Motive (algebraic geometry)

In algebraic geometry, a motive (or sometimes motif, following French usage) denotes 'some essential part of an algebraic variety'. To date, pure motives have been defined, while conjectural mixed motives have not. Pure motives are triples (X, p, m), where X is a smooth projective variety, p : XX is an idempotent correspondence, and m an integer. A morphism from (X, p, m) to (Y, q, n) is given by a correspondence of degree n – m.

As far as mixed motives, following Alexander Grothendieck, mathematicians are working to find a suitable definition which will then provide a "universal" cohomology theory. In terms of category theory, it was intended to have a definition via splitting idempotents in a category of algebraic correspondences. The way ahead for that definition has been blocked for some decades by the failure to prove the standard conjectures on algebraic cycles. This prevents the category from having 'enough' morphisms, as can currently be shown. While the category of motives was supposed to be the universal Weil cohomology much discussed in the years 1960-1970, that hope for it remains unfulfilled. On the other hand, by a quite different route, motivic cohomology now has a technically adequate definition.

Introduction

The theory of motives was originally conjectured as an attempt to unify a rapidly multiplying array of cohomology theories, including Betti cohomology, de Rham cohomology, l-adic cohomology, and crystalline cohomology. The general hope is that equations like

can be put on increasingly solid mathematical footing with a deep meaning. Of course, the above equations are already known to be true in many senses, such as in the sense of CW-complex where "+" corresponds to attaching cells, and in the sense of various cohomology theories, where "+" corresponds to the direct sum.

From another viewpoint, motives continue the sequence of generalizations from rational functions on varieties to divisors on varieties to Chow groups of varieties. The generalization happens in more than one direction, since motives can be considered with respect to more types of equivalence than rational equivalence. The admissiable equivalences are given by the definition of an adequate equivalence relation.

Definition of pure motives

The category of pure motives often proceeds in three steps. Below we describe the case of Chow motives Chow(k), where k is any field.

First step: category of (degree 0) correspondences, Corr(k)

The objects of Corr(k) are simply smooth projective varieties over k. The morphisms are correspondences. They generalize morphisms of varieties X Y, which can be associated with their graphs in X × Y, to fixed dimensional Chow cycles on X × Y.

It will be useful to describe correspondences of arbitrary degree, although morphisms in Corr(k) are correspondences of degree 0. In detail, let X and Y be smooth projective varieties, let \scriptstyle X = \coprod_i X_i be the decomposition of X into connected components, and let di := dim Xi. If rZ, then the correspondences of degree r from X to Y are

Corr^r(k)(X, Y) := \bigoplus_i A^{d_i+r}(X_i \times Y).

Correspondences are often denoted using the "⊢"-notation, e.g., α: XY. For any α ∈ Corrr(X, Y) and β ∈ Corrs(Y, Z), their composition is defined by

\alpha \circ \beta := \pi_{XZ*}(\pi^{*}_{XY}(\alpha) \cdot \pi^{*}_{YZ}(\beta)) \in Corr^{r+s}(X, Z),

where the dot denotes the product in the Chow ring (i.e., intersection).

Returning to constructing the category Corr(k), notice that the composition of degree 0 correspondences is degree 0. Hence we define morphisms of Corr(k) to be degree 0 correspondences.

The association,

F : \begin{array}{rcl}
SmProj(k) & \longrightarrow & Corr(k) \\
X & \longmapsto & X \\
f & \longmapsto & \Gamma_f
\end{array},

where Γf ⊆ X × Y is the graph of f : X → Y, is a functor.

Just like SmProj(k), the category Corr(k) has direct sums (\scriptstyle X \oplus Y := X \coprod Y) and tensor products (XY := X × Y). It is a preadditive category (see the convention for preadditive vs. additive in the preadditive category article.) The sum of morphisms is defined by

\alpha + \beta := (\alpha, \beta) \in A^{*}(X \times X) \oplus A^{*}(Y \times Y) \hookrightarrow A^{*}((X \coprod Y) \times (X \coprod Y)).

Second step: category of pure effective Chow motives, Choweff(k)

The transition to motives is made by taking the pseudo-abelian envelope of Corr(k):

Chow^{eff}(k) := Split(Corr(k)).

In other words, effective Chow motives are pairs of smooth projective varieties X and idempotent correspondences α: XX, and morphisms are of a certain type of correspondence:

Ob(Chow^{eff}(k)) := \{ (X, \alpha) \mbox{ }|\mbox{ } (\alpha : X \vdash X) \in Corr(k) \mbox{ such that } \alpha \circ \alpha = \alpha \}.
Mor((X, \alpha), (Y, \beta)) := \{ f : X \vdash Y \mbox{ }|\mbox{ } f \circ \alpha = f = \beta \circ f \}.

Composition is the above defined composition of correspondences, and the identity morphism of (X, α) is defined to be α : XX.

The association,

h : \begin{array}{rcl}
SmProj(k) & \longrightarrow & Corr(k) \\
X & \longmapsto & [X] := (X, \Delta)_X \\
f & \longmapsto & [f] := \Gamma_f \subset X \times Y
\end{array},

where ΔX := [idX] denotes the diagonal of X × X, is a functor. The motive [X] is often called the motive associated to the variety X.

As intended, Choweff(k) is a pseudo-abelian category. The direct sum of effective motives is given by

([X], \alpha) \oplus ([Y], \beta) := ([X \coprod Y], \alpha + \beta),

The tensor product of effective motives is defined by

([X], \alpha) \otimes ([Y], \beta) := (X \times Y, \pi_X^{*}\alpha \cdot \pi_Y^{*}\beta), \qquad \pi_X : (X \times Y) \times (X \times Y) \to X \times X, \mbox{ and } \pi_Y : (X \times Y) \times (X \times Y) \to Y \times Y.

The tensor product of morphisms may also be defined. Let f1 : (X1, α1) → (Y1, β1) and f2 : (X2, α2) → (Y2, β2) be morphisms of motives. Then let γ1A*(X1 × Y1) and γ2A*(X2 × Y2) be representatives of f1 and f2. Then

f_1 \otimes f_2 : (X_1, \alpha_1) \otimes (X_2, \alpha_2) \vdash (Y_1, \beta_1) \otimes (Y_2, \beta_2), \qquad f_1 \otimes f_2 := \pi^{*}_1 \gamma_1 \cdot \pi^{*}_2 \gamma_2,

where πi : X1 × X2 × Y1 × Y2Xi × Yi are the projections.

Third step: category of pure Chow motives, Chow(k)

To proceed to motives, we adjoin to Choweff(k) a formal inverse (with respect to the tensor product) of a motive called the Lefschetz motive. The effect is that motives become triples instead of pairs. The Lefschetz motive L is

L := (\mathbf{P}^1, \lambda), \qquad \lambda := pt \times \mathbf{P}^1 \in A^1(\mathbf{P}^1 \times \mathbf{P}^1).

If we define the motive 1, called the trivial Tate motive, by 1 := h(Spec(k)), then the pleasant equation

[\mathbf{P}^1] = \mathbf{1} \oplus L

holds, since 1 ≅ (P1, P1 × pt). The tensor inverse of the Lefschetz motive is known as the Tate motive, T := L−1. Then we define the category of pure Chow motives by

Chow(k) := Chow^{eff}(k)[T].

A motive is then a triple (XSmProj(k), p : XX, nZ) such that p ˆ p = p. Morphisms are given by correspondences

f : (X, p, m) \to (Y, q, n), \quad f \in Corr^{n-m}(X, Y) \mbox{ such that } f \circ p = f = q \circ f,

and the composition of morphisms comes from composition of correspondences.

As intended, Chow(k) is a rigid pseudo-abelian category.

Other types of motives

In order to define an intersection product, cycles must be "movable" so we can intersect them in general position. Choosing a suitable equivalence relation on cycles will guarantee that every pair of cycles has an equivalent pair in general position that we can intersect. The Chow groups are defined using rational equivalence, but other equivalences are possible, and each defines a different sort of motive. Examples of equivalences, from strongest to weakest, are

The literature occasionally calls every type of pure motive a Chow motive, in which case a motive with respect to algebraic equivalence would be called a Chow motive modulo algebraic equivalence.

Mixed motives

For a fixed base field k, the category of mixed motives is a conjectural abelian tensor category MM(k), together with a contravariant functor

Var(k) → MM(X)

taking values on all varieties (not just smooth projective ones as it was the case with pure motives). This should be such that motivic cohomology defined by

Ext*MM(1, ?)

coincides with the one predicted by algebraic K-theory, and contains the category of Chow motives in a suitable sense (and other properties). The existence of such a category was conjectured by Beilinson. This category is yet to be constructed.

Instead of constructing such a category, it was proposed by Deligne to first construct a category DM having the properties one expects for the derived category

Db(MM(k)).

Getting MM back from DM would then be accomplished by a (conjectural) motivic t-structure.

The current state of the theory is that we do have a suitable category DM. Already this category is useful in applications. Voevodsky's Fields Medal-winning proof of the Milnor conjecture uses these motives as a key ingredient.

There are different definitions due to Hanamura, Levine and Voevodsky. They are known to be equivalent in most cases and we will give Voevodsky's definition below. The category contains Chow motives as a full subcategory and gives the "right" motivic cohomology. However, Voevodsky also shows that (with integral coefficients) it does not admit a motivic t-structure.

The resulting category is called the category of effective geometric motives. Again, formally inverting the Tate object, one gets the category DM of geometric motives.

Explanation for non-specialists

A commonly applied technique in mathematics is to study objects carrying a particular structure by introducing a category whose morphisms preserve this structure. Then one may ask, when are two given objects isomorphic and ask for a "particularly nice" representative in each isomorphism class. The classification of algebraic varieties, i.e. application of this idea in the case of algebraic varieties, is very difficult due to the highly non-linear structure of the objects. The relaxed question of studying varieties up to birational isomorphism has led to the field of birational geometry. Another way to handle the question is to attach to a given variety X an object of more linear nature, i.e. an object amenable to the techniques of linear algebra, for example a vector space. This "linearization" goes usually under the name of cohomology.

There are several important cohomology theories, which reflect different structural aspects of varieties. The (partly conjectural) theory of motives is an attempt to find a universal way to linearize algebraic varieties, i.e. motives are supposed to provide a cohomology theory that embodies all these particular cohomologies. For example, the genus of a smooth projective curve C which is an interesting invariant of the curve, is an integer, which can be read off the dimension of the first Betti cohomology group of C. So, the motive of the curve should contain the genus information. Of course, the genus is a rather coarse invariant, so the motive of C is more than just this number.

The search for a universal cohomology

Each algebraic variety X has a corresponding motive [X], so the simplest examples of motives are:

These 'equations' hold in many situations, namely for de Rham cohomology and Betti cohomology, l-adic cohomology, the number of points over any finite field, and in multiplicative notation for local zeta-functions.

The general idea is that one motive has the same structure in any reasonable cohomology theory with good formal properties; in particular, any Weil cohomology theory will have such properties. There are different Weil cohomology theories, they apply in different situations and have values in different categories, and reflect different structural aspects of the variety in question:

All these cohomology theories share common properties, e.g. existence of Mayer-Vietoris sequences, homotopy invariance (H*(X)≅H*(X × A1), the product of X with the affine line) and others. Moreover, they are linked by comparison isomorphisms, for example Betti cohomology H*Betti(X, Z/n) of a smooth variety X over C with finite coefficients is isomorphic to l-adic cohomology with finite coefficients.

The theory of motives is an attempt to find a universal theory which embodies all these particular cohomologies and their structures and provides a framework for "equations" like

[projective line] = [line]+[point].

In particular, calculating the motive of any variety X directly gives all the information about the several Weil cohomology theories H*Betti(X), H*DR(X) etc.

Beginning with Grothendieck, people have tried to precisely define this theory for many years.

Motivic cohomology

Motivic cohomology itself had been invented before the creation of mixed motives by means of algebraic K-theory. The above category provides a neat way to (re)define it by

Hn(X, m) := Hn(X, Z(m)) := HomDM(X, Z(m)[n]),

where n and m are integers and Z(m) is the m-th tensor power of the Tate object Z(1), which in Voevodsky's setting is the complex P1pt shifted by –2, and [n] means the usual shift in the triangulated category.

Conjectures related to motives

The standard conjectures were first formulated in terms of the interplay of algebraic cycles and Weil cohomology theories. The category of pure motives provides a categorical framework for these conjectures.

The standard conjectures are commonly considered to be very hard and are open in the general case. Grothendieck, with Bombieri, showed the depth of the motivic approach by producing a conditional (very short and elegant) proof of the Weil conjectures (which are proven by different means by Deligne), assuming the standard conjectures to hold.

For example, the Künneth standard conjecture, which states the existence of algebraic cycles πiX × X inducing the canonical projectors H*(X) → Hi(X) ↣ H*(X) (for any Weil cohomology H) implies that every pure motive M decomposes in graded pieces of weight n: M = ⊕GrnM. The terminology weights comes from a similar decomposition of, say, de-Rham cohomology of smooth projective varieties, see Hodge theory.

Conjecture D, stating the concordance of numerical and homological equivalence, implies the equivalence of pure motives with respect to homological and numerical equivalence. (In particular the former category of motives would not depend on the choice of the Weil cohomology theory). Jannsen (1992) proved the following unconditional result: the category of (pure) motives over a field is abelian and semisimple if and only if the chosen equivalence relation is numerical equivalence.

The Hodge conjecture, may be neatly reformulated using motives: it holds iff the Hodge realization mapping any pure motive with rational coefficients (over a subfield k of C) to its Hodge structure is a full functor H : M(k)QHSQ (rational Hodge structures). Here pure motive means pure motive with respect to homological equivalence.

Similarly, the Tate conjecture is equivalent to: the so-called Tate realization, i.e. ℓ-adic cohomology is a faithful functor H: M(k)QRep(Gal(k)) (pure motives up to homological equivalence, continuous representations of the absolute Galois group of the base field k), which takes values in semi-simple representations. (The latter part is automatic in the case of the Hodge analogue).

Tannakian formalism and motivic Galois group

To motivate the (conjectural) motivic Galois group, fix a field k and consider the functor

finite separable extensions K of knon-empty finite sets with a (continuous) transitive action of the absolute Galois group of k

which maps K to the (finite) set of embeddings of K into an algebraic closure of k. In Galois theory this functor is shown to be an equivalence of categories. Notice that fields are 0-dimensional. Motives of this kind are called Artin motives. By Q-linearizing the above objects, another way of expressing the above is to say that Artin motives are equivalent to finite Q-vector spaces together with an action of the Galois group.

The objective of the motivic Galois group is to extend the above equivalence to higher-dimensional varieties. In order to do this, the technical machinery of Tannakian category theory (going back to Tannaka–Krein duality, but a purely algebraic theory) is used. Its purpose is to shed light on both the Hodge conjecture and the Tate conjecture, the outstanding questions in algebraic cycle theory. Fix a Weil cohomology theory H. It gives a functor from Mnum (pure motives using numerical equivalence) to finite-dimensional Q-vector spaces. It can be shown that the former category is a Tannakian category. Assuming the equivalence of homological and numerical equivalence, i.e. the above standard conjecture D, the functor H is an exact faithful tensor-functor. Applying the Tannakian formalism, one concludes that Mnum is equivalent to the category of representations of an algebraic group G, which is called motivic Galois group.

It is to the theory of motives what the Mumford–Tate group is to Hodge theory. Again speaking in rough terms, the Hodge and Tate conjectures are types of invariant theory (the spaces that are morally the algebraic cycles are picked out by invariance under a group, if one sets up the correct definitions). The motivic Galois group has the surrounding representation theory. (What it is not, is a Galois group; however in terms of the Tate conjecture and Galois representations on étale cohomology, it predicts the image of the Galois group, or, more accurately, its Lie algebra.)

References