You are on page 1of 413

Alexandru Buium Differential Algebraic

Groups of Finite Dimension


Author Alexandru Buium Institute of
Mathematics of the Romanian Academy P.O.
Box 1-764 RO-70700 Bucharest, Romania
Mathematics Subject Classification (1991):
12H05, 14L17, 14K10, 14G05, 20F28 ISBN
3-540-55181-6 Springer-Verlag Berlin
Heidelberg New York ISBN 0-387-55181-6
Springer-Verlag New York Berlin
Heidelberg This work is subject to
copyright. All rights are reserved, whether
the whole or part of the material is
concerned, specifically the rights of
translation, reprinting, re-use of illustrations,
recitation, broadcasting, reproduction on
microfilms or in any other way, and storage

in data banks. Duplication of this publication


or parts thereof is permitted only under the
provisions of the Geman Copyright Law of
September 9, 1965, in its current version,
and permission for use must always be
obtained from Springer-Verlag. Violations
are liable for prosecution under the German
Copyright Law. Springer-Verlag Berlin
Heidelberg 1992 Printed in Germany
Typesetting: Camera ready by author Printing
and binding: Druckhaus Beltz,
Hemsbach/Bergslr. 46/3140-543210 Printed on acid-free paper
INTRODUCTION Differential algebraic
groups are defined roughly speaking as
"groups of solutions of algebraic differential
equations" in the same way In which
algebraic groups are defined as "groups of

solutions of algebraic equations". They were


introduced in modern literature by Cassidy
[C 1] and Kolchin [K2]l their pre-history
goes back however to classical work of S.
Lie, E. Cartan and :J.F. Rltt. Let's
contemplate a few examples before giving
the formal definition (for which we send to
Section 3 of this Introduction). Start with any
linear differential equation: (1) y(n). Sly(nl). 0 ....anY= where the unknown y and the
coefficients a i are, say, meromorphic
functions of the complex variable t. The
difference of any two solutions of (1) is
again a solution of (l)l so the solutions of (1)
form a group with respect to addition. This
provides a first example of *'differential
algebraic group". Slmllarlly, consider the
system xy- I =0 (2) yy.. (y,)2 + ayy' = 0

where the. unknowns x,y and the coefficient a


are once again meromorphlc in t. This system
(extracted from a paper of Cassidy [C]) has
the property that the quotient (Xl/X 2, yl/Y2
) of any two solutions (xlYl), (x2,Y 2) Is
again a solutionl so the solutions of (2) form
a group with respect to the multipllcatlve
group of the hyperbola xy- I = 0 and we are
led to another example of "differential
algebraic group". Examples of a more subtle
nature are provided by the systems (3a)
[y2.x(x= IXx-c)0 x"y -xy +ax'y=0 (3b) /y 2x(x - l)(x - t) = 0 {. =y} - 2(2t - l)(x - t)2x'y
+ 2t(t - l)(x - t)2(x'y - 2x'y') = 0 where the
unknowns x y are still meromorphlc
functions In t the coefficient a is
meromorphlc in t and c 0,1 is a constant in C.
These systems (of which (3b) Is extracted

from a paper of Manin [Ma]) have the


property that If (xl,Yl) and (x2,Y 2) are
dietinct solutions of one of the systems, then
their difference, In the group law of the
elliptic curve ^ defined by the first equation
of that system, is again a solutionl so the
solutions of (3a), (3b] together with the point
vI at infinity (0: l: 0) of A form groups with
respect to the group law of A and lead to
other examples of "differential algebraic
groups". A differential algebraic group will
be called of finite dimension if roughly
speaking its elements "depend on finitaly
many integration constants" rather than on
"arbitrary functions". This Is the case with
the "differential algebraic groups" derived
from (1), (2), (3); on the contrary, for

instance, the group, say, of all matrices x 0 y)


0 I z , xg0 0 0 I satisfying the system = ZX
has its elements depending on the "arbitrary
function" y so It Is an "infinite dimensional
differential algebraic group". Differential
algebraic groups are known today but to a
few mathematicians; and this is because
Kolchin's language [K ITK 2 ] through which
they are studied is known but to a few
mathematicians. Yet differential algebraic
groups certainly deserve a much broader
audience, especially among algebraic
geometers. The scope of the present research
monograph is twofold namely: 1) to provide
an algebraic geometer's introduction to
differential algebraic groups of finite
dimension and 2) to develop a structure and
classification theory for these groups. Un]ess

otherwise stated, all results appearing in this


book are due to the author and were never
published before. The original motivation for
the study o! differential algebraic groups in
[C I] and [K 2] was undoubtedly their
intrinsic beauty and variety. Admittedly no
such group appeared so far to play a role,
say, Jn mathematical physics; but as we hope
to demonstrate in the present work,
differential algebraic groups play a role In
algebraic geometry. This Is already
suggested for instance by the implicit
occurence o! the differential algebraic group
(3b) in Manln's paper [Ma] on the geometric
Mordell conjecture. Moreover as shown in
the author's papers [B6], [B 7] differential
algebraic groups may be used along a line
quite different from Manln's to prove the

geometric and infinitesimal analogues of a


dlophantine conjecture of S. Land (about
inter- sections of subvarieties of abolish
varieties with finite rank subgroups). The
latter application will be presented without
proof ]n an appendix. Other applications and
interesting links with other topics of
algebraic geometry (such as deformations of
algebraic groups and thetr automorphlsms,
toodull spaces of abelian varietlas the
Grothendieck-Mazur-Messlng crystalline
theory of universal extensions of abolish
varieties [MM], a.s.o.) will appear in the
body of the text (and will also be touched in
this Introduction). vii 3o Forrealization
Before explaining our strategy and results in
some detail it will be convenient to provide
the formal definition of differential algebraic

groups with which we will operate in this


book. The frame in which this definition will
be given is that of "differential algebraic
geometry" by which we mean here the
analogue (due to Rltt and Kolchln) of
algebraic geometry Jn which algebraic
equations are replaced by algebraic
differential equations. Let's quickly review
the basic concepts of this geometry, cf. [] [K
1 J, [Cl]; for details we send to the first
section of Chapter 5 of this book. One starts
with a field of characteristic zero equJped
with a derivation 6 .' -+ (i.e. with an additive
map satisfying 5(xy) = (6x)y + x(6y) for all
x,y c,) and set = (x c(; 6 x = 0} the field of
constants. We assume that , is "sufficiently
big" as to "contain" all "solutions of
algebraic differential equations with

coefficients in it"; for the reader familiar


with [K 1 ] what we assume is that is
"universal". Such a field is a quite artificial
being; but since all problems related to
algebraic differential equations can be
"embedded" into "problems over ," the use
of this field appears to be extremely useful
and in any case it greatly simplifies
!anguage. fext one considers the ring of
differential polynomials (shortly 6 polynomials) 'fYl'"' Yn} which by definition
is the ring o:[ polynomials with coefficients
in J, in the indetermJnates Yi .... Yn'Y'i ....
..... yk), ,,(k) E.g. the expressions appearing
in the left hand side of the equations (I), (2),
(3) (where we assume now al,a,t t?., t' -- 1, c
t) are A - polynomials. Now for any finite set
Fi,...,F m of i - polynomials we may consider

their set r. of common zeroes in the arline


space An(l) =n; such a set will be called a closed subset of n, _ closed sets form a
Noetherian topology on ?n called the /, topology which is stronger than the Zariskl
topology. For instance the equation (1)
defines an irreducible I - closed subset of the
arline line AI() =? while the systems (2) and
(3) define irreducible /, - closed subset of
the arline plane A2() --7J 2. Any irreducible
a- closed subset r. in n has a natural sheaf
(for the induced a - topology) of - valued
functions on Jt called the sheaf of A - regular
functions; roughly speaking a function on a open set of Z is called A - regular if, locally
in the I - topology, it is given by a quotient of
/, - polynomials in the coordinates. r.
together with this sheaf will be called an

arline differential algebraic manifold


(shortly, an arline - manifold). An
irreducible ringed space locally isomorphic
to an afine A - manifold will be called a
differential algebraic manifold (or simply a
/, - manifold). Given a A - manifold r., the
'direct limit, over all /, - open sets i2 of g, of
the rings of A - regular functions on t is a
field denoted by < g >. The transcendence
degree of .< : over 2/ will be called the
dimension of g and
VIII intuitively represents the number of
"integration constants" of which the points of
. depend. E.g. the A - closed sets given by
equations (1), (2), (3) have dimensions n, 2,
2 respectively. A map between two 6 manifolds will be called 6 - regular if it is

continuous and pulls back A - reuJar


functions into A - regular functions. /e are
provided thus with a category which has
direct products called the category of Amanifolds. Note that any algebraic .- variety
X (reAspectively any X- rayfiery X o) has a
natural structure of A - manifold which A
we denote by X (respectively by Xo); we
have dimX = dim X and dimX = o0
provided dim X ) 0. Finally we define
differential algebraic groups (or simply A group) as being group objects in the category
of A - manifolds, i.e. 6 - manifolds r whose
set of points is given a group law such that
the multiplication r x - r and the inverse r ( T
are A - regular maps. Differential algebraic
groups of finite allmansion will be simply
called 6o ' groups. The equations (l), (2)

clearly provide examples of 6o ' groups. As


for equations (3) the o ' groups which can be
derived are not the A - closed subsets of
A2() given by (3) but their A - closures in (
p2) which contain an additional point (0: I =
0), the origin of these groups. /e leave open
the question whether our - groups "are" the
same with KoJchin's [K2] one can show that
any 6 - group in our sense "is" a 6 - group in
Kolchin's sense. Moreover one can show that
our Ao- groups "are" precisely Kolchin's
irreducible "- groups of 6- type zero".
Finally note that our concept of "dimension"
corresponds to Rltt's concept of "order" [[]
and also (in case we have "finite
dimension") with KoichJn's concept of
"typical 6 - dimension" [K l I2]. 4. Strategy
Cassidy and Kolchin developed their theory

of differential algebraic groups in analogy


with the theory of algebraic group. Our
viewpoint will be quite different: we will
base our approach on investigating the
relations, not the analogies, between the two
theories. Our strategy has two steps. Step I
will consist in developing a theory of what
we caJ] "algebraic D-groups"{ this will be
done in Chapters 1-4 of the book. Then Step
2 will consist Jn applying the latter theory to
the study of do - groups; this will be done in
Chapter 5. Let's explain the concept of
algebraic D-group; for convenience we will
give in this Introduction a rather restricted
definition o! ]t (which will be "enlarged" in
the body of the text). Let 2, 6,: be as in
Section 3 above and let .D =4[ 6] = i>0, 61
(direct sum) be the - algebra of linear

differential operators generated by /, and .


By an algebraic D-group we will understand
an irreducible algebraic - group G whose
structure sheaf (C of regular functions is
given a structure of sheaf of D - modules
such that the multiplication, comultiplicatlon,
antipode and co-unit are D-module maps; in
other words if G x C -G, S .' G - C are the
group multiplication and inverse and if e G
is the unit then for any regular tunctions ,
defined on some open set of C we have the
formulae IX () ( o s) = () s ((e)) = ( Some
ple might Uke to call such a strture "an
gebralc group with nnection alonR we were
[pired in our termilogy by the paper o
Nichols and eisteller [N]. Note however that
unlike in [N] we do t assume C is afine,
imsin instead that C is o finite ty over !

Algebraic D-groups entirely belong to


"algebraic geomry" (rather than to the RittKolchin "differential aJgebraIc geometry")
hence Step I will inevitably performed In the
field o algebraic geometry) the task o
classifying algebraic D-grou wiiJ somim
quite technlca] but In the end rewarding. Step
2 wJJJ be based on the result that "the
category o 6 - rou" is uivalent to "the o
category o[ gebraJc D- groups"I or any ogroup we shall denote by C() the
corresponding algebraic D-group. Note that
(the underlying group o) appears as the group
all ints G(X) or which the evaluation map
(), + is a D-mule map) actuly appears as a 6
- closed subRroup o G() Note that t unton
field (C()) identiias with < >. Tn Step 2 will
deal with the (sometim t so obvious)

translation o properti o algebraic D-groups


into operties of -grou. In ord to get a 1ing aut
the o corresodence G() l's !k at the exampl
we gan with (equations (1), (2), (3)). I Is
derived from tlon (1) then G() is the
algebraic vector oup C n = Sc[[o,[ 1, ... [n.l ]
uid with t D-mole strte o{ its crdinate gebra
defined by 6%0 = l (and by t condition that 6
is a derivation of[[ o ..... l T is derived from
equation (2) then C() Js C m x C a = Sc[XI,
the product the muttipHcatJve d additive
groups equid with the D-struure deid by 6X=
X In case is derived {tom (3a) C(P) can be
proved o oduc A x C while In e a P is
derived rom (3b), C( is a n-trivIa] extension
o the elHpic curve A by in he Iater case G(r)
does t descend o (because A doesn'O. 5.
Results According to the preceedinR section

the c.[assification problem for A o - groups


reduces
x Xl to answering the following question.'
given an algebraic - group G describe all
structures of algebraic D-groups on G l call
their set P(G/6). Note that P(G/6) Js a
principal homogeneous space for the - linear
space P(G/) of all ?,4- linear maps D: G ' G
satisfying equations O) with D instead of 6.
So we are faced with two problems here
namely: l) What irreducible algebraic [/,groups G admit at least one structure of
algebraic [3 - group ? 2) Describe P(G/?,)
for any such G. Both these problems have a
deformation- theoretic flavour: the first is
related to deformations of the algebraic
groups themselves while the second Js

related to infinitesimal deformations of


automorphlsms of our groups. Let's consider
these problems separately and start by stating
our results on the first of them. THEOREM I.
Let G be an affine algebraic ,-group. Then G
admits a structure of algebraic D - group if
and only If G descends to The proof of
Theorem I will be done Jn Chapter 2 and
wlU involve analytic arguments, specifically
results of Mostow-Hochschildt HI l] and
Harem [Ha]. We already noted that Theorem
I may fail in the non-arline case; we shall
describe in what follows a complete answer
to problem 1) in the commutative (nonaffine) case. The formulation of the next
results requires some familiarity with [Se],
[KO]. Their proofs will be done in Chapter
So assume G is an irreducible commutative

algebraic .- group and make the following


notations: L(G) = Lie agebra of G Xm(G) =
Hom(G,G m) = group of multipllcative
characters of G Xa(G) = Hom(G,G a) =
group of additive characters of G. Recall
[KO] that we dispose of the Gauss-Martin
connection: on the de Rham cohomology
space MiR(A) of the abelian part A of G
(where A -- GfB, B = linear part of G =
maximum linear connected subgroup of We
wiU introduce Jn Chapter 3 a "multlplicatlve
analogue" of the Gauss-Manln. connection
which is a ?,- linear map ,V .' Der , +
MOmgr(H)R(A)m,MIDR(A)) , p ,?p where
M)R(A) m J the first hypercohomology group
of the complex * liA/ ' S^/Zi ' ' ... View now
G as an extension of A by B, let S(G) a and
S(G) m be the images of the natural maps

Xa(B) - HI(( A) and Xm(B) - Pjc(A) [Se]


and let SDR(G) a and SDR(G) m be the
Inverse images of S(G) a and S(G) m via the
"edge morphisms" H)R(A) - HI( A) and
HR(A) m - Plc(A). THEOREM 2. Let G
be an irreducible commutative algebraic ?,/,group. Then G admits a structure of algebraic
D - group if and only if Z?(SDR(G) m)
$DR(G)a V(SDR(G)a) SDR(G) a and In
particular if G is the universal extension
E(A) of an abelJan - variety A by a vector
group then G has a structure of algebraic Dgroup (recall that E(A) is an extension of A
by G8 a, g -- dim A, having no affine
quotient)l this consequence of Theorem 2 is
also a consequence of the GrothendieckMessing-lazur crystalline theory [11]. Note
that the algebraic D - group G(r) associated

to the A - group derived from the equation


(3b) above Is a special case of this
construction! Note also that Theorem 2 says
more generally that any extension of E(A) by
a torus G N has a structure of algebraic D group ! m Theorem 2 will be deduced from a
general duality theorem relating the GaussManin connection and its multipllcative
analogue to the "adjoint connection" on the
Lie algebra of commutatlve algebraic Dgroups. Our duality theorem generalizes
certain aspects of the theory in [MM] and
[BBM] and our proof is quite "elementary"
(although computational Let's pass to
discussing "problem 2)" of "describing
P(C/g/,). It is "well known" that the
automorphlsm funclot Au_. C: { ?J,schemes} + {groups} (Au.._t G)(S) =

Auts.grsch.(G x 5) of an algebraic ,- group G


is not representable in general lBS] (by the
way we will prove in Chapter t that the
restriction of Au% G to 'reduced [,schemes} is representable by a locally
algebraic t_ group Aut G this providing a
positive answer to a question of Borel and
5erre lBS] p. 12). Then P(G/?,/,) is
obviously identified with the Lie algebra
L(AU.t ' G) of the .functor Aut G, which
contains the Lie algebra L(Aut G), but in
general exceeds it (due to
"nonrepresentability of Au__!t over nonreduced schemes"). If G Is commutative
P(G//,) is easily identified with Xa(G)L(G)
(where Der/,L0 G is identified with
((G)L(G)). For noncommutetive G the
analysis of P(G/,) will be quite technical; we

will perform it in some detail for G arline


and get complete results in some special
cases (e.g. in case the radical of G is
nilpotent or if the unipotent radical of G is
commutative), cf. Chapter 2. Let's discuss in
what follows the "splitting" problem for
algebraic D-groups. If G o is any irreducible
algebraic - group one can construct an
algebraic D - group G-G 3(, by letting D
act on (G via I ; any algebraic D - group
isomorphic to one obtained in this way will
be called split. A A o - group will be called
split if it is isomorphic to G o for some
algebraic
XII - roup Go this is equivalent to saying that
G(F ) is split. For instance the A o - group T
defined by equation (l) above is split; for if

1' * * * q'n e, is a fundamental system of


solutions of (l) then we have an lsomorphism
of A - group s (Ga,yC) I' (ci,... ,c n) Zcly i
On the contrary the A o - groups derived
from equations (2) and (3) are not split; for
(3b)this is clear because G() does not
descend to while for (2) this follows
because the D-submodule of [X,x'l,x]
generated by X is infinite dimensional. Now
for any algebraic U- group the set of split
algebraic 13 - group structures on G can be
proved to be a principal homogenous space
for L(AutG) so, at least in case G descends
to, the problem of determining what
algebraic D- group structures on G are split
is equivalent to determining what
elementsKor L(Au__tG) actually lie in L(Aut
G). In particular if G descends to and Au__t

G is representable then any algebraic D group structure on G is split; this is the case
if G is linear reductlve or unlpotent. More
generally we will prove in Chapter 4 the
following= THEOREM :. For an algebraic
D-group G the following are equivalent= 1)
G is split, 2) preserves the (ideal sheaf of
the) unipotent radical U of the linear part of
G. 3) 6 preserves the (Ideal sheaf of). U/U n
[G,G] in G/[G,G]. 4) 6 preserves the
maximum semiahelian subfield of the
function field 7.(G) (which by definition is
the function field of the maximum
semiabelian quotient of GI recall that
"semlabellan" means "extension of an
abelian variety by a torus"). Consequently a
A o - group T is split iff T/[I',I'] is so, iff 6
preserves the maximum semlahellan subfield

of < T >I if in addition I' is a A - closed


subgroup of GL N for some N _ J and we put
i :={tdeal of all functions in ?J,{yij)d,
d=det(Yij), vanising on I'} and {I':} := :=
(yij)d/l then T is split [ff all group-like
elements of the Hopf algebra ( 1'} are killed
by 6. V/e would like to close this
presentation of our main results with a
theorem about A - subgroups of abel Jan
varletiesl we need one more definition. V/e
say thaLa 6 o - group T has no non-trivial
linear representation if any A - regular
homomorphlsm r - GL N is trivial.
THEOREM e. Let ^ be an abel Jan -variety
of dimension g. Then there is a unique 6 o subgroup A Its with the following
properties= 1) it is Zariski dense in A , 2) it
has no non-trivial linear representation.

Moreover as A varies in the modului space g


of principally polarized abellan - varieties ,n
with level n- structure, n _)3, the function A
H. dim ^b varies lower semicontinuously
with respect to the A - topoloLy of ,A-g,n
and assumes all values between and 2. XlII
In case g = I one may say more namely= a)
either A descends toZ, say A = ^o )/,, and
then A4b= Ao(7) so dim A It: = I. b) or ^
does not descend to and in this case dim ^ =
2. Note that the A - roup T defined by (3b) is
nothing but the group Afil=above. On the
other hand the A - group defined by (3a) is
an extension of (Ca,) by A '4k above. 6.
Amp!lficmtlmm Most definitions given so far
in our Introduction will be "enlarged** in the
body of the text and most resuits will be
proved in a generalize d form. For instance

the definitions related 'to - manifolds and A_


groups will be given for "partial" rather than
"ordinary" differential fields i.e. for fields ,
equiped with several commuting derivations.
Algebraic D - groups will be allowed to be
reducible and will he defined for D = KiP]
any "k-algebra of linear differential
operators on a field extension K of k" which
is "built on a Lie K/k-algebra P" (cf. [NW]
or Section 0 below this degree of generality
and abstractness might seem excessive, but it
is motivated in many ways= l) our wish to
deal with A. groups over partial differential
universal fields 2) our wish to "compute",
for a given algebraic K-group C (K
algebraically closed of characteristic zero
containing a field k) the smallest
algebraically closed field of definition K C

of C between k and K; the existence of I G


was proved in [B)] but we will reprove this
In a different way in Chapter 4. 3) our wish
to relate our topic to Deligne's "regular Dmodules". 4) our wish to take a (shy) look at
"algebraic D-groups in positive
characteristic". The reader will appreciate
the usefulness of this more general concept
of algebraic D-group in the light of the
arguments 1)-4) above. 7. Plan The book
opens with a preliminary section 0 which
fixes terminology and conventions. In
Chapter I we present the main concepts
related to algebraic D-groups. Chapters 2
and 3 are devoted to affine, respectively to
commutative algebraic D-groups. Chapter 4
deals with algebraic D-groups which are not
necessarily affine or commutative. Chapter

opens with an introduction to - manifolds


and - groups and then deals with the
"structure" and "moduli" of A - groups.
Internal references to facts not contained in
section 0 will be given in the form (XW,z) or
simply (X,y) where X is the number of the
chapter and y Is the number of the section.
V/ithin the same chapter X we write (y,z)
instead of (X,y,z). References to section 0
will be given in the form (0,z). Each chapter
will begin with a brief account of Its contents
and of its specific conventions.
XlV 8, Prerequisites The reader is assumed
to have only some basic knowledge of
algebraic geometry [Har] and of algebraic
groups [H], [Ro], [Se]. The non-experts in
these fields might still appreciate the results

of this book via our comments and various


examples. No knowledge of the KolchlnCassidy theory [Ki][C j] is assumeall the
elements of tills theory which are relevant
for our approach will be quickly reviewed in
the book. Finally one should note that the
present book is ideologically a continuation
of our previous book iBIS, but it is logically
independent of it. 9, Aknowledgements A
few words cannot express the thanks I owe
to Ellis Kolchin and the members o( his
Columbia Seminar: P. Cassidy, R. Cohn, L.
Goldman, J. Johnson, S. Mortson, P.
Landesman, /. Sit. Their continuous
encouragment, constant support and
numerous suggestions played an important
role Jn my fulfilling the task of writing this
book. I would also like to aknowledge my

debt to H. Harem and O. Laudal for useful


conversations. Last but not least I would like
to thank Camelia Minculescu for her
excellent typing job. June 1989 Revised,
June 1991 A. Buium CONTENTS 0.
TERMINOLOGY AND CONVENTIONS
............................ CHAPTER 1. FIRST
PROPERTIES ...................................... I.
Basic spaces and maps
....................................... 2. Prolongations
and embedclings .................................. 3.
Local finlteness and splitting
.................................. CHAPTER 2. AFFINE
D-GROUP SCHEIES ............................... 1.
The analytic method
......................................... 2. The algebraic
method: direct products .......... ' ..................
3. The algebraic method: semidirect

products ......................... CHAPTER 3.


COMMUTATIVE ALGEBRAIC D-GROUPS
...................... 1. Logarithmic Gauss-Manin
conneotion ............................. 2. Duality
theorem ............................................ 3.
Descent. Regularity ................................ ?
........ CHAPTER a. GENERAL
ALGEBRAIC D-GROUPS ..........................
I. Local finitness criterion
...................................... 2. Representing the
automorphism functor ........................... 3.
Products o! abellan varieties by arline
groups ....................... CHAPTER 5.
APPLICATIONS TO DIFFERENTIAL
ALGEBRAIC GROUPS ......... 1. RittKolchin theory ......................................... 2.
A o-groups versus algebraic D-groups
............................. 3. Structure of Ao-groups

....................................... . Moduli of Aogroups .........................................


APPENDIX A. Link with movable
singularities ............................. APPENDIX
B. Analogue of a diophantine conjecture of
S. Lang ................. APPENDIX C. Final
remarks and questions ................................
REFERENCES
.................................................... INDEX OF
TERMINOLOGY .................................... ,
....... INDEX OF NOTATIONS
............................................. l 7 20 2a 60 66
8.t 87 91 99 106 116 125 139
O. TM]IOLOG AND COIT!ONS (0.1)
Unless otherwise stated rings, fields and
algebras are generally assumed associative
commutative with unit. This will not apply

however to Lie algebras, universal


enveloping algebras or to Hopf algebras (the
latter are understood in the sense of
Sweedler [Sw]). (0.2) Terminology of
algebraic geometry is the standard one (cf.
for instance [Har][DG]). Nevertheless we
make the convention that all schemes
appearing are separated. For any morphism
of schemes f X + we generally denote the
defining sheaf morphism C + f, by the
same letter f. If both X and Y are schemes
over some field K then by a K- f-derivation
we mean a K-derivation of COy into the
COy-module f*x' By a K-variety (K a field )
we will always mean a (separatedl)
geometrically integral K-scheme of finite
type. By a (locally) algebraic K-group we
will mean a geometrically'reduced (locally)

algebraic K-group scheme. For any integral


K-algebra A (respectively for any integral
K-scheme X) we let K(A) (respectively
K(X)) denote the quotient field of A
(respectively of X). For any K-variety X and
any field extension K i/K we denote by X(K
1 ) the set HomK_sch(Spec K 1' X) of K lpoints of X; more generally we write X(Y)
instead of HomK_sch(Y,X) for any Kschemes X,Y. A morphism X-Y of Kvarieties will be called surjectJve if the map
X(K) (K) is surjective. For any K-scheme
X, T X will denote the sheaf DerK X. If X =
$pecA for some K-algebra A we sometimes
write Der(A/K) instead,of OerKA. (0.3) All
arline K-group schemes C are tacitly
assumed to be such that the ring /(C) of
global regular functions is at most countably

generated as a K-algebra. This is a harmless


assumption in view of our applications in
Chapter . The unit in any K-group scheme G
will be denoted by el If G is commutative we
will sometimes write 0 instead of e. For any
locally algebraic K-group G, G will
denote the identity componentl Z(G) will
denote the center and we put Z(C) .'=
(Z(C)) . (0.) For any algebraic C - variety
X (respectively algebraic C - group G) we
denote by X an (resp. C an) the associated
analytic space (respectively the analytic Lie
group). For any analytic manifold we denote
by 1 the analytic tangent bundle. (0.5) Our
terminology of differential algebra is a
combination of terminologies from [C 1] [K
i] [NW] and [Bi], In what follows we shall
review it in some detail and also introduce

some new concepts. Let K/k be a field


extension. By a Lie K/k - algebra [NW] we
mean a K-vector space P which is also a Lie
k-algebra, equipped with a K-linear map a: P
- DerkK of Lie k-algebras such that [Pi,XP2]
= a(pl)()p 2 + [pl,P2 ] for eK, pl,p 2 P
Let's give some basic examples of Lie K/k algebras. EXAMPLE 1. Start with a
derivation d e DerkK; one can associate to it
a Lie K/k - algebra P of dimension I over K
by letting P=Kp (Xp)=Xd, XeK [),lP, X2p] =
(XldX 2 - X2dXl) p , ),l,X2 K So we call
the attention on the fact that this P is far from
being commutatJve! EXAMPLE 2. Start now
with a family of derivations (di)ie l, d i e
DerkK. Then one.can associate to it the "free
Lie K/k-algebra" P built on this family: by

definition P has the property that it contains a


family of elements (Pi)lel with (pl)= d i such
that for any K/k-algebra P' and any family
(P)il' P e P' with l)'(p)--d i there Js a
unique Lie K/k - algebra map f .' P + P' with
f(pl ) = P'i' We leave to the reader the task of
constructing this P. Note that if I consists of
one element this P Is the same with the one In
Example 1. EXAMPLE 3o Assume we are
given a family (di)ie I , d i DerkK such
that [di,d j] = 0 for all i,j I. Then one can
construct a new Lie K/k-algebra P as
follows: we let P have a K-basis (pl)ie] we
let (pi) = d i and define the bracket [, ] by the
formula [,pi,pPj] = X(dlp)pj - p(djX)pi , X,I
K, i,j I This P can be called the "free
lntegrabJe Lie K/k-algebra" built on our
family of derivations. Once again if I

consists of one element, this P coincides with


the one in Example I. EXAMPLE e. Let k
E c K be an intermediate field. Then P -DerEK together with its inclusion = DerEK DerkK is an example of Lie K/k - algebra.
Other remarkable examples of Lie K/k algebras will appear in (I. 1). Given such a P
one' associates to it [NW] a k-algebra of
differential operators D -- by definition K[P]
is the associative k-algebra generated by K
and P subject to the relations Xp = v(X,p) for
X K, p P where v: K x P - P Is the
vector space structure map p3.-Xp= (p)(X)l
forXeK, peP PiP2 ' p2Pl --[pl,P2 ] f pl,P2 e
P Ral[ [NW] tt D K[P] has a K-basis
consisting o all the momlaJs o the form = . ..
Pin wre ej 0 and i I .., I n in some total orr on
a basis (pi)i of P, For instance l-dimenslonal

P = Kp in Example I above then D = E Kp I


(direct sum) i>0 is the ring of "linear
ordinary differential operators" on K
"generated by K and p" multiplication in D
corresponds to "composition of linear
differential operators". This example is quite
familiar: indeed, assume k = C, K = C(t) is
the field of rational functions and (p) -- d/dr;
then D = C(t)[d/dt] is nothing but the
"rational" Weyl algebra. Similarity one can
obtain the "rational" Weyl algebra in several
variables D--C(tl,... ,tn)[d/dtl,... ,d/dtn]
starting with P as in Example 3 above. Since
D is a ring we may speak about D-modules
(always assumed to be left modules). If V Is
a D-module we define its set of P-constants
V D -- x px=0 for all p eP it is a vector space
over the field KD= ), eK; px=0 for all peP},

We sometimes write also V P and K P


instead of V D and K D and speak about Pmodules instead of D-modules. Recall from
[NW] that if V,W are D- modules then V
W V KW and HomK(V,W) hence in
particular o = HomK(V,K), have natural
structures of D - modules and the followin 8
formulae hold p(x,y) = (px,py) for p P,
(x,y) c V W p(xy) = pxey + xepy for peP,
xy eVKW (pfXx) -- p(f(x)) - f(px) for p P,
x e V, f HomK(,W) By a D - algebra
(respectively associative D - algebra, Lie D
- algebra , Hopf D - algebra) we understand
a D - module A which is also a K - algebra
(respectively an associative K - algebra, Lie
K - algebra, Hopf K - algebra) in such a way
that all structure maps are D - module maps
here by "structure maps" we understand

multiplication AK ^ -^, unit K


comultiplicatJon A - A K A and co-unit A K(^ K A and K are viewed with their natural
D - module structure). A D - algebra is
called D - finitely generated if it is generated
as a K - algebra by some finitely 8erierated
D - submodule of it. Note that if P Is as in
Example I (respectively 2, 3) a D - algebra
Is simply a K - algebra A together with a
lifting of the derivation d e DerkK to a
derivation l c DerkA (respectively with
liftings of d i to i DerkA in case of
Example 2 and with pairwise commuting
liftings i Derk^ in case of Example 3). A
D - algebra which is a field will be called a
D - field. Clearly K is a D - field. K will be
called D - algebraically closed if for any D flnitely generated D - algebra A there exists

a D - algebra map A - K. Of course one


would like to "see" an example of Dalgebraically closed field. Unfortunately we
can't "show" any (although one can prove the
existence of enough of them see (0.12)
below). But one should not forget that with a
few exceptions a similar situation occurs
with algebraically closed fields. Note that If
^ 1' A2 are D - algebras then A 1 )KA2
becomes a D - algebra. Clearly if ^ is a D algebra then Im(P- EndkA)C Derk^. If no
confusion arrlses we denote the image of any
p P in DerkA by the same letter p. (0.6)
Following [151] we can define D schemes= these are K - schemes X whose
structure sheaf (.0 X is given a structure of
sheaf of D - algebras (J.e. (x(U) is a D algebra for all open sets U and restriction

maps (x(Ui)+ x(U2) for U2c U 1 are Dallebra maps). For any
D - algebra A, SpecA is a D - scheme.
Clearly D - schemes form a category (whose
morpHisms will be called D- morphisms or
D- maps: these are the morphisms f= X-Y
such that {Yy + f. (. X IS a morphism of
sheaves of D - algebras). This category is
easily seen to have fibre products. Indeed if
X and Y are D - schemes over a D - scheme
Z then the scheme fibre product X x ZY has a
natural D - scheme structure cominig from
the D - module structure of tensor products
(0.)l if p =P then the corresponding
derivation on the structure sheaf of X x zY
will be denoted by p l + ] p (this
convention agrees with formula "p(x y) = px

y + x py" on products of arline pieces=


XoXZoYo , Xo Yo' Zo affine open subsets of
X, Y Z). A D - scheme wUl be called' of D finite type i! it can be covered with fin]rely
many affine open sets U l whose coordinate
rings are D - finlteiy generated. By a D variety we understand a D - scheme X which
is a K - varietyl we shall usually write 1( <
X > instead of K(X). Clearly K < X > is a D
- field. (0.7) Let A be a D - algebra; by a D ideal we mean an Ideal 3 which Is a D submodule of A. For any D - ideal the
quotient A/3 is a D - algebra. Let X be a D scheme; by a closed D - subscheme we mean
a subscheme Y whose sheaf o! ideals is a
sheaf of D-submodules of (X (so Y Is in
particular a D-scheme). By a D-point on a
D-scheme X we mean a (non necessarily

closed) point x X such that the maximal


ideal of ('X,x is a D - ideal; we denote by X
D the set of all D - points of X. l! X,Y are D
- schemes we write XD(Y) instead of
HomD.sch(Y,X); If Y = SpecA for some D algebra A we write XD(A) instead o!
XD(Y). Elements of XD(A) are called D - A
- points. (0.8) Now comes our main concept.
By a D - group scheme we shall mean a
group object In the category of D - schemes
i.e. a D - scheme C which is also a 1( group scheme such that the multiplication G x
KG -G, the antipods G G and the unit Speck
G are D - morphisms. By an algebraic D group we will understand a D - group
scheme which is also an algebraic K- Igorup.
(This is not the concept of algebraic Dgroup from [NW]!). One defines in an

obvious way the notions o! D - subgroup


scheme and algebraic D - subgroup. (0.9) If
X and Y are two D - schemes then to give a
D - map f X -Y is equivalent to giving for
any D-algebra A a map fA "XD(A) 'YD (A)
behaving functorially in A. If both X and Y
are reduced it Is sufficient to define fA only
for reduced A. In particular any D - group
scheme C can be recaptured from the funclot
{D - algebras} - (groups] A CD(A) (0.10) l!
KI/K is a D - field extension (i.e. a
morphism K -K l of D - fields) then one
easily checks that Pl =1(IKP has a natural
structure of Lie Kl/k-algebra and that D l -'=
Ki[P i] identifies with K I K D. IUctreover,
if V is a D-module then K l K V has a natural
structure of Di-module. A similar assertion
holds for algebras, associative algebras, Lie

algebras, Hopf algebras, schemesy group


schems, varieties, aligebraic groups (instead
of modules). If X is a D-variety we will
often write K 1 < X > instead of K 1 < K I
I2}KX >' (0.11) An important remark is'in
order. Namely the above D-schemes and Dvarieties are objects quite close to the 6 schemes and A -varieties considered in [Bi]
more precisely A - schemes over a A - field
K in the sense of [B 1] coinolde with D schemes in the above sense where DK[P], P
= free Lie K/k- algebra on fi, k--K A. But
our algebraic D-groups above are objects
quite different from (although, as we shall
see, deeply related to) the Cassldy-KoJchln
6 - groups [Ci][Ki]. The relation between the
latter concepts will be discussed in Chapter
5. Finally the algebraic D-groups from [NW]

are precisely our arline D-group schemes]


One should also note that Kolchln's /, - fields
[Ki]:K 2] correspond In our language to
fields K equiped with a Lie K/k-algebra P
which has a finite commuting K-basis. (0.12)
Kolchln proved In [K 3] that if P is a Lie
K/k-algebra having a finite commuting Kbasis (char K = 0) and D = KiP] then there
exists a D-field extension Kl/1( such that K 1
is D l-algebraicaUy closed, D l = K 1 K D'
For the reader familiar with [!( 3] note that K
is D-algebraically closed iff (viewed as a A
- field) it Is constrainsally closed in
Kolchin's sense. (0.13) Various other
standard definitions can be given in the
category of D-schemes. We shall need for
instance the concept of left (respectively
right) D-action of a D-group scheme C on a

D-scheme X it is of course a D-map C x XX (respectively X x C - X) satisfying the


usual axioms of an action. Slmllarlly, one
can define D-principal homogeneous spaces
(these are right D-actions X x G - X such that
the induced D-map X x C - X x X is an
lsomorphlsm). One can speak moreover
about D-actions of a D-group. scheme C on
another D-group scheme H and about Dcocycles of C in H. It is useful to have some
notations also. We let PHSD(C/K) be the set
of D-isomorphism classes of D-principal
homogeneous spaces for a D-group scheme
G; it is a pointed set with distinguished
element called I and represented by G itself.
Any D-principal homogeneous space V
representing I will be called trivial; V is
trivia] Jf and only if VD(K) . (0.1) Let's

"recall" from [B 1 ] Chapter I some useful


elementary facts about D-schemes. Let X be
a D-scheme, X noetherian. 1) If X Is
reduced, its irreducible components are Dsubschemes. 2) If Y is an Integral subscheme'
of X its generic point is in X D If and only If
Y Is a D-subscheme. 3) I! charK -- 0 then
Xre d Is a D-subscheme of X. In particular
by 1), 2) above X D . Assume moreover that
K 'Is D-algebraically closed and X IS of Dfinite type. Then XD(K) t, t) If char K = 0
and X is a D-variety then for any maximal
element x in XD the extension K(x)D/K D is
algebraic (where K(x) -- residue field of X
at x). 5) I! f.. X - Y Is a faithfully flat
morphism of integral 1(-schemes and K(Y)
Is a D-subfield of K X ) then Y has a
(unlquel) structure of D-scheme such that f is

a D-map. 6) If f X' - X is an tale morphism of


K-varieties then X' has a unique structure of
D-variety such that f is a D-map.
7) If f: X + Y is a D-map of D-schemes and y
c YD then the fibre f-I(y) has a natural
structure of D-scheme. g) Assume V is a
normal K-variety such that K(V) has a
structure of D-field, D = K[P]. Assume
moreover that dimKP < . Then the locus V O
= [x Vj D/V, x a /V,x } is ZarJski open (so
it is a D-scheme) and V \ V o is the support
of some divisor on V. Recall also that we
have proved in lB l] p. 28 the following:
(O.l) THEOREM. Assume K is algebraically
closed, and X is a projectire D-variety. Then
K D is a field of definition for X (i.e. X is
K-isomorphic to XoKDK for some

projective KD-variety Xo). (0.16) We close


by introducing a concept which will appear
several time in our work. Let P be Lie K/k
algebra, S m a group and 5 a a Lie K-algebra
on which 5 m acts by Lie K-algebra *
automorphisms. By a logarithmic Pconnection on (Sm,S a) we mean a pair (?,?)
of K-linear maps ? = P Derk(Sa,S a) P ?p g?
-' P -' zl(Sm,Sa ) p such that Vp(),x)= ),pX +
a(p)(X)x for all p P, ), K, x cS . Now
Zi(Sm,Sa ) has a natural a structure of Lie Kalgebra induced from that of S a. We say that
(?,?) is lntegrable if for any pi,p 2 P the
following formulae hold; (0.16.1) ?[PI'P2 ] =
[?Pl' ?P2 ] (0.16.2) g[Pl'P2 ] = Pl P2 P2 PI +
['VpI' tVP2 ] Of course condition l) is
equivalent to saying that there exists a
(unique) D-module structure on S a (D =

K[P]) such that px = VpX for all pP and x


c5 a. A "trivial" example of integrahie
logarithmic DerkK-connection is on (K*,K)
where K* acts trivially on K and K is
viewed as an aheiian K-algebra= it is given
by the formulae ?pX = px, lpy = y-lpy for all
p cDerkK , x K, y zK*. Various significant
examples o! logarithmic connections will
appear in (I.1) and (IlL l). Note that if in our
definitions 5 m is trivial and $a is an abelfan
Lie algebra then a logarithmic P-connection
is simply a P-connection on S i.e. a K-linear
map ?: P ' Homk($,$) satisiying the usual
Leibnitz rule. CHAPTER !. FIRST
PROPERTIES Everywhere in this chapter
K/k is a field extension with K algebraically
closed (of arbitrary characteristic) P is a Lie
K/k-algebra and D = K[P] (cf. Section 0)o in

section ! of this chapter we associate to any


K-group scheme G a basic Lie K/k-algebra
P(G) and investigate its "first properties".
Sections 2 and 3 deal with basic concepts of
proJongations local finitehess and splitting.
1. Basic spaces and maps (l.l) Let G be a Kgroup scheme. We shall denote by p::GxC G, S.'G - G, : $pec K - C the multiplication,
antipode and unitl we will also denote by P:
CG" P*GxG 5; (G' S. G ,e the induced
comultipllcatlon, antipode and co-unit (cf.
(0.2)). Define DerkG -- !space of all kderivations of (fig into itselfJ L(G) = v c
DerkG v is K-linear and pv P(G) = p c
DerkG; pK'K, IJP = (P I + 1 p)p, pS = Sp, p
= p} Clearly DerkG is a K-linear space and
a Lie k-algebra. Moreover L(G) is a Lie Kalgebra and P(G) is a Lie K/k-algebra. Both

L(G) and P(G) are Lie k-subalgebras of


DerkG. Define D(G) = K[P(G)]. loreover for
any p P(G) let Kip] be the k-algebra of
differential operators associated to the ldimensional Lie K/k-algebra Kp (clearly
K[p] c KIP]). Note that we made a slight
abuse in our deflnitlons. For Instance the
formula Pp = (p 1 + lp)p in the definition
of P(G) means that the following diagram 'of
abelian sheaves on G is commutative.' (C P '
P* CxG Where p 1 + 1 p was defined in
(0.6). On the other hand the formula pv = (v
l)p in the
definition o! L(G) means that the diagram 7
G P . W. GffGx G [ { W.(v, l) is commutative
where v I is the K-derivation of GxG acin as
v on the "firs coponent" and vanishln 8 on

:he "second component". Such abuses reatly


simplify our computations and will be done
Yrom now on sys{emaYlcally. ote that if
char K > 0 then DerkG = DerKG because K
ls rfect. e can define more generally {or any
intermediate field K o between k and K the
spaces er K G and P(G/K o) as consisting of
all Ko-linear embers of DerkG and P(G)}
clearly P(G/K o) ls Lie K/Ko-algebra and
P(G/K) = Ker( a = P(G) + DerkK). Aprlorl
it ls t clear that L(G) can lden{lfled with the
Lie algebra Lle G o{ G as dellned in [DG]
(Lie G = tangent space of G at e). But of
course, i{ G is an algebraic group, L(G)
coincides with the Lie algebra o{ right
invariant derivations in erKG and hence with
Lie G. e wiU check below that L(G)-Lie G at
least in one ore case namely when G

integralmar{inc (see (1.8)). As {or one


interpretation of P(G) note that there ls a
natural bljectlon be{ween the set of D-grou p
scheme structures on G d the se HLi e
K/k.alg(P, p(G)) In particul&r G has a
structure o{ D(C)-group scheme and this
structure is "universal" in an obvious sense.
This remark is basic for our work. e wlU
usuly investigate first this "universal" D(G)strture and then "specialize" to =bltrary structures. EMPLE. Assume P = Kp is 1dimenalonal} then the set of D-group scheme
structures on G is in bljection with the set
P(G/a(p)) o{ all derivations } in P(G)
inducing the derivation a (p) on K. So P(G/a
(p)) is a prlncll homogeno sce {or P(G/K).
Before computing P(G) for a few sci{lc G's
we nd to prove a series oY general prortles.

(1.2) LEMMA. [P(G), L(G)]c L(G) where t


bracket Is taken }n erkG. P{. Let p P(G),
v L(G). We have: Upv=(pl + lp)pv=(pl + lp)
(vl)p=(pvl +v pvp=(vl)pp=(vl)(pl + lmp)u=
(vpl +v hence t[p,v] = ([p.v] l)l which
proves the lemma. We denote in what
follows by G m, C a the multlplicative and
additive l-dimensional algebraic K-groups.
Recall that G m -- spec K[X,X- 13 pX =
xx K[X,X 'i ] K[X,X' 13 SX= X '1
and that G a =Spec e= 0 For any K-group
scheme we G put Xm(G) =
HOmK_grsch(G,G m) Xa(G) = Horn K
.grsch (G ,G a) Clearly Xm(G ) is a
subgroup of (*(G) and Xa(G) is a K-linear
subspace of (G). (1.3) LEMMA. For any p
P(G), X Xm(G), eXa(G ) and v
L(G) we have X-Ipx Xa(G), peXa(G) and

v K. Proof. The following holds: ppX =


(p ) 1 + 1 p)p( = (p 1 + 1 p)(XX) =
pXX + X pX. Multiplying the above
equality by X '1 )X '1 we get p(x-lpx) =
X'Ipx 1 + I ex-lpx I.e. X-Ipx Xa(G).
Similarily one checks that peXa(G). Finally
we have pv = (v l)p = (v 1)( 1 + 1 ) = v aO 1
Applying {) 1 to the above equality we ge v
= (cO l)Vv = (e lXvao 1) = e(v) K and
the lemma is proved.
10 (l.e) We have an integrable logarithmic
P(G)-connectlon (9,[9) on ((*(G),(G))
defined by the formulae Vpf -- Pf and [Vpg =
g-lpg for all p P(C), f iC(C), g i Ca(C).
Here L(C) is viewed as an abelJan Lie Kalgebra on which 7(G) acts trivially. By
(1.3) above this logarithmic connection

induces an integrahie logarithmic P(C)connection (?[V) on (Xm(G) , Xa(C)) where


Xa(C) is viewed as an abelian Lie K-algebra
on which Xm(C) acts trivially. In what
follows we prepare ourselves to deal with
L(C) for G affirm and with P(C/K) for C
commutative and either affirm or algebraic
over K. (1.5.) Assumed G is an affirm Kgroup scheme. Then we define a linear
topology on DerkG as follows. For any
system of elements fl'" ' fn (C) define Vfj,
... 'fn = {p c DerkG Pfj =... = Pfn = 0} Then
we take as a fundamental system of open
neighbourhoods of the origin in DerkC the
above spaces Vf ... f as fl ... fn and n vary.
Clearly DerkC with this topology is
separated and complete due to bur
convention made in (0.3) this topolo[y has a

countable basis). (1.6) Recall from [DG] that


(with our convention in (0.3)) any arline Kgroup scheme G is the projectire Jim Jr C =
J._ G i of an inverse system u i ... 4-Ci+l--Gl
4-... +G 1 such that each u i is a faithfully flat
morphism of arline algebraic K-groups. In
particular LleC = lira LieC l = li_m._m L(G
i) (!.D Let .. 4-Vi+ l 4-V i 4- .,. 4-V 1 be a
projective system of finite dJmenslonal Kvector spaces with surjectlve connecting
maps and let = Jl.m_m V i be viewed
with its projectire limit topology. Then one
can find a sequence vl,v2,... I (we agree to
assume that the sequence is finite if V has
finite dimension) such that for each index i
there exists an integer n i having the property
that upon letting j be the image of vj in Vp l'
'" n i is a K-basis in V. and . = 0 for all j ) n

i. Call such a sequence (vi) I a pseudo-basis


for V. Clearly any element in can be
written uniquely as a convergent series Z),iv
i with )'i K. If dimkV < m a Pseudo-basis
is simply a basis. (l.g) PROPOSITION. Let
G be an affirm integral K-group scheme and
let G = lilieS_ram G i as in (1.6). Thens l)
For any intermediate field K o between k and
K the spaces DerKoG , P(G/Ko) and L(C)
are closed in DerkG hence they are
complete. 2) Let VlW2, ... be a Pseudo-basis
in li,_m L(G i) (we may view each v i as an
element of 11 DerkC). Then any element in
DerKG can be uniquely written as r. aiv i
with a i (G) (which is a convergent series in
DerkC). 3) An element r. alv I as in 2)
belongs to L(C) if and only if a I K for all l
in particular L(C) - lJ?_m L(Ci) , hence L(C)

identifies wlth Lie G. a) Assume C is


cornmurat[re. Then an element ]:alv i as in
2) belongs to P(G/K) if and only if a i I
Xa(G) for all 1. Proof. 1) is easy and we
omit proof. 2) Let fl "C 4- C l be the i-th
canonical projection. We have DerKC =
l/+mDerK((Ci , fi2c)--lira Hrn(C,K,fi.(.0G)
- - = lim H(Gf,(fl.07 G) x L(Gi)) = lim
((,0(G) x L(Gi)) which immediately implies
assertion 2). 3) Clearly 11mL(Gi)c L(C) so
r.).ivicL(C) If ).jcK. Conversely assume
v=]:ajvi, pv = (v ) l)p. Then w have
r.P(ai)PV I = r.(a I l)(v i e l)p = Z(a i l)pv
i V/e shall be done if we check that an
equality of the type r. mlPV i -- 0 with m i (C
x C) implies rn i = 0 for all i. For each i we
may find elements fl' ' '' fn. in the maximal
ideal m i of TGi,e 1 providing a K-basis In

mi/mi2 and such that Vpfq = 6pq (Krormcker


delta) for all I _( p,q _( nj. Then det(v f )._ 60*- V/e get P q l_Pq_n i i,e' ilm pv f 0 for I
( q ( n i p--I p Pq - - Now der (pVpfq) -p(det (Vpfq)) OlXCl,ex e which implies that
mp= 0 for all I (_ p _( n I. ) If G is
commutative so are all Gi's so for all I we
have pv I = (v i l)p = (1 vi) p and SviS -- -v
i Now for p = t. aiv I we have pp = p(r. aiv
i) = r.P(al)PV i (pO 1 + I p)p -- Z(a i l)
(Vl l)p + r.(l ai)(l vi) p = r.(al 1 + 1
ai)Pv i Sp = r.(Sal)Sv i
12 pS = :aiviS = -r..aiSv 1 p = rdal)V l p=O
So If a i c Xa(G) clearly p c P(G/K).
Conversely, if p c P(G/K) then by the remark
made at point 3) we have a I c Xa(G) and we
are done. (1.9) PROPOSITION. Let G be an

irreducible commutative algebraic K-group.


Then, under the identification DerKG C(G)
L(G) we have P(G/K) = Xa(G) L(G). (l.)).
Proof. Same computation made in (1.8) (with
the "abusive" notatJonal conventions from
Now we are In a postion to compute P(G/K)
for a few special commutatlve G's.
EXAMPLE 0. Let G = G n = Spoc K[X 1' '' '
(n ] be a torus (OX 1 = X i )Xi). Then
P(G/K) = 0. m EXAMPLE 1. Let C =-C n =
SpecK[l,... 'in ] be an algebraic vector group
a (li = [1 I + I [i ). Then a K-basis of
P(C/K) consists of the derivations [ a lj ' 1
_< i,j<_n of K l .... ;n]. EXAMPLE 2. Let G G m xGa= SpecK[x,X-I] (1 --XX, P[ -'[I +
1[). Then a K-basis of ,(G/K) consists of the
derivations [- and [X of K[}(,('I].
EXAMPLE :. Let G=AxG =AxSpecK] (IJ[--

[l.+ l[)whereA is the elliptic a curve with


K(A)= K(x,y), y2= x(x - l)(x -A), A c K.
Then a K-basis of P(G/K) consists of the
derivations and (l.10) Let G be a K-group
scheme. Then G(K) acts by adjoJnt
representation on L(G): for any gc G(K) and
vE: L(G) we put (AdgXv) = LlvLg where Lg
(G* Lg(G is induced by left translation Lg:
G* G defined by g. By (1.2) we dispose of a
K-linear map ?: P(G)+ Derk(L(G),L(G)), p-p defined by pV = [p,v] (bracket in DerkG)
for p c P(G), v c L(G). Then define a Klinear map J,?; P(G)+ ZI(G(K),L(G)), p J, P
by the formula J[Vpg -- LIpLg - p for pc
P(G), gc G(K) The fact that J,?pg c L(G)
follows from the following cornputat[om 13
p(LglpLg - p) = pLgIpLg - pp = (L;I ) l)ppLg
- pp = = (L 1 lp) 1 + 1 ) p)pLg - (p 1 + 1 p)p

= ((L 1 ) l)(p ) 1 + 1 p)(Lg< 1) - (p I + 1 )


p))p = = ((LIpLg - p)O 1) It is easy to check
that the maps V and I,V define an integrable
logarithmic P(G) - connection on (G(K),
L(C))I in particular L(C) hcs a structure of
D(G)-module (in fact of Lie D(G)-algebra)
such that for p c P(G), v c L(C) we have pv
= [p,v] (pv is v multiplied by pin the D(G)moduJe law there is some danger of
confusion here with the composition p so one
should be careful to distinguish between the
two' possible meanings of the symbol pv;
this Js why we sometimes write pV instead
of pv to denote The above structure of D(G)module on L(G) will be sometimes called
the "adjoint connection" on L(G). (I.11)
PROPOSITION. Let G be an integral Kgroup scheme which is either afflne or

algebraic. Then the map J, V .. P(C) +


DerkK Z I(G(K),L(G)) is injectlye.
Moreover for any pc P(G) and g cG(K) we
have J,?pg= 0 if and only if g c CK[p](K)
(recall that this means by definition that g:
Spec K + G is a Kip]-map). Prooo Assume
pc P(C/K) = Ker a is such that [Vpg = 0 for
all g c C(K). Then of course (1.8) holds for
Lleft(C) := (v c DerKC; pv = (1 ev)p}; in
particular picking a pseudo-basis Wl,W2, ...
in Ljeft(G) if G is arline (or simply a basis if
G is algebraic) we may write p = r. atw 1
with a i c (C). Clearly LlwjLg = w l for all I
so the relation 0 = J[Vpg = LlpLg - p implies
r..ajwj = r.(Llai)w i hence by (1.8) a I = -ga i
for all i and g which implies a i c K for all 1,
hence that p c Lieit(G) so pp = (1 p)p.
Applying 1 ) to the latter equality we get p =

(lp)p. But p = p = 0 so p = 0 and


injectlvity follows. Next, for a fixed g c
C(K) we have J,?pg = 0 If and only if Lg: G
+ G is a Kip]-map which happens if and only
if g is a Kip]-map from Spec K to G
(because Lg = p(l g) and g = Lg). Our
proposition is proved. (1.12) REMARK. Let
G be an algebraic K-group and K o be a
field of definition (say C; C O K K, G o a
Ko-group). For each p c Der K K let p*
denote the trivial lifting of a(p) from K
toG GoK (i.e. p* = I alp)). Then te map
G(K)* L(G) defined by g [Vpg coincides
with Kolchin's 1ogarlthmlc p-derivative [K
1 ] p. 39t; an argument for this can be found
in [Bi]p. 25.
14 In what follows we shall use the

universal enveloping algebra of the Lie


algebra of algebraic D-groups G in order to
establish a D-analogue of the classical Lie
correspondence between algebraic
subgroups of G and their Lie algebras. (1.13)
Let L be a Lie D-algebra. Then the universal
enveloping K-algebra U(L) has a natural
structure of associative D-algebra induced
by the D-module structure of the tensor
algebra (0.5). 50 by (0.5) the dual U(L)
(which is a commutative algebra with
convolution product) becomes a D-algebra.
Now for any K-group scheme G there Is a
natural K-algebra map JG: 'G,e ' U(L(G))
defined as follows: for f G,e' JG (f) =7
where for any u U(L(G)), ' (u)= cur (here
U(L(G)) acts on G,e by K-linear
endomorphisms in the natural way). I! in

addition G has a structure of D-group


scheme, then by (1.10) L(G) has an induced
structure of Lie D-algebra and one easily
checks that the map iG above is a map of Dalgebras. (1.14) LEMMA. If char K = 0 and
G is integral and either affine or algebraic
then the map jg :.CG,e- U(L(G)) in
injectlye. In particular, if G is algebraic
dlmKP(G/K) < e. REMARK. If char K > 0,
P(G/K) is in general infinite dimensional
even if G is algebraic, e.g. for G = G a
(Indeed apply (1.9) and the fact that
dimKXa(Ga) -- ). Proof. First it is easy to
check the Lumina for G algebraic. Then i! G
is arline write G = llmG. as in (l.t;)l we have
(G = lira and a commutatlve diagram . I ,e .4.
Li,e (G i,e JGi . U(L(Gi)) G,e JG '
U(L(G)) where JG. and u i are iniective

(for u i use (1.8), 3)). This implies


injectivity of JG' The claim that dimKP(GK)
< m for algebraic G follows form the fact
that P(G/K) embeds into IerKL(G). (1.15)
Recall from [DG] that if G is an (integral, to
simplify) afflne ,K-group scheme and H is a
normal subgroup scheme of G, then the
quotient G/H exists as an afine K-group
scheme. It can be constructed as follows: let
G = 11L.mG 1 as in (l.t9, let 3 H be the ideal
of H in(G), 3 i = ((G i) n H (it is a Hopf
Ideal in (Gi)) and put H i = Spec/(Gi)/1111
then G/H = 11_mGi/H i. Assume now H is
integral. By (l.g), 3) one can easily check
that L(H) Identifies with { v v3 H c 11H}
and by (1.8) we have an exact sequence 0 '
L(H) ' L(G) ' L(G/H) ' 0 15 (1.1)
PROPOSITION. Assume char K = 0 and H

is an integral K-subgroup scheme of an


integral D-group scheme G which is either
affine or algebraic. The following are
euivalent: 1) H is a D-subgroup scheme of
G. 2) L(H) is a D-submodule of L(G). In
particular if G is algebraic then [G,G] and
Z(G) are algebraic D-subgroups of G.
ProoL We have a commutative diagram G,e
Jc U(L(G))O ,e ' U(L()) J with r,s
surjective and JG' JH injectlye (cf. (l.lt) and
(1.15)). If 2) holds then s is a D-algebra map
hence its kernel is aD-ideal in U(L(G)) .
Since Kerr = jl(Ker s) we get that Kerr is a
D-ideal which implies that H is a Dsubscheme, hence a D-subgroup scheme of
G. Conversely, if the latter holds the ideal
sheaf 11FI of H in LO G is a D-ideal sheaf in
(G hence if ve L(H), identifying v with an

element (still denoted by v) in L(G) such that


V3H 3H, we have (for any pe P) pV)11 H
= (p e v)(3H)+ (v p)(gH)c 1114 , so ? pV
L(H). The last assertion of the lemma
follows from the fact that L([G,G])=
[L(G),L(G)] and L(Z(G)) = Z(L(G)) (=
center of L(G)) are clearly D-submodules of
L(G). (1.17) LEMMA. Assume char K -- 0,
G is an integral D-group scheme which Is
either algebraic or arline and H is a normal
(nonnecessary irreducible) K-subgroup
scheme. The following are equivalent 1) H is
a D-subgroup scheme 2) There is a structure
of D-group scheme on G/H such that the
projection G -G/H is a D-map. (Clearly the
D-group scheme structure on G/H in 2) must
be unique I). Proof. 2)=1) is clear by (l.lt0,
7). 1)=2) By (0. It0, 5) it is sufficient to

check that K(G/H) is a D-subfield o! K(G).


I! G Is arline let G = Ii.mGi,G/H =
11._mGi/FI 1 as in (1.15). If G Is algebraic,
put G i = G, H i = H for all t. Letting i: Gi x
H? G l be deduced from the multiplication i l
-' G I x G i G i we have K(G/H) = l+im
K(Gi/Hi) = lim( fe K(Gi); if = f O where
we still denoted (as usual) by i the Induced
map CGi- i, (-0GlXH i' Now if f K(G i) is
16 such that lf -- f l and if p c P then f = fa 1(:
G x H - G having the obvious meaning) so
pf: (p! + 1 ) p)f = (pl + I p)(f )!) -- pf)l But
pf c K(Gj) for some j hence pf K(G/H)
and we are done. (1.18) pROPOSITION.
Assume char K = 0, G is an irreducible
algebralc D-group and H any algebraic Dsubgroup of G. Then the homogenous space

G/H has a natural structure of D-variety such


that the projection map G + G/H is a D-map.
Preof. Same argument as in (1.17). (1.19)
PROPOSITION. Let C = C l x C 2 be a
product of K-group schemes. Then there is a
natural Lie K/k-algebra map s -' P(G
l)XDerkKP(G2)' P(G) and a natural K-linear
map "over" DerkK: li: P(C) P(G
l)XDerkKP(C2 ) such that Rs = identity. In
particular for any p P(G) we have p-slip
P(C/K). Preof. Let E i: G + G l be the
canonical projections (and denote as usual
also by II 1 the induced maps (C. ' Ei.CG)'
Then for p P(C) define li(p) = ((l 2)PE l,
(l l)pE 2) and s(PI'P2) = Pl l + P2
where of course ti Is the co-unit of G l. Then
one checks easily all claims of the
proposition. The next proposition shows that

the conditions defining the elements of P(C),


cf. (l.l), can be weakened in case C Is
algebraic over K. This proposition will play
an important role later. (1.20)
PROPOSITION. Let char K = 0 and G be an
irreducible algebraic K-group. Assume p c
DerkK(C) is such that I) pKcK 2) Vp =(pel +
Ip)v;K(G)*K(GxG) 3) Sp -- pS: K(G)K(G). Then p P(G). Proof. Start with the
remark that if K ilk Is a Kip]-field extension
(0.10) then p extends to some derivation Pl e
DerkKl(G) satisfying the corresponding
properties 1)-3) from the statement ot the
proposition. Since the projection G oKi<i +
G is faithfully flat it Is sufficient by (0.14),
5) to check that Pl z P(GKKi)' So we may
assume that K equipped with the derivation
a(p)cDerkK is Kip]-algebraically closed. By

(0. it0 again, the locus (p)m = (x z G; p(60G,


x) t O/G, x) is a divisor on C. Put U = G\
(P)o; it is a I<[p]-variety.. 17 The set of
K[p]-K.-points UK[p](K) of U is dense in
U(K). Take g UK[p](K) and consider the
corresponding left translation :G gxl p Lg t
GxG sG Its restriction (g x l)'l(p'l(U)n (U x
U)) - g x I i.i(u) n (UxU) " is a K[p]morphism (where note that the source
scheme is an open subset of U which is
nonempty because U n 8'!U). Consequently
the induced morphism Lg=K(C)-K(G)
commutes with p: K(C) - K(C) this showing
that (p). Js Lg-invariant for all g UK[p]
(K). By density this happens for all g c G(K)
so (p)m must be empty, hence pC c (C'
FinalJy, take any g GK[p](K); since Sp -pS, the morphism -1 : SPOcK gxg I GXG

G is a K[p]-morphism and we are done. In


what follows, by an lsogeny we mean a
surjective morphism G'- G of irreducible
algebraic groups having finite kernel. (1.21)
PROPOSITION. Let f .. G'- G be a separable
isogeny. There exists a canonical injectlye
"lifting" map of Lie K/k-algebras f* :p(c)P(G'). Moreover, if C and C' are
commutative and char K = 0 then f* is an
isomorphism. Proof. Let p, S', be the
structure maps of G'. Since f is tale, any p
DerkG lifts uniquely to some p' I DerkG.
Now If p I P(G), IJ'P'-(P' l + lp')l' is a IJK-derivation K(G')- K(G' x G ) vanishing on
K(G)I by separability this derivation must
vanish. Similarfly we get p'S' = S' and
p'' = 'p'. Of course we put f*(p) = p' which
is well defined by the preceedIng remarks

and clearly is an injective Lie K/k-algebra


map from P(C) to P(C'). Now assume char K
= 0 and CG commutative. Then there is an
isogeny g = C - G' such that gf .. C,- C+ C is
the multlpJicetion by some integer N _)l
which we call XN' We get injective maps .
g* f* X N: p(c,) , p(c) * p(c,) is the
identity (this will close our proof). This
follows by noting that N is a We claim that N
map of D(G')-varietlas hence for any p' c
P(G') we have a commutatJve diagram (C') ,
K(C') K(G')
18 showing that the canonical lifting of p' via
l N is p itself. Our proposition is proved.
(1.22) PROPOSmON. Let G be an
Irreducible commutative algebraic D-group.
Then the following hold: 1) Any torsion

point In G(K) of order prime to charK is a


D-point (in other words if N is an integer
prime to char K, N-TOrs(G(K))c GD(K)). 2)
Any torus and any abellan variety contained
in G is an algebraic D-subgroup of G. 3) Any
non,linear irreducible algebraic subgroup of
G contains a non-trivial irreducible
algebraic D-subgroup of G. Proof. l)Let g
lN-Tors(G(K)). Then multiplication by N
(call It IN:G-G) Is a separable isogeny and
),1(0) is a D-subscheme of G, union of
reduced points. By (0.14), each of these
points is a D-point, in particular so Is g. 2)
Let H be either a torus or an abellan variety
contained in G. Then the torsion points of H
whose order is prime to char K is dense in
HI so on any arline open set of G the defining
ideal 3 of H in G is an Intersection of D-

ideals so 3 Itself is a D-ideal and we are


done. 3) Any non-linear Irreducible
algebraic subgroup H of G has Infinitely
many torsion points of order a power o! a
prime , where I does not divide charK. Let
be the Zarlski closure [n H of the group of
thee torsion points. Then M is an algebraic
D-subgroup of H and so will be its
connected component. (1.23) Let's make
some general remarks on the link between
P(G) and the automorphlsm functor of G (a
deeper discussion will be done In Chapter
4). For any K-group scheme G we have the
automorphlsm functor: Aut G: {K-schemes}(groups) defined by (Aut GXS)=
Aut$.gr.sch(G x S) (in particular (Aut G)
(K)= AutG). This functor need not be
representable even If char K = 0 and if G is

an algebraic group (cf. [BS], e.g. if G = G a


x Gin). It is representable however if G is
algebraic reductive (and char K _) 0) cf.
[GD]. Recall from [GD] that for any
contravariant functor A .' {K-scheme} +
{groups} one can define the "Lie algebra"
L(_) ,= Ker (A(n): _(SPOc Kid) * _(Spoc
K)) where K[ ] = K . K e 2 = 0 and the
map II .. K[ ] - K is defined by - 01
L(_A) is a group but not a genuine Lie Kalgebra 'In general. In our specific case
howevery L(Au_._!tC) is easily seen to
identify with P(C/K)= Ker( a :P(G)-DarkK),
in particular It is a genuine Lie K-algebra;
the identification is given of COurse by the
map p +ld + ep (p P(C/K)). There are
remarkable cases when the restriction of ^ut
G to (reduced K-schemes} is representable

by a locally algebraic K groupy call it in this


case ^utG (there is no danger of confusion
with the other meanlng of Aut C which is
(^u____!t G)(K)!). This is the case If char K
= 0 19 and G is linear algebraic [BS] (or
even char K = 0 and G is algebraic, nonnecessarily linear, el. (IV.2) below). If the
above representability properties holds then
there is an induced Lie K-algebra map I = I
G: L(Aut G) L(Aut G) which is not an
lsomorphism in general and which will be
studied in (IV.2). (1.2) For any K-group
scheme the group (^utGXK)=AutG acts on
DerkG by the formula {o,p)-o'lp for any
oeAutGy pe DerkG (as usual we still denote
by the map (-G ' # (C induced by ). Under
this action P(C) and P(C/K ) are globally
invariant. Let us call ? .' P(C)-'

Derk(P(G/K),P(G/K)) , p - V P the K-linear


map defined by pV = [p,v] (bracket taken in
P(G)) for pe P(C), v P(C/K) and define the
K-linear map ,!.V: P(G)- zl(AutG, P(G/K)),
p P by the formida II.V o=o'lpo. p for pz
P(G), oz AutG. One easily checks that
(V,J.V) above P define an integrahie
logarithmic P(G)-connection on (Aut G,
P(G/K)). (1.25) !n way of our study of P(G)
for commutative G we shall be lead to
consider one more example of logarithmic
DerkK-connection (9lt9) whose "additive
part" is the Caussr -Manin connection [KO]
on the first de Rham cohereology space of
abeltan varieties while the "multlpllcatlve
part" lt will be defined and used in Chapter
3. (1.26) We close this section by noting that
if char K = 0 and If K o is a field of

definition for a K-group scheme C (i.e. if we


have a K-group scheme lsomorphism o: G a
G o KoK , G o a Ko-grou p scheme) then
there is a Lie K/k-algebra map Der(K/Ko)*
P(G), p p* where p* =o(1 p)o '1 Is the
trivial lifting of p to G (it depends on o !).
We shall often identify Der(K/K^) with a Lie
K/k-subalgebra of P(C) (after having fixed o
I). It is easy to check that P(G) K is
contained in any algebraically closed field of
definitlon for G containing k. If K p(G) is a
field of definition for C (which is not always
the casey cf. Chapter 3 below l) then we
have a split exact sequence of Lie K/kalgebras 0-' P(G/K) P(G) a Der(K/Kp(G)).,
0 In particular P(G) is the semidirect product
of Ker a and Im a . EXAMPLES Let's
consider once again the Examples 1, 2, 3

after (1.9), where In Example 3 we assume


22 which we are briefly investigating in
what follows. (2.2.) LEMMA. For any field
extension K ilk the map (H)I(K 1) ' H(K 1
) is surjective. Proof. Any field extension K
1 of K admits a D-field structure and we are
immediately done by adjunction. (2.3)
pROPOSITION. Assume G is an irreducible
algebraic D-group. Then: l) There exists a
naturally associated D-cocycie G!. hA :
(L(G)m) = (L(G)m) m such that the
morphism G- G ! deduced from adjunction
induces an isomorphlsm of D-group schemes
between G and the reduced kernel (ker fi
)red of , A. (Here L(G)= L(G !) is of course
identified with its associated algebraic
vector group an. the action of G I on

(L(G)m) is induced by functorality from


the adjolnt action of G on each factor of
L(G)m). 2) There Is a bijection
PHSD(G/I()= L(G)int/G(K) where L(G) lnt - ? piVj - V pjv i . vi,v j] = 0 for all i,j (cf.
notations in (1.10)) and G(K) acts on L() nt
by the "Loewy-typo" [C 1 ] formula: (g,(vl)i)
+ (L;lviLg +V pig)i (g= G(K)) (where - p.g =
L;IpiL 8 - Pi' cf. (1.10)). In particular, if G is
commutative PHSD(G/K) has a natural
strLture of abelfan group. 3) There is an
exact sequence of pointed sets (which In
case G is commutatlve is an exact sequence
of abolish groups): 1 - GD(K)+ G(N) - L(G)
int ' PHSD(G/K) ' 1 In particular, if K is Dalgebraically closed, , A above is surjective.
REMARK. The last surjectivlty assertion is
a generalisation of Kolchln's surjectivhy

theorem for the logarithmic derivative (cf.


K i] PP. 20-a21, or [NW it reduces to it in
case G is split (cf. (3.11) below for the
definition of "split") but our proof here is
different from those in [KI] , ProoL 1) To
define . a it is suffieient by (0.9) to define for
any reduced D-algebra A, coeyeles , A A: (G
I )D(A ) = G!(A I). ((L(G)m) o )D(A ) =
L(G)m A behaving functorially in A. We
shall define. A A on each component to be
given by the map (*) g'" L;IpiLg- Pi (g G!
(A!)) where Pi denotes here the derivation of
{GA corresponding to Pi P while Lg:
CGSA* (GSA is deduced from the left
translation by g. Then L;lpiLg - pie L(G)A
because DerA(CG A) 1 = HmG A(G
A/A' G A ) (DKG) K A and because one
can transse mutatis mutandis he mpuation

made in (1.10) (note that for = K our map (a)


Is ecisely t map gYP from (1.10)). To ove G
a (Kerfi)re d it is sufficient to chk that we
have for any reduced D-algebra A an exa
sequence of pointed sets 1 + HomD.h(Spec
A,G)+ HOmK_sch(SP AG ) g 6A _,. L(G) m
A This can done by adapting mutatis
mutandls the last part of the proo of (1.1 I)
(which deals with the case A =KI). This clos
the oo o 1). To prove 2) we may identiy (in a
n-canonlcal way I) the underlying varieties
of all D-priipal homogenous spaces or G
with G itsel (since K is algebraically closed
!) so to give a D-principal mogenous space
or G i$ ulvalent to giving commuting
derivations dl,... ,dm DerkG tisying the
ormula wd i=(d il +lPi)p (where d i 1 + 1 $
Pi is of course t derivation on G x G induced

by dj on the lrst factor and by Pi on t


second). The ove ndition is ulvalent to w(d i
- pi) = ((d I - pl) l)p hence with d I - p[ e
L(G) (once again we still denote by Pi t
derivatio n in OerkG corrndlng to pi ). Now
two m-uples (all,... ,din), (d,... ,fire) give
isomorphic D-principal homogenous spas i
and only i thee is isomorphism o K-varietl :
G+ G making the ollowing diagram
commutative GxG GxG - +G Such that d
='ldl for all 1. The above diagram shows that
must be the left translation L g (g = (e)
G(K)) and so the latter relation becomes
L;l(di ' Pi)Lg * 'V Pig -- el ' Pi which proves
2). Assertion 3) follows directly from 1) and
2).
24 . Local finitness and s91ittin Here we

come to one of our main concepts (and tools)


namely to "local flnitness" and its relation
with "splitting" and "fields of definition";
this is part of our ideology in [Bi]. We start
with some definitions (where we assume K
algebraically closed as usual and of arbitrary
characteristic while P is any Lie K/k-algebra
and D = K[P]). (3.1) A D-module V is called
finite dimensional if dimKV ( m. A Dmodule is called locally finite If it Is the
union (equivalently the sum) of its finite
dimensional D-submodules. lote that if V and
V/ are locally finite D-modules then V V/
and V V/ are locally finite but HomK(V,W)
won't be so in general and in fact not even V
= HOmK(VsK) will be locally finite in
general A D-module V is split [NW][IB 1] if
V has a K-basis contained in V D,

equivalently if the natural K-linear map K


KDV D * V (which is always injectlye) is
also surjectlve. Of course if char K ) 0 then
V is split if and only if P acts trivially on V
so the concept of "splitting" will be
interesting only in characteristic zero. Any
D-submodule of a split D-module is easily
seen to be split. Moreover, any split DModule is locally finite. Conversely, we
have the following result essentially proved
in [BI], p. S5 (confer also with [T] for a
generaligation and also to (.2) [EMMA.
Assume char K = 0 and V is locally finite Dmodule having at most a countable dimension
over K. Then there exists a D-field extension
K IlK such that KI D -- K D and such that K
1 K V is a split Dr-module (D 1 -- K 1 K
D as In (0.10)). Moreover if we assume K is

D-algebraically closed we may take K 1 -K. Proof. For the sake o! completness we
sketch the argument. It is clear that we may
assume N = dimKV ( m. Then the D-module
structure of V induces a D-variety structure
on GL N by just fixing a basis vt, .... v N in
V.' if pv i = Zaij(P)e j with ajj(p) eK and if
GL N =Spec K[Xjj] d , d = der(Nil) then we
let PXij = kaik(P)Xkj for all l,j,p Now let 8
be a maximal element of (GLN) D and put K
1 = K(g) (residue field at p it Is a D-field
extension of K). By (0.1t)s we have Ki D -K D (because K D is algebraically closed).
Viewing g as Kl-point of CL N, say g -- (glj)
gij e KI it is easy to see that the elements fl''"
fN D l defined by i -- ?.gijfj form a K lbasis of K l V belonging to (K 1. V) (3.3)
A D-algebra A (respectively a Lie D-algebra

L, or a Hopf D-algebra H) is called 25


locally finite if It Is so as a D-modulel A
(respectively L,H) is called split [NW] If
there exists a D-algebra isomorphism A m K
KDAO where A is a KD-algebra and K
KDAo is viewed as a D-algebra with P
acting via i) on K 1 and trlvlally'on 1 A o
(similar definition for I.,H). Here is a
remarkable (trivial) remark (3.) LEMMA.
Let A be a D-algebra (respectively L a Lie
D-algebra of H a Hopf D-algebra). Then A
(respectively L,H) is split as a D-algebra
(respectively L is split as a Lie D-algebra or
H is split as a Hopf D-algebra) if and only If
A (respectively L, H) is split as a D-module.
Proof. Let's a give the proof for H for
instance. Assume H is split as a D-module.
Then it is easy to check that H D is a KD-

subalgebra of H invariant under antiperle


and mapped by H D comultiplication into
HD KD = (H K H) D so that H D has a
naturally induced structure of HD. Hopf KDalgebra. Moreover the natural map KKD H
is clearly a Hopf D-algebra isomorphlsm,
hence H is split as a Hopf D-algebra. The
converse is obvious. (3.5) Let X be a Dscheme; an open set U of X will be called Dinvariant if there is a closed D-subscheme
whose support is X\ U..Now X will be
called locally finite if every point in X has
an affine D-invariant open neighbourhood
whose coordinate algebra is 16rally finite.
Note that if X is locally finite and quasicompact then the D-algebra ((X) is locally
finite; indeed if (Ui) l is a finite covering
with afflne open sets such that C9(U i) is

locally finite then t(X) appears as a Dsubmodule of O((U i) which is locally finite
hence it is locally finite. In particular this
shows that if X is an afflne D-scheme then X
Js locally finite if and only if C/(X) is
locally finite. (3.6) Recall lB 1] that a Dscheme X is called split if there is an
Jsomorphism of D-schemas X K DXo where
X o is a KD-scheme and K DXo is viewed
as a D-scheme with P acting on K 1 via i)
and acting.trivially on . Recll some basic
facts about split D-schemes o 0) Any split Dscheme is locally finite. l)If X and Yo are
KD-scheme and If X=K DXo , Y=K DYe are
viewed with their natural structures of split
D-schemes then any D-morphJsm -'X Y has
the form = K eo for some morphism o Xo*
Yo' In particular XD(K)identifies with

Xo(KD). 2) Assume the D-scheme X is


covered by D-invariant open subsets U i
each of which is split; then X itself is split.
3) Any closed D-subscheme of a split Dscheme is split. ) Any D-invariant open Dsubscheme of a split D-scheme Is split. The
proof of the above assertions can be done as
follows.. first prove 1) for affine schemes
using the fact that D-algebra maps must carry
P-constants into P-constants. Then prove 3)
in the arline case using the fact that any Dsubmodule of a split D-module is split,
26 Now 4) In the arline case follows from 1)
and 3) in the affine case. Next prove 2) by
first noting that the arline case of 4) implies
that for any set of indices 3={Ji,...,jN}, Ug =
Ujl n ... n UjN is afflne and split (use

separation of X) hence by 1) in the affine


case again we may "glue together" the
splittings of the Ui's. Finally one uses 2) to
reduce 1) 3) 4) to the afine case. (3.7)
PROPOSITION. Assume char K = 0 and X
is a locally finite D-variety. Then: 1) Any Dinvariant afflne open set of X is locally
finite. D K D such that KIKX is a split 2)
There exists a D-field extension Ki/K with K
1 = D l-variety, where D 1 = K 1 K D.
Moreover, If K is D-algebraically closed
then X itself is split. Proof. First prove 2).
Cover X by arline D-invariant open subsets
U. such that (U i) Is locally finite. By (3.2)
we may find an exter=ion of D-fields Ki/K
with K? = K D (in case K is D-alebralcally
closed we may take K 1 = K) such that each
K 1 KUi is split (one simply applies (3.2) to

V = V I where V 1 is a finite dimensional


D-submndule of ((U l) generating ((U i) as a
K-algebra). Clearly KiU i are Dl-lnvarlant
open subsets of Klex and we conclude by
assertion 2) in (3.&) that K 1 X is split. To
prove 1) let Ki/K be as in 2) and let Uc X be
an afflne open D-invariant subset. Then K 1
U is Dl-invarlant. By 4) in (3.6) K 1 U is
split, in particular ((K 1 U) IS locally finite
as a Dl-mndule. It follows that O(U) is
locally finite as a D-module as well. Our
proposition is proved. (3.8) REMARK. Let
X be a D-variety, char K = 0. It may happen
that there exists a D-field extension KI/K
with K? = K D such that K 1 K X is a split
Dl-variety (Di=KiKD) but X is not locally
finite: take for lntance X to be an elliptic
curve over K viewed as a Kip]-variety

where p acts on X via a non-zero global


vector field v on X. In this case X contairz
no non-empty Kip]-Invariant open set
(because v is nowhere vanishing) but X
splits over a "strongly normal" extension of
K by (.9) Let G be a K-group scheme. We
denote by P(G,fln) the set of all p P(G)
such that G is.a locally finite Kip]-scheme (if
G is arline this is equivalent to saying that
C(G) is a locally finite K[p]-mndule, cf.
(3.5)). We also put P(G/K,fln) = P(G/K) n
P(G,fln) We will prove later (cf. (IV. .)) that
if G is algebraic (Irreducible) and char K =
0 then P(G,fln) is a Lie K/k-subalgebra of
P(G). But note that if char K = q > 0 then
P(G,fln) (which coincides with P(G/K,fin))
need not be a linear subspace of P(G) even If
G is algebraic. Let's give such an example In

what follows. First it Is easy to check that for


q = 2,3,5 the following formula holds in the
polynomial ring A = (.) (( - [q) + [ - IA)q-l( )
= (Presumably this formula holds for any
prime ql). Assuming formula (.) holds for a
given prime q (e.g. assuming qe {235)) let G
= C a x G m = Spec[Xx'l] 1 = 1 + 1 1X = XX
and define p = DerKC by the ormula p=(-q +
5in -q and 1ong to Xa(G) and 1ong to L(G) It
ollows by (1.9) that p P(G). Now ormula
(,) lmpli that p = and pq-l) = p-l) On the
other hand clearly p = or all 2. Consequently
p P(Gdin). On the other hand consir the
derivation DerKG = q. Clearly X = 0, -1)
= 0, = (-)i+Z E P(G,fin). But + p P(G,fin)
because (3.10) Here IS an example of an afi
(non-gebralc) K-group scheme G char K = 0
such that P(G/K,fin) IS t a linear subsco o

P(G). Take G =ScA, A =Ki i =i ] + 1 l' i = l'i


= 0 (so G an "infinite vector group scheme").
Define= P = il' 2i+1i +-') z = 2i+1 = + 2i- 1
i) i1 - It easy o see using (1.8) that p,
P(G/K,fin); but p P(G/K,in) since or y (p +
)1[ 1) IS a linear orm in whichi+ 1 curs with
coefficient 1. EXAMPLES It is appropriate
m examine here the Examples 123 given ter
(1.9) from the viewpoint of local finitehess.
Consider first G = G n viewed as an
algebraic D-groups (D = KIP], P = Kp) via
the a derivation = * a p +r. aij[i.-. j as Jn
(1.26). Then C is locally finite because for
each d _> I the K-space of all polynomials in
K[ 1' "' n ] of degree _< d Js - stable.
Consider now C =C m x C a viewed as an
algebraic D-group via the derivation = p* +
[XX. Then G is not locally finite because the

D-submodule of K[X,X-I]' generated byx is


infinite dimensional. Finally, consider G = A
x G a viewed as an algebraic D-group via
the derivation = p* +[ Yx ' Then C is not
locally finit for if Jt was then by (3.7) it
would split over a D-field extension of K,
this forcing the subfield K(A)=K(x,y) of K(A
x G a) = K(x,y ) to
28 be preserved by . This is however not the
case since x = y K(A). (311) A D-group
scheme G will be called split if there is a Dgroup scheme isomorphism G = KKDG o
where G o is a KD-group scheme and K
KDGo is viewed as a D-group scheme with
P acting on K & I via a and acting trivially
on G ' By (3.6), l) a D-group scheme G is o
spli:t if and only if it is so as a D-scheme. So

Proposition (3.7) holds if we replace "Dvariety" by ,irreducible algebraic D-group".


For the reader familiar with [K l] let's also
make the remark that if P has a finite
commuting basis Pl' '" 'Pm and if G is a split
irreducible algebraic D-group (G = K
KDGo ) and V is a D-principal hornogenous
space for G then upon viewing the fields K
and K < V ) a A -fields with derivations
(pl),... ,(Pm ) we have that the extension K <
V )/K is G o - primitive in Kolchin's sense
[K l] (cf. also [B i]p. 26). We close by a
general remark on the link between
P(G/K,fin) and the map G in (1.23) (for an
extension of this result confer with (IV.2)
below). (3.12) PROPOSITION. Assume
char K -- 0 and G is an arline algebraic Kgroup. Then the map X: L(Aut G)+ (Au.__t

G)= P(G/K) is injective and its image equals


P(G/K,fln). Proof. Start with a preparation.
Assume W is a finite dimensional vector
subspace o (G) generating (G) as a Kalgebra and let Aut(G,'): {aJfine K-schemes}
+ {groups} be the subunctor of Aut G
defined by S = SpecB {f Aut(G x S/S) f*
=((G)B (C)B preserves WB} We claim
that Aut(G,W) is representablel note that the
affine group scheme Aut(G,W) representing
it will be reduced by [Sw] p. 280. To prove
our claim let W I be the Intersection of all
sub-K-coalgebras of (G) containing Wo= W
+ SW by [Sw] W 1 is a finite dimensional
coalgebra. Now define inductively the
Increasing sequence of subspaces W i of (G)
be the formula Wi+ 1 = I(Wi W 1) for i _) I
and define functors A_o,AIA_2,... A i:

{affine K-schemes} {groups} as iollows.


We let A_o(Spec B) be the group of all Blinear automorphisms o o of W o B such that
oSB=SBo where S B = SI B. For i_ 1, let
Al(SpecB) be the group of all B-linear
automorphisms i of W i B such that Oi(gi_ 1
B) = Wi_ 1 29 Vi I Wi-I BeAi'l(Spec B) and
alp B -- pB(ai.l oi_ 1) where PB = p lB We
have canonical restriction maps R_l * R_i ' 1
for i _> 1. Now clearly all A_.i's are
representable by arline algebraic K-groups
A i hence we have a projectire system .,
AiAi_ 1 *...Ai*A o One checks that
Au.t(G',V/) -- lim_A i. Consequently
Au_t(G,V/) is represented by Spuc(lira (Ai))
and our claim is proved. Let's prove that lm - P(G/K,fin). The inclusion : "is clear.
Conversely if p g P(G/K,fin) we may choose

V/ above such that pV/c W. Then id gp


Au(G,W) (Spec K[, hence we get a
morphism f: 5pec K]* Aut(G,W) such that id.
gp = faG,W where G,W Is the univerl
Aut(G,W)-automphism of G x Aut(G,W).
Now since Aut(G,W) is reduced and since
Au G restricted to (reduced K-schemes} is
reesentable by a 1ocly gebraic K-group B5]
AutG there exists a morphism h:Aut(G,) +
AG such at -h* where G Is the . G,W - G
univers Aut G-automorphsm of G x Aut G.
Consequently id p = (hf)aG hence p lm
. Finally, let's prove that is injtive. Let
AutG =Spec R (by [BS] AutG is fi I) we
may choose a finite dimensional subspace W
of (G) generating (G) as a K-algea such that
(W R) W R (where : (G) R * (G) R is
induced by G ). Exaly ave there exists a

morphism h:Aut(GW) AetG such that h G


:GW' There is also a nat morphism c:
AutG* Aut(G,W) defined at the level of Sints by (AutGNS) Aet(G,W)(S), s where
s* G Consider the affl group scheme A =
Aut(G,W) x Aut G AutG note that the
projtion Pl: A + Aut(G,W) is a closed
embedding. Now the map P c A AutOG
Aut(G,W) equals the map Pl A * Aut(C,W)
for if (f,f,) A(S) is an S-point of A we have
hf = if' (where i .' AutG ' Aut G is the
inclusion) so the image of (f,f') via (cP2)(5)
is a map s Aut(G,W)(S) such that s * = f'i* G
= G,W = f*h* G = fG,V/; consequently s = f
by universality of ^ut(G,W). We get that P2 is
a closed embedding so A is an algebraic
arline group. Since the map P2: A * AutG
induces a bijectlon at the level of K-points

this map is an isomorphlsm. Consequently c


is a closed embedding hence AutG is a
subfunctor of Aut(G,V/) hence of Aunt G and
injectivity of follows.
CHAPTER 2. AFFINE D-GROUP
SCHEMES Throughout this chapter K is
algebraically closed P is any Lie K/kalgebra and D = KIP]. I. The analytic method
Our first main result is: (1.1) THEOREM.
Assume char K = 0 and let G be an
irreducible affine algebraic K-group. Then K
P(C) is a field of definition for G (hence it
coincides with KG). (1.2) The above
statement fails for non-linear G, cf. Chapter
3. Our proof of (1.1) will be analogue to that
of Theorem (1.1) in Chapter 2 of [B 1] in the
sense that we are going to use "birational

quotients", a "Koalaira-Spencer map*' and


an analytic ingredient. In [B l] the analytic
ingredient was the versa] deformation of a
compact complex space. Here the analytic
ingredient is a combination of Theorems
(1.3) and (1.#) below (due to Harem and
Hochschild- -Mostow respectively). To state
the first theorem let's fix some notations.
Assume [: + is an analytic family of
connected complex Lie groups, i.e. a map of
analytic C-manifolds, having connected
fibres, such that one is given analytic - maps
satisfying the usual axioms of multiplication,
inverse and unit. Assume moreover that a) is
simply connected and Vl,... ,v m are
commuting vector fields on giving at each
point a basis of the tangent space and b) v 1,
... v m can be lifted to commuting vector

fields wl,... ,wm on such that p,5 agree


with w 1, .. w m in the sense that for each
w = w i we have: 1 ) (T(g t'g2 )pXw(g
l)'w(g2)) = w(p(g l'g2 )) 2) (TgSXw(g)) =
w(Sg) 3) (TxEXv(x)) = w(E(x)) Then we
have for any (gl,g2) x, for any g C ,' for
any x, (1.3) THEOREM. (Harem, [Ha]).
Under the assumptions above there exists an
analytic -lsomorphism *, o x * (where o is
some fibre of T[) which above each point of
l, a 31 group isomorphism and such that upon
letting v. be the trivial lifting of v i from to o
x. we 1 have (T)(v?) = w i for all i. (l.t)
THEOREM. (Hochschlld-Mostow [HMi]).
Let Gl, G 2 be two irreducible affine an and
an algebraic C-groups. If G i G 2 are
lsomorphic (as analytic Lie groups) then G l
and G 2 are isomorphic (as algebraic

groups). (1.) REMARK. The above


statement fails for non-linear groups (cf.
[HMi[Se. Theorem (l.t) is a consequence of
the theory developed in [HM 1 ] and no idea
of proof will be rlndicated here. We shall
however include the proof (due to Harem) of
(1.3) since it is completely elementary and
fairly elegant: (1.6) Proof of (1.3) (Harem).
By Frobenius, for any x o and any 8o l[ '
l(xo) = o there gg of go in o and an analytic
map exists a neighbourbood of x o in , a
neighbourhood lg o *: go x-- ' over such
that '4) takes v (= trivial lifting of vi from . ,o
*'go x 7go) into go ' w i. A triple (go' ?Jgo'
Lb) as above will be called a "local
solution" at go' It is sufficient to show that
for. a given Xo, the various ?gg *s appearing
in the local solutions at various points 80 o o

can be chosen to contain a fixed open


neighbourhood of x o. Let o -- (x o) and
consider the set r. of aug, o such that there
exists a local solution,?gg,/g, t),,g with c .
One easily checks that r. is an open subgroup
of o (local solutions can be "multiplied" and
"inverted" using p and 5) hence r. = o by
connectivity, which proves the theorem. Next
we need some facts about isomorphism of
Lie algebras. First "recall" the following
trivial representability result: (1.7)
LEMMA, Let R be a ring and L L two Lie Ralgebras which are free and finitaly
generated as R-modules. Then the functor
[SOL,L, from {R-algebras} to {sets} defined
by ISOL,L,() = { - Lie algebra lsomorphisms
R L = i R TM } is representable by a finitaly
generated !-algebra (call it lSOL,L). Exactly

as in [B I] pp. 3-36 the above Lamina


implies the following= (1.) LEMMA. Let C
be an algebraically closed field, S an afflne
C-variety and L a Lie ((S)-algebra which is
free and finitaly generated as an C(S)module. Then there is a construc- tibia subset
Zc S x S such that for any Sl,S 2 c S(C) we
have (Sl,S2) Z(C) if and only if the'Lie Calgebras L (s)C(Sl) and L )((s)C(s2) are
isomorphic. Next we have= (1.9) LEMMA.
Assume C, S, L are as in (1.8), let K be an
algebraically closed extension of C(S) and
assume KL( = smallest algebraically closed
field o! definition of L ((5)K between C and
K, cf. (I. 1.27)) equals the algebraic closure
of C($) in K. Then there exists an open
subset

32 S O c S such that for any s o So(C) the


set { s S(C)t L C(s) L ) C(So)} is finite.
Proof. By (l.S) and [Bi], (1.13), p, 36 there
exists an affine open set 51 c 5 and a
dominant morphism of afflne C-varietlas b .'
Sl+ M such that for any s I Sl(C) we have
'lb(Sl) ={s Si(C)I LC(s) LeC(Sl)} if dim k =
dims I we are done. Assume dim M ( dims 1.
Then we use an argument similar to iV] p.
576. Choose a closed subvariety Nc S 1 with
dJmN = dim 1, let L N be the pull-back of L
on N, let L' be the pull-back of L N on the
affine scheme 31 = S 1 x MN and let L" be
the pull-back of L on 31 Then for any Cpoint X$l(C) one checks that
L'eC(x)=L"C(x). By representability of
ISOL, L" (1.7) there is a generically finite
dominant morphism of finite type of arline

schems 21 with Y integral such that the pullbacks of. L' and L" on Y are Y-isomorphic.
Since Y+ S 1 is generically finite, one can
embed C(Y) over C(S 1) = C(S) into K and
we get that C(N) is a field of definition for L
K between C and K contradicting our
hypothesis. The lamina is proved. Next we
need a Kedaira-Spencer map for irreducible
arline algebraic groups G over nonnecessarily algebraically closed fields F of
characteristic zero containing k. Consider the
following complex (where A -- {(G))'. 0+
DerF(A) ' , DerF(A,AFA) 2 +
DerF(A,AFAFA ) where {) l(p) = pp(pl + I p)p, pc DerF(A,A) {) 2(d) = (d
l)p - (1 ) d)p + (p l)d - (1 p)d, de DerF(A,A
F A) and define Hi(C/F) for i = 1,2 as the
i-th horn ology space of this complex (one

can identify Hi((;/F) wlth the corresponding


Hochschild cohomology spaces of the
adjoint representation of G, cf. [D(;] p. 192,
but we won't need this fact). Clearly Hi((;
FFi/F 1)= Hi(C/F! )FFI for any field
extension Fi/F. Moreover we can define
P(C) and P(C/F) exactly as in (I. l.l) in case
our ground field is not algebraically closed.
(1.10) LEMMA. There Is an exact sequence
0- P(G/F)+ P(G) DerkF-H2(G/F) wherep
is compatible with field extensions Fi/F.
Moreover, we have an identification P(G/F)
= HI(G/F) S (where HI(G/F) S Is the fixed
part of HI(G/F) under the involution v SvS ot
DerF(A,A) which is easily seen to preserve
HI(G/F)). Proof. The assertion about
HI(G/F) S is clear. Let's define p. Since G/F
is smooth and arline, any derivation p DerkF

can be lifted to a k-derivation o! A = ((G).


33 Then one checks immediately that p - (
! + l))p Ker(a 2)c DerF(A,A FA) and the
class of this derivation in H2(G/F) does not
depend on the choice of the lifting call this
class p(p). e must check that Keri0 =
im(P(G)+ DerkF) The Inclusion ':" Is clear.
Converse]y, if 19(p) = 0 then there is a
lifting of p, e DerkA such that p = ( I + I ))1.
Then one immediately checks that .(ss) =
[(ss) l + i (ss)]p Puttlng = (I/2X + SS)
we see that ) lifts p and belongs to P(G) so
our lemma is proved. (1.11) COROLLARY.
If Fi/F is an exension of algebraically closed
fields and C is an irreducible arline
algebraic F-group them P(C FFi/Fi )a
P(G/F) )FF ] and P(G O FFl/Fl,fin)
P(G/F,fin)FF 1 . Proof. The first

isomorphism follows from (1.10). The


second follows from (1.3.12). (1.12)
THEOREM. Assume char K = 0 and C is an
irreducible aifine algebraic K-group. Let C
be an intermediate field between k and K and
let K be the smallest algebraically closed
field of definition for G between C and K (G
COKoK for some Ko-grou p Co). Then the
map I o: DercKo + H2(Go/Ko ) is injectlye.
REMARK, The field C in the above
statement has only an auxiliary role in the
proof. The interesting case of course is that
when C = k (then K o = K G. cf. (I. 1.27)).
We shall give first the proof of (1.12) when
C is the field C of complex numbers. By (I.
1.27) K equals the smallest algebraically
closed field of definition for L(U) (U the
unipotent radical of G) between C and K.

There exist group schemes C.s S and O.s S


(S an arline (:-variety, 0 a closed subgroup
scheme of C) such that K o is the algebraic
closure of K 1 = C(S), C )sKo = Go, 0 SKo
= U (Uounlpotent radical of Go) and the
fibres of I.s S are the unipotent radicals of
the fibres of + S. We may assume the relative
Lie algebra L(O/S) is a free I(S)-module.
Assume Po is not injectire. Then Pl
"DercKI+ H2(CI/KI ) is not injectire (where
C 1 = C SKi ). By (l.10) there exists p
DercK 1 , p 0 which lifts to a derivation
P(Ci). Now both p and can be viewed as
rational vector fields on the C-varieties S
and C respectively. Shrinking S we may
assume both p and are regular everywhere.
Now by (1.9) there exists a Zarisky open set
Soc S such that for any Soe So(C) the set T. s

={s So(C);L(Us) L(U s )} o o


is finite where U s J x sC(s). Let be an
analytic disk in 5 an which is an integral = O
submanifold for p and let := .an x '. Then is
an integral submanifold of an for. By san
(1.3) all fibers o -. are pairwise isomorphic
as complex Lie Groups. By (l.q) the fibres o!
+ S above the points In are pairwise
isomorphJc as algebraic groupsl in
particular all Lie algebras L(U s) with se:
are pairwise isomorphic this contradicting
the finitness of r. So for s o S(C). The
theorem (l.12) is proved for C = C. (1.13)
Proof of Theorem (1.1). It is sufficient to
prove that for any Lie K/k-subalgebra P of
P(G), K p Is a field of definition of G. Case
I. K P is uncountable. Then we can assume

Cc K p. Let K o be the smallest algebraically


closed field of definition for G betveen C
and K, G = Go) K K. We may conclude o by
Inspecting the diagram with exact rows and
colomns (cf. (l. i 2), case C = C and (i. 10))
0 Der K K 0 o P(G) d Derc K P . H2(G/K)
0 (DercKo)DKo K , H2(Go/Ko))K K PoK ,
o o o that line = ImP/hence that K K p, i.e.
that G is defined over K p. Case 2. K p is
countable. Then take an embedding KPc C
and conclude exactly as in [B 1] p. 3
(instea.d of B1], Lemma (1.19) use the fact
which we already know from [B] that the set
of algebraically closed fields of definition
for a linear algebraic group has a minimum
element, el. also (I. 1.27)). (1.1) REMARK.
Using Theorem (l.l) one can immediately
prove Theorem (i.12) for arbitrary C. Now

(1.1) and (I. 1.22) imply: (1.1)


COROLLARY. Assume char K = 0 and let G
be an irreducible affine .agebraic K-group.
Then we have an exact split sequence of Lie
K/k-algebras= 0+ P(G/K)+ P(G) a _
Der(K/KG). 0 EXAMPLE Assume P is a Lie
K/k-algebra, D = KiP] and K D = k K. Then
the "sufficiently general" unipotent algebraic
K-group G of sufficiently big dimension
doesn't admit any structure o algebraic Dgroup. Indeed it is known that such a G is not
defined over k. We shall need in what
foliow's a variation on Hamm's result (I.3)
(1.16) LEMMA. Let be a connected Lie
group and v an analytic vector field on such
that the multiplication x and the inverse - are
equlvarJant (with respect to the vector field
(v,v) on } x and v on ). Then there is a l-

parameter group of analytic group


automorphisms C x + whose derivative at the
origin is v. Proof. Use Hamm's open
subgroup argument 0.3) once again to show
that therb is a disk 0Bc C such that for all g
there exists an analytic map g.'B -, g(0)=g
whose d tangent map TL0; TB + T takes (z a
coordinate in C) into vl this JmmedIataly
implies the lemma. Now recall from [HM 2]
that the connected component of the group
Aut(G an) of analytic group automorphisms
of G an has a natural structure of affJne
algebraic group of which AutG is an
algebraic subgroup (both being viewed as
embedded into GL(L(G))). (1.17)
COROLLARY. Let G be an irreducible
arline algebraic C-group. Then P(G/C)C
L(Aut(Gan)) (as subspaces of H(TGan)).

(1.18) THEOREM. Let char K = 0 and G be


an irreducible arline algebraic K-group.
Then a derivation in P(G) belongs to
P(G,fin) if and only if it preserves the ideal
of the unipotent radical of G. In particular
P(G,fJn) is a Lie K/k-subalgebra of P(G).
We shall provide later (cf. (3.2a) of this
Chapter) a purely algebraic proof of this
result. Here we shall give a proof of it based
on the following analytic result: (1.19)
THEOREM. (Hochschild-Mostow [HM2]).
If G is a connected affine algebraic C-group
then an analytic group automorphlsm
Aut(G an) belongs to Aut C if and only if it
preserves the unJpotent radical U of C.
(1.20) Proof of (1.18). Start with the "only
if" part so assume p e P(G,fin). Then by (I.
3.11) there exists a Kip]-field extension

Ki/K such that KT[P]= K Kip] (call this


field K o) and SUCh that K 1 )K C Js a split
algebraic Kl[p]-grou p so K l K G" K 1 )K
Co for some algebraic o Ko-grou p G o. Let
U o be the unipotent radical of Co; then K 1
KoUo must correspond via the above
lsomorphlsm with K l {)KU. But p clearly
preserves the ideal of K 1 K Uo so it wJJJ o
preserve K 1 )K U hence U itself (by
faithfully flatness of the projection K 1 K G
G). Let"s prove the "if part". It is easy to
reduce the problem to the case K = C. Let
now P P(G) preserve U hence also L(U)
by (1. 1.16). Then by (1.15) we may write G
= K $K Go o
36 (K = KG, G O a Ko-group) so p = p" +
v where p* acts via a on K and acts trivially

on G o while v P(G/C). Since p* clearly


preserves U so does v. By (l.l&) there is a lparameter group of automorphisms C x Gan+
Gan, (t)t C whose derivative at ! = 0 is v.
Then for each t C, t preserves U. By (1.19)tE
AutG for each t hence by (L 3.11)vE P
(G/C,fin). But C(G)is locally finite both as
Der(K/KG)-module and as P(G/K,fin) =
L(Aut G)-module (I. 3.12). Since by (l.ll)
P(G/K,fln)= P(Go/Ko,fin) K we have
[Der(K/KG),P(G/K,fln)]c P(G/K,fin) it
follows that )(G) is locally finite also as a
Der(K/K G) e)P(G/K,fin)-module in
particular as a Kip]-module which proves
the theorem. EXAMPLE Let's re-examine
Example 2 given after (1.9) so let G = G m x
Ga. We have seen that the elements of P(G)
have the form where pz DerkK, a,b z K.

According to Theorem (1.18) the elements of


P(G,fln) have the form p +-[ where pz
DerkK, be K. Indeed the ideal of the
unipotent radical of G is generated by X - I
In K[X, X-1] and the condition that
preserves it is equivalent to a = 0. (1.21)
The following concept will play a role in
Chapter 5. Assume G is a K-group scheme
and assume P(G,fln) is a K-linear subspace
of P(G). An ideal In P(G) will be called a
representative ideal if: a) it is a K-linear
subspace of P(G/K) b) it is a complement of
P(G,fin) in P(G) c) it is Aut G - invariant. As
we shall see in Chapter 3, representative
ideals may not exist even if G IS algebraic
and char K = 0. We shall prove here (using
[HM 2] once again): (1.22) PROPOSITION.
Assume char K = 0 and G is an Irreducible

aflne algebraic K-group. Then P(G) contains


at least one abelisa representative ideal. Our
basic ingredient IS: (1.23) THEOREM
(Hochschlld-Mostow [HM2]), Let G be an
irreducible algebraic C-group. Then AutO(G
an) is the semidirect product of AutC by
some normal algebraic vector subgroup N of
Aut(can). We will also need the following
Jemma whose proof is "obvious": 37 (l.210
LEMMA. Let L be a Lie K-algebra of
dimension n, L 1 a Lie subalgebra of
dimension n I and A a locally algebraic Kgroup acting on L by Lie algebra
automorphisms. Assume K Is an
algebraically closed subfield of K over
which all the above data are defined. Let Y
denote the subset of all K-points in the
Grassmanlan X of (n - nl)-planes of L which

correspond to subspaces L' of L enjoying the


following properties: 1) L' is an abelian
subalgebra of L 2) L is an ideal in L 3) L 1 L'
= L /$) L' is A-invariant, Then Y is locally
closed in X(K) in the obvious Ko-tOpology
of X(K). (1.25) Proof of (1.22). Put K o = K
G And let G -- K K Go' cf. (1.l). It is
sufficient to find o an abel Jan ideal 3 o of
the Lie Ko-algebra P(Co/Ko) complementary
to P(Co/Ko,fin) and ^UtCo-invariant because
then P(G,fJn)=Der(K/Ko)eP(G/K,fin ) (cf.
(1.15), (1.18)) hence 3 = 3o K K will be an
abelJan representative ideal in P(C). Now
by (1.2/$) it IS su:[flclent to o find and ideal
3 of P(G/K) which is abelJan complementary
to P(G/Kfin) and Aut G-invariant. By (1.2/$)
again we may assume (after replacing K by a
field extension of K or by a suitable subflald

of K) that K = C. But then (1.]7) and (I.23)


show that viewing P(C/C) as a subalgebra of
L(Aut(can)) we have that 3 = P(G/C)n L(N)
satisfies our requirements (N as in (I.17)).
$o (1.22) is proved. EXAMPLE Assume G =
G m x C a as in Example 2 (1.9), (1.26). A
representative ideal in P(G) is the K-space 3
o! all derivations aXX, a K Indeed Aut G =
G m acts on K[XX'[,[] by leaving ( fixed and
multiplying with non-zero scalars in K so 3
is Aut G-lnvarlantl it is obviously a
complement of P(G,fln) in P(G) and it is an
ideal in P(G) because p DerkK, a K 2.
Algebraic method= direct The aim of this
section is to investigate by purely algebraic
means the structure of P(G x H) where G,H
are arline (non-necessarily algebraic) Kgroup schemes (char K arbitrary) With H

linearly reductlve (i.e. all rational Hmodules are completely reducible).


lemarkably,
with only a minor extra effort we can make
our arguments work in the (noncommutative) Hopf algebra context [Sw].
Therefore we shall place ourselves (In this
section only!) in this more general frame and
freely borrow from [$w] the Hopf algebra
terminology. As an application we will
provide for instance a description of P(G x
GL N) in case char K = 0 (recall that in this
case GL N is linearly reductive). Let A be a
Hopf K-algebra (K/k as usual a field
extension of arbitrary characteristic with K
algebraically closed) and denote by 'p,S,
the comultiplicatlon, antipode and co-unit of

^. Define ll(^) to be the Lie K/k-algebra of


as! k-derivations p of ^ such that pKc K, pp
= (p ) 1 + I ) p)p, p$ = Sp, pe = k-p (we
wrote lI(^) and not P(A) to avoid confusion
with the space of primitive elements of ^
[$w]). Clearly, if G is an arline K-group
scheme with Hopf K-algebra A = ((G) then
lI(^) = P(G). Our main result will describe
II(A e B) where A, B are Hopf K-algebras
with B co-semisimple. Note that by a result
of Sweedler [$w] if B = CO(H) for some
affine K-group scheme H then B is cosemisimple if and only if H is linearly
reductlvel if moreover A = CO(G) for some
arline K-group scheme G then of course lI(^
B) = P(G x H). We start with a preparation
on centers and cocenters of Hopf algebras.
(2.1) Let C be a K-coalgebra; A coldeal I of

C is called cocentral if Tc(1)q)p = = (q )l)p:


C+ (C/I))C where p is of course the
comultiplication, q: C * C/I is the canonical
surjection and : C (D(C/I)+ (C/i))C is the
twist map (equivalently, If px - z px 1 o C for
all x = C, where : C)C + C)C is the twist
map). If dim C < m, I is cocentra] if and only
if the subalgebra (C/I) of C is central
One easily sees that 1! dim C < then there is
a minimum concentral coideal I C of Cs
indeed let A be the center of C and put I C
= Ker(C = cOO+ ^o). Now we claim that any
co-algebra C (possibly of infinite dimension)
has a minimum co-central coldeal I C.
Indeed, take any family (Ci) i of
subcoalgebras such that C = T.C i and dim C
l < then I C = T. IC. Is easily seen to be the
minimum cocentral coldeal in C. (2.2) ^

coalgebra C will be called cocentral if I C =


Ker E (='= the counit). If dim C < = then C is
cocentral i! and only if C Is a central
algebra. (2.3) Recall from [$w] p. 161 the
following basic properties of simple
coalgebras: if C is a coalgebra, C = r.C F C i
subcoalgebras then any simple subcoalgebra
of C lies in one of the Ci's. Recall that a
coalgebra is called co-sesimple if it is the
sum of Its simple subcoalgebras. The above
propsty implies that if C is co-semlslmple
then any subcoalgebra of C has a
complementary coalgebra and is the sum of
its simple subcoalgebras. (2.) Let B be a
Hopf K-algebra. ^ Hopf ideal :3 is called
cocentral if it Is so as a coldeal. Any Hopf
algebra B has a minimum cocentral Hopf
ideal 3B; indeed put 3 B = B(r. SnlB)B

where is the minimum cocentral coideal in


B. The luotient Bc = B/2E is called the
cocenter of B. 39 (2.) Assume in (2. a)
above that B c is a group algebra and let B
=)B be, he Xm(BC)-gradatlon corresponding
to the coaction of B c on B on the right p: B+
BB c (In analogy with (I. 1.3) we denoted by
Xm(^) rather than by G(^) as in [Sw] the
group of group-like elements of ^); lnce 3 B
Ls cocentral this gradation coincides with
the gradation Corresponding to the left
coaction ,: B+ BCB. We claim that all B's
are subcoalgebras of B. Indeed p: B+ B)B is
equivarlant with respect to the right
coactions of B c on B and B)B (for the latter
take the coactlon p on the second factor)I in
Particular p(B )c B B . Slmilarily
equivarlant with respect to the left coactlon

of B c X on B and BB (for the latter take the


coactlon ), on the first factor)and get p(BX)c
B .B; consequently p()c B X )B and our
claim is proved. Note also that B 0 for all X
Xm(BC indeed, if be B lies above X
Xm(BC) then bx, the X - homogeneous piece
of b, also lies above O hence b x 0. (2.6) ^
Hopf algebra is called co-semlsJmple [Sw]
If It is so as a coalgebra. In [Sw] p. 29a
several characterizations of co-semlsJmple
Hopf algebras are given; In particular as
already noted commutatlve co-semlsimple
Hopf algebras correspond precisely to
linearly reductive affine group schemes.
Note that if B is cosemistmple then the
cocenter B c is a group algebra. Indeed write
B -- e) Ci,C l simple (cocentral)
subcoalgebres. Then B/I B = e (Ci/Ict) -- =

(Ci/kerCi) is generated by IF'oup-llke


elements hence so will be ]C (being a
quotient of /l). (2.7) Let's discuss the notion
of center of Hop! algebra. A subset of a Hopf
K-algebra A will be called central If each
element of it commutes with all elements of
A. Then ^ contains a maximum central sub
Hopf algebra cA (we let CA be the maximum
element of the family of all central, S-stable
$ubcoalgebras of A). /e call CA the center of
Ai It is contained (but apriori not equal to)
the center of the underlying algebra of A.
Note that tf Xa(A) respectively xa(cA)
denote the spaces of primitive elements of A
and CA respectively, then Xa(CA) consists
precisely of the central elements of Xa(A)
(once again we used the notation Xa(A)
instead P(A) as In [$w] In analogy with our

notations in (I. 1.3)). (2.g) Let A and B be


two Hopf K-algebras. Then A )B has a
natural structure of Hopf algebra, As in (I,
1.19) there is a K-linear projection
II(AB)+I(A)x Der K E(B) over DerkK,
defined b y p + (pA,PB) where PA (IA
)B)PiA (where l A ., A+ AB iA( = a I) and
PB defined slmllarlly. Our projection admits
a section (which is a Lie K/k-algebra map)
defined by (Pt'P2) + Pl DI + l )P2' We shall
usually identify II(A) x DerkKE(B) with its
image in (2.) Next assume In (2.8) above that
the cocenter B c is a group algebral by (2.6)
this the case for Instance when B is
cosemlsJmple. Then we shall define a
remarkable Lie K-sub- algebra I(B: A) of
I[(A B) lying In the kernel of the projection
I[(^ B) + l(^) x DerkKT[(B). We proceed as

follows.. consider the Xm(BC)-gradeti_on B


-- B on B defined by the coaction of BC on B
(cf. (2.)) and consider also the space xa(cA)
of prlmive elements of cA. Then for any
group homomorphlsm a
Hom(Xm(BC),xa(cA)) define the K-linear
map Es-.^ B- A )B by
the formula Ea(x b) = r. xa(x) b where b = -b
X is the decomposition of b B into
homogeneous pieces. Then clearly E a is an
A-derivation and using the fac! that all Bs
are subcoalgebras of 13 one checks the fact
that E a ](A B). /e defined a linear map E:
Hom(Xm(BC)xa(CA)) + ](A B) whose
image will be denoted by ](B = A). Since B(
0 for all X (cf. (2.)) it follows that
Hom(Xm(BC), xa(CA)) = I[(B '. A) is an

ahellan Lie K-subalgebra of ](A ) B). Our


main result is the Then (2.10) THEOREM.
Let A and B be Hopf K-algebras with B
cosemisimpie. Then T[(A B) = (l(A)XDerk!
(T[(B)) T[(B: A) (2.11) COROLLARY.
Let G and H be affine !(-group schemes with
H linearly reductive. P(G x H) =
(P(G)XDerkKP(H)I
Homgr(Xm(Z(H)),Xa(C)) (2.12) Proof of
(2.10). We must show that every p Ker(][(A
)13)+ ]I(A)XDerkKl[(B)) lies in ][(B .' A).
Clearly p is K-linear. Step I. We show that p
is an A-derivation. It is sufficient to check
that p(A K)c A K. By (2.3) B--K )M for
some coalgebra M. Consequently for any a A
we may write p(at)a o)! +AM witha oeA.
wegetthat (p(a ) 1)) iJ(ao ) l) A )lVl )A
)M, i(aol) A K )A K On the other hand

I(p(a[))=(Pl + lp)u(al)c (pl)! + I)p)(AK c


AKA K + AMAK +AKAM we get that
p((a I)) = IJ(a o l) hence p(a e l) = a o
I and our claim is proved. Step 2. We show
that for any subcoalgebra C of B we have p(!
< )C)c A )C. Indeed by (2.3) C has a
complementary coalsabra C'. For each b C
we have p(l )b) Xb + A C', XbeAC e et
41 p(p(1 )b)) p(x b) + A {)C' A e)C', p(x
b) A e C A C On the other hand p(p(lb))=
(p)l+lp)p(lb)e(pl +I)pXKC+CK)C
CABKC+KCAB e get that p(p(l b)) =
p(x b) hence p(l b) = x b and we are done.
Step . We show that for any simple
subcoalgebra C of 13 there exists a C
xa(CA) such that for all b C we have p(l
)b) = aC )b. Start with the following general
remark.' if C is any simple K-coalgebra and

L .' C + C is a K-linear map such that IL = (L


) IC) p = (1C L)p '- C + C C then L is a
scalar multiple of the identity. Indeed choose
t g K such that L - tl C Js not lnvertible. The
equalities I(L - tl C ) = ((L = tlc) elc)l =
(Ic(L - tlc))l show that the Image V of L-tl C
is a subcoaJgebra of C since V is not the
whole of C it must be zero hence L = tl C and
our remark is proved. Now let a eA and
La. = (a* ) Ic)Pi C: C +A )C +A C +C
where i C: C +A )C, ic(X) = 1 )x. We claim
that L . is a scalar multiple of 1C. To see this
note first that a* = (a* ) eA)p -' A K which
implies tat PC(a* gJ l C) = (a* 1C )e A e)
IC)IJAC = A )C * C C Now if x C we
get PC(La.X) = pc((a e l)p(l x)) = (a* 1C eA
IC)FA C p(I x) = = (a* 1C eA IcXP IA C
+ IA C P)PA C (1 x) = = (((as Ic)P) eA

IC)A C (1 x) + (a IC)((e A Ic)P)p A e C(1


x) The second term oJ the la*ter sum
vanishes because p projects into 0
A)XDerkK(B). Using the usual sigma
not=tion pC x = ZX(l ) x(2 ) we get PC(La.
x) = ((a* C) c A lC)H x() x(2 )) = =
La.x(j))x(2 ) = (La. JCXUC x) In other
words cLa = (La IC)U.. Similarily starting
with the uality a* = (Aa*)U we get HCLa. =
([CLa.)U By the remark we made L . must
a scalar multiple o the identity and our claim
is proved.
42 Our claim impties that there exists a A
such that p(l b)= a)b for all b c C. All we
have to check now is that a Xa(CA).
Writing 0 = PA B p(l ) b) - (p l A ) B +
l A ) B ) P)P^ B b = = (iA)IBX(PA a- al -

1 where =A)B+BOA is the twist map, we get


acXa(A). Writing xa)b=p(xOb)= = p((! ) b)
(x ])) = ax b for x A we 8at a Xa(CA).
Step 4. We show that whenever C and C' are
two simple subcoalgebras of B contained in
the same B_, we have a C = ac,. IndeXed, by
(2.1) and (2.) the ideal 3 B = Ker(B + B c) Is
generated (as a two-sided ideal) by elements
of the form $nx where n _> 0 and x belongs
to the union of all simple subcoalgebras of
B. This together with Step 3 shows that the
ideal ^3 B is stable under p In particular p
induces a derivation ](A )B c) and we have a
commutatlve diaram Now clearly the
projection of in E(A) x DerkKlI(Be) is still
zero, hence by Step 3 applied to and B c
Instead of C and B we get that (1 e X) = ae X
for some a X A. We have for y b C (pel +

I OXl e I el eq)p(lb) = (POt + IOlOblex )=


=acOb10 + lb On the oth hand (i ] l Oq)pp(l
b) = irom hich e get a C = . 5imilaHly a C' =
d e e done. St 5. !1 of . By (2.3) each B is a
sum of simple subcogebr hence by Step a we
get a 1unction Xm(B ) XaA), X a) such tt p(l
b) = a()b for 1 b B (notethatB B
.0foralJxi, x2becauseby(2.5)0B cB ' 112 ).
Xl X2 X2 (2.13) REMARK. The oo o Step 2
above shows that or any co-semlslmple Hop
K-algebra B, any subcoalgebra C o B and y
pt () we have pC c .C. In particular B is a
locally finite D-module where D =
EXAMPLE The above remark plus (I. 3.2)
show that if char K = 0 and G Is a reductlve
linear algebraic K-group then P(G/K) =
L(^ut G) = L(Int G) = L(G)/L(Z(G)) In
particular for Instance if G = CL N =

SpecK[y,-y-y ], y--(Yij), then P(G/K) has a


K-basis consisting of the derivations Pb=i
(m bimYmj. m b B m jYim)'Yij where b =
(bij) gIN(R ). Consequently for G =
CLNP(C) consists of all derivations of the
form p* + Pb where e I p e DerkK and p#
DerkK[Y,d-e], lifts p and kills y. K 1 Assume
now G=GnxCLN=$pec [[,y,ey], [=([1'' ,in ).
Then by (2.10) P(G) consists of all
derivations of the form: =p*+pa+Pb+Pc
where p e DerkK , Pb is given by the formula
above, Pa = ' a = %) gin(K). PC =(m
m;mXi Yij ']j )' c= (l .... ,On), c i .K It
worths noIng that P(G,fin) consists of all
derivations of the form p* + Pa + Pb' On the
other hand t space o derivations of the orm
Pc is a reesentative ideal in P(C). .
Alsebrale m!l: $eml-t IXXluets Let M be an

afflne K-group scheme, T a diagonalizable


efflne K-group scheme [DG] and 0: M x T -'
M (13(m,t) = m t) a right action of T on Ml to
simplify discussion we always assume in
what follows that O(M) is an integral domain
and Xm(M)= 1. Our aim here is to study the
structure of P(M x pT) where M x pT Is the
semidirect product of M and T constructed
with the help of p. This will enable us in
particular to re-prove algebraically (lolg),
get new information about the map [V: P(G) Hom(Xm(G) , Xa(C)) and compute P(C) in
case the radical of G Is nilpotent or the
unlpotent radical of G is commutative. For
sake of simplicity we agree from now on the
write W(G) instead of Hom(Xm(G) , Xa(G))
for any K-group scheme G. 0.1,.) We shall
think of M x pT as having its underlying

scheme equal to M x T while


multiplication is given by the formula t 2
(ml,tl)(m2, t2) = (m I m2,tlt2) , m I M(K),
t I T(K) hence by the composition (3.1.1)
p:MxTxMxT' TxT lXxl TxMxTxM '
TxMxTxM PT x Pm lxxl TxTxMxM TxM
,MxT where : M x T + M x T, (m,t) =
(P(rn,t),t) = (mr,t) and T is the twist map.
The antipode is given by (m,t) + ((m-l)t' l,t])hence by the composition Sl x s T (3.1.2)
S.'M xT -M xT ' M xT and the unit is given
by MXCT -MxT (3.1.3) : Speck Let r.
denote the group Xm(T) (hence ((T) -- K[Z]
_- group algebra on Z). To P: M x T + there
corresponds an algebra map still called P:
d(M) + ((M) )((T) hence a Z -gradation on
((M);, if we denote by fx :(M)+i(M))( c ((M)

the projection onto the X-component then


PMfx X "X"--) (3.1.a) (a ex) = r-Lx,(a)
X% a(9(), where ;C(M)((T)+((M)iT) is
induced by . we denote by (T, (M)) the set o!
all weights o! T in d)(M) i.e. (T,(M)) = {X Z
d(MLx 0} -- {x z x 0} Since ((M) is
integral, e(T,(M)) is a semigroup. For each X
r. put (Oerk) x _- {p OerkMp(<9())X,c
C(M) or ali ' s Z} _- = {P DerkM; fX-Ix, p =
Pfx' for all X' Z} o cour, e the sum in Uerk
of all (UerkM) X is direct. We "so t [(M) x -[()n (UorkM)X P(M) X --P(M)n (DerkM) x.
Now if p Derk(L9(M)((T))= Derk(M x
pT) is such that pKc K we define for each X
r. endomorphisms p)( Endk)(M ) by the
formula pxa = (, eex)P(a e l), 45 where e( .'
((T) -- K)(' e K( - K is the natural projection
onto the y-component. In other words the

following formula holds: (3.1.) p(a 1) =


y(pxa)e x for all a ed)(M). One checks easily
that Pt Derk(M) and PX DerK(M) for
all X J. We put pO = PX E Derk(U)
Moreover for any X we put Xa(M) X = Xa()
n (M) X. It is easy to check that Xa(M)
=eXa(M) and that we have Xa(M x pT) -Xa(M) i I c (() ((T). Moreover since
Xm(M) = I one chec{<s that Xm.(M x pT) -i Xm(T) c (M) ((T) so we have V/(M x
pT) = Hom(Xm(T), Xa(M) 1) (.2) LEMMA.
Let p Derk(M x pT) such that pK c K as in
(3.1). Then p P(M x T) H and only if the
following conditions are satisfied: 1) For
any X r. we have p(l X) = aXX for
some a X Xa(M) 1 . 2) For any X r. we
have (fx ) l)PM p I = (p I fx 1 + r. fx
ax)PM 3) For any X,X' r., X I we have

(fX,X_ 1 I)pMP X = (pxfy, I)P M a) poe M =


cup ) For any X r. we have r'SMP I SMf+ axf X--'r'f I P Proof. 1) is equivalent to the
act that pp and (p i + ! p)p agree on ] r.
(use(3.1)). We claim that 2)+3) Js equivalent
to the fact that these maps agree on (M) I.
Indeed, for a )(M) we have (using (3.1.1),
(3.1.)): = r.x,((pxa)(l)) x (pa)() (p J
+ ! p)p(a l) = r. px,(fxa(l)) X' a(2)X + + r.
f7a(I) 1 e Px,a(2) XX' + r. fxa(l) I a(2)axe X
Using the fact that 0)(M)<(T)C(M)(P(T) is a
free d)(M)CP(M)-module with basis i ( !
.}( and identifying coefficients we get our
claim.
48 Condition ) is equivalent to pc= ep.
Finally using formula (3. h2) and a
computation similar to the one above we see

that condition 2) is equivalent to p$ = 0.3)


COROLLARY. The map assigning to each p
P(M x iT) the function (pa) (3.1), (3.2)
gives a hi}action between P(M x pT) and the
set of all functions T = Xm(T) + (DerkM) x
(Xa(M)l) X (Pxa) satisfying the following
conditions: a) The map ' a X is a group
homomorphlsm b) plKc K PX L() or 1 X 1
and or y a () here are only finiSely many s
which p 0. c) The relations 2), 3), ), ) rom
(3.2) are salsied (wlh = . p. Les check firs ha
li p P( x T) hen PX L(M) I. Indeed
summinR up he relations (3.2), S) or 1 we
ge p = (p 1) m which means ha p L().
Inruclng laer ormula In (3.2), 3) again and
applying l we gel .lp X = P., whl is wha we
wanted. To conclude R i sufflcien o e t or
any function satisfying a) b) c) above he

{ormula p(a X) = (Zpa X) + aa x a (), X


Z defines elemt p P( x T). (.q) COROLY.
Aume in afions ave ha p P( x T). Then the
lollowing relations ld f pO d a a) ppo = (pO
e I + 1 pO + tX a PI. c) Is (3.2), ). a) lollows
by summing up (3.2) 2) ov 1 d adding o
the obtained relation relations Mpx = (p
l)pM f X I (cf. (3.3)). b) follows by summing
up (3.2) ) over 1 X E. ( There e three
remarkable ma from P(xT) namely P( x )
+L(), p p+ = X I p x + x p + vp, 0. P( x T)
*Derk , p *p=p++pl , Of course he second
map assigns o any p h map a cf. our previous
notations in (3.3) and 47 where we identify
V(M x pT) with Hom(T.,Xa(M)i). We have
the following properties of these maps.' 1)
pOe p(M)epO is SM-invariant =[?p vanishes
on (T,{(M)) (the latter simply means that p

kills the weights, viewed as elements of (M


xiT )3 we shah repeatedly us this expression
Jn what follows). 2) If p+ = 0 then pO
(DerkM)l 3) We have an exact sequence 0 +
P(M) 1 + P(M x pT) + L(M)eW(M x pT) *)
We have {p INIVl x pT)Ip + = 0}
{(p,a) e (DerkM) 1 x Hom(T.,Xa(M)l)
p,a satisfy a), b), c) In (3. t)} 2) The image
of JL? .. p (M x iT)+ W(M x pT) contains the
space of those maps a: T.+ Xa(M)i which
vanish on e(T,((M)). So clearly equality
holds provided P(M x pT) kills the weights.
Proof. 1) follows directly from (3. t). To get
2) note that pO = Pl so introducing (3.#), a)
in (3.2), 2) and appying 1 )gM we get fxp
= pf X for all xwhlch Is what we want. To
get 3) note that the inclusion P(M)lc P(M x
pT) Is given by associating to any Ple INM)I

the function I +(pt,O), X (0,0) for X 1, cf. the


identification In (3.3) hence clearly P(M) 1
Is contained in the kernel of P( x pT) [,(M)
W(M x pT).' The converse inclusion is also
easily seen. Finally t0, ) are easily checked
through the Identification (3.3)-' for 2) put a
X = a00 and PX = 0 for all X. (3.6)
EEM^EK. We will give at the end of (.2)
an example of a situation when P(M x pT)
contarns a derivation not killing the weightsl.
This shows that P(M x pT) can be rather
complicated. In our example T will be G m
(hence T will he algebraic) and M will be
unlpotent (but non-algebraic)[ we dont know
whether it is possible to find such an
example with both T and M algebraic. On the
other hand we have the following.. (.7)
[,EMMA. !n notations above assume ts a

vector group scheme (i.e. ((M) Is the


symmetric algebra over its subspace Xa(M)
of primitive elements). Then P(M x pT) kills
the weights of T In d)(M)o Equivalently, the
image of the map J[.'P(MxpT)+W(MxpT)= =
Hom(Z,Xa(M) l) is precisely the space of all
maps in Hom(T.,Xa(M)i) vanishing on the set
of weights (T,t(M)). Proof. Applying the tvist
map .. (IVl)m((M)+ ((M))(M) to formula (3.
t0, a) and using the fact that TI M = i M we
get IM'p = (1 pO + pO I)l +
hence using (3.#), a) once again we get that
(r. fX & ax)p M = (r.a( Let x e Xa(M) X, x 0
be a primitive element of weight X Ii
applying the latter equality to x we get
XeaX: axX. Ve claim that ax:O; for if it is
not so the latter equality im plies x -- Xa X

Xa(M) 1 which is impossible since )r h


So a X = 0 for all weights of T in Xa(M)
(i.e. for all X such that Xa() X 0). But the
weights of T in (M) are products of weights
of T in Xa(M) and hence a( = 0 for all (
(T,{(M)). (3.8) LEMMA. Assume L(M)
contains a finite dimensional Lie subalgebra
L with the following properties-' a) L(/)( c L
for all ( I and [P(M),L] c L in DerkM. b) (M)
is locally nilpotent as an L-module. c) ((M)
is locally finite as a P(M)-m odule. Put P:
Ker(V: P(U x pT) g(M x pT)). Then ((M x
pT) is locally finite as a P-module; in
particular Ker J c P(M x pT,fin). Proof. First
note that since dmKL< the set =(}(r.
L(M)X0, 1) of all non-trivial weights of T in
L(M) is finite. Next we claim that for any a
(), the K[P]-subm o- dule of (() (T)

generated by a I has finite dimension. Indeed,


by (3.a) for p P we have that pOp(M)
(because a X=0 fgr all () hence Pl-P'P+P()L and PX L for ' I. Since (() is a
locally finite both as a P(M)-m odule and as
an L-module and since [P(M),L] c L it
follows that (() is locally finite as a P(M))Lmodule. Choose a finite dimensional vector
space V of (() containing a (() and stable
under P(M)eL. Hypothesis b) says that for V
there exists an integer N _ I such that ele 2 ...
e N = 0 for all e l, ... ,e N L. Then It is
easy to check using [P(M),L] c L once again
that we have ele 2 . .. en -- 0 for all e 1, ...
,{a n P(M)e L provided card (ie i L)>_
I. Now let (N) be the K-span in (T) of all
products of the form XiX 2 ... X N with X i
u 1) -- . For pl,...,pn p we have

plp2...pn(a)l)= I 2 ...P;na " r'Px I PX2 XnXn1 the sum bein taken for all n-uples (Xl,...Jn)
with XinU. As we have seen, 1 2 n = 0
whenever card {1;' i *} > N.
Consequently K[P](a 1) c (N) and our PX
IPx2 ' ' ' P)'n a - claim is proved. Now we
claim (and this will close the proof) that for
any x e) (T) we have-dimKK[P]x ( .
Indeed write x --r.a I i and let be a finite
dimensional K[P]-submodule of {(IU)
((T) containing all ai's. Then r.(l a)Xl) is a
finite dimensional KiP]-module containin x
and our lemma is proved. (3.9) REMARK. If
in (3.8) above we replace the condition c) by
"c') P(M) = P(l,fin) (in 49 other words (M)
is locally finite as Kip]-module for any
peP(M))" then we still get the second part of
the conclusion namely that "Ker J[?c P(M x

pT,fin)". This follows just by inspecting the


proof of (3.10) LEMMA. Let G be an
irreducible unipotent arline algebraic Kgroup. Then: l) Assume char K ) 0 and G is
commutative. Then P(G) -- P(G,fin). 2)
Assume char K = 0. Then (C)'Is locally finite
as a D(C)-module. Proof. l) By [H] pp. 2 and
63-6, ((G) is locally nilpotent as an L(G)module i.e. for all y ((G) there exists an
Integer N such that 012... aNY=0 for all 01
..... e N eL(.G). If Xx End K (G) denotes the
multiplication in (G) by some element x e
(C) then for any O eL(G) we have [0,Xx] =
).; so by (I. 1.3) if x Xa(C) then [0,] Is the
multiplication by some scalar In K. Now
pick and element p = r. a 6 l = rnie (
eXa(G)eL(C)= P(C/K)= P(G), cf. (I. 1.9)
where ()i is a K-basis of L(G) and a Xa(G).

Then it is easy to see using the above


remarks that K[p]y is contained in the Klinear span in (G) of the set (iai2 ... ain%l2
... In particular dlmKK[p]y < for all y
C(G)and assertion 1) is proved. 2) By [H] p.
231 the image of the map (I. hi3) G(G)+G, e
JG , U(L(c))O coincides with the algebra
B(L(G)) of nilpotent representative functions
on U(L(G)) (recall that by definition B(L(G))
consists of all functionsis on U(L(C))
annihilating some power of the ideal 3 =
L(G)U(L(C)). Now B(L(G)) is a locally
finite D(G)-submodule of U(L(G)) since it is
the union o the finite dimensional D(G)submodules B n = {! .U(L(G)); f vanishes
on 3 n } Since Jc is a map of D(G)-modulas,
d)(G) will be locally finite and we are done.
(3.11) REMARKS. l)(G) is not a locally

finhe D(G)-module even in the case G = G a,


i d char K = q > 0 (if G a = SpecK[F,], Pi =
q P(G) then pi = q hence dimK(P(G)D =
.) so the conclusion in (3.10), 1) cannot be
replaced by "((G) Is locally finite as a D(C)moduJe". 2) It would be interesting to know
if in (3.10), 1) one can drop the
commutattvity assumption; one can formulate
the conjecture that P(G) = P(G,fin) for any
irreducible arline unlpotent algebraic Kgroup G (char K arbitrary). (3.12) Let C be
an irreducible arline algebraic K-group, let
T be a maximal torus of the radical o! G and
let (T,(C))c Xm(T ) be the set of weights
of the action of T on ((G) be Inner
automorphisms. Moreover let (G) be the
subset of Xm(G) of all characters of G
whose

5O restriction to T belongs to (T,(C)); the


elements of (G) will be called the weights of
G. Since the maximal tori of the radical are
all coniugate, (G) does not depend on the
choice of T. We let Wo(G) be the subspace of
W(G) = Hom(Xm(G), Xa(G)) consisting of
all maps vanishing on (G). Here is our main
result in arbitrary characteristic: (3.13)
COROLLARY. Let G be a solvable
irreducible arline algebraic.K-group. 1)
Assume the unipotent radical of G is
commutaive. Then the kernel of is contained
in P(G,fln). 2) Assume the unipotent radical
of G is a vector group. Then P(G) kills the
weights of G. So the image of J[V: P(G)
W(G) equals Wo(G). Proof. By standard
structure theory [H], G -- MTwith U the
unipotent radtcal of G and T a maximal torus.

Applying Lemma (3.8) and Remark (3.9) for


L -- L() together with Lemma (3.10) we get
assertion l) of the Corollary. To check asser,
tJon 2) note that the weights of G as defined
in (3.12) can be identified with the weights
of T on ((M) as defined in (3.1) so we may
conclude by (3.7). (3.1e) REMARK. In
notations o! (3.13) it is not true that P(G,fln)
is contained in Ker (?: P(G) +W(G)). Indeed
take G -- G a x G m as in (I. 3.9); then with
notations from loc. cit. this phenomenon does
not occur in characteristic zero as shown by
the following: (3.15) LEMMA. Assume char
K = 0 and G is an arline integral K-group
scheme. Then P(G,fin) = Ker( ?: P(G)
+W(G)) Proof. Let p P(G,fln) and assume lp
X-' a 0 for some X EXm(G)' By (I. 1.3),
Xa(G) is stable under P(G). By [H] p. $g the

symmetric algebra $(Xa(G)) embeds into


(G) and then of course each homogeneous
component $n(Xa(G)) is a P(G)-submodule
of ((G). By induction we get that pax =(an.
)X lorn_)0 with [n ) Si(Xa(G)). By our
assumption the K-span of the family (pn n is
finite i<n- 1 dimensional' this implies that the
same holds for the family (an+ )n which is
impossible because a n e$n(Xa(G)). The
lemma Is proved. In what follows we devote
ourselves to affine algebraic group in
characteristic zero. (3.16) THEOREM. Let
char K = 0 and let G be an irreducible aine
algebraic K-group. Then: l) P(G,in) = Ker(
JT: D(G) +/(G)) and ((G) is locally finite as
a P(G,fin)-module. 51 2) The image of J[?:
P(G)+ W(G) contains Wo(G). H. ore
precisely there exists a K-linear map E:

Wo(G)e P(G/K), a E a such that 9) E is


the natural inclusion Wo(G)c W(G) and
having the following properties: a) Im E is
an abelian ideal in R(G/K,fJn)elm F(c
D(G/K)). b) For any AutG and a Wo(G)
we haveo'lEa o = Eoa where we still denote
by the induced automorphisms of ((G) and
Wo(G). c) For any pe Der(K/K G) upon
letting G = (G o a KG-grou p) and letting p*
be GOKGK the trivial llftings of p from K to
C(G) = K KG(G o) and to Wo(G) = K
KGWo(Go ) we have [p*,F a] = ERa a for
all a REMARK. Assertion l) can be easily
deduced in fact from our previous result (I.
1.18), cf. (].2) belowl but our proof here for
(3.16) will be purely algebraic. We start
with some preparations (cf. also (3.17)
LEMMA. Assume char K = 0 and g is a Lie

D-algebra of finite dimension. Then the


radical r of g is a Lie D-subalgebra. Proof.
By (l. 3.2) and (I. 3. a) g splits over some Dfield extension K1/K with K o := KiD = K D
so gK l = goK K] for some Lie Ko-algebra
go' Let r be the radical of go' o Both
roKoK l and rKK l coincide with the radical
of gKKl . But r=(r)KKl)n g= = (ro) K Kl}n g
and the latter space is preserved by D. o
(3.18) COROLLARY. Let char K = 0 and G
be an irreducible affine algebraic K-group.
Then the radical R of G is an algebraic
D(G)-subgroup of G. In particular there is a
natural Lie K/k-algebra restriction map
P(G)+ P(R). Proo:L Combine (I. 1.1) and
(3.17) above. (3.19) COROLLARY. Assume
char K -- 0 and M is an irreducible arline
algebraic K-group whose radical is

unipotent. Then ((M) is locally finite as a


D(M)-module. Proof. By (I. 3.2) and (I. 3.#)
we may assume that g = L(G) ts a split Lie
D-algebra so g = goK K (K = KD go = gD
D = D(G)). Let go -- ro + So be a
decamposltlon of go with r e its o radical
and s o a complementary semisimpie Lie
algebra. Then by [H] p. ll2, s--SoK is an
algebraic Lie subalgebra o g,s = L(S), 5c G.
By (l. 1.1;) both 5 and the radical R of G are
algebraic D-subgroups of G; therefore the
multiplication map R x 5+ G is a D-map
hence (G) identifies with a D-submodule of
((R) (S) so we are reduced to proving that
both (R) and (($) are locally finite Dmodules. The assertion about ((R) follows
from (3.10) because R is unlpotent. The
assertion or ((5) ollows for instance from

(2.13) (it also follows from representability


of Au.tS [GDI but the latter fact involves the
whole structure theory of reductive group
schemes over non-reduced basis so our
argument (2. ! l) should be viewed as "the
52 elementary" argument for the local
flnitness of C(S); for another "elementary"
argument see ge shall repeatedly use the
following rernarks (3.20) Let G = G I)G 2 be
a semidirect product of irreducible atfine
algebraic K-groups, char K = 0 and identify
((G) with (Gi)I(G 2) via the multiplication
map G 1 x G2+ G. If Xm(G 1) = I then
Xm(G) identifies with Xrn(G2). I! Xa(G 2) =
0 then Xa(G) = Xa(Gi )G2. Now assume
G2* G 1 is an isogeny. Then the map Xrn(Gl)
Xm(G 2) is injectlve with finite okernel

and the map Xa(G 1 ) Xa(G 2) is an


isornorphlsrn. In particular there is an
induced lsornorphism V(G l ) = V(G2).
(3.21) To prove Theorem (3.16) we fix some
notations let U be the unipotent radical of G,
let H be a maximal reductive subgroup of G
and T the radical of H. By a theorem of
Mostow [H] p. 117, G = UH{ moreover T is
of course a maximal torus of the radical R of
G. Put S = [H,H], U = US and := T = Us4(S x
T). The isogeny S xT- H induces an isogeny (
, G. g/e write I: = Xm(T). Note that
restriction map Xm(G) r. is injectlye and has
finite cokernell Indeed this map is easily
seen to identify with the map Xm(G) =
Xm(H) + Xm(S x T) = Xrn(T) cf. (3.20)
above and we are done also by (3.20). By
(3.15) Ker (V = P(G)+ W(G)) contains

P(G,fln); let's prove it actually equals


P(G,fin). By (I. 1.21) and (3.20) we have a
commutative square P(G) ---W(G) I [I P() --,w() so it is sufiicient to prove that (() is a
locally finite P-module where P -- Ker(l?:
P() + g()). Since T centralizes S, L(M) X -(L(U) L(S)) -- L(U)(c L(U) for all X r., y
1. Due to (3.19) we may apply Lemma (3.8)
to our situation (with L -- L(U)I note that
[P(M), L(U)] c L(U) by (3.18) and (I. 1.16))
so we get that (() Is locally finite as a Pmodule and assertion 1) in (3. i6) is proved.
To prove assertion 2) let's define the map E
Wo(G) - P(G), a + E a as follows. The action
of T on H by the left (or, which is the same
by the right) translations gives a Z-gradation
d() d)(.)X X Let f(:((H)-((H)Yc((H) be
the corresponding projections. Restriction

provides an identification W(G) Hom(r.,X,au)l'{). Then, any a : /o(G)


Horn(Xm(G), Xa(G)) can be viewed as a
homomorphisrn r. Xa(U ) vanishing on the
weights (T,C0(G)) e(T,((M)). Moreover,
identify ((G) with ('(U)((H) via the
multiplication map U x H- G and deitne the
K-linear endomorphisrn ! of ((G) by the
fornula; Ea(xOy )-_ r. x a()r)fY(y), 'x ((U),
y d(H). X Clearly E a Is an ((U)-derlvatton.
To check that E a E P(G) it is sufficient to
check that Its lifting a to d() -- C(U) (S x
T) belongs to P(). But we have Ea(Xz)-i:xa(x)''X(y), x ed'(U), z E(SxT) X where
'Xg((S x T)*(S x T)Xc ((S x T) is the
corresponding projection and d)(S x T)X_- - d)(S)KX. Now a P() by (3.3) (in notations
of (3.3) put a X -- a() and PX -- 0 for all So

our map E Is well-defined and clearly (V)


E Is the inclusion /o(G)c /(G) so Wo(G)c
Is(L?). It is also clear that Im(g) is an abellan
subalgebra of P(G/K). To prove the
remaining assertions in clalrns a) and b) of
(3.16) note that ^ut G is generated by lnt G
and the group Aut(G,H) of all
autornorphisms of G preserving H.
Consequently by (I. 3.12) P(G/K,fin)-- _L(AutG) is generated by L(lnt G) and
L(Aut(G,H)).' So it is sufficient to check that
the following hold for all a Wo(G): (3.21.1)
o' IEao -- E a for all o e !rn(M Int G)
(3.21.2) [d,E a] 0 for all d Im(L(U) L(Int G))
(3.21.3) o' IEao-- Ea for all o ^ut(G,H))
(3.21./) [d,E a] Irn E for all d z L(Aut(G,H))
To prove the assertions above we may
assume G = .. Start with (3.21.1) and

(3.21.2). The action of M on G by inner


autornorphlsrns is given by the corn position
of maps MxMxT a MxMxT b xMxT C)MxT
(m,x,t) : c (mx,m,t) , (rnx,trn'lt'l,t)-+
(mxlrn'lt'l,t) where a and c are induced by
rnultiplications while b is induced by taking
first the antipode on the middle factor and
then applying the action MxT M of T on M
by inner autornorphisms. This irnmediately
implies that for any z (M) we have that the
image of z 1 via the map a' b + c a: (M) )
(9(T) - ((ikl) ((IVI) ((T) belongs to )(M)d)
(M)< {(G)> where < {(G)>c d)(T) Is the Kspan of the group generated by {(G)= e(T,({)
(G)). This Shows that if oand d are as In
(3.21.1) and (3.21.2) we have o(z 1), d(z 1)
(ikl))< (G) >. This immediately implies
that the derivations o-IEao- E a and [d,E a]

Vanish on ((M) 1. On the other hand (since o


is the identity on 1 {3(T) and on Xa(G) and d
is the zero map on these spaces), the above
derivations also vanish on 1 ) (T) hence they
vanish on )(M))d)(T) and (3.21.1), (3.21.2)
are proved. To prove (3.21.3) let o still
denote the
54 restriction of a to M and T. Then for x
((M),) Xm(T)C ((T) we have - IEao(X
X) = o'lEa(Ox eoX) = 'l(a(oX)oX oX) = =
o'l(a())x X = To prove (3.21.) let d still
denote the derivation induced on d(M');
noting that d kills Xm(T) we have for x and
X as above: [d,Ea](x eX) = xd(a(x)) e =
Eda(X X) 5o assertions a) and b) are
proved. Assertion c) can be proved
similarily. (3,22) COROLIAR. Assume In

(3. i6) that either the unipotent radical of G


is commutative or the radical of G is
nilpotent. Then Im(J[V: P(G)+ W(G)) =
Wo(G) and ]m E is an abelian representative
ideal in P(C). Moreover each element of Im
E kills Xa(G). Proof. Assume the uniputent
radical of G Is commutative. By (3.18) we
have a commutative diagram Jc ? P(C) +
w(c) p(l) ,,, w(R) Moreover the right
vertical arrow is injectle and viewing it as
an inclusion we have W(G)n Wo(R) =
Wo(G). By (3.7) Im(J[RV)C Wo(R) so we
obtain that Im(ZGV)C Wo(G). A similar
argument shows that the latter holds If is
nilpent. This and (3.16) show that K P(G) =
G and P(G) = Pl P2 P3 with Pl = Der(/Kc)'
P2 = P(C/K,fin), P3 =Im E and [PI'PI ]c Pl'
[PI'P2 ]c P2' [PI'P3 ]c P3 [P2'P2 ]c P2'

[P2'P3 ]c P3 [P3,P3] -- 0. This and the


AutG-lnvarlance of ImE proved in (3.16)
show that ImE is an abelJan representative
ideal. It clearly kills Xa(G) so we are done.
EXAMPLE 1 Assume char K = 0, H is a
reductlve Irreducible algebraic K-group with
radical T, V is a K-linear space of finite
dimension and let = H GL(V) be a
representation. Put G = VxpH. Then (3.16)
and (3.22) say that P(G) has an abelian
representative ideal which naturally
identifies with the space Wo(G) of all
homomorphisms Xm(H) +(V) H which
vanish on all those y e Xm(H) for which ylT
belongs to the set (TV) of all weights of T in
V . So if [ is the subgroup of Xm(T)
generated by e(T,V) then dim Wo(G) -- (rank
H - rank )dim((V) H) S/here recall that

rank H -- dim T by definition. Consequently


we get: dim L(Aut G) - dim L(Aut C)= (rank
H - rank )dim((V) H) In particular P(G) =
P(G,fin) (equivalently L(Aut G) = L(Aut G))
in each of the following cases a) rank 1' =
rank H b) (v) H = 0 EXAMPLE 2 We
specialize the above example and then say
more. Start with a representation I: GL N
GL(V) and put G = VxpGLN; then P(G)
behaves quite differently according to
whether p factors through PGL N or not.
Indeed= 2e) if p doesn't factor through PGL
N then P(C) = P(C,fin) (indeed in notations
of the previous example rank 1' = rank H =
ll). 213) If p factors through PGL N then
P(C) has an abelian representative ideal of
dimension equal to the dimension of the
space () GLN of fixed elements of p in

V . We may describe this ideal explicitely


as follows. Let G = 5pecK[[,y,d---] , [= ([1
.... 'In )' y = (Yij)I_<i,j_<N where [ is a
basis of V . Then 3 consists of all
derivations of the form: Pf -- f(D( Yij y-S-,),
f-- f(D (V) CLN ZK[ i i,j lj By (3.16)
we have seen that Wo(G) c Im(J[). In what
follows we use the (analltlcally proved)
result (1.1) to "aproximate lm([?) from
above" more precisely to show that Im(J[?)
is contained in the kernel of a natural (Lie
algebra theoretically defined) map W(C) +
H2(L(U),L(U)). So fix an irreducible arline
algebraic K-group G (char K = 0) with
radical R and unlpotent radical U, pick a
maximal torus T of R and put r -- L(R), u =
L(U), t -- L(T). (3.23) We start with the
remark that W(C) naturally identifies with

Hom(u/[g,g] n u,r/u). indeed, identify both


L(G m) and L(G a) with the field K via the
identifications G a = SpecK[t], Grn =Spec
Kit,t-1]. Moreover write G/[G,C;] -- G (a) x
C (m) with C (a) a'vector group and G (m) a
torusl clearly the natural map U/U n [G,G] +
G (a) is an isomorphlsm while the map T +
G (m) is an isoseny. We have identifications
L(G(a)) Hom(L(G(a)), L(Ga))
Hom(G(a),Ga ) -- Xa(G) and L(G(m)) =
Hom(L(G(m)), L(Gm)) -- Xm(G))K. Hence
we have an identification: W(G) -Hom(Xm(G) K, Xa(G)) = Hom(L(G(m)) ,
L(G(a)) ) = = Hom(L(G(a)), L(G(m))) -Hom(u/u n [8,g],r/u) (3.24) Next let's note
that for p P(G) the image of JtVp in Hom(u/u
n [8,g], r/u) (still denoted so) has the

following particularly simple description:


upon letting l[:r + flu be the natural
projection we claim that JVp = l[(px) for all
x = u where c denotes the image of x in u/u n
[g,g: In particular Vp = 0 If and only if pu c u
(this together with (3.16) and (I. 1.16)
provide a purely algebraic proof of (l.lg)).
To check our claim note that under the
Identification in (3.23) we have the
following formula (d x) o .Vp = d(x'lp)) for
all character X:R/U-G m (where dx:L(R/U)L(Gm)= K is the tangent map of X, -1 and
simllarily d(,' lp) is its X PX Xa(G) is
viewed as an additive character U/U n
[G,G]- G a tangent map). Identifying now
L(U/U n [G,G]) (respectively L(R/U)) with a
subspace of I(U/U n [G,G]) (respectively
C(R/U) ) the above formula reads ^

(VpxXx) = (x- lpx) for all x L(U). But x(x


PX) coincides with the image of ( under the
map (R/U)((!)=C(U))C(T) PC(U)aC,0(T)
IET, (U) x ,K and our claim is proved. (3.2)
Now we have the following Lie algebra
theoretic construction. Let r be any Lie algebra, u an ideal in r containing [r,r] and s:
r/u r a Lie algebra section of the projection
r/u. For each f Hom(u/[r,r], r/u), the
bilinear alternating form b(f) -' uxu + u
defined by b(f)(x,y) = [sf,y] - [sf,x] for x,y
u (where , are the images of x,y in u/[r,r]) is
easily seen to be a 2-cocycle of u In u so we
can consider the linear map induced by b: 1:
Horn(u/Jr,r], r/u) + H2(u,u) (3.2) Coming
back to our specific situation where u =
L(U)) r = L(R) and noting that 57 Hom(u/u n
[g,g], r/u) is a sub, pace of Horn(u/Jr,r], r/u)

we get a map (which we still call lB: W(G)H2(u,u) by composing the map lB from
(3.2.s) with the identification isomorphism
W(G) = Hom(u/u n [g,g], r/u) from (3.23).
Then we have: (3.27) PROPOSITION. Let
char K = 0 and C be an irreducible affine
algebraic K-group. Then there is a complex
exact in the first two terms: 0 + P(C,fln) -*
P(C) J[V, W(G) -H2(L(U),L(U)) where U ts
the unlpotent radical of G. Proof. By (3.16)
the only thing to prove is that lm([V)c Ker I
So assume a = J, Vp W(G) fo' some p P(G).
By (l.l) K = K Kip] is a field of definition
for G. 'riting G = Go)Ko K (G o a Ko-grou
p) let's denote by p* the trivial lifting of p
from K to G. Moreover let l :r +r, e 2: r
+r be the projections onto u and t
respectively. Then by (3.2), (3.2) we have

(*) b(VpXx,y) = [e2Px,y] - [e2py,x] for all


x,y u. Projecting the equality p[x,y] = [px,y]
+ [x,py] on u and using the relation () we get
(.) elP[X,y] -- [elPx,y] + [x,elpy] + b([?
pXx,y) Now clearly p+ maps u into u and we
have (...) p#[x,y] = [p*x,y + [x,p*y]
Substracting (a,) from (,) and putting v = elP
- p eEndKU we get b([?pXX,y) -- v[x,y] [vx,y] - [x,vy] which shows that b(J[?p) is a
coboundary and we are done. (3.28)
REMARK. Note that in the abstract frame
(3.25) one can easily check that for any !
ellore(u/it,r], r/u) the form b(f) defines in
fact a (new) Lie algebra multiplication on ul
Coming back to our specific situation (3.2(;)
we see that for any p P(C) we get a "new
Lie algebra multiplication" b(J[ 9p):
uxu'+uon u. It would be interesting to

understand what infor- mation about p


carries this "new Lie algebra multiplication".
We end this section by recording some more
consequences of our results (for a
generalisation to the non-linear case, see (IV.
1.2)): (3.2) COROLLARY. Let char K = 0
and G be an irreducible arline algebraic Kgroup with radical R. Then: 1) The natural
restriction map P(C)/P(G,fin) +
P(R)/P(R,fin) is tnJective. 2) !f i: G +G' is
an isogeny and i#P(G ')- P(G) is the natural
lifting map, then
i (P(G',fin)) = P(G,/in) n i* (P(G')). 3) If the
center of G ts finite then P(G) = P(G,fin). ) If
Xm(G) -- I or Xa(G) = 0 then P(G) = Proof.
1) follows from (3.16) and the fact that the
map W(G)* W(lq) is injectire, cf. (3.21). t)

follows from (3.16). 2) follows from (:).It;)


and the fact that the map W(G')* W(G) is an
isomorphism, cf. (:).20). To check :)) start
with a preparation. Assume V is an Ndimensional E)-module. Then the coordinate
algebra )(GL(V)) of GL() has a natural
structure of Ftopf D-algebra define by
identifying C(GL(V)) with $(gl())[d -1]
where S = "symmetric algebra" and d
$N(gl()) is the "determinant". We claim
that C(GL(V)) is locally finite; indeed
S(gl(V) ) clearly is so and we are done by
noting that d is killed by P (to check this
replace K by some D-field extension of It
such that V splits so V will have a K-basis
contained in V D. Associated to this basis
there is a P-constant basis Xij of gl()l now
d is a polynomial in the Xij's with integer

coefficients so is killed by PI). Coming back


to our group G, let Ad: G+ GL(g), g = L(C)
be Its adjolnt representation. Using the
description of Ad in [H] p. 51 one checks
that Acl : C(GL(g))+ C(G) is a D(G)-algebra
map. Consequently if Z is the center of G,
((G/Z) is a locally finite D(G)-module
(being identified with ((GL(g))/Ker Ad*).
By assertion 2), ((G) must be locally finite
as a Kip]-module for all p P(G) and we are
done. EXAMPLE The example 2x) in (:).22)
can be arranged to provide an example of
arline algebraic K-group G for which: i)
P(G) -- P(G,fin). il) G has a positive
dimensional center. iii) Xm(G) 1. iv) Xa(G)
0. Indeed it is sufficient to choose 10 p I P2
where Pl: GLN* GL(1) is trivial, dim 1
) 0 P2: GLN* GL(V2) does not factor

through PGL N . We can rephrase part of our


results in terms of algebraic D-groups:
(3.30) COROLLARY. Let G be an
irreducible arline algebraic D-group (char K
arbitrary, D KIP]). Then 1) Assume dim P -l G solvable and the unipotent radical of G is
commutative. If the group like elements of
(P(G) are.P-constants then G is locally finite.
2) Assume char K -- 0. Then the radical R of
G is an algebraic D-subgroup of G. loreover
G is locally finite i! and only if R is so, if
and only if all group like elements of (P(C)
are P- constants. 3) Assume char K -- 0 and
i: G * G' is an isogeny of algebraic Dgroups. Then G is locally finite if and only if
G' is so. ) Assume char K -- 0 and the center
of G is finite. Then G is locally finite. )
Assume char K = 0 and either Xm(C) = I or

Xa(G) = 0. Then G is locally finite.


CHAPTER 3. COMMUTAllVE
ALGEBRAIC D-GROUPS In the first
section of this chapter we introduce the
"logarithmic Gauss-Yanin connection" and
the "total Gauss-llanin connection"
associated to it. In the second section we
prove a "duality theorem" laying that the
"total Gauss-Mania connection" on the "total
de Rham space" H)R(A) t of an abelian
variety A is isomorphlc as a D-module with
the continuous dual of the inverse limit of the
Lie algebras of the "relative projective
hulls" of A, viewed with its "adjolnt" Dmodule structure. This implies a precise
description of P(C) for any irreducible corn
m utative algebraic K-group C; note that K

P(C) need not be a field of definition for C!


11e also get from the above duality the
following regularity theorem: let C be an
irreducible algebraic D-group withC(C) = K
(D = K[P], P = DerkK , tr. dug. K/k < 41 then
L(C) is a regular D-module in Deligne's
sense. This together with a "descent
criterion" will be proved in section 3.
Everywhere in this,chapter K is
algebraically closed of characteristic zero
and if not otherwise specified P is any Lie
K/k-algebra and D = K[P]L 1. Logarithmic G
- Manin cmmectlen Let V be a smooth
projective K-variety. Recall from [Ka], [K0]
that there is a natural integrable connection v,
DerkK p Vp called the "Gauss-Manln"
connectionl here HR(V) is the first de Rham
cohomology space of V (see (1.1) for a quick

review o! this). Recall from lillY] that there


is a "m ultipllcative analogue" of HR(V)a :=
HR( ) which is an abellan group called
here HIDR(V)m , see (1.2)I we shall define
a K-linear map v: DerkK +OmgrmR(V) m,
Ve shall call the pair (?,9) the logarithmic
Gauss-Mania DerkK-connection on (H)R(V)
m , H)R(V)a); HIDR() a will be viewed
as an abelian Lie K-algebra on which HR(V)
m acts trivially. 61 (1.1) For convenience of
the reader and also for computational
purposes we recall briefly the construction
of ? ci. [K0]. Let = (Ui) l be an a/fine
covering o V, let C; ts = (C a ), o t c
ts=c;s(zO= H% l'v/K ) a io<11...<I s o'" s be
the standard "ech-de Rham" double complex
of V and let C= C() be the simple complex
associated to C. Recall that the differential

C:+ C 1 Is ilven by the formula a (fl) + (d/i,


fj - fl ), fie H(Ui,/ V) . while the
differential C 1+ C 2 Is given by the formula
a a (u i,xij) + (dr0 t,u j - i - dxij'Xjk - Xik +
xij) where m 1 H(Ui,/K ), Xlj y
definition, the l-st de Rham cohomology
space HR(V) (called here also HR(V) a) is
the l-st hypercohomology K-linear space of
the de Rham complex (called DR(V)a)=
Recall that HR(V) a = HI(c). Recall also
from [Dell] that the spectral sequence
ErrS(a) associated to C degenerates in E I. In
particular we have an exact sequence.'
(1.1.1) 0 ' H 2V/K) - H R(V)a (V) + 0 The
connection V is defined as follows. Letei
Der k (U i) be any lifting of p DerkK. Then V
p takes a class in HIDR()a represented by
(m i' xij) CI Into the class represented by

Here Li : H(UiP,/K)+ H(Uifi/K) is the


"Lie derivative" in the directlone i (recall
that H(Ui2 l V/}= 3/32 where 3=
Ker(/(Ui))d7(Ui)* i(Ui)) so H(Ui /K)hasa
natural structure of K)i].module then Ltd. is
by definition the multiplication with 1 in this
module structure)l moreover is the con'action
between K-derivations and l-forms with
values regular functions. (1.2) Consider now
the complex DR(V) m of abellan sheaves l
dlo_., l 2 ,,, '"V/K " V/K and let HiDR(V)m
denote the l-st hypercohomology group of
this complex cf. for instance [MM] I). 31. So
if C Is the direct limit (over all coverlngs,)
of the double complexes C() associated to
DR(Y) m and and if is the simple complex
associated to the double complex

62 C" then we have nl Now let ErrS(m) be


the spectral sequence associated to C; it does
not degenerate in E l like in the "additive
case" but one can easily check that El(m) =
Pic(V) and st E S(m) = !..I (nV/K) for t_) l,
s_>0 Consequently E201(m) = PicT(V) (see
rG]) and by degeneration in E l of ES(a),
E220(m) = H("2V/K ) It is known
(essentially by [MM]) that dl(m): El(m) '*'
E220(m) is the zero map. For convenience
we check this; it is sufficient to check that
dl(m) is a morphism of algebraic groups
(because PicT(V) Is an extension of a finite
group by an abelian variety while H(i2/K)
is an algebraic vector group so no algebraic
nonzero homomorphJsm can exist between
them). To check algebraicJty of dl(m)
consider for any noetherian reduced K-

algebra R the complex dlog, nl d 2 I +( te R


"V e R/R ---'"V eR/R '" Let ctS(m) R be the
associated double complex and ErtS(m)R the
corresponding spectral sequence, Exactly as
above we have: EI(m)R = Plc(Ve R) =H (2
VR/R)= s t (where we identified H (lV/K)
with its associated algebraic vector group
Spec(S(Hs(t/K))). Clearly dl(m)R: Pic(V
R) + Hl(a R/R) factors through PJc(V e
R)/Pic R = PiCV/K(I ) = (R-points of
Pic/K}. Moreover, since R is reduced the
kernel of Pic/K(R ) HI(y ( R/i ) equals
Pic/K(R ). Analogously we claim that
dl(m)R EI(m)R E0(m)R 0 2 ' + = H v
mR/R ) factors through EI(m)R/Pic R =
Pic//K(R). Indeed El(m) R is a quotient of {a
C10(m)R + C01(m)R; da C20(m)R} 01 and
d 2 (m) R is defined by takin the class o a

into the cl of da. ow J a = (u W.) then -1 -1 '


cocycle and :- ,=y, dy,,. So l a da=(dui' mj'
ui-Yij dYij' YjkYikYlj ) so (Yij) s a ; 1; u
comes from Pic R we have (after suitably
refining the covering) that Ylj = ulu vii
suitable ul's with vii belongin 8 to some ring
of quotients of RI we get (becau dvjj = 0)
hce dj = d(u i + u]duj) But [ + ulduj tch
together to give a 81obal m the c]ass o[ da
vanishes and our claim Js proved. We get
group momphisms behaving functoriaUy in R
1 reded, hence we get a morphism o
algebraic K-grou Pica/K+ H(/K ) (cause
th these grou are retired). 5o the claim that
dl(m)= 0 follows. !n partJiar we get an exact
sequence Now we define p Homgr(HR(V)m,
H(V) a) for p DkK Jet =(U i) be an afine
covering of V let e I Derk(U i) Uftings of p

Jet C() be the simple mmp!ex ociated to the


ubJe mmplex C() and define a group
homomorphism C()+ C() by the formula (
i,Yij)+ (Lieei i,(ej - e 1) Kuj + y[jieiYij) m i
H(Ui'/K )' Yij O"(Ui n Uj). One chks by
hand compation that this morphism is welldefined, sses to homm 1ogy and agrees with
"refining "so we e a Uoup homomphlsm [Vp'
HR(V) m + HR(V> a which dends K-linearly
on p P. The ir (VEV) defines a logarithmic
DerkK.conntJon on (HR(V)m, R(V)a) called
In what follows the logarithmic Gauss-anin
connection. e will prove later that It is
lntegrable f V an abelJan variety (i.e. that
(0.16.2) holds).
(1.]) REMARK. Recall that if K = (: then
HR( ) = H l(van, (:) and HR(V) m = HI(V

an, (:*). Moreover these two groups can be


interpreted in term of differentials of the
second raspact]rely of the third kind on , cf.
for instances N. Katz, invent. Math. ]8, 1-2
(]972) p. 108. We won't need this
interpretation in what follows. (!.4)
REMARK. From the very definitions we
have (in notations of (1.1) and (1.2)) that for
any p c DerkK the following diagram is
commutatlve H=tV/K ),- , p (1.)
REMARK. it follows from [K0], [Ka] (and
indeed from the very definitions) that for any
pc P the K-linear map Ho( tv/K ) V
coincides with the cup product with the
Koalaira-Spencer class p(p)c HI(T^ )
(where p: DerkK+ HI(T^) denotes the
Koalaira-Spencer map). (1.6) In proving
lntegrability of the logarithmic Gauss-Manin

connection (and also in formulating our


"duality theorem" In the next section) we are
led to make a certain construction of linear
algebra whose abstract version we now
describe. Let's consider the following data:
an abelian group Hrn, a K-linear space H a
and a K-linear space H o equipped with a
group homomorphism Jm.; Ho+ Hm and Klinear map ja Ho+ H a (both jm,Ja injectire).
By a total space for these data we will mean
a K-linear space H t equipped with a group
homomorphism i m: Hm+ H t and with a Klinear map 1 a Ha+ H t such that the
following diagram is commutative Jm H o m
H a --' H t a and such that for any other Klinear space H', any group homomorphism 1'
: Hm+ H' and any m K-linear 'map i' H' ' ' a
Ha+ with I m Jm = ia. Ja there is a uniqbe K-

linear map f Ht+ H' such that f i m i m and f i


a-" Ot course (H t, ir, i a) is then unique up to
a canonical 65 isomorphisml it is a sort of
"fibred sum" of H m and H a over H o. total
space always exists? it is constructed as
follows. First put Hm, K = (H m
<zK)/Ker(Ho Z K+ HoKK = H o) and note
that we have an exact sequence of K-linear
spaces 0+ Ho+ Hm,K+ (Hm/Ho) K+ 0 Then
define H t as the cokernel of the map below
0+ Ho--6 Hm, KHa- Ht+ 0 where (x) = (x,x) (we view Jm' Ja as inclusions!). Note that
i a is always injectire and we have dlmKH t
= dlmKH a + rank(Hm/H o) (1.7) Now
assume in situation above that we dispose of
a logarithmic P-connection (.) on (Hm,Ha)
where H a is viewed as an ahellan Lie
algebra on which H m acts trivially, such that

for any pc P we have a cornmurat]re diagram


Jm H H o m Ja Ha Ha P Then we can define
in a canonical way a P-connection on the Klinear space (H m K)oH a by the formula V
p(y e ,,x) = (y (p)), ( ? py) , + pX), p P, y
H m, x c H a, . K . This P-connection
induces a P-connection '. P+ HOmk(HtH t)
which will be called the total connection
associated to our logarithmic connection
(relative to H o !). The basic' (trivial) fact
about this construction is the following: (1.S)
LEMMA. In notations above, the logarithmic
P-connection (VV) on (HmH a) is integrahie
if and only if the associated total connection
? on H t is Integrahie. (1.9) Coming back to
our logarithmic Gauss-Mantn connection
(with H m = HR(V)m, H a = H R(V)a, H
= H GV/K), see Remark (l.t)) we may form

the total de Rham space Ht = HR(V) t


viewed with the ota] Gauss-Mania
connection on it cf. (1.7). This object will
play a key role in the next section.
2. Duality iimmern (2.1) Let C be an
irreducible commutative algebraic K-group,
let B its maximum linear connected subgroup
(which will be called the linear part of C)
and consider A = C/B (which will be called
the abelian part of C). By [$e] there are
naturally associated map m Xm(B)+
Plc(A) a .' Xa(B)+ HI(A ) with m a group
homomorphism and a a K-linear map.
(Recall that they are defined as foUows: one
takes a covering (Ul) i of A and sections s i '.
Ui- C of the projection C * A; then for X
Xm(B) and [ Xa(B) we let re(X)

Pic(A) be represented by the cocycle X


(sj - s i) and a([) HI((? A) be represented by
the cocycle [ (sj - sl). lecall also that the
maps (re,a) uniquely determine the extension
class of C as an extension of A by Finally
recall that given an abelian variety A, an
irTeduclbl.commutatlve algebraic group B
and maps (re,a) as above there exists an
extension C of A by B whose "associatecP'
maps are (m,a). Consider $(C) m -- Im(m)c
Pic(A), $(C} a = Im(a)c Hl(( A) and, with
1I a: Hi(A)a + Hi(( A) and E ; HE(A) m*
Pic(A) as in (1.l) and (1.2), define m
SDR(C) m -- l(S(C)m)C HR(A) m SDR(C) a
= nl(s(C)a o 1 Both 5DR(C)m and SDR(C)a
contain H (fiA/K) and by (1.(;) we may
consider the total space ,,Ote l 5DR(C)t of
the data consisting of 5DR(C)m, 5DR(C) a

and the inclusions Jm o I ' 5DR(C)a. Clearly


5DR(C) t is a K-linear subspace of the total
de + 5DR(C)m' Ja" H (A/K) Rham space
HR(A) t of (1.9). One more definition= we
say that C is a relative prt)jective hull of A ff
the map m is injectlye and the map a is an
isomorphism (the terminology Is motivated
by 5erre's consideration of projeotive hulls
of abelian varieties In.' "Groupes
Proalgebriques", Publ. Math. IHE5, 7 (1960)
our "relative projectire hulls" are
'truncatlons" of the projective hull of A,
enjoying properties similar to those of a
projectlye hull but relative to a given fixed
subgroup Pic(A). We will not go into
details since they are irrelevant for our
purposes). As we shall see in (2.5) below,
(!I(C) -- K if C is a relative projectlye hull

of A. Here is our main result= (2.2)


THEOREM. Fix a Lie K/k-algebra P and D - K[P]. 1) Let A be an abellan K-variety.
Then the logarithmic Causs-Manln DerkKConnection (1.2) is integrahie. In particular
the total de Rham space H)R(A) t has an
induced structure of D-module. 2) Let C be
an irreducible commutative algebraic Kgroup with ((C)= K and abeltan Dart A. Then
C has at most one structure of algebraic Dgroup. It has a structure of algebraic 67 Dgroup if and only if SDE(C)t Is a Dsubmodule of HR(A) t. In particular if C is a
relative projectire hull of A, then C has a
unique structure of algebraic D-group. 3) Let
C be an irreducible algebraic D-group with
(C)= K. Then there is an isomorphism of Dmodules 5DR(C) t L(C) where SDR(C) t

is a D-module via the total Gauss-Manin


connection, (cf. assertion 2) above) and L(C)
is a D-module via adjolnt connection (cf.
(I. 1.10)). a)Let Ci, C 2 be irreducible
algebraic D-groups with C(Ci)= K, i= 1,2.
Then any morphism of algebraic K-groups
Ci+ C 2 is automatically a morphism of
algebraic D-groups. Moreover If both C i are
relative projectlye hulls o A then we have a
commutative diagram 5DR(C2) t : L(C2)
5DR(CI) ! L(Cl) Consequently HR(A) t is
isomorphic as a D-module with the
continuous dual of L(A hull) == lira L(C)
where C runs through the set of relaiive
projection hulls of A (note that these C's do
not form a filtered projective system but their
Lie algebras do I). (2.3) REMARK. Recall
that for any abelian K-variety A there exists

a "universal extension" of A by a vector


group called E(A) (see [MM]); E(A) is
simply the extension.of A by . ._dlmA,
KdimA HI(i27A). Clearly E(A) is a cdim A
defined by any linear lsomorphism 2a a j = a
relative projectlye hull of A (in fact it is the
"smallest" one i.e. a quotient of any relative I
projectire hull). Note that $DR(E(A))m =
HoA/K) and $DR(E(A))a--HR(A) a -HR(A) so $DE(E(A))t = HR(A). Then the
above theorem says that E(A) has a unique
structure of algebraic D-group and we have
an isomorphism of D-modules HR(A)=
L(E(A)) the left hand side being endowed
with the usual Causs-Manin connection,
(1.1). Now (.) is easily seen to be a
consequence of the crystalline theory
developed In ("Crothendleck's Theorem")

and of the duality theorem proved in the last


chapter of [BBM]. But proving Theorem
(2.2) along the linear oi [MM] and [BBM]
seems to us much harder. In any case our
proof of Theorem (2.2) (and hence of (.)) is
much less sophlstJcated Jt is Just a
manipulation of ech cocycles on a Zarlski
open covering of A (no crystalline
background being necessary). It has the
advantage of constructing explicitely the
lifting of the operators pep to E(A) and
indeed to any relative projectire hull o A I In
particular the D-module
lsomorphlsm SDR(C) T L(C) from (2.2)
should be viewed as a "dual" generalisation
of the D-module counterpart o!
Grothendleck's theorem cited above. It also

suggests that the following might hold (?)


Any relative projectlye hull of an abelJan
variety has a crystalline nature. (??)For any
such relative projective hull C the
isomorphism SDR(C)t a L(C) o is "induced"
be the crystalline structures. Note also that
our proof of (2.2) is mainly based on certain
cohomologicai properties of A making very
little use of the group law of A. This permits
to pass from A to other varieties; this will be
explained in a subsequent paper (and makes
an essential diffeence between our approach
and that in [MM]). (2.e) V/e make a
preparation. Assume C is as in (2.1) and
write B -- B m x B a, B m a torus of
dimension M, B a an algebraic vector group
o! dimension N. The projections B + B m
and B + B a induce extensions of A by B m

and B a respectively (call them C m and C a)


which are described by the maps m = (m,0)
and a = (0,a) respectively (in fact C m = C/B
a and C a = C/BIn). l! E' is a linear subspace
of E := Xa(B) then we have E' = Xa(B') for a
well defined quotient B' of B a and the map
B a + .B induces an extension C' of A by B
(C' -- Ca/Ker(B a + B9) whose defining
element a' Hom(E',H (A)) is the
restriction of a to E'. Clearly, if E -- E'
E", upon letting 1', C": a" be the
corresponding objects for E", the natural
map C + C'XAC"XAC m Js an isomorphism.
From now on we fix E'--Ker(a:E + HI(J)A))
and fix E" an arbitrary complement of E' in
E. Since a' = 0 we have C' = B' x A so we
may write C=B'xC 1 C 1 = C" x ACre It wiU
be useful to recall how one obtains C via a

giueing procedure starting from the maps


(re,a). Start by lifting the map m to a
homomorphlsm (mij)ij from r. == Xm(B ) to
the ech group Cl(?,) = itH(U. n U.,I.)
corresponding to a covering = (Ui) i, U i =
SpecA i. /% i I J Slmilarlly lift a to a{near
map (a .),. from E = X.(B) to CI(,,(). V/rlte U
i. = Uin Uj and put lj ,j Z Aij =((Uij). Let'R
be any ring. Letting R[ ] be the group Ralgebra on and R[E] be the symmetric Ralgebra of E and putting R[T',E] = R[T*]R]
[E] we have that C is obtained by giuelng
Spec(Ai[r-,E]) via the Atj - isomorphisms ij:
Atj[T"E]' Alj[r*'E] (2. a. i) eli(X) -mijCX)X for X r r. (2.a.2) ij() = aij() for
all E Clearly C m is described by glueing
Spec(Ai[r']) via (2..1) while C*, C" are
described by gluelng Spec (Ai[F']), Spec

(Ai["]) via (2..2). (2.) LEMMA. In the


above notations we have: 69 1) (C) = K if
and only if the maps m and a are injectlye. 2)
Xa(C) = 0. REMARK. Condition {(C)= K is
equivalent to saying that C has no non-trivial
arline quotient and also equivalent to saying
that C has no non-trivial linear
representation (cf. [Ro] [DG]). ProoJ. Note
first that any additive character C i G a
vanishes on B m hence factors through C/Bm
= C" so 2) follows from i)l let's prove 1).
Assume first that a is not Jnjective so in
notations of (2. a) B' is non-trivial which
implies C(C) K in view of the
decomposition C = B' x C . Similarily, if m
is not injectlye, put r. l = Ker (m). Then
letting T l = Spec (K[r.i]) we have a
surjective homomorphism B T 1 to which

there corresponds a surjective


homomorphism C + C i where C! is a trivial
extension of A by T 1. This implies again
d7(C) K. Conversely, assume m and a are
injective and let's prove that )(C)= K. We
dispose of a commutative diagram with exact
rows= O Bo B* B/Bo* 0 0 C O C Spec(C) 0
where ((C o) = K and B o is the linear part
of C o (cf. [DG], p. 358). From the exact
sequence 0 Ker(u) Co/B C/B Coker (u) 0
and from the fact that Ker (u) and Coker (u)
are arline while A = Co/B o and A = C/B
are abeltan varieties we get that A o A is an
lsogeny. Now look at the commutative
squares fa, Xa(Bo) f Xa(B ) Xm(B ) m
Xrn(Bo ) H]({A)'--HI(IAo ) Plc(A)
isogeny ' Pic(Ao) Since a and m are
injective while fa and fm are surjectJve it

follows that fa and fm are isomorphisms


hence B -- B which implies Spec )(C) =
Spec K and we are done (2.6) LEMMA. Let
G be an irreducible algebraic K-group with
((G) = K, Then the map B:P(C)DerkK is
injectlye and DarkG = L(G)$P(G), In
particular there is at most one structure of
algebraic D-grou p on G. ProoJ. By [Ro] C
is commutative. By (L 1.9) ker(8) = P(C/K)
= Xa(C)iIJL(G ) = 0. To prove that DerkC _-L(C)P(C;), let p DerkK since d)(C)= K
we have pK c K. Let m e be the
7O maxim-' idea, of C,e and let e, PC,e + K - PC,e/me be the cou.it as usual. Then epp:G,e +K is a K-e-derivation hence it is of
the form ev for a unique K-derivation v: G +
G' v L(G). The equality (p - v) = sws that

(p - vXme) m e. e claim that d = p - v


P(G). Indeed he map p G + GxG is an
isomorphism (because p = where t G x G +
G x G (xy) = (xxy) and 2 = G x G + G, 2(x,y)
= y and e Knneth mula to get p2.GxG = G
which implies P- GxG = G )' Then Since the
above derivation (identified with a vector
field in L(G)) vanish at e (recall d(me)C me)
it must be zero so =(di + ld)p Similgily one
prov Sd = dS nsuently de P(G) and we are
ne. (2.7) LEMMA. Let Gi, G 2 be
irreducible gebraic D-grou with (G i) = K, i
= 1,2. Then (with notations rom (I. 2.1)) Pg.
Let fGieG 2 a morphism og algebraic K-grou
and pg P. Then the mphism pg - fp = G2 +
gG1 is a K-g-derivation (where we still
denoted by p the rivatio induced on Gi and
G2 respectively) hence 1ongs to . (c2 ) e

(c2, . ci). L(%) H(C i, Ci) = Menver


upon identHylng p - fp with a vector field on
G2, It must vanish at e G 2 so it vanish
everywhere and we are done. (g) In order to
rorm comarations with coccles, assume
again we e In the situation o (2.) so that we
dispose o maps 71 lifting the maps m and a
which define a given commutative algebraic
K-group C. Moreover r = aij(r ) choose a
basis Xl , ... ,X M of the Z-module r., a Kbasis l '" 'N of E and define i and miSj =
mij(Xs). Upon refining ?,we may assume that
the elements mij lift to some elements
notations being as in (1,2); indeed this is
possible because IIm = HR(A) m + Plc(A)
is surjective (1.2). On the other hand aj
automatically lift to elements (, ai), K er (Cal
(/,). Ca2()) so recall that the following

relations hold: r r r (2'8'1)a datj = nj ' ni m s


-Idms _- s_ s (2'8'1)m ( ij ) ij 0j 0 i Fcr
convenience, let's call a collection (a , mij)
as above a system of adapted cocycles for C.
Then C is obtained by gluelng the affine
schemes SpecAi[{l, ... ,q,Xl,Xi 1, ... ,XM,y
l) (A i = (Ui)) via the isomorphLsm % r
(2.s.2) a %%)-- r + (2.8.2) m %(y) = mjy
(2.) LEMMA. Assume C is a relative
projectlye' hull of A and assume (a? mj)Is a
system of adapted cocycles for C (2.g). Then
for any p e DorkK there exist' vector fields v
k H(TA ), 1 _< k _< N, liftings
DerkA 1 of p to ^i = ((Ui) .and regular
functions r s $ 3k' ' jk' "i =)(Oi ) satisfying
the following system of equations N (2.9.1)
rl a[jvr -- ej- 1 for all i,j (2.9.2)a r r r vkaij
= ek - Ck for all l,j and 1 _< k, r _( n s -I s s

s for alli, jandl <k<N, I <s<M (2'9'2)m (mij)


vkmij = Pjk ' Pik .... Nv r k r_ r (2'9'3)a airj
+ kl kaij = c a l for all i,j and I _< r _< N
,ms,-I m s N s a k s foraUi,jand I <s<M
72 Proof. Since the cup product u: HI(A) e
H(T A) + HI(T A) is an isomorphism one
can find v l, ... ,v n H(TA ) and liftings
of p to derivations of ((U i) such that (2.9. l)
holds. By degeneration in E 1 of E"(a), the
map d: HI(7 A) +HI(gJA/K ) is zero. Hence
the map HI(A)- HI((^) defined by applying v
k to cocycies is zero, in particular there ~r
e(Ui) satisfying exist ik r -_ r r for all l,j,k,r
vkaij jk' ik Similarily, since the map dlog:
.l(O I) + "l(iI/K) sends the classes of
(mj)into zero, one Can find iik (Ui) such that
s,-I s Now recall that we dispose of relations

(2.g.l) a and (2.8.1) m. To conclude the proof


it would be sufficient to check that r v-r ak -s
ak are cocycles. Now we know from (1.1)
and (1.2) that ' + ar (( ) wi +(mill mij)ij are
cocycles so what we have to check is that
are cocycles. Now taking contraction with v
k in (2,8.1) a we get Vk' Vk = vkaij = jk Oik
ik)l glue together to give a constt eK.
SimilarHy (2.8.1) m implies that ml Vk - Pik
= Ck e get which Is a cocycle and simUarily
- a k k s - ak= caikj ' = ffaijvkmi - Uik ij
which is also a cocycle. Our lemma is
proved. (2.10) [EMMA. Assume C is any
irreducible commutative algebraic K-group
(aijmij) is a 73 system of adapted cocycles
for C and assume we have a solution _ r r s
s, Vk# i jkj 'Pjk'Pj ) of the system (2.9.1)(2.9.3). Define for each i a derivation Pi

Derk(Ai[E: ]) by the formula Pt=O i ?kVk .!


(r r. r, 0 s Then the Pi'S glue together to give
a derivation PG DerkG' Preef. An easy
computation using (2.8.2). (2.11)
COROLLARY. Let C be a relative projective
hull of A. Then the map : P(C)* DerkK is an
isomorphism. In particular given any D, we
have that C has a unique structure of
algebraic D-group. Proof. By (2.5) I!)(C) =
K hence by (2.6) 8 is injective. To check
surjectivity of ) take pc DerkK. By (2.9) and
(2.10), p lifts to some derivation PC" DerkG'
By (2.6) again l+ Pl with pie L(G), pie P(G);
clearly Pl lifts p and we are done. PG = PG
(2.12) LEMMA. Let C be an irreducible
commutative algebraic K-group, ((C)= K and
r s (aij,mij) be a system of adapted cocycles
for C. Consider the K-linear space S of all

systems *r^s ^r ^ Oj,pj)l_<r_<N,l_<s_< M


such that 'e L(A)= H(T^) and aj, pe 0(U i)
satisfy the following system.. ^ r ^r ^r (2.12)
a eaij = aj - a i , I _< r _< N (2.12) m (mtj)imj ^s ^s =uj '"i' l <_s_< Moreover, for any
such {'0, ,,r& :s) S consider the
derivations j Der(Aj[E, ]) defined by J
the formula =%. v&r Then: l) The j's glue
together to give a derivation p E L(C) (we
write ) = ('e, .r,..s}); 2) The map assigning to
each element of S the corresponding
derivation in L(C) is a K-linear
isomorphlsm. Proof. 1) is again an
immediate computatlon using (2.g.2). 2)
follows by noting that the map S L(C) dimKs
= dimKL(A) + N + M = dim C = dimKL(C).
is injectlye and that (2.13) To formulate the
next Lemma let's make the following

definition. Assume P is a Lie K/k-algebra


and V, W are two finite dimensional K-linear
spaces with P-connections. By a
74 p-duality between V and W we will mean
a bilinear non-degenerate form <,):VxW K
such that p < x, > -- <px,y> + (x,py> for all
pc P, x c V, y W. We will speak about
orthogonals in the usual sense. Note that a Klinear subspace 1 of is stable under P if
and only if its orthogonal [ W is stable
under P, Moreover the connection on V is
integrable if and only if the connection on W
is so, Here is our main step in proving
Theorem (2.2); (2.111) LEMMA. Assume C
is a relative projective hull of A, P is any Lie
K/k-algebra and D = K[P]. Then there is a Pduality <, > = SDR(C)t x L(C) K where

SDR(C)t has a P-connection induced from


that of H!(A) t (of. (1.g)) and L(C) has a Pconnection defined by the adjolnt connection
(cf. (l.l.10) where C is viewed with its
unique structure of algebraic D-group, cf.
(2.11)). Moreover, for any unipotent linear
algebraic subgroup B l of C we have L(Bi )1
= SDR(C/Bi)t hence we have an induced Pduality <, > .' SDR(C/Bi) t x L(C/B 1) + K r
s Proof. Consider a system of adapted
cocycles (ail,mi) for C as in (2.8) from
where we borrow notations. We also use
notations from (l.l), (1.2) Le ]m() be the
subgroup of Zlm(?Z) := = Ker(C()' C2m())
of those elements (t i,yij ) C lm(tO such
that n = (Fl(m.S.) S)z.z7 l (2' lt' l)m Yij s U I
1 for some integers n s and some zj ((Uj).
Clearly the map ]m() SD](C) m ls. surjective

(due to our assumption that our cocycles are


"adapted"). On the other its kernel is easily
seen to coincide with B()= Im(C(?Z)'
C())so after all we have SDR(C) m = 1
()/Blm(9A) m Of course since the map a .
Xa(B) HI(A) was assumed to be an
isomorphlsm we have SDR(C)a--H)R(A)a-Zla()/Bla(9) where Zat(l,) = Ker(C 1(20 *
C2()), B l(?z) = Im(C(Z) - C(?)). We will
define a Z- bilinear map a a a 75 Let (t l,yij)'
'7Z), (tli,xij). Zla(?)and *note by CO,;[,? .
L<C). derivation defined M in (2.12). e have
= Xa + uj- u i (2'la' l)a xij k ' f someXk K
and uj (Uj). Tn define (using tatls from
(2.1.1) a and (2.1*.l)m) I k K J O course we
must ove that the right hand side o the above
equality is independent o[ j and nce beJonss
to K. But indeed uj - ui =(uj ' u[) =(YjJdYiJ)

= = + zj 0zj Consequently Slmilarly we


5ummJn 8 up the lt two equalitas we get that
the right hand side member o (2.1.2) is
]ndendent o j required. ]t s easy to check that
our Z-bJlinear map nduces a non-degenate
K-bilinear map (, >: SDR(C) t x L(C) + K it
s so ey to check that L(B] = SDR(C/]) t y
Jtent B] c C: Jrst one shows that our
definition o <, > does t dend on choosing the
basis 1 .. 'N o Xa(B) in (2.8) next we
choose a basis with the prorty that Xa(B/ ])
admits 1''" 'N (Ni N) as a sis. Snce
dimensions o L(i and SDR(C/]) t are equal Jt
Js sfJcient to chck that (SDR(C/]) t, L(Bi)> =
0 Whi follows rom definitions using our
special basis. k remains to prove that <, > is
a P-du]ty. This can be done by a tedious (but
not Stralght[orwd) computation with

ocyclas we 8tve t outline o this mputat]on


sJn we [elt it would have en unfair t0 just
"leave it to the reader". Pick any p p
choose a solution
76 ^ r r s sx Vk Ja Jk ai'pjk'pjl of the system
(2.9.1)-(2.9.3) with the 9j's lifting (p)e
DerkK and pick closed l-forms 1 and m j
satisfyin8 k k (mi)-ldm. = s (2.1a.3) da =j -j,
j- mj The whaz we must prove is that the
expression s r r r * equals the sum o (2.1.5)
and (2.1g.6) below= where { } was us to
denote closes in SDE(C) t and we made t
obvious abrevJations e = ( ), .... Now (2.
Jb.b) equals Next (2.1a.5) becomes s . r s
(2.1q.8) (( 0,0,Li.u j + Lj, ej -ei)( j j) +eiaij
+ (mij) J To compute (2.1.6) note irst that
since Derk = L(C) P(G), c. (2.6) we have

,as P } = (Lie bracket in DkC) where PG is


given by the ormula in (2.10) and p Is
defined by the ormula in (2.12). A Ion[
computation yields: [pCp]={[6j ] k e rk r so
(2.1q.6) com Finally let k e K and uj (Uj)
sh that s r r , s,-I ems Xka +uj u i (2. le. 10)
ej-ei)(u j+j)+01aij+[mlj ilj = ' Then by the
very definition (2.1.8) comes j k " ' 77 By
well known properties of Lie derivative we
have ,, ,, +} j[(Liee ) eje ) -- [j ] .) for any lform . In view of this, in order to prove that
(2.1.7) equals (2.1q.9) + (2.1.il) it is s r
sufficient to check the following (with p = e0
j + j): (2. l.12) (vk'- jk-Pjk-Xkxj uj) k Now
introducing (2.9. l), (2.9.3)a, (2.9.3) m Jn
(2.1.J0) we get k Exactly as in the proof of
(2.9) we get that s s Vkj-Pjk eK vkTl r -Ojke
K hence we set (usJn the act that the

cohomoJoy classes o a orm a sJs in HJ(A))


that= Vk' jk'Pj'Xk = uj-a s ui r pK - pj = -o
i - which clearly implies (2. t q. 12) and our
Jemma is proved. (2. J) Let's ss to t proo of
the dualJt theorem (2.2). By (2.]q) or an
relative projectire huh C O A we have an
isomorphJsm of K-linear spaces with Pconnection 5D(C)t L(C) . Assume now that
Cj and C 2 e two relative ojtJve hulls o sa
that c[ is bier than C 2 J 5(C1) m eq contains
S(C2) m a natural Lie aJ[ebra map L(Ci)
L(C2) defined as oJlows. Choose an integer
n I such that nS(C2)mC S(Cj) m (as
sub[roups PJc(A)) and Jet C 3 be the
relative projectire huh e A for which S(C3)
m = nS(C2) m. Tn there is an [so[en C2+ C a
Lie algebra map L(Cj) + L(C3) L(C2) one
can easily check that this map does not dend

on the choice o the inte[er n. Moreover one


checks that the induced square o sces with Pconnection SDE(C]) t' L(C )o SDE(C2) t
(C2) is commutatJve. PsJn to direct limit
we = = ? [(c) o = c c
78 Since the P-connection on L(A hull) is
obviously integrahie o will be the Pconnection on HDR(A)t which proves
assertion 1) in (2.2). Next let C be any
irreducible commutative algebraic K-group
with abelJan part A and with C9(C) = K.
Then by (2.5) the maps m: Xm(B)* Pic(A)
and Xa(B)+ Hl(7 A) (B = linear part of C)
are injectlye so if we let C 1 be the relative
projective hull of A for which S(Ci) m -S(C) m there will be a surjective morphism
Ci* C whose kernel B I is unipotent.

Consider the P-duality from (2.10) < , > .'


SDR(Ci)t x L(Ci)+ K By (2.1) SDR(C)t =
SDR(Ci/Bi)t=L(Bi )1 so SDR(C)t is a Dsubmodule of SDR(Ci)t (equivalently of
HR(A) t) if and only if L(B 1) is a Dsubmodule of L(Ci). By (1.1.16) the latter
happens if and only if B l is an algebraic Dsubgroup of C 1 which by (I. 1.17) happens
if and only if C has an algebraic D-group
structure such that Ci* C is a D-map. By
(2.7) and (2.11) the latter condition is
equivalent to C having an algebraic D-group
structure. This together with (2.6) and (2.11)
proves asertion 2) in Theorem (2.2) as well
as the assertlon 5). Assertion 4) follows
from (2.7) and from the above discussion
involving L(Ahull). Our Theorem (2.2) is
proved. Now we pass to our applications.

(2.16) For any irreducible commutative


algebraic K-group C we let DerSk(C)K be
the Lie K/k-subalgebra of DerkK consisting
of all derivations p DerkK for which %
(5DR(C)m) c SDR(C)a V p(SDR(C)a) c
SDR(C)a where (JLV,?) is the logarithmic
DerkK-connectlon on (HR(A)m,HR(A) a) cf.
(I.). Equivalently Der(C)K is the Lie K/ksubalgebra of P = DerkK consisting of all p
P which send SDR(C) t into Itself (where
we view SDR(C)t as a subspace of the Dmodule H)R(A) t, D = KiP])- (2.17)
COROLLARY. Let C be an irreducible
commutative algebraic K-group. Then there
is a split exact sequence of Lie K/k-algebras
a (C)K * 0 0 Xa(C)L(C)' P(C) Der Proof.
Let as in (2.t) E'= Ker(a: f ' HI(iA)) and E" c
E be a complement of E' in 15. Moreover, let

r., = Ker(m: r. + Pic(A)), Sm= lm(m: r..


PicO(A)), choose a basis of theZ-module
Sm/TorS(Sm) , lift it to a subset of r. and r.,,
be the subgroup generated in r. by this subset;
hence Z' nr.,, = 0, [Z: :,OZ"] < eo and the
restriction of m to :" is injectlye. The
inclusion r.,r.,, c g induces an isogeny B m =
Spec K[r-] ' Bn x B'i where Bn = Spec
KIT"]. Hence we get an isogeny: where B' -SpecKlE'l, B"= SpecK[l"]. To this isogeny
there corresponds an lsogeny C- C l 7g = B'
x B' x C 2 and the maps m 2 and a 2
corresponding to C 2 are injectire. By over
A where C l a m (2.5) (C 2) = K. Clearly
SDR(Ci) t = SDR(C2) t. By (1.1.19) we
have Im(P(Cl)- DerkK) = = Im(P(C2)DerkK) and there is a Lie K/k-algebra map
from the latter space to P(C l) (use the S(C

2) fact that Bn x B is defined over QY). By


(2.2) P(C 2) Der k K so we get a split exact
sequence of Lie K/k-algebras S(C l ) 0+
P(Ci/K) P(C l) 3 Der k K + 0 Finally note
that SDR(C) t -- SDR(C l)t and we conclude
by (LI.21). (2.18) COROLLARY. Let C be an
irreducible commutative algebraic K-group,
P a.Lie K/k-algebra and D: KIP]. Then C has
a structure of algebralc D-group If and only
if a (p) Der(C)K for all p z P. If 9(C) -- K,
this structure is unique and is given by the
formula in (2.10). (2.19) REMARK. One can
give an analytic interpretation of the unique
structure of algebraic D-group on the
universal extension 5(A) of an abelian
variety A (cf. (2.3)) in case k: C and A -- x X
Spec K where L. X is an abelJan scheme and
K is the algebraic closure of k(X). Indeed in

this cae (upon replacing X by an etale open


set of it) we get that E(A) -- x XSpecK
where + X is analytically (but not
algebraically) isomorphic over X an with an
analytically trivial bundle with fibre (C)
2dim (A). Then any analytic vector field v on
X lifts to an analytic vector field on ()an
which agrees with multiplication and inverse
maps. The remarkable fact is that If v is
algebraic then v is also algebraic this is of
course a corollary of (2.2) and indeed of
GroethendJeck's theorem []. Our method has
the advantage o! giving a formula for in
terms of v cf. (2.9) and (2.10). (2.20) /e
close by illustrating (2.18) in the simplest
case. EXAMPLE Let A be an elliptic curve
over K, with j-invariant j z K, let C -- E(A)
be Its universal extension and let p t DerkK ,

pj O. Ve will indicate what is the explicit


lifting of p to P(C), cf. (2.18), (2.10). Cover
A by two arline open sets Ui, U2, choose any
liftings e 1 Derk((U1) , e 2 Derk(J)(U 2)
of p and choose a global vector field e on A.
Then by (2.10) there exist a t )(U 1 n U2) ,
al,l 1 )(Ui) , e2,132 e ((U2) such that the
following hold: (*) Define: ae =e 2 -e l ea -a. 2 - a 1 e2a + ala -' f2 Pl = el + [O- (all+ Il)-Derk(Ul)[F P2 = e2
+ e-(02+ B2)-- Derk((U2)[ and view C as
obid by glueing UixAI and U2xA1, AI=spK[
via the (U 1 n U2)-gebra map :(U 1 nU2IU+U
1 n U2 Then Pl and P2 !u together m ive our
liffin PC of p to P(C). So he problem ot
computing "explicitely" PC for a given A and
p amounts o finding a, 1' BI' satisfying he

system () ave. If A is he smooh projecfiw


mel U 1: SpecK[xy]/(y 2 - x(x- INx- ) hen
this computation can don explicitely. Raher
han perforin his computation (whose final
rult is after all irrelevant) we shall indicate
he seps which should be followed and hen
leave i to he re,der who enjoys
contemplating formulae. A can covered by U
1 and U 2: 5cK[uv]/(u - v(v - uNv - u)) lued
via the formulae U= V=y e may take o course
e = yl solving the system () reduces to two
oblems= a) lilting explicitely p to Derk(U]),
Derk(U2). b) finding (in explicit way) or any
, e(U n U 2) a relation o the form ]{ + {2 =
a2 - al where (Xl, ) eK 2 {(0,0)}, al e(Ul), a
2 e(U2). Indeed by a) we find el, explicite]y
and put a = ( - e])/y. Then by b) we write
expHcJtely = 2 - ;1' where ;] e(U), ;2 e(U2)'

Finally by b) write. expHcJte]y.e2a + ;]a + =


- 1 or me K, e(Ui), 82 e(U2) and deine
Let's deal sepiarely wlh problems a) and b).
For problem a) let p denote e ]JtJn p to
K[x,y] which kJUs xWi then one can
determine, explicitely by a standard
procedure, polynomials M,N, e K[xw] such
that M + a where = y2. x(x - ]Xx - and note
that the derJvati sends the ideal () into itselL
This derivation induces a Hting o p to
DerkU]) constructed analogously. 81 To deal
with 'oblem b) note that (upon denoting by ,
the classes oi x, y modulo (f)) we have ((U l
n U 2)=K[,, '1]--(' K n)+(' K n)+(' K2 n) neZ
neZ neZ Now note that 1) for n_>0 n e KI/,I/]
= 19(U 2) for n < 0 --n d(Ul) for n > 0 xy
_ .n .(U2 ) for n < 0 2n e((Ul ) for n _> 0 / 2n
e (9(U 2) for n < -1 Consequently any

element o! K[,, -1 ] can be explicitely


written as a sum l2[ + a 2 - a 1 where l e K,
all: K[,], a2l: K[/,I/] and we are done (note
that 2/ appears as a "canonical" ech
representative for HI(dA)I). Similar
computations can be performed providing the
explicit liftin 8 of a derivation pc DerkK to
P(C) where C is the extension of A
(=projectire smooth model 2 x G a with (C)
= K corresponding to a given line y -x(xIXx-X)--0, Xl:K, pX0) by C m bundle ((Q Qo ) on A (Q'Qo c A(K)), Descent.
Regularity We continue with our applications
of the preceeding theory. Start with a
preparation. (3.1) Let V be a smooth
projectire K-variety and K an
algebraically closed field o! definition for V.
Then there is an Integrahie Jogarithmic Der

K K-connection on (HI(dV), HI(dV)) defined


as follows. Fix a descent isomorphism V =
Vo)KK, V o a smooth projectlve Kovariety , view V as a split K[Po]-variety , Po
= Dark K and for PPPo define maps-' o vp,
v) by the formulae Vp(falj]) = Epaij]
82 JLV p([mij]) = [mjlPmij] where Exjj]
denotes Cech cohomology class of the
cocycie (xij). The link between the above
logarithmic connection and the logarithmic
Causs-Manin connection is given by the
following: (3.2) LEMMA. With notations
from (1.2) and (3.1), for any p DerKoK
we have the following commutatlve
diagrams: ?p Proof. Use only definitions.
The above lemma shows in partlcular that
the restriction of (V,V) to a logarithmic Der

K K-connection on (Plc(), Hl()) does


not depend on the choice of the "descent
lsomophJsm" OKoK we have fixed.
The following lemma provides another
useful description of our map JV defined in
(3.1): (3.3) LEMMA. With notations from
C3.1) identify HI(;) with the group of Kpoints of the (locally algebraic) Ko-grou p
scheme A = PicVo/K and Identify Hl()
with the group of K-points of'the Lie algebra
L(A). Then the map [Vp:HI(;)- Hl(()
defined for each P e Der K KJn (3.1),
restricted to ^(K) (A= PiC/o/Ko)
identifies with Kolchin's logarithmic O
derivative .p: A(K)+ L(A), see (LI.12).
Proof. To check the above statement it is
useful to adopt a functorial viewpoint. So let
A_: {locally noethertan Ko-schemes} +

{groups} be a contraviarint funclot. Then


one can define for any p e Dar K K the
"logarithmic derivative" O J[A p: A_(K)
LK(A-) (where LK(A_ ) Is the kernel of the
map A_(Spec Kiel)- A_(SpecK) induced by
e 0) by sending any element ge A_(SpecK)
intog-l(A_(l + p))ge A(SpecK[]), where I ep: K[e]+ Kiel is the 83 obvious Ko-aigebra
automorphism of K[ ] and where we still
denoted by g the image of g in A(Spec K ]).
Next note that any morphism of functors as
above A A' "agrees" with J, ^p and j[ A, p.
Finally, it is not hard to check that: 1) If A-Is the funclot of points of an aigebrlc Kogrou p then J. Ap coincides with KoJchlffs
logarithmic derivative[9 (to see this, use for
instance the formalism in rB i P 2) If A_' is
the functor S+ Hl( o x S( ) then [A p

coincides with our [9 P defined Jn The


discussion above plus the fact that A
represents the relative Picard functor [G]
closes our prooL The foliowlng theorem
establishes a description of P(C/K C) for C
commutative (notations as in (l.2), (l.),
(2.1)). Intuitively it says under what
conditions on pe P(C) we have that C
descends to the constant field of p. (3.e)
THEOREII. Let C be an irreducible
commutative algebraic K-group with abelJan
part A and let pc P(C). The following are
equivalent: l) p vanishes on K C
(equivalently, C is defined over KK[P]), 2)
The image of J[9 pe Hom(HIDR(A)m , H
IDR(A)a ) in Hom(SDI(C)m, S(C) a)
vanishes 3) p vanishes on K A and the image
of the map [gpe Hom(Hl(), HI((A)) in

Hom(S(C)m, S(C) a) vanishes. (Of courseJ,


V P in 2) is that defined in (1.2) while[ P in
3) above is the one defined in (3.1)). Proof.
2) =:3). By 2) and (1,#) we get ? p(HO/K)) c
o 1 H A/K ) hence by (i.) the image (p) of p
via the Koalaira-Spencer map vanishes. So p
lifts to some derivation of )A hence by (0.
i5) p vanishes on K A. Now 3) follows by
applying (3.2) to = A, Ko= K A. Write in
what follows S a = S(C) a. 3)==)1). Since
Vp(Sa)C Sa, S a is a K[p]-submodule of the
split Kip]-module HI((^o))KoK (where ^ =
Ao) K K for some abelJan Ko-variety , K o - K A) hence S a itself is O split: Sa=So K K
for some Ko-Subspace SoC HI(iA ). On the
other hand, upon Jetting o o A(K)
(where A = PiCo/K o) be the K-point
corresponding to m() (notation as in (2.1)),

our hypothesis 3) together with Lemma (3.3)


and with (l.l.ll) and (1.3.6), l) show that
XXe A(Ko). This clearly implies that C is
defined over K o i.e. KcC K o so .pK C 0.
Implications i):=3)::=)2) can be proved
using similar arguments. (.) REMARKS. Let
C be an irreducible commutative algebraic
K=group. Then (with usual notsions of this
chapter)=
1) It may happen that neither C nor A ( =
abelfan part of C) are defined over K D(C)
The simplest example of this kind is that ot C
= E(A), cf. (2.3). 2) It may happen that A is
defined over K D(C) but C is not. An
example can be obtained as follows. We let
K be a universal (ordinary) 6-field with
constant field k, we let A be an elliptic

curver over K admitting k as a field of


definition and we let B = G a x G m. Then let
a: Xa(B) = K+ HI(A ) be any linear
isomorphism and let m: Xm(B) --ZPic(A) be such that re(l) is taken via
Kolchin's logarithmic derivative J,?p.
:PicA + HI(C0 A) into a non-zero element
(this is possible slnce by Kolchin's theorem
[? p. Is surjectlve!). Then using (3.3) and
(3.0) one can check that the extension C of A
by B described by the maps (re,a) satisfies
our requirements. 3) If B ( -- linear part of
C) is unlpotent and A is defined over K D(C)
then C itself is defined over K D(C). This
follows directly from (3.t). t) If B is a torus
then C is defined over K D(C). Indeed, by
(3.0) it is sufficient to check that ^ is defined
over K D(C) But by (1.1.22) B is a D(C)-

subgroup of C hence by (I. 1.17) A becomes


an algebraic D(C)-group. /e conclude by
(0.15). /e close by proving a regularity
criterion (in Deligne's sense) for the Lie
algebra of an irreducible commutative
algebraic D-group. (3.6) In the rest of this
section (only) we assume (in addition to the
fact that K is algebralcaUy closed) that k is
also algebraically closed, tr.deg. K/k= m (, P
= DerkK, D = KIP]. Let W be a finite
dimensional D-module: by a model of V we
will understand the following data: a smooth
arline k-variety X of dimension m, an
embedding (over k) of the function field k(X)
of X into K, a vector bundle V x on X, an
integrable connection [Del 2] and a Kisomorphism V VX(xK such that for any
vector field v on X we have a commutative

diagram v Vx ---" X sV V p(v) where


p(v) DerkK is the unique extension of v
and V p(v) is the multiplication with p(v) in
the D-module V. Any (finite dimensional) Dmodule admits a model due to the following:
(3.7) RE!dARK. In the situation of (3.6) If
PI'" "Pro is a commuting K-basis of P and if
xl,...,XnC K then there exists a flnitely
generated k-subalgebra A of K preserved by
PI' '" 'Pro and containing x j, ... ,x n (for a
proof see lB3]). (3.8) Coming back to (3.6)
we say that a finite dimensional D-module V
is regular if it has a model (X,Vx, V) with
regular in DelJgne's sense [Dal 2] p. 90. By
Joc.cit. it is easy to check that if V is regular
then for any other model (X',V(,'), ' must be
regular (indeed by [Del 2] regularity is a
birational concept and does not depend on

passing to finite coverings). Using the latter


remark and the theory In [DeJ 2] p. 90 it
follows that: l) K viewed as a D-module is
regular. 2) If ' *V *" is an exact
sequence of D-modules with V', " regular
then %/ is regular. 3) If %/1 and V 2 are
regular D-modules then V 1 oV 2, V l %/2'
Hm(%/l'%/2) are regular, in particular V
is regular. By 1) above it follows that any
split D-module (of finite dimension) is
regular, since It is a sum of copies of the
D=module K. The "regularity theorem" [DeJ
2] p. IlS implies in particular that if V is a
smooth projectire K-variety then the space
HiE(V) is a regular D-module, with its Dmodule structue given by the Gauss-Manin
connection (l.l). Now here Is the main
consequence of our Theorem (2.2). (3.9)

THEOREM. Let C be an Irreducible


commutative algebraic D-group (D = K[P], P
= DerkK, tr.deg. K/k ( o. Then the D-module
L(C) is regular if and only if the D-module
Xa(C) is regular. Proof. Let U: = Spec d)(C).
By [DC] p. 398, C is a surjective morphism
of algebraic groups and. its kernel C has the
property that ((G) = K. Clearly has an
induced structure of algebraic D-group and C
+U: is a D-map so C is an algebraic Dsubgroup and we have an exact sequence of
D-odules. (3.9.1) 0 +L(G) *L(C) +L( ) +0
Now by (I.1.22) the maximal torus ' of is an
algebraic D=subgroup of is an algebraic
vector group whose space of additive
characters Xa(/) identifies with Xa(C). It is
trivial to check that the nondegenerate
bilinear map Xa(/?) x L(/?) * K, <x,v> = vx

defined in (I. 1.3) is a D-module duality (i.e.


that p(x,v> = (px,v)+ (x,pv) for x Xa(/),
v L(:/')) so we have a D-module
isomorphism Xa(C) = L(/) hence an exact
sequence (3.9.2) 0 +L(') ' L() 'Xa(C) '0
Finally let T be the maximal torus of the
linear part of G; by (1.1.22) it is an algebraic
D-subgroup of C and the quotient [ := C/T is
an alebralc D-group (I. 1.17) with ;() -- K
and With unipotent linear part. Let A denote
the abel Jan part of G; then is a quotient of
the Universal extension E(A) hence L(I:) is a
D-module quotient of L(E(A)). By (3.8)
HR(A ) is a regular D-module hence so is its
dual HIDR(A) which by (2.3) is
lsomorphic to L(E(A)). By
86 again we conclude that L(,) is a regular

D-module and we have an exact sequence of


D-modules (3.9.3) 0 + L(T)+ L(C)+ L()+ 0
Now T and being tori are split algebraic Dgroups (cf. (I. 1.9), (I. 1.26)) hence their Lie
algebras L(T), L(') are split D-modules,
hence are regular. Then the exact sequences
(3.9.2), (3.9.3) show that L(C) is regular if
and only if Xa(C) is so. (:1.10) REMARK. ]f
we consider the theory which we developed
in the present chapter for K a perfect field of
positive characteristic then all the interesting
phenomena we encountered dissappear
because DarkK = 0. But they reappear and in
fact multiply in case K is non-perfect. We
shall come back to this situation in a
subsequent work. CHAPTER If. GENERAL
ALGEBRAIC D-GROUPS In this chapter we
make a synthesis of the affine and

commutative cases which were studied


separately in Chapters 2 and 3 respectively.
We get some results for general (non-affine,
non-commutative) algebraic groups
concerning local finitness (cf. section 1),
automorphism functor (cf. section 2),
representative ideals (cf. section 3).
Throughout this chapter we assume without
any explicit mention (as we did in the
previous chapter) that K Js algebraically
closed of characteristic zero. 1, Local
finitehess crlterlo (1.1) Let G be an
irreducible algebraic K-group. Denote by L
its linear part ( = maximum linear connected
subgroup of G), A its abelian part (A :=
G/L), Z the center of C. lEecall [Ro] that G =
LZ so any normal subgroup of L is normal in
G. Let LJ be the unipotent radical of L

(which will also be called "unipotent radical


of G") H a maximal reductive subgroup of L,
T the radical of H and S = [H,H]. Note that
G/US does not depend on the choice of H.' it
is the maximum semiabelJan quotient of G
and consequently the field K(C/US) will be
called the maximum semiabelian subfield of
K(C). We denote by g, J., u, h, t, s the Lie
algebras corresponding to the groups defined
above. loreover we put 7 := C/[C,C], :=
g/[g,g], :-- u/u n [g,g]c I note that is the Lie
algebra of the unipotent radical of . Here is
our main result (notations as above): (1.2)
THEOREM. Let C be an irreducible
algebraic K-group. The following are
equivalent for p E P(G): 1) p P(G, fin) 2)
p preserves

88 3) p preserves u ) p preserves U 5) p
preserves the subfield K(G/U) o! K(G) 6) p
preserves the subfield K(G/US) of K{G). b)
P(G,fin) Is a Lie K/k-subaigebra of P(G) and
G is 1ocnlly finite as a Dfin=varJety where
Dfin = K[P(C,fin)]. c) K P(G'fin) = K G d)
The natural map P(G)/P(G,fin) P(,)/P(,fin) is
injective. e) lf i=G*G' is an isogeny and i*
:P(G')P(G) is the lifting map then i*
(P(G',fin)) = P(G,fin)n i*(P(G')). Proof.
Assertions d) and e) follow from a).
Assertion c) follows from b) and from (I.
3.7) and 0.1.26). That P(G,fin) is a Lie K/ksubalgebra of P(G) follows from a). Let's
check 1)------)) in a). Indeed, by (1.3.11)
upon replacing K by a Kip]-extension of K
we may suppose G is a split algebraic Kip]group. Then one easily checks that p

preserves U. Now #)::53) in a) follows from


(I. 1.16) while 3)2) is clear. tdoreover a)=5)
follows from (L I. 17) while 5) =) follows
from (L 1.17) and (0. I ), 5). To prove 2) ::1)
and the remaining assertion in b) it is
sufficient to prove that for any Lie K/ksubalgebra P of P(G) such that P13c U (or
equivalently Puc u + [g,g]) we have that G is
locally finite as a D-variety, D = KIP].
Claim 1. u + [g,g] = u + s. Indeed s = Is,s]
being semisimpie hence sc [g,g] hence '5" :
[ollows. Conversely since G/US is
semiabellan, hence commutative, Iu + s) is
commutative so [g,g]c u + s and 'E" follows.
Claim 2. JL is a D-subaigebra of g. Indeed,
by Claim I and by Puc u + [g,g] we get that u
+ s is a D-subaigebra of g. By (Ll.16) US is
an algebraic D-subgroup of G. By (I.1.17)

the semiabelian variety G/US becomes an


algebraic D-group. By (1.1.22) the torus
L/US is an algebraic D-subgroup of G/US
hence L = R-I(L/US) is an algebraic Dsubgroup of G (where ]: G +G/US is the
projection map) and we are done by (I.1.16).
Claim 3. u is a D-subalgebra of L Indeed, by
Claim I we have Puc u + s. On the other hand
by (II.3.17) the radical of (which equals u +
t) is a D-ideal in 4. Hence Puc (u + s) n n (u
+ t) = u and our Claim is proved. Now by
(11.1.15) and (I.I.16)(L) is a locally finite Dmodule while by (1.1.17), 0.1.9) and (0.15)
A -- G/L is a split algebraic D-group. Claim
e. In order to prove that G is a locally finite
D-variety it is sufficient to prove it after
replacing K by some D-field extension of it
K i/K. Indeed since A is locally finite we

can cover it by D-invariant open subsets A 1.


Then Gi, the preimages of A i in G, will be
D-invariant arline open subsets hence by
(1.3.7) (GjNK) will be locally finite
Kl)KD.modules hence ((G i) will be locally
finite D-modules and the claim follows. 89
Now we put a "D-structure" on "Chevalleys
construction" and on our construction from
B 1] p. 97 of an equivarlant completion of
G (cf. also [B 5] for group actions on the
same construction). Let us quickly recall
these constructions. (1.3) One starts by
choosing a finite dimensional K-subspace E
of O(L) containing a set of generators of the
maximal ideal m of {G and such that E Js
invariant under left translations with
elements in L(K). Then put d = dim(E nm), d
Q = projectlye space associated to/ E d qo =

point in Q corresponding to ^ (E n m) and


consider the induced action o:LxQ'eQ,
o(b,q)=bq (the isotropy of qo is triviail).
This construction will be refered to as
Chevalley's construction. Next we defined in
[B 1 ] p. 97 two commuting actions . L x (G x
Q)+ C x Q, (b,(g,q)) = (gb- l,bq) e: G x (G x
Q) + G x Q, e(x,(g,q)) = (xg,q) and proved
there is a cartesian diagram u GxQ v with Z
quasl-projectlve, I[ 1 the projection onto the
first factor, v the natural projection onto the
abelJan part and u is a principal fibre bundle
with group L for the action . Moreover O
descends to an action = G x Z Z and the
lsotropy of x o ;= u(e,qo) Z with respect to is
still trivial. Flnally let 'o c Z be the closure
of the orbit Gx o of x under the action .
We call e: G + Z the map (g) = (g,Xo) then w

* e = v. (l.a) Let,s Introduce D-variety


structures on the objects and maps occurlng
in (1.3). First replacing K by some D-field
extension (which is allowed by Claim ) we
may assume L is split, L = L OK and
taking E = EoK with Eo= (Lo) big enough
and Lo-invariant we may assume in (1.'3)
that E is a D-submodule of )(L). Then Q will
be a D-variety, qo will be a D-point of Q, o,
and {) will be D-maps. Since Z is a
geometric quotient of G x Q by T the
arguments in the proo:[ of 0.1.17) show that
Z has a natural structure of D-variety and u is
a D-map, Since G is a D-point, (e,q o)
is a D-point of G x Q hence x will be a Dpoint o:[ Z. Clearly is a D-map, hence so is
hence '?o will be a D-subvariety of Z. We
claim that Gx o is a D-invariant open

90 91 subset of G- o. To check this note that


u'l(Cxo ) = G x Lqo and u-l(Go ): G x L'o
where L'o = closure of the -orbit Lq o of qo
in G x Q. Hence u' 1(' o \ Gx o) = GX(o' Lq
o) Since L is split so will be E and Q hence
L- o \ Lq o is a D-subscheme of C. Since u is
faithfully flat and arline the ideal of G- o \
Gx o in (Z is the intersection of (1Z with the
ideal of Gx(o \ Lqo) in u, OGxQ hence is a
D-ideal and so Gx o is D-invariant. Now G
o is a projectlye D-variety hence by (0.15)
upon replacing K by a D-field extension of it
(cf. Claim b) we may assume G- o is split.
Since Gx o is D-invariant it follows from
(1.3.6), b) that Gx o (hence G which is Disomorphic to it) is locally finite so the proof
of 2) :::)1) in (1.2), a) is finished. (1.) To
conclude the proof of (1.2) we must check

that a)=6). Exactly as in the proof of ) :::5) it


is sufficient to check that for p c P(G) we
have pu c u if and only if p(u + s) c u + s. But
this can be checked via Claim I and the
arguments used to prove Claim 3 in the proof
above. (1.6) REMARKS. 1) If In (1.2), a)
we assume p c P(G/K) then condition 6) can
be replaced by: "6') p vanishes on K(G/US)"
This follows from (I. 1.9). 2) Note that we
implicitely reproved in (1.2) above the
existence of a minimum algebraically closed
field of definition for C between k and K
(this was shown in [B]). (1.7)
COROLLARY. Let C C t be an isogeny. Then
K G -- KC,. Proof. Let N -- Ker(C + G'). To
check that KC, c KG it is sufficient to check
that N and the inclusion N c G are defined
over K G. Now N Is central hence N lies in

the kernel of the multiplication map n" Z(G)


+ Z(G), n x = nx for some n )_ 1. Since Z(G)
is clearly defined over K G so is Ker I n
(which is finite!) hence so are N and the
inclusion N c G. Conversely to check that K
C c KC ' it is sufficient by (1.2} to check that
for any p' P(C',fin) its lifting p P(G)
belongs to P(G,fln); but this follows from
(1.2), a), 1)3). (1.8) COROLLARY. Let P be
a Lie K/k-algebras and D = KIP]. Moreover
let G be an irreducible algebraic D-group.
Then 1) G is locally finite if and only if
G/[C,G] is locally finite, if and only if the
unJpotent radical of G is an algebraic Dsubgroup of G, if and only if the maximum
semiaboliSh subfield of K<C> is a Dsubfield of 2) If G +G' Is an isogeny of
algebraic D-roups then G is locally finite if

and only if C' is SO. 2. Representin the


automorphism functor Our aim in this section
is to prove the following result (cf. (I. 1.23)
for notations).' (2,1) THEOREM. Let G be
an irreducible algebraic K-group and let
Au__t G .' { K-schemes} + { groups} be its
automorphism funclot. Then: 1) The
restriction of Au._tG to (reduced Kschemes} is representable by a locally
algebraic K-group (call it Aut C). 2) The
identity component AutC is offinc. 3) The
kernel of the natural map AutC AutL x Aura
(L = linear part of C, A = C/L) is finite. ) The
natural map G .' L(Aut G)- L(Au.._t C)=
P(G/K) is inJectire and its Image equals
P(G/K,fin). (2.2) REMARK. According to
the general assumption of this chapter, K in
the above statement is algebraically closed

of characteristic zero. But if we assume K of


characteristic zero not necessarily
algebraically closed and also that G is not
necessarily irreducible then our proof still
leads to conclusions l) and 2) in (2.1), this
providing a positive answer to a question of
Borel and Serra [BS] p. l2. Note that in [BS]
assertion J) was proved for linear G but
much more was proved in this case namely
that AutC is of type ALA i.e. an extension of
an arithmetic group by a linear algebraic
group. It was also observed there that there
exist examples of non-linear (nonconnected!) Cts for which (Au.._tC) (K) is
"discrete" but not an extension of an
arithmetic group by a finite group. In any
case assertion 3) in our Theorem shows that
if Aut L and Aut A are algebraic so is Aut G.

(2.3) Conclusions 1) and 2) in (2.1) are quite


easy to obtain by inspecting our construction
(1.3) and will be derived now while the rest
of the conclusions require further analysis.
As we shall see, in way of proving
assertions 3) and a) we will get a quite
precise description of all objects involved in
case G is commutatlve. Start with any locally
algebraic reduced K-scheme Y acting on Gin
the sense of (i.e. with a "family" of
automorphlsms of G with "parameter space"
Y). Then Y will also act on L and since L is
linear, by [BS] Au__.! L restricted to {
reduced K-schemes} is representable by
some locally algebraic K-group Aut L so
there is an induced map Y+ Aut L.
Assume now (Y)c AutL. Under this
hypothesis we can put an Y-action on the

objects in (1.3). Indeed, in notations from


(1.3) the space E can be chosen to be in
addition AutL-invarlant (because the
natural semidirect product of L by AutL
acts rationally on ((L)) so Y acts naturally on
. One sees that our varieties L,G,Q,Z,, := G
o, [} := [\ (G) are provided then with Yaction, qo and x o are fixed points under
these actions and all our mapso,,e,ll l,V,W;,e
are Y-equivarlant. (2.) Let us prove assertion
1) in the theorem. Since we want to apply the
representability criterion in [BS] p. 1#0 we
first construct a certain connected algebraic
group H as follows. Let I' c G x G x G be
the graph of the multiplication and be the
closure of
92 in x x . By [G] the functor S+

{acAuts(xS);o(DxS)=DxS, (oxaxaXxS)= xS}


is immJately seen to be reprentabie on the
category o locally algebraic K-schemes by a
iocly gebraic K-group H we iet H be as
usual the identity component o H. There is a
natural action q: H x G e G which is
aithul and hence effective in the sense o [BS]
p. Le now Y be y connoted reduced Kscheme o finite type acting on G and Yo be
ch that the corresponding automorphism o G
is given by some ao e Hl in order or Au to
be representable on {reduced K-schemes}
by a locally algebraic K-group Aut G (with
AutOG = Ho) it is sufficient by [BS] p. ]a0
to prove that or any y e Y(K) the correspodin
automorphism o G is given by some point o
H(K). Now both H and Y act on G hence
on L by representability o AuL we et

mphisms : H+AutL y:Y*AutL Since yo ) =


$(0 o) e AutL we get that Y) AutL $o
our discussion in (2.3) applie. In particul Y
acta on letting and globally fixed o there i an
induced orphism Since (yo ) = (o e H we
get that (Y) c H and we are done. To
prove assertion 2) in the theorem, we must
show that H is linear. Let ( + , be a Hequivariant resolution of ; then the map r: ( +
A is nothing but the Albanese map of C. and
is H-equivariant (with respect to the
trivial H-action on A). So H c Ker
(Aut '* Aut(Alb())) which is linear by
[Li] and we are done. In fact assertion 2) can
be proved more "elementary" (i.e. using
neither [Li] nor "equivariant resolution"), cf.
proof of assertion 3) below. To prove
assertions 3) and 0) in (2.1) we need a

preparation. (2,5) Let C be an irreducible


commutative algebraic K-group and let's
borrow notations from (lII.2.0). Recall that C
= B' x C 'L with B' a vector group and
Xa(C/) = 0 moreover B" i$ the unipotent
radical of C 1. We claim that P(C/K) =
(Xa(B') )L(C/)) g P(B'/K) P(C/K ,fin) =
(Xa(B') $ L(B")) e P(B'/K) where we
identify K(C) with K(K(B')eK(CI)). Indeed
the formula for P(C/K) follows from (I. 1.9).
To check the second formula note that by
(1.6) P(C/K,fin) consists Of all pe P(C/K)
93 vanishing on K(Cm). Now since C Z =
C"XAC m we have L(C/) _L(C")XL(A)L(C m) Let us choose a basis of
L(C t) as follows: first take a basis (ai) i of
the image of L(C/) eL(A) and lift it to a
family t(3 i,a i eL(C")XL(A)L(Cm); next let

(()j be a basis of L(B") -- Ker(L(C") + L(A))


and (ekm)k be a basis of L(B m) =
Ker(L(Cm) e I.(A)). Then: is a basis of
L(CI). Identifying K(C) with
K(K(B')OK(C")$K(A)K(Cm)) we may write
any p P(C/K) as where d P(B'/K) and
figjh k e Xa(B') c K(B'), Clearly, if all fi's
and hk'S vanish then p vanishes on K(Cm),
Conversely assume p vanishes on K(Cm),
There exis. ts a family (xj)j xj K(A) with
det(oxj) 0, e set 0 = pxj = je(oix j) so all j's
vanish, Finely there exist elements yq eK(C
m) with de, (myq) 0. e et 0 = pyq = k(myq)
hence hk=0 or aH k and our claim is proved,
(2.6) LEMMA. In notations of (III.2.) we
have 1) The naral map Aut C" Aut A is
injectire. 2) The natural map Aut C m * Aut
B m x Aut A 1s injtive. 3) The natural map

Aut C Aut B m x Aut A is injectire. Prf. To


prove 1) let oeAutC" induce the identity on
A. Viewing C" as obtained by glueing 5p
Ai[[i,...,] , ([1,...,) a basis of E", we get that
o induces A-automorphlsms of these rings.
Let oi = Pik([). Compatibility with gluelags
gives (.) Pik([ + aij) = Pjk(0 + a I t where [ =
([1 ..... )' aikj = ajj(), aij = (aij ..... aij). So
all Pik'S (or fixed k) have the me dree n k. If
n k 2 for me k then Taylor's formula applied
to (e) gives (writing fi instead of Pik and n
instead of nk): .(n) + ? -.t_ (Da k + n-l)(o +
.... (**) f(in)( 0 k a ij = ,Jn)(o + ,n-1)(0 + ,,,
+ a where t dots stand for terms of, degree at
mt n - 2. Looking at degree n terms in () get
fn)= fn)so the coefficients of (fn)) i glue
together to give everywhere defin regular (n)
_ f(n) f(n) eK[[],...,Et]. Looking at terms of

de(tee n - ] in functions on AI consequently fi


-j=
94 95 (**) we get a f(n) k ' fi () [._5k ()aij
f(n- 1)() (n-l) n Upon making a base change
in I:, we may assume that f(n) contains 1 with
coefficient 1. . . n-I ,(n*l) we get (identifying
coefficients of Letting i be the coehctent of 1
in [ I + k = ej a i nai k2 Xkajj - wih X k E K.
Bu the let hand side equs aij(nl + 2Xkk ) so k
k - : Xkrr + bik for Xkr K, bik A i- all k. The
ors of degree I in (Pik)i glue together Pik r
Equa:ion(,) above impli r Injtivity o
E"+HI(A) iplies Xkr=O or r k and Xkk 1.
Hence in particular bjk = bik 81ue toEether
to 8ive a constant b k K. Consuently Pik: k +
bk or all i and k. Since ust have a ixed point
we get b k: 0 for 1 k so identiy. To prove 2)

ake C inducin 8 the identiy on th Bm and A.


Then induces automorphisms oi of Ai[E] for
all i. Sie for X hve oi aiX [ or e Yi E and
some invertible element a i o Aj.
Compatibility with gluein gives aiij(j)Xi:
aij(X) j so Xi j or all i,j. Since restriction to
B is the identity i X so a i = aj or all i,j.
Consequently the ai*s 81ue together givin 8 a
global section is so on Bm nd we conclude
that ff = ldentiy. Assertion 3) ollows from 1)
and 2). (2.7) LEMMA. In notations of (III.2.
a) we have 1) The map AutC + Aut B x Aura
is injectire. 2) AutC = Horn(B',B")aAut B'
where the right hand side acts on C = B*x C
1 by the formula (,aXx,y) = (o(x),y + (x)), e
Horn(B',B"), ( e AutB ,, x I: B*, y C I. In
particular Theorem (2.1) holds for
cornmutative G. Xa(C 1) _- C 1 C 1 C 1

Proof. Since 0 any JsornorphJsm u of C = B'


x must send onto hence sends B" onto B" and
Brn onto Bm in particular o induces
automorphisrns of C"= cl/Brn, Crn = CI/B"
and B' = C/C I (call them (', rn' d). If o
induces the identity on A and B then by (2.6)
a" and Orn must be the identity, while d is
clearly the identity. The isornorphism C :' B'
x C" x ACrn shows that o itself must be the
identity and assertion 1) is proved. Assertion
2) follows immediately from (2.5), (2.6) and
the remarks above by viewing
autornorphlsms of B' x C / as 2 x 2 matrices
whose entries are group homornorphisms. As
for the assertion that (2.1) holds in the
cornmutatJve case, all we have to note is that
from the assertion 2), from (2.5) and from
(1.3.12), (II.3.10) we get: L(AutC):

Horn(L(B'), L(B")) L(AutB ') = =


(Xa(B') L(B") P(B'/K) = -- P(C/K,fin)
Our ]emma is proved. As a by product of the
proof we get.' (2.8) COROLLARY. For C an
irreducible cornmutative algebraic K-group
the cokernel of the map X C: L(^ut C) +
L(Au C) identifies with Xa(B') L(Crn)
(notations as above). (2.9) Let's prove
amertions 3) and a) in (2.1). Put C = Z(G)
so G -- LC. If B is the linear part of C then
15 is of finite index in L nC and L/C nL -'-'^.
Now by (2.7) the map ^ut G + ^ut L x
^ut(C/B) is injectlye and assertion 3) ollows
because C/B + ^ is an isogeny. To prove ) we
proceed in several steps: Claim 1. Any
element of ^utL preserves C n L. Indeed
AutL acts on the finite group Z(L)/Z(L)
hence acts rivially on it. Now C n L is an

intermediate group between Z(L) and Z(L)


which implies our claim. Claim 2. Any
element of AutC preserves C n L. Indeed
AutC preserves B and acts trivially on
C/B hence acts trivially on C n L/B hence
preserves C n L. Claim 3. L(^ut G) = L(^ut
L) x L(^ut (C n L)) L(^u! C) Indeed this
follows frorn the identification AutG =
AutL x AutO(c n L) AutC which in its
turn follows from Claims I and 2. Claim t.
P(G/K,fin) =' P(L/K,fin) x P(C L/K,fin)
P(C/K'fin)' Indeed, any p P(G/K,fin)
preserves L and C (use a splitting argument,
cf. (L3. I 1)) hence induces an element
(pL,PC) of the right hand side of the latter
formula. Clearly the map p +(pL,Pc ) is
injectlve; to prove surjectlvity view G = (L x
C)/(L n C) and apply (LLI7).

96 Now we conclude the proof of z) in our


Theorem. By (I.3.12) the maps XL; L(Aut
L)e P(L/K,fin) and : L(Aut(L n C))e P(L n
C/K,fin) XLn C are lsomorphlsm (note th L n
C is not irreducible in general but in (I.3.12)
we do not assume irreducibilityl) and by
(2.7) so Is XC: L(Aut C)e P(C/K,fin) Claims
3 and t show that lG: L(Aut G)+ P(G/K,fin)
must also be an isomorphism and we are
done. 3. Products of abelien varieties by
affine 8roups As we have seen in Chapter 3
the results on P(G) do not extend from the
arline to the non-arline G at alil But there is
still a case when such an extension is
possible namely when G is the product
between its linear and its abelian part as
shown by the following: O.!) THEOREM.
Let G = L x A, A an abellan E-variety and L

an irreducible arline algebraic K-group.


Then: 1) K ID(G) -- K G and K G is the
algebraic closure of the composltum KAK L.
In particular there is a split exact sequence
of Lie K/k-algebras 0 e P(G/K) e P(G) e
Der(K/KAK L) + 0 2) P(G/K) P(L/K)
(Xa(L) L(A)) 3) P(G/K,fin) = P(L/K,fin) a)
P(G) contains a representative ideal. More
precisely, if is a representative ideal in
P(L) (which always exists by (]I.1.22)) then
(Xa(L) L(A)) is a representative ideal
in P(G). ]n particular if the radical of L is
unipotent then Xa(L) L(A) is a
representative ideal in P(G). 5) Upon
identifying AutG with Aut L x Aut A the
isomorphism in 2) is equivartant (with Aut G
actin 8 on the left hand side and Aut L x Aut
A acting on the right hand side naturally). 6)

Upon fixing "alascent" isomorphisms A =


AOKoK, L = LoKo K, Ko--K G, lettin Po =
Dark K and D o --KiP o] and viewing both
members in assertion 2) as IDo-modules in
the o natural way (in the right hand side we
take the Do-mOdule structure induced by the
natural one on each factor and term) the
lsomorphlsm in 2) is an isomorphism of DomOdules. Proof. 1) follows from (].1.19),
(0.15), (II.i.l) and (].1.25). To prove 2) note
that for the projection map :G--L xA eL we
have gC G =L so K(L) is a D(G) - subfield
of K(G). The restriction map P(G/K)+
P(L/K) which is therefore well defined has
an obvious section given by trivial liftin Po
+ Po Gl' Let p P(G/K) restrict 97 to 0 c
P(L/K). Since p is a vector field on G
tangent to the fibres of ] we have p = r.f i ) e

i where (el) i is a K-basis of H(T A) and


fietiT(L). The equality pcp=(p)J+ l)P)pC is
equivalent then to r'(PLfi ' fi I - 1 fi)PL
PAOi = 0 Choose (xj)j, xj I((A) such that
det(eixj) 0. Applying the above equality to l
xj c K(K(L)K(A)) we get PLfi- fil - lfi=O
for alii hence fi t Xa(L)' Conversely, if fi
Xa(L) one immediately gets r-f i( 0 i z
P(G/K); this proves 2). 3) Clearly P(L/K,fin)
injects via trivial lifting Po e po I into
P(G/K,fin). Conversely, assume p = dl +
ZfiO i z P(G/K,fin) with d c P(L/K), (el) l a
basis of H(T A) and fi Xa(L)' By (1.2) and
(1.6) (and with notations from 1oc. cit.) p
vanishes on K((L x A)/US) c
K(K(L/US)K(A)); this shows that d vanishes
on K(L/US) hence dP(L/K,fin). On the other
hand Jet xj K(A) be such that det(eixj) 0

then 0 = p(l xj) = Zfieix j from which


we get fi = 0 for all i and 3) is proved. a) Let
V be a representative ideal in P(L). Then
putting Po = IDer(K/KG) Pl = P(L/K,fin)
P2= P3 = Xa(L)$ L(A) we have P(C) = Po
Pt P2 e P3 and the following commutation
relations hold: [Pi,Pj] c Pj for all 0_( i_( j_(
3 Indeed to check [Po,P3] c P3 take p
Der(K/Ko), let p, be the trivial lifting of p
from K to C = G O)KOK (G O = L O g Ao, L
O an arline Ko-group, A an abelJan Kovariety) so PG = PL I I p and let v = rfl6i e Xa(L) L(A). Then Pc i A i = = r'fi
)VpOi + ZVp[fiO I z Xa(L) L(A) This
computation already proves assertion 6).
98 To check [Pl,P3] P3' [P2'P3 ]c P3 take
v = r.f i O01 as above and d P(L/K)

(identifying it with dO 1). Then one checks


that [dO l,v] c Zdf i e i = r. Vdf i 0 i
Xa(L)O L(A) .The rest of the commutation
relations are clear. So P2OP3 is an ideal in
P(G) contained in P(G/K) and it is a KLinear complement of P(G,fin) = Poa)PI in
P(G). To conclude the proof of assertion 0)
and to prove assertion 5) of the theorem it is
sufficient to note that for any o = a L x a A
Aut L x Aut A = Aut G the following
formulae hold: OLVO 1 = V and for v = r.f i
0 i as above, a vo=(OLOA)(ZfiOOi)(o lo
1)= = r. fOLf i) ) OAeio l) Xa(L) L(A)
(where we denote as usual byaoAOl. the
induced field automorphisms of K(G), K(A),
K(L)). (3.2) COROLLARY. Assume G is an
irreducible algebraic K-group such that O =
GiG 2 with G i normal irreducible

subgroups, G 1 arline and G 2 an abelian


variety. Then K D(G) = K G. In particular
there is an algebraic D-group structure on G
if and only If K D is a field of definition for
G. Proof. We have an isogeny C 1 x C2+ C
and apply (I. 1.21), (3.1) and (1.7). EXAM
PLE We already considered in (1.1.9) and (I.
1.26) the example of a product G = A x Ga.
With notations as in 1oc. cit. we have that
while P(C/K,fin) = !<[j.. Klyx x is a
representative ideal. CHAPTER 5.
APPLICATIONS TO DIFFERENTIAL
ALGEBRAIC GROUPS I. Ritt-Kolchin
theory The theory and examples of this
section are classical. (1.1) By a f-field we
will understand in what follows a field 'of
characteristic zero equiped with a set A =
{l,,..,m} of pairwise commuting derivations

a I a m a =l,...am)Nm we will writex instead


of 1 '" m X for anylt:', If m= I we say is
ordinary and we write X',X",V",...,X (k)
instead of 6),, 2X, 6 3X ,..., 6 kX. We define
the constant field of ' to be the field 9:6 = {X
iX = 0, l_<i_(m}. EXAMPLES 1) Let (,6
1,...,6 m ) = (Q(Zl,...,Zm), a/a Zl,...,a/a Zn')
be the field of rational functions. Here 9 :A =
Q. More generally one may take 9' any
finitely generated field extension of Q and 6 i
= /a z I where Zl,...,z m is a transcendence
basis of /Q. Note that in this case 9 A/Q is
algebraic. This is of course the most
"geometric" example one can think of{
indeed one may think 9: as the function field
of some smooth arline Q-variety V and 6
1,...,6 rn as corresponding to pairwise
commuting vector fields on V generating its

tangent bundle. 2) Let (,6 1,...,6 m ) =


(Q(Zl,...,zm,exp Zl,...,exp zm), a/a Zl,...,a/a
Zm). Here .A = O too. This should also be
viewed as a "geometric" example, in the
sense that it "comes from" m commuting
vqctor fields on a variety. V which is a
model o! '/Q{ we can be very specific here
by taking V = SpecO[zl.,...,zm,tl,...,t m] and
letting the vector fields be 3) Let :=Q<y>
=Q(yo,Yl,Y2,...) be the fJeJd of rational
functions in infinitely many variables
yo,Yl,Y2,... equiped with the derivation : '- r
defined by 6 Yi = Yi+I' i_> 0. This example
Is very dlilerent in nature from Examples 1,
2 because Yo and its derivatives are
algebraically independent over A =O it
should be viewed as very "non-geometric".
Nevertheless it will be crucial to include this

kind of examples in the theory! #) Let (9:,6


1,...,6 m ) = (Met(R),8/8 Zl,...,a/8 z m)
where Mer(R) is the field of meromorphic
functions on the domain Rc Cm. Here A = C.
5) Let 9r= Met(cS,0) be the quotient field of
the ring {S an of germs o! holornorphlc
functions at (cs,0) and 61 = a/a z i, I < i _<
m. Cs'0 (1.2) By a morphism of A - fields we
will understand a field homomorphism a:
such that
lOO lol for all X r' and I _( 1 _( m. If o is
an inclusion we say that is a A -extension of
"and that ' is a A -subfield of . We say that is
A - finitaly generated over ' if there exist X
i,...X n e such that is generated as a field
extension ofo' by the family ($'Xi)l_<i_<n,
oeNm EXAMPLES We have the

following natural morphisms of A -fields o I


o2 Q , Q(zi,...,z m) s
Q(Zl,...,Zm,eXPZl,...,expz m) + leer((m) +
Mar(cm,0) 3 di/dz i Q s Q<y> - Mar(C,0),
Yi where e Mar(C,0) has the property that
the family consisting of ? and its derivatives
cf arbitrary order are algebraically
independent over Q (such functions are
called transcendentally transcendental and
their existence is easily established). Of the
above A -morphismsOle2e 3 are the ones
which provide A -finitaly generated A extensions. Other remarkable examples of
morphisms of A -fields are provided by
restrictions lEer(R) + lEer(R 1) where R 1
is a subdomain of R. (1.3) Let be a Aextension of ' one says that is semiuniversal
over 'if for any morphism of A -fields o :+ '1

with 1 A -finitaly generated over ' there


exists a morphism I:: r' 1 + making the
following diagram commutatlve A basic
result of Seldenberg [Sell says that Mar(R)
is a semiuniversal extension of Q if Rs say,
is a disk in C. (l.tt) Let be a A -extension of '
one says is universal over ' if is
semiuniversal over any A -finitaly generated
extension of r contained in Such an is always
algebraically closed hence so is its constant
field. ^ A -field will be called universal if It
is universal over Q. ^ basic result of KoJchin
[Kl] says that any A -field ' has a universal
A-extension (which is countable if is so). On
the other hand Seidenberg's results in [Sell
imply that any countable A-field can be Aembedded into Met(cm,0). In particular
Mar(cm,0) contains A-subfields 2/, which

are universal; admittedly no explicit


description of such an 9 seems to be
available. But at least we dispose of
"theoretic" examples of universal A-fields
which are realized as A-fields of
meromorphic functions. (1.) By a A--a!gebra
over a A-field 'one understands an '-algebra
equiped with m commuting derivations on
lifting 61'"" m (and still denoted by 81,..., m
). By a A-ideal in . one understands an ideal
3c such that 6i(3)c 3, I i m. If Sc is a subset
one denotes by ($) the smallest ideal of
containing S, by iS] the smallest A-ideal of
containing 5 and by {S} the smallest perfect
A-ideal of containing S. Note that (S} = {r t:
n such that r n t:[S]). There is an ambiguity of
notation here: if Xl,...,x p E then {Xl,...,x p}
will generally denote the smallest perfect A-

ideal of containing Xl,...,x p rather than the


set whose elements are x.,.,x (unless the
contrary is obvious from context). p A A- 'algebra is called A-finitely generated if there
exist Xl,...,x n E such th'at 6 o is generated as
an algebra by the family ( xi)l_4i_(n,aeNm.
By a A--algebra map we understand an 'aJgebra map which commutes with the
derivations in the obvious way. EXAMPLE .
(a) Let := '{yl,...,yn ) denote the :'-algebra of
polynomials in the indeterminates Yi where I
_( i _( n, o e N m viewed as a A- '-algebra
with the derivations j: + satisfying (a) (a + I)
Yi = Yi (o) and if m = I (ordinary case) we
write for all I <i<ns {8eN m. We write Yi
instead of Yi y',y",y"' instead of
y(l)y(2),y(3). So for instance if m = 1, n= 2
and if A grwe have in '{yl,Y2} the

equality.' (yly . (y)2) -- X,yly + .y,jy + Xyly _


2yy' One calls :'{yl...yn} the ring of Apolynomials in the A-indetermlnates yl,...y n.
To illustrate the concept of A-ideal start with
a A-polynomial F e e'{yl...Wn } where is,
say, ordinary. Then the ideal iF] can be
described of course as iF] = (F, F, 2F,...,
kF,...) but the ideal {F} does not have in
general such an easy description by
generators. We make the important remark
that even if F is irreducible (as a
polynomial) the ideal {F} may fail to be
prime. E.g. take n= 1, Fe{y} F=(y")2-y, [Ri]
p. 2tq then the ideal {F} is not prime
because combining 6 2F and 6 3F we get y,,
(y(3)y(5). 12(y())2 + gy(q). 1) while none
of the factors above belongs to {Fl.
Concerning A-ideals of A-polynomials we

have the following basic results due to Ritt:


102 103 (1.6) THEOREM ([Ri] p. 10). For
any perfect A -ideal 3 in {yl,...,yn } there Is a
finite subset Sc 3 such that 3 = {S}. (1.7)
THEOREM ([Ri] p. i3). Any perfect A ideal in {yj,...,yn } is a finite intersection of
prime A -ideals. The problem of making
(1.6) and (1.7) as "explicit" as possible was
treated in detail by Ritt [Ri]. (l.g) From now
on we fix throughout this Chapter a universal
A -field {, and we denote by =,A its constant
field. Call the set ,n the n-arline space. For
any subset $ of 5J[{ yl,...,y n} we define the
zero locus of S in ,n by the formula Z(S) = {a
e ,n; F() = 0 for all Fe S} The subsets of /,n
of the form Z(S) as above will be called A closed subsets. Conversely for any subset .

of ?jn we define the ideal of - as follows: I )


= { fe /,{ yl,...,yn } F() = 0 for alia e }
Clearly I ) is a perfect A -ideal in { yl,...,yn
} . Due to universality of the correspondence
S* Z(S) establishes a bijection between A closed subsets of ,n and perfect A -ideals in
{ yl,...,yn ] . The A -closed subsets of /n form
a topology called the A -topologyl this
topology is stronger than the Zariski topology
and by (1.6) it is a .Noetherian topology. A
A-closed subset . is irreducible iff l ) is
prime. EXAMPLES l) }n Is an irreducible
A-closed subset of /,n; indeed (n= Z(yls...,y)
and i(x. 2) The A -closed subset Zy") 2 - y)
of , is reducible. 3) The A -closed subset Z(y
2 - y") of is irreducible. (1.9) Let . be an
irreducible A -closed subset of /,n; then . has
the topology induced from the A -topology on

,n which we call the A -topology of r- . Let


be aA -open subset ore (i.e. an open set for
rhea -topology); a function ! :l- ?if, will be
called A -regular at Xoe D if there exist a A
-open neighbourhood o of x o in and A polynomials F,G { yl,...,y with G nowhere
zero on o and f(x) = for all x f2ol f will be
called A-regular on 12 if it is A-regular at
any point of [. Any A -regular function 1 gC
is continuous with respect to the Atopologies on 17 and . Define (i) := {f: i ; f
is A-regular on } It is a A- ,-subalgebra of
the A--algebra of all functions +. In
particular we dispose of the A--algebra ((.)
of all A-regular functions . *. Note on the
other hand that we dispose of the algebra
{r.} := 9{y 1 ..... yn}/l(T) which can be
called the A-coordinate algebra of r. and that

we dispose of an injective A- t/,-algebra


map This map may not be surjective (and this
is a crucial contrast with algebraic
geometry!). Nevertheless it is easily seen to
induce an isomorphism to the quotient fields.
The common quotient field of ,[r.] and {(r.)
is denoted by ,<r.>. Define dim r. = tr.deg.
EXAMPLES 1) If = ?j,n then ,(T.} + (T.) is
easily seen to be an lsomorphism; in other
words ((,n) ={yl,...,yn} ' Note that dim,n = .
More generally (but this is more subtle) we
have shown in lbs] that {.}- ((.) is an
isomorphism if . is Zarlski closed In t/, n
and, as an algebraic varlety Js smooth. We
will not use this result in what follows. 2)
Let .=--Z(y')C,. Then if ,, Xlwe have that 1/(,)](), l/(rl - ) Z{} where I is the image of y in
{}. Note that dim() = i. (1.10) Let F be any

field. By a na've ringed space over F we


will understand a pair (X ) consisting of a
topological space X and of a subsheaf () of
F-algebras of the sheaf F X of all F-valued
functions on the open sets of X. By a
morphism (X, GX)- (Y,y) of naYve ringed
spaces over F we will understand a
continuous map f .' X + Y with the property
that for any open set V c y and any : (y(V) we
have ef (X(f' I(V)). We obtained a category
{na've ringed spaces over F} (l.tl) By an
arline A-manifold we wJiJ understand an
object In the category of naive ringed spaces
over , isomorphic to (.s .) where . is an
irreducible A-closed subset of some ,n and
T. is the sheaf of A-regular functions,*
defined in (1.9). By a A-manifold we will
understand a na've ringed space (T.) over ff,

which is
104 105 Irreducible and can be covered by
finitely many open sets r. l,...:p such that each
(i, CjE i) is an arline A -manifold. For any A
-manifold ]: we define > as the direct limit
of all C7(fl) for c E non-empty A-open sets;
then > is a A-extension of , (and in case ]: is
arline it coincides with the one defined at
(1.9)). We define the dimension of E by
dim]: := tr.deg. By a morphism of We
obtained a category -manifolds} We leave to
the reader the verification that this category
has direct products. As expected if r.c%(, n,
TcL m are A-closed irreducible subsets then
r. xTc?,(, n+m is A-closed and irreducible
and so it has a structure of arline A manifold; this construction is "invariant" and

"globalizes" via "lueing" to get the existence


o! products in the expected way. A group
object in the category of A-manifolds will be
called a fi-group; so a A-group is a Amanifold T equiped with morphisms and
with an element eer satisfying the obvious
group axioms. With the obvious notion of
morphisms of A -groups we obtain a
category { A -groups} A A -manifold of
finite dimension will be called a A omanifold. ^ 6-group of finite dimension will
be called a A o-group. Any A-group in our
sense "provides" an "Irreducible A -group"
in Kolchin's sense [1(2]; It is not clear
whether the converse holds. One can prove
that our 6 o-groups "are" precisely Kolchin's
irreducible A -groups of type zero [K2].
lefore giving a few examples of A-manifolds

and A-groups it is convenient to discuss


more generalities. Let F be an algebraically
closed field. In the rest of this section (only)
F-varieties will always be viewed as naive
ringed spaces (rather than schemes) over F;
so thcy will be identified with their sets of
F-points and their structure sheaf will be the
sheaf of (F-valued) regular functions. Fvarieties will always be assumed
irreducible. (J.J2) Let X be an arline Rvariety and take an embedding i: X - A n =.n;
then X being Zariski closed in ?.n is in
particular A -closed and by a basic result of
Kochin [K l] is irreducible in the A topology so it has by (1.9), (I.11) a structure
of arline A -manifold which we cat] . 'e
leave to the reader the task of checking that
this structure does not depend on the

embedding i. Note that dim ( = o if dim X >


O. Now if X Is any -variety, cover X by
arline Zariski open sets X. and note that the I
A-manifolds i glue together to give a Amanifold structure on X which we call ; will
be called the A-manifold produced from X.
As sets we have = X but the topology on
(called the A,-topology) is stronger than the
(Zariski) topology of X. The A-manifold
does not depend on the arline covering of X
and is "naturally" associated to X. Actually
one may check that X gives a functor {?Zvarieties} { A -manifolds} It worths noting
that if X, are ,-varietles then the map
HomiL_var(X,Y) + Horn A-man () is
injectlye but not surjective in general. Indeed
if X =Y =A 1 then Homa.man( 1, 1) contains
the map f(y)=(y,)2. y3, say, which does not

belong to Homg t_var(AI,Al). The functor


above commutes with products so induces a
functor G + {irreducible algebraic 21gr0ups} - { A-groups} So we already
dispose of a series of examples of Amanifolds and A-groups: EXAMPLES For
any ?,C-variety X we dispose of a Amanifolds (; in particular n, n are Amanifolds. Any irreducible A-closed subset
r. o a A-manifold ( has a structure of Amanifold (the way of seeing this ls rather
trivial and we leave it to the reader). In
particular the A-closed sets below have
natural structures of A-manifolds-' l) Z(y" +
ay' by)c l/o= J, a,b , y a A-indeterminate.
2) Z(xy - [, yy" - (y,)2 + ayy9 c /, 2 _- 2, a ,
x,y two A-indeterminates. 3a) Z(y 2 - x(x IXx - c), x"y - x'y' + ax'y) c ?,L 2 = 2, c , a e

,, x,y as above. 3b) Z(y 2 - x(x - lXx - t),


.y3.2(2t - l)(x - t)2x'y + 2t(t - lXx - t)2(x"y 2x'y'))c 2L 2 = 2, t EL, t' = l, x,y as above. A
) Z(YI2'Y21'Y31'Y32'Y22 ' 1'Y33 '
I'Y3'Y'11 ' Y23Yll )c GL3' Yij 6indeterminates l/i,j I . The irreducibiJity of
these sets is easily established directly in all
cases except 3a) and 3b); the latter cases are
more subtle but nevertheless they can be
treated using lItts theory and we leave them
to the reader, too. Examples l), 2), 3) have
finite dimension while ) has infinite
dimension. Now for any irreducible
algebraic i-group G we dispose oi a 6-group
; in particular Gin, CLN, are A-groups (of
infinite dimension), where A is any non-zero
abelfan ,-variety. It is easy to check that any
irreducible A-closed subgroup o! has a

natural Structure of A-group. On the other


hand if I: is one of the A-closed subsets
defined in 1)-#)
106 107 above then the following hold. If 11
is given by I) then r. a A-closed subgroup of
1 = a' If r. is given by 2) then Ii is a A-closed
subgroup of Z(xy - 1) = m c 2. If r. is g'iven
by t) then r. is a A-closed subgroup of CL 3.
Finally if r. is given by 3a) respectively 3b)
then the A-closure ]' of r. in 2 is a subgroup
of c 2 where A c p2 is the elliptic curve
obtained by taking the Za. riskJ closure of
Z(y 2 - x(x - l)(x - c)) respectively Z(y 2 x(x - l)(x - t)) in p2. All these statements can
be checked by direct computation! We
conclude that in examples 1) 2) r. is a Aogroup, in example #) I: is a A-group while in

examples 3a), 3b) ]' is a Ao-group. As


remarked in the Introduction examples 1) 2)
3a) #) are due to Cassidy and Kochin while
example 3b) is implicit in Manin's paper
[Ma]. Here is another remarkable class of
examples of A-groups of infinite dimension
due to Cassidy an related to work of litt: (i).
(j), c . 5) I' = 2, ,: I' x I' + T, F((Ul,U2),
(Vl,V2)) = (u I + Vl,U 2 + v 2 + i<j aiju I
1 ' aij Many other examples may be found
in Casidy's papers [Ci]. (1.13) In (1.12) we
constructed A-manifolds starting with varleties. Now we shall construct Amanifolds starting with -varieties. Let X
be an arline X-variety, choose an embeddin[
l -' Xo-,ln and compose it with the natural
inclusion n+ 9,l,n; then X o appears as a Aclosed subset of ,n (indeed if X O is the zero

set in ,n of Fi,...,Fp[yl,...,y n] then X o is the


zero set in ,n of FI,...,F p, y'l,...,y' n e {,
{yl,...,yn}). So X O has a natural structure of
arline A-manifold, cf. (1.9), (1.11) which we
call o and which is easily seen to be
independent of the embedding i o. Now if X
O is any (-variety we may cover X o with
arline Zariski open sets Xoi and note v manifold with dimX o=dimX o. As that the
A-manifolds Xoj glue together to give a A o
o ~ . topological spaces X = X but the
structure sheaves are drastcally different
(e.g..If X O = J{ we already saw that e(Xo )
contalns the function f :X', f(l) = 1/( - 1)
where t 7, t l; but of course f is't-valued, not valued so f I ((Xo). v We obtained a functor
X o s Xo {-varieties} + { Ao-manifolds}
inducing a functor C o o {Irreducible

algebraic -groups} + { a o-groups} Any Aogrou p isomorphic to C o for some C o as


above will be called split. Finding criteria
ior a Ao-grou p to be split will be one of our
main concerns. 2. Ao-roups versus algebraic
D-roups (2.1) lecall that we have fixed in the
preceeding section a universal A-tield //with
constant field 1. Then let P .be the free
Integrahie Lie ./ff1-algebra built on 6 l,...,Am
(0.)1 recall that P = J& 1 "' m and for all
i,j and ,la ?L. Moreover let D = [P] recall
that D has an -bis consisting of monomials
EN TM, i.e. D = y m6a with the "obvious"
-8ea structure. ith these notations remark that
the concept of a-extension of in (1.2) is the
same with the concept of D-field in (0.). The
ncept of 6--gebra in (l.) is tbe same with that
of D-algebra in (0.). T ncept of A-ideal in a

6--alEebra in (1.) is the same with the con,pt


of D-ideal in a D-alEebra , cf. (0.7). A 6-gebra is A-finitely 8enerated in the sense of
(1.) iff the corresnding D-algebra Js Dfinitely 8erierated Jn the sense of (0.}). By
(1.6) and (1.7) is D-algebraically closed in
the sense of (0.}). By the ave discussion, any
integral D-scheme of finite ty can be covered
by D-schemes of the form Sc {yl,...,yn}/3
where 3 is some ime a-ideal. (2.2) In order
to relate the Ritt-Kolchin frame (cf. Section I
of this Chapter) to our frame in Chapter 1- it
will be convenit to introduce a bic functor
{integral D-schemes of D-finite type} { amanifolds}, V V This is done follows.
Assume V is an integral D-scheme of Dfinite type then a set we put V A = Horn
D_sch(Spac ,V) Clely V6 identifi with the

subset of V consisting of aJJ p E V for which


the maximal ideal mp of V,p is a A-ideal and
t resid field V,p/mp is a trivial extension of .
In tations oi (0.7) we actually have V =VD().
The subsets of Va of the iorm Ua = =
HOmD.sch(5cU) where U is any ZariskJ
open subset of V are the open sets o a
topology on V a which we call the a-tology.
For any U as ave we t peU Any element f in
(U&) can be Jdent[fied with a function f =
U& + via the formula f(p) = (f rood top) . 5o
we obtained a subsheaf to of the sheaf of 1 valued functions on V . Cl this subsheaf V
then we leave to the reader the tk to eck that
(Y&, V ) is a -maniold and that actually we
lot a functor [rom {integral D-schemes of Dfinite type} to { &-manifolds}. The key int to
bo checked is actual]y the fact that it V =

5pec{yl...yn}]3 for some ime &-ideal 3 then


(V&, V& ) (Z Z) as naive ringed spaces over
108 109 where r. = Z(3) t,n is the zero set
of 3 and (r. is the sheaf o:f A-regular
functions on I: (this shows in particular that
any affine A-manifold has the form (Spec
JR) A where R is an integral A-:finitely
generated A- -algebra. It also worths being
noted that the stalk V A, p of V A at p V A
identi:fles with and that the quotient field /.
(V) of V identi:fies wlth ,(r.:. Consequently
the :functor V V induces a functor {Dvarieties} + { A -manifolds} o Now the
:fupctor V V A commutes with products so it
induces funclots G G A {integral D-group
schemes o:f D-:finite type} + { A-groups}
{irreducible algebraic D-groups} - { Ao-

groups} few easy remarks about the functor


V V A are in order 1) If X is a D-variety then
X 6 is ZarJski dense in X. 2) I:f X - Y is a
surjective D-map of D-varieties then the map
X A e Y A is surjective. 3) l:f X is a Dvariety and Y c X is a closed D-subvariety
then X A n Y = Y A (as subsets o:f x). (2.3)
]t will be use:fuJ to express the functors X +
and X + o defined in (1.12) and (].13)
respectively Jn terms.of the funclot V6
above. Start by recalling that we introduced
in (l. 2.1) a functor {reduced ,-schemes} +
{reduced D-schemes} X k X e, which is an
adjoint for the forgetful funclot V I. Now
note that if X = $pec/J.[yl,...,yn]/3 then X m
=Spec {yl,...,yn}/{3}; this can be checked by
sJm ply checking that HomTj,_sch(V!,Spec ,
[yj,...,yn3) =

HomD_sch(V,Spec{Yl,...,yn}/{3}) for any


reduced D-scheme . But by Kolchins result
quoted in (l.]2) if 3 is prime then so is {3}.
C:onsequently if X is an /,-variety then X is
an integral D-scheme of D-finite type so we
dispose of a funclot {/,-varietles} +
{integral D-schemes of D-finite type} X + X
o We leave to the reader the task to check
that the composition of the above functer
with the funclot in (2.2): {integral Dschemes of D-finite type:} - { A-manifolds}r
V H A iS isomorphic to the :functor in ( I.
12) & (J, -varietie - { A -mani:foida} X ,- X
In other words we have a natural
Isomorphisms 4 X' (Xm)A (2.) Similarly we
disuse o a unctor { -varJie { D-varietie deJn
by X o Xo where Xoe is viewed with the
split D-scheme structure (L 3.6). [t is ey to

check that the comsJtJon o this unor with the


funclot { D-varietJ + { A o-manifold, V V A
is isomorphJc to the functor in (]. J 3): { rJetie * { o-manifoJd, Xo o In other wds e
have natural JsomorphJsms Xo= (Xo) A ()
'Similily y irreducible klgeaic -group G we
have a natur isomorphism o A -groups (G)A
and for any ieducible gebraic -group G
we ve a natur isomphism o A o-grou - (c o e
) Next e epare ourselves to ove that the
te8ory o irreducible algebraic D-roups is
equivalent to the.category of & o-rou. The
Jrst step is the olJowin: (2.6} LEMMA. L a
& -group. The oilowJn[ are equivalent= t) is
a & 2) = C& some irreducible gebraic Droup G. 2)]) Js obvious. J) 2) Pick any aine
Piece = (SpecR)& o T with R a & -finitaly
8enated integral 6 --algebra. It is a fact [B3]

that any such R whose quotient field s finite


transcenden
110 111 degree over contains an element f
R, f 0 such that A := If is fin{rely generated
as a (non-differential) l,-algebra. Performing
a translation we may assume that (Spec ^)&
contains the unit e of I'. So in particular the
multiplication and inverse of I' induce on (V
= SpecA,e) a structure of germ of algebraic
group ("groupuscule algebrique" in the sense
of [U] in fact we have rational maps V x --, and V---V defined at (e,e) respectively
at e enjoying the usual properties of group
law with e es a unit whenever composition
of maps makes sense). Now there exists an
algebraic group G whose germ at the identity
is isomorphic to the germ (V,e) (cf. [U]; this

can be seen as follows.' V has in particular a


normal law of composition in Well's sense
hence by Weil's theorem there is an algebraic
group G and a birational map f: --+G such
that the transported law fore V to G via f
coincides with the law of C. Now one
checks that f must be defined at e, f(e) = e
and i commutes with multiplication and
inverse arround e, so f gives an lsomorphism
of germs (V,e) (G,e)). Transporting the Astructure from V to G via the iso- morphism
(V,e) e (G,e) we get m commuting
derivations pl,...,Pm on (C) which agree with
comultlplication ?.I(G) ' 2(G x G) and
antipode (G) + ?(G) (preserving in fact both
(G,e and its maximal ideal). By (L 1.20) Pie
P(G) so by (I. l.l) G becomes an algebraic
D-group. We see that we have a A-rational

map r--,G& (i.e. a morphism of 6-manifolds


from and open subset of r to G&) which
agrees with multiplication whenever
operations make sense. Exactly as in the case
of algebraic groups such a map must be
everywhere defined on r and our ]emma is
proved. (2.7) LEMMA. Assume G and G'
are irreducible algebraic I-groups. Then the
natural map Homalg. D.gr(G,G,) + Horn A
.gr(G A ,G'&) is bijective. Proof. The
morphism G& + G'i induces a commutative
diagram of local &-rings = %,e, xG,exe-This extends uniquely to a commutative
diagram in which the horizontal arrows are
D.rationai maps of D-varieties (i.e. [-rnaps
defined on open subsets of the sourse).' GxG
---+G' x G' G ..... ,G' The bottom horizontal
arrow must then be an (everywhere defined)

morphism of algebraic D-groups and our


lemma is proved. (2.g) COROLLARY. The
functor G ' G A {irreducible algebraic Dgroups} -+ { Ao-groups } is an equivalence
of categories. Proot. It is just (2.6) + (2.7).
REMARK. We shall fix from now on a
quasi-inverse I' ,.G(T) of the functor in (2.8).
have Horn &_gr (r,r') HOmalg
D.gr(G(r),G(r')) Moreover <r> identifies
with t//,<G(l'>) and (r,e identifies with
Oc(r),e' (2.9) Clearly a do-grou p I' is split if
and only if G(l') is a split algebraic D-group.
But by (I. 3.7) the latter happens if and only
if C.(I') is locally finite. So by (IV, 1.2) upon
letting G = G(r) ! we have r is split iff the
Images of l,...,m 6 c p in P(G) belong to
P(G,fln) (recall that any algebraic D-group
structure on G is determined by a Lie /[-

algebra map P * P(C), cf. (i. 1.1)). (2.10)


LEMMA. For any do-grou p r and any
irreducible algebric -group H we have a
natural bljectJon Horn &.gr (l',lq) = HOmalg
.gr(G(T)!,H) Proof. By (I. 2.1) we have
HOmalg ql. gr(G(r)l,H) HOmD.g r sch(C(
r),H ) Since I' = G(I')& there is an obvious
map HOmD_g r sch(G( I'),H ) - Horn figr(r,lq) Let's construct an inverse to it{ start
with f e Horn &_gr(l',Jq). Then f provides a
&-ring map Composing this map with the
natural map (H ,e '" (.,0H .,,e we get a map
H,e '+ (G(I'),e
112 113 hence a rational map G(I')I+ H
which is easily seen to agree with
multiplication generically and hence is an
(everywhere defined) homomorphism. This

homomorphlsm provides a morphism G(T)+


Hm and our inverse is constructed. (2.11)
LEMMA. For any irreducible algebraic ]'groups G o and H o we have a natural
bijection v v Homalg_gr(Go,H o) = Horn A gr(Go'Ho ) ProoK By (I. 3.6) we have a
bijectlon HOmalg .gr(Go,Ho) = Homalg
D.gr(Go , H o /,) and we conclude by (2.7).
(2.12) One can define in the obvious way
actions of A-groups on A-manifolds (and on
fi-groups) so one can define A-cocycles of a
A-group I' in a A-group I" on which T acts.
Some more definitions: a morphism y -+ I"
of A-groups will be called an embedding if
it is injectlye. A A-group y will be called
linear (respectively abelian) if there is an
embedding I' '+ GL N for some N > I
(respectively I' '+ for some abelfan t,-variety

). A A -group I' will be said to have no nontrivial linear representation if for any N > I
the only morphism y + GLiN is the trivial
one. By (2.10) T has no non-trivial linear
representation if! G(y) has the same propety
i.e. iff (G(I')) =q. Let 1' be a A-group. Two
embedclings o': T'- I', o" = I'"- l' are called
equivalent if there is an isomorphism : T' T" such that o"o = o'. A class of equivalence
of such erabeddings will be called a fisubgroup of y. By abuse we will sometimes
say (given an embedding I"+ Y)that l" is a Asubgroup of I'. (2.13) PROPOSITION. Let 1'
be a Ao-group and let G = G(y) {. Then there
is an embedding I' + C and a A-cocycle C +
(L(G)m) ^ inducing an exact sequence of
pointed sets I - y + G('[{,) + L(G) int -+ 1
where the action of C on (L(G)m)^= (L(C)' m

is induced by adjoint action (L(G) is viewed


as usual as an algebraic vector group) and
L(G) int = ((e I ..... o m) c L(G)m; ? 6i ) ? 6je i * [ei,e j] = 0 for all i,j} where V here
is the adjoint connection (I. 1.10). Proof. It's
just a translation of (I. 2.3) via the yoga of
the present section. REMffkRK. The above
proposition answers (in the finite
dimensional case) ]<ochJn'S question from
the Introduction of [K2] whether differential
algebraic groups can be embedded into
algebraic groups. (2.1) In what follows by a
Hopf fi -/d,-algebra we mean a Hopf 9alebra H which is a fi -,/,-algebra such
that comultiplication, antipode and unit are fi
-algebra maps; this concept is equivalent o
course to that of Hopf D-algebra. There is a
functor induced by that in (1.2); { integral A

-finitely generated commutative Hopf A-algebras}- { arline fi -groups} H (Spec H)fi


Any fi -group isomorphic to an object
contained in the image of the above functor
will be called strictly arline. It can be shown
(but won't be done or' used here) that our
strictly arline A-groups "are" precisely
Cassidy's -algebraic groups with Apolynomial law [C 1] and consequently by
Cassidy's results [Cl] any strictly arline A group is linear. The latter result for fi ogroups will be reproved below. The
following result is due to Cassidy [C3] but
we provide for it a direct proof,' (2.1)
LEMMA. For any two integral fi -finJteJy
genera/ted commutative Hop! fi - ,/,-algebras H and H' there is a bijectlon.' HomHopf
fi. .alg(H',H) = Homfi _gr((Spec H) ,(Spec

H') A ) Proof. The natural map from left to


right is clearly injectlye. To check
surjectivity take a morphism f: (Spec H) +
(Spec H')fi; it induces a morphism Writing
both H and H' as direct limits of fJnitely
generated Hopf algebras [DC] one
immediately sees that the latter morphism is
induced by a morphism of arline group
schemes Spec H- Spec H' hence by a Hopf fi
- t-algebra map H'- H and we are done.
(2.16) Due to (2.15) we may define correctly
(up to an isomorphlsm) for any strictly arline
f -group 1' a f -finitely generated Hop! fi -algebra { I'} such that I' - (Spec?Z( I'} )fi {
this { 1 } can be called the f-coordinate
Hopf algebra of I' (and coincides with
Cassidy's { I'} in [Cl]D. It does not coincide
however with (I' ) in general{ (2.17)

LEMM/, [B3]. Let I' be a strictly arline fi ogroup. Then ( I'} is finitely generated as a
(non-differential){,-algebra. In particular
G(I')=Spec{I'} and ,{l'} =O(G(]')).
Moreover I' is split if and only if /,{ I'} is a
locally finite D-module. Proof. We already
used in (2.6) the fact (proved in B3]) that if l'
=(Spec H)fi then the scheme Spec H contains
a non-empty open subset of finite type over .
A translation argument shows that one can
cover Spec H by arline open sets of finite
type over ,, hence Spec H is of finitey type
over ?d,, hence H is finiteJy generated over .
The remaining assertions follow
114 115 immediately from (2.8)and (2.10).
(2.18) COROLLARY. A A o-group [' Js
strictly arline if and only if G([' )! is afine, if

and only if T is linear. Proof. The first


equivalence follows from (2.17). The second
one follows from (2.17), (2.13) and (2.10).
(2.19) REMARK. By (1.25) we see that
there exist arline Ao-groups T which are not
strictly arline (e.g. I' = A o with A an
abe]ian '-variety; this example was first
discovered by Cassidy). Let's analyse a few
EXAMPLES It worths describing explicitely
the algebraic D-groups associated to the A ogroups introduced in Examples 1), 2), 3)
given after (1.12). 1) Assume first thatZ =
Z(y". ay'. by)c a =' a,b. Then G) = G_ 2 =
Spec[ j2 ] viewed as an algebraic D-group
via the derivation c P(G) defined by [1 =[2
and [ 2 = -I 1 ' a['2 i.e. by where 6 * is the
lifting of 6 to %1 1,[ 2] which kills [ 1 2.
This structure is of course a special case of

the corresponding example in (1.9), (1.26).


To check that G 2 viewed as an algebraic a
D-group as above is indeed isomorphic to Z
we must check that (Ga2)fi = r.. But (Ga2)fi
is by delinition the set of all /,-points p = 1 2
) c G 2 for which the evaluation map ?Z 1 ;
23* ' is a a D-algebra map hence for which a
1 --a 2 a = -t l - 2 i.e. for which a' l =a 2 We
checked that (Ga2) E as sets; the reader may
check that this is an isomorphism oi manifolds. Note that e<Z)=?(y,y') where y is
the class of y.{ y} rnodry,. ay'4 by]. Under
the identification J.(y,y') = 1 2) we have y = [
I' y' = [ 2' ^ 2) Similarfly if we assume E
=Z(xy- l,yy"- (y,)2 + ayy')c G m=Z(xy- l) 6 2
as in fl.12) then one can check as above that
G([)= G m x G a = Spec [X'I[] as In (1.9),
(1.26) viewed as an algebraic D-group via

the derivation P(G m x G a) defined by )t


=rX and i.e. by Once again we have t< ) =.
(y,y,). Under the identification t(y,y,) =(X[)
we have = y and [ = y'/y. 3a) Finely if we let
be the fi -closure in 2 of Z(y 2 - x(x - l)(x c),x"y - x'y' + ax'y)c c2=A 2 then one can eck
as ave tha G()=AXGa=AxSpec] as in (1.9)
(with A = Zariski closure of Z(y 2 - x(x - l)(x
- c)) c 2 in p2) viewed as an algebraic Dgroup via.the derivation P(A x G a)
defined by x = [ y and [ = - hence by =,* *yWe have ) =(x,y,x') under t identification
(x,y,x') =(x,y)we have [ = Note tt in 1 three
exampl o fi o-grou appear embedded in ldimensional algebraic -groups (me ecisely in
a'' and e given by second order difentl
equations. Thee emddings are entirely
different from the emddings provided by

(2.13) above (whi should view as the


"natural" embedclings, in which t ambient firoup should b, x, x,Zx). Conere,ely, in
example I)Z is given in x by the first order
system =0 In example 2). is given in m x a by
the first order system X' -X = 0 ['+a[ =0 In
example 3)I' is given Inkx a by the first order
system x' -[y = 0 ['+a[ =0 We are not yet
reedy to compute G([') for T as in Example
3b) o! (l.i2). See (3.20) for a treatment of it.
But we are ready to compute the & coordinate Hopf algebra =?J.{Z} of the & group g = t I , GL3();c'--O,a'=ca 0
116 117 in Example t)of loc.cit. Indeed, as a
/[-algebra R ={a,b,C}a/[C',a' - ca] =[a,a'
l,c,b,b',b",b"',...] The computation ai I I 0 I 2
= I c I c 2 0 0 0 0 I gives the following

formulae for the comultiplication p '. I + R


R'. a=aa pc=cl+lc pb =ab + b ] p(b9 -cab + ab + b 1 p(b") = c2ab + 2ca b' + a b" +
b" 1 Let R n =[a,a'l,c,b,b',...,b(n)]l it is a
Hopf subalgebra of R so we may consider
the arline algebraic group G n = SpecR n ald
the arline group scheme G = Specie. Put T -[a,a-l]. Then each G n contains T and has a
projection onto T. Moreover ker (Gn+ 1 + C
n) is a vector group for all n. Consequently
each G n is solvable hence a semidirect
product of T by the unlpotent group , b(n) M
n -- Ker(G n + T) = 5pec[c,b,b ,,.., j 5o (3
itself Is a semidirect product M x p T, M =
im M n. Now note the remarkable fact thai a
e Xm(T) is a weight for p but is not killed by
6; this is the example promissed in (I].3.6)!
3. Stntttte e! be-'oups Since we proposed

ourselves not to use Kolchln's powerful


theory [K 2] we are faced ,ith the problem of
constructing a '.ie correspondence" and a
theory of subgroups and quotients for bgroups. We are able to do this only if we
restrict ourselves to o-groups. We begin with
sketching such a theory. (3.1) Let 1" be an
embedding of &o-groups. We shall define a
b-manifold structure on the set T/T' as
follows. By (2.2) G(r')()* G(rX) is injective
so G(')+ G() is a closed immersion hence by
(I. hiS) C(I')/G(I"') has a natural structure o!
D-variety. One checks using (2.2) that the
underlying set of the ao-maniloJd (G(
r)/(G(r')) A coincides with r/r*. So we give
T/T the above Ao-manifold structure. If in
addition T' Is normal in T then by a density
argument (c. (2.2)) we get that C(I")(/) is

normal in C(T)(,) hence G(T') ls normal in


G(r) hence C(I')/G(T') will be an algebraic
D-group so I'/T' will be a Ao-grou p. The
reader can check that in act the above r/r,
satisfies the usual universal property o
homogenous spaces (respectively quotient
groups). (3.2) Clearly (I. 1.16) provides the
desired *'Lie correspondence" for Aogroups; we leave its formulation to the
reader. (3.3) Let's discuss commutators. By
(I. 1.6) If r is a Ao-grou p then [G(I'),G(I')]
is an irreducible algebraic D-subgroup o!
G(T). Then using (2.2) one checks that the
underlying groups of the Ao-grou p
[c(r),G(r)] A coincides with the closure of
[P,T] in r. So we denote the above Ao-grou p
by [r,r]cr (3.e) Let's discuss centers. By (L
1.6) i! 1' Is a &o-group then ZO(G(p)) ts an

irreducible algebraic D-subgroup of G(P).


Using (2.2) one checks that the underlying
group o the Ao-grou p z((r)) coincides
with the connected component Z(l') of the
center of I'. 5o we give Z(r) the above ogroup structure. (3.) Some words about
isogenJes. A morphism of o-groups l' + I"
will be called an isogeny if it Is surjectlve
with inite kernel. By a density argument
again of. (2.2) the homomorphlsm G(I')
(L)+G(T')(,) Is surjective; moreover, if H is
the Identity component of K er(C(l') +G(I"))
by (2.2) we have H = H(/,) n l' is dense In
H(,), this showing that H() consists of one
element only. Hence G(I') +G(l") must also
be an isogeny. Here is one of the main
applications; (3.6) THEOREM. Let I' be Aogrou p. Put G = G(r) and identify ,4 l'> with

(G>. Then 1) l' is split if and only if I'/[ I'1']


is split. 2) is split if and only if the maximum
semlahellan subfield of Z(I') Is a A-subfield
of 1< r>. 3) If r * r is an isogeny then r is
split if and only if I is split. t) If Z(r) =
{e}then r is split. ProeL 1), 2) tollow from
(2.9), (3.3) and (IV. hg). 3) follows from
(2.9), (3.5) and (IV. hS). #) follows from
(2.9), (3'.t) and ([L 3.30).
118 119 (3.7) A linear A-group r will be
called solvabie (respectively nilpotent) if it
is so as an abstract group. Note that a linear
&o-group Y is solvable (respectively
nilpotent) if and only if is so (this :[ollows
from "density", cf. (2.2) and from (2.13)). A
A &-group will be called unlpotent If there
exists an embedding r +GL N whose image

:onsists ot unipotent matrices (this definition


agrees with Cassidy's [C2]). Note that a &ogroup r is unlpotent if and only if (3(T) is so;
indeed if G(r) is unipotent so is F by (2.13).
Conversely if y embeds into GL N and r
consists of unipotent matrices of GLN(/,)
then one can easily show that the action of r
via translations on the algebra /,.{GLN}=
(GLN m) is locally unipotent hence so will
be the action of r on /,{ T} = ((G(r)) hence
so will be the action of G(r)() on ((G(T)) (by
"density" again) and we are done (see [H] p.
64 for background of locally unipotent
actions). Let T be a linear &o-group. By
(2.18) G(r) is arline and by (1I. 3.30) its
radical R(G(T)) is an irreducible algebraic
D-subgroup of G(r). Ve define the radical of
T by the formula R(r) := R(G(r)) We have an

embedding R(r)+ r and it is easy to check


using (2.8) that for any embeddin I - r of
linear &o-groups with r' soJvabJe and
normal Jnr there is a commutative diaEram I
a ,r R(r) with (unique. Here is another
consequence of our theory. (3.8)
THEOREM. ^ linear Ao-group r is split i!
and only if its radical R(T) is split, if and
only if all group like elements of the. Hop!
A-U-algebra {F} are A-constants. Proof. It
follows directly from (3.7), (2.9), (1I. 3.30).
In what follows we briefly discuss abelian
Ao-groups: (3.9) I. EMMA. Let C be an
irreducible algebraic D-group, H an
irreducible'algebraiC ,-group and C - H e C
) + H two morphisms which correspond to
eachother under adjunction (I. 2.1). The
following are equivalent.' 1) The kernel of G

- H eo is trivial. 2) The kernel of G I +H


contains no non-trivial algebraic D-subgroup
of G. 3) The morphism G + fl is an
embedding. Proof. 1) :O3) is clear. 2)==H)
follows since Ker(G-H is an algebraic Dsubgroup of G contained in Ker(G ! - H). 1)------------------:2) Assume M is an
algebraic D-subgroup o! G contained in
Ker(G !+ H). Then both the trivial Dmorphism and the D-morphism :M+G+H m
composed with (H m)I + H give the trivial
morphism. By universality, -- gJ so M is
trivial. (3.10) LED. MA. Let r be a Ao-grou
p. The following are equivalent: I) r is abel
Jan. 2) There is a D- morphism G(r) + A m
with trivial kernel with A an abelfan variety.
3) The linear pert of G(Y)! contains no nontrivial algebraic D-subgroup of G(r). ) Any

morphism from a linear Ao-grou p r' to r is


trivial. Moreover t if the above conditions
hold then the linear part of G(T) ! Is
unipotent. Proof. Let B be the linear part of
G ! (G = G(F)). 3)==)2) follows from (3.9)
applied to the projection C ! G!/B. 2):=H)
follows from (2.10) and (3.9). 2) =3). We
have a commutatIve diagram G "A providing
a commutatlve diagram G $ :A' (c/s) - Since
has a trivial kernel, so has Applying (3.) to
we get our conclusion. Note hat if 2) or 3)
hold G is commutative so by (L 1.22) the
maximal torus B m of B is an algebraic bsubgroup of G hence is trivial so B is
unipotent. 3) ) If 1 '+ r is as in ) then the
image of G( I +G( l ls an algebraic Dsubgroup of B hence is trivial So G( r +G( D
is trivial so P'+ ris trivial. 4) ==)3) Since [

I'Tcl IS linear It Is trivial so r is


commutatlve. By (L 1.22) once again B m is
an algebraic D-subgroup of G. By 4) B m is
trivial. Now assume 3) does not hold hence
there exists an algebraic D-subgroup I o! C
contained in B with H 0. Since H is unfpotent
It Js irreducible. Letting r= H 6 we get a
contradiction. Our lemma is proved.
120 121 (3.11) COROLLARY. 1) Any Aogrou p ]' possess a linear Ao-subgroup r such
that is an abel Jan go-group. 2) Any ^o-group
I' possesses a Ao-subgrou p I' 1 with no nontrivial linear representation such that ['/I' I iS
a linear Ao-grou p. Proof. l) Take I'*= C' b
where C' is the largest irreducible arline
algebraic D-subgroup G(1') and use (3.10).
2) Take I' 1 -- (G l ) g where G [ = Ker(G -

Spec d7(G)) and use [DG] p. (:3.12)


REMARK. The first of the above assertions
answers, in the particular case of ^o-groups ,
Kolchin's question in the Introduction to [!<
2] whether a "ChevalJey structure theorem"
[Ro] holds for differential algebraic groups.
In the case of %-groups one sees that the role
of abelian varieties among algebraic groups
is played by abelian go-groups. This justifies a special study of the latter. Note that we
have the following inclusions of sets modulo
isomorphisms reTresentation modulo
isomorphlsm corem I = commutatlve &ogroups modulo isomorphism It is easy to
show that qab qli,4F(tak e the preimage
via logarithmic derivative A(L) -L(A),
where A Js an abelian -variety defined over
:, of a A -subgroup of L(A)) and o that qe

q1ab (take a split 6o-grou p l' such that


((G(l')) = and G(l') Is not an abelian variety).
In what follows we concentrate on objects of
the set of isomorphism classes of abelian
&o-groups with no non-trivial linear
representation. (3.13) LEMMA. Let F be a
&o-subgroup of I with H an irreducible
algebraic U-group. The following are
equivalent.' 1) Flu Zariski dense in H. 2) The
morphism G([") ! -H (cf. (2.10)) is
surjectJve. Proof. Use definitions. (3.1)
LEMMA. Let A be an abelian i,-variety and
l' +2 a morphism from a ^o-group The
following are equivalent: l) I'ls a Zariski
dense b-subgroup o!/. 2) l'is a -subgroup
of/(' containing the torsion subgroup of A. 3)
The morphism G(l) I -^ (cf. (2.10)) is
surjective, its kernel coincides with the

linear part B of G(T )I and B contains no


non-trivial algebraic D-subgroup of G(I' ).
O. 15) COROLLARY. Any abelian Ao-grou
p I has a ZarJski dense embedding I' binto
some produced abelian variety . Moreover
for any two such embeddings l: I' 1, 2: I' 2
there exists an Jsomorphlsm o: A I + A2 such
that Preef of (3.1tO. In view of (3.9), (3.10)
and (3.13) in order to check that 1):3) it
sufficient to check that under the assumption
that contains no non-trivial algebraic Dsubgroup of G(I') it follows that iV equals
the linear part B of G(I'). Indeed by (I. 1.22)
M must be linear hence Mo= B (actually
by 1oc.cit. B must be an algebraic vector
group). We want to prove M = B; assume
iV/B 0 and look for a contradiction. Take x
iV, x B, Jet N = Iu/BI hence Nx B; choose (B

being divlslblel) be B such that Nb = Nx.


Then x - b Tors(G(I')) hence by (I. 1.22)
again the (finite) subgroup of G(I') generated
by x - b is a non-trivial algebraic Dsubgroup of G(I') contained in M,
contradiction. So the equivalence 1) 4=)3) is
proved. The implication 2)=1) is obvious.
Conversely to check that 1)::)2) it is
sutficient to prove that any torsion point x
A() lifts to a D-point of G(T). But any x as
above lifts to a torsion point of G(I') which
is automatically a D-point due to (I. 1.22)
again. (3. I&) Let I' be an abelian Ao-grou p.
By its embedding dimension we will
understand the dimension of any abelJan variety A such that there exists a ZarJskJ
dense ba ]' . Clearly the embedding
dimension of I' equals the dimension of the

abelian part of G(I') [ (cf. (.l), (.l)). (3.17)


Recall from (IlL l.l), (IlL 1.3) cf. [KO] that
for any abeltan -variety A, H1DR(A) has a
D-module structure induced by Gauss-ManJn
connection. Call DH({21A/) the Dsubmodule of HR(A) generated by
H(i]lA/). Moreover recall from (III. 2.15)
that the universal extension E(A) has a
(unique) structure of algebraic D-group
hence we dispose of a naturally associated
Ao-grou p E(A)g, c. (1.9). Here is one of our
main applications of the theory in (1II. 2):
(3.1) THEOREM. Let A be an abeltan 2variety. Then contains a unique Zariski dense
6o-Subgroup with no non-trivial linear
representation (call it ) and we have the
formula Moreover ^ =W= is a quotient of
E(^) A. ProeL Let I' be any Zariski dense

Ao-subgrou p of , Such l"s correspond


bljecttvely to equivalence classes of
surjectlons C ! +A (G an irreducible
algebraic D-group) whose kernel is the
linear part of C ] and contains no non-trivial
algebraic D-subgroups of G, cf. (3.1a), (2.8)
two such surjectlons C[+A, G-^ are
equivalent if there Is an algebraic D-group
iSomorphism (= (;i + (;2 making the followln
diagram commutative',
122 123 (3. is.l) d ^ Now assume in addition
that equivalent to G( [')! having the same
property). By (3.10) and (III. 2.5) G(I') ! is
an extension of A by a vector group B such
that the map Xa(B) +HI(( A) is iniective. $o
G(T) ! is lsomorphic over A with a quotient
of the universal extension E(A) be some

linear group. By Oil. 2.2) the $urjection


E(A)+G([') is a D-morphism. In view of our
equivalence relation (3.18.1) we may
assume G([')= EGA)/H for some linear
algebraic D-subgroup H of E(A). e claim
that o 1 SDR(E(A)/H)a=DH (fl/tL); indeed
SDR(E(^)/H)a contains H(fllA/,) and
since it is a D-submodule of cf. (m. 2.2), it
must contain Assume o l DH (O./,) is strictly
contained in SDR(E(A)/H) a. Then by (UL
2.14) we have o I .L where = Ker(E(A)+A).
Now DH (/,U.) = L(Hi) for some
intermediate algebraic vector group H l
between H and I. By (I. 1.16) H l is a Dsubgroup of E(A). But this leads to a
contradiction since we would get a nontrivial linear Ao-Subgroup (HI/H) A
embedded .into the abeltan Ao-group

(E(A)/H)A = = DH(alA/%) from which


we get that dim r. For the existence of a
Zariskl dense ouf> [' in with no non-trivial
linear representation we simply let ]'_(E(A)/H) A where H is the maximum linear
agebraiC D-subgroup of E(A) and use (3. l).
(3.19) COROLLARY. Let A be an abellan q,varJety. Then the following are equivalent
(notations as in (3. J7) (3..lg)): I) E(A) . is
an abeltan Ao-grup' 2) ^' = E(^) 3) dim'^ -2dim A (3.20) EXAMPLE. Assume dim A =
! (A an elliptic curve). Then only two
posslbilities may occur for DH(/) namely:
a)DHO(f//,)= o 1 (/). In case a) we get by
(ill. 3.#) that A is defined over , say A =^o ,
A an abeltan (,-variety. Then clearly A= A
o !n case b) we get by (3.19) that A -- E(A)
A and A does not descend to . , Note that in

the latter case A q is the unique Ao-SUbgrou


p of dimension 2 of (lndeed by (3.14) for any
Ao-subgrou p T o! of dimension 2, G(l') is
an extension of A by Ga; it cannot be a trivial
extension by'(l. 3.1) so tt must be the
universal extension E(A)). As an appication
of this unicity remark consider once again the
Ao-Subgroup l' of (where A - Zariski closure
in p2 of .Z(y 2 - x(x - l)(x - t))c /,2 t'= l)
obtained by taking the A-closure Jn 2 of Z(y
2 - x(x - i)(x - t),-y 3 - 2(2t - i)(x - t)2x'y +
2t(t - l)(x - t)2(x"y - 2x'y')) as In Example
3b) in (1.12). Since dim ['--2, by the unicity
remark above we must have [' = A i' = E(A)
A so G([')--E(A) viewed as an algebraic Dgroup via the unique Jilting of 6 Der/, to
P(E(A)) whose explicit formula can be
obtained as explained In the Example after

(IlL 2.19). It would be nice to "see" the


equality [' = A directly, without using the
above "unlcity remark". In particular since
r(l') =(x,y,x') and ,(E(A)) =(x,y,[) (with [ as
in (Ill. 2.1)) it would be nice to "see" the
:formula expressing as a rational function in
xyx . (3.21) REMARK. More generally It
would be interesting to disPOse for any
abeltan -varlety A defined explicitely by
some algebraic equations in the projective
space pn of a method to get algebraic
dl:fferential equations defining A inside A.
(3.22) REMARK. If we :fix an abelian variety A there may exist plenty of Zariski
dense Ao-subgroups of A; they may be of
arbitrary dimension. For example take A an
elliptic curve defined over put G = G N x A
and assuming for simplicity that In ordinary

(A -- {6}) a . let 6 # be the trivia] lifting of 6


from to (3 and e H(T^) be a non-zero
global vector field. Then letting [l,...,[ N be a
-besls of Xa(Ga N) and T 1 ..... be the
corresponding basis of a De.(G). Then by (I.
1.9) P(G)so L(CaN) , put -- 6#+[10+
[2[1+...+[1 defines a structure of
algebraic D-group on G. It is esy to check
that there 1s no intermediate field between ,
(A) and (G) preserved by . Then G A is an
abelian Ao-grou p embedded in . e. Moduli o
ho-Sroupa We propose ourselves to find
descriptions of the set q = ( Ao-groups
modulo isomorphisms} or o:f remarkable
subsets of It. We start with "set-theoretic
descriptions". /e shah continue by putting 6manifold structures on such sets. By (2.8) the
set q/t can be written as a disjoint union rt =

J I qR(G) c
124 125 where G runs through the set of
isomorphism classes of irreducible algebraic
-groups and for any such G we denote ql(G)
the set of isomorphism classes of Ao-groups
I' such that G(I') ! = G. So one reasonable
thing to do (although as we shall see not the
only one) is to fix G and try to describe I(G).
Here are some easy consequences of our
theory: (.1.) LEMMA 1) Assume G is affine.
Then (G) is non-empty if and only i! G is
defined over . 2) Assume G is commutative.
Then 11,(G) is non-empty if and only if _
S(C), (i) uer t.A, for all i. 3) Assume G has
no non-trivial linear representation. Then
9R(G) has at most one element. Proet. Use
(II. 3.31) and (IlL 2.18). Now we claim that

for any irreducible algebraic -group G we


have.' IL(G) - P(c)int/Aut G where P(G) int
is the set of m-uples (pl,...,pm) of pairwise
commuting elements in P(C) such that (pi ) =
(6 i) for I _( i _ m, while AutG acts on P(C)
tnt by the formula (o,(p l,...,pm) ) (o-i -I +
plo,..,O PmO), oe AutG. whereo stands also
for the induced automorphlsm of .(G). To
check our claim note that by (2.8)q(G)
identifies with the set of AutC - orbits in the
set of algebraic D-group structures on G; but
now we are done by (L i.l). The above
description of qqq,(C) is of course not
satisfactory. To get a better picture assume
for a moment that is a field of definition for
G so where G:ls some algebraic -group. * *
acts trivially on (G)). Then Let 8i be the
trivial liftln 8of a(6i) from to G (as usual 6i

the map P(G) m + P(G/,) m (Pl ..... Pm) "(h '


- Induces a bijection qL(G) = P(G/Z)lnt/Aut
C where 1) P(G/) Int is the set of m-uples
l,...Om ) of elements els P(G/ )such that V
,Oj - V ?i + iej ] = 0 for all l,j 2) Aut G acts
by the "Loewy-type" action [Cl]: He (VV) is
the logarithmic P(G)-connection on (Aut
G,P(G/%)) defined in (I. 1.2) so recall that
This dcription is unsatisfactory the first one
cause everything is "hidden" by the "algebraic action" of Aut C; what one should
be looking is a description involving an
algebraic (rather that a " -algebraic") action.
Such a scription will be given in the
Theorem below in a scial case. (.2)
EOREM. L G an irreducible algealc -8roup,
G = G and sume (G) ntains a representative
ideal V. Then we have an identification

where 1L(G) = lnt/Aut C, l) V int = {


(al,...,a m) $ vml V - ,a. - V 6; ai + [ai'aj ] =
0 for all l,j}. 6j i ' 2) Aut G acts on V jnt by
the formula b,(a l,...,am)) + b' la lO,... '
JarnO) Proof. The inclusion vint+ p(G/%) int
induces a map b: vint/Aut G P(G/,)int/Aut G
We must check b Js a bJjection. To check
injectivity assumee = (e l,...em ) and 0' -j,...en) are ^ut C-conjugate elements in
P(CJ,)int, hence ior some element Aut G.
Since V is an Aut C-invariant subspace of
P(G/) we have o--]op v
126 Since P(G,fin) is an Aut G-invariant
subspace of P(G) too, we have o'16 7o - 67
P(G,fin) Since V n P(G,fin) -- 0 we get
a}e =v'lef for all i I' :o'lo for all i. From b)
and (I. 3.0, 1) we get v c Aut G). By a) 0 and

0' are Am -conjugate so, is injectlye. To eck


that is rjecive ake0 = l,...m) P(G/) ira. We
may write 01 = d i + a i with di P(G/,fin), ai
V. We claim that (dl,...,dm) P(G/) nt. Indeed
we have in P(G)the equaliies: Since V is an
ideal In P(G) the lt 3 terms of t6e right hand
side of the above formula belong to V while
he first erm 1ongs to P(G,fin), because
P(G,fin) is a Lie /-subalgebra of P(G), (IV.
1.2). We conclude above that the first term
vanlshes which prov our claim. By this clai
we may consider the gebraic D-group
swucmre on G efine by he map P , P(G) cL
{I. I.D. By {2.) thi* algebraic D-group
tructur on G i ,plit hnc by (2.8) ther exi,ts o
Aut G such that We get that hence 0 is Aut
G-conjugate to O' lalo,...' lamo) V lnt and
our Theorem is proved. (.3) The above

theorem giv a rather satisfactory answer to


the classification problem f linear fi o-grou.
Indeed for any linear fi o-group Y, G = G(F
)! is alline by (2.18). By (II. 1.1) G is
defined over and by (II. 1.22) P(G) contains
a reprentative ideal. So Theorem (0.2)
applim to GI We ge he following bijtion lin
== { linear f o-groups modulo isomorphism}
= (V(G)int/Aut ) G 127 where the disjoint
union is taken over all isomorphism classes
G oi irreducible arline algebraic -groups
defined over J and V(G) is a representative
ideal in P(G). Moreover note that if G is
such that all its weights are killed by P(G)
(e.g. if either the radical of C is nilpotent or
the unipotent radical of G is commutative)
we have by (IL 3.16) and (II. 3.22) that
V(G)int/Aut G;)(, = Wo(G)int/Aut Gij where

Wo(G)int = {(al,...,a m) c Wo(G)i 61*aj =


6;a i for all i,j} (67 being here the trivial
lifting 1 of (61) from to W.o(G)--Wo(Gj) and
AutG acts on Wo(G)int via its action on
Wo(G( ' ). (.) As already noted, by (2.10) a
go-group r has' no non-trivial linear
representation If and only 1! G(r I) has the
same property. Then (4.1) implies that we
have a bijection go-groups with no nontrivial linear representation modulo isomor
phisin s(c) C(G) =[/.sucl'that )(6 i) Der q
for all i, modulo isomorphism (it.5) Assume
G = I. x A where L is an irreducible arline
algebraic group and A is an abelJan variety.
Then by (IV. 3.1)(G) is non-empty if and only
if is a field of definition for both L and A,
say A A.j L L3y ' )q. Assume this is the case
and assume in addition for simpllcJt'y that

P(L) -- P(L,fin) (e.g. the radical of L is


unipotent). Then we have by (IV. 3.1) and
(4.2) that rlt,(G) = (Xa(L) L(A))int/Aut L
x Aut where (Xa(L) L(A)) int = {(al,...,a
m) Xa(L) ) L(A)I 67 aj = 6 a i} ( 67 being
the trivial lifting of (6 i) from , to Xa(L)
L(A) = (Xa(Lv ) ;, L(A)) 1 and Au! L, x
Aut A acts via its natural action on Xa(Lj1 )
L(A: ). (4.6) Let's give a description of
tab, := 71,ab n fil?= [ abellan go'grups
with n non-trivial linear representation
modulo lsomorphism Indeed by (3.19) we
have a decomposition into a disjoint union I
l_<g<_d_<2g ,g where
128 qLd,g = i Ao-groups in 11%ab'= of
dimension d and t embedding dimension g
Moreover by (3.19) we have a bijection f

abelJan -varieties A of dimension g with the


property t d,g" ) = d modulo lsomorphlsm
{t.7) Up to now we discussed only "settheoretic" descriptions of "toodull sets". But
there is an interesting possibility that some of
these moduli sets (or some natural
"coverings" or "pieces" of them) carry
natural structures of A-manifolds. For
instance if G is as in Theorem (3.3) (with
ordinary, to simplify discussion) then qr(G)
is of the form where ^ is an arline locally
algebraic ,-group acting on n via a
representation ^() P , CLn()c GLn(.) where p
is locally algebraic. Now ^ has infinitely
many components in general so our task of
:HndJng A-manifold structures is hopeless
unless we pass from Ln/^() to its "covering"
q,n/^o() (the fibres of the proiection /n/A()

- n/A() are then at most countable). The study


of such quotients Ln/A() was part of a
program initiated by B. Weisfeiler; some
general theorems were proved by him and
lots of conjectures can be made at least or A
reductive. In any case what one should
expect is that, exactly as in the "nondl:ferential" invariant theory only an "open
subset" of ,n/^o() can be equipped with a AmanHol d structure. Fortunately for many
linear algebraic groups G we dispose of a
quite explicit description of the action of
^utC on a representative ideal (cf. (a. 3)) this
leading to a direct easy description of (G) =
V/^utC;g the latter turns out aposter[ori to
contain a "big" piece which is a 6-manifold.
We [lustrate this with one example (the
reader can consider more complicated

examples and try to apply our theory to deal


with them): (t.8) THEOREM. We have a
natural bijection t,(Ga N x GNm) ,
glN(,)/(GLN(Z ) x GLN()) Moreover the
subset 1 = GLN()/CLN() of the set 129
identifies (via logarithmic derivative J'?6"
GLN() * glN( )) with glN(). 5o has a natural
structure of A-arline space , N2. Poog. In
notations of (II. 3.1)) (II. 3.12) we clearly
have: Wo(G) -- W(G) = Horn( 2: N, ,N) =
gin () while AUt(Ga N
xGNm)=GLN(Z)XGLN(). By (#.3)we get
our formula for [(G xCNm ). The rest
follows from KoJchln's surjectivity of [76
(cf. (I. 2.3)). (t.9) We close our search for
"6-manifold structure on moduJJ sets" by
considering abel Jan 6o-groups with no nontrivial linear representation i.e. the set 4%ab'

cf. (#.6). Once again the example of abelJan


varieties from algebraic geometry teaches us
that if one expects toodull one has to
introduce polarizations. In what follows by a
polarized abel Jan Ao-grou p with level n
structure we will understand the equivalence
class of a triple ( r A ,,,) where r is an
abelian Ao=group T +A is a Zarlski dense
embedding (A an abelian -variety), 1 is a
polarization on A and rt Js a level n structure
on A; the triples (l: ri +A l,).i,), i = 1,2 are
called equivalent if l' l --r 2 = r and there is
an isomorphism o.'A 1 +A 2 such that 1 -- c2
' )'2 = )'1 and o(1) -- 2' A triple as above
will be said to be associated to r. There is an
obvious notion of lsomorphism of such
objects. Moreover a triple as above will be
called principally polarized if ). is a

principal DoJarization (see [/u] for


backgraound). Note that given an abel[an 6ogroup r there are at most countably many
triples as above sssociated to r, modulo
isomorphism, cf. (3.1). Let ld,g,n be the set
of lsomorphism classes of principally
polarized abelian Ao-groups with level nstructure having the following properties.' 1)
they have no non-trivial linear
representation, 2) they have dimension d, 3)
they have embedding dimension g. In what
follows we shall prove tht these ld,g,n (for g
_ 2 and n _ 3) have natural structures of Amanifolds. They are non-empty only if g _( d
_( 2g. The case g = I was left aside just
because Jt is too easyl we leave it to the
reader cf. (3.20). On the contrary the
presence of level n-structures (n _ 2) Js a

technical condition which we could not get


rid of. Finally we expect that the case of nonprincipal polarizations can be worked with
along the same lines as we do for principal
onas principal polarizations simplify
discussion at a few points. We will prove in
fact much more namely that dg,n can be
identified with certain subsets of the
corresponding moduli spaces of abel[an
varieties which are locally closed in the G
topology o these moduli spaces (o! course
they will be not locally closed in the
130 131 usual ZariskJ topology). To state our
main result let's make some notations.
Denote by ug,n b the toodull space of
principally polarized abelJan .-varieties
with level n structure [Muj tn tn O o a -

varlety. Then for g < d < 2g the natural maps


tld,g,n + g,n (I' + ,,,n) ' (A,X,rO are injective
by (3. I8) and their images d,g,n are disjoiner
and described as polnts in g,n which
c9rrespond to abel Jan d,g,n = [varieties A
such that dim.,.DH(P./q4) = d t cf. (3. IS),
(a.6); note that %,g,n and d,8, n can be
Identified sets. Here is our main result. (.10)
EOREM. In tations ave, i g 2 and n 3 then: l)
F any g d 2g the sets dgn are n-empty and the
sets ,g,n u g+J,g,n u "' u d,g,n are -ci in gn' 2)
e have ,g,n - d' ,g,n u ... u 2g,g,n = ,n (.11)
Note that the ua]ity =o oUows rom (]I. .).
The rest o the g,g,n g,n aerlons require a
prerafion. (a. 12) Let Y be any -variety
denote by TY V(Ty) Spec(S(/)). It co[ndes
with the langent bundle o Y (vewed a
variety) in case Y [s smooth. Assume or

simplicity this is t case. Aume morver that Y


s defined over and [x an JsomorphJsm Y Yo
Te claim that or each p E P there is a
natural map o A-manolds To define it it is
sufficient to define a D-map Vp Y +(TY)
hence by (0.9) and (l. 2.12)' It is sufficient to
define for any (reduced) D-algebra R maps
Vp .' Y(R) +(TY)(R) behaving- functorially
in R. We define them as follows. For any
morphiSm f $pec R +Y still denote by f the
map induced between the corresponding
sheaves of algebraS, denote by PRc Derl)t
(R) the multiplication by p c P and by p#
Derl (Cy) the trivial Jilting of a(p) from ( to
Y = Yo ( then the difference pR f - fp* will
be an ,-derivation from (fly to f, (flSpecR
hence will induce an y-linear map hence a
morphism SpecR +TY which we 11 pf. More

generall for .any faroill pl,P2,...,P we have a


mphism of A-manifolds (pl, 9p2,...) Y *(TY
x yTY x ...). (.13) Clearly to prove (.10) we
ed a meted to study the "variation" of the Dmodule structure of HR(A) A varies In a
mulJ space. Remarkably this can done by
translating t D-module orti of HR(A) in
terms of the Gauss-Manin connection of any
"deformation of A which is defined over ".
This is what we do next. +Y be a smooth
projectlve morphism of smooth (-varieties,
assume (.l) Let L '* Xo o its fibres are
connected and let II: X *Y be obtained from
X o +Yo by tensorlzatJon with Let moreover
y () be any 9-point of Y and X be the
fibre of X +Y at y. Y By (III. 1.1) we dispose
of a Gauss-Manin connection which is an lJnear map. X X R(Xy)'H)R(Xy))' Y (O.

la.l) V y: De --'*Hom>(,(H d * d On the


other hand one can consider the relative de
Rham sheaf HLi(X/Y):-- = Rl]I((/y)l recall
from [Ka] that thls Is a locally free (ymodule and its formation is compatible with
any'base change, in particular HR(X/Y)e (y)
H)R(XyX(y) -- residue field of Y at y).
Denote by HR(X/Y)y the stalk at y and by ry,
HR(X/Y)y * HR(Xy) the reduction modulo
the maximal ideal my of (y,y. By [Ka] we
also dispose of a Gauss-Manin connection v
(defined in the same way as the one in (III.
1.1)) inducing at y an (flyby - linear map
(.]4.3) ?X/Y,y, DerrbL (,y) +HomtL(H)R(X/
)y ' HR(X/Y)y), o !+ ?/Y'Y The map (4.
lq.l) should be thought of as an "internal"
Gauss-Manin connection while (q. lq. 2) as
an "external" one. Note that we can see T Y

as a subset of TY (cf. (4.12)) it Is of course


the fibre of the Y projection TY+Y at y. In
particular denoting by fy-'$pac+Y the
morphism defining Y Y((.) and al,o the
induced morphIsm fy (yy ' we may consider
by (4.12) for any p P the -point ?pfy TyY c
TY dellned by (.la.a) ?ply = a(p)fy. fyp#
eDer(y,y,() _- TyY
132 133 where p* is the trivial lifting of B
(p) from /, to Y, viewed here as an element
of Der (y,y). Choose moreover any
derivation e Der (y,y) "lifting" ?ply i.e.
with the property that (.lt.5) V f = f e py y
Finally, let p## be the trivial lifting of a(p)
from , to X. It induces an obvious -1inear
endomorphism ?p# of HR(X/Y)(any
derivation of (X lifting a derivation of )

does!). The following lemma shows that the


"internal" data can be recaptured from the
"external" ones. Philosophically thls is
possible because of our hypothesis that the
"deformation" X/Y of X is defined over ,
hence is "sufficiently rich". Y (4.15) I.
EMMA. The following diagram is
commutative vp., </'y , H.(X/Y)y ry I r X
Y HR(Xy ) ? (Yp) HI__(X ). DR y Praof. Put
C/ = $pec(y,y, = X x y , cover by open affine
subsets i = SpecAi' lift 0 to derivations 6 i
DerA ^i' consider the cocycle (6j - e l)
C l(( i)i, T / ) and reduce it modulo my. to
get a cocycle (el - 6 i) CI(Xy,i,T X ) Y n
and the upper bar means "reduction modulo
m "(note that 0j cannot be where Xy,i = Xy i
Y reduced modulo my, so one cannot speak
about j, because 0j may not preserve my;

only the differences ej - 0 i preserve it,


indeed they vanish on it). By (t. it.t) and
(t.l.5) we get the equality a (p)fy = fy(p* + 6)
which shows in particular that p#+ e
preserves my. Hence its liftings p'e+ eiDeiA
i preserve my so they can be reduced modulo
my to get derivations p*# + O i Der ' i
where l )(Xy,i)' We get the equality ej - e i =
(p** + el) - (p** + e i) Now 'choose a
representative (oi,xij)of a class
HR(X/Y)y-- l-fdrms on i and xij R( are H
/C/ ) where ,i :C( i n j). Then ry(VX/Y'Y)T
is represented by (.) (Lie0i 0 i,(0j. el) , J +
Oixij ) x Moreover ( ap))ry is represented by
(p## + 0i) )mj + (p*# OiXlj)) (**) (Liep.# +
01t0i,((p ## + el). + while ry?p. 1] is
represented by (***) (Liep** i' P**Xij) and
our lemma follows. Now for any Pi in a

commuting basis pl,...,p m of P let's fix a 0 i


as in (t.l) I.e. with f = Ply fYOi' (4.16)
LFMM^. In notations of (t. lt) and (3.17) let I
be the subring of End. (HI,u(X/Y).) --. dl '-"
Y generated by . and the maps 7i TM ?p? + ?
/Y'y' I < i < m./non we have the equality ~
el21 DHO(t ./i,) = ry(D H (X/Y)y) Y Proof.
Let D N be the L-linear subspace of D
generated by {, and the products ... (k < N)
and let [N be the -1inear subspace of I
generated by and the PilPi 2 Pi k - products
of the form VilV12 ... Vik (k< N) We check
by induction on N that o 1 (#)N DNH (i]X
/%1, ) = r. (I_ H(l,,.,,). ) y N ^1 y y For N
= 0 this is just the base change theorem.
Assume equality holds for N - I and let o 1 e[
and i any index between I and m. Moreover
let's agree to denote eH ( X/Y)y , N-I X g Y

again by V i (no confusion will m'rise with


the previous notatlons{). Then we have by
a(p i) (.l): ry(?ic = (?j)(ryC ) But by
induction hypothesis (?iXry Z ) e
V.D...H(i]. 1. ... ) CDNH(i]<y/ ) so we
get e o 1 ry( ?lC r DN H which proves the
inclusion ":' in (*)N' The inclusion ,,o,
follows by an entirely similar argument.
134 135 (.17) To formulate the next Jemma
let's Introduce some new constructions. In
notations from (.1) let Y' be an open subset
of Y whose tangent bundle is trivial and fix a
trivialisation T, Ny,. Associated to this
triviaJisation consider the natural map : TY'
+ H(Y',Ty) (:[or each vector v in some
TyY consider the vector field on Y' obtained
by displacing v in the other points of Y' via

the trivialisation of Ty,). Then for any point


(y,v)=(Y,Vl,...,v re) sTY x...x TY, y Y', v i
TyY t Y . Y , m factors we can consider the
-endormorphism V l(Y,V),...,Vm(Y,V) of
H)R(X/Y)y defined by the formula .X/Y,y I
<i< m Vi(Y,V ) = Vp? + v x(Y,Vl) - _ and
define {y,v to be the subring of
End(HIDR(X/Y)y) generated by , and
Vi(Y,v), I _<1< r. Then we have (t.18)
LEMMA. !n notations from (#.17) and (.la)
for any integer d the set .. HO; 1 ' d} Dis
d = {(y,v) TY x x TY; y Y', dimry( y,v
X/Y'y' - Y Y is A.closed in TY x ... x TY. Y
Y Proof.' We may assume Y = Y' shrinking Y
(the question is local) we may assume
H(t.l.,. ) is free and a direct sumsand in
HR(X/Y) which is also free. We also may
write /'/ N TY = Y x , and choose a frame of

vector fields Wl,...,w N on Y. The map : TY'


= Y' x %N + Ho(y,,Ty) has then the form
(y,Xl,...,X N) = ZXiw i The elements of TY x
... x TY are then pairs (y,X) where X = (Xij)
is an m x N matrix with Y Y elemetns in , and
Vi(y,X ) = Vp + j!iXijVw) Finally, fix a
basis el,...,e n of H(fi(/y) as an (y-module
el,...,en,...,e k of HR(X/Y); with respect to
this basis we may write Vp= p+ M i V wi+N
i W. which extends to a basis where Mi, Nj
are matrices with entries in ((Y). All we
have to check is that for any fixed matrix (ai(
of elements in ((Y) and any integer r the
following set Is A-closed: (,,) {(y,X) ey x
2/,raN{ dim<ryde(Zaiaei) where < > means
"-1inear span", (x = (Ol,O2,...) are
sequences of integers of length {(x I and d.=(p . + + '-Vl V2 "' I I jalj(Wj + NJ))

(P2 + M2 jX2j(wj + Nj)) ... Now the


coefficients of the ei's in do(Zaioej) are
polynomials Jn (pi 1) ... )(Pik)). whose
coeffcients are(fixed) functions in ((Y). The
conditions defining (#) are obtained by the
vanishing of certain determinants whose
entries are the coefficients of the ej's Jn
d((.ajoj) evaluated at various points y Y'.
This closes the proof. (e.l) Let's check
assertion 1) in Theorem (q. lO). As well
known the moduli space Y = dg is smooth (n
> 3!) and there is an abelJan scheme E: X +
Y (the "universal amily"). Moreover X,Y, E.
are defined over the prime field Jn particular
over so we are Jn the situation of (.1). The
statement of (/.10), 1) is local on Y so we
may replace Y by some aftins subset where
the tangent bundle is trivial. Noting that in

notations of (/.17), 1ywi) ' H(Ty)lifts (by


its.very definition) the tangent vector v i
TyY we get by (.16) that Y o1 UH(l X ./q)
= ry([(y, Vply ..... Vpmfy)H (X/Y)y) it
follows by (/.9) that Cd,g,n is the inverse
image of Dis d rom (.18) via the map.. Y() +
(TY x ... x TYX[) Y Y y + ('? f ,..., ?pmfy)
Ply The latter map being a map of Amanifolds (/.12), closedness o 0d,g,n in the
A. topology :[oliows from (.l To conclude
the proof of (.10) Jt remains to produce for
each d with g_< d_< 2g an abel Jan variety A
d belonging to rd, g,n. For this let E be an
elliptic curve not defined over the field o!
constants of 61, say, let F be an elliptic curve
defined over 3, and put A d = E d-g x F 2g-d.
(.20) It would be interesting to know whether
Theorem (/.i0) holds for i (n (2 (especially

for n = i since "level l-structure" means


simple "no level structure"l). So assume n _<
2. Then assertion 3) in (. i0) still holds for
Identical reasons. Moreover the sets (.)
Og,g,n u UJ, g+l,g,n u ...u d,g,n are the
projections via ,3 '+ Lg of the corresponding
sets for n= 3. The latter are known to
136 137 be A-closed by (t.10). But in spite
of the fact that the map tg,3 ' tg,n is finite it is
not clear to us whether it takes A.closed sets
into A-closed sets. Note that a "Chevalley
constructibility theorem" holds Jn our context
[B2]B 3] so the sets (,) are at least
constructlble in the A -topology! (4.21) The
loci d,g,n deserve further study. They stratify
(in the A-topology) the moduli space in a
non-trivial way and bare a formal

resemblance with the degeneracy loci in


,,Brill-Noether theory". It would be
interesting to compute their A-dimension
polynomials [K 1 ], to investigate their
irreduclbllity or speciallsation properties
(e.g. is td,g,n contained in the A-closure of
Ld+l,g,n?). APPENDIX A. LINK WITH
MOVABLE SINGULARITIES In this
appendix we discuss the relation between
algebrlc D-groups and A-function fields with
no movable singularity (NMS) in the sense
of [B 1] (recall that by the latter we mean an
extension of D-fields F c E, where D = F[P],
P a Lie F/k-algebra, such that E = F<V>
where V is some wojective D-variety. For
the understanding of the discussion below, a
certain familiarity with [B 1] is preferable.
We shall assume from now on that all Lie

K/k-algebras occuring have a finite


commuting basis and all fields have
characteristic zero; K will be always
algebraically closed. There are easy
examples showing that if P is a Lie K/kalgebra, D = KiP] and G is an algebraic Dgroup then the extension K c K<G> need not
have (NUS); for instance P. Cassidy proved
that this is the case with G =
Spec(K{y}y/[yy" - (y,)2]) (note that G ! = G
a x Gin). On the other hand we can wove:
(R.I) PROPOSrrloN. Let G be an irreducible
algebraic D-group all of whose weights are
P-constant. Then there exist a Picard-Vessiot
extension K i/K and an intermediate D-field
K cEcK <G) such that E/K is split (ie E-K
(EP)) and KI(G>/E is regular (l.e. E is I I 1 ''
- 1 algebraically closed in K i<C>) and

generated by exponential elements (i.e.


elements of K whose logarithmic derivatives
belong to E). For the concept of Picardessiot extension see [K 1 ], Chapter 5. We
will also wove the following Woposltlon
(where we recall that an element x of a Dfield extension E of F Is called a PicardVesslot element if x is contained in a finite
dimensional D-submodule of E). (A.2)
PROPOSITION. Any D-field extension E/F
(with F algebraically closed in D = F[P], P a
Lie F/k-algebra) is generated by Picardessiot elements has (NMS). So we get by
[B 1] p. 103= (A.3) COROLLARY. In
notations of (A. 1) all three extensions K cK
1 cE cKi<G> have (NM5). (A.4) Proo o!
(A.I). Denote by 61,...,6 m a commuting
basis of P and let Pi be the image of 6i in

P(G). Since we assumed all weights of G are


P-constant we get (in notations of (II. 3.16))
that a i := l?i w(G) so by 1oc.cit. we
may write Pi = 1 +di+E a i wre P(G) is the
trlvl lifting oi (i ) from K to G=Go K K
(Ko:KD , Goa Ko-group), d i P(G/K,fin)
and'a Ea is the map rom Wo(G) to P(G)
appearing in (11. 3.1). Exactly as in the
proof of (V. .2) the derivatlo (&[ di) i are
rirwise commuting so by (L 1.1) they define
an gebralc D-group Structure on G. By the
proof of (I. 3.11) we can find an automphism
c Rut(C K i ) where K I/K is a PicardVessiot extension such that So we may
assume the images of i in P(G K 1 ) have the
form ] + Ei Let U and H as in (IL 3.21) and t
E = KI(U); clearly E is eserved by D 1 =
KiKD and E 1 is split. oreover the extension

K i<C) is genoated by elements of the torre 1


y with y O(Ho) x (where H = Ho K K, H
o a Ko-group, notations in (!. 3.21) again)
and any such ly is an exponential element in
the extension Ki<G)/E cause its logarithmic
derivative with respect to Pi is ai(x) e 1 E (i
y G(Ho)X). Our prosition is proved. .5) P
.2). Let E/F a regular extension of D-ields
generated by Picd-Vessiot elemts. Then
clearly a finite set of them suffices so E =
F(X) where X = ScA d A is some Dsubalgebra of E, finitely generated an Falgebra and locally iinite as a D-mule. Let V
c A be a inite dimension D-submodule
generatin 8 A as an F-algebra. Then the
symmetric algebra SV h a natural structure of
D-algebra and the closed embedding X + Y
= Sc SV is a D-map. Give the lynomi algebra

(SV)[t] a structure of D-algebra extension o


5V by lcttJn 8 = 0 for J p P. Since D
prerves the gradation on (SVt] tt follows that
q = Proj((SV)[t becom a D-variety and the
open embeddin 8 Y + is a D-map. Let X be
the Zariski clure of X in . By (0.1), 2) X is a
D-subvariety of O and this closes the proof.
Usin 8 (IV. 3.1) one can prove (in a way
similar to (A.I) above)-' .6) PROSmON. Let
G be an algebraic D-group and assume its
underlying K-group G is the product of a
unittent linear group by an abelJan variety.
Then there exist a Plcard-Yessiot extension
Ki/K (which by [B 1] has (NMS)) and an
intermediate D-field
138 139 Ki c E c Ki<G> such that E/K 1 is
regular and split (hence by [B 1] has (NMS))

and K i<G)/E has (N4S). We close this


Appendix by briefly inspecting the relation
with Painleve's property [GS], [Ida]). We
need an analytic preparation: (A.7) Let f: X
+ Y be a map of analytic complex manifolds
and suppose we are given analytic
commuting vector fields 81,...,6m on Y
lifting to some commuting vector fields 6 on
X. We say that ! has the Painleve property
(with respect to these vector fields) if, upon
Jetting + Y be the universal covering of Y,
there is an analytic -isomorphism XXvC/
Zx'7 Z f'l(yo)' Yo Y' sending into 6. where
is the unique lifting of . from X to X x = i '~ J
' and 6# is the trivial lifting of 6. from Y to Z
x Y Note that if dimY = m and if we throw
away from Y the locus where 61,.,6 m do not
generate the tangent space and from X the

preimage of this locus (we assume that we


are left with some non-empty open
submanifolds!) then we get a foliation on X
transverse to f and bein[ x ire a
"feuilletage de Pamleve de espece" in the
sense of [GS], i.e. having the property that
any path on Y starting from Yo Y can be
lifted in the leaf passing through any point of
f'l(yo). (A.S) Let's assume in what follows
that K is an algebraically closed field
containing C with tr.deg. K/C = m <-- and
put P -- Der cK, D =K[P]. Let pl,...,pm be a
commuting basis of P. Let moreover V be a
D-variety. We say that V is a Palnlev Dvariety if there is a cartesian diagram of C schemes.' S . Spec K with U and S smooth C
- varieties and j dominant and if there are
commuting vector fields 6 6 m I ""' m I '"" on

S lifting to commuting vector fields on U


such that 1) Each (pi) Der C K lifts 6i and
the image of each Pi in Der C (V lifts i' / 6 m
and 2) fan uan + san has the PainIcrc
property (with respect to 61 "'" I ""' m
viewed as analytic vector fields on S an,
uan). Finally let's say that a D-field extension
E/K is a PainIcrc extension if E -- K(V)
where Is some painlevY'D-variety. The
following hold: a) Any smooth projectlye Dvariety is a Pamleve D-varety in particular,
any extensio" E/K with (NM$) is a
Painievextension. This Is a trivial
consequence o! Ehresmann's theorem [3]. b)
A consequence of the above remark and of
our result in [B 1] p. 103 is that any strongly
normal extension of K is a Pmnleve
extension (as it has (NMS)). c) Let G be an

irreducible algebraic D-group. Then K(G)/K


is a Patrdeve extension (more precisely C is
a Painlev D-varietyl). This follows from
Hamm's result (11. 1.3). cf. [Ha]. d) Painlev
extensions need not have (NMS) for they
may have "movable singularities hidden at
infinity". For instance K. Okamoto's work (in
Sero. F. Norguet Fvrier 1977) shows that the
famous Palnleve second order equations
"lead" to Pamleve extensions; these are not
necessarily NMS extensions (as proved by
Nishioka iN]). APPENDIX B. ANALOGUE
OF A DIOPNANTINE CON3ECTURE OF
S. LANG The following result was proved
in [B6] and illustrates the "diophantine
flavour" of Ao-groups: (B.I) THEOREM.
Let A be an abelfan /,-variety with /-trace
zero, X c A a subvari- ety and Y c a Ao-

subgrou p. Then there exist abelfan


subvarieties Bi,...,B m of A and points a I ,...,
a m A such that m 1' nXcL./(B i.a i) cX
i=l The proof relies on analytic arguments
(namely the Palnlevff property in (A.8), c)
plus'a "Big Picard Theorem" due to
Kobayashl-Ochial). Using (B.I) we deduced
in [B 6] the following (B.2) COROLLARY..
Let be an abelfan variety over an
algebraically closed field F of characteristic
zero, assume A has F/k-trace zero for some
algebraically closed subfield k c F, let X c A
be a subvariety, T c A a fJnitely generated
subgroup and Ydiv = {x eA; ] n such that nx
} Then there exist abelJan subvarieties
B 1 .,. B m c A and points a I .,. a m A
such that m 'Ydlv n X c i=kl (Bi + ai) c X
The above statement without the trace

condition was conjectured by Serge LanE;


recently C. Galtings proved a weaker form
S. Lang's conjecture in which Ydiv is
replaced by y. Note also that in [B?] we
proved an "infinitesimal analogue" of (B.I).
APPENDIX C. FINAL REMARKS AND
QUESTIONS Ve start with some intriguing
short questions:
140 141 (C.I) Does there exist non-affine
Ao-groups? (compare with (V. 2. la)). We
think there should be; a candidate would
perhaps be E(A) A where A is an elliptic
curve over whlch Js not defined over v.
(C.2) Let G be an irreducible algebraic Dgroup (D =K[P], char K = 0). Is the linear
part of G defined over KD? This is crucial
for understanding P(G) for arbitrary (non-

linear, non-corn muratlye) G. (C.3) Let G be


an arline algebraic D-group (D =K[P],
charK = 0). Are the weights of G necessarily
P-constants? (see (fl. 3.12) for background).
This is quite significant since we dispose of
a very explicit description o! affJne
algebraic D-groups all of whose weights are
P-constant, cf. (IL 3.16), (V. .3). (C.e) A hard
task is to extend our theory from Chapters 1by shifting from Spec K to a more general kscheme, in particular a k-variety 5 (k = C)
and build a bridge between our theory and
Deligne's theory of linear differential
equations with regular singular points
[DeJ2]. Vector bundles on S with
connections are in fact group schemes over S
(with fibres vector groups) on which the
algebra of differential operator D S on S acts

compatibly with muJtJpHcatJon, inverse and


unit. So the theory we expect is an analogue
of Deligne's in which algebraic vector
groups Ga N are replaced by more general
algebraic groups. More precisely, Deligne's
theory in [Del 2] shows in particular that
there is an equivalence between the
"analytically defined" category (#)top (local
systems of vector spaces on S an } where S
Js a fixed smooth algebraic C - variety and
the "algebraically defined" category (*)alg
{regular D S - modules on S } (here vector
spaces are assumed finite dimensional and
DS-mOdules are assumed 09s-coherent
hence locally free). One is then tempted to
inspect relations between the "analytically
defined" category: (,e)top (local systems of
Lie groups on S an } and the "algebraically

defined" category: (e e)alg {"regular"


algebraic Ds-groups } Here an algebraic Dsgroup should be of course a group scheme
over S on which D S acts compatibly with
multiplication, inverse and unit, while
"regular" should mean at least that the DSmodule o! relative Lie algebras is regular in
Deligne's sense (compare with our regularity
criterion (lIl, 3.9)). Note that local systems
of Lie groups were studied in [Hal. We pose
here the problem of their algebralsation. It is
natural to start with the simplest examples of
Lie groups hich are not vector spaces. Let's
take for instance local systems on S an
whose fibres are (Ce) M (these are
equivalent to representations of ]{j(S an) in
CLM(Z)). What local systems of this type
come from algebraic Ds-groups? Of course,

group shemes which become tori after


passing to a finite etaJe covering of S lead to
finite monodromy groups. But at least for
even M = 2g one can get infinite monodromy
groups by taking non-trivial abel Jan
schemes f: A S of relative dimension g and
then consider their extension E(A)'+S by
(Rife (A)v; there is natural structure of
algebraic Ds-grou p on E(A)/S whose
monodromy is that of Rlf#anZ. Of course the
fibres of E(A)/S are analytically Jsomorphic
to (Ce) {W. There are {owever topological
restrictions :[or I an R fe Z (e.g. if S Js a
curve the latter should be quasi-unipotent at
infinity). This shows that "algebraisatJon"
should not be always possible. It is certainly
not unique either. So the taks is to replace
(ee)to p and (ee)alg by some other related

categories which should be equivalent to


eachother and then to develop a dictionary
between the topological and the algebraic
setting. (C.$) In [B 8] we proved the
lollowing theorem; for any Trreducible
algebraic -group G there is a morphism of /groups L N (for some N > 1) whose kernel is
a (commutative) Ao-group'with no nontrivial linear representation. This shows how
the study of &-groups appearing as &subgroups of G for some algebraic .-group G
"reduces" to the study of linear Zl-groups
and of &o-groups with no non-trivial linear
representations. So we are provided with a
strong apesteriori motivation for our .study
of Ao-groups with n non-trivial linear
representation! [BBM] CsS] [aft [B 3] [s 7]
REFERENCES P. Berthelot, L. Breen, /.

Messing, Thorie de Dieudonn cristalHne II,


Lecture Notes in Math. 930, Springer 1982.
A. Borel, 3.P. Serre, Thormes de finitude en
cohomoloRie galoisienne, Comment. Math.
Helvetici, 39(196#), l 11-16. A. Buium
Differential Function Fields and Moduli oi
Algebraic Varieties, Lecture Notes in Math.
1226, Springer 1986. A. BuJtwn, Ritt
schemes and torsion theory, Pacific 3. IUath.,
98(1982), 281-293. A. Bulm, Splitting
differential algebraic groups, 3. Algebra,
130, 1(1990), 97= 105. A. Bukmm,
Birational moduli and non-abelian
cohomology, l, Compositio Math. 68(1988),
17-202. A. Bulem, Birational moduli and
non-abelian cohomology, II, Compositio
Math. 71(1989), 2#7-263. A. Bulum,
Intersections in jet spaces and a conjecture

of 5. Lang, to appear in Annals of Math. A.


Buium, Infinitesimal study of differential
polynomial functions, Preprint INCREST
1990,
142 143 [B s] [c 1] [c 2] [c 3] [c A] [Del l
[DeI 2 [D] [DC] [OS] [Ha] [Har] A. Buiwn,
Geometry of differential polynomial
functions, Preprint INCREST 1990. P.
Cassidy, Differential algebraic groups,
Amer. 3. Math. 9(1972), 891-95a. P.
Cassidy, Unipotent differential algebraic
groups, in Contributions to Algebra,
Academic Pros, New York 1977. P. Cassidy,
The differential rational representation
algebra of a linear differential algebraic
group, 3. Algebra 37(1975), 223-235. P.
Cassidy, The classification of the semisimpie

differential algebraic groups and the linear


semisimpie differential algebraic Lie
algebras, 3. Algebra P. Deligne , Theoreme
de Lefschetz et Griteros de degenerescence
de suites spectrales, Publ. Math. IHES
95(1968), 107-126. P. Deligne, Ecluations
Differentfolios Points Singullets IReguhers,
Lecture Notes in Math. 163, Springer 1970.
M. Domezra'e, Sch6mas en groupes rductifs,
Bull. Soc. Math. France, 93(1965), 369-a
13. M. Domazure, P. Gabriel, Groupes
Algbriclues, North Holland, 1970. I%.
Grothendieck, Fortdemerits de la Geometric
Algebrque, Sero. Bourbaki 1957-1962. R.
Gerard, A. 5ec, Feuilletages de Palnleve
Bull. Soc. Math. France, 100(1972), t7-72.
A. Grothendleck, M. Domazure, Sche'mas en
groupes I, Lecture Notes in Math. 131,

Springer 1970. H. Harem, Differential


equations and local systems, Conference at
Steklov Institute in Moscow, March 1989. R.
Hartshorne, Algebraic Geometry, Springer
Verlag, 1977. G. Hochschild, Basic Theory
of Algebraic Groups and Lie Algebras,
Springer 1981. [HM l ] G. Hochschild, D.
Megtow, On the algebra of representative
functions of an analytic group, Amer. 3.
Math. 93(1961), 111-136. [HM 2] G.
Hochschild, D. Mostow, Analytic and
rational automorphisms of complex algebraic
groups, 3. Algebra 25, 1(1973), 1t6-152. [3]
[3] [Ka] [KO] [Kei] [K l ] [K] 3. 3ohron
Prolongations of integral domains, 3.
Algebra 9(I) 1985, 173-211. 3.P. 3euanolou,
Equations de Pfaff Algbriques, Lecture
Notes in Math. 708, Springer 1979. N. Katz,

Nilpotent connections and the monodromy


theorem; applications of a result of Turfilia,
Publ. Math. IHES 39(1970), 175-232. N.
Katz, T. Gala, On the differentiation of the de
Rham cohomology classes with respect to
parameters, ;L Kyoto Univ. 8(1968), 119213. W. Keigher, Adjunction and commonads
in differential algebra, Pacific 3. Math.,
59(1975), 99-112. E. Kolchin, Differential
Algebra and Algebraic Groups, Academic
Press, New York 1973. E. Kolchin,
Ditferential Algebraic Groups, Acaclemlc
Press, New York 1985. [K 3] [Ma] [Mo]
[M] [Mu] [NW] [Li] [R] [Ro] [Se] [Sei]
Sw] IT] [U] IV] E. Kolchin, Constrained.
extensions of differential fields, Adv. in
Math. 12(197#), 141-170. :I.1. Manin,
Rational points of algebraic curves over

function fields, lzvestija Akad. Nauk SSR,


Mat. set. t. 27(1963), 1395-1ttO. M.
Matsuda, First order algebraic differential
equations, Lecture Notes in Math. 804,
Springer 19g0. H. Matsumura, F. Oort,
Representability of group funclots and
automorphisms of algebraic schemes, Invent.
Math. t(1967) 1-25. B. Mazur, W. l%!essin8,
Universal Extensions and One Dimensional
Crystalline Cohomology, Lecture Notes in
Math. 370, Springer 197tL D. Mumford,
Geometric Invariant Theory, Springer 1965.
K. Nishioka, A note on the transcendency of
Painleve's first transcendent, Nagoya Math.
3. 109(1988), 63-67. W. Nichols, B.
Weisfeller, Differential formal groups of 3.F.
Ritt, Amer, 3. Math. 1982, 943-1005. D.
Lieherman, Compactness of the Chow

scheme, Sero. Norguet 1976, Lecture Notes


in Math. 670, Springer 197. 2.F. Ritt,
Differential Algebra, Amer. Math. Scl.
Colloq. Publ. M. Rosenlich Some basic
theorems on algebraic groups, Amer. 3.
Math. 78(1956), 01-463. 3.P. Serre, Groupes
Algbrlques et Corps de Classes, Hermann,
Paris 1959. A. SeidenberF,, Abstract
differential algebra and the analytic case I,
II, Proc. AMS 9(1955), 159-16 and
23(1968), 689-691. M. Sweedler, Hopf
Algebras, Benjamin, New York 1969. M.
Takeuchi, A Hop! theoretic approach to the
Plcard Yesslot theory, 3. Algebra. H.
Umemura, Sur leg sous-groupes algebriques
primitifs du groupe de Cremona a variables,
Nagoya Math. 3. 79(1980), #7-67. E.
iehwe 8, Weak positivity and additivity of

the Kodaira dimension II: the ToreIll map,


in: Classification of Algebraic and Analytic
Manifolds, Birkhauser 1983, 567-58a.
INDEX OF TERYlNOLOGY Lie K/kalgebra ................................................. (0.5)
algebra of differential operators
..................................... (0.5) D-module, Dalgebra, P-module .....................................
(0.5) D-scheme
..................................................... (0.6) of Dfinite type ................................................
(0.6) D-variety
..................................................... (0.6) Dideal, D-point ................................................
(0.?) D-group scheme, algebraic D-group
.................................. (0.8) D-actions, Dcocycles ........................................... (0.13)

logarithm ic connection
........................................... (0.16) adjoint
connection .............................................. (I.
1.10) automorphism functor
............................................ (1.1.23) field of
definition ............................................... (I.
1.27) locally finite, split
.............................................. (I.3)
representative Ideal
............................................. (I. 1.21)
weights .......................... : ....................
(I].3.1), (11.3.12) killing the weights
.............................................. (11.3.5) linear
part, abellan part ..........................................
(IV.i.i) logarithmic Gauss-Manln connection
................................. (111.1.2) KoalairaSpencer map ............................................
(111.1,4) total space, total connection

................................ (I11.1.6), (111.1.7)


relative projective hull
........................................... (III.2.1)
universal extension
.............................................. (III.2.3)
regular D-module
............................................... (111.3.6)
maximal semiabelian subfield
........................... : ........... (IV. 1.1) Amanifold ...................................................
(V.I.I 1) A -group, A o_group , dimension
..................................... (V. 1.11) split A
o.group ................................................. (V.
1.13) produced A -group
............................................... (V. I. 12)
embedding dimension
............................................ (V.3.16) moduli
sets ................................................... (V.O

NMS extensions
................................................ Appendix ^
Pa,nleve extensions
.............................................. Appendix ^.
145 K<X>
........................................................ (0.6)
XD()' XD(A)
.................................................. (0.7) KI<X>
........................................................ (0.]0)
PHSD(G/K)
.................................................... (0.13)
(Vfi[V) .......................................................
(0.16) L(G),P(G),P(G/K),DerkG,Lie G (I.
1.1 ) KIp]
......................................................... (1. i.i)
Xm(C)' Xa (c)
.................................................. (I. 1.3)
U(L), JG ......................................................

(I.].13) Au.._t G, Aut G, )'G


............................................... (I. 1.23) KG'
KL ......................................................
(1.1.27) V !, X
....................................................... (1.2.1)
P(G,fin), P(G/K,fin)
............................................. (1.3.9) ]I(A).
before (11.2.1) II(B A)
...................................................... (II.2.9)
W(G) ........................................................
(II..1) (T, ( (lU))
.................................................... (II.. 1)
(G)'Wo(G) ' ' (tl.3.]2) E a
................................................... (11.2.9),
(II.3.16) H l V H l 1 DR ( )a'
DR(V)m,HDR(V) t ............................. (11I.
l.l), (Iii. 1.2), (1II. 1.9) E(A)
........................................................ - S(C) K

.... (III.2.3) uerk


................................................. (Iii. 2.16) /,
<X> .......................................................
(V.I.11) V fi,G fi
...................................................... (V.2.2)
Xo' o ............................................ (V. i. 13)
X,G ........................................................
(V.l.12) r/r,
........................................................ (v.3. i)
[r,r] ........................................................ (v..)
z(r) .......................................................
(v..) R(r) ................... D o f]l
..................................... (V.3.7) H (
A/i,),E(A)a ............................ (V.3.17) A4 k
........................................................ .
(V..i8) , m(C) ....................................... (V.O
b r ab,{ .............. d,g' d,g,n
....................................... (V.t0 INDEX OF

NOTATIONS K[P],V D V P K D lp
................................................... (0.5) (0.6)

You might also like