Lectures On Invariant Theory S

Preface
This book is based on one-semester graduate courses I gave at Michigan in 1994

and 1998, and at Harvard in 1999. A part of the book is borrowed from an earlier
version of my lecture notes which were published by the Seoul National University [22]. The main changes consist of including several chapters on algebraic
invariant theory, simplifying and correcting proofs, and adding more examples
from classical algebraic geometry. The last Lecture of [22] which contains some
applications to construction of moduli spaces has been omitted. The book is literally intended to be a first course in the subject to motivate a beginner to study
more. A new edition of D. Mumfords book Geometric Invariant Theory with appendices by J. Fogarty and F. Kirwan [73] as well as a survey article of V. Popov
and E. Vinberg [89] will help the reader to navigate in this broad and old subject
of mathematics. Most of the results and their proofs discussed in the present book
can be found in the literature. We include some of the extensive bibliography of
the subject (with no claim for completeness). The main purpose of this book is
to give a short and self-contained exposition of the main ideas of the theory. The
sole novelty is including many examples illustrating the dependence of the quotient on a linearization of the action as well as including some basic constructions
in toric geometry as examples of torus actions on affine space. We also give many
examples related to classical algebraic geometry. Each chapter ends with a set of
exercises and bibliographical notes. We assume only minimal prerequisites for
students: a basic knowledge of algebraic geometry covered in the first two chapters of Shafarevichs book [102] and/or Hartshornes book [46], a good knowledge
of multilinear algebra and some rudiments of the theory of linear representations
of groups. Although we often use some of the theory of affine algebraic groups,
the knowledge of the group GL is enough for our purpose.
I am grateful to some of my students for critical remarks and catching numerous mistakes in my lecture notes. Special thanks go to Ana-Maria Castravet,
Mihnea Popa and Janis Stipins.
i
Contents
Preface
Introduction
vii
.
.
.
.
.
1
1
4
10
13
14
.
.
.
.
.
.
17
17
20
21
22
27
27
.
.
.
.
.
.
29
29
32
35
41
45
46
The symbolic method

1.1 First examples . . . . . . .
1.2 Polarization and restitution
1.3 Bracket functions . . . . .
Bibliographical notes . . . . . .
Exercises . . . . . . . . . . . .
.
.
.
.
.
The First Fundamental Theorem

2.1 The omega-operator . . . . .
2.2 The proof . . . . . . . . . .
2.3 Grassmann varieties . . . . .
2.4 The straightening algorithm .
Bibliographical notes . . . . . . .
Exercises . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
Reductive algebraic groups

3.1 The GordanHilbert Theorem
3.2 The unitary trick . . . . . . .
3.3 Affine algebraic groups . . . .
3.4 Nagatas Theorem . . . . . . .
Bibliographical notes . . . . . . . .
Exercises . . . . . . . . . . . . . .
iii
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
iv
4 Hilberts Fourteenth Problem
4.1 The problem . . . . . . . .
4.2 The Weitzenbock Theorem
4.3 Nagatas counterexample .
Exercises . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5 Algebra of covariants
5.1 Examples of covariants . . . . . . . . . .
5.2 Covariants of an action . . . . . . . . . .
5.3 Linear representations of reductive groups
5.4 Dominant weights . . . . . . . . . . . . .
5.5 The CayleySylvester formula . . . . . .
5.6 Standard tableaux again . . . . . . . . . .
Bibliographical notes . . . . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . . . . .
6 Quotients
6.1 Categorical and geometric quotients
6.2 Examples . . . . . . . . . . . . . .
6.3 Rational quotients . . . . . . . . . .
Bibliographical notes . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . .
7 Linearization of actions
7.1 Linearized line bundles . . .
7.2 The existence of linearization
7.3 Linearization of an action . .
Bibliographical notes . . . . . . .
Exercises . . . . . . . . . . . . .
8 Stability
8.1 Stable points . . . . . . . .
8.2 The existence of a quotient
8.3 Examples . . . . . . . . .
Exercises . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
47
47
49
52
62
62
.
.
.
.
.
.
.
.
65
65
69
72
77
79
84
87
88
.
.
.
.
.
91
91
95
98
100
100
.
.
.
.
.
103
103
107
110
112
113
.
.
.
.
.
115
115
117
121
127
127
v
9
Numerical criterion of stability

9.1 The function ) . . .
9.2 The numerical criterion . .
9.3 The proof . . . . . . . . .
9.4 The weight polytope . . .
9.5 Kempf-stability . . . . . .
Exercises . . . . . . . . . . . .
10 Projective hypersurfaces
10.1 Nonsingular hypersurfaces
10.2 Binary forms . . . . . . .
10.3 Plane cubics . . . . . . . .
10.4 Cubic surfaces . . . . . . .
Bibliographical Notes . . . . . .
Exercises . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
129
129
132
133
135
138
142
143
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
145
145
147
153
161
162
162
11 Configurations of linear subspaces

11.1 Stable configurations . . . . .
. . . . . . . . . .
11.2 Points in
11.3 Lines in . . . . . . . . . .
Bibliographical notes . . . . . . . .
Exercises . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
165
165
171
181
183
184
.
.
.
.
.
187
187
190
196
202
202
.
.
.
.
.
.
12 Toric varieties
12.1 Actions of a torus on an affine space
12.2 Fans . . . . . . . . . . . . . . . . .
12.3 Examples . . . . . . . . . . . . . .
Bibliographical notes . . . . . . . . . . .
Exercises . . . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Bibliography
205
Index of Notation
215
Index
217
Introduction
Geometric invariant theory arises in an attempt to construct a quotient of an algebraic variety by an algebraic action of a linear algebraic group . In many
applications is the parametrizing space of certain geometric objects (algebraic
curves, vector bundles, etc.) and the equivalence relation on the objects is defined
by a group action. The main problem here is that the quotient space may
not exist in the category of algebraic varieties. The reason is rather simple. Since
one expects that the canonical projection is a regular map of algebraic varieties and so has closed fibers, all orbits must be closed subsets in the
Zariski topology of . This rarely happens when is not a finite group. A possible solution to this problem is to restrict the action to an invariant open Zariski
subset , as large as possible, so that exists. The geometric invariant
theory (GIT) suggests a method for choosing such a set so that the quotient is a
quasi-projective algebraic variety. The idea goes back to David Hilbert. Suppose
is a linear space and is a linear algebraic group acting on via its
linear representation. The set of polynomial functions on invariant with respect
to this action is a commutative algebra over the ground field. Hilbert proves that
is finitely generated if SL or GL and any set of generators
of defines an invariant regular map from to some affine algebraic variety
whose ring of polynomial functions is isomorphic
contained in affine space
to . By a theorem of Nagata the same is true for any reductive linear algebraic
group. The map has a universal property for -invariant maps of
and is called the categorical quotient. The inverse image of the origin is the
closed subvariety defined by all invariant homogeneous polynomials of positive
degree. It is called the null-cone. Its points cannot be distinguished by invariant
functions; they are called unstable points. The remaining points are called semistable points. When we pass to the projective space associated to , the
images of semi-stable points form an invariant open subset ss and the map
induces a regular map
ss , where (denoted by ss ) is
vii
viii
INTRODUCTION
a projective algebraic variety with the projective coordinate algebra isomorphic

to . In applications considered by Hilbert, parametrizes projective hypersurfaces of certain degree and dimension, and the projective algebraic variety
is the moduli space of these hypersurfaces. The hypersurfaces represented by
unstable points are left out from the moduli space; they are too degenerate. A
nonsingular hypersurface is always represented by a semi-stable point. Since is
a projective variety, it is considered as a compactification of the moduli space
of nonsingular hypersurfaces. The fibers of the map ss ss are
not orbits in general; however, each fiber contains a unique closed orbit so that
ss parametrizes closed orbits in the set of semi-stable points.
Since the equations of the null-cone are hard to find without computing explicitly the ring of invariant polynomials, one uses another approach. This approach
is to describe the set of semi-stable points by using the HilbertMumford numerical criterion of stability. In many cases it allows one to determine the set ss
very explicitly and to distinguish stable points among semi-stable ones. These are
the points whose orbits are closed in ss and whose stabilizer subgroups are
finite. The restriction of the map ss ss to the set of stable points
s is an orbit map s s . It is called a geometric quotient.
More generally, if is a reductive algebraic group acting on a projective algebraic variety , the GIT approach to constructing the quotient consists of the
following steps. First one chooses a linearization of the action, a -equivariant
embedding of into a projective space with a linear action of as above.
The choice of a linearization is a parameter of the construction; it is defined by
a -linearized ample line bundle on . Then one sets ss ss and
defines the categorical quotient ss ss as the restriction of the categorical
quotient ss ss . The image variety ss is a closed subvariety
of ss .
Let us give a brief comment on the content of the book.
In Chapters 1 and 2 we consider the classical example of invariant theory in
which the general linear group GL of a vector space of dimension over
a field acts naturally on the space of homogeneneous polynomials Pol of
some degree . We explain the classical symbolic method which allows one to
identify an invariant polynomial function of degree on this space with an element of the projective coordinate algebra Gr on the Grassmann variety
Gr of -dimensional linear subspaces in in its Plucker embedding. This
interpretation is based on the First Fundamental Theorem of Invariant Theory. The
proof of this theorem uses a rather technical algebraic tool, the so-called Clebsch
omega-operator. We choose this less conceptual approach to show the flavor of the
ix
invariant theory of the nineteenth century. More detailed expositions of the classical invariant theory ([64], [121]) give a conceptual explanation of this operator
via representation theory. The Second Fundamental Theorem of Invariant Theory
is just a statement about the relations between the Plucker coordinates known in
algebraic geometry as the Plucker equations. We use the available computations
of invariants in later chapters to give an explicit description of some of the GIT
quotients arising in classical algebraic geometry.
In Chapter 3 we discuss the problem of finite generatedness of the algebra of
invariant polynomials on the space of a linear rational representation of an algebraic group. We begin with the GordanHilbert theorem and explain the unitary
trick due to Adolf Hurwitz and Hermann Weyl which allows one to prove the
finite generatedness in the case of a semisimple or, more generally, reductive complex algebraic group. Then we introduce the notion of a geometrically reductive
algebraic group and prove Nagatas theorem on finite generatedness of the algebra of invariant polynomials on the space of a linear rational representation of a
reductive algebraic group.
In Chapter 4 we discuss the case of a linear rational representation of a nonreductive algebraic group. We prove a lemma due to Grosshans which allows one to
prove finite generatedness for the restriction of a representation of a reductive algebraic group to a subgroup provided the algebra of regular functions on the
homogeneous space is finitely generated. A corollary of this result is a classical theorem of Weitzenbock about invariants of the additive group. The central
part of this chapter is Nagatas counterexample to Hilberts Fourteenth Problem.
It asks about finite generatedness of the algebra of invariants for an arbitrary algebraic group of linear transformations. We follow the original construction of
Nagata with some simplifications due to R. Steinberg.
Chapter 5 is devoted to covariants of an action. A covariant of an affine algebraic group acting on an algebraic variety is a -equivariant regular map
from to an affine space on which the group acts via its linear representation. The
covariants form an algebra and the main result of the theory is that this algebra is
finitely generated if is reductive. The proof depends heavily on the theory of linear representations of reductive algebraic groups which we review in this chapter.
As an application of this theory we prove the classical Cayley-Sylvester formula
for the dimension of the spaces of covariants and also the Hermite reciprocity.
In Chapter 6 we discuss categorical and geometric quotients of an algebraic
variety under a regular action of an algebraic group. The material is fairly standard
and follows Mumfords book.
Chapter 7 is devoted to linearizations of actions. The main result is that any
INTRODUCTION
algebraic action of a linear algebraic group on a normal quasi-projective algebraic

variety is isomorphic to the restriction of a linear action on a projective space
in which is equivariantly embedded. The proof follows the exposition of the
theory of linearizations from [65].
Chapter 8 is devoted to the concept of stability of algebraic actions and the
construction of categorical and geometric quotients. The material of this chapter
is rather standard and can be found in Mumfords book as well as in many other
books. We include many examples illustrating the dependence of the quotients on
the linearization.
Chapter 9 contains the proof of HilbertMumfords numerical criterion of stability. The only novelty here is that we also include Kempfs notion of stability
and give an example of its application to the theory of moduli of abelian varieties.
The remaining Chapters 1012 are devoted to some examples where the complete description of stable points is available. In Chapter 10 we discuss the case
of hypersurfaces in projective space. We give explicit descriptions of the moduli
spaces of binary forms of degree , plane curves of degree 3 and cubic surfaces.
In Chapter 11 we discuss moduli spaces of ordered collections of linear subspaces
in projective space, in particular of points in or of lines in . The examples discussed in this chapter are related to some of the beautiful constructions of
classical algebraic geometry. In Chapter 12 we introduce toric varieties as GIT
quotients of an open subset of affine space. Some of the constructions discussed
in the preceding chapters admit a nice interpretation in terms of the geometry of
toric varieties. This approach to toric varieties is based on some recent work of D.
Cox ([16]) and M. Audin ([3]).
We will be working over an algebraically closed field sometimes assumed
to be of characteristic zero.
Chapter 1
The symbolic method
1.1
First examples
The notion of an invariant is one of the most general concepts of mathematics.

Whenever a group acts on a set we look for elements which do not
change under the action, i.e., which satisfy for any . For example,
if is a set of functions from a set to a set , and acts on via its action on
and its action on by the formula
then an equivariant function is a function
satisfying

, i.e.,
In the case when acts trivially on , an equivariant function is called an invariant function. It satisfies
Among all invariant functions there exists a universal function, the projection map
from the set to the set of orbits . It satisfies the property

that for any invariant function there exists a unique map
. So if we know the set of orbits , we know all

such that
invariant functions on . We will be concerned with invariants arising in algebra
and algebraic geometry. Our sets and our group will be algebraic varieties and
our invariant functions will be regular maps.
Let us start with some examples.
CHAPTER 1. THE SYMBOLIC METHOD
Example 1.1. Let be a finitely generated algebra over a field

group of its automorphisms. The subset
and let
be a
(1.1)
is a -subalgebra of . It is called the algebra of invariants. This definition

fits the general setting if we let Specm be the affine algebraic variety
over with coordinate ring equal to , and let be the affine line over
. Then elements of can be viewed as regular functions between algebraic varieties. A more general invariant function is an invariant map
between algebraic varieties. If is affine with coordinate ring ,
such a map is defined by a homomorphism of -algebras qrs satisfying
t t for any t . It is clear that such a homomorphism is
equal to the composition of a homomorphism and the natural inclusion
map . Thus if we take Specm we obtain that the map
defined by the inclusion plays the role of the universal function. So it is
natural to assume that is the coordinate ring of the orbit space . However,
we shall quickly convince ourselves that there must be some problems here. The
first one is that the algebra may not be finitely generated over and so does
not define an algebraic variety. This problem can be easily resolved by extending
the category of algebraic varieties to the category of schemes. For any (not necessarily finitely generated) algebra over , we may still consider the subring of
invariants and view any homomorphism of rings as a morphism of
affine schemes Spec Spec . Then the morphism Spec Spec
is the universal invariant function. However, it is preferable to deal with algebraic
varieties rather than to deal with arbitrary schemes, and we will later show that
is always finitely generated if the group is a reductive algebraic group which
acts algebraically on Specm . The second problem is more serious. The affine
algebraic variety Specm rarely coincides with the set of orbits (unless is a
finite group). For example, the standard action of the general linear group GL
on the space has two orbits but no invariant nonconstant functions.

The following is a more interesting example.
Example 1.2. Let GL act by automorphisms on the polynomial algebra
in variables , as follows. For any
the polynomial is equal to the th entry of the matrix
where
Specm
is the matrix with the entries

is the affine space Mat of dimension
(1.2)
. Then, the affine variety

. Its -points can be inter-
1.1 FIRST EXAMPLES
preted as matrices with entries in and we can view elements of as

polynomial functions on the space of matrices. We know from linear algebra that
any such matrix can be reduced to its Jordan form by means of a transformation
(1.2) for an appropriate . Thus any invariant function is uniquely determined
by its values on Jordan matrices. Let be the subspace of diagonal matrices
identified with linear space and let be the algebra of polynomial functions on . Since the set of matrices with diagonal Jordan form is a
Zariski dense subset in the set of all matrices, we see that an invariant function
is uniquely determined by its values on diagonal matrices. Therefore the restriction homomorphism is injective. Since two diagonal matrices with permuted diagonal entries are equivalent, an invariant function must
be a symmetric polynomial in . By the Fundamental Theorem on Symmetric
Functions, such a function can be written uniquely as a polynomial in elementary
symmetric functions in the variables . On the other hand, let be
the coefficients of the characteristic polynomial

q

considered as polynomial functions on Mat , i.e., elements of the ring . Clearly,
the restriction of to is equal to the th elementary symmetric function . So
we see that the image of in coincides with the polynomial subalgebra . This implies that is freely generated by the functions .
So we can identify Specm with affine space . Now consider the universal
map Specm Specm . Its fiber over the point defined by the
maximal ideal is equal to the set of matrices with characteristic

polynomial
. Clearly, this set does not consist of one orbit,
any Jordan matrix with zero diagonal values belongs to this set. Thus Specm
is not the orbit set Specm .
We shall discuss later how to remedy the problem of the construction of the
space of orbits in the category of algebraic varieties. This is the subject of the geometric invariant theory (GIT) with which we will be dealing later. Now we shall
discuss some examples where the algebra of invariants can be found explicitly.
Let be a finite-dimensional vector space over a field and let
GL
be a linear representation of a group in . We consider the associated action of

on the space Pol of degree homogeneous polynomial functions on .
This action is obviously linear. The value of Pol at a vector is given, in
4
terms of the coordinates
by the following expression:
of
with respect to some basis
(1.3)

The direct sum of the vector spaces Pol is equal to the graded algebra of
polynomial functions Pol . Since is infinite (we assumed it to be algebraically

closed), Pol is isomorphic to the polynomial algebra . In more
sophisticated language, Pol is naturally isomorphic to the th symmetric
product of the dual vector space and Pol is isomorphic to the

symmetric algebra .
We will consider the case when Pol and SL be the special
linear group with its linear action on described above. Let Pol Pol .
We can take for coordinates on the space Pol the functions which assign
to a homogeneous form (1.3) its coefficient . So any element from is a polynomial in the . We want to describe the subalgebra of invariants .
The problem of finding is almost two centuries old. Many famous matheor in the vector notation,
maticians of the nineteenth century made a contribution to this problem. Complete

results, however, were obtained only in a few cases. The most complete results
are known in the case

, the case where consists of binary forms of
degree . We write a binary form as
coefficients, and hence elements of

In this case we have

in variables.
1.2
are polynomials
Polarization and restitution
To describe the ring Pol Pol SL one uses the symbolic expression of a
polynomial, which we now explain. We assume that char .
A homogeneous polynomial of degree 2 on a vector space is a quadratic
form. Recall its coordinate-free definition: a map is a quadratic form
if the following two properties are satisfied:
1.2. POLARIZATION AND RESTITUTION

(i)
for any
and any ;

(ii) the map
defined by the formula
is bilinear.
A homogeneous polynomial Pol
way by the following properties:

(i)
(ii)
of degree
can be defined in a similar
, for any and any ;
the map pol defined by the formula

pol

q

r

is multilinear.
Here and throughout we use to denote the set .
As in the case of quadratic forms, we immediately see that the map pol is
a symmetric multilinear form and also that can be reconstructed from pol
by the formula
pol
The symmetric multilinear form pol
symmetric multilinear from
res
is called the restitution of
and
is called the polarization of . For any

the function res defined by
. It is immediately checked that res

pol res
Pol
Since we assumed that char , we obtain that each Pol is equal to

the restitution of a unique symmetric -multilinear form, namely pol .
. We
Assume that is equal to the product of linear forms
have

pol
(1.4)
Here
the permutation group on letters.
denotes
Let
be a basis of and be the dual basis of . Any
can be written in a unique way as . Let Sym be the
vector space of symmetric -multilinear forms on . For any

and any Sym , we have
Taking
res
, we obtain that
Thus any polynomial

Pol can be written uniquely as a sum of monomials . This is the coordinate-dependent definition of a homogeneous
polynomial. Since the polarization map
Sym

is obviously linear, we obtain that Sym has a basis formed by the polariza
tions of monomials . Applying (1.4), we have
pol Pol
pol
If we denote by a th copy of the basis

rewrite the previous expression as
pol
in
, we can
Here, we consider the product of

on q . We have
linear forms on
as an
-multilinear form
pol

(1.5)
If we write , then the right-hand side is equal to

if and zero otherwise.
Note that the polarization allows us to identify Pol with the dual to the
space Pol . To see this, choose a basis of Pol formed by the monomi
als . For any Sym we can set

and then extend the domain of to all homogeneous degree
linearity. Applying (1.5), we get
pol
if
otherwise.
s
Pol

pol
This shows that the map from Pol
polynomials by
to defined by
(1.6)
is a perfect duality, i.e., it defines isomorphisms
Pol Pol
of Pol
Moreover, the monomial
basis
basis
.
Remark 1.1. Note that the coefficients of a polynomial

Pol
Pol
Pol
(1.7)
is dual to the
(1.8)
on . We can view the expression
as a general homogeneous polynomial of degree .

general
formula
Thus we get a strange

general
are equal to the value of
This explains the classical notation of a homogeneous polynomial as a power of a

linear polynomial.
Remark 1.2. One can view a basis vector as a linear differential operator on
Pol which acts on linear functions by . It acts on any polynomial
as the partial derivative . Thus we can identify any poly

nomial Pol with the differential operator by
replacing the variable with . In this way the duality Pol Pol
is defined by the formula
Remark 1.3. For the reader with a deeper knowledge of multilinear algebra, we
recall that there is a natural isomorphism between the linear space Pol and
the th symmetric power of the dual space . The polarization map is

a linear map from to which is bijective when char .
The universal property of tensor product allows one to identify the spaces
and Sym .
.
Let us now consider the case when Pol , where
First recall that a multihomogeneous function of multi-degree on
is a function on which is a homogeneous polynomial function of degree

in each variable; when each , we get the usual definition of a multilinear
function. We denote the linear space of multihomogeneous functions of multidegree by Pol . The symmetric group
acts naturally
on the space Pol by permuting the variables. The subspace of invariant
(symmetric) functions will be denoted by Sym . In particular,
Sym
Sym
Lemma 1.1. We have a natural isomorphism of linear spaces

symb Pol
Pol
Sym
Proof. The polarization map defines an isomorphism

Pol
Pol
Sym
Pol
Using the polarization again we obtain an isomorphism Pol Pol .

Thus any linear function on Pol is a homogeneous polynomial function of
degree on . Thus a multilinear function on Pol can be identified with a
multihomogeneous function on of multi-degree .
Let us make the isomorphism from the preceding lemma more explicit by
using a basis in and its dual basis in . Let
we write each Pol as

be the coordinate functions on Pol , where
in (1.8) with replaced by , so that

. Any Pol Pol

is a polynomial expression in the of degree . Let be
the coordinate functions in each copy of Pol . The polarization pol is a

multilinear expression in the . Now, if we replace with the monomial

in a basis of the th copy of , we obtain the symbolic expression
of
symb Pol
Remark 1.4. The mathematicians of the nineteenth century did not like superscripts and preferred to use different letters for vectors in different copies of the
same space. Thus they would write a general polynomial

of
degree as
and the symbolic expression of a function as an expression in

.
Example 1.3. Let . In this
case Pol consists of quadratic forms in
two variables
. The discriminant
is an obvious invariant of SL . We have
symb
where
pol
q
10
Example 1.4. Let
. The determinant (called the Hankel determinant)
in coefficients of a binary quartic
Pol on the space of binary quartics. It is called
defines a function Pol
the catalecticant. We leave as an exercise to verify that its symbolic expression is

equal to

symb
It is immediate to see that the group GL acts on
by
via its action on
This implies that the catalecticant is invariant with respect to the group SL
1.3
(1.9)
Bracket functions
as a ma
... . . . ...

First, we identify the space Pol with the subspace of the polynomial

algebra consisting of polynomials which are

homogeneous of degree in each set of variables . Next, we identify

the algebra with the algebra Pol Mat of

polynomial functions on the space of matrices Mat . The value of a variable
at a matrix is the -entry of the matrix. The group acts naturally on

the space Mat by

It is convenient to organize the variables
trix of size :
1.3. BRACKET FUNCTIONS
11
where we write a matrix as a collection of its columns. In a similar way the
group acts on Mat by row multiplication. We say that a polynomial
Pol Mat is multihomogeneous of multi-degree if for any ,
and any Mat ,
if the polynomial
We say that
is multiisobaric of multi-weight
on the space Mat is multihomogeneous of multi-degree
function
. Let Pol Mat denote the linear space of polyno
mial functions on the space Mat which are multihomogeneous of multi-degree
and multiisobaric of multi-weight . If
we write ; we use similar notation for the weights.

It follows from the definition that the symbolic expression of any invariant
polynomial from Pol Pol is multilinear. Let us show that it is also multi
isobaric:
Proposition 1.1.
symb Pol
where
SL
Pol
Pol Mat

Proof. We shall consider any Pol Pol as a polynomial in coefficients
of the general polynomial from Pol . For any GL we

can write
where
SL . It is clear that the scalar matrix acts on each element of
the basis of by multiplying it by . Hence it acts on the coordinate function
by multiplying it by and on Pol via multiplication by . Hence it acts

on Pol Pol by multiplication by (recall that .
Therefore we get

Since any GL can be written as an th power, we obtain that
for some homomorphism GL
. Notice that when we fix
and Pol , the function is a polynomial function in
12
entries of the matrix which is homogeneous of degree . Also, we know that

Since is an irreducible polynomial of degree in entries
is a nonnegative power of . Comparing the

of the matrix, we obtain that
degrees we get, for any GL

Since the map symb Pol
Pol Mat is GL -equivariant, we
see that
symb GL
If we take to be the diagonal matrix of the form diag we
immediately obtain that symb is multiisobaric of multi-weight . Also, by

definition of the symbolic expression, symb is multihomogeneous of multidegree . This proves the assertion.
Corollary 1.1. Assume . Then, for
,
Pol Pol SL
An example of a function from Pol Mat is the determinant function

. More generally we define the bracket function det on Mat
whose value on a matrix is equal to the maximal minor formed by the columns
from a subset of . If we will often use its

Pol
classical notation for the minors

det q
It is isobaric of weight but not multihomogeneous if . Using these func

tions one can construct functions from Pol Mat whenever .
This is done as follows.

Definition. A (rectangular) tableau on the set
is a matrix

..
.

of size
.
(1.10)
. ..

with entries in satisfying the inequalities . We say that the tableau

is homogeneous of degree if each , occurs exactly times; clearly
must satisfy the relation
..
BIBLIOGRAPHICAL NOTES
13
An example of a tableau on the set
For each tableau
of size and degree 2 is
as above we define the tableau function on Mat by
We say that is homogeneous of degree if is of degree . It is clear that any
such function belongs to Pol Mat . For example, the symbolic expression
of the determinant of a binary quadratic form from Example 1.3 is equal to .

The symbolic expression

of the catalecticant corresponds to the
function , where
Notice the way a tableau function changes when we apply a transformation
GL : each bracket function is multiplied by . So

each
for
. In
tableau on the set of size
the function
is multiplied by
particular, each such function is an invariant for the group SL of matrices
with determinant equal to 1. Taking linear combinations of homogeneous degree
tableau functions that are invariant with respect to permutation of columns, we
get a lot of examples of elements in Pol Pol SL . In the next chapter we
will prove that any element from this ring is obtained in this way.
Bibliographical notes
The symbolic method for expression of invariants goes back to the earlier days of
theory of algebraic invariants, which originates in the work of A. Cayley of 1846.
It can be found in many classical books on invariant theory ([28], [38], [39], [47],
14
[96]). A modern exposition of the symbolic method can be found in [18], [64], [83].
The theory of polarization of homogeneous forms is a basis of many constructions
in projective algebraic geometry; see for example [14], [39], [97], [98]. For a
modern treatment of some of the geometric applications we refer to [24], [53].
Exercises
1.1 Show that Pol Mat
unless
Pol be the space of quadratic forms on a vector space of

1.2 Let
dimension .
(i) Assume that char
or is odd. Show that Pol SL is generated
(as a -algebra) by the discriminant function whose value at a quadratic form is
equal to the determinant of the matrix defining its polar bilinear form.
(ii) Which level sets of the discriminant function are orbits of SL in ?
1.3 Let Pol . For any and consider the function on
. Show that this function extends to

defined by

and let denote the restriction of the extended function to .
(i) Show that Pol and the pairing
Pol
Pol
is bilinear.
(ii) Assume in . Let Pol
. Show that the function
coincides with pol .
(iii) Show that

with respect to some basis
Pol

defined by
be the linear map
, where
are the coordinates of
1.4 Let be the projective space associated to a vector space of dimension

. We consider each nonzero as a point in . The hypersurface
in is called the polar hypersurface of the hypersurface
with respect to the point . Show that for any the

tangent hyperplane of
at contains the point .
EXERCISES
15
1.5 Consider the bilinear pairing between Pol and Pol defined as in
(1.6). For any Pol Pol denote the value of this pairing at
by . Show that

defines a linear map
(i) for fixed the assignment
ap
Pol Pol

,
,
if is the product of linear polynomials
(ii) for any Pol

(iii)
.
1.6 In the notation of the preceding exercise, Pol is called apolar to a

. Show that
homogeneous form Pol if
(i) is apolar to if and only if ,
(ii) is apolar to if and only if all partial derivatives of

vanish at .
1.7 Consider the linear map ap defined in Exercise 1.5. The matrix
of this map
with respect to the basis in Pol defined by the monomials and the basis
in Pol defined by the monomials is called the catalecticant matrix. Show
that
(i) Show that if

the determinant of the catalecticant matrix is an
invariant on the space Pol (it is called the catalecticant invariant) .

(ii) Show that, if
and , the catalecticant invariant coincides
with the one defined in Example 1.4.
(iii) Find the degree of the catalecticant invariant.
(iv) Show that the catalecticant invariant on the space Pol coincides with
the discriminant invariant.
.
(v) Compute the catalecticant matrix in the case
write
1.8 Let Pol . For any and any
(i) Show that the function

is
multihomogeneous of multi-degree .
(ii) Show that pol .
1.9 Find the symbolic expression for the polynomial

on
the space of binary quartics Pol . Show that it is an invariant for the group
SL .
16
1.10 Find the polarization of the determinant polynomial .

1.11 Let GL be a homomorphism of groups. Assume that is
given by a polynomial in the entries of GL . Prove that there exists a
.
nonnegative integer such that, for all GL ,
Chapter 2
The First Fundamental Theorem
2.1
The omega-operator
We saw in the preceding chapter that the symbolic expressions of the discriminant
of a binary quadratic form and of the catalecticant of a binary quartic are polynomials in the bracket functions. The theorem from the title of this chapter shows
that this is the general case for invariants of homogeneous forms of any degree
and in any number of variables. In fact
we will show more: the bracket functions

generate the algebra Pol Mat SL . Recall that the group SL acts on this
ring via its action on matrices by left multiplication.

We start with some technical lemmas.
For any polynomial let denote the (differential) operator on

obtained by replacing each unknown with the partial derivative
operator (cf. Remark 1.2).

In this section we will use only a special operator of this sort. We take

with unknowns
and let be the determinant function of
the matrix with entries . We denote the corresponding operator by . It is

called the omega-operator or the Cayley operator.
Lemma 2.1.
Proof. First observe that for any permutation
we have

17
(2.1)
CHAPTER 2. THE FIRST FUNDAMENTAL THEOREM
18
where is the sign of the permutation . This immediately gives that

. For any subset of set
Analogously to (2.1) we get

(2.2)
where for any two subsets of of the same cardinality we denote by

the minor of the matrix formed by the rows corresponding to the set and
the columns corresponding to the set . The bar denotes the complementary set
and
t
sign q
r s
rt s
Now applying the chain rule we get
Now recall a well-known formula from multilinear algebra which relates the
minors of a matrix and the minors of its adjoint (also called adjugate in classic
adj (see [8], Chapter 3, 11, exercise 10):
literature) matrix

(2.3)

Applying (2.3) we obtain
2.1. THE OMEGA-OPERATOR
19
Now recall the Laplace formula for the determinant of a square matrix
size :
of
(2.4)
where

is a fixed partition of the set of rows of and is equal to the sign of
the permutation where we assume that the elements of each set are
listed in the increasing order. Applying this formula to we find

where Thus, letting run through the set , we sum up

the expressions to get
where
We leave to the reader as an exercise to verify that

The precise value of the nonzero constant is irrelevant for what follows.
, where each is equal to the

Lemma 2.2. Let

product of linear forms . Then

..
..

..
.
.
.
where the sum is taken over the set
20
Proof. By the chain rule,

multiplying by the sign of the permutation
and summing up
After
over the set of permutations, we get the desired formula from the assertion of the
lemma.
2.2
The proof
Now we are ready to prove the First Fundamental Theorem of Invariant Theory:
Theorem 2.1. The algebra of invariants Pol Mat SL is generated by the
bracket functions .
Proof. Let Pol Mat be the subspace of polynomials which are multiisobaric
of multi-weight
. It is clear that

Pol Mat SL
Pol Mat
SL
So we may assume that an invariant polynomial Pol Mat SL belongs to
Pol Mat . Fix a matrix Mat and consider the assignment
as a function on Mat . It follows from the proof of Proposition 1.1 that

Since is multiisobaric, it is easy to see that can be written as a sum

of products of linear polynomials as in Lemma 2.2, with . Applying the
omega-operator to the left-hand side of the identity times we will be able to get
rid of the variables and get a polynomial in bracket functions. On the other
hand, by Lemma 2.1 we get a scalar multiple of . This proves the theorem.
Let Tab denote the subspace of Pol Mat spanned by tableau func
tions on of size and let Tab hom be its subspace spanned by homo
geneous tableau functions of degree . Recall that, as follows from the definition
of a tableau, . The symmetric group
acts linearly on the space
Tab via its action
on
tableaux
by
permuting
the
elements of the set . We
denote by Tab
the subspace of invariant elements. Clearly,
Tab hom
Tab
2.3. GRASSMANN VARIETIES

Corollary 2.1. Let
21
. We have
Pol Mat
SL

Tab
hom
By Proposition 1.1, the symbolic expression of any

invariant polynomial
SL
SL
from Pol Pol
belongs to Pol Mat , and hence must be a lin
ear combination of tableau functions from Tab . The group
acts naturally on Mat by permuting the columns and hence acts naturally on Pol Mat
leaving the subspaces Pol Mat invariant. Applying Lemma 1.1, we get
Corollary 2.2.
symb Pol
where
2.3
Pol SL
Tab

hom
Grassmann varieties

The ring Pol Mat SL has a nice geometric interpretation. Let Gr be
the Grassmann variety of -dimensional linear subspaces in (or, equivalently,

-dimensional linear projective subspaces of ). Using
the Plucker map
, we can embed Gr in . The projective

coordinates in this projective space are the Plucker coordinates
. Consider the set of ordered -tuples in . Let
be the polynomial ring whose variables are the Plucker coordinates indexed
by elements of the set . We view it as the projective coordinate ring of
. Consider the natural homomorphism
Pol Mat

which assigns to the bracket polynomial . By Theorem
2.1, the
SL
image of this homomorphism is equal to the subring Pol Mat
of invariant
polynomials.
Theorem 2.2. The kernel of is equal to the homogeneous ideal of the Grass
mann variety Gr in its Plucker embedding.
22
be the dense open subset of the affine space Mat formed

by matrices of maximal rank . Consider the map Mat
given by assigning to Mat the values of the bracket

Spec
functions on . Clearly, the corresponding map of the rings of reg
ular functions coincides with . Also it is clear that the image of is contained
in the affine cone Gr
over Gr . The composition of and the canonical projection Gr
Gr is surjective. Let be a homogeneous
polynomial from Ker . Then its restriction to is zero, and hence, since it is
homogeneous, its restriction to the whole of Gr
is zero. Thus belongs
to . Conversely, if belongs to , its restriction to is zero, and hence
because Mat is surjective. Since Gr is a projective

subvariety, is a homogeneous ideal (i.e. generated by homogeneous polyno
mials). Thus it was enough to assume that is homogeneous.
Proof. Let Mat
SL
Gr

The symmetric group
acts naturally on Gr by permuting the coordinates in the space . This corresponds to the action of
on the columns of
matrices from Mat . Let be the subgroup of diagonal matrices in SL . It
acts naturally on Gr by scalar multiplication of columns. Let Gr

be the subspace generated by the cosets of homogeneous polynomials of degree
. Applying Corollary 2.1 and Corollary 2.2, we obtain
Corollary 2.4. Let . Then

Pol Pol SL Gr
Corollary 2.3.
2.4
Pol Mat
The straightening algorithm
We now describe a simple algorithm which allows one to construct a basis of the
space Tab .
Definition. A tableau on the set
is called standard if
of size
..
.

.
. ..

..
for every and .
2.4. THE STRAIGHTENING ALGORITHM
For example,
is standard but
23
is not.
Theorem 2.3. The tableau functions corresponding to standard tableaux form

a basis of the space Tab .
Proof. We will describe the straightening law due to A. Young. It is an algorithm

which allows one to write any tableau function as a linear combination of tableau
functions corresponding to standard tableaux.
We will use the following relation between the bracket functions:
q
(2.5)
Here and are two fixed increasing sequences of num

bers from the set and we assume that in the bracket function ,
the sequence is rearranged to be in increasing order or equal to
zero if two of the numbers are equal.
This relation follows from the observation that the left-hand side, considered
of Mat formed by the columns with
as a function on the subspace
indices , is -multilinear and alternating. Since the exterior power

equals zero, the function must be equal to zero.
is not standard. By permuting the rows of
Suppose a tableau
function
for all . Let be the smallest index such that

we
can assume that
for some . We assume that for . We call the pair

with this property the mark of . Consider equation (2.5) corresponding to
the sequences

Here we assume
that
is put in increasing order. It allows us
the
second sequence
to express as a sum of the products
24
Substituting
this in the product of the bracket functions corresponding to the
rows of , we express as a sum of the such that the mark of each is

greater than the mark of (with respect to the lexicographic order). Continuing
in this way we will be able to write as a sum of standard tableau functions.
This shows that the standard tableau functions span the space Tab . We
skip the proof of their linear independence (see, for example, [48], p. 381).
Corollary 2.5. The homogeneous ideal defining Gr in its Plucker em-
bedding is generated by the quadratic polynomials
where
are increasing sequences of numbers

from the set .
Proof. It is enough to show that any homogeneous polynomial from can
be expressed as a polynomial in the . Let
be the ideal generated by the
polynomials the . It follows from the straightening algorithm that, modulo

, the polynomial is equal to a linear combination of monomials which are
mapped to standard tableau functions in the ring Mat . Since the standard
.
tableau functions are linearly independent, we obtain that

Remark 2.1. The equations defining the Grassmannian Gr are
called the Plucker equations. Corollary 2.3 implies that the Plucker equations describe the basic relations between the bracket functions. This result is sometimes
referred to as the Second Fundamental Theorem of Invariant Theory.
Now we are in business and finally can compute something. We start with the
case . Let us write any degree homogeneous standard tableau in the form
..
..
.
.
where denotes a column vector with coordinates equal to . Let be the

length of this vector. It is clear that

2.4. THE STRAIGHTENING ALGORITHM
25
So if we set , then a standard tableau is determined
by a point with integer coordinates inside of the convex polytope in
defined by the inequalities

Example 2.1. Let . We have
The first nontrivial case is

. We have the unique solution for which
the corresponding standard tableau is
The only nontrivial permutation of two letters changes to

Pol Pol
Next is the case
SL
. Thus
. We have the following solutions:

The corresponding standard tableaux are
Let us see how the group acts on the space Tab
generated by the transpositions . We have
hom.
The group
is
(2.6)
26
By the straightening algorithm,

so that
Similarly, we get
This implies that any -invariant

t combination
of the standard tableau functions
must be equal to
, where
is spanned by

We leave to the reader to verify that this expression is equal to symb , where

(2.7)
This gives that Tab
This is the discriminant of the cubic polynomial
27
Our proof of the First Fundamental Theorem based on the use of the omegaoperator (the Cayley -process) is borrowed from [108]. The -process is also
discussed in [7], [83], [113]. A proof based on the Capelli identity (see the exercises below) can be found in [64], [121]. Another proof using the theory of
representations of the group GL can be found in [18] and [64]. Theorem
2.1 is concerned with invariant polynomial functions on -vectors in a vector
space with respect to the natural representation of SL in . One can
generalize it by considering polynomial functions in vectors in and cov
ectors, i.e. vectors in the dual space . The First Fundamental Theorem
asserts
that the algebra of SL -invariant polynomials on is generated by the bracket functions on the space , bracket functions on the space
, and the functions , whose value at
is equal to . The proof can
be found in [18], [64], [121]. One can also find there a generalization of Theorem
2.1 to invariants with respect to other subgroups of GL .
There is a vast amount of literature devoted to the straightening algorithm and
its various generalizations (see, for example, [17]). We followed the exposition
from [48]. It is not difficult to see that the Plucker equations define set theoretically the Grassmann varieties in their Plucker embedding (see, for example, [40]).
Corollary 2.5 describes the homogeneous ideal of the Grassmannian. As far as I
know the only textbook in algebraic geometry which contains a proof of this fact
is [48]. We refer to [33] for another proof based on the representation theory.
Exercises

2.2 Let be the omega-operator in the polynomial ring . Prove that

for negative integers ,
(i)
(ii)
,
(iii) the function is a solution of the differential equation
in the ring of formal power series .

2.3 For each define the operator acting in Pol Mat by the for
.
mula

2.1 Prove that for any two polynomials
28
(i) Prove that the operators commute with each other and commute with
if .
(ii) Check the following identity (the Capelli identity):
id

id

..
..
..
..
.
.
.
.
id

if

if
2.4 Using the Capelli identity show that the operator Pol Pol
Pol Pol defined by , where symb symb is well
defined and transforms an SL -invariant to an SL -invariant.
2.5 Show that Pol Pol SL is spanned by the catalecticant invariant from
Example 1.4 in Chapter 1.
2.6 Show that Pol Pol SL is generated (as a -algebra) by the discriminant
invariant from Example 2.1.

2.7 Show that Pol Pol SL is equal to , where Pol is the
discriminant of quadratic form. Find symb .
2.8 Let O be the orthogonal group of the vector space equipped with
the standard inner product. Consider
left multipli the action of on Mat by
O
cation. Show that Pol Mat
is generated by the functions whose value
on a matrix is equal to the dot-product of the th and th columns.

2.9 With the notation from

the preceding exercise let O O SL .
Show that Pol Mat O is generated by the functions and the bracket func
tions.

2.10 Show that the field of fractions of the ring Pol Mat SL is a purely tran
scendental extension of of transcendence degree
.
Chapter 3
Reductive algebraic groups
3.1
The GordanHilbert Theorem
In this chapter we consider a class of linear group actions on a vector space for
which the algebra of invariant polynomials Pol is finitely generated. We start
with the case of finite group actions.
Theorem 3.1. Let be a finite group of automorphisms of a finitely generated
-algebra . Then the subalgebra is finitely generated over .
Proof. This follows easily from standard facts from commutative algebra. First
we observe that is integral over . Let be generators of . Let
be the subalgebra of generated by the coefficients of the monic polynomials
such that . Then is a finite -module.
Since is noetherian, is also a finite -module. Since is finitely generated
over , must be finitely generated over .
Let us give another proof of this theorem in the special case when the order
of is prime to the characteristic of and acts on Pol via its
linear action on . In this case leaves invariant the subspace of homogeneous
polynomials of degree so that
Pol
Pol
Let be the ideal in generated by invariant polynomials vanishing at (or,

equivalently, by invariant homogeneous polynomials of positive degree). Applying the Hilbert Basis Theorem, we obtain that the ideal is finitely generated by
29
CHAPTER 3. REDUCTIVE ALGEBRAIC GROUPS
30
a finite set of polynomials
mogeneous of degree
we can write
in . We may assume that each is ho . Then for any homogeneous of degree
for some homogeneous polynomials of degree

operator av defined by the formula
av
Clearly,
id
av
(3.1)
Now consider the
av
Applying the operator av to both sides of (3.1) we get
av
av
By induction we can assume that each invariant homogeneous polynomial of de

gree can be expressed as a polynomial in s. Since av is homogeneous
of degree
, we are done.
Let us give another application of the Hilbert Basis Theorem (it was proven
by Hilbert exactly for this purpose):
Theorem 3.2. (GordanHilbert) The algebra of invariants Pol Pol

finitely generated over .
SL
is
Proof. Let Pol . The proof uses the same idea as the one used in the
second proof of Theorem 3.1. Instead of the averaging operator av we use the
omega-operator . Let Pol SL . Write
SL . By the proof of Proposition

for some Pol
and Pol
1.1 there exists an integer such that, for any ,

The number is called the weight of .
Now, for a general matrix , we have the identity of functions on GL :
3.1. THE GORDANHILBERT THEOREM
31
Now let us apply the omega-operator to both sides times. We get
where is a nonzero constant. Now the assertion follows by showing that the
value of
at is an invariant and using induction on the
degree of the polynomial.
Lemma 3.1. For any
Then
Pol
let
is either zero or an invariant of weight

.
Proof. This is nothing more than the change of variables in differentiation. Let
be a general square matrix of size . We have
Here
denotes the omega-operator in the ring corresponding to the determinant of the matrix where . We use
the formula
r

(3.2)

in the variables . This easily follows from the differfor any polynomial
entiation rules and we leave its proof to the reader. Now plugging in in (3.2)
(although it is not in GL the left-hand side extends to the whole polynomial
ring in the matrix entries) we obtain
This proves the assertion.

Remark 3.1. In fact, the same proof applies to a more general situation when
GL acts on a vector space by means of a rational linear representation (see
the definition of a rational representation in the next section). We have
to use that

SL
in this case
.
for any GL and Pol
32
Remark 3.2. The proof shows that the algebra of invariants Pol SL is generated by a finite generating set of the ideal generated by invariant
homogeneous polynomials of positive degree. Let
be the subset
of common zeros of . Let be the ideal in Pol SL of all polynomials vanishing on . By Hilberts Nullstellensatz, for each there
exists a positive integer such that . Let be homogeneous
generators of . Let be the largest of the degrees of the and be the largest of
the numbers . Then it is easy to see that any invariant homogeneous polynomial
of degree
expressed as a polynomial in q . This implies
can be
that the ring Pol SL is integral over the subring q generated

by q . In fact, it can be shown that it coincides with the integral closure of q in the field of fractions of Pol (see, for example, [113],
Corollary 4.6.2). In Chapter 9 we will learn how to describe the set (it will be
identified with the null-cone) without explicitly computing the ring of invariants.
This gives a constructive approach to finding the algebra of invariants.
3.2
The unitary trick
Let us give another proof of the GordanHilbert Theorem using another device
replacing the averaging operator av due to A. Hurwitz (later called the unitary
trick by H. Weyl). We assume that .
and
SU be its subgroup of unitary matrices. Let
Let SL
act on Pol via its linear representation GL .
Lemma 3.2. (Unitary trick)

Pol
Proof. Let
defined by
Let
Pol
For any
be the function on
Pol
Mat consider the function on

defined by

d

r
Since
we see that for all if and
only if dd for all

and all . The latter is equivalent to the
3.2. THE UNITARY TRICK
33
condition that e for all

and all . Let denote the
space of complex matrices of size with zero trace. Since any SL
can be written as e for some , we see that the condition
(3.3)
is equivalent to being invariant. Next we easily convince ourselves (by using

the chain rule) that the map is linear, so it is enough to check (3.3)
for the set of the which spans . Consider a basis of formed by
the matrices
where . Observe that the same matrices form a basis over of the
subspace of formed by skew-hermitian matrices (i.e. satisfying
t
). Now we repeat the argument replacing by SU . We use
that any can be written in the form e for some . We find that
Pol if and only if
for all . Since the properties
for all and

for all
are equivalent
we are done.
The group
smooth manifold. If and
SU is a compact

, where are real, then is a closed and a bounded

submanifold of defined by the equations
r s r s t
where r s is the Kronecker symbol. This allows one to integrate over it. We
consider any polynomial complex valued function on as a restriction of a poly
nomial function on GL
. For each such function set

d
av
d
where d

d d
Lemma 3.3. For any Pol
the function defined by

av
is -invariant.
34
Proof. For any matrix

with we have
let

and

For any
Here we use block-expressions of these matrices. It is easy to see that
is an orthogonal real matrix of size . Thus the jacobian of the change
of variables s is equal to
. Since is known to be a
connected manifold, the function
is constant; it takes the value at
, so
. Applying the formula for the change of variables in the
integration we get
d
d
d
hence
av
av
d
d
One can generalize the preceding proof to a larger class of groups of complex matrices. What is important in the proof is that such a group contains a
compact subgroup such that the complex Lie algebra of is isomorphic to the
complexification of the real Lie algebra of . Here are examples of such groups,
their compact subgroups, and their corresponding Lie algebras:
GL
SU
O
Lie Mat

Lie
Lie t
Lie
t
These groups satisfy the following property
3.3. AFFINE ALGEBRAIC GROUPS
35
(LR) Let GL be a homomorphism of complex Lie groups, and

Then there exists an invariant subspace such that .
Or, in other words, there exists a -invariant linear function on such
that .
One checks this property by first replacing with its compact subgroup as
above. Taking any linear function with we average it by integration
over to find a nonzero -invariant function with the same property. Then we
apply Lemma 3.3 to ensure that is -invariant.
3.3
Affine algebraic groups
Next we observe that property (LR) from the preceding section can be stated over
any algebraically closed field . Instead of complex Lie groups, we will be dealing
with affine algebraic groups over .
Definition. An affine algebraic group over a field is an affine algebraic variety
over with the structure of a group on its set of points such that the multiplication
map and the inversion map are regular maps.

Although we assume that the reader is familiar with some rudiments of algebraic geometry, we have to fix some terminology which may be slightly different
from the standard textbooks (for example, [102]). We shall use an embeddingfree definition of an affine algebraic variety over an algebraically closed field
. Namely, a set Specm of homomorphisms of a finitely generated algebra without zerodivisors to . The algebra is called the coordinate algebra of and is denoted by (or ). An element can be considered
as a -valued function on whose value at a point is equal to .
Functions on of this form are called regular functions. A point is uniquely
determined by the maximal ideal of functions vanishing at . A choice of
generators of defines a bijection from to a subset of the affine
space Specm identified naturally with the set . This subset

is equal to the set of common zeros of the ideal of relations between the generators. A regular map (or morphism) of affine algebraic varieties is
defined as a map given by composition with a homomorphism of the coordinate
algebras . This makes a category of affine algebraic varieties
over which is equivalent to the dual of the category of finitely generated domains
over . This latter category has direct products defined by the tensor product of
36
-algebras. A subset of of homomorphisms vanishing on an ideal of

is called a closed subset. It can be identified with an affine algebraic variety
, where
rad is the radical of . A point is a closed subSpecm
set corresponding to the maximal ideal of . Closed subsets define a topology
, form a
on , the Zariski topology. Open subsets qq
basis of the topology. Each subset can be identified with an affine algebraic
variety Specm .
A choice of generators of the -algebra defines an isomorphism
from to a closed subset of the affine space . A morphism of affine varieties Specm
Specm corresponding to a surjective homomorphism
of -algebras defines an isomorphism from Specm to a closed subset
of Specm . It is called a closed embedding.
The multiplication and the inversion morphisms defining an affine algebraic group can equivalently be given by homomorphisms of -algebras
which are called the comultiplication and the coinverse.

For any -algebra we define the set of -points of to be the set of
homomorphisms of -algebras . In particular, if
for some
affine algebraic variety , the set

can be identified naturally with the set of
morphisms from to .
Here are some examples of affine algebraic groups which we will be using in
the book.

(a) GL Specm
(a general linear group over
):
GL
where
(b)
is equal to the th entry of the inverse of the matrix .
GL Specm (the multiplicative group over ):

Specm (the additive group over ):
(c) r

GL
37
Other examples of affine algebraic groups can be realized by taking direct products or by taking a closed subvariety which is an affine algebraic group with respect to the restriction of the multiplication and the inverse morphisms (a closed
subgroup). For example, we have
(d) (an affine torus over ),
(e) SL (a special linear group over ).

Affine algebraic groups over form a category. Its morphisms are morphisms
of affine algebraic varieties which induce homomorphisms of the corresponding
group structures. One can prove that any affine algebraic group admits a morphism to the group GL such that it is a closed embedding. In other words, is
isomorphic to a linear algebraic group, i.e., a closed subvariety of GL whose
-points for any -algebra form a subgroup of GL
. If no confusion arises,
we will also drop the subscript in the notation of groups GL , and so on.
From now on all of our groups will be linear algebraic groups and all of our
maps will be morphisms of algebraic varieties.
We define an action of on a variety to be a regular map
satisfying the usual axioms of an action (which can be expressed by the commutativity of some natural diagrams). We call such an action a rational action or, better,
a regular action. In particular, a linear representation GL GL
will be assumed to be given by regular functions on the affine algebraic variety .
Such linear representations are called rational representations.
Let an affine algebraic group act on an affine variety Specm . This
action can be described in terms of the coaction homomorphism

where is the coordinate ring of . It satisfies a bunch of axioms which are
dual to the usual axioms of an action; we leave their statements to the reader.
For any we have

. An element is a homomorphism
and we set

(3.4)
This defines a rational action of on a -algebra , that is, a morphism
Aut . We will continue to denote the subalgebra of invariant elements by .
where

An important property of a rational action is the following.
38
Lemma 3.4. For any , the linear subspace of spanned by the translates
is finite-dimensional.
Proof. This follows immediately from equation (3.4). The set of elements
spanning set.
Note that not every homomorphism of groups
rational action of on .
Aut
is a
arises from a
Example 3.1. Let

on an affine algebraic variety Specm . Let
act

be the corresponding coaction homomorphism. For any
we can write
(3.5)
It is easy to see, using the axioms of an action, that the maps

are the projection operators, i.e., . Denoting the image by we
have
and
(3.6)

This defines a grading on . Conversely, given a grading of , we define by

, where is the th graded part of . This gives a geometric
interpretation of a grading of a commutative -algebra.
Assume now that grading (3.5) on satisfies for and .

Such a grading is called a geometric grading and the corresponding action is called
is a maximal ideal of
a good -action. In this case, the ideal
and hence defines a point of , called the vertex. We set
Specm
Specm
The group
acts on the open set ; the quotient set is denoted by Projm
and is called the projective spectrum of . Assume that is a finitely generated
-algebra with a geometric grading. Choose a set of its homogeneous generators
. If for some , then any acts on by sending
to . Use the generators to identify with a closed subset of defined
by the homogeneous ideal of relations between . The vertex of
. We obtain a natural bijection from Projm to the
becomes the origin in
, where acts by
set Specm
(3.7)
39
In the special case when are algebraically independent (i.e.,

so that with grading defined by , the set
),

is called the weighted projective space with weights . When all the
are equal to 1, we obtain the usual definition of the -dimensional projective space
.
Let be the closed subgroup of
Specm defined by the
ideal
. As an abstract group it is isomorphic to the group of th roots
of 1 in . Let be a graded -algebra and

Aut be the corresponding

Projm
action. It follows from the definition that
The inclusion defines a natural map Specm

which coincides with the quotient map for the action of
that for any ). Let

act on Specm
grading defined by
Specm
on Specm (use

with respect to the
(3.8)
Then
Projm
Specm
Specm
Specm
Projm
It is known that for any finitely generated geometrically graded -algebra there
exists a number such that is generated by elements of degree 1 with respect

to the grading defined by (3.8) (see [9], Chap. III, 1). This implies that Projm
is bijective to a subset of some equal to the set of common zeros of a

homogeneous ideal in the ring of polynomials with the standard
grading.
One can make this statement more precise by defining the category of projective varieties. First of all we notice that for any nonzero homogeneous element

, the subset of Specm of all points not vanishing on does not
contain the vertex and is invariant with respect to the action of
defining the
grading. Since any ideal in is contained in a homogeneous ideal of , the union
of the sets is equal to Specm . So Projm is equal to the union of the
40
subsets
on
of
. If we identify with Specm , the action

corresponds to the (not necessarily geometric) grading defined by
Let r . It is called the homogeneous

localization of
the graded ring with respect to . Any element of

can be written
uniquely in the form . This implies that the image of any point
in is determined by its restriction to . Thus, any point in

is uniquely determined by a homomorphism . This shows that we can
identify with Specm . Since the union of sets of the form is

the whole set Projm , we can define a topology on Projm in which an open
set is a set whose intersection with any set is an open set in its Zariski
topology. The open subsets form a basis of the topology.

A quasi-projective algebraic variety over is defined to be a locally closed
subset (i.e., the intersection of an open subset with a closed subset) of some
Projm . A closed subset is called a projective variety over . For any open subset of Projm we define a regular function on as a function such
that its restriction to any subset is a regular function. Regular functions on form a -algebra which we will denote by . Let Projm
and Projm be two quasi-projective algebraic varieties over . A morphism is defined to be a continuous map from to (with respect
to the induced Zariski topologies) such that for any open subset
and any
, the composition is a regular function on .

For example, any surjective homomorphism of graded algebras
preserving the grading (the latter will be always assumed) defines a closed embedding Specm Specm whose restriction to any subset is a closed
embedding of affine varieties. It corresponds to the homomorphism
. This defines a closed embedding from to and a morphism Projm Projm . In particular, a choice of homogeneous generators of degrees of defines a morphism Projm
which is a closed embedding (i.e., an isomorphism onto a closed subset of the
target space).
One can show (see Exercise 3.6) that any projective algebraic variety is isomorphic to some Projm . Any affine algebraic variety is isomorphic to a quasi
projective algebraic variety because the affine space is isomorphic to an open
subset of
Projm whose complement is the closed subset defined by the ideal . Thus any locally closed subset of an affine variety
3.4. NAGATAS THEOREM
41
is a quasi-projective algebraic variety. We will employ topological terminology

dealing with the Zariski topology of a quasi-projective variety. For example, we
can speak about irreducible, connected quasi-projective algebraic varieties. We
refer the reader to textbooks in algebraic geometry for the notion of a nonsingular
quasi-projective variety.
Note that an algebraic group is irreducible if and only if it is connected; this
follows from Exercise 3.2.
Even when we study rational actions of an algebraic group on an affine algebraic varieties we have to deal with nonaffine quasi-projective algebraic varieties.
Example 3.2. Let be a rational action of an affine algebraic
group on an affine algebraic variety . For any point , we have a regular

map defined by . The fiber of this map over the
point is a closed subgroup of , called the stabilizer subgroup of . It is an
affine algebraic group. The image O of this map is a subset of , called the
orbit of , which is not necessarily closed. However, if is irreducible, the orbit
O is a locally closed subset of , and hence is a quasi-projective algebraic
variety. It follows from the Chevalley Theorem (see [46], p. 94), that the image of
a regular map is a disjoint finite union of locally closed subsets. However, since
is irreducible, the image is irreducible and hence must be a locally closed subset,
i.e., a quasi-projective variety. Of course, the image of an affine variety is not
always affine.
Example 3.3. Let be a closed subgroup of an algebraic group . Consider the

space spanned by the -translates of generators of the ideal defining . By
Lemma 3.4 is finite-dimensional of some dimension . Let and
. Then acts rationally on the Grassmannian variety Gr of

-dimensional subspaces of . One can show that is the subgroup of which
. Thus we can identify the quasi-projective algebraic variety
fixes Gr
O
Gr
with the set of conjugacy classes .
3.4
Nagatas Theorem
Our goal is to prove the following theorem of M. Nagata

Theorem 3.3. Let be a geometrically reductive group which acts rationally on
an affine variety Specm . Then is a finitely generated -algebra.
Let us first explain the notion of a geometrically reductive group.
42
Definition. A linear algebraic group is called linearly reductive if for any rational representation GL and any nonzero invariant vector there
exists a linear -invariant function on such that .
The unitary trick shows that GL and SL and their products are linearly reductive groups over . This is not true anymore for the same groups defined over
a field of characteristic . In fact, even a finite group is not linearly reductive
if its order is not coprime to the characteristic. However, it turns out (Haboushs
Theorem, [44]) that all these groups are geometrically reductive in the following
sense.
Definition. A linear algebraic group is called geometrically reductive if for any
rational representation GL and any nonzero invariant vector there
exists a homogeneous -invariant polynomial on such that .
In fact, one can define the notion of a reductive algebraic group over any field
which will include the groups GL SL O and their products and Haboushs
Theorem asserts that any reductive group is geometrically reductive. We are not
going into the proof of Haboushs Theorem, but let us give the definition of a
reductive affine algebraic group (over an algebraically closed field) without going
into details.
A linear algebraic group is called an algebraic torus (or simply a torus)
if it is isomorphic to . An algebraic group is called solvable if it admits a
composition series of closed normal subgroups whose successive quotients are

abelian groups. Each algebraic group contains a maximal connected solvable
normal subgroup. It is called the radical of . A group is called reductive if its
radical is a torus. A connected linear algebraic group is called semisimple if its
radical is trivial.
Each semisimple group is isomorphic to the direct product of simple algebraic
groups. A simple algebraic group is characterized by the property that it does not
contain proper closed normal subgroups of positive dimension.
There is a complete classification of semisimple affine algebraic groups. Examples of simple groups are the classical groups
SO type

.
There are also some simple groups of exceptional type of types
SL
type SO type Sp type
Every simple algebraic group is isogeneous to one of these groups (i.e., there
exists a surjective homomorphism from one to another with a finite kernel).
We shall start the proof of Nagatas Theorem with the following.
3.4. NAGATAS THEOREM
43
Lemma 3.5. Let a geometrically reductive algebraic group act rationally on a

-algebra leaving an ideal invariant. Consider as a subalgebra of
by means of the injective homomorphism induced by the inclusion
. For any there exists such that . If is
linearly reductive then can be taken to be .
Proof. Let be a nonzero element from , let be its representative in

and let Let be the -invariant subspace of spanned by
the -translates of . By Lemma 3.4 is finite-dimensional and is contained in
the subspace spanned by the s. Let . We have
, where . This shows that any can be
written in the form
for some
We have
and

. Let be the linear map defined by

. This implies that , and, in
for some
particular, the linear map is -invariant. Consider it as an element of
the dual space . The group acts linearly on and is a -invariant element.
Choose a basis of with , and for . Then
we can identify with the affine space by using the dual basis, so that
. By definition of geometrical reductiveness, we can find a -invariant
homogeneous polynomial of degree such that

. We may assume that . Now we can identify with the linear
polynomial , hence belongs to the
ideal of generated by
of belongs to ,
. Since each generator
we see that
modulo . Since (because
is -invariant), we are done.
Now we are ready to finish the proof of Nagatas Theorem. To begin, by
noetherian induction, we may assume that for any nontrivial -invariant ideal
the algebra is finitely generated.
Assume first that is a geometrically graded -algebra (i.e.,
) and that the action of preserves the grading. For example, could be
a polynomial algebra on which acts linearly. The subalgebra inherits the
grading. Suppose is an integral domain. Take a homogeneous element
of positive degree. We have since, for any ,
44

implies that .
integral over
Since is finitely generated and

(Lemma 3.5), we obtain that is
finitely generated. Hence its maximal ideal generated by elements
of positive degree is finitely generated. If we take the set of representatives of its
generators and add to this set, we obtain a set of generators of the ideal
in . But now, using the same inductive (on degree) argument as in the second
proof of Theorem 3.1, we obtain that is a finitely generated algebra.
Now assume that contains a zero-divisor . Then and the annihilator
are nonzero -invariant ideals. As
ideal
above, and are finitely generated. Let be the subring

of generated by representatives of generators of both algebras. It is mapped
surjectively to and . Let be representatives
-module. Since
in of generators of as a
for all , we get , i.e., . Let us show that
. Then we will be done. If , we can find t
such that t (since is mapped surjectively to ). Then
is -invariant implies that . Thus . This implies
t
t
t

as we wanted.
So we are done in the graded case.
Now let us consider the general case. Let be generators of . Consider the -vector space
spanned by -translates of the . It follows from
Lemma 3.4 that is finite-dimensional. Without loss of generality we may as
sume now that is a basis of this space. Let
be the surjective homomorphism defined by . The group acts on
linearly by q , where q . Let be the kernel of .

It is obviously -invariant. We obtain that . By Lemma 3.5, is
integral over . Since we have shown already that is finitely generated, we are almost done (certainly done in the case when is linearly reductive).
By a previous case we may assume that has no zerodivisors. A result from
commutative algebra (see, for example, [26], Corollary 13.3) gives that the inte
gral closure of in the field of fractions of is a finitely
generated -algebra provided that is a finite extension of the field of frac
tions of . Since is integral over this would imply that is
finitely generated (see [26], Exercise 4.3.2). Thus it is enough to show that the
field is a finite extension of the field of fractions of q . Since
is integral over this ring, it is enough to show that is finitely generated as
a field. If is a domain this is obvious (a subfield of a finitely generated field
is finitely generated). In the general case we use the total ring of fractions of ,
45
the localization with respect to the set of nonzerodivisors. For any maximal
ideal of we have since is a domain. This shows that the
field of fractions of is a subfield of . But the latter is a finitely generated
. The proof is now complete.
field equal to the field of fractions of
In the next chapter we will give an example (due to M. Nagata) of a rational linear representation GL of a linear algebraic group such that
Pol is not finitely generated.
The algebra of invariants , where is a reductive algebraic group and is

a finitely generated algebra, inherits many algebraic properties of . We shall not
go into this interesting area of algebraic invariant theory; however, we mention
the following simple but important result.
Proposition 3.1. Let be a reductive algebraic group acting algebraically on a
normal finitely generated -algebra . Then is a normal finitely generated
algebra.
Proof. Recall that a normal ring is a domain integrally closed in its field of fractions. Let be the field of fractions of . It is clear that the field of fractions
of is contained in the field of -invariant elements of . We have to
check that the ring is integrally closed in . Suppose satisfies a monic
equation

with coefficients . Since is normal,

and the assertion
is verified.
The proof of the GordanHilbert Theorem follows the original proof of Hilbert
(see [47]). The proof using the unitary trick can be found in [63], [108], and
[121]. The original proof of Nagatas Theorem can be found in [77]. Our proof
is rather close to the original one. It can be found in [31], [73], [80], and [109]
as well. Haboushs Theorem was a culmination of efforts of many people. There
are other proofs of Haboushs Theorem with more constraints on a group (see a
survey of these results in [73], p. 191).
A good introduction to Lie groups and Lie algebras can be found in [34] or
[84] and [6]; [110], [52] are excellent first courses in algebraic groups.
46
We refer to [89], 3.9 for a survey of results in the spirit of Proposition 3.1.
An interesting question is when the algebra Pol , where is a rational linear representation of a reductive group , is isomorphic to a polynomial algebra.
When is a finite group, a theorem of Chevalley [11] asserts that this happens if
and only if the representation of in is equivalent to a unitary representation
where acts as a group generated by unitary reflections. The classification of
such unitary representations is due to Shephard and Todd ([105]). The classification of pairs with this property when is a connected linear algebraic
group group is known when is simple, or when is semisimple and is its irreducible representation. We refer to [89], 8.7 for the survey of the corresponding
results.
Exercises
3.1 For any abstract finite group construct an affine algebraic -group such that
its group of -points is equal to for any .
3.2. Prove that any affine algebraic group is a nonsingular algebraic variety.
3.3 Show that there are no nontrivial homomorphisms from to r , or in the
other direction.
3.4 Prove that a finite group over a field characteristic is linearly reductive
if and only if its order is prime to . Show that such is always geometrically
reductive.
3.5 Give an example of a nonrational action of an affine algebraic group on an
affine space.
3.6 Prove that any closed subset of Projm is isomorphic to Projm , where
is a homogeneous ideal of .
3.7 Let GL act on Pol via its linear representation in . A polynomial
Pol is called a projective invariant
of weight if, for any and
. Let Pol be the space of projective

any ,
invariants of weight . Show that the graded ring
is finitely generated.
Pol
Chapter 4
Hilberts Fourteenth Problem
4.1
The problem
The assertions about finite generatedness of algebras of invariants are all related
to one of the Hilbert Problems. The precise statement of this problem (number 14
in Hilberts list) is as follows.
Problem 1. Let be a field, and let be its purely transcendental

extension, and let be a field extension contained in . Is the finitely generated?
algebra s
Hilbert himself gave an
affirmative answer to this question in the situation
when
SL where SL acts linearly on (Theorem
3.2 from Chapter 3). The subalgebra is of course the subalgebra
of invariant polynomials SL . A special case of his problem asks
whether the same is true for an arbitrary group acting linearly on the ring of
polynomials. A first counterexample was given by M. Nagata in 1959; we shall
explain it in this chapter. For the reader with a deeper knowledge of algebraic
geometry, which we assume in this book, we give a geometric interpretation of
Hilberts Fourteenth Problem due to O. Zariski.
we can find a normal irreducible algebraic
For any subfield
variety over with field of rational functions isomorphic to . The
inclusion of the fields gives rise to a rational map
Let
subset of
be the closure of the graph of the regular map of the largest open
on which
is defined. Let be the hyperplane at infinity in and
47
CHAPTER 4. HILBERTS FOURTEEN PROBLEM
48
pr pr . This is a closed subset of . By blowing up, if necessary,
we may assume that is the union of codimension 1 irreducible subvarieties .

Let be the Weil divisor on equal to the sum of components such that
; note that could be the zero divisor. Thus for any rational
pr pr
function is regular on
if and only if has poles only
along the irreducible components of . Let be the linear subspace of
such that div
which consists of rational functions
. After
identifying with and

with
(by means of ), we
see that s

is isomorphic to the subalgebra
of . So the problem is reduced to the problem of finite generatedness of the
algebras where is any positive Weil divisor on a normal algebraic variety

.
Assume now that is nonsingular. Then each Weil divisor is a Cartier divisor
for some rational function
and hence can be given locally by an equation
on regular on some open subset
functions must satisfy
on for some . These
. We can

take them to be the

transition functions of a line bundle . Rational functions with poles along
for some . This implies that the functions
must satisfy
satisfy , hence form a section of the line bundle . This shows
that the algebra is equal to the union of the linear subspaces of

the field . Let

Recall that we can view as the space of regular functions on the line
bundle whose restrictions to fibers are monomials of degree . This allows
one to identify the algebra with the algebra . Let be the variety
obtained from by adding the point at infinity in each fiber of . More
precisely, let
be the trivial line bundle. Then the variety can be constructed
as the quotient of the rank 2 vector bundle with the deleted zero
section by the group

acting diagonally on fibers; here the direct sum means
that the transition functions of the vector bundle are chosen to be diagonal matrices

4.2. THE WEITZENBOCK
THEOREM
49
Then we obtain that is equal to the ring where
infinity in . In this way we are led to the following.
is the divisor at
Problem 2. (O. Zariski) Let be a nonsingular algebraic variety and let
an effective divisor on . When is the algebra finitely generated?
be
It can be shown that Nagatas counterexample to the Hilbert problem is of the
form (see Exercise 4.3). It turns out that the algebras are often not
finitely generated. However, if we impose certain conditions on (for example,
that the complete linear system defined by has no base points) then
is finitely generated. One of the fundamental questions in algebraic geometry is
the question of finite generatedness of the ring , where is the canonical

divisor of . This is closely related to the theory of minimal models of algebraic
varieties (see [69]).
4.2
The Weitzenbock Theorem
Let us first discuss the case of algebras of invariants of algebraic groups that are
not necessarily reductive. We will later give an example of Nagata which shows
that is not finitely generated for some nonreductive group . Notice that
according to a result of V. Popov ([87]), if is not finitely generated for some
action of on an affine algebraic variety with , then is not
reductive. In fact, the proof of this result relies on Nagatas counterexample.
Since any affine algebraic group is a closed subgroup of a reductive group
, we may ask how the rings and are related. First of all we have the
following (see [41], [89]).
Lemma 4.1. Let an affine algebraic group act on a finitely generated -algebra
. Then

Here
acts on by left multiplication and acts on itself by right multiplication.
Proof. Let

that
Specm be the affine algebraic variety with . Let

s . Assume . This means
for any r . Let . Then

50
This shows that
satisfies
. Conversely, if
, the function

. We leave to the reader to check that the maps
Thus

are inverse to each other.

Corollary 4.1. Assume that a rational action of on an affine variety extends
to an action of a geometrically reductive group containing and also assume
that is finitely generated. Then is finitely generated.
The algebra can be interpreted as the algebra of regular functions on

the quasi-projective algebraic variety (see Example 3.3). It could be affine,
for example when is a reductive subgroup of a reductive group . It also could
be a projective variety (for example, when GL and contains the subgroup
of upper triangular matrices, or more generally, when is a parabolic subgroup
of a reductive group ). A closed subgroup of affine algebraic group is
called observable if is quasi-affine (i.e., isomorphic to an open subvariety
of an affine variety). An observable subgroup is called a Grosshans subgroup
if is finitely generated.
Theorem 4.1. Let be an observable subgroup of a connected affine algebraic

group . The following properties are equivalent:
(i)
is a Grosshans subgroup;
(ii) there exist a rational linear representation of in a vector space of finite

dimension and a vector such that and the orbit of is
of codimension
in its closure .
Proof. (i) (ii) Let and let Specm . is an irreducible

algebraic variety on which acts (via the action of on ). Consider the canoni
cal morphism such that is

the identity. Since is isomorphic to an open subset of an affine variety ,
the restriction map defines a morphism of affine

4.2. THE WEITZENBOCK
THEOREM
51
varieties
such that the composition
. Since is dominant, this easily implies

is the open embedding
that is an open embedding. So we may assume that is an open subset

of and that the restriction homomorphism is bijective. Let
. This is a closed subset of . Since is a nonsingular irreducible
algebraic variety, is a normal affine variety, i.e., the ring is normal. By
Proposition 3.1 the ring has the same property and hence is a normal
affine variety. In particular, is a Krull domain ([9], Chapter VII, 1) and we
can apply the theory of divisors. It follows from the approximation theorem (loc.
cit., Proposition 9) that one can find a rational function on such that it has a
pole only at one irreducible component of of codimension 1. Thus the rational
function is regular on but not regular on . This contradiction shows

that each irreducible component of is of codimension
. Now, by Lemma
3.5, we can embed into affine space in such a way that acts on via a
linear representation. The closure of the -orbit of is a closed subset of

containing , and hence the complement of the orbit in its closure is of
.
codimension
(ii) (i) Let be the closure of the orbit O O . Replacing by its
normalization, we may assume that O is isomorphic to an open subset of a
normal affine algebraic variety with the complement of O of codimension
.
It remains to use that for each such open subset the restriction map
is bijective (see [26]).
Example 4.1. Let SL and be the subgroup of upper triangular matrices
with diagonal entries equal to 1. Obviously, r . In the natural representation

of in the affine plane , the orbit of of the vector is equal to

and the stabilizer subgroup is equal to . Thus is a Grosshans
subgroup of . More generally, any maximal unipotent subgroup of an affine
algebraic group is a Grosshans subgroup (see [41], Thm. 5.6).
Let r . We know that is not geometrically reductive (Exercise 4.1).
However, we have the following classical result.
Theorem 4.2. (Weitzenbocks Theorem) Assume char . Let r
GL be a rational linear representation. Then the algebra Pol
is finitely
generated.
Proof. To simplify the proof let us assume that . We shall also identify
r with its image in GL ; which is isomorphic to . This can be done since
52
does not contain finite nontrivial subgroups in characteristic zero so is either

trivial or injective. Let be a nonzero element. Since there are no nontrivial rational homomorphisms from to , all eigenvalues of must be equal
to 1. Since is commutative, there is a common eigenvector for all .
Consider the induced action of on . Let be a common eigenvector for

for all . Continuing in
all in this space. Then
this way, we find a basis of such that each is represented by a unipotent
matrix . Consider the differential of the homomorphism GL
at the origin. It is defined by , where . Clearly is a

that
nilpotent matrix. Since , it is easy to see
. By changing basis of , we may assume that is a

and hence
Jordan matrix. Let , where corresponds to a Jordan block
of of
size . It is easy to see that the representation of in defined
is isomorphic to the representation of in Pol obtained

by
by restriction of the natural representation of SL in Pol . Here we con
sider as a subgroup of upper triangular matrices in SL . Thus acts on
by the restriction of the representation of SL in the direct sum of linear

representations in Pol . Now we can apply Lemma 4.1. Observe that any
SL can be reduced after multiplication by some to a

matrix of the form
or
Thus any -invariant regular function on SL is uniquely determined by its val

ues on such matrices. Since the set of such matrices forms a subvariety of SL
isomorphic to , the restriction of functions defines an isomorphism
Since , we conclude that SL is finitely generated. So
we can apply Lemma 4.1 to the pair SL and the representation of SL on
Pol to obtain the assertion of the theorem.
4.3
SL
Nagatas counterexample
Now we are ready to present Nagatas counterexample to the Fourteenth Hilbert

Problem.
4.3. NAGATAS COUNTEREXAMPLE
53
Let be the subgroup of r equal to the set of solutions

system of linear equations
We will specify the coefficients later. The group

by the formula
of a
(4.1)
acts on the affine space
Now let us consider the subgroup
of
. It acts on
by the formula
Both of these groups are identified naturally with subgroups of SL and we en

large by considering the group s . The group is contained in the
subgroup of matrices of the form:

(4.2)
..
..
..
.. . .
..
..
..
..
. .
.
.
.
.
.
.
.

Theorem 4.3. For an appropriate choice of the system of linear equations (4.1)
and the number the algebra of invariants
is not finitely generated.

We start the proof with the following:
54
Lemma 4.2. Assume that the determinant of the matrix

to zero. Then

where
Moreover,
are algebraically independent over .
is not equal
Proof. Under the action of , defined by the matrix (4.2) from above, we have
and, since

, we obtain that
. This shows that
the right-hand side is contained in the left-hand side. Using the assumption on
the coefficients , we can write as a linear combination of
to obtain

The first equality shows that are algebraically independent

over , hence are algebraically independent.

Let be the subgroup of defined by the conditions
Obviously it is isomorphic to r . We see that

Continuing in this way, we eliminate to obtain

Now we throw in the torus part which acts on by multiplying it by . It

is clear that any -invariant rational function in with coefficients in
must be equal to a constant. This proves the lemma.
55
Consider now each column of the matrix as the homoge

neous coordinates of a point in the projective plane . Let be the ideal

in generated by homogeneous polynomials with multiplicity

at each point . If char , this means that all partials of of order
vanish at . In the general case, it means the following. By a linear change of
at
variables we may assume that . Then has multiplicity
if considered as a polynomial in all its nonzero coefficients are homogeneous
polynomials in of degree
.
Lemma 4.3.

Proof. By the preceding lemma, . First

notice that, since for , we have

The intersection of the right-hand side with the field is equal to

. Thus
q
Write any invariant homogeneous polynomial as a sum of monomi
als , where and
. Since each is homogeneous

in of degree 1 and
in
of
degree
,
and
in ,

is homogeneous
of degree
we must have
.
This implies that we can write as a sum , where each

is homogeneous in of degree
. Now write
as a polynomial in whose
are polynomials in . Since the degree
coefficients
, we obtain that each is the
of in is equal to

-homogeneous component of , and hence is a polynomial
in .
It remains to show that s if and only if each

. Assume that none of is equal to zero. After a linear change of
if and only if its coefficients as a polynomial
variables, we obtain that

in are homogeneous polynomials in
56
of degree . Since and are both divisible by

, we have
for any polynomial
in
, we see that,
We leave to the reader to prove the converse.

Next, we need a lemma from algebraic geometry.
be an irreducible plane cubic curve in the projecLet
tive plane over an algebraically closed field . It is known that the set of
nonsingular points of has the structure of an algebraic group (in the case when
is nonsingular this can be found for example in [102], Chapter 3, 3). If is
singular, this is easy to see. The normalization of is isomorphic to and the

projection map
is an isomorphism outside one point (a cuspidal cubic) or
two points (a nodal cubic). The complement of one point in is isomorphic to

the affine line, and hence has a structure of an algebraic group isomorphic to the
and
additive group r . The complement of two points is isomorphic to q
has a structure of an affine algebraic group isomorphic to the multiplicative group
. For example, if char
, any cuspidal cubic is isomorphic to the plane
curve given by the equation

(4.3)
(see Chapter 10). Its singular point is and the set of nonsingular points is
the subset of defined by the equation . The group law is given by
the formula

Each irreducible plane cubic curve has at least one nonsingular inflection point,
i.e., a point where the tangent to the curve has multiplicity of intersection with the
curve is equal to 3 (the only exception are certain cuspidal cubics in characteristic
3, see Chapter 10). Any of these points can be chosen as the zero point of the group
law. In the example (4.3), the point is the unique nonsingular inflection
with respect to the group law by
point. We denote the sum of two points
.
Lemma 4.4. Let be an irreducible plane cubic curve with a nonsingular inflection point taken to be the zero of the group law on the set of nonsingular
points of . Let . Then the order of the sum in the group
law on
is equal to if and only if there exists a homogeneous polynomial
of degree not vanishing identically on with multiplicity at each point
.
57
Proof. We assume that is nonsingular; however, everything we say is valid in

the singular case too. We use the following geometric interpretation of the group
law. Given two nonsingular points and in the line joining them intersects
the curve at the point equal to . Also, for any point its negative is
the third point of intersection of the line joining and with the curve . This
immediately implies that the sum is the unique
point such that there exists
a rational function on with divisor is equal to

. By induction, this
is the unique point such that there exists a rational

implies that
is equal to .
function on whose divisor
Conversely, suppose such exists. Let

the above there
. By
exists a rational function such that

. But

then
. This implies that (otherwise the rational map from
to defined by the function is an isomorphism).

In particular,
that is an -torsion element if and
we obtain
is the divisor of a rational function. Let us

only if
now take . Assume that there exists a polynomial
as in the statement
of the lemma. Let

be the equation of the inflection tangent at the point
. Then the restriction of the rational
to the curve
function on
defines a rational function with

. Thus
is an -torsion element in the group law. Conversely, assume that

the latter
occurs.
By
the
above
there
exists
a
rational
function
with

. By changing the projective coordinates if necessary, we
may assume that the equation of is and that none of the points is the
point with projective coordinates . Then the rational function is regular
on the affine curve
. Hence it can be represented by a polynomial
with nonzero constant term. Homogenizing this polynomial,
we obtain a homogeneous polynomial which
is not divisible by such that
. By Bezouts Theorem,
the curve
cuts out the divisor
the degree of is equal to . Note that is not defined uniquely since we
is a homogeneous
can always add to it a polynomial of the form , where
cuts out the

polynomial of degree . The rational function
same divisor on . Now we have to show that can be chosen in such a way that
has multiplicity at each point . Let
be the local ring of at the point
and let q
be its maximal ideal. Since was assumed to be nonsingular, one
can find a system of generators of q
such that is a local equation of
at . We shall identify the formal completion

of
with the ring of formal
power series in such a way that under the inclusion

the image
58
of is equal to
and the image of is equal to . Let
be the Taylor expansion of the rational function , where is a

homogeneous form of degree in . We denote by the th Taylor poly
nomial . The polynomial has multiplicity at if and only
if . The local ring
is isomorphic to the local ring of at
, and its completion

is isomorphic to q . The image of in is
equal to and the fact that the order of the restriction of to

at is equal to gives that
. This implies that

for some polynomial of degree . Now consider the
-linear map
which assigns to a homogeneous polynomial of degree the element

, where is the th Taylor polynomial of the rational function
at the point . We claim that this map is surjective. Computing the
dimensions of both spaces we find that
Thus it suffices to show that the kernel of the map is one-dimensional. An element
in the kernel defines a homogeneous polynomial of degree which has
at each point . Since we assume that the order of the
multiplicity
sum of the points is exactly , the polynomial must vanish on . Dividing
by and continuing the argument, we see that for some .
This proves the surjectivity. Now, it remains to choose in such a way that its
image under is equal to . Then the th Taylor expansion of

q at is equal to . Thus has multiplicity

at each point .
59
be the equation of the curve cutting out the divisor

. Let be the equation of . For any , the

defines a curve which cuts out the same divisor
polynomial
on . When is equal to the order of the point ,

the pencil of curves is called the Halphen pencil of index (see [15],
Remark
4.1. Let
Chapter 5). One can show that its general member is an irreducible curve with
-tuple points at . The genus of its normalization is equal to 1.
Lemma 4.5. Let be nine distinct nonsingular points on an irreducible

plane cubic s . Assume that their sum in the group law is not a torsion
element.
(i) A homogeneous polynomial of degree which has multiplicity

at each point is divisible by .

(ii) The dimension of the space r of homogeneous polynomials of degree
.
which have multiplicity
at each is equal to
Proof. Assume is not divisible by . By Bezouts Theorem,

. Now
this contradicts Lemma 4.4, so we may write s for some homogeneous
polynomial of degree . Clearly, the multiplicity of at each is equal to
. Applying the lemma again, we find that the sum of the in the group law
is a torsion element unless divides . Continuing in this way we find that
divides . This proves the first assertion.
Let us prove the second one. We may assume that all the points lie in

the affine part . Consider the linear functions
polynomials of degree
on the space of homogeneous
of order of the
which assign to a polynomial
the partial derivatives
dehomogenized polynomial at the point , . Obviously, r

is the space of common zeros of the functions . To check assertion (ii) it suf
fices to show that the functions are linearly independent. The subspace of
common zeros of the restriction of these functions to the space formed by the
polynomials , where , is of dimension 1 (by (i) it con

sists of polynomials proportional to , where is the curve ). Since
, the restriction of the functions to is a linearly inde
pendent set. Therefore the functions are linearly independent.

Now we are ready to prove Theorem 4.3.
Proof. We take and in the equations (4.1) we take to be the co
ordinates of the points which lie in the nonsingular part of an irreducible plane
60
cubic and which do not add up to an -torsion point for any . Also, to
satisfy Lemma 4.2, we assume that the first three points do not lie on a line. This
can always be arranged unless char and is a cuspidal cubic. Assume
that q is finitely generated. By Lemma 4.3, we can find a generating set
of the form , where is a polynomial of some degree

which has multiplicity at the points . By Lemma 4.5(i), .
Choose larger than every and prime to char . By Lemma 4.5(ii), the
dimension of the space
of degree
have mul of polynomials
which
tiplicity
at each is equal to
. On the other
in which vanish on
hand the dimension of the subspace of polynomials
is equal to . Thus there exists a polynomial

which
on the curve . Let us show that cannot be expressed
does not vanish
as a
polynomial in . Consider any monomial . After we replace

with , its degree in is equal to and its degree in
is equal to (here we use that are algebraically independent). Suppose our monomial enters
into a polynomial expression of in the
Thus
generators . Then
Since does not vanish on , we may assume that q if q (in this

case defines ). Thus for all with , and we get that the
only possible case is ,
for one and all other are equal to
zero. Thus for some . This contradicts the choice of .

Remark 4.2. If we take to be the cuspidal cubic over a field of
zero characteristic, and the points with the first three points not on
a line, then the conditions on will always be satisfied unless . In
fact, the group law on has no nonzero torsion points.
Remark 4.3. If we restrict the action only to the group r (not including
the torus), the algebra of invariants is also not finitely generated. This follows
from Nagatas Theorem since the torus is a reductive group. One may ask what
is the smallest such that there exists a rational action of r on a polynomial

algebra for which the algebra of invariants is not finitely generated. Recall that by
Weitzenbocks Theorem,
. Examples with and were given recently
by S. Mukai ([70]).
61
Finally we sketch Nagatas original proof of Theorem 4.3, which leads to a

very interesting conjecture on plane algebraic curves. We keep the previous notations.
denote the
Lemma 4.6. For any homogeneous ideal let
smallest positive integer such that

Assume that is

for all . Then for any natural

chosen to be such that
.
number there exists a natural number such that
be the space of homogeneous polyProof. Let q
nomials of degree in . As we explained in the proof of Lemma 4.5, the

dimension of this space is greater than or equal to

. Thus we seethat
. In view of our assump
tion we must have
q . Since again by assumption

we see that for sufficiently large ,

This implies that
is strictly larger than
Lemma 4.7. The assumptions of the previous lemma are satisfied when
where and the coordinates of the points generate a field of sufficiently
high transcendence degree over .
For the proof we refer to [78]. It is rather hard.
Let us show that the four preceding lemmas imply the assertion. Assume
that the algebra is generated by finitely many polynomials .
as in Lemma 4.3. Let

We can write them in the form

. By
for sufficiently large
Lemma 4.6, we can find
. Obviously cannot be expressed as a

such that
polynomial in the . This contradiction proves the assertion.

The assumption that was crucial in Lemma 4.7. The following conjecture of Nagata is still unsolved.
Conjecture. Let be general points in projective plane. Let
be a plane curve of degree which passes through each with multiplicity .

Then
Here general points means that the sets of points for which
the assertion in the conjecture may be wrong form a proper closed subset in .
62
The relationship between Hilberts Fourteenth Problem and the Zariski Problem
is discussed in [71]. The material about Grosshans subgroups was taken from
[41], see also [89]. The original proof of the Weitzenbock Theorem can be found
in [120]. The case char is discussed in a paper of A. Fauntleroy [29].
The original example of Nagata can be found in [77] (see also [76]). We follow
R. Steinberg ([112]) who was able to simplify essentially the geometric part of
Nagatas proof. The group law on an irreducible singular plane cubic is discussed
in [46], Examples 6.10.2, 6.11.4 and Exercises 6.6, 6.7.
An essentially new example of a linear action with algebra of invariants not
finitely generated can be found in [1]. It is based on an example of P. Roberts
([92]). Nagatas conjecture on plane algebraic curves has not yet been proved.
It has inspired a lot of research in algebraic geometry (see [45] and references
there). It has also an interesting connection with the problem of symplectic sphere
packings (see [67]). It implies that the symplectic 4-ball of radius 1 and volume
1 contains disjoint symplectically embedded 4-balls of total volume arbitrarily
close to 1.
Exercises
4.1 Prove that the additive group r is not geometrically reductive.
4.2 Let
be divisors on a nonsingular variety . Consider the algebra

(i) Show that the algebra is isomorphic to the algebra

for some divisor on some projective bundle over .
(ii) Let
. Show that
is a finitely generated semigroup if is finitely generated.

(iii) Let be a nonsigular projective curve of genus , let t be two
points such that the divisor class of t is not a torsion element in the group of
divisor classes on . Prove that t is not finitely generated.

4.3 Show that the algebra constructed in Nagatas counterexample is isomorphic
to the algebra where is the inverse image of a line under the blow-up
of points in the projective plane and is the exceptional divisor.
EXERCISES
63
4.4 Prove that the algebra is finitely generated if there exists a positive
number such that the complete linear system defined by the line bundle
has no base points.
4.5 Show that the algebra of regular functions on the coset space is isomor
phic to the subalgebra where acts on by left multiplication.
4.6 Let be a closed reductive subgroup of an affine algebraic group which
acts on by left translations. Show that the homogeneous space is affine
and hence is finitely generated.

4.7 Write explicitly the group law on the set of nonsingular points of a nodal cubic
over a field of characteristic different from 2.
4.8 Show that the conjecture of Nagata is not true without the assumption
.
Chapter 5
Algebra of covariants
5.1
Examples of covariants
Let SL act on an affine algebraic variety Specm . Let be its

chapter we shall give a
subgroup of upper triangular unipotent matrices. In this
geometric interpretation of the algebra of invariants . Its elements are called
semiinvariants.
Suppose SL acts linearly on a vector space . Fix a nonzero vector
in and let be the stabilizer of in . Let Pol . For any
there exists SL such that . Define a function on

by

(5.1)
Since
s
, we have
implies

and hence

for some
This shows that this definition does not depend on the choice of and that the
function is well-defined. Also, for any r SL we have

and hence
Therefore is invariant under the natural diagonal action of on

65
CHAPTER 5. ALGEBRA OF COVARIANTS
66
It is clear that is a polynomial function in the first argument. Moreover, if

is homogeneous of degree , then is homogeneous of degree in the first
variable. Let us see that is also polynomial in the second argument. Choose
coordinates to assume that . Let .
Assume . Let

..
.. . .
..
.
.
.
.
to SL and . Thus and

belongs
is a regular function on the open set . Similarly we see that

is regular on the open set . Thus is a rational function which is regular
Clearly,
on . Hence it is regular on the whole of and so is a polynomial function.

Conversely, if is a -invariant polynomial function on , then the
function is an -invariant polynomial function on . It is easy to
see that this establishes an isomorphism of vector spaces:
Pol
Pol
Pol SL
Note that the space Pol Pol
Pol
has a natural bigrading, so that
Pol
Let us specialize this construction by taking
Pol
SL
Pol
Definition. A covariant of degree and order on the space Pol is an element of the space Pol Pol Pol SL . We shall denote this space by
Cov .
The geometric meaning of a covariant Cov

It can be considered as a polynomial map of affine spaces
Pol
Pol
is very simple.
given by homogeneous polynomials of degree . This map is SL -equivariant

with respect to the natural actions of SL on the domain and the target space.
5.1. EXAMPLES OF COVARIANTS

In coordinates:
67
where are homogeneous polynomials of degree
in the coefficients .
On can easily define the symbolic expression of covariants. By polarizing,

an element of Cov Pol Pol Pol SL becomes an SL
invariant polynomial function on the space of matrices Mat which is homoge

neous of degree in each column different from the last one, and is homogeneous
of degree in the last column. Observe that each of the first columns corre

sponds to a basis in . The last one consists of the coordinates
with respect to this basis. There is an analog of the First Fundamental which says that one can write this function as a linear combination of products
of -minors taken from the first columns and dot-products of the last column
with one of the first columns. In each product, each column, except the last
one, appears times, and the last column appears times. This implies that the
number of minors in each product must be equal to
This number is called the weight of a covariant. It has the property that

GL
The symbolic expression for the products is

where . Here each
occur exactly times among them.
Example 5.1. An invariant of degree
, and each number from

is a covariant of degree
and of order .
Example 5.2. The identity map Pol Pol is a covariant of degree 1 and
order . Its weight is equal to zero. Its symbolic expression is .

Example 5.3. Let

r s
Pol

r s
. Let
t
68
The Hessian of is the determinant

Hess
The map Hess r Hess
symbolic expression is
Hess
..
.
..
..
.
(5.2)
is a covariant of degree and order
. Its
We leave it to the reader to check this.
More generally, let be the square matrix with entries
a variables. Take as above and consider the product

polynomial function on Mat . Define the th transvectant as
considered as
as a
where is the omega-operator. The last subscript means that we have to replace
each unknown with . The map is a covariant of degree

. For example,
and order
Hess
Example 5.4. One can combine covariants and invariants to get an invariant. For
example, consider the Hessian of a binary cubic. It is a binary quadric. Take its
discriminant.
result
be an invariant of degree 4; let us compute it. If
The
must
we have

Hess

Discr Hess
This is (up to a constant factor) the discriminant of the binary cubic form from
Chapter 2, Example 2.1.
5.2. COVARIANTS OF AN ACTION
69
Example 5.5. For any two binary forms Pol

Jacobian
Then
5.2
Hess
Pol
define their
is a covariant of degree 3 and order
q .
Covariants of an action
The notion of a covariant of a homogeneous form is a special case of the notion

of a covariant of an arbitrary rational action of an affine algebraic group on an
affine variety Specm . Let GL
be a linear representation
of in a finite-dimensional vector space . We call a -module. A covariant of an action with values in is an equivariant regular map , where
is considered as an affine space. Equivalently, it is a -equivariant homomor . Since any such homomorphism is determined
phism of algebras Pol
by the images of the unknowns, it is defined by a linear map . Let
Hom be the set of such maps. The group acts by the formula
This corresponds to the action on morphisms
given by the formula
A covariant is an invariant element of this space. In the

previous section we
considered the case SL , Pol

and
Pol
with the natural representation of SL . If we take

Pol with the
natural action of on the space of linear functions, we obtain the notion of a
contravariant of order on the space Pol . Another special case is when
Pol
Pol and
Pol . In this case a covariant
is called a concomitant of order . A concomitant of order is called a combi
nant. For example, the resultant of homogeneous polynomials is
a combinant.
Let Hom be the set of covariants with values in a module . It has an obvious structure of an -module. It is called the module of
70
covariants with values in . If char

so that the direct sum
and Pol
Cov
, we can identify the spaces Pol
Hom Pol

Pol
Pol

has a natural structure of a -algebra. It is called the algebra of covariants.
Ap-
plying Nagatas Theorem we obtain

Theorem 5.1. Assume is a geometrically reductive group. Then the algebra of
covariants Cov is a finitely generated -algebra.
Corollary 5.1. Suppose is a geometrically reductive algebraic group acting
rationally on Specm . Then the module of covariants Hom is
finitely generated.
Proof. The algebra Cov is a graded finitely generated -algebra. We
identify with the subalgebra of covariants Cov , where is the trivial
-module. Obviously Cov is a finitely generated -algebra. We may
assume that it is generated by a finite set of homogeneous elements of
positive degrees . Thus there is a surjective homomorphism of graded
-algebras Cov
, where
. Since each

is a finite free -module, its image Cov
is a finitely
generated -module; hence Hom Cov

is finitely generated.
Here is another proof of this result in the case when is linearly reductive,
for example when is reductive over . We use that any rational linear linear
representation of in a finite-dimensional linear space is completely reducible
in the following sense.
Theorem 5.2. Any submodule
(i.e., ).
of
admits a complementary submodule

Proof. Without loss of generality, we may assume that

and is an irreducible submodule, i.e., does not contain any nontrivial proper submodules.
Consider the natural map of -modules Hom Hom . By
5.2. COVARIANTS OF AN ACTION
71
Schurs Lemma, the subspace Hom is one-dimensional. Its inverse
image is a submodule of Hom , and the restriction of

to is a nonzero linear -invariant function. By definition of linear reductivity
there exists a nonzero -invariant vector such that . The linear
map is -invariant and its restriction to is a nonzero automorphism
of irreducible -module . The kernel of the -invariant linear map is the

desired complementary subspace of .
Let and let be the -submodule of generated by invariant
elements. Since is noetherian and is a free -module of finite rank,
is a finitely generated -module. Let be its spanning set. For
any we can write
(5.3)
for some . Since is linearly reductive the -submodule

complementary invariant submodule, i.e., . Let
of
has a
be the projection operator (called the Reynolds operator). It has the property
t t t
In the case we take for the averaging operator over the compact form
.
of . Let be the map defined by
is equal to the image under of the finitely generated -module

and hence it is finitely generated.
Let GL be a finite-dimensional linear rational representation
By (5.3),
of a linearly reductive group. By Theorem 5.2, can be decomposed into a

direct sum of irreducible representations . When is finite, there are only
finitely many irreducible representations (up to isomorphism); in general has
infinitely many nonisomorphic irreducible representations. Let be

a decomposition of into a direct sum of irreducible representations. We have
an isomorphism of -modules:
Irr
Hom
(5.4)
72
where Irr is the set of isomorphism classes of finite-dimensional irreducible

-modules, is a representative of the class , and acts trivially on the space
of linear maps Hom . This isomorphism is defined by the map
Hom
Note that, by Schurs Lemma, when

this gives Hom
if
and otherwise. The dimension of the space Hom is called
the multiplicity of
in and is denoted by mult . It is equal to the number
of direct irreducible summands (or factors) of isomorphic to .
Recall that any element of is contained in a finite-dimensional -invariant

subspace of generated by its -translates (see Lemma 3.4). This allows us to
apply (5.4) to the -module . We have
Irr
Hom
(5.5)
We consider both sides as -modules. By Corollary 5.1 each summand is a

finitely generated -module. Thus we see that any module of covariants for
is contained in as a direct summand.
Example 5.6. Let be a finite abelian group of order prime to char . Then
any irreducible representation of is one-dimensional, and hence is defined by a
GL . For each , let
character
Then (5.5) translates into the equality
The subring of invariants
5.3
corresponds to the trivial character.
Linear representations of reductive groups
Let be a linearly reductive connected affine algebraic group and let

GL be its rational linear representation. Let be a maximal unipotent subgroup of a connected linearly reductive group . The reader unfamiliar with the
5.3. LINEAR REPRESENTATIONS
73
notion may assume that GL or SL , in which case is a subgroup conjugate to the group of unipotent upper triangularupper triangular matrices. We
have seen in section 5.1 that in the case SL the algebra Pol Pol is

isomorphic to the algebra of covariants Cov Pol
. In this section we
shall give a similar interpretation of the algebra
where acts rationally on a
finitely generated -algebra .
For this we have to recall some basic facts about finite-dimensional linear
rational representations of a reductive group . We assume that char .
Let GL
be such a representation. Choose a maximal torus in
(when GL it is a subgroup of diagonal matrices or its conjugate subgroup).
Restricting to we get a linear rational representation GL
. Since
is commutative we can decompose into the direct sum of eigenspaces
where denotes the set of rational character of

algebraic groups , and

Any rational character
of regular functions
, i.e., homomorphisms of
is defined by a homomorphism of the algebras

r
r
r
It is easy to see that it is given by a Laurent monomial
, where
. The monomial is the image of . Also it is easy to see
that the product of characters corresponds to the vector sum of the exponents .
This gives us an isomorphism of abelian groups
Let
Wt
is finite-dimensional, Wt
(5.6)
Since
is a finite set. It is called the set of weights
of .
A rational character is called a root if there exists a nontrivial
homomorphism of algebraic groups r such that, for any and

any ,

74
roots for GL . Each is defined by the

For example, there are

, where
homomorphism which sends to the matrix
.
Let be the set of roots. There is the notion of a positive root. We fix a
Borel subgroup containing (in the case GL we may take to be
the group of upper triangular matrices or its conjugate subgroup) and require that
the image of is contained in . Let be the set of positive roots. Then

, where is the set of negative roots. There

is a finite set of roots

such that any root can be written as a
linear combination of the with nonnegative integer coefficients. They are called
simple roots. The number is called the rank of . In the case SL these are the
roots with
. Under the isomorphism
they correspond
to the vectors , where is the
standard basis of .
Let denote the image of the homomorphism r corresponding to a
root . One can show that the subgroups
are maximal unipotent subgroups of . In the case SL the group

) is the subgroup of upper triangular (resp. lower triangular) matrices.
We have the following.
Lemma 5.1. Let
For every root
Proof. Let
. For any
,
, we have
(resp.
Wt
be the homomorphism defining the action of on
its image is equal to . This means that for any
By definition of a root, we have
(5.7)
5.3. LINEAR REPRESENTATIONS
75
and
Comparing the coefficients of we get Thus equation (5.7) gives

The set defines an order on the set of characters. We say that if

is equal to a linear combination of positive roots with nonnegative coefficients.
Let Wt be a maximal element (not necessary unique) with respect to this
, we have
if It follows from
order. Then, for any
acts identically
. Thus the whole group
(5.7) that acts identically on
on
. On the other hand, by Lemma 5.1, we get
q

, all elements of the form

, leave the subspace

Since
, where
invariant. Since the subset q is Zariski dense in (check this for SL

or GL , where this set consists of matrices with nonzero pivots), all elements of

. Consider
leave invariant. Thus is a -submodule. Let
the -submodule generated by . Obviously it is contained in and
In fact,
the sum
does not change , multiplies by a constant, and sends to

t , where . We consider a complementary subspace
76
to in and choose again a nonzero vector in it to get a submodule
. Continuing in this way we will decompose into the direct sum of
submodules. Each summand has the following properties:
(i) there exists a weight such that ,
(a nonzero vector in is called a highest weight vector),

(ii)
(iii) is the identity representation.

Such a -module is called a highest weight module. It is determined
uniquely (up to isomorphism) by the character (highest weight) and is denoted
by . Thus we infer from the above discussion the following
Theorem 5.3. Every finite-dimensional rational representation of a connected
linearly reductive group is isomorphic to the direct sum of highest weight representations .
Not every weight occurs as a highest weight of some . The ones which
occur are called dominant weights. This set is preserved under taking the dual
module, i.e., for some dominant weight . We will describe
dominant weights in the next section.
Let us return to the situation when a reductive group acts regularly on an
affine algebraic variety Specm . For every dominant weight a homomorphism of -modules is determined by the image of a fixed highest
weight vector of . The set of such images forms an -submodule of

. We have

Hom
It is easy to see that, if is a highest weight vector of and is a highest
weight vector of , the vector is a highest weight vector in an irre .
ducible summand of the representation isomorphic to
This easily implies that the subalgebra of the -algebra generated by the images of highest weight vectors is isomorphic to the direct sum of the -modules
, where runs through the set of dominant weights. Since acts identically
on any highest weight vector we see that
Conversely, if
, by (5.4) can be written uniquely as a sum
, where
each belongs to an irreducible -submodule of . This implies that each
5.4. DOMINANT WEIGHTS
77
is -invariant and hence generates a submodule isomorphic to

dominant weight . This shows that
for some
(5.8)
Since every irreducible representation is isomorphic to some highest weight representation , we can apply (5.4) to obtain an isomorphism of -modules

This gives
Hom
Hom

It follows from the definition of that

is spanned by a
highest weight vector, and hence is one-dimensional. This gives
We will see a little later that
5.4
is a finitely generated algebra.
Dominant weights
Let us now describe dominant weights. For every root there is the dual root
which is a homomorphism
. It is characterized by the property that,
,
for any
and
(i) ,
.
(ii)
For example, when GL and is the subgroup of matrices

, where denotes the matrix with as the th entry and 0 else .
where, the dual root is given by
Note that the composition of a homomorphism
(called a oneparameter subgroup) of and a rational character
can be identified
with an integer. We denote it by

.
Let be the set of one-parameter subgroups. An element of is

given by a homomorphism of algebras of functions
78
It is defined by the images of . Since it defines a homomorphism of groups it is

easy to see that the image of each is a monomial for some . Thus a
one-parameter subgroup is given by a vector . Since each

one-parameter subgroup takes values in a commutative group, we can multiply
them; this of course corresponds to the sum of vectors in . The composition of
a character and a one-parameter subgroup corresponds to the dot-product in .

So it is natural to distinguish the group of characters and the group
of one-parameter subgroups by identifying one of them, say , with and
the other one with the dual group Hom . Then the pairing
from above is equal to
A character
is called a dominant weight if for any positive root
one has
.
Finally, one defines a fundamental weight as a dominant weight with the
property (the Kronecker symbol). Of course, one has to prove first
, are really in . In
that such vectors, which obviously exist in
the case when spans the group of characters of (e.g. SL but not GL ),
a fundamental weight is uniquely determined by this property. Let be the
such that
for all roots
subgroup of which consists of characters
. Choose a basis of and let be the set of
fundamental roots no two of which are congruent modulo the subgroup .

Then any dominant weight can be written uniquely in the form

(5.9)
where .
Any dominant weight from defines a one-dimensional representation . We have rank fundamental representations corre
sponding to the fundamental weights . If is as in (5.9), then is isomorphic
to an irreducible quotient of the tensor product tensored with the
one-dimensional representation defined by the vector .
By writing any dominant weight as a sum of fundamental weights we prove

the result which we promised earlier:
Theorem 5.4. Let be a maximal unipotent group of a reductive group . Assume that
acts rationally on a finitely generated -algebra . Then the subalgebra
of -invariant elements is finitely generated over .
Proof. Since all maximal unipotent subgroups are conjugate, we may assume that
. We know that for each dominant weight the module of covariants
5.5. THE CAYLEYSYLVESTER FORMULA
79
is finitely generated over . Let be the union of the sets of

generators of such modules for , .
Using the equality (5.8) we see that generates
as an -module. Since
is finitely generated by Nagatas Theorem,
must be finitely generated too.
5.5
The CayleySylvester formula
In this section we give an explicit description of irreducible representations for

the group GL . We choose the maximal torus which consists of diagonal matrices diag . The corresponding Borel subgroup is the group of upper
triangular matrices. We have, for any ,
diag
This shows that the characters diag are roots. Under the
isomorphism each corresponds to the vector . So we have

if and only if , we see that consists
roots. Since
of roots with . Simple roots are
diag
The dual roots

are the homomorphisms
defined by
Thus all dual roots can be identified with linear functions
by where is the dual basis to the standard
defined
basis
. A dominant weight must satisfy

which translates into the inequalities . There are
weights

and
fundamental
is generated by the weight

The irreducible representation corresponding to
sentation
GL
is of course the natural repre-
80
We have
Pol
Here the highest weight is the monomial , where is the standard
basis of . All other weights are q with

. The
corresponding subspace is spanned by the monomial . We can write

Pol

Here the highest weight is rsr . When we get

and hence
Pol det
The highest weight here is the monomial .

Consider the case . Let be a two-dimensional vector space. Since
is isomorphic to the representation

GL , we have an isomorphism
of representations:

In particular, as representations of SL . We have one fundamental weight so that any irreducible representation with dominant weight
is isomorphic to
det det Pol det

Let us consider the representation Pol Pol . The space has a basis formed
by monomials in coefficients of a general binary -form
So we can write any monomial of degree

in the basis of :
where
in the as a monomial of degree
81
is the weight of the monomial . This shows that belongs to

the weight space with character . Let
The cardinality of this set is equal to the number of monomials with

weight . Let be a dominant weight. Suppose is a direct
with
summand of Pol Pol . Then for some

. The weights of are the vectors
q This shows that Pol Pol contains
det

summands Pol
det
summands Pol det

and so on. It is known that the generating function for the numbers

summand
Pol
is
equal to the Gaussian polynomial
where
r

t
r
s
s
(see [111]). This gives us

Theorem 5.5. (Plethysm decomposition) Let
phism of representations of GL :

Pol Pol
Pol
where
coefficient of
det
. There is an isomor-
in the polynomial

82
Restricting the representation to the subgroup SL

of SL -representations
we have an isomorphism

Pol Pol
Pol
As a corollary we obtain the Cayley-Sylvester formula for the dimension of

the space of covariants:
Corollary 5.2.
and it is zero if
Cov
is odd.
We also get Hermites Reciprocity :

Theorem 5.6. There is an isomorphism of SL
-modules
Pol Pol Pol Pol
Proof. This follows from the following symmetry property:

t
t by sendThis can be checked by defining the bijection
t
ing a vector from to the vector , where

It follows also from the following property of the Gaussian polynomials:
Corollary 5.3.
Pol
Pol SL
Pol Pol
SL

83
Remark 5.1. The covariant

Pol
Pol
Pol
(5.10)
admits a simple interpretation

in terms of the Veronese map. Let be a linear
space of dimension
. Recall that the Veronese map of degree in dimension
is a regular map
Pol
, where is a linear function on . It is easy to see

given by
that this map is SL -equivariant, where SL acts naturally on and on
Pol . The inverse image under defines an equivariant linear map

When , there is an isomorphism of SL -modules and the map
is the covariant (5.10). Note that the image of the Veronese map (called
Pol
Pol
Pol
the Veronese variety) is always defined by equations of degree 2 (see [102]). The
number of linearly independent equations is equal to
Thus, if the kernel of the map (5.10) is a SL -submodule of the dimen
Pol Pol
Pol
sion given by the above formula.

Remark 5.2. One can strengthen Theorem 5.6 as follows (see [49]). Let
vector space of dimension , and let
Pol Pol
Pol
be a
Pol
. Let

Pol Pol
be the algebra of polynomials on the space Pol
(5.11)
We use the symbolic expression to identify elements of Pol Pol with mul
tihomogeneous functions on of multi-degree (see Lemma 1.1). The product of functions defines bilinear maps
Pol Pol
Pol Pol
Pol Pol
84
which endow with a structure of a graded algebra. The natural

action of
GL on defines an action of GL on both algebras and by au SL is isomorphic to the algetomorphisms

of
graded
algebras.
Notice
that

bra
(Corollary 2.4). Identifying the linear spaces Pol Pol
Pol and Pol Pol Pol (see (1.7)), we get a GL -equivariant
algebra homomorphism:

(5.12)
When , the homomorphism is a GL -equivariant isomorphism of
graded algebras. Hermites Reciprocity only states that all graded pieces are isomorphic as GL -modules.
Example 5.7. Take . We get q

Thus we have the following isomorphism of SL -representations:
Pol Pol
Pol
Using the previous remark this has a simple geometric interpretation. In this case
the Veronese variety is a conic, and the kernel of is one-dimensional. It is
spanned by a quadratic polynomial vanishing on the conic.

Example 5.8. Take
modules
Pol Pol
. Then we have an isomorphism of SL
Pol Pol
Thus quadrics in Pol can be canonically identified with cubics in
Pol . The Veronese curve is a rational space curve of degree
3. It is defined by three linearly independent quadric equations. Thus the kernel

of the projection Pol Pol Pol is equal to the space of quadrics
vanishing on . Using the plethysm decomposition

Pol Pol
we can identify
5.6
, SL
Pol
Pol
-equivariantly, with the space of binary quadratic forms.
Standard tableaux again
Finally let us explain the tableau functions from the point of view of representation theory. Note that any can be embedded (as a representation) into
some tensor power of some copies of . So when we take their symmetric
5.6. STANDARD TABLEAUX AGAIN
85
products and their tensor products we can embed each again into some . So
each irreducible representation is realized as an irreducible submodule of for
some . Let us find them by decomposing into a direct sum of irreducible

representations.
Fix a basis of . For any ordered subset of let
denote the tensor
. A diagonal matrix diag acts
on by multiplying it by the monomial . Writing any element of
as a sum of tensors we
easily see that the weights of our representation
are the vectors
. The weight subspace is spanned by the
tensors , where is obtained from by a permutation of
. A vector is a
dominant weight if
This means that

Assume for the moment that . Then the highest weight vector is . Assume
that . Then is sent by to

. Similarly, is sent to . So in order that
be invariant under we must have , i.e.,

r . If
we must have
must be proportional to

or
or
Now in the case of arbitrary
we do the following: consider a matrix

..
..
..
..
.
.
.
.

Each column represents a basis . We will be taking

minors of order 1 from the first row
86
minors of order 2 from the first 2 rows
minors of order
in such a way that the minors do not have common columns. Of course we compute the minors using the tensor product operation. We first take the product of
the minors in an arbitrary order, but then we reorganize the sum by permuting
the vectors in each decomposable tensor in such a way that each summand has
its upper indices in increasing order. These indices will be our highest weight
vectors.
It is convenient to describe such a vector by a Young diagram. We view a
dominant vector as a partition of . It is described by putting
of length s
boxes in the th row. It has q columns
( ). We fill the boxes with different numbers . Each indicates
which column enters into the minor of the matrix of the corresponding size.
A filled Young diagram is called standard if each row and each column are in
increasing order. Here is an example of a Young diagram for the partition
:
of
1
It turns out that the multiplicity of each in is equal to the number

of standard filled Young diagrams of the shape given by the vector . It is given
by the hook formula
mult
(see [66]).
Example 5.9. We described invariants in Pol Pol by embedding this space
into via the polarization map. Since the space of invariants is contained
87

in the representation
of GL where , the corresponding dom
inant vector is . The representation is of course
one-dimensional. The Young diagram is of rectangular shape with rows and
columns. The number of such diagrams is equal to the dimension

of the space SL . It is not difficult to see that the hook formula gives the
formula

SL

The standard tableaux on the set of size
defined in Chapter 1 correspond to standard Young diagrams which are filled in such a way that if we
write the set as the disjoint union of subsets

, then each column consists of numbers
taken from different subsets of . Moreover, for a homogeneous standard
tableau we have to take exactly numbers from each subset. The general for
mula for the dimension of the space Pol Pol SL is not known for
.
The notion of a covariant of a quantic (i.e., a homogeneous form) goes back to A.
Cayley. It is discussed in all classical books in invariant theory. The fact that a
covariant of a binary form corresponds to a semiinvariant was first discovered by
M. Roberts in 1861 ([91]). It can already be found in Salmons book [97]. The
result that the algebra of covariants of a binary form is finitely generated was first
proved by P. Gordan [38] (see also classical proofs in [28], [39]). A modern proof
can be found in [113]. Theorem 5.4 applied to the action of SL on the
algebra Pol Pol is a generalization of Gordans Theorem. The first proof
of this theorem was given by M. Khadzhiev [61]. Our exposition of the modern
theory of covariants follows [89]. The algebra of
covariants of binary forms of
degree was computed by P. Gordan for

([38]) and by F. von Gall for
degree ([36], [35]) (the proof of completeness of the generating set for
may not be correct). For ternary forms the computations are known only
for forms of degree 3 ([37], [42]) and incomplete for degree 4 ([98], [19]) (a thesis
of Emmy Noether was devoted to such computations). Combinants of two binary
forms of degrees are known in the cases ([96]; see a modern
account of the case in [81]). Also known are combinants of two
ternary forms of degrees

([28]).
88
The theory of linear representations of reductive groups is a subject of numerous textbooks (see, for example, [34], [52]). For the historical account we refer
the reader to [7]. The Cayley-Sylvester formula was first proven by Sylvester in
1878 (see historical notes in [109]). Other proofs of the Cayley-Sylvester formula
can be found in [108], [109], [113]. Hermites Reciprocity goes back to 1854.
One can find more about plethysms for representations of GL in [34]. The relationship between Young diagrams and standard tableaux is discussed in numerous
books (see [64], [113], [121]).
Exercises
5.1 Let Pol Pol be a covariant of degree and order and
Pol Pol SL be an invariant. Consider the composition and compute

its degree and weight.
5.2 Let Hess Pol Pol be the Hessian covariant. Show that it defines
a rational map of degree 3 from the projective space of plane cubic curves to
itself.
[Hint: By a projective transformation reduce a plane cubic to a Hesse form
and evaluate the covariant.]
5.3 Using the symbolic expression of covariants describe all covariants of degree
on the space Pol .
5.4 Find a covariant of degree 2 and order 2 on the space Pol . Describe the
locus of indeterminacy for the corresponding rational map .
5.5 Find the symbolic expression for the transvectant .

5.6 Find all covariants of degree 3 for binary forms.
5.7 Define the th transvectant of homogeneous forms in vari

ables by generalizing the definition of the covariant . Prove that it is a concomitant and find its multi-degree and order.
5.8 Consider the operation of taking the dual hypersurface in projective space.
Show that it defines a contravariant on the space Pol . Find its order and
degree for .
5.9 Let be a plane curve of degree 4. Consider the set of lines which
intersect it in four points which make an anharmonic (or a harmonic) cross-ratio.
Show the set of such lines forms a plane curve in the dual plane. Find its degree
and show that this construction defines a contravariant on the space Pol . Find
its degree.
EXERCISES
89
5.10 Let be a finite group which acts on a finitely generated domain . Assume that the action is faithful (i.e., only acts identically). Show that
for any irreducible representation
of the rank of the module of covariants
. [Hint: Use the fact that each irreducible repreHom is equal to
sentation is contained in the regular representation (realized in the group algebra

of ) with multiplicity equal to its dimension.]
5.11 Let be a finitely generated abelian group and let be its group algebra
over a field . Show
(i) Specm is an affine algebraic group.
if and only if is free.
(ii)
(iii) The group of rational homomorphisms

is naturally isomorphic to , and the group of rational homomorphisms
is isomorphic
to Hom .
(iv) Each closed subgroup of is isomorphic to where is a
factor group of .
(v) There is a bijective correspondence between closed subgroups of
and subgroups of .
5.12 Find the roots, dual roots, dominant weights, and fundamental weights for
the group SL .
5.13 Let
(i) Let
be a representation of with highest weight vector .

be the line spanned by . Show that the stabilizer
is a parabolic subgroup (i.e., a closed subgroup containing a Borel
subgroup).
(ii) Show that the map defines a projective embedding of the homo
geneous space
.
(iii) Consider the case GL and is one of the fundamental weights.
Show that is isomorphic to the Grassmann variety Gr and the map defined
in (ii) is the Plucker embedding.
5.14 In the notation of section 5.1 show that for the group SL
Show that there is an isomorphism of Pol -modules
Pol
5.15 Let be a subgroup of
triangular matrices.
Pol
SL which contains the subgroup
of upper
90

(i) Show that for any highest weight module one has
and the equality takes place if and only if is contained in the stabilizer of a
highest weight vector.
(ii) Let be the set of for which the equality holds. Show that for
any action of on Specm there is an isomorphism of -modules

(iii) Consider the example of from the previous problem and find .
5.16 Let
and char . Show that there is an isomorphism of
SL -modules
Pol
Pol Pol
5.17 Let be as in the previous exercise. Find the decomposition of the GL module Pol Pol into irreducible summands (the ClebschGordan de
composition).
5.18 Find an irreducible representation of GL with highest weight equal to
Chapter 6
Quotients
6.1
Categorical and geometric quotients
Let be an affine algebraic group acting (rationally, as always) on an algebraic

variety over an algebraically closed field . We would like to define the quotient
variety whose points are orbits. As we explained in Chapter 1 this is a
hopeless task due to the existence of nonclosed orbits. So we need to modify the
definition of ; for this we look first at the categorical notion of a quotient
object with respect to an equivalence relation.

. The
Let be a set together with an equivalence relation
canonical map has the universal property with respect to all maps
. Also we have
such that
This equality expresses the property that the

fibers of the map are the equivalence classes. Let us express this in categorical
language. Let be any category with fibered products. We define an equivalence

(or more generally just a
relation on an object as a subobject
morphism ) satisfying the obvious axioms (expressed by means of

as an object in for
commutative diagrams). Then we define a quotient
having the universal property with
which there is a morphism
respect to morphisms such that factors through a morphism

. By definition there is a canonical morphism
(6.1)
Note that, in general, there is no reason to expect that in general the morphism
(6.1) will be an isomorphism or an epimorphism.
91
CHAPTER 6. QUOTIENTS
92
Let be an algebraic action. We say that the pair is

a -variety and often drop from the notation. Let be
the morphism pr . This morphism should be thought of as an equivalence
relation on defined by the action. A -equivariant morphism of -varieties

corresponds to a morphism of sets with an equivalence relation. The definition of
a -equivariant morphism r can be rephrased by saying that the map
the natural
morphism ; this corresponds to
factors through
the property
. This suggests the following definition.
Definition. A categorical quotient of a -variety is a -invariant morphism

such that for any -invariant morphism there exists
a unique morphism
satisfying
s . A categorical quotient is
called a geometric quotient if the image of the morphism equals . We
shall denote the categorical quotient (resp. geometric quotient) by
(resp. ). It is defined uniquely up to isomorphism.
A different approach to defining a geometric quotient is as follows. We know

how to define a geometric quotient as a set; we next discuss topological spaces.
We put the structure of a topological space on so that the canonical projection is continuous. The weakest topology on for which
is open if and only

this will be true is the topology in which a subset
if is open. Then we examine ringed spaces, whose definition is given

in terms of choosing a class of functions on (e.g. regular functions, smooth
, then
functions, analytic functions). If is a function on
must be a function on . It is obviously a

the composition
-invariant function.
Using this remark we can define the structure of a ringed
space on
by setting . This makes
categorical quotient in the category of ringed spaces. Finally, we want the fibers
of to be orbits; this is the condition that the morphism (6.1) is an isomorphism.
Definition. A good geometric quotient of a -variety
phism satisfying the following properties:
(i)
is surjective;
(ii) for any open subset of , the inverse image

is open;
is a
-invariant mor-
is open if and only if
(iii) for any open subset of , the natural homomorphism
is an isomorphism onto the subring of -invariant

functions;
6.1. CATEGORICAL AND GEOMETRIC QUOTIENTS

(iv) the image of
is equal to
93
.
Proposition 6.1. A good geometric quotient is a categorical quotient.

Proof. Let be a -invariant morphism. Pick any affine open cover
of . For any the inverse image will be an open -invariant
subset of . Then we have the obvious inclusion , where

q . Comparing the fibers over points and using property
(iv) (which says that the fibers of are orbits), we conclude that in fact
. By property (ii), is open in . Since is surjective we get an open
cover of . The map is defined by a homomorphism
Since is a -invariant morphism, the image of is contained in the subring
of . This defines a unique homomorphism ,

and hence a unique morphism (because is affine). It is immediately
checked that the maps agree on the intersections , and hence define a
satisfying
unique morphism
.
Proposition 6.2. Let
following properties:
be a
-equivariant morphism satisfying the
(i) for any open subset of , the homomorphism of rings
is an isomorphism onto the subring of -invariant

functions;
is a closed -invariant subset of
(ii) if
(iii) if are closed invariant subsets of

.
then is a closed subset of ;

with
, then
Under these conditions is a categorical quotient. It is a good geometric quotient

if additionally
(iv) the image of
is equal to
Conversely, a good geometric quotient satisfies properties (i)(iv).

Proof. This is similar to the previous proof. With the same notation, let
94
This is a closed -invariant subset of , hence, by (ii), is an
open subset of . Clearly, . Since , by (iii) we
, hence . Now composing the homomorphisms

have
with the restriction homomorphism
we get a homomorphism . Since is

affine this defines a morphism whose composition with
is the map

. Gluing together these morphisms we construct
as in the proof of Proposition 6.1. This shows that is a categorical
quotient.
Let us check that under condition (iv) is a good geometric quotient. First we see that is surjective. Indeed, (i) implies that is dominant and
(iii) implies that is closed. Also property (ii) implies property (ii) of the def
inition of a good geometric quotient. In fact, if is open, then

is closed and -invariant. Since is surjective, its image is equal to and is
closed. Therefore is open. This checks the definition.
Conversely, assume is a good geometric quotient. Properties (i)
and (iv) follow from the definition. Let us check properties (ii) and (iii). The
set is open and invariant. Since the fibers of are orbits,
and hence is open. For the same reason,
and
hence

is closed. Furthermore,

. This checks property (iii).
Corollary 6.1. Under the assumptions from the preceding Proposition, the map
satisfies the following properties:
(i) two points

;
(ii) for each
have the same image in if and only if
the fiber
contains a unique closed orbit.
Proof. In fact, the closures of orbits are closed -invariant subsets in . So if

, . But both sets contain the point
. Conversely, if , we get that and lie in
different fibers. Since the fibers are closed subsets, and lie in different
fibers, and hence they are disjoint. This proves (i). To prove (ii) we notice that by
(i) two closed orbits in the same fiber must have nonempty intersection, but this is
absurd. Since each fiber contains at least one closed orbit, we are done.
Definition. A categorical quotient satisfying properties (i), (ii) and (iii) from
Proposition 6.2 is called a good categorical quotient.
6.2. EXAMPLES
95
Remarks 6.1. 1. Note that condition (ii) in the definition of a good geometric
quotient is satisfied if we require
(ii) for any closed -invariant subset of , the image is closed. Also,
together with condition (iii) this implies the surjectivity of the factor map . In
fact, condition (iii) ensures that the map is dominant, i.e., its image is dense in
. But by (ii), the image of must be closed.
2. Suppose is an irreducible normal -variety over an algebraically closed

field of characteristic , and is a surjective -invariant morphism such
that its fiber over any point is an orbit. Then is a geometric
quotient. The proof is rather technical and we omit it (see [73], Proposition 0.2).
3. The definitions of categorical and geometric quotients are obviously local
in the following sense: If is a -equivariant morphism, and is
an open cover of with the property that each is a categorical

(resp. geometric) quotient, then is a categorical (resp. geometric) quotient.
6.2
Examples
Let us give some examples.

Example 6.1. Let be a finite group considered as an algebraic group over a
field . Assume that is quasi-projective. Then the geometric quotient
always exists. In fact, assume first that is affine. By Theorem 3.1, the algebra
is finitely generated over . Let be an affine algebraic variety with
. By the theorems on lifting of ideals in integral extensions, the
map satisfies properties (ii) and (iii) from Proposition 6.2.
Also, the group acts transitively on the set of prime ideals in which lie
over a fixed prime ideal of (see, for example, [9], Chapter V, 2, Theorem
2). This shows that is an isomorphism.
be quasi-projective but not necessarily affine. Let be the

Now let
be an orbit and let be a homogeneous polynomial
closure of . Let
but not vanishing at any point of . Thus is contained in
vanishing on

an affine subset
. Recall that the complement of a hypersurface in
a projective space is an open affine subset. This implies that , being closed in an
affine set, is affine. Let . This is an open -invariant affine
subset of containing . By letting vary, we get an open affine -invariant

covering of . We already know that each quotient
exists. We will glue the together to obtain the geometric quotient
96
(we refer to the gluing construction of algebric varieties in section 8.2). To do this
we observe first that is affine and is open in and ; this
follows from considering the affine case. Thus we can glue and together
along the open subset r ; we do this for all and . The resulting
algebraic variety is separated. In fact we use that in the affine situation

where acts on
product of the actions. Thus
by the Cartesian
the image of
in is closed,
and, as is easy to see, coincides with . This shows that is

closed. It remains to prove that is quasi-projective; we shall do this later.
Note that, if is not a quasi-projective algebraic variety, may not exist in
the category of algebraic varieties even in the simplest case when is of order
2. The first example of such an action was constructed by M. Nagata ([75]) in
1956 and later a simpler construction was given by H. Hironaka (unpublished).
However, if we assume that each orbit is contained in a -invariant open affine
subset, the previous construction works and exists.
Example 6.2. Let be a finitely generated -algebra with a geometric

grading (see Example 3.1). Consider the corresponding action of
on
Specm . Let be the vertex of defined by the maximal ideal .

Then the open subset
is invariant and the geometric quotient
exists and is isomorphic to the projective variety Projm . We leave the details
to the reader.
Example 6.3. Let be a closed subgroup of an affine algebraic group and
be the coset space (see Example 3.3). The canonical projection is a
good geometric quotient. We omit the proof, referring the reader to [52], IV, 12,
where all conditions of the definition are verified.
Let us show now that the categorical quotient of an affine variety always exists.
We will need the following lemma.
Lemma 6.1. Let be an affine -variety, and let and be two closed
invariant subsets with . Assume is geometrically reductive. Then
there exists a -invariant function such that
Proof. First choose a function , not necessarily -invariant, such that

sum of the ideals defining and
This is easy: since the
is the unit ideal, we can find a function and a function
6.2. EXAMPLES
97

such that
. Then we take . Let be the linear subspace of
spanned by the translates . We know that it is finite-dimensional
(Lemma 3.4); let be a basis. Consider the map

defined by these functions. Clearly, q . The
group acts linearly on the affine space, defining a linear representation. By definition of geometrically reductive groups, we can find a nonconstant -invariant
homogeneous polynomial such that . Then
satisfies the assertion of the lemma.

Now we are ready to prove the following main result of this chapter:
Theorem 6.1. Let be a geometrically reductive group acting on an affine variety
. Then the subalgebra is finitely generated over , and the canonical
morphism Specm is a good categorical quotient.
Proof. The first statement is Nagatas Theorem proven in Chapter 3. To show that
is a good categorical quotient, we apply Proposition 6.2. First of all, property
(i) easily follows from the fact that taking invariants commutes with localizations.
More precisely, if , then ; this is easy and
we skip the proof. Next let be a closed -invariant subset of . Suppose
is not closed. Let . Then

and
are
two closed -invariant subsets of with empty intersection. By the preceding
Lemma, there exists a function such that .
Since for some , we obtain . But this

is absurd since belongs to the closure of . This verifies condition (ii). Now
let and be two disjoint -invariant closed subsets of . As above we find a
function with
. This obviously implies that
. This verifies (iii).
Example 6.4. We have already discussed this example in Chapter 1. Let GL
act on itself by the adjoint action, i.e. . For each matrix GL we

consider the characteristic polynomial
by the formula
Define a regular -equivariant map GL
We claim that this is a categorical quotient. To check this it
is enough to verify that ; this is what we
did in Chapter 1. It is clear that the fiber of does not consist of one orbit, so the
quotient is not a geometric quotient.
98
6.3
Rational quotients
We know that neither nor exists in general. So a natural problem

is to find all possible open subsets of for which the categorical or geometric
quotient exists. Geometric invariant theory gives a solution to this problem when
we additionally assume that the quotient is a quasi-projective algebraic variety.
Let us first show that any open subset for which a geometric quotient
exists must be contained in a certain open subset reg . We will assume in the
sequel that is connected. Otherwise, we consider its connected component
containing the identity element. It is a normal closed subgroup of and the
quotient is a finite group. It is easy to see (see Exercise 6.11) that we can
divide by in two steps: first divide by , and then divide the quotient by the
finite group .
For any point we have a regular map

Clearly the image of this map is the -orbit O of the point .
The set theoretical fiber of this map at a point is denoted by and is called
the isotropy subgroup of in the action . It is a closed subgroup of , hence
coincides
an affine algebraic group. If char , the set theoretical fiber of
with the scheme theoretical fiber (or, in other words, the latter is a reduced closed
subscheme of ). We are not going to prove this; to do so we would have to go
into the theory of group schemes and prove the fundamental result of the theory
that every group scheme over a field of characteristic zero is reduced.
Since all fibers of over points in O are isomorphic (they are conjugate
subgroups of ), the theorem on the dimension of fibers (see [102]) gives

(6.2)
If O O , the complement O O is a proper closed subset of O ,

hence its dimension is strictly less than
any O O and

O . Take

consider its orbit O . Since
O
O , applying (6.2) to we see
that
(6.3)

Let

This is a closed subset of . Consider the second projection pr .
Its fiber over a point is isomorphic to the isotropy subgroup . By the
6.3. RATIONAL QUOTIENTS
99
theorem on the dimension of fibers applied to pr , there exists an open subset reg
for all reg and for all

reg .
of such that
Applying (6.2) we obtain that for any reg the orbit O is closed in reg
. Also, any other orbit in has dimension

and has dimension equal to
. Let be any -invariant open subset of for which

strictly less than
exists. We assume that is irreducible. So
a geometric quotient
reg
and hence some of the orbits in must be of dimension
of have dimension
. By the theorem
on dimension of fibers all fibers
greater than or equal to

and hence all fibers of have dimension equal
to
. Therefore they are contained in reg and hence
reg .
Thus we get a necessary condition for the existence of : must be an
open subset of reg .
Theorem 6.2. (M. Rosenlicht) Assume is irreducible. Then reg contains an
open subset such that a good geometric quotient exists with quasiprojective . The field of rational functions on is isomorphic to the subfield of -invariant rational functions on .
Proof. The proof is easy if we assume additionally that is geometrically reductive and is affine. Let be an algebraic variety with field of rational functions
isomorphic to ; such a always exists since is of finite transcendence degree over . Consider the rational dominant map reg defined by the
inclusion of the fields
. By deleting some subset from reg we find
a -invariant open subset
reg and a regular map from . Replacing by an open subset we may assume that is surjective. This is condition (i)
from the definition of a good geometric quotient. For any open subset
we
have an inclusion
. Since we
see that . Conversely

and hence . Thus we have checked condition (i) of Proposition 6.2. Since is -invariant, the fibers of are unions of orbits. Since any
orbit in reg is closed in reg , it is closed in . By Lemma 6.1 we can separate
closed invariant subsets by functions from . This shows that the fibers of
are orbits. This checks condition (iv). The conditions (ii) and (iii) of Proposition
6.2 are checked by using the argument from the proof of Theorem 6.1.
Let us give an idea for the proof in the general case. For the details we refer
to the original paper of Rosenlicht ([93]; see also [89], 2.3). Since we do not
assume that is affine, even if is geometrically reductive we cannot separate
the closed orbits contained in the fibers of the map . Consider the
generic fiber of as an algebraic variety over the field

100
Let be the algebraic closure of . The group

acts on and the field
have the
of invariant rational functions is isomorphic to . All orbits of
same dimension. Suppose that a group acts on an irreducible quasi-projective
such that all orbits are of the same dimension and closed. We

variety
define a map from to the Chow variety parametrizing closed subsets of of

the same dimension (see [73], Chapter 4, 6) by assigning to a point the
closure of the orbit . If the image is of positive dimension, we can construct
a nonconstant invariant function on by taking the inverse image of a rational
function on the image. Otherwise the image is one point, and we obtain that
consists of one orbit. Applying this argument to
we see that it consists
of one orbit. This implies that there is an open subset of such that each fiber
consists of one orbit. Again deleting a closed subset from we may assume that
is nonsingular. Since the dimension of all orbits is the same, the morphism is
open; this is called Chevalleys criterion (see [6], p. 44). This verifies condition
(ii) of the definition of a good geometric quotient. The remaining conditions have
been checked already.
Corollary 6.2. The transcendence degree of

where
is equal to
Any model of is called a rational quotient of by . We see that

contains an open subset such that a good geometric quotient exists and
coincides with a rational quotient.
The notions of a categorical and geometric quotients are originally due to Mumford ([73]). Many books discuss different versions of these notions (see [63],
[80]). Many interesting results about the structure of fibers of the quotient maps
have been omitted; we refer to [89] for a survey of these results.
Exercises
6.1 Let r act on
map
quotient?
by the formula
. Consider the
. Is it a categorical quotient? If so, is it a geometric
EXERCISES
101
6.2 Let
formula r for some
act on by the
coprime to char . Let with the
positive integers
corresponding geometric grading defined by the action. Show that the geometric

quotient
(see Example 6.2) is isomorphic to a quotient of

by a finite group.
6.3 Let be a graded finitely generated -algebra, and
. Show that, if is coprime to char , , where is a cyclic
group of order .
6.4 Construct a counterexample to Lemma 6.1 when r is the additive group.
6.5 In the notation of Nagatas Theorem show that for any open subset of , the
restriction map is a categorical quotient with respect to the induced

action of .
6.6 Describe the orbits and the fibers of the categorical quotient from Example 6.4
when .
6.7 Show that the categorical quotient of Pol Pol by SL is isomorphic to
. Describe the orbits and the fibers of the categorical quotient.

6.8 Let act on an irreducible affine variety and let be a
invariant morphism to a normal affine variety. Assume that codim
and that there exists an open subset of such that for all the fiber
contains a dense orbit. Show that .
6.9 Let be a finite group of automorphisms of an irreducible algebraic variety.
Prove that .
6.10 Show by example that in general the field of fractions of the ring of
invariants is not equal to . Prove that if is a UFD
and any rational homomorphism
is trivial.
6.11 Let be an algebraic group acting regularly on an algebraic variety and
let be a closed invariant subgroup of finite index. Suppose that a geometric quotient exists. Show that geometric quotients and
exist and . Is the same true without assuming that is of
finite index?
Chapter 7
Linearization of actions
7.1
Linearized line bundles
We have seen already in the proof of Lemma 3.5 that a rational action of an affine
algebraic group on an affine variety can be linearized. This means that
we can -equivariantly embed in affine space on which acts via a linear

representation. We proved this by considering the linear space spanned by the translates of generators of the algebra . In this chapter we will do a similar
construction for a normal projective algebraic variety. This will be our main tool
for constructing quotients.
Recall that a regular map of a projective variety to the projective space

is defined by choosing a line bundle (or equivalently an invertible sheaf of
-modules, or a Cartier divisor ) and a set of its sections . The map
is defined by sending to the point . This point is

well-defined if for any there is a section such that . Often
we will be taking for a basis of the space of sections of .
The condition above says in this case that for any there exists a section
such
that . We say in this case that is base-point-free.
be a map defined by a base-point-free . Of course, it depends
Let
on the choice of a basis; different choices define maps which are the same up to
is a closed embedding,
composition with a projective transformation of . If

is called very ample. If is very ample for some , then is
called ample.
We will often identify with its total space , which comes with a projection ; locally is the product of and the affine line
103
CHAPTER 7. LINEARIZATIONS OF ACTIONS
104
.
Definition. A -linearization of is an action

(i) the diagram
id
such that
is commutative,
(ii) the zero section of is -invariant.
-linearized line bundle (or a line -bundle) over a -variety is a pair
consisting of a line bundle over and its linearization. A morphism of
-linearized line bundles is a -equivariant morphism of line bundles.
It follows from the definition that for any and any the induced
map of the fibers

A
is a linear isomorphism.
We can view the set of such isomorphisms as an isomorphism of line bundles

where we consider as an automorphism
of . The axioms of the
actions translate into the following 1-cocycle condition:

(7.1)
The collection of the isomorphisms can also be viewed as an isomorphism
of vector bundles
pr
The cocycle condition (7.1) is translated into a condition on which can be expressed by some commutative diagrams; this is left to the reader.
Using the definition of linearization by means of an isomorphism it is easy
to define an abelian group structure on the set of line -bundles. If pr
and pr are two line -bundles, we define their tensor
7.1. LINEARIZED LINE BUNDLES
105
product as the line bundle

with the
phism:

pr pr
-linearization given by the isomor-
pr
Here we use the obvious property of the inverse image
The zero element in this group is the trivial line bundle whose linearization
. This is called the
is given by the product id
trivial linearization. The inverse

is equal to with defined as
the inverse of the transpose of . One checks that this again satisfies the cocycle
condition. The structure of an abelian group which we have just defined induces
an abelian group structure on the set of isomorphism classes of line -bundles.
We denote this group by Pic . It comes with the natural homomorphism
Pic
Pic
which is defined by forgetting the linearization.

Let us now describe the kernel of the homomorphism . Observe first that
is an isomorphism of line bundles and pr
if
is a -linearization of , then we can define a -linearization of by setting

is isomorphic to the trivial bundle,
pr . Thus, if
we can replace it by an isomorphic line -bundle to assume that is trivial. This

shows that Ker consists of isomorphism classes of linearizations on the trivial
line bundle .
We denote a point of by . For any ,

where . The function must be a regular function
on which is nowhere vanishing. In other words, . The

axioms of the action give us that

Let us see when two functions define isomorphic linearizations.
(7.2)
Let
be an automorphism of the trivial bundle. It is defined by
106
a formula , where
defined by and if and only if

It commutes with the actions
Or, equivalently, for any ,
Let alg denote the group of functions satisfying (7.2) considered as

a subgroup of the group and let alg be its subgroup con
sisting of functions of the form for some . It follows from the
definition of the group structure on Pic that the product in alg

corresponds to the tensor product of linearized line -bundles. So the above discussion proves the following.
Theorem 7.1. The kernel of the forgetful homomorphism
is isomorphic to the group
Pic
Pic

alg
alg

alg
Note the special case when for any integral -algebra

This happens, for example, when is affine space, or when
proper over . Then
pr
is connected and
and (7.2) gives that

alg
Homalg
the subscript indicating that we are considering rational homomorphisms of algebraic group. The latter group is called the group of rational characters of .
and hence
We studied this group when was a torus. Also we have
alg . Thus we obtain

Corollary 7.1. Assume
pr . Then
Ker

THE EXISTENCE OF LINEARIZATIONS
107
Remark 7.1. According to a theorem of Rosenlicht ([93]), for any two irreducible
algebraic varieties and over an algebraically closed field , the natural homomorphism

is surjective. Let us give a sketch of the proof. First we use that for any irreducible
algebraic variety the group is finitely generated. (This is not difficult
to prove by reducing to the case of a normal variety and then finding a complete
is a divisor. Then for
normal variety containing such that
any the divisor of is contained in the support of and hence
is equal to a linear combination of irreducible components of . This defines an
injective homomorphism from the group to a finitely generated abelian
group.) Now assume we have an invertible function on s . For a fixed

we have a function . Since is a

finitely generated group, the map
modulo must
be constant. Of course to justify this we have to show that this map is given by
an algebraic function; this can be done. So assuming this, we obtain that

is equal to a function up to a multiplicative factor depending on . So

as asserted.
7.2
The existence of linearization
To find conditions for the existence of a -linearization of a line bundle we have to

study the image of the forgetful homomorphism . This consists of isomorphism
classes of line bundles on which admit some -linearization. We start with the
following lemma.
Lemma 7.1. Let be a connected affine algebraic group, and let be an algebraic -variety. A line bundle over admits a -linearization if and only if
there exists an isomorphism of line bundles pr .
Proof. We already know that this condition is necessary, so we show that it is sufficient. Assume that such an isomorphism exists. The problem is that it may not
satisfy the cocycle condition (7.1). Let us interpret as a collection of isomor . When , the unity element, we get an automorphism
phisms
r . It is given by a function . Composing all with ,

we may assume that
id . Now the isomorphisms and
differ by an automorphism of . Denote it by so that we have

108
The cocycle condition means that id . So far we have only that
id for any . Let us identify the automorphism

with an invertible function on . By Rosenlichts Theorem which
we cited in Remark 7.1, we can write . Since

and
, the functions and are
constants. Thus is constant and hence and are constants. This implies
that
Remark 7.2. The existence of an isomorphism pr means that
is a -invariant line bundle. So the preceding lemma asserts that any -invariant
line bundle admits a -linearization provided that is a connected algebraic
group. The assertion is not true if is not connected. For example, assume
that is a finite group. The functions which we considered in the preceding proof form a 2-cocycle of with values in (with trivial action of in
). The obstruction for the existence of a -linearization lies in the cohomology
group . The latter group is called the group of Schur multipliers of .
It has been computed for many groups and, of course, it is not trivial in general.
If we denote the subgroup of -invariant line bundles by Pic , then one has
an exact sequence of abelian groups
Hom
Pic
Pic
(7.3)
Lemma 7.2. Assume that is normal (for example, nonsingular) and is a

connected affine algebraic group. Let . For any line bundle on
we have
pr pr q
Proof. It is enough to show that pr pr for some Pic

and Pic ; then it is immediately checked that and
. To do this we use the following fact about algebraic groups: contains
an open Zariski subset isomorphic to . For GL this follows from
the fact that any matrix with nonzero pivots can be reduced to triangular form by
elementary row transformations. We also use the fact that the homomorphism pr
is an isomorphism (see [46], Chapter 2, Proposition

Pic Pic
6.6). These two facts imply that pr for some line bundle on
(i.e.,
. Let be a Cartier divisor on representing
).
Then the preceding isomorphism implies that there exists a Cartier divisor on
such that pr . For every irreducible component

of its image in is contained in the closed subset
. By the theorem
THE EXISTENCE OF LINEARIZATIONS
109
on the dimension of fibers, the fibers of pr must be of dimension equal
. This easily implies that pr , where . Thus

to
pr for some Weil (and hence Cartier because is nonsingular)

divisor on .
So we have the equality of Cartier divisors pr pr . This translates

into an isomorphism of line bundles pr pr .

Define now a homomorphism Pic
Pic by
pr
where is a chosen point in . Suppose is trivial. By the preceding lemma
applied to pr we obtain that pr . But the
restriction of and pr to are equal. This implies that is trivial, hence
there exists an isomorphism pr . By Lemma 7.1, admits a
-linearization. This proves

Theorem 7.2. Let be a connected affine algebraic group acting on a normal
variety . Then the following sequence of groups is exact
Ker
Pic
Pic
Pic
Corollary 7.2. Under the assumption of the theorem, the image of Pic in
Pic is of finite index. In particular, for any line bundle on there exists a
number such that admits a -linearization.

Proof. Use the fact that for any affine algebraic -group
Pic is finite (see [65], p.74).
the Picard group
Remark 7.3. The assertion that Pic is finite can be checked directly for many
groups. For example, the group is trivial for GL r since these groups
are open subsets of affine space. To compute Pic for PGL SL we use
the following facts. Let be an irreducible hypersurface of degree in . Then

Pic
(7.4)
This isomorphism is defined by restricting a sheaf to an open subset. Another fact,

which is not trivial, is that
Pic
(7.5)
where is the class of a hyperplane section of , provided . This is called
the Lefschetz Theorem on hyperplane sections (see [40], p. 169).
110
Now notice that PGL is isomorphic to

. This gives
the determinant equation
Pic PGL
, where
is given by
On the other hand, SL is isomorphic to the complement of a hyperplane section

of the hypersurface
in
. So when we can apply (7.4) to obtain

Pic SL
There is a notion of a simply connected algebraic group (which makes sense over
an arbitrary algebraically closed field). For all such groups Pic is trivial. Any
is isomorphic to a quotient , where is simply connected and is a finite
abelian group whose dual abelian group is isomorphic to Pic . For example,
. For simple algebraic groups Pic is a subgroup of
SL for PGL
the abelian group

defined by the Cartan
matrix of the root system of the
for different types of simple Lie

Lie algebra of . Here are the values of
algebras:
We refer to [86] for a description of the Picard group of any homogeneous space
.
7.3
Linearization of an action
Now we are ready to prove that any algebraic action on a normal quasi-projective
variety can be linearized. Let be a -linearized line bundle, let
be its space of sections, and let be an affine algebraic group. The group acts
naturally and linearly on by the formula
or, in simplified notation,
(7.6)
LINEARIZATION OF AN ACTION
111
We know that any finite-dimensional subspace of is contained in a invariant finite-dimensional subspace generated by the translates of a basis
of . Thus we obtain a linear representation
GL
Assume that the linear system is base-point-free (i.e., for any
exists such that ). Then defines a regular map

by the formula

q
there
Here we identify a point in with a hyperplane in . Note that although

does not make sense (since it depends on a local trivialization of ), the
equality is well-defined. The representation (7.6) in defines a representation in and the induced projective representation in . It is given
by the formula
where
is a hyperplane in
. Now

q
q
This shows that the map

is -equivariant.
in we obtain a -equivariant rational map
Choosing a basis
If the rational map defined by a basis of is an embedding, then this map is

be an embedding of as a locally
an embedding too. Now let
closed subvariety of projective space. We take . When is large
enough, admits a -linearization. Let be the
image of under the canonical restriction map
. Obviously, is a finite-dimensional base-point-free linear system.

It defines an embedding of into projective
space which is the composition of

and a Veronese map (obtained from the Veronese map
Pol by choosing bases). Replacing with a invariant linear system as above, we obtain a linearization of the action of
on .
112
Theorem 7.3. Let be a quasi-projective normal algebraic variety, acted on by

an irreducible algebraic group . Then there exists a -equivariant embedding
, where acts on via its linear representation GL
Example 7.1. Let PGL act on in the natural

us see
way. Let
that the line bundle is not -linearizable but

is. We view as an open subset of the projective space
complement is the determinant hypersurface given by the equa whose

. The action r is the restriction to of the
tion
rational map

Note that this map is not defined at any point such that
. The restriction of the projection to the set of such points is

a birational map onto the determinant hypersurface (it is an isomorphism over the
in
subset of matrices of corank equal to 1). Since is of codimension
. The
the line bundle is the restriction of a line bundle on
formula for the action shows that this bundle must be pr pr .
is isomorphic to the restriction

Thus restricted to
of
to
. If admits a linearization, we have
pr , and hence the latter line bundle must be trivial. However, by (7.4),
.
it is a generator of the group Pic
The existence of a linearization of some power of a line bundle on a normal complete algebraic variety was first proven in [73] by using the theory of Picard varieties for complete normal varieties. Our proof, which is borrowed from [65], does
not use the theory of Picard varieties and applies to any normal quasi-projective
varieties. One can also consider vector -bundles of arbitrary rank (see for example [99]); however, no generalization of Corollary 7.2 to this case is known to
me.
EXERCISES
113
Exercises
7.1 Let be a line bundle over a connected affine algebraic group. Show that the
complement of the zero section of has the structure of an algebraic group
such that the projection map is a homomorphism of groups with
kernel isomorphic to .
7.2 Let be a connected affine algebraic group. Show that alg is a
homomorphic image of the group . In particular it is trivial if is connected

and complete.
7.3 Use Rosenlichts Theorem from Remark 7.1 to show that any invertible regular
function on a connected affine algebraic group with value 1 at the
defines a rational character of .
unity
7.4 Let be a nonsingular algebraic variety and let be its finite group of automorphisms. Show that the group Pic is isomorphic to the group of invariant Weil divisors modulo linear equivalence defined by -invariant rational
functions. [Hint: Use Hilberts Theorem 90 which asserts that

]
7.5 Let
grading
act on an affine algebraic variety defining the corresponding
of . Let be a projective module of rank 1 over and let be the

associated line bundle on . Show that there is a natural bijective correspondence
between -linearizations of and structures of -graded modules on .
7.6 Show that any line bundle on a normal irreducible variety on which SL
acts admits a unique SL -linearization.
7.7 Let be a -equivariant map, where acts on via its
linear representation. Show that admits a -linearization and
the map is the map given by the line bundle .
7.8 Show that the total space of the line bundle is isomorphic to the

. Describe the unique SL -linearization of in
complement of a point in
terms of an action of the group SL on the total space.
Chapter 8
Stability
8.1
Stable points
From now on we will assume that is a reductive algebraic group acting on an

irreducible algebraic variety . In this chapter we will explain a general construction of quotients due to D. Mumford. The idea is to cover by open affine
-invariant sets and then to construct the categorical quotient by gluing
together the quotients . The latter quotients are defined by Nagatas Theorem. Unfortunately, such a cover does not exist in general. Instead we find such a
cover of some open subset of . So we can define only a partial quotient .
The construction of will depend on a parameter, a choice of a -linearized line
bundle .
Definition. Let be a -linearized line bundle on
(i)
(ii)
(iii)
and
is called semi-stable (with respect to ) if there exists and

such that is affine and contains .
is called stable (with respect to ) if there exists as in (i) and additionally

is finite and all orbits of in are closed.
is called unstable (with respect to ) if it is not semi-stable.
We shall denote the set of semi-stable (respectively stable, unstable) points by
ss
115
us
116
CHAPTER 8. STABILITY
Remark 8.1. 1. Obviously ss and s are open -invariant subsets (but

could be empty).
2. If is ample and is projective, the sets are always affine, so this
condition in the definition of semi-stable points can be dropped. In fact, for any
, so we may assume that is very ample. Let
be a closed embedding defined by some complete linear system associated to .
Then is equal to the inverse image of an affine open subset in which is the
complement to some hyperplane. Since a closed subset of an affine set is affine
we obtain the assertion.
3. The restriction of to ss is ample. This is a consequence of the
following criterion of ampleness: is ample on a variety if and only if there
exists an affine open cover of formed by the sets , where is a global section
of some tensor power of . For the proof we refer to [46], p. 155.
4. The definitions of the sets ss s s s us do not change if we
replace by a positive tensor power (as a -linearized line bundle).
5. Assume is ample. Let ss be a point whose orbit is closed
and whose isotropy subgroup is finite. I claim that s . In fact let
definition of semi-stable points. Then the set

beas inis the
closed in and does not intersect . Since is reductive,
there exists a function such that . One can
show that there exists some number such that extends to a section
of some tensor power of (see [46], Chapter 2, 5.14). Since is irreducible,
this section must be -invariant. Thus and all points in have
zero-dimensional stabilizer. This implies that the orbits of all points in are
closed in . This checks that is stable.
6. In [73] a stable point is called properly stable and in the definition of stability the finiteness of is omitted.
Let us explain the definition of stability in more down-to-earth terms. Assume that is very ample, and embed equivariantly in . We have a equivariant isomorphism of vector spaces
Pol
where is the subspace of Pol

which consists of polynomials vanishing on
. Passing to invariants, we obtain
Pol
Let denote a point in such that . Every

can be represented by a polynomial Pol which is -invariant modulo
8.2. THE EXISTENCE OF A QUOTIENT
117
. In particular, is constant on the orbit of for any point . Clearly

if and only if does not vanish on . So the set of unstable points is
equal to the image in of the set

Pol
This set is called the null-cone of the linear action of in . It is an affine variety
given by a system of homogeneous equations (an affine cone). Let and O
be its orbit in . Suppose O . Then for any -invariant polynomial we
have O . Thus the corresponding point in
is unstable. Conversely, if is unstable, O . In fact, otherwise we can
apply Lemma 6.1 and find an invariant polynomial such that
but
. If we write as a sum of homogeneous polynomials of positive

degree, we find some which does not vanish at . Then is semi-stable. This
interpretation of stability goes back to the original work of D. Hilbert ([47]).
8.2
The existence of a quotient
Let us show that the open subset of semi-stable (respectively stable) points admits
a categorical (respectively geometric) quotient.
First we have to recall the definition of the gluing construction of algebraic
varieties. Let be a finite set of affine algebraic varieties. The gluing data
is a choice of an open affine subset for each , and an isomorphism

for each pair . It is required that
(i) , and is the identity for each ,
and
(ii) for any ,
Let be an equivalence relation on the set defined by if and
and .
only if there exists a pair such that
The assumptions (i) and (ii) show that it is indeed an equivalence relation. Let
be the corresponding factor set and let be the canonical
projection. We equip with the topology for which a subset is open if and
only if is open in the Zariski topology. The restriction of to
defines a homeomorphism of with an open subset of so that
and We also introduce the notion of a regular function on an
118
open subset . By definition, this is a collection of regular functions

on such that for any . Let
is
be the -algebra of regular functions on . The assignment
a sheaf of -algebras, called the structure sheaf of . The pair is an

example of a ringed space, i.e., a topological space equipped with a sheaf of
rings. Ringed spaces form a category. A morphism of ringed spaces
is a continuous map such that for any open subset
and , the composition . Each open subset of
is equipped with the structure of a ringed space whose structure sheaf is
equal to the restriction of to . Each quasi-projective algebraic variety can be
considered as a ringed space, the structure sheaf is the sheaf of regular functions.
It follows from the definition that the ringed space obtained by gluing
of affine varieties is locally isomorphic to an affine variety, i.e., it admits an open
cover by subsets which are isomorphic to affine varieties as ringed spaces; in the
notation from above each open set is isomorphic to . Thus we are led to the
notion of an abstract algebraic variety which is a ringed space locally isomorphic
to an affine algebraic variety. One usually adds a separatedness property which
ensures that the intersection of two open affine subsets is an affine variety. An
abstract algebraic variety is isomorphic to a quasi-projective algebraic variety
if and only if there exists an ample line bundle over which is used to embed
into projective space. We leave it to the reader to define the notion of a line
bundle over an abstract algebraic variety. A useful criterion of ampleness of a line
bundle was given in Remark 8.1.3.
Theorem 8.1. There exists a good categorical quotient
ss ss
and the
There is an open subset
in ss such that s
restriction of to s is a geometric quotient of s by . Moreover there
exists an ample line bundle
on ss such that , restricted
to ss , for some . In particular, ss is a quasi-projective variety.
Proof. Since any open subset of is quasi-compact in the Zariski topology we
can find a finite set of invariant sections of some tensor power of
such that ss is covered by the sets . Obviously we may assume that all
the belong to for some sufficiently large . Let

. For every , we consider the ring of invariant regular functions and let
with as constructed in
8.2. THE EXISTENCE OF A QUOTIENT
119
Nagatas Theorem. For each we can consider as a regular -invariant
function on . Let be the corresponding regular function on the

quotient. Consider the open subset . Obviously

and
This easily implies that both sets

are categorical quotients
of
. By the uniqueness of categorical quotient there is an isomorphism
. It is easy to see that the set of isomorphisms

satisfies the conditions of gluing. So we can glue together the quotients and
the maps to obtain a morphism ss , where ss . To
show that is separated it is enough to observe that it admits an affine open cover
by the sets which satisfies the following properties:

are
affine and is generated by restrictions of functions from and
is generated by
. The latter property follows from the fact
that
restrictions of functions from and . In fact, the separatedness also

follows from the assertion that is quasi-projective. So let us concentrate on
proving the latter.
Note that the cover

of ss is a trivializing cover for the line

bundle obtained by restriction of to ss . In fact, by Remark 8.1.3, is

ample; hence we may assume that some tensor power is very ample. This
implies that is equal to the line bundle for some embedding

section of is equal to the section where is
ss . The
a section of . Thus the open subset is equal to where is an
open subset of isomorphic to affine space. This shows that restricted to
is equal to . However, is isomorphic to the trivial

line bundle since any line bundle over affine space is isomorphic to the trivial
bundle. By fixing some trivializing isomorphisms we can identify the functions
with the transition functions of . As we have shown before,
for some functions . We use the transition functions
to define a line bundle on . Obviously . Let us

show that is ample. First we define some sections by setting for
a fixed and variable . Since for any
the differ by the transition function of , hence is in fact a section

of . Clearly and . As above, since all are affine, we
obtain that is ample. Since ss is obtained by gluing together

good categorical quotients, the morphism is a good categorical quotient.
120
It remains to show that the restriction of to s is a geometric quotient.

By definition s is covered by affine open -invariant sets where acts with
closed orbits. Since is a good categorical quotient, for any s the fiber
consists of one orbit. Thus s is a good geometric quotient.
In the case when is ample and is projective, the following construction of

the categorical quotient ss is equivalent to the previous one.
Proposition 8.1. Assume that
is projective and is ample. Let
Then
ss Proj
In particular, the quotient ss is a projective variety.
Proof. First of all, we observe that by Nagatas Theorem the algebra is finitely
generated. It also has a natural grading, induced by the grading of . Replacing by we may assume that is generated by elements
of degree . Let Projm be the projective subvariety of corresponding to the homogeneous ideal equal to the kernel of the homomorphism
. (The reader should go back to Chapter 3 to recall
the definition of Projm for any finitely generated graded -algebra .) The
elements generate the ideal generated by homogeneous elements of
positive degree. Thus the affine open sets cover ss . On the other
hand the open sets form an open cover of with the prop
erty that . The maps define a morphism ss

which coincides with the categorical quotient defined in the proof of the preceding
theorem.
Remark 8.2. If we assume that is very ample, and embeds in the projective
space = , then we can interpret the null-cone as follows. The

sections from the proof of the preceding proposition define a -equivariant
rational map s . The closed subset of where
,
this map is not defined is exactly the closed subvariety
of equal to
where the bar denotes the image of the null-cone in . So deleting this closed subset from we obtain the set ss and the quotient map
ss ss .
8.3. EXAMPLES
121
Remark 8.3. Note that the morphism ss ss is affine, i.e., inverse

image of an affine open set is affine. There is also the following converse of the
preceding proposition. Let be a -invariant open subset of such that the
exists and is an affine map. Assume

is
geometric quotient
quasi-projective. Then there exists a -linearized line bundle such that

. For the proof we refer to [73], p. 41.
8.3
Examples
Example 8.1. Let and SL acting on naturally via its linear

representation. We know that admits a unique SL linearization
(Exercise 7.7). We also know from Chapter 5 that Pol is an irreducible

representation for . Therefore, for any ,
Pol
This shows that ss

Example 8.2. Let ,
with action defined by the formula

Here
are some integers. We assume that
. Since
Pic
and we have Pic . A -linearized
bundle must be of the form ; it defines a -equivariant Veronese embed
ding , where
. The

group acts on
by the formula
where
is the coordinate in the Veronese space corresponding to the monomial
,
. Now the linearization is given by a linear representation of in the space
which lifts the action in the correspond
sr
sr
ing projective space. Obviously it is defined by the formula
(8.1)

for some integer . Thus the -linearized bundles can be indexed by the pairs
. Denote the corresponding line bundle by . Raising to the
th power as a -linearized bundle corresponds to replacing with .
We know that ss does not change if we replace by . So we may
assume that
, where by definition is defined only for
sr
122
In other words we permit

divisible by t and
r
to be a rational number in formula (8.1) and consider invariant polynomials of

degree a multiple of the denominator of . Here the invariance means that for any
,

. It is obvious that for all
Assume now that
if .
. This implies that ss
or
When , we have

if
if t
. Hence
ss
and
ss Projm
In particular, if , the quotient is the point.
Next, we increase the parameter . If

, we have further
invariant polynomials. For example, if , the monomial
ss
belongs to
.
So
the
set
becomes
larger
and
the
categorical quotient changes. In fact one can show that the quotients do not change
when stays strictly between two different weights ts and do change otherwise.
Example 8.3. Consider the special case of the previous example where t and
t
t . The restriction of the action to is given by the formula

we get
for
If we take

This shows that us . In other words, the set of
semi-stable points is equal to the complement of the hyperplane at infinity
and the point . So it can be identified with
. The quotient
8.3. EXAMPLES
123
is of course . Since the group acts on this set with trivial stabilizers, we
obtain that all orbits are closed and the quotient is a good geometric quotient.
Similar conclusions can be made for any rational . If , we have

The categorical quotient is the same but the set of semi-stable points is differThus
us
ent.
Example 8.4. Let and . Every line bundle is isomorphic to the

trivial bundle . As we saw in Chapter 7, its -linearization is defined
by the formula
where is a homomorphism of algebraic groups. It is easy to see

that any such homomorphism
is given by
the formula s for some integer
. In fact is defined by the image of , and the

condition that this map is a homomorphism implies that the image is a power of
. So let denote the -linearized line bundle which is trivial as a bundle and
whose linearization is given by the formula
A section
of
is given by the formula

for some polynomial
The group acts on the space of
sections by the formula
, where

if and only if
Thus
for all
When , the constant polynomial defines an invariant section of for
any . Thus ss and
Specm Specm
124
Recall that a -action on an affine variety is equivalent to a -grading of its ring

of regular functions; the ring of invariants is the subring
of elements of degree 0
but the variables are not

(see Example 3.1). In our case

necessarily homogeneous. If we can make a linear change of variables such that
they are homogeneous, then the action is given by a formula
In this case we say that the action of on is linearizable . It is an open problem (a very difficult one) whether any action of on affine space is linearizable.
It is known to be true for .
Assume now that . Since we know that the set of semi-stable points
and the quotient do not change when we replace by its tensor power, we may
assume that . Then
The subring
Thus
is a finitely generated algebra over

is a finitely generated ideal in
. Let
generators. Then
ss
ss
Specm
be its homogeneous

s
(see Example 3.1).
where
.
Similar conclusion can be reached in the case
Example 8.5. A special case of the previous example is when

the formula
acts on
by
s
where . If , we get so the quotient is one point.
, we get , so the set of semi-stable points is empty.
If
Finally, if , we get
ss

8.3. EXAMPLES
125
and the construction of the categorical quotient coincides with the construction
of the weighted projective space t t (see Example 3.1). So we see two
different ways to define : as a quotient of and as a quotient of .

Example 8.6. Let be again
and with the action given by the formula

As in the previous example, each -linearized line bundle is isomorphic to the
trivial line bundle with the -linearization defined by an integer . We have

.
However, this time the grading is weighted; the weights are
. Hence
Assume
. Then for any
ss
, and

, defined by
We have a surjection
This shows that

Thus ss is isomorphic to the closed subvariety of
given by the
equation

This is a quadric cone. It has one singular point at the origin.

Assume . Again, without loss of generality we may take
easy to see that
Thus
ss
This set is covered by

. We have

and
. It is
126
We claim that ss
by the equations
of
is isomorphic to a closed subvariety
given

Here we use for homogeneous coordinates in . In fact, this variety
. It is easy
is covered by the two affine open sets given by
to see that . We also verify that these two sets are glued to
gether as they should be according to our construction of the categorical quotient.

Thus we obtain an isomorphism ss . In fact, we have
ss so that is a geometric quotient. Note

that we have a canonical morphism
which is given by the inclusion of the rings . Geometrically it

is induced by the projection . Over the open subset this
morphism is an isomorphism. In fact, is covered by the open subsets
. The inverse image is contained

in the open subset where . Since s we see that induces
an isomorphism . Similarly we treat the other pieces . Over
the origin, the fiber of is isomorphic to . Also, we immediately check that

is a nonsingular variety. Thus is a resolution of singularities of
is called a small resolution because the exceptional set is of codimension

. It. The
reader familiar with the notion of the blowing up will recognize
as the variety obtained by blowing up the closed subvariety of defined by the
equations .
is isomorAssume . Similar arguments show that
given by the equations

phic to the closed subvariety of
We have a morphism
which is an isomorphism over

. The diagram
and whose fiber over
is isomorphic to
127
represents a type of birational transformations between algebraic varieties which

nowadays is called a flip. Note that is not isomorphic to , but they are
isomorphic outside the fibers .
The theory of stable points with respect to an algebraic action was developed in
[73]. There is nothing original in our exposition. The examples given in the chapter show the dependence of the sets of stable points on the choice of linearization
of the action. Although this fact was implicitly acknowledged in [73], the serious
study of this dependence began only recently; see [23], [115] and the references
there. One of the main results of the theory developed in these papers is the finiteness of the set of open subsets which can be realized as the set of semi-stable
points for some linearization.
Exercises
8.1 Let be a homogeneous space with respect to an action of an affine algebraic
group . Assume is not affine. Show that for any Pic the set ss
is empty.
-linearized line bundle is called -effective if ss . Show that

is -effective if both and are -effective.
8.3 Let act on an affine algebraic variety and let

be the corresponding grading. Define

and
similarly
Show
. Let Pic be trivial as a liness bundle.

that there are only three possibilities (up to isomorphism):
s , where (resp. ) is the ideal in generated by
(resp.
). Show that in the first case ss is isomorphic to Specm ,
(resp.
in the second (resp. the third) case ss is isomorphic to Projm
Projm
).
is a nonsingular
8.4 In Example 8.6 show that the fibered product
variety. Its projection to is an isomorphism outside the origin, and the inverse
image of the origin is isomorphic to
. Show that the restrictions of the
to coincide with the two projection maps
projections from to

.
8.2 A
128
8.5 Let be a finite group acting regularly on . Show that for any Pic ,
ss s . Also s if is ample. Show that the assumption of
ampleness is essential (even for the trivial group!).
8.6 Let SL act by conjugation on the affine space
of matrices.
Consider the corresponding action of on the projective space . Find
the sets ss s where Pic .

8.7 Let be a closed -invariant embedding, and let where
is an ample -linearized line bundle on . Assume that is projective and
is linearly reductive, e.g. char( ) = 0. Prove that, for any ,

s

8.8 Consider Example 8.1 with

possible categorical quotients.
s s

and t
. Find all
Chapter 9
Numerical criterion of stability
9.1
The function
In this chapter we shall prove a numerical criterion of stability due to David

Hilbert and David Mumford. It is stated in terms of the restriction of the action
to one-parameter subgroups. The idea of the criterion is as follows. Suppose an
affine algebraic group acts on a projective variety via a linear representation GL . This can be achieved by taking a very ample -linearized
line bundle on . As in Chapter 8, we denote by a representative in

of a point . We know that us if and only if . If is
a subgroup of , then , so one may detect an unstable point by
checking that for some subgroup of . Let us take for the image
of a regular homomorphism (a one-parameter subgroup of ). In
appropriate coordinates it acts by the formula

Suppose all for which are strictly positive. Then the map

can be extended to a regular map
by sending the origin of
to
the origin of . It is clear that the latter belongs to the closure of the orbit of
, hence our point is unstable. Similarly, if all are negative, we change to
to reach the same conclusion. Let us
defined by the formula
set
129
CHAPTER 9. NUMERICAL CRITERION
130
So we can restate the preceding remark by saying that
if there exists in the set
of one-parameter subgroups of such that or

, then
is unstable. In other words, we have a necessary condition for semi
stability:
ss

(9.1)
Assume the preceding condition is satisfied and for some . Let us

show that is not stable. In the preceding notation, let ,
and let , where if , and if . Obviously,
belongs to the closure of the orbit of under the action of the subgroup .
If were stable, then by definition of stability, must be in this orbit. However,
obviously fixes , so that cannot be stable. Thus we obtain a necessary
condition for stable points:

(9.2)
are independent of a choice of
We have to show first that the numbers

coordinates in , and also that the previous condition is sufficient for semistability. Let us start with the former
task and do the latter one in the next section.
Let and be as above. Take a one-parameter subgroup

is equal to the
; for any the corresponding

point
where
point with projective coordinates

if and anything otherwise. Thus when we let go to , we obtain a point
in with
coordinates , where if and only if and
. The precise meaning of let go to is the following. We have a

map
Since
is projective this map can be extended to a unique regular map
Obviously

. Now it is clear that for any

So our point is equal to

We set
9.1. THE FUNCTION
131
that is, is a fixed point for the subgroup of . Also the definition of is
coordinate-free. Furthermore, for any vector over ,
(9.3)
This can be interpreted as follows. Restrict the action of on to the action of defined by . Then has a natural -linearization and, since
is a fixed point, acts on the fiber ; this defines a linear representation
GL We know the geometric interpretation of the total space
of the line bundle . It follows from this that the fiber of the
canonical projection over a point can be identified with
. Thus from (9.3) we get that acts on the fiber by

the character . Hence it acts on the fiber by the

rational character

. This gives us a coordinate-free
definition of . In fact, this

for any -linearized line bundle as
allows one to define
the number
follows. Let
. Then and, as above, there is a
representation
of on the fiber . It is given by an integer which is taken to be
In the case when , we can give another

coordinate-free geomet
ric interpretation of . Let
be the homogeneous ideal
defining in and be the homogeneous

coordinate ring
of . We have Projm . Let Specm

be the affine cone
over . Let and be as above. A one-parameter subgroup as above defines
a morphism
Let be the corresponding homomorphism of the rings of regular

functions. The image of the maximal
ideal defining the vertex of generates
a principal ideal . I claim that

(9.4)
with the canonical homomorphism
In fact, the composition of

is given by the formula , where

Since is generated by the cosets of the , we see that is generated by the
monomials such that . Now the assertion follows from the definition of
.
132
9.2
The numerical criterion
Now we are ready to prove the sufficiency of conditions (9.1) and (9.2). The
following is the main result of this chapter.
Theorem 9.1. Let be a reductive group acting on a projective algebraic variety
. Let be an ample -linearized line bundle on and let . Then
ss
for all
for all
Before starting the proof of the theorem, let us recall the notion of properness
of a map between algebraic varieties. We refer to [46] for the details.
Definition. A regular map of algebraic varieties over an algebraically
closed field is called proper if for any variety over the map id
is closed (i.e., the image of a closed subset is closed). A variety is proper

(or complete) over if the constant map Specm is proper.
We shall use the valuative criterion of properness (see [46]). For any algebraic
variety over , and any -algebra , the set of morphisms of algebraic varieties
Specm can be viewed as the set of points with values in . If
is affine, Hom s as was defined in section 3.3. If is glued
together from affine varieties , and is a field, then is glued together
from the .
Let be a discrete
valuation algebra over with residue -algebra isomorphic
to (e.g., is the algebra of formal power series over ) and let be

its field of fractions. If is glued together from affine varieties , then it is
separated if and only if the natural map is injective (the valuative
criterion of separatedness). In particular, it is always injective for quasi-projective
algebraic varieties, with which we are dealing. A regular map of
varieties over defines a map of -points. In particular,
induces a map , which is
the residue homomorphism
called the residue map. Then the valuative criterion of properness asserts that a
regular map is proper if for any , the natural map
is bijective.
Example 9.1. Any closed subvariety of is proper over . First of all is
proper over . Any -point of comes from a unique -point after multiplying
9.3. THE PROOF
133
its projective coordinates by some power of a generator of the max

imal ideal of . Now, it follows immediately from the definition of properness
that a closed subvariety of a proper variety is proper. On the other hand, an affine
variety is obviously not proper. Let us show that is

not complete. First notice that the point is a -point of
since
q . In fact, it belongs to any open subset
it corresponds to a homomorphism defined
by . However, this point does not come from any -point of . In
for any . Now

fact
and but .
We will need the following fact.
be the ring of formal
Lemma 9.1. (Cartan-Iwahori-Matsumoto) Let
power series with coefficients in and let

be its field of fractions.
For any reductive algebraic group , any element of the set of double cosets
can be represented by a one-parameter subgroup
in the following sense. One considers as a -point of and identifies
with a subfield of by considering the Laurent expansion of rational
functions at the origin of .
Proof. We prove this only for the case GL ; we refer to the original paper of
Iwahori and Matsumoto for the case char (see [55]). In the case of positive
characteristic one has to modify the lemma (see Appendix to Chapter 1 of [73] by
J. Fogarty).
A -point of is a matrix with entries in . We can write it as a matrix
, where GL . Since is a principal ideal domain, we can reduce

the matrix to diagonal form so that where , and
is the diagonal matrix diag . Now we can define a one-parameter

subgroup of by
diag
Then represents the double coset of the point
9.3
as asserted.
The proof
Let us prove Theorem 9.1. We have already proved the necessity of the conditions.
First of all, by replacing with a sufficiently high tensor power, we can place
134
ourselves in the following situation: acts on a projective space by means of

a linear representation GL , is a -invariant closed subvariety of
. We have to prove the following.
Let and s . Then there exists such that

. Moreover, if us then there exists such that

.
, remembering that
From now on we drop from the notation

for all . We have to show that s.
Assume

Suppose s . Choose a point over . Then the map
is not proper. In fact, if it is proper, is closed and the
fiber of over is proper over (Exercise 9.4). Since the fiber is a closed
subvariety of an affine variety, it must consist of finitely many points (Exercise

9.3). This easily implies that is finite and is closed, so that is a stable
point, contradicting the assumption. By the valuative criterion of properness, there
exists an -point of which, viewed as a -point of , has a inverse image
under but does not arise from any -point of . In other
words,
there exists an element such that
.
By Lemma 9.1 we can write , where , and
subgroup . Let be the image
which comes from a one-parameter

of under the reduction homomorphism corresponding to the
natural homomorphism . We can write
The expression in the parentheses is a -point of defined by a one-parameter

subgroup of . Choose a basis in such that the
action of is diagonalized. That is, we may assume that
This is equivalent to

Thus, if we write
, we obtain

Since
, this tells us that
(9.5)
9.4. THE WEIGHT POLYTOPE
135
This implies that if . In fact, the element is reduced to the

identity modulo , hence modulo are constants equal to .
On the other hand, they are equal to modulo for some . This of
course implies that if .
Recalling our definition

of we see that . This contradic
s
tion shows that if for all .
Assume now that for all . We have to show that ss . If
is unstable, and hence we can choose such that

is reduced to zero modulo (this follows immediately from the
proof of the valuative

criterion of properness). Therefore the left-hand
side of

if
. Thus .
(9.5) belongs to and hence we get
This contradiction proves the theorem.
9.4
The weight polytope
Recall from Chapter 5 that a linear representation of a torus

space splits into the direct sum of eigensubspaces
in a vector
where
Also recall from Chapter 5 that there is a natural identification between the sets
and which preserves the natural structures of abelian groups on both
sets. We define the weight set of the representation space by setting

This is a finite subset of . Its convex hull in is called the weight polytope and
is denoted by wt . Let us choose a basis of which is the sum of the bases of
the weight spaces wt . In this basis our representation is defined by a
homomorphism GL given by a formula

(9.6)
..
..
.. . .
..
.
.
.
.
.

wt
136
where we use the vector notation for a monomial

.
Now let be a one-parameter subgroup of . It is given by a
s s for some . Composing the
formula
representation with we have a representation GL given by the
formula
..
.
..
.
..
.
..
..
.
(9.7)
. We define the weight set of by

Let with
setting

(9.8)
wt
We define the weight polytope of by setting

wt convex hull of wt( ) in
(9.9)
If we choose coordinates in as in (9.7) and write then
wt
s we obtain that
Since

wt
between and
(Recall that the natural bilinear pairing
. When we identify and
is defined by the composition
with , it corresponds to the usual dot-product.)
Example 9.2. Let be the subgroup of diagonal matrices in GL . Consider its
natural representation in . Then wt , where are

the
unit basis vectors. Each corresponds to the character diag
The weight space is the coordinate axis . The weight polytope of is the
standard simplex
The weight set of a point with projective coordinates
set Its weight polytope is the subsimplex
given by corresponding to diag
is the
. If is
, then
9.4. THE WEIGHT POLYTOPE

Clearly, one can always find
unstable.
137
such that this number is positive, so all points are
In the case when is a torus we can restate Theorem 9.1 in the following way.
Theorem 9.2. Let be a torus and let

on a projective -variety . Then
ss
s
be an ample
wt s
interior wt
-linearized line bundle
Proof. We use a well-known fact from the theory of convex sets.

Let
be a
closed convex subset of . For any point

interior
(resp.
there exists an affine function such that (resp.
the proof of this fact shows that one can

), and q Moreover,
choose with integral coefficients if is the convex hull of a set of points with
integral coordinates. We refer for the proofs to any textbook on convex sets (see
for example [82]). The result follows.
Now let be any reductive group acting linearly on a projective variety
, and be the restriction to of some positive tensor power of . We
know that any one-parameter subgroup of has its image in a maximal torus of
, and hence can be considered as a one-parameter subgroup of . Now, applying
Theorem 9.1, we obtain
ss
ss s
maximal tori
maximal tori
Here runs over the set of all maximal tori of , and the subscript indicates
the restriction of the action (and the linearization) to .
Let us fix one maximal torus . Then for any other maximal torus , we can
find such that . From the preceding chapter we know that
is semi-stable (resp. stable) with respect to if and only if
(resp. is closed and the stabilizer of in is finite). It immediately follows that this property is satisfied if and only if is semi-stable (resp.
stable) with respect to . This implies that
ss
ss
and similarly for stable points. Putting these together we obtain
138
Theorem 9.3. Let be a maximal torus in . Then
ss
s
9.5
ss s
s
Kempf-stability
To finish this chapter we give a very nice necessary condition for a point to be
unstable in terms of its isotropy subgroup. This is a result due to G. Kempf which
is very important in applications to construction of various moduli spaces in algebraic geometry. Let , where acts on via a linear representation in
. Suppose is unstable. Let be its representative in . We know that
there is a one-parameter subgroup such that

.
We call a destabilizing one-parameter subgroup of . Among all destabiliz
ing one-parameter subgroups of we want to consider those for which
, we should first normalize by diis maximal. Since

viding it by and show that the maximum is defined. Here means
the Euclidean norm in if we choose to identify with ; of course, the
image of could belong to different maximal tori, so we have to proceed more
carefully. First we can fix one maximal torus . For any we can find
such that belongs to . Then we can set

However, we have to check that this definition does not depend on the choice of
as above; equivalently, we have to check that if (i.e.,
belongs to the normalizer of in ). The quotient group

is called the Weyl group of . It is a finite group which acts linearly on .
If GL and is the subgroup of diagonal matrices, we easily check that
can be represented by the matrices , where
is a root. By conjugation, acts on by permutation

of rows and hence it
acts on by permutation of the coordinates. In particular, is
-invariant. In general we choose a norm on which is -invariant;
this is always possible since is finite. This solves our problem of defining
for any . So we set
For any
we define

exists in
9.5. KEMPF-STABILITY
139
is a subgroup of which contains a Borel subgroup. More ,

s
Proof. Again we prove this only for GL . Without loss of generality we may
assume that is a one-parameter subgroup of the group of diagonal matrices and
is given by diag By a further change of basis we may also
assume that
. Let We have

The limit exists if and only if when
. Thus if and only
if whenever
and
. It is easy to see that is a subgroup;
it contains the group of upper triangular matrices and is equal to this group if
s form a set of
. Now the limits
matrices such that

if
. It is immediately checked
that this is the subgroup .
Lemma 9.3. For any ,

Proof. We have, for any ,

where
. It is easy to see that
(see Exercise 9.2(iv)). Therefore, putting
, we obtain

Lemma 9.2.
over, for any
Now
Here we use that centralizes

(i)). This proves the assertion.
and

(see Exercise 9.2
140
Definition. The flag complex of is the set

modulo the following equivalence relation:
of one-parameter subgroups of
such that
It follows from Lemma 9.2 that the function is well-defined as a function
on . Also the function is well-defined on . Now

the idea is
to find a maximum of . It is achieved at a point representing

the one-parameter subgroup which is most responsible for the instability of .
The existence of such a point was conjectured by J. Tits and was proven by
G. Kempf ([59]) and G. Rousseau
([95]). The idea is to show that is strictly
convex on the set of points in representing destabilizing subgroups of and

achieves a maximum on this set.
Theorem 9.4. There exists a one-parameter subgroup
such that
.
is called adapted for the point
All such subgroups represent the same point in
Definition. A one-parameter subgroup

us if it satisfies the assertion of the preceding theorem.
Let be the set of adapted one-parameter

subgroups of . It is an equiv
. We can assign to it the unique
alence class representing one point
parabolic subgroup which we denote by . Of course we have to remember that all of these objects depend on the linearization of the action.
Corollary 9.1. Assume is unstable. Then

. Indeed
Proof. For any and we have

. It follows from the definiBy Theorem 9.4, we must have

. However, it is known that the normalizer of a
tion that
parabolic subgroup is equal to the subgroup.
Corollary 9.2. Assume is semisimple (e.g. SL ) and is not contained
in any proper parabolic subgroup of . Then is semi-stable with respect to any

linearization.
9.5. KEMPF-STABILITY
141
Proof. We use that if is semisimple. Otherwise there is an adapted

one-parameter subgroup which belongs to the center of .
In fact, one can strengthen the preceding corollary by showing that is

closed in if is not contained in any proper parabolic subgroup of . This is
due to Kempf ([59]). To prove it he considers a closed orbit
in O and
proves the existence of a one-parameter subgroup with

.
Next he defines the set of adapted subgroups with this property for which the limit
is reached the fastest. These subgroups define a unique proper parabolic subgroup
and is contained in this subgroup.
Definition.
is called Kempf-stable if
is closed in .
This definition is obviously independent of the choice of

. Note that
stability
Kempf-stability
semi-stability
representing

Indeed, if is closed in ss then is obviously closed in
(otherwise the image in

of a point in the closure belongs to the closure of
in ss). Also is closed in since otherwise a point in its closure
belongs to the null-cone and hence any invariant polynomial will vanish at .
Now if is Kempf-stable, the point cannot belong
to the null-cone. If it does,
we can find a one-parameter subgroup such that

. But then
must belong to , which is absurd since is an orbit.
Thus we can generalize Corollary 9.2 to obtain:
Corollary 9.3. Assume is semisimple and is not contained in any proper
parabolic subgroup of . Then is Kempf-stable.
Example 9.3. This is intended for the reader with some knowledge of the theory
of abelian varieties (see [72]). Let be an abelian variety of dimension over
an algebraically closed field and let be an ample divisor on . One defines
the subgroup of which consists of all points such that .
Here denotes the translation map . Although is obviously

invariant, it does not admit a -linearization. However, one defines a certain
extension group with kernel isomorphic to , with respect to
which admits a linearization. Of course, the subgroup of acts trivially
on . The group is called the theta group of . The linear representation of in is irreducible. As an abstract group is isomorphic to , where diag
142
is the type of the polarization of . For example, when , where

is a principal polarization,
we have , the group of -torsion points,
and . The vector space is isomorphic to the vector space of -valued functions on the finite abelian group ,
and the representation of on this space is called the Schrodinger represen
tation. If we assume that
, then is very ample and can be used to
define a -equivariant embedding of in . Let us now consider an abelian variety with polarization of type and level structure as a triple
, where and are as above, and is an isomorphism
of abelian groups. Each such triple defines a point in the Hilbert scheme

of closed subschemes in . We say that two triples
and are isomorphic if there exists an isomorphism of abelian varieties
such that and . It is easy to see from this
definition that if and only if for

some projective transformation of . One can show that there is an irreducible
component of the Hilbert scheme which contains the points . Since
the

space corresponds to an irreducible representation of the
group , the isotropy subgroup of (equal to ) is not contained

in any proper parabolic subgroup of GL (see Exercise 9.10). Thus is

a Kempf-stable point in . It is also a stable point since its isotropy subgroup is
finite. The set of points in corresponding to smooth schemes is an open subset
of , and is also a GL -invariant subset contained in s . Thus we can con

sider the geometric quotient GL which is a fine moduli scheme for abelian
varieties with polarization of type and a level structure.

Most of the material of this chapter is taken from [73]. Our function
differs by a minus sign from the one studied in Mumfords book [73]. The numerical criterion of stability goes back to D. Hilbert ([47]) who introduced it for the
description of the null-cone for the action of SL on the space of homogeneous
polynomials.
One can give a criterion of stability in terms of the moment map
Lie , where is a maximal compact subgroup

of (SU if SL ).
It is defined by the
formula d , where, for any ,
. Here we fixed a -invariant hermitian norm in . The
criterion states that is semi-stable if and only if belongs to the closure of the
EXERCISES
143
moment map image of the orbit of (see [60]). For more information
about the relationship between GIT and the theory of moment maps we refer to
[62] and Chapter 8 of the
new edition of Mumfords book.

as a function in . One can also get rid of
the
One can consider
dependence on by showing that the function

is
well-defined and can be extended to a function on the vector space
Pic . These functions are used in [23] to define walls and chambers
in the vector space Pic which play an important role in the theory of
variation of GIT quotients.
Exercises
9.1 An algebraic group is called diagonalizable if is generated as algebra by the characters considered as regular functions on .
Prove that a torus is a diagonalizable group and that every connected diagonalizable group is isomorphic to a torus. Give examples of nonconnected diagonalizable groups.
:
for any
, the map Pic defined by the formula
9.2 Check the following properties of the function
(ii)
(iii)
-varieties, and
(iv)

for any

is a homomorphism of groups;
if
is a -equivariant morphism of
;
Pic , then
.
(i)
9.3 Prove that an affine variety over a field

set of points.
is proper if and only if it is a finite
9.4 Prove that a fiber of a proper map is a proper variety. Give an example of a
nonproper map such that all its fibers are proper varieties.
9.5 Prove that acts properly on
s is proper).
9.6 Prove Lemma 9.3 for
(i.e., the map
SL and
s s
144
9.7 Let be an -dimensional

torus acting linearly on a projective space . Show
is a
that Pic and the set of Pic such that ss

finitely generated semigroup of .
9.8 In the notation of Exercise 8.6 from Chapter 8, find the sets ss and s
by using the numerical criterion of stability.
9.9 Suppose is Kempf-stable. Show that its isotropy group is a reductive
subgroup of . [Hint: Use, or prove, the following fact: if is a closed subgroup
of with affine then is reductive.]
9.10 Let be a subgroup of GL such that is irreducible for the natural
action of in . Show that is not contained in any proper parabolic subgroup
of .
9.11 Let be the projective space associated to the space of square ma
trices of size . Consider the action of the group SL on defined by conjugation
of matrices. Using the numerical criterion of stability find the sets of unstable and
stable points.
9.12 Let and let act on via its linear representation.
Consider the
.
flag complex . For any point let
Show that this set is convex.
Chapter 10
Projective hypersurfaces
10.1
Nonsingular hypersurfaces
Let SL act linearly on in the natural way. This action defines an ac
of homogeneous polynomition of on the subspace
als of degree . We view the latter as the affine space , where .
A point of the projective space

is called a hypersurface of degree in . For each nonzero
we denote the corresponding hypersurface by . When is an irreducible
polynomial, it can be identified with the set of zeros of in , which is an irreducible closed subvariety of of dimension
. In general, can be
viewed as the union of irreducible subvarieties of dimension
taken with mul
Hyp
tiplicities. In this chapter we shall try to describe the sets of semi-stable and stable
points for this action. Note that there is no choice for a nontrivial linearization,
since Pic and ; we must take .

Let
Hyp SL
This is a normal unirational variety. According to a classical result of Jordan

and Lie, the group of projective automorphisms of an irreducible hypersurface of
degree is finite (see a modern proof in [85]). This implies that SL acts
on an open nonempty subset with finite stabilizer groups. By Corollary 6.2,
Hyp
SL
145
(10.1)
CHAPTER 10. PROJECTIVE HYPERSURFACES
146
Let be arbitrary. Recall that a hypersurface

gular variety if and only if the equations
Hyp
defines a nonsin-
have no common zeros. Note that, by the Euler formula,
So if char does not divide , the first equation can be eliminated. Let be the
resultant of the polynomials . It is a homogeneous polynomial of degree
in the coefficients of the form . It is called the discriminant of
. Its value at is equal to zero if and only if the have a common zero in
is independent of the choice of coordinates, the hy . Since the latter property
persurface
Hyp is invariant with respect to the action of SL .
This means that for any we have for some . One
immediately verifies that the function is a character of SL . Since
the latter is a simple group, its group of characters is trivial. This implies that
for all , and hence is an invariant polynomial. Since does not vanish on the set of nonsingular hypersurfaces of degree prime to the characteristic,
we obtain
Theorem 10.1. Assume char
semi-stable point of Hyp .
is prime to . Any nonsingular hypersurface is a
If
, one can replace semi-stable with stable. This follows from the
previously observed fact that, under these assumptions, the group of projective
automorphisms of a nonsingular hypersurface is finite.
Example 10.1. Assume
quadrics. The space
and
char . Then Hyp is the space of
is the space of quadratic forms
or equivalently, the space of symmetric matrices
10.2. BINARY FORMS
147
A quadric is nonsingular if and only if the rank of the corresponding

matrix
is the resultant
is equal to . The determinant function on
from above. Thus all nonsingular quadrics are semi-stable. We know that by
a linear change
of variables
every quadratic form can be reduced to the sum of
squares , where the number is equal to the rank of the matrix
from above. In our situation we are allowed to use only linear transformations
with determinant 1 but since we are considering homogeneous forms only up to a
multiplicative factor, the result is the same. We have exactly orbits for the action
of SL on Hyp ; each is determined by the rank of the corresponding nonzero
quadratic form. In fact any invariant nonzero homogeneous polynomial vanishes

on an invariant subvariety of codimension 1 in Hyp , which must consist of
all orbits except the unique open one representing nondegenerate quadratic forms.
By Hilberts Nullstellensatz, this invariant polynomial must be a power of the
discriminant
of the quadratic form. The stabilizer of the quadratic form
is the special orthogonal group SO . Since it is of positive dimension

(if
), there are no stable points.
10.2
Binary forms
Let us consider the case

. The elements of the space
are binary
forms of degree . The corresponding hypersurfaces can be viewed as finite subsets of points in taken with multiplicities (or, equivalently, as effective divisors
on ). Let
Let be the maximal torus of SL which consists of diagonal matrices and is

equal to the image of the one-parameter group
Let us first investigate the stability of with respect to . For this

we will follow the last section of the preceding chapter. We have to compute the
weight set wt . We have
148
The weight set is a subset of the set

Let (resp. ) be the smallest (resp. largest) element of this set.
Obviously,
, where is the maximum power of which
divides . Similarly,
, where is the maximum power of
which divides .
By Theorem 9.2, we know that

if and only if
is semi-stable (resp. stable) with respect to
(resp.
(10.2)
This can be interpreted as follows:
is semi-stable (resp. properly stable) with respect to if and only if the
points and are zeros of of multiplicity (resp. ).

From this we easily deduce
Theorem 10.2. Hyp ss (resp. Hyp
with no roots of multiplicity (resp.
s) is equal to the set of hypersurfaces

).
Proof. Suppose
is semi-stable and has a root of multiplicity
. Let take this point to the point . Then has the
point as a root of multiplicity . This shows that is unstable with
respect to . Hence is unstable with respect to , contradicting the assumption.
Conversely, assume has no roots of multiplicity and is unstable. Then
there exists a maximal torus with respect to which is unstable. Let
for some . Then is unstable with respect to . But then it has one of
the points or as a root of multiplicity . Thus has
and as a root of multiplicity .
A similar argument proves the assertion about stability.
Corollary 10.1. Assume is odd. Then
ss Hyp s
Now assume is even and let
Hyp ss Hyp . This means that
has a root of multiplicity but no roots of multiplicity greater than .
Consider the fiber of the projection Hyp ss Hyp ss containing .
Hyp
Since our categorical quotient is good, the fiber contains a unique closed orbit.
10.2. BINARY FORMS
149
belongs to this orbit if and only if its stabilizer is of positive dimension. Assume
belongs to this orbit. Since any group element stabilizing stabilizes its set
of roots, and it is easy to see that any subset of consisting of more than two
points has a finite stabilizer. Thus, must have only two roots. Since one of
these roots is of multiplicity , the other one is also of multiplicity . Since
any two-point sets on are projectively equivalent, this tells us that
ss
s

where is given by the equation
. In particular,
Hyp ss
Hyp s
where the single point represents the orbit of .
The variety Hyp ss is an irreducible normal projective variety
Hyp
Hyp
of dimension
: by construction of the categorical quotient,
Projm Pol Pol
SL
So it can be explicitly computed if we know the algebra of invariant polynomials

on the space of binary forms of degree .
Let us consider some special cases with small .
If
we have Hyp ss . If we have Hyp s and
Hyp ss consists of subsets of two distinct points in . There is only one orbit
of such subsets.
The set Hyp ss consists of three distinct points in . By a projective trans

is
formation they can be reduced to the points . So the variety
again one point. This also agrees with the fact that Pol Pol SL ,
where is the discriminant invariant (see Exercise 2.6).
The set Hyp s consists of subsets of four distinct points in and the set
Hyp ss consists of closed subsets where has at most double roots.
Since Hyp is an open Zariski subset of the projective space (see Exercise
10.1), and the fibers of the projection Hyp s Hyp s are of dimension
3, we obtain that is a normal, hence nonsingular, curve. Since it is obviously
unirational, it must be isomorphic to . The image of the set of semi-stable but

not properly stable points is one point. If we consider the map
ss
Hyp
as a rational function on Hyp s then we can find its explicit expression as a
rational function in the coordinates of a binary form. To do this we
150
have first to find the algebra of invariants Pol Pol
invariant, the catalecticant
. Another invariant is
(see Example 1.4). Its bracket expression is

of degree 2:
SL . We already know one
Its bracket expression is . One can show that any other invariant must be
a polynomial in and . We will prove this in the next chapter. This agrees
with the fact that . The discriminant of a quartic
polynomial
is an
invariant whose bracket expression is equal to

. It
is a polynomial of degree 6 in the coefficients and we have
Thus the rational function
(10.3)
is invariant with respect to SL and defines a regular map from Hyp
This is the geometric quotient map. The map
ss

is the categorical quotient map. Its fiber over
Hyp
s to
is equal to the union

of orbits of binary forms of degree 4 with double roots (up to a nonzero
scalar
factor). The only closed orbit in this fiber is represented

by
.
Consider the special case when . If char
then each orbit contains a representative of such a form. The value of on

is equal to
The
expression in the denominator is the discriminant of the cubic polynomial
. The reader familiar with the theory of elliptic curves will immediately
recognize this function; it is the absolute invariant of the elliptic curve given in
the Weierstrass form
10.2. BINARY FORMS
151
This coincidence is not accidental. The equation above describes an elliptic curve
as a double cover of branched over four points: the infinity point and the three
. In other words they are the zeros of the
roots of the equation
binary form . Two elliptic curves are isomorphic if and only
if the corresponding sets of four points on are in the same orbit with respect to
the action of SL .
Let
. The algebra of invariants
Pol Pol
SL
can be computed explicitly (see [28]). Let us write a general binary quintic in the
form
(we assume that char

). Then is generated by the following invariants:

There is also one basic relation between these invariants which expresses as
a polynomial in invariants , and . We will consider as a
graded algebra whose grading is defined by the natural grading of Pol Pol
with the degree divided by 2. It follows that there is an isomorphism of graded

algebras
where
is graded by setting
and is a weighted homogeneous polynomial. Let

be the subalgebra of
generated by elements of even degree. Then

is generated by homogeneous
elements of even degree . Since can be expressed as a polynomial

in , we see that
is isomorphic to the graded polynomial algebra
. This implies that
Projm
Projm
152
In particular is a rational surface.
Note that the discriminant of a binary quintic can be expressed via the basic
invariants as follows:
This shows
that the locus of orbits of binary quintics with a double root is equal
to and hence is isomorphic to .
Let
. We will use the explicit description of the algebra of invariants
Pol Pol SL due to A. Clebsch ([12]). For a modern treatment see [54].
is generated by invariants , where the subscript denotes the
degree. The only relation between the basic invariants is
for some polynomial . We will consider as a graded algebra whose grading

is defined by the natural grading of Pol Pol . It follows that there is an
isomorphism of graded algebras

ss
where
is graded by setting

and is a weighted homogeneous polynomial. Arguing as in the preceding ex
ample, we see that
Projm
In particular is a rational three-dimensional variety.

Note that the invariant is the discriminant of a binary sextic, so it vanishes
on the locus of binary sextics with a double root. The complement of this locus
in represents reduced divisors of degree 6 in . It is isomorphic to the

moduli space of genus 2 curves. The isomorphism is defined similarly by
assigning to a genus 2 curve the six branch points of its canonical degree 2 map to
. So we obtain that is isomorphic to the open subset of
where the last coordinate is not equal to zero. Since each point in this subset
is represented by a point s in with

, it follows from the
definition of weighted projective space that
10.3. PLANE CUBICS
153
where a generator of the cyclic group
by the formula
acts on
The image of the origin is the unique singular point of . It represents the
isomorphism class of the hyperelliptic curve corresponding to the binary quintic
of order 5.
. It admits an automorphism
Finally observe that the locus of binary sextics with a multiple root and
are both isomorphic to .
10.3
Plane cubics
Let and . Every homogeneous form of degree 3 in three variables (a

ternary cubic) can be written in the form:
Now let us recall the classification of plane cubic curves. First of all it is easy to
list all reducible curves. They are of the following types:
(1) the union of an irreducible conic and a line intersecting it at two distinct
points;
(2) the union of an irreducible conic and its tangent line;
(3) the union of three nonconcurrent lines;
(4) the union of three concurrent lines;
(5) the union of two lines, one of them double;
(6) one triple line.
Since
all nonsingular conics are projectively equivalent to the conic
and the group of projective automorphisms of the conic acts transitively
on the set of tangents to or on the set of lines intersecting transversally, we

obtain that any curve of type (1) or (2) is projectively equivalent to the curve
(1)
154
(2)
respectively. Since the group of projective transformation of acts transitively on

the set of lines with , we obtain that any curve of type (36) is projectively
equivalent to the curve given by the equation
(4)
(3)
,
,
,
(5)
(6)
respectively. Now let us assume that is irreducible. First let us assume that
is nonsingular. Choose a system of coordinates such that the point is an

inflection point and is the equation of the tangent line at this point. It is
known that any plane curve contains at least one inflection point. Then we can
write the equation as

where is a form of degree and is a form of degree 3. Since the line
intersects the curve at one point, we easily see that the coefficient of at
is
equal to zero. Thus in affine coordinates

, the equation
takes the form
(10.4)

Obviously , so after scaling we may assume .
, we may assume that
Assume char . Replacing with
. If char , by a change of variables , we may assume
that . Thus, we obtain the Weierstrass equation of a nonsingular plane cubic:
(10.5)
char
(10.6)

char
char
(10.7)
The condition that the curve is nonsingular is expressed by , where is the
discriminant defined by
if char
if char
if char

,
.
,
(10.8)
10.3. PLANE CUBICS
155
Two curves are isomorphic if and only if their absolute invariants
if char
if char
if char
are equal.
Now suppose is singular. We may choose
Then the equation is of the form
to be the singular point.

we reduce
(10.9)
(10.10)
to one of two forms:

By a linear
transformation of variables
or . Consider the first case. The singular point is a cusp; the
equation is

Replacing
with
, we may assume that . Since the
curve is irreducible we have
; by scaling we may assume that
and
or .
If char , we see that there are two orbits of cuspidal curves, represented
by the equations
and
All nonsingular points of the first curve are inflection points. The second curve
does not have nonsingular inflection points.
If char , then the curve has only one inflection point with
tangent line given by . Now change the coordinates in such
a way that is the unique nonsingular inflection point, the line is
the tangent line at this point and the singular point is . Then, the equation
reduces to the form
Now we consider the case of nodal curves (when the quadratic form
(10.10) is equal to ) so that the equation is
Changing
to
we reduce the equation to the form
in
156
, so by scaling, we reduce the equation to the form

Clearly,
We leave it to the reader to find a projective isomorphism between this curve and
the curve
if char .
Summarizing, we get the following list of equations of irreducible plane curves
(up to projective transformation):
char :
(7) nonsingular cubic
(8) nodal cubic
(9) cuspidal cubic:

char
(8) nodal cubic
(9) cuspidal cubic:
or
char :
where
10.3. PLANE CUBICS
157
(8) nodal cubic
(9) cuspidal cubic
Let be the diagonal maximal torus in SL . It consists of matrices of the form
The
standard torus
SL s
, we have
acts on
diag
via its natural homomorphism

For each monomial
Pol

Thus each monomial belongs to the eigensubspace , where
is
the character of defined by the vector

It is easy to see that is one-dimensional and is spanned by the monomial
. Thus

wt
It is a set of 10 lattice points in
:
3
T1
T0 T1
T0 T1
T1 T2
T0 T1 T2
T1 T2
T0
T0 T2
2
T0 T2
3
T2
Suppose is unstable with respect to . Then the origin lies outside of the
convex hull of wt . It is easy to see that this is possible only if wt consists
158
of lattice points on one edge of the triangle plus one point nearest to the edge but
not the interior point. After permuting the coordinates we may assume that
It is clear that is a singular point of . In affine coordinates

, the equation looks like
From this we see that the singular point is not an ordinary double point.
It follows from the above classification of plane cubic curves that the following
curves are unstable:
(us1) irreducible cuspidal curve (two orbits if char
);
(us2) the union of an irreducible conic and its tangent line;

(us3) the union of three concurrent lines;
(us4) the union of two lines, one of them double;
(us5) one triple line.
By looking at the equations of the remaining curves and drawing their weight
sets we see that any nonsingular cubic is stable and any singular curve not from
the above list is semi-stable. Note that it is enough to check the numerical criterion
only for one fixed torus. In fact, the property of being nonsingular or have at most
ordinary double points is independent of the chosen coordinates. Thus we have
the following list of semi-stable points:
(ss1) nonsingular cubic (stable point);
(ss2) irreducible nodal curve;
(ss3) the union of an irreducible conic and a line intersecting it at two distinct
points;
(ss4) the union of three non-concurrent lines.
10.3. PLANE CUBICS
159
Consider the quotient map
ss
Hyp
Hyp
ss SL
The dimension of its fibres containing stable curves is equal to 8 (

SL ).
Note that in the process of the previous analysis, we found that curves of types
(ss1), (ss2) and (ss3), each form a single orbit represented by the curves
respectively. Moreover the curves of types (ss2) and (ss3) have stabilizer of pos
itive dimension. In fact the torus , where , stabilizes the
second curve, and the maximal diagonal torus stabilizes the third curve. This
shows that the orbits of curves of types (ss2) and (ss3) are of dimension .
Thus they lie in the closure of some orbit of dimension 8. It cannot be a stable
orbit, hence the only possible case is that it is the orbit of curves of type (ss1).
Hence this orbit is nether closed nor stable.
Hyp ss SL . It is a
Since Hyp is of dimension 9, we obtain
normal projective unirational curve, hence we find that
ss SL
Hyp
Since there is only one closed semi-stable but not stable orbit, namely the set of
three non-concurrent lines, we obtain
s SL
It is easy to see that the orbit of the curve
Hyp
is of dimension . In the
same fibre we find two other orbits: of nodal irreducible cubics (of dimension 8)
and of curves of type (ss2) (of dimension 7). The second orbit lies in the closure
of the first one, and the closed orbit lies in the closure of the second one.
If char , we have five unstable orbits: irreducible cuspidal cubics (of
dimension 8), curves of type (us2) (of dimension 6), of type (us3) (of dimension
5), of type (us4) (of dimension 4), and of type (us5) (of dimension 2). It is easy to
see that the orbit of type (us ) lies in the closure of the orbit of type (us(i-1)).
If char we have two unstable orbits of type (us1), and four other
unstable orbits lying in the closure of these two orbits.
One can give the explicit formula for the quotient map similar to (10.3). In
characteristic , it can be given by the following rational function in the
coefficients (see [98], p. 189192):
160
where
ss
Here we use the following dictionary between our notation of coefficients and
Salmons:
In fact the algebra Pol Pol SL is freely generated by and . If one evalu
ates
and on the curve given in the Weierstrass form from above, we obtain
10.4. CUBIC SURFACES
161
In this special case the value of the function is equal to
This is the absolute invariant of the elliptic curve. Note that we arrived at the same
function by studying the orbits of binary quartics.
10.4
Cubic surfaces
Consider the case . It corresponds to cubic surfaces in . The al

gebra of invariants Pol Pol
was computed by G. Salmon and A. Clebsch
([97]). It is generated by invariants , where the subscript
indicates the degree. The square of the last invariant is expressed as a polynomial

in the first five invariants. In analogy with the case , we find that
In particular, is a rational variety. The invariant corresponding to the

variable with weight is the discriminant. Thus we obtain the following isomorphism for the moduli space cubic of nonsingular cubic surfaces:
cubic
where a generator of the cyclic group
acts on by the formula
The unique singular point of

(see [79]):
cubic
corresponds to the following cubic surface
where
. The automorphism group of this surface is isomorphic to the
dihedral group of order 8.

The subvariety of defined by the equation is isomorphic to
that the latter is isomorphic to ; this is not an accident.

. Recall
the Veronese
If a point of represents six distinct points in , we consider
map to identify them with six points on a nonsingular conic in . Then the linear
162
system of cubics through these points defines a rational map from to . Its
image is a singular cubic representing a point of . The singular point of
this cubic is the image of the conic. Thus we see that the moduli space is
isomorphic to an open subset of the hypersurface in .
The following are the other values of for which the analysis of stability
has been worked out:
s 73 103
104 s 2 q

Bibliographical Notes
The examples of explicit computation of the the quotient spaces given in
this lecture have been known since the nineteenth century (see [30], [38], [96]).
The other known cases are s (see [36], [35] and also [107],
[20]). A modern proof of the completeness of the Clebsch-Salmon list of fundamental invariants of cubic surfaces was given by Beklemishev ([4]). These are
probably the only examples where one can compute the spaces explicitly.
In fact, one can show that the number of generators of the algebra of invariants
on the space of homogeneous polynomials of degree grows very rapidly with
(see [88]).
It is conjectured that all the spaces are rational varieties. In the case of
binary forms, this was proven by F. Bogomolov and P. Katsylo ([5]). The spaces
are known to be rational only in some cases (see [57], [58], [106] and also
a survey of results on rationality in [21]).
Exercises
10.1 Show that Hyp
as subsets of .
. Desribe the sets of semi-stable and stable points
, be four distinct roots of a binary quartic . Let

10.2 Let s
denote the determinant
of the matrix with columns s . The expres
sion is called the cross-ratio of the four points. Prove that
two binary quartics define the same orbit in Hyp if and only if the correspond
ing cross-ratios coincide after we make some permutations of the roots.
EXERCISES
163
10.3 Let be the complement of the quartic in , where is the discriminant of a binary cubic form. Show that is isomorphic to a homogeneous space
SLs , where is a subgroup of order 12.
10.4 Show that there are exactly two orbits in Hyp s with non-trivial stabilizer.
Show that the closures

of these orbits in Hyp are given by the equations
and , where are the polynomials of degree 2 and 3 defined in section

10.2.
10.5 Show that Hyp us is isomorphic to a surface of degree 6 in . Its singular
set is isomorphic to a Veronese curve of degree 4.
10.6 Construct a rational map from to whose image is equal to the
locus of zeroes of the discriminant invariant. Describe the points of indeterminacy

of this map and its inverse.
10.7 Find the orbits of the binary quintics which correspond to singular points of
.
10.8 Find the group of projective automorphisms of a nonsingular cubic curve

(you may assume that char ).
10.9 Find all projective automorphisms of an irreducible cuspidal cubic.
10.10 Perform the analysis of stability in the case and compare the
result with the answer in [73].
10.11 Prove that nonsingular quadrics are semi-stable in all characteristics.
10.12 Show that a plane curve of degree is unstable if it has a singular point of
multiplicity .
Chapter 11
Configurations of linear subspaces
11.1
Stable configurations
In the last two chapters, for typographical reasons, we denote the Grassmannian
Gr of -dimensional linear projective subspaces in by Gr . The

group SL acts naturally on Gr via its linear representation in . In
this lecture we investigate the stability for the diagonal action of on the variety
sr
where
this action.
Lemma 11.1.
Gr
First we have to describe the possible linearizations of
Pic Gr

A generator of this group is the line bundle Gr corresponding to a hyper
plane section in the Plucker embedding of Gr in
Proof. We will represent a point

Gr as a matrix
. . .
..
.. . . . ..
Pic Gr
165
166
CHAPTER 11. CONFIGURATIONS OF LINEAR SUBSPACES
Its rows form a basis of . The Plucker coordinates of

are the maximal
. The open subset of

minors of this matrix formed by the columns
is the affine space . The restriction of any

Gr with
Pic Gr to this
open subset is trivial, so is isomorphic to the line bundle
associated to a divisor equal to a multiple of a hyperplane section. Since any line

bundle admits a unique linearization with respect to SL , the assertion follows.
We use the notation to denote the projective coordinates in (we

order them lexicographically). The value of this coordinate at any

Gr
is equal to the Plucker coordinate of . Since Gr is not contained in a

linear subspace of , the restriction map
Gr
Gr
is injective. One can also show that it is surjective.

For any vector we define a line bundle on sr
pr
Gr
where pr sr Gr
is the -th projection. It follows from Lemma 11.1 that
any line bundle on is isomorphic to for some (use [46], p. 292). Since each
pr is an SL -equivariant morphism, admits a canonical SL -linearization.
Thus

PicSL sr
Also is ample if and only if all are positive. In fact, if some tensor power
of defines a closed embedding r , then the restriction of to any
subvariety isomorphic to a factor is an ample

line bundle. But it is obvious that
this restriction is isomorphic to Gr . The latter is ample if and only if

. Conversely, any with positive (meaning that all s are positive)
is very ample. It defines a projective embedding of r which is equal to the
composition
sr
where the first map is the product of the Plucker embeddings, the second map is
the product of the Veronese embeddings, and the last map is the Segre map.
11.1. STABLE CONFIGURATIONS
167
Now we are ready to describe semi-stable and stable configurations of linear
projective subspaces
sr

Let
. Then
Theorem 11.1.
sr s ) if and only if for any proper linear subspace
r ss
of
(resp.
(resp. the strict inequality holds).
Proof. Let be the maximal diagonal torus

SL
. Each one-parameter sub in
r
group of is defined by diag r
, where t t . By
permuting coordinates we may assume that
(11.1)
Suppose
is semi-stable. Let be the linear
space spanned by the unit vectors and let be the corresponding pro
jective subspace. For any
Gr and any integer , there is a

unique integer for which

and observe that
To see this we list the numbers

since each is a hyperplane in and .
Then we see that each occurs among these numbers and we define
first with .
With this notation we can represent

by a matrix of the form
where for all
matrix that
to be the
(11.2)
from viewing the maximal minors of this

. ifIt is clear
for any value of and .
Now we notice that the projective coordinates of
in the
embedding defined by the line bundle
are equal to the product of
monomials
168
of degree
have
in the Plucker coordinates of
Since for each
as in (11.1) we
sr
it is easy to see that
as in the above. Using

Here are defined for each

if , we can rewrite the previous
that
sum as follows:
Since we want this number to be non-positive (resp. negative) for all , we can
take the special one-parameter subgroup given by
t

qt
It is easy to see that any satisfying (11.1) is a positive linear combination of such
one-parameter subgroups. Plugging in these values of t , we find
resp.
(11.3)
Since any -dimensional linear subspace of is projectively equivalent to , we
obtain the necessary condition for semi-stability or stability stated in the theorem.
is not semi-stable,
It is also sufficient. In fact, if it is satisfied but

. By choosing approwe can find some SL such that
priate coordinates, we may assume that and satisfies

(11.1).
Then we

for
write as a positive linear combination of s to obtain that
some . Then the above computations show that (11.2) does not hold, contradicting our assumption.
11.1. STABLE CONFIGURATIONS
169
Corollary 11.1. Assume that the numbers
Then
r ss
Corollary 11.2.
for any proper subspace
of
of
are coprime.
are equal (in this
and
s s
for any proper subspace
ss
ss s
ss SL
and
Let us rewrite Theorem 11.1 in the case where all

case the linearization is called democratic). We set
ss

. Also,
Let us consider some examples.
. Taking
Example 11.1. Let
to be a point, we get that
can
if is semibe equal to at most

points among
stable with respect to . This is similar to the stability criterion for a binary
form of degree . This is not surprising, since Hyp and
is equal to the inverse image of under the projection Hyp .

Note that if we change to , where
, we get that
is semi-stable.
Example 11.2. Let us take
. Then
is semistable no point is repeated more than

times and no more than
points are on a line
Semi-stability coincides with stability when does not divide .
170
For instance, let us take

. Then stable sextuples of points are all distinct and have at most three collinear. On the other hand, semi-stable but not
stable sextuples have either two coinciding points or four collinear points among
them. It is easy to see that minimal closed orbits of semi-stable but not stable

points are represented by sextuples , where for some
with the remaining four points on a line. Among them there are special orbits
O corresponding to the sextuples with , where
vari . So ss is a four-dimensional
ety, and ss s is isomorphic to the union of 15 curves each
isomorphic to ss SL . Each curve contains three points rep
resented by the orbits O . Each point lies on three curves

and .
Let us consider the subset of s of sextuples such that there
exists an irreducible conic containing the points . Since all irreducible
conics are projectively equivalent, the orbit space s SL is isomorphic to the
orbit space s SL of sextuples of distinct points on . However, as we
will see later, its closure in ss SL is not isomorphic to
ss SL .
. Then we are dealing with

Example 11.3. Let us take
sequences of lines in . Let us apply the criterion of semi-stabilty,
taking
to be first a point, then a line, and finally a plane. In the first case we
obtain
that is, no more than lines intersect at one point.
Taking
to be a line, we obtain
in particular, no more than lines coincide and no more than
intersect a line which is repeated times.
Finally, taking
to be a plane, we get

that is, no more than lines are coplanar.
lines
For example, there are no stable points if . This follows from the fact
that for any four lines in there is a line intersecting all of them. There are no
semi-stable points for

. If , a pair of lines is semi-stable if and only
11.2. POINTS IN
171
if they dont intersect. It is easy to see that by a projective transformation a pair

of skew lines is reduced to the two lines given by the equations
and . Thus we have one orbit. Similarly, if we get one

, and
semi-stable orbit represented by the lines
. If , the
formula for the dimension of the quotient
ss , where is the stabilizer of a
space gives us that
generic point in ss . In our case
since there are no stable orbits. It is
easy to see that

(use that there is a unique quadric through the first
three lines, and the fourth line is determined by two points of intersection with the
quadric; the subgroup of the automorphisms of the quadric which fix two points
and three lines in one ruling is isomorphic to ). We will show later, by explicit
computation of invariants, that
SL
ss
(11.4)
Let us give a geometric reason why this can be true. For any four skew lines in
general position, there exist two lines which intersect them all (they are called
transversals). This is a classical fact which can be proven as follows. Consider
the unique quadric through the first three lines . They belong to one
ruling of lines on . The fourth line intersects at two points t t . The two
transversals are the lines from the other ruling of which pass through t t .
If the fourth line happens to be tangent to , so that t t , we get only one
transversal. Now let be the two transversals. Then we have two ordered sets
of four points on :
This defines a rational map
The proof that this map extends to an isomorphism consists of the study of how
this construction can be extended to degenerate configurations.
11.2
Points in
Let us consider configurations of
points in
. We have
172
. Then
ss (resp. s )
for every proper linear subspace

of

Theorem 11.2. Let
if and only if
(resp. the strict inequality holds).

In particular, if all
, the last condition can be rewritten in the form
(resp.
Corollary 11.3.
ss
, the left-hand side is empty and the assertion is obviously true in

Proof. If
this case. We assume that . Let
gen each subset of

points spans
is an
This is an open nonempty subset of . We know that ss
open subset. So if it is not empty it has nonempty intersection with gen .
If we take a set of points in the intersection, we obtain, since
for each . Conversely,
no two points coincide,
if this condition is satisfied then each point gen is

. In fact, each subspace of dimension contains
semi-stable with respect to
points . Hence
at most

This proves the assertion about the semi-stability. We prove the second assertion
similarly.
11.2. POINTS IN
173
Let
This is called an -dimensional hypersimplex of type . One can restate the
preceding corollary in the following form. Consider the cone over

in
We have the injective map

PicSL
with a subset of . We have

PicSL
PicSL ss
from the left-hand side
In fact, if the first
coordinates of a point
which allows us to identify Pic
SL
are all positive, this follows immediately from Corollary 11.3. Suppose some
of the first coordinates of are equal to zero, say the first coordinates. Then
pr , where pr is the projection to the last
factors, and . By applying Corollary 11.3 to , we obtain
. It is easy to see that

that ss q
ss
pr
ss
and we have a commutative diagram
ss
ss SL
pr
ss
pr
ss SL
where the vertical arrows are quotient maps and the map pr is an isomorphism.
Note that the relative boundary of the convex cone

consists of points

with one of the first coordinates equal to zero, and of points
for some . The intersection of the latter

satisfying
174
part of the boundary with PicSL
consists of line bundles such that
for some . This shows that all points from
gen are
semi-stable but not stable (with respect to ). Since the set of stable points must
be open, it must be empty.
Observe that ss s if and only if there exists a
subspace
of dimension such that
belongs to the hyperplane
This is equivalent to the condition that
where is a nonempty subset of . Let be a connected component of
(called a chamber). One can show that any two line bundles

from the same chamber have the same set of semi-stable points. Suppose
belongs to some and does not lie on other hyperplanes s . Then there

are two chambers with common boundary . We have a commutative

diagram
SL
ss SL
SL
Here s means that we define the stability with respect to any from
. The corner maps are birational morphisms, and the upper arrow is a birational
map (a flip). We refer the reader to [23] for more general and precise results on
this subject.
The spaces
ss SL
can be described explicitly in a few cases. It follows from the construction of the
quotient that
Projm
SL
Projm
Pol SL
11.2. POINTS IN
175
Pol SL by
where . Let us denote the graded algebra
The First Fundamental Theorem tells us how to compute generators of the

graded algebra . We have
Pol Mat
SL
(11.5)
Thus the space is generated by standard tableau functions

of size ,
degree with .
Remark 11.1. Note that the symmetric group acts naturally on , via per
muting the factors. It acts on the graded algebra via its action on the columns
. The quotient is the moduli space of

of matrices of size
(unordered) sets of -points in . In the special case , an unordered set of
-points is the set of zeros of a binary form of degree . Recall that, by the First
Fundamental Theorem, we have an isomorphism
Pol SL Pol Mat SL
Pol
In view of (11.5) we obtain an isomorphism
Pol SL
Pol
Now, if we use Hermite Reciprocity (Theorem 5.6), we get an isomorphism
SL
Pol Pol
It can be shown (see Remark 5.2) that the isomorphisms

phism of graded algebras
define an isomor-
SL
Pol Pol
(11.6)
The projective spectrum of the left-hand side is the variety . The projec
tive spectrum of the right-hand side is the variety Hyp SL Thus

. Then the degree 1 piece of
Example 11.4. Let us start with the case
is spanned by the two functions
and
. The value of the ratio
on the set defined by the coordinate matrix
176
is equal to
This is called the cross-ratio of four ordered points. Two distinct ordered quadruples of points in are projectively equivalent if and only if they have the same
, assuming
cross-ratio. If we choose coordinates in the form s
that none of the points is the infinity point, we obtain

s s

s
s
we get
If
Note that the cross-ratio of four distinct points never takes the values .
The quadruples go to if or . The only closed orbit
in the fiber over consists of configurations with . Similarly,
one describes the fibers

over
and . It is easy to see that the graded algebra
is equal
to and hence is isomorphic to the polynomial algebra
(prove this by following the next example). The permutation group acts
on this algebra as follows:
This easily implies that

Pol Pol
where
SL
Using (11.6) we can identify (up to a constant factor) these invariants with the
invariants and from section 10.2 of Chapter 10.
. The computations here are more involved

Example 11.5. Let
than in the case

which we will discuss in the next example. Here we
to a Del Pezzo surface of

only sketch a proof that the space is isomorphic
degree 5 isomorphic to the blow-up of with center at four points
11.2. POINTS IN
177
no three of which are on a line. The linear system of conics defines a morphism
. Its fibers are conics through the four points . There are three
singular fibers corresponding to three reducible conics. There are four sections of
corresponding to the exceptional curves blown up from the points . Let
us construct a map . If lies on a nonsingular fiber , we
consider the fiber as and assign to the orbit of the five points
in . If lies on a singular fiber, say on the proper transform

of the line passing through the points we assign to the orbit of
, where is the inverse image of the point . If we
assign to it the unique orbit of . Note that under this assignment the
fibration map corresponds to the natural map defined by the projection
. The three points in over which the fiber
is singular are the three special orbits of s , s and . The

section corresponds to the set of orbits of , where .
Example 11.6. Let
is given by a table
. A standard tableau of degree and size
(11.7)
where we use the notation from section 2.4. We have
Set
These numbers satisfy the following inequalities:
The last two inequalities say that each row consists of two different numbers, so
that
178
Setting
determined by a vector

, we obtain that our tableau is completely
satisfying

these inequalities are equivalent to
This gives
which gives
When
solutions. When we have

solutions. Summing up, we get
is equal to
Thus the Hilbert function of the graded ring
This suggests that is isomorphic to a cubic hypersurface in
First of all we have the following generators of :
. This is true.
For every s , the product is a standard tableau function from

. Applying the straightening algorithm, we find
So the standard monomials
can be expressed as polynomials of degree 2 in the . Counting the

number of
standard tableau functions of size

, we find that
In fact, we

11.2. POINTS IN
179
have for any . If we take a tableau function

to tableau (11.7) with , we can write it as
whenever
whenever
, and similarly
if
if
if
if
if
corresponding
,
,
. It is easy to verify that
which gives us the cubic relation
Let
There is a surjective homomorphism of the graded algebras
and comparing the Hilbert functions we see that it is bijective. Thus
Projm is isomorphic to the cubic hypersurface
If we change the variables,
we obtain that
can be given by the equations
180
in which manifest the -symmetry. The cubic hypersurface defined by these

equations is called the Segre cubic primal. It contains 10 nodes (the maximum
possible number for a cubic hypersurface in ) and 15 planes. The nodes correspond to the minimal closed orbits of semi-stable but not stable points. The
singular points can be indexed by the subsets of . For example,
. The planes correspond to the orbits of sextuples with
two coinciding points. They have equations of the form
, where . Each plane contains four

singular points. Each point is contained in 6 planes. The blow-up of the plane at
the four points is naturally isomorphic to (see Exercise 11.7).

. Again we take and try to
Example 11.7. Let and
compute the graded algebra explicitly. We skip the computations ([25], p.17)
and give only the results. First we compute the Hilbert function of the graded
algebra :
This suggests that is generated by five elements of degree 1 and one element
of degree 2 with a relation of degree 4. We have the following.
Generators:
degree 1
degree 2
Relation:
This shows that

projective space
is isomorphic to a hypersurface of degree 4 in the weighted
given by the equation
11.3. LINES IN
If char
181
this can be transformed into the equation
The equation is again symmetric with respect to a linear representation of in

the variables (but not with respect to the standard permutation repre
sentation in ). The quartic hypersurface in given by the equation
is called the Segre quartic primal (or Igusa quartic). It corresponds to the relation
If we fix the points and vary we see that this is of degree 2 in the
. Thus it
coordinates of and vanishes when for some
describes the conic through the points and expresses the condition that
the six points are on a conic. Using the equation , we can exhibit as a
double cover of branched along the Segre quartic hypersurface. In other words,
there is an involution on whose fixed points are the sextuples lying on a conic.
This is the self-association involution. We have a remarkable isomorphism, the
association isomorphism:
It is defined by the isomorphism of the graded algebras

defined

on tableau functions by replacing each determinant

with the deter
minant
. In
, where

the case
, we get an involutive automorphism of the
algebra
which defines the self-association involution of the variety . We refer to
[25] and [27] for the details and for some geometric meanings of the association
isomorphism.
Lines in
11.3
Let us give an algebraic proof of the existence of the isomorphism (11.4). Recall
that Gr is isomorphic to a nonsingular quadric in . Its automorphism group
182
is the complex projective orthogonal group PO O . The natural ac

tion of SL on Gr defines an injective homomorphism from PSL to PO .
Counting the dimensions we see that the image is the connected

component of

the identity of the group PO . It is the subgroup PO whose elements are
represented by orthogonal
matrices with determinant 1. Now the analysis of sta
bility for lines in shows that a semi-stable configuration of lines, considered
as an ordered set of points in , is semi-stable with respect to the

action of SL
ss
in . Thus
is a closed subset of the quotient
O
. The latter
can be computed using the First and the Second Fundamental Theorem of invariant
theory for the orthogonal group. The symmetric bilinear form on the space
defined by the Grassmannian quadric is the wedge product. If is a

vector space equipped with a nondegenerate symmetric bilinear form , then
the algebra of polynomial invariants
of O in the space is generated by the
(see Exercise 2.9, or [121]).

functions defined by
,
This algebra is equal to the algebra of invariants for O unless
when additional invariants are the basic invariants for SL , i.e., the bracket
, there are no relations between the basic invariants.
functions. For
Now
ss O
Pol O
As we saw in Chapter 2, elements of Pol are polynomial functions on
which are homogeneous

of degree in each factor.
Thus the space of invariants
Pol O is spanned by monomials in such that each
appears among exactly times. In our

index
case we have 10 basic invariants . For

we have three mono
mials , where
. For
, we have products of
these three monomials

plus
additionally
the
monomials
which
contain one of the
monomials as its factor. Now observe that the restriction of the function to
the subset of points in lying on the quadric is obviously zero.
Thus, the restriction of the algebra
is freely generated by
to
. Its projective spectrum is
183
Note that a similar computation can be made in the case
(see [117]). In the case , the algebra
Gr s O
and

is generated by the 15 functions , where
, and the determinant function
. The square of is
the determinant of the Gram matrix and hence can be expressed as a
polynomial in the . The subalgebra generated by the functions is

isomorphic to the projective coordinate algebra of a certain nine-dimensional toric
variety (see the next chapter), so that is isomorphic to a double cover of
branched along a hypersurface defined by the equation . The locus of sextuples of lines defined by this hypersurface coincides with the locus of self-polar
sextuples, i.e.,
the sextuples for which there exists a nondegenerate
quadric in such that the set of the polar lines is projectively equiv
alent to . Note the remarkable analogy with the structure of the variety
, where the analog of the polarity involution is the association involution.
The stability criterion for configurations of linear spaces (with respect to the
democratic linearization) was first given by Mumford ([73], Chapter 3). He also
proved that the quotient map for stable configurations of points in is a principal fibration of the group SL . The generalization of the criterion to the case of
arbitrary linearization is straighforward. The cross-ratio invariant is as classical as

can be. Examples 11.6 and 11.7 are taken from [25]. They go back to Coble [13]
who found a beautiful relationship between the moduli spaces of points in and
classical geometry. The book [25] gives a modern exposition of some of the results of Coble. The invariants of lines in are discussed in the book of Sturmfels
([113]). The algebra of SL -invariants on the tensor product of the projective
was studied by
coordinate algebras of four Grassmannians Gr
R. Howe and R. Huang ([51], [50]). They show that this ring is isomorphic to a
polynomial algebra. In the case when

this was first
proved by H. W. Turnbull ([116]). Note that the GIT quotient sr SL
considered in this chapter is isomorphic to the projective spectrum of a subalgebra of the algebra of invariants in the tensor product of the projective coordinate
184
algebras of the Grassmannians; so one needs additional work

to compute the quo
tients. One can also describe all orbits of four lines in (see [22]). The moduli
spaces of five and six lines in and their relationship to the classical algebraic
geometry are discussed in the Ph. D. thesis of D. Vazzana ([117], [118]).
The rationality of the configuration spaces of points is obvious. It is not
known whether
the spaces r ss SL are rational in general. This is known for
lines in ([122]) and, more generally, in the case when
(see [100]).
Exercises
11.1 Prove that the orbit of in is closed but not
stable if and only if there exists a partition of into subsets
such that for any one can find a proper subspace of such that
is the quotient ss isomorphic to
11.2 For what
11.3 Draw a picture of the hypersimplex
cone
.
and describe the chambers of the
11.4 Consider the action of the permutation group on

and show that the
kernel of this action is isomorphic to the group

. Find the orbits whose
stabilizers are of order strictly larger than . Compute the corresponding crossratio.

11.5 Prove that the algebra can be generated by six elements of degree 5 satis
fying five linearly independent quadric relations.
11.6 Show that each projection
(i) Find the points of indeterminacy of .
(ii) Show that is a regular map if
(iii) Construct
rational sections
defines a rational map
of .
11.7 Find the equation (in terms of functions ) of the closure of the locus of
quadruples of lines in which have only one transversal line.
11.8 Prove that is isomorphic to a categorical quotient of the Grassmannian
Gr
with respect to the action of the torus via its standard action on .
EXERCISES
185
Gr which admit
11.9 Prove that the closure of the locus of
a common transversal
line is of codimension 1. Find its equation in terms of
functions .
11.10 Show that Gr is a homogeneous space isomorphic to , where

SL and is its parabolic subgroup of matrices with entries for
.
11.11 Consider the action of SL on via its linear representation in equal
to the direct sum of the two standard two-dimensional representations of SL .
Find stable and semi-stable points of the diagonal action of SL on
with respect to the line bundle . Using the Fundamental Theorem of Invariant
Theory show that ss SL .
11.12 Find stable and semi-stable points in Gr with

respect to the
group SL and linearization

(three lines and two points in ).
11.13 Prove that
(i) the Segre cubic primal is isomorphic to the image of under the rational
map to given by the linear system of quadrics through five points in
general position;
(ii) the nodes of are the images of the lines s joining two points ,
(iii) the planes of are the images of the planes through three points
,
(iv) the blowing up at the points is a resolution of singularities
of with inverse image of each node isomorphic to .

11.14 Let be the Segre quartic primal in . We use the notation from the
preceding exercise. Prove that
(i) is isomorphic to the image of under the rational map
given by the linear system of quartics which pass through the points
with multiplicity 2 and contain the 10 lines s ,

(ii) contains 15 double lines, each line is intersected by three other dou
ble lines (find the meaning of the double
lines and the corresponding points of
intersection in terms of the quotient ss SL ),

(iii) the double lines are the images of the planes under the rational map
,
(iv) the blowing up at the points followed by the blowing up of
the proper transforms of the lines s is a resolution of singularities of ,
(v) is isomorphic to the dual hypersurface of the Segre cubic primal .
action on Gr . Match the minimal

11.15 Describe the orbits of SL in its diagonal
orbits of semi-stable points with points in .
Chapter 12
Toric varieties
12.1
Actions of a torus on an affine space
In this chapter we consider an interesting class of algebraic varieties which arise

as categorical quotients of some open subsets of affine space. These varieties are
generalizations of the projective spaces and admit a very explicit description in
terms of some combinatorial data of convex geometry. In algebraic geometry they
are often used as natural ambient spaces for embeddings of algebraic varieties and
for compactifying moduli spaces. In combinatorics of convex polyhedra they have
served as a powerful tool for proving some of the fundamental conjectures in the
subject.
Let act linearly on by the formula
where

As always we will identify the group with so that we consider the vectors
is trivial and , we have a natural
as characters of . Since Pic
isomorphism (see Chapter 5)
Pic
Let us fix and denote by the corresponding linearized
line bundle. It is the trivial line bundle with the linearization defined by
the formula
187
CHAPTER 12. TORIC VARIETIES
188
We identify its sections with polynomials . A polynomial
defines an invariant section of some nonnegative tensor power if

Here are independent variables. It is clear that belongs to
if and only if is equal to a linear combination of monomials
such that
, or, equivalently,

Let
be the set of nonnegative integral solutions of the system
(12.1)
where the matrix of coefficients is obtained from by adding to it one more

column formed by the vector .
The set of real nonnegative solutions of a linear system of equations forms a
convex polyhedral cone. By definition, this is a subset of given by a system of
linear inequalities
(12.2)

Obviously
any linear equation can be considered as a pair of inequalities
. A convex
polyhedral cone is called a rational convex

polyhedral cone if the vectors can be chosen from (or equivalently from
one can define the dual cone:

). For every polyhedral cone
It is equal to the convex hull of the rays
. It can be shown that
the dual of a rational convex polyhedral cone is a rational convex polyhedral cone.
We have
This shows that any rational polyhedral cone can be defined as a convex hull of a
finite set of positive rays spanned by vectors in .
So we see that the set of vectors satisfying
the system of linear
equations (12.1) is equal to a set of the form for some rational convex
polyhedral cone in . Now we use
12.1. ACTION OF A TORUS
189
Lemma 12.1. (P. Gordan) Let be a rational convex polyhedral cone in

Then is a finitely generated submonoid of .
Proof. Let be spanned by some vectors
. The set

is compact and hence its intersection with is finite. Let be this
intersection. This obviously includes the vectors . We claim that this set generates the monoid
form
. In fact we can write each in the
, where is a nonnegative integer and . Thus
is the sum of some vector and a positive linear

combination of vectors . This proves the assertion.
For any commutative monoid we denote by
its monoid algebra. This
is the free abelian group generated by elements of with the multiplication law
given on the generators by the monoid multiplication. If
we can identify
with the algebra of Laurent polynomials by assigning
to each
the monomial . If is a submonoid
of
we identify
with the subalgebra of
which is generated by
monomials .
Now we can easily construct a natural isomorphism of graded algebras
(12.3)
where is the monoid of nonnegative vectors which satisfy (12.1) for some
, and
is the linear
with .
span of the set of monomials
By Gordans Lemma, is a finitely generated graded algebra. Its homogeneous
part of degree
is .

Let be the ideal
. It can be generated by monomials and

. For each
we choose a minimal set of monomial generators
let . For
each subset of let
and
the open sets
Obviously,
coincide. By definition of semi-stability
ss
190
For any
let
where
(12.4)
(12.5)
We know that the categorical quotient is obtained by gluing together the affine
algebraic varieties with . We will now describe these rings and
their gluing in terms of certain combinatorial structures.
12.2 Fans
Let be the map given by the matrix , then
is a free abelian group of rank rank . Let
((12.5)) is its kernel. It
(12.6)
be the map given by the restriction of linear functions to . Let be
the dual basis of the standard basis of , and let be the
images of these vectors in . For each let be the convex cone in the linear
space
.
be the matrix of size whose rows are
of . If we choose to identify with by means
, then
spanned by the vectors

More explicitly, let
formed by a basis
of the dual basis
This shows that is spanned in
by the columns of with
Lemma 12.2. Let be as in (12.4). Then
12.2. FANS
191
Proof. Obviously is isomorphic to
For each
, where
for some
On the other hand
Lemma 12.3. Let be the set of convex cones

is a face of both and .
. For any
Proof. Let . We want to show that is a common face
of and . Recall that a face of a convex set is the intersection

of with a
hyperplane such that lies in one of the two halfspaces defined by the hyperplane.
We know that is equal to the localization , where

and for . Considering as a linear function on
we have
for
This shows that is identically zero on . On the other hand, it follows from
. This proves the assertion.

Lemma 12.2 that is nonnegative on and on
Definition. A finite collection of rational convex polyhedral cones

in
such that
is a common face of and is called a fan.
In a coordinate-free approach one replaces the space by any real linear

space of finite dimension, then chooses a lattice in , i.e., a finitely generated
abelian subgroup of the additive group of with , and considers rational convex polyhedral cones, i.e., cones spanned by a finite subset of . Then
an -fan is a finite collection of -rational polyhedral cones in satisfying
the property from the above definition. A version of this definition includes in the
fan all faces of all cones .
Let be the dual lattice in the dual space . By Gordans Lemma,
for each the algebra

is finitely generated. Let
192
Specm be the affine variety with isomorphic to . Since

for
is a face in both cones, we obtain that is a

any
localization of each algebra and . This shows that Specm
is isomorphic to an open subset of and . This allows us to glue together
the varieties to obtain a separated (abstract) algebraic variety. It is denoted
by and is called the toric variety associated to the fan . It is not always a
quasi-projective algebraic variety.
isomorphic to .
By definition has a cover by open affine
subsets
Since each algebra

is a subalgebra of
we obtain a
morphism . It is easy to see that this morphism is -equivariant
if one considers the action of on itself by left translations and on by means of
the -grading of each algebra . If no cone contains a linear subspace,
the morphism is an isomorphism onto an open orbit. In general,
always contains an open orbit isomorphic to a factor group of . All toric varieties
are normal and, of course, rational.
Keeping our old notations we obtain

Theorem 12.1. Let be the transpose of the inclusion map
and let be its image. Let be the -fan formed by the cones
Then
ss
Recall that a cone in a linear space is called simplicial if it is spanned by a
part of a basis of . A fan is called simplicial if each is simplicial. The
geometric significance of this property is given by the following result, the proof
of which can be found in [32].
Lemma 12.4. A fan is simplicial if and only if each affine open subset
is isomorphic to the product of a torus and the quotient of an affine space by a
finite abelian group.
In our situation, we have

Proposition 12.1. Let be the toric variety ss . Assume the kernel
of the action homomorphism Aut is finite. Then is simplicial if and

only if
ss s
12.2. FANS
193
Proof. Assume some is not simplicial. We have to show that there exists a
semi-stable but not stable point. Let be the spanning vectors of Since
is not simplicial, for some integers not all of which are zero.
This implies that belongs to the annihilator of in . If we
identify with , then is isomorphic to the submodule spanned by the

rows of the matrix . Thus we can write
for some . This implies that
Let us consider the one-parameter subgroup

vector . It is defined by
for .
corresponding to the
For any and
we have

(12.7)

Take a point , where
if and otherwise. Since
, we see that ss . On the other hand, and
hence is not stable.

Conversely, assume that there exists a semi-stable but not stable point. Arguing as above, we find a one-parameter subgroup such that for all
equal
where . Then has not all coordinates

to zero for and
for all . This gives , hence is
not simplicial.
Since every line bundle on an affine variety is ample, we obtain that the toric
varieties ss are always quasi-projective. Let us find out when
they are projective.

Definition. A fan in a linear space
is called complete if
For the proof of the following basic result we refer to [32].

Lemma 12.5. A fan is complete if and only if the toric variety is complete.
194
Theorem 12.2. Assume that is not the trivial linearized bundle (i.e., )
. The toric variety

ss is projective if and only if
and ss

is not contained in the convex hull of the character vectors
ss that it is equal to the

Proof. It follows from the construction

of
projective spectrum Projm

,
where
is
the
monoid
of solutions
of the
system

(12.1). We have and the inclusion defines a
surjective map Projm Specm . It is easy to see that Projm
is projective if and only if this map is constant, i.e., . The latter is

equivalent to , i.e., the only nonnegative rational combination of
the columns of which is equal to must be the zero combination. If this is not
true, then for some nonnegative integers , and dividing
both sides by we see that is in the convex hull c.h. of
the vectors . Conversely assume that . Without loss of generality we can

assume that span . We can subdivide into simplices to assume that
belongs to the
convex hull of vectors such that among them are
linearly independent. Then the space of solutions of the system of linear equations
is one-dimensional and is generated by a vector . Since
, we can assume that has nonnegative coordinates, and hence

.
Assume ss is projective. Since is not in the convex hull of the
character vectors , there exists a linear function such that

. This is a well-known assertion from the theory of convex sets
(called the Theorem on a Supporting Hyperplane). Obviously we can choose
to be rational, i.e., defined by for some
, i.e., there exists a solution of
. Assume that
for some . Then tq . Let
s

such that
t
We can choose
. For any
we have
(12.8)
Taking the dot-product of both sides with , we obtain
(12.9)
12.2. FANS
195
Consider the action of

The restriction of this action to the open subset

coincides with the action

(12.10)
of
on the weighted projective space
This action contains in its kernel

the finite subgroup of equal to the group of
. The induced action of the torus

points such that r

is isomorphic
to our old action.

Clearly each
is a
linear combination of monomials such that
Comparing this with equations (12.8) and (12.9) we find an isomorphism of vector
spaces
and also an isomorphism of graded algebras
Thus we obtain
ss
(12.11)
ss
ss ss
since each point in the
Obviously
weighted projective space lying on the hyperplane is
with is divisible by ).
unstable (because each
To summarize we obtain
196
Proposition 12.2. Let be the convex hull of the vectors

. Then ss is projective and
ss
ss
for some and
where
by the formula (12.10).
. Assume that
acts on
Applying the numerical criterion of stability we can find the set of unstable
points in t t . It follows from Chapter 9 (up to some modifications
using a weighted projective linearization, i.e. a -equivariant embedding of a

variety into a weighted projective space) that a point is unstable
if and only if the set such that satisfies the
property that does not belong to the convex hull of the vectors t
t t .
t
12.3 Examples
Let us give some examples.
Example 12.1. Let
act on
by the formula
We have

It is easy to see that vectors
form a basis of
, the vectors are equal to
choose the dual basis of

We can take for a new basis of the vectors
. Then
. If we
12.3. EXAMPLES
197
Let us linearize the action by taking the line bundle

have an isomorphism of graded rings
span
. Then we
are the unknowns . Thus
Obviously the minimal generators of the ideal

the cones of our fan are
, where
e3
1
2
e2
3
e1
This is the fan defining the projective space (see [32]). Let us see the corresponding gluing. We can take for a basis of the dual basis of
which is the set of vectors
We easily find
These are the coordinate rings of the standard open subsets of .

Example 12.2. Consider the action of
on
by the formula

198
We have
Let us choose the following basis of
We can express the vectors

as follows:
Choose
tion
of
and consider the monoid
in terms of the dual basis
of nonnegative solutions of the equa-
we have
. If
or
we can subtract
or from to obtain a vector from . If

we have
, and we do the same by subtracting
. This shows that is generated over by and . This
means that the unknowns are the minimal generators of the ideal
.
For any
Thus the fan consists of two cones
span
span
span
The dual cones are
span
The quotient is obtained by gluing together two nonsingular algebraic varieties

with the coordinate algebras
Similarly if we take
we get that the fan consists of two cones
span span
12.3. EXAMPLES
199
The quotient is obtained by gluing together two nonsingular algebraic varieties

with the coordinate algebras
If we now change the linearization by taking we get for
all , hence is generated by . Then we have only one cone spanned

by the four vectors . The toric quotient is isomorphic to the affine variety with
the coordinate algebra
One should compare this with our previous computation of this quotient in
Example 8.6 from Chapter 8. We see here a general phenomenon: two toric varieties and whose fans have the same set of one-dimensional edges of their
cones (called the 1-skeleton of a fan) differ by a special birational modification.
We refer the interested reader to [90] for more details.
Example 12.3. Let consist of the following four cones in
span
span
span
span
This is shown in the following figure.
-e1
3
--e*22
200
We have
hence the action is given by
The variety is obtained by gluing four affine planes with coordinate rings

. This is also seen by
It is easy to see that is isomorphic to the product
observing that
Example 12.4. Recall that the coordinate ring of the Grassmannian Gr

is

SL
isomorphic to Pol Mat

. It is generated by the bracket functions
in GL acts naturally
. The
torus
of
digaonal
matrices
on Mat by multiplying a matrix on the right by a diagonal matrix. It is
easy to see that each function spans an eigensubspace corresponding to the

over Gr
character , where Consider the cone G
r
as a closed subvariety of . Then the torus acts on by multiplying
each coordinate function by . Thus the action is given by the matrix with
columns equal to . Let the linearized line bundle be , where .
It is easy to see that
where
is the set of vectors
where each appears exactly times in the sets . In other words,
is in a bijective
, where
correspondence with the set of tableaux of degree and size

. Let be the restriction of to G r . Then
G r Pol Mat SL Pol SL
This shows that

G
r
ss SL
12.3. EXAMPLES
201
Also, we see that there is a natural closed embedding
ss
The latter quotient is a toric variety of dimension , where depends
.
only on . Let us denote it by . For example, take
We have
It is easy to see that the monoid

of nonnegative integer solutions of the equa
tion
consists of vectors with

. Thus and
. Thus
The embedding
is of course the Veronese embedding.
One can go in the opposite direction by identifying any toric variety with
a categorical quotient of some open subset of an affine space. We state without
proof the following result of D. Cox ([16]).
Theorem 12.3. Let be a toric variety determined by a -fan . To each

one-dimensional edge of the 1-skeleton
of assign a variable and consider the
polynomial algebra
variables. For each cone
generated by these
let

, where is the complementary
set to the 1-skeleton of . Let . Let be the primitive

vectors of the lattice which span one-dimensional edges of the cones from .
, and let be
Let be the matrix whose columns are the vectors
an
matrix whose rows form a basis of the module . Assume
that the vectors span . Then
(i)
with the action of
where are the columns of ,

(ii) is simplicial if and only if
202
Remark 12.1. Note that applying this construction to the toric varieties ob
tained as the quotients ss we obtain ss and the action
is isomorphic to the one we started with. However, in general, ss for
any . One reason for this is that our quotients are always quasi-projective
and there are examples of nonquasi-projective toric varieties. Another reason is
simpler. The fans we are getting from our quotient constructions are full in the
following sense. One cannot extend them to larger fans with the same 1-skeleton.
The torus which acts on has a very nice interpretation. Its character group
is naturally isomorphic to the group Cl of classes of Weil divisors on
.
Also, if the vectors do not span , the assertion is true if we replace by a
diagonalizable algebraic group, an extension of with the help of a finite abelian
group.
The theory of toric varieties is a subject of many books and articles. We refer to
[32] and [82] for the bibliography. The fact that any toric variety can be obtained
as a categorical quotient of an open subset of affine space was first observed by M.
Audin ([3]) and D. Cox ([16]). The relationship between solutions of systems of
linear integral equations, Grobner bases and toric varieties is a subject of the book
[111]. The systematic study of quotients of toric varieties by a torus can be found
in [56]. We refer to [52] and [10] for the theory of variation of a torus quotient
with respect to the linearization.
Exercises
12.1 Consider the action s and take . Show
that the quotient is isomorphic to the blow-up of at the origin. Draw the
corresponding fan.
act on by the formula

Take , where . Show that the quotient is isomorphic to
12.2 Let
the blow-up of the projective plane at three points. Draw the picture of the fan.
EXERCISES
203
12.3 Take a fan in formed by three one-dimensional cones spanned by the

unit vectors . Using Coxs Theorem represent the toric variety as a
geometric quotient.
12.4 A toric variety is nonsingular if and only if each is spanned by a
part of a basis of the lattice . Show that is nonsingular if and only if
the stabilizer of each point of is equal to the same subgroup of .
12.5 Describe the fan and the corresponding toric variety .
12.6 Show that the moduli space of six lines in is isomorphic to a double cover
of the toric variety .
Gr defined by assigning to a
12.7 Consider the isomorphism Gr
linear subspace of a linear space its annihilator in the dual space . Show
that this isomorphism commutes with the action of the torus , and induces
an isomorphism of the quotients
. Show that this isomorphism

coincides with the association isomorphism defined in Chapter 11.
Bibliography
[1] A. ACampo-Neuen, Note on a counterexample to Hilberts fourteenth
problem given by P. Roberts, Indag. Math. 5, 1994, 253257.
[2] D. Allcock, Moduli space of cubic threefolds, J. Alg. Geometry, to appear.
[3] M. Audin, Topology of Torus Actions on Symplectic Manifolds, Progress in
Math. vol. 93, Birkhauser, 1991.
[4] N. Beklemishev, Invariants of cubic forms of four variables, Vestnik
Moskov. Univ., Ser. I Mat. Mekh. 1982, no. 2, 4249; English translation:
Moscow Univ. Bull. 37, 1982, 5462.
[5] F. Bogomolov, P. Katsylo, Rationality of some quotient varieties, Mat.
Sbornik (N.S) 126 , 1985, 584589.
[6] A. Borel, Linear Algebraic Groups, W. A. Benjamin Inc., 1969 (new edition: Springer-Verlag, 1991).
[7] A. Borel, Essays in the History of Lie Groups and Algebraic Groups, History of Mathematics, vol. 21, Amer. Math. Soc., London Math. Soc., 2001.
[8] N. Bourbaki, Algebra, Springer-Verlag, 1989.
[9] N. Bourbaki, Commutative Algebra, Springer-Verlag, 1989.
[10] M. Brion, C. Procesi, Action dun tore dans une variete projective, in Operator Algebras, Unitary Representations, Enveloping Algebras, and Invariant Theory, Progress in Math. vol. 192, Birkhauser, 1990, pp. 509539.
[11] C. Chevalley, Invariants of finite groups generated by reflections, Amer. J.
Math. 77, 1955, 778782.
205
206
BIBLIOGRAPHY
[12] A. Clebsch, Theorie der binaren algebraischen Formen, Teubner, 1872.

[13] A. Coble, Algebraic Geometry and Theta Functions, Colloquium Publ. vol.
10, Amer. Math. Soc. 1929; reprinted by A.M.S., 1982.
[14] J. L. Coolidge, A Treatise on Algebraic Plane Curves, Oxford Univ. Press,
1931; reprinted by Dover Publ. , 1959.
[15] F. Cossec, I. Dolgachev, Enriques Surfaces I, Progress in Math., vol. 76,
Birkhauser, 1985.
[16] D. Cox, The homogeneous coordinate ring of a toric variety, J. Alg. Geom.
4, 1995, 1750.
[17] C. DeConcini, D. Eisenbud, C. Procesi, Young diagrams and determinantal
varieties, Invent. Math. 56, 1980, 129165.
[18] J. A. Dieudonne, J. B. Carrell, Invariant Theory, Old and New, Acad. Press,
1971.
[19] J. Dixmier, On the projective invariants of quartic plane curves, Advances
in Math. 64, 1987, 279304.
[20] J. Dixmier, D. Lazard, Minimal number of of fundamental invariants for
the binary form of degree 7, J. Symb. Computation. 6, 1988, 113115.
[21] I. Dolgachev, Rationality of fields of invariants, in Algebraic Geometry,
Bowdoin, Proc. Symp. Pure Math. vol. 46, 1987, pp. 316.
[22] I. Dolgachev, Introduction to Geometric Invariant Theory, Lect. Notes Series, n. 25, Seoul Nat. Univ., 1994.
[23] I. Dolgachev, Y. Hu, Variations of geometric invariant theory quotients,
Publ. Math. de lIHES, 87, 1998, 551.
[24] I. Dolgachev, V. Kanev, Polar covariants of plane cubics and quartics, Adv.
Math. 98, 1993, 216301.
[25] I. Dolgachev, D. Ortland, Point Sets in Projective Space and Theta Functions, Asterisque, vol. 165, 1989.
[26] D. Eisenbud, Commutative Algebra, Springer-Verlag, 1995.
BIBLIOGRAPHY
207
[27] D. Eisenbud, S. Popescu, The projective geometry of the Gale transform, J.

of Algebra, 230, 2000, 127173.
[28] E. B. Elliot, An Introduction to the Algebra of Quantics, Oxford Univ.
Press, 1895; reprinted by Chelsea Pub. Co. , 1964.
[29] A. Fauntleroy, On Weitzenbocks theorem in positive characteristic, Proc.
Amer. Math. Soc. 64, 1977, 209213.
[30] F. Faa di Bruno, Theorie des formes binaire, Librairie Brero Succ. de P.
Marietti, 1876.
[31] J. Fogarty, Invariant Theory, W. A. Benjamin, Inc., 1969,
[32] W. Fulton, Introduction to Toric Varieties, Princeton Univ. Press, 1993.
[33] W. Fulton, Young Tableaux, London Mathematical Society Student Texts,
vol. 35, Cambridge Univ. Press, 1997.
[34] W. Fulton, J. Harris, Representation Theory, Springer-Verlag, 1991.
[35] F. von Gall, Das vollstandig Formensystem der binaren Form achter Ordnung, Math. Ann. 17, 1880, 3152, 139152.
[36] F. von Gall, Das vollstandig Formensystem der binaren Form
Math. Ann. 31, 1888, 318336.
ter
Ordnung,
[37] P. Gordan, Uber

ternare Formen dritter Grades, Math. Ann. 1, 1869, 90
128.
[38] P. Gordan, Invariantentheorie, Teubner, Leipzig, 1885-87; reprinted by
Chelsea Publ. Co., 1987.
[39] J. H. Grace, A. Young, The Algebra of Invariants, Cambridge Univ. Press,
1903; reprinted by Chelsea Publ. Co. New York, 1965.
[40] Ph. Griffiths, J. Harris, Principles of Algebraic Geometry, John Wiley and
Sons, 1978.
[41] F. G. Grosshans, Algebraic Homogeneous Spaces and Invariant Theory,
Lect. Notes in Math., vol. 1673, Springer-Verlag, 1997.
208
BIBLIOGRAPHY
[42] S. Gundelfinger, Zur Theorie der ternaren cubischen Formen, Math. Ann.
4, 1871, 144168.
[43] G. B. Gurevich, Foundations of the Theory of Algebraic Invariants, P. Noordhoff, 1964.
[44] W. Haboush, Reductive groups are geometrically reductive, Annals of
Math. 102 1975.
[45] B. Harbourne, On Nagatas conjecture, J. Algebra 236, 2001, 692702.
[46] R. Hartshorne, Algebraic Geometry, Springer-Verlag, 1977.
[47] D. Hilbert, Theory of Algebraic Invariants, Cambridge Univ. Press. 1993.
[48] W. V. D. Hodge, D. Pedoe, Methods of Algebraic Geometry, vol. 2, Cambridge Univ. Press, 1952 (reissued in the Cambridge Mathematical Library
in 1994).
[49] R. Howe, The classical groups and invariants of binary forms, in The Mathematical Heritage of Hermann Weyl, Proc. Symp. Pure Math. v. 48, Amer.
Math. Soc. , 1988, pp. 133166.
[50] R. Howe, R. Huang, Projective invariants of four subspaces, Adv. Math.
118, 1996, 295336.
[51] R. Huang, Invariants of sets of linear varieties, Proc. Nat. Acad. Sci. USA
88 1990, 45574560.
[52] J. Humphreys, Linear Algebraic Groups, Springer-Verlag, 1975.
[53] A. Iarrobino, V. Kanev, Power Sums, Gorenstein Algebras and Determinantal Loci, Lect. Notes in Math. vol. 1721, Springer-Verlag, 1999.
[54] J. Igusa, Arithmetic theory of moduli for genus two, Ann. of Math. 72,
1979, 241266.
[55] N. Iwahori, N. Matsumoto, On some Bruhat decompositions and the structure of the Hecke ring of p-adic groups, Publ. Math. de lIHES 25, 1965,
548.
[56] M. Kapranov, B. Sturmfels, A. Zelevinsky, Quotients of toric varieties,
Math. Ann. 290, 1991, 643655.
BIBLIOGRAPHY
209
[57] P. Katsylo, Rationality of the moduli varieties of plane curves of degree ,

Mat. Sbornik (N.S), 136, 1988, 377384; Engl. translation: Math. USSR
Sbor. 64, 1989, 375-381.
[58] P. Katsylo, Rationality of the moduli variety of curves of genus 3, Comment. Math. Helv. 71, 1996, 507524.
[59] G. Kempf, Instability in invariant theory, Ann. of Math. 108, 1978, 299
316.
[60] G. Kempf, L. Ness, The length of vectors in representation spaces, in Algebraic Geometry, Proceedings, Copenhagen 1978, Lect. Notes in Math. vol.
732, Springer-Verlag, 1979, pp. 233243.
[61] D. Khadzhiev, Certain questions in the theory of vector invariants, Mat.
Sbornik (N.S.) 72, 1967, 420435; English translation: Math. USSRSbor.
1, 1967, 383396.
[62] F. Kirwan, Cohomology of Quotients in Symplectic and Algebraic Geometry, Math. Notes, vol. 31, Princeton Univ. Press, 1984.
[63] H. Kraft, Geometrische Methoden in der Invariantentheorie, Vieweg, 1985.
[64] H. Kraft, C. Procesi, Classical Invariant Theory:
http://www.math.unibas.ch.
A Primer,
[65] H. Kraft, P. Slodowy, T.A. Springer (editors), Algebraic Transformation

Groups, DMV Seminar, Bd. 13, Birkhauser Verlag, 1989.
[66] I. MacDonald, Symmetric Functions and Hall Polynomials, Oxford Univ.
Press, 1979.
[67] D. McDuff, L. Polterovich, Symplectic packings and algebraic geometry,
Invent. Math. 115, 1994, 405434.
[68] W. Fr. Meyer, Bericht u ber den gegenwartigen Stand der Invariantentheorie,
Jahresbericht Deut.Math.Ver. 1, 1892, 79292.
[69] S. Mori, Classification of higher-dimensional varieties in Algebraic geometry, Bowdoin, Proc. Symp. Pure Math. vol. 46, Part I, Amer. Math. Soc.,
1987, pp. 269333.
210
BIBLIOGRAPHY
[70] S. Mukai, Counterexamples to Hilberts Fourteenth Problem for the 3dimensional additive group, preprint RIMS-1343, Kyoto Univ., 2001.
[71] D. Mumford, Hilberts fourteenth problem the finite generation of subrings such as rings of invariants, in Mathematical Developments arising
from Hilbert Problems, Proc. Symp. in Pure Math., vol. 28, Amer. Math.
Soc., 1976, 431 444.
[72] D. Mumford, Abelian Varieties, Oxford Univ. Press, 1985.
[73] D. Mumford, J. Fogarty, F. Kirwan, Geometric Invariant Theory, 3d edition, Springer-Verlag, 1994.
[74] D. Mumford, K. Suominen, Introduction to the theory of moduli, in Algebraic Geometry, Oslo 1970, Wolters-Noordhoff, 1970, pp. 171222.
[75] M. Nagata, On the embedding problem of abstract varieties in projective
varieties, Mem. Coll. Sci. Kyoto (A) 30, 1956, 7182.
[76] M. Nagata, On the 14th problem of Hilbert, Amer. J. Math. 81, 1959, 766
772.
[77] M. Nagata, Invariants of a group in an affine ring, J. Math. Kyoto Univ. 3,
1964, 369377.
[78] M. Nagata, Lectures on the Fourteenth Problem of Hilbert, Tata Institute of
Fund. Research, 1965.
[79] I. Naruki, Cross ratio variety as a moduli space of cubic surfaces, Proc.
London Math. Soc. (3) 44, 1982, 130.
[80] P. E. Newstead, Introduction to Moduli Problems and Orbit Spaces, Tata
Institute for Fund. Research, 1978.
[81] P. E. Newstead, Invariants of pencils of binary cubics, Math. Proc. Cambridge Phil. Soc. 89, 1981, 201209.
[82] T. Oda, Convex Bodies and Algebraic Geometry, Springer-Verlag, 1988.
[83] P. Olver, Classical Invariant Theory, London Mathematical Society Student Texts 44, Cambridge Univ. Press, 1999.
BIBLIOGRAPHY
211
[84] A. Onishchik, E. Vinberg, Lie Groups and Algebraic Groups, SpringerVerlag, 1990.
[85] P. Orlik, L. Solomon, Singularities II: Automorphisms of forms, Math. Ann.
231, 1978, 229240.
[86] V. Popov, The Picard group of homogeneous spaces of linear algebraic
groups and one-dimensional homogeneous vector bundles, Izv. Akad. Nauk
SSSR, Ser. Math. 38, 1974, 292322; English translation: Math. USSRIzv.
8, 1975, 301327.
[87] V. Popov, On Hilberts theorem on invariants, Doklady Akad. Nauk SSSR
249, 1979, 551555; English translation: Soviet Math.Dokl. 20, 1979,
13181322.
[88] V. Popov, Groups, Generators and Syzygies and Orbits in Invariant Theory,
Transl. Math. Monographs, vol. 100, Amer. Math. Soc., 1992.
[89] V. L. Popov, E. B. Vinberg, Invariant theory, in Algebraic Geometry IV,
Encycl. Math. Sci., vol. 55, Springer-Verlag, 1994, pp. 123278.
[90] M. Reid, Decomposition of toric morphisms, in Arithmetical Geometry,
vol. II, Birkhauser, 1983, pp. 395418.
[91] M. Roberts, The covariants of a binary quantic of the n-th degree, Quarterly
J. Math. 4, 1861, 168178.
[92] P. Roberts, An infinitely generated symbolic blow-up in a power series ring
and a new counterexample to Hilberts fourteenth problem, emphJ. Algebra
132, 1990, 461473.
[93] M. Rosenlicht, Toroidal algebraic groups, Proc. Amer. Math. Soc. 12, 1961,
984988.
[94] M. Rosenlicht, A remark on quotient spaces, Ann. Acad. Brazil 35, 1963,
487489.
[95] G. Rousseau, Immeubles spheriques et theorie des invariants, Comptes
Rendus Acad. Sci., Paris, Ser. 1 286, 1978, 247250.
[96] G. Salmon, Lessons Introductory to the Modern Higher Algebra, Hodges
and Smith, 1859; reprinted by Chelsea Publ. Co., 1964.
212
BIBLIOGRAPHY
[97] G. Salmon, A Treatise on the Analytic Geometry of Three Dimensions,

Longmans and Green, 19121915; reprinted by Chelsea Publ. Co., 1965.
[98] G. Salmon, A Treatise on the Higher Plane Curves, Hodges, Foster and
Figgis, 1879; reprinted by Chelsea Publ. Co., 1960).
[99] G. Segal, Equivariant K-theory, Publ. Math. de lIHES 34, 1968, 129151.
[100] A. Schofield, Birational classification of moduli spaces, in Infinite Length
Modules (Bielefeld, 1998), Trends Math., Birkhauser, 2000, pp.297309.
[101] C. S. Seshadri, Theory of moduli, in Algebraic Geometry, Arcata, Proc.
Symp. in Pure Math., vol. 29, Amer. Math. Soc., 1975, 263304.
[102] I. Shafarevich, Basic Algebraic Geometry, vols. 1,2, Springer-Verlag, 1994.
[103] J. Shah, A complete moduli space for K3 surfaces of degree 2, Ann. of
Math. 112, 1980, 485510.
[104] J. Shah, Degenerations of K3 surfaces of degree 4, Trans. Amer. Math. Soc.
263, 1981, 271308.
[105] G. Shephard, J. Todd, Finite unitary reflection groups, Canadian J. Math.
6, 1954, 274304.
[106] N. Shepherd-Barron, The rationality of some moduli spaces of plane
curves, Compos. Math. 67, 1988, 5188.
[107] T. Shioda, On the graded ring of invariants of binary octavics, Amer. Math.
J. 89, 1967, 10221046.
[108] I. Schur, Vorlesungen u ber Invariantentheorie, Springer-Verlag, 1968.
[109] T. A. Springer, Invariant Theory, Lect. Notes in Math., vol. 585, Springer
Verlag, 1977.
[110] T. A. Springer Linear Algebraic Groups, Birkhauser, 1998.
[111] R. Stanley, Enumerative Combinatorics, vol. 1, Wadsworth and
Brooks/Cole, 1986.
[112] R. Steinberg, Nagatas example, in Algebraic Groups and Lie Groups,
Cambridge Univ. Press, 1997, pp. 375384.
BIBLIOGRAPHY
213
[113] B. Sturmfels, Algorithms in Invariant Theory, Springer-Verlag, 1993.

[114] B. Sturmfels, Grobner Bases and Convex Polytopes, University Lecture
Series, vol. 8, Amer. Math. Soc., 1996.
[115] M. Thaddeus, Geometric invariant theory and flips, J. Amer. Math. Soc. 9,
1996, 691723.
[116] H.W. Turnbull, The projective invariants of four medials, Proc. Edinburgh
Math. Soc. 7, 1942, 5572.
[117] D. Vazzana, Projections and invariants of lines in projective space, Univ.
Michigan Ph. D. thesis, 1999.
[118] D. Vazzana, Invariants and projections of six lines in projective space,
Trans. Amer. Math. Soc. 353, 2001, 26732688.
[119] R. Weitzenbock, Invariantentheorie, Noordhoff, 1923.
[120] R. Weitzenbock, Uber

die Invarianten von linearen Gruppen, Acta Math.
58, 1932, 231293.
[121] H. Weyl, The Classical Groups, Their Invariants and Representations,
Princeton Univ. Press, 1946; reprinted by Princeton Univ. Press, 1997.
[122] D. Zaitsev, Configurations of linear subspaces and rational invariants,
Michigan Math. J. 46, 1999, 187202.
Index of Notation
Cov , 72
Cov , 68
, 142
GL , 2
GL SL , 37

, 105
Gr , 21
Gr , 167

Hyp , 147
, 21
, 142
Mat , 2

, 17
O , 28
Pic , 107
Pol , 4
Pol Pol , 4
Pol , 3, 5
Projm , 39
SL , 32
SL , 13
, 203
, 20
, 17
Spec , 2
Specm , 35
Sym , 6
Wt , 75
av, 33
, vii
,2
,4
, 147
, 36
, 6
SL , 4
, 100
SU , 32
, 78
, 50
, 176
, 50
, 76
, 50

, 176
, 4
, 76
,
, 32, 36
, 78
Specm , 2
, 36
, 94
ss s s us , 117
sr
, 194
, 167
,
reg 101
ss s , 171
alg alg alg, 108
, 13
,5
215
INDEX
216
, 37
, vii, 14
, 127
, 41
, 105
, 119
, 35
, 108
, 132
, 79
, 12
, 19
, 18
, 33
, 145
, 13
, 39
O , 100
pol , 5
res , 5
wt s wt , 137
wt s wt , 138
symb , 9
, 13
Tab , 20
, 20
Tab
hom

, 17
,4
, 49
, 35
, 21
, 191
, 83
, 21
Index
absolute invariant, 150, 155, 161
action
faithful, 89
linearizable, 124
rational, 37
regular, 37
additive group, 37, 56, 62, 101
adjugate matrix, 18
affine algebraic group, 35
affine cone, 117, 131
algebra
of covariants, 70
of invariants, 2
algebraic group
diagonalizable, 143
exceptional type, 42
geometrically reductive, 42
linear, 37
linear reductive, 42
reductive, 42
semisimple, 42
simple, 42
algebraic torus, 42
algebraic variety
abstract, 118
affine, 35
complete, 132
projective, 40
proper, 132
quasi-affine, 50
quasi-projective, 40
ample, 103
very, 103
ampleness criterion, 116
apolar, 15
association isomorphism, 181
averaging operator, 30, 71
base-point-free
line bundle, 103
linear system, 111
binary form, 4
bracket function, 12, 23
Capelli identity, 28
catalecticant, 10, 13, 15, 28, 150
catalecticant invariant, 15
catalecticant matrix, 15
Cayley -process, 27
Cayley operator, 17
Cayley-Sylvester formula, 82
chamber, 174
Chevalleys criterion, 100
Chow variety, 100
Clebsch-Gordan decomposition, 90
closed embedding, 40
coaction homomorphism, 37
cocycle, 104
combinant, 69, 87
complete reducibility, 70
concomitant, 69
217
INDEX
218
contravariant, 69
convex polyhedral cone, 188
coordinate algebra, 35
covariant, 66
of an action, 69
degree of , 66
order of, 66
cross-ratio, 88, 162, 176
diagonal action, 65
discriminant, 9, 14, 15, 26, 28, 68,
146, 147, 149, 150, 152, 154,
161, 163
equivalence relation, 91
equivariant function, 1
exceptional curve, 177
fan, 191
N-fan, 191
complete, 193
simplicial, 192
flag complex, 140
flip, 127, 174
Fundamental Theorem
First, 20, 67
Second, 24
G-variety, 92
general points, 61
geometric grading, 38
gluing construction, 117
gluing data, 117
good -action, 38
GordanHilbert Theorem, 30
group scheme, 98
Haboushs Theorem, 42
Halphen pencil, 59
Hankel determinant, 10
Hermite Reciprocity, 82
Hesse form, 88
Hessian, 68, 88
highest weight, 76
highest weight module, 76
highest weight vector, 76
Hilberts Problem 14, 47
HilbertMumford criterion, 129
homogeneous localization, 40
hook formula, 86
hypersimplex, 173
hypersurface, 145
Igusa quartic hypersurface, 181
inflection point, 56
invariant function, 1
isogeneous, 42
isotropy subgroup, 98
Jacobian, 69
Kempf-stable, 141
Laplace formula, 19
Laurent monomial, 73
Lefschetz Theorem, 109
line -bundle, 104
-effective, 127
linear algebraic groups, 37
linearization, 104
democratic, 169
trivial, 105
module of covariants, 70
moment map, 142
multi-degree, 8, 11
multi-weight, 11
multihomogeneous, 11
INDEX
multiisobaric, 11
multiplicity, 55, 72
Nagata Theorem, 41
Nagatas conjecture, 61
Nagatas counterexample, 52
normal ring, 45
null-cone, vii, 32, 117, 120
observable subgroup, 50
omega-operator, 17
one-parameter subgroup
adapted, 140
destabilizing, 138
most responsible for instability,
140
of a torus, 77
of an algebraic group, 129
order
of concomitant, 69
of contravariant, 69
parabolic subgroup, 50
Plucker equations, 24
plethysm decomposition, 81
point
-point, 36
polar hypersurface, 14
polarization, 5
polarization map, 6
projective invariant, 46
projective space, 39
projective spectrum, 38
proper map, 132
quadratic form, 4
quantic, 87
quotient
categorical, 92
219
geometric, 92
good categorical, 94
good geometric, 92
radical, 42
rank, 74
rational action
on an algebra, 37
rational character
of a torus, 73
of an algebraic group, 106
rational convex polyhedral cone, 188
rational quotient, 100
rational representation, 37, 73
reductive algebraic group, 42
regular action, 37
regular function, 35
restitution, 5
resultant, 69
Reynolds operator, 71
ring of invariants, 2
ringed space, 118
root, 73
dual, 77
negative, 74
positive, 74
simple, 74
Schrodinger representation, 142
Schur multipliers, 108
Segre cubic primal, 180
Segre quartic primal, 181
self-association involution, 181
semi-stable, 115
semiinvariant, 65
skeleton, 199
small resolution, 126
solvable algebraic group, 42
INDEX
220
stabilizer, 41
stable, 115
properly, 116
straightening law, 23
structure sheaf, 118
symbolic expression, 9
tableau, 12
degree, 12
homogeneous, 12
rectangular, 12
standard, 22
tableau function, 13
homogeneous, 13
ternary cubic, 153
theta group, 141
Tits conjecture, 140
toric variety, 192
transvectant, 68
transversal, 171
unitary trick, 32
unstable, 115
valuative criterion
of properness, 132
of separatedness, 132
variation of quotients, 143
Veronese embedding, 121
Veronese map, 83, 111
Veronese variety, 83
Weierstrass equation, 154
weight, 30, 46, 67
dominant, 76, 78
fundamental, 78
in a representation, 73
of covariant, 67
weight polytope
of a point, 136
of the representation, 135
weight set, 136
weight space, 135
weighted projective
linearization, 196
space, 39, 180
Weitzenbocks Theorem, 51
Weyl group, 138
Young diagram, 86
Zariskis problem, 49
Index
, 12
,5
Cov , 70
Cov , 66
, 140
GL , 2
GL SL , 36
, 103
Gr , 21
Gr , 165
Hyp , 145
, 21
, 140
, 12
,2
, 174
, 48
, 76
Specm , 2
, 36
, 92
, 192
Mat , 2

, 17
O , 28
Pic , 105
Pol , 4
Pol Pol , 4
Pol , 3, 5
Projm , 38
SL , 32
SL , 13
, 201
,6
, 17
Spec , 2
Specm , 35
Sym , 6
Wt , 73
, 74
, 48

, 174
, 4
, 74
,
, 32, 36
ss s s
sr
us
, 115
, 165
reg , 99
ss s , 169
alg alg alg, 106
, 13
, 48
, 145
, 36
, 6
SL , 4
, 98
SU , 32
, 76
, 4
221
INDEX
222
av, 33
, vii
, 37
, vii,
14
, 125
, 40
, 103 , 117
, 35
, 106
, 130

, 79 , 77
, 12
, 19
, 18
, 33
, 143
, 12
, 13
, 39
O , 98
pol , 5
res , 5
wt s wt , 135
wt s wt , 136
symb , 9
, 13
Tab , 20
, 20
Tab
hom

, 17
,4
, 47
, 35
, 21
, 189
, 81

, 21
, 83

Lectures On Invariant Theory S

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Lectures On Invariant Theory S

Uploaded by

Copyright:

Available Formats

Preface

This book is based on one-semester graduate courses I gave at Michigan in 1994

The symbolic method

The First Fundamental Theorem

Reductive algebraic groups

Numerical criterion of stability

11 Configurations of linear subspaces

a projective algebraic variety with the projective coordinate algebra isomorphic

algebraic action of a linear algebraic group on a normal quasi-projective algebraic

The notion of an invariant is one of the most general concepts of mathematics.

then an equivariant function is a function

. So if we know the set of orbits , we know all

CHAPTER 1. THE SYMBOLIC METHOD

Example 1.1. Let be a finitely generated algebra over a field

is a -subalgebra of . It is called the algebra of invariants. This definition

on the space has two orbits but no invariant nonconstant functions.

is the matrix with the entries

. Then, the affine variety

1.1 FIRST EXAMPLES

preted as matrices with entries in and we can view elements of as

be a linear representation of a group in . We consider the associated action of

CHAPTER 1. THE SYMBOLIC METHOD

with respect to some basis

polynomial functions Pol . Since is infinite (we assumed it to be algebraically

product of the dual vector space and Pol is isomorphic to the

maticians of the nineteenth century made a contribution to this problem. Complete

are known in the case

coefficients, and hence elements of

Polarization and restitution

1.2. POLARIZATION AND RESTITUTION

defined by the formula

A homogeneous polynomial Pol

way by the following properties:

can be defined in a similar

, for any and any ;

the map pol defined by the formula

is called the polarization of . For any

. It is immediately checked that res

Since we assumed that char , we obtain that each Pol is equal to

the restitution of a unique symmetric -multilinear form, namely pol .

CHAPTER 1. THE SYMBOLIC METHOD

Thus any polynomial

1.2. POLARIZATION AND RESTITUTION

If we denote by a th copy of the basis

Here, we consider the product of

If we write , then the right-hand side is equal to

als . For any Sym we can set

This shows that the map from Pol

is a perfect duality, i.e., it defines isomorphisms

CHAPTER 1. THE SYMBOLIC METHOD

on . We can view the expression

as a general homogeneous polynomial of degree .

are equal to the value of

This explains the classical notation of a homogeneous polynomial as a power of a

as the partial derivative . Thus we can identify any poly

is defined by the formula

the th symmetric power of the dual space . The polarization map is

is a function on which is a homogeneous polynomial function of degree

Lemma 1.1. We have a natural isomorphism of linear spaces

1.2. POLARIZATION AND RESTITUTION

Proof. The polarization map defines an isomorphism

Using the polarization again we obtain an isomorphism Pol Pol .

using a basis in and its dual basis in . Let