Realization of Voevodsky's motives

by Annette Huber
November 1998

Annette Huber [email protected]

Math. Institut
Einsteinstr. 62
48 149 Munster
Germany
Introduction
The theory of motives has always had two faces. One is the geometric face
where a universal cohomology theory for varieties is cooked up from geo-
metric objects like cycles. The other one is the linear algebra face where
restricting conditions are put on objects of linear algebra like vector spaces
with an operation of the Galois group. The ideal theorem would be an
equivalence of these two approaches.
For pure motives, Grothendieck proposed a geometric construction. The
linear algebra side is covered by l-adic cohomology for all primes l together
with singular cohomology equipped with its Hodge structure. The relation
between the two sides is made by the Tate or the Hodge conjecture which
tell us that geometry and linear algebra should be very close to each other.
However, we neither know whether Grothendieck's construction has the re-
quired properties nor what the image of the category of motives on the linear
algebra side is. I.e., we cannot tell whether a Galois module is motivic just
from checking linear algebra conditions (there are conjectures though).
For mixed motives, Voevodsky's work goes a long way in constructing
the geometric side of the story. The linear algebra side is given by Deligne's
absolute Hodge motives, independently considered by Jannsen under the
name of mixed realizations. By Beilinson's conjectures the interplay be-
tween geometry and linear algebra should be measured by special values of
L-functions of motives. However, we are far from proving the ideal theorem.
The main aim of the present article is to provide the expected functor be-
tween the two sides. More precisely, we construct a realization functor from
Voevodsky's triangulated category of geometrical motives (which should be
thought of as the derived category of mixed motives) to the \derived cate-
gory" of mixed realizations which we constructed in [Hu1]. Indeed, most of
the present article is a follow-up of loc. cit. where the realization functor
was constructed on the category of simplicial varieties. As a direct corollary
we also obtain realizations functors to continuous l-adic cohomology and to
absolute Hodge cohomology. Their existence is not a surprise (cf. [Vo2])
but was not in the literature yet.
We want to mention that Levine has a triangulated category of motives
([Le3]). Over a eld of characteristic zero it is equivalent to Voedvodsky's.
He also constructs realization functors in his setting starting from a dierent

1
set of axioms.
We can show (2.3.6) that the motivic objects in the category of mixed
realizations obtained from Voevodsky's category are contained in the cate-
gory of motivic objects considered before. We do not get new motives on
the linear algebra side.
We can also prove that the Chern classes from higher algebraic K -theory
to mixed realizations factor through Voevodsky's category. The key to this
fact is the computation of the motive of B GL in Voevodsky's category.
Besides these formal insights, the mixed realization functor should be
very useful wherever an attempt is made to prove Beilinson's conjectures on
L-values. They require the construction of elements in motivic cohomology.
In the known cases, Adams eigenspaces of K -theory were used as the deni-
tion of motivic cohomology. The formal properties of motivic cohomology in
the sense of Voevodsky are lot better , e.g. localization sequences involving
singular varieties. Its main advantage is that it allows to do computations
in two variables. Motivic cohomology in the sense of K -theory always xes
the second variable as a Tate motive. We hope that explicit applications
will follow.
Acknowledgments
I would like to thank B. Kahn and M. Spiess who organized an enlightening
Arbeitsgemeinschaft on Voevodsky's work in Oberwolfach. I proted from
discussions with D. Blasius, G. Kings, K. Kunnemann, E. Landvogt, M.
Strauch and J. Wildeshaus. I am particular thankful to B. Kahn and J.
Wildeshaus for their comments on earlier version of this text.

2
Contents
1 Voevodsky's triangulated category of motives 4
1.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Mixed Tate motives . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 More Technical background . . . . . . . . . . . . . . . . . . . 12
2 The realization functor 16
2.1 Axiomatic construction . . . . . . . . . . . . . . . . . . . . . 16
2.2 Review of mixed realizations . . . . . . . . . . . . . . . . . . 27
2.3 The mixed realization functor . . . . . . . . . . . . . . . . . . 30
3 The motive of BGL 34
3.1 Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.2 Motives of some classifying spaces . . . . . . . . . . . . . . . 36
4 Chern classes 38
4.1 K -theory and group cohomology of GL(X ) . . . . . . . . . . 38
4.2 Chern classes into motivic cohomology . . . . . . . . . . . . . 40

3
1 Voevodsky's triangulated category of motives
We start with a quick review of Voevodsky's category. Then we give some
extra notions and properties needed in the present text.
1.1 Review
We repeat the denition of the triangulated category of motives as given by
Voevodsky in [Vo1]. For more details and properties we refer to his paper.
He has developed an integral theory. We mostly need the Q -rational version.
Let k be a xed ground eld of characteristic zero. Let Var be the
category of varieties over k, i.e. separated schemes of nite type over k. Let
Sm be the full subcategory of smooth varieties. Let A = Q or Z be the
coecient ring.
Denition 1.1.1 ([Vo1] 2.1). Let X be a smooth variety and Y a general
variety. A prime correspondence from X to Y is an integral closed sub-
scheme W of X  Y which is nite over X and surjective over a connected
component of X . Let c(X; Y )A be the free A-module generated by prime
correspondences. Its elements are called nite correspondences. Let SmCor
be the category with objects smooth varieties and morphisms given by nite
correspondences.
We want to recall the composition of correspondences. Assume ?1 
X  Y and ?2  Y  Z are prime correspondences. ?1  Z and X  ?2 are
cycles in X  Y  Z . All irreducible components Ci (with reduced structure)
of their intersection are nite over X . In particular they all have the right
dimension. We have
X
(?1  Z ):(X  ?2 ) = ni Ci
where the intersection multiplicity ni of Ci is the usual one in the Chow
group, e.g.[Fu] 20.4. Let  : X  Y  Z ! X  Z be the natural projec-
tion. It induces a map  : c(X; Y  Z ) ! c(X; Z ) which takes a primitive
correspondence C to the closure of its image (C ) times the degree of the
covering C ! (C ). By denition
?

?2  ?1 =  (?1  Z ):(X  ?2 ) :
There is a functor
[
] : Sm ! SmCor :

4
It maps a morphism to its graph.
Note that SmCor is additive. Hence we can consider complexes of objects
in SmCor as well as homotopies between maps of complexes.
Denition 1.1.2 ([Vo1] 2.1.1). The triangulated category of eective ge-
ometrical motives DMeff gm (k; A) is the localization of the homotopy category
b
K (SmCor) with respect to the smallest thick subcategory containing the fol-
lowing:
1. For any smooth scheme X the complex
[X  A 1 ] ! [X ] :
2. For any smooth scheme X and any Zariski-covering X = U [ V the
complex
j 0 +j 0 jU ?jV
[U \ V ] ????
U !V [U ]  [V ] ????! [X ]
where jU0 , jV0 , jU and jV are the obvious inclusions.
The bre product of varieties induces a tensor product structure of DMeff
gm (k; A).
Remark: If the distinction is not important or the setting clear, we will
drop k and A from the notation.
For any smooth variety X , we have the complex [X ] concentrated in
zero. We also have the complex [X ] = [X ] ! [Spec k] sitting in degrees 0
and 1. In this normalization, there is an exact triangle
[X ] ! [X ] ! [Spec k] ! [X ] [1] :
Denition 1.1.3 ([Vo1] after 2.1.3). Let
A(0) = [Spec k] 2 DMeffgm :
It is the unit object for the tensor structure on DMeff
gm . Let
A(1) = [P1 ] [?2] 2 DMeff
gm
be the Tate motive. For k  1 let
A(k) = A(1)
k :
If M is a motive, we put M (k) = M
A(k).

5
 M 7! M (1) is a triangulated functor on DMeff
gm .
Denition 1.1.4 ([Vo1] end of 2.1). The category of geometrical motives
DMgm (k; A) is obtained from DMeff
gm (k; A) by formally inverting the Tate
motive. Explicitly, objects of DMgm are pairs (M; n) with M 2 DMeff gm
and n 2 Z. Morphisms are given by
HomDMgm ((B; n); (C; m)) = klim
0
HomDMeff
gm
(B (n + k); C (n + k)) :
We write A(n) = (A(0); n) for n 2 Z. By construction A(n) = (A(n); 0) for
n  1.
This denition is very much in the spirit of Grothendieck's denition
of pure motives. There is a second category which is a lot more useful in
computations.
Denition 1.1.5 ([Vo1] 3.1.1, 3.1.8, after 3.1.10). A presheaf with trans-
fers on Sm is an additive contravariant functor from SmCor to the category
of abelian groups. It is called a Nisnevich sheaf with transfers if the corre-
sponding presheaf on Sm is a sheaf in the Nisnevich topology (see [Fr] Ch.2).
The category is denoted ShNis (SmCor). A presheaf F on Sm is called ho-
motopy invariant if the natural map F (X ) ! F (X  A 1 ) is an isomorphism
for all smooth varieties X .
The category of motivic complexes DM? (k; A) is the full subcategory
of the derived category D? (ShNis (SmCor)) whose objects have homotopy in-
variant cohomology sheaves.
DM? has a natural t-structure. Its heart is the abelian category of
homotopy invariant Nisnevich sheaves with transfers.
Proposition 1.1.6 ([Vo1] 3.1.2, 3.1.11, 3.1.13). There is a functor
L : SmCor ! ShNis(SmCor)
given by L(X )(U ) = c(U; X )A . It also denes a functor on Var. The nat-
ural inclusion of DM? in D? (ShNis (SmCor)) has a left adjoint. For any
Nisnevich sheaf with transfers F , it is given by the complex C (F )
C?n (F )(X ) = F (X 
n)
where
 is the standard cosimplicial object. The composition C  L induces
a functor
M : DMeff
gm ! DM? :

6
It is fully faithful and hence identies DMeff
gm with a full subcategory of
D M? .
Remark: It is rather formal to show that we get a functor
DMeff gm ! DM? :
However, it is very hard to show that A(1) is quasi-invertible in DMeff
gm
([Vo1] 4.2.4) and hence that DMeff
gm is a full subcategory of D M A
gm (1)
.
is not invertible in DM? .
Remark: Note that M : SmCor ! DM? is induced by a functor to the
category of complexes C 0(SNis (SmCor)).
Denition 1.1.7. The pseudo abelian hull of the image of the composite
functor
M : SmCor ! DMgm ! DM?
is called the category of eective (generalized) Chow motives. The category
of (generalized) Chow motives is obtained from it by formally inverting A(1).
The reason for this terminology will become clear after 1.1.12 below.
DM? is pseudo abelian because the derived category of an abelian category
with enough injectives (as ShNis (SmCor)) is by [Le2] Thm A.5.3. Hence
the pseudo abelian hull of the denition is still a subcategory of DM? .
In particular, A(1)[2] is a Chow motive because it is the cokernel of the
0 1
projector induced by P1 ! Spec k ?! P . By denition

M (P1 ) = A  A(1)[2] :
The splitting is independent of the choice of k-rational point of P1 by ho-
motopy invariance.
Remark: There is some confusion with the older literature. In the category
of Grothendieck motives we would decompose h(P1 ) = Q  Q (?1) and call
h2 (P1 ) = Q (?1) the Lefschetz motive. Its dual then is the Tate motive.
Voevodsky's Tate motive really is the Lefschetz motive in old terminology.
The dierence in signs is only a matter of choice in the denition. It makes
sense because Voevodsky's functor M is covariant whereas Grothendieck's
functor is contravariant.

7
Lemma 1.1.8. The functor M : SmCor ! DM? extends to complexes in
C ? (SmCor). Let f : X ! Y be a morphism of complexes such that all
fn : Xn ! Yn induce isomorphisms M (fn). Then M (f ) is an isomorphism.
Proof. Recall that the functor M is induced by the functor C L which takes
values in the category of bounded above complexes of Nisnevich sheaves
with transfers. More precisely, they are bounded above by 0. It extends by
taking double complexes. There is no problem because in each component
only nitely many direct summands contribute. For the second assertion,
the assumption implies that all H p (C L(fq )) are isomorphisms. A spectral
sequence argument in the surrounding category of Nisnevich sheaves with
transfer shows that M (f ) induces an isomorphism on all H n (C L(f )).
Denition 1.1.9. The full subcategory DMeff
gm? (k; A) of DM? (k; A) gen-
erated by the image of C ? (SmCor) which is triangulated, pseudo abelian and
closed under tensor products is called the category of complexes of eective
generalized Chow motives.
Note that DMeff eff
gm is a full subcategory of DMgm? .
The Tate motive is also used in order to dene motivic cohomology.
Denition 1.1.10 (Voevodsky). Let X be a variety. Then
HMi (X; A(j )) = Hom
DM? (k;A) (M (X ); A(j )[i])
is the motivic cohomology of X .
Morphisms of motives in DM? are \known". At least we relate them
to other theories.
Proposition 1.1.11. Let X and Y be smooth and proper varieties pure of
dimension d and d0 .
HomDM? (M (Y ); M (X )) = CHd (X  Y )
A
where the right hand side denotes cycles of codimension d up to rational
equivalence.
Proof. For a proper variety, M c(X ) = M (X ) (by denition see [Vo1] 4.1).
By [Vo1] 4.2.3 (cf. [Fr] Prop. 4.9 where the sign is correct), 4.2.2 3 and
4.2.5)
HomDM? (Y; X ) = A0;0 (Y; X ) = A?d;0 (Y  X; Spec k)
= HomDM? (M (Y  X ); A(d)[2d]) = CHd (X  Y )A :

8
Corollary 1.1.12 (Voevodsky). The full subcategory of DM? with ob-
jects direct summands of motives M (X ) with smooth and proper X is equiv-
alent to the category of eective Chow motives.
Proof. Clear by denition of the category of Chow motives, e.g. [Sch] 1.2.

Corollary 1.1.13. Let k be a number eld. Then HomDM?(k;Q) (Q ; Q (j )[i])

vanishes in the cases
1. i < 0,
2. i > 1,
3. i = 0, j 6= 0,
4. i = 1, j  0.
If k = Q then HomDM? (Q ; Q (j )[1]) is zero or one-dimensional.
Proof. By [Vo1] 2.2, morphisms of Tate motives are given by higher Chow
groups which are known to be isomorphic to the graded pieces of the -
ltration on higher K -groups. In the case of a number eld the ranks are
known by Borel's result [Bo].
Remark: This theorem will be needed in the computations in the sequel.
Note, however, that we do not need to understand the precise isomorphisms.
1.2 Mixed Tate motives
We now study a particularly simple subcategory of DMgm where the mor-
phisms are understood. The computations in chapter 3 are carried out in
this subcategory. The use of this category for question of this type was sug-
gested to me by Kahn. The main computational tool, the weight ltration
was introduced by Levine in the setting of his triangulated category ([Le1]).
Let k = Q and A = Q .
Denition 1.2.1 (Kahn, Levine). The triangulated category of mixed
Tate motives DMT is the full triangulated category of DMgm generated
by Q (i) for i 2 Z. By DMT N ; DMT N ; DMT [N;M ] we denote the full
triangulated subcategories generated by Q (i) with i  N; i  N; N  i  M
respectively.
The category DMT is closed under tensor products.

9
Lemma 1.2.2. 1. For all N 2 Z, M1 2 DMT N and M2 2 DMT N ?1 ,
we have HomDMT (M1 ; M2 ) = 0.
2. The categories DMT [N;N ] are isomorphic to the category of nite di-
mensional graded Q -vector spaces.
Proof. For Mn = Q (in )[jn ], n = 1; 2, this is a consequence of 1.1.13. If the
assertion is true for two vertices of an exact triangle (in the rst or second
argument), then it is true for the third. This proves 1. For 2. we show by
induction that M 2 DMT [0;0] is direct sum of Q (0)[j ]'s. Consider a triangle
M  M
M! Q (0)ej [j ] ?! Q (0)fj [ j] :
Let j =  jQ(0)ej [j ]. Again by 1.1.13 it maps to Q (0)fj [j ]. Such mor-
phisms are given matrices with rational numbers as entries. Composition
of matrices is composition of morphisms. This means precisely that for
xed j the category of powers of Q (0)[j ] is isomorphic to the category of
nite dimensional Q -vector spaces. We decompose Q (0)ej [j ] = Kj [j ]  Ij [j ],
Q (0)fj [j ] = Ij [j ]  Lj such that j vanishes on Kj and is an isomorphism
on Ij . Then M also stands in the exact triangle
M 0 M
M! Kj [j ] ?! Lj [j ]
hence M 
L L
= Kj [j ]  Lj [j ? 1]. As we have already seen, the morphisms
of such direct sums are the same as the morphisms of graded Q -vector spaces.

Remark: We will construct later (proposition 2.1.7) a faithful bre functor

from DMT [N;N ] to the category of graded Q -vector spaces given by singular
cohomology.
The rst part of this lemma is enough to deduce in a formal way the
existence of an extra structure on DMT .
Proposition 1.2.3 (Levine). For all M 2 DMT and integers N , the
functor
HomDMT (
; M ) : DMT N ! ab
is representable by an object WN M together with a morphism
WN M ! M :
10
 WN is an exact functor DMT ! DMT N . Dually, the functor
HomDMT (M;
) : DMT <N ! ab
is representable by
M ! W<N M :
This denes an exact functor to DMT <N . For each object M there is an
exact triangle
WN M ! M ! W<N M ) :
Let GrN M = W<N +1 WN M be the graded pieces.
Proof. We put N = 0 in order to simplify notations. As always, it is enough
to construct some object W0 M ! M with the universal property. It is
necessarily unique up to unique isomorphism. We start with M = Q (i)[j ].
We put
(
W0 M = M if i  0,
0 else.
It satises the universal property by the lemma.
Now assume W0 is constructed on M1 and M2 . We want to dene it
on M3 which sits in the triangle
M1 ! M2 ! M3 ! M1 [1] :
By the axioms of a triangulated category there is an object W0 M3 sitting in
the triangle
W0 M1 ! W0 M2 ! W0 M3 ! W0 M1 [1]
and a morphism W0 M3 ! M3 such that the obvious morphism of triangles
is dened. Now we have to check the universal property. Let K be an object
with weights less or equal to 0. There are long exact sequences
Hom( K; M1 ) ????! Hom( K; M2 ) ????! Hom( K; M3 ) ????! Hom( [?1]
K ; M1 ) ????! Hom( [?1]
K ; M2 )
x
? x
? x
? x
? x
?
? ? ? ? ?
Hom(
K; W 0 M1 ) ????! Hom( K; W 0 M2 ) ????! Hom(
K; W 0 M3 ) ????! Hom( [?1] 
K ;W 0 M1 ) ????! Hom( [?2] 0
K ;W M2 )
The outer arrows are isomorphisms by assumption. By the ve lemma
the middle arrow is also an isomorphism. The same proof works for the
dual assertion. The universal property allows to construct WN and W<N
on morphisms. For M = Q (i)[j ] the exact triangle clearly exists. By con-
struction and lemma 1.2.4 below we get exact triangles as claimed.
11
Remark: Usually it is not a good idea to dene a functor by choosing the
third vertex of a triangle. In our case it is well-dened because we also have
the universal property working for us.
The above amounts to the construction of the weight coltration on
Tate motives. Application of contravariant realization functors maps it to
the weight ltration on the realization side. Objects of DMT [N;N ] are (tri-
angulated) pure Tate motives of weight 2N depending on conventions.
The category of pure Tate motives of xed weight is the derived category
of a semi-simple abelian category. Obviously there is a \weight spectral
sequence" with initial terms the graded pieces and converging to M .
Lemma 1.2.4. Let D be a triangulated category. Consider the diagram of
exact triangles in D
A1 ???! A2 ???! A3
? ?
? ?
y y

B1 ???! B2 ???! B3 :
? ?
? ?
y y

C1 C2
Then there is an object C3 and morphisms such that the last line and row
are also triangles.
Proof. [BBD] 1.1.11.
1.3 More Technical background
Let again k be a eld of characteristic zero and A = Z or Q .
Denition 1.3.1. Let K be a complex in SmCor. The stupid ltration
n K is dened by
(
(n K )i = K for i  n,
i
0 for i < n.
n K is a subcomplex of K . Its graded pieces are the K i.
Let L be an object in DM? . The canonical truncation nL is the
truncation functor on DM? with respect to the t-structure with heart the
homotopy invariant sheaves with transfers. It is a quotient complex of L.

12
 n is a functor on DM? whereas n does not pass to DMgm . Note
that
nM (K ) = n M (n?1 K ) :
Proposition 1.3.2. Let K ! K 0 be a morphism in C ?(SmCor). Assume
that it induces isomorphisms
n K ! n K 0
in DMgm for all n. Then M (K ) ! M (K 0 ) is an isomorphism in DM? .
Proof. We consider the corresponding objects in the derived category of
Nisnevich sheaves with transfers. We have to check that
H i (M (K )) ! H i (M (K 0 ))
is an isomorphism for all i where H i denotes the cohomological functor with
respect to the natural t-structure. Only nitely many K k contribute to
H i(M (K )), namely those with k  i ? 1. In other words,
H i (M (K )) = H i (M (i?1 K )) :
Hence the assumption on the stupid ltration is enough to prove the quasi-
isomorphism.
Lemma 1.3.3. For K 2 C ?(SmCor) we have in DM?
M (K ) = lim
?! M (nK ) :
Proof. Clearly, the equality holds in the abelian category C ? (ShNis (SmCor)).
We have to prove that it passes to D? (ShNis (SmCor)). Assume we are given
a direct system of morphisms in the derived category
fn : M (n K ) ! L :
They are represented by morphisms of complexes
0
fn
M (n K ) ?! Ln g?n? L
where gn is a quasi-isomorphism. The Ln can be chosen such that the fn0 and
the gn give a direct system up to homotopy. Let L1 the direct limit of the

13
Ln . As direct limits are exact in ShNis (SmCor) it is also quasi-isomorphic
to L. Let
fn00 : M (n K ) ! Ln ! L1 :
They form a direct system up to homotopy. Let si be the stupid ltration
of complexes of Nisnevich sheaves, i.e.
(
(si C )p =
C p for p  i
0 else.
Note that siM (n K ) = si M (K ) for i  n. Hence fn00 induces a direct system
of maps up to homotopy on si M (K ). We can dene their limit f on M (K )
by descending induction. Assume f is dened on snM (K ). f and fn00?1 dier
by a homotopy h : M (K ) ! L? 1 1 . We modify f by hn  d on M (K )n
and extend it to M (K ) using fn00?1 jM (K )n? . This gives a new morphism
n? 1 1

of complexes sn?1 M (K ) ! L1 which is homotopic to fn00?1 . Clearly the

morphism f is homotopic to fn00 on the subcomplex M (n K ) as well.
This lemma allows to reduce all questions on DMeff
gm? (see denition 1.1.9)
to questions in DMgm .
Proposition 1.3.4. Let M (K ) and M (L) be in DMeff
gm? . Then
HomDMeff? (M (K ); M (L)) = lim
? HomDMeff? (M (n K ); M (L)) :
gm gm
Moreover, for xed K and n there is N 2 N such that
HomDMeff? (M (n K ); M (L)) = HomDMgm (M (n K ); M (N L)) :
gm
Proof. The rst equality is nothing but the lemma. In order to show the
existence of N we can assume that K is of the form M (X )[p] for some p
and some smooth variety X . Then by [Fr] 3.3
p (X; M (L)) :
HomDMeff? (M (K ); M (L)) = HZar
gm
X is nite dimensional hence Zariski cohomology has nite cohomological
dimension. Hence for N small enough
p (X; M (L)) = H p (X;  M (L)) = H p (X;  M (
HZar Zar N Zar N N ?1 L)) :

14
Lemma 1.3.5. Let X be a variety such that M (X ) is a mixed Tate motive.
Then >0 Grn M (X ) = 0 for all n.
Let X be in C b (SmCor) such that all Xi are mixed Tate motives. Then
>0 Wn M (X ) = >0 Wn 0 M (X ) :
L
Proof. By assumption Grn M (X ) = ji Q (n)[ji ]. It follows from Hodge
theory, [De1] Thm 8.2.4, that 0  ji  n. Now we use the fact that
>nQ (n) = 0. ([Ka] 2. property C or from 1 M (G nm ) = 0). For the
second assertion it suces to show that
0 Wn (M (X )) = 0
for complexes concentrated in negative degrees. On the level of Grn this
follows from the rst part. By induction on the weight ltration it follows
for all Wn.
Proposition 1.3.6. Let X and Y be objects of C ?(SmCor) such that all
Xi and Yi are mixed Tate motives. Let
f : X  ! Y
be a morphism in DM? . For all integers N and Tate weights n (see 1.2.3)
we assume that kernel and cokernel of the morphism
Grn ?N X ! Grn ?N Y
is direct sum of Q (n)[j ] with j  N . Then f is an isomorphism.
Proof. Recall that ?N +1 Q (n)[j ] = 0 for j  N + n. By assumption the
map
Grn ?N X ! Grn ?N Y
fails to be an isomorphism by direct sums of Q (n)[j ] with j  N . Hence
application of ?N +1 yields that
(*) ?N +1 Grn ?N ?n X ! ?N +1 Grn ?N ?n Y
is an isomorphism for all n and N . We want to show from () that ?N +1 (f )
is an isomorphism for all N . Only nitely many weights occur in ?N X
and ?N Y . Let n0 be the maximal one. The weight ltration has at most
n0 non-trivial steps because all Xi and Yi are eective. By assumption (),
15
?N ?n +1 Grn ?N ?2n (f ) is an isomorphism for all weights up to n0. By
0 0
induction on the exact triangles for the weight ltration (see 1.2.3) and the
ve lemma in each step we deduce that
?N +1 Wn ?N ?2n (f )
0 0

is an isomorphism. By lemma 1.3.5 this implies that

?N +1 Wn ?N (f ) = ?N +1 ?N (f ) = ?N +1 (f )
0

is an isomorphism as claimed.

2 The realization functor

We construct triangulated realization functors from DMgm (k; A) to vari-
ous triangulated categories. On the level of cohomology objects this corre-
sponds e.g. to the l-adic or Hodge realization. In order to stress the logic
we rst give an axiomatic construction. Then we recall the denition of
the \derived" category of mixed realizations. In a last step we apply the
construction to this case. The main result is 2.3.3 and its corollaries.
2.1 Axiomatic construction
Again k is a base eld of characteristic zero. In this section we assume
throughout that A is an abelian category with enough injectives in which
arbitrary direct sums exist and are exact. We start with a contravariant
functor
R~ : Sm ?! C 0 (A) :
The aim of this section is to deduce from it (under the assumption of certain
axioms) a triangulated functor
R : DMgm ?! D+ (A):
Such a functor is called a realization functor. All realization functors induce
by functoriality regulator maps, i.e. transformations of functors from Var to
ab
i (?; Z(n)) ?! H i (?; n) := Hom
HM R D(A) (R(Z(n)); R(?)) :
For technical reason we are also interested in extensions of R to DMeff
gm? .

16
Example: In the case k = C , let Q be the category of Q -vector spaces and
R~ sing the singular cochain complex, i.e., for a smooth variety X , let
R~ sing (X ) = Q [Cont(
 ; X (C ))]_
where X (C ) is considered as a complex manifold,
n is the standard topo-
logical n-simplex, Cont denotes continuous maps of topological spaces and

_ is the Q -dual. By denition
H i (R~sing (X )) = Hsing
i (X (C ); Q ) :

Clearly, R~ sing (X ) is a functor to C 0 (Q ).

Denition 2.1.1. A functor

R~ : Sm ?! C 0 (A)
or
R~ : Var ?! C 0 (A)
has descent for open covers respectively for proper covers, if application of
R~ to the covering map for the nerve of a Cech-covering respectively for a
proper hypercovering yields a quasi-isomorphism . It satises the homotopy
property if for all varieties X the morphism
R~ (X ) ! R~ (X  A 1 )
is a quasi-isomorphism.
The rst realization result is rather obvious:
Proposition 2.1.2.  Assume we have a functor
R~ : SmCor ! C 0 (A)
which has descent for open covers and satises the homotopy property.
Then it extends to an exact functor
R : DMeff
gm? ! D (A) :
+

 Assume in addition that A is a tensor category and R~ is compatible

with tensor products. Then R is a triangulated tensor functor.

17
  Finally, if the functor

R(Q (1)) on D+(A) is an equivalence of
categories, then R extends to a functor
R : DMgm ! D+(A) :
Proof. Applying R~ to objects of C b (SmCor) we get double complexes. We
take the associate simple complex in C + (A). This involves a universal choice
of signs which we x once an for all. R~ is a functor
K b (SmCor) ! K +(A) ! D+ (A) :
(See [Hu1] 2.2.3 for details on signs.) All other statements are immediate.
We obtain
R : DMgm ! D+(A) :
Finally we extend it to DMeff gm? . On objects the same denition as on
bounded complexes works. On morphisms we use the description of lemma
1.3.4 to reduce to the bounded case.
Remark: It is tempting to weaken the assumption of the proposition to
functors with values D(A) or even triangulated categories in general. How-
ever, the above proof does not work in that generality. In fact, all construc-
tions of Chern classes or cycles classes on higher Chow groups - a question
very similar to the above - assume the existence of functorial complexes
rather than objects in a derived category.
Example: (Etale cohomology) Fix a prime l and an integer n. We choose
an injective resolution I of Z=ln on the etale site on the category of smooth
schemes over our base eld k. Let M (X )et
Z=ln be the etale sheacation
of the complex of Nisnevich sheaves modulo ln. We put
R~ ?(X; Z=ln) = HomShet (M (X )et
Z=ln; I ) :
Clearly this is a contravariant functor
SmCor ! C (ab) :
By [Vo1] Prop. 3.2.3 the cohomology of this complex computes the etale
version of motivic cohomology or equivalently etale cohomology of X . Let
R~ (X; Z=ln) be the sheacation of R?(Xk0 ; Z=ln for nite eld extensions
of k in the etale topology. Clearly R~ (X; Z=ln) computes Rp (Z=ln)X if

18
p : X ! Spec k is the structural morphism. From this we get the exact
realization functor
R : DMgm ! Db ((Spec k)et ; Z=ln)
to the derived category of ln-torsion Galois modules. Formulated like this,
many properties of etale cohomology, e.g. localization sequences for smooth
pairs, are a consequence of the existence of the functor. By functoriality, we
get regulator maps
HM i (X; Z(j )) ! H i (X; 
nj ) :
et l

Example: (Continuous etale cohomology) Let R(X; Zl) be the projective

system of complexes R(X; Z=ln) for varying n. It allows to dene an exact
realization functor
R : DMgm ! Db ((Spec k)et ; Zl)
into Ekedahl's triangulated category of Zl-sheaves ([Ek]). By functoriality,
we get regulator maps
HM i (X; Z(j )) ! H i (X; Z (j ))
cont l
into Jannsen's continuous etale cohomology.
The rest of this section will be concerned with giving criteria for the
existence of functors on SmCor.
Proposition 2.1.3. Let R~ : Var ! C 0(A) be a contravariant functor
which is also covariant for all nite and surjective maps between irreducible
varieties.
 We assume that for a nite surjective map f : X 0 ! X the composition
f  ~ 0 f ~
R~ (X ) ?! R(X )) ?! R(X )
equals multiplication with the degree of f .
 We also assume that f is compatible with direct products in the fol-
lowing sense: for f as before and all Y the diagram commutes:
R~ (Y  X 0 ) ??? R~ (X 0 )
? ?
(idY f ) ?
y
?
yf

R~ (Y  X ) ??? R~ (X )

19
  Let g : Y !` X be an arbitrary map and f a closed immersion. We
put Y 0 = Ci the disjoint union of the irreducible components of
X 0 X Y with reduced structure. For all i let ni be the ramication
index of X 0 ! X at Ci (see [Fu] Ex. 4.3.7). We assume that
g0  ~ 0
R~ (Y 0) ??? R(X )
P nifC ??y ?
?
yf
i
g 
R~ (Y ) ??? R~ (X )
also commutes.
Then R~ extends to a functors on SmCor.
Remark: If R~ has descent for open covers
eff
and satises the homotopy prop-
erty, then it extends at least to DMgm . Hence we have potentially two
denitions on singular varieties - the original one and the one from DMgm .
We do not know if they agree in general.
Proof. Let X and Y be smooth connected varieties. Let ? in X  Y be an
irreducible subvariety which is nite and surjective over X . Consider the
diagram
X ?!Y :
It induces a map
R(?) : R~ (Y ) ! R~ (?) ! R~ (X )
by composition of contravariant and covariant functoriality. If ? is the graph
of a morphism f , then R~ (f ) = R(?).
We claim that R is a functor. Let ?1  X  Y and ?2  Y  Z be prime
correspondences. Then
X
(?1  Z ):(X  ?2 ) = ni Ci
is a cycle in X  Y  Z . The Ci are the irreducible components of ?1 Y ?2 .
Let pr be the projection map to X  Z . The prime correspondence Ci is
nite and dominant over its image pr(Ci ). Let di be the degree of this
covering. The composition of correspondences is given by the cycle
X
?2  ?1 = nidi pr(Ci) :
20
 The morphism
R(?2  ?1 ) : R(Z ) ! R(?2  ?1 ) ! R(X )
can equivalently be computed via R(?1 :?2 ). Note that the multiplicities
work out as they have to.
The intersection multiplicity ni is equal to the ramication index of
X  ?2 ! X  Y at Ci (see [Fu] 4.3.7) by loc. cit. 7.1.15 and because Z is
at over k. By assumption
`
R~ ( Ci ) ??? R~ (X  ?2 ) ??? R~ (?2 )
? ? ?
R~(
)?
y
?
y
? ~ 00
yR (
)

R~ (?1 ) ??? R~ (X  Y ) ??? R~ (Y )

commutes. This implies
R(?2  ?1 ) = R~ (?1 )  R~ (?2 ) :

By Deligne's method we can often extend a functor from Sm to Var.

This will allows to weaken our assumptions further.
Convention: Let V2 be a category and V1 a subcategory. Typically, V1
will be Sm and V2 will be Var or SmCor. Given a functor R~ : V1 ! C (A)
we say that it extends to V2 if there is a functor R : V2 ! C (A) and a
natural transformation  : R~ ! R of functors on V1 such that all (?) are
quasi-isomorphisms.

Proposition 2.1.4. Let R~ : Sm ?! C 0(A) be a functor with descent for

proper covers. Then it extends to a functor R~ 0 : Var ?! C 0 (A). R~ 0 has
descent for proper covers on Var as well.
Proof. Let X be a variety. By [De1] 6.2, there exists a proper hypercovering
of X such that all components are smooth. We put
R~ 0 (X ) = lim Tot(R~ (X
))
?!
where X
runs through the inverse system of all proper hypercoverings of X
by smooth varieties. The construction works because direct limits exist in
A. Descent for proper covers is a consequence of the same property for the
original functor.
21
Example: The functor R~ sing has descent for proper covers (e.g.[Hu1] 6.2.2).
By abuse of notation we call the extension also R~ sing .
Lemma 2.1.5. Let A be Q -linear. Let G be a nite group operating on an
object K 2 A. Let K G be the subobject of G-invariant elements, i.e., the
intersection of ker(g ? id) for all g 2 G. Then the functor
G is exact and
there is a canonical section K ! K G of the inclusion.
Proof. The section is given by
 = #1G g :
X

g2G
This is where we need invertibility of #G. Using the section it is easy to
show that short exact sequences remain exact.
Theorem 2.1.6. Let A be a Q -linear abelian category with enough injec-
tives in which arbitrary direct sums exist. Let
R~ : Sm ?! C 0(A)
be a functor. We assume:
 R~ has descent for open and proper covers and satises the homotopy
property (cf. 2.1.1)
Let R~ also denote the extension to Var (cf. 2.1.4).
 For any surjective and nite morphism f : X ! Y between normal
varieties the group G = AutY (X ) operates on R~ (X ) by contravariant
functoriality. If the covering is generically Galois with Galois group
G, then we assume that
 R~ (X )G
R~ (Y ) ! R~ (X ) ?!
is a quasi-isomorphism.
Then R~ extends to a functor
R~ 00 : SmCor ! C 0 (A)
together with a natural transformation of functors on Sm
 : R~ ! R~ 00
22
such the (
) are quasi-isomorphisms. R~ 00 induces a functor
R : DMeff
gm? ! D (A) :
+

Finally assume that A is a tensor category and R~ is compatible with tensor

products. Then R is a triangulated tensor functor. Assume
 The functor

R(Q (1)) on D+(A) is an equivalence of categories.
Then R passes to a functor
R : DMgm ! D+ (A) :
This can be seen as a prototype theorem. The same methods also
works in other settings. An important such case is the realization func-
tor RMR : Sm ! CMR , see theorem 2.3.1.
Remark: Levine also proves a realization theorem in the setting of his tri-
angulated category ([Le3] Part I Ch. V). It has a dierent avour from the
above. His axioms seem to amount to a contravariant functor which is co-
variant for proper maps between smooth schemes plus the existence of cycle
classes.
Before actually proving the theorem, we want to consider our example:
Proposition 2.1.7. R~sing satises all conditions of the theorem. It extends
to a functor
Rsing : DMgm ! D+ (Q ) :
Proof. We have to check several properties of singular cohomology of com-
plex analytic spaces. Proper descent was mentioned above. Descent for
open covers is nothing but the Mayer-Vietoris sequence. The only non-
trivial statement is the isomorphism
i (Y; Q ) ! H i (X; Q )G
Hsing sing
for nite covering maps f : X ! Y (between normal spaces) which are
generically Galois with Galois group G. Note that G operates on X by
functoriality of the normalization of X in K (Y ). By the Leray spectral
sequence it is enough to show the isomorphism for the direct image of the

23
constant sheaf Q (there are no higher direct images for a nite morphism).
Its stalks are given by [GraRem] Ch. 2 x3.3
Y
(f Q )y = Q :
x2f ?1 (x)
The covering group AutY (X ) permutes these factors. On the generic bre
it operates transitively by assumption. G operates transitively on all bres,
e.g. [Ma] Theorem 9.3. Hence (f Q )Gy is at most one-dimensional. As the
image of Q in f f Q is invariant under G it is at least one-dimensional.
Corollary 2.1.8. Let k = Q . Then Rsing identies the category DMT [N;N ]
of Tate motives with xed weight with the category of nite dimensional
graded Q -vector spaces.
Proof. Clear because Rsing (Q (N )) = Q and morphisms are the same.
In the rest of this section, we are going to prove our theorem in several
steps. The main step is the following remark:
Lemma 2.1.9. Let X be a normal variety 0 and R~ a functor on Var as in
the theorem. Let R~ 0 (X ) = lim? ! R~ (X 0 )Aut(X =X ) where the X 0 run through
the category of normal X -schemes which are nite surjective over X with
generically Galois covering map. Then R~ 0 is a contravariant functor on
normal varieties and also covariant for nite surjective maps between them.
The morphism R~ (X ) ! R~ 0 (X ) is a quasi-isomorphism.
Proof. As transition maps in the system we use only nite surjective X -
morphisms which are generically0 Galois. The system is well-dened. By
assumption R~ (X ) ! R~ (X 0 )Aut(X =X ) is a quasi-isomorphism. By exactness
of direct limits in A this remains true in the direct limit. We rst check con-
travariant functoriality. Let X ! Y be an arbitrary morphism of normal
varieties and Y 0 a covering of Y in our direct category. Then the normal-
ization X 0 of Y 0 Y X is also nite and surjective over X . All irreducible
components are generically Galois by the lemma below. We thus get a mor-
phism of direct systems. For covariant functoriality, consider nite surjective
morphism  : X ! Y and a covering X 0 of X in the direct system. There
is another such covering of X 0 which is also generically Galois over Y . Let
 be the projection
R~ (X 0 ) ! R~ (X 0 )Aut(X 0 =Y ) :

24
It induces a map of direct systems and hence
 : R~ 0(X ) ! R~ 0 (Y ) :

Note that in this normalization the composition


 ~0  ~ 0
R~ 0(Y ) ?! R (X ) ?! R (Y )
is the identity.
Lemma 2.1.10. Let Y; Y 0 be normal irreducible varieties in characteristic
zero and let  : Y 0 ! Y be nite surjective and generically Galois with
covering group G. Let
f :?!Y
be a morphism. Then G operates transitively on the bres of
? Y Y 0 ! ? :
Let C be the reduction of an irreducible component of ? Y Y 0 . Then C ! ?
is surjective and generically Galois. The Galois group is the quotient of the
stabilizer of C in G by the xgroup of C .
Proof. First we assume that ? = Spec L is a (not necessarily closed) point
of Y . We can assume that Y = Spec A; Y 0 = Spec B are ane. Now
the assertion is easy to check directly using the fact that B G = A and that
everything is Q -linear. More generally let ? = Spec L where L is a eld. The
reduction of ? Y Y 0 = ? f (?) ?1 (f (?)) is given by ? f (?) ?1 (f (?))red .
Hence the assertion follows from the special case.
Clearly the map ? Y Y 0 ! ? is surjective. G stabilizes the subvariety of
components which dominate ?. Moreover the operation of G is transitive on
all bres, hence the closure of the bre over the generic point is everything.

Proof. (of Theorem) We only have to extend R~ 00 of the lemma to the category
of smooth correspondences. The method is the same as in the proof of 2.1.3.
Let X and Y be smooth connected varieties. Let ? in X Y be an irreducible
subvariety which is nite and surjective over X . Let ?~ be its normalization.
Consider the diagram
X ?~ ! Y :
25
It induces a map
R~ 00(Y ) ! R~ 00(?~ ) ! R~ 00 (X )
by composition of contravariant and covariant functoriality. Let the mor-
phism R(?) be this composition times the degree of ? over Y . If ? is the
graph of a morphism f , then R~ 00 (f ) = R(?) in D+ (A).
We claim that R is a functor. Let ?1  X  Y and ?2  Y  Z be prime
correspondences. Then
X
(?1  Z ):(X  ?2 ) = ni Ci
is a cycle in X  Y  Z . The Ci are the irreducible components of ?1 Y ?2 .
Let pr be the projection map to X  Z . The prime correspondence Ci is
nite and dominant over its image pr(Ci ). Let di be the degree of this
covering. The composition of correspondences is given by the cycle
X
?2  ?1 = nidi pr(Ci) :
P
1 :?2 =
Let ?^ niC~i be the normalization of ?1 :?2. The morphism
R(?2  ?1 ) : R(Z ) ! R(?^
2  ?1 ) ! R(X )

can equivalently be computed via R(?^ 1 :?2 ). Note that the multiplicities
work out as they have to.
The intersection multiplicities ni are equal to the ramication index of
X  ?2 ! X  Y at Ci (see [Fu] 4.3.7) by loc. cit. 7.1.15 and because Z is
at over k. Let ?02 be normal variety, nite and surjective over ?2 which is
Galois over Y . We put
X
?1 :(X  ?02 ) = n0ij Cij0
where the Cij are the irreducible components of ?1 Y ?02 covering Ci and
the n0ij are the ramication indices of X  ?02 ! X  Y . By the last lemma
the covering of ?1 is generically Galois, hence all degrees d(Cij0 =?1 ) = d0
agree. As Y is smooth all ramication indices are equal to
0 )
e0 = dd0
(?#2f=Y
Cij0 g
by the degree formula loc. cit. 4.3.7. By the same degree formula, we can
replace R(?^
1 :?2 ) by R(?1 :X
^  ?02) in the computation of our morphism.
26
 Finally we have to show that the diagram
`
R~ 00 ( C~ij0 ) ??? R~ 00(?~ 02 )
? ?
e0 d0 R~00 (
)?
y
? 0
yd(?2 =Y )R~ 00 (
)

R~ 00 (?~ 1 ) ??? R~ 00 (Y )
commutes. For this we have to go back to the denition of R~ 00 . We use the
original contravariant functoriality and then project to the invariants under
G. We have already seen that the multiplicities t. The rest of the theorem
follows from 2.1.2.
2.2 Review of mixed realizations
We review the basic notions of [Hu1], i.e., dene the category of mixed
realizations and the surrounding triangulated category. Everything in this
section is pure linear algebra. Let k be a eld of characteristic zero which
can be embedded into C . Let S be the set of embeddings.
We rst recall the denition of MR. It is a slight modication of
Jannsen's in [Ja]. It is equivalent to the notion of absolute Hodge motive
which was independently given by Deligne [De2].
Denition 2.2.1 ([Hu1] 11.1.1). An object A in the category of mixed
realizations MR is given by the following data:
 a biltered k-vector space ADR;
 for each prime l a ltered Q l -vector space Al with a continuous oper-
ation of Gk ;
 for each prime l and each  2 S a ltered Q l -vector space A;l ;
 for each  2 S a ltered Q -vector space A ;
 for each  2 S a ltered C -vector space A;C ;
 for each  2 S a ltered isomorphism
IDR; : ADR
 C ?! A;C ;
 for each  2 S a ltered isomorphism
I;C : A
Q C ?! A;C ;

27
  for each  2 S and each prime l a ltered isomorphism
I;l : A
Q Q l ?! A;l ;
 for each prime l and each  2 S a ltered isomorphism
Il; : Al
Q Q l ?! A;l :
Additionally we require that the tuples (A ; ADR ; A;C ; IDR; ; I;C ) give Hodge
structures ([Hu1] 8.1.1) and that the Al are constructible Galois modules
equipped with the ltrations by weights ([Hu1] 9.1.4).
Morphisms of mixed realizations are morphisms of this data compatible
with the comparison isomorphisms.
MR is an abelian category because morphisms are automatically strictly
compatible with all ltrations. Kernels and cokernels are computed compo-
nentwise. Recall that a morphism between ltered objects is called strict if
coimage and image are isomorphic as ltered objects.
Now we need to recall the denition of the category DMR , cf. [Hu1]
11.1.3. It should be thought of as the derived category of MR.
Denition 2.2.2. Let C + be the category with objects given by a tuple of
complexes in the additive categories in the denition of MR plus ltered
quasi-isomorphisms between them. Let CMR be the subcategory of complexes
with strict dierentials whose cohomology objects are in MR. Let DMR be
the localization of the homotopy category of CMR (see [Hu1] 4.1.5) with
respect to the class of quasi-isomorphisms (see [Hu1] 4.1.7).
Lemma 2.2.3. Morphisms of objects in CMR induce strict morphisms on
cohomology. In particular, the category is abelian. DMR is a triangulated
category with t-structure whose heart is MR.
Proof. The rst assertion holds because morphisms in MR are automati-
cally strict. The second is [Hu1] 11.1.4.
We need two lemmas which were implicit in [Hu1] but not stated.
Lemma 2.2.4. Let Ki for i 2 I be a direct system of complexes in CMR.
Assume that all direct limits lim
?! H k (Ki ) exist in MR. Then lim
?! Ki exists
in CMR .
Proof. Direct limits exist in C + . The direct limit functor is exact, hence
strictness of dierentials is preserved and cohomology commutes with the
functor. The second condition on objects in CMR holds by assumption.
28
A direct system where all transition maps are quasi-isomorphisms is a special
case of a direct system to which the lemma applies.
Lemma 2.2.5. Let K  be a bounded below complex of objects in CMR which
are concentrated in positive degrees. Then the total complex Tot K  is in
CMR .
Proof. We can take the total complex in C + as usual. Strictness of dieren-
tials of the total complex is [Hu1] 3.1.8. Clearly the cohomology of the total
complex is obtained from the cohomology of the K i via a spectral sequence.
The boundedness conditions ensure that only nitely many cohomology ob-
jects of the K i contribute to one cohomology object of the total complex.
In particular all vector spaces involved in the denition are indeed nite di-
mensional. We have to check that the ltration on the Galois modules is the
ltration by weights, i.e. that GriW is pure of weight i. We pass to graded
pieces in the spectral sequence. This is possible because the dierentials
are strict. Note that puritity of weight i is stable under subquotients and
extensions. For the Hodge condition we also pass to the weight graded piece.
We now have to check the condition of a pure Hodge structure. Again the
condition is stable under extensions.
Proposition 2.2.6. The category DMR is pseudo-abelian.
Proof. This is not a special case of Levine's result in [Le2] A.5. However,
his proof can be modied so that it works in our case. Let C be an object
of DMR and p : C ! C an idempotent, i.e. p2 = p. p is represented by a
morphism of complexes p(1) : C ! C (1) where C (1) is quasi-isomorphic to C
via c(1) . Choose C~ (1) such that the diagram
p
C ???! C (1)
(1)

? ?
? ?
c(1) y c~ y
(1)

p~
C (1) ???! C~ (1)
(1)

commutes. The equation p2 = p implies that there is a quasi-isomorphism

C~ (1) ! C (2) such that p~(1)  p(1) and c~(1)  p(1) become homotopic. In fact,
there is a whole chain or morphisms of complexes
p(n) : C (n?1) ! C (n)
where all C (n) are quasi-isomorphic to C (x the quasi-isomorphisms c(n)
once and for all), p(n) represents p and p(n+1)  p(n) and c~(n)  p(n) are
29
homotopic. By 2.2.4 the limit over C (n) with transition maps the quasi-
isomorphisms c(n) exists in CMR . Now p is represented by the induced
morphism of complexes
?! C ! lim
(n) (n) (n)
?! p : lim
lim ?! C
Replace C by the limit. By this procedure we have succeeded in representing
p by an endomorphism of a complex such that the identity p2 = p holds up
to homotopy of complexes. From now we can argue precisely as Levine in
loc. cit. Theorem A.5.3. There is one little change, however, because the
maps f (p) and f (id) (notation of loc.cit.) are not homotopy equivalences
but only quasi-isomorphisms. This suces for the argument.
2.3 The mixed realization functor
We proceed by constructing a realization functor from Voevodsky's geomet-
rical motives to mixed realizations.
One of the main results of [Hu1] was the following:
Theorem 2.3.1 (loc. cit. 11.2). There is a contravariant functor
R~ MR : Sm ! CMR
whose cohomology objects compute the mixed realizations of a smooth variety.
Composed with the natural projections to the category of Galois modules
or to the category of Hodge complexes it computes the l-adic realizations
respectively the Hodge realization of a variety.
R~ MR has descent for proper hypercovers, hence the functor extends as
in 2.1.4 (cf. loc. cit. 11.2.2) to all varieties.
Denition 2.3.2 (loc. cit. 11.3.1). Let
DMR (Q (?n); RMR (X )[i])
i (X; n) = Hom
HMR
be the absolute mixed realization cohomology.
As shown in loc. cit. part III this is part of a Bloch-Ogus cohomology
theory. By functoriality, there is a map to absolute Hodge cohomology and
to continuous l-adic cohomology. Note, however, that giving an element in
absolute realization cohomology is stronger than giving elements in these
standard cohomologies. This is parallel to the fact that giving a mixed
realization is stronger than giving a mixed Hodge structure and various
Galois-modules - we also x comparison isomorphisms.
We immediately get:
30
Theorem 2.3.3. Let k be a eld which is embeddable into C and A = Z; Q.
R~ MR extends to contravariant functors
RMR :DMgm (k; A) ! DMR ;
RMR :DMeff
gm? (k; A) ! DMR :
It maps the Tate motive A(n) to Q (?n) (cf. the remark after 1.1.7). In
particular, it induces a transformation of functors
i (X; A(n)) ?! H i (X; n)
HM MR
which is compatible with all structures (products, localization sequences etc.).
Proof. We repeat the proof of section 2.1. CMR itself is not a category of
complexes over an abelian category and certainly arbitrary direct limits do
not exist. However, the constructions of theorem 2.1.6 go through by lemma
2.2.4 and 2.2.5. We only have to check that R~ MR satises the conditions
of theorem 2.1.6. All of them can be checked in the singular component.
Hence they hold by proposition 2.1.7. The realization of the Tate motive
can be computed in DMR , e.g. as the decomposition of of P1 in the category
of Chow motives ([Hu1] 20.2.1). The transformation of functors is nothing
but functoriality.
Remark: The same theorem (with the same proof) also holds for the more
rened functor RMRP with values in the category DMRP which takes into
account the polarizability of the graded pieces with respect to the weight
ltration (see [Hu1] Ch. 21.)
By functoriality (or using the same arguments again), the theorem also
implies the existence of other realization functors.
Corollary 2.3.4. Let Dl be the \derived category" of constructible Q l -sheaves
on Spec(k) in [Ek] (or in the number eld case the rened version in [Hu2]).
Then there is a realization functor
DMgm (k; A) ?! Dl :
It induces a transformation of functors
i (X; A(n)) ?! H i (X; Q (n))
HM cont l
i is continuous etale cohomology respectively the horizontal ver-
where Hcont
sion of [Hu2].

31
Corollary 2.3.5. Let DH be the category of Hodge complexes as in [Be1]
3.2 or [Hu1] 8.1.5. Then there is a realization functor
DMgm (C ; A) ?! DH
It induces a transformation of functors
i (X; A(n)) ?! H i (X; Q (n))
HM H l
where HHi is absolute Hodge cohomology as introduced by Beilinson in [Be1].
Beilinson's category of Hodge complexes diers from Deligne's by decalage
of the weight ltration. Note also that absolute Hodge cohomology agrees
with Deligne cohomology in the good range of indices, see [Be1] 5.7.
Other regulators which we get from this are to De Rham cohomology,
singular cohomology again, and geometric etale cohomology. This is cer-
tainly not a surprise. The existence of such functors is already stated in
[Vo2].
Recall ([Hu1] 22.1.3) that an object of MR is called motivic if it is
subquotient of an object H i (RMR (X )) where X is a complex of varieties
with morphisms formal Q -linear combinations of morphisms of varieties.
Theorem 2.3.6. Let X be an object of DMeff i
gm? . Then H (RMR (X )) is
motivic.
Proof. As H i (RMR (X )) only depends on ?i?1 X we can as well assume
X 2 DMeff gm . First consider the special case of a complex of length one, i.e.
f
X =P[X0 ?! X1 ]. The morphism f in SmCor is nite linear combination
f = i fi with i 2 Q and fi a primitive nite correspondence. Recall
that it is nite over a connected component of X0 . Let Yi = supp(fi ) and Y~i
a normal nite cover of it. It is nite surjective over a connected component
of X0 . We assume that this cover is generically Galois. Let X00 be the union
of those connected components of X0 which are not covered by any Yi . Let
a
X~ 0 = X00 q Y~i :
P
Let fi0 be the projection map Y~i ! X1 and f~ = i fi0. Now we put
~
f
X~  = [X~ 0 ?! X1 ] :

32
By construction of the realization functor for correspondences the diagram
~
R~MR (X~ 0 ) R?????
MR (f ) ~
RMR (X1 )
x x
? ?=
? ?
~ MR (f~)
R~MR (X0 ) R????? R~MR (X1 )
commutes and the left vertical map has a compatible splitting. Hence
RMR (X ) is a direct summand of RMR (X~ ). Clearly cohomology of RMR (X )
is a direct summand of the cohomology of RMR (X~ ).
Now we have to extend this to longer complexes. We do this inductively.
Assume X is a complex in degrees ?n to k with Xi a general variety for
i < 0 and Xi a smooth variety for i  0. We assume that the boundaries
in negative degrees are linear combinations of morphisms of varieties and
the boundaries in positive degrees are nite correspondences. It can be
considered as an object in DM? because the functor M on SmCor is also
dened on Var. Apply the previous construction to X0 ! X1 . This yields
a nite covering X~ 0 of X0 . We have to construct X~ k for k < 0. Let
ffi : Xp ! X0 g
be the set of morphisms which occur as compositions of the morphisms of
varieties making up the complex X . Let
a
X~p = fi X~ 0
where
fi X~ 0 = (Xp X ;fi X~ 0 )red :
0

By lemma 2.1.10 its components are surjective over Xp . The covering group
operates transitively on all bres.
We have to dene boundary morphisms in X~ p such that
X~  ! X
is a morphism of complexes for   0. For simplicity, assume that Xp is con-
nected, let g : Xp ! Xp+1 one of the morphisms occurring in the boundary.
For each of the morphisms fi : Xp+1 ! X0 , we lift g to g(fi ) : (gfi ) X~ 0 !
fi X~ 0 . The coecient of g(fi ) is taken as the coecient of g. It is easy to see
that this yields indeed a complex. The crucial diagram in degrees 0 and 1 is

33
treated as in the in the rst case. At least after application of RMR (X ) it
commutes and has a splitting. RMR (X ) is a direct summand of RMR (X~  )
because the morphism between them is surjective and nite. The splitting
is constructed by projecting to the invariants under the covering group. The
components of X~  are not normal so this is not an application of our axioms
for the existence of a realization functor. But still the same proof works as
in the normal case.
Remark: It is easy to see that the category of motivic objects in MR
remains unchanged when we restrict to Z-linear combination of morphisms
rather than Q -linear ones. Replacing a singular variety by a smooth proper
hypercovering, it is easy to see that we can assume X to be a complex of
smooth varieties in the denition of motivic objects. However, it is not clear
at all whether it might suce to assume that X is a smooth variety rather
than a complex.

3 The motive of BGL

The main result of this chapter is corollary 3.2.5 where we completely deter-
mine the motive of the classifying space of GL. We work over the base Q .
By base change the results follows over any base eld of characteristic zero.
As before let DM? be Voevodsky's category of motivic complexes with ra-
tional coecients. In the next chapter the result will be used in order to
give a very easy construction of Chern classes in motivic cohomology and
check their relation with Chern classes in mixed realization cohomology.
3.1 Set-up
Let G be a connected algebraic group (we only need G = GLn and G = G nm .)
We dene the simplicial variety

EG = G GG GGG

where the face morphisms are induced by the various projections and the
degeneracy maps by the section e. Note that EG is contractible. EG is a
homogeneous space under G with the diagonal action. Put
BG = EG=G :
34
We identify Bi G = Gi . This corresponds to the classical construction of
the classifying space as quotient of the universal cover. EG and BG are
obviously functorial.
We need to understand the motive of G m . Let e : Z(0) ! G m the unit
section. The multiplication map is denoted .
Lemma 3.1.1. The section e induces a decomposition of G m into
M (G m ) 
= Z(0)  M (G m )
in DMgm (Q ; Z). M (G m ) is isomorphic to Z(1)[1] via residue at 0. For
the multiplication map we have
8
<id
> w = 0;
Grw () = >
w = 1;
:
0 w 6= 0; 1:
Proof. Recall that by denition
M (A 1 ) = Z(0) :
Now consider the localization triangle for the smooth pair (G m ; A 1 ):
M (G m ) ???! M (A 1 ) ???! M (0)(1)[2] ;
i.e.
Z(0)  M (G m ) ???! Z(0) ???! :
Z(1)[2]

The last map is an element of H ?2 (Z(1)(Q )) = 0. Hence M (G m ) is iso-

morphic to Z(1)[1]. The decomposition of the multiplication map uses the
commutative diagram for the properties of a left and right unit.
In particular G m is a mixed Tate motive. The same is true for all GLn .
This can be seen by using the stratication given by the Bruhat decompo-
sition. All strata are of the form split torus times some ane space. From
now on we work in the category of mixed Tate motives DMT introduced
in 1.2.1.

35
3.2 Motives of some classifying spaces
Proposition 3.2.1. There is a unique morphism
bi : Q (i)[2i] ! M (B G m )
induced by
Q( i)[2i] 
= (M (G m ) )
i [i] ! M (Bi G m )[i]
Moreover,
M M
= bi : Q( i)[2i] ! M (B G m )
is an isomorphism.
Proof. Consider the exact triangle
M (?i+1 B G m ) ! M (B G m ) ! M (! Bi+1 G m ! Bi G m ! 0)
Clearly we have a map bi to the space on the right. M (?i+1 B G m ) is an
object of DMT [0;i?1] . Hence the composition of bi with the connecting
morphism vanishes by lemma 1.2.2. Then bi lifts to a map to M (B G m ).
Using the same argument again, we see that the lift is unique. Now we are
precisely in the situation of proposition 1.3.6. It is enough to pass to the
weight graded pieces of the subcomplexes M (?N B G m ). The decomposi-
tion of Grw () is known. It determines all dierentials. To compute its
cohomology is a completely combinatorial question. Instead of considering
the combinatorics, we can also quote the result of the computation in the
Hodge realization, e.g. [Hu1] 17.4.1 for n = 1. Either way we see that
Grw ( ) is injective and that the cokernel is a subobject of M (BN G m )[N ].
By 1.3.6  is an isomorphism.
Denition 3.2.2. Let m1; : : : ; mk be simple Tate motives of the form Q (i)[2i].
By the polynomial ring in m1 ; : : : ; mk we mean the motive
M
Q[ m1 ; : : : ; mk ] := m1
e
: : :
mk
ek :
1

e1 ;:::;ek 0
It is not correct to view Q [m1 ; : : : ; mk ] as a ring. There is no multipli-
cation but rather a comultiplication induced by the diagonal. If we apply
the singular cohomology functor to it, we get a true polynomial ring in k
generators.
Remark: M (B G m ) is the polynomial ring in the generator b = Im bi . The
notation is consistent: the image of the map bi is the subobject bi .
36
Corollary 3.2.3. Let T = G nm be a split torus. Then M (BT ) is isomorphic
to the polynomial ring in b1 ; : : : ; bn where bi corresponds to the generator of
the motive of the i-th factor G m in T
Proof. We have already on the simplicial level BT  = (B G m )n . Hence
M (BT ) is isomorphic to the n-fold tensor power of M (B G m ).
Let ci : Q (i)[2i] ! B (T ) be the i-th symmetric polynomial in the gener-
ating maps bk of M (BT ).
Theorem 3.2.4. The object M (B GLn) in DMeff gm? (Q ; Q ) is given by the
commutative polynomial ring generated by ci = Q (i)[2i] for i  n.
Proof. We have dened a map
: Q [c1 ; : : : ; cn ] ! BT ! B GLn :
We claim that it is an isomorphism in DM? . By proposition 1.3.6 it is
enough to consider the weight graded pieces of the nite subcomplexes.
Moreover, the singular realization is faithful on Tate motives of xed weight.
We know that the singular realization of is an isomorphism (e.g. [Du] The-
orem 6.13 and Proposition 8.3). On nite subcomplexes Rsing (?N B GLN ),
the map Rsing ( ) is not an isomorphism but the defect is direct sum of Tate
motives of the form Q (i)[j ] with j  N (because the spectral sequence is
concentrated in the rst quadrant). Hence the assumptions of 1.3.6 hold
and is an isomorphism.
Remark: We only need existence of  in 3.2.1 for this proof and reproof
that it is an isomorphism.
Corollary 3.2.5.
M (B GL) = Q [c1 ; c2 ; : : : ] :
Corollary 3.2.6. Application of RMR yields
RMR (B GL) = Q [c1 ; c2 ; : : : ]
with ci = Q (?i)[?2i]. The splitting is the same as the one constructed in
[Hu1] 17.4.1.

37
Proof. Recall that RMR (Q (i)[2i]) = Q (?i)[?2i]. Hence the equality follows
from the previous corollary. The construction of the splitting of B GL is very
much the same as the construction used in [Hu1] 17.3-17.4. It is enough to
show that the splitting of RMR (B G m ) constructed in [Hu1] 17.3.2 is the
same as ours. Note that we only have to check that the splittings agree in
the l-adic realization because the splitting of the Hodge realization is unique
anyway. The one in loc. cit. is induced by the Chern class of the standard
line bundle on Pn , ours by the cycle class of a point. That they agree is
classic.

4 Chern classes
The aim of this chapter is to show that the higher Chern classes from higher
algebraic K -theory to absolute cohomology of mixed realizations (see [Hu1])
factor over Voevodsky's motivic cohomology.
4.1 K -theory and group cohomology of GL(X )
We start with a review of the results in [Hu1] 18.1-18.2 in a more conceptual
terminology. In this section all schemes are noetherian and regular, e.g.
smooth varieties over k. We denote K(X ) a simplicial set whose homotopy
groups are the K -groups of X .
Denition 4.1.1 ([Hu1] 18.1.1). Let U be a simplicial ane scheme.
Assume that U has nite combinatorial dimension, i.e., is degenerate above
some simplicial degree. Then we dene
K(U ) = holim K(Ui) :
If U is the nerve of an open cover of X , then
K(X ) ! K(U )
is a weak equivalence by the Mayer-Vietoris property of K -theory of regular
schemes. In the ane case K(U ) can be realized as K0 (X )  Z1(B GL(U )).
More generally:
Proposition 4.1.2 (Thomason, [Hu1] 18.1.5).
K(X ) = lim
?! Tot Z  Z1(B GL(U ))

38
where the direct limit runs through all open covers of X . In particular
K0 (X ) = Z  lim
?! 0 Tot Z1(B GL(U ))
?! i Tot Z1(B GL(U )) for i  1.
Ki(X ) = lim
Proof. The weak equivalence follows from the formula in loc. cit. because
direct limits commute with homotopy groups. The explicit calculation fol-
lows from it by the spectral sequence for the total space of a simplicial space.
It converges because all U have nite combinatorial dimension.
The simplicial set lim
?! Tot Z1(B GL(U )) inherits an H -group structure
from the H -group structure on Z1(B GL(Ui )).
Denition 4.1.3. Let X be a regular noetherian scheme. We put
p (GL(X ); Q ) := H p (j lim Tot Z (B GL(U ))j; Q )
HMV ?! 1 
where the right hand side means singular cohomology of the geometric real-
ization. It is called Mayer Vietoris localized group cohomology of GL(X )).
A simpler construction of the same cohomology group will be given be-
low. Note also that in the case of X = Spec A this is not group cohomology
of GL(A) but rather a version such that a long exact Mayer-Vietoris se-
quence for open covers is forced. From the denition, however, the relation
to K -theory is clear:
Proposition 4.1.4. There is a natural map
Kp(X )Q ! HMV p (GL(X ); Q ) :
L 
Its image is the subgroup of primitive elements in HMV (GL(X ); Q ).
Proof. The map is nothing but the Hurewicz map from homotopy groups of
a space to its cohomology. In the case of an H -space the image in rational
cohomology is given by the primitive part, cf. [Lo] A.11.
Now we turn to the promised simpler description of our group cohomol-
ogy. For a set B , we denote by Q [B ] the Q vector space with basis B . Let
U = Spec A be an ane scheme. As in the proof of [Hu1] 18.2.4, the maps
Z1B GL(U ) ! ZZ1B GL(U ) ZB GL(U )

induce isomorphisms on singular cohomology. Hence

H p(jZ1(B GL(U ))j; Q ) = H p (jB GL(U ))j; Q ) = H p (GL(U ); Q )
39
is group cohomology in the usual sense. It is computed by the standard Bar
complex (and this is in fact how the last equality is proved):
H p (GL(U ); Q ) = H ?p(Q [B? GL(U )]) :
(We stick to our convention: all complexes are cohomological ones. A sim-
plicial group is turned into a complex by putting it into negative degrees.)
Proposition 4.1.5. Let X as in denition 4.1.3. There is a natural iso-
morphism
HMVp (GL(X ); Q ) = lim H ?p(Tot Q [B GL(U )])
?! ? 
where the direct system runs through all open covers of X .
Proof. Clear from the above. Note that cohomology commutes with direct
limits.
Rather then constructing Chern classes on higher K -groups, we will con-
struct them on group cohomology of GL(X ).
Remark: If X itself is ane, then the natural map
Z1B GL(X ) ! Tot Z1B GL(U )

induces an isomorphism on all higher homotopy groups but not on 0 . The

space on the left is connected, the one on the right is not in general. They are
certainly not weakly equivalent. Hence there is no reason for it to induce an
isomorphism on singular cohomology. This justies the above remark that
Mayer-Vietoris group cohomology is not the same thing as group cohomol-
ogy.
4.2 Chern classes into motivic cohomology
The construction of Chern classes into motivic cohomology proceeds along
the same lines as for absolute realization cohomology in [Hu1] 18.2.5. The
key observation is that
Bn GL(U ) = HomSm(U; Bn GL):
Theorem 4.2.1. Let X be a smooth variety over k, some xed ground eld
of characteristic zero. There is a natural transformation
p (GL(X ); Q ) ! Hom
HMV DM? (k;Q) (M (X )[p]; M (B GL)) :

40
Proof. We use the description of proposition 4.1.5. Let Q Sm be the cat-
egory of smooth varieties with morphisms formal Q -linear combinations of
morphisms of varieties. An element of
?p
?! H (Tot Q [B? GL(U )]
lim
is represented by a morphism of complexes in Q Sm
U [p] ! B GL :
Note that the functor M from proposition 1.1.6 has values in the category
of complexes of Nisnevich sheaves with transfers, not only in the derived
category. By functoriality, it induces a map
M (U )[p] ! M (B GL)
in DM? . The natural map M (U ) ! M (X ) is a quasi-isomorphism.
Hence we have constructed an element in HomDM? (M (X )[p]; M (B GL))
as claimed. Note that it is well-dened: two representatives dier by a
homotopy of morphisms of complexes.
Corollary 4.2.2. There is a natural map
p (GL(X ); Q )
Kp (X ) ! HMV
M 2j ?p
! HomDM? (M (X )[p]; M (B GL)) ! HM (X; Q (j )) :
For (j; p) 6= (0; 0) let
2j ?p (X; Q (j )) :
cj : Kp (X ) ! HM
c0 on K0 (X ) is given by the above composition plus the natural map
K0 (X ) ?deg
?! Z ! HM
0 (X; Q (0))

mapping 1 to the structural morphism. cj is called j -th motivic Chern class.

Proof. Recall that by corollary 3.2.5
M (B GL) = Q [c1 ; c2 ; : : : ]
with cj = Q (j )[2j ]. The map in the corollary is nothing but the composition
of the transformation in the theorem with the natural projection.

41
Note that our transformation of functors maps primitive elements to primi-
tive elements. Hence we do not loose anything by projecting to the primitive
part of M (B GL) in the corollary.
Corollary 4.2.3. The Chern class
2j ?p (X; Q (j ))
cj : Kp (X ) ! HMR
constructed in [Hu1] 18.2.6 factors through the motivic Chern class.
Proof. The construction of Chern classes in loc. cit. is precisely the one
above with R~ MR replacing the functor M . The compatibility is a direct
consequence of functoriality of our construction.
Remark: For simplicity we have restricted to the case of a smooth variety
in the above. Everything works directly for bounded complexes of smooth
varieties, e.g. smooth simplicial varieties with nite combinatorial dimen-
sion. As in [Hu1] 18.1.3 bounded above complex of smooth varieties can be
treated as the direct limit of its truncations. Singular varieties or complexes
of such can be replaced by a smooth proper hypercovering.
Erratum: We have to correct an inaccuracy in [Hu1] 18.1.5: The variety
has to be assumed smooth. The mistake is that Mayer-Vietoris holds for
K -theory of general varieties only if we allow negative K -groups. The group
?1 A(U: ) (notation of loc. cit.) might not be zero. As as consequence
the arguments in loc. cit. work directly only for smooth simplicial varieties.
However, they extend to the general case again by replacing singular varieties
by smooth proper hypercoverings.

42
References
[Be1] A.A. Beilinson: \Notes on absolute Hodge cohomology", Applica-
tions of algebraic K-theory to algebraic geometry and number the-
ory, Proc. AMS-IMS-SIAM Joint Summer Res. Conf., Boulder/Colo.
1983, Part I, Contemp. Math. 55, 35-68 (1986).
[BBD] A.A. Beilinson, J. Bernstein, P. Deligne: \Faisceaux pervers", in
B. Teissier, J.L. Verdier, \Analyse et Topologie sur les Espaces sin-
guliers" (I), Asterisque 100, Soc. Math. France 1982.
[Bo] A. Borel: \Sur l'homologie et cohomologie des groupes de Lie com-
pacts connexes", Amer. J. Math. 76, 273-342 (1954).
[De1] P. Deligne: \Theorie de Hodge III" Inst. haut. Etud. sci., Publ. Math.
44(1974), 5-77 (1975).
[De2] P. Deligne: \Hodge cycles on abelian varieties", (Notes by J. S.
Milne), in: \Hodge cycles, motives, and Shimura varieties", Lect.
Notes Math. 900, 9-100 (1982).
[Du] J.L. Dupont: \Curvature and characteristic classes", Lecture Notes
in Mathematics 640, Springer-Verlag, (1978).
[Ek] T. Ekedahl: \On the adic formalism", in: \The Grothendieck
Festschrift", Vol. II, Birkhaeuser (1990).
[Fr] Friedlander: \Motivic Complexes of Suslin and Voevodsky", Sem.
Bourbaki 1996-1997, Exp. no. 833.
[Fu] W. Fulton: \Intersection theory", Ergebnisse der Mathematik und
ihrer Grenzgebiete, 3. Folge, Bd. 2, Springer-Verlag, (1984).
[GraRem] H. Grauert, R. Remmert: \Coherent analytic sheaves",
Grundlehren der Mathematischen Wissenschaften, 265, Springer-
Verlag (1984).
[Gro] B. Gross: \The motive of a reductive group", Invent. Math. 130,
No.2, 287-313 (1997)
[Hu1] A. Huber: \Mixed Motives and their Relization in Derived Cate-
gories", Lecture Notes in Mathematics 1604, Springer Verlag (1995).
[Hu2] A. Huber: \Mixed perverse sheaves for schemes over number elds",
Compos. Math. 108, No.1, 107-121 (1997).
43
[Ja] U. Jannsen: \Mixed motives and algebraic K-theory", Lecture Notes
in Mathematics 1400, Springer-Verlag (1990).
[Ka] B. Kahn: \La conjecture de Milnor", Sem. Bourbaki, 49eme annee,
1996-97, no. 834.
[Le1] \Tate motives and the vanishing conjectures for algebraic K -theory",
in: Goerss, P. G. (ed.) et al., \Algebraic K -theory and algebraic
topology", Proceedings of the NATO Advanced Study Institute, Lake
Louise 1991, 167{188, Kluwer Academic Publishers 1993.
[Le2] M. Levine: \Motivic cohomology and algebraic cycles: a categorial
construction I", Preprint 1996.
[Le3] M. Levine:\Mixed Motives", Mathematical Surveys and Monographs
57, AMS (1998).
[Lo] J.-L. Loday: \Cyclic homology", Grundlehren der Mathematischen
Wissenschaften 301, Springer-Verlag (1992).
[Ma] H. Matsumura: \ Commutative Ring theory", Cambridge Studies in
Advanced Mathematics 8, Cambridge University Press (1989).
[Sch] A.J. Scholl: \Classical Motives", in: U. Jannsen (ed.) et al., \Mo-
tives", Proceedings of the summer research conference on motives,
held at the University of Washington, Seattle, WA, USA, July 20-
August 2, 1991. American Mathematical Society, Proc. Symp. Pure
Math. 55, Pt. 1, 163-187 (1994).
[Vo1] V. Voevodsky: \Triangulated categories of motives over a eld",
Preprint 1994.
[Vo2] V. Voevodsky: A Letter to Beilinson, 1992.

44