It was noticed by
Jean-Pierre Serre in the 1960s based on the
Weil conjectures that even singular (possibly reducible) and non-complete algebraic varieties should admit 'virtual Betti numbers'. More precisely, one should be able to assign to any algebraic variety
X a polynomial
PX(
t), called its
virtual Poincaré polynomial, with the properties • If
X is nonsingular and projective (or complete) P_X(t) = \sum \operatorname{rank}(H^n(X))t^n • If
Y is closed algebraic subset of
X and
U =
X \
Y P_X(t)=P_Y(t)+P_U(t) The existence of such polynomials would follow from the existence of an analogue of Hodge structure in the cohomologies of a general (singular and non-complete) algebraic variety. The novel feature is that the
nth cohomology of a general variety looks as if it contained pieces of different weights. This led
Alexander Grothendieck to his conjectural theory of
motives and motivated a search for an extension of Hodge theory, which culminated in the work of
Pierre Deligne. He introduced the notion of a mixed Hodge structure, developed techniques for working with them, gave their construction (based on
Heisuke Hironaka's
resolution of singularities) and related them to the weights on
l-adic cohomology, proving the last part of the
Weil conjectures.
Example of curves To motivate the definition, consider the case of a reducible complex
algebraic curve X consisting of two nonsingular components, X_1 and X_2, which transversally intersect at the points Q_1 and Q_2. Further, assume that the components are not compact, but can be compactified by adding the points P_1, \dots ,P_n. The first cohomology group of the curve
X (with compact support) is dual to the first homology group, which is easier to visualize. There are three types of one-cycles in this group. First, there are elements \alpha_i representing small loops around the punctures P_i. Then there are elements \beta_j that are coming from the first homology of the
compactification of each of the components. The one-cycle in X_k \subset X (k=1,2) corresponding to a cycle in the compactification of this component, is not canonical: these elements are determined modulo the span of \alpha_1, \dots ,\alpha_n. Finally, modulo the first two types, the group is generated by a combinatorial cycle \gamma which goes from Q_1 to Q_2along a path in one component X_1 and comes back along a path in the other component X_2. This suggests that H_1(X) admits an increasing filtration : 0\subset W_0\subset W_1 \subset W_2=H^1(X), whose successive quotients
Wn/
Wn−1 originate from the cohomology of smooth complete varieties, hence admit (pure) Hodge structures, albeit of different weights. Further examples can be found in "A Naive Guide to Mixed Hodge Theory".
Definition of mixed Hodge structure A
mixed Hodge structure on an abelian group H_{\Z} consists of a finite decreasing filtration
Fp on the complex vector space
H (the complexification of H_{\Z}), called the
Hodge filtration and a finite increasing filtration
Wi on the rational vector space H_{\Q} = H_{\Z} \otimes_{\Z} \Q (obtained by extending the scalars to rational numbers), called the
weight filtration, subject to the requirement that the
n-th associated graded quotient of H_{\Q} with respect to the weight filtration, together with the filtration induced by
F on its complexification, is a pure Hodge structure of weight
n, for all integer
n. Here the induced filtration on : \operatorname{gr}_n^{W} H = W_n\otimes\Complex /W_{n-1}\otimes\Complex is defined by : F^p \operatorname{gr}_n^W H = \left (F^p\cap W_n\otimes\Complex +W_{n-1} \otimes \Complex \right )/W_{n-1}\otimes\Complex. One can define a notion of a morphism of mixed Hodge structures, which has to be compatible with the filtrations
F and
W and prove the following: :
Theorem. Mixed Hodge structures form an abelian category. The kernels and cokernels in this category coincide with the usual kernels and cokernels in the category of vector spaces, with the induced filtrations. The total cohomology of a compact Kähler manifold has a mixed Hodge structure, where the
nth space of the weight filtration
Wn is the direct sum of the cohomology groups (with rational coefficients) of degree less than or equal to
n. Therefore, one can think of classical Hodge theory in the compact, complex case as providing a double grading on the complex cohomology group, which defines an increasing filtration
Fp and a decreasing filtration
Wn that are compatible in certain way. In general, the total cohomology space still has these two filtrations, but they no longer come from a direct sum decomposition. In relation with the third definition of the pure Hodge structure, one can say that a mixed Hodge structure cannot be described using the action of the group \Complex^*. An important insight of Deligne is that in the mixed case there is a more complicated noncommutative proalgebraic group that can be used to the same effect using
Tannakian formalism. Moreover, the category of (mixed) Hodge structures admits a good notion of tensor product, corresponding to the product of varieties, as well as related concepts of
inner Hom and
dual object, making it into a
Tannakian category. By
Tannaka–Krein philosophy, this category is equivalent to the category of finite-dimensional representations of a certain group, which Deligne, Milne and et el. has explicitly described, see and . The description of this group was recast in more geometrical terms by . The corresponding (much more involved) analysis for rational pure polarizable Hodge structures was done by .
Mixed Hodge structure in cohomology (Deligne's theorem) Deligne has proved that the
nth cohomology group of an arbitrary algebraic variety has a canonical mixed Hodge structure. This structure is
functorial, and compatible with the products of varieties (
Künneth isomorphism) and the product in cohomology. For a complete nonsingular variety
X this structure is pure of weight
n, and the Hodge filtration can be defined through the
hypercohomology of the truncated de Rham complex. The proof roughly consists of two parts, taking care of noncompactness and singularities. Both parts use the resolution of singularities (due to Hironaka) in an essential way. In the singular case, varieties are replaced by simplicial schemes, leading to more complicated homological algebra, and a technical notion of a Hodge structure on complexes (as opposed to cohomology) is used. Using the theory of
motives, it is possible to refine the weight filtration on the cohomology with rational coefficients to one with integral coefficients. ==Examples==