E 0 are equal, since its Galois conjugates differ by roots of unity. That prime of E 0 E_0 E 0 is fixed by j j j , so α \alpha α and J ( α ) J(\alpha) J ( α ) have the same valuation. The ramification indices are one; consequently v ( A ) = 2 v ( α ) v(A)=2v(\alpha) v ( A ) = 2 v ( α ) is even in k 0 k_0 k 0 . The unramified quadratic norm criterion proves local solvability there. □
Proposition 4.6. For E E E in (28) , every totally even character of exponent dividing 2 a 2^a 2 a and finite conductor dividing g \mathfrak{g} g has a norm-one representative after restriction to E E E .
Proof. Put k = E + k=E^{+} k = E + and use A ∈ k 0 × , + A\in k_{0}^{\times,+} A ∈ k 0 × , + from Lemma 4.5. We check the remaining local norm conditions and then apply the Hasse norm theorem.
Odd primes. At a rational prime ℓ ≡ 1 ( m o d 4 ) \ell\equiv1\pmod{4} ℓ ≡ 1 ( mod 4 ) , i ∈ Q ℓ i\in\mathbb{Q}_{\ell} i ∈ Q ℓ , so the extension is locally split. For ℓ ≡ 3 ( m o d 4 ) \ell\equiv3\pmod{4} ℓ ≡ 3 ( mod 4 ) , put r = v 2 ( ℓ + 1 ) r=v_{2}(\ell+1) r = v 2 ( ℓ + 1 ) . Its residue degree in Q ( ζ 2 b ) + \mathbb{Q}(\zeta_{2^{b}})^{+} Q ( ζ 2 b ) + is
ord Z 2 b × / { ± 1 } ( ℓ ) = 2 max ( 0 , b − r ) . \operatorname{ord}_{\mathbb{Z}_{2^{b}}^{\times}/\{\pm1\}}(\ell)=2^{\max(0,b-r)}. ord Z 2 b × / { ± 1 } ( ℓ ) = 2 m a x ( 0 , b − r ) .
Indeed odd powers of ℓ \ell ℓ are congruent to − 1 -1 − 1 modulo exactly the same power 2 r 2^{r} 2 r , whereas for even m m m , v 2 ( ℓ m − 1 ) = r + v 2 ( m ) v_{2}(\ell^{m}-1)=r+v_{2}(m) v 2 ( ℓ m − 1 ) = r + v 2 ( m ) . If ℓ ∣ N ( g ) \ell\mid N(\mathfrak{g}) ℓ ∣ N ( g ) , our choice b ≥ r + 1 b\ge r+1 b ≥ r + 1 makes the residue degree even. The completion of this real cyclotomic field is unramified at ℓ \ell ℓ , and contains the unramified quadratic extension Q ℓ ( i ) \mathbb{Q}_{\ell}(i) Q ℓ ( i ) . Thus E / k E/k E / k is locally split at those primes, also after adjoining K K K . At all other odd primes the local norm equation already had a solution over k 0 k_{0} k 0 .
Primes above two. Here k / k 0 k/k_{0} k / k 0 has even local degree. Write F t = Q 2 ( ζ 2 t + ζ 2 t − 1 ) F_{t}=\mathbb{Q}_{2}(\zeta_{2^{t}}+\zeta_{2^{t}}^{-1}) F t = Q 2 ( ζ 2 t + ζ 2 t − 1 ) . For a ≥ 2 a\ge2 a ≥ 2 , F a + 2 / F a + 1 F_{a+2}/F_{a+1} F a + 2 / F a + 1 is quadratic and F a + 2 / Q 2 F_{a+2}/\mathbb{Q}_{2} F a + 2 / Q 2 is cyclic; its unique quadratic subfield Q 2 ( 2 ) \mathbb{Q}_{2}(\sqrt{2}) Q 2 ( 2 ) already lies in F a + 1 F_{a+1} F a + 1 . A completion of the quadratic field K K K cannot consume this step: its intersection with F a + 2 F_{a+2} F a + 2 is either Q 2 \mathbb{Q}_{2} Q 2 or that quadratic subfield, and in both cases is already contained in F a + 1 F_{a+1} F a + 1 . For a = 1 a=1 a = 1 , k 0 = K k_{0}=K k 0 = K and [ F 4 : Q 2 ] = 4 [F_{4}:\mathbb{Q}_{2}]=4 [ F 4 : Q 2 ] = 4 , so adjoining F 4 F_{4} F 4 to a completion of K K K has even degree. The bound b ≥ max ( 4 , a + 2 ) b\ge\max(4,a+2) b ≥ max ( 4 , a + 2 ) includes these steps, and further enlargement preserves evenness. The local degree formula for the quadratic Hilbert symbol therefore makes A A A a norm at the primes above two. Positivity handles infinity.
Global norm. The cyclic Hasse norm theorem gives b 1 ∈ E × b_{1}\in E^{\times} b 1 ∈ E × with b 1 j ( b 1 ) = A b_{1}j(b_{1})=A b 1 j ( b 1 ) = A [25 Part III, Corollary 4.8] . In H E HE H E , put z = α / b 1 z=\alpha/b_{1} z = α / b 1 . Then z J ( z ) = 1 zJ(z)=1 z J ( z ) = 1 and β = z n = ρ / b 1 n ∈ E × \beta=z^{n}=\rho/b_{1}^{n}\in E^{\times} β = z n = ρ / b 1 n ∈ E × represents the same restricted character and satisfies β j ( β ) = 1 \beta j(\beta)=1 β j ( β ) = 1 . □
Example 4.7. Take the totally real extension H = K ( 11 ) H=K(\sqrt{11}) H = K ( 11 ) of K = Q ( 3 ) K=\mathbb{Q}(\sqrt{3}) K = Q ( 3 ) . Over E 0 = K ( i ) = Q ( ζ 12 ) E_{0}=K(i)=\mathbb{Q}(\zeta_{12}) E 0 = K ( i ) = Q ( ζ 12 ) its class is [ 11 ] [11] [ 11 ] , with j ( [ 11 ] ) = [ 11 ] = [ 11 ] − 1 j([11])=[11]=[11]^{-1} j ([ 11 ]) = [ 11 ] = [ 11 ] − 1 modulo squares. The two primes over 11 11 11 are j j j -fixed, since their Frobenius is 11 ≡ − 1 ( m o d 12 ) 11\equiv-1\pmod{12} 11 ≡ − 1 ( mod 12 ) . At each, every representative 11 c 2 11c^{2} 11 c 2 has odd valuation, while a norm-one element has valuation zero. Thus no norm-one representative exists. Nevertheless Cl ( O E 0 ) = 0 \operatorname{Cl}(\mathcal{O}_{E_{0}})=0 Cl ( O E 0 ) = 0 : the discriminant is 144 144 144 and Minkowski’s bound is 18 / π 2 < 2 18/\pi^{2}<2 18/ π 2 < 2 .
The product of the three quadratic discriminants gives disc ( H ) = 12 ⋅ 44 ⋅ 33 = 12 2 ⋅ 11 2 \operatorname{disc}(H)=12\cdot44\cdot33=12^{2}\cdot11^{2} disc ( H ) = 12 ⋅ 44 ⋅ 33 = 1 2 2 ⋅ 1 1 2 . Both primes over 11 11 11 ramify, so the relative discriminant and finite conductor are ( 11 ) (11) ( 11 ) . The sufficient level is b = 4 b=4 b = 4 . In E = K ( ζ 16 ) E=K(\zeta_{16}) E = K ( ζ 16 ) , take
b 1 = 3 + 2 i 2 , b 1 j ( b 1 ) = 11 , β = 11 b 1 2 = − 5 − 4 6 11 . b_{1}=\sqrt{3}+2i\sqrt{2},\qquad b_{1}j(b_{1})=11,\qquad\beta=\frac{11}{b_{1}^{2}}=\frac{-5-4\sqrt{6}}{11}. b 1 = 3 + 2 i 2 , b 1 j ( b 1 ) = 11 , β = b 1 2 11 = 11 − 5 − 4 6 . Then β j ( β ) = 1 \beta j(\beta)=1 β j ( β ) = 1 and [ β ] = [ 11 ] [\beta]=[11] [ β ] = [ 11 ] . This solution already lies in K ( ζ 8 ) K(\zeta_{8}) K ( ζ 8 ) , below the uniform sufficient level.
Removing the remaining valuations
We can now remove valuations outside the chosen support while preserving norm one. Only the vanishing of the localized class group is used here.
Lemma 4.8. Let Σ \Sigma Σ be any finite j j j -stable support in the CM coefficient field E E E with Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma})=0 Cl ( R Σ ) = 0 . If ρ j ( ρ ) = 1 \rho j(\rho)=1 ρ j ( ρ ) = 1 and [ ρ ] ∈ Sel n ( E , Σ ) [\rho]\in\operatorname{Sel}_{n}(E,\Sigma) [ ρ ] ∈ Sel n ( E , Σ ) , then its class belongs to W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) .
Proof. Since ρ j ( ρ ) = 1 \rho j(\rho)=1 ρ j ( ρ ) = 1 , its divisor vanishes at j j j -fixed primes. Choosing one prime from each other pair and using the Selmer condition, write
∑ p ∉ Σ v p ( ρ ) [ p ] = n ( D 0 − j D 0 ) \sum_{\mathfrak{p}\notin\Sigma}v_{\mathfrak{p}}(\rho)[\mathfrak{p}]=n(D_{0}-jD_{0}) p ∈ / Σ ∑ v p ( ρ ) [ p ] = n ( D 0 − j D 0 )
for a divisor D 0 D_0 D 0 outside Σ \Sigma Σ . Triviality of Cl ( R Σ ) \operatorname{Cl}(R_\Sigma) Cl ( R Σ ) gives c ∈ E × c \in E^\times c ∈ E × with ∑ p ∉ Σ v p ( c ) [ p ] = D 0 \sum_{\mathfrak p\notin\Sigma} v_{\mathfrak p}(c)[\mathfrak p] = D_0 ∑ p ∈ / Σ v p ( c ) [ p ] = D 0 . Then
β = ρ / ( c / j ( c ) ) n ∈ R Σ × , β j ( β ) = 1 , [ β ] = [ ρ ] . (33) \beta= \rho/(c/j(c))^n \in R_\Sigma^\times,\qquad\beta j(\beta) = 1,\qquad[\beta] = [\rho]. \tag*{(33)} β = ρ / ( c / j ( c ) ) n ∈ R Σ × , β j ( β ) = 1 , [ β ] = [ ρ ] . (33)
This gives the required norm-one unit. □
Proof of Theorem 4.2. 1. Exact norm one. For odd p p p , Lemma 4.3 gives ρ ∈ E × \rho\in E^\times ρ ∈ E × with v ρ = χ ∣ G E v_\rho= \chi|_{G_E} v ρ = χ ∣ G E and ρ j ( ρ ) = 1 \rho j(\rho) = 1 ρ j ( ρ ) = 1 . For p = 2 p = 2 p = 2 , Lemma 4.4 makes ψ \psi ψ totally even; Proposition 4.6 then gives such a ρ \rho ρ over the specified field E E E .
The Selmer condition. The extension E ( ρ n ) / E E(\sqrt[n]{\rho})/E E ( n ρ ) / E is unramified outside the primes above p g \mathfrak p g p g . For p ∉ Σ \mathfrak p \notin\Sigma p ∈ / Σ , choose a normalized valuation w w w above p \mathfrak p p in this extension. Since its ramification index is one,
v p ( ρ ) = n w ( ρ n ) ∈ n Z , [ ρ ] ∈ Sel n ( E , Σ ) . v_{\mathfrak p}(\rho) = n w(\sqrt[n]{\rho}) \in n\mathbb{Z},\qquad[\rho] \in\operatorname{Sel}_n(E,\Sigma). v p ( ρ ) = n w ( n ρ ) ∈ n Z , [ ρ ] ∈ Sel n ( E , Σ ) .
A supported representative. Lemma 4.8 uses Cl ( R Σ ) = 0 \operatorname{Cl}(R_\Sigma) = 0 Cl ( R Σ ) = 0 to replace ρ \rho ρ by a norm-one Σ \Sigma Σ -unit in the same class. Thus [ ρ ] ∈ W Σ ( 1 ) [\rho] \in W_\Sigma^{(1)} [ ρ ] ∈ W Σ ( 1 ) . □
Ordinary Fermat Jacobians
Use the ordinary notation of Section 3.2:
n = p a > 1 , p ∤ 2 Δ , m = Δ n , F = Q ( ζ m ) , E = K ( ζ n ) . n = p^a > 1,\qquad p \nmid2\Delta,\qquad m = \Delta n,\qquad F = \mathbb{Q}(\zeta_m),\qquad E = K(\zeta_n). n = p a > 1 , p ∤ 2Δ , m = Δ n , F = Q ( ζ m ) , E = K ( ζ n ) .
Choose ζ m = e 2 π i / m \zeta_m = e^{2\pi i/m} ζ m = e 2 π i / m , ζ n = ζ m Δ \zeta_n = \zeta_m^\Delta ζ n = ζ m Δ and ζ Δ = ζ m n \zeta_\Delta= \zeta_m^n ζ Δ = ζ m n . Write χ Δ \chi_\Delta χ Δ for the primitive quadratic character modulo Δ \Delta Δ ; τ \tau τ , δ u \delta_u δ u , j j j have their earlier meanings. For an arbitrary ordinary character, Lemma 2.6 first separates the cyclotomic part; we construct the Fermat input for its anti-invariant part.
The classical Fermat decomposition gives one-dimensional character spaces and CM by F F F on the factor primitive in both indices [21 Section 1] [27 Section 4.2] . Restriction to K ⊂ Q ( ζ Δ ) K \subset\mathbb{Q}(\zeta_\Delta) K ⊂ Q ( ζ Δ ) groups these spaces according to its two real embeddings. Their Frobenius determinant quotient is the input A q ( v ) A_q(v) A q ( v ) : CM supplies the eigenvalues, and RM determines their exponents + 1 +1 + 1 or − 1 -1 − 1 .
The curve, its endomorphisms, and its determinant line
Our geometric input is the smooth projective model of the affine curve
C n , Δ : Y n + Z Δ = 1. (34) C_{n,\Delta}: \quad Y^n + Z^\Delta= 1. \tag*{(34)} C n , Δ : Y n + Z Δ = 1. (34)
Write C = C n , Δ C = C_{n,\Delta} C = C n , Δ and J = Jac ( C ) J = \operatorname{Jac}(C) J = Jac ( C ) . Its two coordinate actions will give the cyclotomic and quadratic structures below. The Fermat curve F m : X 1 m + X 2 m = 1 F_m : X_1^m + X_2^m = 1 F m : X 1 m + X 2 m = 1 provides the cover used to compute its character spaces and Frobenius:
F m ⟶ C , ( X 1 , X 2 ) ⟼ ( X 1 Δ , X 2 n ) . (35) F_m \longrightarrow C,\qquad(X_1,X_2) \longmapsto(X_1^\Delta,X_2^n). \tag*{(35)} F m ⟶ C , ( X 1 , X 2 ) ⟼ ( X 1 Δ , X 2 n ) . (35)
The curve C C C has a unique point ∞ \infty ∞ over infinity. The deck transformations S ( Y , Z ) = ( ζ n Y , Z ) S(Y,Z) = (\zeta_nY,Z) S ( Y , Z ) = ( ζ n Y , Z ) and T ( Y , Z ) = ( Y , ζ Δ Z ) T(Y,Z) = (Y,\zeta_\Delta Z) T ( Y , Z ) = ( Y , ζ Δ Z ) act on J F J_F J F by [ P − ∞ ] ↦ [ S ( P ) − ∞ ] [P-\infty] \mapsto[S(P)-\infty] [ P − ∞ ] ↦ [ S ( P ) − ∞ ] and similarly for T T T . They commute, and S n = T Δ = 1 S^n = T^\Delta= 1 S n = T Δ = 1 . On the open subset Y Z ≠ 0 YZ \ne0 Y Z = 0 they are the deck transformations of the finite étale cover
C ∘ ⟶ P F 1 ∖ { 0 , 1 , ∞ } , x = Y n , with group μ n × μ Δ . (36) C^\circ\longrightarrow\mathbb{P}^1_F \setminus\{0,1,\infty\},\qquad x = Y^n,\qquad\text{with group } \mu_n \times\mu_\Delta. \tag*{(36)} C ∘ ⟶ P F 1 ∖ { 0 , 1 , ∞ } , x = Y n , with group μ n × μ Δ . (36)
Fix an auxiliary prime ℓ 0 \ell_0 ℓ 0 and an embedding ι : F ↪ Q ‾ ℓ 0 \iota: F \hookrightarrow\overline{\mathbb{Q}}_{\ell_0} ι : F ↪ Q ℓ 0 . The Tate module and rational Tate module are
T ℓ 0 ( J ) = l i m ← r J [ ℓ 0 r ] ( F ‾ ) , V ℓ 0 ( J ) = T ℓ 0 ( J ) ⊗ Z ℓ 0 Q ℓ 0 . T_{\ell_0}(J) = \varprojlim_r J[\ell_0^r](\overline{F}),\qquad V_{\ell_0}(J) = T_{\ell_0}(J) \otimes_{\mathbb{Z}_{\ell_0}} \mathbb{Q}_{\ell_0}. T ℓ 0 ( J ) = r lim J [ ℓ 0 r ] ( F ) , V ℓ 0 ( J ) = T ℓ 0 ( J ) ⊗ Z ℓ 0 Q ℓ 0 .
The transition maps are multiplication by ℓ 0 \ell_0 ℓ 0 . We work on V = V ℓ 0 ( J ) ⊗ Q ℓ 0 Q ‾ ℓ 0 V = V_{\ell_0}(J) \otimes_{\mathbb{Q}_{\ell_0}} \overline{\mathbb{Q}}_{\ell_0} V = V ℓ 0 ( J ) ⊗ Q ℓ 0 Q ℓ 0 . The prime ℓ 0 \ell_0 ℓ 0 is used only to form the Tate module; it must differ from the residue characteristic of a prime where Frobenius is evaluated. The same convention applies in Section 10. The actions of S , T S,T S , T on torsion points induce commuting Q ℓ 0 \mathbb{Q}_{\ell_0} Q ℓ 0 -linear endomorphisms of V V V . For v ∈ Z n v \in\mathbb{Z}_n v ∈ Z n and c ∈ Z Δ c \in\mathbb{Z}_\Delta c ∈ Z Δ , define
V v , c = { w ∈ V : S w = ι ( ζ n v ) w , T w = ι ( ζ Δ c ) w } . (37) V_{v,c}=\{w\in V:S w=\iota(\zeta_n^v)w,\ T w=\iota(\zeta_\Delta^c)w\}. \tag*{(37)} V v , c = { w ∈ V : S w = ι ( ζ n v ) w , T w = ι ( ζ Δ c ) w } . (37)
The usual character projectors in End Q ‾ ℓ 0 ( V ) \operatorname{End}_{\overline{\mathbb{Q}}_{\ell_0}}(V) End Q ℓ 0 ( V ) are
e v Y = 1 n ∑ b = 0 n − 1 ι ( ζ n − v b ) S b , e c Z = 1 Δ ∑ b = 0 Δ − 1 ι ( ζ Δ − c b ) T b . (38) e_v^Y=\frac{1}{n}\sum_{b=0}^{n-1}\iota(\zeta_n^{-vb})S^b,\qquad e_c^Z=\frac{1}{\Delta}\sum_{b=0}^{\Delta-1}\iota(\zeta_\Delta^{-cb})T^b. \tag*{(38)} e v Y = n 1 b = 0 ∑ n − 1 ι ( ζ n − v b ) S b , e c Z = Δ 1 b = 0 ∑ Δ − 1 ι ( ζ Δ − c b ) T b . (38)
Character orthogonality gives idempotents with e v Y e c Z V = V v , c e_v^Y e_c^Z V=V_{v,c} e v Y e c Z V = V v , c ; their scalar denominators lie in Q ‾ ℓ 0 \overline{\mathbb{Q}}_{\ell_0} Q ℓ 0 . To select the primitive Δ \Delta Δ -characters, set
e Δ p r i m = 1 Δ ∑ b = 0 Δ − 1 r Δ ( b ) T b , r Δ ( b ) = ∑ c ∈ ( Z Δ ) × ζ Δ − c b ∈ Z , J p r i m = im ( Δ e Δ p r i m : J → J ) . (39) e_\Delta^{\mathrm{prim}}=\frac{1}{\Delta}\sum_{b=0}^{\Delta-1}r_\Delta(b)T^b,\qquad r_\Delta(b)=\sum_{c\in(\mathbb{Z}_\Delta)^\times}\zeta_\Delta^{-cb}\in\mathbb{Z},\qquad J^{\mathrm{prim}}=\operatorname{im}(\Delta e_\Delta^{\mathrm{prim}}:J\to J). \tag*{(39)} e Δ prim = Δ 1 b = 0 ∑ Δ − 1 r Δ ( b ) T b , r Δ ( b ) = c ∈ ( Z Δ ) × ∑ ζ Δ − c b ∈ Z , J prim = im ( Δ e Δ prim : J → J ) . (39)
Here r Δ ( b ) = ∑ d ∣ ( Δ , b ) d μ ( Δ / d ) r_\Delta(b)=\sum_{d\mid(\Delta,b)}d\mu(\Delta/d) r Δ ( b ) = ∑ d ∣ ( Δ , b ) d μ ( Δ/ d ) is the Ramanujan sum, with μ \mu μ the Möbius function [2 Chapter 8] . Its integral coefficients make Δ e Δ p r i m \Delta e_\Delta^{\mathrm{prim}} Δ e Δ prim an endomorphism of J J J . Below r = φ ( Δ ) / 2 r=\varphi(\Delta)/2 r = φ ( Δ ) /2 . Define the Gaussian endomorphism
G Δ = ∑ b ∈ ( Z Δ ) × χ Δ ( b ) T b . (40) G_\Delta=\sum_{b\in(\mathbb{Z}_\Delta)^\times}\chi_\Delta(b)T^b. \tag*{(40)} G Δ = b ∈ ( Z Δ ) × ∑ χ Δ ( b ) T b . (40)
For v ∈ ( Z n ) × v\in(\mathbb{Z}_n)^\times v ∈ ( Z n ) × put
M v = e v Y e Δ p r i m V , M v ± = ker ( G Δ ∓ ι ( Δ ) ∣ M v ) . (41) M_v=e_v^Y e_\Delta^{\mathrm{prim}}V,\qquad M_v^\pm=\ker(G_\Delta\mp\iota(\sqrt{\Delta})\mid M_v). \tag*{(41)} M v = e v Y e Δ prim V , M v ± = ker ( G Δ ∓ ι ( Δ ) ∣ M v ) . (41)
The Gaussian sum realizes Δ ∈ K \sqrt{\Delta}\in K Δ ∈ K on these spaces: M v + M_v^+ M v + collects the lines with χ Δ ( c ) = 1 \chi_\Delta(c)=1 χ Δ ( c ) = 1 , and M v − M_v^- M v − those with χ Δ ( c ) = − 1 \chi_\Delta(c)=-1 χ Δ ( c ) = − 1 . Both have dimension r r r , as proved below. Their determinant quotient is the line
L v = ⋀ r M v + ⊗ ( ⋀ r M v − ) ∨ . (42) \mathcal{L}_v=\bigwedge^r M_v^+\otimes\left(\bigwedge^r M_v^-\right)^\vee. \tag*{(42)} L v = ⋀ r M v + ⊗ ( ⋀ r M v − ) ∨ . (42)
Let q \mathfrak{q} q be a prime ideal of F F F over a rational prime ℓ ∤ Δ n \ell\nmid\Delta n ℓ ∤ Δ n , put k q = O F / q k_{\mathfrak{q}}=\mathcal{O}_F/\mathfrak{q} k q = O F / q and q = ∣ k q ∣ q=|k_{\mathfrak{q}}| q = ∣ k q ∣ , and now require ℓ 0 ≠ ℓ \ell_0\ne\ell ℓ 0 = ℓ . The curve has good reduction: normalization of the tame cover with disjoint branch sections 0 , 1 , ∞ 0,1,\infty 0 , 1 , ∞ is smooth. The residue field contains the primitive root ζ ‾ m = ζ m m o d q \overline{\zeta}_m=\zeta_m\bmod\mathfrak{q} ζ m = ζ m mod q , so m ∣ q − 1 m\mid q-1 m ∣ q − 1 . Choose a generator g g g of k q × k_{\mathfrak{q}}^\times k q × and s q ∈ Z m × s_{\mathfrak{q}}\in\mathbb{Z}_m^\times s q ∈ Z m × satisfying
g ( q − 1 ) / m = ζ ‾ m s q . (43) g^{(q-1)/m}=\overline{\zeta}_m^{s_{\mathfrak{q}}}. \tag*{(43)} g ( q − 1 ) / m = ζ m s q . (43)
The power-residue characters are
θ q : k q × ⟶ μ n ( F ) , g t ⟼ ζ n s q t , \theta_{\mathfrak{q}}:k_{\mathfrak{q}}^\times\longrightarrow\mu_n(F),\qquad g^t\longmapsto\zeta_n^{s_{\mathfrak{q}}t}, θ q : k q × ⟶ μ n ( F ) , g t ⟼ ζ n s q t ,
ξ q : k q × ⟶ μ Δ ( F ) , g t ⟼ ζ Δ s q t , t ∈ Z q − 1 . \xi_{\mathfrak{q}}:k_{\mathfrak{q}}^\times\longrightarrow\mu_\Delta(F),\qquad g^t\longmapsto\zeta_\Delta^{s_{\mathfrak{q}}t},\qquad t\in\mathbb{Z}_{q-1}. ξ q : k q × ⟶ μ Δ ( F ) , g t ⟼ ζ Δ s q t , t ∈ Z q − 1 .
For multiplicative characters λ 1 , λ 2 : k q × → Q ‾ × \lambda_1,\lambda_2:k_{\mathfrak{q}}^\times\to\overline{\mathbb{Q}}^\times λ 1 , λ 2 : k q × → Q × , extend each by zero at zero and define the Jacobi sum as follows [5 Section 2.1] . In particular, for v ∈ ( Z n ) × v\in(\mathbb{Z}_n)^\times v ∈ ( Z n ) × and c ∈ ( Z Δ ) × c\in(\mathbb{Z}_\Delta)^\times c ∈ ( Z Δ ) × ,
J ( λ 1 , λ 2 ) = ∑ x ∈ k q λ 1 ( x ) λ 2 ( 1 − x ) , J(\lambda_1,\lambda_2)=\sum_{x\in k_{\mathfrak{q}}}\lambda_1(x)\lambda_2(1-x), J ( λ 1 , λ 2 ) = x ∈ k q ∑ λ 1 ( x ) λ 2 ( 1 − x ) ,
J q , c ( v ) = J ( θ q v , ξ q c ) ∈ F . (44) J_{\mathfrak{q},c}(v)=J(\theta_{\mathfrak{q}}^v,\xi_{\mathfrak{q}}^c)\in F. \tag*{(44)} J q , c ( v ) = J ( θ q v , ξ q c ) ∈ F . (44)
Finally, Frob q \operatorname{Frob}_{\mathfrak{q}} Frob q denotes the q q q -power Frobenius endomorphism of the reduction of J J J . Our Frobenius convention is
Frob q on the Tate module acts as arithmetic Frobenius x ⟼ x q . (45) \operatorname{Frob}_{\mathfrak{q}}\text{ on the Tate module acts as arithmetic Frobenius }x\longmapsto x^q. \tag*{(45)} Frob q on the Tate module acts as arithmetic Frobenius x ⟼ x q . (45)
This convention determines the sign and character indices in the following theorem. Define the RM determinants B q ± ( v ) ∈ F B_q^{\pm}(v) \in F B q ± ( v ) ∈ F by
ι ( B q ± ( v ) ) = det ( Frob q ∣ M v ± ) . \iota(B_q^{\pm}(v))=\det\left(\operatorname{Frob}_q\mid M_v^{\pm}\right). ι ( B q ± ( v )) = det ( Frob q ∣ M v ± ) .
The Jacobi formulas below identify these determinants as algebraic numbers in F F F , independent of ℓ 0 \ell_0 ℓ 0 . Thus ℓ 0 \ell_0 ℓ 0 is not a parameter of B q ± ( v ) B_q^{\pm}(v) B q ± ( v ) or A q ( v ) A_q(v) A q ( v ) .
Theorem 5.1. The endomorphism G Δ G_\Delta G Δ gives real multiplication by K K K on J p r i m J^{\mathrm{prim}} J prim , and dim Q ℓ 0 M v + = dim Q ℓ 0 M v − = r \dim_{\mathbb{Q}_{\ell_0}}M_v^+=\dim_{\mathbb{Q}_{\ell_0}}M_v^-=r dim Q ℓ 0 M v + = dim Q ℓ 0 M v − = r . The Frobenius scalar on L v L_v L v is ι ( A q ( v ) ) \iota(A_q(v)) ι ( A q ( v )) , where
A q ( v ) = ∏ c ∈ ( Z Δ ) × J q , c ( v ) χ Δ ( c ) ∈ E × . (46) A_q(v)=\prod_{c\in(\mathbb{Z}_\Delta)^\times}J_{q,c}(v)^{\chi_\Delta(c)}\in E^\times. \tag*{(46)} A q ( v ) = c ∈ ( Z Δ ) × ∏ J q , c ( v ) χ Δ ( c ) ∈ E × . (46) The two determinants lie in E E E , and
A q ( v ) = B q + ( v ) B q − ( v ) . A_q(v)=\frac{B_q^+(v)}{B_q^-(v)}. A q ( v ) = B q − ( v ) B q + ( v ) . These coefficients satisfy
τ A q ( v ) = A q ( v ) − 1 , j A q ( v ) = A q ( v ) − 1 , δ u A q ( v ) = A q ( u v ) . (47) \tau A_q(v)=A_q(v)^{-1},\qquad jA_q(v)=A_q(v)^{-1},\qquad\delta_uA_q(v)=A_q(uv). \tag*{(47)} τ A q ( v ) = A q ( v ) − 1 , j A q ( v ) = A q ( v ) − 1 , δ u A q ( v ) = A q ( uv ) . (47) For every prime ideal p \mathfrak{p} p of E E E not over ℓ \ell ℓ , v p ( A q ( v ) ) = 0 v_{\mathfrak{p}}(A_q(v))=0 v p ( A q ( v )) = 0 .
The proof combines the classical Fermat decomposition with the Frobenius–Jacobi correspondence [10 Theorem A] [27 Theorem 1.2] . The next three lemmata fix the multiplicities, RM action and sign convention needed for the determinant quotient.
Lemma 5.2. The curve has genus
g ( C ) = ( n − 1 ) ( Δ − 1 ) 2 . (48) g(C)=\frac{(n-1)(\Delta-1)}{2}. \tag*{(48)} g ( C ) = 2 ( n − 1 ) ( Δ − 1 ) . (48) Moreover dim Q ℓ 0 V v , c = 1 \dim_{\mathbb{Q}_{\ell_0}}V_{v,c}=1 dim Q ℓ 0 V v , c = 1 if v , c v,c v , c are both nonzero, and it is zero otherwise.
Proof. Riemann–Hurwitz for the degree-n n n map to the Z \mathbb{Z} Z -line, ramified at the Δ \Delta Δ roots of unity and infinity, gives the genus. The Fermat character decomposition [21 Section 1] , or [27 Propositions 4.8 and 4.11] , assigns a one-dimensional space to each pair with v / n v/n v / n , c / Δ c/\Delta c /Δ , and v / n + c / Δ v/n+c/\Delta v / n + c /Δ nonintegral. Coprimality makes the last condition automatic when v , c ≠ 0 v,c\ne0 v , c = 0 . The two trivial-character parts vanish, equivalently
∑ b = 0 n − 1 S b = 0 , ∑ b = 0 Δ − 1 T b = 0. (49) \sum_{b=0}^{n-1}S^b=0,\qquad\sum_{b=0}^{\Delta-1}T^b=0. \tag*{(49)} b = 0 ∑ n − 1 S b = 0 , b = 0 ∑ Δ − 1 T b = 0. (49) because both complete deck quotients have genus zero. □
The RM structure and its eigenspaces
We write End K 0 ( A ) = End K ( A ) ⊗ Z Q \operatorname{End}_K^0(A)=\operatorname{End}_K(A)\otimes_{\mathbb{Z}}\mathbb{Q} End K 0 ( A ) = End K ( A ) ⊗ Z Q for endomorphisms up to isogeny.
Lemma 5.3. There is an injective homomorphism
r : K ↪ End K 0 ( J p r i m ) , Δ ⟼ G Δ ∣ J p r i m . (50) r:K\hookrightarrow\operatorname{End}_K^0(J^{\mathrm{prim}}),\qquad\sqrt{\Delta}\longmapsto G_\Delta|_{J^{\mathrm{prim}}}. \tag*{(50)} r : K ↪ End K 0 ( J prim ) , Δ ⟼ G Δ ∣ J prim . (50) For primitive c c c , its action on V v , c V_{v,c} V v , c is through the embedding of K K K that sends Δ \sqrt{\Delta} Δ to χ Δ ( c ) ι ( Δ ) \chi_\Delta(c)\iota(\sqrt{\Delta}) χ Δ ( c ) ι ( Δ ) .
Proof. The primitive quadratic Gauss sum for the chosen complex roots is
∑ b ∈ ( Z Δ ) × χ Δ ( b ) ζ Δ b = Δ . (51) \sum_{b\in(\mathbb{Z}_{\Delta})^\times}\chi_{\Delta}(b)\zeta_{\Delta}^{b}=\sqrt{\Delta}. \tag*{(51)} b ∈ ( Z Δ ) × ∑ χ Δ ( b ) ζ Δ b = Δ . (51)
This is the quadratic Gauss-sum evaluation [38 Corollary 4.6] ; in particular its square is Δ \Delta Δ , with the positive sign for Δ > 0 \Delta>0 Δ > 0 in the standard embedding. Replacing b b b by b c − 1 bc^{-1} b c − 1 gives
G Δ ∣ V v , c = χ Δ ( c ) ι ( Δ ) , c ∈ ( Z Δ ) × . (52) G_{\Delta}|_{V_{v,c}}=\chi_{\Delta}(c)\iota(\sqrt{\Delta}),\qquad c\in(\mathbb{Z}_{\Delta})^\times. \tag*{(52)} G Δ ∣ V v , c = χ Δ ( c ) ι ( Δ ) , c ∈ ( Z Δ ) × . (52)
The idempotent e Δ p r i m e_{\Delta}^{\mathrm{prim}} e Δ prim is defined over Q \mathbb{Q} Q , since every Galois automorphism permutes its primitive characters. On its image, T T T satisfies Φ Δ ( T ) = 0 \Phi_{\Delta}(T)=0 Φ Δ ( T ) = 0 , and the action of Q ( ζ Δ ) \mathbb{Q}(\zeta_{\Delta}) Q ( ζ Δ ) is faithful by Lemma 5.2. Equation (52) therefore gives G Δ 2 = Δ G_{\Delta}^{2}=\Delta G Δ 2 = Δ and an embedding of K K K . An automorphism sending ζ Δ \zeta_{\Delta} ζ Δ to ζ Δ b \zeta_{\Delta}^{b} ζ Δ b sends G Δ G_{\Delta} G Δ to χ Δ ( b ) G Δ \chi_{\Delta}(b)G_{\Delta} χ Δ ( b ) G Δ . Hence this endomorphism is fixed by Gal ( Q ‾ / K ) \operatorname{Gal}(\overline{\mathbb{Q}}/K) Gal ( Q / K ) , proving (50) . There are φ ( Δ ) / 2 \varphi(\Delta)/2 φ ( Δ ) /2 primitive indices of each sign, so
M v ± = ⨁ χ Δ ( c ) = ± 1 V v , c , dim M v ± = r . M_{v}^{\pm}=\bigoplus_{\chi_{\Delta}(c)=\pm1}V_{v,c},\qquad\dim M_{v}^{\pm}=r. M v ± = χ Δ ( c ) = ± 1 ⨁ V v , c , dim M v ± = r .
Their coefficient projectors are
e ± = 1 2 ( 1 ± G Δ ι ( Δ ) ) on M v . (53) e_{\pm}=\frac{1}{2}\left(1\pm\frac{G_{\Delta}}{\iota(\sqrt{\Delta})}\right)\quad\text{on }M_{v}. \tag*{(53)} e ± = 2 1 ( 1 ± ι ( Δ ) G Δ ) on M v . (53)
These are coefficient projectors for the two embeddings of K ⊗ Q ‾ ℓ 0 K\otimes\overline{\mathbb{Q}}_{\ell_{0}} K ⊗ Q ℓ 0 , not an asserted isogeny decomposition of J J J . For composite Δ \Delta Δ , the primitive projector remains essential: (49) removes only the trivial character.
Evaluation of Frobenius
Lemma 5.4. On the one-dimensional space V v , c V_{v,c} V v , c ,
Frob q ∣ V v , c = − ι ( J q , c ( v ) ) i d . (54) \operatorname{Frob}_{q}|_{V_{v,c}}=-\iota(J_{q,c}(v))\,\mathrm{id}. \tag*{(54)} Frob q ∣ V v , c = − ι ( J q , c ( v )) id . (54) Moreover
J q , c ( v ) J q , c ( v ) ȷ = q , ∣ σ J q , c ( v ) ∣ = q ( σ : F ↪ C ) . (55) J_{q,c}(v)J_{q,c}(v)^{\jmath}=q,\qquad|\sigma J_{q,c}(v)|=\sqrt{q}\quad(\sigma:F\hookrightarrow\mathbb{C}). \tag*{(55)} J q , c ( v ) J q , c ( v ) = q , ∣ σ J q , c ( v ) ∣ = q ( σ : F ↪ C ) . (55)
Proof. This is the classical Frobenius–Jacobi correspondence [10 Theorem A] , in the character-projector form of [27 Theorem 1.2 and Section 6] . On the fibre above x x x , Frobenius has deck character θ q ( x ) v ξ q ( 1 − x ) c \theta_q(x)^{v}\xi_q(1-x)^{c} θ q ( x ) v ξ q ( 1 − x ) c . Its degree-one trace is the negative of the sum; thus our unsigned Jacobi sum gives − ι ( J q , c ( v ) ) -\iota(J_{q,c}(v)) − ι ( J q , c ( v )) . This records the convention needed here. The norm and absolute values follow from J ( α , β ) J ( α , β ) = q J(\alpha,\beta)J(\alpha,\beta)=q J ( α , β ) J ( α , β ) = q when α , β , α β \alpha,\beta,\alpha\beta α , β , α β are nontrivial [5 Section 2.1] . □ \square □
Proof of Theorem 5.1. The lemmata give (46) : both determinants have sign ( − 1 ) r (-1)^{r} ( − 1 ) r , which cancels in their quotient. The group Gal ( F / E ) \operatorname{Gal}(F/E) Gal ( F / E ) preserves each sign of χ Δ \chi_{\Delta} χ Δ , whereas τ \tau τ interchanges the two signs; this proves membership in E E E and τ A = A − 1 \tau A=A^{-1} τ A = A − 1 . Equation (55) gives j A = A − 1 jA=A^{-1} j A = A − 1 , and coefficient conjugation gives δ u A ( v ) = A ( u v ) \delta_{u}A(v)=A(uv) δ u A ( v ) = A ( uv ) . Integrality and J ȷ J = q J^{\jmath}J=q J J = q restrict every Jacobi divisor, hence the quotient’s divisor, to primes above ℓ \ell ℓ .
The determinant as a quotient of norms
The automorphisms of F / E F/E F / E fix ζ n \zeta_n ζ n and send ζ Δ \zeta_{\Delta} ζ Δ to ζ Δ c \zeta_{\Delta}^{c} ζ Δ c for the indices c c c with χ Δ ( c ) = 1 \chi_{\Delta}(c)=1 χ Δ ( c ) = 1 . The indices of each RM sign occur in pairs c , − c c,-c c , − c , so r r r is even and the Jacobi signs multiply to one. Thus
B q + ( v ) = N F / E J q , 1 ( v ) = ∏ χ Δ ( c ) = 1 J q , c ( v ) , B q − ( v ) = τ ( B q + ( v ) ) , B_q^{+}(v)=N_{F/E}J_{q,1}(v)=\prod_{\chi_{\Delta}(c)=1}J_{q,c}(v),\qquad B_q^{-}(v)=\tau\left(B_q^{+}(v)\right), B q + ( v ) = N F / E J q , 1 ( v ) = χ Δ ( c ) = 1 ∏ J q , c ( v ) , B q − ( v ) = τ ( B q + ( v ) ) ,
A q ( v ) = B q + ( v ) B q − ( v ) . (56) A_q(v)=\frac{B_q^{+}(v)}{B_q^{-}(v)}. \tag*{(56)} A q ( v ) = B q − ( v ) B q + ( v ) . (56)
Thus the determinant quotient is a quotient of coefficient norms. For comparison with normalization by complex conjugation, B q ± ( v ) j ( B q ± ( v ) ) = q r B_q^{\pm}(v)j(B_q^{\pm}(v))=q^r B q ± ( v ) j ( B q ± ( v )) = q r gives
B q + ( v ) j ( B q − ( v ) ) j ( B q + ( v ) ) B q − ( v ) = A q ( v ) 2 , \frac{B_q^{+}(v)j(B_q^{-}(v))}{j(B_q^{+}(v))B_q^{-}(v)}=A_q(v)^2, j ( B q + ( v )) B q − ( v ) B q + ( v ) j ( B q − ( v )) = A q ( v ) 2 ,
[ A q ( v ) ] = [ B q + ( v ) j ( B q − ( v ) ) j ( B q + ( v ) ) B q − ( v ) ] ( n + 1 ) / 2 in E × / E × n . (57) [A_q(v)]=\left[\frac{B_q^{+}(v)j(B_q^{-}(v))}{j(B_q^{+}(v))B_q^{-}(v)}\right]^{(n+1)/2}\quad\text{in }E^{\times}/E^{\times n}. \tag*{(57)} [ A q ( v )] = [ j ( B q + ( v )) B q − ( v ) B q + ( v ) j ( B q − ( v )) ] ( n + 1 ) /2 in E × / E × n . (57)
Here ( n + 1 ) / 2 (n+1)/2 ( n + 1 ) /2 is the inverse of 2 modulo the odd integer n n n . This identity describes the extra quadratic anti-invariance of the ordinary determinant. Proposition 10.7 uses it to compare the ordinary and uniform RM constructions.
Ordinary RM divisors
We compute the divisors of the ordinary determinant coefficients and prove that they span the required anti-invariant part.
Keep n = p α n=p^{\alpha} n = p α , with p p p odd and ( p , Δ ) = 1 (p,\Delta)=1 ( p , Δ ) = 1 . Let Σ \Sigma Σ be a finite Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable set of prime ideals of E E E not over p Δ p\Delta p Δ , and let Σ F \Sigma_F Σ F be all primes of F F F above it. Recall the ordinary divisor lattice of § 3 :
Λ Σ = { Z ∈ Z [ Σ ] : τ Z = − Z , j Z = − Z } , \Lambda_{\Sigma}=\{Z\in\mathbb{Z}[\Sigma]:\tau Z=-Z,\ jZ=-Z\}, Λ Σ = { Z ∈ Z [ Σ ] : τ Z = − Z , j Z = − Z } ,
where Z [ Σ ] = ⨁ p ∈ Σ Z [ p ] \mathbb{Z}[\Sigma]=\bigoplus_{\mathfrak{p}\in\Sigma}\mathbb{Z}[\mathfrak{p}] Z [ Σ ] = ⨁ p ∈ Σ Z [ p ] and the map used in this section is
div E , Σ : R Σ × ⟶ Z [ Σ ] , x ⟼ ∑ p ∈ Σ v p ( x ) [ p ] . \operatorname{div}_{E,\Sigma}:R_{\Sigma}^{\times}\longrightarrow\mathbb{Z}[\Sigma],\qquad x\longmapsto\sum_{\mathfrak{p}\in\Sigma}v_{\mathfrak{p}}(x)[\mathfrak{p}]. div E , Σ : R Σ × ⟶ Z [ Σ ] , x ⟼ p ∈ Σ ∑ v p ( x ) [ p ] .
We abbreviate it to div E \operatorname{div}_E div E on supported elements. In particular its restriction U Σ → Λ Σ U_{\Sigma}\to\Lambda_{\Sigma} U Σ → Λ Σ is the map defined in § 3 .
Theorem 6.1. One has
span Q { div E A q ( v ) : q ∈ Σ F , v ∈ ( Z n ) × } = Λ Σ ⊗ Q . \operatorname{span}_{\mathbb{Q}}\{\operatorname{div}_E A_q(v):q\in\Sigma_F,\ v\in(\mathbb{Z}_n)^{\times}\}=\Lambda_{\Sigma}\otimes\mathbb{Q}. span Q { div E A q ( v ) : q ∈ Σ F , v ∈ ( Z n ) × } = Λ Σ ⊗ Q . There is no restriction on the class number of K K K .
The proof has three steps. Stickelberger’s formula gives the divisors of the Jacobi values. Fourier decomposition reduces their span to eigenvalues proportional to odd Dirichlet L L L -values at zero. Their nonvanishing gives every required component, including at nonsplit primes. This is the classical minus Stickelberger argument [35 ] [27 Section 8] , applied to our quadratic determinant. For the associated twisted Bernoulli determinant, see [14 Theorem 1] , with ( m , f , k , H , ψ ) = ( n , Δ , 1 , { 1 } , χ Δ ) (m,f,k,H,\psi)=(n,\Delta,1,\{1\},\chi_{\Delta}) ( m , f , k , H , ψ ) = ( n , Δ , 1 , { 1 } , χ Δ ) .
The norm and support identities are those of Theorem 5.1 . We state the classical valuation formula at arbitrary conductor for use also in the general case.
Lemma 6.2. Let M > 2 M>2 M > 2 , L = Q ( ζ M ) L=\mathbb{Q}(\zeta_M) L = Q ( ζ M ) , and let q \mathfrak{q} q be a prime of L L L over ℓ ∤ M \ell\nmid M ℓ ∤ M , and put k q = O L / q \mathfrak{k}_{\mathfrak{q}}=\mathcal{O}_L/\mathfrak{q} k q = O L / q , q 0 = ∣ k q ∣ q_0=\lvert\mathfrak{k}_{\mathfrak{q}}\rvert q 0 = ∣ k q ∣ . Choose g , s q g,s_{\mathfrak{q}} g , s q as in (43) , with m = M m=M m = M and q = q 0 q=q_0 q = q 0 . The inverse character, in the coordinates of (43) , is
λ : k q × ⟶ μ M ( L ) , λ ( g t ) = ζ M − s q t . \lambda:\mathfrak{k}_{\mathfrak{q}}^{\times}\longrightarrow\mu_M(L),\qquad\lambda(g^t)=\zeta_M^{-s_{\mathfrak{q}}t}. λ : k q × ⟶ μ M ( L ) , λ ( g t ) = ζ M − s q t . Character powers are formed on k q × \mathfrak{k}_{\mathfrak{q}}^{\times} k q × and then extended by zero. Write { x } = x − ⌊ x ⌋ \{x\}=x-\lfloor x\rfloor { x } = x − ⌊ x ⌋ and assume
r ≠ 0 , t ≠ 0 , r + t ≠ 0 in Z M . r\ne0,\qquad t\ne0,\qquad r+t\ne0\quad\text{in }\mathbb{Z}_M. r = 0 , t = 0 , r + t = 0 in Z M . Put
J q ( r , t ) = ∑ x ∈ k q λ r ( x ) λ t ( 1 − x ) , f r , t ( b ) = { r b M } + { t b M } − { ( r + t ) b M } . J_{\mathfrak{q}}(r,t)=\sum_{x\in\mathfrak{k}_{\mathfrak{q}}}\lambda^r(x)\lambda^t(1-x),\qquad f_{r,t}(b)=\left\{\frac{rb}{M}\right\}+\left\{\frac{tb}{M}\right\}-\left\{\frac{(r+t)b}{M}\right\}. J q ( r , t ) = x ∈ k q ∑ λ r ( x ) λ t ( 1 − x ) , f r , t ( b ) = { M r b } + { M t b } − { M ( r + t ) b } . For σ b ( ζ M ) = ζ M b \sigma_b(\zeta_M)=\zeta_M^b σ b ( ζ M ) = ζ M b , the factorization of its principal ideal, or equivalently its divisor, is
div L J q ( r , t ) = ∑ b ∈ ( Z M ) × f r , t ( b ) [ σ b − 1 q ] . \operatorname{div}_L J_{\mathfrak{q}}(r,t)=\sum_{b\in(\mathbb{Z}_M)^\times}f_{r,t}(b)[\sigma_b^{-1}\mathfrak{q}]. div L J q ( r , t ) = b ∈ ( Z M ) × ∑ f r , t ( b ) [ σ b − 1 q ] . If h h h is the order of ℓ \ell ℓ modulo M M M , the valuation at σ b − 1 q \sigma_b^{-1}\mathfrak{q} σ b − 1 q is consequently ∑ i = 0 h − 1 f r , t ( b ℓ i ) \sum_{i=0}^{h-1}f_{r,t}(b\ell^i) ∑ i = 0 h − 1 f r , t ( b ℓ i ) . Proof. Apply Stickelberger’s Jacobi factorization [5 Theorem 11.2.3] . For positive residue exponent d d d , the convention in [27 Section 8, equation (8.5)] is ∑ b { − d b / M } σ b − 1 \sum_b\{-db/M\}\sigma_b^{-1} ∑ b { − d b / M } σ b − 1 ; substitution of − r , − t -r,-t − r , − t gives the displayed coefficients. Since the decomposition group is ⟨ σ ℓ ⟩ \langle\sigma_\ell\rangle ⟨ σ ℓ ⟩ , the indices b ℓ i b\ell^i b ℓ i represent the same prime and their coefficients add. This includes nonsplit primes. □ \square □
The divisor operator and its Fourier eigenvalues
Let G = Gal ( F / Q ) G=\operatorname{Gal}(F/\mathbb{Q}) G = Gal ( F / Q ) , with σ b ( ζ m ) = ζ m b \sigma_b(\zeta_m)=\zeta_m^b σ b ( ζ m ) = ζ m b . Define
c Δ , n ( b ) = { − Δ b m } + { − n b m } − { − ( Δ + n ) b m } , C Δ , n = ∑ b ∈ ( Z m ) × c Δ , n ( b ) σ b − 1 . (58) c_{\Delta,n}(b)=\left\{\frac{-\Delta b}{m}\right\}+\left\{\frac{-nb}{m}\right\}-\left\{\frac{-(\Delta+n)b}{m}\right\},\qquad C_{\Delta,n}=\sum_{b\in(\mathbb{Z}_m)^\times}c_{\Delta,n}(b)\sigma_b^{-1}. \tag*{(58)} c Δ , n ( b ) = { m − Δ b } + { m − nb } − { m − ( Δ + n ) b } , C Δ , n = b ∈ ( Z m ) × ∑ c Δ , n ( b ) σ b − 1 . (58)
The coefficients are zero or one. In Lemma 6.2, the characters of (44) have exponents − Δ , − n -\Delta,-n − Δ , − n relative to the inverse character of order m m m . Hence
div F J q , 1 ( 1 ) = C Δ , n [ q ] . (59) \operatorname{div}_F J_{\mathfrak{q},1}(1)=C_{\Delta,n}[\mathfrak{q}]. \tag*{(59)} div F J q , 1 ( 1 ) = C Δ , n [ q ] . (59)
This explains both the minus signs in the fractional parts and the inverse Galois indices: b b b records the conjugate prime σ b − 1 q \sigma_b^{-1}\mathfrak{q} σ b − 1 q , with repeated primes summed as in the lemma.
Let H ~ = Gal ( F / E ) \widetilde{H}=\operatorname{Gal}(F/E) H = Gal ( F / E ) and N H ~ = ∑ h ∈ H ~ h N_{\widetilde{H}}=\sum_{h\in\widetilde{H}}h N H = ∑ h ∈ H h . Choose υ ∈ G \upsilon\in G υ ∈ G fixing ζ n \zeta_n ζ n and acting nontrivially on K K K . The norm expression (56) gives
div F A q ( 1 ) = ( 1 − υ ) N H ~ C Δ , n [ q ] . (60) \operatorname{div}_F A_{\mathfrak{q}}(1)=(1-\upsilon)N_{\widetilde{H}}C_{\Delta,n}[\mathfrak{q}]. \tag*{(60)} div F A q ( 1 ) = ( 1 − υ ) N H C Δ , n [ q ] . (60)
For a number field E E E , write Div ( E ) = ⨁ p Z [ p ] \operatorname{Div}(E)=\bigoplus_{\mathfrak{p}}\mathbb{Z}[\mathfrak{p}] Div ( E ) = ⨁ p Z [ p ] , over its nonzero prime ideals. For a finite extension F / E F/E F / E , the standard extension of ideals, written additively on divisors, is the map we denote by
ext F / E : Div ( E ) ⟶ Div ( F ) , [ p ] ⟼ ∑ q ∣ p e ( q / p ) [ q ] . \operatorname{ext}_{F/E}:\operatorname{Div}(E)\longrightarrow\operatorname{Div}(F),\qquad[\mathfrak{p}]\longmapsto\sum_{\mathfrak{q}\mid\mathfrak{p}}e(\mathfrak{q}/\mathfrak{p})[\mathfrak{q}]. ext F / E : Div ( E ) ⟶ Div ( F ) , [ p ] ⟼ q ∣ p ∑ e ( q / p ) [ q ] .
Here [ p ] [\mathfrak{p}] [ p ] is a prime divisor, not an ideal class, and e ( q / p ) e(\mathfrak{q}/\mathfrak{p}) e ( q / p ) is the ramification index. This is the additive form of the ideal factorization p O F = ∏ q ∣ p q e ( q / p ) \mathfrak{p}\mathcal{O}_F=\prod_{\mathfrak{q}\mid\mathfrak{p}}\mathfrak{q}^{e(\mathfrak{q}/\mathfrak{p})} p O F = ∏ q ∣ p q e ( q / p ) [24 Chapter I, Section 8] . It satisfies
ext F / E ( div E ( x ) ) = div F ( x ) ( x ∈ E × ) , \operatorname{ext}_{F/E}(\operatorname{div}_E(x))=\operatorname{div}_F(x)\qquad(x\in E^\times), ext F / E ( div E ( x )) = div F ( x ) ( x ∈ E × ) ,
so it compares the divisor of A q ( 1 ) ∈ E × A_{\mathfrak{q}}(1)\in E^\times A q ( 1 ) ∈ E × with its computation in F F F above. Here all primes in Σ \Sigma Σ are unramified in F / E F/E F / E , so ext F / E [ p ] = ∑ q ∣ p [ q ] \operatorname{ext}_{F/E}[\mathfrak{p}]=\sum_{\mathfrak{q}\mid\mathfrak{p}}[\mathfrak{q}] ext F / E [ p ] = ∑ q ∣ p [ q ] ; after extending scalars to Q \mathbb{Q} Q , this identifies Q [ Σ ] \mathbb{Q}[\Sigma] Q [ Σ ] with Q [ Σ F ] H ~ \mathbb{Q}[\Sigma_F]^{\widetilde{H}} Q [ Σ F ] H . On the subspace where υ \upsilon υ and j j j both act by − 1 -1 − 1 , the factor ( 1 − υ ) N H ~ (1-\upsilon)N_{\widetilde{H}} ( 1 − υ ) N H is multiplication by 2 ∣ H ~ ∣ = φ ( Δ ) 2|\widetilde{H}|=\varphi(\Delta) 2∣ H ∣ = φ ( Δ ) .
Lemma 6.3. Let ψ \psi ψ be an odd character modulo n n n , and regard ϑ = χ Δ ψ \vartheta=\chi_\Delta\psi ϑ = χ Δ ψ as a character of G G G . The eigenvalue of C Δ , n C_{\Delta,n} C Δ , n on the ϑ \vartheta ϑ -eigenspace is
∑ b c Δ , n ( b ) ϑ ( b ) − 1 = ϑ ( − ( Δ + n ) ) L ( 0 , ϑ − 1 ) . (61) \sum_b c_{\Delta,n}(b)\vartheta(b)^{-1}=\vartheta(-(\Delta+n))L(0,\vartheta^{-1}). \tag*{(61)} b ∑ c Δ , n ( b ) ϑ ( b ) − 1 = ϑ ( − ( Δ + n )) L ( 0 , ϑ − 1 ) . (61) The eigenvalue of ( 1 − υ ) N H ~ C Δ , n (1-\upsilon)N_{\widetilde{H}}C_{\Delta,n} ( 1 − υ ) N H C Δ , n on this space is
φ ( Δ ) ϑ ( − ( Δ + n ) ) L ( 0 , ϑ − 1 ) . (62) \varphi(\Delta)\vartheta(-(\Delta+n))L(0,\vartheta^{-1}). \tag*{(62)} φ ( Δ ) ϑ ( − ( Δ + n )) L ( 0 , ϑ − 1 ) . (62)
Proof. The first two terms vanish by summing respectively over χ Δ \chi_\Delta χ Δ and the nontrivial odd character ψ \psi ψ . For any unit d d d mod m m m , substitute y = − d b y=-db y = − d b to obtain
∑ b ∈ ( Z m ) × { − d b m } ϑ ( b ) − 1 = ϑ ( − d ) ∑ y = 1 m y m ϑ ( y ) − 1 = − ϑ ( − d ) L ( 0 , ϑ − 1 ) . \sum_{b\in(\mathbb{Z}_m)^\times}\left\{\frac{-db}{m}\right\}\vartheta(b)^{-1}
=\vartheta(-d)\sum_{y=1}^{m}\frac{y}{m}\vartheta(y)^{-1}
=-\vartheta(-d)L(0,\vartheta^{-1}). b ∈ ( Z m ) × ∑ { m − d b } ϑ ( b ) − 1 = ϑ ( − d ) y = 1 ∑ m m y ϑ ( y ) − 1 = − ϑ ( − d ) L ( 0 , ϑ − 1 ) .
The Bernoulli identity applies also to imprimitive nonprincipal characters [2 Theorem 12.20] . Take d = Δ + n d=\Delta+n d = Δ + n , which is prime to m m m ; the additional factor is ( 1 − υ ) N H ~ = φ ( Δ ) (1-\upsilon)N_{\widetilde{H}}=\varphi(\Delta) ( 1 − υ ) N H = φ ( Δ ) on the required eigenspace. □
Lemma 6.4. If λ \lambda λ is a primitive odd Dirichlet character, then L ( 0 , λ ) ≠ 0 L(0,\lambda)\ne0 L ( 0 , λ ) = 0 . If ψ \psi ψ is odd modulo n = p a n=p^a n = p a and ( p , Δ ) = 1 (p,\Delta)=1 ( p , Δ ) = 1 , the same holds for λ = χ Δ ψ \lambda=\chi_\Delta\psi λ = χ Δ ψ and its inverse, whether or not they are primitive modulo Δ n \Delta n Δ n .
Proof. For a primitive odd character of conductor f f f , let g ( λ ) = ∑ b m o d f λ ( b ) e 2 π i b / f g(\lambda)=\sum_{b\mathrel{\mathrm{mod}} f}\lambda(b)e^{2\pi ib/f} g ( λ ) = ∑ b mod f λ ( b ) e 2 π ib / f . The functional equation and Dirichlet’s nonvanishing theorem give
L ( 0 , λ ) = g ( λ ) i π L ( 1 , λ ‾ ) ≠ 0 ; (63) L(0,\lambda)=\frac{g(\lambda)}{i\pi}L(1,\overline{\lambda})\ne0; \tag*{(63)} L ( 0 , λ ) = iπ g ( λ ) L ( 1 , λ ) = 0 ; (63)
see [26 equations (25.15.5) and (25.15.9)] . The primitive character inducing ψ \psi ψ has conductor p b p^b p b , 1 ≤ b ≤ a 1\le b\le a 1 ≤ b ≤ a , so the primitive character inducing χ Δ ψ \chi_\Delta\psi χ Δ ψ has conductor p b Δ p^b\Delta p b Δ . Enlarging to p a Δ p^a\Delta p a Δ introduces no new prime divisor and therefore no Euler factor [26 equation (25.15.4)] . The primitive nonvanishing consequently applies unchanged. □
Proof of Theorem 6.1. Decompose Q [ Σ F ] \mathbb{Q}[\Sigma_F] Q [ Σ F ] into its prime orbits. The orbit of q q q is the permutation representation Q [ G / G q ] \mathbb{Q}[G/G_q] Q [ G / G q ] , where G q G_q G q is the decomposition group. Since G G G is abelian, its complex characters are exactly those trivial on G q G_q G q , each with multiplicity one. Requiring H ~ \widetilde{H} H to act trivially and υ , j \upsilon,j υ , j to act by − 1 -1 − 1 selects exactly the characters χ Δ ψ \chi_\Delta\psi χ Δ ψ with ψ \psi ψ odd that occur in this permutation representation. Lemmata 6.3 and 6.4 show that the operator in (60) is invertible on this subspace. Its image on the full permutation representation lies in that subspace by the two anti-invariance identities already proved. Since its restriction there is invertible, its image is exactly the required subspace, and is spanned by its values on the prime basis. Those values are precisely div F A q ( 1 ) \operatorname{div}_F A_q(1) div F A q ( 1 ) . Taking the direct sum over the prime orbits and descending through ext F / E \operatorname{ext}_{F/E} ext F / E proves the claim. In particular the assertion holds at nonsplit rational primes; their decomposition groups merely remove some characters from the representation. □
Corollary 6.5. For any finite Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable set Σ \Sigma Σ , put Σ ∘ = { p ∈ Σ : p ∤ p Δ } \Sigma^\circ=\{\mathfrak{p}\in\Sigma:\mathfrak{p}\nmid p\Delta\} Σ ∘ = { p ∈ Σ : p ∤ p Δ } . Then Λ Σ = Λ Σ ∘ \Lambda_\Sigma=\Lambda_{\Sigma^\circ} Λ Σ = Λ Σ ∘ , so the determinant values above Σ ∘ \Sigma^\circ Σ ∘ span Λ Σ ⊗ Q \Lambda_\Sigma\otimes\mathbb{Q} Λ Σ ⊗ Q .
Proof. As in Theorem 4.1 , j j j fixes every prime above p p p and τ \tau τ fixes every prime above a divisor of Δ \Delta Δ . The corresponding anti-invariance forces its coefficient to vanish. Thus deleting these primes leaves Λ Σ \Lambda_{\Sigma} Λ Σ unchanged.
Theorem 6.1 gives rational spanning. § 7 defines the generated subgroup and obtains a basis modulo n n n th powers by internal p p p -saturation.
A basis of the ordinary Kummer module
A subgroup of finite index need not supply every Kummer class: ⟨ u p ⟩ \langle u^p\rangle ⟨ u p ⟩ has trivial image in ⟨ u ⟩ / ⟨ u ⟩ p \langle u\rangle/\langle u\rangle^p ⟨ u ⟩ / ⟨ u ⟩ p . The standard remedy is p p p -saturation [6 Section 4.3, Lemma 4.3, Algorithm 4.9 and Theorem 4.10] . All roots extracted here already lie in E E E ; the extension α n = β \alpha^n=\beta α n = β comes later.
Proposition 7.1. Let U U U be a finitely generated abelian group, written multiplicatively, and let Γ ⊂ U \Gamma\subset U Γ ⊂ U have finite index I I I . Fix a prime p p p . Throughout, the notation Γ s a t \Gamma^{\mathrm{sat}} Γ sat means saturation at this fixed p p p , in the specified ambient group U U U ; put
Γ s a t = { x ∈ U : x p b ∈ Γ for some b ≥ 0 } . (64) \Gamma^{\mathrm{sat}}=\{x\in U:x^{p^b}\in\Gamma\text{ for some }b\geq0\}. \tag*{(64)} Γ sat = { x ∈ U : x p b ∈ Γ for some b ≥ 0 } . (64) Then
Γ s a t / Γ = ( U / Γ ) [ p ∞ ] , p ∤ [ U : Γ s a t ] , [ Γ s a t : Γ ] = p v p ( I ) . \Gamma^{\mathrm{sat}}/\Gamma=(U/\Gamma)[p^\infty],\qquad p\nmid[U:\Gamma^{\mathrm{sat}}],\qquad[\Gamma^{\mathrm{sat}}:\Gamma]=p^{v_p(I)}. Γ sat /Γ = ( U /Γ ) [ p ∞ ] , p ∤ [ U : Γ sat ] , [ Γ sat : Γ ] = p v p ( I ) . For every a ≥ 1 a\geq1 a ≥ 1 , inclusion induces
Γ s a t / ( Γ s a t ) p a → ∼ U / U p a . \Gamma^{\mathrm{sat}}/(\Gamma^{\mathrm{sat}})^{p^a}\xrightarrow{\ \sim\ }U/U^{p^a}. Γ sat / ( Γ sat ) p a ∼ U / U p a . Starting from Γ \Gamma Γ , repeatedly adjoining an element x ∈ U x\in U x ∈ U with x p x^p x p in the current subgroup but x x x outside it reaches Γ s a t \Gamma^{\mathrm{sat}} Γ sat after v p ( I ) v_p(I) v p ( I ) such enlargements.
Proof. The first quotient is the p p p -primary subgroup of U / Γ U/\Gamma U /Γ . Multiplication by p a p^a p a is bijective on the remaining quotient, giving the isomorphism modulo powers. Each enlargement has index p p p ; a nontrivial p p p -primary quotient supplies another element of order p p p . This is the standard finite-index argument [6 Section 4.3, Lemma 4.3] .
The ordinary application. Use the ordinary data of § 3 and § 6 , with Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma})=0 Cl ( R Σ ) = 0 . Enumerate Σ F × Z n × \Sigma_F\times\mathbb{Z}_n^{\times} Σ F × Z n × as ( ( q i , v i ) ) i ((q_i,v_i))_i (( q i , v i ) ) i and define
A i = A q i ( v i ) = ∏ c ∈ Z Δ × J q i , c ( v i ) χ Δ ( c ) ∈ U Σ . (65) A_i=A_{q_i}(v_i)=\prod_{c\in\mathbb{Z}_{\Delta}^{\times}}J_{q_i,c}(v_i)^{\chi_{\Delta}(c)}\in U_{\Sigma}. \tag*{(65)} A i = A q i ( v i ) = c ∈ Z Δ × ∏ J q i , c ( v i ) χ Δ ( c ) ∈ U Σ . (65)
These are the Jacobi determinant quotients of Theorem 5.1 , including their coefficient conjugates. Set
Γ = ⟨ T Σ , A i (all i ) ⟩ ⊂ U Σ , \Gamma=\langle T_{\Sigma},A_i\text{ (all }i)\rangle\subset U_{\Sigma}, Γ = ⟨ T Σ , A i (all i )⟩ ⊂ U Σ ,
L Γ = div E ( Γ ) = ∑ i Z div E ( A i ) ⊆ L Σ ⊆ Λ Σ , L_{\Gamma}=\operatorname{div}_E(\Gamma)=\sum_i\mathbb{Z}\operatorname{div}_E(A_i)\subseteq L_{\Sigma}\subseteq\Lambda_{\Sigma}, L Γ = div E ( Γ ) = i ∑ Z div E ( A i ) ⊆ L Σ ⊆ Λ Σ ,
B = [ Λ Σ : L Γ ] , r = rank Λ Σ . B=[\Lambda_{\Sigma}:L_{\Gamma}],\qquad r=\operatorname{rank}\Lambda_{\Sigma}. B = [ Λ Σ : L Γ ] , r = rank Λ Σ .
Theorem 6.1 makes L Γ L_{\Gamma} L Γ a full-rank lattice. Smith normal form of its valuation matrix computes the finite index B B B . Since ker ( div E ∣ U Σ ) = T Σ ⊂ Γ \ker(\operatorname{div}_E\mid U_{\Sigma})=T_{\Sigma}\subset\Gamma ker ( div E ∣ U Σ ) = T Σ ⊂ Γ ,
U Σ / Γ ≃ L Σ / L Γ , [ U Σ : Γ ] ∣ B . (66) U_{\Sigma}/\Gamma\simeq L_{\Sigma}/L_{\Gamma},\qquad[U_{\Sigma}:\Gamma]\mid B. \tag*{(66)} U Σ /Γ ≃ L Σ / L Γ , [ U Σ : Γ ] ∣ B . (66)
Corollary 7.2. The class map induces
Γ s a t / ( Γ s a t ) n → ∼ U Σ / U Σ n → ∼ W Σ . (67) \Gamma^{\mathrm{sat}}/(\Gamma^{\mathrm{sat}})^n\xrightarrow{\ \sim\ }U_{\Sigma}/U_{\Sigma}^n\xrightarrow{\ \sim\ }W_{\Sigma}. \tag*{(67)} Γ sat / ( Γ sat ) n ∼ U Σ / U Σ n ∼ W Σ . (67)
If z 1 , … , z r z_1,\ldots,z_r z 1 , … , z r lift a Z \mathbb{Z} Z -basis of Γ s a t / T Σ \Gamma^{\mathrm{sat}}/T_\Sigma Γ sat / T Σ , then
( Z n ) r → ∼ W Σ , ( e j ) j ⟼ [ ∏ j z j e j ] . (\mathbb{Z}_n)^r \xrightarrow{\sim} W_\Sigma,\qquad(e_j)_j \longmapsto\left[\prod_j z_j^{e_j}\right]. ( Z n ) r ∼ W Σ , ( e j ) j ⟼ [ j ∏ z j e j ] .
Proof. Apply Propositions 7.1 and 3.6. The torsion contributes nothing because p ∤ ∣ T Σ ∣ p \nmid|T_\Sigma| p ∤ ∣ T Σ ∣ . □ \square □
Corollary 7.3.
Γ s a t = Γ ⟺ p ∤ [ U Σ : Γ ] . \Gamma^{\mathrm{sat}}=\Gamma\quad\Longleftrightarrow\quad p\nmid[U_\Sigma:\Gamma]. Γ sat = Γ ⟺ p ∤ [ U Σ : Γ ] . In particular, p ∤ B p\nmid B p ∤ B suffices. In this case a Kummer basis can be represented by products of the A i A_i A i .
Proof. Use Proposition 7.1 and [ U Σ : Γ ] ∣ B [U_\Sigma:\Gamma]\mid B [ U Σ : Γ ] ∣ B . □ \square □
Sections 16.1 and 16.2 give examples with B = 1 B=1 B = 1 and B = 2 B=2 B = 2 , respectively.
Computing the basis representatives. To compute the z j z_j z j in Corollary 7.2, start with Γ ′ = Γ \Gamma'=\Gamma Γ ′ = Γ and test representatives γ ∈ Γ ′ / Γ ′ p \gamma\in\Gamma'/\Gamma'^p γ ∈ Γ ′ / Γ ′ p by factoring X p − γ X^p-\gamma X p − γ over E E E . If b ∈ E b\in E b ∈ E is a root, put
ξ = b τ ( b ) ∈ μ p , c = b ξ ( p − 1 ) / 2 . \xi=b\tau(b)\in\mu_p,\qquad c=b\xi^{(p-1)/2}. ξ = b τ ( b ) ∈ μ p , c = b ξ ( p − 1 ) /2 .
Then c p = γ c^p=\gamma c p = γ , c τ ( c ) = 1 c\tau(c)=1 c τ ( c ) = 1 and c j ( c ) = 1 cj(c)=1 c j ( c ) = 1 ; the first norm identity follows because τ \tau τ fixes μ p \mu_p μ p , and the second because c j ( c ) cj(c) c j ( c ) is positive and has p p p th power one. Valuations put c c c in U Σ U_\Sigma U Σ . It is the unique root there, since U Σ ∩ μ p = 1 U_\Sigma\cap\mu_p=1 U Σ ∩ μ p = 1 .
Adjoin c c c if its divisor is outside L Γ ′ L_{\Gamma'} L Γ ′ . Membership is tested on divisors because T Σ ⊂ Γ ′ T_\Sigma\subset\Gamma' T Σ ⊂ Γ ′ . The tests are independent of the chosen representatives of Γ ′ / Γ ′ p \Gamma'/\Gamma'^p Γ ′ / Γ ′ p and find every enlargement. Proposition 7.1 bounds their number by v p ( B ) v_p(B) v p ( B ) . Integer basis changes, together with the root equations, give
z j p b j = ξ j ∏ i A i m j i , b j ≥ 0 , ξ j ∈ T Σ , m j i ∈ Z . (68) z_j^{p^{b_j}}=\xi_j\prod_i A_i^{m_{ji}},\qquad b_j\ge0,\quad\xi_j\in T_\Sigma,\quad m_{ji}\in\mathbb{Z}. \tag*{(68)} z j p b j = ξ j i ∏ A i m j i , b j ≥ 0 , ξ j ∈ T Σ , m j i ∈ Z . (68)
Polynomial factorization and integer basis changes are standard [7 Chapters 3–6] .
Selection by reciprocity in the ordinary case
Selection by Artin reciprocity determines the Kummer coordinates of the target from its prescribed Frobenius action. This is the classical Artin map method [15 Section 3] , [8 Section 5.4] , [16 Section 4.2] . We use separating primes rather than compute another ray class group over E E E ; compare [16 Remark 6] . Target membership from § 4 and completeness of the Kummer basis are both essential: together they ensure that these tests determine the prescribed class.
Let n = p a n=p^a n = p a , p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ , and E = K ( ζ n ) E=K(\zeta_n) E = K ( ζ n ) . Inflate the character to m = lcm ( f , τ K f ) m=\operatorname{lcm}(f,\tau_Kf) m = lcm ( f , τ K f ) and assume χ τ = − χ \chi^\tau=-\chi χ τ = − χ ; Lemma 2.6 handles its invariant part separately. Choose the ordinary support Σ \Sigma Σ as in Theorem 4.1, put r = rank Λ Σ r=\operatorname{rank}\Lambda_\Sigma r = rank Λ Σ , and choose representatives z 1 , … , z r z_1,\ldots,z_r z 1 , … , z r from Corollary 7.2. The basis and target membership mean
( Z n ) r → ∼ W Σ , ( e j ) j ⟼ ∏ j = 1 r [ z j ] e j , ν − 1 ( χ ∣ G E ) ∈ W Σ . (\mathbb{Z}_n)^r\xrightarrow{\sim}W_\Sigma,\qquad(e_j)_j\longmapsto\prod_{j=1}^r[z_j]^{e_j},
\qquad
\nu^{-1}(\chi|_{G_E})\in W_\Sigma. ( Z n ) r ∼ W Σ , ( e j ) j ⟼ j = 1 ∏ r [ z j ] e j , ν − 1 ( χ ∣ G E ) ∈ W Σ .
Here ν ( [ x ] ) = ν x \nu([x])=\nu_x ν ([ x ]) = ν x is the Kummer isomorphism (2.6). Thus the unknown is the coordinate vector of a specified class in a known finite module. For N = E ( z 1 n , … , z r n ) N=E(\sqrt[n]{z_1},\ldots,\sqrt[n]{z_r}) N = E ( n z 1 , … , n z r ) , the perfect Kummer pairing is
⟨ , ⟩ : Gal ( N / E ) × W Σ ⟶ μ n , ( g , [ x ] ) ⟼ g ( α x ) α x . (69) \langle\ ,\ \rangle: \operatorname{Gal}(N/E) \times W_{\Sigma} \longrightarrow\mu_n,
\qquad
(g,[x]) \longmapsto\frac{g(\alpha_x)}{\alpha_x}. \tag*{(69)} ⟨ , ⟩ : Gal ( N / E ) × W Σ ⟶ μ n , ( g , [ x ]) ⟼ α x g ( α x ) . (69)
Here α x ∈ N \alpha_x \in N α x ∈ N satisfies α x n = x \alpha_x^n=x α x n = x ; the pairing is independent of the chosen root and representative. The target is specified by
⟨ Frob q , ν − 1 ( χ ∣ G E ) ⟩ = ζ n χ ( [ N E / K q ] m ) \langle\operatorname{Frob}_{\mathfrak q},\nu^{-1}(\chi|_{G_E})\rangle
= \zeta_n^{\chi([\mathcal{N}_{E/K}\mathfrak q]_m)} ⟨ Frob q , ν − 1 ( χ ∣ G E )⟩ = ζ n χ ([ N E / K q ] m )
at primes where the residue symbols are defined. Call a prime of E E E admissible if it lies outside
T = Σ ∪ { p : p ∣ n m O E } . \mathcal{T}=\Sigma\cup\{\mathfrak p:\mathfrak p\mid n m\mathcal{O}_E\}. T = Σ ∪ { p : p ∣ nm O E } .
For a prime q ∉ T \mathfrak q\notin\mathcal{T} q ∈ / T and x ∈ R Σ × x\in R_{\Sigma}^{\times} x ∈ R Σ × , write q = ∣ O E / q ∣ q=|\mathcal{O}_E/\mathfrak q| q = ∣ O E / q ∣ and define the additive residue symbol r q ( x ) ∈ Z n r_{\mathfrak q}(x)\in\mathbb{Z}_n r q ( x ) ∈ Z n by
x ( q − 1 ) / n ≡ ζ n r q ( x ) ( m o d q ) . x^{(q-1)/n}\equiv\zeta_n^{r_{\mathfrak q}(x)}\pmod{\mathfrak q}. x ( q − 1 ) / n ≡ ζ n r q ( x ) ( mod q ) .
Reduction is injective on μ n \mu_n μ n , so n ∣ q − 1 n\mid q-1 n ∣ q − 1 and this exponent is unique. The homomorphism r q : R Σ × → Z n r_{\mathfrak q}:R_{\Sigma}^{\times}\to\mathbb{Z}_n r q : R Σ × → Z n depends only on the Kummer class whenever that class has a supported unit representative. In particular it defines r q : W Σ → Z n r_{\mathfrak q}:W_{\Sigma}\to\mathbb{Z}_n r q : W Σ → Z n . It evaluates the Kummer character at arithmetic Frobenius, by Lemma 2.3 .
Lemma 8.1. Let n = p a n=p^a n = p a for any prime p p p , and let E E E be any number field containing a fixed primitive n n n th root of unity ζ n \zeta_n ζ n . Let V ⊂ E × / E × n V\subset E^{\times}/E^{\times n} V ⊂ E × / E × n be free of rank d d d over Z n \mathbb{Z}_n Z n , with specified basis [ z 1 ] , … , [ z d ] [z_1],\ldots,[z_d] [ z 1 ] , … , [ z d ] represented by supported units. Outside any prescribed finite set containing the primes above n n n and the supports of the z j z_j z j , there are primes q 1 , … , q d \mathfrak q_1,\ldots,\mathfrak q_d q 1 , … , q d such that
V → ∼ ( Z n ) d , [ x ] ⟼ ( r q 1 ( x ) , … , r q d ( x ) ) . V\xrightarrow{\sim}(\mathbb{Z}_n)^d,
\qquad
[x]\longmapsto\bigl(r_{\mathfrak q_1}(x),\ldots,r_{\mathfrak q_d}(x)\bigr). V ∼ ( Z n ) d , [ x ] ⟼ ( r q 1 ( x ) , … , r q d ( x ) ) . They can be found by enumerating primes and keeping a residue row whenever it increases the rank modulo p p p .
Proof. The Kummer pairing identifies
Gal ( E ( z 1 n , … , z d n ) / E ) ≃ Hom ( V , μ n ) ≃ ( Z n ) d . \operatorname{Gal}\left(E\left(\sqrt[n]{z_1},\ldots,\sqrt[n]{z_d}\right)/E\right)
\simeq\operatorname{Hom}(V,\mu_n)\simeq(\mathbb{Z}_n)^d. Gal ( E ( n z 1 , … , n z d ) / E ) ≃ Hom ( V , μ n ) ≃ ( Z n ) d .
Chebotarev supplies every Frobenius vector outside the excluded set [24 Chapter VII, Theorem 13.4] . Consequently the residue rows span modulo p p p , and the procedure reaches rank d d d . A square matrix invertible modulo p p p is invertible modulo p a p^a p a . For d = 0 d=0 d = 0 the empty matrix is the unique isomorphism of the zero module. □
These auxiliary Kummer extensions need not be constructed: all rows are computed in finite residue fields. The lemma gives termination, not a bound asserting that the required primes are small.
Definition 8.2. Enumerate admissible primes of E E E . After examining s s s primes q 1 , … , q s \mathfrak q_1,\ldots,\mathfrak q_s q 1 , … , q s , the residue-testing map and its matrix in the basis [ z 1 ] , … , [ z r ] [z_1],\ldots,[z_r] [ z 1 ] , … , [ z r ] are
W Σ ⟶ ( Z n ) s , [ x ] ⟼ ( r q k ( x ) ) k = 1 s , W_{\Sigma}\longrightarrow(\mathbb{Z}_n)^s,
\qquad
[x]\longmapsto\bigl(r_{\mathfrak q_k}(x)\bigr)_{k=1}^s, W Σ ⟶ ( Z n ) s , [ x ] ⟼ ( r q k ( x ) ) k = 1 s , R s : ( Z n ) r ⟶ ( Z n ) s , R s = ( r q k ( z j ) ) 1 ≤ k ≤ s 1 ≤ j ≤ r . R_s:(\mathbb{Z}_n)^r\longrightarrow(\mathbb{Z}_n)^s,
\qquad
R_s=\bigl(r_{\mathfrak q_k}(z_j)\bigr)_{\substack{1\le k\le s\\1\le j\le r}}. R s : ( Z n ) r ⟶ ( Z n ) s , R s = ( r q k ( z j ) ) 1 ≤ k ≤ s 1 ≤ j ≤ r . Continue until R s R_s R s mod p p p has rank r r r , and choose row indices i 1 , … , i r i_1,\ldots,i_r i 1 , … , i r independent modulo $p. Lemma 8.1 guarantees termination. Define the resulting square matrix and target vector by
C = ( r q i ℓ ( z j ) ) 1 ≤ ℓ , j ≤ r ∈ GL r ( Z n ) , b ℓ = χ ( [ N E / K q i ℓ ] m ) , C=\bigl(r_{\mathfrak q_{i_\ell}}(z_j)\bigr)_{1\le\ell,j\le r}\in\operatorname{GL}_r(\mathbb{Z}_n),
\qquad
b_\ell=\chi([\mathcal{N}_{E/K}\mathfrak q_{i_\ell}]_m), C = ( r q i ℓ ( z j ) ) 1 ≤ ℓ , j ≤ r ∈ GL r ( Z n ) , b ℓ = χ ([ N E / K q i ℓ ] m ) , e = C − 1 b ∈ ( Z n ) r . (70) \boxed{e=C^{-1}b\in(\mathbb{Z}_n)^r.} \tag*{(70)} e = C − 1 b ∈ ( Z n ) r . (70)
Thus C C C is the matrix of the Kummer pairing at the selected primes: ⟨ Frob q i ℓ , [ z j ] ⟩ = ζ n C ℓ j \langle\operatorname{Frob}_{q_i\ell},[z_j]\rangle=\zeta_n^{C_{\ell j}} ⟨ Frob q i ℓ , [ z j ]⟩ = ζ n C ℓ j . Choose representatives 0 ≤ e j < n 0\le e_j<n 0 ≤ e j < n and define
β ( e ) = ∏ j = 1 r z j e j . (71) \beta(e)=\prod_{j=1}^{r}z_j^{e_j}. \tag*{(71)} β ( e ) = j = 1 ∏ r z j e j . (71)
For r = 0 r=0 r = 0 , the product is 1 and the coordinate vector is empty.
The chosen elements z j z_j z j determine β ( e ) \beta(e) β ( e ) itself. Their recorded identities (68) also express a power of it in the original Jacobi values:
β ( e ) p b = ∏ j = 1 r ε j e j p b − b j ∏ i A i ∑ j = 1 r e j m j i p b − b j , b = max j b j . (72) \beta(e)^{p^b}=\prod_{j=1}^{r}\varepsilon_j^{e_jp^{b-b_j}}\prod_i A_i^{\sum_{j=1}^{r}e_jm_{ji}p^{b-b_j}},\qquad b=\max_j b_j. \tag*{(72)} β ( e ) p b = j = 1 ∏ r ε j e j p b − b j i ∏ A i ∑ j = 1 r e j m j i p b − b j , b = j max b j . (72)
Set b = 0 b=0 b = 0 if r = 0 r=0 r = 0 . This identity records the relations between the chosen basis representatives and the A i A_i A i ; the product (71) defines β ( e ) \beta(e) β ( e ) itself. The ideals and units defining χ \chi χ are computed over K K K ; an equation for H χ H_\chi H χ is not an input.
Theorem 8.3. The element β χ = β ( e ) \beta_\chi=\beta(e) β χ = β ( e ) from (70) satisfies
ν β χ = χ ∣ G E . \nu_{\beta_\chi}=\chi|_{G_E}. ν β χ = χ ∣ G E . Its coordinate vector is the unique solution. If χ \chi χ has exact order n n n , then [ β χ ] [\beta_\chi] [ β χ ] has order n n n and
E ( β χ n ) = E H χ , H χ ∩ E = K . E\left(\sqrt[n]{\beta_\chi}\right)=EH_\chi,\qquad H_\chi\cap E=K. E ( n β χ ) = E H χ , H χ ∩ E = K .
Lemma 8.4. For H = H χ H=H_\chi H = H χ , one has H ∩ E = K H\cap E=K H ∩ E = K .
Proof. The stable kernel makes H / Q H/\mathbb{Q} H / Q Galois. On Gal ( H ∩ E / K ) \operatorname{Gal}(H\cap E/K) Gal ( H ∩ E / K ) , conjugation by τ K \tau_K τ K acts by inversion because χ \chi χ is anti-invariant, and trivially because E / Q E/\mathbb{Q} E / Q is abelian. This group has odd order and is therefore trivial. □ \square □
Proof of Theorem 8.3. Target membership and the basis give a unique vector e χ e_\chi e χ representing ν − 1 ( χ ∣ G E ) \nu^{-1}(\chi|_{G_E}) ν − 1 ( χ ∣ G E ) . By the residue law and norm compatibility of Artin reciprocity, C e χ = b Ce_\chi=b C e χ = b . Invertibility therefore gives e = e χ e=e_\chi e = e χ and equality of the entire characters, not merely their values at the chosen primes. The final assertions follow from Kummer theory and Lemma 8.4. □ \square □
Corollary 8.5. Let Γ = ⟨ T Σ , A i ⟩ \Gamma=\langle T_\Sigma,A_i\rangle Γ = ⟨ T Σ , A i ⟩ and r = rank Λ Σ r=\operatorname{rank}\Lambda_\Sigma r = rank Λ Σ . The following are equivalent:
(i) p ∤ [ U Σ : Γ ] p\nmid[U_\Sigma:\Gamma] p ∤ [ U Σ : Γ ] ;
(ii) the classes [ A i ] [A_i] [ A i ] generate W Σ W_\Sigma W Σ ;
(iii) some r r r values A i 1 , … , A i r A_{i_1},\ldots,A_{i_r} A i 1 , … , A i r and r r r admissible primes have residue matrix invertible modulo p p p .
Whenever (iii) holds, these r r r values may be used as the basis representatives z j z_j z j in (70) . No saturation is needed.
Proof. The quotient U Σ / Γ U_\Sigma/\Gamma U Σ /Γ is finite; multiplication by p p p is bijective on it precisely when (i) holds. As p ∤ ∣ T Σ ∣ p\nmid|T_\Sigma| p ∤ ∣ T Σ ∣ , this is equivalent to (ii). A generating set of the free Z n \mathbb{Z}_n Z n -module W Σ W_\Sigma W Σ contains r r r elements whose reductions form a basis modulo p p p ; their classes then form a basis over Z n \mathbb{Z}_n Z n . Lemma 8.1 gives (iii). Conversely, an invertible residue matrix makes the map ( Z n ) r → W Σ (\mathbb{Z}_n)^r\to W_\Sigma ( Z n ) r → W Σ injective. Both groups have order n r n^r n r , proving (ii). □ \square □
Example 8.6. The quintic construction in Section 16.1 uses K = Q ( 2 ) K=\mathbb{Q}(\sqrt{2}) K = Q ( 2 ) , n = 5 n=5 n = 5 and ( z 1 , z 2 ) = ( A 1 , A 2 ) (z_1,z_2)=(A_1,A_2) ( z 1 , z 2 ) = ( A 1 , A 2 ) . The primes
q 1 = ( 31 , 2 − 8 , ζ 5 − 2 ) , q 2 = ( 31 , 2 − 8 , ζ 5 − 4 ) \mathfrak q_1=(31,\sqrt{2}-8,\zeta_5-2),\qquad\mathfrak q_2=(31,\sqrt{2}-8,\zeta_5-4) q 1 = ( 31 , 2 − 8 , ζ 5 − 2 ) , q 2 = ( 31 , 2 − 8 , ζ 5 − 4 )
of E E E both lie above ( 7 + 3 2 ) (7+3\sqrt{2}) ( 7 + 3 2 ) and give
C = ( 2 0 0 4 ) , b = ( 4 4 ) , C − 1 b = ( 2 1 ) in F 5 , β χ = A 1 2 A 2 . C=\begin{pmatrix}2&0\\0&4\end{pmatrix},\qquad b=\begin{pmatrix}4\\4\end{pmatrix},\qquad C^{-1}b=\begin{pmatrix}2\\1\end{pmatrix}\ \text{in } \mathbb{F}_5,\qquad\beta_{\chi}=A_1^2A_2. C = ( 2 0 0 4 ) , b = ( 4 4 ) , C − 1 b = ( 2 1 ) in F 5 , β χ = A 1 2 A 2 .
Their rows separate the rank-two module W Σ W_{\Sigma} W Σ ; they need not generate the auxiliary ray class group. The example also shows that these Jacobi values already represent a Kummer basis.
Ordinary generator by equivariant descent
Let n = p a n=p^a n = p a , p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ , and let χ \chi χ be an anti-invariant ray class character of exact order n n n . Section 8 gives β χ = ∏ j z j χ ~ j ∈ U Σ \beta_{\chi}=\prod_j z_j^{\widetilde{\chi}_j}\in U_{\Sigma} β χ = ∏ j z j χ j ∈ U Σ with ν β χ = χ ∣ G E \nu_{\beta_{\chi}}=\chi|_{G_E} ν β χ = χ ∣ G E , where E = K ( ζ n ) E=K(\zeta_n) E = K ( ζ n ) . Put α n = β χ \alpha^n=\beta_{\chi} α n = β χ and L = E ( α ) = E H χ L=E(\alpha)=EH_{\chi} L = E ( α ) = E H χ . By Lemma 8.4 , [ L : E ] = n [L:E]=n [ L : E ] = n . For b ∈ Z n b\in\mathbb{Z}_n b ∈ Z n define
r b : L ⟶ L , r b ∣ E = id , r b ( α ) = ζ n b α . r_b:L\longrightarrow L,\qquad r_b|_E=\operatorname{id},\qquad r_b(\alpha)=\zeta_n^b\alpha. r b : L ⟶ L , r b ∣ E = id , r b ( α ) = ζ n b α .
Thus Spec L \operatorname{Spec}L Spec L over E E E is a torsor under the constant cyclic group Z n \mathbb{Z}_n Z n : on its n n n geometric points this group acts freely and transitively. To recover the desired field over K K K , we descend this torsor equivariantly , meaning that the semilinear descent maps commute with every r b r_b r b . This is standard Galois descent [17 Section 2.3] ; its computational Kummer form is [15 Lemma 4.1] , [16 Section 4.3] .
Represent u ∈ Z n × u\in\mathbb{Z}_n^\times u ∈ Z n × by 1 ≤ u < n 1\leq u<n 1 ≤ u < n , and let [ u v ] n [uv]_n [ uv ] n denote the representative of its product. Recall δ u ∣ K = id \delta_u|_K=\operatorname{id} δ u ∣ K = id , δ u ( ζ n ) = ζ n u \delta_u(\zeta_n)=\zeta_n^u δ u ( ζ n ) = ζ n u .
Theorem 9.1. There is a unique c u ∈ E × c_u\in E^\times c u ∈ E × satisfying
c u n = δ u ( β χ ) β χ u , c u τ ( c u ) = 1. c_u^n=\frac{\delta_u(\beta_{\chi})}{\beta_{\chi}^u},\qquad c_u\tau(c_u)=1. c u n = β χ u δ u ( β χ ) , c u τ ( c u ) = 1. The maps
d u : L ⟶ L , d u ∣ E = δ u , d u ( α ) = c u α u (73) d_u:L\longrightarrow L,\qquad d_u|_E=\delta_u,\qquad d_u(\alpha)=c_u\alpha^u \tag*{(73)} d u : L ⟶ L , d u ∣ E = δ u , d u ( α ) = c u α u (73) are the equivariant descent to H χ H_{\chi} H χ . In particular,
θ χ = ∑ u ∈ Z n × c u α u = Tr L / H χ ( α ) , K ( θ χ ) = H χ . (74) \theta_{\chi}=\sum_{u\in\mathbb{Z}_n^\times}c_u\alpha^u=\operatorname{Tr}_{L/H_{\chi}}(\alpha),\qquad K(\theta_{\chi})=H_{\chi}. \tag*{(74)} θ χ = u ∈ Z n × ∑ c u α u = Tr L / H χ ( α ) , K ( θ χ ) = H χ . (74)
The exponent u u u in (73) is forced by equivariance: d u r b ( α ) = r b d u ( α ) = ζ n u b c u α u d_ur_b(\alpha)=r_bd_u(\alpha)=\zeta_n^{ub}c_u\alpha^u d u r b ( α ) = r b d u ( α ) = ζ n u b c u α u . The extra norm under τ \tau τ chooses the required descent uniquely; in the general case we will choose it by its Artin character.
The proof
Lemma 9.2. If x ∈ E × n x\in E^{\times n} x ∈ E × n and τ ( x ) = x − 1 \tau(x)=x^{-1} τ ( x ) = x − 1 , there is a unique c ∈ E × c\in E^\times c ∈ E × with c n = x c^n=x c n = x and c τ ( c ) = 1 c\tau(c)=1 c τ ( c ) = 1 . If x j ( x ) = 1 xj(x)=1 x j ( x ) = 1 , then c j ( c ) = 1 cj(c)=1 c j ( c ) = 1 as well.
Proof. For b n = x b^n=x b n = x , put ω = b τ ( b ) ∈ μ n \omega=b\tau(b)\in\mu_n ω = b τ ( b ) ∈ μ n and c = b ω ( n − 1 ) / 2 c=b\omega^{(n-1)/2} c = b ω ( n − 1 ) /2 . This gives the required power and norm. Two such roots differ by ξ ∈ μ n \xi\in\mu_n ξ ∈ μ n with ξ 2 = 1 \xi^2=1 ξ 2 = 1 , so are equal. Finally c j ( c ) cj(c) c j ( c ) is totally positive and its n n n th power is one. □
Covariance (Lemma 2.4 ) makes δ u ( β χ ) / β χ u \delta_u(\beta_{\chi})/\beta_{\chi}^u δ u ( β χ ) / β χ u an n n n th power inverted by τ \tau τ . Thus polynomial factorization over E E E and this lemma compute c u c_u c u without knowing H χ H_{\chi} H χ .
For representatives z j z_j z j lifting a Z \mathbb{Z} Z -basis of Γ s a t / T Σ \Gamma^{\mathrm{sat}}/T_{\Sigma} Γ sat / T Σ as in Corollary 7.2 , Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stability of Γ s a t \Gamma^{\mathrm{sat}} Γ sat gives an alternative:
δ u ( z j ) = ω u , j ∏ ℓ z ℓ S u , ℓ j , S u ∈ GL r ( Z ) , ω u , j ∈ T Σ . \delta_u(z_j)=\omega_{u,j}\prod_{\ell}z_\ell^{S_{u,\ell j}},\qquad S_u\in\operatorname{GL}_r(\mathbb{Z}),\qquad\omega_{u,j}\in T_{\Sigma}. δ u ( z j ) = ω u , j ℓ ∏ z ℓ S u , ℓ j , S u ∈ GL r ( Z ) , ω u , j ∈ T Σ .
Valuations give S u S_u S u , and division gives ω u , j \omega_{u,j} ω u , j . Covariance implies q u = ( S u e − u e ) / n ∈ Z r q_u=(S_u e-ue)/n\in\mathbb{Z}^r q u = ( S u e − u e ) / n ∈ Z r . Since p ∤ ∣ T Σ ∣ p\nmid|T_\Sigma| p ∤ ∣ T Σ ∣ , there is a unique ω u ∈ T Σ \omega_u\in T_\Sigma ω u ∈ T Σ with ω u n = ∏ j ω u , j e j \omega_u^n=\prod_j\omega_{u,j}^{e_j} ω u n = ∏ j ω u , j e j . Hence
c u = ω u ∏ j z j q u , j . (75) c_u=\omega_u\prod_j z_j^{q_{u,j}}. \tag*{(75)} c u = ω u j ∏ z j q u , j . (75)
Thus both β χ \beta_\chi β χ and c u c_u c u are explicit products of the same basis representatives z j ∈ E × z_j\in E^\times z j ∈ E × .
Lemma 9.3. The normalized coefficients satisfy c 1 = 1 c_1=1 c 1 = 1 and
c [ u v ] n = δ u ( c v ) c u v β χ ( u v − [ u v ] n ) / n . c_{[uv]_n}=\delta_u(c_v)c_u^v\beta_\chi^{(uv-[uv]_n)/n}. c [ uv ] n = δ u ( c v ) c u v β χ ( uv − [ uv ] n ) / n .
Proof. The right side has n n n th power δ u v ( β χ ) / β χ [ u v ] n \delta_{uv}(\beta_\chi)/\beta_\chi^{[uv]_n} δ uv ( β χ ) / β χ [ uv ] n and norm one under τ \tau τ . Apply Lemma 9.2 .
Proof of Theorem 9.1 . The power and composition equations give a group D = { d u } \mathcal{D}=\{d_u\} D = { d u } commuting with { r b } \{r_b\} { r b } , and Gal ( L / K ) = ⟨ r 1 ⟩ × D \operatorname{Gal}(L/K)=\langle r_1\rangle\times\mathcal{D} Gal ( L / K ) = ⟨ r 1 ⟩ × D . To identify its fixed field, extend τ \tau τ by τ ^ ( α ) = α − 1 \widehat{\tau}(\alpha)=\alpha^{-1} τ ( α ) = α − 1 . The normalization gives τ ^ d u = d u τ ^ \widehat{\tau}d_u=d_u\widehat{\tau} τ d u = d u τ and τ ^ r b τ ^ − 1 = r − b \widehat{\tau}r_b\widehat{\tau}^{-1}=r_{-b} τ r b τ − 1 = r − b . The subgroup fixing H χ H_\chi H χ is a complement, of the form { r b ( u ) d u } \{r_{b(u)}d_u\} { r b ( u ) d u } . Since anti-invariance makes H χ / Q H_\chi/\mathbb{Q} H χ / Q Galois, this complement is τ ^ \widehat{\tau} τ -stable. Thus b ( u ) = − b ( u ) b(u)=-b(u) b ( u ) = − b ( u ) ; oddness of n n n gives b ( u ) = 0 b(u)=0 b ( u ) = 0 . Its fixed field is therefore exactly H χ H_\chi H χ .
The sum in (74) is invariant under all d u d_u d u . If r b r_b r b fixes it, compare its coefficient of α \alpha α in the E E E -basis 1 , α , … , α n − 1 1,\alpha,\ldots,\alpha^{n-1} 1 , α , … , α n − 1 . Since c 1 = 1 c_1=1 c 1 = 1 , this coefficient gives ζ n b = 1 \zeta_n^b=1 ζ n b = 1 . Thus the trace has trivial stabilizer in Gal ( H χ / K ) \operatorname{Gal}(H_\chi/K) Gal ( H χ / K ) and is primitive.
For an arbitrary ordinary character, recombine its cyclotomic and anti-invariant parts by Lemma 2.6 . The general descent in § 14 uses the same equivariant principle when the norm normalization or the equality H χ ∩ E = K H_\chi\cap E=K H χ ∩ E = K is unavailable.
General Fermat Jacobians
As in § 5 , we take Frobenius determinants on RM eigenspaces of a Fermat quotient. The curve is now C M , Δ : Y M + Z Δ = 1 C_{M,\Delta}:Y^M+Z^\Delta=1 C M , Δ : Y M + Z Δ = 1 , with Δ ∣ M \Delta\mid M Δ ∣ M . The ordinary curve is a quotient of this enlarged curve, and the same Gaussian endomorphism supplies multiplication by K K K . The essential change is the normalization:
B v + B v − ⟶ B v + j ( B v + ) . \frac{B_v^+}{B_v^-}\longrightarrow\frac{B_v^+}{j(B_v^+)}. B v − B v + ⟶ j ( B v + ) B v + .
The first quotient compares the two RM signs; the second compares one RM determinant with its complex conjugate. Only the second always compares equal-dimensional spaces on the enlarged curve.
We define these determinants geometrically, evaluate them by Jacobi sums at good primes, and give explicit CM-ideal inputs at bad primes. The latter are necessary for the uniform computational procedure given here; they avoid evaluating the exact bad-prime Frobenius values. § 11 proves that the resulting divisors span, just as § 6 does in the ordinary case. The basis, reciprocity and descent steps then have the same form.
The curve, its endomorphisms, and its determinant line
Let E / Q E/\mathbb{Q} E / Q be the abelian CM field chosen in § 4.2 , and choose
Δ ∣ M , E ⊆ F = Q ( ζ M ) , t = M / Δ . (76) \Delta\mid M,\qquad E\subseteq F=\mathbb{Q}(\zeta_M),\qquad t=M/\Delta. \tag*{(76)} Δ ∣ M , E ⊆ F = Q ( ζ M ) , t = M /Δ. (76)
One may take M = lcm ( Δ , n ) M=\operatorname{lcm}(\Delta,n) M = lcm ( Δ , n ) at odd p p p , and M = lcm ( Δ , 2 b ) M=\operatorname{lcm}(\Delta,2^b) M = lcm ( Δ , 2 b ) at two, with b b b as in (28) . Fix ζ Δ = ζ M t \zeta_\Delta=\zeta_M^t ζ Δ = ζ M t . Our curve is the smooth projective model of
C = C M , Δ : Y M + Z Δ = 1 , (77) C=C_{M,\Delta}: \quad Y^M+Z^\Delta=1, \tag*{(77)} C = C M , Δ : Y M + Z Δ = 1 , (77)
and J = Jac ( C ) J=\operatorname{Jac}(C) J = Jac ( C ) . Its Fermat cover is
F M : U M + V M = 1 ⟶ C , ( U , V ) ⟼ ( U , V t ) . (78) \mathcal{F}_M: U^M+V^M=1\longrightarrow C,\qquad(U,V)\longmapsto(U,V^t). \tag*{(78)} F M : U M + V M = 1 ⟶ C , ( U , V ) ⟼ ( U , V t ) . (78)
The maps
S : C F ⟶ C F , ( Y , Z ) ⟼ ( ζ M Y , Z ) , T : C F ⟶ C F , ( Y , Z ) ⟼ ( Y , ζ Δ Z ) S:C_F\longrightarrow C_F,\quad(Y,Z)\longmapsto(\zeta_M Y,Z),\qquad T:C_F\longrightarrow C_F,\quad(Y,Z)\longmapsto(Y,\zeta_\Delta Z) S : C F ⟶ C F , ( Y , Z ) ⟼ ( ζ M Y , Z ) , T : C F ⟶ C F , ( Y , Z ) ⟼ ( Y , ζ Δ Z )
act on J F J_F J F by pushing forward degree-zero divisors. No point at infinity is chosen. Over P 1 ∖ { 0 , 1 , ∞ } \mathbf{P}^1\setminus\{0,1,\infty\} P 1 ∖ { 0 , 1 , ∞ } , x = Y M x=Y^M x = Y M is a cover with deck group μ M × μ Δ \mu_M\times\mu_\Delta μ M × μ Δ .
Choose ℓ 0 \ell_0 ℓ 0 outside the rational primes below Σ \Sigma Σ and fix ι : F ↪ Q ‾ ℓ 0 \iota:F\hookrightarrow\overline{\mathbf{Q}}_{\ell_0} ι : F ↪ Q ℓ 0 . On V = V ℓ 0 ( J ) ⊗ Q ℓ 0 Q ‾ ℓ 0 V=V_{\ell_0}(J)\otimes_{\mathbf{Q}_{\ell_0}}\overline{\mathbf{Q}}_{\ell_0} V = V ℓ 0 ( J ) ⊗ Q ℓ 0 Q ℓ 0 , for ( v , b ) ∈ Z M × Z Δ (v,b)\in\mathbf{Z}_M\times\mathbf{Z}_\Delta ( v , b ) ∈ Z M × Z Δ , put
V v , b = { w ∈ V : S w = ι ( ζ M v ) w , T w = ι ( ζ Δ b ) w } . V_{v,b}=\{w\in V: Sw=\iota(\zeta_M^v)w,\ Tw=\iota(\zeta_\Delta^b)w\}. V v , b = { w ∈ V : S w = ι ( ζ M v ) w , T w = ι ( ζ Δ b ) w } .
e v Y = 1 M ∑ a = 0 M − 1 ι ( ζ M − v a ) S a , e b Z = 1 Δ ∑ a = 0 Δ − 1 ι ( ζ Δ − b a ) T a . e_v^Y=\frac{1}{M}\sum_{a=0}^{M-1}\iota(\zeta_M^{-va})S^a,\qquad e_b^Z=\frac{1}{\Delta}\sum_{a=0}^{\Delta-1}\iota(\zeta_\Delta^{-ba})T^a. e v Y = M 1 a = 0 ∑ M − 1 ι ( ζ M − v a ) S a , e b Z = Δ 1 a = 0 ∑ Δ − 1 ι ( ζ Δ − ba ) T a .
These are the projectors of (38) with the first level enlarged: e v Y e b Z V = V v , b e_v^Ye_b^ZV=V_{v,b} e v Y e b Z V = V v , b . The spaces are G F G_F G F -stable because S , T S,T S , T are defined over F F F . Select primitive second indices with (39) :
J 0 = J M , Δ p r i m = im ( Δ e Δ p r i m : J → J ) . J_0=J^{\mathrm{prim}}_{M,\Delta}=\operatorname{im}(\Delta e_\Delta^{\mathrm{prim}}:J\to J). J 0 = J M , Δ prim = im ( Δ e Δ prim : J → J ) .
The rational projector is fixed by coefficient Galois action, so J 0 J_0 J 0 is defined over Q \mathbf{Q} Q . A pair ( v , b ) (v,b) ( v , b ) is admissible if
v ≠ 0 in Z M , b ∈ Z Δ × , v + t b ≠ 0 in Z M . (79) v\ne0\text{ in }\mathbf{Z}_M,\qquad b\in\mathbf{Z}_\Delta^\times,\qquad v+tb\ne0\text{ in }\mathbf{Z}_M. \tag*{(79)} v = 0 in Z M , b ∈ Z Δ × , v + t b = 0 in Z M . (79)
Precisely these pairs give the one-dimensional character spaces of J 0 J_0 J 0 (Lemma 10.3 ). All nonzero first indices are used; restricting v v v to units can lose a divisor component.
The Gaussian endomorphism is unchanged:
G Δ = ∑ a ∈ Z Δ χ Δ ( a ) T a , K ↪ End K 0 ( J 0 ) , Δ ⟼ G Δ . (80) G_\Delta=\sum_{a\in\mathbf{Z}_\Delta}\chi_\Delta(a)T^a,\qquad K\hookrightarrow\operatorname{End}^0_K(J_0),\qquad\sqrt{\Delta}\longmapsto G_\Delta. \tag*{(80)} G Δ = a ∈ Z Δ ∑ χ Δ ( a ) T a , K ↪ End K 0 ( J 0 ) , Δ ⟼ G Δ . (80)
Indeed Lemma 5.3 gives G Δ ∣ V v , b = χ Δ ( b ) ι ( Δ ) G_\Delta|_{V_{v,b}}=\chi_\Delta(b)\iota(\sqrt{\Delta}) G Δ ∣ V v , b = χ Δ ( b ) ι ( Δ ) and G Δ 2 = [ Δ ] G_\Delta^2=[\Delta] G Δ 2 = [ Δ ] . The primitive Q ( ζ Δ ) \mathbf{Q}(\zeta_\Delta) Q ( ζ Δ ) -action is faithful, and coefficient equivariance descends the quadratic endomorphism to K K K , exactly as there. For 1 ≤ v < M 1\le v<M 1 ≤ v < M , define
I v ± = { b ∈ Z Δ × : χ Δ ( b ) = ± 1 , v + t b ≠ 0 ( m o d M ) } , r v ± = ∣ I v ± ∣ , I_v^\pm=\{b\in\mathbf{Z}_\Delta^\times:\chi_\Delta(b)=\pm1,\ v+tb\ne0\pmod M\},\qquad r_v^\pm=|I_v^\pm|, I v ± = { b ∈ Z Δ × : χ Δ ( b ) = ± 1 , v + t b = 0 ( mod M )} , r v ± = ∣ I v ± ∣ ,
M v ± = ker ( G Δ ∓ ι ( Δ ) ∣ e v Y e Δ p r i m V ) = ⨁ b ∈ I v ± V v , b . (81) M_v^\pm=\ker(G_\Delta\mp\iota(\sqrt{\Delta})\mid e_v^Ye_\Delta^{\mathrm{prim}}V)=\bigoplus_{b\in I_v^\pm}V_{v,b}. \tag*{(81)} M v ± = ker ( G Δ ∓ ι ( Δ ) ∣ e v Y e Δ prim V ) = b ∈ I v ± ⨁ V v , b . (81)
These are subspaces of the rational Tate module after extending scalars; they are not asserted to be individual abelian subvarieties. Their projectors are 1 2 ( 1 ± G Δ / ι ( Δ ) ) \frac{1}{2}(1\pm G_\Delta/\iota(\sqrt{\Delta})) 2 1 ( 1 ± G Δ / ι ( Δ )) .
Put r = φ ( Δ ) / 2 r=\varphi(\Delta)/2 r = φ ( Δ ) /2 . Both dimensions equal r r r unless v + t b = 0 v+tb=0 v + t b = 0 for a primitive b b b ; then one equals r − 1 r-1 r − 1 and the other r r r . For example, on C 8 , 8 C_{8,8} C 8 , 8 with v = 1 v=1 v = 1 , the positive index b = 7 b=7 b = 7 is excluded, so ( r 1 + , r 1 − ) = ( 1 , 2 ) (r_1^+,r_1^-)=(1,2) ( r 1 + , r 1 − ) = ( 1 , 2 ) . The ordinary quotient of determinants would have absolute value q − 1 / 2 q^{-1/2} q − 1/2 . In contrast, χ Δ ( − 1 ) = 1 \chi_\Delta(-1)=1 χ Δ ( − 1 ) = 1 gives I − v ± = − I v ± I_{-v}^\pm=-I_v^\pm I − v ± = − I v ± and r − v ± = r v ± r_{-v}^\pm=r_v^\pm r − v ± = r v ± . We therefore use
L v ( 1 ) = det M v + ⊗ ( det M − v + ) ∨ , det M v + = ⋀ r v + M v + . (82) \mathcal{L}_v^{(1)}=\det M_v^+\otimes(\det M_{-v}^+)^\vee,\qquad\det M_v^+=\bigwedge^{r_v^+}M_v^+. \tag*{(82)} L v ( 1 ) = det M v + ⊗ ( det M − v + ) ∨ , det M v + = ⋀ r v + M v + . (82)
Here det 0 = Q ‾ ℓ 0 \det0=\overline{\mathbb{Q}}_{\ell_0} det 0 = Q ℓ 0 . This one-dimensional representation has weight zero. The negative RM sign gives another valid family; either sign alone suffices for spanning. Complex conjugation here acts on the algebraic coefficients, not as an Q ‾ ℓ 0 \overline{\mathbb{Q}}_{\ell_0} Q ℓ 0 -linear operator on V V V .
The Frobenius determinant
For each rational prime ℓ \ell ℓ below Σ \Sigma Σ , choose L ℓ ⊇ F L_\ell\supseteq F L ℓ ⊇ F over which J 0 J_0 J 0 has good reduction and a prime P ℓ ∣ ℓ \mathfrak{P}_\ell\mid\ell P ℓ ∣ ℓ . Set
q ℓ = P ℓ ∩ O F , q ℓ = ∣ O L ℓ / P ℓ ∣ . \mathfrak{q}_\ell=\mathfrak{P}_\ell\cap\mathcal{O}_F,\qquad q_\ell=\lvert\mathcal{O}_{L_\ell}/\mathfrak{P}_\ell\rvert. q ℓ = P ℓ ∩ O F , q ℓ = ∣ O L ℓ / P ℓ ∣ .
At ℓ ∤ M \ell\nmid M ℓ ∤ M one may take L ℓ = F L_\ell=F L ℓ = F ; at every prime one may use Lemma 10.4. Write Frob P ℓ \operatorname{Frob}_{\mathfrak{P}_\ell} Frob P ℓ for the q ℓ q_\ell q ℓ -power Frobenius endomorphism of the reduction, acting on Tate modules by arithmetic Frobenius as in (45) . At a bad prime this means Frobenius on the restriction to G L ℓ G_{L_\ell} G L ℓ after killing inertia. It is not a canonical Frobenius lift on the full ramified G F G_F G F -representation, nor necessarily the operator on inertia invariants defining its local Euler factor over F F F .
Definition 10.1. Define B q ℓ ± ( v ) ∈ F B_{q_\ell}^{\pm}(v)\in F B q ℓ ± ( v ) ∈ F by
ι ( B q ℓ ± ( v ) ) = det ( Frob P ℓ ∣ M v ± ) . (83) \iota\left(B_{q_\ell}^{\pm}(v)\right)=\det\left(\left.\operatorname{Frob}_{\mathfrak{P}_\ell}\right|M_v^{\pm}\right). \tag*{(83)} ι ( B q ℓ ± ( v ) ) = det ( Frob P ℓ ∣ M v ± ) . (83)
Thus Frob P ℓ \operatorname{Frob}_{\mathfrak{P}_\ell} Frob P ℓ acts on det M v ± \det M_v^{\pm} det M v ± by ι ( B q ℓ ± ( v ) ) \iota(B_{q_\ell}^{\pm}(v)) ι ( B q ℓ ± ( v )) . The prime subscript is in F F F , as in § 5 ; the chosen pair ( L ℓ , P ℓ ) (L_\ell,\mathfrak{P}_\ell) ( L ℓ , P ℓ ) is understood throughout. The algebraicity of these determinants follows from Lemma 10.4 and the CM algebra below. For the positive sign put
A q ℓ ( v ) = N F / E ( B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ) . (84) A_{q_\ell}(v)=N_{F/E}\left(\frac{B_{q_\ell}^{+}(v)}{j(B_{q_\ell}^{+}(v))}\right). \tag*{(84)} A q ℓ ( v ) = N F / E ( j ( B q ℓ + ( v )) B q ℓ + ( v ) ) . (84)
and
A Σ = { A q ℓ ( v ) : 1 ≤ v < M , ℓ lies below Σ } . (85) A_\Sigma=\left\{A_{q_\ell}(v):1\leq v<M,\ \ell\text{ lies below }\Sigma\right\}. \tag*{(85)} A Σ = { A q ℓ ( v ) : 1 ≤ v < M , ℓ lies below Σ } . (85)
An empty determinant is 1 1 1 .
The quotient is the Frobenius scalar on L v ( 1 ) \mathcal{L}_v^{(1)} L v ( 1 ) ; A q ℓ ( v ) A_{q_\ell}(v) A q ℓ ( v ) is its field norm. In § 5 the ordinary determinants already lie in E E E , so no further norm is needed. Here Gal ( F / E ) \operatorname{Gal}(F/E) Gal ( F / E ) may move the first index, and N F / E N_{F/E} N F / E puts the input in E E E . This is a coefficient norm, not a CM reflex norm. Equivalently, identifying the determinant scalars with elements of F F F through ι \iota ι ,
A q ℓ ( v ) = N F / E ( det ( Frob P ℓ ∣ M v + ) det ( Frob P ℓ ∣ M − v + ) ) . A_{q_\ell}(v)=N_{F/E}\left(\frac{\det\left(\left.\operatorname{Frob}_{\mathfrak{P}_\ell}\right|M_v^{+}\right)}{\det\left(\left.\operatorname{Frob}_{\mathfrak{P}_\ell}\right|M_{-v}^{+}\right)}\right). A q ℓ ( v ) = N F / E ( det ( Frob P ℓ ∣ M − v + ) det ( Frob P ℓ ∣ M v + ) ) .
Theorem 10.2. On the determinant line,
Frob P ℓ ∣ L v ( 1 ) = ι ( B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ) id . \left.\operatorname{Frob}_{\mathfrak{P}_\ell}\right|_{\mathcal{L}_v^{(1)}}=\iota\left(\frac{B_{q_\ell}^{+}(v)}{j(B_{q_\ell}^{+}(v))}\right)\operatorname{id}. Frob P ℓ ∣ L v ( 1 ) = ι ( j ( B q ℓ + ( v )) B q ℓ + ( v ) ) id . For every 1 ≤ v < M 1\leq v<M 1 ≤ v < M and support prime ℓ \ell ℓ ,
B q ℓ + ( − v ) = j ( B q ℓ + ( v ) ) , B q ℓ + ( v ) j ( B q ℓ + ( v ) ) = q ℓ r v + , B_{q_\ell}^{+}(-v)=j\left(B_{q_\ell}^{+}(v)\right),\qquad B_{q_\ell}^{+}(v)j\left(B_{q_\ell}^{+}(v)\right)=q_\ell^{r_v^{+}}, B q ℓ + ( − v ) = j ( B q ℓ + ( v ) ) , B q ℓ + ( v ) j ( B q ℓ + ( v ) ) = q ℓ r v + , and
A q ℓ ( v ) ∈ U Σ ( 1 ) , j ( A q ℓ ( v ) ) = A q ℓ ( v ) − 1 , A q ℓ ( − v ) = A q ℓ ( v ) − 1 . A_{q_\ell}(v)\in U_\Sigma^{(1)},\qquad j\left(A_{q_\ell}(v)\right)=A_{q_\ell}(v)^{-1},\qquad A_{q_\ell}(-v)=A_{q_\ell}(v)^{-1}. A q ℓ ( v ) ∈ U Σ ( 1 ) , j ( A q ℓ ( v ) ) = A q ℓ ( v ) − 1 , A q ℓ ( − v ) = A q ℓ ( v ) − 1 . Its divisor is supported above ℓ \ell ℓ . These assertions hold also when ℓ ∣ M \ell\mid M ℓ ∣ M , using the specified good-reduction extension.
The proof, like that of Theorem 5.1, uses the Fermat character decomposition. CM supplies a common Frobenius at every prime; the individual eigenvalues below are used to evaluate this determinant and its divisor.
Character decomposition and proof of the determinant theorem.
Lemma 10.3. The space V v , b V_{v,b} V v , b is one-dimensional precisely when v ≠ 0 v \ne0 v = 0 in Z M \mathbf{Z}_M Z M , b ≠ 0 b \ne0 b = 0 in Z Δ \mathbf{Z}_{\Delta} Z Δ , v + t b ≠ 0 v+tb \ne0 v + t b = 0 in Z M \mathbf{Z}_M Z M ; otherwise it is zero. These spaces exhaust V V V , and
2 g ( C M , Δ ) = dim Q ℓ 0 V = ( M − 2 ) ( Δ − 1 ) , 2 dim J 0 = ( M − 2 ) φ ( Δ ) . 2g(C_{M,\Delta})=\dim_{\mathbf{Q}_{\ell_0}}V=(M-2)(\Delta-1),\qquad2\dim J_0=(M-2)\varphi(\Delta). 2 g ( C M , Δ ) = dim Q ℓ 0 V = ( M − 2 ) ( Δ − 1 ) , 2 dim J 0 = ( M − 2 ) φ ( Δ ) .
Proof. The cover (78) identifies ( v , b ) (v,b) ( v , b ) with Fermat indices ( v , t b ) (v,tb) ( v , t b ) . The classical multiplicity-one theorem applies exactly when both indices and their sum are nonzero [21 Section 1] [27 Propositions 4.8 and 4.11] . For each b ≠ 0 b\ne0 b = 0 the two excluded first indices are 0 0 0 and − t b -tb − t b . Counting all second indices, or only primitive ones, gives the two dimensions. □ \square □
The CM algebra is D = Q [ S , T ] ∣ J 0 \mathcal{D}=\mathbf{Q}[S,T]|_{J_0} D = Q [ S , T ] ∣ J 0 . For an admissible pair, its character field is
F v , b = Q ( ζ M v , ζ Δ b ) = Q ( ζ M / gcd ( v , t ) ) ⊆ F , (86) F_{v,b}=\mathbf{Q}(\zeta_M^v,\zeta_\Delta^b)=\mathbf{Q}(\zeta_{M/\gcd(v,t)})\subseteq F, \tag*{(86)} F v , b = Q ( ζ M v , ζ Δ b ) = Q ( ζ M / g c d ( v , t ) ) ⊆ F , (86)
since lcm ( M / gcd ( M , v ) , Δ ) = M / gcd ( v , t ) \operatorname{lcm}(M/\gcd(M,v),\Delta)=M/\gcd(v,t) lcm ( M / g cd( M , v ) , Δ ) = M / g cd( v , t ) . Writing σ u ( ζ M ) = ζ M u \sigma_u(\zeta_M)=\zeta_M^u σ u ( ζ M ) = ζ M u , coefficient conjugation sends ( v , b ) (v,b) ( v , b ) to ( u v , u b ) (uv,ub) ( uv , u b ) . Choose one pair in each such orbit O \mathcal{O} O , and put F O = F v , b F_{\mathcal{O}}=F_{v,b} F O = F v , b . Character orthogonality gives
D ≃ ∏ O F O , dim Q D = 2 dim J 0 , V ℓ 0 ( J 0 ) is free of rank one over D ⊗ Q ℓ 0 . \mathcal{D}\simeq\prod_{\mathcal{O}}F_{\mathcal{O}},\qquad\dim_{\mathbf{Q}}\mathcal{D}=2\dim J_0,\qquad V_{\ell_0}(J_0)\text{ is free of rank one over }\mathcal{D}\otimes\mathbf{Q}_{\ell_0}. D ≃ O ∏ F O , dim Q D = 2 dim J 0 , V ℓ 0 ( J 0 ) is free of rank one over D ⊗ Q ℓ 0 .
Every field F O F_{\mathcal{O}} F O contains Q ( ζ Δ ) \mathbf{Q}(\zeta_\Delta) Q ( ζ Δ ) and hence K K K . The isogeny factors need not be simple; only the rank-one CM action is used.
Lemma 10.4. The variety J 0 J_0 J 0 has good reduction at every finite prime of L = F ( J 0 [ 15 ] ) L=F(J_0[15]) L = F ( J 0 [ 15 ]) . Its residue-field Frobenius belongs to the specialized algebra D \mathcal{D} D .
Proof. CM gives potential good reduction [31 Theorem 7] . At residue characteristic ℓ \ell ℓ , choose r ∈ { 3 , 5 } ∖ { ℓ } r\in\{3,5\}\setminus\{\ell\} r ∈ { 3 , 5 } ∖ { ℓ } . Over L L L , finite inertia acts trivially modulo r r r on T r ( J 0 ) T_r(J_0) T r ( J 0 ) . The kernel 1 + r M 2 dim J 0 ( Z ^ r ) 1+rM_{2\dim J_0}(\widehat{\mathbf{Z}}_r) 1 + r M 2 d i m J 0 ( Z r ) is torsion-free by the r r r -adic logarithm, so inertia is trivial. Néron–Ogg–Shafarevich gives good reduction [31 Theorem 1] .
The deck endomorphisms specialize faithfully and are defined over the residue field. Frobenius therefore centralizes D \mathcal{D} D . The rank-one property makes its Tate-module centralizer D ⊗ Q r \mathcal{D}\otimes\mathbf{Q}_r D ⊗ Q r ; hence the centralizer in the rational endomorphism algebra has dimension at most dim Q D \dim_{\mathbf{Q}}\mathcal{D} dim Q D . Since it contains D \mathcal{D} D , it equals D \mathcal{D} D . □ \square □
This also applies at primes dividing 2 Δ 2\Delta 2Δ : G Δ 2 = [ Δ ] G_\Delta^2=[\Delta] G Δ 2 = [ Δ ] survives specialization and [ Δ ] [\Delta] [ Δ ] is still a nonzero isogeny. Neither smooth reduction of the written affine curve nor separability of [ Δ ] [\Delta] [ Δ ] is required.
For an admissible pair, define the one-dimensional determinant B q ℓ ( v , b ) B_{q_\ell}(v,b) B q ℓ ( v , b ) by evaluating Frob p ℓ ∈ D \operatorname{Frob}_{\mathfrak{p}_\ell}\in\mathcal{D} Frob p ℓ ∈ D on V v , b V_{v,b} V v , b :
B q ℓ ( v , b ) ∈ O F v , b , Frob p ℓ ∣ V v , b = ι ( B q ℓ ( v , b ) ) id , B_{q_\ell}(v,b)\in\mathcal{O}_{F_{v,b}},\qquad\operatorname{Frob}_{\mathfrak{p}_\ell}|_{V_{v,b}}=\iota(B_{q_\ell}(v,b))\operatorname{id}, B q ℓ ( v , b ) ∈ O F v , b , Frob p ℓ ∣ V v , b = ι ( B q ℓ ( v , b )) id ,
B q ℓ ( v , b ) j ( B q ℓ ( v , b ) ) = q ℓ . (87) B_{q_\ell}(v,b)j(B_{q_\ell}(v,b))=q_\ell. \tag*{(87)} B q ℓ ( v , b ) j ( B q ℓ ( v , b )) = q ℓ . (87)
All complex absolute values are q ℓ 1 / 2 q_\ell^{1/2} q ℓ 1/2 , and σ u ( B q ℓ ( v , b ) ) = B q ℓ ( u v , u b ) \sigma_u(B_{q_\ell}(v,b))=B_{q_\ell}(uv,ub) σ u ( B q ℓ ( v , b )) = B q ℓ ( uv , u b ) , with the reduction prime fixed. Consequently
B q ℓ ± ( v ) = ∏ b ∈ I v ± B q ℓ ( v , b ) . B_{q_\ell}^{\pm}(v)=\prod_{b\in I_v^{\pm}}B_{q_\ell}(v,b). B q ℓ ± ( v ) = b ∈ I v ± ∏ B q ℓ ( v , b ) .
This is the character-space evaluation of (83) . For fixed ( L ℓ , p ℓ ) (L_\ell,\mathfrak{p}_\ell) ( L ℓ , p ℓ ) , these algebraic values are independent of the auxiliary prime ℓ 0 \ell_0 ℓ 0 : they evaluate the intrinsic endomorphism $\operatorname{Frob}_{\mathfrak{p}_\ell}\in\mathcal{D} at the algebraic deck characters. The embedding ι \iota ι realizes those same characters on the Tate module. Consequently B q ℓ ± ( v ) B_{\mathfrak{q}_{\ell}}^{\pm}(v) B q ℓ ± ( v ) and A q ℓ ( v ) A_{q_{\ell}}(v) A q ℓ ( v ) do not acquire an additional ℓ 0 \ell_0 ℓ 0 -parameter.
Proof of Theorem 10.2. Coefficient conjugation sends I v + I_v^+ I v + to I − v − I_{-v}^- I − v − , giving j ( B q ℓ + ( v ) ) = B q ℓ + ( − v ) j(B_{\mathfrak{q}_{\ell}}^+(v))=B_{\mathfrak{q}_{\ell}}^+(-v) j ( B q ℓ + ( v )) = B q ℓ + ( − v ) . The product of the Weil identities in (87) gives B q ℓ + ( v ) j ( B q ℓ + ( v ) ) = q ℓ r v + B_{\mathfrak{q}_{\ell}}^+(v)j(B_{\mathfrak{q}_{\ell}}^+(v))=q_{\ell}^{r_v^+} B q ℓ + ( v ) j ( B q ℓ + ( v )) = q ℓ r v + . Thus
B q ℓ + ( v ) j ( B q ℓ + ( v ) ) = B q ℓ + ( v ) 2 q ℓ r v + . \frac{B_{\mathfrak{q}_{\ell}}^+(v)}{j(B_{\mathfrak{q}_{\ell}}^+(v))}=\frac{B_{\mathfrak{q}_{\ell}}^+(v)^2}{q_{\ell}^{r_v^+}}. j ( B q ℓ + ( v )) B q ℓ + ( v ) = q ℓ r v + B q ℓ + ( v ) 2 .
This quotient has norm one under j j j . Taking determinants on (82) gives its Frobenius scalar. Integrality and the Weil norm restrict every individual divisor to primes above ℓ \ell ℓ . Norm to E E E preserves that support and commutes with j j j , proving the assertions. □
More generally σ u ( B q ℓ ε ( v ) ) = B q ℓ ε χ Δ ( u ) ( u v ) \sigma_u(B_{\mathfrak{q}_{\ell}}^\varepsilon(v))=B_{\mathfrak{q}_{\ell}}^{\varepsilon\chi_\Delta(u)}(uv) σ u ( B q ℓ ε ( v )) = B q ℓ ε χ Δ ( u ) ( uv ) . Enlarging the residue field by degree e e e raises every determinant to its e e e th power. The normalized coefficients may consequently change by powers; their saturated Kummer module will not.
Jacobi evaluation at good primes
At ℓ ∤ M \ell\nmid M ℓ ∤ M , take L ℓ = F L_\ell=F L ℓ = F , put k = O F / q ℓ k=\mathcal{O}_F/\mathfrak{q}_\ell k = O F / q ℓ , q = ∣ k ∣ q=|k| q = ∣ k ∣ , and define λ : k × → μ M ( F ) \lambda:k^\times\to\mu_M(F) λ : k × → μ M ( F ) by λ ( x ˉ ) = x ( q − 1 ) / M \lambda(\bar{x})=x^{(q-1)/M} λ ( x ˉ ) = x ( q − 1 ) / M . Extend its character powers by zero at zero. With the convention (44) ,
J q ℓ ( v , t b ) = J ( λ v , λ t b ) = ∑ x ∈ k λ ( x ) v λ ( 1 − x ) t b . (88) J_{q_\ell}(v,tb)=J(\lambda^v,\lambda^{tb})=\sum_{x\in k}\lambda(x)^v\lambda(1-x)^{tb}. \tag*{(88)} J q ℓ ( v , t b ) = J ( λ v , λ t b ) = x ∈ k ∑ λ ( x ) v λ ( 1 − x ) t b . (88)
For an admissible pair,
B q ℓ ( v , b ) = − J q ℓ ( v , t b ) B_{q_\ell}(v,b)=-J_{q_\ell}(v,tb) B q ℓ ( v , b ) = − J q ℓ ( v , t b )
by the same Frobenius–Jacobi formula as in Lemma 5.4 [10 Theorem A] [27 Theorem 1.2 and Section 6] . Thus the determinant input is
A q ℓ ( v ) = N F / E ( ∏ b ∈ I v + J q ℓ ( v , t b ) j ( J q ℓ ( v , t b ) ) ) . (89) A_{q_\ell}(v)=N_{F/E}\left(\prod_{b\in I_v^+}\frac{J_{q_\ell}(v,tb)}{j(J_{q_\ell}(v,tb))}\right). \tag*{(89)} A q ℓ ( v ) = N F / E b ∈ I v + ∏ j ( J q ℓ ( v , t b )) J q ℓ ( v , t b ) . (89)
The determinant signs cancel. For explicit evaluation, choose a generator g g g of k × k^\times k × and a ∈ Z M × a\in\mathbb{Z}_M^\times a ∈ Z M × with g ( q − 1 ) / M = ζ ˉ M a g^{(q-1)/M}=\bar{\zeta}_M^a g ( q − 1 ) / M = ζ ˉ M a ; then λ ( g e ) = ζ M a e \lambda(g^e)=\zeta_M^{ae} λ ( g e ) = ζ M a e . The first index need not be primitive: the two characters and their product need only be nontrivial.
Explicit input at bad primes
At ℓ ∣ M \ell\mid M ℓ ∣ M , the finite-field character of order M M M used above does not exist. This does not rule out smaller-conductor Jacobi descriptions or ramified local formulas. Our uniform algorithm uses the CM ideal of the determinant and computes an equivalent input in E E E ; it does not compute its exact Frobenius eigenvalue. Put G = ( Z / M Z ) × G=(\mathbb{Z}/M\mathbb{Z})^\times G = ( Z / M Z ) × . For an admissible pair, define
h v , b ( u ) = { 1 , { u v / M } + { u b / Δ } < 1 , 0 , { u v / M } + { u b / Δ } > 1 , u ∈ G . (90) h_{v,b}(u)=
\begin{cases}
1, & \{uv/M\}+\{ub/\Delta\}<1,\\
0, & \{uv/M\}+\{ub/\Delta\}>1,
\end{cases}
\qquad u\in G. \tag*{(90)} h v , b ( u ) = { 1 , 0 , { uv / M } + { u b /Δ } < 1 , { uv / M } + { u b /Δ } > 1 , u ∈ G . (90)
Equality cannot occur by (79) . These are the CM-type indicators; § 11.1 explains their geometric origin and proves the ideal formula. For the chosen q ℓ ∣ ℓ \mathfrak{q}_\ell\mid\ell q ℓ ∣ ℓ , compute
H v , ℓ = ∑ u ∈ G ∑ b ∈ I v + h v , b ( u ) [ σ u − 1 q ℓ ] , H_{v,\ell}=\sum_{u\in G}\sum_{b\in I_v^+}h_{v,b}(u)[\sigma_u^{-1}\mathfrak{q}_\ell], H v , ℓ = u ∈ G ∑ b ∈ I v + ∑ h v , b ( u ) [ σ u − 1 q ℓ ] ,
a v , ℓ = ∏ u ∈ G ∏ b ∈ I v + ( σ u − 1 q ℓ ) h v , b ( u ) . (91) a_{v,\ell}=\prod_{u\in G}\prod_{b\in I_v^+}(\sigma_u^{-1}\mathfrak{q}_\ell)^{h_{v,b}(u)}. \tag*{(91)} a v , ℓ = u ∈ G ∏ b ∈ I v + ∏ ( σ u − 1 q ℓ ) h v , b ( u ) . (91)
Repeated primes are combined. Proposition 11.2 gives the determinant formula
div F B q ℓ + ( v ) = m ℓ H v , ℓ , m ℓ = f ( R ℓ / q ℓ ) . \operatorname{div}_{F} B_{q_\ell}^{+}(v)=m_\ell H_{v,\ell},\qquad m_\ell=f(\mathfrak{R}_\ell/q_\ell). div F B q ℓ + ( v ) = m ℓ H v , ℓ , m ℓ = f ( R ℓ / q ℓ ) .
The divisor H v , ℓ H_{v,\ell} H v , ℓ is computable without knowing m ℓ m_\ell m ℓ or constructing L ℓ L_\ell L ℓ . Compute the order h h h of [ a v , ℓ ] [\mathfrak{a}_{v,\ell}] [ a v , ℓ ] in Cl ( O F ) \operatorname{Cl}(\mathcal{O}_{F}) Cl ( O F ) and a generator of its principal power [7 Chapters 6–7] . The explicit input is
( x ) = a v , ℓ h , A q ℓ ′ ( v ) = N F / E ( x j ( x ) ) ∈ U Σ ( 1 ) . (92) (x)=\mathfrak{a}_{v,\ell}^{h},\qquad A'_{q_\ell}(v)=N_{F/E}\left(\frac{x}{j(x)}\right)\in U_{\Sigma}^{(1)}. \tag*{(92)} ( x ) = a v , ℓ h , A q ℓ ′ ( v ) = N F / E ( j ( x ) x ) ∈ U Σ ( 1 ) . (92)
It is an exact algebraic number. Ideal factorization, class-group and principal-ideal arithmetic in F F F suffice; no root-of-unity determination, good-reduction model or residue-degree multiplier is needed. The procedure also works at good primes.
Proposition 10.5. Let h > 0 h>0 h > 0 and x ∈ F × x\in F^\times x ∈ F × satisfy div F ( x ) = h H v , ℓ \operatorname{div}_{F}(x)=hH_{v,\ell} div F ( x ) = h H v , ℓ , and define A q ℓ ′ ( v ) A'_{q_\ell}(v) A q ℓ ′ ( v ) by (92) . Then, for some ξ ∈ μ ( F ) \xi\in\mu(F) ξ ∈ μ ( F ) ,
( x j ( x ) ) m ℓ = ξ ( B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ) h , ( A q ℓ ′ ( v ) ) m ℓ = N F / E ( ξ ) A q ℓ ( v ) h . (93) \left(\frac{x}{j(x)}\right)^{m_\ell}
=\xi\left(\frac{B_{q_\ell}^{+}(v)}{j(B_{q_\ell}^{+}(v))}\right)^{h},
\qquad
\left(A'_{q_\ell}(v)\right)^{m_\ell}=N_{F/E}(\xi)A_{q_\ell}(v)^{h}. \tag*{(93)} ( j ( x ) x ) m ℓ = ξ ( j ( B q ℓ + ( v )) B q ℓ + ( v ) ) h , ( A q ℓ ′ ( v ) ) m ℓ = N F / E ( ξ ) A q ℓ ( v ) h . (93) In particular , div E A q ℓ ′ ( v ) = ( h / m ℓ ) div E A q ℓ ( v ) \operatorname{div}_{E} A'_{q_\ell}(v)=(h/m_\ell)\operatorname{div}_{E} A_{q_\ell}(v) div E A q ℓ ′ ( v ) = ( h / m ℓ ) div E A q ℓ ( v ) . If h h h is the ideal-class order, then
h ∣ m ℓ , B q ℓ + ( v ) j ( B q ℓ + ( v ) ) = ξ 0 ( x j ( x ) ) m ℓ / h , ξ 0 ∈ μ ( F ) . (94) h\mid m_\ell,\qquad
\frac{B_{q_\ell}^{+}(v)}{j(B_{q_\ell}^{+}(v))}
=\xi_{0}\left(\frac{x}{j(x)}\right)^{m_\ell/h},
\qquad\xi_{0}\in\mu(F). \tag*{(94)} h ∣ m ℓ , j ( B q ℓ + ( v )) B q ℓ + ( v ) = ξ 0 ( j ( x ) x ) m ℓ / h , ξ 0 ∈ μ ( F ) . (94) Replacing any or all of the A q ℓ ( v ) A_{q_\ell}(v) A q ℓ ( v ) by A q ℓ ′ ( v ) A'_{q_\ell}(v) A q ℓ ′ ( v ) preserves their rational divisor span. After adjoining μ ( E ) \mu(E) μ ( E ) and taking internal p p p -saturation, each complete family gives W Σ ( 1 ) ( E , p a ) W_{\Sigma}^{(1)}(E,p^{a}) W Σ ( 1 ) ( E , p a ) for every a ≥ 1 a\ge1 a ≥ 1 , including p = 2 p=2 p = 2 .
Proof. The two sides of the first identity, omitting ξ \xi ξ , have divisor m ℓ h ( H v , ℓ − j H v , ℓ ) m_\ell h(H_{v,\ell}-jH_{v,\ell}) m ℓ h ( H v , ℓ − j H v , ℓ ) . Their quotient is a unit with absolute value one at every embedding, since F F F is CM and both normalized elements have norm one under j j j . Kronecker’s theorem makes it a root of unity [24 Chapter I, Proposition 7.1] . Taking the field norm gives the second identity. For the class order, ( B q ℓ + ( v ) ) = a v , ℓ m ℓ (B_{q_\ell}^{+}(v))=\mathfrak{a}_{v,\ell}^{m_\ell} ( B q ℓ + ( v )) = a v , ℓ m ℓ implies h ∣ m ℓ h\mid m_\ell h ∣ m ℓ ; the same argument proves (94) .
Each replaced divisor is a nonzero rational multiple of the original one. Theorem 11.1 therefore gives finite index in U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) for the group Γ \Gamma Γ generated by the chosen family and μ ( E ) \mu(E) μ ( E ) . Proposition 7.1 gives
Γ s a t / ( Γ s a t ) p a ≃ U Σ ( 1 ) / ( U Σ ( 1 ) ) p a ≃ W Σ ( 1 ) ( E , p a ) . \Gamma^{\mathrm{sat}}/(\Gamma^{\mathrm{sat}})^{p^{a}}
\simeq U_{\Sigma}^{(1)}/(U_{\Sigma}^{(1)})^{p^{a}}
\simeq W_{\Sigma}^{(1)}(E,p^{a}). Γ sat / ( Γ sat ) p a ≃ U Σ ( 1 ) / ( U Σ ( 1 ) ) p a ≃ W Σ ( 1 ) ( E , p a ) .
This is saturation inside U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) , not division by h h h or m ℓ m_\ell m ℓ modulo p a p^{a} p a . □
Changing x x x by a unit changes x / j ( x ) x/j(x) x / j ( x ) only by a root of unity. The generated groups, and even their saturated subgroups of E × E^\times E × , need not agree; their classes give the same Kummer module. In particular, a p p p -factor in m ℓ / h m_\ell/h m ℓ / h may change the index removed by saturation. The full group μ ( E ) \mu(E) μ ( E ) must be included, also at two. Section 16.5 works out this input for K = Q ( 6 ) K=\mathbb{Q}(\sqrt{6}) K = Q ( 6 ) at ℓ = p = 3 \ell=p=3 ℓ = p = 3 .
Remark 10.6. The CM ideal formula and the principalization procedure answer different questions. The former computes the ideal of the actual Frobenius; (92) computes an explicit substitute whose divisor is a rational multiple of that of the normalized Frobenius value. It does not claim to recover the actual eigenvalue.
For a chosen good-reduction residue cardinality q ℓ q_\ell q ℓ , the ideal and Weil relation determine each B = B q ℓ ( v , b ) B=B_{q_\ell}(v,b) B = B q ℓ ( v , b ) up to μ ( F v , b ) \mu(F_{v,b}) μ ( F v , b ) . If ( a ) = ( B ) (a)=(B) ( a ) = ( B ) , solve the unit equation
u ∈ O F v , b × , u j ( u ) = q ℓ a j ( a ) . u\in\mathcal{O}_{F_{v,b}}^\times,\qquad u_j(u)=\frac{q_\ell}{a_j(a)}. u ∈ O F v , b × , u j ( u ) = a j ( a ) q ℓ .
A solution exists because B / a B/a B / a is a unit; unit-group generators reduce the equation to a calculation with integer relations. The finite candidate set is { ζ a u : ζ ∈ μ ( F v , b ) } \{\zeta a u:\zeta\in\mu(F_{v,b})\} { ζ a u : ζ ∈ μ ( F v , b )} . The remaining root of unity requires the local CM character or an actual Frobenius computation; the ideal alone does not select it.
Exact ramified formulas do exist. To relate them to our determinants, let ρ v , b , q ℓ : F q ℓ × → Q ‾ ℓ × \rho_{v,b,q_\ell}:F_{q_\ell}^\times\to\overline{\mathbb{Q}}_\ell^\times ρ v , b , q ℓ : F q ℓ × → Q ℓ × be the character on V v , b V_{v,b} V v , b composed with local reciprocity, normalized to send a uniformizer to arithmetic Frobenius in the unramified quotient. For a uniformizer ϖ \varpi ϖ of ( L ℓ ) P ℓ (L_\ell)_{\mathfrak{P}_\ell} ( L ℓ ) P ℓ , compatibility with local norms gives
ι ( B q ℓ ( v , b ) ) = ρ v , b , q ℓ ( N ( L ℓ ) P ℓ / F q ℓ ( ϖ ) ) . \iota(B_{q_\ell}(v,b))=\rho_{v,b,q_\ell}\left(N_{(L_\ell)_{\mathfrak{P}_\ell}/F_{q_\ell}}(\varpi)\right). ι ( B q ℓ ( v , b )) = ρ v , b , q ℓ ( N ( L ℓ ) P ℓ / F q ℓ ( ϖ ) ) .
The restriction over L ℓ L_\ell L ℓ is unramified, so this value does not depend on ϖ \varpi ϖ ; multiplying over b ∈ I v + b\in I_v^+ b ∈ I v + gives ι ( B q ℓ + ( v ) ) \iota(B_{q_\ell}^+(v)) ι ( B q ℓ + ( v )) . Thus a fully normalized local-character formula computes the exact eigenvalue, including its root of unity.
Coleman–McCallum give such local formulas at odd residue characteristic in [12 Theorem 5.3, pp. 83–85] , with the additional 3-adic unit cases treated in [12 Theorem 7.2, pp. 97–98] ; Tsushima gives an elementary proof [37 ] . For residue characteristic greater than three, crystalline Weil-group matrices are computed in [11 Theorems 1.7 and 3.13] . Those formulas are exact; the odd-prime period reformulation in [20 Section 2.2, Theorem 2.4, pp. 1235–1237] is instead modulo roots of unity. At two, [28 ] gives Hilbert-symbol formulas for local components at levels 2 r 2^r 2 r . These results must be applied with their stated hypotheses and reciprocity conventions; a Weil element which permutes character spaces does not supply an individual eigenvalue merely by taking one matrix entry. We do not implement these local evaluations at every mixed level. Our uniform algorithm, including at two, uses the principal-ideal construction (10.17); it computes exact Kummer inputs without first determining the exact Frobenius values.
Comparison with the ordinary determinant
Suppose p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ , n = p a n=p^a n = p a , and M = Δ n M=\Delta n M = Δ n . The ordinary curve is a quotient of the uniform one:
F M → ρ C M , Δ q ∘ ρ ↓ ↓ q C n , Δ = C n , Δ \begin{CD}
\mathcal{F}_M @>{\rho}>> C_{M,\Delta}\\
@V{q\circ\rho}VV @VV{q}V\\
C_{n,\Delta} @= C_{n,\Delta}
\end{CD} F M q ∘ ρ ↓ ⏐ C n , Δ ρ C M , Δ ↓ ⏐ q C n , Δ
Here ρ ( U , V ) = ( U , V n ) \rho(U,V)=(U,V^n) ρ ( U , V ) = ( U , V n ) and
q : C M , Δ ⟶ C n , Δ , ( Y , Z ) ⟼ ( Y M / n , Z ) . (95) q:C_{M,\Delta}\longrightarrow C_{n,\Delta},\qquad(Y,Z)\longmapsto(Y^{M/n},Z). \tag*{(95)} q : C M , Δ ⟶ C n , Δ , ( Y , Z ) ⟼ ( Y M / n , Z ) . (95)
Since q ∗ q ∗ = [ M / n ] q_*q^*=[M/n] q ∗ q ∗ = [ M / n ] , pullback on Jacobians is injective up to isogeny. It commutes with T T T and G Δ G_\Delta G Δ and identifies the ordinary line ( v , b ) (v,b) ( v , b ) with ( Δ v , b ) (\Delta v,b) ( Δ v , b ) . For v ∈ Z n × v\in\mathbb{Z}_n^\times v ∈ Z n × no such pair is excluded. In terms of the uniform RM blocks, the two determinant lines are
L v = det M Δ v + ⊗ ( det M Δ v − ) v , L Δ v ( 1 ) = det M Δ v + ⊗ ( det M Δ v − ) v . \mathcal{L}_v=\det M_{\Delta v}^+\otimes(\det M_{\Delta v}^-)^v,\qquad\mathcal{L}_{\Delta v}^{(1)}=\det M_{\Delta v}^+\otimes(\det M_{\Delta v}^-)^v. L v = det M Δ v + ⊗ ( det M Δ v − ) v , L Δ v ( 1 ) = det M Δ v + ⊗ ( det M Δ v − ) v .
The uniform construction also uses the other nonzero first indices and spans all Λ Σ ( 1 ) ⊗ Q \Lambda_\Sigma^{(1)}\otimes\mathbb{Q} Λ Σ ( 1 ) ⊗ Q ; the ordinary one addresses its additional τ \tau τ -anti-invariant part. In this subsection only, superscripts ord and gen distinguish the two inputs A q ( v ) A_q(v) A q ( v ) .
Proposition 10.7. At a prime q ∣ ℓ q \mid\ell q ∣ ℓ of F F F with ℓ ∤ M \ell\nmid M ℓ ∤ M , use Frob q \operatorname{Frob}_q Frob q in both constructions. Then B q ± ( Δ v ) ∈ E B_q^{\pm}(\Delta v) \in E B q ± ( Δ v ) ∈ E , and
A q o r d ( v ) = B q + ( Δ v ) B q − ( Δ v ) , A q g e n ( Δ v ) = N F / E ( B q + ( Δ v ) j ( B q − ( Δ v ) ) ) . A_q^{\mathrm{ord}}(v)=\frac{B_q^+(\Delta v)}{B_q^-(\Delta v)},\qquad A_q^{\mathrm{gen}}(\Delta v)=N_{F/E}\left(\frac{B_q^+(\Delta v)}{j(B_q^-(\Delta v))}\right). A q ord ( v ) = B q − ( Δ v ) B q + ( Δ v ) , A q gen ( Δ v ) = N F / E ( j ( B q − ( Δ v )) B q + ( Δ v ) ) . Moreover,
B q + ( Δ v ) j ( B q − ( Δ v ) ) j ( B q + ( Δ v ) ) B q − ( Δ v ) = A q o r d ( v ) 2 , \frac{B_q^+(\Delta v)j(B_q^-(\Delta v))}{j(B_q^+(\Delta v))B_q^-(\Delta v)}=A_q^{\mathrm{ord}}(v)^2, j ( B q + ( Δ v )) B q − ( Δ v ) B q + ( Δ v ) j ( B q − ( Δ v )) = A q ord ( v ) 2 , [ A q o r d ( v ) ] = [ B q + ( Δ v ) j ( B q − ( Δ v ) ) j ( B q + ( Δ v ) ) B q − ( Δ v ) ] ( n + 1 ) / 2 in E × / E × n . [A_q^{\mathrm{ord}}(v)]=\left[\frac{B_q^+(\Delta v)j(B_q^-(\Delta v))}{j(B_q^+(\Delta v))B_q^-(\Delta v)}\right]^{(n+1)/2}\quad\text{in }E^\times/E^{\times n}. [ A q ord ( v )] = [ j ( B q + ( Δ v )) B q − ( Δ v ) B q + ( Δ v ) j ( B q − ( Δ v )) ] ( n + 1 ) /2 in E × / E × n . Selecting the same anti-invariant character in the two constructions gives
β χ g e n = b n β χ o r d for some b ∈ E × ; \beta_\chi^{\mathrm{gen}}=b^n\beta_\chi^{\mathrm{ord}}\quad\text{for some }b\in E^\times; β χ gen = b n β χ ord for some b ∈ E × ; equivariant descent with that character gives the same cyclic extension of K K K .
Proof. The group Gal ( F / E ) \operatorname{Gal}(F/E) Gal ( F / E ) fixes the pulled-back ordinary index Δ v \Delta v Δ v and preserves the RM sign, so it fixes B q ± ( Δ v ) B_q^{\pm}(\Delta v) B q ± ( Δ v ) . The involution τ \tau τ interchanges the signs. Both determinants have degree r r r , and
B q ± ( Δ v ) j ( B q ± ( Δ v ) ) = q r , q = ∣ O F / q ∣ . B_q^{\pm}(\Delta v)j(B_q^{\pm}(\Delta v))=q^r,\qquad q=|\mathcal{O}_F/q|. B q ± ( Δ v ) j ( B q ± ( Δ v )) = q r , q = ∣ O F / q ∣.
Dividing these two identities proves the square relation. Since n n n is odd, raising to ( n + 1 ) / 2 (n+1)/2 ( n + 1 ) /2 gives the class identity. Uniqueness of the class selected by the character and descent with that full character give the last assertion. □
The quotient inside the uniform norm already lies in E E E on these ordinary blocks. Its norm is therefore its [ F : E ] [F:E] [ F : E ] th power, so
A q g e n ( Δ v ) τ ( A q g e n ( Δ v ) ) = A q o r d ( v ) 2 [ F : E ] . \frac{A_q^{\mathrm{gen}}(\Delta v)}{\tau(A_q^{\mathrm{gen}}(\Delta v))}=A_q^{\mathrm{ord}}(v)^{2[F:E]}. τ ( A q gen ( Δ v )) A q gen ( Δ v ) = A q ord ( v ) 2 [ F : E ] .
If p ∣ [ F : E ] p\mid[F:E] p ∣ [ F : E ] , this power cannot be inverted modulo n n n ; internal saturation removes the possible loss. For example, Δ = 13 \Delta=13 Δ = 13 , n = 3 n=3 n = 3 gives [ F : E ] = 6 [F:E]=6 [ F : E ] = 6 . At a prime above 79, A q o r d ( 1 ) A_q^{\mathrm{ord}}(1) A q ord ( 1 ) has Jacobi valuations ± 2 \pm2 ± 2 , but its power in (10.23) is a cube. The uniform input is the sixth power of the normalized determinant; its internal cube root is
( B q + ( Δ v ) j ( B q + ( Δ v ) ) ) 2 . \left(\frac{B_q^+(\Delta v)}{j(B_q^+(\Delta v))}\right)^2. ( j ( B q + ( Δ v )) B q + ( Δ v ) ) 2 .
Dividing this by its τ \tau τ -conjugate gives A q o r d ( v ) 4 A_q^{\mathrm{ord}}(v)^4 A q ord ( v ) 4 , representing [ A q o r d ( v ) ] [A_q^{\mathrm{ord}}(v)] [ A q ord ( v )] modulo cubes. Thus different initial values yield compatible selected classes. At two the same RM determinant applies; signatures and the coefficient enlargement still require their separate treatment. For exceptional primes, τ K \tau_K τ K need not extend to E E E while fixing μ n \mu_n μ n ; at two, division by two in the character decomposition is also unavailable. The ordinary hypothesis remains p ∤ 2 Δ p\nmid2\Delta p ∤ 2Δ .
General RM divisors
We prove that the normalized RM determinants of § 10 span the required divisor space. The explicit inputs constructed in § 10.5 have the same span by Proposition 10.5; thus the result applies to the computation at both good and bad primes. As in the ordinary case, this gives finite index; the next section passes to a basis modulo p a p^a p a .
Keep E ⊂ F = Q ( ζ M ) E \subset F = \mathbb{Q}(\zeta_M) E ⊂ F = Q ( ζ M ) , t = M / Δ t = M/\Delta t = M /Δ , and A Σ \mathcal{A}_{\Sigma} A Σ from § 10 . Recall
Λ Σ ( 1 ) = { D ∈ Z [ Σ ] : j D = − D } , U Σ ( 1 ) = { x ∈ R Σ × : x j ( x ) = 1 } , div E ( x ) = ∑ p ∈ Σ v p ( x ) [ p ] . \Lambda_{\Sigma}^{(1)} = \{D \in\mathbb{Z}[\Sigma] : jD = -D\}, \qquad U_{\Sigma}^{(1)} = \{x \in R_{\Sigma}^{\times} : xj(x) = 1\}, \qquad\operatorname{div}_E(x) = \sum_{\mathfrak{p}\in\Sigma} v_{\mathfrak{p}}(x)[\mathfrak{p}]. Λ Σ ( 1 ) = { D ∈ Z [ Σ ] : j D = − D } , U Σ ( 1 ) = { x ∈ R Σ × : x j ( x ) = 1 } , div E ( x ) = p ∈ Σ ∑ v p ( x ) [ p ] .
The valuations are normalized by v p ( p ) = 1 v_{\mathfrak{p}}(\mathfrak{p}) = 1 v p ( p ) = 1 . In contrast to the ordinary lattice, no second anti-invariance condition is imposed.
Theorem 11.1. For every finite Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable support Σ \Sigma Σ ,
span Q { div E A : A ∈ A Σ } = Λ Σ ( 1 ) ⊗ Q . (96) \operatorname{span}_{\mathbb{Q}}\{\operatorname{div}_E A : A \in\mathcal{A}_{\Sigma}\} = \Lambda_{\Sigma}^{(1)} \otimes\mathbb{Q}. \tag*{(96)} span Q { div E A : A ∈ A Σ } = Λ Σ ( 1 ) ⊗ Q . (96) Consequently ⟨ μ ( E ) , A Σ ⟩ \langle\mu(E),\mathcal{A}_{\Sigma}\rangle ⟨ μ ( E ) , A Σ ⟩ has finite index in U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) . Either RM sign may be used, provided all nonzero first indices v v v are used. The statement includes support above 2 and above every prime divisor of M M M .
§ 11.6 gives an alternative using C c : z c = x ( 1 − x ) C_c : z^c = x(1-x) C c : z c = x ( 1 − x ) and only good-reduction Jacobi sums. The proof below uses the fixed RM curve and its determinant divisors.
The proof has three steps.
The Shimura–Taniyama formula computes the Frobenius divisors, including at ramified primes (Proposition 11.2).
After normalization, their coefficient functions d v d_v d v span all odd functions on ( Z M ) × (\mathbb{Z}_M)^\times ( Z M ) × . Additive Fourier transform reduces this to the nonvanishing of L ( 1 , η ) L(1,\eta) L ( 1 , η ) for odd characters (Lemmas 11.4–11.6 ).
Projection to prime divisors and norm to E E E give (11.1) (§ 11.5 ).
The use of L L L -values at one is essential: their extra Euler factors never vanish, so imprimitive characters are included. The ordinary argument at zero needs its more restricted family of characters.
CM types and the Frobenius divisor
Put G = ( Z / M Z ) × G = (\mathbb{Z}/M\mathbb{Z})^\times G = ( Z / M Z ) × , with σ u ( ζ M ) = ζ M u \sigma_u(\zeta_M) = \zeta_M^u σ u ( ζ M ) = ζ M u , and fix one rational prime ℓ \ell ℓ below Σ \Sigma Σ . Let P ℓ \mathfrak{P}_{\ell} P ℓ be the chosen good-reduction prime of § 10 , lying over q ℓ \mathfrak{q}_{\ell} q ℓ in F F F . A good-reduction field containing F F F will be denoted by L L L in this subsection; thus it may be F F F when ℓ ∤ M \ell\nmid M ℓ ∤ M . Set
q ℓ = N P ℓ = ℓ f L , f L = f ( P ℓ / ℓ ) , f F = f ( q ℓ / ℓ ) , m ℓ = f L / f F . q_{\ell} = N\mathfrak{P}_{\ell} = \ell^{f_L}, \qquad f_L = f(\mathfrak{P}_{\ell}/\ell), \qquad f_F = f(\mathfrak{q}_{\ell}/\ell), \qquad m_{\ell} = f_L/f_F. q ℓ = N P ℓ = ℓ f L , f L = f ( P ℓ / ℓ ) , f F = f ( q ℓ / ℓ ) , m ℓ = f L / f F .
Thus m ℓ = f ( P ℓ / q ℓ ) m_{\ell} = f(\mathfrak{P}_{\ell}/\mathfrak{q}_{\ell}) m ℓ = f ( P ℓ / q ℓ ) is common to all eigenspaces of this Frobenius.
The function h v , b h_{v,b} h v , b in (90) is the indicator of the CM type of F v , b F_{v,b} F v , b , inflated to G G G . Indeed the Fermat holomorphic differential with positive indices ( r , s ) (r,s) ( r , s ) occurs precisely when r + s < M r+s<M r + s < M ; here ( r , s ) = ( u v , t u b ) (r,s)=(uv,tub) ( r , s ) = ( uv , t u b ) modulo M M M [21 pp. 1183–1184] . The deck maps act on the Jacobian by pushing forward divisors. Their tangent eigenvalues are the eigenvalues of their pullback on holomorphic differentials; this fixes the CM type used here. In particular
h v , b ( u ) + h v , b ( − u ) = 1. h_{v,b}(u) + h_{v,b}(-u) = 1. h v , b ( u ) + h v , b ( − u ) = 1.
This fixes the sign of the normalized divisor below.
Proposition 11.2. The determinant defined in § 10.2 has divisor
div F B q ℓ + ( v ) = m ℓ H v , ℓ , \operatorname{div}_F B_{\mathfrak{q}_{\ell}}^{+}(v) = m_{\ell} H_{v,\ell}, div F B q ℓ + ( v ) = m ℓ H v , ℓ , where H v , ℓ H_{v,\ell} H v , ℓ is given by (91) . More precisely, for each admissible ( v , b ) (v,b) ( v , b ) ,
div F B q ℓ ( v , b ) = m ℓ ∑ u ∈ G h v , b ( u ) [ σ u − 1 q ℓ ] . (97) \operatorname{div}_F B_{\mathfrak{q}_{\ell}}(v,b) = m_{\ell} \sum_{u\in G} h_{v,b}(u)[\sigma_u^{-1}\mathfrak{q}_{\ell}]. \tag*{(97)} div F B q ℓ ( v , b ) = m ℓ u ∈ G ∑ h v , b ( u ) [ σ u − 1 q ℓ ] . (97)
Repeated occurrences of a prime on the right are added. This formula is valid whether or not ℓ \ell ℓ ramifies in F F F .
Proof. Fix the ℓ \ell ℓ -adic embedding inducing P ℓ \mathfrak{P}_{\ell} P ℓ and put F c = F v , b F_c = F_{v,b} F c = F v , b . For a prime m ∣ ℓ \mathfrak{m} \mid\ell m ∣ ℓ of F c F_c F c , let Φ m \Phi_{\mathfrak{m}} Φ m be the CM-type embeddings inducing m \mathfrak{m} m . The local Shimura–Taniyama formula is
v m ( B q ℓ ( v , b ) ) v m ( q ℓ ) = ∣ Φ m ∣ [ ( F c ) m : Q ℓ ] (98) \frac{v_{\mathfrak{m}}(B_{q_{\ell}}(v,b))}{v_{\mathfrak{m}}(q_{\ell})}
=
\frac{|\Phi_{\mathfrak{m}}|}{[(F_c)_{\mathfrak{m}}:\mathbb{Q}_{\ell}]} \tag*{(98)} v m ( q ℓ ) v m ( B q ℓ ( v , b )) = [( F c ) m : Q ℓ ] ∣ Φ m ∣ (98)
It holds without an unramifiedness hypothesis [36, Section 4, Lemma 5, formula (10); Section 5, Theorem 3]. The numerator and denominator are the dimension and height of the corresponding CM ℓ \ell ℓ -divisible group. The formula therefore also applies to an imprimitive CM type with the rank-one F c F_c F c -action of § 10.3 .
Here is its precise conversion to divisors in F F F . Write e c e_c e c , f c f_c f c for the local ramification index and residue degree in F c F_c F c , and e F e_F e F , f F f_F f F for those in F F F . Since v m ( q ℓ ) = e c f L v_{\mathfrak{m}}(q_{\ell})=e_cf_L v m ( q ℓ ) = e c f L , (98) gives
v m ( B q ℓ ( v , b ) ) = f L f c ∣ Φ m ∣ . v_{\mathfrak{m}}(B_{q_{\ell}}(v,b))=\frac{f_L}{f_c}|\Phi_{\mathfrak{m}}|. v m ( B q ℓ ( v , b )) = f c f L ∣ Φ m ∣.
Extension to F F F multiplies this coefficient by e F / e c e_F/e_c e F / e c . The embeddings of F c F_c F c in the relevant local decomposition coset each have e F f F / ( e c f c ) e_Ff_F/(e_cf_c) e F f F / ( e c f c ) lifts to F F F . Consequently the coefficient of that prime in the right-hand side of (97) is
f L f F e F f F e c f c ∣ Φ m ∣ = e F e c f L f c ∣ Φ m ∣ , \frac{f_L}{f_F}\frac{e_Ff_F}{e_cf_c}|\Phi_{\mathfrak{m}}|
=
\frac{e_F}{e_c}\frac{f_L}{f_c}|\Phi_{\mathfrak{m}}|, f F f L e c f c e F f F ∣ Φ m ∣ = e c e F f c f L ∣ Φ m ∣ ,
which is the required extended valuation. Summing over b ∈ I v + b \in I_v^+ b ∈ I v + proves the determinant formula. The same m ℓ m_{\ell} m ℓ works for every first index, although the orbit fields may differ. □
For clarity, if D ℓ ⊂ G D_{\ell} \subset G D ℓ ⊂ G is the decomposition group of q ℓ q_{\ell} q ℓ , the coefficient at σ a − 1 q ℓ \sigma_a^{-1}q_{\ell} σ a − 1 q ℓ is
m ℓ ∑ u ∈ a D ℓ h v , b ( u ) . m_{\ell}\sum_{u\in aD_{\ell}}h_{v,b}(u). m ℓ u ∈ a D ℓ ∑ h v , b ( u ) .
Thus complete splitting is unnecessary: the CM type is summed over each decomposition coset. At ℓ ∤ M \ell\nmid M ℓ ∤ M , with L = F L=F L = F , this is exactly the Jacobi ideal formula of Lemma 6.2 in the positive-character convention of (88) . At ℓ ∣ M \ell\mid M ℓ ∣ M , the same statement comes from CM reduction, not from a character of order M M M in characteristic ℓ \ell ℓ .
The normalized divisor functions
Fix the positive RM sign and write
S + = { b ∈ Z Δ × : χ Δ ( b ) = 1 } , s = ∣ S + ∣ = φ ( Δ ) / 2. S_+=\{b\in\mathbb{Z}_{\Delta}^{\times}:\chi_{\Delta}(b)=1\},
\qquad
s=|S_+|=\varphi(\Delta)/2. S + = { b ∈ Z Δ × : χ Δ ( b ) = 1 } , s = ∣ S + ∣ = φ ( Δ ) /2.
The set is nonempty and stable under b ↦ − b b\mapsto-b b ↦ − b because K K K is real quadratic. Introduce the odd periodic Bernoulli function
B 1 ( x ) = { { x } − 1 2 , x ∉ Z , 0 , x ∈ Z . B_1(x)=
\begin{cases}
\{x\}-\frac{1}{2}, & x\notin\mathbb{Z},\\
0, & x\in\mathbb{Z}.
\end{cases} B 1 ( x ) = { { x } − 2 1 , 0 , x ∈ / Z , x ∈ Z .
Its value at integers matters: it removes the pairs for which the Fermat eigenspace is zero. For v ∈ Z M v\in\mathbb{Z}_M v ∈ Z M define a function on G G G by
d v ( u ) = 2 ∑ b ∈ S + ( B 1 ( v u / M ) + B 1 ( b u / Δ ) − B 1 ( ( v + t b ) u / M ) ) . (99) d_v(u)=2\sum_{b\in S_+}\left(B_1(vu/M)+B_1(bu/\Delta)-B_1((v+tb)u/M)\right). \tag*{(99)} d v ( u ) = 2 b ∈ S + ∑ ( B 1 ( v u / M ) + B 1 ( b u /Δ ) − B 1 (( v + t b ) u / M ) ) . (99)
Each d v d_v d v has rational values, satisfies d v ( − u ) = − d v ( u ) d_v(-u)=-d_v(u) d v ( − u ) = − d v ( u ) , and d 0 = 0 d_0=0 d 0 = 0 .
Lemma 11.3. For 1 ≤ v < M 1\leq v<M 1 ≤ v < M ,
div F ( B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ) = − m ℓ ∑ u ∈ G d v ( u ) [ σ a − 1 q ℓ ] . (100) \operatorname{div}_F\left(\frac{B_{q_{\ell}^{+}}(v)}{j(B_{q_{\ell}^{+}}(v))}\right)
=
-m_{\ell}\sum_{u\in G}d_v(u)[\sigma_a^{-1}q_{\ell}]. \tag*{(100)} div F ( j ( B q ℓ + ( v )) B q ℓ + ( v ) ) = − m ℓ u ∈ G ∑ d v ( u ) [ σ a − 1 q ℓ ] . (100)
Proof. For nonintegral a , b , a + b a,b,a+b a , b , a + b , put f ( a , b ) = { a } + { b } − { a + b } ∈ { 0 , 1 } f(a,b)=\{a\}+\{b\}-\{a+b\}\in\{0,1\} f ( a , b ) = { a } + { b } − { a + b } ∈ { 0 , 1 } . The CM-type indicator is h = 1 − f h=1-f h = 1 − f , and
h ( a , b ) − h ( − a , − b ) = − 2 ( B 1 ( a ) + B 1 ( b ) − B 1 ( a + b ) ) . h(a,b)-h(-a,-b)=-2\bigl(B_1(a)+B_1(b)-B_1(a+b)\bigr). h ( a , b ) − h ( − a , − b ) = − 2 ( B 1 ( a ) + B 1 ( b ) − B 1 ( a + b ) ) .
Subtract the divisor of j ( B q + ( v ) ) j(B^+_{\mathfrak q}(v)) j ( B q + ( v )) from that of B q + ( v ) B^+_{\mathfrak q}(v) B q + ( v ) in Proposition 11.2. This gives the displayed formula for the admissible terms. If v + t b = 0 v+tb=0 v + t b = 0 in Z M \mathbf Z_M Z M , then B 1 ( v u / M ) + B 1 ( b u / Δ ) = 0 B_1(vu/M)+B_1(bu/\Delta)=0 B 1 ( v u / M ) + B 1 ( b u /Δ ) = 0 and the third term is zero. If v = 0 v=0 v = 0 , the first term is zero and the other two cancel. Thus including all b ∈ S + b\in S_+ b ∈ S + and all v ∈ Z M v\in\mathbf Z_M v ∈ Z M introduces no contribution from an absent eigenspace. □ \square □
It remains to show that the coefficient rows d v d_v d v span all odd functions on G G G . This is a rational spanning assertion, distinct from the integral basis calculation in Section 12.
Fourier transform
Fourier transform is taken in the additive index v ∈ Z M v\in\mathbf Z_M v ∈ Z M ; u u u remains in the multiplicative group G G G . Put
d ^ k ( u ) = ∑ v ∈ Z M d v ( u ) ζ M − k v , k ∈ Z M . \widehat d_k(u)=\sum_{v\in\mathbf Z_M}d_v(u)\zeta_M^{-kv},\qquad k\in\mathbf Z_M. d k ( u ) = v ∈ Z M ∑ d v ( u ) ζ M − k v , k ∈ Z M .
The transform is invertible after extension of scalars to C \mathbb C C , so the families ( d v ) v (d_v)_v ( d v ) v and ( d ^ k ) k (\widehat d_k)_k ( d k ) k have the same row span there. The rows indexed by k ∈ G k\in G k ∈ G will suffice.
Lemma 11.4. For k , u ∈ G k,u\in G k , u ∈ G ,
d ^ k ( u ) = i λ + ( k ) cot ( π k u − 1 / M ) , λ + ( k ) = s − ∑ b ∈ S + ζ Δ k b > 0. (101) \widehat d_k(u)=i\lambda_+(k)\cot(\pi ku^{-1}/M),\qquad\lambda_+(k)=s-\sum_{b\in S_+}\zeta_\Delta^{kb}>0. \tag*{(101)} d k ( u ) = i λ + ( k ) cot ( π k u − 1 / M ) , λ + ( k ) = s − b ∈ S + ∑ ζ Δ k b > 0. (101)
Proof. Pairing b b b with − b -b − b makes ∑ b ∈ S + B 1 ( b u / Δ ) = 0 \sum_{b\in S_+}B_1(bu/\Delta)=0 ∑ b ∈ S + B 1 ( b u /Δ ) = 0 . For k ≠ 0 k\ne0 k = 0 in Z M \mathbf Z_M Z M , differentiation of a finite geometric sum gives
∑ v ∈ Z M B 1 ( v / M ) ζ M − k v = 1 + ζ M k 2 ( 1 − ζ M k ) = i 2 cot ( π k / M ) . \sum_{v\in\mathbf Z_M}B_1(v/M)\zeta_M^{-kv}
=\frac{1+\zeta_M^k}{2(1-\zeta_M^k)}
=\frac{i}{2}\cot(\pi k/M). v ∈ Z M ∑ B 1 ( v / M ) ζ M − k v = 2 ( 1 − ζ M k ) 1 + ζ M k = 2 i cot ( π k / M ) .
In the first term of (11.4), replace v v v by u − 1 v u^{-1}v u − 1 v . In the last term also translate v v v by t b tb t b ; that translation contributes ζ Δ k b \zeta_\Delta^{kb} ζ Δ k b . This proves the identity. Its multiplier is real, and
λ + ( k ) = ∑ b ∈ S + ( 1 − cos ( 2 π k b / Δ ) ) > 0 , \lambda_+(k)=\sum_{b\in S_+}\bigl(1-\cos(2\pi kb/\Delta)\bigr)>0, λ + ( k ) = b ∈ S + ∑ ( 1 − cos ( 2 π k b /Δ ) ) > 0 ,
since k k k and b b b are units modulo Δ \Delta Δ . □ \square □
The multiplier also displays the same quadratic Gauss sum as the RM endomorphism in Section 5. Splitting the indicator of S + S_+ S + as ( 1 + χ Δ ) / 2 (1+\chi_\Delta)/2 ( 1 + χ Δ ) /2 gives
λ + ( k ) = φ ( Δ ) − μ ( Δ ) − χ Δ ( k ) Δ 2 . (102) \lambda_+(k)=\frac{\varphi(\Delta)-\mu(\Delta)-\chi_\Delta(k)\sqrt{\Delta}}{2}. \tag*{(102)} λ + ( k ) = 2 φ ( Δ ) − μ ( Δ ) − χ Δ ( k ) Δ . (102)
Here μ ( Δ ) \mu(\Delta) μ ( Δ ) is the Möbius function: the ordinary character sum is the Ramanujan sum c Δ ( k ) = μ ( Δ ) c_\Delta(k)=\mu(\Delta) c Δ ( k ) = μ ( Δ ) for k k k a unit, and the quadratic sum is χ Δ ( k ) Δ \chi_\Delta(k)\sqrt{\Delta} χ Δ ( k ) Δ [5 Chapter 1] . For the negative RM sign the last minus sign becomes a plus sign. The elementary positivity above handles both; the remaining nonvanishing question concerns the cotangent matrix.
Character eigenvalues and nonvanishing
Consider the endomorphism
T : C G ⟶ C G , ( T f ) ( k ) = ∑ u ∈ G cot ( π k u − 1 / M ) f ( u ) . \mathcal{T} : \mathbb{C}^{G} \longrightarrow\mathbb{C}^{G}, \qquad(\mathcal{T}f)(k) = \sum_{u\in G} \cot(\pi k u^{-1}/M)f(u). T : C G ⟶ C G , ( T f ) ( k ) = u ∈ G ∑ cot ( π k u − 1 / M ) f ( u ) .
Multiplicative characters η : G → C × \eta:G\to\mathbb{C}^{\times} η : G → C × form a basis of C G \mathbb{C}^{G} C G . Substituting a = k u − 1 a=ku^{-1} a = k u − 1 gives the exact eigenvalue formula
T η = κ η η , κ η = ∑ a ∈ G η ( a ) − 1 cot ( π a / M ) . (103) \mathcal{T}\eta=\kappa_{\eta}\eta,\qquad\kappa_{\eta}=\sum_{a\in G}\eta(a)^{-1}\cot(\pi a/M). \tag*{(103)} T η = κ η η , κ η = a ∈ G ∑ η ( a ) − 1 cot ( π a / M ) . (103)
Even characters have eigenvalue zero, by pairing a a a and − a -a − a . For odd characters, the following classical fact is the analytic input to the spanning theorem [18 Proposition 3, pp. 134–135] .
Lemma 11.5. For every odd Dirichlet character η \eta η modulo M M M ,
∑ a ∈ G η ( a ) cot ( π a / M ) = 2 M π L ( 1 , η ) ≠ 0. (104) \sum_{a\in G}\eta(a)\cot(\pi a/M)=\frac{2M}{\pi}L(1,\eta)\ne0. \tag*{(104)} a ∈ G ∑ η ( a ) cot ( π a / M ) = π 2 M L ( 1 , η ) = 0. (104)
Thus κ η = ( 2 M / π ) L ( 1 , η − 1 ) ≠ 0 \kappa_{\eta}=(2M/\pi)L(1,\eta^{-1})\ne0 κ η = ( 2 M / π ) L ( 1 , η − 1 ) = 0 . No primitivity assumption on η \eta η is needed.
Proof. Sum the partial-fraction expansion
π M cot ( π a / M ) = lim R → ∞ ∑ h = − R R 1 a + h M \frac{\pi}{M}\cot(\pi a/M)=\lim_{R\to\infty}\sum_{h=-R}^{R}\frac{1}{a+hM} M π cot ( π a / M ) = R → ∞ lim h = − R ∑ R a + h M 1
against η ( a ) \eta(a) η ( a ) , extended by zero at nonunits. Oddness pairs the positive and negative terms, giving 2 L ( 1 , η ) 2L(1,\eta) 2 L ( 1 , η ) . Dirichlet’s nonvanishing theorem applies [2, p. 149; Chapter 12]. If η ∘ \eta^{\circ} η ∘ of conductor f f f induces η \eta η , then
L ( 1 , η ) = L ( 1 , η ∘ ) ∏ r ∣ M r ∤ f ( 1 − η ∘ ( r ) / r ) . (105) L(1,\eta)=L(1,\eta^{\circ})\prod_{\substack{r\mid M\\r\nmid f}}(1-\eta^{\circ}(r)/r). \tag*{(105)} L ( 1 , η ) = L ( 1 , η ∘ ) r ∣ M r ∤ f ∏ ( 1 − η ∘ ( r ) / r ) . (105)
Each extra factor is nonzero, since ∣ η ∘ ( r ) / r ∣ < 1 |\eta^{\circ}(r)/r|<1 ∣ η ∘ ( r ) / r ∣ < 1 [2 p. 262] . □
Lemma 11.6. The functions d v d_v d v , 1 ≤ v < M 1\le v<M 1 ≤ v < M , span
Q [ G ] − = { f : G → Q : f ( − u ) = − f ( u ) } . \mathbb{Q}[G]^{-}=\{f:G\to\mathbb{Q}:f(-u)=-f(u)\}. Q [ G ] − = { f : G → Q : f ( − u ) = − f ( u )} .
Proof. By (103) and Lemma 11.5, T \mathcal{T} T is invertible on the odd functions and zero on the even functions. Its rank is therefore ∣ G ∣ / 2 |G|/2 ∣ G ∣/2 . The matrix ( d k ( u ) ^ ) k , u ∈ G (\widehat{d_k(u)})_{k,u\in G} ( d k ( u ) ) k , u ∈ G is obtained from the matrix of T \mathcal{T} T by multiplying row k k k by the nonzero scalar i λ + ( k ) i\lambda_{+}(k) i λ + ( k ) . It has the same rank ∣ G ∣ / 2 |G|/2 ∣ G ∣/2 . These rows are linear combinations of all the rows d v d_v d v . Hence the latter have rank at least ∣ G ∣ / 2 |G|/2 ∣ G ∣/2 over C \mathbb{C} C ; their oddness gives the reverse inequality. Since all their entries are rational, their rank over Q \mathbb{Q} Q is also ∣ G ∣ / 2 |G|/2 ∣ G ∣/2 . Finally d 0 = 0 d_0=0 d 0 = 0 , so that row may be discarded. □
Remark 11.7. For the ordinary construction, the divisor eigenvalues are proportional to L ( 0 , ( χ Δ ψ ) − 1 ) L(0,(\chi_{\Delta}\psi)^{-1}) L ( 0 , ( χ Δ ψ ) − 1 ) , where ψ \psi ψ is odd modulo p a p^{a} p a . The primitive conductor contains every prime dividing Δ p a \Delta p^{a} Δ p a , so no additional Euler factor occurs. The uniform construction must allow every odd character modulo M M M ; its extra factors 1 − η ∘ ( r ) 1-\eta^{\circ}(r) 1 − η ∘ ( r ) at zero may vanish. For example, the odd quadratic character χ − 8 \chi_{-8} χ − 8 of conductor 8 8 8 , inflated to modulus 24 24 24 , has
L ( 0 , η ) = 0 , L ( 1 , η ) = 2 3 L ( 1 , η ∘ ) ≠ 0. L(0,\eta)=0,\qquad L(1,\eta)=\frac{2}{3}L(1,\eta^{\circ})\ne0. L ( 0 , η ) = 0 , L ( 1 , η ) = 3 2 L ( 1 , η ∘ ) = 0.
Fourier transform in all first indices replaces this calculation by the cotangent operator, with nonzero factors at one. For primitive odd characters the functional equation (63) relates the two descriptions. All first indices suffice; each need not be indispensable.
Proof of Theorem 11.1
Proof of Theorem 11.1. For the chosen prime q ℓ q_{\ell} q ℓ , define
P ℓ : Q G ⟶ ⨁ q ∣ ℓ Q [ q ] , f ⟼ ∑ u ∈ G f ( u ) [ σ u − 1 q ℓ ] . \mathcal{P}_{\ell}: \mathbb{Q}^{G} \longrightarrow\bigoplus_{q\mid\ell}\mathbb{Q}[q],\qquad f\longmapsto\sum_{u\in G}f(u)[\sigma_{u}^{-1}q_{\ell}]. P ℓ : Q G ⟶ q ∣ ℓ ⨁ Q [ q ] , f ⟼ u ∈ G ∑ f ( u ) [ σ u − 1 q ℓ ] .
It is surjective and intertwines u ↦ − u u\mapsto-u u ↦ − u with complex conjugation. Applying ( 1 − j ) / 2 (1-j)/2 ( 1 − j ) /2 to a preimage therefore proves surjectivity on odd parts.
Lemma 11.6 and (100) now show that the divisors of
B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ( 1 ≤ v < M ) \frac{B_{q_{\ell}}^{+}(v)}{j(B_{q_{\ell}}^{+}(v))}\qquad(1\leq v<M) j ( B q ℓ + ( v )) B q ℓ + ( v ) ( 1 ≤ v < M )
span all odd divisors over ℓ \ell ℓ in F F F . The factor m ℓ m_{\ell} m ℓ is a nonzero integer and does not change their rational span. Equivalently, projection selects exactly the characters of G G G trivial on D ℓ D_{\ell} D ℓ ; this is the same decomposition-group step as in the ordinary proof. If j ∈ D ℓ j\in D_{\ell} j ∈ D ℓ , every prime over ℓ \ell ℓ is fixed by j j j and the odd target is zero. Otherwise its dimension is [ G : D ℓ ] / 2 [G:D_{\ell}]/2 [ G : D ℓ ] /2 . This includes decomposition groups containing wild inertia.
For p \mathfrak{p} p in E E E and q \mathfrak{q} q above it in F F F , norm and extension of divisors are given by
N F / E [ q ] = f ( q / p ) [ p ] , ext F / E [ p ] = ∑ q ∣ p e ( q / p ) [ q ] . N_{F/E}[\mathfrak{q}]=f(\mathfrak{q}/\mathfrak{p})[\mathfrak{p}],\qquad\operatorname{ext}_{F/E}[\mathfrak{p}]=\sum_{\mathfrak{q}\mid\mathfrak{p}}e(\mathfrak{q}/\mathfrak{p})[\mathfrak{q}]. N F / E [ q ] = f ( q / p ) [ p ] , ext F / E [ p ] = q ∣ p ∑ e ( q / p ) [ q ] .
They commute with j j j , and N F / E ext F / E = [ F : E ] N_{F/E}\operatorname{ext}_{F/E}=[F:E] N F / E ext F / E = [ F : E ] . Hence norm is surjective on odd divisors after tensoring with Q \mathbb{Q} Q . Since
A q ℓ ( v ) = N F / E ( B q ℓ + ( v ) j ( B q ℓ + ( v ) ) ) , A_{q_{\ell}}(v)=N_{F/E}\left(\frac{B_{q_{\ell}}^{+}(v)}{j(B_{q_{\ell}}^{+}(v))}\right), A q ℓ ( v ) = N F / E ( j ( B q ℓ + ( v )) B q ℓ + ( v ) ) ,
it gives (11.1), one rational prime at a time.
By Lemma 3.5 , the divisor map on U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) has kernel μ ( E ) \mu(E) μ ( E ) and image a full-rank sublattice of Λ Σ ( 1 ) \Lambda_{\Sigma}^{(1)} Λ Σ ( 1 ) . The spanning just proved therefore gives finite index in the unit group after adjoining μ ( E ) \mu(E) μ ( E ) .
No class-group hypothesis is needed for this theorem; target membership was proved in § 4 . § 12 removes the possible p p p -primary index. In particular, the rational projection ( 1 − j ) / 2 (1-j)/2 ( 1 − j ) /2 used above is not an operation modulo 2 a 2^{a} 2 a .
An alternative using diagonal Fermat quotients
The same divisor components can be obtained using only good reduction, if the curve is allowed to vary with the conductor. For c > 2 c>2 c > 2 , let C c C_c C c be the smooth projective curve with affine equation z c = x ( 1 − x ) z^c=x(1-x) z c = x ( 1 − x ) . The map
F c ⟶ C c , ( U , V ) ⟼ ( U c , U V ) \mathcal{F}_c\longrightarrow C_c,\qquad(U,V)\longmapsto(U^c,UV) F c ⟶ C c , ( U , V ) ⟼ ( U c , U V )
selects the Fermat character pairs ( u , u ) (u,u) ( u , u ) : both character arguments are equal. For u ∈ Z c × u\in\mathbb{Z}_c^{\times} u ∈ Z c × , the indices u , u , 2 u u,u,2u u , u , 2 u are nonzero in Z c \mathbb{Z}_c Z c . At a prime ℓ ∤ c \ell\nmid c ℓ ∤ c these curves have good reduction and their corresponding Frobenius values are the signed Jacobi sums, with the convention of Lemma 5.4.
Put F c = Q ( ζ c ) F_c=\mathbb{Q}(\zeta_c) F c = Q ( ζ c ) and E c = E ∩ F c E_c=E\cap F_c E c = E ∩ F c . For each divisor c > 2 c>2 c > 2 of M M M and each rational support prime ℓ ∤ c \ell\nmid c ℓ ∤ c , choose q ∣ ℓ \mathfrak{q}\mid\ell q ∣ ℓ in F c F_c F c . With the inverse residue character λ \lambda λ of order c c c from Lemma 6.2 , define
J c , q = J ( λ , λ ) = ∑ x ∈ O F c / q λ ( x ) λ ( 1 − x ) , J_{c,\mathfrak{q}}=J(\lambda,\lambda)=\sum_{x\in\mathcal{O}_{F_c}/\mathfrak{q}}\lambda(x)\lambda(1-x), J c , q = J ( λ , λ ) = x ∈ O F c / q ∑ λ ( x ) λ ( 1 − x ) ,
A q diag ( c ) = N F c / E c ( J c , q j ( J c , q ) ) ∈ E c × ⊂ E × , (106) A_{\mathfrak{q}}^{\operatorname{diag}}(c)=N_{F_c/E_c}\left(\frac{J_{c,\mathfrak{q}}}{j(J_{c,\mathfrak{q}})}\right)\in E_c^{\times}\subset E^{\times}, \tag*{(106)} A q diag ( c ) = N F c / E c ( j ( J c , q ) J c , q ) ∈ E c × ⊂ E × , (106)
J Σ = { σ ( A q diag ( c ) ) : c ∣ M , c > 2 , ℓ lies below Σ , ℓ ∤ c , σ ∈ Gal ( E / Q ) } . \mathcal{J}_{\Sigma}=\{\sigma(A_{\mathfrak{q}}^{\operatorname{diag}}(c)):c\mid M,\ c>2,\ \ell\text{ lies below }\Sigma,\ \ell\nmid c,\ \sigma\in\operatorname{Gal}(E/\mathbb{Q})\}. J Σ = { σ ( A q diag ( c )) : c ∣ M , c > 2 , ℓ lies below Σ , ℓ ∤ c , σ ∈ Gal ( E / Q )} .
Here J J J is exactly the sum defined in (44) . All these values lie in U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) .
Theorem 11.8. One has
span Q { div E A : A ∈ J Σ } = Λ Σ ( 1 ) ⊗ Q . \operatorname{span}_{\mathbb{Q}}\{\operatorname{div}_{E} A : A \in\mathcal{J}_{\Sigma}\} = \Lambda_{\Sigma}^{(1)} \otimes\mathbb{Q}. span Q { div E A : A ∈ J Σ } = Λ Σ ( 1 ) ⊗ Q . Thus ⟨ μ ( E ) , J Σ ⟩ \langle\mu(E),\mathcal{J}_{\Sigma}\rangle ⟨ μ ( E ) , J Σ ⟩ , after internal p p p -saturation, also gives W Σ ( 1 ) W_{\Sigma}^{(1)} W Σ ( 1 ) . Every Jacobi sum in this construction is evaluated at a prime of good reduction, including when p = 2 p=2 p = 2 or the support contains primes dividing M M M .
Proof. Fix a rational support prime ℓ \ell ℓ . The prime-divisor representation in E E E , after extension to C \mathbb{C} C , contains exactly the characters of Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) trivial on its decomposition group D ℓ D_{\ell} D ℓ , each with multiplicity one. Let η \eta η be such an odd character, and let η ∘ \eta^{\circ} η ∘ be its primitive Dirichlet character of conductor c c c . It is trivial on inertia at ℓ \ell ℓ ; therefore
ℓ ∤ c , η ∘ ( ℓ ) = 1. \ell\nmid c,\qquad\eta^{\circ}(\ell)=1. ℓ ∤ c , η ∘ ( ℓ ) = 1.
Also c ∣ M c\mid M c ∣ M and c > 2 c>2 c > 2 . This is the reason to use C c C_c C c at this prime: its level c c c is prime to the residue characteristic.
By Lemma 6.2 , the divisor of J c , q J_{c,q} J c , q is obtained by projecting the function
f c ( u ) = 2 { u / c } − { 2 u / c } , u ∈ Z c × , f_c(u)=2\{u/c\}-\{2u/c\},\qquad u\in\mathbb{Z}_c^{\times}, f c ( u ) = 2 { u / c } − { 2 u / c } , u ∈ Z c × ,
to prime divisors. Its primitive odd coefficient is
∑ u ∈ Z c × η ∘ ( u ) f c ( u ) = − L ( 0 , η ∘ ) { 2 − η ∘ ( 2 ) − 1 , c odd , 2 , c even . (107) \sum_{u\in\mathbb{Z}_c^{\times}}\eta^{\circ}(u)f_c(u)
=-L(0,\eta^{\circ})
\begin{cases}
2-\eta^{\circ}(2)^{-1}, & c\ \text{odd},\\
2, & c\ \text{even}.
\end{cases} \tag*{(107)} u ∈ Z c × ∑ η ∘ ( u ) f c ( u ) = − L ( 0 , η ∘ ) { 2 − η ∘ ( 2 ) − 1 , 2 , c odd , c even . (107)
Indeed ∑ u η ∘ ( u ) { u / c } = − L ( 0 , η ∘ ) \sum_u\eta^{\circ}(u)\{u/c\}=-L(0,\eta^{\circ}) ∑ u η ∘ ( u ) { u / c } = − L ( 0 , η ∘ ) . If c c c is odd, substitute 2 u 2u 2 u in the second sum. If c c c is even, primitivity implies 4 ∣ c 4\mid c 4 ∣ c and the two lifts u , u + c / 2 u,u+c/2 u , u + c /2 from Z c / 2 × \mathbb{Z}_{c/2}^{\times} Z c /2 × have opposite character values; their contributions to { 2 u / c } \{2u/c\} { 2 u / c } cancel. The right-hand side is nonzero by Lemma 6.4 . Passing to J / j ( J ) J/j(J) J / j ( J ) multiplies an odd coefficient by two.
Projection to divisors does not kill this coefficient because η ∘ ( ℓ ) = 1 \eta^{\circ}(\ell)=1 η ∘ ( ℓ ) = 1 . Moreover the field cut out by η \eta η lies in both E E E and F c F_c F c , hence in E c E_c E c . Norm to E c E_c E c therefore preserves its component, up to a nonzero integer factor; extension of divisors from E c E_c E c to E E E does likewise. The conjugates of A q diag ( c ) A^{\operatorname{diag}}_q(c) A q diag ( c ) consequently span a space containing the η \eta η -component. Every odd component is obtained in this way. The equality of rational spans follows, and the finite-index and saturation assertions follow as above. □ \square □
At ℓ = 2 \ell=2 ℓ = 2 , every contributing conductor c c c is odd and η ∘ ( 2 ) = 1 \eta^{\circ}(2)=1 η ∘ ( 2 ) = 1 , so the extra factor in (107) is exactly one. If j ∈ D ℓ j\in D_{\ell} j ∈ D ℓ , there is no odd divisor component at that prime. Thus no bad-reduction Frobenius evaluation or principalization is required for this alternative. It uses several curves, choosing their conductors according to the components that actually occur. The fixed RM construction instead keeps Δ ∣ M \Delta\mid M Δ ∣ M , so primes dividing Δ \Delta Δ cannot be avoided in its defining level; § 10.5 handles them while preserving the quadratic multiplication.
A basis of the general Kummer module
The RM divisor theorem gives a finite-index subgroup of U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) . We now apply the same p p p -saturation as in § 7 ; the additional basis class comes from roots of unity. The procedure is standard [6 Section 4.3] . The steps parallel § 7 : compute the divisor lattice, extract the missing p p p th roots in E E E , and read off a basis modulo ν \nu ν . We give the corresponding statements, including the torsion and the criterion for omitting root extraction.
Let n = p a n=p^{a} n = p a , let E / Q E/\mathbb{Q} E / Q be abelian CM with K , μ n ⊂ E K,\mu_{n}\subset E K , μ n ⊂ E , and fix a finite Galois-stable support Σ \Sigma Σ . The module is W Σ ( 1 ) ( E , n ) W_{\Sigma}^{(1)}(E,n) W Σ ( 1 ) ( E , n ) of § 3 . Target membership, with the signature correction at two, is supplied separately by Theorem 4.2 .
What the divisor calculation supplies
For computation, take the Jacobi values at ℓ ∤ M \ell\nmid M ℓ ∤ M and the explicit principalized inputs of § 10.5 at ℓ ∣ M \ell\mid M ℓ ∣ M , and enumerate them as A 1 , … , A m A_{1},\ldots,A_{m} A 1 , … , A m . One may also use actual Frobenius values if available, or the C c C_{c} C c inputs of § 11.6 . In every case these are specified elements of E E E , with computable ideals; no unknown Frobenius scalar occurs in the following algorithms. Put
Γ = ⟨ μ ( E ) , A 1 , … , A m ⟩ ⊂ U Σ ( 1 ) , L Γ = ∑ i = 1 m Z div E ( A i ) , \Gamma=\langle\mu(E),A_{1},\ldots,A_{m}\rangle\subset U_{\Sigma}^{(1)},\qquad L_{\Gamma}=\sum_{i=1}^{m}\mathbb{Z}\operatorname{div}_{E}(A_{i}), Γ = ⟨ μ ( E ) , A 1 , … , A m ⟩ ⊂ U Σ ( 1 ) , L Γ = i = 1 ∑ m Z div E ( A i ) ,
L Γ ⊆ L Σ ( 1 ) ⊆ Λ Σ ( 1 ) , B = [ Λ Σ ( 1 ) : L Γ ] , I = [ U Σ ( 1 ) : Γ ] , r = rank Λ Σ ( 1 ) . L_{\Gamma}\subseteq L_{\Sigma}^{(1)}\subseteq\Lambda_{\Sigma}^{(1)},\qquad B=[\Lambda_{\Sigma}^{(1)}:L_{\Gamma}],\qquad I=[U_{\Sigma}^{(1)}:\Gamma],\qquad r=\operatorname{rank}\Lambda_{\Sigma}^{(1)}. L Γ ⊆ L Σ ( 1 ) ⊆ Λ Σ ( 1 ) , B = [ Λ Σ ( 1 ) : L Γ ] , I = [ U Σ ( 1 ) : Γ ] , r = rank Λ Σ ( 1 ) .
Theorem 11.1 makes B B B finite. Choose primes p 1 , … , p r \mathfrak{p}_{1},\ldots,\mathfrak{p}_{r} p 1 , … , p r , one from each pair { p , j p } ⊂ Σ \{\mathfrak{p},j\mathfrak{p}\}\subset\Sigma { p , j p } ⊂ Σ with j p ≠ p j\mathfrak{p}\ne\mathfrak{p} j p = p . Then
Λ Σ ( 1 ) = ⨁ k = 1 r Z ( [ p k ] − [ j p k ] ) , D = ( v p k ( A i ) ) k , i ∈ M r , m ( Z ) . \Lambda_{\Sigma}^{(1)}=\bigoplus_{k=1}^{r}\mathbb{Z}([\mathfrak{p}_{k}]-[j\mathfrak{p}_{k}]),\qquad D=(v_{\mathfrak{p}_{k}}(A_{i}))_{k,i}\in M_{r,m}(\mathbb{Z}). Λ Σ ( 1 ) = k = 1 ⨁ r Z ([ p k ] − [ j p k ]) , D = ( v p k ( A i ) ) k , i ∈ M r , m ( Z ) .
Primes fixed by j j j contribute no basis divisor. Thus D : Z m → Z r D:\mathbb{Z}^{m}\to\mathbb{Z}^{r} D : Z m → Z r is the matrix of ( a i ) i ↦ div E ( ∏ i = 1 m A i a i ) (a_{i})_{i}\mapsto\operatorname{div}_{E}(\prod_{i=1}^{m}A_{i}^{a_{i}}) ( a i ) i ↦ div E ( ∏ i = 1 m A i a i ) in this fixed basis. Its image is L Γ L_{\Gamma} L Γ , and its Smith normal form computes B B B and products of the A i A_{i} A i whose divisors form a basis of L Γ L_{\Gamma} L Γ .
Because the divisor kernel is the full group μ ( E ) \mu(E) μ ( E ) , already included in Γ \Gamma Γ , Lemma 3.5 gives
U Σ ( 1 ) / Γ ≃ L Σ ( 1 ) / L Γ , B = [ Λ Σ ( 1 ) : L Σ ( 1 ) ] [ U Σ ( 1 ) : Γ ] . (108) U_{\Sigma}^{(1)}/\Gamma\simeq L_{\Sigma}^{(1)}/L_{\Gamma},\qquad B=[\Lambda_{\Sigma}^{(1)}:L_{\Sigma}^{(1)}][U_{\Sigma}^{(1)}:\Gamma]. \tag*{(108)} U Σ ( 1 ) /Γ ≃ L Σ ( 1 ) / L Γ , B = [ Λ Σ ( 1 ) : L Σ ( 1 ) ] [ U Σ ( 1 ) : Γ ] . (108)
The two factors have different meanings. The first measures which formal anti-invariant divisors possess a norm-one generator in E E E . The second measures which such generators are missing from our chosen subgroup. Internal root extraction changes only the second factor. In particular, the condition Cl ( R Σ ) = 0 \operatorname{Cl}(R_{\Sigma})=0 Cl ( R Σ ) = 0 used for target membership does not make every supported divisor principal in O E \mathcal{O}_{E} O E , and does not force the first factor to be one.
Roots in the coefficient field
Lemma 12.1. If x ∈ E × x\in E^{\times} x ∈ E × and x q ∈ U Σ ( 1 ) x^{q}\in U_{\Sigma}^{(1)} x q ∈ U Σ ( 1 ) for an integer q ≥ 1 q\ge1 q ≥ 1 , then x ∈ U Σ ( 1 ) x\in U_{\Sigma}^{(1)} x ∈ U Σ ( 1 ) . Consequently
Γ s a t = { x ∈ E × : x p b ∈ Γ for some b ≥ 0 } ⊂ U Σ ( 1 ) \Gamma^{\mathrm{sat}}=\{x\in E^{\times}:x^{p^{b}}\in\Gamma\text{ for some }b\ge0\}\subset U_{\Sigma}^{(1)} Γ sat = { x ∈ E × : x p b ∈ Γ for some b ≥ 0 } ⊂ U Σ ( 1 ) is the saturation in the intended norm-one group, including at p = 2 p=2 p = 2 .
Proof. For p ∉ Σ \mathfrak{p}\notin\Sigma p ∈ / Σ , the equation q v p ( x ) = 0 qv_{\mathfrak{p}}(x)=0 q v p ( x ) = 0 gives v p ( x ) = 0 v_{\mathfrak{p}}(x)=0 v p ( x ) = 0 . Moreover ( x j ( x ) ) q = 1 (xj(x))^{q}=1 ( x j ( x ) ) q = 1 . At every real embedding of E + E^{+} E + , x j ( x ) xj(x) x j ( x ) is positive, so it equals one. This proves both assertions. □
Unlike in the ordinary group, no correction by a root of unity is needed to restore a second norm equation. On the other hand, a p p p th root is now determined only up to μ p ⊂ E \mu_{p}\subset E μ p ⊂ E . These alternatives are harmless because μ ( E ) ⊂ Γ \mu(E)\subset\Gamma μ ( E ) ⊂ Γ ; roots of unity must nevertheless be kept in the Kummer basis.
Here is a finite version of the root procedure. At an intermediate stage let Γ ′ \Gamma' Γ ′ contain μ ( E ) \mu(E) μ ( E ) and let g 1 , … , g r g_{1},\ldots,g_{r} g 1 , … , g r lift a basis of its divisor lattice. Let ξ \xi ξ generate the p p p -primary part of μ ( E ) \mu(E) μ ( E ) . Representatives for Γ ′ / Γ ′ p \Gamma'/\Gamma'^{p} Γ ′ / Γ ′ p are
γ a = ξ a 0 ∏ i = 1 r g i a i , 0 ≤ a i < p ( 0 ≤ i ≤ r ) . (109) \gamma_{a}=\xi^{a_{0}}\prod_{i=1}^{r}g_{i}^{a_{i}},\qquad0\le a_{i}<p\quad(0\le i\le r). \tag*{(109)} γ a = ξ a 0 i = 1 ∏ r g i a i , 0 ≤ a i < p ( 0 ≤ i ≤ r ) . (109)
In the same fixed divisor basis put
D Γ ′ = ( ν p k ( g i ) ) 1 ≤ k , i ≤ r ∈ M r ( Z ) . D_{\Gamma'}=\left(\nu_{\mathfrak p_k}(g_i)\right)_{1\le k,i\le r}\in M_r(\mathbb{Z}). D Γ ′ = ( ν p k ( g i ) ) 1 ≤ k , i ≤ r ∈ M r ( Z ) .
This matrix represents the injection Z r → Λ Σ ( 1 ) \mathbb{Z}^r\to\Lambda_{\Sigma}^{(1)} Z r → Λ Σ ( 1 ) sending the i i i th standard vector to div E ( g i ) \operatorname{div}_E(g_i) div E ( g i ) . A root in E E E can exist only if D Γ ′ ( a 1 , … , a r ) T ≡ 0 ( m o d p ) D_{\Gamma'}(a_1,\ldots,a_r)^{\mathsf T}\equiv0\pmod p D Γ ′ ( a 1 , … , a r ) T ≡ 0 ( mod p ) . For each remaining candidate, factor X p − γ a X^p-\gamma_a X p − γ a over E E E . A linear factor gives an actual element x ∈ E x\in E x ∈ E with
x p = ξ a 0 ∏ i = 1 r g i a i , div E ( x ) = 1 p ∑ i = 1 r a i div E ( g i ) . x^p=\xi^{a_0}\prod_{i=1}^{r}g_i^{a_i},\qquad\operatorname{div}_E(x)=\frac{1}{p}\sum_{i=1}^{r}a_i\operatorname{div}_E(g_i). x p = ξ a 0 i = 1 ∏ r g i a i , div E ( x ) = p 1 i = 1 ∑ r a i div E ( g i ) .
Accept it only if div E ( x ) ∉ div E ( Γ ′ ) \operatorname{div}_E(x)\notin\operatorname{div}_E(\Gamma') div E ( x ) ∈ / div E ( Γ ′ ) , and replace Γ ′ \Gamma' Γ ′ by ⟨ Γ ′ , x ⟩ \langle\Gamma',x\rangle ⟨ Γ ′ , x ⟩ . Update the lattice basis by integer row or column operations, keeping the corresponding products of actual field elements. Repeat until there is no accepted root. All these are effective number-field and lattice operations [7 Chapters 3–6] .
Proposition 12.2. For I = [ U Σ ( 1 ) : Γ ] I=[U_{\Sigma}^{(1)}:\Gamma] I = [ U Σ ( 1 ) : Γ ] , one has
Γ s a t / Γ = ( U Σ ( 1 ) / Γ ) [ p ∞ ] , [ Γ s a t : Γ ] = p v p ( I ) , p ∤ [ U Σ ( 1 ) : Γ s a t ] . \Gamma^{\mathrm{sat}}/\Gamma=(U_{\Sigma}^{(1)}/\Gamma)[p^\infty],\qquad[\Gamma^{\mathrm{sat}}:\Gamma]=p^{v_p(I)},\qquad p\nmid[U_{\Sigma}^{(1)}:\Gamma^{\mathrm{sat}}]. Γ sat /Γ = ( U Σ ( 1 ) /Γ ) [ p ∞ ] , [ Γ sat : Γ ] = p v p ( I ) , p ∤ [ U Σ ( 1 ) : Γ sat ] . The procedure reaches Γ s a t \Gamma^{\mathrm{sat}} Γ sat after exactly v p ( I ) ≤ v p ( B ) v_p(I)\le v_p(B) v p ( I ) ≤ v p ( B ) accepted roots. The class map induces
Γ s a t / ( Γ s a t ) n ⟶ ∼ U Σ ( 1 ) / ( U Σ ( 1 ) ) n ⟶ ∼ W Σ ( 1 ) . (110) \Gamma^{\mathrm{sat}}/(\Gamma^{\mathrm{sat}})^n\overset{\sim}{\longrightarrow}U_{\Sigma}^{(1)}/(U_{\Sigma}^{(1)})^n\overset{\sim}{\longrightarrow}W_{\Sigma}^{(1)}. \tag*{(110)} Γ sat / ( Γ sat ) n ⟶ ∼ U Σ ( 1 ) / ( U Σ ( 1 ) ) n ⟶ ∼ W Σ ( 1 ) . (110) Proof. Lemma 12.1 puts every tested root in U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) . Since μ ( E ) ⊂ Γ ′ \mu(E)\subset\Gamma' μ ( E ) ⊂ Γ ′ , membership in Γ ′ \Gamma' Γ ′ is equivalent to membership of its divisor in L Γ ′ L_{\Gamma'} L Γ ′ , and one root per candidate suffices. The representatives (109) therefore detect every index-p p p enlargement. Apply Proposition 7.1, (108) and Proposition 3.6. □
Corollary 12.3. Let z 1 , … , z r z_1,\ldots,z_r z 1 , … , z r lift a Z \mathbb{Z} Z -basis of Γ s a t / μ ( E ) \Gamma^{\mathrm{sat}}/\mu(E) Γ sat / μ ( E ) . If p t p^t p t is the order of the p p p -primary subgroup of μ ( E ) \mu(E) μ ( E ) , choose its generator z 0 z_0 z 0 ; then t ≥ a t\ge a t ≥ a and [ z 0 ] [z_0] [ z 0 ] generates μ ( E ) / μ ( E ) n ≃ Z n \mu(E)/\mu(E)^n\simeq\mathbb{Z}_n μ ( E ) / μ ( E ) n ≃ Z n .
The general module has rank r + 1 r+1 r + 1 , with basis
( Z n ) r + 1 ⟶ ∼ W Σ ( 1 ) , ( e 0 , … , e r ) ⟼ [ ξ 0 e 0 ∏ j = 1 r z j e j ] . (111) (\mathbb{Z}_n)^{r+1}\overset{\sim}{\longrightarrow}W_{\Sigma}^{(1)},\qquad(e_0,\ldots,e_r)\longmapsto\left[\xi_0^{e_0}\prod_{j=1}^{r}z_j^{e_j}\right]. \tag*{(111)} ( Z n ) r + 1 ⟶ ∼ W Σ ( 1 ) , ( e 0 , … , e r ) ⟼ [ ξ 0 e 0 j = 1 ∏ r z j e j ] . (111) For j ≥ 1 j\ge1 j ≥ 1 , recording the root equations and integer basis changes gives
z j p b j = ξ j ∏ i A i m j i , b j ≥ 0 , ξ j ∈ μ ( E ) , m j i ∈ Z . (112) z_j^{p^{b_j}}=\xi_j\prod_i A_i^{m_{ji}},\qquad b_j\ge0,\quad\xi_j\in\mu(E),\quad m_{ji}\in\mathbb{Z}. \tag*{(112)} z j p b j = ξ j i ∏ A i m j i , b j ≥ 0 , ξ j ∈ μ ( E ) , m j i ∈ Z . (112) Proof. The free part contributes r r r copies of Z n \mathbb{Z}_n Z n and the cyclic torsion, which contains μ n \mu_n μ n , contributes one. Each adjoining relation is a p p p th-root equation; substitution gives the stated identities. The basis consists of the specified classes [ z j ] [z_j] [ z j ] , not merely their displayed powers. □
The full p p p -primary torsion is essential: ζ 3 \zeta_3 ζ 3 is a cube in Q ( ζ 9 ) \mathbb{Q}(\zeta_9) Q ( ζ 9 ) , whereas ζ 9 \zeta_9 ζ 9 generates the torsion quotient modulo cubes. This accounts for the extra class [ z 0 ] [z_0] [ z 0 ] ; ordinary torsion has order prime to p p p and contributes no class.
Corollary 12.4. With Γ \Gamma Γ , I I I , B B B as above, the following are equivalent:
Γ s a t = Γ ⟺ p ∤ I ⟺ W Σ ( 1 ) = ⟨ [ z 0 ] , [ A 1 ] , … , [ A m ] ⟩ . \Gamma^{\mathrm{sat}}=\Gamma\quad\Longleftrightarrow\quad p\nmid I\quad\Longleftrightarrow\quad W_{\Sigma}^{(1)}=\langle[z_0],[A_1],\ldots,[A_m]\rangle. Γ sat = Γ ⟺ p ∤ I ⟺ W Σ ( 1 ) = ⟨[ z 0 ] , [ A 1 ] , … , [ A m ]⟩ . In particular p ∤ B p\nmid B p ∤ B suffices. In this case, choose products
g j = ∏ i A i m j i ( 1 ≤ j ≤ r ) g_j=\prod_i A_i^{m_{ji}}\qquad(1\le j\le r) g j = i ∏ A i m j i ( 1 ≤ j ≤ r ) whose divisors form a Z \mathbb{Z} Z -basis of L Γ L_{\Gamma} L Γ . Then [ z 0 ] , [ g 1 ] , … , [ g r ] [z_0],[g_1],\ldots,[g_r] [ z 0 ] , [ g 1 ] , … , [ g r ] are a Kummer basis, with no root extraction and with b j = 0 b_j=0 b j = 0 , ξ j = 1 \xi_j=1 ξ j = 1 in (112) .
Proof. Apply Proposition 7.1 to U Σ ( 1 ) U_{\Sigma}^{(1)} U Σ ( 1 ) , using I ∣ B I \mid B I ∣ B and μ ( E ) ⊆ Γ \mu(E) \subseteq\Gamma μ ( E ) ⊆ Γ . A finite quotient has no p p p -torsion exactly when reduction modulo p a p^{a} p a is already surjective. The divisor basis lifts to a basis of Γ / μ ( E ) \Gamma/\mu(E) Γ/ μ ( E ) .
Thus the principalized RM input with p = 3 p=3 p = 3 and B = 16 B=16 B = 16 in § 16.5 needs no internal cube root. One may also certify the criterion by an invertible residue matrix on r + 1 r+1 r + 1 classes selected from [ z 0 ] [z_0] [ z 0 ] , [ A i ] [A_i] [ A i ] , exactly as in Corollary 8.5 . By contrast, the quartic example in § 16.3 requires one internal square root. As in the ordinary case, a particular target can already lie in the image of Γ \Gamma Γ even when the entire module does not.
Remark 12.5. The congruence on D Γ D_{\Gamma} D Γ is necessary for a root in E E E , but does not prove its existence. The divided divisor must be principal and its generator must have the required power. In (108) , saturation changes [ U Σ ( 1 ) : Γ ] [U_{\Sigma}^{(1)}:\Gamma] [ U Σ ( 1 ) : Γ ] , not [ Λ Σ ( 1 ) : Λ Σ ( 1 ) ] [\Lambda_{\Sigma}^{(1)}:\Lambda_{\Sigma}^{(1)}] [ Λ Σ ( 1 ) : Λ Σ ( 1 ) ] . Thus the stopping condition is the absence of further roots in E E E , not an index prime to p p p in the formal divisor lattice.
Selection by reciprocity in the general case
Selection by Artin reciprocity uses the same Kummer pairing and separating-prime argument as § 8 . The general Kummer module includes a torsion basis class. At two we first select the Kummer class of the signature-corrected character ψ \psi ψ , then restore the signature factor. The underlying Artin map method is standard [15 Section 3] , [8 Sections 5.4.3–5.4.5] , [16 Section 4.2] ; here the input basis is supplied by the RM inputs and p p p -power roots taken inside E E E .
Fix χ : G f + → Z n \chi:G_f^+ \to\mathbb{Z}_n χ : G f + → Z n of exact order n = p a n=p^a n = p a . Use ψ , g , E , Σ \psi,\mathfrak{g},E,\Sigma ψ , g , E , Σ from Theorem 4.2 . Put r = rank Λ Σ ( 1 ) r=\operatorname{rank}\Lambda_{\Sigma}^{(1)} r = rank Λ Σ ( 1 ) and use the basis
W Σ ( 1 ) = ⨁ j = 0 r Z n [ z j ] , ν − 1 ( ψ ∣ G E ) ∈ W Σ ( 1 ) W_{\Sigma}^{(1)}=\bigoplus_{j=0}^{r}\mathbb{Z}_n[z_j],\qquad\nu^{-1}(\psi|_{G_E})\in W_{\Sigma}^{(1)} W Σ ( 1 ) = j = 0 ⨁ r Z n [ z j ] , ν − 1 ( ψ ∣ G E ) ∈ W Σ ( 1 )
of (111) . The class of z 0 z_0 z 0 generates μ ( E ) / μ ( E ) n \mu(E)/\mu(E)^n μ ( E ) / μ ( E ) n : one may choose a generator of μ ( E ) \mu(E) μ ( E ) or of its full p p p -primary subgroup. A generator of μ n \mu_n μ n need not suffice. Exclude
T = Σ ∪ { p : p ∣ n g O E } . \mathcal{T}=\Sigma\cup\{ \mathfrak{p}: \mathfrak{p}\mid n\mathfrak{g}\mathcal{O}_E\}. T = Σ ∪ { p : p ∣ n g O E } .
Call the primes outside T \mathcal{T} T admissible. Applying Lemma 8.1 with d = r + 1 d=r+1 d = r + 1 , enumerate these primes and choose q 1 , … , q r + 1 \mathfrak{q}_1,\ldots,\mathfrak{q}_{r+1} q 1 , … , q r + 1 whose residue rows are independent modulo p p p . The resulting isomorphism is
W Σ ( 1 ) ⟶ ( Z n ) r + 1 , [ x ] ⟼ ( r q k ( x ) ) k = 1 r + 1 . W_{\Sigma}^{(1)}\longrightarrow(\mathbb{Z}_n)^{r+1},\qquad[x]\longmapsto\left(r_{\mathfrak{q}_k}(x)\right)_{k=1}^{r+1}. W Σ ( 1 ) ⟶ ( Z n ) r + 1 , [ x ] ⟼ ( r q k ( x ) ) k = 1 r + 1 .
Its matrix in the ordered basis [ z 0 ] , … , [ z r ] [z_0],\ldots,[z_r] [ z 0 ] , … , [ z r ] and the target vector are
C : ( Z n ) r + 1 → ∼ ( Z n ) r + 1 , C k , j + 1 = r q k ( z j ) 1 ≤ k ≤ r + 1 0 ≤ j ≤ r , C:(\mathbb{Z}_n)^{r+1}\xrightarrow{\sim}(\mathbb{Z}_n)^{r+1},\qquad C_{k,j+1}=r_{\mathfrak{q}_k}(z_j)\qquad\substack{1\leq k\leq r+1\\0\leq j\leq r}, C : ( Z n ) r + 1 ∼ ( Z n ) r + 1 , C k , j + 1 = r q k ( z j ) 1 ≤ k ≤ r + 1 0 ≤ j ≤ r ,
b k = ψ ( [ N E / K q k ] g ) , e = C − 1 b , e = ( e 0 , … , e r ) t . (113) b_k=\psi([\mathcal{N}_{E/K}\mathfrak{q}_k]\mathfrak{g}),\qquad e=C^{-1}b,\qquad e=(e_0,\ldots,e_r)^{\mathsf{t}}. \tag*{(113)} b k = ψ ([ N E / K q k ] g ) , e = C − 1 b , e = ( e 0 , … , e r ) t . (113)
For odd p p p , this right-hand side uses ψ = χ \psi=\chi ψ = χ . For p = 2 p=2 p = 2 , with a 0 a_0 a 0 from (26) , one computes
b k = χ ( [ N E / K q k ] g ) − r q k ( a 0 n 0 / 2 ) in Z n . b_k=\chi([\mathcal{N}_{E/K}\mathfrak{q}_k]\mathfrak{g})-r_{\mathfrak{q}_k}\left(a_0^{n_0/2}\right)\quad\text{in }\mathbb{Z}_n. b k = χ ([ N E / K q k ] g ) − r q k ( a 0 n 0 /2 ) in Z n .
Here χ \chi χ is inflated to g \mathfrak{g} g ; the excluded primes include those in the signature correction. Thus the same finite-field residue operation handles both the basis and the correction.
Theorem 13.1. For the unique solution of (113) , choose 0 ≤ e j < n 0\leq e_j<n 0 ≤ e j < n and put
β ψ = z 0 e 0 ∏ j = 1 r z j e j . \beta_\psi=z_0^{e_0}\prod_{j=1}^{r}z_j^{e_j}. β ψ = z 0 e 0 j = 1 ∏ r z j e j .
Then ν β ψ = ψ ∣ G E \nu_{\beta_\psi}=\psi|_{G_E} ν β ψ = ψ ∣ G E , and
β χ = { β ψ , p ≠ 2 , a 0 n / 2 β ψ , p = 2 satisfies ν β χ = χ ∣ G E . (114) \beta_\chi=
\begin{cases}
\beta_\psi, & p\ne2,\\
a_0^{n/2}\beta_\psi, & p=2
\end{cases}
\qquad\text{satisfies}\qquad
\nu_{\beta_\chi}=\chi|_{G_E}. \tag*{(114)} β χ = { β ψ , a 0 n /2 β ψ , p = 2 , p = 2 satisfies ν β χ = χ ∣ G E . (114)
Proof. Theorem 4.2 places ν − 1 ( ψ ∣ G E ) \nu^{-1}(\psi|_{G_E}) ν − 1 ( ψ ∣ G E ) in W Σ ( 1 ) W_\Sigma^{(1)} W Σ ( 1 ) . Evaluation at the selected primes is injective on this entire module because C C C is invertible. Its unique vector is therefore C − 1 b C^{-1}b C − 1 b . At two, a 0 n / 2 a_0^{n/2} a 0 n /2 represents ( n / 2 ) θ (n/2)\theta ( n /2 ) θ , so multiplication by it restores χ \chi χ .
The answer is the displayed product of the specified z j z_j z j . The recorded identities (112) separately express a power of it in the RM inputs. For s 0 ≥ max j b j s_0\geq\max_j b_j s 0 ≥ max j b j , with s 0 = 0 s_0=0 s 0 = 0 if r = 0 r=0 r = 0 ,
β ψ p s 0 = z 0 e 0 p s 0 ∏ j = 1 r ξ j e j p s 0 − b j ∏ i A i ∑ j = 1 r e j m i j p s 0 − b j . (115) \beta_\psi^{p^{s_0}}=z_0^{e_0p^{s_0}}\prod_{j=1}^{r}\xi_j^{e_jp^{s_0-b_j}}\prod_i A_i^{\sum_{j=1}^{r}e_jm_{ij}p^{s_0-b_j}}. \tag*{(115)} β ψ p s 0 = z 0 e 0 p s 0 j = 1 ∏ r ξ j e j p s 0 − b j i ∏ A i ∑ j = 1 r e j m ij p s 0 − b j . (115)
The power identity alone can leave a root-of-unity ambiguity. We do not choose a root of that equation: β ψ \beta_\psi β ψ is already defined by the product of the computed elements z j z_j z j above. At good reduction the A i A_i A i are Jacobi products. At bad reduction they are CM Frobenius values, or the principalized replacements of (93) . For fixed basis and admissible testing primes, one inverse matrix serves every eligible target. No ray class group of E E E is needed for this selection.
The proof of Corollary 8.5 applies here with rank r + 1 r+1 r + 1 : an invertible residue matrix on z 0 z_0 z 0 and r r r raw RM values certifies that they already form the required Kummer basis. Omitting the torsion coordinate would invalidate that rank test.
The examples in § 16.3 and 16.4 give the selected Kummer coordinates, including the torsion coordinate; the second also displays the signature correction. These calculations determine only the restriction to E E E . Its order may be smaller than n n n , even equal to one. In particular, the full algebra E [ X ] / ( X n − β χ ) E[X]/(X^n-\beta_\chi) E [ X ] / ( X n − β χ ) must be kept for the equivariant descent of the next section, which recovers the original character and its field over K K K .
General generator by equivariant descent
Let χ : G f + → Z n \chi:G_f^+\to\mathbb{Z}_n χ : G f + → Z n have exact order n = p α n=p^\alpha n = p α , and put β = β χ \beta=\beta_\chi β = β χ from (114) . The preceding section gives ν β = χ ∣ G E \nu_\beta=\chi|_{G_E} ν β = χ ∣ G E . The remaining step is standard equivariant Galois descent [17 Section 2.3] [16 Section 4.3] : descend the cyclic torsor, select its Artin character, and take an invariant generator. Work in the full algebra
B β = E [ X ] / ( X n − β ) , α = X mod ( X n − β ) , r : B β ⟶ B β , r ∣ E = id , r ( α ) = ζ n α . B_\beta=E[X]/(X^n-\beta),\qquad\alpha=X\mathbin{\operatorname{mod}}(X^n-\beta),\qquad r:B_\beta\longrightarrow B_\beta,\qquad r|_E=\operatorname{id},\quad r(\alpha)=\zeta_n\alpha. B β = E [ X ] / ( X n − β ) , α = X mod ( X n − β ) , r : B β ⟶ B β , r ∣ E = id , r ( α ) = ζ n α .
Its spectrum is a torsor under the constant group Z n \mathbb{Z}_n Z n : r r r acts freely and transitively on its n n n geometric points. The algebra need not be a field, since restriction to E E E can lower the order of χ \chi χ .
The descent equations and the resulting generator
For δ ∈ Gal ( E / K ) \delta\in\operatorname{Gal}(E/K) δ ∈ Gal ( E / K ) choose 1 ≤ u δ < n 1\leq u_\delta<n 1 ≤ u δ < n with δ ( ζ n ) = ζ n u δ \delta(\zeta_n)=\zeta_n^{u_\delta} δ ( ζ n ) = ζ n u δ ; take u 1 = 1 u_1=1 u 1 = 1 . Equivariant descent is a semilinear action of Gal ( E / K ) \operatorname{Gal}(E/K) Gal ( E / K ) on B β B_\beta B β commuting with r r r . Equivariance forces the form
d δ : B β ⟶ B β , d δ ∣ E = δ , d δ ( α ) = c δ α u δ . d_\delta:B_\beta\longrightarrow B_\beta,\qquad d_\delta|_E=\delta,\qquad d_\delta(\alpha)=c_\delta\alpha^{u_\delta}. d δ : B β ⟶ B β , d δ ∣ E = δ , d δ ( α ) = c δ α u δ .
Indeed d δ ( α ) d_\delta(\alpha) d δ ( α ) belongs to the ζ n u δ \zeta_n^{u_\delta} ζ n u δ -eigenspace of r r r . Preserving the relation α n = β \alpha^n=\beta α n = β and composing the maps gives
c 1 = 1 , c δ n = δ ( β ) / β u δ , c δ δ ′ = δ ( c δ ′ ) c δ u δ ′ β ( u δ u δ ′ − u δ δ ′ ) / n . (116) c_1=1,\qquad c_\delta^n=\delta(\beta)/\beta^{u_\delta},\qquad c_{\delta\delta'}=\delta(c_{\delta'})c_\delta^{u_{\delta'}}\beta^{(u_\delta u_{\delta'}-u_{\delta\delta'})/n}. \tag*{(116)} c 1 = 1 , c δ n = δ ( β ) / β u δ , c δ δ ′ = δ ( c δ ′ ) c δ u δ ′ β ( u δ u δ ′ − u δ δ ′ ) / n . (116)
The last power reduces α u δ u δ ′ \alpha^{u_{\delta}u_{\delta'}} α u δ u δ ′ , using α n = β \alpha^n=\beta α n = β . Distinct δ \delta δ may have the same u δ u_{\delta} u δ , so the coefficients are indexed by automorphisms of E / K E/K E / K . For a solution c = ( c δ ) δ c=(c_{\delta})_{\delta} c = ( c δ ) δ , define its fixed algebra
A β , c = { x ∈ B β : d δ ( x ) = x for every δ ∈ Gal ( E / K ) } . A_{\beta,c}=\{x\in B_{\beta}:d_{\delta}(x)=x\text{ for every }\delta\in\operatorname{Gal}(E/K)\}. A β , c = { x ∈ B β : d δ ( x ) = x for every δ ∈ Gal ( E / K )} .
Its cyclic action determines a character χ c : G K → Z n \chi_c:G_K\to\mathbb{Z}_n χ c : G K → Z n , as proved in Proposition 14.2. Fix a K K K -basis b 1 , … , b d b_1,\ldots,b_d b 1 , … , b d of E E E , where d = [ E : K ] d=[E:K] d = [ E : K ] .
Theorem 14.1. Exactly one solution of (14.1) has character χ c = χ \chi_c=\chi χ c = χ . Its fixed algebra is isomorphic to H χ H_{\chi} H χ . It is spanned over K K K by
θ i k = ∑ δ ∈ Gal ( E / K ) δ ( b i ) c δ k α k u δ , 1 ≤ i ≤ d , 0 ≤ k < n . (117) \theta_{ik}=\sum_{\delta\in\operatorname{Gal}(E/K)}\delta(b_i)c_{\delta}^{k} \alpha^{ku_{\delta}},\qquad1\leq i\leq d,\quad0\leq k<n. \tag*{(117)} θ ik = δ ∈ Gal ( E / K ) ∑ δ ( b i ) c δ k α k u δ , 1 ≤ i ≤ d , 0 ≤ k < n . (117) One can choose integers λ i k \lambda_{ik} λ ik such that
H χ = K ( θ χ ) , θ χ = ∑ i , k λ i k θ i k . (118) H_{\chi}=K(\theta_{\chi}),\qquad\theta_{\chi}=\sum_{i,k}\lambda_{ik}\theta_{ik}. \tag*{(118)} H χ = K ( θ χ ) , θ χ = i , k ∑ λ ik θ ik . (118) A proposed combination is primitive precisely when its n n n translates r a ( θ χ ) r^a(\theta_{\chi}) r a ( θ χ ) , a ∈ Z n a\in\mathbb{Z}_n a ∈ Z n , are distinct.
The next two subsections compute this descent and select χ \chi χ . The proof at the end constructs the primitive invariant.
Computing the equivariant descents
For each nonidentity δ \delta δ , factor X n − δ ( β ) / β u δ X^n-\delta(\beta)/\beta^{u_{\delta}} X n − δ ( β ) / β u δ over E E E . Its linear factors give the possible c δ c_{\delta} c δ . Set c 1 = 1 c_1=1 c 1 = 1 and keep exactly the families satisfying the composition equations in (14.1). This is a finite search with at most n d − 1 n^{d-1} n d − 1 candidates, using only the known field E E E and element β ∈ E × \beta\in E^{\times} β ∈ E × . The following standard descent statement proves that the list is complete and nonempty.
Proposition 14.2. Solutions of (14.1) are precisely the equivariant semilinear Gal ( E / K ) \operatorname{Gal}(E/K) Gal ( E / K ) -actions on B β B_{\beta} B β . They correspond bijectively to continuous characters χ c : G K → Z n \chi_c:G_K\to\mathbb{Z}_n χ c : G K → Z n extending ν β \nu_{\beta} ν β . If m = ∣ im χ c ∣ m=|\operatorname{im}\chi_c| m = ∣ im χ c ∣ and L = Q ‾ ker χ c L=\overline{\mathbb{Q}}^{\ker\chi_c} L = Q k e r χ c , then
E ⊗ K A β , c ≃ B β , A β , c ≃ L n / m , [ L : K ] = m . E\otimes_K A_{\beta,c}\simeq B_{\beta},\qquad A_{\beta,c}\simeq L^{n/m},\qquad[L:K]=m. E ⊗ K A β , c ≃ B β , A β , c ≃ L n / m , [ L : K ] = m . In particular, the fixed algebra is a field exactly when the character is surjective.
Proof. Semilinear descent gives the first isomorphism [17 Lemma 2.3.8 and Remark 2.3.10] . The finite étale Galois correspondence [23 Theorems 8.21–8.23] identifies equivariant descents with actions on the n n n points commuting with r r r . Its centralizer is Z n \mathbb{Z}_n Z n , giving exactly the characters extending ν β \nu_{\beta} ν β . Their orbits have stabilizer ker χ c \ker\chi_c ker χ c , hence A β , c ≃ L n / m A_{\beta,c}\simeq L^{n/m} A β , c ≃ L n / m . This gives actual coefficient families: two equivariant identifications over E E E differ by a power of r r r , which commutes with every descent and therefore leaves it unchanged. □
Lemma 14.3. Relative to one solution, all others are indexed by b ∈ Hom ( Gal ( E / K ) , Z n ) b\in\operatorname{Hom}(\operatorname{Gal}(E/K),\mathbb{Z}_n) b ∈ Hom ( Gal ( E / K ) , Z n ) :
d δ ′ = r b ( δ ) d δ , c δ ′ = ζ n u δ b ( δ ) c δ , d'_{\delta}=r^{b(\delta)}d_{\delta},\qquad c'_{\delta}=\zeta_n^{u_{\delta}b(\delta)}c_{\delta}, d δ ′ = r b ( δ ) d δ , c δ ′ = ζ n u δ b ( δ ) c δ , χ c ′ ( g ) = χ c ( g ) − b ( g ∣ E ) . \chi_{c'}(g)=\chi_c(g)-b(g|_E). χ c ′ ( g ) = χ c ( g ) − b ( g ∣ E ) . Changing α \alpha α to a α a\alpha a α , for a ∈ E × a\in E^{\times} a ∈ E × , preserves the character and gives
β ⟼ a n β , c δ ⟼ δ ( a ) a u δ c δ . (119) \beta\longmapsto a^n\beta,\qquad c_{\delta}\longmapsto\frac{\delta(a)}{a^{u_{\delta}}}c_{\delta}. \tag*{(119)} β ⟼ a n β , c δ ⟼ a u δ δ ( a ) c δ . (119) If ( β 1 , c ( 1 ) ) (\beta_1,c^{(1)}) ( β 1 , c ( 1 ) ) and ( β 2 , c ( 2 ) ) (\beta_2,c^{(2)}) ( β 2 , c ( 2 ) ) correspond to χ 1 , χ 2 : G K → Z n \chi_1,\chi_2:G_K\to\mathbb{Z}_n χ 1 , χ 2 : G K → Z n , then
( β 1 β 2 , ( c δ ( 1 ) c δ ( 2 ) ) δ ) \left(\beta_1\beta_2,\left(c_{\delta}^{(1)}c_{\delta}^{(2)}\right)_{\delta}\right) ( β 1 β 2 , ( c δ ( 1 ) c δ ( 2 ) ) δ ) corresponds to χ 1 + χ 2 \chi_1+\chi_2 χ 1 + χ 2 .
In particular the signature character ( n / 2 ) θ (n/2)\theta ( n /2 ) θ at two has data
β a 0 = a 0 n / 2 , c δ , a 0 = a 0 ( 1 − u δ ) / 2 . (120) \beta_{a_0}=a_0^{n/2},\qquad c_{\delta,a_0}=a_0^{(1-u_\delta)/2}. \tag*{(120)} β a 0 = a 0 n /2 , c δ , a 0 = a 0 ( 1 − u δ ) /2 . (120)
Proof. Two extensions of ν β \nu_\beta ν β differ by a homomorphism from Gal ( E / K ) \operatorname{Gal}(E/K) Gal ( E / K ) . Replacing the lift d δ d_\delta d δ by r b ( δ ) d δ r^{b(\delta)}d_\delta r b ( δ ) d δ subtracts b ( δ ) b(\delta) b ( δ ) from the translation character. The displayed coefficients follow by applying the maps to α \alpha α . The remaining statements follow by substitution; (120) uses the root a 0 n \sqrt[n]{a_0} n a 0 of X n − a 0 n / 2 X^n-a_0^{n/2} X n − a 0 n /2 .
After finding one solution, its finite group of twists supplies the whole list without further root searches.
Lemma 14.4. Put e = [ H χ ∩ E : K ] e=[H_\chi\cap E:K] e = [ H χ ∩ E : K ] . Then
B β ≃ E ⊗ K H χ ≃ ( E H χ ) e . (121) B_\beta\simeq E\otimes_K H_\chi\simeq(EH_\chi)^e. \tag*{(121)} B β ≃ E ⊗ K H χ ≃ ( E H χ ) e . (121) If e > 1 e>1 e > 1 , the selected descent permutes these e e e field factors transitively.
Proof. The tensor-product decomposition for two Galois extensions gives (121) . A union of factor orbits determines an idempotent of the fixed algebra. That algebra is the field H χ H_\chi H χ , so there is only one orbit. □ \square □
Thus a single radical field describes only the restricted character; the cyclic torsor supplies all n n n points needed for descent.
Example 14.5. Let K = Q ( 3 ) K=\mathbb{Q}(\sqrt{3}) K = Q ( 3 ) , E = K ( i ) E=K(i) E = K ( i ) , and let χ \chi χ cut out E / K E/K E / K . Its restriction is zero, so β = 1 \beta=1 β = 1 . On E [ α ] / ( α 2 − 1 ) E[\alpha]/(\alpha^2-1) E [ α ] / ( α 2 − 1 ) the two descents
d ± ∣ E = j , d ± ( α ) = ± α d_{\pm}|_E=j,\qquad d_{\pm}(\alpha)=\pm\alpha d ± ∣ E = j , d ± ( α ) = ± α have fixed algebras K × K K\times K K × K and K [ i α ] ≃ E K[i\alpha]\simeq E K [ i α ] ≃ E , respectively. Here α + d − ( α ) = 0 \alpha+d_-(\alpha)=0 α + d − ( α ) = 0 , whereas i α + d − ( i α ) = 2 i α i\alpha+d_-(i\alpha)=2i\alpha i α + d − ( i α ) = 2 i α is primitive; the coefficients b i b_i b i in (117) are therefore necessary.
Selecting the character by its Artin values
For a candidate c c c , compute A β , c A_{\beta,c} A β , c by the linear equations d δ ( x ) = x d_\delta(x)=x d δ ( x ) = x in the K K K -basis b i α k b_i\alpha^k b i α k . Decompose it into fields and choose one factor L L L ; all factors are isomorphic by Proposition 14.2. The next two standard calculations determine whether its character equals χ \chi χ .
Lemma 14.6. If [ L : K ] = p b > 1 [L:K]=p^b>1 [ L : K ] = p b > 1 and L b − 1 L_{b-1} L b − 1 is its subfield of degree p b − 1 p^{b-1} p b − 1 , then
f q ( L / K ) = v q ( D L / K ) − v q ( D L b − 1 / K ) p b − 1 ( p − 1 ) . (122) f_q(L/K)=\frac{v_q(\mathfrak{D}_{L/K})-v_q(\mathfrak{D}_{L_{b-1}/K})}{p^{b-1}(p-1)}. \tag*{(122)} f q ( L / K ) = p b − 1 ( p − 1 ) v q ( D L / K ) − v q ( D L b − 1 / K ) . (122) For L = K L=K L = K , put f q = 0 f_q=0 f q = 0 . In either case
L ⊆ K f + ⟺ f q ( L / K ) ≤ v q ( f ) at every finite q . (123) L\subseteq K_f^+ \Longleftrightarrow f_q(L/K)\leq v_q(f)\quad\text{at every finite }q. \tag*{(123)} L ⊆ K f + ⟺ f q ( L / K ) ≤ v q ( f ) at every finite q . (123)
Proof. Apply the conductor–discriminant formula [30 Chapter VI, Section 3] : subtraction cancels the characters factoring through L b − 1 L_{b-1} L b − 1 , leaving p b − 1 ( p − 1 ) p^{b-1}(p-1) p b − 1 ( p − 1 ) faithful characters of the same conductor. Both real places are already allowed in the strict modulus. □ \square □
Use maximal-order discriminants, computable together with the subfields by [7 Chapters 3–6] . Only primes dividing D L / K \mathfrak{D}_{L/K} D L / K need a conductor test.
Lemma 14.7. Let q \mathfrak q q be a prime of K K K away from n n n , the discriminant of E / K E/K E / K , and the numerators and denominators of β \beta β , c δ c_\delta c δ . Put q = N q q=\mathrm{N}\mathfrak q q = N q , δ = Frob q ( E / K ) \delta=\operatorname{Frob}_{\mathfrak q}(E/K) δ = Frob q ( E / K ) , and choose p ∣ q \mathfrak p\mid\mathfrak q p ∣ q in E E E . Then
ζ n χ c ( q ) ≡ ( β ( q − u δ ) / n c δ ) u δ − 1 ( m o d p ) , (124) \zeta_n^{\chi_c(\mathfrak q)} \equiv\left(\frac{\beta^{(q-u_\delta)/n}}{c_\delta}\right)^{u_\delta^{-1}} \pmod{\mathfrak p}, \tag*{(124)} ζ n χ c ( q ) ≡ ( c δ β ( q − u δ ) / n ) u δ − 1 ( mod p ) , (124) where u δ − 1 u_\delta^{-1} u δ − 1 is an inverse modulo n n n .
Proof. On a geometric root, r a d δ r^a d_\delta r a d δ acts by α ↦ ζ n a u δ c δ α u δ \alpha\mapsto\zeta_n^{au_\delta}c_\delta\alpha^{u_\delta} α ↦ ζ n a u δ c δ α u δ , whereas arithmetic Frobenius reduces to α q \alpha^q α q . Since q ≡ u δ ( m o d n ) q\equiv u_\delta\pmod n q ≡ u δ ( mod n ) , comparison gives the formula. The expression inside parentheses has n n n th power β q / δ ( β ) ≡ 1 \beta^{q/\delta}(\beta)\equiv1 β q / δ ( β ) ≡ 1 , and reduction is injective on μ n \mu_n μ n . This is the usual Kummer Frobenius calculation [16 Proposition 4] , applied in each geometric factor. □
Proposition 14.8. A candidate is the required descent precisely when its field factor satisfies (123) and
χ c ( [ a ℓ ] ) = χ ( [ a ℓ ] ) for generators [ a ℓ ] of G f + . \chi_c([\mathfrak a_\ell])=\chi([\mathfrak a_\ell])\qquad\text{for generators }[\mathfrak a_\ell]\text{ of }G_{\mathfrak f}^{+}. χ c ([ a ℓ ]) = χ ([ a ℓ ]) for generators [ a ℓ ] of G f + . Exactly one candidate passes these tests, and its fixed algebra is isomorphic to H χ H_\chi H χ .
Proof. The conductor test makes χ c \chi_c χ c a character of G f + G_{\mathfrak f}^{+} G f + ; equality on generators gives χ c = χ \chi_c=\chi χ c = χ . Existence and uniqueness follow from Proposition 14.2, and the fixed algebra is a field because χ \chi χ has order n n n . □
Represent the ray class generators by ideals outside the finite excluded set, apply (124) to their prime factors, and add the resulting values. The procedure uses only explicitly constructed fields, not an equation for the unknown H χ H_\chi H χ . Exact finite conductor f \mathfrak f f can additionally be checked by testing that χ \chi χ does not factor through any G f / q + G_{\mathfrak f/\mathfrak q}^{+} G f / q + with q ∣ f \mathfrak q\mid\mathfrak f q ∣ f .
A primitive invariant and comparison with the ordinary case.
Proof of Theorem 14.1 . Proposition 14.8 gives the unique family. Averaging the K K K -basis b i α k b_i\alpha^k b i α k projects onto its fixed algebra, so the sums (117) span H χ H_\chi H χ . Primitivity is equivalent to having n n n distinct translates under r r r . For an explicit finite search, enumerate these sums as v ℓ v_\ell v ℓ and put
θ ( m ) = ∑ ℓ = 0 d n − 1 m ℓ v ℓ . \theta(m)=\sum_{\ell=0}^{dn-1}m^\ell v_\ell. θ ( m ) = ℓ = 0 ∑ d n − 1 m ℓ v ℓ .
Each polynomial r a θ ( m ) − θ ( m ) r^a\theta(m)-\theta(m) r a θ ( m ) − θ ( m ) , 0 < a < n 0<a<n 0 < a < n , is nonzero of degree at most d n − 1 dn-1 d n − 1 . Some 0 ≤ m ≤ ( n − 1 ) ( d n − 1 ) 0\le m\le(n-1)(dn-1) 0 ≤ m ≤ ( n − 1 ) ( d n − 1 ) therefore gives a primitive invariant, verified inside B β B_\beta B β . □
In the ordinary anti-invariant case, the same descent has H χ ∩ E = K H_\chi\cap E=K H χ ∩ E = K and is selected by the norm condition c u τ ( c u ) = 1 c_u\tau(c_u)=1 c u τ ( c u ) = 1 . Theorem 9.1 then gives the primitive unweighted trace, so enumeration and the conductor–Artin tests are unnecessary. The distinction is in how the equivariant descent is selected; the cyclic-torsor interpretation is the same in both cases.
Construction of the strict ray class field
The cyclic generators for prime-power-order characters of the full group G f + G_{\mathfrak f}^{+} G f + generate the strict ray class field. Here the prime power is the order of the character, not its modulus; its conductor may involve several prime ideals. We first compute these characters, then assemble their cyclic fields and describe the Artin action. Finally we give an exact criterion for the compositum attached to separate prime-power moduli to suffice.
Computing the characters of the full modulus
This subsection supplies the separating characters, including their values on the ideals used in the reciprocity tests. The computation is the standard ray-group presentation [8 Section 4.3.1] [16 Section 3] .
Write f = ∏ i f i \mathfrak{f}=\prod_i\mathfrak{f}_i f = ∏ i f i with f i = p i a i \mathfrak{f}_i=\mathfrak{p}_i^{a_i} f i = p i a i for distinct primes p i \mathfrak{p}_i p i , put R i = ( O K / f i ) × R_i=(\mathcal{O}_K/\mathfrak{f}_i)^\times R i = ( O K / f i ) × , and choose ideals a 1 , … , a h \mathfrak{a}_1,\ldots,\mathfrak{a}_h a 1 , … , a h representing an invariant-factor basis Cl + ( K ) = ⨁ j = 1 h Z m j \operatorname{Cl}^{+}(K)=\bigoplus_{j=1}^{h}\mathbb{Z}_{m_j} Cl + ( K ) = ⨁ j = 1 h Z m j . Choose them prime to f \mathfrak{f} f , with
a j m j = ( γ j ) , γ j ≫ 0. (125) \mathfrak{a}_j^{m_j}=(\gamma_j),\qquad\gamma_j\gg0. \tag*{(125)} a j m j = ( γ j ) , γ j ≫ 0. (125)
Put R ‾ f = ( ∏ i R i ) / ⟨ ϵ ‾ ⟩ \overline{R}_{\mathfrak{f}}=(\prod_i R_i)/\langle\overline{\epsilon}\rangle R f = ( ∏ i R i ) / ⟨ ϵ ⟩ . Let x 1 , … , x h x_1,\ldots,x_h x 1 , … , x h denote the standard generators of a free abelian group Z h \mathbb{Z}^h Z h , mapping to the ray classes [ a j ] f [\mathfrak{a}_j]_{\mathfrak{f}} [ a j ] f . Writing finite abelian groups additively for the following presentation, the ray sequence (10) gives
G f + ≅ R ‾ f ⊕ ⨁ j Z x j ⟨ m j x j − [ γ ‾ j ] ⟩ . (126) G_{\mathfrak{f}}^{+}\cong\frac{\overline{R}_{\mathfrak{f}}\oplus\bigoplus_j\mathbb{Z}x_j}{\langle m_jx_j-[\overline{\gamma}_j]\rangle}. \tag*{(126)} G f + ≅ ⟨ m j x j − [ γ j ]⟩ R f ⊕ ⨁ j Z x j . (126)
The class of a residue with totally positive lift b b b maps to [ ( b ) ] f [(b)]_{\mathfrak{f}} [( b ) ] f . Thus the relation m j x j = [ γ ‾ j ] m_jx_j=[\overline{\gamma}_j] m j x j = [ γ j ] records exactly (15.1).
Equivalently, a character to Z n \mathbb{Z}_n Z n is specified by additive local characters ℓ i : R i → Z n \ell_i:R_i\to\mathbb{Z}_n ℓ i : R i → Z n and values y j = χ ( [ a j ] ) y_j=\chi([\mathfrak{a}_j]) y j = χ ([ a j ]) satisfying
∑ i ℓ i ( ϵ ) = 0 , m j y j = ∑ i ℓ i ( γ j ) for every j . (127) \sum_i\ell_i(\epsilon)=0,\qquad m_jy_j=\sum_i\ell_i(\gamma_j)\quad\text{for every }j. \tag*{(127)} i ∑ ℓ i ( ϵ ) = 0 , m j y j = i ∑ ℓ i ( γ j ) for every j . (127)
Evaluation means reduction modulo p i a i \mathfrak{p}_i^{a_i} p i a i ; denominators prime to that ideal are allowed. The first relation is the global unit relation; the others are the chosen ideal-class relations. If a ∏ j a j − n j = ( b ) \mathfrak{a}\prod_j\mathfrak{a}_j^{-n_j}=(b) a ∏ j a j − n j = ( b ) with b ≫ 0 b\gg0 b ≫ 0 , then
χ ( [ a ] ) = ∑ i ℓ i ( b ) + ∑ j n j y j . (128) \chi([\mathfrak{a}])=\sum_i\ell_i(b)+\sum_j n_jy_j. \tag*{(128)} χ ([ a ]) = i ∑ ℓ i ( b ) + j ∑ n j y j . (128)
The relations make this value independent of the choices.
Here is the coordinate map supplied by Smith normal form. Choose an invariant-factor basis g 1 , … , g t g_1,\ldots,g_t g 1 , … , g t of ∏ i R i \prod_iR_i ∏ i R i , with respective orders c 1 , … , c t c_1,\ldots,c_t c 1 , … , c t . Its residue classes and the ideals a j \mathfrak{a}_j a j define a surjection
π : Z t + h ⟶ G f + . \pi:\mathbb{Z}^{t+h}\longrightarrow G_{\mathfrak{f}}^{+}. π : Z t + h ⟶ G f + .
For each g l g_l g l , the image is the principal ray class of a totally positive lift; the last h h h standard vectors map to [ a j ] f [\mathfrak{a}_j]_{\mathfrak{f}} [ a j ] f . Write
ϵ ‾ = ∏ l = 1 t g l u l , γ ‾ j = ∏ l = 1 t g l v l j , u = ( u l ) l , V = ( v l j ) l , j . \overline{\epsilon}=\prod_{l=1}^{t}g_l^{u_l},\qquad\overline{\gamma}_j=\prod_{l=1}^{t}g_l^{v_{lj}},\qquad u=(u_l)_l,\quad V=(v_{lj})_{l,j}. ϵ = l = 1 ∏ t g l u l , γ j = l = 1 ∏ t g l v l j , u = ( u l ) l , V = ( v l j ) l , j .
For N = t + h N=t+h N = t + h , the relation matrix is
A = ( diag ( c 1 , … , c t ) u − V 0 0 diag ( m 1 , … , m h ) ) ∈ M N , N + 1 ( Z ) , im A = ker π . A=\begin{pmatrix}
\operatorname{diag}(c_1,\ldots,c_t)&u&-V\\
0&0&\operatorname{diag}(m_1,\ldots,m_h)
\end{pmatrix}\in M_{N,N+1}(\mathbb{Z}),\qquad\operatorname{im}A=\ker\pi. A = ( diag ( c 1 , … , c t ) 0 u 0 − V diag ( m 1 , … , m h ) ) ∈ M N , N + 1 ( Z ) , im A = ker π .
Smith normal form gives
P ∈ GL N ( Z ) , Q ∈ GL N + 1 ( Z ) , P A Q = ( diag ( d 1 , … , d N ) ∣ 0 ) , d j > 0. P\in\operatorname{GL}_N(\mathbb{Z}),\qquad Q\in\operatorname{GL}_{N+1}(\mathbb{Z}),\qquad PAQ=(\operatorname{diag}(d_1,\ldots,d_N)\mid0),\qquad d_j>0. P ∈ GL N ( Z ) , Q ∈ GL N + 1 ( Z ) , P A Q = ( diag ( d 1 , … , d N ) ∣ 0 ) , d j > 0.
The group is finite, so there are N N N nonzero diagonal entries. The factors d j = 1 d_j=1 d j = 1 may be omitted.
Proposition 15.1. With this notation the explicit isomorphism is
φ : G f + → ∼ ⨁ j = 1 N Z d j , φ ( π ( v ) ) = ( ( P v ) j m o d d j ) j . \varphi:G_{\mathfrak{f}}^{+}\xrightarrow{\sim}\bigoplus_{j=1}^{N}\mathbb{Z}_{d_j},\qquad\varphi(\pi(v))=((Pv)_j\bmod d_j)_j. φ : G f + ∼ j = 1 ⨁ N Z d j , φ ( π ( v )) = (( P v ) j mod d j ) j .
For each p ∣ d j p \mid d_j p ∣ d j , its coordinate character
χ j , p : G f + ⟶ Z p v p ( d j ) , g ⟼ φ j ( g ) m o d p v p ( d j ) \chi_{j,p}: G_{\mathfrak f}^{+} \longrightarrow\mathbb{Z}_{p^{v_p(d_j)}}, \qquad g \longmapsto\varphi_j(g) \bmod p^{v_p(d_j)} χ j , p : G f + ⟶ Z p v p ( d j ) , g ⟼ φ j ( g ) mod p v p ( d j )
has exact order p v p ( d j ) p^{v_p(d_j)} p v p ( d j ) , and ⋂ j , p ker χ j , p = { 1 } \bigcap_{j,p}\ker\chi_{j,p}=\{1\} ⋂ j , p ker χ j , p = { 1 } . In particular all orders p v p ( d j ) p^{v_p(d_j)} p v p ( d j ) and all values on the chosen generators are known before constructing any Kummer module. Equation (128) then evaluates these characters on the testing ideals.
Proof. Smith normal form gives the coordinate maps for the presented group. The Chinese remainder theorem recovers each coordinate from its primary reductions, proving the common-kernel assertion. □
Thus prime-power orders always suffice when the characters are computed on G f + G_{\mathfrak f}^{+} G f + . Restricting to order p p p would not suffice: all maps Z p 2 → Z p \mathbb{Z}_{p^2}\to\mathbb{Z}_p Z p 2 → Z p kill the subgroup p Z p 2 p\mathbb{Z}_{p^2} p Z p 2 . The required orders are the full primary orders of the invariant factors.
Assembly and Artin action.
Theorem 15.2. For the characters χ j , p \chi_{j,p} χ j , p of Proposition 15.1, construct θ χ j , p \theta_{\chi_{j,p}} θ χ j , p by Theorem 14.1. For ordinary anti-invariant characters one may instead use Theorem 9.1. Then
K f + = K ( θ χ j , p : p ∣ d j ) . (129) K_{\mathfrak f}^{+}=K(\theta_{\chi_{j,p}}:p\mid d_j). \tag*{(129)} K f + = K ( θ χ j , p : p ∣ d j ) . (129)
Proof. The characters have trivial common kernel. Apply (12) to their constructed cyclic fields. □
Thus the basis of the Kummer module, reciprocity selection and descent give a field generator for each computed character. Taking their compositum requires no further compatibility among the chosen radicals.
For the Galois action, fix one character χ : G f + → Z n \chi:G_{\mathfrak f}^{+}\to\mathbb{Z}_n χ : G f + → Z n of exact order n n n and put
ρ χ : G f + ⟶ μ n , ρ χ ( g ) = ζ n χ ( g ) . \rho_\chi:G_{\mathfrak f}^{+}\longrightarrow\mu_n,\qquad\rho_\chi(g)=\zeta_n^{\chi(g)}. ρ χ : G f + ⟶ μ n , ρ χ ( g ) = ζ n χ ( g ) .
The map Z n → μ n \mathbb{Z}_n\to\mu_n Z n → μ n , a ↦ ζ n a a\mapsto\zeta_n^a a ↦ ζ n a , identifies these two forms of the same character: χ \chi χ gives the exponents in the linear systems, and ρ χ \rho_\chi ρ χ gives the root-of-unity multipliers. Write the descended generator as θ χ = Θ χ ( α ) \theta_\chi=\Theta_\chi(\alpha) θ χ = Θ χ ( α ) , with Θ χ ( X ) ∈ E [ X ] \Theta_\chi(X)\in E[X] Θ χ ( X ) ∈ E [ X ] and α n = β χ \alpha^n=\beta_\chi α n = β χ in B β B_\beta B β . Identify the selected fixed algebra A β , c A_{\beta,c} A β , c with H χ H_\chi H χ . For ξ ∈ μ n \xi\in\mu_n ξ ∈ μ n , the E E E -automorphism
s ξ : B β ⟶ B β , α ⟼ ξ α s_\xi:B_\beta\longrightarrow B_\beta,\qquad\alpha\longmapsto\xi\alpha s ξ : B β ⟶ B β , α ⟼ ξ α
commutes with descent and restricts to a K K K -automorphism of H χ H_\chi H χ .
Proposition 15.3. For a ∈ I K ( f ) a\in I_K(\mathfrak f) a ∈ I K ( f ) ,
Art f ( [ a ] ) ∣ H χ = s ρ χ ( [ a ] ) ∣ H χ . \operatorname{Art}_{\mathfrak f}([a])|_{H_\chi}=s_{\rho_\chi([a])}|_{H_\chi}. Art f ([ a ]) ∣ H χ = s ρ χ ([ a ]) ∣ H χ . Consequently
Art f ( [ a ] ) ( θ χ ) = Θ χ ( ρ χ ( [ a ] ) α ) . (130) \operatorname{Art}_{\mathfrak f}([a])(\theta_\chi)=\Theta_\chi\bigl(\rho_\chi([a])\alpha\bigr). \tag*{(130)} Art f ([ a ]) ( θ χ ) = Θ χ ( ρ χ ([ a ]) α ) . (130) For the characters in Theorem 15.2, these formulas give the action of Gal ( K f + / K ) \operatorname{Gal}(K_{\mathfrak f}^{+}/K) Gal ( K f + / K ) on all the generators.
Proof. The selected descent has character χ \chi χ , so Artin reciprocity acts through r χ ( [ a ] ) = s ρ χ ( [ a ] ) r^\chi([a])=s_{\rho_\chi([a])} r χ ([ a ]) = s ρ χ ([ a ]) [25 Part III, Theorems 7.8 and 9.2] . Substitute in Θ χ \Theta_\chi Θ χ to obtain the second formula. □
The polynomials in the ordinary and general constructions are, respectively,
Θ χ ( X ) = ∑ u ∈ Z n ∗ c u X u (131) \Theta_\chi(X)=\sum_{u\in\mathbb{Z}_n^*}c_uX^u \tag*{(131)} Θ χ ( X ) = u ∈ Z n ∗ ∑ c u X u (131)
and
Θ χ ( X ) = ∑ i = 1 [ E : K ] ∑ k = 0 n − 1 λ i k ∑ δ ∈ Gal ( E / K ) δ ( b i ) c δ d X k u δ . (132) \Theta_\chi(X)=\sum_{i=1}^{[E:K]}\sum_{k=0}^{n-1}\lambda_{ik}\sum_{\delta\in\operatorname{Gal}(E/K)}\delta(b_i)c_\delta^dX^{ku_\delta}. \tag*{(132)} Θ χ ( X ) = i = 1 ∑ [ E : K ] k = 0 ∑ n − 1 λ ik δ ∈ Gal ( E / K ) ∑ δ ( b i ) c δ d X k u δ . (132)
The n n n values Θ χ ( ξ α ) \Theta_\chi(\xi\alpha) Θ χ ( ξ α ) , ξ ∈ μ n \xi\in\mu_n ξ ∈ μ n , are the distinct conjugates of θ χ \theta_\chi θ χ , and their product polynomial ∏ ξ ∈ μ n ( T − Θ χ ( ξ α ) ) \prod_{\xi\in\mu_n}(T-\Theta_\chi(\xi\alpha)) ∏ ξ ∈ μ n ( T − Θ χ ( ξ α )) lies in K [ T ] K[T] K [ T ] .
What separate prime-power moduli can miss
Assume f ≠ ( 1 ) \mathfrak f\ne(1) f = ( 1 ) and put K 0 = ∏ i K f i + K_0=\prod_iK_{\mathfrak f_i}^+ K 0 = ∏ i K f i + , a compositum. The relevant forgetful map and its kernel are
π f : G f + ⟶ ∏ i G f i + , [ a ] f ⟼ ( [ a ] f i ) i , J f = ker π f . \pi_{\mathfrak f}:G_{\mathfrak f}^+\longrightarrow\prod_iG_{\mathfrak f_i}^+,\qquad[\mathfrak a]_{\mathfrak f}\longmapsto\bigl([\mathfrak a]_{\mathfrak f_i}\bigr)_i,\qquad J_{\mathfrak f}=\ker\pi_{\mathfrak f}. π f : G f + ⟶ i ∏ G f i + , [ a ] f ⟼ ( [ a ] f i ) i , J f = ker π f .
This kernel measures exactly the part of the full ray class field missing from K 0 K_0 K 0 .
Proposition 15.4. For U i = ⟨ ϵ ˉ ⟩ ⊂ R i U_i=\langle\bar\epsilon\rangle\subset R_i U i = ⟨ ϵ ˉ ⟩ ⊂ R i and o i = ∣ U i ∣ o_i=|U_i| o i = ∣ U i ∣ , one has
1 ⟶ J f ⟶ G f + → π f im π f ⟶ 1 , (133) 1\longrightarrow J_{\mathfrak f}\longrightarrow G_{\mathfrak f}^+\xrightarrow{\pi_{\mathfrak f}}\operatorname{im}\pi_{\mathfrak f}\longrightarrow1, \tag*{(133)} 1 ⟶ J f ⟶ G f + π f im π f ⟶ 1 , (133) J f ≃ ( ∏ i U i ) / ⟨ ( ϵ ˉ ) i ⟩ . J_{\mathfrak f}\simeq\left(\prod_iU_i\right)/\langle(\bar\epsilon)_i\rangle. J f ≃ ( i ∏ U i ) / ⟨( ϵ ˉ ) i ⟩ . The image consists of tuples with the same image in Cl + ( K ) \operatorname{Cl}^+(K) Cl + ( K ) . Under Artin reciprocity,
J f = Gal ( K f + / K 0 ) , [ K f + : K 0 ] = ∏ i o i lcm i o i , (134) J_{\mathfrak f}=\operatorname{Gal}(K_{\mathfrak f}^+/K_0),\qquad[K_{\mathfrak f}^+:K_0]=\frac{\prod_i o_i}{\operatorname{lcm}_i o_i}, \tag*{(134)} J f = Gal ( K f + / K 0 ) , [ K f + : K 0 ] = lcm i o i ∏ i o i , (134) K 0 = K f + ⟺ o i are pairwise coprime . K_0=K_{\mathfrak f}^+\Longleftrightarrow o_i\text{ are pairwise coprime}. K 0 = K f + ⟺ o i are pairwise coprime . For a character given by (127) ,
H χ ⊆ K 0 ⟺ ℓ i ( ϵ ˉ ) = 0 for every i . (135) H_\chi\subseteq K_0\Longleftrightarrow\ell_i(\bar\epsilon)=0\quad\text{for every }i. \tag*{(135)} H χ ⊆ K 0 ⟺ ℓ i ( ϵ ˉ ) = 0 for every i . (135)
Proof. In the principal part of (10) , the forgetting map is
( ∏ i R i ) / ⟨ ( ϵ ˉ ) i ⟩ ⟶ ∏ i ( R i / U i ) , \left(\prod_iR_i\right)/\langle(\bar\epsilon)_i\rangle\longrightarrow\prod_i(R_i/U_i), ( i ∏ R i ) / ⟨( ϵ ˉ ) i ⟩ ⟶ i ∏ ( R i / U i ) ,
whose kernel is J f J_{\mathfrak f} J f . The ray maps all induce the identity on Cl + ( K ) \operatorname{Cl}^+(K) Cl + ( K ) . Given ray classes with a common narrow ideal class, divide by an ideal in that class and use the Chinese remainder theorem to lift the resulting principal classes. This proves exactness. The diagonal unit has order lcm i o i \operatorname{lcm}_i o_i lcm i o i , giving the degree formula; the product equals the least common multiple exactly for pairwise coprime orders. Finally χ \chi χ kills J f J_{\mathfrak f} J f precisely when its residue characters kill every U i U_i U i . These are the usual ray-sequence and character arguments [8 Section 4.3.1] . □
The full group requires only ∑ i ℓ i ( ϵ ˉ ) = 0 \sum_i\ell_i(\bar\epsilon)=0 ∑ i ℓ i ( ϵ ˉ ) = 0 ; the separate moduli require each summand to be zero. This is why their compositum can miss characters, although prime-power-order characters of the full group always suffice.
The missing group and the characters needed to detect it can both be computed before the Kummer constructions. If there are s s s prime-power factors of f \mathfrak f f , then
J f ≃ Z s / im ( diag ( o 1 , … , o s ) ∣ 1 ) . J_{\mathfrak f}\simeq\mathbb{Z}^s/\operatorname{im}(\operatorname{diag}(o_1,\ldots,o_s)\mid1). J f ≃ Z s / im ( diag ( o 1 , … , o s ) ∣ 1 ) .
Its Smith normal form gives the invariant factors of the missing relative extension. To locate this group inside G f + G_{\mathfrak f}^+ G f + , take totally positive CRT lifts
w i ≡ ϵ ( m o d f i ) , w i ≡ 1 ( m o d f j ) ( j ≠ i ) . w_i\equiv\epsilon\pmod{\mathfrak f_i},\qquad w_i\equiv1\pmod{\mathfrak f_j}\quad(j\ne i). w i ≡ ϵ ( mod f i ) , w i ≡ 1 ( mod f j ) ( j = i ) .
The classes [ ( w i ) ] f [(w_i)]_{\mathfrak f} [( w i ) ] f generate J f J_{\mathfrak f} J f . Their coordinates under φ \varphi φ and the values χ ( [ ( w i ) ] ) = ℓ i ( ϵ ) \chi([(w_i)])=\ell_i(\epsilon) χ ([( w i )]) = ℓ i ( ϵ ) are obtained from (128) .
Corollary 15.5. For a subset S S S of the primary coordinate characters in Proposition 15.1,
K 0 ( θ χ : χ ∈ S ) = K f + ⟺ J f ∩ ⋂ χ ∈ S ker χ = { 1 } . K_0(\theta_\chi:\chi\in S)=K_{\mathfrak f}^{+}\quad\Longleftrightarrow\quad J_{\mathfrak f}\cap\bigcap_{\chi\in S}\ker\chi=\{1\}. K 0 ( θ χ : χ ∈ S ) = K f + ⟺ J f ∩ χ ∈ S ⋂ ker χ = { 1 } . Such a set can be selected from the computed finite group before constructing any θ χ \theta_\chi θ χ . Its character orders are the known numbers p v p ( d j ) p^{v_p(d_j)} p v p ( d j ) from the full ray-group presentation.
Proof. The subgroup fixing the field on the left is precisely the intersection on the right. The full family of coordinate characters has trivial common kernel, so it always contains a suitable subset. □
One concrete selection is to start with J 0 = J f J_0=J_{\mathfrak f} J 0 = J f , choose a coordinate character χ a \chi_a χ a nonzero on J a − 1 J_{a-1} J a − 1 , and set J a = J a − 1 ∩ ker χ a J_a=J_{a-1}\cap\ker\chi_a J a = J a − 1 ∩ ker χ a . The order decreases at each step until J a = { 1 } J_a=\{1\} J a = { 1 } . All these intersections are computations in the finite presentation above. Afterwards the constructed fields K a = K a − 1 ( θ χ a ) K_a=K_{a-1}(\theta_{\chi_a}) K a = K a − 1 ( θ χ a ) satisfy
Gal ( K f + / K a ) = J a , [ K a : K a − 1 ] = ∣ χ a ( J a − 1 ) ∣ . \operatorname{Gal}(K_{\mathfrak f}^{+}/K_a)=J_a,\qquad[K_a:K_{a-1}]=|\chi_a(J_{a-1})|. Gal ( K f + / K a ) = J a , [ K a : K a − 1 ] = ∣ χ a ( J a − 1 ) ∣.
The order of χ a \chi_a χ a on the full group can exceed its order on J a − 1 J_{a-1} J a − 1 . Thus the invariant factors of J f J_{\mathfrak f} J f alone do not specify the character orders needed for this construction.
Example 15.6. Let K = Q ( 13 ) K=\mathbb{Q}(\sqrt{13}) K = Q ( 13 ) and ϵ = ( 11 + 3 13 ) / 2 \epsilon=(11+3\sqrt{13})/2 ϵ = ( 11 + 3 13 ) /2 . Here h K + = 1 h_K^{+}=1 h K + = 1 , and the unit has orders 6 and 8 at the inert primes 5 and 7. Residue generators can be chosen so that ϵ = g 5 4 = g 7 6 \epsilon=g_5^4=g_7^6 ϵ = g 5 4 = g 7 6 . Writing e 5 , e 7 e_5,e_7 e 5 , e 7 for their ray classes gives
G ( 35 ) + ≅ Z e 5 ⊕ Z e 7 ⟨ 24 e 5 , 48 e 7 , 4 e 5 + 6 e 7 ⟩ ≅ Z 2 ⊕ Z 24 , G ( 5 ) + ≅ Z 4 , G ( 7 ) + ≅ Z 6 . G_{(35)}^{+}\cong\frac{\mathbb{Z}e_5\oplus\mathbb{Z}e_7}{\langle24e_5,48e_7,4e_5+6e_7\rangle}\cong\mathbb{Z}_2\oplus\mathbb{Z}_{24},\qquad G_{(5)}^{+}\cong\mathbb{Z}_4,\qquad G_{(7)}^{+}\cong\mathbb{Z}_6. G ( 35 ) + ≅ ⟨ 24 e 5 , 48 e 7 , 4 e 5 + 6 e 7 ⟩ Z e 5 ⊕ Z e 7 ≅ Z 2 ⊕ Z 24 , G ( 5 ) + ≅ Z 4 , G ( 7 ) + ≅ Z 6 . The character χ ( e 5 ) = 1 \chi(e_5)=1 χ ( e 5 ) = 1 , χ ( e 7 ) = 2 \chi(e_7)=2 χ ( e 7 ) = 2 in Z 8 \mathbb{Z}_8 Z 8 kills the common unit relation since 4 + 6 ⋅ 2 = 0 4+6\cdot2=0 4 + 6 ⋅ 2 = 0 in Z 8 \mathbb{Z}_8 Z 8 . Its value on ( ϵ , 1 ) (\epsilon,1) ( ϵ , 1 ) is 4, so it is nontrivial on the order-two group J ( 35 ) J_{(35)} J ( 35 ) . Hence
[ K 0 : K ] = 24 , [ K 0 H χ : K 0 ] = 2 , K 0 H χ = K ( 35 ) + . [K_0:K]=24,\qquad[K_0H_\chi:K_0]=2,\qquad K_0H_\chi=K_{(35)}^{+}. [ K 0 : K ] = 24 , [ K 0 H χ : K 0 ] = 2 , K 0 H χ = K ( 35 ) + . Thus an order-eight character of the full modulus recovers the missing quadratic extension over K 0 K_0 K 0 . Every character to Z 2 \mathbb{Z}_2 Z 2 or Z 4 \mathbb{Z}_4 Z 4 kills J ( 35 ) J_{(35)} J ( 35 ) : its generator 4 e 5 4e_5 4 e 5 is already a multiple of four. The order of the missing relative extension therefore cannot replace the order of the required full-group character.
The strict Hilbert class field
For f = ( 1 ) \mathfrak f=(1) f = ( 1 ) the result is the classical strict Hilbert class field H K + H_K^{+} H K + , the maximal abelian extension unramified at finite primes [24 Chapter VI, Proposition 6.8] . The following simplification concerns the norm-one representatives, not the definition of that field. Compare the computational constructions in [16 Sections 3, 4.2–4.3] and [9 ] .
Proposition 15.7. Let n = p a n=p^a n = p a , E = Q ( ζ lcm ( Δ , n ) ) E=\mathbb{Q}(\zeta_{\operatorname{lcm}(\Delta,n)}) E = Q ( ζ lcm ( Δ , n ) ) , and let Σ \Sigma Σ be finite and Gal ( E / Q ) \operatorname{Gal}(E/\mathbb{Q}) Gal ( E / Q ) -stable, containing the primes above p p p and all primes of norm at most the Minkowski bound B E B_E B E . For every character χ : Cl + ( K ) → Z n \chi:\operatorname{Cl}^{+}(K)\to\mathbb{Z}_n χ : Cl + ( K ) → Z n , one has
ν − 1 ( χ ∣ G E ) ∈ W Σ ( 1 ) . \nu^{-1}(\chi|_{G_E})\in W_{\Sigma}^{(1)}. ν − 1 ( χ ∣ G E ) ∈ W Σ ( 1 ) . In particular no signature correction or extra dyadic coefficient enlargement is needed.
Proof. The relation a τ K a = ( N a ) a\tau_K a=(N a) a τ K a = ( N a ) implies that τ K \tau_K τ K acts on C l + ( K ) \mathrm{Cl}^{+}(K) Cl + ( K ) by inversion. Reciprocity gives the same action on H K + H_K^+ H K + [25 Part III, Section 6] . Complex conjugation lies in the 2-torsion, which inversion fixes; it is therefore central. Hence each cyclic field H ⊂ H K + H\subset H_K^+ H ⊂ H K + is totally real or CM.
Put L = H E L=HE L = H E and let J J J be complex conjugation. For a Kummer eigenvector α \alpha α of the restricted character, A = α J ( α ) ∈ ( E + ) × , + A=\alpha J(\alpha)\in(E^+)^{\times,+} A = α J ( α ) ∈ ( E + ) × , + . The extension L / E L/E L / E is unramified at finite primes. At an inert unramified prime of E / E + E/E^+ E / E + , the valuations of α \alpha α at all primes above it agree, since its conjugates differ by roots of unity. Thus A A A has even valuation and is a local norm. It is also a norm at split primes and real places. The cyclotomic inertia description [24 Chapter I, Proposition 10.3] shows that E / E + E/E^+ E / E + has at most one finite ramified prime: j j j can lie in inertia at ℓ \ell ℓ only if the prime-to-ℓ \ell ℓ part of the conductor divides two. The product formula for local Artin symbols handles that possible last prime. The Hasse norm theorem gives b ∈ E × b\in E^\times b ∈ E × with b j ( b ) = A bj(b)=A bj ( b ) = A [25 Part III, Theorem 6.15 and Corollary 4.8] . Consequently ( α / b ) n (\alpha/b)^n ( α / b ) n is norm one. Since C l ( R Σ ) = 0 \mathrm{Cl}(R_\Sigma)=0 Cl ( R Σ ) = 0 , Lemma 4.8 puts its class in W Σ ( 1 ) W_\Sigma^{(1)} W Σ ( 1 ) .
Corollary 15.8. Enumerate the coordinate products ∏ j = 0 r z j e j \prod_{j=0}^{r}z_j^{e_j} ∏ j = 0 r z j e j , 0 ≤ e j < n 0\le e_j<n 0 ≤ e j < n , and their descents over the coefficient field above. The compositum of the field factors L L L satisfying d L / K = O K \mathfrak d_{L/K}=\mathcal O_K d L / K = O K is
( H K + ) n C l + ( K ) . (136) (H_K^+)^{n\mathrm{Cl}^+(K)}. \tag*{(136)} ( H K + ) n Cl + ( K ) . (136) If C l + ( K ) ≃ ⨁ j Z m j \mathrm{Cl}^+(K)\simeq\bigoplus_j\mathbb Z_{m_j} Cl + ( K ) ≃ ⨁ j Z m j , take n = p max j v p ( m j ) n=p^{\max_j v_p(m_j)} n = p m a x j v p ( m j ) for each p ∣ ∏ j m j p\mid\prod_jm_j p ∣ ∏ j m j . These exponents are known from its Smith normal form, and the compositum of the resulting fields is H K + H_K^+ H K + .
Proof. The proposition includes every strict Hilbert character of exponent dividing n n n . Conversely the selected cyclic fields are unramified at finite primes by the discriminant criterion [24 Chapter I, Section 9] . Character recovery proves the equality. □
This is the conductor test of Lemma 14.6 with f = ( 1 ) \mathfrak f=(1) f = ( 1 ) : no separate comparisons on ray generators are needed when all unramified characters are included.
Kummer modules and explicit generators: examples
In each example we give the Kummer module, the class selected by reciprocity and a generator obtained by descent. The displayed bases come from the indicated RM curves. All roots of unity use ζ m = e 2 π i / m \zeta_m=e^{2\pi i/m} ζ m = e 2 π i / m .
The symbols log g ( x ) \log_g(x) log g ( x ) below are finite discrete logarithms: g log g ( x ) = x g^{\log_g(x)}=x g l o g g ( x ) = x in the stated finite field, with the exponent taken modulo ∣ g ∣ |g| ∣ g ∣ . They specify the input ray character χ \chi χ , using the residue-group presentation of § 15.1 ; they are not the Kummer coordinates e j e_j e j . Selection computes those coordinates from C e = b Ce=b C e = b . We give the Jacobi expressions producing each good-prime basis, followed by the selected class and descended generator; the arithmetic evaluations are omitted.
An ordinary quintic
16.1. An ordinary quintic. Let K = Q ( b ) K=\mathbb Q(b) K = Q ( b ) , b = 2 b=\sqrt{2} b = 2 , E = K ( ζ 5 ) E=K(\zeta_5) E = K ( ζ 5 ) , n = 5 n=5 n = 5 and f = ( 1025 ) \mathfrak f=(1025) f = ( 1025 ) . For a totally positive u ∈ O K u\in\mathcal O_K u ∈ O K prime to f \mathfrak f f , define r , s , ℓ ± ∈ Z 5 r,s,\ell_\pm\in\mathbb Z_5 r , s , ℓ ± ∈ Z 5 and χ \chi χ by
u 24 ≡ 1 + 5 ( r + s h ) ( m o d 25 ) , ℓ ± = log 6 ( u ∣ h = ± 17 m o d 41 ) ( m o d 5 ) , u^{24}\equiv1+5(r+sh)\pmod{25},\qquad\ell_\pm=\log_6(u\mid_{h=\pm17}\bmod41)\pmod{5}, u 24 ≡ 1 + 5 ( r + s h ) ( mod 25 ) , ℓ ± = log 6 ( u ∣ h = ± 17 mod 41 ) ( mod 5 ) ,
χ ( [ ( u ) ] ) = 4 s + 2 ℓ + + 3 ℓ − . \chi([(u)])=4s+2\ell_++3\ell_-. χ ([( u )]) = 4 s + 2 ℓ + + 3 ℓ − .
For Σ \Sigma Σ the primes above 41, use C 5 , 8 C_{5,8} C 5 , 8 . On F 41 × = ⟨ 6 ⟩ \mathbb F_{41}^{\times}=\langle6\rangle F 41 × = ⟨ 6 ⟩ , let ϑ ( 6 ) = ζ 5 \vartheta(6)=\zeta_5 ϑ ( 6 ) = ζ 5 , ξ ( 6 ) = ζ 8 \xi(6)=\zeta_8 ξ ( 6 ) = ζ 8 and put
J i , c = J ( ϑ i , ξ c ) , A i = J i , 1 J i , 7 J i , 3 J i , 5 ( i = 1 , 2 ) , J_{i,c}=J(\vartheta^i,\xi^c),\qquad A_i=\frac{J_{i,1}J_{i,7}}{J_{i,3}J_{i,5}}\quad(i=1,2), J i , c = J ( ϑ i , ξ c ) , A i = J i , 3 J i , 5 J i , 1 J i , 7 ( i = 1 , 2 ) ,
with J J J as in (44) . These determinant quotients already give a Kummer basis. Put s 0 = ζ 5 + ζ 5 − 1 s_0=\zeta_5+\zeta_5^{-1} s 0 = ζ 5 + ζ 5 − 1 , t 0 = ζ 5 − ζ 5 − 1 t_0=\zeta_5-\zeta_5^{-1} t 0 = ζ 5 − ζ 5 − 1 . Then
z 1 = A 1 = ( − 11 + 16 s 0 + 14 h t 0 + 2 h s 0 t 0 ) / 41 , z 2 = A 2 = ( − 27 − 16 s 0 − 2 h t 0 + 14 h s 0 t 0 ) / 41 , W Σ = Z 5 [ A 1 ] ⊕ Z 5 [ A 2 ] . (137) \begin{aligned}
z_1=A_1&=(-11+16s_0+14ht_0+2hs_0t_0)/41,\\
z_2=A_2&=(-27-16s_0-2ht_0+14hs_0t_0)/41,\\
W_\Sigma&=\mathbb{Z}_5[A_1]\oplus\mathbb{Z}_5[A_2]. \tag*{(137)}
\end{aligned} z 1 = A 1 z 2 = A 2 W Σ = ( − 11 + 16 s 0 + 14 h t 0 + 2 h s 0 t 0 ) /41 , = ( − 27 − 16 s 0 − 2 h t 0 + 14 h s 0 t 0 ) /41 , = Z 5 [ A 1 ] ⊕ Z 5 [ A 2 ] . (137)
Selection by reciprocity gives
( e 1 , e 2 ) = ( 2 , 1 ) , β χ = A 1 2 A 2 . (138) (e_1,e_2)=(2,1),\qquad\beta_\chi=A_1^2A_2. \tag*{(138)} ( e 1 , e 2 ) = ( 2 , 1 ) , β χ = A 1 2 A 2 . (138)
For α 5 = β χ \alpha^5=\beta_\chi α 5 = β χ , descent gives
θ χ = α + α 2 / A 1 + α 3 / ( A 1 A 2 ) + α − 1 . (139) \theta_\chi=\alpha+\alpha^2/A_1+\alpha^3/(A_1A_2)+\alpha^{-1}. \tag*{(139)} θ χ = α + α 2 / A 1 + α 3 / ( A 1 A 2 ) + α − 1 . (139)
Its minimal polynomial over K K K is
X 5 − 10 X 3 + 380 41 X 2 + 185 41 X − 404 1681 . X^5-10X^3+\frac{380}{41}X^2+\frac{185}{41}X-\frac{404}{1681}. X 5 − 10 X 3 + 41 380 X 2 + 41 185 X − 1681 404 .
The cyclic extension H χ = K ( θ χ ) H_\chi=K(\theta_\chi) H χ = K ( θ χ ) is totally real, with conductor ( 1025 ) (1025) ( 1025 ) and relative discriminant ( 1025 ) 4 (1025)^4 ( 1025 ) 4 .
An ordinary cubic without saturation
Let K = Q ( h ) K=\mathbb{Q}(h) K = Q ( h ) , h = 2 h=\sqrt{2} h = 2 , E = K ( ζ 3 ) E=K(\zeta_3) E = K ( ζ 3 ) , n = 3 n=3 n = 3 , s = h ( 2 ζ 3 + 1 ) = − 6 s=h(2\zeta_3+1)=\sqrt{-6} s = h ( 2 ζ 3 + 1 ) = − 6 and f = ( 63 ) \mathfrak{f}=(63) f = ( 63 ) . For a totally positive u = a + b h u=a+bh u = a + bh prime to f \mathfrak{f} f , define r , t , ℓ ± ∈ Z 3 r,t,\ell_{\pm}\in\mathbb{Z}_3 r , t , ℓ ± ∈ Z 3 and χ \chi χ by
u 8 ≡ 1 + 3 ( r + t h ) ( m o d 9 ) , ℓ ± = log 3 ( a ± 3 b m o d 7 ) ( m o d 3 ) , u^8\equiv1+3(r+th)\pmod9,\qquad\ell_{\pm}=\log_3(a\pm3b\mathbin{\mathrm{mod}}7)\pmod3, u 8 ≡ 1 + 3 ( r + t h ) ( mod 9 ) , ℓ ± = log 3 ( a ± 3 b mod 7 ) ( mod 3 ) ,
χ ( [ ( u ) ] ) = t + ℓ + − ℓ − . \chi([(u)])=t+\ell_+-\ell_-. χ ([( u )]) = t + ℓ + − ℓ − .
For Σ \Sigma Σ the primes above 7 7 7 , use C 3 , 8 C_{3,8} C 3 , 8 over k = F 7 ( ϱ ) k=\mathbb{F}_7(\varrho) k = F 7 ( ϱ ) with ϱ 2 = − 1 \varrho^2=-1 ϱ 2 = − 1 and k × = ⟨ 2 + ϱ ⟩ k^\times=\langle2+\varrho\rangle k × = ⟨ 2 + ϱ ⟩ . Let ϑ ( 2 + ϱ ) = ζ 3 \vartheta(2+\varrho)=\zeta_3 ϑ ( 2 + ϱ ) = ζ 3 , ξ ( 2 + ϱ ) = ζ 8 \xi(2+\varrho)=\zeta_8 ξ ( 2 + ϱ ) = ζ 8 and put J c = J ( ϑ , ξ c ) J_c=J(\vartheta,\xi^c) J c = J ( ϑ , ξ c ) . The determinant quotient gives
A = J 1 J 7 J 3 J 5 = 1 − 20 s 49 , W Σ = Z 3 [ A ] , z 1 = A , e 1 = 1 , β χ = A . A=\frac{J_1J_7}{J_3J_5}=\frac{1-20s}{49},\qquad W_\Sigma=\mathbb{Z}_3[A],\qquad z_1=A,\qquad e_1=1,\qquad\beta_\chi=A. A = J 3 J 5 J 1 J 7 = 49 1 − 20 s , W Σ = Z 3 [ A ] , z 1 = A , e 1 = 1 , β χ = A .
Here [ U Σ : ⟨ − 1 , A ⟩ ] = B = 2 [U_\Sigma:\langle-1,A\rangle]=B=2 [ U Σ : ⟨ − 1 , A ⟩] = B = 2 , so no internal cube root is needed. For α 3 = A \alpha^3=A α 3 = A ,
θ χ = α + α − 1 , θ χ 3 − 3 θ χ − 2 49 = 0. \theta_\chi=\alpha+\alpha^{-1},\qquad\theta_\chi^3-3\theta_\chi-\frac{2}{49}=0. θ χ = α + α − 1 , θ χ 3 − 3 θ χ − 49 2 = 0.
The field H χ = K ( θ χ ) H_\chi=K(\theta_\chi) H χ = K ( θ χ ) is totally real and cyclic cubic, with conductor ( 63 ) (63) ( 63 ) and relative discriminant ( 63 ) 2 (63)^2 ( 63 ) 2 .
A dyadic character on the RM curve C 8 , 8 C_{8,8} C 8 , 8
Let K = Q ( h ) K=\mathbb{Q}(h) K = Q ( h ) , h = 2 h=\sqrt{2} h = 2 , E = Q ( ζ 8 ) = K ( i ) E=\mathbb{Q}(\zeta_8)=K(i) E = Q ( ζ 8 ) = K ( i ) , n = 4 n=4 n = 4 and f = ( 17 ) \mathfrak{f}=(17) f = ( 17 ) . For a totally positive generator a + b h a+bh a + bh of an ideal prime to 17 17 17 , define
χ ( [ ( a + b h ) ] ) = log 3 ( a + 6 b ) − log 3 ( a − 6 b ) ( m o d 4 ) , \chi([(a+bh)])=\log_3(a+6b)-\log_3(a-6b)\pmod4, χ ([( a + bh )]) = log 3 ( a + 6 b ) − log 3 ( a − 6 b ) ( mod 4 ) ,
where the logarithms are in F 17 × \mathbb{F}_{17}^{\times} F 17 × . Put w = i h w=ih w = ih , and let Σ \Sigma Σ be the primes above 17 17 17 . For C 8 , 8 C_{8,8} C 8 , 8 , choose λ ( 3 ) = ζ 8 \lambda(3)=\zeta_8 λ ( 3 ) = ζ 8 on F 17 × \mathbb{F}_{17}^{\times} F 17 × . The Jacobi sums and positive RM quotients are
J 2 , 1 = J ( λ 2 , λ ) = 1 − 4 i , J 7 , 7 = J ( λ 7 , λ 7 ) = − 3 + 2 w , J_{2,1}=J(\lambda^2,\lambda)=1-4i,\qquad J_{7,7}=J(\lambda^7,\lambda^7)=-3+2w, J 2 , 1 = J ( λ 2 , λ ) = 1 − 4 i , J 7 , 7 = J ( λ 7 , λ 7 ) = − 3 + 2 w ,
A 1 = J 7 , 7 / j ( J 7 , 7 ) , A 2 = J 2 , 1 J 7 , 7 / j ( J 2 , 1 J 7 , 7 ) . A_1=J_{7,7}/j(J_{7,7}),\qquad A_2=J_{2,1}J_{7,7}/j(J_{2,1}J_{7,7}). A 1 = J 7 , 7 / j ( J 7 , 7 ) , A 2 = J 2 , 1 J 7 , 7 / j ( J 2 , 1 J 7 , 7 ) .
An internal square root gives z 1 = − j ( J 2 , 1 ) J 7 , 7 / 17 z_1=-j(J_{2,1})J_{7,7}/17 z 1 = − j ( J 2 , 1 ) J 7 , 7 /17 and z 2 = A 1 / A 2 z_2=A_1/A_2 z 2 = A 1 / A 2 ; thus z 1 2 = A 1 2 / A 2 z_1^2=A_1^2/A_2 z 1 2 = A 1 2 / A 2 .
Explicitly,
z 0 = ζ 8 , z 1 = − ( 1 + 4 i ) ( − 3 + 2 w ) 17 , z 2 = − 15 + 8 i 17 , z_0=\zeta_8,\qquad z_1=-\frac{(1+4i)(-3+2w)}{17},\qquad z_2=\frac{-15+8i}{17}, z 0 = ζ 8 , z 1 = − 17 ( 1 + 4 i ) ( − 3 + 2 w ) , z 2 = 17 − 15 + 8 i ,
W Σ ( 1 ) = Z 4 [ z 0 ] ⊕ Z 4 [ z 1 ] ⊕ Z 4 [ z 2 ] . W_\Sigma^{(1)}=\mathbb{Z}_4[z_0]\oplus\mathbb{Z}_4[z_1]\oplus\mathbb{Z}_4[z_2]. W Σ ( 1 ) = Z 4 [ z 0 ] ⊕ Z 4 [ z 1 ] ⊕ Z 4 [ z 2 ] .
Selection gives ( e 0 , e 1 , e 2 ) = ( 0 , 2 , − 1 ) (e_0,e_1,e_2)=(0,2,-1) ( e 0 , e 1 , e 2 ) = ( 0 , 2 , − 1 ) and
β χ = z 1 2 z 2 − 1 = 1 − 12 w 17 . \beta_\chi=z_1^2z_2^{-1}=\frac{1-12w}{17}. β χ = z 1 2 z 2 − 1 = 17 1 − 12 w .
For α 4 = β χ \alpha^4=\beta_\chi α 4 = β χ , the character selects the lift d j ( α ) = − α − 1 d_j(\alpha)=-\alpha^{-1} d j ( α ) = − α − 1 . Hence
θ χ = α − α − 1 , θ χ 4 + 4 θ χ 2 + 32 17 = 0. \theta_\chi=\alpha-\alpha^{-1},\qquad\theta_\chi^4+4\theta_\chi^2+\frac{32}{17}=0. θ χ = α − α − 1 , θ χ 4 + 4 θ χ 2 + 17 32 = 0.
The extension H χ / K H_\chi/K H χ / K is cyclic quartic, with finite conductor ( 17 ) (17) ( 17 ) and relative discriminant ( 17 ) 3 (17)^3 ( 17 ) 3 ; both real places ramify. The lift d j ( α ) = α − 1 d_j(\alpha)=\alpha^{-1} d j ( α ) = α − 1 instead gives conductor divisible by the prime above two and is excluded. This is why descent uses the full character, not just its restriction to E E E .
Shintani’s example over Q ( 5 ) \mathbb{Q}(\sqrt{5}) Q ( 5 )
Let K = Q ( d ) K=\mathbb{Q}(d) K = Q ( d ) , d = 5 d=\sqrt{5} d = 5 , ϕ = ( 1 + d ) / 2 \phi=(1+d)/2 ϕ = ( 1 + d ) /2 , ϵ = ϕ 2 \epsilon=\phi^2 ϵ = ϕ 2 and p = ( 4 − d ) \mathfrak{p}=(4-d) p = ( 4 − d ) , of norm 11. The group G p + G^+_{\mathfrak{p}} G p + has order two; let χ \chi χ be its nontrivial character. Take n = 2 n=2 n = 2 , E = Q ( ζ 5 ) E=\mathbb{Q}(\zeta_5) E = Q ( ζ 5 ) and Σ \Sigma Σ the primes above 11. Use the quotient C 10 , 5 → C 5 , 5 C_{10,5}\to C_{5,5} C 10 , 5 → C 5 , 5 , ( Y , Z ) ↦ ( Y 2 , Z ) (Y,Z)\mapsto(Y^2,Z) ( Y , Z ) ↦ ( Y 2 , Z ) ; the coefficient field is unchanged since Q ( ζ 10 ) = E \mathbb{Q}(\zeta_{10})=E Q ( ζ 10 ) = E . On C 5 , 5 C_{5,5} C 5 , 5 take λ ( 2 ) = ζ 5 \lambda(2)=\zeta_5 λ ( 2 ) = ζ 5 on F 11 × \mathbb{F}_{11}^{\times} F 11 × . The Jacobi values J 1 = J ( λ , λ ) J_1=J(\lambda,\lambda) J 1 = J ( λ , λ ) and J 2 = J ( λ 2 , λ ) J_2=J(\lambda^2,\lambda) J 2 = J ( λ 2 , λ ) are
J 1 = − ζ 5 3 − 2 ζ 5 2 + 2 ζ 5 , J 2 = 2 ζ 5 3 + 4 ζ 5 2 + ζ 5 + 2 J_1=-\zeta_5^3-2\zeta_5^2+2\zeta_5,\qquad J_2=2\zeta_5^3+4\zeta_5^2+\zeta_5+2 J 1 = − ζ 5 3 − 2 ζ 5 2 + 2 ζ 5 , J 2 = 2 ζ 5 3 + 4 ζ 5 2 + ζ 5 + 2
and give the basis
z 0 = − ζ 5 , z 1 = J 1 J 2 / 11 , z 2 = J 1 / j ( J 1 ) , W Σ ( 1 ) = ⨁ j = 0 2 Z 2 [ z j ] . z_0=-\zeta_5,\qquad z_1=J_1J_2/11,\qquad z_2=J_1/j(J_1),\qquad W^{(1)}_{\Sigma}=\bigoplus_{j=0}^{2}\mathbb{Z}_2[z_j]. z 0 = − ζ 5 , z 1 = J 1 J 2 /11 , z 2 = J 1 / j ( J 1 ) , W Σ ( 1 ) = j = 0 ⨁ 2 Z 2 [ z j ] .
The corrected character ψ \psi ψ has coordinates ( 1 , 1 , 0 ) (1,1,0) ( 1 , 1 , 0 ) . After restoring its signature factor,
β χ = d z 0 z 1 ≡ 3 ϕ − 2 ( m o d E × 2 ) . (140) \beta_\chi=dz_0z_1\equiv3\phi-2\pmod{E^{\times2}}. \tag*{(140)} β χ = d z 0 z 1 ≡ 3 ϕ − 2 ( mod E × 2 ) . (140)
Descent gives
θ χ = 3 ϕ − 2 , K ( θ χ ) = H χ = K p + . (141) \theta_\chi=\sqrt{3\phi-2},\qquad K(\theta_\chi)=H_\chi=K^+_{\mathfrak{p}}. \tag*{(141)} θ χ = 3 ϕ − 2 , K ( θ χ ) = H χ = K p + . (141)
Its relative discriminant is p \mathfrak{p} p . With the square root positive at σ 1 \sigma_1 σ 1 , Shintani’s evaluated double-sine invariant is
X S h = ϵ + 3 ϕ − 2 2 , K ( X S h ) = K p + . (142) X_{\mathrm{Sh}}=\frac{\epsilon+\sqrt{3\phi-2}}{2},\qquad K(X_{\mathrm{Sh}})=K^+_{\mathfrak{p}}. \tag*{(142)} X Sh = 2 ϵ + 3 ϕ − 2 , K ( X Sh ) = K p + . (142)
See [34 Section 3, Example 2, pp. 191–192] and [43 Section 2, (2.1)–(2.2)] . The minimal polynomial of X S h X_{\mathrm{Sh}} X Sh over Q \mathbb{Q} Q is X 4 − 3 X 3 + 3 X 2 − 3 X + 1 X^4-3X^3+3X^2-3X+1 X 4 − 3 X 3 + 3 X 2 − 3 X + 1 .
An RM input at a bad reduction prime
Let K = Q ( h ) K=\mathbb{Q}(h) K = Q ( h ) , h = 6 h=\sqrt{6} h = 6 , n = 3 n=3 n = 3 and f = ( 9 ) \mathfrak{f}=(9) f = ( 9 ) . The ray class group is Z 6 ⊕ Z 3 \mathbb{Z}_6\oplus\mathbb{Z}_3 Z 6 ⊕ Z 3 , with generators [ ( − 4 + 10 h ) ] [(-4+10h)] [( − 4 + 10 h )] , [ ( 17 − 2 h ) ] [(17-2h)] [( 17 − 2 h )] ; let χ \chi χ take value one on both. Put
E = K ( ζ 3 ) = Q ( s , t ) , s 2 = − 2 , t 2 = − 3 , h = − s t , ζ 3 = − 1 + t 2 , b = 1 + s 1 − s , E=K(\zeta_3)=\mathbb{Q}(s,t),\qquad s^2=-2,\qquad t^2=-3,\qquad h=-st,\qquad\zeta_3=\frac{-1+t}{2},\qquad b=\frac{1+s}{1-s}, E = K ( ζ 3 ) = Q ( s , t ) , s 2 = − 2 , t 2 = − 3 , h = − s t , ζ 3 = 2 − 1 + t , b = 1 − s 1 + s ,
with s , t s,t s , t of positive imaginary part, and let Σ \Sigma Σ be the primes above three. Here the RM curve C 24 , 24 C_{24,24} C 24 , 24 is evaluated through its CM ideal, as in § 10.5 . The principalization can be written out completely. In F = Q ( w ) F=\mathbb{Q}(w) F = Q ( w ) , w = ζ 24 w=\zeta_{24} w = ζ 24 , identify s = w 3 + w 9 s=w^3+w^9 s = w 3 + w 9 , and denote the two primes above three by q + , q − \mathfrak{q}_+,\mathfrak{q}_- q + , q − , where v q + ( 1 + s ) = 2 v_{\mathfrak{q}_+}(1+s)=2 v q + ( 1 + s ) = 2 , and choose q 3 = q + \mathfrak{q}_3=\mathfrak{q}_+ q 3 = q + ; then q − = j ( q + ) \mathfrak{q}_-=j(\mathfrak{q}_+) q − = j ( q + ) . For v = 3 v=3 v = 3 , the positive RM indices are I 3 + = { 1 , 5 , 19 , 23 } I_3^+=\{1,5,19,23\} I 3 + = { 1 , 5 , 19 , 23 } . The CM-type formula gives
H 3 , 3 = 12 [ q + ] + 4 [ q − ] , a 3 , 3 = q + 12 q − 4 = ( 9 ( 1 + s ) 4 ) . H_{3,3}=12[\mathfrak{q}_+]+4[\mathfrak{q}_-],\qquad\mathfrak{a}_{3,3}=\mathfrak{q}_+^{12}\mathfrak{q}_-^4=(9(1+s)^4). H 3 , 3 = 12 [ q + ] + 4 [ q − ] , a 3 , 3 = q + 12 q − 4 = ( 9 ( 1 + s ) 4 ) .
Indeed v q ± ( 3 ) = 2 v_{\mathfrak{q}_{\pm}}(3)=2 v q ± ( 3 ) = 2 , while 1 + s 1+s 1 + s has valuations 2 , 0 2,0 2 , 0 . Thus its ideal class order is one. Choosing x = 9 ( 1 + s ) 4 x=9(1+s)^4 x = 9 ( 1 + s ) 4 gives
x j ( x ) = b 4 , A 1 = N F / E ( x j ( x ) ) = b 8 , [ F : E ] = 2. (143) \frac{x}{j(x)}=b^4,\qquad A_1=N_{F/E}\left(\frac{x}{j(x)}\right)=b^8,\qquad[F:E]=2. \tag*{(143)} j ( x ) x = b 4 , A 1 = N F / E ( j ( x ) x ) = b 8 , [ F : E ] = 2. (143)
This is an explicit algebraic input; no good-reduction extension or bad-prime Frobenius value has been computed. It gives
z 0 = ζ 3 , z 1 = A 1 , W Σ ( 1 ) = Z 3 [ z 0 ] ⊕ Z 3 [ z 1 ] . z_0=\zeta_3,\qquad z_1=A_1,\qquad W_{\Sigma}^{(1)}=\mathbb{Z}_3[z_0]\oplus\mathbb{Z}_3[z_1]. z 0 = ζ 3 , z 1 = A 1 , W Σ ( 1 ) = Z 3 [ z 0 ] ⊕ Z 3 [ z 1 ] .
Here B = 16 B=16 B = 16 , so this input requires no internal cube root. Its relation to the geometric Frobenius quotient is (93) . Selection gives ( e 0 , e 1 ) = ( 0 , 2 ) (e_0,e_1)=(0,2) ( e 0 , e 1 ) = ( 0 , 2 ) and β χ = b 16 \beta_{\chi}=b^{16} β χ = b 16 . For α 3 = β χ \alpha^3=\beta_{\chi} α 3 = β χ , put η = α / b 5 \eta=\alpha/b^5 η = α / b 5 . Then
η 3 = b , θ χ = η + η − 1 , θ χ 3 − 3 θ χ + 2 3 = 0. \eta^3=b,\qquad\theta_{\chi}=\eta+\eta^{-1},\qquad\theta_{\chi}^3-3\theta_{\chi}+\frac{2}{3}=0. η 3 = b , θ χ = η + η − 1 , θ χ 3 − 3 θ χ + 3 2 = 0.
The field H χ = K ( θ χ ) H_{\chi}=K(\theta_{\chi}) H χ = K ( θ χ ) is totally real and cyclic cubic, with conductor ( 9 ) (9) ( 9 ) and relative discriminant ( 81 ) (81) ( 81 ) .
For comparison, the alternative of § 11.6 uses C 8 C_8 C 8 at the same prime three, where this curve has good reduction. In F 9 = F 3 ( ϱ ) \mathbb{F}_9=\mathbb{F}_3(\varrho) F 9 = F 3 ( ϱ ) , ϱ 2 = − 1 \varrho^2=-1 ϱ 2 = − 1 , choose λ ( 2 + ϱ ) = ζ 8 \lambda(2+\varrho)=\zeta_8 λ ( 2 + ϱ ) = ζ 8 . Since E ∩ Q ( ζ 8 ) = Q ( s ) E\cap\mathbb{Q}(\zeta_8)=\mathbb{Q}(s) E ∩ Q ( ζ 8 ) = Q ( s ) , direct Jacobi evaluation gives
J ( λ , λ ) = ( 1 − s ) 2 = − 1 − 2 s , A d i a g = N Q ( ζ 8 ) / Q ( s ) ( J ( λ , λ ) j ( J ( λ , λ ) ) ) = b − 4 . J(\lambda,\lambda)=(1-s)^2=-1-2s,\qquad A^{\mathrm{diag}}=N_{\mathbb{Q}(\zeta_8)/\mathbb{Q}(s)}\left(\frac{J(\lambda,\lambda)}{j(J(\lambda,\lambda))}\right)=b^{-4}. J ( λ , λ ) = ( 1 − s ) 2 = − 1 − 2 s , A diag = N Q ( ζ 8 ) / Q ( s ) ( j ( J ( λ , λ )) J ( λ , λ ) ) = b − 4 .
Thus A 1 = ( A d i a g ) − 2 A_1=(A^{\mathrm{diag}})^{-2} A 1 = ( A diag ) − 2 and [ A 1 ] = [ A d i a g ] [A_1]=[A^{\mathrm{diag}}] [ A 1 ] = [ A diag ] in E × / E × 3 E^{\times}/E^{\times3} E × / E × 3 . Both inputs give the displayed Kummer basis and the same selected extension. The RM route computes a principal ideal generator at level 24; the diagonal route evaluates a Jacobi sum at level 8.
Computing ray characters from ideals and units
This standard algorithm uses the ordinary class group and both real signs, giving the same ray group as the narrow-class presentation (10) [8 Section 4.3.1] [16 Section 3] . It fixes the residue and sign conventions used in the construction.
Proposition A.1. Arithmetic in K K K gives a finite presentation of G f + G_{\mathfrak{f}}^+ G f + and all homomorphisms G f + → Z n G_{\mathfrak{f}}^+\to\mathbb{Z}_n G f + → Z n , with their values on ideals, without constructing a class field. More explicitly:
Choose an invariant-factor presentation of the ordinary ideal class group, with representatives a 1 , … , a s \mathfrak{a}_1,\ldots,\mathfrak{a}_s a 1 , … , a s prime to f \mathfrak{f} f , of orders d i d_i d i , and find a i d i = ( α i ) \mathfrak{a}_i^{d_i}=(\alpha_i) a i d i = ( α i ) . Compute a fundamental unit ϵ 0 \epsilon_0 ϵ 0 , so O K × = ⟨ − 1 , ϵ 0 ⟩ \mathcal{O}_K^\times=\langle-1,\epsilon_0\rangle O K × = ⟨ − 1 , ϵ 0 ⟩ .
Compute the finite abelian group and the homomorphism
A f = ( O K / f ) × × { ± 1 } 2 , A_{\mathfrak{f}}=(\mathcal{O}_K/\mathfrak{f})^\times\times\{\pm1\}^2, A f = ( O K / f ) × × { ± 1 } 2 , a : { x ∈ K × : v p ( x ) = 0 ( p ∣ f ) } ⟶ A f , a:\{x\in K^\times:v_{\mathfrak{p}}(x)=0\ (\mathfrak{p}\mid\mathfrak{f})\}\longrightarrow A_{\mathfrak{f}}, a : { x ∈ K × : v p ( x ) = 0 ( p ∣ f )} ⟶ A f , x ⟼ ( x m o d f , sgn σ 1 ( x ) , sgn σ 2 ( x ) ) . x\longmapsto(x\bmod\mathfrak{f},\operatorname{sgn}\sigma_1(x),\operatorname{sgn}\sigma_2(x)). x ⟼ ( x mod f , sgn σ 1 ( x ) , sgn σ 2 ( x )) . For f = O K \mathfrak{f}=\mathcal{O}_K f = O K the first factor is the trivial group. Take generators and defining relations of A f A_{\mathfrak{f}} A f , add symbols e i e_i e i for a i \mathfrak{a}_i a i , and impose the further relations
a ( − 1 ) = a ( ϵ 0 ) = 0 , d i e i = a ( α i ) . a(-1)=a(\epsilon_0)=0,\qquad d_i e_i=a(\alpha_i). a ( − 1 ) = a ( ϵ 0 ) = 0 , d i e i = a ( α i ) . Here the finite abelian group is written additively. The resulting group is G f + G_{\mathfrak{f}}^+ G f + .
(3) Let R ∈ Mat t × s ( Z ) R \in\operatorname{Mat}_{t \times s}(\mathbb{Z}) R ∈ Mat t × s ( Z ) have these relations as columns, in a list of t t t generators. Thus G f + ≃ Z t / R Z s G_{\mathfrak{f}}^{+} \simeq\mathbb{Z}^{t}/R\mathbb{Z}^{s} G f + ≃ Z t / R Z s . A character is exactly a row vector c ∈ ( Z n ) t c \in(\mathbb{Z}_{n})^{t} c ∈ ( Z n ) t satisfying c R = 0 cR=0 c R = 0 , and, for a column vector x x x ,
χ c ( [ x ] ) = c x ( m o d n ) . (144) \chi_c([x])=cx \pmod n. \tag*{(144)} χ c ([ x ]) = c x ( mod n ) . (144)
Smith normal form lists these characters as in Proposition 15.1.
(4) To evaluate it on a ∈ I K ( f ) \mathfrak{a} \in I_K(\mathfrak{f}) a ∈ I K ( f ) , find m i m_i m i and α ∈ K × \alpha\in K^{\times} α ∈ K × such that
a = ( α ) ∏ i a i m i . \mathfrak{a}=(\alpha)\prod_i\mathfrak{a}_i^{m_i}. a = ( α ) i ∏ a i m i .
The coordinate vector is that of a ( α ) + ∑ i m i e i \mathfrak{a}(\alpha)+\sum_i m_i e_i a ( α ) + ∑ i m i e i in the displayed presentation. Apply the preceding formula.
Proof. The standard ray exact sequence is
A f / α ( O K × ) ↪ G f + ⟶ Cl ( O K ) ⟶ 0 ; A_{\mathfrak{f}}/\alpha(O_K^{\times}) \hookrightarrow G_{\mathfrak{f}}^{+} \longrightarrow\operatorname{Cl}(O_K) \longrightarrow0; A f / α ( O K × ) ↪ G f + ⟶ Cl ( O K ) ⟶ 0 ;
see [8 Section 4.3.1] . Lifting ideal-class generators gives d i e i = a ( α i ) d_i e_i=\mathfrak{a}(\alpha_i) d i e i = a ( α i ) with our principal-ideal convention. The class-group, unit, and principal-ideal computations are standard [7 Chapters 5–6] . □
The CM interpretation of the Frobenius ideal
The construction uses the local CM ideal formula in Proposition 11.2; the proof there is independent of this appendix. We record its relation with the classical reflex norm, which is different from the coefficient norm N F / E N_{F/E} N F / E in (84) .
For the ordinary factor primitive in both deck indices, put F = Q ( ζ Δ n ) F=\mathbb{Q}(\zeta_{\Delta n}) F = Q ( ζ Δ n ) . Its CM type is
Φ = { σ u : u ∈ Z Δ n × , { u / n } + { u / Δ } < 1 } . (145) \Phi=\{\sigma_u:u\in\mathbb{Z}_{\Delta n}^{\times},\ \{u/n\}+\{u/\Delta\}<1\}. \tag*{(145)} Φ = { σ u : u ∈ Z Δ n × , { u / n } + { u /Δ } < 1 } . (145)
Here an automorphism of the curve acts on the Jacobian by pushing forward degree-zero divisors. Its action on the tangent space has the eigenvalues of its pullback on holomorphic differentials. Those differentials give (B.1) [21 pp. 1183–1184] ; the factor need not be simple. The uniform factor has the CM types (90) , with the same action convention.
For this abelian CM field, define
S Φ = { s ∈ Gal ( F / Q ) : s Φ = Φ } , F ∗ = F S Φ , Φ ∗ = { σ − 1 ∣ F ∗ : σ ∈ Φ } , S_{\Phi}=\{s\in\operatorname{Gal}(F/\mathbb{Q}):s\Phi=\Phi\},\qquad F^{*}=F^{S_{\Phi}},\qquad\Phi^{*}=\{\sigma^{-1}|_{F^{*}}:\sigma\in\Phi\}, S Φ = { s ∈ Gal ( F / Q ) : s Φ = Φ } , F ∗ = F S Φ , Φ ∗ = { σ − 1 ∣ F ∗ : σ ∈ Φ } ,
using distinct restrictions in the last set. The reflex norm is
N Φ ∗ : ( F ∗ ) × ⟶ F × , a ⟼ ∏ ψ ∈ Φ ∗ ψ ( a ) , N_{\Phi^{*}}:(F^{*})^{\times}\longrightarrow F^{\times},\qquad a\longmapsto\prod_{\psi\in\Phi^{*}}\psi(a), N Φ ∗ : ( F ∗ ) × ⟶ F × , a ⟼ ψ ∈ Φ ∗ ∏ ψ ( a ) ,
with the analogous definition on ideals. At a prime q ∤ Δ n \mathfrak{q}\nmid\Delta n q ∤ Δ n of F F F , the Jacobi factorization is
N Φ ∗ ( N F / F ∗ q ) O F = ∏ σ ∈ Φ σ − 1 q = ( J q , 1 ( 1 ) ) . (146) N_{\Phi^{*}}(N_{F/F^{*}}\mathfrak{q})O_F=\prod_{\sigma\in\Phi}\sigma^{-1}\mathfrak{q}=(J_{\mathfrak{q},1}(1)). \tag*{(146)} N Φ ∗ ( N F / F ∗ q ) O F = σ ∈ Φ ∏ σ − 1 q = ( J q , 1 ( 1 )) . (146)
Indeed N F / F ∗ q N_{F/F^{*}}\mathfrak{q} N F / F ∗ q extends to ∏ h ∈ S Φ h q \prod_{h\in S_{\Phi}}h\mathfrak{q} ∏ h ∈ S Φ h q ; the reflex type chooses exactly the S Φ S_{\Phi} S Φ -cosets in Φ − 1 \Phi^{-1} Φ − 1 . This proves the middle equality with all prime multiplicities, without assuming complete splitting or a primitive CM type. It is the abelian form of the Shimura–Taniyama ideal formula [32, Chapter II, Section 8; Chapter III, Section 13]. At ramified primes we use its local form (98) , as explained in § 11.1 .
Thus the reflex norm describes the Frobenius ideal from the CM type. The coefficient norm N F / E N_{F/E} N F / E instead puts a normalized determinant value in the chosen Kummer coefficient field. At good primes the normalized determinant is a quotient of Jacobi products; at bad primes § 10.5 computes substitute values by principalizing the CM ideals. No such substitute is asserted to be a Frobenius eigenvalue.
Classical growth of ray towers
Fix a nonzero integral ideal f \mathfrak{f} f , and write
S = { p : p ∣ f } , K a = K f a + , K ∞ = ⋃ a ≥ 1 K a . S=\{p:p\mid\mathfrak{f}\},\qquad K_a=K_{\mathfrak{f}^a}^{+},\qquad K_\infty=\bigcup_{a\geq1}K_a. S = { p : p ∣ f } , K a = K f a + , K ∞ = a ≥ 1 ⋃ K a .
Thus K ∞ K_\infty K ∞ is the maximal abelian extension of K K K unramified at every finite prime outside S S S . We ask which part of this tower can keep growing when the support S S S is fixed.
Say that S S S contains the full support above p p p if every prime of K K K above the rational prime p p p belongs to S S S . Put
T = { p : S contains the full support above p } , K T c y c = K ( Q ∞ ( p ) : p ∈ T ) , (147) T=\{p:S\text{ contains the full support above }p\},
\qquad
K_T^{\mathrm{cyc}}=K\bigl(\mathbb{Q}^{(p)}_\infty:p\in T\bigr), \tag*{(147)} T = { p : S contains the full support above p } , K T cyc = K ( Q ∞ ( p ) : p ∈ T ) , (147)
where Q ∞ ( p ) \mathbb{Q}^{(p)}_\infty Q ∞ ( p ) is the cyclotomic Z ^ p \widehat{\mathbb{Z}}_p Z p -extension of Q \mathbb{Q} Q ; put K T c y c = K K_T^{\mathrm{cyc}}=K K T cyc = K when T T T is empty. An inert or ramified prime in S S S supplies full support above its rational prime. If p p p splits, both primes above p p p must lie in S S S .
The following is a classical consequence of the structure of abelian S S S -ramification [19 Section 1.3, equations (1.1) and (1.3), pp. 252–253] and of cyclotomic Z ^ p \widehat{\mathbb{Z}}_p Z p -extensions [38 Section 13.1] . It is independent of the Fermat construction.
Theorem C.1. The extension K ∞ / K T c y c K_\infty/K_T^{\mathrm{cyc}} K ∞ / K T cyc is finite. There is an integer a 0 ≥ 1 a_0\geq1 a 0 ≥ 1 such that
K ∞ = K a 0 K T c y c , K a = K a 0 ( K a ∩ K T c y c ) ( a ≥ a 0 ) . (148) K_\infty=K_{a_0}K_T^{\mathrm{cyc}},\qquad
K_a=K_{a_0}\bigl(K_a\cap K_T^{\mathrm{cyc}}\bigr)\qquad(a\geq a_0). \tag*{(148)} K ∞ = K a 0 K T cyc , K a = K a 0 ( K a ∩ K T cyc ) ( a ≥ a 0 ) . (148) The tower is eventually constant exactly when T T T is empty.
In particular, after one finite strict ray class field K a 0 K_{a_0} K a 0 is known, every further layer requires only cyclotomic adjunctions. For f = m O K \mathfrak{f}=m\mathcal{O}_K f = m O K , all rational primes dividing m m m have full support, so T = { p : p ∣ m } T=\{p:p\mid m\} T = { p : p ∣ m } . In contrast, a tower supported at just one prime above a split rational prime eventually stops.
Proof. Put G = Gal ( K ∞ / K ) G=\operatorname{Gal}(K_\infty/K) G = Gal ( K ∞ / K ) , and let G p G_p G p be its maximal pro-p p p quotient. The cited structure and rank formulas give a finitely generated Z ^ p \widehat{\mathbb{Z}}_p Z p -module with
rank Z ^ p G p = { 1 , p ∈ T , 0 , p ∉ T . (149) \operatorname{rank}_{\widehat{\mathbb{Z}}_p}G_p=
\begin{cases}
1,&p\in T,\\
0,&p\notin T.
\end{cases} \tag*{(149)} rank Z p G p = { 1 , 0 , p ∈ T , p ∈ / T . (149) Indeed, the rank is the local degree d p = ∑ p ∈ S , p ∣ p [ K p : Q p ] d_p=\sum_{p\in S,\mathfrak{p}\mid p}[K_{\mathfrak{p}}:\mathbb{Q}_p] d p = ∑ p ∈ S , p ∣ p [ K p : Q p ] minus the dimension of the span of the unit logarithms. This dimension is one when d p > 0 d_p>0 d p > 0 : the fundamental unit cannot be a local root of unity. Since K K K is real quadratic, d p d_p d p is two exactly when p ∈ T p\in T p ∈ T . Gras uses unramified real places; allowing their ramification adds only a finite 2 2 2 -group, so the same conclusion holds here, including p = 2 p=2 p = 2 .
The ray sequence (10) also shows that only finitely many G p G_p G p are nontrivial: outside the rational primes below S S S , their orders divide the product of the narrow class number and the orders of the residue-unit groups at S S S . For p ∈ T p\in T p ∈ T , the cyclotomic Z ^ p \widehat{\mathbb{Z}}_p Z p -extension of K K K lies in K ∞ K_\infty K ∞ . Restriction to it is a surjection from G p G_p G p onto Z ^ p \widehat{\mathbb{Z}}_p Z p , with finite kernel by (C.3). The other G p G_p G p are finite. Consequently H = Gal ( K ∞ / K T c y c ) H=\operatorname{Gal}(K_\infty/K_T^{\mathrm{cyc}}) H = Gal ( K ∞ / K T cyc ) is finite.
Let N a = Gal ( K ∞ / K a ) N_a=\operatorname{Gal}(K_\infty/K_a) N a = Gal ( K ∞ / K a ) . Since ⋂ a N a = 1 \bigcap_a N_a=1 ⋂ a N a = 1 , choose a 0 a_0 a 0 with N a 0 ∩ H = 1 N_{a_0}\cap H=1 N a 0 ∩ H = 1 . Galois correspondence gives K ∞ = K a 0 K T c y c K_\infty=K_{a_0}K_T^{\mathrm{cyc}} K ∞ = K a 0 K T cyc . For a ≥ a 0 a\ge a_0 a ≥ a 0 , the elementary identity N a 0 ∩ H N a = N a N_{a_0}\cap HN_a=N_a N a 0 ∩ H N a = N a gives the second formula in (C.2). Finally, K T c y c / K K_T^{\mathrm{cyc}}/K K T cyc / K is infinite exactly when T ≠ ∅ T\ne\varnothing T = ∅ .
The support is fixed throughout. The initial layer K a 0 K_{a_0} K a 0 depends on that support, and the theorem does not specify a minimal a 0 a_0 a 0 .
Institut de Recherche Mathématique Avancée, Université de Strasbourg, CNRS, UMR 7501, 7 rue René Descartes, 67000 Strasbourg, France
Email address: yalkinoglu@math.unistra.fr