Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
This paper concerns a class of complex numbers, Stokes formula insures that the right hand side de-
called periods, that appear naturally when compar- pends only on the classes Hn (M; Q) and
ing two cohomology theories for algebraic varieties n
HdR (X; C) of and .
(the first defined topologically and the second alge- The classical theorem of de Rham asserts that the
n
braically). Our goal is to explain the fundamental pairing (1) is perfect, i.e., that it identifies HdR (M; C)
conjectures of Grothendieck and KontsevichZagier with the space of linear maps from Hn (M; Q) to C.
that give very precise information about the tran-
scendence properties of periods. The notion of mo- 1.2 Periods
tive (due to Grothendieck) plays an important con-
Periods are complex numbers obtained by evaluat-
ceptual role. Finally, we explain a geometric ver-
ing the pairing (1) on special de Rham cohomology
sion of these conjectures. In contrast with the origi-
classes with interesting arithmetic properties.
nal conjectures whose solution seems to lie in a very
Assume that M is an algebraic manifold defined
distant future, if at all it exists, a solution for the
by equations with rational coefficients. In the lan-
geometric conjectures is within reach of the actual
guage of algebraic geometry, we are fixing a smooth
motivic technology.
Q-variety X such that M is the manifold of complex
points in X, which we write simply as M = X(C). In
this case, it makes sense to speak of algebraic differ-
1 Introduction ential forms over M (or X). In general, there are few
of these globally, and one is forced to work locally
1.1 Integration and cohomology for the Zariski topology on X, i.e., to consider the
Let M be a real C -manifold. Let An (M, C) be the sheaf nX of algebraic differential forms of degree n
C-vector space of C -differential forms of degree n on X. Locally for the Zariski topology, a section of
on M. These vector spaces are the components of nX is an element of An (M; C) that can be written
the de Rham complex A (M; C) whose cohomol- as a linear combination of f0 d f1 . . . d fn where
ogy (i.e., the quotient of the space of closed dif- the fi s are regular functions over X. (For exam-
ferential forms by its subspace of exact differen- ple, if M = Cn , the fi s can be written as fractions
tial forms) is the de Rham cohomology of M de- Pi /Qi with Pi s and Qi s polynomials in n variables
with rational coefficients.) Varying the degree, one
noted by HdR (M; C). In practice, for instance if
M is compact, the HdR n
(M; C)s are finite dimen- gets the algebraic de Rham complex X ; this is a
sional vector spaces; in any case, they vanish unless complex of sheaves for the Zariski topology on X.
0 n dim(M). Following Grothendieck, we define the algebraic de
On the other hand, one has the singular chain Rham cohomology of X (or M) to be the Zariski
complex of M, denoted by C (M; Q). For n N, hyper-cohomology of X :
Cn (M; Q) is the Q-vector space with basis consist-
HAdR (X) = HZar (X; X ).
ing of C -maps from the n-th simplex
The elements of HAdR (X) are those special de
n+1 x0 0, . . . , xn 0,
( )
Rham cohomology classes that produce periods when
= (x0 , . . . , xn ) R
n
x0 + + xn = 1 paired with singular homology classes. More pre-
cisely, there is a canonical isomorphism
to M. The singular homology of M, denoted by
HAdR (X) C ' HdR (M; C)
H (M; Q), is the homology of this complex.
Integration of forms yields a well-defined pairing and, in particular, HAdR (X) is a sub-Q-vector space
of HdR (M; C). (Roughly speaking, this inclusion
n
h, i : Hn (M; Q) HdR (M; C) C. (1) is obtained by considering an algebraic differential
form on X as an ordinary differential form on M.)
If = rs=1 a s [ f s : n M] Cn (M; Q) is a
P
Now, restricting the pairing (1), we get a pairing
closed chain and An (M, C) is a closed differen- Sing
tial form, then H (X) HAdR (X) C (3)
Sing
r
X Z where H (X) denotes H (M; Q). (As the notation
h, i = as f s . (2) suggests, the above pairing is canonically associ-
s=1 n ated to the algebraic variety X.)
Definition 1. A period of the algebraic variety X equal to the dimension of the motivic Galois group
is a complex number which is in the image of the of X. Of course, the motivic Galois group of X is
pairing (3). quite a complicated object and its dimension can be
Sing
very hard to compute. Nonetheless, it is easy to
Remark 2. Although H (X) and HAdR (X) are fi- convey that the dimension of a not-so-explicit alge-
nite dimensional Q-vector spaces, the values of the braic group is much easier to compute than the tran-
paring (3) are not rational numbers in general. In- scendence degree of an explicit subfield of C. This
deed, periods are often (expected to be) transcen- is already the case for X = P1 (the projective line):
dental numbers. the motivic Galois group is given by Gm whereas
Remark 3. When X is smooth and proper, its peri- the subfield generated by periods is Q().
ods are called pure. The previous construction, in The conjecture of KontsevichZagier is more am-
the case where X is proper, yields all the pure peri- bitious and goes beyond the above question: it aims
ods. at describing all the algebraic relations among peri-
If X is not necessarily proper, its periods are called ods. Roughly speaking, it says that two periods are
mixed. In contrast with the pure case, the previous equal if and only if there is a geometric reason (see
construction is not expected to give all the mixed pe- Definition 6 for the list of geometric reasons.)
riods. Indeed, more mixed periods are obtained by Remark 4. The conjecture of KontsevichZagier is
considering relative cohomology of pairs (see 2.1). stronger than the conjecture of Grothendieck. This
is not obvious from their statements and will be ex-
1.3 Transcendance plained in 4. However, one can argue that both
It is an important and fascinating problem to under- conjectures are essentially, or morally, equivalent.
stand the arithmetic properties of periods. In the Remark 5. The conjecture of KontsevichZagier is
case of abelian periods (i.e., those obtained by tak- remarkable in its simplicity as it can be stated in
ing n = 1 in (3) with X possibly singular or, equiva- elementary terms. However, in practice, the conjec-
lently, those arising from abelian varieties and more ture of Grothendieck is better suited for deducing
generally 1-motives) much is known thanks to the algebraic independence of periods.
analytic subgroup theorem of Wstholz [16] gen-
eralizing results of Baker. The result of Wstholz 1.4 The geometric version
can be interpreted as follows: every Q-linear rela-
tion between abelian periods is of motivic origin. As it is the case with many deep and difficult prob-
This statement appears explicitly, for example, in lems on numbers, the conjectures of Grothendieck
[17]. We also refer to [8] for a similar interpretation and KontsevichZagier admit geometric (aka., func-
of the result of Wstholz in the context of another tional) analogues that are accessible. Some ideas
(albeit related) conjecture of Grothendieck (in the of the proof of the geometric versions will be dis-
style of the Hodge and Tate conjectures). cussed in 5.
Beside the case of abelian periods, very little is
known (but see [15] for a comprehensive catalogue 2 The KontsevichZagier conjecture
of what transcendence theory knows about periods
and [9] for some spectacular recent advances con- 2.1 The ring of abstract periods
cerning periods of Tate motives, aka., multiple zeta The KontsevichZagier conjecture is best stated by
values). Nonetheless, the conjectural picture is very introducing the ring of abstract periods. Roughly
satisfactory and conjectures of Grothendieck and speaking, the ring of abstract periods is the free
KontsevichZagier yield a precise understanding of Q-vector space generated by formal symbols, one
the ring of periods. Unfortunately, these conjec- for each pairing of a homology class with an alge-
tures seem so desperately out of reach of the present braic de Rham cohomology class, modulo the rela-
mathematics and this is to be taken as my personal tions that come from geometry. If the Kontsevich
opinion I cant think of any other conjecture that Zagier conjecture was true, the ring of abstract peri-
looks as intractable ! ods would be identical to the subring of C generated
A typical question of interest is the following: by periods.
Question. Let X be a smooth Q-variety and let More precisely, consider 5-tuples (X, Z, n, , )
Per(X) be the subfield of C generated by the image where X is a Q-variety (possibly singular), Z X
Sing
of the pairing (3). What is the transcendence degree a closed subvariety, n N, Hn (X, Z) a rel-
of the finitely generated extension Per(X)/Q? ative homology class of the pair (X(C), Z(C)) and
Grothendiecks conjecture gives an answer to this HAdR
n
(X, Z) a relative algebraic de Rham coho-
question: the transcendence degree of Per(X)/Q is mology class. To such a 5-tuple, one associates a
period 2.2 A compact presentation of the ring of ab-
Z stract periods
Ev(X, Z, n, , ) := C. (4) In this paragraph, we will give another presentation
of the ring PKZ that was obtained rather acciden-
Following KontsevichZagier [12], we make the fol- tally by the author.
lowing: In retrospect, this presentation uses less gener-
ators than the presentation of KontsevichZagier.
Definition 6. The ring of abstract (effective) pe- With the notation of Definition 6, one restricts to:
riods, denoted by Peff KZ , is the free Q-vector space X = Spec(A) for A running among tale sub-
generated by symbols [X, Z, n, , ] modulo the fol- Q[z1 , . . . , zn ]-algebras of the ring of convergent
lowing relations: power series with radius strictly larger than 1.
(a) (Additivity) the map (, ) 7 [X, Z, n, , ] is Z X is the normal crossing divisor given by
Sing n
bilinear on Hn (X, Z) HAdR (X, Z). the equation ni=1 zi (zi 1) = 0.
Q
(b) (Base-change) Given a morphism of Q-varieties : [0, 1] X(C) the canonical lift of the
n
f : X 0 X such that f (Z 0 ) Z, a relative sin- obvious incluson [0, 1]n , Cn .
Sing
gular homology class 0 Hn (X 0 , Z 0 ) and a = f dz1 . . . dzn with f A, a top degree
relative algebraic de Rham cohomology class differential form.
HAdRn
(X, Z), we have the relation In return, one needs much less relations: a special
case of the Stokes formula suffices to realize all the
[X, Z, n, f 0 , ] = [X 0 , Z 0 , n, 0 , f ]. geometric relations described in Definition 6.
To write down precisely the above sketch, we in-
(c) (Stokes formula) Given a Q-variety X, closed
troduce some notations. For an integer n N, we
subvarieties Z Y of X, a relative singular
Sing denote by Dn the closed unit polydisc in Cn . Let
homology class Hn (X, Y) and a rela-
O(Dn ) be the ring of convergent power series in the
tive algebraic de Rham cohomology class
n1 system of n variables (z1 , . . . , zn ) with radius of con-
HAdR (Y, Z), we have the relation
vergence strictly larger than 1.
[X, Y, n, , d] = [Y, Z, n 1, , ]. Definition 9. We denote by Oalg (Dn ) the sub-Q-
We denote by vector space of O(Dn ) consisting of those power se-
ries f = f (z1 , . . . , zn ) which are algebraic over the
field Q(z1 , . . . , zn ) of rational functions. We also set
" #
dt
2i := Gm , , 1, t [0, 1] 7 exp(2i t), Oalg (D ) = nN Oalg (Dn ).
S
t
and set PKZ := Peff 1 Definition 10. Let Peff be the quotient of Oalg (D )
KZ [2i ]. This is the ring of ab-
stract periods. by the sub-Q-vector space spanned by elements of
the form
It is clear that the function Ev of (4) induces a f
f |zi =1 + f |zi =0
morphism of Q-algebras zi
for f Oalg (D ) and i N r {0}. We also define P
Ev : PKZ C. (5)
to be Peff [2i1 ] with 2i the class of a well-chosen
(Note that Ev(2i) = 2i.) We can now state: element of Oalg (D1 ) whose integral on [0, 1] is 2i.
R` : MM(k) Rep(Gal(k/k); Q` )
2.3 Over more general base-fields
Fix a base field k and a complex embedding : to the category of `-adic Galois representations. (Af-
k , C. Then most of the previous discussion ex- ter tensoring the source category by Q` and when k
tends to varieties over k. is finitely generated over its prime field, this functor
Indeed, given a pair (X, Z) consisting of a k-variety is also believed to be fully faithful as a consequence
X and closed subvariety Z X, one can still define of the Tate conjecture.)
i
Sing
H (X, Z) and HAdR
(X, Z) and the canonical pair- Given a k-variety X, there are objets HM (X) of
Sing MM(k), called the motives of X, that play the role of
ing H (X, Z) HAdR (X, Z) C.
the universal cohomological invariants attached to
Sing
Remark 14. In contrast with H (X, Z) which is X. Every classical cohomological invariant of X is
still a finite dimensional Q-vector space indepen- i
then obtained from one of the HM (X)s by applying
dently of k and , HAdR
(X, Z) is naturally a k-vector a suitable realization functor. For instance,
space. Also, note that the canonical pairing is now i
R (HM (X)) is the singular cohomology group
perfect in a slightly twisted manner: it induces an i
HSing (X), endowed with its mixed Hodge struc-
Sing
isomorphism HAdR (X, Z)k, C ' hom(H (X, Z), C). ture;
One can also define the ring of abstract periods R` (HM i
(X)) is the `-adic cohomology group
over k, that we denote by PKZ (k, ), together with i
H` (X) endowed with the natural action of the
an evaluation homomorphism Ev : PKZ (k, ) C. absolute Galois group Gal(k/k).
One can wonder to which extent the Kontsevich
Remark 16. When k has characteristic zero, M. Nori
Zagier conjecture is reasonable for general fields.
[13] has constructed a candidate for the category of
We discuss this in the following:
mixed motives. While his construction is not known
Remark 15. When k/Q is algebraic (e.g., k is a num- to satisfy all the expected properties (for instance,
ber field), it is easy to see that PKZ (k, ) = PKZ (Q). the ext-groups between Noris motives are poorly
This shows that the KontsevichZagier conjecture related to Quillen K-groups), its enough for the pur-
holds for k if and only if it holds for Q. pose of the article.
On the other hand, the KontsevichZagier con-
jecture extended to fields of higher transcendence
3.2 The absolute motivic Galois group of a field
degrees is not reasonable unless is general. In-
deed, if (k) contains a transcendental period of a The category MM(k) is expected to share the for-
Q-variety (e.g., (k)), then Ev : PKZ (k, ) C mal properties of MHS and Rep(Gal(k/k); Q` ). For
cannot be injective. (However, see Remark 24 for instance, MM(k) has an exact tensor product and
what is expected without any condition on .) every motive M has a strong dual M . Moreover,
given an embedding : k , C, singular cohomol- Remark 20. By construction G(M) is an algebraic
ogy yields an exact faithful monoidal functor linear group. Moreover, Gmot (k, ) is the inverse
limit of the G(M)s when M runs over larger and
FSing : MM(k) Mod(Q) larger motives.
i i
sending HM (X) to the Q-vector space HSing (X). (In
fact, FSing is just R composed with the forgetful 3.4 Statement of Grothendieck conjecture
functor from MHS to Mod(Q).) This makes MM(k) From now on, we assume that k has characteris-
into a neutralized Tannakian category with fiber func- tic zero. Algebraic de Rham cohomology yields a
tor FSing . functor
A multiplicative operation = ( M ) M on FSing FAdR : MM(k) Mod(k)
is a family of automorphisms M GL(FSing (M)), i i
one for each M MM(k), such that sending HM (X) to the k-vector space HAdR (X). Fix-
for every morphism of motives a : M N, ing an embedding : k , C, the pairing (3) can
one has N FSing (a) = FSing (a) M ; be extended to any motive M yielding a pairing
for motives M and N, one has MN = M FSing (M) FAdR (M) C (8)
N modulo the identification FSing (M N) '
FSing (M) FSing (N). (This is truly and extension of (3): for M = HM
i
(X),
Definition 17. The multiplicative operations of FSing FSing (M) and FAdR (M) are indeed canonically iso-
Sing i
are the Q-rational points of a pro-Q-algebraic group morphic to Hi (X) and HAdR (X).)
Aut (FSing ) called the motivic Galois group (of k) Conjecture 21 (Grothendieck). Assume that k = Q.
and denoted by Gmot (k, ). (Note that this depends Let M be a motive and let Per(M) be the subfield of
on the choice of the complex embedding .) C generated by the image of the pairing (8). Then,
By the Tannaka reconstruction theorem [14], the one has the equality:
functor FSing induces an equivalence of categories
degtr(Per(M)/Q) = dim(G(M)).
eSing : MM(k)
F Rep(Gmot (k, ))
Remark 22. If one is interested in the transcendence
between motives and algebraic representations of degree of the field Per(X) generated by the periods
the motivic Galois group. of a Q-variety X, one should take
Remark 18. One may think about Gmot (k, ) as a 2dim(X)
M
linearization of the absolute Galois group of k. For M= i
HM (X)
instance, there is a continuous morphism i=0
the largest Tannakian subcategory consisting of con- which, by Proposition 42, is equal to Grel (M).
stant motives (i.e., in the image of the pull-back
Remark 45. Theorem 44 is clearly a geometric ana-
MM(k) MM(K)). Then, there is an exact se-
logue of the Grothendieck conjecture. Although it
quence
is a direct corollary of Theorem 40, its precise state-
alg
1 ((K/k)an , ) Aut (FSing |hMi ) ment was obtained during an email exchange with
Daniel Bertrand (and hence, did not appear before
Aut (FSing |hMi0 ) {1} in the literature). It was also independently obtained
alg by Peter Jossen and was probably known to Madhav
where 1 ((K/k)an , ) is the pro-algebraic comple- Nori.
tion of 1 ((K/k)an , ).
Proof. There is a commutative diagram 5.3 Geometric version of KontsevichZagier
As for the Grothendieck conjecture, one can use
alg
1 ((K/k)an ) / Gmot (K, ) / / Gmot (k, ) Theorem 40 to obtain a geometric version of the
KontsevichZagier conjecture. Moreover, working
alg
1 ((K/k)an ) / Aut (FSing |hMi ) / / Aut (FSing |hMi0 ) in the realm of Voevodsky motives, one can give
a very concrete statement in the style of the refor-
alg mulation given in Remark 13. This was achieved
The image of 1 ((K/k)an , ) in Aut (FSing |hMi ) is in [6] and relies on previous work of the author
a normal subgroup N and an algebraic representa- (such as the theory of rigid analytic motives [5] and
tion of Aut (FSing |hMi )/N corresponds to a motive the construction of nearby motives [7, Chapitre 3]).
in hMi whose associated local system is trivial. By We will not discuss the technical details here and
Theorem 40 (and Remark 41), this motive belongs we content ourself with stating the main result of
to hMi0 . [6]. We start by introducing some notation (com-
Definition 43. Let M MM(K) be a motive. The pare with 2.2). Recall that Dn denotes the closed
kernel of the morphism unit polydisc in Cn .
Definition 46. Let Oalg (Dn ) be the sub-C-vector
Aut (FSing |hMi ) Aut (FSing |hMi0 )
space of O(Dn )[[$]][$1 ] consisting of those Lau-
is denoted by Grel (M). This is the relative motivic rent series
Galois group of M. By construction, Grel (M) is a X
closed subgroup of G(M) (cf., Lemma 27). F= fi (z1 , . . . , zn ) $i ,
i>
5.2 Relative motivic Galois groups and func- with coefficients in O(Dn ), which are algebraic over
tional transcendance the field C($, z1 , . . . , zn ). We also set Oalg (D ) =
S n
Keep the situation as above. Given a motive M in nN Oalg (D ).
MM(K), FAdR (M) is naturally a holonomic DK/k -
module with regular singularities. (If M = HM
i
(X), Definition 47. Let P be the quotient of Oalg (D )
the DK/k -module is associated to the GaussManin by the C-vector space spanned by
i
connexion on HAdR (X).) elements of the first kind:
F
Theorem 44. The PicardVessiot extension of K F|zi =1 + F|zi =0
associated to the differential module FAdR (M) has zi
transcendance degree equal to the dimension of the
for F Oalg (D ) and i N r {0};
algebraic group Grel (M).
and elements of the second kind:
Proof. Indeed, by differential Galois theory, the tran- Z !
scendence degree of the PicardVessiot extension g g F
[0,1]
associated to FAdR (M) is equal to the dimension
of its differential Galois group. By the Riemann for g, F Oalg (D ) such that g does not de-
Hilbert correspondence, the latter group has the same g
dimension as the monodromy group of the local pendent on the variable $ (i.e., = 0) and
$
g and F do not depend simultaneously on the [15] M. Waldschmidt. Transcendence of periods: the state of
g F the art. Pure Appl. Math. Q., 2(2):435463, 2006.
variable zi (i.e., = 0) for every i
zi zi [16] G. Wstholz. Algebraische Punkte auf analytischen Un-
N r {0}. tergruppen algebraischer Gruppen. Annals of Mathemat-
ics, 129(3):501517, 1989.
There is an evaluation homomorphism [17] G. Wstholz. Leibniz conjecture, periods and motives.
Colloquium De Giorgi 2009, 3:3342, 2012.
Ev : P C(($)) (9)
X Z !
fi $i .
i> [0,1]n
References
[1] Y. Andr. Une introduction aux motifs. Panoramas et
Synthse 17. SMF, 2004.
[2] Y. Andr. Galois theory, motives and transcendental num-
bers. In Renormalization and Galois theories. IRMA
Lect. in Math. Theo. Phys. 15, pages 165177, 2009.
[3] J. Ayoub. Lalgbre de Hopf et le groupe de Galois mo-
tiviques dun corps de caractristique nulle, I. Journal fr
die reine un angewandte Mathematik, to appear.
[4] J. Ayoub. Lalgbre de Hopf et le groupe de Galois mo-
tiviques dun corps de caractristique nulle, II. Journal
fr die reine un angewandte Mathematik, to appear.
[5] J. Ayoub. Motifs des varits analytiques rigides.
Preprint.
[6] J. Ayoub. Une version relative de la conjecture des pri-
odes de KontsevichZagier. Annals of Mathematics, to
appear.
[7] J. Ayoub. Les six oprations de Grothendieck et le formal-
isme des cycles vanescents dans le monde motivique, II.
Astrisque, 315, 2008.
[8] J.-B. Bost and F. Charles. Some remarks concerning the
Grothendieck period conjecture. Preprint.
[9] F. Brown. Mixed Tate motives over Z. Ann. of Math. (2),
175(2):949976, 2012.
[10] U. Choudhury and M. Gallauer Alves de Souza. In prepa-
ration.
[11] A. Huber and S. Mller-Stach. On the relation between
Nori motives and Kontsevich periods. Preprint.
[12] M. Kontsevich and D. Zagier. Periods. In Mathematics
Unlimited 2001 and Beyond. Springer, pages 771808,
2001.
[13] M. Nori. Lectures at TIFR. Unpublished 32 pages.
[14] N. R. Saavedra. Catgories Tannakiennes. Lecture Notes
in Math. 265. Springer, 1972.