Weil classes and decomposable abelian fourfolds (Q2106525)

We work over $\mathbb{C}$. For a smooth projective variety $X$, denote by $B^p(X) = H^{2p}(X,\mathbb{Q}) \cap H^{p,p}(X)$ the ring of rational Hodge classes of degree $2p$ on $X$, for $p \in \mathbb{Z}_{\geq 0}$. The Hodge conjecture predicts that $B^p(X)$ is spanned by algebraic classes for every smooth projective variety $X$ and every $p \in \mathbb{Z}_{\geq 0}$. This conjecture is known whenever $p \in \{0,1,n-1,n\}$, where $n = \dim(X)$ (hence for all $X$ of $n = \dim(X) \leq 3$, but wide open otherwise. In particular, it is unknown whether abelian fourfolds satisfy the Hodge conjecture. Moonen and Zarhin have proved that if the Hodge conjecture holds for abelian fourfolds of Weil type, then it holds for all abelian fourfolds, see [\textit{B. J. J. Moonen} and \textit{Yu. G. Zarhin}, Duke Math. J. 77, No. 3, 553--581 (1995; Zbl 0874.14034)] and [\textit{B. J. J. Moonen} and \textit{Yu. G. Zarhin}, Math. Ann. 315, No. 4, 711--733 (1999; Zbl 0947.14005)]. This makes the study of abelian fourfolds of Weil type particularly important. The goal of the present paper is, broadly speaking, to consider a very general abelian surface $J$, and study the problem which Hodge classes $\alpha \in B^2(J^2) \otimes_{\mathbb{Q}}\mathbb{C}$ become a linear combination of \textit{Weil classes} (c.f.~Definition 3.2) and the class $\omega_K^2$, for a suitable embedding $\phi: K \hookrightarrow \mathrm{End}_{\mathbb{Q}}(J^2) = M_2(\mathbb{Q})$ by an imaginary quadratic number field $K$. The author proves in Lemma 3.3 that such an embedding provides $J^2$ with the structure of polarized abelian fourfold of Weil type $(J^2, K, \omega_K)$. The first results on the Hodge conjecture for abelian fourfolds of Weil type were obtained by \textit{C. Schoen} [Compos. Math. 65, No. 1, 3--32 (1988; Zbl 0663.14006)]. In a related paper \textit{C. Schoen} [Int. J. Math. 18, No. 5, 585--612 (2007; Zbl 1118.14042)] starts by considering a certain reducible surface $S \subset JC \times JC$, where $JC$ is the Jacobian of a general genus two curve $C$, and shows that $S$ deforms in a specific four-dimensional family of abelian fourfolds of Weil type (see Theorem 1.1 in [loc. cit.]. This establishes the Hodge conjecture for these abelian fourfolds. In the present paper, the author considers and answers the following related questions. For which $\alpha \in B^2(J^2) \otimes\mathbb{C}$ does there exist an embedding of an imaginary quadratic number field $K$ into $\mathrm{End}_{\mathbb{Q}}(J^2)$, such that $\alpha \in B^2_{K, J^2} \otimes\mathbb{C}$? For fixed $\alpha$, how many quadratic $K$ are there for which this holds? Here $ B^2_{K, J^2} \subset B^2(J^2)$ is the subspace spanned by Weil classes and $\omega_K^2$ (c.f.~Equation 1 below). These questions compares to Schoen's results via the following observation: if $\alpha \in B^2_{K, J^2} \otimes\mathbb{C}$ then the parallel transport $\alpha_t \in H^4(A_t,\mathbb{C})$ of $\alpha$ lies in $B^2_{K, A_t} \otimes\mathbb{C}$ for any deformation $(A_t, K, \omega_K)$ of $(J^2, K, \omega_K)$. (Such deformations are parametrized by a $4$-dimensional bounded symmetric domain, as we explain below.) Let us explain the main results of this paper in more detail. Recall that an \textit{abelian variety of Weil type} is a pair $(A, K)$ where $A$ is an abelian variety and $K \subset \mathrm{End}_{\mathbb{Q}}(A)$ is subalgebra with $K =\mathbb{Q}(\sqrt{-d})$ for some $d \in \mathbb{Z}_{>0}$ such that for all $\kappa \in K$, the endomorphism of $T_0A$ defined by the differential of $\kappa$ has eigenvalues $\kappa$ and $\bar \kappa$ with the same multiplicity. A \textit{polarized abelian variety of Weil type} consists of a triple $(A, K, \omega_K)$ where $(A,K)$ is an abelian variety of Weil type and $\omega_K \in B^1(A)$ is a polarization such that $\kappa^\ast(\omega_K) = \kappa \bar \kappa \cdot \omega_K$, for all $\kappa \in K$. Let $n = \dim(A)$, and define a three-dimensional subspace \[ B^n_{K,A} = \mathbb{Q}\cdot \omega_K^n \oplus W_{K,A} \subset B^n(A).\tag{1} \] Here, $W_{K,A} \subset B^n(A)$ is the space of Weil classes of $(A,K)$ (c.f.~Section 2.2 in the paper). Now let $J$ be a very general abelian variety of dimension $n$ over $\mathbb{C}$. Thus, $\mathrm{End}(J) = \mathbb{Z}$ and $\dim_{\mathbb{Q}} B^2(J) = 1$. For any imaginary quadratic field $K$, there exist injective ring homomorphisms \[ \phi \colon K \hookrightarrow \mathrm{End}_{\mathbb{Q}}(J^2) \cong M_2(\mathbb{Q}). \] For each such a ring homomorphism $\phi$, the tuple $(J^2,K)$ is an abelian variety of Weil type (see Lemma 3.3). More precisely, there exists a polarization $\omega_K \in B^1(J^2)$ such that $\left(J^2, K, \omega_K \right)$ is a polarized abelian variety of Weil type. One can deform $(J^2, K, \omega_K)$ to other abelian varieties of Weil type, by changing the complex structure on $H_1(J^2, \mathbb{R})$ in such a way that $K$ acts by $\mathbb{C}$-linear maps on $H_1(J^2,\mathbb{R})$ and $\omega_K$ satisfies the Riemann relations. Such deformations are parametrized by a bounded symmetric domain of dimension $n^2$. For each $(A, K, \omega_K)$ in this family, there is a natural isomorphism $H^{2n}(A,\mathbb{Q}) = H^{2n}(J^2, \mathbb{Q})$ that identifies $B^n_{K,A} \subset H^{2n}(A,\mathbb{Q})$ with $B^n_{K, J^2} \subset H^{2n}(J^2,\mathbb{Q})$. For any Cartan subalgebra $\phi \colon K \hookrightarrow M_2(\mathbb{C})$, the author introduces certain subspaces \[ W_K \subset B_K^n \subset B^n(J^2) \otimes_{\mathbb{Q}} \mathbb{C} \] with $\dim W_K = 2$ and $\dim B_K^n = 3$, see Definition 3.4. If $\phi$ is induced by an embedding $K \hookrightarrow M_2(\mathbb{Q}) = \mathrm{End}_{\mathbb{Q}}(J^2)$, these are the complexifications of the spaces $W_{K,J^2} \subset B_{K, J^2}^n \subset B^n(J^2)$ defined in (1) above. Let $n = 2$. Then $J$ is a very general abelian surface, and since $\dim B^2(J^2) = 6$ by Proposition 3.1, one has \[ \mathbb{P}\left(B^2(J^2) \otimes_{\mathbb{Q}}\mathbb{C} \right) \cong \mathbb{P}^5_{\mathbb{C}}. \] For Cartan subalgebras $K,K' \subset M_2(\mathbb{C})$, the intersection $\mathbb{P} B^2_K \cap\mathbb{P} B^2_{K'} \subset \mathbb{P}( B^2(J^2) \otimes_{\mathbb{Q}}\mathbb{C})$ is a point, and for a fixed Cartan subalgebra $K \subset M_2(\mathbb{C})$, the union $\cup_{K' \subset M_2(\mathbb{C})}\mathbb{P} B_K^2 \cap \mathbb{P} B_{K'}^2$ is a conic $C_K \subset\mathbb{P}B^2_K$ (see Proposition 4.1). The first main theorem of the paper can then be stated as follows. Theorem 4.2. Define \[ Z = \overline{\bigcup_{K \subset M_2(\mathbb{C})} \mathbb{P}B^2_K} \subset \mathbb{P}\left(B^2(J^2) \otimes_{\mathbb{Q}} \mathbb{C}\right) \cong \mathbb{P}^5_{\mathbb{C}}, \] where the union is over all diagonalizable $2$-dimensional Cartan subalgebras $K \subset M_2(\mathbb{C})$. Then $Z$ is a cubic fourfold defined by the determinant of the matrix \[ \begin{pmatrix} 2x_0 & -x_3 & x_1-4x_5 \\ -x_3 & x_1 & x_4 \\ x_1 - 4x_5 & x_4 & 2x_2 \end{pmatrix}. \] The singular locus $\mathrm{Sing}(Z) \subset Z$ is a Veronese surface $S$, and $Z$ is the secant variety of $S$. Finally, for any diagonalizable $2$-dimensional Cartan subalgebra $K \subset M_2(\mathbb{C})$, one has retrieves the conic $C_K \subset \mathbb{P} B^2_K$ defined above as the intersection of $\mathbb{P} B^2_K$ with $S$: one has \[ \mathbb{P} B^2_K \cap S = C_K. \] The second main theorem of this paper is as follows. Theorem 4.6. A class $\alpha \in Z - \mathrm{Sing}(Z) \subset \mathbb{P}(B^2(J^2) \otimes_{\mathbb{Q}}\mathbb{C})$ deforms in at most one four-dimensional family of abelian fourfolds of Weil type. A class $\alpha \in \mathrm{Sing}(Z)$ may deform in infinitely many four-dimensional families of Weil type with distinct imaginary quadratic fields. In [\textit{E. Markman}, J. Eur. Math. Soc. (JEMS) 25, No. 1, 231--321 (2023; Zbl 07665043)], Markman proves the Hodge conjecture for polarized abelian fourfolds of Weil type $(A, K, \omega_K)$ with trivial discriminant, using specific Hodge classes called \emph{Cayley classes}, that deform in Weil type families for any imaginary quadratic field $K$. The author discusses the relation between Markman's and his results in Section 5. In Section 6, the author discusses Hodge classes and algebraic cycles introduced in [\textit{C. Schoen}, Int. J. Math. 18, No. 5, 585--612 (2007; Zbl 1118.14042)] and [\textit{B. van Geemen}, Math. Z. 221, No. 4, 617--631 (1996; Zbl 0862.14029)]. In Section 7, the author proves that the discriminant in $\mathbb{Q}^\ast / \mathrm{Nm}(K^\ast)$ of $(J^2, K, \omega_K)$ is trivial, when $J$ is a very general abelian surface and $K \subset M_2(\mathbb{Q})$ an imaginary quadratic field.