Introduction
Thakur defined function field analogues of Gauss sums in [42 ] and established their fundamental properties, demonstrating a striking parallel with the classical theory. Then in the intervening years Gauss-Thakur sums have appeared as central quantities in various contexts (e.g., see [4 , 5 , 6 , 16 , 17 , 43 , 44 , 45 , 46 , 47 ] ). Gazda and Maurischat [19 ] expanded the definition of Gauss-Thakur sums, built from torsion on an arbitrary t t t -module, and made a crucial connection with modules of special functions, augmenting work of Anglès and Pellarin [5 ] for the Carlitz module. In the present paper we construct and investigate dual Gauss-Thakur sums associated to the a Drinfeld module and its adjoint, rooted in the Drinfeld module’s rigid analytic trivializations and built in part on work of Ferraro [17 ] .
Drinfeld modules
For a finite field F q \mathbb{F}_q F q with q = p m q = p^m q = p m elements, let A = F q [ θ ] A = \mathbb{F}_q[\theta] A = F q [ θ ] denote the ring of polynomials over F q \mathbb{F}_q F q in a variable θ \theta θ . Let k = F q ( θ ) k = \mathbb{F}_q(\theta) k = F q ( θ ) be its fraction field, let k ∞ = F q ( ( θ − 1 ) ) k_{\infty} = \mathbb{F}_{q}((\theta^{-1})) k ∞ = F q (( θ − 1 )) be the completion of k k k at its infinite place, and let C ∞ C_{\infty} C ∞ be the completion of an algebraic closure of k ∞ k_{\infty} k ∞ . We let deg = − ord ∞ : C ∞ → Q ∪ { − ∞ } \deg= -\operatorname{ord}_{\infty} : C_{\infty} \to\mathbb{Q} \cup\{-\infty\} deg = − ord ∞ : C ∞ → Q ∪ { − ∞ } , so for x ∈ C ∞ x \in C_{\infty} x ∈ C ∞ , ∣ x ∣ ∞ = q deg x |x|_{\infty} = q^{\deg x} ∣ x ∣ ∞ = q d e g x , and in particular ∣ θ ∣ ∞ = q |\theta|_{\infty} = q ∣ θ ∣ ∞ = q . We let A + A_{+} A + be the monic elements of A A A .
Let ϕ : A → K [ τ ] \phi: A \to K[\tau] ϕ : A → K [ τ ] be a Drinfeld module of rank r r r defined over an intermediate field k ⊆ K ⊆ C ∞ k \subseteq K \subseteq C_{\infty} k ⊆ K ⊆ C ∞ , where τ \tau τ is the q q q -th power Frobenius endomorphism,
ϕ θ = θ + b 1 τ + ⋯ + b r τ r , b r ≠ 0. (1) \phi_{\theta} = \theta+ b_{1}\tau+ \cdots+ b_{r}\tau^{r}, \qquad b_{r} \ne0. \tag*{(1)} ϕ θ = θ + b 1 τ + ⋯ + b r τ r , b r = 0. (1)
We construct also its adjoint ϕ ∗ : A → K ‾ [ τ − 1 ] \phi^{*} : A \to\overline{K}[\tau^{-1}] ϕ ∗ : A → K [ τ − 1 ] defined over the algebraic closure K ‾ \overline{K} K of K K K (see § 2.4 , § 2.5 ). Both ϕ \phi ϕ and ϕ ∗ \phi^{*} ϕ ∗ can be extended in natural ways (see § 2.7 , § 2.8 ) to F q \mathbb{F}_{q} F q -algebra maps ϕ : k ∞ → C ∞ ( ( τ ) ) \phi: k_{\infty} \to C_{\infty}((\tau)) ϕ : k ∞ → C ∞ (( τ )) and ϕ ∗ : k ∞ → C ∞ ( ( τ ) ) \phi^{*} : k_{\infty} \to C_{\infty}((\tau)) ϕ ∗ : k ∞ → C ∞ (( τ )) , which for α ∈ k ∞ \alpha\in k_{\infty} α ∈ k ∞ we express
ϕ α = ∑ n ( α n ) ϕ τ n , ϕ α ∗ = ∑ n τ n ( α n ) ϕ ∗ . \phi_{\alpha} = \sum_{n} \binom{\alpha}{n}_{\phi}\tau^{n}, \qquad\phi^{*}_{\alpha} = \sum_{n}\tau^{n}\binom{\alpha}{n}_{\phi^{*}}. ϕ α = n ∑ ( n α ) ϕ τ n , ϕ α ∗ = n ∑ τ n ( n α ) ϕ ∗ .
To ϕ \phi ϕ we associate two matrices Υ , Ψ ∈ GL r ( T ) \Upsilon, \Psi\in\operatorname{GL}_{r}(\mathbb{T}) Υ , Ψ ∈ GL r ( T ) , over the Tate algebra T ⊆ C ∞ [ [ t ] ] \mathbb{T} \subseteq C_{\infty}[[t]] T ⊆ C ∞ [[ t ]] of functions converging on the closed unit disk of C ∞ C_{\infty} C ∞ , which are rigid analytic trivializations for the t t t -motive and t t t -comotive of ϕ \phi ϕ respectively as in [2 , 9 , 10 , 11 , 27 , 29 , 32 , 33 , 35 ] . Specializations of these matrices at t = θ t = \theta t = θ produce periods and quasi-periods of ϕ \phi ϕ , and here we consider also specializations at t = ε ∈ F q t = \varepsilon\in\mathbb{F}_{q} t = ε ∈ F q , connected to Gauss-Thakur sums as in [5 , 19 ] (see (5) , (6) , Theorem 6.13 ). As part of these formulations we obtained new power series identities for Υ \Upsilon Υ and Ψ \Psi Ψ in terms of the multiplication coefficients of ϕ \phi ϕ and ϕ ∗ \phi^{*} ϕ ∗ . As a higher rank version of [2 Lem. 5.4.1] , we prove (Proposition 3.16) for N ∈ Z N \in\mathbb{Z} N ∈ Z ,
[ Ψ ( N ) Υ ] 11 = { ∑ m = 0 ∞ ( θ − m − 1 N ) ϕ t m , if N ≥ 0 , − ∑ m = 0 ∞ ( θ − m − 1 − N ) ϕ ∗ t m , if N < 0. (2) \left[\Psi^{(N)}\Upsilon\right]_{11} =
\begin{cases}
\sum_{m=0}^{\infty}\binom{\theta^{-m-1}}{N}_{\phi}t^{m}, & \text{if } N \ge0,\\[1em]
-\sum_{m=0}^{\infty}\binom{\theta^{-m-1}}{-N}_{\phi^{*}}t^{m}, & \text{if } N < 0.
\end{cases} \tag*{(2)} [ Ψ ( N ) Υ ] 11 = ⎩ ⎨ ⎧ ∑ m = 0 ∞ ( N θ − m − 1 ) ϕ t m , − ∑ m = 0 ∞ ( − N θ − m − 1 ) ϕ ∗ t m , if N ≥ 0 , if N < 0. (2)
where for a matrix M M M , we let M i j = [ M ] i j M_{ij} = [M]_{ij} M ij = [ M ] ij denote its ( i , j ) (i,j) ( i , j ) -entry. This identity underlies our main duality result (Theorem A) outlined below.
The first column of Υ \Upsilon Υ and the first row of Ψ \Psi Ψ ,
ω ϕ = ( ω 1 , … , ω r ) T , ζ ϕ = ( ζ 1 , … , ζ r ) , \omega_{\phi} = (\omega_{1}, \ldots, \omega_{r})^{\mathsf{T}}, \qquad\zeta_{\phi} = (\zeta_{1}, \ldots, \zeta_{r}), ω ϕ = ( ω 1 , … , ω r ) T , ζ ϕ = ( ζ 1 , … , ζ r ) ,
consist of special functions (or Anderson generating functions) and dual special functions as defined in [3 , 15 , 16 , 17 , 19 , 24 , 31 , 35 ] , which satisfy certain Frobenius difference equations. As presented here these functions depend on a choice of A A A -basis π 1 , … , π r \pi_{1}, \ldots, \pi_{r} π 1 , … , π r for the period lattice Λ ϕ \Lambda_{\phi} Λ ϕ . See § 3.1 for details.
While ω ϕ \omega_{\phi} ω ϕ and ζ ϕ \zeta_{\phi} ζ ϕ had appeared in the literature in various guises for a while, Ferraro [16 , 17 ] showed that they represent universal objects in a fundamental pairing between special functions and dual special functions (see Theorems 4.3 and 4.7 ). Moreover, after comparing previous work in [10 , 27 , 32 ] , it follows from Ferraro’s results (see Corollary 4.10 ) that each ζ i \zeta_{i} ζ i , previously arising through Anderson generating functions [10 , 27 ] , has a lattice sum incarnation,
ζ i = − ∑ ( a 1 , … , a r ) ∈ A r ′ a i ( t ) a 1 π 1 + ⋯ + a r π r . (3) \zeta_{i} = -\sum_{(a_{1},\ldots,a_{r})\in A^{r}}' \frac{a_{i}(t)}{a_{1}\pi_{1}+\cdots+a_{r}\pi_{r}}. \tag*{(3)} ζ i = − ( a 1 , … , a r ) ∈ A r ∑ ′ a 1 π 1 + ⋯ + a r π r a i ( t ) . (3)
where ∑ ′ \sum' ∑ ′ denotes that the sum omits the 0 0 0 -tuple.
Gauss-Thakur sums and dual Gauss-Thakur sums
For each 1 ≤ i ≤ r 1 \le i \le r 1 ≤ i ≤ r , we define continuous F q \mathbb{F}_q F q -linear functions e i , e i ∗ : k ∞ → C ∞ \mathbf{e}_i, \mathbf{e}_i^*: k_\infty\to C_\infty e i , e i ∗ : k ∞ → C ∞ so that
ω i = ∑ m = 0 ∞ e i ( θ − m − 1 ) t m , ζ i = ∑ m = 0 ∞ e i ∗ ( θ − m − 1 ) t m . \omega_i = \sum_{m=0}^{\infty} \mathbf{e}_i(\theta^{-m-1})t^m,\qquad\zeta_i = \sum_{m=0}^{\infty} \mathbf{e}_i^*(\theta^{-m-1})t^m. ω i = m = 0 ∑ ∞ e i ( θ − m − 1 ) t m , ζ i = m = 0 ∑ ∞ e i ∗ ( θ − m − 1 ) t m .
We have e i ( x ) = exp ϕ ( π i x ) \mathbf{e}_i(x)=\exp_\phi(\pi_i x) e i ( x ) = exp ϕ ( π i x ) , and e i ∗ ( x ) \mathbf{e}_i^*(x) e i ∗ ( x ) is similar to functions defined in [2 §5.2] (though note Remark 8.4). As a matter of course, for monic irreducible ℘ ∈ A \wp\in A ℘ ∈ A and F ℘ = A / ℘ \mathbb{F}_\wp=A/\wp F ℘ = A / ℘ , we have induced injections of A A A -modules,
z ↦ e i ( z / ℘ ) : F ℘ → ϕ [ ℘ ] , z ↦ e i ∗ ( z / ℘ ) : F ℘ → ϕ ∗ [ ℘ ] , z \mapsto\mathbf{e}_i(z/\wp):\mathbb{F}_\wp\to\phi[\wp],\qquad z \mapsto\mathbf{e}_i^*(z/\wp):\mathbb{F}_\wp\to\phi^*[\wp], z ↦ e i ( z / ℘ ) : F ℘ → ϕ [ ℘ ] , z ↦ e i ∗ ( z / ℘ ) : F ℘ → ϕ ∗ [ ℘ ] ,
where ϕ [ ℘ ] \phi[\wp] ϕ [ ℘ ] and ϕ ∗ [ ℘ ] \phi^*[\wp] ϕ ∗ [ ℘ ] are the ℘ \wp ℘ -torsion submodules of ϕ \phi ϕ and ϕ ∗ \phi^* ϕ ∗ respectively. Furthermore,
ϕ [ ℘ ] = ⨁ i = 1 r e i ( ℘ − 1 A ) , ϕ ∗ [ ℘ ] = ⨁ i = 1 r e i ∗ ( ℘ − 1 A ) , \phi[\wp]=\bigoplus_{i=1}^{r}\mathbf{e}_i(\wp^{-1}A),\qquad\phi^*[\wp]=\bigoplus_{i=1}^{r}\mathbf{e}_i^*(\wp^{-1}A), ϕ [ ℘ ] = i = 1 ⨁ r e i ( ℘ − 1 A ) , ϕ ∗ [ ℘ ] = i = 1 ⨁ r e i ∗ ( ℘ − 1 A ) ,
(see Lemma 5.19 ).
Gazda and Maurischat [19 ] defined Gauss-Thakur sums g i ( χ ; a ) g_i(\chi;a) g i ( χ ; a ) and we similarly define their duals g i ∗ ( χ ; a ) g_i^*(\chi;a) g i ∗ ( χ ; a ) in the following way. Letting χ ∈ F ^ q × \chi\in\widehat{\mathbb{F}}_q^{\times} χ ∈ F q × be a multiplicative character and a ∈ A a \in A a ∈ A , for 1 ≤ i ≤ r 1 \le i \le r 1 ≤ i ≤ r ,
g i ( χ ; a ) = − ∑ z ∈ F ℘ × χ ( z ) − 1 e i ( z / ℘ ) , g i ∗ ( χ ; a ) = − ∑ z ∈ F ℘ × χ ( z ) − 1 e i ∗ ( z / ℘ ) . (4) g_i(\chi;a)=-\sum_{z\in\mathbb{F}_\wp^{\times}}\chi(z)^{-1}\mathbf{e}_i(z/\wp),\qquad g_i^*(\chi;a)=-\sum_{z\in\mathbb{F}_\wp^{\times}}\chi(z)^{-1}\mathbf{e}_i^*(z/\wp). \tag*{(4)} g i ( χ ; a ) = − z ∈ F ℘ × ∑ χ ( z ) − 1 e i ( z / ℘ ) , g i ∗ ( χ ; a ) = − z ∈ F ℘ × ∑ χ ( z ) − 1 e i ∗ ( z / ℘ ) . (4)
When ϕ \phi ϕ is the Carlitz module, one recovers Thakur’s Gauss sums from [42 ] . These Gauss-Thakur sums and their duals satisfy a number of properties that parallel classical Gauss sums (see § 6.1 ), particularly that
g i ( χ ; a ) = χ ( a ) g i ( χ ) , g i ∗ ( χ ; a ) = χ ( a ) g i ∗ ( χ ) , g_i(\chi;a)=\chi(a)g_i(\chi),\qquad g_i^*(\chi;a)=\chi(a)g_i^*(\chi), g i ( χ ; a ) = χ ( a ) g i ( χ ) , g i ∗ ( χ ; a ) = χ ( a ) g i ∗ ( χ ) ,
where g i ( χ ) = g i ( χ ; 1 ) g_i(\chi)=g_i(\chi;1) g i ( χ ) = g i ( χ ; 1 ) and g i ∗ ( χ ) = g i ∗ ( χ ; 1 ) g_i^*(\chi)=g_i^*(\chi;1) g i ∗ ( χ ) = g i ∗ ( χ ; 1 ) . Gazda and Maurischat define also a tensor product version, which characterizes Gauss-Thakur sums as a class of special functions for ϕ \phi ϕ , and then dual sums arise as a class of dual special functions (see Lemma 6.6 ).
Special function evaluations
For d = deg ℘ d=\deg\wp d = deg ℘ and a root ε ∈ F q d ⊂ C ∞ \varepsilon\in\mathbb{F}_{q^d}\subset C_\infty ε ∈ F q d ⊂ C ∞ of ℘ \wp ℘ , the map χ ε : A → F q d \chi_\varepsilon:A\to\mathbb{F}_{q^d} χ ε : A → F q d defined by χ ε ( a ) = a ( ε ) \chi_\varepsilon(a)=a(\varepsilon) χ ε ( a ) = a ( ε ) induces an evaluative Dirichlet character modulo ℘ \wp ℘ . Thakur proved in the Carlitz case, and Gazda and Maurischat in general, that g i ( χ ; a ) = 0 g_i(\chi;a)=0 g i ( χ ; a ) = 0 if χ \chi χ is not evaluative. The same holds for g i ∗ ( χ ; a ) g_i^*(\chi;a) g i ∗ ( χ ; a ) , so we primarily consider only g i ( χ k ; a ) g_i(\chi_k;a) g i ( χ k ; a ) and g i ∗ ( χ k ; a ) g_i^*(\chi_k;a) g i ∗ ( χ k ; a ) , where χ k = χ ε q k \chi_k=\chi_{\varepsilon^{q^k}} χ k = χ ε q k .
One of the main theorems of Gazda and Maurischat [19 Thm. 5.2] identifies Gauss-Thakur sums with evaluations of ω ϕ \boldsymbol{\omega}_\phi ω ϕ ,
g i ( χ ε ; ℘ ′ ( θ ) ) = ω i ( ε ) , 1 ≤ i ≤ r , (5) g_i(\chi_\varepsilon;\wp'(\theta))=\omega_i(\varepsilon),\qquad1\le i\le r, \tag*{(5)} g i ( χ ε ; ℘ ′ ( θ )) = ω i ( ε ) , 1 ≤ i ≤ r , (5)
where ℘ ′ ( θ ) \wp'(\theta) ℘ ′ ( θ ) is the derivative of ℘ \wp ℘ . This result extends a theorem of Anglès and Pellarin [5 Thm. 2.9] for the Carlitz module and the Anderson-Thakur function ω C \omega_C ω C (see (71) ). We prove a version of this result (Theorem 6.13) for dual Gauss-Thakur sums in terms of ζ ϕ \boldsymbol{\zeta}_\phi ζ ϕ ,
g i ∗ ( χ ε ; ℘ ′ ( θ ) ) = ζ i ( ε ) , 1 ≤ i ≤ r . (6) g_i^*(\chi_\varepsilon;\wp'(\theta))=\zeta_i(\varepsilon),\qquad1\le i\le r. \tag*{(6)} g i ∗ ( χ ε ; ℘ ′ ( θ )) = ζ i ( ε ) , 1 ≤ i ≤ r . (6)
For k ∈ Z k\in\mathbb{Z} k ∈ Z , we also twist Gauss-Thakur sums,
g i [ k ] ( χ ; a ) = − ∑ z ∈ F ℘ × χ ( z ) − 1 e i ( z / ℘ ) q k , g i ∗ [ k ] ( χ ; a ) = − ∑ z ∈ F ℘ × χ ( z ) − 1 e i ∗ ( z / ℘ ) q k , g_i^{[k]}(\chi;a)=-\sum_{z\in\mathbb{F}_\wp^{\times}}\chi(z)^{-1}\mathbf{e}_i(z/\wp)^{q^k},\qquad g_i^{*[k]}(\chi;a)=-\sum_{z\in\mathbb{F}_\wp^{\times}}\chi(z)^{-1}\mathbf{e}_i^*(z/\wp)^{q^k}, g i [ k ] ( χ ; a ) = − z ∈ F ℘ × ∑ χ ( z ) − 1 e i ( z / ℘ ) q k , g i ∗ [ k ] ( χ ; a ) = − z ∈ F ℘ × ∑ χ ( z ) − 1 e i ∗ ( z / ℘ ) q k ,
and determine (Corollary 6.19) that
g i [ k ] ( χ ε ; O ′ ( θ ) ) = ω i ( k ) ( ε ) , g i ∗ [ k ] ( χ ε ; O ′ ( θ ) ) = ζ i ( k ) ( ε ) , 1 ≤ i ≤ r , g_i^{[k]}(\chi_{\varepsilon};\mathscr{O}'(\theta))=\omega_i^{(k)}(\varepsilon),\qquad g_i^{*[k]}(\chi_{\varepsilon};\mathscr{O}'(\theta))=\zeta_i^{(k)}(\varepsilon),\qquad1\le i\le r, g i [ k ] ( χ ε ; O ′ ( θ )) = ω i ( k ) ( ε ) , g i ∗ [ k ] ( χ ε ; O ′ ( θ )) = ζ i ( k ) ( ε ) , 1 ≤ i ≤ r ,
i.e., that they are evaluations of Frobenius twists (see § 3 ) of ω ϕ \omega_{\phi} ω ϕ and ζ ϕ \zeta_{\phi} ζ ϕ .
Matrices of Gauss–Thakur sums and dual Gauss–Thakur sums
Our main theorem below (Theorem A, and Theorem 7.7 later) is a matrix identity for Gauss–Thakur sums, their duals, and their twists. For χ ∈ F ^ q r \chi\in\widehat{\mathbb{F}}_q^r χ ∈ F q r and a ∈ A a\in A a ∈ A , define
G ( χ ; a ) : = ( g 1 ( χ ; a ) g 1 [ 1 ] ( χ ; a ) ⋯ g 1 [ r − 1 ] ( χ ; a ) g 2 ( χ ; a ) g 2 [ 1 ] ( χ ; a ) ⋯ g 2 [ r − 1 ] ( χ ; a ) ⋮ ⋮ ⋱ ⋮ g r ( χ ; a ) g r [ 1 ] ( χ ; a ) ⋯ g r [ r − 1 ] ( χ ; a ) ) ∈ Mat r × r ( K ‾ ) (7) \mathfrak{G}(\chi;a):=
\begin{pmatrix}
g_1(\chi;a)&g_1^{[1]}(\chi;a)&\cdots&g_1^{[r-1]}(\chi;a)\\
g_2(\chi;a)&g_2^{[1]}(\chi;a)&\cdots&g_2^{[r-1]}(\chi;a)\\
\vdots&\vdots&\ddots&\vdots\\
g_r(\chi;a)&g_r^{[1]}(\chi;a)&\cdots&g_r^{[r-1]}(\chi;a)
\end{pmatrix}\in\operatorname{Mat}_{r\times r}(\overline{K}) \tag*{(7)} G ( χ ; a ) := g 1 ( χ ; a ) g 2 ( χ ; a ) ⋮ g r ( χ ; a ) g 1 [ 1 ] ( χ ; a ) g 2 [ 1 ] ( χ ; a ) ⋮ g r [ 1 ] ( χ ; a ) ⋯ ⋯ ⋱ ⋯ g 1 [ r − 1 ] ( χ ; a ) g 2 [ r − 1 ] ( χ ; a ) ⋮ g r [ r − 1 ] ( χ ; a ) ∈ Mat r × r ( K ) (7)
and
G ∗ ( χ ; a ) : = ( g 1 ∗ ( χ ; a ) g 2 ∗ ( χ ; a ) ⋯ g r ∗ ( χ ; a ) g 1 ∗ [ 1 ] ( χ ; a ) g 2 ∗ [ 1 ] ( χ ; a ) ⋯ g r ∗ [ 1 ] ( χ ; a ) ⋮ ⋮ ⋱ ⋮ g 1 ∗ [ r − 1 ] ( χ ; a ) g 2 ∗ [ r − 1 ] ( χ ; a ) ⋯ g r ∗ [ r − 1 ] ( χ ; a ) ) ∈ Mat r × r ( K ‾ ) . (8) \mathfrak{G}^{*}(\chi;a):=
\begin{pmatrix}
g_1^{*}(\chi;a)&g_2^{*}(\chi;a)&\cdots&g_r^{*}(\chi;a)\\
g_1^{*[1]}(\chi;a)&g_2^{*[1]}(\chi;a)&\cdots&g_r^{*[1]}(\chi;a)\\
\vdots&\vdots&\ddots&\vdots\\
g_1^{*[r-1]}(\chi;a)&g_2^{*[r-1]}(\chi;a)&\cdots&g_r^{*[r-1]}(\chi;a)
\end{pmatrix}\in\operatorname{Mat}_{r\times r}(\overline{K}). \tag*{(8)} G ∗ ( χ ; a ) := g 1 ∗ ( χ ; a ) g 1 ∗ [ 1 ] ( χ ; a ) ⋮ g 1 ∗ [ r − 1 ] ( χ ; a ) g 2 ∗ ( χ ; a ) g 2 ∗ [ 1 ] ( χ ; a ) ⋮ g 2 ∗ [ r − 1 ] ( χ ; a ) ⋯ ⋯ ⋱ ⋯ g r ∗ ( χ ; a ) g r ∗ [ 1 ] ( χ ; a ) ⋮ g r ∗ [ r − 1 ] ( χ ; a ) ∈ Mat r × r ( K ) . (8)
Keeping with previous conventions, we set G ( χ ) : = G ( χ ; 1 ) \mathfrak{G}(\chi):=\mathfrak{G}(\chi;1) G ( χ ) := G ( χ ; 1 ) and G ∗ ( χ ) : = G ∗ ( χ ; 1 ) \mathfrak{G}^{*}(\chi):=\mathfrak{G}^{*}(\chi;1) G ∗ ( χ ) := G ∗ ( χ ; 1 ) . We find that just as above, these matrices are obtained by specializing our rigid analytic trivializations Υ \Upsilon Υ and Ψ \Psi Ψ (see (7.3)),
G ( χ ε ; O ′ ( θ ) ) = O ′ ( ε ) G ( χ ε ) = Υ ( ε ) , G ∗ ( χ ε ; O ′ ( θ ) ) = O ′ ( ε ) G ∗ ( χ ε ) = J Ψ ( τ r − 1 ) ( ε ) , \mathfrak{G}(\chi_{\varepsilon};\mathscr{O}'(\theta))=\mathscr{O}'(\varepsilon)\mathfrak{G}(\chi_{\varepsilon})=\Upsilon(\varepsilon),\qquad
\mathfrak{G}^{*}(\chi_{\varepsilon};\mathscr{O}'(\theta))=\mathscr{O}'(\varepsilon)\mathfrak{G}^{*}(\chi_{\varepsilon})=J\Psi(\tau^{r-1})(\varepsilon), G ( χ ε ; O ′ ( θ )) = O ′ ( ε ) G ( χ ε ) = Υ ( ε ) , G ∗ ( χ ε ; O ′ ( θ )) = O ′ ( ε ) G ∗ ( χ ε ) = J Ψ ( τ r − 1 ) ( ε ) ,
where
J = ( 1 ⋱ 1 ) ∈ Mat r × r ( F q ) . J=\begin{pmatrix}1&&\\&\ddots&\\&&1\end{pmatrix}\in\operatorname{Mat}_{r\times r}(\mathbb{F}_q). J = 1 ⋱ 1 ∈ Mat r × r ( F q ) .
Now for a ∈ A a\in A a ∈ A , define the r d × N rd\times\mathbb{N} r d × N matrix,
Γ ( ϕ ; a ) : = ( G ( χ 0 ; a ) G [ r ] ( χ 0 ; a ) G [ 2 r ] ( χ 0 ; a ) ⋯ G ( χ 1 ; a ) G [ r ] ( χ 1 ; a ) G [ 2 r ] ( χ 1 ; a ) ⋯ ⋮ ⋮ ⋮ G ( χ d − 1 ; a ) G [ r ] ( χ d − 1 ; a ) G [ 2 r ] ( χ d − 1 ; a ) ⋯ ) . (9) \Gamma(\phi;a):=
\begin{pmatrix}
\mathfrak{G}(\chi_0;a)&\mathfrak{G}^{[r]}(\chi_0;a)&\mathfrak{G}^{[2r]}(\chi_0;a)&\cdots\\
\mathfrak{G}(\chi_1;a)&\mathfrak{G}^{[r]}(\chi_1;a)&\mathfrak{G}^{[2r]}(\chi_1;a)&\cdots\\
\vdots&\vdots&\vdots&\\
\mathfrak{G}(\chi_{d-1};a)&\mathfrak{G}^{[r]}(\chi_{d-1};a)&\mathfrak{G}^{[2r]}(\chi_{d-1};a)&\cdots
\end{pmatrix}. \tag*{(9)} Γ ( ϕ ; a ) := G ( χ 0 ; a ) G ( χ 1 ; a ) ⋮ G ( χ d − 1 ; a ) G [ r ] ( χ 0 ; a ) G [ r ] ( χ 1 ; a ) ⋮ G [ r ] ( χ d − 1 ; a ) G [ 2 r ] ( χ 0 ; a ) G [ 2 r ] ( χ 1 ; a ) ⋮ G [ 2 r ] ( χ d − 1 ; a ) ⋯ ⋯ ⋯ . (9)
Define also the N × r d \mathbb{N}\times rd N × r d matrix,
Γ ∗ ( ϕ ∗ ; a ) : = ( G ∗ ( χ 0 ; a ) G ( χ 1 ; a ) ⋯ G ∗ ( χ d − 1 ; a ) G ∗ [ r ] ( χ 0 ; a ) G [ r ] ( χ 1 ; a ) ⋯ G ∗ [ r ] ( χ d − 1 ; a ) G ∗ [ 2 r ] ( χ 0 ; a ) G [ 2 r ] ( χ 1 ; a ) ⋯ G ∗ [ 2 r ] ( χ d − 1 ; a ) ⋮ ⋮ ⋮ ) . (10) \Gamma^{*}(\phi^{*};a):=
\begin{pmatrix}
\mathfrak{G}^{*}(\chi_0;a)&\mathfrak{G}(\chi_1;a)&\cdots&\mathfrak{G}^{*}(\chi_{d-1};a)\\
\mathfrak{G}^{*[r]}(\chi_0;a)&\mathfrak{G}^{[r]}(\chi_1;a)&\cdots&\mathfrak{G}^{*[r]}(\chi_{d-1};a)\\
\mathfrak{G}^{*[2r]}(\chi_0;a)&\mathfrak{G}^{[2r]}(\chi_1;a)&\cdots&\mathfrak{G}^{*[2r]}(\chi_{d-1};a)\\
\vdots&\vdots&\vdots&
\end{pmatrix}. \tag*{(10)} Γ ∗ ( ϕ ∗ ; a ) := G ∗ ( χ 0 ; a ) G ∗ [ r ] ( χ 0 ; a ) G ∗ [ 2 r ] ( χ 0 ; a ) ⋮ G ( χ 1 ; a ) G [ r ] ( χ 1 ; a ) G [ 2 r ] ( χ 1 ; a ) ⋮ ⋯ ⋯ ⋯ ⋮ G ∗ ( χ d − 1 ; a ) G ∗ [ r ] ( χ d − 1 ; a ) G ∗ [ 2 r ] ( χ d − 1 ; a ) . (10)
Here G [ k ] ( χ ; a ) \mathfrak{G}^{[k]}(\chi;a) G [ k ] ( χ ; a ) and G ∗ [ k ] ( χ ; a ) \mathfrak{G}^{*[k]}(\chi;a) G ∗ [ k ] ( χ ; a ) are obtained by twisting each entry of G ( χ ; a ) \mathfrak{G}(\chi;a) G ( χ ; a ) and G ∗ ( χ ; a ) \mathfrak{G}^{*}(\chi;a) G ∗ ( χ ; a ) k k k times respectively as above. For ℓ ∈ N \ell\in\mathbb{N} ℓ ∈ N , we let Γ ( ϕ ; a ) ∈ Mat r d × ℓ r ( K ‾ ) \Gamma(\phi;a)\in\operatorname{Mat}_{rd\times\ell r}(\overline{K}) Γ ( ϕ ; a ) ∈ Mat r d × ℓ r ( K ) be the truncation of Γ ( ϕ ; a ) \Gamma(\phi;a) Γ ( ϕ ; a ) after r ℓ r\ell r ℓ column blocks, and let Γ ∗ ( ϕ ∗ ; a ) ∈ Mat ℓ r × r d ( K ‾ ) \Gamma^{*}(\phi^{*};a)\in\operatorname{Mat}_{\ell r\times rd}(\overline{K}) Γ ∗ ( ϕ ∗ ; a ) ∈ Mat ℓ r × r d ( K ) be the truncation of Γ ∗ ( ϕ ∗ ; a ) \Gamma^{*}(\phi^{*};a) Γ ∗ ( ϕ ∗ ; a ) after r ℓ r\ell r ℓ row blocks. We also set Γ ( ϕ ) : = Γ ( ϕ ; 1 ) \Gamma(\phi):=\Gamma(\phi;1) Γ ( ϕ ) := Γ ( ϕ ; 1 ) , Γ ∗ ( ϕ ∗ ) : = Γ ∗ ( ϕ ∗ ; 1 ) \Gamma^{*}(\phi^{*}):=\Gamma^{*}(\phi^{*};1) Γ ∗ ( ϕ ∗ ) := Γ ∗ ( ϕ ∗ ; 1 ) , etc.
Now for β ∈ k \beta\in k β ∈ k that is regular at ε \varepsilon ε , define the block diagonal matrix,
Q ( β ) : = diag ( β ( ε ) I r , … , β ( ε q d − 1 ) I r ) ∈ Mat r d × r d ( F q d ) , (11) \mathbf{Q}(\beta):=\operatorname{diag}\bigl(\beta(\varepsilon)I_r,\ldots,\beta(\varepsilon^{q^{d-1}})I_r\bigr)\in\operatorname{Mat}_{rd\times rd}(\mathbb{F}_{q^d}), \tag*{(11)} Q ( β ) := diag ( β ( ε ) I r , … , β ( ε q d − 1 ) I r ) ∈ Mat r d × r d ( F q d ) , (11)
where I r I_r I r is the r × r r \times r r × r identity matrix. For β ∈ k \beta\in k β ∈ k define
\mathbf{P}(\beta) \coloneqq
\begin{pmatrix}
\beta& -\left\{\frac{\beta}{1}\right\}_{\phi^*}^{q} & -\left\{\frac{\beta}{2}\right\}_{\phi^*}^{q^2} & -\left\{\frac{\beta}{3}\right\}_{\phi^*}^{q^3} & \cdots\\
\left\{\frac{\beta}{1}\right\}_{\phi} & \beta^q & -\left\{\frac{\beta}{1}\right\}_{\phi^*}^{q^2} & -\left\{\frac{\beta}{2}\right\}_{\phi^*}^{q^3} & \iddots\\
\left\{\frac{\beta}{2}\right\}_{\phi} & \left\{\frac{\beta}{1}\right\}_{\phi}^{q} & \beta^{q^2} & -\left\{\frac{\beta}{1}\right\}_{\phi^*}^{q^3} & \iddots\\
\left\{\frac{\beta}{3}\right\}_{\phi} & \left\{\frac{\beta}{2}\right\}_{\phi}^{q} & \left\{\frac{\beta}{1}\right\}_{\phi}^{q^2} & \beta^{q^3} & \iddots\\
\vdots& \ddots& \ddots& \ddots& \ddots
\end{pmatrix}
\in\operatorname{Mat}_{N \times N}(\overline{K}). \tag*{(12)}
Similarly, let P ℓ ( β ) ∈ Mat r ℓ × r ℓ ( K ‾ ) \mathbf{P}_{\ell}(\beta) \in\operatorname{Mat}_{r\ell\times r\ell}(\overline{K}) P ℓ ( β ) ∈ Mat r ℓ × r ℓ ( K ) be the truncation of P ( β ) \mathbf{P}(\beta) P ( β ) after r ℓ r\ell r ℓ rows and columns. For a ∈ A a \in A a ∈ A , set ⟨ a / ℘ ⟩ = ( a m o d ℘ ) / ℘ ∈ k \langle a/\wp\rangle= (a \bmod\wp)/\wp\in k ⟨ a / ℘ ⟩ = ( a mod ℘ ) / ℘ ∈ k . We then obtain the following matrix duality between Gauss-Thakur sums and their duals (Theorem 7.7).
Theorem A. With the matrices defined as above, we have the following identities.
(a) For a , b ∈ A a, b \in A a , b ∈ A ,
Γ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ; b ) = P ( ⟨ a b / ℘ ⟩ ) . \Gamma^*(\phi^*;a)\cdot Q(\wp')\cdot\Gamma(\phi;b)=\mathbf{P}(\langle ab/\wp\rangle). Γ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ; b ) = P (⟨ ab / ℘ ⟩) . (b) For a , b ∈ A a, b \in A a , b ∈ A and ℓ ∈ N \ell\in\mathbb{N} ℓ ∈ N ,
Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ ; b ) = P ℓ ( ⟨ a b / ℘ ⟩ ) . \Gamma_\ell^*(\phi^*;a)\cdot Q(\wp')\cdot\Gamma_\ell(\phi;b)=\mathbf{P}_\ell(\langle ab/\wp\rangle). Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ ; b ) = P ℓ (⟨ ab / ℘ ⟩) . (c) In particular,
Γ ∗ ( ϕ ∗ ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ) = P ( 1 / ℘ ) . \Gamma^*(\phi^*)\cdot Q(\wp')\cdot\Gamma(\phi)=\mathbf{P}(1/\wp). Γ ∗ ( ϕ ∗ ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ) = P ( 1/ ℘ ) .
1.17. Carlitz module. In §8, we investigate the case of the Carlitz module C, which is the rank 1 Drinfeld module defined by C θ = θ + τ C_\theta=\theta+\tau C θ = θ + τ . The Gauss-Thakur sums g C ( χ ) g_C(\chi) g C ( χ ) for C are then precisely the ones designed by Thakur [42 ] . Thakur proved a number of results about g C ( χ ) g_C(\chi) g C ( χ ) , including identities and ℘ \wp ℘ -divisibility properties [42 Thms. I–IV] , which we summarize in Theorem 8.9. One consequence of Theorem A for C is that
g C ( χ ε ) g C ∗ ( χ ε ) = 1 ( θ − ε ) ℘ ′ ( ε ) 2 . (13) g_C(\chi_\varepsilon)g_C^*(\chi_\varepsilon)=\frac{1}{(\theta-\varepsilon)\wp'(\varepsilon)^2}. \tag*{(13)} g C ( χ ε ) g C ∗ ( χ ε ) = ( θ − ε ) ℘ ′ ( ε ) 2 1 . (13)
and using this identity we can transform Thakur’s results into corresponding results for Carlitz dual Gauss-Thakur sums (Theorem 8.10). Thakur also proved result on the ∞ \infty ∞ -adic sign of g C ( χ ) g_C(\chi) g C ( χ ) in [45 Thm. 3.2] , and we prove a corresponding result for g C ∗ ( χ ) g_C^*(\chi) g C ∗ ( χ ) (Proposition 8.17). We prove the following application of Theorem A (Theorem 8.24).
Theorem B. Let ℘ ∈ A + \wp\in A_+ ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of ℘ \wp ℘ . Let K = k ( e C ( 1 / ℘ ) ) K=k(e_C(1/\wp)) K = k ( e C ( 1/ ℘ )) and H = F q d K H=\mathbb{F}_{q^d}K H = F q d K . Define Γ d ( C ) , Γ d ∗ ( C ∗ ) ∈ GL d ( H ) \Gamma_d(C), \Gamma_d^*(C^*) \in\operatorname{GL}_d(H) Γ d ( C ) , Γ d ∗ ( C ∗ ) ∈ GL d ( H ) as above.
(a) We have in H,
det Γ d ( C ) q − 1 = − ℘ and det Γ d ∗ ( C ∗ ) q − 1 = − 1 ℘ q d . \det\Gamma_d(C)^{q-1}=-\wp\quad\text{and}\quad\det\Gamma_d^*(C^*)^{q-1}=-\frac{1}{\wp^{q^d}}. det Γ d ( C ) q − 1 = − ℘ and det Γ d ∗ ( C ∗ ) q − 1 = − ℘ q d 1 . (b) In the completion H p 0 H_{\mathfrak p_0} H p 0 , where p 0 \mathfrak p_0 p 0 is the unique prime of H above θ − ε ∈ F q d [ θ ] \theta-\varepsilon\in\mathbb{F}_{q^d}[\theta] θ − ε ∈ F q d [ θ ] , by choosing a root ( − ℘ ) 1 / ( q d − 1 ) (-\wp)^{1/(q^d-1)} ( − ℘ ) 1/ ( q d − 1 ) as in Theorem 8.9(c),
det Γ d ( C ) = ( − ℘ ) 1 / ( q − 1 ) and det Γ d ∗ ( C ∗ ) = ( − 1 ) d N ( ℘ ′ ( ε ) ) ⋅ ( − ℘ ) q d / ( q − 1 ) . \det\Gamma_d(C)=(-\wp)^{1/(q-1)}
\quad\text{and}\quad
\det\Gamma_d^*(C^*)=\frac{(-1)^d}{N(\wp'(\varepsilon))\cdot(-\wp)^{q^d/(q-1)}}. det Γ d ( C ) = ( − ℘ ) 1/ ( q − 1 ) and det Γ d ∗ ( C ∗ ) = N ( ℘ ′ ( ε )) ⋅ ( − ℘ ) q d / ( q − 1 ) ( − 1 ) d .
where N : F q d × → F q × \mathrm{N}:\mathbb{F}_{q^d}^{\times}\to\mathbb{F}_q^{\times} N : F q d × → F q × is the norm.
We prove ∞ \infty ∞ -adic identities for det Γ d ( C ) \det\Gamma_d(\mathbb{C}) det Γ d ( C ) and det Γ d ∗ ( C ∗ ) \det\Gamma_d^*(\mathbb{C}^*) det Γ d ∗ ( C ∗ ) in Theorem 8.24 (c), which we have not repeated here for brevity, but can be obtained from Theorem C below as well. As another application of Theorem A for the Carlitz module, we obtain identities for Goss-Dirichlet L L L -functions,
L ( χ k , s ) = ∑ a ∈ A + χ k ( a ) a L(\chi_k,s)=\sum_{a\in A_+}\frac{\chi_k(a)}{a} L ( χ k , s ) = a ∈ A + ∑ a χ k ( a )
by combining Theorem A with Pellarin’s special L L L -value formula [36 Thm. 1] and an identity of Gazda and Maurischat [19 Thm. 6.1] . We obtain a summation identity (Proposition 8.30 ) between L L L -values, twists of Gauss-Thakur sums, the Carlitz period π ~ \widetilde{\pi} π , and Carlitz multiplication coefficients,
∑ k = 0 d − 1 L ( χ k , q i ) q [ j ] ( χ k ; b ) = { π ~ q i { ⟨ b / ℘ ⟩ i − j } C q j if i − j ≥ 0 , − π ~ q i { ⟨ b / ℘ ⟩ j − i } C ∗ q j if i − j < 0. (14) \sum_{k=0}^{d-1}L(\chi_k,q^i)q^{[j]}(\chi_k;b)=
\begin{cases}
\widetilde{\pi}^{q^i}\left\{\frac{\langle b/\wp\rangle}{i-j}\right\}_{\mathbb{C}}^{q^j} & \text{if }i-j\geq0,\\
-\widetilde{\pi}^{q^i}\left\{\frac{\langle b/\wp\rangle}{j-i}\right\}_{\mathbb{C}^*}^{q^j} & \text{if }i-j<0.
\end{cases} \tag*{(14)} k = 0 ∑ d − 1 L ( χ k , q i ) q [ j ] ( χ k ; b ) = ⎩ ⎨ ⎧ π q i { i − j ⟨ b / ℘ ⟩ } C q j − π q i { j − i ⟨ b / ℘ ⟩ } C ∗ q j if i − j ≥ 0 , if i − j < 0. (14)
We explore also connections with Anderson’s log-algebraicity [1 ] in Corollary 8.32.
Determinants in higher rank
In §9, we investigate the determinants of Γ d ( ϕ ) \Gamma_d(\phi) Γ d ( ϕ ) and Γ d ∗ ( ϕ ∗ ) \Gamma_d^*(\phi^*) Γ d ∗ ( ϕ ∗ ) for general ϕ \phi ϕ . As an intermediate result we prove (Proposition 9.7 ) that
∏ j = 0 d − 1 det B ϕ ( χ j ) q − 1 = ( − 1 ) r d ℘ b r q r and ∏ j = 0 d − 1 det B ϕ ∗ ∗ ( χ j ) q − 1 = ( − 1 ) r d b r d ℘ q r , \prod_{j=0}^{d-1}\det\mathcal{B}_{\phi}(\chi_j)^{q-1}=\frac{(-1)^{rd}\wp}{b_r^{q^r}}
\quad\text{and}\quad
\prod_{j=0}^{d-1}\det\mathcal{B}_{\phi^*}^*(\chi_j)^{q-1}=\frac{(-1)^{rd}b_r^d}{\wp^{q^r}}, j = 0 ∏ d − 1 det B ϕ ( χ j ) q − 1 = b r q r ( − 1 ) r d ℘ and j = 0 ∏ d − 1 det B ϕ ∗ ∗ ( χ j ) q − 1 = ℘ q r ( − 1 ) r d b r d ,
which serve as higher rank versions of a result of Thakur [42 Thm. II] (see Theorem 8.9 (b)). The full result is the following (Theorem 9.12 ).
Theorem C. Let ℘ ∈ A + \wp\in A_+ ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of ℘ \wp ℘ . Define Γ d ( ϕ ) , Γ d ∗ ( ϕ ∗ ) ∈ G L d ( K ‾ ) \Gamma_d(\phi),\Gamma_d^*(\phi^*)\in\mathrm{GL}_d(\overline{K}) Γ d ( ϕ ) , Γ d ∗ ( ϕ ∗ ) ∈ GL d ( K ) as in (1.13)–(1.14).
(a) We have
det Γ d ( ϕ ) q − 1 = ( − 1 ) r ℘ b r ( q r d − 1 ) / ( q r − 1 ) and det Γ d ∗ ( ϕ ∗ ) q − 1 = ( − 1 ) r b r ( q r d − 1 ) / ( q r − 1 ) ℘ q r d . \det\Gamma_d(\phi)^{q-1}=\frac{(-1)^r\wp}{b_r^{(q^{rd}-1)/(q^r-1)}}\quad\text{and}\quad\det\Gamma_d^*(\phi^*)^{q-1}=\frac{(-1)^r b_r^{(q^{rd}-1)/(q^r-1)}}{\wp^{q^{rd}}}. det Γ d ( ϕ ) q − 1 = b r ( q r d − 1 ) / ( q r − 1 ) ( − 1 ) r ℘ and det Γ d ∗ ( ϕ ∗ ) q − 1 = ℘ q r d ( − 1 ) r b r ( q r d − 1 ) / ( q r − 1 ) . (b) By choosing ( − 1 ) 1 / ( q − 1 ) (-1)^{1/(q-1)} ( − 1 ) 1/ ( q − 1 ) to have the same sign as π ~ \widetilde{\pi} π and ℘ 1 / ( q − 1 ) \wp^{1/(q-1)} ℘ 1/ ( q − 1 ) to have sign 1 1 1 ,
det Γ d ( ϕ ) = ( − 1 ) r d ( d − 1 ) / 2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) V ( ε ) r ⋅ ℘ 1 / ( q − 1 ) \det\Gamma_d(\phi)=\frac{(-1)^{rd(d-1)/2}\cdot(-1)^{d/(q-1)}\cdot\mu^{(q^{rd}-1)/(q^r-1)}}{V(\varepsilon)^r}\cdot\wp^{1/(q-1)} det Γ d ( ϕ ) = V ( ε ) r ( − 1 ) r d ( d − 1 ) /2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ ℘ 1/ ( q − 1 ) and
det Γ d ∗ ( ϕ ∗ ) = 1 ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ V ( ε ) r ⋅ 1 ℘ q r d / ( q − 1 ) , \det\Gamma_d^*(\phi^*)=\frac{1}{(-1)^{d/(q-1)}\cdot\mu^{(q^{rd}-1)/(q^r-1)}\cdot V(\varepsilon)^r}\cdot\frac{1}{\wp^{q^{rd}/(q-1)}}, det Γ d ∗ ( ϕ ∗ ) = ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ V ( ε ) r 1 ⋅ ℘ q r d / ( q − 1 ) 1 , where μ ∈ K ‾ \mu\in\overline{K} μ ∈ K satisfies det Υ = μ ω C \det\Upsilon=\mu\omega_C det Υ = μ ω C and V ( ε ) V(\varepsilon) V ( ε ) is the Vandermonde determinant V ( ε , … , ε q d − 1 ) V(\varepsilon,\ldots,\varepsilon^{q^{d-1}}) V ( ε , … , ε q d − 1 ) .
Drinfeld modular forms
We apply Theorem C to Drinfeld modular forms of higher rank with values in Tate algebras, building on identities of Gekeler [22 ] , Chen and Gezmiş [12 , 13 ] , Pellarin [36 ] , and Perkins [39 ] . In particular, we show (Corollary 9.23) that Gekeler’s modular forms h r ( z ) h_r(\mathbf{z}) h r ( z ) and Δ r ( z ) \Delta_r(\mathbf{z}) Δ r ( z ) , for z \mathbf{z} z in the Drinfeld upper half space H r \mathcal{H}_r H r , are expressible as determinants of dual Gauss-Thakur sums. For example, we find
Δ r ( z ) ( q r d − 1 ) / ( q r − 1 ) = ( − 1 ) r q q r d ⋅ det Γ d ∗ ( z ) q − 1 , \Delta_r(\mathbf{z})^{(q^{rd}-1)/(q^r-1)}=(-1)^r q^{q^{rd}}\cdot\det\Gamma_d^*(\mathbf{z})^{q^{-1}}, Δ r ( z ) ( q r d − 1 ) / ( q r − 1 ) = ( − 1 ) r q q r d ⋅ det Γ d ∗ ( z ) q − 1 ,
where Γ d ∗ ( z ) = Γ d ∗ ( ϕ z ) \Gamma_d^*(\mathbf{z})=\Gamma_d^*(\phi^z) Γ d ∗ ( z ) = Γ d ∗ ( ϕ z ) is associated to the Drinfeld module ϕ z \phi^z ϕ z . See § 9.3 for details.
Acknowledgments. The authors thank G. H. Ferraro for many helpful discussions on the contents of this paper.
Drinfeld modules and their adjoints
Rigid analytic functions
Let t t t be a variable independent of θ \theta θ . For α ∈ C ∞ \alpha\in\mathbb{C}_\infty α ∈ C ∞ , α ≠ 0 \alpha\ne0 α = 0 , the Tate algebra
T α : = { ∑ m ≥ 0 c m t m ∈ C ∞ [ [ t ] ] : lim m → ∞ ∣ α ∣ ∞ m ∣ c m ∣ ∞ = 0 } \mathbb{T}_\alpha:=\left\{\sum_{m\geq0}c_mt^m\in\mathbb{C}_\infty[[t]]:\lim_{m\to\infty}|\alpha|_\infty^m|c_m|_\infty=0\right\} T α := { m ≥ 0 ∑ c m t m ∈ C ∞ [[ t ]] : m → ∞ lim ∣ α ∣ ∞ m ∣ c m ∣ ∞ = 0 }
consists of power series that converge on the closed disk of radius ∣ α ∣ ∞ |\alpha|_\infty ∣ α ∣ ∞ . The Gauss norm on T α \mathbb{T}_\alpha T α , defined as ∥ ∑ c m t m ∥ = max m { ∣ α ∣ ∞ m ∣ c m ∣ ∞ } \left\|\sum c_mt^m\right\|=\max_m\{|\alpha|_\infty^m|c_m|_\infty\} ∥ ∑ c m t m ∥ = max m { ∣ α ∣ ∞ m ∣ c m ∣ ∞ } , makes T α \mathbb{T}_\alpha T α into a complete C ∞ [ t ] \mathbb{C}_\infty[t] C ∞ [ t ] -algebra. We let T : = T 1 \mathbb{T}:=\mathbb{T}_1 T := T 1 . We further define rings of double-tailed Laurent series,
S α : = { ∑ m ∈ Z c m t m ∈ C ∞ [ [ t , t − 1 ] ] : lim m → ± ∞ ∣ α ∣ ∞ m ∣ c m ∣ ∞ = 0 } , \mathbb{S}_\alpha:=\left\{\sum_{m\in\mathbb{Z}}c_mt^m\in\mathbb{C}_\infty[[t,t^{-1}]]:\lim_{m\to\pm\infty}|\alpha|_\infty^m|c_m|_\infty=0\right\}, S α := { m ∈ Z ∑ c m t m ∈ C ∞ [[ t , t − 1 ]] : m → ± ∞ lim ∣ α ∣ ∞ m ∣ c m ∣ ∞ = 0 } ,
which provide functions that converge on the annular strip { x ∈ C ∞ : ∣ x ∣ ∞ = ∣ α ∣ ∞ } \{x\in\mathbb{C}_\infty:|x|_\infty=|\alpha|_\infty\} { x ∈ C ∞ : ∣ x ∣ ∞ = ∣ α ∣ ∞ } (an affinoid ring domain in the sense of [18 §2.2] ).
Twisted polynomials
We let τ : C ∞ → C ∞ \tau:\mathbb{C}_\infty\to\mathbb{C}_\infty τ : C ∞ → C ∞ denote the q q q -th power Frobenius endomorphism, which we extend to matrices τ : Mat m × n ( C ∞ ) → Mat m × n ( C ∞ ) \tau:\operatorname{Mat}_{m\times n}(\mathbb{C}_\infty)\to\operatorname{Mat}_{m\times n}(\mathbb{C}_\infty) τ : Mat m × n ( C ∞ ) → Mat m × n ( C ∞ ) entry-wise. For an F q \mathbb{F}_q F q -algebra R ⊆ Mat n × n ( C ∞ ) R\subseteq\operatorname{Mat}_{n\times n}(\mathbb{C}_\infty) R ⊆ Mat n × n ( C ∞ ) , we let R [ τ ] R[\tau] R [ τ ] be the ring of twisted polynomials in τ \tau τ over R R R , i.e., the F q \mathbb{F}_q F q -subalgebra of End F q ( C ∞ ) \operatorname{End}_{\mathbb{F}_q}(\mathbb{C}_\infty) End F q ( C ∞ ) generated by R R R and τ \tau τ . Likewise, if R ⊆ Mat n × n ( C ∞ ) R\subseteq\operatorname{Mat}_{n\times n}(\mathbb{C}_\infty) R ⊆ Mat n × n ( C ∞ ) is a perfect F q \mathbb{F}_q F q -subalgebra, we similarly define twisted polynomials R [ τ − 1 ] R[\tau^{-1}] R [ τ − 1 ] in End F q ( C ∞ ) \operatorname{End}_{\mathbb{F}_q}(\mathbb{C}_\infty) End F q ( C ∞ ) . We construct the ring of Laurent polynomials R [ τ , τ − 1 ] R[\tau,\tau^{-1}] R [ τ , τ − 1 ] as well.
The *-operator
If R R R is a perfect field, then the F q \mathbb{F}_q F q -linear involution ∗ : R [ τ , τ − 1 ] → R [ τ , τ − 1 ] *:R[\tau,\tau^{-1}]\to R[\tau,\tau^{-1}] ∗ : R [ τ , τ − 1 ] → R [ τ , τ − 1 ] defined by
( ∑ b i τ i ) ∗ : = ∑ b i 1 / q i τ − i = ∑ τ − i b i , (15) \left(\sum b_i\tau^i\right)^*:=\sum b_i^{1/q^i}\tau^{-i}=\sum\tau^{-i}b_i, \tag*{(15)} ( ∑ b i τ i ) ∗ := ∑ b i 1/ q i τ − i = ∑ τ − i b i , (15)
satisfies the anti-commutative property, ( α β ) ∗ = β ∗ α ∗ (\alpha\beta)^*=\beta^*\alpha^* ( α β ) ∗ = β ∗ α ∗ , for α , β ∈ R [ τ , τ − 1 ] \alpha,\beta\in R[\tau,\tau^{-1}] α , β ∈ R [ τ , τ − 1 ] . As such ∗ * ∗ induces anti-isomorphisms, ∗ : R [ τ ] ⟶ ∼ R [ τ − 1 ] *:R[\tau]\overset{\sim}{\longrightarrow}R[\tau^{-1}] ∗ : R [ τ ] ⟶ ∼ R [ τ − 1 ] and ∗ : R [ τ − 1 ] ⟶ ∼ R [ τ ] *:R[\tau^{-1}]\overset{\sim}{\longrightarrow}R[\tau] ∗ : R [ τ − 1 ] ⟶ ∼ R [ τ ] . We also extend ∗ * ∗ to Laurent series rings ∗ : R ( ( τ ) ) → R ( ( τ − 1 ) ) *:R((\tau))\to R((\tau^{-1})) ∗ : R (( τ )) → R (( τ − 1 )) and ∗ : R ( ( τ − 1 ) ) → R ( ( τ ) ) *:R((\tau^{-1}))\to R((\tau)) ∗ : R (( τ − 1 )) → R (( τ )) in the natural way. For more information, see for example [26 §4.14, 40 , 47 §2.10] .
Drinfeld modules
For an intermediate field k ⊆ K ⊆ C ∞ k\subseteq K\subseteq\mathbb{C}_\infty k ⊆ K ⊆ C ∞ , a Drinfeld module over K K K is defined by an F q \mathbb{F}_q F q -algebra homomorphism ϕ : A → K [ τ ] \phi:A\to K[\tau] ϕ : A → K [ τ ] such that, as in (1) , for r ≥ 1 r\geq1 r ≥ 1 , we have ϕ θ = θ + b 1 τ + ⋯ + b r τ r \phi_\theta=\theta+b_1\tau+\cdots+b_r\tau^r ϕ θ = θ + b 1 τ + ⋯ + b r τ r , b r ≠ 0 b_r\ne0 b r = 0 . In this case r r r is the rank of ϕ \phi ϕ . See [26 , 34 , 47 ] for fundamental properties of Drinfeld modules. For a ∈ A a\in A a ∈ A , we write
ϕ a = ∑ n = 0 r deg a { a n } ϕ τ n , { a n } ϕ ∈ K , \phi_a=\sum_{n=0}^{r\deg a}\left\{\begin{matrix}a\\n\end{matrix}\right\}_{\phi}\tau^n,\qquad\left\{\begin{matrix}a\\n\end{matrix}\right\}_{\phi}\in K, ϕ a = n = 0 ∑ r d e g a { a n } ϕ τ n , { a n } ϕ ∈ K ,
Then ϕ \phi ϕ induces an A A A -module structure on C ∞ \mathbb{C}_{\infty} C ∞ such that for a ∈ A a \in A a ∈ A , x ∈ C ∞ x \in\mathbb{C}_{\infty} x ∈ C ∞ , we have
ϕ a ( x ) = ∑ n = 0 r deg a ( a n ) ϕ x q n . \phi_a(x)=\sum_{n=0}^{r\deg a}\binom{a}{n}_{\phi}x^{q^n}. ϕ a ( x ) = n = 0 ∑ r d e g a ( n a ) ϕ x q n .
The exponential exp ϕ ∈ 1 + K [ [ τ ] ] τ \exp_{\phi}\in1+K[[\tau]]\tau exp ϕ ∈ 1 + K [[ τ ]] τ and logarithm log ϕ ∈ 1 + K [ [ τ ] ] τ \log_{\phi}\in1+K[[\tau]]\tau log ϕ ∈ 1 + K [[ τ ]] τ are the unique series such that
exp ϕ ⋅ log ϕ = log ϕ ⋅ exp ϕ = 1 , exp ϕ ⋅ a = ϕ a ⋅ exp ϕ , log ϕ ⋅ ϕ a = a ⋅ log ϕ , ∀ a ∈ A . (16) \begin{aligned}
\exp_{\phi}\cdot\log_{\phi}&=\log_{\phi}\cdot\exp_{\phi}=1,\\
\exp_{\phi}\cdot a&=\phi_a\cdot\exp_{\phi},\qquad\log_{\phi}\cdot\phi_a=a\cdot\log_{\phi},\qquad\forall a\in A. \tag*{(16)}
\end{aligned} exp ϕ ⋅ log ϕ exp ϕ ⋅ a = log ϕ ⋅ exp ϕ = 1 , = ϕ a ⋅ exp ϕ , log ϕ ⋅ ϕ a = a ⋅ log ϕ , ∀ a ∈ A . (16)
Both exp ϕ \exp_{\phi} exp ϕ and log ϕ \log_{\phi} log ϕ induce functions on z ∈ C ∞ z\in\mathbb{C}_{\infty} z ∈ C ∞ , where exp ϕ ( z ) \exp_{\phi}(z) exp ϕ ( z ) is an entire function and log ϕ ( z ) \log_{\phi}(z) log ϕ ( z ) has a finite radius of convergence (e.g., see [26 Prop. 4.14.2, 29 Cor. 4.5, 34 Lem. 5.1.5] ). If exp ϕ = ∑ i ≥ 0 α i τ i \exp_{\phi}=\sum_{i\geq0}\alpha_i\tau^i exp ϕ = ∑ i ≥ 0 α i τ i and log ϕ = ∑ i ≥ 0 β i τ i \log_{\phi}=\sum_{i\geq0}\beta_i\tau^i log ϕ = ∑ i ≥ 0 β i τ i , then for a variable x x x ,
exp ϕ ⋅ x ⋅ log ϕ = ∑ k ≥ 0 ( x k ) ϕ τ k ⇒ ( x n ) ϕ = ∑ i = 0 n α i β n − i q i x q i ∈ K [ x ] . (17) \exp_{\phi}\cdot x\cdot\log_{\phi}=\sum_{k\geq0}\binom{x}{k}_{\phi}\tau^k
\quad\Rightarrow\quad
\binom{x}{n}_{\phi}=\sum_{i=0}^{n}\alpha_i\beta_{n-i}^{q^i}x^{q^i}\in K[x]. \tag*{(17)} exp ϕ ⋅ x ⋅ log ϕ = k ≥ 0 ∑ ( k x ) ϕ τ k ⇒ ( n x ) ϕ = i = 0 ∑ n α i β n − i q i x q i ∈ K [ x ] . (17)
We note that if a ∈ A a\in A a ∈ A , then as one would expect, ( x n ) ϕ ∣ x = a = ( a n ) ϕ \left.\binom{x}{n}_{\phi}\right|_{x=a}=\binom{a}{n}_{\phi} ( n x ) ϕ x = a = ( n a ) ϕ .
Adjoint Drinfeld modules
When K K K is perfect, the adjoint of ϕ \phi ϕ is the F q \mathbb{F}_q F q -algebra homomorphism ϕ ∗ : A → K [ τ − 1 ] \phi^*:A\to K[\tau^{-1}] ϕ ∗ : A → K [ τ − 1 ] defined by
ϕ a ∗ : = ( ϕ a ) ∗ = ∑ n = 0 r deg a ( a n ) ϕ 1 / q n τ − n : = ∑ n = 0 r deg a τ − n ( a − n ) ϕ ∗ . \phi_a^*:=(\phi_a)^*=\sum_{n=0}^{r\deg a}\binom{a}{n}_{\phi}^{1/q^n}\tau^{-n}:=\sum_{n=0}^{r\deg a}\tau^{-n}\binom{a}{-n}_{\phi^*}. ϕ a ∗ := ( ϕ a ) ∗ = n = 0 ∑ r d e g a ( n a ) ϕ 1/ q n τ − n := n = 0 ∑ r d e g a τ − n ( − n a ) ϕ ∗ .
Note that ( a n ) ϕ = ( a − n ) ϕ ∗ \binom{a}{n}_{\phi}=\binom{a}{-n}_{\phi^*} ( n a ) ϕ = ( − n a ) ϕ ∗ for a ∈ A a\in A a ∈ A and n ∈ Z n\in\mathbb{Z} n ∈ Z . Similar to the Drinfeld module case, ϕ ∗ \phi^* ϕ ∗ induces an A A A -module structure on C ∞ \mathbb{C}_{\infty} C ∞ such that for a ∈ A a\in A a ∈ A and x ∈ C ∞ x\in\mathbb{C}_{\infty} x ∈ C ∞ ,
ϕ a ∗ ( x ) = ∑ n = 0 r deg a ( a n ) ϕ 1 / q n x 1 / q n = ∑ n = 0 r deg a ( a − n ) ϕ ∗ 1 / q n x 1 / q n . \phi_a^*(x)=\sum_{n=0}^{r\deg a}\binom{a}{n}_{\phi}^{1/q^n}x^{1/q^n}
=\sum_{n=0}^{r\deg a}\binom{a}{-n}_{\phi^*}^{1/q^n}x^{1/q^n}. ϕ a ∗ ( x ) = n = 0 ∑ r d e g a ( n a ) ϕ 1/ q n x 1/ q n = n = 0 ∑ r d e g a ( − n a ) ϕ ∗ 1/ q n x 1/ q n .
For basic properties of adjoint modules, see [17 , 26 §4.14, 28 §2.2, §3.2, 40 ] . By applying the ∗ * ∗ operator to (16) , we obtain the companion formulas for exp ϕ ∗ , log ϕ ∗ ∈ 1 + K [ [ τ − 1 ] ] τ − 1 \exp_{\phi}^*,\log_{\phi}^*\in1+K[[\tau^{-1}]]\tau^{-1} exp ϕ ∗ , log ϕ ∗ ∈ 1 + K [[ τ − 1 ]] τ − 1 ,
log ϕ ∗ ⋅ exp ϕ ∗ = exp ϕ ∗ ⋅ log ϕ ∗ = 1 , log ϕ ∗ ⋅ a = ϕ a ∗ ⋅ log ϕ ∗ , exp ϕ ∗ ⋅ ϕ a ∗ = a ⋅ exp ϕ ∗ , ∀ a ∈ A . (18) \begin{aligned}
\log_{\phi}^*\cdot\exp_{\phi}^*&=\exp_{\phi}^*\cdot\log_{\phi}^*=1,\\
\log_{\phi}^*\cdot a&=\phi_a^*\cdot\log_{\phi}^*,\qquad\exp_{\phi}^*\cdot\phi_a^*=a\cdot\exp_{\phi}^*,\qquad\forall a\in A. \tag*{(18)}
\end{aligned} log ϕ ∗ ⋅ exp ϕ ∗ log ϕ ∗ ⋅ a = exp ϕ ∗ ⋅ log ϕ ∗ = 1 , = ϕ a ∗ ⋅ log ϕ ∗ , exp ϕ ∗ ⋅ ϕ a ∗ = a ⋅ exp ϕ ∗ , ∀ a ∈ A . (18)
For z ∈ C ∞ z\in\mathbb{C}_{\infty} z ∈ C ∞ , exp ϕ ∗ ( z ) \exp_{\phi}^*(z) exp ϕ ∗ ( z ) converges everywhere, but log ϕ ∗ ( z ) \log_{\phi}^*(z) log ϕ ∗ ( z ) converges only at z = 0 z=0 z = 0 .
Torsion submodules and exponentiation
For b ∈ A b\in A b ∈ A , b ≠ 0 b\ne0 b = 0 , we have
ϕ [ b ] = { x ∈ K ∣ ϕ b ( x ) = 0 } ≅ ( A / b A ) r , ϕ ∗ [ b ] = { x ∈ K ∣ ϕ b ∗ ( x ) = 0 } ≅ ( A / b A ) r . \phi[b]=\{x\in K\mid\phi_b(x)=0\}\cong(A/bA)^r,\qquad
\phi^*[b]=\{x\in K\mid\phi_b^*(x)=0\}\cong(A/bA)^r. ϕ [ b ] = { x ∈ K ∣ ϕ b ( x ) = 0 } ≅ ( A / b A ) r , ϕ ∗ [ b ] = { x ∈ K ∣ ϕ b ∗ ( x ) = 0 } ≅ ( A / b A ) r .
(see [26 §4.14, 40 §8] ). The kernel Λ ϕ \Lambda_{\phi} Λ ϕ of exp ϕ \exp_{\phi} exp ϕ is a discrete free A A A -submodule of C ∞ \mathbb{C}_{\infty} C ∞ of rank r r r , and we have an induced isomorphism of A A A -modules, exp ϕ : C ∞ / Λ ϕ → ∼ ϕ ( C ∞ ) \exp_{\phi}:\mathbb{C}_{\infty}/\Lambda_{\phi}\xrightarrow{\sim}\phi(\mathbb{C}_{\infty}) exp ϕ : C ∞ / Λ ϕ ∼ ϕ ( C ∞ ) , where ϕ ( C ∞ ) \phi(\mathbb{C}_{\infty}) ϕ ( C ∞ ) denotes C ∞ \mathbb{C}_{\infty} C ∞ with A A A -module structure via ϕ \phi ϕ . This implies that
ϕ [ b ] = exp ϕ ( b − 1 Λ ϕ / Λ ϕ ) . (19) \phi[b]=\exp_{\phi}\left(b^{-1}\Lambda_{\phi}/\Lambda_{\phi}\right). \tag*{(19)} ϕ [ b ] = exp ϕ ( b − 1 Λ ϕ / Λ ϕ ) . (19)
The corresponding map exp ϕ ∗ : C ∞ → ϕ ∗ ( C ∞ ) \exp_{\phi}^*:\mathbb{C}_{\infty}\to\phi^*(\mathbb{C}_{\infty}) exp ϕ ∗ : C ∞ → ϕ ∗ ( C ∞ ) is a bit more subtle but has been studied extensively by Ferraro [17 ] and Poonen [40 ] . We see from (18) that ϕ t o r s ∗ ⊆ ker exp ϕ ∗ \phi^*_{\mathrm{tors}}\subseteq\ker\exp_{\phi}^* ϕ tors ∗ ⊆ ker exp ϕ ∗ , where ϕ t o r s ∗ \phi^*_{\mathrm{tors}} ϕ tors ∗ is the A A A -torsion submodule of ϕ ∗ ( C ∞ ) \phi^*(\mathbb{C}_{\infty}) ϕ ∗ ( C ∞ ) , but it is not an equality. Indeed ker exp ϕ ∗ \ker\exp_{\phi}^* ker exp ϕ ∗ is compact and contains ϕ t o r s ∗ \phi^*_{\mathrm{tors}} ϕ tors ∗ as a dense subset [40 Prop. 20] .
The analogous identity to (19) for ϕ ∗ [ b ] \phi^*[b] ϕ ∗ [ b ] would more naturally, given the identities in (18) , arise from log ϕ ∗ \log_{\phi}^* log ϕ ∗ . However, since log ϕ ∗ \log_{\phi}^* log ϕ ∗ converges only at 0 0 0 , this does not produce results. Nevertheless, we develop a stand-in for (19) using techniques from [2 §5] , which we explore in §5 (see Lemma 5.19 ).
Extension ϕ : k ∞ → C ∞ ( ( τ ) ) \phi: k_\infty\to\mathbb{C}_\infty((\tau)) ϕ : k ∞ → C ∞ (( τ ))
Using (17) we extend ϕ \phi ϕ to an injective F q \mathbb{F}_q F q -algebra homomorphism ϕ : k ∞ → C ∞ ( ( τ ) ) \phi:k_\infty\to\mathbb{C}_\infty((\tau)) ϕ : k ∞ → C ∞ (( τ )) , setting
ϕ α : = ∑ n = 0 ∞ { α n } ϕ τ n , ∀ α ∈ k ∞ . \phi_\alpha:= \sum_{n=0}^{\infty} \left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi}\tau^n,\qquad\forall\alpha\in k_\infty. ϕ α := n = 0 ∑ ∞ { α n } ϕ τ n , ∀ α ∈ k ∞ .
In this way we obtain a formal Drinfeld k ∞ k_\infty k ∞ -module [26 §4.6, 34 §6.5, 41 ] . In particular, { α 0 } ϕ = α \left\{\begin{matrix}\alpha\\ 0\end{matrix}\right\}_{\phi}=\alpha { α 0 } ϕ = α , for all α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ . For convenience we set { α n } ϕ = 0 \left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi}=0 { α n } ϕ = 0 if n < 0 n<0 n < 0 . We note that ϕ θ α = ϕ θ ϕ α = ϕ α ϕ θ \phi_{\theta\alpha}=\phi_\theta\phi_\alpha=\phi_\alpha\phi_\theta ϕ θ α = ϕ θ ϕ α = ϕ α ϕ θ implies for α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ and n ∈ Z n\in\mathbb{Z} n ∈ Z (using b 0 = θ b_0=\theta b 0 = θ ),
{ θ α n } ϕ = ∑ j = 0 r b j { α n − j } ϕ q j = ∑ j = 0 r b j q n − j { α n − j } ϕ . (20) \left\{\begin{matrix}\theta\alpha\\ n\end{matrix}\right\}_{\phi}
=\sum_{j=0}^{r}b_j\left\{\begin{matrix}\alpha\\ n-j\end{matrix}\right\}_{\phi}^{q^j}
=\sum_{j=0}^{r}b_j^{q^{n-j}}\left\{\begin{matrix}\alpha\\ n-j\end{matrix}\right\}_{\phi}. \tag*{(20)} { θ α n } ϕ = j = 0 ∑ r b j { α n − j } ϕ q j = j = 0 ∑ r b j q n − j { α n − j } ϕ . (20)
As { x n } ∈ K [ x ] \left\{\begin{matrix}x\\ n\end{matrix}\right\}\in K[x] { x n } ∈ K [ x ] by (17) , each coefficient function of ϕ : k ∞ → C ∞ ( ( τ ) ) \phi:k_\infty\to\mathbb{C}_\infty((\tau)) ϕ : k ∞ → C ∞ (( τ )) is continuous.
Extension ϕ ∗ : k ∞ → C ∞ ( ( τ ) ) \phi^*: k_\infty\to\mathbb{C}_\infty((\tau)) ϕ ∗ : k ∞ → C ∞ (( τ ))
As defined by Ferraro [17 Prop. 7.16] , we extend ϕ ∗ \phi^* ϕ ∗ to ϕ ∗ : k ∞ → C ∞ ( ( τ ) ) \phi^*:k_\infty\to\mathbb{C}_\infty((\tau)) ϕ ∗ : k ∞ → C ∞ (( τ )) . The definition here matches Ferraro’s Φ ^ \widehat{\Phi} Φ , though as we are working with A = F q [ θ ] A=\mathbb{F}_q[\theta] A = F q [ θ ] , we obtain explicit identities as well. For a ∈ A a\in A a ∈ A of degree d d d ,
τ r d ϕ a ∗ = ∑ n = 0 r d { a − n } ϕ ∗ q r d − n τ r d − n = ∑ n = 0 r d { a n − r d } ϕ ∗ q n τ n . (21) \tau^{rd}\phi_a^*=\sum_{n=0}^{rd}\left\{\begin{matrix}a\\ -n\end{matrix}\right\}_{\phi^*}^{q^{rd-n}}\tau^{rd-n}
=\sum_{n=0}^{rd}\left\{\begin{matrix}a\\ n-rd\end{matrix}\right\}_{\phi^*}^{q^n}\tau^n. \tag*{(21)} τ r d ϕ a ∗ = n = 0 ∑ r d { a − n } ϕ ∗ q r d − n τ r d − n = n = 0 ∑ r d { a n − r d } ϕ ∗ q n τ n . (21)
so then τ r d ϕ a ∗ \tau^{rd}\phi_a^* τ r d ϕ a ∗ is invertible in C ∞ [ [ τ ] ] \mathbb{C}_\infty[[\tau]] C ∞ [[ τ ]] , as { a − r d } ϕ ∗ ≠ 0 \left\{\begin{matrix}a\\ -rd\end{matrix}\right\}_{\phi^*}\ne0 { a − r d } ϕ ∗ = 0 . Thus in C ∞ ( ( τ ) ) \mathbb{C}_\infty((\tau)) C ∞ (( τ )) we find
( ϕ a ∗ ) − 1 = { a − r d } ϕ ∗ − 1 τ − r d − { a − r d } ϕ ∗ − q − 1 { a − r d + 1 } ϕ ∗ q τ − r d + 1 + ⋯ . (22) (\phi_a^*)^{-1}
=\left\{\begin{matrix}a\\ -rd\end{matrix}\right\}_{\phi^*}^{-1}\tau^{-rd}
-\left\{\begin{matrix}a\\ -rd\end{matrix}\right\}_{\phi^*}^{-q-1}
\left\{\begin{matrix}a\\ -rd+1\end{matrix}\right\}_{\phi^*}^{q}\tau^{-rd+1}+\cdots. \tag*{(22)} ( ϕ a ∗ ) − 1 = { a − r d } ϕ ∗ − 1 τ − r d − { a − r d } ϕ ∗ − q − 1 { a − r d + 1 } ϕ ∗ q τ − r d + 1 + ⋯ . (22)
Setting
ϕ 1 / a ∗ : = ( ϕ a ∗ ) − 1 ∈ C ∞ ( ( τ ) ) , \phi^*_{1/a}:=(\phi_a^*)^{-1}\in\mathbb{C}_\infty((\tau)), ϕ 1/ a ∗ := ( ϕ a ∗ ) − 1 ∈ C ∞ (( τ )) ,
we obtain a well-defined F q \mathbb{F}_q F q -algebra homomorphism ϕ ∗ : k → C ∞ ( ( τ ) ) \phi^*:k\to\mathbb{C}_\infty((\tau)) ϕ ∗ : k → C ∞ (( τ )) , such that ϕ a / b ∗ = ϕ a ∗ ⋅ ϕ 1 / b ∗ \phi^*_{a/b}=\phi^*_a\cdot\phi^*_{1/b} ϕ a / b ∗ = ϕ a ∗ ⋅ ϕ 1/ b ∗ , for a , b ∈ A a,b\in A a , b ∈ A , b ≠ 0 b\ne0 b = 0 (cf. [26 Prop. 5.5.3] ). Furthermore, since for α ∈ k \alpha\in k α ∈ k we have ϕ α ∗ ∈ C ∞ [ [ τ ] ] τ − r deg α \phi^*_\alpha\in\mathbb{C}_\infty[[\tau]]\tau^{-r\deg\alpha} ϕ α ∗ ∈ C ∞ [[ τ ]] τ − r d e g α , it follows that the coefficient functions of ϕ ∗ \phi^* ϕ ∗ are continuous. We thus define { α n } ϕ ∗ \left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi^*} { α n } ϕ ∗ for α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ and n ∈ Z n\in\mathbb{Z} n ∈ Z by
ϕ α ∗ : = ∑ n = − r deg α ∞ { α n } ϕ ∗ q n τ n = ∑ n = − r deg α ∞ τ n { α n } ϕ ∗ . \phi^*_\alpha:=\sum_{n=-r\deg\alpha}^{\infty}\left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi^*}^{q^n}\tau^n
=\sum_{n=-r\deg\alpha}^{\infty}\tau^n\left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi^*}. ϕ α ∗ := n = − r d e g α ∑ ∞ { α n } ϕ ∗ q n τ n = n = − r d e g α ∑ ∞ τ n { α n } ϕ ∗ .
By convention we set { α n } ϕ ∗ = 0 \left\{\begin{matrix}\alpha\\ n\end{matrix}\right\}_{\phi^*}=0 { α n } ϕ ∗ = 0 for n < − r deg α n<-r\deg\alpha n < − r deg α . Since ϕ θ α ∗ = ϕ α ∗ ϕ θ ∗ = ϕ θ ∗ ϕ α ∗ \phi^*_{\theta\alpha}=\phi^*_\alpha\phi^*_\theta=\phi^*_\theta\phi^*_\alpha ϕ θ α ∗ = ϕ α ∗ ϕ θ ∗ = ϕ θ ∗ ϕ α ∗ for α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ , similar to (20) we have for n ∈ Z n\in\mathbb{Z} n ∈ Z ,
{ θ α n } ϕ ∗ = ∑ j = 0 r b j { α n + j } ϕ ∗ q j = ∑ j = 0 r b j q − n − j { α n + j } ϕ ∗ . (23) \left\{\begin{matrix}\theta\alpha\\ n\end{matrix}\right\}_{\phi^*}
=\sum_{j=0}^{r}b_j\left\{\begin{matrix}\alpha\\ n+j\end{matrix}\right\}_{\phi^*}^{q^j}
=\sum_{j=0}^{r}b_j^{q^{-n-j}}\left\{\begin{matrix}\alpha\\ n+j\end{matrix}\right\}_{\phi^*}. \tag*{(23)} { θ α n } ϕ ∗ = j = 0 ∑ r b j { α n + j } ϕ ∗ q j = j = 0 ∑ r b j q − n − j { α n + j } ϕ ∗ . (23)
We note that if α = ∑ m ≥ m 0 c m θ − m ∈ k ∞ \alpha=\sum_{m\ge m_0}c_m\theta^{-m}\in k_\infty α = ∑ m ≥ m 0 c m θ − m ∈ k ∞ , c m ∈ F q c_m\in\mathbb{F}_q c m ∈ F q , then
ϕ α ∗ = ∑ m = m 0 ∞ c m ϕ θ − m ∗ = ∑ m = m 0 ∞ c m ∑ n = r m ∞ τ n { θ − m n } ϕ ∗ . \phi_\alpha^*=\sum_{m=m_0}^{\infty}c_m\phi_{\theta^{-m}}^*=\sum_{m=m_0}^{\infty}c_m\sum_{n=rm}^{\infty}\tau^n\left\{\frac{\theta^{-m}}{n}\right\}_{\phi^*}. ϕ α ∗ = m = m 0 ∑ ∞ c m ϕ θ − m ∗ = m = m 0 ∑ ∞ c m n = r m ∑ ∞ τ n { n θ − m } ϕ ∗ .
A word of caution is that generally ϕ α ∗ ≠ ( ϕ α ) ∗ \phi_\alpha^*\ne(\phi_\alpha)^* ϕ α ∗ = ( ϕ α ) ∗ for α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ , unless α ∈ A \alpha\in A α ∈ A . If we had extended ϕ ∗ \phi^* ϕ ∗ to a homomorphism k ∞ → C ∞ ( ( τ − 1 ) ) k_\infty\to\mathbb{C}_\infty((\tau^{-1})) k ∞ → C ∞ (( τ − 1 )) , arguing as in § 2.7 , such an identity would hold for all α ∈ k ∞ \alpha\in k_\infty α ∈ k ∞ . However, the definition of ϕ ∗ : k ∞ → C ∞ ( ( τ ) ) \phi^*:k_\infty\to\mathbb{C}_\infty((\tau)) ϕ ∗ : k ∞ → C ∞ (( τ )) here is the right one for our future considerations (e.g., see Proposition 3.16 and Theorem 7.7 ).
Lemma 2.18. Let a , b ∈ A a,b\in A a , b ∈ A , with b ≠ 0 b\ne0 b = 0 . Then { a / b m } ϕ ∗ q m ∈ A [ b 1 , … , b r , b − 1 ] \left\{\frac{a/b}{m}\right\}_{\phi^*}^{q^m}\in A[b_1,\ldots,b_r,b^{-1}] { m a / b } ϕ ∗ q m ∈ A [ b 1 , … , b r , b − 1 ] for all m ≥ 0 m\ge0 m ≥ 0 .
Proof. Let R = A [ b 1 , … , b r , b − 1 ] R=A[b_1,\ldots,b_r,b^{-1}] R = A [ b 1 , … , b r , b − 1 ] . For b ∈ A b\in A b ∈ A , recall that for n ∈ Z n\in\mathbb{Z} n ∈ Z , { b − n } ϕ ∗ = { b n } ϕ ∗ \left\{\frac{b}{-n}\right\}_{\phi^*}=\left\{\frac{b}{n}\right\}_{\phi^*} { − n b } ϕ ∗ = { n b } ϕ ∗ , and so { b − n } ϕ ∗ ∈ A [ b 1 , … , b r ] \left\{\frac{b}{-n}\right\}_{\phi^*}\in A[b_1,\ldots,b_r] { − n b } ϕ ∗ ∈ A [ b 1 , … , b r ] . Furthermore, if deg b = d \deg b=d deg b = d , then { b − r d } ϕ ∗ = b r 1 + q + ⋯ + q deg a − 1 ∈ R × \left\{\frac{b}{-rd}\right\}_{\phi^*}=b_r^{1+q+\cdots+q^{\deg a-1}}\in R^\times { − r d b } ϕ ∗ = b r 1 + q + ⋯ + q d e g a − 1 ∈ R × . Therefore by (2.15), we have τ r d ϕ b ∗ ∈ R [ [ τ ] ] × \tau^{rd}\phi_b^*\in R[[\tau]]^\times τ r d ϕ b ∗ ∈ R [[ τ ] ] × , and so ϕ 1 / b ∗ = ( τ r d ϕ b ∗ ) − 1 τ − r d ∈ R [ [ τ ] ] \phi_{1/b}^*=(\tau^{rd}\phi_b^*)^{-1}\tau^{-rd}\in R[[\tau]] ϕ 1/ b ∗ = ( τ r d ϕ b ∗ ) − 1 τ − r d ∈ R [[ τ ]] . Since ϕ 1 / b ∗ = ∑ m ≥ 0 τ m { 1 / b m } ϕ ∗ \phi_{1/b}^*=\sum_{m\ge0}\tau^m\left\{\frac{1/b}{m}\right\}_{\phi^*} ϕ 1/ b ∗ = ∑ m ≥ 0 τ m { m 1/ b } ϕ ∗ , the result holds for 1 / b 1/b 1/ b . In general,
ϕ a / b ∗ = ∑ k = r ( deg b − deg a ) ∞ τ k { a / b k } ϕ ∗ = ϕ 1 / b ∗ ⋅ ϕ a ∗ = ∑ m = r deg b ∞ ∑ n = 0 r deg a τ m − n { 1 / b m } ϕ ∗ { a − n } ϕ ∗ q n . \phi_{a/b}^*=\sum_{k=r(\deg b-\deg a)}^\infty\tau^k\left\{\frac{a/b}{k}\right\}_{\phi^*}
=\phi_{1/b}^*\cdot\phi_a^*
=\sum_{m=r\deg b}^{\infty}\sum_{n=0}^{r\deg a}\tau^{m-n}\left\{\frac{1/b}{m}\right\}_{\phi^*}\left\{\frac{a}{-n}\right\}_{\phi^*}^{q^n}. ϕ a / b ∗ = k = r ( d e g b − d e g a ) ∑ ∞ τ k { k a / b } ϕ ∗ = ϕ 1/ b ∗ ⋅ ϕ a ∗ = m = r d e g b ∑ ∞ n = 0 ∑ r d e g a τ m − n { m 1/ b } ϕ ∗ { − n a } ϕ ∗ q n .
For m − n ≥ 0 m-n\ge0 m − n ≥ 0 , when we raise a given coefficient in the final sum to the q m − n q^{m-n} q m − n -th power, the result follows from the result for 1 / b 1/b 1/ b and the fact that { a − n } ϕ ∗ ∈ A [ b 1 , … , b r ] \left\{\frac{a}{-n}\right\}_{\phi^*}\in A[b_1,\ldots,b_r] { − n a } ϕ ∗ ∈ A [ b 1 , … , b r ] . □ \square □
Rigid analytic trivializations
For i ∈ Z i\in\mathbb{Z} i ∈ Z , we define Frobenius twisting
f = ∑ c m t m ↦ f ( i ) : = ∑ c m q i t m : C ∞ ( ( t ) ) → C ∞ ( ( t ) ) , f=\sum c_mt^m\mapsto f^{(i)}:=\sum c_m^{q^i}t^m:\mathbb{C}_\infty((t))\to\mathbb{C}_\infty((t)), f = ∑ c m t m ↦ f ( i ) := ∑ c m q i t m : C ∞ (( t )) → C ∞ (( t )) ,
which we extend to matrices in C ∞ ( ( t ) ) \mathbb{C}_\infty((t)) C ∞ (( t )) entry-wise. Frobenius twisting induces automorphisms of the Tate algebra T \mathbb{T} T . For an F q [ t ] \mathbb{F}_q[t] F q [ t ] -subalgebra S S S of C ∞ ( ( t ) ) \mathbb{C}_\infty((t)) C ∞ (( t )) , we define twisted polynomials S [ τ , τ − 1 ] S[\tau,\tau^{-1}] S [ τ , τ − 1 ] subject to the relation, τ f = f ( 1 ) τ \tau f=f^{(1)}\tau τ f = f ( 1 ) τ , f ∈ S f\in S f ∈ S . In this way, for η = ∑ B i τ i ∈ Mat d ( S ) [ τ , τ − 1 ] \eta=\sum B_i\tau^i\in\operatorname{Mat}_d(S)[\tau,\tau^{-1}] η = ∑ B i τ i ∈ Mat d ( S ) [ τ , τ − 1 ] and F ∈ Mat d × n ( C ∞ ( ( t ) ) ) F\in\operatorname{Mat}_{d\times n}(\mathbb{C}_\infty((t))) F ∈ Mat d × n ( C ∞ (( t ))) , we have the operation
η ( F ) : = ∑ i B i F ( i ) ∈ Mat d × n ( C ∞ ( ( t ) ) ) . \eta(F):=\sum_iB_iF^{(i)}\in\operatorname{Mat}_{d\times n}(\mathbb{C}_\infty((t))). η ( F ) := i ∑ B i F ( i ) ∈ Mat d × n ( C ∞ (( t ))) .
3.1. Special functions. Fix a Drinfeld module ϕ \phi ϕ of rank r r r defined over C ∞ \mathbb{C}_\infty C ∞ as in (1.2). A special function and a dual special function for ϕ \phi ϕ are f f f and g ∈ T g\in\mathbb{T} g ∈ T respectively satisfying
ϕ a ( f ) = a ( t ) f , ϕ a ∗ ( g ) = a ( t ) g , ∀ a ∈ A . \phi_a(f)=a(t)f,\qquad\phi_a^*(g)=a(t)g,\qquad\forall a\in A. ϕ a ( f ) = a ( t ) f , ϕ a ∗ ( g ) = a ( t ) g , ∀ a ∈ A .
Special functions for Drinfeld modules and t t t -modules have been studied extensively in various contexts, especially related to Anderson generating functions (e.g., see [10 , 15 , 16 , 17 , 19 , 24 , 29 , 31 , 32 , 35 ] ). Dual special functions arise via Frobenius difference equations with respect to τ − 1 \tau^{-1} τ − 1 (e.g., see [2 , 10 , 32 , 35 ] ), but more recently Ferraro [17 ] has provided new insights and directions for them (e.g., see Theorems 4.3 and 4.7).
3.2. Anderson generating functions. For u ∈ C ∞ u\in\mathbb{C}_\infty u ∈ C ∞ , we let
ω u : = ω u ( t ) : = ∑ m = 0 ∞ exp ϕ ( u θ m + 1 ) t m ∈ T . \omega_u:=\omega_u(t):=\sum_{m=0}^{\infty}\exp_\phi\left(\frac{u}{\theta^{m+1}}\right)t^m\in\mathbb{T}. ω u := ω u ( t ) := m = 0 ∑ ∞ exp ϕ ( θ m + 1 u ) t m ∈ T .
Defined originally by Anderson (see [24 ] ), the function ω u \omega_u ω u satisfies the principal functional equation ϕ t ( ω u ) = t ω u + exp ϕ ( u ) \phi_t(\omega_u)=t\omega_u+\exp_\phi(u) ϕ t ( ω u ) = t ω u + exp ϕ ( u ) . For λ ∈ Λ ϕ \lambda\in\Lambda_\phi λ ∈ Λ ϕ , we then find that
ϕ a ( ω λ ) = a ( t ) ω λ , ∀ a ∈ A . (24) \phi_a(\omega_\lambda)=a(t)\omega_\lambda,\qquad\forall\,a\in A. \tag*{(24)} ϕ a ( ω λ ) = a ( t ) ω λ , ∀ a ∈ A . (24)
Thus ω λ \omega_\lambda ω λ is a special function for ϕ \phi ϕ as defined above, and moreover, all special functions for ϕ \phi ϕ arise in this way [35 §4.2] (see also [15 Rem. 6.3] ). Pellarin [35 §4.2] showed further that ω u \omega_u ω u has a partial fraction decomposition,
ω u = ∑ n = 0 ∞ α n u q n θ q n − t , u ∈ C ∞ , \omega_u=\sum_{n=0}^{\infty}\frac{\alpha_n u^{q^n}}{\theta^{q^n}-t},\qquad u\in\mathbb{C}_\infty, ω u = n = 0 ∑ ∞ θ q n − t α n u q n , u ∈ C ∞ ,
where exp ϕ = ∑ n ≥ 0 α n τ n \exp_\phi=\sum_{n\geq0}\alpha_n\tau^n exp ϕ = ∑ n ≥ 0 α n τ n , converging in T \mathbb{T} T with respect to the Gauss norm. And so res t = θ ( ω u d t ) = − u \operatorname{res}_{t=\theta}(\omega_u\,dt)=-u res t = θ ( ω u d t ) = − u . For extensions to higher dimensional t t t -modules see [31 , 32 §4] .
Rigid analytic trivializations
We set
Θ : = ( 0 1 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ 1 ( t − θ ) / b r − b 1 / b r ⋯ − b r − 1 / b r ) \Theta:=\left(
\begin{array}{ccccc}
0 & 1 & \cdots& 0 \\
\vdots& \vdots& \ddots& \vdots\\
0 & 0 & \cdots& 1 \\
(t-\theta)/b_r & -b_1/b_r & \cdots& -b_{r-1}/b_r
\end{array}
\right) Θ := 0 ⋮ 0 ( t − θ ) / b r 1 ⋮ 0 − b 1 / b r ⋯ ⋱ ⋯ ⋯ 0 ⋮ 1 − b r − 1 / b r
and
Φ : = ( 0 1 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ 1 ( t − θ ) / b r ( − r ) − b 1 ( − 1 ) / b r ( − r ) ⋯ − b r − 1 ( − r + 1 ) / b r ( − r ) ) , \Phi:=\left(
\begin{array}{ccccc}
0 & 1 & \cdots& 0 \\
\vdots& \vdots& \ddots& \vdots\\
0 & 0 & \cdots& 1 \\
(t-\theta)/b_r^{(-r)} & -b_1^{(-1)}/b_r^{(-r)} & \cdots& -b_{r-1}^{(-r+1)}/b_r^{(-r)}
\end{array}
\right), Φ := 0 ⋮ 0 ( t − θ ) / b r ( − r ) 1 ⋮ 0 − b 1 ( − 1 ) / b r ( − r ) ⋯ ⋱ ⋯ ⋯ 0 ⋮ 1 − b r − 1 ( − r + 1 ) / b r ( − r ) ,
which represent multiplication by τ \tau τ (respectively by τ − 1 \tau^{-1} τ − 1 ) on the t t t -motive of ϕ \phi ϕ (respectively the t t t -comotive of ϕ \phi ϕ ). Though we will not pursue the motivic aspects in this paper directly, the reader is directed to [10 , 20 , 27 , 32 , 35 ] for more information.
We now fix an A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r of Λ ϕ \Lambda_\phi Λ ϕ , and let ω i : = ω π i \omega_i:=\omega_{\pi_i} ω i := ω π i for i = 1 , … , r i=1,\ldots,r i = 1 , … , r . Setting
Υ : = ( ω 1 ω 1 ( 1 ) ⋯ ω 1 ( r − 1 ) ⋮ ⋮ ⋮ ω r ω r ( 1 ) ⋯ ω r ( r − 1 ) ) (25) \Upsilon:=\left(
\begin{array}{cccc}
\omega_1 & \omega_1^{(1)} & \cdots& \omega_1^{(r-1)} \\
\vdots& \vdots& & \vdots\\
\omega_r & \omega_r^{(1)} & \cdots& \omega_r^{(r-1)}
\end{array}
\right) \tag*{(25)} Υ := ω 1 ⋮ ω r ω 1 ( 1 ) ⋮ ω r ( 1 ) ⋯ ⋯ ω 1 ( r − 1 ) ⋮ ω r ( r − 1 ) (25)
Pellarin [35 §4.2] proved that Υ ∈ GL r ( T ) \Upsilon\in\operatorname{GL}_r(\mathbb{T}) Υ ∈ GL r ( T ) , and moreover we observe that
Υ ( 1 ) = Υ Θ T , (26) \Upsilon^{(1)}=\Upsilon\Theta^{\mathsf{T}}, \tag*{(26)} Υ ( 1 ) = Υ Θ T , (26)
where Θ T \Theta^{\mathsf{T}} Θ T is the transpose of Θ \Theta Θ . Otherwise, as in [10 §3.4, 27 §2.5.2, 35 ] , if we let
V : = ( b 1 b 2 ( − 1 ) ⋯ b r − 1 ( − r + 2 ) b r ( − r + 1 ) b 2 b 3 ( − 1 ) ⋯ b r ( − r + 2 ) ⋮ ⋮ ⋱ ⋮ ⋮ ⋱ b r − 1 b r ( − 1 ) b r ) , (27) V:=\left(
\begin{array}{ccccc}
b_1 & b_2^{(-1)} & \cdots& b_{r-1}^{(-r+2)} & b_r^{(-r+1)} \\
b_2 & b_3^{(-1)} & \cdots& b_r^{(-r+2)} \\
\vdots& \vdots& \ddots\\
\vdots& \vdots& \ddots\\
b_{r-1} & b_r^{(-1)} \\
b_r
\end{array}
\right), \tag*{(27)} V := b 1 b 2 ⋮ ⋮ b r − 1 b r b 2 ( − 1 ) b 3 ( − 1 ) ⋮ ⋮ b r ( − 1 ) ⋯ ⋯ ⋱ ⋱ b r − 1 ( − r + 2 ) b r ( − r + 2 ) b r ( − r + 1 ) , (27)
then it follows that V ( − 1 ) Φ = Θ T V V^{(-1)}\Phi=\Theta^{\mathsf{T}}V V ( − 1 ) Φ = Θ T V . Setting
Ψ : = − V − 1 ( Υ ( 1 ) ) − 1 , (28) \Psi:=-V^{-1}\left(\Upsilon^{(1)}\right)^{-1}, \tag*{(28)} Ψ := − V − 1 ( Υ ( 1 ) ) − 1 , (28)
we see that
Ψ ( − 1 ) = Φ Ψ . (29) \Psi^{(-1)}=\Phi\Psi. \tag*{(29)} Ψ ( − 1 ) = ΦΨ. (29)
It follows from [2 Prop. 3.1.3] that the entries of Ψ \Psi Ψ are entire functions on C ∞ \mathbb{C}_{\infty} C ∞ , and we see that Ψ \Psi Ψ has the form
Ψ = ( ζ 1 ⋯ ζ r ζ 1 ( − 1 ) ⋯ ζ r ( − 1 ) ⋮ ⋮ ζ 1 ( − r + 1 ) ⋯ ζ r ( − r + 1 ) ) , (30) \Psi=
\begin{pmatrix}
\zeta_1 & \cdots& \zeta_r \\
\zeta_1^{(-1)} & \cdots& \zeta_r^{(-1)} \\
\vdots& & \vdots\\
\zeta_1^{(-r+1)} & \cdots& \zeta_r^{(-r+1)}
\end{pmatrix}, \tag*{(30)} Ψ = ζ 1 ζ 1 ( − 1 ) ⋮ ζ 1 ( − r + 1 ) ⋯ ⋯ ⋯ ζ r ζ r ( − 1 ) ⋮ ζ r ( − r + 1 ) , (30)
thus defining ζ 1 , … , ζ r ∈ T \zeta_1,\ldots,\zeta_r\in\mathbb{T} ζ 1 , … , ζ r ∈ T . Now (3.9) implies ζ 1 , … , ζ r \zeta_1,\ldots,\zeta_r ζ 1 , … , ζ r are dual special functions for ϕ \phi ϕ , and moreover they form an F q [ t ] \mathbb{F}_q[t] F q [ t ] -basis of all dual special functions [33 Prop. 3.4.7(b)] .
3.11. The matrix V V V . The matrix in (3.7) represents the Hartl-Juschka pairing between the t t t -comotive of ϕ \phi ϕ and its module of biderivations [27 §2.5.2] (see also [32 §4] for additional constructions). We find through calculation [27 Eq. (2.5.12), 32 Ex. 4.117] ,
V ( − 1 ) Φ = Θ T V = ( t − θ 0 ⋯ 0 0 b 2 ( − 1 ) ⋯ b r ( − r + 1 ) ⋮ ⋮ ⋱ 0 0 b r ( − 1 ) 0 ⋯ 0 ) . (31) V^{(-1)}\Phi=\Theta^{\mathsf{T}}V=
\begin{pmatrix}
t-\theta& 0 & \cdots& 0 \\
0 & b_2^{(-1)} & \cdots& b_r^{(-r+1)} \\
\vdots& \vdots& \ddots& 0 \\
0 & b_r^{(-1)} & 0 & \cdots& 0
\end{pmatrix}. \tag*{(31)} V ( − 1 ) Φ = Θ T V = t − θ 0 ⋮ 0 0 b 2 ( − 1 ) ⋮ b r ( − 1 ) ⋯ ⋯ ⋱ 0 0 b r ( − r + 1 ) 0 ⋯ 0 . (31)
From (3.6) and (3.8), we see that Ψ Υ = − V − 1 ( Θ T ) − 1 \Psi\Upsilon=-V^{-1}(\Theta^{\mathsf{T}})^{-1} ΨΥ = − V − 1 ( Θ T ) − 1 . Combining this with the preceding identity yields the following.
Lemma 3.13. We have
Ψ Υ = − ( Θ T V ) − 1 = − ( V ( − 1 ) Φ ) − 1 = ( 1 θ − t 0 ⋯ 0 0 0 ⋯ − 1 b r ( − 1 ) ⋮ ⋮ ⋱ ⋮ 0 − 1 b r ( − r + 1 ) ⋯ ∗ ) . \Psi\Upsilon=-(\Theta^{\mathsf{T}}V)^{-1}=-(V^{(-1)}\Phi)^{-1}=
\begin{pmatrix}
\frac{1}{\theta-t} & 0 & \cdots& 0 \\
0 & 0 & \cdots& \frac{-1}{b_r^{(-1)}} \\
\vdots& \vdots& \ddots& \vdots\\
0 & \frac{-1}{b_r^{(-r+1)}} & \cdots& *
\end{pmatrix}. ΨΥ = − ( Θ T V ) − 1 = − ( V ( − 1 ) Φ ) − 1 = θ − t 1 0 ⋮ 0 0 0 ⋮ b r ( − r + 1 ) − 1 ⋯ ⋯ ⋱ ⋯ 0 b r ( − 1 ) − 1 ⋮ ∗ .
Remark 3.14. (1) All the identities to this point have depended on our choices of the defining homomorphism for ϕ \phi ϕ and then the A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r for A ϕ A_\phi A ϕ . However, these are the only significant choices we make, and we design our future constructions to be consistent with them. (2) The inclusion of the negative sign in (3.8) makes Ψ \Psi Ψ differ slightly from previous incarnations, such as in [10 , 32 ] , but it aligns better with Ferraro [17 ] .
For N ∈ Z N\in\mathbb{Z} N ∈ Z , using (3.9) we note that Ψ ( N ) Ψ − 1 = ( Φ − 1 ) ( N ) ⋯ ( Φ − 1 ) ( 1 ) \Psi^{(N)}\Psi^{-1}=(\Phi^{-1})^{(N)}\cdots(\Phi^{-1})^{(1)} Ψ ( N ) Ψ − 1 = ( Φ − 1 ) ( N ) ⋯ ( Φ − 1 ) ( 1 ) when N ≥ 0 N\geq0 N ≥ 0 , and Ψ ( N ) Ψ − 1 = Φ ( N + 1 ) ⋯ Φ ( − 1 ) Φ \Psi^{(N)}\Psi^{-1}=\Phi^{(N+1)}\cdots\Phi^{(-1)}\Phi Ψ ( N ) Ψ − 1 = Φ ( N + 1 ) ⋯ Φ ( − 1 ) Φ when N < 0 N<0 N < 0 . Then applying Lemma 3.13,
[ Ψ ( N ) Υ ] 11 = 1 θ − t { [ ( Φ − 1 ) ( N ) ⋯ ( Φ − 1 ) ( 1 ) ] 11 , if N ≥ 0 , [ Φ ( N + 1 ) ⋯ Φ ( − 1 ) Φ ] 11 , if N < 0. (32) \left[\Psi^{(N)}\Upsilon\right]_{11}
=
\frac{1}{\theta-t}
\begin{cases}
\left[(\Phi^{-1})^{(N)}\cdots(\Phi^{-1})^{(1)}\right]_{11}, & \text{if } N\geq0,\\
\left[\Phi^{(N+1)}\cdots\Phi^{(-1)}\Phi\right]_{11}, & \text{if } N<0.
\end{cases} \tag*{(32)} [ Ψ ( N ) Υ ] 11 = θ − t 1 { [ ( Φ − 1 ) ( N ) ⋯ ( Φ − 1 ) ( 1 ) ] 11 , [ Φ ( N + 1 ) ⋯ Φ ( − 1 ) Φ ] 11 , if N ≥ 0 , if N < 0. (32)
This yields the following proposition, which serves as a generalization of [2 Lem. 5.4.1] to arbitrary Drinfeld A A A -modules. One should also compare with [17 Thm. 7.18] .
Proposition 3.16. For N ∈ Z N \in\mathbb{Z} N ∈ Z ,
[ Ψ ( N ) Γ ] 11 = { ∑ m = 0 ∞ { θ − m − 1 N } ϕ t m , if N ≥ 0 , − ∑ m = 0 ∞ { θ − m − 1 − N } ϕ ∗ t m , if N < 0. \left[\Psi^{(N)}\Gamma\right]_{11} =
\begin{cases}
\sum_{m=0}^{\infty} \left\{\begin{matrix}\theta^{-m-1}\\N\end{matrix}\right\}_{\phi} t^{m}, & \text{if } N \ge0,\\[8pt]
-\sum_{m=0}^{\infty} \left\{\begin{matrix}\theta^{-m-1}\\-N\end{matrix}\right\}_{\phi^*} t^{m}, & \text{if } N < 0.
\end{cases} [ Ψ ( N ) Γ ] 11 = ⎩ ⎨ ⎧ ∑ m = 0 ∞ { θ − m − 1 N } ϕ t m , − ∑ m = 0 ∞ { θ − m − 1 − N } ϕ ∗ t m , if N ≥ 0 , if N < 0.
Proof. Let S N ( t ) \mathcal{S}_{N}(t) S N ( t ) denote the left-hand side of the identity in the proposition, and likewise let T N ( t ) \mathcal{T}_{N}(t) T N ( t ) denote the right-hand side. We let ζ ϕ : = ( ζ 1 , … , ζ r ) ∈ Mat 1 × r ( T ) \boldsymbol{\zeta}_{\phi}:=(\zeta_{1},\ldots,\zeta_{r})\in\operatorname{Mat}_{1\times r}(\mathbb{T}) ζ ϕ := ( ζ 1 , … , ζ r ) ∈ Mat 1 × r ( T ) and ω ϕ : = ( ω 1 , … , ω r ) T ∈ Mat r × 1 ( T ) \boldsymbol{\omega}_{\phi}:=(\omega_{1},\ldots,\omega_{r})^{\mathsf{T}}\in\operatorname{Mat}_{r\times1}(\mathbb{T}) ω ϕ := ( ω 1 , … , ω r ) T ∈ Mat r × 1 ( T ) , which are the first row of Ψ \Psi Ψ and first column of Γ \Gamma Γ respectively. We note that the r r r cases where − r + 1 ≤ N ≤ 0 -r+1\le N\le0 − r + 1 ≤ N ≤ 0 are precisely obtained by considering the first column of Ψ Γ \Psi\Gamma ΨΓ in Lemma 3.13 . Indeed in these cases,
S N ( t ) = [ Ψ Γ ] − N + 1 , 1 = { 1 θ − t , if N = 0 , 0 if − r + 1 ≤ N ≤ − 1. (33) \mathcal{S}_{N}(t)=[\Psi\Gamma]_{-N+1,1}=
\begin{cases}
\frac{1}{\theta-t}, & \text{if } N=0,\\
0 & \text{if } -r+1\le N\le-1.
\end{cases} \tag*{(33)} S N ( t ) = [ ΨΓ ] − N + 1 , 1 = { θ − t 1 , 0 if N = 0 , if − r + 1 ≤ N ≤ − 1. (33)
When N = 0 N=0 N = 0 , we have T 0 ( t ) = ∑ m = 0 ∞ θ − m − 1 t m = 1 / ( θ − t ) = S 0 ( t ) \mathcal{T}_{0}(t)=\sum_{m=0}^{\infty}\theta^{-m-1}t^{m}=1/(\theta-t)=\mathcal{S}_{0}(t) T 0 ( t ) = ∑ m = 0 ∞ θ − m − 1 t m = 1/ ( θ − t ) = S 0 ( t ) , with this equality occurring in T \mathbb{T} T . Recalling from § 2.8 that { θ − m − 1 − N } ϕ ∗ = 0 \left\{\begin{matrix}\theta^{-m-1}\\-N\end{matrix}\right\}_{\phi^*}=0 { θ − m − 1 − N } ϕ ∗ = 0 for N > − r ( m + 1 ) N>-r(m+1) N > − r ( m + 1 ) , i.e., for m > − N / r − 1 m>-N/r-1 m > − N / r − 1 , it follows that T N ( t ) = 0 = S N ( t ) \mathcal{T}_{N}(t)=0=\mathcal{S}_{N}(t) T N ( t ) = 0 = S N ( t ) for − r + 1 ≤ N ≤ − 1 -r+1\le N\le-1 − r + 1 ≤ N ≤ − 1 .
Let N ∈ Z N\in\mathbb{Z} N ∈ Z . We claim that S N \mathcal{S}_{N} S N satisfies two recursions,
S N = 1 t − θ ∑ i = 1 r b i S N − i ( i ) = 1 t − θ ( N ) ∑ i = 1 r b i ( N − i ) S N − i . (34) \mathcal{S}_{N}=\frac{1}{t-\theta}\sum_{i=1}^{r}b_{i}\mathcal{S}_{N-i}^{(i)}=\frac{1}{t-\theta^{(N)}}\sum_{i=1}^{r}b_{i}^{(N-i)}\mathcal{S}_{N-i}. \tag*{(34)} S N = t − θ 1 i = 1 ∑ r b i S N − i ( i ) = t − θ ( N ) 1 i = 1 ∑ r b i ( N − i ) S N − i . (34)
Since ϕ θ ( ω ϕ ) = t ω ϕ \phi_{\theta}(\boldsymbol{\omega}_{\phi})=t\boldsymbol{\omega}_{\phi} ϕ θ ( ω ϕ ) = t ω ϕ , we have ω ϕ = ( t − θ ) − 1 ∑ i = 1 r b i ω ϕ ( i ) \boldsymbol{\omega}_{\phi}=(t-\theta)^{-1}\sum_{i=1}^{r}b_{i}\boldsymbol{\omega}_{\phi}^{(i)} ω ϕ = ( t − θ ) − 1 ∑ i = 1 r b i ω ϕ ( i ) , from which
S N = ζ ϕ ( N ) ω ϕ = 1 t − θ ∑ i = 1 r b i ζ ϕ ( N ) ω ϕ ( i ) = 1 t − θ ∑ i = 1 r b i S N − i ( i ) \mathcal{S}_{N}=\boldsymbol{\zeta}_{\phi}^{(N)}\boldsymbol{\omega}_{\phi}=\frac{1}{t-\theta}\sum_{i=1}^{r}b_{i}\boldsymbol{\zeta}_{\phi}^{(N)}\boldsymbol{\omega}_{\phi}^{(i)}=\frac{1}{t-\theta}\sum_{i=1}^{r}b_{i}\mathcal{S}_{N-i}^{(i)} S N = ζ ϕ ( N ) ω ϕ = t − θ 1 i = 1 ∑ r b i ζ ϕ ( N ) ω ϕ ( i ) = t − θ 1 i = 1 ∑ r b i S N − i ( i )
the first recursion in (34) . Likewise, ϕ θ ∗ ( ζ ϕ ) = t ζ ϕ \phi_{\theta}^{*}(\boldsymbol{\zeta}_{\phi})=t\boldsymbol{\zeta}_{\phi} ϕ θ ∗ ( ζ ϕ ) = t ζ ϕ implies ζ ϕ = ( t − θ ) − 1 ∑ i = 1 r b i ( − i ) ζ ϕ ( − i ) \boldsymbol{\zeta}_{\phi}=(t-\theta)^{-1}\sum_{i=1}^{r}b_{i}^{(-i)}\boldsymbol{\zeta}_{\phi}^{(-i)} ζ ϕ = ( t − θ ) − 1 ∑ i = 1 r b i ( − i ) ζ ϕ ( − i ) . Since S N = ζ ϕ ( N ) ω \mathcal{S}_{N}=\boldsymbol{\zeta}_{\phi}^{(N)}\boldsymbol{\omega} S N = ζ ϕ ( N ) ω , we obtain the second recursion in (34) .
We observe that either recursion in (34) uniquely determines { S N } N ∈ Z \{\mathcal{S}_{N}\}_{N\in\mathbb{Z}} { S N } N ∈ Z from the base information in (33) . Indeed for N > 0 N>0 N > 0 , S N \mathcal{S}_{N} S N is determined uniquely from S N − 1 , … , S N − r \mathcal{S}_{N-1},\ldots,\mathcal{S}_{N-r} S N − 1 , … , S N − r using either recursion. By rearranging and reindexing the second sum in (34) ,
S N = t − θ ( N + r ) b r ( N ) S N + r − 1 b r ( N ) ∑ i = 1 r − 1 b i ( N + r − i ) S N + r − i , N ∈ Z . \mathcal{S}_{N}=\frac{t-\theta^{(N+r)}}{b_{r}^{(N)}}\mathcal{S}_{N+r}-\frac{1}{b_{r}^{(N)}}\sum_{i=1}^{r-1}b_{i}^{(N+r-i)}\mathcal{S}_{N+r-i},\qquad N\in\mathbb{Z}. S N = b r ( N ) t − θ ( N + r ) S N + r − b r ( N ) 1 i = 1 ∑ r − 1 b i ( N + r − i ) S N + r − i , N ∈ Z .
One can make a similar rearrangement using the first sum in (34) , which we omit. Nevertheless for N ≤ − r N\le-r N ≤ − r , we have that either recursion in (34) defines S N \mathcal{S}_{N} S N in terms of S N + 1 , … , S N + r \mathcal{S}_{N+1},\ldots,\mathcal{S}_{N+r} S N + 1 , … , S N + r , starting with the base cases in (33) .
Let N > 0 N>0 N > 0 . Noting that { 1 N } ϕ = 0 \left\{\begin{matrix}1\\N\end{matrix}\right\}_{\phi}=0 { 1 N } ϕ = 0 for N > 0 N>0 N > 0 , it follows from (20) and the base cases for − r + 1 ≤ N ≤ − 1 -r+1\le N\le-1 − r + 1 ≤ N ≤ − 1 that T N \mathcal{T}_{N} T N satisfies the two recursions,
T N = 1 t − θ ∑ i = 1 r b i T N − i ( i ) = 1 t − θ ( N ) ∑ i = 1 r b i ( N − i ) T N − i . (35) \mathcal{T}_{N}=\frac{1}{t-\theta}\sum_{i=1}^{r}b_{i}\mathcal{T}_{N-i}^{(i)}=\frac{1}{t-\theta^{(N)}}\sum_{i=1}^{r}b_{i}^{(N-i)}\mathcal{T}_{N-i}. \tag*{(35)} T N = t − θ 1 i = 1 ∑ r b i T N − i ( i ) = t − θ ( N ) 1 i = 1 ∑ r b i ( N − i ) T N − i . (35)
Now let N < 0 N < 0 N < 0 . Noting that { 1 N } ϕ ∗ = { 1 − N } ϕ = 0 \left\{\frac{1}{N}\right\}_{\phi^*}=\left\{\frac{1}{-N}\right\}_{\phi}=0 { N 1 } ϕ ∗ = { − N 1 } ϕ = 0 for N < 0 N < 0 N < 0 , it follows from substituting n = − N n=-N n = − N in (23) and the base cases for − r + 1 ≤ N ≤ − 1 -r+1 \le N \le-1 − r + 1 ≤ N ≤ − 1 that T N \mathcal{T}_N T N , for N < 0 N < 0 N < 0 , also satisfies (35) . Therefore, { S N } N ∈ Z \{\mathcal{S}_N\}_{N\in\mathbb{Z}} { S N } N ∈ Z and { T N } N ∈ Z \{\mathcal{T}_N\}_{N\in\mathbb{Z}} { T N } N ∈ Z satisfy the same recursions with the same initial conditions, so as noted above we must have S N = T N \mathcal{S}_N=\mathcal{T}_N S N = T N for all N ∈ Z N\in\mathbb{Z} N ∈ Z . □ \square □
Remark 3.20. For all M , N ∈ Z M,N\in\mathbb{Z} M , N ∈ Z and 1 ≤ i , j ≤ r 1\le i,j\le r 1 ≤ i , j ≤ r , Proposition 3.16 permits calculation of [ Ψ ( N ) Υ ( M ) ] i j [\Psi^{(N)}\Upsilon^{(M)}]_{ij} [ Ψ ( N ) Υ ( M ) ] ij . Indeed, using (25) and (30) , we have
[ Ψ ( N ) Υ ( M ) ] i j = [ Ψ ( N − i + 1 ) Υ ( M + j − 1 ) ] 11 = [ Ψ ( N − M − i − j + 2 ) Υ ] 11 ( M + j − 1 ) . [\Psi^{(N)}\Upsilon^{(M)}]_{ij}=[\Psi^{(N-i+1)}\Upsilon^{(M+j-1)}]_{11}=[\Psi^{(N-M-i-j+2)}\Upsilon]_{11}^{(M+j-1)}. [ Ψ ( N ) Υ ( M ) ] ij = [ Ψ ( N − i + 1 ) Υ ( M + j − 1 ) ] 11 = [ Ψ ( N − M − i − j + 2 ) Υ ] 11 ( M + j − 1 ) .
Remark 3.21. When considering the N ≥ 0 N\ge0 N ≥ 0 case of Proposition 3.16, we note that Chen used a similar recursion to (34) in [11 Lem. 3.1] to prove that S N = B N / ( θ − t ) \mathcal{S}_N=\mathcal{B}_N/(\theta-t) S N = B N / ( θ − t ) , where { B N } \{\mathcal{B}_N\} { B N } is a sequence of rational functions defined in [15 ] using the combinatorics of shadowed partitions. A consequence of Proposition 3.16 is that
B N θ − t = ∑ m = 0 ∞ { θ m − 1 N } ϕ t m , N ≥ 0 , \frac{\mathcal{B}_N}{\theta-t}=\sum_{m=0}^{\infty}\left\{\begin{matrix}\theta^{m-1}\\N\end{matrix}\right\}_{\phi}t^m,\qquad N\ge0, θ − t B N = m = 0 ∑ ∞ { θ m − 1 N } ϕ t m , N ≥ 0 ,
which gives a new description of B N \mathcal{B}_N B N as a power series. Indeed the recursion in (34) and the corresponding one in [11 Lem. 3.1] are the characterizing recursions for B N \mathcal{B}_N B N in [15 Lem. 6.12] , and one can check using (17) that ( θ − t ) S N ∣ t = θ = β N (\theta-t)\mathcal{S}_N|_{t=\theta}=\beta_N ( θ − t ) S N ∣ t = θ = β N as in [15 Eq. (6.5)] .
Universal Anderson eigenvectors
We now consider results of Ferraro [17 ] , which provide remarkable extensions of work of Poonen [40 ] , particularly focusing on universal Anderson eigenvectors and dual Anderson eigenvectors. We compare them to the rigid analytic trivializations reviewed in §3.4, leading to new identities for ζ 1 , … , ζ r ∈ T \zeta_1,\ldots,\zeta_r\in\mathbb{T} ζ 1 , … , ζ r ∈ T from (30) (see Corollary 4.10).
Modules of eigenvectors
Because we are considering the case A = F q [ θ ] A=\mathbb{F}_q[\theta] A = F q [ θ ] , as opposed to general base rings A A A , several aspects of [17 ] simplify (see also [17 §7.5] ). For an A A A -module M M M , we let M ^ : = Hom F q c o n t ( M , F q ) \widehat{M}:=\operatorname{Hom}^{\mathrm{cont}}_{\mathbb{F}_q}(M,\mathbb{F}_q) M := Hom F q cont ( M , F q ) denote its Pontryagin dual, and for a pair of A A A -modules M , N M,N M , N , with N N N locally compact, their topological tensor product is M ⊗ ^ N : = Hom F q c o n t ( N ^ , M ^ ) M\widehat{\otimes}N:=\operatorname{Hom}^{\mathrm{cont}}_{\mathbb{F}_q}(\widehat{N},\widehat{M}) M ⊗ N := Hom F q cont ( N , M ) . See [17 §2, 40 §7] . Under these definitions it follows that
C ∞ ⊗ ^ A ≅ T , ( θ ⊗ 1 ↦ θ , 1 ⊗ θ ↦ t ) (36) \mathbb{C}_{\infty}\widehat{\otimes}A\cong\mathbb{T,\qquad}(\theta\otimes1\mapsto\theta,\ 1\otimes\theta\mapsto t) \tag*{(36)} C ∞ ⊗ A ≅ T , ( θ ⊗ 1 ↦ θ , 1 ⊗ θ ↦ t ) (36)
as topological A ⊗ F q A A\otimes_{\mathbb{F}_q}A A ⊗ F q A -modules. Now for a discrete A A A -module M M M , we set
S f ϕ ( M ) : = Hom A c o n t ( M ^ , ϕ ( C ∞ ) ) ≅ { α ∈ ϕ ( C ∞ ) ⊗ ^ M ∣ ( ϕ θ ⊗ 1 ) α = ( 1 ⊗ θ ) α } , \mathop{\mathrm{Sf}}\nolimits_{\phi}(M):=\operatorname{Hom}^{\mathrm{cont}}_A(\widehat{M},\phi(\mathbb{C}_{\infty}))\cong\{\alpha\in\phi(\mathbb{C}_{\infty})\widehat{\otimes}M\mid(\phi_{\theta}\otimes1)\alpha=(1\otimes\theta)\alpha\}, Sf ϕ ( M ) := Hom A cont ( M , ϕ ( C ∞ )) ≅ { α ∈ ϕ ( C ∞ ) ⊗ M ∣ ( ϕ θ ⊗ 1 ) α = ( 1 ⊗ θ ) α } ,
S f ϕ ∗ ( M ) : = Hom A c o n t ( M ^ , ϕ ∗ ( C ∞ ) ) ≅ { α ∈ ϕ ∗ ( C ∞ ) ⊗ ^ M ∣ ( ϕ θ ∗ ⊗ 1 ) α = ( 1 ⊗ θ ) α } . \mathop{\mathrm{Sf}}\nolimits_{\phi^*}(M):=\operatorname{Hom}^{\mathrm{cont}}_A(\widehat{M},\phi^*(\mathbb{C}_{\infty}))\cong\{\alpha\in\phi^*(\mathbb{C}_{\infty})\widehat{\otimes}M\mid(\phi^*_{\theta}\otimes1)\alpha=(1\otimes\theta)\alpha\}. Sf ϕ ∗ ( M ) := Hom A cont ( M , ϕ ∗ ( C ∞ )) ≅ { α ∈ ϕ ∗ ( C ∞ ) ⊗ M ∣ ( ϕ θ ∗ ⊗ 1 ) α = ( 1 ⊗ θ ) α } .
to be the A A A -modules of Anderson eigenvectors and dual Anderson eigenvectors of M M M (see [17 §3, §6] ). And indeed we see that S f ϕ ( A ) \mathop{\mathrm{Sf}}\nolimits_{\phi}(A) Sf ϕ ( A ) and S f ϕ ∗ ( A ) \mathop{\mathrm{Sf}}\nolimits_{\phi^*}(A) Sf ϕ ∗ ( A ) are precisely the modules of special functions and dual special functions of ϕ \phi ϕ contained in T \mathbb{T} T , as defined in §3.1 (see also [19 ] ). Considering S f ϕ \mathop{\mathrm{Sf}}\nolimits_{\phi} Sf ϕ and S f ϕ ∗ \mathop{\mathrm{Sf}}\nolimits_{\phi^*} Sf ϕ ∗ as functors from A A A -modules to A A A -modules, one of the main results of [17 ] is the following. We recall the convention that a sum ∑ a ∈ G S a \sum_{a\in G}S_a ∑ a ∈ G S a over an additive group G G G denotes that index a = 0 a=0 a = 0 is omitted.
Theorem 4.3 (Ferraro [17 Thm. 3.10 & Thm. 6.4] ). The functor S f ϕ \mathop{\mathrm{Sf}}\nolimits_{\phi} Sf ϕ is isomorphic to Hom A ( Hom A ( Λ ϕ , A d θ ) , − ) \operatorname{Hom}_A(\operatorname{Hom}_A(\Lambda_{\phi},A\,d\theta),-) Hom A ( Hom A ( Λ ϕ , A d θ ) , − ) , and S f ϕ ∗ \mathop{\mathrm{Sf}}\nolimits_{\phi^*} Sf ϕ ∗ is isomorphic to Hom A ( Λ ϕ , − ) \operatorname{Hom}_A(\Lambda_{\phi},-) Hom A ( Λ ϕ , − ) .
(a) Let η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) \eta_1,\ldots,\eta_r \in\operatorname{Hom}_A(\Lambda_\phi,A) η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) denote the dual A A A -basis of π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r . Then the universal element in Sf ϕ ( Hom A ( Λ ϕ , A d θ ) ) \operatorname{Sf}_\phi(\operatorname{Hom}_A(\Lambda_\phi,A\,d\theta)) Sf ϕ ( Hom A ( Λ ϕ , A d θ )) is
ω ϕ : = ∑ i = 1 r ∑ m = 0 ∞ exp ϕ ( π i θ m + 1 ) ⊗ θ m η i d θ ∈ ϕ ( C ∞ ) ⊗ ^ Hom A ( Λ ϕ , A d θ ) . \omega_\phi:= \sum_{i=1}^{r}\sum_{m=0}^{\infty}\exp_\phi\left(\frac{\pi_i}{\theta^{m+1}}\right)\otimes\theta^m\eta_i\,d\theta\in\phi(\mathbb{C}_\infty)\widehat{\otimes}\operatorname{Hom}_A(\Lambda_\phi,A\,d\theta). ω ϕ := i = 1 ∑ r m = 0 ∑ ∞ exp ϕ ( θ m + 1 π i ) ⊗ θ m η i d θ ∈ ϕ ( C ∞ ) ⊗ Hom A ( Λ ϕ , A d θ ) .
(b) The universal element in Sf ϕ ∗ ( Λ ϕ ) \operatorname{Sf}_{\phi^*}(\Lambda_\phi) Sf ϕ ∗ ( Λ ϕ ) is
ζ ϕ : = − ∑ λ ∈ Λ ϕ ′ λ − 1 ⊗ λ ∈ ϕ ∗ ( C ∞ ) ⊗ ^ Λ ϕ . \zeta_\phi:=-\sum_{\lambda\in\Lambda_\phi}'\lambda^{-1}\otimes\lambda\in\phi^*(\mathbb{C}_\infty)\widehat{\otimes}\Lambda_\phi. ζ ϕ := − λ ∈ Λ ϕ ∑ ′ λ − 1 ⊗ λ ∈ ϕ ∗ ( C ∞ ) ⊗ Λ ϕ .
Anderson eigenvectors
Theorem 4.3(a) implies that for a discrete A A A -module M M M ,
Sf ϕ ( M ) = { ∑ i = 0 r ∑ m = 0 ∞ exp ϕ ( π θ m + 1 ) ⊗ θ m μ i | μ 1 , … , μ r ∈ M } . \operatorname{Sf}_\phi(M)=\left\{\sum_{i=0}^{r}\sum_{m=0}^{\infty}\exp_\phi\left(\frac{\pi}{\theta^{m+1}}\right)\otimes\theta^m\mu_i\ \middle|\ \mu_1,\ldots,\mu_r\in M\right\}. Sf ϕ ( M ) = { i = 0 ∑ r m = 0 ∑ ∞ exp ϕ ( θ m + 1 π ) ⊗ θ m μ i μ 1 , … , μ r ∈ M } .
In particular by (36) , ω 1 , … , ω r \omega_1,\ldots,\omega_r ω 1 , … , ω r form an F q [ t ] \mathbb{F}_q[t] F q [ t ] -basis of Sf ϕ ( A ) \operatorname{Sf}_\phi(A) Sf ϕ ( A ) , as noted in §3.2.
Dual Anderson eigenvectors
Theorem 4.3(b) implies for a discrete A A A -module M M M ,
Sf ϕ ∗ ( M ) = { − ∑ λ ∈ Λ ϕ ′ λ − 1 ⊗ β ( λ ) | β ∈ Hom A ( Λ ϕ , M ) } . \operatorname{Sf}_{\phi^*}(M)=\left\{-\sum_{\lambda\in\Lambda_\phi}'\lambda^{-1}\otimes\beta(\lambda)\ \middle|\ \beta\in\operatorname{Hom}_A(\Lambda_\phi,M)\right\}. Sf ϕ ∗ ( M ) = ⎩ ⎨ ⎧ − λ ∈ Λ ϕ ∑ ′ λ − 1 ⊗ β ( λ ) β ∈ Hom A ( Λ ϕ , M ) ⎭ ⎬ ⎫ .
Therefore, with η i : Λ ϕ → A \eta_i:\Lambda_\phi\to A η i : Λ ϕ → A as above, we define for i = 1 , … , r i=1,\ldots,r i = 1 , … , r ,
ξ i : = − ∑ λ ∈ Λ ϕ ′ λ − 1 ⊗ η i ( λ ) ∈ T , \xi_i:=-\sum_{\lambda\in\Lambda_\phi}'\lambda^{-1}\otimes\eta_i(\lambda)\in\mathbb{T}, ξ i := − λ ∈ Λ ϕ ∑ ′ λ − 1 ⊗ η i ( λ ) ∈ T ,
and then ξ 1 , … , ξ r \xi_1,\ldots,\xi_r ξ 1 , … , ξ r form an F q [ t ] \mathbb{F}_q[t] F q [ t ] -basis of Sf ϕ ∗ ( A ) ⊆ T \operatorname{Sf}_{\phi^*}(A)\subseteq\mathbb{T} Sf ϕ ∗ ( A ) ⊆ T . If we consider the dual basis relation of η i ( π j ) = δ i j \eta_i(\pi_j)=\delta_{ij} η i ( π j ) = δ ij for 1 ≤ i , j ≤ r 1\leq i,j\leq r 1 ≤ i , j ≤ r , then we can rewrite the ξ i \xi_i ξ i as
ξ i = − ∑ ( a 1 , … , a r ) ∈ A r a i ( t ) a 1 π 1 + ⋯ + a r π r , 1 ≤ i ≤ r . (37) \xi_i=-\sum_{(a_1,\ldots,a_r)\in A^r}\frac{a_i(t)}{a_1\pi_1+\cdots+a_r\pi_r},\qquad1\leq i\leq r. \tag*{(37)} ξ i = − ( a 1 , … , a r ) ∈ A r ∑ a 1 π 1 + ⋯ + a r π r a i ( t ) , 1 ≤ i ≤ r . (37)
Theorem 4.7 (Ferraro [17, §7]). The canonical A A A -bilinear pairing, ( λ , α ) ↦ α ( λ ) : Λ ϕ × Hom A ( Λ ϕ , A d θ ) → A d θ (\lambda,\alpha)\mapsto\alpha(\lambda):\Lambda_\phi\times\operatorname{Hom}_A(\Lambda_\phi,A\,d\theta)\to A\,d\theta ( λ , α ) ↦ α ( λ ) : Λ ϕ × Hom A ( Λ ϕ , A d θ ) → A d θ , induces a C ∞ ⊗ F q A \mathbb{C}_\infty\otimes_{\mathbb{F}_q}A C ∞ ⊗ F q A -bilinear pairing
( α , β ) ↦ α ⋅ β : C ∞ ⊗ ^ Λ ϕ × C ∞ ⊗ ^ Hom A ( Λ ϕ , A d θ ) → ( C ∞ ⊗ ^ A ) d t , (38) (\alpha,\beta)\mapsto\alpha\cdot\beta:\mathbb{C}_\infty\widehat{\otimes}\Lambda_\phi\times\mathbb{C}_\infty\widehat{\otimes}\operatorname{Hom}_A(\Lambda_\phi,A\,d\theta)\to(\mathbb{C}_\infty\widehat{\otimes}A)\,dt, \tag*{(38)} ( α , β ) ↦ α ⋅ β : C ∞ ⊗ Λ ϕ × C ∞ ⊗ Hom A ( Λ ϕ , A d θ ) → ( C ∞ ⊗ A ) d t , (38) under which
ζ ϕ ⋅ ω ϕ ( k ) = { d ( 1 ⊗ θ ) θ ⊗ 1 − 1 ⊗ θ if k = 0 , 0 if 1 ≤ k ≤ r − 1. \zeta_\phi\cdot\omega_\phi^{(k)}=
\begin{cases}
\dfrac{d(1\otimes\theta)}{\theta\otimes1-1\otimes\theta} & \text{if } k=0,\\
0 & \text{if } 1\leq k\leq r-1.
\end{cases} ζ ϕ ⋅ ω ϕ ( k ) = ⎩ ⎨ ⎧ θ ⊗ 1 − 1 ⊗ θ d ( 1 ⊗ θ ) 0 if k = 0 , if 1 ≤ k ≤ r − 1. where ω ϕ ( k ) \omega_\phi^{(k)} ω ϕ ( k ) is the k k k -th twist under τ ⊗ id \tau\otimes\operatorname{id} τ ⊗ id .
As in the proof of Proposition 3.16 , we recall ζ ϕ = ( ζ 1 , … , ζ r ) ∈ Mat 1 × r ( T ) \zeta_\phi=(\zeta_1,\ldots,\zeta_r)\in\operatorname{Mat}_{1\times r}(\mathbb{T}) ζ ϕ = ( ζ 1 , … , ζ r ) ∈ Mat 1 × r ( T ) and ω ϕ = ( ω 1 , … , ω r ) T ∈ Mat r × 1 ( T ) \omega_\phi=(\omega_1,\ldots,\omega_r)^{\mathsf{T}}\in\operatorname{Mat}_{r\times1}(\mathbb{T}) ω ϕ = ( ω 1 , … , ω r ) T ∈ Mat r × 1 ( T ) , and we further let ξ ϕ : = ( ξ 1 , … , ξ r ) ∈ Mat 1 × r ( T ) \xi_\phi:=(\xi_1,\ldots,\xi_r)\in\operatorname{Mat}_{1\times r}(\mathbb{T}) ξ ϕ := ( ξ 1 , … , ξ r ) ∈ Mat 1 × r ( T ) . Now under our choice of A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r for Λ ϕ \Lambda_\phi Λ ϕ , the C ∞ [ t ] \mathbb{C}_\infty[t] C ∞ [ t ] -bilinear pairing induced by (4.8), Mat 1 × r ( T ) × Mat r × 1 ( T ) → T \operatorname{Mat}_{1\times r}(\mathbb{T})\times\operatorname{Mat}_{r\times1}(\mathbb{T})\to\mathbb{T} Mat 1 × r ( T ) × Mat r × 1 ( T ) → T , is simply the dot product (canonically identifying T ≅ T d t \mathbb{T}\cong\mathbb{T}\,dt T ≅ T d t ). In particular, Theorem 4.7 implies
ξ ϕ ⋅ ω ϕ ( k ) = { 1 θ − t if k = 0 , 0 if 1 ≤ k ≤ r − 1. (39) \xi_\phi\cdot\omega_\phi^{(k)}=
\begin{cases}
\dfrac{1}{\theta-t} & \text{if } k=0,\\
0 & \text{if } 1\leq k\leq r-1.
\end{cases} \tag*{(39)} ξ ϕ ⋅ ω ϕ ( k ) = ⎩ ⎨ ⎧ θ − t 1 0 if k = 0 , if 1 ≤ k ≤ r − 1. (39)
The following corollary shows that our dual special functions ζ 1 , … , ζ r \zeta_1,\ldots,\zeta_r ζ 1 , … , ζ r from (30) are in fact the same as ξ 1 , … , ξ r \xi_1,\ldots,\xi_r ξ 1 , … , ξ r , i.e., ζ ϕ = ξ ϕ \zeta_\phi=\xi_\phi ζ ϕ = ξ ϕ .
Corollary 4.10. Under our choice of A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r for Λ ϕ \Lambda_\phi Λ ϕ , we have
ζ i = ξ i = − ∑ ( a 1 , … , a r ) ∈ A r ′ a i ( t ) a 1 π 1 + ⋯ + a r π r , ∀ i = 1 , … , r . \zeta_i=\xi_i=-\sum_{(a_1,\ldots,a_r)\in A^r}^{\prime}\frac{a_i(t)}{a_1\pi_1+\cdots+a_r\pi_r},\qquad\forall i=1,\ldots,r. ζ i = ξ i = − ( a 1 , … , a r ) ∈ A r ∑ ′ a 1 π 1 + ⋯ + a r π r a i ( t ) , ∀ i = 1 , … , r . Proof. We note that (39) can be translated into the matrix form, ( ξ 1 , … , ξ r ) ⋅ Υ = ( 1 / ( θ − t ) , 0 , … , 0 ) (\xi_1,\ldots,\xi_r)\cdot\Upsilon=(1/(\theta-t),0,\ldots,0) ( ξ 1 , … , ξ r ) ⋅ Υ = ( 1/ ( θ − t ) , 0 , … , 0 ) . Since Υ ∈ GL r ( T ) \Upsilon\in\operatorname{GL}_r(\mathbb{T}) Υ ∈ GL r ( T ) , we have
( ξ 1 , … , ξ r ) = ( 1 θ − t , 0 , … , 0 ) Υ − 1 = first row of ( Ψ Υ ) Υ − 1 = ( ζ 1 , … , ζ r ) , (\xi_1,\ldots,\xi_r)=\left(\frac{1}{\theta-t},0,\ldots,0\right)\Upsilon^{-1}=\text{first row of }(\Psi\Upsilon)\Upsilon^{-1}=(\zeta_1,\ldots,\zeta_r), ( ξ 1 , … , ξ r ) = ( θ − t 1 , 0 , … , 0 ) Υ − 1 = first row of ( ΨΥ ) Υ − 1 = ( ζ 1 , … , ζ r ) , where the second equality follows from Lemma 3.13 . We are done by (37) .
The maps e λ e_\lambda e λ and e η ∗ e_\eta^* e η ∗
Residue maps
For a field F F F and a double-tailed Laurent series
x = ∑ j ∈ Z c j z − j ∈ F [ [ z , z − 1 ] ] , x=\sum_{j\in\mathbb{Z}}c_jz^{-j}\in F[[z,z^{-1}]], x = j ∈ Z ∑ c j z − j ∈ F [[ z , z − 1 ]] ,
we have the F F F -linear map R z : F [ [ z , z − 1 ] ] → F R_z:F[[z,z^{-1}]]\to F R z : F [[ z , z − 1 ]] → F given by
R z ( x ) : = c 1 . R_z(x):=c_1. R z ( x ) := c 1 .
We note that if t t t is the canonical uniformizing parameter on P 1 / F \mathbb{P}^1/F P 1 / F vanishing at 0 0 0 and if, for f ∈ F ( t ) f\in F(t) f ∈ F ( t ) , the differential f d t f\,dt f d t has expansion at ∞ \infty ∞ , f d t = ∑ j ≥ j 0 c j t − j d t f\,dt=\sum_{j\geq j_0}c_jt^{-j}\,dt f d t = ∑ j ≥ j 0 c j t − j d t , then
c 1 = R t ( ∑ j ≥ j 0 c j t − j ) = − Res ∞ ( f d t ) . (40) c_1=R_t\left(\sum_{j\geq j_0}c_jt^{-j}\right)=-\operatorname{Res}_{\infty}(f\,dt). \tag*{(40)} c 1 = R t ( j ≥ j 0 ∑ c j t − j ) = − Res ∞ ( f d t ) . (40)
where the latter is the standard residue of a differential form on an algebraic curve.
The maps e λ e_\lambda e λ
For λ ∈ Λ ϕ \lambda\in\Lambda_\phi λ ∈ Λ ϕ , define the A A A -module homomorphism
e λ : = ( x ↦ exp ϕ ( λ x ) ) : k ∞ → ϕ ( C ∞ ) . (41) e_\lambda:=\left(x\mapsto\exp_\phi(\lambda x)\right):k_\infty\to\phi(\mathbb{C}_\infty). \tag*{(41)} e λ := ( x ↦ exp ϕ ( λ x ) ) : k ∞ → ϕ ( C ∞ ) . (41)
As such, ker e λ ⊇ A \ker e_\lambda\supseteq A ker e λ ⊇ A for each λ ∈ Λ ϕ ∖ { 0 } \lambda\in\Lambda_\phi\setminus\{0\} λ ∈ Λ ϕ ∖ { 0 } . Moreover, the induced map
λ ↦ e λ : Λ ϕ → Hom A ( k ∞ / A , ϕ ( C ∞ ) ) (42) \lambda\mapsto e_\lambda:\Lambda_\phi\to\operatorname{Hom}_A(k_\infty/A,\phi(\mathbb{C}_\infty)) \tag*{(42)} λ ↦ e λ : Λ ϕ → Hom A ( k ∞ / A , ϕ ( C ∞ )) (42)
is injective and A A A -linear. For our choice of A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r of Λ ϕ \Lambda_\phi Λ ϕ , we set
e i ( x ) : = e π i ( x ) , 1 ≤ i ≤ r , e_i(x):=e_{\pi_i}(x),\qquad1\leq i\leq r, e i ( x ) := e π i ( x ) , 1 ≤ i ≤ r ,
for which ker e i = A \ker e_i=A ker e i = A . In particular for each b ∈ A b\in A b ∈ A , b ≠ 0 b\ne0 b = 0 , it follows from (19) that
ϕ [ b ] = ⨁ i = 1 r e i ( b − 1 A ) . (43) \phi[b]=\bigoplus_{i=1}^{r}e_i\left(b^{-1}A\right). \tag*{(43)} ϕ [ b ] = i = 1 ⨁ r e i ( b − 1 A ) . (43)
Now for x ∈ k ∞ x\in k_\infty x ∈ k ∞ , we can form the product x ( t ) ω λ ( t ) ∈ C ∞ [ [ t , t − 1 ] ] x(t)\omega_\lambda(t)\in\mathbb{C}_\infty[[t,t^{-1}]] x ( t ) ω λ ( t ) ∈ C ∞ [[ t , t − 1 ]] . Indeed, for example by the argument in [32 Rem. 4.29] , for α ∈ C ∞ \alpha\in\mathbb{C}_\infty α ∈ C ∞ with 1 < ∣ α ∣ ∞ < ∣ θ ∣ ∞ 1<|\alpha|_\infty<|\theta|_\infty 1 < ∣ α ∣ ∞ < ∣ θ ∣ ∞ , we have x ( t ) , ω λ ( t ) ∈ S α x(t),\omega_\lambda(t)\in S_\alpha x ( t ) , ω λ ( t ) ∈ S α as in § 2.1 , and so x ( t ) ω λ ( t ) ∈ S α ⊆ C ∞ [ [ t , t − 1 ] ] x(t)\omega_\lambda(t)\in S_\alpha\subseteq\mathbb{C}_\infty[[t,t^{-1}]] x ( t ) ω λ ( t ) ∈ S α ⊆ C ∞ [[ t , t − 1 ]] . We then have the following.
Lemma 5.7. Let λ ∈ Λ ϕ \lambda\in\Lambda_\phi λ ∈ Λ ϕ . For x ∈ k ∞ x\in k_\infty x ∈ k ∞ , we have e λ ( x ) = R t ( x ( t ) ω λ ( t ) ) e_\lambda(x)=R_t(x(t)\omega_\lambda(t)) e λ ( x ) = R t ( x ( t ) ω λ ( t )) .
Proof. Let x = ∑ j ≥ j 0 c j θ − j x=\sum_{j\ge j_0}c_j\theta^{-j} x = ∑ j ≥ j 0 c j θ − j . From the definition of ω λ \omega_\lambda ω λ in §3.2, we have
ω λ ( t ) = ∑ m ≥ 0 e λ ( 1 / θ m + 1 ) t m . (44) \omega_\lambda(t)=\sum_{m\ge0}e_\lambda(1/\theta^{m+1})t^m. \tag*{(44)} ω λ ( t ) = m ≥ 0 ∑ e λ ( 1/ θ m + 1 ) t m . (44)
Thus, the coefficient of t − 1 t^{-1} t − 1 in x ( t ) ω λ ( t ) x(t)\omega_\lambda(t) x ( t ) ω λ ( t ) is
∑ m = 0 ∞ c m + 1 e λ ( 1 / θ m + 1 ) = e λ ( ∑ j ≥ 1 c j θ − j ) = e λ ( x ) , \sum_{m=0}^{\infty}c_{m+1}e_\lambda(1/\theta^{m+1})=e_\lambda\left(\sum_{j\ge1}c_j\theta^{-j}\right)=e_\lambda(x), m = 0 ∑ ∞ c m + 1 e λ ( 1/ θ m + 1 ) = e λ ( j ≥ 1 ∑ c j θ − j ) = e λ ( x ) ,
the first equality by the continuity and F q \mathbb{F}_q F q -linearity of e λ e_\lambda e λ , the second since e λ ( A ) = 0 e_\lambda(A)=0 e λ ( A ) = 0 . □ \square □
The maps e η ∗ e_\eta^* e η ∗
As in §5.3, we would like to define A A A -module homomorphisms
e η ∗ : k ∞ ⟶ ϕ ∗ ( C ∞ ) , η ∈ Hom A ( Λ ϕ , A ) , e_\eta^*:k_\infty\longrightarrow\phi^*(\mathbb{C}_\infty),\qquad\eta\in\operatorname{Hom}_A(\Lambda_\phi,A), e η ∗ : k ∞ ⟶ ϕ ∗ ( C ∞ ) , η ∈ Hom A ( Λ ϕ , A ) ,
that satisfy similar properties such as (5.4)–(5.8), and particularly Lemma 5.7 . However, as indicated in § 2.6 , the series exp ϕ ∗ \exp_\phi^* exp ϕ ∗ and log ϕ ∗ \log_\phi^* log ϕ ∗ are inadequate to the task.
Instead, let η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) \eta_1,\ldots,\eta_r\in\operatorname{Hom}_A(\Lambda_\phi,A) η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) denote the dual basis of π 1 , … , π r ∈ Λ ϕ \pi_1,\ldots,\pi_r\in\Lambda_\phi π 1 , … , π r ∈ Λ ϕ . For η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) , writing η = s 1 η 1 + ⋯ + s r η r \eta=s_1\eta_1+\cdots+s_r\eta_r η = s 1 η 1 + ⋯ + s r η r for s 1 , … , s r ∈ A s_1,\ldots,s_r\in A s 1 , … , s r ∈ A , we set
ζ η ( t ) : = ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = s 1 ( t ) ζ 1 ( t ) + ⋯ + s r ( t ) ζ r ( t ) . (45) \zeta_\eta(t):=\boldsymbol{\zeta}_\phi\cdot(s_1(t),\ldots,s_r(t))^{\mathsf T}=s_1(t)\zeta_1(t)+\cdots+s_r(t)\zeta_r(t). \tag*{(45)} ζ η ( t ) := ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = s 1 ( t ) ζ 1 ( t ) + ⋯ + s r ( t ) ζ r ( t ) . (45)
We note that ζ i ( t ) = ζ η i ( t ) \zeta_i(t)=\zeta_{\eta_i}(t) ζ i ( t ) = ζ η i ( t ) , for 1 ≤ i ≤ r 1\le i\le r 1 ≤ i ≤ r . The reader will recognize that Theorem 4.3 of Ferraro necessitates indexing ζ η ( t ) \zeta_\eta(t) ζ η ( t ) by Hom A ( Λ ϕ , A ) \operatorname{Hom}_A(\Lambda_\phi,A) Hom A ( Λ ϕ , A ) , rather than simply by Λ ϕ \Lambda_\phi Λ ϕ . It is also clear from the following lemma.
Lemma 5.11. For η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) , the function ζ η ( t ) ∈ T \zeta_\eta(t)\in\mathbb{T} ζ η ( t ) ∈ T is independent of the choice of A A A -basis π 1 , … , π r \pi_1,\ldots,\pi_r π 1 , … , π r .
Proof. Suppose π 1 ′ , … , π r ′ \pi'_1,\ldots,\pi'_r π 1 ′ , … , π r ′ is an A A A -basis of Λ ϕ \Lambda_\phi Λ ϕ with corresponding dual basis η 1 ′ , … , η r ′ \eta'_1,\ldots,\eta'_r η 1 ′ , … , η r ′ , and let ζ η ′ ( t ) \zeta'_\eta(t) ζ η ′ ( t ) denote (5.10) with respect to this basis. Then for some B ∈ GL r ( A ) B\in\operatorname{GL}_r(A) B ∈ GL r ( A ) ,
( π 1 ′ , … , π r ′ ) T = B ( π 1 , … , π r ) T , B T ( η 1 ′ , … , η r ′ ) T = ( η 1 , … , η r ) T . (\pi'_1,\ldots,\pi'_r)^{\mathsf T}=B(\pi_1,\ldots,\pi_r)^{\mathsf T},\qquad B^{\mathsf T}(\eta'_1,\ldots,\eta'_r)^{\mathsf T}=(\eta_1,\ldots,\eta_r)^{\mathsf T}. ( π 1 ′ , … , π r ′ ) T = B ( π 1 , … , π r ) T , B T ( η 1 ′ , … , η r ′ ) T = ( η 1 , … , η r ) T .
Let Υ ′ ∈ GL r ( T ) \Upsilon'\in\operatorname{GL}_r(\mathbb{T}) Υ ′ ∈ GL r ( T ) be defined as in (25) with respect to π 1 ′ , … , π r ′ \pi'_1,\ldots,\pi'_r π 1 ′ , … , π r ′ . Taking B ( t ) = B ∣ θ = t ∈ GL r ( F q [ t ] ) B(t)=B|_{\theta=t}\in\operatorname{GL}_r(\mathbb{F}_q[t]) B ( t ) = B ∣ θ = t ∈ GL r ( F q [ t ]) , it follows from (24) that Υ ′ = B ( t ) Υ \Upsilon'=B(t)\Upsilon Υ ′ = B ( t ) Υ . The corresponding rigid analytic trivialization Ψ ′ \Psi' Ψ ′ as in (28) is then
Ψ ′ = − V − 1 ( Υ ′ ( 1 ) ) − 1 = − V − 1 ( Υ ( 1 ) ) − 1 B ( t ) − 1 = Ψ B ( t ) − 1 . \Psi'=-V^{-1}\left(\Upsilon'^{(1)}\right)^{-1}=-V^{-1}\left(\Upsilon^{(1)}\right)^{-1}B(t)^{-1}=\Psi B(t)^{-1}. Ψ ′ = − V − 1 ( Υ ′ ( 1 ) ) − 1 = − V − 1 ( Υ ( 1 ) ) − 1 B ( t ) − 1 = Ψ B ( t ) − 1 .
from which we see that for ζ ϕ ′ = ( ζ 1 ′ , … , ζ r ′ ) \boldsymbol{\zeta}'_\phi=(\zeta'_1,\ldots,\zeta'_r) ζ ϕ ′ = ( ζ 1 ′ , … , ζ r ′ ) , the first row of Ψ ′ \Psi' Ψ ′ , we have ζ ϕ = ζ ϕ ′ ⋅ B ( t ) \boldsymbol{\zeta}_\phi=\boldsymbol{\zeta}'_\phi\cdot B(t) ζ ϕ = ζ ϕ ′ ⋅ B ( t ) . Thus, if η = ( s 1 , … , s r ) ⋅ ( η 1 , … , η r ) T \eta=(s_1,\ldots,s_r)\cdot(\eta_1,\ldots,\eta_r)^{\mathsf T} η = ( s 1 , … , s r ) ⋅ ( η 1 , … , η r ) T , then by definition, ζ η ( t ) = ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T \zeta_\eta(t)=\boldsymbol{\zeta}_\phi\cdot(s_1(t),\ldots,s_r(t))^{\mathsf T} ζ η ( t ) = ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T . On the other hand, η = ( s 1 , … , s r ) B T ( η 1 ′ , … , η r ′ ) T \eta=(s_1,\ldots,s_r)B^{\mathsf T}(\eta'_1,\ldots,\eta'_r)^{\mathsf T} η = ( s 1 , … , s r ) B T ( η 1 ′ , … , η r ′ ) T , which implies
ζ η ′ ( t ) = ζ ϕ ′ ⋅ B ( t ) ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = ζ η ( t ) . \zeta'_\eta(t)=\boldsymbol{\zeta}'_\phi\cdot B(t)\cdot(s_1(t),\ldots,s_r(t))^{\mathsf T}=\boldsymbol{\zeta}_\phi\cdot(s_1(t),\ldots,s_r(t))^{\mathsf T}=\zeta_\eta(t). ζ η ′ ( t ) = ζ ϕ ′ ⋅ B ( t ) ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = ζ ϕ ⋅ ( s 1 ( t ) , … , s r ( t ) ) T = ζ η ( t ) .
□ \square □
Now taking inspiration from [2 §5.2.5] , we define for η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) and x ∈ k ∞ x\in k_\infty x ∈ k ∞ ,
e η ∗ ( x ) : = Res t ( x ( t ) ζ η ( t ) ) , (46) e_\eta^*(x):=\operatorname{Res}_t\bigl(x(t)\zeta_\eta(t)\bigr), \tag*{(46)} e η ∗ ( x ) := Res t ( x ( t ) ζ η ( t ) ) , (46)
noting as in §5.3 that x ( t ) ζ η ( t ) ∈ S α ⊆ C ∞ [ [ t , t − 1 ] ] x(t)\zeta_\eta(t)\in S_\alpha\subseteq\mathbb{C}_\infty[[t,t^{-1}]] x ( t ) ζ η ( t ) ∈ S α ⊆ C ∞ [[ t , t − 1 ]] for any α ∈ C ∞ \alpha\in\mathbb{C}_\infty α ∈ C ∞ with 1 < ∣ α ∣ ∞ < ∣ θ ∣ ∞ 1<|\alpha|_\infty<|\theta|_\infty 1 < ∣ α ∣ ∞ < ∣ θ ∣ ∞ . Thus by definition e η ∗ e_\eta^* e η ∗ satisfies the conclusion of Lemma 5.7 , and by Lemma 5.11 its definition does not depend on the representation of η \eta η with respect to our A A A -basis of Λ ϕ \Lambda_\phi Λ ϕ . Because ζ η ( t ) ∈ C ∞ [ [ t ] ] \zeta_\eta(t) \in C_\infty[[t]] ζ η ( t ) ∈ C ∞ [[ t ]] , the definition of R t R_t R t implies ker e η ∗ ⊇ A \ker e_\eta^* \supseteq A ker e η ∗ ⊇ A , and as in (42) , the map
η ↦ e η ∗ : Hom A ( Λ ϕ , A ) ⟶ Hom A ( k ∞ / A , ϕ ∗ ( C ∞ ) ) \eta\mapsto e_\eta^*: \operatorname{Hom}_A(\Lambda_\phi,A) \longrightarrow\operatorname{Hom}_A(k_\infty/A,\phi^*(C_\infty)) η ↦ e η ∗ : Hom A ( Λ ϕ , A ) ⟶ Hom A ( k ∞ / A , ϕ ∗ ( C ∞ ))
is injective and A A A -linear. By (46) , taking x = θ − m − 1 x=\theta^{-m-1} x = θ − m − 1 yields
ζ η ( t ) = ∑ m = 0 ∞ e η ∗ ( 1 / θ m + 1 ) t m (47) \zeta_\eta(t)=\sum_{m=0}^{\infty}e_\eta^*(1/\theta^{m+1})t^m \tag*{(47)} ζ η ( t ) = m = 0 ∑ ∞ e η ∗ ( 1/ θ m + 1 ) t m (47)
which gives the companion identity to (44) . For an A A A -basis π 1 , … , π r ∈ Λ ϕ \pi_1,\ldots,\pi_r\in\Lambda_\phi π 1 , … , π r ∈ Λ ϕ and corresponding dual basis η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) \eta_1,\ldots,\eta_r\in\operatorname{Hom}_A(\Lambda_\phi,A) η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) , we set
e i ∗ : = e η i ∗ : k ∞ ⟶ ϕ ∗ ( C ∞ ) , 1 ≤ i ≤ r , e_i^*:=e_{\eta_i}^*:k_\infty\longrightarrow\phi^*(C_\infty),\qquad1\leq i\leq r, e i ∗ := e η i ∗ : k ∞ ⟶ ϕ ∗ ( C ∞ ) , 1 ≤ i ≤ r ,
and so if η = s 1 η 1 + ⋯ + s r η r \eta=s_1\eta_1+\cdots+s_r\eta_r η = s 1 η 1 + ⋯ + s r η r with s 1 , … , s r ∈ A s_1,\ldots,s_r\in A s 1 , … , s r ∈ A , then e η ∗ = ϕ s 1 ∗ ∘ e 1 ∗ + ⋯ + ϕ s r ∗ ∘ e r ∗ e_\eta^*=\phi_{s_1}^*\circ e_1^*+\cdots+\phi_{s_r}^*\circ e_r^* e η ∗ = ϕ s 1 ∗ ∘ e 1 ∗ + ⋯ + ϕ s r ∗ ∘ e r ∗ .
Lemma 5.14. Let η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) . For x ∈ k ∞ x\in k_\infty x ∈ k ∞ ,
e η ∗ ( x ) = − ∑ λ ∈ Λ ϕ ′ R θ ( x η ( λ ) ) λ . e_\eta^*(x)=-\sum_{\lambda\in\Lambda_\phi}'\frac{R_\theta(x\eta(\lambda))}{\lambda}. e η ∗ ( x ) = − λ ∈ Λ ϕ ∑ ′ λ R θ ( x η ( λ )) .
Proof. By forgetting A A A -linearity, we have an inclusion Hom A ( Λ ϕ , A ) ↪ Hom F q ( Λ ϕ , A ) \operatorname{Hom}_A(\Lambda_\phi,A)\hookrightarrow\operatorname{Hom}_{\mathbb{F}_q}(\Lambda_\phi,A) Hom A ( Λ ϕ , A ) ↪ Hom F q ( Λ ϕ , A ) , and by picking an F q \mathbb{F}_q F q -basis of A A A , we have Hom F q ( Λ ϕ , A ) ≅ ⨁ j ≥ 0 Hom F q ( Λ ϕ , F q ) \operatorname{Hom}_{\mathbb{F}_q}(\Lambda_\phi,A)\cong\bigoplus_{j\geq0}\operatorname{Hom}_{\mathbb{F}_q}(\Lambda_\phi,\mathbb{F}_q) Hom F q ( Λ ϕ , A ) ≅ ⨁ j ≥ 0 Hom F q ( Λ ϕ , F q ) . Taking { θ j } \{\theta^j\} { θ j } as this basis, the induced isomorphism is given by
η ↦ ( R θ ( θ − 1 η ( ⋅ ) ) , R θ ( θ − 2 η ( ⋅ ) ) , … ) . \eta\mapsto\left(R_\theta(\theta^{-1}\eta(\mathord\cdot)),R_\theta(\theta^{-2}\eta(\mathord\cdot)),\ldots\right). η ↦ ( R θ ( θ − 1 η ( ⋅ )) , R θ ( θ − 2 η ( ⋅ )) , … ) .
Since R θ ( x η ( λ ) ) ∈ F q R_\theta(x\eta(\lambda))\in\mathbb{F}_q R θ ( x η ( λ )) ∈ F q and Λ \Lambda Λ is a finitely generated lattice over A A A in C ∞ C_\infty C ∞ , the sum
∑ λ ∈ Λ ϕ ′ R θ ( x η ( λ ) ) λ , η ∈ Hom A ( Λ ϕ , A ) , x ∈ k ∞ , \sum_{\lambda\in\Lambda_\phi}'\frac{R_\theta(x\eta(\lambda))}{\lambda},\qquad\eta\in\operatorname{Hom}_A(\Lambda_\phi,A),\ x\in k_\infty, λ ∈ Λ ϕ ∑ ′ λ R θ ( x η ( λ )) , η ∈ Hom A ( Λ ϕ , A ) , x ∈ k ∞ ,
converges in C ∞ C_\infty C ∞ . Now if η = s 1 η 1 + ⋯ + s r η r \eta=s_1\eta_1+\cdots+s_r\eta_r η = s 1 η 1 + ⋯ + s r η r with s 1 , … , s r ∈ A s_1,\ldots,s_r\in A s 1 , … , s r ∈ A , then for j ≥ 0 j\geq0 j ≥ 0 ,
− ∑ λ ∈ Λ ϕ ′ R θ ( θ − j − 1 η ( λ ) ) λ = − ∑ i = 1 r ∑ ( a 1 , … , a r ) ∈ A r ′ R t ( t − j − 1 s i ( t ) a i ( t ) ) a 1 π 1 + ⋯ + a r π r = ∑ i = 1 r R t ( t − j − 1 s i ( t ) ζ i ( t ) ) (Corollary 4.10) = R t ( t − j − 1 ζ η ( t ) ) (Equation (5.10)) = e η ∗ ( 1 / θ j + 1 ) (Equation (5.12)). \begin{aligned}
-\sum_{\lambda\in\Lambda_\phi}'\frac{R_\theta(\theta^{-j-1}\eta(\lambda))}{\lambda}
=-\sum_{i=1}^{r}\sum_{(a_1,\ldots,a_r)\in A^r}'\frac{R_t(t^{-j-1}s_i(t)a_i(t))}{a_1\pi_1+\cdots+a_r\pi_r} \\
&=\sum_{i=1}^{r}R_t\left(t^{-j-1}s_i(t)\zeta_i(t)\right)
\qquad\text{(Corollary 4.10)} \\
&=R_t\left(t^{-j-1}\zeta_\eta(t)\right)
\qquad\text{(Equation (5.10))} \\
&=e_\eta^*(1/\theta^{j+1})
\qquad\text{(Equation (5.12)).}
\end{aligned} − λ ∈ Λ ϕ ∑ ′ λ R θ ( θ − j − 1 η ( λ )) = − i = 1 ∑ r ( a 1 , … , a r ) ∈ A r ∑ ′ a 1 π 1 + ⋯ + a r π r R t ( t − j − 1 s i ( t ) a i ( t )) = i = 1 ∑ r R t ( t − j − 1 s i ( t ) ζ i ( t ) ) (Corollary 4.10) = R t ( t − j − 1 ζ η ( t ) ) (Equation (5.10)) = e η ∗ ( 1/ θ j + 1 ) (Equation (5.12)).
The result then follows by extending F q \mathbb{F}_q F q -linearly and continuously to k ∞ k_\infty k ∞ . □
5.15. Determining ker e η ∗ \ker e_\eta^* ker e η ∗ . We recall results of Poonen [40 Thm. 10] and Ferraro [17 Prop. 5.18] . As mentioned in §2.10, we know from [40 Prop. 20] that ϕ t o r s ∗ \phi^*_{\mathrm{tors}} ϕ tors ∗ is a dense subset of the compact space ker exp ϕ ∗ \ker\exp_{\phi}^* ker exp ϕ ∗ . By [17 Lem. 5.3] , for β ∈ ker exp ϕ ∗ \beta\in\ker\exp_{\phi}^* β ∈ ker exp ϕ ∗ there is a unique h β ∈ C ∞ [ [ τ ] ] h_\beta\in C_\infty[[\tau]] h β ∈ C ∞ [[ τ ]] that converges on all of C ∞ C_\infty C ∞ such that ( 1 − τ ) h β = β exp ϕ (1-\tau)h_\beta=\beta\exp_\phi ( 1 − τ ) h β = β exp ϕ . Poonen [40 Thm. 10] then proved (see also [17 Thm. 5.5] ) that the map
β ↦ h β ∣ Λ ϕ : ker exp ϕ ∗ → ∼ Λ ^ ϕ (48) \beta\mapsto h_\beta|_{\Lambda_\phi}:\ker\exp_\phi^*\xrightarrow{\sim}\widehat{\Lambda}_\phi \tag*{(48)} β ↦ h β ∣ Λ ϕ : ker exp ϕ ∗ ∼ Λ ϕ (48)
is an A A A -linear homeomorphism. Furthermore, Ferraro [17 Prop. 5.18] proved that
β = − ∑ λ ∈ Λ ϕ ′ h β ( λ ) λ , ∀ β ∈ ker exp ϕ ∗ . (49) \beta=-\sum_{\lambda\in\Lambda_\phi}'\frac{h_\beta(\lambda)}{\lambda},\qquad\forall\ \beta\in\ker\exp_\phi^*. \tag*{(49)} β = − λ ∈ Λ ϕ ∑ ′ λ h β ( λ ) , ∀ β ∈ ker exp ϕ ∗ . (49)
Fixing η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) , for x ∈ k ∞ x\in k_\infty x ∈ k ∞ the F q \mathbb{F}_q F q -linear map
γ x = ( λ ↦ R θ ( x η ( λ ) ) : Λ → F q ) \gamma_x=\left(\lambda\mapsto R_\theta(x\eta(\lambda)):\Lambda\to\mathbb{F}_q\right) γ x = ( λ ↦ R θ ( x η ( λ )) : Λ → F q )
is an element of Λ ^ ϕ \widehat{\Lambda}_\phi Λ ϕ . Thus by (5.16) there is a unique β x ∈ ker exp ϕ ∗ \beta_x\in\ker\exp_\phi^* β x ∈ ker exp ϕ ∗ so that γ x = h β x ∣ Λ ϕ \gamma_x=h_{\beta_x}|_{\Lambda_\phi} γ x = h β x ∣ Λ ϕ . Then by Lemma 5.14 and (5.17), e η ∗ ( x ) = β x e_\eta^*(x)=\beta_x e η ∗ ( x ) = β x . Therefore,
ker e η ∗ = { x ∈ k ∞ ∣ R θ ( x η ( λ ) ) = 0 , ∀ λ ∈ Λ ϕ } . \ker e_\eta^*=\{x\in k_\infty\mid R_\theta(x\eta(\lambda))=0,\ \forall\ \lambda\in\Lambda_\phi\}. ker e η ∗ = { x ∈ k ∞ ∣ R θ ( x η ( λ )) = 0 , ∀ λ ∈ Λ ϕ } .
Now let us assume that η ≠ 0 \eta\ne0 η = 0 . In this case im η = b A ⊆ A \operatorname{im}\eta=bA\subseteq A im η = b A ⊆ A for some b ∈ A b\in A b ∈ A , b ≠ 0 b\ne0 b = 0 . Then
x ∈ ker e η ∗ ⟺ R θ ( x b a ) = 0 , ∀ a ∈ A . x\in\ker e_\eta^*\quad\Longleftrightarrow\quad R_\theta(xba)=0,\qquad\forall\ a\in A. x ∈ ker e η ∗ ⟺ R θ ( x ba ) = 0 , ∀ a ∈ A .
We claim this implies that
ker e η ∗ = b − 1 A . (50) \ker e_\eta^*=b^{-1}A. \tag*{(50)} ker e η ∗ = b − 1 A . (50)
Indeed, b − 1 A ⊆ ker e η ∗ b^{-1}A\subseteq\ker e_\eta^* b − 1 A ⊆ ker e η ∗ is clear. On the other hand, if b x ∉ A bx\notin A b x ∈ / A , for some x ∈ k ∞ x\in k_\infty x ∈ k ∞ , then b x = a + ∑ j ≥ 1 c j θ − j bx=a+\sum_{j\ge1}c_j\theta^{-j} b x = a + ∑ j ≥ 1 c j θ − j for some a ∈ A a\in A a ∈ A and with c j 0 ≠ 0 c_{j_0}\ne0 c j 0 = 0 for some j 0 ≥ 1 j_0\ge1 j 0 ≥ 1 . However, then R θ ( b x θ j 0 − 1 ) = c j 0 ≠ 0 R_\theta(bx\theta^{j_0-1})=c_{j_0}\ne0 R θ ( b x θ j 0 − 1 ) = c j 0 = 0 , whence b x ∉ ker e η ∗ bx\notin\ker e_\eta^* b x ∈ / ker e η ∗ .
Lemma 5.19. For an A A A -basis π 1 , … , π r ∈ Λ ϕ \pi_1,\ldots,\pi_r\in\Lambda_\phi π 1 , … , π r ∈ Λ ϕ and corresponding dual basis η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) \eta_1,\ldots,\eta_r\in\operatorname{Hom}_A(\Lambda_\phi,A) η 1 , … , η r ∈ Hom A ( Λ ϕ , A ) , the following hold.
(a) For each i i i , ker e η i ∗ = A \ker e_{\eta_i}^*=A ker e η i ∗ = A .
(b) For b ∈ A b\in A b ∈ A , b ≠ 0 b\ne0 b = 0 ,
ϕ ∗ [ b ] = ⨁ i = 1 r e η i ∗ ( b − 1 A ) . \phi^*[b]=\bigoplus_{i=1}^{r}e_{\eta_i}^*(b^{-1}A). ϕ ∗ [ b ] = i = 1 ⨁ r e η i ∗ ( b − 1 A ) .
Proof. Part (a) follows from (5.18), since η i ( Λ ) = A \eta_i(\Lambda)=A η i ( Λ ) = A for each i i i . For (b), the case that deg b = 0 \deg b=0 deg b = 0 is clear, so we then assume deg b ≥ 1 \deg b\ge1 deg b ≥ 1 . We first note that e η i ∗ ( b − 1 A ) ≅ A / b A e_{\eta_i}^*(b^{-1}A)\cong A/bA e η i ∗ ( b − 1 A ) ≅ A / b A as A A A -modules by part (a). Second, we let N = ∑ i = 1 r e η i ∗ ( b − 1 A ) N=\sum_{i=1}^{r}e_{\eta_i}^*(b^{-1}A) N = ∑ i = 1 r e η i ∗ ( b − 1 A ) , and we show this sum is direct. Suppose that e η 1 ∗ ( b − 1 s 1 ) + ⋯ + e η r ∗ ( b − 1 s r ) = 0 e_{\eta_1}^*(b^{-1}s_1)+\cdots+e_{\eta_r}^*(b^{-1}s_r)=0 e η 1 ∗ ( b − 1 s 1 ) + ⋯ + e η r ∗ ( b − 1 s r ) = 0 for some s 1 , … , s r ∈ A s_1,\ldots,s_r\in A s 1 , … , s r ∈ A with gcd ( s 1 , … , s r ) = s \gcd(s_1,\ldots,s_r)=s g cd( s 1 , … , s r ) = s . Therefore, if we let η = s 1 η 1 + ⋯ + s r η r \eta=s_1\eta_1+\cdots+s_r\eta_r η = s 1 η 1 + ⋯ + s r η r , then
0 = ϕ s 1 ∗ e η 1 ∗ ( 1 / b ) + ⋯ + ϕ s r ∗ e η r ∗ ( 1 / b ) = e η ∗ ( 1 / b ) . 0=\phi_{s_1}^*e_{\eta_1}^*(1/b)+\cdots+\phi_{s_r}^*e_{\eta_r}^*(1/b)=e_\eta^*(1/b). 0 = ϕ s 1 ∗ e η 1 ∗ ( 1/ b ) + ⋯ + ϕ s r ∗ e η r ∗ ( 1/ b ) = e η ∗ ( 1/ b ) .
This implies that b − 1 A ⊆ ker e η ∗ b^{-1}A\subseteq\ker e_\eta^* b − 1 A ⊆ ker e η ∗ , whereas by (5.18), ker e η ∗ = s − 1 A \ker e_\eta^*=s^{-1}A ker e η ∗ = s − 1 A . Hence b ∣ s i b\mid s_i b ∣ s i for each i i i , and so e η i ∗ ( b − 1 s i ) = 0 e_{\eta_i}^*(b^{-1}s_i)=0 e η i ∗ ( b − 1 s i ) = 0 for each i i i , as desired. □
Gauss-Thakur sums and dual Gauss-Thakur sums
We review definitions of Gauss-Thakur sums of Gazda and Maurischat [19 ] for Drinfeld modules, which extend Thakur’s original constructions [42 , 46 , 47 Ch. 4] . At the same time we define their dual counterparts, built from torsion on the adjoint Drinfeld module.
6.1. Evaluative characters. We fix a monic irreducible polynomial ℘ ∈ A \wp\in A ℘ ∈ A of degree d d d , and we let F ℘ : = A / ℘ \mathbb{F}_\wp:=A/\wp F ℘ := A / ℘ . A multiplicative character χ : F ℘ → F ‾ q × \chi:\mathbb{F}_\wp\to\overline{\mathbb{F}}_q^\times χ : F ℘ → F q × extends in the usual way to a Dirichlet character modulo p \mathfrak{p} p , χ : A → F ‾ q \chi: A \to\overline{\mathbb{F}}_q χ : A → F q , where χ ( a ) = 0 \chi(a) = 0 χ ( a ) = 0 if p ∣ a \mathfrak{p} \mid a p ∣ a . By selecting ε ∈ F q d ⊆ F ‾ q \varepsilon\in\mathbb{F}_{q^d} \subseteq\overline{\mathbb{F}}_q ε ∈ F q d ⊆ F q that is a root of p \mathfrak{p} p , we define the evaluative character ,
χ ε : = ( a ↦ a ( ε ) ) : A → F q d , \chi_\varepsilon:= \left(a \mapsto a(\varepsilon)\right) : A \to\mathbb{F}_{q^d}, χ ε := ( a ↦ a ( ε ) ) : A → F q d ,
and we find that F ^ p × = ⟨ χ ε ⟩ \widehat{\mathbb{F}}_{\mathfrak{p}}^\times= \langle\chi_\varepsilon\rangle F p × = ⟨ χ ε ⟩ . If ε \varepsilon ε has been fixed, then to ease notation, for k ∈ Z k \in\mathbb{Z} k ∈ Z we will sometimes write χ k : = χ ε q k \chi_k := \chi_{\varepsilon q^k} χ k := χ ε q k .
Gauss-Thakur sums
Let ϕ \phi ϕ be a Drinfeld module of rank r r r as in (1.2). For a Dirichlet character χ : A → F q d \chi: A \to\mathbb{F}_{q^d} χ : A → F q d modulo p \mathfrak{p} p , A A A -module homomorphisms ψ : A → ϕ [ p ] \psi: A \to\phi[\mathfrak{p}] ψ : A → ϕ [ p ] and ψ ∗ : A → ϕ ∗ [ p ] \psi^* : A \to\phi^*[\mathfrak{p}] ψ ∗ : A → ϕ ∗ [ p ] , and a ∈ A a \in A a ∈ A , the Gauss-Thakur sum of [19 ] and its dual sum are
G ( χ , ψ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 ⊗ ψ ( a z ) ∈ F q d ⊗ F q ϕ ( C ∞ ) , G(\chi,\psi;a) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}\chi(z)^{-1}\otimes\psi(az) \in\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi(C_\infty), G ( χ , ψ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 ⊗ ψ ( a z ) ∈ F q d ⊗ F q ϕ ( C ∞ ) ,
G ∗ ( χ , ψ ∗ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 ⊗ ψ ∗ ( a z ) ∈ F q d ⊗ F q ϕ ∗ ( C ∞ ) . G^*(\chi,\psi^*;a) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}\chi(z)^{-1}\otimes\psi^*(az) \in\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi^*(C_\infty). G ∗ ( χ , ψ ∗ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 ⊗ ψ ∗ ( a z ) ∈ F q d ⊗ F q ϕ ∗ ( C ∞ ) .
We set G ( χ , ψ ) : = G ( χ , ψ ; 1 ) G(\chi,\psi) := G(\chi,\psi;1) G ( χ , ψ ) := G ( χ , ψ ; 1 ) and G ∗ ( χ , ψ ∗ ) : = G ∗ ( χ , ψ ∗ ; 1 ) G^*(\chi,\psi^*) := G^*(\chi,\psi^*;1) G ∗ ( χ , ψ ∗ ) := G ∗ ( χ , ψ ∗ ; 1 ) , and we have for a , b ∈ A a,b \in A a , b ∈ A ,
G ( χ , ψ ; a b ) = G ( χ , ϕ a ∘ ψ ; b ) = ( χ ( a ) ⊗ 1 ) G ( χ , ψ ; b ) , G ∗ ( χ , ψ ∗ ; a b ) = G ∗ ( χ , ϕ a ∗ ∘ ψ ∗ ; b ) = ( χ ( a ) ⊗ 1 ) G ∗ ( χ , ψ ∗ ; b ) . (51) \begin{aligned}
G(\chi,\psi;ab) &= G(\chi,\phi_a\circ\psi;b) = (\chi(a)\otimes1)G(\chi,\psi;b),\\
G^*(\chi,\psi^*;ab) &= G^*(\chi,\phi_a^*\circ\psi^*;b) = (\chi(a)\otimes1)G^*(\chi,\psi^*;b). \tag*{(51)}
\end{aligned} G ( χ , ψ ; ab ) G ∗ ( χ , ψ ∗ ; ab ) = G ( χ , ϕ a ∘ ψ ; b ) = ( χ ( a ) ⊗ 1 ) G ( χ , ψ ; b ) , = G ∗ ( χ , ϕ a ∗ ∘ ψ ∗ ; b ) = ( χ ( a ) ⊗ 1 ) G ∗ ( χ , ψ ∗ ; b ) . (51)
Composing with the product map F q d ⊗ F q C ∞ → C ∞ \mathbb{F}_{q^d}\otimes_{\mathbb{F}_q} C_\infty\to C_\infty F q d ⊗ F q C ∞ → C ∞ , we also have the tensorless sums (as named in [19 Rem. 4.3] ),
g ( χ , ψ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 ψ ( a z ) , g ∗ ( χ , ψ ∗ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 ψ ∗ ( a z ) , g(\chi,\psi;a) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}\chi(z)^{-1}\psi(az),\qquad g^*(\chi,\psi^*;a) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}\chi(z)^{-1}\psi^*(az), g ( χ , ψ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 ψ ( a z ) , g ∗ ( χ , ψ ∗ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 ψ ∗ ( a z ) ,
which take values in K ‾ \overline{K} K . We also set g ( χ , ψ ) : = g ( χ , ψ ; 1 ) g(\chi,\psi) := g(\chi,\psi;1) g ( χ , ψ ) := g ( χ , ψ ; 1 ) and g ∗ ( χ , ψ ∗ ) = g ∗ ( χ , ψ ∗ ; 1 ) g^*(\chi,\psi^*) = g^*(\chi,\psi^*;1) g ∗ ( χ , ψ ∗ ) = g ∗ ( χ , ψ ∗ ; 1 ) , which for a , b ∈ A a,b \in A a , b ∈ A satisfy,
g ( χ , ψ ; a b ) = g ( χ , ϕ a ∘ ψ ; b ) = χ ( a ) g ( χ , ψ ; b ) , g ∗ ( χ , ψ ∗ ; a b ) = g ∗ ( χ , ϕ a ∗ ∘ ψ ∗ ; b ) = χ ( a ) g ∗ ( χ , ψ ∗ ; b ) . (52) \begin{aligned}
g(\chi,\psi;ab) &= g(\chi,\phi_a\circ\psi;b) = \chi(a)g(\chi,\psi;b),\\
g^*(\chi,\psi^*;ab) &= g^*(\chi,\phi_a^*\circ\psi^*;b) = \chi(a)g^*(\chi,\psi^*;b). \tag*{(52)}
\end{aligned} g ( χ , ψ ; ab ) g ∗ ( χ , ψ ∗ ; ab ) = g ( χ , ϕ a ∘ ψ ; b ) = χ ( a ) g ( χ , ψ ; b ) , = g ∗ ( χ , ϕ a ∗ ∘ ψ ∗ ; b ) = χ ( a ) g ∗ ( χ , ψ ∗ ; b ) . (52)
As in [19 §4] , we further have the F q d \mathbb{F}_{q^d} F q d -linear vector spaces of special elements,
G ( ϕ , χ ) : = { G ∈ F q d ⊗ F q ϕ ( C ∞ ) ∣ ( 1 ⊗ a ) G = ( χ ( a ) ⊗ 1 ) G , ∀ a ∈ A } , G ∗ ( ϕ ∗ , χ ) : = { G ∗ ∈ F q d ⊗ F q ϕ ∗ ( C ∞ ) ∣ ( 1 ⊗ a ) G ∗ = ( χ ( a ) ⊗ 1 ) G ∗ , ∀ a ∈ A } . (53) \begin{aligned}
\mathcal{G}(\phi,\chi) &:= \{G\in\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi(C_\infty)\mid(1\otimes a)G=(\chi(a)\otimes1)G,\ \forall a\in A\},\\
\mathcal{G}^*(\phi^*,\chi) &:= \{G^*\in\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi^*(C_\infty)\mid(1\otimes a)G^*=(\chi(a)\otimes1)G^*,\ \forall a\in A\}. \tag*{(53)}
\end{aligned} G ( ϕ , χ ) G ∗ ( ϕ ∗ , χ ) := { G ∈ F q d ⊗ F q ϕ ( C ∞ ) ∣ ( 1 ⊗ a ) G = ( χ ( a ) ⊗ 1 ) G , ∀ a ∈ A } , := { G ∗ ∈ F q d ⊗ F q ϕ ∗ ( C ∞ ) ∣ ( 1 ⊗ a ) G ∗ = ( χ ( a ) ⊗ 1 ) G ∗ , ∀ a ∈ A } . (53)
We see in Lemma 6.6 that G ( ϕ , χ ) \mathcal{G}(\phi,\chi) G ( ϕ , χ ) and G ∗ ( ϕ ∗ , χ ) \mathcal{G}^*(\phi^*,\chi) G ∗ ( ϕ ∗ , χ ) are really spaces of Gauss-Thakur sums and dual Gauss-Thakur sums.
As in [19 Ex. 4.4] , all additive characters ψ : A → ϕ [ p ] \psi: A \to\phi[\mathfrak{p}] ψ : A → ϕ [ p ] , ψ ∗ : A → ϕ ∗ [ p ] \psi^* : A \to\phi^*[\mathfrak{p}] ψ ∗ : A → ϕ ∗ [ p ] , have the form
ψ γ : = ( a ↦ ϕ a ( γ ) ) : A → ϕ [ p ] , γ ∈ ϕ [ p ] ; ψ δ ∗ : = ( a ↦ ϕ a ∗ ( δ ) ) : A → ϕ ∗ [ p ] , δ ∈ ϕ ∗ [ p ] . \psi_\gamma:= \left(a\mapsto\phi_a(\gamma)\right) : A\to\phi[\mathfrak{p}],\ \gamma\in\phi[\mathfrak{p}];\qquad\psi_\delta^* := \left(a\mapsto\phi_a^*(\delta)\right) : A\to\phi^*[\mathfrak{p}],\ \delta\in\phi^*[\mathfrak{p}]. ψ γ := ( a ↦ ϕ a ( γ ) ) : A → ϕ [ p ] , γ ∈ ϕ [ p ] ; ψ δ ∗ := ( a ↦ ϕ a ∗ ( δ ) ) : A → ϕ ∗ [ p ] , δ ∈ ϕ ∗ [ p ] .
Also for a multiplicative character χ ∈ F ^ p × \chi\in\widehat{\mathbb{F}}_{\mathfrak{p}}^\times χ ∈ F p × , define F q d \mathbb{F}_{q^d} F q d -linear maps, η χ : F q d ⊗ F q ϕ [ p ] → F q d ⊗ F q ϕ [ p ] \eta_\chi: \mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi[\mathfrak{p}] \to\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi[\mathfrak{p}] η χ : F q d ⊗ F q ϕ [ p ] → F q d ⊗ F q ϕ [ p ] and η χ ∗ : F q d ⊗ F q ϕ ∗ [ p ] → F q d ⊗ F q ϕ ∗ [ p ] \eta_\chi^* : \mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi^*[\mathfrak{p}] \to\mathbb{F}_{q^d}\otimes_{\mathbb{F}_q}\phi^*[\mathfrak{p}] η χ ∗ : F q d ⊗ F q ϕ ∗ [ p ] → F q d ⊗ F q ϕ ∗ [ p ] by
η χ ( α ) : = − ∑ z ∈ F p × ( χ ( z ) − 1 ⊗ z ) α , η χ ∗ ( β ) : = − ∑ z ∈ F p × ( χ ( z ) − 1 ⊗ z ) β . \eta_\chi(\alpha) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}(\chi(z)^{-1}\otimes z)\alpha,\qquad\eta_\chi^*(\beta) := -\sum_{z\in\mathbb{F}_{\mathfrak{p}}^\times}(\chi(z)^{-1}\otimes z)\beta. η χ ( α ) := − z ∈ F p × ∑ ( χ ( z ) − 1 ⊗ z ) α , η χ ∗ ( β ) := − z ∈ F p × ∑ ( χ ( z ) − 1 ⊗ z ) β .
We then have the following lemma of Gazda and Maurischat, with added dual Gauss-Thakur sum variant. The proof for dual Gauss-Thakur sums is exactly the same with minor changes, so we omit it.
Lemma 6.6 (Gazda-Maurischat [19 Lem. 4.5] ). Fix a multiplicative character χ ∈ F ^ p × \chi\in\widehat{\mathbb{F}}_{\mathfrak{p}}^{\times} χ ∈ F p × . With notation as above the following hold.
(a) The map η χ \eta_{\chi} η χ is an F q d \mathbb{F}_{q}^{d} F q d -linear projection onto G ( ϕ , χ ) \mathcal{G}(\phi,\chi) G ( ϕ , χ ) , and η χ ∗ \eta_{\chi}^{*} η χ ∗ is an F q d \mathbb{F}_{q}^{d} F q d -linear projection onto G ∗ ( ϕ ∗ , χ ) \mathcal{G}^{*}(\phi^{*},\chi) G ∗ ( ϕ ∗ , χ ) .
(b) For γ ∈ ϕ [ p ] \gamma\in\phi[\mathfrak{p}] γ ∈ ϕ [ p ] , δ ∈ ϕ ∗ [ p ] \delta\in\phi^{*}[\mathfrak{p}] δ ∈ ϕ ∗ [ p ] , we have η χ ( 1 ⊗ γ ) = G ( χ , ψ γ ) \eta_{\chi}(1 \otimes\gamma) = G(\chi,\psi_{\gamma}) η χ ( 1 ⊗ γ ) = G ( χ , ψ γ ) or η χ ∗ ( 1 ⊗ δ ) = G ∗ ( χ , ψ δ ∗ ) \eta_{\chi}^{*}(1 \otimes\delta) = G^{*}(\chi,\psi_{\delta}^{*}) η χ ∗ ( 1 ⊗ δ ) = G ∗ ( χ , ψ δ ∗ ) .
(c) Every Gauss-Thakur sum G ( χ , ψ ) G(\chi,\psi) G ( χ , ψ ) is in G ( ϕ , χ ) \mathcal{G}(\phi,\chi) G ( ϕ , χ ) , and G ( ϕ , χ ) \mathcal{G}(\phi,\chi) G ( ϕ , χ ) is generated as an F q d \mathbb{F}_{q}^{d} F q d -vector space by the Gauss-Thakur sums.
(d) Likewise, every dual Gauss-Thakur sum G ∗ ( χ , ψ ∗ ) G^{*}(\chi,\psi^{*}) G ∗ ( χ , ψ ∗ ) is in G ∗ ( ϕ ∗ , χ ) \mathcal{G}^{*}(\phi^{*},\chi) G ∗ ( ϕ ∗ , χ ) , and G ∗ ( ϕ ∗ , χ ) \mathcal{G}^{*}(\phi^{*},\chi) G ∗ ( ϕ ∗ , χ ) is generated as an F q d \mathbb{F}_{q}^{d} F q d -vector space by the dual Gauss-Thakur sums.
We also have the following vanishing lemma for non-evaluative characters, whose proof for dual Gauss-Thakur sums is identical to the one in [19 ] .
Proposition 6.7 (Gazda-Maurischat [19 Prop. 4.8(1)] ). Fix a multiplicative character χ ∈ F ^ p × \chi\in\widehat{\mathbb{F}}_{\mathfrak{p}}^{\times} χ ∈ F p × . If χ \chi χ is not evaluative, then G ( ϕ , χ ) = { 0 } \mathcal{G}(\phi,\chi)=\{0\} G ( ϕ , χ ) = { 0 } and G ∗ ( ϕ ∗ , χ ) = { 0 } \mathcal{G}^{*}(\phi^{*},\chi)=\{0\} G ∗ ( ϕ ∗ , χ ) = { 0 } ; in particular, for additive characters ψ : A → ϕ [ p ] \psi:A\to\phi[\mathfrak{p}] ψ : A → ϕ [ p ] or ψ ∗ : A → ϕ ∗ [ p ] \psi^{*}:A\to\phi^{*}[\mathfrak{p}] ψ ∗ : A → ϕ ∗ [ p ] , we have
G ( χ , ψ ) = G ∗ ( χ , ψ ∗ ) = 0 , g ( χ , ψ ) = g ∗ ( χ , ψ ∗ ) = 0. G(\chi,\psi)=G^{*}(\chi,\psi^{*})=0,\qquad g(\chi,\psi)=g^{*}(\chi,\psi^{*})=0. G ( χ , ψ ) = G ∗ ( χ , ψ ∗ ) = 0 , g ( χ , ψ ) = g ∗ ( χ , ψ ∗ ) = 0.
Gauss-Thakur sums with respect to a basis
As in §§3–5, we choose an A A A -basis π 1 , … , π r \pi_{1},\ldots,\pi_{r} π 1 , … , π r of Λ ϕ \Lambda_{\phi} Λ ϕ . Then the maps ψ i : A → ϕ [ p ] \psi_{i}:A\to\phi[\mathfrak{p}] ψ i : A → ϕ [ p ] , ψ i ∗ : A → ϕ ∗ [ p ] \psi_{i}^{*}:A\to\phi^{*}[\mathfrak{p}] ψ i ∗ : A → ϕ ∗ [ p ] defined by
ψ i ( z ) : = e i ( z / p ) , ψ i ∗ ( z ) : = e i ∗ ( z / p ) , 1 ≤ i ≤ r , \psi_{i}(z):=e_{i}(z/\mathfrak{p}),\qquad\psi_{i}^{*}(z):=e_{i}^{*}(z/\mathfrak{p}),\qquad1\leq i\leq r, ψ i ( z ) := e i ( z / p ) , ψ i ∗ ( z ) := e i ∗ ( z / p ) , 1 ≤ i ≤ r ,
are A A A -module homomorphisms each with kernel p A \mathfrak{p}A p A (for the latter by Lemma 5.19(a)). If λ = a 1 π 1 + ⋯ + a r π r \lambda=a_{1}\pi_{1}+\cdots+a_{r}\pi_{r} λ = a 1 π 1 + ⋯ + a r π r or η = a 1 η 1 + ⋯ + a r η r \eta=a_{1}\eta_{1}+\cdots+a_{r}\eta_{r} η = a 1 η 1 + ⋯ + a r η r , for a 1 , … , a r ∈ A a_{1},\ldots,a_{r}\in A a 1 , … , a r ∈ A , we similarly have ψ λ : A → ϕ [ p ] \psi_{\lambda}:A\to\phi[\mathfrak{p}] ψ λ : A → ϕ [ p ] and ψ η ∗ : A → ϕ ∗ [ p ] \psi_{\eta}^{*}:A\to\phi^{*}[\mathfrak{p}] ψ η ∗ : A → ϕ ∗ [ p ] defined by
ψ λ ( z ) : = e λ ( z / p ) = ∑ i = 1 r ψ i ( a i z ) , ψ η ∗ ( z ) : = e η ∗ ( z / p ) = ∑ i = 1 r ψ i ∗ ( a i z ) . \psi_{\lambda}(z):=e_{\lambda}(z/\mathfrak{p})=\sum_{i=1}^{r}\psi_{i}(a_{i}z),\qquad\psi_{\eta}^{*}(z):=e_{\eta}^{*}(z/\mathfrak{p})=\sum_{i=1}^{r}\psi_{i}^{*}(a_{i}z). ψ λ ( z ) := e λ ( z / p ) = i = 1 ∑ r ψ i ( a i z ) , ψ η ∗ ( z ) := e η ∗ ( z / p ) = i = 1 ∑ r ψ i ∗ ( a i z ) .
We note that ψ λ \psi_{\lambda} ψ λ depends on the congruence class of λ \lambda λ modulo p Λ ϕ \mathfrak{p}\Lambda_{\phi} p Λ ϕ , and likewise ψ η ∗ \psi_{\eta}^{*} ψ η ∗ depends on the class of η \eta η modulo p Hom A ( Λ ϕ , A ) \mathfrak{p}\operatorname{Hom}_{A}(\Lambda_{\phi},A) p Hom A ( Λ ϕ , A ) . Moreover, by (5.6) and Lemma 5.19,
Hom A ( A , ϕ [ p ] ) = Span A ( ψ 1 , … , ψ r ) , Hom A ( A , ϕ ∗ [ p ] ) = Span A ( ψ 1 ∗ , … , ψ r ∗ ) , \operatorname{Hom}_{A}(A,\phi[\mathfrak{p}])=\operatorname{Span}_{A}(\psi_{1},\ldots,\psi_{r}),\qquad\operatorname{Hom}_{A}(A,\phi^{*}[\mathfrak{p}])=\operatorname{Span}_{A}(\psi_{1}^{*},\ldots,\psi_{r}^{*}), Hom A ( A , ϕ [ p ]) = Span A ( ψ 1 , … , ψ r ) , Hom A ( A , ϕ ∗ [ p ]) = Span A ( ψ 1 ∗ , … , ψ r ∗ ) ,
and so every ψ ∈ Hom A ( A , ϕ [ p ] ) \psi\in\operatorname{Hom}_{A}(A,\phi[\mathfrak{p}]) ψ ∈ Hom A ( A , ϕ [ p ]) (respectively, ψ ∗ ∈ Hom A ( A , ϕ ∗ [ p ] ) \psi^{*}\in\operatorname{Hom}_{A}(A,\phi^{*}[\mathfrak{p}]) ψ ∗ ∈ Hom A ( A , ϕ ∗ [ p ]) ) is of the form ψ λ \psi_{\lambda} ψ λ for some λ ∈ Λ ϕ \lambda\in\Lambda_{\phi} λ ∈ Λ ϕ (respectively, ψ η ∗ \psi_{\eta}^{*} ψ η ∗ for some η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_{A}(\Lambda_{\phi},A) η ∈ Hom A ( Λ ϕ , A ) ).
Thus, as a matter of notation, we define the Gauss-Thakur sums for λ ∈ Λ ϕ \lambda\in\Lambda_{\phi} λ ∈ Λ ϕ and a ∈ A a\in A a ∈ A ,
G λ ( χ ; a ) : = G ( χ , ψ λ ; a ) = − ∑ z ∈ F p × χ ( z ) − 1 ⊗ e λ ( a z / p ) , G λ ( χ ) : = G ( χ , ψ λ ; 1 ) , (54) G_{\lambda}(\chi;a):=G(\chi,\psi_{\lambda};a)=-\sum_{z\in\mathbb{F}_{\mathfrak{p}}^{\times}}\chi(z)^{-1}\otimes e_{\lambda}(az/\mathfrak{p}),\qquad G_{\lambda}(\chi):=G(\chi,\psi_{\lambda};1), \tag*{(54)} G λ ( χ ; a ) := G ( χ , ψ λ ; a ) = − z ∈ F p × ∑ χ ( z ) − 1 ⊗ e λ ( a z / p ) , G λ ( χ ) := G ( χ , ψ λ ; 1 ) , (54)
g λ ( χ ; a ) : = g ( χ , ψ λ ; a ) = − ∑ z ∈ F p × χ ( z ) − 1 e λ ( a z / p ) , g λ ( χ ) : = g ( χ , ψ λ ; 1 ) . g_{\lambda}(\chi;a):=g(\chi,\psi_{\lambda};a)=-\sum_{z\in\mathbb{F}_{\mathfrak{p}}^{\times}}\chi(z)^{-1}e_{\lambda}(az/\mathfrak{p}),\qquad g_{\lambda}(\chi):=g(\chi,\psi_{\lambda};1). g λ ( χ ; a ) := g ( χ , ψ λ ; a ) = − z ∈ F p × ∑ χ ( z ) − 1 e λ ( a z / p ) , g λ ( χ ) := g ( χ , ψ λ ; 1 ) .
We will let G i ( χ ; a ) : = G π i ( χ ; a ) G_{i}(\chi;a):=G_{\pi_{i}}(\chi;a) G i ( χ ; a ) := G π i ( χ ; a ) and g i ( χ ; a ) : = g π i ( χ ; a ) g_{i}(\chi;a):=g_{\pi_{i}}(\chi;a) g i ( χ ; a ) := g π i ( χ ; a ) for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r . Necessarily we find for λ = a 1 π 1 + ⋯ + a r π r ∈ Λ ϕ \lambda=a_{1}\pi_{1}+\cdots+a_{r}\pi_{r}\in\Lambda_{\phi} λ = a 1 π 1 + ⋯ + a r π r ∈ Λ ϕ , G λ ( χ ) = ∑ i = 1 r G i ( χ ; a i ) G_{\lambda}(\chi)=\sum_{i=1}^{r}G_{i}(\chi;a_{i}) G λ ( χ ) = ∑ i = 1 r G i ( χ ; a i ) and g λ ( χ ) = ∑ i = 1 r g i ( χ ; a i ) g_{\lambda}(\chi)=\sum_{i=1}^{r}g_{i}(\chi;a_{i}) g λ ( χ ) = ∑ i = 1 r g i ( χ ; a i ) . By Lemma 6.6(c), G ( ϕ , χ ) \mathcal{G}(\phi,\chi) G ( ϕ , χ ) is generated over F q d \mathbb{F}_{q}^{d} F q d by { G λ ( χ ) : λ ∈ Λ ϕ } \{G_{\lambda}(\chi):\lambda\in\Lambda_{\phi}\} { G λ ( χ ) : λ ∈ Λ ϕ } .
We also define dual Gauss-Thakur sums for η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_{A}(\Lambda_{\phi},A) η ∈ Hom A ( Λ ϕ , A ) and a ∈ A a\in A a ∈ A ,
G η ∗ ( χ ; a ) : = G ∗ ( χ , ψ η ∗ ; a ) = − ∑ z ∈ F p × χ ( z ) − 1 ⊗ e η ∗ ( a z / p ) , G η ∗ ( χ ) : = G ∗ ( χ , ψ η ∗ ; 1 ) , (55) G_{\eta}^{*}(\chi;a):=G^{*}(\chi,\psi_{\eta}^{*};a)=-\sum_{z\in\mathbb{F}_{\mathfrak{p}}^{\times}}\chi(z)^{-1}\otimes e_{\eta}^{*}(az/\mathfrak{p}),\qquad G_{\eta}^{*}(\chi):=G^{*}(\chi,\psi_{\eta}^{*};1), \tag*{(55)} G η ∗ ( χ ; a ) := G ∗ ( χ , ψ η ∗ ; a ) = − z ∈ F p × ∑ χ ( z ) − 1 ⊗ e η ∗ ( a z / p ) , G η ∗ ( χ ) := G ∗ ( χ , ψ η ∗ ; 1 ) , (55)
g η ∗ ( χ ; a ) : = g ∗ ( χ , ψ η ∗ ; a ) = − ∑ z ∈ F q ∗ χ ( z ) − 1 e η ∗ ( a z / p ) , g η ∗ ( χ ) : = g ∗ ( χ , ψ η ∗ ) . g_{\eta}^{*}(\chi;a) := g^{*}(\chi,\psi_{\eta}^{*};a) = -\sum_{z\in\mathbb{F}_{q}^{*}}\chi(z)^{-1}e_{\eta}^{*}(az/\mathfrak{p}), \qquad g_{\eta}^{*}(\chi) := g^{*}(\chi,\psi_{\eta}^{*}). g η ∗ ( χ ; a ) := g ∗ ( χ , ψ η ∗ ; a ) = − z ∈ F q ∗ ∑ χ ( z ) − 1 e η ∗ ( a z / p ) , g η ∗ ( χ ) := g ∗ ( χ , ψ η ∗ ) .
As above, we let G i ∗ ( χ ; a ) : = G η i ∗ ( χ ; a ) G_{i}^{*}(\chi;a) := G_{\eta_{i}}^{*}(\chi;a) G i ∗ ( χ ; a ) := G η i ∗ ( χ ; a ) and g i ∗ ( χ ; a ) : = g η i ∗ ( χ ; a ) g_{i}^{*}(\chi;a) := g_{\eta_{i}}^{*}(\chi;a) g i ∗ ( χ ; a ) := g η i ∗ ( χ ; a ) for 1 ≤ i ≤ r 1 \leq i \leq r 1 ≤ i ≤ r , and for η = s 1 η 1 + ⋯ + s r η r ∈ Hom ( Λ φ , A ) \eta=s_{1}\eta_{1}+\cdots+s_{r}\eta_{r}\in\operatorname{Hom}(\Lambda_{\varphi},A) η = s 1 η 1 + ⋯ + s r η r ∈ Hom ( Λ φ , A ) , G η ∗ ( χ ) = ∑ i = 1 r G i ∗ ( χ ; s i ) G_{\eta}^{*}(\chi)=\sum_{i=1}^{r}G_{i}^{*}(\chi;s_{i}) G η ∗ ( χ ) = ∑ i = 1 r G i ∗ ( χ ; s i ) and g η ∗ ( χ ) = ∑ i = 1 r g i ∗ ( χ ; s i ) g_{\eta}^{*}(\chi)=\sum_{i=1}^{r}g_{i}^{*}(\chi;s_{i}) g η ∗ ( χ ) = ∑ i = 1 r g i ∗ ( χ ; s i ) . By Lemma 6.6(d), G ∗ ( φ ∗ , χ ) G^{*}(\varphi^{*},\chi) G ∗ ( φ ∗ , χ ) is generated over F q d \mathbb{F}_{q^{d}} F q d by { G η ∗ ( χ ) : η ∈ Hom A ( Λ φ , A ) } \{G_{\eta}^{*}(\chi):\eta\in\operatorname{Hom}_{A}(\Lambda_{\varphi},A)\} { G η ∗ ( χ ) : η ∈ Hom A ( Λ φ , A )} .
(Dual) Gauss-Thakur sums and evaluations of (dual) special functions
We now investigate analogues of one of Gazda and Maurischat’s main results [19 Thm. 5.2] in the context of dual Gauss-Thakur sums. Such results go back to Anglès and Pellarin [5 Thm. 2.9] , who proved identities between Thakur’s Gauss sums for the Carlitz module and evaluations of the Anderson-Thakur special function (see § 8 ).
For ε ∈ F q d \varepsilon\in\mathbb{F}_{q^{d}} ε ∈ F q d a fixed root of ℘ ( t ) = 0 \wp(t)=0 ℘ ( t ) = 0 we have the evaluative character χ ε : A → F q d \chi_{\varepsilon}:A\to\mathbb{F}_{q^{d}} χ ε : A → F q d as in §6.1. We also define χ ε : C ∞ ⊗ ^ A → F q d ⊗ F q C ∞ \chi_{\varepsilon}:C_{\infty}\widehat{\otimes}A\to\mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}C_{\infty} χ ε : C ∞ ⊗ A → F q d ⊗ F q C ∞ by the composition
χ ε : C ∞ ⊗ ^ A → 1 ⊗ χ ε C ∞ ⊗ F q F q d → ∼ F q d ⊗ F q C ∞ . \chi_{\varepsilon}: C_{\infty}\widehat{\otimes} A \xrightarrow{1\otimes\chi_{\varepsilon}} C_{\infty}\otimes_{\mathbb{F}_{q}}\mathbb{F}_{q^{d}} \xrightarrow{\sim} \mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}C_{\infty}. χ ε : C ∞ ⊗ A 1 ⊗ χ ε C ∞ ⊗ F q F q d ∼ F q d ⊗ F q C ∞ .
where the latter map is the canonical isomorphism that swaps the tensor product. If we identify T \mathbb{T} T with C ∞ ⊗ ^ A C_{\infty}\widehat{\otimes} A C ∞ ⊗ A as in (36) , then multiplication composed with χ ε \chi_{\varepsilon} χ ε ,
f ↦ f ( ε ) : T → χ ε F q d ⊗ F q C ∞ → C ∞ , (56) f\mapsto f(\varepsilon):\mathbb{T}\xrightarrow{\chi_{\varepsilon}}\mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}C_{\infty}\to C_{\infty}, \tag*{(56)} f ↦ f ( ε ) : T χ ε F q d ⊗ F q C ∞ → C ∞ , (56)
is simply evaluation at t = ε t=\varepsilon t = ε . For a ∈ A a\in A a ∈ A , let a ′ ( θ ) a'(\theta) a ′ ( θ ) be its derivative with respect to θ \theta θ .
Theorem 6.13. Fix an evaluative character χ ε : A → F q d \chi_{\varepsilon}:A\to\mathbb{F}_{q^{d}} χ ε : A → F q d modulo ℘ \wp ℘ as in §6.1.
(a) (Gazda-Maurischat [19 Thm. 5.2] ) For λ ∈ Λ φ \lambda\in\Lambda_{\varphi} λ ∈ Λ φ ,
χ ε ( ω λ ( t ) ) = G λ ( χ ε ; ℘ ′ ( θ ) ) and ω λ ( ε ) = g λ ( χ ε ; ℘ ′ ( θ ) ) . \chi_{\varepsilon}(\omega_{\lambda}(t))=G_{\lambda}(\chi_{\varepsilon};\wp'(\theta)) \quad\text{and}\quad\omega_{\lambda}(\varepsilon)=g_{\lambda}(\chi_{\varepsilon};\wp'(\theta)). χ ε ( ω λ ( t )) = G λ ( χ ε ; ℘ ′ ( θ )) and ω λ ( ε ) = g λ ( χ ε ; ℘ ′ ( θ )) . (b) For η ∈ Hom A ( Λ φ , A ) \eta\in\operatorname{Hom}_{A}(\Lambda_{\varphi},A) η ∈ Hom A ( Λ φ , A ) ,
χ ε ( ζ η ( t ) ) = G η ∗ ( χ ε ; ℘ ′ ( θ ) ) and ζ η ( ε ) = g η ∗ ( χ ε ; ℘ ′ ( θ ) ) . \chi_{\varepsilon}(\zeta_{\eta}(t))=G_{\eta}^{*}(\chi_{\varepsilon};\wp'(\theta)) \quad\text{and}\quad\zeta_{\eta}(\varepsilon)=g_{\eta}^{*}(\chi_{\varepsilon};\wp'(\theta)). χ ε ( ζ η ( t )) = G η ∗ ( χ ε ; ℘ ′ ( θ )) and ζ η ( ε ) = g η ∗ ( χ ε ; ℘ ′ ( θ )) .
Proof. In principle the proof of (b) can follow similar lines to [19 Thm. 5.2] , though since we are restricting to the case A = F q [ θ ] A=\mathbb{F}_{q}[\theta] A = F q [ θ ] , rather than the case of general A A A as in [19 ] , we instead appeal to identities of Anglès and Pellarin [4 , 5 , 36 ] to obtain a more direct proof.
We first note that the second equality in (b) follows from the first. Next we consider
1 ⊗ e η ∗ : F q d ⊗ F q k ∞ ⟶ F q d ⊗ F q φ ∗ ( C ∞ ) , 1\otimes e_{\eta}^{*}:\mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}k_{\infty}\longrightarrow\mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}\varphi^{*}(C_{\infty}), 1 ⊗ e η ∗ : F q d ⊗ F q k ∞ ⟶ F q d ⊗ F q φ ∗ ( C ∞ ) ,
which is F q d ⊗ F q A \mathbb{F}_{q^{d}}\otimes_{\mathbb{F}_{q}}A F q d ⊗ F q A -linear and continuous with respect to the sup-norm. Then we have
χ ε ( ζ η ( t ) ) = ( 1 ⊗ e η ∗ ) ( ∑ m = 0 ∞ ε m ⊗ θ − m − 1 ) = ( 1 ⊗ e η ∗ ) ( 1 1 ⊗ θ − ε ⊗ 1 ) . (57) \chi_{\varepsilon}(\zeta_{\eta}(t))=(1\otimes e_{\eta}^{*})\left(\sum_{m=0}^{\infty}\varepsilon^{m}\otimes\theta^{-m-1}\right)=(1\otimes e_{\eta}^{*})\left(\frac{1}{1\otimes\theta-\varepsilon\otimes1}\right). \tag*{(57)} χ ε ( ζ η ( t )) = ( 1 ⊗ e η ∗ ) ( m = 0 ∑ ∞ ε m ⊗ θ − m − 1 ) = ( 1 ⊗ e η ∗ ) ( 1 ⊗ θ − ε ⊗ 1 1 ) . (57)
On the other hand by (51) and the definition of χ ε \chi_{\varepsilon} χ ε ,
G η ∗ ( χ ε ; ℘ ′ ( θ ) ) = − ( ℘ ′ ( ε ) ⊗ 1 ) ∑ z ∈ F q ∗ z ( ε ) − 1 ⊗ e η ∗ ( z ( θ ) / ℘ ( θ ) ) = − ( 1 ⊗ e η ∗ ) ( ( ℘ ′ ( ε ) ⊗ ℘ ( θ ) − 1 ) ∑ z ∈ A deg z < d z ( ε ) − 1 ⊗ z ( θ ) ) . (58) \begin{aligned}
G_{\eta}^{*}(\chi_{\varepsilon};\wp'(\theta))=-(\wp'(\varepsilon)\otimes1)\sum_{z\in\mathbb{F}_{q}^{*}}\!z(\varepsilon)^{-1}\otimes e_{\eta}^{*}(z(\theta)/\wp(\theta)) \\
&=-(1\otimes e_{\eta}^{*})\left(\left(\wp'(\varepsilon)\otimes\wp(\theta)^{-1}\right)\sum_{\substack{z\in A\\ \deg z<d}}z(\varepsilon)^{-1}\otimes z(\theta)\right). \tag*{(58)}
\end{aligned} G η ∗ ( χ ε ; ℘ ′ ( θ )) = − ( ℘ ′ ( ε ) ⊗ 1 ) z ∈ F q ∗ ∑ z ( ε ) − 1 ⊗ e η ∗ ( z ( θ ) / ℘ ( θ )) = − ( 1 ⊗ e η ∗ ) ( ℘ ′ ( ε ) ⊗ ℘ ( θ ) − 1 ) z ∈ A d e g z < d ∑ z ( ε ) − 1 ⊗ z ( θ ) . (58)
Now for independent variables x x x , y y y over F q \mathbb{F}_q F q , it follows from [4 Prop. 10] that, for k ≥ 0 k \ge0 k ≥ 0 , we have in F q ( x ) [ y ] = F q ( x ) ⊗ F q F q [ y ] \mathbb{F}_q(x)[y]=\mathbb{F}_q(x)\otimes_{\mathbb{F}_q}\mathbb{F}_q[y] F q ( x ) [ y ] = F q ( x ) ⊗ F q F q [ y ] ,
∑ z ∈ A deg z = k z ( x ) − 1 ⊗ z ( y ) = − ( 1 ⊗ y − x ⊗ 1 ) ( 1 ⊗ y − x q ⊗ 1 ) ⋯ ( 1 ⊗ y − x q k − 1 ⊗ 1 ) ( x − x q ) ( x − x q 2 ) ⋯ ( x − x q k ) ⊗ 1 . (59) \sum_{\substack{z\in A\\ \deg z=k}} z(x)^{-1}\otimes z(y)=-\frac{(1\otimes y-x\otimes1)(1\otimes y-x^q\otimes1)\cdots(1\otimes y-x^{q^{k-1}}\otimes1)}{(x-x^q)(x-x^{q^2})\cdots(x-x^{q^k})\otimes1}. \tag*{(59)} z ∈ A d e g z = k ∑ z ( x ) − 1 ⊗ z ( y ) = − ( x − x q ) ( x − x q 2 ) ⋯ ( x − x q k ) ⊗ 1 ( 1 ⊗ y − x ⊗ 1 ) ( 1 ⊗ y − x q ⊗ 1 ) ⋯ ( 1 ⊗ y − x q k − 1 ⊗ 1 ) . (59)
Moreover, an induction implies that
∑ z ∈ A deg z ≤ k ′ z ( x ) − 1 ⊗ z ( y ) = − ( 1 ⊗ y − x q ⊗ 1 ) ( 1 ⊗ y − x q 2 ⊗ 1 ) ⋯ ( 1 ⊗ y − x q k ⊗ 1 ) ( x − x q ) ( x − x q 2 ) ⋯ ( x − x q k ) ⊗ 1 . \sum_{\substack{z\in A\\ \deg z\le k}}' z(x)^{-1}\otimes z(y)=-\frac{(1\otimes y-x^q\otimes1)(1\otimes y-x^{q^2}\otimes1)\cdots(1\otimes y-x^{q^k}\otimes1)}{(x-x^q)(x-x^{q^2})\cdots(x-x^{q^k})\otimes1}. z ∈ A d e g z ≤ k ∑ ′ z ( x ) − 1 ⊗ z ( y ) = − ( x − x q ) ( x − x q 2 ) ⋯ ( x − x q k ) ⊗ 1 ( 1 ⊗ y − x q ⊗ 1 ) ( 1 ⊗ y − x q 2 ⊗ 1 ) ⋯ ( 1 ⊗ y − x q k ⊗ 1 ) .
(The reader will also observe that these are partial sums and partial products in the specialization of Pellarin’s L L L -series at s = 1 s=1 s = 1 in [36 Thm. 1] .) In our setting, this implies
∑ z ∈ A deg z < d ′ z ( ε ) − 1 ⊗ z ( θ ) = − ( 1 ⊗ θ − ε q ⊗ 1 ) ( 1 ⊗ θ − ε q 2 ⊗ 1 ) ⋯ ( 1 ⊗ θ − ε q d − 1 ⊗ 1 ) ( ε − ε q ) ( ε − ε q 2 ) ⋯ ( ε − ε q d − 1 ) ⊗ 1 = − 1 ⊗ ℘ ( θ ) ( 1 ⊗ θ − ε ⊗ 1 ) ( ℘ ′ ( ε ) ⊗ 1 ) . \begin{aligned}
\sum_{\substack{z\in A\\ \deg z<d}}' z(\varepsilon)^{-1}\otimes z(\theta)=-\frac{(1\otimes\theta-\varepsilon^q\otimes1)(1\otimes\theta-\varepsilon^{q^2}\otimes1)\cdots(1\otimes\theta-\varepsilon^{q^{d-1}}\otimes1)}{(\varepsilon-\varepsilon^q)(\varepsilon-\varepsilon^{q^2})\cdots(\varepsilon-\varepsilon^{q^{d-1}})\otimes1} \\
&=-\frac{1\otimes\wp(\theta)}{(1\otimes\theta-\varepsilon\otimes1)(\wp'(\varepsilon)\otimes1)}.
\end{aligned} z ∈ A d e g z < d ∑ ′ z ( ε ) − 1 ⊗ z ( θ ) = − ( ε − ε q ) ( ε − ε q 2 ) ⋯ ( ε − ε q d − 1 ) ⊗ 1 ( 1 ⊗ θ − ε q ⊗ 1 ) ( 1 ⊗ θ − ε q 2 ⊗ 1 ) ⋯ ( 1 ⊗ θ − ε q d − 1 ⊗ 1 ) = − ( 1 ⊗ θ − ε ⊗ 1 ) ( ℘ ′ ( ε ) ⊗ 1 ) 1 ⊗ ℘ ( θ ) .
Combining this with (57) –(58) , we see that
G η ∗ ( χ ε ; ℘ ′ ( θ ) ) = ( 1 ⊗ e η ∗ ) ( 1 1 ⊗ θ − ε ⊗ 1 ) = χ ε ( ζ η ( t ) ) . G_\eta^*(\chi_\varepsilon;\wp'(\theta))=(1\otimes e_\eta^*)\left(\frac{1}{1\otimes\theta-\varepsilon\otimes1}\right)=\chi_\varepsilon(\zeta_\eta(t)). G η ∗ ( χ ε ; ℘ ′ ( θ )) = ( 1 ⊗ e η ∗ ) ( 1 ⊗ θ − ε ⊗ 1 1 ) = χ ε ( ζ η ( t )) .
Twists of (dual) Gauss-Thakur sums
For λ ∈ Λ ϕ \lambda\in\Lambda_\phi λ ∈ Λ ϕ , η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) , k ∈ Z k\in\mathbb{Z} k ∈ Z , and a ∈ A a\in A a ∈ A , let
g λ [ k ] ( χ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 e λ ( a z / ℘ ) q k , g η ∗ [ k ] ( χ ; a ) : = − ∑ z ∈ F p × χ ( z ) − 1 e η ∗ ( a z / ℘ ) q k . g_\lambda^{[k]}(\chi;a):=-\sum_{z\in\mathbb{F}_\mathfrak{p}^{\times}}\chi(z)^{-1}e_\lambda(az/\wp)^{q^k},\qquad g_\eta^{*[k]}(\chi;a):=-\sum_{z\in\mathbb{F}_\mathfrak{p}^{\times}}\chi(z)^{-1}e_\eta^*(az/\wp)^{q^k}. g λ [ k ] ( χ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 e λ ( a z / ℘ ) q k , g η ∗ [ k ] ( χ ; a ) := − z ∈ F p × ∑ χ ( z ) − 1 e η ∗ ( a z / ℘ ) q k .
This definition can be made for the tensored version of Gauss-Thakur sums as well, but we do not need this generality. The following corollary shows that twists of Gauss-Thakur sums satisfy a similar special function property to that of Gauss-Thakur sums in (52) .
Corollary 6.18. Fix a multiplicative character χ ∈ F ^ p × \chi\in\widehat{\mathbb{F}}_\mathfrak{p}^{\times} χ ∈ F p × , k ∈ Z k\in\mathbb{Z} k ∈ Z , and a ∈ A a\in A a ∈ A . For λ ∈ Λ ϕ \lambda\in\Lambda_\phi λ ∈ Λ ϕ and η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_A(\Lambda_\phi,A) η ∈ Hom A ( Λ ϕ , A ) , the following hold.
(a) If χ \chi χ is not evaluative as in §6.1, then g λ [ k ] ( χ ; a ) = 0 g_\lambda^{[k]}(\chi;a)=0 g λ [ k ] ( χ ; a ) = 0 and g η ∗ [ k ] ( χ ; a ) = 0 g_\eta^{*[k]}(\chi;a)=0 g η ∗ [ k ] ( χ ; a ) = 0 .
(b) We have g λ [ k ] ( χ ; a ) = χ ( a ) g λ [ k ] ( χ ; 1 ) g_\lambda^{[k]}(\chi;a)=\chi(a)g_\lambda^{[k]}(\chi;1) g λ [ k ] ( χ ; a ) = χ ( a ) g λ [ k ] ( χ ; 1 ) and g η ∗ [ k ] ( χ ; a ) = χ ( a ) g η ∗ [ k ] ( χ ; 1 ) g_\eta^{*[k]}(\chi;a)=\chi(a)g_\eta^{*[k]}(\chi;1) g η ∗ [ k ] ( χ ; a ) = χ ( a ) g η ∗ [ k ] ( χ ; 1 ) .
Proof. We write only the dual case, as the argument is the same for both cases. Then
g η ∗ [ k ] ( χ ; a ) = − ∑ z ∈ F p × χ ( z ) − 1 e η ∗ ( a z / ℘ ) q k = ( − ∑ z ∈ F p × χ ( z ) − q − k e η ∗ ( a z / ℘ ) ) q k = g η ∗ ( χ q − k ; a ) q k = ( χ q − k ( a ) g η ∗ ( χ q − k ; 1 ) ) q k = χ ( a ) g η ∗ [ k ] ( χ ; 1 ) , \begin{aligned}
g_\eta^{*[k]}(\chi;a)=-\sum_{z\in\mathbb{F}_\mathfrak{p}^{\times}}\chi(z)^{-1}e_\eta^*(az/\wp)^{q^k}=\left(-\sum_{z\in\mathbb{F}_\mathfrak{p}^{\times}}\chi(z)^{-q^{-k}}e_\eta^*(az/\wp)\right)^{q^k} \\
&=g_\eta^*(\chi^{q^{-k}};a)^{q^k}=\left(\chi^{q^{-k}}(a)g_\eta^*(\chi^{q^{-k}};1)\right)^{q^k}=\chi(a)g_\eta^{*[k]}(\chi;1),
\end{aligned} g η ∗ [ k ] ( χ ; a ) = − z ∈ F p × ∑ χ ( z ) − 1 e η ∗ ( a z / ℘ ) q k = − z ∈ F p × ∑ χ ( z ) − q − k e η ∗ ( a z / ℘ ) q k = g η ∗ ( χ q − k ; a ) q k = ( χ q − k ( a ) g η ∗ ( χ q − k ; 1 ) ) q k = χ ( a ) g η ∗ [ k ] ( χ ; 1 ) ,
where the fourth equality follows from (52) and the last by essentially undoing the first and second. This proves (b). For (a), if χ \chi χ is not evaluative, then neither is χ q − k \chi^{q^{-k}} χ q − k , in which case the above calculation and Proposition 6.7 imply the desired vanishing. □
Corollary 6.19. Fix an evaluative character χ ε : A → F q d \chi_{\varepsilon}: A \to\mathbb{F}_{q^{d}} χ ε : A → F q d modulo ℘ \wp ℘ as in §6.1. Then for k ∈ Z k \in\mathbb{Z} k ∈ Z the following hold. For λ ∈ Λ ϕ \lambda\in\Lambda_{\phi} λ ∈ Λ ϕ and η ∈ Hom A ( Λ ϕ , A ) \eta\in\operatorname{Hom}_{A}(\Lambda_{\phi}, A) η ∈ Hom A ( Λ ϕ , A ) ,
ω λ ( k ) ( ε ) = g λ [ k ] ( χ ε ; ℘ ′ ( θ ) ) , ζ η ( k ) ( ε ) = g η ∗ [ k ] ( χ ε ; ℘ ′ ( θ ) ) . \omega_{\lambda}^{(k)}(\varepsilon)=g_{\lambda}^{[k]}(\chi_{\varepsilon};\wp'(\theta)),\qquad\zeta_{\eta}^{(k)}(\varepsilon)=g_{\eta}^{*[k]}(\chi_{\varepsilon};\wp'(\theta)). ω λ ( k ) ( ε ) = g λ [ k ] ( χ ε ; ℘ ′ ( θ )) , ζ η ( k ) ( ε ) = g η ∗ [ k ] ( χ ε ; ℘ ′ ( θ )) .
Proof. Once again, the proof is similar for both cases so we proceed with only the dual case. Working as in the proof of the previous corollary,
g η ∗ [ k ] ( χ ε ; ℘ ′ ( θ ) ) = g η ∗ ( χ ε q − k ; ℘ ′ ( θ ) ) q k = ( ζ η ( ε q − k ) ) q k = ζ η ( k ) ( ε ) . g_{\eta}^{*[k]}(\chi_{\varepsilon};\wp'(\theta))=g_{\eta}^{*}(\chi_{\varepsilon q^{-k}};\wp'(\theta))^{q^{k}}=\left(\zeta_{\eta}(\varepsilon^{q^{-k}})\right)^{q^{k}}=\zeta_{\eta}^{(k)}(\varepsilon). g η ∗ [ k ] ( χ ε ; ℘ ′ ( θ )) = g η ∗ ( χ ε q − k ; ℘ ′ ( θ ) ) q k = ( ζ η ( ε q − k ) ) q k = ζ η ( k ) ( ε ) .
Duality of Gauss-Thakur sums
As one might expect the duality results of Ferraro [17 ] in Theorems 4.3 and 4.7 induce a duality between Gauss-Thakur sums and dual Gauss-Thakur sums. This is also reflected in the interconnections between rigid analytic trivializations for the t t t -motive and t t t -comotive of a Drinfeld module, as in Lemma 3.13 and Proposition 3.16 . We explore this duality in detail in this section, and our first result is the following.
Theorem 7.1. Fix ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + irreducible, deg ℘ = d \deg\wp=d deg ℘ = d , and let ε ∈ F q d ⊆ C ∞ \varepsilon\in\mathbb{F}_{q^{d}}\subseteq\mathbb{C}_{\infty} ε ∈ F q d ⊆ C ∞ be a root of ℘ \wp ℘ . For 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r the following hold.
(a) We have
∑ i = 1 r g i ∗ ( χ ε ) g i ( χ ε ) = 1 ℘ ′ ( ε ) 2 ( θ − ε ) . \sum_{i=1}^{r}g_{i}^{*}(\chi_{\varepsilon})g_{i}(\chi_{\varepsilon})=\frac{1}{\wp'(\varepsilon)^{2}(\theta-\varepsilon)}. i = 1 ∑ r g i ∗ ( χ ε ) g i ( χ ε ) = ℘ ′ ( ε ) 2 ( θ − ε ) 1 .
(b) For 0 ≤ j , k ≤ r − 1 0\leq j,k\leq r-1 0 ≤ j , k ≤ r − 1 with 1 ≤ j + k ≤ r − 1 1\leq j+k\leq r-1 1 ≤ j + k ≤ r − 1 ,
∑ i = 1 r g i ∗ [ − j ] ( χ ε ) g i [ k ] ( χ ε ) = 0. \sum_{i=1}^{r}g_{i}^{*[-j]}(\chi_{\varepsilon})g_{i}^{[k]}(\chi_{\varepsilon})=0. i = 1 ∑ r g i ∗ [ − j ] ( χ ε ) g i [ k ] ( χ ε ) = 0.
Proof. Let ω ϕ = ( ω 1 , … , ω r ) T \boldsymbol{\omega}_{\phi}=(\omega_{1},\ldots,\omega_{r})^{\mathsf{T}} ω ϕ = ( ω 1 , … , ω r ) T be the first column of Υ \Upsilon Υ as in (25) and ζ ϕ = ( ζ 1 , … , ζ r ) \boldsymbol{\zeta}_{\phi}=(\zeta_{1},\ldots,\zeta_{r}) ζ ϕ = ( ζ 1 , … , ζ r ) be the first row of Ψ \Psi Ψ as in (30) . By Theorem 6.13 and Corollaries 6.18 and 6.19,
ζ i ( j ) ( ε ) = g i ∗ [ j ] ( χ ε ; ℘ ′ ( θ ) ) = ℘ ′ ( ε ) g i ∗ [ j ] ( χ ε ) , ω i ( k ) ( ε ) = g i [ k ] ( χ ε ; ℘ ′ ( θ ) ) = ℘ ′ ( ε ) g i [ k ] ( χ ε ) , \zeta_{i}^{(j)}(\varepsilon)=g_{i}^{*[j]}(\chi_{\varepsilon};\wp'(\theta))=\wp'(\varepsilon)g_{i}^{*[j]}(\chi_{\varepsilon}),\qquad\omega_{i}^{(k)}(\varepsilon)=g_{i}^{[k]}(\chi_{\varepsilon};\wp'(\theta))=\wp'(\varepsilon)g_{i}^{[k]}(\chi_{\varepsilon}), ζ i ( j ) ( ε ) = g i ∗ [ j ] ( χ ε ; ℘ ′ ( θ )) = ℘ ′ ( ε ) g i ∗ [ j ] ( χ ε ) , ω i ( k ) ( ε ) = g i [ k ] ( χ ε ; ℘ ′ ( θ )) = ℘ ′ ( ε ) g i [ k ] ( χ ε ) ,
for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r and j , k ∈ Z j,k\in\mathbb{Z} j , k ∈ Z . Therefore, for j , k j,k j , k , with 0 ≤ j , k ≤ r − 1 0\leq j,k\leq r-1 0 ≤ j , k ≤ r − 1 , we have
∑ i = 1 r g i ∗ [ − j ] ( χ ε ) g i [ k ] ( χ ε ) = 1 ℘ ′ ( ε ) 2 ⋅ ζ ϕ ( − j ) ( ε ) ⋅ ω ϕ ( k ) ( ε ) = 1 ℘ ′ ( ε ) 2 ⋅ [ Ψ Υ ] j + 1 , k + 1 ∣ t = ε , \sum_{i=1}^{r}g_{i}^{*[-j]}(\chi_{\varepsilon})g_{i}^{[k]}(\chi_{\varepsilon})=\frac{1}{\wp'(\varepsilon)^{2}}\cdot\boldsymbol{\zeta}_{\phi}^{(-j)}(\varepsilon)\cdot\boldsymbol{\omega}_{\phi}^{(k)}(\varepsilon)=\frac{1}{\wp'(\varepsilon)^{2}}\cdot\left[\Psi\Upsilon\right]_{j+1,k+1}\big|_{t=\varepsilon}, i = 1 ∑ r g i ∗ [ − j ] ( χ ε ) g i [ k ] ( χ ε ) = ℘ ′ ( ε ) 2 1 ⋅ ζ ϕ ( − j ) ( ε ) ⋅ ω ϕ ( k ) ( ε ) = ℘ ′ ( ε ) 2 1 ⋅ [ ΨΥ ] j + 1 , k + 1 t = ε ,
and the result follows from Lemma 3.13 .
Matrices of (dual) Gauss-Thakur sums
We next prove a more uniform result in Theorem 7.7, which does not depend on the choice of ε \varepsilon ε . For χ ∈ F ^ q × \chi\in\widehat{\mathbb{F}}_{q}^{\times} χ ∈ F q × and a ∈ A a\in A a ∈ A , we recall the definitions of G ( χ ; a ) \mathcal{G}(\chi;a) G ( χ ; a ) and G ∗ ( χ ; a ) ∈ Mat r × r ( K ‾ ) \mathcal{G}^{*}(\chi;a)\in\operatorname{Mat}_{r\times r}(\overline{K}) G ∗ ( χ ; a ) ∈ Mat r × r ( K ) from (1.11)–(1.12). For reasons of space we do not repeat their definitions here. Keeping with previous conventions, we set G ( χ ) : = G ( χ ; 1 ) \mathcal{G}(\chi):=\mathcal{G}(\chi;1) G ( χ ) := G ( χ ; 1 ) and G ∗ ( χ ) : = G ∗ ( χ ; 1 ) \mathcal{G}^{*}(\chi):=\mathcal{G}^{*}(\chi;1) G ∗ ( χ ) := G ∗ ( χ ; 1 ) . We note as in §1.10, Theorem 6.13 and Corollaries 6.18 and 6.19 imply
G ( χ ε ; ℘ ′ ( θ ) ) = ℘ ′ ( ε ) G ( χ ε ) = Υ ( ε ) , G ∗ ( χ ε ; ℘ ′ ( θ ) ) = ℘ ′ ( ε ) G ∗ ( χ ε ) = J Ψ ( r − 1 ) ( ε ) . (60) \mathcal{G}(\chi_{\varepsilon};\wp'(\theta))=\wp'(\varepsilon)\mathcal{G}(\chi_{\varepsilon})=\Upsilon(\varepsilon),\qquad\mathcal{G}^{*}(\chi_{\varepsilon};\wp'(\theta))=\wp'(\varepsilon)\mathcal{G}^{*}(\chi_{\varepsilon})=J\Psi^{(r-1)}(\varepsilon). \tag*{(60)} G ( χ ε ; ℘ ′ ( θ )) = ℘ ′ ( ε ) G ( χ ε ) = Υ ( ε ) , G ∗ ( χ ε ; ℘ ′ ( θ )) = ℘ ′ ( ε ) G ∗ ( χ ε ) = J Ψ ( r − 1 ) ( ε ) . (60)
For a ∈ A a\in A a ∈ A , define Γ ( ϕ ; a ) ∈ Mat r d × N ( K ‾ ) \Gamma(\phi;a)\in\operatorname{Mat}_{rd\times N}(\overline{K}) Γ ( ϕ ; a ) ∈ Mat r d × N ( K ) and Γ ∗ ( ϕ ∗ ; a ) ∈ Mat N × r d ( K ‾ ) \Gamma^{*}(\phi^{*};a)\in\operatorname{Mat}_{N\times rd}(\overline{K}) Γ ∗ ( ϕ ∗ ; a ) ∈ Mat N × r d ( K ) as in (1.13)–(1.14). Recall also the definitions of Γ ℓ ( ϕ ; a ) \Gamma_{\ell}(\phi;a) Γ ℓ ( ϕ ; a ) and Γ ℓ ∗ ( ϕ ∗ ; a ) \Gamma_{\ell}^{*}(\phi^{*};a) Γ ℓ ∗ ( ϕ ∗ ; a ) from §1.10, and set Γ ( ϕ ) : = Γ ( ϕ ; 1 ) \Gamma(\phi):=\Gamma(\phi;1) Γ ( ϕ ) := Γ ( ϕ ; 1 ) and Γ ∗ ( ϕ ∗ ) : = Γ ∗ ( ϕ ∗ ; 1 ) \Gamma^{*}(\phi^{*}):=\Gamma^{*}(\phi^{*};1) Γ ∗ ( ϕ ∗ ) := Γ ∗ ( ϕ ∗ ; 1 ) . Further for β ∈ k \beta\in k β ∈ k that is regular at ε \varepsilon ε , recall
Q ( β ) : = diag ( β ( ε ) I r , … , β ( ε q d − 1 ) I r ) ∈ Mat r d × r d ( F q d ) , Q(\beta):=\operatorname{diag}\left(\beta(\varepsilon)I_{r},\ldots,\beta(\varepsilon^{q^{d-1}})I_{r}\right)\in\operatorname{Mat}_{rd\times rd}(\mathbb{F}_{q^{d}}), Q ( β ) := diag ( β ( ε ) I r , … , β ( ε q d − 1 ) I r ) ∈ Mat r d × r d ( F q d ) ,
@ JSImport as in (11) . For a , b ∈ A a,b \in A a , b ∈ A , we note that Γ ( ϕ ; a b ) = Q ( b ) Γ ( ϕ ; a ) \Gamma(\phi;ab)=\mathbf{Q}(b)\Gamma(\phi;a) Γ ( ϕ ; ab ) = Q ( b ) Γ ( ϕ ; a ) and Γ ∗ ( ϕ ∗ ; a b ) = Γ ∗ ( ϕ ∗ ; a ) Q ( b ) \Gamma^*(\phi^*;ab)=\Gamma^*(\phi^*;a)\mathbf{Q}(b) Γ ∗ ( ϕ ∗ ; ab ) = Γ ∗ ( ϕ ∗ ; a ) Q ( b ) by Corollary 6.18. Also, by Theorem 6.13 and Corollary 6.19,
Γ ( ϕ ; a ) = Q ( a / ℘ ′ ) Γ ( ϕ ; ℘ ′ ( θ ) ) = Q ( a / ℘ ′ ) ( Γ ( ε ) Γ ( r ) ( ε ) ⋯ ⋮ ⋮ Γ ( ε q d − 1 ) Γ ( r ) ( ε q d − 1 ) ⋯ ) (61) \Gamma(\phi;a)=\mathbf{Q}(a/\wp')\Gamma(\phi;\wp'(\theta))=\mathbf{Q}(a/\wp')\begin{pmatrix}
\Gamma(\varepsilon) & \Gamma^{(r)}(\varepsilon) & \cdots\\
\vdots& \vdots& \\
\Gamma(\varepsilon^{q^{d-1}}) & \Gamma^{(r)}(\varepsilon^{q^{d-1}}) & \cdots
\end{pmatrix} \tag*{(61)} Γ ( ϕ ; a ) = Q ( a / ℘ ′ ) Γ ( ϕ ; ℘ ′ ( θ )) = Q ( a / ℘ ′ ) Γ ( ε ) ⋮ Γ ( ε q d − 1 ) Γ ( r ) ( ε ) ⋮ Γ ( r ) ( ε q d − 1 ) ⋯ ⋯ (61)
and
Γ ∗ ( ϕ ∗ ; a ) = Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ ) ) Q ( a / ℘ ′ ) = ( J Ψ ( r − 1 ) ( ε ) ⋯ J Ψ ( r − 1 ) ( ε q d − 1 ) J Ψ ( 2 r − 1 ) ( ε ) ⋯ J Ψ ( 2 r − 1 ) ( ε q d − 1 ) ⋮ ⋮ ) Q ( a / ℘ ′ ) . (62) \Gamma^*(\phi^*;a)=\Gamma^*(\phi^*;\wp'(\theta))\mathbf{Q}(a/\wp')=
\begin{pmatrix}
\mathcal{J}\Psi^{(r-1)}(\varepsilon) & \cdots& \mathcal{J}\Psi^{(r-1)}(\varepsilon^{q^{d-1}}) \\
\mathcal{J}\Psi^{(2r-1)}(\varepsilon) & \cdots& \mathcal{J}\Psi^{(2r-1)}(\varepsilon^{q^{d-1}}) \\
\vdots& & \vdots
\end{pmatrix}\mathbf{Q}(a/\wp'). \tag*{(62)} Γ ∗ ( ϕ ∗ ; a ) = Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ )) Q ( a / ℘ ′ ) = J Ψ ( r − 1 ) ( ε ) J Ψ ( 2 r − 1 ) ( ε ) ⋮ ⋯ ⋯ J Ψ ( r − 1 ) ( ε q d − 1 ) J Ψ ( 2 r − 1 ) ( ε q d − 1 ) ⋮ Q ( a / ℘ ′ ) . (62)
Lastly, for β ∈ k \beta\in k β ∈ k recall P ( β ) ∈ Mat N × N ( K ‾ ) \mathbf{P}(\beta) \in\operatorname{Mat}_{N \times N}(\overline{K}) P ( β ) ∈ Mat N × N ( K ) as in (12) , which written more explicitly, for i , j ≥ 1 i,j \ge1 i , j ≥ 1 , satisfies
[ P ( β ) ] i j = { { β i − j } ϕ q j − 1 if i − j ≥ 0 , − { β j − i } ϕ ∗ q j − 1 if i − j < 0. (63) [\mathbf{P}(\beta)]_{ij}=
\begin{cases}
\left\{\begin{matrix}\beta\\ i-j\end{matrix}\right\}_{\phi}^{q^{j-1}} & \text{if } i-j \ge0,\\
-\left\{\begin{matrix}\beta\\ j-i\end{matrix}\right\}_{\phi^*}^{q^{j-1}} & \text{if } i-j<0.
\end{cases} \tag*{(63)} [ P ( β ) ] ij = ⎩ ⎨ ⎧ { β i − j } ϕ q j − 1 − { β j − i } ϕ ∗ q j − 1 if i − j ≥ 0 , if i − j < 0. (63)
Similarly, let P ℓ ( β ) ∈ Mat r ℓ × r ℓ ( K ‾ ) \mathbf{P}_{\ell}(\beta) \in\operatorname{Mat}_{r\ell\times r\ell}(\overline{K}) P ℓ ( β ) ∈ Mat r ℓ × r ℓ ( K ) be the truncation of P ( β ) \mathbf{P}(\beta) P ( β ) after r ℓ r\ell r ℓ rows and columns. The following theorem is the main result of this paper. It is similar to [17 Thm. 7.18] , but made more explicit in this case where A = F q [ θ ] A=\mathbb{F}_q[\theta] A = F q [ θ ] . Namely, it is a duality result between the Gauss-Thakur sums and their twists present in the Γ ( ϕ ) \Gamma(\phi) Γ ( ϕ ) matrix and the dual Gauss-Thakur sums and their twists present in the Γ ∗ ( ϕ ) \Gamma^*(\phi) Γ ∗ ( ϕ ) matrix. Now, for x = ∑ j ≥ j 0 c j θ − j ∈ k ∞ x=\sum_{j\geq j_0}c_j\theta^{-j}\in k_\infty x = ∑ j ≥ j 0 c j θ − j ∈ k ∞ , we write ⟨ x ⟩ : = ∑ j ≥ 1 c j θ − j ∈ θ − 1 F q [ [ θ − 1 ] ] \langle x\rangle:=\sum_{j\geq1}c_j\theta^{-j}\in\theta^{-1}\mathbb{F}_q[[\theta^{-1}]] ⟨ x ⟩ := ∑ j ≥ 1 c j θ − j ∈ θ − 1 F q [[ θ − 1 ]] . For a ∈ A a\in A a ∈ A , we take a m o d ℘ a\bmod\wp a mod ℘ for the remainder of a a a divided by ℘ \wp ℘ with deg ( a m o d ℘ ) < deg ℘ \deg(a\bmod\wp)<\deg\wp deg ( a mod ℘ ) < deg ℘ . Thus, ⟨ a / ℘ ⟩ = ( a m o d ℘ ) / ℘ \langle a/\wp\rangle=(a\bmod\wp)/\wp ⟨ a / ℘ ⟩ = ( a mod ℘ ) / ℘ .
Theorem 7.7. With the matrices defined as above, we have the following identities.
(a) For a , b ∈ A a,b\in A a , b ∈ A ,
Γ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ; b ) = P ( ⟨ a b / ℘ ⟩ ) . \Gamma^*(\phi^*;a)\cdot\mathbf{Q}(\wp')\cdot\Gamma(\phi;b)=\mathbf{P}(\langle ab/\wp\rangle). Γ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ; b ) = P (⟨ ab / ℘ ⟩) . (b) For a , b ∈ A a,b\in A a , b ∈ A and ℓ ∈ N \ell\in\mathbb{N} ℓ ∈ N ,
Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ ; b ) = P ℓ ( ⟨ a b / ℘ ⟩ ) . \Gamma^*_{\ell}(\phi^*;a)\cdot\mathbf{Q}(\wp')\cdot\Gamma_{\ell}(\phi;b)=\mathbf{P}_{\ell}(\langle ab/\wp\rangle). Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ ; b ) = P ℓ (⟨ ab / ℘ ⟩) . (c) In particular,
Γ ∗ ( ϕ ∗ ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ) = P ( 1 / ℘ ) . \Gamma^*(\phi^*)\cdot\mathbf{Q}(\wp')\cdot\Gamma(\phi)=\mathbf{P}(1/\wp). Γ ∗ ( ϕ ∗ ) ⋅ Q ( ℘ ′ ) ⋅ Γ ( ϕ ) = P ( 1/ ℘ ) .
Proof. Parts (b) and (c) follow directly from (a). To prove (a), we fix i , j ≥ 1 i,j\geq1 i , j ≥ 1 and note
[ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] i j = [ Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ ) ) Q ( a b / ℘ ′ ) Γ ( ϕ ; ℘ ′ ( θ ) ) ] i j . (64) \left[\Gamma^*(\phi^*;a)\mathbf{Q}(\wp')\Gamma(\phi;b)\right]_{ij}
=
\left[\Gamma^*(\phi^*;\wp'(\theta))\mathbf{Q}(ab/\wp')\Gamma(\phi;\wp'(\theta))\right]_{ij}. \tag*{(64)} [ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] ij = [ Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ )) Q ( ab / ℘ ′ ) Γ ( ϕ ; ℘ ′ ( θ )) ] ij . (64)
We note that J Ψ ( r − 1 ) = ( ζ ϕ , ζ ϕ ( 1 ) , … , ζ ϕ ( r − 1 ) ) T \mathcal{J}\Psi^{(r-1)}=(\zeta_{\phi},\zeta_{\phi}^{(1)},\ldots,\zeta_{\phi}^{(r-1)})^{\mathsf{T}} J Ψ ( r − 1 ) = ( ζ ϕ , ζ ϕ ( 1 ) , … , ζ ϕ ( r − 1 ) ) T , and so (62) implies that the i i i -th row of Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ ) ) \Gamma^*(\phi^*;\wp'(\theta)) Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ )) is ( ζ ϕ ( i − 1 ) ( ε ) , ζ ϕ ( i − 1 ) ( ε q ) , … , ζ ϕ ( i − 1 ) ( ε q d − 1 ) ) (\zeta_{\phi}^{(i-1)}(\varepsilon),\zeta_{\phi}^{(i-1)}(\varepsilon^q),\ldots,\zeta_{\phi}^{(i-1)}(\varepsilon^{q^{d-1}})) ( ζ ϕ ( i − 1 ) ( ε ) , ζ ϕ ( i − 1 ) ( ε q ) , … , ζ ϕ ( i − 1 ) ( ε q d − 1 )) . Likewise, by (61) , the j j j -th column of Γ ( ϕ ; ℘ ′ ( θ ) ) \Gamma(\phi;\wp'(\theta)) Γ ( ϕ ; ℘ ′ ( θ )) is ( ω ϕ ( j − 1 ) ( ε ) , ω ϕ ( j − 1 ) ( ε q ) , … , ω ϕ ( j − 1 ) ( ε q d − 1 ) ) T (\omega_{\phi}^{(j-1)}(\varepsilon),\omega_{\phi}^{(j-1)}(\varepsilon^q),\ldots,\omega_{\phi}^{(j-1)}(\varepsilon^{q^{d-1}}))^{\mathsf{T}} ( ω ϕ ( j − 1 ) ( ε ) , ω ϕ ( j − 1 ) ( ε q ) , … , ω ϕ ( j − 1 ) ( ε q d − 1 ) ) T . From the definition of Q ( β ) \mathbf{Q}(\beta) Q ( β ) as a diagonal matrix, it follows that
[ Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ ) ) ⋅ Q ( a b / ℘ ′ ) ⋅ Γ ( ϕ ; ℘ ′ ( θ ) ) ] i j = ∑ k = 1 d a ( ε q k − 1 ) b ( ε q k − 1 ) ℘ ′ ( ε q k − 1 ) ⋅ ζ ϕ ( i − 1 ) ( ε q k − 1 ) ⋅ ω ϕ ( j − 1 ) ( ε q k − 1 ) , (65) \begin{aligned}
\left[\Gamma^*(\phi^*;\wp'(\theta))\cdot\mathbf{Q}(ab/\wp')\cdot\Gamma(\phi;\wp'(\theta))\right]_{ij} \\
&= \sum_{k=1}^{d} \frac{a(\varepsilon^{q^{k-1}})b(\varepsilon^{q^{k-1}})}{\wp'(\varepsilon^{q^{k-1}})} \cdot\zeta_{\phi}^{(i-1)}(\varepsilon^{q^{k-1}})\cdot\omega_{\phi}^{(j-1)}(\varepsilon^{q^{k-1}}), \tag*{(65)}
\end{aligned} [ Γ ∗ ( ϕ ∗ ; ℘ ′ ( θ )) ⋅ Q ( ab / ℘ ′ ) ⋅ Γ ( ϕ ; ℘ ′ ( θ )) ] ij = k = 1 ∑ d ℘ ′ ( ε q k − 1 ) a ( ε q k − 1 ) b ( ε q k − 1 ) ⋅ ζ ϕ ( i − 1 ) ( ε q k − 1 ) ⋅ ω ϕ ( j − 1 ) ( ε q k − 1 ) , (65)
and since ζ ϕ ( i − 1 ) ω ϕ ( j − 1 ) \zeta_{\phi}^{(i-1)}\omega_{\phi}^{(j-1)} ζ ϕ ( i − 1 ) ω ϕ ( j − 1 ) is the upper-left entry of Ψ ( i − 1 ) Γ ( j − 1 ) \Psi^{(i-1)}\Gamma^{(j-1)} Ψ ( i − 1 ) Γ ( j − 1 ) ,
= ∑ k = 1 d a ( ε q k − 1 ) b ( ε q k − 1 ) ℘ ′ ( ε q k − 1 ) ⋅ [ Ψ ( i − 1 ) ( ε q k − 1 ) ⋅ Γ ( j − 1 ) ( ε q k − 1 ) ] 11 = ∑ k = 1 d a ( t ) b ( t ) m o d ℘ ( t ) ℘ ′ ( t ) ⋅ [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 ∣ t = ε q k − 1 . \begin{aligned}
= \sum_{k=1}^{d}\frac{a(\varepsilon^{q^{k-1}})b(\varepsilon^{q^{k-1}})}{\wp'(\varepsilon^{q^{k-1}})}\cdot\left[\Psi^{(i-1)}(\varepsilon^{q^{k-1}})\cdot\Gamma^{(j-1)}(\varepsilon^{q^{k-1}})\right]_{11} \\
&= \sum_{k=1}^{d}\frac{a(t)b(t)\mathbin{\mathrm{mod}}\wp(t)}{\wp'(t)}\cdot\left[\Psi^{(i-j)}\Gamma^{(j-1)}\right]_{11}\bigg|_{t=\varepsilon^{q^{k-1}}}.
\end{aligned} = k = 1 ∑ d ℘ ′ ( ε q k − 1 ) a ( ε q k − 1 ) b ( ε q k − 1 ) ⋅ [ Ψ ( i − 1 ) ( ε q k − 1 ) ⋅ Γ ( j − 1 ) ( ε q k − 1 ) ] 11 = k = 1 ∑ d ℘ ′ ( t ) a ( t ) b ( t ) mod ℘ ( t ) ⋅ [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 t = ε q k − 1 .
By their definitions in §3.4, the entries of Ψ \Psi Ψ and Γ \Gamma Γ are in T ∩ K ‾ [ t ] \mathbb{T}\cap\overline{K}[t] T ∩ K [ t ] . Thus, [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 = ∑ m ≥ 0 c m t m \left[\Psi^{(i-j)}\Gamma^{(j-1)}\right]_{11}=\sum_{m\geq0}c_mt^m [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 = ∑ m ≥ 0 c m t m with c m ∈ K c_m\in K c m ∈ K and ∣ c m ∣ ∞ → 0 \lvert c_m\rvert_{\infty}\to0 ∣ c m ∣ ∞ → 0 . Letting ρ ( t ) = a ( t ) b ( t ) m o d ℘ ( t ) ∈ F q [ t ] \rho(t)=a(t)b(t)\mathbin{\mathrm{mod}}\wp(t)\in\mathbb{F}_q[t] ρ ( t ) = a ( t ) b ( t ) mod ℘ ( t ) ∈ F q [ t ] ,
∑ k = 1 d ρ ( t ) ℘ ′ ( t ) ⋅ [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 ∣ t = ε q k − 1 = ∑ m = 0 ∞ c m ∑ k = 1 d ρ ( t ) ℘ ′ ( t ) t m ∣ t = ε q k − 1 (66) \sum_{k=1}^{d}\frac{\rho(t)}{\wp'(t)}\cdot\left[\Psi^{(i-j)}\Gamma^{(j-1)}\right]_{11}\bigg|_{t=\varepsilon^{q^{k-1}}}
=\sum_{m=0}^{\infty}c_m\sum_{k=1}^{d}\frac{\rho(t)}{\wp'(t)}t^m\bigg|_{t=\varepsilon^{q^{k-1}}} \tag*{(66)} k = 1 ∑ d ℘ ′ ( t ) ρ ( t ) ⋅ [ Ψ ( i − j ) Γ ( j − 1 ) ] 11 t = ε q k − 1 = m = 0 ∑ ∞ c m k = 1 ∑ d ℘ ′ ( t ) ρ ( t ) t m t = ε q k − 1 (66)
Now as a function on P 1 / F ‾ q \mathbb{P}^1/\overline{\mathbb{F}}_q P 1 / F q , ( ρ ( t ) / ℘ ( t ) ) t m (\rho(t)/\wp(t))t^m ( ρ ( t ) / ℘ ( t )) t m can only have poles at ∞ \infty ∞ and at the zeros of ℘ ( t ) \wp(t) ℘ ( t ) . Hence, by (5.2) and the residue theorem,
Res t ( ρ ( t ) ℘ ( t ) t m ) = − Res ∞ ( ρ ( t ) ℘ ( t ) t m d t ) = ∑ k = 1 d Res t = ε q k − 1 ( ρ ( t ) ℘ ( t ) t m d t ) . (67) \operatorname{Res}_{t}\left(\frac{\rho(t)}{\wp(t)}t^m\right)=-\operatorname{Res}_{\infty}\left(\frac{\rho(t)}{\wp(t)}t^m\,dt\right)=\sum_{k=1}^{d}\operatorname{Res}_{t=\varepsilon^{q^{k-1}}}\left(\frac{\rho(t)}{\wp(t)}t^m\,dt\right). \tag*{(67)} Res t ( ℘ ( t ) ρ ( t ) t m ) = − Res ∞ ( ℘ ( t ) ρ ( t ) t m d t ) = k = 1 ∑ d Res t = ε q k − 1 ( ℘ ( t ) ρ ( t ) t m d t ) . (67)
As ( ρ ( t ) / ℘ ( t ) ) t m (\rho(t)/\wp(t))t^m ( ρ ( t ) / ℘ ( t )) t m has at most simple poles at ε q k − 1 \varepsilon^{q^{k-1}} ε q k − 1 , 1 ≤ k ≤ d 1\leq k\leq d 1 ≤ k ≤ d , we have
Res t = ε q k − 1 ( ρ ( t ) ℘ ( t ) t m d t ) = ( t − ε q k − 1 ) ρ ( t ) ℘ ( t ) t m ∣ t = ε q k − 1 = ρ ( t ) ℘ ′ ( t ) t m ∣ t = ε q k − 1 . (68) \operatorname{Res}_{t=\varepsilon^{q^{k-1}}}\left(\frac{\rho(t)}{\wp(t)}t^m\,dt\right)=\left.\frac{(t-\varepsilon^{q^{k-1}})\rho(t)}{\wp(t)}t^m\right|_{t=\varepsilon^{q^{k-1}}}=\left.\frac{\rho(t)}{\wp'(t)}t^m\right|_{t=\varepsilon^{q^{k-1}}}. \tag*{(68)} Res t = ε q k − 1 ( ℘ ( t ) ρ ( t ) t m d t ) = ℘ ( t ) ( t − ε q k − 1 ) ρ ( t ) t m t = ε q k − 1 = ℘ ′ ( t ) ρ ( t ) t m t = ε q k − 1 . (68)
Combining (7.8)–(7.12), we have
[ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] i j = ∑ m = 0 ∞ c m R t ( ρ ( t ) t m ℘ ( t ) ) . \left[\Gamma^*(\phi^*;a)\mathsf{Q}(\wp')\Gamma(\phi;b)\right]_{ij}=\sum_{m=0}^{\infty}c_m\operatorname{R}_{t}\left(\frac{\rho(t)t^m}{\wp(t)}\right). [ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] ij = m = 0 ∑ ∞ c m R t ( ℘ ( t ) ρ ( t ) t m ) .
Note that R t ( ( ρ ( t ) / ℘ ( t ) ) t m ) ∈ F q \operatorname{R}_{t}((\rho(t)/\wp(t))t^m)\in\mathbb{F}_q R t (( ρ ( t ) / ℘ ( t )) t m ) ∈ F q is the coefficient of t − m − 1 t^{-m-1} t − m − 1 in the F q ( ( t − 1 ) ) \mathbb{F}_q((t^{-1})) F q (( t − 1 )) -expansion of ρ ( t ) / ℘ ( t ) \rho(t)/\wp(t) ρ ( t ) / ℘ ( t ) . Thus writing ρ ( t ) / ℘ ( t ) = ∑ m = δ ∞ a m t − m \rho(t)/\wp(t)=\sum_{m=\delta}^{\infty}a_mt^{-m} ρ ( t ) / ℘ ( t ) = ∑ m = δ ∞ a m t − m , with δ = − deg ( ρ ( t ) / ℘ ( t ) ) \delta=-\deg(\rho(t)/\wp(t)) δ = − deg ( ρ ( t ) / ℘ ( t )) and a m ∈ F q a_m\in\mathbb{F}_q a m ∈ F q , and noting that δ > 0 \delta>0 δ > 0 , we obtain
[ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] i j = ∑ m = δ ∞ a m c m − 1 = ∑ m = δ ∞ a m { { θ − m } ϕ q j − 1 , if i − j ≥ 0 , − { θ − m } ϕ ∗ q j − 1 , if i − j < 0. \left[\Gamma^*(\phi^*;a)\mathsf{Q}(\wp')\Gamma(\phi;b)\right]_{ij}
=\sum_{m=\delta}^{\infty}a_mc_{m-1}
=\sum_{m=\delta}^{\infty}a_m
\begin{cases}
\left\{\theta^{-m}\right\}_{\phi}^{q^{j-1}}, & \text{if }i-j\geq0,\\
-\left\{\theta^{-m}\right\}_{\phi^*}^{q^{j-1}}, & \text{if }i-j<0.
\end{cases} [ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] ij = m = δ ∑ ∞ a m c m − 1 = m = δ ∑ ∞ a m { { θ − m } ϕ q j − 1 , − { θ − m } ϕ ∗ q j − 1 , if i − j ≥ 0 , if i − j < 0.
where the second equality follows from the definition of c m c_m c m and Proposition 3.16. Because ϕ : k ∞ → C ∞ ( τ ) \phi:k_\infty\to\mathbb{C}_\infty(\tau) ϕ : k ∞ → C ∞ ( τ ) and ϕ ∗ : k ∞ → C ∞ ( τ ) \phi^*:k_\infty\to\mathbb{C}_\infty(\tau) ϕ ∗ : k ∞ → C ∞ ( τ ) have continuous coefficient functions,
[ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] i j = { { ρ ( θ ) ℘ ( θ ) } ϕ q j − 1 , if i − j ≥ 0 , − { ρ ( θ ) ℘ ( θ ) } ϕ ∗ q j − 1 , if i − j < 0 , \left[\Gamma^*(\phi^*;a)\mathsf{Q}(\wp')\Gamma(\phi;b)\right]_{ij}
=
\begin{cases}
\left\{\dfrac{\rho(\theta)}{\wp(\theta)}\right\}_{\phi}^{q^{j-1}}, & \text{if }i-j\geq0,\\
-\left\{\dfrac{\rho(\theta)}{\wp(\theta)}\right\}_{\phi^*}^{q^{j-1}}, & \text{if }i-j<0,
\end{cases} [ Γ ∗ ( ϕ ∗ ; a ) Q ( ℘ ′ ) Γ ( ϕ ; b ) ] ij = ⎩ ⎨ ⎧ { ℘ ( θ ) ρ ( θ ) } ϕ q j − 1 , − { ℘ ( θ ) ρ ( θ ) } ϕ ∗ q j − 1 , if i − j ≥ 0 , if i − j < 0 ,
as desired. ∎
Remark 7.13. Note that for − r deg ℘ < i − j < 0 -r\deg\wp< i-j < 0 − r deg ℘ < i − j < 0 , [ P ( 1 / ℘ ) ] i j = − { 1 / ℘ i − j } ϕ ∗ q i − j = 0 [\mathbf{P}(1/\wp)]_{ij}=-\left\{\begin{matrix}1/\wp\\i-j\end{matrix}\right\}_{\phi^*}^{q^{i-j}}=0 [ P ( 1/ ℘ ) ] ij = − { 1/ ℘ i − j } ϕ ∗ q i − j = 0 by conventions established in §2.14. This implies that for ℓ ≤ deg ℘ \ell\leq\deg\wp ℓ ≤ deg ℘ , and a , b ∈ A a,b\in A a , b ∈ A such that a b m o d ℘ = 1 ab\bmod\wp=1 ab mod ℘ = 1 , the matrix
Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ , b ) = P ℓ ( 1 / ℘ ) ∈ GL r ℓ ( K ‾ ) , \Gamma_\ell^*(\phi^*;a)\cdot\mathbf{Q}(\wp')\cdot\Gamma_\ell(\phi,b)=\mathbf{P}_\ell(1/\wp)\in\operatorname{GL}_{r\ell}(\overline{K}), Γ ℓ ∗ ( ϕ ∗ ; a ) ⋅ Q ( ℘ ′ ) ⋅ Γ ℓ ( ϕ , b ) = P ℓ ( 1/ ℘ ) ∈ GL r ℓ ( K ) ,
is lower triangular with non-zero diagonal entries ℘ − q i − 1 \wp^{-q^{i-1}} ℘ − q i − 1 , and thus is invertible.
Carlitz module
In this section we restrict to the case of the Carlitz module, C : A → A [ τ ] \mathrm{C}:A\to A[\tau] C : A → A [ τ ] defined by C θ = θ + τ \mathrm{C}_\theta=\theta+\tau C θ = θ + τ . In this case the Gauss-Thakur sums g ( χ ε ; a ) g(\chi_\varepsilon;a) g ( χ ε ; a ) are precisely the sums defined and investigated by Thakur in [42 ] .
8.1. Essentials of the Carlitz module. The Carlitz module is rank 1 1 1 , and its fundamental period, the Carlitz period , is
π ~ = − ( − θ ) q / ( q − 1 ) ∏ n = 1 ∞ ( 1 − θ 1 − q n ) − 1 ∈ k ∞ ( ( − θ ) 1 / ( q − 1 ) ) . (69) \widetilde{\pi}=-(-\theta)^{q/(q-1)}\prod_{n=1}^{\infty}\left(1-\theta^{1-q^n}\right)^{-1}\in k_\infty\left((- \theta)^{1/(q-1)}\right). \tag*{(69)} π = − ( − θ ) q / ( q − 1 ) n = 1 ∏ ∞ ( 1 − θ 1 − q n ) − 1 ∈ k ∞ ( ( − θ ) 1/ ( q − 1 ) ) . (69)
For more details about the Carlitz module, see [26 Ch. 3, 34 Ch. 3, 47 Ch. 2] . The matrices Υ \Upsilon Υ and Ψ \Psi Ψ are then 1 × 1 1\times1 1 × 1 , given by
Υ = ω C = ( − θ ) 1 / ( q − 1 ) ∏ n = 0 ∞ ( 1 − t θ q n ) − 1 , Ψ = − Ω C = − ( − θ ) − q / ( q − 1 ) ∏ n = 1 ∞ ( 1 − t θ q n ) , (70) \Upsilon=\omega_{\mathrm{C}}=(-\theta)^{1/(q-1)}\prod_{n=0}^{\infty}\left(1-\frac{t}{\theta^{q^n}}\right)^{-1},\qquad\Psi=-\Omega_{\mathrm{C}}=-(-\theta)^{-q/(q-1)}\prod_{n=1}^{\infty}\left(1-\frac{t}{\theta^{q^n}}\right), \tag*{(70)} Υ = ω C = ( − θ ) 1/ ( q − 1 ) n = 0 ∏ ∞ ( 1 − θ q n t ) − 1 , Ψ = − Ω C = − ( − θ ) − q / ( q − 1 ) n = 1 ∏ ∞ ( 1 − θ q n t ) , (70)
for a fixed choice of ( − θ ) 1 / ( q − 1 ) (-\theta)^{1/(q-1)} ( − θ ) 1/ ( q − 1 ) . The function ω C \omega_{\mathrm{C}} ω C is the Anderson-Thakur function defined in [3 §2.5] , and Ω C \Omega_{\mathrm{C}} Ω C was defined in [2 §3.1] . In the context of §3.2 and §5.9,
ω π ~ ( t ) = ω C ( t ) = ∑ m = 0 ∞ e π ~ ( 1 / θ m + 1 ) t m , ζ η ~ ( t ) = − Ω C ( t ) = ∑ m = 0 ∞ e η ~ ∗ ( 1 / θ m + 1 ) t m , \omega_{\widetilde{\pi}}(t)=\omega_{\mathrm{C}}(t)=\sum_{m=0}^{\infty}e_{\widetilde{\pi}}(1/\theta^{m+1})t^m,\qquad\zeta_{\widetilde{\eta}}(t)=-\Omega_{\mathrm{C}}(t)=\sum_{m=0}^{\infty}e_{\widetilde{\eta}}^*(1/\theta^{m+1})t^m, ω π ( t ) = ω C ( t ) = m = 0 ∑ ∞ e π ( 1/ θ m + 1 ) t m , ζ η ( t ) = − Ω C ( t ) = m = 0 ∑ ∞ e η ∗ ( 1/ θ m + 1 ) t m ,
where η ~ : Λ C → A \widetilde{\eta}:\Lambda_{\mathrm{C}}\to A η : Λ C → A is defined by η ~ ( a π ~ ) = a \widetilde{\eta}(a\widetilde{\pi})=a η ( a π ) = a (cf. [2 §5.1, 14 §2.5] ). We also have ω C ( 1 ) = ( t − θ ) ω C \omega_{\mathrm{C}}^{(1)}=(t-\theta)\omega_{\mathrm{C}} ω C ( 1 ) = ( t − θ ) ω C , Ω C ( − 1 ) = ( t − θ ) Ω C \Omega_{\mathrm{C}}^{(-1)}=(t-\theta)\Omega_{\mathrm{C}} Ω C ( − 1 ) = ( t − θ ) Ω C , and Ω C = 1 / ω C ( 1 ) \Omega_{\mathrm{C}}=1/\omega_{\mathrm{C}}^{(1)} Ω C = 1/ ω C ( 1 ) , which then provide directly Proposition 3.16 in this case that − Ω C ⋅ ω C = 1 / ( θ − t ) -\Omega_{\mathrm{C}}\cdot\omega_{\mathrm{C}}=1/(\theta-t) − Ω C ⋅ ω C = 1/ ( θ − t ) .
Remark 8.4. Throughout this section we let e = e π ~ \mathbf{e}=\mathbf{e}_{\widetilde{\pi}} e = e π and e ∗ = e η ~ \mathbf{e}^*=\mathbf{e}_{\widetilde{\eta}} e ∗ = e η , though as a warning to the reader, we note that this notation for e ∗ \mathbf{e}^* e ∗ is different from that in [2 §5.2] . Indeed if e 1 ∗ : k ∞ → C ∗ ( C ∞ ) \mathbf{e}_1^*:k_\infty\to\mathbb{C}^*(\mathbb{C}_\infty) e 1 ∗ : k ∞ → C ∗ ( C ∞ ) is the function defined in [2 ] , then e ∗ ( x ) = − e 1 ∗ ( x ) q \mathbf{e}^*(x)=-\mathbf{e}_1^*(x)^q e ∗ ( x ) = − e 1 ∗ ( x ) q . The negative sign arises because ζ η ~ = − Ω C \zeta_{\widetilde{\eta}}=-\Omega_{\mathrm{C}} ζ η = − Ω C , and so also e ∗ \mathbf{e}^* e ∗ differs by a sign from e ~ \widetilde{\mathbf{e}} e in [14 §3.1] .
8.5. Gauss-Thakur sums and their duals for C. Fix ℘ ∈ A + \wp\in A_+ ℘ ∈ A + , irreducible of degree d d d , and let ε ∈ F q d ⊂ C ∞ \varepsilon\in\mathbb{F}_{q^d}\subset\mathbb{C}_\infty ε ∈ F q d ⊂ C ∞ be one of its roots. As presented in §6.2, we have Gauss-Thakur sums, for χ ∈ F ^ q × \chi\in\widehat{\mathbb{F}}_q^\times χ ∈ F q × and a ∈ A a\in A a ∈ A ,
g ( χ ; a ) : = g π ~ ( χ ; a ) , g ∗ ( χ ; a ) : = g η ~ ∗ ( χ ; a ) . g(\chi;a):=g_{\widetilde{\pi}}(\chi;a),\qquad g^*(\chi;a):=g_{\widetilde{\eta}}^*(\chi;a). g ( χ ; a ) := g π ( χ ; a ) , g ∗ ( χ ; a ) := g η ∗ ( χ ; a ) .
As usual g ( χ ) = g ( χ ; 1 ) g(\chi)=g(\chi;1) g ( χ ) = g ( χ ; 1 ) and g ∗ ( χ ) = g ∗ ( χ ; 1 ) g^*(\chi)=g^*(\chi;1) g ∗ ( χ ) = g ∗ ( χ ; 1 ) . By Theorem 6.13,
℘ ′ ( ε ) g ( χ ε ) = ω C ( ε ) , ℘ ′ ( ε ) g ∗ ( χ ε ) = − Ω C ( ε ) , (71) \wp'(\varepsilon)g(\chi_\varepsilon)=\omega_{\mathrm{C}}(\varepsilon),\qquad\wp'(\varepsilon)g^*(\chi_\varepsilon)=-\Omega_{\mathrm{C}}(\varepsilon), \tag*{(71)} ℘ ′ ( ε ) g ( χ ε ) = ω C ( ε ) , ℘ ′ ( ε ) g ∗ ( χ ε ) = − Ω C ( ε ) , (71)
(the first originally due to Anglès and Pellarin [5 Thm. 2.9] ). Also by Theorem 7.1(a),
g ( χ ε ) g ∗ ( χ ε ) = 1 ( θ − ε ) ℘ ′ ( ε ) 2 . (72) g(\chi_\varepsilon)g^*(\chi_\varepsilon)=\frac{1}{(\theta-\varepsilon)\wp'(\varepsilon)^2}. \tag*{(72)} g ( χ ε ) g ∗ ( χ ε ) = ( θ − ε ) ℘ ′ ( ε ) 2 1 . (72)
from which we see in particular that g ( χ ε ) ≠ 0 g(\chi_{\varepsilon}) \ne0 g ( χ ε ) = 0 and g ∗ ( χ ε ) ≠ 0 g^{*}(\chi_{\varepsilon}) \ne0 g ∗ ( χ ε ) = 0 .
Now Thakur established several theorems for Gauss-Thakur sums in [42 ] , which parallel results for classical Gauss sums. Let K = k ( e ( 1 / ℘ ) ) K = k(e(1/\wp)) K = k ( e ( 1/ ℘ )) , the ℘ \wp ℘ -th Carlitz cyclotomic field, and let H = F q d K H = \mathbb{F}_{q^{d}}K H = F q d K . Then Gal ( H / k ) ≅ Gal ( F q d / F q ) × Gal ( K / k ) \operatorname{Gal}(H/k) \cong\operatorname{Gal}(\mathbb{F}_{q^{d}}/\mathbb{F}_{q}) \times\operatorname{Gal}(K/k) Gal ( H / k ) ≅ Gal ( F q d / F q ) × Gal ( K / k ) , where Gal ( F q d / F q ) = ⟨ σ ⟩ \operatorname{Gal}(\mathbb{F}_{q^{d}}/\mathbb{F}_{q}) = \langle\sigma\rangle Gal ( F q d / F q ) = ⟨ σ ⟩ is generated by the q q q -th power Frobenius and Gal ( K / k ) = { ρ a : a ∈ ( A / ℘ ) × } ≅ ( A / ℘ ) × \operatorname{Gal}(K/k) = \{\rho_a : a \in(A/\wp)^{\times}\} \cong(A/\wp)^{\times} Gal ( K / k ) = { ρ a : a ∈ ( A / ℘ ) × } ≅ ( A / ℘ ) × .
By [25 Thm. 3.2, 26 Thm. 1.7.11] , we have K = k ( e ( 1 / ℘ ) ) = k ( e ∗ ( 1 / ℘ ) ) K = k(e(1/\wp)) = k(e^{*}(1/\wp)) K = k ( e ( 1/ ℘ )) = k ( e ∗ ( 1/ ℘ )) , and moreover, by [26 Prop. 4.14.13, 34 Thm. 7.1.12, 40 Cor. 11] , for any a , b ∈ ( A / ℘ ) × a,b \in(A/\wp)^{\times} a , b ∈ ( A / ℘ ) × ,
ρ a ( e ( b / ℘ ) ) = e ( a b / ℘ ) = C a ( e ( b / ℘ ) ) , ρ a − 1 ( e ∗ ( b / ℘ ) ) = e ∗ ( a b / ℘ ) = C a ∗ ( e ∗ ( b / ℘ ) ) . (73) \rho_a(e(b/\wp)) = e(ab/\wp) = C_a(e(b/\wp)), \qquad\rho_a^{-1}(e^{*}(b/\wp)) = e^{*}(ab/\wp) = C_a^{*}(e^{*}(b/\wp)). \tag*{(73)} ρ a ( e ( b / ℘ )) = e ( ab / ℘ ) = C a ( e ( b / ℘ )) , ρ a − 1 ( e ∗ ( b / ℘ )) = e ∗ ( ab / ℘ ) = C a ∗ ( e ∗ ( b / ℘ )) . (73)
(see also [14 Prop. 3.1.6] ). Finally, for 0 ≤ k ≤ d − 1 0 \le k \le d-1 0 ≤ k ≤ d − 1 , let p k ⊆ O H \mathfrak{p}_k \subseteq\mathcal{O}_H p k ⊆ O H be the unique (totally ramified) prime ideal above θ − ε q k \theta-\varepsilon^{q^k} θ − ε q k in F q d [ θ ] \mathbb{F}_{q^d}[\theta] F q d [ θ ] , and normalize ord p k \operatorname{ord}_{\mathfrak{p}_k} ord p k so that ord p k ( ℘ ) = 1 \operatorname{ord}_{\mathfrak{p}_k}(\wp)=1 ord p k ( ℘ ) = 1 .
Theorem 8.9 (Thakur [42 Thms. I–IV] ). Let ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of ℘ \wp ℘ . Let K = k ( e ( 1 / ℘ ) ) K = k(e(1/\wp)) K = k ( e ( 1/ ℘ )) and H = F q d K H = \mathbb{F}_{q^d}K H = F q d K .
(a) Let 0 ≤ k ≤ d − 1 0 \le k \le d-1 0 ≤ k ≤ d − 1 . For σ ∈ Gal ( H / k ) \sigma\in\operatorname{Gal}(H/k) σ ∈ Gal ( H / k ) as above, σ ( g ( χ k ) ) = g ( χ k + 1 ) \sigma(g(\chi_k)) = g(\chi_{k+1}) σ ( g ( χ k )) = g ( χ k + 1 ) . For a ∈ ( A / ℘ ) × a \in(A/\wp)^{\times} a ∈ ( A / ℘ ) × , ρ a ( g ( χ k ) ) = a ( ε q k ) g ( χ k ) \rho_a(g(\chi_k)) = a(\varepsilon^{q^k})g(\chi_k) ρ a ( g ( χ k )) = a ( ε q k ) g ( χ k ) .
(b) We have
∏ k = 0 d − 1 g ( χ k ) q − 1 = ( − 1 ) d ℘ . \prod_{k=0}^{d-1} g(\chi_k)^{q-1} = (-1)^d\wp. k = 0 ∏ d − 1 g ( χ k ) q − 1 = ( − 1 ) d ℘ . (c) (Stickelberger’s Theorem) There is a unique root ( − ℘ ) 1 / ( q d − 1 ) (-\wp)^{1/(q^d-1)} ( − ℘ ) 1/ ( q d − 1 ) in the completion H p 0 H_{\mathfrak{p}_0} H p 0 such that ord p 0 ( e ( 1 / ℘ ) − ( − ℘ ) 1 / ( q d − 1 ) ) > 1 / ( q d − 1 ) \operatorname{ord}_{\mathfrak{p}_0}(e(1/\wp)-(-\wp)^{1/(q^d-1)}) > 1/(q^d-1) ord p 0 ( e ( 1/ ℘ ) − ( − ℘ ) 1/ ( q d − 1 ) ) > 1/ ( q d − 1 ) . In this case,
ord p 0 ( g ( χ ε ) ( − ℘ ) 1 / ( q d − 1 ) − 1 ) > 0 , \operatorname{ord}_{\mathfrak{p}_0}\left(\frac{g(\chi_{\varepsilon})}{(-\wp)^{1/(q^d-1)}}-1\right)>0, ord p 0 ( ( − ℘ ) 1/ ( q d − 1 ) g ( χ ε ) − 1 ) > 0 , and more generally, for $1 \le k \le d-1,
ord p 0 ( g ( χ k ) ( − ℘ ) q k / ( q d − 1 ) − 1 ( ε q k − θ ) ( ε q k − 1 − θ ) q ⋯ ( ε q − θ ) q k − 1 ) > 0. \operatorname{ord}_{\mathfrak{p}_0}\left(\frac{g(\chi_k)}{(-\wp)^{q^k/(q^d-1)}}-\frac{1}{(\varepsilon^{q^k}-\theta)(\varepsilon^{q^{k-1}}-\theta)^{q}\cdots(\varepsilon^q-\theta)^{q^{k-1}}}\right)>0. ord p 0 ( ( − ℘ ) q k / ( q d − 1 ) g ( χ k ) − ( ε q k − θ ) ( ε q k − 1 − θ ) q ⋯ ( ε q − θ ) q k − 1 1 ) > 0. (d) As ideals in O H \mathcal{O}_H O H , for $0 \le k \le d-1,
g ( χ k ) ⋅ O H = ∏ j = 0 d − 1 p j ( k − j ) m o d d . g(\chi_k)\cdot\mathcal{O}_H=\prod_{j=0}^{d-1}\mathfrak{p}_j^{(k-j)\bmod d}. g ( χ k ) ⋅ O H = j = 0 ∏ d − 1 p j ( k − j ) mod d . (e) For 0 ≤ k ≤ d − 1 0 \le k \le d-1 0 ≤ k ≤ d − 1 , the degree of g ( χ k ) g(\chi_k) g ( χ k ) at any infinite place of H H H is 1 / ( q − 1 ) 1/(q-1) 1/ ( q − 1 ) .
The companion results for dual Gauss-Thakur sums for the Carlitz module are the following. They can be proved in the same manner as in [42 ] directly from properties of g ∗ ( χ ε ) g^{*}(\chi_{\varepsilon}) g ∗ ( χ ε ) , but in large part we instead apply Theorem 7.1 via (72) . The trickiest part to derive is the analogue of Stickelberger’s theorem in (c).
Theorem 8.10. Let ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of ℘ \wp ℘ . Let K = k ( e ∗ ( 1 / ℘ ) ) = k ( e ( 1 / ℘ ) ) K = k(e^{*}(1/\wp)) = k(e(1/\wp)) K = k ( e ∗ ( 1/ ℘ )) = k ( e ( 1/ ℘ )) and H = F q d K H = \mathbb{F}_{q^d}K H = F q d K .
(a) Let 0 ≤ k ≤ d − 1 0 \le k \le d-1 0 ≤ k ≤ d − 1 . For σ ∈ Gal ( H / k ) \sigma\in\operatorname{Gal}(H/k) σ ∈ Gal ( H / k ) as above, σ ( g ∗ ( χ k ) ) = g ∗ ( χ k + 1 ) \sigma(g^{*}(\chi_k)) = g^{*}(\chi_{k+1}) σ ( g ∗ ( χ k )) = g ∗ ( χ k + 1 ) . For a ∈ ( A / ℘ ) × a \in(A/\wp)^{\times} a ∈ ( A / ℘ ) × , ρ a − 1 ( g ∗ ( χ k ) ) = a ( ε − q k ) g ∗ ( χ k ) \rho_a^{-1}(g^{*}(\chi_k)) = a(\varepsilon^{-q^k})g^{*}(\chi_k) ρ a − 1 ( g ∗ ( χ k )) = a ( ε − q k ) g ∗ ( χ k ) .
(b) We have
∏ k = 0 d − 1 g ∗ ( χ k ) q − 1 = ( − 1 ) d ℘ q . \prod_{k=0}^{d-1} g^{*}(\chi_k)^{q-1} = \frac{(-1)^d}{\wp^q}. k = 0 ∏ d − 1 g ∗ ( χ k ) q − 1 = ℘ q ( − 1 ) d .
(c) (Stickelberger’s Theorem) For the same root ( − ℘ ) 1 / ( q d − 1 ) (-\wp)^{1/(q^{d}-1)} ( − ℘ ) 1/ ( q d − 1 ) in Theorem 8.9 (c), we have ord p 0 ( e ∗ ( 1 / ℘ ) + ℘ ′ ( ϵ ) − 1 ( − ℘ ) − q d / ( q d − 1 ) ) > − q d / ( q d − 1 ) \operatorname{ord}_{\mathfrak{p}_0}\left(e^*(1/\wp)+\wp'(\epsilon)^{-1}(-\wp)^{-q^d/(q^d-1)}\right)>-q^d/(q^d-1) ord p 0 ( e ∗ ( 1/ ℘ ) + ℘ ′ ( ϵ ) − 1 ( − ℘ ) − q d / ( q d − 1 ) ) > − q d / ( q d − 1 ) . In this case,
ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ ) ) ( − ℘ ) q d / ( q d − 1 ) − 1 ) > 0 , \operatorname{ord}_{\mathfrak{p}_0}\left(g^*(\chi_\epsilon)(-\wp'(\epsilon))(-\wp)^{q^d/(q^d-1)}-1\right)>0, ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ )) ( − ℘ ) q d / ( q d − 1 ) − 1 ) > 0 ,
and more generally, for 1 ≤ k ≤ d − 1 1\leq k\leq d-1 1 ≤ k ≤ d − 1 ,
ord p 0 ( g ∗ ( χ k ) ( − ℘ ′ ( ϵ q k ) ) ( − ℘ ) ( q d + k ) / ( q d − 1 ) ( ϵ − θ ) q k ( ϵ q − θ ) q k − 1 ⋯ ( ϵ q k − 1 − θ ) q − 1 ) > 0. \operatorname{ord}_{\mathfrak{p}_0}\left(\frac{g^*(\chi_k)(-\wp'(\epsilon^{q^k}))(-\wp)^{(q^d+k)/(q^d-1)}}{(\epsilon-\theta)^{q^k}(\epsilon^q-\theta)^{q^{k-1}}\cdots(\epsilon^{q^{k-1}}-\theta)^q}-1\right)>0. ord p 0 ( ( ϵ − θ ) q k ( ϵ q − θ ) q k − 1 ⋯ ( ϵ q k − 1 − θ ) q g ∗ ( χ k ) ( − ℘ ′ ( ϵ q k )) ( − ℘ ) ( q d + k ) / ( q d − 1 ) − 1 ) > 0.
(d) As fractional ideals of O H O_H O H in H H H , for 0 ≤ k ≤ d − 1 0\leq k\leq d-1 0 ≤ k ≤ d − 1 ,
g ∗ ( χ k ) ⋅ O H = ∏ j = 0 d − 1 p j − q d − ( ( j − k ) m o d d ) . g^*(\chi_k)\cdot O_H=\prod_{j=0}^{d-1}\mathfrak{p}_j^{-q^{d-((j-k)\bmod d)}}. g ∗ ( χ k ) ⋅ O H = j = 0 ∏ d − 1 p j − q d − (( j − k ) mod d ) .
(e) For 0 ≤ k ≤ d − 1 0\leq k\leq d-1 0 ≤ k ≤ d − 1 , the degree of g ∗ ( χ k ) g^*(\chi_k) g ∗ ( χ k ) at any infinite place of H H H is − q / ( q − 1 ) -q/(q-1) − q / ( q − 1 ) .
Proof. Part (a) follows from the definitions of σ \sigma σ , ρ a ∈ Gal ( H / k ) \rho_a\in\operatorname{Gal}(H/k) ρ a ∈ Gal ( H / k ) and from (73) . For (b), we combine (72) with Theorem 8.9 (b) and find
∏ k = 0 d − 1 g ∗ ( χ k ) q − 1 = 1 ( − 1 ) d ℘ ⋅ ∏ k = 0 d − 1 1 ( θ − ϵ q k ) q − 1 ⋅ 1 ℘ ′ ( ϵ q k ) 2 ( q − 1 ) . \prod_{k=0}^{d-1}g^*(\chi_k)^{q-1}=\frac{1}{(-1)^d\wp}\cdot\prod_{k=0}^{d-1}\frac{1}{(\theta-\epsilon^{q^k})^{q-1}}\cdot\frac{1}{\wp'(\epsilon^{q^k})^{2(q-1)}}. k = 0 ∏ d − 1 g ∗ ( χ k ) q − 1 = ( − 1 ) d ℘ 1 ⋅ k = 0 ∏ d − 1 ( θ − ϵ q k ) q − 1 1 ⋅ ℘ ′ ( ϵ q k ) 2 ( q − 1 ) 1 .
In the product on the right, the first terms multiply together to give ℘ − ( q − 1 ) \wp^{-(q-1)} ℘ − ( q − 1 ) . The second terms multiply to give N ( ℘ ′ ( ϵ ) 2 ( q − 1 ) ) N(\wp'(\epsilon)^{2(q-1)}) N ( ℘ ′ ( ϵ ) 2 ( q − 1 ) ) , where N = N F q d / F q : F q d × → F q × N=N_{\mathbb{F}_{q^d}/\mathbb{F}_q}:\mathbb{F}_{q^d}^{\times}\to\mathbb{F}_q^{\times} N = N F q d / F q : F q d × → F q × is the norm. But this is 1 as it is the ( q − 1 ) (q-1) ( q − 1 ) -st power of an element of F q × \mathbb{F}_q^{\times} F q × .
For (c), substituting (72) into the first main inequality in Theorem 8.9 (c), we obtain
ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ) 1 / ( q d − 1 ) ( θ − ϵ ) ℘ ′ ( ϵ ) 2 1 − 1 ) > 0. (74) \operatorname{ord}_{\mathfrak{p}_0}\left(\frac{g^*(\chi_\epsilon)(-\wp)^{1/(q^d-1)}(\theta-\epsilon)\wp'(\epsilon)^2}{1}-1\right)>0. \tag*{(74)} ord p 0 ( 1 g ∗ ( χ ϵ ) ( − ℘ ) 1/ ( q d − 1 ) ( θ − ϵ ) ℘ ′ ( ϵ ) 2 − 1 ) > 0. (74)
We note that (72) implies ord p 0 ( g ( χ ϵ ) ) + ord p 0 ( g ∗ ( χ ϵ ) ) = − ord p 0 ( θ − ϵ ) = − 1 \operatorname{ord}_{\mathfrak{p}_0}(g(\chi_\epsilon))+\operatorname{ord}_{\mathfrak{p}_0}(g^*(\chi_\epsilon))=-\operatorname{ord}_{\mathfrak{p}_0}(\theta-\epsilon)=-1 ord p 0 ( g ( χ ϵ )) + ord p 0 ( g ∗ ( χ ϵ )) = − ord p 0 ( θ − ϵ ) = − 1 . Thus, as ord p 0 ( g ( χ ϵ ) ) = 1 / ( q d − 1 ) \operatorname{ord}_{\mathfrak{p}_0}(g(\chi_\epsilon))=1/(q^d-1) ord p 0 ( g ( χ ϵ )) = 1/ ( q d − 1 ) by Theorem 8.9 (c), ord p 0 ( g ∗ ( χ ϵ ) ) = − q d / ( q d − 1 ) \operatorname{ord}_{\mathfrak{p}_0}(g^*(\chi_\epsilon))=-q^d/(q^d-1) ord p 0 ( g ∗ ( χ ϵ )) = − q d / ( q d − 1 ) . Multiplying the expression in (74) by g ∗ ( χ ϵ ) ( − ℘ ) q d / ( q d − 1 ) ℘ ′ ( ϵ ) g^*(\chi_\epsilon)(-\wp)^{q^d/(q^d-1)}\wp'(\epsilon) g ∗ ( χ ϵ ) ( − ℘ ) q d / ( q d − 1 ) ℘ ′ ( ϵ ) , which has p 0 \mathfrak{p}_0 p 0 -adic valuation 0, yields
ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ ) ) ( − ℘ ) q d / ( q d − 1 ) − ℘ ( θ ) ( θ − ϵ ) ℘ ′ ( ϵ ) ) > 0. \operatorname{ord}_{\mathfrak{p}_0}\left(g^*(\chi_\epsilon)(-\wp'(\epsilon))(-\wp)^{q^d/(q^d-1)}-\frac{\wp(\theta)}{(\theta-\epsilon)\wp'(\epsilon)}\right)>0. ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ )) ( − ℘ ) q d / ( q d − 1 ) − ( θ − ϵ ) ℘ ′ ( ϵ ) ℘ ( θ ) ) > 0.
Since ℘ ( θ ) = ℘ ′ ( ϵ ) ( θ − ϵ ) + O ( ( θ − ϵ ) 2 ) \wp(\theta)=\wp'(\epsilon)(\theta-\epsilon)+O((\theta-\epsilon)^2) ℘ ( θ ) = ℘ ′ ( ϵ ) ( θ − ϵ ) + O (( θ − ϵ ) 2 ) in F q d [ θ ] \mathbb{F}_{q^d}[\theta] F q d [ θ ] , we then obtain
ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ ) ) ( − ℘ ) q d / ( q d − 1 ) − 1 ) > 0. (75) \operatorname{ord}_{\mathfrak{p}_0}\left(g^*(\chi_\epsilon)(-\wp'(\epsilon))(-\wp)^{q^d/(q^d-1)}-1\right)>0. \tag*{(75)} ord p 0 ( g ∗ ( χ ϵ ) ( − ℘ ′ ( ϵ )) ( − ℘ ) q d / ( q d − 1 ) − 1 ) > 0. (75)
Now for k ∈ Z k\in\mathbb{Z} k ∈ Z , since C θ ∗ = θ + τ − 1 C_\theta^*=\theta+\tau^{-1} C θ ∗ = θ + τ − 1 ,
ϵ q k g ∗ ( χ k ) = g ∗ ( χ k ; θ ) = − ∑ z ∈ F q d × z ( ϵ q k ) − 1 e ∗ ( θ z / ℘ ) = − ∑ z ∈ F q d × z ( ϵ q k ) − 1 ( θ e ∗ ( 1 / ℘ ) + e ∗ ( z / ℘ ) 1 / q ) = θ g ∗ ( χ k ) + g ∗ ( χ k + 1 ) 1 / q . \begin{aligned}
\epsilon^{q^k}g^*(\chi_k)=g^*(\chi_k;\theta)=-\sum_{z\in\mathbb{F}_{q^d}^{\times}}z(\epsilon^{q^k})^{-1}e^*(\theta z/\wp) \\
&=-\sum_{z\in\mathbb{F}_{q^d}^{\times}}z(\epsilon^{q^k})^{-1}\left(\theta e^*(1/\wp)+e^*(z/\wp)^{1/q}\right)=\theta g^*(\chi_k)+g^*(\chi_{k+1})^{1/q}.
\end{aligned} ϵ q k g ∗ ( χ k ) = g ∗ ( χ k ; θ ) = − z ∈ F q d × ∑ z ( ϵ q k ) − 1 e ∗ ( θ z / ℘ ) = − z ∈ F q d × ∑ z ( ϵ q k ) − 1 ( θ e ∗ ( 1/ ℘ ) + e ∗ ( z / ℘ ) 1/ q ) = θ g ∗ ( χ k ) + g ∗ ( χ k + 1 ) 1/ q .
Thus,
g ∗ ( χ k ) q = g ∗ ( χ k + 1 ) ( ϵ q k − θ ) q . (76) g^*(\chi_k)^q=\frac{g^*(\chi_{k+1})}{(\epsilon^{q^k}-\theta)^q}. \tag*{(76)} g ∗ ( χ k ) q = ( ϵ q k − θ ) q g ∗ ( χ k + 1 ) . (76)
Using the convention that an empty product is 1, we see that (8.12) is the k = 0 k=0 k = 0 version of the inequality in (c). Thus proceeding by induction on k k k , assume the inequality in (c) holds for k < d − 1 k<d-1 k < d − 1 . Raising the expression in (c) for k k k to the q q q -th power and substituting in (8.13), we obtain
g ∗ ( χ k ) q ( − ℘ ′ ) ( ε q k + 1 ) ( − ℘ ) q d + k + 1 / ( q d − 1 ) ( ε − θ ) q k + 1 ( ε q − θ ) q k ⋯ ( ε q k − θ ) q 2 = g ∗ ( χ k + 1 ) ( − ℘ ′ ) ( ε q k + 1 ) ( − ℘ ) q d + k + 1 / ( q d − 1 ) ( ε − θ ) q k + 1 ( ε q − θ ) q k ⋯ ( ε q k − θ ) q , \frac{g^*(\chi_k)^q(-\wp')(\varepsilon^{q^{k+1}})(-\wp)^{q^{d+k+1}/(q^d-1)}}{(\varepsilon-\theta)^{q^{k+1}}(\varepsilon^q-\theta)^{q^k}\cdots(\varepsilon^{q^k}-\theta)^{q^2}}=\frac{g^*(\chi_{k+1})(-\wp')(\varepsilon^{q^{k+1}})(-\wp)^{q^{d+k+1}/(q^d-1)}}{(\varepsilon-\theta)^{q^{k+1}}(\varepsilon^q-\theta)^{q^k}\cdots(\varepsilon^{q^k}-\theta)^q}, ( ε − θ ) q k + 1 ( ε q − θ ) q k ⋯ ( ε q k − θ ) q 2 g ∗ ( χ k ) q ( − ℘ ′ ) ( ε q k + 1 ) ( − ℘ ) q d + k + 1 / ( q d − 1 ) = ( ε − θ ) q k + 1 ( ε q − θ ) q k ⋯ ( ε q k − θ ) q g ∗ ( χ k + 1 ) ( − ℘ ′ ) ( ε q k + 1 ) ( − ℘ ) q d + k + 1 / ( q d − 1 ) ,
which yields the desired inequality in (c).
For (d), fix 0 ≤ k ≤ d − 1 0\leq k\leq d-1 0 ≤ k ≤ d − 1 . For 0 ≤ j ≤ d − 1 0\leq j\leq d-1 0 ≤ j ≤ d − 1 , we obtain from (72) that
ord p j ( g ∗ ( χ k ) ) + ord p j ( g ( χ k ) ) = { 0 if j ≠ k , − 1 if j = k . \operatorname{ord}_{\mathfrak p_j}(g^*(\chi_k))+\operatorname{ord}_{\mathfrak p_j}(g(\chi_k))=
\begin{cases}
0 & \text{if }j\ne k,\\
-1 & \text{if }j=k.
\end{cases} ord p j ( g ∗ ( χ k )) + ord p j ( g ( χ k )) = { 0 − 1 if j = k , if j = k .
Using this together with Theorem 8.9 (d), we find
g ∗ ( χ k ) ⋅ O H = p k − q d ⋅ ∏ j = 0 j ≠ k d − 1 p j q ( k − j ) m o d d = ∏ j = 0 d − 1 p j q d − ( ( j − k ) m o d d ) . g^*(\chi_k)\cdot\mathcal{O}_H=\mathfrak p_k^{-q^d}\cdot\prod_{\substack{j=0\\j\ne k}}^{d-1}\mathfrak p_j^{q^{(k-j)\bmod d}}=\prod_{j=0}^{d-1}\mathfrak p_j^{q^{d-((j-k)\bmod d)}}. g ∗ ( χ k ) ⋅ O H = p k − q d ⋅ j = 0 j = k ∏ d − 1 p j q ( k − j ) mod d = j = 0 ∏ d − 1 p j q d − (( j − k ) mod d ) .
Note that the powers on p j \mathfrak p_j p j are affected by our normalization of the valuation ord p j \operatorname{ord}_{\mathfrak p_j} ord p j . Part (e) follows directly from Theorem 8.9 (e) and (72) . □
Remark 8.14. One can show that with the chosen root ( − ℘ ) 1 / ( q d − 1 ) (-\wp)^{1/(q^d-1)} ( − ℘ ) 1/ ( q d − 1 ) in Theorem 8.9 (c),
ord p 0 ( ε ∗ ( 1 / ℘ ) + ( − ℘ ) − q d / ( q d − 1 ) ℘ ′ ( ε ) ) > − q d q d − 1 . \operatorname{ord}_{\mathfrak p_0}\left(\varepsilon^*(1/\wp)+\frac{(-\wp)^{-q^d/(q^d-1)}}{\wp'(\varepsilon)}\right)>-\frac{q^d}{q^d-1}. ord p 0 ( ε ∗ ( 1/ ℘ ) + ℘ ′ ( ε ) ( − ℘ ) − q d / ( q d − 1 ) ) > − q d − 1 q d .
We have omitted the details, which are similar to those in [42 Lem. II] , for brevity.
∞ \infty ∞ -adic signs of Gauss-Thakur sums
We have seen in Theorems 8.9 (e) and 8.10 (e) that the ∞ \infty ∞ -adic degrees of Carlitz module Gauss-Thakur sums and their duals are explicit. These align with [45 Thm. 1.2] , where Thakur relates the ∞ \infty ∞ -adic size of g ( χ ε ) g(\chi_\varepsilon) g ( χ ε ) to the ℘ \wp ℘ -torsion on C C C with largest norm. We can also develop information about the signs of g ( χ ε ) g(\chi_\varepsilon) g ( χ ε ) and g ∗ ( χ ε ) g^*(\chi_\varepsilon) g ∗ ( χ ε ) . Suppose we have fixed a decomposition C ∞ = F q × × ( 1 / θ ) Q × U 1 ( C ∞ ) \mathbb{C}_\infty=\mathbb{F}_q^\times\times(1/\theta)^{\mathbb{Q}}\times U_1(\mathbb{C}_\infty) C ∞ = F q × × ( 1/ θ ) Q × U 1 ( C ∞ ) , where consistent roots of θ \theta θ have been chosen and U 1 ( C ∞ ) U_1(\mathbb{C}_\infty) U 1 ( C ∞ ) is the group of 1-units. Projecting onto F q × \mathbb{F}_q^\times F q × defines a sign function, sgn : C ∞ × → F q × \operatorname{sgn}:\mathbb{C}_\infty^\times\to\mathbb{F}_q^\times sgn : C ∞ × → F q × , that we fix throughout. In this way all rational powers of θ \theta θ have sign 1, and moreover, the group ( 1 / θ ) Q × U 1 ( C ∞ ) (1/\theta)^{\mathbb{Q}}\times U_1(\mathbb{C}_\infty) ( 1/ θ ) Q × U 1 ( C ∞ ) of sign 1 elements is uniquely divisible. Also, by (69) ,
sgn ( π ~ ) = − ( − 1 ) q / ( q − 1 ) = ( − 1 ) 1 / ( q − 1 ) . (77) \operatorname{sgn}(\widetilde{\pi})=-(-1)^{q/(q-1)}=(-1)^{1/(q-1)}. \tag*{(77)} sgn ( π ) = − ( − 1 ) q / ( q − 1 ) = ( − 1 ) 1/ ( q − 1 ) . (77)
where throughout we fix this choice of root of − 1 -1 − 1 so that (69) –(70) are consistent. Results on the signs of Gauss-Thakur sums are then due to Thakur.
Proposition 8.17 (Thakur [45 Thm. 2.3] ). For 0 ≤ k ≤ d − 1 0\leq k\leq d-1 0 ≤ k ≤ d − 1 ,
sgn ( g ( χ k ) ) = ( − 1 ) 1 / ( q − 1 ) ℘ ′ ( ε q k ) , sgn ( g ∗ ( χ k ) ) = 1 ( − 1 ) 1 / ( q − 1 ) ℘ ′ ( ε q k ) . \operatorname{sgn}(g(\chi_k))=\frac{(-1)^{1/(q-1)}}{\wp'(\varepsilon^{q^k})},\qquad\operatorname{sgn}(g^*(\chi_k))=\frac{1}{(-1)^{1/(q-1)}\wp'(\varepsilon^{q^k})}. sgn ( g ( χ k )) = ℘ ′ ( ε q k ) ( − 1 ) 1/ ( q − 1 ) , sgn ( g ∗ ( χ k )) = ( − 1 ) 1/ ( q − 1 ) ℘ ′ ( ε q k ) 1 .
Proof. For completeness we prove the case of sgn ( g ∗ ( χ ε ) ) \operatorname{sgn}(g^*(\chi_\varepsilon)) sgn ( g ∗ ( χ ε )) , though it can be derived from [45 Thm. 2.3] and (72) . The argument is similar in spirit to [45 Thm. 2.3] (see also [47 Thm. 4.4.1] ). For x ∈ k ∞ x\in k_\infty x ∈ k ∞ with ∣ x ∣ ∞ < 1 |x|_\infty<1 ∣ x ∣ ∞ < 1 , [2 Pf. of Lem. 5.2.6] implies that ∣ e ∗ ( x ) ∣ ∞ = ∣ θ ∣ − q deg x / ( q − 1 ) |e^*(x)|_\infty= |\theta|^{-q^{\deg x}/(q-1)} ∣ e ∗ ( x ) ∣ ∞ = ∣ θ ∣ − q d e g x / ( q − 1 ) , (see also [14 Eq. (2.5.9)] , keep in mind Remark 8.4). For z ∈ A z \in A z ∈ A , deg z < deg φ \deg z < \deg\varphi deg z < deg φ , it follows that deg e ∗ ( z / φ ) = − q d − deg z / ( q − 1 ) \deg e^*(z/\varphi) = -q^{d-\deg z}/(q-1) deg e ∗ ( z / φ ) = − q d − d e g z / ( q − 1 ) is maximized when deg z = d − 1 \deg z = d-1 deg z = d − 1 , and in that case
sgn ( e ∗ ( z / φ ) ) = sgn ( z ) sgn ( e ∗ ( θ − 1 ) ) = sgn ( z ) sgn ( − ( − θ ) − q / ( q − 1 ) ) , (78) \operatorname{sgn}\left(e^*(z/\varphi)\right) = \operatorname{sgn}(z)\operatorname{sgn}\left(e^*(\theta^{-1})\right) = \operatorname{sgn}(z)\operatorname{sgn}\left(-(-\theta)^{-q/(q-1)}\right), \tag*{(78)} sgn ( e ∗ ( z / φ ) ) = sgn ( z ) sgn ( e ∗ ( θ − 1 ) ) = sgn ( z ) sgn ( − ( − θ ) − q / ( q − 1 ) ) , (78)
by (8.3). Therefore,
g ∗ ( χ ε ) = − ∑ z ∈ A deg z < d ′ z ( ε ) − 1 e ∗ ( z / φ ) = − ∑ z ∈ A deg z = d − 1 z ( ε ) − 1 e ∗ ( z / φ ) + E , deg E < − q q − 1 . g^*(\chi_\varepsilon)=-\sum_{\substack{z\in A\\\deg z<d}}'z(\varepsilon)^{-1}e^*(z/\varphi)=-\sum_{\substack{z\in A\\\deg z=d-1}}z(\varepsilon)^{-1}e^*(z/\varphi)+E,\qquad\deg E<-\frac{q}{q-1}. g ∗ ( χ ε ) = − z ∈ A d e g z < d ∑ ′ z ( ε ) − 1 e ∗ ( z / φ ) = − z ∈ A d e g z = d − 1 ∑ z ( ε ) − 1 e ∗ ( z / φ ) + E , deg E < − q − 1 q .
By Theorem 8.10 (e), this initial sum is non-zero, so by (8.16) and (8.18),
sgn ( g ∗ ( χ ε ) ) = − sgn ( ∑ z ∈ A deg z = d − 1 z ( ε ) − 1 e ∗ ( z / φ ) ) = − 1 ( − 1 ) 1 / ( q − 1 ) ⋅ sgn ( ∑ z ∈ A deg z = d − 1 z ( θ ) z ( ε ) ) . \operatorname{sgn}\left(g^*(\chi_\varepsilon)\right)=-\operatorname{sgn}\left(\sum_{\substack{z\in A\\\deg z=d-1}}z(\varepsilon)^{-1}e^*(z/\varphi)\right)=\frac{-1}{(-1)^{1/(q-1)}}\cdot\operatorname{sgn}\left(\sum_{\substack{z\in A\\\deg z=d-1}}\frac{z(\theta)}{z(\varepsilon)}\right). sgn ( g ∗ ( χ ε ) ) = − sgn z ∈ A d e g z = d − 1 ∑ z ( ε ) − 1 e ∗ ( z / φ ) = ( − 1 ) 1/ ( q − 1 ) − 1 ⋅ sgn z ∈ A d e g z = d − 1 ∑ z ( ε ) z ( θ ) .
By (6.16), this latter sum is − ( θ − ε ) ⋯ ( θ − ε q d − 2 ) / ( ( ε − ε q ) ⋯ ( ε − ε q d − 1 ) ) -(\theta-\varepsilon)\cdots(\theta-\varepsilon^{q^{d-2}})/((\varepsilon-\varepsilon^q)\cdots(\varepsilon-\varepsilon^{q^{d-1}})) − ( θ − ε ) ⋯ ( θ − ε q d − 2 ) / (( ε − ε q ) ⋯ ( ε − ε q d − 1 )) , and so sgn ( g ∗ ( χ ε ) ) = ( − 1 ) − 1 / ( q − 1 ) / φ ( ε ) \operatorname{sgn}(g^*(\chi_\varepsilon))=(-1)^{-1/(q-1)}/\varphi(\varepsilon) sgn ( g ∗ ( χ ε )) = ( − 1 ) − 1/ ( q − 1 ) / φ ( ε ) . Replacing ε \varepsilon ε by ε q k \varepsilon^{q^k} ε q k yields the general result. □ \square □
Matrices of (dual) Gauss-Thakur sums for the Carlitz module
In the context of Theorem 7.7 , consider the d × d d\times d d × d matrices,
Γ d ( C ) = ( g ( χ 0 ) ⋯ g [ d − 1 ] ( χ 0 ) ⋮ ⋮ g ( χ d − 1 ) ⋯ g [ d − 1 ] ( χ d − 1 ) ) , Γ d ∗ ( C ∗ ) = ( g ∗ ( χ 0 ) ⋯ g ∗ ( χ d − 1 ) ⋮ ⋮ g ∗ [ d − 1 ] ( χ 0 ) ⋯ g ∗ [ d − 1 ] ( χ d − 1 ) ) . \Gamma_d(\mathbb{C})=
\begin{pmatrix}
g(\chi_0)&\cdots&g^{[d-1]}(\chi_0)\\
\vdots&&\vdots\\
g(\chi_{d-1})&\cdots&g^{[d-1]}(\chi_{d-1})
\end{pmatrix},
\qquad
\Gamma_d^*(\mathbb{C}^*)=
\begin{pmatrix}
g^*(\chi_0)&\cdots&g^*(\chi_{d-1})\\
\vdots&&\vdots\\
g^{*[d-1]}(\chi_0)&\cdots&g^{*[d-1]}(\chi_{d-1})
\end{pmatrix}. Γ d ( C ) = g ( χ 0 ) ⋮ g ( χ d − 1 ) ⋯ ⋯ g [ d − 1 ] ( χ 0 ) ⋮ g [ d − 1 ] ( χ d − 1 ) , Γ d ∗ ( C ∗ ) = g ∗ ( χ 0 ) ⋮ g ∗ [ d − 1 ] ( χ 0 ) ⋯ ⋯ g ∗ ( χ d − 1 ) ⋮ g ∗ [ d − 1 ] ( χ d − 1 ) .
Theorem 7.7 (c) and Remark 7.13 imply
Γ d ∗ ( C ∗ ) ⋅ Q ( φ ′ ) ⋅ Γ d ( C ) = ( 1 / φ 0 0 ⋯ 0 { 1 / φ 1 } C 1 / φ q 0 ⋯ 0 ⋮ ⋱ ⋱ ⋮ { 1 / φ d − 2 } C ⋯ { 1 / φ 1 } C 1 / φ q d − 2 0 { 1 / φ d − 1 } C { 1 / φ d − 2 } C ⋯ { 1 / φ 1 } C 1 / φ q d − 2 1 / φ q d − 1 ) . (79) \Gamma_d^*(\mathbb{C}^*)\cdot\mathbb{Q}(\varphi')\cdot\Gamma_d(\mathbb{C})=
\begin{pmatrix}
1/\varphi&0&0&\cdots&0\\
\left\{\frac{1/\varphi}{1}\right\}_{\mathbb{C}}&1/\varphi^q&0&\cdots&0\\
\vdots&\ddots&\ddots&&\vdots\\
\left\{\frac{1/\varphi}{d-2}\right\}_{\mathbb{C}}&\cdots&\left\{\frac{1/\varphi}{1}\right\}_{\mathbb{C}}1/\varphi^{q^{d-2}}&0\\
\left\{\frac{1/\varphi}{d-1}\right\}_{\mathbb{C}}&\left\{\frac{1/\varphi}{d-2}\right\}_{\mathbb{C}}&\cdots&\left\{\frac{1/\varphi}{1}\right\}_{\mathbb{C}}1/\varphi^{q^{d-2}}&1/\varphi^{q^{d-1}}
\end{pmatrix}. \tag*{(79)} Γ d ∗ ( C ∗ ) ⋅ Q ( φ ′ ) ⋅ Γ d ( C ) = 1/ φ { 1 1/ φ } C ⋮ { d − 2 1/ φ } C { d − 1 1/ φ } C 0 1/ φ q ⋱ ⋯ { d − 2 1/ φ } C 0 0 ⋱ { 1 1/ φ } C 1/ φ q d − 2 ⋯ ⋯ ⋯ 0 { 1 1/ φ } C 1/ φ q d − 2 0 0 ⋮ 1/ φ q d − 1 . (79)
A natural problem to consider is to evaluate the determinants of Γ d ( C ) \Gamma_d(\mathbb{C}) Γ d ( C ) and Γ d ∗ ( C ∗ ) \Gamma_d^*(\mathbb{C}^*) Γ d ∗ ( C ∗ ) , which are nonzero by (8.20). To proceed, since ω ( 1 ) = ( t − θ ) ω \omega^{(1)}=(t-\theta)\omega ω ( 1 ) = ( t − θ ) ω , we have
ω C ( j ) = ω C ∏ ℓ = 0 j − 1 ( t − θ q ℓ ) , j ≥ 1. \omega_{\mathbb{C}}^{(j)}=\omega_{\mathbb{C}}\prod_{\ell=0}^{j-1}(t-\theta^{q^\ell}),\qquad j\geq1. ω C ( j ) = ω C ℓ = 0 ∏ j − 1 ( t − θ q ℓ ) , j ≥ 1.
Thus by Theorem 6.13 and Corollary 6.19, as well as (8.6), we find, by evaluating both sides of of this identity at t = ε q k t=\varepsilon^{q^k} t = ε q k , k ∈ Z k\in\mathbb{Z} k ∈ Z , and dividing by φ ′ ( ε q k ) \varphi'(\varepsilon^{q^k}) φ ′ ( ε q k ) , that
g [ j ] ( χ k ) = g ( χ k ) ∏ ℓ = 0 j − 1 ( ε q k − θ q ℓ ) . g^{[j]}(\chi_k)=g(\chi_k)\prod_{\ell=0}^{j-1}(\varepsilon^{q^k}-\theta^{q^\ell}). g [ j ] ( χ k ) = g ( χ k ) ℓ = 0 ∏ j − 1 ( ε q k − θ q ℓ ) .
Rewriting Γ d ( C ) \Gamma_d(\mathrm{C}) Γ d ( C ) using this identity, we can now pull out the common factor of g ( χ k ) g(\chi_k) g ( χ k ) , 0 ≤ k ≤ d − 1 0 \le k \le d-1 0 ≤ k ≤ d − 1 , in each row to obtain the determinant relation
det Γ d ( C ) = ∏ k = 0 d − 1 g ( χ k ) ⋅ det ( 1 ε − θ ⋯ ∏ ℓ = 0 d − 1 ( ε − θ q ℓ ) 1 ε q − θ ⋯ ∏ ℓ = 0 d − 1 ( ε q − θ q ℓ ) ⋮ ⋮ ⋮ 1 ε q d − 1 − θ ⋯ ∏ ℓ = 0 d − 1 ( ε q d − 1 − θ q ℓ ) ) . (80) \det\Gamma_d(\mathrm{C})=\prod_{k=0}^{d-1}g(\chi_k)\cdot\det\begin{pmatrix}1&\varepsilon-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon-\theta^{q^\ell})\\1&\varepsilon^q-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon^q-\theta^{q^\ell})\\ \vdots&\vdots&&\vdots\\1&\varepsilon^{q^{d-1}}-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon^{q^{d-1}}-\theta^{q^\ell})\end{pmatrix}. \tag*{(80)} det Γ d ( C ) = k = 0 ∏ d − 1 g ( χ k ) ⋅ det 1 1 ⋮ 1 ε − θ ε q − θ ⋮ ε q d − 1 − θ ⋯ ⋯ ⋯ ∏ ℓ = 0 d − 1 ( ε − θ q ℓ ) ∏ ℓ = 0 d − 1 ( ε q − θ q ℓ ) ⋮ ∏ ℓ = 0 d − 1 ( ε q d − 1 − θ q ℓ ) . (80)
By performing column manipulations of this determinant we find that
V ( ε ) : = det ( 1 ε − θ ⋯ ∏ ℓ = 0 d − 1 ( ε − θ q ℓ ) 1 ε q − θ ⋯ ∏ ℓ = 0 d − 1 ( ε q − θ q ℓ ) ⋮ ⋮ ⋮ 1 ε q d − 1 − θ ⋯ ∏ ℓ = 0 d − 1 ( ε q d − 1 − θ q ℓ ) ) = det ( 1 ε ⋯ ε d − 1 1 ε q ⋯ ε ( d − 1 ) q ⋮ ⋮ ⋮ 1 ε q d − 1 ⋯ ε ( d − 1 ) q d − 1 ) , (81) V(\varepsilon):=\det\begin{pmatrix}1&\varepsilon-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon-\theta^{q^\ell})\\1&\varepsilon^q-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon^q-\theta^{q^\ell})\\ \vdots&\vdots&&\vdots\\1&\varepsilon^{q^{d-1}}-\theta&\cdots&\prod_{\ell=0}^{d-1}(\varepsilon^{q^{d-1}}-\theta^{q^\ell})\end{pmatrix}
=\det\begin{pmatrix}1&\varepsilon&\cdots&\varepsilon^{d-1}\\1&\varepsilon^q&\cdots&\varepsilon^{(d-1)q}\\ \vdots&\vdots&&\vdots\\1&\varepsilon^{q^{d-1}}&\cdots&\varepsilon^{(d-1)q^{d-1}}\end{pmatrix}, \tag*{(81)} V ( ε ) := det 1 1 ⋮ 1 ε − θ ε q − θ ⋮ ε q d − 1 − θ ⋯ ⋯ ⋯ ∏ ℓ = 0 d − 1 ( ε − θ q ℓ ) ∏ ℓ = 0 d − 1 ( ε q − θ q ℓ ) ⋮ ∏ ℓ = 0 d − 1 ( ε q d − 1 − θ q ℓ ) = det 1 1 ⋮ 1 ε ε q ⋮ ε q d − 1 ⋯ ⋯ ⋯ ε d − 1 ε ( d − 1 ) q ⋮ ε ( d − 1 ) q d − 1 , (81)
which is the evaluation V ( ε , … , ε q d − 1 ) V(\varepsilon,\ldots,\varepsilon^{q^{d-1}}) V ( ε , … , ε q d − 1 ) of the Vandermonde determinant V ( t 1 , … , t d ) V(t_1,\ldots,t_d) V ( t 1 , … , t d ) . Notably V ( ε ) ∈ F q d V(\varepsilon)\in\mathbb{F}_{q^d} V ( ε ) ∈ F q d , and V ( ε ) 2 = ( − 1 ) d ( d − 1 ) / 2 N ( ℘ ′ ( ε ) ) ∈ F q × V(\varepsilon)^2=(-1)^{d(d-1)/2}N(\wp'(\varepsilon))\in\mathbb{F}_q^\times V ( ε ) 2 = ( − 1 ) d ( d − 1 ) /2 N ( ℘ ′ ( ε )) ∈ F q × . Since we have V ( ε ) = ∏ 0 ≤ i < j ≤ d − 1 ( ε q j − ε q i ) V(\varepsilon)=\prod_{0\le i<j\le d-1}(\varepsilon^{q^j}-\varepsilon^{q^i}) V ( ε ) = ∏ 0 ≤ i < j ≤ d − 1 ( ε q j − ε q i ) , we observe that V ( ε ) q = ( − 1 ) d − 1 V ( ε ) V(\varepsilon)^q=(-1)^{d-1}V(\varepsilon) V ( ε ) q = ( − 1 ) d − 1 V ( ε ) . Thus right away, combining (8.21) with Theorem 8.9 (b),
det Γ d ( C ) q − 1 = ( − 1 ) d ℘ ⋅ V ( ε ) q − 1 = − ℘ . (82) \det\Gamma_d(\mathrm{C})^{q-1}=(-1)^d\wp\cdot V(\varepsilon)^{q-1}=-\wp. \tag*{(82)} det Γ d ( C ) q − 1 = ( − 1 ) d ℘ ⋅ V ( ε ) q − 1 = − ℘ . (82)
We extract this root ℘ \wp ℘ -adically and ∞ \infty ∞ -adically in the following theorem.
Theorem 8.24. Let ℘ ∈ A + \wp\in A_+ ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of ℘ \wp ℘ . Let K = k ( e ( 1 / ℘ ) ) K=k(e(1/\wp)) K = k ( e ( 1/ ℘ )) and H = F q d K H=\mathbb{F}_{q^d}K H = F q d K . Define Γ d ( C ) , Γ d ∗ ( C ∗ ) ∈ G L d ( H ) \Gamma_d(\mathrm{C}),\Gamma_d^*(\mathrm{C}^*)\in\mathrm{GL}_d(H) Γ d ( C ) , Γ d ∗ ( C ∗ ) ∈ GL d ( H ) as in §8.19.
(a) We have in H H H ,
det Γ d ( C ) q − 1 = − ℘ and det Γ d ∗ ( C ∗ ) q − 1 = − 1 ℘ q d . \det\Gamma_d(\mathrm{C})^{q-1}=-\wp\quad\text{and}\quad\det\Gamma_d^*(\mathrm{C}^*)^{q-1}=-\frac{1}{\wp^{q^d}}. det Γ d ( C ) q − 1 = − ℘ and det Γ d ∗ ( C ∗ ) q − 1 = − ℘ q d 1 . (b) (℘ \wp ℘ -adic Considerations) In H p 0 H_{\mathfrak p_0} H p 0 , by choosing the root ( − ℘ ) 1 / ( q d − 1 ) (-\wp)^{1/(q^d-1)} ( − ℘ ) 1/ ( q d − 1 ) as in Theorem 8.9 (c),
det Γ d ( C ) = ( − ℘ ) 1 / ( q − 1 ) and det Γ d ∗ ( C ∗ ) = ( − 1 ) d N ( ℘ ′ ( ε ) ) ⋅ ( − ℘ ) q d / ( q − 1 ) . \det\Gamma_d(\mathrm{C})=(-\wp)^{1/(q-1)}\quad\text{and}\quad\det\Gamma_d^*(\mathrm{C}^*)=\frac{(-1)^d}{N(\wp'(\varepsilon))\cdot(-\wp)^{q^d/(q-1)}}. det Γ d ( C ) = ( − ℘ ) 1/ ( q − 1 ) and det Γ d ∗ ( C ∗ ) = N ( ℘ ′ ( ε )) ⋅ ( − ℘ ) q d / ( q − 1 ) ( − 1 ) d . (c) (∞ \infty ∞ -adic Considerations) In C ∞ C_\infty C ∞ , by choosing ( − 1 ) 1 / ( q − 1 ) (-1)^{1/(q-1)} ( − 1 ) 1/ ( q − 1 ) as in (77) and ℘ 1 / ( q − 1 ) \wp^{1/(q-1)} ℘ 1/ ( q − 1 ) to have sign 1 1 1 ,
det Γ d ( C ) = ( − 1 ) d ( d − 1 ) / 2 ( − 1 ) d / ( q − 1 ) V ( ε ) ⋅ ℘ 1 / ( q − 1 ) , \det\Gamma_d(\mathrm{C})=\frac{(-1)^{d(d-1)/2}(-1)^{d/(q-1)}}{V(\varepsilon)}\cdot\wp^{1/(q-1)}, det Γ d ( C ) = V ( ε ) ( − 1 ) d ( d − 1 ) /2 ( − 1 ) d / ( q − 1 ) ⋅ ℘ 1/ ( q − 1 ) , and
det Γ d ∗ ( C ∗ ) = 1 ( − 1 ) d / ( q − 1 ) V ( ε ) ⋅ 1 ℘ q d / ( q − 1 ) . \det\Gamma_d^*(\mathrm{C}^*)=\frac{1}{(-1)^{d/(q-1)}V(\varepsilon)}\cdot\frac{1}{\wp^{q^d/(q-1)}}. det Γ d ∗ ( C ∗ ) = ( − 1 ) d / ( q − 1 ) V ( ε ) 1 ⋅ ℘ q d / ( q − 1 ) 1 .
Proof. The first identity in (a) was proved in (82) . Now det Q ( ℘ ′ ) = N ( ℘ ′ ( ε ) ) ∈ F q × \det Q(\wp')=N(\wp'(\varepsilon))\in\mathbb{F}_q^\times det Q ( ℘ ′ ) = N ( ℘ ′ ( ε )) ∈ F q × , and so taking determinants of (79) ,
N ( ℘ ′ ( ε ) ) ⋅ det Γ d ∗ ( C ∗ ) ⋅ det Γ d ( C ) = 1 ℘ ( q d − 1 ) / ( q − 1 ) . (83) N(\wp'(\varepsilon))\cdot\det\Gamma_d^*(\mathrm{C}^*)\cdot\det\Gamma_d(\mathrm{C})=\frac{1}{\wp^{(q^d-1)/(q-1)}}. \tag*{(83)} N ( ℘ ′ ( ε )) ⋅ det Γ d ∗ ( C ∗ ) ⋅ det Γ d ( C ) = ℘ ( q d − 1 ) / ( q − 1 ) 1 . (83)
Taking a ( q − 1 ) (q-1) ( q − 1 ) -st power and substituting the first part of (a) yields the second part.
For the first part of (b), noting that θ ≡ ε ( m o d p 0 ) \theta\equiv\varepsilon\pmod{p_0} θ ≡ ε ( mod p 0 ) , we rewrite the identity in Theorem 8.9 (c) for 0 ≤ k ≤ d − 1 0 \leq k \leq d-1 0 ≤ k ≤ d − 1 as
ord p 0 ( g ( χ k ) ( − ℘ ) q k / ( q d − 1 ) − 1 ( ε q k − ε ) ( ε q k − ε q ) ⋯ ( ε q k − ε q k − 1 ) ) > 0. \operatorname{ord}_{p_0}\left(\frac{g(\chi_k)}{(-\wp)^{q^k/(q^d-1)}}-\frac{1}{(\varepsilon^{q^k}-\varepsilon)(\varepsilon^{q^k}-\varepsilon^q)\cdots(\varepsilon^{q^k}-\varepsilon^{q^{k-1}})}\right)>0. ord p 0 ( ( − ℘ ) q k / ( q d − 1 ) g ( χ k ) − ( ε q k − ε ) ( ε q k − ε q ) ⋯ ( ε q k − ε q k − 1 ) 1 ) > 0.
Multiplying over all k k k , we obtain
ord p 0 ( ∏ k = 0 d − 1 g ( χ k ) ( − ℘ ) q k / ( q d − 1 ) − 1 V ( ε ) ) > 0 ⟹ ( 8.21 ) ord p 0 ( det Γ d ( C ) ( − ℘ ) 1 / ( q − 1 ) − 1 ) > 0. \operatorname{ord}_{p_0}\left(\prod_{k=0}^{d-1}\frac{g(\chi_k)}{(-\wp)^{q^k/(q^d-1)}}-\frac{1}{\mathrm{V}(\varepsilon)}\right)>0
\mathrel{\overset{(8.21)}{\Longrightarrow}}
\operatorname{ord}_{p_0}\left(\frac{\det\Gamma_d(\mathrm{C})}{(-\wp)^{1/(q-1)}}-1\right)>0. ord p 0 ( k = 0 ∏ d − 1 ( − ℘ ) q k / ( q d − 1 ) g ( χ k ) − V ( ε ) 1 ) > 0 ⟹ ( 8.21 ) ord p 0 ( ( − ℘ ) 1/ ( q − 1 ) det Γ d ( C ) − 1 ) > 0.
Thus in H p 0 H_{p_0} H p 0 , det Γ d ( C ) = ( − ℘ ) 1 / ( q − 1 ) \det\Gamma_d(\mathrm{C})=(-\wp)^{1/(q-1)} det Γ d ( C ) = ( − ℘ ) 1/ ( q − 1 ) , with root chosen as in Theorem 8.9 (c). Substituting this into (83) yields the second part of (b).
For the first part of (c), it suffices from (a) to verify that sign of det Γ d ( C ) \det\Gamma_d(\mathrm{C}) det Γ d ( C ) is correct. Applying Proposition 8.17 to (80) ,
sgn ( det Γ d ( C ) ) = V ( ε ) ∏ k = 0 d − 1 sgn ( g ( χ k ) ) = ( − 1 ) d / ( q − 1 ) V ( ε ) N ( ℘ ′ ( ε ) ) . \operatorname{sgn}\left(\det\Gamma_d(\mathrm{C})\right)=\mathrm{V}(\varepsilon)\prod_{k=0}^{d-1}\operatorname{sgn}(g(\chi_k))=\frac{(-1)^{d/(q-1)}\mathrm{V}(\varepsilon)}{\mathrm{N}(\wp'(\varepsilon))}. sgn ( det Γ d ( C ) ) = V ( ε ) k = 0 ∏ d − 1 sgn ( g ( χ k )) = N ( ℘ ′ ( ε )) ( − 1 ) d / ( q − 1 ) V ( ε ) .
which simplifies to the desired expression as V ( ε ) 2 = ( − 1 ) d ( d − 1 ) / 2 N ( ℘ ′ ( ε ) ) \mathrm{V}(\varepsilon)^2=(-1)^{d(d-1)/2}\mathrm{N}(\wp'(\varepsilon)) V ( ε ) 2 = ( − 1 ) d ( d − 1 ) /2 N ( ℘ ′ ( ε )) . The expression for det Γ d ∗ ( C ) \det\Gamma_d^*(\mathrm{C}) det Γ d ∗ ( C ) then follows from (83) . □
Summation identities for Dirichlet L L L -series and Gauss-Thakur sums
In [19 ] , Gazda and Maurischat analyzed the Dirichlet L L L -series of Goss type,
L ( χ ε , s ) = ∑ a ∈ A + χ ε ( a ) a s , L(\chi_\varepsilon,s)=\sum_{a\in A_+}\frac{\chi_\varepsilon(a)}{a^s}, L ( χ ε , s ) = a ∈ A + ∑ a s χ ε ( a ) ,
especially in the case s = q i s=q^i s = q i , i ≥ 0 i\geq0 i ≥ 0 , where [19, Eq. (6.1.1) and Pf. of Thm. 6.1 imply
L ( χ ε , q i ) = − π ~ q i ( ε − θ q i ) ⋯ ( ε − θ q ) ( ε − θ ) ℘ ′ ( ε ) g ( χ ε ) , i ≥ 0. (84) L(\chi_\varepsilon,q^i)=-\frac{\widetilde{\pi}^{q^i}}{(\varepsilon-\theta^{q^i})\cdots(\varepsilon-\theta^q)(\varepsilon-\theta)\wp'(\varepsilon)g(\chi_\varepsilon)},\qquad i\geq0. \tag*{(84)} L ( χ ε , q i ) = − ( ε − θ q i ) ⋯ ( ε − θ q ) ( ε − θ ) ℘ ′ ( ε ) g ( χ ε ) π q i , i ≥ 0. (84)
See also [4 p. 225, 36 Cor. 5] . In light of our previous calculations, we can express L ( χ ε , q i ) L(\chi_\varepsilon,q^i) L ( χ ε , q i ) in terms of dual Gauss-Thakur sums.
Applying Theorem 7.7 (a) for the Carlitz module, we have for b ∈ A b\in A b ∈ A and j , k ∈ Z j,k\in\mathbb{Z} j , k ∈ Z ,
∑ k = 0 d − 1 ℘ ′ ( ε q k ) g ∗ [ i ] ( χ k ) g [ j ] ( χ k ; b ) = { ⟨ b / ℘ ⟩ q j ⟨ i − j ⟩ C if i − j ≥ 0 , − ⟨ b / ℘ ⟩ q j ⟨ j − i ⟩ C ∗ if i − j < 0. (85) \sum_{k=0}^{d-1}\wp'(\varepsilon^{q^k})g^{*[i]}(\chi_k)g^{[j]}(\chi_k;b)=
\begin{cases}
\left\langle b/\wp\right\rangle^{q^j}\left\langle i-j\right\rangle_{\mathrm{C}} & \text{if } i-j\geq0,\\
-\left\langle b/\wp\right\rangle^{q^j}\left\langle j-i\right\rangle_{\mathrm{C}^*} & \text{if } i-j<0.
\end{cases} \tag*{(85)} k = 0 ∑ d − 1 ℘ ′ ( ε q k ) g ∗ [ i ] ( χ k ) g [ j ] ( χ k ; b ) = { ⟨ b / ℘ ⟩ q j ⟨ i − j ⟩ C − ⟨ b / ℘ ⟩ q j ⟨ j − i ⟩ C ∗ if i − j ≥ 0 , if i − j < 0. (85)
Now by Pellarin [36, Thm. 1],
∑ a ∈ A + a ( t ) a = − π ~ ( t − θ ) ω C ( t ) = − π ~ Ω C ( t ) , \sum_{a\in A_+}\frac{a(t)}{a}=-\frac{\widetilde{\pi}}{(t-\theta)\omega_{\mathrm{C}}(t)}=-\widetilde{\pi}\Omega_{\mathrm{C}}(t), a ∈ A + ∑ a a ( t ) = − ( t − θ ) ω C ( t ) π = − π Ω C ( t ) ,
and so by Corollary 6.19, we have a dual Gauss-Thakur sum version of (84) for j ≥ 0 j\geq0 j ≥ 0 ,
L ( χ ε , q i ) = ∑ a ∈ A + a ( ε ) a q i = − π ~ q i Ω C ( i ) ( ε ) = ℘ ′ ( ε ) g ∗ [ i ] ( χ ε ) π ~ q i . (86) L(\chi_\varepsilon,q^i)=\sum_{a\in A_+}\frac{a(\varepsilon)}{a q^i}=-\widetilde{\pi}^{q^i}\Omega_{\mathrm{C}}^{(i)}(\varepsilon)=\wp'(\varepsilon)g^{*[i]}(\chi_\varepsilon)\widetilde{\pi}^{q^i}. \tag*{(86)} L ( χ ε , q i ) = a ∈ A + ∑ a q i a ( ε ) = − π q i Ω C ( i ) ( ε ) = ℘ ′ ( ε ) g ∗ [ i ] ( χ ε ) π q i . (86)
Combining this with (85) we have the following.
Proposition 8.30. Let ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^{d}} ε ∈ F q d . Then for b ∈ A b \in A b ∈ A and i , j ∈ Z i,j \in\mathbb{Z} i , j ∈ Z ,
∑ k = 0 d − 1 L ( χ k , q i ) g [ j ] ( χ k ; b ) = { π q i { ⟨ b / ℘ ⟩ i − j } C q j if i − j ≥ 0 , − π ~ q i { ⟨ b / ℘ ⟩ j − i } C ∗ q j if i − j < 0. \sum_{k=0}^{d-1}L(\chi_k,q^i)g^{[j]}(\chi_k;b)=
\begin{cases}
\pi^{q^i}\left\{\begin{matrix}\langle b/\wp\rangle\\ i-j\end{matrix}\right\}_{C}^{q^j} & \text{if }i-j\geq0,\\
-\widetilde{\pi}^{q^i}\left\{\begin{matrix}\langle b/\wp\rangle\\ j-i\end{matrix}\right\}_{C^{*}}^{q^j} & \text{if }i-j<0.
\end{cases} k = 0 ∑ d − 1 L ( χ k , q i ) g [ j ] ( χ k ; b ) = ⎩ ⎨ ⎧ π q i { ⟨ b / ℘ ⟩ i − j } C q j − π q i { ⟨ b / ℘ ⟩ j − i } C ∗ q j if i − j ≥ 0 , if i − j < 0.
As a corollary we evaluate some log-algebraic series introduced by Anderson [1 §4.1] ,
λ m ( x ) : = ∑ a ∈ A + e ( a x ) m a , x ∈ k ∞ , m ≥ 0. \lambda_m(x):=\sum_{a\in A_{+}}\frac{\mathrm{e}(ax)^m}{a},\qquad x\in k_{\infty},\ m\geq0. λ m ( x ) := a ∈ A + ∑ a e ( a x ) m , x ∈ k ∞ , m ≥ 0.
By Anderson’s log-algebraicity theorem for the Carlitz module [1 Prop. 8] , for each m ≥ 0 m\geq0 m ≥ 0 there is S m ( t ) ∈ A [ t ] S_m(t)\in A[t] S m ( t ) ∈ A [ t ] such that for b ∈ A b\in A b ∈ A ,
exp C ( λ m ( b / ℘ ) ) = S m ( e ( b / ℘ ) ) . (87) \exp_C\left(\lambda_m(b/\wp)\right)=S_m\left(\mathrm{e}(b/\wp)\right). \tag*{(87)} exp C ( λ m ( b / ℘ ) ) = S m ( e ( b / ℘ ) ) . (87)
Anderson related λ m ( b / ℘ ) \lambda_m(b/\wp) λ m ( b / ℘ ) to Gauss-Thakur sums [1 §4.6] (see also [30 Eq. (3.10)] , but there is missing negative sign there) by
λ m ( b / ℘ ) = − ∑ χ ∈ F ^ q × ( ∑ a ∈ F q × χ − 1 ( a ) e ( a b / ℘ ) m ) L ( 1 , χ ) . \lambda_m(b/\wp)=-\sum_{\chi\in\widehat{\mathbb{F}}_q^{\times}}\left(\sum_{a\in\mathbb{F}_q^{\times}}\chi^{-1}(a)\mathrm{e}(ab/\wp)^m\right)L(1,\chi). λ m ( b / ℘ ) = − χ ∈ F q × ∑ a ∈ F q × ∑ χ − 1 ( a ) e ( ab / ℘ ) m L ( 1 , χ ) .
When m = q j m=q^j m = q j , the inner sum is simply g [ j ] ( χ ; b ) g^{[j]}(\chi;b) g [ j ] ( χ ; b ) , which vanishes if χ \chi χ is not evaluative.
Corollary 8.32. With notation as above, for j ≥ 0 j\geq0 j ≥ 0 and b ∈ A b\in A b ∈ A with deg b < deg ℘ \deg b<\deg\wp deg b < deg ℘ ,
λ q j ( b / ℘ ) = ∑ k = 0 d − 1 L ( χ k , 1 ) g [ j ] ( χ k ; b ) = { π ~ b / ℘ if j = 0 , − π ~ { b / ℘ j } C ∗ q j if j ≥ 1. \lambda_{q^j}(b/\wp)=\sum_{k=0}^{d-1}L(\chi_k,1)g^{[j]}(\chi_k;b)=
\begin{cases}
\widetilde{\pi}b/\wp& \text{if }j=0,\\
-\widetilde{\pi}\left\{\begin{matrix}b/\wp\\ j\end{matrix}\right\}_{C^{*}}^{q^j} & \text{if }j\geq1.
\end{cases} λ q j ( b / ℘ ) = k = 0 ∑ d − 1 L ( χ k , 1 ) g [ j ] ( χ k ; b ) = ⎩ ⎨ ⎧ π b / ℘ − π { b / ℘ j } C ∗ q j if j = 0 , if j ≥ 1.
Remark 8.33. By [1 Prop. 8(II)] , S 1 ( t ) = t S_1(t)=t S 1 ( t ) = t , and so for j = 0 j=0 j = 0 , the formula for λ 1 ( b / ℘ ) \lambda_1(b/\wp) λ 1 ( b / ℘ ) is immediately consistent with (8.31). When j ≥ 1 j\geq1 j ≥ 1 , [1 Prop. 8(VII)] implies that S q j ( t ) = 0 S_{q^j}(t)=0 S q j ( t ) = 0 identically. The formula for λ q j ( b / ℘ ) \lambda_{q^j}(b/\wp) λ q j ( b / ℘ ) is also consistent with (8.31) by Lemma 2.18.
Remark 8.34. One should compare these results with work of Anglès and Tavares Ribeiro [6 §4.2] , where they consider special values of ℘ \wp ℘ -adic Goss-Dirichlet L L L -functions, also in terms of Gauss-Thakur sums and log-algebraicity.
Applications to general Drinfeld modules
The determinants of Γ d ( ϕ ) \Gamma_d(\phi) Γ d ( ϕ ) and Γ d ∗ ( ϕ ∗ ) \Gamma_d^{*}(\phi^{*}) Γ d ∗ ( ϕ ∗ )
For an intermediate field k ⊆ K ⊆ C ∞ k\subseteq K\subseteq C_{\infty} k ⊆ K ⊆ C ∞ , fix a rank r r r Drinfeld module ϕ : A → K [ τ ] \phi:A\to K[\tau] ϕ : A → K [ τ ] as in (1.2). We wish to compute the determinants,
det Γ d ( ϕ ) = det Q ( 1 / ℘ ′ ) det ( Υ ( ε ) ⋯ Υ ( r ( d − 1 ) ) ( ε ) ⋮ ⋱ ⋮ Υ ( ε q d − 1 ) ⋯ Υ ( r ( d − 1 ) ) ( ε q d − 1 ) ) \det\Gamma_d(\phi)=\det Q(1/\wp')\det\begin{pmatrix}
\Upsilon(\varepsilon)&\cdots&\Upsilon^{(r(d-1))}(\varepsilon)\\
\vdots&\ddots&\vdots\\
\Upsilon(\varepsilon^{q^{d-1}})&\cdots&\Upsilon^{(r(d-1))}(\varepsilon^{q^{d-1}})
\end{pmatrix} det Γ d ( ϕ ) = det Q ( 1/ ℘ ′ ) det Υ ( ε ) ⋮ Υ ( ε q d − 1 ) ⋯ ⋱ ⋯ Υ ( r ( d − 1 )) ( ε ) ⋮ Υ ( r ( d − 1 )) ( ε q d − 1 )
and
det Γ d ∗ ( ϕ ∗ ) = det Q ( 1 / ℘ ′ ) det ( J Ψ ( r − 1 ) ( ε ) ⋯ J Ψ ( r − 1 ) ( ε q d − 1 ) ⋮ ⋮ J Ψ ( d r − 1 ) ( ε ) ⋯ J Ψ ( d r − 1 ) ( ε q d − 1 ) ) . \det\Gamma_d^*(\phi^*)=\det Q(1/\wp')\det\begin{pmatrix}
J\Psi^{(r-1)}(\varepsilon)&\cdots&J\Psi^{(r-1)}(\varepsilon^{q^{d-1}})\\
\vdots&&\vdots\\
J\Psi^{(dr-1)}(\varepsilon)&\cdots&J\Psi^{(dr-1)}(\varepsilon^{q^{d-1}})
\end{pmatrix}. det Γ d ∗ ( ϕ ∗ ) = det Q ( 1/ ℘ ′ ) det J Ψ ( r − 1 ) ( ε ) ⋮ J Ψ ( d r − 1 ) ( ε ) ⋯ ⋯ J Ψ ( r − 1 ) ( ε q d − 1 ) ⋮ J Ψ ( d r − 1 ) ( ε q d − 1 ) .
which we recall from (7.4)–(7.5). By (26) , det Υ ( 1 ) = det Υ det Θ T = ( − 1 ) r − 1 ( t − θ ) / b r ⋅ det Υ \det\Upsilon^{(1)}=\det\Upsilon\det\Theta^{\mathrm{T}}=(-1)^{r-1}(t-\theta)/b_r\cdot\det\Upsilon det Υ ( 1 ) = det Υ det Θ T = ( − 1 ) r − 1 ( t − θ ) / b r ⋅ det Υ . Therefore,
det Υ = μ ω C (88) \det\Upsilon=\mu\omega_C \tag*{(88)} det Υ = μ ω C (88)
for some μ ∈ K ‾ \mu\in\overline{K} μ ∈ K with μ q − 1 = ( − 1 ) r − 1 / b r \mu^{q-1}=(-1)^{r-1}/b_r μ q − 1 = ( − 1 ) r − 1 / b r . Indeed, both ω C \omega_C ω C and ( det Υ ) / μ (\det\Upsilon)/\mu ( det Υ ) / μ satisfy the same difference equation, f ( 1 ) = ( t − θ ) f f^{(1)}=(t-\theta)f f ( 1 ) = ( t − θ ) f , and since Υ ∈ GL r ( T ) \Upsilon\in\operatorname{GL}_r(\mathbb{T}) Υ ∈ GL r ( T ) by [35 §4.2] , the two functions can differ by only an F q \mathbb{F}_q F q -multiple (e.g., see [3 Lem. 2.5.4] ). Also by Lemma 3.13,
det Ψ = ( − 1 ) r ( r − 1 ) / 2 b r ( 1 − q r ) / ( 1 − q ) ( θ − t ) det Υ = ( − 1 ) r ( r + 1 ) / 2 Ω C μ q 1 / ( r − 1 ) . (89) \det\Psi=\frac{(-1)^{r(r-1)/2}b_r^{(1-q^r)/(1-q)}}{(\theta-t)\det\Upsilon}
=\frac{(-1)^{r(r+1)/2}\Omega_C}{\mu^{q^{1/(r-1)}}}. \tag*{(89)} det Ψ = ( θ − t ) det Υ ( − 1 ) r ( r − 1 ) /2 b r ( 1 − q r ) / ( 1 − q ) = μ q 1/ ( r − 1 ) ( − 1 ) r ( r + 1 ) /2 Ω C . (89)
Our main result on det Γ d ( ϕ ) \det\Gamma_d(\phi) det Γ d ( ϕ ) and det Γ d ∗ ( ϕ ∗ ) \det\Gamma_d^*(\phi^*) det Γ d ∗ ( ϕ ∗ ) is Theorem 9.12 below.
The determinants of G ϕ ( χ ε ) \mathcal{G}_{\phi}(\chi_\varepsilon) G ϕ ( χ ε ) and G ϕ ∗ ∗ ( χ ε ) \mathcal{G}_{\phi^*}^*(\chi_\varepsilon) G ϕ ∗ ∗ ( χ ε )
Recall G ϕ ( χ ε ) \mathcal{G}_{\phi}(\chi_\varepsilon) G ϕ ( χ ε ) and G ϕ ∗ ∗ ( χ ε ) \mathcal{G}_{\phi^*}^*(\chi_\varepsilon) G ϕ ∗ ∗ ( χ ε ) from (1.11)–(1.12), where we have included ϕ \phi ϕ and ϕ ∗ \phi^* ϕ ∗ in the notation to indicate the dependence on ϕ \phi ϕ , and let g C ( χ ε ) g_C(\chi_\varepsilon) g C ( χ ε ) and g C ∗ ∗ ( χ ε ) g_{C^*}^*(\chi_\varepsilon) g C ∗ ∗ ( χ ε ) denote the Gauss–Thakur sums for the Carlitz module as in §8.5. Then Corollary 6.19, together with (60) and (9.2)–(9.3), implies for 0 ≤ j ≤ d − 1 0\leq j\leq d-1 0 ≤ j ≤ d − 1 ,
det G ϕ ( χ j ) = det Υ ( ε q j ) ℘ ′ ( ε q j ) = μ g C ( χ j ) ℘ ′ ( ε q j ) r − 1 . (90) \det\mathcal{G}_{\phi}(\chi_j)=\frac{\det\Upsilon(\varepsilon^{q^j})}{\wp'(\varepsilon^{q^j})}
=\frac{\mu g_C(\chi_j)}{\wp'(\varepsilon^{q^j})^{r-1}}. \tag*{(90)} det G ϕ ( χ j ) = ℘ ′ ( ε q j ) det Υ ( ε q j ) = ℘ ′ ( ε q j ) r − 1 μ g C ( χ j ) . (90)
and
det G ϕ ∗ ∗ ( χ j ) = ( − 1 ) r ( r − 1 ) / 2 det Ψ ( r − 1 ) ( ε q j ) ℘ ′ ( ε q j ) r = ( − 1 ) r − 1 g C ∗ ∗ [ r − 1 ] ( χ j ) μ ℘ ′ ( ε q j ) r − 1 . (91) \det\mathcal{G}_{\phi^*}^*(\chi_j)=\frac{(-1)^{r(r-1)/2}\det\Psi^{(r-1)}(\varepsilon^{q^j})}{\wp'(\varepsilon^{q^j})^r}
=\frac{(-1)^{r-1}g_{C^*}^{*[r-1]}(\chi_j)}{\mu\wp'(\varepsilon^{q^j})^{r-1}}. \tag*{(91)} det G ϕ ∗ ∗ ( χ j ) = ℘ ′ ( ε q j ) r ( − 1 ) r ( r − 1 ) /2 det Ψ ( r − 1 ) ( ε q j ) = μ ℘ ′ ( ε q j ) r − 1 ( − 1 ) r − 1 g C ∗ ∗ [ r − 1 ] ( χ j ) . (91)
Proposition 9.7. With notation as above,
∏ j = 0 d − 1 det G ϕ ( χ j ) q − 1 = ( − 1 ) r d ℘ b r d and ∏ j = 0 d − 1 det G ϕ ∗ ∗ ( χ j ) q − 1 = ( − 1 ) r d b r d ℘ q r . \prod_{j=0}^{d-1}\det\mathcal{G}_{\phi}(\chi_j)^{q-1}=\frac{(-1)^{rd}\wp}{b_r^d}
\quad\text{and}\quad
\prod_{j=0}^{d-1}\det\mathcal{G}_{\phi^*}^*(\chi_j)^{q-1}=\frac{(-1)^{rd}b_r^d}{\wp^{q^r}}. j = 0 ∏ d − 1 det G ϕ ( χ j ) q − 1 = b r d ( − 1 ) r d ℘ and j = 0 ∏ d − 1 det G ϕ ∗ ∗ ( χ j ) q − 1 = ℘ q r ( − 1 ) r d b r d .
Proof. Multiplying (90) over the roots of ℘ \wp ℘ and raising to the ( q − 1 ) (q-1) ( q − 1 ) -st power, we obtain
∏ j = 0 d − 1 det G ϕ ( χ j ) q − 1 = μ ( q − 1 ) d N ( ℘ ′ ( ε ) ) ( q − 1 ) ( r − 1 ) ∏ j = 0 d − 1 g C ( χ j ) q − 1 = ( − 1 ) r d ℘ b r d , \prod_{j=0}^{d-1}\det\mathcal{G}_{\phi}(\chi_j)^{q-1}
=\frac{\mu^{(q-1)d}}{N(\wp'(\varepsilon))^{(q-1)(r-1)}}\prod_{j=0}^{d-1}g_C(\chi_j)^{q-1}
=\frac{(-1)^{rd}\wp}{b_r^d}, j = 0 ∏ d − 1 det G ϕ ( χ j ) q − 1 = N ( ℘ ′ ( ε ) ) ( q − 1 ) ( r − 1 ) μ ( q − 1 ) d j = 0 ∏ d − 1 g C ( χ j ) q − 1 = b r d ( − 1 ) r d ℘ ,
where the final equality follows from Theorem 8.9(b) and that μ q − 1 = ( − 1 ) r − 1 / b r \mu^{q-1}=(-1)^{r-1}/b_r μ q − 1 = ( − 1 ) r − 1 / b r and N ( ℘ ′ ( ε ) ) ∈ F q × N(\wp'(\varepsilon))\in\mathbb{F}_q^\times N ( ℘ ′ ( ε )) ∈ F q × . For the G ϕ ∗ ∗ \mathcal{G}_{\phi^*}^* G ϕ ∗ ∗ case, note that for 0 ≤ j ≤ d − 1 0\leq j\leq d-1 0 ≤ j ≤ d − 1 , g C ∗ ∗ [ r − 1 ] ( χ ε q j ) = g C ∗ ∗ ( χ ε q j − r + 1 ) q r g_{C^*}^{*[r-1]}(\chi_{\varepsilon^{q^j}})=g_{C^*}^*(\chi_{\varepsilon^{q^j-r+1}})^{q^r} g C ∗ ∗ [ r − 1 ] ( χ ε q j ) = g C ∗ ∗ ( χ ε q j − r + 1 ) q r , as in the proof of Corollary 6.18. Then using (91) ,
∏ j = 0 d − 1 det G ϕ ∗ ∗ ( χ j ) q − 1 = 1 μ ( q − 1 ) d N ( ℘ ′ ( ε ) ) ( q − 1 ) ( r − 1 ) ∏ j = 0 d − 1 g C ∗ ∗ ( χ j ) ( q − 1 ) q r − 1 = ( − 1 ) r d b r d ℘ q r , \prod_{j=0}^{d-1}\det\mathcal{G}_{\phi^*}^*(\chi_j)^{q-1}
=\frac{1}{\mu^{(q-1)d}N(\wp'(\varepsilon))^{(q-1)(r-1)}}\prod_{j=0}^{d-1}g_{C^*}^*(\chi_j)^{(q-1)q^{r-1}}
=\frac{(-1)^{rd}b_r^d}{\wp^{q^r}}, j = 0 ∏ d − 1 det G ϕ ∗ ∗ ( χ j ) q − 1 = μ ( q − 1 ) d N ( ℘ ′ ( ε ) ) ( q − 1 ) ( r − 1 ) 1 j = 0 ∏ d − 1 g C ∗ ∗ ( χ j ) ( q − 1 ) q r − 1 = ℘ q r ( − 1 ) r d b r d ,
where the last equality follows from Theorem 8.10(b). □
We now analyze det Γ d ( ϕ ) \det\Gamma_d(\phi) det Γ d ( ϕ ) in a manner similar to Theorem 8.24. By (26) , we find
Υ ( k ) = Υ Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( k − 1 ) , k ≥ 0. (92) \Upsilon^{(k)}=\Upsilon\Theta^{\mathrm{T}}(\Theta^{\mathrm{T}})^{(1)}\cdots(\Theta^{\mathrm{T}})^{(k-1)},\qquad k\geq0. \tag*{(92)} Υ ( k ) = Υ Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( k − 1 ) , k ≥ 0. (92)
Lemma 9.9. As matrices in Mat r ( K [ t ] ) \operatorname{Mat}_{r}(K[t]) Mat r ( K [ t ]) , we have
Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) = B 1 ⋅ t + C 1 , \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(r-1)}=\mathbf{B}_{1}\cdot t+\mathbf{C}_{1}, Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) = B 1 ⋅ t + C 1 , where B 1 , C 1 ∈ Mat r ( K ) \mathbf{B}_{1},\mathbf{C}_{1}\in\operatorname{Mat}_{r}(K) B 1 , C 1 ∈ Mat r ( K ) and B 1 \mathbf{B}_{1} B 1 is upper triangular with [ B 1 ] i i = 1 / b r q i − 1 [\mathbf{B}_{1}]_{ii}=1/b_{r}^{q^{i-1}} [ B 1 ] ii = 1/ b r q i − 1 for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r . Moreover, for k ≥ 1 k\geq1 k ≥ 1 ,
Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r k − 1 ) = B k ⋅ t k + C k ( t ) , \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(rk-1)}=\mathbf{B}_{k}\cdot t^{k}+\mathbf{C}_{k}(t), Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r k − 1 ) = B k ⋅ t k + C k ( t ) , where B k = B 1 B 1 ( r ) ⋯ B 1 ( r ( k − 1 ) ) \mathbf{B}_{k}=\mathbf{B}_{1}\mathbf{B}_{1}^{(r)}\cdots\mathbf{B}_{1}^{(r(k-1))} B k = B 1 B 1 ( r ) ⋯ B 1 ( r ( k − 1 )) is upper triangular with [ B k ] i i = b r − ( q r k + i − 1 − q i − 1 ) / ( q r − 1 ) [\mathbf{B}_{k}]_{ii}=b_{r}^{-(q^{rk+i-1}-q^{i-1})/(q^{r}-1)} [ B k ] ii = b r − ( q r k + i − 1 − q i − 1 ) / ( q r − 1 ) for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r and where C k ( t ) ∈ Mat r ( K [ t ] ) \mathbf{C}_{k}(t)\in\operatorname{Mat}_{r}(K[t]) C k ( t ) ∈ Mat r ( K [ t ]) has entries with degree < k <k < k in t t t .
Proof. (cf. [27, Ex. 2.5.16]) We claim that for 0 ≤ ℓ ≤ r − 1 0\leq\ell\leq r-1 0 ≤ ℓ ≤ r − 1 ,
Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( ℓ ) = V ℓ + W ℓ , \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(\ell)}=\mathbf{V}_{\ell}+\mathbf{W}_{\ell}, Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( ℓ ) = V ℓ + W ℓ , where V ℓ , W ℓ ∈ Mat r ( K [ t ] ) \mathbf{V}_{\ell},\mathbf{W}_{\ell}\in\operatorname{Mat}_{r}(K[t]) V ℓ , W ℓ ∈ Mat r ( K [ t ]) have the form
V ℓ = ( 0 ⋯ t / b r ⋯ 1 ⋯ ⋯ t / b r q ℓ ⋮ ⋱ ⋱ ⋮ 0 ⋯ 1 0 ) , W ℓ = ( 0 ⋯ 0 ∗ ∗ ⋯ ∗ ⋮ ⋮ ⋮ ⋱ ∗ ⋮ ⋮ ⋮ ⋱ ⋱ ⋮ 0 ⋯ 0 ∗ ⋯ ⋯ ∗ ) , \mathbf{V}_{\ell}=
\begin{pmatrix}
0 & \cdots& t/b_{r} & \cdots\\
1 & \cdots& \cdots& t/b_{r}^{q^{\ell}}\\
\vdots& \ddots& \ddots& \vdots\\
0 & \cdots& 1 & 0
\end{pmatrix},
\qquad
\mathbf{W}_{\ell}=
\begin{pmatrix}
0 & \cdots& 0 & * & * & \cdots& *\\
\vdots& & \vdots& \vdots& \ddots& & *\\
\vdots& & \vdots& \vdots& \ddots& \ddots& \vdots\\
0 & \cdots& 0 & * & \cdots& \cdots& *
\end{pmatrix}, V ℓ = 0 1 ⋮ 0 ⋯ ⋯ ⋱ ⋯ t / b r ⋯ ⋱ 1 ⋯ t / b r q ℓ ⋮ 0 , W ℓ = 0 ⋮ ⋮ 0 ⋯ ⋯ 0 ⋮ ⋮ 0 ∗ ⋮ ⋮ ∗ ∗ ⋱ ⋱ ⋯ ⋯ ⋱ ⋯ ∗ ∗ ⋮ ∗ , and where the entries of W ℓ \mathbf{W}_{\ell} W ℓ have degree at most 1 1 1 in t t t and if j − i ≤ r − 1 − ℓ j-i\leq r-1-\ell j − i ≤ r − 1 − ℓ then [ W ℓ ] i j ∈ K [\mathbf{W}_{\ell}]_{ij}\in K [ W ℓ ] ij ∈ K . That is, the entries in the upper triangle in the diagram above for W ℓ \mathbf{W}_{\ell} W ℓ can be linear in t t t , and the ones in the lower trapezoid are all in K K K . Note that in V ℓ \mathbf{V}_{\ell} V ℓ the non-zero super-diagonal starts in the r − ℓ r-\ell r − ℓ entry of the first row, and the non-zero sub-diagonal starts in the ℓ + 2 \ell+2 ℓ + 2 entry of the first column. Thus V r − 1 \mathbf{V}_{r-1} V r − 1 is a diagonal matrix with diagonal entries t / b r , t / b r q , … , t / b r q r − 1 t/b_{r},t/b_{r}^{q},\ldots,t/b_{r}^{q^{r-1}} t / b r , t / b r q , … , t / b r q r − 1 . Now since
Θ T = ( 0 ⋯ 0 ( t − θ ) / b r 1 ⋯ 0 − b 1 / b r ⋮ ⋱ ⋮ ⋮ 0 ⋯ 1 − b r − 1 / b r ) = ( t / b r 1 ⋱ 1 ) + ( 0 ⋯ 0 − θ / b r ⋮ ⋮ ⋮ 0 ⋯ 0 − b r − 1 / b r ) , \Theta^{\mathrm T}=
\begin{pmatrix}
0 & \cdots& 0 & (t-\theta)/b_{r}\\
1 & \cdots& 0 & -b_{1}/b_{r}\\
\vdots& \ddots& \vdots& \vdots\\
0 & \cdots& 1 & -b_{r-1}/b_{r}
\end{pmatrix}
=
\begin{pmatrix}
t/b_{r} & & \\
1 & \ddots& \\
& & 1
\end{pmatrix}
+
\begin{pmatrix}
0 & \cdots& 0 & -\theta/b_{r}\\
\vdots& & \vdots& \vdots\\
0 & \cdots& 0 & -b_{r-1}/b_{r}
\end{pmatrix}, Θ T = 0 1 ⋮ 0 ⋯ ⋯ ⋱ ⋯ 0 0 ⋮ 1 ( t − θ ) / b r − b 1 / b r ⋮ − b r − 1 / b r = t / b r 1 ⋱ 1 + 0 ⋮ 0 ⋯ ⋯ 0 ⋮ 0 − θ / b r ⋮ − b r − 1 / b r , we see that the claim is true for ℓ = 0 \ell=0 ℓ = 0 . Suppose that it is true for ℓ < r − 1 \ell<r-1 ℓ < r − 1 . Then
Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( ℓ + 1 ) = ( V ℓ + W ℓ ) ( V 0 ( ℓ + 1 ) + W 0 ( ℓ + 1 ) ) . \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(\ell+1)}
=(\mathbf{V}_{\ell}+\mathbf{W}_{\ell})(\mathbf{V}_{0}^{(\ell+1)}+\mathbf{W}_{0}^{(\ell+1)}). Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( ℓ + 1 ) = ( V ℓ + W ℓ ) ( V 0 ( ℓ + 1 ) + W 0 ( ℓ + 1 ) ) . One checks that V ℓ V 0 ( ℓ + 1 ) = V ℓ + 1 \mathbf{V}_{\ell}\mathbf{V}_{0}^{(\ell+1)}=\mathbf{V}_{\ell+1} V ℓ V 0 ( ℓ + 1 ) = V ℓ + 1 . Letting W ℓ + 1 : = W ℓ V 0 ( ℓ + 1 ) + V ℓ W 0 ( ℓ + 1 ) + W ℓ W 0 ( ℓ + 1 ) \mathbf{W}_{\ell+1}:=\mathbf{W}_{\ell}\mathbf{V}_{0}^{(\ell+1)}+\mathbf{V}_{\ell}\mathbf{W}_{0}^{(\ell+1)}+\mathbf{W}_{\ell}\mathbf{W}_{0}^{(\ell+1)} W ℓ + 1 := W ℓ V 0 ( ℓ + 1 ) + V ℓ W 0 ( ℓ + 1 ) + W ℓ W 0 ( ℓ + 1 ) , it is straightforward to verify using the induction hypothesis that each of W ℓ V 0 ( ℓ + 1 ) \mathbf{W}_{\ell}\mathbf{V}_{0}^{(\ell+1)} W ℓ V 0 ( ℓ + 1 ) , V ℓ W 0 ( ℓ + 1 ) \mathbf{V}_{\ell}\mathbf{W}_{0}^{(\ell+1)} V ℓ W 0 ( ℓ + 1 ) , and W ℓ W 0 ( ℓ + 1 ) \mathbf{W}_{\ell}\mathbf{W}_{0}^{(\ell+1)} W ℓ W 0 ( ℓ + 1 ) satisfy the defining conditions for W ℓ + 1 \mathbf{W}_{\ell+1} W ℓ + 1 , and the claim is proved. Therefore, Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) = V r − 1 + W r − 1 = : B 1 ⋅ t + C 1 \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(r-1)}=\mathbf{V}_{r-1}+\mathbf{W}_{r-1}=:\mathbf{B}_{1}\cdot t+\mathbf{C}_{1} Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) = V r − 1 + W r − 1 =: B 1 ⋅ t + C 1 , where B 1 , C 1 ∈ Mat r ( K ) \mathbf{B}_{1},\mathbf{C}_{1}\in\operatorname{Mat}_{r}(K) B 1 , C 1 ∈ Mat r ( K ) satisfy the desired conditions in the lemma. The second part of the lemma then follows from the first by taking r r r -twists of Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) \Theta^{\mathrm T}(\Theta^{\mathrm T})^{(1)}\cdots(\Theta^{\mathrm T})^{(r-1)} Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( r − 1 ) and multiplying. □ \square □
Using the matrices defined in Lemma 9.9, for variables t 1 , … , t d t_1,\ldots,t_d t 1 , … , t d , let
D ( t 1 , … , t d ) : = ( I r B 1 t 1 + C 1 ⋯ B d − 1 t 1 d − 1 + C d − 1 ( t 1 ) I r B 1 t 2 + C 1 ⋯ B d − 1 t 2 d − 1 + C d − 1 ( t 2 ) ⋮ ⋮ ⋮ I r B 1 t d + C 1 ⋯ B d − 1 t d d − 1 + C d − 1 ( t d ) ) ∈ Mat r d ( K [ t 1 , … , t d ] ) , \mathbf{D}(t_1,\ldots,t_d) := \begin{pmatrix}
I_r & \mathbf{B}_1t_1+\mathbf{C}_1 & \cdots& \mathbf{B}_{d-1}t_1^{d-1}+\mathbf{C}_{d-1}(t_1) \\
I_r & \mathbf{B}_1t_2+\mathbf{C}_1 & \cdots& \mathbf{B}_{d-1}t_2^{d-1}+\mathbf{C}_{d-1}(t_2) \\
\vdots& \vdots& & \vdots\\
I_r & \mathbf{B}_1t_d+\mathbf{C}_1 & \cdots& \mathbf{B}_{d-1}t_d^{d-1}+\mathbf{C}_{d-1}(t_d)
\end{pmatrix} \in\operatorname{Mat}_{rd}(K[t_1,\ldots,t_d]), D ( t 1 , … , t d ) := I r I r ⋮ I r B 1 t 1 + C 1 B 1 t 2 + C 1 ⋮ B 1 t d + C 1 ⋯ ⋯ ⋯ B d − 1 t 1 d − 1 + C d − 1 ( t 1 ) B d − 1 t 2 d − 1 + C d − 1 ( t 2 ) ⋮ B d − 1 t d d − 1 + C d − 1 ( t d ) ∈ Mat r d ( K [ t 1 , … , t d ]) ,
and let
Δ = Δ ( t 1 , … , t d ) : = det D ( t 1 , … , t d ) . \Delta=\Delta(t_1,\ldots,t_d):=\det\mathbf{D}(t_1,\ldots,t_d). Δ = Δ ( t 1 , … , t d ) := det D ( t 1 , … , t d ) .
Similar to the simplifications in (81) , as each B k \mathbf{B}_k B k , 1 ≤ k ≤ d − 1 1\leq k\leq d-1 1 ≤ k ≤ d − 1 , is upper triangular with non-zero diagonal entries, we can perform column manipulations of this determinant to eliminate the C k ( t i ) \mathbf{C}_k(t_i) C k ( t i ) terms as their entries have degrees < k <k < k in t i t_i t i , and we find
Δ = det ( I r B 1 t 1 ⋯ B d − 1 t 1 d − 1 I r B 1 t 2 ⋯ B d − 1 t 2 d − 1 ⋮ ⋮ ⋮ I r B 1 t d ⋯ B d − 1 t d d − 1 ) = det ( I r t 1 I r ⋯ t 1 d − 1 I r I r t 2 I r ⋯ t 2 d − 1 I r ⋮ ⋮ ⋮ I r t d I r ⋯ t d d − 1 I r ) det ( I r 0 ⋯ 0 0 B 1 ⋯ 0 ⋮ ⋮ ⋱ ⋮ 0 0 ⋯ B d − 1 ) . \Delta=\det\begin{pmatrix}
I_r & \mathbf{B}_1t_1 & \cdots& \mathbf{B}_{d-1}t_1^{d-1}\\
I_r & \mathbf{B}_1t_2 & \cdots& \mathbf{B}_{d-1}t_2^{d-1}\\
\vdots& \vdots& & \vdots\\
I_r & \mathbf{B}_1t_d & \cdots& \mathbf{B}_{d-1}t_d^{d-1}
\end{pmatrix}
=\det\begin{pmatrix}
I_r & t_1I_r & \cdots& t_1^{d-1}I_r\\
I_r & t_2I_r & \cdots& t_2^{d-1}I_r\\
\vdots& \vdots& & \vdots\\
I_r & t_dI_r & \cdots& t_d^{d-1}I_r
\end{pmatrix}
\det\begin{pmatrix}
I_r & 0 & \cdots& 0\\
0 & \mathbf{B}_1 & \cdots& 0\\
\vdots& \vdots& \ddots& \vdots\\
0 & 0 & \cdots& \mathbf{B}_{d-1}
\end{pmatrix}. Δ = det I r I r ⋮ I r B 1 t 1 B 1 t 2 ⋮ B 1 t d ⋯ ⋯ ⋯ B d − 1 t 1 d − 1 B d − 1 t 2 d − 1 ⋮ B d − 1 t d d − 1 = det I r I r ⋮ I r t 1 I r t 2 I r ⋮ t d I r ⋯ ⋯ ⋯ t 1 d − 1 I r t 2 d − 1 I r ⋮ t d d − 1 I r det I r 0 ⋮ 0 0 B 1 ⋮ 0 ⋯ ⋯ ⋱ ⋯ 0 0 ⋮ B d − 1 .
After performing an equal number of column and row swaps on the first determinant after the first equality, we then obtain
Δ = V ( t 1 , … , t d ) r det B 1 ⋯ det B d − 1 = b r − ( q r d − 1 ) / ( q r − 1 ) − d / ( q − 1 ) ⋅ V ( t 1 , … , t d ) r , \Delta=V(t_1,\ldots,t_d)^r\det\mathbf{B}_1\cdots\det\mathbf{B}_{d-1}=b_r^{-(q^{rd}-1)/(q^r-1)-d/(q-1)}\cdot V(t_1,\ldots,t_d)^r, Δ = V ( t 1 , … , t d ) r det B 1 ⋯ det B d − 1 = b r − ( q r d − 1 ) / ( q r − 1 ) − d / ( q − 1 ) ⋅ V ( t 1 , … , t d ) r ,
where V ( t 1 , … , t d ) V(t_1,\ldots,t_d) V ( t 1 , … , t d ) is the Vandermonde determinant in t 1 , … , t d t_1,\ldots,t_d t 1 , … , t d , and where we have used that det B k = b r ( q r k − 1 ) / ( q r − 1 ) \det\mathbf{B}_k=b_r^{(q^{rk}-1)/(q^r-1)} det B k = b r ( q r k − 1 ) / ( q r − 1 ) by Lemma 9.9. Substituting b r = ( − 1 ) r − 1 / μ q − 1 b_r=(-1)^{r-1}/\mu^{q-1} b r = ( − 1 ) r − 1 / μ q − 1 (and keeping track of signs), we obtain
Δ = μ ( ( q r d − 1 ) / ( q r − 1 ) − d ) ⋅ V ( t 1 , … , t d ) r . (93) \Delta=\mu^{((q^{rd}-1)/(q^r-1)-d)}\cdot V(t_1,\ldots,t_d)^r. \tag*{(93)} Δ = μ (( q r d − 1 ) / ( q r − 1 ) − d ) ⋅ V ( t 1 , … , t d ) r . (93)
For 0 ≤ j ≤ d − 1 0\leq j\leq d-1 0 ≤ j ≤ d − 1 , once we divide (92) by φ ′ ( ε q j ) \varphi'(\varepsilon^{q^j}) φ ′ ( ε q j ) , we obtain from Corollary 6.19,
G ϕ [ k ] ( χ j ) = G ϕ ( χ j ) ⋅ Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( k − 1 ) ∣ t = ε q j , k ≥ 0. \mathcal{G}_{\phi}^{[k]}(\chi_j)=\mathcal{G}_{\phi}(\chi_j)\cdot\Theta^{\mathsf{T}}(\Theta^{\mathsf{T}})^{(1)}\cdots(\Theta^{\mathsf{T}})^{(k-1)}\bigg|_{t=\varepsilon^{q^j}},\qquad k\geq0. G ϕ [ k ] ( χ j ) = G ϕ ( χ j ) ⋅ Θ T ( Θ T ) ( 1 ) ⋯ ( Θ T ) ( k − 1 ) t = ε q j , k ≥ 0.
Taking Γ d ( ϕ ) \Gamma_d(\phi) Γ d ( ϕ ) written in block form as in (9) , this then implies
Γ d ( ϕ ) = ( G ϕ ( χ 0 ) ⋯ 0 ⋮ ⋱ ⋮ 0 ⋯ G ϕ ( χ d − 1 ) ) D ( ε , … , ε q d − 1 ) . (94) \Gamma_d(\phi)=\begin{pmatrix}
\mathcal{G}_{\phi}(\chi_0)&\cdots&0\\
\vdots&\ddots&\vdots\\
0&\cdots&\mathcal{G}_{\phi}(\chi_{d-1})
\end{pmatrix}\mathbf{D}(\varepsilon,\ldots,\varepsilon^{q^{d-1}}). \tag*{(94)} Γ d ( ϕ ) = G ϕ ( χ 0 ) ⋮ 0 ⋯ ⋱ ⋯ 0 ⋮ G ϕ ( χ d − 1 ) D ( ε , … , ε q d − 1 ) . (94)
Theorem 9.12. Let φ ∈ A + \varphi\in A_+ φ ∈ A + be irreducible of degree d d d , and fix a root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^d} ε ∈ F q d of φ \varphi φ . Define Γ d ( ϕ ) , Γ d ∗ ( ϕ ∗ ) ∈ GL d ( K ‾ ) \Gamma_d(\phi),\ \Gamma_d^*(\phi^*)\in\operatorname{GL}_d(\overline{K}) Γ d ( ϕ ) , Γ d ∗ ( ϕ ∗ ) ∈ GL d ( K ) as in §7.2.
(a) We have
det Γ d ( ϕ ) q − 1 = ( − 1 ) r J φ b r ( q r d − 1 ) / ( q r − 1 ) and det Γ d ∗ ( ϕ ∗ ) q − 1 = ( − 1 ) r b r ( q r d − 1 ) / ( q r − 1 ) J q r d . \det\Gamma_d(\phi)^{q-1}=\frac{(-1)^r\mathcal{J}_{\varphi}}{b_r^{(q^{rd}-1)/(q^r-1)}}\qquad\text{and}\qquad\det\Gamma_d^*(\phi^*)^{q-1}=\frac{(-1)^rb_r^{(q^{rd}-1)/(q^r-1)}}{\mathcal{J}^{q^{rd}}}. det Γ d ( ϕ ) q − 1 = b r ( q r d − 1 ) / ( q r − 1 ) ( − 1 ) r J φ and det Γ d ∗ ( ϕ ∗ ) q − 1 = J q r d ( − 1 ) r b r ( q r d − 1 ) / ( q r − 1 ) . (b) *By choosing ( − 1 ) 1 / ( q − 1 ) (-1)^{1/(q-1)} ( − 1 ) 1/ ( q − 1 ) as in (77) and J 1 / ( q − 1 ) \mathcal{J}^{1/(q-1)} J 1/ ( q − 1 ) to have sign 1 1 1 ,
det Γ d ( ϕ ) = ( − 1 ) r d ( d − 1 ) / 2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) V ( ε ) r ⋅ J 1 / ( q − 1 ) \det\Gamma_d(\phi)=\frac{(-1)^{rd(d-1)/2}\cdot(-1)^{d/(q-1)}\cdot\mu^{(q^{rd}-1)/(q^r-1)}}{V(\varepsilon)^r}\cdot\mathcal{J}^{1/(q-1)} det Γ d ( ϕ ) = V ( ε ) r ( − 1 ) r d ( d − 1 ) /2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ J 1/ ( q − 1 ) and
det Γ d ∗ ( ϕ ∗ ) = 1 ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ V ( ε ) r ⋅ 1 J q r d / ( q − 1 ) . \det\Gamma_d^*(\phi^*)=\frac{1}{(-1)^{d/(q-1)}\cdot\mu^{(q^{rd}-1)/(q^r-1)}\cdot V(\varepsilon)^r}\cdot\frac{1}{\mathcal{J}^{q^{rd}/(q-1)}}. det Γ d ∗ ( ϕ ∗ ) = ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ V ( ε ) r 1 ⋅ J q r d / ( q − 1 ) 1 .
Proof. Let V ( ε ) V(\varepsilon) V ( ε ) be defined as in (81) . Then combining (93) –(94) , we have
det Γ d ( ϕ ) = μ ( q r d − 1 ) / ( q r − 1 ) − d V ( ε ) r ∏ j = 0 d − 1 det Φ ϕ ( χ j ) . (95) \det\Gamma_d(\phi)=\mu^{(q^{rd}-1)/(q^r-1)-d}V(\varepsilon)^r\prod_{j=0}^{d-1}\det\Phi_\phi(\chi_j). \tag*{(95)} det Γ d ( ϕ ) = μ ( q r d − 1 ) / ( q r − 1 ) − d V ( ε ) r j = 0 ∏ d − 1 det Φ ϕ ( χ j ) . (95)
We also recall from Theorem 7.7(c) that
det Γ d ( ϕ ) ⋅ det Γ d ∗ ( ϕ ∗ ) ⋅ N ( g ′ ( ε ) ) r = det P d ( 1 / g ) = g − ( q r d − 1 ) / ( q − 1 ) . (96) \det\Gamma_d(\phi)\cdot\det\Gamma_d^*(\phi^*)\cdot\mathrm{N}(\mathfrak{g}'(\varepsilon))^r=\det P_d(1/\mathfrak{g})=\mathfrak{g}^{-(q^{rd}-1)/(q-1)}. \tag*{(96)} det Γ d ( ϕ ) ⋅ det Γ d ∗ ( ϕ ∗ ) ⋅ N ( g ′ ( ε ) ) r = det P d ( 1/ g ) = g − ( q r d − 1 ) / ( q − 1 ) . (96)
Though (a) follows from (b), we can prove it simply by raising (95) to the ( q − 1 ) (q-1) ( q − 1 ) -st power and applying Proposition 9.7. Recalling that μ = ( − 1 ) r − 1 / b r \mu=(-1)^{r-1}/b_r μ = ( − 1 ) r − 1 / b r and V ( ε ) q − 1 = ( − 1 ) d − 1 V(\varepsilon)^{q-1}=(-1)^{d-1} V ( ε ) q − 1 = ( − 1 ) d − 1 , a straightforward calculation yields the identity for det Γ d ( ϕ ) q − 1 \det\Gamma_d(\phi)^{q-1} det Γ d ( ϕ ) q − 1 in (a). The identity for det Γ d ∗ ( ϕ ∗ ) q − 1 \det\Gamma_d^*(\phi^*)^{q-1} det Γ d ∗ ( ϕ ∗ ) q − 1 in (a) follows from (96) . For (b), we combine (90) with (95) and find
det Γ d ( ϕ ) = μ ( q r d − 1 ) / ( q r − 1 ) V ( ε ) r N ( g ′ ( ε ) ) r − 1 ∏ j = 0 d − 1 g C ( χ j ) , \det\Gamma_d(\phi)=\frac{\mu^{(q^{rd}-1)/(q^r-1)}V(\varepsilon)^r}{\mathrm{N}(\mathfrak{g}'(\varepsilon))^{r-1}}\prod_{j=0}^{d-1}g_C(\chi_j), det Γ d ( ϕ ) = N ( g ′ ( ε ) ) r − 1 μ ( q r d − 1 ) / ( q r − 1 ) V ( ε ) r j = 0 ∏ d − 1 g C ( χ j ) ,
and then after applying (80) and Theorem 8.24(c),
= ( − 1 ) d ( d − 1 ) / 2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ V ( ε ) r − 2 N ( g ′ ( ε ) ) r − 1 ⋅ g 1 / ( q − 1 ) . =(-1)^{d(d-1)/2}\cdot(-1)^{d/(q-1)}\cdot\mu^{(q^{rd}-1)/(q^r-1)}\cdot\frac{V(\varepsilon)^{r-2}}{\mathrm{N}(\mathfrak{g}'(\varepsilon))^{r-1}}\cdot\mathfrak{g}^{1/(q-1)}. = ( − 1 ) d ( d − 1 ) /2 ⋅ ( − 1 ) d / ( q − 1 ) ⋅ μ ( q r d − 1 ) / ( q r − 1 ) ⋅ N ( g ′ ( ε ) ) r − 1 V ( ε ) r − 2 ⋅ g 1/ ( q − 1 ) .
Using that N ( g ′ ( ε ) ) = ( − 1 ) d ( d − 1 ) / 2 V ( ε ) 2 \mathrm{N}(\mathfrak{g}'(\varepsilon))=(-1)^{d(d-1)/2}V(\varepsilon)^2 N ( g ′ ( ε )) = ( − 1 ) d ( d − 1 ) /2 V ( ε ) 2 completes the proof for det Γ d ( ϕ ) \det\Gamma_d(\phi) det Γ d ( ϕ ) , and then the identity for det Γ d ∗ ( ϕ ∗ ) \det\Gamma_d^*(\phi^*) det Γ d ∗ ( ϕ ∗ ) follows from (96) . □
Special values of Drinfeld modular forms in Tate algebras
We explore connections between Drinfeld modular forms of arbitrary rank with values in T \mathbb{T} T and dual Gauss-Thakur sums. We hope the methods of the present section shed additional light on certain modular form identities. For precise definitions of Drinfeld modular forms of higher rank, see [7 , 8 , 12 , 22 , 23 ] , and for modular forms in Tate algebras [36 , 37 , 38 , 39 ] .
Throughout we primarily adopt the conventions and notation of Chen and Gezmiş in [12 ] for Drinfeld modular forms. For r ≥ 2 r\ge2 r ≥ 2 , let H r = P r − 1 ( C ∞ ) ∖ { k ∞ -rational hyperplanes } \mathbb{H}^r=\mathbb{P}^{r-1}(\mathbb{C}_\infty)\setminus\{k_\infty\text{-rational hyperplanes}\} H r = P r − 1 ( C ∞ ) ∖ { k ∞ -rational hyperplanes } be the Drinfeld upper half space. For z = ( z 1 , … , z r ) T ∈ H r \boldsymbol{z}=(z_1,\ldots,z_r)^{\mathsf{T}}\in\mathbb{H}^r z = ( z 1 , … , z r ) T ∈ H r , z 1 , … , z r z_1,\ldots,z_r z 1 , … , z r are k ∞ k_\infty k ∞ -linearly independent, which we normalize so that z r = 1 z_r=1 z r = 1 . For z ∈ H r \boldsymbol{z}\in\mathbb{H}^r z ∈ H r , define Λ z = A z 1 + ⋯ + A z r ⊆ C ∞ \Lambda_{\boldsymbol{z}}=Az_1+\cdots+Az_r\subseteq\mathbb{C}_\infty Λ z = A z 1 + ⋯ + A z r ⊆ C ∞ , and let ϕ z \phi^{\boldsymbol{z}} ϕ z be the Drinfeld module associated to Λ z \Lambda_{\boldsymbol{z}} Λ z ,
ϕ θ z : = θ + b 1 ( z ) τ + ⋯ + b r ( z ) τ r . \phi_\theta^{\boldsymbol{z}}:=\theta+b_1(\boldsymbol{z})\tau+\cdots+b_r(\boldsymbol{z})\tau^r. ϕ θ z := θ + b 1 ( z ) τ + ⋯ + b r ( z ) τ r .
As functions on H r \mathbb{H}^r H r , the coefficients b i ( z ) b_i(\boldsymbol{z}) b i ( z ) are then Drinfeld modular forms of rank r r r , weight q i − 1 q^i-1 q i − 1 , and type 0 for G L r ( A ) \mathrm{GL}_r(A) GL r ( A ) [7 Prop. 3.4, 12 Ex. 2.8, 22 Ex. 1.12] . The leading coefficient b r ( z ) b_r(\boldsymbol{z}) b r ( z ) is the discriminant form Δ r ( z ) \Delta_r(\boldsymbol{z}) Δ r ( z ) of Gekeler as in [22 §3] . Gekeler [21 §4, 22 Thm. 3.8] further defined a rank r r r modular form h r ( z ) h_r(\boldsymbol{z}) h r ( z ) , which has weight ( q r − 1 ) / ( q − 1 ) (q^r-1)/(q-1) ( q r − 1 ) / ( q − 1 ) and type 1, such that (see also [8 Thm. 5.3(2), 12 Ex. 2.8(iii)] )
Δ r ( z ) = b r ( z ) = ( − 1 ) r − 1 π ~ q r − 1 h r ( z ) q − 1 . (97) \Delta_r(\boldsymbol{z})=b_r(\boldsymbol{z})=(-1)^{r-1}\widetilde{\pi}^{q^r-1}h_r(\boldsymbol{z})^{q-1}. \tag*{(97)} Δ r ( z ) = b r ( z ) = ( − 1 ) r − 1 π q r − 1 h r ( z ) q − 1 . (97)
We have Eisenstein series of weight k k k and type 0 (see [7 Ex. 3.1, 12 Ex. 2.8(i), 22 §1] ),
E k ( z ) = ∑ ( a 1 , … , a r ) ∈ A r ′ 1 ( a 1 z 1 + ⋯ + a r z r ) k . E_k(\boldsymbol{z})=\sum_{(a_1,\ldots,a_r)\in A^r}'\frac{1}{(a_1z_1+\cdots+a_rz_r)^k}. E k ( z ) = ( a 1 , … , a r ) ∈ A r ∑ ′ ( a 1 z 1 + ⋯ + a r z r ) k 1 .
Right away we see that if we take the dual special functions ζ i ( z , t ) ∈ T \zeta_i(\mathbf{z},t)\in\mathbb{T} ζ i ( z , t ) ∈ T , 1 ≤ i ≤ r 1\le i\le r 1 ≤ i ≤ r , for φ z \varphi^{\mathbf{z}} φ z as in Corollary 4.10, then
E q j − 1 ( z ) = − ∑ i = 1 r ζ i ( ζ i ( z , t ) ) ( j ) ∣ t = θ , j ≥ 0. E_{q^j-1}(\mathbf{z})=-\left.\sum_{i=1}^{r}\zeta_i\left(\zeta_i(\mathbf{z},t)\right)^{(j)}\right|_{t=\theta},\qquad j\ge0. E q j − 1 ( z ) = − i = 1 ∑ r ζ i ( ζ i ( z , t ) ) ( j ) t = θ , j ≥ 0.
Chen and Gezmiş [12 ] investigate Eisenstein series taking values in T \mathbb{T} T , originally constructed by Pellarin [36 §2] and Perkins [39 §4] . For k ≥ 0 k\ge0 k ≥ 0 , set
E k ( z , t ) : = ( ∑ ( a 1 , … , a r ) ∈ A r ′ a 1 ( t ) ( a 1 z 1 + ⋯ + a r z r ) k , … , ∑ ( a 1 , … , a r ) ∈ A r ′ a r ( t ) ( a 1 z 1 + ⋯ + a r z r ) k ) T , \mathcal{E}_k(\mathbf{z},t):=\left(\sum_{(a_1,\ldots,a_r)\in A^r}^{\prime}\frac{a_1(t)}{(a_1z_1+\cdots+a_rz_r)^k},\ldots,\sum_{(a_1,\ldots,a_r)\in A^r}^{\prime}\frac{a_r(t)}{(a_1z_1+\cdots+a_rz_r)^k}\right)^{\mathsf{T}}, E k ( z , t ) := ( a 1 , … , a r ) ∈ A r ∑ ′ ( a 1 z 1 + ⋯ + a r z r ) k a 1 ( t ) , … , ( a 1 , … , a r ) ∈ A r ∑ ′ ( a 1 z 1 + ⋯ + a r z r ) k a r ( t ) T ,
which, for fixed z \mathbf{z} z , is a vector in T r \mathbb{T}^r T r . Let Υ ( z , t ) , Ψ ( z , t ) ∈ GL r ( T ) \Upsilon(\mathbf{z},t),\Psi(\mathbf{z},t)\in\operatorname{GL}_r(\mathbb{T}) Υ ( z , t ) , Ψ ( z , t ) ∈ GL r ( T ) be defined with respect to φ z \varphi^{\mathbf{z}} φ z as in (3.5), (3.8), and let ω ( z , t ) \omega(\mathbf{z},t) ω ( z , t ) and ζ ( z , t ) \zeta(\mathbf{z},t) ζ ( z , t ) be their first column and row respectively. Then again by Corollary 4.10,
E q j ( z , t ) = ( − ζ ( z , t ) ( j ) ) T . (98) \mathcal{E}_{q^j}(\mathbf{z},t)=\left(-\zeta(\mathbf{z},t)^{(j)}\right)^{\mathsf{T}}. \tag*{(98)} E q j ( z , t ) = ( − ζ ( z , t ) ( j ) ) T . (98)
Also, we note that Υ ( z , t ) \Upsilon(\mathbf{z},t) Υ ( z , t ) is exactly the same as F ( z , t ) \mathcal{F}(\mathbf{z},t) F ( z , t ) in [12 Eq. (3.1)] , and moreover,
det Υ ( z , t ) = π ~ − ( q r − 1 ) / ( q − 1 ) h r ( z ) − 1 ω C ( t ) , (99) \det\Upsilon(\mathbf{z},t)=\widetilde{\pi}^{-(q^r-1)/(q-1)}h_r(\mathbf{z})^{-1}\omega_C(t), \tag*{(99)} det Υ ( z , t ) = π − ( q r − 1 ) / ( q − 1 ) h r ( z ) − 1 ω C ( t ) , (99)
by [12 Prop. 3.4] . Thus, we see that
μ ( z ) : = π ~ − ( q r − 1 ) / ( q − 1 ) h r ( z ) − 1 (100) \mu(\mathbf{z}):=\widetilde{\pi}^{-(q^r-1)/(q-1)}h_r(\mathbf{z})^{-1} \tag*{(100)} μ ( z ) := π − ( q r − 1 ) / ( q − 1 ) h r ( z ) − 1 (100)
is the value of μ \mu μ in (9.2) and elsewhere, as a function of z \mathbf{z} z . Let C ( z , t ) \mathcal{C}(\mathbf{z},t) C ( z , t ) be the cofactor matrix of Υ ( z , t ) \Upsilon(\mathbf{z},t) Υ ( z , t ) , so that C ( z , t ) T Υ ( z , t ) = det Υ ( z , t ) I r \mathcal{C}(\mathbf{z},t)^{\mathsf{T}}\Upsilon(\mathbf{z},t)=\det\Upsilon(\mathbf{z},t)I_r C ( z , t ) T Υ ( z , t ) = det Υ ( z , t ) I r .
Theorem 9.20. For fixed z ∈ H r \mathbf{z}\in\mathbb{H}^r z ∈ H r the following identities hold in T r \mathbb{T}^r T r .
(a) (Pellarin [36 Thm. 8] , case r = 2 r=2 r = 2 ): For z ∈ C ∞ ∖ k ∞ \mathbf{z}\in\mathbb{C}_\infty\setminus k_\infty z ∈ C ∞ ∖ k ∞ .
( ∑ ( a , b ) ∈ A 2 ′ a ( t ) a z + b , ∑ ( a , b ) ∈ A 2 ′ b ( t ) a z + b ) = π ~ q + 1 h 2 ( z , 1 ) ( t − θ ) ω C ( t ) ⋅ ( ω 1 ( 1 ) ( t ) , − ω z ( 1 ) ( t ) ) . \left(\sum_{(a,b)\in A^2}^{\prime}\frac{a(t)}{az+b},\sum_{(a,b)\in A^2}^{\prime}\frac{b(t)}{az+b}\right)=\frac{\widetilde{\pi}^{q+1}h_2(z,1)}{(t-\theta)\omega_C(t)}\cdot\left(\omega_1^{(1)}(t),-\omega_z^{(1)}(t)\right). ( a , b ) ∈ A 2 ∑ ′ a z + b a ( t ) , ( a , b ) ∈ A 2 ∑ ′ a z + b b ( t ) = ( t − θ ) ω C ( t ) π q + 1 h 2 ( z , 1 ) ⋅ ( ω 1 ( 1 ) ( t ) , − ω z ( 1 ) ( t ) ) . (b) (Chen-Gezmiş [12 Thm. 3.2] , case r ≥ 2 r\ge2 r ≥ 2 ): For z ∈ H r \mathbf{z}\in\mathbb{H}^r z ∈ H r .
E 1 ( z , t ) T = π ~ ( q r − 1 ) / ( q − 1 ) h r ( z ) ( t − θ ) ω C ( t ) ⋅ ( C ( z , t ) 11 , … , C ( z , t ) r 1 ) . \mathcal{E}_1(\mathbf{z},t)^{\mathsf{T}}=\frac{\widetilde{\pi}^{(q^r-1)/(q-1)}h_r(\mathbf{z})}{(t-\theta)\omega_C(t)}\cdot\left(\mathcal{C}(\mathbf{z},t)_{11},\ldots,\mathcal{C}(\mathbf{z},t)_{r1}\right). E 1 ( z , t ) T = ( t − θ ) ω C ( t ) π ( q r − 1 ) / ( q − 1 ) h r ( z ) ⋅ ( C ( z , t ) 11 , … , C ( z , t ) r 1 ) .
Alternative proof of Theorem 9.20. Part (a) is simply a special case of (b). For (b) we need [12 Prop. 3.4] to pin down the sign of det Υ ( z , t ) \det\Upsilon(\mathbf{z},t) det Υ ( z , t ) as a function of z \mathbf{z} z as in (9.18) (cf. (9.2), (9.19)). In any case, similar to the proof of Corollary 4.10,
E 1 ( z , t ) T = − ζ ( z , t ) = ( − 1 , 0 , … , 0 ) Ψ ( z , t ) = ( 1 t − θ , 0 , … , 0 ) C ( z , t ) T det Υ ( z , t ) − 1 , \mathcal{E}_1(\mathbf{z},t)^{\mathsf{T}}=-\zeta(\mathbf{z},t)=(-1,0,\ldots,0)\Psi(\mathbf{z},t)=\left(\frac{1}{t-\theta},0,\ldots,0\right)\mathcal{C}(\mathbf{z},t)^{\mathsf{T}}\det\Upsilon(\mathbf{z},t)^{-1}, E 1 ( z , t ) T = − ζ ( z , t ) = ( − 1 , 0 , … , 0 ) Ψ ( z , t ) = ( t − θ 1 , 0 , … , 0 ) C ( z , t ) T det Υ ( z , t ) − 1 ,
using Lemma 3.13 at the end (cf. [12 Prop. 3.18] ). The result follows from (99) . □
As usual, fix irreducible ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + of degree d d d with root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^{d}} ε ∈ F q d . For a character χ ∈ F ^ q × \chi\in\widehat{\mathbb{F}}_{q}^{\times} χ ∈ F q × , 1 ≤ i ≤ r 1 \le i \le r 1 ≤ i ≤ r and z ∈ H r \mathbf{z} \in\mathbb{H}^{r} z ∈ H r , we will denote g i ( z , χ ) g_{i}(\mathbf{z},\chi) g i ( z , χ ) , g i ∗ ( z , χ ) g_{i}^{*}(\mathbf{z},\chi) g i ∗ ( z , χ ) , G ( z , χ ) \mathcal{G}(\mathbf{z},\chi) G ( z , χ ) , G ∗ ( z , χ ) \mathcal{G}^{*}(\mathbf{z},\chi) G ∗ ( z , χ ) , Γ ( z ) \Gamma(\mathbf{z}) Γ ( z ) , Γ ∗ ( z ) \Gamma^{*}(\mathbf{z}) Γ ∗ ( z ) , etc. for the corresponding constructions on Gauss-Thakur sums and their duals as in §§6–7, with respect to the Drinfeld modules ϕ z \phi^{\mathbf{z}} ϕ z and their adjoints ( ϕ z ) ∗ (\phi^{\mathbf{z}})^{*} ( ϕ z ) ∗ . The connection with dual Gauss-Thakur sums arrives through evaluation, as by Corollary 6.19 and (9.17),
E q j ( z , ε ) = − ( g 1 ∗ [ j ] ( z , χ ε ) ℘ ′ ( ε ) , … , g r ∗ [ j ] ( z , χ ε ) ℘ ′ ( ε ) ) T . (101) \mathcal{E}_{qj}(\mathbf{z},\varepsilon)=-\left(\frac{g_{1}^{*[j]}(\mathbf{z},\chi_{\varepsilon})}{\wp'(\varepsilon)},\ldots,\frac{g_{r}^{*[j]}(\mathbf{z},\chi_{\varepsilon})}{\wp'(\varepsilon)}\right)^{\mathsf{T}}. \tag*{(101)} E q j ( z , ε ) = − ( ℘ ′ ( ε ) g 1 ∗ [ j ] ( z , χ ε ) , … , ℘ ′ ( ε ) g r ∗ [ j ] ( z , χ ε ) ) T . (101)
From (9.6), Proposition 9.7, and Theorem 9.12, we immediately obtain identities for h r ( z ) h_{r}(\mathbf{z}) h r ( z ) and Δ r ( z ) \Delta_{r}(\mathbf{z}) Δ r ( z ) in terms of dual Gauss-Thakur sums.
Corollary 9.23. Fix ℘ ∈ A + \wp\in A_{+} ℘ ∈ A + irreducible of degree d d d and root ε ∈ F q d \varepsilon\in\mathbb{F}_{q^{d}} ε ∈ F q d . For z ∈ H r \mathbf{z} \in\mathbb{H}^{r} z ∈ H r ,
(a) h r ( z ) = ( − 1 ) r − 1 ℘ ′ ( ε ) r − 1 π ~ ( q r − 1 ) / ( q − 1 ) g C ∗ [ r − 1 ] ( χ ε ) ⋅ det G ∗ ( z , χ ε ) , h_{r}(\mathbf{z})=\frac{(-1)^{r-1}\wp'(\varepsilon)^{r-1}}{\widetilde{\pi}^{(q^{r}-1)/(q-1)}}g_{\mathcal{C}}^{*[r-1]}(\chi_{\varepsilon})\cdot\det\mathcal{G}^{*}(\mathbf{z},\chi_{\varepsilon}), h r ( z ) = π ( q r − 1 ) / ( q − 1 ) ( − 1 ) r − 1 ℘ ′ ( ε ) r − 1 g C ∗ [ r − 1 ] ( χ ε ) ⋅ det G ∗ ( z , χ ε ) ,
h r ( z ) ( q r d − 1 ) / ( q r − 1 ) = ( − 1 ) d / ( q − 1 ) V ( ε ) ℘ ( ε ) q r d / ( q − 1 ) π ~ ( q r d − 1 ) / ( q − 1 ) ⋅ det Γ d ∗ ( z ) ; h_{r}(\mathbf{z})^{(q^{rd}-1)/(q^{r}-1)}=\frac{(-1)^{d/(q-1)}V(\varepsilon)\wp(\varepsilon)^{q^{rd}/(q-1)}}{\widetilde{\pi}^{(q^{rd}-1)/(q-1)}}\cdot\det\Gamma_{d}^{*}(\mathbf{z}); h r ( z ) ( q r d − 1 ) / ( q r − 1 ) = π ( q r d − 1 ) / ( q − 1 ) ( − 1 ) d / ( q − 1 ) V ( ε ) ℘ ( ε ) q r d / ( q − 1 ) ⋅ det Γ d ∗ ( z ) ;
(b) Δ r ( z ) d = ( − 1 ) r d ℘ ( ε ) q r ∏ j = 0 d − 1 det G ∗ ( z , χ j ) q − 1 , \Delta_{r}(\mathbf{z})^{d}=(-1)^{rd}\wp(\varepsilon)^{q^{r}}\prod_{j=0}^{d-1}\det\mathcal{G}^{*}(\mathbf{z},\chi_{j})^{q-1}, Δ r ( z ) d = ( − 1 ) r d ℘ ( ε ) q r ∏ j = 0 d − 1 det G ∗ ( z , χ j ) q − 1 ,
Δ r ( z ) ( q r d − 1 ) / ( q r − 1 ) = ( − 1 ) r ℘ ( ε ) q r d ⋅ det Γ d ∗ ( z ) q − 1 \Delta_{r}(\mathbf{z})^{(q^{rd}-1)/(q^{r}-1)}=(-1)^{r}\wp(\varepsilon)^{q^{rd}}\cdot\det\Gamma_{d}^{*}(\mathbf{z})^{q-1} Δ r ( z ) ( q r d − 1 ) / ( q r − 1 ) = ( − 1 ) r ℘ ( ε ) q r d ⋅ det Γ d ∗ ( z ) q − 1 .
References
Department of Mathematics, Texas A&M University, College Station, TX 77843, U.S.A.
Email address , A. Bassil: athenabassil@tamu.edu
Email address , M. Papanikolas: papanikolas@tamu.edu