Hodge Structures of Type (n, 0, ..., 0, n)

Abstract

Hodge structures of type (n, 0, . . . , 0, n) Burt Totaro Completing earlier work by Albert, Shimura found all the possible endomor- phism algebras (tensored with the rationals) for complex abelian varieties of a given dimension [12, Theorem 5]. In five exceptional cases, every abelian variety on which a certain algebra acts has “extra endomorphisms”, so that the full endomorphism algebra is bigger than expected. Complex abelian varieties X up to isogeny are equivalent to polarizable Q-Hodge structures of weight 1, with Hodge numbers (n, n) (where n is the dimension of X). In this paper, we generalize Shimura’s classification to determine all the possible endomorphism algebras for polarizable Q-Hodge structures with Hodge numbers (n, 0, . . . , 0, n). For Hodge structures of odd weight, the answer is the same as for abelian varieties. For Hodge structures of even weight, the answer is similar but not identical. The proof combines ideas from Shimura with Green-Griffiths-Kerr’s approach to computing Mumford-Tate groups [4, Proposition VI.A.5]. As with abelian varieties, the most interesting feature of the classification is that in certain cases, every Hodge structure on which a given algebra acts must have extra endomorphisms. (Throughout this discussion, we only consider polarizable Hodge structures.) One known case (pointed out to me by Beauville) is that every Q- Hodge structure with Hodge numbers (1, 0, 1) has endomorphisms by an imaginary quadratic field and hence is of complex multiplication (CM) type, meaning that its Mumford-Tate group is commutative. More generally, every Q-Hodge structure with Hodge numbers (n, 0, n) that has endomorphisms by a totally real field F of degree n has endomorphisms by a totally imaginary quadratic extension field of F , and hence is of CM type. Another case, which seems to be new, is that a Q-Hodge structure V with Hodge numbers (2, 0, 2) that has endomorphisms by an imaginary quadratic field F 0 must have endomorphisms by a quaternion algebra over Q. In this case, V need not be of CM type; there is a period space isomorphic to CP 1 of Hodge structures of this type, whereas there are only countably many Hodge structures of CM type. To motivate the results of this paper on endomorphism algebras, consider the geometric origin of Hodge structures. A Hodge structure comes from geometry if it is a summand of the cohomology of a smooth complex projective variety defined by an algebraic correspondence. Griffiths found (“Griffiths transversality”) that a family of Hodge structures coming from geometry can vary only in certain directions, expressed by the notion of a variation of Hodge structures [15, Theorem 10.2]. In particular, any variation of Hodge structures of weight at least 2 with Hodge numbers (n, 0, . . . , 0, n) (so there is at least one 0) is locally constant; more generally, this holds whenever there are no two adjacent nonzero Hodge numbers. This has the remarkable consequence that only countably many Hodge structures of weight at least 2 with Hodge numbers (n, 0, . . . , 0, n) come from geometry. Very little is