This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Twisted Fermat curves over
totally real fields

(November 24, 2003)

1353 \currannalsline1622005

Acknowledgements.
\twoauthorsAdrian DiaconuYe Tian \institutionColumbia University, New York, NY

McGill University, Montreal, Quebec, Canada

1 Introduction

Let pp be a prime number, FF a totally real field such that [F(μp):F]=2[F(\mu_{p}):F]=2 and [F:][F:{\mathbb{Q}}] is odd. For δF×\delta\in F^{\times}, let [δ][\ \delta\ ] denote its class in F×/F×pF^{\times}/F^{\times p}. In this paper, we show

\demo

Main Theorem There are infinitely many classes [δ]F×/F×p[\ \delta\ ]\in F^{\times}/F^{\times p} such that the twisted affine Fermat curves

Wδ:Xp+Yp=δW_{\delta}:\quad X^{p}+Y^{p}=\delta

have no FF-rational points.

\demo

Remark It is clear that if [δ]=[δ][\ \delta\ ]=[\ \delta^{\prime}\ ], then WδW_{\delta} is isomorphic to WδW_{\delta^{\prime}} over FF. For any δF×,\delta\in F^{\times}, Wδ/FW_{\delta}/F has rational points locally everywhere. \Enddemo

To obtain this result, consider the smooth open affine curve:

Cδ:Vp=U(δU),C_{\delta}:V^{p}=U(\delta-U),

and the morphism:

ψδ:WδCδ;(x,y)(xp,xy).\psi_{\delta}:W_{\delta}\longrightarrow C_{\delta};\quad(x,y)\longmapsto(x^{p},xy).

Let CδJδC_{\delta}\rightarrow J_{\delta} be the Jacobian embedding of Cδ/FC_{\delta}/F defined by the point (0,0)(0,0). We will show that:

  1. 1.

    If L(1,Jδ/F)0L(1,J_{\delta}/F)\neq 0, then Jδ(F)J_{\delta}(F) is a finite group (cf. Theorem 2.1. of §2).

    The proof is based on Zhang’s extension of the Gross-Zagier formula to totally real fields and on Kolyvagin’s technique of Euler systems. One might use techniques of congruence of modular forms to remove the restriction that the degree [F:][F:{\mathbb{Q}}] is odd.

  2. 2.

    There are infinitely many classes [δ][\ \delta\ ] such that L(1,Jδ/F)0L(1,J_{\delta}/F)\neq 0 (cf. Theorem 3.1. of §3; see also 2.2.4.).

    The proof is based on the theory of double Dirichlet series. The condition that [F(μp):F]=2[F(\mu_{p}):F]=2 is essential for the technique we use here.

Combining (1)(1) and (2)(2), one can see that the set

Π:={[δ]F×/F×p|Jδ(F)is torsion}\Pi:=\Big{\{}[\ \delta\ ]\in F^{\times}/F^{\times p}\ \Big{|}\ J_{\delta}(F)\ \text{is torsion}\Big{\}}

is infinite.

\Subsec

Proof of the Main Theorem assuming (1)(1) and (2)(2) For any δF×\delta\in F^{\times}, consider the twisting isomorphism (defined over F(δp)F(\sqrt[p]{\delta})):

ιδ:CδC1;(u,v)(u/δ,v/δ2p).\iota_{\delta}:C_{\delta}\longrightarrow C_{1};\quad(u,v)\longmapsto(u/\delta,v/\sqrt[p]{\delta^{2}}).

Define ηδ:JδJ1\eta_{\delta}:J_{\delta}\longrightarrow J_{1} to be the homomorphism associated to ιδ\iota_{\delta}.

Let Σδ\Sigma_{\delta} denote the set ιδ(Cδ(F))\iota_{\delta}\left(C_{\delta}(F)\right). It is easy to see that:

  1. (i)

    Σδ=Σδ\Sigma_{\delta}=\Sigma_{\delta^{\prime}}, if [δ]=[δ][\ \delta\ ]=[\ \delta^{\prime}\ ],

  2. (ii)

    ΣδΣδ={(0,0),(1,0)}\Sigma_{\delta}\cap\Sigma_{\delta^{\prime}}=\{(0,0),(1,0)\}, otherwise.

For any δF×\delta\in F^{\times} with [δ]Π[\ \delta\ ]\in\Pi, and [δ]1[\ \delta\ ]\neq 1, the diagram

Wδ(F)ψδCδ(F)Jδ(F)ιδηδC1(F(δp))J1(F(δp))\begin{array}[]{cccccccccccccc}W_{\delta}(F)\ \stackrel{{\scriptstyle\psi_{\delta}}}{{\longrightarrow}}&C_{\delta}(F)&\hookrightarrow&J_{\delta}(F)\\ &\Biggr{\downarrow}\hbox to0.0pt{$\displaystyle\iota_{\delta}$\hss}&&\Biggr{\downarrow}\hbox to0.0pt{$\displaystyle\eta_{\delta}$\hss}\\ &C_{1}(F(\sqrt[p]{\delta}))&\hookrightarrow&J_{1}(F(\sqrt[p]{\delta}))\end{array}

commutes.

Since the set

δF×J1(F(δp))torJ1(F¯)\bigcup_{\delta\in F^{\times}}J_{1}(F(\sqrt[p]{\delta}))_{{\mathrm{tor}}}\ \subset\ J_{1}(\overline{F})

is finite by the Northcott theorem, the set [δ]ΠΣδ\displaystyle{\bigcup_{[\ \delta\ ]\in\Pi}\Sigma_{\delta}} is finite. Thus, for all but finitely many [δ]Π{[1]}[\ \delta\ ]\in\Pi\setminus\{[1]\}, Σδ={(0,0),(1,0)}\Sigma_{\delta}=\{(0,0),(1,0)\}, and therefore WδW_{\delta} has no FF-rational points. ∎

\demo

Remark Our method is, in fact, effective: for any [δ]F×/F×p[\ \delta\ ]\in F^{\times}/F^{\times p}, let

Supp(p)([δ])={𝔭prime of F|pv𝔭(δ)}.{\mathrm{Supp}}^{(p)}\left([\ \delta\ ]\right)=\left\{{\mathfrak{p}}\ \text{prime of $F$}\ \Big{|}\ p\nmid v_{\mathfrak{p}}(\delta)\right\}.

Let LL^{\prime} be the Galois closure of F(μp)F(\mu_{p}), and let SS be the set of places of FF above 2DL/2D_{L^{\prime}/{\mathbb{Q}}}, where DL/D_{L^{\prime}/{\mathbb{Q}}} is the discriminant of L/L^{\prime}/{\mathbb{Q}}. If Supp(p)([δ]){\mathrm{Supp}}^{(p)}\left([\ \delta\ ]\right) is not contained in SS and L(1,Jδ)0L(1,J_{\delta})\neq 0, then the twisted Fermat curve WδW_{\delta} has no FF-rational points (see Proposition 2.2).

\demo

Acknowledgment We would like to thank D. Goldfeld, S. Friedberg,J. Hoffstein, H. Jacquet, V. A. Kolyvagin, L. Szpiro for their help and encouragement, and the referees for useful remarks and suggestions. In particular, we are grateful to S. Zhang, who suggested the problem to us, for many helpful conversations. The second author was partially supported by the Clay Mathematics Institute.

2 Arithmetic methods

Fix δF×𝒪F\delta\in F^{\times}\cap{\mathcal{O}}_{F} such that (δ,p)=1(\delta,p)=1. Let ζ=ζp\zeta=\zeta_{p} be a primitive pp-th root of unity. The abelian variety JδJ_{\delta} is absolutely simple, of dimension g=p12\displaystyle{g=\frac{p-1}{2}}, and has complex multiplication by [ζ]{\mathbb{Z}}[\zeta] over the field F(μp)F(\mu_{p}). In this section we show:

Theorem 2.1

If L(1,Jδ/F)0L(1,J_{\delta}/F)\neq 0, then Jδ(F)J_{\delta}(F) is finite.

Notation. In this section, for an abelian group MM, set M^=Mpp\widehat{M}=M\otimes_{\mathbb{Z}}\prod_{p}{\mathbb{Z}}_{p} where pp runs over all primes. For any ring RR, let R×R^{\times} denote the group of invertible elements. For any ideal 𝔞{\mathfrak{a}} of F,F, denote the norm NF/(𝔞){\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{a}}) by N𝔞{\mathrm{N}}{\mathfrak{a}}. Let 𝔸{\mathbb{A}} denote the adele ring of FF, and 𝔸f{\mathbb{A}}_{f} its finite part. Sometimes, we shall not distinguish a finite place from its corresponding prime ideal.

\Subsec

The Hilbert newform associated to JδJ_{\delta} We first recall some facts about LL-functions of twisted Fermat curves over arbitrary number fields (see [14], [32]). Let FF be any number field, L=F(μp),L=F(\mu_{p}), L0=(μp),L_{0}={\mathbb{Q}}(\mu_{p}), and F0=L0FF_{0}=L_{0}\cap F.

For any place ww of LL, denote by w0w_{0} and vv its restrictions to (μp){\mathbb{Q}}(\mu_{p}) and FF, respectively. Let χw0\chi_{w_{0}} and χw\chi_{w} be the pp-th power residue symbols on L0×L_{0}^{\times} and L×,L^{\times}, respectively, given by class field theory. Then χw=χw0NL/(μp)\chi_{w}=\chi_{w_{0}}\circ{\mathrm{N}}_{L/{\mathbb{Q}}(\mu_{p})}. The Jacobi sum

j(χw,χw)=a𝒪L/wa0,1χw(a)χw(1a)j(\chi_{w},\chi_{w})=-\sum_{\begin{subarray}{c}{a\in{\mathcal{O}}_{L}/w}\\ {a\neq 0,1}\end{subarray}}\chi_{w}(a)\chi_{w}(1-a)

is an integer in L0L_{0} satisfying j(χw,χw)=j(χw0,χw0)iw/w0j(\chi_{w},\chi_{w})=j(\chi_{w_{0}},\chi_{w_{0}})^{i_{w/w_{0}}} and the Stickelberger relation:

(j(χw0,χw0))=i=1p12σi1(w0)\left(j(\chi_{w_{0}},\chi_{w_{0}})\right)=\prod_{i=1}^{\frac{p-1}{2}}\sigma_{i}^{-1}(w_{0})

as an ideal in L0.L_{0}. Here, iw/w0i_{w/w_{0}} is the inertial degree for w/w0w/w_{0}, and σiGal(L0/)\sigma_{i}\in{\mathrm{Gal}}(L_{0}/{\mathbb{Q}}) is the image of ii under the isomorphism (/p)×Gal(L0/)({\mathbb{Z}}/p{\mathbb{Z}})^{\times}\longrightarrow{\mathrm{Gal}}(L_{0}/{\mathbb{Q}}).

Since δ𝒪F\delta\in{\mathcal{O}}_{F} is coprime to pp, CδC_{\delta} has good reduction at ww for any wpδw\nmid p\delta. We know that the zeta-function of the reduction Cδ~\widetilde{C_{\delta}} of CδC_{\delta} at a place vv of FF is

Z(Cδ~,T)=Pv(T)(1T)(1NvT),Z(\widetilde{C_{\delta}},T)=\frac{P_{v}(T)}{(1-T)(1-{\mathrm{N}}vT)},

with

Pv(T)=w|vσ(1χw(δ2)σj(χw,χw)σTfv),P_{v}(T)=\prod_{w|v}\prod_{\sigma}(1-\chi_{w}(\delta^{2})^{\sigma}j(\chi_{w},\chi_{w})^{\sigma}T^{f_{v}}),

where fvf_{v} is the order of Nv{\mathrm{N}}v modulo p,p, and σ\sigma runs over representatives in Gal((μp)/){\mathrm{Gal}}({\mathbb{Q}}(\mu_{p})/{\mathbb{Q}}) of Gal(F0/){\mathrm{Gal}}(F_{0}/{\mathbb{Q}}). Then the number of points on Jδ~\tilde{J_{\delta}} (the reduction of JδJ_{\delta} at vv) is Pv(1)P_{v}(1).

Now we give a bound on torsion points of Jδ(F)J_{\delta}(F). Let FF^{\prime} be the Galois closure of F/F/{\mathbb{Q}}, and assume that FL0=FL0.F\cap L_{0}=F^{\prime}\cap L_{0}. This assumption is satisfied if FF is as in the main theorem, or FF is Galois over {\mathbb{Q}}. Let L=F(μp),L^{\prime}=F^{\prime}(\mu_{p}), and let q2DL/q\nmid 2D_{L^{\prime}/{\mathbb{Q}}} be a prime. Let \ell be a prime for which there exists a place w|w^{\prime}|\ell of LL^{\prime} such that FrobL0/F0(w|L0){\mathrm{Frob}}_{L_{0}/F_{0}}(w^{\prime}|_{L_{0}}) is a generator of Gal(L0/F0){\mathrm{Gal}}(L_{0}/F_{0}), FrobF/F0(w|F)=1{\mathrm{Frob}}_{F^{\prime}/F_{0}}(w^{\prime}|_{F^{\prime}})=1 and Frob(μq)/(w|(μq))=1{\mathrm{Frob}}_{{\mathbb{Q}}(\mu_{q})/{\mathbb{Q}}}(w^{\prime}|_{{\mathbb{Q}}(\mu_{q})})=1. Then, 1modq\ell\equiv 1\mod q. Let v,v, ww and w0w_{0} be the places of F,F, LL and L0L_{0}, respectively, below ww^{\prime}. Then, vv is inert in L/FL/F and iw/w0=1.i_{w/w_{0}}=1. We have

Pv(1)=σ(1χw(δ2)σj(χw,χw)σ).P_{v}(1)=\prod_{\sigma}(1-\chi_{w}(\delta^{2})^{\sigma}j(\chi_{w},\chi_{w})^{\sigma}).

Since vv is inert in L/FL/F and δF×,\delta\in F^{\times}, we have χw(δ2)=1.\chi_{w}(\delta^{2})=1. Using the Stickelberger relation and the fact that j(χw0,χw0)1mod(1ζp)2j(\chi_{w_{0}},\chi_{w_{0}})\equiv 1\mod(1-\zeta_{p})^{2}, one can show thatj(χw,χw)=f,j(\chi_{w},\chi_{w})=-\ell^{f}, for f=p12[F0:].f=\frac{p-1}{2[F_{0}:{\mathbb{Q}}]}. Then, Pv(1)=(1+f)[F0:]2[F0:]modq.P_{v}(1)=(1+\ell^{f})^{[F_{0}:{\mathbb{Q}}]}\equiv 2^{[F_{0}:{\mathbb{Q}}]}\mod q. Consequently, there are no qq-torsion points in Jδ(F)J_{\delta}(F).

Similarly, for the case q|2DL/q|2D_{L^{\prime}/{\mathbb{Q}}}, let cq1c_{q}\geq 1 be the smallest positive integer such that there is a σGal(L(μqcq)/)\sigma\in{\mathrm{Gal}}(L^{\prime}(\mu_{q^{c_{q}}})/{\mathbb{Q}}) for which σ|L\sigma|_{L} is a generator of Gal(L/F){\mathrm{Gal}}(L/F), σ|F=1\sigma|_{F^{\prime}}=1, and the restriction of σ\sigma to Gal((μqcq)/){\mathrm{Gal}}({\mathbb{Q}}(\mu_{q^{c_{q}}})/{\mathbb{Q}}) has order greater than f=p12[F0:].f=\frac{p-1}{2[F_{0}:{\mathbb{Q}}]}. Then, Pv(1)/ 0modqcq[F0:]P_{v}(1)\equiv\hskip-9.5pt/\ 0\mod q^{c_{q}[F_{0}:{\mathbb{Q}}]}. Let MM be definedby M:=q|2DL/qcq[F0:].M:=\prod_{q|2D_{L^{\prime}/{\mathbb{Q}}}}q^{c_{q}[F_{0}:{\mathbb{Q}}]}. It follows that Jδ(F)torJδ[M],J_{\delta}(F)_{\mathrm{tor}}\subset J_{\delta}[M], the subgroup of MM-torsion points of Jδ(F¯)J_{\delta}(\overline{F}).

Let FF be a totally real field as in the main theorem. We have:

Proposition 2.2

Let SS be the set of places of FF above 2DL/2D_{L^{\prime}/{\mathbb{Q}}}. If Supp(p)([δ])\ {\mathrm{Supp}}^{(p)}\left([\ \delta\ ]\right) is not contained in SS and L(1,Jδ/F)0,L(1,J_{\delta}/F)\neq 0, then the twistedFermat curve WδW_{\delta} has no FF-rational points.

Let FF be as in the introduction. Then F0=(μp)+F_{0}={\mathbb{Q}}(\mu_{p})^{+} is the maximal totally real subfield of L0=(μp).L_{0}={\mathbb{Q}}(\mu_{p}). By the reciprocity law, one can see that wχw(δ2)w\mapsto\chi_{w}(\delta^{2}) defines a Hecke character, which we denote by χ[δ2].\chi_{[\delta^{2}]}. It depends only on the class of δ2\delta^{2} and has conductor above δ.\delta. By Weil [32], the map wj(χw,χw)NL/w12w\mapsto j(\chi_{w},\chi_{w}){\mathrm{N}}_{L/{\mathbb{Q}}}w^{-\frac{1}{2}} also defines a Hecke character on L,L, denoted by ψ,\psi, which has conductor above p.p. Thus, we have a (unitary) Hecke character on LL,

χ[δ2]ψ:𝔸L××,\chi_{[\delta^{2}]}\psi:{\mathbb{A}}_{L}^{\times}\longrightarrow{\mathbb{C}}^{\times},

which is not of the form ϕNL/F,\phi\circ{\mathrm{N}}_{L/F}, for any Hecke character ϕ\phi over F.F. Then, there exists a unique holomorphic Hilbert newform f/Ff/F of pure weight 22 with trivial central character such that,

Lv(s,f/F)=w|vLw(s1/2,χ[δ2]ψ),L_{v}(s,f/F)=\prod_{w|v}L_{w}(s-1/2,\chi_{[\delta^{2}]}\psi),

for all places vv of F.F. Actually, the field over {\mathbb{Q}} generated by the Hecke eigenvalues attached to ff is F0=(μp)+,F_{0}={\mathbb{Q}}(\mu_{p})^{+}, and for the CM abelian variety Jδ,J_{\delta}, we have

L(s,Jδ/F)\displaystyle L(s,J_{\delta}/F) =σGal(L0/)/Gal(L0/F0)L(s1/2,χ[δ2]σψσ)\displaystyle=\prod_{\sigma\in{\mathrm{Gal}}(L_{0}/{\mathbb{Q}})\big{/}{\mathrm{Gal}}(L_{0}/F_{0})}L(s-1/2,\chi_{[\delta^{2}]}^{\sigma}\psi^{\sigma})
=σ:F0L(s,fσ/F).\displaystyle=\prod_{\sigma:F_{0}\hookrightarrow{\mathbb{C}}}L(s,f^{\sigma}/F).

Note that L(s,Jδ)L(s,J_{\delta}) only depends on the class [δ][\ \delta\ ] of δ,\delta, and the above equality holds for any local factor.

\Subsec

A nonvanishing result Let π\pi be the automorphic representation associated to f,f, and let NN be its conductor. Let S0S_{0} be any finite set of places of F,F, including all infinite places and the places dividing N.N. Choose a quadratic Hecke character ξ\xi corresponding to a totally imaginary quadratic extension of F,F, unramified at N,N, where ξ(N)(1)g=1\xi(N)\cdot(-1)^{g}=-1 (since FF is of odd degree, we have (1)g=1(-1)^{g}=-1); i.e., the epsilon factor of L(s,πξ)L(s,\pi\otimes\xi) is 1.-1. Let 𝒟(ξ;S0){\mathcal{D}}(\xi;S_{0}) denote the set of quadratic characters χ\chi of F×/𝔸F×,F^{\times}/{\mathbb{A}}_{F}^{\times}, for which χv=ξv,\chi_{v}=\xi_{v}, for all vS0.v\in S_{0}. With the above notation and assumptions, by a theorem of Friedberg and Hoffstein [11], there exist infinitely many quadratic characters χ𝒟(ξ;S0)\chi\in{\mathcal{D}}(\xi;S_{0}) such that L(s,πχ)L(s,\pi\otimes\chi) has a simple zero at the center s=1/2.s=1/2.

Choose such a χ,\chi, and let KK be the totally imaginary quadratic extension of FF associated to it. The conductor of χ\chi is coprime to N,N, and the LL-function L(s,f/K)=L(s1/2,π)L(s1/2,πχ)L(s,f/K)=L(s-1/2,\pi)L(s-1/2,\pi\otimes\chi) has a simple zero at s=1.s=1. Let dd denote the discriminant of K/F.K/F.

\Subsec

Zhang’s formula \SubsubsecThe (N,K)(N,K)-type Shimura curves Let 𝒪{\mathcal{O}} be the subalgebra of {\mathbb{C}} over {\mathbb{Z}} generated by the eigenvalues of ff under the Hecke operators. In our case, 𝒪=[ζ+ζ1]{\mathcal{O}}={\mathbb{Z}}[\zeta+\zeta^{-1}] is the ring of integers of F0.F_{0}. In [33] (see also [5], [6]), Zhang constructs a Shimura curve XX of (N,K)(N,K)-type, and proves that there exists a unique abelian subvariety AA of the Jacobian Jac(X){\mathrm{Jac}}(X) of dimension [𝒪:]=g,[{\mathcal{O}}:{\mathbb{Z}}]=g, such that

Lv(s,A)=σ:𝒪Lv(s,fσ/F),L_{v}(s,A)=\prod_{\sigma:{\mathcal{O}}\hookrightarrow{\mathbb{C}}}L_{v}(s,f^{\sigma}/F),

for all places vv of F.F. By the construction of f,f, it follows that Lv(s,A/F)=Lv(s,Jδ/F)L_{v}(s,A/F)=L_{v}(s,J_{\delta}/F) for all places vv of F.F. Therefore, by the isogeny conjecture proved by Faltings, AA is isogenous to JδJ_{\delta} over F.F. In particular, the complex multiplication by 𝒪(μp)+{\mathcal{O}}\subset{\mathbb{Q}}(\mu_{p})^{+} on AA is defined over F.F.

Now, let us recall the constructions of XX and A.A.

The LL-function of πχ\pi\otimes\chi satisfies the functional equation

L(1s,πχ)=(1)|Σ|NF/(Nd)2s1L(s,πχ),L(1-s,\pi\otimes\chi)=(-1)^{\left|\Sigma\right|}{\mathrm{N}}_{F/{\mathbb{Q}}}(Nd)^{2s-1}L(s,\pi\otimes\chi),

where Σ=Σ(N,K)\Sigma=\Sigma(N,K) is the following set of places of F:F:

Σ(N,K)={v|v|,orχv(N)=1}.\Sigma(N,K)=\left\{v\ \Big{|}\ v|\infty,\ \text{or}\ \chi_{v}(N)=-1\right\}.

Since the sign of the functional equation is 1,-1, by our choice of K,K, the cardinality of Σ\Sigma is odd. Let τ\tau be any real place of F.F. Then, we have:

  1. 1.

    Up to isomorphism, there exists a unique quaternion algebra BB such that BB is ramified at exactly the places in Σ\{τ}\Sigma\backslash\{\tau\};

  2. 2.

    There exist embeddings ρ:KB\rho:K\hookrightarrow B over F.F.

From now on, we fix an embedding ρ:KB\rho:K\rightarrow B over F.F.

Let GG denote the algebraic group over F,F, which is an inner form of PGL2{\mathrm{PGL}}_{2} with G(F)B×/F×.G(F)\cong B^{\times}/F^{\times}. The group G(Fτ)PGL2()G(F_{\tau})\cong{\mathrm{PGL}}_{2}({\mathbb{R}}) acts on ±=.{\mathcal{H}}^{\pm}={\mathbb{C}}\setminus{\mathbb{R}}. Now, for any open compact subgroup UU of G(𝔸f),G({\mathbb{A}}_{f}), we have an analytic space

SU()=G(F)+\+×G(𝔸f)/U,S_{U}({\mathbb{C}})=G(F)_{+}\backslash{\mathcal{H}}^{+}\times G({\mathbb{A}}_{f})/U,

where G(F)+G(F)_{+} denotes the subgroup of elements in G(F)G(F) with positive determinant via τ.\tau.

Shimura has shown that SU()S_{U}({\mathbb{C}}) is the set of complex points of an algebraic curve SU,S_{U}, which descends canonically to FF (as a subfield of {\mathbb{C}} via τ\tau). The curve SUS_{U} over FF is independent of the choice of τ.\tau.

There exists an order R0R_{0} of BB containing 𝒪K{\mathcal{O}}_{K} with reduced discriminant N.N. One can choose R0R_{0} as follows. Let 𝒪B{\mathcal{O}}_{B} be a maximal order of BB containing 𝒪K,{\mathcal{O}}_{K}, and let 𝒩{\mathcal{N}} be an ideal of 𝒪K{\mathcal{O}}_{K} such that

NK/F𝒩discB/F=N,{\mathrm{N}}_{K/F}{\mathcal{N}}\cdot{\mathrm{disc}}_{B/F}=N,

where discB/F{\mathrm{disc}}_{B/F} is the reduced discriminant of 𝒪B{\mathcal{O}}_{B} over 𝒪F.{\mathcal{O}}_{F}. Then, we take

R0=𝒪K+𝒩𝒪B.R_{0}={\mathcal{O}}_{K}+{\mathcal{N}}\cdot{\mathcal{O}}_{B}.

Take U=vRv×/𝒪v×.U=\prod_{v}R^{\times}_{v}/{\mathcal{O}}_{v}^{\times}. The corresponding Shimura curve X:=SUX:=S_{U} is compact.

Let ξPic(X)\xi\in\mathrm{Pic}(X)\otimes{\mathbb{Q}} be the unique class whose degree is 11 on each connected component and such that,

Tmξ=deg(Tm)ξ,{\mathrm{T}}_{m}\xi=\deg({\mathrm{T}}_{m})\xi,

for all integral ideals mm of 𝒪F{\mathcal{O}}_{F} coprime to Nd.Nd. Here, the Tm{\mathrm{T}}_{m} are the Hecke operators.

\Subsubsec

Gross-Zagier-Zhang formula Now, we define the basic class in Jac(X)(K),{\mathrm{Jac}}(X)(K)\otimes{\mathbb{Q}}, where Jac(X){\mathrm{Jac}}(X) is the connected component of Pic(X),\mathrm{Pic}(X), from the CM-points on the curve X.X. The CM points corresponding to KK on XX form a set:

𝒞:G(F)+G(F)+h0×G(𝔸f)/UT(F)G(𝔸f)/U;[(h0,g)][g],{\mathcal{C}}:\ G(F)_{+}\setminus G(F)_{+}\cdot h_{0}\times G({\mathbb{A}}_{f})/U\cong T(F)\setminus G({\mathbb{A}}_{f})/U;\qquad[(h_{0},g)]\leftrightarrow[g],

where h0+h_{0}\in{\mathcal{H}}^{+} is the unique fixed point of the torus T(F)=K×/F×.T(F)=K^{\times}/F^{\times}.

For a CM point z=[g]𝒞,z=[g]\in{\mathcal{C}}, represented by gG(𝔸f),g\in G({\mathbb{A}}_{f}), let

Φg:KB^,tg1ρ(t)g.\Phi_{g}:K\longrightarrow\widehat{B},\qquad t\longmapsto g^{-1}\rho(t)g.

Then, End(z):=Φg1(R0^){\mathrm{End}}(z):=\Phi_{g}^{-1}(\widehat{R_{0}}) is an order of K,K, say 𝒪n=𝒪F+n𝒪K,{\mathcal{O}}_{n}={\mathcal{O}}_{F}+n{\mathcal{O}}_{K}, for a (unique) ideal nn of F.F. The ideal n,n, called the conductor of z,z, is independent of the choice of the representative g.g. By Shimura’s theory, every CM point of conductor nn is defined over the abelian extension HnH_{n}^{\prime} of KK corresponding to K×K^×/F^×𝒪^n×K^{\times}\setminus\widehat{K}^{\times}/\widehat{F}^{\times}\widehat{{\mathcal{O}}}_{n}^{\times} via class field theory.

Let P1P_{1} be a CM point in XX of conductor 1,1, which is defined over H1,H_{1}^{\prime}, the abelian extension of KK corresponding to K×K^×/F^×𝒪^K×.K^{\times}\setminus\widehat{K}^{\times}/\widehat{F}^{\times}\widehat{{\mathcal{O}}}_{K}^{\times}. The divisor P=Gal(H1/K)P1P={\mathrm{Gal}}(H_{1}^{\prime}/K)\cdot P_{1} together with the Hodge class defines a class

x:=[Pdeg(P)ξ]Jac(X)(K),x:=[P-\deg(P)\xi]\in{\mathrm{Jac}}(X)(K)\otimes{\mathbb{Q}},

where degP\deg P is the multi-degree of PP on the geometric components. Let xfx_{f} be the ff-typical component of x.x. In [34], Zhang generalized the Gross-Zagier formula to the totally real field case, by proving that

L(1,f/K)=2g+1N(d)f2xf2,L^{\prime}(1,f/K)=\frac{2^{g+1}}{\sqrt{{\mathrm{N}}(d)}}\cdot\|f\|^{2}\cdot\|x_{f}\|^{2},

where f2\|f\|^{2} is computed on the invariant measure on

PGL2(F)g×PGL2(𝔸f)/U0(N){\mathrm{PGL}}_{2}(F)\setminus{\mathcal{H}}^{g}\times{\mathrm{PGL}}_{2}({\mathbb{A}}_{f})/U_{0}(N)

induced by dxdy/y2dxdy/y^{2} on g,{\mathcal{H}}^{g}, and where

U0(N)={(abcd)GL2(𝒪^F)|cN^}GL2(F^),U_{0}(N)=\left\{\begin{pmatrix}a&b\\ c&d\end{pmatrix}\in{\mathrm{GL}}_{2}(\widehat{{\mathcal{O}}}_{F})\big{|}c\in\widehat{N}\right\}\subset{\mathrm{GL}}_{2}(\widehat{F}),

and xf2\|x_{f}\|^{2} is the Neron-Tate pairing of xfx_{f} with itself.

\Subsubsec

The equivalence of nonvanishing of LL-factors For any σ:F,\sigma:F\hookrightarrow{\mathbb{C}}, it is known by a result of Shimura that L(1,f/F)0L(1,f/F)\neq 0 is equivalent to L(1,fσ/F)0.L(1,f^{\sigma}/F)\break\neq 0. One can also show this using Zhang’s formula above. To see this, assume L(1,f/F)0.L(1,f/F)\neq 0. Then, xf0,\|x_{f}\|\neq 0, and therefore, xfσ0.\|x_{f^{\sigma}}\|\neq 0. It follows that L(1,fσ/K)0.L^{\prime}(1,f^{\sigma}/K)\neq 0. Since L(1,f/F)0,L(1,f/F)\neq 0, the LL-function L(s,fσ/F)L(s,f^{\sigma}/F) has a positive sign in its functional equation. Thus, L(1,fσ/F)0.L(1,f^{\sigma}/F)\neq 0. In fact, to obtain our main theorem, we do not need this equivalence, but we may see that Theorem 3.1 is equivalent to statement (2) in the introduction.

\Subsec

The Euler system of CM points We now assume that L(1,χ[δ2]ψ)0,L(1,\chi_{[\delta^{2}]}\psi)\neq 0, or equivalently, L(1,f/F)0.L(1,f/F)\neq 0. Then by the equivalence of nonvanishing of L(1,fσ)L(1,f^{\sigma}) for all embeddings σ:F,\sigma:F\hookrightarrow{\mathbb{C}}, we have that L(1,Jδ/F)0.L(1,J_{\delta}/F)\neq 0. By Zhang’s formula, we also know that xf0.\|x_{f}\|\neq 0.

Let 𝒩{\mathcal{N}} be the set of square-free integral ideals of FF whose prime divisors are inert in KK and coprime to Nd.Nd. For any n𝒩,n\in{\mathcal{N}}, define

Hn=|nHHn,H1=H1.H_{n}=\prod_{\ell|n}H^{\prime}_{\ell}\subset H_{n}^{\prime},\qquad H_{1}=H_{1}^{\prime}.

Let unu_{n} denote the cardinality of (𝒪^n×K×F^×)/𝒪^F×.(\widehat{{\mathcal{O}}}_{n}^{\times}\cap K^{\times}\widehat{F}^{\times})/\widehat{{\mathcal{O}}}_{F}^{\times}. Then, H/H1H_{\ell}/H_{1} is a cyclic extension of degree t()=N()+1u1/u.t(\ell)=\frac{{\mathrm{N}}(\ell)+1}{u_{1}/u_{\ell}}.

For each n𝒩,n\in{\mathcal{N}}, let PnP_{n} be a CM point of order nn such that PnP_{n} is contained in TPm{\mathrm{T}}_{\ell}P_{m} if n=m𝒩n=m\ell\in{\mathcal{N}} and \ell is a prime ideal of F.F. Let yn=TrHn/Hnπ(Pn)A(Hn),y_{n}={\mathrm{Tr}}_{H_{n}^{\prime}/H_{n}}\pi(P_{n})\in A(H_{n}), where π\pi is a morphism from XX to Jac(X){\mathrm{Jac}}(X) defined by a multiple of the Hodge class.

The points {yn}n𝒩\{y_{n}\}_{n\in{\mathcal{N}}} form an Euler system (see [29, Prop. 7.5], or [33, Lemma 7.2.2]) so that, for any n=m𝒩n=m\ell\in{\mathcal{N}} with \ell a prime ideal of F,F,

  1. 1.

    un1σGal(Hn/Hm)ynσ=um1aym\displaystyle{{u_{n}}^{-1}\sum_{\sigma\in{\mathrm{Gal}}(H_{n}/H_{m})}y_{n}^{\sigma}={u_{m}}^{-1}a_{\ell}y_{m}};

  2. 2.

    For any prime ideal λm\lambda_{m} of HmH_{m} above ,\ell, and for λn\lambda_{n} the unique prime above λm,\lambda_{m},

    Frobλmymynmodλn;{\mathrm{Frob}}_{\lambda_{m}}y_{m}\equiv y_{n}\mod\lambda_{n};
  3. 3.

    The class xfx_{f} is equal to yK:=trH1/Ky1y_{K}:={\mathrm{tr}}_{H_{1}/K}y_{1} in (A(K))/×.\big{(}A(K)\otimes{\mathbb{Q}}\big{)}\big{/}{\mathbb{Q}}^{\times}.

Theorem 2.1 follows with the nontrivial Euler system by Kolyvagin’s standard argument (see [21], [23], [13], and [33, Th. A]).

up

3 Analytic methods

Let r=4r=4 or an odd prime, and let L=F(ζr),L=F(\zeta_{r}), with [L:F]=2.[L:F]=2. Let ψ\psi be a unitary Hecke character of L.L. In this section, we show:

Theorem 3.1

There are infinitely many classes δF×/F×r\delta\in F^{\times}/F^{\times r} such that L(12,χ[δ]ψ)L\left(\frac{1}{2},\chi_{[\ \delta\ ]}\psi\right) does not vanish.

Let ρ\rho be a unitary Hecke character of F.F. The purpose of this section is to construct a perfect double Dirichlet series Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) similar to an Asai-Flicker-Patterson type Rankin-Selberg convolution, which possesses meromorphic continuation to 2{\mathbb{C}}^{2} and functional equations. Then, Theorem 3.13.1 will follow from the analytic properties of Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) (when r=4r=4, see [7]). To do this, it is necessary to recall the Fisher-Friedberg symbol in [9].

\Subsec

The rr-th power residue symbol Let SS^{\prime} be a finite set of non-archimedean places of LL containing all places dividing r,r, and such that the ring of SS^{\prime}-integers 𝒪LS{\mathcal{O}}_{L}^{S^{\prime}} has class number one. We shall also assume that SS^{\prime} is closed under conjugation and that ψ\psi and ρ\rho are both unramified outside S.S^{\prime}.

Let SS_{\infty} denote the set of all archimedean places of L,L, and set S=SS.S=S^{\prime}\cup S_{\infty}. Let IL(S)I_{L}(S) (resp. L(S){\mathcal{I}}_{L}(S)) denote the group of fractional ideals (resp. the set of all integral ideals) of 𝒪L{\mathcal{O}}_{L} coprime to S.S^{\prime}. In [9], Fisher and Friedberg have shown that the rr-th order symbol χn\chi_{n} can be extended to IL(S)I_{L}(S) i.e., χ𝔫(𝔪)\chi_{\mathfrak{n}}(\mathfrak{m}) is defined for 𝔪,\mathfrak{m}, 𝔫IL(S).\mathfrak{n}\in I_{L}(S). Let us recall their construction.

For a non-archimedean place vS,v\in S^{\prime}, let 𝔓v\mathfrak{P}_{v} denote the corresponding ideal of L.L. Define 𝔠=vS𝔓vrv\mathfrak{c}=\prod_{v\in S^{\prime}}\mathfrak{P}_{v}^{r_{v}} with rv=1r_{v}=1 if ordv(r)=0{\mathrm{ord}}_{v}(r)=0, and rvr_{v} sufficiently large such that, for aLv,a\in L_{v}, ordv(a1)rv{\mathrm{ord}}_{v}(a-1)\geq r_{v} implies that a(Lv×)ra\in(L^{\times}_{v})^{r}. Let PL(𝔠)IL(S)P_{L}({\mathfrak{c}})\subset I_{L}(S) be the subgroup of principal ideals (α)(\alpha) with α1mod𝔠,\alpha\equiv 1\mod{\mathfrak{c}}, and let H𝔠=IL(S)/PL(𝔠)H_{\mathfrak{c}}=I_{L}(S)/P_{L}({\mathfrak{c}}) be the ray class group modulo 𝔠.{\mathfrak{c}}. Set R𝔠=H𝔠/r,R_{\mathfrak{c}}=H_{\mathfrak{c}}\otimes{\mathbb{Z}}/r{\mathbb{Z}}, and write the finite group R𝔠R_{\mathfrak{c}} as a direct product of cyclic groups. Choose a generator for each, and let 𝔈0{\mathfrak{E}}_{0} be a set of ideals of 𝒪L,{\mathcal{O}}_{L}, prime to S,S, which represent these generators. For each 𝔢0𝔈0,{\mathfrak{e}}_{0}\in{\mathfrak{E}}_{0}, choose m𝔢0L×m_{{\mathfrak{e}}_{0}}\in L^{\times} such that 𝔢0𝒪LS=m𝔢0𝒪LS.{\mathfrak{e}}_{0}{\mathcal{O}}_{L}^{S^{\prime}}=m_{{\mathfrak{e}}_{0}}{\mathcal{O}}_{L}^{S^{\prime}}. Let 𝔈{\mathfrak{E}} be a full set of representatives for R𝔠R_{\mathfrak{c}} of the form 𝔢0𝔈0𝔢0λ𝔢0.\prod_{{\mathfrak{e}}_{0}\in{\mathfrak{E}}_{0}}{\mathfrak{e}}_{0}^{\lambda_{{\mathfrak{e}}_{0}}}. Note that 𝔢𝒪LS=m𝔢𝒪LS{\mathfrak{e}}{\mathcal{O}}_{L}^{S^{\prime}}=m_{\mathfrak{e}}{\mathcal{O}}_{L}^{S^{\prime}} for all 𝔢𝔈.{\mathfrak{e}}\in{\mathfrak{E}}. Without loss, we suppose that 𝒪LS𝔈{\mathcal{O}}_{L}^{S^{\prime}}\in{\mathfrak{E}} and m𝒪LS=1.m_{{\mathcal{O}}_{L}^{S^{\prime}}}=1.

Let 𝔪,𝔫IL(S){\mathfrak{m}},{\mathfrak{n}}\in I_{L}(S) be coprime. Write 𝔪=(m)𝔢𝔤r{\mathfrak{m}}=(m){\mathfrak{e}}{\mathfrak{g}}^{r} with 𝔢𝔈,{\mathfrak{e}}\in{\mathfrak{E}}, mL×m\in L^{\times}, m1mod𝔠m\equiv 1\mod{\mathfrak{c}} and 𝔤IL(S),{\mathfrak{g}}\in I_{L}(S), (𝔤,𝔫)=1.({\mathfrak{g}},{\mathfrak{n}})=1. Then the rr-th power residue symbol (mm𝔢𝔫)r\left(\frac{mm_{\mathfrak{e}}}{{\mathfrak{n}}}\right)_{r} is defined. If 𝔪=(m)𝔢𝔤r{\mathfrak{m}}=(m^{\prime}){\mathfrak{e}}^{\prime}{\mathfrak{g}}^{{}^{\prime}r} is another such decomposition, then 𝔢=𝔢{\mathfrak{e}}^{\prime}={\mathfrak{e}} and (mm𝔢𝔫)r=(mm𝔢𝔫)r.\left(\frac{m^{\prime}m_{{\mathfrak{e}}^{\prime}}}{{\mathfrak{n}}}\right)_{r}=\left(\frac{mm_{\mathfrak{e}}}{{\mathfrak{n}}}\right)_{r}.

In view of this, the rr-th power residue symbol (𝔪𝔫)r\left(\frac{{\mathfrak{m}}}{{\mathfrak{n}}}\right)_{r} is defined to be(mm𝔢𝔫)r,\left(\frac{mm_{\mathfrak{e}}}{{\mathfrak{n}}}\right)_{r}, and the character χ𝔪\chi_{\mathfrak{m}} is defined by χ𝔪(𝔫)=(𝔪𝔫)r.\chi_{\mathfrak{m}}({\mathfrak{n}})=\left(\frac{{\mathfrak{m}}}{{\mathfrak{n}}}\right)_{r}. This extension of the rr-th power residue symbol depends on the above choices. Let S𝔪S_{\mathfrak{m}} denote the support of the conductor of χ𝔪.\chi_{\mathfrak{m}}. It can be easily checked that if 𝔪=𝔪𝔞r{\mathfrak{m}}={\mathfrak{m}}^{\prime}{\mathfrak{a}}^{r}, then χ𝔪(𝔫)=χ𝔪(𝔫)\chi_{\mathfrak{m}}({\mathfrak{n}})=\chi_{{\mathfrak{m}}^{\prime}}({\mathfrak{n}}) whenever both are defined. This allows one to extend χ𝔪\chi_{\mathfrak{m}} to a character of all ideals of IL(SS𝔪).I_{L}(S\cup S_{\mathfrak{m}}).

The extended symbol possesses a reciprocity law: if 𝔪,𝔫IL(S){\mathfrak{m}},{\mathfrak{n}}\in I_{L}(S) are coprime, then α(𝔪,𝔫)=χ𝔪(𝔫)χ𝔫(𝔪)1\alpha({\mathfrak{m}},{\mathfrak{n}})=\chi_{\mathfrak{m}}({\mathfrak{n}})\chi_{\mathfrak{n}}({\mathfrak{m}})^{-1} depends only on the images of 𝔪,𝔫{\mathfrak{m}},{\mathfrak{n}} in R𝔠.R_{\mathfrak{c}}.

In our situation, we also need the following lemma:

Lemma 3.2

The natural morphism

IF(S)/PF(𝔠)IL(S)/PL(𝔠)I_{F}(S)/P_{F}({\mathfrak{c}})\longrightarrow I_{L}(S)/P_{L}({\mathfrak{c}})

has kernel of order a power of 2.2.

\Proof

If [𝔫][{\mathfrak{n}}] is in the kernel, i.e., 𝔫=(α){\mathfrak{n}}=(\alpha) in IL(S)I_{L}(S) is a principal ideal with α1mod𝔠,\alpha\equiv 1\mod{\mathfrak{c}}, then α/α¯\alpha/\overline{\alpha} is a root of unity with α/α¯1mod𝔠.\alpha/\overline{\alpha}\equiv 1\mod{\mathfrak{c}}. Now let WW be the set of roots of unity in LL which are 1mod𝔠.\equiv 1\mod{\mathfrak{c}}. Let W0W_{0} be the subset of WW of elements of the form u/u¯u/\overline{u} for some unit uu in 𝒪L{\mathcal{O}}_{L} and u1mod𝔠.u\equiv 1\mod{\mathfrak{c}}. It is clear that W0W2.W_{0}\supset W^{2}. Then, the map

Ker(IF(S)/PF(𝔠)IL(S)/PL(𝔠))W/W0;𝔫α/α¯{\mathrm{Ker}}\left(I_{F}(S)/P_{F}({\mathfrak{c}})\rightarrow I_{L}(S)/P_{L}({\mathfrak{c}})\right)\longrightarrow W/W_{0};\qquad{\mathfrak{n}}\longmapsto\alpha/\overline{\alpha}

is obviously injective; i.e., the order of the kernel of the natural map in this lemma is a power of 2.2. \Endproof

Since rr is odd, using the lemma, we may choose a suitable set 𝔈0{\mathfrak{E}}_{0} of representatives since the beginning such that if 𝔪IF(S),{\mathfrak{m}}\in I_{F}(S), then the decomposition 𝔪=(m)𝔢𝔤r{\mathfrak{m}}=(m){\mathfrak{e}}{\mathfrak{g}}^{r} is such that mF×m\in F^{\times}, 𝔢,𝔤IF(S).{\mathfrak{e}},{\mathfrak{g}}\in I_{F}(S).

Using the symbol χ𝔫,\chi_{\mathfrak{n}}, we shall construct a perfect double Dirichlet series Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) (i.e., possessing meromorphic continuation to 2{{\mathbb{C}}}^{2}) of type:

(3.1) Z(s,w;ψ;ρ)=ZS(s,w;ψ;ρ)=𝔫F(S)LS(s,ψχ𝔫)ρ(𝔫)NF/(𝔫)w,Z(s,w;\psi;\rho)\;=\;Z_{S}(s,w;\psi;\rho)\;\;=*\sum_{\mathfrak{n}\in{\mathcal{I}}_{F}(S)}L_{S}(s,\psi\,\chi_{\mathfrak{n}})\,\rho(\mathfrak{n})\,{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{-w},

where the sum is over the set of all integral ideals of 𝒪F{\mathcal{O}}_{F} coprime to S,S^{\prime}, for 𝔫F(S)\mathfrak{n}\in{\mathcal{I}}_{F}(S) square-free, the function LS(s,ψχ𝔫)L_{S}(s,\psi\,\chi_{\mathfrak{n}}) is precisely the Hecke LL-function attached to ψχ𝔫\psi\,\chi_{\mathfrak{n}} with the Euler factors at all places in SS removed, and where * is a certain normalizing factor. For an arbitrary 𝔫F(S),\mathfrak{n}\in{\mathcal{I}}_{F}(S), write 𝔫=𝔫1𝔫2r\mathfrak{n}=\mathfrak{n}_{1}\mathfrak{n}_{2}^{r} with 𝔫1\mathfrak{n}_{1} rr-th power free. If LS(s,ψχ𝔫1)L_{S}(s,\psi\,\chi_{\mathfrak{n}_{1}}) denotes the Hecke LL-series associated to ψχ𝔫1\psi\,\chi_{\mathfrak{n}_{1}} with the Euler factors at all places in SS removed, then LS(s,ψχ𝔫)L_{S}(s,\psi\,\chi_{\mathfrak{n}}) is defined as LS(s,ψχ𝔫1)L_{S}(s,\psi\,\chi_{\mathfrak{n}_{1}}) multiplied by a Dirichlet polynomial whose complexity grows with the divisibility of 𝔫\mathfrak{n} by powers (see (3.10),(3.10), (3.12)(3.12) and (3.13)(3.13) for precise definitions).

Based on the analytic properties of Z(s,w;ψ;ρ)Z(s,w;\psi;\rho), we show the following result which is stronger than Theorem 3.1.

Theorem 3.3

1) There exist infinitely many rr-th power free ideals 𝔫1{\mathfrak{n}}_{1} in F(S){\mathcal{I}}_{F}(S) with trivial image in R𝔠R_{\mathfrak{c}} for which the special value LS(12,χ𝔫ψ)L_{S}(\frac{1}{2},\chi_{\mathfrak{n}}\psi) does not vanish.

2) Let κ𝔠\kappa_{\mathfrak{c}} denote the number of characters of R𝔠R_{\mathfrak{c}} whose restrictions to FF are also characters of the ideal class group of FF, and let κ\kappa be the residue of the Dedekind zeta function ζF(s)\zeta_{F}(s) at s=1s=1. Then for x,x\rightarrow\infty,

(3.2) NF/(𝔫)<x𝔫F(S)𝔫=(n)[𝔫]= 1LS(12,χ𝔫ψ)κκ𝔠hF|R𝔠|LS(1,ψ)LS(r2,ψr)LS(r2+1,ψr)vinFvS(1qv1)x,\sum_{\begin{subarray}{c}{{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})<x}\\ {\mathfrak{n}\in{\mathcal{I}}_{F}(S)}\\ {{\mathfrak{n}}\,=\,(n)}\\ {[{\mathfrak{n}}]\,=\,1}\end{subarray}}L_{S}\left(\frac{1}{2},\,\chi_{\mathfrak{n}}\psi\right)\;\sim\;\frac{\kappa\cdot\kappa_{\mathfrak{c}}}{h_{F}\cdot|R_{\mathfrak{c}}|}\,\frac{L_{S}(1,\psi)\,L_{S}(\frac{r}{2},\psi^{r})}{L_{S}(\frac{r}{2}+1,\psi^{r})}\prod_{\begin{subarray}{c}{v\;\text{in}\;F}\\ {v\in S^{\prime}}\end{subarray}}\left(1-q_{v}^{-1}\right)\cdot x,\

where [𝔫][{\mathfrak{n}}] denotes the image of the ideal 𝔫{\mathfrak{n}} in R𝔠R_{\mathfrak{c}}.

Remarks. i) By the above definition of the extended rr-th power residue symbol, it is easy to see that the first part of this theorem is equivalent to Theorem 3.1.3.1.

ii) In fact, by a well-known result of Waldspurger [30], it will follow that LS(12,χ𝔫ψ)0,L_{S}(\frac{1}{2},\chi_{\mathfrak{n}}\psi)\geq 0, for 𝔫F(S),𝔫=(n){\mathfrak{n}}\in{\mathcal{I}}_{F}(S),\,{\mathfrak{n}}=(n) and trivial image in R𝔠.R_{\mathfrak{c}}. We will see this in the course of the proof of Theorem 3.3.3.3.

iii) Following [8], by a simple sieving process, one can prove the more familiar variant of the above asymptotic formula where the sum is restricted to square-free principal ideals.

\Subsec

The series Zaux(s,w;ψ;ρ)Z_{\rm aux}(s,w;\psi;\rho) and metaplectic Eisenstein series To obtain the correct definition of Z(s,w;ψ;ρ)Z(s,w;\psi;\rho), let G0(𝔫,𝔪),G_{0}({\mathfrak{n}},\,{\mathfrak{m}}), for 𝔪,{\mathfrak{m}}, 𝔫L(S),{\mathfrak{n}}\in{\mathcal{I}}_{L}(S), be given by

(3.3) G0(𝔫,𝔪)=vordv(𝔫)=kordv(𝔪)=lG0(𝔭vk,𝔭vl),G_{0}({\mathfrak{n}},\,{\mathfrak{m}})\;\;\;=\prod_{\begin{subarray}{c}v\\ {{\mathrm{ord}}_{v}({\mathfrak{n}})=k}\\ {{\mathrm{ord}}_{v}({\mathfrak{m}})=l}\end{subarray}}G_{0}(\mathfrak{p}_{v}^{k},\mathfrak{p}_{v}^{l}),\

where, for k,k, l0,l\geq 0,

(3.4) G0(𝔭vk,𝔭vl)={1if l=0,qvk2if k+1=l; l0(modr),qvk12if k+1=l; l>0; l0(modr),qvl21(qv1)if kl; l>0; l0(modr),0otherwise.G_{0}({\mathfrak{p}}_{v}^{k},\,{\mathfrak{p}}_{v}^{l})\,=\,\begin{cases}1&\text{if $l=0,$}\\ q_{v}^{\frac{k}{2}}&\text{if $k+1=l;$ $l\not\equiv 0\pmod{r},$}\\ -\,q_{v}^{\frac{k-1}{2}}&\text{if $k+1=l;$ $l>0;$ $l\equiv 0\pmod{r},$}\\ q_{v}^{\frac{l}{2}-1}(q_{v}-1)&\text{if $k\geq l;$ $l>0;$ $l\equiv 0\pmod{r},$}\\ 0&\text{otherwise.}\end{cases}\

Here qvq_{v} denotes the absolute value of the norm of v.v. Also, let G(χ𝔪1)G(\chi_{{\mathfrak{m}}_{1}}^{*}) (where 𝔪1{\mathfrak{m}}_{1} denotes the rr-th power free part of 𝔪{\mathfrak{m}} and χ𝔞(𝔟):=χ𝔟(𝔞)\chi_{\mathfrak{a}}^{*}(\mathfrak{b}):=\chi_{\mathfrak{b}}(\mathfrak{a})) be the normalized Gauss sum appearing in the functional equation of the (primitive) Hecke LL-function associated to χ𝔪.\chi_{{\mathfrak{m}}}^{*}. If 𝔫{\mathfrak{n}}^{*} denotes the part of 𝔫{\mathfrak{n}} coprime to 𝔪1,{\mathfrak{m}}_{1}, then set

G(𝔫,𝔪):=χ𝔪1(𝔫)¯G(χ𝔪1)G0(𝔫,𝔪).G({\mathfrak{n}},{\mathfrak{m}})\,:=\,\overline{\chi_{{\mathfrak{m}}_{1}}^{*}({\mathfrak{n}}^{*})}\,G(\chi_{{\mathfrak{m}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}}).

Now, let ψ\psi be as above. For 𝔫L(S){\mathfrak{n}}\in{\mathcal{I}}_{L}(S) and Re(s)>1,{\mathrm{Re}}(s)>1, let ΨS(s,𝔫,ψ)\Psi_{S}(s,{\mathfrak{n}},\psi) be the absolutely convergent Dirichlet series defined by

ΨS(s,𝔫,ψ)=LS(rsr2+1,ψr)𝔪L(S)ψ(𝔪)G(𝔫,𝔪)NL/(𝔪)s.\Psi_{S}(s,{\mathfrak{n}},\psi)\;=\;L_{S}\left(rs-\frac{r}{2}+1,\psi^{r}\right)\sum_{{\mathfrak{m}}\in{\mathcal{I}}_{L}(S)}\frac{\psi({\mathfrak{m}})G({\mathfrak{n}},{\mathfrak{m}})}{{\mathrm{N}}_{L/\mathbb{Q}}({\mathfrak{m}})^{s}}.

This series can be realized as a Fourier coefficient of a metaplectic Eisenstein series on the rr-fold cover of GL(2){\mathrm{GL}}(2) (see [18] and [24]). It follows as in Selberg [28], or alternatively, from Langlands’ general theory of Eisenstein series [25] that ΨS(s,𝔫,ψ)\Psi_{S}(s,{\mathfrak{n}},\psi) has meromorphic continuation to {\mathbb{C}} with only one possible (simple) pole at s=12+1r.s=\frac{1}{2}+\frac{1}{r}. Moreover, this function is bounded when |Im(s)||{\mathrm{Im}}(s)| is large in vertical strips, and satisfies a functional equation as s1ss\to 1-s (see Kazhdan-Patterson [18, Cor. II.2.4]).

For Re(s),Re(w)>1,{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)>1, let Zaux(s,w;ψ;ρ)Z_{\rm aux}(s,w;\psi;\rho) be the auxiliary double Dirichlet series defined by

(3.5) Zaux(s,w;ψ;ρ)=𝔫F(S)ΨS(s,𝔫,ψ)ρ(𝔫)NF/(𝔫)w.Z_{\rm aux}(s,w;\psi;\rho)\;\;=\sum_{\mathfrak{n}\in{\mathcal{I}}_{F}(S)}\frac{\Psi_{S}(s,{\mathfrak{n}},\psi)\rho(\mathfrak{n})}{{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{w}}.

Let ρ~\tilde{\rho} be the Hecke character of LL given by ρ~=ρNL/F.\tilde{\rho}=\rho\,\circ\,{\mathrm{N}}_{L/F}. As we shall shortly see, Zaux(s,w;ψρ~;ρ¯)Z_{\rm aux}(s,w;\psi\,\tilde{\rho};\overline{\rho}) is the type of object that constitutes a building block in the process of constructing the perfect double Dirichlet series Z(s,w;ψ;ρ).Z(s,w;\psi;\rho). Set

Γaux(s,ψρ~)=vSj=1r1Lv(s12+jr,ψvρ~v),\Gamma_{\rm aux}^{*}(s,\psi\,\tilde{\rho})\;=\prod_{v\in S_{\infty}}\;\prod_{j=1}^{r-1}L_{v}\Big{(}s-\frac{1}{2}+\frac{j}{r},\psi_{v}\,\tilde{\rho}_{v}\Big{)},

and let

Z^aux(s,w;ψρ~;ρ¯):=Γaux(s,ψρ~)Zaux(s,w;ψρ~;ρ¯).\widehat{Z}_{\rm aux}(s,w;\psi\,\tilde{\rho};\bar{\rho})\,:=\,\Gamma_{\rm aux}^{*}(s,\psi\,\tilde{\rho})\cdot Z_{\rm aux}(s,w;\psi\,\tilde{\rho};\bar{\rho}).

Let 1{\mathcal{R}}_{1} be the tube region in 2{\mathbb{C}}^{2} whose base 1{\mathcal{B}}_{1} is the convex region in 2{\mathbb{R}}^{2} which lies strictly above the polygonal contour determined by (0,2),(0,2), (1,1)(1,1), and the rays y=2x+2y=-2x+2 for x0x\leq 0 and y=1y=1 for x1x\geq 1. As a simple consequence of the analytic properties of ΨS(s,𝔫,ψ)\Psi_{S}(s,{\mathfrak{n}},\psi) (𝔫L(S){\mathfrak{n}}\in{\mathcal{I}}_{L}(S)), we have the following:

Proposition 3.4

The double Dirichlet series Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho}) is holomorphic in 1,{\mathcal{R}}_{1}, unless ψrρ~r=1\psi^{r}\tilde{\rho}^{r}=1 when it has only one simple pole at s=12+1rs=\frac{1}{2}+\frac{1}{r}. Furthermore, Z^aux(s,w;ψρ~,ρ¯)\widehat{Z}_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho}) satisfies the functional equation

(3.6) Z^aux(s,w;ψρ~,ρ¯)vS(1(ψρ~)r(πv)qvrsr21)\displaystyle\widehat{Z}_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho})\,\cdot\prod_{v\in S^{\prime}}\left(1-(\psi\tilde{\rho})^{-r}(\pi_{v})\,q_{v}^{rs-\frac{r}{2}-1}\right)
=η,τAη,τ(ψ,ρ)(1s)Z^aux(1s,2s+w1;ψ1ρ~1η,ψρτ),\displaystyle\hskip 70.0pt=\;\sum_{\eta,\,\tau}\,A_{\eta,\,\tau}^{(\psi,\,\rho)}(1-s)\,\widehat{Z}_{\rm aux}(1-s,2s+w-1;\psi^{-1}\tilde{\rho}^{-1}\eta,\psi\,\rho\,\tau),

where each Aη,τ(ψ,ρ)(s)A_{\eta,\,\tau}^{(\psi,\,\rho)}(s) is a polynomial in the variables qvs,qvsq_{v}^{s},\,q_{v}^{-s} (vS),(v\in S^{\prime}), and the sum is over a finite set of idéle class characters η\eta and τ,\tau, unramified outside SS and with orders dividing r.r.

\Subsec

The double Dirichlet series Z~(s,w;ψ;ρ)\widetilde{Z}(s,w;\psi;\rho) It turns out that the function Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho}) possesses another functional equation. To describe it, we introduce a new double Dirichlet series Z~(s,w;ψ;ρ)\widetilde{Z}(s,w;\psi;\rho) defined for Re(s),Re(w)>1{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)\break>1 by

(3.7)
Z~(s,w;ψ;ρ)=LS(rs+rw+1r,ψrρ~r)𝔪L(S)𝔪imaginaryψ(𝔪)LS(w,χ𝔪ρ)NL/(𝔪)s\displaystyle\widetilde{Z}(s,w;\psi;\rho)\;=\;L_{S}(rs+rw+1-r,\psi^{r}\tilde{\rho}^{r})\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{m}}-\text{imaginary}}\end{subarray}}\frac{\psi({\mathfrak{m}})\,L_{S}(w,\,\chi_{{\mathfrak{m}}}^{*}\,\rho)}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}}
𝔥F(S)(ψρ)(𝔥)χ𝔪(𝔥1)NF/(𝔥)2s1NF/(𝔥)wvordv(𝔥0)>0[(χ𝔪ρ)(πv)qvwqv1]\displaystyle\qquad\cdot\sum_{{\mathfrak{h}}\in{\mathcal{I}}_{F}(S)}\frac{(\psi\rho)({\mathfrak{h}})\,\chi_{{\mathfrak{m}}}^{*}({\mathfrak{h}}_{1})}{{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{h}})^{2s-1}\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{h}})^{w}}\;\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{0})>0}\end{subarray}}\left[\,(\chi_{{\mathfrak{m}}}^{*}\,\rho)(\pi_{v})\,q_{v}^{-w}\,-\;q_{v}^{-1}\,\right]
vordv(NL/F(𝔪))>0ordv(𝔥2)>0(1qv1)vsplit inLordv(NL/F(𝔪))=0ordv(𝔥2)>0[(χ𝔪ρ)(πv)qvw1+ 1 2qv1]\displaystyle\qquad\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))>0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}(1-q_{v}^{-1})\prod_{\begin{subarray}{c}{v-\text{split in}\ L}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))=0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}\left[\,(\chi_{{\mathfrak{m}}}^{*}\,\rho)(\pi_{v})\,q_{v}^{-w-1}\,+\,1\,-\,2q_{v}^{-1}\,\right]
vinert inLordv(𝔥2)>0[ 1(χ𝔪ρ)(πv)qvw1].\displaystyle\qquad\cdot\prod_{\begin{subarray}{c}{v-\text{inert in}\ L}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}\left[\,1\,-\;(\chi_{{\mathfrak{m}}}^{*}\,\rho)(\pi_{v})\,q_{v}^{-w-1}\,\right].

In the above formula, an ideal 𝔪L(S)\mathfrak{m}\in{\mathcal{I}}_{L}(S) is called imaginary, if it has no divisor in F(S),{\mathcal{I}}_{F}(S), other than 𝒪F{\mathcal{O}}_{F}. The function LS(w,χ𝔪ρ)L_{S}(w,\,\chi_{{\mathfrak{m}}}^{*}\,\rho) represents the LL-series defined over FF (not necessarily primitive) associated to χ𝔪ρ\chi_{{\mathfrak{m}}}^{*}\,\rho with the Euler factors corresponding to places removed in SS. Also, all the products are over places of FF, πv\pi_{v} is the local parameter of FvF_{v} (FvF_{v} denoting the completion of FF at vv), and qvq_{v} is the absolute value of the norm in FF of v.v.

Let 2{\mathcal{R}}_{2} denote the tube region in 2{\mathbb{C}}^{2} whose base 2{\mathcal{B}}_{2} is the convex region in 2{\mathbb{R}}^{2} which lies strictly above the polygonal contour determined by (1,1),(1,1), (32,0)(\frac{3}{2},0) and the rays y=x+32y=-x+\frac{3}{2} for y0y\leq 0 and x=1x=1 for y1.y\geq 1. Recall that LS(w,χ𝔪ρ)L_{S}(w,\,\chi_{{\mathfrak{m}}}^{*}\,\rho) differs from a primitive LL-series by only finitely many Euler factors (i.e., the factors corresponding to places in SS and to places vv for which ordv(NL/F(𝔪))0(modr){\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))\equiv 0\pmod{r}). Applying the functional equation of LS(w,χ𝔪ρ)L_{S}(w,\,\chi_{{\mathfrak{m}}}^{*}\,\rho) and some standard estimates, one can easily show that the function Z~(s,w;ψ;ρ)\widetilde{Z}(s,w;\psi;\rho) is holomorphic in 2,{\mathcal{R}}_{2}, unless ρ=1\rho=1 where it has only one simple pole at w=1w=1. The following proposition gives the functional equation connecting the double Dirichlet series Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho}) and Z~(s,w;ψ;ρ).\widetilde{Z}(s,w;\psi;\rho).

Proposition 3.5

The function Z~(s,w;ψ;ρ)\widetilde{Z}(s,w;\psi;\rho) is holomorphic in 2,{\mathcal{R}}_{2}, unless ρ\rho is the trivial character when it has a simple pole at w=1w=1. Furthermore, for Re(s),Re(w)>1,{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)>1, there exist the functional equations

(3.8) vSLv(1w,ρv)vS(1ρr(πv)qvrw)Z~(s+w12,1w;ψ;ρ)=vSLv(w,ρv1)τBτ(ρ)(w)Zaux(s,w;ψρ~τ,ρ¯),\prod_{v\in S_{\infty}}L_{v}\left(1-w,\rho_{v}\right)\,\cdot\prod_{v\in S^{\prime}}\left(1-\rho^{-r}(\pi_{v})\,q_{v}^{-rw}\right)\cdot\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho)\\ =\prod_{v\in S_{\infty}}L_{v}\left(w,\rho_{v}^{-1}\right)\,\cdot\sum_{\tau}B_{\tau}^{(\rho)}(w)\,Z_{\rm aux}(s,w;\psi\tilde{\rho}\,\tau,\bar{\rho}),

and

(3.9) vSLv(w,ρv1)vS(1ρr(πv)qvrwr)Zaux(s,w;ψρ~,ρ¯)\displaystyle\prod_{v\in S_{\infty}}L_{v}\left(w,\rho_{v}^{-1}\right)\,\cdot\prod_{v\in S^{\prime}}\left(1-\rho^{r}(\pi_{v})\,q_{v}^{rw-r}\right)\cdot Z_{\rm aux}(s,w;\psi\tilde{\rho},\bar{\rho})
=vSLv(1w,ρv)τCτ(ρ)(1w)Z~(s+w12,1w;ψτ;ρ),\displaystyle\hskip 28.0pt=\prod_{v\in S_{\infty}}L_{v}\bigl{(}1-w,\rho_{v}\bigr{)}\,\cdot\sum_{\tau}C_{\tau}^{(\rho)}(1-w)\,\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi\,\tau;\rho),

where, as before, Bτ(ρ)(w),B_{\tau}^{(\rho)}(w), Cτ(ρ)(w)C_{\tau}^{(\rho)}(w) are polynomials in the variables qvw,qvwq_{v}^{w},\,q_{v}^{-w} (vS).(v\in S^{\prime}). The above products are over the places of kk corresponding to those in S,S, and the sums are over a finite set of id\́thinspaceele class characters τ,\tau, unramified outside SS and orders dividing r.r.

The proof of this proposition will be given in the next section.

Let α\alpha and β\beta be the involutions on 2{\mathbb{C}}^{2} given by

α:(s,w)(1s,2s+w1)andβ:(s,w)(s+w12,1w).\alpha:(s,w)\rightarrow(1-s,2s+w-1)\quad\hbox{and}\quad\beta:(s,w)\rightarrow(s+w-{\scriptstyle\frac{1}{2}},1-w).

It can be easily checked that these involutions generate the dihedral group D8D_{8} of order 8.8. It follows directly from Propositions 3.23.2 and 3.33.3 that bothZ~(s+w12,1w;ψ;ρ)\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho) and Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\tilde{\rho},\bar{\rho}) can be continued to 12{\mathcal{R}}_{1}\,\cup\,{\mathcal{R}}_{2}. Clearly, this applies to Zaux(s,w;ψ,ρ)Z_{\rm aux}(s,w;\psi,\rho) (replace ψ\psi by ψρ~1\psi\tilde{\rho}^{-1} and ρ\rho by ρ¯\bar{\rho}). It follows from the functional equation (3.6)(3.6) that Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\tilde{\rho},\bar{\rho}) can be continued to 12α(2),{\mathcal{R}}_{1}\,\cup\,{\mathcal{R}}_{2}\,\cup\,\alpha({\mathcal{R}}_{2}), and hence, by (3.8),(3.8), the function Z~(s+w12,1w;ψ;ρ)\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho) continues to this region. The double Dirichlet series Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\widetilde{\rho},\overline{\rho}) may have only one simple pole in 2{\mathcal{R}}_{2}, namely w=1w=1, and this pole occurs only if ρ\rho is the trivial character. This fact follows easily by inspection of the proof of Proposition 3.3 (see §3.1). Then from the functional equation (3.6), one can see that Zaux(s,w;ψρ~,ρ¯)Z_{\rm aux}(s,w;\psi\widetilde{\rho},\overline{\rho}) may have a pole only at w=22sw=2-2s in α(2)\alpha({\mathcal{R}}_{2}), provided ψr|𝒪Fρr\psi^{r}|_{{\mathcal{O}}_{F}}\cdot\rho^{r} is trivial. The last fact also applies to Z~(s+w12,1w;ψ,ρ)\widetilde{Z}(s+w-\frac{1}{2},1-w;\psi,\rho), by the functional equation β\beta in (3.8).

\Subsec

The double Dirichlet series Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) To define the perfect double Dirichlet series Z(s,w;ψ;ρ),Z(s,w;\psi;\rho), let LS(s,χ𝔫ψ),L_{S}(s,\chi_{\mathfrak{n}}\psi), for 𝔫F(S),{\mathfrak{n}}\in{\mathcal{I}}_{F}(S), be given by

LS(s,χ𝔫ψ):=LS(s,χ𝔫1ψ)P𝔫(s,ψ),L_{S}(s,\,\chi_{\mathfrak{n}}\psi)\,:=\,L_{S}(s,\,\chi_{{\mathfrak{n}}_{1}}\psi)P_{{\mathfrak{n}}}(s,\,\psi),

where 𝔫1{\mathfrak{n}}_{1} denotes the rr-th power free part of 𝔫,{\mathfrak{n}}, and P𝔫(s,ψ)P_{{\mathfrak{n}}}(s,\psi) is the Dirichlet polynomial defined by

(3.10)
P𝔫(s,ψ)=vordv(𝔫1)>0(1+ψ(πv)qv12s++ψ(πv)ordv(𝔫)1qv(ordv(𝔫)1)(12s))\displaystyle P_{{\mathfrak{n}}}(s,\,\psi)=\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{n}}_{1})>0}\end{subarray}}\Biggr{(}1+\psi(\pi_{v})\,q_{v}^{1-2s}+\cdots+\psi(\pi_{v})^{{\mathrm{ord}}_{v}({\mathfrak{n}})-1}q_{v}^{({\mathrm{ord}}_{v}({\mathfrak{n}})-1)(1-2s)}\Biggr{)}
vordv(𝔫)=rμvinert inL((1ψ(πv)qv2s)(1+ψ(πv)qv12s+\displaystyle\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{n}})=r\mu}\\ {v-\text{inert in}\;L}\end{subarray}}\Biggr{(}\Big{(}1-\psi(\pi_{v})\,q_{v}^{-2s}\Big{)}\left(1+\psi(\pi_{v})\,q_{v}^{1-2s}+\cdots\right.
+ψ(πv)rμ1qv(rμ1)(12s))+ψ(πv)rμqvrμ(12s)(1+qv1))\displaystyle\left.\qquad\qquad+\psi(\pi_{v})^{r\mu-1}\,q_{v}^{(r\mu-1)(1-2s)}\right)+\psi(\pi_{v})^{r\mu}\,q_{v}^{r\mu(1-2s)}\left(1+q_{v}^{-1}\right)\Biggr{)}
vordv(𝔫)=rωv=vv¯inL((1(χ𝔫1ψ)(πv)qvs)(1(χ𝔫1ψ)(πv¯)qvs)(1+ψ(πv)qv12s+\displaystyle\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{n}})=r\omega}\\ {v=v^{\prime}\bar{v}^{\prime}\;\text{in}\;L}\end{subarray}}\Biggr{(}(1-(\chi_{{\mathfrak{n}}_{1}}\psi)(\pi_{v^{\prime}})\,q_{v}^{-s})(1-(\chi_{{\mathfrak{n}}_{1}}\psi)(\pi_{\bar{v}^{\prime}})\,q_{v}^{-s})\big{(}1+\psi(\pi_{v})\,q_{v}^{1-2s}+\cdots
+ψ(πv)rω1qv(rω1)(12s))+ψ(πv)rωqvrω(12s)(1qv1)).\displaystyle\qquad\qquad+\psi(\pi_{v})^{r\omega-1}\,q_{v}^{(r\omega-1)(1-2s)}\big{)}+\psi(\pi_{v})^{r\omega}\,q_{v}^{r\omega(1-2s)}\left(1-q_{v}^{-1}\right)\Biggr{)}.

Here the products are over places vv of FF, and πv\pi_{v} denotes the local parameter of Fv.F_{v}. It can be seen that these polynomials satisfy a functional equation as s1s,s\to 1-s, and that we have the estimate

(3.11) P𝔫(s,ψ)εNF/(𝔫)ε(ε>0,Re(s)12).P_{{\mathfrak{n}}}(s,\,\psi)\ll_{\varepsilon}{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{\varepsilon}\;\;\;\;\;\;\;\;\;\;(\varepsilon>0,\;{\mathrm{Re}}(s)\geq{\scriptstyle\frac{1}{2}}).

Furthermore, if ψ(𝔪¯)=ψ(𝔪)¯,\psi(\overline{{\mathfrak{m}}})=\overline{\psi({\mathfrak{m}})}, for 𝔪L(S),{\mathfrak{m}}\in{\mathcal{I}}_{L}(S), then P𝔫(s,ψ)0,P_{{\mathfrak{n}}}(s,\,\psi)\geq 0, for ss\in{\mathbb{R}}. Later, we shall specialize ψ\psi to be (essentially) a normalized Jacobi sum, which obviously satisfies this property.

For Re(s),Re(w)>1,{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)>1, we define Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) as

Z(s,w;ψ;ρ)\displaystyle\qquad Z(s,w;\psi;\rho) =\displaystyle= ZS(s,w;ψ;ρ)\displaystyle Z_{S}(s,w;\psi;\rho)
=\displaystyle= LS(rs+rw+1r,ψrρ~r)𝔫F(S)LS(s,χ𝔫ψ)ρ(𝔫)NF/(𝔫)w.\displaystyle L_{S}(rs+rw+1-r,\,\psi^{r}\tilde{\rho}^{r})\sum_{\mathfrak{n}\in{\mathcal{I}}_{F}(S)}\frac{L_{S}(s,\,\chi_{\mathfrak{n}}\psi)\rho(\mathfrak{n})}{{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{w}}.

Applying the functional equation and the convexity bound of LS(s,χ𝔫ψ)L_{S}(s,\chi_{\mathfrak{n}}\psi)(𝔫F(S){\mathfrak{n}}\in{\mathcal{I}}_{F}(S)), we see that Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) is holomorphic in 1,{\mathcal{R}}_{1}, if the character ψr\psi^{r} is nontrivial. Representing the normalizing factor LS(rs+rw+1r,ψrρ~r)L_{S}(rs+rw+1-r,\psi^{r}\tilde{\rho}^{r}) by its Dirichlet series, then after multiplying and reorganizing, we can write Z(s,w;ψ;ρ)Z(s,w;\psi;\rho) as

(3.13) Z(s,w;ψ;ρ)=𝔫F(S)LS(s,χ𝔫1ψ)Q𝔫(s,ψ)ρ(𝔫)NF/(𝔫)w,Z(s,w;\psi;\rho)\;\;=\sum_{\mathfrak{n}\in{\mathcal{I}}_{F}(S)}\frac{L_{S}(s,\,\chi_{{\mathfrak{n}}_{1}}\psi)\,Q_{{\mathfrak{n}}}(s,\,\psi)\,\rho(\mathfrak{n})}{{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{w}},

where Q𝔫(s,ψ),Q_{{\mathfrak{n}}}(s,\psi), for 𝔫F(S),{\mathfrak{n}}\in{\mathcal{I}}_{F}(S), is a new set of Dirichlet polynomials which can be easily expressed in terms of P𝔫(s,ψ).P_{{\mathfrak{n}}}(s,\psi).

Referring to the definition of Z~(s,w;ψ;ρ)\widetilde{Z}(s,w;\psi;\rho) given in (3.7)(3.7), replace LS(w,χ𝔪ρ)L_{S}(w,\chi_{{\mathfrak{m}}}^{*}\,\rho) by its Dirichlet series, the sum being over 𝔫,{\mathfrak{n}}, say. For fixed 𝔪L(S){\mathfrak{m}}\in{\mathcal{I}}_{L}(S) imaginary, and 𝔫F(S),{\mathfrak{n}}\in{\mathcal{I}}_{F}(S), collect the terms contributing to (χ𝔪ρ)(𝔫)NF/(𝔫)w.(\chi_{{\mathfrak{m}}}^{*}\,\rho)({\mathfrak{n}})\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{-w}. Switching the order of summation, we obtain:

Proposition 3.6

For Re(s),Re(w)>1{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)>1,

(3.14) Z(s,w;ψ;ρ)=LS(2s,ψ)Z~(s,w;ψ;ρ),Z(s,w;\psi;\rho)\,=\,L_{S}(2s,\psi)\widetilde{Z}(s,w;\psi;\rho),

where the LL-function is defined over FF.

Assuming both ψr\psi^{r} and ψrρ~r\psi^{r}\tilde{\rho}^{r} to be nontrivial, we see from Proposition 3.43.4 that

LS(2s+2w1,ψ)Z~(s+w12,1w;ψ;ρ)L_{S}(2s+2w-1,\psi)\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho)

continues to β(1),\beta({\mathcal{R}}_{1}), and hence, from the above discussion, it continues to 1β(1)2α(2){\mathcal{R}}_{1}\,\cup\beta({\mathcal{R}}_{1})\,\cup\,{\mathcal{R}}_{2}\,\cup\,\alpha({\mathcal{R}}_{2}). Note that the convex closure of this tube region is 2{\mathbb{C}}^{2}. As ψrρ~r1\psi^{r}\tilde{\rho}^{r}\neq 1, and therefore, by Propositions 3.23.2 and 3.33.3, the function Z~(s+w12,1w;ψ;ρ)\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho) does not have a pole at s=12+1rs=\frac{1}{2}+\frac{1}{r}, one can easily check that the only possible poles of LS(2s+2w1,ψ)Z~(s+w12,1w;ψ;ρ)L_{S}(2s+2w-1,\psi)\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho) are the hyperplanes w=0w=0 and w=22sw=2-2s. Clearly, both are simple poles, and they may occur only if ρ\rho and ψr|𝒪Fρr\psi^{r}|_{{\mathcal{O}}_{F}}\cdot\rho^{r} are both trivial.

Consequently, by the convexity theorem for holomorphic functions of several complex variables (see [16]) and by Proposition 3.4,3.4, we have the following:

Theorem 3.7

When ψr\psi^{r} and ψrρ~r\psi^{r}\tilde{\rho}^{r} are nontrivial, the function

(w1)(2s+w2)Z(s,w;ψ;ρ)(w-1)(2s+w-2)Z(s,w;\psi;\rho)

has analytic continuation to 2,{\mathbb{C}}^{2}, and for any fixed s,s, it is (as a function of the variable ww) of order one.

The fact that, for any fixed s,s, the above function is of order one follows as in [8, Prop. 3.113.11].

By Proposition 3.43.4 and (3.7),(3.7), one finds that, for Re(s)>12,{\mathrm{Re}}(s)>\frac{1}{2}, (3.15)

Resw=1Z(s,w;ψ;1)=LS(2s,ψ)LS(rs+1,ψr)\displaystyle\underset{w=1}{{\mathrm{Res}}}\ Z(s,w;\psi;1)=L_{S}(2s,\psi)\,L_{S}(rs+1,\psi^{r})
vinFvS[(1qv1)𝔪L(S)𝔪imaginary(κψ(𝔪)rv|𝔪(1qv1)NL/(𝔪)rs𝔥F(S)ψ(𝔥)rNF/(𝔥)2rs\displaystyle\qquad\cdot\prod_{\begin{subarray}{c}{v\;\text{in}\;F}\\ {v\in S^{\prime}}\end{subarray}}\Biggr{[}\left(1-q_{v}^{-1}\right)\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{m}}-\text{imaginary}}\end{subarray}}\Biggr{(}\frac{\kappa\,\psi({\mathfrak{m}})^{r}\,\prod_{v^{\prime}|{\mathfrak{m}}}\left(1-q_{v^{\prime}}^{-1}\right)}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{rs}}\sum_{{\mathfrak{h}}\in{\mathcal{I}}_{F}(S)}\frac{\psi({\mathfrak{h}})^{r}}{{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{h}})^{2rs}}
vordv(NL/F(𝔪))>0ordv(𝔥)>0( 1qv1)vsplit inLordv(NL/F(𝔪))=0ordv(𝔥)>0( 1qv1)2vinert in Lordv(𝔥)>0(1qv2))]\displaystyle\qquad\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))>0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}})>0}\end{subarray}}\big{(}\,1\,-\,q_{v}^{-1}\,\big{)}\prod_{\begin{subarray}{c}{v-\text{split in}\ L}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))=0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}})>0}\end{subarray}}\big{(}\,1\,-\,q_{v}^{-1}\,\big{)}^{2}\prod_{\begin{subarray}{c}{v-\text{inert in $L$}}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}})>0}\end{subarray}}\big{(}1\,-\;q_{v}^{-2}\,\big{)}\Biggr{)}\Biggr{]}
=κLS(2s,ψ)LS(rs,ψr)vinFvS(1qv1),\displaystyle\quad=\kappa L_{S}(2s,\psi)\,L_{S}(rs,\psi^{r})\prod_{\begin{subarray}{c}{v\;\text{in}\;F}\\ {v\in S^{\prime}}\end{subarray}}\left(1-q_{v}^{-1}\right),

where κ\kappa denotes the residue at w=1w=1 of the Dedekind zeta-function ζF(w).\zeta_{F}(w).

We are now in the position to give the proof of Theorem 3.3.3.3.

\demo

Proof of Theorem 3.33.3 As before, let ρ=ρv\rho=\prod\rho_{v} be a unitary Hecke character of FF unramified outside S.S. We further assume that ρ\rho is of finite order. For Re(s),Re(w)>1,{\mathrm{Re}}(s),\,{\mathrm{Re}}(w)>1, consider the double Dirichlet series Z1(s,w;ψ;ρ)Z_{1}(s,w;\psi;\rho) defined by

(3.16). Z1(s,w;ψ;ρ)=𝔫F(S)𝔫=(n)[𝔫]= 1LS(s,χ𝔫1ψ)Q𝔫(s,ψ)ρ(𝔫)NF/(𝔫)w.Z_{1}(s,w;\psi;\rho)\;\;\,=\;\sum_{\begin{subarray}{c}{{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\\ {{\mathfrak{n}}\,=\,(n)}\\ {[{\mathfrak{n}}]\,=\,1}\end{subarray}}\frac{L_{S}(s,\,\chi_{{\mathfrak{n}}_{1}}\psi)\,Q_{{\mathfrak{n}}}(s,\,\psi)\,\rho(\mathfrak{n})}{{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{w}}.

By expressing this function as

Z1(s,w;ψ;ρ)=1hF|R𝔠|ρ1,ρ2Z(s,w;ψ;ρρ1ρ^2),Z_{1}(s,w;\psi;\rho)\;=\;\frac{1}{h_{F}\cdot|R_{\mathfrak{c}}|}\,\sum_{\rho_{1},\,\rho_{2}}\,Z(s,w;\psi;\rho\rho_{1}\widehat{\rho}_{2}),

where ρ1\rho_{1} ranges over the characters of the ideal class group of FF, ρ2\rho_{2} ranges over the characters of R𝔠,R_{\mathfrak{c}}, and ρ^2\widehat{\rho}_{2} is the restriction of ρ2\rho_{2} to FF, it follows from Theorem 3.53.5 that Z1(s,w;ψ;ρ)Z_{1}(s,w;\psi;\rho) is holomorphic on 2,{\mathbb{C}}^{2}, except for w=1w=1 and w=22sw=2-2s, where it might have simple poles. Furthermore,

limw1(w1)2Z1(12,w;ψ;ρ)=lim(s,w)(12,1)(w1)(2s+w2)Z1(s,w;ψ;ρ)= 0,\lim_{w\rightarrow 1}(w-1)^{2}\,Z_{1}({\scriptstyle\frac{1}{2}},w;\psi;\rho)\;\;\;=\lim_{(s,w)\rightarrow(\frac{1}{2},1)}(w-1)(2s+w-2)Z_{1}(s,w;\psi;\rho)\,=\,0,

and, therefore, Z1(12,w;ψ;1)Z_{1}(\frac{1}{2},w;\psi;1) has at most a simple pole at w=1.w=1. To compute its residue, recall the functional equation satisfied by L(s,χ𝔫1ψ)L(s,\chi_{{\mathfrak{n}}_{1}}\psi) with 𝔫1F(S){\mathfrak{n}}_{1}\in{\mathcal{I}}_{F}(S) rr-th power free (see [31, Ch. VII, §7]). Combining this with the functional equation of the polynomial Q𝔫(s,ψ)Q_{{\mathfrak{n}}}(s,\psi) (𝔫F(S){\mathfrak{n}}\in{\mathcal{I}}_{F}(S)), we find that

LS(s,χ𝔫1ψ)Q𝔫(s,ψ)\displaystyle L_{S}(s,\,\chi_{{\mathfrak{n}}_{1}}\psi)\,Q_{{\mathfrak{n}}}(s,\,\psi) =\displaystyle= ε(s,χ𝔫1ψ)LS(1s,χ𝔫1ψ)Q𝔫(1s,ψ)\displaystyle\varepsilon(s,\,\chi_{{\mathfrak{n}}_{1}}\psi)\cdot L_{S}(1-s,\,\chi_{{\mathfrak{n}}_{1}}\psi)\,Q_{{\mathfrak{n}}}(1-s,\,\psi)
vSLv(1s,ψv)Lv(s,ψv)vSLv(1s,(χ𝔫1ψ)v)Lv(s,(χ𝔫1ψ)v).\displaystyle\cdot\prod_{v\in S_{\infty}}\frac{L_{v}(1-s,\,\psi_{v})}{L_{v}(s,\,\psi_{v})}\,\cdot\prod_{v\in S^{\prime}}\frac{L_{v}\left(1-s,\,(\chi_{{\mathfrak{n}}_{1}}\psi)_{v}\right)}{L_{v}\left(s,\,(\chi_{{\mathfrak{n}}_{1}}\psi)_{v}\right)}.

A simple local computation shows that ε(12,χ𝔫1ψ)=ψ(𝔫)ε(12,ψ)\varepsilon(\frac{1}{2},\chi_{{\mathfrak{n}}_{1}}\psi)=\psi({\mathfrak{n}})\varepsilon(\frac{1}{2},\psi). It immediately follows that Z1(s,w;ψ;1)Z_{1}(s,w;\psi;1) satisfies the functional equation

(3.17) vSLv(s,ψv)vS(1ψr(πv)qvrsr)Z1(s,w;ψ;1)\displaystyle\prod_{v\in S_{\infty}}L_{v}(s,\psi_{v})\,\cdot\prod_{v\in S^{\prime}}\Big{(}1-\psi^{r}(\pi_{v})\,q_{v}^{rs-r}\Big{)}\cdot Z_{1}(s,w;\psi;1)
=vSLv(1s,ψv)ρDρ(ψ)(1s)Z1(1s,2s+w1;ψ;ρ),\displaystyle\qquad=\;\prod_{v\in S_{\infty}}L_{v}(1-s,\psi_{v})\,\cdot\sum_{\rho}D_{\rho}^{(\psi)}(1-s)\,Z_{1}(1-s,2s+w-1;\psi;\rho),

where Dρ(ψ)(s)D_{\rho}^{(\psi)}(s) are polynomials in the variables qvs,qvs,q_{v}^{s},\,q_{v}^{-s}, vS,v\in S^{\prime}, and the sum is over a finite set of idéle class characters ρ,\rho, unramified outside SS and orders dividing rr. As rr is odd, and ψ\psi, restricted to the group of principal ideals of FF, is quadratic and nontrivial, it follows that Z1(s,w;ψ;1)Z_{1}(s,w;\psi;1) does not have a pole at w=22sw=2-2s. Then (3.15)(3.15) yields

(3.18) Resw=1Z1(12,w;ψ;1)=κκ𝔠hF|R𝔠|LS(1,ψ)LS(r2,ψr)vinFvS(1qv1),\underset{w=1}{{\mathrm{Res}}}\ Z_{1}\left(\frac{1}{2},w;\psi;1\right)=\frac{\kappa\cdot\kappa_{\mathfrak{c}}}{h_{F}\cdot|R_{\mathfrak{c}}|}L_{S}(1,\psi)L_{S}\left(\frac{r}{2},\psi^{r}\right)\prod_{\begin{subarray}{c}{v\,\text{in}\,F}\\ {v\in S^{\prime}}\end{subarray}}(1-q_{v}^{-1}),

where κ𝔠\kappa_{\mathfrak{c}} denotes the number of characters of R𝔠R_{\mathfrak{c}} whose restrictions to FF are also characters of the ideal class group of FF.

To complete the proof, we define the double Dirichlet series Z0(s,w;ψ;ρ)Z_{0}(s,w;\psi;\rho) by simply replacing in (3.16)(3.16) the polynomial Q𝔫(s,ψ)Q_{\mathfrak{n}}(s,\psi) by P𝔫(s,ψ)P_{\mathfrak{n}}(s,\psi) defined in (3.10)(3.10). Note that

Z0(s,w;ψ;ρ)=1hF|R𝔠|ρ1,ρ2Z(s,w;ψ;ρρ1ρ2)LS(rs+rw+1r,ψrρ~rρ~1r),Z_{0}(s,w;\psi;\rho)=\frac{1}{h_{F}\cdot|R_{\mathfrak{c}}|}\sum_{\rho_{1},\rho_{2}}\frac{Z(s,w;\psi;\rho\rho_{1}\rho_{2})}{L_{S}(rs+rw+1-r,\psi^{r}\widetilde{\rho}^{r}\widetilde{\rho}^{r}_{1})},

and therefore, Z0(s,w;ψ;ρ)Z_{0}(s,w;\psi;\rho) may have additional poles at the zeros of the incomplete LL-functions LS(rs+rw+1r,ψrρ~rρ~1r)L_{S}(rs+rw+1-r,\psi^{r}\widetilde{\rho}^{r}\widetilde{\rho}^{r}_{1}). It is well-known that these zeros occur in the region Re(s+w)<1.{\mathrm{Re}}(s+w)<1. In particular, the function Z0(12,w;ψ;1)Z_{0}(\frac{1}{2},w;\psi;1) is holomorphic for Re(w)>12,{\mathrm{Re}}(w)>\frac{1}{2}, except for w=1w=1, where it has a simple pole. Using (3.18)(3.18), we can compute its residue as

(3.19) Resw=1Z0(12,w;ψ;1)=κκ𝔠hF|R𝔠|LS(1,ψ)LS(r2,ψr)LS(r2+1,ψr)vinFvS(1qv1)>0.\underset{w=1}{{\mathrm{Res}}}\ Z_{0}\left(\frac{1}{2},w;\psi;1\right)\,=\,\frac{\kappa\cdot\kappa_{\mathfrak{c}}}{h_{F}\cdot|R_{\mathfrak{c}}|}\,\frac{L_{S}(1,\psi)\,L_{S}(\frac{r}{2},\psi^{r})}{L_{S}(\frac{r}{2}+1,\psi^{r})}\prod_{\begin{subarray}{c}{v\;\text{in}\;F}\\ {v\in S^{\prime}}\end{subarray}}\left(1-q_{v}^{-1}\right)>0.

This implies that LS(12,χ𝔫1ψ)0L_{S}(\frac{1}{2},\chi_{{\mathfrak{n}}_{1}}\psi)\neq 0 for infinitely many rr-th power free ideals 𝔫1{\mathfrak{n}}_{1} in F(S){\mathcal{I}}_{F}(S) with trivial image in R𝔠R_{\mathfrak{c}}, which is the first assertion of Theorem 3.3.

For the remaining part, one needs to apply a Tauberian theorem. To keep the argument as simple as possible, note first that, as ψ(𝔪¯)=ψ(𝔪)¯,\psi(\overline{{\mathfrak{m}}})=\overline{\psi({\mathfrak{m}})}, for 𝔪L(S),{\mathfrak{m}}\in{\mathcal{I}}_{L}(S), we have P𝔫(s,ψ)0,P_{\mathfrak{n}}(s,\psi)\geq 0, for s.s\in{\mathbb{R}}. On the other hand, by the comment made right after Lemma 3.2,3.2, any rr-th power free ideal 𝔫1{\mathfrak{n}}_{1} in F(S){\mathcal{I}}_{F}(S) with trivial image in R𝔠R_{\mathfrak{c}} can be decomposed as 𝔫1=(n1)𝔤r{\mathfrak{n}}_{1}=(n_{1}){\mathfrak{g}}^{r} with n1F×,n11mod𝔠n_{1}\in F^{\times},\,\,n_{1}\equiv 1\mod{\mathfrak{c}} and 𝔤IF(S).{\mathfrak{g}}\in I_{F}(S). By definition, the character χ𝔫1\chi_{{\mathfrak{n}}_{1}} coincides with the classical rr-th power residue symbol χn1\chi_{n_{1}} given by class field theory. It follows that the incomplete LL-series LS(s,χ𝔫1ψ)L_{S}(s,\chi_{{\mathfrak{n}}_{1}}\psi) differs from the complete Hecke LL-series associated to L(s,χn1ψ)L(s,\chi_{n_{1}}\psi) by only finitely many local factors. Recall that the latter is the LL-series associated to a Hilbert modular form. As the set SS^{\prime} is closed under conjugation, it follows from a well-known result of Waldspurger [31] that LS(12,χ𝔫ψ)0,L_{S}(\frac{1}{2},\chi_{\mathfrak{n}}\psi)\geq 0, for 𝔫F(S),𝔫=(n){\mathfrak{n}}\in{\mathcal{I}}_{F}(S),\,\,{\mathfrak{n}}=(n) and trivial image in R𝔠.R_{\mathfrak{c}}. Hence, the function Z0(12,w;ψ;1),Z_{0}({\scriptstyle\frac{1}{2}},w;\psi;1), for (w)>1,\mathfrak{R}(w)>1, is given by a Dirichlet series with nonnegative coefficients. The second part of Theorem 3.33.3 now follows from the Wiener-Ikehara Tauberian theorem. \Endproof

Remark. With some additional effort, one can exhibit an error term on the order of O(xθ)O(x^{\theta}) with θ<1\theta<1 in the asymptotic formula (3.2).(3.2). Also, the remark following Theorem 3.33.3 implies that the Hecke LL-series LS(12,χ𝔫1ψ)0L_{S}(\frac{1}{2},\chi_{{\mathfrak{n}}_{1}}\psi)\neq 0 for infinitely many square-free principal ideals (n)(n) in F(S){\mathcal{I}}_{F}(S) with trivial image in R𝔠.R_{\mathfrak{c}}. Any such ideal has a generator nFn\in F with n1mod𝔠.n\equiv 1\mod{\mathfrak{c}}.

\Subsec

Proof of Proposition 3.33.3 Recall that for 𝔞L(S),\mathfrak{a}\in{\mathcal{I}}_{L}(S), we defined χ𝔞\chi_{\mathfrak{a}}^{*} by χ𝔞(𝔟):=χ𝔟(𝔞)\chi_{\mathfrak{a}}^{*}(\mathfrak{b}):=\chi_{\mathfrak{b}}(\mathfrak{a}) (𝔟L(S)\mathfrak{b}\in{\mathcal{I}}_{L}(S)). Note that every ideal 𝔪{\mathfrak{m}} of 𝒪L{\mathcal{O}}_{L} can be uniquely decomposed as 𝔪=𝔪𝔥,{\mathfrak{m}}={\mathfrak{m}}^{\prime}{\mathfrak{h}}, where 𝔪{\mathfrak{m}}^{\prime} is an imaginary ideal of 𝒪L,{\mathcal{O}}_{L}, and 𝔥{\mathfrak{h}} is a real ideal; that is, 𝔥𝒪F.{\mathfrak{h}}\in{\mathcal{O}}_{F}. For 𝔪L(S){\mathfrak{m}}\in{\mathcal{I}}_{L}(S) imaginary and rr-th power free, let ε(w,(χ𝔪ρ)1)\varepsilon(w,(\chi_{{\mathfrak{m}}}^{*}\,\rho)^{-1}) denote the epsilon-factor in the functional equation of L(w,(χ𝔪ρ)1)L(w,(\chi_{{\mathfrak{m}}}^{*}\,\rho)^{-1}) (as a Hecke LL-function of FF). Also, for 𝔪L(S){\mathfrak{m}}\in{\mathcal{I}}_{L}(S) imaginary and 𝔥F(S),{\mathfrak{h}}\in{\mathcal{I}}_{F}(S), coprime and rr-th power free, let G(χ𝔪𝔥)G(\chi_{{\mathfrak{m}}{\mathfrak{h}}}^{*}) be the normalized Gauss sum in the functional equation of the Hecke LL-function (of the field LL) associated to χ𝔪𝔥,\chi_{{\mathfrak{m}}{\mathfrak{h}}}^{*}, i.e., ε(12,χ𝔪𝔥).\varepsilon(\frac{1}{2},\chi_{{\mathfrak{m}}{\mathfrak{h}}}^{*}). We set 𝔪0{\mathfrak{m}}_{0} and 𝔥0{\mathfrak{h}}_{0} to be the product of all distinct prime ideals dividing 𝔪{\mathfrak{m}} and 𝔥,{\mathfrak{h}}, respectively.

The following lemma is a consequence of a standard local computation. The details will be omitted.

Lemma 3.8

Let 𝔪{\mathfrak{m}} and 𝔥{\mathfrak{h}} be integral ideals as above. Assume that the images of 𝔪𝔥{\mathfrak{m}}{\mathfrak{h}} and 𝔪{\mathfrak{m}} in R𝔠R_{\mathfrak{c}} are 𝔢\mathfrak{e} and 𝔢,{\mathfrak{e}}^{\prime}, respectively. Then,

G(χ𝔪𝔥)ε(12,(χ𝔪ρ)1)=C𝔢,𝔢,ρη(𝔢)1η(𝔪1𝔥1)ρ~(𝔪0)1χ𝔪(𝔥0)χ𝔥(𝔪0)χ𝔪(𝔪¯0)1,G(\chi_{{\mathfrak{m}}{\mathfrak{h}}}^{*})\,\varepsilon\left(\frac{1}{2},(\chi_{{\mathfrak{m}}}^{*}\,\rho)^{-1}\right)\\ =\,C_{{\mathfrak{e}},\,{\mathfrak{e}}^{\prime},\,\rho}\cdot\,\eta({\mathfrak{e}})^{-1}\eta({\mathfrak{m}}_{1}{\mathfrak{h}}_{1})\,\tilde{\rho}({\mathfrak{m}}_{0})^{-1}\,\chi_{{\mathfrak{m}}}^{*}({\mathfrak{h}}_{0})\,\chi_{{\mathfrak{h}}}^{*}({\mathfrak{m}}_{0})\,\chi_{{\mathfrak{m}}}^{*}(\overline{{\mathfrak{m}}}_{0})^{-1},

where ρ~=ρNL/F,\tilde{\rho}=\rho\,\circ\,{\mathrm{N}}_{L/F}, C𝔢,𝔢,ρC_{{\mathfrak{e}},\,{\mathfrak{e}}^{\prime},\,\rho} is a constant depending on just 𝔢,{\mathfrak{e}}, 𝔢{\mathfrak{e}}^{\prime} and ρ,\rho, and η\eta is a Hecke character unramified outside SS and order dividing r.r. Furthermore, if 𝔢{\mathfrak{e}}^{\prime} is replaced by 𝔢′′{\mathfrak{e}}^{\prime\prime} with 𝔢/𝔢′′{\mathfrak{e}}^{\prime}/{\mathfrak{e}}^{\prime\prime} a real ideal, then both C𝔢,𝔢,ρC_{{\mathfrak{e}},\,{\mathfrak{e}}^{\prime},\,\rho} and η\eta do not change.

Proof of Proposition 3.33.3. Using (3.5),(3.5), we have

(3.20)
Zaux(s,w;ψρ~,ρ¯)\displaystyle Z_{\rm aux}(s,w;\psi\,\tilde{\rho},\bar{\rho})
=𝔫F(S)ΨS(s,𝔫,ψρ~)ρ(𝔫)¯NF/(𝔫)w\displaystyle\qquad=\sum_{\mathfrak{n}\in{\mathcal{I}}_{F}(S)}\frac{\Psi_{S}(s,{\mathfrak{n}},\psi\,\tilde{\rho})\,\overline{\rho(\mathfrak{n})}}{{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{w}}
=LS(rsr2+1,ψrρ~r)𝔪L(S)𝔫F(S)(ψρ~)(𝔪)ρ(𝔫)¯G(𝔫,𝔪)NL/(𝔪)sNF/(𝔫)w\displaystyle\qquad=L_{S}\left(rs-\frac{r}{2}+1,\psi^{r}\tilde{\rho}^{r}\right)\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\end{subarray}}\frac{(\psi\,\tilde{\rho})({\mathfrak{m}})\,\overline{\rho(\mathfrak{n})}\,G({\mathfrak{n}},{\mathfrak{m}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}\,{\mathrm{N}}_{F/{\mathbb{Q}}}(\mathfrak{n})^{w}}
=LS(rsr2+1,ψrρ~r)𝔪L(S)𝔫F(S)(ψρ~)(𝔪)ρ(𝔫)¯χ𝔪1(𝔫)¯G(χ𝔪1)G0(𝔫,𝔪)NL/(𝔪)sNF/(𝔫)w,\displaystyle\qquad=L_{S}\left(rs-\frac{r}{2}+1,\psi^{r}\tilde{\rho}^{r}\right)\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\end{subarray}}\frac{(\psi\,\tilde{\rho})({\mathfrak{m}})\,\overline{\rho(\mathfrak{n})}\,\,\overline{\chi_{{\mathfrak{m}}_{1}}^{*}({\mathfrak{n}}^{*})}\,G(\chi_{{\mathfrak{m}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{w}},

where 𝔫{\mathfrak{n}}^{*} denotes the part of 𝔫{\mathfrak{n}} coprime to 𝔪1.{\mathfrak{m}}_{1}. In the last sum, replace 𝔪{\mathfrak{m}} by 𝔪𝔥{\mathfrak{m}}{\mathfrak{h}} with 𝔪L(S){\mathfrak{m}}\in{\mathcal{I}}_{L}(S) imaginary and 𝔥{\mathfrak{h}} real. As we shall see, the only contribution to the sum comes from 𝔪{\mathfrak{m}} and 𝔥{\mathfrak{h}} for which their rr-th power free parts 𝔪1{\mathfrak{m}}_{1} and 𝔥1{\mathfrak{h}}_{1} are coprime. Then, we have

(3.21) 𝔪L(S)𝔫F(S)(ψρ~)(𝔪)ρ(𝔫)¯χ𝔪1(𝔫)¯G(χ𝔪1)G0(𝔫,𝔪)NL/(𝔪)sNF/(𝔫)w=𝔪L(S)𝔪imaginary(ψρ~)(𝔪)NL/(𝔪)s\displaystyle\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\end{subarray}}\frac{(\psi\,\tilde{\rho})({\mathfrak{m}})\,\overline{\rho(\mathfrak{n})}\,\,\overline{\chi_{{\mathfrak{m}}_{1}}^{*}({\mathfrak{n}}^{*})}\,G(\chi_{{\mathfrak{m}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{w}}\;\;\;\;\,=\sum_{\begin{subarray}{c}{{\mathfrak{m}}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{m}}-\text{imaginary}}\end{subarray}}\frac{(\psi\,\tilde{\rho})({\mathfrak{m}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}}
𝔥L(S)𝔫F(S)𝔥real(ψρ~)(𝔥)ρ(𝔫)¯χ𝔪1𝔥1(𝔫)¯G(χ𝔪1𝔥1)G0(𝔫,𝔪𝔥)NL/(𝔥)sNF/(𝔫)w.\displaystyle\qquad\qquad\cdot\sum_{\begin{subarray}{c}{{\mathfrak{h}}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\\ {{\mathfrak{h}}-\text{real}}\end{subarray}}\frac{(\psi\,\tilde{\rho})({\mathfrak{h}})\,\overline{\rho(\mathfrak{n})}\,\,\overline{\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*}({\mathfrak{n}}^{*})}\,G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}}{\mathfrak{h}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{h}})^{s}\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{w}}.

Next, we separate the contribution of 𝔥{\mathfrak{h}} in the inner sum. To do so, let 𝔪1{\mathfrak{m}}_{1} denote the rr-th power free part of an ideal 𝔪L(S),\mathfrak{m}\in{\mathcal{I}}_{L}(S), and set 𝔪0{\mathfrak{m}}_{0} to be the product of all distinct prime ideals dividing 𝔪1,{\mathfrak{m}}_{1}, and

𝔪2:=vordv(𝔪)=rev𝔭vrev.{\mathfrak{m}}_{2}\;\;\;\;:=\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{m}})=re_{v}}\end{subarray}}{\mathfrak{p}}_{v}^{re_{v}}.

For fixed 𝔪,{\mathfrak{m}}, 𝔫{\mathfrak{n}} and 𝔥{\mathfrak{h}} as above, let 𝔭v{\mathfrak{p}}_{v} be a prime ideal of LL dividing 𝔥0.{\mathfrak{h}}_{0}. Upon replacing this prime ideal by its conjugate, we can assume that ordv(𝔪)=0.{\mathrm{ord}}_{v}({\mathfrak{m}})=0. Recall that

G0(𝔫,𝔪)=vordv(𝔫)=kordv(𝔪)=lG0(𝔭vk,𝔭vl),G_{0}({\mathfrak{n}},\,{\mathfrak{m}})\;\;\;=\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{n}})=k}\\ {{\mathrm{ord}}_{v}({\mathfrak{m}})=l}\end{subarray}}G_{0}({\mathfrak{p}}_{v}^{k},{\mathfrak{p}}_{v}^{l}),

where G0(𝔭vk,𝔭vl)G_{0}({\mathfrak{p}}_{v}^{k},\,{\mathfrak{p}}_{v}^{l}) is given by (3.4).(3.4). As ordv(𝔪𝔥)=ordv(𝔥)0(modr){\mathrm{ord}}_{v}({\mathfrak{m}}{\mathfrak{h}})={\mathrm{ord}}_{v}({\mathfrak{h}})\not\equiv 0\pmod{r} (this condition implying that ordv(𝔫)=ordv(𝔥)1{\mathrm{ord}}_{v}({\mathfrak{n}})={\mathrm{ord}}_{v}({\mathfrak{h}})-1), and 𝔫F(S),{\mathfrak{n}}\in{\mathcal{I}}_{F}(S), we can decompose 𝔫=(𝔥/𝔥0𝔥2)𝔫{\mathfrak{n}}=({\mathfrak{h}}/{\mathfrak{h}}_{0}{\mathfrak{h}}_{2}){\mathfrak{n}}^{\prime} with 𝔫F(S){\mathfrak{n}}^{\prime}\in{\mathcal{I}}_{F}(S) coprime to 𝔥1.{\mathfrak{h}}_{1}. Also, we have

ordv(𝔫)\displaystyle{\mathrm{ord}}_{v}({\mathfrak{n}}) =\displaystyle= ordv¯(𝔫)ordv¯(𝔪𝔥)1\displaystyle{\mathrm{ord}}_{\bar{v}}({\mathfrak{n}})\,\geq\,{\mathrm{ord}}_{\bar{v}}({\mathfrak{m}}{\mathfrak{h}})-1
=\displaystyle= ordv¯(𝔪)+ordv(𝔥)1=ordv¯(𝔪)+ordv(𝔫),\displaystyle\,{\mathrm{ord}}_{\bar{v}}({\mathfrak{m}})+{\mathrm{ord}}_{v}({\mathfrak{h}})-1\,=\,{\mathrm{ord}}_{\bar{v}}({\mathfrak{m}})+{\mathrm{ord}}_{v}({\mathfrak{n}}),

which implies ordv¯(𝔪)=0.{\mathrm{ord}}_{\bar{v}}({\mathfrak{m}})=0. It immediately follows that 𝔪{\mathfrak{m}} and 𝔥1{\mathfrak{h}}_{1} are coprime. Then, by (3.4),(3.4), we can write

G(χ𝔪1𝔥1)G0(𝔫,𝔪𝔥)\displaystyle G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}}{\mathfrak{h}})\, =\displaystyle= G(χ𝔪1𝔥1)G0(𝔥𝔥0𝔥2,𝔥𝔥2)G0(𝔫,𝔪𝔥2)\displaystyle G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})\,G_{0}\left(\frac{{\mathfrak{h}}}{{\mathfrak{h}}_{0}{\mathfrak{h}}_{2}},\frac{{\mathfrak{h}}}{{\mathfrak{h}}_{2}}\right)G_{0}({\mathfrak{n}}^{\prime},{\mathfrak{m}}{\mathfrak{h}}_{2})
=\displaystyle= G(χ𝔪1𝔥1)NL/(𝔥𝔥0𝔥2)12G0(𝔫,𝔪𝔥2).\displaystyle G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})\,{\mathrm{N}}_{L/{\mathbb{Q}}}\left(\frac{{\mathfrak{h}}}{{\mathfrak{h}}_{0}{\mathfrak{h}}_{2}}\right)^{\frac{1}{2}}G_{0}({\mathfrak{n}}^{\prime},{\mathfrak{m}}{\mathfrak{h}}_{2}).

Furthermore, we have

G0(𝔫,𝔪𝔥2)=vordv(𝔫)=kvordv(𝔪)=lvordv(𝔥2)=revG0(𝔭vkv,𝔭vlv+rev)\displaystyle\!{\bf}{\bf}G_{0}({\mathfrak{n}}^{\prime},{\mathfrak{m}}{\mathfrak{h}}_{2})=\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathfrak{n}}^{\prime})=k_{v}}\\ {{\mathrm{ord}}_{v}({\mathfrak{m}})=l_{v}}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})=re_{v}}\end{subarray}}G_{0}({\mathfrak{p}}_{v}^{k_{v}},{\mathfrak{p}}_{v}^{l_{v}+re_{v}})
=vlv0(r)kv+1=lv+revG0(𝔭vkv,𝔭vlv+rev)vlv0(r)kv+1lv+revG0(𝔭vkv,𝔭vlv+rev)\displaystyle\!=\prod_{\begin{subarray}{c}{v}\\ {l_{v}\not\equiv 0\,(r)}\\ {k_{v}+1=l_{v}+re_{v}}\end{subarray}}G_{0}({\mathfrak{p}}_{v}^{k_{v}},{\mathfrak{p}}_{v}^{l_{v}+re_{v}})\cdot\prod_{\begin{subarray}{c}{v}\\ {l_{v}\equiv 0\,(r)}\\ {k_{v}+1\geq l_{v}+re_{v}}\end{subarray}}G_{0}({\mathfrak{p}}_{v}^{k_{v}},{\mathfrak{p}}_{v}^{l_{v}+re_{v}})
=vlv0(r)kv+1=lv+revqv(lv1)+rev2vlv0(r)kv+1=lv+rev>0qvlv+rev22vlv0(r)kvlv+rev>0qvlv+rev2(1qv1)\displaystyle\!=\prod_{\begin{subarray}{c}{v}\\ {l_{v}\not\equiv 0\,(r)}\\ {k_{v}+1=l_{v}+re_{v}}\end{subarray}}q_{v}^{\frac{(l_{v}-1)+re_{v}}{2}}\cdot\prod_{\begin{subarray}{c}{v}\\ {l_{v}\equiv 0\,(r)}\\ {k_{v}+1=l_{v}+re_{v}>0}\end{subarray}}-\;\,q_{v}^{\frac{l_{v}+re_{v}-2}{2}}\cdot\prod_{\begin{subarray}{c}{v}\\ {l_{v}\equiv 0\,(r)}\\ {k_{v}\geq l_{v}+re_{v}>0}\end{subarray}}q_{v}^{\frac{l_{v}+re_{v}}{2}}(1-q_{v}^{-1})
=NL/(𝔪𝔥2𝔪0)12vlv0(r)kv+1=lv+rev>0qv1vlv0(r)kvlv+rev>0(1qv1).\displaystyle\!={\mathrm{N}}_{L/{\mathbb{Q}}}\left(\frac{{\mathfrak{m}}{\mathfrak{h}}_{2}}{{\mathfrak{m}}_{0}}\right)^{\frac{1}{2}}\cdot\prod_{\begin{subarray}{c}{v}\\ {l_{v}\equiv 0\,(r)}\\ {k_{v}+1=l_{v}+re_{v}>0}\end{subarray}}-\;\,q_{v}^{-1}\cdot\prod_{\begin{subarray}{c}{v}\\ {l_{v}\equiv 0\,(r)}\\ {k_{v}\geq l_{v}+re_{v}>0}\end{subarray}}(1-q_{v}^{-1}).

One can decompose 𝔫{\mathfrak{n}}^{\prime} as

𝔫\displaystyle{\mathfrak{n}}^{\prime} =\displaystyle= 𝔫1NL/F(𝔪𝔪0)𝔥2\displaystyle{\mathfrak{n}}_{1}\cdot\,{\mathrm{N}}_{L/F}\left(\frac{{\mathfrak{m}}}{{\mathfrak{m}}_{0}}\right)\cdot\,{\mathfrak{h}}_{2}
vcomplexlv0(r);lv¯=0lv+rev>0αv:=1+kvlvrev0NL/F(𝔭v)αv1vrealev>0βv:=1+kvrev0𝔮vβv1,\displaystyle\cdot\prod_{\begin{subarray}{c}{v-\text{complex}}\\ {l_{v}\equiv 0\,(r);\;l_{\bar{v}}=0}\\ {l_{v}+re_{v}>0}\\ {\alpha_{v}:=1+k_{v}-l_{v}-re_{v}\geq 0}\end{subarray}}{\mathrm{N}}_{L/F}({\mathfrak{p}}_{v})^{\alpha_{v}-1}\;\;\;\;\cdot\prod_{\begin{subarray}{c}{v-\text{real}}\\ {e_{v}>0}\\ {\beta_{v}:=1+k_{v}-re_{v}\geq 0}\end{subarray}}{\mathfrak{q}}_{v}^{\beta_{v}-1},

with 𝔫1{\mathfrak{n}}_{1} coprime to 𝔪𝔥.{\mathfrak{m}}{\mathfrak{h}}. Here, if vv is complex such that lv=lv¯=0,l_{v}=l_{\bar{v}}=0, then one chooses either vv or v¯,\bar{v}, but not both. As 𝔫=(𝔥/𝔥0𝔥2)𝔫,{\mathfrak{n}}=({\mathfrak{h}}/{\mathfrak{h}}_{0}{\mathfrak{h}}_{2}){\mathfrak{n}}^{\prime}, we also have

𝔫\displaystyle{\mathfrak{n}} =\displaystyle= 𝔫1NL/F(𝔪𝔪0)𝔥𝔥0\displaystyle{\mathfrak{n}}_{1}\cdot\,{\mathrm{N}}_{L/F}\left(\frac{{\mathfrak{m}}}{{\mathfrak{m}}_{0}}\right)\cdot\,\frac{{\mathfrak{h}}}{{\mathfrak{h}}_{0}}
vcomplexlv0(r);lv¯=0lv+rev>0αv:=1+kvlvrev0NL/F(𝔭v)αv1vrealev>0βv:=1+kvrev0𝔮vβv1.\displaystyle\cdot\prod_{\begin{subarray}{c}{v-\text{complex}}\\ {l_{v}\equiv 0\,(r);\;l_{\bar{v}}=0}\\ {l_{v}+re_{v}>0}\\ {\alpha_{v}:=1+k_{v}-l_{v}-re_{v}\geq 0}\end{subarray}}{\mathrm{N}}_{L/F}({\mathfrak{p}}_{v})^{\alpha_{v}-1}\;\;\;\;\cdot\prod_{\begin{subarray}{c}{v-\text{real}}\\ {e_{v}>0}\\ {\beta_{v}:=1+k_{v}-re_{v}\geq 0}\end{subarray}}{\mathfrak{q}}_{v}^{\beta_{v}-1}.

Recall that 𝔫{\mathfrak{n}}^{*} denotes the part of 𝔫{\mathfrak{n}} coprime to 𝔪1𝔥1.{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}. It follows that

𝔫\displaystyle{\mathfrak{n}}^{*} =\displaystyle= 𝔫1(𝔪¯𝔪¯0𝔪¯2)NL/F(𝔪2)𝔥2\displaystyle{\mathfrak{n}}_{1}\cdot\,\left(\frac{\overline{{\mathfrak{m}}}}{\overline{{\mathfrak{m}}}_{0}\overline{{\mathfrak{m}}}_{2}}\right)\cdot\,{\mathrm{N}}_{L/F}({\mathfrak{m}}_{2})\cdot{\mathfrak{h}}_{2}
vcomplexlv0(r);lv¯=0lv+rev>0αv:=1+kvlvrev0NL/F(𝔭v)αv1vrealev>0βv:=1+kvrev0𝔮vβv1.\displaystyle\cdot\prod_{\begin{subarray}{c}{v-\text{complex}}\\ {l_{v}\equiv 0\,(r);\;l_{\bar{v}}=0}\\ {l_{v}+re_{v}>0}\\ {\alpha_{v}:=1+k_{v}-l_{v}-re_{v}\geq 0}\end{subarray}}{\mathrm{N}}_{L/F}({\mathfrak{p}}_{v})^{\alpha_{v}-1}\;\;\;\cdot\prod_{\begin{subarray}{c}{v-\text{real}}\\ {e_{v}>0}\\ {\beta_{v}:=1+k_{v}-re_{v}\geq 0}\end{subarray}}{\mathfrak{q}}_{v}^{\beta_{v}-1}.

Combining all these with (4.26),(4.26), we obtain

𝔪L(S)𝔪imaginary(ψρ~)(𝔪)NL/(𝔪)s𝔥L(S)𝔫F(S)𝔥real(ψρ~)(𝔥)ρ(𝔫)¯χ𝔪1𝔥1(𝔫)¯G(χ𝔪1𝔥1)G0(𝔫,𝔪𝔥)NL/(𝔥)sNF/(𝔫)w=𝔪L(S)𝔪imaginaryψ(𝔪)ρ~(𝔪0)χ𝔪1(𝔪¯𝔪¯0)¯NL/(𝔪0)w12NL/(𝔪)s+w12𝔥F(S)(ψρ)(𝔥)ρ(𝔥0)NF/(𝔥0)w1χ𝔥1(𝔪)χ𝔥1(𝔪0)1G(χ𝔪1𝔥1)NF/(𝔥)2s+w1vordv(NL/F(𝔪1))>0ordv(𝔥2)>0(1qv1)vordv(NL/F(𝔪2))>0ordv(𝔥)=0[(χ𝔪1ρ)(πv)qvw1+(1qv1)αv0((χ𝔪1ρ)1(πv)qvw)αv]vordv(NL/F(𝔪2))>0ordv(𝔥2)>0[(χ𝔪1ρ)(πv)qvw1(1qv1)+(1qv1)2αv0((χ𝔪1ρ)1(πv)qvw)αv]vsplit in Lordv(NL/F(𝔪))=0ordv(𝔥2)>0[(χ𝔪1ρ)(πv)qvw2+(1qv1)2αv0((χ𝔪1ρ)1(πv)qvw)αv]\thinspace\begin{aligned} &\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{m}}-\text{imaginary}}\end{subarray}}\frac{(\psi\tilde{\rho})({\mathfrak{m}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s}}\;\sum_{\begin{subarray}{c}{{\mathfrak{h}}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{n}}\in{\mathcal{I}}_{F}(S)}\\ {{\mathfrak{h}}-\text{real}}\end{subarray}}\frac{(\psi\tilde{\rho})({\mathfrak{h}})\,\overline{\rho(\mathfrak{n})}\,\,\overline{\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*}({\mathfrak{n}}^{*})}\,G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})\,G_{0}({\mathfrak{n}},{\mathfrak{m}}{\mathfrak{h}})}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{h}})^{s}\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}})^{w}}\\ &=\sum_{\begin{subarray}{c}{\mathfrak{m}\in{\mathcal{I}}_{L}(S)}\\ {{\mathfrak{m}}-\text{imaginary}}\end{subarray}}\frac{\psi({\mathfrak{m}})\tilde{\rho}({\mathfrak{m}}_{0})\,\overline{\chi_{{\mathfrak{m}}_{1}}^{*}\left(\frac{\overline{{\mathfrak{m}}}}{\overline{{\mathfrak{m}}}_{0}}\right)}\,{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}}_{0})^{w-\frac{1}{2}}}{{\mathrm{N}}_{L/{\mathbb{Q}}}({\mathfrak{m}})^{s+w-\frac{1}{2}}}\\ &\cdot\sum_{{\mathfrak{h}}\in{\mathcal{I}}_{F}(S)}\frac{(\psi\rho)({\mathfrak{h}})\,\rho({\mathfrak{h}}_{0})\,{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{h}}_{0})^{w-1}\,\chi_{{\mathfrak{h}}_{1}}^{*}({\mathfrak{m}})\,\chi_{{\mathfrak{h}}_{1}}^{*}({\mathfrak{m}}_{0})^{-1}G(\chi_{{\mathfrak{m}}_{1}{\mathfrak{h}}_{1}}^{*})}{{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{h}})^{2s+w-1}}\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}_{1}))>0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}(1-q_{v}^{-1})\\ &\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}_{2}))>0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}})=0}\end{subarray}}\Big{[}\,-\;(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)(\pi_{v})\,q_{v}^{w-1}\,+\;(1-q_{v}^{-1})\;\cdot\sum_{\alpha_{v}\geq 0}(\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}(\pi_{v})\,q_{v}^{-w}\,)^{\alpha_{v}}\Big{]}\\ &\cdot\prod_{\begin{subarray}{c}{v}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}_{2}))>0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}\Big{[}\,-\;(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)(\pi_{v})\,q_{v}^{w-1}(1-q_{v}^{-1})\;+\;(1-q_{v}^{-1})^{2}\,\cdot\sum_{\alpha_{v}\geq 0}(\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}(\pi_{v})\,q_{v}^{-w}\,)^{\alpha_{v}}\Big{]}\\ &\cdot\prod_{\begin{subarray}{c}{v-\text{split in $L$}}\\ {{\mathrm{ord}}_{v}({\mathrm{N}}_{L/F}({\mathfrak{m}}))=0}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}\Big{[}\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)(\pi_{v})\,q_{v}^{w-2}\,+\;(1-q_{v}^{-1})^{2}\,\cdot\sum_{\alpha_{v}\geq 0}(\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}(\pi_{v})\,q_{v}^{-w}\,)^{\alpha_{v}}\Big{]}\end{aligned}
vinert in Lordv(𝔥2)>0[(χ𝔪1ρ)(πv)qvw2+(1qv2)βv0((χ𝔪1ρ)1(πv)qvw)βv]\displaystyle\cdot\prod_{\begin{subarray}{c}{v-\text{inert in $L$}}\\ {{\mathrm{ord}}_{v}({\mathfrak{h}}_{2})>0}\end{subarray}}\Big{[}\,-\;(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)(\pi_{v})\,q_{v}^{w-2}\,+\;(1-q_{v}^{-2})\,\cdot\sum_{\beta_{v}\geq 0}(\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}(\pi_{v})\,q_{v}^{-w}\,)^{\beta_{v}}\Big{]}
𝔫1F(S)(𝔫1,𝔪𝔥)= 1ρ(𝔫1)¯χ𝔪1(𝔫1)¯NF/(𝔫1)w.\displaystyle\cdot\sum_{\begin{subarray}{c}{{\mathfrak{n}}_{1}\in{\mathcal{I}}_{F}(S)}\\ {({\mathfrak{n}}_{1},{\mathfrak{m}}{\mathfrak{h}})\,=\,1}\end{subarray}}\;\frac{\overline{\rho({\mathfrak{n}}_{1})}\,\,\overline{\chi_{{\mathfrak{m}}_{1}}^{*}({\mathfrak{n}}_{1})}}{{\mathrm{N}}_{F/{\mathbb{Q}}}({\mathfrak{n}}_{1})^{w}}.

Note that the last sum represents an incomplete Hecke LL-function. After evaluating the geometric series inside the last four products, the missing Euler factors corresponding to places of FF dividing NL/F(𝔪2)𝔥2{\mathrm{N}}_{L/F}({\mathfrak{m}}_{2}){\mathfrak{h}}_{2} can be incorporated. Also, multiply and divide by the Euler factors corresponding to places of FF dividing 𝔥0,{\mathfrak{h}}_{0}, forcing in this way LS(w,(χ𝔪1ρ)1)L_{S}(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}) to appear.

Let R𝔠+R_{\mathfrak{c}}^{+} be the subgroup of R𝔠R_{\mathfrak{c}} generated by the images (in R𝔠R_{\mathfrak{c}}) of all real fractional ideals of LL coprime to S.S^{\prime}. Let 𝔢{\mathfrak{e}}^{\prime} be a fixed element of R𝔠R_{\mathfrak{c}} which is the image of an imaginary ideal 𝔪L(S).{\mathfrak{m}}\in{\mathcal{I}}_{L}(S). Replacing ψ\psi by ψτ1τ2\psi\tau_{1}\tau_{2} with τ1\tau_{1} and τ2\tau_{2} characters of R𝔠R_{\mathfrak{c}} and R𝔠/R𝔠+,R_{\mathfrak{c}}/R_{\mathfrak{c}}^{+}, respectively, and making a standard linear combination, one can restrict the first two sums over ideals 𝔪{\mathfrak{m}} and 𝔥,{\mathfrak{h}}, for which the image of 𝔪1{\mathfrak{m}}_{1} in R𝔠R_{\mathfrak{c}} is 𝔢{\mathfrak{e}}^{\prime} modulo R𝔠+R_{\mathfrak{c}}^{+} and the image of 𝔪1𝔥1{\mathfrak{m}}_{1}{\mathfrak{h}}_{1} is a fixed element 𝔢{\mathfrak{e}} of R𝔠R_{\mathfrak{c}}.

Now, invoke the functional equation of L(w,(χ𝔪1ρ)1).L(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}). It is well-known, see [31], that the incomplete Hecke LL-function (defined over FF)

LS(w,(χ𝔪1ρ)1)=vSLv(w,(χ𝔪1ρ)v1)=vS[1(χ𝔪1ρ)v1(πv)qvw]1L_{S}\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}\right)\,=\,\prod_{v\not\in S}L_{v}\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}^{-1}\right)\,=\,\prod_{v\not\in S}\,\big{[}1\,-\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}^{-1}(\pi_{v})\,q_{v}^{-w}\big{]}^{-1}

satisfies the functional equation

LS(w,(χ𝔪1ρ)1)\displaystyle L_{S}\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}\right) =\displaystyle= ε(w,(χ𝔪1ρ)1)LS(1w,χ𝔪1ρ)\displaystyle\varepsilon\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)^{-1}\right)\cdot L_{S}\left(1-w,\,\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho\right)
vSLv(1w,ρv)Lv(w,ρv1)vSLv(1w,(χ𝔪1ρ)v)Lv(w,(χ𝔪1ρ)v1).\displaystyle\cdot\prod_{v\in S_{\infty}}\frac{L_{v}\left(1-w,\,\rho_{v}\right)}{L_{v}\left(w,\,\rho_{v}^{-1}\right)}\cdot\prod_{v\in S^{\prime}}\frac{L_{v}\left(1-w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}\right)}{L_{v}\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}^{-1}\right)}.

Replace ψ\psi by ψη1,\psi\,\eta^{-1}, and combine the above functional equation with Lemma 3.6.3.6. Here Re(s){\mathrm{Re}}(s) is taken sufficiently large to ensure convergence. Using the Fisher-Friedberg extension of the reciprocity law [9], one can see that

χ𝔪1(𝔪¯)¯χ𝔥1(𝔪)=C𝔢,𝔢^χ𝔪(𝔥1),\overline{\chi_{{\mathfrak{m}}_{1}}^{*}(\overline{{\mathfrak{m}}})}\,\chi_{{\mathfrak{h}}_{1}}^{*}({\mathfrak{m}})\,=\,C_{{\mathfrak{e}},\,\widehat{{\mathfrak{e}}^{\prime}}}^{\prime}\cdot\chi_{{\mathfrak{m}}}^{*}({\mathfrak{h}}_{1}),

where C𝔢,𝔢^C_{{\mathfrak{e}},\,\widehat{{\mathfrak{e}}^{\prime}}}^{\prime} is a constant depending on just 𝔢{\mathfrak{e}} and the class 𝔢^\widehat{{\mathfrak{e}}^{\prime}} in R𝔠/R𝔠+.R_{\mathfrak{c}}/R_{\mathfrak{c}}^{+}. Also, note that

vS(1ρr(πv)qvrw)1Lv(1w,(χ𝔪1ρ)v)Lv(w,(χ𝔪1ρ)v1)\prod_{v\in S^{\prime}}\Bigl{(}1-\rho^{-r}(\pi_{v})\,q_{v}^{-rw}\Bigr{)}^{-1}\cdot\;\,\frac{L_{v}\left(1-w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}\right)}{L_{v}\left(w,\,(\chi_{{\mathfrak{m}}_{1}}^{*}\,\rho)_{v}^{-1}\right)}

is the inverse of a polynomial in the variables qvw,qvwq_{v}^{w},\,q_{v}^{-w} corresponding to places vSv\in S^{\prime} of the totally real field F.F. The characters involved in its coefficients are trivial on real ideals. Now, the functional equation (3.8)(3.8) immediately follows, after we replace ψ\psi with ψτ,\psi\tau, where τ\tau ranges over a finite set of idéle class characters unramified outside SS and orders dividing r,r, and make a combination such that the above product over vSv\in S^{\prime} disappears.

Starting from the definition of

vS(1ρr(πv)qvrwr)1Z~(s+w12,1w;ψ;ρ),\prod_{v\in S^{\prime}}\Bigl{(}1-\rho^{r}(\pi_{v})\,q_{v}^{rw-r}\Bigr{)}^{-1}\cdot\;\;\widetilde{Z}(s+w-{\scriptstyle\frac{1}{2}},1-w;\psi;\rho),

one can easily check (3.9)(3.9) by reversing the above argument. ∎

References

  • 911
  • [1] \nameD. Bump, Automorphic Forms and Representations, Cambridge Studies in Advanced Math55, Cambridge Univ. Press, Cambridge (1997).
  • [2] \nameD. Bump, S. Friedberg, and \nameJ. Hoffstein, Eisenstein series on the metaplectic group and nonvanishing theorems for automorphic LL-functions and their derivatives, Ann. of Math131 (1990), 53–127.
  • [3] \bibline, Nonvanishing theorems for LL-functions of modular forms and their derivatives, Invent. Math102 (1990), 543–618.
  • [4] \bibline, On some applications of automorphic forms to number theory, Bull. Amer. Math. Soc33 (1996), 157–175.
  • [5] \nameH. Carayol, Sur les representations \ell-adiques associées aux formes modulaires de Hilbert, Ann. Sci. École Norm. Sup19 (1986), 409–468.
  • [6] \bibline, Sur la mauvaise reduction des courbes de Shimura, Compositio Math59 (1986), 151–230.
  • [7] \nameA. Diaconu, Mean square values of Hecke LL-series formed with rr-th order characters, Invent. Math157 (2004), 635–684.
  • [8] \nameA. Diaconu, D. Goldfeld, and \nameJ. Hoffstein, Multiple Dirichlet series and moments of zeta and LL-functions, Compositio Math139 (2003), 297–360.
  • [9] \nameB. Fisher and \nameS. Friedberg, Double Dirichlet series over function fields, Compositio Math140 (2004), 613–630.
  • [10] \bibline, Sums of twisted GL(2){\rm GL}(2) LL-functions over function fields, Duke Math. J117 (2003), 543–570.
  • [11] \nameS. Friedberg and \nameJ. Hoffstein, Nonvanishing theorems for automorphic LL-functions on GL(2){\rm GL}(2), Ann. of Math142 (1995), 385–423.
  • [12] \nameS. Friedberg, J. Hoffstein, and \nameD. Lieman, Double Dirichlet series and the nn-th order twists of Hecke LL-series, Math. Ann327 (2003), 315–338.
  • [13] \nameB. Gross, Kolyvagin’s work on modular elliptic curves, LL-functions and arithmetic, Proc. Durham Sympos. (1989), 235–256, London Math. Soc. Lecture Note Ser153, Cambridge Univ. Press, Cambridge (1991).
  • [14] \nameB. Gross and \nameD. Rohrlich, Some results on the Mordell-Weil group of the Jacobian of the Fermat curve, Invent. Math44 (1978), 201–224.
  • [15] \nameJ. Hoffstein, Eisenstein series and theta functions on the metaplectic group, in Theta Functions: From the Classical to the Modern, 65–104, CRM Proc. Lect. Notes (M. Ram Murty ed.), A. M. S., Providence, R.I. (1993).
  • [16] \nameL. Hörmander, An Introduction to Complex Analysis in Several Variables, Van Nostrand, Princeton, N.J. (1966).
  • [17] \nameH. Jacquet and \nameR. P. Langlands, Automorphic Forms on GL2{\mathrm{GL}}_{2}, Lectures Notes in Math114, Springer-Verlag, New York (1971).
  • [18] \nameD. Kazhdan and \nameS. Patterson, Metaplectic forms, Publ. Math.  I.H.E.S59 (1984), 35–142.
  • [19] \nameW. Kohnen and \nameD. Zagier, Values of LL-series of modular forms at the center of the critical strip, Invent. Math64 (1981), 175–198.
  • [20] \nameV. A. Kolyvagin, Euler Systems, The Grothendieck FestschriftII, 435–483, Progr. Math87, Birkhäuser Boston, Boston, MA (1990).
  • [21] \bibline, Finiteness of E()E({\mathbb{Q}}) and X(E,)\hbox{\cyr X}(E,{\mathbb{Q}}) for a subclass of Weil curves (Russian), Izv. Akad. Nauk SSSR Ser. Mat52 (1988), 522–540, 670–671; English transl. in Math. USSR-Izv32 (1989), 523–541.
  • [22] \nameV. A. Kolyvagin and \nameD. Yu. Logachev, Finiteness of the Shafarevich-Tate group and the group of rational points for some modular abelian varieties (Russian) Algebra i Anal. 1 (1989), 171–196; English transl. in Leningrad Math. J1 (1990), 1229–1253.
  • [23] \bibline, Finiteness of X over totally real fields (Russian), Izv. Akad. Nauk SSSR Ser. Mat55 (1991), 851–876; English transl. in Math. USSR-Izv39 (1992), 829–853.
  • [24] \nameT. Kubota, On Automorphic Forms and the Reciprocity Law in a Number Field,Kinokuniya Book Store Co., Tokyo, 1969.
  • [25] \nameR. P. Langlands, On the Functional Equations Satisfied by Eisenstein Series, Lecture Notes in Math544, Springer-Verlag, New York (1976).
  • [26] \nameD. Lieman, Nonvanishing of LL-series associated to cubic twists of elliptic curves, Ann. of Math140 (1994), 81–108.
  • [27] \nameJ. Milne, Arithmetic Duality Theorems, Perspectives in Mathematics, Academic Press, New York (1986).
  • [28] \nameA. Selberg, Discontinuous groups and harmonic analysis, Proc. Internat. Congr. Mathematicians (Stockholm, 1962), Inst. Mittag-Leffler, Djursholm (1963), 177–189.
  • [29] \nameY. Tian, Euler systems of CM points on Shimura curves, Ph.D. Thesis, Columbia University (2003).
  • [30] \nameJ. L. Waldspurger, Sur les coefficients de Fourier des forms modulaires de poids demi-entier, J. Math. Pure Appl60 (1981), 375–484.
  • [31] \nameA. Weil, Basic Number Theory, Springer-Verlag, New York (1995).
  • [32] \bibline, Jacobi sums as “Grossencharaktere”, Trans. Amer. Math. Soc73 (1952), 487–495.
  • [33] \nameS. Zhang, Heights of Heegner points on Shimura curves, Ann. of Math153 (2001), 23–147.
  • [34] \bibline, Gross-Zagier formula for GL(2){\mathrm{GL}}(2). II, in Heegner Points and Rankin LL-series(H. Darmon and S. Zhang, eds.), 191–214, Math. Sci. Res. Inst. Publ49, Cambridge Univ. Press, Cambridge, 2004. \Endrefs