Andre - Theorie de L'ambiguite
Andre - Theorie de L'ambiguite
Andre - Theorie de L'ambiguite
Q=Q) := lim
K
Gal (K=Q)
is the compact (profinite) group of automorphisms of the field
Q of all algebraic
numbers.
This is the process which the philosopher A. Lautman called ``la montee vers
l'absolu'' [21, III]: a single mathematical object, Gal (
Q=Q).
Let now X be complex plane deprived of the points 0; 1, so that p
1
(X) is a free
group on the loops g
0
and g
1
around 0 and 1 respectively.
According to Grothendieck and Belyi, the absolute arithmetic Galois group
Gal (
Q=Q) acts (
3
) on the absolute geometric Galois group ^ p
1
(X), and this action
is faithful: Gal (
Q=Q), and it is still unknown whether the two coincide (cf: [22]).
3. - Galois ambiguities in ``transcendental analysis''. Differential Galois
theory.
Let us now turn to some recent applications of Galois's ideas in ``transcendental
analysis'', in the spirit of Galois's lines quoted at the beginning of this paper.
3.1. - The beginnings.
Such ambiguities already puzzled Gauss in his work on the arithmetico-
geometric mean and hypergeometric functions. The Gauss hypergeometric
function
F(a; b; c; x) = 1
a:b
1:c
x
a(a 1):b(b 1)
1:2:c(c 1)
x
2
(
3
) the action depends on the choice of a rational base point +, for instance 1=2, or
better, a tangential base point; but the outer action is canonical.
(
4
) consisting in a unit in the profinite free group on one generator together with an
element of the derived group of the profinite free group on two generators.
(
5
) which comes from the consideration of the configuration of five points on the
projective line.
AMBIGUITY THEORY, OLD AND NEW 5
is a solution of the hypergeometric differential equation
x(1 x)F
//
(c (a b 1)x)F
/
abF = 0:
The arithmetico-geometric mean is related to F(
1
2
;
1
2
; 1; x). On the other hand,
F(
1
4
;
1
4
; 1; 4x(1 x)) converges inside the lemniscate
[4x(1 x)[ < 1
and satisfies there the same hypergeometric differential equation with para-
meters a = b = 1=2; c = 1. Actually
F
1
4
;
1
4
; 1; 4x(1 x)
= F
1
2
;
1
2
; 1; x
in the left region of the lemniscate, whereas
F
1
4
;
1
4
; 1; 4x(1 x)
= F
1
2
;
1
2
; 1; 1 x
in the right region.
This kind of problems was to be elucidated by Riemann's study [27] of
monodromy (that is, in essence, the study of the action of p
1
(C 0; 1; +) on the
solutions of the hypergeometric equation).
In the same vein, one can also mention the following classical works (cf. [15]
for a detailed account of this story):
1) Schwarz' classification of hypergeometric differential equations with al-
gebraic solutions: by looking at the the Galois groups of the corresponding Galois
extensions of C(x), Schwarz reduces it to the classification of finite subgroups of
GL
2
[29],
2) Jordan's celebrated theorem on finite subgroups of GL
m
, which arose, via
Galois's philosophy, from the study of algebraic solutions of differential equa-
tions of order m [17],
3) much of Klein's work, which he described himself as a try to ``blend Galois
with Riemann''.
3.2. - Differential Galois theory.
This theory, initiated by Picard, Vessiot, Drach and later Kolchin, realizes
Galois's program(
6
):
(
6
) a notable precursor of differential Galois theory was Liouville, who not only
exhumed and advertised Galois' work but also made the first steps towards Galois'
dreams by studying the conditions under which a linear differential equation can be solved
by quadratures.
6 YVES ANDRE
1
e )
:
^
f
i
(x)
for suitable polynomials P
i
's and formal power series f
i
's. The exponential scal-
ings act on the factors e
P
i
(x
1
e)
, formal monodromies on the factors x
l
i
log
k
i
x.
Over K = C(x) (analytic local case), one encounters a
3.3.3. 3rd type of Galois ambiguity: Stokes phenomenon. (
8
) In general, di-
vergent series
^
f
i
occur in the solutions. They may be resummed, using a Borel-
Laplace type process, in suitable sectors bissected by an arbitrary fixed half-line
` (except for finitely many `'s, the singular directions).
EXAMPLE. - L : x
2
@y y = x: The formal solution ^ y =
P
0
( 1)
n
n!x
n1
can
resummed (in a sector bissected by ` = R
) to s
0
e
t=x
1t
dt:
For a singular line (` = R
4.2. - Motives.
Let X be a smooth algebraic variety over Q, and let Y be a closed (possibly
reducible) subvariety. Periods arise as entries of a matrix of the comparison
isomorphism, given by integration of algebraic diffential forms over chains, be-
tween algebraic De Rham and ordinary Betti relative cohomology:
H
DR
(X; Y) C
$
X;Y
H
B
(X; Y) C: (4)
This is where motives enter the stage. Motives are intermediate between alge-
braic varieties and their linear invariants (cohomology): they are of algebro-
geometric nature on one hand, but they are supposed to play the role of a uni-
versal cohomology for algebraic varieties and thus to enjoy the same formalism
on the other hand. One expects the existence of an abelian category MM of mixed
motives (over Q, with rational coefficients), and of a functor
h : Var (Q) MM
(from the category of algebraic varieties over Q) which plays the role of universal
cohomology. The morphisms in MM should be related to algebraic corre-
spondences. In addition, the cartesian product on Var (Q) corresponds via h to a
certain tensor product on MM, which makes MM into a tannakian category,
i.e. it has the same formal properties as the category of representations of a
group.
The cohomologies H
DR
and H
B
factor through h, giving rise to two -functors
H
DR
; H
B
: MM Vec
Q
with values in the category of finite-dimensional Q-vector spaces. Moreover,
corresponding to (4), there is a isomorphism in Vec
C
$
M
: H
DR
(M) C H
B
(M) C (5)
which is -functorial in the motive M. The entries of a matrix of $
M
with respect
to some basis of the Q-vector space H
DR
(M) (resp. H
B
(M)) are the periods of M.
4.3. - Motivic Galois groups and period torsors.
Let M) be the tannakian subcategory of MM generated by a motive M: its
objets are given by algebraic constructions on M (sums, subquotients, duals,
tensor products).
One defines the motivic Galois group of M to be the group scheme
G
mot
(M) := Aut
H
B[M)
(6)
of automorphisms of the restriction of the -functor H
B
to M).
AMBIGUITY THEORY, OLD AND NEW 11
This is a linear algebraic group over Q: in heuristic terms, G
mot
(M) is just the
Zariski-closed subgroup of GL(H
B
(M)) consisting of matrices which preserve
motivic relations in the algebraic constructions on H
B
(M).
Similarly, one can consider both H
DR
and H
B
, and define the period torsor of
M to be
P
mot
(M) := Isom
(H
DR[M)
; H
B[M)
) Var (Q) (7)
of isomorphisms of the restrictions of the -functors H
DR
and H
B
to M). This is
a torsor under G
mot
(M), and it has a canonical complex point:
$
M
P
mot
(M)(C): (8)
4.4. - Grothendieck's period conjecture.
It asserts that $
M
is a generic point:
the smallest algebraic subvariety of P
mot
(M) defined over Q and containing
$ is P
mot
(M) itself.
In more heuristic terms, this means that any polynomial relations with ra-
tional coefficients between periods should be of motivic origin (the relations of
motivic origin being precisely those which define P
mot
(M)).
The conjecture is also equivalent to: P
mot
(M) is connected (over Q) and
tr: deg
Q
Q[periods(M)] = dim G
mot
(M): (9)
EXAMPLE. - For the motive of P
1
, the motivic Galois group is the mutiplicative
group G
m
, a period is 2pi, and Grothendieck's conjecture amounts to the
transcendence of p.
REMARKS. - 1) There is an analogy between equations (3) and (9), which can
be more made more precise if one considers a family of smooth algebraic vari-
eties (or motives): the variation of periods is then controlled by a linear differ-
ential equation (the so-called Picard-Fuchs equation. For the relation between
its differential Galois group and the motivic Galois groups of the fibers, we refer
to [2, 6].
2) By definition, periods are convergent integrals of a certain type. They
can be transformed by algebraic changes of variable, or using additivity of the
integral, or using Stokes formula.
Kontsevich has conjectured that any polynomial relation with rational
coefficients between periods can be obtained by way of these elementary op-
erations from calculus (cf: [19]). Using ideas of Nori and the expected equiva-
lence of various motivic settings, it can be shown that this conjecture is actually
equivalent to Grothendieck's conjecture (cf: [1, ch. 23]).
12 YVES ANDRE
.
2) Let x be a non-zero rational number of absolute value than 1. Then
a = F(
1
2
;
1
2
; 1; x) is, up to multiplication by 2pi, a period of the elliptic curve Y
2
=
X(X 1)(X x). Except for finitely many exceptions (
11
) x Q, G
a
= GL
2
(Q), and
the conjugates of a are mF(
1
2
;
1
2
; 1; x) nF(
1
2
;
1
2
; 1; 1 x); m; n Q, not both 0.
3) Let s be an odd integer > 1. Then z(s) :=
P
n
s
is a period of a so-called
mixed Tate motive over Z. Grothendieck's period conjecture for this type of
motives would imply that p and z(3); z(5); . . . are algebrically independent, that
the conjugates of z(s) are z(s) r(pi)
s
; r Q; and that G
z(s)
is a semi-direct
product of Q
by Q.
4) More generally, multiple zeta values
z(s
1
; . . . ; s
k
) =
X
n
1
>>n
k
>0
1
n
s
1
. . . n
s
k
k
= q
1>t
1
>>t
s
>0
v
s
1
. . v
s
k
;
(where s
i
> 0; s
1
> 1; s = s
1
s
k
; v
0
=
dt
t
; v
1
=
dt
1t
; v
r
= v
r1
0
. v
1
for
r > 1) are also periods for mixed Tate motives over Z. Many polynomial relations
are known among these numbers (double shuffle and regularization relations,
Drinfeld's associator relations); so far, all these relations have been proven to be
of motivic origin, as predicted by Grothendieck's period conjecture.
(
11
) the so-called singular moduli, corresponding to elliptic curves with complex
multiplication.
AMBIGUITY THEORY, OLD AND NEW 13
The ``absolute'' motivic Galois group corresponding to multiple zeta values
(taken altogether) is expected to be
U
+
(3; 5; 7 . . . ) = U(3; 5; 7 . . . ) G
m
(10)
where U(3; 5; 7 . . . ) is prounipotent group whose Lie algebra, graded by the G
m
-
action, is the free graded Lie algebra with one generator in each odd degree
3; 5; 7 . . ..
The fact that multiple zeta values satisfy Drinfeld's associator relations
provides a homomorphism
U
+
(3; 5; 7 . . . ) GT (11)
where GT denotes the ``proalgebraic version'' of the (profinite) Grothendieck-
Teichmuller group mentioned in 2.3. It is still unknown whether this is an
isomorphism (cf: [1, ch. 25]).
5. - Galois ambiguities in quantum field theory.
In the last years, Galois ideas have barged into quantum field theory. We end
this survey by just giving a glimpse of the main steps of these developments (see
[7], [8]).
5.1. D. Kreimer, A. Connes [10]: conceptual understanding of (reg-dim) per-
turbative renormalization of a QFT in dimension D in terms of Birkhoff de-
composition of loops g(z) in a certain pro-unipotent group U attached to the QFT
via the construction of a Hopf algebra on suitable Feynman diagrams; here z is
the complexified dimension minus D.
The postulated independence of g
YVES, Une introduction aux motifs (motifs purs, motifs mixtes, pe riodes).
Panoramas et Syntheses 17. Societe Mathe matique de France, Paris, 2004.
[2] ANDRE
LE