Beruflich Dokumente
Kultur Dokumente
Managing Editor:
M.HAZEWINKEL
Centre for Mathematics and Computer Science, Amsterdam, The Netherlands
Volume 361
Nilpotent Lie Algebras
by
Michel Ooze
Department ofMathematics,
Universite de Haute Alsace,
Mulhouse-Colmar, France
and
Yusupdjan Khakimdjanov
lrutitute of Mathematics,
Uwekistan Academy of Sciences,
Tachkent, Uzbekistan
PREFACE xiii
§I Basic notions 77
1.1. 9-modules and representations 77
1.2. The space of cochains 78
1.3. The coboundary operator 79
104. The cohomology space 80
1.5. Exact sequence 81
§II Some interpretations of the space W(g, V) for i = 0,1,2,3 84
11.1. The spaces HOC9,V) 84
11.2. The spaces HIC9,V) 84
11.3. The spaces HZC9, V) 86
1104. The spaces H'C9,9) 89
§I11 Cohomology of filtered and graded Lie algebras 91
III. 1. Graded Ue algebras. Filtered Ue algebras 91
11.2. Graded and filtered 9-modules 92
111.3. Graduation and filtration of the cohomology spaces 93
§ IV Spectral sequences 95
IV.l. Definition 95
IV.2. Some properties 95
IV.3. Exact sequence associated to a filtered complex 97
§V The Hochschild-Se"e spectral sequence 99
§ VI. Cohomological calculus and computers 101
VI.1. The MATHEMATICA Program 102
V1.2. How to use this program 110
BIBLIOGRAPHY 329
INDEX 335
PREFACE
Nilpotent Ue algebras have played an Important role over the last ye!US : either
In the domain at Algebra when one considers Its role In the classlftcation problems of
Ue algebras, or In the domain of geometry since one knows the place of nilmanlfolds
In the Illustration, the description and representation of specific situations.
The first fondamental results In the study of nilpotent Ue algebras are obvlsouly,
due to Umlauf. In his thesis (leipZig, 1991), he presented the first non trlvlal
classifications. The systematic study of real and complex nilpotent Ue algebras was
Independently begun by D1xmler and Morozov. Complete classifications In dimension
less than or equal to six were given and the problems regarding superior dimensions
brought to light, such as problems related to the existence from seven up, of an
infinity of non Isomorphic complex nilpotent Ue algebras. One can also find these
losts (for complex and real algebras) In the books about differential geometry by
Vranceanu. A more formal approach within the frame of algebraiC geometry was
developed by Michele Vergne. The variety of Ue algebraiC laws Is an affine algebraic
variety and the nilpotent laws constitute a Zarlski's closed subset In this view the role
of Irreduclbl~ components appears naturally as well the determination or estimate of
their numbers.
Theoritical physiCiSts, Interested In the links between diverse mechanics have
developed the Idea of contractions of Ue algebras (Segal, Inonu, Wlgner). That Idea was
In fact very convenient In the determination of components. Others tried to combine
it with the formal notion about deformations which was Introduced by Gerstenhaber.
xiv
Truly the link between these two notions was only described a few years ago. One can
read it in this book. The linear approach to defonnations is situated in the frame of
spaces of cohomology: a deformation is a fonnal series which fonnally saUsfies
jacobi's identities. The linear part of it is thus acocycle for wlues in the adjoint module
(see Chevalley cohomology). Hence the problem of the existence in a given cocycle,
of a defonnation having that cocycle as its linear part. In this frame rigid algebras
appear also naturally. They are the ones which are invariant by defonnations. Nljenhuis
and Richardson have shown that the laws which have a trivial second group of
Chevalley's cohomology are rigid. However they are not all of the same kind :
NiJenhuis and Richardson give an example of a rigid law having a non-trivial
cohomology (in fact, as we wt1l see it here, the class of rigid laws with a non trivial
cohomology Is far from being small).
It Is Important to know about rigid laws : the orbit (t.e the set of all Isomorphic
laws) of a rigid is an open subset in the variety of Ue algebraiC laws and its closure is an
irreducible component of Zarlski. An estimate of the number of isomorphic classes
of rigid laws gives thus an estimate of the number of irreducible components. This
approach has been taken by Michele Vergne in the frame of nilpotent laws. She has
developed a cohomology adapted to nilpotent defonnations, described components
for large dimensions and introduced inCidentally the notion of flUfonn algebras.
Unfortunately we do not know yet of rigid nilpotent laws, and the problem of the
existence of rigid nilpotent Ue algebras In the subvariety of nilpotent laws is also being
considered.
The aim of this book is to put together all the results which were obtained
either In the cohomological study of nilpotent algebras or In the problems of
classification and the problems of deformations. The notion of derivations of
nilpotent Ue algebras is also eludied It elips naturally into the cohomological domain
(derivations are linted to the first group of cohomology). Their interest Is not only
algebraic: Riemanian spaces with a left invariant homogeneous metric often have an
isometric group the algebra of which can be identified with one of orthogonal
derivations. The study of infinitesimal frame : it has the advantage of simplyfylng the
approach and thus eliminate its formal aspect. It also allows to solve the geometrical
problems : what are the conditions for a cocycle (or for afonnal tangent vector) to be
Integrable, mean the first term of a deformation (or so that this formal vector be a
vector of the cone of tangents). One can In the infinitesimal scape describe the duality
contraction - deformation perfecdy. We focus also on two classes of nilpotent Ue
xv
algebras, which are more or less models of nilpotent Ue algebras: those are filiform
algebras (the least commutative ones) and the characteristically nilpotent Ue algebras
(all the derivations are nilpotent).
This book was made possible thanks for the collaboration of Professor Yu.
Khakimdjanov, who worked at the UHA as a visiting professor during the yeatS 1991-
94. We hope that this fruitful exchange will continue ....
Colmar - Mulhouse
23 october 1994
CHAPTER 1
The aim of this chapter is to recall some fundamental notions concerning finite-
dimensional Lie algebras. We also present the principal classes of these algebras and
mainly the more interesting classes of nilpotent Lie algebras. First we study Lie
algebras on an arbitrary field, although the larger part of the book is devoted to
complex nilpotent Lie algebras.
Remarks
1. The identity (2) is called the Jacobi identity.
2 Chapter 1
1.2. Examples
1; Every vector space a with the bracket [x,y] - 0, for all x and y in a, is a Ue algebra.,
called an Abe11an Ue algebra
2. Let V be the real vector space Ii'. The bracket defined by the cross-product of the
vectors of Ji3 defines on Ji3 a non-Abelian Ue algebra.
[A,B] - AB - BA
satisfies conditions 1 and 2. Then, nnCK) with this bracket is a Ue algebra, and.
denoted as glCn,K).
6. Let V be a C2k+D-dimensional vector space and Cel '''., e2k+I) a basis ofv.
The brackets defined by
Generalities 3
Examples
A-(~~) ,
where N and P are n-square symmetric matrices and the matrices M and Q satisfy
tM - - Q. Then, sp(2n, K) is a Ue subalgebra of glCn,K) called a symplectic algebra..
Remark. It is easy to enlarge these examples by considerin'g, not only the set of
matrices, but endomorphisms of agiven finite-dimensional vector space. Therefore,
we can define the following Ue algebras
(1) sl(Y) - {f E End V : tref) - O}, where trCf) is the trace of f, that is the sum of the
diagonal entries of a matrix of f (this is independent of the choice of basis of D.
1.5. Ideals
Examples
1. Let q be a Ue algebra. Then {OJ is an ideal.
Generalities 5
4. Let 9 - glen, CC) and h - slen, CC). Then fa. is an ideal of 9. In fact, every matrix A can
be written as A - al + A l' where A1 is in slen, CC) and I is the identity matrix. Let B
be in slen, CC). We have
[A, BJ - [aI + AI' B] - [AI' BJ E s1Cn, tC:) •
Remark. If 'L and 1 are ideals of aUe algebra 9, then 'L + 1 - {x + y, X E 'L and y E 1} ,
['L, 1J- {l: [xl> Y2J; xl E 'L, Y2 E 1} and 'L (] 1 are ideals of 9·
Let 9 be a Ue algebra and 'L an ideal of 9. The quotient vector space 9/'L is a Ue
algebra for the multiplication defined by [x + 'L , Y + 'L] - [x, yJ + 'L . This is
unambiguous since if x + 'L = x' + 'L, Y+ 'L = y' + 'L, then [x' + 'L, y' + 'LJ- [x, yJ + 'L.
1.7. Homomorphisms
is called a homomorphism if <p[x,yJ- [q(x), cpCy)] for all x and yin 9. We note that the
first bracket is in 9 and the second is in 91 .
Examples
1t
We note that, if 1.. - Ker cp, then 9/Ker cp is isomorphic to 1m cp, the Image of cpo The
Generalities 7
Let 91 and 92 be two tie algebras over K. We can build a new tie algebra by
9 - 91 + 92 - {(xl' X2): Xl E 91 ,X2 E 92}
with the multiplication
[(xl' X2) , (Yl> Y2)) - ([Xl' X2) ,[Y1, Y2 D .
We can remark that 91 and I:h are ideals of 9. We shall say that 1:1 is the direct sum of
91 and 92 and we shall denote 9 - 91 E9 92 .
Now suppose that 1. and J are ideals of a given tie algebra 9, such that 9 -'L+ J (direct
sum of vector subspaces). Then, 9 Is isomorphic to the direct sum 1. E9 J .
It is easy to see that the set Der 9 of all derivations of 9 Is a vector subspace of End 9.
Proof.
(f 0 g - g 0 f) [x, Y1 - fog [x, Y1- g 0 f [x,y1
- f[[gCx),yl + [x,g(y)]J - g[[fCx),yl + [x,fCy)]]
- [f ogCx),y] + fgCx),fCy)] + [fCx),gCy)]+ [x,f ogCy)]
- [go fCx),y] - [fcx),fCy)] - [gCx),fCy)]- [x,g 0 fCy)]
- [Cfog-gof)(x),y]+ [x,Cfog-gof) (y)]
Then, Der q is a subalgebra of gl(q).
Examples
Let q be a lie algebra over K and x E q. The endomorphism ofq defined by y ~ [x, yl
is denoted ad x. The Jacobi identity implies that ad x is a derivation of q for all x In q,
called ithe nner derivation. All others derivations of q are called outer.
An ideal" of a lie algebra 9 is called characteristic if fCL) is included in " for all
derivations f of 9. We note that every ideal is invariant for any inner derivation of 9.
But, generally, it is not invariant for any outer derivation. We can see easily that every
characteristic ideal J of a characteristic ideal" of 9 is also a characteristic ideal of 9.
likewise, if "1 and"2 are characteristic ideals of 9, ["1' 'L:z] is also a characteristic ideal
of9·
We wantto compute the derivations of a finite direct sum 9 - EEl 9t of lie algebras 9i
over K. We can consider every derivation fl of 91 as a derivation of 9 by :
flx) - flCxl' ... , xJ - fl:x~ for all x = CXI' ... , xJ in q.
Let fbe in DeqJ. Then f - ~f-l fl + ~f"l fij where fl EEnd 91 ,flJ EHom (91 , 9V are de-
fined by fl - 1t1 0 f 0 1t1 ' flj - 1tj 0 f 0 1t1 where 1t1 is the canonical projection of 9 on
91' These endomorphisms fb flJ are derivations of 9. We shall prove that flj E DC91,9)).
Let x - [y, zl in [91' 911 . Then
*
flj(x) - flj [y, z] - [flj (y), z] + [y, flj (z)] - 0 if i j and flj [91 ,91] - 0 .
Suppose now that flj(91) is not contained in 2(9j) . There is x '# 0 in 91 such that y-f lj (x)
yeo 2(9j) .WechooseZin9j Sadsfying[z,yJ '#O.Then
[flj' adz] (x) - [z, y] '# 0
and the nontrivial derivation [flj,ad z] is in D(91 , 9j). However, this derivation is inner
(Proposition 4). This Is Impossible If I '# j.
Let 91 and 92 be two Ue algebras over K. A IJe algebra 9 over K Is called a semidirect
sum of 91 by 92 If 9 - 91 + 92 (direct sum of vector spaces) where 91 is an ideal
and 92 a subalgebra of 9·
The direct sum corresponds to the particular case where 92 Is also an Ideal. We use
the same notation which we will summarize If it is necessary.
Generalities 11
V'k~o
C09 - 9 ,
{
Ck+19 - [Ck9 ,9] . }
Note that this sequence decreases more slowly that the derived sequence and the tie
algebra ct+19/ct+~ Is an ideal of ct9/C i+~ contained in its center.
Co(9) - 0,
Ck(9) - {X E 9 : [x, 9) c C k-l (9)} V k ~ 1.
We have
{OJ - Co(9) c C1(9) c ... c Ck (9) c ...
We note that C{9) is the center of 9 and Ck+1(9) is the reciprocal image for the
canonical embedding of 9 on 9ICk(9) of the center of 9ICk(~ .
The matrices of the elements of beD) relative to an adapted basis of the flag are
triangular. This algebra is similar to that described in example 3.
Proof.
(a) If L is a sub algebra of IJ, then nil is contained in n'IJ.
(b) Let 1t: Ij ~ Ij / 1 be the canonical homomorphism. As it exists, an integer k such
as1)K(IJ/t) -0, then 1tlDKIJ)-0. Thus DKlJct andas t issolvable, DKIj isalso
solvable and therefore IJ is solvable.
(c) We can deduce this result from the property (b) usingthe fact that the ideal
1+J / J is isomorphic to 1 / 1nJ .
Property (c) implies that there exists a unique maximal solvable ideal of IJ. This ideal is
called the radical of IJ.
Examples.
1. Every Abelian algebra is nilpotent With a nilindex equal to 1.
14 Chapter 1
Proof.
(a) This is the same as the proof in the previous section.
(b) Suppose that 9/2(9) is nilpotent. There is an integer k such that 0'<:9!Z(9)) - 0 so
cI<cJ is contained in Z(9) and dt+ 19 - {OJ.
(c) If g is nilpotent with a nilindex equal to k, then ck+19 '# {OJ and Z(9) => dt-~.
(d) Suppose that the nilindex of I is k and this one of J is 1. From the Jacobi identities,
CPCI+J) is included in C£P121+ 1Cj) where [p/2J denotes the integer part of p/2. As k-l is
less than dim 9 for p - 2dim 9, then CP+l(l+J) - {OJ.
I I
Infact, C 9 cD 9 .
Proposition 10. A lie algebra 9 Is nilpotent with a nillndex k if and only if for every
"0, Xl' ... , xk E 9, we have
[XO,[Xl,[X2, ... [Xk-l,Xk]]] ...... ]-0.
Proof of the theorem. The necessary condition follows from Pproposition 6. For
proving the sufficient condition, we use an induction. Suppose that the theorem is
true for the tie algebras of dimensions less than n. Let 9 be an n-dimensional tie
algebra such that ad x is nilpotent for all x. From Engel's lemma, it exists that y E 9
, y*"O such that [x,y] - 0 'itx E 9. This prove that the center Z(9) of 9 is nontrivial.
Let 91 the quotient algebra 9 1 Z(9) , by hypothesis 91 Is nilpotent and from
Proposition 7, 9 Is nilpotent.
Theorem 3 ( Lie's theorem). Let I} be a solvable lie algebra on K and <I> a linear
representation of I} In a vector space V. Then there exists a flag D - (V - V0::> VI
::> ... ::> V n - {OJ ) of V such that b(D) ::><l>Cg) where beD) is the lie algebra of
endomorphlsms of V related with the flag D.
This algebra. &CD) is described in example 4 (1.2). We can also enunciate the theorem
as :
Theorem 4. With the hypothesis of lie's theorem, if V =I- {O}, there is a vector v =I- 0
in V such that
«XXv) - 'i.I.x)v , 'v'XEI}.
The eqUivalence between these theorems can be proved using an induction on the
dimension of V.
18 Chapter 1
Proof of Theorem 3.
First, we prove the following lemma:
because 1. => n9 => [t , 9] . From the hypothesis, there is a nontrivial vector v in V and
a linear fonn A. on 1. such that
fJ)(a)(v) - A.(a)v , V' ae 1..
We have W - { we V such that fl»(a) (v) - A.(a) (av) V' a e 1. }. It is a nontrivial
subspace of V. As before in the previous lemma, we have
A{x,a] - 0 V'xe 9 and ae 1. .
But
fJ)(a) 0 fJ)(x) (w) - tll(a,x] (w) + q,(x) 0 <I(a) (w)
then fl»(xXw) e W and W is invarlantin respect to the action of 9. Let x e 9, x Ii!: 1.. As
fJ)(x)(W) is contained in W, there is an eigenvector vo ;/: 0, Vo Ii!: W. This vector is also
an eigenvector for every linear operator fl»(y), y e Kx + I - 9. This proves the lemma.
Consequences
1. We can deduce from Ue's theorem the existence of a basis such that the matrices of
the endomorphisms with respect to this basis are triangular.
2. We consider the adjoint representation of 9. Then as before in the previous case,
there is a basis of 9 such that every adjoint operator ad x admits a matricial
triangular representation with respect to this basis.
Corollary 3. The Ue algebra 9 is solvable if and only if the derived algebra n(9) is
nilpotent
Proof. We suppose that 9 is solvable. From Ue's theorem applied to the adjoint
representation, we deduce that there exists a sequence
9=>1.1=>~=>"'=>'Ln-(0} .
where 1., are ideals of 9. If x e n~, then ad xC1.,) is contained in 1.'+1' The endo-
morphism ad x is nilpotent. Its restriction to n 19 is also nilpotent. From Engel's
theorem, :1)19 is nilpotent
20 Chapter 1
Proof. Suppose that TrCxy) - 0 for all y In 9 and x In D9. For proving the solvability of
9. It is sufficient to prove that D9 is nilpotent. From Engel's lemma. we can prove that
x Is nilpotent In D9. This Is a consequence of the following lemma:
tent. But, Tr([u, v] z) - Tr(u[v, z]) - Tr([v, z]u) for every u, v, z in gl(V) . In particular,
if z eM, u, v e q, we have [v, z] e [q,q] and Tr([v, z] u) - o. This gives Tr([u, v] z)
- 0 and Tr(x y) - 0 'r:j y eM. From the lemma, x is nilpotent this ends the proof
after we have seen that the converse is a direct consequence of Ue's theorem.
The principal topics of this book concern the study of nilpotent tie algebras. So we
shall give only a short survey of the theory of semisimple tie algebras Cthe reader
interested in this subject can read the classical books, for example [SelJ). The aim of
this survey is to present the necessary tools in order to study some important classes
of nilpotent subalgebras of semisimple complex Ue algebras.
Definition 7. A non-Abelian Lie algebra q is called simple if its only ideals are q
and {a}.
Examples
1. ste2, tC) is simple.
2. We will see later that every classical tie algebraslCn, tC), soCn, «:;), spC2n, «:;) Csee 104)
are simple.
22 Chapter 1
Recall that the radical of a Ue algebra9 is the unique maximal solvable ideal of 9. It Is
denoted by rad 9.
For example, the bilinear form BCx,y) - TrCad x 0 ad y) is invariant and symmetric.
This form is called the KIlllng form.
Proposition 12. Let 'L be an Ideal of the Ue algebra 9. Then the orthogonal space 'L.L
of 'L with respect to he KIllIng form B Is an Ideal of 9.
consider f - ad x 0 ad y for x E a. and y E 9. Then a.::::> f(9) and f(a.) - {O}, this gives
f2 - 0 and TrCD - o. This proves that a. c 'L = Ker B = {O}.
Corollary 1. Let 9 be a semisimple Ue algebra and 'L an ideal of 9. Then 9 - 'L $ 'Ll.
(direct sum of Ue algebras) where V is the orthogonal of 'L with respect to the
Killing form of 9.
First, we recall the notions of representations of tie algebra 9 and the corresponding
24 Chapter 1
notion of 9 -modules.
Let 9 be a Ue algebra. A representation of 9 In a complex vector space V Is a
homomorphism of Ue algebras cI> : 9 ~ gICV). A vector space V endowed with an
operation 9 x V ~ V (denoted x . v) is called a 9-module if we have
Q) (ax + by)v - a(xv)+ b(yv) , a, be <C, x, y E 9, v E V ,
Cit) x(av + bw) - a(xv) + bCxw) , a, b E <C, x, Y E 9, v E V ,
aID [x, yJv - xyv - yxv, x, Y E 9, v E V .
Remark. 9 may be considered as a9-module for the operation xy - [x, yl. Note that 9
may be asemlslmple 9-module without being a semisimple Ue algebra; for example
in the case 9 - <C.
We can define the nllradicalof the Ue algebra 9 as the largest nilpotent Ideal of 9. We
Generalities 25
can also define the notion of the nilpotent radical of 9 as the intersection of the
kernels of the finite dimension irreducible representations of 9. This ideal is
contained in the nilradical and we can prove lhat it coincides with:09 n rad 9.
Theorem 8 (Levi). Every Ue algebra 9 is the semi-direct sum ofits radical rad 9 and a
semisimple subalgebra s :
9 - rad 9 E9 s.
Proof.
1st case. : [9, rad 9]- {OJ • Then rad 9 is the center of 9 and the adjoint representation
of 9 is semlsimple (we can Identify this representation with the representation of
91Z(9) = 9/rad 9 which Is semisimple). Then:09 is also semisimple and 9 - rad 9 E9
:09·
2fJd case: [9, rad 9] '* (OJ and the only ideals of 9 contained In rad 9 are {OJ and rad 9.
Then [9, rad 9] - rad 9 • [rad 9. rad 9] - (OJ and the center of 9 is (OJ.
'*
Let M be defined by M = (f E End 9: f(9) c rad 9; f(x) - A.(f)x A.(f) 0 V X E rad 9}
26 Chapter 1
o(x) fo E P for all x in 9. The mapping x ~ "(Pl(O(X) fo) is a linear mapping of 9 into
mapping of 9 into rad 9 . Then its restriction on rad 9 is the identity mapping.
Hence the Kernel 5 of this mapping satisfies 9 - rad 9 + 5 Cvectorial direct sum). As 5
- {x E 9: oCx) fo - O}, 5 is a suba1gebra of 9.
Levi decomposiUon.
then there exists an automorphism (J of 9 such that 0(5 1) - 52' Moreover, we can
choose (J of the form (J - expCad x), where x in the nilradical of 9. This result is due to
Malcev.
2. From Levi's theorem, the general description of Ue algebras is deduced. from the
description of semisimple Ue algebras, solvable Ue algebras, and the representation
of semisimple Ue algebras as derivations of the radical. The class of semisimple Ue
algebras is well known (see later). But the class of solvable and nilpotent Ue algebras is
very difficult to describe. Moreover, solvable Ue algebras exist that are never radical
of an unsolvable Ue algebra wuh a nontrivial representation of the Levi subalgebra.
We shall see that every derivation of 9 is nilpotent Then Der 9 does not contain a
semisimple subalgebra.
28 Chapter 1
Definition 11. The element x E 9 Is ca1led regular If the dimension of the space 9 0 x
18mlnlmal:
dim 90x - min (dim 90y).
ye ':J
Recall that the normalizer of 9° x in 9 is the set of the elements y of 9 such that
[y , cfx] c cfx .
We can note that there always exists a nontrivial Carlan subalgebras for every
semisimple Ue algebra. Moreover, all the Carlan subalgebras of a semisimple Ue
Generalities 29
algebra 9 are isomorphic. They are also conjugated in the following sense: let hI and
h2 be two Cartan subalgebras ; there is an automorphism cr of 9 of the form exp(ad y)
with Y E 9 such that ~ - o(h l )·
Consequences
1. All the Cartan subalgebras of 9 have the same dimension r. This dimension is calied
thenmkof 9.
Let r be the rank of the semlslmple Ue algebra I}. It Is possible to find a basis
(ai' ... , a) of fl.- such that
1. <Xt e A , I - 1 ,..., r ,
2. Every root a can be decomposed as a sum
a - kl <Xt + k2 <Xz + ... + ~ afl
and the coefficients kt are all positive Integers or all negative Integers.
A root system S - (ai' ... , a) satisfying the previous conditions Is called a system of
simple roots.
Now, we fix a system of simple roots. The elements of this system are called simple
roots. A root a Is called positive Crespo negative) If the coefficients of a In
decomposition In the simple roots are positive Crespo negative).
Generalities 31
We denote A+ the set of positive roots, and A_ the set of negative roots. Then:
A - A+ u A_, A_ - - A+ .
From a fixed simple roots system S - Cex l , ... , ex), we can provide fl.- with a partial
relation by stating that y > v <=> y - v has a decomposition as a sum of simple
roots exl with positive coefficients. of course, A Inherits this partial order relation.
For each I, we put HI - H IX! and we choose elements XI E 91X!, YI E 9- IX! such thaI
[xl' YI] - HI . Finally, we put nCi, j) - CXtCH~ . The matrix formed by the numbers nCi,p Is
called the Canan matrix of the given system s. We note that nCi,p Is an integer ~ 0 If
*
I j and n(l, I) - 2.
Theorem 9. The vectors CHI' ~, YI) I - 1 ,... , r satisfy the fol1owing conditions
9 is generated by (X h Y h HI) ,
The Weyt basis is a basis of 9 defined by the vectors (XI' YI' H~ and the above
32 Chapter 1
Let S - Cal' ... , a) be a simple roots system. These roots will be the vertices of a
graph. Two vertices will be formed by nCi, j) . n(j, 1) edges. These integer numbers are
equal to 0, 1, 2 or 3. We put the inequality sign> on the multiple edge to indicate
which of the two adjacent roots is the longer; the absence of an inequality sign means
that two adjacent roots have the same length. The length of a root a l is given by
< ai' a l > where < , > is the restricted Killingform to fl.., Thus, the graph labelled is
called the Dynkin diagram.
Theorem 10. All the Dynkin diagrams of the simple complex LIe algebras are the
foHowing:
al a2 a3 ~ ar_1
Dn (n 2: 4) 0----0--0- - --
a r
al a3 a4 as 0<6
~
£6
~
al 0<3 0<4 as 016 0<7 0<8
£8
I
~
0<2
a1 a2
G2 <::l=$O
Remark. Indeed the Dynkin diagrams An "'" Ea correspond to roots systems and
therefore to simple complex tie algebras, We note that two isomorphic simple tie
algebras admit the same diagram and, conversely, two simple tie algebras having the
same Dynkin diagram are isomorphic, Hence, the classification of complex simple tie
algebras is deduced from theorem 9, It is sufficient to give the corresponding tie
algebra for each diagram :
These algebras are described in Section 104. The diagrams E6, E7, E8, F4 and G2
correspond to the exceptional simple Lie algebras. These algebra admit also matrlcial
representations. For example, the simple tie algebra G2 may be presented as a
derivation algebra of a Cayley algebra.
Vl.l. Definition
Theorem 11. Let q be a Lie algebra and r - rad q its radical. The following four sets
are identical :
(a) the largest nilpotent ideal ofq,
(b) the largest nilpotent ideal of rad q,
(c) the set of x E rad q such that ad qX is nilpotent,
(d) the set of x E rad q such that <dr x is nilpotent
The ideal defined in this theorem is called the nilradlcal of q and denoted nil q.
Corollary
1. The nilpotent radical Dq n rad q of q coincides with [rad q, qJ, hence it is
included in nil q .
2. For every f of Der q, f(rad g) c nil q .
Generalities 35
Algebraic Lie algebras are the Lie algebras of algebraic Lie groups (see [Bo]). They
constitute an interesting class of Lie algebras containing, in particular, semlsimple Lie
algebras and nilpotent Lie algebras. The algebra of all derivations of a Lie algebra Is also
algebraiC. For these Lie algebras, we have a similar decomposition to levi's
decomposition.
Theorem 12. Let 9 be an algebraiC Ue algebra.. Then there is a reductive subalgebra. c;a.
co The nilindex of 9
Examples
3. If s - dim 9 - 1, we say that 9 is filiform. The study of these algebras can also be
found In the Chapter 2.
o 1 s
9 - C 9::::> C 9::::> ... ::::> C 9 - {O},
(Cl' C;2, ••. , c s ) ~ (db ... , dp) <=> 3 i such that cJ - ~ for J < i and Ci > d; .
Examples
1. c(9) = (n-l, 1)
There is X in 9 such that (ad X)n-2 "* 0, and the Jordan form of the matrix of ad X is
o 1 0 ... 0
o 0 1 ... 0
o 0 0 ... 1
o 0 0 ... 0
38 Chapter 1
2. c(9) - 0, 1, ... , 1)
For every X in 9 - 1)9, the Jordan matrix of ad X is zero. Then 9 is Abelian.
CD the dimensions of cohomologtcal spaces (for the trivial values in a cohomology, the
cohomology with coefficients 9-module, in the adjoint module, etc.),
Generalities 39
(Iv) Dixmier's invariant. It is the dimension of a maximal orbit for the coadjoint
representation (see Chapter 7).
CHAPTER 2
Proof. Recall that cOO is the ordered sequence of the dimension of Jordan blocks of
ad X. If c(X) - (n - 1 , D, there is a Jordan basis (X - Xl' ... , Xn) of ad X satisfying
[Xl , ~] - ~-l for i - 3, ... , n
and C 19 is generated by {X 2 , X3 , .•• , Xn_l} for 2$; i. Then the tie algebra 9 is filifonn.
Conversely, we choose a basis {X o' Xl' ... , ~) corresponding to the filtration
9 - CO(9) :::> Cr9) :::> Cl:9) :::> ••• :::> Cn -\9) .
Remark. The filifonn tie algebra have a maximal nilindex. These algebras are the "less"
nilpotent.
Examples
1. Let (,n be the (n + D-dimensional tie algebra defined by [XO' XI] - XI+I for i - 1,
... , n - 1 where (X o' ... , Xn) is a basis of (,n' the undefined brackets being zero. This
algebra is filifonn. This is, in a certain manner, the simplest filifonn Ue algebra.
2. Let Qn be the (n + D-dimensional nilpotent Ue algebra defined in the basis (X o' ... ,
Xn) by
[Xo ,XI] = XI+I , i - 1, ... , n - 1 ; [XI, Xn-t] = (- 1)IXn , i = 1, ..., n - 1 .
This algebra is also filifonn.
4. Let Wn be the tie algebra whose brackets, in the basis (X o, ... , Xn), are:
°,
[X XI] - XI+l , i - 1, ..., n - 1 ,
6(i-l)!(j-l) <.. <
[XI,XJ]= (" ')' XI+j+I, 1_I<J_n-1 i+j+1$;n.
1+J •
We shall see later that there are only two, up to isomorphism, (n+ I)-dimensional Ue
algebras, n ~ 11, whose brackets satisfy the relations:
[X 0 , XI] - XI+ 1, i-I, ... , n - 1
[XI,X~-aljXI+j+1 with l~i<j~n-l and i+j+l~n.
the linear map fl : 9t -+ ~ being not identically null. There is Xo E 91 such that
fl(X o) * 0, 'if i, 1 ~ i ~ n-l . One can choose the vectors XI E 91 ' with XI - AI YI'
AI * 0 ,so that they satisfy the conditions
Nilpotent Lie Algebras 43
Corollary. If n is odd, then there are only, up to isomorphism, two (n+ 1)-
dimensional graded filiform Lie algebras. These algebras are Ln and Qn _
Examples
This Ue algebra is two-step nilpotent. It is, in a way, a model within these two-step
nilpotent Ue algebras. It is characterized by the following property :
Obviously, we have [xl' XJ] - [XI' Yk] - 0 . Then the Ue algebra 9 is entirely defined
by the only brackets : p
[Yi> YJJ = L a~J Xk 1 ~i<j ~n-p .
k-I
jacobi's identities are all satisfied and the constant aklJ are free. Intrinsically this means
that the structure of 9 is defined by a surjective linear map
cp: NV ~ Z(9) .
The dual point of view permits a concrete analysiS. The dual map
cp. : Z(9)· ~ A2CU)·
Nilpotent Lie Algebras 45
Is Injective. The data of this mapping Cand then the data of the bracket of 9) Is the same
as the data of a subspace of dimension p (p - dim ZCrp) in A2y . If Cal' ... , ap, ~1'
... , ~n-p) is the dual basis of a given basis (Xl' ... , Xp' Y 1, •.• , Yn-p), then the structural
equations of 9 Cthe dual Jacobi conditions) are :
dUk - L
J.g<j~-p
~j 13t,,~ .
The bilinear forms C9 1 - da 1 , ••• , 9k - dak) are two-forms on Y . They generate the
subspace cp"C ZCrp") . It appears clearly, in this writing, that the data of the two-forms
91 is eqUivalent to the data of the constants Cakij).
At first Sight, the problem of classifying two-step nilpotent tie algebras appears to be
easy to solve. Recall that this problem was first approached by Umlauf In 1891 In his
thesis. Today, it is always an open problem. At the end of this chapter, we will give the
rare results concerning this problem. If we introduce a new and strong hypothesis as
the two-step nilpotency, we can hope for a simplification of this problem. Indeed,
this is not the case. From the previous results, we can affirm :
The first example of a nilpotent algebra whose derivations are degenerated was given
by Dixmier and tister in 1957 [D.L.]. It is the follOwing 8-dimensional tie algebra whose
Nilpotent Lie Algebras 47
brackets are
[Xl> X2l - Xs ' [~, X4l-~,
[Xl' X3l - X() , [X2, X()l - -X7 ,
[Xl' X4l - X7 ' [~, X4l - -XS '
[Xl' Xsl - -Xs , [X3, XSl - -X7 ,
[~, X 3l- X 8 , [X4,~l--X8'
This example was the starting point of a study of a new class of tie algebras called
characteristically nilpotent tie algebras.
Definition 3. Let 9 be a nilpotent Lie algebra and Der 9 its derivations algebra. The
Lie algebra 9 is called characteristically nilpotent if every derivation f in Der 9 is a
nilpotent endomorphism.
Let 9[1) = Der if..9) - { Y E 9 such that Y - f(X), f E Der 9 ,X E 9 } and, more gener-
ally,
tit] = Der 9 (tit-I]), i > 1 .
The Ue algebra 9 is characteristically nilpotent If and only if there is k E JRII such that
9 1kl - {OJ.
This sequence 9[1) generalizes the central descending sequence; here we use the set of
all derivations instead of the set of inner derivations.
Theorem 2 [LTl. Let 9 be a nilpotent Lie algebra of a dimension more than 2. Then 9
is characteristically nilpotent if and only if the Lie algebra Der 9 is nilpotent
Note that the nilpotence of Der 9 doesn't directly imply the nilpotence of its
elements, but only the nilpotence of the adjoint operators. The proof of the theorem
is based on the follOwing lemma.
Lemma. Let L be a nilpotent Lie algebra such that L - 'Ll €a 'Lz where 'Ll and 'L2
48 Chapter 2
and 12 - Vo .
Then 9 is a direct vectorial sum of'Ll and'L2, which are ideals of 9, 'Ll being centtal.
The previous lemma implies 'L t - {OJ or'L2 - {OJ.
if'Ll - {OJ, then 9 - 'L2 - Vo and every derivation of 9 is nilpotent.
if'L2 - (OJ, then 9 - 'L l . This implies that 9 Is Abelian and Der 9 - glCn,rc:).
As Der 9 Is nilpotent, we deduce that n - 1, and we find a contradiction.
Nilpotent Lie Algebras 49
111.3. Examples
Theorem 3. Let 9 - E9Pi~l 9i be a Lie algebra where the 9i are ideals of 9· Then 9
is characteristically nilpotent if and only if each ideal 91 is characteristically nilpotent.
50 Chapter 2
where
As we have also fCZC~) c Z(9) for all f E Der 9, we deduce that every derivation is
nilpotent.
p
Conversely, we suppose that 9 - $ 9i is characteristically nilpotent.
i=1
Each derivation of 91 can be extended naturally to a derivation of 9. Then this
derivation Is nilpotent.
Recall what is a Borel subalgebra. First we fix the notation : fI. is a Cartan subalgebra of
9, A the set of roots corresponding to fl., S a basis of A Cthe simple roots), A+
Crespo A_) the set of positive Crespo negative) roots Crecall that A - A+ u A_ ) and
9 a - {X E 9 / [X, Hl - cxCH)X, 'v' HE fl.} .
The space 9 a has a dimension one. We choose a nonnull vector Xa in 9 a . A Borel
subalgebra & of 9 is a solvable maximal subalgebra of 9. It is conjugate, up to an
inner automorphism, with a subalgebra of the follOwing type :
&' - fI. E9 L
9a·
aeL\..
These algebras were first studied by G.B. Gurevich [GUll but in the case where the
simple algebra 9 is of type Ar • The motivation of this study is founded on the theory
of complex homogeneous spaces: let M be a compact homogeneous space M - G/H
G being a complex Ue group and M a closed subgroup. If 9 and fI. are the lie
algebras corresponding to G and H, the normalizer of fI. in 9 is parabolic. This
result has been established by Tits. It permits us to translate the study of
52 Chapter 2
We note that, for every standard algebra t (not necessarly nilpotent) whose nonnalizer
has the follOwing fonn
In the preceding chapter, we introduced that the partial order relation on the dual 11. *
of the Cartan subalgebra 11. : COl ~ IDz ¢::> 0)1 - IDz is a linear combination of simple roots
with nonnegative coefficients.
Proposition 4. Let t be a standard aJsebra such that Its nonnalizer can be expressed
as 11. e 19a. Suppose that a and ~ are posItive roots with a S ~ .
aeD.lu~+
Proof. We consider in each subspace 9 a a nonzero vector Xa. As 'Y ~ a , one can write
'Y - a + 'Yl + ... + "I'k where the 'Yi are positive roots. One can order again the terms of
this sum 'Y - 'YiJ + ... + 'Ylo + a + 'Yio+l + ... + r.. '
such that every partially connected sum
beginning with 'Yt, are positive roolS. Each vector Xli is in the nonnaliz.er of t. Then the
image of the vector X a by ad X n. 0 ad X"A._1 0 ••• 0 ad X t".1 0 ad X "fIl + ... + 'YI,. is in the
A description of standard algebras is presented in the general case in urn 1]. Here we
study only the nilpotent standard algebras.
Nilpotent Ue Algebras 53
Let R be a subset of 8+ that consists of pairwise incomparable roots (for the partial
order in h.' given in the preceding section). One puts :
Rl- {ae 8+/a;;::~ for ~e R}.
The subspace n - L 9a is a nilpotent subalgebra of 9.
aER,
8' C 8 2 +C 8+. Let R be the subset of 8' constituted of all minimal elements, always for
the partial order law. These elements are pairwise incomparable. Then R defines n.
Lemma 1. Suppose that ai' ~, a" a 1 + ~, a 1 + ~ +a, are roots such that a 1 + a,
and ~+ a, are nonzero. Then one of these comblnal1ons a 1 +a, or ~ + a, is a
root.
The proof Is a direct consequence of the JacoblldenUty concerning (Xal,Xa2,Xa,).
Lemma 3. Let a and ~ be two roots such that the decomposIl1on of a as the sum of
the basIc roots contains all the roots of sli Cthe set of extremal roots of ~) with a
mull1plicity greater than that of ~. Then a ~ ~ .
Proof. We put a - ~ + 11 + ... + 'Yk- ~c ...- ~ with 'YI and ~ In S and 'Y1 ~ ~ for
every I and j. By hypothesis, the roots ~1' ... , ~ are not In sli. Suppose now that
a is not greater Cor equal to) than ~ Ci.e. m ~ 1). We are going to show that v - ~ + 'Y1
+ ... + 'Yk Is a root of A and that ~1' .•. , l;.n are not in Sv. For k - O. the result is
obvious. We suppose k ~ 1 and we make an inducUon on k. One has :
k m
C-) (~.~) = (~,a) + ~ (~.-'Yi) + ~ (~.~J > 0,
1-1 j-1
where C ,) is the usual product on the set of the roots associated with the Killing-
Cartan form.
m
As (p, (X) ~ 0 and ~ (~.~j) ~ 0 Cif not ~ - (X and ~ -1: ~j are roots, this Is opposite to
j-1
m ~ 1 or to ~ 1 ••••• ~ m ~ S~. from C-) there Is an integer r. 1 ~ r ~ k such that one has
(~ , - 'Yr) > 0 . So ~ + 'Yr E A. From Lemma 2. ~ 1 ,...• ~m ~ sP + Yr. We apply the
induction hypothesIs for proving the previous assertion.
m
So. one has (V,v) =(v,a) + ~ (V'~i) > 0 and ~lo ••• ,~m ~ sv. One deduce (v,a) > 0
1-1
Nilpotent Lie Algebras 55
and ~ - V- a - ~ 1 + ... + ~m is In the set /l. We can suppose that each paxUal sum
~1 + ... + ~ of~ is In A. Let" - ~1 + ... + ~-l" Then a - -" -~ + v. From Lemma 2,
one has V - " - V - ~1 - '" - ~_I Is aroot. By a simple induction. we canafflnn that
V - ~I E A. As ~ It!: sv, this Is Impossible. We have proved the lemma.
This shows that the decomposition of a in simple roots contains one element V of
SI (by definition. v Is an extremal element for ~ E R). But N(n) Is a parabolic
subalgebra. Then X-v E N(n). More so we have ~ - v E A or ~= v. If ~ - V E A,
then
[Xj30 X-v] - N 'X~-v E n With N ~ o. If ~ =v, then [Xj30 X-v] - [Xv. X-v] E n, but this
vector is also in the Cartan subalgebra. These two cases lead to a contradiction.
SI - U Sp.
PeR
Conversely, every nilpotent standard subalgebra Is conjugated to a subalgebra n
a standard nilpotent lle algebra, then the quotient 9+/11. Is also nilpotent
This process permits us to construct a class of nilpotent Ue algebras. Every Ue algebra
of this class, and every standard nilpotent tie algebra Is given by a subset R c 11+ that
consists of pairwise incomparable roots. If we consider the subset R c 11+ whose
elements are pairwise incomparable and containing an the decomposition In simple
roots) at least three distinct simple roots, then we obtain a subclass of nilpotent lle
algebras. These algebras are nothing but the nilpotent lle algebras with maximal rank
studied by G. Favre and L Santharoubane [FS.].
subalgebrans' We easdyshow that n ens' for all S· E cIIR Then n 1- nns, for an
0 0
(ii) ~ (i).
Let be n - ()-l,nnl where I1t is a complete nilpotent stanadard subalgebra associated.
to the system SI of simple roots. Consider a root ~E R. Suppose that Its
decomposition ~ - klal + ... + kmam contains acoefficient kl strlcdy greater than 1.
As the vector X~ is in n - ()-l,nnl , there is a root 11 in SI such that ~ ~ 11 for alii, 1 ~ i
~ n. Then for the root ~ - al + ... + <Xu! E R+, we shall also have ~. ~ 110 Thus, the
vector X~ Is In each nl for alii and, 1hus,1s In n. This Is Impossible because po < ~o
Note. If the semislmple algebra 9 Is of type An' condition en) is necessarily satisfied
The decomposition of n as an Intersection of the complete subalgebra Is always true.
This result was proved by Gurevich [GU].
The classification given In this section (and in the follOwing sections) concerns only
complex lJe algebras. By convention, the undefined brackets are null, except those
which emanate from the anticommutativlty.
58 Chapter 2
Dimension 1
fl, 11 - A.t - the Abeban algebra of dimension 1.
Dimension 2
Dimension 3
From dJmension 4, we don't write the decomposable algebras, i.e. the tie algebras
which are dJrect sums of proper ideals.
Dimension 4
Dimension 5
Dimension 6
The previous tables show that, for dimensions less than 6, there is a finite number of
Isomorphism classes of complex nilpotent tie algebras. But, for dimension 7 and
more, we are going to find families with one or more parameters of nonisomorphic
tie algebras. This Is the same as saying that the variety of Isomorphism classes has
nontrivial dimension. Finding one-parameter families of Isomorphic classes of 7
dimensional nilpotent algebra is not the only difficulty. There Is, in a variety of
isomorphic classes, some isolated points. So every table of dimension 7 is very
difficult to establsh. Now, we have 3 or 4 tables given by various authors.
Unfortunately, It is not easy to compare these lists because of the techniques used
and the Invariants are very different Ccharacteristic sequence, rank, extension). The
links between the invariants used are not evident. We begin by giving all the one-
parameter families.
Theorem 7. There are six families with one parameter nilpotent Ue algebras of
dJmensJon 7, palrwlse not isomorphic:
1-4,6,7
[X 4 ,X7 ] - X2 [X2'~] - cxXs
[X 2X 7] - [X,X 4 ]:;:: Xs
Remark: Each Ue algebra of the family ~l,a Is filiform, and the Ue algebras of the
family n 72 ,a are characterlsdca1ly nilpotent
Lie algebras whose characteristic sequence is c(9) - (6,1). Filiform Lie algebras.
n'
7 .. [Xl,Xt] - Xt-l , 3SiS7,
[X 4,X7] - X2 ;
[X6,X7] - X2 + ~ ,
62 Chapter 2
,
fI, 6 .. [X1,Xt] - Xt-l ' 3SiS7,
[Xs,X,] - X2 , [X6'X,] - - X3 ,
fI, , .
7 • [X1,Xt] - xt-l ' 3SiS7,
[X6'X,] - X2 ,
,
fI, 9 .• [X1,Xt] - Xt-l ' 4 SIS 7 ,
[X6,X7] - X3
[X5,X6] - -t X3
[X2,X6] - t X3
[X5,X6] - - t X3
[X5,X7] - ~ X3+ t ~
[X5'~] - X3 ; [X6,X7 ] - X2 + X5 ;
[X 5X,] - X4 ,
[X s , X 71 - ~; [X6,X7 1 -XS'
,
n ,8·. [X1,Xt] - Xt-l , 4SiS7,
[X 2,Xti]- X, ; [X 2'X7]- X,+Xt •
n42 .,.44
7 .. - .,.'6 e a 1
7 .. - n16 ea 1 n 743 .. - .,.26 ea1
, .
.,.4S. - .,.46 ea1 , .
.,.46 • - ",S6 ea1
n 753 •• 6
-n6 ffia l
n 756 .• -nl~ffial
[X 3,X7] - X5 , [X",x7] - ~;
[X3'Xt] - Xs '
nf51 - n? $a2
n153 n154
7 7
In the next chapter, we shall present other classifications. In particular we shall de-
scribe all the filiform Lie algebras of rank nonzero, i.e. the uncharacteristically
nilpotent tie algebras (see Chapter 7).
CHAPTER 3
I. BASIC NOTIONS
In the first chapter, we glanced at the notions of representations of Lie algebras and
the corresponding notions of 9-modules. Here we will study all those notions which
can be utilized for the theories of cohomology. First, we recall all the definitions.
Let 9 be a Lie algebra on a field K. An n-dimensional vector space V over the same
field K is called a 9-module if a bilinear map <p : 9 x V ~ V satisfying the condition
q(lx, yl, v) - q(x,q(y, v)) - q(y, q(x,v)), 'V x, y E 9 and v E V,
is given. We write gv instead of q(g,v), if no disarray is possible. In other words, a
9-module V corresponds to a representation of 9 on V, i.e. a homomorphism of Lie
algebras ell : 9 ~ gl (V) defined by
cIICg) (v) - q(g, v) - gv .
For any IJ-modules V and W, the vector spaces V $ W, V ® W and Hom(V, W) can
be considered as 9-modules by putting:
g(v+ w) - gv+ gw,
78 ChapterS
g Cv ® w) - gv ® w + v ® gw ,
Cgf) Cv) - gCf(v)) - fCgv),
where ve V,weW,ge I}andfe HomCV, W).
The structure of al}-module on a tensor product Induces one on the exterior products
AJlV (the quotient of the tensor product ®PV - V ® ... ® V by the subspace generated
by all the v ® v). Note that AV - I A'V is a fini~mensional (because V has a finite
dimension) graded algebra over K with AOV - K, AIV - V. It is clear that Anv m
one-dimensional(n - dim V) andAkV - {OJ for k > n. If W is a subspace of V, then
APW can be identified canonically with a subspace of APV.
Examples
(t) Let V be a vector space over K. It can be considered as a I}-module for all Ue
algebras 9 by putting gv - 0 for all g In I} and v In V. This 9-module Is called trivial.
(ll) Let V be a IJ-module. An element v In V Is called I}-inw.dantlf gv - 0 for all gin I}.
Let be V9 the set of all9-lnvariant elements. Then V9 is al}-submodule of V, which m
trivial. This submodule Is maximal within the trivial 9-submodules of V.
OIl) Every Ue algebra I} can be considered as a I}-module by putting gl82 - [g1' 82] for
all gl and 82 In I}. The corresponding representation of I} on I} Is nothing but the
adjoint representation 4l(g1) (g2) - ad gl (g2). This example plays an Important role.
as the space of the p-cochain of I}. Then, CPCI}, V) - Hom CAPIJ, V) for p ~ 1,
COCI}, V) - V, and forp < 0, we putCPCI}, V) - {OJ.
Let C·CI}, V) be the direct sum of CP(I}, V). The elements of C"CI}, V) are called cochains
of I} with values in V ; those of CPCI}, V) are said to be of degree p.
We provide the space CPCI}, V) of a I}-module structure :
(x4») (Xl, ... , Xp) - x4» (Xl, ... , Xp) - 1: 4»(Xl ,... , XI-l, [X, XI] , ... , Xp)
lSiSp
for aU X, Xl ,... , Xp in I} and 4» in CPCI}, V). This structure of al}-module can be expanded
to the space C"CI}, V). The corresponding representation wiD be denoted by S.
For all y in ':J, we denote iCy) as the endomorphism of C·CI}, V) which maps each
subspace CPCI}, V) into cP-l(I}, V) and given by
(iCy) 4») (Xl'''.' Xp-l) - «y ,Xl'···' Xp-l) .
for 4» in CP(I}, V), P ~ 1, and Xl'.'" "P+l in I}, where the symbol A over a letter means
that it is omitted. This endomorphism d maps CPCI}, V) into CP+l(I}, V). Its restriction
to CP(I}, V) wiD be denoted d p.
80 Chapter 3
Proof. The first Identity is nothing but the deflniUons of the endomorphisms d and i.
We prove the last by an induction on p. Let «I> be a O-cochain. We have
(9Cx) d«l» (y) - x(d«l» (y)- (d«l»([x,y» =x(y«l» -[x, y] «I>=y(x«l» - (d9(x) «1» (y)
We suppose now that the relation is true until the Indice p-l, i.e. dC9CX)«I» - 9Cx) d«l>
for all «I> In CP-1(9, V). One considers an element «I> of CP(9, V). We have
l(y)(d9Cx) - 9CX)d)«I>
o
One deduces: dd«l> - 0 'if «I> e C (9,V) and, by inducUon,
dd«l>- 0 'if«l>e CP(9,V).
So dd-o.
The elements of the image BP(9, V) of ~-1 are called coboundarles of degree p (or p-
coboundarles) with values In V. From Theorem 1, we have BP(9,V) c ZP(9, V). The
quotient space
HP(9, V) - ZP(9, V) / BP(9, V)
a- H·(9, V) ~ W(9, W) .
It is called a homomorphism associated to a .
From this sequence, we can define another exact sequence of the cochain spaces :
a' 13'
o ~ C"(IJ, V1) ~ C"(IJ, V2) ~ C"(IJ, V3) ~ 0 .
In fact, let cI> be such that u"C cI» - 0, i.e. U 0 cI> - 0 . Suppose that cI> Is a p-cochain, then
vJ -
U(cI>(V1' ... , v~) - 0 for all v1 ,... , vp. As U is injective, «v1 ,... , 0 for all v 1•... , vp
and cI> - 0 . The exactness of the terms of the sequence could be verified In this way.
Passing to the cohomology spaces, we find the long sequence of homomorphisms
~ ~ So
o ~ HO (IJ,V1) ~Ho (IJ,V 2) ~Ho (IJ,V3) ~H1 (IJ,V1) ~
-131 51
~ H1 (IJ,V 2) ~ H1 (IJ,V3) ~H2 (IJ,V1) ~ ..... .
Ker ~P+1 - 1m Up+1 . So there is a cochain pin Cp+1CIJ, V1) such that <X.p+1Cp) - d'l'. But
dCup+1Cr)) - up+ldp) - 0 then dp - o. Hence, p is in Zp+1CIJ,V 1) and the mapping Op+1 is
lefined by op(q;) - p, where p Is the class of p in HP+1(IJ,V 1). This Is well defined be
cause the class p of p doesn It depend of the choice of the representative <p In the class
q;. In fact, we suppose that <p + d v Is another representative of q;. The cochain '1'1 In
In CP(IJ,V2), whose Image by ~P Is <p + dv. can be written as '1'1 - 'I' + ell; . Then
d'l'l - d'l'and this defines the same cochain p in Cp+1CIJ,V1).
Theorem 2. Let
IXo
-IIa
o ~ HO(9,V 1) ~ HO(9,v2) ~HO(9,V,) ~
Proof. It Is sufficient to prove that ao Is Injecdve, 1m Sp - Ker ap+l and 1m /3 p - Ker lip .
By definldon,
HOC9,V1) - ZOC9,V 1) - {VE VI: 9.Vl - 0 , 'V 9 E 9} .
By hypothesis, (X Is Injecdve. Then ao Is also Injecdve.
Let q; be In HP(9,V,) . By the construcdon of lip, we have seq;) - pwith pin Hp+l(9,V 1) ,
and (Xp+l(P) - dw with ~(w) - <p, Win CP(9,V2). The class of cohomology of ap+l(p)
is nonzero and 1m Sp is contained In Ker ap+l .
Now we consider 'if in Ker ap+l and 'II as a representadve of \if. We have (Xp+l(W) - c41
where Ills an element of CP(9, V2). Furthennore, one has :
d(J3pCJ,D) - ~p+ICdJ,D - J3p+1C«P+1CW)) - o.
Then ~p(l1) is In ZP(9,V,). We denote by vthe class of /3p(l1) In HP(9,V,). By the deftni·
don of 8p, we have 8p(V) - 'if then 1m 8p - Ker ~ 1 .
We can prove the second reladon 1m /3 p - Ker lip by using the same arguments, and the
theorem is proved.
itt lit
~ HI (9,VlW) ~ H2 (9,W) ~ ... ~ .
84 Chapter 3
11.2.1. V= K
We consider K as a trivial I}-module. The spaces HICI},K) are nothing but the
cohomological spaces for the De Rham cohomology of the left invariant forms on the
connected Ue group G associated to I}.
First interpretation. The space Z IeI},I}) Is the space of the derivations of I} and the
Cohomology of Lie Algebras 85
space Bl(9,9) is the space of the inner derivations of 9. Then, the space Hl(9,9) can
be interpreted as the space of the "outer" derivations of the tie algebra 9. The
particular case H 1(9,9) - {O} corresponds to the tie algebra 9 whose derivations are
inner. This is the case for semisimple tie algebras and for parabolic subalgebras of
semisimple tie algebras.
Second interpretation. The space H 1(9,9) can be interpreted as the set of the one-
dimensional extensions classes of the Ue algebra 9.
and
are eqUivalent if there is a homomorphism <p: ij ~ 91 of tie algebras such that the
diagram
Is commutative.
Let ex be a cocycle in 2 1(9,9). To this cocycle corresponds the one-dirnenslonal
extension of 'J :
112
o ~ 9 ~ ij ~ 9 E9 K ~ K ~ 0,
where i : 9 ~ q = 9 E9 K Is the natural Injection I(x) - (x, 0) and 1t2 Is the projection on K
86 ChapterS
o ~9~~-9E9K ~K
We can look upon this space as a set of classes of the cohomology of the 2-left
invariant forms on the connected tie group G associated with 9. Recall that if G is
compact, then we obtain the De Rham cohomology of G. We can also interpret
H2(9,K) as the set of classes of one-dlmensional central extensions of the tie algebra 9.
A one-dlmenslonal central extension of 9 is an exact sequence
0~K~9~9~O
o ~K-?9-KE99 -?9~0
Cohomology of Us Algebras 87
Remark. Let CCOI'.... COn) be a basis of 9- and ~ - 1: CIJt CIlj 'Ilk the structural
1\
The space Z2(9.9) can be Interpreted as the space of the infinitesimal deformadons of
the Ue algebra 9. and two cohomologeous cocycles define their equivalenttnfiniteslmal
deformadons. This permits us to Interpret the space H2(9.9) as the set of Infinitesimal
deformadon classes of 9. We will devote one chapter for the general study of
deformadons. However. In order to understand better this interpretadon of H2(9.9).
we give here some ways of approaching the topic.
This last condition means that all the coefficients of the fonnal serie
J.lt (J.lt (x,y),z) + J.ll (J.ll (y,z),x) + J.lt (J.lt (z.x),y)
are zero.
If we develop this fonnal identity, then the first relations are :
Cal) J.l(J.l(x,y),Z) + J.l(J.l(y,z),x) + J.l(J.l(z.x),y) - 0
It is the jacobi's identity for J.l.
Ca2) J.l(<P1 (x,y),z) + J.l(<Pl (y,z),x) + J.l(<P1 (z.x),y) + <PI (J.l(x,y),z) +
+ <P1 (J.l(y,z),x) + <PI (J.l(z,x),y) =d<Pl - o.
This relation is nothing but d<P1 - 0 and <PI is in 22(9,9).
So the first tenn <PI of the defonnation J.lt of J.l , called the infinitesimal defonnation
associated to J.lt Is an element of 22(9,9),
Two defonnations J.lt and Yt of J.l are called equivalenuf there is a family (ft) of
nondegenerate endomorphisms of the vector space 9 of the fonn
f t (x) - X + L
tl9 (x),
1>1
where 91 is an endomorphism of the vector space 9, such that
Yt (x,y) - f t- 1 (J.lt (f t (x),f. (y»), for all x and y in 9 .
This space coincides with the space of the obstruction of the deformations of I}. We
consider an element cP of the space H 2(1},1}). Are the infinitesimal deformations of this
class derived with respect to some deformations of I}? It is not the general case and
we shall indicate some necessary conditions related to H3(1} ,I}) for this to take place.
Let Ilt be a deformation of the tie algebra law Jl of I} and CPl its infinitesimal
deformation.
By writing Ilt - Jl + t CPl + t2 'P2 + ... , and using jacobi's conditions for Ilt, we obtain
k+I=I+2i
CPlOCP2i+l+ L CPk 0 CPI + 1. CPI 0 CPI = - dcp21+2
k+I=I+2i 2
k"l
90 Chapter 3
The first relation shows that <PI is in Z2(9.cp. The second relation concerns the
3-cochains <PI 0 <PI . We can see that if <PI is in Z2(9.9). then <PI 0 <PI is In Z3(9.9). and the
cohomologlcal class of <PI 0 <Pl corresponds to the classical product on the c1asses of
cohomology. We denote this by [<PI ]2. So this second relation couldbe interpreted E
a necessary condition for <Pl E Z2(9.9) to be a infinitesimal deformation of a
deformation J.lt of !.liS that [<PI]2 - 0 .
A:J we can find examples showtng that this condition is not sufficient, we can read the
thUd relation as
<Pl 0 cpz - -d <Il3 •
This shows that the 3-cochaiil <Pl cpz is a 3-coboundary (it is a 3-cocycle because cpz
0
verifies d cpz - <PI 0 <PI ). Thus. the cohomologlcal class of <PI 0 cpz vanishes in H3(9.9).
From the construction of cpz. this class doesn't depend of the choice of cpz. It is the
MMassey cube" of the class of <PI . We can denote It by [<Pl ]3 • Continuing the argument,
we see that for the existence of the deformation it is necessary that all the Massey
products of <Pl vanish; all of these powers are in the space H3(9.9). (Note that the
ith..power product is defined when the kth..power products are zero for k < I).
Conclusions. The conditions related to <Pl show that all Massey powers are zero. It is
clear that all these conditions are satisfied when H3(9.9) - 0 . We can conclude that if
H3(9.9) - O. every cocyc1e <Pl in H2(9.9) is the infinitesimal deformation of a
deformation (formal) of the law !.l of 9.
An element <P in <pl is called integrable if there is a deformation J.lt of !.l such that
J.lt-!.l+t<pl+t2 <p2+ ...
Remarks.
1. In fact, It is sufficient that a subspace of H3(9.9) vanishes for the integrability of any
cocycle <Pl In Z2(9.9). This subspace H3~9.1}) has been studied by Rauch [Raul.
2. We can also write the necessary and sufficient conditions for the integrability of an
Cohomology of Ue Algebras 91
element CP1 of Z2(9,9) In perturbation theory. Note that, In this case, we can entirely
solve the conditions system on CPl. As we can prove that the conditions of integrability
for perturbation theory and deformation theory are the same, we find another
approach of the space H3~9,9) (see Chapter 5).
Sl - 9 if I s: n1 ,
Sl - {oJ if I ~ n2 .
{ Fk(gr 9) - $ Si} .
I~
It is isomorphic to the filtration of the initial lie algebra 9 given by the sequence Sk .
Now we consider a filtered lie algebra 9 - UJd SI , with the filtration descending.
This filtration is called finite if there are m 1 and m2 in Z:: such that
VI - V if i:S: m 1 ,
VI - {O} if i ~ m2 •
The link between filtration and graduation is established in the same way as for Ue
algebras :
(1) Let V - EBlell V I a graded IJ-module. We put Fk(V) - EBI~ VI .
of a graded gr IJ-module :
Let ij E S; and vE VJ ; we put ij.v = IJv (this is correcdy defined).
Let IJ - EBleiB IJI be a graded Ue algebra and V = <:9 1e11: VI a graded IJ-module.
We can graduate the space of the j-cochains by putting CJ (IJ,v) - <:9 Cl(IJ,V), where
kell
c~ (IJ,V) - {c E CJ(IJ,V) : C(3.tl, ... ,a~) E Vh+ ...+lj+k, al E IJI} .
The coboundary operator d is in respect to the graduation, i.e. :
d(cl(IJ,v») c cr 1 (IJ,V) .
This permits us to graduate the cocyde and cobord. spaces :
94 Chapter 3
ZJ(9,V) - E9 Zl(9,V),
keZ
sJ(9,V) = E9 Bl(9,V),
keZ
d(9,V) - U Fk d(9,V) ,
keZ
where
As d(Fk c'(9,V» c Fk CJ+l(9, V), we can provide the spaces of the cocycles of the coboun-
daries and of cohomology by the filtration:
ZJ(9,V) - U Fk zJ(9,V) ,
keZ
BJ(9,V) - U Fk sJ{Ij,V) ,
keZ
Note: The correspondence between graduation and filtration that we have considered
for the 9-modules and Ue algebras, passes on the spaces of cocycles, coboundaries,
and cohomology.
Cohomology of Lie Algebras 95
N. SPECTRAL SEQUENCES
N.I. Definition
dy 0 <1r - 0,
dr(Ef,q) c ErrA-r+l,
PA Ker dP,q
(4) E - r with dfA is the restriction of d r to Ef,q .
r+l 1m d p-q,q+r-l
r
Remark. We often add the condition EtA - (OJ if one of the two indices Cor the two) is
negative. Such a spectral sequence is denoted as being of the first quadrant
Let m be an integer. We consider a spectral sequence (Er,dy)reN such that El,q - {O},
if we have p < m or q < m (the case m - 0 corresponds to the spectral sequence of the
first quadrant).
The condition (4) of the definition of aspectral means that Er+l is the cohomology of
the complex (Er,dr). Then Er+ IP,q is a subquotient of EtA. We show that EtA is also a
subquotient of EoP,q. For this we construct the sequence :
o c Bci,q c sf,q c ... c S~A c Z:,q c ... c zf,q c zci,q c Eci,q
such that:
96 Chapter 3
rn p,q
Tr : Z rp,q -+ S
Z f,q
p,q =Er+p,q1 the canonical prol--tion
J"'.... ,
r
zf,q - (cpC,q)-1 (z f,q),
then
E p,q = Zr,q = E p,q
- -Bf,q - r+1
Cohomology of Lie Algebras 97
We have previously defined the notion of a filtered 9-module. Here we use the more
general notion of a filtered module when the action of 9 doesn't occur.
is called filtered if all the modules AP are filtered and if the coboundary operators
respect these filtrations, that is :
{ forall p,
}
We can extract from such complex of modules the follOwing complexes:
dO d1 d2
Aq:O~AOq~Alq~A2q~ ... ,
We note H"CA) as the cohomology of the complex A, and H"CAq) as the cohomology
of the complex Aq :
HPCA) = ZP(A)
BPCA)
98 Chapter 3
P n dp+q-l (A q-r·
Brp,q - AP+q P+q-l)
In pardcular :
Z p+l,q-l - A p+q
-1 p+l '
BP,q - A p+q
-1 p+l '
Z p,q - A p+l
o P'
dFPCJ>+<I(I},V) c FPCp+q+~I},V) .
iss isomorphic to
and we have
In this notation, H·CI},h;V) denotes the cohomology of I} modulo h. with values in V (or
relative cohomology). The corresponding cochains verify:
c(al ,... , aq) - 0 if al E h. and dc(al ,... , aq+l) - 0 if al E h. .
It is the same as saying
cq(I},h.;V) - Hom k(A\~)' Y) .
The spectral sequence constructed in this way is called the Hochschild-Serre sequence
[H-S).
100 Chapter 3
Applications
This element doesn't depend upon the choice of HmCI},V)IJ. We denote tm+l as the
homomorphism
tm+l: Hn\'L,V)9 --+ Hm+i~' V'L)
that we have constructed. Therefore, we have the exact sequence :
o --+ Hm (.'l,
'L
v'l)
--+ H m (I},V) --+ Hm ('L,V)9 --+ Hm+l (.'l,V'L) --+ H m+1 (I},V)
1m rm tm+l 'L 1m+l
In this sequence, the homomorphism 1m is defined by looking upon the cochains of
2.
'L
in v'l as cochains of I} in V.
As for the homomorphism rm ' this is a homomorphism of restriction.
We suppose now that Hn(l,v) - 0 for 2 ~n ~m and m> 1. In this case, we have
the following exact sequence
Here the homomorphism r'm is the restriction to the first argument of the chosen
cocycle representing the given cohomological class for the ideal I. The
homomorp hi sm d 12 corresponds to d 2: E2m-l,m --+ E2m+lO
' .
In particular, if
Cohomology of Lie Algebras 101
then
2. Factorisation theorem
To end this chapter we shall discuss some loglcials of formal calculus which have been
developed. For example, Mapple and Mathematlca, etc. These IOglcials permit us to use
personal computers for the study of the cohomology of tie algebras. of course, the
calculus related with the cohomology is linear, but the dimension of the underlying
spaces and the rank of used matrices are very big, and this implies a quick limitation.
With these IOglcials we can determine without toil, not only the dimensions of the
cohomologlcal spaces, but also the basis of these spaces and also for dimensions of
large enough tie algebras (this depends only of the possibility of the computer). Here
we present a progrn.m using Mathematica which permits :
C1) the verification of the Jacobi Identities;
(2) the calculus of the dimensions of B2(IJ,IJ) and z2(lJ, IJ} ;
(3) the determination of a basis of H2(IJ, IJ) ;
C4) the verification of the nilpotency.
102 Chapter 3
mu: :usage =
"muL._] is the ley of Lie algebra"
Adj : : usage =
"Adj [V] compute the adj oint of the vector y,
Adj [k] is the adjoint of the vector x[k],"
Jacobi: :usage "Jacobi[ ] gives the Jacobi conditions,"
XilPot: :usage "XilPot[ ] gives the nilpotence conditions,"
DilllB2: :usage "DimB2[ ] computes the dimension of B2(mu.mu),"
DimZ2: :usage "DimZ2[ ] computes the dimension of Z2(mu.mu) ,"
DimH2: :usage "DimB2[] cOllputes the dimension of B2(mu.mu),"
BaseDer: :usage "BaseDer[] determines a base of Der(mu),"
BaseB2: : usage "BaseB2[ ] determines a base of B2(mu.mu),"
BaseZ2: :usage "BaseZ2[ ] determines a base of Z2(mu.mu),"
BaseB2: :usage "BaseB2[ ] determines a base of B2(mu.mu),"
·{'x. 'n};
Begin[" 'Private''']
mu[O. x_] := 0;
mu[x_. 0] := 0;
mu[x_. x_] := 0;
mu[x_. y_] := -Jau[y. x] I; OrderedQ[{y.x}];
mu[x_+y_. z_] := mu[x. z] + mu[y. z];
mu[z_. x_+y_] := muEz. x] + muEz. y];
mu[a_ x_, y_] := a mu[x,yl.
mu[x_. &_ y_] := a au[x,Y]j
(* ADJOIITS *)
(* JACOBI COIOITIOIS *)
Jacobi[] :=
Module [{i, j, k, LRel} ,
LRel = {};
For[i = 1, i <= n-2, i++,
For[j = i+1, j <= n-1, j++,
For[k = j+1, k <= n, k++,
LRel = Union[LRel, Jacobi[i, j, k]]
];
] ;
] ;
LRel
]
(* IILPOTEICE COIDITIOIS *)
lilPot[] :=
Module[{i, LRel, cadj} ,
LRel = {};
For[i = 1, i <= n, i++, LRel = Union[LRel, lilPot[i]]];
LRel
lilPot[i_Integer] :=
Module [{pol , lrel, laux, lambda},
pol = Det [lambda IdentityMatrix[n] - Adj [i]];
Select [Expand[CoefficientList [pol, lambda]],
Function [!lumberQ [#]]]
104 ChapterS
Desp[ee_]:=
Module [{eex, lv, v, rs},
eex = Expand[ee];
Iv = Reverse[Flattenrvariables[eex]]];
If[Length[lv] > 0,
v = Iv[[l]];
rs = Flatten[Solve[eex == 0, v]] [[1]];
Set [Evaluate[rs[[l]]] , rs[[2]]]
]
DimB2[ ]:=
Module[{A, a, i, j, f, Df, Base},
Base=Array[x, {n}];
A=Array[a, {n,n}];
f[v_]:=Coord[v] . A . Base;
Df[i_, j_] := mu[f[x[i]], x[j]] +
mu[x[i], f[x[j]]] -
f[mu[x[i], x[j]]];
n·2-Length[Flatten[Variables[A]]]
Cohomology of Ue Algebras 105
DimZ2[ ]:=
Hodule[{A, TA, a, i, j, k, f, Df, Base},
Base = Array[x, n];
A = Array [a, {n,n,n}];
TA = Transpose[A, {1,3,2}];
f[u_, v_] := Coord[u] . TA . Coord[v] . Base;
];
] ;
] ;
(* BASE OF H2 *)
LDif[genV_List. genL_List] :=
Module [{dim. v. p. 1n1. i. j. base. genv. sol}.
d = Dimensions [First [genV]] ;
base = RowReduce[Map[Flatten. genL]];
genv = Map[Flatten. genV];
dim = Length[genv[[l]]];
p = U;
For[i = 1. i <= Length[base]. i++.
j 1;
While[j <= dim AA base[[i. j]] == O. j++];
If[j > dim. base = Drop[base. {ill. AppendTo[p. j]];
];
ini = Length[base] + 1;
BaseDer [] : =
Module [{A. a. i. j. f. Df. base}.
base =
Array[x. in}];
A = Array [a. {n.n}];
f[v_] := Coord[v] . A . base;
Df[L. j_] := mu[f[x[i]], x[j]] +
mu[x [i] • f[x[j]]] -
f[mu[x[i]. x[j]]];
] ;
ParToBase [A]
ParToBase[lExPar_List] :=
Module [{lPar. dim. i. j}.
lPar = Flatten[Variables[lExPar]];
dim = Length[lPar];
VectorBase[i_] := lExPar I.
Table[lPar[[j]] -> If[j == i. 1.0].
{j. 1. dim}
] ;
Array [VectorBase. dimJ
BaseB2[] :=
Module[{BBaseEnd}.
BaseEnd = Map[Function[Partition[#. n]].
IdentityMatrix[n-2]
] ;
Map [Delta. LDif[BaseEnd. BaseDer[ ]]]
108 Chapter 3
Delta[A_List] :=
Module[{f, Df, base, Dfij},
base = Array[x, {n}];
f[v_] := Coord[v] . A . base;
Df[i_, j_] : = mu[f [x [i)], x [j]] +
mu[x[i] , f[x[j]]] -
f [mu [x [i) , x[j]]] ;
BaseZ2[ ] :=
Module [{A, TA, a, i, j, k, f, Df, Base},
Base = Array[x, n];
A = Array[a, {n,n,n}];
TA = Transpose[A, {1,3,2}];
f[u_, v_] := Coord[u] . TA . Coord[v] . Base;
];
J;
] ;
ParToBase [A]
]
End[ ]
EndPackage [ ]
110 ChapterS
First, one Introduces the constants of the structure by giving the dimensions of the Ue
algebra and writing the nontrivial brackets :
mu[x!l],x[jll - C1jkx[k] .
For example, for the flve-dimenslonal Heisenberg algebra, we write:
n - 5,
mu[X[2],x[3]] - x[l],
mu[x[4],x[5]] - x[l].
Now, the procedure Jacobi[ I returns all the Jacobi equations which are not satisfied. In
this example, Jacobi[ ] gives { l.
If we want compute the dimension of B2 or Z2 or H2, It Is suffldent to write dimB2[ ]
or dImZ2[ ] . or dImH2[ ]
To expllddy obtain a basis for these spaces, write baseH2I ], or baseZ2l lor baseB2[ ].
CHAPTER 4
COHOMOLOGY OF SOME
NILPOTENT LIE ALGEBRAS
Recall (see Chapter 2) that a (n+ 1)-dimenslon filiform algebra has a maximal Index of
nilpotency equal to n. Among all these algebras. only four play an Important role. They
are described In the basis (eo.el' ...•en ) by :
Proposition 1. The linear-mappings ad eo, ad el ,00', ad en_l ,h2' h3' ... ,bn-l, tl' t2 and t3
of the space Ln, with
hk(eJ-e'+k forany l~f~n-k,
Proof. It Is easy to verify that these mappings are the derivations of In and that they are
hneadyindependent
Let d e Der Ln . The Ue algebra Ln admits a natural graduation Ln - E9ie z(Ln>, , where
(In), is the space generated by the vectors (e,) for 2 ~ i ~ n, while (In)l is generated by
eo and e l . The other spaces are reduced to (0). This graduation is associated to the
descending central sequence (see Chapter 2). The derivation d can be decomposed
with respect to the graduation:
d- do + d l + 00. + ~ with ~ E DerIn .
for some 90. 9'0, 9t. 9' 11 92 ,•••, On e K . It follows from the equality
do [a, b] -[do (a), b] -[a, do (b)] - 0,
for a - e 1 and b - ek for 2 S k < n, that S' 1 - o. This same equatton, with a - eo and
b - ek for 1 S k < n, yields 9k+1- 90 + 9k . Considering the values CO, 1, 0), 0, 0, 0), and
(0,0, 1) for the trlpleCS o• 9 1, 9'0)' we arrive at t l , ~,and '3, respectively. Thus, we have
do - 91tl + 92t2 + 9'0 t::s .
Now consider d k e Der Ln with k ~ 1 . It is obvious that
Then, d'kCeO) - d'k(el) - O. We will show that d'k - O. Indeed, suppose d'k -:I- 0 .
Then there exists an integer m such that d'kCem) -:I- 0 and d'k(e~ - 0 for I < m
(clearly m ~ 2) . This Implies that
d'k!:eo, em.l] - [d'k(eO) ,em-I] -[eo, d'k(em-l)] - d'k(e m) -:I- 0 ,
In fact
as d - d_ 1 + do + ... + do with
diE Der{W n>1 - 1 :s I :s n ,
dl{W n>j C (W n)l+j .
Note that the Ideals CICWn) are conserved by the derivations; these Ideals are
characteristic. In the fonnal decomposition of d, d - Ea1ezd l , we have dk - 0 for k ~ -2 .
r
and
a = 0 with 0 ~ k ~ n-4 and k;t 1 for I - 1 and j- 2 .
This proves the proposition.
Proposition 3. Let tl' tz, ~, h5, ..., hzk-l be the elements of EndCQn) defined by
tlCeO) - - e1
tlCe~ - el if 1 SiS n-l
tlCe n) - 2en
tleo) - eo + e 1
t2Ce~ - (i-1)el if 1 SIS n-l
tlerJ - Cn-2)en
Then the endomorphisms ad e l Ci - O, ... ,n-1), tl' t2 and h 3 , h 5 , ... , h 2k-1 form a basis
of DerCQn)'
Proof. The independence and the existence of these endomorphisms in DerCQn) can
be easily verified.
One grades Qn by the filtration associated to the descending central sequence Cas for
Ln}. If d is in DerCQrJ, it can be decomposed as
d - do + ... + ~-l
with d l E DerCQ~ and dICQn)J c CQn)I+J for 0 SiS n-l and is j S n .
The derivation do verifies
116 Chapter 4
doCe o) - no eo + ~o e l
doCe l ) - al eo + ~l e l
doCe~ - Clj el if
This Identity, calculated for the pair Ceo,e~, gives a l+ l - ~l + lao (for i - 1,... ,n-2) and
Un - no + a n-l - ~o Cfor I - n-1). For the pair Cel,en_l)' one obtains an - ~l + an-I.
Then, all the coefficients can be ex primed with only a o and ~l. The derivations
corresponding to Cno - 1, ~l - 0) and Cno - 0, ~l - 1) are nothing but t2 and t l .
Now, consider the derivations ds with n-l ~ s ~ l.
Onehas:
dsCeo) - no e s + I
if l~i~n-k
ifnot .
Even If we add to ds a multiple of ad e s ' we can suppose <Xo - o. The Identity of the
derivation applies to the pairs Ceo, e~, and with 1 ~ i < n-s-l gives:
a l - <Xz - ••• - an-s .
The same identity applies to the pairs Cel,e~ with 1~, j ~ n-s, i+j - n-s, but for
s < n-l gives :
(-I)lal - (-I)lal where I+j - n-s.
If s Is even, we have necessary al - 0 and ds - o. If s Is odd., the Identity Is obviously
satisfied because I and j have the same parity (recall that n Is odd).
Then ds - 0 If s is even,
- a l es+1
dsCe~ lfsisodd.
We can conclude that
dzs'+l - h2S '+1 with 1 ~ s' ~ k-l and recall that d l - al ad eo.
The proposition Is proved.
Proof. Note that these endomorphlsms are the derivations of ~ and form a free
linear system. Let d be a derivation of Rn. We consider the graduation of Rn given
by Rn - E9 te z(Rn>t , where (Rn>t is 1he subspace of Rn generated by the vector el-l • 1-1..n+]
and (Rn)j - (OJ for J S 0 and J ~ n+2. Again. we note that this graduation Is not
associated to the filtration defined by the central descending sequence. Each Ideal
Cf(R.~Is stable for the derivations andls generated by the vectors e l+l •... , en assoon
as I ~ 1. Then. the derivation d can be written:
d - dol + do + ... + ~ •
with d t e DerCRn) , dt:Rn:>j c CRn)t+j
The same identity, applied to the pair (el,e Z)' gives CXz - a 4 • Then, we deduce a o - 0
and d l - 0 . Q.E.D.
Then Der n is the vectorial direct sum of the weighting subspaces VA' where the
elements of VA are the vectors having A. as their weight. If d Is anontrlvlal element of
VA' then d(Xa ) - c .Xp, where A. -
- a. This pennlts us to consider only the weighting
~
derivations for the description of Der n, I.e. the derivations belonging to VA for a
weight A. of the type A- ~ - a, where a, ~ are In d.
119
Proof. If n is Abelian. the theorem Is obvious. Suppose that n is not Abelian. Let d be
aderlvation with weight A. - P- a, where a and p are In .:1'. We disUnguishthree cases
CO 1.-p-a-O;
(II) 1.- p-ae.:1;
ClIO 1.-p-a~.:1.
Case (i). For all Indices i such that Xat....1 is In n. we choose an element h In Fa. such that
ad h (X at....1) - f (X at....1). Then the derivation d ' - f - ad h is rendered as zero for
every vector XCXi,rtl of n. We can suppose, by adding, of necessity, a derivation of D2,
that d' is zero on aU the vectors Xa of (n-{n-nD n ZCn). By using the fact that d' is a
derivation, we can easdy prove that d'CXa ) - 0 for every Xa of n.
the smallest Indice such that Xa..m is In n at is easy to see that such a root exists). Then
we verify that dCX) - 0 for every Xv In n.
Case (iii). Let d be a derivation of n such that dCXa) - ad X~, p - a e .:1, Xa ' X~ e n.
Besides, suppose that d is not In D 2 • Then
A. - (1.cp+acp+l +...+ a't - a p - ... - au) with <p S 't S P S u
or
120 Chapter 4
A. - (alP+alP+I +...+ a't - a p - ..• - au) with <P < 't < p-l .
or
~ - (alP +...+ ai-I) and a - (al+l +... + an)
in the first case, and
where 1 - <Pl < CP.z < ... < 'Pm S r , kl < k2 < ... < ksn S r+ 1 , 'Pi S kl .
If t, s, p are integers satisfying
kl < t ~ S k2 , P > <Pm , Xs,p En,
then the endomorphism of n defined by
~X (Xl,.) - Xs,p ,d(X (Xis) - Xl,:
is in Der n. We designate by D, the set of all these derivations.
Now suppose that R contains an element of the fonn acp,r+l" We put
R - {a lPl,kl,alpz,k2 ' ... , alPm.r+l }
with <Pl < CP.z < ... < <Pm S r , kl < k2 < ... < ksn - r+ 1 , 'Pi S kl .
If t, s, p are integers satisfying
Is kl , 'Pm-l < t S p S'Pm , x"t En,
CohomoIogyd some nIpoIent I..ie algebras 121
This theorem can be obtained direcdy by using the lemma and the definitions of 0,
and 04.
In this secdon. we shall describe the cohomology spaces with values In the adjoint
module of the "model" filiform tie algebra which is denoted ~ and which Is deflned,
in the basis eeO.e 1•••• ,ert. by the brackets [eo.ell - e l+ 1 1- 1•...•n. If F;:2e~.~) are the
elements of the filtration obtained by the central descending and ascending sequences:
then the space Fo2 2el.n.Ln) Is Identified as the "tangent spaces" at the point 1.0 to the set
of nilpotent tie algebras. This geometrical description will be developed In a later
chapter.
'I'(el,el+l)- {
el if i-k,
0 if not , }
\jJCeo,e~ - 0 ,IS i S n .
Proof. Let 'I' be a cocycle satisfying the hypothesis of the proposition, and we have :
'I'(ehel) - adeo('I'(ehel-l» -'I'(el+hel-l) .
This relation shows that one of the following conditions is satisfied
'll(ehel) - 0 if
(1) k<l<j;
(2) I<j~k;
This shows the existence and the unicity of the cocycle '1'.
We now construct the cocycles 'l'k,s for aliI ~ k ~ n-l and2k ~ s~ n (we need2k ~ s in
order to ensure that q{el' e n +l) - 0 for aliI ~ i ~ n) by putting
es If 1- k }
'l'k,s(ehel+ 1) - {
o if I *k
Proposition 6. The cocycles <j>IJ and 'l'k,s with 1 S i < j ~ n and 2k+ 1 ~ 1 ~ n form a
basis of FOZ2(~,~).
Proof. Let ~ E FOZ2(L ,L ) be a cocycle compatible with the grading that satisfies
CohomoIogyof some ~ Liea/geblas 123
Ii CCIn)f,CIn)f) c CIn)f+J . SubtracUng, If necessary, the cocycles <Pt", we may assume that
IiCeo,e.> - 0 for all 1 SIS n. If IiCef,el +1) - 0 for all 1 SIS n-l ,then we have Ii - o.
Let k be the minimal positive Integer such that IiCek,ek+1) - A.es with A.;e 0, A. E C and
s ~ 2k+ 1. Then the cocycle Ii' - Ii - A.'I'k,s sallsftes the condition ~. Cek,ek+1) - O. If
~. ;e 0, then we repeat the argument for Ii' . Continuing in this way, we eventually find
that ~ is a linear combination of cocycles of the form 'IIk,s. The linear independence of
the cocycles 'PtJ and 'l'k,s Is obvious, and the proof Is complete.
Corollary 1. The dimension of the space FoZ2CL,IJ is equal to C3n2-4n+ 1)/4 If n is odd,
and to (30 2-40)/4 If n If even.
The proof of Corollary 2 can be deduced from the fact that the cocycle <Pt", 1 ~ I < j ~
n, and the cocycle '1'1" are elements of the space FoB2CLo,Ln). In fact, fPtJ - df and
'1'1" - dg, where the endomorphisms f and g are defined by
fCe m) - e)-l If m - I ,
fCe m) - 0 if not,
~em) -eo If m - I ,
~em) - 0 ifnot .
The cohomology classes of the other elements of FoZ2CLo,Ln) are nonzero and form a
free system.
This study gives rise to the works of Kostant which gave a description of the
cohomology space H"Cn,V), where fl. is the nihadical of a parabolic subalgebra p of a
complex semlslmple Ue algebra 9, and V is an Irreducible 9-module restricted to fl.. If
124 Chapter 4
9 is simple, then for V, one can consider the Ue algebra9 with respect to the adjoint
action, and we obtain a description of H-Cn,9). However, for many applications Cthe
study of the automorphisms of the Ue algebra n, the problem of classifying Ue
algebras with a given radical, and algebraiC Ue algebras with a given nIlra.dlcal, the study
of the deformations of the Ue algebra n, and the study of a variety of Ue structures of a
vector space) it would be useful to have a description of the cohomology space
H-Cn,n). As Kostant observed by modifying his arguments somewhat, one could
obtain a description of the space H1Cm,m), where m is the nllradicalof a Borel
subalgebra of a Ue algebra 9. This would give a description of the derivation algebra
Derm.
In the follOwing two sections, we describe H 1Cn,n) and H2Cn,n) for the nUradical n of
an arbitrary parabolic subalgebra p of a complex simple Ue algebra 9. The prototype
Is the Heisenberg algebra. To save on language, we will speak of the nllradical of
parabolic algebras instead of the nllradical of parabolic subalgebras of simple
complex algebras.
Let us recall some classical notions. Let 9 be acomplex semisimple Ue algebra of finite
dimension, fI. a Cartan subalgebra of it, /1 a root system with respect to fl., and.
S a basis of /1 Ca system of simple roots). We have Introduced a partial ordering in fl.-
relative to which the positive elements are linear combinations of roots In S with
positive coefficients. This ordering divldes /1 Into aset of positive roots /1+ and aset of
negative roots /1-, so that /1 - /1+ u /1- .
We denote by 9 a the root space corresponding to the root <X.. We shall assume that for
each <X. E /1, some nonzero element e a E 9 a Is fixed.
Let Sl be a subsystem of s, /10 the set of roots that can be expressed In terms of S1'
Cohomology of some nilpotent Lie algebras 125
and fl.' = (fl. - fl.o) n fl.+. Then the subalgebra p - h. G) L I}a is parabolic, and
ae~u/1'
Is a reductive Levi subalgebra of It, and Its nilradical has the fonn fI. - L I}a
<xe/1'
Recall that any parabolic subalgebra Is conjugate to the subalgebra p for some
subsystem SI of S.
Let S - {aI' ... , an} and let ~ E fl.. We consider a decomposition kl a l + ... + k n an of
the root ~ Into simple roots. We call the Integer lC~) - kl + ... + k n the length of the
root ~. We consider the subset S -SI - {<XI,. ... ,<XI.} . We call the integer In(~) - ~, + ... + kl,
the fl.-length (or the S-Sllength) of ~. In the case S - S-Sl ' the concepts of length and
the fl.-length of roots obviously coincide. Since an arbitrary root decomposes into
simple roots with either positive or negative coefficients, we have
Recall that the length of a negative root is negative, then the I}I for I < 0 are not
necessary zero. In fact, if I}i ~ {OJ for i > 0, then I}-t ~ {OJ .
We call the resulting grading the natural grading of the lie algebra I} with respect to fl..
The subalgebra fI. = G) fl.i Is also ll-graded if one sets
IE!
Setting
126 Chapter 4
Fk(9) - $ 9i , Fk(n) - $ ni ,
i~k i.,k
we obtain the filtrations FCip and Fen) of the tie algebras 9 and n. We note that Fen)
coincides with the natural filtration of the nilpotent tie algebra n, i.e.,
Remark. All subspaces 91' I E 3::, are s-modules with respect to the adjoint action Cthe
subalgebra s coincides with 90)' Since s is reductive, these s-modules are semisimple.
In addition, the n-modules n, 9, and 9/n are semisimple s-modules. In what follows,
we identify the s-module 9J'n with the submodule $ iSO 9i in 9 complementary to n.
Now, the gradings of the tie algebras n and 9 define 3::-gradings in the spaces of
cocoons, cocycles, and coboundarles :
Cl (n,t) - $ C/ (n,t) ;
iEZ
zl (n,t) - $ Zi (n,t) ; B(n,t) - $ B! (n,t) ;
iEZ iEZ
where t - n, 9, or 91'n. This grading is compatible with the coboundary operator, and
we also obtain a 3::-grading of the cohomology space:
HI (n,t) - $ Hli (n,t) .
iEZ
where <p Is an Imbedding and 'II Is the natural homomorphism. This sequence generates
a long exact sequence
.
Ii 1-1
.
<PJ
.
IJIJ
... ~ HJ-l (n,.yn) ~ HJ (n,n) ~ HJ (n,l}) ~ HJ (n, .yn) ~ ....
We will note the space 1m <P*J c HJ(n,l}) by HJfund(n, n)_ It is called the fundamental
cohomology space. We can consider the space HJfunin, n) as a sub-s-module of
HJ(n,n) by considering a supplementary of 1m d·l-l ( HJ(n,l}/n)) in HJ(n,n) .
TABLE!
IJ 5 51
Al (Xl
An ,n ~2 (Xl
An,n ~2 (Xn
Cn,n ~2 (Xl
TABLE 2
IJ 5 51
An ,n ~2 (Xl
C n ,n ~2 (Xl, (X2
CohorooIogyd some ~ Liea/geblas 129
f.(e,)- {
eSg(II) , if 'Y - a
o, if 'Y¢ a }.
Sa Is a reflection of the space s" that carries A into Itself and such that sa(a) - - a .
The spaces of n.-Invariants elements in n and in'} with respect to the adjoint action
coincide (up to imbedding In '}) and are equal to the center zen). This means that
HO(n,tp - HOCn,n) and the mapping 8"0 in the sequence Is Injective.
(b) We consider the given long exact sequence. We know that
1m CP"l - Hlfund(n,n) ,
and, from the proof of (a),
H1 fund(n,n) - H1Cn,,}).
This means that the mapping cpo 1 is surjective and 1m'll" 1 - O. Since the long sequence
is exact, we have Ker 8"1 - 1m'll" 1 - 0, i.e., the mapping 8" 1 is injective. This proves
the theorem.
Remark. It is well known that Z1(n,n) coincides with the derivation algebra Der n.
Theorem 1, in essence, gives a description of Der n as an s-module.
In addition, it turns out that H2 fund(n,n) contains all the cocycles in H2(n,n) that
preserve the filtration, I.e.,
Now we examine the cases listed In Table 1 and describe the space H2(n,n). We
denote simple roots by a1 , ..., Un and number them according to Chapter 1.
(a) 9 - A1 and S Sl - (a 1) •
(c) 9 - Cn , en ~ 2) and S Sl - {a 1} •
The Ue algebra n is isomorphic to the Heisenberg algebra Hn-1 andone can Imbed it
Into An as the nilradical of the parabolic subalgebra defined by the subset S,
Sl - {a1,Un} and use Theorem 1.
As shown in the previous section, the description of the space H2(n,n) reduces to the
calculal10n of H1(n,CJ/n). In this section, we shall prove a number of lemmas that are
necessary for the description of H1(n,CJ/n) . In order to understand the ideaof the
calculations, we shall briefly describe their order.
same action which are semisimple. We call elements of the components of the
decomposition of these h.-modules into simple submodules weIght elements
Ccochains, cocycles, coboundarles, cohomologies). Let
n
This means that g is a weight cochain with P- L (XI • Since any cochatn (resp. cocycle,
1
coboundary, cohomology) in our case is a linear combination of cochains of the form
C·), to describe the space of cochatns Ccocycles, coboundaries, cohomologies) it m
enough to find all the weight cochatns that correspond to a weight A. of the form
L
~- ~ ai, where aland ~are roots,e lXl e n,andell e t.
Lemma 1. Let f be an element of C l(n, 9) such that fen) does not lie ~ FI 9 for
some i and let df(n) c FI 9 (d Is the coboundaryoperator). Then there exists
an element ae n l such that fCa) ¢o.
Proof. Assume the contrary, let fen l ) - 0 and let r be the smallest integer such that
fen~ ¢ O. We choose an element x e nr for which fex) ¢ O. Since x can be expressed ll'!iI
a linear combination of root vectors and since f is linear, we can assume that x - e p'
where ~ is a root with n-Iength r. Since r> 1, we can decompose ~ into the sum
~I + ~2 ,where ~I e Ak, ~2 e Am' 1 S k, m < r , and k + m - r. We have
By the hypothesis of the lemma, df (elh,elJJ E F1 9. Since ~epl'e/32] - NploP2 ' f (e p) is anoo-
zero element that does not lie in Ff9, this is possible only if f (e 13,) ¢ 0 or f (e~) ¢ O.
But this contradicts the fact that r is minimal. This proves the lemma.
Proof. By Lemma 1, there exists an e a E n1 such that fCea) "# o. Since f is a weight map
In C 1-1Cn,9), we have fCe a) - ae_~ for some ~ E nl_1 and for some nonzero a E C . By
the hypothesis of the lemma, dfCea,e~ E n. On the other hand, df E a2_ICn,9) and
dfCe a . e~ E no = s. which is not contained in n. Therefore, dfCe a . e~ - 0, and so
f[ea,e~]-[f(ea),e~]-[ea,f(e~)]-O .
If a + ~ ~ L1, then - a [e_~,e~] - [ea,f(e~)] - 0 • Since the first tenn on the left hand
side of this equality Is not equal to zero and lies in fI., the second tenn must also lie in fl..
Consequently, f(e~ - be_a for some nonzero b E C. We have ahp - bha , where hy
denotes the element leY' e_yl in fl., 'Y E L1. But this equality Is possible only If a - ~ and
a-b. This proves the lemma.
Proof. Itis c1earthatf (F2n) eF1n - nandf (n1) eno - s. Therefore, df(e~l,e~2) En
for any e ~1 ' e ~ En. This proves the lemma.
Now we construct some cocycles in the space 21_ 2(n,o/n) which, as will be proved in
the follOwing subsections, generate this space.
Cohomology of some nilpotent Lie algebras 133
Lemma 4. Let e a E nl be such that ,for any root y E do, we have a + y ~ d, and let
fEe lCn,9/n) be such that fCe a) - e_ a and fCe~) - 0 for any~, ~ =F- a. Then
f E ZlCn,o/n).
Proof. It is enough to verify the containment dfCev,e,;> E n, in the case when one of the
roots is equal to a Cif v = 't = a, then obviously df(e a , ea) = 0). We have
dfCea,e,;> - [e_a,etl .
The element [e_a,etl can fail to lie in n only when e t E n l and't - a E d. Let't - a - y.
Since e t , e a E n l , we have y E do and a + y - 't E d, which contradicts the hypothesis
of the lemma. This proves the lemma.
Remark 6. For the hypothesis of the lemma to hold, it is necessary and sufficient that
a E S- 51 and that all the simple roots connected with a in the Dynkin diagram also
lie in S-SI' In the case when n is the nilradical of a Borel subaigebraCi.e., in the case
when 5 - SI - S), the hypothesis of the lemma holds for all elements e a E n l .
Lemma 5. Let e a E n 1 be such that the set Ra of roots y E do with a + y E dis non-
empty, where for any yE Ra we have 2a+ yE d, and let f E c l (n,9/n) be such
that
f (e a) - e_a, f (e2a+y) - ~. [e_a,e a+y]
a,a+y
with f (e ~) - 0 for ~ =F- a ,2a + y. If 2a + y = a + (a + y) is the unique decomposition i
into the sum of two roots in dl forany y E R a , then f E ZI(n, 9/n) .
Proof. It is enough to show that dfCev,e,;> E n in the case when one of the roots v or 't
Csay v) is equal to a. We have
If e t E nl, where i ~ 2, then obviously df(ea,e t ) En. But if 't E dl and't =F- a + y,
where y E do, then df(ea,e t } = 0 E n. Let't - a + 1- Then
df(ea,et) - Na,a+yf(ea+t)-[e_a,et]
by the hypothesis of the lemma. This proves the lemma.
134 Chapter 4
Remark. In the case when fI. is the nilradical of a Borel subalgebra, there do not exist
elements e a E fl.I that satisfy the hypothesis of Lemma 5. Such elements also do not
exist in the case 9 - slCfI.,C).
Lemma 6. Let e a ' e~ E fl. I, a+ ~ E A, be such that ~ + "( eo A for any "( E Ao, andlet
f E CI(fI.,o/fI.) be such that
f(ea) - e_~ , f(ea+~) - t-. [e_~,e~] ,
a,~
Proof. A direct check shows that df(ea,e~) - 0 E fI. . If neither of the roots 't or v is
equal 10 a or ~, then obviously dfCe't,e) - 0 E fl.. It remains to examine the case
case when t or v (say t) is equal 10 a (or ~) . We have
dfCea,e) - - [e_~,e).
This expression can fail to lie in fI. only in the case when e y E n I and v - ~ - "( E Ao .
But this case is impossible Since, by the hypothesis of the lemma, ~ + "( eo A for any ~
where "( E ~ . This proves the lemma.
Lemma 7. Let e ex ' e~ E fl.1' a + ~ E A, be such that the set Rp ofroots "( E Ao with
~ + "( E A is nonempty, where for any "( E RW we have a + ~ + "( E A (except,
perhaps, for the case a - ~ + "(J, and let
f(ea) - e_~ , f(£ex+~y) - ~. [e_~,e~+y]
ex,ex+y
with f(e o) - 0 for a¢ a, a+ ~ + "( where "(E R~ U {O}. If a+ ~ + "(- a+ (~+"() is the
unique decomposition into the sum of two roots in Al for any "( E R~, then we have
fEZ I(fI.,o/fI.).
Proof. It is clear that df(ey,e~ - 0 for all v and 't, neither of which is equal 10 a or
a + ~ + "( , where "( E ~. Let v - a and 't ¢ a + ~ + ,,(, where "( E R~. We have
If e't E fl.I ' where i 2: 2 , then obviously dCea,e) E fl.. But if e't E fl.I and't ¢ ~ + "( , where
"( E ~, then f(ea,e) - o. Lett - ~ + ,,(for some "(E Ao. In this case also
CohomoIogyof some nlpotent Lie algebras 135
Remark. In the case when 11. is the nUradical of a Borel subalgebra., Lemma 6 is satisfied
by any simple roots a , ~ e S, connected by an edge in the Dynkin diagram, and there
do not exist roots a and ~ that satisfy Lemma 7.
Lemma 8. Let be e a e 11.1 be such that the set Raofroots ye .10, where a + yea,
Is nonempty but there exists y e Ra such that 2a + y ~ .1 and let f be a weighted
element In c1(n.,91n.) such that fCe a ) =ae a , a* O. Then f ~ Z1(n.,9/n.) .
Lemma 9. Let e a e 11.1 and e pen., a + ~ e a, be such that the set R p of all the roots
ye U.11 with ~ + ye a' Is nonempty, but there exlstsye R~ such that a+{3 +r ~ .1,
ISO
a * ~ + Y andlet f be a welghtelement In Z1(n., 9/11.) such that fCea) - 3e-IJ' Then
f e B1(n., 9/11.).
~+ a + y ~ .1, a * ~ + y, we have
136 Chapter 4
This expression cannot be zero, since hl3--'t + cha+'t "* 0 in view of the fact that we have
~ - 't "* a + 't Cotherwise a + ~ - 2Ca + 't) E .1., which is impossible). This means that
e1f'Cee, e) e: nand f e: ZlCn, 9/n). We have a similar situation in the second case. Thus,
if for some ee E n 1 we have f'Ce e) "* 0 , we obtain a contradiction. Let f'Ce e) - 0 for all ee
E n 1. According to the corollary of Lemma 1, f' - o. This is eqUivalent to the fact that
f - g E B1Cn, 9/n). This proves the lemma.
Lemma 10. Let e a E n 1 and e~ En, a + ~ E .1., and suppose that there exists
yE (~dl) U{OJ such that~ + y, a+ ~+ y E .1. and a+ ~+ y- 't+ v forsome't and v
in A' different from a and ~ + y. Then anycocyclef in Zl (n,9/n) such that f (ea) - e_j3
is cohomologous to zero.
CohorooIogy of some nilpotent Lie aIgebtas 137
Proof. We first consider the case when a. + ~ - 't + V , where 't, v E d' ,'t *- a. , ~, and
we consider an element g - a ad. e_a_~ in B1(n, 9/n) ; we choose the number a so that
f(e~ - gCe~ (it is clear that 't == a + ~ - v). For the cocycle f' - f - 8, we have
f'(e t, e v) - Nt,v f'(ea+~) -[e t , f'(e v)] .
for some constarlts No, N' and N". This expression lies in n n h and is therefore equal
to zero, i.e.
(1)
for some nonzero numbers N 1 and N2 (if, for example, N1 = 0, then h t - N2ha' where
t *- a., which is impossible).
Similar arguments for v instead. of 't lead. to the fact that either
f E B1(n,9/n), or
hv - N3h~ + N4 ha (2)
if
138 Chapter 4
n
L kt(alat ,
a - 1-1 (4)
is a decomposition of the positive root ainto simple roots (the kt(a) are nonnegaUve
integers). It follows from (1)....(4) that
t - NIP + N2 a, v - N,P + N4 a ,
where, since t + V - a + P and t c# a ,p, we have N, - 1 - N1 and N4 - 1 - N2 .
In addition, the numbers NI , N2 , N" andN4 are different from zero and one, and we
can assume they are rational.
Let us show that all the simple roots in the decomposition of a + P into simple roots
occur with mUltiplicity ~ 2. To do this, we consider several possible cases :
(2) 0 < NI < 1 and N2 > 1 All simple roots having mUltiplicity 1 in the
decomposition of P are con1ained in a as well (otherwise these simple roots would
occur in the decomposition of t with nonintegra! coefficients, which is impossible). In
addition, since 0 < N, < 1 andN4 < 0, from the second equality in (.5), all the simple
roots in the decomposition of a are con1ainedin P (since v is a positive root, all the
simple roots occur in the decomposition with mUltiplicity ~ 0). The simple roots that
occur in the decomposition of P but not in the decomposition of a, must have
mUltiplicity ~ 2 in P (otherwise, since 0 < N I < I, these roots would occur in the
decomposition of t with nonintegra! coefficients). Thus, a + P con1ains all simple roots
in its decomposition with multiplicity ~ 2.
(3) N1 < 0 and 0 < N2 < 1. In this case N, - 1 - N I > 1 and 0 < N" < 1, and, by
analogywith case (2), all the simple roots in the decomposition of a + P occur with
multiplicity ~ 2.
(4) N1 < 0 and N2 > 1. In this case from the first equality in (.5), all roots in the
decomposition of b occur in the decomposition of a. Since N, > 0 and N" < 0, by the
CohoImIogyof sane nIpotent Lie aJgebtas 139
second equality In (13), all roots In the decomposition of a occur In the decomposition
of 13. Thus, a and 13 have the same simple roots in their decompositions Into simple
roots. Consequendy, they all occur with mUltiplicity ~ 21n a + 13.
The cases N l' N2 > I and N l' N2 < 0 are impossible, since 't, v < a + 13 (with respect to
the partial order introduced in h.") and since 't, v E A+.
We note that there existthree cases In all when a root from A+ in the decomposition
Into simple roots contains all slnlple roots with multiplicity ~ 2 :
m 9 - E8 anda+ 13= 8is a maximal root;
CiO 9 - FIj and a + p = 8is a maximal root;
CliO 9 - G 2 and a + P= 8 is a maximal root
In all cases a slngularroot (see, for example, [KH2]), i.e., a simple root connected with a
minimal root in an extended Dynkin diagram), occurs in the decomposition of 8 with
multiplicity 2, but all other positive roots occur with multiplicity ~ 1. This means that
in the decomposition of the roots, a, 13, 't, and v, a singular root occurs with
mUltiplicity 1. This is possible only In the case when we have N 1 + N2 - 1 in (5).
Since N1 ¢ N2 (otherwise a + 13 - 2't E A, which is impossible), It follows that
Nl N2 I-N1-N2¢O
II-N1 I-N2
and we have
13 - N't 't + N' 2 V a - C1 - N'l) 't + C1 - N' 2) v , (6)
Let us consider a simple root in the decomposition of 8 that occurs with mUltiplicity 3
(such roots exist in all three cases). This simple root ao can occur in the
decomposition of a and 13 with multiplicities 0 and 3 or 1 and 2 (up to rearrangement).
In the first case, 't contains (in the decomposition into simple roots) <Xo with
multiplicity lor 2 (multiplicities 0 and3 are impossible, since N2 ¢ 0, 1). Thus, up to
replacing the equalities (5) by (6), we can assume that a and 13 contain ao with
multipliCity I and 2, respectively. Then the number N 1 + 2N2 , which gives the
140 Chapter 4
multiplicity of the root 0.0 in the decomposition of't, can be equal to 0, 1, 2, or 3. Since
N1 + N2 - 1, the only possible case is N2 - 2, N1 - - 1 (or N2 - -1, N1 - 2). Assuming
'3 "# G 2, we consider a simple root ~o in the decomposition of () occurring with
multiplicity 4. Denoting the multiplicities of the occurrence of ~o in the
decomposition of ~ and <X. by x and y, respectively, we have
-x+2y-ao, 2x-y-bo ,
where the pair (~, b o) can assume the values (4, 0), (3, 1), (1,3), and (0,4).
For all these values, system (7) does not admit integral solutions. We have concluded,
therefore, that case (b) is impossible if '3 "# G 2.
In the case '3 - G 2, the only situation in which (5) and (6) hold is
<X. - 0. 1 + 0.2 ' ~ - 2 0. 1 + ~ , 't - 0.2 ' V - 3 0. 1 + 0.2
(we use Bourbaki's notation [BOU31 for simple roots).
Butin this case, for 'Y - - 0. 1 we have ~ + 'Y E ~ and <X. + ~ + 'Y ~ ~,andby Lemma 9,
f' is a nonzero element in H1(n, .yn).
Thus, we have proved the lemma in the case when <X. + ~ = 't + v, where 't, v E ~. and
't "# 0., ~.
Now let <X. + ~ + 'Y - 't + v, where 'Y E U1g) ~I and 't"# <X. , ~ + 'Y but the root <X. + ~
cannot be represented as a different sum of this type. We consider a cocycle g ,
g - a ad e-a_p in Z len, '3/n) cohomologous to zero and choose a such that
f(ep+y) - g(e a +y). For the difference f' - f - g we have
f'(e,(,e v ) - N 1f'(e a+p+y) - [f'(e,(), e v] - [e,(, f'(e v )] - 0
If fl (e'() "# 0 (or fl (e v )"# 0), then - <X. - ~ +t - - A. E ~ and f'(e'() - ce_a_p+,( for
some nonzero c E (C. This means that <X. + ~ = 't + A. is a different representation of
<X. + ~ as a sum of two roots in ~, which is impossible by hypothesis. Therefore,
and we have
CohorrcIogyof some nHpotent Lie a/gebIas 141
Proof. Let f be a nonzero (weight) cocycle in Z\(n, 'lIn) where i < - 2. According
to Lemma 1, there exist a E III and ~ E UI~IlI, a + ~ E Il, such that f(e a ) - ae_~,
a;t: o. We consider the decomposition ~ - ~1 + ~2' where ~1' ~2 E Il'. If one of the ~1
The proof of the foI1owing lemma is similar to the proof of Lemma 10, and we omit it
Lemma 11. Let e a E n1 be such that the set R a of roots y E .1.0 with a + y EllIs non-
empty, where for any y ERa' we have 2a + YEA, and we suppose that there exists Y
E Ra u {OJ such that2a + Y- 1: + v for some 1: and v in III different from a and a + y. If
f E Zl(n, 9/n) is a weightcocyde such thatf(e a ) - e_ a , then f E B1(n,9/n).
n 2 ={(a,-a)/aE..1\, Ra- 0 },
Remarks. 1. In the case S - Sl - S (I.e., in the case when n is the nUradJcal of a Borel
subalgebra), we have n 1 - n, - Os - 0. In addition, the pair (a,-a) lies in ~ if and only
if a E S, and the pair Ca, -~) lies in !l.t if and only if a and ~ lie in S and are connected
by an edge in the Dynkin diagram.
f ,",Je~)
...,~ . _ { h a, if 't - a } , where 0)0 - (a,ho) E no ;
0, if't;t: a
e-j3, if 't - a
If 't¢a
CohorooIogy of some nilpotent Lie aIgebtas 143
e~ If't-a
0, If 't * a, 2a+y
e--J3' If't-a
0, If 't * a, a+~
if't-a
0, if 't * a, a+~y
Let
5
where 1 ~ i ~ 6, and let E - UI-O EI . It follows from Lemmas 3 - 7 that each element of E is
contained in Z l(n, IJ/n) and Is not cohomologous to zero. These elements define
nonzero elements In Hl(n, IJ/n). In what follows, for simplidty, we shall assume that
E c H~n., rI n} and identify cocycles that differ by an element in B~n., rI n} .
Let 9 E HI_ 2(n, cy'n) . We consider a root vector e a "f;. 0 (the existence of such an
element follows from Lemma 1). Then 9(e a) - ae_~ for some nonzero a E c: and for a
root ~ E d l with - a - ~ = A.. By Lemma 2, two cases are possible:
(a) ~ = a;
(b) a + ~ E d.
Ca) According to Lemmas 8 and 11, this case can also be divided into two subcases :
(al) Ra - 0;
Ca2) Ra "# 0, but 2a + 'Y E d and 2a + 'Y "# 't + v, where 't, v E d l and 't, v "f;. a ,
a + 'Y for all 'Y ERa·
In subcase (al), we choose an element f in <E2> such that f(e a ) - ae_a> and we
consider the difference g' - g - f .
In subcase~) we choose an analogous element f in < E3> and consider the difference
g-g- f . In both subcases, g is an element of Hl(n, 9/n) that vanishes at e a and
continues to vanish at those elements e t E nl at which g vanishes.
(b) According to Lemmas 9 and 10, this case also splits into two subcases, and in each
of these subcases one can choose an element f in <E4> or < E5> such that the
difference g' - g - f vanishes at e a and continues to vanish at all elements e t E n l for
which gCeJ - o.
Thus, in all cases, the number of root vectors in n l at which g' vanishes is one less
than the same number for g. If g'(eJ "f;. 0 for some e t E n l , then, repeating the
argument, we can find an f 1 in E2 u E3 u E4 U ES for which g" - g - f 1 vanishes at e"C.
Continuing in this way, we find an element ik) - ik-n - fk_l in HI(n, 9/n) that
vanishes on all of n 1.
According to Lemma 1, this is possible only if ik) - o. This means that g can be
expressed as a linear combination of elements in E2, E3• E4• and E5. This proves the
theorem.
CohonrJIogyd some niIpoIerI Lie algebras 145
Proposition 7. (a) If A. Is not a root, then EAo Is linearly Independent In H l(n, c.:J/n)
(b) If A.Isarootln A_1' then the system EAoIsJinearlydependentln H1(n, c.:J/n) .
(c) If A. is a root In ..:1_2' then EAols linearly lndependentln H1(n, c.:J/n) If and only If
MAo - a" u Ll:s where MAo - {(a. - 13) / a.,f3 E ..:11 ,A. - -a. - f3}.
Proof. (a) The weight elements of B1(n, c.:J/n) obviously have the form g - a ad e a '
where a. E u,so ..:1,. Therefore, any nonzero linear combination of elements in EAo
cannot lie in B1(n, c.:J/n) ,since A. G!: ..:1.
(b) Let g - adeAo E B1(n, c.:J/n), where A. E ..:1_ 1. For any root a. in ..:11 for which
a. + A. - - 13 E ..:1, the pair (a.,-f3) lies in 0 1. Therefore, one can find a linear combination
f of elements in Eo UE1 such that the values of the maps f and g are equal at all root
vectors e a E n 1 . We consider their difference f' - f -g, whichliesinZ 1(n, c.:J/n). The
value of f' at any element in n 1, by the choice of f', is equal to zero. According to
Lemma 1, this is possible only if f'(n)cn, i.e. if f' - o. Thus a nonzero linear
combination of elements in EA.lies in B1(n, c.:J/n) , i.e. is equal to zero in H1(n, c.:J/n).
(c) The proof of the sufficiency is completely analogous to the proof of (b). Let us
prove the neceSSity. If MAo ~ Ut u Os, then there exists a pair (a.,-f3) in MAo' where
A. - -a. - f3 , that does not lie in Ut u Os. For this pair, we have fi,.e a ) ~ N e~ ~ 0 , and
any nonzero linear combination f of elements in E1 is equal to zero at ea. Since any
weight element of Weight A. in B1(n, c.:J/n) is a multiple of g,f cannot lie in B1(n, c.:J/n);
that is, EAo is linearly independent This proves the lemma.
146 Chapter 4
Corollary 1 (from the proof). If A is a root and EA is linearly dependent (J.e., either
AE d..1' or A E A-2 and MA - 0.. u as), then :
(a) If A E A-2, any collection E' A that l1es properly in EA is linearly Independent
(b) If A E A_I any collection E' A'::) Eo n EA that lies properly in EA is linearly
Independent.
This description is based on the study of the s-module Hl(n,V), where V is a simple
9-module whose structure has been given in Kostant's works (see [KOD.
148 Chapter 4
Let n be the nilradical of a parabolic subalgebra of the simple complex algebra fJ,
defied by SI c S. Consider a simple fJ-module V of dominating weight g (see Chapter
1) and the corresponding Weyl group of fJ noted W.
We have seen that the spaces cl(n,V), Zl(n,V), Blen,V) and Hl(n,V) are provided to
structures of s-modules, where s is the LevI subalgebra of fJ given by
We put
We have
HJ(n,V) - L Wa,
aeW.
where Wa Is the simple s-module with domlnatlng weight ;a = a(p + A.) - P (A. is
the weight of V).
Let vo(J..) be a vector of V which Is nonzero and which has a weight equal to 00..) and let
(al ,... , ay be the ordered elements of U(J"
Then the cohomology classes of the cocycIe fa defined by
f a{X al ,..., XCIj) - va(A.)
is a primitive vector (I.e. anonzero vector corresponding to the dominating weight) of
the s-module W(J"
CohorTKJ/ogyof SO/7I9 nipotent Lie algebras 149
Remarks. 1. For the description of these structures, Kostant shows that the
dominating weight ~ - O<p+/J - p of the s-modules Wa are all different from each
other and distinct to the dominating weights of the simple components of the s-
module &n,V).
We recall that aroot a is singular if 3 - a E S, where 31s a maximal root. The simple tie
algebras of type Ar (with r > 2) owns two singular roots denoted a l and a r in
Bourbaki's terminology. The other types have only one singular root. But in these
cases, the mUltiplicity in the decomposition of 3 of the singular root is equal to two.
Now If ~ is a positive root, then 3 - Pis a root if and only if the singular root (or one
singular root in the Ar case) belongs to the decomposition of p, and Its multipliCity is
equal to 1 otherwise ~ - 3.
Following is the table of maximal roots, of singular roots and the expanded Dynkin
diagram for each type of simple complex tie algebras. Also we precise a number y(9)
which we wID use later.
150 Chapter 4
TABLE 1
Type of Singular
Dyukin extended diagram Maximal root y(g)
algebra roots
An
(n ~ 2) ~
al a2 On-Ian
al +a2+···+a" Ql,On 5
,
al a2 a3 0,,_1 an
Bn
(n~ 3)
L---~ al + 2a2 + ... + 2a" a2 3
-6
-6 al a2 Qn-lOn
en o::::;x:o---o- -- ~ ~ 2al + ···+2a,,_1 +a" al 4
(n ~ 2)
(n~4)
Dn
T---<::::-'-6
al + 2a 2 + ... + 2a n_2+
+an_1 +an
a2 3
al a3 a4 aG as
Es al + 2a2 + 2a3 + 3a4+
~
a2 2
+2a5+ as
-6 al a3 a4 aG a6 a7
E7 2al + 2a2 + 3aa + 4a4+
"
~
al 2
+3aG+ 2a6 + a7
la2
al a3 a4 0'5 a6 a7 as
Es ,.. ,.. ,... ,... 2al + 3a2 + 4a3 + 60'4+
I
"
~ "
~ ~ ~
f'o
as 2
+5a5 + 4as + 3a7 + 2as
la2 -6
-6 al a2 a3 a4
2al + 3a2 + 4a3 + 2a4
F4 ~ al 2
al a2 -6
G2 (;E3I$O--O 3al + 2a2 aa
CchomoIogyof some nlpotent Lie aJgebtas 151
In the first case, U a - {~1' ~2} andln the second case U a - {~1' ~2 + kI3 1}, where k Is
the largest Integer such that ~2 + k13 1 e A.
r(r-l)
- - + 2r - tl - t2 + t,
2
where tl ls the number or roots of S - SI whlch are on the boundazy of the Dynkin
dlagzam, ~ the number of pairs of roots of S - SI whlch are fastenedln the Dynkin
diagram and t, - 0 for the Ue aJsebras IJ of type A r, Br' Cr' 02' F4 and t, - 1 for the
other types.
The proof Is obvious after we have noted that If ~1 and ~2 satisfy the conditions of case
Ca), the reflexlons s PI and s liz commute.
The proof Is established by considering the tables of roots before the description of
roots tied to the elements of W2r We use same considerations for the follOwing
lemmas :
TABLE 2
9 S SI wt ~l j}z Y
A2 al,a2 0
al 0
An On
0
Cn al 0
~
154 Chapter 4
Theorem 6. If 9 is in the list of Table 2, then the s-modu/e H2 fondCn,n) is the direct
sum of s-submodules generated by the cocycles VI of the following form :
(e~l' ep.z) ~ eJ .
Now, one fixes for each complex simple tie algebra the Integer Y(9) whose value Is
given In the Table I.
We suppose that n is the nilradJcal of a Borel subalgebra CS 1 - 121). In this case, the
previous theorem can be formulated more precisely.
CohomoIogyd some n4:lotetJt Lie a/gebIas 155
TABLE 3
Uu lPa..1I\x4 a. a. Ii H~.nl
Proof. Point (a) emanates direcdy from Theorem 3 (III.5.2) and from Proposition
(III.5.2).
Point (b) emanates from Theorems 1 and 2 CIII.5.2).
for every nilradical of a parabolic subalgebra (with some exceptions). But each spaces
H2 fond(n,n) and Hl(n,9/n ) is very precisely described in the previous sections. If we
concentrate this study on the case when n is the nilradical of a Borel subalgebra, and if
9 is not of the types indicated in the previous theorem at the point (a), we reclaim the
theorem of Leger-Luks.
Theorem 9.
The dimension of the simple s-module W 0"1 whose dominating Weight Ais equal to
Then, we have
Theorem 10.
Remark. By a direct calculus, we can give the dimension of the space H2CH 1,H l ) Cwe
can also find the dimension of these spaces by considering H 1 as the nilradical of a
Borel subalgebra of A2. As for the algebra H2 , the calculus is also made using the Weyl
fonnula.
Theorem 11.
Proof. In effect we have dim Z2CH 1,H l ) - 0 , dim B2CH 1,H l ) - 3 and dim Von - 2 if
k- 2.
Cohomology of some nilpotent Lie aIgebtas 159
Let B be the subalgebra of M composed of upper triangular matrices, I.e. the matrices
A- Caq) such that <1jj - 0 for I > j. This subalgebra Is called a Borel subalgebra of M.
The set of simple roots is naturally denoted by S. So S - {(Xl. ~ '(X3 , ••• , <Xr ,•••). Each
root (X - (XIj' where (XIEA) - a. -an, can be decomposed into the sum of simple roots :
The unit matrix (root vector) associated to a root (X is denoted Xa and the
unidimensional space generated by Xa is denoted Ma' We define a partial order on the
set H' of the linear form on H by putting A. > 0 if and only If A. can be expressed as a
linear combination of simple roots with positive coefficients. It Is clear that each root
Is equal to a linear combination of simple roots either with the coefficient 1 or with the
coefficient -I, so that we obtain a patltlon of A Into positive (A+) and negative (A-)
roots: A - A+ U A-.
Let R be the subset of A+ that consists of the roots which are pairwise Incomparable
roots for the previous partial order relation. We put L - Ll3eb' M 13 where A' is the set
A' - {~E A+ 13 YERas ~ ~ Y} .
It Is clear that L Is a standard nllsubalgebra. Note that every standard nilsubalgebra of M
can be obtained as this, from some set R of roots pairwise Incomparable.
We consider in L, the natural filtration corresponding to the descending central
sequence
For example, if f is a weighting operator satisfying fCX a) - axf3' where a:¢: 0, then its
weight A is equal to ~ - (X.. Every operator of L can be fonnally decomposed Into a
sum Ceventually infinite) of weighting operators.
of root vectors. This means that only a finite number of weighting components of d
are not equal to zero. But, two weighting operators, having distinct weights, transfonn
a given root vector into two distinct root vectors Clf the images are not zero). This
proves the lemma.
identified M(n) to fl,(n,C)) and it is defined by the system R(n) - {<XtJ E R/ iJ:S; n} (see
Chapter 2). It is clear that the derived subalgebra of LCn) is nothing but D(L) (') LCn),
and we have
Proof. From Lemma 1, It Is sufficient to find an Integer n such that d(L(n» c L(n).
Let Xa be a root vector such that d(Xa) - ax~ with a :#: o. We consider the follOwing
three possible cases for weight 'Y - ~ - a of the derivation d:
Lemma 4. Let dbe a weightingdedvatlon of L such that d/[L,LJ:f:. o. Then there exists
a poSitive integer N such that d/LCn) is a dedvatlon of LCn) such that d/[LCn),LCn)J :f:. 0
foraJ1 n, n > N.
Proof. Let d[x,yJ *" 0, x, Y E L. We choose N as so to the matrices x and y lie in L(N).
From Lemma 2, d/ LCN) is a derivation of L(N).
the center of L. Then there exists an integer N such that d(LCn)) cL Z(LCn))
164 Chapter 4
for all n, n ~ N.
Proof. let x E L such that dCx) - y ~ Z(I.). let z be in L saUsfying [y,zl :¢: O. We can find
an Integer N such that x, y and z lie In LCN). By virtue of lemma 4, we can regard d/LCN)
as a derivation. Moreover d/LCNXx) - y ~ Z(LCn)). The lemma is proved.
let M be the IJe algebra of matrices A - (aq), IJ ~ 1, such that each line and each column
contains only a finite number of nonzero elements. Then L Is a subalgebra of M. We
note N ~L) as the nonnallzer of L In M. If x E N ~I.), the operator ad x Is an endomor-
phism of L which belongs to Der CU. Then the set D 1 - {ad x / x E N M<L)} Is a subset of
Suppose now that R Is a finite set and contains roots of the form a 1,k' We order the
elements of R as a 1J1 , a 12J2 ,••• ,aIoJo, where 1 < ~ < ... < ~. For every pair of roots
a- <X'J and ~ - ~k,m' where it S j S k S jr and ~ < m, such that Xa and Xp lie In L, we
Coht:JnrJIogy of some nilpotent Lie aJgebtas 165
Proof. We put a - a lJ and /3 - ak,m' By hypothesis, j "" m and i "" k. We consider the
root Y- am,s and we choose s to be large enough so that Xy ELand s > j. This choice is
possible because R contains an infinity of elements. As the set of inner derivations is
an ideal of DerCI..), then [d,adX~ - d'is an inner derivation. We have
d'CXa) - Cd 0 adXy - adXy 0 d)(Xa ) - ax't
where 'C - Y+ /3 ~ ak s' The difference 'C - a is not a root because k "" i and j "" s. If a"" 0,
we obtain a contradiction, d' being an inner derivation.
Proof. If R doesn't contain a root of the fonn a 1,k, then Theorem 13 follows from
Theorem 11 and the fact that ZCL) - 0 .
Let al,jl E R andlet d be a nonzero weighting derivation that doesn't lie in 0 1, From
Lemma 3 (IVA), there is an Integer N such that el/LCn) E OerCLCn)) andel/LCn) *- adx for
all x E LCn) and for all n, n ~ N. From the studies of derivations of fInlte-dlmensional
standard algebras C§ 1.5), the follOwing two cases are possible:
CO d/[LCn),LCn)] - 0 and d(1.(n) c Z(1.(n) ,
RCn) - {al,jl ,a i2j2' ... , a i,j,}' with 1 < i2 < ... < is·
Coho!ooIogy of some n8potent Lie algebras 167
We choose an integer n sufflciendy large for be sure that [L(n),L(n)] n Z(L(n)) - {OJ and
(X ~ ~ (it is possible that R is infinite). Then the weight of d is not equal to zero. From
Lemma 1, if ~ - (X ~ .6, then d - 0, which is inconsistent
Consider the vector space V of Infinite dimension provided with a countable basis
{eo,e 1, .•. ,ek, ...J. It is possible to make a structure of a tie algebra, noted Lo , from the
bracket
Remarks. (1) The algebras Lt are topolOgically nilpotent for the topology associated to
the filtration generated by this graduation.
We accept this proposition without proof; for an example, one can read [FU].
As Goncharova has composed all the cohomological spaces of ~ with values in (C (see
also [FUD, one deduces the follOwing proposition.
Proposition 10.
In fact this proposition has been proved direcdy by A. Fialovski [FI]. More precisely,
he gave a precise description of the cocycles of a basis of H2(L v L1).
i - 2, 3 and J~ 4 }
otherwise
L 3 (el,ej)-0 'r;f J
L3 (e2,e3) - 8e2
L3 (e2,e4) - 4e3
L3 (e3,e4) - - 10e4
L3 (et,ej)- {
(_1)1 ct2 (J - i + 3) ej+I-3, as i - 2,3,4 and
0, otherwise
f-4(el,ej)-O 'd J
f-4 (e2,e3) - - 14el f-4 (e2,e4) - 0 ; f-4 (e2,e5) - 8e3
f -4 (e3,e4) - - 24e3 f-4 (e3,e5) - - 16e4
f-4 (e4,e5) - 18e5
CohorooIogyof some nIpotent Lie aJgebtas 169
By using the MATHEMATICA program given In Chapter 3. we can easely compute. for
each nilpotent Ue algebras n of dimension less than 7. the dimensions of the spaces
Hl(n,n) for 1- 0, 1 and 2 and also the dimension of the spacesHl(n,CHor
dim n. Also this programm pennlts to obtain more Infonnations as the basis of these
spaces.
CHAPTER 5
1.1. Definition
°
T(ebeJ) - L T~J ek .
k-l
Identification of T with its structure constants determines, once the base {eJ is fixed,
n'-n2
an isomorphism between fi2•1 and (C 2 •
Definition 1. Two tensors T and T' of Tn2 •1 are called isomorphic if it exists
f E GLCn,(C) such that
TI (x,y) - f- 1 (T(fCx),fCy») , "i/ x, Y E (Cn .
To simplify, we will note that T' - f-l 0 T(f, f). This formula corresponds to that for a
change of basis. So the two tensors T and l' are isomorphic when the two tensors
have the same constants of structure in comparison with the two bases of (Cn.
We will denote by OCT) (or G(T)lf it is necessary to define the group G) the orbit ofT
under the action of GLCn,(C), i.e. is the set of isomorphic tensors of T.
Example. Let To be the tensor of T2 2,1 given by TOCe 1,e2) - el' We showthat To is
rigid. Every tensor of T of T2 2,l is expressed by
If f is the isomorphism with matrix Cat? ' then the tensor T' - f-1 0 TCf, f) is written as
If Ca,b) *" CO,O), we can always find an isomorphism uy so that we have T'Ce 1,e2) -
TOCel,e2) - e1' Particularly, if T is in a neighborhood of To, i.e. is if Ca,b) is neaI(1,O), then
Ca,b) *" CO,O) and T is isomorphic to To, which shows the rigidity of this tensor.
Consequence. Every tensor T of T2 2,l is either isomorphic to the nul tensor, or
isomorphic to the tensor To'
Definition 3. A lie algebraic law Jl on (Cn is a tensor of Tn2 ,1 verifying the Jacobi
identity.
~~x,y),z) + ~~y,z),x) + ~~z,x),y) - 0
for every x, y and z in (Cn.
We fix, once for ever, a basis Cel) of (Cn. We will note Ckij as the structure constants of
The Variety of Lie Algebras 173
J1 relative to this basis. Jacobi's Identity can be reduced to the polynomial e<pidons
system:
n
(S) 1:
1-1
cIIJCslk+c1jkcslI+c1kiCsIJ-O, lSsSn, lSI<j<kSn.
So we can consider a tie algebraic law J1 as a tensor J1 - (Ck~, the CklJ of which verifies
(S).
Remark.Until now, we have only spoken of tie algebras and a tie algebra has been
defined like a space vector 9 , provlded with a square brackets [ , 1verifying the Jacobi
Identity. If one identifies the vectorial 9 to ten and the brackets to a tie algebraiC Jaw,
the tie algebra Is nothing but a pair 9 - CJ1,te n), where J1 is a tie algebraic law. It Is clear
that most of the time, we wIllldentlfy 9 to its law.
One ldentlftes the law J1 with Its structure constants. The set lJl of the tie algebraic laws
~
Is then the subset of ~ 2 - T'2,1 defined by the system of polynomial equations (S).
n n"-n2
So L Is provlded with a structure of the algebraic variety Imbedded In tC2 .
to provlde Ln with the Zarlskl topology which Is less fine than the preceding topology.
We recall that a Zarlskl closed set Is defined by a finite number of polynomial
equations. In this chapter, somewhat contrary to the uses, we will consider Ln with a
metric topology. When It will be necessary to restrict oneself to the Zarlskl topology,
we will predse It
174 ChapterS
Examples.
Since every tensorT - CT 112, T2 l2) verifies (S), we Immediately deduce that 0 - 1"22,1'
Here the algebraiC variety L2 is nothing but a complex plane of dimension 2.
2. The variety L3
A tensor ~ - (c~J 1 sIs J s 3 and 1 S k s 3 veri8es the Jacobi conditions, If and only If we
have
The variety {,3 Is an algebraiC variety of «:9 defined by the system s,. We will see later
that this variety Is a reunion of two irreducible algebraiC components. The tangent
geometry of {,3 is more difficult to describe, because of the existence of singular
points. For example, the point ~ defined by C3 12 - I, c klj - 0 for the other Indices I, J,
k (~ corresponds to the Heisenberg law H3) Is a singular point
(A prime ideal is an ideal which verifies pq E I ~ pel or q E I.) One provides Spec(R)
Approximately, it remains to define the regular functions on Spec(R). The fibre reo of
the corresponding sheaf above the point I corresponds to RI.
We consider the ring R - C[X1, ... , XN ] of complex polynomials with N - (o3_n2:>12. Let
J be the ideal of R generated by the Jacobi polynomials of the system (S) and let
A - R/1. The scheme Ln is an affine scheme constructed from Spec A. We will not
distinguish, at least in the notal1ons, the scheme Ln with the variety lJl.
176 Chapter 5
Definition 4. A scheme CSpecCR),n 1s reduced 1f the local dng RCI) has no n11potent
1deal for all I in SpecCR).
This means that doesn't exist non trivial polynomial f such that Cf)p is in the Ideal of
theJacobi.
Let J.lo - (aktf be a point of the space Ln. The tangent space in J.lo to the scheme LD Cwe
can identify J.lo with a point of the scheme) is defined by the linear equations which
correspond to the homogeneous part with degree 1 of the Jacobi equations centred
on the point J.lo.
Examples.
(1) If J.lo - CO) is the Abelian law, so the tangent space Is eN, where N-Cn'-n 2)/2
because there is no component of degree lin the Jacobi equations.
(2) J.lo - (a'12 - 1, aklj - 0 otherwise) e V. The equations translated on this point are
written
and the equations of the plan tangent to the scheme L' on J.lo are
The Variety of Ue AJgebtaS 177
It is an 8-plane.
Theorem 1. The tangent plane at the poInt J.Io to the scheme LO coIncIdes with the
space Z2(J.Io,J,Io) of the 2-cocycles for the cohomology of J.Io with values In the adjoint
module.
In the notation Z2(J.Io,J.Io) we identify, to agree with the precedJng chapters, the law J.Io
with 1he Ue algebra 90 - (J.Io,ICO).
which is nothing else than the linearization of (S) translated. on 1he point J.Io - (C~.
The action of GL(n,IC) on 'Jb2,1 restricted to lJ1 induces an action on the variety of the
IJe algebra laws : two laws Jll and Jl2 are isomorphic, if it exists, f E G L(n,IC), such that
Jl2 - (--10 J1Cf,O.
Lemma 1. For every Jl E I,n Cor in 1'02,1) the orbIt OCID Is provided with the
structure of a dJfferentiable mando/d.
Effectively, OCID is the image through the action of the Ue group GL(n,lC) of the point
178 ChapterS
~ (considered here as a point of the vector space Tn2,1). The Isotropy subgroup of ~,
colnddes with the group of automorphlsms of~. Then It Is a closed subgroup and the
orbit Is Isomorphic with the homogeneous space GLCn,C) / IsoCJD. This proves the
lemma.
Theorem 2. Let ~ be in Ln. Then the tangent space at the point ~ on the orbit O(JD
colnddes with the space of coboundaries B2(~,JD.
Reca1I that the elements of B2(~,JD are coboundaries of the form !p - 'Ov,.f, where f
Is in g1Cn,tC) with
'Ol1f(X , V,) - ~f(X) , Y) + ~X , f(Y» - f(~X , V»~ .
Proof. Let ~l be a point close to ~ in O(JD (for the topology of the manifold O(JD which
coincides with the topology inducedby TD2,1) In OCJD. Then ~I - (-I 0 ~f,f) with f
near to the Identity In g1Cn,C). We put f - Id + g, where g Is a linear mapping given by a
matrix which belongs to the neighborhood of 0 in gICn,C). Knowing that, one can WIire
The algebraic variety Ln can be provided with two structures of topological spaces by
considering either the metric topology or the Zariski topology which, recall, is less
fine than the first. We can also consider the orbits, either like differentiable manifolds
provided with the metric topology or like topological spaces provided with the
Zariski topology.
Proposition 2. The two topologic spaces OCJ,D (metric topology) and OCJ,D (ZariskJ
topology) are homeomorphic.
This results from the fact that O(J,D is differentiable linear (see, for example, Mumford
"the Red Book").
According to the preceding proposition, it means that 0(/1) is open for the metric
topology.
Proposition 3. Only a flnited number of open orbits (or of isomorphic classes of rigid
laws) exist in Ln.
Examples.
1. Only one open orbit exists in L2, which is defined in /1 by
/1(el,e2) - 0
(Recall the implicit convention: the undefined brackets are nul).
In this case, L2 is hreducible and coincides with 0(/1) .
180 ChapterS
IlCe2,e3) - e 1 •
It is the simple Ue algebra law slC2,C). We will see later that all simple laws are rigid.
Ill(e 1,e2) - e2 ,
Let JIo be in LD. The orbit OCJlo) identifies itself with G LCn,C)/AutCJIo), where AUtCJ.Io) is
the group of isomorphisms of JIo. We have
We note that the Ue algebra of the Ue group AUtCJIo) is nothing more than the algebra
of derivations DerCJIo).
Examples.
Cl) If Ilo is Abelian, DerCllo) - glCn,C). Then
The Variety of Lie AJgebtas 181
dim 0(110) - 0 .
dim Der(J1o) - 6 ,
dim 0(110) - 3 .
(3) If Ilo is the law of simple algebra sl(2,C), then dim Der(J1o) - 3 and dim 0(1lo) - 6.
We can see very easdy that, after dimension 3, the variety LD is not yet irreducible. The
problem of determining the components Is also naturally presented.
Unfortunately, this problem Is a little difficult to tackle. We wdl see at the end of this
chapter that this determination Is only explicit for dimensions less than 7. The
problem is dlsplacedby an estimation of this number of components. But here also,
we need refined techniques, and the approaches are difficult. Nevertheless, the next
proposition allows us to envisage an estimation of the number of components.
Proposition 5. Let J.1o be a rigid law in {JI. Then the closure in the Zariski sense
But there exist component given by nonrigid laws. We can consider some "rigid"
families parametrized by one, two or several parameters. In theses caces, the orbits of
theses families give components. The foHowing proposition precise such orbits.
m. CONTRACTIONS. DEFORMATIONS
exist, itis inside Ln and is called a conlraclion of ~. We will note it, for example as ~oo.
So
Proposition 7. Let ~oo be a conlraclion of ~, then ~oo Is In the closure of the orbit
of~.
algebra. This noUon of contracUon was developed someUmes later by Inonu and
Wigner with the parUculariZaUon : In their works, they suppose that one subalgebra.
remains fixed throughout the contracdon. As this noUon Is alitde bit restricUve, It has
been generalized by Saletan, Levi-Nahas, and some others. If we carry to extremes the
generalizations of Inonu and Wigner, we will probably discover the original noUon of
contra.cUons!
In particular, we have
~cP - J.Io 0 cP - cP 0 J.Io as soon as J.Io E (,n and cP E "2,1 and Il 0 Il- 0
which implies
Proposition 8. A 2-cocycJe CP1 for a given Jaw J.Io dellnes a Ilnear defonnation if and
onJylfcp1 lsalawof (,n.
Proposition 9. Let f.1t - J.1o + 1: t l CPt be a defonnation of J.Io. Then the Ilrst tenn CP1ls
a 2-cocycJe for J.Io, i.e. 8J.1O CPl - 0
Effectively, 811o'P1 corresponds to the coefficient of the linear part of the formal Jacobi
Identity.
Remark. We can always consider a deformation f.1t of J.Io as a tie algebra on the field
C((O) of the formal sedes In t. This,moreover, permits us to generalize this notion In
consldedng not C((t)) but the dng of formal sedes with some variables as the space of
1he scalars. So one Is led to the very general notion of deformations parametrized by
C-complete local algebras A, so AlmAn Is of finite dimenSion, where mA Is the
The Variety of Lie Algebras 185
maximal ideal of A.
This generalization allows us to foresee a tie with the notion of contIaction. But we
prefer to present here amore direct aspect and most topologic than the perturbation
for tackling this problem.
This results from the Artin theorem relative to the polynomial developments with
coeffidents In Integer series.
(N.R)
The system (N.R) can be Interpreted as the necessary and suffidentconditions for a
vector <P1 of the tangent plan of Zariski in J.Io to the scheme Ln being the first term of a
deformation.
If this system admits a solution, we wiD say that the linear development J.Io + t <P1 is
Integrable. We also will say that <P1 is integrable.
186 Chapter 5
Theorem 3. Let ~ E (}l such that there exists an open U containing ~ whose
elements are sums of convergentdeformatlons of ~. Then a vector tangent<Pl to the
scheme (}lin ~ is tangent to the variety Ln if and only If there is a sequence of
tensors (<1>2, ... , CJln, ...) of Tn2,1 as CN.R) Is satIsfled.
Consequence. Let T2 be the smallest vector subspace containing the cone of the
tangents to the variety Ln in ~. If T2 Is strlcdy contalnedln Z2(~,~) , then the
scheme Ln Is not reduced In ~.
where ft-1 Is the formal Inverse of ft' Is called a deformation equivalent to the
deformation J.lt of 1he law JJo.
To give a precise sense to the object which represents a law close to a given law, and to
tum the formal obstacle of the deformations, we will Introduce the notion of
perturbation, which is nothing more than a law infinitely close to a given IawCthen it Is
a concept marvellously adapted. to the metric topology of LR). This notion Is based on
the notion of inftniteslmals. We will briefly recall the few basic notions of the theory of
the complex infinitesimal.
The infinlteslmals are Introduced In an axiomatic way into the classical axiom of the set
theory. First, we will describe this by summing up the consequences of this theory on
complex space vectors.
In the fteldof complex numbers C, there exist both standard and nonslandard
elements. Every sca1arconstructed. by formulas ofclassical mathematics is standard.;
for example I, 10, 10- 10, i are slandards Cusuallyevery set classically defined Is standard;
and also 1he fteklC Is Slandard.).
nonstandard elements would be standard and Axiom I would be at fault On the other
hand, there does not exist any subset of C constltutedby all the nonstandard elements
of C , because these elements are not, by nature, constructed by the formulas of the
classical set theory.
Saying that the two elements a and b of C are infinitely close, means that their
dJfference is infinitesimal ; we will write a ; b.
of course, this affirmation keeps a theorem with in the nonstandard frame, because
this extension is conservative: every theorem keeps a theorem in the extension and
every theorem of the extension expressible in the language of the initial theory, is a
theorem of this. If we suppose every variable to be standaJd, we will write : "every
complex standard polynomial admits a standard root".
The axiom of transfer permits us to affirm that this sentence is also a theorem
(non expressible within the classical theory of the sets) ; without this axiom, we only
would write "every complex standaJd polynomial admits a root", without we can
The variety of lie Algebras 189
We will currently use the transfer In "the opposite sense". For example, we suppose
that we want to prove C'J) i we can prove within the (nonstandard) frame, that every
standard polynomial admits a standard root. With the transfer principle, we deduce
that (1) Is true. The rewards of this approach reflect perfectly the qulntescence of the
nonstandard method: to demonstrate the transference of (1), we can use all the tools
that are produced by nonstandard analysis, which are the classical tools (within the
nonstandard framework we can dlrecdy demonstration en), to which are added the
specific Nonstandard tools. As we are richer the proof of en transfered can only be
more simple.
We easily prove the natural rules of calculus concerning the Infinitesimal and infinitely
large elements :
CI»+CI»-CI»,
Cl»xCl»-CI» ,
LxCl»-CI»,
L+L-L,
lJ+CI»-lJ,
lJxlJ-lJ ,
where CI» designates every Infinitesimal, L designates every limited and lJ an Infinitely
large. On the other hand, one can not pretend about the result of the operations of the
type:
CI»/CI»,
lJ+lJ,
Cl»xlJ,
We consider a standard norm In Cn (for example. that associated to Ihe usual Hermitian
product). If Y is Infinitesimal (resp. limited, resp. infinitely large). its norm is infinitesi-
mal Cresp. limited. resp. infinitely large).
Usually. the vectors Y and oy are linearly Independent. The aim of the next section is
to determine the relations between these two vectors.
This theorem is the basis of all calculus relative to the perturbations of a point in a
vectorial plane.
Comments
2) The vector V defines In a single way the flag (V l'CV l' V2)' ... , CV l' V 2' ..• , V k)). This
proceeds direcdy from relations Cu). The most esthetic illustration of this geometry
linked to alimlted vector Is based on the Euclidean plan. If we take a limited point M
situated on a standam smooth curve, the flag linked to this point is nothing other than
the flag given by the tangent vector, the osculator plane, the second osculator plane,
etc.
The flag (V 1,eV1,V2), ..•, (V1, V2 , ••. , ViJ) which is canonically linked to the vector V, Is
defined as follows: we consider a Cnonstandam vectorial) Straight line with a director
192 Chapter 5
vector V-OV. It Is Infinitely close to a unique standard straight line whose director
vector is V 1. We consider now a vector plane containing these straight lines. Let us
note that If these two straight lines are the same, the length of the decomposition is 1.
Otherwise, the plan is well determined. It is standard or not. If it is (standard), the
length of the decomposition is 2 and (Vi' V2) is a standard basis of this plane. If it Is
nonstandald, It Is infinitely close to a unI<Jle standard. plane Cits shadow) and (V l' V2) Is
a basis of this standard. plane, etc.
Proposition ll. The length of V colndcies with the dJmensIon of the smallest standard
vectodal subspace containing the InfInItesImal vector V-OV .
with ~ standard (eventually null) and Wi == O. Let us note V 1 as the standard vector
(ai' a.:z, ... , ~ and Withe infinitesimal vector (wi' w2' ••• , wri. By construction, the
vector V 1 is not zero, the component corresponding to £1 is equal to 1. If Wi Is zero,
decomposition Is achieved, it has a length equal to 1. Otherwise we repeat the above
mentioned procedure to the vector Wi whose component corresponding to £1 Is
Lemma. The inflnitesimals el and (Xl are equivalent, which means (Xl leI is
appreciable (this implies that (Xl is nonnuID.
Proof of the lemma. If (Xl leI is not appreciable, it is infinitely large or infinitesimal.
In the first case, we divide the identity en by (Xl and then take the shadow of the two
members. We get WI - 0, which is the contrary to the hypothesis made on the vectors
of the decomposition. In the second case, it is the vector V I which is zero. So the
lemma is proved.
So, the two vectors are equal and limited and their shadows are equal, this gives:
We are led to comparing the Infinitesimais Ez. all and a 21. The different situations are
condensed In the next scheme:
All these cases disprove the hypothesis. The only possibilities are
We can put
V2 ai W 1 + ai W 2 ,
-
We are now led to comparing the Inflnlte5lmals .:" a 21, a 22 and a 2,. The hypothesis
relative to the vecto!"S V 1 and W 1 Implies that the quantities a2/e, are not Infinitely
large. So .:, Is equivalent to one of the a2 1 , this gives the linear relations between V,
and W 1 for I S 3. This process decreases the number of terms In the left part of the
Identity (5). Before r steps (r S k) , we get :
k-1
Vr-1- ~ a1r-1W1 wl1h a!~l;tOands1andard.
1-1
We put
with
196 ChapterS
But the vectors W l' ... , Ws are standard and lineary independent. As Vk is also
standard, its decomposition in the frame WI' ... , Ws has standard coefficients. Every
scalar A.. , i - 1, ... , k and A.kak+1' <Xs is standard. As the a1 are inflnitesimals,
•••, A.kak+1 ...
we must have ak+1 - 0 , this gives the equivalence between the two decompositions.
In the first section, we exposed just enough for an infinitesimal study of the
perturbation problems of Ue algebra laws or, more generally, of the perturbation
problems in linear algebra and in algebraic geometry. of course, nonstandard analysts
does not confine itself only to these simple studies. The axiomatic presentation,
established by E. Nelson and called I.S.T. (for Internal Set Theory), is a universal
presentation of nonstandard analysis. But it can be good in precise application of
A.N .s., like here in the study of the perturbations of algebraiC laws, to group and to
understand what, in the Nelson axiomatic, playa fundamental role.
Let us take, for example, the properties of elementary classical analysis; the Nelson
axiomatic of course permits a wonderful use of inflnitesimals ; but somebody wtJl
reproach us for using too rich an axiom for the use we want, hence the temptation to
find a minimal model (but with risks limited to its own preoccupations). The
infinitesimals in the algebraiC problems which interest us, in fact intervene only in
situations of comparison, and the essentialness of the Nelson axiomatic consists in the
existence of inflnitesimals and into taking the shadowoflimited elements. Perhaps the
most characteristic example of this situation concerns the interpretations of the
famous theorem of Hironaka : that we can summarize as follows: let us suppose that
The Variety of Lie Algebras 197
AXIOM I (Idealization)
vhere 'ff resumes •for all standard •and ~. for all finite standards I
This is translated by : let R. be a classical formula; for finding an element x satisfying
R.Cx,y) for all standarr:l y it is necessary and suffldent that for all finite and standarr:l
parts F wecanflnd x - XF such thatR.Cx,F).
As applications, we deduce from (I) the existence of the infinitesimals in C (that was
announced as an axiom In the first part).
This most surprising consequence pennits us to show that all infinite standard sets
contain at least a standard element or that in a standard finite set, all elements are
standard. This will be the key for the study of rigid algebras by infinitesimal methods.
Now let us give the reason for this : every algebraiC variety admits a finite number of
irreducible components; if this variety is standard, then the components are standard.
198 Chapter 5
AXIOM S (Standardization)
This axiom permits us to construct standard elements from an internal propriety (i.e.
is expressible in a classical language, which does not use either the standard predicate
or its derivative) or external propriety (which is not internal). From this axiom, we
deduce the existence of the shadow of a limited complex element
AXIOM T (Transference)
Proof. As the sum and the external product are continuous operations, we can write
for limited vectors of V : "(x+y) - Ox + oy andif ex. is alimited scalar, "(ax) - °ex.Ox. This
permits us to provide OY with the structure of a vector space. To look for its
dimension, one considers a standard scalar product in (CD and an orthonormal basis of
V. Every vector of this basis is limited, admits a shadow, and its orthogonality implies
that these shadows are independent and form a basis of OY .
Remark. This theorem does not translate anything other than the compactness of the
Grassmannian manifold, although here no topology of this variety has been defined.
The Variety of Lie AIgebtas 199
1) Perturbations of polynomials
follows:
Theorem 6. The roots of Q(X) are infinitely close to the roots of P(X) and the
multiplicity of the (standard) root of P(X) is the sum of the multiplicIties of the roots
of QeX) which are inflnltely close to the given root of peX).
This theorem is the basis of complex algebraic geometry, which one easily understand
by its easy formulation and by the evident proof in the nonstandard. frame that the
approach of some problems of algebraic geometry, via the N.S.A., can be easily made
(see, for example, [G04D.
Let E be a complex (or real) standard. vector space with a finite and standard
dimension. The space EndCE) of the endomorphisms of E is also standard and we can
define, as previously, a natural notion of perturbation in this space. Then we can
observe the role of the decomposition for the limited vectors of EndCE).
200 Chapter 5
( 0 If g Is a perturbation off, It verifies gOO ;; f(X) for all X limited In E. In fact, a linear
mapping being conUnuous, we have for all limited vectors X : fCOX) ;; f(X). From this
wededuce:
00 We consider as1andard basis (e~ of E. Then, If gls a perturbation off, the matrix of
g relative to (e~ Is infinitely close to the s1andard matrix of f. In parUcular, the
characteristic polynomials of g and f are infinitely close. From the preceding remarks,
we can affirm
Within the framework of the nons1andard analysis, we are led to call s1andard a Ue
algebra whose cons1ants of structure relative to a S1andard basis of (Cn (n S1andard) are
s1andard..
This can Induce some confusion with the notion of s1andard nilpotent Ue algebra
Introduced in Chapter 2. It is difficult to modify the terminology of one or other of
these notions without loosing the relation with the eXisting literature. As the risk of
confusion between s1andard algebras (about the viewpoint of N.S.A.) and standan:l
algebras (like the nilpotent algebras of a parabolic algebra), absolutely excluded in this
treatise (we will neither talk about standard algebra, nor of s1andard nonstandan:l
algebra !!D, we will conserve for the two notions the word s1andard when necessary,
we explain what it means.
The Variety of lie Algebras 201
2) Decomposition of a perturbation
Let J.1 be a perturbation of the standanllaw ~ on cn. As these laws are elements of the
standanl vector space Tn2 ,1 of tensors (2.1) on cn, we can apply the theorem of the
decomposition of limited vectors.
Theorem 7. Let J.1 be a perturbation of the standard Ue algebra. J.1o on cn. Then there
are Independent standard mappings 'Pl ••••• <ilk in 'Jh2 ,1 and there are Infiniteslmals
£1' ••• , Et such that
We will see In the next section the relations which must verify the tensors CJlt , so the
Jacobi conditions are satisfied for J.1. We note now the difference between the
deformations and the perturbations. not from the conceptual viewpoint (these two
objects must represent a close law), but from the viewpOint of the presentation: a
perturbation is a law of Ln and is written as a finite sum of Independent tensors and a
deformation Is an Infinite formal series.
Let f be an Isomorphism (not standanl of cn) and ~ a standanllaw of Ln. The law
J.1- f- l 0 ~(f.f) Is certainly nonstandanlandls Isomorphic to ~. If this law is limited
(I.e. admits constants of structure limited with respect to astandanlbasls of CD), then
It admits a (standani) shadow 1hatwe will denote J.1oo j It Is a law of Ln.
202 ChapterS
This results from the definition of the contraction. So, conserving the above-
mentioned notations, we note that If Iloo (which Is standard) Is a contraction of the
slandaJd law ~, then there Is a perturbation Il of Iloo Isomorphic to the standard law ~
Let Il be a law of (,11 andOCw Its orbit. We have seen that the tangent spaceTIlOCw to
the variety O(W at the point Il, Is Identified with the space B2(/l,W.
The space of cocycles Z2(Il,W Is Identified with the tangent space at the point /l to the
scheme (,11, otherwise Z2(Il,W Is the fonnal tangent space at /l to 1Jl.
We must now detennlne the equations of the tangent plane to the variety Ln at a
regular point and the equations of the tangent cone In a singular point
Let J.It - /l + tCPl + t 2CJl2 + ... + tllc!>n + ... be adefonnation of the law /l of Ln. We write that
The Valiety of Lie Algebras 203
J1t formally verifies the Jacobi identity. First, we recall the notations in1roduced
previously. If Cj) and 'I' are in Tn2,l' \Ve note by Cj) ° 'l'the trllineu mapping vitb. values in
Cn defined by
~o Cj) - allCj).
+ tzP+l( 1:
l+j-2p+l
Cj)10Cj)j ) + ... - 0,
l<j
~Cj)I- 0,
t Cj)]'oCj)1 + ~CJl2 ,
CE.D.)
1:
l+j-2p
1 Cj)pOCj)p + a llCj)2p - 0 ,
Cj)MlJ + -2
kp
1:
l+j-2p+l
Cj)t«Pj + a llCj)2p+l - 0 .
The system CE.D.) , which has an infinity of equations, can be interpreted as the
necessuyandsufficiantcondltionsthat an element Cj)1 E Z2C~,~ is the first term of a
one parameter famdy, i.e. a defonnatlon of the law ~.
204 Chapter 5
i.e. [CPl ° CPl~ - 0, where [ 13 designates the cohomology class in the space H 'C"qD.
We suppose the first obstruction to be cleared up. Then the second obstruction is
written
This is interpreted in the followtngmanner : we give an element CPl E 'Jn2 ,l such that
511CPl - o. The first obstruction is :
In order that CPl is the first term of a deformation, It is necessary that [CPl ° <Jl.2~ - 0 for
every representative <Jl.2 of the coboundary CPl 0 CPl. It is clear that this does not depend
on the choice of <Jl.2.
Applications.
1. CPl defines an infinitesimal deformation of Jl if and only if
511CPl - 0 ,
CPPCPl - o.
2. CPl is the first term of a deformation of degree 2 Ci.e. all the coefficients of tl are null
for i ~ 3) if and only if
511CPl - 0 ,
[CPPCPl] - 0 ,
Theorem 9. Let CPl be in Z2CJ.l.,J,D. Then the obstructions for CPl' being the first term
of a defonnation, are In the space H3CJ.I.,~.
Proof. Let J.I.t - J.I. + tCPl + ... + tPcPp + .... Let us suppose that all obstructions up to order
k are cleared up, that is
where
'lip - L
I+j-p
CPloCPj + o)1CPP (+ -21 CPpOCPp if p is even)
I<p
with i + J - k + 1 and k - 2s or to
0- O)1CPk+l + CPl 0 CPk + ... + CPs-l ° CPs+l + ~ CPs 0 CPs
ifk - 2s-1.
In order that these equations be satisfied, it is necessary that
I+j~+l
y. CJ\0'PI (or 1: qlj0'PI + -2
1 'PsoCJls)
I<j
is a coboundary B3CJ.I.,J,D.
We can show direcdy that, if the obstructions are satisfied up to the order k, then
1: qljo{j>j (or 1: qlj0'PI + ~ CJlsoC/ls) is a cocycle of Z~J.I.,J.I.).
where the CI\ are standard and linearly independent In ')hz, I" Let us write that J.I.' verifies
the Jacobi Identities
J.l.1 - J.l.1 ° J.l.1 - J.l.0 J.I. + £ IS IL'P 1 + ••• + £1 ..• £kSIL'Pk + ~ell'Pl ° 'PI + •.. +
+ ell·" £k'Pl ° 'Pk + ~ell elz 'Pz ° 'Pz + ... + ell elz ..• £k 'P2 ° 'Pk + ... +
~'PI- o.
(E.P.) £z SIL'PZ + ·.. +£Z·.. £k SlL'Pk+2"£I'PI°'PI+·"+
1
2
z z
1. £I£Z···£k'Pk°'Pk -0 •
So the first term 'PI of a perturbation Is an element of Z2(J.I.,~. The system (E.P.) Is
Interpreted like this.
Theorem 10. An element 'PI of ZZ(J.I.,~ 1s the f1rst term of a perturbation J.I.' of J.I. if
and only1f there are b1llnearstandard 1ndependent forms cpz ••••• 'PIt andlnfin1teslmals
£1 •••.• ex such that (E.P.) 1ssatisfled.
The variety of Lie algebras 207
Remark. One will note that the resolution of the obstructions to the prolongation of a
cocycle Into a perturbation, makes to occur only a finite number of mapping, but the
prolongaUon of this cocycle in a defonnation is traduced by an infinity of conditions.
Theorem 11. Nontandard linear equation (E.P.) equIvalent to a standard finIte linear
system.
Proof. We will solve the equation (E.P.) by playing on the index k of the
decomposition of J.1' • This integer is called the length of J.1'.
(E.P.)
So an element CPl of Z2(J.1,J.1} Is the fIrst term of a perturbation with length Ilf and
only If CPl E Ln.
(We will compare this result with that of the infinitesimal defonnation.)
We have
aJ.1cP1 - 0
(E.P.)
208 Chapter 5
Let us suppose £2 ~ o. Then a lL'P2 - 0 implies 'P1 0 'P1 - 'P1 0 'P2 - 'P2 0 'P2 - 0 and
J.1 + £1'P1 is a perturbation of J.L
Suppose a l1'P2 ~ 0 . Then the vectors ('P1 0'P1 , 'P1 0<1>2 , 'P2 0'P2 , a l1<1>2) fonn a system of
rank 1. Effectively, if the rank of these vectors is equal to 3, there is a linear relation
with standard coefficients between these vectors :
a1 'P1 0 'P1 + a2 'P1 0 'P2 + a, 'P2 0 'P2 + a ... aIL 'P2 - 0 , al standard.
Before we interpret this system, let us introduce some notations relative to the
products of cohomology classes.
The variety of Lie algebras 209
o - a"cll(X },X2.X,.X4) -
=JI.(X 1.cll(X 2. X,.X4» - JI.(X 2.cll(X 1.X,.X4» +
+ JI.(X ,.cll(X ltX2.X4» - JI.(X4.cll(X ltX2.X,» - cll(JI.(X 1.X2).X,.X4) +
+ cll(JI.(X2.X4).X1.X,) - cll(JI.(X,.X4).XltX2) •
Lemma 2. Let [q>21, - 0 and let 'I' be a representative of q> 0 q> in 8'(JI.,JL). i.e.
a"'I' = q> 0 q>. Then q> 0 'I' E Z'(JI.,JL).
As the class of q> 0 'I' only depends on q> • we will denote it [<P'1,. This class exists only
iHq>21, -o. Let us suppose that [<P'1, - o. Let p be suchthat q>o q> = ap. Then we have
a(q>o p.+ q>o '1'0 V) - o. The class of cohomology does not depend on the choice of p
(nor of V). One will denote it [qtl1,. The system (E.P) for the perturbations of length 2
is also interpreted as :
Proposition 14. Let q>1 E Z2(JI., JI.). Then q>1 is the first tetm of a perturbation of
length 2 If and only If
(1) [qrJ - [cp'J - [q>4J - 0,
(2) the tank of the representatives of these classes in 8'(JI.,JL) is equal to 1.
210 Chapter 5
Remarks
1. The first terms of the perturbations of length 1 are directing vectors of the
generative lines to (J1 in ~ One can so write again the results relative to the
perturbations of length 1 :
Let be 'P1 E Z2(Ji.JD. Then 'P1 Is a gene13live lJne In Ji to (,D If and only If 'P1 0 'P1 - o.
2. In the case of the perturbations of length 2. the term 'P1 represents the tangent
vector to the curve described by the laws Ji- Jio + E1 'P1 + E1 E2 'P2. where 'P1 and 'P2
are fixed and c 1 - E2 / (a+bE2+cEl) Cfrom the equation E.P J. Such a curve Is a rational
CUIve at is cubic). the tangent vector of which In Jils given by 'Pl.
This equation represents a linear combination of the standard vectors 'PI 0 'Pj wuh
Infinitesimal coefficients.
Definition 12. The rank of Jils the rank of vectors {8J.l~ •...•8J.l'Pit • 'PI 0 'P1 .'P1 0 ~ •••••
Theorem 12. Let Ji be a perturbation of length k. Then the rank of Jils equal to the
rank of the vectors (<Pt 0 ~ 1 SIS j S k-l.
Proof. Let co be a linear form whose kernel contains the vectors (<1\ 0 ~ ,
After division by the Infinitesimal having the smallest module and by taking the
shadow, one obtains
The variety of Lie algebras 211
Then the vectors 511OCP2, ... , CPk 0 <i>k are In the kernel of 0) for every fonn 0) whose kernel
contains the vectors (CPI 0 CPJ) lSI S j S k-l. Then the vectors (511O<i>t) I - 2 •...,k and (CPI 0 CPk)
1- l •...,k. are in the space generated by (CPlo CPJ) l:S i :S j S k-l. This proves the theorem.
Consequence. Let CPl be a vector of the tangentfonnal plane In ~ to Ln. Then If CPl Is a
tangent vector to a curve In Ln passing through ~ (or CPl Is a vector of the tangent
line), we have:
Theorem 13. Let CPl E Z2(~.~) and Jet CPl be a tangent vector at ~ to Ln. Then we
have:
1: C~rk
~(X"XI) - k>1 •
f (Xl) - E2X l •
f(Xv- EXI with E::O. i-2....,2p+l .
Theorem 14. Every Ue algebra of dJmenslon 2p+ 1 provided with a contact fonn can
be contracted on the HeIsenberg algebra Hp'
The perturbation ~' of the Jaw of Heisenberg algebra has for its first term the cocycle
CPl defined by
The variety of Lie algebras 213
Proposition 15. Let F be an irreducJble closed set of 1..0 and ~ a point of F such
that dim F - dim Z2(~,~. Then
(1) ~ is a simple point of the scheme 1..0
(2) F is the only irreducible component containing ~.
identified to the Zariski tangent space of 1..0 (that is, the tangent space in ~ to the
scheme 1..°), the Zariski tangent space in C at the point ~ is contained in Z2(~,~. Then
dim C :s dim Z2(~,~. But
dim F - dim Z2(~,~ :s dim C :s dim Z2(~,~
which gives dim F - dim C and F - C . QED.
214 Chapter 5
2 2 --
(1) n-2 L is irreductible : L - O(a2) ,
where ~ is the solvable algebra defined by [Xl'~] - Xl" The dimension of the
component is 2.
(v) F.s, where F S Is the set of Ue algebras con1a!nlng one Abelian Ideal
dimension 4 (dim 20),
(vi) G.s, where G.s Is the set of Ue algebras whose radical Is fIlIfonn and of
dimension 4 (dim 20),
(vii) M.s, where M.s Is the set of Ue algebras whose nilrad!calls of dimension
4 and s Isomorphic to the direct sum HI ED C, where Hils the
3-dimenslonal Heisenberg algebras.
Of) O(sl(2,C) ED HI) (semldirect sum obtained from the action of sl(2,C)
on HI) (dim 30),
Ov) LO (sl(2,C) ED D2) (dim 30) (D2 is the iIreducible sl(2,C) module of
dimension 2),
(viii) RO (n.of). where n.of Is the 4-dimenslonal fIllfonn algebra (dim 30),
Ox) RO (C.s) (dim 30),
216 Chapter 5
where RO(a.) (resp. Lo(a.) ) Is the set of solvable Ue algebras Cresp. of Ue algebras )
I} - "' E9 t, where", Is the nilradlcal, t Is Abelian and reductive on 9 Cresp.
I} - "' E9 t E9 s, where s is a Levi subalgebraJ whose maximal Ideal with a nilpotent
tadicalls Isomorphic to a..
(dim 45),
Sketch of the proof. One begins to determine the components of the variety RD of
solvable Ue algebras. Let us agaJn take the notations introduced in this theorem. Let I}
be aUe algebra and a - D19 E9 fI" where fl,is the nilradical of 9. It is clear that a is the
greatest ideal of 9 whose radical is nilpotent. We denote by L"(a) Crespo R"(a)) the set
of the laws of 1,0 (resp. R D) which have a as the greatest ideal with a nilpotent radical.
In the next section, we will see the part which these algebras play in the
detennination of rigid Ue algebras.
Now, we note that if n :5; 7, then allUe algebras 9 E L"(a) are close (in the Zariski
sense) to a decomposable algebra of L"(a). This is verified by using the classification
of nilpotent Ue algebras of dimensions less than 6. We are led to introduce the subset
LO D(a) (resp. ~ D(a)) of L"(a) (resp. ~(a)) formed by decomposable algebras. The
previous remark mentions the density of LO(a) in L"(a).
218 Chapter 5
Lemma 1. The components of Rn(a) for n S 7 are the closed subset ROCa) with
dim a< n.
Proof. One directly verifies that there does not exist, for n S 7, any component which
contains only nilpotent laws.
The fact that the Rg(a) are component results from proposition CVI.l).
As for the determination of components of Ln, this is made by studying the spaces
LO nCa) and the determination of the orbits of the algebra 9 containing sIC2,IC) as Ue
subalgebras, and constructed from the irreductible representations of sIC2,IC).
VI.2.1. Definition 14. Let be ~ E Ln. One says that ~ is rigid if its orbit O(~) is a
Zariski open set
We have previously shown the equivalence of these definitions. The interest of rigid
Ue algebras in the determination of the components Is revealed in the next
proposition.
Proposition 16. If ~ is rigid, then the Zartski closure of its orbit is an irreductible
component of the variety Ln.
Corollary
1. There does exist only a finite number (for a fixed dimension) of isomorphic dasses
of rigid laws.
2. Every rigid law is isomorphic to a standard law.
The variety of Lie algebras 219
We now propose to determine the rigid laws. As there Is only a finite number of
Isomorphic classes, we have here a good family to classify.
Theorem 16. Let J.l E Ln such that H2(J.l,J.L} - O. Then J.lis tlgld.
Proof. We have seen that the formal tangent space In J.l to Ln Is Z2(J.l,J.L} and the
tangent space In J.l to O(J.l) Is B2(J.l,J.L}. If H2(J.l,J.L} - 0, then Z2(J.l,J.L} Is the tangent space
to Ln. As It Is equal to B2(J.l,J.L}, so the theorem of the Inverse functions shows that 0 (J.l)
Is open In Ln. From where we have proved the theorem.
CO, 0, 1, ... , 9), then mllx admits for eigenvalues (0, A", A. 1' ... , ~), where A, Is infinitely
close to l. But every nontrivial standa!d vector Y - aXo + bx verifies ~YCx.) - c,X,
, I - 1,... ,9 and C,:F. o. The Jacobi conditions Imply that A" - 0 and mllx Is
diagonaliZable, the eigenvalues being (0,0, A.1' ""~) With A, == I. One chooses abasls
(X, Uo, Ul""'U~ of eigenvectors of m~X verifying U, == X, ,I - 0, ... ,9. Jacobi'S
r
Identities show that the vector J.LCu1, U Isanelgenvectorfor the eigenvalues A, +\ If
this vector Is nonnull, then A, + A., Is an eigenvalue Infinitely close to (I + p. One
deduces the next system:
220 Chapter 5
Ao-O,
t..l + t..2 - t..3, t..2 + t..4 - t..6,
t..l + t..4 - t..5, t..2 + t..5 - t..7,
Note that IlCUI,Uy"# 0 as soon as lloC~,xy"# o. The resolution of this linear system is
written
t..2 - 2t..1'
t..'\ - 3t..1'
t..1 - t..4 + (i-l)t..l' i - 5, ...,9.
Now we can calculate the dimension of H2(f.1o,f.1o) by using either a program of formal
calculus,or a reduction via the Hochschild-Serre series, to the calculus of H2(n,n),
where n is the nilra.cUcal. One show
dim H2(~o'f.1o) - 1 .
Conclusion. There are rigid Ue algebras whose second group of cohomology Is not
null.
Let us note that the above-mentioned example is the most simple example known of
rigid algebra with nonnull cohomology. We could show directly that all rigid Ue
algebra of a dimension less than 8 have a trivial group of cohomologies.
We don't expose the classical proof of this theorem. We prefer to prove directly the
rigidity of 9 to understand better the phenomen on which Implies the rigidity of
simple and semisimple Ue algebras.
Proof. Let f.1o be a law of a semisimple complex tie algebra. Then every perturbation
~ of f.1o is also semisimple. Effectively, If Ko (resp. K) designates a KillingCartan form
of f.1o (resp. K), the form K verifies K == Ko. Then it is nondegenerated and verifies
OK - Ko. As ~is semisimple, ltadmits a Weyl base (U a , Va' Za)aeD verifying:
222 Chapter 5
and the scalars a(H~, and other constants of structure only depend on the weights of
the roots. One knows that a(Ha) ¢ 0 and that
K(Ya,Zp) - 0, K(Ya,Yp) - 0,
K(Ua,Zp) - 0, K(Ya,Y a ) - -2,
K(Ua,Yp)-O, K(Za,Zp)-O,
K(Ua,U a ) - - a(HaL K(Za,Za) - -2.
The vectors UIXI (i - 1, ...,r) , Ya (a E A+) and Za (a E A+) generate the Ue algebra J.L
This basis is orthogonal for K. It also generates the real compact form J.l.1I of J.I. and the
Killing form KII of J.l.1I correspond to the restriction of J.I. to J.l.II, the real law
corresponding to J.I.. Remember that KII is a scalar product
For example we can refer to the preceding chapters. The constants of structure of
simple algebras only depend on the roots, and on the lengths of these roots.
Lemma. The shadow of the Weyl basis of J.l.ls a Weyl basis of Ilo.
We show, firstly, that °CJ.l.II) is the real compact form of Ilo. Effectively, if 911 Crespo
90 II) designates the real compact subalgebra of the algebra 9 Crespo 90), we have :
The variety of Lie algebras 223
The vectors
U' a - U a yl a _ Ya Z' a _ Z a
a ' b ' c
have a nonnuU shadow in the algebra. 0 (9 11). But
As aCHa) is standard and strictly positive, c/ab is limited. We show also that b/ac and
c/ab are limited. Let us suppose that c/ab is infinitesimal. Then
But
nonnulshadows in 130m - 09 11. As the shadows of the vectors of the Weyl basis Ua,
Ya , Za ' a > 0 are orthogonal for the scalar product (- Ko m), they are also
independent As dim 90 II - dim 9 11 , then we have a basis of 90 m and a complex basis
for 90. So the constants of structure of 9 with respect to the Weyl basis are standard.
They stay by placing their shadows. We recover, therefore, a Weyl basis of 90. This
proves the lemma.
The theorem is deduced direcdy from this proof. The constants of structure of 9 with
respect to the Weyl basis are equal to the constants of structure of 90 with respect to
the Weyl basiS, given by the shadows of the Weyl basis of 9. Therefore, these two
algebras are isomorphic.
In fact, b is defined from the roots system. As this is rigid (from the precedent
lemma), the rigidity of b will be deduced from this.
Remark. The rigidity of the Borel subalgebras is shown classically by the Nijenhuis-
Richardson Theorem. But the nullity of H2 is, to our mind, a consequence of a
relatively long and han:! calculus.
In the preceding section, we have shown the importance of the rigid algebras in the
study of the components of a variety of Ue algebra laws. We propose to approach
here the classification of solvable rigid Ue algebras and to develop a technique for the
construction of these algebras.
The variety of Lie algebras 225
Examples.
1. Every algebraiC tie algebra is decomposable.
2. Let 9 be the tie algebra defined by the next law:
Jl(X2 , X4) - - ~ X4 + X3
The nilradical of 9 is generated by (Xl' X3, ... , X2p). It is isomorphic to the abelian
algebra (C2p-l. Every supplementary subalgebraof n is one dimensional. It is abelian
and possesses a basis fonned by a vector of the fonn X2 + U with U En. As adOS +
U) is never diagonalizable, this algebra is not decomposable.
Let us remark that this algebra is not rigid. In fact, consider the perturbation Jl' of Jl
defined by
We will use this result for the construction of rigid solvable Cnonnllpotent) lle
algebras.
Let ~ be a law of Ln whose associated lle algebra. 9 Is solvable, non nilpotent and
decomposable.
As ad. X Is diagonalizable and n Is ad. X-invariant, one can find a basis (Xl' X2 , ••• ,
Xp+q ,..., Xn - X) of eigenvectors of ad. X such that Cx 1 ' X2 ,... , Xp+ct Is a basis of n,
(X p+q +1 ,••• , Xn) a baSis of t and Cx p+1 ,••., Xn) a basis of VoCX).
We will note this system by SOO or, more simply, S, when the regular vector X Is
fixed.
Theorem 19 (ofthe rank), If the Ue algebra IJ Is rigid, then for allreguJar vector X,
one has rankCSCX») - dim n - 1 ,
Proof, In fact, the eigenvalues A.I of ad X are solutions of SCX); this results in the Jacobi
equations:
2- case , The solutions of SCX) verifies a l - 0 for i - p+1 "'" n-1. The constants of
structure of ~ can be written like this :
~(X I ,Xj) - al X for i - p+q+ 1 "'" n and j - 1 "", P with aJ '" 0 when i S j S P ,
~(XI,Xj) - L
lSkSp
~jXk for i-p+l"",p+q and j-1"",p,
~XI , Xj) - L
lSkSp+q
a1\j Xk for 1 S i < j S P with a~j - 0 if aln + aln ",a~ ,
228 Chapter 5
Vp+q+l - (a1p+q+1 , ••• , aPp+q+1' 0,0,...,0) ,..., Vn - (a~ ,..., aPn ,0,...,0) .
Let cp be the bilinear alternative mapping defined by q(X, Xl) - ~IXI i-I,... , P , the
other products being null. We will show that the perturbation J1 + ecp is not
isomorphic to J1 as soon as e is a not zero infinitesimal. Suppose that 9 is rigid; the law
J1 + Ecp is in the orbit of J1 and cp represents a tangent vector to this orbit at the point JL
As the tangent space to the orbit coincides with the space B2(J1,J!} of2-coboundartes
for the Chevalley cohomology of 9, we must have cp - alLh where aIL is the
coboundary operator of this cohomology and h an endomorphism of c n . We put
p+q n
p+q
o-l:bkalkj ISi,jSp,
p+l
p+q
o - l: hk alkj p+ 1 S i ,j S p+q
p+1
This shows that the vector Y -1:;"1 hi XI verifies J.1(Y, XI) - ~I XI for i-I, ..., P
and J.1(y, XI) for I ~ p+ 1 . As for the vector Z - Y - ~+q+1 hi X" It is In the nilradica1
and verifies
The variety of Lie algebras 229
This is impossible and the vectors V, Vp+q+ 1 ,... , Vn are not linearly independent
Corollary 1. II 9 is rigid, the rank of the linear system 01 roots does not depend of
the choice of the vectors (Xl ,... , ~) .
Corollary 2. If 9 is rigid, there is a basis (~+q+l ,... , Xn )of t such that the
eigenvalues 01 the operators ad. X. are integers.
In fact, if we give a basis of t, the eigenvalues of the operators ad. X, for every X in this
basis, are solutions of the system seX). As this has integer coefficients, we deduce the
corollary about this.
Remark. The above-mentioned corollary does not imply the rationality of the
solvable rigid lie algebras. In the next section, we will give an example of nonrational
rigid algebra.
In this section, we will consider the consequences of the previous theorem on the
classification of solvable rigid lie algebras whose nilradical is filiform. The choice of
this propriety concerning the nilradical is not haphazard: a good approach of the
classification of the nilpotent algebras is studying of the characteristic sequence which
is an invariant related to the problem of the perturbations. Then the characteristic
sequence is classified by a natural order relation and the filiform nilpotent lie algebras
correspond to the nilpotent algebras whose characteristic sequence is maximal. So
the classification of the rigids begins naturally by considering by filiform nilradical.
230 Chapter 5
In this whole secdon, we wtll suppose that I} Is solvable. not nilpotent, decomposable
and the nllradical n Is filiform. Let us suppose that there Is a basis (Xl' .... Xn) of I}
such that
(a) X - Xn is a regular vector of I}.
Such a basis always exists. If the nilradical n Is a defonoadon of the fillfono graded
algebra.1n defined by a cocycle whose minimal homogeneity degree Is at least ecpd
to 1 en is not a defonoadon of 1he flilfono algebra. Qn)'
Note that. recendy • we have examlnated all rigid Ue algebras whose nllradical is
fllifono and eliminated this previous hypothesis concerning the basis ([G.A)). But the
method Is the same. This general case is based on the classlftcadon of flUfono Ue
algebras having a nonnull rank. We prefer to present here the parUcular case
correponding to the existence of an operator ad X which Is diagonalible on t and
jordanizable on n.
Proposition 18. Let I} be a solvable rigid. Ue algebra whose nilradlcalis fIlifonn and
which is provided with an adapted basis. Then the external torus is of dimension 1
or 2.
Proof. Consider the linear system of roots S corresponding to the basis of the
previous lemma. Its rank is greater or equal to n-'. Indeed, from the definidon of this
system, the equaUons Xl + "I - "1+1 • 2 SiS k-l • fono a subsystem of S. More.
S contains the equaUons Xk+J - O. 1 S JS n-k-l (see the proof of the theorem of
rank). SO the rank is greater than by n-'.
The variety of Lie algebras 231
Theorem 20. Every solvable rigid lie algebra of dimenslon n, n 2: 4, whose nJbadJcal Is
flMorm and of codJmenslon 2, Is IsomorphJc to the lie algebra defIned by
As the rank of every system of roots Is equal to n-3, the nontrivial brackets are the
square brackets resulUng from the ftlifonnlty of the nilradica1.
The proof of the rigidity of the above-menUoned law,just as the other laws which
appear In this work, Is direct : one sbJdies a perturbaUon, we compute the rank of the
assoda1e<ilinear system and one concludes that the perturbaUon Is Isomorphic to the
given standard law.
Theorem 21. Let 1.:1 be a rigId lie algebra whose nilradicalis fl1Jform and of
codimenslon 1. There Is a regular vector X E t such that the eIgenvalues of the
operator ad X are (1, k, k+ 1 ,..., n+k-3, 0) where k Is a posltlve Integer.
Proof. Effecuvely, let X be a regular vector and (XI' ...' Xn_l ) abasisofelgenvectors
of ad X given by Lemma 1. Let (AI ,... , ~_I ,0) be the corresponding eigenvalues. If
Al - 0, the Jacobi ldenUUes Imply At - A, for 2 ~ I, j ~ 0-1. As 1.:1 Is nonnilpotent,
one of the eigenvalues Is nonnull. This shows that ~ ¢ 0 and we have ~(Xl' XJ) - 0
2 ~ I < j . The rank of the associated linear system Is equal to n-2 and the Ue
algebra 1.:1 cannot be rigid. So Al ¢ 0 and the eigenvalues At are, two by two disUnct
232 Chapter 5
From the theorem of the rank, we deduce the existence of two integers i and j such
that !lCX1 , Xj) * 0 . As the product of two eigenvectors is also an eigenvector, there is
an index to such that !lCX 2 , XjO) is a nonnull eigenvector of ad X. The index h of the
corresponding eigenvalue ~ verifies h ~ jo.
Let us remember the hypothesis (H) : 9 is rIIJId solvable and its nilradical n is of
codimension 1 (other cases have been studied). We note X as a regular vector and
suppose that its eigenvectors (1, k , k+ 1 ,... , n+k-3 ,0) with k E if" . We suppose
thatk ~n-4.
Theorem 22.
1) If n - k+4 and n O!: 6 then 9 is Isomorphic to the Ue algebra defined by :
[Xn' XI] - X I ; [Xn, XI] - (k+i-2)XI , 2 S; i S; n-1 ; [X I , XI] - X I+I , 2 S; i S; n-2
[X2 , X,] - Xn-I .
Remarks
CO The proof of this theorem only goes through the resolution of the equations
applied to the A; deduced from the Jacobi equations.
cli) This theorem gives a start to the classification of rigid Ue algebras whose nilradica1
is filiform and of codimension 1. The case of a codimension greater or equal to 2 has
The variety of Lie algebras 233
been deduced. In the previous section. The case of the codimenslon 0 (of rigid filiform
algebras) has been studied by Carles. In an unpublished work, Carles demonstrated
that such algebras do not exist (see also [GAD.
(iiI) As for the general classification of rigid Ue algebras whose nilradicals are of
codimenslon 1, it Is necessary to examine only the rools system (1 • k • k+ 1 •...• n+k-2
• 0). As k Is an Integer. the most natural process Is to vary this parameter k In the finite
Interval determlnated by the above-mentioned theorem.
VTI.6.2. Case k = 1
VTI.6.3. Case k = 2
These algebras have been partially studied by Bratzlavsky and Carles. In [BR21.
Bratzlavsky considers the nilpotent Ue algebras n defined by [~. XJl - ~J XIJ with
I+J < n-l - dim n and supposes the follOwing conditions: ~J ¢ 0 for 2 S J S n-'
and ~ ¢ o. Then he construcls the solvable algebras of dimension n whose
nilradical Is n by putting [X. Xl] - I~ for I - 1 •...• n-l . These algebras are
decomposable by construction (9 - n E9 tCX) and are complete : HO(9 • 9) -
H 1(9 .9) - (O}. carles shows that. for n ~ 12, rigids exisls with H2(9. 9) ¢ (O} • These
algebras. In the terminology of the previous section. correspond to the system 0 • 2 •
2 •...• n-l .0). We want to generalize this work.
Theorem 23.
(j) If n S 5, 9 is not rigid.
(Ii) If n - 6 and if 9 is rigid, so it is isomorphic to the next lie algebra
Proof. We know that there is a basis (Xl' ... ' Xn_l ,Xl such that [X, XI) - i XI for
1 SiS n-l and [Xl' XI) - Xl+l for 2 SiS n-2. This theorem is a consequence ofthe
theorem of the rank.
Lemma. The JacobI Identities of weight p are the same for all the Ue algebras
verlfylng the given hypothesis as soon as dim n ~ p.
pS8 no reIa1ions
P - 11
P - 12
Theorem 24. The only rlgJd Ue algebras of dJmenslon 13 verlfylng the hypothesIs (H)
are lsomorphJc to one of the foDowing algebras
It) ri13: [X, Xt] -txt 1 SIS 12 i [Xl, Xt] - Xt+1 2 SIS 11 ;
[X2 , X3] - X5 ; [X2'~] - X6 ; [X2' X5] - 9/10X7 ; [X2' X6] - 4!5Xs ;
[X2 , X7] - 517'X, ; [X2' Xs] - 9/14XlO ; [X2'~] - 7/12Xl l ;
Proof. If Al '# ° then we can 1ake Al - 1 and the solutions of the system are :
CAl' ~. ~. A4) - (1,0,0,0) which corresponds to the algebra 9 113
CAl'~'~' A4) - (1,VI0,V70,V420) which corresponds to the algebra 9213
- 2~ + (2° degree) - °,
2a4 + (2° degree) - °,
Jacobi equations with weight less than 14. Preparation of the induction
Jacobi equations with weight p S 14 are parametrized by (AI' ... ' A5), ; also these
parameters are solutions of the equations of weight p S 12. These last ones being
solved, we will write only the simplified equations of weight 13 and 14 by considering
the solutions found
Then the solutions are (Al' A2. A3. A4, A5) - (0,0,0,0,0) which corresponds with
the nonrlgld algebra
(Al' A2, A3, A4, A5) - (0,0,0,0,1) which corresponds to another
rlgid nonlsomorphic algebra which Is noled 9'14 '
Remark. The hypotheses made by carles and 8ratdavsky correspond in the above-
mentioned study to the case Al - 1 .
238 Chapter 5
The study of Jacobi equaUons of weight 14, by the previous remark, is reduced. to the
study of the case A,1 - 0 ; this implies ~ - A., - 0 . These equaUons are wlitten :
Then A,4 - 0 and the only soluUon (A,I' A,2, A,3, A,4, A,s) - (0,0,0,0, 1) deflnes the
rlgid tie algebra 9 3 15 •
Conclusion. The 14-dimenslonal rigid lie algebras are isomorphic to one of the four
fol1owing algebras :
9 114 - tED f13 ' 9 214 - tED w13 ' 9'14 and 94 15 .
The 15-dimenslonal rigid lie algebras are isomorphic to one of the three foJJowing
algebras:
9 115 - tED f14 , 9 215 - tED w14 and 9'15
Theorem 25. Every solvable rigid lie algebra 9 - n ED t such that the nJlradica1 n Is
fllifonn and ofcodJmension 1 and admitting a regular vector X whose roots are equal
to Cl, 2, 3, ..., n, 0), n ~ 12, is isomorphic to one of the fol1owing lie algebras :
The variety of Lie algebras 239
ali) case n - 2p + 1
rf2P+2 associated to (AI, ... , Ap-I) - (0, ... ,0, 1) ,
l:(i2P+2 associated to (Al> ..., Ap-I) - (0, ... ,0, 1, (p-l) (p-2) /2) ;
ali) casen-2p+2,
rf2P+2 associated to (Al> ..., Ap-I) - (0,0, ... , 0, 1) .
1st case: n - 2p . We suppose that the Jacobi system only admits as asolution CAl' ... '
Ap-~ - CO, ... , 0, I) or (AI' ... ' Ap-2) - (0, ... , 0, 0) . This is the only rigid Ue algebra
corresponding to (Ap ... , Ap-2) - (0, ..., 0, I). Then, if n-2p+ 1, the solutions (AI, ... ,Ap-
2,Ap-I) of the Jacobi identities must verify Al - ... - AP-3 - 0. The equations of weight p
- 2p+l are reduced to AP-2 [Cp-l)Cp-2) AP-2 - 2 Ap-ll- °and this implies CAl'···' Ap-2'
Ap-l) - (0, ... , 0, 0) or (AI' ... ' Ap-2' Ap-l) - (0, ... , 0, 1, (p-2)(p-l)12). The algebras
corresponding to the two last solutions are rigid.
2° case: n - 2p + 1
The hypothesis of induction is : the only solutions of the Jacobi equations are CAl' ... '
Ap-2' '1>-1) - (0, ... , 0, 0) or (0, 0, ... , 1) or CAl' ... ' Ap-2' ~-l) - CO, ... , 0, 1, (p-2)(p-I)/2).
Only the algebras corresponding to the solutions (AI' ... ' Ap-2' Ap _l) - (0, 0, ... , I) and
(Ap ... , '1>-2' Ap-l) - (0, ... , 0, 1, (p-2)(p-l)l2) are rigid. Let us take n - 2p+2. The
equaUons of weight p - 2p+ 1 are reduced to :
240 Chapter 5
2P -I )
(
AP-2 I.
I-I
(p-i)(p-i-l) AP-2 - 2(p-l)Ap-1 -0,
We can direcdy verify that there Is no nilpotent rigid Ue algebra of dimension less or
equal to 8. The approach developed in the previous section permits us to construct an
of the solvable rigid Ue algebras, as soon as the dimension is no greater than 8. In fact,
we can determine the other rigid Ue algebras; the only obstacle we can meet comes
from the large number of nonlsomorphic rigid algebras which appear.
Notation: In the next list, the regular vector is noted X and the eigenvectors
associated.1D the eigenvalue ~ - i are noted YI ' Y'. , ....
Dimension 2
Dimension 4
Dimension 5
J1l(X,Y o)-O,
J1l (X , YI) - YI, J1l (Yo, yll) - yll , J1s1 (Y I, yll) - y 2 ,
J1l (X , yll) - yll , J1sl (y0 , Y2) - Y2 , J1sl (X, Y2) - 2Y2 ·
The variety of Ue algebras 241
Dimension 6
Dimension 7
1 1 X 1
117 - 115 1l2'
~f (X , y. 1) - yll 1 , ~f (Yo, y 2) - y 2,
~f (X , y 2) - 2Y 2 ,
~.f(x.y,)-3Y" ~.f (Y 1 , y 4) - y 5 •
~j (X, yl,) - 3yl"
~.f (X • y 4) - 4Y 4 ,
~.f (X , y 5) - 5Y 5 •
~t (X, y 5) - 5Y 5, ~t(YI,Y6)-Y7'
~t (X, Y6) - 6Y6,
~t (X , y 7) - 7Y 7 .
The variety of Lie algebras 243
We use here the notations of chapter 4, where we have given a description of the
space H2(n,n) in decomposing it in to the sum of the two subspaces H1(n,9/n) and
H2 fund(n,n).
Let us first study the integrability of certain infinitesimal defonnations which belong
to Hl(n,91n).
Lemma 2. Every lnflnlteslm.al deformation cp of the form dfml with CI\ En., i- 0, 1
(see chapter 4), is linearly Integrable (It means ~ + 'I' verifIes the Jacobi conditions,
where ~ Is the law of the Ue algebra n).
Yt - Xa - t.[X p,X_p] ,
Lemma 4. Let a, peS, with a + peA, Rp - 0 and let A(a,p) be a subset of A'
formed by the roots, which can be expressed by a and p. Then the subspace
m- L I}a
aeA'-A(a,~)
Proof. It is clear that the subset A' - A(a,p) of A' is closed with respect to the
addition of the roots. It is also clear that 't+ v e A' - A(a,p), If't e A', v e A' - A(a,p).
Then m is an ideal of n. Suppose that x.~ does not normalize the ideal m. Then there
is a root vector Xv so v - p e Ao u A(a,p). We remark that the case v - p e Ao
Is excluded, because R~ - 0. The case v - p e A(a,p) is also excluded,since
ve A'- A(a,p).
where aye C for 'YeA. We can easily show, that Q.t is a subalgebra of I}. Let
m- L I}a.
aeAI-A(a,~)
246 Chapter5
f(x,y) - ~ a.y. e y + y.
yeA(a,p)
It is an isomorphism and the law ~<lt verifies the Jacobi idenUUes ; so n<lt is a bnear
deformaUon of n.. This proves the lemma.
fw{XYI'X')'Z) - Xw(1I) ,
where ais the maximal root
(b) a + ~ t! d.
Suppose that a + ~ E d. It is easy to see that the endomorphism h of m defined by
D - th+ adX a
We have D E Der m. The semicIJrect product C. D ED m, defined by
[D,X] - D(X) - th(X) + ad Xa(X) ,
where X E mdefines a lie algebra defloted "'t. Consider an algebra structure, noted
",mt ,on the linear space underlying the '" associated to the product
Now suppose that : a + ~ t! d. It Is clear that at least one of the simple roots a, ~,for
Proposition 19. Let 11. be the nilradical of a Borel subalgebm of a simple Lie algebm 9
of lank> 1 (it means S1 - s, card(S) > 1). Then, there is a system of cocycles
(infinitesimal deformations)
{ <PI I i-I, ... , dim H2(11.,11.) }
with <PI 0 <PI - 0 such that the system formed by the cohomology classes of these
cocycles forms a basis of the space H2(11.,11.).
We can suppose that the condition a + 2~ ~ !!.. is fulfilled, because the element f 004 of
H 1(n,IJI11.) is cohomologous to f 00'4' where 0)'4 - (~, -a). Lemmas 2, 3, and 5 show
that all of these cocycles (infinitesimal deformations) are linearly integrable. From the
description of the space H2 fund(11.,11.) (see Chapter 4), every element z of this space is
expressed by
where 3w e «:. From Lemma 6, every cocycle fw verifies fwJw - o. This means that it is
linearly integrable. Then we can choose a basis of the space H2 fund(n.,11.) formed by
the linearly integrable cocycles. This gives the proposition.
Theorem 26. Let 11. the nilradical of a Borel subalgebm of 9. Then the tangent space
to the scheme Ln coincides to the tangent space to the corresponding reduced
scheme at the point 11..
isomorphic to the simple algebra so(r+ l,e) c 9, fonned by the antlsymetrlc matrix.
Then we obtain afamilyn t of the subalgebras of 9 verifying no - n, n t == soCr+ l,e) for
t #- o. This family gives us a defonnation of n such that its linear part Is z.
CHAPTER 6
It Is an algebraic subset of Ln. If we fix a basis of (Cn the laws of Ue algebras are
identified with their constants of structure which verify some polynomial relations
(see Chapter 5). The nilpotent laws whose nilindex is less or equal to P are also given
by polynomial relations. Then Nnpis aZariski closed subset of Ln. We alsodesignate
the corresponding affine schema by Nnp.
The nilindex of a n-dimensional nilpotent Ue algebra does not exceed n-l. Then
Nnn. l contains all the n-dimensional complex nilpotent Ue algebra laws. For
Simplicity, we note Nnn_l by Nn.
Variety of Nilpotent Lie Algebras 251
Let I} - (CD,/lO) be anllpotent tie algebra. We wantto detennlne the tangent space to
the point /lo END. Suppose that the nillndex of I} Is equal to p. We consider the
flltradon of I} given by the descending central sequence: FIJ - CI-ltj and the ftllradon
of I}, viewed as an adjolntl}-module, given by TIJ - Cp-I+11}. These fUtradons Induce
fUtradons on the spaces of cochains, cocycles, coboundarles and cohomology spaces
(we have studied all these spaces In Chapter 3). We follow the notadons of this
chapter. Let
J.LCt) - /lo + t<lll + t2~ + ...
be adefonnadon of/lo In the varletyNDq wlthq~p. We know that <Ill E Z2(/lo,/lJ. As
J.LCt) Is a law with a nillndex equal 10 q, we have
<Ill (X1,/lO(X2,/lO( .../lO(Xct b xq) ...) + /lo (x t.<IlI (X2,/lO(X3' /lo( ...Xq) ...»). .. )
+ ... + /lO(X1,/lO(X2,.../lO(Xq-2,<IlI(Xct 1, xq) ...») - O.
From this equality, we deduce the existence of a coboundary df E B2(J.1,~ such that
the cocycle '1'1 - <Ill - df verifies 'l'1(F11},FJIJ) c Fi+j+p-qIJ ,that Is <Ill E Fp-q Z2(/lo,/lO)
an this notadon I} Is the tie algebra associated to J.1o). Then, we have the following
proposidon
Remark. The space W Is the tangent space In J.1o to the schema NDq. If this schema Is
252 Chapter 6
reduced at the point ~ , then W coincides with the tangent space in J.In to the variety
Nnq .
In this section, we study some components of the varietyNn. Let j='n be the subset of
Nn of filiform Ue algebras. As f"n - Nn - Nnn-2' this subset is a Zariski open subset of
Nn. Each component of f"n determines a component of Nn.
J.l - ~ + V with V E Z2(J1o,J1o). As J.l verifies the Jacobi conditions, the cocycle VaSl
verifies these conditions, and V E {,II. We will denote by ~ the Ue algebra
corresponding to Il- J10 + vCthat is ~ - (Cn. ~ + V)).
Now consider the filiform Ue algebra Ln - (Cn+1, ~) definedin the basis (eO,e 1,... ,e.)
by
Lemma 1. Let 'I' - I ~,s'l'k,s be In FoHZC~,V with ~,s"# 0 for a palr Ck, s) saJislyJng
s - 2k+ 1 < n . Then the InfinItesimal de/onnatlon 'I' Is not Integrable.
Proof. A cocycle 'l'is integrable if and only if the 3-cocycle 'I' 0 'I' belongs to B3C~,V.
Let ko be the smallest integer satisfying the hypothesis of the lemma. The description
of the space FoHzCLn,Ln) gives us ko ;j!: 2. Let us determine the coefficient of
ez~zin the expression ('1'.'1') (el.elQ,.eko+ tl- From the choice of ko, this coefficient Is equal
From Lemma 1, the study of Nn+l around the point Ln Is brought back to problems
of the integrability of Infinitesimal deformation 'P . These cocycles are combinations
of the cocycles 'l'k,s with 2k+ 1 < s < n and 'l'r,n with n - 2r+ 1 Cif n Is odd). We denote
by X the linear space generated by these cocycles. As these cocycles are linearly
Independents, every 'I' E X Is written
Let M be the affine algebraiC variety In X defined by the relation 'I' 0 'I' - O. The
decomposition
with
254 Chapter 6
x, - H2,(Ln.Ln} If I ~ 1.
Xo - 0 It n Is even.
Xo - C('I1r,n) If n - 21'+-1 Is odd.
Let J10 be the law of Ln. A filiform law Il E :f'Il+l always admits an adap1ed basis (see
Chap1er 2) and then can be described as being of the form Il - J10 + ~ with ~ E
J.1(el.X2(J.1» - L (X1(J.1)XI(J.1) •
w
el(J.L,) - (X3(J.L,)eo - el •
el+ 1(J.L,) - (adJ.1(eo))i e 1(J.L,) .
Then the elements eo. e 1(J.L,). e2(J.L,) ..... en(J.L,) determine an adapted basis of the algebra.
J.1 and the mapping J.1 -+ (eO.e1(J.L,), ....en(J.L,)) is a rational mapping. Then, we have
J.1 0 J.1- q(J.L,) 0 (J.1o+~(J.L,)) where q(J.L,) e GLCCn+l) and ~(J.L,) eM. The mapping
J.1 -+ ~J.L,) is rational. Thus. ~U n C1) is an bred.uclble set conlalning C. This bnpUes
then
C 1 - GLCCn+1) 0 (J.1o + C)
In this section. we consider only the case n ~ 12. The study of the variety Nn for
n s: 11 will be presented. in section VII.
Ifn Is odd.
Ifn Is odd.
Variety of Nilpotent Lie Algebras 257
(4) M4 ' which is defined only if n is odd, is the closed set of M given by
Lemma 1. The closed subsets Ml , M2, M3 (and M 4,lf n 1s odd) are nontrivial subsets
of M, strictly contalned 1n M.
Proof. We put
'PI = 'If1,4 ,
(n-2)/2
q>z =L ak,2k+2 'lfk,2k+2 with al,4 - 1 ,
k-l
The existence of the filifonn algebras Rn and W n (Chapter 2) shows that the cocycles
'PI and q>z are linearly Integrable, i.e. 'PI 0 'PI - 0 and q>z 0 q>z - O. It is obvious that also
we have 'P3 0 'P3 - 0 and 'P4 0 'P4 - O. Then, 'Pi E MI for j - 1, 2, 3, 4. Moreover, 'Pi eo
Mj if 1"* J. This proves the lemma.
By writing that the coefficients of es, e lO ande u are zero, we obtain the follOwing
relations:
258 Chapter 6
The solution of this system is the union of the three straight-lines in the 4-dimensional
spaces parametrized by (a1,4' az,6' a,,8' a4,l~ whose equations are (t,O,O,o), (0,0,0,t) and
(t, 1~ '7~ '4~O) .We note that the cocycle '11- 1: ak,s 'IIk,s E FoH2(Ln,Ln) with a1,4 -
-az.6 - ~,8 - 0 and ~.10 "# 0 is not integrable if n > 11. For verifying this, it is sufficient
to consider the coefficient of e l2 in ('110'11) (e1,e4,eS)' This coefficient is equal to C3.j,lO)2
and is not equal to zero. But an element of B'(In,In) doesn't possess this property.
By successively using the following relations
('II.'II)(ebe40eS) - 0 ,
('II.'II)(ebe.50e6) - 0 ,
(a) if n is even, the vector (a1,4, a2,6, ... , an~2 , n) is equal to one of the following vectors :
(b) if n is odd, the vector (al,4, a2,6, ... ,ant, n-1 ,ant. n) is equal to one of the following:
Theorem 1. The variety :fm (and then also the variety Nm), with m ~ 12, contains at
least three Irreducible components If m Is odd, and at least four irreducible
components If m Is even.
Suppose n ~ 11. We denote UI as the ZariskJ open subset of Ml' defined by the
inequall.on a l ... '¢ 0 (we use the previous notadons). It Is clear that Rn e UI'
b) (cpo-3)a22,so
a3.2so-" - 2
a I."
260 Chapter 6
We suppose the converse and let", be a cocycle such that ~,s ~ 0, where
r - s - 2k - 1 < 280 - 11. We can suppose that r is the minimal value satlsfytngtbis
hypothesis and that k is the maximal value such that s - 2k - 1 - r. By writUng that the
coefficient of e 2k+l+r In ",o'!'Ce 1,ek_1,eIJ Is equal to 0, we obtain ~,4 ~,2k+l+r - O. As
~,4 ~ 0, then ~,2k+l+r - O. This is impossible. We deduce the case (a). To verlfy point
(b), It is sufficient to consider the relation ('1'0"') Ce 1,e2,e,) - 0 and to write that
the coefficient of e2s0-4 Is zero. To end the proof, we can show that ak,2k+1+r - 0 if
If k > 3 and r - 2 So - 11. Let k > 3 be and r - 2 So - 11. We consider the relation
<", 0"') (el'ek_1,eIJ - 0 • The previous assertion permits us to write ~,2k+l+r - O.
Proof. If n ~ 14, the lemma [s obvious. We suppose now that the Index s satisfies
s S n; 7 wi1h a2,s ~ 0 and denote by So the minimal value of such Index. The relation
Variety of Nilpotent Lie Algebras 261
('1/ 0 '1/) (e he"eS) - 0 gives, by writing that the coefficient of e3S0-7 is zero,
and
(- to+3) a2,so a"to + 2a1..~a4,3s0-8 = 0 if s - 7 .
From Lemmas 1 and 2, we have a2,so - O.
This depends of the inequalities So > n~ 7 and 3so - 7 < n which can be verified simul-
taneously if So = n~8 .
Lemma 4. We consider the arbitrary scalars 0.1,4. ai,S • .... a 1,n .0.2,50 ..... a2,n satisfyins
So - [n;lO] and 0.1,4 ¢ 0 . Then there is only one cocycJe 'I' - 1: ak,s'l'k,s of U 1 such that
a1,! - 0.1,1 anda2J - a2J·
Proof. We put
n n n
'1/ - L 0.105 'l/1,s + L
s-4 S-So
0.205 '1/205 + L
8-210-4
X,,s 'I/,,s +X4,n 'l/4,n
We choose the values x3,s in order that the condition ('1/0'1/) (el'e 2,e,) - 0 is verified By
writting that the coefficients of vectors e28o-4, e2so-3' __ ., en are zero. we obtain the va-
lues of the variables X"S, where s - 2s 0-4, ... , n. They have the desired form. In
particular, we have:
In other cases, we put X4,n - O. Lemmas 1, 2 and 3 show that the equalities
ey depend on the previous relations and of ('" 0 ",) (e 1,e2,e,) - O. The others relations
(",0 ",) (el,ej,e r ) can be verlfted directly. This gives the lemma.
Corollary 1. The variety U lis lneducible and Its dJmenslon Is equal to 2n - [n16] .
Corollary 2. The dimenslon of the tangent space at the point 0 to the variety U liS
equa/b3 (n - [Df]) lin EO or2 mod 3 and Is equal b3 (n- [n;7]) + 1 lin E 1 mod 3.
Theorem 2. Let n ~ 11. The point Rn of the variety Nn+l Is a simple pOint. The
dlmenslon of the JrreducJble component e of N
n+l
passlngthrough Rnlsequa/b
n 2 + 2n - n;16. Moreover, aJJ the points (Inlt, ole with '" E Ulareslmple points.
Variety of Nilpotent Lie Algebras 263
Proof. We put
"'1 = L lSisk
~ "'1,21+2.
where
k- 0-:/ . al - 1 • aj - (4 + f)aj+l • J - l .....k-l .
f!.eo) - eo •
f!.e~ - el + el+r ' i-I •...• n-r ;
f!.e~ - ej • j - n-r+ 1 •...• n .
We have g((L.t~\) - (L.t~ • where '" is a cocycle verifying the hypothesis of the previous
lemma. Now suppose that r - 2. In this case. it is sufficient to consider the linear
transformation h E G defined by the following relations
hCeo) - eo + el • hCel) - el •
hCe~ - [eo,hC~+~] + "'lCel.hCei-l))
Lemma 2. Let 2 S r S n-l0 and n 2: 12 and consider the following cocycle given by
(XI,ir) - - ~+ (r+;~:+6) ,
(XI,':r) - ~ ,
(XI,ir) - 0 ,
) 1 2 (r+2) + 6 (r-l)
(X2,tCr - 420 - (r+ 5)(r+6)(r+ 7)(r+8) , (r+3)(r+4)(r+5)(r+6)(r+7)(r+8) ,
(X ':r) _ 4 + r+6
2, - 35 (r+ 7) (r+8) ,
Variety of Nilpotent Lie Algebras 265
a2.{.r) .11
, 35
a3,l.r) • 0 ,
i ) 1 2 (r+3) + 6 (r+1)
a3, r • 420 - (r+6)(r+7)(r+8)(r+9) (r+5)(r+6)(r+7)(r+8)(r+9)'
a2J,,,,"fr) • 2....
35
We have
for all values r ~ 2. The verification of this assertion is reduced to the verification of the
nonexistence of integer roots greater than 2 of the foUowing polynomial:
The supremum of the positive roots of this polynomial is equal to 50. An easy
calculation shows that fer) *" 0 for all intergers r, 2 S r S 50. Then the solutions of this
system generate a l-dimensional vector space in the 4-dimension vector space. The
relaUons
permits us to compute "s, "6, ..., "m from the parameters xl' ":2, x" x4 • This means
that the space of the solutions of the homogeneous system
266 Chapter 6
is, at most, of dimension 1. The previous lemmas have shown that this dimension Is
exacdyone.
Lemma 3. Letn-9 S r S n-7, n ~ 11 and <PI the cocyc1edeflnedin Lemma 2. Then the
dimension of the space of solutions of the equaJIon <PI 0 X +X0 <PI - ObeJonging to
H2 r(~'~) is equal 10 2.
Proof. Let
x= L XI 'l'1,21+l+r E H2.{Ln,Ln)
lSISm
be a cocyc1e saUsfylng the equation <PI 0 x+ x0 <PI - 0 ( m - n-;-l) . In the case r - n-9,
this equation gives a system of 2 linear homogeneous equaUons at 4 indetennlnates
(see the proof of Lemma 2). In the cases r - n-8 and r - n-7, we obtain an
homogeneous linear equation at , indeterminates. In all cases, the dimension of the
space of the soluuons Is equal to 2. Moreover, we can take XI and ~ as free
parameters. This gives the lemma.
Let n-9 S; r S; n-7 and n ~ 11. We denote by CPr and <p'r the cocycles of H2 r(~'~)
satisfying the equation <PI 0 x + x 0 <PI - 0 and such that
Lemma S. Suppose that the hypothesis of Lemma 3 is satisfied and let 1r+1 be a
cocycJe of Z'r+1CLn,Ln). Then there is a unique solution of the equation
CP1 0 X + X 0 CP1 - 1r+1 belonging to C2.-CLn,Lr) and such that xCe1,e2) - xCe2,e3) - o.
See Lemma 3.
Let 'I' - '1'1 + '1'2 + ... + 'I' n-3 be a cocycle belonging to u 2 with '1'1 E H21 (Ln,Ln) .
Lemmas 2, 3,4, and5 show that the cocycle 'I' can be uniquely written 'I' = '1'0 + '1'-,
"here
and C'II-)r is the unique solution satisfying Lemma 4 or 5 if n-9 S r S n-7 with
1r+1 = L
I+j- r+1
('11I.'I'j+'IIj.'I'I).
The cocycle '1'0 is called the homogeneous part of the cocycle '1'. It is clear that every
cocycle is determined by Its homogeneous part, i.e., if we have a cocycle '1'0 which is a
linear combination of the cocycles
268 Chapter 6
and if the coefficient of <PI is different of zero, then there is one and only one element
'II of U2 whose homogeneous part is '110.
This reasoning shows that the variety U2 can be defined by the follOwing relauons :
n-2
3.t,2I+r - (X1,l' al,3+r + 'Y1,l' , i - 2, ... , 2 ,IS r S n-l0 ;
Here the elements 'YI~ , 'Y3,r , 'Y4~ are polynomials whose variables are
These polynomials have a degree greater than 2. The linear combination of cocydes
<PI ,'P2 , ... , 'Pn-3 • !p'n-5 ' .u, !p'n-9 generates the tangent space at the point Ln to the variety
U2 • As La E G{AJ for every filifonn algebra A E ~+l • we have the follOwing theorem:
Theorem 3. Let n;:o; 11. There exists a unique irreducible component of the variety
Nn+l containing W n. This component is smooth (each point is a simple point) and
its dimension is equal to n 2 + n + 2.
Theorem 4. Let A+0 - Ao n A+ and let p be the halfsum ofpositive roots lying in A+0
Then
(a) If 9 - Aror Cr andlfS 1 - {ru, where aisaslngularroot, we have
dim G(n) - r2 + n + dim Zen) - dim I} •
(b) In other cases,
where ais the maximal root and sa the reflexion (symmetry) with respect to a.
Proof. We define a homomorphism
f:p ~Dern
by putting fCx) - ad x In. This mapping is surjective in case (a) from the description
of Der 11. (see Chapter n. Now assertion (a) results in the equality
dim p - dim I} - dim 11. and Ker f - Zen). For case (b), we put
Ip -{adx/n , x E p}.
Dern-~e(e Wa).
aes,.
The centralizer of the Ue algebra n In p coincides with the center Zen). Then we have
The dimension of the module Wa can be calculated. using the Weyl formula [SE1] :
dim Wa - fl (1 + (Sa(a)-a;y») .
+ (p,y)
yeAo
The corollary can be proved as this: consider the Inclusion of Hk In the a1gebral} -
Cr> where r - k + 1 (n - Hk if S1 - (al, where a is a singular root). From the theorem
under consideration, we have :
Variety of Nilpotent Lie Algebras 271
Theorem 5. Let n be the nilradicaJ ofa parabollc subaJgebra ofa complex simple Ue
algebra, p be the nilindex of n, and n - dim n, w.lth p S n-3. Then the Zariskl
tangent space at the point n to the schema LD coincides w.lth the Zariskl tangent
space at the same point n to the subschema N Dp+2.
Remarks.
1. For some Ue algebras n, we can accurately state this theorem. For example, If
n - Hk Is the Heisenberg algebra with k > I, then
and the tangent space LD colnddes with the tangent space to NDP+l at the pOint n.
If2p + 1 S n, then the tangentspace to (,n coincides w.lth the tangentspace to ND2p at
thepoint n.
272 Chapter 6
This result is deduced dJrectly from the relation F-pH2(n,n) - H2(n,n) (see Chapter 3).
where
W 2 being the subset constituted of the elements CO of the Weyl group W such that
coer) (').1+ cA' and card(CO(A-) (') .1+) =2, P being the half sum of positive roots,
PI the halfsum of the roots of .1+0 , and Qpls the last Idea/of the descendingcenllal
sequence of n (Qp - Cp-1n).
where x, yen., X E End 11.. This impbes that the image of (df)Mo coincides with the
space constituted from the cocycles <p E 2 2(11.,11.) whose cohomologic class Is in
FoH2(11.,11.). This space Is the tangent space to the schema Nnp at the point 11.. As
the Image of (df)M1 belongs to the tangent space T(NRp) of the variety N np at the
point 11., we have
where (Nnp)red is the reduced schema. Moreover, the mapping (df)Mo Is surjective.
Let M - (g1' 'II) be a pOint of the variety G x FoH2(11.,11.) sufficiendynear to the point
Od,O) for the metric topology. It Is clear that
Moreover, we have
dim (Im(~)S dim TOO (N~)S dim T{(Mol (N~) - dim (Im(dfMo) .
Then we have
then is stahle in a neighborhood of the point Mo. In this neighborhood, the image of
the mapping f is a smooth subvariety whose dimension is equal to the dimension of
the tangent space of Nnp at the point f(M o). As the dimension of the variety Nn p at a
point M is less than or equal to the dimension of the tangent space at the same point,
and as this dimension is stahle in a neighborhood of Mo , we have
) - dim Tn (Nn)
dim (Nn pn p.
This means that the point n. of the variety Nnp is simple. Then, as the scheme Nnpis
reduced, it is smooth at the point n.. To determinate the dimension of the variety Nnp
, we note that
(see Chapter 4). The description of the space H2(n..9) and the formula for the
dimension of G(n.) imply the required result.
The subset of filiform laws is an open set (irredUcible or not) in N°. From the
previous paragraph, we have:
If n - 7, the variety N 7 is union component such that one of them contains the open
set of fillfonn laws. In particular, this shows lhat this open set is Irreducible.
The proof is supported by the results which follows. Let (Xl' X2, ... , X~ be a basis of
!CO and J.1 the n-dimensionaille algebra law defined by :
Proof. Let us consider a nilpotent perturbation J.1' of J.1 (see Chapter 5). As the
characteristic sequence of J.1is (n-2, 1, n, Its J.1' Is equal to (0-2, 1, 1) or (0-1, 1). In this
last case, J.1' Is filifonn. We want to prove that J.1' is flUfonn. It is sufflclentto prove
Variety of Nilpotent Lie Algebras 277
that If Its center Is one-dimenslonal. then J1' Is not fllifonn. Let us suppose that the
center of J1' Is one-d1rnenslonal. Consider a basis of this center given by a vector Y,
with Y, =X, . The derived subalgebra of J1 Is generated by (~-1'X,. X2). As this
...•
subalgebra Is also of dimension n-2, we can find a basis CY0-1' ... , Y" Y2) for this
satisfying YI = XI. As the vector X1 Is a characteristic vector for J1, there exists a
=
characteristic vector Y1 for J1' such that Y1 Xl and J1'CYl'Y~ -YI-4 for I ~4. This gives
J1' (Y 1, Y,) - ! elYI andJl.' (Y 1. Y2) - ! ~YI
I.? -
with al =0 and el =0. The vector
space generated by the vector Y, Is the center of J1. This gives £Ie, =0 for 1*"3 and
J1' (Y2.Y,) is Infinitely large. Then J1 is not filiform and the component does not cut
the open set of these filiform laws.
Let 9 0 be the n-dimensional complex Ue algebra whose law J1 is defined in the basis
(eo' ...• em_I' fl' ...• fk' g1' ...• gk' h1' ...• hkl. where n - 3k + m by :
This Ue algebra is adirect sum of the model filiform Ue algebra Ln and of k patterns
of the 3-dimenslonal Heisenberg algebra.
'IIr,s,t (f~s) - h t
llr,s,t (fr,fs) - h t
for 1 S r, s, t S k and r ¢ s ¢ t
~r,s,t (g~s) - h t
for 1 S r, s, t S k and r ¢ s ¢ t
pt,u(eo.ftl - hu , p't,u(eo,gtl - hu
Indeed, if J.L + 'IIis in the orbit of J.L, then'll is acoboundary, I.e. 'II =lif with f E gl(n).
As, no cocycle of n is a coboundary, the laws J.L + 'II are not isomorphic to J.L.
Now, let Co be an algebraic component passing through 9 n and containing the family
Proposition 4. We have :
Indeed, dim n - MCk, m). From the previous lemma, this number Is a minimum of
the number of nonorbltal parameters of Co.
In effect, consider Y E 9, such that ~X"Y) -:j:. 0, where JL Is the law of 9. Then
&P(X 1,X2 ,Y) -:j:. o.
We know that If Im Is the model filifonn algebra, we have dim 22CIm,Im) :s; 2m2 -
2m. By looking at the structure of 9n' we see that exists an Ideal 1m Isomorphic to Im.
On the other hand, the dimension of the components passing through 9n Is smaller
than dim 22C9n,9~. Now Lemma 3 Implies:
Proof of the theorem 9. From Propositions 4 and 5, we can give the boundaries of
the dimensions of the component Co. Suppose now that k ~ m. Then, for n
suffidendy large, we have :
280 Chapter 6
If k varies between n/3 and n/4, we reveal a minoration of the number of (not
filiform) components of Nn. This boundary is of the order n/74. This gives the
theorem.
CHAPTER 7
CHARACTERISTICALLY NILPOTENT
LIE ALGEBRAS
Then 'I' - 'l'r + 'l'rH + ... + '1'1 with '1'1 E HI~In,In) , 'l'r ~ 0 and r ~ 1. We have called. the
component 'l'r the sill coqcIe of '1'. It verifies 'l'r. 'l'r - 0 Then
A large part of the proof of this theorem is based on the follOwing lemmas :
First, we prove some of the lemmas Intervening In the proof of this theorem.
l
91,el -
) { el, If 1 - 0 ,
o , If lSISn •
} 92 (el ) - { eo • If 1- 0 •
Cl+r)el • if lSiSn • }
Proof. Leta E Vk, bE Vm • Onehas
283
l~l ~l
[a.
do([a,b]) -[do(a},b] -[a,do(b}] - [1: d1(a},b] + 1: d 1(b}]-1: d1([a,b]) +
. l~l
The left part of this equality belongs to the space Vk+rn as soon as the right part
belongs to the space ED V t . This means that do E Der Ln . The descrtpUon of the Ue
t>k+m
algebra Der In (see Chapter 4) permits us to write
By comparing the coefficients to e2k+l+r in 1he two parts of this idendty, we see 1hat
A.l - (r+ I) ~ and therefore do - 'AtiJ2 + A-,6 1 • Q.E.D.
Lemma 3. Let A' - (LrJ'Vr be the sill algebra of the Ue algebra A - (Ln)'V . Then we
have d 'OE Der A'
Proof. We have
d'O ([a,b] + vr(a,b»- [dlO (a),b] - Vr(d I O(a),b) -[a,dlo (b)] - Vr (a,dlo (b» -
Consider this Identity for a - es ' s ~ 1 and b - em , m ~ 1. Then the left hand Is In
V s+m+n as soon as the light hand belongs to E9 V s+m+r+ 1 . This Implies that the left part
1~1
Is equal to zero. Now consider this Identity for a - eo and b - em . Then we have
vrCa,b) - 0 . As d' 0 E Der In, also In this case, the left part Is equal to zero. This proves
that dlo E Der AI •
Lemma 5. Let A - CIn)", be a filiform Ue algebra with 'I' E F1H2(In,In) and Jet
d - Ad' 0 + d 1 + ... + dy. be a derlvation of A with A ~ 0 ( we use the notations 01
Lemma 4). Then, there Is a basis fo' f l' ... , fn of the space V such that
[fo, f l] - f l+1 , 1 ~I ~n-l ;
Proof. We have
We put fo - e o+ aIel + ... +~en and we choose the scalars a l •..., ~ for having
d(f~ - Afo . This Is possible because the numbers r and A are different to O. By
putting
Lemma 6. Suppose that the hypotheses of Lemma 5 are satisfied. Then there Is a basis
Zo, Zl' ..., Zn of the space V such that
Zoo - fo'
Zl - f1 + a1f2 + ... + a n-1f n ,
-----------,
It Is easy to choose aI' ..., ao-l as such, as the basis Cz:t) satisfies the required
conditions.
Proof of the theorem. Let A - (10).'1' be a filiform tie algebra with 'I' E F1H2c1o,1o)
and consider A' - (Ln>'I'r Its sill algebra. Suppose that A' Is Isomorphic to A There is
286 Chapter 7
This derivation Is no nilpotent. Then A' and thus A, Is not characteristically nilpotent.
As A - (10)'11 Is not characteristically nilpotent, there Is a derivation d of A which is
written d - A.d.' 0 + d 1 + ... + du (notations of Lemma 4). We choose a basis Zo, 2:1, ... , Zn
satisfying the conclusion of Lemma 6. In this basis, the Ue algebra A Is written ~
(10)'11' I.e. gCA) - (1o)'V' ' where g is In G UV) and Is defined by zt - gCel ) , I - 0, 1, ..., n).
The cocycle'll' also belongs to F1H2C1o,1o) and Its sill cocycle is the same as the sill
cocyde 'IIr of'll. Show that 'II' - 'IIr. If It Is not the case, we have
Hs2(Ln,Ln) Ii S2«L n)'IIr ,(LJ'IIr) ,then this cocyc1e Is unique up a nontrivial fa.ctDr.
Proof. Let
CharacterisIica Njpotsnt Lie Algebras 287
where (ev is the given basis of V (Section D, because this basis satisfies
As q>"# 2r, we have f(ell - Ael+s+r ,IS is n-S+r and ",does not depend of the index i.
The lemma is proved.
Lemma 2. Let n ;;:: 7 andJet U a open serin the variety Nn with A E U Then there
always there exists a characteristically nilpotent Ue algebra belonging to U.
combination of cocycles 't and 'l'n-6 where 't Is a cocycle linearly Integrable belonging
to F IH2(Ln, Ln), Is also linearly integrable. Then every tie algebra of the fonn (L n )1I/(t)
with'i'Ct) - 'l'r + t'l'n-6 , t :# 0 , Is not isomorphic to the sill algebra (Ln)Vr .
From Theorem 1, the tie algebra (Ln¥l) Is characteristically nilpotent. By considering the
nonnull values of t, when t tends to 0, t :# 0, we obtain characteristically nilpotent tie
algebras belonging to a full neighborhood of A.
(3)We consider a cocycle q> E H2n-5(Ln,Ln) and not cohomologous to zero, q> being
considered as a cocycle of the tie algebra (Ln)Vr (this Is possible from Lemma D.
Then the tie algebra (~lvF'P Is characteristically nilpotent for all values of t, t :# o.
(4) We can always flnd a cocycIe <pof H 2n_2(L n,Ln) which does not belong 10 S2«Ln)", ,(Ln)",).
For example, If r > n-5 and 'l'r - a'l'I,o-2 + ~'I'2,o, a :# 0, then we take cp - 'l'2,n-1. If r - n-5
with a - 0 , then we take cp - '1'1,0-3. The tie algebra (Ln)VFtCP Is characteristically nil-
potent for all values of t different to O. Q.E.D.
Proof of Theorem 2. Recall that an external torus of the derivation of a tie algebra L is
an Abelian subalgebra of Der L whose elements are semlsimple. We know that there is
only a finite number of conjugation classes of maximal torus for a nilpotent tie algebra
L (see Chapter O. Let Tj , 0 $; I $; k, be representatives of these conjugation classes,
. the nu11 torus. Conslde r the closed subset NTIn+1 of N
To bemg n+1 whose elements are
n+1
he T I-invariant laws ; the union of the orbits G Ln+ 1(<<:) x NT I for 1 $; i $; k Is a construe·
tIble subset, i.e. a finite union of locally closed subsets. It Is complementary to a
subsetof Nn+ 1 fonned by characteristically nilpotent laws. Then this last subset is also
289
a constructible subset of ND+l. Now if the theorem is not true, there is a nonempty
open set of C formed by non characteristically nilpotent tie algebras. From the lemma
2, this is impossible. Q.E.D.
Let 9 be a simple Ue algebra such that its type is different to A I' A 2, A 3, A 4, A 5, B2, B3,
C3, C4, D4, and G 2 . Let n be the radical of a Borel subalgebra of 9 (we use here the
notations of Chapters 4 and 6). Recall that, in this case, the spaces FoH2(n,n) and
FIH~n, n) coincide. The follOwing cocycles fa ,[3 , (ex,~) E E ,defined by
have
In the next Chapter (as In the previous chapter), we wiU Identify the cocycles with
their cohomology classes.
Let <p E FoH 2 (n.,n) . It Is easy to verify that <p 0 <p - 0, I.e. the cocycle <p Is linearly
Integrable. The IJe algebra. defined In the underlying space to n by the bracket
Is denoted by nIP'
Let <p- 1: 'A.rJro bean element of FoH2(n,n) such that 'A.ro:;l!:O forall OlE E. Then
IDEE
p-l
g-d-do -1:
1-1
dl
Then
First, consider the case '} of type Ar ( r > 5). For the pairs
(a,b) - (eOlt , ellk+OIt+l) and (a,b) - (eOlt+l , eOlt+OIt+l)
1 < k < n-l, we have :
291
Ci) The projections of the elements [a,gCb)] and [gCa), bl on 91i' where a is a maximal
root, are null because a - <lk. and a - ak+l are notin A.
Cit) From the description of FoH2(n,n), we have q(a, b) - !..eli for some A E ce. Thus
gCep Ca),b)) - 0, because g - d - do - d 1+ ... + dp-l.
afO ep(g(a),b) - 0, because g(a) and be [n,n] .
By writing that the coefficient of eli is in the left part of the above identity, we obtain
Then ak(h) - ak+l (h) for all 1 < k < 1-1 , because Am * o. Moreover, we have
Now consider the pairs (a,b) - (e lll ,ea,) and (a,b) - (ea,.2 ,ear) .
The same reasoning gives:
This implies
For other Ue algebras 9 of a type different to Art we can show, uSlngthe same
arguments, that nip Is chara.cterisdcally nilpotent We will restrict ourselves to the case
9 - E6·
This gives
S(h) - 2al (h) - a, (h) - S(h) - al (h) - 2a, (h) - S(h) - 2a, (h) - a4 (h) -
- S(h) - 2a4 (h) - as (h) - S(h) - 2as (h) - a6 (h) - S(h) - al (h) - 2a2 (h) - 0 ,
11rus
Moreover, the descripdon of the cocycles of the space FoH2(n,n) are, In these cases,
different to the descripdons of the other cases (see Chapter 1). However, we can
extend the results of Theorem 3 by studying each one of these pardcul3.r cases.
Characteristica Njpotent Lie AIgebtas 293
and <Jl:z given in Table 4, and characterized by the fact that J.I. + t CPl does not satisfy
the Jacobi conditions (J.I. is the law of n). Then the Ue algebra m whose BW 1: is the
defonnation 1: - J.I. + tCPl +t2CP2 with t:;t 0, is characteristically nilpotent.
Remark. If 9 is a tie algebra of type AI' ~, ~, 8 2, G2, then we have FoH 2(n,n) - O.
Tab1e4
CJlI
9 CPl ~
i~
alvl + l: blllj
1-1
ea, /\ e ll2 ~ alb 7
C4 . e l>-Il1-1l2 0
N 1l1+1l2 , &-CX r CX2
al,b 7 :;t0
294 Chapter 7
-- r2+r-2
dim G(F) - n 2 - n - 2r + 1 +--::--
2
(Recall that G(F) is the orbit of F with respect to the action of GL(n)).
CHAPTER 8
APPLICATIONS TO DIFFERENTIAL
GEOMETRY:
THE NILMANIFOLDS
Introduction
A tie group is, roughly speaking, an analytic manifold with a group structure such that
the group operations
(x,y) ~ Xy-l
are analytic.
If G is a tie group, then the connected component passing through the identity of G is
a connected tie subgroup of G.
We note by La: x ~ ax the left translation by a in G and Ra the right translation bya.
A vector field X on G is called left invariant if it satisfies
(LSXx - Xax ,
where La- is the Jacobian (or the tangent mapping) of f. As
X,YE 9.
exp:9 ~G
297
of the Ue algebra 9 of G In the Ue group G sends straight hnes throught the origin In 9
Into one-parameter subgroups of G. This mapping Is defined as this: let X E 9 .
There exists a unique analytic homomorphism 9 of IIlnlo G such that 9(0) - X. We put
exp X - 9(1). So we have exp t X - 9Ct) andlt Is a one-parameter subgroup of G. We
have the formula
Now let x - exp X and y - exp Y be two arbitrary elements of the Ue group G.
where X and Y are In the Ue algebra 9. The campbell-Hausdorff series:
where the summation extends over all nonnegative Integers PI • q • i-I ..... m • m - 1.
2.... and PI + q':t: 0 (exceptm - 0 and 1 which have been written explicitly)deflnes the
multlphcation In a neighborhood of the groug G. Consequently. an arbitrary Ue algebra.
9 (in finite dimension) determines a unique Ue grouP. up to local isomorphism. such
that its Ue algebra is 9.
298 Chapter 8
Example. Let 9 be the one-dimensional (Abelian) Ue algebra. Then 1"X9) contains the
additive group of reals (it is the universal covering) and the multiplicative group of
complex numbers (group of rotations of a circles).
The Ue algebra 9 of a tie group G has been defined as the vector space of left invariant
vector fields on G. The dual space rf can be identified as the left invariant pfaffian
fonns on G. Let ol be a differential fonn on G. It is left invariant if
where X, Y are in 9 and ol e 9 . We can also extend this differentiation for every p-Ieft
invariant fonn on G (i.e. for the elements of AI1J-).
The NIman1ok:Js 299
H~9) - 9 '
H 1(9) _ 2 1(9) ,
B1(9)
Often this cohomology is called the cohomology of left invariant forms on the Ile
group G. In general, this cohomology doesn't coincides with the De Rham
cohomology of G. But, if G is a compact Ile group, then these cohomologies
coincide. The nilpotent case is examined in the next section.
Remark. The Maurer Cartan equations. We can read the constants of the structure of
a Ue algebra 9 by writing the Maurer Cartan equations on 9-.
Let (X 1 ,... , ~) be a basis of 9 such that
Consider (001 , ••• , COr) as the dual basis of 9-. Then, we have
and these equations again givethe constants of the structure of 9. Note that jacobi's
identities correspond to d(d~) - 0 k - 1 ,..., n .
300 Chapter 8
D. THE NILMANIFOLDS
Usually, the research of unlfonn subgroups passes through the research of lattice, i.e
discrete, subgroups such that G/r carries a finite Invariant measure. If every discrete
unlfonn subgroup is a lattice, the converse is not usually true (see, for example,
"Discrete subgroups of Ue groups" by Raghunatan [RAGD.
connected and simply connected, such a representation always exists. It can be also
supposed free. If risaclosed subgroup of G, then G/r is compact if and onlyiffor
any representation p : G ~ GLCn, lID, pCG) andpCD have the same closure for the
Zarlski topology, in the algebraic group GLCn , lID.
Theorem 2. Let G a simply connected nilpotent Ue group and letIJ be its Ue algebra.
Then G admits a discrete unifonn subgroup if and only if IJ is a rational Ue algebra, i.e.
9 admits a basis with respect to which the constants of structure are rational.
From the classifications of nilpotent Ue algebras of dimension less than seven, we can
deduce the following result
Proposition 1. Every complex (and reaD nilpotent Ue algebra of dimension less than 6
is rational. Every complex nilpotent nonrational Ue algebra of dimension 7 belongs to
a parametrized family of nonisomorphic Ue algebras.
302 Chapter 8
We have identified the elements of 9- with the left invariant forms on a Ue group G
whose Ue algebra is 9. Then we can see how, within this framework, the classical
differential invariants be have themselves. The most important of them is probably
the Cartan class of a pfaffian form.
The class (or CartaIl class) of ro is the codimension in 9 of the vector space
Ker ronh.ro
dro eX I y) - - ro rx I yl.
303
Lemma. If the class of c.o Is odd, there Is a basis (c.o - c.o1, ID.2, ••• , COn) of '.t such that
dc.o - ~ 1\ co, + ... + ~ 1\ ~p+1'
If the class ofc.o Is even and equal to 2p, there Is a basis Cc.o- c.ol' ..., COn) of 9- such that
This lemma is a direct consequence of the classification of the alternated bilinear form
on a real or complex vector space. From this lemma, we can easilysee that the dual
space (Ker c.o f"I h oo )- is generated by (ID.2P+2' ..., COn) if the class is 2p+ 1, or by
(ID.2P+1' ..., COn) if the class is 2p.
Proof of the proposition. Let (c.o - c.ol , ... , COn) be a basis of 9 reducing to the canonical
form dc.o(Lemma 3). Suppose, first, that c.o E Z(9)-. If(X 1, X2 ' .••, Xn) is the basis of9
such that (c.o1, ... , COn) is the dual basis, then Xl E Z(9) and [X, Yl - 0, V Y . Thus,
c1Co)) - 2p implies dc.a::x, X2) - 1, i.e. c.o [X , X21-:t: 0 . This is impossible andclCc.o) =
2p+ 1. Suppose now that c.a::X) - 0 , V X E Z(I}), the matrix of the operator ad X I is
nilpotent and nontrivial (we suppose c.o -:t: 0). Jordan's reduction of this operator shows
that the equation [Xl' Yl - Xl + U ,where U is independent of Xl' has no solution.
Then dc.a::XI ' Y) - 0 for all Yand c.o has an odd class.
This denies the fact that (dro)P Is a volume fOnD on 1t(H00) n G/r but explains the
previous proposition.
Proof. Suppose that dim Z(9) ~ 2. Then there are two independent vectors Xl andX 2
such that [~Yl- o. ';f Y E 9 • j - 1 and2 . If 00 E 9·. then I(xpdoo - 0 for j - 1 and2
andd(ro) S; 2p.
Example. The Heisenberg algebra lip Is a contact Ue algebra. Indeed. contacts of the
structure of lip are given by
1- 1..... P.
Proof. Let 9 be a contact tie algebra. Let (Xl' ... ,X 2p+ l) be a basts of 9 such that the
dualbasls (001' ••• , CO:2p+l) satisfies
The limit-point 9 00 ' which is a contraction of 9 , is the Heisenberg algebra.. Then every
contact tie algebra. can be contracted on the Heisenberg algebra. This proves the first
part of the theorem. As the existence of a contact form is an "open" property, every
perturbation of the Heisenberg algebra. also is a contact tie algebra.. Then we can
306 ChapterS
Frst, it is Important not to blend a Ue algebra with a symplectic structure and the
symplectic algebra spCn).
m.3.1 Definitiui'i
(1) d9 - 0,
(2) 9P ¢O, where dim 9 - 2p .
Example. If there is alinearfonn 0>0n9 of maximal class equal to 2p, then 9 - drolsa
symplectic fonn.
A Ue algebra provided with such a symplectic fonn Is called Frobeniusian.
Indeed, the class of a linear fonn on a nilpotent tie algebra is always odd.
- dJmension 4 (we describe the constants of structure from the Maurer Cartan
equations)
dOlJ. - 0 i - 2, 3, 4
dco, - co 1 A CO2
dC04 - CO 1 A C03
- dJmension 6
dC04 - CO 1 A CO,
dC04 - CO 1 A CO,
"'6,5
dro, - COl A CO2
dros - CO 1 A C04
"'6,6
d~ - C01ACO:z
dOlt - C01A~
dCOs- C02A~
"'6,7
dco, - CO 1 A CO2
"'6,8
dro, - COl A CO2
dro4 - CO 1 A CO,
"'6,9
dro4 - CO 1 A CO2
ns.l ell
dro, - co I A 0)2
dol" - 0) I A 0),
"5,2 ffi m
d0>3 - 0>1/\ 0>2
The nol1on of double extension has been defined by Medina and Revoy [M-Rl for the
312 Chapter 8
construction of tie groups provided with invariant metrlcs and also with a symplectic
structure.
X, Ye 91 .
This mapping «1>1 corresponds to a 2-cocycle for the scalar cohomology of the left
symmetric algebra 9 51' whose product satisfies
(x . y) . Z - x.(y . Z) - (y . x) . Z - Y.(X . Z)
and the Ue algebra structure 91 is related with the algebra structure on 9 51 by
[x , y] - X . Y - Y . X
(91 and915 have the same underlying vector space). Consider the adjoint operator f"l
Offl with respect to 0)1 :
We can find a 2-cocycle «I> of the tie algebra 91 for the cohomology W(91' 91) by
putting
X, Ye 91.
Let be 1.1 - !lW E9 91 the central extension of 91 by !lW (~ is the scalar field of 91)
9- ~V €a 1.1
[V, ul- 0
[V , xl- - oo'(W , X)u - f 1(X) , X E 91'
W-V2.
Every nilpotent symplectic lie algebra can be obtained this way from the trivial tie
algebra of dimension O.
is an isometry.
The metric g defines a scalar product, also noted g, on the lie algebra 9. We can find a
314 ChapterS
Infinitesimal version
( oIx1 yz 1
001
Now, we study the IJe algebra of Killing veclor fields for the metric gA:
Let
a~l _0,
ax
~
A;2X."l + ( 1 + A;2X2 )~2
- - Ax -~3 - 0 ,
ay ay
Then, by integrating:
~ (
." - - x + 3
'l2~2)~1
az - 1\."2
"I
ay + 9(y ,z)
X2 ~1
.
Wededuce:
2
d ~1 _ 0 and dIv _ d9 _ 0 .
ayaz az az
dIv
- - 0; --~1.
de
ay ay
Then
~1- ay+b,
'If- d ,
The NImanifokJs 317
~2--ax+d
~3 - - M. X2 + Aa y2 + Aby + c
2 2
The tie algebra of infinitesimal isometries is generated by the following 4 vector fields :
a
X- -+ Ay-
a Y--
a Z--,
a
ax az dy az
Note that the fields ex, Y ,Z) form basis of IJ - Hl and correspond to the left
translations.
Let X and Y two vector fields on G. We define V x Y as the unique vector field on G
satisfying
where(el ,... , en) is an orthononnal frame of G for the metric g. They verify:
If (ro1 ,.... ron) is the dual basis. we define the symbol :rJ kI by: •
rol- 1:.
k
:rJ kI cJt •
The Riemann curvature tensor of the connection V associates to each pair of vector
fields X and Y the linear transfonnation
Remark. If g is a left invariant metric on G, then the first Bianchi identity corresponds
to the Jacobi identities on the tie algebra CJ. Indeed, these relations, written on a
othonormal left invariant basis (~, are relations concerning the constants of the
structure of g related to ~.
Finally, recall the expressions of the curvature forms and of the Ricci tensor. The
curvature forms of G are the 2-forms
They satisfy
Z -+ R(X , Y) Z .
Let g be a left invariant metric on G, CJ the tie algebra of G, {el} an orthonormal basis
of CJ, and Cl~ the corresponding constants of structure. Then
320 ChapterS
where TrC designates the trace of the operator C and where Cl is the operator whose
matrtx in the basis {ell is (C~yk,j.
n
pg - _lTrCI CJ -ITr ICI CJ+ 1
2 2
1: (clkr d kr )
4 k,r-l
n
and the scalar curvature 't - 1: PH
1-1
is given by
n n
't- _11: T r Cp C p _1 1: Tr ICpC p .
2 p-l 4 p-l
Theorem 5. Suppose that the Ue algebra 9 is nilpotent and not Abelian. For every left
invadant metric, there exist a direction X such that p(X, X) > 0 and a direction Y
such that p(Y, Y) < 0 .
Indeed, let X oF- 0 be in Z(9) (') [9,9] . Then p(X) > O. For finding a vector Y giving a
negative curvature, we can take a vector orthogonal to [9 ,9] but not belonging to Z(9).
Such a vector exists because 9 is nilpotent
The determination of the Ue group of isometries for a left invariant metric is easy
enough. The computations result in a theorem of Wolf generalized by Wilson [WI]. Let
{ell be an orthonormal basis of 9. For every derivation f of 9, we consider the matrix
322 ChapterS
MCt) with respect to the basis {el}' Then f corresponds to an infinitesimal isometry if
MCf) is an orthogonal matrix.
Wilson's theorem affirms that every isometry for a left invariant metric g on the
nilpotent group G is in AutC9) the group of automorphisms of 9. This implies that the
infinitesimal isometries are derivations f of 9 such that a matrix of f related to an
orthonormal basis is orthogonal.
In the following section, we compute this group for the left invariantmetrics on the
Heisenberg group.
Another interesting example consists in considering a nilpotent tie group such that
every derivation is orthogonal with respect to a given basis of 9. A partial answeris
given by considering the quatemlonian Heisenberg algebra. This Ue algebra is given by
where (001, ~, ~ , (Xl ' ~ , ~ , (X4) Is the dual basis of .~t. Consider the corresponding
basis (X 1 , X2 , X3 ' Y 1 , Y 2 , Y3) In 9 and the metric
2 2 __ 2 2 2 2 2
g- 00 1 + 00 2 , 013 + (X I + (X 2 + (X 3 + (X 4 .
Let f be a derivation of 9. The matrix of f related to the basis (X 1 ' X2 ' X3 ' Y 1 ' Y2 ' Y3'
Y4) is
(X -a -b
a 2(X -c
b -c 2(X
0 0 0 (X -d -e -f
0 0 0 d (X -h -i
0 0 0 e h (X -j
0 0 0 f j (X
The NimanifoIds 323
o -a -b
a 0 -c
b -c 0
0 0 0 0 ~ -e -
0 0 0 d 0 -h -i
0 0 0 e h 0 -j
0 0 0 f 0
1 Xl Xp Z
o o o YI
o 0 0 1 YP
o 0 0 o 1
There is a basis (Xl' ... , X p ' Yl' ... , Yp ' Z) of the Heisenberg algebra ~ whose
corresponding left invariant vector fields are
324 ChapterS
d
Z--,
dz
where (xt, YI' z) are the global coordinates of Kp defined above. These fields define the
structure of the tie algebra lfp because they verify
The left invariantforrn (a.1' ... , «po ~1' ..., ~P' m) which define the dual basis of ~ are
writen
We have seen this In the previous chapter but from an other point of view. We again
give the proof using here the matricial representaUon~
Let f be a derivaUon of~. A matrix of f with respect to the basis (XI' VI' Z) has the
following fonn
M-( H~ )
where A, B. C, D are square matrices of the order p, B and C being symmetric, and D
saUsfying the relaUon D - - tA + a.I.
The decomposlUon
The /IIImanIokJs 325
M-(
A
B -tA
C 0
0 oo aI
0 0
0
J
000 p J.1 a
From the classification of Williamson (see, for example, the book of Arnold on the
classical mechanic), we can reduce g to the form
(here < ,> denotes the product induced on m from 9 by Identification of m with T.,M
and ~ the tie algebra of K).
Suppose now that M - Kp and g - gA.l....,Ap .
ProposiUon 7.For every left invariant metric gA.l,''''Ap , the Riemannian space
(Up, gA.lo....Ap) is homogeneous naturally reductive.
327
other products are zero. Then 1. - ~ E9 m, where ~ Is the Abelian algebra generated
by the RI gives the naturally reductive decomposition.
Theorem 7. Let G be a nilpotent Ue group such that for every left lnvarlantmetrlc g,
the Riemannian space(G, g) is ahomogeneous naturally reductive space. Then the Ue
algebra I} ofG is a dJrect sum of Heisenberg and Abellan algebras.
The proof is based on the notion of Riemannian contractions. We study some models
of nilpotent Riemannian Ue groups and, by deformation, we can look all the other
nilpotent groups (see [GP2].
This theorem shows the importance of the riemannian Heisenberg group, because the
behavior of the isometry group Is analogous to the group of Isometries of the
Riemannian Abelian group which corresponds to the Euclidean geometry.
BIBLIOGRAPHY
fAGH 2] Ancochea-Bermudez).M., Gaze M, Hakimjanov Yu. Sur les composantes de]a varlele
des aJgebres de Ue nllpotentes.
C.R.A.Sc. Paris, t 314 CD, 1992,657-660.
[0111 Olxmler J. Sur les representallons unltalres des groupes de Ue nlJpoll:nll:s Ill.
Canad.J. Math., 1958, v. 10,321-348.
[012) Olxmler J. CohomolosJe des aIgebres de Ue nlJpoll:ntes.
Acta ScI. Mat szeged, 1955, v. 16, 246-250.
[fS) Favre G., Santharoubane LJ. NlJpoll:nt Ue algebras of maximal rank and of classical
simple type.
Preprlnt.
[GO 11 Goze M. Bouyakoub K. Sur les aJgebres de Ue munies d'une forme symplectlque.
Rend. Unlv. Cagliart, V. LVII (1), 1987.
IKH 61 HaklmjanOV You (Khaklmdjanov Yu). Sur les varieles d'a/gebre de Ue.
Commun. Algebra, 1990, v. 18, N° 4, 1147-1187.
[KH 91 Haklmjanov You (Khaklmdjanov Yu). Varletes des lois d'algehres de Ue nilpotentes.
Geometriae dedicata, 1991, v. 40, N° 3, 269-295.
ILL 21 Leger G., Luks E. Cohomology and weight systems for nHpotent Ue algebras.
Bull. A.M.5., 1974, v. 50, N° 1, 77-80.
Bibliography 333
[LU 11 Luks E. lie algebra with only Inner deI1vations need be complete.
J. of Algebra, 1970, v. 15, N" 2, 280-282.
[LU 21 Luks E. What Is the typIcal nilpotent lie algebra ?
In ·Computers In nonassoclattve rings and algebras", Academic Press, 1977.
[NR 21 Nljenhuls A., Richardson R.W. Cohomology and deformations of algebraic structures.
Bull. AM.S., 1964, V. 70,406-411.
[NR 31 Nljenhuls A., Richardson R.W. Cohomology and deformations In graded lie algebras.
Bull. AM.s., 1966, V. 72, Nl, 1-29.
[RI 1] Richardson R.W. A I1gIdIty theorem for subalgebras of lie and assocIative algebra.
Illinois J. Math., 1967, V. 11, 92-110.
[SEMI 5emlna\re ·Sophus Ue". Theone des ~res de Ue. Topologle des groupes de Ue.
Parts : ENS. 1954-1955.
[WI Williams f. Laplace operators and the h-module structure of certain cohomology
sroups.
Trans. AMS., 1975, V. 197, N° I, 1-57.
scheme 174
semidirect sum 10
semisimple Lie algebra 21,28
sill algebra 282
sill cocyc1e 282
simple lie algebra 21,32
simple module 24
simple root 30
singular root 129
solvable lie algebras 12
special linear algebra 3
special orthogonal Lie algebra 3
spectral sequences 95
standardisation 188
standard Lie algebra 50,52
subalgebra (of lie algebra) 3
symplectic geometry 306
symplectic Lie algebra 4
symplectic structure 306