Given a polarized abelian scheme with action by a ring, and a projective finitely presented module over that ring, Serre's tensor construction produces a new abelian scheme. We show that to equip these abelian schemes with polarizations it's enough to equip the projective modules with non-degenerate positive-definite hermitian forms. As an application, we relate certain moduli spaces of principally polarized abelian schemes with action by the ring of integers of a CM field. More specifically, we consider integral models of zero-dimensional Shimura varieties associated to compact unitary groups. We show that all abelian schemes in such moduli spaces are,étale locally over their base schemes, Serre constructions of CM abelian schemes with positive-definite hermitian modules. We also describe the morphisms between such objects in terms of morphisms between the constituent data, and formulate these relations as an isomorphism of algebraic stacks.
IntroductionLet R be a ring, possibly non-commutative, and free of finite rank over . Let (A, ι) be an abelian scheme A over a base S, with an injective ring homomorphism ι : R ֒→ End S (A) giving an R-action on A. Take M to be a projective finitely presented right R-module. Serre's tensor construction associates to this data a new abelian scheme M ⊗ R A over S, which is characterized by its functor of points Sch /S → Ab, T → M ⊗ R A(T ) (Definition 1). The map A → M ⊗ R A is functorial in A and M , and preserves many desirable properties of A. This suggests the possibility of using it to relate families of abelian schemes. In order to do this, we first need to equip M ⊗ R A with extra structures, in particular a polarization [20,6] that is compatible with the R-action in the following sense.Assume R is equipped with a positive involution r → r * (Definition 4). Then the pair (A, ι) has a dual (A ∨ , ι ∨ ), where A ∨ is the dual abelian scheme of A, and ι ∨ (r) = ι(r * ) ∨ , for r ∈ R. A polarization λ :The dual module M ∨ = Hom R (M, R) has a natural right R-module structure, with r ∈ R acting on f ∈ M ∨ by (f · r)(m) = r * f (m). Then R-linear maps h : M → M ∨ may be identified with sesquilinear forms H : M × M → R via H(m, m ′ ) = h(m)(m ′ ). Such a map h is called hermitian if H(m, m ′ ) = H(m ′ , m) * , and non-degenerate if it's an isomorphism. Since M É ≃ R n É , we may identify h with an element of H n (R É ), the set of n × n hermitian matrices with entries in R É . Then H n (R É ) ⊗ Ê is a formally real Jordan algebra (Definition 8) over Ê. We say h is positive-definite if its image under H n (R É ) ⊂ H n (R É ) ⊗ Ê is positive (Definition 15). This notion does not depend on the choice of isomorphism M É ≃ R n É (Lemma 16).Theorem A. Suppose (A, ι, λ) consists of an abelian scheme A/S, an R-action ι : R ֒→ End(A), and R-linear polarization λ : A → A ∨ . Let h : M → M ∨ be R-linear. The map h ⊗ λ : M ⊗ R A → M ∨ ⊗ R A ∨ is a polarization on M ⊗ R A if and only if h is a positive definite R-valued hermitian form.The above is Theorem 17 in the main text. That the abel...