Introduction
The Bers quasiform Ψ N ( x , y ) \Psi_N(x,y) Ψ N ( x , y ) (N ≥ 2 N\geq2 N ≥ 2 ; see [16 ] ) is a Poincaré series over a Schottky group built from π N = d x N d y 1 − N x − y \pi_N=\frac{dx^Ndy^{1-N}}{x-y} π N = x − y d x N d y 1 − N , multiplied by an auxiliary factor ∏ j = 0 2 N − 2 y − A j x − A j \prod_{j=0}^{2N-2}\frac{y-A_j}{x-A_j} ∏ j = 0 2 N − 2 x − A j y − A j . The factor destroys periodicity in y y y and produces an Eichler-cocycle defect. The kernel has weight ( N , 1 − N ) = ( 1 2 , 1 2 ) (N,1-N)=(\frac{1}{2},\frac{1}{2}) ( N , 1 − N ) = ( 2 1 , 2 1 ) at N = 1 2 N=\frac{1}{2} N = 2 1 , the weight of a Szegő kernel, and there are 2 N − 1 = 0 2N-1=0 2 N − 1 = 0 auxiliary points. This paper treats that case, without formal continuation in N N N .
Relation to earlier work
Poincaré-type series ∑ ( ⋯ ) γ ′ w \sum(\cdots)_{\gamma}'{}^{w} ∑ ( ⋯ ) γ ′ w over Schottky groups go back to Burnside and Poincaré. For integral weights w ≥ 1 w\geq1 w ≥ 1 and their role in the theory of automorphic forms for Schottky groups see [4 , 7 , 11 ] . Schottky-series expressions are also standard in the operator formalism of string theory. The Bers quasiform on a Schottky-uniformized surface, its properties and use in expansions of meromorphic N N N -forms are developed in [16 ] . This paper is the source of the genus g g g Zhu recursion for vertex operator algebras of [17 ] (see Remark 4.7 for what that recursion is). The name “Szegő kernel” also designates the reproducing kernel of Hardy spaces on domains in C n \mathbb{C}^n C n . For recent work in that setting (closed formulas and asymptotic expansions on balls and Reinhardt domains) see [5 , 8 ] . Those are different objects from the spinor Szegő kernel on a compact Riemann surface studied here (a section of L ⊠ L L\boxtimes L L ⊠ L with a pole on the diagonal, not the reproducing kernel of a Hardy space), and we do not explore a relation. To our knowledge the following is specific to weight 1 2 \frac{1}{2} 2 1 : the series needs no auxiliary points and is invariant under both variables. It is a Szegő kernel (the unique one, by H 0 = 0 H^0=0 H 0 = 0 ) for each of the N g N_g N g lifted spin structures. Its convergence, behaviour at ∞ \infty ∞ and expansion in the Schottky multipliers can be controlled explicitly.
Our results
Let us describe the results of this paper. We proved (modulo the quoted classical facts on theta functions and prime forms, [9 , 13 ] ): § 2 –3 except Remark 3.20, Theorems 4.1 and 5.1, Proposition 6.1. The estimates of Sections 3.1–§ 3.2 and Theorem 5.1 are uniform on the domain (H); Theorem 3.15 extends the qualitative results to every Schottky group satisfying the intrinsic condition (E) below.
We proved, depending on a hypothesis: Corollary 4.4 (on (A)) and Proposition 4.5 (on (Af f _{\mathrm{ff}} ff ). Neither hypothesis is established here for correlation functions obtained by a Schottky construction. These are statements about meromorphic sections of L ε L_{\varepsilon} L ε , not about the vertex operator superalgebra.
2020 Mathematics Subject Classification . Primary 30F10, 17B69; Secondary 14H55, 81T40.
Key words and phrases. Bers quasiform, Szegő kernel, spin structure, Schottky group, vertex operator superalgebra, Zhu recursion.
We conjecture: the dictionary between lifts and theta characteristics for g ≥ 2 g \ge2 g ≥ 2 (see Remark 3.20); convergence and continuation beyond (E); Conjecture 4.6 (general-VOSA genus g g g recursion).
Not treated in this paper: the 2 2 g − N g 2^{2g}-N_g 2 2 g − N g remaining spin structures (this set contains both even spin structures outside our family and all odd ones); zero modes; super-moduli.
Notations
s ε ( x , y ) s_\varepsilon(x,y) s ε ( x , y ) is the scalar kernel in the coordinates ( x , y ) (x,y) ( x , y ) , S ε = s ε ( x , y ) ( d x d y ) 1 / 2 S_\varepsilon=s_\varepsilon(x,y)(dx\,dy)^{1/2} S ε = s ε ( x , y ) ( d x d y ) 1/2 the corresponding section of L ε ⊠ L ε L_\varepsilon\boxtimes L_\varepsilon L ε ⊠ L ε , and s ^ ε ( a , b ) \hat{s}_\varepsilon(\mathbf{a},\mathbf{b}) s ^ ε ( a , b ) (a , b ∈ C 2 ∖ 0 \mathbf{a},\mathbf{b}\in\mathbb{C}^2\setminus0 a , b ∈ C 2 ∖ 0 , Section 3) the bihomogeneous representative of the same section, which treats ∞ \infty ∞ like any other point. Domains attached to fixed Schottky data p = ( w , λ ) p=(w,\lambda) p = ( w , λ ) carry the subscript p p p . Statements uniform in the data are formulated on total spaces (see § 2 ).
Our hypotheses
We assume the following:
(H) 0 < r < d / ( 8 ( 2 g − 1 ) ) 0<r<d/(8(2g-1)) 0 < r < d / ( 8 ( 2 g − 1 )) : condition on the Schottky data, used for all quantitative estimates (Lemmas 3.2 , 3.3 , Theorems 3.5 , 5.1).
(D): the discs D c D_c D c are pairwise disjoint (so that Γ \Gamma Γ is a classical Schottky group; used in § 2 ).
(E): ∑ γ ~ ∥ γ ~ ∥ − 1 < ∞ \sum_{\widetilde{\gamma}}\|\widetilde{\gamma}\|^{-1}<\infty ∑ γ ∥ γ ∥ − 1 < ∞ , intrinsic.
We have the following implications: (H) leads to (D) and (H) leads to (E) for every $g\ge 1 ({xref}Theorem 3.15 (iii)). Critical exponent $<\frac{1}{2}$ leads to (E). (E) together with (D) gives all qualitative results of {xref}Section 3, in every genus $g\ge 1$. Genus one does not require (H) since {xref}Proposition 3.19 treats it. (A), $(A_{\mathrm{ff}})$ are hypotheses on vertex-algebraic correlators, used only in {xref}Section 4.2 and {xref}Section 6`. They are not implied by, and do not imply, any of (H), (D), (E).
Main Theorem
Let Γ \Gamma Γ be a Schottky group of genus g ≥ 1 g\ge1 g ≥ 1 with data ( w , λ ) (w,\lambda) ( w , λ ) as in § 2 satisfying (D) and (E) of § 3.3 (for g ≥ 2 g\ge2 g ≥ 2 this holds in particular under the open condition (H)), and let ε \varepsilon ε be one of the N g N_g N g lifts (see Proposition 2.1). Then:
Theorems 3.5, 3.15: S ε ( x , y ) = ∑ γ ~ ∈ Γ ~ π ( γ ~ x , y ) S_\varepsilon(x,y)=\sum_{\widetilde{\gamma}\in\widetilde{\Gamma}}\pi(\widetilde{\gamma}x,y) S ε ( x , y ) = ∑ γ ∈ Γ π ( γ x , y ) converges absolutely and locally uniformly for x , y ∈ Ω 0 ⊂ C ^ x,y\in\Omega_0\subset\widehat{\mathbb{C}} x , y ∈ Ω 0 ⊂ C (the point ∞ \infty ∞ and its orbit included, in homogeneous coordinates) with y ∉ Γ x y\notin\Gamma x y ∈ / Γ x . Under (H) it is moreover jointly holomorphic in the parameters on the total space. The bound s ε = O ( 1 / ∣ x ∣ ) s_\varepsilon=O(1/|x|) s ε = O ( 1/∣ x ∣ ) as x → ∞ x\to\infty x → ∞ holds only locally uniformly in y ∈ Ω 0 ∖ Γ ∞ y\in\Omega_0\setminus\Gamma\infty y ∈ Ω 0 ∖ Γ∞ .
Propositions 3.6, 3.7: S ε ( γ x , y ) = j ( γ ~ , x ~ ) − 1 S ε ( x , y ) S_\varepsilon(\gamma x,y)=j(\widetilde{\gamma},\widetilde{x})^{-1}S_\varepsilon(x,y) S ε ( γ x , y ) = j ( γ , x ) − 1 S ε ( x , y ) , S ε ( x , γ y ) = j ( γ ~ , y ~ ) − 1 S ε ( x , y ) S_\varepsilon(x,\gamma y)=j(\widetilde{\gamma},\widetilde{y})^{-1}S_\varepsilon(x,y) S ε ( x , γ y ) = j ( γ , y ) − 1 S ε ( x , y ) and S ε ( y , x ) = − S ε ( x , y ) S_\varepsilon(y,x)=-S_\varepsilon(x,y) S ε ( y , x ) = − S ε ( x , y ) , with no additive Eichler-type defect.
Proposition 3.6: S ε S_\varepsilon S ε descends to a global meromorphic section of L ε ⊠ L ε L_\varepsilon\boxtimes L_\varepsilon L ε ⊠ L ε on Σ × Σ \Sigma\times\Sigma Σ × Σ .
Proposition 3.7 (ii): its only singularity is a simple pole along the diagonal Δ \Delta Δ , with principal part the canonical section of ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ≅ O Σ (L_\varepsilon\boxtimes L_\varepsilon)(\Delta)|_\Delta\cong\mathcal{O}_\Sigma ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ≅ O Σ .
Theorem 3.10: H 0 ( Σ , L ε ) = 0 H^0(\Sigma,L_\varepsilon)=0 H 0 ( Σ , L ε ) = 0 , and S ε S_\varepsilon S ε is the unique meromorphic section of L ε ⊠ L ε L_\varepsilon\boxtimes L_\varepsilon L ε ⊠ L ε with properties (2)–(4).
Proposition 3.16, Corollary 3.18: S ε S_\varepsilon S ε is the theta-function Szegő kernel of the characteristic δ ( ε ) \delta(\varepsilon) δ ( ε ) , and R ε = 12 c 1 R_\varepsilon=12c_1 R ε = 12 c 1 is the Bergman projective connection given by 6 ∑ i , j ∂ i ∂ j log θ [ δ ( ε ) ] ζ i ζ j 6\sum_{i,j}\partial_i\partial_j\log\theta[\delta(\varepsilon)]\,\zeta_i\zeta_j 6 ∑ i , j ∂ i ∂ j log θ [ δ ( ε )] ζ i ζ j .
Proposition 3.19: for g = 1 g=1 g = 1 the kernel and the relation ε ↦ δ ( ε ) \varepsilon\mapsto\delta(\varepsilon) ε ↦ δ ( ε ) are given.
Remark 1.1. Genus zero is empty since there is no Schottky group. Sections 2–4 and Section 6 hold for every g ≥ 1 g\ge1 g ≥ 1 under (D) and (E) (Theorem 3.15, Proposition 3.19 ). The quantitative part (Lemmas 3.2 , 3.3 , Theorem 3.5 and Theorem 5.1, whose constants use r / d ≤ 1 4 r/d\le\frac{1}{4} r / d ≤ 4 1 ) is written for g ≥ 2 g\ge2 g ≥ 2 . At genus one the constants change, and Theorem 3.15 and Proposition 3.19 are used instead.
The vertex operator superalgebra statements depend on the following assumptions: the fermionic Zhu-type reduction and the free fermion Pfaffian formula are proved only under hypotheses (A), ( A f f ) (A_{\mathrm{ff}}) ( A ff ) on genus g g g correlation functions which are not established for Schottky-sewn correlators, and the general VOSA genus g g g recursion is a conjecture (see [17 ] ).
Schottky data, lifts and spin structures
Let I = I + ∪ ( − I + ) \mathcal{I}=\mathcal{I}_+\cup(-\mathcal{I}_+) I = I + ∪ ( − I + ) , I + = { 1 , … , g } \mathcal{I}_+=\{1,\ldots,g\} I + = { 1 , … , g } . We deal with the following data: points w c ∈ C w_c\in\mathbb{C} w c ∈ C (c ∈ I c\in\mathcal{I} c ∈ I ), and λ a ∈ C × \lambda_a\in\mathbb{C}^{\times} λ a ∈ C × (a ∈ I + a\in\mathcal{I}_+ a ∈ I + , in particular λ a ≠ 0 \lambda_a\ne0 λ a = 0 ), with λ − a : = − λ a \lambda_{-a}:=-\lambda_a λ − a := − λ a and ρ a : = − λ a \rho_a:=-\lambda_a ρ a := − λ a , ρ − a : = ρ a \rho_{-a}:=\rho_a ρ − a := ρ a . Put d : = min c ≠ c ′ ∣ w c − w c ′ ∣ d:=\min_{c\ne c'}|w_c-w_{c'}| d := min c = c ′ ∣ w c − w c ′ ∣ , r c : = ∣ λ c ∣ = ∣ ρ c ∣ 1 / 2 r_c:=|\lambda_c|=|\rho_c|^{1/2} r c := ∣ λ c ∣ = ∣ ρ c ∣ 1/2 , r : = max c r c r:=\max_c r_c r := max c r c (thus ∣ ρ ∣ max = r 2 |\rho|_{\max}=r^2 ∣ ρ ∣ m a x = r 2 ), and let (H) denote the condition
0 < r ≤ d 8 ( 2 g − 1 ) . 0<r\le\frac{d}{8(2g-1)}. 0 < r ≤ 8 ( 2 g − 1 ) d .
The parameter space is { ( w c ) c ∈ I ∈ C 2 g distinct } × ( C × ) I + \{(w_c)_{c\in I}\in\mathbb{C}^{2g}\text{ distinct}\}\times(\mathbb{C}^{\times})^{I_+} {( w c ) c ∈ I ∈ C 2 g distinct } × ( C × ) I + , and (H) singles out an open subset P \mathcal{P} P of it; λ a = 0 \lambda_a=0 λ a = 0 is not a point of the parameter space (see Remark 3.12). A point of P \mathcal{P} P is written as p = ( w , λ ) p=(w,\lambda) p = ( w , λ ) , and d = d ( p ) d=d(p) d = d ( p ) , r c = r c ( p ) r_c=r_c(p) r c = r c ( p ) , r = r ( p ) r=r(p) r = r ( p ) are continuous functions on P \mathcal{P} P .
We have the following convention on parameter-dependent domains: every domain below built from w c w_c w c , r c r_c r c (namely K \mathcal{K} K , K d \mathcal{K}_d K d , D D D , Ω 0 \Omega_0 Ω 0 ) depends on p p p and is written K p \mathcal{K}_p K p , etc., when this matters. Statements claiming uniformity or joint holomorphy in p p p are formulated on the total spaces, e.g., K : = { ( p , z ) ∈ P × C ^ : z = ∞ or ∣ z − w c ( p ) ∣ ≥ r c ( p ) / 2 for all c } \mathcal{K}:=\{(p,z)\in\mathcal{P}\times\widehat{\mathbb{C}}:z=\infty\text{ or }|z-w_c(p)|\ge r_c(p)/2\text{ for all }c\} K := {( p , z ) ∈ P × C : z = ∞ or ∣ z − w c ( p ) ∣ ≥ r c ( p ) /2 for all c } , but never on a product U × K × K U\times\mathcal{K}\times\mathcal{K} U × K × K . Let γ a ( z ) = w − a + ρ a / ( z − w a ) \gamma_a(z)=w_{-a}+\rho_a/(z-w_a) γ a ( z ) = w − a + ρ a / ( z − w a ) , γ − a : = γ a − 1 \gamma_{-a}:=\gamma_a^{-1} γ − a := γ a − 1 , i.e., γ − a ( u ) = w a + ρ a / ( u − w − a ) \gamma_{-a}(u)=w_a+\rho_a/(u-w_{-a}) γ − a ( u ) = w a + ρ a / ( u − w − a ) . The matrix
γ ~ a : = 1 λ a ( w − a ρ a − w a w − a 1 − w a ) \widetilde{\gamma}_a:=\frac{1}{\lambda_a}\begin{pmatrix}w_{-a}&\rho_a-w_aw_{-a}\\1&-w_a\end{pmatrix} γ a := λ a 1 ( w − a 1 ρ a − w a w − a − w a )
has determinant − ρ a / λ a 2 = 1 -\rho_a/\lambda_a^2=1 − ρ a / λ a 2 = 1 , thus γ ~ a ∈ S L 2 ( C ) \widetilde{\gamma}_a\in\mathrm{SL}_2(\mathbb{C}) γ a ∈ SL 2 ( C ) lifts γ a \gamma_a γ a . For γ ~ = ( a b c d ) ∈ S L 2 ( C ) \widetilde{\gamma}=\begin{pmatrix}a&b\\c&d\end{pmatrix}\in\mathrm{SL}_2(\mathbb{C}) γ = ( a c b d ) ∈ SL 2 ( C ) , j ( γ ~ , z ) : = ( c z + d ) − 1 j(\widetilde{\gamma},z):=(cz+d)^{-1} j ( γ , z ) := ( cz + d ) − 1 is, by definition, a factor of automorphy: a function on S L 2 ( C ) × C \mathrm{SL}_2(\mathbb{C})\times\mathbb{C} SL 2 ( C ) × C (not a globally defined square root of γ ′ \gamma' γ ′ ) satisfying the cocycle identity j ( γ ~ μ ~ , z ) = j ( γ ~ , μ z ) j ( μ ~ , z ) j(\widetilde{\gamma}\widetilde{\mu},z)=j(\widetilde{\gamma},\mu z)j(\widetilde{\mu},z) j ( γ μ , z ) = j ( γ , μ z ) j ( μ , z ) . Only the induced action on sections (as in the spin bundle below) is globally meaningful. One has j ( γ ~ , z ) 2 = γ ′ ( z ) j(\widetilde{\gamma},z)^2=\gamma'(z) j ( γ , z ) 2 = γ ′ ( z ) and j ( γ ~ − 1 , γ z ) = j ( γ ~ , z ) − 1 j(\widetilde{\gamma}^{-1},\gamma z)=j(\widetilde{\gamma},z)^{-1} j ( γ − 1 , γ z ) = j ( γ , z ) − 1 , and for all c ∈ I c\in I c ∈ I
j ( γ ~ c , z ) = λ c z − w c . (1) j(\widetilde{\gamma}_c,z)=\frac{\lambda_c}{z-w_c}. \tag*{(1)} j ( γ c , z ) = z − w c λ c . (1)
Thus, the pair ( ρ a , sign ε a (\rho_a,\operatorname{sign}\varepsilon_a ( ρ a , sign ε a relative to a fixed branch of ρ a 1 / 2 ) \rho_a^{1/2}) ρ a 1/2 ) is encoded in the single parameter λ a \lambda_a λ a . Changing ε a \varepsilon_a ε a is λ a ↦ − λ a \lambda_a\mapsto-\lambda_a λ a ↦ − λ a . Let Γ ~ = ⟨ γ ~ a ⟩ ≅ Γ \widetilde{\Gamma}=\langle\widetilde{\gamma}_a\rangle\cong\Gamma Γ = ⟨ γ a ⟩ ≅ Γ (free, thus this is a homomorphic section; Γ ~ \widetilde{\Gamma} Γ and Γ \Gamma Γ are kept distinct).
Standard facts
In this subsection we recall standard facts from [11 , 13 ] . Let D c = { ∣ z − w c ∣ ≤ r c } D_c=\{|z-w_c|\le r_c\} D c = { ∣ z − w c ∣ ≤ r c } and consider the condition
(D) the 2 g 2g 2 g discs D c D_c D c are pairwise disjoint.
Condition (H) implies (D), since d > 2 r d>2r d > 2 r there. All of this subsection and Proposition 2.1 below use only (D).
Let us assume (D). γ c \gamma_c γ c maps { ∣ z − w c ∣ > r c } \{|z-w_c|>r_c\} { ∣ z − w c ∣ > r c } into D − c D_{-c} D − c . Hence Γ \Gamma Γ is a free classical Schottky group. D ‾ : = C ^ ∖ ⋃ c D c ∘ \overline{D}:=\widehat{\mathbb{C}}\setminus\bigcup_c D_c^\circ D := C ∖ ⋃ c D c ∘ is a closed fundamental domain with side pairings γ a : ∂ D a → ∂ D − a \gamma_a:\partial D_a\to\partial D_{-a} γ a : ∂ D a → ∂ D − a , so that Ω 0 = Γ D ‾ \Omega_0=\Gamma\overline{D} Ω 0 = Γ D . Every point of the region of discontinuity Ω 0 ⊂ C ^ \Omega_0\subset\widehat{\mathbb{C}} Ω 0 ⊂ C is a Γ \Gamma Γ -translate of a point of D ‾ \overline{D} D . Boundary points may have two such representatives, which does not affect the existence statements used below. Note that D ‾ ⊂ C ^ \overline{D}\subset\widehat{\mathbb{C}} D ⊂ C contains ∞ \infty ∞ . Hence, ∞ ∈ Ω 0 \infty\in\Omega_0 ∞ ∈ Ω 0 , and the orbit Γ ∞ ⊂ Ω 0 \Gamma\infty\subset\Omega_0 Γ∞ ⊂ Ω 0 consists of ∞ \infty ∞ together with the finite points γ ( ∞ ) \gamma(\infty) γ ( ∞ ) , γ ≠ 1 \gamma\ne1 γ = 1 (for instance γ c ( ∞ ) = w − c \gamma_c(\infty)=w_{-c} γ c ( ∞ ) = w − c ). The image of ∞ \infty ∞ in Σ \Sigma Σ is an ordinary point of Σ \Sigma Σ . The scalar trivialization by ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 used below is singular along Γ ∞ \Gamma\infty Γ∞ , which is why ∞ \infty ∞ is treated in homogeneous coordinates in Section 3. Σ = Ω 0 / Γ \Sigma=\Omega_0/\Gamma Σ = Ω 0 /Γ has genus g g g . Every γ ≠ 1 \gamma\ne1 γ = 1 in Γ \Gamma Γ is loxodromic with both fixed points in the limit set Λ \Lambda Λ , thus the action of Γ \Gamma Γ on Ω 0 \Omega_0 Ω 0 is free and properly discontinuous (standard for classical Schottky groups, loc. cit.). Consequently, for fixed x ∈ Ω 0 x\in\Omega_0 x ∈ Ω 0 , the orbit Γ x \Gamma x Γ x is a locally finite subset of Ω 0 \Omega_0 Ω 0 . It accumulates only on the limit set Λ ⊂ ⋃ c D c ∘ \Lambda\subset\bigcup_c D_c^\circ Λ ⊂ ⋃ c D c ∘ , which is disjoint from Ω 0 \Omega_0 Ω 0 . Each point of Γ x \Gamma x Γ x has a unique preimage group element, since the action is free. We will also use proper discontinuity in the following form: for compact K , K ′ ⊂ Ω 0 K,K'\subset\Omega_0 K , K ′ ⊂ Ω 0 the set { γ ∈ Γ : γ K ∩ K ′ ≠ ∅ } \{\gamma\in\Gamma:\gamma K\cap K'\ne\varnothing\} { γ ∈ Γ : γ K ∩ K ′ = ∅ } is finite. In particular, the family of graphs { ( x , γ x ) } γ ∈ Γ \{(x,\gamma x)\}_{\gamma\in\Gamma} {( x , γ x ) } γ ∈ Γ in Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 is locally finite, not only each single orbit. Every point of Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 has a neighbourhood K × K ′ K\times K' K × K ′ meeting only finitely many graphs.
The spin bundle
L ε L_\varepsilon L ε is the line bundle on Σ \Sigma Σ whose sections are ψ ( z ) ( d z ) 1 / 2 \psi(z)(dz)^{1/2} ψ ( z ) ( d z ) 1/2 on Ω 0 ∖ Γ ∞ \Omega_0\setminus\Gamma\infty Ω 0 ∖ Γ∞ with ψ ( γ z ) j ( γ ~ , z ) = ψ ( z ) \psi(\gamma z)j(\widetilde{\gamma},z)=\psi(z) ψ ( γ z ) j ( γ , z ) = ψ ( z ) for γ ~ ∈ Γ ~ \widetilde{\gamma}\in\widetilde{\Gamma} γ ∈ Γ . The trivialization by ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 is singular at Γ ∞ \Gamma\infty Γ∞ . Across these points a section is extended by the transformation law, equivalently, (see Section 3) L ε L_\varepsilon L ε is the quotient by Γ ~ ⊂ S L 2 ( C ) \widetilde{\Gamma}\subset\mathrm{SL}_2(\mathbb{C}) Γ ⊂ SL 2 ( C ) of the restriction to Ω 0 \Omega_0 Ω 0 of the S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) -equivariant bundle K P 1 1 / 2 K_{\mathbb{P}^1}^{1/2} K P 1 1/2 . Since j 2 = γ ′ j^2=\gamma' j 2 = γ ′ , L ε ⊗ 2 ≅ K Σ L_\varepsilon^{\otimes2}\cong K_\Sigma L ε ⊗ 2 ≅ K Σ , and deg L ε = g − 1 \deg L_\varepsilon=g-1 deg L ε = g − 1 . The next proposition concerns the family of spin structures.
Proposition 2.1. Let us assume the condition (D). Then:
(a) The N g N_g N g lifts of Γ \Gamma Γ (equivalently, the N g N_g N g sign choices λ a ↦ ± λ a \lambda_a\mapsto\pm\lambda_a λ a ↦ ± λ a ) give N g N_g N g pairwise non-isomorphic spin structures L ε L_\varepsilon L ε .
(b) They are the spin structures L L L for which the pull-back π ∗ L \pi^*L π ∗ L to Ω 0 \Omega_0 Ω 0 is the trivial spin structure, i.e., for which d z dz d z has a single valued square root ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 as a section of π ∗ L \pi^*L π ∗ L over Ω 0 \Omega_0 Ω 0 .
(c) For the marking with α a = ∂ D a \alpha_a=\partial D_a α a = ∂ D a their characteristics are δ = [ 0 ; δ ′ ′ ] \delta=[0;\delta^{\prime\prime}] δ = [ 0 ; δ ′′ ] , δ ′ ′ ∈ { 0 , 1 2 } g \delta^{\prime\prime}\in\{0,\frac{1}{2}\}^{g} δ ′′ ∈ { 0 , 2 1 } g , all even (proven in Proposition 3.16 (b)).
We do not discuss here which characteristic corresponds to which sign vector for g ≥ 2 g\geq2 g ≥ 2 (see Remark 3.20). The remaining 2 2 g − N g 2^{2g}-N_g 2 2 g − N g spin structures, both even structures not of this form and every odd structure, are not treated. Note that in (b) above we do not claim that π ∗ L \pi^*L π ∗ L is trivial as a holomorphic line bundle, which holds for every L L L because Ω 0 \Omega_0 Ω 0 is a plane domain, nor does every spin structure admit such a square root.
Proof. (a) Lifts: a lift of γ a \gamma_a γ a to S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) is ± γ ~ a \pm\widetilde{\gamma}_a ± γ a , i.e., λ a ↦ ± λ a \lambda_a\mapsto\pm\lambda_a λ a ↦ ± λ a . Since Γ \Gamma Γ is free on γ 1 , … , γ g \gamma_1,\ldots,\gamma_g γ 1 , … , γ g , every assignment of lifts of the generators extends uniquely to a homomorphic section Γ ~ → S L 2 ( C ) \widetilde{\Gamma}\to\mathrm{SL}_2(\mathbb{C}) Γ → SL 2 ( C ) . This gives the N g N_g N g groups Γ ~ \widetilde{\Gamma} Γ .
Bundles: since j ~ ( γ ~ , z ) 2 = γ ′ ( z ) \widetilde{j}(\widetilde{\gamma},z)^2=\gamma'(z) j ( γ , z ) 2 = γ ′ ( z ) and j ~ \widetilde{j} j satisfies the cocycle identity, γ ~ ⋅ ( z , v ) : = ( γ z , j ~ ( γ ~ , z ) − 1 v ) \widetilde{\gamma}\cdot(z,v):=(\gamma z,\widetilde{j}(\widetilde{\gamma},z)^{-1}v) γ ⋅ ( z , v ) := ( γ z , j ( γ , z ) − 1 v ) is an action on Ω 0 × C \Omega_0\times\mathbb{C} Ω 0 × C , free and properly discontinuous on the base. L ε : = ( Ω 0 × C ) / Γ ~ L_\varepsilon:=(\Omega_0\times\mathbb{C})/\widetilde{\Gamma} L ε := ( Ω 0 × C ) / Γ is a line bundle on Σ \Sigma Σ with L ε ⊗ 2 ≅ K Σ L_\varepsilon^{\otimes2}\cong K_\Sigma L ε ⊗ 2 ≅ K Σ . The section v = 1 v=1 v = 1 trivializes the pull-back, and corresponds to ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 . Two lifts differ by a character χ : Γ → { ± 1 } \chi:\Gamma\to\{\pm1\} χ : Γ → { ± 1 } , and L ε ′ ≅ L ε ⊗ F χ L_{\varepsilon'}\cong L_\varepsilon\otimes F_\chi L ε ′ ≅ L ε ⊗ F χ with F χ F_\chi F χ the flat unitary line bundle of χ \chi χ , viewed as a character of π 1 ( Σ ) \pi_1(\Sigma) π 1 ( Σ ) through the surjection π 1 ( Σ ) → Γ \pi_1(\Sigma)\to\Gamma π 1 ( Σ ) → Γ . Distinct unitary characters of π 1 ( Σ ) \pi_1(\Sigma) π 1 ( Σ ) give non-isomorphic holomorphic line bundles, (injectivity of the map from unitary characters to Pic 0 \operatorname{Pic}^0 Pic 0 ), thus the N g N_g N g bundles L ε L_\varepsilon L ε are pairwise non-isomorphic.
(b) Spin structures on Σ \Sigma Σ form a torsor under H 1 ( Σ ; Z / 2 ) H^1(\Sigma;\mathbb{Z}/2) H 1 ( Σ ; Z /2 ) , and the restriction of a spin structure to Ω 0 \Omega_0 Ω 0 is trivial iff its class pulled back to Ω 0 \Omega_0 Ω 0 coincides with that of ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 . I.e., iff its holonomy along each generator α a = ∂ D a \alpha_a=\partial D_a α a = ∂ D a of π 1 ( Ω 0 ) \pi_1(\Omega_0) π 1 ( Ω 0 ) equals that of ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 (which is trivial, ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 being single valued on Ω 0 \Omega_0 Ω 0 ). Namely, 2 2 g / 2 g = N g 2^{2g}/2^g=N_g 2 2 g / 2 g = N g classes satisfy this (g g g independent conditions), and similarly the L ε L_\varepsilon L ε .
(c) is Proposition 3.16 (b).
Convention
For the fixed marking ( α a = ∂ D a , β a (\alpha_a=\partial D_a,\beta_a ( α a = ∂ D a , β a as in Proposition 3.16 ), theta characteristics δ ∈ ( 1 2 Z / Z ) 2 g \delta\in(\frac{1}{2}\mathbb{Z}/\mathbb{Z})^{2g} δ ∈ ( 2 1 Z / Z ) 2 g correspond bijectively to spin structures L δ L_\delta L δ (the spin structure whose Szegő kernel, when θ [ δ ] ( 0 ) ≠ 0 \theta[\delta](0)\ne0 θ [ δ ] ( 0 ) = 0 , is given by the formula of Proposition 3.16 ). For each lift ε \varepsilon ε we let δ ( ε ) \delta(\varepsilon) δ ( ε ) be the unique characteristic with L δ ( ε ) ≅ L ε L_{\delta(\varepsilon)}\cong L_\varepsilon L δ ( ε ) ≅ L ε . This is defined purely through the isomorphism class of the bundle L ε L_\varepsilon L ε (i.e., no choice of sign or branch). In what follows, we use δ ( ε ) \delta(\varepsilon) δ ( ε ) of this form. The numerical relation between the sign vector of ε \varepsilon ε and δ ′ ′ ( ε ) \delta^{\prime\prime}(\varepsilon) δ ′′ ( ε ) is a separate question (Proposition 3.19 for $g=1, {xref}Remark 3.20 for $g\ge2) on which no proof in this paper depends. We also mention the classical fact that (Riemann singularity theorem in its form for characteristics, see [13 , 9 ] )
θ [ δ ] ( 0 ) = 0 iff h 0 ( Σ , L δ ) > 0 , \theta[\delta](0)=0\ \text{ iff }\ h^0(\Sigma,L_\delta)>0, θ [ δ ] ( 0 ) = 0 iff h 0 ( Σ , L δ ) > 0 ,
valid for every characteristic δ \delta δ (even or odd), and we do prove it again here.
Let π ( x , y ) = ( d x d y ) 1 / 2 ( x − y ) − 1 \pi(x,y)=(dx\,dy)^{1/2}(x-y)^{-1} π ( x , y ) = ( d x d y ) 1/2 ( x − y ) − 1 . By g x − g y = x − y ( c x + d ) ( c y + a ) g x-g y=\frac{x-y}{(cx+d)(cy+a)} g x − g y = ( c x + d ) ( cy + a ) x − y it is jointly invariant, π ( g ~ x , g ~ y ) = π ( x , y ) \pi(\widetilde{g}x,\widetilde{g}y)=\pi(x,y) π ( g x , g y ) = π ( x , y ) for g ~ ∈ S L 2 ( C ) \widetilde{g}\in\mathrm{SL}_2(\mathbb{C}) g ∈ SL 2 ( C ) . In coordinates: j ~ ( g ~ , x ) j ~ ( g ~ , y ) / ( g x − g y ) = 1 / ( x − y ) \widetilde{j}(\widetilde{g},x)\widetilde{j}(\widetilde{g},y)/(gx-gy)=1/(x-y) j ( g , x ) j ( g , y ) / ( g x − g y ) = 1/ ( x − y ) .
Definition 3.1. For x , y ∈ Ω 0 x,y\in\Omega_0 x , y ∈ Ω 0 with y ∉ Γ x y\notin\Gamma x y ∈ / Γ x ,
s ε ( x , y ) : = ∑ γ ~ ∈ Γ ~ j ~ ( γ ~ , x ) γ x − y , S ε : = s ε ( x , y ) ( d x d y ) 1 / 2 . s_\varepsilon(x,y):=\sum_{\widetilde{\gamma}\in\widetilde{\Gamma}}\frac{\widetilde{j}(\widetilde{\gamma},x)}{\gamma x-y},
\qquad
S_\varepsilon:=s_\varepsilon(x,y)(dx\,dy)^{1/2}. s ε ( x , y ) := γ ∈ Γ ∑ γ x − y j ( γ , x ) , S ε := s ε ( x , y ) ( d x d y ) 1/2 .
We will use the following conventions: γ \gamma γ denotes the Möbius transformation underlying γ ~ \widetilde{\gamma} γ , j ~ ( γ ~ , x ) = ( c γ ~ x + d γ ~ ) − 1 \widetilde{j}(\widetilde{\gamma},x)=(c\widetilde{\gamma}x+d\widetilde{\gamma})^{-1} j ( γ , x ) = ( c γ x + d γ ) − 1 , and the first variable x x x is the one acted on by Γ ~ \widetilde{\Gamma} Γ , the second variable y y y the point at which the pole sits (y = γ x y=\gamma x y = γ x ). The homogeneous formulation below (eq. (2) ) is the same object. With x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) , y = ( y , 1 ) \mathbf{y}=(y,1) y = ( y , 1 ) one has [ γ ~ x , y ] = j ~ ( γ ~ , x ) − 1 ( γ x − y ) [\widetilde{\gamma}\mathbf{x},\mathbf{y}]=\widetilde{j}(\widetilde{\gamma},x)^{-1}(\gamma x-y) [ γ x , y ] = j ( γ , x ) − 1 ( γ x − y ) , thus the term 1 / [ γ ~ x , y ] 1/[\widetilde{\gamma}\mathbf{x},\mathbf{y}] 1/ [ γ x , y ] equals j ~ ( γ ~ , x ) / ( γ x − y ) \widetilde{j}(\widetilde{\gamma},x)/(\gamma x-y) j ( γ , x ) / ( γ x − y ) with no sign. The pairing is skew symmetric, [ x , y ] = − [ y , x ] [\mathbf{x},\mathbf{y}]=-[\mathbf{y},\mathbf{x}] [ x , y ] = − [ y , x ] , which is the source of the skew-symmetry of Proposition 3.7 (i). A change of chart ( x , 1 ) = x ( 1 , u ) (x,1)=x(1,u) ( x , 1 ) = x ( 1 , u ) , u = 1 / x u=1/x u = 1/ x , is absorbed by bihomogeneity of degree ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) (the constant ± i \pm i ± i coming from d x = − u − 2 d u dx=-u^{-2}du d x = − u − 2 d u enters only the identification of K 1 / 2 K^{1/2} K 1/2 with O ( − 1 ) \mathcal{O}(-1) O ( − 1 ) and is the same for all terms). That this series converges on the stated set is the content of Theorem 3.5 (until then it is only a formal expression). The definition makes sense only for finite x , y x,y x , y , and the point ∞ ∈ Ω 0 \infty\in\Omega_0 ∞ ∈ Ω 0 is dealt with by the following reformulation.
Homogeneous coordinates (treatment of ∞ \infty ∞ )
For x = ( x 1 , x 2 ) \mathbf{x}=(x_1,x_2) x = ( x 1 , x 2 ) , y = ( y 1 , y 2 ) ∈ C 2 ∖ 0 \mathbf{y}=(y_1,y_2)\in\mathbb{C}^2\setminus0 y = ( y 1 , y 2 ) ∈ C 2 ∖ 0 set ⟨ x , y ⟩ : = x 1 y 2 − x 2 y 1 \langle\mathbf{x},\mathbf{y}\rangle:=x_1y_2-x_2y_1 ⟨ x , y ⟩ := x 1 y 2 − x 2 y 1 . Then ⟨ g ~ x , g ~ y ⟩ = ⟨ x , y ⟩ \langle\widetilde{g}\mathbf{x},\widetilde{g}\mathbf{y}\rangle=\langle\mathbf{x},\mathbf{y}\rangle ⟨ g x , g y ⟩ = ⟨ x , y ⟩ for g ~ ∈ SL 2 ( C ) \widetilde{g}\in\operatorname{SL}_2(\mathbb{C}) g ∈ SL 2 ( C ) . The point x ∈ C x\in\mathbb{C} x ∈ C is represented by x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) , the point ∞ \infty ∞ by x = ( 1 , 0 ) \mathbf{x}=(1,0) x = ( 1 , 0 ) , and a point x x x of the chart at infinity by ( 1 , u ) (1,u) ( 1 , u ) , u = 1 / x u=1/x u = 1/ x . A section of K P 1 1 / 2 ≅ O ( − 1 ) K_{\mathbb{P}^1}^{1/2}\cong\mathcal{O}(-1) K P 1 1/2 ≅ O ( − 1 ) over V ⊂ C ^ V\subset\widehat{\mathbb{C}} V ⊂ C is a function f f f on the cone over V V V with f ( t x ) = t − 1 f ( x ) f(t\mathbf{x})=t^{-1}f(\mathbf{x}) f ( t x ) = t − 1 f ( x ) . In the chart x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) it is f ( x , 1 ) ( d x ) 1 / 2 f(x,1)(dx)^{1/2} f ( x , 1 ) ( d x ) 1/2 and in the chart x = ( 1 , u ) \mathbf{x}=(1,u) x = ( 1 , u ) it is f ( 1 , u ) ( d u ) 1 / 2 f(1,u)(du)^{1/2} f ( 1 , u ) ( d u ) 1/2 (a constant factor ± i \pm i ± i from d x = − u − 2 d u dx=-u^{-2}du d x = − u − 2 d u is absorbed in the identification K 1 / 2 ≅ O ( − 1 ) K^{1/2}\cong\mathcal{O}(-1) K 1/2 ≅ O ( − 1 ) and plays no role). For g ~ = ( α β κ ς ) ∈ SL 2 ( C ) \widetilde{g}=\begin{pmatrix}\alpha&\beta\\\kappa&\varsigma\end{pmatrix}\in\operatorname{SL}_2(\mathbb{C}) g = ( α κ β ς ) ∈ SL 2 ( C ) one has g ~ ( x , 1 ) = ( κ x + ς ) ( g x , 1 ) = j ( g ~ , x ) − 1 ( g x , 1 ) \widetilde{g}(x,1)=(\kappa x+\varsigma)(gx,1)=j(\widetilde{g},x)^{-1}(gx,1) g ( x , 1 ) = ( κ x + ς ) ( g x , 1 ) = j ( g , x ) − 1 ( g x , 1 ) . Hence f ( g ~ x ) = j ( g ~ , x ) f ( g x , 1 ) f(\widetilde{g}\mathbf{x})=j(\widetilde{g},x)f(gx,1) f ( g x ) = j ( g , x ) f ( g x , 1 ) for x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) . Invariance f ( g ~ x ) = f ( x ) f(\widetilde{g}\mathbf{x})=f(\mathbf{x}) f ( g x ) = f ( x ) is the automorphy law ψ ( g x ) j ( g ~ , x ) = ψ ( x ) \psi(gx)j(\widetilde{g},x)=\psi(x) ψ ( g x ) j ( g , x ) = ψ ( x ) defining L ε L_\varepsilon L ε . Thus, L ε = O ( − 1 ) ∣ Ω 0 / Γ ~ L_\varepsilon=\mathcal{O}(-1)|_{\Omega_0}/\widetilde{\Gamma} L ε = O ( − 1 ) ∣ Ω 0 / Γ , and sections of L ε ~ \widetilde{L_\varepsilon} L ε are the Γ ~ \widetilde{\Gamma} Γ -invariant functions f ~ \widetilde{f} f . Define, for x , y ∈ C 2 ∖ 0 \mathbf{x},\mathbf{y}\in\mathbb{C}^2\setminus0 x , y ∈ C 2 ∖ 0 over Ω 0 \Omega_0 Ω 0 with [ y ] ∉ Γ [ x ] [\mathbf{y}]\notin\Gamma[\mathbf{x}] [ y ] ∈ / Γ [ x ] ,
s ^ ε ( x , y ) : = ∑ γ ~ ∈ Γ ~ 1 ⟨ γ ~ x , y ⟩ , s ^ ε ( t x , t ′ y ) = t − 1 t ′ − 1 s ^ ε ( x , y ) . (2) \widehat{s}_\varepsilon(\mathbf{x},\mathbf{y}):=\sum_{\widetilde{\gamma}\in\widetilde{\Gamma}}\frac{1}{\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle},\qquad\widehat{s}_\varepsilon(t\mathbf{x},t'\mathbf{y})=t^{-1}{t'}^{-1}\widehat{s}_\varepsilon(\mathbf{x},\mathbf{y}). \tag*{(2)} s ε ( x , y ) := γ ∈ Γ ∑ ⟨ γ x , y ⟩ 1 , s ε ( t x , t ′ y ) = t − 1 t ′ − 1 s ε ( x , y ) . (2)
For x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) , y = ( y , 1 ) \mathbf{y}=(y,1) y = ( y , 1 ) one computes ⟨ γ ~ x , y ⟩ = j ( γ ~ , x ) − 1 ( γ x − y ) \langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle=j(\widetilde{\gamma},x)^{-1}(\gamma x-y) ⟨ γ x , y ⟩ = j ( γ , x ) − 1 ( γ x − y ) . Thus, s ^ ε ( ( x , 1 ) , ( y , 1 ) ) = s ε ( x , y ) \widehat{s}_\varepsilon((x,1),(y,1))=s_\varepsilon(x,y) s ε (( x , 1 ) , ( y , 1 )) = s ε ( x , y ) is the series of Definition 3.1. At x = ( 1 , 0 ) \mathbf{x}=(1,0) x = ( 1 , 0 ) (the point ∞ \infty ∞ ) s ^ ε \widehat{s}_\varepsilon s ε is the same series in the other chart, and ⟨ γ ~ ( 1 , 0 ) , ( y , 1 ) ⟩ = \langle\widetilde{\gamma}(1,0),(y,1)\rangle= ⟨ γ ( 1 , 0 ) , ( y , 1 )⟩ = (first column of γ ~ \widetilde{\gamma} γ ) paired with ( y , 1 ) = a γ ~ − c γ ~ y = c γ ~ ( γ ∞ − y ) (y,1)=a_{\widetilde{\gamma}}-c_{\widetilde{\gamma}}y=c_{\widetilde{\gamma}}(\gamma\infty-y) ( y , 1 ) = a γ − c γ y = c γ ( γ ∞ − y ) (for γ ~ = ( a γ ~ ∗ c γ ~ ∗ ) \widetilde{\gamma}=\begin{pmatrix}a_{\widetilde{\gamma}}&*\\c_{\widetilde{\gamma}}&*\end{pmatrix} γ = ( a γ c γ ∗ ∗ ) , c γ ~ ≠ 0 c_{\widetilde{\gamma}}\ne0 c γ = 0 for γ ~ ≠ 1 \widetilde{\gamma}\ne1 γ = 1 ).
Let K = K p = { z ∈ C : ∣ z − w c ∣ ≥ r c / 2 for all c } K=K_p=\{z\in\mathbb{C}:|z-w_c|\ge r_c/2\text{ for all }c\} K = K p = { z ∈ C : ∣ z − w c ∣ ≥ r c /2 for all c } , K ^ : = K ∪ { ∞ } ⊂ C ^ \widehat{K}:=K\cup\{\infty\}\subset\widehat{\mathbb{C}} K := K ∪ { ∞ } ⊂ C , a compact set containing D ‾ \overline{D} D in its interior, and ϖ : = 2 r / d ≤ 1 4 ( 2 g − 1 ) \varpi:=2r/d\le\frac{1}{4(2g-1)} ϖ := 2 r / d ≤ 4 ( 2 g − 1 ) 1 . Set W : = max c ∣ w c ∣ W:=\max_c|w_c| W := max c ∣ w c ∣ and X 0 : = 2 ( W + r ) X_0:=2(W+r) X 0 := 2 ( W + r ) . A normalized representative of a point α ∈ K ^ \alpha\in\widehat{K} α ∈ K is x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) if α = x \alpha=x α = x is finite with ∣ x ∣ ≤ 2 X 0 |x|\le2X_0 ∣ x ∣ ≤ 2 X 0 , and x = ( 1 , u ) \mathbf{x}=(1,u) x = ( 1 , u ) with ∣ u ∣ ≤ 1 / X 0 |u|\le1/X_0 ∣ u ∣ ≤ 1/ X 0 (u = 0 u=0 u = 0 for α = ∞ \alpha=\infty α = ∞ ) if ∣ α ∣ ≥ X 0 |\alpha|\ge X_0 ∣ α ∣ ≥ X 0 or α = ∞ \alpha=\infty α = ∞ . Points with X 0 ≤ ∣ α ∣ ≤ 2 X 0 X_0\le|\alpha|\le2X_0 X 0 ≤ ∣ α ∣ ≤ 2 X 0 have both. Two charts are then needed: no single normalization of x \mathbf{x} x is bounded near ∞ \infty ∞ and at the same time bounded away from 0 0 0 on all of K ^ \widehat{K} K using a single formula.
Let x ∈ K x\in K x ∈ K and let γ ~ = γ ~ c 1 ⋯ γ ~ c n ∈ Γ ~ \widetilde{\gamma}=\widetilde{\gamma}_{c_1}\cdots\widetilde{\gamma}_{c_n}\in\widetilde{\Gamma} γ = γ c 1 ⋯ γ c n ∈ Γ (n ≥ 1 n\ge1 n ≥ 1 ) be a reduced word, i.e., γ = γ c 1 ∘ ⋯ ∘ γ c n \gamma=\gamma_{c_1}\circ\cdots\circ\gamma_{c_n} γ = γ c 1 ∘ ⋯ ∘ γ c n with c i + 1 ≠ − c i c_{i+1}\ne-c_i c i + 1 = − c i for 1 ≤ i ≤ n − 1 1\le i\le n-1 1 ≤ i ≤ n − 1 (so that γ \gamma γ first applies γ c n \gamma_{c_n} γ c n to x x x , then γ c n − 1 \gamma_{c_{n-1}} γ c n − 1 , and so on, ending with γ c 1 \gamma_{c_1} γ c 1 ). Then we have
Lemma 3.2. ∣ j ( γ ~ , x ) ∣ ≤ 2 ϖ n − 1 |j(\widetilde{\gamma},x)|\le2\varpi^{n-1} ∣ j ( γ , x ) ∣ ≤ 2 ϖ n − 1 . Moreover ∣ γ x − w − c 1 ∣ ≤ 2 r c 1 |\gamma x-w_{-c_1}|\le2r_{c_1} ∣ γ x − w − c 1 ∣ ≤ 2 r c 1 for n = 1 n=1 n = 1 and ≤ r c 1 ϖ ≤ r c 1 / 4 \le r_{c_1}\varpi\le r_{c_1}/4 ≤ r c 1 ϖ ≤ r c 1 /4 for n ≥ 2 n\ge2 n ≥ 2 . There are 2 g ( 2 g − 1 ) n − 1 2g(2g-1)^{n-1} 2 g ( 2 g − 1 ) n − 1 reduced words of length n n n .
Proof. Set z n : = x z_n:=x z n := x and, for i = n , n − 1 , … , 1 i=n,n-1,\ldots,1 i = n , n − 1 , … , 1 , z i − 1 : = γ c i ( z i ) z_{i-1}:=\gamma_{c_i}(z_i) z i − 1 := γ c i ( z i ) . Thus, z 0 = γ c 1 ( γ c 2 ( ⋯ γ c n ( x ) ⋯ ) ) = γ x z_0=\gamma_{c_1}(\gamma_{c_2}(\cdots\gamma_{c_n}(x)\cdots))=\gamma x z 0 = γ c 1 ( γ c 2 ( ⋯ γ c n ( x ) ⋯ )) = γ x .
By the cocycle identity j ( γ ~ , z n ) = ∏ i = 1 n j ( γ ~ c i , z i ) j(\widetilde{\gamma},z_n)=\prod_{i=1}^n j(\widetilde{\gamma}_{c_i},z_i) j ( γ , z n ) = ∏ i = 1 n j ( γ c i , z i ) (the factor j ( γ ~ c i , z i ) j(\widetilde{\gamma}_{c_i},z_i) j ( γ c i , z i ) is evaluated at z i z_i z i , the point on which γ ~ c i \widetilde{\gamma}_{c_i} γ c i acts before γ ~ c i − 1 , … , γ ~ c 1 \widetilde{\gamma}_{c_{i-1}},\ldots,\widetilde{\gamma}_{c_1} γ c i − 1 , … , γ c 1 are applied). We estimate the factors in the order i = n , n − 1 , … , 1 i=n,n-1,\ldots,1 i = n , n − 1 , … , 1 . I.e., starting from x = z n x=z_n x = z n and working outward. By (1), ∣ j ( γ ~ c n , z n ) ∣ = r c n / ∣ z n − w c n ∣ ≤ 2 |j(\widetilde{\gamma}_{c_n},z_n)|=r_{c_n}/|z_n-w_{c_n}|\le2 ∣ j ( γ c n , z n ) ∣ = r c n /∣ z n − w c n ∣ ≤ 2 (as z n = x ∈ K z_n=x\in K z n = x ∈ K , ∣ z n − w c n ∣ ≥ r c n / 2 |z_n-w_{c_n}|\ge r_{c_n}/2 ∣ z n − w c n ∣ ≥ r c n /2 ) and ∣ z n − 1 − w − c n ∣ = r c n 2 / ∣ z n − w c n ∣ ≤ 2 r |z_{n-1}-w_{-c_n}|=r_{c_n}^2/|z_n-w_{c_n}|\le2r ∣ z n − 1 − w − c n ∣ = r c n 2 /∣ z n − w c n ∣ ≤ 2 r . For $1\le i\le n-1: c i ≠ − c i + 1 c_i\ne-c_{i+1} c i = − c i + 1 and ∣ w − c i + 1 − w c i ∣ ≥ d |w_{-c_{i+1}}-w_{c_i}|\ge d ∣ w − c i + 1 − w c i ∣ ≥ d , the bound ∣ z i − w − c i + 1 ∣ ≤ 2 r c i + 1 ≤ 2 r |z_i-w_{-c_{i+1}}|\le2r_{c_{i+1}}\le2r ∣ z i − w − c i + 1 ∣ ≤ 2 r c i + 1 ≤ 2 r already established gives ∣ z i − w c i ∣ ≥ d − 2 r ≥ d / 2 |z_i-w_{c_i}|\ge d-2r\ge d/2 ∣ z i − w c i ∣ ≥ d − 2 r ≥ d /2 by (H). Then ∣ j ( γ ~ c i , z i ) ∣ = r c i / ∣ z i − w c i ∣ ≤ 2 r c i / d ≤ ϖ |j(\widetilde{\gamma}_{c_i},z_i)|=r_{c_i}/|z_i-w_{c_i}|\le2r_{c_i}/d\le\varpi ∣ j ( γ c i , z i ) ∣ = r c i /∣ z i − w c i ∣ ≤ 2 r c i / d ≤ ϖ and ∣ z i − 1 − w − c i ∣ = r c i 2 / ∣ z i − w c i ∣ ≤ r c i ϖ ≤ r c i / 4 |z_{i-1}-w_{-c_i}|=r_{c_i}^2/|z_i-w_{c_i}|\le r_{c_i}\varpi\le r_{c_i}/4 ∣ z i − 1 − w − c i ∣ = r c i 2 /∣ z i − w c i ∣ ≤ r c i ϖ ≤ r c i /4 (using ϖ ≤ 1 4 ( 2 g − 1 ) ≤ 1 4 \varpi\le\frac{1}{4(2g-1)}\le\frac14 ϖ ≤ 4 ( 2 g − 1 ) 1 ≤ 4 1 ). Multiplying the n n n bounds ∣ j ( γ ~ c n , z n ) ∣ ≤ 2 |j(\widetilde{\gamma}_{c_n},z_n)|\le2 ∣ j ( γ c n , z n ) ∣ ≤ 2 and ∣ j ( γ ~ c i , z i ) ∣ ≤ ϖ |j(\widetilde{\gamma}_{c_i},z_i)|\le\varpi ∣ j ( γ c i , z i ) ∣ ≤ ϖ (i < n i<n i < n ) gives ∣ j ( γ ~ , x ) ∣ ≤ 2 ϖ n − 1 |j(\widetilde{\gamma},x)|\le2\varpi^{n-1} ∣ j ( γ , x ) ∣ ≤ 2 ϖ n − 1 . The case i = 1 i=1 i = 1 of the second estimate gives ∣ z 0 − w − c 1 ∣ = ∣ γ x − w − c 1 ∣ ≤ r c 1 ϖ ≤ r c 1 / 4 |z_0-w_{-c_1}|=|\gamma x-w_{-c_1}|\le r_{c_1}\varpi\le r_{c_1}/4 ∣ z 0 − w − c 1 ∣ = ∣ γ x − w − c 1 ∣ ≤ r c 1 ϖ ≤ r c 1 /4 for n > 2 n>2 n > 2 (for n = 1 n=1 n = 1 , ∣ γ x − w − c 1 ∣ ≤ 2 r c 1 |\gamma x-w_{-c_1}|\le2r_{c_1} ∣ γ x − w − c 1 ∣ ≤ 2 r c 1 ). Counting reduced words of length n n n over the alphabet Γ ~ \widetilde{\Gamma} Γ (∣ Γ ~ ∣ = 2 g |\widetilde{\Gamma}|=2g ∣ Γ ∣ = 2 g ) with no two consecutive inverse letters is the standard count 2 g ( 2 g − 1 ) n − 1 2g(2g-1)^{n-1} 2 g ( 2 g − 1 ) n − 1 . ▢
The next lemma describes the chart-normalized word estimate.
Lemma 3.3. Let α , β ∈ K ^ \alpha,\beta\in\widehat{K} α , β ∈ K with normalized representatives x , y \mathbf{x},\mathbf{y} x , y , and let γ ~ = γ ~ c 1 ⋯ γ ~ c n ∈ Γ ~ \widetilde{\gamma}=\widetilde{\gamma}_{c_1}\cdots\widetilde{\gamma}_{c_n}\in\widetilde{\Gamma} γ = γ c 1 ⋯ γ c n ∈ Γ be a reduced word of length n ≥ 2 n\ge2 n ≥ 2 . Then
∣ ⟨ γ ~ x , y ⟩ ∣ − 1 ≤ C ∗ ϖ n − 2 , C ∗ : = 16 d + 1 + r . |\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle|^{-1}\le C_*\varpi^{n-2},\qquad C_*:=\frac{16}{d}+1+r. ∣ ⟨ γ x , y ⟩ ∣ − 1 ≤ C ∗ ϖ n − 2 , C ∗ := d 16 + 1 + r .
Note that the constant depends on the data only through d d d and r r r . It does not involve r min r_{\min} r m i n , α \alpha α or β \beta β . It is not coordinate-invariant. The normalized representatives ( x , 1 ) (x,1) ( x , 1 ) , ( 1 , u ) (1,u) ( 1 , u ) refer to the fixed affine coordinate on C ^ \widehat{\mathbb{C}} C . Thus, C ∗ C_* C ∗ adds quantities of different dimensions, 1 / d 1/d 1/ d and 1 + r 1+r 1 + r . But this is not a problem since it is repaired by measuring lengths in units of d d d .
Proof. Set z n : = α z_n := \alpha z n := α and z i − 1 : = γ c i ( z i ) z_{i-1} := \gamma_{c_i}(z_i) z i − 1 := γ c i ( z i ) for i = n , … , 1 i = n, \ldots, 1 i = n , … , 1 . Since γ c \gamma_c γ c maps C ∖ D c ∘ \mathbb{C}\setminus D_c^\circ C ∖ D c ∘ into D − c ⊂ C D_{-c}\subset\mathbb{C} D − c ⊂ C , all z i z_i z i with i ≤ n − 1 i\le n-1 i ≤ n − 1 are finite. For v = ( v 1 , v 2 ) ∈ C 2 \mathbf{v}=(v_1,v_2)\in\mathbb{C}^2 v = ( v 1 , v 2 ) ∈ C 2 and c ∈ I c\in I c ∈ I ,
γ ~ c v = 1 λ c ( w − c v 1 + ( ρ c − w c w − c ) v 2 , v 1 − w c v 2 ) . \widetilde{\gamma}_c\mathbf{v}=\frac{1}{\lambda_c}\left(w_{-c}v_1+(\rho_c-w_cw_{-c})v_2,\ v_1-w_cv_2\right). γ c v = λ c 1 ( w − c v 1 + ( ρ c − w c w − c ) v 2 , v 1 − w c v 2 ) .
so that γ ~ c ( μ ( z , 1 ) ) = μ z − w c λ c ( γ c z , 1 ) \widetilde{\gamma}_c\bigl(\mu(z,1)\bigr)=\mu\frac{z-w_c}{\lambda_c}\bigl(\gamma_cz,1\bigr) γ c ( μ ( z , 1 ) ) = μ λ c z − w c ( γ c z , 1 ) . Applying this n n n times (the first time to x \mathbf{x} x ),
γ ~ x = μ 1 μ 2 ⋯ μ n ( z 0 , 1 ) , μ i = z i − w c i λ c i , 1 ≤ i ≤ n − 1 , (3) \widetilde{\gamma}\mathbf{x}=\mu_1\mu_2\cdots\mu_n\,(z_0,1),\qquad\mu_i=\frac{z_i-w_{c_i}}{\lambda_{c_i}},\qquad1\le i\le n-1, \tag*{(3)} γ x = μ 1 μ 2 ⋯ μ n ( z 0 , 1 ) , μ i = λ c i z i − w c i , 1 ≤ i ≤ n − 1 , (3)
where μ n \mu_n μ n is the second component of γ ~ c n x \widetilde{\gamma}_{c_n}\mathbf{x} γ c n x , z 0 = γ x z_0=\gamma\mathbf{x} z 0 = γ x .
(a) The first part. If x = ( x , 1 ) \mathbf{x}=(x,1) x = ( x , 1 ) with x ∈ K x\in K x ∈ K , then μ n = ( x − w c n ) / λ c n \mu_n=(x-w_{c_n})/\lambda_{c_n} μ n = ( x − w c n ) / λ c n . Thus, ∣ μ n ∣ ≥ 1 2 |\mu_n|\ge\frac12 ∣ μ n ∣ ≥ 2 1 , and
∣ z n − 1 − w − c n ∣ = r c n 2 ∣ x − w c n ∣ ≤ 2 r c n . |z_{n-1}-w_{-c_n}|=\frac{r_{c_n}^2}{|x-w_{c_n}|}\le2r_{c_n}. ∣ z n − 1 − w − c n ∣ = ∣ x − w c n ∣ r c n 2 ≤ 2 r c n .
If x = ( 1 , u ) \mathbf{x}=(1,u) x = ( 1 , u ) with ∣ u ∣ ≤ 1 / X 0 |u|\le1/X_0 ∣ u ∣ ≤ 1/ X 0 (then α = 1 / u \alpha=1/u α = 1/ u belongs outside all D c D_c D c , since X 0 > W + r X_0>W+r X 0 > W + r ), then μ n = ( 1 − u w c n ) / λ c n \mu_n=(1-uw_{c_n})/\lambda_{c_n} μ n = ( 1 − u w c n ) / λ c n with ∣ 1 − u w c n ∣ ≥ 1 − W / X 0 ≥ 1 2 |1-uw_{c_n}|\ge1-W/X_0\ge\frac12 ∣1 − u w c n ∣ ≥ 1 − W / X 0 ≥ 2 1 . Thus ∣ μ n ∣ ≥ 1 / ( 2 r c n ) ≥ 1 / ( 2 r ) |\mu_n|\ge1/(2r_{c_n})\ge1/(2r) ∣ μ n ∣ ≥ 1/ ( 2 r c n ) ≥ 1/ ( 2 r ) . Moreover, z n − 1 − w − c n = ρ c n u / ( 1 − u w c n ) z_{n-1}-w_{-c_n}=\rho_{c_n}u/(1-uw_{c_n}) z n − 1 − w − c n = ρ c n u / ( 1 − u w c n ) . Hence, ∣ z n − 1 − w − c n ∣ ≤ 2 r c n 2 / X 0 ≤ r c n |z_{n-1}-w_{-c_n}|\le2r_{c_n}^2/X_0\le r_{c_n} ∣ z n − 1 − w − c n ∣ ≤ 2 r c n 2 / X 0 ≤ r c n as X 0 ≥ 2 r X_0\ge2r X 0 ≥ 2 r .
(b) The later parts. Let 1 ≤ i ≤ n − 1 1\le i\le n-1 1 ≤ i ≤ n − 1 and assume ∣ z i − w − c i + 1 ∣ ≤ 2 r |z_i-w_{-c_{i+1}}|\le2r ∣ z i − w − c i + 1 ∣ ≤ 2 r which is true for i = n − 1 i=n-1 i = n − 1 by (a). Since c i ≠ − c i + 1 c_i\ne-c_{i+1} c i = − c i + 1 we have ∣ w c i − w − c i + 1 ∣ ≥ d |w_{c_i}-w_{-c_{i+1}}|\ge d ∣ w c i − w − c i + 1 ∣ ≥ d . Thus, ∣ z i − w c i ∣ ≥ d − 2 r ≥ d / 2 |z_i-w_{c_i}|\ge d-2r\ge d/2 ∣ z i − w c i ∣ ≥ d − 2 r ≥ d /2 by (H). Hence, ∣ μ i ∣ ≥ d / ( 2 r c i ) ≥ d / ( 2 r ) = ϖ − 1 |\mu_i|\ge d/(2r_{c_i})\ge d/(2r)=\varpi^{-1} ∣ μ i ∣ ≥ d / ( 2 r c i ) ≥ d / ( 2 r ) = ϖ − 1 and
∣ z i − 1 − w − c i ∣ = r c i 2 ∣ z i − w c i ∣ ≤ 2 r c i 2 d ≤ r c i ϖ ≤ r c i 4 ( ≤ 2 r ) , |z_{i-1}-w_{-c_i}|=\frac{r_{c_i}^2}{|z_i-w_{c_i}|}\le\frac{2r_{c_i}^2}{d}\le r_{c_i}\varpi\le\frac{r_{c_i}}4\ (\le2r), ∣ z i − 1 − w − c i ∣ = ∣ z i − w c i ∣ r c i 2 ≤ d 2 r c i 2 ≤ r c i ϖ ≤ 4 r c i ( ≤ 2 r ) ,
which is the hypothesis for i − 1 i-1 i − 1 . In particular, ∣ z 0 − w − c 1 ∣ ≤ r c 1 / 4 |z_0-w_{-c_1}|\le r_{c_1}/4 ∣ z 0 − w − c 1 ∣ ≤ r c 1 /4 and ∣ μ 1 ∣ ≥ d / ( 2 r c 1 ) |\mu_1|\ge d/(2r_{c_1}) ∣ μ 1 ∣ ≥ d / ( 2 r c 1 ) .
(c) The pairing with y \mathbf{y} y . If y = ( y , 1 ) \mathbf{y}=(y,1) y = ( y , 1 ) with y ∈ K y\in K y ∈ K , then ⟨ ( z 0 , 1 ) , y ⟩ = z 0 − y \langle(z_0,1),\mathbf{y}\rangle=z_0-y ⟨( z 0 , 1 ) , y ⟩ = z 0 − y , and ∣ y − w − c 1 ∣ ≥ r c 1 / 2 |y-w_{-c_1}|\ge r_{c_1}/2 ∣ y − w − c 1 ∣ ≥ r c 1 /2 together with (b) gives ∣ z 0 − y ∣ ≥ r c 1 / 4 |z_0-y|\ge r_{c_1}/4 ∣ z 0 − y ∣ ≥ r c 1 /4 . If y = ( 1 , v ) \mathbf{y}=(1,v) y = ( 1 , v ) with ∣ v ∣ ≤ 1 / X 0 |v|\le1/X_0 ∣ v ∣ ≤ 1/ X 0 , then ⟨ ( z 0 , 1 ) , y ⟩ = z 0 v − 1 \langle(z_0,1),\mathbf{y}\rangle=z_0v-1 ⟨( z 0 , 1 ) , y ⟩ = z 0 v − 1 and ∣ z 0 ∣ ≤ W + r = X 0 / 2 |z_0|\le W+r=X_0/2 ∣ z 0 ∣ ≤ W + r = X 0 /2 , thus ∣ z 0 v − 1 ∣ ≥ 1 2 |z_0v-1|\ge\frac12 ∣ z 0 v − 1∣ ≥ 2 1 .
(d) Conclusion. By (3), ∣ ⟨ γ ~ x , y ⟩ ∣ = ∣ μ n ∣ ⋅ ∏ i = 2 n − 1 ∣ μ i ∣ ⋅ ∣ μ 1 ∣ ⋅ ∣ ⟨ ( z 0 , 1 ) , y ⟩ ∣ |\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle|=|\mu_n|\cdot\prod_{i=2}^{n-1}|\mu_i|\cdot|\mu_1|\cdot|\langle(z_0,1),\mathbf{y}\rangle| ∣ ⟨ γ x , y ⟩ ∣ = ∣ μ n ∣ ⋅ ∏ i = 2 n − 1 ∣ μ i ∣ ⋅ ∣ μ 1 ∣ ⋅ ∣ ⟨( z 0 , 1 ) , y ⟩ ∣ , where the middle product has n − 2 n-2 n − 2 factors, each greater or equal ϖ − 1 \varpi^{-1} ϖ − 1 . By (b),(c): ∣ μ 1 ∣ ∣ ⟨ ( z 0 , 1 ) , y ⟩ ∣ ≥ d 2 r c 1 ⋅ r c 1 4 = d 8 |\mu_1||\langle(z_0,1),\mathbf{y}\rangle|\ge\frac{d}{2r_{c_1}}\cdot\frac{r_{c_1}}4=\frac d8 ∣ μ 1 ∣∣ ⟨( z 0 , 1 ) , y ⟩ ∣ ≥ 2 r c 1 d ⋅ 4 r c 1 = 8 d if y = ( y , 1 ) \mathbf{y}=(y,1) y = ( y , 1 ) , and ≥ d 2 r ⋅ 1 2 = d 4 r \ge\frac d{2r}\cdot\frac12=\frac d{4r} ≥ 2 r d ⋅ 2 1 = 4 r d if y = ( 1 , v ) \mathbf{y}=(1,v) y = ( 1 , v ) . Together with (a) this gives, for the four combinations of charts,
∣ ⟨ γ ~ x , y ⟩ ∣ − 1 ≤ ϖ n − 2 × { 2 ⋅ 8 d = 16 d , x , y both in the first chart , 2 ⋅ 4 r d ≤ 1 3 , x first , y second , 2 r ⋅ 8 d ≤ 2 3 , x second , y first , 2 r ⋅ 4 r d ≤ r 3 , both second , |\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle|^{-1}\le\varpi^{n-2}\times
\begin{cases}
2\cdot\frac8d=\frac{16}{d},&\mathbf{x},\mathbf{y}\text{ both in the first chart},\\
2\cdot\frac{4r}{d}\le\frac13,&\mathbf{x}\text{ first},\ \mathbf{y}\text{ second},\\
2r\cdot\frac8d\le\frac23,&\mathbf{x}\text{ second},\ \mathbf{y}\text{ first},\\
2r\cdot\frac{4r}{d}\le\frac r3,&\text{both second},
\end{cases} ∣ ⟨ γ x , y ⟩ ∣ − 1 ≤ ϖ n − 2 × ⎩ ⎨ ⎧ 2 ⋅ d 8 = d 16 , 2 ⋅ d 4 r ≤ 3 1 , 2 r ⋅ d 8 ≤ 3 2 , 2 r ⋅ d 4 r ≤ 3 r , x , y both in the first chart , x first , y second , x second , y first , both second ,
where we used r / d ≤ 1 24 r/d\le\frac1{24} r / d ≤ 24 1 from (H). Each right-hand side is little or equal C ∗ ϖ n − 2 C_*\varpi^{n-2} C ∗ ϖ n − 2 . The four cases are collected in the following table, where L x L_x L x is the lower bound for ∣ μ n ∣ |\mu_n| ∣ μ n ∣ from (a), L y L_y L y the lower bound for ∣ μ 1 ∣ ∣ ⟨ ( z 0 , 1 ) , y ⟩ ∣ |\mu_1||\langle(z_0,1),\mathbf{y}\rangle| ∣ μ 1 ∣∣ ⟨( z 0 , 1 ) , y ⟩ ∣ from (b)-(c), and the bound for ∣ ⟨ γ ~ x , y ⟩ ∣ − 1 / ϖ n − 2 |\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle|^{-1}/\varpi^{n-2} ∣ ⟨ γ x , y ⟩ ∣ − 1 / ϖ n − 2 is 1 / ( L x L y ) 1/(L_xL_y) 1/ ( L x L y ) :
Table 1.
The second and fourth rows use ∣ μ 1 ∣ ≥ d / ( 2 r ) |\mu_1|\ge d/(2r) ∣ μ 1 ∣ ≥ d / ( 2 r ) and ∣ z 0 v − 1 ∣ ≥ 1 2 |z_0v-1|\ge\frac12 ∣ z 0 v − 1∣ ≥ 2 1 . The first and third use ∣ μ 1 ∣ ≥ d / ( 2 r c 1 ) |\mu_1|\ge d/(2r_{c_1}) ∣ μ 1 ∣ ≥ d / ( 2 r c 1 ) and ∣ z 0 − y ∣ ≥ r c 1 / 4 |z_0-y|\ge r_{c_1}/4 ∣ z 0 − y ∣ ≥ r c 1 /4 . □
Remark 3.4. In this Remark we discuss constants and conventions in Lemmas 3.2-3.3. The bounds ∣ γ ~ x ∣ ≤ 2 ϖ n − 1 |\widetilde{\gamma}\mathbf{x}|\le2\varpi^{n-1} ∣ γ x ∣ ≤ 2 ϖ n − 1 and ∣ ⟨ γ ~ x , y ⟩ ∣ − 1 ≤ C ∗ ϖ n − 2 |\langle\widetilde{\gamma}\mathbf{x},\mathbf{y}\rangle|^{-1}\le C_*\varpi^{n-2} ∣ ⟨ γ x , y ⟩ ∣ − 1 ≤ C ∗ ϖ n − 2 hold for every reduced word of length n n n with constants depending only on the Schottky data through d d d , r r r and ϖ \varpi ϖ (C ∗ = 16 / d + 1 + r C_*=16/d+1+r C ∗ = 16/ d + 1 + r ). They do not depend on the compact subset or on the word, and no constant has been enlarged to absorb short words. The identity element (n = 0 n=0 n = 0 ) is excluded from the reduced words of the lemmas and is treated separately. It contributes the term 1 / ⟨ x , y ⟩ = 1 / ( x − y ) 1/\langle\mathbf{x},\mathbf{y}\rangle=1/(x-y) 1/ ⟨ x , y ⟩ = 1/ ( x − y ) (which carries the diagonal pole), and the words of length 1 1 1 are likewise summed separately (see Lemma 3.2 with n = 1 n=1 n = 1 ). In Theorem 3.5 the tail ∑ n ≥ 2 \sum_{n\ge2} ∑ n ≥ 2 is thus dominated by C ∗ ∑ n ≥ 2 2 g ( 2 g − 1 ) n − 1 ϖ n − 2 ≤ C ∗ 2 g ( 2 g − 1 ) ∑ k ≥ 0 q k C_*\sum_{n\ge2}2g(2g-1)^{n-1}\varpi^{n-2}\le C_*2g(2g-1)\sum_{k\ge0}q^k C ∗ ∑ n ≥ 2 2 g ( 2 g − 1 ) n − 1 ϖ n − 2 ≤ C ∗ 2 g ( 2 g − 1 ) ∑ k ≥ 0 q k , q = ( 2 g − 1 ) ϖ ≤ 1 4 q=(2g-1)\varpi\le\frac14 q = ( 2 g − 1 ) ϖ ≤ 4 1 . The only dependence on the compact set K ^ p \widehat{K}_p K p is through the finitely many terms n ≤ 1 n\le1 n ≤ 1 and the distance to the polar loci. The same convention is used in Theorem 5.1, whose remainder constant 2 g ( 2 g − 1 ) ∣ ρ ∣ max / d 3 2g(2g-1)|\rho|_{\max}/d^3 2 g ( 2 g − 1 ) ∣ ρ ∣ m a x / d 3 is uniform on G d G_d G d . The compact set enters only through K d , p K_{d,p} K d , p , i.e., distance ≥ d / 2 \ge d/2 ≥ d /2 from all w c w_c w c . It is not claimed it is uniform up to the pole loci x ∈ Γ y x \in\Gamma y x ∈ Γ y (the expansion is stated on K d , p × K d , p K_{d,p} \times K_{d,p} K d , p × K d , p only).
The theorem below states the convergence and meromorphicity on Ω 0 × Ω 0 \Omega_0 \times\Omega_0 Ω 0 × Ω 0 , including ∞ \infty ∞ . Let us assume (H). Let c γ ~ c_{\widetilde{\gamma}} c γ denote the lower-left entry of γ ~ ∈ Γ ~ \widetilde{\gamma} \in\widetilde{\Gamma} γ ∈ Γ (thus c γ ~ ≠ 0 c_{\widetilde{\gamma}} \ne0 c γ = 0 for γ ~ ≠ 1 \widetilde{\gamma} \ne1 γ = 1 , since Γ \Gamma Γ acts freely and γ ∞ ≠ ∞ \gamma\infty\ne\infty γ ∞ = ∞ ).
Theorem 3.5. (i) Uniform tail on the total space. Let U ⊂ P U \subset\mathfrak{P} U ⊂ P be compact. For every p ∈ U p \in U p ∈ U and all α , β ∈ K ^ p \alpha,\beta\in\widehat{K}_p α , β ∈ K p with normalized representatives x , y x,y x , y ,
∑ n ≥ 2 ∑ ∣ γ ~ ∣ = n ∣ ⟨ γ ~ x , y ⟩ ∣ − 1 ≤ 8 3 g ( 2 g − 1 ) sup U C ∗ < ∞ . \sum_{n\ge2}\sum_{\lvert\widetilde{\gamma}\rvert=n}\lvert\langle\widetilde{\gamma}x,y\rangle\rvert^{-1}\le\frac{8}{3}g(2g-1)\sup_U C_*<\infty. n ≥ 2 ∑ ∣ γ ∣ = n ∑ ∣⟨ γ x , y ⟩ ∣ − 1 ≤ 3 8 g ( 2 g − 1 ) U sup C ∗ < ∞. Let U U U denote the interior of the total space { ( p , α , β ) : p ∈ P , α , β ∈ K ^ p } \{(p,\alpha,\beta):p\in\mathfrak{P},\ \alpha,\beta\in\widehat{K}_p\} {( p , α , β ) : p ∈ P , α , β ∈ K p } , in the coordinates ( x , y ) (x,y) ( x , y ) , ( u , y ) (u,y) ( u , y ) , ( x , v ) (x,v) ( x , v ) or ( u , v ) (u,v) ( u , v ) , u = 1 / x u=1/x u = 1/ x , v = 1 / y v=1/y v = 1/ y , according to the charts. Let P 1 : = { β = α } \mathcal{P}_1:=\{\beta=\alpha\} P 1 := { β = α } , P c : = { β = γ c α } \mathcal{P}_c:=\{\beta=\gamma_c\alpha\} P c := { β = γ c α } , c ∈ I c\in I c ∈ I , be the polar graphs in U U U . Then:
(a) the tail T : = ∑ ∣ γ ~ ∣ ≥ 2 ⟨ γ ~ x , y ⟩ − 1 T:=\sum_{\lvert\widetilde{\gamma}\rvert\ge2}\langle\widetilde{\gamma}x,y\rangle^{-1} T := ∑ ∣ γ ∣ ≥ 2 ⟨ γ x , y ⟩ − 1 converges absolutely and uniformly on U U U (on the part over U U U ) and is holomorphic on all of U U U , including across the polar graphs (Weierstrass);
(b) the terms γ ~ = 1 \widetilde{\gamma}=1 γ = 1 and γ ~ = γ ~ c \widetilde{\gamma}=\widetilde{\gamma}_c γ = γ c (c ∈ I c\in I c ∈ I ) are explicit rational functions of ( p , x , y ) (p,x,y) ( p , x , y ) , holomorphic on U ∖ U P U\setminus U^{\mathcal{P}} U ∖ U P with a simple pole along their respective graph;
(c) consequently, s ^ ε = ⟨ x , y ⟩ − 1 + ∑ c ⟨ γ ~ c x , y ⟩ − 1 + T \widehat{s}_{\varepsilon}=\langle x,y\rangle^{-1}+\sum_c\langle\widetilde{\gamma}_c x,y\rangle^{-1}+T s ε = ⟨ x , y ⟩ − 1 + ∑ c ⟨ γ c x , y ⟩ − 1 + T is meromorphic on U U U with simple poles along the graphs P 1 \mathcal{P}_1 P 1 , P c \mathcal{P}_c P c , and holomorphic on U ∖ U P U\setminus U^{\mathcal{P}} U ∖ U P .
(ii) Global convergence, invariance, meromorphicity. Let us assume (H) and g ≥ 2 g\ge2 g ≥ 2 . For x , y ∈ C 2 ∖ 0 x,y\in\mathbb{C}^2\setminus0 x , y ∈ C 2 ∖ 0 over Ω 0 \Omega_0 Ω 0 with [ y ] ∉ Γ [ x ] [y]\notin\Gamma[x] [ y ] ∈ / Γ [ x ] the series (2) converges absolutely, and
s ^ ε ( λ ~ x , μ ~ y ) = s ^ ε ( x , y ) , λ ~ , μ ~ ∈ Γ ~ . (4) \widehat{s}_{\varepsilon}(\widetilde{\lambda}x,\widetilde{\mu}y)=\widehat{s}_{\varepsilon}(x,y),\qquad\widetilde{\lambda},\widetilde{\mu}\in\widetilde{\Gamma}. \tag*{(4)} s ε ( λ x , μ y ) = s ε ( x , y ) , λ , μ ∈ Γ . (4) The function s ^ ε \widehat{s}_{\varepsilon} s ε is meromorphic on the cone over Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 (the points ∞ \infty ∞ and Γ ∞ \Gamma\infty Γ∞ included), with polar set { [ y ] ∈ Γ [ x ] } \{[y]\in\Gamma[x]\} {[ y ] ∈ Γ [ x ]} , all poles simple. Equivalently, for fixed y y y away from the orbit of the pole, x ↦ s ε ( x , y ) x\mapsto s_{\varepsilon}(x,y) x ↦ s ε ( x , y ) is meromorphic on Ω 0 \Omega_0 Ω 0 with simple poles at the points of Γ y \Gamma y Γ y . These project to the single point [ y ] ∈ Σ [y]\in\Sigma [ y ] ∈ Σ , i.e., to the diagonal of Σ × Σ \Sigma\times\Sigma Σ × Σ . The only singularities in any chart are these poles: the point x = ∞ x=\infty x = ∞ is not singular for y ∉ Γ ∞ y\notin\Gamma\infty y ∈ / Γ∞ , and there are no coordinate-dependent apparent singularities. For finite points x , y x,y x , y with λ x , μ y ≠ ∞ \lambda x,\mu y\ne\infty λ x , μ y = ∞ the identity (4) reads
s ε ( λ x , μ y ) j ( λ ~ , x ) j ( μ ~ , y ) = s ε ( x , y ) . (5) s_{\varepsilon}(\lambda x,\mu y)j(\widetilde{\lambda},x)j(\widetilde{\mu},y)=s_{\varepsilon}(x,y). \tag*{(5)} s ε ( λ x , μ y ) j ( λ , x ) j ( μ , y ) = s ε ( x , y ) . (5) (iii) Behaviour at ∞ \infty ∞ . Set s ∞ ( y ) : = s ^ ε ( ( 1 , 0 ) , ( y , 1 ) ) = 1 + ∑ γ ~ ≠ 1 ( c γ ~ ( γ ∞ − y ) ) − 1 s^\infty(y):=\widehat{s}_{\varepsilon}((1,0),(y,1))=1+\sum_{\widetilde{\gamma}\ne1}(c_{\widetilde{\gamma}}(\gamma\infty-y))^{-1} s ∞ ( y ) := s ε (( 1 , 0 ) , ( y , 1 )) = 1 + ∑ γ = 1 ( c γ ( γ ∞ − y ) ) − 1 . It is meromorphic in y ∈ Ω 0 ∖ { ∞ } y\in\Omega_0\setminus\{\infty\} y ∈ Ω 0 ∖ { ∞ } , its poles are the points γ ∞ \gamma\infty γ ∞ , γ ≠ 1 \gamma\ne1 γ = 1 , all simple, with residue − 1 / c γ ~ -1/c_{\widetilde{\gamma}} − 1/ c γ at γ ∞ \gamma\infty γ ∞ . For y ∈ Ω 0 ∖ Γ ∞ y\in\Omega_0\setminus\Gamma\infty y ∈ Ω 0 ∖ Γ∞ ,
s ε ( x , y ) = s ∞ ( y ) x + O ( ∣ x ∣ − 2 ) , ∣ x ∣ → ∞ , s_{\varepsilon}(x,y)=\frac{s^\infty(y)}{x}+O(\lvert x\rvert^{-2}),\qquad\lvert x\rvert\to\infty, s ε ( x , y ) = x s ∞ ( y ) + O (∣ x ∣ − 2 ) , ∣ x ∣ → ∞ , locally uniformly in y ∈ Ω 0 ∖ Γ ∞ y\in\Omega_0\setminus\Gamma\infty y ∈ Ω 0 ∖ Γ∞ . In particular s ε ( x , y ) = O ( 1 / ∣ x ∣ ) s_{\varepsilon}(x,y)=O(1/\lvert x\rvert) s ε ( x , y ) = O ( 1/ ∣ x ∣) uniformly for y y y in compact subsets of Ω 0 ∖ Γ ∞ \Omega_0\setminus\Gamma\infty Ω 0 ∖ Γ∞ , and s ε ( ⋅ , y ) s_{\varepsilon}(\mathord{\cdot},y) s ε ( ⋅ , y ) is regular at x = ∞ x=\infty x = ∞ in the coordinate u = 1 / x u=1/x u = 1/ x for such y y y .
Note that in (i) no holomorphy across the polar graphs is claimed. For fixed p p p the statement is meromorphicity on the interior of K ^ p × K ^ p \widehat{K}_p\times\widehat{K}_p K p × K p , depending holomorphically on p p p , locally uniformly on P \mathfrak{P} P . For y ∈ Γ ∞ y\in\Gamma\infty y ∈ Γ∞ , (iii) fails. For γ ∞ ≠ ∞ \gamma\infty\ne\infty γ ∞ = ∞ , γ ≠ 1 \gamma\ne1 γ = 1 , then lim ∣ x ∣ → ∞ s ε ( x , y ) = − c γ ≠ 0 \lim_{\lvert x\rvert\to\infty}s_{\varepsilon}(x,y)=-c_{\gamma}\ne0 lim ∣ x ∣ → ∞ s ε ( x , y ) = − c γ = 0 . For instance, for y = w − c = γ c ( ∞ ) y=w_{-c}=\gamma_c(\infty) y = w − c = γ c ( ∞ ) the length-one summand equals j ( γ ~ c , x ) / ( γ c x − w − c ) = λ c / ρ c = − 1 / λ c j(\widetilde{\gamma}_c,x)/(\gamma_c x-w_{-c})=\lambda_c/\rho_c=-1/\lambda_c j ( γ c , x ) / ( γ c x − w − c ) = λ c / ρ c = − 1/ λ c , independent of x x x . Neither is the bound O ( 1 / ∣ x ∣ ) O(1/\lvert x\rvert) O ( 1/ ∣ x ∣) uniform as y → γ ∞ y\to\gamma\infty y → γ ∞ , since x s ε ( x , y ) → s ∞ ( y ) x\,s_{\varepsilon}(x,y)\to s^\infty(y) x s ε ( x , y ) → s ∞ ( y ) and ∣ s ∞ ( y ) ∣ → ∞ \lvert s^\infty(y)\rvert\to\infty ∣ s ∞ ( y )∣ → ∞ . The pairs ( ∞ , γ ∞ ) (\infty,\gamma\infty) ( ∞ , γ ∞ ) are points of the polar set { y ∈ Γ x } \{y\in\Gamma x\} { y ∈ Γ x } , i.e., lie over the diagonal of Σ × Σ \Sigma\times\Sigma Σ × Σ .
Proof. (i) Lemma 3.3 bounds each of the 2 g ( 2 g − 1 ) n − 1 2g(2g-1)^{n-1} 2 g ( 2 g − 1 ) n − 1 words of length n ≥ 2 n\ge2 n ≥ 2 by C ∗ ϖ n − 2 C_*\varpi^{n-2} C ∗ ϖ n − 2 . Since q : = ( 2 g − 1 ) ϖ ≤ 1 / 4 q:=(2g-1)\varpi\le1/4 q := ( 2 g − 1 ) ϖ ≤ 1/4 the sum is at most 2 g ( 2 g − 1 ) C ∗ ∑ m ≥ 0 q m ≤ 8 3 g ( 2 g − 1 ) C ∗ 2g(2g-1)C_*\sum_{m\ge0}q^m\le\frac{8}{3}g(2g-1)C_* 2 g ( 2 g − 1 ) C ∗ ∑ m ≥ 0 q m ≤ 3 8 g ( 2 g − 1 ) C ∗ , and C ∗ , ϖ C_*,\varpi C ∗ , ϖ are continuous on the compact set U U U . Every term is holomorphic in ( p , x , y ) (p,x,y) ( p , x , y ) (the entries of γ ~ \widetilde{\gamma} γ are polynomial in w c w_c w c , λ c ± 1 \lambda_c^{\pm1} λ c ± 1 ) and, by Lemma 3.3, nowhere zero in the denominator on K ^ p × K ^ p \widehat{K}_p\times\widehat{K}_p K p × K p . The Weierstrass theorem gives holomorphy of the sum. The statements on the terms n = 0 , 1 n=0,1 n = 0 , 1 and on the polar set are immediate: ⟨ x , y ⟩ = 0 \langle x,y\rangle=0 ⟨ x , y ⟩ = 0 iff β = α \beta=\alpha β = α , and ⟨ γ ~ c x , y ⟩ = 0 \langle\widetilde{\gamma}_c x,y\rangle=0 ⟨ γ c x , y ⟩ = 0 iff β = γ c α \beta=\gamma_c\alpha β = γ c α .
(ii) Uniformity from (i) is used only on the compact set K ^ p × K ^ p \widehat{K}_p\times\widehat{K}_p K p × K p for fixed p p p (or p ∈ U p\in U p ∈ U compact). The reindexing below is applied only to series that converge absolutely there, and no limiting operation is exchanged with the summation otherwise. Let α , β ∈ Ω 0 \alpha,\beta\in\Omega_0 α , β ∈ Ω 0 , β ∉ Γ α \beta\notin\Gamma\alpha β ∈ / Γ α . By the standard facts of § 2 (Ω 0 = Γ D \Omega_0=\Gamma D Ω 0 = Γ D ) write α = λ α 0 \alpha=\lambda\alpha_0 α = λ α 0 , β = μ β 0 \beta=\mu\beta_0 β = μ β 0 with α 0 , β 0 ∈ D ⊂ int K ^ \alpha_0,\beta_0\in D\subset\operatorname{int}\widehat{K} α 0 , β 0 ∈ D ⊂ int K and λ , μ ∈ Γ \lambda,\mu\in\Gamma λ , μ ∈ Γ . Let λ ~ , μ ~ ∈ Γ ~ \widetilde{\lambda},\widetilde{\mu}\in\widetilde{\Gamma} λ , μ ∈ Γ be the lifts and x 0 , y 0 x_0,y_0 x 0 , y 0 normalized representatives of α 0 , β 0 \alpha_0,\beta_0 α 0 , β 0 . Then β 0 ∉ Γ α 0 \beta_0\notin\Gamma\alpha_0 β 0 ∈ / Γ α 0 (otherwise β = μ β 0 ∈ μ Γ α 0 = Γ α \beta=\mu\beta_0\in\mu\Gamma\alpha_0=\Gamma\alpha β = μ β 0 ∈ μ Γ α 0 = Γ α ). For every γ ~ ∈ Γ ~ \widetilde{\gamma}\in\widetilde{\Gamma} γ ∈ Γ , by S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) -invariance of ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ ,
⟨ γ ~ λ ~ x 0 , μ ~ y 0 ⟩ = ⟨ μ ~ − 1 γ ~ λ ~ x 0 , y 0 ⟩ \langle\widetilde{\gamma}\widetilde{\lambda}x_0,\widetilde{\mu}y_0\rangle=\langle\widetilde{\mu}^{-1}\widetilde{\gamma}\widetilde{\lambda}x_0,y_0\rangle ⟨ γ λ x 0 , μ y 0 ⟩ = ⟨ μ − 1 γ λ x 0 , y 0 ⟩
and γ ~ ↦ κ ~ : = μ ~ − 1 γ ~ λ ~ \widetilde{\gamma}\mapsto\widetilde{\kappa}:=\widetilde{\mu}^{-1}\widetilde{\gamma}\widetilde{\lambda} γ ↦ κ := μ − 1 γ λ is a bijection of Γ ~ \widetilde{\Gamma} Γ onto itself (left and right translations). Thus, the series for s ^ ε ( λ ~ x 0 , μ ~ y 0 ) \widehat{s}_{\varepsilon}(\widetilde{\lambda}x_0,\widetilde{\mu}y_0) s ε ( λ x 0 , μ y 0 ) is a rearrangement of the series for s ^ ε ( x 0 , y 0 ) \widehat{s}_{\varepsilon}(x_0,y_0) s ε ( x 0 , y 0 ) , which converges absolutely by (i) (together with the finitely many terms n ≤ 1 n\le1 n ≤ 1 , all finite as β 0 ∉ Γ α 0 \beta_0\notin\Gamma\alpha_0 β 0 ∈ / Γ α 0 ). Hence, we have absolute convergence and s ^ ε ( λ ~ x 0 , μ ~ y 0 ) = s ^ ε ( x 0 , y 0 ) \widehat{s}_{\varepsilon}(\widetilde{\lambda}x_0,\widetilde{\mu}y_0)=\widehat{s}_{\varepsilon}(x_0,y_0) s ε ( λ x 0 , μ y 0 ) = s ε ( x 0 , y 0 ) . As every pair of vectors over Ω 0 \Omega_0 Ω 0 is ( t λ ~ 1 x 0 , t ′ μ ~ 1 y 0 ) (t\widetilde{\lambda}_1x_0,t'\widetilde{\mu}_1y_0) ( t λ 1 x 0 , t ′ μ 1 y 0 ) and s ^ ε \widehat{s}_{\varepsilon} s ε is bihomogeneous of degree ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) , this proves (4) (i.e., replace λ ~ 1 , μ ~ 1 \widetilde{\lambda}_1,\widetilde{\mu}_1 λ 1 , μ 1 by λ ~ λ ~ 1 , μ ~ μ ~ 1 \widetilde{\lambda}\widetilde{\lambda}_1,\widetilde{\mu}\widetilde{\mu}_1 λ λ 1 , μ μ 1 ).
Meromorphicity. Near ( α , β ) (\alpha,\beta) ( α , β ) write ( x , y ) = ( λ ~ x ′ , μ ~ y ′ ) (\mathbf{x},\mathbf{y})=(\widetilde{\lambda}\mathbf{x}',\widetilde{\mu}\mathbf{y}') ( x , y ) = ( λ x ′ , μ y ′ ) with ( x ′ , y ′ ) (\mathbf{x}',\mathbf{y}') ( x ′ , y ′ ) in a neighbourhood of ( x 0 , y 0 ) (x_0,y_0) ( x 0 , y 0 ) contained in int K ^ × int K ^ \operatorname{int}\widehat{K}\times\operatorname{int}\widehat{K} int K × int K . By (4) and (i), s ^ ε ( x , y ) = s ^ ε ( x ′ , y ′ ) \widehat{s}_{\varepsilon}(\mathbf{x},\mathbf{y})=\widehat{s}_{\varepsilon}(\mathbf{x}',\mathbf{y}') s ε ( x , y ) = s ε ( x ′ , y ′ ) is meromorphic there.
Poles. The fixed points of the (loxodromic) elements γ ≠ 1 \gamma\ne1 γ = 1 belong to the limit set, thus inside Ω 0 \Omega_0 Ω 0 the graphs { y = γ x } \{y=\gamma x\} { y = γ x } , γ ∈ Γ \gamma\in\Gamma γ ∈ Γ , are pairwise disjoint. By local finiteness of the family of graphs (i.e., proper discontinuity, § 2 : only finitely many γ \gamma γ have γ K ∩ K ′ ≠ ∅ \gamma K\cap K'\ne\varnothing γ K ∩ K ′ = ∅ for given compacta K , K ′ ⊂ Ω 0 K,K'\subset\Omega_0 K , K ′ ⊂ Ω 0 ), any point of Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 has a neighbourhood meeting at most one of them. At a point of the graph { y = γ x } \{y=\gamma x\} { y = γ x } one term of the series (that of γ ~ \widetilde{\gamma} γ ) is singular and has a simple pole. The remaining terms sum to a function holomorphic there (by the same rearrangement and (i)). Thus, there is no cancellation and the pole is simple. For finite points, the relations s ^ ε ( ( x , 1 ) , ( y , 1 ) ) = s ε ( x , y ) \widehat{s}_{\varepsilon}((x,1),(y,1))=s_{\varepsilon}(x,y) s ε (( x , 1 ) , ( y , 1 )) = s ε ( x , y ) and g ~ ( x , 1 ) = j ( g ~ , x ) − 1 ( g x , 1 ) \widetilde{g}(x,1)=j(\widetilde{g},x)^{-1}(gx,1) g ( x , 1 ) = j ( g , x ) − 1 ( g x , 1 ) turn (4) into (5) by bihomogeneity.
(iii) For γ ~ ≠ 1 \widetilde{\gamma}\ne1 γ = 1 one has ⟨ γ ~ ( 1 , 0 ) , ( y , 1 ) ⟩ = a γ ~ − c γ ~ y = c γ ~ ( γ ∞ − y ) \langle\widetilde{\gamma}(1,0),(y,1)\rangle=a_{\widetilde{\gamma}}-c_{\widetilde{\gamma}}y=c_{\widetilde{\gamma}}(\gamma\infty-y) ⟨ γ ( 1 , 0 ) , ( y , 1 )⟩ = a γ − c γ y = c γ ( γ ∞ − y ) , which gives the series for s ∞ s^\infty s ∞ . By (ii) (i.e., the pair ( [ ( 1 , 0 ) ] , [ ( y , 1 ) ] ) ([(1,0)],[(y,1)]) ([( 1 , 0 )] , [( y , 1 )]) is polar iff y ∈ Γ ∞ y\in\Gamma\infty y ∈ Γ∞ ) it is meromorphic and has simple poles with residue − 1 / c γ ~ -1/c_{\widetilde{\gamma}} − 1/ c γ at γ ∞ \gamma\infty γ ∞ . Let y 0 ∈ Ω 0 ∖ Γ ∞ y_0\in\Omega_0\setminus\Gamma\infty y 0 ∈ Ω 0 ∖ Γ∞ . The point ( [ ( 1 , 0 ) ] , [ ( y 0 , 1 ) ] ) ([(1,0)],[(y_0,1)]) ([( 1 , 0 )] , [( y 0 , 1 )]) is not polar. Thus, by (ii) the function ( u , y ) ↦ s ^ ε ( ( 1 , u ) , ( y , 1 ) ) (u,y)\mapsto\widehat{s}_{\varepsilon}((1,u),(y,1)) ( u , y ) ↦ s ε (( 1 , u ) , ( y , 1 )) is holomorphic near ( 0 , y 0 ) (0,y_0) ( 0 , y 0 ) . As ( x , 1 ) = x ( 1 , u ) (x,1)=x(1,u) ( x , 1 ) = x ( 1 , u ) with u = 1 / x u=1/x u = 1/ x , bihomogeneity gives s ε ( x , y ) = u s ^ ε ( ( 1 , u ) , ( y , 1 ) ) = u ( s ∞ ( y ) + O ( u ) ) s_{\varepsilon}(x,y)=u\widehat{s}_{\varepsilon}((1,u),(y,1))=u(s^\infty(y)+O(u)) s ε ( x , y ) = u s ε (( 1 , u ) , ( y , 1 )) = u ( s ∞ ( y ) + O ( u )) , locally uniformly in y y y . Now let y 0 = γ ∞ y_0=\gamma\infty y 0 = γ ∞ , γ ~ ≠ 1 \widetilde{\gamma}\ne1 γ = 1 . Thus, y 0 = a / c y_0=a/c y 0 = a / c . Near ( 0 , y 0 ) (0,y_0) ( 0 , y 0 ) the only singular summand of s ^ ε ( ( 1 , u ) , ( y , 1 ) ) \widehat{s}_{\varepsilon}((1,u),(y,1)) s ε (( 1 , u ) , ( y , 1 )) is the one of γ ~ \widetilde{\gamma} γ . The others sum to a function h ( u , y ) h(u,y) h ( u , y ) holomorphic there. Indeed, by the rearrangement of (ii) with λ = 1 \lambda=1 λ = 1 , μ = γ \mu=\gamma μ = γ , in the variables x = ( 1 , u ) \mathbf{x}=(1,u) x = ( 1 , u ) and y ′ = γ ~ − 1 ( y , 1 ) \mathbf{y}'=\widetilde{\gamma}^{-1}(y,1) y ′ = γ − 1 ( y , 1 ) , which lie near ( ( 1 , 0 ) , ( 1 , 0 ) ) ∈ int K ^ × int K ^ ((1,0),(1,0))\in\operatorname{int}\widehat{K}\times\operatorname{int}\widehat{K} (( 1 , 0 ) , ( 1 , 0 )) ∈ int K × int K , the summand of γ ~ \widetilde{\gamma} γ becomes the term 1 / ⟨ x , y ′ ⟩ 1/\langle\mathbf{x},\mathbf{y}'\rangle 1/ ⟨ x , y ′ ⟩ of length 0 0 0 . The terms of length 1 1 1 are regular there (since γ c ∞ = w − c ≠ ∞ \gamma_c\infty=w_{-c}\ne\infty γ c ∞ = w − c = ∞ ), and the tail is holomorphic by (i). A computation, using a − c y 0 = 0 a-cy_0=0 a − c y 0 = 0 and a d − b c = 1 ad-bc=1 a d − b c = 1 , gives ⟨ γ ~ ( 1 , u ) , ( y 0 , 1 ) ⟩ = ( a − c y 0 ) + ( b − d y 0 ) u = − u / c \langle\widetilde{\gamma}(1,u),(y_0,1)\rangle=(a-cy_0)+(b-dy_0)u=-u/c ⟨ γ ( 1 , u ) , ( y 0 , 1 )⟩ = ( a − c y 0 ) + ( b − d y 0 ) u = − u / c , so that s ε ( x , y 0 ) = − c / u + h ( u ) = − c x + u → − ∞ s_{\varepsilon}(x,y_0)=-c/u+h(u)=-cx+u\to-\infty s ε ( x , y 0 ) = − c / u + h ( u ) = − c x + u → − ∞ as u → 0 u\to0 u → 0 . Finally, x s ε ( x , y ) → s ∞ ( y ) x\,s_{\varepsilon}(x,y)\to s^\infty(y) x s ε ( x , y ) → s ∞ ( y ) as ∣ x ∣ → ∞ |x|\to\infty ∣ x ∣ → ∞ for y ∉ Γ ∞ y\notin\Gamma\infty y ∈ / Γ∞ , and s ∞ s^\infty s ∞ has a pole at ∞ \infty ∞ .
Next, we deal with equivariance and global descent.
Proposition 3.6. For λ , μ ∈ Γ ~ \lambda,\mu\in\widetilde{\Gamma} λ , μ ∈ Γ and finite x , y ∈ Ω 0 x,y\in\Omega_0 x , y ∈ Ω 0 with y ∉ Γ x y\notin\Gamma x y ∈ / Γ x and λ x , μ y ≠ ∞ \lambda x,\mu y\ne\infty λ x , μ y = ∞ ,
s ε ( λ x , y ) j ( λ , x ) = s ε ( x , y ) , s ε ( x , μ y ) j ( μ , y ) = s ε ( x , y ) . s_{\varepsilon}(\lambda x,y)j(\lambda,x)=s_{\varepsilon}(x,y),\qquad s_{\varepsilon}(x,\mu y)j(\mu,y)=s_{\varepsilon}(x,y). s ε ( λ x , y ) j ( λ , x ) = s ε ( x , y ) , s ε ( x , μ y ) j ( μ , y ) = s ε ( x , y ) . More generally, in homogeneous form, s ^ ε ( λ ~ x , y ) = s ^ ε ( x , y ) = s ^ ε ( x , μ ~ y ) \widehat{s}_{\varepsilon}(\widetilde{\lambda}\mathbf{x},\mathbf{y})=\widehat{s}_{\varepsilon}(\mathbf{x},\mathbf{y})=\widehat{s}_{\varepsilon}(\mathbf{x},\widetilde{\mu}\mathbf{y}) s ε ( λ x , y ) = s ε ( x , y ) = s ε ( x , μ y ) at all points of Ω 0 \Omega_0 Ω 0 , including ∞ \infty ∞ and Γ ∞ \Gamma\infty Γ∞ , and there is no additive defect χ [ γ ] ( x , y ) \chi[\gamma](x,y) χ [ γ ] ( x , y ) . Consequently, S ε S_{\varepsilon} S ε descends to a meromorphic section of L ε ⊠ L ε L_{\varepsilon}\boxtimes L_{\varepsilon} L ε ⊠ L ε on Σ × Σ \Sigma\times\Sigma Σ × Σ , holomorphic off the diagonal Δ \Delta Δ , with a simple pole along Δ \Delta Δ and no other singularity.
Proof. The invariance is (4) with μ ~ = 1 \widetilde{\mu}=1 μ = 1 or λ ~ = 1 \widetilde{\lambda}=1 λ = 1 , and the finite-point forms are (5) .
Descent: as explained at the beginning of this section, sections of L ε ⊠ L ε L_{\varepsilon}\boxtimes L_{\varepsilon} L ε ⊠ L ε over an open subset of Σ × Σ \Sigma\times\Sigma Σ × Σ are Γ ~ × Γ ~ \widetilde{\Gamma}\times\widetilde{\Gamma} Γ × Γ -invariant functions on the cover over its preimage in Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 that are bihomogeneous of degree ( − 1 , − 1 ) (-1,-1) ( − 1 , − 1 ) . Hence, s ^ ε \widehat{s}_{\varepsilon} s ε is a section of L ε ⊠ L ε L_{\varepsilon}\boxtimes L_{\varepsilon} L ε ⊠ L ε over the complement of the image of its polar set. That image is the diagonal Δ ⊂ Σ × Σ \Delta\subset\Sigma\times\Sigma Δ ⊂ Σ × Σ : y ∈ Γ x y\in\Gamma x y ∈ Γ x means that x , y x,y x , y have the same class in Σ \Sigma Σ . Ω 0 × Ω 0 → Σ × Σ \Omega_0\times\Omega_0\to\Sigma\times\Sigma Ω 0 × Ω 0 → Σ × Σ is a local biholomorphism since Γ \Gamma Γ acts freely and properly discontinuously. Meromorphicity across Δ \Delta Δ and simplicity of the pole were proven in Theorem 3.5 (ii) at every point of Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 , in particular, at the points lying over x = ∞ x=\infty x = ∞ or y = ∞ y=\infty y = ∞ , where the scalar trivialization by ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 is not available and the homogeneous form is used. Note that we do not claim regularity of s ε s_{\varepsilon} s ε at x = ∞ x=\infty x = ∞ . The pairs ( ∞ , γ ∞ ) (\infty,\gamma\infty) ( ∞ , γ ∞ ) are ordinary points of the polar divisor, and Theorem 3.5 (iii) describes how s ε s_{\varepsilon} s ε behaves there.
Here below we discuss the skew-symmetry, local form, and projective connection.
Proposition 3.7.
(i) s ^ ε ( y , x ) = − s ^ ε ( x , y ) \widehat{s}_{\varepsilon}(y,x)=-\widehat{s}_{\varepsilon}(x,y) s ε ( y , x ) = − s ε ( x , y ) for all x x x , y y y over Ω 0 \Omega_0 Ω 0 with [ y ] ∉ Γ [ x ] [y]\notin\Gamma[x] [ y ] ∈ / Γ [ x ] . In particular, s ε ( y , x ) = − s ε ( x , y ) s_{\varepsilon}(y,x)=-s_{\varepsilon}(x,y) s ε ( y , x ) = − s ε ( x , y ) for finite points.
(ii) Canonical form: with π ( x , y ) = ( d x d y ) 1 / 2 x − y \pi(x,y)=\frac{(dx\,dy)^{1/2}}{x-y} π ( x , y ) = x − y ( d x d y ) 1/2 in any local coordinate near Δ \Delta Δ , S ε − π S_{\varepsilon}-\pi S ε − π is a holomorphic section of ( L ε ⊠ L ε ) ( Δ ) (L_{\varepsilon}\boxtimes L_{\varepsilon})(\Delta) ( L ε ⊠ L ε ) ( Δ ) near Δ \Delta Δ , and its restriction to Δ \Delta Δ is the global constant 0 ∈ Γ ( Σ , O Σ ) = C 0\in\Gamma(\Sigma,\mathcal{O}_{\Sigma})=\mathbb{C} 0 ∈ Γ ( Σ , O Σ ) = C . Coordinate form: in any local coordinate this reads s ε ( x , y ) = 1 x − y + c 1 ( y ) ( x − y ) + O ( ( x − y ) 2 ) s_{\varepsilon}(x,y)=\frac{1}{x-y}+c_1(y)(x-y)+O((x-y)^2) s ε ( x , y ) = x − y 1 + c 1 ( y ) ( x − y ) + O (( x − y ) 2 ) . The coefficient c 0 ( y ) c_0(y) c 0 ( y ) of ( x − y ) 0 (x-y)^0 ( x − y ) 0 which represents the intrinsic restriction above in that coordinate vanishes identically.
(iii) Let z z z and w w w be two local coordinates, w = w ( z ) w=w(z) w = w ( z ) , w ′ = d w / d z w'=dw/dz w ′ = d w / d z , and let c 1 ( z ) c_1^{(z)} c 1 ( z ) , c 1 ( w ) c_1^{(w)} c 1 ( w ) denote the coefficient c 1 c_1 c 1 of part (ii) computed in the coordinates z z z and w w w respectively. Then c 1 ( z ) ( y ) = w ′ ( y ) 2 c 1 ( w ) ( w ( y ) ) + 1 12 { w ; z } ( y ) c_1^{(z)}(y)=w'(y)^2c_1^{(w)}(w(y))+\frac{1}{12}\{w;z\}(y) c 1 ( z ) ( y ) = w ′ ( y ) 2 c 1 ( w ) ( w ( y )) + 12 1 { w ; z } ( y ) . Thus, R ε : = 12 c 1 R_{\varepsilon}:=12c_1 R ε := 12 c 1 is a projective connection: R ε ( z ) = w ′ 2 R ε ( w ) + { w ; z } R_{\varepsilon}^{(z)}=w'^2R_{\varepsilon}^{(w)}+\{w;z\} R ε ( z ) = w ′2 R ε ( w ) + { w ; z } .
Note that the restriction of ( L ε ⊠ L ε ) ( Δ ) (L_{\varepsilon}\boxtimes L_{\varepsilon})(\Delta) ( L ε ⊠ L ε ) ( Δ ) to Δ \Delta Δ in (ii) is an element of Γ ( Σ , ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ) \Gamma(\Sigma,(L_{\varepsilon}\boxtimes L_{\varepsilon})(\Delta)|_{\Delta}) Γ ( Σ , ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ) , and ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ≅ L ε ⊗ 2 ⊗ N Δ ≅ K Σ ⊗ T Σ ≅ O Σ (L_{\varepsilon}\boxtimes L_{\varepsilon})(\Delta)|_{\Delta}\cong L_{\varepsilon}^{\otimes2}\otimes N_{\Delta}\cong K_{\Sigma}\otimes T_{\Sigma}\cong\mathcal{O}_{\Sigma} ( L ε ⊠ L ε ) ( Δ ) ∣ Δ ≅ L ε ⊗ 2 ⊗ N Δ ≅ K Σ ⊗ T Σ ≅ O Σ canonically since L ε ⊗ 2 ≅ K Σ L_{\varepsilon}^{\otimes2}\cong K_{\Sigma} L ε ⊗ 2 ≅ K Σ and N Δ ≅ T Σ N_{\Delta}\cong T_{\Sigma} N Δ ≅ T Σ for the diagonal in Σ × Σ \Sigma\times\Sigma Σ × Σ . In the coordinate form, vanishing of the coefficient c 0 ( y ) c_0(y) c 0 ( y ) is what “no constant term” means. It is coordinate-independent because the restriction to Δ \Delta Δ described above is defined without reference to a coordinate.
Proof. (i) For γ ~ ∈ Γ ~ \widetilde{\gamma}\in\widetilde{\Gamma} γ ∈ Γ , ⟨ γ ~ x , y ⟩ = − ⟨ y , γ ~ x ⟩ = − ⟨ γ ~ − 1 y , x ⟩ \langle\widetilde{\gamma}x,y\rangle=-\langle y,\widetilde{\gamma}x\rangle=-\langle\widetilde{\gamma}^{-1}y,x\rangle ⟨ γ x , y ⟩ = − ⟨ y , γ x ⟩ = − ⟨ γ − 1 y , x ⟩ using antisymmetry of ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ and S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) -invariance. Summing over γ ~ \widetilde{\gamma} γ and reindexing γ ~ ↦ γ ~ − 1 \widetilde{\gamma}\mapsto\widetilde{\gamma}^{-1} γ ↦ γ − 1 , which is allowed by the absolute convergence of Theorem 3.5 (ii), gives s ^ ε ( x , y ) = − s ^ ε ( y , x ) \widehat{s}_{\varepsilon}(x,y)=-\widehat{s}_{\varepsilon}(y,x) s ε ( x , y ) = − s ε ( y , x ) .
(ii) Every point of Σ \Sigma Σ has a finite representative x 0 ∈ Ω 0 ∖ { ∞ } x_0\in\Omega_0\setminus\{\infty\} x 0 ∈ Ω 0 ∖ { ∞ } . The class of ∞ \infty ∞ is also represented by w − c = γ c ( ∞ ) w_{-c}=\gamma_c(\infty) w − c = γ c ( ∞ ) . Thus, by Proposition 3.6 it suffices to work near a diagonal point ( x 0 , x 0 ) (x_0,x_0) ( x 0 , x 0 ) of finite points, in the coordinate z = x z=x z = x . By Theorem 3.5 (ii) the only singular summand of the series near ( x 0 , x 0 ) (x_0,x_0) ( x 0 , x 0 ) is the one with γ ~ = 1 \widetilde{\gamma}=1 γ = 1 . Note that for γ ≠ 1 \gamma\ne1 γ = 1 one has x 0 ≠ γ x 0 x_0\ne\gamma x_0 x 0 = γ x 0 , since the action is free, and the graphs { y = γ x } \{y=\gamma x\} { y = γ x } are pairwise disjoint. Thus, s ε − 1 x − y s_{\varepsilon}-\frac{1}{x-y} s ε − x − y 1 is holomorphic there. Write s ε = 1 x − y + c 0 ( y ) + c 1 ( y ) ( x − y ) + ⋯ s_{\varepsilon}=\frac{1}{x-y}+c_0(y)+c_1(y)(x-y)+\cdots s ε = x − y 1 + c 0 ( y ) + c 1 ( y ) ( x − y ) + ⋯ . Skew-symmetry at order ( x − y ) 0 (x-y)^0 ( x − y ) 0 gives c 0 = − c 0 c_0=-c_0 c 0 = − c 0 , thus c 0 = 0 c_0=0 c 0 = 0 .
(iii) Since S ε S_{\varepsilon} S ε is a well-defined section of L ε ⊠ L ε L_{\varepsilon}\boxtimes L_{\varepsilon} L ε ⊠ L ε , s ( z ) ( x , y ) = A s ( w ) ( w ( x ) , w ( y ) ) s^{(z)}(x,y)=A\,s^{(w)}(w(x),w(y)) s ( z ) ( x , y ) = A s ( w ) ( w ( x ) , w ( y )) with A = ( w ′ ( x ) w ′ ( y ) ) 1 / 2 A=(w'(x)w'(y))^{1/2} A = ( w ′ ( x ) w ′ ( y ) ) 1/2 with branch fixed by continuity along the diagonal. Set ξ : = x − y \xi:=x-y ξ := x − y , a : = w ′ ′ ( y ) / w ′ ( y ) a:=w''(y)/w'(y) a := w ′′ ( y ) / w ′ ( y ) , b : = w ′ ′ ′ ( y ) / w ′ ( y ) b:=w'''(y)/w'(y) b := w ′′′ ( y ) / w ′ ( y ) , and expand at fixed y y y :
w ′ ( x ) w ′ ( y ) = 1 + a ξ + b 2 ξ 2 + O ( ξ 3 ) , w ( x ) − w ( y ) w ′ ( y ) ξ = 1 + a 2 ξ + b 6 ξ 2 + O ( ξ 3 ) . \frac{w'(x)}{w'(y)}=1+a\xi+\frac{b}{2}\xi^2+O(\xi^3),\qquad\frac{w(x)-w(y)}{w'(y)\xi}=1+\frac{a}{2}\xi+\frac{b}{6}\xi^2+O(\xi^3). w ′ ( y ) w ′ ( x ) = 1 + a ξ + 2 b ξ 2 + O ( ξ 3 ) , w ′ ( y ) ξ w ( x ) − w ( y ) = 1 + 2 a ξ + 6 b ξ 2 + O ( ξ 3 ) .
Since ( 1 + e ) − 2 = 1 − 2 e + 3 e 2 − ⋯ (1+e)^{-2}=1-2e+3e^2-\cdots ( 1 + e ) − 2 = 1 − 2 e + 3 e 2 − ⋯ , the second expansion gives ( w ( x ) − w ( y ) w ′ ( y ) ξ ) − 2 = 1 − a ξ + ( 3 4 a 2 − b 3 ) ξ 2 + O ( ξ 3 ) \left(\frac{w(x)-w(y)}{w'(y)\xi}\right)^{-2}=1-a\xi+\left(\frac{3}{4}a^2-\frac{b}{3}\right)\xi^2+O(\xi^3) ( w ′ ( y ) ξ w ( x ) − w ( y ) ) − 2 = 1 − a ξ + ( 4 3 a 2 − 3 b ) ξ 2 + O ( ξ 3 ) , and therefore
w ′ ( x ) w ′ ( y ) ( w ( x ) − w ( y ) ) 2 = 1 ξ 2 ( 1 + a ξ + b 2 ξ 2 ) ( 1 − a ξ + ( 3 4 a 2 − b 3 ) ξ 2 ) + O ( ξ ) = 1 ξ 2 ( 1 + ( b 6 − a 2 4 ) ξ 2 ) + O ( ξ ) = 1 ξ 2 + 1 6 { w ; z } ( y ) + O ( ξ ) , \begin{aligned}
\frac{w'(x)w'(y)}{(w(x)-w(y))^2}=\frac{1}{\xi^2}\left(1+a\xi+\frac{b}{2}\xi^2\right)\left(1-a\xi+\left(\frac{3}{4}a^2-\frac{b}{3}\right)\xi^2\right)+O(\xi) \\
&=\frac{1}{\xi^2}\left(1+\left(\frac{b}{6}-\frac{a^2}{4}\right)\xi^2\right)+O(\xi)=\frac{1}{\xi^2}+\frac{1}{6}\{w;z\}(y)+O(\xi),
\end{aligned} ( w ( x ) − w ( y ) ) 2 w ′ ( x ) w ′ ( y ) = ξ 2 1 ( 1 + a ξ + 2 b ξ 2 ) ( 1 − a ξ + ( 4 3 a 2 − 3 b ) ξ 2 ) + O ( ξ ) = ξ 2 1 ( 1 + ( 6 b − 4 a 2 ) ξ 2 ) + O ( ξ ) = ξ 2 1 + 6 1 { w ; z } ( y ) + O ( ξ ) ,
where { w ; z } : = w ′ ′ ′ w ′ − 3 2 ( w ′ ′ w ′ ) 2 = b − 3 2 a 2 \{w;z\}:=\frac{w'''}{w'}-\frac{3}{2}\left(\frac{w''}{w'}\right)^2=b-\frac{3}{2}a^2 { w ; z } := w ′ w ′′′ − 2 3 ( w ′ w ′′ ) 2 = b − 2 3 a 2 . The coefficient of ξ − 1 \xi^{-1} ξ − 1 vanishes because the two terms ± a ξ \pm a\xi ± a ξ cancel in the product. Taking the square root, with the branch A / ( w ( x ) − w ( y ) ) ∼ + 1 / ξ A/(w(x)-w(y))\sim+1/\xi A / ( w ( x ) − w ( y )) ∼ + 1/ ξ ,
A w ( x ) − w ( y ) = 1 ξ ( 1 + 1 12 { w ; z } ( y ) ξ 2 + O ( ξ 3 ) ) = 1 ξ + 1 12 { w ; z } ( y ) ξ + O ( ξ 2 ) . \frac{A}{w(x)-w(y)}=\frac{1}{\xi}\left(1+\frac{1}{12}\{w;z\}(y)\xi^2+O(\xi^3)\right)=\frac{1}{\xi}+\frac{1}{12}\{w;z\}(y)\xi+O(\xi^2). w ( x ) − w ( y ) A = ξ 1 ( 1 + 12 1 { w ; z } ( y ) ξ 2 + O ( ξ 3 ) ) = ξ 1 + 12 1 { w ; z } ( y ) ξ + O ( ξ 2 ) .
with no ξ 0 \xi^0 ξ 0 term, in agreement with part (ii). Now insert s ( w ) ( w ( x ) , w ( y ) ) = 1 w ( x ) − w ( y ) + c 1 ( w ) ( w ( y ) ) ( w ( x ) − w ( y ) ) + O ( ξ 2 ) s^{(w)}(w(x),w(y))=\frac{1}{w(x)-w(y)}+c_1^{(w)}(w(y))(w(x)-w(y))+O(\xi^2) s ( w ) ( w ( x ) , w ( y )) = w ( x ) − w ( y ) 1 + c 1 ( w ) ( w ( y )) ( w ( x ) − w ( y )) + O ( ξ 2 ) (using part (ii) in the coordinate w w w ) into s ( z ) = A s ( w ) s^{(z)}=A s^{(w)} s ( z ) = A s ( w ) and use A ( w ( x ) − w ( y ) ) = w ′ ( y ) 2 ξ + O ( ξ 2 ) A(w(x)-w(y))=w'(y)^2\xi+O(\xi^2) A ( w ( x ) − w ( y )) = w ′ ( y ) 2 ξ + O ( ξ 2 ) to get
1 ξ + c 1 ( z ) ( y ) ξ + O ( ξ 2 ) = 1 ξ + ( 1 12 { w ; z } ( y ) + w ′ ( y ) 2 c 1 ( w ) ( w ( y ) ) ) ξ + O ( ξ 2 ) . \frac{1}{\xi}+c_1^{(z)}(y)\xi+O(\xi^2)=\frac{1}{\xi}+\left(\frac{1}{12}\{w;z\}(y)+w'(y)^2c_1^{(w)}(w(y))\right)\xi+O(\xi^2). ξ 1 + c 1 ( z ) ( y ) ξ + O ( ξ 2 ) = ξ 1 + ( 12 1 { w ; z } ( y ) + w ′ ( y ) 2 c 1 ( w ) ( w ( y )) ) ξ + O ( ξ 2 ) .
Comparing the coefficients of ξ \xi ξ gives the required transformation law, and multiplying by 12 gives R ε ( z ) = w ′ 2 R ε ( w ) + { w ; z } R_{\varepsilon}^{(z)}=w'^2R_{\varepsilon}^{(w)}+\{w;z\} R ε ( z ) = w ′2 R ε ( w ) + { w ; z } .
This is an ordinary projective connection. In this paper we do not consider its counterparts such as super-projective connection that would need a supergeometric structure, cf. [12 , 14 ] .
Nonsingularity, uniqueness, identification
Let us define the canonical Szegő kernel.
Definition 3.8. For a spin structure L L L , a Szegő kernel is a meromorphic section S L S_L S L of L ⊠ L L\boxtimes L L ⊠ L on Σ × Σ \Sigma\times\Sigma Σ × Σ , holomorphic off Δ \Delta Δ , with S L − ( d x d y ) 1 / 2 x − y S_L-\frac{(dx\,dy)^{1/2}}{x-y} S L − x − y ( d x d y ) 1/2 holomorphic near Δ \Delta Δ in every local coordinate.
This is the intrinsic meaning of “simple pole with unit principal part”. The polar part is the canonical section 1 1 1 of ( L ⊠ L ) ( Δ ) ∣ Δ ≅ L 2 ⊗ T ≅ O (L\boxtimes L)(\Delta)|_\Delta\cong L^2\otimes T\cong\mathcal{O} ( L ⊠ L ) ( Δ ) ∣ Δ ≅ L 2 ⊗ T ≅ O . It is coordinate independent by Proposition 3.7 (ii). We now distinguish: (a) the series of Definition 3.1; (b) the descended section S ε S_\varepsilon S ε of Proposition 3.6 ; and (c) a canonical kernel S L S_L S L as above.
The theorem below concerns the nonsingularity, uniqueness, and identification. Let us assume (H) with g ≥ 2 g\ge2 g ≥ 2 or, more generally, the hypotheses of Theorem 3.15. Recall Proposition 3.6 . The statements of the theorem below concern only the N g N_g N g lifted spin structures L ε L_\varepsilon L ε . In particular, we do not consider the remaining spin structures (cf. Remark 3.20), for some of which H 0 ≠ 0 H^0\ne0 H 0 = 0 can occur. The vanishing H 0 ( Σ , L ε ) = 0 H^0(\Sigma,L_\varepsilon)=0 H 0 ( Σ , L ε ) = 0 is obtained in the proof from the existence of S ε S_\varepsilon S ε , a consequence of the convergence hypotheses.
Theorem 3.10. (i) H 0 ( Σ , L ε ) = 0 H^0(\Sigma,L_\varepsilon)=0 H 0 ( Σ , L ε ) = 0 for every one of the N g N_g N g lifts, for all parameters satisfying (H). (ii) If S ′ S' S ′ is any section of L ε ⊠ L ε L_\varepsilon\boxtimes L_\varepsilon L ε ⊠ L ε with the properties of a Szegő kernel, then S ′ = S ε S'=S_\varepsilon S ′ = S ε . Thus, (a)=(b)=(c) mentioned above.
(i) is a statement about that family on this domain only. We do not claim that all even spin structures are nonsingular.
Proof. (i) Let L = L ε L=L_\varepsilon L = L ε and suppose k : = h 0 ( L ) > 0 k:=h^0(L)>0 k := h 0 ( L ) > 0 . Choose 0 ≠ ϕ ∈ H 0 ( L ) 0\ne\phi\in H^0(L) 0 = ϕ ∈ H 0 ( L ) . Its zero divisor has degree deg L = g − 1 \deg L=g-1 deg L = g − 1 , hence is finite. We choose p ∈ Σ p\in\Sigma p ∈ Σ outside it. The sections of L ( − p ) L(-p) L ( − p ) are the sections of L L L vanishing at p p p , i.e., the kernel of the evaluation map H 0 ( L ) → L ∣ p H^0(L)\to L|_p H 0 ( L ) → L ∣ p , which has rank ≤ 1 \le1 ≤ 1 , and ϕ ∉ H 0 ( L ( − p ) ) \phi\notin H^0(L(-p)) ϕ ∈ / H 0 ( L ( − p )) . Hence, h 0 ( L ( − p ) ) = k − 1 h^0(L(-p))=k-1 h 0 ( L ( − p )) = k − 1 . Riemann–Roch for L ( p ) L(p) L ( p ) , of degree g g g , gives h 0 ( L ( p ) ) = h 1 ( L ( p ) ) h^0(L(p))=h^1(L(p)) h 0 ( L ( p )) = h 1 ( L ( p )) . Serre duality with K Σ ⊗ L − 1 ≅ L K_\Sigma\otimes L^{-1}\cong L K Σ ⊗ L − 1 ≅ L gives h 1 ( L ( p ) ) = h 0 ( L ( − p ) ) = k − 1 h^1(L(p))=h^0(L(-p))=k-1 h 1 ( L ( p )) = h 0 ( L ( − p )) = k − 1 . Hence, h 0 ( L ( p ) ) = k = h 0 ( L ) h^0(L(p))=k=h^0(L) h 0 ( L ( p )) = k = h 0 ( L ) , and the injection H 0 ( L ) ↪ H 0 ( L ( p ) ) H^0(L)\hookrightarrow H^0(L(p)) H 0 ( L ) ↪ H 0 ( L ( p )) is onto. Every section of L ( p ) L(p) L ( p ) is regular at p p p . On the other hand, take y ∈ Ω 0 y\in\Omega_0 y ∈ Ω 0 with [ y ] = p [y]=p [ y ] = p . By Proposition 3.6 the kernel S ε ( ⋅ , y ) S_\varepsilon(\cdot,y) S ε ( ⋅ , y ) is a global meromorphic section of L L L on Σ \Sigma Σ whose only singularity is a simple pole at p p p with nonzero principal part (Proposition 3.7 (ii)), i.e., a section of L ( p ) L(p) L ( p ) not regular at p p p . This contradiction shows k = 0 k=0 k = 0 . Proposition 3.6 and Theorem 3.5 (ii) provide the global existence of S ε ( ⋅ , y ) S_\varepsilon(\cdot,y) S ε ( ⋅ , y ) , including across the points over ∞ \infty ∞ .
(ii) The difference S ′ ( ⋅ , y ) − S ε ( ⋅ , y ) S'(\cdot,y)-S_\varepsilon(\cdot,y) S ′ ( ⋅ , y ) − S ε ( ⋅ , y ) is holomorphic on Σ \Sigma Σ since there is no pole at x = y x=y x = y by the principal-part condition. Thus, it vanishes by (i).
Remark 3.13. S ε S_{\varepsilon} S ε is a function of λ a \lambda_a λ a , not of ρ a = − λ a 2 \rho_a=-\lambda_a^2 ρ a = − λ a 2 . The sign of λ a \lambda_a λ a labels the lift. Continuing λ a ↦ e i φ λ a \lambda_a\mapsto e^{i\varphi}\lambda_a λ a ↦ e i φ λ a along φ ∈ [ 0 , π ] \varphi\in[0,\pi] φ ∈ [ 0 , π ] assuming (H) (a half-turn around the excluded point λ a = 0 \lambda_a=0 λ a = 0 ) exchanges the two lifts of γ a \gamma_a γ a . Hence, it permutes the N g N_g N g kernels { S ε } \{S_{\varepsilon}\} { S ε } . Continuation of S ε S_{\varepsilon} S ε beyond the domains of Section 3.3 (for instance, across the theta-null divisor θ [ δ ] ( 0 ) = 0 \theta[\delta](0)=0 θ [ δ ] ( 0 ) = 0 , where by Theorem 3.10 the series cannot converge in the sense of Theorem 3.15) is not treated here (see § 7 ).
Convergence beyond (H). Möbius covariance
Condition (H) is a coordinate-dependent, small neighbourhood of the maximal degeneration. The following facts show that the results of Section 3 do not depend on it. The statement on the Möbius covariance is the following. For h ∈ S L 2 ( C ) h\in\mathrm{SL}_2(\mathbb{C}) h ∈ SL 2 ( C ) set a lift of h Γ h − 1 h\Gamma h^{-1} h Γ h − 1 as Γ ~ h : = h Γ ~ h − 1 \widetilde{\Gamma}^{h}:=h\widetilde{\Gamma}h^{-1} Γ h := h Γ h − 1 . Then
Proposition 3.14. s ^ Γ ~ h ( h x , h y ) = s ^ Γ ~ ( x , y ) \widehat{s}_{\widetilde{\Gamma}^{h}}(hx,hy)=\widehat{s}_{\widetilde{\Gamma}}(x,y) s Γ h ( h x , h y ) = s Γ ( x , y ) whenever one side converges absolutely.
In particular, the 3 g 3g 3 g parameters ( w , λ ) (w,\lambda) ( w , λ ) are redundant by the action of P S L 2 ( C ) \mathrm{PSL}_2(\mathbb{C}) PSL 2 ( C ) , and the kernels S ε S_{\varepsilon} S ε are objects attached to a point of Schottky space together with a lift.
Proof. γ ~ ↦ h γ ~ h − 1 \widetilde{\gamma}\mapsto h\widetilde{\gamma}h^{-1} γ ↦ h γ h − 1 is a bijection Γ ~ → Γ ~ h \widetilde{\Gamma}\to\widetilde{\Gamma}^{h} Γ → Γ h , and ⟨ h γ ~ h − 1 h x , h y ⟩ = ⟨ h γ ~ x , h y ⟩ = ⟨ γ ~ x , y ⟩ \langle h\widetilde{\gamma}h^{-1}hx,hy\rangle=\langle h\widetilde{\gamma}x,hy\rangle=\langle\widetilde{\gamma}x,y\rangle ⟨ h γ h − 1 h x , h y ⟩ = ⟨ h γ x , h y ⟩ = ⟨ γ x , y ⟩ by S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) -invariance of ⟨ ⋅ , ⋅ ⟩ \langle\cdot,\cdot\rangle ⟨ ⋅ , ⋅ ⟩ . □ \square □
Let ∥ ⋅ ∥ \|\cdot\| ∥ ⋅ ∥ be the operator norm on C 2 \mathbb{C}^{2} C 2 (Euclidean). We consider the condition
∑ γ ~ ∈ Γ ~ ∥ γ ~ ∥ − 1 < ∞ . \sum_{\widetilde{\gamma}\in\widetilde{\Gamma}}\|\widetilde{\gamma}\|^{-1}<\infty. γ ∈ Γ ∑ ∥ γ ∥ − 1 < ∞.
It does not depend on the choice of lift since the lifts differ by signs, and it is invariant under conjugation in S L 2 ( C ) \mathrm{SL}_2(\mathbb{C}) SL 2 ( C ) (norms change by bounded factors), hence intrinsic to Γ \Gamma Γ . Under conditions (D) and (E) the convergence is described by
Theorem 3.15. Let Γ \Gamma Γ be a Schottky group of genus g ≥ 1 g\geq1 g ≥ 1 as in Section 2, satisfying (D), and let Γ ~ \widetilde{\Gamma} Γ be a lift satisfying (E). Then:
For compact K , K ′ ⊂ Ω 0 K,K'\subset\Omega_0 K , K ′ ⊂ Ω 0 there are a finite set F ⊂ Γ ~ F\subset\widetilde{\Gamma} F ⊂ Γ and η > 0 \eta>0 η > 0 such that ∣ ⟨ γ ~ x , y ⟩ ∣ ≥ η ∥ γ ~ ∥ |\langle\widetilde{\gamma}x,y\rangle|\geq\eta\|\widetilde{\gamma}\| ∣ ⟨ γ x , y ⟩ ∣ ≥ η ∥ γ ∥ for all γ ~ ∉ F \widetilde{\gamma}\notin F γ ∈ / F and all unit vectors x , y x,y x , y over K , K ′ K,K' K , K ′ . Hence, the series (2) converges absolutely and locally uniformly on { ( x , y ) over Ω 0 × Ω 0 : [ y ] ∉ Γ [ x ] } \{(x,y)\text{ over }\Omega_0\times\Omega_0:[y]\notin\Gamma[x]\} {( x , y ) over Ω 0 × Ω 0 : [ y ] ∈ / Γ [ x ]} (the points ∞ \infty ∞ and Γ ∞ \Gamma\infty Γ∞ included), is meromorphic on the cone over Ω 0 × Ω 0 \Omega_0\times\Omega_0 Ω 0 × Ω 0 , with simple poles along { [ y ] ∈ Γ [ x ] } \{[y]\in\Gamma[x]\} {[ y ] ∈ Γ [ x ]} , and satisfies the invariance (4) and the gluing relation (5).
Consequently, Proposition 3.6, Proposition 3.7 (i)–(iii), Theorem 3.10 (in particular H 0 ( Σ , L ε ) = 0 H^0(\Sigma,L_{\varepsilon})=0 H 0 ( Σ , L ε ) = 0 ), Propositions 3.16, 3.17 and Corollary 3.18, Theorem 4.1, Proposition 4.5, Proposition 6.1 hold, with the same proofs, for every such Γ ~ \widetilde{\Gamma} Γ .
For every g ≥ 1 g\geq1 g ≥ 1 , condition (H) implies (D) and (E). If the critical exponent δ ( Γ ) < 1 2 \delta(\Gamma)<\frac{1}{2} δ ( Γ ) < 2 1 (the abscissa of convergence of ∑ γ e − s d ( o , γ o ) \sum_{\gamma}e^{-s d(o,\gamma o)} ∑ γ e − s d ( o , γ o ) for the hyperbolic distance in H 3 \mathbb{H}^3 H 3 ), then (E) holds.
Note that condition (H) is meaningful for all g\geq 1. In the proof the constants of Lemma 3.3 are not used. Instead, we use only Lemma 3.2, valid for all g\geq 1. Genus one does not need (H): it is covered by (E) (Proposition 3.19).
Proof. (i) Fix K K K , K ′ K' K ′ and η 0 > 0 \eta_0>0 η 0 > 0 so small that the closed chordal η 0 \eta_0 η 0 -neighbourhood K ′ ′ K'' K ′′ of K ′ K' K ′ belongs to Ω 0 \Omega_0 Ω 0 (Ω 0 \Omega_0 Ω 0 is open). By proper discontinuity the set F 1 : = { γ ~ : γ K ∩ K ′ ′ ≠ ∅ } F_1:=\{\widetilde{\gamma}:\gamma K\cap K''\ne\varnothing\} F 1 := { γ : γ K ∩ K ′′ = ∅ } is finite. For γ ~ ∉ F 1 \widetilde{\gamma}\notin F_1 γ ∈ / F 1 , γ x \gamma x γ x has chordal distance ≥ η 0 \geq\eta_0 ≥ η 0 from y y y for x ∈ K , y ∈ K ′ x\in K,y\in K' x ∈ K , y ∈ K ′ . For unit vectors one has ∣ ⟨ u , v ⟩ ∣ = ∣ u ∣ ∣ v ∣ χ ( [ u ] , [ v ] ) |\langle u,v\rangle|=|u||v|\chi([u],[v]) ∣ ⟨ u , v ⟩ ∣ = ∣ u ∣∣ v ∣ χ ([ u ] , [ v ]) with χ ∈ [ 0 , 1 ] \chi\in[0,1] χ ∈ [ 0 , 1 ] the chordal distance (Lagrange identity ∣ u 1 v 2 − u 2 v 1 ∣ 2 = ∣ u ∣ 2 ∣ v ∣ 2 − ∣ u 1 v ‾ 1 + u 2 v ‾ 2 ∣ 2 |u_1v_2-u_2v_1|^2=|u|^2|v|^2-|u_1\overline{v}_1+u_2\overline{v}_2|^2 ∣ u 1 v 2 − u 2 v 1 ∣ 2 = ∣ u ∣ 2 ∣ v ∣ 2 − ∣ u 1 v 1 + u 2 v 2 ∣ 2 ). Thus, ∣ ⟨ γ ~ x , y ⟩ ∣ ≥ η 0 ∣ γ ~ x ∣ |\langle\widetilde{\gamma}x,y\rangle|\geq\eta_0|\widetilde{\gamma}x| ∣ ⟨ γ x , y ⟩ ∣ ≥ η 0 ∣ γ x ∣ for γ ∉ F 1 \gamma\notin F_1 γ ∈ / F 1 . Next, let σ 1 ≥ σ 2 = σ 1 − 1 \sigma_1\geq\sigma_2=\sigma_1^{-1} σ 1 ≥ σ 2 = σ 1 − 1 be the singular values of γ ~ \widetilde{\gamma} γ (det γ ~ = 1 \det\widetilde{\gamma}=1 det γ = 1 ), v 1 , v 2 v_1,v_2 v 1 , v 2 orthonormal right singular vectors. Denoting x = c 1 v 1 + c 2 v 2 x=c_1v_1+c_2v_2 x = c 1 v 1 + c 2 v 2 , we obtain ∣ γ ~ x ∣ 2 = σ 1 2 ∣ c 1 ∣ 2 + σ 2 2 ∣ c 2 ∣ 2 ≥ ∥ γ ~ ∥ 2 χ ( [ x ] , [ v 2 ] ) 2 |\widetilde{\gamma}x|^2=\sigma_1^2|c_1|^2+\sigma_2^2|c_2|^2\geq\|\widetilde{\gamma}\|^2\chi([x],[v_2])^2 ∣ γ x ∣ 2 = σ 1 2 ∣ c 1 ∣ 2 + σ 2 2 ∣ c 2 ∣ 2 ≥ ∥ γ ∥ 2 χ ([ x ] , [ v 2 ] ) 2 again by Lagrange identity, ∣ c 1 ∣ 2 = χ ( [ x ] , [ v 2 ] ) |c_1|^2=\chi([x],[v_2]) ∣ c 1 ∣ 2 = χ ([ x ] , [ v 2 ]) .
Consider the following claim: there are a finite F 2 F_2 F 2 and η 1 > 0 \eta_1>0 η 1 > 0 with χ ( [ x ] , [ v 2 ( γ ~ ) ] ) ≥ η 1 \chi([x],[v_2(\widetilde{\gamma})])\geq\eta_1 χ ([ x ] , [ v 2 ( γ )]) ≥ η 1 for γ ~ ∉ F 2 \widetilde{\gamma}\notin F_2 γ ∈ / F 2 , x x x over K K K . Otherwise there are distinct γ ~ n \widetilde{\gamma}_n γ n and x n ∈ K x_n\in K x n ∈ K with χ ( x n , [ v 2 ( γ ~ n ) ] ) → 0 \chi(x_n,[v_2(\widetilde{\gamma}_n)])\to0 χ ( x n , [ v 2 ( γ n )]) → 0 . Passing to a subsequence, x n → b ∈ K x_n\to b\in K x n → b ∈ K , [ v 2 ( γ ~ n ) ] → b [v_2(\widetilde{\gamma}_n)]\to b [ v 2 ( γ n )] → b and γ ~ n / ∥ γ ~ n ∥ → A = u 1 v 1 ∗ \widetilde{\gamma}_n/\|\widetilde{\gamma}_n\|\to A=u_1v_1^* γ n /∥ γ n ∥ → A = u 1 v 1 ∗ of rank one (here ∥ γ ~ n ∥ → ∞ \|\widetilde{\gamma}_n\|\to\infty ∥ γ n ∥ → ∞ since Γ ~ \widetilde{\Gamma} Γ is discrete), with ker A = b \ker A=b ker A = b . For y ≠ b y\ne b y = b we get γ n y → a : = [ u 1 ] \gamma_ny\to a:=[u_1] γ n y → a := [ u 1 ] , with the γ n y \gamma_ny γ n y pairwise distinct points of an orbit of a point of Ω 0 \Omega_0 Ω 0 , hence a ∈ Λ a\in\Lambda a ∈ Λ . Applying the same reasoning to γ ~ n − 1 \widetilde{\gamma}_n^{-1} γ n − 1 (whose dominant left singular vector is v 2 ( γ ~ n ) v_2(\widetilde{\gamma}_n) v 2 ( γ n ) ) gives b ∈ Λ b\in\Lambda b ∈ Λ . This contradicts b ∈ K ⊂ Ω 0 b\in K\subset\Omega_0 b ∈ K ⊂ Ω 0 . Set F : = F 1 ∪ F 2 F:=F_1\cup F_2 F := F 1 ∪ F 2 and η : = η 0 η 1 \eta:=\eta_0\eta_1 η := η 0 η 1 . Then ∣ ⟨ γ ~ x , y ⟩ ∣ ≥ η ∥ γ ~ ∥ |\langle\widetilde{\gamma}x,y\rangle|\geq\eta\|\widetilde{\gamma}\| ∣ ⟨ γ x , y ⟩ ∣ ≥ η ∥ γ ∥ for γ ~ ∉ F \widetilde{\gamma}\notin F γ ∈ / F , and (E) gives uniform absolute convergence of the sum over γ ~ ∉ F \widetilde{\gamma}\notin F γ ∈ / F on K × K ′ K\times K' K × K ′ . The finitely many remaining terms are meromorphic. Bihomogeneity gives the same for vectors of bounded norm. Invariance (4) follows by reindexing γ ~ ↦ μ ~ − 1 γ ~ λ ~ \widetilde{\gamma} \mapsto\widetilde{\mu}^{-1}\widetilde{\gamma}\widetilde{\lambda} γ ↦ μ − 1 γ λ , which is allowed by absolute convergence, and (5) by the computation in Theorem 3.5 (ii).
Poles: at a point ( α , γ 0 α ) (\alpha,\gamma_0\alpha) ( α , γ 0 α ) only the term of γ ~ 0 \widetilde{\gamma}_0 γ 0 is singular (due to freeness of the action). It has a simple pole, and the sum of the others is holomorphic nearby (finitely many terms of F F F with γ ≠ γ 0 \gamma\ne\gamma_0 γ = γ 0 , and others are the uniformly convergent).
(ii) The proofs of the listed statements use only the invariance (4), the meromorphicity of s ^ ε \widehat{s}_{\varepsilon} s ε on the cone over Ω 0 × Ω 0 \Omega_0 \times\Omega_0 Ω 0 × Ω 0 with simple poles along { y ∈ Γ x } \{y \in\Gamma x\} { y ∈ Γ x } and local finiteness of the family of graphs. All of these are now provided by (i). For g = 1 g=1 g = 1 the Riemann-Roch argument of Theorem 3.10 applies: deg L ε = 0 \deg L_{\varepsilon}=0 deg L ε = 0 and the zero divisor of a nonzero section is empty.
(iii) That (H) implies (D) was noted above. Let x 0 : = w 1 + d / 2 x_0 := w_1+d/2 x 0 := w 1 + d /2 . Then ∣ x 0 − w c ∣ ≥ d / 2 ≥ r c / 2 |x_0-w_c| \ge d/2 \ge r_c/2 ∣ x 0 − w c ∣ ≥ d /2 ≥ r c /2 for all c c c , i.e., x 0 ∈ K x_0 \in K x 0 ∈ K . By Lemma 3.2, ∣ c γ ~ x 0 + d γ ~ ∣ = ∣ j ( γ ~ , x 0 ) ∣ − 1 ≥ 2 ϖ − ( n − 1 ) |c_{\widetilde{\gamma}}x_0+d_{\widetilde{\gamma}}|=|j(\widetilde{\gamma},x_0)|^{-1} \ge2\varpi^{-(n-1)} ∣ c γ x 0 + d γ ∣ = ∣ j ( γ , x 0 ) ∣ − 1 ≥ 2 ϖ − ( n − 1 ) for ∣ γ ~ ∣ = n ≥ 1 |\widetilde{\gamma}|=n \ge1 ∣ γ ∣ = n ≥ 1 , and ∥ γ ~ ∥ ≥ ∣ γ ~ ( x 0 , 1 ) ∣ / ∣ ( x 0 , 1 ) ∣ ≥ ∣ c γ ~ x 0 + d γ ~ ∣ / ( 1 + ∣ x 0 ∣ ) \|\widetilde{\gamma}\| \ge|\widetilde{\gamma}(x_0,1)|/|(x_0,1)| \ge|c_{\widetilde{\gamma}}x_0+d_{\widetilde{\gamma}}|/(1+|x_0|) ∥ γ ∥ ≥ ∣ γ ( x 0 , 1 ) ∣/∣ ( x 0 , 1 ) ∣ ≥ ∣ c γ x 0 + d γ ∣/ ( 1 + ∣ x 0 ∣ ) . Therefore, ∑ γ ~ ≠ 1 ∥ γ ~ ∥ − 1 ≤ 2 ( 1 + ∣ x 0 ∣ ) ∑ n ≥ 1 2 g ( 2 g − 1 ) n − 1 ϖ n − 1 ≤ 4 g ( 1 + ∣ x 0 ∣ ) ∑ m ≥ 0 4 − m < ∞ \sum_{\widetilde{\gamma}\ne1}\|\widetilde{\gamma}\|^{-1} \le2(1+|x_0|)\sum_{n\ge1}2g(2g-1)^{n-1}\varpi^{n-1} \le4g(1+|x_0|)\sum_{m\ge0}4^{-m}<\infty ∑ γ = 1 ∥ γ ∥ − 1 ≤ 2 ( 1 + ∣ x 0 ∣ ) ∑ n ≥ 1 2 g ( 2 g − 1 ) n − 1 ϖ n − 1 ≤ 4 g ( 1 + ∣ x 0 ∣ ) ∑ m ≥ 0 4 − m < ∞ . For the last claim, ∥ γ ~ ∥ 2 ≍ e d ( o , γ o ) \|\widetilde{\gamma}\|^2 \asymp e^{d(o,\gamma o)} ∥ γ ∥ 2 ≍ e d ( o , γ o ) for a base point o ∈ H 3 o \in\mathbb{H}^3 o ∈ H 3 , thus ∑ ∥ γ ~ ∥ − 1 ≍ ∑ γ e − d ( o , γ o ) / 2 \sum\|\widetilde{\gamma}\|^{-1}\asymp\sum_{\gamma}e^{-d(o,\gamma o)/2} ∑ ∥ γ ∥ − 1 ≍ ∑ γ e − d ( o , γ o ) /2 , which converges when 1 2 > δ ( Γ ) \frac{1}{2}>\delta(\Gamma) 2 1 > δ ( Γ ) (definition of the critical exponent, see [15 ] ). □
A few remarks are in order. (1) Numerically (genus two, random data with r / d r/d r / d up to 0.3, far outside (H)), the series converges geometrically in the word length and the relation (5) holds to the accuracy of the truncation. (2) By Theorem 3.10 and the classical criterion θ [ δ ] ( 0 ) = 0 \theta[\delta](0)=0 θ [ δ ] ( 0 ) = 0 iff h 0 > 0 h^0>0 h 0 > 0 , (E) fails at every Schottky group for which θ [ δ ( ε ) ] ( 0 ) = 0 \theta[\delta(\varepsilon)](0)=0 θ [ δ ( ε )] ( 0 ) = 0 . Thus the locus where (E) holds avoids those theta-null divisors. (3) Whether (E) fails when δ ( Γ ) ≥ 1 2 \delta(\Gamma)\ge\frac{1}{2} δ ( Γ ) ≥ 2 1 is not claimed.
Relation to the prime form and the canonical bidifferential
Normalization: fix the marking α a = ∂ D a \alpha_a=\partial D_a α a = ∂ D a , β a \beta_a β a a path joining ∂ D − a \partial D_{-a} ∂ D − a to ∂ D a \partial D_a ∂ D a through D \mathcal{D} D . Let ζ = ( ζ 1 , … , ζ g ) \zeta=(\zeta_1,\ldots,\zeta_g) ζ = ( ζ 1 , … , ζ g ) be the holomorphic 1-forms normalized by ∮ α a ζ b = δ a b \oint_{\alpha_a}\zeta_b=\delta_{ab} ∮ α a ζ b = δ ab , τ a b = ∮ β a ζ b \tau_{ab}=\oint_{\beta_a}\zeta_b τ ab = ∮ β a ζ b , thus Im τ > 0 \operatorname{Im}\tau>0 Im τ > 0 , and
θ [ δ ] ( z ) = ∑ n ∈ Z g exp ( π i ( n + δ ′ ) T τ ( n + δ ′ ) + 2 π i ( n + δ ′ ) T ( z + δ ′ ′ ) ) . \theta[\delta](z)=\sum_{n\in\mathbb{Z}^g}\exp\left(\pi i(n+\delta')^T\tau(n+\delta')+2\pi i(n+\delta')^T(z+\delta'')\right). θ [ δ ] ( z ) = n ∈ Z g ∑ exp ( π i ( n + δ ′ ) T τ ( n + δ ′ ) + 2 π i ( n + δ ′ ) T ( z + δ ′′ ) ) .
Let B ( x , y ) B(x,y) B ( x , y ) be the canonical bidifferential normalized by ∮ α a B ( ⋅ , y ) = 0 \oint_{\alpha_a}B(\cdot,y)=0 ∮ α a B ( ⋅ , y ) = 0 . We denote by E E E the prime form, E ( x , y ) = ( x − y ) ( d x ) − 1 / 2 ( d y ) − 1 / 2 ( 1 + O ( ( x − y ) 2 ) ) E(x,y)=(x-y)(dx)^{-1/2}(dy)^{-1/2}(1+O((x-y)^2)) E ( x , y ) = ( x − y ) ( d x ) − 1/2 ( d y ) − 1/2 ( 1 + O (( x − y ) 2 )) . Thus, B ( x , y ) = ∂ x ∂ y log E d x d y = d x d y ( x − y ) 2 + ⋯ B(x,y)=\partial_x\partial_y\log E\,dx\,dy=\frac{dx\,dy}{(x-y)^2}+\cdots B ( x , y ) = ∂ x ∂ y log E d x d y = ( x − y ) 2 d x d y + ⋯ (note that ∂ x ∂ y log ( x − y ) = + ( x − y ) − 2 \partial_x\partial_y\log(x-y)=+(x-y)^{-2} ∂ x ∂ y log ( x − y ) = + ( x − y ) − 2 ). Since E E E is skew symmetric, E ( x , y ) = ( x − y ) ( d x d y ) − 1 / 2 ( 1 + e 2 ( y ) ( x − y ) 2 + O ( ( x − y ) 3 ) ) E(x,y)=(x-y)(dx\,dy)^{-1/2}(1+e_2(y)(x-y)^2+O((x-y)^3)) E ( x , y ) = ( x − y ) ( d x d y ) − 1/2 ( 1 + e 2 ( y ) ( x − y ) 2 + O (( x − y ) 3 )) , i.e., there is no linear term. We use the following standard facts [9 , 13 ] . Note that B = ∂ x ∂ y log E B=\partial_x\partial_y\log E B = ∂ x ∂ y log E . As a function of x x x in a fixed local coordinate, E ( x , y ) E(x,y) E ( x , y ) is single valued along α \alpha α -cycles and is multiplied by − exp ( − π i τ b b − 2 π i ∫ y x ζ b ) -\exp(-\pi i\tau_{bb}-2\pi i\int_y^x\zeta_b) − exp ( − π i τ bb − 2 π i ∫ y x ζ b ) along β b \beta_b β b . For θ [ δ ] ( 0 ) ≠ 0 \theta[\delta](0)\ne0 θ [ δ ] ( 0 ) = 0 , the Szegő kernel of L δ L_\delta L δ is the expression in Proposition 3.16. The following proposition is due to Fay [9 , 13 ] .
Proposition 3.16. Assume the hypotheses of Theorem 3.15 (in particular (H) with g ≥ 2 g\ge2 g ≥ 2 , or g = 1 g=1 g = 1 and (E)).
(a) For δ = δ ( ε ) \delta=\delta(\varepsilon) δ = δ ( ε ) (Convention of Section 2) one has θ [ δ ] ( 0 ) ≠ 0 \theta[\delta](0)\ne0 θ [ δ ] ( 0 ) = 0 (by Theorem 3.10 (i) and the criterion above), and
S ε ( x , y ) = θ [ δ ] ( ∫ y x ζ ) θ [ δ ] ( 0 ) E ( x , y ) . S_{\varepsilon}(x,y)=\frac{\theta[\delta]\left(\int_y^x\zeta\right)}{\theta[\delta](0)E(x,y)}. S ε ( x , y ) = θ [ δ ] ( 0 ) E ( x , y ) θ [ δ ] ( ∫ y x ζ ) . (b) δ ′ = 0 \delta'=0 δ ′ = 0 , i.e., δ ( ε ) = [ 0 ; δ ′ ′ ] \delta(\varepsilon)=[0;\delta''] δ ( ε ) = [ 0 ; δ ′′ ] with δ ′ ′ ∈ { 0 , 1 2 } g \delta''\in\left\{0,\frac{1}{2}\right\}^g δ ′′ ∈ { 0 , 2 1 } g . In particular δ ( ε ) \delta(\varepsilon) δ ( ε ) is even.
For odd characteristics (not in our family) θ [ δ ] ( 0 ) = 0 \theta[\delta](0)=0 θ [ δ ] ( 0 ) = 0 and the formula is unavailable.
Proof. (a) By the quoted fact the right-hand side is a Szegő kernel of L δ ≅ L ε L_\delta\cong L_\varepsilon L δ ≅ L ε , and by Theorem 3.10 (ii) it equals S ε S_\varepsilon S ε . (b) Let x x x run once around α b = ∂ D b \alpha_b=\partial D_b α b = ∂ D b in Ω 0 \Omega_0 Ω 0 with y y y fixed. The scalar function s ε ( ⋅ , y ) s_\varepsilon(\cdot,y) s ε ( ⋅ , y ) is single valued on Ω 0 \Omega_0 Ω 0 (the loop is closed in Ω 0 \Omega_0 Ω 0 , where the trivialization by ( d z ) 1 / 2 (dz)^{1/2} ( d z ) 1/2 is global). On the right-hand side, ∫ y x ζ \int_y^x\zeta ∫ y x ζ increases by δ b c \delta_{bc} δ b c , thus θ [ δ ] \theta[\delta] θ [ δ ] is multiplied by e 2 π i δ b ′ e^{2\pi i\delta'_b} e 2 π i δ b ′ , while E ( x , y ) E(x,y) E ( x , y ) is single valued along α b \alpha_b α b . Hence e 2 π i δ b ′ = 1 e^{2\pi i\delta'_b}=1 e 2 π i δ b ′ = 1 for all b b b , i.e., δ ′ = 0 \delta'=0 δ ′ = 0 . Then 4 δ ′ ⋅ δ ′ ′ = 0 4\delta'\cdot\delta''=0 4 δ ′ ⋅ δ ′′ = 0 and δ \delta δ is even. □
The identification of the bundle with L δ ( ε ) L_{\delta(\varepsilon)} L δ ( ε ) is the Convention of Section 2. Let us check normalization. As y → x y\to x y → x , θ [ δ ] ( ∫ y x ζ ) → θ [ δ ] ( 0 ) \theta[\delta](\int_y^x\zeta)\to\theta[\delta](0) θ [ δ ] ( ∫ y x ζ ) → θ [ δ ] ( 0 ) and E ( x , y ) → ( x − y ) ( d x ) − 1 / 2 ( d y ) − 1 / 2 E(x,y)\to(x-y)(dx)^{-1/2}(dy)^{-1/2} E ( x , y ) → ( x − y ) ( d x ) − 1/2 ( d y ) − 1/2 by the stated normalization of E E E . Thus, the right-hand side tends to
θ [ δ ] ( 0 ) θ [ δ ] ( 0 ) ( x − y ) ( d x ) − 1 / 2 ( d y ) − 1 / 2 = ( d x d y ) 1 / 2 x − y , \frac{\theta[\delta](0)}{\theta[\delta](0)(x-y)(dx)^{-1/2}(dy)^{-1/2}}=\frac{(dx\,dy)^{1/2}}{x-y}, θ [ δ ] ( 0 ) ( x − y ) ( d x ) − 1/2 ( d y ) − 1/2 θ [ δ ] ( 0 ) = x − y ( d x d y ) 1/2 ,
which is the principal part of S ε S_{\varepsilon} S ε fixed in Definition 3.1 with no extra sign, confirming that the stated normalization of E E E is the one compatible with our S ε S_{\varepsilon} S ε .
Review of conventions
(a) Characteristics: with the theta function displayed above, θ [ δ ] ( z + e b ) = e 2 π i δ b ′ θ [ δ ] ( z ) \theta[\delta](z+e_b)=e^{2\pi i\delta'_b}\theta[\delta](z) θ [ δ ] ( z + e b ) = e 2 π i δ b ′ θ [ δ ] ( z ) for the b b b -th unit vector e b e_b e b (periods along α b \alpha_b α b ). This is the multiplier used in Proposition 3.16 (b). Only the intrinsic bundle isomorphism L δ ( ε ) ≅ L ε L_{\delta(\varepsilon)}\cong L_{\varepsilon} L δ ( ε ) ≅ L ε is used for δ ( ε ) \delta(\varepsilon) δ ( ε ) with no sign convention for the lifts entering.
(b) Parity and sign: for even δ \delta δ , θ [ δ ] ( − z ) = θ [ δ ] ( z ) \theta[\delta](-z)=\theta[\delta](z) θ [ δ ] ( − z ) = θ [ δ ] ( z ) and E ( y , x ) = − E ( x , y ) E(y,x)=-E(x,y) E ( y , x ) = − E ( x , y ) . Thus, the right-hand side of Proposition 3.16 (a) is skew symmetric in ( x , y ) (x,y) ( x , y ) as required by Proposition 3.7 (i).
(c) Overall sign: the sign of the prime form is fixed here by E ( x , y ) ∼ ( x − y ) ( d x ) − 1 / 2 ( d y ) − 1 / 2 E(x,y)\sim(x-y)(dx)^{-1/2}(dy)^{-1/2} E ( x , y ) ∼ ( x − y ) ( d x ) − 1/2 ( d y ) − 1/2 , equivalently by the requirement that S ε S_{\varepsilon} S ε have principal part + ( d x d y ) 1 / 2 x − y +\frac{(dx\,dy)^{1/2}}{x-y} + x − y ( d x d y ) 1/2 (Definition 3.1). The footnote in [18 ] , Section 2, mentions that their prime form differs by a factor − 1 -1 − 1 from that of [13 , 9 ] . With the opposite sign for E E E one obtains the same kernel with the opposite sign in the formula.
(d) The formula for S ε S_{\varepsilon} S ε is also confirmed independently at genus one (Proposition 3.19 ), including the sign and the normalization of the prime form.
Proposition 3.17. S ε ( x , y ) 2 − B ( x , y ) = ∑ i , j c i j ζ i ( x ) ζ j ( y ) S_{\varepsilon}(x,y)^2-B(x,y)=\sum_{i,j}c_{ij}\zeta_i(x)\zeta_j(y) S ε ( x , y ) 2 − B ( x , y ) = ∑ i , j c ij ζ i ( x ) ζ j ( y ) is a holomorphic bidifferential.
Proof. The statement comes from using H 0 ( Σ × Σ , K ⊠ K ) ≅ H 0 ( Σ , K ) ⊗ H 0 ( Σ , K ) H^0(\Sigma\times\Sigma,K\boxtimes K)\cong H^0(\Sigma,K)\otimes H^0(\Sigma,K) H 0 ( Σ × Σ , K ⊠ K ) ≅ H 0 ( Σ , K ) ⊗ H 0 ( Σ , K ) , a standard consequence of the Künneth formula, together with the choice of basis ζ i \zeta_i ζ i . S ε 2 S_{\varepsilon}^2 S ε 2 is a section of K ⊠ K K\boxtimes K K ⊠ K (signs square to 1 by Proposition 3.6 ) with principal part d x d y ( x − y ) 2 \frac{dx\,dy}{(x-y)^2} ( x − y ) 2 d x d y near Δ \Delta Δ and, by Proposition 3.7 (ii), there is no term of order ( x − y ) − 1 (x-y)^{-1} ( x − y ) − 1 . The same holds for B B B by definition of the canonical bidifferential. Hence, S ε 2 − B S_{\varepsilon}^2-B S ε 2 − B is holomorphic on Σ × Σ \Sigma\times\Sigma Σ × Σ , and the displayed decomposition follows from the Künneth isomorphism. □
Let us describe now projective connection and Bergman projective connection.
Corollary 3.18. Let S B S_B S B be the Bergman projective connection, defined by
B ( x , y ) = d x d y ( x − y ) 2 + 1 6 S B ( y ) d x d y + O ( x − y ) B(x,y)=\frac{dx\,dy}{(x-y)^2}+\frac{1}{6}S_B(y)\,dx\,dy+O(x-y) B ( x , y ) = ( x − y ) 2 d x d y + 6 1 S B ( y ) d x d y + O ( x − y ) in a local coordinate. Then in every local coordinate
R ε = 12 c 1 = S B + 6 ∑ i , j c i j ζ i ζ j , R_{\varepsilon}=12c_1=S_B+6\sum_{i,j}c_{ij}\zeta_i\zeta_j, R ε = 12 c 1 = S B + 6 i , j ∑ c ij ζ i ζ j , with c i j c_{ij} c ij as in Proposition 3.17 . Moreover c i j = ∂ i ∂ j log θ [ δ ] ( 0 ) c_{ij}=\partial_i\partial_j\log\theta[\delta](0) c ij = ∂ i ∂ j log θ [ δ ] ( 0 ) , δ = δ ( ε ) \delta=\delta(\varepsilon) δ = δ ( ε ) , where ∂ i = ∂ / ∂ z i \partial_i=\partial/\partial z_i ∂ i = ∂ / ∂ z i and z i = ∫ ζ i z_i=\int\zeta_i z i = ∫ ζ i , at least as quadratic differentials, i.e., ∑ c i j ζ i ζ j = ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ζ j \sum c_{ij}\zeta_i\zeta_j=\sum\partial_i\partial_j\log\theta[\delta](0)\,\zeta_i\zeta_j ∑ c ij ζ i ζ j = ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ζ j .
Proof. By Proposition 3.7 (ii), S ε 2 = ( x − y ) − 2 + 2 c 1 ( y ) + O ( x − y ) S_{\varepsilon}^2=(x-y)^{-2}+2c_1(y)+O(x-y) S ε 2 = ( x − y ) − 2 + 2 c 1 ( y ) + O ( x − y ) (the cross term 2 c 0 / ( x − y ) 2c_0/(x-y) 2 c 0 / ( x − y ) vanishes). B B B has the expansion above. Thus ∑ c i j ζ i ( x ) ζ j ( y ) = 2 c 1 ( y ) − 1 6 S B ( y ) \sum c_{ij}\zeta_i(x)\zeta_j(y)=2c_1(y)-\frac{1}{6}S_B(y) ∑ c ij ζ i ( x ) ζ j ( y ) = 2 c 1 ( y ) − 6 1 S B ( y ) , which is the first formula. For the second, set ξ = x − y \xi=x-y ξ = x − y and use Proposition 3.16 (a). Since δ \delta δ is even, θ [ δ ] ( ∫ y x ζ ) / θ [ δ ] ( 0 ) = 1 + 1 2 ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ( y ) ζ j ( y ) ξ 2 + O ( ξ 3 ) \theta[\delta](\int_y^x\zeta)/\theta[\delta](0)=1+\frac{1}{2}\sum\partial_i\partial_j\log\theta[\delta](0)\zeta_i(y)\zeta_j(y)\xi^2+O(\xi^3) θ [ δ ] ( ∫ y x ζ ) / θ [ δ ] ( 0 ) = 1 + 2 1 ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ( y ) ζ j ( y ) ξ 2 + O ( ξ 3 ) , and E − 2 = ξ − 2 ( d x d y ) ( 1 − 2 e 2 ξ 2 + ⋯ ) E^{-2}=\xi^{-2}(dx\,dy)(1-2e_2\xi^2+\cdots) E − 2 = ξ − 2 ( d x d y ) ( 1 − 2 e 2 ξ 2 + ⋯ ) while B = ∂ x ∂ y log E = ξ − 2 − 2 e 2 + O ( ξ ) B=\partial_x\partial_y\log E=\xi^{-2}-2e_2+O(\xi) B = ∂ x ∂ y log E = ξ − 2 − 2 e 2 + O ( ξ ) , thus E − 2 − B = O ( ξ ) E^{-2}-B=O(\xi) E − 2 − B = O ( ξ ) . Hence, S ε 2 − B = E − 2 ( ( θ [ δ ] / θ [ δ ] ( 0 ) ) 2 − 1 ) + ( E − 2 − B ) = ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ( y ) ζ j ( y ) + O ( ξ ) S_{\varepsilon}^2-B=E^{-2}\left((\theta[\delta]/\theta[\delta](0))^2-1\right)+(E^{-2}-B)=\sum\partial_i\partial_j\log\theta[\delta](0)\zeta_i(y)\zeta_j(y)+O(\xi) S ε 2 − B = E − 2 ( ( θ [ δ ] / θ [ δ ] ( 0 ) ) 2 − 1 ) + ( E − 2 − B ) = ∑ ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ( y ) ζ j ( y ) + O ( ξ ) . □
Since S B S_B S B is a projective connection, this is consistent with Proposition 3.7 (iii). For g = 1 g=1 g = 1 it is checked in Proposition 3.19 (c) below.
Genus one
Let g = 1 g=1 g = 1 and let Γ ~ = ⟨ γ ~ ⟩ \widetilde{\Gamma}=\langle\widetilde{\gamma}\rangle Γ = ⟨ γ ⟩ with γ ~ \widetilde{\gamma} γ loxodromic of eigenvalues κ ± 1 \kappa^{\pm1} κ ± 1 , 0 < ∣ κ ∣ < 1 0<|\kappa|<1 0 < ∣ κ ∣ < 1 . This is the general form of a lifted genus one Schottky group. (D) is empty and (E) holds because ∥ γ ~ n ∥ ∼ ∣ κ ∣ − ∣ n ∣ \|\widetilde{\gamma}^n\|\sim|\kappa|^{-|n|} ∥ γ n ∥ ∼ ∣ κ ∣ − ∣ n ∣ . Set q = κ 2 q=\kappa^2 q = κ 2 and fix τ \tau τ with e 2 π i τ = q e^{2\pi i\tau}=q e 2 π i τ = q , Im τ > 0 \operatorname{Im}\tau>0 Im τ > 0 . By Proposition 3.14 we may conjugate so that γ ~ = diag ( κ , κ − 1 ) \widetilde{\gamma}=\operatorname{diag}(\kappa,\kappa^{-1}) γ = diag ( κ , κ − 1 ) , i.e., γ x = q x \gamma x=qx γ x = q x , Ω 0 = C × \Omega_0=\mathbb{C}^{\times} Ω 0 = C × , Σ = C × / q Z \Sigma=\mathbb{C}^{\times}/q^{\mathbb{Z}} Σ = C × / q Z . Then we have
Proposition 3.19. (a) For x , y ∈ C × x,y\in\mathbb{C}^{\times} x , y ∈ C × , y ∉ q Z x y\notin q^{\mathbb{Z}}x y ∈ / q Z x : S ε ( x , y ) = ∑ n ∈ Z κ n q n x − y S_{\varepsilon}(x,y)=\sum_{n\in\mathbb{Z}}\frac{\kappa^n}{q^n x-y} S ε ( x , y ) = ∑ n ∈ Z q n x − y κ n .
(b) Define δ ′ ′ ∈ { 0 , 1 2 } \delta^{\prime\prime} \in\left\{0,\frac{1}{2}\right\} δ ′′ ∈ { 0 , 2 1 } by κ = − e 2 π i δ ′ ′ e i π τ \kappa=-e^{2\pi i\delta^{\prime\prime}}e^{i\pi\tau} κ = − e 2 π i δ ′′ e iπ τ (note that ( κ e − i π τ ) 2 = 1 (\kappa e^{-i\pi\tau})^2=1 ( κ e − iπ τ ) 2 = 1 ). Then with x = e 2 π i u x=e^{2\pi i u} x = e 2 π i u , y = e 2 π i v y=e^{2\pi i v} y = e 2 π i v ,
s ε ( x , y ) = 1 2 π i e π i ( u + v ) θ [ 0 ; δ ′ ′ ] ( u − v ) θ 1 ′ ( 0 ) θ [ 0 ; δ ′ ′ ] ( 0 ) θ 1 ( u − v ) , θ 1 : = θ [ 1 2 ; 1 2 ] , ′ = d d u , s_{\varepsilon}(x,y)=\frac{1}{2\pi i e^{\pi i(u+v)}}\frac{\theta[0;\delta^{\prime\prime}](u-v)\theta_1'(0)}{\theta[0;\delta^{\prime\prime}](0)\theta_1(u-v)},\qquad\theta_1:=\theta\left[\frac{1}{2};\frac{1}{2}\right],\quad{}'=\frac{d}{du}, s ε ( x , y ) = 2 π i e π i ( u + v ) 1 θ [ 0 ; δ ′′ ] ( 0 ) θ 1 ( u − v ) θ [ 0 ; δ ′′ ] ( u − v ) θ 1 ′ ( 0 ) , θ 1 := θ [ 2 1 ; 2 1 ] , ′ = d u d ,
where the right side is independent of the branches of u , v u,v u , v .
(c) In the coordinate x x x , R ε ( y ) = y − 2 [ 1 2 − 1 4 π 2 ( 6 θ [ 0 ; δ ′ ′ ] ′ ′ ( 0 ) θ [ 0 ; δ ′ ′ ] ( 0 ) − 2 θ 1 ′ ′ ′ ( 0 ) θ 1 ′ ( 0 ) ) ] R_{\varepsilon}(y)=y^{-2}\left[\frac{1}{2}-\frac{1}{4\pi^2}\left(6\frac{\theta[0;\delta^{\prime\prime}]^{\prime\prime}(0)}{\theta[0;\delta^{\prime\prime}](0)}-2\frac{\theta_1^{\prime\prime\prime}(0)}{\theta_1'(0)}\right)\right] R ε ( y ) = y − 2 [ 2 1 − 4 π 2 1 ( 6 θ [ 0 ; δ ′′ ] ( 0 ) θ [ 0 ; δ ′′ ] ′′ ( 0 ) − 2 θ 1 ′ ( 0 ) θ 1 ′′′ ( 0 ) ) ] , and S B = − 2 θ 1 ′ ′ ( 0 ) / θ 1 ′ ( 0 ) S_B=-2\theta_1^{\prime\prime}(0)/\theta_1'(0) S B = − 2 θ 1 ′′ ( 0 ) / θ 1 ′ ( 0 ) in the coordinate u u u .
Note that it follows from (b) that Proposition 3.16 holds with δ ( ε ) = [ 0 ; δ ′ ′ ] \delta(\varepsilon)=[0;\delta^{\prime\prime}] δ ( ε ) = [ 0 ; δ ′′ ] , and κ ↦ − κ \kappa\mapsto-\kappa κ ↦ − κ (the other lift) exchanges δ ′ ′ = 0 \delta^{\prime\prime}=0 δ ′′ = 0 and δ ′ ′ = 1 2 \delta^{\prime\prime}=\frac{1}{2} δ ′′ = 2 1 . Replacing τ \tau τ by τ + 1 \tau+1 τ + 1 (another marking, β ↦ β + α \beta\mapsto\beta+\alpha β ↦ β + α ) exchanges the two values as well.
Proof. (a) ⟨ γ ~ n ( x , 1 ) , ( y , 1 ) ⟩ = κ n x − κ − n y \langle\widetilde{\gamma}^n(x,1),(y,1)\rangle=\kappa^n x-\kappa^{-n}y ⟨ γ n ( x , 1 ) , ( y , 1 )⟩ = κ n x − κ − n y .
(b) Let Φ ( x , y ) \Phi(x,y) Φ ( x , y ) be the right-hand side. Using θ [ 0 ; δ ′ ′ ] ( z + τ ) = e − π i τ − 2 π i ( z + δ ′ ′ ) θ [ 0 ; δ ′ ′ ] ( z ) \theta[0;\delta^{\prime\prime}](z+\tau)=e^{-\pi i\tau-2\pi i(z+\delta^{\prime\prime})}\theta[0;\delta^{\prime\prime}](z) θ [ 0 ; δ ′′ ] ( z + τ ) = e − π i τ − 2 π i ( z + δ ′′ ) θ [ 0 ; δ ′′ ] ( z ) , θ 1 ( z + τ ) = − e − π i τ − 2 π i z θ 1 ( z ) \theta_1(z+\tau)=-e^{-\pi i\tau-2\pi iz}\theta_1(z) θ 1 ( z + τ ) = − e − π i τ − 2 π i z θ 1 ( z ) , the one-periodicity of θ [ 0 ; δ ′ ′ ] \theta[0;\delta^{\prime\prime}] θ [ 0 ; δ ′′ ] and the sign change of θ 1 \theta_1 θ 1 under z ↦ z + 1 z\mapsto z+1 z ↦ z + 1 (compensated by e π i ( u + v ) ↦ − e π i ( u + v ) e^{\pi i(u+v)}\mapsto-e^{\pi i(u+v)} e π i ( u + v ) ↦ − e π i ( u + v ) ), one finds that Φ \Phi Φ is well defined on C × × C × \mathbb{C}^{\times}\times\mathbb{C}^{\times} C × × C × . Also, Φ ( q x , y ) = − e − 2 π i δ ′ ′ e − i π τ Φ ( x , y ) = κ − 1 Φ ( x , y ) \Phi(qx,y)=-e^{-2\pi i\delta^{\prime\prime}}e^{-i\pi\tau}\Phi(x,y)=\kappa^{-1}\Phi(x,y) Φ ( q x , y ) = − e − 2 π i δ ′′ e − iπ τ Φ ( x , y ) = κ − 1 Φ ( x , y ) and likewise Φ ( x , q y ) = κ − 1 Φ ( x , y ) \Phi(x,qy)=\kappa^{-1}\Phi(x,y) Φ ( x , q y ) = κ − 1 Φ ( x , y ) , which is the gluing relation (5) (j ( γ ~ , x ) = κ j(\widetilde{\gamma},x)=\kappa j ( γ , x ) = κ ). Φ \Phi Φ has simple poles on y ∈ q Z x y\in q^{\mathbb{Z}}x y ∈ q Z x , and near y = x y=x y = x , θ 1 ( u − v ) / θ 1 ′ ( 0 ) = ( x − y ) / ( 2 π i y ) ( 1 + O ( x − y ) ) \theta_1(u-v)/\theta_1'(0)=(x-y)/(2\pi iy)(1+O(x-y)) θ 1 ( u − v ) / θ 1 ′ ( 0 ) = ( x − y ) / ( 2 π i y ) ( 1 + O ( x − y )) . Thus, Φ = 1 x − y + O ( 1 ) \Phi=\frac{1}{x-y}+O(1) Φ = x − y 1 + O ( 1 ) . Now fix y y y . Then D ( x ) : = s ε ( x , y ) − Φ ( x , y ) D(x):=s_{\varepsilon}(x,y)-\Phi(x,y) D ( x ) := s ε ( x , y ) − Φ ( x , y ) has no pole at x = y x=y x = y and, by the common automorphy law, none at q n x q^n x q n x . Thus, D D D is holomorphic on C × \mathbb{C}^{\times} C × with D ( q x ) = κ − 1 D ( x ) D(qx)=\kappa^{-1}D(x) D ( q x ) = κ − 1 D ( x ) . Writing D = ∑ a m x m D=\sum a_mx^m D = ∑ a m x m gives a m ( κ 2 m + 1 − 1 ) = 0 a_m(\kappa^{2m+1}-1)=0 a m ( κ 2 m + 1 − 1 ) = 0 , and κ 2 m + 1 ≠ 1 \kappa^{2m+1}\ne1 κ 2 m + 1 = 1 for every integer m m m (0 < ∣ κ ∣ < 1 0<|\kappa|<1 0 < ∣ κ ∣ < 1 ), thus D = 0 D=0 D = 0 .
(c) From (b), in the coordinate u u u : c 1 ( u ) = 1 2 θ [ 0 ; δ ′ ′ ] ′ ′ ( 0 ) θ [ 0 ; δ ′ ′ ] ( 0 ) − θ 1 ′ ′ ′ ( 0 ) 6 θ 1 ′ ( 0 ) c_1^{(u)}=\frac{1}{2}\frac{\theta[0;\delta^{\prime\prime}]^{\prime\prime}(0)}{\theta[0;\delta^{\prime\prime}](0)}-\frac{\theta_1^{\prime\prime\prime}(0)}{6\theta_1'(0)} c 1 ( u ) = 2 1 θ [ 0 ; δ ′′ ] ( 0 ) θ [ 0 ; δ ′′ ] ′′ ( 0 ) − 6 θ 1 ′ ( 0 ) θ 1 ′′′ ( 0 ) and R ( u ) = 12 c 1 ( u ) R^{(u)}=12c_1^{(u)} R ( u ) = 12 c 1 ( u ) . B = − ∂ u 2 log θ 1 ( u − v ) B=-\partial_u^2\log\theta_1(u-v) B = − ∂ u 2 log θ 1 ( u − v ) gives S B ( u ) = − 2 θ 1 ′ ′ ( 0 ) / θ 1 ′ ( 0 ) S_B^{(u)}=-2\theta_1^{\prime\prime}(0)/\theta_1'(0) S B ( u ) = − 2 θ 1 ′′ ( 0 ) / θ 1 ′ ( 0 ) . For u = log x / ( 2 π i ) u=\log x/(2\pi i) u = log x / ( 2 π i ) , u ′ = 1 / ( 2 π i x ) u'=1/(2\pi ix) u ′ = 1/ ( 2 π i x ) and { u ; x } = 1 2 x 2 \{u;x\}=\frac{1}{2x^2} { u ; x } = 2 x 2 1 , thus Proposition 3.7 (iii) gives R ( x ) = − 1 4 π 2 x 2 R ( u ) + 1 2 x 2 R^{(x)}=-\frac{1}{4\pi^2x^2}R^{(u)}+\frac{1}{2x^2} R ( x ) = − 4 π 2 x 2 1 R ( u ) + 2 x 2 1 , which is the required formula. □
In particular, the genus one kernel for δ ′ ′ = 0 \delta^{\prime\prime}=0 δ ′′ = 0 and δ ′ ′ = 1 2 \delta^{\prime\prime}=\frac{1}{2} δ ′′ = 2 1 is the classical torus Szegő kernel of the two even characteristics with trivial α \alpha α -holonomy. The other two characteristics of the torus are not in the family.
Remark 3.20. For g ≥ 2 g\ge2 g ≥ 2 we have the following. Let κ a \kappa_a κ a be the eigenvalue of γ ~ a \widetilde{\gamma}_a γ a of modulus < 1 <1 < 1 . Then κ a = − λ a / ( w a − w − a ) ( 1 + O ( r / d ) ) \kappa_a=-\lambda_a/(w_a-w_{-a})(1+O(r/d)) κ a = − λ a / ( w a − w − a ) ( 1 + O ( r / d )) . At genus one, (b) gives κ = − e 2 π i δ ′ ′ e i π τ \kappa=-e^{2\pi i\delta^{\prime\prime}}e^{i\pi\tau} κ = − e 2 π i δ ′′ e iπ τ . For g ≥ 2 g\ge2 g ≥ 2 we expect, for the marking of § 2 , e 2 π i δ a ′ ′ = − κ a e − i π τ a a ( 1 + O ( r / d ) ) e^{2\pi i\delta_a^{\prime\prime}}=-\kappa_a e^{-i\pi\tau_{aa}}(1+O(r/d)) e 2 π i δ a ′′ = − κ a e − iπ τ aa ( 1 + O ( r / d )) , where the left side is ± 1 \pm1 ± 1 . The only evidence we have is a formal one: the part of the first-order term of Theorem 5.1 that is odd in λ a \lambda_a λ a , namely λ a ( 1 ( x − w a ) ( w − a − y ) − 1 ( x − w − a ) ( w a − y ) ) = λ a ( x − y ) ( w a − w − a ) ( x − w a ) ( x − w − a ) ( y − w a ) ( y − w − a ) \lambda_a\left(\frac{1}{(x-w_a)(w_{-a}-y)}-\frac{1}{(x-w_{-a})(w_a-y)}\right)=\lambda_a\frac{(x-y)(w_a-w_{-a})}{(x-w_a)(x-w_{-a})(y-w_a)(y-w_{-a})} λ a ( ( x − w a ) ( w − a − y ) 1 − ( x − w − a ) ( w a − y ) 1 ) = λ a ( x − w a ) ( x − w − a ) ( y − w a ) ( y − w − a ) ( x − y ) ( w a − w − a ) , agrees with the term of θ [ δ ] ( ∫ y x ζ ) / ( θ [ δ ] ( 0 ) E ) \theta[\delta]\left(\int_y^x\zeta\right)/(\theta[\delta](0)E) θ [ δ ] ( ∫ y x ζ ) / ( θ [ δ ] ( 0 ) E ) odd in e 2 π i δ a ′ ′ e^{2\pi i\delta_a^{\prime\prime}} e 2 π i δ a ′′ to leading order, e 2 π i δ a ′ ′ e i π τ a a ( w a − w − a ) 2 ( x − y ) / ( ( x − w a ) ( x − w − a ) ( y − w a ) ( y − w − a ) ) e^{2\pi i\delta_a^{\prime\prime}}e^{i\pi\tau_{aa}}(w_a-w_{-a})^2(x-y)/((x-w_a)(x-w_{-a})(y-w_a)(y-w_{-a})) e 2 π i δ a ′′ e iπ τ aa ( w a − w − a ) 2 ( x − y ) / (( x − w a ) ( x − w − a ) ( y − w a ) ( y − w − a )) , if λ a = e 2 π i δ a ′ ′ e i π τ a a ( w a − w − a ) \lambda_a=e^{2\pi i\delta_a^{\prime\prime}}e^{i\pi\tau_{aa}}(w_a-w_{-a}) λ a = e 2 π i δ a ′′ e iπ τ aa ( w a − w − a ) , which is the claimed relation.
Reduction formulas
The theorem below describes the residue reconstruction. Assume the hypotheses of Theorem 3.15 (in particular, (H)). Let H H H be a meromorphic section of L ε L_{\varepsilon} L ε on Σ \Sigma Σ , with poles of arbitrary (finite) order at a finite set of distinct points p 1 , … , p n p_1,\ldots,p_n p 1 , … , p n of Σ \Sigma Σ (not of Ω 0 \Omega_0 Ω 0 : each p i p_i p i denotes a single point of the surface, not a choice of lift), and let x ∈ Σ ∖ { p 1 , … , p n } x\in\Sigma\setminus\{p_1,\ldots,p_n\} x ∈ Σ ∖ { p 1 , … , p n } .
Theorem 4.1. Then
H ( x ) = ∑ i = 1 n Res y = p i ( S ε ( x , y ) H ( y ) ) , H(x)=\sum_{i=1}^{n}\operatorname{Res}_{y=p_i}\left(S_{\varepsilon}(x,y)H(y)\right), H ( x ) = i = 1 ∑ n Res y = p i ( S ε ( x , y ) H ( y ) ) ,
an identity of sections of L ε L_{\varepsilon} L ε in the variable x x x . For fixed x x x , S ε ( x , y ) H ( y ) S_{\varepsilon}(x,y)H(y) S ε ( x , y ) H ( y ) has ordinary scalar residues at each p i p_i p i .
Note that the formula above represents a product of two sections of L ε L_{\varepsilon} L ε in y y y . Hence, it is a meromorphic 1-form on Σ \Sigma Σ in y y y , by Proposition 3.6 . The right-hand side, as x x x varies, assembles into a section of L ε L_{\varepsilon} L ε because S ε ( x , ⋅ ) S_{\varepsilon}(x,\cdot) S ε ( x , ⋅ ) is.
Proof. By Proposition 3.6 the automorphy factors j ( μ ~ , y ) j(\widetilde{\mu},y) j ( μ , y ) of S ε S_{\varepsilon} S ε and of H H H cancel in the product. Thus, η ( y ) : = S ε ( x , y ) H ( y ) \eta(y):=S_{\varepsilon}(x,y)H(y) η ( y ) := S ε ( x , y ) H ( y ) descends to a well-defined meromorphic 1-form on Σ \Sigma Σ , with poles only at y = x y=x y = x and y = p i y=p_i y = p i (i = 1 , … , n i=1,\ldots,n i = 1 , … , n ). These are finitely many points, since H H H has finitely many poles by hypothesis and x x x is one further point. In a local coordinate near y = x y=x y = x , writing H ( y ) = h ( y ) ( d y ) 1 / 2 H(y)=h(y)(\mathrm{d}y)^{1/2} H ( y ) = h ( y ) ( d y ) 1/2 and S ε ( x , y ) = s ε ( x , y ) ( d x d y ) 1 / 2 S_\varepsilon(x,y)=s_\varepsilon(x,y)(\mathrm{d}x\,\mathrm{d}y)^{1/2} S ε ( x , y ) = s ε ( x , y ) ( d x d y ) 1/2 , η ( y ) = s ε ( x , y ) h ( y ) d y ( d x ) 1 / 2 = ( 1 x − y + O ( x − y ) ) h ( y ) d y ( d x ) 1 / 2 \eta(y)=s_\varepsilon(x,y)h(y)\,\mathrm{d}y\,(\mathrm{d}x)^{1/2}=\left(\frac{1}{x-y}+O(x-y)\right)h(y)\,\mathrm{d}y\,(\mathrm{d}x)^{1/2} η ( y ) = s ε ( x , y ) h ( y ) d y ( d x ) 1/2 = ( x − y 1 + O ( x − y ) ) h ( y ) d y ( d x ) 1/2 (Proposition 3.7 (ii)). Thus, using skew symmetry of 1 x − y \frac{1}{x-y} x − y 1 , we obtain Res y = x η = − h ( x ) ( d x ) 1 / 2 = − H ( x ) \operatorname{Res}_{y=x}\eta=-h(x)(\mathrm{d}x)^{1/2}=-H(x) Res y = x η = − h ( x ) ( d x ) 1/2 = − H ( x ) . The residue theorem on the compact surface Σ \Sigma Σ (sum of all residues of a meromorphic 1-form is 0) gives − H ( x ) + ∑ i Res y = p i η = 0 -H(x)+\sum_i\operatorname{Res}_{y=p_i}\eta=0 − H ( x ) + ∑ i Res y = p i η = 0 , i.e., the required claim. □
Remark 4.2. The proof of Theorem 4.1 applies the residue theorem on the compact surface Σ \Sigma Σ to the meromorphic 1-form y ↦ S ε ( x , y ) ϕ ( y ) y\mapsto S_\varepsilon(x,y)\,\phi(y) y ↦ S ε ( x , y ) ϕ ( y ) . It is a section of L ε ⊗ L ε − 1 ⊗ K = K L_\varepsilon\otimes L_\varepsilon^{-1}\otimes K=K L ε ⊗ L ε − 1 ⊗ K = K , hence a meromorphic 1-form for ϕ \phi ϕ a section of L ε L_\varepsilon L ε . Its total residue is zero, and its poles are the pole y = x y=x y = x of S ε ( x , ⋅ ) S_\varepsilon(x,\cdot) S ε ( x , ⋅ ) and the poles of ϕ \phi ϕ . There are no others since S ε ( x , ⋅ ) S_\varepsilon(x,\cdot) S ε ( x , ⋅ ) is holomorphic off x x x (see Theorem 3.5 ). In a chart the sum of residues is computed on small circles about these points. Thus, no contour at ∞ \infty ∞ contributes: ∞ \infty ∞ is an ordinary point of Σ \Sigma Σ and the homogeneous formulation (eq. (2) ) has no singularity there. Corollary 4.3 uses the same residue normalization: Res y = x j \operatorname{Res}_{y=x_j} Res y = x j of a 1-form is chart-independent, and the half weight factors transform with the same j j j -multiplier on both sides.
In the coordinate form we have
Corollary 4.3. With H = h ( z ) ( d z ) 1 / 2 H=h(z)(\mathrm{d}z)^{1/2} H = h ( z ) ( d z ) 1/2 , c m ( i ) : = Res y = p i ( y − p i ) m h ( y ) d y c_m^{(i)}:=\operatorname{Res}_{y=p_i}(y-p_i)^m h(y)\,\mathrm{d}y c m ( i ) := Res y = p i ( y − p i ) m h ( y ) d y (thus h = ∑ m c m ( i ) ( y − p i ) − m − 1 + regular part ) h=\sum_m c_m^{(i)}(y-p_i)^{-m-1}+\text{regular part}) h = ∑ m c m ( i ) ( y − p i ) − m − 1 + regular part ) and ∂ y ( m ) : = 1 m ! ∂ y m \partial_y^{(m)}:=\frac{1}{m!}\partial_y^m ∂ y ( m ) := m ! 1 ∂ y m :
h ( x ) = ∑ i ∑ m = 0 M i c m ( i ) ∂ y ( m ) s ε ( x , y ) ∣ y = p i , h(x)=\sum_i\sum_{m=0}^{M_i}c_m^{(i)}\left.\partial_y^{(m)}s_\varepsilon(x,y)\right|_{y=p_i}, h ( x ) = i ∑ m = 0 ∑ M i c m ( i ) ∂ y ( m ) s ε ( x , y ) y = p i ,
where M i M_i M i is pole order of H H H at p i p_i p i , finite by hypothesis.
The coefficients c m ( i ) c_m^{(i)} c m ( i ) depend on the coordinate. Theorem 4.1 is the invariant statement. There is no holomorphic remainder because h 0 ( L ε ) = 0 h^0(L_\varepsilon)=0 h 0 ( L ε ) = 0 (see Theorem 3.10 (i)). The notation ∂ y ( m ) : = 1 m ! ∂ y m \partial_y^{(m)}:=\frac{1}{m!}\partial_y^m ∂ y ( m ) := m ! 1 ∂ y m is for an ordinary derivative in y y y (not an exterior/Lie derivative).
VOSA correlation functions
Conventions: V = V ‾ 0 ⊕ V ‾ 1 V=\overline{V}_0\oplus\overline{V}_1 V = V 0 ⊕ V 1 is a Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z -graded VOSA, ∣ v ∣ ∈ Z / 2 Z |v|\in\mathbb{Z}/2\mathbb{Z} ∣ v ∣ ∈ Z /2 Z for v ∈ V ∣ v ∣ v\in V_{|v|} v ∈ V ∣ v ∣ homogeneous. A vertex operator is Y ( v , x ) = ∑ n v ( n ) x − n − 1 Y(v,x)=\sum_n v(n)x^{-n-1} Y ( v , x ) = ∑ n v ( n ) x − n − 1 . ψ ∈ V 1 / 2 ∩ V ‾ 1 \psi\in V_{1/2}\cap\overline{V}_1 ψ ∈ V 1/2 ∩ V 1 is an odd primary vector with truncation, ψ ( n ) v = 0 \psi(n)v=0 ψ ( n ) v = 0 for n ≫ 0 n\gg0 n ≫ 0 for each v v v . Locality/OPE: ψ ( x ) Y ( v , y ) = ∑ n ≥ 0 ( x − y ) − n − 1 Y ( ψ ( n ) v , y ) + regular part \psi(x)Y(v,y)=\sum_{n\geq0}(x-y)^{-n-1}Y(\psi(n)v,y)+\text{regular part} ψ ( x ) Y ( v , y ) = ∑ n ≥ 0 ( x − y ) − n − 1 Y ( ψ ( n ) v , y ) + regular part , a finite polar sum. The sum truncates by the assumption on ψ ( n ) v \psi(n)v ψ ( n ) v . We assume also the following hypotheses.
Hypothesis (A).
For v 1 , … , v n ∈ V v_1,\ldots,v_n\in V v 1 , … , v n ∈ V homogeneous and distinct points y 1 , … , y n ∈ Σ y_1,\ldots,y_n\in\Sigma y 1 , … , y n ∈ Σ , the genus g g g correlation function H ( x ) : = ⟨ ψ ( x ) Y ( v 1 , y 1 ) ⋯ Y ( v n , y n ) ⟩ Σ , ε H(x):=\langle\psi(x)Y(v_1,y_1)\cdots Y(v_n,y_n)\rangle_{\Sigma,\varepsilon} H ( x ) := ⟨ ψ ( x ) Y ( v 1 , y 1 ) ⋯ Y ( v n , y n ) ⟩ Σ , ε (i.e., the unnormalized correlation function: no division by the genus g g g partition function Z g Z_g Z g is performed, thus Z g ≠ 0 Z_g\neq0 Z g = 0 is not assumed; all conditions below are linear in H H H , hence they hold for H H H if and only if they hold for H / Z g H/Z_g H / Z g whenever Z g ≠ 0 Z_g\neq0 Z g = 0 ) is a meromorphic section of L ε L_\varepsilon L ε in x x x , holomorphic away from { y 1 , … , y n } \{y_1,\ldots,y_n\} { y 1 , … , y n } , and at each y k y_k y k its principal part is the one dictated by the OPE above. In a local coordinate near y k y_k y k , we have
H ( x ) = ∑ m ≥ 0 ( x − y k ) − m − 1 ( − 1 ) ∑ l < k ∣ v l ∣ ⟨ Y ( v 1 , y 1 ) ⋯ Y ( ψ ( m ) v k , y k ) ⋯ Y ( v n , y n ) ⟩ + O ( ( x − y k ) 0 ) . H(x)=\sum_{m\geq0}(x-y_k)^{-m-1}(-1)^{\sum_{l<k}|v_l|}\langle Y(v_1,y_1)\cdots Y(\psi(m)v_k,y_k)\cdots Y(v_n,y_n)\rangle+O((x-y_k)^0). H ( x ) = m ≥ 0 ∑ ( x − y k ) − m − 1 ( − 1 ) ∑ l < k ∣ v l ∣ ⟨ Y ( v 1 , y 1 ) ⋯ Y ( ψ ( m ) v k , y k ) ⋯ Y ( v n , y n )⟩ + O (( x − y k ) 0 ) .
The sign ( − 1 ) ∑ l < k ∣ v l ∣ (-1)^{\sum_{l<k}|v_l|} ( − 1 ) ∑ l < k ∣ v l ∣ arising from anticommuting the odd field ψ ( x ) \psi(x) ψ ( x ) past Y ( v l , y l ) Y(v_l,y_l) Y ( v l , y l ) , l < k l<k l < k (Koszul sign rule for Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z -graded vertex operators; no further sign occurs from the ordering of the modes ψ ( m ) \psi(m) ψ ( m ) , as these act on the single vector v k v_k v k ). Under condition (A), the Fermionic Zhu-type reduction is given by Corollary 4.4.
Corollary 4.4.
⟨ ψ ( x ) ∏ k Y ( v k , y k ) ⟩ = ∑ k = 1 n ( − 1 ) ∑ l < k ∣ v l ∣ ∑ m ≥ 0 ∂ y ( m ) s ε ( x , y ) ∣ y = y k ⟨ ⋯ Y ( ψ ( m ) v k , y k ) ⋯ ⟩ . \left\langle\psi(x)\prod_kY(v_k,y_k)\right\rangle=\sum_{k=1}^{n}(-1)^{\sum_{l<k}|v_l|}\sum_{m\geq0}\left.\partial_y^{(m)}s_\varepsilon(x,y)\right|_{y=y_k}\left\langle\cdots Y(\psi(m)v_k,y_k)\cdots\right\rangle. ⟨ ψ ( x ) k ∏ Y ( v k , y k ) ⟩ = k = 1 ∑ n ( − 1 ) ∑ l < k ∣ v l ∣ m ≥ 0 ∑ ∂ y ( m ) s ε ( x , y ) y = y k ⟨ ⋯ Y ( ψ ( m ) v k , y k ) ⋯ ⟩ .
Proof. Theorem 4.1 and Corollary 4.3, with c m ( k ) c_m^{(k)} c m ( k ) read from the OPE and the Koszul sign from moving ψ \psi ψ past the insertions v l v_l v l , l < k l<k l < k . □
Hypothesis (Af f _{\mathrm{ff}} ff ).
For the free fermion VOSA ( ψ ( 0 ) ψ = 1 , ψ ( m ) ψ = 0 (\psi_{(0)}\psi=1,\ \psi_{(m)}\psi=0 ( ψ ( 0 ) ψ = 1 , ψ ( m ) ψ = 0 for m ≥ 1 m\geq1 m ≥ 1 ): for every m ≥ 0 m\geq0 m ≥ 0 the unnormalized correlator F m ( x 1 , … , x m ) = ⟨ ψ ( x 1 ) ⋯ ψ ( x m ) ⟩ Σ , ε F_m(x_1,\ldots,x_m)=\langle\psi(x_1)\cdots\psi(x_m)\rangle_{\Sigma,\varepsilon} F m ( x 1 , … , x m ) = ⟨ ψ ( x 1 ) ⋯ ψ ( x m ) ⟩ Σ , ε is a meromorphic section of L ε ⊠ m L_\varepsilon^{\boxtimes m} L ε ⊠ m on Σ m \Sigma^m Σ m , holomorphic away from the diagonals { x i = x j } \{x_i=x_j\} { x i = x j } , with at most simple poles along the diagonals, so that in each variable x i x_i x i it is a meromorphic section of L ε L_\varepsilon L ε on Σ \Sigma Σ , regular except at the points x j x_j x j (j ≠ i j \ne i j = i ); and (in x 1 x_1 x 1 ) its principal part ( − 1 ) j x 1 − x j F m − 2 ( x ^ 1 , x ^ j ) \frac{(-1)^j}{x_1-x_j}F_{m-2}(\hat{x}_1,\hat{x}_j) x 1 − x j ( − 1 ) j F m − 2 ( x ^ 1 , x ^ j ) at x j x_j x j ; F 0 = Z g F_0=Z_g F 0 = Z g is a constant, the genus g g g partition function, which may be any complex number (no division by Z g Z_g Z g , and no assumption Z g ≠ 0 Z_g \ne0 Z g = 0 ).
Proposition 4.5. (1) Assume the hypotheses of Theorem 3.15. For each constant F 0 = Z F_0=Z F 0 = Z there is one family { F m } m ≥ 0 \{F_m\}_{m\geq0} { F m } m ≥ 0 satisfying ( A f f ) (A_{\mathrm{ff}}) ( A ff ) , namely F m = 0 F_m=0 F m = 0 for m m m odd and F 2 n = Z ⋅ Pf [ S ε ( x i , x j ) ] i < j F_{2n}=Z\cdot\operatorname{Pf}[S_\varepsilon(x_i,x_j)]_{i<j} F 2 n = Z ⋅ Pf [ S ε ( x i , x j ) ] i < j .
(2) If, in addition, ( A f f ) (A_{\mathrm{ff}}) ( A ff ) holds for the correlation functions F m = ⟨ ψ ( x 1 ) ⋯ ψ ( x m ) ⟩ Σ , ε F_m=\langle\psi(x_1)\cdots\psi(x_m)\rangle_{\Sigma,\varepsilon} F m = ⟨ ψ ( x 1 ) ⋯ ψ ( x m ) ⟩ Σ , ε of the free fermion VOSA then those correlation functions equal Z Z Z multiplied by the Pfaffians above, where Z Z Z is the partition function (normalized correlators are obtained by dividing by Z Z Z if Z ≠ 0 Z\ne0 Z = 0 ).
Note that (1) is a statement purely about meromorphic sections of L ε L_\varepsilon L ε satisfying the formulated analytic conditions with no reference to the genus g g g correlation functions for an actual free fermion VOSA satisfying ( A f f ) (A_{\mathrm{ff}}) ( A ff ) .
Proof. Suppose { F m } \{F_m\} { F m } satisfies ( A f f ) (A_{\mathrm{ff}}) ( A ff ) . We show by induction on m m m that it must be Z Z Z multiplied by the Pfaffian family, and separately that this family does satisfy ( A f f ) (A_{\mathrm{ff}}) ( A ff ) .
For odd $m: assume F k = 0 F_k=0 F k = 0 for all odd k < m k<m k < m . In the variable x 1 x_1 x 1 , poles of F m F_m F m can occur only at the x j x_j x j , with principal parts proportional to F m − 2 F_{m-2} F m − 2 , which is odd and hence 0. Thus, F m F_m F m is holomorphic in x 1 x_1 x 1 , i.e., a section of H 0 ( L ε ) = 0 H^0(L_\varepsilon)=0 H 0 ( L ε ) = 0 (Theorem 3.10 (i)), and F m = 0 F_m=0 F m = 0 . Note that for m = 1 m=1 m = 1 there are no poles.
For even $m: by Theorem 4.1 in x 1 x_1 x 1 ,
F m = ∑ j ≥ 2 Res y = x j S ε ( x 1 , y ) F m ( y , … ) = ∑ j ≥ 2 ( − 1 ) j S ε ( x 1 , x j ) F m − 2 ( x ^ 1 , x ^ j ) , F_m=\sum_{j\geq2}\operatorname{Res}_{y=x_j}S_\varepsilon(x_1,y)F_m(y,\ldots)=\sum_{j\geq2}(-1)^jS_\varepsilon(x_1,x_j)F_{m-2}(\hat{x}_1,\hat{x}_j), F m = j ≥ 2 ∑ Res y = x j S ε ( x 1 , y ) F m ( y , … ) = j ≥ 2 ∑ ( − 1 ) j S ε ( x 1 , x j ) F m − 2 ( x ^ 1 , x ^ j ) ,
which is Z Z Z multiplied by the expansion of Pf [ S ε ( x i , x j ) ] \operatorname{Pf}[S_\varepsilon(x_i,x_j)] Pf [ S ε ( x i , x j )] along its first row, given F m − 2 = Z Pf F_{m-2}=Z\operatorname{Pf} F m − 2 = Z Pf . For m = 2 m=2 m = 2 this gives F 2 = Z S ε ( x 1 , x 2 ) F_2=ZS_\varepsilon(x_1,x_2) F 2 = Z S ε ( x 1 , x 2 ) . Conversely, Z Pf Z\operatorname{Pf} Z Pf is skew symmetric and, by first-row expansion and Proposition 3.7 (i), meromorphic with the simple poles and principal parts in every variable. □
Free fermion correlation functions are expressed through Szegő kernels for specific sewing constructions of genus one and two in [18 , 19 , 20 , 21 ] . The Szegő kernel on a genus two surface obtained by sewing two tori or by self-sewing a torus. The partition function and n n n -point generating functions of the rank-two free fermion VOSA as determinants of such kernels, in the sewing parameters and with the normalization of the partition function used there. Those constructions and normalizations are not the Schottky construction of this paper. We do not claim here identification with those schemes. Whether the Schottky-sewn correlators of the free fermion VOSA satisfy ( A f f ) (A_{\mathrm{ff}}) ( A ff ) is an open question at every genus g ≥ 2 g\geq2 g ≥ 2 . Proposition 4.5 is therefore depending on condition ( A f f ) (A_{\mathrm{ff}}) ( A ff ) .
Conjecture 4.6. For an arbitrary VOSA V V V , hypothesis (A) holds at every genus g ≥ 2 g\geq2 g ≥ 2 on (H), and Corollary 4.4 is the odd part of a genus g g g Zhu recursion extending [17 ] from VOAs to VOSAs (see Remark 4.7).
Remark 4.7. Let us recall the Schottky recursion procedure in Tuite-Welby. The genus g g g recursion in the Schottky scheme is developed by Tuite and Welby [17 ] , for vertex operator algebras (even VOAs, with their modules). They derive a Zhu-type recursion for n n n -point correlation functions on a genus g g g surface formally sewn from a sphere through Schottky parameters, with the Bers quasiforms of [16 ] (here, the forms Ψ N \Psi_N Ψ N , N ≥ 2 N\geq2 N ≥ 2 , of integer weight) as kernels. As far as we can verify, [17 ] does not treat vertex operator superalgebras, half-integral weights, spin structures, Szegő kernels or the case N = 1 2 N=\frac{1}{2} N = 2 1 , and the Schottky sewing there is formal (it is not shown to converge in the sense of Theorem 3.5 ). Thus [17 ] supplies the VOA model for Conjecture 4.6 and for Corollary 4.4, but the VOSA statements here are new and depending on conditions (A), ( A f f ) (A_{\mathrm{ff}}) ( A ff ) .
Expansion in the Schottky multipliers
For p ∈ P p\in\mathfrak{P} p ∈ P set ∣ ρ ∣ max : = max a ∈ A ∣ ρ a ∣ = r ( p ) 2 |\rho|_{\max}:=\max_{a\in\mathcal{A}}|\rho_a|=r(p)^2 ∣ ρ ∣ m a x := max a ∈ A ∣ ρ a ∣ = r ( p ) 2 and K d , p : = { z ∈ C : ∣ z − w α ∣ ≥ d ( p ) / 2 for all α } \mathcal{K}_{d,p}:=\{z\in\mathbb{C}:|z-w_\alpha|\geq d(p)/2\text{ for all }\alpha\} K d , p := { z ∈ C : ∣ z − w α ∣ ≥ d ( p ) /2 for all α } . Since d / 2 ≥ r c / 2 d/2\geq r_c/2 d /2 ≥ r c /2 under (H), K d , p ⊂ K p \mathcal{K}_{d,p}\subset\mathcal{K}_p K d , p ⊂ K p . The total space over which uniformity is claimed is G d : = { ( p , x , y ) : p ∈ P , x , y ∈ K d , p } \mathcal{G}_d:=\{(p,x,y):p\in\mathfrak{P},\ x,y\in\mathcal{K}_{d,p}\} G d := {( p , x , y ) : p ∈ P , x , y ∈ K d , p } . Expansion in λ \lambda λ is given by the
Theorem 5.1. Assume g ≥ 2 g\geq2 g ≥ 2 and (H). For every p = ( w , λ ) ∈ P p=(w,\lambda)\in\mathfrak{P} p = ( w , λ ) ∈ P and all x , y ∈ K d , p x,y\in\mathcal{K}_{d,p} x , y ∈ K d , p ,
S ε ( x , y ) = 1 x − y + ∑ a ∈ I + λ a [ 1 ( x − w a ) ( w a − y ) + ρ a − 1 ( x − w a ) ( w a − y ) + ρ a ] + R ( x , y ) , S_\varepsilon(x,y)=\frac{1}{x-y}+\sum_{a\in I_+}\lambda_a\left[\frac{1}{(x-w_a)(w_a-y)+\rho_a}-\frac{1}{(x-w_a)(w_a-y)+\rho_a}\right]+R(x,y), S ε ( x , y ) = x − y 1 + a ∈ I + ∑ λ a [ ( x − w a ) ( w a − y ) + ρ a 1 − ( x − w a ) ( w a − y ) + ρ a 1 ] + R ( x , y ) ,
where
∣ R ( x , y ) ∣ ≤ 64 3 ⋅ 144 143 ⋅ g ( 2 g − 1 ) ∣ ρ ∣ max d 3 ≤ 22 g ( 2 g − 1 ) ∣ ρ ∣ max d 3 . (6) |R(x,y)| \le\frac{64}{3}\cdot\frac{144}{143}\cdot\frac{g(2g-1)|\rho|_{\max}}{d^3}\le\frac{22g(2g-1)|\rho|_{\max}}{d^3}. \tag*{(6)} ∣ R ( x , y ) ∣ ≤ 3 64 ⋅ 143 144 ⋅ d 3 g ( 2 g − 1 ) ∣ ρ ∣ m a x ≤ d 3 22 g ( 2 g − 1 ) ∣ ρ ∣ m a x . (6)
The constant is absolute. The bound depends on the data only through g g g , d d d and ∣ ρ ∣ max |\rho|_{\max} ∣ ρ ∣ m a x , and holds on all of G d \mathbb{G}_d G d . The bracketed terms are exact (given by the contribution of the words of length one). R ( x , y ) = O ( ∣ λ ∣ max 2 ) R(x,y)=O(|\lambda|_{\max}^2) R ( x , y ) = O ( ∣ λ ∣ m a x 2 ) as ∣ λ ∣ max : = max a ∣ λ a ∣ → 0 |\lambda|_{\max}:=\max_a|\lambda_a|\to0 ∣ λ ∣ m a x := max a ∣ λ a ∣ → 0 with g g g and d d d held fixed, so that (H) continues to hold along the limit. s ε s_\varepsilon s ε and R R R are holomorphic in ( p , x , y ) (p,x,y) ( p , x , y ) on the interior of G d \mathbb{G}_d G d by Theorem 3.5 (i).
Proof. Length-one words are γ ~ c \widetilde{\gamma}_c γ c , c ∈ I c\in\mathcal{I} c ∈ I . By (1) and γ a x − y = ( w − a − y ) ( x − w a ) + ρ a x − w a \gamma_a x-y=\frac{(w_{-a}-y)(x-w_a)+\rho_a}{x-w_a} γ a x − y = x − w a ( w − a − y ) ( x − w a ) + ρ a , the c = a c=a c = a term is λ a ( x − w a ) ( w − a − y ) + ρ a \frac{\lambda_a}{(x-w_a)(w_{-a}-y)+\rho_a} ( x − w a ) ( w − a − y ) + ρ a λ a , and the c = − a c=-a c = − a term (with λ − a = − λ a \lambda_{-a}=-\lambda_a λ − a = − λ a ) is the negative of the second bracket entry. Their sum is the bracket sum in the statement, and R ( x , y ) : = s ε ( x , y ) − 1 x − y − [ bracket sum ] = ∑ n ≥ 2 [ sum of the length- n terms ] R(x,y):=s_\varepsilon(x,y)-\frac{1}{x-y}-[\text{bracket sum}]=\sum_{n\ge2}[\text{sum of the length-}n\text{ terms}] R ( x , y ) := s ε ( x , y ) − x − y 1 − [ bracket sum ] = ∑ n ≥ 2 [ sum of the length- n terms ] .
Bound on K d , p K_{d,p} K d , p . Let x ∈ K d , p x\in K_{d,p} x ∈ K d , p . In the proof of Lemma 3.2 the outermost factor satisfies ∣ j ( γ ~ c n , x ) ∣ = r c n ∣ x − w c n ∣ ≤ 2 r c n d ≤ ϖ |j(\widetilde{\gamma}_{c_n},x)|=\frac{r_{c_n}}{|x-w_{c_n}|}\le\frac{2r_{c_n}}{d}\le\varpi ∣ j ( γ c n , x ) ∣ = ∣ x − w c n ∣ r c n ≤ d 2 r c n ≤ ϖ . Now ∣ x − w c n ∣ ≥ d / 2 |x-w_{c_n}|\ge d/2 ∣ x − w c n ∣ ≥ d /2 is available, instead of the generic ≥ r c n / 2 \ge r_{c_n}/2 ≥ r c n /2 , while the remaining n − 1 n-1 n − 1 factors are ≤ ϖ \le\varpi ≤ ϖ as before. Hence, for every reduced word of length n ≥ 1 n\ge1 n ≥ 1 ,
∣ j ( γ ~ , x ) ∣ ≤ ϖ n , x ∈ K d , p . (7) |j(\widetilde{\gamma},x)|\le\varpi^n,\qquad x\in K_{d,p}. \tag*{(7)} ∣ j ( γ , x ) ∣ ≤ ϖ n , x ∈ K d , p . (7)
Let now y ∈ K d , p y\in K_{d,p} y ∈ K d , p and n ≥ 2 n\ge2 n ≥ 2 . By Lemma 3.2 , ∣ γ x − w − c 1 ∣ ≤ r c 1 ϖ ≤ r ϖ < d 8 ( 2 g − 1 ) ⋅ 1 4 ( 2 g − 1 ) ≤ d 288 |\gamma x-w_{-c_1}|\le r_{c_1}\varpi\le r\varpi<\frac{d}{8(2g-1)}\cdot\frac{1}{4(2g-1)}\le\frac{d}{288} ∣ γ x − w − c 1 ∣ ≤ r c 1 ϖ ≤ r ϖ < 8 ( 2 g − 1 ) d ⋅ 4 ( 2 g − 1 ) 1 ≤ 288 d (as g ≥ 2 g\ge2 g ≥ 2 ), while ∣ y − w − c 1 ∣ ≥ d / 2 |y-w_{-c_1}|\ge d/2 ∣ y − w − c 1 ∣ ≥ d /2 ; hence ∣ γ x − y ∣ ≥ d 2 − d 288 = 143 d 288 |\gamma x-y|\ge\frac d2-\frac{d}{288}=\frac{143d}{288} ∣ γ x − y ∣ ≥ 2 d − 288 d = 288 143 d . With (7), each term of length n n n is bounded by 288 143 d ϖ n \frac{288}{143d}\varpi^n 143 d 288 ϖ n , there are 2 g ( 2 g − 1 ) n − 1 2g(2g-1)^{n-1} 2 g ( 2 g − 1 ) n − 1 such words, and q : = ( 2 g − 1 ) ϖ ≤ 1 4 q:=(2g-1)\varpi\le\frac14 q := ( 2 g − 1 ) ϖ ≤ 4 1 . Therefore,
∣ R ( x , y ) ∣ ≤ 288 143 d ∑ n ≥ 2 2 g ( 2 g − 1 ) n − 1 ϖ n = 576 g 143 d ( 2 g − 1 ) ∑ n ≥ 2 q n ≤ 576 g 143 d ( 2 g − 1 ) ⋅ 4 3 q 2 , |R(x,y)|\le\frac{288}{143d}\sum_{n\ge2}2g(2g-1)^{n-1}\varpi^n=\frac{576g}{143d(2g-1)}\sum_{n\ge2}q^n\le\frac{576g}{143d(2g-1)}\cdot\frac43q^2, ∣ R ( x , y ) ∣ ≤ 143 d 288 n ≥ 2 ∑ 2 g ( 2 g − 1 ) n − 1 ϖ n = 143 d ( 2 g − 1 ) 576 g n ≥ 2 ∑ q n ≤ 143 d ( 2 g − 1 ) 576 g ⋅ 3 4 q 2 ,
using ∑ n ≥ 2 q n = q 2 / ( 1 − q ) ≤ 4 3 q 2 \sum_{n\ge2}q^n=q^2/(1-q)\le\frac43q^2 ∑ n ≥ 2 q n = q 2 / ( 1 − q ) ≤ 3 4 q 2 . Since q 2 = ( 2 g − 1 ) 2 ϖ 2 = 4 ( 2 g − 1 ) 2 ∣ ρ ∣ max d 2 q^2=(2g-1)^2\varpi^2=\frac{4(2g-1)^2|\rho|_{\max}}{d^2} q 2 = ( 2 g − 1 ) 2 ϖ 2 = d 2 4 ( 2 g − 1 ) 2 ∣ ρ ∣ m a x ,
∣ R ( x , y ) ∣ ≤ 576 g 143 d ( 2 g − 1 ) ⋅ 4 3 ⋅ 4 ( 2 g − 1 ) 2 ∣ ρ ∣ max d 2 = 9216 429 g ( 2 g − 1 ) ∣ ρ ∣ max d 3 = 64 3 144 143 g ( 2 g − 1 ) ∣ ρ ∣ max d 3 , |R(x,y)|\le\frac{576g}{143d(2g-1)}\cdot\frac43\cdot\frac{4(2g-1)^2|\rho|_{\max}}{d^2}=\frac{9216}{429}\frac{g(2g-1)|\rho|_{\max}}{d^3}=\frac{64}{3}\frac{144}{143}\frac{g(2g-1)|\rho|_{\max}}{d^3}, ∣ R ( x , y ) ∣ ≤ 143 d ( 2 g − 1 ) 576 g ⋅ 3 4 ⋅ d 2 4 ( 2 g − 1 ) 2 ∣ ρ ∣ m a x = 429 9216 d 3 g ( 2 g − 1 ) ∣ ρ ∣ m a x = 3 64 143 144 d 3 g ( 2 g − 1 ) ∣ ρ ∣ m a x ,
and 9216 429 = 21.48 … ≤ 22 \frac{9216}{429}=21.48\ldots\le22 429 9216 = 21.48 … ≤ 22 . This is (6). Letting ∣ λ ∣ max → 0 |\lambda|_{\max}\to0 ∣ λ ∣ m a x → 0 with g g g , d d d fixed gives ∣ ρ ∣ max = ∣ λ ∣ max 2 → 0 |\rho|_{\max}=|\lambda|_{\max}^2\to0 ∣ ρ ∣ m a x = ∣ λ ∣ m a x 2 → 0 , hence R = O ( ∣ λ ∣ max 2 ) R=O(|\lambda|_{\max}^2) R = O ( ∣ λ ∣ m a x 2 ) . Holomorphicity is Theorem 3.5 (i). □
The bracket is skew symmetric under x ↔ y x\leftrightarrow y x ↔ y , as required by Proposition 3.7 (i). The expansion is in the Schottky multipliers λ a \lambda_a λ a (the multiplier of γ a \gamma_a γ a is ≈ λ a 2 / ( w a − w − a ) 2 \approx\lambda_a^2/(w_a-w_{-a})^2 ≈ λ a 2 / ( w a − w − a ) 2 ). It is not a sewing expansion in the vertex-algebraic sense. Half integer powers of a sewing parameter also occur in the sewing expansions of [18 ] . λ a \lambda_a λ a plays a role analogous to a square root of a sewing parameter. We do not claim identification of the two schemes. The constants of Theorem 5.1 are 64 3 ⋅ 144 143 = 9216 429 ≈ 21.48 ≤ 22 \frac{64}{3}\cdot\frac{144}{143}=\frac{9216}{429}\approx21.48\le22 3 64 ⋅ 143 144 = 429 9216 ≈ 21.48 ≤ 22 .
Stress tensor
Fermion OPE convention
The free fermion VOSA is generated by ψ ∈ V 1 / 2 ∩ V ‾ 1 \psi\in V_{1/2}\cap\overline{V}_1 ψ ∈ V 1/2 ∩ V 1 with ψ ( 0 ) ψ = 1 \psi(0)\psi=1 ψ ( 0 ) ψ = 1 , ψ ( n ) ψ = 0 \psi(n)\psi=0 ψ ( n ) ψ = 0 (n ≥ 1 n\ge1 n ≥ 1 ), and Virasoro vector ω V = 1 2 ψ ( − 2 ) ψ ∈ V 2 \omega_V=\frac12\psi(-2)\psi\in V_2 ω V = 2 1 ψ ( − 2 ) ψ ∈ V 2 . This normalization is fixed by requiring the standard genus zero OPE
T ( z ) ψ ( w ) ∼ 1 / 2 ( z − w ) 2 ψ ( w ) + 1 z − w ∂ ψ ( w ) , (8) T(z)\psi(w)\sim\frac{1/2}{(z-w)^2}\psi(w)+\frac{1}{z-w}\partial\psi(w), \tag*{(8)} T ( z ) ψ ( w ) ∼ ( z − w ) 2 1/2 ψ ( w ) + z − w 1 ∂ ψ ( w ) , (8)
c = 1 2 c=\frac12 c = 2 1 , which is equivalent (by the state-field correspondence and ω V = 1 2 ψ ( − 2 ) ψ \omega_V=\frac12\psi(-2)\psi ω V = 2 1 ψ ( − 2 ) ψ ) to ψ ( x ) ψ ( y ) ∼ 1 x − y 1 + ( x − y ) ( 1 2 ψ ( − 2 ) ψ ) ( y ) + O ( ( x − y ) 2 ) \psi(x)\psi(y)\sim\frac{1}{x-y}1+\left(x-y\right)\left(\frac12\psi(-2)\psi\right)(y)+O\left((x-y)^2\right) ψ ( x ) ψ ( y ) ∼ x − y 1 1 + ( x − y ) ( 2 1 ψ ( − 2 ) ψ ) ( y ) + O ( ( x − y ) 2 ) . I.e., to ∂ ψ ψ = 1 2 T \partial\psi\psi={\frac12}T ∂ ψ ψ = 2 1 T where ∂ ψ ψ : ( y ) : = lim x → y ( ∂ x 1 x − y ) ψ ( x ) ψ ( y ) \partial\psi\psi:(y):=\lim_{x\to y}\left(\partial_x\frac{1}{x-y}\right)\psi(x)\psi(y) ∂ ψ ψ : ( y ) := lim x → y ( ∂ x x − y 1 ) ψ ( x ) ψ ( y ) is the standard normal-ordered product. We assume ( A f f ) (A_{\mathrm{ff}}) ( A ff ) at m = 2 m=2 m = 2 , i.e., Proposition 4.5 by specializing Proposition 4.5. In particular ⟨ ψ ( x ) ψ ( y ) ⟩ Σ ε = Z g s ε ( x , y ) ( d x d y ) 1 / 2 \langle\psi(x)\psi(y)\rangle_{\Sigma_\varepsilon}=Z_g\,s_\varepsilon(x,y)(dx\,dy)^{1/2} ⟨ ψ ( x ) ψ ( y ) ⟩ Σ ε = Z g s ε ( x , y ) ( d x d y ) 1/2 for the unnormalized correlator. Then ⟨ T ⟩ = 1 2 ⟨ : ∂ ψ ψ : ⟩ \langle T\rangle=\frac12\langle:\partial\psi\psi:\rangle ⟨ T ⟩ = 2 1 ⟨ : ∂ ψ ψ : ⟩ , and when Z g ≠ 0 Z_g\ne0 Z g = 0 the normalized one-point function is ⟨ T ⟩ / Z g \langle T\rangle/Z_g ⟨ T ⟩ / Z g .
Finally, having ( A f f ) (A_{\mathrm{ff}}) ( A ff ) at m = 2 m=2 m = 2 and the OPE normalization ψ ( x ) ψ ( y ) ∼ 1 x − y \psi(x)\psi(y)\sim\frac{1}{x-y} ψ ( x ) ψ ( y ) ∼ x − y 1 , T = 1 2 : ∂ ψ ψ : T=\frac12:\partial\psi\psi: T = 2 1 : ∂ ψ ψ : , let us discuss the stress tensor.
Proposition 6.1. Under the normalization above, the unnormalized and (for Z g ≠ 0 Z_g\ne0 Z g = 0 ) normalized one-point functions of T T T are
⟨ T ( y ) ⟩ = Z g ⋅ 1 2 lim x → y s ε ( x , y ) − 1 x − y x − y , ⟨ T ( y ) ⟩ Z g = 1 2 c 1 ( y ) = 1 24 R ε ( y ) . \langle T(y)\rangle=Z_g\cdot\frac12\lim_{x\to y}\frac{s_\varepsilon(x,y)-\frac{1}{x-y}}{x-y},\qquad\frac{\langle T(y)\rangle}{Z_g}=\frac12c_1(y)=\frac{1}{24}R_\varepsilon(y). ⟨ T ( y )⟩ = Z g ⋅ 2 1 x → y lim x − y s ε ( x , y ) − x − y 1 , Z g ⟨ T ( y )⟩ = 2 1 c 1 ( y ) = 24 1 R ε ( y ) .
Proof. By Proposition 3.7 (ii), s ε ( x , y ) − 1 x − y = c 1 ( y ) ( x − y ) + O ( ( x − y ) 2 ) s_{\varepsilon}(x,y)-\frac{1}{x-y}=c_{1}(y)(x-y)+O((x-y)^{2}) s ε ( x , y ) − x − y 1 = c 1 ( y ) ( x − y ) + O (( x − y ) 2 ) near Δ \Delta Δ , thus lim x → y ∂ x ( s ε ( x , y ) − 1 x − y ) = c 1 ( y ) \lim_{x\to y}\partial_{x}\left(s_{\varepsilon}(x,y)-\frac{1}{x-y}\right)=c_{1}(y) lim x → y ∂ x ( s ε ( x , y ) − x − y 1 ) = c 1 ( y ) . I.e., ⟨ : ∂ ψ ψ : ( y ) ⟩ = c 1 ( y ) \langle:\partial\psi\psi:(y)\rangle=c_{1}(y) ⟨ : ∂ ψ ψ : ( y )⟩ = c 1 ( y ) and, after dividing by Z g ≠ 0 Z_{g}\ne0 Z g = 0 , ⟨ T ( y ) ⟩ / Z g = 1 2 c 1 ( y ) \langle T(y)\rangle/Z_{g}=\frac{1}{2}c_{1}(y) ⟨ T ( y )⟩ / Z g = 2 1 c 1 ( y ) , which is the displayed formula. Equivalently, it is obtained by dividing s ε − 1 x − y = c 1 ( y ) ( x − y ) + O ( ( x − y ) 2 ) s_{\varepsilon}-\frac{1}{x-y}=c_{1}(y)(x-y)+O((x-y)^{2}) s ε − x − y 1 = c 1 ( y ) ( x − y ) + O (( x − y ) 2 ) by ( x − y ) (x-y) ( x − y ) before taking x → y x\to y x → y , as in the statement. Without this division by ( x − y ) (x-y) ( x − y ) the limit would be 0 0 0 , thus the division is essential. By Proposition 3.7 (iii), R ε = 12 c 1 R_{\varepsilon}=12c_{1} R ε = 12 c 1 is a projective connection. Thus, ( ⟨ T ⟩ / Z g ) ( z ) = 1 2 c 1 ( z ) = w ′ 2 ⋅ 1 2 c 1 ( w ) + 1 12 { w ; z } = w ′ 2 ( ⟨ T ⟩ / Z g ) ( w ) + 1 12 { w ; z } \left(\langle T\rangle/Z_{g}\right)^{(z)}=\frac{1}{2}c_{1}^{(z)}=w^{\prime2}\cdot\frac{1}{2}c_{1}^{(w)}+\frac{1}{12}\{w;z\}=w^{\prime2}\left(\langle T\rangle/Z_{g}\right)^{(w)}+\frac{1}{12}\{w;z\} ( ⟨ T ⟩ / Z g ) ( z ) = 2 1 c 1 ( z ) = w ′2 ⋅ 2 1 c 1 ( w ) + 12 1 { w ; z } = w ′2 ( ⟨ T ⟩ / Z g ) ( w ) + 12 1 { w ; z } . This is the anomalous transformation law of a stress tensor of central charge c = 1 2 c=\frac{1}{2} c = 2 1 under z ↦ w z\mapsto w z ↦ w , consistent with (8) .
Consistency check of sign and scalar: with ψ ( x ) ψ ( y ) ∼ ( x − y ) − 1 \psi(x)\psi(y)\sim(x-y)^{-1} ψ ( x ) ψ ( y ) ∼ ( x − y ) − 1 and T = 1 2 : ∂ ψ ψ : T=\frac{1}{2}:\partial\psi\psi: T = 2 1 : ∂ ψ ψ : (c = 1 2 c=\frac{1}{2} c = 2 1 ), the one point function is 1 2 lim x → y ∂ x ( ⟨ ψ ( x ) ψ ( y ) ⟩ / Z − 1 x − y ) = 1 2 c 1 ( y ) \frac{1}{2}\lim_{x\to y}\partial_{x}\left(\langle\psi(x)\psi(y)\rangle/Z-\frac{1}{x-y}\right)=\frac{1}{2}c_{1}(y) 2 1 lim x → y ∂ x ( ⟨ ψ ( x ) ψ ( y )⟩ / Z − x − y 1 ) = 2 1 c 1 ( y ) , using ⟨ ψ ( x ) ψ ( y ) ⟩ = Z s ε ( x , y ) \langle\psi(x)\psi(y)\rangle=Zs_{\varepsilon}(x,y) ⟨ ψ ( x ) ψ ( y )⟩ = Z s ε ( x , y ) (hypothesis ( A f f ) (A_{\mathrm{ff}}) ( A ff ) , m = 2 m=2 m = 2 ). Under a coordinate change the anomaly of 1 2 c 1 = 1 24 R ε \frac{1}{2}c_{1}=\frac{1}{24}R_{\varepsilon} 2 1 c 1 = 24 1 R ε is 1 24 { ϕ , y } = c 12 { ϕ , y } \frac{1}{24}\{\phi,y\}=\frac{c}{12}\{\phi,y\} 24 1 { ϕ , y } = 12 c { ϕ , y } with c = 1 2 c=\frac{1}{2} c = 2 1 , which fixes both the sign and the scalar factor. This agrees with the diagonal expansion of Remark 3.9 .
Corollary 6.2. ⟨ T ⟩ ε / Z g = 1 24 ( S B + 6 ∑ i , j ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ζ j ) \langle T\rangle_{\varepsilon}/Z_{g}=\frac{1}{24}\left(S_{B}+6\sum_{i,j}\partial_{i}\partial_{j}\log\theta[\delta](0)\,\zeta_{i}\zeta_{j}\right) ⟨ T ⟩ ε / Z g = 24 1 ( S B + 6 ∑ i , j ∂ i ∂ j log θ [ δ ] ( 0 ) ζ i ζ j ) when Z g ≠ 0 Z_{g}\ne0 Z g = 0 (under ( A f f ) (A_{\mathrm{ff}}) ( A ff ) ), by Corollary 3.18 . For g = 1 g=1 g = 1 this is the expression of Proposition 3.19 (c).
Remark 6.3. Let us see that no Bers-type moduli operator exists in our formulation. No analogue of the Bers moduli-derivative operator is defined here: S ε ( x , w c ) S_{\varepsilon}(x,w_{c}) S ε ( x , w c ) has weight 1 2 \frac{1}{2} 2 1 in w c w_{c} w c whereas ∂ w c \partial_{w_{c}} ∂ w c has weight − 1 -1 − 1 , thus a faithful analogue would require odd partners of weight − 3 2 -\frac{3}{2} − 2 3 (super-moduli, cf. [14 ] ).
Open problems and conclusions
We collect here the proven and conjectural statements of this paper. We proved, for the N g N_{g} N g lifted spin structures only: global convergence and meromorphicity of the half weight Poincaré series on ( H ) (\mathbb{H}) ( H ) (estimates and gluing) and, qualitatively, whenever (E) holds, in any genus g ≥ 1 g\ge1 g ≥ 1 ; periodicity and skew-symmetry; H 0 ( L ε ) = 0 H^{0}(L_{\varepsilon})=0 H 0 ( L ε ) = 0 ; identification of (a) the series, (b) the descended section, (c) the canonical Szegő kernel and the theta-function kernel; the projective connection and its relation to the Bergman projective connection; the genus one kernel and dictionary; residue reconstruction for arbitrary pole orders; the expansion in the Schottky multipliers with remainder; the stress-tensor formula.
Beyond (E): for δ ( Γ ) < 1 2 \delta(\Gamma)<\frac{1}{2} δ ( Γ ) < 2 1 convergence is proven (Theorem 3.15). It is an open question whether (E) fails when δ ( Γ ) ≥ 1 2 \delta(\Gamma)\ge\frac{1}{2} δ ( Γ ) ≥ 2 1 , and whether a Hecke-type regularization ∑ γ ~ ∣ j ( γ ~ , x e ) ∣ 2 s ( ⋯ ) \sum_{\widetilde{\gamma}}|j(\widetilde{\gamma},x_{e})|^{2s}(\cdots) ∑ γ ∣ j ( γ , x e ) ∣ 2 s ( ⋯ ) , continued analytically to s = 0 s=0 s = 0 , extends S ε S_{\varepsilon} S ε across the locus where (E) fails, for instance, across the theta-null divisor θ [ δ ] ( 0 ) = 0 \theta[\delta](0)=0 θ [ δ ] ( 0 ) = 0 . The dictionary for g ≥ 2 g\ge2 g ≥ 2 : determine ε ↦ δ ′ ( ε ) \varepsilon\mapsto\delta^{\prime}(\varepsilon) ε ↦ δ ′ ( ε ) . Remark 3.20 gives the expected form.
Grandin Hypothesis ( A f f ) (A_{\mathrm{ff}}) ( A ff ) for the Schottky scheme: whether the Schottky-sewn correlation functions of the free fermion VOSA satisfy ( A f f ) (A_{\mathrm{ff}}) ( A ff ) is (to our knowledge) an open question at every genus g ≥ 2 g\ge2 g ≥ 2 . The genus one and genus two sewing schemes of [18 , 19 , 20 , 21 ] are different constructions, and no identification of the schemes is claimed here. Corollary 4.4 depends on (A) and Proposition 4.5 on ( A f f ) (A_{\mathrm{ff}}) ( A ff ) , neither of which is established for Schottky-sewn correlators.
The other 2 2 g − N g 2^{2g}-N_{g} 2 2 g − N g spin structures: (both even structures outside the N g N_{g} N g -family and all odd ones), zero modes, super-moduli. We do not treated it in this paper. It presumably requires either a different choice of generators/markings for each spin structure, or the super-Schottky formalism of [14 ] . Super-moduli analogue of the Bers operator: see Remark 6.3.
It remains conjectural: the dictionary for g ≥ 2 g\ge2 g ≥ 2 , continuation beyond (E), and the general-VOSA recursion (Conjecture 4.6).
The material of this paper is also useful in other areas of condensed matter theory [6 , 22 , 23 , 24 , 10 ] , and foliation theory [1 , 2 , 3 , 25 , 26 , 27 ] .
Acknowledgments
The author is supported by the Institute of Mathematics, Academy of Sciences of the Czech Republic (RVO 67985840).
Data Availability. Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.
Declarations
This paper arose in close collaboration with ChatGPT, an AI assistant developed by OpenAI. The collaboration involved exploring proof strategies, refining arguments, locating references, and drafting.
Revision and writing of final text have been done by the author. The author takes full responsibility for the mathematical content and the final text.
Conflict of interest. The authors have no conflicts of interest to declare that are relevant to the content of this article.
References
Institute of Mathematics, Czech Academy of Sciences, Žitná 25, Prague, Czech Republic
Email address: zuevsky@yahoo.com