Introduction
Converse theorems characterize automorphic representations through the analytic properties of their twisted L L L -functions. In their usual formulation, the functional equations involve specified global ϵ \epsilon ϵ -factors, assembled from the corresponding local factors. In this paper, we allow these functional equations to hold up to nonzero scalar factors depending on the twisting character. Our hypothesis is that the scalar factors remain unchanged under unramified twists. We prove that they are then independent of the twisting character, and deduce automorphy with the prescribed unramified local components. This extends the work of Booker [3 ] to global fields.
Theorem 1.1. Let F F F be a global field, i.e., a number field or a function field over a finite field. Let A F \mathbb{A}_F A F denote the ring of adèles of F F F and let π = ⨂ v π v \pi= \bigotimes_v \pi_v π = ⨂ v π v be an irreducible, admissible generic representation of G L 2 ( A F ) GL_2(\mathbb{A}_F) G L 2 ( A F ) with automorphic central character χ π \chi_\pi χ π . For every idèle class character χ \chi χ of F F F which is unramified at all finite places v v v where π v \pi_v π v is ramified, assume that the products Λ ( s , π × χ ) = ∏ v L ( s , π v × χ v ) \Lambda(s,\pi\times\chi) = \prod_v L(s,\pi_v \times\chi_v) Λ ( s , π × χ ) = ∏ v L ( s , π v × χ v ) and Λ ( s , π ~ × χ − 1 ) = ∏ v L ( s , π ~ v × χ v − 1 ) \Lambda(s,\widetilde{\pi} \times\chi^{-1}) = \prod_v L(s,\widetilde{\pi}_v \times\chi_v^{-1}) Λ ( s , π × χ − 1 ) = ∏ v L ( s , π v × χ v − 1 ) converge absolutely in some right half plane, continue to entire functions of finite order, and satisfy the functional equation
Λ ( s , π × χ ) = ϵ χ ϵ ( s , π × χ ) Λ ( 1 − s , π ~ × χ − 1 ) , (1) \Lambda(s,\pi\times\chi) = \epsilon_\chi\epsilon(s,\pi\times\chi)\Lambda(1-s,\widetilde{\pi} \times\chi^{-1}), \tag*{(1)} Λ ( s , π × χ ) = ϵ χ ϵ ( s , π × χ ) Λ ( 1 − s , π × χ − 1 ) , (1) where ϵ ( s , π × χ ) \epsilon(s,\pi\times\chi) ϵ ( s , π × χ ) is the product of the corresponding local ϵ \epsilon ϵ -factors, defined in [8 ] with respect to a fixed nontrivial additive character ψ = ⨂ v ψ v \psi= \bigotimes_v \psi_v ψ = ⨂ v ψ v of F \ A F F\backslash\mathbb{A}_F F \ A F , and ϵ χ ∈ C × \epsilon_\chi\in\mathbb{C}^{\times} ϵ χ ∈ C × is independent of s s s and satisfies ϵ χ ω = ϵ χ \epsilon_{\chi\omega} = \epsilon_\chi ϵ χ ω = ϵ χ for every unramified idèle class character ω \omega ω in the sense of Definition 2.1. Then there exists a unique isobaric automorphic representation Π = ⨂ v Π v \Pi= \bigotimes_v \Pi_v Π = ⨂ v Π v of G L 2 ( A F ) GL_2(\mathbb{A}_F) G L 2 ( A F ) such that π v ≅ Π v \pi_v \cong\Pi_v π v ≅ Π v for all non-archimedean places v v v where π v \pi_v π v is unramified.
Remarks.
The existence of twisted functional equations and the identification of their precise ϵ \epsilon ϵ -factors are distinct issues in applications of converse theorems. For example, in the function field setting, Grothendieck’s theory supplies functional equations for the twists of the L L L -function of an elliptic curve, while applying Weil’s converse theorem
requires a further comparison with its prescribed functional equations; see [15 §3.2] . Our theorem replaces the identification of each scalar multiplier by a compatibility condition under unramified twists.
(2) If we define the equivalence relation χ ∼ χ ′ ↔ χ ′ χ − 1 \chi\sim\chi' \leftrightarrow\chi'\chi^{-1} χ ∼ χ ′ ↔ χ ′ χ − 1 is unramified, then we may unambiguously talk about the number ϵ [ χ ] \epsilon_{[\chi]} ϵ [ χ ] associated with the equivalence class [ χ ] [\chi] [ χ ] .
(3) If ϵ χ = c ≠ 0 \epsilon_\chi= c \ne0 ϵ χ = c = 0 is independent of χ \chi χ , the conclusion follows from a slight modification of the proof of the converse theorem in [10 §2] . We explain this modification, together with the passage to the unique isobaric representation with the prescribed unramified components, in Lemma 4.8. A variant allowing a fixed nonzero scalar in the functional equations is also discussed for G L 3 \mathrm{GL}_3 GL 3 in [9 Theorem 13.7] . It would be interesting to extend Theorem 1.1 to G L 3 \mathrm{GL}_3 GL 3 by showing that invariance under unramified twists again forces the multipliers to be constant.
(4) The principal step is to show that the multipliers ϵ χ \epsilon_\chi ϵ χ are independent of χ \chi χ . We normalize the auxiliary dual function introduced below so that the corresponding multipliers become ϵ χ / ϵ 1 \epsilon_\chi/\epsilon_1 ϵ χ / ϵ 1 . Starting with the case χ ∼ 1 \chi\sim1 χ ∼ 1 , we use Mellin inversion and finite Fourier analysis to show that every normalized multiplier equals 1 1 1 . Thus the original multipliers are all equal to ϵ 1 \epsilon_1 ϵ 1 .
Our proof draws on Weil’s use of character twists and Mellin inversion [13 Theorem 4.3.15] and on Booker’s approach to recovering root numbers [3 ] . Over a general global field, additional arguments are needed to handle divisor classes and the possible vanishing of Whittaker coefficients. These issues are addressed through a sequence of lemmas developed in the course of the proof. We now outline the method.
Suppose
f ( z ) = ∑ n ≥ 1 a n e 2 π i n z . f(z) = \sum_{n \ge1} a_n e^{2\pi i n z}. f ( z ) = n ≥ 1 ∑ a n e 2 π in z .
where the coefficients a n a_n a n have at most polynomial growth. Thus f f f is holomorphic on the upper half plane H \mathbb{H} H , and the series converges absolutely and uniformly on compact subsets of H \mathbb{H} H . Fix integers N ≥ 1 N \ge1 N ≥ 1 and even k ≥ 2 k \ge2 k ≥ 2 . Consider the completed L L L -function
Λ N ( s , f ) = N s / 2 ( 2 π ) − s Γ ( s ) ∑ n ≥ 1 a n n − s . \Lambda_N(s,f) = N^{s/2}(2\pi)^{-s}\Gamma(s)\sum_{n \ge1} a_n n^{-s}. Λ N ( s , f ) = N s /2 ( 2 π ) − s Γ ( s ) n ≥ 1 ∑ a n n − s .
Hecke’s Mellin inversion argument shows that Λ N ( s , f ) \Lambda_N(s,f) Λ N ( s , f ) extends to an entire function, bounded in vertical strips, satisfying
Λ N ( s , f ) = i k Λ N ( k − s , f ) , \Lambda_N(s,f) = i^k\Lambda_N(k-s,f), Λ N ( s , f ) = i k Λ N ( k − s , f ) ,
if and only if
f ( − 1 N z ) = N k / 2 z k f ( z ) . f\left(-\frac{1}{Nz}\right) = N^{k/2}z^k f(z). f ( − N z 1 ) = N k /2 z k f ( z ) .
Thus a functional equation yields a transformation law for f f f . Applying the same principle to character twists of f f f is the starting point of the classical converse-theorem method. The twist by a Dirichlet character χ \chi χ is defined as
f χ ( z ) = ∑ n ≥ 1 a n χ ( n ) e 2 π i n z . f_\chi(z) = \sum_{n \ge1} a_n\chi(n)e^{2\pi i n z}. f χ ( z ) = n ≥ 1 ∑ a n χ ( n ) e 2 π in z .
For a primitive character χ \chi χ modulo q q q , this can be expressed as an average of translates,
f χ ( z ) = 1 τ ( χ ‾ ) ∑ a m o d q χ ‾ ( a ) f ( z + a q ) , τ ( χ ‾ ) = ∑ a m o d q χ ‾ ( a ) e 2 π i a / q . f_{\chi}(z)=\frac{1}{\tau(\overline{\chi})}\sum_{a\bmod q}\overline{\chi}(a)f\left(z+\frac{a}{q}\right),\qquad\tau(\overline{\chi})=\sum_{a\bmod q}\overline{\chi}(a)e^{2\pi ia/q}. f χ ( z ) = τ ( χ ) 1 a mod q ∑ χ ( a ) f ( z + q a ) , τ ( χ ) = a mod q ∑ χ ( a ) e 2 π ia / q .
We use an analogous construction for Whittaker sums on G L 2 ( A F ) \mathrm{GL}_{2}(\mathbb{A}_{F}) GL 2 ( A F ) . For π = ⨂ v π v \pi=\bigotimes_v\pi_v π = ⨂ v π v as in Theorem 1.1 , let n \mathfrak n n be its conductor. Retain the global additive character ψ = ⨂ v ψ v \psi=\bigotimes_v\psi_v ψ = ⨂ v ψ v fixed above. At each non-archimedean place, choose d v ∈ F × d_v\in F^\times d v ∈ F × such that the conductor of ψ v \psi_v ψ v is d v − 1 o v d_v^{-1}\mathfrak{o}_v d v − 1 o v . Throughout the paper, let S 0 S_0 S 0 denote the set of archimedean places of F F F when F F F is a number field. When F F F is a function field, fix a place v ∞ v_\infty v ∞ at which π v ∞ \pi_{v_\infty} π v ∞ is unramified and ψ v ∞ \psi_{v_\infty} ψ v ∞ has conductor o v ∞ \mathfrak{o}_{v_\infty} o v ∞ , and put S 0 = { v ∞ } S_0=\{v_\infty\} S 0 = { v ∞ } . Take d v = 1 d_v=1 d v = 1 at the places in S 0 S_0 S 0 , and put
d = ( d v ) v , d = div S 0 ( d ) . \mathbf{d}=(d_v)_v,\qquad\mathfrak d=\operatorname{div}^{S_0}(\mathbf{d}). d = ( d v ) v , d = div S 0 ( d ) .
Let ξ ↦ W ξ \xi\mapsto W_\xi ξ ↦ W ξ denote the Whittaker model of π \pi π with respect to ψ \psi ψ . See § 2 for details. Write
a ( y ) = ( y 0 0 1 ) , u ( x ) = ( 1 x 0 1 ) . a(y)=\begin{pmatrix}y&0\\0&1\end{pmatrix},\qquad u(x)=\begin{pmatrix}1&x\\0&1\end{pmatrix}. a ( y ) = ( y 0 0 1 ) , u ( x ) = ( 1 0 x 1 ) .
To a pure tensor ξ = ⨂ v ξ v \xi=\bigotimes_v\xi_v ξ = ⨂ v ξ v , associate the Whittaker sum
ϕ ξ ( g ) = ∑ γ ∈ F × W ξ ( a ( γ ) g ) . \phi_\xi(g)=\sum_{\gamma\in F^\times}W_\xi(a(\gamma)g). ϕ ξ ( g ) = γ ∈ F × ∑ W ξ ( a ( γ ) g ) .
We fix the normalized essential vectors ξ v = ξ v 0 \xi_v=\xi_v^0 ξ v = ξ v 0 for v ∉ S 0 v\notin S_0 v ∈ / S 0 , with
W ξ v 0 ( a ( d v − 1 ) ) = 1 , W_{\xi_v^0}(a(d_v^{-1}))=1, W ξ v 0 ( a ( d v − 1 )) = 1 ,
and allow ξ v \xi_v ξ v to vary over the smooth vectors for v ∈ S 0 v\in S_0 v ∈ S 0 . With the vectors outside S 0 S_0 S 0 fixed, we write ϕ ξ S 0 \phi_{\xi_{S_0}} ϕ ξ S 0 for ϕ ξ \phi_\xi ϕ ξ , where ξ S 0 = ⨂ v ∈ S 0 ξ v \xi_{S_0}=\bigotimes_{v\in S_0}\xi_v ξ S 0 = ⨂ v ∈ S 0 ξ v .
We call an effective divisor m \mathfrak m m admissible if its support is disjoint from S 0 ∪ Supp ( n ) S_0\cup\operatorname{Supp}(\mathfrak n) S 0 ∪ Supp ( n ) . For such a modulus, put
X ( m ) = { unitary id e ˋ le class characters ω : f ω S 0 ∣ m } , X(\mathfrak m)=\{\text{unitary idèle class characters }\omega:\mathfrak f_\omega^{S_0}\mid\mathfrak m\}, X ( m ) = { unitary id e ˋ le class characters ω : f ω S 0 ∣ m } ,
where f ω S 0 \mathfrak f_\omega^{S_0} f ω S 0 denotes the part of the conductor of ω \omega ω supported outside S 0 S_0 S 0 . No restriction is imposed at the places in S 0 S_0 S 0 .
For an admissible modulus m = ∏ v ∈ S p v n v \mathfrak m=\prod_{v\in S}\mathfrak p_v^{n_v} m = ∏ v ∈ S p v n v , put w v = ϖ v − n v w_v=\varpi_v^{-n_v} w v = ϖ v − n v for v ∈ S v\in S v ∈ S and w v = 0 w_v=0 w v = 0 otherwise. For ω ∈ X ( m ) \omega\in X(\mathfrak m) ω ∈ X ( m ) , define the twist
ϕ ω , m ( g ) = ∫ ∏ v ∈ S o v × ϕ ξ ( g u ( w u ) ) ω S ( u ) d × u , \phi_{\omega,\mathfrak m}(g)=\int_{\prod_{v\in S}\mathfrak{o}_v^\times}\phi_{\xi}(g\,u(wu))\omega_S(u)\,d^\times u, ϕ ω , m ( g ) = ∫ ∏ v ∈ S o v × ϕ ξ ( g u ( w u )) ω S ( u ) d × u ,
where ω S ( u ) = ∏ v ∈ S ω v ( u v ) \omega_S(u)=\prod_{v\in S}\omega_v(u_v) ω S ( u ) = ∏ v ∈ S ω v ( u v ) and the measure is normalized so that each o v × \mathfrak{o}_v^\times o v × has unit volume. The local Mellin calculation then introduces the factor
∫ o v × ω v ( u ) χ v ( u ) − 1 d × u = { 1 , ω v ∣ o v × = χ v ∣ o v × , 0 , otherwise . \int_{\mathfrak{o}_v^\times}\omega_v(u)\chi_v(u)^{-1}\,d^\times u=
\begin{cases}
1,&\omega_v|_{\mathfrak{o}_v^\times}=\chi_v|_{\mathfrak{o}_v^\times},\\
0,&\text{otherwise}.
\end{cases} ∫ o v × ω v ( u ) χ v ( u ) − 1 d × u = { 1 , 0 , ω v ∣ o v × = χ v ∣ o v × , otherwise .
Together with the essential-vector calculation outside S ∪ S 0 S\cup S_0 S ∪ S 0 , this isolates precisely the characters χ ∼ ω \chi\sim\omega χ ∼ ω (cf. Lemma 3.5). The local integrals arising from these twists and their dual counterparts are not directly related by the usual local functional equation. For v ∈ S v\in S v ∈ S , we introduce normalized polynomials D m ( s , π v × χ v ) D_{\mathfrak m}(s,\pi_v\times\chi_v) D m ( s , π v × χ v ) and establish a functional equation for them (Lemma 3.8). This supplies the local identities needed to deduce, by Mellin inversion, identities between the twisted Whittaker sums from the assumed global functional equations. After normalizing the scalar factors by ϵ [ 1 ] \epsilon[1] ϵ [ 1 ] , write ϵ [ ω ] \epsilon[\omega] ϵ [ ω ] for the resulting factors, so that ϵ [ 1 ] = 1 \epsilon[1] = 1 ϵ [ 1 ] = 1 . Mellin inversion of the functional equations, followed by summation over X ( m ) / X ( 1 ) X(\mathfrak{m})/X(1) X ( m ) / X ( 1 ) , leads to a finite Fourier transform. More precisely, put
R m = ∏ v ∈ S o v / p v n v , K m = R m × , Ψ m ( x ) = ∏ v ∈ S ψ v ( d v − 1 ϖ ‾ v − n v x v ) . R_{\mathfrak{m}} = \prod_{v\in S} \mathfrak{o}_v/\mathfrak{p}_v^{n_v}, \qquad K_{\mathfrak{m}} = R_{\mathfrak{m}}^{\times}, \qquad\Psi_{\mathfrak{m}}(x) = \prod_{v\in S} \psi_v(d_v^{-1}\overline{\varpi}_v^{-n_v}x_v). R m = v ∈ S ∏ o v / p v n v , K m = R m × , Ψ m ( x ) = v ∈ S ∏ ψ v ( d v − 1 ϖ v − n v x v ) .
Restriction to local units identifies X ( m ) / X ( 1 ) X(\mathfrak{m})/X(1) X ( m ) / X ( 1 ) with K ^ m \widehat{K}_{\mathfrak{m}} K m . Writing ϵ ρ \epsilon_\rho ϵ ρ for the factor corresponding to ρ ∈ K ^ m \rho\in\widehat{K}_{\mathfrak{m}} ρ ∈ K m , define
ϵ ^ m ( u ) = ∑ ρ ∈ K ^ m ϵ ρ ρ − 1 ( u ) , S m ( x ) = 1 ∣ K m ∣ ∑ u ∈ K m ϵ ^ m ( u ) Ψ m ( ( 1 − u ) x ) . \widehat{\epsilon}_{\mathfrak{m}}(u) = \sum_{\rho\in\widehat{K}_{\mathfrak{m}}} \epsilon_\rho\rho^{-1}(u), \qquad S_{\mathfrak{m}}(x) = \frac{1}{|K_{\mathfrak{m}}|}\sum_{u\in K_{\mathfrak{m}}}\widehat{\epsilon}_{\mathfrak{m}}(u)\Psi_{\mathfrak{m}}((1-u)x). ϵ m ( u ) = ρ ∈ K m ∑ ϵ ρ ρ − 1 ( u ) , S m ( x ) = ∣ K m ∣ 1 u ∈ K m ∑ ϵ m ( u ) Ψ m (( 1 − u ) x ) .
The central assertion is (see Proposition 4.6 )
S m ( x ) = 1 ∀ x ∈ R m . S_{\mathfrak{m}}(x) = 1 \qquad\forall x\in R_{\mathfrak{m}}. S m ( x ) = 1 ∀ x ∈ R m .
Fix an effective divisor H \mathfrak{H} H , supported outside S 0 S_0 S 0 and coprime to n \mathfrak{n} n , and let h h h be its representing idèle, with components 1 1 1 at S 0 S_0 S 0 . We first treat admissible moduli m \mathfrak{m} m coprime to H \mathfrak{H} H with trivial ray class modulo n H \mathfrak{n}\mathfrak{H} nH . Comparing two Whittaker expansions and varying the smooth vectors at S 0 S_0 S 0 to isolate individual terms, we obtain
c ~ S 0 ( γ n h f m ) ( S m ( d γ n h f m ) − 1 ) = 0 \widetilde{c}^{S_0}(\gamma n h f_{\mathfrak{m}})\left(S_{\mathfrak{m}}(d\gamma n h f_{\mathfrak{m}})-1\right)=0 c S 0 ( γ nh f m ) ( S m ( d γ nh f m ) − 1 ) = 0
whenever d γ n h f m d\gamma n h f_{\mathfrak{m}} d γ nh f m is integral outside S 0 S_0 S 0 . Here c ~ S 0 \widetilde{c}^{S_0} c S 0 denotes the product of the dual Whittaker coefficients outside S 0 S_0 S 0 . Thus the desired identity holds when the accompanying coefficient is nonzero.
To pass to arbitrary admissible moduli, we first use the functional equations for an auxiliary character twist to obtain an effective divisor D \mathfrak{D} D satisfying
[ D ] = [ n d 2 ] in Cl ( o S 0 ) , ( D , m ) = 1 , c ~ S 0 ( D d − 1 ) ≠ 0. [\mathfrak{D}] = [\mathfrak{n}\mathfrak{d}^{2}] \quad\text{in } \operatorname{Cl}(\mathfrak{o}^{S_0}), \qquad(\mathfrak{D},\mathfrak{m})=1, \qquad\widetilde{c}^{S_0}(\mathfrak{D}\mathfrak{d}^{-1})\ne0. [ D ] = [ n d 2 ] in Cl ( o S 0 ) , ( D , m ) = 1 , c S 0 ( D d − 1 ) = 0.
We then choose an effective divisor H \mathfrak{H} H such that
[ H ] = [ d ] , ( H , n m D ) = 1. [\mathfrak{H}] = [\mathfrak{d}], \qquad(\mathfrak{H},\mathfrak{n}\mathfrak{m}\mathfrak{D})=1. [ H ] = [ d ] , ( H , nmD ) = 1.
In particular,
[ d n H ] = [ D ] . [\mathfrak{d}\mathfrak{n}\mathfrak{H}] = [\mathfrak{D}]. [ dnH ] = [ D ] .
Given x ∈ R m x\in R_{\mathfrak{m}} x ∈ R m , the Chinese remainder theorem and the local Hecke relations then allow us to construct a multiple M \mathfrak{M} M of m \mathfrak{m} m with trivial ray class modulo n H \mathfrak{n}\mathfrak{H} nH , together with γ ∈ F × \gamma\in F^\times γ ∈ F × such that
c ~ S 0 ( γ n h f M ) ≠ 0. \widetilde{c}^{S_0}(\gamma n h f_{\mathfrak{M}})\ne0. c S 0 ( γ nh f M ) = 0.
The construction also ensures that
y = d γ n h f m y=d\gamma n h f_{\mathfrak{m}} y = d γ nh f m
is integral outside S 0 S_0 S 0 and represents x x x in R m R_{\mathfrak{m}} R m . The comparison identity therefore gives
S M ( d γ n h f M ) = 1. S_{\mathfrak{M}}(d\gamma n h f_{\mathfrak{M}})=1. S M ( d γ nh f M ) = 1.
The compatibility relation
S M ( f M f m − 1 y ) = S m ( y ) , S_{\mathfrak{M}}(f_{\mathfrak{M}}f_{\mathfrak{m}}^{-1}y)=S_{\mathfrak{m}}(y), S M ( f M f m − 1 y ) = S m ( y ) ,
valid when y y y is integral at every place dividing M \mathfrak{M} M , transfers the resulting identity back to the prescribed residue class x x x . This proves S m ( x ) = 1 S_{\mathfrak{m}}(x)=1 S m ( x ) = 1 for every x ∈ R m x\in R_{\mathfrak{m}} x ∈ R m and every admissible modulus m \mathfrak{m} m .
Finally, finite Fourier analysis shows that ϵ [ ω ] = 1 \epsilon_{[\omega]}=1 ϵ [ ω ] = 1 for every ω ∈ X ( m ) \omega\in X(\mathfrak{m}) ω ∈ X ( m ) and every admissible modulus m \mathfrak{m} m . Undoing the normalization, the original scalar factors are independent of the twisting character. The constant-scalar variant of the converse theorem (see Lemma 4.8 ) then yields the required isobaric automorphic representation.
Acknowledgements. The function field case of this work formed part of the first author’s Ph.D. thesis [1 ] at the University of Iowa. The present paper develops that work further and treats both number fields and function fields. We thank Andrew Booker for many helpful discussions and for explaining his work [3 ] . We are also grateful to Min Lee for her support and interest in this project.
Preliminaries
Let F F F be a global field, either a number field or a function field of one variable over its field of constants F q \mathbb{F}_q F q . If F F F is a number field, let o F \mathfrak{o}_F o F denote its ring of integers. For each place v v v of F F F , let F v F_v F v denote the completion of F F F at v v v . For a nonarchimedean place v v v , let o v \mathfrak{o}_v o v denote the ring of integers of F v F_v F v , p v \mathfrak{p}_v p v its maximal ideal, and q v q_v q v the cardinality of o v / p v \mathfrak{o}_v/\mathfrak{p}_v o v / p v . Fix a uniformizer ϖ v \varpi_v ϖ v , and normalize the absolute value by ∥ ϖ v ∥ v = q v − 1 \lVert\varpi_v\rVert_v=q_v^{-1} ∥ ϖ v ∥ v = q v − 1 .
Let D F \mathcal{D}_F D F denote the divisor group of F F F , the free abelian group on its nonarchimedean places. Its elements are formal sums ∑ v < ∞ n v v \sum_{v<\infty}n_vv ∑ v < ∞ n v v , where n v ∈ Z n_v\in\mathbb{Z} n v ∈ Z and n v = 0 n_v=0 n v = 0 for almost all v v v . We also write such a divisor multiplicatively as ∏ v < ∞ p v n v \prod_{v<\infty}\mathfrak{p}_v^{n_v} ∏ v < ∞ p v n v , and refer to p v \mathfrak{p}_v p v as a prime divisor. A divisor is called positive if n v ≥ 0 n_v\geq0 n v ≥ 0 for every v v v . We write a ⪰ 0 \mathfrak{a}\succeq0 a ⪰ 0 for a positive divisor a \mathfrak{a} a , and denote the monoid of positive divisors by D F + \mathcal{D}_F^+ D F + . The zero divisor is written 1 1 1 in multiplicative notation.
When F F F is a number field, D F \mathcal{D}_F D F is naturally identified with the group of nonzero fractional ideals of o F \mathfrak{o}_F o F , and D F + \mathcal{D}_F^+ D F + with the monoid of nonzero integral ideals. When F F F is a function field, write q v = q d v q_v=q^{d_v} q v = q d v and call d v d_v d v the degree of v v v . The degree of ∏ v < ∞ p v n v \prod_{v<\infty}\mathfrak{p}_v^{n_v} ∏ v < ∞ p v n v is then ∑ v n v d v \sum_v n_vd_v ∑ v n v d v . Each idèle x = ( x v ) v ∈ A F × x=(x_v)_v\in\mathbb{A}_F^\times x = ( x v ) v ∈ A F × determines a divisor
div ( x ) = ∏ v < ∞ p v ord v ( x v ) . \operatorname{div}(x)=\prod_{v<\infty}\mathfrak{p}_v^{\operatorname{ord}_v(x_v)}. div ( x ) = v < ∞ ∏ p v ord v ( x v ) .
Define N ( p v ) = q v N(\mathfrak{p}_v)=q_v N ( p v ) = q v and extend this multiplicatively, so that
N ( a ) = ∏ v < ∞ q v n v ( a = ∏ v < ∞ p v n v ) . N(\mathfrak{a})=\prod_{v<\infty}q_v^{n_v}\qquad\left(\mathfrak{a}=\prod_{v<\infty}\mathfrak{p}_v^{n_v}\right). N ( a ) = v < ∞ ∏ q v n v ( a = v < ∞ ∏ p v n v ) .
Let S ∞ S_\infty S ∞ denote the set of archimedean places of F F F . If F F F is a number field, put S 0 = S ∞ S_0=S_\infty S 0 = S ∞ . If F F F is a function field, choose a place v ∞ v_\infty v ∞ at which the representation π \pi π of Theorem 1.1 is unramified, and put S 0 = { v ∞ } S_0=\{v_\infty\} S 0 = { v ∞ } . In this case S ∞ S_\infty S ∞ is empty. For any finite set of places S S S containing S 0 S_0 S 0 , let
o F S = { x ∈ F : ord v ( x ) ≥ 0 for every v ∉ S } \mathfrak{o}_F^S=\{x\in F:\operatorname{ord}_v(x)\geq0\text{ for every }v\notin S\} o F S = { x ∈ F : ord v ( x ) ≥ 0 for every v ∈ / S }
denote the ring of S S S -integers in F F F . In particular, o F S 0 = o F \mathfrak{o}_F^{S_0}=\mathfrak{o}_F o F S 0 = o F in the number-field case; in the function-field case, its elements may have poles only at v ∞ v_\infty v ∞ . Put A S = o F S A_S=\mathfrak{o}_F^S A S = o F S , and let D F S \mathcal{D}_F^S D F S denote the group of divisors supported outside S S S . For a divisor m \mathfrak{m} m , we write m S \mathfrak{m}^{S} m S to denote
m S = ∏ v ∉ S p v ord v ( m ) . \mathfrak{m}^{S}=\prod_{v\notin S}\mathfrak{p}_{v}^{\operatorname{ord}_{v}(\mathfrak{m})}. m S = v ∈ / S ∏ p v ord v ( m ) .
Define
Cl ( A S ) = D F S / { div S ( x ) : x ∈ F × } . \operatorname{Cl}(A_{S})=\mathcal{D}_{F}^{S}/\{\operatorname{div}^{S}(x):x\in F^{\times}\}. Cl ( A S ) = D F S / { div S ( x ) : x ∈ F × } .
We denote the class of a \mathfrak{a} a by [ a ] [\mathfrak{a}] [ a ] .
For an effective divisor b = ∏ v ∉ S p v b v \mathfrak{b}=\prod_{v\notin S}\mathfrak{p}_{v}^{b_{v}} b = ∏ v ∈ / S p v b v , let D F S ( b ) \mathcal{D}_{F}^{S}(\mathfrak{b}) D F S ( b ) be the subgroup of divisors supported outside S ∪ Supp ( b ) S\cup\operatorname{Supp}(\mathfrak{b}) S ∪ Supp ( b ) , and put
P b ( A S ) = { div S ( x ) : x ∈ F × , x − 1 ∈ p v b v o v for every v ∣ b } . P_{\mathfrak{b}}(A_{S})=\{\operatorname{div}^{S}(x):x\in F^{\times},\ x-1\in\mathfrak{p}_{v}^{b_{v}}\mathfrak{o}_{v}\text{ for every }v\mid\mathfrak{b}\}. P b ( A S ) = { div S ( x ) : x ∈ F × , x − 1 ∈ p v b v o v for every v ∣ b } .
The ray class group modulo b \mathfrak{b} b is
Cl b ( A S ) = D F S ( b ) / P b ( A S ) . \operatorname{Cl}_{\mathfrak{b}}(A_{S})=\mathcal{D}_{F}^{S}(\mathfrak{b})/P_{\mathfrak{b}}(A_{S}). Cl b ( A S ) = D F S ( b ) / P b ( A S ) .
No sign conditions are imposed at real places. We denote the ray class of a \mathfrak{a} a by [ a ] b [\mathfrak{a}]_{\mathfrak{b}} [ a ] b . Both groups are finite. When S = S 0 S=S_{0} S = S 0 , we write simply A = o F S 0 A=\mathfrak{o}_{F}^{S_{0}} A = o F S 0 .
For each nonarchimedean place v v v and integer n ≥ 0 n\geq0 n ≥ 0 , put
U v ( n ) = { o v × , n = 0 , 1 + p v n , n > 0. U_{v}^{(n)}=
\begin{cases}
\mathfrak{o}_{v}^{\times}, & n=0,\\
1+\mathfrak{p}_{v}^{n}, & n>0.
\end{cases} U v ( n ) = { o v × , 1 + p v n , n = 0 , n > 0.
Let χ = ⨂ v χ v \chi=\bigotimes_{v}\chi_{v} χ = ⨂ v χ v be a continuous quasi-character of F × \ A F × F^{\times}\backslash\mathbb{A}_{F}^{\times} F × \ A F × , which we call an idèle class character. For each nonarchimedean place v v v , let a ( χ v ) a(\chi_{v}) a ( χ v ) be the least integer n ≥ 0 n\geq0 n ≥ 0 such that χ v \chi_{v} χ v is trivial on U v ( n ) U_{v}^{(n)} U v ( n ) . The conductor of χ \chi χ is the positive divisor
f χ = ∏ v < ∞ p v a ( χ v ) . \mathfrak{f}_{\chi}=\prod_{v<\infty}\mathfrak{p}_{v}^{a(\chi_{v})}. f χ = v < ∞ ∏ p v a ( χ v ) .
We represent this divisor by the idèle f χ \mathfrak{f}_{\chi} f χ whose component at a nonarchimedean place v v v is ϖ v a ( χ v ) \varpi_{v}^{a(\chi_{v})} ϖ v a ( χ v ) and whose archimedean components are 1 1 1 .
If v ∤ f χ v\nmid\mathfrak{f}_{\chi} v ∤ f χ , the value
χ ( p v ) : = χ v ( ϖ v ) \chi(\mathfrak{p}_{v}):=\chi_{v}(\varpi_{v}) χ ( p v ) := χ v ( ϖ v )
is independent of the choice of uniformizer. These values extend multiplicatively to a character of the group of divisors supported away from f χ \mathfrak{f}_{\chi} f χ . We also denote this character by χ \chi χ , and extend it to a function on D F + \mathcal{D}_{F}^{+} D F + by setting χ ( a ) = 0 \chi(\mathfrak{a})=0 χ ( a ) = 0 whenever a \mathfrak{a} a and f χ \mathfrak{f}_{\chi} f χ have a common prime divisor.
Definition 2.1. An idèle class character χ \chi χ is said to be unramified if χ v \chi_{v} χ v is trivial on o v × \mathfrak{o}_{v}^{\times} o v × for every v ∉ S 0 v\notin S_{0} v ∈ / S 0 .
Remark. Thus “unramified” means unramified outside S 0 S_{0} S 0 . In the function field case, ramification at v ∞ v_{\infty} v ∞ is permitted.
For a positive divisor f \mathfrak{f} f supported outside S 0 S_{0} S 0 , put
X ( f ) = { unitary id e ˋ le class characters ω : f ω S 0 ∣ f } , f ω S 0 = ∏ v ∉ S 0 p v a ( ω v ) . X(\mathfrak{f})=\{\text{unitary idèle class characters }\omega:\mathfrak{f}_{\omega}^{S_{0}}\mid\mathfrak{f}\},\qquad
\mathfrak{f}_{\omega}^{S_{0}}=\prod_{v\notin S_{0}}\mathfrak{p}_{v}^{a(\omega_{v})}. X ( f ) = { unitary id e ˋ le class characters ω : f ω S 0 ∣ f } , f ω S 0 = v ∈ / S 0 ∏ p v a ( ω v ) .
No local restriction is imposed at the places in S 0 S_{0} S 0 . In particular, X ( 1 ) X(1) X ( 1 ) is the group of unitary idèle class characters unramified outside S 0 S_{0} S 0 .
We next fix notation for subgroups of G L 2 \mathrm{GL}_{2} GL 2 . Let B = T U B=TU B = T U be the Borel subgroup of upper triangular matrices, where T T T is the diagonal torus and U U U is the upper unipotent subgroup. Let B 1 ⊂ B B_{1}\subset B B 1 ⊂ B be the mirabolic subgroup consisting of matrices whose lower-right entry is $1.
For each place v v v , put G v = G L 2 ( F v ) G_v=\mathrm{GL}_2(F_v) G v = GL 2 ( F v ) . For a finite set S S S of places, write
F S = ∏ v ∈ S F v , G S = ∏ v ∈ S G v . F_S=\prod_{v\in S}F_v,\qquad G_S=\prod_{v\in S}G_v. F S = v ∈ S ∏ F v , G S = v ∈ S ∏ G v .
and let A F S A_F^S A F S denote the restricted product ∏ v ∉ S ′ F v \prod_{v\notin S}'F_v ∏ v ∈ / S ′ F v . Set
G S = G L 2 ( A F S ) = ∏ v ∉ S ′ G v . G^S=\mathrm{GL}_2(A_F^S)=\prod_{v\notin S}'G_v. G S = GL 2 ( A F S ) = v ∈ / S ∏ ′ G v .
Then
G L 2 ( A F ) = G S × G S . \mathrm{GL}_2(A_F)=G_S\times G^S. GL 2 ( A F ) = G S × G S .
We identify G S G_S G S with its image under the embedding g S ↦ ( g S , 1 ) g_S\mapsto(g_S,1) g S ↦ ( g S , 1 ) . For a function f f f on G ( A F ) G(A_F) G ( A F ) and g S ∈ G S g_S\in G_S g S ∈ G S , we write f ( g S ) f(g_S) f ( g S ) for f ( ( g S , 1 ) ) f((g_S,1)) f (( g S , 1 )) . Conversely, if f S f_S f S is a function on G S G_S G S and g ∈ G ( A F ) g\in G(A_F) g ∈ G ( A F ) , we write f S ( g ) f_S(g) f S ( g ) for f S ( g S ) f_S(g_S) f S ( g S ) , where g S g_S g S is the projection of g g g to G S G_S G S .
When F F F is a number field, write
F ∞ = ∏ v ∣ ∞ F v , G ∞ = ∏ v ∣ ∞ G v . F_\infty=\prod_{v\mid\infty}F_v,\qquad G_\infty=\prod_{v\mid\infty}G_v. F ∞ = v ∣ ∞ ∏ F v , G ∞ = v ∣ ∞ ∏ G v .
and let K ∞ = ∏ v ∣ ∞ K v K_\infty=\prod_{v\mid\infty}K_v K ∞ = ∏ v ∣ ∞ K v , where each K v K_v K v is a fixed maximal compact subgroup of G v G_v G v . Let g ∞ \mathfrak{g}_\infty g ∞ be the real Lie algebra of G ∞ G_\infty G ∞ , and let U ( g ∞ , C ) U(\mathfrak{g}_\infty,\mathbb{C}) U ( g ∞ , C ) denote the universal enveloping algebra of its complexification. In the function field case, G ∞ G_\infty G ∞ and K ∞ K_\infty K ∞ are understood to be trivial. In particular, G S 0 = G ∞ G_{S_0}=G_\infty G S 0 = G ∞ in the number field case, whereas G S 0 = G v ∞ G_{S_0}=G_{v_\infty} G S 0 = G v ∞ in the function field case.
For each nonarchimedean place v v v , put K v = G L 2 ( o v ) K_v=\mathrm{GL}_2(\mathfrak{o}_v) K v = GL 2 ( o v ) . For an integer m ≥ 1 m\geq1 m ≥ 1 , define
K 1 , v ( p v m ) = { g ∈ G L 2 ( o v ) : g ≡ ( ∗ ∗ 0 1 ) ( m o d p v m ) } , K_{1,v}(\mathfrak{p}_v^m)=\left\{g\in\mathrm{GL}_2(\mathfrak{o}_v):g\equiv
\begin{pmatrix}
* & *\\
0 & 1
\end{pmatrix}\pmod{\mathfrak{p}_v^m}\right\}, K 1 , v ( p v m ) = { g ∈ GL 2 ( o v ) : g ≡ ( ∗ 0 ∗ 1 ) ( mod p v m ) } ,
and
K 0 , v ( p v m ) = { g ∈ G L 2 ( o v ) : g ≡ ( ∗ ∗ 0 ∗ ) ( m o d p v m ) } . K_{0,v}(\mathfrak{p}_v^m)=\left\{g\in\mathrm{GL}_2(\mathfrak{o}_v):g\equiv
\begin{pmatrix}
* & *\\
0 & *
\end{pmatrix}\pmod{\mathfrak{p}_v^m}\right\}. K 0 , v ( p v m ) = { g ∈ GL 2 ( o v ) : g ≡ ( ∗ 0 ∗ ∗ ) ( mod p v m ) } .
For m = 0 m=0 m = 0 , put K 1 , v ( 1 ) = K 0 , v ( 1 ) = K v K_{1,v}(1)=K_{0,v}(1)=K_v K 1 , v ( 1 ) = K 0 , v ( 1 ) = K v . For m ≥ 1 m\geq1 m ≥ 1 , the lower-right entry induces an isomorphism
K 0 , v ( p v m ) / K 1 , v ( p v m ) ≃ ( o v / p v m ) × . K_{0,v}(\mathfrak{p}_v^m)/K_{1,v}(\mathfrak{p}_v^m)\simeq(\mathfrak{o}_v/\mathfrak{p}_v^m)^\times. K 0 , v ( p v m ) / K 1 , v ( p v m ) ≃ ( o v / p v m ) × .
In particular, K 1 , v ( p v m ) K_{1,v}(\mathfrak{p}_v^m) K 1 , v ( p v m ) is normal in K 0 , v ( p v m ) K_{0,v}(\mathfrak{p}_v^m) K 0 , v ( p v m ) .
Let K f = ∏ v < ∞ K v K_f=\prod_{v<\infty}K_v K f = ∏ v < ∞ K v . For a = ∏ v < ∞ p v m v ∈ O F + \mathfrak{a}=\prod_{v<\infty}\mathfrak{p}_v^{m_v}\in\mathcal{O}_F^+ a = ∏ v < ∞ p v m v ∈ O F + , put
K i ( a ) = ∏ v < ∞ K i , v ( p v m v ) , i = 0 , 1. K_i(\mathfrak{a})=\prod_{v<\infty}K_{i,v}(\mathfrak{p}_v^{m_v}),\qquad i=0,1. K i ( a ) = v < ∞ ∏ K i , v ( p v m v ) , i = 0 , 1.
These are compact open subgroups of the group of finite adelic points, with
K 1 ( a ) ⊆ K 0 ( a ) ⊆ K f . K_1(\mathfrak{a})\subseteq K_0(\mathfrak{a})\subseteq K_f. K 1 ( a ) ⊆ K 0 ( a ) ⊆ K f .
We also write
K i S 0 ( a ) = ∏ v ∉ S 0 K i , v ( p v m v ) , i = 0 , 1. K_i^{S_0}(\mathfrak{a})=\prod_{v\notin S_0}K_{i,v}(\mathfrak{p}_v^{m_v}),\qquad i=0,1. K i S 0 ( a ) = v ∈ / S 0 ∏ K i , v ( p v m v ) , i = 0 , 1.
Whittaker models
Throughout this paper, we fix a non-trivial additive character ψ = ⨂ ψ v \psi= \bigotimes\psi_v ψ = ⨂ ψ v of F \ A F F\backslash\mathbb{A}_F F \ A F such that ψ w ( x ) = e − 2 π i x \psi_w(x) = e^{-2\pi i x} ψ w ( x ) = e − 2 π i x when F w = R F_w = \mathbb{R} F w = R , ψ w ( x ) = e − 2 π i ( x + x ˉ ) \psi_w(x) = e^{-2\pi i(x+\bar{x})} ψ w ( x ) = e − 2 π i ( x + x ˉ ) when F w = C F_w = \mathbb{C} F w = C . For each nonarchimedean place v v v , choose d v ∈ F v × d_v \in F_v^\times d v ∈ F v × such that
cond ( ψ v ) = d v − 1 o v \operatorname{cond}(\psi_v) = d_v^{-1}\mathfrak{o}_v cond ( ψ v ) = d v − 1 o v
where the conductor is the largest fractional ideal on which ψ v \psi_v ψ v is trivial. Let δ v = ord v ( d v ) \delta_v = \operatorname{ord}_v(d_v) δ v = ord v ( d v ) . We may take d v = 1 d_v = 1 d v = 1 at almost all places. Put d = ( d v ) v \mathbf{d} = (d_v)_v d = ( d v ) v , taking d v = 1 d_v = 1 d v = 1 at the archimedean places. For a nonarchimedean place v v v , the character
ψ v ∘ ( x ) = ψ v ( d v − 1 x ) \psi_v^\circ(x) = \psi_v(d_v^{-1}x) ψ v ∘ ( x ) = ψ v ( d v − 1 x )
has conductor o v \mathfrak{o}_v o v . We use ψ v ∘ \psi_v^\circ ψ v ∘ for local calculations when convenient, while retaining ψ \psi ψ in all global calculations.
Let π = ⨂ v π v \pi= \bigotimes_v \pi_v π = ⨂ v π v be an irreducible admissible representation of GL 2 ( A F ) \operatorname{GL}_2(\mathbb{A}_F) GL 2 ( A F ) . At an archimedean place v v v , we work with the smooth Fréchet globalization of moderate growth V π v ∞ V_{\pi_v}^{\infty} V π v ∞ of the underlying Harish-Chandra module V π v V_{\pi_v} V π v ; see [2 ] . When discussing smooth vectors and the action of G v G_v G v , we use π v \pi_v π v also to denote this globalization.
We say that π \pi π is generic if each π v \pi_v π v admits a nonzero Whittaker functional with respect to ψ v \psi_v ψ v , required to be continuous at an archimedean place. Equivalently, there is a nonzero G v G_v G v -intertwining map
V π v ⟶ Ind U ( F v ) G v ψ v , ξ v ′ ⟼ W ξ v ′ , V_{\pi_v} \longrightarrow\operatorname{Ind}_{U(F_v)}^{G_v}\psi_v,\qquad\xi'_v \longmapsto W_{\xi'_v}, V π v ⟶ Ind U ( F v ) G v ψ v , ξ v ′ ⟼ W ξ v ′ ,
where the domain is understood to be V π v ∞ V_{\pi_v}^{\infty} V π v ∞ at an archimedean place. Here the induced representation consists of smooth functions satisfying
W ( ( 1 x 0 1 ) g ) = ψ v ( x ) W ( g ) , W\left(\begin{pmatrix}1 & x \\ 0 & 1\end{pmatrix}g\right)=\psi_v(x)W(g), W ( ( 1 0 x 1 ) g ) = ψ v ( x ) W ( g ) ,
with G v G_v G v acting by right translation. The intertwining map is unique up to scalar multiplication, and its image W ( π v , ψ v ) \mathcal{W}(\pi_v,\psi_v) W ( π v , ψ v ) is called the Whittaker model of π v \pi_v π v with respect to ψ v \psi_v ψ v . If ψ v ′ \psi'_v ψ v ′ is another nontrivial additive character of F v F_v F v , there is a unique a ∈ F v × a \in F_v^\times a ∈ F v × such that ψ v ′ ( x ) = ψ v ( a − 1 x ) \psi'_v(x)=\psi_v(a^{-1}x) ψ v ′ ( x ) = ψ v ( a − 1 x ) . For W ∈ W ( π v , ψ v ) W \in\mathcal{W}(\pi_v,\psi_v) W ∈ W ( π v , ψ v ) , put
( L ( a ) W ) ( g ) = W ( ( a − 1 0 0 1 ) g ) . (L(a)W)(g)=W\left(\begin{pmatrix}a^{-1} & 0 \\ 0 & 1\end{pmatrix}g\right). ( L ( a ) W ) ( g ) = W ( ( a − 1 0 0 1 ) g ) .
Then L ( a ) L(a) L ( a ) gives a G v G_v G v -equivariant bijection W ( π v , ψ v ) ⟶ W ( π v , ψ v ′ ) \mathcal{W}(\pi_v,\psi_v)\longrightarrow\mathcal{W}(\pi_v,\psi'_v) W ( π v , ψ v ) ⟶ W ( π v , ψ v ′ ) .
We write
W ( π , ψ ) = ⨂ v W ( π v , ψ v ) \mathcal{W}(\pi,\psi)=\bigotimes_v\mathcal{W}(\pi_v,\psi_v) W ( π , ψ ) = v ⨂ W ( π v , ψ v )
for the associated global Whittaker model. The restricted tensor product is taken with respect to the spherical Whittaker functions normalized to have value 1 1 1 at the identity at almost all nonarchimedean places. For a pure tensor ξ = ⨂ v ξ v \xi=\bigotimes_v\xi_v ξ = ⨂ v ξ v , the corresponding Whittaker function is
W ξ ( g ) = ∏ v W ξ v ( g v ) . W_\xi(g)=\prod_v W_{\xi_v}(g_v). W ξ ( g ) = v ∏ W ξ v ( g v ) .
Similarly, we form W ( π ~ , ψ ) \mathcal{W}(\widetilde{\pi},\psi) W ( π , ψ ) for the contragredient representation π ~ = ⨂ v π ~ v \widetilde{\pi}=\bigotimes_v\widetilde{\pi}_v π = ⨂ v π v .
Fix
w = ( 0 1 − 1 0 ) . w=\begin{pmatrix}0 & 1 \\ -1 & 0\end{pmatrix}. w = ( 0 − 1 1 0 ) .
For W ∈ W ( π v , ψ v ) W \in\mathcal{W}(\pi_v,\psi_v) W ∈ W ( π v , ψ v ) , define
W ~ ( g ) = W ( w t g − 1 ) . \widetilde{W}(g)=W(w^{t}g^{-1}). W ( g ) = W ( w t g − 1 ) .
Then W ↦ W ~ W \mapsto\widetilde{W} W ↦ W is a linear bijection
W ( π v , ψ v ) ⟶ W ( π ~ v , ψ v ) . \mathcal{W}(\pi_v,\psi_v)\longrightarrow\mathcal{W}(\widetilde{\pi}_v,\psi_v). W ( π v , ψ v ) ⟶ W ( π v , ψ v ) .
With the convention ( R ( h ) W ) ( g ) = W ( g h ) (R(h)W)(g)=W(gh) ( R ( h ) W ) ( g ) = W ( g h ) , we have
R ( h ) W ~ ξ = W ~ ξ ′ , ξ ′ = π v ( t h − 1 ) ξ . R(h)\widetilde{W}_{\xi}=\widetilde{W}_{\xi'},\qquad\xi'=\pi_v(t h^{-1})\xi. R ( h ) W ξ = W ξ ′ , ξ ′ = π v ( t h − 1 ) ξ .
Moreover,
L ( a ) W ~ = χ π v ( a − 1 ) L ( a ) W ~ , a ∈ F v × . \widetilde{L(a)W}=\chi_{\pi_v}(a^{-1})L(a)\widetilde{W},\qquad a\in F_v^\times. L ( a ) W = χ π v ( a − 1 ) L ( a ) W , a ∈ F v × .
New vectors and the essential Whittaker function
Assume that π \pi π is generic. For each nonarchimedean place v v v , normalize the multiplicative Haar measure d × a d^\times a d × a by vol ( o v × , d × a ) = 1 \operatorname{vol}(\mathfrak{o}_v^\times,d^\times a)=1 vol ( o v × , d × a ) = 1 . Recall that ψ v \psi_v ψ v is assumed to have conductor o v \mathfrak{o}_v o v .
Let n ( π v ) n(\pi_v) n ( π v ) be the smallest nonnegative integer such that V π v V_{\pi_v} V π v contains a nonzero vector fixed by K 1 , v ( p v n ( π v ) ) K_{1,v}(\mathfrak{p}_v^{n(\pi_v)}) K 1 , v ( p v n ( π v ) ) . The space of such vectors is one-dimensional, and its nonzero elements are called newvectors ; see [7 ] . The integer n ( π v ) n(\pi_v) n ( π v ) is the conductor exponent of π v \pi_v π v . It also determines the exponent in the local ϵ \epsilon ϵ -factor [11 ] :
ϵ ( s , π v , ψ v ∘ ) = ϵ ( π v , ψ v ∘ ) q v n ( π v ) ( 1 / 2 − s ) , ϵ ( π v , ψ v ∘ ) : = ϵ ( 1 2 , π v , ψ v ∘ ) . \epsilon(s,\pi_v,\psi_v^\circ)=\epsilon(\pi_v,\psi_v^\circ)q_v^{n(\pi_v)(1/2-s)},\qquad\epsilon(\pi_v,\psi_v^\circ):=\epsilon\left(\frac{1}{2},\pi_v,\psi_v^\circ\right). ϵ ( s , π v , ψ v ∘ ) = ϵ ( π v , ψ v ∘ ) q v n ( π v ) ( 1/2 − s ) , ϵ ( π v , ψ v ∘ ) := ϵ ( 2 1 , π v , ψ v ∘ ) .
The conductor of π \pi π is the positive divisor
n = ∏ v < ∞ p v n ( π v ) . \mathfrak n=\prod_{v<\infty}\mathfrak p_v^{n(\pi_v)}. n = v < ∞ ∏ p v n ( π v ) .
Let n = ( n v ) v \mathfrak n=(n_v)_v n = ( n v ) v be the idele defined by n v = ϖ v n ( π v ) n_v=\varpi_v^{n(\pi_v)} n v = ϖ v n ( π v ) at nonarchimedean places and n v = 1 n_v=1 n v = 1 at archimedean places.
For each nonarchimedean place v v v , choose the normalized newvector ξ v 0 \xi_v^0 ξ v 0 by requiring W ξ v 0 ( ( d v − 1 1 ) ) = 1. W e c a l l W_{\xi_v^0}\left(\begin{pmatrix}d_v^{-1}&\\&1\end{pmatrix}\right)=1. We call W ξ v 0 ( ( d v − 1 1 ) ) = 1. W ec a l l W_{\xi_v^0}$ the essential Whittaker function and write
W π v = W ξ v 0 W_{\pi_v}=W_{\xi_v^0} W π v = W ξ v 0
to emphasize the underlying representation. We use the analogous notation W π ~ v W_{\widetilde{\pi}_v} W π v for the essential Whittaker function of the contragredient representation.
For an unramified quasi-character χ v \chi_v χ v of F v × F_v^\times F v × , one has
∫ F v × W π v ( ( a d v − 1 1 ) ) χ v ( a ) ∥ a ∥ v s − 1 / 2 d × a = L ( s , π v ⊗ χ v ) . (2) \int_{F_v^\times}W_{\pi_v}\left(\begin{pmatrix}a d_v^{-1}&\\&1\end{pmatrix}\right)\chi_v(a)\lVert a\rVert_v^{s-1/2}\,d^\times a=L(s,\pi_v\otimes\chi_v). \tag*{(2)} ∫ F v × W π v ( ( a d v − 1 1 ) ) χ v ( a ) ∥ a ∥ v s − 1/2 d × a = L ( s , π v ⊗ χ v ) . (2)
The integral converges for ℜ ( s ) \Re(s) ℜ ( s ) sufficiently large, and the identity holds elsewhere by meromorphic continuation. If χ v \chi_v χ v is ramified, the integral vanishes, since W π v ( ( a 1 ) ) W_{\pi_v}\left(\begin{pmatrix}a&\\&1\end{pmatrix}\right) W π v ( ( a 1 ) ) is invariant under multiplication of a a a by o v × \mathfrak{o}_v^\times o v × .
Suppose now that χ v \chi_v χ v has conductor exponent m v = a ( χ v ) > 0 m_v=a(\chi_v)>0 m v = a ( χ v ) > 0 . Then
∫ F v × W π v ( ( a d v − 1 1 ) ( ϖ v − m v 1 ) ) χ v ( a ) ∥ a ∥ v s − 1 / 2 d × a = G ( χ v , ψ v , ϖ v − m v − δ v ) , (3) \int_{F_v^\times}W_{\pi_v}\left(\begin{pmatrix}a d_v^{-1}&\\&1\end{pmatrix}\begin{pmatrix}\varpi_v^{-m_v}&\\&1\end{pmatrix}\right)\chi_v(a)\lVert a\rVert_v^{s-1/2}\,d^\times a=\mathcal{G}(\chi_v,\psi_v,\varpi_v^{-m_v-\delta_v}), \tag*{(3)} ∫ F v × W π v ( ( a d v − 1 1 ) ( ϖ v − m v 1 ) ) χ v ( a ) ∥ a ∥ v s − 1/2 d × a = G ( χ v , ψ v , ϖ v − m v − δ v ) , (3)
as in [5 ] , where
G ( χ v , ψ v , y ) = ∫ o v × χ v ( u ) ψ v ( y u ) d × u . \mathcal{G}(\chi_v,\psi_v,y)=\int_{\mathfrak{o}_v^\times}\chi_v(u)\psi_v(yu)\,d^\times u. G ( χ v , ψ v , y ) = ∫ o v × χ v ( u ) ψ v ( y u ) d × u .
As above, the integral identity is initially understood in a right half-plane and then by meromorphic continuation.
For m v > 0 m_v > 0 m v > 0 , the Gauss sums satisfy
G ( χ v , ψ v , ϖ ‾ v − m v − δ v ) G ( χ v − 1 , ψ v , ϖ ‾ v − m v − δ v ) = q v − m v ( 1 − q v − 1 ) − 2 χ v ( − 1 ) . (4) \mathcal{G}(\chi_v,\psi_v,\overline{\varpi}_v^{-m_v-\delta_v})\mathcal{G}(\chi_v^{-1},\psi_v,\overline{\varpi}_v^{-m_v-\delta_v})=q_v^{-m_v}(1-q_v^{-1})^{-2}\chi_v(-1). \tag*{(4)} G ( χ v , ψ v , ϖ v − m v − δ v ) G ( χ v − 1 , ψ v , ϖ v − m v − δ v ) = q v − m v ( 1 − q v − 1 ) − 2 χ v ( − 1 ) . (4)
Put
κ v = q v m v / 2 ( 1 − q v − 1 ) G ( χ v − 1 , ψ v , ϖ ‾ v − m v − δ v ) . \kappa_v=q_v^{m_v/2}(1-q_v^{-1})\mathcal{G}(\chi_v^{-1},\psi_v,\overline{\varpi}_v^{-m_v-\delta_v}). κ v = q v m v /2 ( 1 − q v − 1 ) G ( χ v − 1 , ψ v , ϖ v − m v − δ v ) .
Then ∣ κ v ∣ = 1 |\kappa_v|=1 ∣ κ v ∣ = 1 , and (4) gives
G ( χ v , ψ v , ϖ ‾ v − m v − δ v ) = κ v − 1 ( 1 − q v − 1 ) − 1 q v − m v / 2 χ v ( − 1 ) . \mathcal{G}(\chi_v,\psi_v,\overline{\varpi}_v^{-m_v-\delta_v})=\kappa_v^{-1}(1-q_v^{-1})^{-1}q_v^{-m_v/2}\chi_v(-1). G ( χ v , ψ v , ϖ v − m v − δ v ) = κ v − 1 ( 1 − q v − 1 ) − 1 q v − m v /2 χ v ( − 1 ) .
The relation with the abelian local ϵ \epsilon ϵ -factor is given by [14 ] :
ϵ ( s , χ v , ψ v ∘ ) = χ v ( ϖ v m v ) q v − m v ( s − 1 / 2 ) q v m v / 2 ( 1 − q v − 1 ) G ( χ v − 1 , ψ v , ϖ ‾ v − m v − δ v ) . (5) \epsilon(s,\chi_v,\psi_v^\circ)=\chi_v(\varpi_v^{m_v})q_v^{-m_v(s-1/2)}q_v^{m_v/2}(1-q_v^{-1})\mathcal{G}(\chi_v^{-1},\psi_v,\overline{\varpi}_v^{-m_v-\delta_v}). \tag*{(5)} ϵ ( s , χ v , ψ v ∘ ) = χ v ( ϖ v m v ) q v − m v ( s − 1/2 ) q v m v /2 ( 1 − q v − 1 ) G ( χ v − 1 , ψ v , ϖ v − m v − δ v ) . (5)
The local functional equation
Fix a place v v v of F F F and let W v ∈ W ( π v , ψ v ) W_v\in\mathcal{W}(\pi_v,\psi_v) W v ∈ W ( π v , ψ v ) . For a smooth character χ v \chi_v χ v of F v × F_v^\times F v × consider the integral
Ψ v ( s ; W v , χ v ) = ∫ F v × W v ( ( x 1 ) ) χ v ( x ) ∥ x ∥ v s − 1 / 2 d × x \Psi_v(s;W_v,\chi_v)=\int_{F_v^\times}W_v\left(\begin{pmatrix}x& \\ &1\end{pmatrix}\right)\chi_v(x)\lVert x\rVert_v^{s-1/2}\,\mathrm{d}^{\times}x Ψ v ( s ; W v , χ v ) = ∫ F v × W v ( ( x 1 ) ) χ v ( x ) ∥ x ∥ v s − 1/2 d × x
originally defined for $\operatorname{Re}(s)\gg 0. It follows from [8 ] that
Ψ v ( s ; W v , χ v ) L ( s , π v × χ v ) = Φ v ( s ; W v , χ v ) \frac{\Psi_v(s;W_v,\chi_v)}{L(s,\pi_v\times\chi_v)}=\Phi_v(s;W_v,\chi_v) L ( s , π v × χ v ) Ψ v ( s ; W v , χ v ) = Φ v ( s ; W v , χ v )
is an entire function of s s s for all v v v . If v v v is non-archimedean, then Φ v ( s ; W v , χ v ) ∈ C [ q v s , q v − s ] \Phi_v(s;W_v,\chi_v)\in\mathbb{C}[q_v^s,q_v^{-s}] Φ v ( s ; W v , χ v ) ∈ C [ q v s , q v − s ] and for almost all v v v , Φ v ( s ; W v , χ v ) ≡ 1 \Phi_v(s;W_v,\chi_v)\equiv1 Φ v ( s ; W v , χ v ) ≡ 1 . We also have the analogous object Φ v ( s ; W ~ v , χ v − 1 ) \Phi_v(s;\widetilde{W}_v,\chi_v^{-1}) Φ v ( s ; W v , χ v − 1 ) associated to the pair ( π ~ v , χ v − 1 ) (\widetilde{\pi}_v,\chi_v^{-1}) ( π v , χ v − 1 ) of contragredient representations. The local functional equation then asserts that (see loc.cit.)
Φ v ( 1 − s ; W ~ v , χ v − 1 ) = ϵ ( s , π v × χ v , ψ v ) Φ v ( s ; W v , χ v ) (6) \Phi_v(1-s;\widetilde{W}_v,\chi_v^{-1})=\epsilon(s,\pi_v\times\chi_v,\psi_v)\Phi_v(s;W_v,\chi_v) \tag*{(6)} Φ v ( 1 − s ; W v , χ v − 1 ) = ϵ ( s , π v × χ v , ψ v ) Φ v ( s ; W v , χ v ) (6)
for all W v W_v W v .
Lemmas
Initial setup
Let π = ⨂ v π v \pi=\bigotimes_v\pi_v π = ⨂ v π v be as in Theorem 1.1. We first define the auxiliary functions. The normalization described in Remark (4) will be made after Lemma 3.4.
Let Φ ( π ) \Phi(\pi) Φ ( π ) denote the space of functions of the form
φ ξ ′ ( g ) = ∑ γ ∈ F × W ξ ′ ( ( γ 1 ) g ) , ξ ′ ∈ V π . \varphi_{\xi'}(g)=\sum_{\gamma\in F^\times}W_{\xi'}\left(\begin{pmatrix}\gamma& \\ &1\end{pmatrix}g\right),\qquad\xi'\in V_\pi. φ ξ ′ ( g ) = γ ∈ F × ∑ W ξ ′ ( ( γ 1 ) g ) , ξ ′ ∈ V π .
The group G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) acts on Φ ( π ) \Phi(\pi) Φ ( π ) by right translation, and the map ξ ′ ↦ φ ξ ′ \xi'\mapsto\varphi_{\xi'} ξ ′ ↦ φ ξ ′ is an isomorphism of G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) -representations.
Retain the set S 0 S_0 S 0 fixed in §2. For each v ∉ S 0 v\notin S_0 v ∈ / S 0 , fix the normalized essential vector ξ v 0 ∈ V π v \xi_v^0\in V_{\pi_v} ξ v 0 ∈ V π v as in §2.2. For v ∈ S 0 v\in S_0 v ∈ S 0 , let ξ v \xi_v ξ v be an arbitrary smooth vector, and write
ξ = ξ S 0 ⊗ ξ S 0 , ξ S 0 = ⨂ v ∉ S 0 ξ v 0 , ξ S 0 = ⨂ v ∈ S 0 ξ v . \xi=\xi^{S_0}\otimes\xi_{S_0},\qquad\xi^{S_0}=\bigotimes_{v\notin S_0}\xi_v^0,\qquad\xi_{S_0}=\bigotimes_{v\in S_0}\xi_v. ξ = ξ S 0 ⊗ ξ S 0 , ξ S 0 = v ∈ / S 0 ⨂ ξ v 0 , ξ S 0 = v ∈ S 0 ⨂ ξ v .
At archimedean places, smooth vectors are understood to belong to the globalization fixed above. Consider the function ϕ = ϕ ξ S 0 : = ϕ ξ \phi=\phi_{\xi_{S_0}}:=\phi_{\xi} ϕ = ϕ ξ S 0 := ϕ ξ . It satisfies:
(i) ϕ ( γ g ) = ϕ ( g ) \phi(\gamma g)=\phi(g) ϕ ( γ g ) = ϕ ( g ) for γ ∈ B 1 ( F ) \gamma\in B_1(F) γ ∈ B 1 ( F ) ;
(ii) ϕ ( g z ) = χ π ( z ) ϕ ( g ) \phi(gz)=\chi_\pi(z)\phi(g) ϕ ( g z ) = χ π ( z ) ϕ ( g ) for a central element z z z , identified with an idèle in A F × \mathbb A_F^\times A F × ;
(iii) ϕ ( g k ) = ϕ ( g ) \phi(gk)=\phi(g) ϕ ( g k ) = ϕ ( g ) for k ∈ K 1 0 ( n ) k\in K_1^0(\mathfrak n) k ∈ K 1 0 ( n ) .
For each v v v , put
ξ ~ v = π v ( ( ϖ v 0 0 1 ) ) ξ v . \widetilde{\xi}_v=\pi_v\left(\begin{pmatrix}\varpi_v&0\\0&1\end{pmatrix}\right)\xi_v. ξ v = π v ( ( ϖ v 0 0 1 ) ) ξ v .
where ξ v = ξ v 0 \xi_v=\xi_v^0 ξ v = ξ v 0 for v ∉ S 0 v\notin S_0 v ∈ / S 0 , and set ξ ~ = ⨂ v ξ ~ v \widetilde{\xi}=\bigotimes_v\widetilde{\xi}_v ξ = ⨂ v ξ v . Define
ϕ ~ ξ ( g ) = ∑ γ ∈ F × W ~ ξ ( ( γ 0 0 1 ) g ) . \widetilde{\phi}_{\xi}(g)=\sum_{\gamma\in F^\times}\widetilde{W}_{\xi}\left(\begin{pmatrix}\gamma&0\\0&1\end{pmatrix}g\right). ϕ ξ ( g ) = γ ∈ F × ∑ W ξ ( ( γ 0 0 1 ) g ) .
One checks that
ϕ ~ ξ ( g ) = χ π ( det g ) − 1 ϕ ~ ξ ( g ( 0 1 − 1 0 ) ) . \widetilde{\phi}_{\xi}(g)=\chi_\pi(\det g)^{-1}\widetilde{\phi}_{\xi}\left(g\begin{pmatrix}0&1\\-1&0\end{pmatrix}\right). ϕ ξ ( g ) = χ π ( det g ) − 1 ϕ ξ ( g ( 0 − 1 1 0 ) ) .
The function ϕ ~ = ϕ ~ ξ S 0 : = ϕ ~ ξ \widetilde{\phi}=\widetilde{\phi}_{\xi_{S_0}}:=\widetilde{\phi}_{\xi} ϕ = ϕ ξ S 0 := ϕ ξ satisfies (i) and (iii), and satisfies (ii) with χ π \chi_\pi χ π replaced by χ π − 1 \chi_\pi^{-1} χ π − 1 . The function ϕ \phi ϕ has the Fourier expansion
ϕ ξ S 0 ( ( y x 0 1 ) ) = ∑ γ ∈ F × c S 0 ( γ y ) W ξ S 0 ( ( γ y 0 0 1 ) ) ψ ( γ x ) , y ∈ A F × , x ∈ A F , (7) \phi_{\xi_{S_0}}\left(\begin{pmatrix}y&x\\0&1\end{pmatrix}\right)
=
\sum_{\gamma\in F^\times}c^{S_0}(\gamma y)W_{\xi_{S_0}}\left(\begin{pmatrix}\gamma y&0\\0&1\end{pmatrix}\right)\psi(\gamma x),
\qquad y\in\mathbb A_F^\times,\quad x\in\mathbb A_F, \tag*{(7)} ϕ ξ S 0 ( ( y 0 x 1 ) ) = γ ∈ F × ∑ c S 0 ( γ y ) W ξ S 0 ( ( γ y 0 0 1 ) ) ψ ( γ x ) , y ∈ A F × , x ∈ A F , (7)
where
c S 0 ( z ) = W ξ S 0 ( ( z 0 0 1 ) ) . c^{S_0}(z)=W_{\xi_{S_0}}\left(\begin{pmatrix}z&0\\0&1\end{pmatrix}\right). c S 0 ( z ) = W ξ S 0 ( ( z 0 0 1 ) ) .
Similarly,
ϕ ~ ξ S 0 ( ( y x 0 1 ) ) = ∑ γ ∈ F × c ~ S 0 ( γ y ) W ~ ξ S 0 ( ( γ y 0 0 1 ) ) ψ ( γ x ) , (8) \widetilde{\phi}_{\xi_{S_0}}\left(\begin{pmatrix}y&x\\0&1\end{pmatrix}\right)
=
\sum_{\gamma\in F^\times}\widetilde{c}^{S_0}(\gamma y)\widetilde{W}_{\xi_{S_0}}\left(\begin{pmatrix}\gamma y&0\\0&1\end{pmatrix}\right)\psi(\gamma x), \tag*{(8)} ϕ ξ S 0 ( ( y 0 x 1 ) ) = γ ∈ F × ∑ c S 0 ( γ y ) W ξ S 0 ( ( γ y 0 0 1 ) ) ψ ( γ x ) , (8)
where
c ~ S 0 ( z ) = W ~ ξ ~ S 0 ( ( z 0 0 1 ) ) . \widetilde{c}^{S_0}(z)=\widetilde{W}_{\widetilde{\xi}_{S_0}}\left(\begin{pmatrix}z&0\\0&1\end{pmatrix}\right). c S 0 ( z ) = W ξ S 0 ( ( z 0 0 1 ) ) .
Here ξ ~ v = ξ v \widetilde{\xi}_v=\xi_v ξ v = ξ v for v ∈ S 0 v\in S_0 v ∈ S 0 , since n v = 1 \mathfrak n_v=1 n v = 1 at these places. For v ∉ S 0 v\notin S_0 v ∈ / S 0 , W ~ ξ ~ v \widetilde{W}_{\widetilde{\xi}_v} W ξ v is a nonzero scalar multiple of the essential Whittaker function W π ~ v W_{\widetilde{\pi}_v} W π v . Consequently, both c S 0 ( z ) c^{S_0}(z) c S 0 ( z ) and c ~ S 0 ( z ) \widetilde{c}^{S_0}(z) c S 0 ( z ) depend only on m = div S 0 ( z ) \mathfrak m=\operatorname{div}^{S_0}(z) m = div S 0 ( z ) . They vanish unless
ord v ( d v z v ) ≥ 0 for every v ∉ S 0 , \operatorname{ord}_v(\mathfrak d_v z_v)\geq0\qquad\text{for every }v\notin S_0, ord v ( d v z v ) ≥ 0 for every v ∈ / S 0 ,
or, equivalently, unless div S 0 ( d z ) \operatorname{div}^{S_0}(\mathfrak d z) div S 0 ( d z ) is effective. We regard these coefficients as functions on divisors supported outside S 0 S_0 S 0 , and write c S 0 ( m ) c^{S_0}(\mathfrak m) c S 0 ( m ) and c ~ S 0 ( m ) \widetilde{c}^{S_0}(\mathfrak m) c S 0 ( m ) .
We call ( ϕ , ϕ ~ ) = ( ϕ ξ S 0 , ϕ ~ ξ S 0 ) (\phi,\widetilde{\phi})=(\phi_{\xi_{S_0}},\widetilde{\phi}_{\xi_{S_0}}) ( ϕ , ϕ ) = ( ϕ ξ S 0 , ϕ ξ S 0 ) a dual pair . Their Mellin transforms recover the two sides of the functional equations for twists by idèle class characters unramified outside S 0 S_0 S 0 .
For a local Whittaker function W W W and a quasi-character χ v \chi_v χ v of F v × F_v^\times F v × , write
Ψ ( s ; W , χ v ) = ∫ F v × W ( ( a 0 0 1 ) ) χ v ( a ) ∣ a ∣ v s − 1 / 2 d × a \Psi(s;W,\chi_v)=\int_{F_v^\times}W\left(\begin{pmatrix}a&0\\0&1\end{pmatrix}\right)\chi_v(a)\lvert a\rvert_v^{s-1/2}\,d^\times a Ψ ( s ; W , χ v ) = ∫ F v × W ( ( a 0 0 1 ) ) χ v ( a ) ∣ a ∣ v s − 1/2 d × a
initially for ℜ ( s ) \Re(s) ℜ ( s ) sufficiently large and elsewhere by meromorphic continuation. For an idèle class character χ \chi χ , put
C ψ S 0 ( s , χ ) = ∏ v ∉ S 0 χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s . C_{\psi}^{S_0}(s,\chi)=\prod_{v\notin S_0}\chi_v(\mathfrak d_v)^{-1}\lvert\mathfrak d_v\rvert_v^{1/2-s}. C ψ S 0 ( s , χ ) = v ∈ / S 0 ∏ χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s .
This is a finite product and non-vanishing for every s s s . For ℜ ( s ) \Re(s) ℜ ( s ) sufficiently large, unfolding and (2.1) gives
∫ F × \ A F × ϕ ( ( t 1 ) ) χ ( t ) ∥ t ∥ s − 1 / 2 d × t = C ψ S 0 ( s , χ ) ( ∏ v ∈ S 0 Ψ ( s ; W ξ v , χ v ) ) L S 0 ( s , π × χ ) . (9) \int_{F^\times\backslash\mathbb{A}_F^\times}\phi\left(\begin{pmatrix}t\\1\end{pmatrix}\right)\chi(t)\lVert t\rVert^{s-1/2}\,d^\times t=C_{\psi}^{S_0}(s,\chi)\left(\prod_{v\in S_0}\Psi(s;W_{\xi_v},\chi_v)\right)L^{S_0}(s,\pi\times\chi). \tag*{(9)} ∫ F × \ A F × ϕ ( ( t 1 ) ) χ ( t ) ∥ t ∥ s − 1/2 d × t = C ψ S 0 ( s , χ ) ( v ∈ S 0 ∏ Ψ ( s ; W ξ v , χ v ) ) L S 0 ( s , π × χ ) . (9)
where
L S 0 ( s , π × χ ) = ∏ v ∉ S 0 L ( s , π v × χ v ) . L^{S_0}(s,\pi\times\chi)=\prod_{v\notin S_0}L(s,\pi_v\times\chi_v). L S 0 ( s , π × χ ) = v ∈ / S 0 ∏ L ( s , π v × χ v ) .
On the other hand, the definition of ξ ~ \widetilde{\xi} ξ gives
W ~ ξ ~ ( g ( n 1 ) ) = W ~ ξ ~ ( g ) . \widetilde{W}_{\widetilde{\xi}}\left(g\begin{pmatrix}n\\1\end{pmatrix}\right)=\widetilde{W}_{\widetilde{\xi}}(g). W ξ ( g ( n 1 ) ) = W ξ ( g ) .
In particular,
W ~ ξ ~ ( ( n t − 1 1 ) ) = W ~ ξ ~ ( ( t − 1 1 ) ) . \widetilde{W}_{\widetilde{\xi}}\left(\begin{pmatrix}nt^{-1}\\1\end{pmatrix}\right)=\widetilde{W}_{\widetilde{\xi}}\left(\begin{pmatrix}t^{-1}\\1\end{pmatrix}\right). W ξ ( ( n t − 1 1 ) ) = W ξ ( ( t − 1 1 ) ) .
Thus, for ℜ ( s ) \Re(s) ℜ ( s ) sufficiently negative, unfolding and then changing variables gives
∫ F × \ A F × ϕ ~ ( ( n t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1 / 2 d × t = ∫ A F × W ~ ξ ~ ( ( t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1 / 2 d × t = ∏ v Ψ ( 1 − s ; W ~ ξ ~ v , χ v − 1 ) . (10) \begin{aligned}
\int_{F^\times\backslash\mathbb{A}_F^\times}\widetilde{\phi}\left(\begin{pmatrix}nt^{-1}\\1\end{pmatrix}\right)\chi(t)\lVert t\rVert^{s-1/2}\,d^\times t
=\int_{\mathbb{A}_F^\times}\widetilde{W}_{\widetilde{\xi}}\left(\begin{pmatrix}t^{-1}\\1\end{pmatrix}\right)\chi(t)\lVert t\rVert^{s-1/2}\,d^\times t \\
&=\prod_v\Psi(1-s;\widetilde{W}_{\widetilde{\xi}_v},\chi_v^{-1}). \tag*{(10)}
\end{aligned} ∫ F × \ A F × ϕ ( ( n t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1/2 d × t = ∫ A F × W ξ ( ( t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1/2 d × t = v ∏ Ψ ( 1 − s ; W ξ v , χ v − 1 ) . (10)
For v ∉ S 0 v\notin S_0 v ∈ / S 0 , the local functional equation (2.5) gives
Ψ ( 1 − s ; W ~ ξ ~ v , χ v − 1 ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s ϵ ( s , π v × χ v , ψ v ) × L ( 1 − s , π ~ v × χ v − 1 ) . \Psi(1-s;\widetilde{W}_{\widetilde{\xi}_v},\chi_v^{-1})=\chi_v(d_v)^{-1}|d_v|_v^{1/2-s}\epsilon(s,\pi_v\times\chi_v,\psi_v)\times L(1-s,\widetilde{\pi}_v\times\chi_v^{-1}). Ψ ( 1 − s ; W ξ v , χ v − 1 ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s ϵ ( s , π v × χ v , ψ v ) × L ( 1 − s , π v × χ v − 1 ) .
Put
ϵ S 0 ( s , π × χ , ψ ) = ∏ v ∉ S 0 ϵ ( s , π v × χ v , ψ v ) . \epsilon^{S_0}(s,\pi\times\chi,\psi)=\prod_{v\notin S_0}\epsilon(s,\pi_v\times\chi_v,\psi_v). ϵ S 0 ( s , π × χ , ψ ) = v ∈ / S 0 ∏ ϵ ( s , π v × χ v , ψ v ) .
We conclude that
∫ F × \ A F × ϕ ~ ( ( n t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1 / 2 d × t = C ψ S 0 ( s , χ ) ( ∏ v ∈ S 0 Ψ ( 1 − s ; W ~ ξ ~ v , χ v − 1 ) ) ϵ S 0 ( s , π × χ , ψ ) L S 0 ( 1 − s , π ~ × χ − 1 ) . (11) \int_{F^\times\backslash\mathbb{A}_F^\times}\widetilde{\phi}\left(\begin{pmatrix}nt^{-1}\\1\end{pmatrix}\right)\chi(t)\lVert t\rVert^{s-1/2}\,d^\times t=C_{\psi}^{S_0}(s,\chi)\left(\prod_{v\in S_0}\Psi(1-s;\widetilde{W}_{\widetilde{\xi}_v},\chi_v^{-1})\right)\epsilon^{S_0}(s,\pi\times\chi,\psi)L^{S_0}(1-s,\widetilde{\pi}\times\chi^{-1}). \tag*{(11)} ∫ F × \ A F × ϕ ( ( n t − 1 1 ) ) χ ( t ) ∥ t ∥ s − 1/2 d × t = C ψ S 0 ( s , χ ) ( v ∈ S 0 ∏ Ψ ( 1 − s ; W ξ v , χ v − 1 ) ) ϵ S 0 ( s , π × χ , ψ ) L S 0 ( 1 − s , π × χ − 1 ) . (11)
The Mellin identities are understood beyond their initial domains of convergence by meromorphic continuation.
The lemmas
We begin with the following result on Mellin inversion (cf. [8 Lemma 11.3.1] ). For a continuous function f f f on F × \ A F × F^\times\backslash\mathbb{A}_F^\times F × \ A F × and an idèle class character χ \chi χ , define
M ( s , f , χ ) = ∫ F × \ A F × f ( a ) χ ( a ) ∥ a ∥ s − 1 / 2 d × a \mathcal{M}(s,f,\chi)=\int_{F^\times\backslash\mathbb{A}_F^\times}f(a)\chi(a)\lVert a\rVert^{s-1/2}\,d^\times a M ( s , f , χ ) = ∫ F × \ A F × f ( a ) χ ( a ) ∥ a ∥ s − 1/2 d × a
whenever the integral converges absolutely.
Lemma 3.1. Let f 1 f_1 f 1 and f 2 f_2 f 2 be continuous functions on F × \ A F × F^\times\backslash\mathbb{A}_F^\times F × \ A F × . Suppose there is a constant c > 0 c>0 c > 0 such that, for every unitary idèle class character χ \chi χ , M ( s , f 1 , χ ) \mathcal{M}(s,f_1,\chi) M ( s , f 1 , χ ) converges absolutely for ℜ ( s ) > c \Re(s)>c ℜ ( s ) > c , and M ( s , f 2 , χ ) \mathcal{M}(s,f_2,\chi) M ( s , f 2 , χ ) converges absolutely for ℜ ( s ) < − c \Re(s)<-c ℜ ( s ) < − c . For each χ \chi χ , assume that the functions represented by these integrals continue to the same entire function of s s s , bounded in every vertical strip of finite width. Then f 1 = f 2 f_1=f_2 f 1 = f 2 .
Following [16 Chapter IV] , a pair ( f , e ) ∈ A F × × A F (\mathbf{f},\mathbf{e})\in\mathbb{A}_F^\times\times\mathbb{A}_F ( f , e ) ∈ A F × × A F is called reduced if it satisfies the following conditions. At each nonarchimedean place v v v ,
f v = ϖ v n v , n v ≥ 0 , e v ∈ o v , max { ∥ f v ∥ v , ∥ e v ∥ v } = 1. f_v=\varpi_v^{n_v},\qquad n_v\geq0,\qquad e_v\in\mathfrak{o}_v,\qquad\max\{\lVert f_v\rVert_v,\lVert e_v\rVert_v\}=1. f v = ϖ v n v , n v ≥ 0 , e v ∈ o v , max {∥ f v ∥ v , ∥ e v ∥ v } = 1.
In particular, e v ∈ o v × e_v\in\mathfrak{o}_v^\times e v ∈ o v × whenever n v > 0 n_v>0 n v > 0 . At an archimedean place v v v , we require f v ∈ R > 0 f_v\in\mathbb{R}_{>0} f v ∈ R > 0 and e v ∈ F v e_v\in F_v e v ∈ F v , with
f v f v ‾ + e v e v ‾ = 1. f_v\overline{f_v}+e_v\overline{e_v}=1. f v f v + e v e v = 1.
Every h ∈ B 1 ( A F ) h\in B_1(\mathbb{A}_F) h ∈ B 1 ( A F ) can be written as
h = ( t f t e 1 ) , t ∈ A F × , h=\begin{pmatrix}tf&te\\ &1\end{pmatrix},\qquad t\in\mathbb{A}_F^\times, h = ( t f t e 1 ) , t ∈ A F × ,
for a reduced pair ( f , e ) (\mathbf{f},\mathbf{e}) ( f , e ) . We call
div ( f ) = ∏ v ∤ ∞ p v n v \operatorname{div}(\mathbf{f})=\prod_{v\nmid\infty}\mathfrak{p}_v^{n_v} div ( f ) = v ∤ ∞ ∏ p v n v
the conductor of the reduced pair.
Given a reduced pair ( f , e ) (\mathbf{f},\mathbf{e}) ( f , e ) , we choose e ′ ∈ A F \mathbf{e}'\in\mathbb{A}_F e ′ ∈ A F with e v ′ = − e v ‾ e'_v=-\overline{e_v} e v ′ = − e v at every archimedean place. Its components at nonarchimedean places are chosen according to the situation. If the conductor of ( f , e ) (\mathbf{f},\mathbf{e}) ( f , e ) is coprime to n \mathfrak{n} n , choose e v ′ ∈ o v e'_v\in\mathfrak{o}_v e v ′ ∈ o v such that
∥ 1 + n v e v e v ′ ∥ v ≤ ∥ f v ∥ v (12) \lVert1+n_ve_ve'_v\rVert_v\leq\lVert f_v\rVert_v \tag*{(12)} ∥ 1 + n v e v e v ′ ∥ v ≤ ∥ f v ∥ v (12)
at every nonarchimedean place. If the conductor of ( f , e ) (f,e) ( f , e ) is a multiple of n \mathfrak{n} n , choose e v ′ ∈ o v e'_v\in\mathfrak{o}_v e v ′ ∈ o v such that
∥ 1 + e v e v ′ ∥ v ≤ ∥ f v ∥ v . (13) \lVert1+e_ve'_v\rVert_v\leq\lVert f_v\rVert_v. \tag*{(13)} ∥ 1 + e v e v ′ ∥ v ≤ ∥ f v ∥ v . (13)
Put
k v = ( ( 1 + e v e v ′ ) / f v − e v ′ − e v f v ) , b v = ( f v e v 1 ) , b v ′ = ( f v e v ′ 1 ) . (14) k_v=\begin{pmatrix}(1+e_ve'_v)/f_v&-e'_v\\-e_v&f_v\end{pmatrix},\qquad
b_v=\begin{pmatrix}f_v&e_v\\&1\end{pmatrix},\qquad
b'_v=\begin{pmatrix}f_v&e'_v\\&1\end{pmatrix}. \tag*{(14)} k v = ( ( 1 + e v e v ′ ) / f v − e v − e v ′ f v ) , b v = ( f v e v 1 ) , b v ′ = ( f v e v ′ 1 ) . (14)
In the function field case, k v ∞ ∈ GL 2 ( o v ∞ ) k_{v_\infty}\in\operatorname{GL}_2(\mathfrak{o}_{v_\infty}) k v ∞ ∈ GL 2 ( o v ∞ ) . At an archimedean place, k v k_v k v belongs to SO ( 2 ) \operatorname{SO}(2) SO ( 2 ) if F v = R F_v=\mathbb{R} F v = R and to SU ( 2 ) \operatorname{SU}(2) SU ( 2 ) if F v = C F_v=\mathbb{C} F v = C . We take the corresponding maximal compact subgroup K v K_v K v to be O ( 2 ) \operatorname{O}(2) O ( 2 ) or U ( 2 ) \operatorname{U}(2) U ( 2 ) , respectively.
With w = ( 0 1 − 1 0 ) w=\begin{pmatrix}0&1\\-1&0\end{pmatrix} w = ( 0 − 1 1 0 ) , direct calculation gives
b v = w ( b v ′ k v ) w − 1 . b_v=w(b'_vk_v)w^{-1}. b v = w ( b v ′ k v ) w − 1 .
Consequently, writing
k S 0 = ( k v ) v ∈ S 0 , b S 0 = ( b v ) v ∈ S 0 , b S 0 ′ = ( b v ′ ) v ∈ S 0 , k_{S_0}=(k_v)_{v\in S_0},\qquad b_{S_0}=(b_v)_{v\in S_0},\qquad b'_{S_0}=(b'_v)_{v\in S_0}, k S 0 = ( k v ) v ∈ S 0 , b S 0 = ( b v ) v ∈ S 0 , b S 0 ′ = ( b v ′ ) v ∈ S 0 ,
we have
b S 0 = w ( b S 0 ′ k S 0 ) w − 1 . b_{S_0}=w(b'_{S_0}k_{S_0})w^{-1}. b S 0 = w ( b S 0 ′ k S 0 ) w − 1 .
When convenient, we write k S 0 ( f , e ) k_{S_0}(f,e) k S 0 ( f , e ) , suppressing its dependence on the chosen e ′ e' e ′ .
For a reduced pair ( f , e ) (f,e) ( f , e ) , put
f ( t ) = ϕ ξ S 0 ( ( t f t e 1 ) ) , t ∈ A F × . f(t)=\phi_{\xi_{S_0}}\left(\begin{pmatrix}tf&te\\&1\end{pmatrix}\right),\qquad t\in\mathbb{A}_F^\times. f ( t ) = ϕ ξ S 0 ( ( t f t e 1 ) ) , t ∈ A F × .
Lemma 3.2. Let ( f , e ) (\mathfrak f,\mathfrak e) ( f , e ) be a reduced pair whose conductor m \mathfrak m m is coprime to n \mathfrak n n . Then f ∣ m S 0 \mathfrak f\mid\mathfrak m^{S_0} f ∣ m S 0 . When this divisibility holds, it is given by
C ψ S 0 ( s , χ ) ∏ v ∉ S 0 v ∤ m L ( s , π v × χ v ) ∏ v ∉ S 0 v ∣ m Ψ ( s ; L ( d v ) W b v ⋅ ξ v , χ v ) ∏ v ∈ S 0 Ψ ( s ; W b v ⋅ ξ v , χ v ) . C_{\psi}^{S_0}(s,\chi)\prod_{\substack{v\notin S_0\\v\nmid\mathfrak m}}L(s,\pi_v\times\chi_v)\prod_{\substack{v\notin S_0\\v\mid\mathfrak m}}\Psi(s;L(d_v)W_{b_v\cdot\xi_v},\chi_v)\prod_{v\in S_0}\Psi(s;W_{b_v\cdot\xi_v},\chi_v). C ψ S 0 ( s , χ ) v ∈ / S 0 v ∤ m ∏ L ( s , π v × χ v ) v ∈ / S 0 v ∣ m ∏ Ψ ( s ; L ( d v ) W b v ⋅ ξ v , χ v ) v ∈ S 0 ∏ Ψ ( s ; W b v ⋅ ξ v , χ v ) . where b v = ( f v e v 1 ) b_v=\left(\begin{smallmatrix}\mathfrak f_v&\mathfrak e_v\\1&\end{smallmatrix}\right) b v = ( f v 1 e v ) . The formula is initially valid for ℜ ( s ) \Re(s) ℜ ( s ) sufficiently large.
Proof. Unfolding gives
M ( s , f , χ ) = ∫ A F × W ξ ( ( t f t e 1 ) ) χ ( t ) ∣ t ∣ s − 1 / 2 d × t = ∏ v Ψ ( s ; W b v ⋅ ξ v , χ v ) . \begin{aligned}
\mathcal M(s,\mathfrak f,\chi)=\int_{A_{\mathbb F}^{\times}}W_\xi\left(\left(\begin{smallmatrix}t\mathfrak f& t\mathfrak e\\1&\end{smallmatrix}\right)\right)\chi(t)|t|^{s-1/2}\,d^\times t \\
&=\prod_v\Psi(s;W_{b_v\cdot\xi_v},\chi_v).
\end{aligned} M ( s , f , χ ) = ∫ A F × W ξ ( ( t f 1 t e ) ) χ ( t ) ∣ t ∣ s − 1/2 d × t = v ∏ Ψ ( s ; W b v ⋅ ξ v , χ v ) .
Note that, for any W W W , we have
Ψ ( s ; W , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s Ψ ( s ; L ( d v ) W , χ v ) (15) \Psi(s;W,\chi_v)=\chi_v(d_v)^{-1}|d_v|_v^{1/2-s}\Psi(s;L(d_v)W,\chi_v) \tag*{(15)} Ψ ( s ; W , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s Ψ ( s ; L ( d v ) W , χ v ) (15)
If v ∉ S 0 v\notin S_0 v ∈ / S 0 and v ∤ m v\nmid\mathfrak m v ∤ m , then b v ⋅ ξ v = ξ v b_v\cdot\xi_v=\xi_v b v ⋅ ξ v = ξ v . The local integral vanishes unless χ v \chi_v χ v is unramified. In the latter case, changing variables and using (15) and (2) gives
Ψ ( s ; W ξ v , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s L ( s , π v × χ v ) . \Psi(s;W_{\xi_v},\chi_v)=\chi_v(d_v)^{-1}|d_v|_v^{1/2-s}L(s,\pi_v\times\chi_v). Ψ ( s ; W ξ v , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s L ( s , π v × χ v ) .
Suppose v ∉ S 0 v\notin S_0 v ∈ / S 0 and v ∣ m v\mid\mathfrak m v ∣ m . Write f v = ϖ v n v \mathfrak f_v=\varpi_v^{n_v} f v = ϖ v n v , where n v > 0 n_v>0 n v > 0 . Since the pair is reduced, e v ∈ o v × \mathfrak e_v\in\mathfrak o_v^\times e v ∈ o v × , and
W b v ⋅ ξ v ( ( t 1 ) ) = ψ v ( t e v ) W ξ v ( ( t f v 1 ) ) . W_{b_v\cdot\xi_v}\left(\left(\begin{smallmatrix}t\\1\end{smallmatrix}\right)\right)=\psi_v(t\mathfrak e_v)W_{\xi_v}\left(\left(\begin{smallmatrix}t\mathfrak f_v\\1\end{smallmatrix}\right)\right). W b v ⋅ ξ v ( ( t 1 ) ) = ψ v ( t e v ) W ξ v ( ( t f v 1 ) ) .
This expression is unchanged when t t t is replaced by t u tu t u , for u ∈ 1 + f v o v u\in1+\mathfrak f_v\mathfrak o_v u ∈ 1 + f v o v . Indeed, the Whittaker function is invariant under multiplication by units. If it is nonzero, then
d v t f v ∈ o v . d_vt\mathfrak f_v\in\mathfrak o_v. d v t f v ∈ o v .
For u ∈ 1 + f v o v u\in1+\mathfrak f_v\mathfrak o_v u ∈ 1 + f v o v , we therefore have
t e v ( u − 1 ) ∈ d v − 1 o v . t\mathfrak e_v(u-1)\in d_v^{-1}\mathfrak o_v. t e v ( u − 1 ) ∈ d v − 1 o v .
Since ψ v \psi_v ψ v is trivial on d v − 1 o v d_v^{-1}\mathfrak o_v d v − 1 o v , the additive character is unchanged as well. Consequently,
Ψ ( s ; W b v ⋅ ξ v , χ v ) = χ v ( u ) − 1 Ψ ( s ; W b v ⋅ ξ v , χ v ) . \Psi(s;W_{b_v\cdot\xi_v},\chi_v)=\chi_v(u)^{-1}\Psi(s;W_{b_v\cdot\xi_v},\chi_v). Ψ ( s ; W b v ⋅ ξ v , χ v ) = χ v ( u ) − 1 Ψ ( s ; W b v ⋅ ξ v , χ v ) .
Thus the local integral vanishes unless χ v \chi_v χ v is trivial on 1 + p v n v 1+\mathfrak p_v^{n_v} 1 + p v n v , equivalently a ( χ v ) ≤ n v a(\chi_v)\le n_v a ( χ v ) ≤ n v . The stated form follows from (3.9). There is no additional restriction at places in S 0 S_0 S 0 . These observations prove the lemma. □ \square □
An analogous result holds for
f ~ ( t ) = χ π ( f ) ( R ( k S 0 ) ϕ ~ S 0 ) ( ( n t − 1 f n t − 1 e 1 ) ) . \widetilde f(t)=\chi_\pi(\mathfrak f)\bigl(R(k_{S_0})\widetilde\phi_{S_0}\bigr)\left(\left(\begin{smallmatrix}nt^{-1}\mathfrak f&nt^{-1}\mathfrak e\\1&\end{smallmatrix}\right)\right). f ( t ) = χ π ( f ) ( R ( k S 0 ) ϕ S 0 ) ( ( n t − 1 f 1 n t − 1 e ) ) .
Lemma 3.3. With notation as in Lemma 3.2 , and e ′ \mathfrak e' e ′ chosen as in (12) , the Mellin transform M ( s , f ~ , χ ) \mathcal M(s,\widetilde{\mathfrak f},\chi) M ( s , f , χ ) vanishes unless
f ~ χ S 0 ∣ m S 0 . \widetilde{\mathfrak f}^{S_0}_{\chi}\mid\mathfrak m^{S_0}. f χ S 0 ∣ m S 0 .
When this divisibility holds, it equals
C ψ S 0 ( s , χ ) ∏ v ∉ S 0 v ∤ m ( ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π ~ v × χ v − 1 ) ) C_{\psi}^{S_0}(s,\chi)\prod_{\substack{v\notin S_0\\v\nmid\mathfrak m}}\left(\epsilon(s,\pi_v\times\chi_v,\psi_v)L(1-s,\widetilde{\pi}_v\times\chi_v^{-1})\right) C ψ S 0 ( s , χ ) v ∈ / S 0 v ∤ m ∏ ( ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π v × χ v − 1 ) )
× ∏ v ∉ S 0 v ∣ m ( ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W b v ⋅ ξ v , χ v − 1 ) ) {}\times\prod_{\substack{v\notin S_0\\v\mid\mathfrak m}}\left((\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}\Psi\left(1-s;L(d_v)W_{b_v\cdot\xi_v},\chi_v^{-1}\right)\right) × v ∈ / S 0 v ∣ m ∏ ( ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W b v ⋅ ξ v , χ v − 1 ) )
× ∏ v ∈ S 0 Ψ ( 1 − s ; W ~ b v ⋅ ξ v , χ v − 1 ) , {}\times\prod_{v\in S_0}\Psi\left(1-s;\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1}\right), × v ∈ S 0 ∏ Ψ ( 1 − s ; W b v ⋅ ξ v , χ v − 1 ) ,
where
b v = ( f v e v 1 ) . b_v=\binom{f_v e_v}{1}. b v = ( 1 f v e v ) .
The formula is initially valid for ℜ ( s ) \Re(s) ℜ ( s ) sufficiently negative.
Proof. Unfolding and the local matrix calculations used in the proof of Lemma 3.6 , with the averaging operators omitted, give
M ( s , f ~ , χ ) = ∏ v Ψ ( 1 − s ; W ~ b v ⋅ ξ v , χ v − 1 ) . \mathcal{M}(s,\widetilde{f},\chi)=\prod_v\Psi\left(1-s;\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1}\right). M ( s , f , χ ) = v ∏ Ψ ( 1 − s ; W b v ⋅ ξ v , χ v − 1 ) .
Here the factors χ π v ( f v ) − 1 \chi_{\pi_v}(f_v)^{-1} χ π v ( f v ) − 1 arising from those calculations cancel the prefactor χ π ( f ) \chi_\pi(f) χ π ( f ) in the definition of f ~ \widetilde{f} f .
For v ∉ S 0 v\notin S_0 v ∈ / S 0 , put
C v ( s , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s . C_v(s,\chi_v)=\chi_v(d_v)^{-1}|d_v|_v^{1/2-s}. C v ( s , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s .
If v ∤ m v\nmid\mathfrak m v ∤ m , then b v ⋅ ξ v = ξ v b_v\cdot\xi_v=\xi_v b v ⋅ ξ v = ξ v . The local functional equation and Lemma 3.2 show that the local integral vanishes unless χ v \chi_v χ v is unramified. When χ v \chi_v χ v is unramified, it equals
C v ( s , χ v ) ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π ~ v × χ v − 1 ) . C_v(s,\chi_v)\epsilon(s,\pi_v\times\chi_v,\psi_v)L(1-s,\widetilde{\pi}_v\times\chi_v^{-1}). C v ( s , χ v ) ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π v × χ v − 1 ) .
Now suppose v ∉ S 0 v\notin S_0 v ∈ / S 0 and v ∣ m v\mid\mathfrak m v ∣ m . For any W ∈ W ( π v , ψ v ) W\in\mathcal{W}(\pi_v,\psi_v) W ∈ W ( π v , ψ v ) , a change of variables gives
Ψ ( 1 − s ; W ~ , χ v − 1 ) = C v ( s , χ v ) − 1 Ψ ( 1 − s ; L ( d v ) W , χ v − 1 ) . \Psi(1-s;\widetilde{W},\chi_v^{-1})=C_v(s,\chi_v)^{-1}\Psi(1-s;L(d_v)W,\chi_v^{-1}). Ψ ( 1 − s ; W , χ v − 1 ) = C v ( s , χ v ) − 1 Ψ ( 1 − s ; L ( d v ) W , χ v − 1 ) .
Using
L ( d v ) W ~ = χ π v ( d v ) L ( d v ) W ~ , L(d_v)\widetilde{W}=\chi_{\pi_v}(d_v)\widetilde{L(d_v)W}, L ( d v ) W = χ π v ( d v ) L ( d v ) W ,
we obtain
Ψ ( 1 − s ; W ~ , χ v − 1 ) = C v ( s , χ v ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W ~ , χ v − 1 ) . \Psi(1-s;\widetilde{W},\chi_v^{-1})=C_v(s,\chi_v)(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}\Psi(1-s;\widetilde{L(d_v)W},\chi_v^{-1}). Ψ ( 1 − s ; W , χ v − 1 ) = C v ( s , χ v ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W , χ v − 1 ) .
Apply this with W = W b v ⋅ ξ v W=W_{b_v\cdot\xi_v} W = W b v ⋅ ξ v . Moreover, the local functional equation transfers the vanishing established in Lemma 3.2 to the dual integral. Thus this factor vanishes unless
a ( χ v ) ≤ ord v ( m ) . a(\chi_v)\leq\operatorname{ord}_v(\mathfrak m). a ( χ v ) ≤ ord v ( m ) .
At the places in S 0 S_0 S 0 , we retain the local integrals without modification. Multiplying the local identities and using
∏ v ∉ S 0 C v ( s , χ v ) = C ψ S 0 ( s , χ ) \prod_{v\notin S_0}C_v(s,\chi_v)=C_{\psi}^{S_0}(s,\chi) v ∈ / S 0 ∏ C v ( s , χ v ) = C ψ S 0 ( s , χ )
proves the lemma. □
Lemma 3.4. Under the hypotheses of Theorem 1.1 , let ( ϕ , ϕ ~ ) (\phi,\widetilde{\phi}) ( ϕ , ϕ ) be associated to any choice of smooth vectors at the places in S 0 S_0 S 0 . Let ( f , e ) (\mathfrak{f},\mathfrak{e}) ( f , e ) be a reduced pair whose conductor is supported in S 0 S_0 S 0 , and choose e ′ \mathfrak{e}' e ′ satisfying (3.6). Put ϵ = ϵ [ 1 ] \epsilon=\epsilon_{[1]} ϵ = ϵ [ 1 ] . Then, for every t ∈ A F × t\in\mathbb{A}_{F}^{\times} t ∈ A F × ,
ϕ ( ( t f t e 1 1 ) ) = ϵ χ π ( f ) ( R ( k S 0 ) ϕ ~ ) ( ( n 1 − 1 f n 1 − 1 e ′ 1 1 ) ) , \phi\left(\begin{pmatrix} t\mathfrak{f} & t\mathfrak{e} \\ 1 & 1 \end{pmatrix}\right)=\epsilon\chi_{\pi}(\mathfrak{f})\left(R(k_{S_0})\widetilde{\phi}\right)\left(\begin{pmatrix} n_{1}^{-1}\mathfrak{f} & n_{1}^{-1}\mathfrak{e}' \\ 1 & 1 \end{pmatrix}\right), ϕ ( ( t f 1 t e 1 ) ) = ϵ χ π ( f ) ( R ( k S 0 ) ϕ ) ( ( n 1 − 1 f 1 n 1 − 1 e ′ 1 ) ) , where k S 0 = k S 0 ( f , e ) k_{S_0}=k_{S_0}(\mathfrak{f},\mathfrak{e}) k S 0 = k S 0 ( f , e ) .
Proof. Let f \mathfrak{f} f and f ~ \widetilde{\mathfrak{f}} f be as in Lemmas 3.2 and 3.3. Since the conductor of ( f , e ) (\mathfrak{f},\mathfrak{e}) ( f , e ) is supported in S 0 S_0 S 0 , both Mellin transforms vanish unless χ \chi χ is unramified outside S 0 S_0 S 0 . For such a character, the hypotheses of Theorem 1.1 give ϵ χ = ϵ [ 1 ] = ϵ \epsilon_{\chi}=\epsilon_{[1]}=\epsilon ϵ χ = ϵ [ 1 ] = ϵ .
Put
b v = ( f v e v 1 1 ) . b_v=\begin{pmatrix} f_v & e_v \\ 1 & 1 \end{pmatrix}. b v = ( f v 1 e v 1 ) .
By Lemma 3.2,
M ( s , f , χ ) = C ψ S 0 ( s , χ ) Λ ( s , π × χ ) ∏ v ∈ S 0 Ψ ( s ; W b v ⋅ ξ v , χ v ) L ( s , π v × χ v ) . \mathcal{M}(s,\mathfrak{f},\chi)=C_{\psi}^{S_0}(s,\chi)\Lambda(s,\pi\times\chi)\prod_{v\in S_0}\frac{\Psi(s;W_{b_v\cdot\xi_v,\chi_v})}{L(s,\pi_v\times\chi_v)}. M ( s , f , χ ) = C ψ S 0 ( s , χ ) Λ ( s , π × χ ) v ∈ S 0 ∏ L ( s , π v × χ v ) Ψ ( s ; W b v ⋅ ξ v , χ v ) .
Similarly, Lemma 3.3 and the local functional equations at the places in S 0 S_0 S 0 give
M ( s , f ~ , χ ) = C ψ S 0 ( s , χ ) ϵ ( s , π × χ , ψ ) Λ ( 1 − s , π ~ × χ − 1 ) \mathcal{M}(s,\widetilde{\mathfrak{f}},\chi)=C_{\psi}^{S_0}(s,\chi)\epsilon(s,\pi\times\chi,\psi)\Lambda(1-s,\widetilde{\pi}\times\chi^{-1}) M ( s , f , χ ) = C ψ S 0 ( s , χ ) ϵ ( s , π × χ , ψ ) Λ ( 1 − s , π × χ − 1 )
× ∏ v ∈ S 0 Ψ ( s ; W b v ⋅ ξ v , χ v ) L ( s , π v × χ v ) . {}\times\prod_{v\in S_0}\frac{\Psi(s;W_{b_v\cdot\xi_v,\chi_v})}{L(s,\pi_v\times\chi_v)}. × v ∈ S 0 ∏ L ( s , π v × χ v ) Ψ ( s ; W b v ⋅ ξ v , χ v ) .
The assumed global functional equation therefore implies
M ( s , f , χ ) = ϵ M ( s , f ~ , χ ) \mathcal{M}(s,\mathfrak{f},\chi)=\epsilon\mathcal{M}(s,\widetilde{\mathfrak{f}},\chi) M ( s , f , χ ) = ϵ M ( s , f , χ )
as meromorphically continued functions. Together with the vanishing when χ \chi χ is ramified outside S 0 S_0 S 0 , this establishes the identity for every unitary idèle class character χ \chi χ . Applying Lemma 3.1 to f \mathfrak{f} f and ϵ f ~ \epsilon\widetilde{\mathfrak{f}} ϵ f gives f ( t ) = ϵ f ~ ( t ) f(t)=\epsilon\widetilde{f}(t) f ( t ) = ϵ f ( t ) , as required. □
In Lemma 3.4, we used only the functional equations for twists by characters unramified outside S 0 S_0 S 0 . To extract information from the remaining functional equations by Mellin inversion, we introduce twists of the auxiliary functions.
Fix a positive divisor m \mathfrak{m} m supported away from n \mathfrak{n} n and S 0 S_0 S 0 , and put S = Supp ( m ) S=\operatorname{Supp}(\mathfrak{m}) S = Supp ( m ) . Henceforth, we call such a modulus admissible . Write n v = ord v ( m ) n_v=\operatorname{ord}_v(\mathfrak{m}) n v = ord v ( m ) and choose the idèle f m \mathfrak{f}_{\mathfrak{m}} f m with components
f m , v = { ϖ v n v , v ∈ S , 1 , v ∉ S . \mathfrak{f}_{\mathfrak{m},v}=
\begin{cases}
\varpi_v^{n_v}, & v\in S,\\
1, & v\notin S.
\end{cases} f m , v = { ϖ v n v , 1 , v ∈ S , v ∈ / S .
For ω ∈ X ( m ) \omega\in X(\mathfrak{m}) ω ∈ X ( m ) , define local vectors by ξ ω , v = ξ v 0 \xi_{\omega,v}=\xi_v^0 ξ ω , v = ξ v 0 for v ∉ S ∪ S 0 v\notin S\cup S_0 v ∈ / S ∪ S 0 , by
ξ ω , v = ∫ o v × ω v ( u ) π v ( ( 1 u ϖ v − n v 1 ) ) ξ v 0 d × u \xi_{\omega,v}=\int_{\mathfrak{o}_v^\times}\omega_v(u)\pi_v\left(\begin{pmatrix}1&u\varpi_v^{-n_v}\\&1\end{pmatrix}\right)\xi_v^0\,d^\times u ξ ω , v = ∫ o v × ω v ( u ) π v ( ( 1 u ϖ v − n v 1 ) ) ξ v 0 d × u
for v ∈ S v\in S v ∈ S , and by ξ ω , v = ξ v \xi_{\omega,v}=\xi_v ξ ω , v = ξ v for v ∈ S 0 v\in S_0 v ∈ S 0 . Here each o v × \mathfrak{o}_v^\times o v × has unit volume, and ξ v \xi_v ξ v is the chosen smooth vector at v ∈ S 0 v\in S_0 v ∈ S 0 . Put ξ ω = ⨂ v ξ ω , v \xi_\omega=\bigotimes_v\xi_{\omega,v} ξ ω = ⨂ v ξ ω , v and define
ϕ ω ( g ) = ϕ ω , m , ξ S 0 ( g ) : = ϕ ξ ω ( g ) = ∑ γ ∈ F × W ξ ω ( ( γ 1 ) g ) . \phi_\omega(g)=\phi_{\omega,\mathfrak{m},\xi_{S_0}}(g):=\phi_{\xi_\omega}(g)=\sum_{\gamma\in F^\times}W_{\xi_\omega}\left(\begin{pmatrix}\gamma&\\&1\end{pmatrix}g\right). ϕ ω ( g ) = ϕ ω , m , ξ S 0 ( g ) := ϕ ξ ω ( g ) = γ ∈ F × ∑ W ξ ω ( ( γ 1 ) g ) .
Let w ∈ A F \mathbf{w} \in\mathbb{A}_{F} w ∈ A F have components w v = ϖ v − n v w_v=\varpi_v^{-n_v} w v = ϖ v − n v for v ∈ S v \in S v ∈ S and w v = 0 w_v=0 w v = 0 otherwise. Write
ω S ( u ) = ∏ v ∈ S ω v ( u v ) , u ∈ ∏ v ∈ S o v × . \omega_S(u)=\prod_{v\in S}\omega_v(u_v),\qquad u\in\prod_{v\in S}o_v^\times. ω S ( u ) = v ∈ S ∏ ω v ( u v ) , u ∈ v ∈ S ∏ o v × .
Then
ϕ ω ( g ) = ∑ γ ∈ F × ∫ ∏ v ∈ S o v × W ξ ( ( γ 1 ) g ( 1 w u 1 ) ) ω S ( u ) d × u . (16) \phi_\omega(g)=\sum_{\gamma\in F^\times}\int_{\prod_{v\in S}o_v^\times}W_\xi\left(\begin{pmatrix}\gamma& \\ 1&\end{pmatrix}g\begin{pmatrix}1&w^u\\ &1\end{pmatrix}\right)\omega_S(u)\,d^\times u. \tag*{(16)} ϕ ω ( g ) = γ ∈ F × ∑ ∫ ∏ v ∈ S o v × W ξ ( ( γ 1 ) g ( 1 w u 1 ) ) ω S ( u ) d × u . (16)
where w u \mathbf{w}u w u has component w v u v w_vu_v w v u v at v ∈ S v\in S v ∈ S and zero elsewhere.
The construction depends only on ω v ∣ o v × \omega_v|_{o_v^\times} ω v ∣ o v × for v ∈ S v\in S v ∈ S . Consequently, it depends only on the class [ ω ] ∈ X ( m ) / X ( 1 ) [\omega]\in X(\mathfrak{m})/X(1) [ ω ] ∈ X ( m ) / X ( 1 ) , and we also write ϕ [ ω ] \phi_{[\omega]} ϕ [ ω ] . Its Fourier expansion is
ϕ [ ω ] ( ( y x 1 ) ) = ∑ γ ∈ F × W ξ ( ( γ y 1 ) ) ( ∏ v ∈ S G ( ω v , ψ v , γ y v ϖ ‾ v − n v ) ) ψ ( γ x ) . \phi_{[\omega]}\left(\begin{pmatrix}y&x\\1&\end{pmatrix}\right)
=\sum_{\gamma\in F^\times}W_\xi\left(\begin{pmatrix}\gamma y&\\1&\end{pmatrix}\right)
\left(\prod_{v\in S}\mathcal{G}(\omega_v,\psi_v,\gamma y_v\overline{\varpi}_v^{-n_v})\right)\psi(\gamma x). ϕ [ ω ] ( ( y 1 x ) ) = γ ∈ F × ∑ W ξ ( ( γ y 1 ) ) ( v ∈ S ∏ G ( ω v , ψ v , γ y v ϖ v − n v ) ) ψ ( γ x ) .
Thus the Fourier coefficients of ϕ [ ω ] \phi_{[\omega]} ϕ [ ω ] are obtained from those of ϕ \phi ϕ by multiplication by the indicated product of Gauss integrals.
We define the twist of the normalized dual function by
ϕ ~ ω ( g ) = ∫ ∏ v ∈ S o v × ϕ ~ ( g ( 1 w u 1 ) ) ω S ( u ) d × u . (17) \widetilde{\phi}_\omega(g)=\int_{\prod_{v\in S}o_v^\times}\widetilde{\phi}\left(g\begin{pmatrix}1&w^u\\&1\end{pmatrix}\right)\omega_S(u)\,d^\times u. \tag*{(17)} ϕ ω ( g ) = ∫ ∏ v ∈ S o v × ϕ ( g ( 1 w u 1 ) ) ω S ( u ) d × u . (17)
In terms of the Whittaker function, this is
ϕ ~ ω ( g ) = ∑ γ ∈ F × ∫ ∏ v ∈ S o v × W ~ ξ ~ ( ( γ 1 ) g ( 1 w u 1 ) ) ω S ( u ) d × u . \widetilde{\phi}_\omega(g)=\sum_{\gamma\in F^\times}\int_{\prod_{v\in S}o_v^\times}\widetilde{W}_{\widetilde{\xi}}\left(\begin{pmatrix}\gamma&\\1&\end{pmatrix}g\begin{pmatrix}1&w^u\\&1\end{pmatrix}\right)\omega_S(u)\,d^\times u. ϕ ω ( g ) = γ ∈ F × ∑ ∫ ∏ v ∈ S o v × W ξ ( ( γ 1 ) g ( 1 w u 1 ) ) ω S ( u ) d × u .
For v ∈ S v\in S v ∈ S , put
ξ ~ ω , v = ∫ o v × ω v ( u ) π v ( ( 1 0 − w v u 1 ) ) ξ ~ v d × u , \widetilde{\xi}_{\omega,v}=\int_{o_v^\times}\omega_v(u)\pi_v\left(\begin{pmatrix}1&0\\-w_vu&1\end{pmatrix}\right)\widetilde{\xi}_v\,d^\times u, ξ ω , v = ∫ o v × ω v ( u ) π v ( ( 1 − w v u 0 1 ) ) ξ v d × u ,
and put ξ ~ ω , v = ξ ~ v \widetilde{\xi}_{\omega,v}=\widetilde{\xi}_v ξ ω , v = ξ v for v ∉ S v\notin S v ∈ / S . Then
ϕ ~ ω ( g ) = ∑ γ ∈ F × W ~ ξ ~ ω ( ( γ 1 ) g ) , ξ ~ ω = ⨂ v ξ ~ ω , v . \widetilde{\phi}_\omega(g)=\sum_{\gamma\in F^\times}\widetilde{W}_{\widetilde{\xi}_\omega}\left(\begin{pmatrix}\gamma&\\1&\end{pmatrix}g\right),
\qquad
\widetilde{\xi}_\omega=\bigotimes_v\widetilde{\xi}_{\omega,v}. ϕ ω ( g ) = γ ∈ F × ∑ W ξ ω ( ( γ 1 ) g ) , ξ ω = v ⨂ ξ ω , v .
This construction also depends only on [ ω ] [\omega] [ ω ] .
Finally, for ( f , e ) (f,e) ( f , e ) and e ′ e' e ′ as in Lemma 3.4, put
f [ ω ] ( t ) = ϕ ω ( ( t f t e 1 ) ) f_{[\omega]}(t)=\phi_\omega\left(\begin{pmatrix}tf&te\\1&\end{pmatrix}\right) f [ ω ] ( t ) = ϕ ω ( ( t f 1 t e ) )
and
f ~ [ ω ] ( t ) = χ π ( m ) χ π ( f ) ( R ( k S 0 ) ϕ ~ ω − 1 ) ( ( n f 2 t − 1 n f 2 t − 1 e ′ 1 ) ) . \widetilde{f}_{[\omega]}(t)=\chi_\pi(\mathfrak{m})\chi_\pi(f)\left(R(k_{S_0})\widetilde{\phi}_{\omega^{-1}}\right)\left(\begin{pmatrix}nf_2t^{-1}&nf_2t^{-1}e'\\1&\end{pmatrix}\right). f [ ω ] ( t ) = χ π ( m ) χ π ( f ) ( R ( k S 0 ) ϕ ω − 1 ) ( ( n f 2 t − 1 1 n f 2 t − 1 e ′ ) ) .
Lemma 3.5. The Mellin transform M ( s , f [ ω ] , χ ) \mathcal{M}(s,f_{[\omega]},\chi) M ( s , f [ ω ] , χ ) vanishes unless χ ∼ ω \chi\sim\omega χ ∼ ω . When this condition holds, it equals
C ψ S 0 ( s , χ ) ∏ v ∉ S ∪ S 0 L ( s , π v × χ v ) ∏ v ∈ S Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) ∏ v ∈ S 0 Ψ ( s ; W b v ⋅ ξ v , χ v ) . C_{\psi}^{S_0}(s,\chi)
\prod_{v\notin S\cup S_0}L(s,\pi_v\times\chi_v)
\prod_{v\in S}\Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v)
\prod_{v\in S_0}\Psi(s;W_{b_v\cdot\xi_v},\chi_v). C ψ S 0 ( s , χ ) v ∈ / S ∪ S 0 ∏ L ( s , π v × χ v ) v ∈ S ∏ Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) v ∈ S 0 ∏ Ψ ( s ; W b v ⋅ ξ v , χ v ) .
where
b v = ( f v e v 1 ) , u v = ( 1 ϖ ‾ v − n v 1 ) , n v = ord v ( m ) . b_v=\begin{pmatrix}f_v&e_v\\&1\end{pmatrix},
\qquad
u_v=\begin{pmatrix}1&\overline{\varpi}_v^{-n_v}\\&1\end{pmatrix},
\qquad
n_v=\operatorname{ord}_v(\mathfrak{m}). b v = ( f v e v 1 ) , u v = ( 1 ϖ v − n v 1 ) , n v = ord v ( m ) .
Proof. We first work with Re ( s ) \operatorname{Re}(s) Re ( s ) sufficiently large that the integrals below converge absolutely. As in the proof of Lemma 3.2 , unfolding gives
M ( s , f [ ω ] , χ ) = ∫ F × \ A F × f [ ω ] ( t ) χ ( t ) ∥ t ∥ s − 1 / 2 d × t = ∏ v Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) . \mathcal{M}(s,f_{[\omega]},\chi)=\int_{F^\times\backslash\mathbb{A}_F^\times}f_{[\omega]}(t)\chi(t)\lVert t\rVert^{s-1/2}\,d^\times t=\prod_v\Psi(s;W_{b_v\cdot\xi_{\omega,v}},\chi_v). M ( s , f [ ω ] , χ ) = ∫ F × \ A F × f [ ω ] ( t ) χ ( t ) ∥ t ∥ s − 1/2 d × t = v ∏ Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) .
Since the conductor of ( f , e ) (f,e) ( f , e ) is supported in S 0 S_0 S 0 , we have f v = 1 f_v=1 f v = 1 and e v ∈ o v e_v\in\mathfrak{o}_v e v ∈ o v for every v ∉ S 0 v\notin S_0 v ∈ / S 0 .
If v ∉ S ∪ S 0 v\notin S\cup S_0 v ∈ / S ∪ S 0 , then ξ ω , v = ξ v = ξ v 0 \xi_{\omega,v}=\xi_v=\xi_v^0 ξ ω , v = ξ v = ξ v 0 and b v = u ( e v ) b_v=u(e_v) b v = u ( e v ) fixes ξ v \xi_v ξ v . Consequently,
Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) = { χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s L ( s , π v × χ v ) , if χ v is unramified , 0 , otherwise . \Psi(s;W_{b_v\cdot\xi_{\omega,v}},\chi_v)=\begin{cases}\chi_v(d_v)^{-1}\lvert d_v\rvert_v^{1/2-s}L(s,\pi_v\times\chi_v), & \text{if }\chi_v\text{ is unramified},\\0, & \text{otherwise}.\end{cases} Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) = { χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s L ( s , π v × χ v ) , 0 , if χ v is unramified , otherwise .
Now let v ∈ S v\in S v ∈ S . The vector ξ v = ξ v 0 \xi_v=\xi_v^0 ξ v = ξ v 0 is spherical, since S S S is disjoint from the conductor of π \pi π . Moreover, b v = u ( e v ) b_v=u(e_v) b v = u ( e v ) fixes ξ v \xi_v ξ v and commutes with every u ( ϖ v − n v u ) u(\varpi_v^{-n_v}u) u ( ϖ v − n v u ) . It therefore fixes ξ ω , v \xi_{\omega,v} ξ ω , v as well. By the definition of ξ ω , v \xi_{\omega,v} ξ ω , v ,
Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) = ∫ F v × ∫ o v × ω v ( u ) W ξ v ( ( t 1 0 1 1 ) ( 1 ϖ v − n v u 0 1 ) ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × u d × t = ∫ F v × ∫ o v × ω v ( u ) W ξ v ( ( t u 0 1 1 ) ( 1 ϖ v − n v 0 1 ) ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × u d × t . (18) \begin{aligned}
\Psi(s;W_{b_v\cdot\xi_{\omega,v}},\chi_v)
&=\int_{F_v^\times}\int_{\mathfrak{o}_v^\times}\omega_v(u)W_{\xi_v}\left(\begin{pmatrix}t_1&0\\1&1\end{pmatrix}\left(1\ \varpi_v^{-n_v}u\atop0\ \ 1\right)\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times u\,d^\times t\\
&=\int_{F_v^\times}\int_{\mathfrak{o}_v^\times}\omega_v(u)W_{\xi_v}\left(\begin{pmatrix}tu&0\\1&1\end{pmatrix}\left(1\ \varpi_v^{-n_v}\atop0\ \ 1\right)\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times u\,d^\times t. \tag*{(18)}
\end{aligned} Ψ ( s ; W b v ⋅ ξ ω , v , χ v ) = ∫ F v × ∫ o v × ω v ( u ) W ξ v ( ( t 1 1 0 1 ) ( 0 1 1 ϖ v − n v u ) ) χ v ( t ) ∥ t ∥ v s − 1/2 d × u d × t = ∫ F v × ∫ o v × ω v ( u ) W ξ v ( ( t u 1 0 1 ) ( 0 1 1 ϖ v − n v ) ) χ v ( t ) ∥ t ∥ v s − 1/2 d × u d × t . (18)
The second equality follows from
( t 0 1 1 ) ( 1 ϖ v − n v u 0 1 ) = ( t u 0 1 1 ) ( 1 ϖ v − n v 0 1 ) ( u − 1 0 0 1 ) , \begin{pmatrix}t&0\\1&1\end{pmatrix}\begin{pmatrix}1&\varpi_v^{-n_v}u\\0&1\end{pmatrix}
=
\begin{pmatrix}tu&0\\1&1\end{pmatrix}\begin{pmatrix}1&\varpi_v^{-n_v}\\0&1\end{pmatrix}\begin{pmatrix}u^{-1}&0\\0&1\end{pmatrix}, ( t 1 0 1 ) ( 1 0 ϖ v − n v u 1 ) = ( t u 1 0 1 ) ( 1 0 ϖ v − n v 1 ) ( u − 1 0 0 1 ) ,
because the last matrix fixes ξ v \xi_v ξ v .
Interchanging the integrals and replacing t t t by t u − 1 tu^{-1} t u − 1 in the inner integral, we obtain
Ψ ( s ; W ξ ω , v , χ v ) = ( ∫ o v × ω v ( u ) χ v − 1 ( u ) d × u ) Ψ ( s ; W u v ⋅ ξ v , χ v ) . \Psi(s;W_{\xi_{\omega,v}},\chi_v)=\left(\int_{\mathfrak{o}_v^\times}\omega_v(u)\chi_v^{-1}(u)\,d^\times u\right)\Psi(s;W_{u_v\cdot\xi_v},\chi_v). Ψ ( s ; W ξ ω , v , χ v ) = ( ∫ o v × ω v ( u ) χ v − 1 ( u ) d × u ) Ψ ( s ; W u v ⋅ ξ v , χ v ) .
Since o v × \mathfrak{o}_v^\times o v × has volume 1 1 1 , character orthogonality gives
∫ o v × ω v ( u ) χ v − 1 ( u ) d × u = { 1 , if χ v ω v − 1 is unramified , 0 , otherwise . \int_{\mathfrak{o}_v^\times}\omega_v(u)\chi_v^{-1}(u)\,d^\times u=\begin{cases}1, & \text{if }\chi_v\omega_v^{-1}\text{ is unramified},\\0, & \text{otherwise}.\end{cases} ∫ o v × ω v ( u ) χ v − 1 ( u ) d × u = { 1 , 0 , if χ v ω v − 1 is unramified , otherwise .
In particular,
Ψ ( s ; W ξ ω , v , χ v ) = Ψ ( s ; W u v ⋅ ξ v , χ v ) (19) \Psi(s;W_{\xi_{\omega,v}},\chi_v)=\Psi(s;W_{u_v\cdot\xi_v},\chi_v) \tag*{(19)} Ψ ( s ; W ξ ω , v , χ v ) = Ψ ( s ; W u v ⋅ ξ v , χ v ) (19)
whenever χ v ω v − 1 \chi_v\omega_v^{-1} χ v ω v − 1 is unramified. Changing variables
Ψ ( s ; W u v ⋅ ξ v , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) . \Psi(s;W_{u_v\cdot\xi_v},\chi_v)=\chi_v(d_v)^{-1}\lvert d_v\rvert_v^{1/2-s}\Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v). Ψ ( s ; W u v ⋅ ξ v , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) .
Finally, if v ∈ S 0 v\in S_0 v ∈ S 0 , then ξ ω , v = ξ v \xi_{\omega,v}=\xi_v ξ ω , v = ξ v , so the local factor remains
Ψ ( s ; W b v ⋅ ξ v , χ v ) . \Psi(s;W_{b_v\cdot\xi_v},\chi_v). Ψ ( s ; W b v ⋅ ξ v , χ v ) .
Since ω \omega ω is unramified outside S ∪ S 0 S\cup S_0 S ∪ S 0 , the conditions obtained above are together equivalent to χ ω − 1 \chi\omega^{-1} χ ω − 1 being unramified outside S 0 S_0 S 0 , that is, to χ ∼ ω \chi\sim\omega χ ∼ ω . Multiplying the local factors proves the assertion in the domain of absolute convergence, and hence wherever the Mellin transform admits meromorphic continuation. □
Thus the unipotent average defining ϕ [ ω ] \phi_{[\omega]} ϕ [ ω ] isolates the twists by characters χ ∼ ω \chi\sim\omega χ ∼ ω .
Lemma 3.6. The Mellin transform M ( s , f ~ [ ω ] , χ ) \mathcal{M}(s,\widetilde{f}_{[\omega]},\chi) M ( s , f [ ω ] , χ ) vanishes unless χ ∼ ω \chi\sim\omega χ ∼ ω . When this condition holds, it equals
C ψ S 0 ( s , χ ) ∏ v ∉ S ∪ S 0 ( ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π ~ v × χ v − 1 ) ) C_{\psi}^{S_0}(s,\chi)\prod_{v\notin S\cup S_0}\left(\epsilon(s,\pi_v\times\chi_v,\psi_v)L(1-s,\widetilde{\pi}_v\times\chi_v^{-1})\right) C ψ S 0 ( s , χ ) v ∈ / S ∪ S 0 ∏ ( ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π v × χ v − 1 ) ) × ∏ v ∈ S ( ( χ π v χ v 2 ) ( ϖ ‾ v ) q v − n v ( 2 s − 1 ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W ~ b v ⋅ ξ v , χ v − 1 ) ) {}\times\prod_{v\in S}\left((\chi_{\pi_v}\chi_v^2)(\overline{\varpi}_v)q_v^{-n_v(2s-1)}(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}\Psi\left(1-s;L(d_v)\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1}\right)\right) × v ∈ S ∏ ( ( χ π v χ v 2 ) ( ϖ v ) q v − n v ( 2 s − 1 ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W b v ⋅ ξ v , χ v − 1 ) ) × ∏ v ∈ S 0 Ψ ( 1 − s ; W ~ b v ⋅ ξ v , χ v − 1 ) . {}\times\prod_{v\in S_0}\Psi\left(1-s;\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1}\right). × v ∈ S 0 ∏ Ψ ( 1 − s ; W b v ⋅ ξ v , χ v − 1 ) . where
u ‾ v = ( − 1 e v ϖ v n v 1 ) , n v = ord v ( m ) . \overline{u}_v=\left(-\frac{1}{e_v} \varpi_v^{n_v}\ 1\right),\qquad n_v=\operatorname{ord}_v(\mathfrak{m}). u v = ( − e v 1 ϖ v n v 1 ) , n v = ord v ( m ) .
Proof. We first work with Re ( s ) \operatorname{Re}(s) Re ( s ) sufficiently negative that the integrals below converge absolutely. Unfolding gives
M ( s , f ~ [ ω ] , χ ) = χ π ( m ) χ π ( f ) ∏ v I v . \mathcal{M}(s,\widetilde{f}_{[\omega]},\chi)=\chi_{\pi}(\mathfrak{m})\chi_{\pi}(\mathfrak{f})\prod_v I_v. M ( s , f [ ω ] , χ ) = χ π ( m ) χ π ( f ) v ∏ I v .
where
I v = { ∫ F v × W ~ ξ ω − 1 , v ( ( n v f m , v 2 t − 1 1 ) b v ′ ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × t , v ∉ S 0 , ∫ F v × W ~ ξ v ( ( t − 1 1 ) b v ′ k v ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × t , v ∈ S 0 . I_v=
\begin{cases}
\int_{F_v^\times}\widetilde{W}_{\xi\omega^{-1},v}\left(\begin{pmatrix}n_vf_{m,v}^2t^{-1}& \\ &1\end{pmatrix}b'_v\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times t,&v\notin S_0,\\
\int_{F_v^\times}\widetilde{W}_{\xi_v}\left(\begin{pmatrix}t^{-1}& \\ &1\end{pmatrix}b'_vk_v\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times t,&v\in S_0.
\end{cases} I v = ⎩ ⎨ ⎧ ∫ F v × W ξ ω − 1 , v ( ( n v f m , v 2 t − 1 1 ) b v ′ ) χ v ( t ) ∥ t ∥ v s − 1/2 d × t , ∫ F v × W ξ v ( ( t − 1 1 ) b v ′ k v ) χ v ( t ) ∥ t ∥ v s − 1/2 d × t , v ∈ / S 0 , v ∈ S 0 .
Here b v ′ b'_v b v ′ and
k v = ( ( 1 + e v e v ′ ) / f v − e v ′ − e v f v ) k_v=\begin{pmatrix}(1+e_ve'_v)/f_v&-e'_v\\-e_v&f_v\end{pmatrix} k v = ( ( 1 + e v e v ′ ) / f v − e v − e v ′ f v )
are as in (3.8).
Suppose first that v ∉ S ∪ S 0 v\notin S\cup S_0 v ∈ / S ∪ S 0 . Then ξ ω − 1 , v ~ = ξ ~ v \widetilde{\xi\omega^{-1},v}=\widetilde{\xi}_v ξ ω − 1 , v = ξ v , f m , v = 1 f_{m,v}=1 f m , v = 1 , and f v = 1 f_v=1 f v = 1 . Since e v ′ ∈ o v e'_v\in\mathfrak{o}_v e v ′ ∈ o v , we have
t v ′ b v ′ − 1 ⋅ ξ ~ v = ( 1 − e v ′ 1 ) ( ϖ v n v 1 ) ⋅ ξ v = ( ϖ v n v 1 ) ( 1 − n v e v ′ 1 ) ⋅ ξ ~ v = ξ ~ v . t'_vb_v^{\prime-1}\cdot\widetilde{\xi}_v=
\begin{pmatrix}1&\\-e'_v&1\end{pmatrix}
\begin{pmatrix}\varpi_v^{n_v}&\\&1\end{pmatrix}\cdot\xi_v
=
\begin{pmatrix}\varpi_v^{n_v}&\\&1\end{pmatrix}
\begin{pmatrix}1&\\-n_ve'_v&1\end{pmatrix}\cdot\widetilde{\xi}_v
=\widetilde{\xi}_v. t v ′ b v ′ − 1 ⋅ ξ v = ( 1 − e v ′ 1 ) ( ϖ v n v 1 ) ⋅ ξ v = ( ϖ v n v 1 ) ( 1 − n v e v ′ 1 ) ⋅ ξ v = ξ v .
The last equality follows from the K 1 , v ( p v n ( π v ) ) K_{1,v}(\mathfrak{p}_v^{n(\pi_v)}) K 1 , v ( p v n ( π v ) ) -invariance of ξ v \xi_v ξ v . Consequently,
I v = ∫ F v × W ~ ξ ~ v ( ( ϖ v n v t − 1 1 ) ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × t = ∫ F v × W ~ ξ ~ v ( ( t − 1 1 ) ) χ v ( t ) ∥ t ∥ v s − 1 / 2 d × t = Ψ ( 1 − s ; W ~ ξ ~ v , χ v − 1 ) . \begin{aligned}
I_v=\int_{F_v^\times}\widetilde{W}_{\widetilde{\xi}_v}\left(\begin{pmatrix}\varpi_v^{n_v}t^{-1}&\\&1\end{pmatrix}\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times t \\
&=\int_{F_v^\times}\widetilde{W}_{\widetilde{\xi}_v}\left(\begin{pmatrix}t^{-1}&\\&1\end{pmatrix}\right)\chi_v(t)\lVert t\rVert_v^{s-1/2}\,d^\times t \\
&=\Psi(1-s;\widetilde{W}_{\widetilde{\xi}_v},\chi_v^{-1}).
\end{aligned} I v = ∫ F v × W ξ v ( ( ϖ v n v t − 1 1 ) ) χ v ( t ) ∥ t ∥ v s − 1/2 d × t = ∫ F v × W ξ v ( ( t − 1 1 ) ) χ v ( t ) ∥ t ∥ v s − 1/2 d × t = Ψ ( 1 − s ; W ξ v , χ v − 1 ) .
As in §3.1, this vanishes unless χ v \chi_v χ v is unramified, and in that case it equals
χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π ~ v × χ v − 1 ) . \chi_v(d_v)^{-1}|d_v|_v^{1/2-s}\epsilon(s,\pi_v\times\chi_v,\psi_v)L(1-s,\widetilde{\pi}_v\times\chi_v^{-1}). χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s ϵ ( s , π v × χ v , ψ v ) L ( 1 − s , π v × χ v − 1 ) .
Next suppose that v ∈ S 0 v\in S_0 v ∈ S 0 . Then n v = f m , v = 1 n_v=f_{m,v}=1 n v = f m , v = 1 and ξ ω − 1 , v ~ = ξ ~ v \widetilde{\xi\omega^{-1},v}=\widetilde{\xi}_v ξ ω − 1 , v = ξ v . The identity
b v ′ k v = w b v w − 1 = f v t v ′ b v ′ − 1 b'_vk_v=wb_vw^{-1}=f_v\,t'_v b_v^{\prime-1} b v ′ k v = w b v w − 1 = f v t v ′ b v ′ − 1
gives
W ~ ξ v ( g b v ′ k v ) = χ π v ( f v ) − 1 W ~ b v ⋅ ξ v ( g ) . \widetilde{W}_{\xi_v}(gb'_v k_v)=\chi_{\pi_v}(f_v)^{-1}\widetilde{W}_{b_v\cdot\xi_v}(g). W ξ v ( g b v ′ k v ) = χ π v ( f v ) − 1 W b v ⋅ ξ v ( g ) .
Changing variables y = t − 1 y=t^{-1} y = t − 1 therefore yields
I v = χ π v ( f v ) − 1 Ψ ( 1 − s ; W ~ b v ⋅ ξ v , χ v − 1 ) . I_v=\chi_{\pi_v}(f_v)^{-1}\Psi\left(1-s;\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1}\right). I v = χ π v ( f v ) − 1 Ψ ( 1 − s ; W b v ⋅ ξ v , χ v − 1 ) .
This argument applies to every place in S 0 S_0 S 0 and requires no compact invariance of ξ v \xi_v ξ v .
Finally, suppose that v ∈ S v\in S v ∈ S . Here n v = f v = 1 n_v=f_v=1 n v = f v = 1 , and
ξ ~ ω − 1 , v = ∫ o v × ω v ( u ) − 1 π v ( ( 1 − ϖ v n v 1 ) u ) ξ v d × u . \widetilde{\xi}_{\omega^{-1},v}=\int_{\mathfrak{o}_v^\times}\omega_v(u)^{-1}\pi_v\left(\begin{pmatrix} \frac{1}{-\varpi_v^{n_v}} & 1 \end{pmatrix}u\right)\xi_v\,d^\times u. ξ ω − 1 , v = ∫ o v × ω v ( u ) − 1 π v ( ( − ϖ v n v 1 1 ) u ) ξ v d × u .
The matrix
t v ′ − 1 = ( − 1 e v ′ 1 ) t'_v{}^{-1}=\left(\begin{matrix}-\frac{1}{e'_v}&1\end{matrix}\right) t v ′ − 1 = ( − e v ′ 1 1 )
fixes the spherical vector ξ v \xi_v ξ v and commutes with the lower unipotent matrices in this integral. It therefore fixes ξ ~ ω − 1 , v \widetilde{\xi}_{\omega^{-1},v} ξ ω − 1 , v . Using the relation between right translation and Whittaker duality, we obtain
I v = ∫ F v × ∫ o v × ω v ( u ) − 1 W ~ ξ v ( ( f m , v 2 t − 1 1 ) ( 1 ϖ v − n v u 1 ) ) χ v ( t ) ∣ t ∣ v s − 1 / 2 d × u d × t . I_v=\int_{F_v^\times}\int_{\mathfrak{o}_v^\times}\omega_v(u)^{-1}\widetilde{W}_{\xi_v}\left(\left(\begin{matrix}f_{m,v}^2t^{-1}&\\&1\end{matrix}\right)\left(\begin{matrix}1&\varpi_v^{-n_v}u\\&1\end{matrix}\right)\right)\chi_v(t)\lvert t\rvert_v^{s-1/2}\,d^\times u\,d^\times t. I v = ∫ F v × ∫ o v × ω v ( u ) − 1 W ξ v ( ( f m , v 2 t − 1 1 ) ( 1 ϖ v − n v u 1 ) ) χ v ( t ) ∣ t ∣ v s − 1/2 d × u d × t .
Since W ~ ξ v \widetilde{W}_{\xi_v} W ξ v is right invariant under ( u − 1 1 ) \left(\begin{matrix}u^{-1}&\\&1\end{matrix}\right) ( u − 1 1 ) for u ∈ o v × u\in\mathfrak{o}_v^\times u ∈ o v × , the matrix identity used in (3.12), followed by the substitution t = r u t=ru t = r u , gives
I v = ( ∫ o v × ω v ( u ) − 1 χ v ( u ) d × u ) × ∫ F v × W ~ ξ v ( ( f m , v 2 r − 1 1 ) u v ) χ v ( r ) ∣ r ∣ v s − 1 / 2 d × r . I_v=\left(\int_{\mathfrak{o}_v^\times}\omega_v(u)^{-1}\chi_v(u)\,d^\times u\right)
\times\int_{F_v^\times}\widetilde{W}_{\xi_v}\left(\left(\begin{matrix}f_{m,v}^2r^{-1}&\\&1\end{matrix}\right)u_v\right)\chi_v(r)\lvert r\rvert_v^{s-1/2}\,d^\times r. I v = ( ∫ o v × ω v ( u ) − 1 χ v ( u ) d × u ) × ∫ F v × W ξ v ( ( f m , v 2 r − 1 1 ) u v ) χ v ( r ) ∣ r ∣ v s − 1/2 d × r .
The unit integral is zero unless χ v ω v − 1 \chi_v\omega_v^{-1} χ v ω v − 1 is unramified, in which case it is 1. Under this condition, using
R ( u v ) W ~ ξ v = W ~ u ‾ v ⋅ ξ v R(u_v)\widetilde{W}_{\xi_v}=\widetilde{W}_{\overline{u}_v\cdot\xi_v} R ( u v ) W ξ v = W u v ⋅ ξ v
and changing variables y = f m , v 2 r − 1 y=f_{m,v}^2r^{-1} y = f m , v 2 r − 1 , we conclude that
I v = χ v ( ϖ v 2 n v ) q v − n v ( 2 s − 1 ) Ψ ( 1 − s ; W ~ u ‾ v ⋅ ξ v , χ v − 1 ) . I_v=\chi_v(\varpi_v^{2n_v})q_v^{-n_v(2s-1)}\Psi\left(1-s;\widetilde{W}_{\overline{u}_v\cdot\xi_v},\chi_v^{-1}\right). I v = χ v ( ϖ v 2 n v ) q v − n v ( 2 s − 1 ) Ψ ( 1 − s ; W u v ⋅ ξ v , χ v − 1 ) .
Put
C v ( s , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1 / 2 − s . C_v(s,\chi_v)=\chi_v(d_v)^{-1}\lvert d_v\rvert_v^{1/2-s}. C v ( s , χ v ) = χ v ( d v ) − 1 ∣ d v ∣ v 1/2 − s .
Arguing as in Lemma 3.3, we have
Ψ ( 1 − s ; W ~ u ‾ v ⋅ ξ v , χ v − 1 ) = C v ( s , χ v ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W ~ u ‾ v ⋅ ξ v , χ v − 1 ) . \Psi\left(1-s;\widetilde{W}_{\overline{u}_v\cdot\xi_v},\chi_v^{-1}\right)
=C_v(s,\chi_v)(\chi_{\pi_v}\chi_v^2)(d_v)\lvert d_v\rvert_v^{2s-1}\Psi\left(1-s;L(d_v)\widetilde{W}_{\overline{u}_v\cdot\xi_v},\chi_v^{-1}\right). Ψ ( 1 − s ; W u v ⋅ ξ v , χ v − 1 ) = C v ( s , χ v ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 Ψ ( 1 − s ; L ( d v ) W u v ⋅ ξ v , χ v − 1 ) .
The local vanishing conditions are together equivalent to χ ∼ ω \chi\sim\omega χ ∼ ω . Moreover,
χ π ( f ) = ∏ v ∈ S 0 χ π v ( f v ) , χ π ( m ) = ∏ v ∈ S χ π v ( ϖ v n v ) . \chi_\pi(f)=\prod_{v\in S_0}\chi_{\pi_v}(f_v),\qquad
\chi_\pi(\mathfrak{m})=\prod_{v\in S}\chi_{\pi_v}(\varpi_v^{n_v}). χ π ( f ) = v ∈ S 0 ∏ χ π v ( f v ) , χ π ( m ) = v ∈ S ∏ χ π v ( ϖ v n v ) .
Thus the factors involving f v f_v f v cancel, and multiplying the remaining local factors proves the stated formula in the domain of absolute convergence. The identity then holds wherever the Mellin transform admits meromorphic continuation. □
Remark. For v ∈ S v \in S v ∈ S , the Whittaker functions appearing in Lemmas 3.5 and 3.6 are not, in general, dual to each other. Indeed, since left and right translations commute, the Whittaker function in Lemma 3.5 is
L ( d v ) W u v ⋅ ξ v = R ( u v ) L ( d v ) W ξ v . L(\mathrm{d}_v)W_{u_v\cdot\xi_v}=R(u_v)L(\mathrm{d}_v)W_{\xi_v}. L ( d v ) W u v ⋅ ξ v = R ( u v ) L ( d v ) W ξ v .
Its dual is
R ( u ‾ v ) L ( d ‾ v ) W ξ v , R(\overline{u}_v)L(\overline{\mathrm{d}}_v)W_{\xi_v}, R ( u v ) L ( d v ) W ξ v ,
whereas the Whittaker function in Lemma 3.6 is
L ( d v ) W u ‾ v ⋅ ξ v = R ( u v ) L ( d ‾ v ) W ξ v . L(\mathrm{d}_v)W_{\overline{u}_v\cdot\xi_v}=R(u_v)L(\overline{\mathrm{d}}_v)W_{\xi_v}. L ( d v ) W u v ⋅ ξ v = R ( u v ) L ( d v ) W ξ v .
Thus the local functional equation (2.5) does not directly relate the two local integrals. We derive the required relation below.
The local functional equation for v ∈ S v \in S v ∈ S
Suppose n v = ord v ( m ) n_v=\operatorname{ord}_v(\mathfrak{m}) n v = ord v ( m ) , u v = u ( ϖ ‾ v − n v ) u_v=u(\overline{\varpi}_v^{-n_v}) u v = u ( ϖ v − n v ) and u ‾ v = u ‾ ( − ϖ ‾ v − n v ) \overline{u}_v=\overline{u}(-\overline{\varpi}_v^{-n_v}) u v = u ( − ϖ v − n v ) as before. In this section, we only consider characters χ v \chi_v χ v so that χ v ω v − 1 \chi_v\omega_v^{-1} χ v ω v − 1 is unramified. In particular, this means n v ≥ m v = cond ( χ v ) n_v \geq m_v=\operatorname{cond}(\chi_v) n v ≥ m v = cond ( χ v ) . Since S S S is disjoint from the support of n \mathfrak{n} n , we also have that π v \pi_v π v is unramified. Put
D m ( s , π v × χ v ) = χ v ( ϖ ‾ v m v − n v ) G ( χ v , ψ v , d v − 1 ϖ ‾ v − m v ) Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) L ( s , π v × χ v ) . D_{\mathfrak{m}}(s,\pi_v\times\chi_v)=
\frac{\chi_v(\overline{\varpi}_v^{m_v-n_v})}
{\mathcal{G}(\chi_v,\psi_v,\mathrm{d}_v^{-1}\overline{\varpi}_v^{-m_v})}
\frac{\Psi(s;L(\mathrm{d}_v)W_{u_v\cdot\xi_v},\chi_v)}
{L(s,\pi_v\times\chi_v)}. D m ( s , π v × χ v ) = G ( χ v , ψ v , d v − 1 ϖ v − m v ) χ v ( ϖ v m v − n v ) L ( s , π v × χ v ) Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) .
Here L ( d v ) W ξ v L(\mathrm{d}_v)W_{\xi_v} L ( d v ) W ξ v is the normalized essential Whittaker function for the additive character x ↦ ψ v ( d v − 1 x ) x\mapsto\psi_v(\mathrm{d}_v^{-1}x) x ↦ ψ v ( d v − 1 x ) , whose conductor is o v \mathfrak{o}_v o v .
The normalizing factor can also be written as
χ v ( ϖ ‾ v − n v ) χ v ( ϖ ‾ v − m v ) G ( χ v , ψ v , ϖ ‾ v − m v − δ v ) . \frac{\chi_v(\overline{\varpi}_v^{-n_v})}
{\chi_v(\overline{\varpi}_v^{-m_v})\mathcal{G}(\chi_v,\psi_v,\overline{\varpi}_v^{-m_v-\delta_v})}. χ v ( ϖ v − m v ) G ( χ v , ψ v , ϖ v − m v − δ v ) χ v ( ϖ v − n v ) .
For m v ≥ 1 m_v\geq1 m v ≥ 1 ,
χ v ( ϖ ‾ v − m v ) G ( χ v , ψ v , ϖ ‾ v − m v − δ v ) = q v − m v / 2 ( 1 − q v − 1 ) − 1 χ v ( d v ) ϵ ( 1 2 , χ v − 1 , ψ v ) . \chi_v(\overline{\varpi}_v^{-m_v})\mathcal{G}(\chi_v,\psi_v,\overline{\varpi}_v^{-m_v-\delta_v})
=q_v^{-m_v/2}(1-q_v^{-1})^{-1}\chi_v(\mathrm{d}_v)\epsilon\left(\frac{1}{2},\chi_v^{-1},\psi_v\right). χ v ( ϖ v − m v ) G ( χ v , ψ v , ϖ v − m v − δ v ) = q v − m v /2 ( 1 − q v − 1 ) − 1 χ v ( d v ) ϵ ( 2 1 , χ v − 1 , ψ v ) .
When m v = 0 m_v=0 m v = 0 , the denominator equals 1.
Lemma 3.7. D m ( s , π v × χ v ) D_{\mathfrak{m}}(s,\pi_v\times\chi_v) D m ( s , π v × χ v ) belongs to C [ q v − s ] \mathbb{C}[q_v^{-s}] C [ q v − s ] .
Proof. We initially work in a right half-plane of absolute convergence.
Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = ∫ F v × ( L ( d v ) W ξ v ) ( a ( t ) ) ψ v ( d v − 1 ϖ ‾ v − n v t ) χ v ( t ) ∣ t ∣ v s − 1 / 2 d × t = ∑ r ≥ 0 G ( χ v , ψ v , ϖ ‾ v r − n v − δ v ) ( L ( d v ) W ξ v ) ( a ( ϖ ‾ v r ) ) χ v ( ϖ ‾ v r ) q v − r ( s − 1 / 2 ) . \begin{aligned}
\Psi(s;L(\mathrm{d}_v)W_{u_v\cdot\xi_v},\chi_v)
=\int_{F_v^\times}(L(\mathrm{d}_v)W_{\xi_v})(a(t))\psi_v(\mathrm{d}_v^{-1}\overline{\varpi}_v^{-n_v}t)\chi_v(t)|t|_v^{s-1/2}\,\mathrm{d}^\times t \\
&=\sum_{r\geq0}\mathcal{G}(\chi_v,\psi_v,\overline{\varpi}_v^{r-n_v-\delta_v})(L(\mathrm{d}_v)W_{\xi_v})(a(\overline{\varpi}_v^r))\chi_v(\overline{\varpi}_v^r)q_v^{-r(s-1/2)}.
\end{aligned} Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = ∫ F v × ( L ( d v ) W ξ v ) ( a ( t )) ψ v ( d v − 1 ϖ v − n v t ) χ v ( t ) ∣ t ∣ v s − 1/2 d × t = r ≥ 0 ∑ G ( χ v , ψ v , ϖ v r − n v − δ v ) ( L ( d v ) W ξ v ) ( a ( ϖ v r )) χ v ( ϖ v r ) q v − r ( s − 1/2 ) .
For r ≥ 0 r\geq0 r ≥ 0 , put
λ r = q v r / 2 ( L ( d v ) W ξ v ) ( a ( ϖ ‾ v r ) ) , \lambda_r=q_v^{r/2}(L(\mathrm{d}_v)W_{\xi_v})(a(\overline{\varpi}_v^r)), λ r = q v r /2 ( L ( d v ) W ξ v ) ( a ( ϖ v r )) ,
and set λ − 1 = 0 \lambda_{-1}=0 λ − 1 = 0 . Thus λ 0 = 1 \lambda_0=1 λ 0 = 1 . Suppose first that m v = 0 m_v=0 m v = 0 , and put γ v = χ v ( ϖ v ) \gamma_v=\chi_v(\varpi_v) γ v = χ v ( ϖ v ) . Since x ↦ ψ v ( d v − 1 x ) x\mapsto\psi_v(d_v^{-1}x) x ↦ ψ v ( d v − 1 x ) has conductor o v \mathfrak{o}_v o v , we have
G ( χ v , ψ v , ϖ v r − n v − δ v ) = { 1 , r ≥ n v , − 1 q v − 1 , r = n v − 1 , 0 , r < n v − 1. \mathcal{G}(\chi_v,\psi_v,\varpi_v^{r-n_v-\delta_v})=
\begin{cases}
1, & r\ge n_v,\\
-\dfrac{1}{q_v-1}, & r=n_v-1,\\
0, & r<n_v-1.
\end{cases} G ( χ v , ψ v , ϖ v r − n v − δ v ) = ⎩ ⎨ ⎧ 1 , − q v − 1 1 , 0 , r ≥ n v , r = n v − 1 , r < n v − 1.
Moreover,
L ( s , π v × χ v ) = ∑ r ≥ 0 λ r γ v r q v − r s . L(s,\pi_v\times\chi_v)=\sum_{r\ge0}\lambda_r\gamma_v^r q_v^{-rs}. L ( s , π v × χ v ) = r ≥ 0 ∑ λ r γ v r q v − r s .
Consequently,
Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = − q v q v − 1 λ n v − 1 γ v n v − 1 q v − ( n v − 1 ) s + L ( s , π v × χ v ) − ∑ r = 0 n v − 2 λ r γ v r q v − r s . (20) \begin{aligned}
\Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v) \\
&=-\frac{q_v}{q_v-1}\lambda_{n_v-1}\gamma_v^{n_v-1}q_v^{-(n_v-1)s}+L(s,\pi_v\times\chi_v)-\sum_{r=0}^{n_v-2}\lambda_r\gamma_v^r q_v^{-rs}. \tag*{(20)}
\end{aligned} Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = − q v − 1 q v λ n v − 1 γ v n v − 1 q v − ( n v − 1 ) s + L ( s , π v × χ v ) − r = 0 ∑ n v − 2 λ r γ v r q v − r s . (20)
The final sum is empty when n v = 1 n_v=1 n v = 1 . Since the reciprocal of L ( s , π v × χ v ) L(s,\pi_v\times\chi_v) L ( s , π v × χ v ) is a polynomial in q v − s q_v^{-s} q v − s , and the normalizing factor defining D m D_m D m is independent of s s s , this proves the assertion when m v = 0 m_v=0 m v = 0 .
Now suppose that m v ≥ 1 m_v\ge1 m v ≥ 1 . The Gauss integral
G ( χ v , ψ v , ϖ v r − n v − δ v ) \mathcal{G}(\chi_v,\psi_v,\varpi_v^{r-n_v-\delta_v}) G ( χ v , ψ v , ϖ v r − n v − δ v )
vanishes unless r = n v − m v r=n_v-m_v r = n v − m v . Hence
Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = G ( χ v , ψ v , ϖ v − m v − δ v ) ( L ( d v ) W ξ v ) ( a ( ϖ v n v − m v ) ) \Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v)
=\mathcal{G}(\chi_v,\psi_v,\varpi_v^{-m_v-\delta_v})(L(d_v)W_{\xi_v})(a(\varpi_v^{n_v-m_v})) Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = G ( χ v , ψ v , ϖ v − m v − δ v ) ( L ( d v ) W ξ v ) ( a ( ϖ v n v − m v ))
× χ v ( ϖ v n v − m v ) q v − ( n v − m v ) ( s − 1 2 ) . \qquad\qquad\times\chi_v(\varpi_v^{n_v-m_v})q_v^{-(n_v-m_v)(s-\frac{1}{2})}. × χ v ( ϖ v n v − m v ) q v − ( n v − m v ) ( s − 2 1 ) .
Since π v \pi_v π v is unramified and χ v \chi_v χ v is ramified, L ( s , π v × χ v ) = 1 L(s,\pi_v\times\chi_v)=1 L ( s , π v × χ v ) = 1 . The normalizing factors therefore cancel, giving
D m ( s , π v × χ v ) = λ n v − m v q v − ( n v − m v ) s . (21) D_m(s,\pi_v\times\chi_v)=\lambda_{n_v-m_v}q_v^{-(n_v-m_v)s}. \tag*{(21)} D m ( s , π v × χ v ) = λ n v − m v q v − ( n v − m v ) s . (21)
Since n v ≥ m v n_v\ge m_v n v ≥ m v , this is a polynomial in q v − s q_v^{-s} q v − s .
Lemma 3.8. One has the functional equation
D m ( s , π v × χ v ) = q v ( n v − m v ) ( 1 − 2 s ) χ π v ( ϖ v n v − m v ) D m ( 1 − s , π ~ v × χ v − 1 ) . D_m(s,\pi_v\times\chi_v)=q_v^{(n_v-m_v)(1-2s)}\chi_{\pi_v}(\varpi_v^{n_v-m_v})D_m(1-s,\widetilde{\pi}_v\times\chi_v^{-1}). D m ( s , π v × χ v ) = q v ( n v − m v ) ( 1 − 2 s ) χ π v ( ϖ v n v − m v ) D m ( 1 − s , π v × χ v − 1 ) .
Proof. Since π v \pi_v π v is unramified, the normalized spherical Whittaker functions satisfy
L ( d v ) W π ~ v = L ( d v ) W ξ v . L(d_v)W_{\widetilde{\pi}_v}=L(d_v)W_{\xi_v}. L ( d v ) W π v = L ( d v ) W ξ v .
Moreover,
( L ( d v ) W π ~ v ) ( g ) = χ π v ( det g ) − 1 ( L ( d v ) W ξ v ) ( g ) . (L(d_v)W_{\widetilde{\pi}_v})(g)=\chi_{\pi_v}(\det g)^{-1}(L(d_v)W_{\xi_v})(g). ( L ( d v ) W π v ) ( g ) = χ π v ( det g ) − 1 ( L ( d v ) W ξ v ) ( g ) .
If m v ≥ 1 m_v\ge1 m v ≥ 1 , put k = n v − m v k=n_v-m_v k = n v − m v . By (3.15) and the preceding identity,
D m ( 1 − s , π ~ v × χ v − 1 ) = χ π v ( ϖ v k ) − 1 λ k q v − k ( 1 − s ) . D_m(1-s,\widetilde{\pi}_v\times\chi_v^{-1})=\chi_{\pi_v}(\varpi_v^k)^{-1}\lambda_kq_v^{-k(1-s)}. D m ( 1 − s , π v × χ v − 1 ) = χ π v ( ϖ v k ) − 1 λ k q v − k ( 1 − s ) .
Multiplying by q v k ( 1 − 2 s ) χ π v ( ϖ v k ) q_v^{k(1-2s)}\chi_{\pi_v}(\varpi_v^k) q v k ( 1 − 2 s ) χ π v ( ϖ v k ) gives λ k q v − k s = D m ( s , π v × χ v ) \lambda_kq_v^{-ks}=D_m(s,\pi_v\times\chi_v) λ k q v − k s = D m ( s , π v × χ v ) , as required.
Suppose now that m v = 0 m_v=0 m v = 0 , and write n = n v ≥ 1 n=n_v\ge1 n = n v ≥ 1 . Let { α v , β v } \{\alpha_v,\beta_v\} { α v , β v } be the Satake parameters associated with π v \pi_v π v . With the notation of Lemma 3.7, we have
λ r = ∑ j = 0 r α v r − j β v j , r ≥ 0. \lambda_r=\sum_{j=0}^{r}\alpha_v^{r-j}\beta_v^j,\quad r\ge0. λ r = j = 0 ∑ r α v r − j β v j , r ≥ 0.
We have
L ( s , π v × χ v ) = ( 1 − λ 1 γ v q v − s + χ π v ( ϖ v ) γ v 2 q v − 2 s ) − 1 L(s,\pi_v\times\chi_v)=\left(1-\lambda_1\gamma_vq_v^{-s}+\chi_{\pi_v}(\varpi_v)\gamma_v^2q_v^{-2s}\right)^{-1} L ( s , π v × χ v ) = ( 1 − λ 1 γ v q v − s + χ π v ( ϖ v ) γ v 2 q v − 2 s ) − 1
with χ π v ( ϖ v ) = α v β v \chi_{\pi_v}(\varpi_v)=\alpha_v\beta_v χ π v ( ϖ v ) = α v β v , and the recurrence relation λ r λ 1 = λ r + 1 + χ π v ( ϖ v ) λ r − 1 \lambda_r\lambda_1=\lambda_{r+1}+\chi_{\pi_v}(\varpi_v)\lambda_{r-1} λ r λ 1 = λ r + 1 + χ π v ( ϖ v ) λ r − 1 for r ≥ 1 r\ge1 r ≥ 1 . It follows from the latter relation that
∑ r = 0 n − 2 λ r γ v r q v − r s = ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s + 1 ) ∑ r = 0 ∞ λ r γ v r q v − r s = ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s + 1 ) L ( s , π v × χ v ) . \begin{aligned}
\sum_{r=0}^{n-2}\lambda_r\gamma_v^rq_v^{-rs}
=\left(\lambda_{n-2}\gamma_v^n\chi_{\pi_v}(\varpi_v)q_v^{-ns}-\lambda_{n-1}\gamma_v^{n-1}q_v^{-(n-1)s}+1\right)\sum_{r=0}^{\infty}\lambda_r\gamma_v^rq_v^{-rs} \\
&=\left(\lambda_{n-2}\gamma_v^n\chi_{\pi_v}(\varpi_v)q_v^{-ns}-\lambda_{n-1}\gamma_v^{n-1}q_v^{-(n-1)s}+1\right)L(s,\pi_v\times\chi_v).
\end{aligned} r = 0 ∑ n − 2 λ r γ v r q v − r s = ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s + 1 ) r = 0 ∑ ∞ λ r γ v r q v − r s = ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s + 1 ) L ( s , π v × χ v ) .
Inserting this into (20) , we obtain
Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = − q v q v − 1 λ n − 1 γ v n − 1 q v − ( n − 1 ) s − L ( s , π v × χ v ) ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s ) . \Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v)
=-\frac{q_v}{q_v-1}\lambda_{n-1}\gamma_v^{n-1}q_v^{-(n-1)s}
-L(s,\pi_v\times\chi_v)\left(\lambda_{n-2}\gamma_v^n\chi_{\pi_v}(\varpi_v)q_v^{-ns}-\lambda_{n-1}\gamma_v^{n-1}q_v^{-(n-1)s}\right). Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) = − q v − 1 q v λ n − 1 γ v n − 1 q v − ( n − 1 ) s − L ( s , π v × χ v ) ( λ n − 2 γ v n χ π v ( ϖ v ) q v − n s − λ n − 1 γ v n − 1 q v − ( n − 1 ) s ) .
Therefore D m ( s , π v × χ v ) D_m(s,\pi_v\times\chi_v) D m ( s , π v × χ v ) equals
− q v q v − 1 λ n − 1 γ v − 1 q v − ( n − 1 ) s ( 1 − λ 1 γ v q v − s + χ π v ( ϖ v ) γ v 2 q v − 2 s ) − λ n − 2 χ π v ( ϖ v ) q v − n s + λ n − 1 γ v − 1 q v − ( n − 1 ) s . -\frac{q_v}{q_v-1}\lambda_{n-1}\gamma_v^{-1}q_v^{-(n-1)s}\left(1-\lambda_1\gamma_vq_v^{-s}+\chi_{\pi_v}(\varpi_v)\gamma_v^2q_v^{-2s}\right)-\lambda_{n-2}\chi_{\pi_v}(\varpi_v)q_v^{-ns}+\lambda_{n-1}\gamma_v^{-1}q_v^{-(n-1)s}. − q v − 1 q v λ n − 1 γ v − 1 q v − ( n − 1 ) s ( 1 − λ 1 γ v q v − s + χ π v ( ϖ v ) γ v 2 q v − 2 s ) − λ n − 2 χ π v ( ϖ v ) q v − n s + λ n − 1 γ v − 1 q v − ( n − 1 ) s .
Collecting like powers of q v − s q_v^{-s} q v − s in the above expression and using the recurrence relation we obtain (22)
D m ( s , π v × χ v ) = q v − ( n − 1 ) s q v − 1 ( λ n q v 1 − s − λ n − 1 γ v − 1 + λ n − 2 χ π v ( ϖ v ) q v s − λ n − 1 χ π v ( ϖ v ) γ v q v 1 − 2 s ) . (22) D_m(s,\pi_v\times\chi_v)=\frac{q_v^{-(n-1)s}}{q_v-1}\left(\lambda_nq_v^{1-s}-\lambda_{n-1}\gamma_v^{-1}+\lambda_{n-2}\chi_{\pi_v}(\varpi_v)q_v^s-\lambda_{n-1}\chi_{\pi_v}(\varpi_v)\gamma_vq_v^{1-2s}\right). \tag*{(22)} D m ( s , π v × χ v ) = q v − 1 q v − ( n − 1 ) s ( λ n q v 1 − s − λ n − 1 γ v − 1 + λ n − 2 χ π v ( ϖ v ) q v s − λ n − 1 χ π v ( ϖ v ) γ v q v 1 − 2 s ) . (22)
Put A v = χ π v ( ϖ v ) A_v=\chi_{\pi_v}(\varpi_v) A v = χ π v ( ϖ v ) . Replacing π v \pi_v π v by π ~ v \widetilde{\pi}_v π v and χ v \chi_v χ v by χ v − 1 \chi_v^{-1} χ v − 1 sends
λ r ⟼ A v − r λ r , A v ⟼ A v − 1 , γ v ⟼ γ v − 1 . \lambda_r\longmapsto A_v^{-r}\lambda_r,\qquad A_v\longmapsto A_v^{-1},\qquad\gamma_v\longmapsto\gamma_v^{-1}. λ r ⟼ A v − r λ r , A v ⟼ A v − 1 , γ v ⟼ γ v − 1 .
Substituting these into (22) gives
D m ( 1 − s , π ~ v × χ v − 1 ) = A v − n q v ( n − 1 ) ( s − 1 ) q v − 1 ( λ n q v s − A v λ n − 1 γ v + A v λ n − 2 q v s − 1 − λ n − 1 γ v − 1 q v 2 s − 1 ) . D_m(1-s,\widetilde{\pi}_v\times\chi_v^{-1})
=\frac{A_v^{-n}q_v^{(n-1)(s-1)}}{q_v-1}\left(\lambda_nq_v^s-A_v\lambda_{n-1}\gamma_v+A_v\lambda_{n-2}q_v^{s-1}-\lambda_{n-1}\gamma_v^{-1}q_v^{2s-1}\right). D m ( 1 − s , π v × χ v − 1 ) = q v − 1 A v − n q v ( n − 1 ) ( s − 1 ) ( λ n q v s − A v λ n − 1 γ v + A v λ n − 2 q v s − 1 − λ n − 1 γ v − 1 q v 2 s − 1 ) .
Multiplying by A v n q v n ( 1 − 2 s ) A_v^nq_v^{n(1-2s)} A v n q v n ( 1 − 2 s ) and comparing with (22) proves the asserted functional equation.
Lemma 3.9. Under the hypothesis of Theorem 1.1, for ω ∈ X ( m ) \omega\in X(\mathfrak m) ω ∈ X ( m ) , t ∈ A F × t\in\mathbb A_F^\times t ∈ A F × , and f , e , e ′ f,e,e' f , e , e ′ as in Lemma 3.4, we have
φ ω ( ( t f t e 1 1 ) ) = ϵ [ ω ] ω S ( − 1 ) χ π ( m ) χ π ( f ) ( R ( k S 0 ) φ ~ ω − 1 ) ( ( n f m 2 t − 1 f n f m 2 t − 1 e ′ 1 1 ) ) . (23) \varphi_\omega\left(\begin{pmatrix} tf & te \\ 1 & 1 \end{pmatrix}\right)
=\epsilon[\omega]\omega_S(-1)\chi_\pi(\mathfrak m)\chi_\pi(f)\left(R(k_{S_0})\widetilde{\varphi}_{\omega^{-1}}\right)\left(\begin{pmatrix} nf_{\mathfrak m}^{2}t^{-1}f & nf_{\mathfrak m}^{2}t^{-1}e' \\ 1 & 1 \end{pmatrix}\right). \tag*{(23)} φ ω ( ( t f 1 t e 1 ) ) = ϵ [ ω ] ω S ( − 1 ) χ π ( m ) χ π ( f ) ( R ( k S 0 ) φ ω − 1 ) ( ( n f m 2 t − 1 f 1 n f m 2 t − 1 e ′ 1 ) ) . (23)
Proof. Let χ \chi χ be a unitary idèle class character. By Lemmas 3.5 and 3.6 , both Mellin transforms vanish unless χ ∼ ω \chi\sim\omega χ ∼ ω . Suppose that χ ∼ ω \chi\sim\omega χ ∼ ω .
For v ∈ S v \in S v ∈ S , put m v = cond ( χ v ) m_v = \operatorname{cond}(\chi_v) m v = cond ( χ v ) . The definition of D m D_m D m and Lemma 3.8 give
( χ π v χ v 2 ) ( ω ‾ v m v ) q v − m v ( 2 s − 1 ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × Ψ ( 1 − s ; L ( d v ) W ~ u ˉ v ⋅ ξ v , χ v − 1 ) L ( 1 − s , π ~ v × χ v − 1 ) = Ξ v ( s , π v × χ v , ψ v ) Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) L ( s , π v × χ v ) . (\chi_{\pi_v}\chi_v^2)(\overline{\omega}_v^{m_v})q_v^{-m_v(2s-1)}(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}
\times\frac{\Psi(1-s;L(d_v)\widetilde{W}_{\bar{u}_v\cdot\xi_v},\chi_v^{-1})}{L(1-s,\widetilde{\pi}_v\times\chi_v^{-1})}
= \Xi_v(s,\pi_v\times\chi_v,\psi_v)\frac{\Psi(s;L(d_v)W_{u_v\cdot\xi_v},\chi_v)}{L(s,\pi_v\times\chi_v)}. ( χ π v χ v 2 ) ( ω v m v ) q v − m v ( 2 s − 1 ) ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × L ( 1 − s , π v × χ v − 1 ) Ψ ( 1 − s ; L ( d v ) W u ˉ v ⋅ ξ v , χ v − 1 ) = Ξ v ( s , π v × χ v , ψ v ) L ( s , π v × χ v ) Ψ ( s ; L ( d v ) W u v ⋅ ξ v , χ v ) .
where
Ξ v ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × ( χ π v χ v 2 ) ( ω ‾ v m v ) q v − m v ( 2 s − 1 ) G ( χ v − 1 , ψ v , d v − 1 ω ‾ v − m v ) G ( χ v , ψ v , d v − 1 ω ‾ v − m v ) . \Xi_v(s,\pi_v\times\chi_v,\psi_v)=(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}
\times(\chi_{\pi_v}\chi_v^2)(\overline{\omega}_v^{m_v})q_v^{-m_v(2s-1)}
\frac{\mathcal{G}(\chi_v^{-1},\psi_v,d_v^{-1}\overline{\omega}_v^{-m_v})}{\mathcal{G}(\chi_v,\psi_v,d_v^{-1}\overline{\omega}_v^{-m_v})}. Ξ v ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × ( χ π v χ v 2 ) ( ω v m v ) q v − m v ( 2 s − 1 ) G ( χ v , ψ v , d v − 1 ω v − m v ) G ( χ v − 1 , ψ v , d v − 1 ω v − m v ) .
Here we have used
L ( d v ) W ~ ξ v = L ( d v ) W π ~ v L(d_v)\widetilde{W}_{\xi_v}=L(d_v)W_{\widetilde{\pi}_v} L ( d v ) W ξ v = L ( d v ) W π v
and
L ( d v ) W ~ u ˉ v ⋅ ξ v = R ( u v ) L ( d v ) W ~ ξ v . L(d_v)\widetilde{W}_{\bar{u}_v\cdot\xi_v}=R(u_v)L(d_v)\widetilde{W}_{\xi_v}. L ( d v ) W u ˉ v ⋅ ξ v = R ( u v ) L ( d v ) W ξ v .
If m v = 0 m_v=0 m v = 0 , both Gauss integrals are 1. Since π v \pi_v π v and χ v \chi_v χ v are unramified,
ϵ ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 . \epsilon(s,\pi_v\times\chi_v,\psi_v)=(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}. ϵ ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 .
Also χ v ( − 1 ) = 1 \chi_v(-1)=1 χ v ( − 1 ) = 1 , so
Ξ v ( s , π v × χ v , ψ v ) = ϵ ( s , π v × χ v , ψ v ) χ v ( − 1 ) . \Xi_v(s,\pi_v\times\chi_v,\psi_v)=\epsilon(s,\pi_v\times\chi_v,\psi_v)\chi_v(-1). Ξ v ( s , π v × χ v , ψ v ) = ϵ ( s , π v × χ v , ψ v ) χ v ( − 1 ) .
If m v ≥ 1 m_v\geq1 m v ≥ 1 , formulas (4) and (5) , together with the multiplicativity of local ϵ \epsilon ϵ -factors for the unramified representation π v \pi_v π v give
( χ π v χ v 2 ) ( ω ‾ v m v ) q v − m v ( 2 s − 1 ) G ( χ v − 1 , ψ v , d v − 1 ω ‾ v − m v ) G ( χ v , ψ v , d v − 1 ω ‾ v − m v ) = ϵ ( s , π v × χ v , ψ v ∘ ) χ v ( − 1 ) . (\chi_{\pi_v}\chi_v^2)(\overline{\omega}_v^{m_v})q_v^{-m_v(2s-1)}
\frac{\mathcal{G}(\chi_v^{-1},\psi_v,d_v^{-1}\overline{\omega}_v^{-m_v})}{\mathcal{G}(\chi_v,\psi_v,d_v^{-1}\overline{\omega}_v^{-m_v})}
=\epsilon(s,\pi_v\times\chi_v,\psi_v^{\circ})\chi_v(-1). ( χ π v χ v 2 ) ( ω v m v ) q v − m v ( 2 s − 1 ) G ( χ v , ψ v , d v − 1 ω v − m v ) G ( χ v − 1 , ψ v , d v − 1 ω v − m v ) = ϵ ( s , π v × χ v , ψ v ∘ ) χ v ( − 1 ) .
Since
ϵ ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × ϵ ( s , π v × χ v , ψ v ( d v − 1 ⋅ ) ) , \epsilon(s,\pi_v\times\chi_v,\psi_v)=(\chi_{\pi_v}\chi_v^2)(d_v)|d_v|_v^{2s-1}\times\epsilon(s,\pi_v\times\chi_v,\psi_v(d_v^{-1}\cdot)), ϵ ( s , π v × χ v , ψ v ) = ( χ π v χ v 2 ) ( d v ) ∣ d v ∣ v 2 s − 1 × ϵ ( s , π v × χ v , ψ v ( d v − 1 ⋅ )) ,
it follows that
Ξ v ( s , π v × χ v , ψ v ) = ϵ ( s , π v × χ v , ψ v ) χ v ( − 1 ) \Xi_v(s,\pi_v\times\chi_v,\psi_v)=\epsilon(s,\pi_v\times\chi_v,\psi_v)\chi_v(-1) Ξ v ( s , π v × χ v , ψ v ) = ϵ ( s , π v × χ v , ψ v ) χ v ( − 1 )
in both cases.
At each v ∈ S 0 v\in S_0 v ∈ S 0 , the local functional equation (6) gives
Ψ ( 1 − s ; W ~ b v ⋅ ξ v , χ v − 1 ) L ( 1 − s , π ~ v × χ v − 1 ) = ϵ ( s , π v × χ v , ψ v ) Ψ ( s ; W b v ⋅ ξ v , χ v ) L ( s , π v × χ v ) . \frac{\Psi(1-s;\widetilde{W}_{b_v\cdot\xi_v},\chi_v^{-1})}{L(1-s,\widetilde{\pi}_v\times\chi_v^{-1})}
=\epsilon(s,\pi_v\times\chi_v,\psi_v)\frac{\Psi(s;W_{b_v\cdot\xi_v},\chi_v)}{L(s,\pi_v\times\chi_v)}. L ( 1 − s , π v × χ v − 1 ) Ψ ( 1 − s ; W b v ⋅ ξ v , χ v − 1 ) = ϵ ( s , π v × χ v , ψ v ) L ( s , π v × χ v ) Ψ ( s ; W b v ⋅ ξ v , χ v ) .
Combining these identities with Lemmas 3.5 and 3.6, and noting that the same factor C ψ S 0 ( s , χ ) \mathcal{C}_{\psi}^{S_0}(s,\chi) C ψ S 0 ( s , χ ) occurs in both Mellin transforms, we obtain
M ( s , f ~ [ ω ] , χ ) Λ ( 1 − s , π ~ × χ − 1 ) = ( ∏ v ∈ S χ v ( − 1 ) ) ϵ ( s , π × χ , ψ ) M ( s , f [ ω ] , χ ) Λ ( s , π × χ ) . \frac{\mathcal{M}(s,\widetilde{f}_{[\omega]},\chi)}{\Lambda(1-s,\widetilde{\pi}\times\chi^{-1})}
=
\left(\prod_{v\in S}\chi_v(-1)\right)\epsilon(s,\pi\times\chi,\psi)
\frac{\mathcal{M}(s,f_{[\omega]},\chi)}{\Lambda(s,\pi\times\chi)}. Λ ( 1 − s , π × χ − 1 ) M ( s , f [ ω ] , χ ) = ( v ∈ S ∏ χ v ( − 1 ) ) ϵ ( s , π × χ , ψ ) Λ ( s , π × χ ) M ( s , f [ ω ] , χ ) .
These are identities of meromorphic functions.
Since χ ∼ ω \chi\sim\omega χ ∼ ω , we have
ϵ χ = ϵ [ ω ] , ∏ v ∈ S χ v ( − 1 ) = ω S ( − 1 ) . \epsilon_\chi=\epsilon_{[\omega]},\qquad
\prod_{v\in S}\chi_v(-1)=\omega_S(-1). ϵ χ = ϵ [ ω ] , v ∈ S ∏ χ v ( − 1 ) = ω S ( − 1 ) .
Using the assumed global functional equation (1.1) and ω S ( − 1 ) 2 = 1 \omega_S(-1)^2=1 ω S ( − 1 ) 2 = 1 , it follows that
M ( s , f [ ω ] , χ ) = ϵ [ ω ] ω S ( − 1 ) M ( s , f ~ [ ω ] , χ ) . \mathcal{M}(s,f_{[\omega]},\chi)=\epsilon_{[\omega]}\omega_S(-1)\mathcal{M}(s,\widetilde{f}_{[\omega]},\chi). M ( s , f [ ω ] , χ ) = ϵ [ ω ] ω S ( − 1 ) M ( s , f [ ω ] , χ ) .
Together with the vanishing when χ ≁ ω \chi\not\sim\omega χ ∼ ω , this gives the identity for every unitary idèle class character χ \chi χ . Applying Lemma 3.1 yields (3.17). □
We now make the normalization described in Remark (4). Put c = ϵ [ 1 ] ≠ 0 c=\epsilon_{[1]}\ne0 c = ϵ [ 1 ] = 0 . We replace each auxiliary dual function by c c c times that function and each ϵ χ \epsilon_\chi ϵ χ by ϵ χ / c \epsilon_\chi/c ϵ χ / c . We retain the same notation for the normalized functions and multipliers; thus ϵ [ 1 ] = 1 \epsilon_{[1]}=1 ϵ [ 1 ] = 1 from this point onward. The local Whittaker functions and local ϵ \epsilon ϵ -factors remain unchanged.
Proof of Theorem 1.1
We retain f , e , e ′ f,e,e' f , e , e ′ as in the previous lemma, and put
g = ( t f t e 0 1 ) . g=\begin{pmatrix}{}^t f & {}^t e\\ 0 & 1\end{pmatrix}. g = ( t f 0 t e 1 ) .
For an admissible m \mathfrak{m} m as before, let S = Supp ( m ) S=\operatorname{Supp}(\mathfrak{m}) S = Supp ( m ) . Then S ∩ S 0 = ∅ S\cap S_0=\varnothing S ∩ S 0 = ∅ . Let
G = X ( m ) , H = X ( 1 ) . G=X(\mathfrak{m}),\qquad H=X(1). G = X ( m ) , H = X ( 1 ) .
With n v = ord v ( m ) n_v=\operatorname{ord}_v(\mathfrak{m}) n v = ord v ( m ) , let
R m = ∏ v ∈ S o v / p v n v , K m = R m × , N m = ∣ K m ∣ , Ψ m ( x ) = ∏ v ∈ S ψ v ( ϖ v − n v − δ v x v ) . R_{\mathfrak{m}}=\prod_{v\in S}\mathfrak{o}_v/\mathfrak{p}_v^{n_v},\qquad
K_{\mathfrak{m}}=R_{\mathfrak{m}}^\times,\qquad
N_{\mathfrak{m}}=|K_{\mathfrak{m}}|,\qquad
\Psi_{\mathfrak{m}}(x)=\prod_{v\in S}\psi_v(\varpi_v^{-n_v-\delta_v}x_v). R m = v ∈ S ∏ o v / p v n v , K m = R m × , N m = ∣ K m ∣ , Ψ m ( x ) = v ∈ S ∏ ψ v ( ϖ v − n v − δ v x v ) .
The group K m K_{\mathfrak{m}} K m may also be realized as
K m = ∏ v ∈ S K v , K_{\mathfrak{m}}=\prod_{v\in S}K_v, K m = v ∈ S ∏ K v ,
where K v = o v × / ( 1 + p v n v ) K_v=\mathfrak{o}_v^\times/(1+\mathfrak{p}_v^{n_v}) K v = o v × / ( 1 + p v n v ) .
Let K ^ m = ∏ v ∣ m K ^ v \widehat{K}_{\mathfrak{m}}=\prod_{v\mid\mathfrak{m}}\widehat{K}_v K m = ∏ v ∣ m K v denote the corresponding character group. Restriction to the local unit group defines a surjective homomorphism from
G ⟶ K ^ m , ω ⟼ ω S , G\longrightarrow\widehat{K}_{\mathfrak{m}},\qquad
\omega\longmapsto\omega_S, G ⟶ K m , ω ⟼ ω S ,
where
ω S ( ( u v ) v ∈ S ) = ∏ v ∈ S ω v ( u v ) . \omega_S((u_v)_{v\in S})=\prod_{v\in S}\omega_v(u_v). ω S (( u v ) v ∈ S ) = v ∈ S ∏ ω v ( u v ) .
Its kernel is precisely H H H and consequently
G / H ≃ K ^ m . G/H\simeq\widehat{K}_{\mathfrak{m}}. G / H ≃ K m .
For each ω S = ∏ v ∈ S ω v ∈ K ^ m \omega_S=\prod_{v\in S}\omega_v\in\widehat{K}_m ω S = ∏ v ∈ S ω v ∈ K m , choose a representative ω ∈ G \omega\in G ω ∈ G . The functions ϕ [ ω ] \phi_{[\omega]} ϕ [ ω ] and ϕ ~ [ ω ] \widetilde{\phi}_{[\omega]} ϕ [ ω ] , as well as the multiplier ϵ [ ω ] \epsilon_{[\omega]} ϵ [ ω ] , are independent of this choice. Take any finite idèle b = ( b v ) b=(b_v) b = ( b v ) so that b v ∈ o v × b_v\in\mathfrak{o}_v^{\times} b v ∈ o v × for all v ∈ S v\in S v ∈ S , and b v = 1 b_v=1 b v = 1 otherwise, multiply both sides of (3.17) by ω ( b ) \omega(b) ω ( b ) and sum over all ω ∈ G / H \omega\in G/H ω ∈ G / H . One checks that
∑ ω ∈ G / H ω ( b ) ϕ [ ω ] ( g ) = ∑ ω ∈ G / H ∑ γ ∈ F × ∫ ∏ v ∈ S o v × W ξ ( ( γ 1 1 ) g ( 1 w u 1 ) ) ω ( u b ) d × u = ∑ γ ∈ F × ∏ v ′ ∉ S W ξ v ′ ( ( γ 1 1 ) g v ′ ) × ∏ v ∈ S I v ( γ , g v ) . (24) \begin{aligned}
\sum_{\omega\in G/H}\omega(b)\phi_{[\omega]}(g)
=
\sum_{\omega\in G/H}\sum_{\gamma\in F^{\times}}\int_{\prod_{v\in S}\mathfrak{o}_v^{\times}}W_{\xi}\left(\begin{pmatrix}\gamma&\\1&1\end{pmatrix}g\begin{pmatrix}1&wu\\&1\end{pmatrix}\right)\omega(ub)\,d^{\times}u \\
&=
\sum_{\gamma\in F^{\times}}\prod_{v'\notin S}W_{\xi_{v'}}\left(\begin{pmatrix}\gamma&\\1&1\end{pmatrix}g_{v'}\right)\times\prod_{v\in S}I_v(\gamma,g_v). \tag*{(24)}
\end{aligned} ω ∈ G / H ∑ ω ( b ) ϕ [ ω ] ( g ) = ω ∈ G / H ∑ γ ∈ F × ∑ ∫ ∏ v ∈ S o v × W ξ ( ( γ 1 1 ) g ( 1 w u 1 ) ) ω ( u b ) d × u = γ ∈ F × ∑ v ′ ∈ / S ∏ W ξ v ′ ( ( γ 1 1 ) g v ′ ) × v ∈ S ∏ I v ( γ , g v ) . (24)
where
I v ( γ , g v ) = ∫ o v × W ξ v ( ( γ 1 1 ) g v ( 1 w v u b v − 1 1 ) ) ( ∑ ω v ∈ K ^ v ω v ( u ) ) d × u . I_v(\gamma,g_v)=\int_{\mathfrak{o}_v^{\times}}W_{\xi_v}\left(\begin{pmatrix}\gamma&\\1&1\end{pmatrix}g_v\begin{pmatrix}1&w_vu b_v^{-1}\\&1\end{pmatrix}\right)\left(\sum_{\omega_v\in\widehat{K}_v}\omega_v(u)\right)d^{\times}u. I v ( γ , g v ) = ∫ o v × W ξ v ( ( γ 1 1 ) g v ( 1 w v u b v − 1 1 ) ) ω v ∈ K v ∑ ω v ( u ) d × u .
But ∑ ω v ∈ K ^ v ω v ( u v ) = 0 \sum_{\omega_v\in\widehat{K}_v}\omega_v(u_v)=0 ∑ ω v ∈ K v ω v ( u v ) = 0 unless u v ∈ U v ( n v ) u_v\in U_v^{(n_v)} u v ∈ U v ( n v ) . Consequently, for v ∈ S v\in S v ∈ S ,
I v ( γ , g v ) = W ξ v ( ( γ 1 1 ) g v ( 1 w v b v − 1 1 ) ) . I_v(\gamma,g_v)=W_{\xi_v}\left(\begin{pmatrix}\gamma&\\1&1\end{pmatrix}g_v\begin{pmatrix}1&w_vb_v^{-1}\\&1\end{pmatrix}\right). I v ( γ , g v ) = W ξ v ( ( γ 1 1 ) g v ( 1 w v b v − 1 1 ) ) .
Inserting this into equation (4.1) and using the definition of ϕ \phi ϕ , we obtain
∑ ω ω ( b ) ϕ [ ω ] ( ( t f t e 1 1 ) ) = ϕ ( ( t f t e 1 1 ) ( 1 w b − 1 1 ) ) . (25) \sum_{\omega}\omega(b)\phi_{[\omega]}\left(\begin{pmatrix}tf&te\\1&1\end{pmatrix}\right)
=
\phi\left(\begin{pmatrix}tf&te\\1&1\end{pmatrix}\begin{pmatrix}1&wb^{-1}\\&1\end{pmatrix}\right). \tag*{(25)} ω ∑ ω ( b ) ϕ [ ω ] ( ( t f 1 t e 1 ) ) = ϕ ( ( t f 1 t e 1 ) ( 1 w b − 1 1 ) ) . (25)
On the other hand, the right hand side of (3.17) after summation is given by
χ π ( f ) χ π ( m ) ∫ ∏ v ∈ S o v × ( ∑ ω ∈ G / H ϵ [ ω ] ω − 1 ( u ) ) ( R ( k S 0 ) ϕ ~ ) ( ( n m f 2 t − 1 1 n m f 2 t − 1 e ′ ) ( 1 w u b 1 ) ) d × u . \chi_{\pi}(f)\chi_{\pi}(m)\int_{\prod_{v\in S}\mathfrak{o}_v^{\times}}
\left(\sum_{\omega\in G/H}\epsilon_{[\omega]}\omega^{-1}(u)\right)
\left(R(k_{S_0})\widetilde{\phi}\right)\left(\begin{pmatrix}n_mf_2^{t-1}&\\1&n_mf_2^{t-1}e'\end{pmatrix}\begin{pmatrix}1&wub\\&1\end{pmatrix}\right)d^{\times}u. χ π ( f ) χ π ( m ) ∫ ∏ v ∈ S o v × ω ∈ G / H ∑ ϵ [ ω ] ω − 1 ( u ) ( R ( k S 0 ) ϕ ) ( ( n m f 2 t − 1 1 n m f 2 t − 1 e ′ ) ( 1 w u b 1 ) ) d × u .
Inserting the Fourier expansion for ϕ ~ \widetilde{\phi} ϕ (cf. (3.2)) into the above equation results in
χ π ( f ) χ π ( m ) ∑ γ ∈ F × W ~ S 0 ( ( γ t S 0 − 1 1 1 ) b S 0 ′ k S 0 ) ψ ~ S 0 ( γ n m f 2 t − 1 ) × ( ∫ ∏ v ∈ S o v × ( ∑ ω S ∈ K ^ m ϵ [ ω ] ω S − 1 ( u ) ) Ψ m ( − d γ n m f 2 t − 1 ( u − 1 ) b ) d × u ) × ψ ( − γ n m f 2 t − 1 w b ) ψ S 0 ( γ n m f 2 t − 1 e ′ ) . (26) \begin{aligned}
\chi_{\pi}(f)\chi_{\pi}(m)\sum_{\gamma\in F^{\times}}\widetilde{W}_{S_0}\left(\begin{pmatrix}\gamma t_{S_0}^{-1}&\\1&1\end{pmatrix}b'_{S_0}k_{S_0}\right)\widetilde{\psi}^{S_0}(\gamma n_mf_2^{t-1}) \\
&\times\left(\int_{\prod_{v\in S}\mathfrak{o}_v^{\times}}
\left(\sum_{\omega_S\in\widehat{K}_m}\epsilon_{[\omega]}\omega_S^{-1}(u)\right)
\Psi_m(-d\gamma n_mf_2^{t-1}(u-1)b)\,d^{\times}u\right)
\times\psi(-\gamma n_mf_2^{t-1}wb)\psi^{S_0}(\gamma n_mf_2^{t-1}e'). \tag*{(26)}
\end{aligned} χ π ( f ) χ π ( m ) γ ∈ F × ∑ W S 0 ( ( γ t S 0 − 1 1 1 ) b S 0 ′ k S 0 ) ψ S 0 ( γ n m f 2 t − 1 ) × ∫ ∏ v ∈ S o v × ω S ∈ K m ∑ ϵ [ ω ] ω S − 1 ( u ) Ψ m ( − d γ n m f 2 t − 1 ( u − 1 ) b ) d × u × ψ ( − γ n m f 2 t − 1 w b ) ψ S 0 ( γ n m f 2 t − 1 e ′ ) . (26)
For x ∈ R m x\in R_m x ∈ R m , put
S m ( x ) = ∫ ∏ v ∈ S o v × ϵ ^ m ( u ) Ψ m ( − ( u − 1 ) x ) d × u , S_m(x)=\int_{\prod_{v\in S}\mathfrak{o}_v^{\times}}\widehat{\epsilon}_m(u)\Psi_m(-(u-1)x)\,d^{\times}u, S m ( x ) = ∫ ∏ v ∈ S o v × ϵ m ( u ) Ψ m ( − ( u − 1 ) x ) d × u ,
where ϵ ^ m ( u ) = ∑ ω S ∈ K ^ m ϵ [ ω ] ω S − 1 ( u ) \widehat{\epsilon}_m(u)=\sum_{\omega_S\in\widehat{K}_m}\epsilon_{[\omega]}\omega_S^{-1}(u) ϵ m ( u ) = ∑ ω S ∈ K m ϵ [ ω ] ω S − 1 ( u ) . It becomes a finite sum
S m ( x ) = 1 N m ∑ u ∈ K m ϵ ^ m ( u ) Ψ m ( ( 1 − u ) x ) . S_m(x)=\frac{1}{N_m}\sum_{u\in K_m}\widehat{\epsilon}_m(u)\Psi_m((1-u)x). S m ( x ) = N m 1 u ∈ K m ∑ ϵ m ( u ) Ψ m (( 1 − u ) x ) .
Now, we can express (26) as
χ π ( f ) χ π ( m ) ∑ γ ∈ F × W ~ S 0 ( ( γ t S 0 − 1 1 ) b S 0 ′ k S 0 ) c ~ S 0 ( γ n m 2 f t − 1 ) (27) \chi_\pi(\mathfrak f)\chi_\pi(\mathfrak m)\sum_{\gamma\in F^\times}\widetilde W_{S_0}\left(\begin{pmatrix}\gamma t_{S_0}^{-1} & \\ 1 & \end{pmatrix}b'_{S_0}k_{S_0}\right)\widetilde c^{S_0}\left(\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}\right)
\qquad \tag*{(27)} χ π ( f ) χ π ( m ) γ ∈ F × ∑ W S 0 ( ( γ t S 0 − 1 1 ) b S 0 ′ k S 0 ) c S 0 ( γ n m 2 f t − 1 ) (27)
× S m ( d γ n m 2 f t − 1 b ) ψ ( − γ n m 2 f t − 1 w b ) ψ S 0 ( γ n m 2 t − 1 e ′ ) . {}\times S_{\mathfrak m}\left(d\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}b\right)\psi\left(-\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}wb\right)\psi^{S_0}\left(\gamma\mathfrak n_{\mathfrak m}^{2}t^{-1}e'\right). × S m ( d γ n m 2 f t − 1 b ) ψ ( − γ n m 2 f t − 1 w b ) ψ S 0 ( γ n m 2 t − 1 e ′ ) .
We absorb the scalar c c c used to normalize the auxiliary dual function into its coefficients c ~ S 0 \widetilde c^{S_0} c S 0 , retaining the same notation. The local Whittaker functions remain unchanged. Now, equating (25) and (27) , we obtain
ϕ ( ( t 1 f t 1 e 1 1 ) w b 1 − 1 ) = χ π ( f ) χ π ( m ) ∑ γ ∈ F × W ~ S 0 ( ( γ t S 0 − 1 1 ) b S 0 ′ k S 0 ) c ~ S 0 ( γ n m 2 f t − 1 ) (28) \phi\left(\begin{pmatrix}t_1^f t_1^e & \\ 1 & 1\end{pmatrix}w b_1^{-1}\right)=\chi_\pi(\mathfrak f)\chi_\pi(\mathfrak m)\sum_{\gamma\in F^\times}\widetilde W_{S_0}\left(\begin{pmatrix}\gamma t_{S_0}^{-1} & \\ 1 & \end{pmatrix}b'_{S_0}k_{S_0}\right)\widetilde c^{S_0}\left(\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}\right) \tag*{(28)} ϕ ( ( t 1 f t 1 e 1 1 ) w b 1 − 1 ) = χ π ( f ) χ π ( m ) γ ∈ F × ∑ W S 0 ( ( γ t S 0 − 1 1 ) b S 0 ′ k S 0 ) c S 0 ( γ n m 2 f t − 1 ) (28)
× S m ( d γ n m 2 f t − 1 b ) ψ ( − γ n m 2 f t − 1 w b ) ψ S 0 ( γ n m 2 t − 1 e ′ ) . {}\times S_{\mathfrak m}\left(d\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}b\right)\psi\left(-\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}wb\right)\psi^{S_0}\left(\gamma\mathfrak n_{\mathfrak m}^{2}t^{-1}e'\right). × S m ( d γ n m 2 f t − 1 b ) ψ ( − γ n m 2 f t − 1 w b ) ψ S 0 ( γ n m 2 t − 1 e ′ ) .
Only those γ \gamma γ for which
div S 0 ( d γ n m 2 f t − 1 ) ≥ 0 \operatorname{div}^{S_0}\left(d\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}\right)\ge0 div S 0 ( d γ n m 2 f t − 1 ) ≥ 0
can contribute, by the support of c ~ S 0 \widetilde c^{S_0} c S 0 . For these terms, the argument
d γ n m 2 f t − 1 b d\gamma\mathfrak n_{\mathfrak m}^{2}\mathfrak f t^{-1}b d γ n m 2 f t − 1 b
defines an element of R m R_{\mathfrak m} R m .
A special case of the automorphy relation
Fix an effective divisor H \mathfrak H H , supported outside S 0 S_0 S 0 and coprime to n \mathfrak n n . Choose the idèle h \mathfrak h h with components
h v = { ϖ v ord v ( H ) , v ∉ S 0 , 1 , v ∈ S 0 . h_v=
\begin{cases}
\varpi_v^{\operatorname{ord}_v(\mathfrak H)}, & v\notin S_0,\\
1, & v\in S_0.
\end{cases} h v = { ϖ v ord v ( H ) , 1 , v ∈ / S 0 , v ∈ S 0 .
Let m \mathfrak m m be an admissible modulus coprime to H \mathfrak H H , and put S = Supp ( m ) S=\operatorname{Supp}(\mathfrak m) S = Supp ( m ) . At every nonarchimedean place v v v , set
f 1 , v = f m , v , e 1 , v = h v , t 1 , v = h v − 1 , f_{1,v}=f_{\mathfrak m,v},\qquad e_{1,v}=h_v,\qquad t_{1,v}=h_v^{-1}, f 1 , v = f m , v , e 1 , v = h v , t 1 , v = h v − 1 ,
and
e 1 , v ′ = { − 1 , v ∈ S , 0 , v ∉ S . e'_{1,v}=
\begin{cases}
-1, & v\in S,\\
0, & v\notin S.
\end{cases} e 1 , v ′ = { − 1 , 0 , v ∈ S , v ∈ / S .
At each archimedean place w w w , set
f 1 , w = e 1 , w = 1 2 , e 1 , w ′ = − 1 2 , t 1 , w = 2 . f_{1,w}=e_{1,w}=\frac{1}{\sqrt{2}},\qquad e'_{1,w}=-\frac{1}{\sqrt{2}},\qquad t_{1,w}=\sqrt{2}. f 1 , w = e 1 , w = 2 1 , e 1 , w ′ = − 2 1 , t 1 , w = 2 .
Then ( f 1 , e 1 ) (f_1,e_1) ( f 1 , e 1 ) is a reduced pair of conductor m \mathfrak m m , and e 1 ′ e'_1 e 1 ′ satisfies (3.6). Moreover,
( t 1 f t 1 e 1 1 ) = ( f m h − 1 1 1 ) . \begin{pmatrix}t_1^f t_1^e & \\ 1 & 1\end{pmatrix}
=
\begin{pmatrix}f_{\mathfrak m}h^{-1} & \\ 1 & 1\end{pmatrix}. ( t 1 f t 1 e 1 1 ) = ( f m h − 1 1 1 ) .
Lemma 4.1. Retain H \mathfrak H H , h \mathfrak h h and m \mathfrak m m as above. Suppose that there exists M ∈ F × M\in F^\times M ∈ F × such that
div S 0 ( M ) = m , M ≡ 1 ( m o d n H ) . \operatorname{div}^{S_0}(M)=\mathfrak m,\qquad M\equiv1\pmod{\mathfrak n\mathfrak H}. div S 0 ( M ) = m , M ≡ 1 ( mod nH ) . Put
b ′ = u ( − 1 ) a ( n h f m − 1 ) . b'=u(-1)a\left(\mathfrak n\mathfrak h\mathfrak f_{\mathfrak m}^{-1}\right). b ′ = u ( − 1 ) a ( nh f m − 1 ) . Then, for every choice of smooth vectors ξ S 0 \xi_{S_0} ξ S 0 ,
ϕ ~ ξ S 0 ( w b ′ ) = ϕ ~ ξ S 0 ( b ′ ) . \widetilde\phi_{\xi_{S_0}}(wb')=\widetilde\phi_{\xi_{S_0}}(b'). ϕ ξ S 0 ( w b ′ ) = ϕ ξ S 0 ( b ′ ) .
Proof. The assertion is unchanged by multiplying the dual function by a nonzero scalar. We therefore work with the original, unscaled dual function ϕ = ϕ ~ ξ \phi=\widetilde{\phi}_{\xi} ϕ = ϕ ξ , and write ϵ \epsilon ϵ for the original multiplier ϵ [ 1 ] \epsilon_{[1]} ϵ [ 1 ] . By Lemma 3.4 , we have
ϕ ( a ( n t − 1 ) ) = ϵ ϕ ~ ( a ( t ) ) . \phi\left(a\left(nt^{-1}\right)\right)=\epsilon\widetilde{\phi}(a(t)). ϕ ( a ( n t − 1 ) ) = ϵ ϕ ( a ( t )) .
which in turn, using
w a ( t ) ( 0 1 − n 0 ) = − t a ( n t − 1 ) , wa(t)\begin{pmatrix}0 & 1 \\ -n & 0\end{pmatrix}=-t\,a\left(nt^{-1}\right), w a ( t ) ( 0 − n 1 0 ) = − t a ( n t − 1 ) ,
gives
ϕ ~ ( w a ( t ) ) = ϕ ξ ( a ( n t − 1 ) ) = ϵ ϕ ~ ( a ( t ) ) , t ∈ A F × . \widetilde{\phi}(wa(t))=\phi_{\xi}\left(a\left(nt^{-1}\right)\right)=\epsilon\widetilde{\phi}(a(t)),\qquad t\in A_F^\times. ϕ ( w a ( t )) = ϕ ξ ( a ( n t − 1 ) ) = ϵ ϕ ( a ( t )) , t ∈ A F × .
For h ∈ G S 0 h\in G_{S_0} h ∈ G S 0 , right translation of ϕ ~ \widetilde{\phi} ϕ by h h h corresponds to replacing ξ S 0 \xi_{S_0} ξ S 0 by π S 0 ( t h − 1 ) ξ S 0 \pi_{S_0}(t h^{-1})\xi_{S_0} π S 0 ( t h − 1 ) ξ S 0 . Here n S 0 = 1 n_{S_0}=1 n S 0 = 1 , and the vectors outside S 0 S_0 S 0 remain unchanged. It follows that
ϕ ~ ( w a ( t ) h ) = ϵ ϕ ~ ( a ( t ) h ) t ∈ A F × , h ∈ G S 0 . (29) \widetilde{\phi}(wa(t)h)=\epsilon\widetilde{\phi}(a(t)h)\qquad t\in A_F^\times,\quad h\in G_{S_0}. \tag*{(29)} ϕ ( w a ( t ) h ) = ϵ ϕ ( a ( t ) h ) t ∈ A F × , h ∈ G S 0 . (29)
Write A = o F S 0 A=\mathfrak{o}^{S_0}_F A = o F S 0 and put
F ( h ) = ϕ ~ ( a ( n h ) h ) , h ∈ G S 0 . \mathcal{F}(h)=\widetilde{\phi}(a(nh)h),\qquad h\in G_{S_0}. F ( h ) = ϕ ( a ( nh ) h ) , h ∈ G S 0 .
We claim that
F ( u ( r ) S 0 h ) = F ( h ) , r ∈ n H A , (30) \mathcal{F}(u(r)_{S_0}h)=\mathcal{F}(h),\qquad r\in n\mathfrak{H}A, \tag*{(30)} F ( u ( r ) S 0 h ) = F ( h ) , r ∈ n H A , (30)
F ( u ˉ ( s ) S 0 h ) = F ( h ) , s ∈ A . (31) \mathcal{F}(\bar{u}(s)_{S_0}h)=\mathcal{F}(h),\qquad s\in A. \tag*{(31)} F ( u ˉ ( s ) S 0 h ) = F ( h ) , s ∈ A . (31)
Since ( n h ) S 0 = 1 (nh)_{S_0}=1 ( nh ) S 0 = 1 , left invariance under u ( r ) ∈ B ( F ) u(r)\in B(F) u ( r ) ∈ B ( F ) and right invariance under K 1 S 0 ( n ) K^{S_0}_1(\mathfrak{n}) K 1 S 0 ( n ) give
ϕ ~ ( a ( n h ) u ( r ) S 0 h ) = ϕ ~ ( u ( r ) S 0 a ( n h ) h ) = ϕ ~ ( u ( − r ) S 0 a ( n h ) h ) = ϕ ~ ( a ( n h ) u ( − r / ( n h ) ) S 0 h ) = ϕ ~ ( a ( n h ) h ) , \begin{aligned}
\widetilde{\phi}(a(nh)u(r)_{S_0}h)&=\widetilde{\phi}(u(r)_{S_0}a(nh)h)\\
&=\widetilde{\phi}(u(-r)^{S_0}a(nh)h)\\
&=\widetilde{\phi}(a(nh)u(-r/(nh))^{S_0}h)\\
&=\widetilde{\phi}(a(nh)h),
\end{aligned} ϕ ( a ( nh ) u ( r ) S 0 h ) = ϕ ( u ( r ) S 0 a ( nh ) h ) = ϕ ( u ( − r ) S 0 a ( nh ) h ) = ϕ ( a ( nh ) u ( − r / ( nh ) ) S 0 h ) = ϕ ( a ( nh ) h ) ,
proving (4.7). Here
u ( − r / ( n h ) ) S 0 ∈ K 1 S 0 ( n ) . u(-r/(nh))^{S_0}\in K^{S_0}_1(\mathfrak{n}). u ( − r / ( nh ) ) S 0 ∈ K 1 S 0 ( n ) .
Similarly, using (4.6),
ϵ F ( u ˉ ( s ) S 0 h ) = ϕ ~ ( w a ( n h ) u ˉ ( s ) S 0 h ) = ϕ ~ ( u ( − s ) S 0 w a ( n h ) h ) = ϕ ~ ( u ( s ) S 0 w a ( n h ) h ) = ϕ ~ ( w a ( n h ) u ˉ ( − s n h ) S 0 h ) = ϕ ~ ( w a ( n h ) h ) = ϵ F ( h ) . \begin{aligned}
\epsilon\mathcal{F}(\bar{u}(s)_{S_0}h)&=\widetilde{\phi}(wa(nh)\bar{u}(s)_{S_0}h)\\
&=\widetilde{\phi}(u(-s)_{S_0}wa(nh)h)\\
&=\widetilde{\phi}(u(s)^{S_0}wa(nh)h)\\
&=\widetilde{\phi}(wa(nh)\bar{u}(-snh)^{S_0}h)\\
&=\widetilde{\phi}(wa(nh)h)=\epsilon\mathcal{F}(h).
\end{aligned} ϵ F ( u ˉ ( s ) S 0 h ) = ϕ ( w a ( nh ) u ˉ ( s ) S 0 h ) = ϕ ( u ( − s ) S 0 w a ( nh ) h ) = ϕ ( u ( s ) S 0 w a ( nh ) h ) = ϕ ( w a ( nh ) u ˉ ( − s nh ) S 0 h ) = ϕ ( w a ( nh ) h ) = ϵ F ( h ) .
Since H \mathfrak{H} H is effective,
u ˉ ( − s n h ) S 0 ∈ K 1 S 0 ( n ) . \bar{u}(-snh)^{S_0}\in K^{S_0}_1(\mathfrak{n}). u ˉ ( − s nh ) S 0 ∈ K 1 S 0 ( n ) .
Together with ϵ ≠ 0 \epsilon\ne0 ϵ = 0 , this proves (4.8).
Now put h = a ( M S 0 ) h=a(M_{S_0}) h = a ( M S 0 ) and define the idèle δ \delta δ by
δ v = { M v f m , v − 1 , v ∉ S 0 , 1 , v ∈ S 0 . \delta_v=
\begin{cases}
M_vf_{m,v}^{-1}, & v\notin S_0,\\
1, & v\in S_0.
\end{cases} δ v = { M v f m , v − 1 , 1 , v ∈ / S 0 , v ∈ S 0 .
By hypothesis, δ v ∈ o v × \delta_v\in o_v^\times δ v ∈ o v × for v ∉ S 0 v\notin S_0 v ∈ / S 0 , and
a ( n h f m − 1 ) = a ( n h / M ) h a ( δ ) . a(nhf_m^{-1})=a(nh/M)h\,a(\delta). a ( nh f m − 1 ) = a ( nh / M ) h a ( δ ) .
Consequently,
ϕ ~ ( b ′ ) = ϕ ~ ( a ( n h f m − 1 ) ) = ϕ ~ ( a ( n h / M ) h a ( δ ) ) = ϕ ~ ( a ( n h ) h ) = F ( h ) . \widetilde{\phi}(b')=\widetilde{\phi}\left(a(nhf_m^{-1})\right)
=\widetilde{\phi}\left(a(nh/M)h\,a(\delta)\right)
=\widetilde{\phi}\left(a(nh)h\right)=\mathcal{F}(h). ϕ ( b ′ ) = ϕ ( a ( nh f m − 1 ) ) = ϕ ( a ( nh / M ) h a ( δ ) ) = ϕ ( a ( nh ) h ) = F ( h ) .
On the other hand, the identity
w u ( − 1 ) a ( M − 1 ) = M − 1 a ( M ) w u ( − M ) wu(-1)a(M^{-1})=M^{-1}a(M)wu(-M) w u ( − 1 ) a ( M − 1 ) = M − 1 a ( M ) w u ( − M )
gives
ϕ ~ ( w b ′ ) = ϕ ~ ( w u ( − M ) a ( n h ) h ) . \widetilde{\phi}(wb')=\widetilde{\phi}\left(wu(-M)a(nh)h\right). ϕ ( w b ′ ) = ϕ ( w u ( − M ) a ( nh ) h ) .
Put
k M = ( 1 ( 1 − M ) / ( n h ) − n h M ) . k_M=\begin{pmatrix}1 & (1-M)/(nh)\\ -nh & M\end{pmatrix}. k M = ( 1 − nh ( 1 − M ) / ( nh ) M ) .
Direct multiplication shows that
w u ( − M ) a ( n h ) = u ( 1 ) a ( n h ) k M . wu(-M)a(nh)=u(1)a(nh)k_M. w u ( − M ) a ( nh ) = u ( 1 ) a ( nh ) k M .
Our assumptions on M M M imply k M S 0 ∈ K 1 S 0 ( n ) k_M^{S_0}\in K_1^{S_0}(\mathfrak n) k M S 0 ∈ K 1 S 0 ( n ) , while
( k M ) S 0 = u ‾ ( − 1 ) S 0 u ( 1 − M ) S 0 . (k_M)_{S_0}=\overline{u}(-1)_{S_0}u(1-M)_{S_0}. ( k M ) S 0 = u ( − 1 ) S 0 u ( 1 − M ) S 0 .
As − 1 ∈ A -1\in A − 1 ∈ A and 1 − M ∈ n S A 1-M\in\mathfrak n\mathfrak S A 1 − M ∈ nS A , (4.7) and (4.8) now give
ϕ ~ ( w b ′ ) = F ( ( k M ) S 0 h ) = F ( h ) = ϕ ~ ( b ′ ) . \widetilde{\phi}(wb')=\mathcal{F}((k_M)_{S_0}h)=\mathcal{F}(h)=\widetilde{\phi}(b'). ϕ ( w b ′ ) = F (( k M ) S 0 h ) = F ( h ) = ϕ ( b ′ ) .
Proposition 4.2. Let π \pi π be as in Theorem 1.1 , and retain S \mathfrak S S and h \mathbf h h as above. Let m \mathbf m m be an admissible modulus coprime to S \mathfrak S S for which there exists M ∈ F × M\in F^\times M ∈ F × satisfying
div S 0 ( M ) = m , M ≡ 1 ( m o d n S ) . \operatorname{div}^{S_0}(M)=\mathbf m,\qquad M\equiv1\pmod{\mathfrak n\mathfrak S}. div S 0 ( M ) = m , M ≡ 1 ( mod nS ) . With f 1 , e 1 , e 1 ′ \mathbf f_1,\mathbf e_1,\mathbf e'_1 f 1 , e 1 , e 1 ′ and t 1 t_1 t 1 chosen as above, we have
ϕ ξ S 0 ( ( t 1 f 1 t 1 e 1 1 1 ) ) = χ π ( f 1 ) ( R ( k S 0 ) ϕ ~ ξ S 0 ) ( ( n t 1 − 1 f 1 n t 1 − 1 e 1 ′ 1 1 ) ) , \phi_{\xi_{S_0}}\left(\begin{pmatrix}t_1f_1&t_1e_1\\1&1\end{pmatrix}\right)
=\chi_\pi(f_1)\left(R(k_{S_0})\widetilde{\phi}_{\xi_{S_0}}\right)\left(\begin{pmatrix}nt_1^{-1}f_1&nt_1^{-1}e'_1\\1&1\end{pmatrix}\right), ϕ ξ S 0 ( ( t 1 f 1 1 t 1 e 1 1 ) ) = χ π ( f 1 ) ( R ( k S 0 ) ϕ ξ S 0 ) ( ( n t 1 − 1 f 1 1 n t 1 − 1 e 1 ′ 1 ) ) , where k S 0 = k S 0 ( f 1 , e 1 ) k_{S_0}=k_{S_0}(f_1,e_1) k S 0 = k S 0 ( f 1 , e 1 ) , for every choice of smooth vectors ξ S 0 \xi_{S_0} ξ S 0 .
Proof. Write ( ϕ , ϕ ~ ) = ( ϕ ξ S 0 , ϕ ~ ξ S 0 ) (\phi,\widetilde{\phi})=(\phi_{\xi_{S_0}},\widetilde{\phi}_{\xi_{S_0}}) ( ϕ , ϕ ) = ( ϕ ξ S 0 , ϕ ξ S 0 ) , with the normalization fixed above. By construction,
( t 1 f 1 t 1 e 1 1 1 ) = ( f m h − 1 1 1 1 ) . \begin{pmatrix}t_1f_1&t_1e_1\\1&1\end{pmatrix}
=
\begin{pmatrix}f_mh^{-1}&1\\1&1\end{pmatrix}. ( t 1 f 1 1 t 1 e 1 1 ) = ( f m h − 1 1 1 1 ) .
Left B 1 ( F ) B_1(F) B 1 ( F ) -invariance and Lemma 3.4, applied to the diagonal matrix, therefore give
ϕ ( ( t 1 f 1 t 1 e 1 1 1 ) ) = ϕ ( ( f m h − 1 0 1 1 ) ) = ϕ ~ ( ( n h f m − 1 0 1 1 ) ) . \phi\left(\begin{pmatrix}t_1f_1&t_1e_1\\1&1\end{pmatrix}\right)
=\phi\left(\begin{pmatrix}f_mh^{-1}&0\\1&1\end{pmatrix}\right)
=\widetilde{\phi}\left(\begin{pmatrix}nhf_m^{-1}&0\\1&1\end{pmatrix}\right). ϕ ( ( t 1 f 1 1 t 1 e 1 1 ) ) = ϕ ( ( f m h − 1 1 0 1 ) ) = ϕ ( ( nh f m − 1 1 0 1 ) ) .
To evaluate the right-hand side of the asserted identity, put
B = ( n t 1 − 1 f 1 n t 1 − 1 e 1 ′ 1 1 ) , f = n t 1 − 1 , e = f 1 − 1 ( n t 1 − 1 e 1 ′ + 1 ) . B=\begin{pmatrix} nt_1^{-1}f_1 & nt_1^{-1}e'_1 \\ 1 & 1 \end{pmatrix}, \qquad f=nt_1^{-1}, \qquad e=f_1^{-1}(nt_1^{-1}e'_1+1). B = ( n t 1 − 1 f 1 1 n t 1 − 1 e 1 ′ 1 ) , f = n t 1 − 1 , e = f 1 − 1 ( n t 1 − 1 e 1 ′ + 1 ) .
Since R ( k S 0 ) ϕ ~ R(k_{S_0})\widetilde{\phi} R ( k S 0 ) ϕ is left B 1 ( F ) B_1(F) B 1 ( F ) -invariant, we have
( R ( k S 0 ) ϕ ~ ) ( B ) = ( R ( k S 0 ) ϕ ~ ) ( b ) , b = ( f 1 f f 1 e 1 1 ) = u ( 1 ) B . (32) (R(k_{S_0})\widetilde{\phi})(B)=(R(k_{S_0})\widetilde{\phi})(b), \qquad b=\begin{pmatrix} f_1f & f_1e \\ 1 & 1 \end{pmatrix}=u(1)B. \tag*{(32)} ( R ( k S 0 ) ϕ ) ( B ) = ( R ( k S 0 ) ϕ ) ( b ) , b = ( f 1 f 1 f 1 e 1 ) = u ( 1 ) B . (32)
Our choices give
f w = e w = 1 2 ( w ∣ ∞ ) , f_w=e_w=\frac{1}{\sqrt{2}}(w\mid\infty), f w = e w = 2 1 ( w ∣ ∞ ) ,
and, at finite places,
f v = n v h v , e v = { 0 , v ∈ S , 1 , v ∉ S . f_v=n_vh_v,\qquad e_v=\begin{cases}0,&v\in S,\\1,&v\notin S.\end{cases} f v = n v h v , e v = { 0 , 1 , v ∈ S , v ∈ / S .
Here n v = h v = 1 n_v=h_v=1 n v = h v = 1 for v ∈ S v\in S v ∈ S .
Choose e ′ e' e ′ by
e v ′ = { − 1 / 2 , v ∣ ∞ , − f m , v , v ∈ S , − 1 , v < ∞ , v ∉ S . e'_v=\begin{cases}-1/\sqrt{2},&v\mid\infty,\\-f_{m,v},&v\in S,\\-1,&v<\infty,\ v\notin S.\end{cases} e v ′ = ⎩ ⎨ ⎧ − 1/ 2 , − f m , v , − 1 , v ∣ ∞ , v ∈ S , v < ∞ , v ∈ / S .
Define
b ′ = ( f 1 − 1 f 1 − 1 e 1 ′ 1 ) , z = ( f 1 0 0 f 1 ) , k ′ = ( − e ′ − ( 1 + e e ′ ) / f f e ) . b'=\begin{pmatrix} f_1^{-1}f_1^{-1}e'_1 & \\ 1 & \end{pmatrix},\qquad
z=\begin{pmatrix} f_1 & 0 \\ 0 & f_1 \end{pmatrix},\qquad
k'=\begin{pmatrix} -e' & -(1+ee')/f \\ f & e \end{pmatrix}. b ′ = ( f 1 − 1 f 1 − 1 e 1 ′ 1 ) , z = ( f 1 0 0 f 1 ) , k ′ = ( − e ′ f − ( 1 + e e ′ ) / f e ) .
Then
b = w b ′ k ′ z . (33) b=wb'k'z. \tag*{(33)} b = w b ′ k ′ z . (33)
Moreover, the local choices above give
k ′ S 0 ∈ K 1 S 0 ( n ) , k S 0 ′ = k S 0 − 1 , k'^{S_0}\in K^{S_0}_1(\mathfrak n),\qquad k'_{S_0}=k_{S_0}^{-1}, k ′ S 0 ∈ K 1 S 0 ( n ) , k S 0 ′ = k S 0 − 1 ,
and
b ′ = u ( − 1 ) a ( n h m − 1 ) . b'=u(-1)a(nh_{m}^{-1}). b ′ = u ( − 1 ) a ( n h m − 1 ) .
Indeed, at v ∉ S ∪ S 0 v\notin S\cup S_0 v ∈ / S ∪ S 0 ,
k v ′ = ( n v − 1 h v − 1 0 1 1 ) ∈ K 1 , v ( p v n ( π v ) ) , k'_v=\begin{pmatrix} n_v^{-1}h_v^{-1} & 0 \\ 1 & 1 \end{pmatrix}\in K_{1,v}\bigl(\mathfrak p_v^{n(\pi_v)}\bigr), k v ′ = ( n v − 1 h v − 1 1 0 1 ) ∈ K 1 , v ( p v n ( π v ) ) ,
while at v ∈ S v\in S v ∈ S ,
k v ′ = ( f m , v − 1 1 0 ) ∈ GL 2 ( o v ) . k'_v=\begin{pmatrix} f_{m,v} & -1 \\ 1 & 0 \end{pmatrix}\in\operatorname{GL}_2(\mathfrak o_v). k v ′ = ( f m , v 1 − 1 0 ) ∈ GL 2 ( o v ) .
Using (33) , we obtain
( R ( k S 0 ) ϕ ~ ) ( b ) = ϕ ~ ( w b ′ k ′ z k S 0 ) = χ π ( f 1 ) − 1 ϕ ~ ( w b ′ k ′ S 0 ) = χ π ( f 1 ) − 1 ϕ ~ ( w b ′ ) = χ π ( f 1 ) − 1 ϕ ~ ( b ′ ) , \begin{aligned}
(R(k_{S_0})\widetilde{\phi})(b)
&=\widetilde{\phi}(wb'k'z k_{S_0})\\
&=\chi_\pi(f_1)^{-1}\widetilde{\phi}(wb'k'^{S_0})\\
&=\chi_\pi(f_1)^{-1}\widetilde{\phi}(wb')\\
&=\chi_\pi(f_1)^{-1}\widetilde{\phi}(b'),
\end{aligned} ( R ( k S 0 ) ϕ ) ( b ) = ϕ ( w b ′ k ′ z k S 0 ) = χ π ( f 1 ) − 1 ϕ ( w b ′ k ′ S 0 ) = χ π ( f 1 ) − 1 ϕ ( w b ′ ) = χ π ( f 1 ) − 1 ϕ ( b ′ ) ,
where the last equality follows from Lemma 4.1.
Finally, left invariance under u ( 1 ) ∈ B 1 ( F ) u(1)\in B_1(F) u ( 1 ) ∈ B 1 ( F ) gives
ϕ ~ ( b ′ ) = ϕ ~ ( u ( 1 ) b ′ ) = ϕ ~ ( ( n h m − 1 1 ) ) . \widetilde{\phi}(b')=\widetilde{\phi}(u(1)b')=\widetilde{\phi}\left(\begin{pmatrix} nh_m^{-1} & \\ 1 & \end{pmatrix}\right). ϕ ( b ′ ) = ϕ ( u ( 1 ) b ′ ) = ϕ ( ( n h m − 1 1 ) ) .
Combining this with (32) and the first displayed identity proves the proposition.
Lemma 4.3. For any irreducible admissible generic representation π \pi π of G L 2 ( A F ) \mathrm{GL}_{2}(\mathbb{A}_{F}) GL 2 ( A F ) , let L \mathcal{L} L be a nonzero fractional ideal of A = o F S 0 A = \mathfrak{o}_{F}^{S_{0}} A = o F S 0 . Put Γ = L ∖ { 0 } \Gamma= \mathcal{L} \setminus\{0\} Γ = L ∖ { 0 } . For γ 0 ∈ Γ \gamma_{0} \in\Gamma γ 0 ∈ Γ and g S 0 ∈ G S 0 g_{S_{0}} \in G_{S_{0}} g S 0 ∈ G S 0 , there exists a pure tensor ξ S 0 \xi_{S_{0}} ξ S 0 of smooth vectors such that
W ξ S 0 ( a ( γ ) g S 0 ) ) = { 1 , γ = γ 0 , 0 , γ ∈ Γ ∖ { γ 0 } . W_{\xi_{S_{0}}}(a(\gamma)g_{S_{0})) =
\begin{cases}
1, & \gamma= \gamma_{0},\\
0, & \gamma\in\Gamma\setminus\{\gamma_{0}\}.
\end{cases}} W ξ S 0 ( a ( γ ) g S 0 )) = { 1 , 0 , γ = γ 0 , γ ∈ Γ ∖ { γ 0 } .
Proof. By the product formula, the diagonal image of L \mathcal{L} L in F S 0 = ∏ v ∈ S 0 F v F_{S_{0}} = \prod_{v \in S_{0}} F_{v} F S 0 = ∏ v ∈ S 0 F v is discrete. Fix γ 0 ∈ Γ \gamma_{0} \in\Gamma γ 0 ∈ Γ . We may choose relatively compact open neighborhoods V v V_{v} V v of γ 0 \gamma_{0} γ 0 in F v F_{v} F v , with 0 ∉ V v 0 \notin V_{v} 0 ∈ / V v , such that
( ∏ v ∈ S 0 V v ) ∩ Γ = { γ 0 } . \left(\prod_{v \in S_{0}} V_{v}\right) \cap\Gamma= \{\gamma_{0}\}. ( v ∈ S 0 ∏ V v ) ∩ Γ = { γ 0 } .
Choose Schwartz–Bruhat functions h v h_{v} h v on F v F_{v} F v , supported in V v V_{v} V v , with h v ( γ 0 ) = 1 h_{v}(\gamma_{0}) = 1 h v ( γ 0 ) = 1 , and put h = ⨂ v ∈ S 0 h v h = \bigotimes_{v \in S_{0}} h_{v} h = ⨂ v ∈ S 0 h v . Thus
h ( γ ) = { 1 , γ = γ 0 , 0 , γ ∈ Γ ∖ { γ 0 } . h(\gamma) =
\begin{cases}
1, & \gamma= \gamma_{0},\\
0, & \gamma\in\Gamma\setminus\{\gamma_{0}\}.
\end{cases} h ( γ ) = { 1 , 0 , γ = γ 0 , γ ∈ Γ ∖ { γ 0 } .
Let
κ ( x ) = ∫ F S 0 h ( y ) ψ S 0 ( − x y ) d y , \kappa(x) = \int_{F_{S_{0}}} h(y)\psi_{S_{0}}(-xy)\,dy, κ ( x ) = ∫ F S 0 h ( y ) ψ S 0 ( − x y ) d y ,
be its Fourier transform with respect to the self-dual additive Haar measure d y dy d y . Then κ \kappa κ is also a Schwartz–Bruhat function, and Fourier inversion gives
h ( y ) = ∫ F S 0 κ ( x ) ψ S 0 ( x y ) d x . h(y) = \int_{F_{S_{0}}} \kappa(x)\psi_{S_{0}}(xy)\,dx. h ( y ) = ∫ F S 0 κ ( x ) ψ S 0 ( x y ) d x .
Choose
W v ′ ∈ W ( π v , ψ v ) , W v ′ ( a ( γ 0 ) ) = 1 ( v ∈ S 0 ) , W'_{v} \in W(\pi_{v},\psi_{v}), \qquad W'_{v}(a(\gamma_{0})) = 1 \quad(v \in S_{0}), W v ′ ∈ W ( π v , ψ v ) , W v ′ ( a ( γ 0 )) = 1 ( v ∈ S 0 ) ,
and put W ′ = ⨂ v ∈ S 0 W v ′ W' = \bigotimes_{v \in S_{0}} W'_{v} W ′ = ⨂ v ∈ S 0 W v ′ . Let
W h ′ ( g ) = ∫ F S 0 κ ( x ) W ′ ( g u ( x ) ) d x . W'_{h}(g) = \int_{F_{S_{0}}} \kappa(x)W'(g u(x))\,dx. W h ′ ( g ) = ∫ F S 0 κ ( x ) W ′ ( g u ( x )) d x .
This defines an element of W ( π S 0 , ψ S 0 ) W(\pi_{S_{0}},\psi_{S_{0}}) W ( π S 0 , ψ S 0 ) . Indeed, at nonarchimedean places it is a finite linear combination of right translates of W ′ W' W ′ . At archimedean places, moderate growth of the smooth globalization and rapid decay of κ \kappa κ imply convergence in the smooth Fréchet topology [2 §2.5] . Using a ( y ) u ( x ) = u ( y x ) a ( y ) a(y)u(x) = u(yx)a(y) a ( y ) u ( x ) = u ( y x ) a ( y ) , we obtain
W h ′ ( a ( y ) ) = ∫ F S 0 κ ( x ) ψ S 0 ( y x ) W ′ ( a ( y ) ) d x = h ( y ) W ′ ( a ( y ) ) , y ∈ F S 0 × . W'_{h}(a(y)) = \int_{F_{S_{0}}} \kappa(x)\psi_{S_{0}}(yx)W'(a(y))\,dx
= h(y)W'(a(y)), \qquad y \in F_{S_{0}}^{\times}. W h ′ ( a ( y )) = ∫ F S 0 κ ( x ) ψ S 0 ( y x ) W ′ ( a ( y )) d x = h ( y ) W ′ ( a ( y )) , y ∈ F S 0 × .
Consequently,
W h ′ ( a ( γ ) ) = { 1 , γ = γ 0 , 0 , γ ∈ Γ ∖ { γ 0 } . W'_{h}(a(\gamma)) =
\begin{cases}
1, & \gamma= \gamma_{0},\\
0, & \gamma\in\Gamma\setminus\{\gamma_{0}\}.
\end{cases} W h ′ ( a ( γ )) = { 1 , 0 , γ = γ 0 , γ ∈ Γ ∖ { γ 0 } .
Since κ = ⨂ v ∈ S 0 κ v \kappa=\bigotimes_{v\in S_0}\kappa_v κ = ⨂ v ∈ S 0 κ v ,
W h ′ = ⨂ v ∈ S 0 W v , h v ′ . W'_h=\bigotimes_{v\in S_0}W'_{v,h_v}. W h ′ = v ∈ S 0 ⨂ W v , h v ′ .
We may therefore choose a smooth vector
ξ S 0 ( γ 0 ) = ⨂ v ∈ S 0 ξ v ( γ 0 ) \xi_{S_0}^{(\gamma_0)}=\bigotimes_{v\in S_0}\xi_v^{(\gamma_0)} ξ S 0 ( γ 0 ) = v ∈ S 0 ⨂ ξ v ( γ 0 )
such that
R ( g S 0 ) W ξ S 0 ( γ 0 ) = W h ′ . R(g_{S_0})W_{\xi_{S_0}^{(\gamma_0)}}=W'_h. R ( g S 0 ) W ξ S 0 ( γ 0 ) = W h ′ .
This proves the assertion. □
Lemma 4.4. Put d = div S 0 ( d ) \mathfrak{d}=\operatorname{div}^{S_0}(\mathfrak{d}) d = div S 0 ( d ) . For every admissible modulus m \mathfrak{m} m , there exists an effective divisor D \mathfrak{D} D , supported outside S 0 S_0 S 0 , such that
[ D ] = [ n d 2 ] in C l ( o F S 0 ) , ( D , m ) = 1 , c ~ S 0 ( D d − 1 ) ≠ 0. [\mathfrak{D}]=[\mathfrak{n}\mathfrak{d}^2]\quad\text{in }\mathrm{Cl}(\mathfrak{o}_F^{S_0}),\qquad(\mathfrak{D},\mathfrak{m})=1,\qquad\widetilde{c}^{S_0}(\mathfrak{D}\mathfrak{d}^{-1})\ne0. [ D ] = [ n d 2 ] in Cl ( o F S 0 ) , ( D , m ) = 1 , c S 0 ( D d − 1 ) = 0. Proof. Put A = o F S 0 A=\mathfrak{o}_F^{S_0} A = o F S 0 and T = Supp ( m ) T=\operatorname{Supp}(\mathfrak{m}) T = Supp ( m ) . Let h h h be the order of the finite group C l ( A ) \mathrm{Cl}(A) Cl ( A ) . For each v ∈ T v\in T v ∈ T , choose a sufficiently large positive integer e v e_v e v divisible by h h h , and a character ρ v \rho_v ρ v of o v × \mathfrak{o}_v^\times o v × of conductor exponent e v e_v e v . Choose a unitary idèle class character θ \theta θ unramified outside T ∪ S 0 T\cup S_0 T ∪ S 0 such that
θ v ∣ o v × = ρ v ( v ∈ T ) , \theta_v|_{\mathfrak{o}_v^\times}=\rho_v\qquad(v\in T), θ v ∣ o v × = ρ v ( v ∈ T ) , and put σ = π ⊗ θ \sigma=\pi\otimes\theta σ = π ⊗ θ .
Since π v \pi_v π v is unramified for v ∈ T v\in T v ∈ T ,
L ( s , σ v ) = L ( s , σ ~ v ) = 1 , n ( σ v ) = 2 e v . L(s,\sigma_v)=L(s,\widetilde{\sigma}_v)=1,\qquad n(\sigma_v)=2e_v. L ( s , σ v ) = L ( s , σ v ) = 1 , n ( σ v ) = 2 e v . Outside T ∪ S 0 T\cup S_0 T ∪ S 0 , twisting by θ v \theta_v θ v preserves the conductor. Thus the conductor of σ \sigma σ outside S 0 S_0 S 0 is
N θ = n ∏ v ∈ T p v 2 e v . \mathfrak{N}_\theta=\mathfrak{n}\prod_{v\in T}\mathfrak{p}_v^{2e_v}. N θ = n v ∈ T ∏ p v 2 e v . By our choice of the e v e_v e v ,
(4.11)
[ N θ ] = [ n ] in C l ( A ) . [\mathfrak{N}_\theta]=[\mathfrak{n}]\quad\text{in }\mathrm{Cl}(A). [ N θ ] = [ n ] in Cl ( A ) . Let N θ \mathfrak{N}_\theta N θ represent N θ \mathfrak{N}_\theta N θ , with components 1 1 1 at S 0 S_0 S 0 .
For every idèle class character η \eta η unramified outside S 0 S_0 S 0 , the character θ η \theta\eta θ η is an allowed twist in Theorem 1.1. The original functional-equation scalar is independent of η \eta η ; denote it by ϵ θ o r i g ≠ 0 \epsilon_\theta^{\mathrm{orig}}\ne0 ϵ θ orig = 0 . Consequently, the base-case Mellin-inversion argument applies to σ \sigma σ and this family of characters. Only that argument is used here, not the conclusion of Theorem 1.1 for σ \sigma σ .
Choose normalized essential vectors for σ v \sigma_v σ v at v ∉ S 0 v\notin S_0 v ∈ / S 0 and arbitrary smooth vectors at S 0 S_0 S 0 . Denote the resulting functions by ϕ ξ S 0 σ \phi_{\xi_{S_0}}^\sigma ϕ ξ S 0 σ and ϕ ~ ξ S 0 σ \widetilde{\phi}_{\xi_{S_0}}^\sigma ϕ ξ S 0 σ , using the original, unscaled dual function. Applying the base-case identity at t = d − 1 t=\mathfrak{d}^{-1} t = d − 1 gives
(4.12)
ϕ ξ S 0 σ ( a ( d − 1 ) ) = ϵ θ o r i g ϕ ~ ξ S 0 σ ( a ( N θ d ) ) . \phi_{\xi_{S_0}}^\sigma(a(\mathfrak{d}^{-1}))=\epsilon_\theta^{\mathrm{orig}}\widetilde{\phi}_{\xi_{S_0}}^\sigma(a(\mathfrak{N}_\theta\mathfrak{d})). ϕ ξ S 0 σ ( a ( d − 1 )) = ϵ θ orig ϕ ξ S 0 σ ( a ( N θ d )) . Since d S 0 = 1 \mathfrak{d}_{S_0}=1 d S 0 = 1 , the Fourier expansion is
ϕ ξ S 0 σ ( a ( d − 1 ) ) = ∑ γ ∈ A ∖ { 0 } c σ S 0 ( γ d − 1 ) W ξ S 0 σ ( a ( γ ) ) . \phi_{\xi_{S_0}}^\sigma(a(\mathfrak{d}^{-1}))=\sum_{\gamma\in A\setminus\{0\}}c_\sigma^{S_0}(\gamma\mathfrak{d}^{-1})W_{\xi_{S_0}}^\sigma(a(\gamma)). ϕ ξ S 0 σ ( a ( d − 1 )) = γ ∈ A ∖ { 0 } ∑ c σ S 0 ( γ d − 1 ) W ξ S 0 σ ( a ( γ )) .
Indeed, the support condition outside S 0 S_0 S 0 requires ord v ( γ ) ≥ 0 \operatorname{ord}_v(\gamma) \ge0 ord v ( γ ) ≥ 0 . The essential-vector normalization gives
C σ S 0 ( d − 1 ) = 1. C_{\sigma}^{S_0}(d^{-1}) = 1. C σ S 0 ( d − 1 ) = 1.
Apply Lemma 4.3 to σ \sigma σ , with L = A \mathcal{L} = A L = A , γ 0 = 1 \gamma_0 = 1 γ 0 = 1 , and g S 0 = 1 g_{S_0} = 1 g S 0 = 1 . We may choose a pure tensor ξ S 0 \xi_{S_0} ξ S 0 so that only the term γ = 1 \gamma= 1 γ = 1 survives, with value 1. Hence
ϕ ξ S 0 σ ( a ( d − 1 ) ) = 1. \phi_{\xi_{S_0}}^\sigma(a(d^{-1})) = 1. ϕ ξ S 0 σ ( a ( d − 1 )) = 1.
It follows from (4.12) that
ϕ ~ ξ S 0 σ ( a ( N θ d ) ) ≠ 0. \widetilde{\phi}_{\xi_{S_0}}^\sigma(a(N_\theta d)) \ne0. ϕ ξ S 0 σ ( a ( N θ d )) = 0.
Therefore there exists γ ∈ F × \gamma\in F^\times γ ∈ F × such that
C ~ σ S 0 ( γ N θ d ) ≠ 0. \widetilde{C}_{\sigma}^{S_0}(\gamma N_\theta d) \ne0. C σ S 0 ( γ N θ d ) = 0.
Put
D = div S 0 ( γ N θ d 2 ) . \mathcal{D} = \operatorname{div}^{S_0}(\gamma N_\theta d^2). D = div S 0 ( γ N θ d 2 ) .
By the support of the essential Whittaker functions, this divisor is effective. Equation (4.11) gives
[ D ] = [ N θ d 2 ] = [ n d 2 ] . [\mathcal{D}] = [N_\theta d^2] = [n d^2]. [ D ] = [ N θ d 2 ] = [ n d 2 ] .
For v ∈ T v \in T v ∈ T , the identity L ( s , σ ~ v ) = 1 L(s,\widetilde{\sigma}_v) = 1 L ( s , σ v ) = 1 implies
( L ( d v ) W σ ~ v ) ( a ( y ) ) = 1 o v × ( y ) . (L(d_v)W_{\widetilde{\sigma}_v})(a(y)) = \mathbf{1}_{\mathfrak{o}_v^\times}(y). ( L ( d v ) W σ v ) ( a ( y )) = 1 o v × ( y ) .
Equivalently,
W σ ~ v ( a ( y ) ) = 1 o v × ( d v y ) . W_{\widetilde{\sigma}_v}(a(y)) = \mathbf{1}_{\mathfrak{o}_v^\times}(d_v y). W σ v ( a ( y )) = 1 o v × ( d v y ) .
The local factor in C ~ σ S 0 \widetilde{C}_{\sigma}^{S_0} C σ S 0 is a nonzero scalar multiple of this function. Thus its nonvanishing implies
ord v ( γ N θ , v d v 2 ) = 0 ( v ∈ T ) . \operatorname{ord}_v(\gamma N_{\theta,v}d_v^2) = 0 \qquad(v \in T). ord v ( γ N θ , v d v 2 ) = 0 ( v ∈ T ) .
Consequently, ( D , m ) = 1 (\mathcal{D},\mathfrak{m}) = 1 ( D , m ) = 1 .
Outside T ∪ S 0 T \cup S_0 T ∪ S 0 , the normalized essential diagonal coefficients for σ ~ v = π ~ v ⊗ θ v − 1 \widetilde{\sigma}_v = \widetilde{\pi}_v \otimes\theta_v^{-1} σ v = π v ⊗ θ v − 1 differ from those for π ~ v \widetilde{\pi}_v π v by nonzero character factors. At v ∈ T v \in T v ∈ T , the argument y = γ N θ , v d v y = \gamma N_{\theta,v}d_v y = γ N θ , v d v satisfies d v y ∈ o v × d_v y \in\mathfrak{o}_v^\times d v y ∈ o v × . At such an argument, both normalized essential Whittaker functions have value 1. The corresponding dual local factors are therefore nonzero. It follows that
C ~ S 0 ( γ N θ d ) ≠ 0. \widetilde{C}^{S_0}(\gamma N_\theta d) \ne0. C S 0 ( γ N θ d ) = 0.
Since
div S 0 ( γ N θ d ) = D d − 1 , \operatorname{div}^{S_0}(\gamma N_\theta d) = \mathcal{D}d^{-1}, div S 0 ( γ N θ d ) = D d − 1 ,
the assertion follows. □
Conclusion of the proof of the main theorem
We keep H \mathfrak{H} H and its representing idèle h h h as in §4.1. Fix an admissible modulus m \mathfrak{m} m coprime to H \mathfrak{H} H such that [ m ] H = 1 [\mathfrak{m}]_{\mathfrak{H}}=1 [ m ] H = 1 in Cl H ( A ) \operatorname{Cl}_{\mathfrak{H}}(A) Cl H ( A ) . Thus there exists M ∈ F × M\in F^\times M ∈ F × satisfying
div S 0 ( M ) = m , M ≡ 1 ( m o d n H ) . \operatorname{div}^{S_0}(M)=\mathfrak{m},\qquad M\equiv1\pmod{\mathfrak{n}\mathfrak{H}}. div S 0 ( M ) = m , M ≡ 1 ( mod nH ) .
We retain f 1 , e 1 , e 1 ′ , t 1 \mathbf{f}_1,\mathbf{e}_1,\mathbf{e}'_1,t_1 f 1 , e 1 , e 1 ′ , t 1 from §4.1, and put
z m = n t 1 − 1 e 1 ′ , q v = a ( t 1 , v − 1 f 1 , v ) k v ( v ∈ S 0 ) . z_{\mathfrak{m}}=nt_1^{-1}\mathbf{e}'_1,\qquad q_v=a(t_{1,v}^{-1}f_{1,v})k_v\quad(v\in S_0). z m = n t 1 − 1 e 1 ′ , q v = a ( t 1 , v − 1 f 1 , v ) k v ( v ∈ S 0 ) .
Write g S 0 = ( q v ) v ∈ S 0 g_{S_0}=(q_v)_{v\in S_0} g S 0 = ( q v ) v ∈ S 0 and set
Γ m = { γ ∈ F × : ord v ( d v γ n v h v f m , v ) ≥ 0 for every v ∉ S 0 } . \Gamma_{\mathfrak{m}}=\left\{\gamma\in F^\times:\operatorname{ord}_v(d_v\gamma n_vh_vf_{\mathfrak{m},v})\geq0\text{ for every }v\notin S_0\right\}. Γ m = { γ ∈ F × : ord v ( d v γ n v h v f m , v ) ≥ 0 for every v ∈ / S 0 } .
We suppress the dependence of Γ m \Gamma_{\mathfrak{m}} Γ m on the fixed divisor H \mathfrak{H} H . The coefficient c ~ S 0 ( γ n h f m ) \widetilde{c}^{S_0}(\gamma nhf_{\mathfrak{m}}) c S 0 ( γ nh f m ) vanishes outside Γ m \Gamma_{\mathfrak{m}} Γ m . For γ ∈ Γ m \gamma\in\Gamma_{\mathfrak{m}} γ ∈ Γ m , the argument of S m S_{\mathfrak{m}} S m is the image of
( d v γ n v h v f m , v ) v ∈ S (d_v\gamma n_vh_vf_{\mathfrak{m},v})_{v\in S} ( d v γ n v h v f m , v ) v ∈ S
in R m R_{\mathfrak{m}} R m .
In (4.5), take b = 1 b=1 b = 1 and replace t t t by t f m tf_{\mathfrak{m}} t f m . At every v ∉ S 0 v\notin S_0 v ∈ / S 0 , choose
f v = 1 , e v ′ = 0 , e v = { 0 , v ∈ S , h v , v ∉ S . f_v=1,\qquad e'_v=0,\qquad e_v=
\begin{cases}
0,&v\in S,\\
h_v,&v\notin S.
\end{cases} f v = 1 , e v ′ = 0 , e v = { 0 , h v , v ∈ S , v ∈ / S .
At the places in S 0 S_0 S 0 , take
( f v , e v , e v ′ ) = ( f 1 , v , e 1 , v , e 1 , v ′ ) . (f_v,e_v,e'_v)=(f_{1,v},e_{1,v},e'_{1,v}). ( f v , e v , e v ′ ) = ( f 1 , v , e 1 , v , e 1 , v ′ ) .
Finally, set t = t 1 t=t_1 t = t 1 .
With these choices, the matrix on the left is
( t 1 f 1 t 1 e 1 0 1 ) . \begin{pmatrix}
t_1f_1&t_1e_1\\
0&1
\end{pmatrix}. ( t 1 f 1 0 t 1 e 1 1 ) .
and the translated matrix on the dual side is
( n t 1 − 1 f 1 n t 1 − 1 e 1 ′ 0 1 ) . \begin{pmatrix}
nt_1^{-1}f_1&nt_1^{-1}e'_1\\
0&1
\end{pmatrix}. ( n t 1 − 1 f 1 0 n t 1 − 1 e 1 ′ 1 ) .
Consequently,
ϕ ξ S 0 ( ( t 1 f 1 t 1 e 1 1 1 ) ) = χ π ( f 1 ) ∑ γ ∈ Γ m c ~ S 0 ( γ n h f m ) S m ( d γ n h f m ) ψ ( γ z m ) W ~ ξ S 0 ( a ( γ ) g S 0 ) . (34) \phi_{\xi_{S_0}}\left(\begin{pmatrix}
t_1f_1&t_1e_1\\
1&1
\end{pmatrix}\right)
=\chi_\pi(f_1)\sum_{\gamma\in\Gamma_{\mathfrak{m}}}\widetilde{c}^{S_0}(\gamma nhf_{\mathfrak{m}})S_{\mathfrak{m}}(d\gamma nhf_{\mathfrak{m}})\psi(\gamma z_{\mathfrak{m}})\widetilde{W}_{\xi_{S_0}}(a(\gamma)g_{S_0}). \tag*{(34)} ϕ ξ S 0 ( ( t 1 f 1 1 t 1 e 1 1 ) ) = χ π ( f 1 ) γ ∈ Γ m ∑ c S 0 ( γ nh f m ) S m ( d γ nh f m ) ψ ( γ z m ) W ξ S 0 ( a ( γ ) g S 0 ) . (34)
On the other hand, expanding the right-hand side of Proposition 4.2 gives the same expression without the factor S m S_{\mathfrak{m}} S m . Subtracting these identities and dividing by the nonzero scalar χ π ( f 1 ) \chi_\pi(f_1) χ π ( f 1 ) yields
∑ γ ∈ Γ m c ~ S 0 ( γ n h f m ) ( S m ( d γ n h f m ) − 1 ) × ψ ( γ z m ) W ~ ξ S 0 ( a ( γ ) g S 0 ) = 0. (35) \sum_{\gamma\in\Gamma_{\mathfrak{m}}}\widetilde{c}^{S_0}(\gamma nhf_{\mathfrak{m}})\left(S_{\mathfrak{m}}(d\gamma nhf_{\mathfrak{m}})-1\right)
\times\psi(\gamma z_{\mathfrak{m}})\widetilde{W}_{\xi_{S_0}}(a(\gamma)g_{S_0})=0. \tag*{(35)} γ ∈ Γ m ∑ c S 0 ( γ nh f m ) ( S m ( d γ nh f m ) − 1 ) × ψ ( γ z m ) W ξ S 0 ( a ( γ ) g S 0 ) = 0. (35)
This holds for every smooth vector ξ S 0 \xi_{S_0} ξ S 0 in the space of π S 0 \pi_{S_0} π S 0 .
The set Γ m \Gamma_{\mathfrak{m}} Γ m is the set of nonzero elements of the fractional ideal
L m = ( d n H m ) − 1 \mathcal{L}_{\mathfrak{m}}=(\mathfrak{d}\mathfrak{n}\mathfrak{H}\mathfrak{m})^{-1} L m = ( dnHm ) − 1
of A. Apply Lemma 4.3 to π ~ \widetilde{\pi} π , with this fractional ideal, γ 0 ∈ Γ m \gamma_0 \in\Gamma_{\mathfrak m} γ 0 ∈ Γ m and g S 0 g_{S_0} g S 0 . The resulting pure tensor of Whittaker functions can be realized as W ~ ξ S 0 \widetilde{W}_{\xi_{S_0}} W ξ S 0 for a suitable pure tensor ξ S 0 \xi_{S_0} ξ S 0 . Applying (4.14) to this vector, we obtain
C ~ S 0 ( γ 0 n h f m ) ( S m ( d γ 0 n h f m ) − 1 ) ψ ( γ 0 z m ) = 0. \widetilde{\mathcal{C}}^{S_0}(\gamma_0 n h f_{\mathfrak m})\left(S_{\mathfrak m}(d\gamma_0 n h f_{\mathfrak m})-1\right)\psi(\gamma_0 z_{\mathfrak m})=0. C S 0 ( γ 0 nh f m ) ( S m ( d γ 0 nh f m ) − 1 ) ψ ( γ 0 z m ) = 0.
Since the additive character never vanishes and γ 0 \gamma_0 γ 0 was arbitrary, it follows that
C ~ S 0 ( γ n h f m ) ( S m ( d γ n h f m ) − 1 ) = 0 ( γ ∈ Γ m ) . (36) \widetilde{\mathcal{C}}^{S_0}(\gamma n h f_{\mathfrak m})\left(S_{\mathfrak m}(d\gamma n h f_{\mathfrak m})-1\right)=0 \qquad(\gamma\in\Gamma_{\mathfrak m}). \tag*{(36)} C S 0 ( γ nh f m ) ( S m ( d γ nh f m ) − 1 ) = 0 ( γ ∈ Γ m ) . (36)
Lemma 4.5. For any admissible moduli m ∣ m ′ \mathfrak m\mid\mathfrak m' m ∣ m ′ , suppose
τ = f m ′ f m − 1 . \tau=f_{\mathfrak m'}f_{\mathfrak m}^{-1}. τ = f m ′ f m − 1 . If y v ∈ o v y_v\in\mathfrak{o}_v y v ∈ o v for every v ∣ m ′ v\mid\mathfrak m' v ∣ m ′ , then
S m ′ ( τ y ) = S m ( y ) , S_{\mathfrak m'}(\tau y)=S_{\mathfrak m}(y), S m ′ ( τ y ) = S m ( y ) , where the arguments are interpreted in R m ′ R_{\mathfrak m'} R m ′ and R m R_{\mathfrak m} R m , respectively.
Proof. Let
r : K m ′ ⟶ K m r:\mathcal{K}_{\mathfrak m'}\longrightarrow\mathcal{K}_{\mathfrak m} r : K m ′ ⟶ K m
be the natural reduction map, and put J = ker r J=\ker r J = ker r . This map is surjective, so that N m ′ = ∣ J ∣ N m N_{\mathfrak m'}=|J|N_{\mathfrak m} N m ′ = ∣ J ∣ N m . First, note that
Ψ m ′ ( ( 1 − u ′ ) τ y ) = Ψ m ( ( 1 − r ( u ′ ) ) y ) . \Psi_{\mathfrak m'}((1-u')\tau y)=\Psi_{\mathfrak m}((1-r(u'))y). Ψ m ′ (( 1 − u ′ ) τ y ) = Ψ m (( 1 − r ( u ′ )) y ) .
Indeed, if a v = ord v ( m ) a_v=\operatorname{ord}_v(\mathfrak m) a v = ord v ( m ) and a v ′ = ord v ( m ′ ) a'_v=\operatorname{ord}_v(\mathfrak m') a v ′ = ord v ( m ′ ) , then τ v = ϖ v a v ′ − a v \tau_v=\varpi_v^{a'_v-a_v} τ v = ϖ v a v ′ − a v . Thus, at v ∣ m v\mid\mathfrak m v ∣ m ,
ψ v ( d v − 1 ϖ v − a v ′ ( 1 − u v ′ ) τ v y v ) = ψ v ( d v − 1 ϖ v − a v ( 1 − u v ′ ) y v ) . \psi_v\left(d_v^{-1}\varpi_v^{-a'_v}(1-u'_v)\tau_v y_v\right)
=\psi_v\left(d_v^{-1}\varpi_v^{-a_v}(1-u'_v)y_v\right). ψ v ( d v − 1 ϖ v − a v ′ ( 1 − u v ′ ) τ v y v ) = ψ v ( d v − 1 ϖ v − a v ( 1 − u v ′ ) y v ) .
At v ∣ m ′ v\mid\mathfrak m' v ∣ m ′ with v ∤ m v\nmid\mathfrak m v ∤ m , the corresponding factor is
ψ v ( d v − 1 ( 1 − u v ′ ) y v ) = 1 , \psi_v\left(d_v^{-1}(1-u'_v)y_v\right)=1, ψ v ( d v − 1 ( 1 − u v ′ ) y v ) = 1 ,
since ( 1 − u v ′ ) y v ∈ o v (1-u'_v)y_v\in\mathfrak{o}_v ( 1 − u v ′ ) y v ∈ o v . Thus
S m ′ ( τ y ) = 1 N m ′ ∑ u ′ ∈ K m ′ ϵ ^ m ′ ( u ′ ) Ψ m ′ ( ( 1 − u ′ ) τ y ) = 1 N m ′ ∑ u ∈ K m Ψ m ( ( 1 − u ) y ) ∑ { u ′ : r ( u ′ ) = u } ϵ ^ m ′ ( u ′ ) . S_{\mathfrak m'}(\tau y)
=\frac{1}{N_{\mathfrak m'}}\sum_{u'\in\mathcal{K}_{\mathfrak m'}}\widehat{\epsilon}_{\mathfrak m'}(u')\Psi_{\mathfrak m'}((1-u')\tau y)
=\frac{1}{N_{\mathfrak m'}}\sum_{u\in\mathcal{K}_{\mathfrak m}}\Psi_{\mathfrak m}((1-u)y)\sum_{\{u':r(u')=u\}}\widehat{\epsilon}_{\mathfrak m'}(u'). S m ′ ( τ y ) = N m ′ 1 u ′ ∈ K m ′ ∑ ϵ m ′ ( u ′ ) Ψ m ′ (( 1 − u ′ ) τ y ) = N m ′ 1 u ∈ K m ∑ Ψ m (( 1 − u ) y ) { u ′ : r ( u ′ ) = u } ∑ ϵ m ′ ( u ′ ) .
To evaluate the inner sum, expand ϵ ^ m ′ \widehat{\epsilon}_{\mathfrak m'} ϵ m ′ :
∑ { u ′ : r ( u ′ ) = u } ϵ ^ m ′ ( u ′ ) = ∑ ρ ′ ∈ K ^ m ′ ϵ [ ρ ′ ] ∑ { u ′ : r ( u ′ ) = u } ρ ′ − 1 ( u ′ ) . \sum_{\{u':r(u')=u\}}\widehat{\epsilon}_{\mathfrak m'}(u')
=\sum_{\rho'\in\widehat{\mathcal{K}}_{\mathfrak m'}}\epsilon_{[\rho']}\sum_{\{u':r(u')=u\}}{\rho'}^{-1}(u'). { u ′ : r ( u ′ ) = u } ∑ ϵ m ′ ( u ′ ) = ρ ′ ∈ K m ′ ∑ ϵ [ ρ ′ ] { u ′ : r ( u ′ ) = u } ∑ ρ ′ − 1 ( u ′ ) .
The inner sum vanishes unless ρ ′ \rho' ρ ′ is trivial on J J J . In that case, ρ ′ = ρ ∘ r \rho'=\rho\circ r ρ ′ = ρ ∘ r for a unique ρ ∈ K ^ m \rho\in\widehat{\mathcal{K}}_{\mathfrak m} ρ ∈ K m , and the inner sum equals ∣ J ∣ ρ − 1 ( u ) |J|\rho^{-1}(u) ∣ J ∣ ρ − 1 ( u ) . Moreover, since there is a common lift ω ∈ X ( m ) \omega\in X(\mathfrak m) ω ∈ X ( m ) of ρ \rho ρ and ρ ∘ r \rho\circ r ρ ∘ r , we have
ϵ [ ρ ∘ r ] = ϵ [ ρ ] . \epsilon_{[\rho\circ r]}=\epsilon_{[\rho]}. ϵ [ ρ ∘ r ] = ϵ [ ρ ] .
Consequently,
∑ { u ′ : r ( u ′ ) = u } ϵ ^ m ′ ( u ′ ) = ∣ J ∣ ϵ ^ m ( u ) . \sum_{\{u':r(u')=u\}}\widehat{\epsilon}_{\mathfrak m'}(u')=|J|\widehat{\epsilon}_{\mathfrak m}(u). { u ′ : r ( u ′ ) = u } ∑ ϵ m ′ ( u ′ ) = ∣ J ∣ ϵ m ( u ) .
Substitution now gives
S m ′ ( τ y ) = ∣ J ∣ N m ′ ∑ u ∈ K m ϵ ^ m ( u ) Ψ m ( ( 1 − u ) y ) = 1 N m ∑ u ∈ K m ϵ ^ m ( u ) Ψ m ( ( 1 − u ) y ) = S m ( y ) . \begin{aligned}
S_{\mathfrak m'}(\tau y)=\frac{|J|}{N_{\mathfrak m'}}\sum_{u\in K_{\mathfrak m}}\widehat{\epsilon}_{\mathfrak m}(u)\Psi_{\mathfrak m}((1-u)y) \\
&=\frac{1}{N_{\mathfrak m}}\sum_{u\in K_{\mathfrak m}}\widehat{\epsilon}_{\mathfrak m}(u)\Psi_{\mathfrak m}((1-u)y)=S_{\mathfrak m}(y).
\end{aligned} S m ′ ( τ y ) = N m ′ ∣ J ∣ u ∈ K m ∑ ϵ m ( u ) Ψ m (( 1 − u ) y ) = N m 1 u ∈ K m ∑ ϵ m ( u ) Ψ m (( 1 − u ) y ) = S m ( y ) .
as required. □
Proposition 4.6. For every admissible modulus m \mathfrak m m ,
S m ( x ) = 1 ∀ x ∈ R m . S_{\mathfrak m}(x)=1\qquad\forall x\in R_{\mathfrak m}. S m ( x ) = 1 ∀ x ∈ R m . Proof. Fix x ∈ R m x\in R_{\mathfrak m} x ∈ R m . Choose D \mathfrak D D as in Lemma 4.4. Thus
[ D ] = [ n d 2 ] , ( D , m ) = 1 , c ~ S 0 ( D d − 1 ) ≠ 0. [\mathfrak D]=[\mathfrak n\mathfrak d^{2}],\qquad(\mathfrak D,\mathfrak m)=1,\qquad\widetilde{c}^{S_0}(\mathfrak D\mathfrak d^{-1})\ne0. [ D ] = [ n d 2 ] , ( D , m ) = 1 , c S 0 ( D d − 1 ) = 0. Choose an effective divisor H \mathfrak H H such that
[ H ] = [ d ] in Cl ( A ) , ( H , m n D ) = 1. [\mathfrak H]=[\mathfrak d]\ \text{in }\operatorname{Cl}(A),\qquad(\mathfrak H,\mathfrak m\mathfrak n\mathfrak D)=1. [ H ] = [ d ] in Cl ( A ) , ( H , mnD ) = 1. Such a choice is possible by the Chinese remainder theorem: choose 0 ≠ b ∈ d − 1 0\ne b\in\mathfrak d^{-1} 0 = b ∈ d − 1 whose image in d − 1 / p d − 1 \mathfrak d^{-1}/\mathfrak p\mathfrak d^{-1} d − 1 / p d − 1 is nonzero at every prime p \mathfrak p p dividing m n D \mathfrak m\mathfrak n\mathfrak D mnD , and put H = ( b ) d \mathfrak H=(b)\mathfrak d H = ( b ) d . Let h h h represent H \mathfrak H H as in §4.1.
Since [ d n H ] = [ D ] [\mathfrak d\mathfrak n\mathfrak H]=[\mathfrak D] [ dnH ] = [ D ] , choose δ ∈ F × \delta\in F^\times δ ∈ F × such that
( δ ) d n H = D (\delta)\mathfrak d\mathfrak n\mathfrak H=\mathfrak D ( δ ) dnH = D as divisors of A A A .
Another application of the Chinese remainder theorem to the A A A -module m − 1 \mathfrak m^{-1} m − 1 , choose 0 ≠ a ∈ m − 1 0\ne a\in\mathfrak m^{-1} 0 = a ∈ m − 1 such that
d v δ a n v h v f m , v ≡ x v ( m o d p v ord v ( m ) ) ( v ∣ m ) , d_v\delta a\,n_vh_vf_{\mathfrak m,v}\equiv x_v\pmod{\mathfrak p_v^{\operatorname{ord}_v(\mathfrak m)}}\qquad(v\mid\mathfrak m), d v δ a n v h v f m , v ≡ x v ( mod p v ord v ( m ) ) ( v ∣ m ) , a ≡ 1 ( m o d p v ord v ( n H ) ) ( v ∣ n H ) , a\equiv1\pmod{\mathfrak p_v^{\operatorname{ord}_v(\mathfrak n\mathfrak H)}}\qquad(v\mid\mathfrak n\mathfrak H), a ≡ 1 ( mod p v ord v ( nH ) ) ( v ∣ nH ) , and
a ≡ 1 ( m o d p v ) ( v ∣ D , v ∤ n H ) . a\equiv1\pmod{\mathfrak p_v}\qquad(v\mid\mathfrak D,\ v\nmid\mathfrak n\mathfrak H). a ≡ 1 ( mod p v ) ( v ∣ D , v ∤ nH ) . These conditions are compatible: the supports of m \mathfrak m m and n H D \mathfrak n\mathfrak H\mathfrak D nHD are disjoint, and d v δ n v h v d_v\delta n_vh_v d v δ n v h v is a unit for every v ∣ m v\mid\mathfrak m v ∣ m , since
( δ ) d n H = D . (\delta)\mathfrak d\mathfrak n\mathfrak H=\mathfrak D. ( δ ) dnH = D . Thus, the first family prescribes residue classes in the localizations of m − 1 \mathfrak m^{-1} m − 1 .
In particular, a a a is integral at every place outside S 0 ∪ Supp ( m ) S_0\cup\operatorname{Supp}(\mathfrak m) S 0 ∪ Supp ( m ) and is a unit at every prime dividing n H D \mathfrak n\mathfrak H\mathfrak D nHD . Hence
I = ( a ) m \mathfrak I=(a)\mathfrak m I = ( a ) m is integral and satisfies
( I , n H D ) = 1. (\mathfrak I,\mathfrak n\mathfrak H\mathfrak D)=1. ( I , nHD ) = 1. Moreover, since a ≡ 1 ( m o d n H ) a\equiv1\pmod{\mathfrak n\mathfrak H} a ≡ 1 ( mod nH ) ,
[ I ] n H = [ m ] n H . [\mathfrak I]_{\mathfrak n\mathfrak H}=[\mathfrak m]_{\mathfrak n\mathfrak H}. [ I ] nH = [ m ] nH . Putting
γ = δ a , y = d γ n h f m , \gamma=\delta a,\qquad y=d\gamma nhf_{\mathfrak m}, γ = δ a , y = d γ nh f m , we have
div S 0 ( y ) = D I . \operatorname{div}^{S_0}(y)=\mathfrak D\mathfrak I. div S 0 ( y ) = DI .
Consequently, y y y is integral at every place outside S 0 S_0 S 0 and satisfies
y v ≡ x v ( m o d p v n v ) ( v ∣ m ) , y_v \equiv x_v \pmod{\mathfrak{p}_v^{n_v}} \qquad(v \mid\mathfrak{m}), y v ≡ x v ( mod p v n v ) ( v ∣ m ) ,
where n v = ord v ( m ) n_v=\operatorname{ord}_v(\mathfrak{m}) n v = ord v ( m ) .
Let d d d be the order of [ m ] n H [\mathfrak{m}]_{\mathfrak{n}_{\mathfrak{H}}} [ m ] n H in Cl n H ( A ) \operatorname{Cl}_{\mathfrak{n}_{\mathfrak{H}}}(A) Cl n H ( A ) . Choose an integer E ≥ 2 E \ge2 E ≥ 2 divisible by d d d . This ensures
[ m ] n H E = 1. [\mathfrak{m}]_{\mathfrak{n}_{\mathfrak{H}}}^{E}=1. [ m ] n H E = 1.
Since
[ J ] n H = [ m ] n H , [\mathfrak{J}]_{\mathfrak{n}_{\mathfrak{H}}}=[\mathfrak{m}]_{\mathfrak{n}_{\mathfrak{H}}}, [ J ] n H = [ m ] n H ,
it follows that
M = m J E − 1 \mathfrak{M}=\mathfrak{m}\mathfrak{J}^{E-1} M = m J E − 1
is trivial in Cl n H ( A ) \operatorname{Cl}_{\mathfrak{n}_{\mathfrak{H}}}(A) Cl n H ( A ) . Consequently, there exists M ∈ F × M\in F^\times M ∈ F × such that
div S 0 ( M ) = M , M ≡ 1 ( m o d n H ) , \operatorname{div}^{S_0}(M)=\mathfrak{M}, \qquad M\equiv1 \pmod{\mathfrak{n}_{\mathfrak{H}}}, div S 0 ( M ) = M , M ≡ 1 ( mod n H ) ,
as required by Proposition 4.2. Also, ( M , n H ) = 1 (\mathfrak{M},\mathfrak{n}_{\mathfrak{H}})=1 ( M , n H ) = 1 .
Second, we also require
( L ( d v ) W π ~ v ) ( a ( ϖ v E r v ) ) ≠ 0 ( v ∣ J ) , (L(\mathfrak{d}_v)W_{\widetilde{\pi}_v})(a(\varpi_v^{E r_v}))\ne0 \qquad(v\mid\mathfrak{J}), ( L ( d v ) W π v ) ( a ( ϖ v E r v )) = 0 ( v ∣ J ) ,
where r v = ord v ( J ) r_v=\operatorname{ord}_v(\mathfrak{J}) r v = ord v ( J ) . Since ( J , n ) = 1 (\mathfrak{J},\mathfrak{n})=1 ( J , n ) = 1 , the representation π v \pi_v π v is unramified. Let α v , β v \alpha_v,\beta_v α v , β v be its Satake parameters, and write ζ v = α v / β v \zeta_v=\alpha_v/\beta_v ζ v = α v / β v . If α v ≠ β v \alpha_v\ne\beta_v α v = β v , then
λ E r v = α v E r v + 1 − β v E r v + 1 α v − β v = β v E r v ζ v E r v + 1 − 1 ζ v − 1 . \lambda_{E r_v}=\frac{\alpha_v^{E r_v+1}-\beta_v^{E r_v+1}}{\alpha_v-\beta_v}
=\beta_v^{E r_v}\frac{\zeta_v^{E r_v+1}-1}{\zeta_v-1}. λ E r v = α v − β v α v E r v + 1 − β v E r v + 1 = β v E r v ζ v − 1 ζ v E r v + 1 − 1 .
Thus λ E r v = 0 \lambda_{E r_v}=0 λ E r v = 0 precisely when ζ v E r v + 1 = 1 \zeta_v^{E r_v+1}=1 ζ v E r v + 1 = 1 . If ζ v \zeta_v ζ v is not a root of unity, this cannot occur. If ζ v ≠ 1 \zeta_v\ne1 ζ v = 1 has finite order d v d_v d v , we additionally require d v ∣ E d_v\mid E d v ∣ E . Then
ζ v E r v + 1 = ζ v ≠ 1 , \zeta_v^{E r_v+1}=\zeta_v\ne1, ζ v E r v + 1 = ζ v = 1 ,
so again λ E r v ≠ 0 \lambda_{E r_v}\ne0 λ E r v = 0 . Finally, if α v = β v \alpha_v=\beta_v α v = β v , then
λ E r v = ( E r v + 1 ) α v E r v ≠ 0. \lambda_{E r_v}=(E r_v+1)\alpha_v^{E r_v}\ne0. λ E r v = ( E r v + 1 ) α v E r v = 0.
There are only finitely many primes dividing J \mathfrak{J} J . We may therefore choose E ≥ 2 E\ge2 E ≥ 2 to be divisible by d d d and by every order d v d_v d v arising above. This choice ensures both that
[ M ] n H = 1 [\mathfrak{M}]_{\mathfrak{n}_{\mathfrak{H}}}=1 [ M ] n H = 1
and that
λ E r v ≠ 0 ( v ∣ J ) . \lambda_{E r_v}\ne0 \qquad(v\mid\mathfrak{J}). λ E r v = 0 ( v ∣ J ) .
The corresponding coefficients for π ~ v \widetilde{\pi}_v π v are also nonzero, since
λ ~ E r v = ( α v β v ) − E r v λ E r v . \widetilde{\lambda}_{E r_v}=(\alpha_v\beta_v)^{-E r_v}\lambda_{E r_v}. λ E r v = ( α v β v ) − E r v λ E r v .
We can now use (36) with m = M \mathfrak{m}=\mathfrak{M} m = M and the auxiliary divisor H \mathfrak{H} H chosen above. Since
div S 0 ( d γ n H f M ) = D J E , \operatorname{div}^{S_0}(d\gamma\mathfrak{n}_{\mathfrak{H}}f_{\mathfrak{M}})=\mathfrak{D}\mathfrak{J}^{E}, div S 0 ( d γ n H f M ) = D J E ,
we have γ ∈ Γ M \gamma\in\Gamma_{\mathfrak{M}} γ ∈ Γ M and
div S 0 ( γ n H f M ) = D J E Q − 1 . \operatorname{div}^{S_0}(\gamma\mathfrak{n}_{\mathfrak{H}}f_{\mathfrak{M}})=\mathfrak{D}\mathfrak{J}^{E}\mathfrak{Q}^{-1}. div S 0 ( γ n H f M ) = D J E Q − 1 .
We compare the local factors of
c ~ S 0 ( D J E Q − 1 ) \widetilde{c}^{S_0}(\mathfrak{D}\mathfrak{J}^{E}\mathfrak{Q}^{-1}) c S 0 ( D J E Q − 1 )
with those of c ~ S 0 ( D D − 1 ) ≠ 0 \widetilde{c}^{S_0}(\mathcal{D}\mathcal{D}^{-1}) \ne0 c S 0 ( D D − 1 ) = 0 . Indeed, for v ∉ S 0 v \notin S_0 v ∈ / S 0 with v ∤ I v \nmid\mathcal{I} v ∤ I , we have
ord v ( D I E D − 1 ) = ord v ( D D − 1 ) . \operatorname{ord}_v(\mathcal{D}\mathcal{I}^{E}\mathcal{D}^{-1})=\operatorname{ord}_v(\mathcal{D}\mathcal{D}^{-1}). ord v ( D I E D − 1 ) = ord v ( D D − 1 ) .
Since the local factor depends only on this valuation, it is unchanged and remains nonzero.
For v ∣ I v \mid\mathcal{I} v ∣ I , the coprimality ( D , I ) = 1 (\mathcal{D},\mathcal{I})=1 ( D , I ) = 1 gives
ord v ( D ) = 0 , ord v ( D I E ) = E r v . \operatorname{ord}_v(\mathcal{D})=0,\qquad\operatorname{ord}_v(\mathcal{D}\mathcal{I}^{E})=E r_v. ord v ( D ) = 0 , ord v ( D I E ) = E r v .
Thus, after the translation by L ( d v ) L(\mathbf{d}_v) L ( d v ) , the relevant essential Whittaker value is at a ( ϖ E r v ) a(\varpi^{E r_v}) a ( ϖ E r v ) . Our choice of E E E ensures that this value, and hence the corresponding dual local factor, is nonzero. Consequently,
c ~ S 0 ( γ n h f M ) = c ~ S 0 ( D I E D − 1 ) ≠ 0. \widetilde{c}^{S_0}(\gamma n h f_{\mathcal{M}})=\widetilde{c}^{S_0}(\mathcal{D}\mathcal{I}^{E}\mathcal{D}^{-1})\ne0. c S 0 ( γ nh f M ) = c S 0 ( D I E D − 1 ) = 0.
From (36) with modulus M \mathcal{M} M , we have
c ~ S 0 ( γ n h f M ) ( S M ( d γ n h f M ) − 1 ) = 0 , \widetilde{c}^{S_0}(\gamma n h f_{\mathcal{M}})\left(S_{\mathcal{M}}(\mathrm{d}\gamma n h f_{\mathcal{M}})-1\right)=0, c S 0 ( γ nh f M ) ( S M ( d γ nh f M ) − 1 ) = 0 ,
from which we conclude
S M ( d γ n h f M ) = 1. S_{\mathcal{M}}(\mathrm{d}\gamma n h f_{\mathcal{M}})=1. S M ( d γ nh f M ) = 1.
Finally, put
τ = f M f m − 1 , y = d γ n h f m . \tau=f_{\mathcal{M}}f_{\mathfrak m}^{-1},\qquad y=\mathrm{d}\gamma n h f_{\mathfrak m}. τ = f M f m − 1 , y = d γ nh f m .
Since div S 0 ( y ) = D I \operatorname{div}^{S_0}(y)=\mathcal{D}\mathcal{I} div S 0 ( y ) = D I , the integrality requirement in Lemma 4.5 is satisfied. That lemma and the choice of a a a give
1 = S M ( τ y ) = S m ( y ) = S m ( x ) . 1=S_{\mathcal{M}}(\tau y)=S_{\mathfrak m}(y)=S_{\mathfrak m}(x). 1 = S M ( τ y ) = S m ( y ) = S m ( x ) .
Since x x x was arbitrary, the proposition follows. □ \square □
Corollary 4.7. For every admissible modulus m \mathfrak m m and ω ∈ X ( m ) \omega\in X(\mathfrak m) ω ∈ X ( m ) , the normalized scalar ϵ [ ω ] \epsilon_{[\omega]} ϵ [ ω ] equals 1.
Proof. By the proposition, S m ( x ) = 1 S_{\mathfrak m}(x)=1 S m ( x ) = 1 for every x ∈ R m x\in R_{\mathfrak m} x ∈ R m . Fix u 0 ∈ K m u_0\in K_{\mathfrak m} u 0 ∈ K m . Multiplying the defining identity for S m ( x ) S_{\mathfrak m}(x) S m ( x ) by Ψ m ( ( u 0 − 1 ) x ) \Psi_{\mathfrak m}((u_0-1)x) Ψ m (( u 0 − 1 ) x ) and averaging over R m R_{\mathfrak m} R m , we obtain
1 { 1 } ( u 0 ) = 1 ∣ R m ∣ ∑ x ∈ R m Ψ m ( ( u 0 − 1 ) x ) \mathbf{1}_{\{1\}}(u_0)=\frac{1}{|R_{\mathfrak m}|}\sum_{x\in R_{\mathfrak m}}\Psi_{\mathfrak m}((u_0-1)x) 1 { 1 } ( u 0 ) = ∣ R m ∣ 1 x ∈ R m ∑ Ψ m (( u 0 − 1 ) x ) = 1 N m ∑ u ∈ K m ϵ ^ m ( u ) ( 1 ∣ R m ∣ ∑ x ∈ R m Ψ m ( ( u 0 − u ) x ) ) =\frac{1}{N_{\mathfrak m}}\sum_{u\in K_{\mathfrak m}}\widehat{\epsilon}_{\mathfrak m}(u)\left(\frac{1}{|R_{\mathfrak m}|}\sum_{x\in R_{\mathfrak m}}\Psi_{\mathfrak m}((u_0-u)x)\right) = N m 1 u ∈ K m ∑ ϵ m ( u ) ( ∣ R m ∣ 1 x ∈ R m ∑ Ψ m (( u 0 − u ) x ) ) = ϵ ^ m ( u 0 ) N m , =\frac{\widehat{\epsilon}_{\mathfrak m}(u_0)}{N_{\mathfrak m}}, = N m ϵ m ( u 0 ) , where the first and last equalities follow from additive character orthogonality and the nondegeneracy of the pairing ( a , x ) ↦ Ψ m ( a x ) (a,x)\mapsto\Psi_{\mathfrak m}(ax) ( a , x ) ↦ Ψ m ( a x ) . Thus
ϵ ^ m ( u ) = N m 1 { 1 } ( u ) . \widehat{\epsilon}_{\mathfrak m}(u)=N_{\mathfrak m}\mathbf{1}_{\{1\}}(u). ϵ m ( u ) = N m 1 { 1 } ( u ) . From this, using multiplicative Fourier inversion, we obtain
ϵ [ ω ] = 1 N m ∑ u ∈ K m ϵ ^ m ( u ) ω S ( u ) = 1. \epsilon_{[\omega]}=\frac{1}{N_{\mathfrak m}}\sum_{u\in K_{\mathfrak m}}\widehat{\epsilon}_{\mathfrak m}(u)\omega^{S}(u)=1. ϵ [ ω ] = N m 1 u ∈ K m ∑ ϵ m ( u ) ω S ( u ) = 1.
We use the following variant of the converse theorem of [10 ] for G L ( 2 ) \mathrm{GL}(2) GL ( 2 ) , allowing a fixed nonzero scalar in the functional equations. The same modification applies to G L ( N ) \mathrm{GL}(N) GL ( N ) , provided the scalar is independent of the twisting representation and the remaining hypotheses of the cited converse theorem are satisfied.
Lemma 4.8. Let π = ⨂ v ′ π v \pi= \bigotimes_v' \pi_v π = ⨂ v ′ π v be an irreducible admissible generic representation of G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) with automorphic central character, and let Σ \Sigma Σ be a finite set of finite places. Suppose that, for every unitary idele class character χ \chi χ unramified at all places in Σ \Sigma Σ , the completed L L L -functions
Λ ( s , π × χ ) , Λ ( s , π ~ × χ − 1 ) \Lambda(s,\pi\times\chi), \qquad\Lambda(s,\widetilde{\pi} \times\chi^{-1}) Λ ( s , π × χ ) , Λ ( s , π × χ − 1 ) converge absolutely in a right half-plane, extend to entire functions bounded in finite vertical strips, and satisfy
Λ ( s , π × χ ) = c ϵ ( s , π × χ , ψ ) Λ ( 1 − s , π ~ × χ − 1 ) , \Lambda(s,\pi\times\chi)=c\epsilon(s,\pi\times\chi,\psi)\Lambda(1-s,\widetilde{\pi} \times\chi^{-1}), Λ ( s , π × χ ) = cϵ ( s , π × χ , ψ ) Λ ( 1 − s , π × χ − 1 ) , where c ≠ 0 c \ne0 c = 0 is independent of χ \chi χ . Then there exists a unique isobaric automorphic representation Π \Pi Π of G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) such that Π v ≃ π v \Pi_v \simeq\pi_v Π v ≃ π v at every nonarchimedean place v ∉ Σ v \notin\Sigma v ∈ / Σ where π v \pi_v π v is unramified.
Proof. We follow the proof of the converse theorem in [10 § 2] , with N = 2 N=2 N = 2 and auxiliary character η = 1 \eta=1 η = 1 , retaining the scalar in the functional equations. We use the notation and local test vectors of that proof. In particular, fix the prescribed vectors ξ Σ \xi^\Sigma ξ Σ at the places in Σ \Sigma Σ , allow ξ Σ \xi^\Sigma ξ Σ to vary, and put ξ = ξ 0 Σ ⊗ ξ Σ \xi=\xi_0^\Sigma\otimes\xi^\Sigma ξ = ξ 0 Σ ⊗ ξ Σ . Let U ξ U_\xi U ξ and V ξ V_\xi V ξ be the two mirabolic Whittaker sums constructed there, and let
G 0 = K 0 , Σ ( a ) G Σ G_0=K_{0,\Sigma}(a)G^\Sigma G 0 = K 0 , Σ ( a ) G Σ
be the subgroup occurring in their proof. The functional equations and Mellin inversion now give
U ξ ( g ) = c V ξ ( g ) ( g ∈ G 0 ) , U_\xi(g)=cV_\xi(g)\qquad(g\in G_0), U ξ ( g ) = c V ξ ( g ) ( g ∈ G 0 ) ,
in place of U ξ ( g ) = V ξ ( g ) U_\xi(g)=V_\xi(g) U ξ ( g ) = V ξ ( g ) . The same scalar occurs for every ξ Σ \xi^\Sigma ξ Σ .
Let
P 0 = P ( F ) ∩ G 0 , Q 0 = Q ( F ) ∩ G 0 , P_0=P(F)\cap G_0,\qquad Q_0=Q(F)\cap G_0, P 0 = P ( F ) ∩ G 0 , Q 0 = Q ( F ) ∩ G 0 ,
where P P P and Q Q Q are the mirabolic subgroups used in the cited proof. The function U ξ U_\xi U ξ is left P 0 P_0 P 0 -invariant by construction. For q ∈ Q 0 q\in Q_0 q ∈ Q 0 and g ∈ G 0 g\in G_0 g ∈ G 0 , the left Q ( F ) Q(F) Q ( F ) -invariance of V ξ V_\xi V ξ gives
U ξ ( q g ) = c V ξ ( q g ) = c V ξ ( g ) = U ξ ( g ) . U_\xi(qg)=cV_\xi(qg)=cV_\xi(g)=U_\xi(g). U ξ ( q g ) = c V ξ ( q g ) = c V ξ ( g ) = U ξ ( g ) .
Thus U ξ ∣ G 0 U_\xi|_{G_0} U ξ ∣ G 0 is invariant under P 0 P_0 P 0 and Q 0 Q_0 Q 0 . It is also invariant under rational central elements, since the central character is automorphic. The remaining generation, approximation, and representation-theoretic arguments in the cited proof are unchanged by the fixed scalar. Thus we obtain an irreducible automorphic representation Π ′ \Pi' Π ′ with Π v ′ ≃ π v \Pi'_v\simeq\pi_v Π v ′ ≃ π v for every v ∉ Σ v\notin\Sigma v ∈ / Σ .
Finally, let Π \Pi Π be the isobaric automorphic representation associated to the cuspidal support of Π ′ \Pi' Π ′ . By the argument in the proof of [6 Lemma 3.5] , this replacement preserves the unramified local components. Hence Π v ≃ Π v ′ ≃ π v \Pi_v\simeq\Pi'_v\simeq\pi_v Π v ≃ Π v ′ ≃ π v at every nonarchimedean place v ∉ Σ v\notin\Sigma v ∈ / Σ where π v \pi_v π v is unramified. The uniqueness of Π \Pi Π follows from strong multiplicity one for isobaric automorphic representations.
By Corollary 4.7 , the normalized scalar factors are all equal to 1 1 1 ; see Remark 4 in the Introduction. Undoing the normalization, we find that the original factors ϵ χ \epsilon_\chi ϵ χ are equal to a fixed nonzero constant, independent of χ \chi χ . The required boundedness in vertical strips follows by Phragmén–Lindelöf. Taking
Σ = { v < ∞ : π v is ramified } , \Sigma= \{v < \infty: \pi_v \text{ is ramified}\}, Σ = { v < ∞ : π v is ramified } ,
the hypotheses of Lemma 4.8 are therefore satisfied, and its conclusion gives the required isobaric automorphic representation. This completes the proof of Theorem 1.1 .
Email address: shantanuagr123@gmail.com
Email address: mkrish1939@gmail.com