Introduction
Height is one of the most important concepts in number theory, with earliest prototype dating back to the method of infinite descent by Fermat. A century ago, Mordell systemically used the notion of height to prove the famous Mordell theorem on the finite generation of rational points on elliptic curves. The next highlight moment of height was Faltings’ proof of the Mordell conjecture and the Gross–Zagier formula toward the Birch and Swinnerton-Dyer conjecture, both in the 1980s.
Recall that the Gross–Zagier formula [14 ] computes the central derivative of the Rankin–Selberg L L L -function of a cusp newform of weight 2 and trivial Nebentypus and a ring class group character of an imaginary quadratic field, jointly under the so-called Heegner condition, through the height of the so-called Heegner point. Under the philosophy of this formula, one expects an arithmetic interpretation of the central derivative of L ( s , σ , χ ) L(s,\sigma,\chi) L ( s , σ , χ ) , where σ \sigma σ is a cuspidal automorphic representation of G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) with F F F a totally real field such that σ ∞ \sigma_\infty σ ∞ is a discrete series, and χ \chi χ is a Hecke character of a CM extension E / F E/F E / F , jointly satisfying
self-duality condition : the central character of σ \sigma σ is inverse to the restriction of χ \chi χ to A F × \mathbb{A}_F^\times A F × (which would imply that L ( s , σ , χ ) L(s,\sigma,\chi) L ( s , σ , χ ) is self-dual);
odd-parity condition : the root number of L ( s , σ , χ ) L(s,\sigma,\chi) L ( s , σ , χ ) is − 1 -1 − 1 (so that the central vanishing order is odd);
weight-balancing condition : the weight of χ \chi χ at every complex embedding of E E E is at most the weight of σ \sigma σ at the underlying real place of F F F minus 2.
When σ \sigma σ has parallel weight 2 (so that χ \chi χ must be a Dirichlet character), such expectation has been achieved by Yuan–Zhang–Zhang [39 ] unconditionally. In higher weight case, S. Zhang [Zha97] obtained a formula, still under the Heegner condition (so that F ≅ Q F \cong\mathbb{Q} F ≅ Q ), for σ \sigma σ of arbitrary even weight and trivial central character and χ \chi χ a (ring) class group character, through the Beilinson–Bloch height of the so-called Heegner cycle. However, unlike the weight 2 case, Zhang’s formula crucially relies on the Heegner condition as the construction involves the Kuga–Sato variety, which does not exist for other quaternionic Shimura curves.1 For this reason, there is almost no further progress in the case of general weights since Zhang’s work, except for a recent preprint [23 ] where the authors allow the character to have nonzero weights (but still under the Heegner condition).
In this article, we will give an answer to the question of finding an arithmetic interpretation of the central derivative of L ( s , σ , χ ) L(s,\sigma,\chi) L ( s , σ , χ ) in the above generality. Note that, it is possible that the weights of σ \sigma σ may have different parities, in which case σ \sigma σ is not cohomological hence does not appear in the cohomology of quaternionic Shimura curves. For this reason, it is necessary to encode σ \sigma σ and χ \chi χ simultaneously, which leads naturally to the unitary Shimura curves.
Special cycles in motivic cohomology
Let E / F E/F E / F be a CM field. We develop a theory of special cycles and generating functions of general weights for incoherent totally positive definite hermitian spaces over A E / A F A_E/A_F A E / A F (Definition 2.15), parallel to the original Kudla–Millson theory of the trivial weight for incoherent totally positive definite quadratic spaces over A F A_F A F [18 ] .
To reduce unnecessary technical complication, in the main part of the article we will assume that E E E contains an imaginary quadratic field (see Remark 6.7 for general CM fields). From now on, we assume this and choose an imaginary quadratic subfield E 0 ⊆ E E_0 \subseteq E E 0 ⊆ E .
Let V V V be an incoherent totally positive definite hermitian space over A E / A F A_E/A_F A E / A F of rank m ≥ 1 m \ge1 m ≥ 1 ; and put H : = U ( V ) H := U(V) H := U ( V ) .
In § 3.2 , we construct a projective system of smooth quasi-projective varieties { X L } L \{X_L\}_L { X L } L over E E E indexed by neat open compact subgroups L L L of H ( A F ∞ ) H(A_F^\infty) H ( A F ∞ ) , whose base change along any given complex embedding of E E E recovers usual Shimura varieties for certain unitary groups, together with a relative (Chow) motive M L M_L M L over X L X_L X L with coefficients in E E E functorial in the level L L L , which we call the standard motive . In § 3.4 , we use M L M_L M L to construct, for every hermitian E E E -weight λ \lambda λ (Definition 2.7), a relative motive M L , λ M_{L,\lambda} M L , λ over X L X_L X L with coefficients in E λ E^\lambda E λ – the field of definition of λ \lambda λ as a subfield of C \mathbb{C} C – functorial in the level L L L , which we call the automorphic motive of weight λ \lambda λ . The motive M L , λ M_{L,\lambda} M L , λ is an explicit motivic object underlying all coefficient systems of weight λ \lambda λ in the following sense: for every prime ℓ \ell ℓ , the ℓ \ell ℓ -adic étale realization of M L , λ M_{L,\lambda} M L , λ is naturally the ℓ \ell ℓ -adic local system over X L X_L X L of weight λ \lambda λ ; and the similar property holds for Hodge realizations.
Consider a positive integer r r r and a hermitian E E E -weight λ \lambda λ that is r r r -balanced (Definition 2.7). In § 3.7 we define, for every r r r -tuple x ∈ ( V ∞ ) r x \in(V^\infty)^r x ∈ ( V ∞ ) r with T ( x ) ∈ Herm r ( F ) + T(x) \in\operatorname{Herm}_r(F)^+ T ( x ) ∈ Herm r ( F ) + (Notation 2.6), an element
Z λ ( x ) L ∈ S λ + ( Q r ) ∨ ⊗ Q H 2 r ( X L , M L , λ ( r ) ) , Z^\lambda(x)_L \in S_{\lambda^+}(\mathbb{Q}^r)^\vee\otimes_{\mathbb{Q}} H^{2r}(X_L,M_{L,\lambda}(r)), Z λ ( x ) L ∈ S λ + ( Q r ) ∨ ⊗ Q H 2 r ( X L , M L , λ ( r )) ,
called motivic special class , for every level L L L , in which S λ + ( Q r ) S_{\lambda^+}(\mathbb{Q}^r) S λ + ( Q r ) is the (algebraic) representation of GL r ( Q ) Hom ( E , C ) \operatorname{GL}_r(\mathbb{Q})^{\operatorname{Hom}(E,\mathbb{C})} GL r ( Q ) Hom ( E , C ) of weight λ + \lambda^+ λ + (Notation 2.1) and H ∙ ( X L , M L , λ ( r ) ) H^\bullet(X_L,M_{L,\lambda}(r)) H ∙ ( X L , M L , λ ( r )) denotes the motivic cohomology of the motive M L , λ ( r ) M_{L,\lambda}(r) M L , λ ( r ) over X L X_L X L . Once constructing special classes, we can formulate generating functions. Denote by G r G_r G r the quasi-split unitary group of rank 2 r 2r 2 r over E / F E/F E / F . For every Schwartz function ϕ ∞ ∈ S ( ( V ∞ ) r ) L \phi^\infty\in\mathcal{S}((V^\infty)^r)^L ϕ ∞ ∈ S (( V ∞ ) r ) L for some level L L L , we have the formal series
Z λ ( ϕ ∞ ) L : = ∑ T ∈ Herm r ( F ) + Z T λ ( ϕ ∞ ) L Z^\lambda(\phi^\infty)_L := \sum_{T \in\operatorname{Herm}_r(F)^+} Z_T^\lambda(\phi^\infty)_L Z λ ( ϕ ∞ ) L := T ∈ Herm r ( F ) + ∑ Z T λ ( ϕ ∞ ) L
as a function on G r ( A F ) G_r(A_F) G r ( A F ) (or rather its metaplectic double cover when m m m is odd) valued in the complex vector space S λ + ( C r ) ∨ ⊗ R H 2 r ( X L , M L , λ ( r ) ) S_{\lambda^+}(\mathbb{C}^r)^\vee\otimes_{\mathbb{R}} H^{2r}(X_L,M_{L,\lambda}(r)) S λ + ( C r ) ∨ ⊗ R H 2 r ( X L , M L , λ ( r )) , which we call the motivic generating function (see Definition 3.35 for the precise definition). Note that when λ = 0 \lambda=0 λ = 0 is the trivial weight, S 0 ( C r ) = C S_0(\mathbb{C}^r)=\mathbb{C} S 0 ( C r ) = C , E 0 = Q E^0=\mathbb{Q} E 0 = Q , and M L , 0 M_{L,0} M L , 0 is the unit motive over X L X_L X L so that H 2 r ( X L , M L , 0 ( r ) ) = CH r ( X L ) ⊗ Q H^{2r}(X_L,M_{L,0}(r))=\operatorname{CH}^r(X_L)\otimes\mathbb{Q} H 2 r ( X L , M L , 0 ( r )) = CH r ( X L ) ⊗ Q ; in this case, Z 0 ( x ) L Z^0(x)_L Z 0 ( x ) L recovers the usual special cycle and Z 0 ( ϕ ∞ ) L Z^0(\phi^\infty)_L Z 0 ( ϕ ∞ ) L recovers the usual generating function, both in the hermitian case (see [25 ] or [20 ] ). Thus, it is natural to propose the following conjecture.
Conjecture 1.1 (Modularity of motivic generating function, Conjecture 3.36 ). *Let r r r be a positive integer such that λ \lambda λ is r r r -balanced. For every level L L L and every Schwartz function ϕ ∞ ∈ S ( ( V ∞ ) r ) L \phi^\infty\in\mathcal{S}((V^\infty)^r)^L ϕ ∞ ∈ S (( V ∞ ) r ) L , the motivic generating function Z λ ( ϕ ∞ ) L Z^\lambda(\phi^\infty)_L Z λ ( ϕ ∞ ) L is given by a holomorphic vector-valued hermitian Siegel modular form (of
genus r r r with respect to the CM extension E / F E/F E / F ) for the representation S λ + ( C r ) ∨ ⊗ δ − m / 2 S_{\lambda+}(\mathbb{C}^{r})^{\vee} \otimes\delta^{-m/2} S λ + ( C r ) ∨ ⊗ δ − m /2 (see Notation 2.13(4) for δ \delta δ ) with coefficients in H 2 r ( X L , M L , λ ( r ) ) ⊗ E λ C H^{2r}(X_L,M_{L,\lambda}(r)) \otimes_{E_\lambda} \mathbb{C} H 2 r ( X L , M L , λ ( r )) ⊗ E λ C .
Unlike the case for the trivial weight, even for r = 1 r=1 r = 1 (or even for r = 1 r=1 r = 1 and m = 2 m=2 m = 2 ), we do not have a proof of the above conjecture. Nevertheless, assuming the above conjecture, one may use Z ℓ ( φ ∞ ) L Z^{\ell}(\varphi^{\infty})_L Z ℓ ( φ ∞ ) L as the theta kernel to define what we call motivic arithmetic theta lift : For every form φ ∈ A λ + m / 2 ( G r ) c u s p h o l \varphi\in\mathcal{A}^{m/2}_{\lambda+}(G_r)^{\mathrm{hol}}_{\mathrm{cusp}} φ ∈ A λ + m /2 ( G r ) cusp hol (Definition 2.14) and every φ ∞ ∈ S ( ( V ∞ ) r ) L \varphi^{\infty} \in\mathcal{S}((V^{\infty})^r)^L φ ∞ ∈ S (( V ∞ ) r ) L , we define
Θ φ ∞ λ ( φ ) L : = ∫ G r ( F ) \ G r ( A F ) tr λ ( Z ℓ ( φ ∞ ) L ( g ) ⊗ φ ( g ) ‾ ) d g ∈ H 2 r ( X L , M L , λ ( r ) ) ⊗ E λ C . \Theta^{\lambda}_{\varphi^{\infty}}(\varphi)_L := \int_{G_r(F)\backslash G_r(\mathbb{A}_F)} \operatorname{tr}_{\lambda}\left(Z^{\ell}(\varphi^{\infty})_L(g) \otimes\overline{\varphi(g)}\right)\,dg \in H^{2r}(X_L,M_{L,\lambda}(r)) \otimes_{E_\lambda} \mathbb{C}. Θ φ ∞ λ ( φ ) L := ∫ G r ( F ) \ G r ( A F ) tr λ ( Z ℓ ( φ ∞ ) L ( g ) ⊗ φ ( g ) ) d g ∈ H 2 r ( X L , M L , λ ( r )) ⊗ E λ C .
where tr λ : S λ ( C r ) ∨ ⊗ C S λ ( C r ) ∨ ‾ → C \operatorname{tr}_{\lambda}: S_{\lambda}(\mathbb{C}^{r})^{\vee} \otimes_{\mathbb{C}} \overline{S_{\lambda}(\mathbb{C}^{r})^{\vee}} \to\mathbb{C} tr λ : S λ ( C r ) ∨ ⊗ C S λ ( C r ) ∨ → C is the natural pairing induced from the standard hermitian form on C r \mathbb{C}^{r} C r . One of the main outcomes of § 4 is to construct, when m = 2 r m=2r m = 2 r and the system { X L } L \{X_L\}_L { X L } L is proper, a level-independent height pairing
⟨ Θ φ 1 ∞ λ ( φ 1 ) , Θ φ 2 ∞ λ ( φ 2 ) ⟩ λ \left\langle\Theta^{\lambda}_{\varphi^{\infty}_1}(\varphi_1),\Theta^{\lambda}_{\varphi^{\infty}_2}(\varphi_2)\right\rangle_{\lambda} ⟨ Θ φ 1 ∞ λ ( φ 1 ) , Θ φ 2 ∞ λ ( φ 2 ) ⟩ λ
between two motivic arithmetic theta lifts (see Remark 1.3 below), so that we may formulate the following conjecture.
Conjecture 1.2 (Motivic arithmetic inner product formula, Conjecture 4.11 ). Suppose that m = 2 r m=2r m = 2 r , that the hermitian E E E -weight λ \lambda λ is r r r -balanced, and that the system { X L } L \{X_L\}_L { X L } L is proper. Let ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ (Definition 2.27 ). For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) of automorphic forms in ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + h o l ( G r ) c u s p (S_{\lambda+}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\mathcal{A}_{\pi})\cap\mathcal{A}^{\mathrm{hol}}_{\lambda+}(G_r)_{\mathrm{cusp}} ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + hol ( G r ) cusp , and every pair φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) \varphi^{\infty}=(\varphi^{\infty}_1,\varphi^{\infty}_2) φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) of Schwartz functions in S ( ( V ∞ ) r ) \mathcal{S}((V^{\infty})^r) S (( V ∞ ) r ) ,
⟨ Θ φ 1 ∞ λ ( φ 1 ) , Θ φ 2 ∞ λ ( φ 2 ) ⟩ λ = L ′ ( 1 2 , π ) b 2 r ( 0 ) ⋅ C ( λ ) ⋅ Z f i n ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ‾ ) \left\langle\Theta^{\lambda}_{\varphi^{\infty}_1}(\varphi_1),\Theta^{\lambda}_{\varphi^{\infty}_2}(\varphi_2)\right\rangle_{\lambda}
=
\frac{L'(\frac{1}{2},\pi)}{b_{2r}(0)}\cdot C(\lambda)\cdot\mathcal{Z}^{\sharp}_{\mathrm{fin}}\left(\varphi,\varphi^{\infty}_1\otimes\overline{\varphi^{\infty}_2}\right) ⟨ Θ φ 1 ∞ λ ( φ 1 ) , Θ φ 2 ∞ λ ( φ 2 ) ⟩ λ = b 2 r ( 0 ) L ′ ( 2 1 , π ) ⋅ C ( λ ) ⋅ Z fin ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ) holds. Here, C ( λ ) ∈ R × C(\lambda)\in\mathbb{R}^{\times} C ( λ ) ∈ R × is the real number of sign ( − 1 ) r [ F : Q ] (-1)^{r[F:\mathbb{Q}]} ( − 1 ) r [ F : Q ] from Proposition 2.29 , and Z f i n ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ‾ ) \mathcal{Z}^{\sharp}_{\mathrm{fin}}\left(\varphi,\varphi^{\infty}_1\otimes\overline{\varphi^{\infty}_2}\right) Z fin ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ) is the normalized doubling zeta integral introduced in Definition 2.28 .
Remark 1.3. Similar to Beilinson’s height pairing on Chow groups [4 ] (see also [20 §4] ), the height pairing on the motivic cohomology is actually defined on a certain subspace
H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ⊆ H 2 r ( X L , M L , λ ( r ) ) H^{2r}(X_L,M_{L,\lambda}(r))^{\langle\ell\rangle}\subseteq H^{2r}(X_L,M_{L,\lambda}(r)) H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ⊆ H 2 r ( X L , M L , λ ( r ))
depending on the choice of a (sufficiently large) rational prime ℓ \ell ℓ , and takes values in C ⊗ Q ℓ \mathbb{C}\otimes\mathbb{Q}_{\ell} C ⊗ Q ℓ a priori. The above (global) height pairing can also be decomposed into a sum of local heights over all places of E E E for certain representatives of classes, similar to the case of constant coefficients. Gladly, when r = 1 r=1 r = 1 and m = 2 m=2 m = 2 (so that X L X_L X L is a curve), such local heights coincide with the one defined and studied by Brylinski in [6 ] . In some sense, our construction fulfills an expectation of Brylinski,2 at least for unitary Shimura curves, that there should be a global theory overseeing these local heights.
Moreover, in the case where r = 1 r=1 r = 1 and m = 2 m=2 m = 2 , the subspace H 2 ( X L , M L , λ ( 1 ) ) ⟨ ℓ ⟩ H^2(X_L,M_{L,\lambda}(1))^{\langle\ell\rangle} H 2 ( X L , M L , λ ( 1 ) ) ⟨ ℓ ⟩ is independent of ℓ \ell ℓ (which coincides with H 2 ( X L , M L , λ ( 1 ) ) H^2(X_L,M_{L,\lambda}(1)) H 2 ( X L , M L , λ ( 1 )) when λ ≠ 0 \lambda\ne0 λ = 0 ) and the height pairing takes values in C \mathbb{C} C that is also independent of ℓ \ell ℓ (Theorem 4.15 ).
Remark 1.4. The idea of formulating the motivic arithmetic inner product formula can also be used to formulate the arithmetic Gan–Gross–Prasad conjecture [13 §27] for cohomological cuspidal automorphic representations of G L n ( A E ) × G L n + 1 ( A E ) \mathrm{GL}_n(\mathbb{A}_E)\times\mathrm{GL}_{n+1}(\mathbb{A}_E) GL n ( A E ) × GL n + 1 ( A E ) of general interlacing weights [27 Definition 4.1] .
Results for unitary groups of two variables
In this section, we assume r = 1 r=1 r = 1 and m = 2 m=2 m = 2 . When λ = 0 \lambda=0 λ = 0 , both of the modularity conjecture and the (motivic) arithmetic inner product formula have been confirmed in [25 , 26 ] .3
Here, we assume λ ≠ 0 \lambda\ne0 λ = 0 . Though we cannot prove the modularity conjecture now hence cannot define the motivic arithmetic theta lifts unconditionally, we can prove a numerical analogue of the modularity conjecture in terms of the height pairing of Fourier coefficients of generating functions. Using such modularity, we can produce sufficiently many doubling kernel functions that encode central L L L -derivatives of holomorphic tempered cuspidal automorphic representations of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ and root number − 1 -1 − 1 , which would imply the motivic arithmetic inner product formula assuming the modularity conjecture.
Recall that for every t ∈ F + t\in F^+ t ∈ F + , we have the t t t -th Fourier coefficient Z t λ ( ϕ ∞ ) L Z_t^\lambda(\phi^\infty)_L Z t λ ( ϕ ∞ ) L as a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) valued in H 2 ( X L , M L , λ ( 1 ) ) ⊗ E λ C H^2(X_L,M_{L,\lambda}(1))\otimes_{E_\lambda}\mathbb{C} H 2 ( X L , M L , λ ( 1 )) ⊗ E λ C , among which Z 0 λ ( ϕ ∞ ) L = 0 Z_0^\lambda(\phi^\infty)_L=0 Z 0 λ ( ϕ ∞ ) L = 0 as λ ≠ 0 \lambda\ne0 λ = 0 . Consider a finite set ◊ \Diamond ◊ of rational primes containing all those ramified in E E E and an open compact subgroup L ◊ L^\Diamond L ◊ of H ( A F ∞ , ◊ ) H(\mathbb{A}_F^{\infty,\Diamond}) H ( A F ∞ , ◊ ) that is the stabilizer of a self-dual lattice of V ∞ , ◊ V^{\infty,\Diamond} V ∞ , ◊ . For every nonempty subset ⧫ \blacklozenge ⧫ of places of F F F above ◊ \Diamond ◊ containing all those nonsplit in E E E ,4 we define a H ( F ◊ ) H(F_\Diamond) H ( F ◊ ) -stable subspace
S ( ( V ∞ ) 2 , L ) ⧫ L ◊ ⊆ S ( ( V ∞ ) 2 , L ) L ◊ \mathcal{S}((V^\infty)^2,\mathbb{L})_{\blacklozenge}^{L^\Diamond}\subseteq\mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\Diamond} S (( V ∞ ) 2 , L ) ⧫ L ◊ ⊆ S (( V ∞ ) 2 , L ) L ◊
in Definition 5.29 , and denote by S ( ( V ∞ ) 2 , L ) ⧫ L ◊ \mathcal{S}((V^\infty)^2,\mathbb{L})_{\blacklozenge}^{L^\Diamond} S (( V ∞ ) 2 , L ) ⧫ L ◊ the G 1 ( A F ∞ ) 2 G_1(\mathbb{A}_F^\infty)^2 G 1 ( A F ∞ ) 2 -stable (via the diagonal Weil representation) subspace generated by S ( ( V ∞ ) 2 , L ) ⧫ L ◊ \mathcal{S}((V^\infty)^2,\mathbb{L})_{\blacklozenge}^{L^\Diamond} S (( V ∞ ) 2 , L ) ⧫ L ◊ . The subspace S ( ( V ∞ ) 2 , L ) ⧫ L ◊ \mathcal{S}((V^\infty)^2,\mathbb{L})_{\blacklozenge}^{L^\Diamond} S (( V ∞ ) 2 , L ) ⧫ L ◊ is large enough in the sense that for every generic irreducible representation π ∞ \pi^\infty π ∞ of G 1 ( A F ∞ ) G_1(\mathbb{A}_F^\infty) G 1 ( A F ∞ ) , the natural restriction map
Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) L ◊ , π ∞ ⊠ ( π ∞ ) ∨ ) ⟶ Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) ⧫ L ◊ , π ∞ ⊠ ( π ∞ ) ∨ ) \operatorname{Hom}_{G_1(\mathbb{A}_F^\infty)^2}\left(\mathcal{S}((V^\infty)^2,\mathbb{C})^{L^\Diamond},\pi^\infty\boxtimes(\pi^\infty)^\vee\right)\longrightarrow\operatorname{Hom}_{G_1(\mathbb{A}_F^\infty)^2}\left(\mathcal{S}((V^\infty)^2,\mathbb{C})_{\blacklozenge}^{L^\Diamond},\pi^\infty\boxtimes(\pi^\infty)^\vee\right) Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) L ◊ , π ∞ ⊠ ( π ∞ ) ∨ ) ⟶ Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) ⧫ L ◊ , π ∞ ⊠ ( π ∞ ) ∨ )
is an isomorphism, which has dimension at most one by the multiplicity one property for degenerate principal series.
We assume that the system { X L } L \{X_L\}_L { X L } L is proper. As pointed out in Remark 1.3, we have a height pairing
H 2 ( X L , M L , λ ( 1 ) ) × H 2 ( X L , M L , λ ( 1 ) ) ⟶ C H^2(X_L,M_{L,\lambda}(1))\times H^2(X_L,M_{L,\lambda}(1))\longrightarrow\mathbb{C} H 2 ( X L , M L , λ ( 1 )) × H 2 ( X L , M L , λ ( 1 )) ⟶ C
for every level L L L as a hermitian form with coefficients in E λ E^\lambda E λ .
Theorem 1.5 (Theorem 6.1 ). Assume that the system { X L } L \{X_L\}_L { X L } L is proper. There is a unique G 1 ( A F ∞ ) 2 G_1(\mathbb{A}_F^\infty)^2 G 1 ( A F ∞ ) 2 -equivariant complex linear map
I : S ( ( V ∞ ) 2 , C ) ⧫ L ◊ ⟶ A λ + ( G 1 ) c u s p h o l ⊗ C A λ + ( G 1 ) c u s p h o l ‾ \mathfrak{I}:\mathcal{S}((V^\infty)^2,\mathbb{C})_{\blacklozenge}^{L^\Diamond}\longrightarrow\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}}\otimes_{\mathbb{C}}\overline{\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}}} I : S (( V ∞ ) 2 , C ) ⧫ L ◊ ⟶ A λ + ( G 1 ) cusp hol ⊗ C A λ + ( G 1 ) cusp hol such that for every pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^\infty=(\phi_1^\infty,\phi_2^\infty) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of Schwartz functions in S ( V ∞ , C ) L ◊ \mathcal{S}(V^\infty,\mathbb{C})^{L^\Diamond} S ( V ∞ , C ) L ◊ satisfying ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ∈ S ( ( V ∞ ) 2 , C ) ⧫ L ◊ \phi_1^\infty\otimes\overline{\phi_2^\infty}\in\mathcal{S}((V^\infty)^2,\mathbb{C})_{\blacklozenge}^{L^\Diamond} ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S (( V ∞ ) 2 , C ) ⧫ L ◊ , the identity
I ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) ( g 1 , g 2 ) = vol ♮ ( L ) ⋅ ∑ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L \mathfrak{I}(\phi_1^\infty\otimes\overline{\phi_2^\infty})(g_1,g_2)=\operatorname{vol}^{\natural}(L)\cdot\sum_{t_1\in(F^\times)^+}\sum_{t_2\in(F^\times)^+}\left\langle Z_{t_1}^\lambda(\phi_1^\infty)_L(g_1),Z_{t_2}^\lambda(\phi_2^\infty)_L(g_2)\right\rangle_{\lambda,L} I ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) ( g 1 , g 2 ) = vol ♮ ( L ) ⋅ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L holds with absolute convergence on the right-hand side for every ( g 1 , g 2 ) ∈ G 1 ( A F ) 2 (g_1,g_2)\in G_1(\mathbb{A}_F)^2 ( g 1 , g 2 ) ∈ G 1 ( A F ) 2 and every neat open compact subgroup L L L of H ( A F ∞ ) H(\mathbb{A}_F^\infty) H ( A F ∞ ) fixing both ϕ 1 ∞ \phi_1^\infty ϕ 1 ∞ and ϕ 2 ∞ \phi_2^\infty ϕ 2 ∞ .
The theorem below provides a variant of the motivic arithmetic inner product formula, showing that the functional I \mathfrak{I} I is indeed a doubling kernel function for the central L L L -derivative.
Theorem 1.6 (Theorem 6.3). *Assume that the system { X L } L \{X_L\}_L { X L } L is proper. Let ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ . For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) of automorphic forms in A π ∩ A λ , + ( G 1 ) c u s p h o l \mathcal{A}_\pi\cap\mathcal{A}_{\lambda,+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} A π ∩ A λ , + ( G 1 ) cusp hol and every Φ ∞ ∈ S ( ( V ∞ ) 2 , C ) ( ⋄ ) L \Phi^\infty\in\mathcal{S}((V^\infty)^2,\mathbb{C})^{L}_{(\diamond)} Φ ∞ ∈ S (( V ∞ ) 2 , C ) ( ⋄ ) L
∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 Φ ~ ( Φ ∞ ) ( g 1 , g 2 ) ⋅ φ 1 ( g 1 ) φ 2 ( g 2 ) d g 1 d g 2 = L ′ ( 1 2 , π ) b 2 ( 0 ) ⋅ ( − π 4 ) [ F : Q ] ⋅ Z f i n ♭ ( φ , Φ ∞ ) \iint_{(G_1(F)\backslash G_1(\mathbb{A}_F))^2} \widetilde{\Phi}(\Phi^\infty)(g_1,g_2)\cdot\varphi_1(g_1)\varphi_2(g_2)\,dg_1\,dg_2 = \frac{L'(\frac{1}{2},\pi)}{b_2(0)}\cdot\left(\frac{-\pi}{4}\right)^{[F:\mathbb{Q}]}\cdot\mathcal{Z}^{\flat}_{\mathrm{fin}}(\varphi,\Phi^\infty) ∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 Φ ( Φ ∞ ) ( g 1 , g 2 ) ⋅ φ 1 ( g 1 ) φ 2 ( g 2 ) d g 1 d g 2 = b 2 ( 0 ) L ′ ( 2 1 , π ) ⋅ ( 4 − π ) [ F : Q ] ⋅ Z fin ♭ ( φ , Φ ∞ ) holds.⁵
By Lemma 2.12 , the above theorem indeed fulfills our original objective of finding an arithmetic interpretation of the central derivative of L ( s , σ , χ ) L(s,\sigma,\chi) L ( s , σ , χ ) at the beginning of the introduction by taking π \pi π such that its base change to G L 2 ( A E ) \mathrm{GL}_2(\mathbb{A}_E) GL 2 ( A E ) is B C ( σ ) ⊗ χ \mathrm{BC}(\sigma)\otimes\chi BC ( σ ) ⊗ χ .
Corollary 1.7 (Corollary 6.6). Let the situation be as in the above theorem. Assume the modularity of motivic generating function for V V V . Then the motivic arithmetic inner product holds for ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) .
Remark 1.8. Careful readers may notice that we exclude one case in our theory (for r = 1 r=1 r = 1 and m = 2 m=2 m = 2 ), namely, the case where the system { X L } L \{X_L\}_L { X L } L is not proper, or equivalently, F = Q F=\mathbb{Q} F = Q and V ∞ V^\infty V ∞ is split. We expect that our theory works in this case as well with extra work on the boundary extension. Note that, this case is closely related to the one studied in [Zha97] and [23 ] , and our theory should be equivalent to the cycle construction in the two articles under their assumptions. We leave the details to the future.
Beilinson–Bloch conjecture for motivic cohomology
Let V V V be an incoherent totally positive definite hermitian space over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F of rank 2 r 2r 2 r . For every hermitian E E E -weight λ \lambda λ , put
H 2 r ( X , M λ ( r ) ) : = l i m → L H 2 r ( X L , M L , λ ( r ) ) \mathrm{H}^{2r}(X,\mathrm{M}_{\lambda}(r)):=\varinjlim_L \mathrm{H}^{2r}(X_L,\mathrm{M}_{L,\lambda}(r)) H 2 r ( X , M λ ( r )) := L lim H 2 r ( X L , M L , λ ( r ))
as a module over E λ [ H ( A F ∞ ) ] E^\lambda[\mathrm{H}(\mathbb{A}_F^\infty)] E λ [ H ( A F ∞ )] .
The conjecture below can be regarded as an analogue of the Birch–Swinnerton-Dyer conjecture (or rather the Beilinson–Bloch conjecture) for motivic cohomology of automorphic motives.
Conjecture 1.9 (Beilinson–Bloch, Conjecture 4.14). Suppose that the hermitian E E E -weight λ \lambda λ is balanced. Suppose that the system { X L } L \{X_L\}_L { X L } L is proper.
(1) The E λ [ H ( A F ∞ ) ] E^\lambda[\mathrm{H}(\mathbb{A}_F^\infty)] E λ [ H ( A F ∞ )] -module H 2 r ( X , M λ ( r ) ) \mathrm{H}^{2r}(X,\mathrm{M}_{\lambda}(r)) H 2 r ( X , M λ ( r )) is admissible and semisimple.
(2) For every holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ , the relation
dim C Hom H ( A F ∞ ) ( π ~ ∞ , H 2 r ( X , M λ ( r ) ) ⊗ E λ C ) = ord s = 1 / 2 L ( s , Π π ~ ∞ ) \dim_{\mathbb{C}}\operatorname{Hom}_{\mathrm{H}(\mathbb{A}_F^\infty)}\left(\widetilde{\pi}^{\infty},\mathrm{H}^{2r}(X,\mathrm{M}_{\lambda}(r))\otimes_{E^\lambda}\mathbb{C}\right)=\operatorname{ord}_{s=1/2}L(s,\Pi_{\widetilde{\pi}^\infty}) dim C Hom H ( A F ∞ ) ( π ∞ , H 2 r ( X , M λ ( r )) ⊗ E λ C ) = ord s = 1/2 L ( s , Π π ∞ ) holds for every irreducible admissible representation π ~ ∞ \widetilde{\pi}^{\infty} π ∞ of H ( A F ∞ ) \mathrm{H}(\mathbb{A}_F^\infty) H ( A F ∞ ) that is nearly equivalent to π ∞ \pi^\infty π ∞ . Here, Π π ~ ∞ \Pi_{\widetilde{\pi}^\infty} Π π ∞ is the unique isobaric factor of the automorphic base change of π \pi π to E E E (as an isobaric automorphic representation of G L 2 r ( A E ) \mathrm{GL}_{2r}(\mathbb{A}_E) GL 2 r ( A E ) ) determined by π ~ ∞ \widetilde{\pi}^\infty π ∞ (see Remark 4.13).*
For π \pi π in (2) above, there is, by the local theta dichotomy, a unique up to isomorphism totally positive definite hermitian space V ( π ) V(\pi) V ( π ) over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F such that the theta lift of π ∞ \pi^\infty π ∞ to V ( π ) ∞ V(\pi)^\infty V ( π ) ∞ does not vanish.
Theorem 1.6 has a corollary that can be regarded as an evidence toward the above conjecture when ord s = 1 / 2 L ( s , π ) = 1 \operatorname{ord}_{s=1/2}L(s,\pi)=1 ord s = 1/2 L ( s , π ) = 1 . Once again we assume r = 1 r=1 r = 1 , and λ \lambda λ nontrivial and 1-balanced. For every intermediate field E λ ⊆ L ⊆ C E^\lambda\subseteq L \subseteq\mathbb{C} E λ ⊆ L ⊆ C and every level L L L , we define a subquotient S C λ ( X L , L ) \mathrm{SC}_{\lambda}(X_L,\mathcal{L}) SC λ ( X L , L ) of H 2 ( X L , M L , λ ) ⊗ E λ L \mathrm{H}^2(X_L,\mathrm{M}_{L,\lambda})\otimes_{E^\lambda}L H 2 ( X L , M L , λ ) ⊗ E λ L in Definition 6.4, which is, roughly speaking, the subspace generated by motivic special classes modulo the radical with respect to the height pairing.
Corollary 1.10 (Corollary 6.5). Let ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) \mathrm{G}_{1}(\mathbb{A}_{F}) G 1 ( A F ) of weight λ \lambda λ such that V : = V ( π ) V:=V(\pi) V := V ( π ) is incoherent with the associated system { X L } L \{X_{L}\}_{L} { X L } L proper. Let E ♮ ⊆ L π ⊆ C \mathbb{E}^{\natural}\subseteq L_{\pi}\subseteq\mathbb{C} E ♮ ⊆ L π ⊆ C be the field of definition of π ∞ \pi^{\infty} π ∞ . If ord s = 1 / 2 L ( s , π ) = 1 \operatorname{ord}_{s=1/2}L(s,\pi)=1 ord s = 1/2 L ( s , π ) = 1 , then for every finite set ⋄ \diamond ⋄ of rational primes containing all those ramified in E E E such that π ∞ , ⋄ \pi^{\infty,\diamond} π ∞ , ⋄ is spherical6 and every open compact subgroup L ⋄ L^{\diamond} L ⋄ of H ( A F ∞ , ⋄ ) H(\mathbb{A}_{F}^{\infty,\diamond}) H ( A F ∞ , ⋄ ) that is the stabilizer of a self-dual lattice of V ∞ , ⋄ V^{\infty,\diamond} V ∞ , ⋄ ,
lim L ⋄ S C ‾ λ ( X L ⋄ ⋅ L ⋄ , L π ) ker χ π ≠ 0 \lim_{L^{\diamond}} \overline{\mathrm{SC}}_{\lambda}(X_{L^{\diamond}\cdot L^{\diamond}},L_{\pi})_{\ker\chi_{\pi}}\ne0 L ⋄ lim SC λ ( X L ⋄ ⋅ L ⋄ , L π ) k e r χ π = 0 holds, where χ π : L π [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] → L π \chi_{\pi}:L_{\pi}[L^{\diamond}\backslash H(\mathbb{A}_{F}^{\infty,\diamond})/L^{\diamond}]\to L_{\pi} χ π : L π [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] → L π is the Satake character of π ∞ , ⋄ \pi^{\infty,\diamond} π ∞ , ⋄ .
Structure of the article
The article consists of five chapters after the introduction and one appendix.
In Chapter 2 2 2 , we collect some notions and facts from representation theory and automorphic forms. In §2.1 2.1 2.1 , we review the construction of algebraic representations of unitary groups of given weights via a way that is symmetric in the standard representation and its dual. In §2.2 2.2 2.2 , we apply the previous construction to hermitian spaces over CM fields and introduce further objects. In §2.3 2.3 2.3 , we review the notion of cohomological automorphic representations of G L m ( A E ) \mathrm{GL}_{m}(\mathbb{A}_{E}) GL m ( A E ) of hermitian type of given weights and their relation with Rankin–Selberg pairs ( σ , χ ) (\sigma,\chi) ( σ , χ ) appearing at the beginning of the introduction when m = 2 m=2 m = 2 . In §2.4 2.4 2.4 , we collect the notation for automorphic forms on quasi-split unitary groups of even ranks. At last, we review the doubling method in the theory of theta lifting and deduce some formulae concerning derivatives of L L L -functions in §2.5 2.5 2.5 .
In Chapter 3 3 3 , we define several key concepts in our theory, including standard (relative) motives, automorphic (relative) motives, motivic special classes, motivic generating functions, and motivic arithmetic theta lifting. In §3.1 3.1 3.1 , we introduce a CM moduli problem that is auxiliary to our construction of the unitary moduli schemes in the next section. In §3.2 3.2 3.2 , we define the unitary moduli schemes over the (abstract) CM field E E E and construct what we called the standard (relative) motive over those moduli schemes. In §3.3 3.3 3.3 , we review the construction of special morphisms for unitary moduli schemes. In §3.4 3.4 3.4 , we construct automorphic (relative) motives over unitary moduli schemes of given weights, using the standard motive constructed previously. We compute the étale realization and Hodge realization of those automorphic motives in §3.5 3.5 3.5 and §3.6 3.6 3.6 , respectively. In §3.7 3.7 3.7 , we define motivic special classes and motivic generating functions, extending Kudla’s special cycles and generating functions in the trivial weight case; we also formulate the conjecture on the modularity of motivic generating functions. At the end, we review the construction of canonical smooth integral models of unitary moduli schemes in the hyperspecial case in §3.8 3.8 3.8 , which will be used later.
In Chapter 4 4 4 , we construct and study a height pairing on motivic cohomology of those automorphic motives constructed in the previous chapter. In §4.1 4.1 4.1 , we define such height pairing via Beilinson’s height pairing on Chow groups (that is, motivic cohomology of the unit motive) with respect to a (nice) auxiliary rational prime ℓ \ell ℓ . In §4.2 4.2 4.2 , we show that such height pairing has a decomposition into local heights for certain representatives of classes, similar to the height pairing on Chow groups. In §4.3 4.3 4.3 , we use the height pairing constructed previously to formulate the motivic arithmetic inner product formula assuming the modularity of generating functions. Finally in §4.4 4.4 4.4 , we show that over unitary moduli curves, the height pairing on motivic cohomology is independent of ℓ \ell ℓ .
In Chapter 5 5 5 , we compute local heights of motivic special classes in various situations. In §5.1 5.1 5.1 , we relate such local height at an inert place to the derivative of Whittaker coefficients of certain Eisenstein series in both smooth and semistable cases, by reducing the computation to the trivial weight case, which is already known by Li–Zhang [22 ] . In §5.2 5.2 5.2 , we show that over a split place, such local height vanishes just like in the trivial weight case. In §5.3 5.3 5.3 , we state our result on local heights on unitary moduli curves at a general nonsplit nonarchimedean place, and prove it in §5.4 5.4 5.4 and §5.5 5.5 5.5 , respectively, according to whether the hermitian space at this place is split or not. In § 5.6 , we collect our results on local heights on unitary moduli curves at an archimedean place from the previous work [28 ] . In the last section § 5.7 , we identify a sufficiently large subspace of nonarchimedean Schwartz functions for which we can show the modularity of motivic doubling kernel functions later; we also reprove the arithmetic inner product formula (for the trivial weight) obtained in [26 ] by a simpler argument.
In Chapter 6, we prove our main results over unitary moduli curves when the weight is balanced and nontrivial. In § 6.1 , we show the modularity of motivic doubling kernel functions defined by height pairing of motivic special classes with respect to the Schwartz functions introduced in § 5.7 . In § 6.2 , we prove the formula computing derivatives of L L L -functions of cohomological automorphic representations of GL 2 ( A E ) \operatorname{GL}_{2}(A_{E}) GL 2 ( A E ) of hermitian type and deduce its corollaries.
In Appendix A, we collect some known facts concerning relative motives. In §A.1, we review the notion of relative motives or, more generally, motivic sheaves, together with their étale and Hodge realizations. In §A.2, we review the motivic Künneth decomposition of abelian schemes and identify its realizations.
Notation and conventions
Generalities:
The imaginary unit is denoted by i \mathrm{i} i ; and the natural constant is denoted by e \mathrm{e} e .
We denote by arg : C × → C × \operatorname{arg}\colon\mathbb{C}^{\times} \to\mathbb{C}^{\times} arg : C × → C × the argument character.
For b ∈ N = { 0 , 1 , 2 , … } b \in\mathbb{N} = \{0,1,2,\ldots\} b ∈ N = { 0 , 1 , 2 , … } , denote by S b \mathfrak{S}_{b} S b the group of permutations on { 1 , … , b } \{1,\ldots,b\} { 1 , … , b } . For a finite tuple b = ( b u ) u b=(b_{u})_{u} b = ( b u ) u of elements in N \mathbb{N} N , put S b : = ∏ u S b u \mathfrak{S}_{b}:=\prod_{u}\mathfrak{S}_{b_{u}} S b := ∏ u S b u .
We use standard notations from the category theory. The category of sets is denoted by Set \operatorname{Set} Set . For a category C \mathrm{C} C , we denote by C o p \mathrm{C}^{\mathrm{op}} C op its opposite category. For another category D \mathrm{D} D , we denote by Fun ( C , D ) \operatorname{Fun}(\mathrm{C},\mathrm{D}) Fun ( C , D ) the category of functors from C \mathrm{C} C to D \mathrm{D} D . In particular, we denote by PC : = Fun ( C o p , Set ) \operatorname{PC}:=\operatorname{Fun}(\mathrm{C}^{\mathrm{op}},\operatorname{Set}) PC := Fun ( C op , Set ) the category of presheaves on C \mathrm{C} C , which contains C \mathrm{C} C as a full subcategory by the Yoneda embedding.
In a category with endofunctors like “Tate twist” − ( n ) -(n) − ( n ) and “degree shift” − [ n ] -[n] − [ n ] for n ∈ Z n \in\mathbb{Z} n ∈ Z , we put − ⟨ n ⟩ : = − ( n ) [ 2 n ] -\langle n\rangle:=-(n)[2n] − ⟨ n ⟩ := − ( n ) [ 2 n ] as usual.
All rings are commutative and unital; and ring homomorphisms preserve units. For a (topological) ring L L L , a (topological) L L L -ring is a (topological) ring R R R together with a (continuous) ring homomorphism from L L L to R R R . However, we use the word algebra in the general sense, which is not necessarily commutative or unital.
If a base ring is not specified in the tensor operation ⊗ \otimes ⊗ , then it is Z \mathbb{Z} Z , unless otherwise noted.
All hermitian spaces (over an étale quadratic extension E / F E/F E / F of rings) are assumed to be of finite rank and nondegenerate. For a hermitian space V V V , we usually denote by ( , ) V (~,~)_{V} ( , ) V its hermitian form. When 2 ∈ F × 2 \in F^{\times} 2 ∈ F × , we have the moment matrix
T ( x ) = ( 1 2 ( x i , x j ) V ) 1 ≤ i , j ≤ r T(x)=\left(\frac{1}{2}(x_{i},x_{j})_{V}\right)_{1\leq i,j\leq r} T ( x ) = ( 2 1 ( x i , x j ) V ) 1 ≤ i , j ≤ r
for a tuple x = ( x 1 , … , x r ) ∈ V r x=(x_{1},\ldots,x_{r})\in V^{r} x = ( x 1 , … , x r ) ∈ V r ; and for an r r r -by-r r r matrix T T T with entries in E E E , put
V T r : = { x ∈ V r ∣ T ( x ) = T } . V_{T}^{r}:=\{x\in V^{r}\mid T(x)=T\}. V T r := { x ∈ V r ∣ T ( x ) = T } .
Geometry:
Sch / T ′ , Sch / T f l . p r o j , Sch / T s m . p r o j , Sch / T s m . q p r o j \operatorname{Sch}'_{/T},\quad\operatorname{Sch}^{\mathrm{fl.proj}}_{/T},\quad\operatorname{Sch}^{\mathrm{sm.proj}}_{/T},\quad\operatorname{Sch}^{\mathrm{sm.qproj}}_{/T} Sch / T ′ , Sch / T fl.proj , Sch / T sm.proj , Sch / T sm.qproj
the full subcategories of locally Noetherian, flat projective, smooth projective, and smooth quasi-projective schemes over T T T , respectively. When T = Spec R T=\operatorname{Spec}R T = Spec R is affine, we also write Sch / R ′ \operatorname{Sch}'_{/R} Sch / R ′ for Sch / T ′ \operatorname{Sch}'_{/T} Sch / T ′ .
For every positive integer r r r , denote by Mat r \operatorname{Mat}_r Mat r the scheme of r r r -by-r r r matrices, GL r ⊆ Mat r \operatorname{GL}_r \subseteq\operatorname{Mat}_r GL r ⊆ Mat r the subscheme of invertible ones, and 1 r ∈ GL r ( Z ) 1_r \in\operatorname{GL}_r(\mathbb{Z}) 1 r ∈ GL r ( Z ) the unit matrix.
Denote by Mfd \operatorname{Mfd} Mfd the category of complex manifolds. For a field k k k together with an embedding u : k → C u:k\to\mathbb{C} u : k → C , we have a functor − u : Sch / k s m , q p r o j → Mfd -^{u}:\operatorname{Sch}^{\mathrm{sm,qproj}}_{/k}\to\operatorname{Mfd} − u : Sch / k sm , qproj → Mfd sending X X X to its underlying complex manifold.
Number Theory: Let K K K be a number field.
Denote by V K \mathcal{V}_K V K the set of all places of K K K . For a finite set ◊ \Diamond ◊ of places of Q \mathbb{Q} Q , we denote by V K ( ◊ ) \mathcal{V}_K^{(\Diamond)} V K ( ◊ ) the subset of V K \mathcal{V}_K V K of places above ◊ \Diamond ◊ ; when ◊ = { ⋄ 1 , ⋄ 2 , … } \Diamond=\{\diamond_1,\diamond_2,\ldots\} ◊ = { ⋄ 1 , ⋄ 2 , … } is written explicitly, we simply write V K ( ⋄ 1 , ⋄ 2 , … ) \mathcal{V}_K^{(\diamond_1,\diamond_2,\ldots)} V K ( ⋄ 1 , ⋄ 2 , … ) instead of V K ( { ⋄ 1 , ⋄ 2 , … } ) \mathcal{V}_K^{(\{\diamond_1,\diamond_2,\ldots\})} V K ({ ⋄ 1 , ⋄ 2 , … }) . Put V K f i n : = V K ∖ V K ( ∞ ) \mathcal{V}^{\mathrm{fin}}_K:=\mathcal{V}_K\setminus\mathcal{V}^{(\infty)}_K V K fin := V K ∖ V K ( ∞ ) .
For a nonarchimedean place v v v of K K K , we write ∥ v ∥ \lVert v\rVert ∥ v ∥ for the cardinality of the residue field of K v K_v K v and p ( v ) p(v) p ( v ) for the underlying rational prime of v v v .
Denote by A \mathbb{A} A the ring of adèles of Q \mathbb{Q} Q , and put A K : = A ⊗ K \mathbb{A}_K:=\mathbb{A}\otimes K A K := A ⊗ K .
Denote by ψ : Q \ A → C × \psi:\mathbb{Q}\backslash\mathbb{A}\to\mathbb{C}^{\times} ψ : Q \ A → C × the standard additive automorphic character, that is, the unique character such that ψ ∞ ( x ) = e 2 π i x \psi_\infty(x)=e^{2\pi i x} ψ ∞ ( x ) = e 2 π i x . Put ψ K : = ψ ∘ Tr K / Q : K \ A K → C × \psi_K:=\psi\circ\operatorname{Tr}_{K/\mathbb{Q}}:K\backslash\mathbb{A}_K\to\mathbb{C}^{\times} ψ K := ψ ∘ Tr K / Q : K \ A K → C × .
We write
K ⋄ : = K ⊗ Q ⋄ K_\diamond:=K\otimes\mathbb{Q}_\diamond K ⋄ := K ⊗ Q ⋄ for a place ⋄ ∈ { ∞ , 2 , 3 , 5 , … } \diamond\in\{\infty,2,3,5,\ldots\} ⋄ ∈ { ∞ , 2 , 3 , 5 , … } of Q \mathbb{Q} Q ,
O K , ( p ) : = O K ⊗ Z ( p ) \mathcal{O}_{K,(\mathit{p})}:=\mathcal{O}_K\otimes\mathbb{Z}_{(p)} O K , ( p ) := O K ⊗ Z ( p ) and O K , p : = O K ⊗ Z p \mathcal{O}_{K,p}:=\mathcal{O}_K\otimes\mathbb{Z}_p O K , p := O K ⊗ Z p for a rational prime p ∈ { 2 , 3 , 5 , … } p\in\{2,3,5,\ldots\} p ∈ { 2 , 3 , 5 , … } ,
more generally, for a finite set ◊ \Diamond ◊ of rational primes, we put O K , ( ◊ ) : = ⋂ p ∈ ◊ O K , ( p ) \mathcal{O}_{K,(\Diamond)}:=\bigcap_{p\in\Diamond}\mathcal{O}_{K,(p)} O K , ( ◊ ) := ⋂ p ∈ ◊ O K , ( p ) (in K K K ).
When K K K is a CM field, we put M ‾ : = M ⊗ K , − K \overline{M}:=M\otimes_{K,-}K M := M ⊗ K , − K for an object M M M in a K K K -linear category, where − - − denotes the complex conjugation.
Notation 1.11. Consider a number field L L L contained in C \mathbb{C} C . For an L ∞ L_\infty L ∞ -linear object V V V (e.g., vector space, category, ...), we denote by
κ : V → V ♭ \kappa:V\to V^\flat κ : V → V ♭ the natural projection map, where V ♭ V^\flat V ♭ is the C \mathbb{C} C -linear factor of V V V corresponding to the natural contraction of coefficients L ∞ = L ⊗ R → C L_\infty=L\otimes\mathbb{R}\to\mathbb{C} L ∞ = L ⊗ R → C .
Definition 1.12. Let K K K be a number field. For a reductive group H H H over K K K and a finite set ◊ \Diamond ◊ of places of K K K including all archimedean ones, we define H k ( H ( A K ◊ ) ) H_k(H(\mathbb{A}_K^\Diamond)) H k ( H ( A K ◊ )) to be the category
whose objects are neat open compact subgroups L L L of H ( A K ◊ ) H(\mathbb{A}_K^\Diamond) H ( A K ◊ ) ; and
in which a morphism from L L L to L ′ L' L ′ is an element h ∈ L \ H ( A K ◊ ) / L ′ h\in L\backslash H(\mathbb{A}_K^\Diamond)/L' h ∈ L \ H ( A K ◊ ) / L ′ satisfying h − 1 L h ⊆ L ′ h^{-1}Lh\subseteq L' h − 1 L h ⊆ L ′ .
Definition 1.13. Let C \mathcal{C} C be a symmetric monoidal category with product ⊗ \otimes ⊗ and unit object 1 \mathbb{1} 1 . Recall that an object A A A of C \mathcal{C} C is dualizable if there exist another object A ∨ ∈ C A^\vee\in\mathcal{C} A ∨ ∈ C together with two maps
ev A : A ⊗ A ∨ → 1 , coev A : 1 → A ∨ ⊗ A , \operatorname{ev}_A:A\otimes A^\vee\to\mathbb{1},\qquad\operatorname{coev}_A:\mathbb{1}\to A^\vee\otimes A, ev A : A ⊗ A ∨ → 1 , coev A : 1 → A ∨ ⊗ A , called the evaluation and coevaluation maps , respectively, satisfying that the composite maps
A → 1 A ⊗ coev A A ⊗ A ∨ ⊗ A → ev A ⊗ 1 A A , A ∨ → coev A ⊗ 1 A ∨ A ∨ ⊗ A ⊗ A ∨ → 1 A ∨ ⊗ ev A A ∨ A\xrightarrow{1_A\otimes\operatorname{coev}_A}A\otimes A^\vee\otimes A\xrightarrow{\operatorname{ev}_A\otimes1_A}A,\qquad A^\vee\xrightarrow{\operatorname{coev}_A\otimes1_{A^\vee}}A^\vee\otimes A\otimes A^\vee\xrightarrow{1_{A^\vee}\otimes\operatorname{ev}_A}A^\vee A 1 A ⊗ coev A A ⊗ A ∨ ⊗ A ev A ⊗ 1 A A , A ∨ coev A ⊗ 1 A ∨ A ∨ ⊗ A ⊗ A ∨ 1 A ∨ ⊗ ev A A ∨ are both identity maps. The data A ∨ A^\vee A ∨ , ev A \operatorname{ev}_A ev A , coev A \operatorname{coev}_A coev A , if exist, are unique up to unique isomorphisms. When C \mathcal{C} C is closed, a map A ⊗ A ∨ → 1 A\otimes A^\vee\to\mathbb{1} A ⊗ A ∨ → 1 is an evaluation map if and only if the induced maps
A → Hom ( A ∨ , 1 ) , A ∨ → Hom ( A , 1 ) A\to\operatorname{Hom}(A^\vee,\mathbb{1}),\qquad A^\vee\to\operatorname{Hom}(A,\mathbb{1}) A → Hom ( A ∨ , 1 ) , A ∨ → Hom ( A , 1 ) are both isomorphisms.
Definition 1.14. Denote by Cat \operatorname{Cat} Cat the ( 2 , 1 ) (2,1) ( 2 , 1 ) -category of categories. Consider an (ordinary) category C \mathcal{C} C and a ( 2 , 1 ) (2,1) ( 2 , 1 ) -functor F : C o p → Cat F:\mathcal{C}^{\mathrm{op}}\to\operatorname{Cat} F : C op → Cat . For every subcategory C ′ \mathcal{C}' C ′ of C \mathcal{C} C ,
we denote by F ( C ′ ) \mathrm{F}(\mathcal{C}') F ( C ′ ) the category whose objects are pairs ( c , A c ) (c,A_c) ( c , A c ) with c c c an object in C ′ \mathcal{C}' C ′ and A c A_c A c an object in F ( c ) \mathrm{F}(c) F ( c ) and morphisms from ( c , A c ) (c,A_c) ( c , A c ) to ( c ′ , A c ′ ′ ) (c',A_{c'}') ( c ′ , A c ′ ′ ) are pairs ( f , A f ) (f,A_f) ( f , A f ) with f : c → c ′ f:c\to c' f : c → c ′ an edge in C ′ \mathcal{C}' C ′ and A f : A c → F ( f ) A c ′ ′ A_f:A_c\to\mathrm{F}(f)A_{c'}' A f : A c → F ( f ) A c ′ ′ an edge in F ( c ) \mathrm{F}(c) F ( c ) ;7
we say that an edge ( f , A f ) (f,A_f) ( f , A f ) in F ( C ′ ) \mathrm{F}(\mathcal{C}') F ( C ′ ) is Cartesian if A f A_f A f is an isomorphism;
we say that a functor D → F ( C ′ ) \mathcal{D}\to\mathrm{F}(\mathcal{C}') D → F ( C ′ ) is Cartesian if the image of every edge is Cartesian.
Acknowledgements. The author would like to thank Fangzhou Jin, Andreas Mihatsch, and Lei Wu for helpful discussion. The author is partially supported by the National Key R&D Program of China No. 2022YFA1005300.
AI disclosure. AI was not used in this work.
Preliminaries on representation theory
In this chapter, we collect some notions and facts from representation theory and automorphic forms. Let r r r and m m m be positive integers.
Algebraic representations of unitary groups
We start from the notation of weights.
Notation 2.1. Let m ≥ 0 m \ge0 m ≥ 0 be an integer.
Denote by W m \mathbb{W}_m W m the set of tuple λ = ( b 1 , b 2 , … , b m ) \lambda=(b_1,b_2,\ldots,b_m) λ = ( b 1 , b 2 , … , b m ) of non-increasing integers. For λ ∈ W m \lambda\in\mathbb{W}_m λ ∈ W m , put b ( λ ) : = ∑ i = 1 m b i b(\lambda):=\sum_{i=1}^{m}b_i b ( λ ) := ∑ i = 1 m b i . We denote the zero tuple by 0 \mathbf{0} 0 (when the length m m m is clear).
Denote by λ ‾ : W m → W m \overline{\phantom{\lambda}}:\mathbb{W}_m\to\mathbb{W}_m λ : W m → W m the involution sending λ \lambda λ to λ ‾ : = ( − b m , … , − b 2 , − b 1 ) \overline{\lambda}:=(-b_m,\ldots,-b_2,-b_1) λ := ( − b m , … , − b 2 , − b 1 ) .
For λ ∈ W m \lambda\in\mathbb{W}_m λ ∈ W m , denote by 0 ≤ s = s ( λ ) ≤ m 0\le s=s(\lambda)\le m 0 ≤ s = s ( λ ) ≤ m the largest integer such that b s > 0 b_s>0 b s > 0 (s = 0 s=0 s = 0 if b 1 ≤ 0 b_1\le0 b 1 ≤ 0 ).
Denote by W m + \mathbb{W}_m^+ W m + the subset of W m \mathbb{W}_m W m consisting of λ \lambda λ such that b m ≥ 0 b_m\ge0 b m ≥ 0 . We will adopt the convention that an element λ ∈ W m + \lambda\in\mathbb{W}_m^+ λ ∈ W m + is naturally an element in W r + \mathbb{W}_r^+ W r + for every integer r ≥ m r\ge m r ≥ m by adding zeros at the end.
For λ ∈ W m \lambda\in\mathbb{W}_m λ ∈ W m , put λ + : = ( b 1 , … , b s ) ∈ W s + \lambda^+:=(b_1,\ldots,b_s)\in\mathbb{W}_s^+ λ + := ( b 1 , … , b s ) ∈ W s + with s = s ( λ ) s=s(\lambda) s = s ( λ ) .
Let E / F E/F E / F be an étale quadratic extension of rings with x ‾ ∈ Aut ( E / F ) \overline{\phantom{x}}\in\operatorname{Aut}(E/F) x ∈ Aut ( E / F ) the nontrivial involution. We first recall the construction of algebraic representations of general linear groups.
Construction 2.2. Consider a free F F F -module U U U of rank r r r and an element λ ∈ W r + \lambda\in\mathbb{W}_r^+ λ ∈ W r + such that b ! b! b ! is invertible in F F F , where b : = b ( λ ) b:=b(\lambda) b := b ( λ ) . Let π λ ∈ Z [ 1 / b ! ] [ S b ] \pi_\lambda\in\mathbb{Z}[1/b!][S_b] π λ ∈ Z [ 1/ b !] [ S b ] be the idempotent corresponding to the standard filling that assigns 1 , 2 , … , b 1,2,\ldots,b 1 , 2 , … , b from left to right starting with the first row then moving to the second row, etc (see, for example, [12 §4.2] ), which gives rise to a direct summand
S λ ( U ) : = π λ ( T b ( U ) ) S_\lambda(U):=\pi_\lambda\left(T^b(U)\right) S λ ( U ) := π λ ( T b ( U ) )
of T b ( U ) : = U ⊗ b T^b(U):=U^{\otimes b} T b ( U ) := U ⊗ b of F [ G L F ( U ) ] F[\mathrm{GL}_F(U)] F [ GL F ( U )] -modules. In particular, when F F F is a field of characteristic zero, S λ ( U ) S_\lambda(U) S λ ( U ) is an irreducible representation of G L F ( U ) ≃ G L r ( F ) \mathrm{GL}_F(U)\simeq\mathrm{GL}_r(F) GL F ( U ) ≃ GL r ( F ) of highest weight λ \lambda λ .
In general, for a finite set U \mathcal{U} U and a tuple λ = ( λ u ) u ∈ U ∈ ( W r + ) U \lambda=(\lambda_u)_{u\in\mathcal{U}}\in(\mathbb{W}_r^+)^{\mathcal{U}} λ = ( λ u ) u ∈ U ∈ ( W r + ) U such that b u ! b_u! b u ! (b u : = b ( λ u ) b_u:=b(\lambda_u) b u := b ( λ u ) ) is invertible in F F F for every u ∈ U u\in\mathcal{U} u ∈ U , we have
T b ( U ) : = ⨂ u ∈ U T b u ( U ) T^b(U):=\bigotimes_{u\in\mathcal{U}}T^{b_u}(U) T b ( U ) := u ∈ U ⨂ T b u ( U )
(tensor over F F F ) and similarly S λ ( U ) S_\lambda(U) S λ ( U ) , together with the projector
π λ : = ⨂ u ∈ U π λ u . \pi_\lambda:=\bigotimes_{u\in\mathcal{U}}\pi_{\lambda_u}. π λ := u ∈ U ⨂ π λ u .
We then recall the construction of algebraic representations of unitary groups.
Construction 2.3. Consider a (nondegenerate) hermitian space V V V over E / F E/F E / F of rank m m m and an F F F -ring L \mathbb{L} L such that the set U \mathcal{U} U of F F F -ring homomorphisms from E E E to L \mathbb{L} L is nonempty (hence consists of two elements). Consider a pair λ ∈ W m U \lambda\in\mathcal{W}^{\mathcal{U}}_{m} λ ∈ W m U satisfying λ u ‾ = λ u ‾ \lambda_{\overline{u}}=\overline{\lambda_{u}} λ u = λ u for every u ∈ U u \in\mathcal{U} u ∈ U . Put b = ( b u : = b ( λ u + ) ) u ∈ U ∈ N U \mathbf{b}=(b_{u}:=b(\lambda^{+}_{u}))_{u\in\mathcal{U}}\in\mathbb{N}^{\mathcal{U}} b = ( b u := b ( λ u + ) ) u ∈ U ∈ N U and
B λ : = m min u ∈ U { b u } ⋅ ∏ u ∈ U b u ! . B_{\lambda}:=m^{\min_{u\in\mathcal{U}}\{b_{u}\}}\cdot\prod_{u\in\mathcal{U}}b_{u}!. B λ := m m i n u ∈ U { b u } ⋅ u ∈ U ∏ b u ! .
Assume that B λ B_{\lambda} B λ is invertible in L \mathbb{L} L . Put
T b ( V ) : = ⨂ u ∈ U T b u ( V u ) T_{\mathbf{b}}(V):=\bigotimes_{u\in\mathcal{U}}T^{b_{u}}(V_{u}) T b ( V ) := u ∈ U ⨂ T b u ( V u )
(tensor over L \mathbb{L} L ) as an algebraic representation of U ( V ) \mathrm{U}(V) U ( V ) over L \mathbb{L} L , where U ( V ) \mathrm{U}(V) U ( V ) acts on V u : = V ⊗ E , u L V_{u}:=V\otimes_{E,u}\mathbb{L} V u := V ⊗ E , u L via the homomorphism u u u . For every pair I = ( i u ) u I=(i_{u})_{u} I = ( i u ) u with 1 ≤ i u ≤ b u 1\leq i_{u}\leq b_{u} 1 ≤ i u ≤ b u , we have the contraction operator
C I : T b ( V ) ⟶ T b − 1 ( V ) C_{I}:T_{\mathbf{b}}(V)\longrightarrow T_{\mathbf{b}-1}(V) C I : T b ( V ) ⟶ T b − 1 ( V )
by “pairing” the i u i_{u} i u -th factor of T b u ( V u ) T^{b_{u}}(V_{u}) T b u ( V u ) for the two elements u ∈ U u\in\mathcal{U} u ∈ U , and the expansion operator
A I : T b − 1 ( V ) ⟶ T b ( V ) A_{I}:T_{\mathbf{b}-1}(V)\longrightarrow T_{\mathbf{b}}(V) A I : T b − 1 ( V ) ⟶ T b ( V )
by inserting the coevaluation map L ⟶ ⨂ u ∈ U V u \mathbb{L}\longrightarrow\bigotimes_{u\in\mathcal{U}}V_{u} L ⟶ ⨂ u ∈ U V u (Definition 1.13) to the i u i_{u} i u -th spot of T b u ( V u ) T^{b_{u}}(V_{u}) T b u ( V u ) for u ∈ U u\in\mathcal{U} u ∈ U . If we put
T [ b ] ( V ) : = ⋂ I ker C I , T_{[\mathbf{b}]}(V):=\bigcap_{I}\ker C_{I}, T [ b ] ( V ) := I ⋂ ker C I ,
then
T b ( V ) = T [ b ] ( V ) ⊕ ( ⨁ r = 1 min u ∈ U { b u } T [ b ] ( V ) b − r ) . (1) T_{\mathbf{b}}(V)=T_{[\mathbf{b}]}(V)\oplus\left(\bigoplus_{r=1}^{\min_{u\in\mathcal{U}}\{b_{u}\}}T_{[\mathbf{b}]}(V)_{\mathbf{b}-r}\right). \tag*{(1)} T b ( V ) = T [ b ] ( V ) ⊕ r = 1 ⨁ m i n u ∈ U { b u } T [ b ] ( V ) b − r . (1)
where
T [ b ] ( V ) b − r : = ∑ I 1 ∘ ⋯ ∘ I r A I 1 ∘ ⋯ ∘ A I r ( T [ b − r ] ( V ) ) . T_{[\mathbf{b}]}(V)_{\mathbf{b}-r}:=\sum_{I_{1}\circ\cdots\circ I_{r}}A_{I_{1}}\circ\cdots\circ A_{I_{r}}\left(T_{[\mathbf{b}-r]}(V)\right). T [ b ] ( V ) b − r := I 1 ∘ ⋯ ∘ I r ∑ A I 1 ∘ ⋯ ∘ A I r ( T [ b − r ] ( V ) ) .
We call T [ b ] ( V ) T_{[\mathbf{b}]}(V) T [ b ] ( V ) the harmonic direct summand of T b ( V ) T_{\mathbf{b}}(V) T b ( V ) .
It is clear that the decomposition (1) is stable under the action of both U ( V ) \mathrm{U}(V) U ( V ) and S b \mathfrak{S}_{\mathbf{b}} S b . In particular, we may define
S [ λ ] ( V ) : = ( ⨂ u ∈ U π λ u + ) ( T [ b ] ( V ) ) S_{[\lambda]}(V):=\left(\bigotimes_{u\in\mathcal{U}}\pi_{\lambda_{u}^{+}}\right)\left(T_{[\mathbf{b}]}(V)\right) S [ λ ] ( V ) := ( u ∈ U ⨂ π λ u + ) ( T [ b ] ( V ) )
(Construction 2.2 with r r r replaced by m m m ), which is a direct summand of the L [ U ( V ) ] \mathbb{L}[\mathrm{U}(V)] L [ U ( V )] -module T [ b ] ( V ) T_{[\mathbf{b}]}(V) T [ b ] ( V ) hence T b ( V ) T^{\mathbf{b}}(V) T b ( V ) .
Remark 2.4. Suppose that L \mathbb{L} L is a field of characteristic zero. Take an element u ∈ U u\in\mathcal{U} u ∈ U . Put V u : = V ⊗ E , u L V_{u}:=V\otimes_{E,u}\mathbb{L} V u := V ⊗ E , u L and similarly for V u ‾ V_{\overline{u}} V u .
The element u ∈ U u\in\mathcal{U} u ∈ U induces a homomorphism U ( V ) → G L ( V u ) ≃ G L m ( L ) \mathrm{U}(V)\to\mathrm{GL}(V_{u})\simeq\mathrm{GL}_{m}(\mathbb{L}) U ( V ) → GL ( V u ) ≃ GL m ( L ) ; and it is well-known that S [ λ ] ( V ) S_{[\lambda]}(V) S [ λ ] ( V ) extends by scalar to an irreducible representation of G L m ( L ) \mathrm{GL}_{m}(\mathbb{L}) GL m ( L ) of highest weight λ u \lambda_{u} λ u .
The hermitian form induces an isomorphism V u ‾ ≃ V u ∨ V_{\overline{u}}\simeq V_{u}^{\vee} V u ≃ V u ∨ . The map ( ⋀ m V u ) ⊗ L V u ∨ → ⋀ m − 1 V u (\bigwedge^{m}V_{u})\otimes_{\mathbb{L}}V_{u}^{\vee}\to\bigwedge^{m-1}V_{u} ( ⋀ m V u ) ⊗ L V u ∨ → ⋀ m − 1 V u given by pairing the last two factors induces an isomorphism V u ∨ ≃ ( ⋀ m − 1 V u ) ⊗ L ( ⋀ m V u ) ∨ V_{u}^{\vee}\simeq(\bigwedge^{m-1}V_{u})\otimes_{\mathbb{L}}(\bigwedge^{m}V_{u})^{\vee} V u ∨ ≃ ( ⋀ m − 1 V u ) ⊗ L ( ⋀ m V u ) ∨ , which further induces an isomorphism
T b ( V u ) ≃ T b u ( V u ) ⊗ L T b u ‾ ( ⋀ m − 1 V u ) ⊗ L ( ⋀ m V u ) − b u ‾ . T_{\mathbf{b}}(V_{u})\simeq T^{b_{u}}(V_{u})\otimes_{\mathbb{L}}T^{b_{\overline{u}}}\left(\bigwedge^{m-1}V_{u}\right)\otimes_{\mathbb{L}}\left(\bigwedge^{m}V_{u}\right)^{-b_{\overline{u}}}. T b ( V u ) ≃ T b u ( V u ) ⊗ L T b u ( ⋀ m − 1 V u ) ⊗ L ( ⋀ m V u ) − b u .
In particular, T b ( V u ) T_{\mathbf{b}}(V_{u}) T b ( V u ) is canonically a subspace of T b u + ( m − 1 ) b u ‾ ( V u ) ⊗ L ( ⋀ m V u ) − b u ‾ T^{b_{u}+(m-1)b_{\overline{u}}}(V_{u})\otimes_{\mathbb{L}}(\bigwedge^{m}V_{u})^{-b_{\overline{u}}} T b u + ( m − 1 ) b u ( V u ) ⊗ L ( ⋀ m V u ) − b u , in which S b u + ( m − 1 ) b u ‾ \mathfrak{S}_{b_{u}+(m-1)b_{\overline{u}}} S b u + ( m − 1 ) b u acts on the latter via the first tensor factor. There exists a unique idempotent δ λ u , λ u ‾ ∈ Q [ S b u + ( m − 1 ) b u ‾ ] \delta_{\lambda_{u},\lambda_{\overline{u}}}\in\mathbb{Q}[\mathfrak{S}_{b_{u}+(m-1)b_{\overline{u}}}] δ λ u , λ u ∈ Q [ S b u + ( m − 1 ) b u ] , depending only on the ordered pair ( λ u , λ u ‾ ) (\lambda_{u},\lambda_{\overline{u}}) ( λ u , λ u ) , such that T [ λ ] ( V ) T_{[\lambda]}(V) T [ λ ] ( V ) is the kernel of δ λ u , λ u ‾ \delta_{\lambda_{u},\lambda_{\overline{u}}} δ λ u , λ u .
Construction 2.5. Let the situation be as in Construction 2.3 and r r r a positive integer satisfying r ≥ s ( λ u ) r \ge s(\lambda_u) r ≥ s ( λ u ) for every u ∈ U u \in\mathcal{U} u ∈ U . For every tuple x = ( x 1 , … , x r ) ∈ V r x=(x_1,\ldots,x_r)\in V^r x = ( x 1 , … , x r ) ∈ V r , we construct a map
μ [ λ ] ( x ) : S [ λ ] ( V ) ∨ → S λ + ( L r ) ∨ \mu_{[\lambda]}(x): S_{[\lambda]}(V)^{\vee}\to S_{\lambda_+}(L^r)^{\vee} μ [ λ ] ( x ) : S [ λ ] ( V ) ∨ → S λ + ( L r ) ∨
(notation from Construction 2.2) as follows. Dualizing the r r r components of x x x gives an E E E -linear map
μ ( x ) : V ∨ → ( E ∨ ) r , \mu(x): V^{\vee}\to(E^{\vee})^r, μ ( x ) : V ∨ → ( E ∨ ) r ,
so that we have the induced map μ ( x ) u : V u ∨ → ( L u ∨ ) r \mu(x)_u:V_u^{\vee}\to(L_u^{\vee})^r μ ( x ) u : V u ∨ → ( L u ∨ ) r for each u ∈ U u\in\mathcal{U} u ∈ U . Taking tensor products, we have the L L L -linear map
μ b ( x ) : T b ( V ) ∨ → T b ( ( L r ) ∨ ) = T b ( ( L r ) ∨ ) . \mu_b(x):T_b(V)^{\vee}\to T^b((L^r)^{\vee})=T^b((L^r)^{\vee}). μ b ( x ) : T b ( V ) ∨ → T b (( L r ) ∨ ) = T b (( L r ) ∨ ) .
Composing with the dual of the natural map T b ( V ) → T [ b ] ( V ) T_b(V)\to T_{[b]}(V) T b ( V ) → T [ b ] ( V ) , we obtain the map
μ [ b ] ( x ) : T [ b ] ( V ) ∨ → T b ( L r ) ∨ , \mu_{[b]}(x):T_{[b]}(V)^{\vee}\to T^b(L^r)^{\vee}, μ [ b ] ( x ) : T [ b ] ( V ) ∨ → T b ( L r ) ∨ ,
which is S b S_b S b -equivariant. The desired map μ [ λ ] ( x ) \mu_{[\lambda]}(x) μ [ λ ] ( x ) is obtained by applying π λ + \pi_{\lambda_+} π λ + on both sides.
Hermitian spaces over CM extensions
From now to the end of this article, we fix a CM extension E / F E/F E / F of number fields.
Notation 2.6. The following notation will be used throughout the article.
Denote by E ⊂ C \mathbb{E}\subset\mathbb{C} E ⊂ C the normal closure of E E E in C \mathbb{C} C , so that Hom ( E , C ) = Hom ( E , E ) \operatorname{Hom}(E,\mathbb{C})=\operatorname{Hom}(E,\mathbb{E}) Hom ( E , C ) = Hom ( E , E ) .
Denote by D E ∈ N D_E\in\mathbb{N} D E ∈ N the absolute discriminant of E E E .
Denote by χ E / F : F × \ A F × → { ± 1 } \chi_{E/F}:F^{\times}\backslash\mathbb{A}_F^{\times}\to\{\pm1\} χ E / F : F × \ A F × → { ± 1 } the quadratic character associated with the extension E / F E/F E / F . For every place v v v of F F F , define a local L L L -factor
b m , v ( s ) : = ∏ i = 1 m L ( 2 s + i , ( χ E / F ) v i ) . b_{m,v}(s):=\prod_{i=1}^{m}L(2s+i,(\chi_{E/F})_v^i). b m , v ( s ) := i = 1 ∏ m L ( 2 s + i , ( χ E / F ) v i ) .
For every place v v v of F F F , put E v : = E ⊗ F F v E_v:=E\otimes_F F_v E v := E ⊗ F F v and O E v : = O F v ⊗ O F O E \mathcal{O}_{E_v}:=\mathcal{O}_{F_v}\otimes_{\mathcal{O}_F}\mathcal{O}_E O E v := O F v ⊗ O F O E .
For every place v v v of F F F , define
d ( v ) : = { 1 , v is split in E ; 2 , v is nonsplit in E . d(v):=
\begin{cases}
1, & v\text{ is split in }E;\\
2, & v\text{ is nonsplit in }E.
\end{cases} d ( v ) := { 1 , 2 , v is split in E ; v is nonsplit in E .
For every finite set ⋄ \diamond ⋄ of places of F F F , put d ( ⋄ ) : = ∑ v ∈ ⋄ d ( v ) d(\diamond):=\sum_{v\in\diamond}d(v) d ( ⋄ ) := ∑ v ∈ ⋄ d ( v ) .
For every prime v v v of F F F , denote by p v \mathfrak{p}_v p v the maximal ideal of O F v \mathcal{O}_{F_v} O F v .
Denote by Herm r ⊂ Res O E / O F Mat r \operatorname{Herm}_r\subset\operatorname{Res}_{\mathcal{O}_E/\mathcal{O}_F}\operatorname{Mat}_r Herm r ⊂ Res O E / O F Mat r the O F \mathcal{O}_F O F -scheme of r r r -by-r r r hermitian matrices with respect to O E / O F \mathcal{O}_E/\mathcal{O}_F O E / O F . Put Herm r ∘ : = Herm r ∩ Res O E / O F GL r \operatorname{Herm}_r^{\circ}:=\operatorname{Herm}_r\cap\operatorname{Res}_{\mathcal{O}_E/\mathcal{O}_F}\operatorname{GL}_r Herm r ∘ := Herm r ∩ Res O E / O F GL r .
Denote by Herm r ( F ) + ⊂ Herm r ( F ) \operatorname{Herm}_r(F)^+\subset\operatorname{Herm}_r(F) Herm r ( F ) + ⊂ Herm r ( F ) the subset of totally semi-positive definite hermitian matrices, and put Herm r ∘ ( F ) + : = Herm r ( F ) + ∩ Herm r ∘ ( F ) \operatorname{Herm}_r^{\circ}(F)^+:=\operatorname{Herm}_r(F)^+\cap\operatorname{Herm}_r^{\circ}(F) Herm r ∘ ( F ) + := Herm r ( F ) + ∩ Herm r ∘ ( F ) .
Definition 2.7. An E E E -weight of rank m m m is a tuple λ = ( λ u ) u ∈ W m Hom ( E , C ) \lambda=(\lambda_u)_u\in\mathcal{W}_m^{\operatorname{Hom}(E,\mathbb{C})} λ = ( λ u ) u ∈ W m Hom ( E , C ) .
We say that an E E E -weight λ \lambda λ is hermitian (with respect to E / F E/F E / F ) if λ u ˉ = λ u ‾ \lambda_{\bar{u}}=\overline{\lambda_u} λ u ˉ = λ u holds for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) .
We say that an E E E -weight λ \lambda λ is r r r -balanced (for an integer r r r ) if s ( λ u ) ≤ r s(\lambda_u)\le r s ( λ u ) ≤ r for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , and is balanced if it is ⌊ m / 2 ⌋ \lfloor m/2\rfloor ⌊ m /2 ⌋ -balanced where m m m is the rank of λ \lambda λ .
We denote by E λ ⊆ C E^\lambda\subseteq\mathbb{C} E λ ⊆ C the field of definition of λ \lambda λ , that is, the smallest subfield of C \mathbb{C} C such that Aut ( C / E λ ) \operatorname{Aut}(\mathbb{C}/E^\lambda) Aut ( C / E λ ) acts trivially on λ \lambda λ with respect to the natural action of Aut ( C / Q ) \operatorname{Aut}(\mathbb{C}/\mathbb{Q}) Aut ( C / Q ) on the set of E E E -weights of rank m m m ; E λ E^\lambda E λ is indeed contained in E \mathbb{E} E .
In what follows, we write λ v \lambda_v λ v for the pair ( λ u ) v ( u ) = v (\lambda_u)_{v(u)=v} ( λ u ) v ( u ) = v for every v ∈ W F ( ∞ ) v\in\mathcal{W}_F^{(\infty)} v ∈ W F ( ∞ ) .
Take a hermitian E E E -weight λ \lambda λ of rank m m m that is r r r -balanced. Put
B λ : = ∏ v ∈ B F ( ∞ ) B λ v (2) B_{\lambda}:=\prod_{v\in\mathcal{B}_{F}^{(\infty)}}B_{\lambda_{v}} \tag*{(2)} B λ := v ∈ B F ( ∞ ) ∏ B λ v (2)
(see Construction 2.3).
Construction 2.8. Consider a Z [ ( D E B λ ) − 1 ] \mathbb{Z}[(D_{E}B_{\lambda})^{-1}] Z [( D E B λ ) − 1 ] -ring R R R and a hermitian space V V V over O E ⊗ R / O F ⊗ R \mathcal{O}_{E}\otimes R/\mathcal{O}_{F}\otimes R O E ⊗ R / O F ⊗ R of rank m m m . In (2–4) below, we further assume that [ E : E ι ] [\mathbb{E}:\mathbb{E}^{\iota}] [ E : E ι ] is invertible on R R R .
(1) Put
S [ λ ] ( V ) : = ⨂ v ∈ B F ( ∞ ) S [ λ v ] ( V ) S_{[\lambda]}(V):=\bigotimes_{v\in\mathcal{B}_{F}^{(\infty)}}S_{[\lambda_{v}]}(V) S [ λ ] ( V ) := v ∈ B F ( ∞ ) ⨂ S [ λ v ] ( V ) (tensor over O E ⊗ R ) a s a n \mathcal{O}_{E}\otimes R) as an O E ⊗ R ) a s an (\mathcal{O}_{E}\otimes R)[U(V)(\mathcal{O}_{F}\otimes R)]$-module, where S [ λ v ] ( V ) S_{[\lambda_{v}]}(V) S [ λ v ] ( V ) is from Construction 2.3 with E / F = O E ⊗ R / O F ⊗ R E/F=\mathcal{O}_{E}\otimes R/\mathcal{O}_{F}\otimes R E / F = O E ⊗ R / O F ⊗ R and L = O E ⊗ R \mathbb{L}=\mathcal{O}_{E}\otimes R L = O E ⊗ R (as an O F ⊗ R \mathcal{O}_{F}\otimes R O F ⊗ R -ring via v v v ).
(2) For every γ ∈ Gal ( E / Q ) \gamma\in\operatorname{Gal}(E/\mathbb{Q}) γ ∈ Gal ( E / Q ) , its action on O E \mathcal{O}_{E} O E induces an isomorphism
γ [ λ v ] : S [ λ v ] ( V ) ⊗ O E , γ O E → ∼ S [ ( γ λ ) γ v ] ( V ) \gamma_{[\lambda_{v}]}:S_{[\lambda_{v}]}(V)\otimes_{\mathcal{O}_{E,\gamma}}\mathcal{O}_{E}\xrightarrow{\sim}S_{[(\gamma\lambda)_{\gamma v}]}(V) γ [ λ v ] : S [ λ v ] ( V ) ⊗ O E , γ O E ∼ S [( γ λ ) γ v ] ( V ) for every v ∈ B F ( ∞ ) v\in\mathcal{B}_{F}^{(\infty)} v ∈ B F ( ∞ ) , which further induces an isomorphism
γ [ λ ] : S [ λ ] ( V ) ⊗ O E , γ O E → ∼ S [ γ λ ] ( V ) \gamma_{[\lambda]}:S_{[\lambda]}(V)\otimes_{\mathcal{O}_{E,\gamma}}\mathcal{O}_{E}\xrightarrow{\sim}S_{[\gamma\lambda]}(V) γ [ λ ] : S [ λ ] ( V ) ⊗ O E , γ O E ∼ S [ γ λ ] ( V ) by taking tensor product over all v ∈ B F ( ∞ ) v\in\mathcal{B}_{F}^{(\infty)} v ∈ B F ( ∞ ) . In particular, if γ ∈ Gal ( E / E ι ) \gamma\in\operatorname{Gal}(E/E^{\iota}) γ ∈ Gal ( E / E ι ) , then γ [ λ ] \gamma_{[\lambda]} γ [ λ ] is an O E ι \mathcal{O}_{E^{\iota}} O E ι -linear automorphism of S [ λ ] ( V ) S_{[\lambda]}(V) S [ λ ] ( V ) , from which we obtain an O E ι \mathcal{O}_{E^{\iota}} O E ι -linear action of Gal ( E / E ι ) \operatorname{Gal}(E/E^{\iota}) Gal ( E / E ι ) on S [ λ ] ( V ) S_{[\lambda]}(V) S [ λ ] ( V ) . Define
V λ : = S [ λ ] ( V ) Gal ( E / E ι ) V_{\lambda}:=S_{[\lambda]}(V)^{\operatorname{Gal}(E/E^{\iota})} V λ := S [ λ ] ( V ) Gal ( E / E ι ) as an ( O E ι ⊗ R ) [ U ( V ) ( O F ⊗ R ) ] (\mathcal{O}_{E^{\iota}}\otimes R)[U(V)(\mathcal{O}_{F}\otimes R)] ( O E ι ⊗ R ) [ U ( V ) ( O F ⊗ R )] -module, so that S [ λ ] ( V ) = V λ ⊗ O E ι O E S_{[\lambda]}(V)=V_{\lambda}\otimes_{\mathcal{O}_{E^{\iota}}}\mathcal{O}_{E} S [ λ ] ( V ) = V λ ⊗ O E ι O E .
(3) Take an element x ∈ V ∨ x\in V^{\vee} x ∈ V ∨ . We have the map
μ [ λ ] ( x ) : S [ λ ] ( V ) ∨ ⟶ S λ + ( ( O E ⊗ R ) r ) ∨ \mu_{[\lambda]}(x):S_{[\lambda]}(V)^{\vee}\longrightarrow S_{\lambda+}((\mathcal{O}_{E}\otimes R)^{r})^{\vee} μ [ λ ] ( x ) : S [ λ ] ( V ) ∨ ⟶ S λ + (( O E ⊗ R ) r ) ∨ by taking the tensor product of μ [ λ v ] ( x ) \mu_{[\lambda_{v}]}(x) μ [ λ v ] ( x ) over all v ∈ B F ( ∞ ) v\in\mathcal{B}_{F}^{(\infty)} v ∈ B F ( ∞ ) , where
μ [ λ v ] ( x ) : S [ λ v ] ( V ) ∨ ⟶ S λ v + ( ( O E ⊗ R ) r ) ∨ \mu_{[\lambda_{v}]}(x):S_{[\lambda_{v}]}(V)^{\vee}\longrightarrow S_{\lambda_{v}+}((\mathcal{O}_{E}\otimes R)^{r})^{\vee} μ [ λ v ] ( x ) : S [ λ v ] ( V ) ∨ ⟶ S λ v + (( O E ⊗ R ) r ) ∨ is the map from Construction 2.5 with E / F = O E ⊗ R / O F ⊗ R E/F=\mathcal{O}_{E}\otimes R/\mathcal{O}_{F}\otimes R E / F = O E ⊗ R / O F ⊗ R and L = O E ⊗ R \mathbb{L}=\mathcal{O}_{E}\otimes R L = O E ⊗ R . It is straightforward to show the map μ [ λ ] ( x ) \mu_{[\lambda]}(x) μ [ λ ] ( x ) is Gal ( E / E ι ) \operatorname{Gal}(E/E^{\iota}) Gal ( E / E ι ) -equivariant, hence restricts to a map
μ λ ( x ) : ( V λ ) ∨ ⟶ S λ + ( ( O E ι ⊗ R ) r ) ∨ . \mu_{\lambda}(x):(V_{\lambda})^{\vee}\longrightarrow S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee}. μ λ ( x ) : ( V λ ) ∨ ⟶ S λ + (( O E ι ⊗ R ) r ) ∨ . Finally, denote by
μ λ ( x ) : S λ + ( ( O E ι ⊗ R ) r ) ⟶ V λ \mu^{\lambda}(x):S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})\longrightarrow V_{\lambda} μ λ ( x ) : S λ + (( O E ι ⊗ R ) r ) ⟶ V λ the dual of μ λ ( x ) \mu_{\lambda}(x) μ λ ( x ) .
(4) We define a function Ψ V 2 r λ \Psi_{V^{2r}}^{\lambda} Ψ V 2 r λ on V 2 r V^{2r} V 2 r valued in S λ + ( ( O E ι ⊗ R ) r ) ∨ ⊗ O E ι ⊗ R S λ + ( ( O E ι ⊗ R ) r ) ∨ S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee}\otimes_{\mathcal{O}_{E^{\iota}}\otimes R}S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee} S λ + (( O E ι ⊗ R ) r ) ∨ ⊗ O E ι ⊗ R S λ + (( O E ι ⊗ R ) r ) ∨ such that for every x = ( x 1 , x 2 ) ∈ V 2 r = ( V r ) 2 x=(x_{1},x_{2})\in V^{2r}=(V^{r})^{2} x = ( x 1 , x 2 ) ∈ V 2 r = ( V r ) 2 ,
Ψ V 2 r λ ( x ) = μ λ ( x 2 ) ∘ μ λ ( x 1 ) ∈ Hom O E ι ⊗ R ( S λ + ( ( O E ι ⊗ R ) r ) , S λ + ( ( O E ι ⊗ R ) r ) ∨ ) \Psi_{V^{2r}}^{\lambda}(x)=\mu_{\lambda}(x_{2})\circ\mu^{\lambda}(x_{1})\in\operatorname{Hom}_{\mathcal{O}_{E^{\iota}}\otimes R}\left(S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r}),S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee}\right) Ψ V 2 r λ ( x ) = μ λ ( x 2 ) ∘ μ λ ( x 1 ) ∈ Hom O E ι ⊗ R ( S λ + (( O E ι ⊗ R ) r ) , S λ + (( O E ι ⊗ R ) r ) ∨ ) = S λ + ( ( O E ι ⊗ R ) r ) ∨ ⊗ O E ι ⊗ R S λ + ( ( O E ι ⊗ R ) r ) ∨ , =S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee}\otimes_{\mathcal{O}_{E^{\iota}}\otimes R}S_{\lambda+}((\mathcal{O}_{E^{\iota}}\otimes R)^{r})^{\vee}, = S λ + (( O E ι ⊗ R ) r ) ∨ ⊗ O E ι ⊗ R S λ + (( O E ι ⊗ R ) r ) ∨ , where we have identified V λ V_{\lambda} V λ with ( V λ ) ∨ (V_{\lambda})^{\vee} ( V λ ) ∨ using the hermitian form on V V V .
We recall the following definition, which will be used in latter chapters.
Definition 2.9. Let v v v be a nonarchimedean place of F F F and V V V a hermitian space over E v / F v E_v/F_v E v / F v . For an O E v \mathcal{O}_{E_v} O E v -lattice Λ \Lambda Λ of V V V , put
Λ ∨ : = { x ∈ V ∣ Tr E v / F v ( x , y ) V ∈ O F v for ∀ y ∈ Λ } . \Lambda^\vee:= \{x \in V \mid\operatorname{Tr}_{E_v/F_v}(x,y)_V \in\mathcal{O}_{F_v}\text{ for }\forall y \in\Lambda\}. Λ ∨ := { x ∈ V ∣ Tr E v / F v ( x , y ) V ∈ O F v for ∀ y ∈ Λ } . We say that
Λ \Lambda Λ is an integral lattice if Λ ⊆ Λ ∨ \Lambda\subseteq\Lambda^\vee Λ ⊆ Λ ∨ ;
Λ \Lambda Λ is a vertex lattice if Λ ∨ / Λ \Lambda^\vee/\Lambda Λ ∨ /Λ is annihilated by a uniformizer of E v E_v E v (and we define the index t ( Λ ) t(\Lambda) t ( Λ ) of Λ \Lambda Λ to be the length of Λ ∨ / Λ \Lambda^\vee/\Lambda Λ ∨ /Λ as an O E v \mathcal{O}_{E_v} O E v -module);
Λ \Lambda Λ is an self-dual or almost self-dual lattice if it is a vertex lattice with t ( Λ ) t(\Lambda) t ( Λ ) equal to 0 0 0 or 1 1 1 .
Cohomological automorphic representations
Definition 2.10. For a hermitian E E E -weight λ \lambda λ of rank m m m , we define H A ( λ ) \mathrm{HA}(\lambda) HA ( λ ) to be the set of isomorphism classes of automorphic representations Π \Pi Π of G L m ( A E ) \mathrm{GL}_m(\mathbb{A}_E) GL m ( A E ) such that
Π \Pi Π is an isobaric sum of mutually non-isomorphic conjugate self-dual cuspidal automorphic representations;
for every u ∈ Hom ( E , C ) u \in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , Π v ( u ) \Pi_{v(u)} Π v ( u ) , regarded as an irreducible representation of G L m ( C ) \mathrm{GL}_m(\mathbb{C}) GL m ( C ) via u u u is isomorphic to the principal series ⊞ i = 1 m arg 2 b u i + m + 1 − 2 i \boxplus_{i=1}^{m}\operatorname{arg}^{2b_{ui}+m+1-2i} ⊞ i = 1 m arg 2 b u i + m + 1 − 2 i , where λ u = ( b u 1 , … , b u m ) \lambda_u=(b_{u1},\ldots,b_{um}) λ u = ( b u 1 , … , b u m ) .
We relate the set H A ( λ ) \mathrm{HA}(\lambda) HA ( λ ) to the classical Rankin–Selberg setting when m = 2 m=2 m = 2 .
Definition 2.11. We introduce the following concepts.
A Rankin–Selberg pair is a pair ( σ , χ ) (\sigma,\chi) ( σ , χ ) in which σ \sigma σ is a cuspidal automorphic representation of G L 2 ( A F ) \mathrm{GL}_2(\mathbb{A}_F) GL 2 ( A F ) and χ \chi χ is an automorphic character of A E × \mathbb{A}_E^\times A E × , satisfying χ σ ⋅ χ ∣ A F × = 1 \chi_\sigma\cdot\chi|_{\mathbb{A}_F^\times}=1 χ σ ⋅ χ ∣ A F × = 1 , where χ σ \chi_\sigma χ σ denotes the central character of σ \sigma σ .
For a Rankin–Selberg pair ( σ , χ ) (\sigma,\chi) ( σ , χ ) , we may associate it with a root number ϵ ( σ , χ ) ∈ { ± 1 } \epsilon(\sigma,\chi)\in\{\pm1\} ϵ ( σ , χ ) ∈ { ± 1 } . We say that ( σ , χ ) (\sigma,\chi) ( σ , χ ) is coherent/incoherent if ϵ ( σ , χ ) = + 1 / − 1 \epsilon(\sigma,\chi)=+1/-1 ϵ ( σ , χ ) = + 1/ − 1 .
A Rankin–Selberg weight is a pair ( k , l ) (k,l) ( k , l ) in which k = ( k v ) v ∈ N V F ( ∞ ) k=(k_v)_v\in\mathbb{N}^{\mathscr{V}_F^{(\infty)}} k = ( k v ) v ∈ N V F ( ∞ ) and l = ( l u ) u ∈ Z Hom ( E , C ) l=(l_u)_u\in\mathbb{Z}^{\operatorname{Hom}(E,\mathbb{C})} l = ( l u ) u ∈ Z Hom ( E , C ) satisfying l u + l u ˉ = 0 l_u+l_{\bar u}=0 l u + l u ˉ = 0 and l u ≡ k v ( u ) ( m o d 2 ) l_u\equiv k_{v(u)}\pmod2 l u ≡ k v ( u ) ( mod 2 ) for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) .
We say that a Rankin–Selberg weight ( k , l ) (k,l) ( k , l ) is balanced if ∣ l u ∣ ≤ k v ( u ) |l_u|\leq k_{v(u)} ∣ l u ∣ ≤ k v ( u ) for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) .
For a Rankin–Selberg weight ( k , l ) (k,l) ( k , l ) , denote by R S ( k , l ) \mathrm{RS}(k,l) RS ( k , l ) the set of isomorphism classes of Rankin–Selberg pairs ( σ , χ ) (\sigma,\chi) ( σ , χ ) satisfying that (a) for every v ∈ V F ( ∞ ) v\in\mathscr{V}_F^{(\infty)} v ∈ V F ( ∞ ) , σ v \sigma_v σ v is the (unitary) discrete series representation of G L 2 ( R ) \mathrm{GL}_2(\mathbb{R}) GL 2 ( R ) of weight k v + 2 k_v+2 k v + 2 ; and (b) for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , χ u \chi_u χ u , regarded as a character of C × \mathbb{C}^\times C × via u u u , is arg l u \operatorname{arg}^{l_u} arg l u .
For a Rankin–Selberg weight ( k , l ) (k,l) ( k , l ) , define an element λ ( k , l ) = ( b u 1 , b u 2 ) u ∈ Z 2 Hom ( E , C ) \lambda_{(k,l)}=(b_{u1},b_{u2})_u\in\mathbb{Z}_2^{\operatorname{Hom}(E,\mathbb{C})} λ ( k , l ) = ( b u 1 , b u 2 ) u ∈ Z 2 Hom ( E , C ) by the formula
b u 1 = k v ( u ) + l u 2 , b u 2 = − k v ( u ) + l u 2 b_{u1}=\frac{k_{v(u)}+l_u}{2},\qquad b_{u2}=\frac{-k_{v(u)}+l_u}{2} b u 1 = 2 k v ( u ) + l u , b u 2 = 2 − k v ( u ) + l u
for u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . It is clear that the assignment ( k , l ) ↦ λ ( k , l ) (k,l)\mapsto\lambda_{(k,l)} ( k , l ) ↦ λ ( k , l ) defines a bijection between Rankin–Selberg weights and hermitian E E E -weights of rank 2 2 2 , under which the two balanced subsets coincide.
Lemma 2.12. The assignment ( σ , χ ) ↦ B C ( σ ) ⊗ χ (\sigma,\chi)\mapsto\mathrm{BC}(\sigma)\otimes\chi ( σ , χ ) ↦ BC ( σ ) ⊗ χ defines a map R S ( k , l ) → H A ( λ ( k , l ) ) \mathrm{RS}(k,l)\to\mathrm{HA}(\lambda_{(k,l)}) RS ( k , l ) → HA ( λ ( k , l ) ) , which is surjective. Here, B C \mathrm{BC} BC denotes the quadratic automorphic base change from F F F to E E E .
Proof. We start from showing that Π : = B C ( σ ) ⊗ χ \Pi:=\mathrm{BC}(\sigma)\otimes\chi Π := BC ( σ ) ⊗ χ indeed belongs to H A ( λ ( k , l ) ) \mathrm{HA}(\lambda_{(k,l)}) HA ( λ ( k , l ) ) . Part (2) of Definition 2.10 is clear. It remains to check (1). For every automorphic representation Σ \Sigma Σ over E E E , denote by Σ − \Sigma^{-} Σ − its conjugation with respect to E / F E/F E / F . There are two cases.
First, B C ( σ ) \mathrm{BC}(\sigma) BC ( σ ) remains cuspidal so that Π \Pi Π is cuspidal. Then
Π ∨ ≃ B C ( σ ∨ ) ⊗ χ − 1 ≃ B C ( σ ⊗ χ σ − 1 ) ⊗ χ − 1 ≃ B C ( σ ) ⊗ ( χ σ − 1 ∘ Nm E / F ) ⊗ χ − 1 ≃ B C ( σ ) ⊗ ( χ ∘ Nm E / F ) ⊗ χ − 1 ≃ B C ( σ ) ⊗ χ − ≃ B C ( σ ) − ⊗ χ − ≃ Π − . (3) \begin{aligned}
\Pi^\vee\simeq\mathrm{BC}(\sigma^\vee)\otimes\chi^{-1}
\simeq\mathrm{BC}(\sigma\otimes\chi_\sigma^{-1})\otimes\chi^{-1}
\simeq\mathrm{BC}(\sigma)\otimes(\chi_\sigma^{-1}\circ\operatorname{Nm}_{E/F})\otimes\chi^{-1} \\
&\simeq\mathrm{BC}(\sigma)\otimes(\chi\circ\operatorname{Nm}_{E/F})\otimes\chi^{-1}
\simeq\mathrm{BC}(\sigma)\otimes\chi^{-}
\simeq\mathrm{BC}(\sigma)^{-}\otimes\chi^{-}
\simeq\Pi^{-}. \tag*{(3)}
\end{aligned} Π ∨ ≃ BC ( σ ∨ ) ⊗ χ − 1 ≃ BC ( σ ⊗ χ σ − 1 ) ⊗ χ − 1 ≃ BC ( σ ) ⊗ ( χ σ − 1 ∘ Nm E / F ) ⊗ χ − 1 ≃ BC ( σ ) ⊗ ( χ ∘ Nm E / F ) ⊗ χ − 1 ≃ BC ( σ ) ⊗ χ − ≃ BC ( σ ) − ⊗ χ − ≃ Π − . (3)
Second, BC ( σ ) \operatorname{BC}(\sigma) BC ( σ ) is the isobaric sum of two (distinct) automorphic characters (hence so is Π \Pi Π ). Then we must have BC ( σ ) ≃ χ ~ ⊞ χ ~ ∼ \operatorname{BC}(\sigma) \simeq\widetilde{\chi} \boxplus\widetilde{\chi}^{\,\sim} BC ( σ ) ≃ χ ⊞ χ ∼ for some automorphic character χ ~ \widetilde{\chi} χ of A E × \mathbb{A}_{E}^{\times} A E × , hence ω σ = χ ~ ∣ A F × ⋅ χ E / F \omega_{\sigma}=\widetilde{\chi}|_{\mathbb{A}_{F}^{\times}}\cdot\chi_{E/F} ω σ = χ ∣ A F × ⋅ χ E / F . It follows that χ ~ ⋅ χ ∣ A F × = χ ~ ∼ ⋅ χ ∣ A F × = χ E / F \widetilde{\chi}\cdot\chi|_{\mathbb{A}_{F}^{\times}}=\widetilde{\chi}^{\,\sim}\cdot\chi|_{\mathbb{A}_{F}^{\times}}=\chi_{E/F} χ ⋅ χ ∣ A F × = χ ∼ ⋅ χ ∣ A F × = χ E / F , which implies that ( χ ~ ⋅ χ ) ∨ ≃ ( χ ~ ⋅ χ ) − (\widetilde{\chi}\cdot\chi)^{\vee}\simeq(\widetilde{\chi}\cdot\chi)^{-} ( χ ⋅ χ ) ∨ ≃ ( χ ⋅ χ ) − and ( χ ~ ∼ ⋅ χ ) ∨ ≃ ( χ ~ ∼ ⋅ χ ) − (\widetilde{\chi}^{\,\sim}\cdot\chi)^{\vee}\simeq(\widetilde{\chi}^{\,\sim}\cdot\chi)^{-} ( χ ∼ ⋅ χ ) ∨ ≃ ( χ ∼ ⋅ χ ) − .
Then we show the surjectivity. By twisting an automorphic character of E E E satisfying Definition 2.11(5b), we may assume without loss of generality that l = 0 l=0 l = 0 (that is, l u = 0 l_u=0 l u = 0 for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) ). Take an element Π ∈ H A ( λ ( k , 0 ) ) \Pi\in\mathrm{HA}(\lambda_{(k,0)}) Π ∈ HA ( λ ( k , 0 ) ) . Again, there are two cases.
First, Π \Pi Π is cuspidal. The central character χ Π \chi_{\Pi} χ Π is a Dirichlet character of E E E satisfying χ Π ⋅ χ Π − = 1 \chi_{\Pi}\cdot\chi_{\Pi}^{-}=1 χ Π ⋅ χ Π − = 1 . Thus, one may find a Dirichlet character χ \chi χ of E E E such that χ Π = χ / χ − \chi_{\Pi}=\chi/\chi^{-} χ Π = χ / χ − , which implies that ( Π ⊗ χ − ) ∨ ≃ Π ⊗ χ − 1 (\Pi\otimes\chi^{-})^{\vee}\simeq\Pi\otimes\chi^{-1} ( Π ⊗ χ − ) ∨ ≃ Π ⊗ χ − 1 . Thus, we may find a cuspidal automorphic representation σ \sigma σ of GL 2 ( A F ) \operatorname{GL}_{2}(\mathbb{A}_{F}) GL 2 ( A F ) such that BC ( σ ) ≃ Π ⊗ χ − 1 \operatorname{BC}(\sigma)\simeq\Pi\otimes\chi^{-1} BC ( σ ) ≃ Π ⊗ χ − 1 , or equivalently, Π ≃ BC ( σ ) ⊗ χ \Pi\simeq\operatorname{BC}(\sigma)\otimes\chi Π ≃ BC ( σ ) ⊗ χ . It is clear that ( σ , χ ) ∈ RS ( k , 0 ) (\sigma,\chi)\in\operatorname{RS}(k,0) ( σ , χ ) ∈ RS ( k , 0 ) .
Second, Π ≃ χ 1 ⊞ χ 2 \Pi\simeq\chi_{1}\boxplus\chi_{2} Π ≃ χ 1 ⊞ χ 2 is an isobaric sum of two distinct conjugate self-dual automorphic characters. Then the quotient χ 1 / χ 2 − \chi_{1}/\chi_{2}^{-} χ 1 / χ 2 − is a Dirichlet character of E E E satisfying ( χ 1 / χ 2 − ) ( χ 1 / χ 2 − ) − = 1 (\chi_{1}/\chi_{2}^{-})(\chi_{1}/\chi_{2}^{-})^{-}=1 ( χ 1 / χ 2 − ) ( χ 1 / χ 2 − ) − = 1 . Thus, one may find a Dirichlet character χ \chi χ of E E E such that χ 1 / χ 2 − = χ / χ − \chi_{1}/\chi_{2}^{-}=\chi/\chi^{-} χ 1 / χ 2 − = χ / χ − . Put χ ~ : = χ 1 χ − 1 \widetilde{\chi}:=\chi_{1}\chi^{-1} χ := χ 1 χ − 1 , so that Π ≃ ( χ ~ ⊞ χ ~ − ) ⊗ χ \Pi\simeq(\widetilde{\chi}\boxplus\widetilde{\chi}^{-})\otimes\chi Π ≃ ( χ ⊞ χ − ) ⊗ χ . As χ 1 ≠ χ 2 \chi_{1}\ne\chi_{2} χ 1 = χ 2 , one can find a cuspidal automorphic representation σ \sigma σ of GL 2 ( A F ) \operatorname{GL}_{2}(\mathbb{A}_{F}) GL 2 ( A F ) such that BC ( σ ) ≃ χ ~ ⊞ χ ~ − \operatorname{BC}(\sigma)\simeq\widetilde{\chi}\boxplus\widetilde{\chi}^{-} BC ( σ ) ≃ χ ⊞ χ − , or equivalently, Π ≃ BC ( σ ) ⊗ χ \Pi\simeq\operatorname{BC}(\sigma)\otimes\chi Π ≃ BC ( σ ) ⊗ χ . It is clear that ( σ , χ ) ∈ RS ( k , 0 ) (\sigma,\chi)\in\operatorname{RS}(k,0) ( σ , χ ) ∈ RS ( k , 0 ) .
The lemma is proved. □
Automorphic forms for unitary groups
Let W r W_r W r be the standard skew-hermitian space over E / F E/F E / F of dimension 2 r 2r 2 r , whose skew-hermitian form is given by the matrix
w r : = ( 0 1 r − 1 r 0 ) . w_r:=\begin{pmatrix} 0 & 1_r \\ -1_r & 0 \end{pmatrix}. w r := ( 0 − 1 r 1 r 0 ) .
Put G r : = U ( W r ) G_r:=U(W_r) G r := U ( W r ) as a reductive group over F F F , equipped with homomorphisms
m : Res E / F GL r ⟶ G r , a ⟼ ( a t a ‾ − 1 ) , m:\operatorname{Res}_{E/F}\operatorname{GL}_r\longrightarrow G_r,\qquad a\longmapsto\begin{pmatrix} a & \\ & {}^{t}\overline{a}^{-1}\end{pmatrix}, m : Res E / F GL r ⟶ G r , a ⟼ ( a t a − 1 ) ,
n : Herm r ⟶ G r , b ⟼ ( 1 r b 1 r ) , n:\operatorname{Herm}_r\longrightarrow G_r,\qquad b\longmapsto\begin{pmatrix} 1_r & b \\ & 1_r\end{pmatrix}, n : Herm r ⟶ G r , b ⟼ ( 1 r b 1 r ) ,
of algebraic groups over F F F , whose image we denote by M r M_r M r and N r N_r N r , respectively.
Notation 2.13. We will frequently use the following notation related to G r G_r G r .
For every v ∈ V F f i n v\in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin , we denote by K r , v K_{r,v} K r , v the maximal open compact subgroup of G r ( F v ) G_r(F_v) G r ( F v ) that is the stabilizer of O E v 2 r \mathcal{O}_{E_v}^{2r} O E v 2 r .
For every v ∈ V F ( ∞ ) v\in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) , the map
( U 1 , U 2 ) ⟼ u − 1 1 2 ( U 1 + U 2 − i U 1 + i U 2 i U 1 − i U 2 U 1 + U 2 ) , (U_1,U_2)\longmapsto u^{-1}\frac{1}{2}\begin{pmatrix} U_1+U_2 & -iU_1+iU_2 \\ iU_1-iU_2 & U_1+U_2 \end{pmatrix}, ( U 1 , U 2 ) ⟼ u − 1 2 1 ( U 1 + U 2 i U 1 − i U 2 − i U 1 + i U 2 U 1 + U 2 ) ,
where u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) is an arbitrary element above v v v , identifies U ( r ) × U ( r ) U(r)\times U(r) U ( r ) × U ( r ) with a maximal compact subgroup K r , v K_{r,v} K r , v of G r ( F v ) G_r(F_v) G r ( F v ) that is independent of the choice of u u u . Moreover, the assignment ( U 1 , U 2 ) ↦ det ( U 1 U 2 − 1 ) (U_1,U_2)\mapsto\det(U_1U_2^{-1}) ( U 1 , U 2 ) ↦ det ( U 1 U 2 − 1 ) defines a character δ v : K r , v → C × \delta_v:K_{r,v}\to\mathbb{C}^{\times} δ v : K r , v → C × , which is also independent of the choice of u u u .
For every finite subset S ⊆ V F S\subseteq\mathcal{V}_{F} S ⊆ V F , denote by G ~ r ( F S ) \widetilde{G}_r(F_S) G r ( F S ) the metaplectic double cover of G r ( F S ) G_r(F_S) G r ( F S ) . Globally, denote by G ~ r ( A F ) \widetilde{G}_r(\mathbb{A}_{F}) G r ( A F ) the metaplectic double cover of G r ( A F ) G_r(\mathbb{A}_{F}) G r ( A F ) .
Put
K r , ∞ : = ∏ v ∈ V F ( ∞ ) K r , v , δ : = ∏ v ∈ V F ( ∞ ) δ v : K r , ∞ ⟶ C × . K_{r,\infty}:=\prod_{v\in\mathcal{V}_{F}^{(\infty)}}K_{r,v},\qquad\delta:=\prod_{v\in\mathcal{V}_{F}^{(\infty)}}\delta_v:K_{r,\infty}\longrightarrow\mathbb{C}^{\times}. K r , ∞ := v ∈ V F ( ∞ ) ∏ K r , v , δ := v ∈ V F ( ∞ ) ∏ δ v : K r , ∞ ⟶ C × .
Denote by
δ 1 / 2 : K ~ r , ∞ ⟶ C × \delta^{1/2}:\widetilde{K}_{r,\infty}\longrightarrow\mathbb{C}^{\times} δ 1/2 : K r , ∞ ⟶ C ×
the pullback of det \det det along the map C × → C × \mathbb{C}^{\times} \to\mathbb{C}^{\times} C × → C × sending z z z to z 2 z^{2} z 2 , so that K ~ r , ∞ \widetilde{K}_{r,\infty} K r , ∞ is naturally a maximal compact subgroup of G r ( F ∞ ) G_{r}(F_{\infty}) G r ( F ∞ ) .
(5) For every tuple b = ( b u ) u ∈ N Hom ( E , C ) b=(b_{u})_{u}\in\mathbb{N}^{\operatorname{Hom}(E,\mathbb{C})} b = ( b u ) u ∈ N Hom ( E , C ) , we have the representation T b ( C r ) T^{b}(\mathbb{C}^{r}) T b ( C r ) (Construction 2.2) of S b × K r , ∞ \mathcal{S}_{b}\times K_{r,\infty} S b × K r , ∞ , where the action of K r , ∞ K_{r,\infty} K r , ∞ is the unique one such that for every k ∈ K r , ∞ k\in K_{r,\infty} k ∈ K r , ∞ and every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , if we write u ( k ) = ( U 1 , U 2 ) ∈ U ( r ) × U ( r ) u(k)=(U_{1},U_{2})\in U(r)\times U(r) u ( k ) = ( U 1 , U 2 ) ∈ U ( r ) × U ( r ) , then U 1 U_{1} U 1 acts on T b u ( C r ) T^{b^{u}}(\mathbb{C}^{r}) T b u ( C r ) via itself as an element of GL r ( C ) \operatorname{GL}_{r}(\mathbb{C}) GL r ( C ) and U 2 U_{2} U 2 acts on T b u ( C r ) T^{b_{u}}(\mathbb{C}^{r}) T b u ( C r ) via the element U 2 U_{2} U 2 of GL r ( C ) \operatorname{GL}_{r}(\mathbb{C}) GL r ( C ) .
(6) For every v ∈ V F v\in\mathcal{V}_{F} v ∈ V F , we define a Haar measure d g v dg_{v} d g v on G r ( F v ) G_{r}(F_{v}) G r ( F v ) as follows:
for v ∈ V F f i n v\in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin , d g v dg_{v} d g v is the Haar measure under which K r , v K_{r,v} K r , v has volume 1 1 1 ;
for v ∈ V F ( ∞ ) v\in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) , d g v dg_{v} d g v is the product of the measure on K v K_{v} K v of total volume 1 1 1 and the standard hyperbolic measure on G ( F v ) / K v G(F_{v})/K_{v} G ( F v ) / K v (see, for example, [11 §2.1] ).
Put d g : = ∏ v d g v dg:=\prod_{v}dg_{v} d g := ∏ v d g v , which is a Haar measure on G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) .
(7) Denote by A ( G r ) \mathcal{A}(G_{r}) A ( G r ) the space of ( g r , ∞ , K ~ r , ∞ ) (\mathfrak{g}_{r,\infty},\widetilde{K}_{r,\infty}) ( g r , ∞ , K r , ∞ ) -finite automorphic forms on G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) , where g r , ∞ \mathfrak{g}_{r,\infty} g r , ∞ denotes the complexified Lie algebra of G r , ∞ G_{r,\infty} G r , ∞ . Denote by A ( G r ) c u s p \mathcal{A}(G_{r})_{\mathrm{cusp}} A ( G r ) cusp and A ( G r ) h o l \mathcal{A}(G_{r})^{\mathrm{hol}} A ( G r ) hol the subspaces of A ( G r ) \mathcal{A}(G_{r}) A ( G r ) of cusp and holomorphic forms, respectively. Put A ( G r ) c u s p h o l : = A ( G r ) c u s p ∩ A ( G r ) h o l \mathcal{A}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}}:=\mathcal{A}(G_{r})_{\mathrm{cusp}}\cap\mathcal{A}(G_{r})^{\mathrm{hol}} A ( G r ) cusp hol := A ( G r ) cusp ∩ A ( G r ) hol .
Definition 2.14. Consider an element λ = ( λ u ) u ∈ ( Z r + ) Hom ( E , C ) \lambda=(\lambda_{u})_{u}\in(\mathbb{Z}_{r}^{+})^{\operatorname{Hom}(E,\mathbb{C})} λ = ( λ u ) u ∈ ( Z r + ) Hom ( E , C ) . Construction 2.2 provides us with a direct summand S λ ( C r ) S_{\lambda}(\mathbb{C}^{r}) S λ ( C r ) of T b ( λ ) ( C r ) T^{b(\lambda)}(\mathbb{C}^{r}) T b ( λ ) ( C r ) as an (irreducible) representation of K r , ∞ K_{r,\infty} K r , ∞ , which we denote by ρ λ \rho_{\lambda} ρ λ (also regarded as a representation of K ~ r , ∞ \widetilde{K}_{r,\infty} K r , ∞ by inflation).
Let w w w be a half-integer (or an integer).
(1) Denote by A λ w ( G r ) \mathcal{A}_{\lambda}^{w}(G_{r}) A λ w ( G r ) the subspace of
Hom C ( S λ ( C r ) , A ( G r ) ) = S λ ( C r ) ∨ ⊗ C A ( G r ) \operatorname{Hom}_{\mathbb{C}}(S_{\lambda}(\mathbb{C}^{r}),\mathcal{A}(G_{r}))=S_{\lambda}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\mathcal{A}(G_{r}) Hom C ( S λ ( C r ) , A ( G r )) = S λ ( C r ) ∨ ⊗ C A ( G r )
consisting of elements φ \varphi φ (written as a homomorphism) satisfying
φ ( ρ λ ( k ) v ) = δ − w ( k ) ⋅ R k ( φ ( v ) ) , ∀ v ∈ S λ ( C r ) , ∀ k ∈ K ~ r , ∞ , \varphi(\rho_{\lambda}(k)v)=\delta^{-w}(k)\cdot R_{k}(\varphi(v)),\qquad\forall v\in S_{\lambda}(\mathbb{C}^{r}),\ \forall k\in\widetilde{K}_{r,\infty}, φ ( ρ λ ( k ) v ) = δ − w ( k ) ⋅ R k ( φ ( v )) , ∀ v ∈ S λ ( C r ) , ∀ k ∈ K r , ∞ ,
where R k R_{k} R k denotes the right translation on A ( G r ) \mathcal{A}(G_{r}) A ( G r ) by $k.8 S i m i l a r l y , w e d e f i n e Similarly, we define S imi l a r l y , w e d e f in e \mathcal{A}_{\lambda}^{w}(G_{r})_{\mathrm{cusp}}, , , \mathcal{A}_{\lambda}^{w}(G_{r})^{\mathrm{hol}}a n d and an d \mathcal{A}_{\lambda}^{w}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}}. W h e n . When . W h e n w=r, w e s i m p l y w r i t e , we simply write , w es im pl y w r i t e \mathcal{A}_{\lambda}(G_{r})f o r for f or \mathcal{A}_{\lambda}^{r}(G_{r})$.
(2) Define a hermitian pairing
( , ) G r : A λ w ( G r ) c u s p × A λ w ( G r ) c u s p → C (\ ,\ )_{G_{r}}:\mathcal{A}_{\lambda}^{w}(G_{r})_{\mathrm{cusp}}\times\mathcal{A}_{\lambda}^{w}(G_{r})_{\mathrm{cusp}}\to\mathbb{C} ( , ) G r : A λ w ( G r ) cusp × A λ w ( G r ) cusp → C
by the formula
( φ 1 , φ 2 ) G r : = ∫ G r ( F ) \ G r ( A F ) tr λ ( φ 1 ( g ) ⊗ φ 2 ( g ) ‾ ) d g , (\varphi_{1},\varphi_{2})_{G_{r}}:=\int_{G_{r}(F)\backslash G_{r}(\mathbb{A}_{F})}\operatorname{tr}_{\lambda}\bigl(\varphi_{1}(g)\otimes\overline{\varphi_{2}(g)}\bigr)\,dg, ( φ 1 , φ 2 ) G r := ∫ G r ( F ) \ G r ( A F ) tr λ ( φ 1 ( g ) ⊗ φ 2 ( g ) ) d g ,
where
tr λ : S λ ( C r ) ∨ ⊗ C S λ ( C r ) ∨ ‾ → C \operatorname{tr}_{\lambda}:S_{\lambda}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\overline{S_{\lambda}(\mathbb{C}^{r})^{\vee}}\to\mathbb{C} tr λ : S λ ( C r ) ∨ ⊗ C S λ ( C r ) ∨ → C
is the natural pairing induced from the standard hermitian form on C r \mathbb{C}^{r} C r , and d g dg d g is the measure in Notation 2.13(6). The integral ( φ 1 , φ 2 ) G r (\varphi_{1},\varphi_{2})_{G_{r}} ( φ 1 , φ 2 ) G r extends to the case where one of φ 1 , φ 2 \varphi_{1},\varphi_{2} φ 1 , φ 2 belongs to A λ w ( G r ) \mathcal{A}_{\lambda}^{w}(G_{r}) A λ w ( G r ) .
(3) The representation ( ρ λ , S λ ( C r ) ) (\rho_{\lambda},S_{\lambda}(\mathbb{C}^{r})) ( ρ λ , S λ ( C r )) restricted to K ~ r , ∞ ∩ M r ( F ∞ ) \widetilde{K}_{r,\infty}\cap M_{r}(F_{\infty}) K r , ∞ ∩ M r ( F ∞ ) extends naturally to a representation of M r ( F ∞ ) M_{r}(F_{\infty}) M r ( F ∞ ) in the way that m ( a ) ∈ M r ( F ∞ ) m(a)\in M_{r}(F_{\infty}) m ( a ) ∈ M r ( F ∞ ) acts on the factor T b u ( C r ) T^{b^{u}}(\mathbb{C}^{r}) T b u ( C r ) via u ( a ) ∈ GL r ( C ) u(a)\in\operatorname{GL}_{r}(\mathbb{C}) u ( a ) ∈ GL r ( C ) . More generally, we have a map
ρ λ : G ~ r ( F ∞ ) → Aut C ( S λ ( C r ) ) \rho_{\lambda}: \widetilde{G}_{r}(F_{\infty})\to\operatorname{Aut}_{\mathbb{C}}(S_{\lambda}(\mathbb{C}^{r})) ρ λ : G r ( F ∞ ) → Aut C ( S λ ( C r ))
sending g ∈ G ~ r ( F ∞ ) g\in\widetilde{G}_{r}(F_{\infty}) g ∈ G r ( F ∞ ) to ρ λ ( m ( a ) ) ρ λ ( k ) \rho_{\lambda}(m(a))\rho_{\lambda}(k) ρ λ ( m ( a )) ρ λ ( k ) if we write g = n ( b ) m ( a ) k g=n(b)m(a)k g = n ( b ) m ( a ) k for b ∈ Herm r ( F ∞ ) b\in\operatorname{Herm}_{r}(F_{\infty}) b ∈ Herm r ( F ∞ ) , a ∈ GL r ( E ∞ ) a\in\operatorname{GL}_{r}(E_{\infty}) a ∈ GL r ( E ∞ ) , and $k\in\widetilde{K}_{r,\infty}.9 $
Doubling method
Consider a hermitian space V V V over A E / A F \mathbb{A}_{E}/\mathbb{A}_{F} A E / A F of rank m m m . Put H : = U ( V ) H := U(V) H := U ( V ) as a reductive group over A F \mathbb{A}_{F} A F .
Definition 2.15. We say that a hermitian space V V V over A E / A F \mathbb{A}_{E}/\mathbb{A}_{F} A E / A F is coherent/incoherent if χ E / F ( det V ) = + 1 / − 1 \chi_{E/F}(\det V) = +1/-1 χ E / F ( det V ) = + 1/ − 1 .
Definition 2.16. Let R R R be a ring and ⋄ ⊆ S F f i n \diamond\subseteq\mathbb{S}_{F}^{\mathrm{fin}} ⋄ ⊆ S F fin a finite subset.
(1) We denote by S ( V ⋄ , R ) \mathcal{S}(V_{\diamond},R) S ( V ⋄ , R ) and S ( ( V ∞ , ⋄ ) r , R ) \mathcal{S}((V^{\infty,\diamond})^{r},R) S (( V ∞ , ⋄ ) r , R ) the R R R -module of R R R -valued Schwartz functions (namely, locally constant functions with compact support) on V ⋄ r V_{\diamond}^{r} V ⋄ r and ( V ∞ , ⋄ ) r (V^{\infty,\diamond})^{r} ( V ∞ , ⋄ ) r , respectively.
(2) Suppose that 1 ≤ r ≤ m 1 \leq r \leq m 1 ≤ r ≤ m . We say that a Schwartz function in S ( V ⋄ r , R ) \mathcal{S}(V_{\diamond}^{r},R) S ( V ⋄ r , R ) is regularly supported if its support is contained in the set
{ x ∈ V ⋄ r ∣ T ( x ) ∈ Herm r ∘ ( F ⋄ ) } . \left\{x \in V_{\diamond}^{r} \mid T(x) \in\operatorname{Herm}_{r}^{\circ}(F_{\diamond})\right\}. { x ∈ V ⋄ r ∣ T ( x ) ∈ Herm r ∘ ( F ⋄ ) } .
(3) Suppose that 1 ≤ r ≤ m 1 \leq r \leq m 1 ≤ r ≤ m and ⋄ = { v } \diamond= \{v\} ⋄ = { v } in which v v v is nonsplit in E E E . We say that a Schwartz function in S ( V v r , R ) \mathcal{S}(V_{v}^{r},R) S ( V v r , R ) is deeply regularly supported if its support is contained in the set
{ x ∈ V v r ∣ T ( x ) + ϖ v d Herm r ( O F v ) ⊆ Herm r ∘ ( F v ) } , \left\{x \in V_{v}^{r} \mid T(x) + \varpi_{v}^{d}\operatorname{Herm}_{r}(\mathcal{O}_{F_{v}}) \subseteq\operatorname{Herm}_{r}^{\circ}(F_{v})\right\}, { x ∈ V v r ∣ T ( x ) + ϖ v d Herm r ( O F v ) ⊆ Herm r ∘ ( F v ) } ,
where d d d is the smallest integer such that ψ F , v \psi_{F,v} ψ F , v is trivial on ϖ v d O F v \varpi_{v}^{d}\mathcal{O}_{F_{v}} ϖ v d O F v .
We suppress R R R from the notation when it is C \mathbb{C} C .
We review the notion of Weil representation. For every v ∈ S F v \in\mathbb{S}_{F} v ∈ S F , we have the Weil representation ω r , V v \omega_{r,V_{v}} ω r , V v of G ~ r ( F v ) × H ( F v ) \widetilde{G}_{r}(F_{v}) \times H(F_{v}) G r ( F v ) × H ( F v ) on the space S ( V v r ) \mathcal{S}(V_{v}^{r}) S ( V v r ) with respect to the additive character ψ F , v \psi_{F,v} ψ F , v . When m m m is even, the representation factors through G r ( F v ) × H ( F v ) G_{r}(F_{v}) \times H(F_{v}) G r ( F v ) × H ( F v ) ,10 whose formula we review as follows: Taking ϕ ∈ S ( V v r ) \phi\in\mathcal{S}(V_{v}^{r}) ϕ ∈ S ( V v r ) ,
ω r , V v ( m ( a ) ) ϕ ( x ) = ∣ det a ∣ E v m / 2 ⋅ ϕ ( x a ) ; \omega_{r,V_{v}}(m(a))\phi(x) = \lvert\det a\rvert_{E_{v}}^{m/2} \cdot\phi(xa); ω r , V v ( m ( a )) ϕ ( x ) = ∣ det a ∣ E v m /2 ⋅ ϕ ( x a ) ;
ω r , V v ( n ( b ) ) ϕ ( x ) = ψ F , v ( tr ( T ( x ) b ) ) ⋅ ϕ ( x ) ; \omega_{r,V_{v}}(n(b))\phi(x) = \psi_{F,v}(\operatorname{tr}(T(x)b)) \cdot\phi(x); ω r , V v ( n ( b )) ϕ ( x ) = ψ F , v ( tr ( T ( x ) b )) ⋅ ϕ ( x ) ;
ω r , V v ( w r ) ϕ ( x ) = γ V v r ⋅ ϕ ^ ( x ) , \omega_{r,V_{v}}(w_{r})\phi(x) = \gamma_{V_{v}}^{r} \cdot\widehat{\phi}(x), ω r , V v ( w r ) ϕ ( x ) = γ V v r ⋅ ϕ ( x ) ,
where γ V v \gamma_{V_{v}} γ V v is the Weil constant of V v V_{v} V v (with respect to ψ F , v \psi_{F,v} ψ F , v );
ω r , V v ( h ) ϕ ( x ) = ϕ ( h − 1 x ) . \omega_{r,V_{v}}(h)\phi(x) = \phi(h^{-1}x). ω r , V v ( h ) ϕ ( x ) = ϕ ( h − 1 x ) .
When V V V is clear, we simply write ω r , v \omega_{r,v} ω r , v for ω r , V v \omega_{r,V_{v}} ω r , V v . Taking (restricted) tensor product, we have ω r , ∞ \omega_{r,\infty} ω r , ∞ for all archimedean places or ω r \omega_{r} ω r for all places.
Remark 2.17. Let { S v ∣ v ∈ S F } \{S_{v} \mid v \in\mathbb{S}_{F}\} { S v ∣ v ∈ S F } be a collection of finite-dimensional complex vector spaces indexed by S F \mathbb{S}_{F} S F such that S v = C S_{v} = \mathbb{C} S v = C for all but finitely many v v v .
(1) For every v ∈ S F v \in\mathbb{S}_{F} v ∈ S F , we regard S ( V v r ) ⊗ C S v \mathcal{S}(V_{v}^{r}) \otimes_{\mathbb{C}} S_{v} S ( V v r ) ⊗ C S v as the space of S v S_{v} S v -valued Schwartz functions on V v r V_{v}^{r} V v r , hence receives a similar Weil representation ω r , v \omega_{r,v} ω r , v . Definition 2.16 extends to this setting as well.
(2) Take S : = ⨂ v S v S := \bigotimes_{v} S_{v} S := ⨂ v S v . We regard S ( V r ) ⊗ C S \mathcal{S}(V^{r}) \otimes_{\mathbb{C}} S S ( V r ) ⊗ C S as
( ⨂ v ∈ S F ( S ( V v r ) ⊗ C S v ) ) , \left(\bigotimes_{v \in\mathbb{S}_{F}} \left(\mathcal{S}(V_{v}^{r}) \otimes_{\mathbb{C}} S_{v}\right)\right), ( v ∈ S F ⨂ ( S ( V v r ) ⊗ C S v ) ) ,
so that a decomposable Schwartz function ϕ ∈ S ( V r ) ⊗ C S \phi\in\mathcal{S}(V^{r}) \otimes_{\mathbb{C}} S ϕ ∈ S ( V r ) ⊗ C S is of the form ⨂ v ϕ v \bigotimes_{v}\phi_{v} ⨂ v ϕ v in which ϕ v ∈ S ( V v r ) ⊗ C S v \phi_{v} \in\mathcal{S}(V_{v}^{r}) \otimes_{\mathbb{C}} S_{v} ϕ v ∈ S ( V v r ) ⊗ C S v .
Notation 2.18. We collect some notation concerning the doubling skew-hermitian space
W r □ : = W r ⊕ ( − W r ) . W_r^{\square}:=W_r\oplus(-W_r). W r □ := W r ⊕ ( − W r ) . Let G r □ G_r^{\square} G r □ be the unitary group of W r □ W_r^{\square} W r □ , which contains G r × G r G_r\times G_r G r × G r canonically. We now take a basis { e 1 □ , … , e 4 r □ } \{e_1^{\square},\ldots,e_{4r}^{\square}\} { e 1 □ , … , e 4 r □ } of W r □ W_r^{\square} W r □ by the formula
e i □ = e i , e r + i □ = − e i − , e 2 r + i □ = e r + i , e 3 r + i □ = e r + i − e_i^{\square}=e_i,\qquad e_{r+i}^{\square}=-e_i^{-},\qquad e_{2r+i}^{\square}=e_{r+i},\qquad e_{3r+i}^{\square}=e_{r+i}^{-} e i □ = e i , e r + i □ = − e i − , e 2 r + i □ = e r + i , e 3 r + i □ = e r + i − (where { e 1 − , … , e 2 r − } \{e_1^{-},\ldots,e_{2r}^{-}\} { e 1 − , … , e 2 r − } is the standard basis of − W r -W_r − W r ) for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r , under which we may identify W r □ W_r^{\square} W r □ with W 2 r W_{2r} W 2 r and G r □ G_r^{\square} G r □ with G 2 r G_{2r} G 2 r . Finally, put
w r : = ( 1 r 1 r − 1 r 1 r 1 r 1 r ) ∈ G r □ ( F ) . \mathbf{w}_r:=\begin{pmatrix}
& &1_r&\\
&1_r&&\\
-1_r&1_r&&\\
&&1_r&1_r
\end{pmatrix}\in G_r^{\square}(F). w r := − 1 r 1 r 1 r 1 r 1 r 1 r ∈ G r □ ( F ) . Then P r □ w r ( G r × G r ) P_r^{\square}\mathbf{w}_r(G_r\times G_r) P r □ w r ( G r × G r ) is Zariski open in G r □ G_r^{\square} G r □ .
From now to the end of this section, we assume m = 2 r m=2r m = 2 r . Take a collection { S v ∣ v ∈ V F } \{S_v\mid v\in\mathcal{V}_F\} { S v ∣ v ∈ V F } as in Remark 2.17 and put S : = ⨂ v S v \mathcal{S}:=\bigotimes_v S_v S := ⨂ v S v . For every Schwartz function Φ ∈ S ( V 2 r ) ⊗ C S \Phi\in\mathcal{S}(V^{2r})\otimes_{\mathbb{C}}\mathcal{S} Φ ∈ S ( V 2 r ) ⊗ C S , we have the Siegel–Weil section f Φ ( s ) f_{\Phi}^{(s)} f Φ ( s ) (see [20 §3] for details) which is an S \mathcal{S} S -valued function on P 2 r ( F ) \ G 2 r ( A F ) P_{2r}(F)\backslash G_{2r}(\mathbb{A}_F) P 2 r ( F ) \ G 2 r ( A F ) , and further the Eisenstein series
E ( s , g , Φ ) : = E ( g , f Φ ( s ) ) = ∑ γ ∈ P 2 r ( F ) \ G 2 r ( F ) f Φ ( s ) ( γ g ) , g ∈ G 2 r ( A F ) E(s,g,\Phi):=E(g,f_{\Phi}^{(s)})=\sum_{\gamma\in P_{2r}(F)\backslash G_{2r}(F)}f_{\Phi}^{(s)}(\gamma g),\qquad g\in G_{2r}(\mathbb{A}_F) E ( s , g , Φ ) := E ( g , f Φ ( s ) ) = γ ∈ P 2 r ( F ) \ G 2 r ( F ) ∑ f Φ ( s ) ( γ g ) , g ∈ G 2 r ( A F )
which is meromorphic in s s s and holomorphic on the imaginary line.
For every place v v v of F F F , T □ ∈ Herm 2 r ∘ ( F v ) T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F_v) T □ ∈ Herm 2 r ∘ ( F v ) , and Φ v ∈ S ( V v 2 r ) ⊗ C S v \Phi_v\in\mathcal{S}(V_v^{2r})\otimes_{\mathbb{C}}S_v Φ v ∈ S ( V v 2 r ) ⊗ C S v , we have the local Whittaker function on G 2 r ( F v ) G_{2r}(F_v) G 2 r ( F v ) with parameter s ∈ C s\in\mathbb{C} s ∈ C as
W T □ ( s , g , Φ v ) : = ∫ Herm 2 r ( F v ) f Φ v ( s ) ( w 2 r n ( b ) g ) ψ F , v ( tr b T □ ) − 1 d b (4) W_{T^{\square}}(s,g,\Phi_v):=\int_{\operatorname{Herm}_{2r}(F_v)}f_{\Phi_v}^{(s)}(w_{2r}n(b)g)\psi_{F,v}(\operatorname{tr}bT^{\square})^{-1}\,db \tag*{(4)} W T □ ( s , g , Φ v ) := ∫ Herm 2 r ( F v ) f Φ v ( s ) ( w 2 r n ( b ) g ) ψ F , v ( tr b T □ ) − 1 d b (4)
by meromorphic continuation (which is indeed entire), where d b db d b is the self-dual measure on Herm 2 r ( F v ) \operatorname{Herm}_{2r}(F_v) Herm 2 r ( F v ) with respect to ψ F , v \psi_{F,v} ψ F , v .
Notation 2.19. Consider an element T □ ∈ Herm 2 r ( F ) T^{\square}\in\operatorname{Herm}_{2r}(F) T □ ∈ Herm 2 r ( F ) .
(1) Put ∂ T □ = ( T 1 , T 2 ) ∈ Herm r ( F ) 2 \partial T^{\square}=(T_1,T_2)\in\operatorname{Herm}_r(F)^2 ∂ T □ = ( T 1 , T 2 ) ∈ Herm r ( F ) 2 if
T = ( T 1 ∗ ∗ T 2 ) , T=\begin{pmatrix}T_1&*\\ *&T_2\end{pmatrix}, T = ( T 1 ∗ ∗ T 2 ) , which belongs to ( Herm r ( F ) + ) 2 (\operatorname{Herm}_r(F)^+)^2 ( Herm r ( F ) + ) 2 if T □ ∈ Herm 2 r ∘ ( F ) + T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F)^+ T □ ∈ Herm 2 r ∘ ( F ) + .
(2) If T □ ∈ Herm 2 r ∘ ( F ) T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F) T □ ∈ Herm 2 r ∘ ( F ) , then we put
Diff ( T □ , V ) : = { v ∈ V F ∣ ( V v 2 r ) T = ∅ } , \operatorname{Diff}(T^{\square},V):=\{v\in\mathcal{V}_F\mid(V_v^{2r})_{T}=\varnothing\}, Diff ( T □ , V ) := { v ∈ V F ∣ ( V v 2 r ) T = ∅ } , which is a finite set of even/odd cardinality when V V V is coherent/incoherent, consisting of places of F F F that are nonsplit in E E E .
Proposition 2.20. Let ⋄ \diamond ⋄ be a nonempty finite subset of V F f i n \mathcal{V}_F^{\mathrm{fin}} V F fin . Let Φ = ⊗ v Φ v ∈ S ( V 2 r ) ⊗ C S \Phi=\otimes_v\Phi_v\in\mathcal{S}(V^{2r})\otimes_{\mathbb{C}}S Φ = ⊗ v Φ v ∈ S ( V 2 r ) ⊗ C S be a decomposable Schwartz function such that Φ ⋄ \Phi_{\diamond} Φ ⋄ is regularly supported (Definition 2.16).
(1) For every g ∈ P 2 r ( F ⋄ ) G 2 r ( A F ) g\in P_{2r}(F_{\diamond})G_{2r}(\mathbb{A}_F) g ∈ P 2 r ( F ⋄ ) G 2 r ( A F ) , we have
E ( 0 , g , Φ ) = ∑ T □ ∈ Herm 2 r ( F ) ∏ v ∈ V F W T □ ( 0 , g v , Φ v ) . E(0,g,\Phi)=\sum_{T^{\square}\in\operatorname{Herm}_{2r}(F)}\prod_{v\in\mathcal{V}_F}W_{T^{\square}}(0,g_v,\Phi_v). E ( 0 , g , Φ ) = T □ ∈ Herm 2 r ( F ) ∑ v ∈ V F ∏ W T □ ( 0 , g v , Φ v ) .
(2) Suppose that ⧫ \blacklozenge ⧫ has cardinality at least 2 and that V V V is incoherent (Definition 2.15) so that E ( 0 , g , Φ ) ≡ 0 E(0,g,\Phi)\equiv0 E ( 0 , g , Φ ) ≡ 0 . Then for every g ∈ P 2 r ( F ⧫ ) G 2 r ( A F ⧫ ) g\in P_{2r}(F_{\blacklozenge})G_{2r}(\mathbb{A}_{F}^{\blacklozenge}) g ∈ P 2 r ( F ⧫ ) G 2 r ( A F ⧫ ) , we have
E ′ ( 0 , g , Φ ) = ∑ v ∈ B F C ( g , Φ ) v . E'(0,g,\Phi)=\sum_{v\in\mathcal{B}_{F}}\mathfrak{C}(g,\Phi)_{v}. E ′ ( 0 , g , Φ ) = v ∈ B F ∑ C ( g , Φ ) v .
where11
C ( g , Φ ) v : = ∑ T ⧫ ∈ Herm 2 r ( F ) Diff ( T ⧫ , V ) = { v } W T ⧫ ′ ( 0 , g v , Φ v ) ∏ v ′ ∈ B F ∖ { v } W T ⧫ ( 0 , g v ′ , Φ v ′ ) . \mathfrak{C}(g,\Phi)_{v}:=\sum_{\substack{T^{\blacklozenge}\in\operatorname{Herm}_{2r}(F)\\ \operatorname{Diff}(T^{\blacklozenge},V)=\{v\}}}W'_{T^{\blacklozenge}}(0,g_{v},\Phi_{v})\prod_{v'\in\mathcal{B}_{F}\setminus\{v\}}W_{T^{\blacklozenge}}(0,g_{v'},\Phi_{v'}). C ( g , Φ ) v := T ⧫ ∈ Herm 2 r ( F ) Diff ( T ⧫ , V ) = { v } ∑ W T ⧫ ′ ( 0 , g v , Φ v ) v ′ ∈ B F ∖ { v } ∏ W T ⧫ ( 0 , g v ′ , Φ v ′ ) .
(3) Suppose that for every v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ , v v v is nonsplit in $E such that Φ v \Phi_{v} Φ v is deeply regularly supported (Definition 2.16), and that V V V is incoherent so that E ( 0 , g , Φ ) ≡ 0 E(0,g,\Phi)\equiv0 E ( 0 , g , Φ ) ≡ 0 . Then for every g ∈ G 2 r ( A F ⧫ ) g\in G_{2r}(\mathbb{A}_{F}^{\blacklozenge}) g ∈ G 2 r ( A F ⧫ ) , we have
E ′ ( 0 , g , Φ ) = ∑ v ∈ B F C ( g , Φ ) v , E'(0,g,\Phi)=\sum_{v\in\mathcal{B}_{F}}\mathfrak{C}(g,\Phi)_{v}, E ′ ( 0 , g , Φ ) = v ∈ B F ∑ C ( g , Φ ) v ,
where C ( g , Φ ) v \mathfrak{C}(g,\Phi)_{v} C ( g , Φ ) v is the one in (2).
Proof. Parts (1) and (2) are proved in [26 Section 2B] . Now we show (3). Write E T ⧫ ( s , g , Φ ) E_{T^{\blacklozenge}}(s,g,\Phi) E T ⧫ ( s , g , Φ ) for the T ⧫ T^{\blacklozenge} T ⧫ -th Fourier coefficient of E ( s , g , Φ ) E(s,g,\Phi) E ( s , g , Φ ) , so that
E ( s , g , Φ ) = ∑ T ⧫ ∈ Herm 2 r ( F ) E T ⧫ ( s , g , Φ ) . E(s,g,\Phi)=\sum_{T^{\blacklozenge}\in\operatorname{Herm}_{2r}(F)}E_{T^{\blacklozenge}}(s,g,\Phi). E ( s , g , Φ ) = T ⧫ ∈ Herm 2 r ( F ) ∑ E T ⧫ ( s , g , Φ ) .
It suffices to show that E T ⧫ ( s , g , Φ ) ≡ 0 E_{T^{\blacklozenge}}(s,g,\Phi)\equiv0 E T ⧫ ( s , g , Φ ) ≡ 0 for T ⧫ ∉ Herm 2 r ∘ ( F ) T^{\blacklozenge}\notin\operatorname{Herm}^{\circ}_{2r}(F) T ⧫ ∈ / Herm 2 r ∘ ( F ) and g ∈ G 2 r ( A F ⧫ ) g\in G_{2r}(\mathbb{A}_{F}^{\blacklozenge}) g ∈ G 2 r ( A F ⧫ ) . Without loss of generality, we may assume g = 1 4 r g=1_{4r} g = 1 4 r . Take an element T ⧫ ∈ Herm 2 r ( F ) ∖ Herm 2 r ∘ ( F ) T^{\blacklozenge}\in\operatorname{Herm}_{2r}(F)\setminus\operatorname{Herm}^{\circ}_{2r}(F) T ⧫ ∈ Herm 2 r ( F ) ∖ Herm 2 r ∘ ( F ) . By [26 Lemma 2.2] , we have
E T ⧫ ( s , 1 4 r , Φ ) = ∏ v ∈ B F W T ⧫ ( s , 1 4 r , Φ v ) E_{T^{\blacklozenge}}(s,1_{4r},\Phi)=\prod_{v\in\mathcal{B}_{F}}W_{T^{\blacklozenge}}(s,1_{4r},\Phi_{v}) E T ⧫ ( s , 1 4 r , Φ ) = v ∈ B F ∏ W T ⧫ ( s , 1 4 r , Φ v )
for Re s ≫ 0 \operatorname{Re}s\gg0 Re s ≫ 0 . By Lemma 2.21(1) below, we have for v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ (which is nonempty), W T ⧫ ( s , 1 4 r , Φ v ) ≡ 0 W_{T^{\blacklozenge}}(s,1_{4r},\Phi_{v})\equiv0 W T ⧫ ( s , 1 4 r , Φ v ) ≡ 0 , so that E T ⧫ ( s , g , Φ ) ≡ 0 E_{T^{\blacklozenge}}(s,g,\Phi)\equiv0 E T ⧫ ( s , g , Φ ) ≡ 0 . Part (3) is proved. □
Lemma 2.21. Let v v v be a nonarchimedean place of F F F that is nonsplit in E E E . Consider a deeply regularly supported Schwartz function Φ v ∈ S ( V v 2 r ) ⊗ C S v \Phi_{v}\in\mathcal{S}(V_{v}^{2r})\otimes_{\mathbb{C}}S_{v} Φ v ∈ S ( V v 2 r ) ⊗ C S v (Definition 2.16).
(1) We have
W T ⧫ ( s , 1 4 r , Φ v ) ≡ 0 W_{T^{\blacklozenge}}(s,1_{4r},\Phi_{v})\equiv0 W T ⧫ ( s , 1 4 r , Φ v ) ≡ 0 for every T ⧫ ∈ ( Herm 2 r ( F v ) ∖ Herm 2 r ∘ ( F v ) ) + ϖ d Herm 2 r ( O F v ) T^{\blacklozenge}\in\left(\operatorname{Herm}_{2r}(F_{v})\setminus\operatorname{Herm}^{\circ}_{2r}(F_{v})\right)+\varpi^{d}\operatorname{Herm}_{2r}(\mathcal{O}_{F_{v}}) T ⧫ ∈ ( Herm 2 r ( F v ) ∖ Herm 2 r ∘ ( F v ) ) + ϖ d Herm 2 r ( O F v ) , where d d d is the integer from Definition 2.16(3).
(2) Let V v ⋆ V_{v}^{\star} V v ⋆ be the other (up to isomorphism) hermitian space over E v / F v E_{v}/F_{v} E v / F v of rank 2 r 2r 2 r that is not isomorphic to V v V_{v} V v . There exists a regularly supported Schwartz function Φ v ⋆ ∈ S ( ( V v ⋆ ) 2 r ) ⊗ C S v \Phi_{v}^{\star}\in\mathcal{S}((V_{v}^{\star})^{2r})\otimes_{\mathbb{C}}S_{v} Φ v ⋆ ∈ S (( V v ⋆ ) 2 r ) ⊗ C S v satisfying that
W T ⧫ ( 0 , 1 4 r , Φ v ⋆ ) = W T ⧫ ′ ( 0 , 1 4 r , Φ v ) W_{T^{\blacklozenge}}(0,1_{4r},\Phi_{v}^{\star})=W'_{T^{\blacklozenge}}(0,1_{4r},\Phi_{v}) W T ⧫ ( 0 , 1 4 r , Φ v ⋆ ) = W T ⧫ ′ ( 0 , 1 4 r , Φ v ) holds for every T ⧫ T^{\blacklozenge} T ⧫ in the image in the moment map T : ( V v ⋆ ) 2 r ⟶ Herm 2 r ( F v ) T:(V_{v}^{\star})^{2r}\longrightarrow\operatorname{Herm}_{2r}(F_{v}) T : ( V v ⋆ ) 2 r ⟶ Herm 2 r ( F v ) .
Proof. By (2.3), we have
W T ⧫ ( s , 1 4 r , Φ v ) = ∫ Herm 2 r ( F v ) f Φ v ( s ) ( w 2 r n ( b ) ) ψ F , v ( tr b T ⧫ ) − 1 d b = ∫ Herm 2 r ( F v ) ( ∫ V v 2 r ψ F , v ( tr b T ( x ) ) Φ v ( x ) d x ) h ( w 2 r n ( b ) ) s ψ F , v ( − tr b T ⧫ ) d b , \begin{aligned}
W_{T^{\blacklozenge}}(s,1_{4r},\Phi_{v})=
\int_{\operatorname{Herm}_{2r}(F_{v})}f_{\Phi_{v}}^{(s)}(w_{2r}n(b))\psi_{F,v}(\operatorname{tr}bT^{\blacklozenge})^{-1}\,db \\
&=\int_{\operatorname{Herm}_{2r}(F_{v})}\left(\int_{V_{v}^{2r}}\psi_{F,v}(\operatorname{tr}bT(x))\Phi_{v}(x)\,dx\right)h(w_{2r}n(b))^{s}\psi_{F,v}(-\operatorname{tr}bT^{\blacklozenge})\,db,
\end{aligned} W T ⧫ ( s , 1 4 r , Φ v ) = ∫ Herm 2 r ( F v ) f Φ v ( s ) ( w 2 r n ( b )) ψ F , v ( tr b T ⧫ ) − 1 d b = ∫ Herm 2 r ( F v ) ( ∫ V v 2 r ψ F , v ( tr b T ( x )) Φ v ( x ) d x ) h ( w 2 r n ( b ) ) s ψ F , v ( − tr b T ⧫ ) d b ,
where h h h is the height function with respect to the standard Iwasawa decomposition. Since h ( w 2 r n ( b ) ) h(w_{2r}n(b)) h ( w 2 r n ( b )) is invariant under Herm 2 r ( O F v ) \operatorname{Herm}_{2r}(\mathcal{O}_{F_v}) Herm 2 r ( O F v ) , the above expression equals
= ∫ Herm 2 r ( F v ) / Herm 2 r ( O F v ) h ( w 2 r n ( b ′ ) ) s ( ∫ V v 2 r Φ v ( x ) ( ∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ ) ) d b ) d x ) d b ′ . =\int_{\operatorname{Herm}_{2r}(F_v)/\operatorname{Herm}_{2r}(\mathcal{O}_{F_v})}h(w_{2r}n(b'))^s\left(\int_{V_v^{2r}}\Phi_v(x)\left(\int_{\operatorname{Herm}_{2r}(\mathcal{O}_{F_v})}\psi_{F,v}(\operatorname{tr}b(T(x)-T^\square))\,db\right)\,dx\right)\,db'. = ∫ Herm 2 r ( F v ) / Herm 2 r ( O F v ) h ( w 2 r n ( b ′ ) ) s ( ∫ V v 2 r Φ v ( x ) ( ∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ )) d b ) d x ) d b ′ .
For (1), suppose that T □ ∈ ( Herm 2 r ( F v ) ∖ Herm 2 r ∘ ( F v ) ) + ϖ d Herm 2 r ( O F v ) T^\square\in\left(\operatorname{Herm}_{2r}(F_v)\setminus\operatorname{Herm}_{2r}^{\circ}(F_v)\right)+\varpi^d\operatorname{Herm}_{2r}(\mathcal{O}_{F_v}) T □ ∈ ( Herm 2 r ( F v ) ∖ Herm 2 r ∘ ( F v ) ) + ϖ d Herm 2 r ( O F v ) . For x x x in the support of Φ v \Phi_v Φ v , the character b ↦ ψ F , v ( tr b ( T ( x ) − T □ ) ) b\mapsto\psi_{F,v}(\operatorname{tr}b(T(x)-T^\square)) b ↦ ψ F , v ( tr b ( T ( x ) − T □ )) on Herm 2 r ( O F v ) \operatorname{Herm}_{2r}(\mathcal{O}_{F_v}) Herm 2 r ( O F v ) is nontrivial, which implies that
∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ ) ) d b = 0. \int_{\operatorname{Herm}_{2r}(\mathcal{O}_{F_v})}\psi_{F,v}(\operatorname{tr}b(T(x)-T^\square))\,db=0. ∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ )) d b = 0.
Thus, W T □ ( s , 1 4 r , Φ v ) ≡ 0 W_{T^\square}(s,1_{4r},\Phi_v)\equiv0 W T □ ( s , 1 4 r , Φ v ) ≡ 0 .
For (2), note that the function
( x , T □ ) ∈ V v 2 r × Herm 2 r ∘ ( F v ) ⟼ ∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ ) ) d b (x,T^\square)\in V_v^{2r}\times\operatorname{Herm}_{2r}^{\circ}(F_v)\longmapsto\int_{\operatorname{Herm}_{2r}(\mathcal{O}_{F_v})}\psi_{F,v}(\operatorname{tr}b(T(x)-T^\square))\,db ( x , T □ ) ∈ V v 2 r × Herm 2 r ∘ ( F v ) ⟼ ∫ Herm 2 r ( O F v ) ψ F , v ( tr b ( T ( x ) − T □ )) d b
is locally constant, and compactly supported for every fixed x x x . Since Φ v \Phi_v Φ v is compactly supported, the assignment T □ ↦ W T □ ( s , 1 4 r , Φ v ) T^\square\mapsto W_{T^\square}(s,1_{4r},\Phi_v) T □ ↦ W T □ ( s , 1 4 r , Φ v ) is uniformly locally constant and compactly supported on Herm 2 r ∘ ( F v ) \operatorname{Herm}_{2r}^{\circ}(F_v) Herm 2 r ∘ ( F v ) with respect to s s s . It follows that the assignment T □ ↦ W T □ ′ ( 0 , 1 4 r , Φ v ) T^\square\mapsto W'_{T^\square}(0,1_{4r},\Phi_v) T □ ↦ W T □ ′ ( 0 , 1 4 r , Φ v ) is locally constant and compactly supported on Herm 2 r ∘ ( F v ) \operatorname{Herm}_{2r}^{\circ}(F_v) Herm 2 r ∘ ( F v ) . By (1), this assignment extends (by zero) to a locally constant and compactly supported function on Herm 2 r ( F v ) \operatorname{Herm}_{2r}(F_v) Herm 2 r ( F v ) . Thus, we can find a regularly supported Schwartz function Φ v ⋆ ∈ S ( ( V v ⋆ ) 2 r ) ⊗ C S v \Phi_v^\star\in\mathcal{S}((V_v^\star)^{2r})\otimes_{\mathbb{C}}S_v Φ v ⋆ ∈ S (( V v ⋆ ) 2 r ) ⊗ C S v whose pushforward along the moment map coincides with the assignment T □ ↦ W T □ ′ ( 0 , 1 4 r , Φ v ) T^\square\mapsto W'_{T^\square}(0,1_{4r},\Phi_v) T □ ↦ W T □ ′ ( 0 , 1 4 r , Φ v ) for T □ T^\square T □ in the image of T : ( V v ⋆ ) 2 r → Herm 2 r ( F v ) T:(V_v^\star)^{2r}\to\operatorname{Herm}_{2r}(F_v) T : ( V v ⋆ ) 2 r → Herm 2 r ( F v ) .
The lemma is proved.
Definition 2.22. Consider a pair ( T 1 , T 2 ) (T_1,T_2) ( T 1 , T 2 ) of elements in Herm r ( F ) \operatorname{Herm}_r(F) Herm r ( F ) and a decomposable Schwartz function Φ ∈ S ( V 2 r ) ⊗ C S \Phi\in\mathcal{S}(V^{2r})\otimes_{\mathbb{C}}S Φ ∈ S ( V 2 r ) ⊗ C S .
(1) Suppose that V V V is coherent. We put
E T 1 , T 2 ( g , Φ ) : = ∑ T □ ∈ Herm 2 r ∘ ( F ) ∂ T □ = ( T 1 , T 2 ) ∏ v ∈ V F W T □ ( 0 , g v , Φ v ) . E_{T_1,T_2}(g,\Phi):=\sum_{\substack{T^\square\in\operatorname{Herm}_{2r}^{\circ}(F)\\ \partial T^\square=(T_1,T_2)}}\prod_{v\in\mathcal{V}_F}W_{T^\square}(0,g_v,\Phi_v). E T 1 , T 2 ( g , Φ ) := T □ ∈ Herm 2 r ∘ ( F ) ∂ T □ = ( T 1 , T 2 ) ∑ v ∈ V F ∏ W T □ ( 0 , g v , Φ v ) .
(2) Suppose that V V V is incoherent. For every v ∈ V F v\in\mathcal{V}_F v ∈ V F , we put
E T 1 , T 2 ( g , Φ ) v : = ∑ T □ ∈ Herm 2 r ∘ ( F ) Diff ( T □ , V ) = { v } ∂ T □ = ( T 1 , T 2 ) W T □ ′ ( 0 , g v , Φ v ) ∏ v ′ ∈ V F ∖ { v } W T □ ( 0 , g v ′ , Φ v ′ ) . \mathcal{E}_{T_1,T_2}(g,\Phi)_v:=\sum_{\substack{T^\square\in\operatorname{Herm}_{2r}^{\circ}(F)\\ \operatorname{Diff}(T^\square,V)=\{v\}\\ \partial T^\square=(T_1,T_2)}}W'_{T^\square}(0,g_v,\Phi_v)\prod_{v'\in\mathcal{V}_F\setminus\{v\}}W_{T^\square}(0,g_{v'},\Phi_{v'}). E T 1 , T 2 ( g , Φ ) v := T □ ∈ Herm 2 r ∘ ( F ) Diff ( T □ , V ) = { v } ∂ T □ = ( T 1 , T 2 ) ∑ W T □ ′ ( 0 , g v , Φ v ) v ′ ∈ V F ∖ { v } ∏ W T □ ( 0 , g v ′ , Φ v ′ ) .
Take a balanced hermitian E E E -weight λ \lambda λ of rank 2 r 2r 2 r (Definition 2.7) and suppose that V V V is totally positive definite. Apply Remark 2.17 with S v : = S λ ∨ ( C r ) ⊗ C S λ ∨ ( C r ) S_v:=S_{\lambda}^{\vee}(\mathbb{C}^r)\otimes_{\mathbb{C}}S_{\lambda}^{\vee}(\mathbb{C}^r) S v := S λ ∨ ( C r ) ⊗ C S λ ∨ ( C r ) for v ∈ V F ( ∞ ) v\in\mathcal{V}_F^{(\infty)} v ∈ V F ( ∞ ) and S v = C S_v=\mathbb{C} S v = C for v ∈ V F f i n v\in\mathcal{V}_F^{\mathrm{fin}} v ∈ V F fin .
Definition 2.23. Define a function
Φ V ∞ 2 r λ : = κ V ∞ 2 r λ ⋅ Φ V ∞ 2 r 0 : V ∞ 2 r ⟶ S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ‾ ∨ . \Phi_{V_\infty^{2r}}^\lambda:=\kappa_{V_\infty^{2r}}^\lambda\cdot\Phi_{V_\infty^{2r}}^0:V_\infty^{2r}\longrightarrow S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{\mathbb{C}}\overline{S_{\lambda+}(\mathbb{C}^r)}^\vee. Φ V ∞ 2 r λ := κ V ∞ 2 r λ ⋅ Φ V ∞ 2 r 0 : V ∞ 2 r ⟶ S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ .
(Notation 1.11 for κ \kappa κ with L = E λ L=\mathbb{E}^\lambda L = E λ ), where Ψ V ∞ 2 r λ \Psi_{V_\infty^{2r}}^\lambda Ψ V ∞ 2 r λ is from Construction 2.8(4) and Φ V ∞ 2 r 0 \Phi_{V_\infty^{2r}}^0 Φ V ∞ 2 r 0 is the standard Gaussian function on V ∞ 2 r V_\infty^{2r} V ∞ 2 r given by the formula
Φ V ∞ 2 r 0 ( x ) = ∏ v ∈ V F ( ∞ ) e − 2 π tr T ( x v ) . \Phi_{V_\infty^{2r}}^0(x)=\prod_{v\in\mathcal{V}_F^{(\infty)}}e^{-2\pi\operatorname{tr}T(x_v)}. Φ V ∞ 2 r 0 ( x ) = v ∈ V F ( ∞ ) ∏ e − 2 π tr T ( x v ) .
In particular, Φ V ∞ 2 r λ \Phi_{V_\infty^{2r}}^\lambda Φ V ∞ 2 r λ is an element in S ( V ∞ 2 r ) ⊗ C S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ‾ ∨ \mathcal{S}(V_\infty^{2r})\otimes_{\mathbb{C}}S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{\mathbb{C}}\overline{S_{\lambda+}(\mathbb{C}^r)}^\vee S ( V ∞ 2 r ) ⊗ C S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ .
Remark 2.24. The Schwartz function Φ V ∞ 2 r λ \Phi_{V^{2r}_{\infty}}^{\lambda} Φ V ∞ 2 r λ can be decomposed as the product of local ones over archimedean places of F F F .
Take an element v ∈ V F ( ∞ ) v \in\mathscr{V}^{(\infty)}_{F} v ∈ V F ( ∞ ) . For every element x ∈ V v r x \in V_v^r x ∈ V v r , we have the map
μ [ λ v ] ( x ) : S [ λ v ] ( V v ) ∨ → S λ + ∨ ( C r ) \mu_{[\lambda_v]}(x): S_{[\lambda_v]}(V_v)^{\vee} \to S_{\lambda_+}^{\vee}(\mathbb{C}^r) μ [ λ v ] ( x ) : S [ λ v ] ( V v ) ∨ → S λ + ∨ ( C r )
from Construction 2.5 with E / F = E v / F v E/F=E_v/F_v E / F = E v / F v and L = C \mathbb{L}=\mathbb{C} L = C (as the natural F v F_v F v -ring), and its dual
μ [ λ v ] ( x ) : S λ + ( C r ) → S [ λ v ] ( V v ) . \mu^{[\lambda_v]}(x): S_{\lambda_+}(\mathbb{C}^r) \to S_{[\lambda_v]}(V_v). μ [ λ v ] ( x ) : S λ + ( C r ) → S [ λ v ] ( V v ) .
Define a function
Φ λ v : V v 2 r → S λ + ∨ ( C r ) ⊗ C S λ + ∨ ( C r ) \Phi^{\lambda_v}: V_v^{2r} \to S_{\lambda_+}^{\vee}(\mathbb{C}^r) \otimes_{\mathbb{C}} S_{\lambda_+}^{\vee}(\mathbb{C}^r) Φ λ v : V v 2 r → S λ + ∨ ( C r ) ⊗ C S λ + ∨ ( C r )
such that for ( x 1 , x 2 ) ∈ ( V v r ) 2 = V v 2 r (x_1,x_2) \in(V_v^r)^2=V_v^{2r} ( x 1 , x 2 ) ∈ ( V v r ) 2 = V v 2 r ,
Φ λ v ( x 1 , x 2 ) = e − 2 π tr T ( x 1 , x 2 ) ⋅ μ [ λ v ] ( x 2 ) ‾ ∘ μ [ λ v ] ( x 1 ) . \Phi^{\lambda_v}(x_1,x_2)=e^{-2\pi\operatorname{tr} T(x_1,x_2)}\cdot\overline{\mu_{[\lambda_v]}(x_2)}\circ\mu^{[\lambda_v]}(x_1). Φ λ v ( x 1 , x 2 ) = e − 2 π tr T ( x 1 , x 2 ) ⋅ μ [ λ v ] ( x 2 ) ∘ μ [ λ v ] ( x 1 ) .
Then we have
Φ V ∞ 2 r λ = ⨂ v ∈ V F ( ∞ ) Φ λ v . \Phi^{\lambda}_{V^{2r}_{\infty}}=\bigotimes_{v\in\mathscr{V}^{(\infty)}_{F}}\Phi^{\lambda_v}. Φ V ∞ 2 r λ = v ∈ V F ( ∞ ) ⨂ Φ λ v .
Lemma 2.25. For every pair ( k 1 , k 2 ) ∈ K r , ∞ × K r , ∞ (k_1,k_2)\in K_{r,\infty}\times K_{r,\infty} ( k 1 , k 2 ) ∈ K r , ∞ × K r , ∞ , we have
ω 2 r , ∞ ( k 1 , k 2 ) Φ V ∞ 2 r λ = δ r ( k 1 ) δ r ( k 2 ) ⋅ ( ρ λ + ( k 1 ) ∨ ‾ ⊗ ρ λ + ( k 2 ) ) ( Φ V ∞ 2 r λ ) . \omega_{2r,\infty}(k_1,k_2)\Phi^{\lambda}_{V^{2r}_{\infty}}
=\delta^r(k_1)\delta^r(k_2)\cdot\left(\overline{\rho_{\lambda_+}(k_1)^{\vee}}\otimes\rho_{\lambda_+}(k_2)\right)(\Phi^{\lambda}_{V^{2r}_{\infty}}). ω 2 r , ∞ ( k 1 , k 2 ) Φ V ∞ 2 r λ = δ r ( k 1 ) δ r ( k 2 ) ⋅ ( ρ λ + ( k 1 ) ∨ ⊗ ρ λ + ( k 2 ) ) ( Φ V ∞ 2 r λ ) .
Proof. This follows from [28 Lemma 3.7] with m = 2 r m=2r m = 2 r and q = 0 q=0 q = 0 . □
Definition 2.26. Take a place v ∈ V F ( ∞ ) v\in\mathscr{V}^{(\infty)}_{F} v ∈ V F ( ∞ ) .
Denote by π ( λ v ) \pi(\lambda_v) π ( λ v ) the unique up to isomorphism holomorphic discrete series representation of G r ( F v ) G_r(F_v) G r ( F v ) such that its base change BC ( π ( λ v ) ) \operatorname{BC}(\pi(\lambda_v)) BC ( π ( λ v )) to E v E_v E v (as an irreducible admissible representation of GL 2 r ( E v ) \operatorname{GL}_{2r}(E_v) GL 2 r ( E v ) ) is the normalized induction of
arg 2 b u , 1 + 2 r − 1 ⊠ arg 2 b u , 2 + 2 r − 3 ⊠ ⋯ ⊠ arg 2 b u , 2 r − 1 + 3 − 2 r ⊠ arg 2 b u , 2 r + 1 − 2 r , λ u = ( b u , 1 , … , b u , 2 r ) \operatorname{arg}^{2b_{u,1}+2r-1}\boxtimes\operatorname{arg}^{2b_{u,2}+2r-3}\boxtimes\cdots\boxtimes\operatorname{arg}^{2b_{u,2r-1}+3-2r}\boxtimes\operatorname{arg}^{2b_{u,2r+1}-2r},
\qquad\lambda_u=(b_{u,1},\ldots,b_{u,2r}) arg 2 b u , 1 + 2 r − 1 ⊠ arg 2 b u , 2 + 2 r − 3 ⊠ ⋯ ⊠ arg 2 b u , 2 r − 1 + 3 − 2 r ⊠ arg 2 b u , 2 r + 1 − 2 r , λ u = ( b u , 1 , … , b u , 2 r )
if we identify E v E_v E v with C \mathbb{C} C via u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) above v v v , where arg : C × → C × \operatorname{arg}:\mathbb{C}^{\times}\to\mathbb{C}^{\times} arg : C × → C × denotes the argument character.
The local L L L -function L ( s , π ( λ v ) ) L(s,\pi(\lambda_v)) L ( s , π ( λ v )) of π ( λ v ) \pi(\lambda_v) π ( λ v ) is defined to be L ( s , BC ( π ( λ v ) ) ) L(s,\operatorname{BC}(\pi(\lambda_v))) L ( s , BC ( π ( λ v ))) , so that
L ( λ v ) : = L ( 1 / 2 , π ( λ v ) ) = ∏ v ( u ) = v ∏ i = 1 r 2 ( 2 π ) − ( b u , i + r + 1 − i ) Γ ( b u , i + r + 1 − i ) ∈ R > 0 . L(\lambda_v):=L(1/2,\pi(\lambda_v))=\prod_{v(u)=v}\prod_{i=1}^{r}2(2\pi)^{-(b_{u,i+r+1-i})}\Gamma(b_{u,i}+r+1-i)\in\mathbb{R}_{>0}. L ( λ v ) := L ( 1/2 , π ( λ v )) = v ( u ) = v ∏ i = 1 ∏ r 2 ( 2 π ) − ( b u , i + r + 1 − i ) Γ ( b u , i + r + 1 − i ) ∈ R > 0 .
Denote by ξ λ v : G r ( F v ) → S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ‾ ∨ \xi^{\lambda_v}:G_r(F_v)\to S_{\lambda_+}(\mathbb{C}^r)^{\vee}\otimes_{\mathbb{C}}\overline{S_{\lambda_+}(\mathbb{C}^r)}^{\vee} ξ λ v : G r ( F v ) → S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ the vector-valued matrix coefficient satisfying
ξ λ v ( g ) = ( φ 1 , π ( λ v ) ( g ) φ 2 ) π ( λ v ) \xi^{\lambda_v}(g)=(\varphi_1,\pi(\lambda_v)(g)\varphi_2)_{\pi(\lambda_v)} ξ λ v ( g ) = ( φ 1 , π ( λ v ) ( g ) φ 2 ) π ( λ v )
for arbitrary K r , v K_{r,v} K r , v -invariant vectors φ 1 ∈ S λ + ( C r ) ∨ ⊗ π ( λ v ) \varphi_1\in S_{\lambda_+}(\mathbb{C}^r)^{\vee}\otimes\pi(\lambda_v) φ 1 ∈ S λ + ( C r ) ∨ ⊗ π ( λ v ) and φ 2 ∈ S λ + ( C r ) ‾ ∨ ⊗ π ( λ v ) ∨ \varphi_2\in\overline{S_{\lambda_+}(\mathbb{C}^r)}^{\vee}\otimes\pi(\lambda_v)^{\vee} φ 2 ∈ S λ + ( C r ) ∨ ⊗ π ( λ v ) ∨ (each exists uniquely up to a constant) satisfying tr λ v ( φ 1 , φ 2 ) π ( λ v ) = 1 \operatorname{tr}_{\lambda_v}(\varphi_1,\varphi_2)_{\pi(\lambda_v)}=1 tr λ v ( φ 1 , φ 2 ) π ( λ v ) = 1 . Put
Z ( λ v ) : = ∫ G r ( F v ) ξ λ v ( g ) ‾ ⋅ f Φ λ v ( 0 ) ( w r ( g , 1 2 r ) ) d g Z(\lambda_v):=\int_{G_r(F_v)}\overline{\xi^{\lambda_v}(g)}\cdot f_{\Phi^{(0)}_{\lambda_v}}(w_r(g,1_{2r}))\,dg Z ( λ v ) := ∫ G r ( F v ) ξ λ v ( g ) ⋅ f Φ λ v ( 0 ) ( w r ( g , 1 2 r )) d g
(see Notation 2.13(6) for the measure d g dg d g and Remark 2.24 for Φ λ v \Phi^{\lambda_v} Φ λ v ), which is absolutely convergent and depends only on λ v \lambda_v λ v . Since the map g ∈ G r ( F v ) ↦ ( − 1 ) r ⋅ f Φ λ v ( 0 ) ( w r ( g , 1 2 r ) ) g\in G_r(F_v)\mapsto(-1)^r\cdot f_{\Phi^{(0)}_{\lambda_v}}(w_r(g,1_{2r})) g ∈ G r ( F v ) ↦ ( − 1 ) r ⋅ f Φ λ v ( 0 ) ( w r ( g , 1 2 r )) is a matrix coefficient of positive type, ( − 1 ) r Z ( λ v ) ∈ R ≥ 0 (-1)^rZ(\lambda_v)\in\mathbb{R}_{\ge0} ( − 1 ) r Z ( λ v ) ∈ R ≥ 0 . On the other hand, Z ( λ v ) Z(\lambda_v) Z ( λ v ) is nonzero by [11 Theorem 1.3.1] .12 It follows that Z ( λ v ) ∈ ( − 1 ) r R > 0 Z(\lambda_v)\in(-1)^r\mathbb{R}_{>0} Z ( λ v ) ∈ ( − 1 ) r R > 0 .
Definition 2.27. We say that a pair ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) , in which π \pi π is an irreducible representation of G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) and A π \mathcal{A}_{\pi} A π is a subspace of A ( G r ) c u s p \mathcal{A}(G_{r})_{\mathrm{cusp}} A ( G r ) cusp on which π \pi π realizes,13 is a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) of weight λ \lambda λ if
π v \pi_{v} π v is tempered for every v ∈ V F v \in\mathcal{V}_{F} v ∈ V F ;
π v \pi_{v} π v is a discrete series representation for every v ∈ V F ( ∞ ) v \in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) ;
( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) c u s p h o l (S_{\lambda+}(\mathbb{C}^{r})^{\vee} \otimes_{\mathbb{C}} \mathcal{A}_{\pi}) \cap\mathcal{A}_{\lambda+}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}} ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) cusp hol (Definition 2.14(1)) is nonzero.
Note that (2) plus (3) is equivalent to that π v ≃ π ( λ v ) \pi_{v} \simeq\pi(\lambda_{v}) π v ≃ π ( λ v ) for every v ∈ V F ( ∞ ) v \in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) .
Define the (global) L L L -function for π \pi π to be the Euler product
L ( s , π ) : = ∏ v ∈ V F L ( s , π v ) , L(s,\pi) := \prod_{v\in\mathcal{V}_{F}} L(s,\pi_{v}), L ( s , π ) := v ∈ V F ∏ L ( s , π v ) ,
in which L ( s , π v ) L(s,\pi_{v}) L ( s , π v ) is the doubling L L L -function defined in [38 Theorem 5.2] for v ∈ V F f i n v \in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin and L ( s , π v ) = L ( s , π ( λ v ) ) L(s,\pi_{v})=L(s,\pi(\lambda_{v})) L ( s , π v ) = L ( s , π ( λ v )) for v ∈ V F ( ∞ ) v \in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) . The L L L -function L ( s , π ) L(s,\pi) L ( s , π ) satisfies a self-dual functional equation with the center s = 1 / 2 s=1/2 s = 1/2 and the epsilon factor ϵ ( s , π ) \epsilon(s,\pi) ϵ ( s , π ) . Put ϵ ( π ) : = ϵ ( 1 / 2 , π ) ∈ { ± 1 } \epsilon(\pi):=\epsilon(1/2,\pi)\in\{\pm1\} ϵ ( π ) := ϵ ( 1/2 , π ) ∈ { ± 1 } .
Take a holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) of G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) of weight λ \lambda λ .
Definition 2.28. Consider a pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_{1},\varphi_{2}) φ = ( φ 1 , φ 2 ) of automorphic forms in ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) c u s p h o l (S_{\lambda+}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\mathcal{A}_{\pi})\cap\mathcal{A}_{\lambda+}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}} ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) cusp hol and a Schwartz function Φ ∞ ∈ S ( ( V ∞ ) 2 r ) \Phi^{\infty}\in\mathcal{S}((V^{\infty})^{2r}) Φ ∞ ∈ S (( V ∞ ) 2 r ) . For every finite set ⋄ \diamond ⋄ of nonarchimedean places of F F F , we have the ⋄ \diamond ⋄ -local doubling zeta integral
Z ⋄ ( s , φ , Φ ∞ ) : = ∫ G r ( F ⋄ ) ( φ 1 , π ( g ) φ 2 ) G r ⋅ f Φ ∞ ( s ) ( w r ( g , 1 2 r ) ) d g \mathcal{Z}_{\diamond}(s,\varphi,\Phi^{\infty}) := \int_{G_{r}(F_{\diamond})}(\varphi_{1},\pi(g)\varphi_{2})_{G_{r}}\cdot f_{\Phi^{\infty}}^{(s)}(w_{r}(g,1_{2r}))\,dg Z ⋄ ( s , φ , Φ ∞ ) := ∫ G r ( F ⋄ ) ( φ 1 , π ( g ) φ 2 ) G r ⋅ f Φ ∞ ( s ) ( w r ( g , 1 2 r )) d g (Notation 2.13(6) and Definition 2.14(2)) and its normalized version
Z ⋄ ♮ ( s , φ , Φ ∞ ) : = ( L ( s + 1 2 , π ⋄ ) b 2 r , ⋄ ( s ) ) − 1 ⋅ Z ⋄ ( s , φ , Φ ∞ ) , \mathcal{Z}_{\diamond}^{\natural}(s,\varphi,\Phi^{\infty}) := \left(\frac{L(s+\frac{1}{2},\pi_{\diamond})}{b_{2r,\diamond}(s)}\right)^{-1}\cdot\mathcal{Z}_{\diamond}(s,\varphi,\Phi^{\infty}), Z ⋄ ♮ ( s , φ , Φ ∞ ) := ( b 2 r , ⋄ ( s ) L ( s + 2 1 , π ⋄ ) ) − 1 ⋅ Z ⋄ ( s , φ , Φ ∞ ) , in which the latter has a holomorphic continuation to the entire complex plane. Here, b 2 r , ⋄ : = ∏ v ∈ ⋄ b 2 r , v b_{2r,\diamond}:=\prod_{v\in\diamond}b_{2r,v} b 2 r , ⋄ := ∏ v ∈ ⋄ b 2 r , v (Notation 2.6(3)).
By [38 Proposition 7.1 & (7.2)] , the function Z ⋄ ♮ ( s , φ , Φ ∞ ) \mathcal{Z}_{\diamond}^{\natural}(s,\varphi,\Phi^{\infty}) Z ⋄ ♮ ( s , φ , Φ ∞ ) stabilizes when ⋄ \diamond ⋄ enlarges, which we denote by Z f i n ♮ ( s , φ , Φ ∞ ) \mathcal{Z}_{\mathrm{fin}}^{\natural}(s,\varphi,\Phi^{\infty}) Z fin ♮ ( s , φ , Φ ∞ ) . Finally, put Z f i n ♮ ( φ , Φ ∞ ) : = Z f i n ♮ ( 0 , φ , Φ ∞ ) \mathcal{Z}_{\mathrm{fin}}^{\natural}(\varphi,\Phi^{\infty}):=\mathcal{Z}_{\mathrm{fin}}^{\natural}(0,\varphi,\Phi^{\infty}) Z fin ♮ ( φ , Φ ∞ ) := Z fin ♮ ( 0 , φ , Φ ∞ ) .
Proposition 2.29. Let λ \lambda λ be a balanced hermitian E E E -weight λ \lambda λ of rank 2 r 2r 2 r , and ( π , V π ) (\pi,V_{\pi}) ( π , V π ) a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_{r}(\mathbb{A}_{F}) G r ( A F ) of weight λ \lambda λ . Suppose that V V V is totally positive definite. For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_{1},\varphi_{2}) φ = ( φ 1 , φ 2 ) of automorphic forms in ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) c u s p h o l (S_{\lambda+}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\mathcal{A}_{\pi})\cap\mathcal{A}_{\lambda+}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}} ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) cusp hol and every Schwartz function Φ ∞ ∈ S ( ( V ∞ ) 2 r ) \Phi^{\infty}\in\mathcal{S}((V^{\infty})^{2r}) Φ ∞ ∈ S (( V ∞ ) 2 r ) ,
∫ G r ( F ) \ G r ( A F ) tr λ ( φ 2 ( g 2 ) ⊗ ∫ G r ( F ) \ G r ( A F ) tr λ ( E ′ ( 0 , ( g 1 , g 2 ) , Φ V ∞ ∨ 2 r Φ ∞ ) ⊗ φ 1 ( g 1 ) ) d g 1 ) d g 2 = L ′ ( 1 2 , π ) b 2 r ( 0 ) ⋅ C ( λ ) ⋅ Z f i n ♮ ( φ , Φ ∞ ) \int_{G_{r}(F)\backslash G_{r}(\mathbb{A}_{F})}\operatorname{tr}_{\lambda}\left(\varphi_{2}(g_{2})\otimes\int_{G_{r}(F)\backslash G_{r}(\mathbb{A}_{F})}\operatorname{tr}_{\lambda}\left(E'(0,(g_{1},g_{2}),\Phi_{V^{\vee}_{\infty}}^{2r}\Phi^{\infty})\otimes\varphi_{1}(g_{1})\right)\,dg_{1}\right)\,dg_{2}
=
\frac{L'(\frac{1}{2},\pi)}{b_{2r}(0)}\cdot C(\lambda)\cdot\mathcal{Z}_{\mathrm{fin}}^{\natural}(\varphi,\Phi^{\infty}) ∫ G r ( F ) \ G r ( A F ) tr λ ( φ 2 ( g 2 ) ⊗ ∫ G r ( F ) \ G r ( A F ) tr λ ( E ′ ( 0 , ( g 1 , g 2 ) , Φ V ∞ ∨ 2 r Φ ∞ ) ⊗ φ 1 ( g 1 ) ) d g 1 ) d g 2 = b 2 r ( 0 ) L ′ ( 2 1 , π ) ⋅ C ( λ ) ⋅ Z fin ♮ ( φ , Φ ∞ ) holds. Here, C ( λ ) : = ∏ v ∈ V F ( ∞ ) C ( λ v ) C(\lambda):=\prod_{v\in\mathcal{V}_{F}^{(\infty)}}C(\lambda_{v}) C ( λ ) := ∏ v ∈ V F ( ∞ ) C ( λ v ) , and
C ( λ v ) : = b 2 r , v ( 0 ) ⋅ L ( λ v ) − 1 ⋅ Z ( λ v ) ∈ ( − 1 ) r R > 0 C(\lambda_{v}):=b_{2r,v}(0)\cdot L(\lambda_{v})^{-1}\cdot Z(\lambda_{v})\in(-1)^{r}\mathbb{R}_{>0} C ( λ v ) := b 2 r , v ( 0 ) ⋅ L ( λ v ) − 1 ⋅ Z ( λ v ) ∈ ( − 1 ) r R > 0 in which the three factors are from Notation 2.6(3), Definition 2.26(2), and Definition 2.26(3), respectively (in order).
Proof. This follows from the same argument for [20 Proposition 3.7] together with Remark 2.24. □
Motivic special cycles and motivic generating functions
In this chapter, we define several key objects in our theory, including standard (relative) motives, automorphic (relative) motives, motivic special classes, motivic generating functions, and motivic arithmetic theta lifting.
From now to the end of this article except Remark 6.7, we assume that the CM field E E E contains an imaginary quadratic field, for which we choose one and denote it by E 0 E_0 E 0 . Recall from Notation 2.6 that E ⊆ C \mathbb{E}\subseteq\mathbb{C} E ⊆ C denotes the normal closure of E E E in C \mathbb{C} C , so that Hom ( E , C ) = Hom ( E , E ) \operatorname{Hom}(E,\mathbb{C})=\operatorname{Hom}(E,\mathbb{E}) Hom ( E , C ) = Hom ( E , E ) .
An auxiliary moduli problem
Put T : = Res E 0 / Q G m T:=\operatorname{Res}_{E_0/\mathbb{Q}}\mathbb{G}_m T := Res E 0 / Q G m .
Definition 3.1. We define a functor
Y ∙ : Hk ( T ( A ∞ ) ) ⟶ PSch / E 0 ′ Y_\bullet:\operatorname{Hk}(T(\mathbb{A}^{\infty}))\longrightarrow\operatorname{PSch}'_{/E_0} Y ∙ : Hk ( T ( A ∞ )) ⟶ PSch / E 0 ′
(Definition 1.12 ) such that for every objects U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) and S ∈ Sch / E 0 ′ S\in\operatorname{Sch}'_{/E_0} S ∈ Sch / E 0 ′ , Y U ( S ) Y_U(S) Y U ( S ) is the set of equivalence classes of triples ( A 0 , λ 0 , η ˉ 0 ) (A_0,\lambda_0,\bar{\eta}_0) ( A 0 , λ 0 , η ˉ 0 ) , where
A 0 A_0 A 0 is an object of AV ( S , E 0 ) \operatorname{AV}(S,E_0) AV ( S , E 0 ) (Definition A.5) such that tr A 0 ( e ) = e \operatorname{tr}_{A_0}(e)=e tr A 0 ( e ) = e for e ∈ E 0 e\in E_0 e ∈ E 0 ;
λ 0 : A 0 → A 0 ∨ \lambda_0:A_0\to A_0^\vee λ 0 : A 0 → A 0 ∨ is a polarization that turns the complex conjugation of E 0 E_0 E 0 into the Rosati involution;
η ˉ 0 \bar{\eta}_0 η ˉ 0 is a U U U -level structure, that is, for a chosen geometric point s s s on every connected component of S S S , a π 1 ( S , s ) \pi_1(S,s) π 1 ( S , s ) -invariant U U U -orbit of isomorphisms
η 0 : A E 0 ∞ → ∼ H 1 e ˊ t ( A 0 s , A ∞ ) \eta_0:A_{E_0}^{\infty}\xrightarrow{\sim}H_1^{\mathrm{\acute{e}t}}(A_{0s},A^{\infty}) η 0 : A E 0 ∞ ∼ H 1 e ˊ t ( A 0 s , A ∞ )
of A E 0 ∞ A_{E_0}^{\infty} A E 0 ∞ -modules.
Two objects ( A 0 , λ 0 , η ˉ 0 ) (A_0,\lambda_0,\bar{\eta}_0) ( A 0 , λ 0 , η ˉ 0 ) and ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ) (A'_0,\lambda'_0,\bar{\eta}'_0) ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ) are equivalent15 if there exists an isomorphism φ 0 : A 0 → A 0 ′ \varphi_0:A_0\to A'_0 φ 0 : A 0 → A 0 ′ in the category AV ( S , E 0 ) \operatorname{AV}(S,E_0) AV ( S , E 0 ) satisfying that
A morphism t ∈ U \ T ( A ∞ ) / U ′ t\in U\backslash T(\mathbb{A}^{\infty})/U' t ∈ U \ T ( A ∞ ) / U ′ maps Y U ( S ) Y_U(S) Y U ( S ) to Y U ′ ( S ) Y_{U'}(S) Y U ′ ( S ) by changing η ˉ 0 \bar{\eta}_0 η ˉ 0 to η ˉ 0 ∘ t \bar{\eta}_0\circ t η ˉ 0 ∘ t .
It is well-known that Y ∙ Y_\bullet Y ∙ is representable and factors through the full subcategory S c h / E 0 s m , p r o j \mathrm{Sch}^{\mathrm{sm,proj}}_{/E_0} Sch / E 0 sm , proj of P S c h / E 0 ′ \mathrm{PSch}'_{/E_0} PSch / E 0 ′ . Moreover, if we denote by A 0 , U ∈ A V ( Y U , E 0 ) A_{0,U} \in\mathrm{AV}(Y_U,E_0) A 0 , U ∈ AV ( Y U , E 0 ) the universal object, then we obtain a functor
A 0 , ∙ / Y ∙ : H k ( T ( A ∞ ) ) ⟶ A V ( S c h / E 0 s m , p r o j , E 0 ) A_{0,\bullet}/Y_\bullet: \mathrm{Hk}(T(\mathbb{A}^{\infty})) \longrightarrow\mathrm{AV}(\mathrm{Sch}^{\mathrm{sm,proj}}_{/E_0},E_0) A 0 , ∙ / Y ∙ : Hk ( T ( A ∞ )) ⟶ AV ( Sch / E 0 sm , proj , E 0 )
over Y ∙ Y_\bullet Y ∙ that is Cartesian (Definition 1.14 ).
Remark 3.2. For every object U ∈ H k ( T ( A ∞ ) ) U \in\mathrm{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) , the group T ( A ∞ ) / U T(\mathbb{A}^{\infty})/U T ( A ∞ ) / U acts on Y U Y_U Y U via modifying η ˉ 0 \bar{\eta}_0 η ˉ 0 , which factors through the double quotient T ( Q ) \ T ( A ∞ ) / U T(\mathbb{Q})\backslash T(\mathbb{A}^{\infty})/U T ( Q ) \ T ( A ∞ ) / U . By the theory of complex multiplication, the induced action of T ( Q ) \ T ( A ∞ ) / U T(\mathbb{Q})\backslash T(\mathbb{A}^{\infty})/U T ( Q ) \ T ( A ∞ ) / U on Y U Y_U Y U is free, and the quotient scheme is Spec E 0 \operatorname{Spec} E_0 Spec E 0 .
Standard motives over unitary Shimura varieties
From now to the end of this chapter, we consider a totally positive definite incoherent hermitian space V V V over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F of rank m m m (Definition 2.15), and put H : = U ( V ) H := \mathrm{U}(V) H := U ( V ) .
Definition 3.3. We define a functor
X ∙ , ∙ : H k ( H ( A F ∞ ) ) × H k ( T ( A ∞ ) ) ⟶ P S c h / E ′ X_{\bullet,\bullet} : \mathrm{Hk}(H(\mathbb{A}^{\infty}_F)) \times\mathrm{Hk}(T(\mathbb{A}^{\infty})) \longrightarrow\mathrm{PSch}'_{/E} X ∙ , ∙ : Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) ⟶ PSch / E ′
(Definition 1.12) such that for every objects ( L , U ) ∈ H k ( H ( A F ∞ ) ) × H k ( T ( A ∞ ) ) (L,U) \in\mathrm{Hk}(H(\mathbb{A}^{\infty}_F)) \times\mathrm{Hk}(T(\mathbb{A}^{\infty})) ( L , U ) ∈ Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) and S ∈ S c h / E ′ S \in\mathrm{Sch}'_{/E} S ∈ Sch / E ′ , X L , U ( S ) X_{L,U}(S) X L , U ( S ) is the set of equivalence classes of sextuples ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) (A_0,\lambda_0,\bar{\eta}_0; A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) ,16 where
( A 0 , λ 0 , η ˉ 0 ) (A_0,\lambda_0,\bar{\eta}_0) ( A 0 , λ 0 , η ˉ 0 ) is an object of Y U ( S ) Y_U(S) Y U ( S ) ;
A A A is an object of A V ( S , E ) \mathrm{AV}(S,E) AV ( S , E ) (Definition A.5) such that tr A ( e ) = m Tr E / E 0 ( e ) − e + e ˉ \operatorname{tr}_A(e)=m\operatorname{Tr}_{E/E_0}(e)-e+\bar{e} tr A ( e ) = m Tr E / E 0 ( e ) − e + e ˉ for e ∈ E e \in E e ∈ E ;
λ : A → A ∨ \lambda:A\to A^\vee λ : A → A ∨ is a polarization that turns the complex conjugation of E E E into the Rosati involution;
η ˉ \bar{\eta} η ˉ is a hermitian L L L -level structure, that is, for a chosen geometric point s s s on every connected component of S S S , a π 1 ( S , s ) \pi_1(S,s) π 1 ( S , s ) -invariant L L L -orbit of isometries
η : V ∞ → ∼ Hom A E 0 ∞ ( H ^ 1 ( A 0 s , A ∞ ) ⊗ A F ∞ A ∞ , H ^ 1 ( A s , A ∞ ) ) \eta: V^{\infty} \xrightarrow{\sim} \operatorname{Hom}_{A^\infty_{E_0}}(\widehat{\mathrm{H}}_1(A_{0s},A^\infty)\otimes_{A^\infty_F} A^\infty,\widehat{\mathrm{H}}_1(A_s,A^\infty)) η : V ∞ ∼ Hom A E 0 ∞ ( H 1 ( A 0 s , A ∞ ) ⊗ A F ∞ A ∞ , H 1 ( A s , A ∞ ))
of hermitian spaces over A E ∞ / A F ∞ \mathbb{A}^{\infty}_E/\mathbb{A}^{\infty}_F A E ∞ / A F ∞ (see, for example, [29 Construction 3.4.4] for the hermitian structure on the right-hand side).
Two objects ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) (A_0,\lambda_0,\bar{\eta}_0;A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) and ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ; A ′ , λ ′ , η ˉ ′ ) (A'_0,\lambda'_0,\bar{\eta}'_0;A',\lambda',\bar{\eta}') ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ; A ′ , λ ′ , η ˉ ′ ) are equivalent if there exist isomorphisms φ 0 : A 0 → A 0 ′ \varphi_0:A_0\to A'_0 φ 0 : A 0 → A 0 ′ in the category A V ( S , E 0 ) \mathrm{AV}(S,E_0) AV ( S , E 0 ) and φ : A → A ′ \varphi:A\to A' φ : A → A ′ in the category A V ( S , E ) \mathrm{AV}(S,E) AV ( S , E ) satisfying that
φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 \varphi_0^\vee\circ\lambda'_0 \circ\varphi_0 φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 and φ ∨ ∘ λ ′ ∘ φ \varphi^\vee\circ\lambda' \circ\varphi φ ∨ ∘ λ ′ ∘ φ are the same multiple of λ 0 \lambda_0 λ 0 and λ \lambda λ , respectively;
the U U U -orbits φ 0 ∗ ∘ η ˉ 0 \varphi_{0*}\circ\bar{\eta}_0 φ 0 ∗ ∘ η ˉ 0 and η ˉ 0 ′ \bar{\eta}'_0 η ˉ 0 ′ coincide;
the L L L -orbit of maps v ↦ φ ∗ ∘ η ˉ ( v ) ∘ ( φ 0 ∗ ) − 1 v\mapsto\varphi_*\circ\bar{\eta}(v)\circ(\varphi_{0*})^{-1} v ↦ φ ∗ ∘ η ˉ ( v ) ∘ ( φ 0 ∗ ) − 1 for v ∈ V ∞ v\in V^{\infty} v ∈ V ∞ coincides with η ˉ ′ \bar{\eta}' η ˉ ′ .
A morphism ( h , t ) ∈ L \ H ( A F ∞ ) / L ′ × U \ T ( A ∞ ) / U ′ (h,t)\in L\backslash H(\mathbb{A}^{\infty}_F)/L'\times U\backslash T(\mathbb{A}^{\infty})/U' ( h , t ) ∈ L \ H ( A F ∞ ) / L ′ × U \ T ( A ∞ ) / U ′ maps X L , U ( S ) X_{L,U}(S) X L , U ( S ) to X L ′ , U ′ ( S ) X_{L',U'}(S) X L ′ , U ′ ( S ) by changing η ˉ 0 \bar{\eta}_0 η ˉ 0 to η ˉ 0 ∘ t \bar{\eta}_0\circ t η ˉ 0 ∘ t and η ˉ \bar{\eta} η ˉ to η ˉ ∘ h \bar{\eta}\circ h η ˉ ∘ h .
By [32 Theorem 4.4] , X ∙ , ∙ X_{\bullet,\bullet} X ∙ , ∙ factors through the full subcategory S c h / E s m , q p r o j \mathrm{Sch}^{\mathrm{sm,qproj}}_{/E} Sch / E sm , qproj of P S c h / E ′ \mathrm{PSch}'_{/E} PSch / E ′ . If we denote by
( A 0 , L , U , A L , U ) ∈ A V ( X L , U , E 0 ) × A V ( X L , U , E ) (A_{0,L,U},A_{L,U})\in\mathrm{AV}(X_{L,U},E_0)\times\mathrm{AV}(X_{L,U},E) ( A 0 , L , U , A L , U ) ∈ AV ( X L , U , E 0 ) × AV ( X L , U , E )
the universal object, then we obtain functors
A 0 , ∙ , ∙ / X ∙ , ∙ : H k ( H ( A F ∞ ) ) × H k ( T ( A ∞ ) ) ⟶ A V ( S c h / E s m , q p r o j , E 0 ) , A_{0,\bullet,\bullet}/X_{\bullet,\bullet}:\mathrm{Hk}(H(\mathbb{A}^{\infty}_F))\times\mathrm{Hk}(T(\mathbb{A}^{\infty}))\longrightarrow\mathrm{AV}(\mathrm{Sch}^{\mathrm{sm,qproj}}_{/E},E_0), A 0 , ∙ , ∙ / X ∙ , ∙ : Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) ⟶ AV ( Sch / E sm , qproj , E 0 ) ,
A ∙ , ∙ / X ∙ , ∙ : H k ( H ( A F ∞ ) ) × H k ( T ( A ∞ ) ) ⟶ A V ( S c h / E s m , q p r o j , E ) A_{\bullet,\bullet}/X_{\bullet,\bullet}:\mathrm{Hk}(H(\mathbb{A}^{\infty}_F))\times\mathrm{Hk}(T(\mathbb{A}^{\infty}))\longrightarrow\mathrm{AV}(\mathrm{Sch}^{\mathrm{sm,qproj}}_{/E},E) A ∙ , ∙ / X ∙ , ∙ : Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) ⟶ AV ( Sch / E sm , qproj , E )
over X ∙ , ∙ X_{\bullet,\bullet} X ∙ , ∙ that are Cartesian (Definition 1.14). By definition, we have a natural morphism
X L , U ⟶ Y U ⊗ E 0 E (5) X_{L,U}\longrightarrow Y_U\otimes_{E_0}E \tag*{(5)} X L , U ⟶ Y U ⊗ E 0 E (5)
for every object ( L , U ) ∈ H k ( H ( A F ∞ ) ) × H k ( T ( A ∞ ) ) (L,U)\in\mathrm{Hk}(H(\mathbb{A}^{\infty}_F))\times\mathrm{Hk}(T(\mathbb{A}^{\infty})) ( L , U ) ∈ Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) , along which A 0 , L , U A_{0,L,U} A 0 , L , U is the pullback of A 0 , U ⊗ E 0 E A_{0,U}\otimes_{E_0}E A 0 , U ⊗ E 0 E to X L , U X_{L,U} X L , U .
Take an object L ∈ Hk ( H ( A F ∞ ) ) L \in\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) L ∈ Hk ( H ( A F ∞ )) . For every U ∈ Hk ( T ( A ∞ ) ) U \in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) , the group T ( A ∞ ) / U T(\mathbb{A}^{\infty})/U T ( A ∞ ) / U acts on X L , U X_{L,U} X L , U via modifying η ˉ 0 \bar{\eta}_{0} η ˉ 0 , which also factors through the quotient
Ξ U : = T ( Q ) \ T ( A ∞ ) / U . \Xi_{U} := T(\mathbb{Q})\backslash T(\mathbb{A}^{\infty})/U. Ξ U := T ( Q ) \ T ( A ∞ ) / U .
Indeed, for every element t ∈ E × t \in E^{\times} t ∈ E × , the isomorphisms in AV ( S , E 0 ) \operatorname{AV}(S,E_{0}) AV ( S , E 0 ) and AV ( S , E ) \operatorname{AV}(S,E) AV ( S , E ) given by the action of t t t give an equivalence between ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) (A_{0},\lambda_{0},\bar{\eta}_{0}; A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) and ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) (A_{0},\lambda_{0},\bar{\eta}_{0}\circ t; A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) . It is clear that the morphism (5) is equivariant with respect to the action of Ξ U \Xi_{U} Ξ U . Thus, by Remark 3.2, the action of Ξ U \Xi_{U} Ξ U on X L , U X_{L,U} X L , U is free. Since X L , U X_{L,U} X L , U is smooth and quasi-projective over Spec E \operatorname{Spec} E Spec E , the quotient X L , U / Ξ U X_{L,U}/\Xi_{U} X L , U / Ξ U is a scheme. It is clear that for U ′ ⊆ U U' \subseteq U U ′ ⊆ U in Hk ( T ( A ∞ ) ) \operatorname{Hk}(T(\mathbb{A}^{\infty})) Hk ( T ( A ∞ )) , the natural morphism X L , U ′ / Ξ U ′ → X L , U / Ξ U X_{L,U'}/\Xi_{U'} \to X_{L,U}/\Xi_{U} X L , U ′ / Ξ U ′ → X L , U / Ξ U is an isomorphism.
To summarize, we obtain a functor
X ∙ : Hk ( H ( A F ∞ ) ) ⟶ Sch / E s m . q p r o j . X_{\bullet}\colon\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) \longrightarrow\operatorname{Sch}^{\mathrm{sm.qproj}}_{/E}. X ∙ : Hk ( H ( A F ∞ )) ⟶ Sch / E sm.qproj .
such that for every L L L , X L = X L , U / Ξ U X_{L}=X_{L,U}/\Xi_{U} X L = X L , U / Ξ U – the quotient scheme obtained from the above process, which is canonically independent of the choice of U ∈ Hk ( T ( A ∞ ) ) U \in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) . In what follows, we denote by
π L , U : X L , U ⟶ X L (6) \pi_{L,U}\colon X_{L,U}\longrightarrow X_{L} \tag*{(6)} π L , U : X L , U ⟶ X L (6)
the natural quotient morphism.
Remark 3.4. We discuss the relation between X ∙ X_{\bullet} X ∙ and the usual Shimura variety. Take an element v ∈ V F ( ∞ ) v \in\mathcal{V}^{(\infty)}_{F} v ∈ V F ( ∞ ) . Denote by V ( v ) V(v) V ( v ) the unique hermitian space over E / F E/F E / F that has signature ( m − 1 , 1 ) (m-1,1) ( m − 1 , 1 ) at v v v and such that V ( v ) ⊗ F A F ∞ V(v)\otimes_{F}\mathbb{A}^{\infty}_{F} V ( v ) ⊗ F A F ∞ is isometric to V ⊗ F A F ∞ V\otimes_{F}\mathbb{A}^{\infty}_{F} V ⊗ F A F ∞ . Put H ( v ) : = U ( V ( v ) ) H(v):=U(V(v)) H ( v ) := U ( V ( v )) , and fix an isometry V ( v ) ⊗ F A F ∞ ≃ V ∞ V(v)\otimes_{F}\mathbb{A}^{\infty}_{F}\simeq V^{\infty} V ( v ) ⊗ F A F ∞ ≃ V ∞ under which we identify H ( A F ∞ ) H(\mathbb{A}^{\infty}_{F}) H ( A F ∞ ) with H ( v ) ( A F ∞ ) H(v)(\mathbb{A}^{\infty}_{F}) H ( v ) ( A F ∞ ) . For every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) above v v v , we have a Shimura variety functor17
Sh ( H ( v ) ) u : Hk ( H ( A F ∞ ) ) ⟶ Sch / u ( E ) s m . q p r o j . \operatorname{Sh}(H(v))^{u}\colon\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F}))\longrightarrow\operatorname{Sch}^{\mathrm{sm.qproj}}_{/u(E)}. Sh ( H ( v ) ) u : Hk ( H ( A F ∞ )) ⟶ Sch / u ( E ) sm.qproj .
after we identify E ⊗ F , v R E\otimes_{F,v}\mathbb{R} E ⊗ F , v R with C \mathbb{C} C using u u u (see, for example, [29 Section 3.2] ). Then we have a canonical isomorphism
X ∙ ⊗ E , u u ( E ) → ∼ Sh ( H ( v ) ) u (7) X_{\bullet}\otimes_{E,u}u(E)\xrightarrow{\sim}\operatorname{Sh}(H(v))^{u} \tag*{(7)} X ∙ ⊗ E , u u ( E ) ∼ Sh ( H ( v ) ) u (7)
in Fun ( Hk ( H ( A F ∞ ) ) , Sch / u ( E ) s m . q p r o j ) \operatorname{Fun}(\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})),\operatorname{Sch}^{\mathrm{sm.qproj}}_{/u(E)}) Fun ( Hk ( H ( A F ∞ )) , Sch / u ( E ) sm.qproj ) . In particular, for every L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) L ∈ Hk ( H ( A F ∞ )) ,
X L u = H ( v ) ( F ) \ ( D u × H ( A F ∞ ) / L ) X^{u}_{L}=H(v)(F)\backslash\left(\mathcal{D}^{u}\times H(\mathbb{A}^{\infty}_{F})/L\right) X L u = H ( v ) ( F ) \ ( D u × H ( A F ∞ ) / L )
(as complex manifolds) in which D u \mathcal{D}^{u} D u denotes the open subset of V ( v ) ⊗ E , u C V(v)\otimes_{E,u}\mathbb{C} V ( v ) ⊗ E , u C of negative definite lines.
Take an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) L ∈ Hk ( H ( A F ∞ )) . For every U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) , we have the relative motives (see §A.2)18
h 1 ( A 0 , L , U / X L , U , Q ) ∈ Mot ( X L , U , E 0 ) , h_{1}(A_{0,L,U}/X_{L,U},\mathbb{Q})\in\operatorname{Mot}(X_{L,U},E_{0}), h 1 ( A 0 , L , U / X L , U , Q ) ∈ Mot ( X L , U , E 0 ) ,
h 1 ( A L , U / X L , U , Q ) ∈ Mot ( X L , U , E ) , h_{1}(A_{L,U}/X_{L,U},\mathbb{Q})\in\operatorname{Mot}(X_{L,U},E), h 1 ( A L , U / X L , U , Q ) ∈ Mot ( X L , U , E ) ,
hence
M L , U : = Hom E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) ∈ Mot ( X L , U , E ) . M_{L,U}:=\operatorname{Hom}_{E_{0}}\left(h_{1}(A_{0,L,U}/X_{L,U},\mathbb{Q}),h_{1}(A_{L,U}/X_{L,U},\mathbb{Q})\right)\in\operatorname{Mot}(X_{L,U},E). M L , U := Hom E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) ∈ Mot ( X L , U , E ) .
We would like to descend the above motive to X L X_{L} X L . For every t ∈ T ( A ∞ ) / U t\in T(\mathbb{A}^{\infty})/U t ∈ T ( A ∞ ) / U , denote by
⟨ t ⟩ : X L , U ⟶ X L , U \langle t\rangle\colon X_{L,U}\longrightarrow X_{L,U} ⟨ t ⟩ : X L , U ⟶ X L , U
the induced (iso)morphism. By definition, the pullback of the universal data ⟨ t ⟩ ∗ ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) \langle t\rangle^{*}(A_{0},\lambda_{0},\bar{\eta}_{0};A,\lambda,\bar{\eta}) ⟨ t ⟩ ∗ ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) is equivalent to ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) (A_{0},\lambda_{0},\bar{\eta}_{0}\circ t;A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) . Denote by φ t , 0 : A 0 , L , U → ⟨ t ⟩ ∗ A 0 , L , U \varphi_{t,0}\colon A_{0,L,U}\to\langle t\rangle^{*}A_{0,L,U} φ t , 0 : A 0 , L , U → ⟨ t ⟩ ∗ A 0 , L , U and φ t : A L , U → ⟨ t ⟩ ∗ A L , U \varphi_{t}\colon A_{L,U}\to\langle t\rangle^{*}A_{L,U} φ t : A L , U → ⟨ t ⟩ ∗ A L , U the unique isomorphisms in AV ( X L , U , E 0 ) \operatorname{AV}(X_{L,U},E_{0}) AV ( X L , U , E 0 ) and AV ( X L , U , E ) \operatorname{AV}(X_{L,U},E) AV ( X L , U , E ) , respectively, that give the unique equivalence from ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) (A_{0},\lambda_{0},\bar{\eta}_{0}\circ t;A,\lambda,\bar{\eta}) ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ) to ⟨ t ⟩ ∗ ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) \langle t\rangle^{*}(A_{0},\lambda_{0},\bar{\eta}_{0};A,\lambda,\bar{\eta}) ⟨ t ⟩ ∗ ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ) . Then
⟨ t ⟩ ∗ ( φ t ′ ) ∘ φ t = φ t t ′ . \langle t\rangle^{*}(\varphi_{t'})\circ\varphi_{t}=\varphi_{tt'}. ⟨ t ⟩ ∗ ( φ t ′ ) ∘ φ t = φ t t ′ .
for every t ′ ∈ T ( A ∞ ) / U t' \in T(\mathbb{A}^{\infty})/U t ′ ∈ T ( A ∞ ) / U . Denote by Ξ U \Xi_U Ξ U the groupoid given by Ξ U \Xi_U Ξ U . We have a functor
X L , U / Ξ U : Ξ U ⟶ S c h / E s m . q p r o j X_{L,U}/\Xi_U : \Xi_U \longrightarrow\mathrm{Sch}^{\mathrm{sm.qproj}}_{/E} X L , U / Ξ U : Ξ U ⟶ Sch / E sm.qproj
sending the unique object to X L , U X_{L,U} X L , U and t ∈ Ξ U t \in\Xi_U t ∈ Ξ U to ⟨ t ⟩ \langle t\rangle ⟨ t ⟩ .
For t ∈ T ( A ∞ ) / U t \in T(\mathbb{A}^{\infty})/U t ∈ T ( A ∞ ) / U , we define a map
α t : ⟨ t ⟩ ∗ M L , U ⟶ M L , U \alpha_t:\langle t\rangle^*M_{L,U} \longrightarrow M_{L,U} α t : ⟨ t ⟩ ∗ M L , U ⟶ M L , U
as the composition of isomorphisms
⟨ t ⟩ ∗ M L , U → ∼ Hom ~ E 0 ( ⟨ t ⟩ ∗ h 1 ( A 0 , L , U / X L , U , Q ) , ⟨ t ⟩ ∗ h 1 ( A L , U / X L , U , Q ) ) → ∼ ⟨ t ⟩ ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ⟨ t ⟩ ∗ h 1 ( A L , U / X L , U , Q ) ⟶ h 1 ( ⟨ t ⟩ ∗ A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( ⟨ t ⟩ ∗ A L , U / X L , U , Q ) → φ t , 0 ∗ ( φ t ) ∗ − 1 h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) → ∼ Hom ~ E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) = M L , U , \begin{aligned}
\langle t\rangle^*M_{L,U}
&\xrightarrow{\sim} \widetilde{\operatorname{Hom}}_{E_0}\left(\langle t\rangle^*h_1(A_{0,L,U}/X_{L,U},\mathbb{Q}),\langle t\rangle^*h_1(A_{L,U}/X_{L,U},\mathbb{Q})\right)\\
&\xrightarrow{\sim} \langle t\rangle^*h_1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}\langle t\rangle^*h_1(A_{L,U}/X_{L,U},\mathbb{Q})\\
&\longrightarrow h^1(\langle t\rangle^*A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h_1(\langle t\rangle^*A_{L,U}/X_{L,U},\mathbb{Q})\\
&\xrightarrow{\varphi_{t,0}^*(\varphi_t)_*^{-1}} h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h_1(A_{L,U}/X_{L,U},\mathbb{Q})\\
&\xrightarrow{\sim}\widetilde{\operatorname{Hom}}_{E_0}\left(h_1(A_{0,L,U}/X_{L,U},\mathbb{Q}),h_1(A_{L,U}/X_{L,U},\mathbb{Q})\right)=M_{L,U},
\end{aligned} ⟨ t ⟩ ∗ M L , U ∼ Hom E 0 ( ⟨ t ⟩ ∗ h 1 ( A 0 , L , U / X L , U , Q ) , ⟨ t ⟩ ∗ h 1 ( A L , U / X L , U , Q ) ) ∼ ⟨ t ⟩ ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ⟨ t ⟩ ∗ h 1 ( A L , U / X L , U , Q ) ⟶ h 1 (⟨ t ⟩ ∗ A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 (⟨ t ⟩ ∗ A L , U / X L , U , Q ) φ t , 0 ∗ ( φ t ) ∗ − 1 h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ∼ Hom E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) = M L , U ,
in which we have used the fact that the canonical maps
φ t , 0 ∗ : h 1 ( ⟨ t ⟩ ∗ A 0 , L , U / X L , U , Q ) ⟶ h 1 ( A 0 , L , U / X L , U , Q ) , \varphi_{t,0}^*:h^1(\langle t\rangle^*A_{0,L,U}/X_{L,U},\mathbb{Q})\longrightarrow h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}), φ t , 0 ∗ : h 1 (⟨ t ⟩ ∗ A 0 , L , U / X L , U , Q ) ⟶ h 1 ( A 0 , L , U / X L , U , Q ) ,
( φ t ) ∗ : h 1 ( A L , U / X L , U , Q ) ⟶ h 1 ( ⟨ t ⟩ ∗ A L , U / X L , U , Q ) (\varphi_t)_*:h_1(A_{L,U}/X_{L,U},\mathbb{Q})\longrightarrow h_1(\langle t\rangle^*A_{L,U}/X_{L,U},\mathbb{Q}) ( φ t ) ∗ : h 1 ( A L , U / X L , U , Q ) ⟶ h 1 (⟨ t ⟩ ∗ A L , U / X L , U , Q )
are isomorphisms. Then it is easy to verify that
α t ∘ ⟨ t ⟩ ∗ ( α t ′ ) = α t t ′ \alpha_t\circ\langle t\rangle^*(\alpha_{t'})=\alpha_{tt'} α t ∘ ⟨ t ⟩ ∗ ( α t ′ ) = α t t ′
for every t ′ ∈ T ( A ∞ ) / U t'\in T(\mathbb{A}^{\infty})/U t ′ ∈ T ( A ∞ ) / U . Moreover, for t ∈ E 0 × t\in E_0^\times t ∈ E 0 × , α t \alpha_t α t is the identity isomorphism. Thus, if we denote by Mot ( X L , U / Ξ U , E ) \operatorname{Mot}(X_{L,U}/\Xi_U,E) Mot ( X L , U / Ξ U , E ) the category of Cartesian functors Ξ U → Mot ( S c h / E s m . q p r o j , E ) \Xi_U\to\operatorname{Mot}(\mathrm{Sch}^{\mathrm{sm.qproj}}_{/E},E) Ξ U → Mot ( Sch / E sm.qproj , E ) above X L , U / Ξ U X_{L,U}/\Xi_U X L , U / Ξ U , then M L , U M_{L,U} M L , U together with the isomorphisms { α t } \{\alpha_t\} { α t } define an object
M L , U / Ξ U ∈ Mot ( X L , U / Ξ U , E ) . M_{L,U}/\Xi_U\in\operatorname{Mot}(X_{L,U}/\Xi_U,E). M L , U / Ξ U ∈ Mot ( X L , U / Ξ U , E ) .
By Lemma 3.7 below, M L , U / Ξ U M_{L,U}/\Xi_U M L , U / Ξ U is an object of Mot ( X L , E ) \operatorname{Mot}(X_L,E) Mot ( X L , E ) . By the same lemma, for U ′ ⊆ U U'\subseteq U U ′ ⊆ U in Hk ( T ( A ∞ ) ) \operatorname{Hk}(T(\mathbb{A}^{\infty})) Hk ( T ( A ∞ )) , the natural map M L , U / Ξ U → M L , U ′ / Ξ U ′ M_{L,U}/\Xi_U\to M_{L,U'}/\Xi_{U'} M L , U / Ξ U → M L , U ′ / Ξ U ′ is an isomorphism in Mot ( X L , E ) \operatorname{Mot}(X_L,E) Mot ( X L , E ) .
Definition 3.5. We define the standard motive over X L X_L X L to be the object M L , U / Ξ U M_{L,U}/\Xi_U M L , U / Ξ U of Mot ( X L , E ) \operatorname{Mot}(X_L,E) Mot ( X L , E ) defined above, which is independent of U U U , and we denote it by M L M_L M L .
Remark 3.6. In general, neither h 1 ( A 0 , L , U / X L , U , Q ) h_1(A_{0,L,U}/X_{L,U},\mathbb{Q}) h 1 ( A 0 , L , U / X L , U , Q ) nor h 1 ( A L , U / X L , U , Q ) h_1(A_{L,U}/X_{L,U},\mathbb{Q}) h 1 ( A L , U / X L , U , Q ) descends to a relative motive over X L X_L X L .
It is straightforward to verify that
M ∙ : Hk ( H ( A F ∞ ) ) ⟶ Mot ( S c h / E s m . q p r o j , E ) M_{\bullet}:\operatorname{Hk}(H(\mathbb{A}^{\infty}_F))\longrightarrow\operatorname{Mot}(\mathrm{Sch}^{\mathrm{sm.qproj}}_{/E},E) M ∙ : Hk ( H ( A F ∞ )) ⟶ Mot ( Sch / E sm.qproj , E )
is a Cartesian functor above X ∙ X_{\bullet} X ∙ .
Lemma 3.7. Let the notation be as above. The natural functor Mot ( X L , E ) → Mot ( X L , U / Ξ U , E ) \operatorname{Mot}(X_L,E)\to\operatorname{Mot}(X_{L,U}/\Xi_U,E) Mot ( X L , E ) → Mot ( X L , U / Ξ U , E ) induced by pullback is an equivalence of categories.
Proof. By construction, the natural morphism X L , U → X L × E 0 Y U X_{L,U}\to X_L\times_{E_0}Y_U X L , U → X L × E 0 Y U is an isomorphism. We would like to apply [16 Theorem 2.17] with Π = T ( Q ) \ T ( A ∞ ) \Pi=T(\mathbb{Q})\backslash T(\mathbb{A}^{\infty}) Π = T ( Q ) \ T ( A ∞ ) . Let B Π B\Pi B Π be the classifying stack of the profinite group Π \Pi Π , that is, the category of finite continuous Π \Pi Π -sets. We have a functor B Π → S c h / E 0 B\Pi\to\mathrm{Sch}/E_0 B Π → Sch / E 0 that sends finite coproducts to finite products and such that for every U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) , Y ( Π / U ) = Y U Y(\Pi/U)=Y_U Y ( Π/ U ) = Y U . Now let A \mathcal{A} A be the contravariant pseudo-functor from B Π B\Pi B Π to the 2-category of pseudo-abelian categories that sends a finite continuous Π \Pi Π -set S S S to Mot ( X L × E 0 Y ( S ) , Q ) \operatorname{Mot}(X_L\times_{E_0}Y(S),\mathbb{Q}) Mot ( X L × E 0 Y ( S ) , Q ) . By [16 Theorem 2.17(b)] , we know that A \mathcal{A} A is a stack over B Π B\Pi B Π , which implies the lemma. □
Remark 3.8. It follows from Remark 3.4 that X ∙ X_{\bullet} X ∙ does not depend on the choice of E 0 E_0 E 0 (an imaginary quadratic field contained in E E E ). However, we expect but do not know how to prove that M ∙ M_{\bullet} M ∙ also does not depend on such a choice. This is why our theory relies on the choice of E 0 E_0 E 0 .
In the rest of this section, we construct a map
M L ⊗ E M ‾ L ⟶ 1 X L , E , (8) M_L \otimes_{E} \overline{M}_L \longrightarrow\mathbb{1}_{X_L,E}, \tag*{(8)} M L ⊗ E M L ⟶ 1 X L , E , (8)
where 1 X L , E \mathbb{1}_{X_L,E} 1 X L , E denotes the unit object of Mot ( X L , E ) \operatorname{Mot}(X_L,E) Mot ( X L , E ) , which turns out to be an evaluation map (Definition 1.13 ).
Take an arbitrary object U ∈ Hk ( T ( A ∞ ) ) U \in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) . Using the isomorphism in Lemma A.7, we have successive isomorphisms in the category Mot ( X L , U , E ) \operatorname{Mot}(X_{L,U},E) Mot ( X L , U , E ) :
M L , U → ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) → ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) → ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ . (9) \begin{aligned}
M_{L,U} &\xrightarrow{\sim} h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0} h^1(A_{L,U}/X_{L,U},\mathbb{Q}) \\
&\xrightarrow{\sim} h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0} h^1(A_{L,U}/X_{L,U},\mathbb{Q}) \\
&\xrightarrow{\sim} h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0} h^1(A^\vee_{L,U}/X_{L,U},\mathbb{Q})\langle1\rangle. \tag*{(9)}
\end{aligned} M L , U ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ . (9)
Using Lemma A.7 again, we have
M L , U ∨ ≃ h 1 ( A 0 , L , U / X L , U , Q ) ∨ ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ∨ ⟨ − 1 ⟩ → ∼ h 1 ( A 0 , L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) → ∼ h 1 ( A 0 , L , U ∨ / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ⟨ 1 ⟩ . \begin{aligned}
M^\vee_{L,U} &\simeq h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})^\vee\otimes_{E_0} h^1(A^\vee_{L,U}/X_{L,U},\mathbb{Q})^\vee\langle-1\rangle\\
&\xrightarrow{\sim} h^1(A^\vee_{0,L,U}/X_{L,U},\mathbb{Q})\langle1\rangle\otimes_{E_0} h^1(A^\vee_{L,U}/X_{L,U},\mathbb{Q}) \\
&\xrightarrow{\sim} h^1(A^\vee_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0} h^1(A_{L,U}/X_{L,U},\mathbb{Q})\langle1\rangle.
\end{aligned} M L , U ∨ ≃ h 1 ( A 0 , L , U / X L , U , Q ) ∨ ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ∨ ⟨ − 1 ⟩ ∼ h 1 ( A 0 , L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ∼ h 1 ( A 0 , L , U ∨ / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ⟨ 1 ⟩ .
Now the polarizations λ 0 \lambda_0 λ 0 and λ \lambda λ induce isomorphisms
λ 0 ∗ : h 1 ( A 0 , L , U ∨ / X L , U , Q ) → ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 , − E 0 , \lambda_0^*: h^1(A^\vee_{0,L,U}/X_{L,U},\mathbb{Q}) \xrightarrow{\sim} h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0,-} E_0, λ 0 ∗ : h 1 ( A 0 , L , U ∨ / X L , U , Q ) ∼ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 , − E 0 ,
( λ − 1 ) ∗ : h 1 ( A L , U / X L , U , Q ) → ∼ h 1 ( A L , U ∨ / X L , U , Q ) ⊗ E , − E (\lambda^{-1})^*: h^1(A_{L,U}/X_{L,U},\mathbb{Q}) \xrightarrow{\sim} h^1(A^\vee_{L,U}/X_{L,U},\mathbb{Q}) \otimes_{E,-} E ( λ − 1 ) ∗ : h 1 ( A L , U / X L , U , Q ) ∼ h 1 ( A L , U ∨ / X L , U , Q ) ⊗ E , − E
in the categories Mot ( X L , U , E 0 ) \operatorname{Mot}(X_{L,U},E_0) Mot ( X L , U , E 0 ) and Mot ( X L , U , E ) \operatorname{Mot}(X_{L,U},E) Mot ( X L , U , E ) , respectively. Thus, we obtain an isomorphism
M L , U ∨ → ∼ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ) ⊗ E , − E → ∼ M L , U ⊗ E , − E (10) M^\vee_{L,U} \xrightarrow{\sim} \left(h^1(A_{0,L,U}/X_{L,U},\mathbb{Q}) \otimes_{E_0} h^1(A^\vee_{L,U}/X_{L,U},\mathbb{Q})\langle1\rangle\right) \otimes_{E,-} E \xrightarrow{\sim} M_{L,U} \otimes_{E,-} E \tag*{(10)} M L , U ∨ ∼ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ) ⊗ E , − E ∼ M L , U ⊗ E , − E (10)
in the category Mot ( X L , U , E ) \operatorname{Mot}(X_{L,U},E) Mot ( X L , U , E ) . It is straightforward to check that the isomorphism (10) is indeed an isomorphism in Mot ( X L , U / Ξ ‾ U , E ) \operatorname{Mot}(X_{L,U}/\overline{\Xi}_U,E) Mot ( X L , U / Ξ U , E ) , hence descends to an isomorphism M L ∨ → ∼ M ‾ L M^\vee_L \xrightarrow{\sim} \overline{M}_L M L ∨ ∼ M L , which is independent of the choice of U U U . In particular, we obtain the evaluation map (8) .
Special morphisms of unitary Shimura varieties
Consider a tuple x = ( x 1 , … , x r ) ∈ ( V ∞ ) r x=(x_1,\ldots,x_r)\in(V^\infty)^r x = ( x 1 , … , x r ) ∈ ( V ∞ ) r .
Definition 3.9. For every object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}^{\infty}_F)) L ∈ Hk ( H ( A F ∞ )) , we define a functor
Z ( x ) L , ∙ : Hk ( T ( A ∞ ) ) ⟶ PSch E ′ Z(x)_{L,\bullet}:\operatorname{Hk}(T(\mathbb{A}^{\infty}))\longrightarrow\operatorname{PSch}'_E Z ( x ) L , ∙ : Hk ( T ( A ∞ )) ⟶ PSch E ′
such that for every objects U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) and S ∈ Sch E ′ S\in\operatorname{Sch}'_E S ∈ Sch E ′ , Z ( x ) L , U ( S ) Z(x)_{L,U}(S) Z ( x ) L , U ( S ) is the set of equivalence classes of ( 6 + r ) (6+r) ( 6 + r ) -tuples ( A 0 , λ 0 , η ‾ 0 ; A , λ , η ‾ ; ν 1 , … , ν r ) (A_0,\lambda_0,\overline{\eta}_0;A,\lambda,\overline{\eta};\nu_1,\ldots,\nu_r) ( A 0 , λ 0 , η 0 ; A , λ , η ; ν 1 , … , ν r ) , where
( A 0 , λ 0 , η ‾ 0 ; A , λ , η ‾ ) (A_0,\lambda_0,\overline{\eta}_0;A,\lambda,\overline{\eta}) ( A 0 , λ 0 , η 0 ; A , λ , η ) is an object of X L , U ( S ) X_{L,U}(S) X L , U ( S ) ;
ν 1 , … , ν r \nu_1,\ldots,\nu_r ν 1 , … , ν r are morphisms from A 0 A_0 A 0 to A A A in the category AV ( S , E 0 ) \operatorname{AV}(S,E_0) AV ( S , E 0 ) satisfying that (on every connected component of S S S ) there exists an element η ∈ η ‾ \eta\in\overline{\eta} η ∈ η such that η ( x i ) \eta(x_i) η ( x i ) coincides with the induced map ν i ∗ : H e ˊ t 1 ( A 0 s , A ∞ ) ⊗ A F ∞ A F ∞ ⟶ H e ˊ t 1 ( A s , A ∞ ) \nu_{i*}:H^1_{\mathrm{\acute et}}(A_{0s},\mathbb{A}^{\infty})\otimes_{\mathbb{A}^{\infty}_F}\mathbb{A}^{\infty}_F\longrightarrow H^1_{\mathrm{\acute et}}(A_s,\mathbb{A}^{\infty}) ν i ∗ : H e ˊ t 1 ( A 0 s , A ∞ ) ⊗ A F ∞ A F ∞ ⟶ H e ˊ t 1 ( A s , A ∞ ) for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r .
Two objects ( A 0 , λ 0 , η ‾ 0 ; A , λ , η ‾ ; ν 1 , … , ν r ) (A_0,\lambda_0,\overline{\eta}_0;A,\lambda,\overline{\eta};\nu_1,\ldots,\nu_r) ( A 0 , λ 0 , η 0 ; A , λ , η ; ν 1 , … , ν r ) and ( A 0 ′ , λ 0 ′ , η ‾ 0 ′ ; A ′ , λ ′ , η ‾ ′ ; ν 1 ′ , … , ν r ′ ) (A'_0,\lambda'_0,\overline{\eta}'_0;A',\lambda',\overline{\eta}';\nu'_1,\ldots,\nu'_r) ( A 0 ′ , λ 0 ′ , η 0 ′ ; A ′ , λ ′ , η ′ ; ν 1 ′ , … , ν r ′ ) are equivalent if there exist isomorphisms φ 0 : A 0 ⟶ A 0 ′ \varphi_0:A_0\longrightarrow A'_0 φ 0 : A 0 ⟶ A 0 ′ in the category AV ( S , E 0 ) \operatorname{AV}(S,E_0) AV ( S , E 0 ) and φ : A ⟶ A ′ \varphi:A\longrightarrow A' φ : A ⟶ A ′ in the category AV ( S , E ) \operatorname{AV}(S,E) AV ( S , E ) that induce equivalence between ( A 0 , λ 0 , η ‾ 0 ; A , λ , η ‾ ) (A_0,\lambda_0,\overline{\eta}_0;A,\lambda,\overline{\eta}) ( A 0 , λ 0 , η 0 ; A , λ , η ) and ( A 0 ′ , λ 0 ′ , η ‾ 0 ′ ; A ′ , λ ′ , η ‾ ′ ) (A'_0,\lambda'_0,\overline{\eta}'_0;A',\lambda',\overline{\eta}') ( A 0 ′ , λ 0 ′ , η 0 ′ ; A ′ , λ ′ , η ′ ) and such that φ ∘ ν i = ν i ′ ∘ φ 0 \varphi\circ\nu_i=\nu'_i\circ\varphi_0 φ ∘ ν i = ν i ′ ∘ φ 0 for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r .
A morphism t ∈ U \ T ( A ∞ ) / U ′ t\in U\backslash T(\mathbb{A}^{\infty})/U' t ∈ U \ T ( A ∞ ) / U ′ maps Z ( x ) L , U ( S ) Z(x)_{L,U}(S) Z ( x ) L , U ( S ) to Z ( x ) L , U ′ ( S ) Z(x)_{L,U'}(S) Z ( x ) L , U ′ ( S ) by changing η ‾ 0 \overline{\eta}_0 η 0 to η ‾ 0 ∘ t \overline{\eta}_0\circ t η 0 ∘ t .
It is clear that Z ( x ) L , ∙ Z(x)_{L,\bullet} Z ( x ) L , ∙ is representable and factors through the full subcategory Sch / E s m , q p r o j \operatorname{Sch}^{\mathrm{sm,qproj}}_{/E} Sch / E sm , qproj of PSch / E ′ \operatorname{PSch}'_{/E} PSch / E ′ . For every U ∈ H k ( T ( A ∞ ) ) U \in Hk(T(\mathbb{A}^{\infty})) U ∈ H k ( T ( A ∞ )) , we have the natural morphism
ς ( x ) L , U : Z ( x ) L , U ⟶ X L , U \varsigma(x)_{L,U}: Z(x)_{L,U} \longrightarrow X_{L,U} ς ( x ) L , U : Z ( x ) L , U ⟶ X L , U
by forgetting ( ν 1 , … , ν r ) (\nu_1,\ldots,\nu_r) ( ν 1 , … , ν r ) , which is finite and unramified. For a morphism ( h , t ) ∈ L \ H ( A F ∞ ) / L ′ × U \ T ( A ∞ ) / U ′ (h,t) \in L\backslash H(\mathbb{A}^{\infty}_F)/L' \times U\backslash T(\mathbb{A}^{\infty})/U' ( h , t ) ∈ L \ H ( A F ∞ ) / L ′ × U \ T ( A ∞ ) / U ′ , we have the induced morphism
⟨ h , t ⟩ : Z ( x ) L , U ⟶ Z ( h − 1 x ) L ′ , U ′ \langle h,t\rangle: Z(x)_{L,U} \longrightarrow Z(h^{-1}x)_{L',U'} ⟨ h , t ⟩ : Z ( x ) L , U ⟶ Z ( h − 1 x ) L ′ , U ′
sending ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ; ν 1 , … , ν r ) (A_0,\lambda_0,\bar{\eta}_0;A,\lambda,\bar{\eta};\nu_1,\ldots,\nu_r) ( A 0 , λ 0 , η ˉ 0 ; A , λ , η ˉ ; ν 1 , … , ν r ) to ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ∘ h ; ν 1 , … , ν r ) (A_0,\lambda_0,\bar{\eta}_0\circ t;A,\lambda,\bar{\eta}\circ h;\nu_1,\ldots,\nu_r) ( A 0 , λ 0 , η ˉ 0 ∘ t ; A , λ , η ˉ ∘ h ; ν 1 , … , ν r ) , which fits into the commutative diagram
Z ( x ) L , U → ⟨ h , t ⟩ Z ( h − 1 x ) L ′ , U ′ ς ( x ) L , U ↓ ↓ ς ( x ) L ′ , U ′ X L , U → ⟨ h , t ⟩ X L ′ , U ′ \begin{CD}
Z(x)_{L,U} @>{\langle h,t\rangle}>> Z(h^{-1}x)_{L',U'} \\
@V{\varsigma(x)_{L,U}}VV @VV{\varsigma(x)_{L',U'}}V \\
X_{L,U} @>{\langle h,t\rangle}>> X_{L',U'}
\end{CD} Z ( x ) L , U ς ( x ) L , U ↓ ⏐ X L , U ⟨ h , t ⟩ ⟨ h , t ⟩ Z ( h − 1 x ) L ′ , U ′ ↓ ⏐ ς ( x ) L ′ , U ′ X L ′ , U ′
in which the bottom morphism is from the functor X ∙ , ∙ X_{\bullet,\bullet} X ∙ , ∙ .
The natural action of T ( A ∞ ) / U T(\mathbb{A}^{\infty})/U T ( A ∞ ) / U on Z ( x ) L , U Z(x)_{L,U} Z ( x ) L , U also factors through Ξ U \Xi_U Ξ U under which the morphism ς ( x ) L , U \varsigma(x)_{L,U} ς ( x ) L , U is equivariant. Similar to X L X_L X L , we have the quotient scheme Z ( x ) L : = Z ( x ) L , U / Ξ U Z(x)_L := Z(x)_{L,U}/\Xi_U Z ( x ) L := Z ( x ) L , U / Ξ U , which is independent of the choice of U U U , together with the induced morphism
ς ( x ) L : Z ( x ) L ⟶ X L (11) \varsigma(x)_L: Z(x)_L \longrightarrow X_L \tag*{(11)} ς ( x ) L : Z ( x ) L ⟶ X L (11)
fitting into the commutative diagrams
Z ( x ) L , U → ς ( x ) L , U X L , U π ( x ) L , U ↓ ↓ π L , U Z ( x ) L → ς ( x ) L X L (12) \begin{CD}
Z(x)_{L,U} @>{\varsigma(x)_{L,U}}>> X_{L,U} \\
@V{\pi(x)_{L,U}}VV @VV{\pi_{L,U}}V \\
Z(x)_L @>{\varsigma(x)_L}>> X_L \tag*{(12)}
\end{CD} Z ( x ) L , U π ( x ) L , U ↓ ⏐ Z ( x ) L ς ( x ) L , U ς ( x ) L X L , U ↓ ⏐ π L , U X L (12)
and
Z ( x ) L → ⟨ h ⟩ Z ( h − 1 x ) L ′ ς ( x ) L ↓ ↓ ς ( x ) L ′ X L → ⟨ h ⟩ X L ′ \begin{CD}
Z(x)_L @>{\langle h\rangle}>> Z(h^{-1}x)_{L'} \\
@V{\varsigma(x)_L}VV @VV{\varsigma(x)_{L'}}V \\
X_L @>{\langle h\rangle}>> X_{L'}
\end{CD} Z ( x ) L ς ( x ) L ↓ ⏐ X L ⟨ h ⟩ ⟨ h ⟩ Z ( h − 1 x ) L ′ ↓ ⏐ ς ( x ) L ′ X L ′
in which the bottom morphism is from the functor X ∙ X_\bullet X ∙ . The morphism ς ( x ) L \varsigma(x)_L ς ( x ) L is finite and unramified, and depends only on the coset L x Lx Lx .
Remark 3.10. The scheme Z ( x ) L Z(x)_L Z ( x ) L is nonempty only when T ( x ) T(x) T ( x ) belongs to Herm r ( F ) + \operatorname{Herm}_r(F)^+ Herm r ( F ) + whose rank equals rank x \operatorname{rank}x rank x . In this case, dim X L − dim Z ( x ) L = rank x \dim X_L-\dim Z(x)_L=\operatorname{rank}x dim X L − dim Z ( x ) L = rank x .
Remark 3.11. We discuss the relation between ς ( x ) L \varsigma(x)_L ς ( x ) L and the special morphisms of Shimura varieties when T ( x ) ∈ Herm r ∘ ( F ) + T(x)\in\operatorname{Herm}^{\circ}_r(F)^+ T ( x ) ∈ Herm r ∘ ( F ) + . Take an element u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) with v : = v ( u ) ∈ V F ( ∞ ) v:=v(u)\in V_F^{(\infty)} v := v ( u ) ∈ V F ( ∞ ) , and fix the data as in Remark 3.4. We may choose a tuple y = ( y 1 , … , y r ) ∈ V ( v ) r \mathbf{y}=(y_1,\ldots,y_r)\in V(v)^r y = ( y 1 , … , y r ) ∈ V ( v ) r and an element h ∈ H ( A F ∞ ) h\in H(\mathbb{A}^{\infty}_F) h ∈ H ( A F ∞ ) such that y i = h x i y_i=h x_i y i = h x i holds in V ∞ V^\infty V ∞ for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r . Denote by V ( v ) y V(v)^{\mathbf{y}} V ( v ) y the orthogonal complement of y \mathbf{y} y in V ( v ) V(v) V ( v ) and put H ( v ) y : = U ( V ( v ) y ) H(v)^{\mathbf{y}}:=U(V(v)^{\mathbf{y}}) H ( v ) y := U ( V ( v ) y ) . Then the composite morphism
Z ( x ) L ⊗ E , u u ( E ) → ⟨ h ⟩ − 1 Z ( y ) h L h − 1 ⊗ E , u u ( E ) → ς ( y ) h L h − 1 ⊗ E , u u ( E ) X h L h − 1 ⊗ E , u u ( E ) → ( 3.3 ) Sh ( H ( v ) ) h L h − 1 u Z(x)_L\otimes_{E,u}u(E)\xrightarrow{\langle h\rangle^{-1}}Z(y)_{hLh^{-1}}\otimes_{E,u}u(E)\xrightarrow{\varsigma(y)_{hLh^{-1}}\otimes_{E,u}u(E)}X_{hLh^{-1}}\otimes_{E,u}u(E)\xrightarrow{(3.3)}\operatorname{Sh}(H(v))^{u}_{hLh^{-1}} Z ( x ) L ⊗ E , u u ( E ) ⟨ h ⟩ − 1 Z ( y ) h L h − 1 ⊗ E , u u ( E ) ς ( y ) h L h − 1 ⊗ E , u u ( E ) X h L h − 1 ⊗ E , u u ( E ) ( 3.3 ) Sh ( H ( v ) ) h L h − 1 u
factors through the morphism
Sh ( H ( v ) y ) ( h L h − 1 ) y u ⟶ Sh ( H ( v ) ) h L h − 1 u \operatorname{Sh}(H(v)^{\mathbf{y}})^{u}_{(hLh^{-1})^{\mathbf{y}}}\longrightarrow\operatorname{Sh}(H(v))^{u}_{hLh^{-1}} Sh ( H ( v ) y ) ( h L h − 1 ) y u ⟶ Sh ( H ( v ) ) h L h − 1 u
functorial in L L L (uniquely), where ( h L h − 1 ) y : = h L h − 1 ∩ H ( v ) y ( A ∞ ) (hLh^{-1})^{\mathbf{y}}:=hLh^{-1}\cap H(v)^{\mathbf{y}}(\mathbb{A}^{\infty}) ( h L h − 1 ) y := h L h − 1 ∩ H ( v ) y ( A ∞ ) , and induces an isomorphism
Z ( x ) L ⊗ E , u u ( E ) → ∼ Sh ( H ( v ) y ) ( h L h − 1 ) y u . Z(x)_L\otimes_{E,u}u(E)\xrightarrow{\sim}\operatorname{Sh}(H(v)^{\mathbf{y}})^{u}_{(hLh^{-1})^{\mathbf{y}}}. Z ( x ) L ⊗ E , u u ( E ) ∼ Sh ( H ( v ) y ) ( h L h − 1 ) y u .
As a consequence, we have the following commutative diagram
Z ( x ) L ⊗ E , u u ( E ) → ς ( x ) L ⊗ E , u u ( E ) X L ⊗ E , u u ( E ) ≃ ↓ ≃ ↓ Sh ( H ( v ) ) ( h L h − 1 ) L v u → Sh ( H ( v ) ) h L h − 1 u → ⟨ h ⟩ Sh ( H ( v ) ) L u (13) \begin{CD}
Z(x)_L \otimes_{E,u} u(E) @>{\varsigma(x)_L\otimes_{E,u}u(E)}>> X_L\otimes_{E,u}u(E) \\
@V{\simeq}VV @V{\simeq}VV \\
\operatorname{Sh}(H(v))^u_{(h_Lh^{-1})_L^v} @>>> \operatorname{Sh}(H(v))^u_{h_Lh^{-1}} @>{\langle h\rangle}>> \operatorname{Sh}(H(v))^u_L \tag*{(13)}
\end{CD} Z ( x ) L ⊗ E , u u ( E ) ≃ ↓ ⏐ Sh ( H ( v ) ) ( h L h − 1 ) L v u ς ( x ) L ⊗ E , u u ( E ) X L ⊗ E , u u ( E ) ≃ ↓ ⏐ Sh ( H ( v ) ) h L h − 1 u ⟨ h ⟩ Sh ( H ( v ) ) L u (13)
in Sch / u ( E ) s m . q p r o j \operatorname{Sch}^{\mathrm{sm.qproj}}_{/u(E)} Sch / u ( E ) sm.qproj .
We now construct, when T ( x ) ∈ Herm r ( F ) + T(x)\in\operatorname{Herm}_r(F)^+ T ( x ) ∈ Herm r ( F ) + , a canonical map
μ ( x ) L : ς ( x ) L ∗ M L ∨ ⟶ 1 Z ( x ) L , E ⊕ r (14) \mu(x)_L:\varsigma(x)^*_L M^\vee_L\longrightarrow\mathbb{1}^{\oplus r}_{Z(x)_L,E} \tag*{(14)} μ ( x ) L : ς ( x ) L ∗ M L ∨ ⟶ 1 Z ( x ) L , E ⊕ r (14)
in Mot ( Z ( x ) L , E ) \operatorname{Mot}(Z(x)_L,E) Mot ( Z ( x ) L , E ) . By (12) , it suffices to construct a map μ ( x ) L , U : ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ 1 Z ( x ) L , U , E ⊕ r \mu(x)_{L,U}:\varsigma(x)^*_{L,U}\pi^*_{L,U}M^\vee_L\longrightarrow\mathbb{1}^{\oplus r}_{Z(x)_{L,U},E} μ ( x ) L , U : ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ 1 Z ( x ) L , U , E ⊕ r compatible with the descent data, that is, a map in Mot ( Z ( x ) L , U / Ξ U , E ) \operatorname{Mot}(Z(x)_{L,U}/\Xi_U,E) Mot ( Z ( x ) L , U / Ξ U , E ) . Since π L , U \pi_{L,U} π L , U is étale, the natural map π L , U ∗ M L ∨ → ( π L , U ∗ M L ) ∨ \pi^*_{L,U}M^\vee_L\to(\pi^*_{L,U}M_L)^\vee π L , U ∗ M L ∨ → ( π L , U ∗ M L ) ∨ is an isomorphism. Thus, we have
ς ( x ) L , U ∗ π L , U ∗ M L ∨ ≃ ς ( x ) L , U ∗ ( π L , U ∗ M L ) ∨ ≃ ς ( x ) L , U ∗ M L , U ∨ ≃ ς ( x ) L , U ∗ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ) ≃ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ς ( x ) L , U ∗ h 1 ( A L , U / X L , U , Q ) ≃ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( ς ( x ) L , U ∗ A L , U / Z ( x ) L , U , Q ) . \begin{aligned}
\varsigma(x)^*_{L,U}\pi^*_{L,U}M^\vee_L\simeq\varsigma(x)^*_{L,U}(\pi^*_{L,U}M_L)^\vee\\
&\simeq\varsigma(x)^*_{L,U}M^\vee_{L,U} \\
&\simeq\varsigma(x)^*_{L,U}\left(h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h^1(A_{L,U}/X_{L,U},\mathbb{Q})\right) \\
&\simeq\varsigma(x)^*_{L,U}h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}\varsigma(x)^*_{L,U}h^1(A_{L,U}/X_{L,U},\mathbb{Q}) \\
&\simeq\varsigma(x)^*_{L,U}h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h^1(\varsigma(x)^*_{L,U}A_{L,U}/Z(x)_{L,U},\mathbb{Q}).
\end{aligned} ς ( x ) L , U ∗ π L , U ∗ M L ∨ ≃ ς ( x ) L , U ∗ ( π L , U ∗ M L ) ∨ ≃ ς ( x ) L , U ∗ M L , U ∨ ≃ ς ( x ) L , U ∗ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A L , U / X L , U , Q ) ) ≃ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ς ( x ) L , U ∗ h 1 ( A L , U / X L , U , Q ) ≃ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( ς ( x ) L , U ∗ A L , U / Z ( x ) L , U , Q ) .
Note that over Z ( x ) L , U Z(x)_{L,U} Z ( x ) L , U , we have homomorphisms ν i : ς ( x ) L , U ∗ A 0 , L , U → ς ( x ) L , U ∗ A L , U \nu_i:\varsigma(x)^*_{L,U}A_{0,L,U}\to\varsigma(x)^*_{L,U}A_{L,U} ν i : ς ( x ) L , U ∗ A 0 , L , U → ς ( x ) L , U ∗ A L , U for 1 ≤ i ≤ r 1\leq i\leq r 1 ≤ i ≤ r , which induce maps
ν i ∗ : h 1 ( ς ( x ) L , U ∗ A L , U / Z ( x ) L , U , Q ) ⟶ h 1 ( ς ( x ) L , U ∗ A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F \nu_i^*:h^1(\varsigma(x)^*_{L,U}A_{L,U}/Z(x)_{L,U},\mathbb{Q})\longrightarrow h^1(\varsigma(x)^*_{L,U}A_{0,L,U}/Z(x)_{L,U},\mathbb{Q})\otimes_{\mathbb{Q}}F ν i ∗ : h 1 ( ς ( x ) L , U ∗ A L , U / Z ( x ) L , U , Q ) ⟶ h 1 ( ς ( x ) L , U ∗ A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F
in Mot ( Z ( x ) L , U , E ) \operatorname{Mot}(Z(x)_{L,U},E) Mot ( Z ( x ) L , U , E ) . Taking sum, we obtain a map
ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ( h 1 ( ς ( x ) L , U ∗ A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F ) ⊕ r \varsigma(x)^*_{L,U}\pi^*_{L,U}M^\vee_L\longrightarrow\varsigma(x)^*_{L,U}h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}\left(h^1(\varsigma(x)^*_{L,U}A_{0,L,U}/Z(x)_{L,U},\mathbb{Q})\otimes_{\mathbb{Q}}F\right)^{\oplus r} ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ ς ( x ) L , U ∗ h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 ( h 1 ( ς ( x ) L , U ∗ A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F ) ⊕ r
→ ∼ ς ( x ) L , U ∗ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F ) ⊕ r \xrightarrow{\sim}\varsigma(x)^*_{L,U}\left(h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h^1(A_{0,L,U}/Z(x)_{L,U},\mathbb{Q})\otimes_{\mathbb{Q}}F\right)^{\oplus r} ∼ ς ( x ) L , U ∗ ( h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A 0 , L , U / Z ( x ) L , U , Q ) ⊗ Q F ) ⊕ r
Composing with the evaluation map
h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A 0 , L , U / Z ( x ) L , U , Q ) ⟶ 1 X L , U , E 0 , h^1(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{E_0}h^1(A_{0,L,U}/Z(x)_{L,U},\mathbb{Q})\longrightarrow\mathbb{1}_{X_{L,U},E_0}, h 1 ( A 0 , L , U / X L , U , Q ) ⊗ E 0 h 1 ( A 0 , L , U / Z ( x ) L , U , Q ) ⟶ 1 X L , U , E 0 ,
we obtain
μ ( x ) L , U : ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ ς ( x ) L , U ∗ 1 X L , U , E ⊕ r = 1 Z ( x ) L , U , E ⊕ r . \mu(x)_{L,U}:\varsigma(x)^*_{L,U}\pi^*_{L,U}M^\vee_L\longrightarrow\varsigma(x)^*_{L,U}\mathbb{1}^{\oplus r}_{X_{L,U},E}=\mathbb{1}^{\oplus r}_{Z(x)_{L,U},E}. μ ( x ) L , U : ς ( x ) L , U ∗ π L , U ∗ M L ∨ ⟶ ς ( x ) L , U ∗ 1 X L , U , E ⊕ r = 1 Z ( x ) L , U , E ⊕ r .
It is straightforward to check that μ ( x ) L , U \mu(x)_{L,U} μ ( x ) L , U is compatible with descent data for Ξ U \Xi_U Ξ U , hence descends to a map μ ( x ) L \mu(x)_L μ ( x ) L (14) we look for.
Remark 3.12. The special morphism ς ( x ) L \varsigma(x)_L ς ( x ) L depends only on the span of components of x x x . However, the map μ ( x ) \mu(x) μ ( x ) indeed depends on x x x .
Automorphic motives over unitary Shimura varieties
Take a hermitian E E E -weight λ \lambda λ of rank m m m (Definition 2.7), and recall from Definition 2.7(3) that E λ ⊆ E E^\lambda\subseteq E E λ ⊆ E is the field of definition of λ \lambda λ . In this section, we define a Cartesian functor M ∙ , λ : Hk ( H ( A F ∞ ) ) ⟶ Mot ( Sch E s m . q p r o j , E λ ) M_{\bullet,\lambda}:\operatorname{Hk}(H(\mathbb{A}_F^\infty))\longrightarrow\operatorname{Mot}(\operatorname{Sch}^{\mathrm{sm.qproj}}_E,E^\lambda) M ∙ , λ : Hk ( H ( A F ∞ )) ⟶ Mot ( Sch E sm.qproj , E λ ) over X ∙ X_\bullet X ∙ .
Notation 3.13. For an E E E -linear object M M M in a pseudo-abelian category C \mathcal{C} C , we have a unique decomposition
M ⊗ Q E = ⨁ u ∈ Hom ( E , E ‾ ) M [ u ] M\otimes_{\mathbb{Q}}E=\bigoplus_{u\in\operatorname{Hom}(E,\overline{E})}M[u] M ⊗ Q E = u ∈ Hom ( E , E ) ⨁ M [ u ]
in the category of E \mathbb{E} E -linear objects in C \mathcal{C} C , in which M [ u ] M_{[u]} M [ u ] is the direct summand on which E E E acts via the embedding u : E → E u:E\to\mathbb{E} u : E → E .
Put b = ( b u : = b ( λ u + ) ) u ∈ N Hom ( E , C ) b=(b_u:=b(\lambda_u^+))_{u}\in\mathbb{N}^{\operatorname{Hom}(E,\mathcal{C})} b = ( b u := b ( λ u + ) ) u ∈ N Hom ( E , C ) and b v : = ( b u ) v ( u ) = v b_v:=(b_u)_{v(u)=v} b v := ( b u ) v ( u ) = v for every v ∈ B F ( ∞ ) v\in\mathcal{B}_{F}^{(\infty)} v ∈ B F ( ∞ ) . Similarly, we have b ˉ \bar{b} b ˉ and b ˉ v \bar{b}_v b ˉ v for λ ˉ \bar{\lambda} λ ˉ instead of λ \lambda λ .
Take an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(A_F^{\infty})) L ∈ Hk ( H ( A F ∞ )) . Take an element v ∈ B F ( ∞ ) v\in\mathcal{B}_{F}^{(\infty)} v ∈ B F ( ∞ ) . Put
T b v ( M L ) : = ⨂ v ( u ) = v T b u ( M L [ u ] ) T_{b_v}(M_L):=\bigotimes_{v(u)=v}T^{b_u}(M_{L[u]}) T b v ( M L ) := v ( u ) = v ⨂ T b u ( M L [ u ] )
(tensor over E \mathbb{E} E ) as an object in Mot ( X L , E ) \operatorname{Mot}(X_L,\mathbb{E}) Mot ( X L , E ) , which admits a natural action by S b v \mathfrak{S}_{b_v} S b v by switching factors. The evaluation map (8) induces an evaluation map
⨂ v ( u ) = v M L [ u ] ⟶ 1 X L , E , \bigotimes_{v(u)=v}M_{L[u]}\longrightarrow\mathbb{1}_{X_L,\mathbb{E}}, v ( u ) = v ⨂ M L [ u ] ⟶ 1 X L , E ,
(tensor over E \mathbb{E} E ) which further induces
T b v ( M L ) ⊗ E T b ˉ v ( M L ) ⟶ 1 X L , E ; T_{b_v}(M_L)\otimes_{\mathbb{E}}T_{\bar{b}_v}(M_L)\longrightarrow\mathbb{1}_{X_L,\mathbb{E}}; T b v ( M L ) ⊗ E T b ˉ v ( M L ) ⟶ 1 X L , E ;
for every pair I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } I=(i_u)_{v(u)=v}\in\prod_{v(u)=v}\{1,\ldots,b_u\} I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } , a contraction map
C I : T b v ( M L ) ⟶ T b v − 1 ( M L ) C_I:T_{b_v}(M_L)\longrightarrow T_{b_v-1}(M_L) C I : T b v ( M L ) ⟶ T b v − 1 ( M L )
by evaluating i u i_u i u -th factors of T b u ( M L [ u ] ) T^{b_u}(M_{L[u]}) T b u ( M L [ u ] ) for v ( u ) = v v(u)=v v ( u ) = v (set T b v − 1 ( M L ) = 0 T_{b_v-1}(M_L)=0 T b v − 1 ( M L ) = 0 if b u − 1 < 0 b_u-1<0 b u − 1 < 0 for some u u u ).
Lemma 3.14. The following statements hold.
(1) The map
∑ I C I : T b v ( M L ) ⟶ ⨁ I ∈ ∏ v ( u ) = v { 1 , … , b u } T b v − 1 ( M L ) \sum_I C_I:T_{b_v}(M_L)\longrightarrow\bigoplus_{I\in\prod_{v(u)=v}\{1,\ldots,b_u\}}T_{b_v-1}(M_L) I ∑ C I : T b v ( M L ) ⟶ I ∈ ∏ v ( u ) = v { 1 , … , b u } ⨁ T b v − 1 ( M L ) admits kernel in the category Mot ( X L , E ) \operatorname{Mot}(X_L,\mathbb{E}) Mot ( X L , E ) , which we denote by T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) .
(2) The induced map
T b v ( M L ) ⟶ T ~ b ˉ v ( M L ) ∨ ⟶ T [ b ˉ v ] ( M L ) ∨ T_{b_v}(M_L)\longrightarrow\widetilde{T}_{\bar{b}_v}(M_L)^\vee\longrightarrow T_{[\bar{b}_v]}(M_L)^\vee T b v ( M L ) ⟶ T b ˉ v ( M L ) ∨ ⟶ T [ b ˉ v ] ( M L ) ∨ admits kernel as well, which we denote by T [ b ˉ v ] ⊥ ( M L ) T^\perp_{[\bar{b}_v]}(M_L) T [ b ˉ v ] ⊥ ( M L ) .
(3) The induced map
T [ b v ] ( M L ) ⊕ T [ b v ] ⊥ ( M L ) ⟶ T b v ( M L ) T_{[b_v]}(M_L)\oplus T^\perp_{[b_v]}(M_L)\longrightarrow T_{b_v}(M_L) T [ b v ] ( M L ) ⊕ T [ b v ] ⊥ ( M L ) ⟶ T b v ( M L ) is an isomorphism.
Proof. For the proof of this lemma, we need to use étale realization in the next section until Definition 3.21 and the conservativity property for abelian motives in [37 ] .
Choose a rational prime ℓ \ell ℓ and take an element u ∈ Hom ( E , E ) u\in\operatorname{Hom}(E,\mathbb{E}) u ∈ Hom ( E , E ) with v ( u ) = v v(u)=v v ( u ) = v . Similar to Remark 2.4(2) (with L = E ℓ L=\mathbb{E}_{\ell} L = E ℓ ), the map
( ⋀ m M L , ℓ [ u ] ) ⊗ E ℓ M L , ℓ [ u ] ∨ ⟶ ⋀ m − 1 M L , ℓ [ u ] (15) \left(\bigwedge^m M_{L,\ell}[u]\right)\otimes_{\mathbb{E}_{\ell}}M_{L,\ell}[u]^\vee\longrightarrow\bigwedge^{m-1}M_{L,\ell}[u] \tag*{(15)} ( ⋀ m M L , ℓ [ u ] ) ⊗ E ℓ M L , ℓ [ u ] ∨ ⟶ ⋀ m − 1 M L , ℓ [ u ] (15)
(see §3.5 for the notation) given by pairing the last two factors induces an isomorphism
M L , ℓ [ u ] ∨ ⟶ ( ⋀ m − 1 M L , ℓ [ u ] ) ⊗ E ℓ ( ⋀ m M L , ℓ [ u ] ) ∨ (16) M_{L,\ell}[u]^\vee\longrightarrow\left(\bigwedge^{m-1}M_{L,\ell}[u]\right)\otimes_{\mathbb{E}_{\ell}}\left(\bigwedge^m M_{L,\ell}[u]\right)^\vee \tag*{(16)} M L , ℓ [ u ] ∨ ⟶ ( ⋀ m − 1 M L , ℓ [ u ] ) ⊗ E ℓ ( ⋀ m M L , ℓ [ u ] ) ∨ (16)
which further induces an isomorphism
T b v ( M L , ℓ ) ⟶ T b u ( M L , ℓ [ u ] ) ⊗ E ℓ T b ˉ u ( ⋀ m − 1 M L , ℓ [ u ] ) ⊗ E ℓ T b ˉ u ( ⋀ m M L , ℓ [ u ] ) − b u (17) T_{b_v}(M_{L,\ell})\longrightarrow T^{b_u}(M_{L,\ell}[u])\otimes_{\mathbb{E}_{\ell}}T^{\bar{b}_u}\left(\bigwedge^{m-1}M_{L,\ell}[u]\right)\otimes_{\mathbb{E}_{\ell}}T^{\bar{b}_u}\left(\bigwedge^m M_{L,\ell}[u]\right)^{-b_u} \tag*{(17)} T b v ( M L , ℓ ) ⟶ T b u ( M L , ℓ [ u ]) ⊗ E ℓ T b ˉ u ( ⋀ m − 1 M L , ℓ [ u ] ) ⊗ E ℓ T b ˉ u ( ⋀ m M L , ℓ [ u ] ) − b u (17)
so that T [ b v ] ( M L , ℓ ) T_{[b_v]}(M_{L,\ell}) T [ b v ] ( M L , ℓ ) (Definition 3.21) is the kernel of the idempotent δ λ u , λ ˉ u ∈ Q [ S b u + ( m − 1 ) b ˉ u ] \delta_{\lambda_u,\bar{\lambda}_u}\in\mathbb{Q}[\mathfrak{S}_{b_u+(m-1)\bar{b}_u}] δ λ u , λ ˉ u ∈ Q [ S b u + ( m − 1 ) b ˉ u ] (as in Remark 2.4(2)) via the action of S b u + ( m − 1 ) b ˉ u \mathfrak{S}_{b_u+(m-1)\bar{b}_u} S b u + ( m − 1 ) b ˉ u on T b u + ( m − 1 ) b ˉ u ( M L , ℓ [ u ] ) ⊗ E ℓ T b ˉ u ( ⋀ m M L , ℓ [ u ] ) − b u T^{b_u+(m-1)\bar{b}_u}(M_{L,\ell}[u])\otimes_{\mathbb{E}_{\ell}}T^{\bar{b}_u}( \bigwedge^m M_{L,\ell}[u])^{-b_u} T b u + ( m − 1 ) b ˉ u ( M L , ℓ [ u ]) ⊗ E ℓ T b ˉ u ( ⋀ m M L , ℓ [ u ] ) − b u . In particular, T [ b v ] ( M L , ℓ ) T_{[b_v]}(M_{L,\ell}) T [ b v ] ( M L , ℓ ) is a direct summand of T b v ( M L , ℓ ) T_{b_v}(M_{L,\ell}) T b v ( M L , ℓ ) .
Now we repeat the same process for motives. We have the similar map
( ⋀ m M L [ u ] ) ⊗ E M L [ u ] ∨ → ⋀ m − 1 M L [ u ] (18) \left(\bigwedge^{m} M_{L[u]}\right) \otimes_{\mathbb{E}} M_{L[u]}^{\vee} \to\bigwedge^{m-1} M_{L[u]} \tag*{(18)} ( ⋀ m M L [ u ] ) ⊗ E M L [ u ] ∨ → ⋀ m − 1 M L [ u ] (18)
in the category Mot ( X L , E ) \operatorname{Mot}(X_L,\mathbb{E}) Mot ( X L , E ) , which induces a map
M L [ u ] ∨ → ( ⋀ m − 1 M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) ∨ , (19) M_{L[u]}^{\vee} \to\left(\bigwedge^{m-1} M_{L[u]}\right) \otimes_{\mathbb{E}} \left(\bigwedge^{m} M_{L[u]}\right)^{\vee}, \tag*{(19)} M L [ u ] ∨ → ( ⋀ m − 1 M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) ∨ , (19)
which further induces a map
T b v ( M L ) → T b u ( M L [ u ] ) ⊗ E T b ˉ u ( ⋀ m − 1 M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u (20) T_{b_v}(M_L) \to T^{b^u}(M_{L[u]}) \otimes_{\mathbb{E}} T^{\bar b_u}\left(\bigwedge^{m-1}M_{L[u]}\right) \otimes_{\mathbb{E}} \left(\bigwedge^{m}M_{L[u]}\right)^{-\bar b_u} \tag*{(20)} T b v ( M L ) → T b u ( M L [ u ] ) ⊗ E T b ˉ u ( ⋀ m − 1 M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u (20)
(here, ( ⋀ m M L [ u ] ) − b ˉ u \left(\bigwedge^{m}M_{L[u]}\right)^{-\bar b_u} ( ⋀ m M L [ u ] ) − b ˉ u really means T b ˉ u ( ⋀ m M L [ u ] ) ∨ T^{\bar b_u}\left(\bigwedge^{m}M_{L[u]}\right)^{\vee} T b ˉ u ( ⋀ m M L [ u ] ) ∨ ). By Lemma 3.20 , the (ℓ \ell ℓ -adic) étale realizations of (18) , (19) and (20) are canonically (15) , (16) and (17) , respectively. By [37 Theorem 4.4] , (20) is an isomorphism since (17) is. Note that the target of (20) is naturally a direct summand of T b u + ( m − 1 ) b u ( M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u T^{b^u+(m-1)b_u}\left(M_{L[u]}\right) \otimes_{\mathbb{E}} \left(\bigwedge^{m}M_{L[u]}\right)^{-\bar b_u} T b u + ( m − 1 ) b u ( M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u . Now we define T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) to be the kernel of δ λ u , λ u ∈ Q [ S b u + ( m − 1 ) b u ] \delta_{\lambda_u,\lambda_u \in\mathbb{Q}[S_{b_u+(m-1)b_u}]} δ λ u , λ u ∈ Q [ S b u + ( m − 1 ) b u ] . Using [37 Theorem 4.4] and the situation for (ℓ \ell ℓ -adic) étale realizations, we know that the natural map T [ b v ] ( M L ) → T b u + ( m − 1 ) b u ( M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u T_{[b_v]}(M_L) \to T^{b^u+(m-1)b_u}(M_{L[u]}) \otimes_{\mathbb{E}} \left(\bigwedge^{m}M_{L[u]}\right)^{-\bar b_u} T [ b v ] ( M L ) → T b u + ( m − 1 ) b u ( M L [ u ] ) ⊗ E ( ⋀ m M L [ u ] ) − b ˉ u factors through T b v ( M L ) T_{b_v}(M_L) T b v ( M L ) ; and that the induced map T [ b v ] ( M L ) → T b v ( M L ) T_{[b_v]}(M_L) \to T_{b_v}(M_L) T [ b v ] ( M L ) → T b v ( M L ) is the kernel of ∑ I C I \sum_I C_I ∑ I C I . Part (1) is proved. Part (2) and (3) follows from a similar argument using the obvious case for (ℓ \ell ℓ -adic) étale realizations and [37 Theorem 4.4] . □
It is clear that the action of S ~ b v \widetilde{S}_{b_v} S b v on T b v ( M L ) T_{b_v}(M_L) T b v ( M L ) induces an action on T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) . Applying the idempotent π λ v : = ⨂ v ( u ) = v π λ u ∈ Q [ S b v ] \pi_{\lambda_v}:=\bigotimes_{v(u)=v}\pi_{\lambda_u}\in\mathbb{Q}[S_{b_v}] π λ v := ⨂ v ( u ) = v π λ u ∈ Q [ S b v ] , we obtain a direct summand
S [ λ v ] ( M L ) : = π λ v ( T [ b v ] ( M L ) ) S_{[\lambda_v]}(M_L):=\pi_{\lambda_v}\left(T_{[b_v]}(M_L)\right) S [ λ v ] ( M L ) := π λ v ( T [ b v ] ( M L ) )
of T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) .
The remaining steps are motivic analogue of Construction 2.8(1,2). Taking into account all v ∈ V F ( ∞ ) v\in\mathcal{V}^{(\infty)}_F v ∈ V F ( ∞ ) , we define
S [ λ ] ( M L ) : = ⨂ v ∈ V F ( ∞ ) S [ λ v ] ( M L ) S_{[\lambda]}(M_L):=\bigotimes_{v\in\mathcal{V}^{(\infty)}_F}S_{[\lambda_v]}(M_L) S [ λ ] ( M L ) := v ∈ V F ( ∞ ) ⨂ S [ λ v ] ( M L )
(tensor over E \mathbb{E} E ) as an object of Mot ( X L , E ) \operatorname{Mot}(X_L,\mathbb{E}) Mot ( X L , E ) . For every γ ∈ Gal ( E / Q ) \gamma\in\operatorname{Gal}(\mathbb{E}/\mathbb{Q}) γ ∈ Gal ( E / Q ) , its action on E \mathbb{E} E induces an isomorphism
γ [ λ v ] : S [ λ v ] ( M L ) ⊗ E , γ E → ∼ S [ ( γ λ ) γ v ] ( M L ) \gamma_{[\lambda_v]}:S_{[\lambda_v]}(M_L)\otimes_{\mathbb{E},\gamma}\mathbb{E}\xrightarrow{\sim}S_{[(\gamma\lambda)_{\gamma v}]}(M_L) γ [ λ v ] : S [ λ v ] ( M L ) ⊗ E , γ E ∼ S [( γ λ ) γ v ] ( M L )
for every v ∈ V F ( ∞ ) v\in\mathcal{V}^{(\infty)}_F v ∈ V F ( ∞ ) , which further induces an isomorphism
γ [ λ ] : S [ λ ] ( M L ) ⊗ E , γ E → ∼ S [ γ λ ] ( M L ) \gamma_{[\lambda]}:S_{[\lambda]}(M_L)\otimes_{\mathbb{E},\gamma}\mathbb{E}\xrightarrow{\sim}S_{[\gamma\lambda]}(M_L) γ [ λ ] : S [ λ ] ( M L ) ⊗ E , γ E ∼ S [ γ λ ] ( M L )
by taking the tensor products over all v ∈ V F ( ∞ ) v\in\mathcal{V}^{(\infty)}_F v ∈ V F ( ∞ ) . In particular, for γ ∈ Gal ( E / E λ ) \gamma\in\operatorname{Gal}(\mathbb{E}/\mathbb{E}^{\lambda}) γ ∈ Gal ( E / E λ ) , γ [ λ ] \gamma_{[\lambda]} γ [ λ ] is an E λ \mathbb{E}^{\lambda} E λ -linear automorphism of S [ λ ] ( M L ) S_{[\lambda]}(M_L) S [ λ ] ( M L ) . It is straightforward to check that this defines a homomorphism from Gal ( E / E λ ) \operatorname{Gal}(\mathbb{E}/\mathbb{E}^{\lambda}) Gal ( E / E λ ) to Aut E λ ( S [ λ ] ( M L ) ) \operatorname{Aut}_{\mathbb{E}^{\lambda}}(S_{[\lambda]}(M_L)) Aut E λ ( S [ λ ] ( M L )) . Thus, we may define
M L , λ : = S [ λ ] ( M L ) Gal ( E / E λ ) ∈ Mot ( X L , E λ ) , M_{L,\lambda}:=S_{[\lambda]}(M_L)^{\operatorname{Gal}(\mathbb{E}/\mathbb{E}^{\lambda})}\in\operatorname{Mot}(X_L,\mathbb{E}^{\lambda}), M L , λ := S [ λ ] ( M L ) Gal ( E / E λ ) ∈ Mot ( X L , E λ ) ,
so that the natural map M L , λ ⊗ E λ E → S [ λ ] ( M L ) M_{L,\lambda}\otimes_{\mathbb{E}^{\lambda}}\mathbb{E}\to S_{[\lambda]}(M_L) M L , λ ⊗ E λ E → S [ λ ] ( M L ) is an isomorphism. For general γ ∈ Gal ( E / Q ) \gamma\in\operatorname{Gal}(\mathbb{E}/\mathbb{Q}) γ ∈ Gal ( E / Q ) , the isomorphism γ [ λ ] \gamma_{[\lambda]} γ [ λ ] induces an isomorphism
γ λ : M L , λ ⊗ E λ , γ E γ λ → ∼ M L , γ λ \gamma_{\lambda}:M_{L,\lambda}\otimes_{\mathbb{E}^{\lambda},\gamma}\mathbb{E}^{\gamma\lambda}\xrightarrow{\sim}M_{L,\gamma\lambda} γ λ : M L , λ ⊗ E λ , γ E γ λ ∼ M L , γ λ
in the category Mot ( X L , E γ λ ) \operatorname{Mot}(X_L,\mathbb{E}^{\gamma\lambda}) Mot ( X L , E γ λ ) .
Definition 3.15. We define the automorphic (relative) motive of weight λ \lambda λ over X L X_L X L to be M L , λ M_{L,\lambda} M L , λ . It is straightforward to check that
M ∙ , λ : Hk ( H ( A F ∞ ) ) → Mot ( Sch / E s m . q p r o j , E λ ) M_{\bullet,\lambda}:\operatorname{Hk}\left(H(\mathbb{A}^{\infty}_F)\right)\to\operatorname{Mot}\left(\operatorname{Sch}^{\mathrm{sm.qproj}}_{/E},\mathbb{E}^{\lambda}\right) M ∙ , λ : Hk ( H ( A F ∞ ) ) → Mot ( Sch / E sm.qproj , E λ )
is a Cartesian functor over X ∙ X_\bullet X ∙ .
Remark 3.16. The construction of the functor M ∙ , λ \mathrm{M}_{\bullet,\lambda} M ∙ , λ a priori depends on the choice of E 0 E_0 E 0 . We expect that it does not; however, we do not know how to prove yet.
Lemma 3.14 implies that the map
T [ b v ] ( M L ) ⊗ E T [ b ‾ v ] ( M L ) ⟶ 1 X L , E \mathrm{T}_{[b_v]}(\mathrm{M}_L) \otimes_{\mathbb{E}} \mathrm{T}_{[\overline{b}_v]}(\mathrm{M}_L) \longrightarrow\mathbb{1}_{X_L,\mathbb{E}} T [ b v ] ( M L ) ⊗ E T [ b v ] ( M L ) ⟶ 1 X L , E
induced from (3.10) is also an evaluation map. Since the above map is S b v \mathfrak{S}_{b_v} S b v -invariant, the induced map
S [ λ v ] ( M L ) ⊗ E S [ λ ‾ v ] ( M L ) ⟶ 1 X L , E \mathrm{S}_{[\lambda_v]}(\mathrm{M}_L) \otimes_{\mathbb{E}} \mathrm{S}_{[\overline{\lambda}_v]}(\mathrm{M}_L) \longrightarrow\mathbb{1}_{X_L,\mathbb{E}} S [ λ v ] ( M L ) ⊗ E S [ λ v ] ( M L ) ⟶ 1 X L , E
is an evaluation map. Taking products over all v ∈ V F ( ∞ ) v \in\mathcal{V}_{F}^{(\infty)} v ∈ V F ( ∞ ) , the induced map
S [ λ ] ( M L ) ⊗ E S [ λ ‾ ] ( M L ) ⟶ 1 X L , E \mathrm{S}_{[\lambda]}(\mathrm{M}_L) \otimes_{\mathbb{E}} \mathrm{S}_{[\overline{\lambda}]}(\mathrm{M}_L) \longrightarrow\mathbb{1}_{X_L,\mathbb{E}} S [ λ ] ( M L ) ⊗ E S [ λ ] ( M L ) ⟶ 1 X L , E
is an evaluation map. Since the above map is Gal ( E / E λ ) \operatorname{Gal}(E/E^\lambda) Gal ( E / E λ ) -equivariant, the induced map
e L , λ : M L , λ ⊗ E λ M L , λ ‾ ⟶ 1 X L , E λ (21) e_{L,\lambda}\colon\mathrm{M}_{L,\lambda} \otimes_{E^\lambda} \mathrm{M}_{L,\overline{\lambda}} \longrightarrow\mathbb{1}_{X_L,E^\lambda} \tag*{(21)} e L , λ : M L , λ ⊗ E λ M L , λ ⟶ 1 X L , E λ (21)
is an evaluation map.
In the rest of this section, we will relate (21) to the Poincaré duality pairing on the (product of) universal abelian scheme, which will be used in §4.1 for the definition of height pairing.
Choose an object U ∈ Hk ( T ( A ∞ ) ) U \in\operatorname{Hk}(T(A^\infty)) U ∈ Hk ( T ( A ∞ )) . Put M L , U , λ : = π L , U ∗ M L , λ \mathrm{M}_{L,U,\lambda}:=\pi_{L,U}^{*}\mathrm{M}_{L,\lambda} M L , U , λ := π L , U ∗ M L , λ . Then M L , U , λ \mathrm{M}_{L,U,\lambda} M L , U , λ can be obtained from M L , U \mathrm{M}_{L,U} M L , U via the same process as M L \mathrm{M}_L M L . In particular, M L , U , λ \mathrm{M}_{L,U,\lambda} M L , U , λ is naturally an E λ E^\lambda E λ -linear direct summand of
⨂ u ∈ Hom ( E , E ) T b u ( M L , U [ u ] ) \bigotimes_{u\in\operatorname{Hom}(E,E)} \mathrm{T}^{b_u}(\mathrm{M}_{L,U}[u]) u ∈ Hom ( E , E ) ⨂ T b u ( M L , U [ u ])
(tensor over E \mathbb{E} E ), and we have an evaluation map
e L , U , λ : M L , U , λ ⊗ E λ M L , U , λ ‾ ⟶ 1 X L , U , E λ e_{L,U,\lambda}\colon\mathrm{M}_{L,U,\lambda} \otimes_{E^\lambda} \mathrm{M}_{L,U,\overline{\lambda}} \longrightarrow\mathbb{1}_{X_{L,U},E^\lambda} e L , U , λ : M L , U , λ ⊗ E λ M L , U , λ ⟶ 1 X L , U , E λ
as the restriction of the evaluation map
⨂ u ∈ Hom ( E , E ) T b u ( M L , U [ u ] ) ⊗ ⨂ u ∈ Hom ( E , E ) T b u ( M L , U ∨ [ u ] ) ⟶ 1 X L , U , E \bigotimes_{u\in\operatorname{Hom}(E,E)} \mathrm{T}^{b_u}(\mathrm{M}_{L,U}[u]) \otimes\bigotimes_{u\in\operatorname{Hom}(E,E)} \mathrm{T}^{b_u}(\mathrm{M}_{L,U}^{\vee}[u]) \longrightarrow\mathbb{1}_{X_{L,U},\mathbb{E}} u ∈ Hom ( E , E ) ⨂ T b u ( M L , U [ u ]) ⊗ u ∈ Hom ( E , E ) ⨂ T b u ( M L , U ∨ [ u ]) ⟶ 1 X L , U , E
(tensor over E \mathbb{E} E ) induced from (10) . The latter is in turn the restriction of the evaluation map
M L , U ⊗ ∣ b ∣ ⊗ Q M L , U ∨ ⊗ ∣ b ∣ ⟶ 1 X L , U , Q \mathrm{M}_{L,U}^{\otimes|b|} \otimes_{\mathbb{Q}} \mathrm{M}_{L,U}^{\vee\otimes|b|} \longrightarrow\mathbb{1}_{X_{L,U},\mathbb{Q}} M L , U ⊗ ∣ b ∣ ⊗ Q M L , U ∨⊗ ∣ b ∣ ⟶ 1 X L , U , Q
after base change to E \mathbb{E} E , where ∣ b ∣ : = ∑ u ∈ Hom ( E , E ) b u |b|:=\sum_{u\in\operatorname{Hom}(E,E)}b_u ∣ b ∣ := ∑ u ∈ Hom ( E , E ) b u . By (9) , M L , U \mathrm{M}_{L,U} M L , U is a natural direct summand of
h 1 ( A 0 , L , U / X L , U , Q ) ⊗ Q h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ∈ Mot ( X L , U , Q ) , \mathrm{h}^{1}(A_{0,L,U}/X_{L,U},\mathbb{Q})\otimes_{\mathbb{Q}}\mathrm{h}^{1}(A_{L,U}^{\vee}/X_{L,U},\mathbb{Q})\langle1\rangle\in\operatorname{Mot}(X_{L,U},\mathbb{Q}), h 1 ( A 0 , L , U / X L , U , Q ) ⊗ Q h 1 ( A L , U ∨ / X L , U , Q ) ⟨ 1 ⟩ ∈ Mot ( X L , U , Q ) ,
so that M L , U ⊗ ∣ b ∣ \mathrm{M}_{L,U}^{\otimes|b|} M L , U ⊗ ∣ b ∣ is a natural direct summand of
h 1 ( A 0 , L , U / X L , U , Q ) ⊗ ∣ b ∣ ⊗ Q h 1 ( A L , U ∨ / X L , U , Q ) ⊗ ∣ b ∣ ⟨ ∣ b ∣ ⟩ . \mathrm{h}^{1}(A_{0,L,U}/X_{L,U},\mathbb{Q})^{\otimes|b|}\otimes_{\mathbb{Q}}\mathrm{h}^{1}(A_{L,U}^{\vee}/X_{L,U},\mathbb{Q})^{\otimes|b|}\langle|b|\rangle. h 1 ( A 0 , L , U / X L , U , Q ) ⊗ ∣ b ∣ ⊗ Q h 1 ( A L , U ∨ / X L , U , Q ) ⊗ ∣ b ∣ ⟨ ∣ b ∣ ⟩ .
The latter is a natural direct summand of
h 2 ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ , \mathrm{h}^{2|b|}(B_{L,U}^{|b|}/X_{L,U},\mathbb{Q})\langle|b|\rangle, h 2∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ,
where
B L , U : = A 0 , L , U × X L , U A L , U ∨ B_{L,U}:=A_{0,L,U}\times_{X_{L,U}}A_{L,U}^{\vee} B L , U := A 0 , L , U × X L , U A L , U ∨
and
B L , U n : = B L , U × X L , U ⋯ × X L , U B L , U ⏟ n (22) B_{L,U}^{n}:=\underbrace{B_{L,U}\times_{X_{L,U}}\cdots\times_{X_{L,U}}B_{L,U}}_{n} \tag*{(22)} B L , U n := n B L , U × X L , U ⋯ × X L , U B L , U (22)
for n ≥ 0 n\geq0 n ≥ 0 . Note that the relative dimension of B L , U / X L , U B_{L,U}/X_{L,U} B L , U / X L , U is g : = m [ F : Q ] + 1 g:=m[F:\mathbb{Q}]+1 g := m [ F : Q ] + 1 .
Definition 3.17. Define two maps in the category Mot ( X L , U , E λ ) \operatorname{Mot}(X_{L,U},E^\lambda) Mot ( X L , U , E λ ) : (1) the straightening map
st L , U λ , + : M L , U , λ → h 2 ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q E λ \operatorname{st}_{L,U}^{\lambda,+}: M_{L,U,\lambda} \to h^{2|b|}\left(B_{L,U}^{|b|}/X_{L,U},\mathbb{Q}\right)\langle|b| \rangle\otimes_{\mathbb{Q}} E^{\lambda} st L , U λ , + : M L , U , λ → h 2∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q E λ
that is induced from the above discussion, and
(2) the complimentary straightening map
st L , U λ , − : = ( ( λ 0 × λ − 1 ) ∣ b ∣ ) ∗ ∘ F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ ⟨ ∣ b ∣ ⟩ ∘ st L , U λ , + : M L , U , λ → h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ( g − 1 ) ∣ b ∣ ⟩ ⊗ Q E λ , \operatorname{st}_{L,U}^{\lambda,-}:=\left((\lambda_0\times\lambda^{-1})^{|b|}\right)^*\circ F_{B_{L,U}^{|b|}/X_{L,U}}^{2(g-1)|b|}\langle|b|\rangle\circ\operatorname{st}_{L,U}^{\lambda,+}:M_{L,U,\lambda}\to h^{2(g-1)|b|}\left(B_{L,U}^{|b|}/X_{L,U},\mathbb{Q}\right)\langle(g-1)|b|\rangle\otimes_{\mathbb{Q}}E^{\lambda}, st L , U λ , − := ( ( λ 0 × λ − 1 ) ∣ b ∣ ) ∗ ∘ F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ ⟨ ∣ b ∣ ⟩ ∘ st L , U λ , + : M L , U , λ → h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨( g − 1 ) ∣ b ∣ ⟩ ⊗ Q E λ ,
where F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ F_{B_{L,U}^{|b|}/X_{L,U}}^{2(g-1)|b|} F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ stands for the relative Fourier transform (A.2).
Lemma 3.18. Let the notation be as above. We have
π L , U ∗ e L , λ = 1 ∣ Ξ U ∣ ⋅ ( e B ∣ b ∣ ⊗ Q E λ ) ∘ ( st L , U λ , + ⊗ st L , U λ ˉ , − ) ∘ ( π L , U ⊗ π L , U ) ∗ , \pi_{L,U}^{*}e_{L,\lambda}=\frac{1}{|\Xi_U|}\cdot\left(e_{B}^{|b|}\otimes_{\mathbb{Q}}E^{\lambda}\right)\circ\left(\operatorname{st}_{L,U}^{\lambda,+}\otimes\operatorname{st}_{L,U}^{\bar{\lambda},-}\right)\circ\left(\pi_{L,U}\otimes\pi_{L,U}\right)^*, π L , U ∗ e L , λ = ∣ Ξ U ∣ 1 ⋅ ( e B ∣ b ∣ ⊗ Q E λ ) ∘ ( st L , U λ , + ⊗ st L , U λ ˉ , − ) ∘ ( π L , U ⊗ π L , U ) ∗ , where
e B ∣ b ∣ : h 2 ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ( g − 1 ) ∣ b ∣ ⟩ → 1 X L , U , Q e_{B}^{|b|}:h^{2|b|}\left(B_{L,U}^{|b|}/X_{L,U},\mathbb{Q}\right)\langle|b|\rangle\otimes_{\mathbb{Q}}h^{2(g-1)|b|}\left(B_{L,U}^{|b|}/X_{L,U},\mathbb{Q}\right)\langle(g-1)|b|\rangle\to\mathbf{1}_{X_{L,U},\mathbb{Q}} e B ∣ b ∣ : h 2∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨( g − 1 ) ∣ b ∣ ⟩ → 1 X L , U , Q denotes the Poincaré duality map.
Proof. This follows from the construction and the simple observation that e L , U , λ ∘ ( π L , U ⊗ π L , U ) ∗ = ∣ Ξ U ∣ ⋅ π L , U ∗ e L , λ e_{L,U,\lambda}\circ(\pi_{L,U}\otimes\pi_{L,U})^*=|\Xi_U|\cdot\pi_{L,U}^*e_{L,\lambda} e L , U , λ ∘ ( π L , U ⊗ π L , U ) ∗ = ∣ Ξ U ∣ ⋅ π L , U ∗ e L , λ . □
Remark 3.19. If one forgets about the descent from X L , U X_{L,U} X L , U to X L X_L X L , then the construction of the standard and the automorphic (relative) motives can be carried over to a much more general context: Let S S S be a regular Noetherian scheme on which at least one rational prime is invertible, and consider two pairs ( A 0 , λ 0 ) (A_0,\lambda_0) ( A 0 , λ 0 ) and ( A , λ ) (A,\lambda) ( A , λ ) in which
A 0 A_0 A 0 is an object of AV ( S , E 0 ) \operatorname{AV}(S,E_0) AV ( S , E 0 ) of relative dimension one, and λ 0 : A 0 → A 0 ∨ \lambda_0:A_0\to A_0^{\vee} λ 0 : A 0 → A 0 ∨ is a polarization that turns the complex conjugation of E 0 E_0 E 0 into the Rosati involution;
A A A is an object of AV ( S , E ) \operatorname{AV}(S,E) AV ( S , E ) of relative dimension m [ F : Q ] m[F:\mathbb{Q}] m [ F : Q ] , and λ : A → A ∨ \lambda:A\to A^{\vee} λ : A → A ∨ is a polarization that turns the complex conjugation of E E E into the Rosati involution.
Put
M : = Hom E 0 ( h 1 ( A 0 / S , Q ) , h 1 ( A / S , Q ) ) ∈ Mot ( S , E ) , M:=\operatorname{Hom}_{E_0}\left(h^1(A_0/S,\mathbb{Q}),h^1(A/S,\mathbb{Q})\right)\in\operatorname{Mot}(S,E), M := Hom E 0 ( h 1 ( A 0 / S , Q ) , h 1 ( A / S , Q ) ) ∈ Mot ( S , E ) ,
together with the canonical evaluation map M ⊗ E M ˉ → 1 S , E M\otimes_E\bar{M}\to\mathbf{1}_{S,E} M ⊗ E M ˉ → 1 S , E similar to (10) . Applying the process of obtaining M L , λ M_{L,\lambda} M L , λ from M L M_L M L in § 3.4 to M M M , we obtain an object
M λ ∈ Mot ( S , E λ ) , M_{\lambda}\in\operatorname{Mot}(S,E^{\lambda}), M λ ∈ Mot ( S , E λ ) ,
which we call the abstract automorphic (relative) motive over S S S of weight λ \lambda λ associated with ( A 0 , λ 0 ) (A_0,\lambda_0) ( A 0 , λ 0 ) and ( A , λ ) (A,\lambda) ( A , λ ) .
Put B : = A 0 × S A ∨ B:=A_0\times_S A^{\vee} B := A 0 × S A ∨ and
B n : = B × S ⋯ × S B ⏟ n B^n:=\underbrace{B\times_S\cdots\times_S B}_{n} B n := n B × S ⋯ × S B
for n ≥ 0 n\geq0 n ≥ 0 . Note that the relative dimension of B / S B/S B / S is g : = m [ F : Q ] + 1 g:=m[F:\mathbb{Q}]+1 g := m [ F : Q ] + 1 . Similar to Definition 3.17, we have two maps
st λ , + : M λ → h 2 ∣ b ∣ ( B ∣ b ∣ / S , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q E λ , \operatorname{st}^{\lambda,+}:M_{\lambda}\to h^{2|b|}\left(B^{|b|}/S,\mathbb{Q}\right)\langle|b|\rangle\otimes_{\mathbb{Q}}E^{\lambda}, st λ , + : M λ → h 2∣ b ∣ ( B ∣ b ∣ / S , Q ) ⟨ ∣ b ∣ ⟩ ⊗ Q E λ ,
st λ , − : M λ → h 2 ( g − 1 ) ∣ b ∣ ( B ∣ b ∣ / S , Q ) ⟨ ( g − 1 ) ∣ b ∣ ⟩ ⊗ Q E λ \operatorname{st}^{\lambda,-}:M_{\lambda}\to h^{2(g-1)|b|}\left(B^{|b|}/S,\mathbb{Q}\right)\langle(g-1)|b|\rangle\otimes_{\mathbb{Q}}E^{\lambda} st λ , − : M λ → h 2 ( g − 1 ) ∣ b ∣ ( B ∣ b ∣ / S , Q ) ⟨( g − 1 ) ∣ b ∣ ⟩ ⊗ Q E λ
in the category Mot ( S , E λ ) \operatorname{Mot}(S,E^{\lambda}) Mot ( S , E λ ) , which further induce two maps
st λ , + : H 2 r ( S , M λ ( r ) ) → CH r + ∣ b ∣ ( B ∣ b ∣ ) ⊗ E λ , \operatorname{st}^{\lambda,+}:H^{2r}(S,M_{\lambda}(r))\to\operatorname{CH}^{r+|b|}(B^{|b|})\otimes E^{\lambda}, st λ , + : H 2 r ( S , M λ ( r )) → CH r + ∣ b ∣ ( B ∣ b ∣ ) ⊗ E λ ,
st λ , − : H 2 r ( S , M λ ( r ) ) → CH r + ( g − 1 ) ∣ b ∣ ( B ∣ b ∣ ) ⊗ E λ \operatorname{st}^{\lambda,-}:H^{2r}(S,M_{\lambda}(r))\to\operatorname{CH}^{r+(g-1)|b|}(B^{|b|})\otimes E^{\lambda} st λ , − : H 2 r ( S , M λ ( r )) → CH r + ( g − 1 ) ∣ b ∣ ( B ∣ b ∣ ) ⊗ E λ
on the level of motivic cohomology for every r ≥ 0 r\geq0 r ≥ 0 by Remark A.2(1).
Étale realizations of automorphic relative motives
Fix a rational prime ℓ \ell ℓ . In this section, we compute the image of M ∙ , λ \mathbf{M}_{\bullet,\lambda} M ∙ , λ under the ℓ \ell ℓ -adic étale realization functor r ℓ : D ( X ∙ , E λ ) → D e ˊ t ( X ∙ , E ℓ λ ) r_{\ell}: \mathrm{D}(X_{\bullet}, E^{\lambda}) \to\mathrm{D}_{\mathrm{\acute{e}t}}(X_{\bullet}, E^{\lambda}_{\ell}) r ℓ : D ( X ∙ , E λ ) → D e ˊ t ( X ∙ , E ℓ λ ) (Notation A.3).
Take an object L ∈ H k ( H ( A F ∞ ) ) L \in\mathrm{Hk}(H(\mathbb{A}_{F}^{\infty})) L ∈ Hk ( H ( A F ∞ )) . We first consider the étale realization of the standard motive M L \mathbf{M}_{L} M L .
Choose an object U ∈ H k ( T ( A ∞ ) ) U \in\mathrm{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) . We have ℓ \ell ℓ -adic complexes
h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) ∈ D e ˊ t ( X L , U , E 0 , ℓ ) , h^{\mathrm{\acute{e}t}}_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}_{\ell}) \in\mathrm{D}_{\mathrm{\acute{e}t}}(X_{L,U}, E_{0,\ell}), h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) ∈ D e ˊ t ( X L , U , E 0 , ℓ ) ,
h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ∈ D e ˊ t ( X L , U , E ℓ ) h^{\mathrm{\acute{e}t}}_{1}(A_{L,U}/X_{L,U}, \mathbb{Q}_{\ell}) \in\mathrm{D}_{\mathrm{\acute{e}t}}(X_{L,U}, E_{\ell}) h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ∈ D e ˊ t ( X L , U , E ℓ )
from Remark A.8(1). Put
M L , U , ℓ : = Hom ‾ E 0 , ℓ ( h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ) ∈ D e ˊ t ( X L , U , E ℓ ) , \mathbf{M}_{L,U,\ell} := \underline{\operatorname{Hom}}_{E_{0,\ell}}\left(h^{\mathrm{\acute{e}t}}_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}_{\ell}), h^{\mathrm{\acute{e}t}}_{1}(A_{L,U}/X_{L,U}, \mathbb{Q}_{\ell})\right) \in\mathrm{D}_{\mathrm{\acute{e}t}}(X_{L,U}, E_{\ell}), M L , U , ℓ := Hom E 0 , ℓ ( h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ) ∈ D e ˊ t ( X L , U , E ℓ ) ,
which is indeed an object in D e ˊ t ∅ ( X L , U , E ℓ ) \mathrm{D}^{\varnothing}_{\mathrm{\acute{e}t}}(X_{L,U}, E_{\ell}) D e ˊ t ∅ ( X L , U , E ℓ ) , that is, an ℓ \ell ℓ -adic sheaf (as the above internal hom is concentrated in degree zero). Similar to the case of M L , U \mathbf{M}_{L,U} M L , U , we have an object
M L , U , ℓ / / Ξ U ∈ D e ˊ t ∅ ( X L , U / / Ξ U , E ℓ ) , \mathbf{M}_{L,U,\ell}/\!\!/\Xi_{U} \in\mathrm{D}^{\varnothing}_{\mathrm{\acute{e}t}}(X_{L,U}/\!\!/\Xi_{U}, E_{\ell}), M L , U , ℓ / / Ξ U ∈ D e ˊ t ∅ ( X L , U / / Ξ U , E ℓ ) ,
where the target denotes the category of Cartesian functors Ξ U → D ℓ e ˊ t ( S c h / E s m . q p r o j , E ℓ ) \Xi_{U} \to\mathrm{D}^{\mathrm{\acute{e}t}}_{\ell}(\mathrm{Sch}^{\mathrm{sm.qproj}}_{/E}, E_{\ell}) Ξ U → D ℓ e ˊ t ( Sch / E sm.qproj , E ℓ ) above X L , U / / Ξ U X_{L,U}/\!\!/\Xi_{U} X L , U / / Ξ U . Then by the usual descent in abelian category, we obtain an object
M L , ℓ : = M L , U , ℓ / / Ξ U ∈ D e ˊ t ∅ ( X L , U / / Ξ U , E ℓ ) = D e ˊ t ∅ ( X L , E ℓ ) , \mathbf{M}_{L,\ell} := \mathbf{M}_{L,U,\ell}/\!\!/\Xi_{U} \in\mathrm{D}^{\varnothing}_{\mathrm{\acute{e}t}}(X_{L,U}/\!\!/\Xi_{U}, E_{\ell}) = \mathrm{D}^{\varnothing}_{\mathrm{\acute{e}t}}(X_{L}, E_{\ell}), M L , ℓ := M L , U , ℓ / / Ξ U ∈ D e ˊ t ∅ ( X L , U / / Ξ U , E ℓ ) = D e ˊ t ∅ ( X L , E ℓ ) ,
which is independent of the choice of U U U . Indeed, M L , ℓ \mathbf{M}_{L,\ell} M L , ℓ is a local system over X L X_{L} X L of rank m m m with coefficients in E ℓ E_{\ell} E ℓ . Similar to (3.4), we have an evaluation map
M L , ℓ ⊗ E ℓ M ‾ L , ℓ ⟶ 1 X L , E ℓ e ˊ t . (23) \mathbf{M}_{L,\ell} \otimes_{E_{\ell}} \overline{\mathbf{M}}_{L,\ell} \longrightarrow\mathbbm{1}^{\mathrm{\acute{e}t}}_{X_{L},E_{\ell}}. \tag*{(23)} M L , ℓ ⊗ E ℓ M L , ℓ ⟶ 1 X L , E ℓ e ˊ t . (23)
where 1 X L , E ℓ e ˊ t \mathbbm{1}^{\mathrm{\acute{e}t}}_{X_{L},E_{\ell}} 1 X L , E ℓ e ˊ t denotes the unit object of D e ˊ t ( X L , E ℓ ) \mathrm{D}_{\mathrm{\acute{e}t}}(X_{L}, E_{\ell}) D e ˊ t ( X L , E ℓ ) .
Lemma 3.20. The étale realization r ℓ M L r_{\ell}\mathbf{M}_{L} r ℓ M L is canonically isomorphic to M L , ℓ \mathbf{M}_{L,\ell} M L , ℓ .
Proof. It suffices to construct, for U ∈ H k ( T ( A ∞ ) ) U \in\mathrm{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) , a canonical isomorphism r ℓ M L , U ≃ M L , U , ℓ r_{\ell}\mathbf{M}_{L,U} \simeq\mathbf{M}_{L,U,\ell} r ℓ M L , U ≃ M L , U , ℓ that is Ξ U \Xi_{U} Ξ U -equivariant. From Remark A.8(1), we have
r ℓ h 1 ( A 0 , L , U / X L , U , Q ) ≃ h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , r ℓ h 1 ( A L , U / X L , U , Q ) ≃ h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) r_{\ell}h_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}) \simeq h^{\mathrm{\acute{e}t}}_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}_{\ell}), \qquad r_{\ell}h_{1}(A_{L,U}/X_{L,U}, \mathbb{Q}) \simeq h^{\mathrm{\acute{e}t}}_{1}(A_{L,U}/X_{L,U}, \mathbb{Q}_{\ell}) r ℓ h 1 ( A 0 , L , U / X L , U , Q ) ≃ h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , r ℓ h 1 ( A L , U / X L , U , Q ) ≃ h 1 e ˊ t ( A L , U / X L , U , Q ℓ )
canonically. Since h 1 ( A 0 , L , U / X L , U , Q ) h_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}) h 1 ( A 0 , L , U / X L , U , Q ) is compact, we have a canonical isomorphism
r ℓ M L , U = r ℓ Hom ‾ E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) ≃ Hom ‾ E 0 , ℓ ( h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ) = M L , U , ℓ \begin{aligned}
r_{\ell}\mathbf{M}_{L,U} = r_{\ell}\underline{\operatorname{Hom}}_{E_{0}}\left(h_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}), h_{1}(A_{L,U}/X_{L,U}, \mathbb{Q})\right) \\
&\simeq\underline{\operatorname{Hom}}_{E_{0,\ell}}\left(h^{\mathrm{\acute{e}t}}_{1}(A_{0,L,U}/X_{L,U}, \mathbb{Q}_{\ell}), h^{\mathrm{\acute{e}t}}_{1}(A_{L,U}/X_{L,U}, \mathbb{Q}_{\ell})\right) = \mathbf{M}_{L,U,\ell}
\end{aligned} r ℓ M L , U = r ℓ Hom E 0 ( h 1 ( A 0 , L , U / X L , U , Q ) , h 1 ( A L , U / X L , U , Q ) ) ≃ Hom E 0 , ℓ ( h 1 e ˊ t ( A 0 , L , U / X L , U , Q ℓ ) , h 1 e ˊ t ( A L , U / X L , U , Q ℓ ) ) = M L , U , ℓ
by [2 Théorème 6.6] . It is clear that the above isomorphism is Ξ U \Xi_{U} Ξ U -equivariant, which implies the lemma. □
Take an element v ∈ V ( ∞ ) v \in\mathcal{V}^{(\infty)} v ∈ V ( ∞ ) . Put
T b v ( M L , ℓ ) : = ⨂ ν ( u ) = v M L , ℓ [ u ] \mathbf{T}_{b_{v}}(\mathbf{M}_{L,\ell}) := \bigotimes_{\nu(u)=v} \mathbf{M}_{L,\ell}[u] T b v ( M L , ℓ ) := ν ( u ) = v ⨂ M L , ℓ [ u ]
(tensor over E ℓ E_{\ell} E ℓ ), which is an object of D e ˊ t ∅ ( X L , E ℓ ) \mathrm{D}^{\varnothing}_{\mathrm{\acute{e}t}}(X_{L}, E_{\ell}) D e ˊ t ∅ ( X L , E ℓ ) and admits a natural action by S b v \mathfrak{S}_{b_{v}} S b v by switching factors. The evaluation map (3.19) induces an evaluation map
⨂ ν ( u ) = v M L , ℓ [ u ] ⟶ 1 X L , E ℓ e ˊ t , \bigotimes_{\nu(u)=v} \mathbf{M}_{L,\ell}[u] \longrightarrow\mathbbm{1}^{\mathrm{\acute{e}t}}_{X_{L},E_{\ell}}, ν ( u ) = v ⨂ M L , ℓ [ u ] ⟶ 1 X L , E ℓ e ˊ t ,
(tensor over E ℓ E_{\ell} E ℓ ) which further induces
T b v ( M L , ℓ ) ⊗ E ℓ T b v ( M L , ℓ ) ⟶ 1 X L , E ℓ e ˊ t , \mathbf{T}_{b_{v}}(\mathbf{M}_{L,\ell}) \otimes_{E_{\ell}} \mathbf{T}_{b_{v}}(\mathbf{M}_{L,\ell}) \longrightarrow\mathbbm{1}^{\mathrm{\acute{e}t}}_{X_{L},E_{\ell}}, T b v ( M L , ℓ ) ⊗ E ℓ T b v ( M L , ℓ ) ⟶ 1 X L , E ℓ e ˊ t ,
for every pair I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } I=(i_u)_{v(u)=v}\in\prod_{v(u)=v}\{1,\ldots,b_u\} I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } , a contraction map
C I , ℓ : T b v ( M L , ℓ ) ⟶ T b v − 1 ( M L , ℓ ) C_{I,\ell}:T_{b_v}(M_{L,\ell})\longrightarrow T_{b_v-1}(M_{L,\ell}) C I , ℓ : T b v ( M L , ℓ ) ⟶ T b v − 1 ( M L , ℓ )
by evaluating i u i_u i u -th factors of T b u ( M L , ℓ [ u ] ) T^{b_u}(M_{L,\ell}[u]) T b u ( M L , ℓ [ u ]) for v ( u ) = v v(u)=v v ( u ) = v (set T b v − 1 ( M L , ℓ ) = 0 T_{b_v-1}(M_{L,\ell})=0 T b v − 1 ( M L , ℓ ) = 0 if b u − 1 < 0 b_u-1<0 b u − 1 < 0 for some u u u ). It follows from Lemma 3.20 that we have a canonical isomorphism
r ℓ C I ≃ C I , ℓ r_\ell C_I\simeq C_{I,\ell} r ℓ C I ≃ C I , ℓ
for every pair I ∈ ∏ v ( u ) = v { 1 , … , b u } I\in\prod_{v(u)=v}\{1,\ldots,b_u\} I ∈ ∏ v ( u ) = v { 1 , … , b u } .
Definition 3.21. Define T [ b v ] ( M L , ℓ ) T_{[b_v]}(M_{L,\ell}) T [ b v ] ( M L , ℓ ) to be the kernel of the natural map
⨁ I C I , ℓ : T b v ( M L , ℓ ) ⟶ ⨁ ∏ v ( u ) = v { 1 , … , b u } T b v − 1 ( M L , ℓ ) \bigoplus_I C_{I,\ell}:T_{b_v}(M_{L,\ell})\longrightarrow\bigoplus_{\prod_{v(u)=v}\{1,\ldots,b_u\}}T_{b_v-1}(M_{L,\ell}) I ⨁ C I , ℓ : T b v ( M L , ℓ ) ⟶ ∏ v ( u ) = v { 1 , … , b u } ⨁ T b v − 1 ( M L , ℓ ) in the abelian category D e ˊ t ∨ ( X L , E ℓ ) D^\vee_{\mathrm{\acute et}}(X_L,E_\ell) D e ˊ t ∨ ( X L , E ℓ ) , which admits a natural action by S b v \mathfrak{S}_{b_v} S b v . Similar to Remark 2.4(2), T [ b v ] ( M L , ℓ ) T_{[b_v]}(M_{L,\ell}) T [ b v ] ( M L , ℓ ) is a direct summand of T b v ( M L , ℓ ) T_{b_v}(M_{L,\ell}) T b v ( M L , ℓ ) , called the harmonic direct summand .
By Lemma 3.14, we have a canonical isomorphism
r ℓ T [ b v ] ( M L ) ≃ T [ b v ] ( M L , ℓ ) . r_\ell T_{[b_v]}(M_L)\simeq T_{[b_v]}(M_{L,\ell}). r ℓ T [ b v ] ( M L ) ≃ T [ b v ] ( M L , ℓ ) .
By the similar process of constructing S [ λ v ] ( M L ) S_{[\lambda_v]}(M_L) S [ λ v ] ( M L ) , S [ λ ] ( M L ) S_{[\lambda]}(M_L) S [ λ ] ( M L ) , and M L , λ M_{L,\lambda} M L , λ from T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) , we obtain the ℓ \ell ℓ -adic analogues S [ λ v ] ( M L , ℓ ) S_{[\lambda_v]}(M_{L,\ell}) S [ λ v ] ( M L , ℓ ) , S [ λ ] ( M L , ℓ ) ∈ D e ˊ t ∨ ( X L , E ℓ ) S_{[\lambda]}(M_{L,\ell})\in D^\vee_{\mathrm{\acute et}}(X_L,E_\ell) S [ λ ] ( M L , ℓ ) ∈ D e ˊ t ∨ ( X L , E ℓ ) , and M L , ℓ , λ ∈ D e ˊ t ∨ ( X L , E ℓ λ ) M_{L,\ell,\lambda}\in D^\vee_{\mathrm{\acute et}}(X_L,E^\lambda_\ell) M L , ℓ , λ ∈ D e ˊ t ∨ ( X L , E ℓ λ ) from T [ b v ] ( M L , ℓ ) T_{[b_v]}(M_{L,\ell}) T [ b v ] ( M L , ℓ ) , with canonical isomorphisms:
r ℓ S [ λ v ] ( M L ) ≃ S [ λ v ] ( M L , ℓ ) , r ℓ S [ λ ] ( M L ) ≃ S [ λ ] ( M L , ℓ ) , r ℓ M L , λ ≃ M L , ℓ , λ . (24) r_\ell S_{[\lambda_v]}(M_L)\simeq S_{[\lambda_v]}(M_{L,\ell}),\qquad r_\ell S_{[\lambda]}(M_L)\simeq S_{[\lambda]}(M_{L,\ell}),\qquad r_\ell M_{L,\lambda}\simeq M_{L,\ell,\lambda}. \tag*{(24)} r ℓ S [ λ v ] ( M L ) ≃ S [ λ v ] ( M L , ℓ ) , r ℓ S [ λ ] ( M L ) ≃ S [ λ ] ( M L , ℓ ) , r ℓ M L , λ ≃ M L , ℓ , λ . (24)
Remark 3.22. The evaluation map e L , λ e_{L,\lambda} e L , λ (21) induces an evaluation map
e L , ℓ , λ : M L , ℓ , λ ⊗ E ℓ λ M L , ℓ , λ ⟶ 1 X L , E ℓ λ e ˊ t (25) e_{L,\ell,\lambda}:M_{L,\ell,\lambda}\otimes_{E^\lambda_\ell}M_{L,\ell,\lambda}\longrightarrow\mathbbm{1}^{\mathrm{\acute et}}_{X_L,E^\lambda_\ell} \tag*{(25)} e L , ℓ , λ : M L , ℓ , λ ⊗ E ℓ λ M L , ℓ , λ ⟶ 1 X L , E ℓ λ e ˊ t (25)
in D e ˊ t ∨ ( Sch m . q p r o j / E ℓ λ ) D^\vee_{\mathrm{\acute et}}(\operatorname{Sch}^{\mathrm{m.qproj}}/E^\lambda_\ell) D e ˊ t ∨ ( Sch m.qproj / E ℓ λ ) for every L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}^{\infty}_F)) L ∈ Hk ( H ( A F ∞ )) .
To end this section, we give an alternative description of M L , ℓ , λ M_{L,\ell,\lambda} M L , ℓ , λ . Choose an O E , ℓ O_{E,\ell} O E , ℓ -lattice Λ ℓ \Lambda_\ell Λ ℓ of V ℓ V_\ell V ℓ such that L ℓ L_\ell L ℓ is contained in the stabilizer of Λ ℓ \Lambda_\ell Λ ℓ in H ( F ℓ ) H(F_\ell) H ( F ℓ ) . We review below the well-known construction of constructing a O E , ℓ O_{E,\ell} O E , ℓ -linear local system over X L X_L X L from Λ ℓ \Lambda_\ell Λ ℓ .
Construction 3.23. For every integer n ≥ 1 n\geq1 n ≥ 1 , denote by L n ⊆ L L_n\subseteq L L n ⊆ L the subgroup that induces the trivial action on Λ ℓ / ℓ n Λ ℓ \Lambda_\ell/\ell^n\Lambda_\ell Λ ℓ / ℓ n Λ ℓ (via ℓ \ell ℓ -adic components), so that we have an O E , ℓ O_{E,\ell} O E , ℓ -linear sheaf
Λ ℓ , n , L : = ( X L n × Λ ℓ / ℓ n Λ ℓ ) / ( L / L n ) \Lambda_{\ell,n,L}:=(X_{L_n}\times\Lambda_\ell/\ell^n\Lambda_\ell)/(L/L_n) Λ ℓ , n , L := ( X L n × Λ ℓ / ℓ n Λ ℓ ) / ( L / L n )
over X L X_L X L , in which the action is given by ( x , v ) . h = ( x h , h − 1 v ) (x,v).h=(xh,h^{-1}v) ( x , v ) . h = ( x h , h − 1 v ) . Then we obtain an obvious projective system ( Λ ℓ , n , L ) n (\Lambda_{\ell,n,L})_n ( Λ ℓ , n , L ) n of O E , ℓ O_{E,\ell} O E , ℓ -linear sheaves over X L X_L X L ; and we put
Λ ℓ , L : = lim ⟵ n ≥ 1 Λ ℓ , n , L , \Lambda_{\ell,L}:=\lim_{\substack{\longleftarrow\\n\geq1}}\Lambda_{\ell,n,L}, Λ ℓ , L := ⟵ n ≥ 1 lim Λ ℓ , n , L ,
which is an O E , ℓ O_{E,\ell} O E , ℓ -linear local system over X L X_L X L .
It is well-known that the E ℓ E_\ell E ℓ -linear local V ℓ , L : = Λ ℓ , L ⊗ Z ℓ Q ℓ V_{\ell,L}:=\Lambda_{\ell,L}\otimes_{\mathbb{Z}_\ell}\mathbb{Q}_\ell V ℓ , L := Λ ℓ , L ⊗ Z ℓ Q ℓ on X L X_L X L does not depend on the choice of Λ ℓ \Lambda_\ell Λ ℓ , and that the hermitian level structures at ℓ \ell ℓ -adic places in Definition 3.3 induce a canonical isomorphism M L , ℓ ≃ V ℓ , L M_{L,\ell}\simeq V_{\ell,L} M L , ℓ ≃ V ℓ , L . Repeating the same process for M L , ℓ M_{L,\ell} M L , ℓ , we obtain E ℓ E_\ell E ℓ -linear local systems S [ λ v ] ( V ℓ , L ) S_{[\lambda_v]}(V_{\ell,L}) S [ λ v ] ( V ℓ , L ) for v ∈ V F ( ∞ ) v\in\mathcal{V}^{(\infty)}_F v ∈ V F ( ∞ ) and S [ λ ] ( V ℓ , L ) S_{[\lambda]}(V_{\ell,L}) S [ λ ] ( V ℓ , L ) , and finally the E ℓ λ E^\lambda_\ell E ℓ λ -linear local system V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ on X L X_L X L , with canonical isomorphisms:
S [ λ v ] ( M L , ℓ ) ≃ S [ λ v ] ( V ℓ , L ) , S [ λ ] ( M L , ℓ ) ≃ S [ λ ] ( V ℓ , L ) , M L , ℓ , λ ≃ V ℓ , L , λ . (26) S_{[\lambda_v]}(M_{L,\ell})\simeq S_{[\lambda_v]}(V_{\ell,L}),\qquad S_{[\lambda]}(M_{L,\ell})\simeq S_{[\lambda]}(V_{\ell,L}),\qquad M_{L,\ell,\lambda}\simeq V_{\ell,L,\lambda}. \tag*{(26)} S [ λ v ] ( M L , ℓ ) ≃ S [ λ v ] ( V ℓ , L ) , S [ λ ] ( M L , ℓ ) ≃ S [ λ ] ( V ℓ , L ) , M L , ℓ , λ ≃ V ℓ , L , λ . (26)
Remark 3.24. When ℓ \ell ℓ avoids certain divisibility, the E ℓ λ E_\ell^\lambda E ℓ λ -linear local system V ℓ , λ V_{\ell,\lambda} V ℓ , λ has an integral model that can be constructed from an O E λ , ℓ \mathcal{O}_{E^\lambda,\ell} O E λ , ℓ -lattice via the similar process in Construction 3.23. Indeed, suppose that ℓ ∤ D E B λ [ E : E λ ] \ell\nmid D_EB_\lambda[E:E^\lambda] ℓ ∤ D E B λ [ E : E λ ] (see (2) for B λ B_\lambda B λ ). Furthermore, suppose that L ℓ L_\ell L ℓ is contained in the stabilizer of a self-dual lattice (Definition 2.9) Λ ℓ \Lambda_\ell Λ ℓ of V ℓ V_\ell V ℓ . Applying Construction 2.8(2) to R = Z ℓ R=\mathbb{Z}_\ell R = Z ℓ and V = Λ ℓ V=\Lambda_\ell V = Λ ℓ , we obtain an O E λ , ℓ \mathcal{O}_{E^\lambda,\ell} O E λ , ℓ -lattice Λ ℓ , λ : = ( Λ ℓ ) λ \Lambda_{\ell,\lambda}:=(\Lambda_\ell)_\lambda Λ ℓ , λ := ( Λ ℓ ) λ of V ℓ , λ : = ( V ℓ ) λ V_{\ell,\lambda}:=(V_\ell)_\lambda V ℓ , λ := ( V ℓ ) λ . Applying the similar process in Construction 3.23, we obtain an O E λ , ℓ \mathcal{O}_{E^\lambda,\ell} O E λ , ℓ -linear local system Λ ℓ , λ , L \Lambda_{\ell,\lambda,L} Λ ℓ , λ , L from Λ ℓ , λ \Lambda_{\ell,\lambda} Λ ℓ , λ , so that we have a canonical isomorphism
Λ ℓ , λ , L ⊗ Z ℓ Q ℓ ≃ V ℓ , L , λ (27) \Lambda_{\ell,\lambda,L}\otimes_{\mathbb{Z}_\ell}\mathbb{Q}_\ell\simeq V_{\ell,L,\lambda} \tag*{(27)} Λ ℓ , λ , L ⊗ Z ℓ Q ℓ ≃ V ℓ , L , λ (27)
of E ℓ λ E_\ell^\lambda E ℓ λ -linear local systems over X L X_L X L .
Hodge realizations of automorphic relative motives
Fix an element u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . In this section, we compute the image of M ∙ , λ M_{\bullet,\lambda} M ∙ , λ under the u u u -Hodge realization functor r u : D ( X ∙ , E λ ) → D h s ( X ∙ u , E ∞ λ ) r^u:D(X_\bullet,E^\lambda)\to D_{\mathrm{hs}}(X_\bullet^u,E^\lambda_\infty) r u : D ( X ∙ , E λ ) → D hs ( X ∙ u , E ∞ λ ) (Notation A.4).
Take an object L ∈ Hk ( H ( A ∞ ) ) L\in\operatorname{Hk}(H(A^\infty)) L ∈ Hk ( H ( A ∞ )) . We first consider the u u u -Hodge realization of the standard motive M L M_L M L . Choose an object U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(A^\infty)) U ∈ Hk ( T ( A ∞ )) . We have complexes of relative Hodge structures
h 1 h s ( A 0 , L , U u / X L , U u , R ) ∈ D h s ( X L , U u , E 0 , ∞ ) , h^{\mathrm{hs}}_1(A^u_{0,L,U}/X^u_{L,U},\mathbb{R})\in D_{\mathrm{hs}}(X^u_{L,U},E_{0,\infty}), h 1 hs ( A 0 , L , U u / X L , U u , R ) ∈ D hs ( X L , U u , E 0 , ∞ ) ,
h 1 h s ( A L , U u / X L , U u , R ) ∈ D h s ( X L , U u , E ∞ ) h^{\mathrm{hs}}_1(A^u_{L,U}/X^u_{L,U},\mathbb{R})\in D_{\mathrm{hs}}(X^u_{L,U},E_\infty) h 1 hs ( A L , U u / X L , U u , R ) ∈ D hs ( X L , U u , E ∞ )
from Remark A.8(2). Put
M L , U u : = Hom E 0 , ∞ ( h 1 h s ( A 0 , L , U u / X L , U u , R ) , h 1 h s ( A L , U u / X L , U u , R ) ) ∈ D h s ( X L , U u , E ∞ ) , M^u_{L,U}:=\operatorname{Hom}_{E_{0,\infty}}\left(h^{\mathrm{hs}}_1(A^u_{0,L,U}/X^u_{L,U},\mathbb{R}),h^{\mathrm{hs}}_1(A^u_{L,U}/X^u_{L,U},\mathbb{R})\right)\in D_{\mathrm{hs}}(X^u_{L,U},E_\infty), M L , U u := Hom E 0 , ∞ ( h 1 hs ( A 0 , L , U u / X L , U u , R ) , h 1 hs ( A L , U u / X L , U u , R ) ) ∈ D hs ( X L , U u , E ∞ ) ,
which is indeed an object in D h s ♡ ( X L , U u , E ∞ ) D^\heartsuit_{\mathrm{hs}}(X^u_{L,U},E_\infty) D hs ♡ ( X L , U u , E ∞ ) , that is, a relative Hodge structure with coefficients in E ∞ E_\infty E ∞ (as the above Hom E 0 , ∞ \operatorname{Hom}_{E_{0,\infty}} Hom E 0 , ∞ is concentrated in degree zero). Similar to the case of M L , U M_{L,U} M L , U , we have an object
M L , U / Ξ U u ∈ D h s ♡ ( X L , U / Ξ U u , E ∞ ) , M^u_{L,U/\Xi_U}\in D^\heartsuit_{\mathrm{hs}}(X^u_{L,U/\Xi_U},E_\infty), M L , U / Ξ U u ∈ D hs ♡ ( X L , U / Ξ U u , E ∞ ) ,
where the target denotes the category of Cartesian functors Ξ U → D h s ♡ ( Mfd , E ∞ ) \Xi_U\to D^\heartsuit_{\mathrm{hs}}(\operatorname{Mfd},E_\infty) Ξ U → D hs ♡ ( Mfd , E ∞ ) above X L , U / Ξ U u X^u_{L,U/\Xi_U} X L , U / Ξ U u . Then by the usual descent in abelian category, we obtain an object
M L u : = M L , U / Ξ U u ∈ D h s ♡ ( X L u / Ξ U , E ∞ ) = D h s ♡ ( X L u , E ∞ ) , M^u_L:=M^u_{L,U/\Xi_U}\in D^\heartsuit_{\mathrm{hs}}(X^u_L/\Xi_U,E_\infty)=D^\heartsuit_{\mathrm{hs}}(X^u_L,E_\infty), M L u := M L , U / Ξ U u ∈ D hs ♡ ( X L u / Ξ U , E ∞ ) = D hs ♡ ( X L u , E ∞ ) ,
which is independent of the choice of U U U . Indeed, M L u M^u_L M L u is a relative Hodge structure over X L u X^u_L X L u of rank m m m with coefficients in E ∞ E_\infty E ∞ . Similar to (8) , we have an evaluation map
M L u ⊗ E ∞ M L u ‾ ⟶ 1 X L u , E ∞ h s . (28) M^u_L\otimes_{E_\infty}\overline{M^u_L}\longrightarrow\mathbb{1}^{\mathrm{hs}}_{X^u_L,E_\infty}. \tag*{(28)} M L u ⊗ E ∞ M L u ⟶ 1 X L u , E ∞ hs . (28)
where 1 X L u , E ∞ h s \mathbb{1}^{\mathrm{hs}}_{X^u_L,E_\infty} 1 X L u , E ∞ hs denotes the unit object of D h s ( X L u , E ∞ ) D_{\mathrm{hs}}(X^u_L,E_\infty) D hs ( X L u , E ∞ ) . Using [35 ] , we obtain the following lemma similar to Lemma 3.20.
Lemma 3.25. The u u u -Hodge realization r u M L r^uM_L r u M L is canonically isomorphic to M L u M^u_L M L u .
Take an element v ∈ S F ( ∞ ) v\in\mathfrak{S}^{(\infty)}_F v ∈ S F ( ∞ ) . Put
T b v ( M L u ) : = ⨂ v ( u ) = v T v b u ( M L u [ u ] ) T_{b_v}(M^u_L):=\bigotimes_{v(u)=v}T^{b_u}_v(M^u_L[u]) T b v ( M L u ) := v ( u ) = v ⨂ T v b u ( M L u [ u ])
(tensor over E ∞ E_\infty E ∞ ), which is an object of D h s ♡ ( X L u , E ∞ ) D^\heartsuit_{\mathrm{hs}}(X^u_L,E_\infty) D hs ♡ ( X L u , E ∞ ) and admits a natural action by S b v \mathfrak{S}_{b_v} S b v by switching factors. The evaluation map (28) induces an evaluation map
⨂ v ( u ) = v M L u [ u ] ⟶ 1 X L u , E ∞ h s , \bigotimes_{v(u)=v}M^u_L[u]\longrightarrow\mathbb{1}^{\mathrm{hs}}_{X^u_L,E_\infty}, v ( u ) = v ⨂ M L u [ u ] ⟶ 1 X L u , E ∞ hs ,
(tensor over E ∞ E_\infty E ∞ ) which further induces
T b v ( M L u ) ⊗ E ∞ T b ‾ v ( M L u ) ⟶ 1 X L u , E ∞ h s ; T_{b_v}(M^u_L)\otimes_{E_\infty}T_{\overline{b}_v}(M^u_L)\longrightarrow\mathbb{1}^{\mathrm{hs}}_{X^u_L,E_\infty}; T b v ( M L u ) ⊗ E ∞ T b v ( M L u ) ⟶ 1 X L u , E ∞ hs ;
for every pair I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } I=(i_u)_{v(u)=v}\in\prod_{v(u)=v}\{1,\ldots,b_u\} I = ( i u ) v ( u ) = v ∈ ∏ v ( u ) = v { 1 , … , b u } , a contraction map
C I u : T b v ( M L u ) ⟶ T b v − 1 ( M L u ) C_I^u:T_{b_v}(M_L^u)\longrightarrow T_{b_v-1}(M_L^u) C I u : T b v ( M L u ) ⟶ T b v − 1 ( M L u )
by evaluating i u i_u i u -th factors of T b u ( M L u [ i u ] ) T^{b_u}(M_L^u[i_u]) T b u ( M L u [ i u ]) for v ( u ) = v v(u)=v v ( u ) = v (set T b v − 1 ( M L u ) = 0 T_{b_v-1}(M_L^u)=0 T b v − 1 ( M L u ) = 0 if b u − 1 < 0 b_u-1<0 b u − 1 < 0 for some u u u ).
It follows from Lemma 3.25 that we have a canonical isomorphism
r C I ≃ C I u {}^rC_I\simeq C_I^u r C I ≃ C I u
for every pair I ∈ ∏ v ( u ) = v { 1 , … , b u } I\in\prod_{v(u)=v}\{1,\ldots,b_u\} I ∈ ∏ v ( u ) = v { 1 , … , b u } .
Definition 3.26. Define T [ b v ] ( M L u ) T_{[b_v]}(M_L^u) T [ b v ] ( M L u ) to be the kernel of the natural map
⨁ I C I u : T b v ( M L u ) ⟶ ⨁ ∏ v ( u ) = v { 1 , … , b u } T b v − 1 ( M L u ) \bigoplus_I C_I^u:T_{b_v}(M_L^u)\longrightarrow\bigoplus_{\prod_{v(u)=v}\{1,\ldots,b_u\}}T_{b_v-1}(M_L^u) I ⨁ C I u : T b v ( M L u ) ⟶ ∏ v ( u ) = v { 1 , … , b u } ⨁ T b v − 1 ( M L u )
in the abelian category D h s ∨ ( X L u , E ∞ ) D_{\mathrm{hs}}^\vee(X_L^u,\mathcal{E}_\infty) D hs ∨ ( X L u , E ∞ ) , which admits a natural action by S b v \mathfrak{S}_{b_v} S b v .
Similar to Remark 2.4(2), T [ b v ] ( M L u ) T_{[b_v]}(M_L^u) T [ b v ] ( M L u ) is a direct summand of T b v ( M L u ) T_{b_v}(M_L^u) T b v ( M L u ) . Thus, by Lemma 3.14, we have a canonical isomorphism
r T [ b v ] ( M L ) ≃ T [ b v ] ( M L u ) . {}^rT_{[b_v]}(M_L)\simeq T_{[b_v]}(M_L^u). r T [ b v ] ( M L ) ≃ T [ b v ] ( M L u ) .
By the similar process of constructing S [ λ v ] ( M L ) S_{[\lambda_v]}(M_L) S [ λ v ] ( M L ) , S [ λ ] ( M L ) S_{[\lambda]}(M_L) S [ λ ] ( M L ) , and M L , λ M_{L,\lambda} M L , λ from T [ b v ] ( M L ) T_{[b_v]}(M_L) T [ b v ] ( M L ) , we obtain the u u u -Hodge analogues S [ λ v ] ( M L u ) S_{[\lambda_v]}(M_L^u) S [ λ v ] ( M L u ) , S [ λ ] ( M L u ) ∈ D h s ∨ ( X L u , E ∞ ) S_{[\lambda]}(M_L^u)\in D_{\mathrm{hs}}^\vee(X_L^u,\mathcal{E}_\infty) S [ λ ] ( M L u ) ∈ D hs ∨ ( X L u , E ∞ ) , and M L , λ u ∈ D h s ∨ ( X L u , E ∞ u ) M_{L,\lambda}^u\in D_{\mathrm{hs}}^\vee(X_L^u,\mathcal{E}_\infty^u) M L , λ u ∈ D hs ∨ ( X L u , E ∞ u ) from T [ b v ] ( M L u ) T_{[b_v]}(M_L^u) T [ b v ] ( M L u ) , with canonical isomorphisms:
r S [ λ v ] ( M L ) ≃ S [ λ v ] ( M L u ) , r S [ λ ] ( M L ) ≃ S [ λ ] ( M L u ) , r M L , λ ≃ M L , λ u . (29) {}^rS_{[\lambda_v]}(M_L)\simeq S_{[\lambda_v]}(M_L^u),\qquad{}^rS_{[\lambda]}(M_L)\simeq S_{[\lambda]}(M_L^u),\qquad{}^rM_{L,\lambda}\simeq M_{L,\lambda}^u. \tag*{(29)} r S [ λ v ] ( M L ) ≃ S [ λ v ] ( M L u ) , r S [ λ ] ( M L ) ≃ S [ λ ] ( M L u ) , r M L , λ ≃ M L , λ u . (29)
Remark 3.27. The evaluation map e L , λ e_{L,\lambda} e L , λ (21) induces an evaluation map
e L , λ u : M L , λ u ⊗ E ∞ u M L , λ u ⟶ 1 X L u , E ∞ u h s (30) e_{L,\lambda}^u:M_{L,\lambda}^u\otimes_{\mathcal{E}_\infty^u}M_{L,\lambda}^u\longrightarrow\mathbb{1}_{X_L^u,\mathcal{E}_\infty^u}^{\mathrm{hs}} \tag*{(30)} e L , λ u : M L , λ u ⊗ E ∞ u M L , λ u ⟶ 1 X L u , E ∞ u hs (30)
in D h s ∨ ( M f d , E ∞ u ) D_{\mathrm{hs}}^\vee(\mathrm{Mfd},\mathcal{E}_\infty^u) D hs ∨ ( Mfd , E ∞ u ) for every L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(A_F^\infty)) L ∈ Hk ( H ( A F ∞ )) .
To end this section, we give an explicit description of the complex relative Hodge structure [28 Definition 4.3] κ M L , λ u ∈ D h s ∨ ( X L u , C ) \kappa M_{L,\lambda}^u\in D_{\mathrm{hs}}^\vee(X_L^u,\mathbb{C}) κ M L , λ u ∈ D hs ∨ ( X L u , C ) (see Notation 1.11 for κ \kappa κ ) via the complex uniformization
X L u = H ( v ) ( F ) \ D u × H ( v ) ( A F ∞ ) / L X_L^u=H(v)(F)\backslash\mathcal{D}^u\times H(v)(A_F^\infty)/L X L u = H ( v ) ( F ) \ D u × H ( v ) ( A F ∞ ) / L
in Remark 3.4, where v : = v ( u ) v:=v(u) v := v ( u ) .
In [28 §3] , we have constructed the complex representation
S [ λ ] ( V ( v ) ∞ ) : = ⨂ v ∈ V F ( ∞ ) S [ λ v ] ( V ( v ) v ) S_{[\lambda]}(V(v)_\infty):=\bigotimes_{v\in\mathcal{V}_F^{(\infty)}}S_{[\lambda_v]}(V(v)_v) S [ λ ] ( V ( v ) ∞ ) := v ∈ V F ( ∞ ) ⨂ S [ λ v ] ( V ( v ) v )
(tensor over C \mathbb{C} C ) as a representation of H ( v ) ( F ∞ ) H(v)(F_\infty) H ( v ) ( F ∞ ) . Recall that
S [ λ v ] ( V ( v ) v ) = ( ⨂ v ( u ) = v π λ u + ) ( T [ b v ] ( V ( v ) v ) ) , S_{[\lambda_v]}(V(v)_v)=\left(\bigotimes_{v(u)=v}\pi_{\lambda_u}^+\right)\left(T_{[b_v]}(V(v)_v)\right), S [ λ v ] ( V ( v ) v ) = v ( u ) = v ⨂ π λ u + ( T [ b v ] ( V ( v ) v ) ) ,
where T [ b v ] ( V ( v ) v ) T_{[b_v]}(V(v)_v) T [ b v ] ( V ( v ) v ) is the harmonic direct summand of
T b v ( V ( v ) v ) : = ⨂ v ( u ) = v T b u ( V ( v ) u ) T_{b_v}(V(v)_v):=\bigotimes_{v(u)=v}T^{b_u}(V(v)_u) T b v ( V ( v ) v ) := v ( u ) = v ⨂ T b u ( V ( v ) u )
similar to (1) , where V ( v ) u : = V ( v ) ⊗ E , u C V(v)_u:=V(v)\otimes_{E,u}\mathbb{C} V ( v ) u := V ( v ) ⊗ E , u C .
Definition 3.28. We define complex Hodge structures on the local system V ( v ) u V(v)_u V ( v ) u over D u \mathcal{D}^u D u . Put V u u : = V ( v ) u ⊗ C O D u \mathcal{V}_u^u:=V(v)_u\otimes_{\mathbb{C}}\mathcal{O}_{\mathcal{D}^u} V u u := V ( v ) u ⊗ C O D u .
For u ∈ Hom ( E , C ) ∖ { u , u ˉ } u\in\operatorname{Hom}(E,\mathbb{C})\setminus\{u,\bar{u}\} u ∈ Hom ( E , C ) ∖ { u , u ˉ } , define V u u \mathcal{V}_u^u V u u to be the unique complex Hodge structure on V ( v ) u V(v)_u V ( v ) u satisfying
( V u u ) ( 0 , 0 ) = ( V u u ) ∞ . (\mathcal{V}_u^u)^{(0,0)}=(\mathcal{V}_u^u)^\infty. ( V u u ) ( 0 , 0 ) = ( V u u ) ∞ .
For u = u ‾ u=\overline{u} u = u , denote by V D u − \mathcal{V}_{\mathcal{D}^{u}}^{-} V D u − the tautological complex subbundle of negative lines of ( V u ) ∞ = V ( ν ) u ⊗ C O D u (\mathcal{V}^{u})^{\infty} = V(\nu)_{u} \otimes_{\mathbb{C}} \mathcal{O}_{\mathcal{D}^{u}} ( V u ) ∞ = V ( ν ) u ⊗ C O D u over (the underlying smooth manifold of) D u \mathcal{D}^{u} D u , which is a holomorphic subbundle of ( V u ) ∞ (\mathcal{V}^{u})^{\infty} ( V u ) ∞ , and by V D u + \mathcal{V}_{\mathcal{D}^{u}}^{+} V D u + its orthogonal complement. Define V u \mathcal{V}^{u} V u to be the unique complex Hodge structure on V ( ν ) u V(\nu)_{u} V ( ν ) u satisfying that ( V u u ) ( 0 , 0 ) = V D u + (\mathcal{V}^{u}_{u})^{(0,0)} = V_{\mathcal{D}^{u}}^{+} ( V u u ) ( 0 , 0 ) = V D u + and ( V u u ) ( 1 , − 1 ) = V D u − (\mathcal{V}^{u}_{u})^{(1,-1)} = V_{\mathcal{D}^{u}}^{-} ( V u u ) ( 1 , − 1 ) = V D u − .19
For u = u ‾ u=\overline{u} u = u , define V u u \mathcal{V}^{u}_{u} V u u to be the complex conjugation of V u u \mathcal{V}^{u}_{u} V u u , which is a complex Hodge structure on V ( ν ) u = V ( ν ) u ‾ V(\nu)_{u}=V(\nu)_{\overline{u}} V ( ν ) u = V ( ν ) u . Note that V u u \mathcal{V}^{u}_{u} V u u is also canonically isomorphic to the dual of V u u \mathcal{V}^{u}_{u} V u u induced by the hermitian form on V ( ν ) u V(\nu)_{u} V ( ν ) u .
Taking tensor products of V u u \mathcal{V}^{u}_{u} V u u for u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) induces a complex Hodge structure V λ u \mathcal{V}^{u}_{\lambda} V λ u on the local system S [ λ ] ( V ( ν ) ∞ ) S_{[\lambda]}(V(\nu)_{\infty}) S [ λ ] ( V ( ν ) ∞ ) over D u \mathcal{D}^{u} D u . The complex Hodge structures V u u \mathcal{V}^{u}_{u} V u u and V λ u \mathcal{V}^{u}_{\lambda} V λ u are all H ( ν ) ( F ∞ ) H(\nu)(F_{\infty}) H ( ν ) ( F ∞ ) -invariant and descend to the complex unitary Shimura variety
X L u = H ( ν ) ( F ) \ D u × H ( ν ) ( A F ∞ ) / L , X^{u}_{L}=H(\nu)(F)\backslash\mathcal{D}^{u}\times H(\nu)(\mathbb{A}^{\infty}_{F})/L, X L u = H ( ν ) ( F ) \ D u × H ( ν ) ( A F ∞ ) / L ,
for which we denote by V u , L u \mathcal{V}^{u}_{u,L} V u , L u and V λ , L u \mathcal{V}^{u}_{\lambda,L} V λ , L u , respectively.
We now construct a canonical isomorphism
κ ( M L u [ u ] ) ⟶ ~ V u , L u (31) \kappa\left(M^{u}_{L}[u]\right)\mathrel{\widetilde{\longrightarrow}}\mathcal{V}^{u}_{u,L} \tag*{(31)} κ ( M L u [ u ] ) ⟶ V u , L u (31)
of complex Hodge structures over X L u X^{u}_{L} X L u for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . To do this, it suffices to construct such an isomorphism over X L , U u X^{u}_{L,U} X L , U u for an arbitrarily chosen object U ∈ Hk ( T ( A ∞ ) ) U\in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) that is Ξ U \Xi_{U} Ξ U -equivariant. Recall that M L , U = π L , U ∗ M L M_{L,U}=\pi^{*}_{L,U}M_{L} M L , U = π L , U ∗ M L and put V u , L , U u : = ( π L , U u ) ∗ V u , L u \mathcal{V}^{u}_{u,L,U}:=(\pi^{u}_{L,U})^{*}\mathcal{V}^{u}_{u,L} V u , L , U u := ( π L , U u ) ∗ V u , L u . Note that
X L , U u = ( H ( ν ) ( F ) \ D u × H ( ν ) ( A F ∞ ) / L ) × ( T ( Q ) \ T ( A ∞ ) / U ) . X^{u}_{L,U}=\left(H(\nu)(F)\backslash\mathcal{D}^{u}\times H(\nu)(\mathbb{A}^{\infty}_{F})/L\right)\times\left(T(\mathbb{Q})\backslash T(\mathbb{A}^{\infty})/U\right). X L , U u = ( H ( ν ) ( F ) \ D u × H ( ν ) ( A F ∞ ) / L ) × ( T ( Q ) \ T ( A ∞ ) / U ) .
Choose a purely imaginary element j ∈ E 0 × j\in E_{0}^{\times} j ∈ E 0 × and denote by W 0 W_{0} W 0 the space E 0 E_{0} E 0 equipped with the rational skew-hermitian form ⟨ x , y ⟩ ↦ Tr E 0 / Q j x y \langle x,y\rangle\mapsto\operatorname{Tr}_{E_{0}/\mathbb{Q}}jxy ⟨ x , y ⟩ ↦ Tr E 0 / Q j x y . Then W : = W 0 ⊗ E 0 V W:=W_{0}\otimes_{E_{0}}V W := W 0 ⊗ E 0 V has an induced rational skew-hermitian form using the hermitian form on V V V . Over the complex manifold D L , U u : = D u × H ( ν ) ( A F ∞ ) / L × T ( A ∞ ) / U \mathcal{D}^{u}_{L,U}:=\mathcal{D}^{u}\times H(\nu)(\mathbb{A}^{\infty}_{F})/L\times T(\mathbb{A}^{\infty})/U D L , U u := D u × H ( ν ) ( A F ∞ ) / L × T ( A ∞ ) / U , the relative homology groups H 1 ( A 0 , L , U u / D L , U u , Q ) H_{1}(A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}) H 1 ( A 0 , L , U u / D L , U u , Q ) and H 1 ( A L , U u / D L , U u , Q ) H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}) H 1 ( A L , U u / D L , U u , Q ) can be trivialized as
W 0 ≃ H 1 ( A 0 , L , U u / D L , U u , Q ) , W ≃ H 1 ( A L , U u / D L , U u , Q ) (32) W_{0}\simeq H_{1}(A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}),\qquad W\simeq H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}) \tag*{(32)} W 0 ≃ H 1 ( A 0 , L , U u / D L , U u , Q ) , W ≃ H 1 ( A L , U u / D L , U u , Q ) (32)
as E 0 E_{0} E 0 -linear and E E E -linear spaces equipped with rational skew-hermitian forms, respectively. Then we obtain an isomorphism
Hom E 0 ( H 1 ( A 0 , L , U u / D L , U u , Q ) , H 1 ( A L , U u / D L , U u , Q ) ) ≃ V ( ν ) (33) \operatorname{Hom}_{E_{0}}\left(H_{1}(A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}),H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q})\right)\simeq V(\nu) \tag*{(33)} Hom E 0 ( H 1 ( A 0 , L , U u / D L , U u , Q ) , H 1 ( A L , U u / D L , U u , Q ) ) ≃ V ( ν ) (33)
of E E E -linear local systems over D L , U u \mathcal{D}^{u}_{L,U} D L , U u . Take an element u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . By Lemma 3.29 below, the above isomorphism induces an isomorphism κ ( M L , U u [ u ] ) ⟶ V u , L , U u \kappa(M^{u}_{L,U}[u])\longrightarrow\mathcal{V}^{u}_{u,L,U} κ ( M L , U u [ u ]) ⟶ V u , L , U u of complex Hodge structures over D L , U u \mathcal{D}^{u}_{L,U} D L , U u , which is clearly H ( ν ) ( F ) × T ( Q ) H(\nu)(F)\times T(\mathbb{Q}) H ( ν ) ( F ) × T ( Q ) -invariant hence descends to an isomorphism over X L , U u X^{u}_{L,U} X L , U u . The resulting isomorphism κ ( M L , U u [ u ] ) ⟶ V u , L , U u \kappa(M^{u}_{L,U}[u])\longrightarrow\mathcal{V}^{u}_{u,L,U} κ ( M L , U u [ u ]) ⟶ V u , L , U u over X L , U u X^{u}_{L,U} X L , U u does not depend on the choice of the trivialization (3.28) and is Ξ U \Xi_{U} Ξ U -equivariant, so that we obtain a canonical isomorphism (3.27).
Finally, the isomorphisms (3.27) for all u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) induce an isomorphism
κ M L , λ u ⟶ ~ V λ , L u (34) \kappa M^{u}_{L,\lambda}\mathrel{\widetilde{\longrightarrow}}\mathcal{V}^{u}_{\lambda,L} \tag*{(34)} κ M L , λ u ⟶ V λ , L u (34)
in the category D h s ( X L u , C ) D^{\mathrm{hs}}(X^{u}_{L},\mathbb{C}) D hs ( X L u , C ) .
Lemma 3.29. For every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , the isomorphism
Hom E 0 ( H 1 ( A 0 , L , U u / D L , U u , Q ) , H 1 ( A L , U u / D L , U u , Q ) ) ⊗ E , u C ⟶ ~ V ( ν ) u \operatorname{Hom}_{E_{0}}\left(H_{1}(A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}),H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q})\right)\otimes_{E,u}\mathbb{C}\mathrel{\widetilde{\longrightarrow}}V(\nu)_{u} Hom E 0 ( H 1 ( A 0 , L , U u / D L , U u , Q ) , H 1 ( A L , U u / D L , U u , Q ) ) ⊗ E , u C ⟶ V ( ν ) u induced from (3.29) is an isomorphism of complex Hodge structures κ ( M L , U u [ u ] ) ⟶ V u , L , U u \kappa(M^{u}_{L,U}[u])\longrightarrow\mathcal{V}^{u}_{u,L,U} κ ( M L , U u [ u ]) ⟶ V u , L , U u over D L , U u \mathcal{D}^{u}_{L,U} D L , U u .
Proof. First, we realize that the underlying complex local system of κ ( M L , U u [ u ] ) \kappa(\mathcal{M}^{u}_{L,U}[u]) κ ( M L , U u [ u ]) is indeed the left-hand side of the isomorphism in question.
Recall that we have the Hodge sequence
0 ⟶ ω ( A L , U u / D L , U u ) ∨ ⟶ H 1 ( A L , U u / D L , U u , C ) ⟶ Lie A L , U u / D L , U u ⟶ 0 0 \longrightarrow\omega(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U})^{\vee} \longrightarrow H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{C}) \longrightarrow\operatorname{Lie}A^{u}_{L,U}/\mathcal{D}^{u}_{L,U} \longrightarrow0 0 ⟶ ω ( A L , U u / D L , U u ) ∨ ⟶ H 1 ( A L , U u / D L , U u , C ) ⟶ Lie A L , U u / D L , U u ⟶ 0
of complex local systems, which is the direct sum of u u u -isotypic components
0 ⟶ ω ( A L , U u / D L , U u ) ∨ [ u ] ⟶ H 1 ( A L , U u / D L , U u , Q ) ⊗ E , u C ⟶ Lie A L , U u / D L , U u [ u ] ⟶ 0 (35) 0 \longrightarrow\omega(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U})^{\vee}[u] \longrightarrow H_{1}(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U},\mathbb{Q}) \otimes_{E,u} \mathbb{C} \longrightarrow\operatorname{Lie}A^{u}_{L,U}/\mathcal{D}^{u}_{L,U}[u] \longrightarrow0 \tag*{(35)} 0 ⟶ ω ( A L , U u / D L , U u ) ∨ [ u ] ⟶ H 1 ( A L , U u / D L , U u , Q ) ⊗ E , u C ⟶ Lie A L , U u / D L , U u [ u ] ⟶ 0 (35)
for u ∈ Hom ( E , C ) u \in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . The sequence (35) gives a complex Hodge structure W \mathcal{W} W on W u : = W ⊗ E , u C W_{u} := W \otimes_{E,u} \mathbb{C} W u := W ⊗ E , u C over D L , U u \mathcal{D}^{u}_{L,U} D L , U u of weight − 1 -1 − 1 satisfying
W ( 0 , − 1 ) = ω ( A L , U u / D L , U u ) ∨ [ u ] , W ( − 1 , 0 ) = Lie A L , U u / D L , U u [ u ] , \mathcal{W}^{(0,-1)} = \omega(A^{u}_{L,U}/\mathcal{D}^{u}_{L,U})^{\vee}[u], \qquad\mathcal{W}^{(-1,0)} = \operatorname{Lie}A^{u}_{L,U}/\mathcal{D}^{u}_{L,U}[u], W ( 0 , − 1 ) = ω ( A L , U u / D L , U u ) ∨ [ u ] , W ( − 1 , 0 ) = Lie A L , U u / D L , U u [ u ] ,
which is nothing but κ ( h h s ( A L , U u / X L , U u , R ) [ u ] ) \kappa\left(h^{\mathrm{hs}}(A^{u}_{L,U}/X^{u}_{L,U},\mathbb{R})[u]\right) κ ( h hs ( A L , U u / X L , U u , R ) [ u ] ) .
Similarly, from A 0 , L , U A_{0,L,U} A 0 , L , U , we obtain a complex Hodge structure W 0 \mathcal{W}_{0} W 0 on W 0 , u : = W 0 ⊗ E 0 , u C W_{0,u} := W_{0} \otimes_{E_{0},u} \mathbb{C} W 0 , u := W 0 ⊗ E 0 , u C over D L , U u \mathcal{D}^{u}_{L,U} D L , U u of weight − 1 -1 − 1 satisfying
W 0 ( 0 , − 1 ) = ω ( A 0 , L , U u / D L , U u ) ∨ [ u ] , W 0 ( − 1 , 0 ) = Lie A 0 , L , U u / D L , U u [ u ] , \mathcal{W}_{0}^{(0,-1)} = \omega(A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U})^{\vee}[u], \qquad\mathcal{W}_{0}^{(-1,0)} = \operatorname{Lie}A^{u}_{0,L,U}/\mathcal{D}^{u}_{L,U}[u], W 0 ( 0 , − 1 ) = ω ( A 0 , L , U u / D L , U u ) ∨ [ u ] , W 0 ( − 1 , 0 ) = Lie A 0 , L , U u / D L , U u [ u ] ,
which is nothing but κ ( h h s ( A 0 , L , U u / X L , U u , R ) [ u ] ) \kappa\left(h^{\mathrm{hs}}(A^{u}_{0,L,U}/X^{u}_{L,U},\mathbb{R})[u]\right) κ ( h hs ( A 0 , L , U u / X L , U u , R ) [ u ] ) .
Together, we obtain that κ ( M L , U u [ u ] ) = Hom E 0 ( W 0 , W ) \kappa(\mathcal{M}^{u}_{L,U}[u]) = \operatorname{Hom}_{E_{0}}(\mathcal{W}_{0},\mathcal{W}) κ ( M L , U u [ u ]) = Hom E 0 ( W 0 , W ) . It is easy to see that under the canonical isomorphism Hom C ( W 0 , u , W u ) = V ( ν ) u \operatorname{Hom}_{\mathbb{C}}(W_{0,u},W_{u}) = V(\nu)_{u} Hom C ( W 0 , u , W u ) = V ( ν ) u , we have Hom E 0 ( W 0 , W ) = V u , L , U u \operatorname{Hom}_{E_{0}}(\mathcal{W}_{0},\mathcal{W}) = \mathcal{V}^{u}_{u,L,U} Hom E 0 ( W 0 , W ) = V u , L , U u as complex Hodge structures over D L , U u \mathcal{D}^{u}_{L,U} D L , U u .
The lemma is proved. □
Remark 3.30. The various realizations of the automorphic motive M ∙ , λ \mathbf{M}_{\bullet,\lambda} M ∙ , λ (Definition 3.15) studied in this and the previous sections can be summarized in the following picture:
V λ u → g l o b a l i z a t i o n V λ , ∙ u ⟵ ( 3.30 ) κ M ∙ , λ u ⟵ ( 3.25 ) r u M ∙ , λ → r ℓ ( 3.20 ) M ∙ , ℓ , λ → ≃ ( 3.22 ) V ℓ , ∙ , λ ⟵ − ⊗ Z ℓ Q ℓ ( 3.23 ) Λ ℓ , λ , ∙ \mathcal{V}^{u}_{\lambda} \xrightarrow{\mathrm{globalization}} \mathcal{V}^{u}_{\lambda,\bullet}
\longleftarrow^{\kappa}_{(3.30)} \mathbf{M}^{u}_{\bullet,\lambda}
\longleftarrow^{r^{u}}_{(3.25)} \mathbf{M}_{\bullet,\lambda}
\xrightarrow{r_{\ell}}_{(3.20)} \mathbf{M}_{\bullet,\ell,\lambda}
\xrightarrow{\simeq}_{(3.22)} V_{\ell,\bullet,\lambda}
\longleftarrow_{-\otimes_{\mathbb{Z}_{\ell}}\mathbb{Q}_{\ell}}^{(3.23)}
\Lambda_{\ell,\lambda,\bullet} V λ u globalization V λ , ∙ u ⟵ ( 3.30 ) κ M ∙ , λ u ⟵ ( 3.25 ) r u M ∙ , λ r ℓ ( 3.20 ) M ∙ , ℓ , λ ≃ ( 3.22 ) V ℓ , ∙ , λ ⟵ − ⊗ Z ℓ Q ℓ ( 3.23 ) Λ ℓ , λ , ∙
in which the dashed arrows require ℓ ∤ D E B λ [ E : E ′ ] \ell\nmid D_{E}B_{\lambda}[E:E^{\prime}] ℓ ∤ D E B λ [ E : E ′ ] and that L ℓ L_{\ell} L ℓ is contained in the stabilizer of the (self-dual) lattice Λ ℓ \Lambda_{\ell} Λ ℓ (see Remark 3.24).
Motivic special classes and motivic generating functions
Take a hermitian E E E -weight λ \lambda λ of rank m m m (Definition 2.7) and consider a positive integer r r r such that λ \lambda λ is r r r -balanced (Definition 2.7(2)).
We construct, for every tuple x = ( x 1 , … , x r ) ∈ ( V ∘ ) r x=(x_{1},\ldots,x_{r}) \in(V^{\circ})^{r} x = ( x 1 , … , x r ) ∈ ( V ∘ ) r satisfying T ( x ) ∈ Herm r ( F ) + T(x) \in\operatorname{Herm}_{r}(F)^{+} T ( x ) ∈ Herm r ( F ) + a motivic special class , which is an E λ E^{\lambda} E λ -linear map
Z λ ( x ) L : S λ + ( ( E ∨ ) r ) ⟶ H 2 r ( X L , M L , λ ( r ) ) Z^{\lambda}(x)_{L}: S_{\lambda^{+}}\left((E^{\vee})^{r}\right) \longrightarrow H^{2r}(X_{L},\mathcal{M}_{L,\lambda}(r)) Z λ ( x ) L : S λ + ( ( E ∨ ) r ) ⟶ H 2 r ( X L , M L , λ ( r ))
(see Construction 2.2 for the source) for every L ∈ Hk ( H ( A F ∞ ) ) L \in\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) L ∈ Hk ( H ( A F ∞ )) .
Recall from (14) that we have the canonical map
μ ( x ) L : S ( x ) ∗ M L ∨ ⟶ 1 Z ( x ) L , E Φ r \mu(x)_{L}: \mathscr{S}(x)^{*}\mathcal{M}^{\vee}_{L} \longrightarrow\mathbb{1}^{\Phi r}_{Z(x)_{L},E} μ ( x ) L : S ( x ) ∗ M L ∨ ⟶ 1 Z ( x ) L , E Φ r
in Mot ( Z ( x ) L , E ) \operatorname{Mot}(Z(x)_{L},E) Mot ( Z ( x ) L , E ) . Replace 1 Z ( x ) L , E \mathbb{1}_{Z(x)_{L},E} 1 Z ( x ) L , E by 1 Z ( x ) L , E ∨ \mathbb{1}^{\vee}_{Z(x)_{L},E} 1 Z ( x ) L , E ∨ by the tautological duality of the unit object, so that we have the induced map
μ ( x ) L [ u ] : S ( x ) L ∗ ( M L [ u ] ) ∨ ⟶ ( 1 Z ( x ) L , E ∨ ) Φ r [ u ] = ( 1 Z ( x ) L , E ∨ ) Φ r \mu(x)_{L}[u]: \mathscr{S}(x)^{*}_{L}(\mathcal{M}_{L}[u])^{\vee} \longrightarrow(\mathbb{1}^{\vee}_{Z(x)_{L},E})^{\Phi r}[u] = (\mathbb{1}^{\vee}_{Z(x)_{L},E})^{\Phi r} μ ( x ) L [ u ] : S ( x ) L ∗ ( M L [ u ] ) ∨ ⟶ ( 1 Z ( x ) L , E ∨ ) Φ r [ u ] = ( 1 Z ( x ) L , E ∨ ) Φ r
(Notation 3.13) for every u ∈ Hom ( E , C ) u \in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . Taking tensor products, we have the map
μ b ( x ) L : S ( x ) L ∗ T b ( M L ) ∨ → T b ( 1 Z ( x ) L , E ⊕ r ) ∨ = ( T b ( 1 Z ( x ) L , E ⊕ r ) ) ∨ \mu_b(x)_L:\mathcal{S}(x)_L^*T_b(M_L)^\vee\to T^b\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}}^{\oplus r}\right)^\vee=\left(T^b\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}}^{\oplus r}\right)\right)^\vee μ b ( x ) L : S ( x ) L ∗ T b ( M L ) ∨ → T b ( 1 Z ( x ) L , E ⊕ r ) ∨ = ( T b ( 1 Z ( x ) L , E ⊕ r ) ) ∨
in Mot ( Z ( x ) L , E ) \operatorname{Mot}(Z(x)_L,\mathbb{E}) Mot ( Z ( x ) L , E ) . Composing with the dual of the natural map T b ( M L ) → T [ b ] ( M L ) T_b(M_L)\to T_{[b]}(M_L) T b ( M L ) → T [ b ] ( M L ) , we obtain the map
μ [ b ] ( x ) L : S ( x ) L ∗ T [ b ] ( M L ) ∨ → T b ( 1 Z ( x ) L , E ⊕ r ) ∨ , \mu_{[b]}(x)_L:\mathcal{S}(x)_L^*T_{[b]}(M_L)^\vee\to T^b\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}}^{\oplus r}\right)^\vee, μ [ b ] ( x ) L : S ( x ) L ∗ T [ b ] ( M L ) ∨ → T b ( 1 Z ( x ) L , E ⊕ r ) ∨ ,
which is S b S_b S b -equivariant. Applying π λ + \pi_{\lambda}^{+} π λ + , we obtain the map
μ [ λ ] ( x ) L : S ( x ) L ∗ S [ λ ] ( M L ) ∨ → S λ + ( 1 Z ( x ) L , E ⊕ r ) ∨ . \mu_{[\lambda]}(x)_L:\mathcal{S}(x)_L^*S_{[\lambda]}(M_L)^\vee\to S_{\lambda+}\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}}^{\oplus r}\right)^\vee. μ [ λ ] ( x ) L : S ( x ) L ∗ S [ λ ] ( M L ) ∨ → S λ + ( 1 Z ( x ) L , E ⊕ r ) ∨ .
It is straightforward to verify that μ [ λ ] ( x ) L \mu_{[\lambda]}(x)_L μ [ λ ] ( x ) L is Gal ( E / E λ ) \operatorname{Gal}(E/E^{\lambda}) Gal ( E / E λ ) -equivariant, hence induces a map
μ λ ( x ) L : S ( x ) L ∗ M L , λ ∨ → S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨ \mu_{\lambda}(x)_L:\mathcal{S}(x)_L^*M_{L,\lambda}^{\vee}\to S_{\lambda+}\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)^\vee μ λ ( x ) L : S ( x ) L ∗ M L , λ ∨ → S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨
in the category Mot ( Z ( x ) L , E λ ) \operatorname{Mot}(Z(x)_L,\mathbb{E}^{\lambda}) Mot ( Z ( x ) L , E λ ) .21
Denote by D − D_- D − the Verdier duality functor in D ( − , E λ ) D(-,\mathbb{E}^{\lambda}) D ( − , E λ ) . We have the induced map
D Z ( x ) L μ λ ( x ) L : D Z ( x ) L ( S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨ ) → D Z ( x ) L ( S ( x ) L ∗ M L , λ ∨ ) D_{Z(x)_L}\mu_{\lambda}(x)_L:D_{Z(x)_L}\left(S_{\lambda+}\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)^\vee\right)\to D_{Z(x)_L}\left(\mathcal{S}(x)_L^*M_{L,\lambda}^{\vee}\right) D Z ( x ) L μ λ ( x ) L : D Z ( x ) L ( S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨ ) → D Z ( x ) L ( S ( x ) L ∗ M L , λ ∨ )
in the category D ( Z ( x ) L , E λ ) D(Z(x)_L,\mathbb{E}^{\lambda}) D ( Z ( x ) L , E λ ) . By the bi-duality property [Ayo14a, Theorem 3.10], we have canonical isomorphisms
D Z ( x ) L ( S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨ ) ≃ S ( x ) L ∗ ( D X L S λ + ( 1 X L , E λ ⊕ r ) ∨ ) , D_{Z(x)_L}\left(S_{\lambda+}\left(\mathbbm{1}_{Z(x)_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)^\vee\right)\simeq\mathcal{S}(x)_L^*\left(D_{X_L}S_{\lambda+}\left(\mathbbm{1}_{X_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)^\vee\right), D Z ( x ) L ( S λ + ( 1 Z ( x ) L , E λ ⊕ r ) ∨ ) ≃ S ( x ) L ∗ ( D X L S λ + ( 1 X L , E λ ⊕ r ) ∨ ) ,
D Z ( x ) L ( S ( x ) L ∗ M L , λ ∨ ) ≃ S ( x ) L ∗ ( D X L M L , λ ) . D_{Z(x)_L}\left(\mathcal{S}(x)_L^*M_{L,\lambda}^{\vee}\right)\simeq\mathcal{S}(x)_L^*\left(D_{X_L}M_{L,\lambda}\right). D Z ( x ) L ( S ( x ) L ∗ M L , λ ∨ ) ≃ S ( x ) L ∗ ( D X L M L , λ ) .
Since X L X_L X L is regular, we have canonical isomorphisms
D X L S λ + ( 1 X L , E λ ⊕ r ) ∨ ≃ S λ + ( 1 X L , E λ ⊕ r ) ⟨ dim X L ⟩ , D X L M L , λ ∨ ≃ M L , λ ⟨ dim X L ⟩ . D_{X_L}S_{\lambda+}\left(\mathbbm{1}_{X_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)^\vee\simeq S_{\lambda+}\left(\mathbbm{1}_{X_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)\langle\dim X_L\rangle,\qquad D_{X_L}M_{L,\lambda}^{\vee}\simeq M_{L,\lambda}\langle\dim X_L\rangle. D X L S λ + ( 1 X L , E λ ⊕ r ) ∨ ≃ S λ + ( 1 X L , E λ ⊕ r ) ⟨ dim X L ⟩ , D X L M L , λ ∨ ≃ M L , λ ⟨ dim X L ⟩ .
Combining together, D Z ( x ) L μ λ ( x ) L D_{Z(x)_L}\mu_{\lambda}(x)_L D Z ( x ) L μ λ ( x ) L is indeed a map
μ λ ( x ) L : S λ + ( ( E λ ) r ) ⊗ S ( x ) L ∗ 1 X L , Q = S ( x ) L ∗ S λ + ( 1 X L , E λ ⊕ r ) → S ( x ) L ∗ M L , λ (36) \mu^{\lambda}(x)_L:S_{\lambda+}((\mathbb{E}^{\lambda})^r)\otimes\mathcal{S}(x)_L^*\mathbbm{1}_{X_L,\mathbb{Q}}=\mathcal{S}(x)_L^*S_{\lambda+}\left(\mathbbm{1}_{X_L,\mathbb{E}^{\lambda}}^{\oplus r}\right)\to\mathcal{S}(x)_L^*M_{L,\lambda} \tag*{(36)} μ λ ( x ) L : S λ + (( E λ ) r ) ⊗ S ( x ) L ∗ 1 X L , Q = S ( x ) L ∗ S λ + ( 1 X L , E λ ⊕ r ) → S ( x ) L ∗ M L , λ (36)
in the category D ( Z ( x ) L , E λ ) D(Z(x)_L,\mathbb{E}^{\lambda}) D ( Z ( x ) L , E λ ) .
Apply the motivic cohomology H 2 rank x ( Z ( x ) L , − ( rank x ) ) H^{2\operatorname{rank}x}(Z(x)_L,-(\operatorname{rank}x)) H 2 rank x ( Z ( x ) L , − ( rank x )) to μ λ ( x ) L \mu^{\lambda}(x)_L μ λ ( x ) L , we have the induced map
S λ + ( ( E λ ) r ) ⊗ H 2 rank x ( Z ( x ) L , S ( x ) L ∗ 1 X L , Q ( rank x ) ) → H 2 rank x ( Z ( x ) L , S ( x ) L ∗ M L , λ ( rank x ) ) . S_{\lambda+}((\mathbb{E}^{\lambda})^r)\otimes H^{2\operatorname{rank}x}\left(Z(x)_L,\mathcal{S}(x)_L^*\mathbbm{1}_{X_L,\mathbb{Q}}(\operatorname{rank}x)\right)\to H^{2\operatorname{rank}x}\left(Z(x)_L,\mathcal{S}(x)_L^*M_{L,\lambda}(\operatorname{rank}x)\right). S λ + (( E λ ) r ) ⊗ H 2 rank x ( Z ( x ) L , S ( x ) L ∗ 1 X L , Q ( rank x ) ) → H 2 rank x ( Z ( x ) L , S ( x ) L ∗ M L , λ ( rank x ) ) .
For the source, since Z ( x ) L Z(x)_L Z ( x ) L is regular, we have canonically
H 2 rank x ( Z ( x ) L , S ( x ) L ∗ 1 X L , Q ( rank x ) ) = H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) . H^{2\operatorname{rank}x}\left(Z(x)_L,\mathcal{S}(x)_L^*\mathbbm{1}_{X_L,\mathbb{Q}}(\operatorname{rank}x)\right)=H^0\left(Z(x)_L,\mathbbm{1}_{Z(x)_L,\mathbb{Q}}\right). H 2 rank x ( Z ( x ) L , S ( x ) L ∗ 1 X L , Q ( rank x ) ) = H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) .
For the target, we have the natural map
H 2 rank x ( Z ( x ) L , S ( x ) L ∗ M L , λ ( rank x ) ) → H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) , H^{2\operatorname{rank}x}\left(Z(x)_L,\mathcal{S}(x)_L^*M_{L,\lambda}(\operatorname{rank}x)\right)\to H^{2\operatorname{rank}x}_{Z[x]_L}\left(X_L,M_{L,\lambda}(\operatorname{rank}x)\right), H 2 rank x ( Z ( x ) L , S ( x ) L ∗ M L , λ ( rank x ) ) → H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) ,
where Z [ x ] L Z[x]_L Z [ x ] L denotes the image of the finite unramified morphism S ( x ) L \mathcal{S}(x)_L S ( x ) L (11) which is a Zariski closed subset of X L X_L X L . Together, we obtain the map
S λ + ( ( E λ ) r ) ⊗ H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) → H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) (37) S_{\lambda+}((\mathbb{E}^{\lambda})^r)\otimes H^0\left(Z(x)_L,\mathbbm{1}_{Z(x)_L,\mathbb{Q}}\right)\to H^{2\operatorname{rank}x}_{Z[x]_L}\left(X_L,M_{L,\lambda}(\operatorname{rank}x)\right) \tag*{(37)} S λ + (( E λ ) r ) ⊗ H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) → H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) (37)
induced from μ λ ( x ) L \mu^{\lambda}(x)_L μ λ ( x ) L .
Definition 3.31. Take a tuple x = ( x 1 , … , x r ) ∈ ( V ∨ ) r x=(x_1,\ldots,x_r)\in(V^{\vee})^r x = ( x 1 , … , x r ) ∈ ( V ∨ ) r satisfying T ( x ) ∈ Herm r ( F ) + T(x)\in\operatorname{Herm}_r(F)^+ T ( x ) ∈ Herm r ( F ) + . (1) Define the refined motivic special class z λ ( x ) L z^{\lambda}(x)_L z λ ( x ) L to be the composite map
S λ + ( ( E λ ) r ) → S λ + ( ( E λ ) r ) ⊗ H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) → (3.33) H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) , S_{\lambda+}((\mathbb{E}^{\lambda})^r)\to S_{\lambda+}((\mathbb{E}^{\lambda})^r)\otimes H^0\left(Z(x)_L,\mathbbm{1}_{Z(x)_L,\mathbb{Q}}\right)\xrightarrow{\text{\scriptsize(3.33)}}H^{2\operatorname{rank}x}_{Z[x]_L}\left(X_L,M_{L,\lambda}(\operatorname{rank}x)\right), S λ + (( E λ ) r ) → S λ + (( E λ ) r ) ⊗ H 0 ( Z ( x ) L , 1 Z ( x ) L , Q ) (3.33) H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) ,
in which the first one is induced by tensoring with the fundamental class of Z ( x ) L Z(x)_L Z ( x ) L .
(2) Define the motivic special class z λ ( x ) L z^\lambda(x)_L z λ ( x ) L to be the composite map
S λ + ( ( E λ ) r ) → z λ ( x ) L H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x ) ) ⟶ H 2 rank x ( X L , M L , λ ( rank x ) ) → ∧ e r − rank x H 2 r ( X L , M L , λ ( r ) ) , S_{\lambda+}((E^\lambda)^r) \xrightarrow{z^\lambda(x)_L} H^{2\operatorname{rank}x}_{Z[x]_L}(X_L,M_{L,\lambda}(\operatorname{rank}x)) \longrightarrow H^{2\operatorname{rank}x}(X_L,M_{L,\lambda}(\operatorname{rank}x)) \xrightarrow{\wedge e^{r-\operatorname{rank}x}} H^{2r}(X_L,M_{L,\lambda}(r)), S λ + (( E λ ) r ) z λ ( x ) L H Z [ x ] L 2 rank x ( X L , M L , λ ( rank x )) ⟶ H 2 rank x ( X L , M L , λ ( rank x )) ∧ e r − rank x H 2 r ( X L , M L , λ ( r )) ,
in which e ∈ H 2 ( X L , I X L , Q ( 1 ) ) e \in H^2(X_L,\mathbb{I}_{X_L,\mathbb{Q}}(1)) e ∈ H 2 ( X L , I X L , Q ( 1 )) denotes the class given by the dual Hodge line bundle over X L X_L X L .
Once motivic special classes are constructed, we may define generating functions using those classes. Take an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) and a Schwartz function ϕ ∞ ∈ S ( ( V ∞ ) r ) L \phi^\infty\in\mathcal{S}((V^\infty)^r)_L ϕ ∞ ∈ S (( V ∞ ) r ) L .
Definition 3.33. Take an element T ∈ Herm r ( F ) + T\in\operatorname{Herm}_r(F)^+ T ∈ Herm r ( F ) + . Put
Z T [ ϕ ∞ ] L : = ⋃ x ∈ ( V ∞ ) T r ϕ ∞ ( x ) ≠ 0 Z [ x ] L . Z_T[\phi^\infty]_L:=\bigcup_{\substack{x\in(V^\infty)^r_T\\ \phi^\infty(x)\ne0}}Z[x]_L. Z T [ ϕ ∞ ] L := x ∈ ( V ∞ ) T r ϕ ∞ ( x ) = 0 ⋃ Z [ x ] L . which is a finite union, as a Zariski closed subset of X L X_L X L .
(1) When T ∈ Herm r ∘ ( F ) + T\in\operatorname{Herm}_r^\circ(F)^+ T ∈ Herm r ∘ ( F ) + , put
Z T λ ( ϕ ∞ ) L : = ∑ x ∈ L \ ( V ∞ ) T r ϕ ∞ ( x ) ⋅ z λ ( x ) L ∈ S λ + ( C r ) ∨ ⊗ E λ H Z T [ ϕ ∞ ] L 2 r ( X L , M L , λ ( r ) ) . Z_T^\lambda(\phi^\infty)_L:=\sum_{x\in L\backslash(V^\infty)^r_T}\phi^\infty(x)\cdot z^\lambda(x)_L\in S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{E^\lambda}H^{2r}_{Z_T[\phi^\infty]_L}(X_L,M_{L,\lambda}(r)). Z T λ ( ϕ ∞ ) L := x ∈ L \ ( V ∞ ) T r ∑ ϕ ∞ ( x ) ⋅ z λ ( x ) L ∈ S λ + ( C r ) ∨ ⊗ E λ H Z T [ ϕ ∞ ] L 2 r ( X L , M L , λ ( r )) . where the tensor in the target is taken with respect to the inclusion E λ ⊆ C E^\lambda\subseteq\mathbb{C} E λ ⊆ C . We extend Z T λ ( ϕ ∞ ) L Z_T^\lambda(\phi^\infty)_L Z T λ ( ϕ ∞ ) L to a function on G ~ r ( A F ) \widetilde{G}_r(\mathbb{A}_F) G r ( A F ) via the formula
g ∈ G ~ r ( A F ) ↦ Z T λ ( ϕ ∞ ) L ( g ) : = ω r , ∞ ( g ∞ ) ϕ V ∞ 0 ( T ) ⋅ ρ λ + ( g ∞ ) ∨ ( Z T λ ( ω r ∞ ( g ∞ ) ϕ ∞ ) L ) g\in\widetilde{G}_r(\mathbb{A}_F)\mapsto Z_T^\lambda(\phi^\infty)_L(g):=\omega_{r,\infty}(g_\infty)\phi^0_{V_\infty}(T)\cdot\rho_{\lambda+}(g_\infty)^\vee\left(Z_T^\lambda(\omega_r^\infty(g^\infty)\phi^\infty)_L\right) g ∈ G r ( A F ) ↦ Z T λ ( ϕ ∞ ) L ( g ) := ω r , ∞ ( g ∞ ) ϕ V ∞ 0 ( T ) ⋅ ρ λ + ( g ∞ ) ∨ ( Z T λ ( ω r ∞ ( g ∞ ) ϕ ∞ ) L ) ∈ S λ + ( C r ) ∨ ⊗ E λ H Z T [ ω r ∞ ( g ∞ ) ϕ ∞ ] L 2 r ( X L , M L , λ ( r ) ) , \in S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{E^\lambda}H^{2r}_{Z_T[\omega_r^\infty(g^\infty)\phi^\infty]_L}(X_L,M_{L,\lambda}(r)), ∈ S λ + ( C r ) ∨ ⊗ E λ H Z T [ ω r ∞ ( g ∞ ) ϕ ∞ ] L 2 r ( X L , M L , λ ( r )) , where ρ λ + \rho_{\lambda+} ρ λ + is the map from Definition 2.14(3) and ϕ V ∞ 0 \phi^0_{V_\infty} ϕ V ∞ 0 is standard Gaussian function on V ∞ r V_\infty^r V ∞ r .
(2) For T T T general, put
Z T λ ( ϕ ∞ ) L : = ∑ x ∈ L \ ( V ∞ ) T r ϕ ∞ ( x ) ⋅ z λ ( x ) L ∈ S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r ) ) . Z_T^\lambda(\phi^\infty)_L:=\sum_{x\in L\backslash(V^\infty)^r_T}\phi^\infty(x)\cdot z^\lambda(x)_L\in S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{E^\lambda}H^{2r}(X_L,M_{L,\lambda}(r)). Z T λ ( ϕ ∞ ) L := x ∈ L \ ( V ∞ ) T r ∑ ϕ ∞ ( x ) ⋅ z λ ( x ) L ∈ S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r )) . Similar to (1), we extend Z T λ ( ϕ ∞ ) L Z_T^\lambda(\phi^\infty)_L Z T λ ( ϕ ∞ ) L to a function on G ~ r ( A F ) \widetilde{G}_r(\mathbb{A}_F) G r ( A F ) valued in S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r ) ) S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{E^\lambda}H^{2r}(X_L,M_{L,\lambda}(r)) S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r )) via the formula
g ∈ G ~ r ( A F ) ↦ Z T λ ( ϕ ∞ ) L ( g ) : = ω r , ∞ ( g ∞ ) ϕ V ∞ 0 ( T ) ⋅ ρ λ + ( g ∞ ) ∨ ( Z T λ ( ω r ∞ ( g ∞ ) ϕ ∞ ) L ) . g\in\widetilde{G}_r(\mathbb{A}_F)\mapsto Z_T^\lambda(\phi^\infty)_L(g):=\omega_{r,\infty}(g_\infty)\phi^0_{V_\infty}(T)\cdot\rho_{\lambda+}(g_\infty)^\vee\left(Z_T^\lambda(\omega_r^\infty(g^\infty)\phi^\infty)_L\right). g ∈ G r ( A F ) ↦ Z T λ ( ϕ ∞ ) L ( g ) := ω r , ∞ ( g ∞ ) ϕ V ∞ 0 ( T ) ⋅ ρ λ + ( g ∞ ) ∨ ( Z T λ ( ω r ∞ ( g ∞ ) ϕ ∞ ) L ) .
Lemma 3.34. For every element t ∈ C [ L \ H ( A F ) / L ] \mathfrak{t}\in\mathbb{C}[L\backslash H(\mathbb{A}_F)/L] t ∈ C [ L \ H ( A F ) / L ] , we have t Z T λ ( ϕ ∞ ) L = Z T λ ( t ϕ ∞ ) L \mathfrak{t}Z_T^\lambda(\phi^\infty)_L=Z_T^\lambda(\mathfrak{t}\phi^\infty)_L t Z T λ ( ϕ ∞ ) L = Z T λ ( t ϕ ∞ ) L .
Proof. This follows from the same proof of [20 Lemma 4.4] (see also [10 Remark 4.51] ). □ \square □
Definition 3.35. Define the motivic generating function Z λ ( ϕ ∞ ) L Z^\lambda(\phi^\infty)_L Z λ ( ϕ ∞ ) L to be the formal sum
Z λ ( ϕ ∞ ) L : = ∑ T ∈ Herm r ( F ) + Z T λ ( ϕ ∞ ) L Z^\lambda(\phi^\infty)_L:=\sum_{T\in\operatorname{Herm}_r(F)^+}Z_T^\lambda(\phi^\infty)_L Z λ ( ϕ ∞ ) L := T ∈ Herm r ( F ) + ∑ Z T λ ( ϕ ∞ ) L as a function on G ~ r ( A F ) \widetilde{G}_r(\mathbb{A}_F) G r ( A F ) valued in S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r ) ) S_{\lambda+}(\mathbb{C}^r)^\vee\otimes_{E^\lambda}H^{2r}(X_L,M_{L,\lambda}(r)) S λ + ( C r ) ∨ ⊗ E λ H 2 r ( X L , M L , λ ( r )) .
Conjecture 3.36 (Modularity of motivic generating function). Let r r r be a positive integer such that λ \lambda λ is r r r -balanced. For every element ϕ ∞ ∈ S ( ( V ∞ ) r ) L \phi^\infty\in\mathcal{S}((V^\infty)^r)_L ϕ ∞ ∈ S (( V ∞ ) r ) L , the motivic generating function $Z^\lambda(\phi^\infty)_L is an absolutely convergent sum of terms in a finite-dimensional subspace, which further belongs to A λ + m / 2 ( G ~ r ) h o l ⊗ E λ H 2 r ( X L , M L , λ ( r ) ) A_{\lambda+}^{m/2}(\widetilde{G}_r)^{\mathrm{hol}}\otimes_{E^\lambda}H^{2r}(X_L,M_{L,\lambda}(r)) A λ + m /2 ( G r ) hol ⊗ E λ H 2 r ( X L , M L , λ ( r )) (Definition 2.14(1)).
When λ = 0 \lambda=0 λ = 0 so that H 2 r ( X L , M L , λ ( r ) ) = CH r ( X L ) ⊗ Q H^{2r}(X_L,M_{L,\lambda}(r))=\operatorname{CH}^r(X_L)\otimes\mathbb{Q} H 2 r ( X L , M L , λ ( r )) = CH r ( X L ) ⊗ Q , the function Z 0 ( ϕ ∞ ) L Z^0(\phi^\infty)_L Z 0 ( ϕ ∞ ) L recovers Kudla’s generating function in the hermitian case and Conjecture 3.36 recovers Kudla’s modularity conjecture in the hermitian case.22
For a general weight λ \lambda λ , Conjecture 3.36 is not known even for r = 1 r=1 r = 1 and m = 2 m=2 m = 2 .
Assuming the conjecture, Z λ ( ϕ ∞ ) L Z^\lambda(\phi^\infty)_L Z λ ( ϕ ∞ ) L further belongs to A λ + m / 2 ( G r ) c u s p ⊗ E λ H 2 r ( X L , M L , λ ( r ) ) \mathcal{A}^{m/2}_{\lambda_+}(G_r)_{\mathrm{cusp}}\otimes_{E_\lambda}H^{2r}(X_L,M_{L,\lambda}(r)) A λ + m /2 ( G r ) cusp ⊗ E λ H 2 r ( X L , M L , λ ( r )) if r = max u { s ( λ u ) } r=\max_u\{s(\lambda_u)\} r = max u { s ( λ u )} .
The definition below generalizes [25 Definition 3.8] or [20 Definition 4.8] .
Definition 3.38. Assume Conjecture 3.36 for r r r . For every form ϕ ∈ A λ + m / 2 ( G r ) c u s p \phi\in\mathcal{A}^{m/2}_{\lambda_+}(G_r)_{\mathrm{cusp}} ϕ ∈ A λ + m /2 ( G r ) cusp , we define the motivic arithmetic theta lift of weight λ \lambda λ of ϕ \phi ϕ , with respect to a Schwartz function φ ∞ ∈ S ( ( V ∞ ) r ) \varphi^\infty\in\mathcal{S}((V^\infty)^r) φ ∞ ∈ S (( V ∞ ) r ) , to be
Θ φ ∞ λ ( ϕ ) L : = ( Z λ ( ϕ ∞ ) L , ϕ ) G r ∈ H 2 r ( X L , M L , λ ( r ) ) ⊗ E λ C . \Theta^{\lambda}_{\varphi^\infty}(\phi)_L := \left(Z^\lambda(\phi^\infty)_L,\phi\right)_{G_r}\in H^{2r}(X_L,M_{L,\lambda}(r))\otimes_{E_\lambda}\mathbb{C}. Θ φ ∞ λ ( ϕ ) L := ( Z λ ( ϕ ∞ ) L , ϕ ) G r ∈ H 2 r ( X L , M L , λ ( r )) ⊗ E λ C . (Definition 2.14(2)).
Canonical smooth integral models
In this section, we recall the construction of canonical smooth integral models at good primes. Take a finite set ⋄ \diamond ⋄ of rational primes.
Definition 3.41. We define a functor
Y ∙ , ( ⋄ ) : Hk ( T ( A ∞ , ⋄ ) ) ⟶ PSch ′ / O E 0 , ( ⋄ ) Y_{\bullet,(\diamond)}:\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond}))\longrightarrow\operatorname{PSch}'/\mathcal{O}_{E_0,(\diamond)} Y ∙ , ( ⋄ ) : Hk ( T ( A ∞ , ⋄ )) ⟶ PSch ′ / O E 0 , ( ⋄ ) such that for every objects U ⋄ ∈ Hk ( T ( A ∞ , ⋄ ) ) U^\diamond\in\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) U ⋄ ∈ Hk ( T ( A ∞ , ⋄ )) and S ∈ Sch / O E 0 , ( ⋄ ) S\in\operatorname{Sch}/\mathcal{O}_{E_0,(\diamond)} S ∈ Sch / O E 0 , ( ⋄ ) , Y U ⋄ , ( ⋄ ) ( S ) Y_{U^\diamond,(\diamond)}(S) Y U ⋄ , ( ⋄ ) ( S ) is the set of equivalence classes of triples ( A 0 , λ 0 , η ˉ 0 ⋄ ) (A_0,\lambda_0,\bar{\eta}_0^\diamond) ( A 0 , λ 0 , η ˉ 0 ⋄ ) , where
A 0 A_0 A 0 is an object of AV ( S , O E 0 , ( ⋄ ) ) \operatorname{AV}(S,\mathcal{O}_{E_0,(\diamond)}) AV ( S , O E 0 , ( ⋄ ) ) such that tr A 0 ( e ) = e \operatorname{tr}_{A_0}(e)=e tr A 0 ( e ) = e for e ∈ O E 0 , ( ⋄ ) e\in\mathcal{O}_{E_0,(\diamond)} e ∈ O E 0 , ( ⋄ ) ;
λ 0 : A 0 ⟶ A 0 ∨ \lambda_0:A_0\longrightarrow A_0^\vee λ 0 : A 0 ⟶ A 0 ∨ is a ⋄ \diamond ⋄ -principal polarization that turns the complex conjugation of O E 0 , ( ⋄ ) \mathcal{O}_{E_0,(\diamond)} O E 0 , ( ⋄ ) into the Rosati involution;
η ˉ 0 ⋄ \bar{\eta}_0^\diamond η ˉ 0 ⋄ is a U ⋄ U^\diamond U ⋄ -level structure, that is, for a chosen geometric point s s s on every connected component of S S S , a π 1 ( S , s ) \pi_1(S,s) π 1 ( S , s ) -invariant U ⋄ U^\diamond U ⋄ -orbit of isomorphisms
η 0 ⋄ : A E 0 ∞ , ⋄ ⟶ H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) \eta_0^\diamond:\mathbb{A}_{E_0}^{\infty,\diamond}\longrightarrow\mathrm{H}_{1}^{\mathrm{\acute{e}t}}(A_{0s},\mathbb{A}^{\infty,\diamond}) η 0 ⋄ : A E 0 ∞ , ⋄ ⟶ H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) of A E 0 ∞ , ⋄ \mathbb{A}_{E_0}^{\infty,\diamond} A E 0 ∞ , ⋄ -modules.
Two objects ( A 0 , λ 0 , η ˉ 0 ⋄ ) (A_0,\lambda_0,\bar{\eta}_0^\diamond) ( A 0 , λ 0 , η ˉ 0 ⋄ ) and ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ) (A'_0,\lambda'_0,\bar{\eta}'^\diamond_0) ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ) are equivalent if there exists an isomorphism φ 0 : A 0 ⟶ A 0 ′ \varphi_0:A_0\longrightarrow A'_0 φ 0 : A 0 ⟶ A 0 ′ in the category AV ( S , O E 0 , ( ⋄ ) ) \operatorname{AV}(S,\mathcal{O}_{E_0,(\diamond)}) AV ( S , O E 0 , ( ⋄ ) ) satisfying that
φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 \varphi_0^\vee\circ\lambda'_0\circ\varphi_0 φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 is a prime-to-⋄ \diamond ⋄ multiple of λ 0 \lambda_0 λ 0 ;
the U ⋄ U^\diamond U ⋄ -orbits φ 0 ∗ ∘ η ˉ 0 ⋄ \varphi_{0*}\circ\bar{\eta}_0^\diamond φ 0 ∗ ∘ η ˉ 0 ⋄ and η ˉ 0 ′ ⋄ \bar{\eta}'^\diamond_0 η ˉ 0 ′ ⋄ coincide.
A morphism t ∈ U ′ ⋄ \ T ( A ∞ , ⋄ ) / U ⋄ t\in U'^\diamond\backslash T(\mathbb{A}^{\infty,\diamond})/U^\diamond t ∈ U ′ ⋄ \ T ( A ∞ , ⋄ ) / U ⋄ maps Y U ⋄ , ( ⋄ ) ( S ) Y_{U^\diamond,(\diamond)}(S) Y U ⋄ , ( ⋄ ) ( S ) to Y U ′ ⋄ , ( ⋄ ) ( S ) Y_{U'^\diamond,(\diamond)}(S) Y U ′ ⋄ , ( ⋄ ) ( S ) by changing η ˉ 0 ⋄ \bar{\eta}_0^\diamond η ˉ 0 ⋄ to η ˉ 0 ⋄ ∘ t \bar{\eta}_0^\diamond\circ t η ˉ 0 ⋄ ∘ t .
It is well-known that Y ∙ , ( ⋄ ) Y_{\bullet,(\diamond)} Y ∙ , ( ⋄ ) is representable and factors through the full subcategory Sch s m . p r o j / O E 0 , ( ⋄ ) \operatorname{Sch}^{\mathrm{sm.proj}}/\mathcal{O}_{E_0,(\diamond)} Sch sm.proj / O E 0 , ( ⋄ ) of PSch ′ / O E 0 , ( ⋄ ) \operatorname{PSch}'/\mathcal{O}_{E_0,(\diamond)} PSch ′ / O E 0 , ( ⋄ ) . Moreover, if we denote by
A 0 , U ⋄ , ( ⋄ ) ∈ AV ( Y U ⋄ , ( ⋄ ) , O E 0 , ( ⋄ ) ) A_{0,U^\diamond,(\diamond)}\in\operatorname{AV}(Y_{U^\diamond,(\diamond)},\mathcal{O}_{E_0,(\diamond)}) A 0 , U ⋄ , ( ⋄ ) ∈ AV ( Y U ⋄ , ( ⋄ ) , O E 0 , ( ⋄ ) ) the universal object, then we obtain a functor
A 0 , ∙ , ( ⋄ ) / Y ∙ , ( ⋄ ) : Hk ( T ( A ∞ , ⋄ ) ) ⟶ AV ( Sch s m . p r o j / O E 0 , ( ⋄ ) , O E 0 , ( ⋄ ) ) A_{0,\bullet,(\diamond)}/Y_{\bullet,(\diamond)}:\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond}))\longrightarrow\operatorname{AV}(\operatorname{Sch}^{\mathrm{sm.proj}}/\mathcal{O}_{E_0,(\diamond)},\mathcal{O}_{E_0,(\diamond)}) A 0 , ∙ , ( ⋄ ) / Y ∙ , ( ⋄ ) : Hk ( T ( A ∞ , ⋄ )) ⟶ AV ( Sch sm.proj / O E 0 , ( ⋄ ) , O E 0 , ( ⋄ ) ) over Y ∙ , ( ⋄ ) Y_{\bullet,(\diamond)} Y ∙ , ( ⋄ ) that is Cartesian.
Definition 3.42. We define a functor
X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) ⟶ PSch ′ / O E , ( ⋄ ) X_{\bullet,\bullet,(\diamond)}:\operatorname{Hk}(H(\mathbb{A}^{\infty,\diamond}_F))\times\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond}))\longrightarrow\operatorname{PSch}'/\mathcal{O}_{E,(\diamond)} X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) ⟶ PSch ′ / O E , ( ⋄ ) such that for every objects ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) (L^\diamond,U^\diamond)\in\operatorname{Hk}(H(\mathbb{A}^{\infty,\diamond}_F))\times\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) and S ∈ Sch ′ / O E , ( ⋄ ) S\in\operatorname{Sch}'/\mathcal{O}_{E,(\diamond)} S ∈ Sch ′ / O E , ( ⋄ ) , X L ⋄ , U ⋄ , ( ⋄ ) ( S ) X_{L^\diamond,U^\diamond,(\diamond)}(S) X L ⋄ , U ⋄ , ( ⋄ ) ( S ) is the set of equivalence classes of sextuples ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) (A_0,\lambda_0,\bar{\eta}_0^\diamond;A,\lambda,\bar{\eta}^\diamond) ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) , where
( A 0 , λ 0 , η ˉ 0 ⋄ ) (A_0,\lambda_0,\bar{\eta}_0^\diamond) ( A 0 , λ 0 , η ˉ 0 ⋄ ) is an object of Y U ⋄ , ( ⋄ ) ( S ) Y_{U^\diamond,(\diamond)}(S) Y U ⋄ , ( ⋄ ) ( S ) ;
A A A is an object of AV ( S , O E , ( ⋄ ) ) \operatorname{AV}(S,\mathcal{O}_{E,(\diamond)}) AV ( S , O E , ( ⋄ ) ) such that tr A ( e ) = m Tr E / E 0 ( e ) − e + e ˉ \operatorname{tr}_A(e)=m\operatorname{Tr}_{E/E_0}(e)-e+\bar{e} tr A ( e ) = m Tr E / E 0 ( e ) − e + e ˉ for e ∈ O E , ( ⋄ ) e\in\mathcal{O}_{E,(\diamond)} e ∈ O E , ( ⋄ ) ;
λ : A ⟶ A ∨ \lambda:A\longrightarrow A^\vee λ : A ⟶ A ∨ is a ⋄ \diamond ⋄ -principal polarization that turns the complex conjugation of O E , ( ⋄ ) \mathcal{O}_{E,(\diamond)} O E , ( ⋄ ) into the Rosati involution;
η ˉ ⋄ \bar{\eta}^\diamond η ˉ ⋄ is a hermitian L ⋄ L^\diamond L ⋄ -level structure, that is, for a chosen geometric point s s s on every connected component of S S S , a π 1 ( S , s ) \pi_1(S,s) π 1 ( S , s ) -invariant L ⋄ L^\diamond L ⋄ -orbit of isometries
η ⋄ : V ∞ , ⋄ ⟶ Hom A E 0 ∞ , ⋄ ( H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) ⊗ A E 0 ∞ , ⋄ A F ∞ , ⋄ , H 1 e ˊ t ( A s , A ∞ , ⋄ ) ) \eta^\diamond:V^{\infty,\diamond}\longrightarrow\operatorname{Hom}_{\mathbb{A}_{E_0}^{\infty,\diamond}}\left(\mathrm{H}_{1}^{\mathrm{\acute{e}t}}(A_{0s},\mathbb{A}^{\infty,\diamond})\otimes_{\mathbb{A}_{E_0}^{\infty,\diamond}}\mathbb{A}_F^{\infty,\diamond},\mathrm{H}_{1}^{\mathrm{\acute{e}t}}(A_s,\mathbb{A}^{\infty,\diamond})\right) η ⋄ : V ∞ , ⋄ ⟶ Hom A E 0 ∞ , ⋄ ( H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) ⊗ A E 0 ∞ , ⋄ A F ∞ , ⋄ , H 1 e ˊ t ( A s , A ∞ , ⋄ ) ) of hermitian spaces over A E ∞ , ⋄ / A F ∞ , ⋄ \mathbb{A}_E^{\infty,\diamond}/\mathbb{A}_F^{\infty,\diamond} A E ∞ , ⋄ / A F ∞ , ⋄ .
Two objects ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) (A_0,\lambda_0,\bar{\eta}_0^\diamond;A,\lambda,\bar{\eta}^\diamond) ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) and ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ; A ′ , λ ′ , η ˉ ′ ⋄ ) (A'_0,\lambda'_0,\bar{\eta}'^\diamond_0;A',\lambda',\bar{\eta}'^\diamond) ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ; A ′ , λ ′ , η ˉ ′ ⋄ ) are equivalent if there exist isomorphisms φ 0 : A 0 ⟶ A 0 ′ \varphi_0:A_0\longrightarrow A'_0 φ 0 : A 0 ⟶ A 0 ′ in the category AV ( S , O E 0 , ( ⋄ ) ) \operatorname{AV}(S,\mathcal{O}_{E_0,(\diamond)}) AV ( S , O E 0 , ( ⋄ ) ) and φ : A ⟶ A ′ \varphi:A\longrightarrow A' φ : A ⟶ A ′ in the category AV ( S , O E , ( ⋄ ) ) \operatorname{AV}(S,\mathcal{O}_{E,(\diamond)}) AV ( S , O E , ( ⋄ ) ) satisfying that
the U ⋄ U^{\diamond} U ⋄ -orbits φ 0 ∗ ∘ η ^ ⋄ \varphi_{0*} \circ\widehat{\eta}^{\diamond} φ 0 ∗ ∘ η ⋄ and η ^ 0 ⋄ \widehat{\eta}_{0}^{\diamond} η 0 ⋄ coincide;
the L ⋄ L^{\diamond} L ⋄ -orbit of maps v ↦ φ ∗ ∘ η ^ ⋄ ( v ) ∘ ( φ 0 ∗ ) − 1 v \mapsto\varphi_* \circ\widehat{\eta}^{\diamond}(v) \circ(\varphi_{0*})^{-1} v ↦ φ ∗ ∘ η ⋄ ( v ) ∘ ( φ 0 ∗ ) − 1 for v ∈ V ∞ , ⋄ v \in V^{\infty,\diamond} v ∈ V ∞ , ⋄ coincides with η ^ ⋄ \widehat{\eta}^{\diamond} η ⋄ .
A morphism ( h , t ) ∈ L ⋄ \ H ( A F ∞ , ⋄ ) / L ′ ⋄ × U ⋄ \ T ( A ∞ , ⋄ ) / U ′ ⋄ (h,t) \in L^{\diamond}\backslash H(\mathbb{A}_{F}^{\infty,\diamond})/L^{\prime\diamond} \times U^{\diamond}\backslash T(\mathbb{A}^{\infty,\diamond})/U^{\prime\diamond} ( h , t ) ∈ L ⋄ \ H ( A F ∞ , ⋄ ) / L ′ ⋄ × U ⋄ \ T ( A ∞ , ⋄ ) / U ′ ⋄ maps X L ⋄ , U ⋄ , ( ⋄ ) ( S ) X_{L^{\diamond},U^{\diamond},(\diamond)}(S) X L ⋄ , U ⋄ , ( ⋄ ) ( S ) to X L ′ ⋄ , U ′ ⋄ , ( ⋄ ) ( S ) X_{L^{\prime\diamond},U^{\prime\diamond},(\diamond)}(S) X L ′ ⋄ , U ′ ⋄ , ( ⋄ ) ( S ) by changing η ^ 0 ⋄ \widehat{\eta}_{0}^{\diamond} η 0 ⋄ to η ^ 0 ⋄ ∘ t \widehat{\eta}_{0}^{\diamond} \circ t η 0 ⋄ ∘ t and η ^ ⋄ \widehat{\eta}^{\diamond} η ⋄ to η ^ ⋄ ∘ h \widehat{\eta}^{\diamond} \circ h η ⋄ ∘ h .
Now we suppose that every prime in ⋄ \diamond ⋄ is unramified in E E E and V v V_v V v is split for every v ∈ O F ( ⋄ ) v \in\mathcal{O}_{F}^{(\diamond)} v ∈ O F ( ⋄ ) .
By [32 Theorem 4.4] , X ∙ , ∙ , ( ⋄ ) X_{\bullet,\bullet,(\diamond)} X ∙ , ∙ , ( ⋄ ) factors through the full subcategory Sch / O E , ( ⋄ ) s m . q p r o j \operatorname{Sch}^{\mathrm{sm.qproj}}_{/O_{E,(\diamond)}} Sch / O E , ( ⋄ ) sm.qproj of PSch / O E , ( ⋄ ) ′ \operatorname{PSch}^{\prime}_{/O_{E,(\diamond)}} PSch / O E , ( ⋄ ) ′ . Moreover, if we denote by
( A 0 , L ⋄ , U ⋄ , ( ⋄ ) , A L ⋄ , U ⋄ , ( ⋄ ) ) ∈ AV ( X L ⋄ , U ⋄ , ( ⋄ ) , O E 0 , ( ⋄ ) ) × AV ( X L ⋄ , U ⋄ , ( ⋄ ) , O E , ( ⋄ ) ) (A_{0,L^{\diamond},U^{\diamond},(\diamond)}, A_{L^{\diamond},U^{\diamond},(\diamond)}) \in\operatorname{AV}(X_{L^{\diamond},U^{\diamond},(\diamond)}, O_{E_0,(\diamond)}) \times\operatorname{AV}(X_{L^{\diamond},U^{\diamond},(\diamond)}, O_{E,(\diamond)}) ( A 0 , L ⋄ , U ⋄ , ( ⋄ ) , A L ⋄ , U ⋄ , ( ⋄ ) ) ∈ AV ( X L ⋄ , U ⋄ , ( ⋄ ) , O E 0 , ( ⋄ ) ) × AV ( X L ⋄ , U ⋄ , ( ⋄ ) , O E , ( ⋄ ) )
the universal object, then we obtain functors
A 0 , ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ ) ) × Hk ( T ( A ∞ ) ) → AV ( Sch / O E , ( ⋄ ) s m . q p r o j , O E 0 , ( ⋄ ) ) , A_{0,\bullet,\bullet,(\diamond)}/X_{\bullet,\bullet,(\diamond)}: \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) \times\operatorname{Hk}(T(\mathbb{A}^{\infty})) \to\operatorname{AV}(\operatorname{Sch}^{\mathrm{sm.qproj}}_{/O_{E,(\diamond)}}, O_{E_0,(\diamond)}), A 0 , ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) → AV ( Sch / O E , ( ⋄ ) sm.qproj , O E 0 , ( ⋄ ) ) ,
A ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ ) ) × Hk ( T ( A ∞ ) ) → AV ( Sch / O E , ( ⋄ ) s m . q p r o j , O E , ( ⋄ ) ) A_{\bullet,\bullet,(\diamond)}/X_{\bullet,\bullet,(\diamond)}: \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) \times\operatorname{Hk}(T(\mathbb{A}^{\infty})) \to\operatorname{AV}(\operatorname{Sch}^{\mathrm{sm.qproj}}_{/O_{E,(\diamond)}}, O_{E,(\diamond)}) A ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) → AV ( Sch / O E , ( ⋄ ) sm.qproj , O E , ( ⋄ ) )
over X ∙ , ∙ , ( ⋄ ) X_{\bullet,\bullet,(\diamond)} X ∙ , ∙ , ( ⋄ ) that are Cartesian. By definition, we have a natural morphism
X L ⋄ , U ⋄ , ( ⋄ ) → Y U ⋄ , ( ⋄ ) ⊗ O E 0 O E X_{L^{\diamond},U^{\diamond},(\diamond)} \to Y_{U^{\diamond},(\diamond)} \otimes_{O_{E_0}} O_E X L ⋄ , U ⋄ , ( ⋄ ) → Y U ⋄ , ( ⋄ ) ⊗ O E 0 O E
for every object ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ ) ) × Hk ( T ( A ∞ ) ) (L^{\diamond},U^{\diamond}) \in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) \times\operatorname{Hk}(T(\mathbb{A}^{\infty})) ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ )) × Hk ( T ( A ∞ )) , along which A 0 , L ⋄ , U ⋄ , ( ⋄ ) A_{0,L^{\diamond},U^{\diamond},(\diamond)} A 0 , L ⋄ , U ⋄ , ( ⋄ ) is the pullback of A 0 , U ⋄ , ( ⋄ ) ⊗ O E 0 O E A_{0,U^{\diamond},(\diamond)} \otimes_{O_{E_0}} O_E A 0 , U ⋄ , ( ⋄ ) ⊗ O E 0 O E to X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) .
If we choose an open compact subgroup L ⋄ L_{\diamond} L ⋄ of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) that is the stabilizer of a self-dual lattice of V ⋄ V_{\diamond} V ⋄ and denote by U ⋄ U_{\diamond} U ⋄ the maximal open compact subgroup of T ( Q ⋄ ) T(\mathbb{Q}_{\diamond}) T ( Q ⋄ ) , then it is clear from the construction that
X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) E = X L ⋄ L ⋄ , U ⋄ U ⋄ X_{L^{\diamond},U^{\diamond},(\diamond)} \otimes_{O_{E,(\diamond)}} E = X_{L_{\diamond}L^{\diamond},U_{\diamond}U^{\diamond}} X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) E = X L ⋄ L ⋄ , U ⋄ U ⋄
functorial in ( L ⋄ , U ⋄ ) (L^{\diamond},U^{\diamond}) ( L ⋄ , U ⋄ ) .
Remark 3.43. Take an object L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ ) ) L^{\diamond} \in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,\diamond})) L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ )) . For every U ⋄ ∈ Hk ( T ( A ∞ , ⋄ ) ) U^{\diamond} \in\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) U ⋄ ∈ Hk ( T ( A ∞ , ⋄ )) , the group T ( A ∞ , ⋄ ) / U ⋄ T(\mathbb{A}^{\infty,\diamond})/U^{\diamond} T ( A ∞ , ⋄ ) / U ⋄ acts on X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) via modifying η ^ 0 ⋄ \widehat{\eta}_{0}^{\diamond} η 0 ⋄ , which also factors through the quotient
Ξ U ⋄ : = O E 0 , ( ⋄ ) × \ T ( A ∞ , ⋄ ) / U ⋄ \Xi_{U^{\diamond}} := O_{E_0,(\diamond)}^{\times}\backslash T(\mathbb{A}^{\infty,\diamond})/U^{\diamond} Ξ U ⋄ := O E 0 , ( ⋄ ) × \ T ( A ∞ , ⋄ ) / U ⋄
by a way that is similar to the case for X L , U X_{L,U} X L , U , so that the action of Ξ U ⋄ \Xi_{U^{\diamond}} Ξ U ⋄ on X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) is free. Since X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) is smooth and quasi-projective over Spec O E , ( ⋄ ) \operatorname{Spec} O_{E,(\diamond)} Spec O E , ( ⋄ ) , the quotient X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ X_{L^{\diamond},U^{\diamond},(\diamond)}/\Xi_{U^{\diamond}} X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ is again a smooth and quasi-projective scheme over Spec O E , ( ⋄ ) \operatorname{Spec} O_{E,(\diamond)} Spec O E , ( ⋄ ) . It is clear that for U ′ ⋄ ⊆ U ⋄ U^{\prime\diamond} \subseteq U^{\diamond} U ′ ⋄ ⊆ U ⋄ in Hk ( T ( A ∞ , ⋄ ) ) \operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) Hk ( T ( A ∞ , ⋄ )) , the natural morphism X L ⋄ , U ′ ⋄ , ( ⋄ ) / Ξ U ′ ⋄ → X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ X_{L^{\diamond},U^{\prime\diamond},(\diamond)}/\Xi_{U^{\prime\diamond}} \to X_{L^{\diamond},U^{\diamond},(\diamond)}/\Xi_{U^{\diamond}} X L ⋄ , U ′ ⋄ , ( ⋄ ) / Ξ U ′ ⋄ → X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ is an isomorphism.
To summarize, we obtain a functor
X ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ ) ) → Sch / O E , ( ⋄ ) s m . q p r o j X_{\bullet,(\diamond)}: \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,\diamond})) \to\operatorname{Sch}^{\mathrm{sm.qproj}}_{/O_{E,(\diamond)}} X ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ )) → Sch / O E , ( ⋄ ) sm.qproj
such that for every L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ ) ) L^{\diamond} \in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,\diamond})) L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ )) , X L ⋄ , ( ⋄ ) = X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ X_{L^{\diamond},(\diamond)} = X_{L^{\diamond},U^{\diamond},(\diamond)}/\Xi_{U^{\diamond}} X L ⋄ , ( ⋄ ) = X L ⋄ , U ⋄ , ( ⋄ ) / Ξ U ⋄ – the quotient scheme obtained from the above process, which is canonically independent of the choice of U ⋄ ∈ Hk ( T ( A ∞ , ⋄ ) ) U^{\diamond} \in\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) U ⋄ ∈ Hk ( T ( A ∞ , ⋄ )) . By construction, it is clear that
X L ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) E = X L ⋄ L ⋄ X_{L^{\diamond},(\diamond)} \otimes_{O_{E,(\diamond)}} E = X_{L_{\diamond}L^{\diamond}} X L ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) E = X L ⋄ L ⋄
functorial in L ⋄ L^{\diamond} L ⋄ .
Height pairing on motivic cohomology
In this chapter, we construct and study a height pairing on motivic cohomology of those automorphic motives constructed in the previous chapter. We
fix a totally positive definite incoherent hermitian space V V V over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F of rank m ≥ 1 m \ge1 m ≥ 1 satisfying that the set V ( V ) : = { v ∈ V F ∣ V v is anisotropic } \mathcal{V}(\mathbb{V}) := \{v \in\mathbb{V}_F \mid V_v\text{ is anisotropic}\} V ( V ) := { v ∈ V F ∣ V v is anisotropic } , which contains V F ( ∞ ) \mathbb{V}_F^{(\infty)} V F ( ∞ ) , is not a singleton (in particular, the E E E -schemes { X L } L \{X_L\}_L { X L } L are projective);
fix a hermitian E E E -weight λ \lambda λ of rank m m m (Definition 2.7).
Put H : = U ( V ) H := \mathrm{U}(V) H := U ( V ) . In what follows, we denote by − u : S c h / E → S c h / E u -_{u} : \mathrm{Sch}/E \to\mathrm{Sch}/E_u − u : Sch / E → Sch / E u the base change functor for every place u u u of E E E .
Definition of height pairing
Take a rational prime ℓ \ell ℓ .
Notation 4.1. For integers q , r q,r q , r , an object L ∈ H k ( H ( A F ∞ ) ) L \in\mathrm{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) , and a Zariski closed subset Z Z Z of X L X_L X L , we put
H Z q ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ : = ker ( H Z q ( X L , M L , λ ( r ) ) ⟶ ∏ u ∈ V E f i n ∖ V E ( ℓ ) H e ˊ t q ( X L , u , M L , ℓ , λ ( r ) ) ) . \mathrm{H}^{q}_{Z}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle} := \ker\left(\mathrm{H}^{q}_{Z}(X_L,\mathcal{M}_{L,\lambda}(r)) \longrightarrow\prod_{u\in\mathcal{V}_E^{\mathrm{fin}}\setminus\mathcal{V}_E^{(\ell)}} \mathrm{H}^{q}_{\mathrm{\acute{e}t}}(X_{L,u},\mathcal{M}_{L,\ell,\lambda}(r))\right). H Z q ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ := ker H Z q ( X L , M L , λ ( r )) ⟶ u ∈ V E fin ∖ V E ( ℓ ) ∏ H e ˊ t q ( X L , u , M L , ℓ , λ ( r )) .
Suppose that ℓ \ell ℓ is unramified in E E E . The goal of this section is to define, for every object L ∈ H k ( H ( A F ∞ ) ) L \in\mathrm{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) such that L ℓ L_\ell L ℓ is the stabilizer of a self-dual lattice of V ℓ V_\ell V ℓ , a canonical hermitian24 pairing
⟨ , ⟩ λ , L ℓ : ( ⨁ r = 1 m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) × ( ⨁ r = 1 m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) → C ⊗ Q ℓ (38) \langle\ ,\ \rangle_{\lambda,L}^{\ell} : \left(\bigoplus_{r=1}^{m-1}\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\right)\times\left(\bigoplus_{r=1}^{m-1}\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\right)\to\mathbb{C}\otimes\mathbb{Q}_{\ell} \tag*{(38)} ⟨ , ⟩ λ , L ℓ : ( r = 1 ⨁ m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) × ( r = 1 ⨁ m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) → C ⊗ Q ℓ (38)
that vanishes on H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ \mathrm{H}^{2r_1}(X_L,\mathcal{M}_{L,\lambda}(r_1))^{\langle\ell\rangle}\times\mathrm{H}^{2r_2}(X_L,\mathcal{M}_{L,\lambda}(r_2))^{\langle\ell\rangle} H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ unless when r 1 + r 2 = m r_1+r_2=m r 1 + r 2 = m .
Choose an object U U U of H k ( T ( A ∞ ) ) \mathrm{Hk}(T(\mathbb{A}^\infty)) Hk ( T ( A ∞ )) such that U ℓ U_\ell U ℓ is maximal so that we have the natural quotient map π L , U : X L , U → X L \pi_{L,U}: X_{L,U}\to X_L π L , U : X L , U → X L (6) . Put M L , U , λ : = π L , U ∗ M L , λ \mathcal{M}_{L,U,\lambda}:=\pi_{L,U}^{*}\mathcal{M}_{L,\lambda} M L , U , λ := π L , U ∗ M L , λ as in §3.4. The straightening map s t L , U λ , + \mathrm{st}_{L,U}^{\lambda,+} st L , U λ , + and complementary straightening map s t L , U λ , − \mathrm{st}_{L,U}^{\lambda,-} st L , U λ , − (Definition 3.17) induce E λ E^\lambda E λ -linear maps (with the same notation)
s t L , U λ , + : H 2 r ( X L , U , M L , U , λ ( r ) ) → H 2 r + 2 ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ∣ b ∣ ) ) , \mathrm{st}_{L,U}^{\lambda,+}:\mathrm{H}^{2r}(X_{L,U},\mathcal{M}_{L,U,\lambda}(r))\to\mathrm{H}^{2r+2|b|}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(r+|b|)), st L , U λ , + : H 2 r ( X L , U , M L , U , λ ( r )) → H 2 r + 2∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ∣ b ∣ )) ,
s t L , U λ , − : H 2 r ( X L , U , M L , U , λ ( r ) ) → H 2 r + 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ( g − 1 ) ∣ b ∣ ) ) , \mathrm{st}_{L,U}^{\lambda,-}:\mathrm{H}^{2r}(X_{L,U},\mathcal{M}_{L,U,\lambda}(r))\to\mathrm{H}^{2r+2(g-1)|b|}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(r+(g-1)|b|)), st L , U λ , − : H 2 r ( X L , U , M L , U , λ ( r )) → H 2 r + 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ( g − 1 ) ∣ b ∣ )) ,
respectively. For every term H H H in the above two maps, one may define the subspace H ⟨ ℓ ⟩ H^{\langle\ell\rangle} H ⟨ ℓ ⟩ similarly as in Notation 4.1. It follows easily that we have induced E λ E^\lambda E λ -linear maps
s t L , U λ , + ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → H 2 r + 2 ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ∣ b ∣ ) ) ⟨ ℓ ⟩ , \mathrm{st}_{L,U}^{\lambda,+}\circ\pi_{L,U}^{*}:\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\to\mathrm{H}^{2r+2|b|}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(r+|b|))^{\langle\ell\rangle}, st L , U λ , + ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → H 2 r + 2∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ∣ b ∣ ) ) ⟨ ℓ ⟩ ,
s t L , U λ , − ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → H 2 r + 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ( g − 1 ) ∣ b ∣ ) ) ⟨ ℓ ⟩ , \mathrm{st}_{L,U}^{\lambda,-}\circ\pi_{L,U}^{*}:\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\to\mathrm{H}^{2r+2(g-1)|b|}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(r+(g-1)|b|))^{\langle\ell\rangle}, st L , U λ , − ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → H 2 r + 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( r + ( g − 1 ) ∣ b ∣ ) ) ⟨ ℓ ⟩ ,
respectively. Note that for every integer s ≥ 0 s\ge0 s ≥ 0 , under the natural isomorphism
H 2 s ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( s ) ) = C H s ( B L , U ∣ b ∣ ) ⊗ E λ , \mathrm{H}^{2s}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(s))=\mathrm{CH}^{s}(B_{L,U}^{|b|})\otimes E^\lambda, H 2 s ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( s )) = CH s ( B L , U ∣ b ∣ ) ⊗ E λ ,
the subspace H 2 s ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( s ) ) ⟨ ℓ ⟩ \mathrm{H}^{2s}(B_{L,U}^{|b|},\mathbb{1}_{B_{L,U}^{|b|},E^\lambda}(s))^{\langle\ell\rangle} H 2 s ( B L , U ∣ b ∣ , 1 B L , U ∣ b ∣ , E λ ( s ) ) ⟨ ℓ ⟩ coincides with C H s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ \mathrm{CH}^{s}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda CH s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ , where C H s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ \mathrm{CH}^{s}(B_{L,U}^{|b|})^{\langle\ell\rangle} CH s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ was defined in [20 Notation 2.5(C3)] . In summary, we have induced E λ E^\lambda E λ -linear maps
s t L , U λ , + ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → C H r + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ , \mathrm{st}_{L,U}^{\lambda,+}\circ\pi_{L,U}^{*}:\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\to\mathrm{CH}^{r+|b|}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda, st L , U λ , + ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → CH r + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ,
s t L , U λ , − ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → C H r + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ , \mathrm{st}_{L,U}^{\lambda,-}\circ\pi_{L,U}^{*}:\mathrm{H}^{2r}(X_L,\mathcal{M}_{L,\lambda}(r))^{\langle\ell\rangle}\to\mathrm{CH}^{r+(g-1)|b|}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda, st L , U λ , − ∘ π L , U ∗ : H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ → CH r + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ,
respectively.
By the discussion in §3.8, we know that B L , U ∣ b ∣ B^{|b|}_{L,U} B L , U ∣ b ∣ has smooth projective reduction at every ℓ \ell ℓ -adic place of E E E . Thus, we may apply Beilinson’s construction of the height pairing in [4 §4] (see also [20 §6] ) to obtain a map
⟨ , ⟩ B L , U ∣ b ∣ ℓ : ( ⨁ s = 1 ∣ g ∣ b ∣ + m − 1 ∣ C H s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C ) × ( ⨁ s = 1 ∣ g ∣ b ∣ + m − 1 ∣ C H s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C ) ⟶ C ⊗ Q ℓ \langle\ ,\ \rangle_{B^{|b|}_{L,U}}^\ell:\left(\bigoplus_{s=1}^{|g|b|+m-1|}\mathrm{CH}^{s}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{C}\right)\times\left(\bigoplus_{s=1}^{|g|b|+m-1|}\mathrm{CH}^{s}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{C}\right)\longrightarrow\mathbb{C}\otimes\mathbb{Q}_{\ell} ⟨ , ⟩ B L , U ∣ b ∣ ℓ : s = 1 ⨁ ∣ g ∣ b ∣ + m − 1∣ CH s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C × s = 1 ⨁ ∣ g ∣ b ∣ + m − 1∣ CH s ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C ⟶ C ⊗ Q ℓ
that is hermitian and vanishes on C H s 1 ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C × C H s 2 ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C \mathrm{CH}^{s_1}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{C}\times\mathrm{CH}^{s_2}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{C} CH s 1 ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C × CH s 2 ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ C unless when s 1 + s 2 = g ∣ b ∣ + m = dim B L , U ∣ b ∣ + 1 s_1+s_2=g|b|+m=\dim B^{|b|}_{L,U}+1 s 1 + s 2 = g ∣ b ∣ + m = dim B L , U ∣ b ∣ + 1 . Put
⟨ , ⟩ λ , L ℓ : = 1 [ Ξ U ] ⋅ ⟨ , ⟩ B L , U ∣ b ∣ , E ℓ ∘ ( ( st L , U ℓ , + ∘ π L , U ∗ ) × ( st L , U ℓ , − ∘ π L , U ∗ ) ) \langle\ ,\ \rangle_{\lambda,L}^\ell:=\frac{1}{[\Xi_U]}\cdot\langle\ ,\ \rangle_{B^{|b|}_{L,U},E}^\ell\circ\left(\left(\operatorname{st}_{L,U}^{\ell,+}\circ\pi_{L,U}^{*}\right)\times\left(\operatorname{st}_{L,U}^{\ell,-}\circ\pi_{L,U}^{*}\right)\right) ⟨ , ⟩ λ , L ℓ := [ Ξ U ] 1 ⋅ ⟨ , ⟩ B L , U ∣ b ∣ , E ℓ ∘ ( ( st L , U ℓ , + ∘ π L , U ∗ ) × ( st L , U ℓ , − ∘ π L , U ∗ ) )
: ( ⨁ r = 1 m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) × ( ⨁ r = 1 m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) ⟶ C ⊗ Q ℓ . :\left(\bigoplus_{r=1}^{m-1}\mathrm{H}^{2r}(X_L,M_{L,\lambda}(r))^{\langle\ell\rangle}\right)\times\left(\bigoplus_{r=1}^{m-1}\mathrm{H}^{2r}(X_L,M_{L,\lambda}(r))^{\langle\ell\rangle}\right)\longrightarrow\mathbb{C}\otimes\mathbb{Q}_{\ell}. : ( r = 1 ⨁ m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) × ( r = 1 ⨁ m − 1 H 2 r ( X L , M L , λ ( r ) ) ⟨ ℓ ⟩ ) ⟶ C ⊗ Q ℓ .
Lemma 4.2. The map ⟨ , ⟩ λ , L ℓ \langle\ ,\ \rangle_{\lambda,L}^\ell ⟨ , ⟩ λ , L ℓ does not depend on the choice of U U U (hence justifying the disappearance of U U U in the notation). Moreover,
(1) it is E ℓ \mathbb{E}^{\ell} E ℓ -linear in the first variable and complex-conjugate symmetric;
(2) it vanishes on H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ \mathrm{H}^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times\mathrm{H}^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle} H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ unless when r 1 + r 2 = m r_1+r_2=m r 1 + r 2 = m .
Proof. Note that for an open compact subgroup U ′ ⊆ U U'\subseteq U U ′ ⊆ U with U ℓ ′ = U ℓ U'_{\ell}=U_{\ell} U ℓ ′ = U ℓ , the natural transition map B L , U ′ ∣ b ∣ → B L , U ∣ b ∣ B^{|b|}_{L,U'}\to B^{|b|}_{L,U} B L , U ′ ∣ b ∣ → B L , U ∣ b ∣ is finite étale of degree [ Ξ U ′ / Ξ U ] [\Xi_{U'}/\Xi_U] [ Ξ U ′ / Ξ U ] . Then the independence of U U U of ⟨ , ⟩ λ , L ℓ \langle\ ,\ \rangle_{\lambda,L}^\ell ⟨ , ⟩ λ , L ℓ follows from the functoriality of Beilinson’s height pairing [4 4.0.3] .
Part (2) follows from the similar property for ⟨ , ⟩ B L , U ∣ b ∣ , E ℓ \langle\ ,\ \rangle_{B^{|b|}_{L,U},E}^\ell ⟨ , ⟩ B L , U ∣ b ∣ , E ℓ .
For part (1), the E ℓ \mathbb{E}^{\ell} E ℓ -linearity in the first variable is obvious. It remains to show the complex-conjugate symmetry, that is, for every pair ( c 1 , c 2 ) ∈ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ (c_1,c_2)\in\mathrm{H}^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times\mathrm{H}^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle} ( c 1 , c 2 ) ∈ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ with r 1 + r 2 = m r_1+r_2=m r 1 + r 2 = m , we have
⟨ c 2 , c 1 ⟩ λ , L ℓ = ⟨ c 1 , c 2 ⟩ λ , L ℓ ‾ . \langle c_2,c_1\rangle_{\lambda,L}^\ell=\overline{\langle c_1,c_2\rangle_{\lambda,L}^\ell}. ⟨ c 2 , c 1 ⟩ λ , L ℓ = ⟨ c 1 , c 2 ⟩ λ , L ℓ .
For i = 1 , 2 i=1,2 i = 1 , 2 , put c i ± : = st L , U ℓ , ± ( π L , U ∗ ( c i ) ) c_i^{\pm}:=\operatorname{st}_{L,U}^{\ell,\pm}(\pi_{L,U}^{*}(c_i)) c i ± := st L , U ℓ , ± ( π L , U ∗ ( c i )) . It suffices to show that
⟨ c 2 + , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ ‾ . \langle c_2^{+},c_1^{-}\rangle_{B^{|b|}_{L,U}}^\ell=\overline{\langle c_1^{+},c_2^{-}\rangle_{B^{|b|}_{L,U}}^\ell}. ⟨ c 2 + , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ .
For this, note that the map
C H r + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ Q ⟶ C H r + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ Q \mathrm{CH}^{r+|b|}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{Q}\longrightarrow\mathrm{CH}^{r+(g-1)|b|}(B^{|b|}_{L,U})^{\langle\ell\rangle}\otimes\mathbb{Q} CH r + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ Q ⟶ CH r + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ Q
for every integer r ≥ 0 r\geq0 r ≥ 0 induced from the map
( ( λ 0 × λ − 1 ) ∣ b ∣ ) ∗ ∘ F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ : ⟨ ∣ b ∣ ⟩ : h 2 ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⟶ h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ( g − 1 ) ∣ b ∣ ⟩ ((\lambda_0\times\lambda^{-1})|_{b|})^{*}\circ F_{B^{|b|}_{L,U}/X_{L,U}}^{2(g-1)|b|}\colon\langle|b|\rangle\colon h^{2|b|}(B^{|b|}_{L,U}/X_{L,U},\mathbb{Q})\langle|b|\rangle\longrightarrow h^{2(g-1)|b|}(B^{|b|}_{L,U}/X_{L,U},\mathbb{Q})\langle(g-1)|b|\rangle (( λ 0 × λ − 1 ) ∣ b ∣ ) ∗ ∘ F B L , U ∣ b ∣ / X L , U 2 ( g − 1 ) ∣ b ∣ : ⟨ ∣ b ∣ ⟩ : h 2∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨ ∣ b ∣ ⟩ ⟶ h 2 ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ / X L , U , Q ) ⟨( g − 1 ) ∣ b ∣ ⟩
of relative motives over X L , U X_{L,U} X L , U is simply the pullback given by the correspondence p ⊗ 2 ( g − 1 ) ∣ b ∣ p^{\otimes2(g-1)|b|} p ⊗ 2 ( g − 1 ) ∣ b ∣ , where p ∈ C H 1 ( B L , U ∣ b ∣ × X L , U B L , U ∣ b ∣ ) ⊗ Q p\in\mathrm{CH}^{1}(B^{|b|}_{L,U}\times_{X_{L,U}}B^{|b|}_{L,U})\otimes\mathbb{Q} p ∈ CH 1 ( B L , U ∣ b ∣ × X L , U B L , U ∣ b ∣ ) ⊗ Q is the relative Poincaré line bundle (after identifying C H 1 ( B L , U ∣ b ∣ × X L , U B L , U ∣ b ∣ ) ⊗ Q \mathrm{CH}^{1}(B^{|b|}_{L,U}\times_{X_{L,U}}B^{|b|}_{L,U})\otimes\mathbb{Q} CH 1 ( B L , U ∣ b ∣ × X L , U B L , U ∣ b ∣ ) ⊗ Q with C H 1 ( B L , U ∣ b ∣ × X L , U B L , U ∨ ∣ b ∣ ) ⊗ Q \mathrm{CH}^{1}(B^{|b|}_{L,U}\times_{X_{L,U}}B^{|b|}_{L,U^{\vee}})\otimes\mathbb{Q} CH 1 ( B L , U ∣ b ∣ × X L , U B L , U ∨ ∣ b ∣ ) ⊗ Q via ( λ 0 × λ − 1 ) ∣ b ∣ (\lambda_0\times\lambda^{-1})|_{b|} ( λ 0 × λ − 1 ) ∣ b ∣ ). In particular, c i − = ( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c i + c_i^{-}=(p^{\otimes2(g-1)|b|})^{*}c_i^{+} c i − = ( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c i + for i = 1 , 2 i=1,2 i = 1 , 2 by Definition 3.17. On the other hand, p p p is symmetric, which implies that
⟨ c 2 + , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ ( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c 2 − , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 2 − , ( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 2 , c 1 + ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ ‾ , \langle c_2^{+},c_1^{-}\rangle_{B^{|b|}_{L,U}}^\ell=\langle(p^{\otimes2(g-1)|b|})^{*}c_2^{-},c_1^{-}\rangle_{B^{|b|}_{L,U}}^\ell=\langle c_2^{-},(p^{\otimes2(g-1)|b|})^{*}c_1^{-}\rangle_{B^{|b|}_{L,U}}^\ell=\langle c_2,c_1^{+}\rangle_{B^{|b|}_{L,U}}^\ell=\overline{\langle c_1^{+},c_2^{-}\rangle_{B^{|b|}_{L,U}}^\ell}, ⟨ c 2 + , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c 2 − , c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 2 − , ( p ⊗ 2 ( g − 1 ) ∣ b ∣ ) ∗ c 1 − ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 2 , c 1 + ⟩ B L , U ∣ b ∣ ℓ = ⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ ,
in which the second identity uses [4 4.0.3] .
The lemma is proved.
We end this section by the following easy lemma.
Lemma 4.3. For another object L ′ ⊆ L L' \subseteq L L ′ ⊆ L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) Hk ( H ( A F ∞ )) satisfying L ℓ ′ = L ℓ L'_{\ell}=L_{\ell} L ℓ ′ = L ℓ , if we denote by τ L ′ , L : X L ′ → X L \tau_{L',L}:X_{L'}\to X_L τ L ′ , L : X L ′ → X L the transition map, then
⟨ τ L ′ , L ∗ ( c 1 ) , τ L ′ , L ∗ ( c 2 ) ⟩ λ , L ℓ = ∣ L / L ′ ∣ ⋅ ⟨ c 1 , c 2 ⟩ λ , L ′ ℓ \left\langle\tau_{L',L}^{*}(c_1),\tau_{L',L}^{*}(c_2)\right\rangle_{\lambda,L}^{\ell}=|L/L'|\cdot\langle c_1,c_2\rangle_{\lambda,L'}^{\ell} ⟨ τ L ′ , L ∗ ( c 1 ) , τ L ′ , L ∗ ( c 2 ) ⟩ λ , L ℓ = ∣ L / L ′ ∣ ⋅ ⟨ c 1 , c 2 ⟩ λ , L ′ ℓ holds for every pair ( c 1 , c 2 ) ∈ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ (c_1,c_2)\in\mathrm{H}^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times\mathrm{H}^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle} ( c 1 , c 2 ) ∈ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ .
Proof. The lemma follows from [4 4.0.3] and the fact that the natural morphism X L ′ , U → X L , U X_{L',U}\to X_{L,U} X L ′ , U → X L , U is finite étale of degree ∣ L / L ′ ∣ |L/L'| ∣ L / L ′ ∣ for every object U U U of Hk ( T ( A F ∞ ) ) \operatorname{Hk}(T(\mathbb{A}_{F}^{\infty})) Hk ( T ( A F ∞ )) such that U ℓ U_{\ell} U ℓ is maximal. □
Decomposition into local heights
Let ℓ \ell ℓ be an arbitrary rational prime. Take an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) L ∈ Hk ( H ( A F ∞ )) . Fix a pair ( r 1 , r 2 ) (r_1,r_2) ( r 1 , r 2 ) of positive integers satisfying r 1 + r 2 = m r_1+r_2=m r 1 + r 2 = m and a pair Z = ( Z 1 , Z 2 ) Z=(Z_1,Z_2) Z = ( Z 1 , Z 2 ) of disjoint Zariski closed subsets of X L X_L X L . In this section, we define local height pairings on H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ \mathrm{H}^{2r_1}_{Z_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times\mathrm{H}^{2r_2}_{Z_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle} H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ indexed by elements in ( V E f i n ∖ V E ( ℓ ) ) ∪ Hom ( E , C ) (\mathcal{V}_{E}^{\mathrm{fin}}\setminus\mathcal{V}_{E}^{(\ell)})\cup\operatorname{Hom}(E,\mathbb{C}) ( V E fin ∖ V E ( ℓ ) ) ∪ Hom ( E , C ) . Write U i : = X L ∖ Z i U_i:=X_L\setminus Z_i U i := X L ∖ Z i for i = 1 , 2 i=1,2 i = 1 , 2 .
We start from the nonarchimedean case. Take an element u ∈ V E f i n ∖ V E ( ℓ ) u\in\mathcal{V}_{E}^{\mathrm{fin}}\setminus\mathcal{V}_{E}^{(\ell)} u ∈ V E fin ∖ V E ( ℓ ) , and we will define a pairing
⟨ , ⟩ λ , L , Z , u ℓ : H e ˊ t , Z 1 , u 2 r 1 ( X L , u , M L , ℓ , λ ( r 1 ) ) 0 × H e ˊ t , Z 2 , u 2 r 2 ( X L , u , M L , ℓ , λ ( r 2 ) ) 0 → E u ℓ , \langle\ ,\ \rangle_{\lambda,L,Z,u}^{\ell}:\mathrm{H}^{2r_1}_{\mathrm{\acute{e}t},Z_1,u}(X_{L,u},M_{L,\ell,\lambda}(r_1))^{0}\times\mathrm{H}^{2r_2}_{\mathrm{\acute{e}t},Z_2,u}(X_{L,u},M_{L,\ell,\lambda}(r_2))^{0}\to E_{u}^{\ell}, ⟨ , ⟩ λ , L , Z , u ℓ : H e ˊ t , Z 1 , u 2 r 1 ( X L , u , M L , ℓ , λ ( r 1 ) ) 0 × H e ˊ t , Z 2 , u 2 r 2 ( X L , u , M L , ℓ , λ ( r 2 ) ) 0 → E u ℓ ,
in which
H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i ) ) 0 : = ker ( H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i ) ) → H e ˊ t 2 r i ( X L , u , M L , ℓ , λ ( r i ) ) ) , \mathrm{H}^{2r_i}_{\mathrm{\acute{e}t},Z_i,u}(X_{L,u},M_{L,\ell,\lambda}(r_i))^{0}:=\operatorname{ker}\left(\mathrm{H}^{2r_i}_{\mathrm{\acute{e}t},Z_i,u}(X_{L,u},M_{L,\ell,\lambda}(r_i))\to\mathrm{H}^{2r_i}_{\mathrm{\acute{e}t}}(X_{L,u},M_{L,\ell,\lambda}(r_i))\right), H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i ) ) 0 := ker ( H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i )) → H e ˊ t 2 r i ( X L , u , M L , ℓ , λ ( r i )) ) ,
as follows. Take a pair ( c 1 , c 2 ) (c_1,c_2) ( c 1 , c 2 ) from the source. By the Gysin long exact sequence, we may choose an element γ i ∈ H e ˊ t 2 r i − 1 ( U i , u , M L , ℓ , λ ( r i ) ) \gamma_i\in\mathrm{H}^{2r_i-1}_{\mathrm{\acute{e}t}}(U_{i,u},M_{L,\ell,\lambda}(r_i)) γ i ∈ H e ˊ t 2 r i − 1 ( U i , u , M L , ℓ , λ ( r i )) that maps to c i c_i c i under the coboundary map
H e ˊ t 2 r i − 1 ( U i , u , M L , ℓ , λ ( r i ) ) → H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i ) ) \mathrm{H}^{2r_i-1}_{\mathrm{\acute{e}t}}(U_{i,u},M_{L,\ell,\lambda}(r_i))\to\mathrm{H}^{2r_i}_{\mathrm{\acute{e}t},Z_i,u}(X_{L,u},M_{L,\ell,\lambda}(r_i)) H e ˊ t 2 r i − 1 ( U i , u , M L , ℓ , λ ( r i )) → H e ˊ t , Z i , u 2 r i ( X L , u , M L , ℓ , λ ( r i ))
for i = 1 , 2 i=1,2 i = 1 , 2 . We then have the cup product
γ 1 ∪ γ 2 ∈ H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , M L , ℓ , λ ( r 1 ) ⊗ E ℓ λ M L , ℓ , λ ( r 2 ) ) . \gamma_1\cup\gamma_2\in\mathrm{H}^{2m-2}_{\mathrm{\acute{e}t}}(U_{1,u}\cap U_{2,u},M_{L,\ell,\lambda}(r_1)\otimes_{E_{\ell}^{\lambda}}M_{L,\ell,\lambda}(r_2)). γ 1 ∪ γ 2 ∈ H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , M L , ℓ , λ ( r 1 ) ⊗ E ℓ λ M L , ℓ , λ ( r 2 )) .
If we denote by γ 1 ∨ γ 2 \gamma_1\vee\gamma_2 γ 1 ∨ γ 2 the image of γ 1 ∪ γ 2 \gamma_1\cup\gamma_2 γ 1 ∪ γ 2 under the map
H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , M L , ℓ , λ ( r 1 ) ⊗ E ℓ λ M L , ℓ , λ ( r 2 ) ) → H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , E u ℓ ( m ) ) \mathrm{H}^{2m-2}_{\mathrm{\acute{e}t}}(U_{1,u}\cap U_{2,u},M_{L,\ell,\lambda}(r_1)\otimes_{E_{\ell}^{\lambda}}M_{L,\ell,\lambda}(r_2))\to\mathrm{H}^{2m-2}_{\mathrm{\acute{e}t}}(U_{1,u}\cap U_{2,u},E_{u}^{\ell}(m)) H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , M L , ℓ , λ ( r 1 ) ⊗ E ℓ λ M L , ℓ , λ ( r 2 )) → H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , E u ℓ ( m ))
induced by (3.21), then ⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ \langle c_1,c_2\rangle_{\lambda,L,Z,u}^{\ell} ⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ is defined to be the image of γ 1 ∨ γ 2 \gamma_1\vee\gamma_2 γ 1 ∨ γ 2 under the composite map
H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , E u ℓ ( m ) ) → H e ˊ t 2 m − 1 ( X L , u , E u ℓ ( m ) ) → H 1 ( Spec E u , E u ℓ ( 1 ) ) = E u ℓ , \mathrm{H}^{2m-2}_{\mathrm{\acute{e}t}}(U_{1,u}\cap U_{2,u},E_{u}^{\ell}(m))\to\mathrm{H}^{2m-1}_{\mathrm{\acute{e}t}}(X_{L,u},E_{u}^{\ell}(m))\to\mathrm{H}^{1}(\operatorname{Spec}E_u,E_{u}^{\ell}(1))=E_{u}^{\ell}, H e ˊ t 2 m − 2 ( U 1 , u ∩ U 2 , u , E u ℓ ( m )) → H e ˊ t 2 m − 1 ( X L , u , E u ℓ ( m )) → H 1 ( Spec E u , E u ℓ ( 1 )) = E u ℓ ,
in which the first map is the coboundary map in the Mayer–Vietoris exact sequence for the Zariski open covering X L , u = U 1 , u ∪ U 2 , u X_{L,u}=U_{1,u}\cup U_{2,u} X L , u = U 1 , u ∪ U 2 , u , and the second map is the trace map for the proper smooth structural morphism X L , u → Spec E u X_{L,u}\to\operatorname{Spec}E_u X L , u → Spec E u . It is easy to check that ⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ \langle c_1,c_2\rangle_{\lambda,L,Z,u}^{\ell} ⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ does not depend on the choices of γ 1 , γ 2 \gamma_1,\gamma_2 γ 1 , γ 2 .
We then consider the archimedean case. Take an element u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , and we will define a pairing
⟨ , ⟩ λ , L , Z u : H h s , Z 1 u 2 r 1 ( X L u , M L , λ u ( r 1 ) ) 0 × H h s , Z 2 u 2 r 2 ( X L u , M L , λ u ( r 2 ) ) 0 → C , \langle\ ,\ \rangle_{\lambda,L,Z}^{u}:\mathrm{H}^{2r_1}_{\mathrm{hs},Z_1^{u}}(X_L^{u},M_{L,\lambda}^{u}(r_1))^{0}\times\mathrm{H}^{2r_2}_{\mathrm{hs},Z_2^{u}}(X_L^{u},M_{L,\lambda}^{u}(r_2))^{0}\to\mathbb{C}, ⟨ , ⟩ λ , L , Z u : H hs , Z 1 u 2 r 1 ( X L u , M L , λ u ( r 1 ) ) 0 × H hs , Z 2 u 2 r 2 ( X L u , M L , λ u ( r 2 ) ) 0 → C ,
in which
H h s , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) 0 : = ker ( H h s , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) → H h s 2 r i ( X L u , M L , λ u ( r i ) ) ) , \mathrm{H}^{2r_i}_{\mathrm{hs},Z_i^{u}}(X_L^{u},M_{L,\lambda}^{u}(r_i))^{0}:=\operatorname{ker}\left(\mathrm{H}^{2r_i}_{\mathrm{hs},Z_i^{u}}(X_L^{u},M_{L,\lambda}^{u}(r_i))\to\mathrm{H}^{2r_i}_{\mathrm{hs}}(X_L^{u},M_{L,\lambda}^{u}(r_i))\right), H hs , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) 0 := ker ( H hs , Z i u 2 r i ( X L u , M L , λ u ( r i )) → H hs 2 r i ( X L u , M L , λ u ( r i )) ) ,
as follows. Take a pair ( c 1 , c 2 ) (c_1,c_2) ( c 1 , c 2 ) from the source. By the Gysin long exact sequence, we may choose an element γ i ∈ H h s 2 r i − 1 ( U i u , M L , λ u ( r i ) ) \gamma_i\in\mathrm{H}^{2r_i-1}_{\mathrm{hs}}(U_i^{u},M_{L,\lambda}^{u}(r_i)) γ i ∈ H hs 2 r i − 1 ( U i u , M L , λ u ( r i )) that maps to c i c_i c i under the coboundary map
H h s 2 r i − 1 ( U i u , M L , λ u ( r i ) ) → H h s , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) \mathrm{H}^{2r_i-1}_{\mathrm{hs}}(U_i^{u},M_{L,\lambda}^{u}(r_i))\to\mathrm{H}^{2r_i}_{\mathrm{hs},Z_i^{u}}(X_L^{u},M_{L,\lambda}^{u}(r_i)) H hs 2 r i − 1 ( U i u , M L , λ u ( r i )) → H hs , Z i u 2 r i ( X L u , M L , λ u ( r i ))
for i = 1 , 2 i=1,2 i = 1 , 2 . We then have the cup product
γ 1 ∪ γ 2 ∈ H h s 2 m − 2 ( U 1 u ∩ U 2 u , M L , λ u ( r 1 ) ⊗ E ∞ M L , λ u ( r 2 ) ) . \gamma_1\cup\gamma_2\in\mathrm{H}^{2m-2}_{\mathrm{hs}}(U_1^{u}\cap U_2^{u},M_{L,\lambda}^{u}(r_1)\otimes_{E_{\infty}}M_{L,\lambda}^{u}(r_2)). γ 1 ∪ γ 2 ∈ H hs 2 m − 2 ( U 1 u ∩ U 2 u , M L , λ u ( r 1 ) ⊗ E ∞ M L , λ u ( r 2 )) .
If we denote by γ ˉ 1 ∨ γ ˉ 2 \bar{\gamma}_1 \vee\bar{\gamma}_2 γ ˉ 1 ∨ γ ˉ 2 the image of γ 1 ∪ γ ˉ 2 \gamma_1 \cup\bar{\gamma}_2 γ 1 ∪ γ ˉ 2 under the map
H h s 2 m − 2 ( U 1 u ∩ U 2 u , M L , λ u ( r 1 ) ⊗ E ∞ M L , λ ˉ u ( r 2 ) ) ⟶ H h s 2 m − 2 ( U 1 u ∩ U 2 u , E ∞ u ( m ) ) H_{\mathrm{hs}}^{2m-2}(U_1^u \cap U_2^u, M_{L,\lambda}^u(r_1) \otimes_{E_\infty} M_{L,\bar{\lambda}}^u(r_2)) \longrightarrow H_{\mathrm{hs}}^{2m-2}(U_1^u \cap U_2^u, E_\infty^u(m)) H hs 2 m − 2 ( U 1 u ∩ U 2 u , M L , λ u ( r 1 ) ⊗ E ∞ M L , λ ˉ u ( r 2 )) ⟶ H hs 2 m − 2 ( U 1 u ∩ U 2 u , E ∞ u ( m ))
induced by (30) , then ⟨ c 1 , c 2 ⟩ λ , L , Z u \langle c_1,c_2\rangle_{\lambda,L,Z}^u ⟨ c 1 , c 2 ⟩ λ , L , Z u is defined to be the image of γ 1 ∨ γ ˉ 2 \gamma_1 \vee\bar{\gamma}_2 γ 1 ∨ γ ˉ 2 under the composite map
H h s 2 m − 2 ( U 1 u ∩ U 2 u , E ∞ u ( m ) ) ⟶ H h s 2 m − 1 ( X L , u , E ∞ u ( m ) ) ⟶ H 1 ( Spec E u , E ∞ u ( 1 ) ) = E ∞ u → κ C , H_{\mathrm{hs}}^{2m-2}(U_1^u \cap U_2^u, E_\infty^u(m)) \longrightarrow H_{\mathrm{hs}}^{2m-1}(X_{L,u}, E_\infty^u(m)) \longrightarrow H^1(\operatorname{Spec} E_u,E_\infty^u(1)) = E_\infty^u \xrightarrow{\kappa} \mathbb{C}, H hs 2 m − 2 ( U 1 u ∩ U 2 u , E ∞ u ( m )) ⟶ H hs 2 m − 1 ( X L , u , E ∞ u ( m )) ⟶ H 1 ( Spec E u , E ∞ u ( 1 )) = E ∞ u κ C ,
in which the first map is the coboundary map in the Mayer–Vietoris exact sequence for the Zariski open covering X L u = U 1 u ∪ U 2 u X_L^u=U_1^u\cup U_2^u X L u = U 1 u ∪ U 2 u , the second map is the trace map for the projective complex manifold X L u X_L^u X L u , and the last map κ \kappa κ is the contraction (Notation 1.11). It is easy to check that ⟨ c 1 , c 2 ⟩ λ , L , Z u \langle c_1,c_2\rangle_{\lambda,L,Z}^u ⟨ c 1 , c 2 ⟩ λ , L , Z u does not depend on the choices of γ 1 , γ 2 \gamma_1,\gamma_2 γ 1 , γ 2 .
Lemma 4.4. For every integer r ≥ 0 r\geq0 r ≥ 0 , the natural map
H h s 2 r ( X L u , M L , λ u ( r ) ) ⟶ H 2 r ( X L u , M L , λ u ( r ) ) H_{\mathrm{hs}}^{2r}(X_L^u,M_{L,\lambda}^u(r))\longrightarrow H^{2r}(X_L^u,M_{L,\lambda}^u(r)) H hs 2 r ( X L u , M L , λ u ( r )) ⟶ H 2 r ( X L u , M L , λ u ( r )) is injective, where M L , λ u M_{L,\lambda}^u M L , λ u denotes the underlying E ∞ u E_\infty^u E ∞ u -linear local system of M L , λ u M_{L,\lambda}^u M L , λ u .
Proof. This lemma follows from the essentially same proofs of [6 Lemmas 2.2&2.3] . We sketch the argument. Since the injectivity property does not depend what coefficients we use, we may forget the E ∞ u E_\infty^u E ∞ u -linear structure and regard M L , λ u M_{L,\lambda}^u M L , λ u as a real polarized Hodge structure of weight zero. Write X X X for X L u X_L^u X L u and M M M for M L , λ u M_{L,\lambda}^u M L , λ u for short. We have an exact sequence
⋯ ⟶ H 2 r − 1 ( X , M ‾ ( r ) ) ⟶ H 2 r − 1 ( X , Ω X ∙ ( M ) / Fil r Ω X ∙ ( M ) ) ⟶ H h s 2 r ( X , M ( r ) ) ⟶ H 2 r ( X , M ‾ ( r ) ) ⟶ ⋯ \cdots\longrightarrow H^{2r-1}(X,\underline{M}(r))\longrightarrow H^{2r-1}(X,\Omega_X^\bullet(M)/\operatorname{Fil}^r\Omega_X^\bullet(M))\longrightarrow H_{\mathrm{hs}}^{2r}(X,M(r))\longrightarrow H^{2r}(X,\underline{M}(r))\longrightarrow\cdots ⋯ ⟶ H 2 r − 1 ( X , M ( r )) ⟶ H 2 r − 1 ( X , Ω X ∙ ( M ) / Fil r Ω X ∙ ( M )) ⟶ H hs 2 r ( X , M ( r )) ⟶ H 2 r ( X , M ( r )) ⟶ ⋯
where Ω X ∙ ( M ) : = Ω X ∙ ⊗ R M \Omega_X^\bullet(M):=\Omega_X^\bullet\otimes_{\mathbb{R}}M Ω X ∙ ( M ) := Ω X ∙ ⊗ R M is a filtered complex of sheaves on X X X . It is known that H 2 r − 1 ( X , M ‾ ( r ) ) H^{2r-1}(X,\underline{M}(r)) H 2 r − 1 ( X , M ( r )) admits a natural pure real Hodge structure W W W of weight 2 r − 1 2r-1 2 r − 1 so that H 2 r − 1 ( X , Ω X ∙ ( M ) / Fil r Ω X ∙ ( M ) ) = W C / Fil r W C H^{2r-1}(X,\Omega_X^\bullet(M)/\operatorname{Fil}^r\Omega_X^\bullet(M))=W_{\mathbb{C}}/\operatorname{Fil}^rW_{\mathbb{C}} H 2 r − 1 ( X , Ω X ∙ ( M ) / Fil r Ω X ∙ ( M )) = W C / Fil r W C and the first map in the above sequence is given by the natural map W ( r ) ⟶ W C / Fil r W C W(r)\longrightarrow W_{\mathbb{C}}/\operatorname{Fil}^rW_{\mathbb{C}} W ( r ) ⟶ W C / Fil r W C . Then the lemma follows since W ( r ) ⟶ W C / Fil r W C W(r)\longrightarrow W_{\mathbb{C}}/\operatorname{Fil}^rW_{\mathbb{C}} W ( r ) ⟶ W C / Fil r W C is a bijection. □
By the above lemma and the Betti-étale comparison theorem, we know that for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) , the Hodge realization map
H Z i 2 r i ( X L , M L , λ ( r i ) ) ⟨ ℓ ⟩ ⟶ H h s , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) H_{\mathrm{Z}_i}^{2r_i}(X_L,M_{L,\lambda}(r_i))^{\langle\ell\rangle}\longrightarrow H_{\mathrm{hs},\mathrm{Z}_i^u}^{2r_i}(X_L^u,M_{L,\lambda}^u(r_i)) H Z i 2 r i ( X L , M L , λ ( r i ) ) ⟨ ℓ ⟩ ⟶ H hs , Z i u 2 r i ( X L u , M L , λ u ( r i ))
factors through the subspace H h s , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) 0 H_{\mathrm{hs},\mathrm{Z}_i^u}^{2r_i}(X_L^u,M_{L,\lambda}^u(r_i))^0 H hs , Z i u 2 r i ( X L u , M L , λ u ( r i ) ) 0 . The similar statement obviously holds for every u ∈ V E f i n ∖ V E ( ℓ ) u\in\mathcal{V}_E^{\mathrm{fin}}\setminus\mathcal{V}_E^{(\ell)} u ∈ V E fin ∖ V E ( ℓ ) . In particular, we may define a pairing
⟨ , ⟩ λ , L , Z ℓ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ C ⊗ Q ℓ (39) \langle\, ,\,\rangle_{\lambda,L,Z}^{\ell}:H_{\mathrm{Z}_1}^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times H_{\mathrm{Z}_2}^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle}\longrightarrow\mathbb{C}\otimes\mathbb{Q}_{\ell} \tag*{(39)} ⟨ , ⟩ λ , L , Z ℓ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ C ⊗ Q ℓ (39)
as
( ∑ u ∈ V E f i n ∖ V E ( ℓ ) log ∥ u ∥ ⋅ ⟨ , ⟩ λ , L , Z , u ℓ ) + ( ∑ u ∈ Hom ( E , C ) ⟨ , ⟩ λ , L , Z u ) . (40) \left(\sum_{u\in\mathcal{V}_E^{\mathrm{fin}}\setminus\mathcal{V}_E^{(\ell)}}\log\lVert u\rVert\cdot\langle\, ,\,\rangle_{\lambda,L,Z,u}^{\ell}\right)+\left(\sum_{u\in\operatorname{Hom}(E,\mathbb{C})}\langle\, ,\,\rangle_{\lambda,L,Z}^u\right). \tag*{(40)} u ∈ V E fin ∖ V E ( ℓ ) ∑ log ∥ u ∥ ⋅ ⟨ , ⟩ λ , L , Z , u ℓ + u ∈ Hom ( E , C ) ∑ ⟨ , ⟩ λ , L , Z u . (40)
Proposition 4.5. Let ( r 1 , r 2 ) (r_1,r_2) ( r 1 , r 2 ) and Z = ( Z 1 , Z 2 ) Z=(Z_1,Z_2) Z = ( Z 1 , Z 2 ) be as above.
The sum (40) is finite, uniformly for all elements in the source.
Suppose that ℓ \ell ℓ is unramified in E E E , L ℓ L_\ell L ℓ is the stabilizer of a self-dual lattice of V ℓ V_\ell V ℓ , and that the Zariski closures of Z 1 Z_1 Z 1 and Z 2 Z_2 Z 2 in X L , ( ℓ ) X_{L,(\ell)} X L , ( ℓ ) do not intersect as well. Then the pairing ⟨ , ⟩ λ , L , Z \langle\, ,\,\rangle_{\lambda,L,Z} ⟨ , ⟩ λ , L , Z (39) coincides with the composition of the natural map
H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ H_{\mathrm{Z}_1}^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times H_{\mathrm{Z}_2}^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle}\longrightarrow H^{2r_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times H^{2r_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle} H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ H 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ and the pairing ⟨ , ⟩ λ , L ℓ \langle\, ,\,\rangle_{\lambda,L}^{\ell} ⟨ , ⟩ λ , L ℓ (38) .
Proof. Denote by ◊ L , Z \Diamond_{L,Z} ◊ L , Z the smallest set of rational primes such that for every rational prime p ∉ ◊ L , Z p\notin\Diamond_{L,Z} p ∈ / ◊ L , Z ,
L p L_p L p is the stabilizer of a self-dual lattice V p V_p V p ;
the Zariski closures of Z 1 Z_1 Z 1 and Z 2 Z_2 Z 2 in X L , ( p ) X_{L,(p)} X L , ( p ) do not intersect.
Then ◊ L , Z \Diamond_{L,Z} ◊ L , Z is a finite set. Choose an object U ∈ Hk ( T ( A ∞ ) ) U \in\operatorname{Hk}(T(\mathbb{A}^{\infty})) U ∈ Hk ( T ( A ∞ )) such that U p U_p U p is maximal for every rational prime p ∉ ◊ L , Z p \notin\Diamond_{L,Z} p ∈ / ◊ L , Z . Denote by Z U : = ( Z U , 1 , Z U , 2 ) Z_U := (Z_{U,1},Z_{U,2}) Z U := ( Z U , 1 , Z U , 2 ) and D U : = ( D U , 1 , D U , 2 ) D_U := (D_{U,1},D_{U,2}) D U := ( D U , 1 , D U , 2 ) the inverse images of Z = ( Z 1 , Z 2 ) Z=(Z_1,Z_2) Z = ( Z 1 , Z 2 ) in X L , U X_{L,U} X L , U and B L , U ∣ b ∣ B_{L,U}^{|b|} B L , U ∣ b ∣ {(xref)}(3.18), respectively. The maps st L , U λ , + \operatorname{st}_{L,U}^{\lambda,+} st L , U λ , + and st L , U λ , − \operatorname{st}_{L,U}^{\lambda,-} st L , U λ , − in Definition 3.17 induce E λ E^\lambda E λ -linear maps
st L , U λ , + ∘ π L , U ∗ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ ⟶ CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ , \operatorname{st}_{L,U}^{\lambda,+}\circ\pi_{L,U}^{*}:H^{2r_1}_{Z_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\longrightarrow\operatorname{CH}^{r_1+|b|}_{D_{U,1}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda, st L , U λ , + ∘ π L , U ∗ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ ⟶ CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ,
st L , U λ , − ∘ π L , U ∗ : H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ . \operatorname{st}_{L,U}^{\lambda,-}\circ\pi_{L,U}^{*}:H^{2r_2}_{Z_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle}\longrightarrow\operatorname{CH}^{r_2+(g-1)|b|}_{D_{U,2}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda. st L , U λ , − ∘ π L , U ∗ : H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ .
respectively. Since D U , 1 D_{U,1} D U , 1 and D U , 2 D_{U,2} D U , 2 do not intersect, we have a (hermitian) local height
⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ : CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ × CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ⟶ E ℓ λ \langle\ ,\ \rangle_{B_{L,U}^{|b|},D_{L,U},u}^{\ell}:\operatorname{CH}^{r_1+|b|}_{D_{U,1}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda\times\operatorname{CH}^{r_2+(g-1)|b|}_{D_{U,2}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda\longrightarrow E_\ell^\lambda ⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ : CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ × CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ⟶ E ℓ λ
for every u ∈ V F f i n u\in\mathcal{V}^{\mathrm{fin}}_F u ∈ V F fin , and a (hermitian) local height
⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ : CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ × CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ⟶ C \langle\ ,\ \rangle_{B_{L,U}^{|b|},D_{L,U},u}^{\ell}:\operatorname{CH}^{r_1+|b|}_{D_{U,1}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda\times\operatorname{CH}^{r_2+(g-1)|b|}_{D_{U,2}}(B_{L,U}^{|b|})^{\langle\ell\rangle}\otimes E^\lambda\longrightarrow\mathbb{C} ⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ : CH D U , 1 r 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ × CH D U , 2 r 2 + ( g − 1 ) ∣ b ∣ ( B L , U ∣ b ∣ ) ⟨ ℓ ⟩ ⊗ E λ ⟶ C
for every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) .
Now Lemma 3.18 implies that for u ∈ V F f i n ∖ V F ( ℓ ) u\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\ell)}_F u ∈ V F fin ∖ V F ( ℓ ) , the pairing
⟨ , ⟩ λ , L , Z , u ℓ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ E ℓ λ \langle\ ,\ \rangle_{\lambda,L,Z,u}^{\ell}:H^{2r_1}_{Z_1}(X_L,M_{L,\lambda}(r_1))^{\langle\ell\rangle}\times H^{2r_2}_{Z_2}(X_L,M_{L,\lambda}(r_2))^{\langle\ell\rangle}\longrightarrow E_\ell^\lambda ⟨ , ⟩ λ , L , Z , u ℓ : H Z 1 2 r 1 ( X L , M L , λ ( r 1 ) ) ⟨ ℓ ⟩ × H Z 2 2 r 2 ( X L , M L , λ ( r 2 ) ) ⟨ ℓ ⟩ ⟶ E ℓ λ
coincides with
1 ∣ E U ∣ ⋅ ⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ ∘ ( st L , U λ , + ∘ π L , U ∗ ) × ( st L , U λ , − ∘ π L , U ∗ ) , \frac{1}{|E_U|}\cdot\langle\ ,\ \rangle_{B_{L,U}^{|b|},D_{L,U},u}^{\ell}\circ(\operatorname{st}_{L,U}^{\lambda,+}\circ\pi_{L,U}^{*})\times(\operatorname{st}_{L,U}^{\lambda,-}\circ\pi_{L,U}^{*}), ∣ E U ∣ 1 ⋅ ⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ ∘ ( st L , U λ , + ∘ π L , U ∗ ) × ( st L , U λ , − ∘ π L , U ∗ ) ,
and that the similar statement holds for u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . Note that
⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ ≡ 0 , u ∈ V F f i n ∖ V F ( ◊ L , Z ) (41) \langle\ ,\ \rangle_{B_{L,U}^{|b|},D_{L,U},u}^{\ell}\equiv0,\qquad u\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\Diamond_{L,Z})}_F \tag*{(41)} ⟨ , ⟩ B L , U ∣ b ∣ , D L , U , u ℓ ≡ 0 , u ∈ V F fin ∖ V F ( ◊ L , Z ) (41)
due to the presence of the smooth projective integral model
B L p , U p , ( p ) ∣ b ∣ : = B L p , U p , ( p ) × X L p , U p , ( p ) ⋯ × X L p , U p , ( p ) B L p , U p , ( p ) B_{L^p,U^p,(p)}^{|b|}:=B_{L^p,U^p,(p)}\times_{X_{L^p,U^p,(p)}}\cdots\times_{X_{L^p,U^p,(p)}}B_{L^p,U^p,(p)} B L p , U p , ( p ) ∣ b ∣ := B L p , U p , ( p ) × X L p , U p , ( p ) ⋯ × X L p , U p , ( p ) B L p , U p , ( p )
of B L , U ∣ b ∣ B_{L,U}^{|b|} B L , U ∣ b ∣ , where B L p , U p , ( p ) : = A 0 , L p , U p , ( p ) × X L p , U p , ( p ) A L p , U p , ( p ) ∨ B_{L^p,U^p,(p)}:=A_{0,L^p,U^p,(p)}\times_{X_{L^p,U^p,(p)}}A_{L^p,U^p,(p)}^\vee B L p , U p , ( p ) := A 0 , L p , U p , ( p ) × X L p , U p , ( p ) A L p , U p , ( p ) ∨ from § 3.8 (with ◊ = { p } \Diamond=\{p\} ◊ = { p } ). Now (1) follows from the above discussion and (4.4).
For (2), the condition implies that ℓ ∉ ◊ L , Z \ell\notin\Diamond_{L,Z} ℓ ∈ / ◊ L , Z . Then it follows from the above discussion, (4.4), and the decomposition of ⟨ , ⟩ B L , U ∣ b ∣ ℓ \langle\ ,\ \rangle_{B_{L,U}^{|b|}}^\ell ⟨ , ⟩ B L , U ∣ b ∣ ℓ into local heights.
The proposition is proved. □
At the end of this section, we describe a way to compute the local height ⟨ , ⟩ λ , L , Z , u ℓ \langle\ ,\ \rangle_{\lambda,L,Z,u}^{\ell} ⟨ , ⟩ λ , L , Z , u ℓ for u ∈ V F f i n ∖ V F ( ℓ ) u\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\ell)}_F u ∈ V F fin ∖ V F ( ℓ ) in the presence of nice regular models. Fix a prime u ∈ V F f i n ∖ V F ( ℓ ) u\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\ell)}_F u ∈ V F fin ∖ V F ( ℓ ) and an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}^{\infty}_F)) Hk ( H ( A F ∞ )) .
Assumption 4.6. There is a functor
X ∙ : Hk ( H ( A F ∞ , p ) ) ⟶ Sch / O E u f l . p r o j \mathcal{X}_{\bullet}:\operatorname{Hk}(H(\mathbb{A}^{\infty,p}_F))\longrightarrow\operatorname{Sch}^{\mathrm{fl.proj}}_{/O_{E_u}} X ∙ : Hk ( H ( A F ∞ , p )) ⟶ Sch / O E u fl.proj that maps objects to regular schemes and arrows to finite étale morphisms, satisfying
X L ′ , p ⊗ O E u E u = X L p ′ ⊗ E E u \mathcal{X}_{L',p}\otimes_{O_{E_u}}E_u=X_{L'_p}\otimes_E E_u X L ′ , p ⊗ O E u E u = X L p ′ ⊗ E E u functorial in L ′ L' L ′ . To ease notation, we will write X L ′ X_{L'} X L ′ for X L ′ , p \mathcal{X}_{L',p} X L ′ , p as long as L p ′ = L p L'_p=L_p L p ′ = L p , and write Y L ′ Y_{L'} Y L ′ for the special fiber of X L ′ \mathcal{X}_{L'} X L ′ .
Under Assumption 4.6, the definition of the E ℓ λ E_\ell^\lambda E ℓ λ -linear local system V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ from §3.5 extends to X L \mathcal{X}_L X L as well, so that its restriction to the generic fiber canonically identifies with M L , ℓ , λ M_{L,\ell,\lambda} M L , ℓ , λ by (3.22).
Lemma 4.7. Suppose that we are given the functor in Assumption 4.6. Let ( C 1 , C 2 ) (C_1,C_2) ( C 1 , C 2 ) be a pair of elements in H e ˊ t , Z 1 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r ) ) × H e ˊ t , Z 2 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r ) ) H_{\mathrm{\acute et},Z_{1,u}\cup Y_L}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\times H_{\mathrm{\acute et},Z_{2,u}\cup Y_L}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r)) H e ˊ t , Z 1 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r )) × H e ˊ t , Z 2 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r )) satisfying that
if we denote by c i c_i c i the image of C i C_i C i in H e ˊ t , Z i , u 2 r ( X L , u , M L , ℓ , λ ( r ) ) H_{\mathrm{\acute et},Z_{i,u}}^{2r}(\mathcal{X}_{L,u},M_{L,\ell,\lambda}(r)) H e ˊ t , Z i , u 2 r ( X L , u , M L , ℓ , λ ( r )) , then c i c_i c i belongs to H e ˊ t , Z i , u 2 r ( X L , u , M L , ℓ , λ ( r ) ) 0 H_{\mathrm{\acute et},Z_{i,u}}^{2r}(\mathcal{X}_{L,u},M_{L,\ell,\lambda}(r))^0 H e ˊ t , Z i , u 2 r ( X L , u , M L , ℓ , λ ( r ) ) 0 for i = 1 , 2 i=1,2 i = 1 , 2 ;
the image of either C 1 C_1 C 1 or C 2 C_2 C 2 in H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) H_{\mathrm{\acute et}}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r)) H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) vanishes.
Denote by C 1 ∙ λ C 2 C_1\mathbin{\mathop{\bullet}^{\lambda}} C_2 C 1 ∙ λ C 2 the image of ( C 1 , C 2 ) (C_1,C_2) ( C 1 , C 2 ) under the composite map
H e ˊ t , Z 1 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r ) ) × H e ˊ t , Z 2 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r ) ) ⟶ H e ˊ t , Y L 4 r ( X L , O E λ , ℓ ( 2 r ) ) ⟶ H e ˊ t , Spec κ u 4 r ( Spec O E u , O E λ , ℓ ( 1 ) ) = O E λ , ℓ , H_{\mathrm{\acute et},Z_{1,u}\cup Y_L}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\times H_{\mathrm{\acute et},Z_{2,u}\cup Y_L}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))
\longrightarrow H_{\mathrm{\acute et},Y_L}^{4r}(\mathcal{X}_L,\mathcal{O}_{E_\lambda,\ell}(2r))
\longrightarrow H_{\mathrm{\acute et},\operatorname{Spec}\kappa_u}^{4r}(\operatorname{Spec}O_{E_u},\mathcal{O}_{E_\lambda,\ell}(1))=O_{E_\lambda,\ell}, H e ˊ t , Z 1 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r )) × H e ˊ t , Z 2 , u ∪ Y L 2 r ( X L , V ℓ , L , λ ( r )) ⟶ H e ˊ t , Y L 4 r ( X L , O E λ , ℓ ( 2 r )) ⟶ H e ˊ t , Spec κ u 4 r ( Spec O E u , O E λ , ℓ ( 1 )) = O E λ , ℓ , in which the first arrow is the cup product induced by the natural hermitian pairing on V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ , the second arrow is the trace map for X L / O E u \mathcal{X}_L/O_{E_u} X L / O E u , and κ u \kappa_u κ u denotes the residue field of E u E_u E u . Then
⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ = C 1 ∙ λ C 2 \langle c_1,c_2\rangle_{\lambda,L,Z,u}^{\ell}=C_1\mathbin{\mathop{\bullet}^{\lambda}} C_2 ⟨ c 1 , c 2 ⟩ λ , L , Z , u ℓ = C 1 ∙ λ C 2 holds.
Proof. The proof of this lemma follows from the same argument for [20 Lemma B.4] , together with the obvious analogue of [20 Remark B.5] . □
Motivic arithmetic inner product formula
In this section, we assume m = 2 r m=2r m = 2 r and that λ \lambda λ is balanced (Definition 2.7).
Let V V V be as in previous sections (but of rank m = 2 r m=2r m = 2 r ). Assume Conjecture 3.36 for V V V so that one can apply Definition 3.38.
Definition 4.8. Let ℓ \ell ℓ be a rational prime.
For an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_F^\infty)) Hk ( H ( A F ∞ )) , we say that ℓ \ell ℓ is L L L -good if ℓ ∤ D E \ell\nmid D_E ℓ ∤ D E and L ℓ L_\ell L ℓ is the stabilizer of a self-dual lattice of V ℓ V_\ell V ℓ .
For a finite collection φ ∞ = ( φ 1 ∞ , φ 2 ∞ , … ) \varphi^\infty=(\varphi_1^\infty,\varphi_2^\infty,\ldots) φ ∞ = ( φ 1 ∞ , φ 2 ∞ , … ) of elements in S ( ( V ∞ ) r ) \mathcal{S}((V^\infty)^r) S (( V ∞ ) r ) , we say that ℓ \ell ℓ is φ ∞ \varphi^\infty φ ∞ -good if there exists an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_F^\infty)) Hk ( H ( A F ∞ )) that fixes φ i ∞ \varphi_i^\infty φ i ∞ for every i i i such that ℓ \ell ℓ is L L L -good.
Consider a holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ (Definition 2.27). Take a pair ϕ = ( ϕ 1 , ϕ 2 ) \phi=(\phi_1,\phi_2) ϕ = ( ϕ 1 , ϕ 2 ) of automorphic forms in ( S + ( C r ) ⊗ C A π ) ∩ A + c u s p λ ( G r ) (\mathcal{S}_+(\mathbb{C}^r)\otimes_{\mathbb{C}}\mathcal{A}_\pi)\cap\mathcal{A}_{+\mathrm{cusp}}^\lambda(G_r) ( S + ( C r ) ⊗ C A π ) ∩ A + cusp λ ( G r ) and a pair φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) \varphi^\infty=(\varphi_1^\infty,\varphi_2^\infty) φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) of Schwartz functions in S ( ( V ∞ ) r ) \mathcal{S}((V^\infty)^r) S (( V ∞ ) r ) . Consider a rational prime ℓ \ell ℓ that is φ ∞ \varphi^\infty φ ∞ -good. We define, using (4.1), an element
⟨ Θ ϕ 1 ∞ ( ϕ 1 ) , Θ ϕ 2 ∞ ( ϕ 2 ) ⟩ λ ℓ ∈ C ⊗ Q ℓ \left\langle\Theta_{\phi_1^\infty}(\phi_1),\Theta_{\phi_2^\infty}(\phi_2)\right\rangle_\lambda^\ell\in\mathbb{C}\otimes\mathbb{Q}_\ell ⟨ Θ ϕ 1 ∞ ( ϕ 1 ) , Θ ϕ 2 ∞ ( ϕ 2 ) ⟩ λ ℓ ∈ C ⊗ Q ℓ
depending only on φ ∞ \varphi^\infty φ ∞ and ϕ \phi ϕ as follows.
Choose an intermediate number field E λ ⊆ L ⊆ C E^\lambda\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C over which π ∞ \pi^\infty π ∞ is defined. Choose an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_F^\infty)) Hk ( H ( A F ∞ )) that fixes φ 1 ∞ \varphi_1^\infty φ 1 ∞ and φ 2 ∞ \varphi_2^\infty φ 2 ∞ such that ℓ \ell ℓ is L L L -good (which is possible by definition). Choose a finite set ◊ \Diamond ◊ of rational primes containing ℓ \ell ℓ , such that for every v ∈ X F f i n ∖ S ( ◊ ) v\in\mathbb{X}_F^{\mathrm{fin}}\setminus\mathbb{S}^{(\Diamond)} v ∈ X F fin ∖ S ( ◊ ) , ( π v ) K r , v ≠ { 0 } (\pi_v)^{K_{r,v}}\ne\{0\} ( π v ) K r , v = { 0 } (Notation 2.13(1)) and L v L_v L v is the stabilizer of a self-dual lattice of V v V_v V v . In particular, π ∞ , ◊ \pi^{\infty,\Diamond} π ∞ , ◊ gives rise to a Satake character
χ π : L [ L ◊ \ H ( A F ∞ , ◊ ) / L ◊ ] ⟶ L (42) \chi_\pi:\mathbb{L}[L^\Diamond\backslash H(\mathbb{A}_F^{\infty,\Diamond})/L^\Diamond]\longrightarrow\mathbb{L} \tag*{(42)} χ π : L [ L ◊ \ H ( A F ∞ , ◊ ) / L ◊ ] ⟶ L (42)
using the canonical isomorphism Z [ L ◊ \ H ( A F ∞ , ◊ ) / L ◊ ] ≃ Z [ K r ◊ \ G r ( A F ∞ , ◊ ) / K r ◊ ] \mathbb{Z}[L^\Diamond\backslash H(\mathbb{A}_F^{\infty,\Diamond})/L^\Diamond]\simeq\mathbb{Z}[K_r^\Diamond\backslash G_r(\mathbb{A}_F^{\infty,\Diamond})/K_r^\Diamond] Z [ L ◊ \ H ( A F ∞ , ◊ ) / L ◊ ] ≃ Z [ K r ◊ \ G r ( A F ∞ , ◊ ) / K r ◊ ] of spherical Hecke algebras.
Lemma 4.9. If we denote by L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ \mathbb{L}[\mathbb{L}^{\diamond}\mathrm{H}(A_{F}^{\infty,\diamond})/\mathbb{L}^{\diamond}]_{\lambda,L}^{\langle\ell\rangle} L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ the subring of L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] \mathbb{L}[\mathbb{L}^{\diamond}\mathrm{H}(A_{F}^{\infty,\diamond})/\mathbb{L}^{\diamond}] L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] consisting of elements that annihilate
∏ u ∈ V E f i n ∖ V E ∞ H e ˊ t 2 r ( X L , u , M L , ℓ , λ ( r ) ) ⊗ E λ L \prod_{u\in\mathcal{V}_{E}^{\mathrm{fin}}\setminus\mathcal{V}_{E}^{\infty}} \mathrm{H}_{\mathrm{\acute{e}t}}^{2r}(X_{L,u},M_{L,\ell,\lambda}(r))\otimes_{E_{\lambda}}L u ∈ V E fin ∖ V E ∞ ∏ H e ˊ t 2 r ( X L , u , M L , ℓ , λ ( r )) ⊗ E λ L (under pullback), then L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ ∖ ker χ π \mathbb{L}[\mathbb{L}^{\diamond}\mathrm{H}(A_{F}^{\infty,\diamond})/\mathbb{L}^{\diamond}]_{\lambda,L}^{\langle\ell\rangle}\setminus\ker\chi_{\pi} L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ ∖ ker χ π is nonempty. Proof. This follows from the same proof of [20 Proposition 6.9(2)] . □
Choose a pair ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) of elements in L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ ∖ ker χ π \mathbb{L}[\mathbb{L}^{\diamond}\mathrm{H}(A_{F}^{\infty,\diamond})/\mathbb{L}^{\diamond}]_{\lambda,L}^{\langle\ell\rangle}\setminus\ker\chi_{\pi} L [ L ⋄ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , L ⟨ ℓ ⟩ ∖ ker χ π , so that s i ∗ Θ ϕ i ∞ λ ( φ i ) L s_{i}^{*}\Theta_{\phi_{i}^{\infty}}^{\lambda}(\varphi_{i})_{L} s i ∗ Θ ϕ i ∞ λ ( φ i ) L belongs to H 2 r i ( X L , M L , λ ( r i ) ) ⟨ ℓ ⟩ \mathrm{H}^{2r_{i}}(X_{L},M_{L,\lambda}(r_{i}))^{\langle\ell\rangle} H 2 r i ( X L , M L , λ ( r i ) ) ⟨ ℓ ⟩ for i = 1 , 2 i=1,2 i = 1 , 2 . We set
⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ : = vol ♯ ( L ) ⋅ ( χ π ( s 1 ) χ π ( s 2 ) ) − 1 ⋅ ⟨ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ ∈ C ⊗ Q ℓ , \left\langle\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1}),\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})\right\rangle_{\lambda}^{\ell}:=\operatorname{vol}^{\sharp}(L)\cdot\left(\chi_{\pi}(s_{1})\chi_{\pi}(s_{2})\right)^{-1}\cdot\left\langle s_{1}^{*}\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1})_{L},s_{2}^{*}\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})_{L}\right\rangle_{\lambda,L}^{\ell}\in\mathbb{C}\otimes\mathbb{Q}_{\ell}, ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ := vol ♯ ( L ) ⋅ ( χ π ( s 1 ) χ π ( s 2 ) ) − 1 ⋅ ⟨ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ ∈ C ⊗ Q ℓ ,
where vol ♯ ( L ) \operatorname{vol}^{\sharp}(L) vol ♯ ( L ) is the normalized volume of L L L [LL21, Definition 3.8].
Lemma 4.10. The quantity ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ \left\langle\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1}),\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})\right\rangle_{\lambda}^{\ell} ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ does not depend on the choices of L \mathbb{L} L , L L L (that fixes φ ∞ \varphi^{\infty} φ ∞ with prescribed L ℓ L_{\ell} L ℓ as above), ⋄ \diamond ⋄ , and ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) .
Proof. The independence of the choice of L \mathbb{L} L is obvious.
We first show the independence of ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) . Choose another pair ( s 1 ′ , s 2 ′ ) (s_{1}^{\prime},s_{2}^{\prime}) ( s 1 ′ , s 2 ′ ) with respect to a possibly different set ⋄ ′ \diamond^{\prime} ⋄ ′ . By Lemma 3.34 , we have
s i ′ ∗ s i ∗ Θ ϕ i ∞ λ ( φ i ) L = s i ∗ Θ ϕ i ∞ λ ( s i ′ ∗ φ i ) L s_{i}^{\prime*}s_{i}^{*}\Theta_{\phi_{i}^{\infty}}^{\lambda}(\varphi_{i})_{L}=s_{i}^{*}\Theta_{\phi_{i}^{\infty}}^{\lambda}(s_{i}^{\prime*}\varphi_{i})_{L} s i ′ ∗ s i ∗ Θ ϕ i ∞ λ ( φ i ) L = s i ∗ Θ ϕ i ∞ λ ( s i ′ ∗ φ i ) L for i = 1 , 2 i=1,2 i = 1 , 2 . By [25 Proposition A.5] , we have
s i ∗ Θ ϕ i ∞ λ ( s i ′ ∗ φ i ) L = s i ∗ Θ s i ′ ∗ ϕ i ∞ λ ( φ i ) L = χ π ( s i ′ ) ‾ ⋅ s i ∗ Θ ϕ i ∞ λ ( φ i ) L . s_{i}^{*}\Theta_{\phi_{i}^{\infty}}^{\lambda}(s_{i}^{\prime*}\varphi_{i})_{L}=s_{i}^{*}\Theta_{s_{i}^{\prime*}\phi_{i}^{\infty}}^{\lambda}(\varphi_{i})_{L}=\overline{\chi_{\pi}(s_{i}^{\prime})}\cdot s_{i}^{*}\Theta_{\phi_{i}^{\infty}}^{\lambda}(\varphi_{i})_{L}. s i ∗ Θ ϕ i ∞ λ ( s i ′ ∗ φ i ) L = s i ∗ Θ s i ′ ∗ ϕ i ∞ λ ( φ i ) L = χ π ( s i ′ ) ⋅ s i ∗ Θ ϕ i ∞ λ ( φ i ) L . It follows that
( χ π ( s 1 ) χ π ( s 2 ) ‾ ) − 1 ⋅ ⟨ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ \left(\overline{\chi_{\pi}(s_{1})\chi_{\pi}(s_{2})}\right)^{-1}\cdot\left\langle s_{1}^{*}\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1})_{L},s_{2}^{*}\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})_{L}\right\rangle_{\lambda,L}^{\ell} ( χ π ( s 1 ) χ π ( s 2 ) ) − 1 ⋅ ⟨ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ = ( χ π ( s 1 ′ s 1 ) χ π ( s 2 ′ s 2 ) ‾ ) − 1 ⋅ ⟨ s 1 ′ ∗ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ′ ∗ s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ =\left(\overline{\chi_{\pi}(s_{1}^{\prime}s_{1})\chi_{\pi}(s_{2}^{\prime}s_{2})}\right)^{-1}\cdot\left\langle s_{1}^{\prime*}s_{1}^{*}\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1})_{L},s_{2}^{\prime*}s_{2}^{*}\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})_{L}\right\rangle_{\lambda,L}^{\ell} = ( χ π ( s 1 ′ s 1 ) χ π ( s 2 ′ s 2 ) ) − 1 ⋅ ⟨ s 1 ′ ∗ s 1 ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ′ ∗ s 2 ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ = ( χ π ( s 1 s 1 ′ ) χ π ( s 2 s 2 ′ ) ‾ ) − 1 ⋅ ⟨ s 1 ∗ s 1 ′ ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ s 2 ′ ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ =\left(\overline{\chi_{\pi}(s_{1}s_{1}^{\prime})\chi_{\pi}(s_{2}s_{2}^{\prime})}\right)^{-1}\cdot\left\langle s_{1}^{*}s_{1}^{\prime*}\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1})_{L},s_{2}^{*}s_{2}^{\prime*}\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})_{L}\right\rangle_{\lambda,L}^{\ell} = ( χ π ( s 1 s 1 ′ ) χ π ( s 2 s 2 ′ ) ) − 1 ⋅ ⟨ s 1 ∗ s 1 ′ ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ∗ s 2 ′ ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ = ( χ π ( s 1 ′ ) χ π ( s 2 ′ ) ‾ ) − 1 ⋅ ⟨ s 1 ′ ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ′ ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ . =\left(\overline{\chi_{\pi}(s_{1}^{\prime})\chi_{\pi}(s_{2}^{\prime})}\right)^{-1}\cdot\left\langle s_{1}^{\prime*}\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1})_{L},s_{2}^{\prime*}\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})_{L}\right\rangle_{\lambda,L}^{\ell}. = ( χ π ( s 1 ′ ) χ π ( s 2 ′ ) ) − 1 ⋅ ⟨ s 1 ′ ∗ Θ ϕ 1 ∞ λ ( φ 1 ) L , s 2 ′ ∗ Θ ϕ 2 ∞ λ ( φ 2 ) L ⟩ λ , L ℓ . Thus, we obtain the independence of ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) . The independence of ⋄ \diamond ⋄ follows from the independence of ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) . Finally, the independence of L L L follows from Lemma 4.3 the independence of ⋄ \diamond ⋄ .
The lemma is proved. □
Conjecture 4.11 (Motivic arithmetic inner product formula). Suppose that the hermitian E E E -weight λ \lambda λ is of rank 2 r 2r 2 r and is balanced. Let ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_{r}(A_{F}) G r ( A F ) of weight λ \lambda λ (Definition 2.27 ). For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_{1},\varphi_{2}) φ = ( φ 1 , φ 2 ) of automorphic forms in ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) c u s p h o l (\mathcal{S}_{\lambda^{+}}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}\mathcal{A}_{\pi})\cap\mathcal{A}_{\lambda^{+}}(G_{r})_{\mathrm{cusp}}^{\mathrm{hol}} ( S λ + ( C r ) ∨ ⊗ C A π ) ∩ A λ + ( G r ) cusp hol and every pair φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) \varphi^{\infty}=(\varphi_{1}^{\infty},\varphi_{2}^{\infty}) φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) of Schwartz functions in S ( ( V ∞ ) r ) \mathcal{S}((V^{\infty})^{r}) S (( V ∞ ) r ) ,
⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ = L ′ ( 1 2 , π ) b 2 r ( 0 ) ⋅ C ( λ ) ⋅ Z f i n ♯ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) \left\langle\Theta_{\phi_{1}^{\infty}}^{\lambda}(\varphi_{1}),\Theta_{\phi_{2}^{\infty}}^{\lambda}(\varphi_{2})\right\rangle_{\lambda}^{\ell}=\frac{L^{\prime}\left(\frac{1}{2},\pi\right)}{b_{2r}(0)}\cdot C(\lambda)\cdot\mathfrak{Z}_{\mathrm{fin}}^{\sharp}(\varphi,\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}}) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ = b 2 r ( 0 ) L ′ ( 2 1 , π ) ⋅ C ( λ ) ⋅ Z fin ♯ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) holds for all rational primes ℓ \ell ℓ that are φ ∞ \varphi^{\infty} φ ∞ -good, in which C ( λ ) C(\lambda) C ( λ ) is the real constant from Proposition 2.29 , and Z f i n ♯ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) \mathfrak{Z}_{\mathrm{fin}}^{\sharp}(\varphi,\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}}) Z fin ♯ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) is introduced in Definition 2.28 . In particular, the left-hand side belongs to C \mathbb{C} C and is independent of ℓ \ell ℓ .
Remark 4.12. We have the following remarks concerning the above conjecture.
By Theorem 4.15 below, we know a priori that, when r = 1 r=1 r = 1 ,
⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ \left\langle\Theta_{\phi^\infty_1}^{\lambda}(\varphi_1),\Theta_{\phi^\infty_2}^{\lambda}(\varphi_2)\right\rangle_{\lambda}^{\ell} ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ
takes value in C \mathbb{C} C and is independent of ℓ \ell ℓ .
By the local theta dichotomy, for every v ∈ S F v\in\mathbb{S}_F v ∈ S F , there exists a unique up to isomorphism hermitian space V ( π v ) V(\pi_v) V ( π v ) over E v E_v E v of rank m = 2 r m=2r m = 2 r , such that the local theta lift of π v \pi_v π v to U ( V ( π v ) ) U(V(\pi_v)) U ( V ( π v )) is nontrivial. Put V ( π ) : = ⨁ v V ( π v ) V(\pi):=\bigoplus_v V(\pi_v) V ( π ) := ⨁ v V ( π v ) as a hermitian space over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F . Then the condition that λ \lambda λ is balanced implies that V ( π ) V(\pi) V ( π ) is totally positive definite. Moreover, V ( π ) V(\pi) V ( π ) is coherent/incoherent (Definition 2.15) if and only if ϵ ( π ) = + 1 / − 1 \epsilon(\pi)=+1/-1 ϵ ( π ) = + 1/ − 1 (Definition 2.27 ).
If V V V is not isomorphic to V ( π ) V(\pi) V ( π ) , then both terms
⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ , Z f i n ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ‾ ) \left\langle\Theta_{\phi_1^\infty}^{\lambda}(\varphi_1),\Theta_{\phi_2^\infty}^{\lambda}(\varphi_2)\right\rangle_{\lambda}^{\ell},\qquad\mathcal{Z}_{\mathrm{fin}}^{\sharp}(\varphi,\varphi_1^\infty\otimes\overline{\varphi_2^\infty}) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ λ ℓ , Z fin ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ )
always vanish by the theta dichotomy. In particular, when ϵ ( π ) = 1 \epsilon(\pi)=1 ϵ ( π ) = 1 , both sides in the conjecture are zero.
When ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 (so that L ( 1 2 , π ) = 0 L(\frac{1}{2},\pi)=0 L ( 2 1 , π ) = 0 ) and V = V ( π ) V=V(\pi) V = V ( π ) ,25 one can always find φ \varphi φ and ϕ ∞ \phi^\infty ϕ ∞ such that
Z f i n ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ‾ ) ≠ 0. \mathcal{Z}_{\mathrm{fin}}^{\sharp}(\varphi,\varphi_1^\infty\otimes\overline{\varphi_2^\infty})\ne0. Z fin ♯ ( φ , φ 1 ∞ ⊗ φ 2 ∞ ) = 0.
26 Thus, the conjecture indeed provides a formula for the central L L L -derivative
L ′ ( 1 2 , π ) . L'\left(\frac{1}{2},\pi\right). L ′ ( 2 1 , π ) .
When “everything is unramified”, that is, E / F E/F E / F is everywhere unramified, ( φ 0 ∞ : = ) φ 1 ∞ = φ 2 ∞ (\varphi_0^\infty:=)\varphi_1^\infty=\varphi_2^\infty ( φ 0 ∞ := ) φ 1 ∞ = φ 2 ∞ is the characteristic function of Λ r \Lambda^r Λ r for a self-dual lattice Λ \Lambda Λ of V ∞ V^\infty V ∞ ,27 ( φ 0 : = ) φ 1 = φ 2 (\varphi_0:=)\varphi_1=\varphi_2 ( φ 0 := ) φ 1 = φ 2 is fixed by ∏ v ∈ S F ∞ K r , v \prod_{v\in\mathbb{S}_F}^{\infty}K_{r,v} ∏ v ∈ S F ∞ K r , v , the conjecture particularly implies that
⟨ Θ φ 0 ∞ λ ( φ 0 ) , Θ φ 0 ∞ λ ( φ 0 ) ⟩ λ ℓ = L ′ ( 1 2 , π ) b 2 r ( 0 ) ⋅ C ( λ ) ⋅ ( φ 0 , φ 0 ) G r , \left\langle\Theta_{\varphi_0^\infty}^{\lambda}(\varphi_0),\Theta_{\varphi_0^\infty}^{\lambda}(\varphi_0)\right\rangle_{\lambda}^{\ell}
=
\frac{L'(\frac{1}{2},\pi)}{b_{2r}(0)}\cdot C(\lambda)\cdot(\varphi_0,\varphi_0)_{G_r}, ⟨ Θ φ 0 ∞ λ ( φ 0 ) , Θ φ 0 ∞ λ ( φ 0 ) ⟩ λ ℓ = b 2 r ( 0 ) L ′ ( 2 1 , π ) ⋅ C ( λ ) ⋅ ( φ 0 , φ 0 ) G r ,
where ( , ) G r (\ ,\ )_{G_r} ( , ) G r is the inner product from Definition 2.14(2).
To end this section, we formula an analogue of the Beilinson–Bloch conjecture for the motivic cohomology
H 2 r ( X , M λ ( r ) ) : = l i m → L H 2 r ( X L , M L , λ ( r ) ) H^{2r}(X,M_\lambda(r)):=\varinjlim_L H^{2r}(X_L,M_{L,\lambda}(r)) H 2 r ( X , M λ ( r )) := L lim H 2 r ( X L , M L , λ ( r ))
as a module over E λ [ H ( A F ∞ ) ] E^\lambda[H(\mathbb{A}_F^\infty)] E λ [ H ( A F ∞ )] .
Remark 4.13. Consider a holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ with Π : = BC ( π ) \Pi:=\operatorname{BC}(\pi) Π := BC ( π ) its base change to E E E , which is a hermitian isobaric automorphic representation of GL 2 r ( A E ) \operatorname{GL}_{2r}(\mathbb{A}_E) GL 2 r ( A E ) in the sense of [5 Definition 1.5] . For every irreducible admissible representation π ~ ∞ \widetilde{\pi}^{\infty} π ∞ of H ( A F ∞ ) H(\mathbb{A}_F^\infty) H ( A F ∞ ) that is nearly equivalent to π ∞ \pi^\infty π ∞ , one has an associated isobaric factor Π π ~ ∞ \Pi_{\widetilde{\pi}^{\infty}} Π π ∞ of Π \Pi Π , determined via the way in [20 Lemma 3.5] .
Conjecture 4.14 (Beilinson–Bloch). Suppose that the hermitian E E E -weight λ \lambda λ is of rank 2 r 2r 2 r and is balanced. Let V V V be a totally positive definite incoherent hermitian space over A E / A F \mathbb{A}_E/\mathbb{A}_F A E / A F of rank 2 r 2r 2 r satisfying ∣ V ( V ) ∣ > 1 |\mathcal{V}(V)|>1 ∣ V ( V ) ∣ > 1 .
The E λ [ H ( A F ∞ ) ] E^\lambda[H(\mathbb{A}_F^\infty)] E λ [ H ( A F ∞ )] -module H 2 r ( X , M λ ( r ) ) H^{2r}(X,M_\lambda(r)) H 2 r ( X , M λ ( r )) is admissible and semisimple.
(2) For every holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) of Gr ( A F ) \operatorname{Gr}(\mathbb{A}_{F}) Gr ( A F ) of weight λ \lambda λ , the relation
dim C Hom H ( A F ∞ ) ( π ~ ∞ , H 2 r ( X , M λ ( r ) ) ⊗ E λ C ) = ord s = 1 / 2 L ( s , Π π ~ ∞ ) \dim_{\mathbb{C}}\operatorname{Hom}_{H(\mathbb{A}_{F}^{\infty})}\left(\widetilde{\pi}^{\infty},H^{2r}(X,M_{\lambda}(r))\otimes_{E_{\lambda}}\mathbb{C}\right)=\operatorname{ord}_{s=1/2}L(s,\Pi_{\widetilde{\pi}^{\infty}}) dim C Hom H ( A F ∞ ) ( π ∞ , H 2 r ( X , M λ ( r )) ⊗ E λ C ) = ord s = 1/2 L ( s , Π π ∞ )
holds for every irreducible admissible representation π ~ ∞ \widetilde{\pi}^{\infty} π ∞ of H ( A F ∞ ) H(\mathbb{A}_{F}^{\infty}) H ( A F ∞ ) that is nearly equivalent to π ∞ \pi^{\infty} π ∞ .
For ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) in Conjecture 4.14 (2) satisfying ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 and ∣ V ( V ( π ) ) ∣ > 1 \lvert\mathcal{V}(V(\pi))\rvert>1 ∣ V ( V ( π ))∣ > 1 (and we take V = V ( π ) V=V(\pi) V = V ( π ) ), the assignment ( ϕ ∞ , φ ) ↦ Θ ϕ ∞ λ ( φ ) (\phi^{\infty},\varphi)\mapsto\Theta_{\phi^{\infty}}^{\lambda}(\varphi) ( ϕ ∞ , φ ) ↦ Θ ϕ ∞ λ ( φ ) indeed induces an element in Hom H ( A F ∞ ) ( θ ( π ∞ ) , H 2 r ( X , M λ ( r ) ) ⊗ E λ C ) \operatorname{Hom}_{H(\mathbb{A}_{F}^{\infty})}\left(\theta(\pi^{\infty}),H^{2r}(X,M_{\lambda}(r))\otimes_{E_{\lambda}}\mathbb{C}\right) Hom H ( A F ∞ ) ( θ ( π ∞ ) , H 2 r ( X , M λ ( r )) ⊗ E λ C ) . Thus, Conjecture 4.11 provides with the following evidence:
ord s = 1 / 2 L ( s , π ) = 1 ⟹ dim C Hom H ( A F ∞ ) ( θ ( π ∞ ) , H 2 r ( X , M λ ( r ) ) ⊗ E λ C ) ≥ 1 \operatorname{ord}_{s=1/2}L(s,\pi)=1\quad\Longrightarrow\quad\dim_{\mathbb{C}}\operatorname{Hom}_{H(\mathbb{A}_{F}^{\infty})}\left(\theta(\pi^{\infty}),H^{2r}(X,M_{\lambda}(r))\otimes_{E_{\lambda}}\mathbb{C}\right)\geq1 ord s = 1/2 L ( s , π ) = 1 ⟹ dim C Hom H ( A F ∞ ) ( θ ( π ∞ ) , H 2 r ( X , M λ ( r )) ⊗ E λ C ) ≥ 1
toward Conjecture 4.14 (2). Indeed, in this case, Π θ ( π ∞ ) \Pi_{\theta(\pi^{\infty})} Π θ ( π ∞ ) is the unique isobaric factor of Π \Pi Π with global root number − 1 -1 − 1 .
Independence-of-ℓ \ell ℓ for height pairing over unitary Shimura curves
In this section, we assume m = 2 m=2 m = 2 .
Theorem 4.15. For every object L L L of Hk ( H ( A ∞ ) ) \operatorname{Hk}(H(\mathbb{A}^{\infty})) Hk ( H ( A ∞ )) , the height pairing ⟨ , ⟩ λ , L ℓ \langle\ ,\ \rangle_{\lambda,L}^{\ell} ⟨ , ⟩ λ , L ℓ ((38) (for L L L -good ℓ \ell ℓ in the sense of Definition 4.8(1)) takes values in C \mathbb{C} C and is independent of ℓ \ell ℓ .
When λ = 0 \lambda=0 λ = 0 , the above theorem is well-known as the height pairing coincides with the Néron–Tate height pairing. It remains to consider the case where λ ≠ 0 \lambda\ne0 λ = 0 . In this case,
H 2 ( X L , M L , λ ( 1 ) ) ⟨ ℓ ⟩ = H 2 ( X L , M L , λ ( 1 ) ) \mathrm{H}^{2}(X_{L},M_{L,\lambda}(1))^{\langle\ell\rangle}=\mathrm{H}^{2}(X_{L},M_{L,\lambda}(1)) H 2 ( X L , M L , λ ( 1 ) ) ⟨ ℓ ⟩ = H 2 ( X L , M L , λ ( 1 ))
since H e ˊ t 2 ( X L , u , r ℓ M L , λ ( 1 ) ) \mathrm{H}^{2}_{\acute{\mathrm{e}}\mathrm{t}}(X_{L,u},r_{\ell}M_{L,\lambda}(1)) H e ˊ t 2 ( X L , u , r ℓ M L , λ ( 1 )) vanishes for every u ∈ V E f i n ∖ V E ⟨ ℓ ⟩ u\in\mathcal{V}_{E}^{\mathrm{fin}}\setminus\mathcal{V}_{E}^{\langle\ell\rangle} u ∈ V E fin ∖ V E ⟨ ℓ ⟩ , so at least the source of ⟨ , ⟩ λ , L ℓ \langle\ ,\ \rangle_{\lambda,L}^{\ell} ⟨ , ⟩ λ , L ℓ is independent of ℓ \ell ℓ .
Definition 4.16. For every v ∈ V F f i n v\in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin , we define an integer t v t_{v} t v (depending also on V V V ) as follows:
Table 4.16.
Take a finite set ⋄ \diamond ⋄ of rational primes. The definition below extends Definition 3.42 .
Definition 4.17. We define a functor
X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) ⟶ PSch / O E , ( ⋄ ) ′ X^{\mathrm{b}}_{\bullet,\bullet,(\diamond)}:\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,\diamond}))\times\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond}))\longrightarrow\operatorname{PSch}^{\prime}_{/O_{E,(\diamond)}} X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) ⟶ PSch / O E , ( ⋄ ) ′ such that for every objects ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) (L^{\diamond},U^{\diamond})\in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,\diamond}))\times\operatorname{Hk}(T(\mathbb{A}^{\infty,\diamond})) ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) and S ∈ Sch / O E , ( ⋄ ) ′ S\in\operatorname{Sch}^{\prime}_{/O_{E,(\diamond)}} S ∈ Sch / O E , ( ⋄ ) ′ , X L ⋄ , U ⋄ , ( ⋄ ) b ( S ) X^{\mathrm{b}}_{L^{\diamond},U^{\diamond},(\diamond)}(S) X L ⋄ , U ⋄ , ( ⋄ ) b ( S ) is the set of equivalence classes of sextuples ( A 0 , λ 0 , η ‾ 0 ⋄ ; A , λ , η ‾ ⋄ ) (A_{0},\lambda_{0},\overline{\eta}_{0}^{\diamond};A,\lambda,\overline{\eta}^{\diamond}) ( A 0 , λ 0 , η 0 ⋄ ; A , λ , η ⋄ ) , where
( A 0 , λ 0 , η ‾ 0 ⋄ ) (A_{0},\lambda_{0},\overline{\eta}_{0}^{\diamond}) ( A 0 , λ 0 , η 0 ⋄ ) is an object of Y U ⋄ , ( ⋄ ) ( S ) Y_{U^{\diamond},(\diamond)}(S) Y U ⋄ , ( ⋄ ) ( S ) (Definition 3.41 );
A A A is an object of AV ( S , O E , ( ⋄ ) ) \operatorname{AV}(S,O_{E,(\diamond)}) AV ( S , O E , ( ⋄ ) ) (Definition A.5) satisfying the Eisenstein condition and the signature ( 1 , 1 ) (1,1) ( 1 , 1 ) condition [[30 Definition 6.2] ] for
A = { ϕ ∈ Hom ( E , E a c ) ∣ ϕ ∣ E 0 = id E 0 } ∖ { id E } , B = { ϕ ∈ Hom ( E , E a c ) ∣ ϕ ∣ E 0 = id E 0 } ∪ { c } ; A=\{\phi\in\operatorname{Hom}(E,E^{\mathrm{ac}})\mid\phi|_{E_{0}}=\operatorname{id}_{E_{0}}\}\setminus\{\operatorname{id}_{E}\},\qquad B=\{\phi\in\operatorname{Hom}(E,E^{\mathrm{ac}})\mid\phi|_{E_{0}}=\operatorname{id}_{E_{0}}\}\cup\{c\}; A = { ϕ ∈ Hom ( E , E ac ) ∣ ϕ ∣ E 0 = id E 0 } ∖ { id E } , B = { ϕ ∈ Hom ( E , E ac ) ∣ ϕ ∣ E 0 = id E 0 } ∪ { c } ; λ : A → A ∨ \lambda:A\to A^{\vee} λ : A → A ∨ is a polarization that turns the complex conjugation of O E , ( ⋄ ) O_{E,(\diamond)} O E , ( ⋄ ) into the Rosati involution, such that for every v ∈ V F f i n v\in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin , ker λ [ p v ∞ ] \ker\lambda[p_{v}^{\infty}] ker λ [ p v ∞ ] is a semisimple O E v O_{E_{v}} O E v -module of length t v t_{v} t v ;
η ˉ ⋄ \bar{\eta}^{\diamond} η ˉ ⋄ is a hermitian L ⋄ L^{\diamond} L ⋄ -level structure, that is, for a chosen geometric point s s s on every connected component of S S S , a π 1 ( S , s ) \pi_1(S,s) π 1 ( S , s ) -invariant L ⋄ L^{\diamond} L ⋄ -orbit of isometries
η ⋄ : V ∞ , ⋄ → ∼ Hom A E ⋄ , ⋄ ( H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) ⊗ A ∞ , ⋄ A F ∞ , ⋄ , H 1 e ˊ t ( A s , A ∞ , ⋄ ) ) \eta^{\diamond}: V^{\infty,\diamond} \xrightarrow{\sim} \operatorname{Hom}_{A_E^{\diamond,\diamond}}\left(\mathrm{H}_1^{\mathrm{\acute{e}t}}(A_{0s}, A^{\infty,\diamond}) \otimes_{A^{\infty,\diamond}} A_F^{\infty,\diamond}, \mathrm{H}_1^{\mathrm{\acute{e}t}}(A_s, A^{\infty,\diamond})\right) η ⋄ : V ∞ , ⋄ ∼ Hom A E ⋄ , ⋄ ( H 1 e ˊ t ( A 0 s , A ∞ , ⋄ ) ⊗ A ∞ , ⋄ A F ∞ , ⋄ , H 1 e ˊ t ( A s , A ∞ , ⋄ ) )
of hermitian spaces over A E ⋄ , ⋄ / A F ⋄ , ⋄ A_E^{\diamond,\diamond}/A_F^{\diamond,\diamond} A E ⋄ , ⋄ / A F ⋄ , ⋄ .
Two objects ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) (A_0, \lambda_0, \bar{\eta}_0^{\diamond}; A, \lambda, \bar{\eta}^{\diamond}) ( A 0 , λ 0 , η ˉ 0 ⋄ ; A , λ , η ˉ ⋄ ) and ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ; A ′ , λ ′ , η ˉ ′ ⋄ ) (A'_0, \lambda'_0, \bar{\eta}'^{\diamond}_0; A', \lambda', \bar{\eta}'^{\diamond}) ( A 0 ′ , λ 0 ′ , η ˉ 0 ′ ⋄ ; A ′ , λ ′ , η ˉ ′ ⋄ ) are equivalent if there exist isomorphisms φ 0 : A 0 → A 0 ′ \varphi_0: A_0 \to A'_0 φ 0 : A 0 → A 0 ′ in the category A V ( S , O E 0 , ( ⋄ ) ) \mathrm{AV}(S,O_{E_0,(\diamond)}) AV ( S , O E 0 , ( ⋄ ) ) and φ : A → A ′ \varphi: A \to A' φ : A → A ′ in the category A V ( S , O E , ( ⋄ ) ) \mathrm{AV}(S,O_{E,(\diamond)}) AV ( S , O E , ( ⋄ ) ) satisfying that
φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 \varphi_0^\vee\circ\lambda'_0 \circ\varphi_0 φ 0 ∨ ∘ λ 0 ′ ∘ φ 0 and φ ∨ ∘ λ ′ ∘ φ \varphi^\vee\circ\lambda' \circ\varphi φ ∨ ∘ λ ′ ∘ φ are the same prime-to-⋄ \diamond ⋄ multiple of λ 0 \lambda_0 λ 0 and λ \lambda λ , respectively;
the U 0 ⋄ U^{\diamond}_0 U 0 ⋄ -orbits φ 0 ∗ ∘ η ˉ 0 ⋄ \varphi_{0*} \circ\bar{\eta}^{\diamond}_0 φ 0 ∗ ∘ η ˉ 0 ⋄ and η ˉ 0 ′ ⋄ \bar{\eta}'^{\diamond}_0 η ˉ 0 ′ ⋄ coincide;
the L ⋄ L^{\diamond} L ⋄ -orbit of maps v ↦ φ ∗ ∘ η ˉ ⋄ ( v ) ∘ ( φ 0 ∗ ) − 1 v \mapsto\varphi_* \circ\bar{\eta}^{\diamond}(v) \circ(\varphi_{0*})^{-1} v ↦ φ ∗ ∘ η ˉ ⋄ ( v ) ∘ ( φ 0 ∗ ) − 1 for v ∈ V ∞ , ⋄ v \in V^{\infty,\diamond} v ∈ V ∞ , ⋄ coincides with η ˉ ′ ⋄ \bar{\eta}'^{\diamond} η ˉ ′ ⋄ .
The values on morphisms are defined similarly as in Definition 3.42 .
Note that over S ⊗ O E , ( ⋄ ) E S \otimes_{O_{E,(\diamond)}} E S ⊗ O E , ( ⋄ ) E , the Eisenstein condition plus the signature ( 1 , 1 ) (1,1) ( 1 , 1 ) condition is equivalent to that tr A ( e ) = 2 Tr E / E 0 ( e ) − e + e ˉ \operatorname{tr}_{A}(e) = 2\operatorname{Tr}_{E/E_0}(e) - e + \bar{e} tr A ( e ) = 2 Tr E / E 0 ( e ) − e + e ˉ for e ∈ O E , ( ⋄ ) e \in O_{E,(\diamond)} e ∈ O E , ( ⋄ ) by [30 Example 6.3] . It follows that if we choose an open compact subgroup L b ⋄ L^{\diamond}_b L b ⋄ of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) that is optimal (see Definition 4.18 below) and denote by U ⋄ U_{\diamond} U ⋄ the maximal open compact subgroup of T ( Q ⋄ ) T(\mathbb{Q}_{\diamond}) T ( Q ⋄ ) , then it is clear from the construction that
X L b ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) E = X L b ⋄ , U ⋄ U ⋄ b X^{b}_{L^{\diamond}_b,U_{\diamond},(\diamond)} \otimes_{O_{E,(\diamond)}} E = X^{b}_{L^{\diamond}_b,U_{\diamond}U^{\diamond}} X L b ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) E = X L b ⋄ , U ⋄ U ⋄ b
functorial in ( L ⋄ , U ⋄ ) (L^{\diamond}, U^{\diamond}) ( L ⋄ , U ⋄ ) .
Definition 4.18. We say that an open compact subgroup L ⋄ b L^{b}_{\diamond} L ⋄ b of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) is optimal if ( L ⋄ b ) v (L^{b}_{\diamond})_v ( L ⋄ b ) v is the stabilizer of a (necessarily unique) vertex lattice of V v V_v V v of index t v t_v t v for every v ∈ B F ( ⋄ ) v \in\mathcal{B}^{(\diamond)}_F v ∈ B F ( ⋄ ) .
Similar to Definition 3.42 , X ∙ , ∙ , ( ⋄ ) b X^b_{\bullet,\bullet,(\diamond)} X ∙ , ∙ , ( ⋄ ) b factors through the full subcategory S c h / O E , ( ⋄ ) f l . p r o j \mathrm{Sch}^{\mathrm{fl.proj}}_{/O_{E,(\diamond)}} Sch / O E , ( ⋄ ) fl.proj of P S c h / O E , ( ⋄ ) ′ \mathrm{PSch}'_{/O_{E,(\diamond)}} PSch / O E , ( ⋄ ) ′ , and we have universal objects
A 0 , ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) → AV ( S c h / O E , ( ⋄ ) f l . p r o j , O E 0 , ( ⋄ ) ) , A_{0,\bullet,\bullet,(\diamond)}/X^b_{\bullet,\bullet,(\diamond)}:
\operatorname{Hk}(H(A_F^{\infty,\diamond})) \times\operatorname{Hk}(T(A^{\infty,\diamond})) \to\operatorname{AV}(\mathrm{Sch}^{\mathrm{fl.proj}}_{/O_{E,(\diamond)}}, O_{E_0,(\diamond)}), A 0 , ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) → AV ( Sch / O E , ( ⋄ ) fl.proj , O E 0 , ( ⋄ ) ) ,
A ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) → AV ( S c h / O E , ( ⋄ ) f l . p r o j , O E , ( ⋄ ) ) A_{\bullet,\bullet,(\diamond)}/X^b_{\bullet,\bullet,(\diamond)}:
\operatorname{Hk}(H(A_F^{\infty,\diamond})) \times\operatorname{Hk}(T(A^{\infty,\diamond})) \to\operatorname{AV}(\mathrm{Sch}^{\mathrm{fl.proj}}_{/O_{E,(\diamond)}}, O_{E,(\diamond)}) A ∙ , ∙ , ( ⋄ ) / X ∙ , ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) → AV ( Sch / O E , ( ⋄ ) fl.proj , O E , ( ⋄ ) )
over X ∙ , ∙ , ( ⋄ ) b X^b_{\bullet,\bullet,(\diamond)} X ∙ , ∙ , ( ⋄ ) b that are Cartesian (Definition 1.14 ), and the functor
X ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ ) ) → S c h / O E , ( ⋄ ) f l . p r o j X^b_{\bullet,(\diamond)}:
\operatorname{Hk}(H(A_F^{\infty,\diamond})) \to\mathrm{Sch}^{\mathrm{fl.proj}}_{/O_{E,(\diamond)}} X ∙ , ( ⋄ ) b : Hk ( H ( A F ∞ , ⋄ )) → Sch / O E , ( ⋄ ) fl.proj
such that for every L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ ) ) L^{\diamond} \in\operatorname{Hk}(H(A_F^{\infty,\diamond})) L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ )) , X L ⋄ , ( ⋄ ) b = X L ⋄ , U ⋄ , ( ⋄ ) b / Ξ U ⋄ X^b_{L^{\diamond},(\diamond)} = X^b_{L^{\diamond},U^{\diamond},(\diamond)}/\Xi_{U^{\diamond}} X L ⋄ , ( ⋄ ) b = X L ⋄ , U ⋄ , ( ⋄ ) b / Ξ U ⋄ . By construction, it is clear that
X L ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) E = X L ⋄ b X^b_{L^{\diamond},(\diamond)} \otimes_{O_{E,(\diamond)}} E = X^b_{L^{\diamond}} X L ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) E = X L ⋄ b
functorial in L ⋄ L^{\diamond} L ⋄ .
Remark 4.19. Choose an object U ⋄ ∈ Hk ( T ( A ∞ , ⋄ ) ) U^{\diamond} \in\operatorname{Hk}(T(A^{\infty,\diamond})) U ⋄ ∈ Hk ( T ( A ∞ , ⋄ )) and an optimal open compact subgroup L ⋄ b L^b_{\diamond} L ⋄ b of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) . In this remark, we recall the basic uniformization of X ∙ , U ⋄ , ( ⋄ ) b X^b_{\bullet,U^{\diamond},(\diamond)} X ∙ , U ⋄ , ( ⋄ ) b at a place u ∈ B E ( ⋄ ) u \in\mathcal{B}^{(\diamond)}_E u ∈ B E ( ⋄ ) with p = p ( u ) p = p(u) p = p ( u ) the underlying rational prime.
Denote by E ˘ u \breve{E}_u E ˘ u a complete maximal unramified extension of E u E_u E u and κ u \kappa_u κ u its residue field, which is an algebraic closure of F p \mathbb{F}_p F p . Choose a framing , that is, an element
P ⋆ : = ( A 0 ⋆ , λ 0 ⋆ , η ˉ 0 ⋄ , ⋆ ; A ⋆ , λ ⋆ , η ⋄ , ⋆ ) ∈ l i m ← L ⋄ X L ⋄ , U ⋄ , ( ⋄ ) b ( O E ˘ u ) P^\star:= (A^\star_0,\lambda^\star_0,\bar{\eta}^{\diamond,\star}_0; A^\star,\lambda^\star,\eta^{\diamond,\star}) \in\varprojlim_{L^{\diamond}} X^b_{L^{\diamond},U^{\diamond},(\diamond)}(O_{\breve{E}_u}) P ⋆ := ( A 0 ⋆ , λ 0 ⋆ , η ˉ 0 ⋄ , ⋆ ; A ⋆ , λ ⋆ , η ⋄ , ⋆ ) ∈ L ⋄ lim X L ⋄ , U ⋄ , ( ⋄ ) b ( O E ˘ u )
whose reduction is basic, where the inverse limit is taken over all L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ ) ) L^{\diamond} \in\operatorname{Hk}(H(A_F^{\infty,\diamond})) L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ )) with respect to inclusions. The framing P ⋆ P^\star P ⋆ gives rise to (U1): a totally positive definite hermitian space
V ⋆ : = Hom O E , ( ⋄ ) ( A ⋄ ⋆ ⊗ O E ˇ u κ u , A ⋆ ⊗ O E ˇ u κ u ) ⊗ Q V^{\star}:=\operatorname{Hom}_{O_{E,(\diamond)}}\left(A^{\star}_{\diamond}\otimes_{O_{\check{E}_{u}}}\kappa_{u},A^{\star}\otimes_{O_{\check{E}_{u}}}\kappa_{u}\right)\otimes\mathbb{Q} V ⋆ := Hom O E , ( ⋄ ) ( A ⋄ ⋆ ⊗ O E ˇ u κ u , A ⋆ ⊗ O E ˇ u κ u ) ⊗ Q
over E / F E/F E / F (and put H ⋆ : = U ( V ⋆ ) H^{\star}:=U(V^{\star}) H ⋆ := U ( V ⋆ ) );
(U2): an isometry ( V ⋆ ) ∞ , ⋄ ≃ V ∞ , ⋄ (V^{\star})^{\infty,\diamond}\simeq V^{\infty,\diamond} ( V ⋆ ) ∞ , ⋄ ≃ V ∞ , ⋄ given by η ⋄ ⋆ \eta^{\diamond\star} η ⋄⋆ , and an ( L ⋄ b ) v (L_{\diamond}^{b})_{v} ( L ⋄ b ) v -class of isometries V ⋆ ⊗ F F v ≃ V ⊗ A F F v V^{\star}\otimes_{F}F_{v}\simeq V\otimes_{A_{F}}F_{v} V ⋆ ⊗ F F v ≃ V ⊗ A F F v for every v ∈ V F ( ⋄ ) ∖ { v ( u ) } v\in\mathcal{V}_{F}^{(\diamond)}\setminus\{v(u)\} v ∈ V F ( ⋄ ) ∖ { v ( u )} ;
(U3): a relative Rapoport–Zink space, that is, a formal scheme N b \mathcal{N}^{b} N b locally of finite type over O E ˇ u O_{\check{E}_{u}} O E ˇ u classifying quasi-isogenies of ( A ⋆ [ p v ( u ) ] , λ ⋆ [ p v ( u ) ] ) ⊗ O E ˇ u κ u (A^{\star}[p_{v(u)}],\lambda^{\star}[p_{v(u)}])\otimes_{O_{\check{E}_{u}}}\kappa_{u} ( A ⋆ [ p v ( u ) ] , λ ⋆ [ p v ( u ) ]) ⊗ O E ˇ u κ u of signature ( 1 , 1 ) (1,1) ( 1 , 1 ) and type t v ( u ) t_{v(u)} t v ( u ) [30 (5.20)] .
Denote by X L ⋄ , U ⋄ b \mathfrak{X}^{b}_{L_{\diamond},U^{\diamond}} X L ⋄ , U ⋄ b the formal completion of X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) O E ˇ u \mathfrak{X}^{b}_{L_{\diamond},U^{\diamond},(\diamond)}\otimes_{O_{E,(\diamond)}}O_{\check{E}_{u}} X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) O E ˇ u along the basic locus. By [30 Theorem 6.8] , we have the following isomorphism of basic uniformization
X L ⋄ , U ⋄ b ≃ ( H ⋆ ( F ) × O E 0 , ( ⋄ ) × ) \ ( N b × H ( A F ∞ , v ( u ) ) / ( L ⋄ b L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ ) (43) \mathfrak{X}^{b}_{L_{\diamond},U^{\diamond}}\simeq\left(H^{\star}(F)\times O^{\times}_{E_{0,(\diamond)}}\right)\backslash\left(\mathcal{N}^{b}\times H(A_{F}^{\infty,v(u)})/(L_{\diamond}^{b}L^{\diamond})^{v(u)}\times T(A^{\infty,\diamond})/U^{\diamond}\right) \tag*{(43)} X L ⋄ , U ⋄ b ≃ ( H ⋆ ( F ) × O E 0 , ( ⋄ ) × ) \ ( N b × H ( A F ∞ , v ( u ) ) / ( L ⋄ b L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ ) (43)
functorial in L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ ) ) L^{\diamond}\in\operatorname{Hk}(H(A_{F}^{\infty,\diamond})) L ⋄ ∈ Hk ( H ( A F ∞ , ⋄ )) , together with a quasi-isogeny
A ⋆ ⊗ O E ˇ u ( N L ⋄ , U ⋄ b ⊗ O E ˇ u κ u ) ⟶ A L ⋄ , U ⋄ , ( ⋄ ) ⊗ X L ⋄ , U ⋄ , ( ⋄ ) b ( X L ⋄ , U ⋄ b ⊗ O E ˇ u κ u ) (44) A^{\star}\otimes_{O_{\check{E}_{u}}}\left(\mathcal{N}^{b}_{L_{\diamond},U^{\diamond}}\otimes_{O_{\check{E}_{u}}}\kappa_{u}\right)\longrightarrow A_{L_{\diamond},U^{\diamond},(\diamond)}\otimes_{\mathfrak{X}^{b}_{L_{\diamond},U^{\diamond},(\diamond)}}\left(\mathfrak{X}^{b}_{L_{\diamond},U^{\diamond}}\otimes_{O_{\check{E}_{u}}}\kappa_{u}\right) \tag*{(44)} A ⋆ ⊗ O E ˇ u ( N L ⋄ , U ⋄ b ⊗ O E ˇ u κ u ) ⟶ A L ⋄ , U ⋄ , ( ⋄ ) ⊗ X L ⋄ , U ⋄ , ( ⋄ ) b ( X L ⋄ , U ⋄ b ⊗ O E ˇ u κ u ) (44)
preserving polarizations, where
N L ⋄ , U ⋄ b : = N b × H ( A F ∞ , v ( u ) ) / ( L ⋄ b L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ . \mathcal{N}^{b}_{L_{\diamond},U^{\diamond}}:=\mathcal{N}^{b}\times H(A_{F}^{\infty,v(u)})/(L_{\diamond}^{b}L^{\diamond})^{v(u)}\times T(A^{\infty,\diamond})/U^{\diamond}. N L ⋄ , U ⋄ b := N b × H ( A F ∞ , v ( u ) ) / ( L ⋄ b L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ .
Note that when V v ( u ) V_{v(u)} V v ( u ) is nonsplit, the entire special fiber X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u \mathfrak{X}^{b}_{L_{\diamond},U^{\diamond},(\diamond)}\otimes_{O_{E,(\diamond)}}\kappa_{u} X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u is basic.
Definition 4.20. Let R R R be a Dedekind domain with the generic point η \eta η and X \mathcal{X} X a scheme over R R R locally of finite type.
We denote by X f l \mathcal{X}^{\mathrm{fl}} X fl the flat closure of X \mathcal{X} X in its generic fiber X η \mathcal{X}_{\eta} X η .
For a finite étale cover X ′ → X η \mathcal{X}'\to\mathcal{X}_{\eta} X ′ → X η over η \eta η in which X ′ \mathcal{X}' X ′ is smooth of dimension one, we define the regular closure of X \mathcal{X} X in X ′ \mathcal{X}' X ′ to be the canonical resolution of (isolated) singularities of the normalization of X f l \mathcal{X}^{\mathrm{fl}} X fl in X ′ \mathcal{X}' X ′ .
Note that the regular closure is regular and admits a natural map to X \mathcal{X} X .
We have the similar definition when R R R is a complete local ring, X \mathcal{X} X is a formal scheme over R R R locally of finite type, and X ′ → X η \mathcal{X}'\to\mathcal{X}_{\eta} X ′ → X η is a finite étale cover (of rigid analytic spaces) over η \eta η in which X ′ \mathcal{X}' X ′ is a smooth rigid analytic space of dimension one.
Now we consider an open compact subgroup L ⋄ L_{\diamond} L ⋄ of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) together with an optimal one L ⋄ b L^{b}_{\diamond} L ⋄ b containing L ⋄ L_{\diamond} L ⋄ .
Definition 4.21. We define a functor
X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) ⟶ Sch / O E , ( ⋄ ) f l , p r o j \mathcal{X}_{\bullet,\bullet,(\diamond)}:\operatorname{Hk}(H(A_{F}^{\infty,\diamond}))\times\operatorname{Hk}(T(A^{\infty,\diamond}))\longrightarrow\operatorname{Sch}^{\mathrm{fl,proj}}_{/O_{E,(\diamond)}} X ∙ , ∙ , ( ⋄ ) : Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) ⟶ Sch / O E , ( ⋄ ) fl , proj
such that for every objects ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ ) ) × Hk ( T ( A ∞ , ⋄ ) ) (L^{\diamond},U^{\diamond})\in\operatorname{Hk}(H(A_{F}^{\infty,\diamond}))\times\operatorname{Hk}(T(A^{\infty,\diamond})) ( L ⋄ , U ⋄ ) ∈ Hk ( H ( A F ∞ , ⋄ )) × Hk ( T ( A ∞ , ⋄ )) , X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) is the regular closure of X L ⋄ , U ⋄ , ( ⋄ ) b X^{b}_{L_{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) b in X L ⋄ L ⋄ , U ⋄ U ⋄ X_{L_{\diamond}L^{\diamond},U_{\diamond}U^{\diamond}} X L ⋄ L ⋄ , U ⋄ U ⋄ .29
Choose an object U ⋄ ∈ Hk ( T ( A ∞ , ⋄ ) ) U^{\diamond}\in\operatorname{Hk}(T(A^{\infty,\diamond})) U ⋄ ∈ Hk ( T ( A ∞ , ⋄ )) and take a place u ∈ V E ( p ) u\in\mathcal{V}_{E}^{(p)} u ∈ V E ( p ) as above. Denote by N η → N η b \mathcal{N}_{\eta}\to\mathcal{N}^{b}_{\eta} N η → N η b the classifying space of L v ( u ) L_{v(u)} L v ( u ) -level structure over N η b \mathcal{N}^{b}_{\eta} N η b —the rigid analytic space of N b \mathcal{N}^{b} N b , which is a finite étale cover, and define N \mathcal{N} N to be the regular closure of N b \mathcal{N}^{b} N b in N η \mathcal{N}_{\eta} N η . If we denote by X L ⋄ , U ⋄ \mathfrak{X}_{L_{\diamond},U^{\diamond}} X L ⋄ , U ⋄ the formal completion of X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˇ u X_{L^{\diamond},U^{\diamond},(\diamond)}\otimes_{O_{E,(\diamond)}}O_{\check{E}_{u}} X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˇ u along the basic locus, then since taking regular closure is functorial and commutes with taking formalization, (4.6) implies the following isomorphism of basic uniformization
X L ⋄ , U ⋄ ≃ ( H ⋆ ( F ) × O E 0 , ( ⋄ ) × ) \ ( N × H ( A F ∞ , v ( u ) ) / ( L ⋄ L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ ) (45) \mathfrak{X}_{L_{\diamond},U^{\diamond}}\simeq\left(H^{\star}(F)\times O^{\times}_{E_{0,(\diamond)}}\right)\backslash\left(\mathcal{N}\times H(A_{F}^{\infty,v(u)})/(L_{\diamond}L^{\diamond})^{v(u)}\times T(A^{\infty,\diamond})/U^{\diamond}\right) \tag*{(45)} X L ⋄ , U ⋄ ≃ ( H ⋆ ( F ) × O E 0 , ( ⋄ ) × ) \ ( N × H ( A F ∞ , v ( u ) ) / ( L ⋄ L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ ) (45)
functorial in L ⋄ ∈ H k ( H ( A F ∞ , ⋄ ) ) L^{\diamond} \in H^k(H(A_F^{\infty,\diamond})) L ⋄ ∈ H k ( H ( A F ∞ , ⋄ )) .
Proof of Theorem 4.15. As we have explained, it suffices to consider the case where λ ≠ 0 \lambda\ne0 λ = 0 . Take ⋄ \diamond ⋄ to be the smallest set of rational primes that contains those ramified in E E E such that L ⋄ L^{\diamond} L ⋄ is the stabilizer of a self-dual lattice of V ∞ , ⋄ V^{\infty,\diamond} V ∞ , ⋄ . We may shrink L ⋄ L^{\diamond} L ⋄ so that it is contained in an optimal open compact subgroup L b L^{b} L b of H ( F ⋄ ) H(F_{\diamond}) H ( F ⋄ ) . We still denote by A 0 , ∙ , ∙ , ( ⋄ ) A_{0,\bullet,\bullet,(\diamond)} A 0 , ∙ , ∙ , ( ⋄ ) and A ∙ , ∙ , ( ⋄ ) A_{\bullet,\bullet,(\diamond)} A ∙ , ∙ , ( ⋄ ) their restrictions to X ∙ , ∙ , ( ⋄ ) X_{\bullet,\bullet,(\diamond)} X ∙ , ∙ , ( ⋄ ) .
Choose an object U ⋄ ∈ H k ( T ( A ∞ , ⋄ ) ) U^{\diamond} \in H^k(T(\mathbb{A}^{\infty,\diamond})) U ⋄ ∈ H k ( T ( A ∞ , ⋄ )) and put U = U ⋄ U ⋄ U=U_{\diamond}U^{\diamond} U = U ⋄ U ⋄ with U ⋄ U_{\diamond} U ⋄ the maximal open compact subgroup of T ( Q ⋄ ) T(\mathbb{Q}_{\diamond}) T ( Q ⋄ ) . Recall that we have the map
st L , U λ , + : H 2 ( X L , U , M L , U , λ ( 1 ) ) ⟶ CH 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⊗ E λ . \operatorname{st}_{L,U}^{\lambda,+}: H^2(X_{L,U},M_{L,U,\lambda}(1)) \longrightarrow\operatorname{CH}^{1+|b|}(B_{L,U}^{|b|}) \otimes E^{\lambda}. st L , U λ , + : H 2 ( X L , U , M L , U , λ ( 1 )) ⟶ CH 1 + ∣ b ∣ ( B L , U ∣ b ∣ ) ⊗ E λ .
which is canonically a direct summand. By Remark 3.19, we may extend the relative motive M L ⋄ , U ⋄ , λ M_{L^{\diamond},U^{\diamond},\lambda} M L ⋄ , U ⋄ , λ to an object in Mot ( X L ⋄ , U ⋄ , ( ⋄ ) , E λ ) \operatorname{Mot}(X_{L^{\diamond},U^{\diamond},(\diamond)},E^{\lambda}) Mot ( X L ⋄ , U ⋄ , ( ⋄ ) , E λ ) (with the same notation) as the abstract automorphic motive over X L ⋄ , U ⋄ , ( ⋄ ) X_{L^{\diamond},U^{\diamond},(\diamond)} X L ⋄ , U ⋄ , ( ⋄ ) of weight λ \lambda λ associated with ( A 0 , L ⋄ , U ⋄ , ( ⋄ ) , λ 0 ) (A_{0,L^{\diamond},U^{\diamond},(\diamond)},\lambda_0) ( A 0 , L ⋄ , U ⋄ , ( ⋄ ) , λ 0 ) and ( A L ⋄ , U ⋄ , ( ⋄ ) , λ ) (A_{L^{\diamond},U^{\diamond},(\diamond)},\lambda) ( A L ⋄ , U ⋄ , ( ⋄ ) , λ ) , together with a canonical direct summand map
st L ⋄ , U ⋄ , ( ⋄ ) λ , + : H 2 ( X L ⋄ , U ⋄ , ( ⋄ ) , M L , U , λ ( 1 ) ) ⟶ CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ (46) \operatorname{st}_{L^{\diamond},U^{\diamond},(\diamond)}^{\lambda,+}: H^2(X_{L^{\diamond},U^{\diamond},(\diamond)},M_{L,U,\lambda}(1)) \longrightarrow\operatorname{CH}^{1+|b|}(B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|}) \otimes E^{\lambda} \tag*{(46)} st L ⋄ , U ⋄ , ( ⋄ ) λ , + : H 2 ( X L ⋄ , U ⋄ , ( ⋄ ) , M L , U , λ ( 1 )) ⟶ CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ (46)
by Remark A.2(1), where B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ : = A 0 , L ⋄ , U ⋄ , ( ⋄ ) × X L ⋄ , U ⋄ , ( ⋄ ) A L ⋄ , U ⋄ , ( ⋄ ) ∨ B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|}:=A_{0,L^{\diamond},U^{\diamond},(\diamond)}\times_{X_{L^{\diamond},U^{\diamond},(\diamond)}}A_{L^{\diamond},U^{\diamond},(\diamond)}^{\vee} B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ := A 0 , L ⋄ , U ⋄ , ( ⋄ ) × X L ⋄ , U ⋄ , ( ⋄ ) A L ⋄ , U ⋄ , ( ⋄ ) ∨ .
Now take a pair ( c 1 , c 2 ) (c_1,c_2) ( c 1 , c 2 ) of elements in H 2 ( X L , U , M L , λ ( 1 ) ) H^2(X_{L,U},M_{L,\lambda}(1)) H 2 ( X L , U , M L , λ ( 1 )) . For i = 1 , 2 i=1,2 i = 1 , 2 , put c i ± : = st L , U λ , ± ( π L , U ∗ ( c i ) ) c_i^{\pm}:=\operatorname{st}_{L,U}^{\lambda,\pm}(\pi_{L,U}^{*}(c_i)) c i ± := st L , U λ , ± ( π L , U ∗ ( c i )) . By the construction of (38) , the goal is to show that
⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ \left\langle c_1^{+},c_2^{-}\right\rangle_{B_{L,U}^{|b|}}^{\ell} ⟨ c 1 + , c 2 − ⟩ B L , U ∣ b ∣ ℓ
takes value in C \mathbb{C} C and is independent of ℓ \ell ℓ for every ℓ ∉ ⋄ \ell\notin\diamond ℓ ∈ / ⋄ . Now, using the direct summand map st L ⋄ , U ⋄ , ( ⋄ ) λ , + \operatorname{st}_{L^{\diamond},U^{\diamond},(\diamond)}^{\lambda,+} st L ⋄ , U ⋄ , ( ⋄ ) λ , + above, we may assume that c 1 + c_1^{+} c 1 + extends to a class c 1 , ( ⋄ ) ∈ CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ c_{1,(\diamond)}\in\operatorname{CH}^{1+|b|}(B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|})\otimes E^{\lambda} c 1 , ( ⋄ ) ∈ CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ that is contained in the image of st L ⋄ , U ⋄ , ( ⋄ ) λ , + \operatorname{st}_{L^{\diamond},U^{\diamond},(\diamond)}^{\lambda,+} st L ⋄ , U ⋄ , ( ⋄ ) λ , + . Choose an arbitrary extension c 2 , ( ⋄ ) ∈ CH 1 + ( g − 1 ) ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ c_{2,(\diamond)}\in\operatorname{CH}^{1+(g-1)|b|}(B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|})\otimes E^{\lambda} c 2 , ( ⋄ ) ∈ CH 1 + ( g − 1 ) ∣ b ∣ ( B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ ) ⊗ E λ of c 2 − c_2^{-} c 2 − . By the moving lemma, we may assume that c 1 , ( ⋄ ) c_{1,(\diamond)} c 1 , ( ⋄ ) and c 2 , ( ⋄ ) c_{2,(\diamond)} c 2 , ( ⋄ ) are represented by algebraic cycles C 1 , ( ⋄ ) C_{1,(\diamond)} C 1 , ( ⋄ ) and C 2 , ( ⋄ ) C_{2,(\diamond)} C 2 , ( ⋄ ) on B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|} B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ (with coefficients in E λ E^{\lambda} E λ ) with disjoint support on B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ B_{L^{\diamond},U^{\diamond},(\diamond)}^{|b|} B L ⋄ , U ⋄ , ( ⋄ ) ∣ b ∣ . For i = 1 , 2 i=1,2 i = 1 , 2 , put C i : = C i , ( ⋄ ) ∣ B L , U ∣ b ∣ C_i:=C_{i,(\diamond)}|_{B_{L,U}^{|b|}} C i := C i , ( ⋄ ) ∣ B L , U ∣ b ∣ , whose support is denoted by D i D_i D i . By the decomposition formula for the usual Beilinson–Bloch height pairing, it suffices to show that for every u ∈ X E f i n u\in\mathfrak{X}^{\mathrm{fin}}_E u ∈ X E fin ,
⟨ C 1 , C 2 ⟩ B L , U ∣ b ∣ , D , u ℓ \left\langle C_1,C_2\right\rangle_{B_{L,U}^{|b|},D,u}^{\ell} ⟨ C 1 , C 2 ⟩ B L , U ∣ b ∣ , D , u ℓ
takes value in E λ E^{\lambda} E λ and is independent of ℓ \ell ℓ for every ℓ ∉ ⋄ \ell\notin\diamond ℓ ∈ / ⋄ , where D = ( D 1 , D 2 ) D=(D_1,D_2) D = ( D 1 , D 2 ) . When u ∉ X E ( ⋄ ) u\notin\mathfrak{X}^{(\diamond)}_E u ∈ / X E ( ⋄ ) , this holds since B L , U ∣ b ∣ B_{L,U}^{|b|} B L , U ∣ b ∣ has smooth projective reduction at u u u . It remains to consider places in X E ( ⋄ ) \mathfrak{X}^{(\diamond)}_E X E ( ⋄ ) .
Take an element u ∈ X E ( ⋄ ) u\in\mathfrak{X}^{(\diamond)}_E u ∈ X E ( ⋄ ) with v = v ( u ) v=v(u) v = v ( u ) . Write
X L ⋄ , U ⋄ : = X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u , B L ⋄ , U ⋄ : = B L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u , X_{L^{\diamond},U^{\diamond}}:=X_{L^{\diamond},U^{\diamond},(\diamond)}\otimes_{\mathcal{O}_{E,(\diamond)}}\mathcal{O}_{\bar{E}_u},\qquad B_{L^{\diamond},U^{\diamond}}:=B_{L^{\diamond},U^{\diamond},(\diamond)}\otimes_{\mathcal{O}_{E,(\diamond)}}\mathcal{O}_{\bar{E}_u}, X L ⋄ , U ⋄ := X L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u , B L ⋄ , U ⋄ := B L ⋄ , U ⋄ , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u ,
and C i : = C i , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u C_i:=C_{i,(\diamond)}\otimes_{\mathcal{O}_{E,(\diamond)}}\mathcal{O}_{\bar{E}_u} C i := C i , ( ⋄ ) ⊗ O E , ( ⋄ ) O E ˉ u for i = 1 , 2 i=1,2 i = 1 , 2 . Below we will regard H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) H^2(X_{L^{\diamond},U^{\diamond}},M_{L,U,\lambda}(1)) H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 )) as a direct summand of CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ ∣ b ∣ ) ⊗ E λ \operatorname{CH}^{1+|b|}(B_{L^{\diamond},U^{\diamond}}^{|b|})\otimes E^{\lambda} CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ ∣ b ∣ ) ⊗ E λ via (46) . For every ℓ ∉ ⋄ \ell\notin\diamond ℓ ∈ / ⋄ , put
H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) ⟨ ℓ ⟩ : = ker ( H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) ⟶ H e ˊ t 2 ( X L ⋄ , U ⋄ , r ℓ M L , U , λ ( 1 ) ) ) . H^2(X_{L^{\diamond},U^{\diamond}},M_{L,U,\lambda}(1))^{\langle\ell\rangle}:=\ker\left(H^2(X_{L^{\diamond},U^{\diamond}},M_{L,U,\lambda}(1))\longrightarrow H^2_{\mathrm{\acute{e}t}}(X_{L^{\diamond},U^{\diamond}},r_{\ell}M_{L,U,\lambda}(1))\right). H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) ⟨ ℓ ⟩ := ker ( H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 )) ⟶ H e ˊ t 2 ( X L ⋄ , U ⋄ , r ℓ M L , U , λ ( 1 )) ) .
Similar to Lemma 3.20 and (26) , we have a canonical isomorphism r ℓ M L , U , λ ≃ V L , λ r_{\ell}M_{L,U,\lambda}\simeq V_{L,\lambda} r ℓ M L , U , λ ≃ V L , λ over X L ⋄ , U ⋄ X_{L^{\diamond},U^{\diamond}} X L ⋄ , U ⋄ . Our strategy is the following: Find an algebraic cycle C 1 † C_1^{\dagger} C 1 † on B L ⋄ , U ⋄ ∣ b ∣ B_{L^{\diamond},U^{\diamond}}^{|b|} B L ⋄ , U ⋄ ∣ b ∣ with coefficients in E λ E^{\lambda} E λ whose restriction to B L , U ∣ b ∣ ⊗ E E ˉ u B_{L,U}^{|b|}\otimes_E\bar{E}_u B L , U ∣ b ∣ ⊗ E E ˉ u coincides with that of C 1 C_1 C 1 , such that its image in CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ ∣ b ∣ ) ⊗ E λ \operatorname{CH}^{1+|b|}(B_{L^{\diamond},U^{\diamond}}^{|b|})\otimes E^{\lambda} CH 1 + ∣ b ∣ ( B L ⋄ , U ⋄ ∣ b ∣ ) ⊗ E λ belongs to the subspace H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) ⟨ ℓ ⟩ H^2(X_{L^{\diamond},U^{\diamond}},M_{L,U,\lambda}(1))^{\langle\ell\rangle} H 2 ( X L ⋄ , U ⋄ , M L , U , λ ( 1 ) ) ⟨ ℓ ⟩ for every ℓ ∉ ⋄ \ell\notin\diamond ℓ ∈ / ⋄ . Then by [20 Proposition B.10] ,
⟨ C 1 , C 2 ⟩ B L , U ∣ b ∣ , D , u ℓ = C 1 † ⋅ C ‾ 2 . \left\langle C_1,C_2\right\rangle_{B_{L,U}^{|b|},D,u}^{\ell}=C_1^{\dagger}\cdot\overline{C}_2. ⟨ C 1 , C 2 ⟩ B L , U ∣ b ∣ , D , u ℓ = C 1 † ⋅ C 2 .
hence is in E ℓ E^{\ell} E ℓ and is independent of ℓ \ell ℓ for every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ . By Lemma 4.22(2) below, it suffices to verify that the cycle class of C 1 + C^{+}_{1} C 1 + in H e ˊ t 2 ( C , V ℓ , λ ( 1 ) ) H^{2}_{\mathrm{\acute et}}(C, V_{\ell,\lambda}(1)) H e ˊ t 2 ( C , V ℓ , λ ( 1 )) vanishes for every irreducible component C C C of X L ⋄ , U ⋄ ⊗ O E ˘ u κ u \mathcal{X}_{L^{\diamond},U^{\diamond}} \otimes_{\mathcal{O}_{\breve{E}_{u}}} \kappa_{u} X L ⋄ , U ⋄ ⊗ O E ˘ u κ u contained in the basic locus and every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ .30
Denote by M λ ⋆ M^{\star}_{\lambda} M λ ⋆ the abstract automorphic motive (Remark 3.19) over κ u \kappa_{u} κ u of weight λ \lambda λ associated with ( A 0 ⋆ , λ 0 ⋆ ) (A^{\star}_{0},\lambda^{\star}_{0}) ( A 0 ⋆ , λ 0 ⋆ ) and ( A ⋆ , λ ⋆ ) (A^{\star},\lambda^{\star}) ( A ⋆ , λ ⋆ ) . The two maps s t λ , + \mathrm{st}^{\lambda,+} st λ , + and s t λ , − \mathrm{st}^{\lambda,-} st λ , − from the same remark induce a hermitian pairing
H 0 ( Spec κ u , M λ ⋆ ) × H 0 ( Spec κ u , M λ ⋆ ) → E ℓ H^{0}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda}) \times H^{0}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda}) \to E^{\ell} H 0 ( Spec κ u , M λ ⋆ ) × H 0 ( Spec κ u , M λ ⋆ ) → E ℓ
via the Poincaré duality on ( B ⋆ ) ∣ b ∣ (B^{\star})^{|b|} ( B ⋆ ) ∣ b ∣ , where B ⋆ : = A 0 ⋆ × κ u ( A ⋆ ) ∨ B^{\star}:=A^{\star}_{0}\times_{\kappa_{u}}(A^{\star})^{\vee} B ⋆ := A 0 ⋆ × κ u ( A ⋆ ) ∨ . Put V λ ⋆ : = H n u m 0 ( Spec κ u , M λ ⋆ ) V^{\star}_{\lambda}:=H^{0}_{\mathrm{num}}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda}) V λ ⋆ := H num 0 ( Spec κ u , M λ ⋆ ) – the E ℓ E^{\ell} E ℓ -linear space of numerical equivalence classes of H 0 ( Spec κ u , M λ ⋆ ) H^{0}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda}) H 0 ( Spec κ u , M λ ⋆ ) under the above pairing.31 By Lemma 4.23 below, the natural map V λ ⋆ ⊗ Q ℓ → r ℓ M λ ⋆ V^{\star}_{\lambda}\otimes\mathbb{Q}_{\ell}\to r_{\ell}M^{\star}_{\lambda} V λ ⋆ ⊗ Q ℓ → r ℓ M λ ⋆ is an isomorphism for every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ . On the other hand, the level structure η ⋆ \eta^{\star} η ⋆ (at ℓ \ell ℓ alone) induces an isomorphism r ℓ M λ ⋆ ≃ V ℓ , λ : = ( V ℓ ) λ r_{\ell}M^{\star}_{\lambda}\simeq V_{\ell,\lambda}:=(V_{\ell})_{\lambda} r ℓ M λ ⋆ ≃ V ℓ , λ := ( V ℓ ) λ . In summary, the ℓ \ell ℓ -adic realization map induces an isomorphism
V λ ⋆ ⊗ Q ℓ ≃ V ℓ , λ (47) V^{\star}_{\lambda}\otimes\mathbb{Q}_{\ell}\simeq V_{\ell,\lambda} \tag*{(47)} V λ ⋆ ⊗ Q ℓ ≃ V ℓ , λ (47)
for every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ , depending only on the framing P ⋆ P^{\star} P ⋆ .
Denote by N L ⋄ , U ⋄ N_{L^{\diamond},U^{\diamond}} N L ⋄ , U ⋄ the normalization of the induced reduced subscheme of
N L ⋄ , U ⋄ : = N × H ( A F ∞ , v ( u ) ) / ( L ⋄ L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ , \mathcal{N}_{L^{\diamond},U^{\diamond}}:=\mathcal{N}\times H(A_{F}^{\infty,v(u)})/(L_{\diamond}L^{\diamond})^{v(u)}\times T(A^{\infty,\diamond})/U^{\diamond}, N L ⋄ , U ⋄ := N × H ( A F ∞ , v ( u ) ) / ( L ⋄ L ⋄ ) v ( u ) × T ( A ∞ , ⋄ ) / U ⋄ ,
and by ∣ N L ⋄ , U ⋄ ∣ |N_{L^{\diamond},U^{\diamond}}| ∣ N L ⋄ , U ⋄ ∣ the set of its irreducible components. Consider an element C ∈ ∣ N L ⋄ , U ⋄ ∣ C\in|N_{L^{\diamond},U^{\diamond}}| C ∈ ∣ N L ⋄ , U ⋄ ∣ with π C : C → Spec κ u \pi_{C}:C\to\operatorname{Spec}\kappa_{u} π C : C → Spec κ u the (smooth projective) structure map. Using the maps s t λ , + \mathrm{st}^{\lambda,+} st λ , + and s t λ , − \mathrm{st}^{\lambda,-} st λ , − in Remark 3.19 (with S = C S=C S = C ), we obtain a hermitian pairing
H 2 ( C , M L , U , λ ( 1 ) ) × H 2 ( C , M L , U , λ ( 1 ) ) → E ℓ H^{2}(C,M_{L,U,\lambda}(1))\times H^{2}(C,M_{L,U,\lambda}(1))\to E^{\ell} H 2 ( C , M L , U , λ ( 1 )) × H 2 ( C , M L , U , λ ( 1 )) → E ℓ
via the Poincaré duality on B L ⋄ , U ⋄ ∣ b ∣ × X L ⋄ , U ⋄ C \mathcal{B}^{|b|}_{L^{\diamond},U^{\diamond}}\times_{\mathcal{X}_{L^{\diamond},U^{\diamond}}}C B L ⋄ , U ⋄ ∣ b ∣ × X L ⋄ , U ⋄ C . Denote by H n u m 2 ( C , M L , U , λ ( 1 ) ) H^{2}_{\mathrm{num}}(C,M_{L,U,\lambda}(1)) H num 2 ( C , M L , U , λ ( 1 )) the E ℓ E^{\ell} E ℓ -linear space of numerical equivalence classes of H 2 ( C , M L , U , λ ( 1 ) ) H^{2}(C,M_{L,U,\lambda}(1)) H 2 ( C , M L , U , λ ( 1 )) under the above pairing. By [17 Theorem 3.5] , the map
H n u m 2 ( C , M L , U , λ ( 1 ) ) ⊗ Q ℓ → H e ˊ t 2 ( C , r ℓ M L , U , λ ( 1 ) ) (48) H^{2}_{\mathrm{num}}(C,M_{L,U,\lambda}(1))\otimes\mathbb{Q}_{\ell}\to H^{2}_{\mathrm{\acute et}}(C,r_{\ell}M_{L,U,\lambda}(1)) \tag*{(48)} H num 2 ( C , M L , U , λ ( 1 )) ⊗ Q ℓ → H e ˊ t 2 ( C , r ℓ M L , U , λ ( 1 )) (48)
induced by the ℓ \ell ℓ -adic realization is injective for every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ . The universal quasi-isogeny (44) gives rise to an isomorphism M L , U , λ ∣ C ≃ π C ⋆ M λ ⋆ M_{L,U,\lambda}|_{C}\simeq\pi_{C}^{\star}M^{\star}_{\lambda} M L , U , λ ∣ C ≃ π C ⋆ M λ ⋆ , which in particular gives rise to an isomorphism r ℓ M L , U , λ ∣ C ≃ π C ⋆ V ℓ , λ r_{\ell}M_{L,U,\lambda}|_{C}\simeq\pi_{C}^{\star}V_{\ell,\lambda} r ℓ M L , U , λ ∣ C ≃ π C ⋆ V ℓ , λ . Then the map (48) can be written as
H n u m 2 ( C , π C ⋆ M λ ⋆ ( 1 ) ) ⊗ Q ℓ → H e ˊ t 2 ( C , V ℓ , λ ( 1 ) ) = H e ˊ t 2 ( C , Q ℓ ( 1 ) ) ⊗ Q ℓ V ℓ , λ ≃ V ℓ , λ . H^{2}_{\mathrm{num}}(C,\pi_{C}^{\star}M^{\star}_{\lambda}(1))\otimes\mathbb{Q}_{\ell}\to H^{2}_{\mathrm{\acute et}}(C,V_{\ell,\lambda}(1))=H^{2}_{\mathrm{\acute et}}(C,\mathbb{Q}_{\ell}(1))\otimes_{\mathbb{Q}_{\ell}}V_{\ell,\lambda}\simeq V_{\ell,\lambda}. H num 2 ( C , π C ⋆ M λ ⋆ ( 1 )) ⊗ Q ℓ → H e ˊ t 2 ( C , V ℓ , λ ( 1 )) = H e ˊ t 2 ( C , Q ℓ ( 1 )) ⊗ Q ℓ V ℓ , λ ≃ V ℓ , λ .
Consider the Gysin map
H n u m 0 ( Spec κ u , M λ ⋆ ) → H n u m 2 ( C , π C ⋆ M λ ⋆ ( 1 ) ) (49) H^{0}_{\mathrm{num}}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda})\to H^{2}_{\mathrm{num}}(C,\pi_{C}^{\star}M^{\star}_{\lambda}(1)) \tag*{(49)} H num 0 ( Spec κ u , M λ ⋆ ) → H num 2 ( C , π C ⋆ M λ ⋆ ( 1 )) (49)
at an arbitrary closed point of C C C (which does not depend on the choice of such a point). By the injectivity of (48) and the isomorphism (47) , the consecutive maps
H n u m 0 ( Spec κ u , M λ ⋆ ) ⊗ Q ℓ → H n u m 2 ( C , π C ⋆ M λ ⋆ ( 1 ) ) ⊗ Q ℓ → H e ˊ t 2 ( C , Q ℓ ( 1 ) ) ⊗ Q ℓ V ℓ , λ H^{0}_{\mathrm{num}}(\operatorname{Spec}\kappa_{u},M^{\star}_{\lambda})\otimes\mathbb{Q}_{\ell}\to H^{2}_{\mathrm{num}}(C,\pi_{C}^{\star}M^{\star}_{\lambda}(1))\otimes\mathbb{Q}_{\ell}\to H^{2}_{\mathrm{\acute et}}(C,\mathbb{Q}_{\ell}(1))\otimes_{\mathbb{Q}_{\ell}}V_{\ell,\lambda} H num 0 ( Spec κ u , M λ ⋆ ) ⊗ Q ℓ → H num 2 ( C , π C ⋆ M λ ⋆ ( 1 )) ⊗ Q ℓ → H e ˊ t 2 ( C , Q ℓ ( 1 )) ⊗ Q ℓ V ℓ , λ
are both isomorphisms for every ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ , which implies that (49) is also an isomorphism.
By (45) and the isomorphism (49) , the image of the natural composite map
H 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) ⟶ ⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H 2 ( C , M L , U , λ ( 1 ) ) H^{2}(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1)) \longrightarrow\bigoplus_{C\in|N_{L^\circ,U^\circ}|} H^{2}(C,M_{L,U,\lambda}(1)) H 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 )) ⟶ C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H 2 ( C , M L , U , λ ( 1 ))
⟶ ⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H n u m 2 ( C , M L , U , λ ( 1 ) ) → ∼ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) (50) \longrightarrow\bigoplus_{C\in|N_{L^\circ,U^\circ}|} H^{2}_{\mathrm{num}}(C,M_{L,U,\lambda}(1)) \xrightarrow{\sim} \operatorname{Fun}(|N_{L^\circ,U^\circ|},V_\lambda^*) \tag*{(50)} ⟶ C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H num 2 ( C , M L , U , λ ( 1 )) ∼ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) (50)
is contained in Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ \operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*)^\Gamma Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ , where Γ : = H ∗ ( F ) × O E 0 , ( ⋄ ) × \Gamma:= H^*(F)\times\mathcal{O}_{E_0,(\diamond)}^\times Γ := H ∗ ( F ) × O E 0 , ( ⋄ ) × acts on V λ ∗ V_\lambda^* V λ ∗ via the projection H ∗ ( F ) × O E 0 , ( ⋄ ) × → H ∗ ( F ) H^*(F)\times\mathcal{O}_{E_0,(\diamond)}^\times\to H^*(F) H ∗ ( F ) × O E 0 , ( ⋄ ) × → H ∗ ( F ) . On the other hand, if we denote by Y L ∘ , U ∘ Y_{L^\circ,U^\circ} Y L ∘ , U ∘ the special fiber of X L ∘ , U ∘ \mathcal{X}_{L^\circ,U^\circ} X L ∘ , U ∘ , then we have the Gysin map
ζ : H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) ⟶ H 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) . \zeta:H^2_{Y_{L^\circ,U^\circ}}(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1))\longrightarrow H^2(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1)). ζ : H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 )) ⟶ H 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 )) .
Composing ζ \zeta ζ with (50) , we obtain the map
H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ . (51) H^2_{Y_{L^\circ,U^\circ}}(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1))\longrightarrow\operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*)^\Gamma. \tag*{(51)} H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 )) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ . (51)
We claim that (51) is surjective. Assuming the claim, we may find an element
C 1 ′ ∈ H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) C'_1\in H^2_{Y_{L^\circ,U^\circ}}(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1)) C 1 ′ ∈ H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ))
whose image under (51) coincides with the image of C 1 C_1 C 1 under (50) . Then C 1 † : = C 1 − ζ C 1 ′ C_1^\dagger:=C_1-\zeta C'_1 C 1 † := C 1 − ζ C 1 ′ works for our strategy.
It remains to show the claim. Composing the natural composite map
⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H 0 ( Spec κ u , M λ ∗ ) → ⊕ π C ∗ ⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H 0 ( C , M L , U , λ ) ⟶ H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ) ) \bigoplus_{C\in|N_{L^\circ,U^\circ}|}H^0(\operatorname{Spec}\kappa_u,M_\lambda^*)\xrightarrow{\oplus\pi_C^*}\bigoplus_{C\in|N_{L^\circ,U^\circ}|}H^0(C,M_{L,U,\lambda})\longrightarrow H^2_{Y_{L^\circ,U^\circ}}(\mathcal{X}_{L^\circ,U^\circ},M_{L,U,\lambda}(1)) C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H 0 ( Spec κ u , M λ ∗ ) ⊕ π C ∗ C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H 0 ( C , M L , U , λ ) ⟶ H Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , M L , U , λ ( 1 ))
with (51) , we obtain the map
⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H 0 ( Spec κ u , M λ ∗ ) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) . (52) \bigoplus_{C\in|N_{L^\circ,U^\circ}|}H^0(\operatorname{Spec}\kappa_u,M_\lambda^*)\longrightarrow\operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*). \tag*{(52)} C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H 0 ( Spec κ u , M λ ∗ ) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) . (52)
It suffices to show that (52) is surjective. For every abelian group A A A , denote by Fun c ( ∣ N L ∘ , U ∘ ∣ , A ) ⊆ Fun ( ∣ N L ∘ , U ∘ ∣ , A ) \operatorname{Fun}_{c}(|N_{L^\circ,U^\circ}|,A)\subseteq\operatorname{Fun}(|N_{L^\circ,U^\circ}|,A) Fun c ( ∣ N L ∘ , U ∘ ∣ , A ) ⊆ Fun ( ∣ N L ∘ , U ∘ ∣ , A ) the subgroup of finitely supported functions (valued in A A A ), for which we have the map
Δ : Z [ ∣ N L ∘ , U ∘ ∣ ] ⟶ Fun c ( ∣ N L ∘ , U ∘ ∣ , Z ) \Delta:\mathbb{Z}[|N_{L^\circ,U^\circ}|]\longrightarrow\operatorname{Fun}_{c}(|N_{L^\circ,U^\circ}|,\mathbb{Z}) Δ : Z [ ∣ N L ∘ , U ∘ ∣ ] ⟶ Fun c ( ∣ N L ∘ , U ∘ ∣ , Z )
given by the intersection in (the regular formal scheme) N L ∘ , U ∘ \mathfrak{N}_{L^\circ,U^\circ} N L ∘ , U ∘ , and the map
av Γ : Fun c ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ \operatorname{av}^{\Gamma}:\operatorname{Fun}_{c}(|N_{L^\circ,U^\circ}|,V_\lambda^*)\longrightarrow\operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*)^\Gamma av Γ : Fun c ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ
sending f f f to ∑ γ ∈ Γ γ f \sum_{\gamma\in\Gamma}\gamma f ∑ γ ∈ Γ γ f . Then (52) is given by the composition
⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H 0 ( Spec κ u , M λ ∗ ) ⟶ ⨁ C ∈ ∣ N L ∘ , U ∘ ∣ H n u m 0 ( Spec κ u , M λ ∗ ) = Z [ ∣ N L ∘ , U ∘ ∣ ] ⊗ V λ ∗ \bigoplus_{C\in|N_{L^\circ,U^\circ}|}H^0(\operatorname{Spec}\kappa_u,M_\lambda^*)\longrightarrow\bigoplus_{C\in|N_{L^\circ,U^\circ}|}H^0_{\mathrm{num}}(\operatorname{Spec}\kappa_u,M_\lambda^*)=\mathbb{Z}[|N_{L^\circ,U^\circ}|]\otimes V_\lambda^* C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H 0 ( Spec κ u , M λ ∗ ) ⟶ C ∈ ∣ N L ∘ , U ∘ ∣ ⨁ H num 0 ( Spec κ u , M λ ∗ ) = Z [ ∣ N L ∘ , U ∘ ∣ ] ⊗ V λ ∗
→ Δ ⊗ V λ ∗ Fun c ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) → av Γ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ . \xrightarrow{\Delta\otimes V_\lambda^*}\operatorname{Fun}_{c}(|N_{L^\circ,U^\circ}|,V_\lambda^*)\xrightarrow{\operatorname{av}^{\Gamma}}\operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*)^\Gamma. Δ ⊗ V λ ∗ Fun c ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) av Γ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ .
Since the first map is surjective, it suffices to show that av Γ ∘ ( Δ ⊗ V λ ∗ ) \operatorname{av}^{\Gamma}\circ(\Delta\otimes V_\lambda^*) av Γ ∘ ( Δ ⊗ V λ ∗ ) is surjective, or rather the induced map
Δ λ : ( Z [ ∣ N L ∘ , U ∘ ∣ ] ⊗ V λ ∗ ) Γ ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ \Delta_\lambda:\bigl(\mathbb{Z}[|N_{L^\circ,U^\circ}|]\otimes V_\lambda^*\bigr)_\Gamma\longrightarrow\operatorname{Fun}(|N_{L^\circ,U^\circ}|,V_\lambda^*)^\Gamma Δ λ : ( Z [ ∣ N L ∘ , U ∘ ∣ ] ⊗ V λ ∗ ) Γ ⟶ Fun ( ∣ N L ∘ , U ∘ ∣ , V λ ∗ ) Γ
is surjective. However, by a similar argument, Δ λ ⊗ Q ℓ \Delta_\lambda\otimes\mathbb{Q}_\ell Δ λ ⊗ Q ℓ (for an arbitrary rational prime ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ ) can be canonically identified with the ℓ \ell ℓ -adic Gysin map
ζ ℓ : H e ˊ t , Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , r ℓ M L , U , λ ( 1 ) ) ⟶ H e ˊ t 2 ( X L ∘ , U ∘ , r ℓ M L , U , λ ( 1 ) ) . \zeta_\ell:H^2_{\mathrm{\acute et},Y_{L^\circ,U^\circ}}(\mathcal{X}_{L^\circ,U^\circ},r_\ell M_{L,U,\lambda}(1))\longrightarrow H^2_{\mathrm{\acute et}}(\mathcal{X}_{L^\circ,U^\circ},r_\ell M_{L,U,\lambda}(1)). ζ ℓ : H e ˊ t , Y L ∘ , U ∘ 2 ( X L ∘ , U ∘ , r ℓ M L , U , λ ( 1 )) ⟶ H e ˊ t 2 ( X L ∘ , U ∘ , r ℓ M L , U , λ ( 1 )) .
Note that the cokernel of ζ ℓ \zeta_\ell ζ ℓ is contained in the Gal ( E ~ u / E u ) \operatorname{Gal}(\widetilde{E}_u/E_u) Gal ( E u / E u ) -finite subspace of H e ˊ t 2 ( X L , U ⊗ E E E ~ u , r ℓ M L , U , λ ( 1 ) ) H^2_{\mathrm{\acute et}}(X_{L,U}\otimes_E E\widetilde{E}_u,r_\ell M_{L,U,\lambda}(1)) H e ˊ t 2 ( X L , U ⊗ E E E u , r ℓ M L , U , λ ( 1 )) , which vanishes when λ ≠ 0 \lambda\ne0 λ = 0 . It follows that av Γ ∘ ( Δ ⊗ V λ ∗ ) \operatorname{av}^{\Gamma}\circ(\Delta\otimes V_\lambda^*) av Γ ∘ ( Δ ⊗ V λ ∗ ) hence (51) are surjective.
The theorem is finally proved! □
Lemma 4.22. Let u ∈ M E ( ⋄ ) u \in\mathcal{M}^{(\diamond)}_E u ∈ M E ( ⋄ ) be an element such that V v ( u ) V_{v(u)} V v ( u ) is split.
The base change ( X L ⋄ , U ⋄ , ( ⋄ ) b ) f l ⊗ O E , ( ⋄ ) O E u (X^{b}_{L^\diamond,U^\diamond,(\diamond)})^{\mathrm{fl}} \otimes_{O_{E,(\diamond)}} O_{E_u} ( X L ⋄ , U ⋄ , ( ⋄ ) b ) fl ⊗ O E , ( ⋄ ) O E u is smooth over O E u O_{E_u} O E u .
For every irreducible component C C C of X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u X^{b}_{L^\diamond,U^\diamond,(\diamond)} \otimes_{O_{E,(\diamond)}} \kappa_u X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u that is not contained in the basic locus, we have
H e ˊ t 2 ( C , V ℓ , L , λ ( 1 ) ) = 0 H^2_{\mathrm{\acute{e}t}}(C,V_{\ell,L,\lambda}(1))=0 H e ˊ t 2 ( C , V ℓ , L , λ ( 1 )) = 0 for every ℓ ∉ ⋄ \ell\notin\diamond ℓ ∈ / ⋄ when λ ≠ 0 \lambda\ne0 λ = 0 .
Proof. For (1), it is straightforward to see that X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) O E u X^{b}_{L^\diamond,U^\diamond,(\diamond)} \otimes_{O_{E,(\diamond)}} O_{E_u} X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) O E u is already smooth away from the basic locus. Thus, by (43) , it suffices to show that ( N b ) f l (\mathcal{N}^{b})^{\mathrm{fl}} ( N b ) fl is a formal open unit disc of relative dimension one. This is known when v ( u ) v(u) v ( u ) splits in E E E . When v ( u ) v(u) v ( u ) is nonsplit in E E E , there is a natural map from the formal moduli M \mathcal{M} M of Lubin–Tate formal O E u O_{E_u} O E u -divisible groups to N b \mathcal{N}^{b} N b sending Y Y Y to Y ⊕ Y ‾ Y \oplus\overline{Y} Y ⊕ Y with the natural polarization λ \lambda λ , which factors through ( N b ) f l (\mathcal{N}^{b})^{\mathrm{fl}} ( N b ) fl as M \mathcal{M} M is flat and induces an isomorphism between generic fibers – this implies that ( N b ) f l ≃ M (\mathcal{N}^{b})^{\mathrm{fl}} \simeq\mathcal{M} ( N b ) fl ≃ M which is a formal open unit disc of relative dimension one.
For (2), denote by C b C^{b} C b the image of C C C in X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u X^{b}_{L^\diamond,U^\diamond,(\diamond)} \otimes_{O_{E,(\diamond)}} \kappa_u X L ⋄ , U ⋄ , ( ⋄ ) b ⊗ O E , ( ⋄ ) κ u , which is a smooth irreducible curve by (1). Denote by H d e r H^{\mathrm{der}} H der denotes the derived subgroup of H H H . It follows by (1) that the monodromy group of V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ over C b C^{b} C b is an open compact subgroup of H d e r ( F ℓ ) H^{\mathrm{der}}(F_\ell) H der ( F ℓ ) , acting via the representation ( V ℓ ) λ (V_\ell)_\lambda ( V ℓ ) λ which remains irreducible when restricting to H d e r ( F ℓ ) H^{\mathrm{der}}(F_\ell) H der ( F ℓ ) . It follows that H e ˊ t 2 ( C , V ℓ , L , λ ( 1 ) ) = 0 H^2_{\mathrm{\acute{e}t}}(C,V_{\ell,L,\lambda}(1))=0 H e ˊ t 2 ( C , V ℓ , L , λ ( 1 )) = 0 . □
Lemma 4.23. In Remark 3.19 suppose that S = Spec κ S=\operatorname{Spec}\kappa S = Spec κ for an algebraically closed field κ \kappa κ . If the E E E -vector space Hom E 0 ( A 0 , A ) \operatorname{Hom}_{E_0}(A_0,A) Hom E 0 ( A 0 , A ) is of dimension m m m , then the natural map
H 0 ( Spec κ , M λ ) ⊗ Q ℓ ⟶ H e ˊ t 0 ( Spec κ , r ℓ M λ ) = r ℓ M λ H^0(\operatorname{Spec}\kappa,\mathcal{M}_\lambda)\otimes\mathbb{Q}_\ell\longrightarrow H^0_{\mathrm{\acute{e}t}}(\operatorname{Spec}\kappa,r_\ell\mathcal{M}_\lambda)=r_\ell\mathcal{M}_\lambda H 0 ( Spec κ , M λ ) ⊗ Q ℓ ⟶ H e ˊ t 0 ( Spec κ , r ℓ M λ ) = r ℓ M λ is surjective for every rational prime ℓ \ell ℓ invertible on κ \kappa κ .
Proof. Take a rational prime ℓ \ell ℓ that is invertible on κ \kappa κ . It is clear that
r ℓ M = Hom E 0 , ℓ ( h e ˊ t 1 ( A 0 , Q ℓ ) , h e ˊ t 1 ( A , Q ℓ ) ) r_\ell\mathcal{M}=\operatorname{Hom}_{E_{0,\ell}}\left(h^1_{\mathrm{\acute{e}t}}(A_0,\mathbb{Q}_\ell),h^1_{\mathrm{\acute{e}t}}(A,\mathbb{Q}_\ell)\right) r ℓ M = Hom E 0 , ℓ ( h e ˊ t 1 ( A 0 , Q ℓ ) , h e ˊ t 1 ( A , Q ℓ ) )
in which h e ˊ t 1 ( − , Q ℓ ) = V ℓ ( − ) [ − 1 ] h^1_{\mathrm{\acute{e}t}}(-,\mathbb{Q}_\ell)=V_\ell(-)[-1] h e ˊ t 1 ( − , Q ℓ ) = V ℓ ( − ) [ − 1 ] is nothing but the rational ℓ \ell ℓ -adic Tate module placed in degree 1. It is well-known that the natural map
Hom E 0 ( A 0 , A ) ⊗ Q ℓ ⟶ Hom E 0 , ℓ ( V ℓ ( A 0 ) , V ℓ ( A ) ) (53) \operatorname{Hom}_{E_0}(A_0,A)\otimes\mathbb{Q}_\ell\longrightarrow\operatorname{Hom}_{E_{0,\ell}}(V_\ell(A_0),V_\ell(A)) \tag*{(53)} Hom E 0 ( A 0 , A ) ⊗ Q ℓ ⟶ Hom E 0 , ℓ ( V ℓ ( A 0 ) , V ℓ ( A )) (53)
is injective. Since the target is a free E ℓ E_\ell E ℓ -module of rank m m m , it follows that (53) is an isomorphism. On the other hand, (53) factors as
Hom E 0 ( A 0 , A ) ⊗ Q ℓ ⟶ Hom E 0 ( h 1 ( A 0 ) , h 1 ( A ) ) ⊗ Q ℓ = H 0 ( Spec κ , M ) ⊗ Q ℓ ⟶ Hom E 0 , ℓ ( V ℓ ( A 0 ) , V ℓ ( A ) ) , \operatorname{Hom}_{E_0}(A_0,A)\otimes\mathbb{Q}_\ell\longrightarrow\operatorname{Hom}_{E_0}(h^1(A_0),h^1(A))\otimes\mathbb{Q}_\ell=H^0(\operatorname{Spec}\kappa,\mathcal{M})\otimes\mathbb{Q}_\ell\longrightarrow\operatorname{Hom}_{E_{0,\ell}}(V_\ell(A_0),V_\ell(A)), Hom E 0 ( A 0 , A ) ⊗ Q ℓ ⟶ Hom E 0 ( h 1 ( A 0 ) , h 1 ( A )) ⊗ Q ℓ = H 0 ( Spec κ , M ) ⊗ Q ℓ ⟶ Hom E 0 , ℓ ( V ℓ ( A 0 ) , V ℓ ( A )) ,
which implies that the map
H 0 ( Spec κ , M ) ⊗ Q ℓ ⟶ r ℓ M H^0(\operatorname{Spec}\kappa,\mathcal{M})\otimes\mathbb{Q}_\ell\longrightarrow r_\ell\mathcal{M} H 0 ( Spec κ , M ) ⊗ Q ℓ ⟶ r ℓ M
of E ℓ E_\ell E ℓ -modules is surjective. Applying Construction 2.8(2) to H 0 ( Spec κ , M ) H^0(\operatorname{Spec}\kappa,\mathcal{M}) H 0 ( Spec κ , M ) (resp. r ℓ M r_\ell\mathcal{M} r ℓ M ), we obtain an E λ l E_\lambda^l E λ l -module (resp. E λ l E_\lambda^l E λ l -module) H 0 ( Spec κ , M ) λ H^0(\operatorname{Spec}\kappa,\mathcal{M})_\lambda H 0 ( Spec κ , M ) λ (resp. ( r ℓ M ) λ (r_\ell\mathcal{M})_\lambda ( r ℓ M ) λ ), such that the induced map
H 0 ( Spec κ , M ) λ ⊗ Q ℓ ⟶ ( r ℓ M ) λ H^0(\operatorname{Spec}\kappa,\mathcal{M})_\lambda\otimes\mathbb{Q}_\ell\longrightarrow(r_\ell\mathcal{M})_\lambda H 0 ( Spec κ , M ) λ ⊗ Q ℓ ⟶ ( r ℓ M ) λ
is surjective. However, ( r ℓ M ) λ (r_\ell\mathcal{M})_\lambda ( r ℓ M ) λ is canonically isomorphic to r ℓ M λ r_\ell\mathcal{M}_\lambda r ℓ M λ and the above map factors naturally as
H 0 ( Spec κ , M ) λ ⊗ Q ℓ ⟶ H 0 ( Spec κ , M λ ) ⊗ Q ℓ ⟶ r ℓ M λ . H^0(\operatorname{Spec}\kappa,\mathcal{M})_\lambda\otimes\mathbb{Q}_\ell\longrightarrow H^0(\operatorname{Spec}\kappa,\mathcal{M}_\lambda)\otimes\mathbb{Q}_\ell\longrightarrow r_\ell\mathcal{M}_\lambda. H 0 ( Spec κ , M ) λ ⊗ Q ℓ ⟶ H 0 ( Spec κ , M λ ) ⊗ Q ℓ ⟶ r ℓ M λ .
The lemma follows. □
Computation of local heights of motivic special classes
In this chapter, we compute local heights of motivic special classes in various situations. We fix a totally positive definite incoherent hermitian space V V V over A E / A F \mathbb{A}_{E}/\mathbb{A}_{F} A E / A F of rank 2 r 2r 2 r satisfying that the set B ( V ) \mathcal{B}(V) B ( V ) is not a singleton (and put H : = U ( V ) H := U(V) H := U ( V ) as always), and a balanced hermitian E E E -weight λ \lambda λ of rank 2 r 2r 2 r (Definition 2.7).
To simplify notation, write Ψ ∞ λ \Psi_{\infty}^{\lambda} Ψ ∞ λ and Φ ∞ λ \Phi_{\infty}^{\lambda} Φ ∞ λ for Ψ V ∞ 2 r λ \Psi_{V_{\infty}^{2r}}^{\lambda} Ψ V ∞ 2 r λ and Φ V ∞ 2 r λ \Phi_{V_{\infty}^{2r}}^{\lambda} Φ V ∞ 2 r λ from Construction 2.8(4) and Definition 2.23, respectively. It is clear that as functions on V ∞ 2 r V_{\infty}^{2r} V ∞ 2 r valued in S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ S_{\lambda+}(\mathbb{C}^{r})^{\vee} \otimes_{\mathbb{C}} S_{\lambda+}(\mathbb{C}^{r})^{\vee} S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ , both Ψ ∞ λ \Psi_{\infty}^{\lambda} Ψ ∞ λ and Φ ∞ λ \Phi_{\infty}^{\lambda} Φ ∞ λ factor through the moment map T : V ∞ 2 r → Herm 2 r ( F ∞ ) + T: V_{\infty}^{2r} \to\operatorname{Herm}_{2r}(F_{\infty})^{+} T : V ∞ 2 r → Herm 2 r ( F ∞ ) + .
Explicit local heights at inert places
In this section, we take a nonarchimedean place u u u of E E E such that v : = v ( u ) v := v(u) v := v ( u ) is inert in E E E and p : = p ( u ) p := p(u) p := p ( u ) is odd.
Consider an object L ∈ Hk ( H ( A F ∞ ) ) L \in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) L ∈ Hk ( H ( A F ∞ )) such that
L v L_{v} L v is the stabilizer of a self-dual or almost self-dual lattice (Definition 2.9), say Λ v \Lambda_{v} Λ v , of V v V_{v} V v , depending on whether V v V_{v} V v is split or nonsplit;
L v ′ L_{v'} L v ′ is the stabilizer of a vertex lattice (Definition 2.9), say Λ v ′ \Lambda_{v'} Λ v ′ , for every place v ′ ∈ B F ( p ) v' \in\mathcal{B}_{F}^{(p)} v ′ ∈ B F ( p ) that is nonsplit in E E E .
Let
X ∙ : Hk ( H ( A F ∞ , p ) ) ⟶ Sch / O E u f l . p r o j \mathcal{X}_{\bullet}: \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty,p})) \longrightarrow\operatorname{Sch}_{/O_{E_{u}}}^{\mathrm{fl.proj}} X ∙ : Hk ( H ( A F ∞ , p )) ⟶ Sch / O E u fl.proj
be the functor (with respect to L L L ) constructed in [22 §11.2 or §11.3] (see Remark 5.1 below), satisfying Assumption 4.6 with the further property that X L ′ \mathcal{X}_{L'} X L ′ is smooth or strictly semistable over O E u O_{E_{u}} O E u , depending on whether V v V_{v} V v is split or not.
Remark 5.1. The integral model X L ′ = X L ′ , p \mathcal{X}_{L'} = \mathcal{X}_{L',p} X L ′ = X L ′ , p is constructed in the following way: First, choose an object U p U^{p} U p of Hk ( T ( A F ∞ , p ) ) \operatorname{Hk}(T(\mathbb{A}_{F}^{\infty,p})) Hk ( T ( A F ∞ , p )) . In [22 §11.2 or §11.3] 32 the authors constructed a moduli problem X L ′ , U p \mathcal{X}_{L',U^{p}} X L ′ , U p classifying similar data in Definition 3.42 (together with the Eisenstein condition), hence admits a natural free action of Ξ U p \Xi_{U^{p}} Ξ U p . We then put X L ′ , p : = X L ′ , U p / Ξ U p \mathcal{X}_{L',p} := \mathcal{X}_{L',U^{p}}/\Xi_{U^{p}} X L ′ , p := X L ′ , U p / Ξ U p , which is canonically independent of the choice of U p U^{p} U p .
Proposition 5.2. Let L ∈ Hk ( H ( A F ∞ ) ) L \in\operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) L ∈ Hk ( H ( A F ∞ )) be as above. Consider
an intermediate number field E λ ⊆ L ⊆ C E^{\lambda} \subseteq L \subseteq\mathbb{C} E λ ⊆ L ⊆ C ,
a finite set ⋄ \diamond ⋄ of rational primes containing p p p and a nonempty subset ⧫ ⊆ B F ( ⋄ ) ∖ { v } \blacklozenge\subseteq\mathcal{B}_{F}^{(\diamond)} \setminus\{v\} ⧫ ⊆ B F ( ⋄ ) ∖ { v } ,
a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^{\infty} = (\phi_{1}^{\infty},\phi_{2}^{\infty}) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of decomposable Schwartz functions in S ( ( V ∞ ) r , L ) L \mathcal{S}((V^{\infty})^{r},L)^{L} S (( V ∞ ) r , L ) L satisfying that ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v ′ (\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}})_{v'} ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) v ′ is regularly supported (Definition 2.16) for v ′ ∈ ⧫ v' \in\blacklozenge v ′ ∈ ⧫ ; and ( ϕ i ∞ ) v ′ (\phi_{i}^{\infty})_{v'} ( ϕ i ∞ ) v ′ is the characteristic function of Λ v ′ r \Lambda_{v'}^{r} Λ v ′ r for i = 1 , 2 i=1,2 i = 1 , 2 for v ′ ∈ B F ( p ) v' \in\mathcal{B}_{F}^{(p)} v ′ ∈ B F ( p ) that is nonsplit in E E E ,
a rational prime ℓ ≠ p \ell\ne p ℓ = p , and
a pair ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) of elements in L [ L ⋄ \ H ( A F ∞ ) / L ⋄ ] L[L^{\diamond}\backslash H(\mathbb{A}_{F}^{\infty})/L^{\diamond}] L [ L ⋄ \ H ( A F ∞ ) / L ⋄ ] that annihilate both H e ˊ t 2 r ( X L , u , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\acute{e}t}^{2r}(X_{L,u},V_{\ell,L,\lambda}(r))\otimes_{E_{\lambda}}L H e ˊ t 2 r ( X L , u , V ℓ , L , λ ( r )) ⊗ E λ L and H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\acute{e}t}^{2r}(X_{L},V_{\ell,L,\lambda}(r))\otimes_{E_{\lambda}}L H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L .
(1) Suppose that V v V_{v} V v is split. Then for every pair ( T 1 , T 2 ) (T_{1},T_{2}) ( T 1 , T 2 ) of elements in Herm r ( F ) + \operatorname{Herm}_{r}(F)^{+} Herm r ( F ) + (so that Z = ( Z 1 : = Z T 1 [ s 1 ϕ 1 ∞ ] L , Z 2 : = Z T 2 [ s 2 ϕ 2 ∞ ] L ) Z = (Z_{1} := Z_{T_{1}}[s_{1}\phi_{1}^{\infty}]_{L}, Z_{2} := Z_{T_{2}}[s_{2}\phi_{2}^{\infty}]_{L}) Z = ( Z 1 := Z T 1 [ s 1 ϕ 1 ∞ ] L , Z 2 := Z T 2 [ s 2 ϕ 2 ∞ ] L ) (Definition 3.33) is a disjoint pair) and every pair ( g 1 , g 2 ) ∈ G r ( F ∞ ) 2 (g_{1},g_{2}) \in G_{r}(F_{\infty})^{2} ( g 1 , g 2 ) ∈ G r ( F ∞ ) 2 ,
log ∥ u ∥ ⋅ vol □ ( L ) ⋅ ⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ L , Z , u ℓ = E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) v ) \log\lVert u \rVert\cdot\operatorname{vol}^{\square}(L) \cdot\left\langle s_{1}^{*}Z_{T_{1}}^{\lambda}(\phi_{1}^{\infty})_{L}(g_{1}),s_{2}^{*}Z_{T_{2}}^{\lambda}(\phi_{2}^{\infty})_{L}(g_{2}) \right\rangle_{L,Z,u}^{\ell}
= \mathcal{E}_{T_{1},T_{2}}\left((g_{1},g_{2}),\Phi_{\infty}^{\lambda}\cdot(s_{1}\phi_{1}^{\infty}\otimes\overline{s_{2}\phi_{2}^{\infty}})_{v}\right) log ∥ u ∥ ⋅ vol □ ( L ) ⋅ ⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ L , Z , u ℓ = E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) v ) holds in S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ ⊗ Q Q ℓ S_{\lambda+}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}S_{\lambda+}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{Q}}\mathbb{Q}_{\ell} S λ + ( C r ) ∨ ⊗ C S λ + ( C r ) ∨ ⊗ Q Q ℓ , where Z T i λ ( ϕ i ∞ ) L Z_{T_{i}}^{\lambda}(\phi_{i}^{\infty})_{L} Z T i λ ( ϕ i ∞ ) L and E T 1 , T 2 \mathcal{E}_{T_{1},T_{2}} E T 1 , T 2 are introduced in Definition 3.33(1) and Definition 2.22(2), respectively.
(2) Suppose that V v V_v V v is nonsplit. Then for every pair ( T 1 , T 2 ) (T_1,T_2) ( T 1 , T 2 ) of elements in Herm r ∘ ( F ) + \operatorname{Herm}_{r}^{\circ}(F)^{+} Herm r ∘ ( F ) + (so that Z = ( Z 1 : = Z T 1 [ s 1 φ 1 ∞ ] L , Z 2 : = Z T 2 [ s 2 φ 2 ∞ ] L ) Z=(Z_1:=Z_{T_1}[s_1\varphi_1^{\infty}]_L,Z_2:=Z_{T_2}[s_2\varphi_2^{\infty}]_L) Z = ( Z 1 := Z T 1 [ s 1 φ 1 ∞ ] L , Z 2 := Z T 2 [ s 2 φ 2 ∞ ] L ) is a disjoint pair) and every pair ( g 1 , g 2 ) ∈ G r ( F ∞ ) 2 (g_1,g_2)\in G_r(F_{\infty})^2 ( g 1 , g 2 ) ∈ G r ( F ∞ ) 2 ,
log ∥ u ∥ ⋅ vol ♮ ( L ) ⋅ ⟨ s 1 ∗ Z T 1 λ ( φ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z T 2 λ ( φ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 φ 1 ∞ ⊗ s 2 φ 2 ∞ ) ) v − log ∥ u ∥ ∥ u ∥ r − 1 E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 φ 1 ∞ ⊗ s 2 φ 2 ∞ ) v ⋅ Φ v ⋆ ) \begin{aligned}
\log\lVert u\rVert\cdot\operatorname{vol}^{\natural}(L)\cdot\left\langle s_1^{*}Z_{T_1}^{\lambda}(\varphi_1^{\infty})_L(g_1),s_2^{*}Z_{T_2}^{\lambda}(\varphi_2^{\infty})_L(g_2)\right\rangle_{\lambda,L,Z,u}^{\ell} \\
&=\mathscr{E}_{T_1,T_2}\left((g_1,g_2),\Phi_{\infty}^{\lambda}\cdot(s_1\varphi_1^{\infty}\otimes s_2\varphi_2^{\infty})\right)_v-\frac{\log\lVert u\rVert}{\lVert u\rVert^r-1}E_{T_1,T_2}\left((g_1,g_2),\Phi_{\infty}^{\lambda}\cdot(s_1\varphi_1^{\infty}\otimes s_2\varphi_2^{\infty})^v\cdot\Phi_v^{\star}\right)
\end{aligned} log ∥ u ∥ ⋅ vol ♮ ( L ) ⋅ ⟨ s 1 ∗ Z T 1 λ ( φ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z T 2 λ ( φ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 φ 1 ∞ ⊗ s 2 φ 2 ∞ ) ) v − ∥ u ∥ r − 1 log ∥ u ∥ E T 1 , T 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 φ 1 ∞ ⊗ s 2 φ 2 ∞ ) v ⋅ Φ v ⋆ )
holds in S + λ ( C r ) ∨ ⊗ C S + λ ( C r ) ∨ ⊗ Q ℓ S_{+}^{\lambda}(\mathbb{C}^{r})^{\vee}\otimes_{\mathbb{C}}S_{+}^{\lambda}(\mathbb{C}^{r})^{\vee}\otimes\mathbb{Q}_{\ell} S + λ ( C r ) ∨ ⊗ C S + λ ( C r ) ∨ ⊗ Q ℓ , where E T 1 , T 2 E_{T_1,T_2} E T 1 , T 2 is introduced in Definition 2.22(1) and Φ v ⋆ \Phi_v^{\star} Φ v ⋆ is the characteristic function of ( Λ v ⋆ ) 2 r (\Lambda_v^{\star})^{2r} ( Λ v ⋆ ) 2 r in which Λ v ⋆ \Lambda_v^{\star} Λ v ⋆ denotes the (unique up to isomorphism) hermitian space over O E v / O F v O_{E_v}/O_{F_v} O E v / O F v of rank 2 r 2r 2 r .
Proof. We show (2) and leave the similar and less involved case (1) to readers. By Lemma 2.25 and Definition 3.33 (1), it suffices to consider the case where g i = m ( a i ) g_i=m(a_i) g i = m ( a i ) with a i ∈ GL r ( E ∞ ) a_i\in\operatorname{GL}_r(E_{\infty}) a i ∈ GL r ( E ∞ ) for i = 1 , 2 i=1,2 i = 1 , 2 . Furthermore, since for every T □ ∈ Herm 2 r ∘ ( F ) + T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F)^{+} T □ ∈ Herm 2 r ∘ ( F ) + ,
( ρ λ + ( m ( a 1 ) ) ∨ ⊗ ρ λ + ( m ( a 2 ) ) ∨ ) Φ ∞ λ ( T □ ) = Φ ∞ λ ( a t T □ a ) \left(\rho_{\lambda^{+}}(m(a_1))^{\vee}\otimes\rho_{\lambda^{+}}(m(a_2))^{\vee}\right)\Phi_{\infty}^{\lambda}(T^{\square})=\Phi_{\infty}^{\lambda}(a^{t}T^{\square}a) ( ρ λ + ( m ( a 1 ) ) ∨ ⊗ ρ λ + ( m ( a 2 ) ) ∨ ) Φ ∞ λ ( T □ ) = Φ ∞ λ ( a t T □ a )
where a = ( a 1 0 0 a 2 ) a=\begin{pmatrix}a_1&0\\0&a_2\end{pmatrix} a = ( a 1 0 0 a 2 ) , we may just assume g 1 = g 2 = 1 2 r g_1=g_2=1_{2r} g 1 = g 2 = 1 2 r .
Write Φ v : = ( φ 1 ∞ ) v ⊗ ( φ 2 ∞ ) v \Phi_v:=(\varphi_1^{\infty})_v\otimes(\varphi_2^{\infty})_v Φ v := ( φ 1 ∞ ) v ⊗ ( φ 2 ∞ ) v (which is the characteristic function of Λ v 2 r \Lambda_v^{2r} Λ v 2 r ) and Φ ∞ , v : = s 1 ( ( φ 1 ∞ ) v ) ⊗ s 2 ( ( φ 2 ∞ ) v ) \Phi_{\infty,v}:=s_1((\varphi_1^{\infty})^v)\otimes s_2((\varphi_2^{\infty})^v) Φ ∞ , v := s 1 (( φ 1 ∞ ) v ) ⊗ s 2 (( φ 2 ∞ ) v ) for short.
For every element T □ ∈ Herm 2 r ∘ ( F v ) T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F_v) T □ ∈ Herm 2 r ∘ ( F v ) whose determinant has odd valuation, put
χ ( T □ ) : = b 2 r , v ( 0 ) log ∥ u ∥ ( ∥ u ∥ 2 r − 1 ∥ u ∥ + 1 W T □ ′ ( 0 , 1 4 r , Φ v ) − log ∥ u ∥ ∥ u ∥ + 1 W T □ ( 0 , 1 4 r , Φ v ⋆ ) ) , \chi(T^{\square}):=\frac{b_{2r,v}(0)}{\log\lVert u\rVert}\left(\frac{\lVert u\rVert^{2r}-1}{\lVert u\rVert+1}W_{T^{\square}}^{\prime}(0,1_{4r},\Phi_v)-\frac{\log\lVert u\rVert}{\lVert u\rVert+1}W_{T^{\square}}(0,1_{4r},\Phi_v^{\star})\right), χ ( T □ ) := log ∥ u ∥ b 2 r , v ( 0 ) ( ∥ u ∥ + 1 ∥ u ∥ 2 r − 1 W T □ ′ ( 0 , 1 4 r , Φ v ) − ∥ u ∥ + 1 log ∥ u ∥ W T □ ( 0 , 1 4 r , Φ v ⋆ ) ) ,
which is indeed an integer. By the definition of vol ♮ ( L ) \operatorname{vol}^{\natural}(L) vol ♮ ( L ) in [20 Definition 3.8] , (2) reduces to the identity
⟨ s 1 ∗ Z T 1 λ ( φ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( φ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = ∑ T □ ∈ Herm 2 r ∘ ( F ) + ∂ T □ = ( T 1 , T 2 ) κ Ψ ∞ λ ( T □ ) ⋅ χ ( T □ ) ⋅ ( ∑ x ∈ V T □ ∞ , v Φ ∞ , v ( x ) ) , (54) \left\langle s_1^{*}Z_{T_1}^{\lambda}(\varphi_1^{\infty})_L,s_2^{*}Z_{T_2}^{\lambda}(\varphi_2^{\infty})_L\right\rangle_{\lambda,L,Z,u}^{\ell}
=
\sum_{\substack{T^{\square}\in\operatorname{Herm}_{2r}^{\circ}(F)^{+}\\ \partial T^{\square}=(T_1,T_2)}}\kappa\Psi_{\infty}^{\lambda}(T^{\square})\cdot\chi(T^{\square})\cdot\left(\sum_{x\in V^{\infty,v}_{T^{\square}}}\Phi^{\infty,v}(x)\right), \tag*{(54)} ⟨ s 1 ∗ Z T 1 λ ( φ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( φ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = T □ ∈ Herm 2 r ∘ ( F ) + ∂ T □ = ( T 1 , T 2 ) ∑ κ Ψ ∞ λ ( T □ ) ⋅ χ ( T □ ) ⋅ x ∈ V T □ ∞ , v ∑ Φ ∞ , v ( x ) , (54)
where we recall that Ψ ∞ λ \Psi_{\infty}^{\lambda} Ψ ∞ λ is from Construction 2.8(4), which factors through the moment map. Note that on the right-hand side of (54) , if T □ T^{\square} T □ has even valuation at v v v , then ( V ∞ , v ) T □ 2 r (V^{\infty,v})_{T^{\square}}^{2r} ( V ∞ , v ) T □ 2 r is empty.
For every integer 1 ≤ s ≤ 2 r 1\le s\le2r 1 ≤ s ≤ 2 r and every element x v ∈ ( V ∞ , v ) s x^v\in(V^{\infty,v})^s x v ∈ ( V ∞ , v ) s with T = T ( x v ) ∈ Herm s ∘ ( F ) + T=T(x^v)\in\operatorname{Herm}_{s}^{\circ}(F)^{+} T = T ( x v ) ∈ Herm s ∘ ( F ) + , we have a natural finite unramified morphism ς T ( x v ) L : Z T ( x v ) L → X L \varsigma_{T}(x^v)_L:Z_{T}(x^v)_L\to X_L ς T ( x v ) L : Z T ( x v ) L → X L whose generic fiber is
∐ x = x v x v ∈ ( V ∞ ) s x v ∈ L v \ Λ v Z ( x ) L , u ⟶ X L , u \coprod_{\substack{x=x_vx^v\in(V^{\infty})^s\\ x_v\in L_v\backslash\Lambda_v}}Z(x)_L,u\longrightarrow X_{L,u} x = x v x v ∈ ( V ∞ ) s x v ∈ L v \ Λ v ∐ Z ( x ) L , u ⟶ X L , u
via moduli interpretation of X L , U X_{L,U} X L , U (after choosing U U U as in Remark 5.1), and denote by Z T [ x v ] L Z_T[x^v]_L Z T [ x v ] L its image as a Zariski closed subset of X L X_L X L . It also carries a map
μ T ( x v ) L : ς T ( x v ) L ∗ V L , ℓ ∨ ⟶ ( 1 Z T ( x v ) L , E ℓ ) ⊕ s (55) \mu_T(x^v)_L:\varsigma_T(x^v)_L^{*}V_{L,\ell}^{\vee}\longrightarrow\left(\mathbb{1}_{Z_T(x^v)_L,E_\ell}\right)^{\oplus s} \tag*{(55)} μ T ( x v ) L : ς T ( x v ) L ∗ V L , ℓ ∨ ⟶ ( 1 Z T ( x v ) L , E ℓ ) ⊕ s (55)
similar to (14) .
Now take x v = ( x 1 v , … , x r v ) ∈ ( V ∞ , v ) r x^v=(x_1^v,\ldots,x_r^v)\in(V^{\infty,v})^r x v = ( x 1 v , … , x r v ) ∈ ( V ∞ , v ) r with T = T ( x v ) ∈ Herm r ∘ ( F ) + T=T(x^v)\in\operatorname{Herm}_{r}^{\circ}(F)^{+} T = T ( x v ) ∈ Herm r ∘ ( F ) + and t 1 , … , t r t_1,\ldots,t_r t 1 , … , t r diagonal entries of T T T . For 1 ≤ i ≤ r 1\le i\le r 1 ≤ i ≤ r , write C t i ( x i v ) L ∈ H e ˊ t , Z t i [ x i v ] L 2 ( X L , Q ℓ ( 1 ) ) C_{t_i}(x_i^v)_L\in H_{\mathrm{\acute et},Z_{t_i}[x_i^v]_L}^{2}(X_L,\mathbb{Q}_{\ell}(1)) C t i ( x i v ) L ∈ H e ˊ t , Z t i [ x i v ] L 2 ( X L , Q ℓ ( 1 )) for the refined absolute ℓ \ell ℓ -adic cycle class of Z t i ( x i v ) L Z_{t_i}(x_i^v)_L Z t i ( x i v ) L , so that we have
C t 1 ( x 1 v ) L ∪ ⋯ ∪ C t r ( x r v ) L ∈ ⨁ T ′ ∈ Herm r ∘ ( F ) + T ′ ∼ T H e ˊ t , Z T ′ [ x v ] L 2 r ( X L , Q ℓ ( r ) ) . C_{t_1}(x_1^v)_L\cup\cdots\cup C_{t_r}(x_r^v)_L\in\bigoplus_{\substack{T'\in\operatorname{Herm}_{r}^{\circ}(F)^{+}\\T'\sim T}}H_{\mathrm{\acute et},Z_{T'[x^v]_L}^{2r}(X_L,\mathbb{Q}_{\ell}(r)).} C t 1 ( x 1 v ) L ∪ ⋯ ∪ C t r ( x r v ) L ∈ T ′ ∈ Herm r ∘ ( F ) + T ′ ∼ T ⨁ H e ˊ t , Z T ′ [ x v ] L 2 r ( X L , Q ℓ ( r )) .
where T ′ ≃ T T' \simeq T T ′ ≃ T means that they have the same diagonal entries. Define C T ( x v ) L C_T(x^v)_L C T ( x v ) L to be its component in H e ˊ t , Z T [ x v ] L 2 r ( X L , Q ℓ ( r ) ) H^{2r}_{\mathrm{\acute et},\mathcal{Z}_T[x^v]_L}(\mathcal{X}_L,\mathbb{Q}_{\ell}(r)) H e ˊ t , Z T [ x v ] L 2 r ( X L , Q ℓ ( r )) . Similar to (36) , the map (5.2) induces maps
μ T λ ( x v ) L : S λ + ( ( E λ ) r ) ⊗ S T ( x v ) L ! ⊗ Q ℓ e ˊ t ⟶ S T ( x v ) L ! V ℓ , L , λ , \mu_T^\lambda(x^v)_L:S_{\lambda_+}((E^\lambda)^r)\otimes S_T(x^v)_L{}^!\mathop{\otimes}_{\mathbb{Q}_\ell} \!{}_{\mathrm{\acute et}} \longrightarrow S_T(x^v)_L^!\,V_{\ell,L,\lambda}, μ T λ ( x v ) L : S λ + (( E λ ) r ) ⊗ S T ( x v ) L ! ⊗ Q ℓ e ˊ t ⟶ S T ( x v ) L ! V ℓ , L , λ ,
hence
S λ + ( ( E λ ) r ) ⊗ H e ˊ t , Z T [ x v ] L 2 r ( X L , Q ℓ ( r ) ) ⟶ H e ˊ t , Z T [ x v ] L 2 r ( X L , V ℓ , L , λ ( r ) ) . (56) S_{\lambda_+}((E^\lambda)^r)\otimes H^{2r}_{\mathrm{\acute et},\mathcal{Z}_T[x^v]_L}(\mathcal{X}_L,\mathbb{Q}_\ell(r))\longrightarrow H^{2r}_{\mathrm{\acute et},\mathcal{Z}_T[x^v]_L}(\mathcal{X}_L,V_{\ell,L,\lambda}(r)). \tag*{(56)} S λ + (( E λ ) r ) ⊗ H e ˊ t , Z T [ x v ] L 2 r ( X L , Q ℓ ( r )) ⟶ H e ˊ t , Z T [ x v ] L 2 r ( X L , V ℓ , L , λ ( r )) . (56)
Denote by
C T λ ( x v ) L : S λ + ( ( E λ ) r ) ⟶ H e ˊ t , Z T [ x v ] L 2 r ( X L , V ℓ , L , λ ( r ) ) C_T^\lambda(x^v)_L:S_{\lambda_+}((E^\lambda)^r)\longrightarrow H^{2r}_{\mathrm{\acute et},\mathcal{Z}_T[x^v]_L}(\mathcal{X}_L,V_{\ell,L,\lambda}(r)) C T λ ( x v ) L : S λ + (( E λ ) r ) ⟶ H e ˊ t , Z T [ x v ] L 2 r ( X L , V ℓ , L , λ ( r ))
the image of id ⊗ C T ( x v ) L \operatorname{id}\otimes C_T(x^v)_L id ⊗ C T ( x v ) L under (56) .
For i = 1 , 2 i=1,2 i = 1 , 2 , put
Z T i [ s i ϕ i ∞ ] L : = ∐ x i v ∈ ( V ∞ , v ) T i ( s i ϕ i ∞ , v , x i v ) ≠ 0 Z T i [ x i v ] L \mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L:=\coprod_{\substack{x_i^v\in(V^{\infty,v})_{T_i}\\(s_i\phi_i^{\infty,v},x_i^v)\ne0}}\mathcal{Z}_{T_i}[x_i^v]_L Z T i [ s i ϕ i ∞ ] L := x i v ∈ ( V ∞ , v ) T i ( s i ϕ i ∞ , v , x i v ) = 0 ∐ Z T i [ x i v ] L
whose generic fiber coincides with Z i , u Z_{i,u} Z i , u , and put
C T i λ ( s i ϕ i ∞ ) L : = ∑ x i v ∈ L λ \ ( V ∞ , v ) T i ( s i ϕ i ∞ , v ) ( x i v ) ⋅ C T i λ ( x i v ) L ∈ S λ + ( L ∨ ) ⊗ E λ H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( X L , V ℓ , L , λ ( r ) ) C_{T_i}^\lambda(s_i\phi_i^\infty)_L:=\sum_{x_i^v\in L\lambda\backslash(V^{\infty,v})_{T_i}}(s_i\phi_i^{\infty,v})(x_i^v)\cdot C_{T_i}^\lambda(x_i^v)_L\in S_{\lambda_+}(L^\vee)\otimes_{E_\lambda}H^{2r}_{\mathrm{\acute et},\mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L}(\mathcal{X}_L,V_{\ell,L,\lambda}(r)) C T i λ ( s i ϕ i ∞ ) L := x i v ∈ L λ \ ( V ∞ , v ) T i ∑ ( s i ϕ i ∞ , v ) ( x i v ) ⋅ C T i λ ( x i v ) L ∈ S λ + ( L ∨ ) ⊗ E λ H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( X L , V ℓ , L , λ ( r ))
whose natural image in H e ˊ t , Z i , u 2 r ( X L , u , V ℓ , L , λ ( r ) ) H^{2r}_{\mathrm{\acute et},Z_{i,u}}(\mathcal{X}_{L,u},V_{\ell,L,\lambda}(r)) H e ˊ t , Z i , u 2 r ( X L , u , V ℓ , L , λ ( r )) coincides with that of s i ∗ Z T i λ ( ϕ i ∞ ) L s_i^*Z_{T_i}^\lambda(\phi_i^\infty)_L s i ∗ Z T i λ ( ϕ i ∞ ) L .
Since s 1 s_1 s 1 annihilates H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H^{2r}_{\mathrm{\acute et}}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\otimes_{E_\lambda}L H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L , by Lemma 4.7, we have
⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u = C T 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C T 2 λ ( s 2 ϕ 2 ∞ ) L . \left\langle s_1^*Z_{T_1}^\lambda(\phi_1^\infty)_L,s_2^*Z_{T_2}^\lambda(\phi_2^\infty)_L\right\rangle_{\lambda,L,Z,u}
=C_{T_1}^\lambda(s_1\phi_1^\infty)_L\bullet C_{T_2}^\lambda(s_2\phi_2^\infty)_L. ⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u = C T 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C T 2 λ ( s 2 ϕ 2 ∞ ) L .
Note that the intersection of Z T 1 [ s 1 ϕ 1 ∞ ] L \mathcal{Z}_{T_1}[s_1\phi_1^\infty]_L Z T 1 [ s 1 ϕ 1 ∞ ] L and Z T 2 [ s 2 ϕ 2 ∞ ] L \mathcal{Z}_{T_2}[s_2\phi_2^\infty]_L Z T 2 [ s 2 ϕ 2 ∞ ] L is contained in the basic locus of X L \mathcal{X}_L X L .
Denote by V ⋆ V^\star V ⋆ the unique totally positive definite hermitian space over E / F E/F E / F such that V ⋆ ⊗ F A F ∞ , v V^\star\otimes_F A_F^{\infty,v} V ⋆ ⊗ F A F ∞ , v isometric to V ⊗ A F A F ∞ , v V\otimes_{A_F}A_F^{\infty,v} V ⊗ A F A F ∞ , v . Put H ⋆ : = U ( V ⋆ ) H^\star:=U(V^\star) H ⋆ := U ( V ⋆ ) , and fix an isometry V ⋆ ⊗ F A F ∞ , v ≃ V ⊗ A F A F ∞ , v V^\star\otimes_F A_F^{\infty,v}\simeq V\otimes_{A_F}A_F^{\infty,v} V ⋆ ⊗ F A F ∞ , v ≃ V ⊗ A F A F ∞ , v under which we identify H ( A F ∞ , v ) H(A_F^{\infty,v}) H ( A F ∞ , v ) with H ⋆ ( A F ∞ , v ) H^\star(A_F^{\infty,v}) H ⋆ ( A F ∞ , v ) . For every L ′ p ∈ H k ( H ( A F ∞ , p ) ) L'\mathfrak{p}\in H^k(H(A_F^{\infty,p})) L ′ p ∈ H k ( H ( A F ∞ , p )) , denote by X L ′ \mathcal{X}_{L'} X L ′ the formal completion of X L ′ ⊗ O E u O E ˘ u \mathcal{X}_{L'}\otimes_{O_{E_u}}O_{\breve{E}_u} X L ′ ⊗ O E u O E ˘ u along its basic locus, where E ˘ u \breve{E}_u E ˘ u denotes a complete maximal unramified extension of E u E_u E u . Then we have the uniformization isomorphism
X L ′ ≃ H ⋆ ( F ) \ ( N × H ( A F ∞ , v ) / L ′ v ) \mathcal{X}_{L'}\simeq H^\star(F)\backslash\left(\mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v}\right) X L ′ ≃ H ⋆ ( F ) \ ( N × H ( A F ∞ , v ) / L ′ v )
functorial in L ′ L' L ′ , where N \mathcal{N} N is a certain formal Rapoport–Zink space over Spf O E ˘ u \operatorname{Spf}O_{\breve{E}_u} Spf O E ˘ u (see [22 §13.1] for more details). The E ℓ λ E_\ell^\lambda E ℓ λ -linear local system V ℓ , L ′ , λ V_{\ell,L',\lambda} V ℓ , L ′ , λ over X L ′ \mathcal{X}_{L'} X L ′ has a compatible uniformization:
V ℓ , L ′ , λ ∣ X L ′ = H ⋆ ( F ) \ ( N × H ( A F ∞ , v ) / L ′ v × V ℓ , λ ) , V_{\ell,L',\lambda}|_{\mathcal{X}_{L'}}=H^\star(F)\backslash\left(\mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v}\times V_{\ell,\lambda}\right), V ℓ , L ′ , λ ∣ X L ′ = H ⋆ ( F ) \ ( N × H ( A F ∞ , v ) / L ′ v × V ℓ , λ ) ,
in which H ⋆ ( F ) H^\star(F) H ⋆ ( F ) acts on V ℓ , λ : = ( V ℓ ) λ V_{\ell,\lambda}:=(V_\ell)_\lambda V ℓ , λ := ( V ℓ ) λ via the natural map H ⋆ ( F ) → H ( F ℓ ) H^\star(F)\to H(F_\ell) H ⋆ ( F ) → H ( F ℓ ) .
For i = 1 , 2 i=1,2 i = 1 , 2 , if we denote by Z T i [ s i ϕ i ∞ ] L \mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L Z T i [ s i ϕ i ∞ ] L the inverse image of Z T i [ s i ϕ i ∞ ] L \mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L Z T i [ s i ϕ i ∞ ] L in N × H ( A F ∞ , v ) / L ′ v \mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v} N × H ( A F ∞ , v ) / L ′ v , then C T i λ ( s i ϕ i ∞ ) L ∣ X L ′ C_{T_i}^\lambda(s_i\phi_i^\infty)_L|_{\mathcal{X}_{L'}} C T i λ ( s i ϕ i ∞ ) L ∣ X L ′ is represented by the H ⋆ ( F ) H^\star(F) H ⋆ ( F ) -invariant class
∑ x i ⋆ ∈ ( V ⋆ ) T i ( ∑ h ∈ H ( A F ∞ , v ) / L ′ v s i ϕ i ∞ , v ( h − 1 x i ⋆ ) ⋅ C ( x i , v ) h ) ⊗ μ λ ( x i ⋆ ) \sum_{x_i^\star\in(V^\star)_{T_i}}\left(\sum_{h\in H(A_F^{\infty,v})/L'^{\,v}}s_i\phi_i^{\infty,v}(h^{-1}x_i^\star)\cdot\mathcal{C}(x_{i,v})_h\right)\otimes\mu^\lambda(x_i^\star) x i ⋆ ∈ ( V ⋆ ) T i ∑ h ∈ H ( A F ∞ , v ) / L ′ v ∑ s i ϕ i ∞ , v ( h − 1 x i ⋆ ) ⋅ C ( x i , v ) h ⊗ μ λ ( x i ⋆ )
∈ H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r ) ) ⊗ ( S λ + ( ( E λ ) r ) ⊗ E λ V λ ) = S λ + ( ( E λ ) r ) ⊗ E λ ( H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r ) ) ⊗ Q ℓ V ℓ , λ ) , \begin{aligned}
\in H^{2r}_{\mathrm{\acute et},\mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L}\left(\mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v},\mathbb{Q}_\ell(r)\right)\otimes\left(S_{\lambda_+}((E^\lambda)^r)\otimes_{E_\lambda}V_\lambda\right) \\
&=S_{\lambda_+}((E^\lambda)^r)\otimes_{E_\lambda}\left(H^{2r}_{\mathrm{\acute et},\mathcal{Z}_{T_i}[s_i\phi_i^\infty]_L}\left(\mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v},\mathbb{Q}_\ell(r)\right)\otimes_{\mathbb{Q}_\ell}V_{\ell,\lambda}\right),
\end{aligned} ∈ H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r ) ) ⊗ ( S λ + (( E λ ) r ) ⊗ E λ V λ ) = S λ + (( E λ ) r ) ⊗ E λ ( H e ˊ t , Z T i [ s i ϕ i ∞ ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r ) ) ⊗ Q ℓ V ℓ , λ ) ,
where C ( x i , v ) h ∈ H e ˊ t , Z T i [ x i v ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r ) ) \mathcal{C}(x_{i,v})_h\in H^{2r}_{\mathrm{\acute et},\mathcal{Z}_{T_i}[x_i^v]_L}(\mathcal{N}\times H(A_F^{\infty,v})/L'^{\,v},\mathbb{Q}_\ell(r)) C ( x i , v ) h ∈ H e ˊ t , Z T i [ x i v ] L 2 r ( N × H ( A F ∞ , v ) / L ′ v , Q ℓ ( r )) denotes the ℓ \ell ℓ -adic cycle class of the derived special cycle of x i , v x_{i,v} x i , v on N \mathcal{N} N defined in [22 Definition 2.4.1] but put in the component indexed by h h h ; and $\mu^\lambda is defined in Construction 2.8(3) (with R = Q R = \mathbb{Q} R = Q ). Combining this with [22 Theorem 10.5.1 & Remark 10.5.4] and using the argument for [20 Proposition 8.1] , we obtain
C T 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C T 2 λ ( s 2 ϕ 2 ∞ ) L = ∑ T ′ ∈ Herm 2 r ∘ ( F ) + ∂ T ′ = ( T 1 , T 2 ) κ Ψ ∞ λ ( T ′ ) ⋅ χ ( T ′ ) ⋅ ( ∑ x ∈ V ′ \ ( V ∞ , ∘ ) T ′ 2 r Φ ∞ , ∘ ( x ) ) . C_{T_1}^{\lambda}(s_1\phi_1^\infty)_L \mathbin{\bullet} C_{T_2}^{\lambda}(s_2\phi_2^\infty)_L = \sum_{\substack{T^\prime\in\operatorname{Herm}^{\circ}_{2r}(F)^+\\ \partial T^\prime=(T_1,T_2)}} \kappa\Psi^\lambda_\infty(T^\prime)\cdot\chi(T^\prime)\cdot\left(\sum_{x\in V^\prime\backslash(V^{\infty,\circ})^{2r}_{T^\prime}}\Phi^{\infty,\circ}(x)\right). C T 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C T 2 λ ( s 2 ϕ 2 ∞ ) L = T ′ ∈ Herm 2 r ∘ ( F ) + ∂ T ′ = ( T 1 , T 2 ) ∑ κ Ψ ∞ λ ( T ′ ) ⋅ χ ( T ′ ) ⋅ x ∈ V ′ \ ( V ∞ , ∘ ) T ′ 2 r ∑ Φ ∞ , ∘ ( x ) .
The proposition is proved. □
Remark 5.3. By the similar argument for [20 Lemma 8.2] , Proposition 5.2 holds as well if we replace ( ϕ 1 ∞ , ϕ 2 ∞ ) (\phi_1^\infty,\phi_2^\infty) ( ϕ 1 ∞ , ϕ 2 ∞ ) by ( ω r ∞ ( g 1 ) ϕ 1 ∞ , ω r ∞ ( g 2 ) ϕ 2 ∞ ) (\omega_r^\infty(g_1)\phi_1^\infty,\omega_r^\infty(g_2)\phi_2^\infty) ( ω r ∞ ( g 1 ) ϕ 1 ∞ , ω r ∞ ( g 2 ) ϕ 2 ∞ ) for every pair ( g 1 , g 2 ) ∈ G r ( A F ∞ , ∘ ) 2 (g_1,g_2)\in G_r(\mathbb{A}_F^{\infty,\circ})^2 ( g 1 , g 2 ) ∈ G r ( A F ∞ , ∘ ) 2 .
Remark 5.4. Let ( π , A π ) (\pi,A_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ (Definition 2.27). Suppose that π ∞ \pi^\infty π ∞ is defined over L \mathbb{L} L and that ( π v ) K r , v ≠ { 0 } (\pi_v)^{K_{r,v}}\ne\{0\} ( π v ) K r , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ∘ v\in\mathcal{V}_F^{\mathrm{fin}}\setminus\mathcal{V}_F^{\circ} v ∈ V F fin ∖ V F ∘ .
Suppose that V v V_v V v is split. By Lemma 4.9 and the proper base change, there exist elements in L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] ∖ ker χ π \mathbb{L}[\mathbb{L}^{\circ}\backslash H(\mathbb{A}_F^{\infty,\circ})/\mathbb{L}^{\circ}]\setminus\ker\chi_\pi L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] ∖ ker χ π (4.5) that annihilate H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2r}(X_{\mathbb{L}},V_{\ell,\mathbb{L},\lambda}(r))\otimes_{E_\lambda}\mathbb{L} H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L and support only at places split in E E E .
Suppose that V v V_v V v is nonsplit. By the similar argument for [20 Lemma 9.2] , there exist elements in L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] ∖ ker χ π \mathbb{L}[\mathbb{L}^{\circ}\backslash H(\mathbb{A}_F^{\infty,\circ})/\mathbb{L}^{\circ}]\setminus\ker\chi_\pi L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] ∖ ker χ π that annihilate H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2r}(X_{\mathbb{L}},V_{\ell,\mathbb{L},\lambda}(r))\otimes_{E_\lambda}\mathbb{L} H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L and support only at places split in E E E .
Vanishing local heights at split places
In this section, we take a nonarchimedean place u u u of E E E such that v : = v ( u ) v:=v(u) v := v ( u ) is split in E E E .
Consider an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) such that
L v = ker ( U ( Λ v ) → U ( Λ v / ν ∘ ) ) L_v=\ker\bigl(U(\Lambda_v)\to U(\Lambda_v/\nu^\circ)\bigr) L v = ker ( U ( Λ v ) → U ( Λ v / ν ∘ ) ) for some self-dual lattice Λ v \Lambda_v Λ v of V v V_v V v and some integer δ ≥ 0 \delta\ge0 δ ≥ 0 .
when r > 1 r>1 r > 1 , L v ′ L_{v^\prime} L v ′ is the stabilizer of a vertex lattice (Definition 2.9) for every place v ′ ∈ V F ( p ) v^\prime\in\mathcal{V}_F^{(p)} v ′ ∈ V F ( p ) that is nonsplit in E E E , where p : = p ( u ) p:=p(u) p := p ( u ) .
Let
X ∙ : Hk ( H ( A F ∞ , ∘ ) ) ⟶ Sch f l . p r o j / O E u \mathcal{X}_\bullet:\operatorname{Hk}(H(\mathbb{A}_F^{\infty,\circ}))\longrightarrow\operatorname{Sch}^{\mathrm{fl.proj}}/\mathcal{O}_{E_u} X ∙ : Hk ( H ( A F ∞ , ∘ )) ⟶ Sch fl.proj / O E u
be the functor constructed in [22 §11.5] ,33 satisfying Assumption 4.6.
Proposition 5.5. Let L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) be as above. Consider
an intermediate number field E λ ⊆ L ⊆ C E^\lambda\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C ,
a finite set ⋄ \diamond ⋄ of rational primes containing p p p and a nonempty subset ⧫ ⊆ V F ( ⋄ ) ∖ { v } \blacklozenge\subseteq\mathcal{V}_F^{(\diamond)}\setminus\{v\} ⧫ ⊆ V F ( ⋄ ) ∖ { v } ,
a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^\infty=(\phi_1^\infty,\phi_2^\infty) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of decomposable Schwartz functions in S ( ( V ∞ ) ∘ , L ) L \mathcal{S}((V^\infty)^{\circ},L)^L S (( V ∞ ) ∘ , L ) L satisfying that ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v ′ (\phi_1^\infty\otimes\overline{\phi_2^\infty})_{v^\prime} ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) v ′ is regularly supported (Definition 2.16) for v ′ ∈ ⧫ v^\prime\in\blacklozenge v ′ ∈ ⧫ ,
a rational prime ℓ ≠ p \ell\ne p ℓ = p , and
a pair ( s 1 , s 2 ) (s_1,s_2) ( s 1 , s 2 ) of elements in L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] \mathbb{L}[\mathbb{L}^{\circ}\backslash H(\mathbb{A}_F^{\infty,\circ})/\mathbb{L}^{\circ}] L [ L ∘ \ H ( A F ∞ , ∘ ) / L ∘ ] that annihilate both H e ˊ t 2 r ( X L , u , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2r}(X_{\mathbb{L},u},V_{\ell,\mathbb{L},\lambda}(r))\otimes_{E_\lambda}L H e ˊ t 2 r ( X L , u , V ℓ , L , λ ( r )) ⊗ E λ L and H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2r}(X_{\mathbb{L}},V_{\ell,\mathbb{L},\lambda}(r))\otimes_{E_\lambda}L H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L .
Then for every pair ( T 1 , T 2 ) (T_1,T_2) ( T 1 , T 2 ) of elements in Herm r ∘ ( F ) + \operatorname{Herm}^{\circ}_{r}(F)^+ Herm r ∘ ( F ) + (so that Z = ( Z 1 : = Z T 1 [ s 1 ϕ 1 ∞ ] L , Z 2 : = Z T 2 [ s 2 ϕ 2 ∞ ] L ) Z=(Z_1:=Z_{T_1}[s_1\phi_1^\infty]_L,Z_2:=Z_{T_2}[s_2\phi_2^\infty]_L) Z = ( Z 1 := Z T 1 [ s 1 ϕ 1 ∞ ] L , Z 2 := Z T 2 [ s 2 ϕ 2 ∞ ] L ) (Definition 3.33) is a disjoint pair),
⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = 0 \left\langle s_1^*Z_{T_1}^{\lambda}(\phi_1^\infty)_L,s_2^*Z_{T_2}^{\lambda}(\phi_2^\infty)_L\right\rangle^\ell_{\lambda,L,Z,u}=0 ⟨ s 1 ∗ Z T 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z T 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = 0 holds.
Proof. Denote by Z i \mathcal{Z}_i Z i the Zariski closure of Z i , u Z_{i,u} Z i , u in X L \mathcal{X}_L X L for i = 1 , 2 i=1,2 i = 1 , 2 . Then s 1 Z 1 s_1\mathcal{Z}_1 s 1 Z 1 and s 2 Z 2 s_2\mathcal{Z}_2 s 2 Z 2 have empty intersection by (the proof of) [20 Lemma 7.2] . For i = 1 , 2 i=1,2 i = 1 , 2 , we may find a class
C i ∈ H e ˊ t , Z i 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L C_i \in H_{\mathrm{\acute{e}t},\mathcal{Z}_i}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\otimes_{E_\lambda}L C i ∈ H e ˊ t , Z i 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L
whose image in H e ˊ t , Z i , u 2 r ( X L , u , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute{e}t},Z_{i,u}}^{2r}(\mathcal{X}_{L,u},V_{\ell,L,\lambda}(r))\otimes_{E_\lambda}L H e ˊ t , Z i , u 2 r ( X L , u , V ℓ , L , λ ( r )) ⊗ E λ L coincides with that of Z T i λ ( ϕ i ∞ ) L Z_{T_i}^{\lambda}(\phi_i^\infty)_L Z T i λ ( ϕ i ∞ ) L . Since the image of s 1 ∗ C 1 s_1^*C_1 s 1 ∗ C 1 in H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute{e}t}}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\otimes_{E_\lambda}L H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L vanishes, the proposition follows from (the L L L -sesquilinear extension of) Lemma 4.7. □ \square □
Remark 5.6. Let ( π , A π ) (\pi,A_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G r ( A F ) G_r(\mathbb{A}_F) G r ( A F ) of weight λ \lambda λ (Definition 2.27). Suppose that π ∞ \pi^\infty π ∞ is defined over L L L and ( π v ) K r , v ≠ { 0 } (\pi_v)^{K_{r,v}}\ne\{0\} ( π v ) K r , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ∞ v\in\mathfrak{V}_F^{\mathrm{fin}}\setminus\mathfrak{V}_F^{\infty} v ∈ V F fin ∖ V F ∞ . By the similar argument for [21 Theorem 4.21] , there exist elements in L [ L ∞ \ H ( A F ∞ ) / L × ] ∖ ker χ π \mathbb{L}[\mathbb{L}^\infty\backslash H(\mathbb{A}_F^\infty)/\mathbb{L}^\times]\setminus\ker\chi_\pi L [ L ∞ \ H ( A F ∞ ) / L × ] ∖ ker χ π (4.5) that annihilate H e ˊ t 2 r ( X L , V ℓ , L , λ ( r ) ) ⊗ E λ L H_{\mathrm{\acute{e}t}}^{2r}(\mathcal{X}_L,V_{\ell,L,\lambda}(r))\otimes_{E_\lambda}L H e ˊ t 2 r ( X L , V ℓ , L , λ ( r )) ⊗ E λ L .
Nonarchimedean local heights over unitary moduli curves
In this section, we assume r = 1 r=1 r = 1 and take a nonarchimedean place u u u of E E E such that v : = v ( u ) v:=v(u) v := v ( u ) is nonsplit in E E E . Put p : = p ( u ) p:=p(u) p := p ( u ) .
For every intermediate field E λ ⊆ L ⊆ C E^\lambda\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C , we canonically identify both S λ + ( L ) S_{\lambda^+}(L) S λ + ( L ) and S λ + ( L ) S_{\lambda^+}(L) S λ + ( L ) with L L L as L L L -vector spaces.
Definition 5.7. For every v ∈ V λ ⊗ E λ C v\in V_\lambda\otimes_{E^\lambda}\mathbb{C} v ∈ V λ ⊗ E λ C , we have a polynomial function P [ v ] P[v] P [ v ] on V ∞ V_\infty V ∞ given by the formula
P [ v ] ( x ) = ( μ λ ( x ) , v ) V λ ∈ C , x ∈ V ∞ , P[v](x)=(\mu^\lambda(x),v)_{V_\lambda}\in\mathbb{C},\qquad x\in V_\infty, P [ v ] ( x ) = ( μ λ ( x ) , v ) V λ ∈ C , x ∈ V ∞ ,
where μ λ \mu^\lambda μ λ is defined in Construction 2.8(3) (with R = R R=\mathbb{R} R = R ).
Definition 5.8. For a field L L L containing E λ E^\lambda E λ , a finite subset ⋄ \diamond ⋄ of Q F f i n \mathfrak{Q}_F^{\mathrm{fin}} Q F fin , and an object L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ∈ Hk ( H ( A F ∞ )) , we define the special Eisenstein ideal of L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] \mathbb{L}[\mathbb{L}^\diamond\backslash H(\mathbb{A}_F^{\infty,\diamond})/\mathbb{L}^\diamond] L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] with respect to the weight λ \lambda λ , denoted by L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ \mathbb{L}[\mathbb{L}^\diamond\backslash H(\mathbb{A}_F^{\infty,\diamond})/\mathbb{L}^\diamond]^\lambda L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ , to be
when λ = 0 \lambda=0 λ = 0 , the (two-sided) ideal consisting of elements that are of degree zero and supported on double cosets with unit determinants;
when λ ≠ 0 \lambda\ne0 λ = 0 , the entire ring L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] \mathbb{L}[\mathbb{L}^\diamond\backslash H(\mathbb{A}_F^{\infty,\diamond})/\mathbb{L}^\diamond] L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] .
In both cases, elements in L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ \mathbb{L}[\mathbb{L}^\diamond\backslash H(\mathbb{A}_F^{\infty,\diamond})/\mathbb{L}^\diamond]^\lambda L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ annihilate H e ˊ t 2 ( X L ⊗ E E a c , V ℓ , L , λ ( 1 ) ) ⊗ E λ L H_{\mathrm{\acute{e}t}}^2(X_L\otimes_EE^{\mathrm{ac}},V_{\ell,L,\lambda}(1))\otimes_{E^\lambda}L H e ˊ t 2 ( X L ⊗ E E ac , V ℓ , L , λ ( 1 )) ⊗ E λ L .
Denote by V ( v ) V(v) V ( v ) the unique totally positive definite hermitian space over E / F E/F E / F such that V ( v ) ⊗ F A F ∞ , v V(v)\otimes_F\mathbb{A}_F^{\infty,v} V ( v ) ⊗ F A F ∞ , v is isometric to V ⊗ A F A F ∞ , v V\otimes_{A_F}\mathbb{A}_F^{\infty,v} V ⊗ A F A F ∞ , v . Put H ( v ) : = U ( V ( v ) ) H(v):=U(V(v)) H ( v ) := U ( V ( v )) , and fix an isometry V ( v ) ⊗ F A F ∞ , v ≃ V ⊗ A F A F ∞ , v V(v)\otimes_F\mathbb{A}_F^{\infty,v}\simeq V\otimes_{A_F}\mathbb{A}_F^{\infty,v} V ( v ) ⊗ F A F ∞ , v ≃ V ⊗ A F A F ∞ , v under which we identify H ( A F ∞ , v ) H(\mathbb{A}_F^{\infty,v}) H ( A F ∞ , v ) with H ( v ) ( A F ∞ , v ) H(v)(\mathbb{A}_F^{\infty,v}) H ( v ) ( A F ∞ , v ) .
Proposition 5.9. Consider
an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_F^\infty)) Hk ( H ( A F ∞ )) ,
an intermediate number field E λ ⊆ L ⊆ C E^\lambda\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C ,
a finite set ⋄ \diamond ⋄ of rational primes containing p p p and a nonempty subset ⋄ ⊆ Q ⟨ ⋄ ⟩ ∖ { v } \diamond\subseteq\mathfrak{Q}^{\langle\diamond\rangle}\setminus\{v\} ⋄ ⊆ Q ⟨ ⋄ ⟩ ∖ { v } ,
a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^\infty=(\phi_1^\infty,\phi_2^\infty) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of decomposable Schwartz functions in S ( V ∞ , L ) L \mathcal{S}(V^\infty,\mathbb{L})^L S ( V ∞ , L ) L satisfying that ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v (\phi_1^\infty\otimes\overline{\phi_2^\infty})_v ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) v is regularly supported (Definition 2.16) for v ′ ∈ ⋄ ∪ { v } v'\in\diamond\cup\{v\} v ′ ∈ ⋄ ∪ { v } ; and ϕ i , v ∞ \phi_{i,v}^\infty ϕ i , v ∞ is regularly supported for i = 1 , 2 i=1,2 i = 1 , 2 .
(1) Suppose that V v V_v V v is split. There exists a Schwartz function Φ v ⋆ ∈ S ( V ( v ) 2 , L ) \Phi_v^\star\in\mathcal{S}(V(v)^2,L) Φ v ⋆ ∈ S ( V ( v ) 2 , L ) such that for every pair ( s 1 , s 2 ) (s_1,s_2) ( s 1 , s 2 ) in L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ \mathbb{L}[\mathbb{L}^\diamond\backslash H(\mathbb{A}_F^{\infty,\diamond})/\mathbb{L}^\diamond]^\lambda L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] λ and every pair ( t 1 , t 2 ) (t_1,t_2) ( t 1 , t 2 ) in ( F × ) + (F^\times)^+ ( F × ) + (so that Z = ( Z 1 : = Z t 1 [ s 1 ϕ 1 ∞ ] L , Z 2 : = Z t 2 [ s 2 ϕ 2 ∞ ] L ) Z=(Z_1:=Z_{t_1}[s_1\phi_1^\infty]_L,Z_2:=Z_{t_2}[s_2\phi_2^\infty]_L) Z = ( Z 1 := Z t 1 [ s 1 ϕ 1 ∞ ] L , Z 2 := Z t 2 [ s 2 ϕ 2 ∞ ] L ) (Definition 3.33) is a disjoint pair),
vol ♭ ( L ) ⋅ ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E t 1 , t 2 ( g 1 , g 2 ) Φ v λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) v ⋅ Φ v ⋆ \operatorname{vol}^{\flat}(L)\cdot\left\langle s_1^*Z_{t_1}^{\lambda}(\phi_1^\infty)_L(\mathfrak{g}_1),s_2^*Z_{t_2}^{\lambda}(\phi_2^\infty)_L(\mathfrak{g}_2)\right\rangle_{\lambda,L,Z,u}^{\ell}
=
E_{t_1,t_2}(\mathfrak{g}_1,\mathfrak{g}_2)\Phi_v^\lambda\cdot(s_1\phi_1^\infty\otimes s_2\overline{\phi_2^\infty})_v\cdot\Phi_v^\star vol ♭ ( L ) ⋅ ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E t 1 , t 2 ( g 1 , g 2 ) Φ v λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) v ⋅ Φ v ⋆ holds (in C ⊗ Q R \mathbb{C}\otimes_{\mathbb{Q}}\mathbb{R} C ⊗ Q R ) for every pair ( g 1 , g 2 ) ∈ G 1 ( F ∞ ) 2 (\mathfrak{g}_1,\mathfrak{g}_2)\in G_1(F_\infty)^2 ( g 1 , g 2 ) ∈ G 1 ( F ∞ ) 2 and every prime ℓ ≠ p \ell\ne p ℓ = p , where E t 1 , t 2 E_{t_1,t_2} E t 1 , t 2 is introduced in Definition 2.22(1).
(2) Suppose that V v V_v V v is nonsplit. Consider further a pair ( s 1 , s 2 ) (s_1,s_2) ( s 1 , s 2 ) in L [ L ⋄ \ H ( A F ∞ ) / L ] ⋄ L[L^\diamond\backslash H(\mathbb{A}_F^\infty)/L]^\diamond L [ L ⋄ \ H ( A F ∞ ) / L ] ⋄ . For every element t 2 ∈ ( F × ) + t_2\in(F^\times)^+ t 2 ∈ ( F × ) + , there exists a V λ ⊗ E λ L V_\lambda\otimes_{E_\lambda}L V λ ⊗ E λ L -valued Schwartz function φ t 2 ⋄ \varphi_{t_2}^\diamond φ t 2 ⋄ on V ( v ) ∞ V(v)^\infty V ( v ) ∞ (depending on the data so far) such that
vol ♮ ( L ) ⋅ ⟨ s 1 ⋄ Z t 1 λ ( φ 1 ⋄ ) L ( g 1 ) , s 2 ⋄ Z t 2 λ ( φ 2 ⋄ ) L ⟩ L , Z , u ℓ = ∑ x 1 ⋄ ∈ E ( V ( v ) ) t 1 ω 1 , ∞ ( g 1 ) ( P [ φ t 2 ⋄ ( x 1 ⋄ ) ] ⋅ φ ∞ 0 ( x 1 ⋄ ) ) \operatorname{vol}^{\natural}(L)\cdot\left\langle s_1^\diamond Z_{t_1}^{\lambda}(\varphi_1^\diamond)_L(g_1),s_2^\diamond Z_{t_2}^{\lambda}(\varphi_2^\diamond)_L\right\rangle_{L,\mathbb{Z},u}^{\ell}
=
\sum_{x_1^\diamond\in E(V(v))_{t_1}}\omega_{1,\infty}(g_1)\left(P[\varphi_{t_2}^\diamond(x_1^\diamond)]\cdot\varphi_\infty^0(x_1^\diamond)\right) vol ♮ ( L ) ⋅ ⟨ s 1 ⋄ Z t 1 λ ( φ 1 ⋄ ) L ( g 1 ) , s 2 ⋄ Z t 2 λ ( φ 2 ⋄ ) L ⟩ L , Z , u ℓ = x 1 ⋄ ∈ E ( V ( v ) ) t 1 ∑ ω 1 , ∞ ( g 1 ) ( P [ φ t 2 ⋄ ( x 1 ⋄ )] ⋅ φ ∞ 0 ( x 1 ⋄ ) )
holds (in C ⊗ Q ℓ \mathbb{C}\otimes\mathbb{Q}_\ell C ⊗ Q ℓ ) for every t 1 ∈ ( F × ) + t_1\in(F^\times)^+ t 1 ∈ ( F × ) + , every element g 1 ∈ G 1 ( F ∞ ) g_1\in G_1(F_\infty) g 1 ∈ G 1 ( F ∞ ) and every prime ℓ ≠ p \ell\ne p ℓ = p .
We will prove the above proposition in the next two sections (as Proposition 5.17 and Proposition 5.22 ), according to whether V v V_v V v is split or nonsplit.
Remark 5.10. The conclusion in Proposition 5.9 (1) is stronger than the conclusion in Proposition 5.9 (2). The reason we are not able to obtain the similar statement as in (1) when V v V_v V v is nonsplit is that, in this case, it is difficult to control flat extensions of special classes in the integral model X L \mathcal{X}_L X L we use below.
In the rest of this section, we make some common preparation for both cases. Fix a complete maximal unramified extension E ˘ u \breve{E}_u E ˘ u of E u E_u E u , with κ \kappa κ its residue field. Put R 0 : = O E u R_0:=O_{E_u} R 0 := O E u and R : = O E ˘ u R:=O_{\breve{E}_u} R := O E ˘ u for short.
By the projection formula for local heights, we may assume, without lost of generality, that the following holds.
Assumption 5.11. There exist integers a < b a<b a < b such that
L p L_p L p is contained in an optimal open compact subgroup L p b L_p^b L p b (Definition 4.18 with ⋄ = { p } \diamond=\{p\} ⋄ = { p } ) such that L v L_v L v is the subgroup of L v b L_v^b L v b consisting of elements that act trivially on Λ v / p v b − a Λ v \Lambda_v/p_v^{b-a}\Lambda_v Λ v / p v b − a Λ v , where Λ v \Lambda_v Λ v is the lattice of V v V_v V v that L v b L_v^b L v b stabilizes;
( φ i ⋄ ) v (\varphi_i^\diamond)_v ( φ i ⋄ ) v is a function on p v a Λ v / p v b Λ v p_v^a\Lambda_v/p_v^b\Lambda_v p v a Λ v / p v b Λ v (that is, supported on p v a Λ v p_v^a\Lambda_v p v a Λ v and invariant under p v b Λ v p_v^b\Lambda_v p v b Λ v ) for i = 1 , 2 i=1,2 i = 1 , 2 .
Applying Definition 4.21 to L p ⊆ L p b L_p\subseteq L_p^b L p ⊆ L p b , we obtain the functor
X ∙ , ∙ , ( p ) : Hk ( H ( A F ∞ , p ) ) × Hk ( T ( A F ∞ , p ) ) ⟶ Sch f l , p r o j / O E , ( p ) , \mathcal{X}_{\bullet,\bullet,(p)}:\operatorname{Hk}(H(\mathbb{A}_F^{\infty,p}))\times\operatorname{Hk}(T(\mathbb{A}_F^{\infty,p}))\longrightarrow\operatorname{Sch}^{\mathrm{fl,proj}}/O_{E,(p)}, X ∙ , ∙ , ( p ) : Hk ( H ( A F ∞ , p )) × Hk ( T ( A F ∞ , p )) ⟶ Sch fl , proj / O E , ( p ) ,
and put X L ′ , U : = X L ′ , U p , ( p ) ⊗ O E , ( p ) R 0 \mathcal{X}_{L',U}:=\mathcal{X}_{L',U_p,(p)}\otimes_{O_{E,(p)}}R_0 X L ′ , U := X L ′ , U p , ( p ) ⊗ O E , ( p ) R 0 for every L ′ ∈ Hk ( H ( A F ∞ ) ) L'\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ′ ∈ Hk ( H ( A F ∞ )) with L p ′ = L p L'_p=L_p L p ′ = L p and U p U_p U p maximal.
Choose an object U p U^p U p of Hk ( T ( A F ∞ , p ) ) \operatorname{Hk}(T(\mathbb{A}_F^{\infty,p})) Hk ( T ( A F ∞ , p )) and a framing P ⋆ P^\star P ⋆ as in Remark 4.19. Put X L ′ : = X L ′ , U / Ξ U p \mathcal{X}_{L'}:=\mathcal{X}_{L',U}/\Xi_{U_p} X L ′ := X L ′ , U / Ξ U p (which is canonically independent of the choice of U p U^p U p ). Denote by X ^ L ′ \widehat{\mathcal{X}}_{L'} X L ′ the formal completion of X L ′ ⊗ R 0 R \mathcal{X}_{L'}\otimes_{R_0}R X L ′ ⊗ R 0 R along the basic locus. Then (45) implies the following isomorphism of basic uniformization
X ^ L ′ ≃ H ∗ ( F ) \ ( N b − a × H ( A F ∞ ) / L ′ v ) (57) \widehat{\mathcal{X}}_{L'}\simeq H^\ast(F)\backslash\left(\mathcal{N}^{b-a}\times H(\mathbb{A}_F^\infty)/L'^{v}\right) \tag*{(57)} X L ′ ≃ H ∗ ( F ) \ ( N b − a × H ( A F ∞ ) / L ′ v ) (57)
functorial in L ′ ⋄ ∈ Hk ( H ( A F ∞ ) ) L'^\diamond\in\operatorname{Hk}(H(\mathbb{A}_F^\infty)) L ′ ⋄ ∈ Hk ( H ( A F ∞ )) , where N b − a \mathcal{N}^{b-a} N b − a is described in Definition 5.12 below. Note that when V v V_v V v is nonsplit, the entire special fiber X L ′ ⊗ R 0 κ \mathcal{X}_{L'}\otimes_{R_0}\kappa X L ′ ⊗ R 0 κ is basic.
Definition 5.12. Put ( X ⋆ , λ ⋆ ) : = ( A ⋆ [ p ∞ ] , λ ⋆ [ p ∞ ] ) (X^\star,\lambda^\star):=(A^\star[p^\infty],\lambda^\star[p^\infty]) ( X ⋆ , λ ⋆ ) := ( A ⋆ [ p ∞ ] , λ ⋆ [ p ∞ ]) by abuse of notation on λ ⋆ \lambda^\star λ ⋆ and similarly ( X 0 ⋆ , λ 0 ⋆ ) : = ( A 0 ⋆ [ p ∞ ] , λ 0 ⋆ [ p ∞ ] ) (X_0^\star,\lambda_0^\star):=(A_0^\star[p^\infty],\lambda_0^\star[p^\infty]) ( X 0 ⋆ , λ 0 ⋆ ) := ( A 0 ⋆ [ p ∞ ] , λ 0 ⋆ [ p ∞ ]) . Over N ♭ \mathcal{N}^\flat N ♭ (see Remark 4.19 (U3)), we have the universal quasi-isogeny
ρ : ( X , λ ) ⊗ R κ ⟶ ( X ⋆ , λ ⋆ ) ⊗ R ( N ♭ ⊗ R κ ) \rho:(X,\lambda)\otimes_R\kappa\longrightarrow(X^\star,\lambda^\star)\otimes_R(\mathcal{N}^\flat\otimes_R\kappa) ρ : ( X , λ ) ⊗ R κ ⟶ ( X ⋆ , λ ⋆ ) ⊗ R ( N ♭ ⊗ R κ )
of polarized O F v O_{F_v} O F v -divisible groups with R 0 R_0 R 0 -actions. Write N η ♭ \mathcal{N}_\eta^\flat N η ♭ for the generic fiber of N ♭ \mathcal{N}^\flat N ♭ as a smooth rigid analytic space over R [ 1 / p ] R[1/p] R [ 1/ p ] of dimension one.
For every integer δ ≥ 0 \delta\geq0 δ ≥ 0 , denote by N η δ → N η ♭ \mathcal{N}_\eta^\delta\to\mathcal{N}_\eta^\flat N η δ → N η ♭ the finite étale cover classifying isometries
η v : Λ v / p v δ Λ v → ∼ Hom O E 0 , p ( X 0 ⋆ [ p v δ ] ( S ) , X [ p v δ ] ( S ) ) \eta_v:\Lambda_v/p_v^\delta\Lambda_v\xrightarrow{\sim}\operatorname{Hom}_{O_{E_0,p}}\left(X_0^\star[p_v^\delta](S),X[p_v^\delta](S)\right) η v : Λ v / p v δ Λ v ∼ Hom O E 0 , p ( X 0 ⋆ [ p v δ ] ( S ) , X [ p v δ ] ( S ) )
over any rigid analytic space S S S over N η ♭ \mathcal{N}_\eta^\flat N η ♭ of finite type, where the hermitian form on the latter module is induced from the polarizations λ 0 ⋆ \lambda_0^\star λ 0 ⋆ and λ \lambda λ , and define N δ \mathcal{N}^\delta N δ to be the regular closure of N ♭ \mathcal{N}^\flat N ♭ in N η δ \mathcal{N}_\eta^\delta N η δ (Definition 4.20). Then N δ \mathcal{N}^\delta N δ is a regular formal scheme over R R R of relative dimension one.
Definition 5.13. Take a nonzero element x ⋆ ∈ V v ⋆ x^{\star} \in V^{\star}_{v} x ⋆ ∈ V v ⋆ and an integer δ ≥ 0 \delta\ge0 δ ≥ 0 .
(1) We denote by Z a b ( x ⋆ ) \mathcal{Z}^{b}_{a}(x^{\star}) Z a b ( x ⋆ ) the maximal closed subscheme of N b \mathcal{N}^{b} N b over which the composite quasi-map
ϖ − a ⋅ ρ − 1 ∘ x ⋆ : X 0 ⋆ ⊗ R κ ⟶ A ⋆ [ p ∞ ] ⊗ R κ ⟶ X × N b Spec κ \varpi^{-a}\cdot\rho^{-1}\circ x^{\star}:X^{\star}_{0}\otimes_{R}\kappa\longrightarrow A^{\star}[p^{\infty}]\otimes_{R}\kappa\longrightarrow X\times_{\mathcal{N}^{b}}\operatorname{Spec}\kappa ϖ − a ⋅ ρ − 1 ∘ x ⋆ : X 0 ⋆ ⊗ R κ ⟶ A ⋆ [ p ∞ ] ⊗ R κ ⟶ X × N b Spec κ
lifts to a map of O F v \mathcal{O}_{F_{v}} O F v -divisible groups for some hence every uniformizer ϖ \varpi ϖ of F v F_{v} F v , where the last base change is with respect to the closed point Spec κ → N b \operatorname{Spec}\kappa\to\mathcal{N}^{b} Spec κ → N b given by the framing P ⋆ P^{\star} P ⋆ . Write Z a δ ( x ⋆ ) \mathcal{Z}^{\delta}_{a}(x^{\star}) Z a δ ( x ⋆ ) for the inverse image of Z a b ( x ⋆ ) \mathcal{Z}^{b}_{a}(x^{\star}) Z a b ( x ⋆ ) in N δ \mathcal{N}^{\delta} N δ .
(2) For every element x ∈ p v a Λ v / p v a + δ Λ v x\in p_{v}^{a}\Lambda_{v}/p_{v}^{a+\delta}\Lambda_{v} x ∈ p v a Λ v / p v a + δ Λ v , define Z a δ ( x ⋆ , x ) \mathcal{Z}^{\delta}_{a}(x^{\star},x) Z a δ ( x ⋆ , x ) to be the maximal vertical divisor contained in Z a δ ( x ⋆ ) \mathcal{Z}^{\delta}_{a}(x^{\star}) Z a δ ( x ⋆ ) such that over the generic fiber,
( ϖ − a ⋅ ρ − 1 ∘ x ⋆ ) [ p v δ ] = η v ( ϖ − a ⋅ x ) (\varpi^{-a}\cdot\rho^{-1}\circ x^{\star})[p_{v}^{\delta}]=\eta_{v}(\varpi^{-a}\cdot x) ( ϖ − a ⋅ ρ − 1 ∘ x ⋆ ) [ p v δ ] = η v ( ϖ − a ⋅ x )
holds for some hence every uniformizer ϖ \varpi ϖ of F v F_{v} F v .
(3) For every L L L -valued function ϕ \phi ϕ on p v a Λ v / p v a + δ Λ v p_{v}^{a}\Lambda_{v}/p_{v}^{a+\delta}\Lambda_{v} p v a Λ v / p v a + δ Λ v , put
Z a δ ( x ⋆ , ϕ ) : = ∑ x ∈ p v a Λ v / p v a + δ Λ v ϕ ( x ) ⋅ Z a δ ( x ⋆ , x ) ∈ Div ( N δ ) ⊗ L . \mathcal{Z}^{\delta}_{a}(x^{\star},\phi):=\sum_{x\in p_{v}^{a}\Lambda_{v}/p_{v}^{a+\delta}\Lambda_{v}}\phi(x)\cdot\mathcal{Z}^{\delta}_{a}(x^{\star},x)\in\operatorname{Div}(\mathcal{N}^{\delta})\otimes L. Z a δ ( x ⋆ , ϕ ) := x ∈ p v a Λ v / p v a + δ Λ v ∑ ϕ ( x ) ⋅ Z a δ ( x ⋆ , x ) ∈ Div ( N δ ) ⊗ L .
Lemma 5.14. Consider an integer δ ≥ 0 \delta\ge0 δ ≥ 0 and an element x ⋆ ∈ V v ⋆ x^{\star}\in V^{\star}_{v} x ⋆ ∈ V v ⋆ with T ( x ⋆ ) ≠ 0 T(x^{\star})\ne0 T ( x ⋆ ) = 0 . For every integer N ≥ 1 N\ge1 N ≥ 1 , there exists an open neighbourhood U N U_{N} U N of x ⋆ x^{\star} x ⋆ in V v ⋆ V^{\star}_{v} V v ⋆ , such that for every y ⋆ ∈ U N y^{\star}\in U_{N} y ⋆ ∈ U N , Z a δ ( x ⋆ , x ) \mathcal{Z}^{\delta}_{a}(x^{\star},x) Z a δ ( x ⋆ , x ) and Z a δ ( y ⋆ , x ) \mathcal{Z}^{\delta}_{a}(y^{\star},x) Z a δ ( y ⋆ , x ) have the same image in Div ( N δ ⊗ R R / p N ) \operatorname{Div}(\mathcal{N}^{\delta}\otimes_{R}R/p^{N}) Div ( N δ ⊗ R R / p N ) for every x ∈ p v a Λ v / p v a + δ Λ v x\in p_{v}^{a}\Lambda_{v}/p_{v}^{a+\delta}\Lambda_{v} x ∈ p v a Λ v / p v a + δ Λ v , where Div ( N δ ⊗ R R / p N ) \operatorname{Div}(\mathcal{N}^{\delta}\otimes_{R}R/p^{N}) Div ( N δ ⊗ R R / p N ) denotes the abelian group of Cartier divisors on N δ ⊗ R R / p N \mathcal{N}^{\delta}\otimes_{R}R/p^{N} N δ ⊗ R R / p N .
Proof. Without lost of generality, we may assume a = 0 a=0 a = 0 and suppress it from the subscripts.
We first consider the case where δ = 0 \delta=0 δ = 0 . Denote by H H H the (topological) subgroup of GU ( V v ⋆ ) ( F v ) \operatorname{GU}(V^{\star}_{v})(F_{v}) GU ( V v ⋆ ) ( F v ) consisting of elements whose similitudes are in O F v × \mathcal{O}^{\times}_{F_{v}} O F v × . Then H H H acts on ( A ⋆ [ p ∞ ] , λ ∞ ⋆ ) (A^{\star}[p^{\infty}],\lambda^{\star}_{\infty}) ( A ⋆ [ p ∞ ] , λ ∞ ⋆ ) by quasi-automorphisms of height zero, which induces a continuous action on N b \mathcal{N}^{b} N b by composition hence on N 0 \mathcal{N}^{0} N 0 by the functoriality of taking regular closure. Take an element x ⋆ ∈ V v ⋆ x^{\star}\in V^{\star}_{v} x ⋆ ∈ V v ⋆ with T ( x ⋆ ) ≠ 0 T(x^{\star})\ne0 T ( x ⋆ ) = 0 . Then for h ∈ H h\in H h ∈ H , h Z 0 ( x ⋆ ) = Z 0 ( h x ⋆ ) h\mathcal{Z}^{0}(x^{\star})=\mathcal{Z}^{0}(hx^{\star}) h Z 0 ( x ⋆ ) = Z 0 ( h x ⋆ ) by Definition 5.13. For every N ≥ 1 N\ge1 N ≥ 1 , there exists an open compact subgroup I N I_{N} I N of H H H such that the image of h Z 0 ( x ⋆ ) h\mathcal{Z}^{0}(x^{\star}) h Z 0 ( x ⋆ ) in Div ( N 0 ⊗ R R / p N ) \operatorname{Div}(\mathcal{N}^{0}\otimes_{R}R/p^{N}) Div ( N 0 ⊗ R R / p N ) does not depend on h ∈ I N h\in I_{N} h ∈ I N . Since { I ⊂ H ∣ I open compact subgroups of H } \{I\subset H\mid I\text{ open compact subgroups of }H\} { I ⊂ H ∣ I open compact subgroups of H } is a topological basis of x ⋆ x^{\star} x ⋆ , the lemma follows in this case.
For general δ \delta δ , the action of H H H on N η 0 \mathcal{N}^{0}_{\eta} N η 0 lifts naturally to a (continuous) action on N η δ \mathcal{N}^{\delta}_{\eta} N η δ . By the functoriality of taking regular closure, we obtain a continuous action of H H H on N δ \mathcal{N}^{\delta} N δ lifting that on N 0 \mathcal{N}^{0} N 0 . Take an element x ⋆ ∈ V v ⋆ x^{\star}\in V^{\star}_{v} x ⋆ ∈ V v ⋆ with T ( x ⋆ ) ≠ 0 T(x^{\star})\ne0 T ( x ⋆ ) = 0 . Then for h ∈ H h\in H h ∈ H and every x ∈ Λ v / p δ Λ v x\in\Lambda_{v}/p^{\delta}\Lambda_{v} x ∈ Λ v / p δ Λ v , h Z δ ( x ⋆ , x ) = Z δ ( h x ⋆ , x ) h\mathcal{Z}^{\delta}(x^{\star},x)=\mathcal{Z}^{\delta}(hx^{\star},x) h Z δ ( x ⋆ , x ) = Z δ ( h x ⋆ , x ) by Definition 5.13. The lemma then follows in general by the same reason as above. □ \square □
Proof of Proposition 5.9(1)
In this section, we assume V v V_{v} V v split. We keep the notation from § 5.3 and take δ : = b − a \delta:=b-a δ := b − a .
Lemma 5.15. For every nonzero element x ⋆ ∈ V v ⋆ x^{\star}\in V^{\star}_{v} x ⋆ ∈ V v ⋆ and every element x ∈ p v a Λ v / p v b Λ v x\in p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v} x ∈ p v a Λ v / p v b Λ v , there is a unique element Z a δ ( x ⋆ , x ) † ∈ Div ( N δ ) ⊗ Q \mathcal{Z}^{\delta}_{a}(x^{\star},x)^{\dagger}\in\operatorname{Div}(\mathcal{N}^{\delta})\otimes\mathbb{Q} Z a δ ( x ⋆ , x ) † ∈ Div ( N δ ) ⊗ Q satisfying that
(a) Z a δ ( x ⋆ , x ) † − Z a δ ( x ⋆ , x ) \mathcal{Z}^{\delta}_{a}(x^{\star},x)^{\dagger}-\mathcal{Z}^{\delta}_{a}(x^{\star},x) Z a δ ( x ⋆ , x ) † − Z a δ ( x ⋆ , x ) is a ( Q ) (\mathbb{Q}) ( Q ) -linear combination of irreducible components of the induced reduced subscheme of N δ \mathcal{N}^{\delta} N δ ;
(b) for every irreducible component C C C of the induced reduced subscheme of N δ \mathcal{N}^{\delta} N δ , Z a δ ( x ⋆ , x ) † ∣ C \mathcal{Z}^{\delta}_{a}(x^{\star},x)^{\dagger}|_{C} Z a δ ( x ⋆ , x ) † ∣ C has degree zero.
Proof. By Lemma 4.22(1), ( N b ) f l (\mathcal{N}^{b})^{\mathrm{fl}} ( N b ) fl is formally smooth over R R R (of relative dimension one). Since N δ \mathcal{N}^{\delta} N δ is the canonical resolution of (isolated) singularities of the normalization of ( N b ) f l (\mathcal{N}^{b})^{\mathrm{fl}} ( N b ) fl in N η δ \mathcal{N}^{\delta}_{\eta} N η δ , it is well-known that the intersection matrix of the reduced fiber of N δ \mathcal{N}^{\delta} N δ is negative definite. The lemma then follows. □ \square □
For every L L L -valued function ϕ \phi ϕ on p v a Λ v / p v b Λ v p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v} p v a Λ v / p v b Λ v , put
Z a δ ( x ⋆ , ϕ ) † : = ∑ x ∈ p v a Λ v / p v b Λ v ϕ ( x ) ⋅ Z a δ ( x ⋆ , x ) † ∈ Div ( N δ ) ⊗ L . \mathcal{Z}^{\delta}_{a}(x^{\star},\phi)^{\dagger}:=\sum_{x\in p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v}}\phi(x)\cdot\mathcal{Z}^{\delta}_{a}(x^{\star},x)^{\dagger}\in\operatorname{Div}(\mathcal{N}^{\delta})\otimes L. Z a δ ( x ⋆ , ϕ ) † := x ∈ p v a Λ v / p v b Λ v ∑ ϕ ( x ) ⋅ Z a δ ( x ⋆ , x ) † ∈ Div ( N δ ) ⊗ L .
Lemma 5.16. There exists a unique L L L -valued Schwartz function Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] on ( V v ⋆ ) 2 (V_{v}^{\star})^{2} ( V v ⋆ ) 2 such that for every pair ( x 1 ⋆ , x 2 ⋆ ) ∈ ( V v ⋆ ∖ { 0 } ) 2 (x_{1}^{\star},x_{2}^{\star})\in(V_{v}^{\star}\setminus\{0\})^{2} ( x 1 ⋆ , x 2 ⋆ ) ∈ ( V v ⋆ ∖ { 0 } ) 2 ,
Φ v [ a , b ] ( x 1 ⋆ , x 2 ⋆ ) = Z a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) † ∙ Z a δ ( x 2 ⋆ , ϕ 2 , v ∞ ) † \Phi_{v}^{[a,b]}(x_{1}^{\star},x_{2}^{\star})=\mathcal{Z}_{a}^{\delta}(x_{1}^{\star},\phi_{1,v}^{\infty})^{\dagger}\mathbin{\bullet}\mathcal{Z}_{a}^{\delta}(x_{2}^{\star},\phi_{2,v}^{\infty})^{\dagger} Φ v [ a , b ] ( x 1 ⋆ , x 2 ⋆ ) = Z a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) † ∙ Z a δ ( x 2 ⋆ , ϕ 2 , v ∞ ) † = ∑ x 1 ∈ p v a Λ v / p v b Λ v ∑ x 2 ∈ p v a Λ v / p v b Λ v ϕ 1 , v ∞ ( x 1 ) ϕ 2 , v ∞ ( x 2 ) ‾ ⋅ Z a δ ( x 1 ⋆ , x 1 ) † ∙ Z a δ ( x 2 ⋆ , x 2 ) † . =\sum_{x_{1}\in p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v}}\sum_{x_{2}\in p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v}}\phi_{1,v}^{\infty}(x_{1})\overline{\phi_{2,v}^{\infty}(x_{2})}\cdot\mathcal{Z}_{a}^{\delta}(x_{1}^{\star},x_{1})^{\dagger}\mathbin{\bullet}\mathcal{Z}_{a}^{\delta}(x_{2}^{\star},x_{2})^{\dagger}. = x 1 ∈ p v a Λ v / p v b Λ v ∑ x 2 ∈ p v a Λ v / p v b Λ v ∑ ϕ 1 , v ∞ ( x 1 ) ϕ 2 , v ∞ ( x 2 ) ⋅ Z a δ ( x 1 ⋆ , x 1 ) † ∙ Z a δ ( x 2 ⋆ , x 2 ) † . in which ∙ \mathbin{\bullet} ∙ denotes the intersection pairing on N δ \mathcal{N}^{\delta} N δ as an L L L -hermitian form. Moreover, Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] is invariant under the action of H ⋆ ( F v ) H^{\star}(F_{v}) H ⋆ ( F v ) .
Here, note that since ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v (\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}})_{v} ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) v is regularly supported, the intersection Z a δ ( x 1 ⋆ , x 1 ) ∩ Z a δ ( x 2 ⋆ , x 2 ) \mathcal{Z}_{a}^{\delta}(x_{1}^{\star},x_{1})\cap\mathcal{Z}_{a}^{\delta}(x_{2}^{\star},x_{2}) Z a δ ( x 1 ⋆ , x 1 ) ∩ Z a δ ( x 2 ⋆ , x 2 ) is finite if ϕ 1 , v ∞ ( x 1 ) ϕ 2 , v ∞ ( x 2 ) ‾ ≠ 0 \phi_{1,v}^{\infty}(x_{1})\overline{\phi_{2,v}^{\infty}(x_{2})}\ne0 ϕ 1 , v ∞ ( x 1 ) ϕ 2 , v ∞ ( x 2 ) = 0 .
Proof. It is clear that if Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] exists, then it is unique, L L L -valued, invariant under H ⋆ ( F v ) H^{\star}(F_{v}) H ⋆ ( F v ) . The continuity of Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] on ( V v ⋆ ∖ { 0 } ) 2 (V_{v}^{\star}\setminus\{0\})^{2} ( V v ⋆ ∖ { 0 } ) 2 follows from Lemma 5.14.
The lemma follows if we can show that Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] is compactly supported in ( V v ⋆ ∖ { 0 } ) 2 (V_{v}^{\star}\setminus\{0\})^{2} ( V v ⋆ ∖ { 0 } ) 2 . Indeed, Z a δ ( x ⋆ , x ) \mathcal{Z}_{a}^{\delta}(x^{\star},x) Z a δ ( x ⋆ , x ) hence Z a δ ( x ⋆ , x ) † \mathcal{Z}_{a}^{\delta}(x^{\star},x)^{\dagger} Z a δ ( x ⋆ , x ) † vanish unless ( x ⋆ , x ⋆ ) ≡ ( x , x ) ( m o d p v 2 b ) (x^{\star},x^{\star})\equiv(x,x)\pmod{p_{v}^{2b}} ( x ⋆ , x ⋆ ) ≡ ( x , x ) ( mod p v 2 b ) . Now as both ϕ 1 , v ∞ \phi_{1,v}^{\infty} ϕ 1 , v ∞ and ϕ 2 , v ∞ \phi_{2,v}^{\infty} ϕ 2 , v ∞ are regularly supported, the support of Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] is contained in the subset D ( 2 b , 2 a ] 2 D^{2}_{(2b,2a]} D ( 2 b , 2 a ] 2 , where
D ( 2 b , 2 a ] 2 : = { x ⋆ ∈ V v ⋆ ∣ ( x ⋆ , x ⋆ ) ∈ p v 2 a ∖ p v 2 b } . D_{(2b,2a]}^{2}:=\{x^{\star}\in V_{v}^{\star}\mid(x^{\star},x^{\star})\in p_{v}^{2a}\setminus p_{v}^{2b}\}. D ( 2 b , 2 a ] 2 := { x ⋆ ∈ V v ⋆ ∣ ( x ⋆ , x ⋆ ) ∈ p v 2 a ∖ p v 2 b } . Since V v ⋆ V_{v}^{\star} V v ⋆ is anisotropic, D ( 2 b , 2 a ] D_{(2b,2a]} D ( 2 b , 2 a ] is compact in V v ⋆ ∖ { 0 } V_{v}^{\star}\setminus\{0\} V v ⋆ ∖ { 0 } .
The lemma is proved. □
Proposition 5.17. Suppose that we are in the situation of Proposition 5.9 (1) together with Assumption 5.11. If we put
Φ v ⋆ : = vol ( L v ) vol ( H ⋆ ( F v ) ) Φ v [ a , b ] , \Phi_{v}^{\star}:=\frac{\operatorname{vol}(L_{v})}{\operatorname{vol}(H^{\star}(F_{v}))}\Phi_{v}^{[a,b]}, Φ v ⋆ := vol ( H ⋆ ( F v )) vol ( L v ) Φ v [ a , b ] , then for every pair ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) in L [ L × \ H ( A F ∞ ) / L × ] L[L^{\times}\backslash H(\mathbb{A}_{F}^{\infty})/L^{\times}] L [ L × \ H ( A F ∞ ) / L × ] and every pair ( t 1 , t 2 ) (t_{1},t_{2}) ( t 1 , t 2 ) in ( F × ) + (F^{\times})^{+} ( F × ) + ,
vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ⋆ Z t 2 ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) v ⋅ Φ v ⋆ ) \operatorname{vol}^{\natural}(L)\cdot\left\langle s_{1}^{\star}Z_{t_{1}}(\phi_{1}^{\infty})_{L}(g_{1}),s_{2}^{\star}Z_{t_{2}}(\phi_{2}^{\infty})_{L}(g_{2})\right\rangle_{\lambda,L,Z,u}^{\ell}=E_{t_{1},t_{2}}((g_{1},g_{2}),\Phi_{\infty}^{\lambda}\cdot(s_{1}\phi_{1}^{\infty}\otimes s_{2}\overline{\phi_{2}^{\infty}})^{v}\cdot\Phi_{v}^{\star}) vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ⋆ Z t 2 ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , Z , u ℓ = E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) v ⋅ Φ v ⋆ ) holds (in C ⊗ Q ℓ \mathbb{C}\otimes\mathbb{Q}_{\ell} C ⊗ Q ℓ ) for every pair ( g 1 , g 2 ) ∈ G 1 ( F ∞ ) 2 (g_{1},g_{2})\in G_{1}(F_{\infty})^{2} ( g 1 , g 2 ) ∈ G 1 ( F ∞ ) 2 and every prime ℓ ≠ p \ell\ne p ℓ = p .
Take a rational prime ℓ ≠ p \ell\ne p ℓ = p . Note that the E ℓ λ E_{\ell}^{\lambda} E ℓ λ -linear local system V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ over X L X_{L} X L has a uniformization:
V ℓ , L , λ ∣ X L = H ⋆ ( F ) \ ( N δ × H ( A F ∞ , v ) / L × × V ℓ , λ ) V_{\ell,L,\lambda}|_{X_{L}}=H^{\star}(F)\backslash\left(\mathcal{N}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{\times}\times V_{\ell,\lambda}\right) V ℓ , L , λ ∣ X L = H ⋆ ( F ) \ ( N δ × H ( A F ∞ , v ) / L × × V ℓ , λ )
compatible with (57) , in which H ⋆ ( F ) H^{\star}(F) H ⋆ ( F ) acts on V ℓ , λ : = ( V ℓ ) λ V_{\ell,\lambda}:=(V_{\ell})_{\lambda} V ℓ , λ := ( V ℓ ) λ via the natural map H ⋆ ( F ) → H ( F ℓ ) H^{\star}(F)\to H(F_{\ell}) H ⋆ ( F ) → H ( F ℓ ) .
Construction 5.18. Consider a Schwartz function ϕ ∞ ∈ S ( V ∞ , L ) \phi^{\infty}\in\mathcal{S}(V^{\infty},L) ϕ ∞ ∈ S ( V ∞ , L ) of the form ϕ v ⊗ ϕ ∞ , v \phi_{v}\otimes\phi^{\infty,v} ϕ v ⊗ ϕ ∞ , v in which ϕ v \phi_{v} ϕ v is a regularly supported L L L -valued function on p v a Λ v / p v b Λ v p_{v}^{a}\Lambda_{v}/p_{v}^{b}\Lambda_{v} p v a Λ v / p v b Λ v .
For x v ⋆ ∈ V v ⋆ ∖ { 0 } x_{v}^{\star}\in V_{v}^{\star}\setminus\{0\} x v ⋆ ∈ V v ⋆ ∖ { 0 } , put
Z ( x v ⋆ , ϕ v ) h : = Z a δ ( x v ⋆ , ϕ v ) † × { h } ∈ Div ( N δ × H ( A F ∞ , v ) / L ∨ ) ⊗ L , \mathcal{Z}(x_{v}^{\star},\phi_{v})_{h}:=\mathcal{Z}_{a}^{\delta}(x_{v}^{\star},\phi_{v})^{\dagger}\times\{h\}\in\operatorname{Div}(\mathcal{N}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{\vee})\otimes L, Z ( x v ⋆ , ϕ v ) h := Z a δ ( x v ⋆ , ϕ v ) † × { h } ∈ Div ( N δ × H ( A F ∞ , v ) / L ∨ ) ⊗ L , and denote by
C ( x v ⋆ , ϕ v ) h ∈ H e ˊ t , ∣ Z ( x v ⋆ , ϕ v ) h ∣ 2 ( N δ × H ( A F ∞ , v ) / L ∨ , L ℓ ( 1 ) ) \mathcal{C}(x_{v}^{\star},\phi_{v})_{h}\in H_{\mathrm{\acute et},|\mathcal{Z}(x_{v}^{\star},\phi_{v})_{h}|}^{2}(\mathcal{N}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{\vee},L_{\ell}(1)) C ( x v ⋆ , ϕ v ) h ∈ H e ˊ t , ∣ Z ( x v ⋆ , ϕ v ) h ∣ 2 ( N δ × H ( A F ∞ , v ) / L ∨ , L ℓ ( 1 )) its refined ℓ \ell ℓ -adic cycle class.
For every t ∈ ( F × ) + t\in(F^{\times})^{+} t ∈ ( F × ) + , define Z t ( ϕ ∞ ) L \mathcal{Z}_{t}(\phi^{\infty})_{L} Z t ( ϕ ∞ ) L to be the unique element in Div ( X L ⊗ R 0 R ) ⊗ L \operatorname{Div}(X_{L}\otimes_{R_{0}}R)\otimes L Div ( X L ⊗ R 0 R ) ⊗ L supported on the basic locus whose pullback to N δ × H ( A F ∞ , v ) / L ∨ \mathcal{N}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{\vee} N δ × H ( A F ∞ , v ) / L ∨ is given by
Z t ( ϕ ∞ ) L : = ∑ x ⋆ ∈ V t ⋆ ∑ h ∈ H ( A F ∞ , v ) / L ∨ ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ Z ( x v ⋆ , ϕ v ) h . \mathcal{Z}_{t}(\phi^{\infty})_{L}:=\sum_{x^{\star}\in V_{t}^{\star}}\sum_{h\in H(\mathbb{A}_{F}^{\infty,v})/L^{\vee}}\phi^{\infty,v}(h^{-1}x^{\star})\cdot\mathcal{Z}(x_{v}^{\star},\phi_{v})_{h}. Z t ( ϕ ∞ ) L := x ⋆ ∈ V t ⋆ ∑ h ∈ H ( A F ∞ , v ) / L ∨ ∑ ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ Z ( x v ⋆ , ϕ v ) h .
(3) For every t ∈ ( F × ) + t \in(F^{\times})^{+} t ∈ ( F × ) + , denote by
C t λ ( ϕ ∞ ) L ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) ⊗ E λ L C_{t}^{\lambda}(\phi^{\infty})_{L} \in H_{\mathrm{\acute et},|Z_{t}(\phi^{\infty})_{L}|}^{2}(\mathcal{X}_{L}\otimes_{R_{0}}R,V_{\ell,L,\lambda}(1))\otimes_{E_{\lambda}}L C t λ ( ϕ ∞ ) L ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 )) ⊗ E λ L
the unique element whose restriction to M δ × H ( A F ∞ , v ) / L v \mathcal{M}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{v} M δ × H ( A F ∞ , v ) / L v is given by the H ∗ ( F ) H^{\ast}(F) H ∗ ( F ) -invariant class
∑ x ∗ ∈ V t ∗ ( ∑ h ∈ H ( A F ∞ , v ) / L v ϕ ∞ , v ( h − 1 x ∗ ) ⋅ Z ( x v ∗ , ϕ v ) h ) ⊗ μ λ ( x ∗ ) \sum_{x^{\ast}\in V_{t}^{\ast}}\left(\sum_{h\in H(\mathbb{A}_{F}^{\infty,v})/L^{v}}\phi^{\infty,v}(h^{-1}x^{\ast})\cdot\mathcal{Z}(x^{\ast}_{v},\phi_{v})_{h}\right)\otimes\mu^{\lambda}(x^{\ast}) x ∗ ∈ V t ∗ ∑ h ∈ H ( A F ∞ , v ) / L v ∑ ϕ ∞ , v ( h − 1 x ∗ ) ⋅ Z ( x v ∗ , ϕ v ) h ⊗ μ λ ( x ∗ )
∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( M δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 ) ) ⊗ E ℓ V ℓ ∗ = H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( M δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 ) ) ⊗ E ℓ V ℓ , λ \in H_{\mathrm{\acute et},|Z_{t}(\phi^{\infty})_{L}|}^{2}(\mathcal{M}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{v},\mathcal{L}_{\ell}(1))\otimes_{E_{\ell}}V_{\ell}^{\ast}=H_{\mathrm{\acute et},|Z_{t}(\phi^{\infty})_{L}|}^{2}(\mathcal{M}^{\delta}\times H(\mathbb{A}_{F}^{\infty,v})/L^{v},\mathcal{L}_{\ell}(1))\otimes_{E_{\ell}}V_{\ell,\lambda} ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( M δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 )) ⊗ E ℓ V ℓ ∗ = H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( M δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 )) ⊗ E ℓ V ℓ , λ , where μ λ \mu^{\lambda} μ λ is defined in Construction 2.8(3) (with R = Q R=\mathbb{Q} R = Q ).
Lemma 5.19. Let ϕ ∞ \phi^{\infty} ϕ ∞ be as in Construction 5.18. For every s ∈ L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] 1 s\in L[L^{\diamond}\backslash H(\mathbb{A}_{F}^{\infty,\diamond})/L^{\diamond}]^{1} s ∈ L [ L ⋄ \ H ( A F ∞ , ⋄ ) / L ⋄ ] 1 (Definition 5.8) and every t ∈ ( F × ) + t\in(F^{\times})^{+} t ∈ ( F × ) + , the image of C t λ ( s ϕ ∞ ) L C_{t}^{\lambda}(s\phi^{\infty})_{L} C t λ ( s ϕ ∞ ) L in H e ˊ t 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2}(\mathcal{X}_{L}\otimes_{R_{0}}R,V_{\ell,L,\lambda}(1))\otimes_{E_{\lambda}}L H e ˊ t 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 )) ⊗ E λ L vanishes.
Proof. It suffices to show that for every (reduced) irreducible component C C C of X L ⊗ R 0 κ \mathcal{X}_{L}\otimes_{R_{0}}\kappa X L ⊗ R 0 κ , the image in H e ˊ t 2 ( C , V ℓ , L , λ ( 1 ) ) ⊗ E λ L H_{\mathrm{\acute et}}^{2}(C,V_{\ell,L,\lambda}(1))\otimes_{E_{\lambda}}L H e ˊ t 2 ( C , V ℓ , L , λ ( 1 )) ⊗ E λ L vanishes. There are two cases.
First suppose that C C C is contained in the basic locus. Then the vanishing follows from the property (b) of Lemma 5.15.
Second suppose that C C C is not contained in the basic locus. If λ = 0 \lambda=0 λ = 0 , then s s s preserves and annihilates H e ˊ t 2 ( C , L ℓ ( 1 ) ) H_{\mathrm{\acute et}}^{2}(C,\mathcal{L}_{\ell}(1)) H e ˊ t 2 ( C , L ℓ ( 1 )) by Lemma 4.22(1) (with ⋄ = { p } \diamond=\{p\} ⋄ = { p } ). If λ ≠ 0 \lambda\ne0 λ = 0 , we have H e ˊ t 2 ( C , V ℓ , L , λ ( 1 ) ) = 0 H_{\mathrm{\acute et}}^{2}(C,V_{\ell,L,\lambda}(1))=0 H e ˊ t 2 ( C , V ℓ , L , λ ( 1 )) = 0 by Lemma 4.22(2) (with ⋄ = { p } \diamond=\{p\} ⋄ = { p } ).
The lemma is proved. □
Proof of Proposition 5.17. Similar to the proof of Proposition 5.2, we may assume g 1 = g 2 = 1 2 g_{1}=g_{2}=1_{2} g 1 = g 2 = 1 2 .
By construction, for i = 1 , 2 i=1,2 i = 1 , 2 , the natural image of the class C t i λ ( s i ϕ i ∞ ) L C_{t_{i}}^{\lambda}(s_{i}\phi_{i}^{\infty})_{L} C t i λ ( s i ϕ i ∞ ) L from Construction 5.18(3) in H e ˊ t , Z i , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) H_{\mathrm{\acute et},Z_{i,u}\otimes_{R_{0}}R}^{2}(X_{L,u}\otimes_{R_{0}}R,V_{\ell,L,\lambda}(1)) H e ˊ t , Z i , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 )) coincides with that of s i ∗ Z t i λ ( ϕ i ∞ ) L s_{i}^{\ast}Z_{t_{i}}^{\lambda}(\phi_{i}^{\infty})_{L} s i ∗ Z t i λ ( ϕ i ∞ ) L . Then by Lemma 4.7 (or rather its analogue after base change to R R R ) and Lemma 5.19, we have
⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C t 2 λ ( s 2 ϕ 2 ∞ ) L . \left\langle s_{1}^{\ast}Z_{t_{1}}^{\lambda}(\phi_{1}^{\infty})_{L},s_{2}^{\ast}Z_{t_{2}}^{\lambda}(\phi_{2}^{\infty})_{L}\right\rangle_{\lambda,L,Z,u}^{\ell}=C_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})_{L}\mathbin{\bullet} C_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})_{L}. ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C t 2 λ ( s 2 ϕ 2 ∞ ) L .
Applying the argument for [20 Proposition 8.1] to Construction 5.18, we obtain
C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C t 2 λ ( s 2 ϕ 2 ∞ ) L = ∑ T ∈ Herm 2 ( F ) + ∂ T = ( t 1 , t 2 ) κ T Φ ∞ ( T □ ) ⋅ Φ v [ a , b ] ( T □ ) ⋅ ( ∑ x ∈ V ∗ \ ( V ∞ , v ) T 2 Φ ∞ , v ( x ) ) . C_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})_{L}\mathbin{\bullet} C_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})_{L}
=
\sum_{\substack{T\in\operatorname{Herm}_{2}(F)^{+}\\ \partial T=(t_{1},t_{2})}}
\kappa_{T}^{\Phi^{\infty}}(T^{\square})\cdot\Phi_{v}^{[a,b]}(T^{\square})\cdot
\left(\sum_{x\in V^{\ast}\backslash(V^{\infty,v})_{T}^{2}}\Phi^{\infty,v}(x)\right). C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ C t 2 λ ( s 2 ϕ 2 ∞ ) L = T ∈ Herm 2 ( F ) + ∂ T = ( t 1 , t 2 ) ∑ κ T Φ ∞ ( T □ ) ⋅ Φ v [ a , b ] ( T □ ) ⋅ x ∈ V ∗ \ ( V ∞ , v ) T 2 ∑ Φ ∞ , v ( x ) .
Here, we note that Φ v [ a , b ] \Phi_{v}^{[a,b]} Φ v [ a , b ] factors through the moment map T : ( V v ∗ ) 2 → Herm 2 ( F v ) T:(V_{v}^{\ast})^{2}\to\operatorname{Herm}_{2}(F_{v}) T : ( V v ∗ ) 2 → Herm 2 ( F v ) by Lemma 5.16. Then the proposition follows from the local Siegel–Weil formula. □
Proof of Proposition 5.9(2)
In this section, we assume V v V_{v} V v nonsplit. We keep the notation from §5.3 and take δ : = b − a \delta:=b-a δ := b − a .
Fix an element x 2 ∗ ∈ V ∗ x_{2}^{\ast}\in V^{\ast} x 2 ∗ ∈ V ∗ with T ( x 2 ∗ ) = t 2 T(x_{2}^{\ast})=t_{2} T ( x 2 ∗ ) = t 2 , and denote by ( H ∗ ) x 2 ∗ (H^{\ast})_{x_{2}^{\ast}} ( H ∗ ) x 2 ∗ the stabilizer of x 2 ∗ x_{2}^{\ast} x 2 ∗ in H ∗ H^{\ast} H ∗ . Consider the set
S ( x 2 ∗ ) : = { h ∈ H ( A F ∞ , v ) ∣ s 2 ϕ 2 ∞ , v ( h − 1 x 2 ∗ ) ≠ 0 } . S(x_{2}^{\ast}):=\{h\in H(\mathbb{A}_{F}^{\infty,v})\mid s_{2}\phi_{2}^{\infty,v}(h^{-1}x_{2}^{\ast})\ne0\}. S ( x 2 ∗ ) := { h ∈ H ( A F ∞ , v ) ∣ s 2 ϕ 2 ∞ , v ( h − 1 x 2 ∗ ) = 0 } .
Then the double coset
( H ∗ ) x 2 ∗ ( F ) \ S ( x 2 ∗ ) / L v (H^{\ast})_{x_{2}^{\ast}}(F)\backslash S(x_{2}^{\ast})/L^{v} ( H ∗ ) x 2 ∗ ( F ) \ S ( x 2 ∗ ) / L v
is finite, for which we fix a set { h 1 , … , h n } \{h_{1},\ldots,h_{n}\} { h 1 , … , h n } of representatives.
Take an index 1 ≤ j ≤ n 1\le j\le n 1 ≤ j ≤ n . We
denote by
π j : N δ × { h j } ⟶ Γ j \ ( N δ × h j L ∨ / L ∨ ) \pi_j:\mathfrak{N}^{\delta}\times\{h_j\}\longrightarrow\Gamma_j\backslash(\mathfrak{N}^{\delta}\times h_jL^{\vee}/L^{\vee}) π j : N δ × { h j } ⟶ Γ j \ ( N δ × h j L ∨ / L ∨ ) the natural projection map;
denote by X L , j \mathcal{X}_{L,j} X L , j the unique connected component of X L ⊗ R 0 R \mathcal{X}_L\otimes_{R_0}R X L ⊗ R 0 R such that the formal completion along its special fiber Y L , j : = X L , j ⊗ R κ Y_{L,j}:=\mathcal{X}_{L,j}\otimes_R\kappa Y L , j := X L , j ⊗ R κ coincides with Γ j \ ( N δ × h j L ∨ / L ∨ ) \Gamma_j\backslash(\mathfrak{N}^{\delta}\times h_jL^{\vee}/L^{\vee}) Γ j \ ( N δ × h j L ∨ / L ∨ ) ;
denote by N j δ N_j^{\delta} N j δ the normalization of the induced reduced subscheme of N δ × { h j } = N δ × h j L ∨ / L ∨ \mathfrak{N}^{\delta}\times\{h_j\}=\mathfrak{N}^{\delta}\times h_jL^{\vee}/L^{\vee} N δ × { h j } = N δ × h j L ∨ / L ∨ , and by ∣ N j δ ∣ \lvert N_j^{\delta}\rvert ∣ N j δ ∣ the set of its irreducible components.
For every abelian group A A A , denote by Func c ( ∣ N j δ ∣ , A ) ⊆ Fun ( ∣ N j δ ∣ , A ) \operatorname{Func}_c(\lvert N_j^{\delta}\rvert,A)\subseteq\operatorname{Fun}(\lvert N_j^{\delta}\rvert,A) Func c (∣ N j δ ∣ , A ) ⊆ Fun (∣ N j δ ∣ , A ) the subgroup of finitely supported functions (valued in A A A ), for which we have the map
Δ : Z [ ∣ N j δ ∣ ] ⟶ Func c ( ∣ N j δ ∣ , Z ) \Delta:\mathbb{Z}[\lvert N_j^{\delta}\rvert]\longrightarrow\operatorname{Func}_c(\lvert N_j^{\delta}\rvert,\mathbb{Z}) Δ : Z [∣ N j δ ∣] ⟶ Func c (∣ N j δ ∣ , Z )
given by the intersection in (the regular formal scheme) N δ × { h j } \mathfrak{N}^{\delta}\times\{h_j\} N δ × { h j } , and the map
av Γ j : Func c ( ∣ N j δ ∣ , V λ ⋆ ) ⟶ Fun ( ∣ N j δ ∣ , V λ ⋆ ) Γ j \operatorname{av}^{\Gamma_j}:\operatorname{Func}_c(\lvert N_j^{\delta}\rvert,V_\lambda^\star)\longrightarrow\operatorname{Fun}(\lvert N_j^{\delta}\rvert,V_\lambda^\star)^{\Gamma_j} av Γ j : Func c (∣ N j δ ∣ , V λ ⋆ ) ⟶ Fun (∣ N j δ ∣ , V λ ⋆ ) Γ j
sending f f f to ∑ γ ∈ Γ j γ f \sum_{\gamma\in\Gamma_j}\gamma f ∑ γ ∈ Γ j γ f . The composition av Γ j ∘ ( Δ ⊗ V λ ⋆ ) \operatorname{av}^{\Gamma_j}\circ(\Delta\otimes V_\lambda^\star) av Γ j ∘ ( Δ ⊗ V λ ⋆ ) factors through uniquely a map
Δ λ , j : ( Z [ ∣ N j δ ∣ ] ⊗ V λ ⋆ ) Γ j ⟶ Fun ( ∣ N j δ ∣ , V λ ⋆ ) Γ j . \Delta_{\lambda,j}:\left(\mathbb{Z}[\lvert N_j^{\delta}\rvert]\otimes V_\lambda^\star\right)_{\Gamma_j}\longrightarrow\operatorname{Fun}(\lvert N_j^{\delta}\rvert,V_\lambda^\star)^{\Gamma_j}. Δ λ , j : ( Z [∣ N j δ ∣] ⊗ V λ ⋆ ) Γ j ⟶ Fun (∣ N j δ ∣ , V λ ⋆ ) Γ j .
We also have the map
∇ j : Div ( N δ × { h j } ) ⟶ Func c ( ∣ N j δ ∣ , Z ) \nabla_j:\operatorname{Div}(\mathfrak{N}^{\delta}\times\{h_j\})\longrightarrow\operatorname{Func}_c(\lvert N_j^{\delta}\rvert,\mathbb{Z}) ∇ j : Div ( N δ × { h j }) ⟶ Func c (∣ N j δ ∣ , Z )
given by the intersection in N δ × { h j } \mathfrak{N}^{\delta}\times\{h_j\} N δ × { h j } .
Lemma 5.20. For every L L L -valued function ϕ \phi ϕ on p v a Λ v / p v b Λ v p_v^a\Lambda_v/p_v^b\Lambda_v p v a Λ v / p v b Λ v and every element x ⋆ ∈ V ⋆ x^\star\in V^\star x ⋆ ∈ V ⋆ , there exists an element
F λ ( x ⋆ , ϕ ) j ∈ ( L [ ∣ N j δ ∣ ] ⊗ V λ ⋆ ) Γ j \mathcal{F}^{\lambda}(x^\star,\phi)_j\in\left(L[\lvert N_j^{\delta}\rvert]\otimes V_\lambda^\star\right)_{\Gamma_j} F λ ( x ⋆ , ϕ ) j ∈ ( L [∣ N j δ ∣] ⊗ V λ ⋆ ) Γ j such that
Δ λ , j F λ ( x ⋆ , ϕ ) j − av Γ j ( C ( x ⋆ , ϕ ) j ⊗ μ λ ( x ⋆ ) ) \Delta_{\lambda,j}\mathcal{F}^{\lambda}(x^\star,\phi)_j-\operatorname{av}^{\Gamma_j}\left(C(x^\star,\phi)_j\otimes\mu^\lambda(x^\star)\right) Δ λ , j F λ ( x ⋆ , ϕ ) j − av Γ j ( C ( x ⋆ , ϕ ) j ⊗ μ λ ( x ⋆ ) ) is a constant function on ∣ N j δ ∣ \lvert N_j^{\delta}\rvert ∣ N j δ ∣ (which must be the zero function when λ ≠ 0 \lambda\ne0 λ = 0 ). Here,
C ( x ⋆ , ϕ ) j : = ∇ j ( D v a ( x ⋆ , ϕ ) × { h j } ) ∈ Func c ( ∣ N j δ ∣ , L ) C(x^\star,\phi)_j:=\nabla_j\left(\mathcal{D}_v^a(x^\star,\phi)\times\{h_j\}\right)\in\operatorname{Func}_c(\lvert N_j^{\delta}\rvert,L) C ( x ⋆ , ϕ ) j := ∇ j ( D v a ( x ⋆ , ϕ ) × { h j } ) ∈ Func c (∣ N j δ ∣ , L ) , where D v a ( x ⋆ , ϕ ) × { h j } ∈ Div ( N δ × { h j } ) ⊗ L \mathcal{D}_v^a(x^\star,\phi)\times\{h_j\}\in\operatorname{Div}(\mathfrak{N}^{\delta}\times\{h_j\})\otimes L D v a ( x ⋆ , ϕ ) × { h j } ∈ Div ( N δ × { h j }) ⊗ L is from Definition 5.13(3);
μ λ \mu^\lambda μ λ is defined in Construction 2.8(3) (with R = Q R=\mathbb{Q} R = Q ).
Proof. Choose an auxiliary rational prime ℓ ≠ p \ell\ne p ℓ = p . By Remark 5.23 below, the map Δ λ , j ⊗ Q ℓ \Delta_{\lambda,j}\otimes\mathbb{Q}_\ell Δ λ , j ⊗ Q ℓ is identified with the ℓ \ell ℓ -adic Gysin map (59) . Note that the cokernel of (59) , which is same as the image of the natural map
H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 ) ) ⟶ H e ˊ t 2 ( X L , j [ 1 / p ] , V ℓ , L , λ ( 1 ) ) , H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j},V_{\ell,L,\lambda}(1))\longrightarrow H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j}[1/p],V_{\ell,L,\lambda}(1)), H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 )) ⟶ H e ˊ t 2 ( X L , j [ 1/ p ] , V ℓ , L , λ ( 1 )) , is the Gal ( E ~ u / E u ) \operatorname{Gal}(\widetilde{E}_u/E_u) Gal ( E u / E u ) -finite subspace of H e ˊ t 2 ( X L , j [ 1 / p ] , V ℓ , L , λ ( 1 ) ) H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j}[1/p],V_{\ell,L,\lambda}(1)) H e ˊ t 2 ( X L , j [ 1/ p ] , V ℓ , L , λ ( 1 )) . There are two cases.
First suppose that λ ≠ 0 \lambda\ne0 λ = 0 . Then the Gal ( E ~ u / E u ) \operatorname{Gal}(\widetilde{E}_u/E_u) Gal ( E u / E u ) -finite subspace of H e ˊ t 2 ( X L , j [ 1 / p ] , V ℓ , L , λ ( 1 ) ) H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j}[1/p],V_{\ell,L,\lambda}(1)) H e ˊ t 2 ( X L , j [ 1/ p ] , V ℓ , L , λ ( 1 )) vanishes, which implies that Δ λ , j ⊗ Q ℓ \Delta_{\lambda,j}\otimes\mathbb{Q}_\ell Δ λ , j ⊗ Q ℓ hence Δ λ , j \Delta_{\lambda,j} Δ λ , j are surjective. The lemma then follows.
Second suppose that λ = 0 \lambda=0 λ = 0 . Then the Gal ( E ~ u / E u ) \operatorname{Gal}(\widetilde{E}_u/E_u) Gal ( E u / E u ) -finite subspace of H e ˊ t 2 ( X L , j [ 1 / p ] , V ℓ , L , λ ( 1 ) ) H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j}[1/p],V_{\ell,L,\lambda}(1)) H e ˊ t 2 ( X L , j [ 1/ p ] , V ℓ , L , λ ( 1 )) is generated by the image of constant functions on ∣ N j δ ∣ \lvert N_j^{\delta}\rvert ∣ N j δ ∣ under the identification H e ˊ t 2 ( X L , j , Q ℓ ( 1 ) ) ≃ Fun ( ∣ N j δ ∣ , Q ℓ ) Γ j H^2_{\mathrm{\acute{e}t}}(\mathcal{X}_{L,j},\mathbb{Q}_\ell(1))\simeq\operatorname{Fun}(\lvert N_j^{\delta}\rvert,\mathbb{Q}_\ell)^{\Gamma_j} H e ˊ t 2 ( X L , j , Q ℓ ( 1 )) ≃ Fun (∣ N j δ ∣ , Q ℓ ) Γ j from Remark 5.23 below. The lemma then follows. □ \square □
Denote by N δ N^\delta N δ the normalization of the induced reduced subscheme of N δ \mathfrak{N}^{\delta} N δ , and by ∣ N δ ∣ \lvert N^\delta\rvert ∣ N δ ∣ the set of its irreducible components. For every 1 ≤ j ≤ n 1\le j\le n 1 ≤ j ≤ n , we fix an element
F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ∈ L [ ∣ N δ ∣ ] ⊗ V λ ⋆ \mathcal{F}^{\lambda}(x_2^\star,\phi_{2,v}^{\infty})_j\in L[\lvert N^\delta\rvert]\otimes V_\lambda^\star F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ∈ L [∣ N δ ∣] ⊗ V λ ⋆
such that F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } \mathcal{F}^{\lambda}(x_2^\star,\phi_{2,v}^{\infty})_j\times\{h_j\} F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } lifts F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j \mathcal{F}^{\lambda}(x_2^\star,\phi_{2,v}^{\infty})_j F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j from Lemma 5.20 .
Lemma 5.21. For every 1 ≤ j ≤ n 1 \le j \le n 1 ≤ j ≤ n , there exists a unique V λ ⊗ E λ L V_\lambda\otimes_{E_\lambda} L V λ ⊗ E λ L -valued Schwartz function ϕ v , j [ a , b ] \phi^{[a,b]}_{v,j} ϕ v , j [ a , b ] on V v ⋆ V_v^\star V v ⋆ such that for every x 1 ⋆ ∈ V v ⋆ ∖ { 0 } x_1^\star\in V_v^\star\setminus\{0\} x 1 ⋆ ∈ V v ⋆ ∖ { 0 } ,
ϕ v , j [ a , b ] ( x 1 ⋆ ) = ( D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) ∙ D a δ ( x 2 ⋆ , ϕ 2 , v ∞ ) ) ⋅ μ λ ( x 2 ⋆ ) − D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) ⋅ F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j . \phi^{[a,b]}_{v,j}(x_1^\star)=\left(\mathcal{D}^{\delta}_{a}(x_1^\star,\phi^\infty_{1,v})\bullet\mathcal{D}^{\delta}_{a}(x_2^\star,\phi^\infty_{2,v})\right)\cdot\mu^\lambda(x_2^\star)-\mathcal{D}^{\delta}_{a}(x_1^\star,\phi^\infty_{1,v})\cdot\mathcal{F}^\lambda(x_2^\star,\phi^\infty_{2,v})_j. ϕ v , j [ a , b ] ( x 1 ⋆ ) = ( D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) ∙ D a δ ( x 2 ⋆ , ϕ 2 , v ∞ ) ) ⋅ μ λ ( x 2 ⋆ ) − D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) ⋅ F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j . in which ∙ \bullet ∙ denotes the intersection pairing on N δ \mathcal{N}^\delta N δ as an L L L -hermitian form.
Proof. The continuity of ϕ v , j [ a , b ] \phi^{[a,b]}_{v,j} ϕ v , j [ a , b ] on V v ⋆ ∖ { 0 } V_v^\star\setminus\{0\} V v ⋆ ∖ { 0 } follows from Lemma 5.14. Since ϕ 1 , v ∞ \phi^\infty_{1,v} ϕ 1 , v ∞ is regularly supported, D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) = 0 \mathcal{D}^{\delta}_{a}(x_1^\star,\phi^\infty_{1,v})=0 D a δ ( x 1 ⋆ , ϕ 1 , v ∞ ) = 0 unless x 1 ⋆ ∈ D ( 2 b , 2 a ] : = { x ⋆ ∈ V v ⋆ ∣ ( x ⋆ , x ⋆ ) ∈ p v 2 a ∖ p v 2 b } x_1^\star\in D_{(2b,2a]}:=\{x^\star\in V_v^\star\mid(x^\star,x^\star)\in\mathfrak{p}_v^{2a}\setminus\mathfrak{p}_v^{2b}\} x 1 ⋆ ∈ D ( 2 b , 2 a ] := { x ⋆ ∈ V v ⋆ ∣ ( x ⋆ , x ⋆ ) ∈ p v 2 a ∖ p v 2 b } . Thus, ϕ v , j [ a , b ] \phi^{[a,b]}_{v,j} ϕ v , j [ a , b ] extends uniquely to a V λ ⊗ E λ L V_\lambda\otimes_{E_\lambda}L V λ ⊗ E λ L -valued continuous function on V v ⋆ V_v^\star V v ⋆ . It remains to show that ϕ v , j [ a , b ] \phi^{[a,b]}_{v,j} ϕ v , j [ a , b ] is compactly supported. Indeed, this follows from the fact that for every t ∈ F v × t\in F_v^\times t ∈ F v × , the subset
{ x ⋆ ∈ ( V ⋆ ) t ∣ D a 0 ( x ⋆ , 0 ) ∩ Σ j ≠ ∅ } \{x^\star\in(V^\star)_t\mid\mathcal{D}^{0}_{a}(x^\star,0)\cap\Sigma_j\ne\varnothing\} { x ⋆ ∈ ( V ⋆ ) t ∣ D a 0 ( x ⋆ , 0 ) ∩ Σ j = ∅ } is compact, where Σ j \Sigma_j Σ j is the image of the support of F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j \mathcal{F}^\lambda(x_2^\star,\phi^\infty_{2,v})_j F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j in N 0 \mathcal{N}^0 N 0 . The lemma is proved. □
For 1 ≤ j ≤ n 1\le j\le n 1 ≤ j ≤ n , we also define a function ϕ j ∞ , v \phi^{\infty,v}_j ϕ j ∞ , v on V ∞ , v V^{\infty,v} V ∞ , v by the formula
ϕ j ∞ , v ( x ) = s 1 ϕ 1 ∞ ( h j − 1 x ) ⋅ s 2 ϕ 2 ∞ ( h j − 1 x 2 ⋆ ) , x ∈ V ∞ , v , \phi^{\infty,v}_j(x)=s_1\phi^\infty_1(h_j^{-1}x)\cdot s_2\phi^\infty_2(h_j^{-1}x_2^\star),\qquad x\in V^{\infty,v}, ϕ j ∞ , v ( x ) = s 1 ϕ 1 ∞ ( h j − 1 x ) ⋅ s 2 ϕ 2 ∞ ( h j − 1 x 2 ⋆ ) , x ∈ V ∞ , v ,
which is an element in S ( V ∞ , v , L ) \mathcal{S}(V^{\infty,v},L) S ( V ∞ , v , L ) .
Proposition 5.22. Suppose that we are in the situation of Proposition 5.9 (2) together with Assumption 5.11. If we put
ϕ t 2 ∞ : = vol ♮ ( L ) ⋅ ∑ j = 1 n ϕ j ∞ , v ⊗ ϕ v , j [ a , b ] , \phi^{\infty}_{t_2}:=\operatorname{vol}^{\natural}(L)\cdot\sum_{j=1}^{n}\phi^{\infty,v}_j\otimes\phi^{[a,b]}_{v,j}, ϕ t 2 ∞ := vol ♮ ( L ) ⋅ j = 1 ∑ n ϕ j ∞ , v ⊗ ϕ v , j [ a , b ] , as a V λ ⊗ E λ L V_\lambda\otimes_{E_\lambda}L V λ ⊗ E λ L -valued Schwartz function on V ⋆ ⊗ F A F ∞ V^\star\otimes_F\mathbb{A}^{\infty}_F V ⋆ ⊗ F A F ∞ , then for every t 1 ∈ ( F × ) + t_1\in(F^\times)^+ t 1 ∈ ( F × ) + ,
vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = ∑ x 1 ⋆ ∈ V ( v ) t 1 ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ⋆ ) ] ⋅ ϕ V , ∞ ∞ ( x 1 ⋆ ) ) \operatorname{vol}^{\natural}(L)\cdot\left\langle s_1^\star Z^{\lambda}_{t_1}(\phi^\infty_1)_L(g_1),s_2^\star Z^{\lambda}_{t_2}(\phi^\infty_2)_L\right\rangle^{\ell}_{\lambda,L,Z,u}
=\sum_{x_1^\star\in V(v)_{t_1}}\omega_{1,\infty}(g_1)\left(P[\phi^{\infty}_{t_2}(x_1^\star)]\cdot\phi^\infty_{V,\infty}(x_1^\star)\right) vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = x 1 ⋆ ∈ V ( v ) t 1 ∑ ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ⋆ )] ⋅ ϕ V , ∞ ∞ ( x 1 ⋆ ) ) holds (in C ⊗ Q ℓ \mathbb{C}\otimes\mathbb{Q}_\ell C ⊗ Q ℓ ) for every element g 1 ∈ G 1 ( F ∞ ) g_1\in G_1(F_\infty) g 1 ∈ G 1 ( F ∞ ) and every prime ℓ ≠ p \ell\ne p ℓ = p .
Take a rational prime ℓ ≠ p \ell\ne p ℓ = p . Note that the E ℓ λ E_\ell^\lambda E ℓ λ -linear local system V ℓ , L , λ V_{\ell,L,\lambda} V ℓ , L , λ over X L X_L X L has a uniformization:
V ℓ , L , λ ∣ X L ′ = H ⋆ ( F ) \ ( N δ × H ( A F ∞ , v ) / L ′ ∨ × V ℓ , λ ) (58) V_{\ell,L,\lambda}|_{X'_L}=H^\star(F)\backslash\left(\mathcal{N}^\delta\times H(\mathbb{A}^{\infty,v}_F)/L'^\vee\times V_{\ell,\lambda}\right) \tag*{(58)} V ℓ , L , λ ∣ X L ′ = H ⋆ ( F ) \ ( N δ × H ( A F ∞ , v ) / L ′ ∨ × V ℓ , λ ) (58)
compatible with (57) , in which H ⋆ ( F ) H^\star(F) H ⋆ ( F ) acts on V ℓ , λ : = ( V ℓ ) λ V_{\ell,\lambda}:=(V_\ell)_\lambda V ℓ , λ := ( V ℓ ) λ via the natural map H ⋆ ( F ) → H ( F ℓ ) H^\star(F)\to H(F_\ell) H ⋆ ( F ) → H ( F ℓ ) .
Remark 5.23. By (58) , the pullback map
H e ˊ t 2 ( L L , j , V ℓ , L , λ ( 1 ) ) ⟶ ∏ C ∈ ∣ N j δ ∣ H e ˊ t 2 ( C , V ℓ , L , λ ( 1 ) ) = Fun ( ∣ N j δ ∣ , V ℓ , L , λ ) H^2_{\mathrm{\acute et}}(\mathcal{L}_{L,j},V_{\ell,L,\lambda}(1))\longrightarrow\prod_{C\in|N_j^\delta|}H^2_{\mathrm{\acute et}}(C,V_{\ell,L,\lambda}(1))=\operatorname{Fun}(|N_j^\delta|,V_{\ell,L,\lambda}) H e ˊ t 2 ( L L , j , V ℓ , L , λ ( 1 )) ⟶ C ∈ ∣ N j δ ∣ ∏ H e ˊ t 2 ( C , V ℓ , L , λ ( 1 )) = Fun ( ∣ N j δ ∣ , V ℓ , L , λ )
induces an isomorphism
H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 ) ) ≃ Fun ( ∣ N j δ ∣ , V ℓ , L , λ ) Γ j . H^2_{\mathrm{\acute et}}(X_{L,j},V_{\ell,L,\lambda}(1))\simeq\operatorname{Fun}(|N_j^\delta|,V_{\ell,L,\lambda})^{\Gamma_j}. H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 )) ≃ Fun ( ∣ N j δ ∣ , V ℓ , L , λ ) Γ j .
On the other hand, the Gysin map
Z [ ∣ N j δ ∣ ] ⊗ V ℓ , L , λ = ⨁ C ∈ ∣ N j δ ∣ H e ˊ t 0 ( C , V ℓ , L , λ ) ⟶ H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 ) ) \mathbb{Z}[|N_j^\delta|]\otimes V_{\ell,L,\lambda}=\bigoplus_{C\in|N_j^\delta|}H^0_{\mathrm{\acute et}}(C,V_{\ell,L,\lambda})\longrightarrow H^2_{\mathrm{\acute et},Y_{L,j}}(X_{L,j},V_{\ell,L,\lambda}(1)) Z [ ∣ N j δ ∣ ] ⊗ V ℓ , L , λ = C ∈ ∣ N j δ ∣ ⨁ H e ˊ t 0 ( C , V ℓ , L , λ ) ⟶ H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 ))
induces an isomorphism
( Z [ ∣ N j δ ∣ ] ⊗ V ℓ λ ) Γ j ≃ H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 ) ) . \left(\mathbb{Z}[|N_j^\delta|]\otimes V_{\ell}^\lambda\right)_{\Gamma_j}\simeq H^2_{\mathrm{\acute et},Y_{L,j}}(X_{L,j},V_{\ell,L,\lambda}(1)). ( Z [ ∣ N j δ ∣ ] ⊗ V ℓ λ ) Γ j ≃ H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 )) .
Then it is clear that under the above two isomorphisms, the Gysin map
H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 ) ) ⟶ H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 ) ) (59) H^2_{\mathrm{\acute et},Y_{L,j}}(X_{L,j},V_{\ell,L,\lambda}(1))\longrightarrow H^2_{\mathrm{\acute et}}(X_{L,j},V_{\ell,L,\lambda}(1)) \tag*{(59)} H e ˊ t , Y L , j 2 ( X L , j , V ℓ , L , λ ( 1 )) ⟶ H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 )) (59)
coincides with Δ λ , j ⊗ Q ℓ \Delta_{\lambda,j}\otimes\mathbb{Q}_\ell Δ λ , j ⊗ Q ℓ .
The construction below is parallel to Construction 5.18, except that we use the naïve horizontal divisor Z a δ ( x v ⋆ , ϕ v ) \mathcal{Z}^{\delta}_{a}(x_v^\star,\phi_v) Z a δ ( x v ⋆ , ϕ v ) .
Construction 5.24. Consider a Schwartz function ϕ ∞ ∈ S ( V ∞ , L ) \phi^\infty\in\mathcal{S}(V^\infty,\mathbb{L}) ϕ ∞ ∈ S ( V ∞ , L ) of the form ϕ v ⊗ ϕ ∞ , v \phi_v \otimes\phi^{\infty,v} ϕ v ⊗ ϕ ∞ , v in which ϕ v \phi_v ϕ v is a regularly supported L \mathbb{L} L -valued function on p v a Λ v / p v b Λ v p_v^a\Lambda_v/p_v^b\Lambda_v p v a Λ v / p v b Λ v .
(1) For x v ⋆ ∈ V v ⋆ ∖ { 0 } x_v^\star\in V_v^\star\setminus\{0\} x v ⋆ ∈ V v ⋆ ∖ { 0 } , put
Z ( x v ⋆ , ϕ v ) h : = Z a δ ( x v ⋆ , ϕ v ) × { h } ∈ Div ( N δ × H ( A F ∞ , v ) / L v ) ⊗ L \mathcal{Z}(x_v^\star,\phi_v)_h := \mathcal{Z}^{\delta}_a(x_v^\star,\phi_v)\times\{h\}\in\operatorname{Div}(\mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v)\otimes\mathbb{L} Z ( x v ⋆ , ϕ v ) h := Z a δ ( x v ⋆ , ϕ v ) × { h } ∈ Div ( N δ × H ( A F ∞ , v ) / L v ) ⊗ L (Definition 5.13(3)), and denote by
C ( x v ⋆ , ϕ v ) h ∈ H e ˊ t , ∣ Z ( x v ⋆ , ϕ v ) h ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 ) ) \mathcal{C}(x_v^\star,\phi_v)_h\in H^2_{\mathrm{\acute et},|\mathcal{Z}(x_v^\star,\phi_v)_h|}(\mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v,\mathbb{L}_{\ell}(1)) C ( x v ⋆ , ϕ v ) h ∈ H e ˊ t , ∣ Z ( x v ⋆ , ϕ v ) h ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 )) its refined ℓ \ell ℓ -adic cycle class.
(2) For every t ∈ ( F × ) + t\in(F^\times)^+ t ∈ ( F × ) + , define Z t ( ϕ ∞ ) L \mathcal{Z}_t(\phi^\infty)_{\mathbb{L}} Z t ( ϕ ∞ ) L to be the unique element in Div ( X L ⊗ R 0 R ) ⊗ L \operatorname{Div}(\mathcal{X}_{\mathbb{L}}\otimes_{R_0}R)\otimes\mathbb{L} Div ( X L ⊗ R 0 R ) ⊗ L supported on the basic locus whose pullback to N δ × H ( A F ∞ , v ) / L v \mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v N δ × H ( A F ∞ , v ) / L v is given by
Z t ( ϕ ∞ ) L : = ∑ x ⋆ ∈ V t ⋆ ∑ h ∈ H ( A F ∞ , v ) / L v ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ Z ( x v ⋆ , ϕ v ) h . \mathcal{Z}_t(\phi^\infty)_{\mathbb{L}}:=\sum_{x^\star\in V_t^\star}\sum_{h\in H(\mathbb{A}^{\infty,v}_F)/L^v}\phi^{\infty,v}(h^{-1}x^\star)\cdot\mathcal{Z}(x_v^\star,\phi_v)_h. Z t ( ϕ ∞ ) L := x ⋆ ∈ V t ⋆ ∑ h ∈ H ( A F ∞ , v ) / L v ∑ ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ Z ( x v ⋆ , ϕ v ) h . (3) For every t ∈ ( F × ) + t\in(F^\times)^+ t ∈ ( F × ) + , denote by
C t ( ϕ ∞ ) L ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) ⊗ E λ L \mathcal{C}_t(\phi^\infty)_{\mathbb{L}}\in H^2_{\mathrm{\acute et},|\mathcal{Z}_t(\phi^\infty)_{\mathbb{L}}|}(\mathcal{X}_{\mathbb{L}}\otimes_{R_0}R,V_{\ell,\mathbb{L},\lambda}(1))\otimes_{E_\lambda}\mathbb{L} C t ( ϕ ∞ ) L ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 )) ⊗ E λ L the unique element whose restriction to N δ × H ( A F ∞ , v ) / L v \mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v N δ × H ( A F ∞ , v ) / L v is given by the H ⋆ ( F ) H^\star(F) H ⋆ ( F ) -invariant class
∑ x ⋆ ∈ V t ⋆ ( ∑ h ∈ H ( A F ∞ , v ) / L v ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ C ( x v ⋆ , ϕ v ) h ) ⊗ μ λ ( x ⋆ ) \sum_{x^\star\in V_t^\star}\left(\sum_{h\in H(\mathbb{A}^{\infty,v}_F)/L^v}\phi^{\infty,v}(h^{-1}x^\star)\cdot\mathcal{C}(x_v^\star,\phi_v)_h\right)\otimes\mu^\lambda(x^\star) x ⋆ ∈ V t ⋆ ∑ h ∈ H ( A F ∞ , v ) / L v ∑ ϕ ∞ , v ( h − 1 x ⋆ ) ⋅ C ( x v ⋆ , ϕ v ) h ⊗ μ λ ( x ⋆ ) ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 ) ) ⊗ E λ V λ ⋆ = H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 ) ) ⊗ E λ V ℓ , λ , \in H^2_{\mathrm{\acute et},|\mathcal{Z}_t(\phi^\infty)_{\mathbb{L}}|}(\mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v,\mathbb{L}_{\ell}(1))\otimes_{E_\lambda}V_\lambda^\star
=H^2_{\mathrm{\acute et},|\mathcal{Z}_t(\phi^\infty)_{\mathbb{L}}|}(\mathcal{N}^{\delta}\times H(\mathbb{A}^{\infty,v}_F)/L^v,\mathbb{L}_{\ell}(1))\otimes_{E_\lambda}V_{\ell,\lambda}, ∈ H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 )) ⊗ E λ V λ ⋆ = H e ˊ t , ∣ Z t ( ϕ ∞ ) L ∣ 2 ( N δ × H ( A F ∞ , v ) / L v , L ℓ ( 1 )) ⊗ E λ V ℓ , λ , where μ λ \mu^\lambda μ λ is defined in Construction 2.8(3) (with R = Q R=\mathbb{Q} R = Q ).
For every 1 ≤ j ≤ n 1\leq j\leq n 1 ≤ j ≤ n , denote by
C λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j ∈ H e ˊ t , π j ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( X j , V ℓ , λ ( 1 ) ) ⊗ E λ L \mathcal{C}^{\lambda}(x_{2,v}^\star,\phi_{2,v}^{\infty})_j\in H^2_{\mathrm{\acute et},\pi_j|\mathcal{Z}(x_{2,v}^\star,\phi_{2,v}^{\infty})_{h_j}|}(\mathcal{X}_j,V_{\ell,\lambda}(1))\otimes_{E_\lambda}\mathbb{L} C λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j ∈ H e ˊ t , π j ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( X j , V ℓ , λ ( 1 )) ⊗ E λ L the pushforward of the class
C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) \mathcal{C}(x_{2,v}^\star,\phi_{2,v}^{\infty})_{h_j}\otimes\mu^\lambda(x_{2}^\star) C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) ∈ H e ˊ t , ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( N δ × { h j } , L ℓ ( 1 ) ) ⊗ E λ V λ ⋆ = H e ˊ t , ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( N δ × { h j } , L ℓ ( 1 ) ) ⊗ E λ V ℓ , λ \in H^2_{\mathrm{\acute et},|\mathcal{Z}(x_{2,v}^\star,\phi_{2,v}^{\infty})_{h_j}|}(\mathcal{N}^{\delta}\times\{h_j\},\mathbb{L}_{\ell}(1))\otimes_{E_\lambda}V_\lambda^\star
=H^2_{\mathrm{\acute et},|\mathcal{Z}(x_{2,v}^\star,\phi_{2,v}^{\infty})_{h_j}|}(\mathcal{N}^{\delta}\times\{h_j\},\mathbb{L}_{\ell}(1))\otimes_{E_\lambda}V_{\ell,\lambda} ∈ H e ˊ t , ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( N δ × { h j } , L ℓ ( 1 )) ⊗ E λ V λ ⋆ = H e ˊ t , ∣ Z ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ∣ 2 ( N δ × { h j } , L ℓ ( 1 )) ⊗ E λ V ℓ , λ (Construction 5.24(1)) along π j \pi_j π j .
Lemma 5.25. The natural image of
∑ j = 1 n s 2 ϕ 2 , v ∞ ( h j − 1 x 2 ⋆ ) ( C λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j ) (60) \sum_{j=1}^{n}s_{2}^{\phi_{2,v}^{\infty}}(h_j^{-1}x_2^\star)\left(\mathcal{C}^{\lambda}(x_{2,v}^\star,\phi_{2,v}^{\infty})_j-\mathcal{F}^{\lambda}(x_{2,v}^\star,\phi_{2,v}^{\infty})_j\right) \tag*{(60)} j = 1 ∑ n s 2 ϕ 2 , v ∞ ( h j − 1 x 2 ⋆ ) ( C λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j ) (60) in H e ˊ t 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) ⊗ E λ L H^2_{\mathrm{\acute et}}(\mathcal{X}_{\mathbb{L}}\otimes_{R_0}R,V_{\ell,\mathbb{L},\lambda}(1))\otimes_{E_\lambda}\mathbb{L} H e ˊ t 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 )) ⊗ E λ L vanishes. Here, for the term F λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j \mathcal{F}^{\lambda}(x_{2,v}^\star,\phi_{2,v}^{\infty})_j F λ ( x 2 , v ⋆ , ϕ 2 , v ∞ ) j from Lemma 5.20, we have used the map ( Z [ ∣ N j δ ∣ ] ⊗ V λ ⋆ ) Γ j ⟶ H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 ) ) (\mathbb{Z}[|N_j^\delta|]\otimes V_\lambda^\star)_{\Gamma_j}\longrightarrow H^2_{\mathrm{\acute et}}(\mathcal{X}_{\mathbb{L},j},V_{\ell,\mathbb{L},\lambda}(1)) ( Z [ ∣ N j δ ∣ ] ⊗ V λ ⋆ ) Γ j ⟶ H e ˊ t 2 ( X L , j , V ℓ , L , λ ( 1 )) from Remark 5.23.
Proof. When λ ≠ 0 \lambda\ne0 λ = 0 , this follows immediately from Lemma 5.20 and Remark 5.23. When λ = 0 \lambda=0 λ = 0 , the image of (5.7) in H e ˊ t 2 ( X L , u ⊗ R 0 R , L ℓ ( 1 ) ) H^2_{\mathrm{\acute et}}(\mathcal{X}_{\mathbb{L},u}\otimes_{R_0}R,\mathbb{L}_{\ell}(1)) H e ˊ t 2 ( X L , u ⊗ R 0 R , L ℓ ( 1 )) vanishes, which implies that the total degree of (5.7) as an L ℓ \mathbb{L}_{\ell} L ℓ -valued function on the set of vertical irreducible components on each connected component of X L ⊗ R 0 R \mathcal{X}_{\mathbb{L}}\otimes_{R_0}R X L ⊗ R 0 R is zero. Combining this with Lemma 5.20, the lemma follows. □
Proof of Proposition 5.22. Similar to the proof of Proposition 5.2 , we may assume g 1 = 1 2 g_1=1_2 g 1 = 1 2 .
Put
C t 2 λ ( s 2 ϕ 2 ∞ ) L : = ∑ j = 1 n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) . (61) C_{t_2}^{\lambda}(s_2\phi_2^\infty)_L:=\sum_{j=1}^{n}s_2\phi_2^{\infty,v}(h_j^{-1}x_2^\star)\cdot\left(C^\lambda(x_2^\star,\phi_{2,v}^\infty)_j-F^\lambda(x_2^\star,\phi_{2,v}^\infty)_j\right). \tag*{(61)} C t 2 λ ( s 2 ϕ 2 ∞ ) L := j = 1 ∑ n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) . (61)
regarded as an element in H e ˊ t , ( Z 2 , u ∪ Y L ) ⊗ R 0 R 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) H_{\mathrm{\acute et},(Z_{2,u}\cup Y_L)\otimes_{R_0}R}^{2}(X_L\otimes_{R_0}R,V_{\ell,L,\lambda}(1)) H e ˊ t , ( Z 2 , u ∪ Y L ) ⊗ R 0 R 2 ( X L ⊗ R 0 R , V ℓ , L , λ ( 1 )) , so that, by construction, its natural image in H e ˊ t , Z 2 , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) H_{\mathrm{\acute et},Z_{2,u}\otimes_{R_0}R}^{2}(X_{L,u}\otimes_{R_0}R,V_{\ell,L,\lambda}(1)) H e ˊ t , Z 2 , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 )) coincides with that of s 2 ⋆ Z t 2 1 ( ϕ 2 ∞ ) L s_2^\star Z_{t_2}^{1}(\phi_2^\infty)_L s 2 ⋆ Z t 2 1 ( ϕ 2 ∞ ) L . On the other hand, the natural image of the class C t 1 λ ( s 1 ϕ 1 ∞ ) L C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L C t 1 λ ( s 1 ϕ 1 ∞ ) L from Construction 5.24(3) in H e ˊ t , Z 1 , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 ) ) H_{\mathrm{\acute et},Z_{1,u}\otimes_{R_0}R}^{2}(X_{L,u}\otimes_{R_0}R,V_{\ell,L,\lambda}(1)) H e ˊ t , Z 1 , u ⊗ R 0 R 2 ( X L , u ⊗ R 0 R , V ℓ , L , λ ( 1 )) coincides with that of s 1 ⋆ Z t 1 1 ( ϕ 1 ∞ ) L s_1^\star Z_{t_1}^{1}(\phi_1^\infty)_L s 1 ⋆ Z t 1 1 ( ϕ 1 ∞ ) L .
Then by Lemma 4.7 (or rather its analogue after base change to R R R ) and Lemma 5.25, we have
⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L . \left\langle s_1^\star Z_{t_1}^{\lambda}(\phi_1^\infty)_L,s_2^\star Z_{t_2}^{\lambda}(\phi_2^\infty)_L\right\rangle_{\lambda,L,Z,u}^{\ell}
=C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}C_{t_2}^{\lambda}(s_2\phi_2^\infty)_L. ⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L .
By (61) ,
C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L = ∑ j = 1 n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) . C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}C_{t_2}^{\lambda}(s_2\phi_2^\infty)_L
=\sum_{j=1}^{n}s_2\phi_2^{\infty,v}(h_j^{-1}x_2^\star)\cdot C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}\left(C^\lambda(x_2^\star,\phi_{2,v}^\infty)_j-F^\lambda(x_2^\star,\phi_{2,v}^\infty)_j\right). C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L = j = 1 ∑ n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) .
Note that
C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j = ( π j ) ∗ ( C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) ) , C^\lambda(x_2^\star,\phi_{2,v}^\infty)_j=(\pi_j)_*\left(\mathcal{C}(x_{2,v}^\star,\phi_{2,v}^\infty)_{h_j}\otimes\mu^\lambda(x_2^\star)\right), C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j = ( π j ) ∗ ( C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) ) ,
and
F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j = ( π j ) ∗ ( F ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } ) . F^\lambda(x_2^\star,\phi_{2,v}^\infty)_j=(\pi_j)_*\left(\mathcal{F}(x_2^\star,\phi_{2,v}^\infty)_j\times\{h_j\}\right). F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j = ( π j ) ∗ ( F ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } ) .
It follows by the projection formula that
C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) = ( π j ) ∗ C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( ( C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) ) − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } ) , (62) \begin{aligned}
C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}\left(C^\lambda(x_2^\star,\phi_{2,v}^\infty)_j-F^\lambda(x_2^\star,\phi_{2,v}^\infty)_j\right) \\
&=(\pi_j)_*C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}\left(\left(\mathcal{C}(x_{2,v}^\star,\phi_{2,v}^\infty)_{h_j}\otimes\mu^\lambda(x_2^\star)\right)-\mathcal{F}^\lambda(x_2^\star,\phi_{2,v}^\infty)_j\times\{h_j\}\right), \tag*{(62)}
\end{aligned} C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( C λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j ) = ( π j ) ∗ C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ ( ( C ( x 2 , v ⋆ , ϕ 2 , v ∞ ) h j ⊗ μ λ ( x 2 ⋆ ) ) − F λ ( x 2 ⋆ , ϕ 2 , v ∞ ) j × { h j } ) , (62)
where the second ∙ λ \mathbin{\bullet^\lambda} ∙ λ is the analogous pairing over N δ × { h j } \mathfrak{N}^{\delta}\times\{h_j\} N δ × { h j } . By Construction 5.24(3),
( π j ) ∗ C t 1 λ ( s 1 ϕ 1 ∞ ) L = ∑ x 1 ⋆ ∈ V t ⋆ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( C ( x 1 , v ⋆ , ϕ 1 , v ∞ ) h j ⊗ μ λ ( x 1 ⋆ ) ) . (\pi_j)_*C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L
=\sum_{x_1^\star\in V_{t}^{\star}}s_1\phi_1^{\infty,v}(h_j^{-1}x_1^\star)\cdot\left(\mathcal{C}(x_{1,v}^\star,\phi_{1,v}^\infty)_{h_j}\otimes\mu^\lambda(x_1^\star)\right). ( π j ) ∗ C t 1 λ ( s 1 ϕ 1 ∞ ) L = x 1 ⋆ ∈ V t ⋆ ∑ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( C ( x 1 , v ⋆ , ϕ 1 , v ∞ ) h j ⊗ μ λ ( x 1 ⋆ ) ) .
It follows by Lemma 5.21 that
( 5.9 ) = ∑ x 1 ⋆ ∈ V t ⋆ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ . (5.9)=\sum_{x_1^\star\in V_t^\star}s_1\phi_1^{\infty,v}(h_j^{-1}x_1^\star)\cdot\left(\mu^\lambda(x_1^\star),\phi_{v,j}^{[a,b]}(x_{1,v}^\star)\right)_{V_\lambda^\star}. ( 5.9 ) = x 1 ⋆ ∈ V t ⋆ ∑ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ .
Therefore,
C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L = ∑ j = 1 n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ∑ x 1 ⋆ ∈ V t ⋆ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ = ∑ x 1 ⋆ ∈ V t ⋆ ∑ j = 1 n s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ = vol ♮ ( L ) − 1 ⋅ ∑ x 1 ⋆ ∈ V t ⋆ ( μ λ ( x 1 ⋆ ) , ϕ t 2 ∞ ( x 1 ⋆ ) ) V λ ⋆ . \begin{aligned}
C_{t_1}^{\lambda}(s_1\phi_1^\infty)_L\mathbin{\bullet^\lambda}C_{t_2}^{\lambda}(s_2\phi_2^\infty)_L
=\sum_{j=1}^{n}s_2\phi_2^{\infty,v}(h_j^{-1}x_2^\star)\sum_{x_1^\star\in V_t^\star}s_1\phi_1^{\infty,v}(h_j^{-1}x_1^\star)\cdot\left(\mu^\lambda(x_1^\star),\phi_{v,j}^{[a,b]}(x_{1,v}^\star)\right)_{V_\lambda^\star} \\
&=\sum_{x_1^\star\in V_t^\star}\sum_{j=1}^{n}s_1\phi_1^{\infty,v}(h_j^{-1}x_1^\star)s_2\phi_2^{\infty,v}(h_j^{-1}x_2^\star)\cdot\left(\mu^\lambda(x_1^\star),\phi_{v,j}^{[a,b]}(x_{1,v}^\star)\right)_{V_\lambda^\star} \\
&=\operatorname{vol}^{\natural}(L)^{-1}\cdot\sum_{x_1^\star\in V_t^\star}\left(\mu^\lambda(x_1^\star),\phi_{t_2}^{\infty}(x_1^\star)\right)_{V_\lambda^\star}.
\end{aligned} C t 1 λ ( s 1 ϕ 1 ∞ ) L ∙ λ C t 2 λ ( s 2 ϕ 2 ∞ ) L = j = 1 ∑ n s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) x 1 ⋆ ∈ V t ⋆ ∑ s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ = x 1 ⋆ ∈ V t ⋆ ∑ j = 1 ∑ n s 1 ϕ 1 ∞ , v ( h j − 1 x 1 ⋆ ) s 2 ϕ 2 ∞ , v ( h j − 1 x 2 ⋆ ) ⋅ ( μ λ ( x 1 ⋆ ) , ϕ v , j [ a , b ] ( x 1 , v ⋆ ) ) V λ ⋆ = vol ♮ ( L ) − 1 ⋅ x 1 ⋆ ∈ V t ⋆ ∑ ( μ λ ( x 1 ⋆ ) , ϕ t 2 ∞ ( x 1 ⋆ ) ) V λ ⋆ .
so that
vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = ∑ x 1 ⋆ ∈ V t ⋆ ( μ λ ( x 1 ⋆ ) , ϕ t 2 ∞ ( x 1 ⋆ ) ) V λ ⋆ . \operatorname{vol}^{\natural}(L)\cdot\left\langle s_1^\star Z_{t_1}^{\lambda}(\phi_1^\infty)_L,s_2^\star Z_{t_2}^{\lambda}(\phi_2^\infty)_L\right\rangle_{\lambda,L,Z,u}^{\ell}
=\sum_{x_1^\star\in V_t^\star}\left(\mu^\lambda(x_1^\star),\phi_{t_2}^{\infty}(x_1^\star)\right)_{V_\lambda^\star}. vol ♮ ( L ) ⋅ ⟨ s 1 ⋆ Z t 1 λ ( ϕ 1 ∞ ) L , s 2 ⋆ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z , u ℓ = x 1 ⋆ ∈ V t ⋆ ∑ ( μ λ ( x 1 ⋆ ) , ϕ t 2 ∞ ( x 1 ⋆ ) ) V λ ⋆ .
The proposition is proved. □
Archimedean local heights over unitary moduli curves
In this section, we assume r = 1 r=1 r = 1 and take an element u ∈ Hom ( E , C ) u \in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) . Put v : = v ( u ) ∈ V F ( ∞ ) v:=v(u)\in\mathcal{V}_{F}^{(\infty)} v := v ( u ) ∈ V F ( ∞ ) . As in Remark 3.4, denote by V ( v ) V(v) V ( v ) the unique hermitian space over E / F E/F E / F that has signature ( 1 , 1 ) (1,1) ( 1 , 1 ) at v v v and such that V ( v ) ⊗ F A F ∞ V(v)\otimes_{F}A_{F}^{\infty} V ( v ) ⊗ F A F ∞ is isometric to V ⊗ F A F ∞ V\otimes_{F}A_{F}^{\infty} V ⊗ F A F ∞ . Put H ( v ) : = U ( V ( v ) ) H(v):=U(V(v)) H ( v ) := U ( V ( v )) , and fix an isometry V ( v ) ⊗ F A F ∞ ≃ V ∞ V(v)\otimes_{F}A_{F}^{\infty}\simeq V^{\infty} V ( v ) ⊗ F A F ∞ ≃ V ∞ under which we identify H ( A F ∞ ) H(A_{F}^{\infty}) H ( A F ∞ ) with H ( v ) ( A F ∞ ) H(v)(A_{F}^{\infty}) H ( v ) ( A F ∞ ) .
As before, for every intermediate field E λ ⊆ L ⊆ C \mathbb{E}^{\lambda}\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C , we canonically identify both S λ + ( L ) S_{\lambda^{+}}(L) S λ + ( L ) and S λ + ( L ‾ ) ‾ \overline{S_{\lambda^{+}}(\overline{L})} S λ + ( L ) with L L L as L L L -vector spaces.
Remark 5.26. Since we have identified both S λ + ( C ) S_{\lambda^{+}}(\mathbb{C}) S λ + ( C ) and S λ + ( C ) ‾ \overline{S_{\lambda^{+}}(\mathbb{C})} S λ + ( C ) with C \mathbb{C} C , now A λ + ( G 1 ) \mathcal{A}_{\lambda^{+}}(G_{1}) A λ + ( G 1 ) (Definition 2.14) is simply the subspace of A ( G 1 ) \mathcal{A}(G_{1}) A ( G 1 ) consisting of ( g 1 , ∞ , K 1 , ∞ ) (\mathfrak{g}_{1,\infty},K_{1,\infty}) ( g 1 , ∞ , K 1 , ∞ ) -finite automorphic forms φ \varphi φ on G 1 ( A F ) G_{1}(A_{F}) G 1 ( A F ) satisfying
R k ( φ ) = ρ λ + ( k ) δ ( k ) ⋅ φ R_{k}(\varphi)=\rho_{\lambda^{+}}(k)\delta(k)\cdot\varphi R k ( φ ) = ρ λ + ( k ) δ ( k ) ⋅ φ
for every k ∈ K 1 , ∞ k\in K_{1,\infty} k ∈ K 1 , ∞ , where ρ λ + \rho_{\lambda^{+}} ρ λ + is the character of K 1 , ∞ K_{1,\infty} K 1 , ∞ determined by λ + \lambda^{+} λ + given in Definition 2.14.
Proposition 5.27. Consider
an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(A_{F}^{\infty})) Hk ( H ( A F ∞ )) ,
a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^{\infty}=(\phi_{1}^{\infty},\phi_{2}^{\infty}) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of decomposable Schwartz functions in S ( V ∞ ) L \mathcal{S}(V^{\infty})^{L} S ( V ∞ ) L satisfying that ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v i s (\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}})_{\mathrm{vis}} ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) vis is regularly supported (Definition 2.16) for some v ∈ V F f i n v\in\mathcal{V}_{F}^{\mathrm{fin}} v ∈ V F fin ,
a pair ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) of elements in C [ L v \ H ( A F ∞ , v ) / L v ] \mathbb{C}[L^{v}\backslash H(A_{F}^{\infty,v})/L^{v}] C [ L v \ H ( A F ∞ , v ) / L v ] ,
an open compact subgroup K ∞ K^{\infty} K ∞ of G 1 ( A F ∞ ) G_{1}(A_{F}^{\infty}) G 1 ( A F ∞ ) that fixes ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ .
For every element t 2 ∈ ( F × ) + t_{2}\in(F^{\times})^{+} t 2 ∈ ( F × ) + , there exists an element φ t 2 F M ∈ A λ + ( G 1 ) K ∞ \varphi_{t_{2}}^{\mathrm{FM}}\in\mathcal{A}_{\lambda^{+}}(G_{1})^{K^{\infty}} φ t 2 FM ∈ A λ + ( G 1 ) K ∞ (depending on the data so far) such that
(a) for every t 1 ∈ ( F × ) + t_{1}\in(F^{\times})^{+} t 1 ∈ ( F × ) + (so that Z = ( Z 1 : = Z t 1 [ s 1 ϕ 1 ∞ ] L , Z 2 : = Z t 2 [ s 2 ϕ 2 ∞ ] L ) Z=(Z_{1}:=Z_{t_{1}}[s_{1}\phi_{1}^{\infty}]_{L},Z_{2}:=Z_{t_{2}}[s_{2}\phi_{2}^{\infty}]_{L}) Z = ( Z 1 := Z t 1 [ s 1 ϕ 1 ∞ ] L , Z 2 := Z t 2 [ s 2 ϕ 2 ∞ ] L ) (Definition 3.33) is a disjoint pair),
vol ♮ ( L ) ⋅ ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z = 1 2 G t 1 , t 2 ( ( g 1 , 1 2 ) , Φ λ , ∞ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) v ) + ( φ t 2 F M ) t 1 ( g 1 ) , \operatorname{vol}^{\natural}(L)\cdot\langle s_{1}^{*}Z_{t_{1}}^{\lambda}(\phi_{1}^{\infty})_{L}(g_{1}),s_{2}^{*}Z_{t_{2}}^{\lambda}(\phi_{2}^{\infty})_{L}\rangle_{\lambda,L,Z}
=\frac{1}{2}\mathcal{G}_{t_{1},t_{2}}\left((g_{1},1_{2}),\Phi^{\lambda,\infty}\cdot(s_{1}\phi_{1}^{\infty}\otimes s_{2}\overline{\phi_{2}^{\infty}})_{v}\right)+(\varphi_{t_{2}}^{\mathrm{FM}})_{t_{1}}(g_{1}), vol ♮ ( L ) ⋅ ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z = 2 1 G t 1 , t 2 ( ( g 1 , 1 2 ) , Φ λ , ∞ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) v ) + ( φ t 2 FM ) t 1 ( g 1 ) , (b) for every t 1 ∈ F ∖ ( F × ) + t_{1}\in F\setminus(F^{\times})^{+} t 1 ∈ F ∖ ( F × ) + ,
( φ t 2 F M ) t 1 ( g 1 ) = 0 (\varphi_{t_{2}}^{\mathrm{FM}})_{t_{1}}(g_{1})=0 ( φ t 2 FM ) t 1 ( g 1 ) = 0 hold (in C \mathbb{C} C ) for every g 1 ∈ G 1 ( F ∞ ) g_{1}\in G_{1}(F_{\infty}) g 1 ∈ G 1 ( F ∞ ) , where ( φ t 2 F M ) t 1 (\varphi_{t_{2}}^{\mathrm{FM}})_{t_{1}} ( φ t 2 FM ) t 1 denotes the t 1 t_{1} t 1 -th Fourier coefficient of φ t 2 F M \varphi_{t_{2}}^{\mathrm{FM}} φ t 2 FM (as a function on P 1 ( F ) \ G 1 ( A F ) P_{1}(F)\backslash G_{1}(A_{F}) P 1 ( F ) \ G 1 ( A F ) ).
Proof. For i = 1 , 2 i=1,2 i = 1 , 2 and t i ∈ ( F × ) + t_{i}\in(F^{\times})^{+} t i ∈ ( F × ) + , we have a function F t i λ ( ϕ i ∞ ) F_{t_{i}}^{\lambda}(\phi_{i}^{\infty}) F t i λ ( ϕ i ∞ ) on G 1 ( F ∞ ) G_{1}(F_{\infty}) G 1 ( F ∞ ) valued in A 1 , 1 ( X L u , S [ λ ] ( V ( v ) ∞ ) ) \mathcal{A}^{1,1}(X_{L}^{u},S_{[\lambda]}(V(v)_{\infty})) A 1 , 1 ( X L u , S [ λ ] ( V ( v ) ∞ )) defined by the formula
F t i λ ( ϕ i ∞ ) ( g ) = ρ λ + ( k ) δ ( k ) ∑ ( x , h ) ∈ H ( v ) ( F ) \ V ( v ) t i × H ( A F ∞ ) / L ϕ i ∞ ( h − 1 x ) ⋅ φ λ ( x a , h ) L ⋅ ∏ v ∈ Hom ( F , R ) e 2 π i ( t i b v ) F_{t_{i}}^{\lambda}(\phi_{i}^{\infty})(g)=\rho_{\lambda^{+}}(k)\delta(k)\sum_{(x,h)\in H(v)(F)\backslash V(v)_{t_{i}}\times H(A_{F}^{\infty})/L}\phi_{i}^{\infty}(h^{-1}x)\cdot\varphi_{\lambda}(xa,h)_{L}\cdot\prod_{v\in\operatorname{Hom}(F,\mathbb{R})}e^{2\pi i(t_{i}b_{v})} F t i λ ( ϕ i ∞ ) ( g ) = ρ λ + ( k ) δ ( k ) ( x , h ) ∈ H ( v ) ( F ) \ V ( v ) t i × H ( A F ∞ ) / L ∑ ϕ i ∞ ( h − 1 x ) ⋅ φ λ ( x a , h ) L ⋅ v ∈ Hom ( F , R ) ∏ e 2 π i ( t i b v )
if we write g = m ( a ) n ( b ) k g=m(a)n(b)k g = m ( a ) n ( b ) k with a ∈ E ∞ × a\in E_{\infty}^{\times} a ∈ E ∞ × , b ∈ F ∞ b\in F_{\infty} b ∈ F ∞ , and k ∈ K 1 , ∞ k\in K_{1,\infty} k ∈ K 1 , ∞ . Here, φ λ ( x a , h ) L ∈ A 1 , 1 ( X L u , S [ λ ] ( V ( v ) ∞ ) ) \varphi_{\lambda}(xa,h)_{L}\in\mathcal{A}^{1,1}(X_{L}^{u},S_{[\lambda]}(V(v)_{\infty})) φ λ ( x a , h ) L ∈ A 1 , 1 ( X L u , S [ λ ] ( V ( v ) ∞ )) is the S [ λ ] ( V ( v ) ∞ ) S_{[\lambda]}(V(v)_{\infty}) S [ λ ] ( V ( v ) ∞ ) -valued Funke–Millson (Kudla–Millson if λ = 0 \lambda=0 λ = 0 ) form constructed above [28 Definition 3.9] . Furthermore, we have another function G t i λ ( ϕ i ∞ ) G_{t_{i}}^{\lambda}(\phi_{i}^{\infty}) G t i λ ( ϕ i ∞ ) on G 1 ( F ∞ ) G_{1}(F_{\infty}) G 1 ( F ∞ ) valued in the space of S [ λ ] ( V ( v ) ∞ ) S_{[\lambda]}(V(v)_{\infty}) S [ λ ] ( V ( v ) ∞ ) -valued Green functions on X L u X_{L}^{u} X L u defined by the formula
G t i λ ( ϕ i ∞ ) ( g ) = ρ λ + ( k ) δ ( k ) ∑ ( x , h ) ∈ H ( v ) ( F ) \ V ( v ) t i × H ( A F ∞ ) / L ϕ i ∞ ( h − 1 x ) ⋅ g λ ( x a , h ) L ⋅ ∏ v ∈ Hom ( F , R ) e 2 π i ( t i b v ) , G_{t_{i}}^{\lambda}(\phi_{i}^{\infty})(g)=\rho_{\lambda^{+}}(k)\delta(k)\sum_{(x,h)\in H(v)(F)\backslash V(v)_{t_{i}}\times H(A_{F}^{\infty})/L}\phi_{i}^{\infty}(h^{-1}x)\cdot g_{\lambda}(xa,h)_{L}\cdot\prod_{v\in\operatorname{Hom}(F,\mathbb{R})}e^{2\pi i(t_{i}b_{v})}, G t i λ ( ϕ i ∞ ) ( g ) = ρ λ + ( k ) δ ( k ) ( x , h ) ∈ H ( v ) ( F ) \ V ( v ) t i × H ( A F ∞ ) / L ∑ ϕ i ∞ ( h − 1 x ) ⋅ g λ ( x a , h ) L ⋅ v ∈ Hom ( F , R ) ∏ e 2 π i ( t i b v ) ,
where g λ ( x a , h ) L g_{\lambda}(xa,h)_{L} g λ ( x a , h ) L is defined above [28 Lemma 8.4] , so that G t i λ ( ϕ i ∞ ) ( g ) G_{t_{i}}^{\lambda}(\phi_{i}^{\infty})(g) G t i λ ( ϕ i ∞ ) ( g ) is a Green function for the S [ λ ] ( V ( v ) ∞ ) S_{[\lambda]}(V(v)_{\infty}) S [ λ ] ( V ( v ) ∞ ) -valued divisor Z t i ( ϕ i ∞ ) ( g ) Z_{t_{i}}(\phi_{i}^{\infty})(g) Z t i ( ϕ i ∞ ) ( g ) with tail form F t i λ ( ϕ i ∞ ) ( g ) F_{t_{i}}^{\lambda}(\phi_{i}^{\infty})(g) F t i λ ( ϕ i ∞ ) ( g ) [28 Proposition 8.6] .
For i = 1 , 2 i=1,2 i = 1 , 2 and t i ∈ ( F × ) + t_{i}\in(F^{\times})^{+} t i ∈ ( F × ) + , fix a harmonic Green function G t i λ ( s i ϕ i ∞ ) ∘ G_{t_{i}}^{\lambda}(s_{i}\phi_{i}^{\infty})^{\circ} G t i λ ( s i ϕ i ∞ ) ∘ (that is, Green function with vanishing tail form [28 Definition 5.9] ) for Z t i ( s i ϕ i ∞ ) Z_{t_{i}}(s_{i}\phi_{i}^{\infty}) Z t i ( s i ϕ i ∞ ) , so that G t i λ ( s i ϕ i ∞ ) ∘ − G t i λ ( s i ϕ i ∞ ) ( 1 2 ) G_{t_{i}}^{\lambda}(s_{i}\phi_{i}^{\infty})^{\circ}-G_{t_{i}}^{\lambda}(s_{i}\phi_{i}^{\infty})(1_{2}) G t i λ ( s i ϕ i ∞ ) ∘ − G t i λ ( s i ϕ i ∞ ) ( 1 2 ) is a smooth function on X L u X_{L}^{u} X L u .
By [28 Proposition 3.10] , there exists a unique element
F λ ( ϕ 1 ∞ ) ∈ A λ + ( G 1 ) K ∞ ⊗ C A 1 , 1 ( X L μ , S [ λ ] ( V ( v ) ∞ ) ) F^{\lambda}(\phi_{1}^{\infty}) \in\mathcal{A}_{\lambda+}(G_{1})^{K^{\infty}} \otimes_{\mathbb{C}} \mathcal{A}^{1,1}(X_{L}^{\mu},S_{[\lambda]}(V(v)_{\infty})) F λ ( ϕ 1 ∞ ) ∈ A λ + ( G 1 ) K ∞ ⊗ C A 1 , 1 ( X L μ , S [ λ ] ( V ( v ) ∞ ))
satisfying
F λ ( ϕ 1 ∞ ) F^{\lambda}(\phi_{1}^{\infty}) F λ ( ϕ 1 ∞ ) is supported on G 1 ( F ) G 1 ( F ∞ ) K ∞ G_{1}(F)G_{1}(F_{\infty})K^{\infty} G 1 ( F ) G 1 ( F ∞ ) K ∞ .
for every t 1 ∈ ( F × ) + t_{1} \in(F^{\times})^{+} t 1 ∈ ( F × ) + , F λ ( ϕ 1 ∞ ) t 1 ∣ G 1 ( F ∞ ) = F t 1 λ ( ϕ 1 ∞ ) F^{\lambda}(\phi_{1}^{\infty})_{t_{1}|G_{1}(F_{\infty})}=F^{\lambda}_{t_{1}}(\phi_{1}^{\infty}) F λ ( ϕ 1 ∞ ) t 1 ∣ G 1 ( F ∞ ) = F t 1 λ ( ϕ 1 ∞ ) ;
for every t 1 ∈ F ∖ ( F × ) + t_{1} \in F \setminus(F^{\times})^{+} t 1 ∈ F ∖ ( F × ) + , F λ ( ϕ 1 ∞ ) t 1 ∣ G 1 ( F ∞ ) = 0 F^{\lambda}(\phi_{1}^{\infty})_{t_{1}|G_{1}(F_{\infty})}=0 F λ ( ϕ 1 ∞ ) t 1 ∣ G 1 ( F ∞ ) = 0 .
Put
φ t 2 F M : = vol ♮ ( L ) 2 ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) ∈ A λ + ( G 1 ) K ∞ . \varphi_{t_{2}}^{\mathrm{FM}}:=\frac{\operatorname{vol}^{\natural}(L)}{2}\int_{X_{L}^{\mu}}\left(F_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty}),G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})^{\vee}-G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2})\right)_{S_{[\lambda]}(V(v)_{\infty})}\in\mathcal{A}_{\lambda+}(G_{1})^{K^{\infty}}. φ t 2 FM := 2 vol ♮ ( L ) ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) ∈ A λ + ( G 1 ) K ∞ .
We claim that the proposition holds with this function φ t 2 F M \varphi_{t_{2}}^{\mathrm{FM}} φ t 2 FM . Property (2) is obvious. It remains to show (1). It is clear that we may assume g 1 = m ( a ) g_{1}=m(a) g 1 = m ( a ) for a ∈ E ∞ × a\in E_{\infty}^{\times} a ∈ E ∞ × .
By [28 Definition 5.12 & Remark 5.13] , we have
⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z u = 1 2 ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ . \left\langle s_{1}^{*}Z_{t_{1}}^{\lambda}(\phi_{1}^{\infty})_{L}(g_{1}),s_{2}^{*}Z_{t_{2}}^{\lambda}(\phi_{2}^{\infty})_{L}\right\rangle_{\lambda,L,Z}^{u}=\frac{1}{2}\int_{X_{L}^{\mu}}G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})^{\vee}*G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})^{\vee}. ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z u = 2 1 ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ .
where ∗ * ∗ is defined above [28 Definition 5.12] . We compute the difference
∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ( m ( a ) ) ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) = ∫ X L μ ( G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ − G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ( m ( a ) ) , δ Z t 2 λ ( s 2 ϕ 2 ∞ ) ) S [ λ ] ( V ( v ) ∞ ) − ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a ) ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) . \begin{aligned}
\int_{X_{L}^{\mu}}G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})^{\vee}*G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})^{\vee}-\int_{X_{L}^{\mu}}G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})^{\vee}(m(a))*G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2}) \\
&=\int_{X_{L}^{\mu}}\left(G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})^{\vee}-G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})^{\vee}(m(a)),\delta Z_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})\right)_{S_{[\lambda]}(V(v)_{\infty})}-\int_{X_{L}^{\mu}}\left(F_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})(m(a)),G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2})\right)_{S_{[\lambda]}(V(v)_{\infty})}.
\end{aligned} ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ( m ( a )) ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) = ∫ X L μ ( G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ − G t 1 λ ( s 1 ϕ 1 ∞ ) ∨ ( m ( a )) , δ Z t 2 λ ( s 2 ϕ 2 ∞ ) ) S [ λ ] ( V ( v ) ∞ ) − ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a )) , G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) .
which, by Stokes’ formula, equals
∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a ) ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ ) S [ λ ] ( V ( v ) ∞ ) − ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a ) ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) = ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a ) ) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) = 2 vol ♮ ( L ) − 1 φ t 2 F M ( m ( a ) ) . \begin{aligned}
\int_{X_{L}^{\mu}}\left(F_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})(m(a)),G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})^{\vee}\right)_{S_{[\lambda]}(V(v)_{\infty})}-\int_{X_{L}^{\mu}}\left(F_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})(m(a)),G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2})\right)_{S_{[\lambda]}(V(v)_{\infty})} \\
&=\int_{X_{L}^{\mu}}\left(F_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})(m(a)),G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})^{\vee}-G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2})\right)_{S_{[\lambda]}(V(v)_{\infty})} \\
&=2\operatorname{vol}^{\natural}(L)^{-1}\varphi_{t_{2}}^{\mathrm{FM}}(m(a)).
\end{aligned} ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a )) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ ) S [ λ ] ( V ( v ) ∞ ) − ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a )) , G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) = ∫ X L μ ( F t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a )) , G t 2 λ ( s 2 ϕ 2 ∞ ) ∨ − G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) ) S [ λ ] ( V ( v ) ∞ ) = 2 vol ♮ ( L ) − 1 φ t 2 FM ( m ( a )) .
Thus, (1) is reduced to the identity
vol ♮ ( L ) ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a ) ) ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) = E t 1 , t 2 ( ( m ( a ) , 1 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) ) v . \operatorname{vol}^{\natural}(L)\int_{X_{L}^{\mu}}G_{t_{1}}^{\lambda}(s_{1}\phi_{1}^{\infty})(m(a))*G_{t_{2}}^{\lambda}(s_{2}\phi_{2}^{\infty})(1_{2})=\mathcal{E}_{t_{1},t_{2}}\left((m(a),1_{2}),\Phi_{\infty}^{\lambda}\cdot(s_{1}\phi_{1}^{\infty}\otimes s_{2}\phi_{2}^{\infty})\right)_{v}. vol ♮ ( L ) ∫ X L μ G t 1 λ ( s 1 ϕ 1 ∞ ) ( m ( a )) ∗ G t 2 λ ( s 2 ϕ 2 ∞ ) ( 1 2 ) = E t 1 , t 2 ( ( m ( a ) , 1 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) ) v .
However, this is exactly [28 Theorem 8.10] . □
Proposition 5.28. Consider
an object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_{F}^{\infty})) Hk ( H ( A F ∞ )) ,
a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^{\infty}=(\phi_{1}^{\infty},\phi_{2}^{\infty}) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of decomposable Schwartz functions in S ( V ∞ ) L \mathcal{S}(V^{\infty})^{L} S ( V ∞ ) L satisfying that ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) v (\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}})_{v} ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) v is regularly supported (Definition 2.16) for some v ∈ V F f i n v\in\mathbb{V}_{F}^{\mathrm{fin}} v ∈ V F fin ,
a pair ( s 1 , s 2 ) (s_{1},s_{2}) ( s 1 , s 2 ) in [ L × \ H ( A F ∞ ) / L × ] [\mathbb{L}^{\times}\backslash H(\mathbb{A}_{F}^{\infty})/\mathbb{L}^{\times}] [ L × \ H ( A F ∞ ) / L × ] ,
an open compact subgroup K ∞ K^{\infty} K ∞ of G 1 ( A F ∞ ) G_{1}(\mathbb{A}_{F}^{\infty}) G 1 ( A F ∞ ) that fixes ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ .
For every element φ ∈ A λ + ( G 1 ) c u s p \varphi\in\mathcal{A}_{\lambda+}(G_{1})_{\mathrm{cusp}} φ ∈ A λ + ( G 1 ) cusp invariant under K ∞ K^{\infty} K ∞ , the identity
vol ♮ ( L ) ∫ Ω φ ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z u d g 1 \operatorname{vol}^{\natural}(L)\int_{\Omega}\overline{\varphi(g_{1})}\sum_{t_{1}\in(F^{\times})^{+}}\left\langle s_{1}^{*}Z_{t_{1}}^{\lambda}(\phi_{1}^{\infty})_{L}(g_{1}),s_{2}^{*}Z_{t_{2}}^{\lambda}(\phi_{2}^{\infty})_{L}\right\rangle_{\lambda,L,Z}^{u}\,dg_{1} vol ♮ ( L ) ∫ Ω φ ( g 1 ) t 1 ∈ ( F × ) + ∑ ⟨ s 1 ∗ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z t 2 λ ( ϕ 2 ∞ ) L ⟩ λ , L , Z u d g 1 = 1 2 ∫ Ω φ ( g 1 ) ‾ ⋅ E t 1 , t 2 ( ( g 1 , 1 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) ) v d g 1 . =\frac{1}{2}\int_{\Omega}\overline{\varphi(g_{1})}\cdot\mathcal{E}_{t_{1},t_{2}}\left((g_{1},1_{2}),\Phi_{\infty}^{\lambda}\cdot(s_{1}\phi_{1}^{\infty}\otimes s_{2}\phi_{2}^{\infty})\right)_{v}\,dg_{1}. = 2 1 ∫ Ω φ ( g 1 ) ⋅ E t 1 , t 2 ( ( g 1 , 1 2 ) , Φ ∞ λ ⋅ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ) ) v d g 1 .
holds in which both sides are absolutely convergent, where Ω ⊆ G 1 ( F ∞ ) \Omega\subseteq G_1(F_\infty) Ω ⊆ G 1 ( F ∞ ) is an arbitrary fundamental region for the arithmetic subgroup G 1 ( F ) ∩ K ∘ G_1(F) \cap K^\circ G 1 ( F ) ∩ K ∘ .
Proof. This is [28 Theorem 8.11] . □
Abundance of nonarchimedean Schwartz functions
We continue to assume r = 1 r=1 r = 1 . Consider
a finite set ⋄ \diamond ⋄ of rational primes containing all those ramified in E E E and all those underlying V ( V ) \mathcal{V}(V) V ( V ) ;
an open compact subgroup L ⋄ L^\diamond L ⋄ of H ( A F ∞ ) H(\mathbb{A}_F^\infty) H ( A F ∞ ) that is the stabilizer of a self-dual lattice of V ⋄ V^\diamond V ⋄ ;
a subset ⧫ ⊆ V F ( ⋄ ) \blacklozenge\subseteq\mathcal{V}_F^{(\diamond)} ⧫ ⊆ V F ( ⋄ ) containing all those nonsplit in E E E and satisfying d ( ⧫ ) ≥ 2 d(\blacklozenge)\geq2 d ( ⧫ ) ≥ 2 (Notation 2.6(5)).
Definition 5.29. For every field L \mathbb{L} L of characteristic zero, we define S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond}_{\blacklozenge} S (( V ∞ ) 2 , L ) ⧫ L ⋄ to be the subspace of S ( ( V ∞ ) 2 , L ) L ⋄ \mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond} S (( V ∞ ) 2 , L ) L ⋄ spanned by decomposable functions Φ ∞ = ⨂ v Φ v ∞ \Phi^\infty=\bigotimes_v\Phi_v^\infty Φ ∞ = ⨂ v Φ v ∞ satisfying
(1) for v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ , Φ v ∞ \Phi_v^\infty Φ v ∞ is regularly supported;
(2) for v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ nonsplit in E E E , Φ v ∞ \Phi_v^\infty Φ v ∞ is deeply regularly supported;
(3) for v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ nonsplit in E E E , Φ v ∞ \Phi_v^\infty Φ v ∞ is regularly supported along both variables of V v 2 V_v^2 V v 2 .35
See Definition 2.16 for the above terminologies. It is clear that S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond}_{\blacklozenge} S (( V ∞ ) 2 , L ) ⧫ L ⋄ is stable under the action of H ( F ⋄ ) H(F_\diamond) H ( F ⋄ ) . Furthermore, we denote by
S ( ( V ∞ ) 2 , L ) ( ⧫ ) L ⋄ ⊆ S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond}_{(\blacklozenge)} \subseteq\mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond}_{\blacklozenge} S (( V ∞ ) 2 , L ) ( ⧫ ) L ⋄ ⊆ S (( V ∞ ) 2 , L ) ⧫ L ⋄ the G 1 ( A ∞ ) 2 G_1(\mathbb{A}^\infty)^2 G 1 ( A ∞ ) 2 -stable subspace generated by S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \mathcal{S}((V^\infty)^2,\mathbb{L})^{L^\diamond}_{\blacklozenge} S (( V ∞ ) 2 , L ) ⧫ L ⋄ .
Lemma 5.30. Let ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ (Definition 2.27) and L ⊆ C L\subseteq\mathbb{C} L ⊆ C a subfield over which π \pi π is defined. Suppose that V = V ( π ) V=V(\pi) V = V ( π ) as in Remark 2.27(1) and ( π v ) K 1 , v ≠ { 0 } (\pi_v)^{K_{1,v}}\neq\{0\} ( π v ) K 1 , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ( ⋄ ) v\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\diamond)}_F v ∈ V F fin ∖ V F ( ⋄ ) . Then one can find a pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) of automorphic forms in A π ∩ A λ , + ( G 1 ) c u s p h o l \mathcal{A}_\pi\cap\mathcal{A}_{\lambda,+}(G_1)^{\mathrm{hol}}_{\mathrm{cusp}} A π ∩ A λ , + ( G 1 ) cusp hol and a pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^\infty=(\phi_1^\infty,\phi_2^\infty) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of Schwartz functions in S ( V ∞ , L ) \mathcal{S}(V^\infty,L) S ( V ∞ , L ) satisfying ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \phi_1^\infty\otimes\phi_2^\infty\in\mathcal{S}((V^\infty)^2,L)^{L^\diamond}_{\blacklozenge} ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S (( V ∞ ) 2 , L ) ⧫ L ⋄ , such that
Z f i n ⋄ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) ≠ 0 \mathcal{Z}^{\diamond}_{\mathrm{fin}}(\varphi,\phi_1^\infty\otimes\phi_2^\infty)\neq0 Z fin ⋄ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) = 0 (Definition 2.28).
Proof. By [20 Proposition 3.13] , we may find φ \varphi φ and ϕ ∞ \phi^\infty ϕ ∞ as in the statement satisfying Definition 5.29(1) such that Z f i n ⋄ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) ≠ 0 \mathcal{Z}^{\diamond}_{\mathrm{fin}}(\varphi,\phi_1^\infty\otimes\phi_2^\infty)\neq0 Z fin ⋄ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) = 0 .
For each v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ nonsplit in E E E , we may replace ϕ i ∞ \phi_i^\infty ϕ i ∞ by ω 1 , v ( m ( a i ) ) ϕ i ∞ \omega_{1,v}(m(a_i))\phi_i^\infty ω 1 , v ( m ( a i )) ϕ i ∞ (and same for φ i \varphi_i φ i ) with a i ∈ E v × a_i\in E_v^\times a i ∈ E v × sufficiently close to zero to further meet 5.29(2) without changing the value of Z f i n ⋄ \mathcal{Z}^{\diamond}_{\mathrm{fin}} Z fin ⋄ .
Finally, for each v ∈ ⧫ v\in\blacklozenge v ∈ ⧫ nonsplit in E E E , we may choose elements b 1 , b 2 ∈ F v b_1,b_2\in F_v b 1 , b 2 ∈ F v such that φ i ′ : = ( π v ( n ( b i ) ) − 1 ) φ i ≠ 0 \varphi_i' := (\pi_v(n(b_i))-1)\varphi_i\neq0 φ i ′ := ( π v ( n ( b i )) − 1 ) φ i = 0 for i = 1 , 2 i=1,2 i = 1 , 2 . Put ϕ i ∞ ′ : = ω 1 , v ( n ( b i ) − 1 ) ϕ i ∞ \phi_i^{\infty\prime}:=\omega_{1,v}(n(b_i)-1)\phi_i^\infty ϕ i ∞′ := ω 1 , v ( n ( b i ) − 1 ) ϕ i ∞ for i = 1 , 2 i=1,2 i = 1 , 2 . Then ϕ 1 ∞ ′ ⊗ ϕ 2 ∞ ′ \phi_1^{\infty\prime}\otimes\phi_2^{\infty\prime} ϕ 1 ∞′ ⊗ ϕ 2 ∞′ belongs to S ( ( V ∞ ) 2 , L ) ⧫ L ⋄ \mathcal{S}((V^\infty)^2,L)^{L^\diamond}_{\blacklozenge} S (( V ∞ ) 2 , L ) ⧫ L ⋄ , satisfying Z f i n ⋄ ( φ ′ , ϕ 1 ∞ ′ ⊗ ϕ 2 ∞ ′ ) ≠ 0 \mathcal{Z}^{\diamond}_{\mathrm{fin}}(\varphi',\phi_1^{\infty\prime}\otimes\phi_2^{\infty\prime})\neq0 Z fin ⋄ ( φ ′ , ϕ 1 ∞′ ⊗ ϕ 2 ∞′ ) = 0 , where φ ′ = ( φ 1 ′ , φ 2 ′ ) \varphi'=(\varphi_1',\varphi_2') φ ′ = ( φ 1 ′ , φ 2 ′ ) .
The lemma is proved. □
At the end of this section, we explain that the ingredients we obtained so far are enough to give a proof of Conjecture 4.11 when λ = 0 \lambda=0 λ = 0 (and r = 1 r=1 r = 1 ), simplifying to some degree the one given in [26 ] ; see Remark 5.31 below.
For ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) in Conjecture 4.11 , we choose ⋄ \diamond ⋄ so that ( π v ) K 1 , v ≠ { 0 } (\pi_v)^{K_{1,v}}\neq\{0\} ( π v ) K 1 , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ( ⋄ ) v\in\mathcal{V}^{\mathrm{fin}}_F\setminus\mathcal{V}^{(\diamond)}_F v ∈ V F fin ∖ V F ( ⋄ ) . We may assume V = V ( π ) V=V(\pi) V = V ( π ) as in Remark 4.12(2), since otherwise both sides of the identity in Conjecture 4.11 are zero. In particular, the global root number ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 .
Note that Conjecture 3.36 is known when λ = 0 \lambda=0 λ = 0 by [25 Theorem 3.5] . It follows that for every pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^\infty=(\phi_1^\infty,\phi_2^\infty) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of Schwartz functions in S ( V ∞ ) L \mathcal{S}(V^\infty)^L S ( V ∞ ) L (with L ⋄ L^\diamond L ⋄ the stabilizer of a self-dual lattice) and every pair s ‾ = ( s 1 , s 2 ) \underline{s}=(s_1,s_2) s = ( s 1 , s 2 ) of elements in C [ L ∨ \ H ( A F ∞ ) / L ∨ ] 0 \mathbb{C}[L^{\vee}\backslash H(\mathbb{A}_{F}^{\infty})/L^{\vee}]^{0} C [ L ∨ \ H ( A F ∞ ) / L ∨ ] 0 , the function I ϕ ‾ ∞ s ‾ \mathcal{I}_{\underline{\phi}^{\infty}}^{\underline{s}} I ϕ ∞ s on G 1 ( A F ) × G 1 ( A F ) G_1(\mathbb{A}_F)\times G_1(\mathbb{A}_F) G 1 ( A F ) × G 1 ( A F ) defined by the formula
I ϕ ‾ ∞ s ‾ ( g 1 , g 2 ) : = vol ♮ ( L ) ⟨ s 1 ∗ Z 0 ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z 0 ( ϕ 2 ∞ ) L ( g 2 ) ⟩ 0 , L \mathcal{I}_{\underline{\phi}^{\infty}}^{\underline{s}}(g_1,g_2):=\operatorname{vol}^{\natural}(L)\left\langle s_1^{*}Z^{0}(\phi_1^{\infty})_L(g_1),s_2^{*}Z^{0}(\phi_2^{\infty})_L(g_2)\right\rangle_{0,L} I ϕ ∞ s ( g 1 , g 2 ) := vol ♮ ( L ) ⟨ s 1 ∗ Z 0 ( ϕ 1 ∞ ) L ( g 1 ) , s 2 ∗ Z 0 ( ϕ 2 ∞ ) L ( g 2 ) ⟩ 0 , L
belongs to the space A 0 ( G 1 ) h o l ⊗ C A 0 ( G 1 ) h o l ‾ \mathcal{A}_0(G_1)^{\mathrm{hol}}\otimes_{\mathbb{C}}\overline{\mathcal{A}_0(G_1)^{\mathrm{hol}}} A 0 ( G 1 ) hol ⊗ C A 0 ( G 1 ) hol .
Now suppose that
(a) ϕ ∞ \phi^{\infty} ϕ ∞ is decomposable such that ϕ 1 ∞ , ⋄ ⊗ ϕ 2 ∞ , ⋄ ‾ ∈ S ( ( V ∞ ) 2 , L ) L ∨ \phi_1^{\infty,\diamond}\otimes\overline{\phi_2^{\infty,\diamond}}\in\mathcal{S}((V^{\infty})^2,L)^{L^{\vee}} ϕ 1 ∞ , ⋄ ⊗ ϕ 2 ∞ , ⋄ ∈ S (( V ∞ ) 2 , L ) L ∨ for some ⧫ \blacklozenge ⧫ as above, and that ϕ 1 ∞ , ⋄ = ϕ 2 ∞ , ⋄ \phi_1^{\infty,\diamond}=\phi_2^{\infty,\diamond} ϕ 1 ∞ , ⋄ = ϕ 2 ∞ , ⋄ is the characteristic function of the self-dual lattice L ∨ L^{\vee} L ∨ stabilizes;
(b) both s 1 s_1 s 1 and s 2 s_2 s 2 are supported at places split in E E E .
Using the height decomposition formula (Proposition 4.5 ) and combining Proposition 5.2 (1) (plus Remark 5.3), Proposition 5.5, Proposition 5.9 and Proposition 5.28 for various local heights, we conclude that for every g 2 ∈ G 1 ( A F ) g_2\in G_1(\mathbb{A}_F) g 2 ∈ G 1 ( A F ) , the difference
I ϕ ‾ ∞ s ‾ ( − , g 2 ) − E ′ ( 0 , ( − , g 2 ) , Φ 0 ⊗ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) ) , \mathcal{I}_{\underline{\phi}^{\infty}}^{\underline{s}}(-,g_2)-E'(0,(-,g_2),\Phi^0\otimes(s_1\phi_1^{\infty}\otimes\overline{s_2\phi_2^{\infty}})), I ϕ ∞ s ( − , g 2 ) − E ′ ( 0 , ( − , g 2 ) , Φ 0 ⊗ ( s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ )) ,
as an element in A 0 ( G 1 ) h o l \mathcal{A}_0(G_1)^{\mathrm{hol}} A 0 ( G 1 ) hol , is perpendicular to A π ∩ A 0 ( G 1 ) c u s p h o l \mathcal{A}_{\pi}\cap\mathcal{A}_0(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} A π ∩ A 0 ( G 1 ) cusp hol under the inner product in Definition 2.14(2).36 Then by Proposition 2.29 and Remark 2.30 (2), we obtain the identity
vol ♮ ( L ) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ 0 , L = L ′ ( 1 2 , π ) b 2 ( 0 ) ⋅ ( − π 4 ) [ F : Q ] ⋅ Z f i n ♮ ( φ , s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ ‾ ) \operatorname{vol}^{\natural}(L)\left\langle\Theta_{\phi_1^{\infty}}^{\lambda}(\varphi_1),\Theta_{\phi_2^{\infty}}^{\lambda}(\varphi_2)\right\rangle_{0,L}
=
\frac{L'(\frac{1}{2},\pi)}{b_2(0)}\cdot\left(-\frac{\pi}{4}\right)^{[F:\mathbb{Q}]}\cdot\mathcal{Z}_{\mathrm{fin}}^{\natural}(\varphi,s_1\phi_1^{\infty}\otimes\overline{s_2\phi_2^{\infty}}) vol ♮ ( L ) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ 0 , L = b 2 ( 0 ) L ′ ( 2 1 , π ) ⋅ ( − 4 π ) [ F : Q ] ⋅ Z fin ♮ ( φ , s 1 ϕ 1 ∞ ⊗ s 2 ϕ 2 ∞ )
for ϕ ∞ \phi^{\infty} ϕ ∞ satisfying (a), s ‾ \underline{s} s satisfying (b), and every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) in A π ∩ A 0 ( G 1 ) c u s p h o l \mathcal{A}_{\pi}\cap\mathcal{A}_0(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} A π ∩ A 0 ( G 1 ) cusp hol satisfying
(c) both φ 1 \varphi_1 φ 1 and φ 2 \varphi_2 φ 2 are invariant under K 1 , v K_{1,v} K 1 , v (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ( ⋄ ) v\in\mathcal{V}_{F}^{\mathrm{fin}}\setminus\mathcal{V}_{F}^{(\diamond)} v ∈ V F fin ∖ V F ( ⋄ ) .
Now we can choose s ‾ \underline{s} s satisfying (b) such that s i s_i s i fixes φ i \varphi_i φ i for i = 1 , 2 i=1,2 i = 1 , 2 . Then we arrive at the identity
vol ♮ ( L ) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ 0 , L = L ′ ( 1 2 , π ) b 2 ( 0 ) ⋅ ( − π 4 ) [ F : Q ] ⋅ Z f i n ♮ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) (63) \operatorname{vol}^{\natural}(L)\left\langle\Theta_{\phi_1^{\infty}}^{\lambda}(\varphi_1),\Theta_{\phi_2^{\infty}}^{\lambda}(\varphi_2)\right\rangle_{0,L}
=
\frac{L'(\frac{1}{2},\pi)}{b_2(0)}\cdot\left(-\frac{\pi}{4}\right)^{[F:\mathbb{Q}]}\cdot\mathcal{Z}_{\mathrm{fin}}^{\natural}(\varphi,\phi_1^{\infty}\otimes\overline{\phi_2^{\infty}}) \tag*{(63)} vol ♮ ( L ) ⟨ Θ ϕ 1 ∞ λ ( φ 1 ) , Θ ϕ 2 ∞ λ ( φ 2 ) ⟩ 0 , L = b 2 ( 0 ) L ′ ( 2 1 , π ) ⋅ ( − 4 π ) [ F : Q ] ⋅ Z fin ♮ ( φ , ϕ 1 ∞ ⊗ ϕ 2 ∞ ) (63)
By Lemma 5.30 , we may find φ \varphi φ satisfying (a) and ϕ ∞ \phi^{\infty} ϕ ∞ satisfying (c) such that Z f i n ♮ ( φ , ϕ ∞ ) ≠ 0 \mathcal{Z}_{\mathrm{fin}}^{\natural}(\varphi,\phi^{\infty})\ne0 Z fin ♮ ( φ , ϕ ∞ ) = 0 . Finally, by the multiplicity one property [26 Proposition 6.7] , (5.10) holds for arbitrary pairs φ \varphi φ and ϕ ∞ \phi^{\infty} ϕ ∞ .
Remark 5.31. The proof given above simplifies the one given in [26 ] in the following three aspects:
We have simplified the argument for nonarchimedean local heights at bad reductions without using the “change of Shimura data” argument in [26 ] ; this is possible partially due to the recent advances on integral models of unitary Shimura varieties.
We have used the notion of special Eisenstein ideal to annihilate certain degenerate terms that appeared in the computation of nonarchimedean local heights at bad reductions in [26 ] .
We have avoided the step of holomorphic projection for the archimedean computation by using Proposition 5.28 , an idea originally adopted in [20 ] .
Proof of main results
In this chapter, we prove our main results. Fix a totally positive definite incoherent hermitian space V V V over A E / A F A_E/A_F A E / A F of rank 2 satisfying that the set D ( V ) \mathfrak{D}(V) D ( V ) is not a singleton (and put H : = U ( V ) H:=U(V) H := U ( V ) as always), and a nontrivial balanced hermitian E E E -weight λ \lambda λ of rank 2 (Definition 2.7).
Modularity of motivic doubling kernel functions
By Theorem 4.15 , for every object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) Hk ( H ( A F ∞ )) , we have a well-defined height pairing
⟨ , ⟩ λ , L : H 2 ( X L , M L , λ ( 1 ) ) × H 2 ( X L , M L , λ ( 1 ) ) ⟶ C . \langle\ ,\ \rangle_{\lambda,L}: H^{2}(X_{L},M_{L,\lambda}(1)) \times H^{2}(X_{L},M_{L,\lambda}(1)) \longrightarrow\mathbb{C}. ⟨ , ⟩ λ , L : H 2 ( X L , M L , λ ( 1 )) × H 2 ( X L , M L , λ ( 1 )) ⟶ C .
As in §5.7, we consider
a finite set ⋄ \diamond ⋄ of rational primes containing all those ramified in E E E and all those underlying B ( V ) \mathscr{B}(V) B ( V ) ;
an open compact subgroup L ⋄ L^{\diamond} L ⋄ of H ( A F ∞ , ⋄ ) H(\mathbb{A}^{\infty,\diamond}_{F}) H ( A F ∞ , ⋄ ) that is the stabilizer of a self-dual lattice of V ⋄ , ∞ V^{\diamond,\infty} V ⋄ , ∞ ;
a subset ⧫ ⊆ B F ( ⋄ ) \blacklozenge\subseteq\mathscr{B}^{(\diamond)}_{F} ⧫ ⊆ B F ( ⋄ ) containing all those nonsplit in E E E and satisfying d ( ⧫ ) ≥ 2 d(\blacklozenge) \ge2 d ( ⧫ ) ≥ 2 (Notation 2.6(5)).
Recall that we have the G 1 ( A F ∞ ) 2 G_{1}(\mathbb{A}^{\infty}_{F})^{2} G 1 ( A F ∞ ) 2 -stable subspace S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ ⊆ S ( ( V ∞ ) 2 , C ) L ⋄ \mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} \subseteq\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}} S (( V ∞ ) 2 , C ) ⧫ L ⋄ ⊆ S (( V ∞ ) 2 , C ) L ⋄ from Definition 5.29 .
Theorem 6.1. There is a unique G 1 ( A F ∞ ) 2 G_{1}(\mathbb{A}^{\infty}_{F})^{2} G 1 ( A F ∞ ) 2 -equivariant complex linear map
I ~ : S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ ⟶ A λ + ( G 1 ) c u s p h o l ⊗ C A λ + ( G 1 ) c u s p h o l ‾ \widetilde{\mathcal{I}}:\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge}\longrightarrow\mathcal{A}_{\lambda+}(G_{1})^{\mathrm{hol}}_{\mathrm{cusp}}\otimes_{\mathbb{C}}\overline{\mathcal{A}_{\lambda+}(G_{1})^{\mathrm{hol}}_{\mathrm{cusp}}} I : S (( V ∞ ) 2 , C ) ⧫ L ⋄ ⟶ A λ + ( G 1 ) cusp hol ⊗ C A λ + ( G 1 ) cusp hol (Remark 5.26) such that for every pair ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^{\infty}=(\phi_{1}^{\infty},\phi_{2}^{\infty}) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) of Schwartz functions in S ( V ∞ , C ) L ⋄ \mathcal{S}(V^{\infty},\mathbb{C})^{L^{\diamond}} S ( V ∞ , C ) L ⋄ satisfying ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ∈ S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ \phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}}\in\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S (( V ∞ ) 2 , C ) ⧫ L ⋄ , the identity
I ~ ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ) ( g 1 , g 2 ) = vol ♮ ( L ) ⋅ ∑ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L \widetilde{\mathcal{I}}(\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}})(g_{1},g_{2})=\operatorname{vol}^{\natural}(L)\cdot\sum_{t_{1}\in(F^{\times})^{+}}\sum_{t_{2}\in(F^{\times})^{+}}\left\langle Z^{\lambda}_{t_{1}}(\phi_{1}^{\infty})_{L}(g_{1}),Z^{\lambda}_{t_{2}}(\phi_{2}^{\infty})_{L}(g_{2})\right\rangle_{\lambda,L} I ( ϕ 1 ∞ ⊗ ϕ 2 ∞ ) ( g 1 , g 2 ) = vol ♮ ( L ) ⋅ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L holds with absolute convergence on the right-hand side for every ( g 1 , g 2 ) ∈ G 1 ( A F ⋄ ) 2 (g_{1},g_{2})\in G_{1}(\mathbb{A}^{\diamond}_{F})^{2} ( g 1 , g 2 ) ∈ G 1 ( A F ⋄ ) 2 and every L ∈ Hk ( H ( A F ∞ ) ) L\in\operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) L ∈ Hk ( H ( A F ∞ )) fixing both ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ and ϕ 2 ∞ \phi_{2}^{\infty} ϕ 2 ∞ .
Proof. The uniqueness is clear since G 1 ( F ) G 1 ( A F ∞ ) G_{1}(F)G_{1}(\mathbb{A}^{\infty}_{F}) G 1 ( F ) G 1 ( A F ∞ ) is dense in G 1 ( A F ) G_{1}(\mathbb{A}_{F}) G 1 ( A F ) , and the functions of the form ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ \phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}} ϕ 1 ∞ ⊗ ϕ 2 ∞ in the statement generate S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ \mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} S (( V ∞ ) 2 , C ) ⧫ L ⋄ as a G 1 ( A F ∞ ) 2 G_{1}(\mathbb{A}^{\infty}_{F})^{2} G 1 ( A F ∞ ) 2 -module.
For every Φ ∞ ∈ S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ \Phi^{\infty}\in\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} Φ ∞ ∈ S (( V ∞ ) 2 , C ) ⧫ L ⋄ , we may express it as
Φ ∞ = ∑ { ϕ ∞ } ω 1 ∞ ( g 1 ′ ) ϕ 1 ∞ ⊗ ω 1 ∞ ( g 2 ′ ) ϕ 2 ∞ ‾ . (64) \Phi^{\infty}=\sum_{\{\phi^{\infty}\}}\omega_{1}^{\infty}(g_{1}')\phi_{1}^{\infty}\otimes\overline{\omega_{1}^{\infty}(g_{2}')\phi_{2}^{\infty}}. \tag*{(64)} Φ ∞ = { ϕ ∞ } ∑ ω 1 ∞ ( g 1 ′ ) ϕ 1 ∞ ⊗ ω 1 ∞ ( g 2 ′ ) ϕ 2 ∞ . (64)
in which the sum is taken over finitely many pairs ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) \phi^{\infty}=(\phi_{1}^{\infty},\phi_{2}^{\infty}) ϕ ∞ = ( ϕ 1 ∞ , ϕ 2 ∞ ) in S ( V ∞ , C ) L ⋄ \mathcal{S}(V^{\infty},\mathbb{C})^{L^{\diamond}} S ( V ∞ , C ) L ⋄ satisfying ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ∈ S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ \phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}}\in\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S (( V ∞ ) 2 , C ) ⧫ L ⋄ with ( g 1 ′ , g 2 ′ ) ∈ G 1 ( A F ∞ ) 2 (g_{1}',g_{2}')\in G_{1}(\mathbb{A}^{\infty}_{F})^{2} ( g 1 ′ , g 2 ′ ) ∈ G 1 ( A F ∞ ) 2 . For ( g 1 , g 2 ) ∈ G 1 ( A F ) 2 (g_{1},g_{2})\in G_{1}(\mathbb{A}_{F})^{2} ( g 1 , g 2 ) ∈ G 1 ( A F ) 2 , put
I ~ ( Φ ∞ ) ( g 1 , g 2 ) : = vol ♮ ( L ) ∑ { ϕ ∞ } ∑ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 g 1 ′ ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 g 2 ′ ) ⟩ λ , L , \widetilde{\mathcal{I}}(\Phi^{\infty})(g_{1},g_{2}):=\operatorname{vol}^{\natural}(L)\sum_{\{\phi^{\infty}\}}\sum_{t_{1}\in(F^{\times})^{+}}\sum_{t_{2}\in(F^{\times})^{+}}\left\langle Z^{\lambda}_{t_{1}}(\phi_{1}^{\infty})_{L}(g_{1}g_{1}'),Z^{\lambda}_{t_{2}}(\phi_{2}^{\infty})_{L}(g_{2}g_{2}')\right\rangle_{\lambda,L}, I ( Φ ∞ ) ( g 1 , g 2 ) := vol ♮ ( L ) { ϕ ∞ } ∑ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 g 1 ′ ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 g 2 ′ ) ⟩ λ , L ,
where L L L is some object of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}^{\infty}_{F})) Hk ( H ( A F ∞ )) fixing both ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ and ϕ 2 ∞ \phi_{2}^{\infty} ϕ 2 ∞ for every ϕ ∞ \phi^{\infty} ϕ ∞ appearing in the above sum.
Assume the absolute convergence on the right-hand side. Then by Lemma 4.3, I ~ ( Φ ∞ ) ( g 1 , g 2 ) \widetilde{\mathcal{I}}(\Phi^{\infty})(g_{1},g_{2}) I ( Φ ∞ ) ( g 1 , g 2 ) does not depend on the choice of L L L ; from this, it is straightforward to see that I ~ ( Φ ∞ ) ( g 1 , g 2 ) \widetilde{\mathcal{I}}(\Phi^{\infty})(g_{1},g_{2}) I ( Φ ∞ ) ( g 1 , g 2 ) also does not depend on the choice of the expression of (6.1).
Thus, it remains to show that for Φ ∞ = ϕ 1 ∞ ⊗ ϕ 2 ∞ ‾ ∈ S ( ( V ∞ ) 2 , C ) ⧫ L ⋄ \Phi^{\infty}=\phi_{1}^{\infty}\otimes\overline{\phi_{2}^{\infty}}\in\mathcal{S}((V^{\infty})^{2},\mathbb{C})^{L^{\diamond}}_{\blacklozenge} Φ ∞ = ϕ 1 ∞ ⊗ ϕ 2 ∞ ∈ S (( V ∞ ) 2 , C ) ⧫ L ⋄ with ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ and ϕ 2 ∞ \phi_{2}^{\infty} ϕ 2 ∞ in S ( V ∞ , C ) L ⋄ \mathcal{S}(V^{\infty},\mathbb{C})^{L^{\diamond}} S ( V ∞ , C ) L ⋄ , the sum
vol ♮ ( L ) ∑ t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L \operatorname{vol}^{\natural}(L)\sum_{t_{1}\in(F^{\times})^{+}}\sum_{t_{2}\in(F^{\times})^{+}}\left\langle Z^{\lambda}_{t_{1}}(\phi_{1}^{\infty})_{L}(g_{1}),Z^{\lambda}_{t_{2}}(\phi_{2}^{\infty})_{L}(g_{2})\right\rangle_{\lambda,L} vol ♮ ( L ) t 1 ∈ ( F × ) + ∑ t 2 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L
is absolutely convergent for ( g 1 , g 2 ) ∈ G 1 ( A F ⋄ ) 2 (g_{1},g_{2})\in G_{1}(\mathbb{A}^{\diamond}_{F})^{2} ( g 1 , g 2 ) ∈ G 1 ( A F ⋄ ) 2 and extends to a function in A λ + ( G 1 ) c u s p h o l ⊗ C A λ + ( G 1 ) c u s p h o l ‾ \mathcal{A}_{\lambda+}(G_{1})^{\mathrm{hol}}_{\mathrm{cusp}}\otimes_{\mathbb{C}}\overline{\mathcal{A}_{\lambda+}(G_{1})^{\mathrm{hol}}_{\mathrm{cusp}}} A λ + ( G 1 ) cusp hol ⊗ C A λ + ( G 1 ) cusp hol . However, this is equivalent to the following claim: If K ∞ ⊂ G 1 ( A F ∞ ) K^{\infty}\subset G_{1}(\mathbb{A}^{\infty}_{F}) K ∞ ⊂ G 1 ( A F ∞ ) is an open compact subgroup that fixes both ϕ 1 ∞ \phi_{1}^{\infty} ϕ 1 ∞ and ϕ 2 ∞ \phi_{2}^{\infty} ϕ 2 ∞ , then
(1) for every t 2 ∈ ( F × ) + t_{2}\in(F^{\times})^{+} t 2 ∈ ( F × ) + and g 2 ∈ G 1 ( A F ) g_{2}\in G_{1}(\mathbb{A}_{F}) g 2 ∈ G 1 ( A F ) , the sum
vol ♮ ( L ) ∑ t 1 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L \operatorname{vol}^{\natural}(L)\sum_{t_{1}\in(F^{\times})^{+}}\left\langle Z^{\lambda}_{t_{1}}(\phi_{1}^{\infty})_{L}(g_{1}),Z^{\lambda}_{t_{2}}(\phi_{2}^{\infty})_{L}(g_{2})\right\rangle_{\lambda,L} vol ♮ ( L ) t 1 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L
is absolutely convergent for g 1 ∈ G 1 ( A F ⋄ ) g_{1}\in G_{1}(\mathbb{A}^{\diamond}_{F}) g 1 ∈ G 1 ( A F ⋄ ) and extends to a function in ( A λ + ( G 1 ) c u s p h o l ) K ∞ (\mathcal{A}_{\lambda+}(G_{1})^{\mathrm{hol}}_{\mathrm{cusp}})^{K^{\infty}} ( A λ + ( G 1 ) cusp hol ) K ∞ ; (2) for every t 1 ∈ ( F × ) + t_1 \in(F^\times)^+ t 1 ∈ ( F × ) + and g 1 ∈ G 1 ( A F ⧫ ) g_1 \in G_1(\mathbb{A}_F^\blacklozenge) g 1 ∈ G 1 ( A F ⧫ ) , the sum
vol ♮ ( L ) ∑ t 2 ∈ ( F × ) + ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L \operatorname{vol}^{\natural}(L)\sum_{t_2\in(F^\times)^+}\left\langle Z_{t_1}^{\lambda}(\phi_1^\infty)_L(g_1),Z_{t_2}^{\lambda}(\phi_2^\infty)_L(g_2)\right\rangle_{\lambda,L} vol ♮ ( L ) t 2 ∈ ( F × ) + ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L
is absolutely convergent for g 2 ∈ G 1 ( A F ⧫ ) g_2 \in G_1(\mathbb{A}_F^\blacklozenge) g 2 ∈ G 1 ( A F ⧫ ) and extends to a function in ( A λ + ( G 1 ) c u s p h o l ) K ∞ (\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}})^{K^\infty} ( A λ + ( G 1 ) cusp hol ) K ∞ .
We only show the property (1) by symmetry, that is,
∑ t 1 ∈ ( F × ) + I ~ t 1 ( g 1 ) (65) \sum_{t_1\in(F^\times)^+}\widetilde{\mathcal{I}}_{t_1}(g_1) \tag*{(65)} t 1 ∈ ( F × ) + ∑ I t 1 ( g 1 ) (65)
is absolutely convergent and extends to a function in ( A λ + ( G 1 ) c u s p h o l ) K ∞ (\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}})^{K^\infty} ( A λ + ( G 1 ) cusp hol ) K ∞ , where
I ~ t 1 ( g 1 ) : = vol ♮ ( L ) ⋅ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , \widetilde{\mathcal{I}}_{t_1}(g_1):=\operatorname{vol}^{\natural}(L)\cdot\left\langle Z_{t_1}^{\lambda}(\phi_1^\infty)_L(g_1),Z_{t_2}^{\lambda}(\phi_2^\infty)_L(g_2)\right\rangle_{\lambda,L}, I t 1 ( g 1 ) := vol ♮ ( L ) ⋅ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L ,
which extends uniquely to a smooth function on G 1 ( A F ⧫ ) K ∞ G_1(\mathbb{A}_F^\blacklozenge)K^\infty G 1 ( A F ⧫ ) K ∞ . It is clear from its expression that it suffices to show that (65) is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) . Without loss of generality, we may assume that L L L is of the form L ⋄ L ⋄ L_{\diamond}L^{\diamond} L ⋄ L ⋄ (with L ⋄ L^{\diamond} L ⋄ the given one at the beginning of this section).
Choose an auxiliary rational prime ℓ ≠ ⋄ \ell\ne\diamond ℓ = ⋄ that splits in E 0 E_0 E 0 . For g 1 ∈ G 1 ( A F ⧫ ) K ∞ g_1 \in G_1(\mathbb{A}_F^\blacklozenge)K^\infty g 1 ∈ G 1 ( A F ⧫ ) K ∞ , the supports of Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) Z_{t_1}^{\lambda}(\phi_1^\infty)_L(g_1) Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) and Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) Z_{t_2}^{\lambda}(\phi_2^\infty)_L(g_2) Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) have empty intersection. By Proposition 4.5 and [20 Lemma 7.2] , we have for every t 1 ∈ ( F × ) + t_1 \in(F^\times)^+ t 1 ∈ ( F × ) + that
I ~ t 1 ( g 1 ) = vol ♮ ( L ) ( ∑ u ∈ B E f i n ∖ B E ( 0 ) log ∥ u ∥ ⋅ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , u ℓ ) \widetilde{\mathcal{I}}_{t_1}(g_1)=\operatorname{vol}^{\natural}(L)\left(\sum_{u\in\mathbb{B}_E^{\mathrm{fin}}\setminus\mathbb{B}_E^{(0)}}\log\lVert u\rVert\cdot\left\langle Z_{t_1}^{\lambda}(\phi_1^\infty)_L(g_1),Z_{t_2}^{\lambda}(\phi_2^\infty)_L(g_2)\right\rangle_{\lambda,L,u}^{\ell}\right) I t 1 ( g 1 ) = vol ♮ ( L ) u ∈ B E fin ∖ B E ( 0 ) ∑ log ∥ u ∥ ⋅ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L , u ℓ
+ vol ♮ ( L ) ( ∑ u ∈ Hom ( E , C ) ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L u ) , \qquad+\operatorname{vol}^{\natural}(L)\left(\sum_{u\in\operatorname{Hom}(E,\mathbb{C})}\left\langle Z_{t_1}^{\lambda}(\phi_1^\infty)_L(g_1),Z_{t_2}^{\lambda}(\phi_2^\infty)_L(g_2)\right\rangle_{\lambda,L}^{u}\right), + vol ♮ ( L ) u ∈ Hom ( E , C ) ∑ ⟨ Z t 1 λ ( ϕ 1 ∞ ) L ( g 1 ) , Z t 2 λ ( ϕ 2 ∞ ) L ( g 2 ) ⟩ λ , L u ,
where we have suppress the pair Z \mathbb{Z} Z of supports in the subscripts as it is clear from the context.
Denote by ⧫ ( 1 ) \blacklozenge^{(1)} ⧫ ( 1 ) and ⧫ ( 2 ) \blacklozenge^{(2)} ⧫ ( 2 ) the subsets of ⧫ \blacklozenge ⧫ consisting of v v v that is nonsplit in E E E such that V v V_v V v is split and nonsplit, respectively. For every t 1 ∈ ( F × ) + t_1 \in(F^\times)^+ t 1 ∈ ( F × ) + , put
K t 1 ( g 1 ) : = ∑ v ∈ B F ∖ ( ⧫ ( 1 ) ∪ ⧫ ( 2 ) ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v \mathcal{K}_{t_1}(g_1):=\sum_{v\in\mathbb{B}_F\setminus(\blacklozenge^{(1)}\cup\blacklozenge^{(2)})}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi_{\infty}^{\lambda}\cdot\Phi^\infty)_v K t 1 ( g 1 ) := v ∈ B F ∖ ( ⧫ ( 1 ) ∪ ⧫ ( 2 ) ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v
+ ∑ v ∈ ⧫ ( 1 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v , Φ v ⋆ ) \qquad+\sum_{v\in\blacklozenge^{(1)}}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi_{\infty}^{\lambda}\cdot\Phi^{\infty,v},\Phi_v^\star) + v ∈ ⧫ ( 1 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v , Φ v ⋆ )
+ ∑ v ∈ ⧫ ( 2 ) ∑ x 1 ⋆ ∈ V ( v ) t 1 ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ⋆ ) ] ⋅ ϕ V v ∞ 0 ) ( x 1 ⋆ ) \qquad+\sum_{v\in\blacklozenge^{(2)}}\sum_{x_1^\star\in V(v)_{t_1}}\omega_{1,\infty}(g_1)\left(P[\phi_{t_2}^{\infty}(x_1^\star)]\cdot\phi_{V_v^\infty}^{0}\right)(x_1^\star) + v ∈ ⧫ ( 2 ) ∑ x 1 ⋆ ∈ V ( v ) t 1 ∑ ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ⋆ )] ⋅ ϕ V v ∞ 0 ) ( x 1 ⋆ )
+ ∑ u ∈ Hom ( E , C ) ( φ t 2 F M ) t 1 ( g 1 ) , \qquad+\sum_{u\in\operatorname{Hom}(E,\mathbb{C})}(\varphi_{t_2}^{\mathrm{FM}})_{t_1}(g_1), + u ∈ Hom ( E , C ) ∑ ( φ t 2 FM ) t 1 ( g 1 ) ,
in which the terms Φ v ⋆ \Phi_v^\star Φ v ⋆ , P [ ϕ t 2 ∞ ( x 1 ⋆ ) ] P[\phi_{t_2}^{\infty}(x_1^\star)] P [ ϕ t 2 ∞ ( x 1 ⋆ )] , and φ t 2 F M \varphi_{t_2}^{\mathrm{FM}} φ t 2 FM are from Proposition 5.9 (1), Proposition 5.9 (2), and Proposition 5.27 , respectively. By Proposition 5.2 (1) (plus Remark 5.3), Proposition 5.5, Proposition 5.9 and Proposition 5.27 , we have
I ~ t 1 ( g 1 ) = K t 1 ( g 1 ) \widetilde{\mathcal{I}}_{t_1}(g_1)=\mathcal{K}_{t_1}(g_1) I t 1 ( g 1 ) = K t 1 ( g 1 )
for every t 1 ∈ ( F × ) + t_1 \in(F^\times)^+ t 1 ∈ ( F × ) + and g 1 ∈ G 1 ( A F ⧫ ) g_1 \in G_1(\mathbb{A}_F^\blacklozenge) g 1 ∈ G 1 ( A F ⧫ ) . Thus, it is equivalent to showing that
∑ t 1 ∈ ( F × ) + K t 1 ( g 1 ) (66) \sum_{t_1\in(F^\times)^+}\mathcal{K}_{t_1}(g_1) \tag*{(66)} t 1 ∈ ( F × ) + ∑ K t 1 ( g 1 ) (66)
is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) . We analyze term by term.
(1) By Proposition 2.20(2,3), we know that
∑ t 1 ∈ ( F × ) + ∑ v ∈ V F E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v \sum_{t_1\in(F^\times)^+}\sum_{v\in V_F}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v t 1 ∈ ( F × ) + ∑ v ∈ V F ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v
is the (full) Fourier expansion of E ′ ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) t 2 E'(0,(g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_{t_2} E ′ ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) t 2 (regarded as an automorphic form on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) in the g 1 g_1 g 1 -variable), hence is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) . On the other hand, by Lemma 2.21(2), we know that
∑ t 1 ∈ ( F × ) + ∑ v ∈ V F ( 1 ) ∪ V F ( 2 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ α ⋅ Φ ∞ ) v \sum_{t_1\in(F^\times)^+}\sum_{v\in V_F^{(1)}\cup V_F^{(2)}}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\alpha_\infty\cdot\Phi^\infty)_v t 1 ∈ ( F × ) + ∑ v ∈ V F ( 1 ) ∪ V F ( 2 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ α ⋅ Φ ∞ ) v
is the (full) Fourier expansion of a doubling Eisenstein series (regarded as an automorphic form on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) in the g 1 g_1 g 1 -variable), hence is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) .
(2) For every v ∈ V F ( 1 ) v\in V_F^{(1)} v ∈ V F ( 1 ) ,
∑ t 1 ∈ ( F × ) + E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) \sum_{t_1\in(F^\times)^+}E_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi^\star_v) t 1 ∈ ( F × ) + ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ )
is the (full) Fourier expansion of E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 E(0,(g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi^\star_v)_{t_2} E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 (regarded as an automorphic form on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) in the g 1 g_1 g 1 -variable), hence is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) .
(3) For every v ∈ V F ( 2 ) v\in V_F^{(2)} v ∈ V F ( 2 ) ,
∑ t 1 ∈ ( F × ) + ∑ x 1 ∗ ∈ V ( v ) t 1 ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ∗ ) ] ⋅ ϕ V ∞ 0 ) ( x 1 ∗ ) \sum_{t_1\in(F^\times)^+}\sum_{x_1^\ast\in V(v)_{t_1}}\omega_{1,\infty}(g_1)\bigl(P[\phi^\infty_{t_2}(x_1^\ast)]\cdot\phi^0_{V_\infty}\bigr)(x_1^\ast) t 1 ∈ ( F × ) + ∑ x 1 ∗ ∈ V ( v ) t 1 ∑ ω 1 , ∞ ( g 1 ) ( P [ ϕ t 2 ∞ ( x 1 ∗ )] ⋅ ϕ V ∞ 0 ) ( x 1 ∗ )
is the (full) Fourier expansion of the theta series defined by the Schwartz function
x ⟼ ( P [ ϕ t 2 ∞ ( x ∞ ) ] ⋅ ϕ V ∞ 0 ) ( x ∞ ) , x\longmapsto\bigl(P[\phi^\infty_{t_2}(x^\infty)]\cdot\phi^0_{V_\infty}\bigr)(x_\infty), x ⟼ ( P [ ϕ t 2 ∞ ( x ∞ )] ⋅ ϕ V ∞ 0 ) ( x ∞ ) ,
hence is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) .
(4) For every u ∈ Hom ( E , C ) u\in\operatorname{Hom}(E,\mathbb{C}) u ∈ Hom ( E , C ) ,
∑ t 1 ∈ ( F × ) + ( φ t 2 F M ) t 1 ( g 1 ) \sum_{t_1\in(F^\times)^+}(\varphi^{\mathrm{FM}}_{t_2})_{t_1}(g_1) t 1 ∈ ( F × ) + ∑ ( φ t 2 FM ) t 1 ( g 1 )
is the (full) Fourier expansion of φ t 2 F M \varphi^{\mathrm{FM}}_{t_2} φ t 2 FM by Proposition 5.27 (b), hence is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) .
Combining the four parts together, it follows that (66) is absolutely convergent and extends to a function on G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) that is left invariant under G 1 ( F ) G_1(F) G 1 ( F ) .
The theorem is proved. □
Central derivatives of L L L -functions
In this section, we consider a holomorphic tempered cuspidal automorphic representation ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ (Definition 2.27 , also Remark 5.26).
Let ( ⋄ , L ⋄ , ⋄ ) (\diamond,L^\diamond,\diamond) ( ⋄ , L ⋄ , ⋄ ) be as in § 6.1 such that ( π v ) K 1 , v ≠ { 0 } (\pi_v)^{K_{1,v}}\ne\{0\} ( π v ) K 1 , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ( 0 ) v\in V_F^{\mathrm{fin}}\setminus V_F^{(0)} v ∈ V F fin ∖ V F ( 0 ) . We define a functional
Z : ( A π ∩ A λ + ( G 1 ) c u s p h o l ) 2 × S ( ( V ∞ ) 2 , C ) L ( ⋄ ) ⋄ ⟶ C \mathcal{Z}:\bigl(\mathcal{A}_\pi\cap\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}}\bigr)^2\times\mathcal{S}((V^\infty)^2,\mathbb{C})^{L^\diamond_{(\diamond)}}\longrightarrow\mathbb{C} Z : ( A π ∩ A λ + ( G 1 ) cusp hol ) 2 × S (( V ∞ ) 2 , C ) L ( ⋄ ) ⋄ ⟶ C
(Definition 5.29 ) by the formula
Z ( φ , Φ ∞ ) : = ∫ G 1 ( F ) \ G 1 ( A F ) ∫ G 1 ( F ) \ G 1 ( A F ) I ~ ( Φ ∞ ) ( g 1 , g 2 ) ⋅ φ 1 ( g 1 ) ‾ φ 2 ( g 2 ) d g 1 d g 2 (67) \mathcal{Z}(\varphi,\Phi^\infty):=\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\widetilde{\mathcal{I}}(\Phi^\infty)(g_1,g_2)\cdot\overline{\varphi_1(g_1)}\varphi_2(g_2)\,dg_1\,dg_2 \tag*{(67)} Z ( φ , Φ ∞ ) := ∫ G 1 ( F ) \ G 1 ( A F ) ∫ G 1 ( F ) \ G 1 ( A F ) I ( Φ ∞ ) ( g 1 , g 2 ) ⋅ φ 1 ( g 1 ) φ 2 ( g 2 ) d g 1 d g 2 (67)
for φ = ( φ 1 , φ 2 ) ∈ ( A π ∩ A + λ ( G 1 ) c u s p h o l ) 2 \varphi=(\varphi_1,\varphi_2)\in(\mathcal{A}_{\pi}\cap\mathcal{A}_{+}^{\lambda}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}})^2 φ = ( φ 1 , φ 2 ) ∈ ( A π ∩ A + λ ( G 1 ) cusp hol ) 2 and Φ ∞ ∈ S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ \Phi^{\infty}\in\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}} Φ ∞ ∈ S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , where V \mathcal{V} V is from Theorem 6.1 . It is clear that Z \mathcal{Z} Z is an element in
Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) . \operatorname{Hom}_{G_1(\mathbb{A}^{\infty}_{F})^2}\left(\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}},\pi^{\infty}\boxtimes(\pi^{\infty})^{\vee}\right). Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) .
Lemma 6.2. We have
dim C Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ 1. \dim_{\mathbb{C}}\operatorname{Hom}_{G_1(\mathbb{A}^{\infty}_{F})^2}\left(\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}},\pi^{\infty}\boxtimes(\pi^{\infty})^{\vee}\right)\leq1. dim C Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ 1. Moreover, it equals one if and only if V = V ( π ) V=V(\pi) V = V ( π ) (Remark 4.12(2)).
Proof. Since the space S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ \mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}} S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ admits a G 1 ( A F ∞ ) 2 G_1(\mathbb{A}^{\infty}_{F})^2 G 1 ( A F ∞ ) 2 -invariant inner product, S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ \mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}} S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ is a G 1 ( A F ∞ ) 2 G_1(\mathbb{A}^{\infty}_{F})^2 G 1 ( A F ∞ ) 2 -equivariant direct summand of S ( ( V ∞ ) 2 , C ) L ⋄ \mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}} S (( V ∞ ) 2 , C ) L ⋄ . It follows from [26 Proposition 6.7] or [20 Proposition 3.6] that
dim C Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ dim C Hom G 1 ( A F ∞ ) 2 ( S ( ( V ∞ ) 2 , C ) L ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ 1 , \dim_{\mathbb{C}}\operatorname{Hom}_{G_1(\mathbb{A}^{\infty}_{F})^2}\left(\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}},\pi^{\infty}\boxtimes(\pi^{\infty})^{\vee}\right)\leq\dim_{\mathbb{C}}\operatorname{Hom}_{G_1(\mathbb{A}^{\infty}_{F})^2}\left(\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}},\pi^{\infty}\boxtimes(\pi^{\infty})^{\vee}\right)\leq1, dim C Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ dim C Hom G 1 ( A F ∞ ) 2 ( S (( V ∞ ) 2 , C ) L ⋄ , π ∞ ⊠ ( π ∞ ) ∨ ) ≤ 1 ,
and it equals zero if V ≠ V ( π ) V\ne V(\pi) V = V ( π ) . When V = V ( π ) V=V(\pi) V = V ( π ) , the above Hom space is nonzero by Lemma 5.30. The lemma is proved. □
Theorem 6.3. Let ( π , A π ) (\pi,\mathcal{A}_{\pi}) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) G_1(\mathbb{A}_{F}) G 1 ( A F ) of weight λ \lambda λ . For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) of automorphic forms in A π ∩ A + λ ( G 1 ) c u s p h o l \mathcal{A}_{\pi}\cap\mathcal{A}_{+}^{\lambda}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} A π ∩ A + λ ( G 1 ) cusp hol and every Φ ∞ ∈ S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ \Phi^{\infty}\in\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}} Φ ∞ ∈ S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ ,
Z ( φ , Φ ∞ ) = L ′ ( 1 2 , π ) b 2 ( 0 ) ⋅ ( π 4 ) [ F : Q ] ⋅ Z f i n ( φ , Φ ∞ ) \mathcal{Z}(\varphi,\Phi^{\infty})=\frac{L'(\frac{1}{2},\pi)}{b_2(0)}\cdot\left(\frac{\pi}{4}\right)^{[F:\mathbb{Q}]}\cdot\mathcal{Z}^{\mathrm{fin}}(\varphi,\Phi^{\infty}) Z ( φ , Φ ∞ ) = b 2 ( 0 ) L ′ ( 2 1 , π ) ⋅ ( 4 π ) [ F : Q ] ⋅ Z fin ( φ , Φ ∞ ) (Definition 2.28 ) holds.
Proof. By Lemma 6.2, it suffices to consider the case where V = V ( π ) V=V(\pi) V = V ( π ) since otherwise both sides vanish. In particular, we may assume ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 (Remark 4.12(2)). Similar to the proof of Theorem 6.1 , it suffices to consider the case for Φ ∞ = φ 1 ∞ ⊗ φ 2 ∞ ‾ ∈ S ( ( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ \Phi^{\infty}=\varphi_1^{\infty}\otimes\overline{\varphi_2^{\infty}}\in\mathcal{S}((V^{\infty})^2,\mathbb{C})^{L^{\diamond}_{\langle\diamond\rangle}} Φ ∞ = φ 1 ∞ ⊗ φ 2 ∞ ∈ S (( V ∞ ) 2 , C ) L ⟨ ⋄ ⟩ ⋄ with φ 1 ∞ \varphi_1^{\infty} φ 1 ∞ and φ 2 ∞ \varphi_2^{\infty} φ 2 ∞ in S ( V ∞ , C ) L ⋄ \mathcal{S}(V^{\infty},\mathbb{C})^{L^{\diamond}} S ( V ∞ , C ) L ⋄ .
Take elements t 2 ∈ ( F × ) + t_2\in(F^{\times})^{+} t 2 ∈ ( F × ) + and g 2 ∈ G 1 ( A F ⋄ ) g_2\in G_1(\mathbb{A}^{\diamond}_{F}) g 2 ∈ G 1 ( A F ⋄ ) . Put I ~ t 2 , g 2 ( Φ ∞ ) ( g 1 ) : = I ( Φ ∞ ) t 2 ( g 1 , g 2 ) \widetilde{\mathcal{I}}_{t_2,g_2}(\Phi^{\infty})(g_1):=\mathcal{I}(\Phi^{\infty})_{t_2}(g_1,g_2) I t 2 , g 2 ( Φ ∞ ) ( g 1 ) := I ( Φ ∞ ) t 2 ( g 1 , g 2 ) as a function in g 1 ∈ G 1 ( A F ) g_1\in G_1(\mathbb{A}_{F}) g 1 ∈ G 1 ( A F ) , so that we have
I ~ t 2 , g 2 ( Φ ∞ ) ( g 1 ) = ∑ t 1 ∈ ( F × ) + I ~ t 2 , g 2 ( Φ ∞ ) t 1 ( g 1 ) . \widetilde{\mathcal{I}}_{t_2,g_2}(\Phi^{\infty})(g_1)=\sum_{t_1\in(F^{\times})^{+}}\widetilde{\mathcal{I}}_{t_2,g_2}(\Phi^{\infty})_{t_1}(g_1). I t 2 , g 2 ( Φ ∞ ) ( g 1 ) = t 1 ∈ ( F × ) + ∑ I t 2 , g 2 ( Φ ∞ ) t 1 ( g 1 ) .
As in the proof of Theorem 6.1 , we denote by ⋄ ( 1 ) \diamond^{(1)} ⋄ ( 1 ) and ⋄ ( 2 ) \diamond^{(2)} ⋄ ( 2 ) the subsets of ⋄ \diamond ⋄ consisting of v v v that is nonsplit in E E E such that V v V_v V v is split and nonsplit, respectively. Take an element v ∈ ⋄ ( 2 ) v\in\diamond^{(2)} v ∈ ⋄ ( 2 ) , by the theory of theta series, there exists a unique element θ t 2 , g 2 ( v ) ( Φ ∞ ) ∈ A + λ ( G 1 ) c u s p h o l \theta^{(v)}_{t_2,g_2}(\Phi^{\infty})\in\mathcal{A}_{+}^{\lambda}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} θ t 2 , g 2 ( v ) ( Φ ∞ ) ∈ A + λ ( G 1 ) cusp hol (which is a theta series) such that for g 1 ∈ G 1 ( A F ⋄ ) g_1\in G_1(\mathbb{A}^{\diamond}_{F}) g 1 ∈ G 1 ( A F ⋄ ) ,
θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) = ∑ x 1 ∗ ∈ V ( v ) ω 1 , ∞ ( g 1 , ∞ ) ( [ P φ 1 ∞ ( g 1 ∞ ) φ 2 ∞ ( x 1 ∗ ) ] ⋅ φ V ∞ 0 ) ( x 1 ∗ ) \theta^{(v)}_{t_2,g_2}(\Phi^{\infty})(g_1)=\sum_{x_1^{*}\in V(v)}\omega_{1,\infty}(g_{1,\infty})\left([P\varphi_1^{\infty}(g_1^{\infty})\varphi_2^{\infty}(x_1^{*})]\cdot\varphi_{V_{\infty}}^{0}\right)(x_1^{*}) θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) = x 1 ∗ ∈ V ( v ) ∑ ω 1 , ∞ ( g 1 , ∞ ) ( [ P φ 1 ∞ ( g 1 ∞ ) φ 2 ∞ ( x 1 ∗ )] ⋅ φ V ∞ 0 ) ( x 1 ∗ )
from Proposition 5.22 .37 Then by Proposition 5.2 (1) (plus Remark 5.3), Proposition 5.5, Proposition 5.9 (1), Proposition 5.22 and Proposition 5.28 , we have
∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ S ~ t 2 , g 2 ( Φ ∞ ) ( g 1 ) d g 1 \int_{G_1(F)\backslash G_1(\mathbb{A}_F)} \overline{\varphi_1(g_1)}\widetilde{\mathcal{S}}_{t_2,g_2}(\Phi^\infty)(g_1)\,\mathrm{d}g_1 ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) S t 2 , g 2 ( Φ ∞ ) ( g 1 ) d g 1
= ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 + ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 + ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ v ∈ ◊ ( 2 ) θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 . \begin{aligned}
={}&\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{t_1\in(F^\times)^+}\sum_{v\in\Diamond^{(1)}\cup\Diamond^{(2)}}
\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v\,\mathrm{d}g_1\\
&+\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{t_1\in(F^\times)^+}\sum_{v\in\Diamond^{(1)}}
E_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star)\,\mathrm{d}g_1\\
&+\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{v\in\Diamond^{(2)}}\theta^{(v)}_{t_2,g_2}(\Phi^\infty)(g_1)\,\mathrm{d}g_1.
\end{aligned} = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 + ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 + ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) v ∈ ◊ ( 2 ) ∑ θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 .
in which the terms Φ v ⋆ \Phi_v^\star Φ v ⋆ is from Proposition 5.9 (1).
We have three sorts of vanishing:
first, by Lemma 2.21(2), for every v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) v\in\Diamond^{(1)}\cup\Diamond^{(2)} v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) , there exists a regularly supported Schwartz function Φ v ⋆ ∈ S ( ( V v ⋆ ) 2 ) \Phi_v^\star\in\mathcal{S}((V_v^\star)^2) Φ v ⋆ ∈ S (( V v ⋆ ) 2 ) such that
∑ t 1 ∈ ( F × ) + E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v = ∑ t 1 ∈ ( F × ) + E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) , \sum_{t_1\in(F^\times)^+}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v
=
\sum_{t_1\in(F^\times)^+}E_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star), t 1 ∈ ( F × ) + ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v = t 1 ∈ ( F × ) + ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) ,
which implies that
∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 = ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 = ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ⋅ E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 d g 1 , \begin{aligned}
&\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{t_1\in(F^\times)^+}\sum_{v\in\Diamond^{(1)}\cup\Diamond^{(2)}}
\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v\,\mathrm{d}g_1\\
={}&\sum_{v\in\Diamond^{(1)}\cup\Diamond^{(2)}}\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}
\overline{\varphi_1(g_1)}
\sum_{t_1\in(F^\times)^+}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v\,\mathrm{d}g_1\\
={}&\sum_{v\in\Diamond^{(1)}\cup\Diamond^{(2)}}\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}
\overline{\varphi_1(g_1)}\cdot E(0,(g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star)_{t_2}\,\mathrm{d}g_1,
\end{aligned} = = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∑ ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 v ∈ ◊ ( 1 ) ∪ ◊ ( 2 ) ∑ ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ⋅ E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 d g 1 ,
in which every integration in the last summation vanishes as ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 ;
second,
∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ v ∈ ◊ ( 1 ) ∑ t 1 ∈ ( F × ) + E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 = ∑ v ∈ ◊ ( 1 ) ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ⋅ E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 d g 1 , \begin{aligned}
&\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{t_1\in(F^\times)^+}\sum_{v\in\Diamond^{(1)}}
E_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star)\,\mathrm{d}g_1\\
={}&\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{v\in\Diamond^{(1)}}\sum_{t_1\in(F^\times)^+}
E_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star)\,\mathrm{d}g_1\\
={}&\sum_{v\in\Diamond^{(1)}}\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}
\overline{\varphi_1(g_1)}\cdot E(0,(g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^{\infty,v}\cdot\Phi_v^\star)_{t_2}\,\mathrm{d}g_1,
\end{aligned} = = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ v ∈ ◊ ( 1 ) ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) v ∈ ◊ ( 1 ) ∑ t 1 ∈ ( F × ) + ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) d g 1 v ∈ ◊ ( 1 ) ∑ ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ⋅ E ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ , v ⋅ Φ v ⋆ ) t 2 d g 1 ,
in which every integration in the last summation vanishes as ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 ;
third,
∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ v ∈ ◊ ( 2 ) θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 = ∑ v ∈ ◊ ( 2 ) ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ⋅ θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 , \int_{G_1(F)\backslash G_1(\mathbb{A}_F)}\overline{\varphi_1(g_1)}
\sum_{v\in\Diamond^{(2)}}\theta^{(v)}_{t_2,g_2}(\Phi^\infty)(g_1)\,\mathrm{d}g_1
=
\sum_{v\in\Diamond^{(2)}}\int_{G_1(F)\backslash G_1(\mathbb{A}_F)}
\overline{\varphi_1(g_1)}\cdot\theta^{(v)}_{t_2,g_2}(\Phi^\infty)(g_1)\,\mathrm{d}g_1, ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) v ∈ ◊ ( 2 ) ∑ θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 = v ∈ ◊ ( 2 ) ∑ ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ⋅ θ t 2 , g 2 ( v ) ( Φ ∞ ) ( g 1 ) d g 1 ,
in which every integration in the last summation vanishes as ϵ ( π ) = − 1 \epsilon(\pi)=-1 ϵ ( π ) = − 1 .
It follows that
∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ I ~ t 2 , g 2 ( Φ ∞ ) ( g 1 ) d g 1 = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) ‾ ∑ t 1 ∈ ( F × ) + ∑ v ∈ V F E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 . \begin{aligned}
\int_{G_1(F)\backslash G_1(\mathbb{A}_F)} \overline{\varphi_1(g_1)}\widetilde{\mathcal{I}}_{t_2,g_2}(\Phi^\infty)(g_1)\,\mathrm{d}g_1 \\
&=\int_{G_1(F)\backslash G_1(\mathbb{A}_F)} \overline{\varphi_1(g_1)}\sum_{t_1\in(F^\times)^+}\sum_{v\in\mathcal{V}_F}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v\,\mathrm{d}g_1.
\end{aligned} ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) I t 2 , g 2 ( Φ ∞ ) ( g 1 ) d g 1 = ∫ G 1 ( F ) \ G 1 ( A F ) φ 1 ( g 1 ) t 1 ∈ ( F × ) + ∑ v ∈ V F ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v d g 1 .
If we sum over all t 2 ∈ ( F + ) × t_2\in(F^+)^\times t 2 ∈ ( F + ) × and take the convolution against φ 2 \varphi_2 φ 2 , we obtain that
Z ( φ , Φ ∞ ) = ∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 ∑ ( t 1 , t 2 ) ∈ ( ( F × ) + ) 2 ∑ v ∈ V F E t 1 , t 2 ( ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v ⋅ φ 1 ( g 1 ) ‾ φ 2 ( g 2 ) d g 1 d g 2 , \mathcal{Z}(\varphi,\Phi^\infty)=\iint_{(G_1(F)\backslash G_1(\mathbb{A}_F))^2}\sum_{(t_1,t_2)\in((F^\times)^+)^2}\sum_{v\in\mathcal{V}_F}\mathcal{E}_{t_1,t_2}((g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)_v\cdot\overline{\varphi_1(g_1)}\varphi_2(g_2)\,\mathrm{d}g_1\,\mathrm{d}g_2, Z ( φ , Φ ∞ ) = ∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 ( t 1 , t 2 ) ∈ (( F × ) + ) 2 ∑ v ∈ V F ∑ E t 1 , t 2 (( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) v ⋅ φ 1 ( g 1 ) φ 2 ( g 2 ) d g 1 d g 2 ,
which by Proposition 2.20(2,3) equals
∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 E ′ ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) ‾ ⋅ φ 1 ( g 1 ) ‾ φ 2 ( g 2 ) d g 1 d g 2 . \iint_{(G_1(F)\backslash G_1(\mathbb{A}_F))^2}\overline{E'(0,(g_1,g_2),\Phi^\lambda_\infty\cdot\Phi^\infty)}\cdot\overline{\varphi_1(g_1)}\varphi_2(g_2)\,\mathrm{d}g_1\,\mathrm{d}g_2. ∬ ( G 1 ( F ) \ G 1 ( A F ) ) 2 E ′ ( 0 , ( g 1 , g 2 ) , Φ ∞ λ ⋅ Φ ∞ ) ⋅ φ 1 ( g 1 ) φ 2 ( g 2 ) d g 1 d g 2 .
Then the theorem follows from Proposition 2.29 and Remark 2.30(2). □
Definition 6.4. For every intermediate field E λ ⊆ L ⊆ C E^\lambda\subseteq L\subseteq\mathbb{C} E λ ⊆ L ⊆ C and every object L L L of Hk ( H ( A F ∞ ) ) \operatorname{Hk}(H(\mathbb{A}_F^\infty)) Hk ( H ( A F ∞ )) , we
define SC λ ( X L , L ) \operatorname{SC}_\lambda(X_L,\mathbb{L}) SC λ ( X L , L ) to be the subspace of H 2 ( X L , M L , λ ) ⊗ E λ L H^2(X_L,M_{L,\lambda})\otimes_{E^\lambda}L H 2 ( X L , M L , λ ) ⊗ E λ L spanned by motivic special classes z λ ( x ) L z^\lambda(x)_L z λ ( x ) L for x ∈ V ∞ x\in V^\infty x ∈ V ∞ satisfying T ( x ) ∈ F + T(x)\in F^+ T ( x ) ∈ F + ;
denote by SC λ ∘ ( X L , L ) ⊆ SC λ ( X L , L ) \operatorname{SC}^\circ_\lambda(X_L,\mathbb{L})\subseteq\operatorname{SC}_\lambda(X_L,\mathbb{L}) SC λ ∘ ( X L , L ) ⊆ SC λ ( X L , L ) the subspace contained in the radical of the height pairing
⟨ , ⟩ λ , L : H 2 ( X L , M L , λ ) ⊗ E λ L × H 2 ( X L , M L , λ ) ⊗ E λ L ⟶ C \langle\ ,\ \rangle_{\lambda,L}:H^2(X_L,M_{L,\lambda})\otimes_{E^\lambda}L\times H^2(X_L,M_{L,\lambda})\otimes_{E^\lambda}L\longrightarrow\mathbb{C} ⟨ , ⟩ λ , L : H 2 ( X L , M L , λ ) ⊗ E λ L × H 2 ( X L , M L , λ ) ⊗ E λ L ⟶ C extended to a hermitian form with coefficients in L L L ;
put
SC ‾ λ ( X L , L ) : = SC λ ( X L , L ) SC λ ∘ ( X L , L ) , \overline{\operatorname{SC}}_\lambda(X_L,\mathbb{L}):=\frac{\operatorname{SC}_\lambda(X_L,\mathbb{L})}{\operatorname{SC}^\circ_\lambda(X_L,\mathbb{L})}, SC λ ( X L , L ) := SC λ ∘ ( X L , L ) SC λ ( X L , L ) , which receives a natural action by L [ L \ H ( A F ∞ ) / L ] L[L\backslash H(\mathbb{A}_F^\infty)/L] L [ L \ H ( A F ∞ ) / L ] .
Corollary 6.5. Let ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ such that V ( π ) V(\pi) V ( π ) (Remark 4.12(2)) is incoherent (Definition 2.15) satisfying ∣ V ( V ( π ) ) ∣ > 1 |\mathcal{V}(V(\pi))|>1 ∣ V ( V ( π )) ∣ > 1 . Let E λ ⊆ L π ⊆ C E^\lambda\subseteq L_\pi\subseteq\mathbb{C} E λ ⊆ L π ⊆ C be the field of definition of π ∞ \pi^\infty π ∞ . If L ′ ( 1 2 , π ) ≠ 0 L'(\frac{1}{2},\pi)\ne0 L ′ ( 2 1 , π ) = 0 , then for V = V ( π ) V=V(\pi) V = V ( π ) , every finite set ⋄ \diamond ⋄ of rational primes containing all those ramified in $E such that ( π v ) K 1 , v ≠ { 0 } (\pi_v)_{K_{1,v}}\ne\{0\} ( π v ) K 1 , v = { 0 } (Notation 2.13(1)) for every v ∈ V F f i n ∖ V F ⋄ v\in\mathcal{V}_F^{\mathrm{fin}}\setminus\mathcal{V}_F^\diamond v ∈ V F fin ∖ V F ⋄ , and every open compact subgroup L ⋄ L^\diamond L ⋄ of H(\mathbb{A}_F^{\infty}) that is the stabilizer of a self-dual lattice of V_F^{\infty,\diamond}$,
l i m → L ⋄ SC ‾ λ ( X L ⋄ L ⋄ , L π ) m π ≠ 0 \varinjlim_{L^\diamond}\overline{\operatorname{SC}}_\lambda(X_{L^\diamond L^\diamond},L_\pi)_{m_\pi}\ne0 L ⋄ lim SC λ ( X L ⋄ L ⋄ , L π ) m π = 0 holds, where m π : = ker χ π ⊆ L π [ L ⋄ \ H ( A F ∞ ) / L ⋄ ] m_\pi:=\ker\chi_\pi\subseteq L_\pi[L^\diamond\backslash H(\mathbb{A}_F^\infty)/L^\diamond] m π := ker χ π ⊆ L π [ L ⋄ \ H ( A F ∞ ) / L ⋄ ] from (42) .
Proof. This follows from Theorem 6.3 and Lemma 5.30. □
The corollary below confirms the (conjectural) motivic arithmetic inner product formula (Conjecture 4.11) for r = 1 r=1 r = 1 , assuming the modularity of motivic generating functions (Conjecture 3.36).
Corollary 6.6. Let ( π , A π ) (\pi,\mathcal{A}_\pi) ( π , A π ) be a holomorphic tempered cuspidal automorphic representation of G 1 ( A F ) G_1(\mathbb{A}_F) G 1 ( A F ) of weight λ \lambda λ . Assume Conjecture 3.36 for V V V so that one can apply Definition 3.38. For every pair φ = ( φ 1 , φ 2 ) \varphi=(\varphi_1,\varphi_2) φ = ( φ 1 , φ 2 ) of automorphic forms in A π ∩ A λ + ( G 1 ) c u s p h o l \mathcal{A}_\pi\cap\mathcal{A}_{\lambda+}(G_1)_{\mathrm{cusp}}^{\mathrm{hol}} A π ∩ A λ + ( G 1 ) cusp hol , and very pair φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) \varphi^\infty=(\varphi_1^\infty,\varphi_2^\infty) φ ∞ = ( φ 1 ∞ , φ 2 ∞ ) of Schwartz functions in S ( V ∞ ) \mathcal{S}(V^\infty) S ( V ∞ ) ,
⟨ Θ φ 1 ∞ ( φ 1 ) , Θ φ 2 ∞ ( φ 2 ) ⟩ λ = L ′ ( 1 2 , π ) b 2 ( 0 ) ⋅ ( − π 4 ) [ F : Q ] ⋅ Z f i n 3 ( φ , φ 1 ∞ ⊗ φ 2 ∞ ) \left\langle\Theta_{\varphi_1^\infty}(\varphi_1),\Theta_{\varphi_2^\infty}(\varphi_2)\right\rangle_\lambda
=\frac{L'(\frac{1}{2},\pi)}{b_2(0)}\cdot\left(-\frac{\pi}{4}\right)^{[F:\mathbb{Q}]}\cdot\mathcal{Z}_{\mathrm{fin}}^3(\varphi,\varphi_1^\infty\otimes\varphi_2^\infty) ⟨ Θ φ 1 ∞ ( φ 1 ) , Θ φ 2 ∞ ( φ 2 ) ⟩ λ = b 2 ( 0 ) L ′ ( 2 1 , π ) ⋅ ( − 4 π ) [ F : Q ] ⋅ Z fin 3 ( φ , φ 1 ∞ ⊗ φ 2 ∞ ) holds.
Proof. By [26 Proposition 6.7] or [20 Proposition 3.6] , it suffices to consider the case where V = V ( π ) V = V(\pi) V = V ( π ) since otherwise both sides vanish; and in this case, it suffices to show the identity in the statement for a single choice of data with Z f i n ▲ ( φ , φ 1 0 ⊗ φ 2 0 ) ≠ 0 \mathcal{Z}_{\mathrm{fin}}^{\blacktriangle}(\varphi,\varphi_1^0\otimes\varphi_2^0) \ne0 Z fin ▲ ( φ , φ 1 0 ⊗ φ 2 0 ) = 0 . Then the corollary follows from Theorem 6.3 and Lemma 5.30 . □
Remark 6.7. In this remark, we explain how to modify our theory when E E E is a general CM field.
We fix an abstract CM data of E E E , which is a pair ( E ~ / E , tr ) (\widetilde{E}/E,\operatorname{tr}) ( E / E , tr ) in which
E ~ / E \widetilde{E}/E E / E is a finite extension with E ~ \widetilde{E} E also a CM field; and
tr : E → E ~ \operatorname{tr}\colon E \to\widetilde{E} tr : E → E is a function such that for some hence every embedding u ~ : E ~ → C \widetilde{u}\colon\widetilde{E} \to\mathbb{C} u : E → C , u ~ ∘ tr \widetilde{u}\circ\operatorname{tr} u ∘ tr is the trace function of a CM type of E E E .
Note that the case in the main text corresponds to the abstract CM data in which E ~ = E \widetilde{E}=E E = E and tr \operatorname{tr} tr is the composition of tr E / E 0 \operatorname{tr}_{E/E_0} tr E / E 0 and the inclusion E 0 ↪ E E_0\hookrightarrow E E 0 ↪ E .
We modify the functor Y ∙ Y_\bullet Y ∙ in Definition 3.1 to classify triples ( A 0 , λ 0 , η ‾ 0 ) (A_0,\lambda_0,\overline{\eta}_0) ( A 0 , λ 0 , η 0 ) over E ~ \widetilde{E} E in which A 0 A_0 A 0 is an object of AV ( S , E ) \operatorname{AV}(S,E) AV ( S , E ) (for S ∈ Sch / E ~ ′ S\in\operatorname{Sch}^{\prime}_{/\widetilde{E}} S ∈ Sch / E ′ ) such that tr A 0 ( e ) = tr ( e ) \operatorname{tr}_{A_0}(e)=\operatorname{tr}(e) tr A 0 ( e ) = tr ( e ) for e ∈ E e\in E e ∈ E ; and the others are similar.38 We then modify the functor X ∙ , ∙ X_{\bullet,\bullet} X ∙ , ∙ in Definition 3.3 to classify sextuples ( A 0 , λ 0 , η ‾ 0 ; A , λ , η ‾ ) (A_0,\lambda_0,\overline{\eta}_0;A,\lambda,\overline{\eta}) ( A 0 , λ 0 , η 0 ; A , λ , η ) over E ~ \widetilde{E} E (rather than E E E ) in which A A A is an object of AV ( S , E ) \operatorname{AV}(S,E) AV ( S , E ) (for S ∈ Sch / E ~ ′ S\in\operatorname{Sch}^{\prime}_{/\widetilde{E}} S ∈ Sch / E ′ ) such that tr A ( e ) = m tr ( e ) − e + e ‾ \operatorname{tr}_A(e)=m\operatorname{tr}(e)-e+\overline{e} tr A ( e ) = m tr ( e ) − e + e for e ∈ E e\in E e ∈ E ; and the others are similar. By the similar descent process, we obtain a functor
X ∙ : Hk ( H ( A F ∞ ) ) ⟶ Sch / E ~ s m . q p r o j X_\bullet\colon\operatorname{Hk}(H(A_F^\infty))\longrightarrow\operatorname{Sch}^{\mathrm{sm.qproj}}_{/\widetilde{E}} X ∙ : Hk ( H ( A F ∞ )) ⟶ Sch / E sm.qproj
such that for every u ~ ∈ Hom ( E ~ , C ) \widetilde{u}\in\operatorname{Hom}(\widetilde{E},\mathbb{C}) u ∈ Hom ( E , C ) , X ∙ ⊗ E ~ , u ~ E ~ ≃ Sh ( H ( v ) ) μ ⊗ u ( E ) u ~ ( E ~ ) X_\bullet\otimes_{\widetilde{E},\widetilde{u}} \widetilde{E}\simeq\operatorname{Sh}(H(v))^\mu\otimes_{u(E)}\widetilde{u}(\widetilde{E}) X ∙ ⊗ E , u E ≃ Sh ( H ( v ) ) μ ⊗ u ( E ) u ( E ) where v : = u ~ ∣ F v:=\widetilde{u}|_F v := u ∣ F and u : = u ~ ∣ E u:=\widetilde{u}|_E u := u ∣ E , together with a Cartesian functor
M ∙ : Hk ( H ( A F ∞ ) ) ⟶ Mot ( Sch / E ~ s m . q p r o j , E ) M_\bullet\colon\operatorname{Hk}(H(A_F^\infty))\longrightarrow\operatorname{Mot}(\operatorname{Sch}^{\mathrm{sm.qproj}}_{/\widetilde{E}},E) M ∙ : Hk ( H ( A F ∞ )) ⟶ Mot ( Sch / E sm.qproj , E )
above X ∙ X_\bullet X ∙ as the standard motive .
The remaining part is essentially same to the case in the main text, except that in the definition of the height pairing ⟨ , ⟩ λ , L \langle\ ,\ \rangle_{\lambda,L} ⟨ , ⟩ λ , L (38) (or rather ⟨ , ⟩ λ , L \langle\ ,\ \rangle_{\lambda,L} ⟨ , ⟩ λ , L by Theorem 4.15 ) we divide by [ E ~ : E ] [\widetilde{E}:E] [ E : E ] at the end.
Our theory a priori depends on the choice of the abstract CM data. For a general CM field E E E , our approach may not produce an optimal theory over the base field E E E since it is not true that we can always find an abstract CM data with E ~ = E \widetilde{E}=E E = E .
Appendix A. Relative motives
In the appendix, we collect some known facts concerning relative motives.
Relative motives and their realizations
Let L L L be a Q \mathbb{Q} Q -ring that is finite-dimensional as a Q \mathbb{Q} Q -vector space. For every place ⋄ ∈ { ∞ , 2 , 3 , 5 , … } \diamond\in\{\infty,2,3,5,\ldots\} ⋄ ∈ { ∞ , 2 , 3 , 5 , … } of Q \mathbb{Q} Q , put L ⋄ : = L ⊗ Q ⋄ L_\diamond:=L\otimes\mathbb{Q}_\diamond L ⋄ := L ⊗ Q ⋄ .
Notation A.1. For every Noetherian scheme S S S , we have the following.
Denote by Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) the category of relative (Chow) motives over S S S with coefficients in Q \mathbb{Q} Q (see, for example, [31 Definition 8.1.4] 39 ), which is a closed symmetric monoidal Q \mathbb{Q} Q -linear pseudo-abelian category.40 Denote by Mot ( S , L ) \operatorname{Mot}(S,L) Mot ( S , L ) the category of L L L -linear objects in Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) , which is a closed symmetric monoidal L L L -linear pseudo-abelian category.
(2) Denote by D ( S , L ) \mathrm{D}(S,L) D ( S , L ) the category of constructible étale motivic sheaves over S S S with coefficients in L L L (see [7 ] for the theory), which is a closed symmetric monoidal L L L -linear triangulated category whose unit element we denote by 1 S , L \mathbb{1}_{S,L} 1 S , L .41
(3) For every object M M M in D ( S , L ) \mathrm{D}(S,L) D ( S , L ) and integers i , j i,j i , j , we have the motivic cohomology
H i ( S , M ( j ) ) : = Hom D ( S , L ) ( 1 S , L , M [ i ] ( j ) ) H^{i}(S,M(j)) := \operatorname{Hom}_{\mathrm{D}(S,L)}(\mathbb{1}_{S,L},M[i](j)) H i ( S , M ( j )) := Hom D ( S , L ) ( 1 S , L , M [ i ] ( j ))
as an L L L -module, and for every closed subset T ⊆ S T\subseteq S T ⊆ S the motivic cohomology with support
H T i ( S , M ( j ) ) : = H i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j ) ) , H^{i}_{T}(S,M(j)) := H^{i+1}(S,\operatorname{cone}(M\to Ra_{*}a^{*}M)(j)), H T i ( S , M ( j )) := H i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j )) ,
where a : S ∖ T → S a:S\setminus T\to S a : S ∖ T → S denotes the open complement.
Remark A.2. Suppose that S S S is a regular Noetherian scheme.
(1) For every smooth projective morphism f : X → S f:X\to S f : X → S , we have a canonical isomorphism
H 2 r ( S , R f ∗ 1 X , L ( r ) ) ≃ CH r ( X ) ⊗ L H^{2r}(S,Rf_{*}\mathbb{1}_{X,L}(r))\simeq\operatorname{CH}^{r}(X)\otimes L H 2 r ( S , R f ∗ 1 X , L ( r )) ≃ CH r ( X ) ⊗ L
by [Dég18, Remark 1.3.5(4)] (and the fact that the motivic absolutely purity is known for coefficients in Q \mathbb{Q} Q hence L L L ).42
(2) Part (1) implies that the assignment sending ( X , id ) (X,\operatorname{id}) ( X , id ) for every smooth projective morphism f : X → S f:X\to S f : X → S to R f ∗ 1 X , L Rf_{*}\mathbb{1}_{X,L} R f ∗ 1 X , L induces a fully faithful L L L -linear functor Mot ( S , L ) → D ( S , L ) \operatorname{Mot}(S,L)\to\mathrm{D}(S,L) Mot ( S , L ) → D ( S , L ) .
Notation A.3. For every Noetherian scheme S S S and every rational prime ℓ \ell ℓ that is invertible on S S S , we have the following.
(1) Put L ℓ : = L ⊗ Q ℓ L_{\ell}:=L\otimes\mathbb{Q}_{\ell} L ℓ := L ⊗ Q ℓ and denote by D e ˊ t ( S , L ℓ ) \mathrm{D}_{\text{ét}}(S,L_{\ell}) D e ˊ t ( S , L ℓ ) the derived category of L ℓ L_{\ell} L ℓ -linear étale sheaves over S S S , which is a closed symmetric monoidal L ℓ L_{\ell} L ℓ -linear triangulated category whose unit element we denote by 1 S , L ℓ e ˊ t \mathbb{1}^{\text{ét}}_{S,L_{\ell}} 1 S , L ℓ e ˊ t . Denote by D e ˊ t △ ( S , L ℓ ) \mathrm{D}^{\vartriangle}_{\text{ét}}(S,L_{\ell}) D e ˊ t △ ( S , L ℓ ) the heart of D e ˊ t ( S , L ℓ ) \mathrm{D}_{\text{ét}}(S,L_{\ell}) D e ˊ t ( S , L ℓ ) with respect to the usual t-structure.
(2) For every object M M M in D e ˊ t ( S , L ℓ ) \mathrm{D}_{\text{ét}}(S,L_{\ell}) D e ˊ t ( S , L ℓ ) and integers i , j i,j i , j , we have the usual étale cohomology
H e ˊ t i ( S , M ( j ) ) : = Hom D e ˊ t ( S , L ℓ ) ( 1 S , L ℓ e ˊ t , M [ i ] ( j ) ) H^{i}_{\text{ét}}(S,M(j)) := \operatorname{Hom}_{\mathrm{D}_{\text{ét}}(S,L_{\ell})}(\mathbb{1}^{\text{ét}}_{S,L_{\ell}},M[i](j)) H e ˊ t i ( S , M ( j )) := Hom D e ˊ t ( S , L ℓ ) ( 1 S , L ℓ e ˊ t , M [ i ] ( j ))
as an L ℓ L_{\ell} L ℓ -module, and for every closed subset T ⊆ S T\subseteq S T ⊆ S the étale cohomology with support
H e ˊ t , T i ( S , M ( j ) ) : = H e ˊ t i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j ) ) , H^{i}_{\text{ét},T}(S,M(j)) := H^{i+1}_{\text{ét}}(S,\operatorname{cone}(M\to Ra_{*}a^{*}M)(j)), H e ˊ t , T i ( S , M ( j )) := H e ˊ t i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j )) ,
where a : S ∖ T → S a:S\setminus T\to S a : S ∖ T → S denotes the open complement.
(3) We have the étale realization functor
r ℓ : D ( S , L ) → D e ˊ t ( S , L ℓ ) , r_{\ell}:\mathrm{D}(S,L)\to\mathrm{D}_{\text{ét}}(S,L_{\ell}), r ℓ : D ( S , L ) → D e ˊ t ( S , L ℓ ) ,
compatible with the six operations [2 ] . In particular, for every morphism f : X → S f:X\to S f : X → S of finite type, r ℓ R f ∗ 1 X , L = R f ∗ 1 X , L ℓ e ˊ t r_{\ell}Rf_{*}\mathbb{1}_{X,L}=Rf_{*}\mathbb{1}^{\text{ét}}_{X,L_{\ell}} r ℓ R f ∗ 1 X , L = R f ∗ 1 X , L ℓ e ˊ t .
Notation A.4. For every complex manifold S S S , we have the following.
(1) Put L ∞ : = L ⊗ R L_{\infty}:=L\otimes\mathbb{R} L ∞ := L ⊗ R and denote by D h s ( S , L ∞ ) \mathrm{D}_{\mathrm{hs}}(S,L_{\infty}) D hs ( S , L ∞ ) the derived category of L ∞ L_{\infty} L ∞ -linear objects in the category of (real) mixed Hodge modules over S S S [34 ] , which is a closed symmetric monoidal L ∞ L_{\infty} L ∞ -linear triangulated category whose unit element we denote by 1 S , L ∞ h s \mathbb{1}^{\mathrm{hs}}_{S,L_{\infty}} 1 S , L ∞ hs . Denote by D h s △ ( S , L ∞ ) \mathrm{D}^{\vartriangle}_{\mathrm{hs}}(S,L_{\infty}) D hs △ ( S , L ∞ ) the heart of D h s ( S , L ∞ ) \mathrm{D}_{\mathrm{hs}}(S,L_{\infty}) D hs ( S , L ∞ ) with respect to the usual t-structure.
(2) For every object M M M in D h s ( S , L ∞ ) \mathrm{D}_{\mathrm{hs}}(S,L_{\infty}) D hs ( S , L ∞ ) and integers i , j i,j i , j , we have the absolute Hodge cohomology43
H h s i ( S , M ( j ) ) : = Hom D h s ( S , L ∞ ) ( 1 S , L ∞ h s , M [ i ] ( j ) ) H^{i}_{\mathrm{hs}}(S,M(j)) := \operatorname{Hom}_{\mathrm{D}_{\mathrm{hs}}(S,L_{\infty})}(\mathbb{1}^{\mathrm{hs}}_{S,L_{\infty}},M[i](j)) H hs i ( S , M ( j )) := Hom D hs ( S , L ∞ ) ( 1 S , L ∞ hs , M [ i ] ( j ))
as an L ∞ L_{\infty} L ∞ -module, and for every closed subset T ⊆ S T \subseteq S T ⊆ S the absolute Hodge cohomology with support
H h s , T i ( S , M ( j ) ) : = H h s i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j ) ) , \mathrm{H}^{i}_{\mathrm{hs},T}(S,\mathrm{M}(j)) := \mathrm{H}^{i+1}_{\mathrm{hs}}(S,\operatorname{cone}(\mathrm{M} \to\mathrm{R}a_{*}a^{*}\mathrm{M})(j)), H hs , T i ( S , M ( j )) := H hs i + 1 ( S , cone ( M → R a ∗ a ∗ M ) ( j )) ,
where a : S ∖ T → S a:S \setminus T \to S a : S ∖ T → S denotes the open complement.
(3) Consider a field k k k together with an embedding u : k → C u:k \to\mathbb{C} u : k → C . For every smooth quasi-projective scheme S S S over k k k , we have the Hodge realization functor
r u : D ( S , L ) → D h s ( S u , L ∞ ) , r^{u}: \mathrm{D}(S,L) \to\mathrm{D}_{\mathrm{hs}}(S^{u},L_{\infty}), r u : D ( S , L ) → D hs ( S u , L ∞ ) ,
compatible with the six operations [15 , 35 ] . In particular, for every smooth quasi-projective morphism f : X → S f:X \to S f : X → S ,
r u R f ∗ 1 X , L = R f ∗ u 1 X u , L ∞ h s . r^{u}\mathrm{R}f_{*}\mathbb{1}_{X,L}=\mathrm{R}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{X^{u},L_{\infty}}. r u R f ∗ 1 X , L = R f ∗ u 1 X u , L ∞ hs .
Motives of abelian schemes
Definition A.5. Consider a ring R R R and an R R R -scheme T T T .
We have a natural functor AV ( − , R ) : ( Sch / T ) o p → Cat \operatorname{AV}(-,R):(\operatorname{Sch}/T)^{\mathrm{op}} \to\operatorname{Cat} AV ( − , R ) : ( Sch / T ) op → Cat such that for every T T T -scheme S S S , AV ( S , R ) \operatorname{AV}(S,R) AV ( S , R ) is the category of abelian schemes A A A over T T T together with a unit-preserving algebra homomorphism
R → End S ( A ) ⊗ Z [ p − 1 ∣ p ∈ R × ] . R \to\operatorname{End}_{S}(A) \otimes\mathbb{Z}[p^{-1}\mid p\in R^{\times}]. R → End S ( A ) ⊗ Z [ p − 1 ∣ p ∈ R × ] .
For S ∈ Sch / T S \in\operatorname{Sch}/T S ∈ Sch / T and A ∈ AV ( S , R ) A \in\operatorname{AV}(S,R) A ∈ AV ( S , R ) , write tr A ( r ) : = tr ( r ∣ Lie S A ) ∈ O ( S ) \operatorname{tr}_{A}(r):=\operatorname{tr}(r\mid\operatorname{Lie}_{S}A)\in\mathcal{O}(S) tr A ( r ) := tr ( r ∣ Lie S A ) ∈ O ( S ) for r ∈ R r\in R r ∈ R .
Let S S S be a regular Noetherian scheme. Consider an abelian scheme A / S A/S A / S of relative dimension g g g . We have the corresponding cohomological and homological relative motives
h ∗ ( A / S , Q ) , h ∗ ( A / S , Q ) h^{*}(A/S,\mathbb{Q}),\qquad h_{*}(A/S,\mathbb{Q}) h ∗ ( A / S , Q ) , h ∗ ( A / S , Q )
in Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) , together with the evaluation map (Definition 1.13)
h ∗ ( A / S , Q ) ⊗ h ∗ ( A / S , Q ) → 1 S , Q . (68) h^{*}(A/S,\mathbb{Q})\otimes h_{*}(A/S,\mathbb{Q})\to\mathbb{1}_{S,\mathbb{Q}}. \tag*{(68)} h ∗ ( A / S , Q ) ⊗ h ∗ ( A / S , Q ) → 1 S , Q . (68)
By [9 Corollary 3.2] (see Remark A.6 below), there is a canonical decomposition44
h ∗ ( A / S , Q ) = ⨁ i = 0 2 g h i ( A / S , Q ) h^{*}(A/S,\mathbb{Q})=\bigoplus_{i=0}^{2g}h^{i}(A/S,\mathbb{Q}) h ∗ ( A / S , Q ) = i = 0 ⨁ 2 g h i ( A / S , Q )
in the category Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) such that n ∗ n^{*} n ∗ acts on h i ( A / S , Q ) h^{i}(A/S,\mathbb{Q}) h i ( A / S , Q ) by multiplication with n i n^{i} n i for every n ∈ Z n\in\mathbb{Z} n ∈ Z . It follows by (A.1) that there is another canonical decomposition
h ∗ ( A / S , Q ) = ⨁ i = 0 2 g h i ( A / S , Q ) h_{*}(A/S,\mathbb{Q})=\bigoplus_{i=0}^{2g}h_{i}(A/S,\mathbb{Q}) h ∗ ( A / S , Q ) = i = 0 ⨁ 2 g h i ( A / S , Q )
in the category Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) such that n ∗ n_{*} n ∗ acts on h i ( A / S , Q ) h_{i}(A/S,\mathbb{Q}) h i ( A / S , Q ) by multiplication with n i n^{i} n i for every n ∈ Z n\in\mathbb{Z} n ∈ Z . By [9 Proposition 2.11] (and the following remark), the (relative) Fourier transform F A / S F_{A/S} F A / S induces an isomorphism
F A / S i : h 2 g − i ( A / S , Q ) → ∼ h i ( A ∨ / S , Q ) ⟨ i − g ⟩ . (69) F^{i}_{A/S}:h^{2g-i}(A/S,\mathbb{Q})\xrightarrow{\sim}h^{i}(A^{\vee}/S,\mathbb{Q})\langle i-g\rangle. \tag*{(69)} F A / S i : h 2 g − i ( A / S , Q ) ∼ h i ( A ∨ / S , Q ) ⟨ i − g ⟩ . (69)
in the category Mot ( S , Q ) \operatorname{Mot}(S,\mathbb{Q}) Mot ( S , Q ) for every 0 ≤ i ≤ 2 g 0\leq i\leq2g 0 ≤ i ≤ 2 g .
Remark A.6. The article [9 ] only considered the case where S S S is smooth quasi-projective over a base field. However, as pointed out in [19 Remark 1.1] , all results in [9 ] remain true as long as the Chow ring (with coefficients in Q \mathbb{Q} Q ) is defined for every smooth projective scheme over S S S , which is case when S S S is a regular Noetherian scheme by Remark A.2(1).
We define a map
h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟶ 1 S , Q ⟨ − 1 ⟩ (70) \mathrm{h}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{1}(A^{\vee}/S,\mathbb{Q}) \longrightarrow\mathbb{1}_{S,\mathbb{Q}}\langle-1\rangle \tag*{(70)} h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟶ 1 S , Q ⟨ − 1 ⟩ (70)
to be the composition of the isomorphism
1 ⊗ ( F A / S 1 ) − 1 : h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟶ h ~ 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟨ g − 1 ⟩ \mathbb{1} \otimes(\mathrm{F}^{1}_{A/S})^{-1}: \mathrm{h}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{1}(A^{\vee}/S,\mathbb{Q}) \longrightarrow\widetilde{\mathrm{h}}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{2g-1}(A/S,\mathbb{Q})\langle g-1\rangle 1 ⊗ ( F A / S 1 ) − 1 : h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟶ h 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟨ g − 1 ⟩
and the Poincaré duality map
h 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟶ 1 S , Q ⟨ − g ⟩ . \mathrm{h}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{2g-1}(A/S,\mathbb{Q}) \longrightarrow\mathbb{1}_{S,\mathbb{Q}}\langle-g\rangle. h 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟶ 1 S , Q ⟨ − g ⟩ .
Lemma A.7. The map (A.3) induces an isomorphism h 1 ( A ∨ / S , Q ) ⟶ h 1 ( A / S , Q ) ∨ ⟨ − 1 ⟩ \mathrm{h}^{1}(A^{\vee}/S,\mathbb{Q}) \longrightarrow\mathrm{h}^{1}(A/S,\mathbb{Q})^{\vee}\langle-1\rangle h 1 ( A ∨ / S , Q ) ⟶ h 1 ( A / S , Q ) ∨ ⟨ − 1 ⟩ .
Proof. It suffices to show that the induced map
h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟨ 1 ⟩ ⟶ 1 S , Q \mathrm{h}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{1}(A^{\vee}/S,\mathbb{Q})\langle1\rangle\longrightarrow\mathbb{1}_{S,\mathbb{Q}} h 1 ( A / S , Q ) ⊗ h 1 ( A ∨ / S , Q ) ⟨ 1 ⟩ ⟶ 1 S , Q
is an evaluation map. By construction, this map is isomorphic to the map
h 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟨ g ⟩ ⟶ 1 S , Q \mathrm{h}^{1}(A/S,\mathbb{Q}) \otimes\mathrm{h}^{2g-1}(A/S,\mathbb{Q})\langle g\rangle\longrightarrow\mathbb{1}_{S,\mathbb{Q}} h 1 ( A / S , Q ) ⊗ h 2 g − 1 ( A / S , Q ) ⟨ g ⟩ ⟶ 1 S , Q
induced from the Poincaré duality map, which is an evaluation map. □
Remark A.8. In this remark, we identify the étale and Hodge realizations of h ∗ ( A / S , Q ) \mathrm{h}^{*}(A/S,\mathbb{Q}) h ∗ ( A / S , Q ) and h ∗ ( A / S , Q ) \mathrm{h}_{*}(A/S,\mathbb{Q}) h ∗ ( A / S , Q ) . Write f : A ⟶ S f:A\longrightarrow S f : A ⟶ S for the structure morphism.
Take a rational prime ℓ \ell ℓ that is invertible on S S S . By the compatibility of six operations for the étale realization (Notation A.3), we have
r ℓ h ∗ ( A / S , Q ) = R f ∗ 1 A , Q ℓ e ˊ t . r_{\ell}\mathrm{h}^{*}(A/S,\mathbb{Q})=\mathrm{R}f_{*}\mathbb{1}^{\text{ét}}_{A,\mathbb{Q}_{\ell}}. r ℓ h ∗ ( A / S , Q ) = R f ∗ 1 A , Q ℓ e ˊ t .
For the right-hand side, the corresponding decomposition is
R f ∗ 1 A , Q ℓ e ˊ t = ⨁ i = 0 2 g R i f ∗ 1 A , Q ℓ e ˊ t [ − i ] . \mathrm{R}f_{*}\mathbb{1}^{\text{ét}}_{A,\mathbb{Q}_{\ell}}=\bigoplus_{i=0}^{2g}\mathrm{R}^{i}f_{*}\mathbb{1}^{\text{ét}}_{A,\mathbb{Q}_{\ell}}[-i]. R f ∗ 1 A , Q ℓ e ˊ t = i = 0 ⨁ 2 g R i f ∗ 1 A , Q ℓ e ˊ t [ − i ] .
It follows that
r ℓ h i ( A / S , Q ) = h e ˊ t i ( A / S , Q ℓ ) : = R i f ∗ 1 A , Q ℓ e ˊ t [ − i ] r_{\ell}\mathrm{h}^{i}(A/S,\mathbb{Q})=\mathrm{h}^{i}_{\text{ét}}(A/S,\mathbb{Q}_{\ell}):=\mathrm{R}^{i}f_{*}\mathbb{1}^{\text{ét}}_{A,\mathbb{Q}_{\ell}}[-i] r ℓ h i ( A / S , Q ) = h e ˊ t i ( A / S , Q ℓ ) := R i f ∗ 1 A , Q ℓ e ˊ t [ − i ]
for every 0 ≤ i ≤ 2 g 0\leq i\leq2g 0 ≤ i ≤ 2 g . See the remark after [9 Corollary 3.2] . Dually, we have
r ℓ h i ( A / S , Q ) = h i e ˊ t ( A / S , Q ℓ ) : = ( R i f ∗ 1 A , Q ℓ e ˊ t ) ∨ [ i ] r_{\ell}\mathrm{h}_{i}(A/S,\mathbb{Q})=\mathrm{h}^{\text{ét}}_{i}(A/S,\mathbb{Q}_{\ell}):=\left(\mathrm{R}^{i}f_{*}\mathbb{1}^{\text{ét}}_{A,\mathbb{Q}_{\ell}}\right)^{\vee}[i] r ℓ h i ( A / S , Q ) = h i e ˊ t ( A / S , Q ℓ ) := ( R i f ∗ 1 A , Q ℓ e ˊ t ) ∨ [ i ]
for every 0 ≤ i ≤ 2 g 0\leq i\leq2g 0 ≤ i ≤ 2 g .
Suppose that S S S is smooth quasi-projective over a field k k k with an embedding u : k ⟶ C u:k\longrightarrow\mathbb{C} u : k ⟶ C . By the compatibility of six operations for the Hodge realization (Notation A.4), we have
r ℓ h ∗ ( A / S , Q ) = R f ∗ u 1 A R u h s . r_{\ell}\mathrm{h}^{*}(A/S,\mathbb{Q})=\mathrm{R}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{A^{u}_{\mathbb{R}}}. r ℓ h ∗ ( A / S , Q ) = R f ∗ u 1 A R u hs .
For the right-hand side, the corresponding decomposition is
R f ∗ u 1 A R u h s = ⨁ i = 0 2 g R i f ∗ u 1 A R u h s [ − i ] . \mathrm{R}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{A^{u}_{\mathbb{R}}}=\bigoplus_{i=0}^{2g}\mathrm{R}^{i}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{A^{u}_{\mathbb{R}}}[-i]. R f ∗ u 1 A R u hs = i = 0 ⨁ 2 g R i f ∗ u 1 A R u hs [ − i ] .
It follows that
r ℓ h i ( A / S , Q ) = h h s i ( A / S , R ) : = R i f ∗ u 1 A R u h s [ − i ] r_{\ell}\mathrm{h}^{i}(A/S,\mathbb{Q})=\mathrm{h}^{i}_{\mathrm{hs}}(A/S,\mathbb{R}):=\mathrm{R}^{i}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{A^{u}_{\mathbb{R}}}[-i] r ℓ h i ( A / S , Q ) = h hs i ( A / S , R ) := R i f ∗ u 1 A R u hs [ − i ]
for every 0 ≤ i ≤ 2 g 0\leq i\leq2g 0 ≤ i ≤ 2 g . Dually, we have
r ℓ h i ( A / S , Q ) = h i h s ( A / S , R ) : = ( R i f ∗ u 1 A R u h s ) ∨ [ i ] r_{\ell}\mathrm{h}_{i}(A/S,\mathbb{Q})=\mathrm{h}^{\mathrm{hs}}_{i}(A/S,\mathbb{R}):=\left(\mathrm{R}^{i}f^{u}_{*}\mathbb{1}^{\mathrm{hs}}_{A^{u}_{\mathbb{R}}}\right)^{\vee}[i] r ℓ h i ( A / S , Q ) = h i hs ( A / S , R ) := ( R i f ∗ u 1 A R u hs ) ∨ [ i ]
for every $0\leq i\leq 2g.
References
Institute for Advanced Study in Mathematics, Zhejiang University, Hangzhou 310058, China
Email address: liuyf0719@zju.edu.cn