Académique Documents
Professionnel Documents
Culture Documents
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 rst 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 afne 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 rst 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 afne 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.
Contents
Preface Introduction 1 The symbolic method 1.1 First examples . . . . . . . 1.2 Polarization and restitution 1.3 Bracket functions . . . . . Bibliographical notes . . . . . . Exercises . . . . . . . . . . . . i vii 1 1 4 10 13 14 17 17 20 21 22 27 27 29 29 32 35 41 45 46
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
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 Afne algebraic groups . . . . 3.4 Nagatas Theorem . . . . . . . Bibliographical notes . . . . . . . . Exercises . . . . . . . . . . . . . . iii
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
iv 4 Hilberts Fourteenth Problem 4.1 The problem . . . . . . . . 4.2 The Weitzenb ck Theorem o 4.3 Nagatas counterexample . Bibliographical notes . . . . . . 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
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
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 . . . . . . . . . Bibliographical notes . . . . . . Exercises . . . . . . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
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 . . . . . . Bibliographical notes . . . . . . Exercises . . . . . . . . . . . . 129 129 132 133 135 138 142 143 145 145 147 153 161 162 162 165 165 171 181 183 184 187 187 190 196 202 202 205 215 217
10 Projective hypersurfaces 10.1 Nonsingular hypersurfaces 10.2 Binary forms . . . . . . . 10.3 Plane cubics . . . . . . . . 10.4 Cubic surfaces . . . . . . . Bibliographical Notes . . . . . . Exercises . . . . . . . . . . . .
11 Congurations of linear subspaces 11.1 Stable congurations . . . . . . . . . . . . . . . 11.2 Points in 11.3 Lines in . . . . . . . . . . Bibliographical notes . . . . . . . . Exercises . . . . . . . . . . . . . .
12 Toric varieties 12.1 Actions of a torus on an afne space 12.2 Fans . . . . . . . . . . . . . . . . . 12.3 Examples . . . . . . . . . . . . . . Bibliographical notes . . . . . . . . . . . Exercises . . . . . . . . . . . . . . . . . Bibliography Index of Notation Index
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
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 dened 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 bers, all orbits must be closed subsets in the Zariski topology of . This rarely happens when is not a nite 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 eld. Hilbert proves that is nitely generated if SL or GL and any set of generators of denes an invariant regular map from to some afne algebraic variety contained in afne space whose ring of polynomial functions is isomorphic 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 dened 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 ss images of semi-stable points form an invariant open subset and the map ss ss induces a regular map , where (denoted by ) is vii
@ 8 555 3 9 7766 4
D E)
)
DE)
D E)
$#
F@
" !
'%
F@
" &%
"
D H) CGF
8 A ( 2
"
CB
) 0(
viii
INTRODUCTION
a projective algebraic variety with the projective coordinate algebra isomorphic parametrizes projective hyperto . In applications considered by Hilbert, surfaces 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 compactication of the moduli space ss ss of nonsingular hypersurfaces. The bers of the map are not orbits in general; however, each ber 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 nd 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 numerss ical criterion of stability. In many cases it allows one to determine the set very explicitly and to distinguish stable points among semi-stable ones. These are ss and whose stabilizer subgroups are the points whose orbits are closed in ss ss nite. The restriction of the map to the set of stable points s s s is an orbit map . 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 dened by ss a -linearized ample line bundle on . Then one sets ss and ss ss as the restriction of the categorical denes the categorical quotient ss ss quotient . The image variety ss is a closed subvariety ss of . 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 eld 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 Pl cker embedding. This u 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 avor of the
F@
D H)
F@
DX) W U
D H)
D Q E) SR2(
"
D E)
e fV c dDb` U V a `
D E) I" E) D $ H) " H) D D D E)
D E)
D P)
I"
D T)
D E)
D E) " E) D
D E)
D b` U
D E)
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 Pl cker coordinates known in u algebraic geometry as the Pl cker equations. We use the available computations u 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 nite 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 nite 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 nite 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 nite 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 nitely generated. A corollary of this result is a classical theorem of Weitzenb ck about invariants of the additive group. The central o part of this chapter is Nagatas counterexample to Hilberts Fourteenth Problem. It asks about nite generatedness of the algebra of invariants for an arbitrary algebraic group of linear transformations. We follow the original construction of Nagata with some simplications due to R. Steinberg. Chapter 5 is devoted to covariants of an action. A covariant of an afne algebraic group acting on an algebraic variety is a -equivariant regular map from to an afne 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 nitely 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
g X
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 , plane curves of degree 3 and cubic surfaces. spaces of binary forms of degree 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 afne 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 eld sometimes assumed to be of characteristic zero.
q
i ph
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 for any . For example, change under the action, i.e., which satisfy 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
Among all invariant functions there exists a universal function, the projection map from the set to the set of orbits . It satises the property that for any invariant function there exists a unique map such that . So if we know the set of orbits , we know all 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. 1
" F
t 5 ut w
@ 0 "
D GD w ( x
( x pTw
x w ( D x D x 3 w 9XD !9Tw
satisfying , i.e.,
@ s ( sx By9Tw
( SF
"
is a -subalgebra of . It is called the algebra of invariants. This denition ts the general setting if we let Specm be the afne algebraic variety over with coordinate ring equal to , and let be the afne 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 afne with coordinate ring , such a map is dened by a homomorphism of -algebras satisfying for any . 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 dened 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 rst one is that the algebra may not be nitely generated over and so does not dene 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 nitely generated) algebra over , we may still consider the subring of invariants and view any homomorphism of rings as a morphism of afne 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 nitely generated if the group is a reductive algebraic group which acts algebraically on Specm . The second problem is more serious. The afne algebraic variety Specm rarely coincides with the set of orbits (unless is a nite 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. GL act by automorphisms on the polynomial algebra Example 1.2. Let in variables , as follows. For any the polynomial is equal to the th entry of the matrix
D y1
where Specm
is the matrix with the entries is the afne space Mat of dimension
D d 1
(1.2)
D V
d 1
} 0 "
3l A
"
1 srqp " n o
$#
"D1
mhg
Example 1.1. Let be a nitely generated algebra over a eld group of its automorphisms. The subset
and let
be a (1.1)
" Rn
"D1
d 1
D H (
D x o!&(wvD u ! w D t o t " p
d 1
1 " d 1
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 identied 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 . On the other hand, let be symmetric functions in the variables the coefcients of the characteristic polynomial
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 afne space . Now consider the universal map Specm Specm . Its ber over the point dened 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 nite-dimensional vector space over a eld 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
D d 1
7s
DX
c 7667 V 555 3 a
7766 555 3
D T
t
c 7676 V 555 3 a
" hf
d{
" d 1
D T e
7s
U U
The direct sum of the vector spaces Pol is equal to the graded algebra of polynomial functions Pol . Since is innite (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 coefcient . So any element from is a polynomial in the . We want to describe the subalgebra of invariants . The problem of nding is almost two centuries old. Many famous mathematicians 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
are polynomials
1.2
SL To describe the ring Pol Pol 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 is a quadratic form form. Recall its coordinate-free denition: a map if the following two properties are satised:
DvDX) W
d 1 1 1 ( 1 D ) ( D X ` c 555 3 a p 676 V
Y 1 Y W xx 3 3 g ( D3 5 3W zg x 3 W dg W vv d
V "
g D ) W D X) W
5 g
x x x 3 v 4g
(1.3)
V DT e e 6 7 ( D
e d px ( D 555 3 666 ( ( ) D T e d 1 D X
of
D 555 v666 3
D 555 3 7676
Do r
Do e z r 1
D T
DW 555 e1 666 e1
5 ;
is bilinear.
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
res
pol res
Since we assumed that char , we obtain that each Pol the restitution of a unique symmetric -multilinear form, namely Assume that is equal to the product of linear forms have pol
` ( DD 5 pv
Pol
is equal to pol . . We
D ) e
t D TX
V &f " D
D 6567 ( D D 5 5 V " e f
e r Dq 9 (D e 5667 f D 55 3 e V " e fD
pol
j 555 u` 667 ff
D 555 5 666 D
V t
t #
(i)
, for any
and any
V t '
t u#
(i)
D ( D d D ( D e d
for any
and any
D e e ! 3 xxx D3 D
(1.4)
denotes the permutation group on letters. be a basis of and be the dual basis of can be written in a unique way as . Let Sym vector space of symmetric -multilinear forms on . For any and any Sym , we have
is obviously linear, we obtain that Sym has a basis formed by the polarizations of monomials . Applying (1.4), we have
D X e
DT e " D T e
Thus any polynomial Pol can be written uniquely as a sum of mono. This is the coordinate-dependent denition of a homogeneous mials polynomial. Since the polarization map
D v7676 d! 5 5 5 D x x x D
( D D X
(D e 7667 f 555 3 D T e t
e D D e
3 3 6 e ( ( ( e rz rz
6 ( xx D e e x E9 ' x x x D 3 e vx ' xx 3
Here Let
D v666 3 555 e
Taking
( e f (xxx ( 3
res
D T e 3p D 5 xxxp 4g 3p
pol Pol
pol
, we obtain that
Sym
x x x p
t b
. Any be the
Do z e D e 5 T
e D X e t g ( ` g D e D x 3 l e l ( xx l l xx ( x l3 bD e (o X e e D Do
Pol Pol Pol
D l (o D o
D D Do e
` ( D D 'X e
6V Vxxx3
( sD x D x 3 x x 3 xx l l
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 monomials . For any Sym we can set
D v6656 (D xxx 55 DT e t xxx Do e Do e D T e j e 7667 3 yj e 7676 3 f 555 f ( 555 x73V Vx x xx xx dx 3 &( x l l e e e t 5 jzD 565656 3 D 76756 3 SfG(D 75667 D dxxx! ( 55 55 e q
linear forms on as an
D 7676 555 3
and then extend the domain of to all homogeneous degree linearity. Applying (1.5), we get
D 7676 3 555
pol
pol
pol
Pol
if otherwise.
to
pol
of a polynomial of Pol Pol dened by -multilinear form is dual to the , we can (1.7) (1.6) (1.5) 7
Pol
(1.8)
on . We can view the expression as a general homogeneous polynomial of degree . general Thus we get a strange formula
general
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 which acts on linear functions by . It acts on any polynomial Pol as the partial derivative . Thus we can identify any polynomial Pol with the differential operator by replacing the variable with . In this way the duality Pol Pol is dened 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 denition 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
symb Pol
Pol
Sym
4Y
o T D e r &(D ` V D D X e
D e Y 5676 3Y 55 ( )
5 e
5 D o ) W W
3 p
o X
D e 5 T) D X) W W
( `
( (mDx
D ) W
` ( 5 D D
t ED 76576 5 5 3 g ( D T
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 identied 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 be the coordinate functions on Pol , where we write each Pol as 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
pol
symb where
Dzo) W (
1 e (
( X) W oD
5 D o ) W zW
DY 56576 Y 5 DX) W W
xxx (
5 3
D X) W 1 ` g(PD 1 ) W D
D )
3
D ) W
( CD
o X)
Pol
Pol
Sym
Pol
DD 5 vX) W
( X) W DD
D 555 z v7676 3 ( `
D v7676 3 555
o )
555 667
10
( XY
denes a function Pol Pol on the space of binary quartics. It is called 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 by
$ & &3 (Y '3 5555 666 t g " c g 565656 gV a D V 5 D % D %# D # X" $ $ ( D
This implies that the catalecticant is invariant with respect to the group SL
1.3
Bracket functions
eDo dzV e e D c e v766 e 3 2 555 555 v7676 3 c 555 e v7676 D o ) W W 55 e G675 3 5 3 ) e3 5 5 0 1 G675 3 555 555 555 e v666 e 3 v7676 3 676 3 3
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 with the algebra Pol Mat of the algebra 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
D V
3 g 3 4 3 g g
acts on
"
e e 555 3 " a ( 666 m63 c " 666 m5"fCD 667 3 u 555 3 ax 555
g g g g g g 3 g g g 3
!
( v1
G( D
11
where we write a matrix as a collection of its columns. In a similar way the acts on Mat by row multiplication. We say that a polynomial group Pol Mat is multihomogeneous of multi-degree if for any , and any Mat ,
is multiisobaric of multi-weight if the polynomial on the space Mat is multihomogeneous of multi-degree . Let Pol Mat denote the linear space of polynomial 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 denition that the symbolic expression of any invariant polynomial from Pol Pol is multilinear. Let us show that it is also multiisobaric: Proposition 1.1.
9A B CD
where
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 by multiplication by (recall that . on Pol Pol Therefore we get
Since any GL can be written as an th power, we obtain that for some homomorphism GL . Notice that when we x and Pol , the function is a polynomial function in
D vD x 3
( x hTw
D V
D w pw x " D T) W o V V " D P D V t H Iw D ( x D ( 5 W e w Bw W e w Bx w
symb Pol
Pol
SL
Pol Mat
Pol
o VSt t
D e Y 75676 Y ( e Y Y 55 3 ( e Yp(fxx&3Y x ( D D e Y 7676 Y 7676 3 x 555 555 3 e A u W WD e D 555 z v7667 3 x e A 9@ D 1 " 1 D 555 7766 3 e D e 55 3 5 Dc " 6566 " 666 m7"a W (c " 75766 p" " z 3 " 667 m7"a 5 5 555 3 555 3 8 e tc e " 75667 3m7"av1 ( 55 D e Do V D e Y 7676 Y 555 3 W e dD DD p vX) W DD T) W e 3 e D V DD vT) W ) t # w e
t w D P
12
entries of the matrix which is homogeneous of degree . Also, we know that Since is an irreducible polynomial of degree in entries of the matrix, we obtain that is a nonnegative power of . Comparing the degrees we get, for any GL ,
Pol Mat
is GL 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 denition of the symbolic expression, symb is multihomogeneous of multidegree . This proves the assertion.
R S`
An example of a function from Pol Mat is the determinant function . More generally we dene 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 classical notation for the minors det
3 pv
5 ( q EY ` ` h Y ph X
i p
with entries in satisfying the inequalities is homogeneous of degree if each , must satisfy the relation
. . .
A c
..
. . .
f ge
of size
(1.10)
( 9 pY `
A R Y`
It is isobaric of weight but not multihomogeneous if tions one can construct functions from Pol Mat This is done as follows.
j 666 {X 555 f (
W e dD
55 665
55 665 3
3 3
c d`ab
Pol
Pol
3D 3 z f DD 5 j ( X) W
Y `
. Then, for
SL
Y `
t w w D ( D A D e " DD mT) W e D ( x 5 A w Bw D D
DV t w Dw P w
Q C
x Xw
D D 5 W e w ( w P c #`a 1
X
-equivariant, we
eY
U" 1 VX
BIBLIOGRAPHICAL NOTES
13
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 for each tableau on the set of size the function is multiplied by . In 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 SL get a lot of examples of elements in Pol Pol . 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],
9a c
D V Dw w
s t
( h
DD p vX) W
H
s v
c 555 3 5 7667 a
c 555 v 7676 3 a
f ge
f r
h r
5
on Mat by
r r
fe f
h r
d Dr Dr D W e A D
c $5a ab
c `a a a a a
of size
and degree 2 is
s v
3 p
u c
( t s c #`a
s t
D V c
s v
t w Y
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
Pol be the space of quadratic forms on a vector space of 1.2 Let dimension . SL (i) Assume that char or is odd. Show that Pol is generated (as a -algebra) by the discriminant function whose value at a quadratic form is equal to the determinant of the matrix dening its polar bilinear form. in ? (ii) Which level sets of the discriminant function are orbits of SL
Pol
Pol
Pol
Pol dened by
(A
, where .
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 .
gt
D H)
D 555 3 4g 666 g
D 5 5 5 z7667 3 3 ( D 4g p G A D
coincides with pol . (iii) Show that with respect to some basis
D X) 3 W
D D
xxx q
V " W ) " T) W D
j |) f o u) V
. For any
consider the function on . Show that this function extends to denote the restriction of the extended function to . Pol and the pairing
and
D T)
D X
) t B
9A
unless
3 ( 3 p v!4Y pe
j ( A f
x( D |yV D T) W e W D V D E)
x( Y
( D ( D E)
U"
EXERCISES
15
1.5 Consider the bilinear pairing between Pol and Pol dened as in Pol Pol denote the value of this pairing at (1.6). For any by . Show that (i) for xed the assignment denes a linear map
ap
Pol ,
Pol
Pol
if
. Pol is called apolar to a 1.6 In the notation of the preceding exercise, homogeneous form Pol if . Show that (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 dened in Exercise 1.5. The matrix of this map with respect to the basis in Pol dened by the monomials and the basis dened by the monomials is called the catalecticant matrix. Show in Pol 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 dened 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 1.8 Let Pol . For any and any write
(i) Show that the function is multihomogeneous of multi-degree . (ii) Show that pol . 1.9 Find the symbolic expression for the polynomial on . Show that it is an invariant for the group the space of binary quartics Pol SL .
D V D V g4r 3gdggv(G g D ( 3 3 D e Y 7667 Y 555 3 e U"D 7676 f W W 555 3 e puzW W e ( D e e x f3 u xx 3 5 3 W z5765 W ) t e 666 D ) e t 555 3
Do ) e
D X x xx Dv D ( D Do ) e @ I@ e D X)
( ` % Do ) e
D T) e
" D o ) D U" Do ) t
D ) e
) ( D) e ( G`
D X) e
o o D) h'pf ( ) t xxx 3 ( D i !H t
D X t
D ) e
D
16
1.10 Find the polarization of the determinant polynomial . GL be a homomorphism of groups. Assume that is 1.11 Let given by a polynomial in the entries of GL . Prove that there exists a nonnegative integer such that, for all GL , .
P
w D
(Dw D t w P V D V t Gw
o V
" D iV
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 SL generate the algebra Pol Mat . 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.
17
D 5 3 T
h4(D
s 5 5 s vD D s ( T 665
Lemma 2.1.
g
we have (2.1)
(T kf
D V
P
5675 3 3 5
D 555 3 8 666 i
e l D
H 555 9%7667 ( H
3 W D p 5 W o 4X h u l l
T
( D q l X 675 X D 3 k5 5
k
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 literature) matrix adj (see [8], Chapter 3, 11, exercise 10):
n p
13
D 1
( p n 1 D1 (v1 1
% zW iW 3 p l 3 l #d 6{ l #w 6{ #w ~ } ~ } ~ w ~ w #16|y v } y ~6{{#w 3 v61#w 6#w y ~ }{ y ~ }{ y ~ w ~ }{z xw 3 3 ( ~ }{z xw v ~~ }{ y w ~ }{z xw 3 6|y @6#w 3 6|%y v
~ }{ y 6#w
( ~ } { z x w v w ~ } { z x w 6|%Cy ~ 6|1y v
W s r t
r u q
uD t D 5 'vD x {Cg
( D X
h
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
p o n
D v X
h ( D l
D H c a m D X k
T
( D z T
g
D D 5 X g j i x x T i jx l c
a
( D X ( D X
g
where is the sign of the permutation . This immediately gives that . For any subset of set
18
sign
(2.3) (2.2)
D q
j h D 555 f ( 5 #` h 666 3 h{r 55 g 665 v3 g xx ( D X D $z x!D 3 3 ) 55 3 g 665 3 3 g 0 1 3 #` #` 76756 ( gP g p p( 55 c e H 667 3 p3 B 555 3 a V t xxx ( D s s 5 5 s Gs D s ( D 6765 D 5
5 l 765 3 55
%W zW i ( D V
W i d
55 665
W dD
l 7766 9 555 3
h
i 4 D l X 7676 X 555 3
h ( D l1
Now recall the Laplace formula for the determinant of a square matrix size :
where
is a xed partition of the set of rows of and is equal to the sign of where we assume that the elements of each set are the permutation we nd listed in the increasing order. Applying this formula to
X x x x 3 X ( u@pc a
D l 665 9 55 3
where
. . .
..
. . .
20
After multiplying by the sign of the permutation and summing up over the set of permutations, we get the desired formula from the assertion of the lemma.
2.2
Now we are ready to prove the First Fundamental Theorem of Invariant Theory: Theorem 2.1. The algebra of invariants Pol Mat bracket functions .
Proof. Let Pol Mat be the subspace of polynomials which are multiisobaric of multi-weight . It is clear that
A
SL So we may assume that an invariant polynomial Pol Mat 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 functions on of size and let Tab be its subspace spanned by homohom geneous tableau functions of degree . Recall that, as follows from the denition 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,
B
Tab
Tab
hom
Dyw 1x U" w e l D
c #`a
The proof
e l D
SL SL
(#` D14x w D 5 1
D e
D e
Pol Mat
SL
Pol Mat
5 l
D x e
w D ( D 1x 6w
is generated by the
D e
A ( D e l e A D c 555 v 666 3 a e Y D e x t h1
D e f D e Y ` ( $ c #`a D x e
9w
D e
21
Pol Mat
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 linear 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.
where
2.3
Grassmann varieties
D b` ( D 3 D XvD e V 3 e eV
e l D
SL has a nice geometric interpretation. Let Gr be The ring Pol Mat the Grassmann variety of -dimensional linear subspaces in (or, equivalently, -dimensional linear projective subspaces of ). Using the Pl cker map u , we can embed Gr in . The projective coordinates in this projective space are the Pl cker coordinates u . Consider the set of ordered -tuples in . Let be the polynomial ring whose variables are the Pl cker coordinates u 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. of is equal to the homogeneous ideal of the GrassTheorem 2.2. The kernel mann variety Gr in its Pl cker embedding. u
W c a Db` V c #`a i3 h
D e l cv 766 3 a 555 D e
" cD a V pdb` |
D b` | D b` 9|
D b`
symb Pol
Pol
SL
Tab
hom
D e
e D x
D e x e l A W D
D x e
( D T) W DD l
( A W D e l
SL
W e A D
W e (
. We have
DD l vT) W D b` 9|
Y ( $ `
`h Ex i x x " D y
DD e v'V
22
Mat by matrices of maximal rank . Consider the map Spec given by assigning to Mat the values of the bracket functions on . Clearly, the corresponding map of the rings of regular functions coincides with . Also it is clear that the image of is contained in the afne 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 polynomials). Thus it was enough to assume that is homogeneous. Pol 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
2.4
We now describe a simple algorithm which allows one to construct a basis of the space Tab .
c d`ab
p 3
h y
is called standard if
. . .
A c
55 665 3
3 3
c #`a
of size ..
. . .
c A dDb` | D V e
a V
D b` cD 5 b`
cD A b`
3 5 5 f e c 665 s
a V ( D e l
Corollary 2.3.
( D A " e H e
Db` |
o }
0
D b` | }
e H
aV ( DvD V W e l Y ( $ `
t 1
Db` 9| D b` |
" e
D " j f b` D D b`
Db` | e e fV
y
e H
formed
( D wX o e
23
For example,
is not. Theorem 2.3. The tableau functions 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:
and are two xed increasing sequences of numHere 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. Suppose a tableau function is not standard. By permuting the rows of we can assume that for all . Let be the smallest index such 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
c
Here we assume that the second sequence is put in increasing order. It allows us as a sum of the products to express
D 3 7676 3 555 c 3 56566 3 a c #`a 5 D3 555 pd 666 3 D 555 3 666 3 x 3 pu D 5 ( c3 555 3 dpd 666 p 3 7766 3 3 7766 3 a q 555 ac 555 3 pd
i V
p l 3
h l
3 pd D V
s v
c s v
is standard but
r
3 p 3
D e D
c
(2.5)
h 3 E
D V3 3 555 pd 666 3 pd p 3
c
D x
24
c
Substituting this in the product of the bracket functions corresponding to the as a sum of the such that the mark of each is rows of , we express 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 dening Gr bedding is generated by the quadratic polynomials in its Pl cker emu
H
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 dening the Grassmannian Gr are called the Pl cker equations. Corollary 2.3 implies that the Pl cker equations deu u scribe 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 nally can compute something. We start with the case . Let us write any degree homogeneous standard tableau in the form
f ge f f
where denotes a column vector with coordinates equal to . Let length of this vector. It is clear that
D ( Y 5 Y4` Ty ( g 3 Y Y p( g 3 g p(
. . .
. . .
D b`
D x e e
e Ht c e
D b`
a V
e H
s t
g
e g 3 3 e g
3 g 33 g
( W
s t `a a ab
e e g
s t
3 3 g
3 g e 3
G(
e H
be the
D
h
5
f f
f r f f r fe
ab a a a a `a
( w
c
D D D ( 5 4 D r r x D 3 #
. We have the following solutions: Pol Pol
s t
s t
D zx
D 5 v44s`
r ( l!
So if we set , then a standard tableau is determined in by a point with integer coordinates inside of the convex polytope dened by the inequalities
D b` Y
The rst nontrivial case is . We have the unique solution the corresponding standard tableau is
3 p e
3 p Y ( 5 ! e
Y h
h e
555 ` 7676 h( 3
. We have
to
Let us see how the group acts on the space Tab . We have generated by the transpositions
h r
f D 5 j ( l vD V 5
h e
` b
r h
h e
(
c
|( `
t D e 7676 3 # 5 5 5
r ( lTY
( 3 g
SL
hom .
for which
(2.6)
( s Dr st ( st t C t 4 D r s Df D Dr r h c
25
is
a a `a f a a
f f f f
f r fe
a a `a a a
ab
(
f r f f r f f r fe
ab
(
c
h
r r
f f f f
fe
ab a a a a `a
( 3
c
(
( D b` r
g 3 g g ( 5 3 g 3 4r 3 4r pb
This is the discriminant of the cubic polynomial (2.7)
5 t s
s v
s v
s r s t v r
s t st s t
( v d s D r ( v d s D r ( v d s D r
Similarly, we get
5 t t t s s s c c r ac ac a $a $d5r9X (
( s D t 4 ( c5ra 5r $a 9( c a c ac ac r c a D 4
s t
( t 4 s D ( s D t 4
so that
c ac a c ac r a ( c ac r 5r9 $dp$ a
By the straightening algorithm,
Y ( vw
( t Eg
26
s t
s tzt
s g ( t ph
is spanned by
, where
BIBLIOGRAPHICAL NOTES
27
Bibliographical notes
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 covectors, 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 nd 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 difcult to see that the Pl cker equations dene set theoretiu cally the Grassmann varieties in their Pl cker embedding (see, for example, [40]). u 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.
H ` g g
Exercises
2.2 Let be the omega-operator in the polynomial ring . Prove that (i) for negative integers , (ii) , is a solution of the differential equation (iii) the function in the ring of formal power series .
g ( g 4wCg
D e
e &D o )
D V D e z) e t D e 7676 3 Do ) 5 5 5 ` h B ` h h c a h h e )
D X) `
e )
E cD7P aV c a p 3 3 D 3 T D 3 T
o )
D X)
3 ( ( D D 3 z T g s D s ( 5 5 s vD T 765 g
D T)
5666 55 3 e X) Do
( vg
by the for-
28
(i) Prove that the operators commute with each other and commute with if . (ii) Check the following identity (the Capelli identity):
0 0 e 33 3 e
if if
2.4 Using the Capelli identity show that the operator Pol Pol Pol Pol dened by , where symb symb is welldened and transforms an SL -invariant to an SL -invariant. SL 2.5 Show that Pol Pol is spanned by the catalecticant invariant from Example 1.4 in Chapter 1. SL is generated (as a -algebra) by the discriminant 2.6 Show that Pol Pol invariant from Example 2.1. SL 2.7 Show that Pol Pol 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 the action of on Mat by left multipliO 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 . O Show that Pol Mat is generated by the functions and the bracket functions. SL 2.10 Show that the eld of fractions of the ring Pol Mat is a purely tran. scendental extension of of transcendence degree
DV
DD v " DD vT) W
V V T) " D
3 3 3
QDV
c a e
D ` e l D
3
. . .
. . .
..
. id
. . .
c a (yDV
g
g ( DH
D )
id
id
0 1
5665 5 55 665
e l D
l D X)
DD p vX) l vD V D
D D V
e l D
e 3 e e D ` e V
e 3 e
D V
DD T) 3 W
( &
x(
) ) ) ) )
In this chapter we consider a class of linear group actions on a vector space for which the algebra of invariant polynomials Pol is nitely generated. We start with the case of nite group actions. Theorem 3.1. Let be a nite group of automorphisms of a nitely generated -algebra . Then the subalgebra is nitely 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 coefcients of the monic polynomials such that . Then is a nite -module. Since is noetherian, is also a nite -module. Since is nitely generated over , must be nitely 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 nitely generated by 29
D X
( 1
( d X D
d 1
n V H n c aint'D H pn Y
Clearly, av id
av
By induction we can assume that each invariant homogeneous polynomial of degree can be expressed as a polynomial in s. Since av is homogeneous , we are done. of degree 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 nitely generated over .
Pol . The proof uses the same idea as the one used in the Proof. Let second proof of Theorem 3.1. Instead of the averaging operator av we use the SL omega-operator . Let Pol . Write
for some Pol and Pol 1.1 there exists an integer such that, for any
SL
D X)
The number is called the weight of . Now, for a general matrix , we have the identity of functions on GL
DD ! vT) W
xx x
av
av
5 D
d 1 ( D 5 y1
av
SL
d yt 1
#`B ` xx 3 ( x 3 B 5D w 3 D 3 ( D ( d by1
d 1
R #` 7676 555 3
in . We may assume that each is ho. Then for any homogeneous of degree
( G
DT) W V
1 p " 1
( 0
` i
` i
is
31
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.
Then
Here denotes the omega-operator in the ring sponding to the determinant of the matrix where the formula
g
for any polynomial in the variables . This easily follows from the differentiation rules and we leave its proof to the reader. Now plugging in in (3.2) the left-hand side extends to the whole polynomial (although it is not in GL 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 denition of a rational representation in the next section). We have to use that SL in this case for any GL and Pol .
Proof. This is nothing more than the change of variables in differentiation. Let be a general square matrix of size . We have
D X) ( Hw } D y} DD gD vy} 'r@ ( DD y} # g D C} @ vdH yC} ( c 56656 @ 7766 H 665 V 5 555 55 a D D 5 fw v ( Dv6fw D Dx D D 6w g ( D x vDw D g D 'fw % ( DD x vw 'w g w D ( Dx
D l T
t &
DD x D 5 w 'w v
D D ( Dx 5 x v x
D V
t #w
g ( D 4C
w D
Pol
let
D x D D Dw 'w v
.
( Pw
D T
3 p
DD x Xw w v D ( x Bw
( D D u
g
D V
32
SL Remark 3.2. The proof shows that the algebra of invariants Pol is genof the ideal generated by invariant erated by a nite generating set homogeneous polynomials of positive degree. Let be the subset SL of common zeros of . Let be the ideal in Pol 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 can be expressed as a polynomial in . This implies SL that the ring Pol is integral over the subring generated by . In fact, it can be shown that it coincides with the integral closure of in the eld 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 identied with the null-cone) without explicitly computing the ring of invariants. This gives a constructive approach to nding the algebra of invariants.
3.2
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
for all
St
t ug D x
Since only if
t (D 5 D u 2 D
( D C
2 r
( D g ( D 2 g 9
Let
be the function on
dened by
D 5 x
( D v 2
Pol
. For any
DX
D T
"Bf
( dD X
( 0m
D T
DT (
SL and SU be its subgroup of unitary matrices. Let via its linear representation GL .
D l T
X t
7766 555 3
( lPV X
D qU
Y } 666 555 3
33
condition that e for all and all . Let denote the with zero trace. Since any SL space of complex matrices of size can be written as e for some , we see that the condition
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
. Observe that the same matrices form a basis over of the 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 nd that Pol if and only if for all . Since the properties for all and for all are equivalent we are done.
m t D m$w
where is the Kronecker symbol. This allows one to integrate over it. We consider any polynomial complex valued function on as a restriction of a polynomial function on GL . For each such function set
( D
av
( 3Ew
where d
av
is
-invariant.
DD x 7Xw
Pol
the function
dened by
( D w
D w
w w w D
U h g ( sF h t h s r vww r w
(D DT t 8 H H 5 HIw (w f3
3 p
submanifold of
H HH Iw (w
H(7 H w ( m
The group
SU is a compact smooth manifold. If and , where are real, then is a closed and a bounded dened by the equations
DU
( iw
D qU
D t D qU t
where subspace
t w
D E
D urw
t m
t G
D " Et p ( D ( Ew t U U t D ( D 2Tx
D qU ( (p t p D X t ( n m t bw ( D U D U h i T h d{w
(3.3)
D U
sr
(R$w H w (
we have
is an orthogonal real matrix of size . Thus the jacobian of the change of variables 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
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 complexication of the real Lie algebra of . Here are examples of such groups, their compact subgroups, and their corresponding Lie algebras: GL SU O O Lie Lie Lie Lie
Mat
t t
D H(w H (w H ( H
m
D x
DD x s ww
D 5 w n D x | n
D (w wIw H ( H
H HH
(CDsw vs Xw DD x x
r U" s ( Ehr U U $ 9
s s e
H HH
s ( s xD HH w H w D HH H sx w 9X( ( w
m
. For any
( Dx s
(vx DD ( w n ( w 6w n Dx s
m t D 9w
( r
D D D qU D
(Hw Dxwsw
s w x s(
0r
( m ( h 4 D (m ( h D
U" ow
( bs
ws
35
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 rst replacing with its compact subgroup as above. Taking any linear function with we average it by integration over to nd a nonzero -invariant function with the same property. Then we apply Lemma 3.3 to ensure that is -invariant.
3.3
Next we observe that property (LR) from the preceding section can be stated over any algebraically closed eld . Instead of complex Lie groups, we will be dealing with afne algebraic groups over . Denition. An afne algebraic group over a eld is an afne 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 x some terminology which may be slightly different from the standard textbooks (for example, [102]). We shall use an embeddingfree denition of an afne algebraic variety over an algebraically closed eld . Namely, a set Specm of homomorphisms of a nitely 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 denes a bijection from to a subset of the afne space Specm identied 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 afne algebraic varieties is dened as a map given by composition with a homomorphism of the coordinate algebras . This makes a category of afne algebraic varieties over which is equivalent to the dual of the category of nitely generated domains over . This latter category has direct products dened by the tensor product of
t V
Dg
w m
) ( lh)
V p " 1 1 ut g
h "
R D p
" q
c a V
D y1
D
D
D T)
m
(
" D T@
hf " V
x( D f 5 j p d ) "
E
o
36
-algebras. A subset of of homomorphisms vanishing on an ideal of is called a closed subset. It can be identied with an afne algebraic variety Specm , where rad is the radical of . A point is a closed subset corresponding to the maximal ideal of . Closed subsets dene a topology on , the Zariski topology. Open subsets , form a basis of the topology. Each subset can be identied with an afne algebraic variety Specm . A choice of generators of the -algebra denes an isomorphism from to a closed subset of the afne space . A morphism of afne varieties Specm Specm corresponding to a surjective homomorphism of -algebras denes an isomorphism from Specm to a closed subset of Specm . It is called a closed embedding. The multiplication and the inversion morphisms dening an afne algebraic group can equivalently be given by homomorphisms of -algebras
which are called the comultiplication and the coinverse. of -points of to be the set of For any -algebra we dene the set homomorphisms of -algebras . In particular, if for some afne algebraic variety , the set can be identied naturally with the set of morphisms from to . Here are some examples of afne algebraic groups which we will be using in the book. (a) GL Specm (a general linear group over ): GL where (b)
GL
is equal to the th entry of the inverse of the matrix . GL Specm (the multiplicative group over ):
( D 5 &e
(c)
Specm
( D H o
D En
" D
dgP
D
( D m ( D C pe o m l V Dc a !V (y o o lm l e D m ( D D c 3 aV ( l 3 D
D DD c 3 P v
D m
3 p
H o ( D
D X
D
D m "D
D
DEn
D m U
D ) @
" D
( X
m
U Dc a 4T1
" D1
( D m l
DXIy1 D 1 V V
( l
( l e
( l
1 n "
D
V V
37
Other examples of afne algebraic groups can be realized by taking direct products or by taking a closed subvariety which is an afne algebraic group with respect to the restriction of the multiplication and the inverse morphisms (a closed subgroup). For example, we have (d) (an afne torus over ), (e) SL (a special linear group over ). Afne algebraic groups over form a category. Its morphisms are morphisms of afne algebraic varieties which induce homomorphisms of the corresponding group structures. One can prove that any afne 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 dene 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, GL GL a regular action. In particular, a linear representation will be assumed to be given by regular functions on the afne algebraic variety . Such linear representations are called rational representations. Let an afne algebraic group act on an afne variety Specm . This action can be described in terms of the coaction homomorphism
where is the coordinate ring of . It satises 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
and we set
d 1 " Bf
This denes a rational action of on a -algebra , that is, a morphism Aut . We will continue to denote the subalgebra of invariant elements by An important property of a rational action is the following.
gD 5 4w
( D Gg o
t w
V D vw (g w D
t E
where
. An element
is a homomorphism
(3.4) .
D V
" I f
"D
D y1
( ) D
e l
(
" h
D m
1 D
g 4
(Dg o
" 1 p o
1 t D &Tg
l e
D U w " V
1 t ug
D X
(
D y1
38
It is easy to see, using the axioms of an action, that the maps are the projection operators, i.e., . Denoting the image have and
This denes a grading on . Conversely, given a grading of , we dene 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 satises for and . Such a grading is called a geometric grading and the corresponding action is called a good -action. In this case, the ideal is a maximal ideal of and hence denes 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 nitely 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 dened by the homogeneous ideal of relations between . The vertex of becomes the origin in . We obtain a natural bijection from Projm to the Specm , where acts by set (3.7)
D m1 1 U" 4g yg 1 v " 1
D 1
V ( Ge1
D1 3 p A 1
( R
v667 555 e t
j f D 5 6 y1
j !1 f (
( m1 e
g 5 4
5 m1
g ( D 4vg
( D 1 o
( p1
( D g o 1 ut g 1c a V ( D 3 e hv1 ( 4G
act on an afne algebraic variety Specm . Let be the corresponding coaction homomorwe can write (3.5)
by
D y1
"
Aut
arises from a
4g
Proof. This follows immediately from equation (3.4). The set of elements spanning set.
D y1
o
R @
4g
1 t ug
( o
1u t
g vg o ( D
1 B 1 q'1
t D w g w j v7676 f 555 V
is a
" 1 G o
we
(3.6)
39
is called the weighted projective space with weights . When all the are equal to 1, we obtain the usual denition of the -dimensional projective space . be the closed subgroup of Specm dened by the Let 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 action. It follows from the denition that
The inclusion denes a natural map Specm which coincides with the quotient map for the action of that for any ). Let act on Specm grading dened by Then Projm
Specm Specm
Specm Projm
It is known that for any nitely generated geometrically graded -algebra there exists a number such that is generated by elements of degree 1 with respect to the grading dened 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 dening 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 dening 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
D y1 o D 1 o
Dy1
j P f ( 1 U
5 D 1 V D o vD vy1 o
D 1 " e Dc 3 V a ( e U 7676 7 555 f 3 ( D o V4q9j p A y1 p 1 t V 1 o " D 1 o c8! 666 p!V 555 a Dy1 5 1
( o 4D 1 o V VD ( o 4vy1 o
1 5 p( 1
1 ( ( 1
D V 8
D y1
Dy1 o
v7676 555 1 V 1
),
( 555 D 676 1 t ( ( D 1 D 1
B
c 7667 ! V ( 1 555 a 1 U D !
1 V DB
1 t
t dB
D V
40 subsets on of
Let . 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 dene 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 dened 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 dene 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 dened 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) denes a closed embedding Specm Specm whose restriction to any subset is a closed embedding of afne varieties. It corresponds to the homomorphism . This denes a closed embedding from to and a morphism Projm Projm . In particular, a choice of homogeneous genof denes a morphism Projm erators of degrees 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 afne algebraic variety is isomorphic to a quasiprojective algebraic variety because the afne space is isomorphic to an open of Projm whose complement is the closed subsubset set dened by the ideal . Thus any locally closed subset of an afne variety
D D D t
D 1
" 1
V v " %
@ B &%
Dc a T1
B
D
% B D
D D 555 a c 7667 ! V
1 D 1
D 1
D ! o 4v! G( VD D D 1
B D CEn
@ I " DEn V %
Dy1
cD a in
D y%
D 1
D
1
B
e @
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 afne algebraic varieties we have to deal with nonafne quasi-projective algebraic varieties. be a rational action of an afne algebraic Example 3.2. Let group on an afne algebraic variety . For any point , we have a regular map dened by . The ber of this map over the point is a closed subgroup of , called the stabilizer subgroup of . It is an afne 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 nite 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 afne variety is not always afne. Example 3.3. Let be a closed subgroup of an algebraic group . Consider the space spanned by the -translates of generators of the ideal dening . By Lemma 3.4 is nite-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 xes Gr . Thus we can identify the quasi-projective algebraic variety 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 afne variety Specm . Then is a nitely generated -algebra. Let us rst explain the notion of a geometrically reductive group.
Duf U Q ) G( g
t
f
g
D D ( D v w w d 1
"
y2'
w
D1
Df Duf
" C )
w
B D Chw
( BU U
42
Denition. A linear algebraic group is called linearly reductive if for any raGL and any nonzero invariant vector there tional representation 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 dened over a eld of characteristic . In fact, even a nite 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. Denition. 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 dene the notion of a reductive algebraic group over any eld 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 denition of a reductive afne algebraic group (over an algebraically closed eld) 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 classication of semisimple afne algebraic groups. Examples of simple groups are the classical groups
"
There are also some simple groups of exceptional type of types . Every simple algebraic group is isogeneous to one of these groups (i.e., there exists a surjective homomorphism from one to another with a nite kernel). We shall start the proof of Nagatas Theorem with the following.
SL
type
SO
type
Sp
type
SO
5D
type
x( D
x( D
D n
D T)
DT)
3 p
"
" RC
R
D 1
3 p
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 nite-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
w ot B
and
. Let
for some . This implies that , and, in 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 afne space by using the dual basis, so that . By denition of geometrical reductiveness, we can nd 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 . Since each generator of belongs to , we see that modulo . Since (because is -invariant), we are done. Now we are ready to nish the proof of Nagatas Theorem. To begin, by noetherian induction, we may assume that for any nontrivial -invariant ideal the algebra is nitely generated. Assume rst 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 is an integral domain. Take a homogeneous element grading. Suppose of positive degree. We have since, for any ,
( w t D 1 d 1 t d 1
d 1 t 'D 667 f D 666 G W g 555 3 5 5 5 3 B w X 1 X 666 555 g } } 555 3 ( 555 3 W CD 666 GD 7766 D W3 G 9 xx x W 3 } 2 ( x( D z5667z Y D } 7665 7 } 55 5 5 3 D z666 555 ( A o ) 0 w t g i ( 3 ) D 7766 555 3 o ) o ) V ! " ) HH mH iv w ( D ( DD w t HH H gDD gD ( D wD ( gD w ( D w ( H CD w g D u C w
) t B ( DD H w ( D g (w C w
U"
d 1
HH
d 1 QX$ d t W g 1 d e$ d 1 1 Q 1 1 V g
d v( d p 1 1 Q 1
R oY
1 ( 1
g ( D Hw w t 2Eg Ifw 4g g g ( D 5 4 wg o Fg
Y d $y1 g D t d 1
d $y1 D
V t
d D$y1
V ( he1
44
. Since is nitely generated and (Lemma 3.5), we obtain that is integral over nitely generated. Hence its maximal ideal generated by elements of positive degree is nitely 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 nitely generated algebra. Now assume that contains a zero-divisor . Then and the annihilator ideal are nonzero -invariant ideals. As above, and are nitely generated. Let be the subring of generated by representatives of generators of both algebras. It is mapped surjectively to and . Let be representatives in of generators of as a -module. Since for all , we get , i.e., . Let us show that . Then we will be done. If , we can nd such that (since is mapped surjectively to ). Then is -invariant implies that . Thus . This implies as we wanted. So we are done in the graded case. be generators of . ConNow let us consider the general case. Let sider the -vector space spanned by -translates of the . It follows from Lemma 3.4 that is nite-dimensional. Without loss of generality we may assume now that is a basis of this space. Let be the surjective homomorphism dened by . The group acts on linearly by , where . 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 nitely 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 integral closure of in the eld of fractions of is a nitely generated -algebra provided that is a nite extension of the eld of fractions of . Since is integral over this would imply that is nitely generated (see [26], Exercise 4.3.2). Thus it is enough to show that the eld is a nite extension of the eld of fractions of . Since is integral over this ring, it is enough to show that is nitely generated as a eld. If is a domain this is obvious (a subeld of a nitely generated eld is nitely generated). In the general case we use the total ring of fractions of ,
D d 1 d 1 1
c ( ! 666 6! ie! t ( t Eg 555 3 a n t dD Tt n v( t Cg 1 t d 11p4 d 1 Q n 1 pvt t g d 1vg nvt t t c ! 666 6! i( d 1 555 3 a n d 1 t z! D D t 2w ( zhz w d D 1 n Q En 1 tizBeD w d Q d 1 d Q p4 d 1 1 1 1 d 1
d p4 d 1 1 1
d q9 d r r Q
7667 555 3
1 p
D d 1
d yp4y1 D 1
d 1
d b4 d p( d 4 d 1 1 Q 1 1 1 d t 1 ( mD D w 1 B )
d X$ d r r Q
D 666 555 3
d r$ d r Q q w ( D
d Q $ d r r
D 1
d 1
BIBLIOGRAPHICAL NOTES
45
the localization with respect to the set of nonzerodivisors. For any maximal we have since is a domain. This shows that the ideal of eld of fractions of is a subeld of . But the latter is a nitely generated eld equal to the eld of fractions of . The proof is now complete. 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 nitely generated. The algebra of invariants , where is a reductive algebraic group and is a nitely 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 nitely generated -algebra . Then is a normal nitely generated algebra. Proof. Recall that a normal ring is a domain integrally closed in its eld of fractions. Let be the eld of fractions of . It is clear that the eld of fractions of is contained in the eld of -invariant elements of . We have to check that the ring is integrally closed in . Suppose satises a monic equation with coefcients is veried. . Since is normal, and the assertion
m
Bibliographical notes
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 rst courses in algebraic groups.
d 4g d 1( d Eut m Q 1 1 1 t ( xx 3 'g x 3 g
St
d 1
d 1
1 Q bpy1 1 D X) 1
1 Q ( d
m
"
d 1
d 1
d 1
d T) D
d 1
46
We refer to [89], 3.9 for a survey of results in the spirit of Proposition 3.1. , where is a rational linAn interesting question is when the algebra Pol ear representation of a reductive group , is isomorphic to a polynomial algebra. When is a nite 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 reections. The classication of such unitary representations is due to Shephard and Todd ([105]). The classication 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 nite group construct an afne algebraic -group such that its group of -points is equal to for any . 3.2. Prove that any afne algebraic group is a nonsingular algebraic variety. to , or in the 3.3 Show that there are no nontrivial homomorphisms from other direction. 3.4 Prove that a nite group over a eld 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 afne algebraic group on an afne space. is isomorphic to Projm , where 3.6 Prove that any closed subset of Projm is a homogeneous ideal of . via its linear representation in . A polynomial 3.7 Let GL act on Pol Pol is called a projective invariant of weight if, for any and any , . Let Pol be the space of projective invariants of weight . Show that the graded ring Pol
is nitely generated.
&t w
D $1
dA DT
d DX)
V 4
dA D T
D 1
D w D
DT
( Dx H#w
D )
t D T
The assertions about nite 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. be its purely transcendental Problem 1. Let be a eld, and let extension, and let be a eld extension contained in . Is the algebra nitely generated? Hilbert himself gave an afrmative answer to this question in the situation SL when where SL acts linearly on (Theorem 3.2 from Chapter 3). The subalgebra is of course the subalgebra SL of invariant polynomials . A special case of his problem asks whether the same is true for an arbitrary group acting linearly on the ring of polynomials. A rst 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 nd a normal irreducible algebraic For any subeld variety over with eld of rational functions isomorphic to . The inclusion of the elds gives rise to a rational map
m
47
Let subset of
B }
be the closure of the graph of the regular map of the largest open on which is dened. Let be the hyperplane at innity in and
c 7766 V 555 3 a
D 666 V 555 3
D V
l c 667 V 555 3 a c 75676 3a!Vim Q 55 l D 766 m 555 3 V ( " 5 # 4u D 667 B 555 3 V
m
D 7667 V 555 3
48
pr pr . This is a closed subset of . By blowing up, if necessary, is the union of codimension 1 irreducible subvarieties . we may assume that Let be the Weil divisor on equal to the sum of components such that pr pr ; note that could be the zero divisor. Thus for any rational function is regular on if and only if has poles only along the irreducible components of . Let be the linear subspace of which consists of rational functions such that div . After identifying with and with (by means of ), we see that is isomorphic to the subalgebra
of . So the problem is reduced to the problem of nite 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 and hence can be given locally by an equation for some rational function on regular on some open subset . These functions must satisfy on for some . We can take them to be the transition functions of a line bundle . Rational functions with poles along must satisfy for some . This implies that the functions 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 eld . Let
Recall that we can view as the space of regular functions on the line bundle whose restrictions to bers 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 innity in each ber 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 bers; here the direct sum means that the transition functions of the vector bundle are chosen to be diagonal matrices
D
o p
3
c 7766 V 555 3 a D
D 3
Di ` g
D y% oU oD ) TuQ % ot w B % (
5 D
D i `
3 ' !
e ( D #
Dugm
( D i
D i o D
o
3
D# D V
D V
wp(
3
D V
g
49
It can be shown that Nagatas counterexample to the Hilbert problem is of the (see Exercise 4.3). It turns out that the algebras are often not form nitely generated. However, if we impose certain conditions on (for example, that the complete linear system dened by has no base points) then is nitely generated. One of the fundamental questions in algebraic geometry is , where is the canonical the question of nite generatedness of the ring divisor of . This is closely related to the theory of minimal models of algebraic varieties (see [69]).
4.2
Let us rst 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 is not nitely generated for some nonreductive group . Notice that that according to a result of V. Popov ([87]), if is not nitely generated for some action of on an afne algebraic variety with , then is not reductive. In fact, the proof of this result relies on Nagatas counterexample. Since any afne 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]).
that
Proof. Let
Specm
be the afne algebraic variety with . Let . Assume . This means for any . Let . Then
p
1 ( D
Here
acts on
dD 1 D 5 y p (
D i o
Problem 2. (O. Zariski) Let be a nonsingular algebraic variety and let an effective divisor on . When is the algebra nitely generated?
D i o
1 ( D
Then we obtain that is equal to the ring innity in . In this way we are led to the following.
D r
D i o
D i o
"
d 1
d 1
D i o
where
is the divisor at
D # o
be
d 1
50
Thus
Corollary 4.1. Assume that a rational action of on an afne variety extends to an action of a geometrically reductive group containing and also assume that is nitely generated. Then is nitely generated. can be interpreted as the algebra of regular functions on The algebra the quasi-projective algebraic variety (see Example 3.3). It could be afne, 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 afne algebraic group is called observable if is quasi-afne (i.e., isomorphic to an open subvariety of an afne variety). An observable subgroup is called a Grosshans subgroup if is nitely generated. Theorem 4.1. Let be an observable subgroup of a connected afne algebraic group . The following properties are equivalent:
Proof. (i) (ii) Let and let Specm . is an irreducible algebraic variety on which acts (via the action of on ). Consider the canonical morphism such that is is isomorphic to an open subset of an afne variety , the identity. Since the restriction map denes a morphism of afne
xp )
vkg (
(ii) there exist a rational linear representation of dimension and a vector such that of codimension in its closure .
x v
) t
(i)
5 D v w vD w ( x g g
( D x D Xw x $ # g g g
p
1 St ( B D
g
( x # D w %
d 1 p X u D D t ( D x H w 3 H fw % w # g
p
D X
1 t g g
. Conversely, if
, the function
D Tw x
( D v w
D
p
D X
51
varieties such that the composition is the open embedding . Since is dominant, this easily implies 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 afne variety, i.e., the ring is normal. By Proposition 3.1 the ring has the same property and hence is a normal afne 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 nd 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 afne 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 . (i) Let be the closure of the orbit O O . Replacing by its (ii) normalization, we may assume that O is isomorphic to an open subset of a normal afne 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, . In the natural representation of in the afne 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 afne algebraic group is a Grosshans subgroup (see [41], Thm. 5.6). Let . We know that is not geometrically reductive (Exercise 4.1). However, we have the following classical result. is nitely
Proof. To simplify the proof let us assume that . We shall also identify with its image in GL ; which is isomorphic to . This can be done since
"
Theorem 4.2. (Weitzenb cks Theorem) Assume char o . Let GL be a rational linear representation. Then the algebra Pol generated.
" D
"
) 0
g D z (
D)
"
DX Dg g
g
( D V
Dug
g
" D ( V (
r
( g
g (
"~
1 DX
g X
g $X A
(
"
g $X
( r
Dg
y
j T A f
p} (
D X)
D y%
52
does not contain nite nontrivial subgroups in characteristic zero so is either be a nonzero element. Since there are no nontrivial or injective. Let trivial 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 all in this space. Then for all . Continuing in this way, we nd a basis of such that each is represented by a unipotent matrix . Consider the differential of the homomorphism GL at the origin. It is dened by , where . Clearly is a nilpotent matrix. Since , it is easy to see that and hence . By changing basis of , we may assume that is a Jordan matrix. Let , where corresponds to a Jordan block of of size . It is easy to see that the representation of in dened by is isomorphic to the representation of in Pol obtained by restriction of the natural representation of SL in Pol . Here we consider 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
or
Thus any -invariant regular function on SL is uniquely determined by its values on such matrices. Since the set of such matrices forms a subvariety of SL isomorphic to , the restriction of functions denes an isomorphism SL
Since , we conclude that SL is nitely generated. So SL and the representation of SL on we can apply Lemma 4.1 to the pair Pol to obtain the assertion of the theorem.
4.3
Nagatas counterexample
Now we are ready to present Nagatas counterexample to the Fourteenth Hilbert Problem.
% t
SL
DV
D V D V D V D V DV ) ) n ) D 1Xnp(6 1 HD 12 n D u W 4W 12 In ( 3 "R
t hw
( 5 D '
t pw
D D D f 5 $j y A ( D
Vty g w ( D 4 V V
o V
D x(
t pw )
D A
D V D V
Y 3 m
j yA f
3 D V p h) ( D f ( $j A
(Y
D 1
t pw
( w
V )
to a
Both of these groups are identied naturally with subgroups of SL and we enlarge by considering the group . The group is contained in the subgroup of matrices of the form:
D 7676 555 3
Theorem 4.3. For an appropriate choice of the system of linear equations (4.1) and the number the algebra of invariants
. . . . . .
. . .
555 555 55 6766675 555 555 55 6766675 567666766675 55 555 555 555 55 555 555 555 555 55 67666766675
" x H ( s
( &pz
3p e u
t wD
7766 6 555 3
( G"
3 p
5 r &(
( 4g
r
H
of
. It acts on
) ) ) ) ) ) ) ) )
. . .
(4.2) (4.1) of a 53
H
The rst equality shows that are algebraically independent over , hence are algebraically independent. Let be the subgroup of dened by the conditions Obviously it is isomorphic to . We see that
the right-hand side is contained in the left-hand side. Using the assumption on the coefcients , we can write as a linear combination of to obtain
fg 3 D
Lemma 4.2. Assume that the determinant of the matrix to zero. Then
Now we throw in the torus part which acts on by multiplying it by . It with coefcients in is clear that any -invariant rational function in must be equal to a constant. This proves the lemma.
"
D } } '} V 3
5 D3 D 3
p
( xx x ( B A
} } } 3 } '} V 3 } } '} V 3
@ r 2( P
} ( D ( } o w ( D e o w
@ @ ( w o w
3 pv ( m 4g
Proof. Under the action of , dened by the matrix (4.2) from above, we have
and, since
Moreover,
where
54
, we obtain that
V V
} } } 3
55
Consider now each column of the matrix as the homogein the projective plane . Let be the ideal neous coordinates of a point 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 variables we may assume that . Then has multiplicity at if considered as a polynomial in all its nonzero coefcients are homogeneous polynomials in of degree . Lemma 4.3.
Write any invariant homogeneous polynomial as a sum of monomials , where and . Since each is homogeneous in of degree 1 and in of degree , and is homogeneous of degree in , 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 coefcients are polynomials in . Since the degree of in is equal to , we obtain that each is the -homogeneous component of , and hence is a polynomial in . It remains to show that if and only if each . Assume that none of is equal to zero. After a linear change of if and only if its coefcients as a polynomial variables, we obtain that in are homogeneous polynomials in
} } g 7 v( I } 3 g 3 } g ( } g H 7y7 vzI e D b` t g D b` t e c @ st e D } } '} e a V 3 @ e dD } 3 e } } e dD } } '} e 3 3 @ e } } '} ` h( 3 3 e W dD } } } e e 3 Yp(iE4D U ` 3 x uiE4D 3 x iU D 3 U ( U ` D @ U U @ e } } 3 } C} G t H` 3 dW @ c a V t 3 a V c a 5 c 3 } } }p( d q@ V c } } } 3 V 3 a D } } '} V 3 3 a V ( E E 5 5 E3 5 c 3 F@ 6566 3 @ 676 y } } }ppc 3 F@ 7667 3 F@ 7766 i V 555 5 5 5 E3 555 3 a r @ } ( CG!H ( De } } } {!c @ ( d @ V 3 V Q a V c a e e DD 3 e ( dc a 5 0b` t f e D } } '} @ !V C
. First
, we have
is equal to
`i `
` G
D b`
D 4g
( D V D g g g 3 ` } D ( u y
} '} 3
c } } }!V 3 a
We leave to the reader to prove the converse. Next, we need a lemma from algebraic geometry. Let be an irreducible plane cubic curve in the projective plane over an algebraically closed eld . 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 afne line, and hence has a structure of an algebraic group isomorphic to the and additive group . The complement of two points is isomorphic to has a structure of an afne 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 the subset of dened by the equation the formula
"
Each irreducible plane cubic curve has at least one nonsingular inection 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 is the unique nonsingular inection law. In the example (4.3), the point point. We denote the sum of two points with respect to the group law by . Lemma 4.4. Let be an irreducible plane cubic curve with a nonsingular inection 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 .
P Q"
"
j q 3 A f
3
3 p S
3
"
5 D D H H D H @ H D @ ( 9 9 @ ( D zz ( 3 {p
"
c a V 5 q@ t e V F"
"tT D zu "
r x (
D b` t I zI
" t
and
in
c @ V a
, we see that,
( 3 D ! "
D V
S 5 5 5 T666 3 R
R S`
"
"
" hF "
"
` r
"
P Q"
"
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 implies that is the unique point such that there exists a rational function on whose divisor is equal to . Conversely, suppose such exists. Let . By the above there exists a rational function such that . But then . This implies that (otherwise the rational map from to dened by the function is an isomorphism).
B y c c @ !V aa " q @ ( 9 " q ` g " e 9 D gx r ` pr 9G g gx ` r DS xx ( T x 3 ` p D z!9 p9v 3 H j XpX3" ( f D zz ( m! R `` S xx deD T x 3 ` ( D e x 3 XV ` xx R `` dcyD ( D S xx b x 3 ` WV e e " R ( | e ` ( a{U RU ` iBeD x 3 ` xx xx ` x 3 " 3 H ( ( D w 6H C9! W V w RD Y U xx ( D 4H x 3 Cw V xx ( x 3 CH X " R 4DU x xx 3 ! D WV 765 3 55 RU e "
In particular, we obtain that is an -torsion element if and only if is the divisor of a rational function. Let us now take . Assume that there exists a polynomial as in the statement of the lemma. Let be the equation of the inection tangent at the point . Then the restriction of the rational function on to the curve denes 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 afne 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 the curve cuts out the divisor . By B zouts Theorem, e the degree of is equal to . Note that is not dened uniquely since we can always add to it a polynomial of the form , where is a homogeneous polynomial of degree . The rational function cuts out the 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 be its maximal ideal. Since was assumed to be nonsingular, one can nd a system of generators of 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
"
"
D 4
58
be the Taylor expansion of the rational function , where is a homogeneous form of degree in . We denote by the th Taylor polynomial . 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 . 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
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 nd that
Thus it sufces to show that the kernel of the map is one-dimensional. An element which has in the kernel denes a homogeneous polynomial of degree 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 at is equal to . Thus has multiplicity at each point .
V t
r ` !r
of degree
e g 9 w fQ g Q f
"
D X@ w
"
r ` lr
Fw
g P ( ec # E3 f aw D v` D b&666 3 $ S #555 # g 3 e ( g
c aw
e 9X
@ V a
D T@ w g 3 p
" e c ! f 3 a V
@
( Hw
D 4 `
of
is equal to
e 9vqP D g
e #g D '667 3 S 555
is equal to
. Let
59
be the equation of the curve cutting out the divisor . Let be the equation of . For any , the polynomial denes a curve which cuts out the same divisor on . When is equal to the order of the point , the pencil of curves is called the Halphen pencil of index (see [15], 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. be nine distinct nonsingular points on an irreducible Lemma 4.5. Let plane cubic . Assume that their sum in the group law is not a torsion element. which has multiplicity (i) A homogeneous polynomial of degree at each point is divisible by . (ii) The dimension of the space of homogeneous polynomials of degree which have multiplicity at each is equal to . Proof. Assume is not divisible by . By B zouts Theorem, e . Now this contradicts Lemma 4.4, so we may write for some homogeneous polynomial of degree . Clearly, the multiplicity of at each is equal to . Applying the lemma again, we nd that the sum of the in the group law is a torsion element unless divides . Continuing in this way we nd that divides . This proves the rst assertion. Let us prove the second one. We may assume that all the points lie in the afne part . Consider the linear functions on the space of homogeneous polynomials of degree which assign to a polynomial the partial derivatives of order of the dehomogenized polynomial at the point , . Obviously, is the space of common zeros of the functions . To check assertion (ii) it sufces 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, where is the curve ). Since sists of polynomials proportional to , the restriction of the functions to is a linearly independent 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 coordinates of the points which lie in the nonsingular part of an irreducible plane
S xx T x
e `
Vt
"
H `r ( 4h h i ` F p e D U F W D 3
D g g g 3
HW
HW )
H ( s2
e (I e c ! 3 a V t
"
`r h
D q
H
W r)
W 9
e
r ` &r
S 55 5 T7766 3 S 5 5 5 T667 3
( s!k"
( F p e D p F e D ` 3 e W
` ( qU
x( p
D q ` " e ( e (
D xx x 3 ` e b S S xx b x 3 ` Y F p e D 7766 555 3 `r Y ` `
60
"
cubic and which do not add up to an -torsion point for any . Also, to satisfy Lemma 4.2, we assume that the rst three points do not lie on a line. This and is a cuspidal cubic. Assume can always be arranged unless char that is nitely generated. By Lemma 4.3, we can nd 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 polynomials of degree which have multiplicity at each is equal to . On the other hand the dimension of the subspace of polynomials in which vanish on is equal to . Thus there exists a polynomial which does not vanish on the curve . Let us show that cannot be expressed 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 generators . Then Thus
Since does not vanish on , we may assume that if (in this case denes ). 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 eld of zero characteristic, and the points with the rst three points not on a line, then the conditions on will always be satised unless . In fact, the group law on has no nonzero torsion points. Remark 4.3. If we restrict the action only to the group (not including the torus), the algebra of invariants is also not nitely 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 on a polynomial algebra for which the algebra of invariants is not nitely generated. Recall that by . Examples with and were given recently Weitzenb cks Theorem, o by S. Mukai ([70]).
r
`r HU U
"
( 3 4g pS
3 p e ) t
e 9
` ( Hr qU
3 ( {p
3 p e ) `r( F p e D q F e D ` ` r 3 D V
Y Y ( ` 5 H` ( ` 9U e r } } } 3 } } '} 9YU 3
s % x x W 3 % 4W 8 x e 9
x ( Y ( qY
H
r (
( YD `r 5 wHU
"
S 5 5 Tu86765 3
D ( g g !
D V
` 3 p e ) H` ` H` 5 5 6567 &( e 98 d q@ V c a R
"
`r 4(
"
"
Y ` 9H e 98 8 e 8
` e D r
e 98
eD
!%
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. Lemma 4.6. For any homogeneous ideal smallest positive integer such that chosen to be such that for all number there exists a natural number such that
Proof. Let be the space of homogeneous polynomials 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 see that . In view of our assumption we must have . Since again by assumption we see that for sufciently large ,
Lemma 4.7. The assumptions of the previous lemma are satised when where and the coordinates of the points generate a eld of sufciently 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 nitely many polynomials . We can write them in the form as in Lemma 4.3. Let . By Lemma 4.6, we can nd for sufciently large such that . Obviously cannot be expressed as a 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 .
V s ( U s D D f ` 8 b` f ` h D f 5 8 Db` t vb` t f i U '` t h ( DD h h f U S$vb` R ` DD 2 e i h U ( D `DD u qb$vb` h 2 U 'b$vb` i e u h D `DD `D b i h ` UH4v Y Y D D D b` Y D b` Q W 3 a V ( W D c } } }!qb` D f` D f ` x ( 8 b` U ` R DD Bb` t R ` hY U x( W 3 a V Q 5 j f2c } } }!X c } } '} p 3 a V B
let is strictly larger than .
#`
( D X@
D
"
D 7766 3 555
h i
Df |
8 d ( t e e e y ( 5 #` 3 p
` x3U
vU s (
u7667 3 5 5 5
8 D | j t e f e f h i 0w 5v U
d @ V c a
62
Bibliographical notes
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 Weitzenb ck Theorem can be found o 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 nitely 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.
x( D dV
Exercises
4.2 Let be divisors on a nonsingular variety
r
(i) Show that the algebra is isomorphic to the algebra for some divisor on some projective bundle over . (ii) Let . Show that is a nitely generated semigroup if is nitely generated. (iii) Let be a nonsigular projective curve of genus , let be two points such that the divisor class of is not a torsion element in the group of divisor classes on . Prove that is not nitely generated. 4.3 Show that the algebra constructed in Nagatas counterexample is isomorphic where is the inverse image of a line under the blow-up to the algebra of points in the projective plane and is the exceptional divisor.
D i o
t t z g
j j
5 D l p$ l y l l
D Xe x o
EXERCISES
63
4.4 Prove that the algebra is nitely generated if there exists a positive such that the complete linear system dened by the line bundle number has no base points. 4.5 Show that the algebra of regular functions on the coset space is isomorphic to the subalgebra where acts on by left multiplication. 4.6 Let be a closed reductive subgroup of an afne algebraic group which acts on by left translations. Show that the homogeneous space is afne and hence is nitely generated. 4.7 Write explicitly the group law on the set of nonsingular points of a nodal cubic over a eld of characteristic different from 2. 4.8 Show that the conjecture of Nagata is not true without the assumption .
8
` q U
g
g
D i o
D
D X
Examples of covariants
% D 1
SL act on an afne algebraic variety Specm . Let be its Let subgroup of upper triangular unipotent matrices. In this chapter we shall give a 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 . Dene a function on by (5.1) implies and hence for some , we have
p
Since
65
) u
Dx x ( D 5 Xw X w Xx w
Therefore
( xHw Hw $H7(dD 3 (w w
This shows that this denition does not depend on the choice of is well-dened. Also, for any SL we have function and hence
on
D ( x ( D w ( D 5 x pD Tw vD x H w x 3 H fw x H w X x H w
w # C(w ( H
Dhw
t
x ( x # ( 5 D Xw D Tw % D x H w
( $E7x w
D )
( D 9X9 3 wIw H x ( D 5 D Xw Xx
t H r(w
D X)
t w
9x 3 (sP$x 3 w H w (
D X)
(
j y w f ) j m!# f ) t ) $
g t #
66
It is clear that is a polynomial function in the rst argument. Moreover, if is homogeneous of degree , then is homogeneous of degree in the rst variable. Let us see that is also polynomial in the second argument. Choose coordinates to assume that . Let . Assume . Let
. . .
. . .
. . .
..
belongs to SL and . Thus and 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 . Hence it is regular on the whole of and so is a polynomial function. on 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:
D hw
Pol
Pol
SL
Pol
Denition. A covariant of degree and order on the space Pol is an eleSL ment of the space Pol Pol Pol . We shall denote this space by Cov . The geometric meaning of a covariant Cov It can be considered as a polynomial map of afne spaces Pol Pol
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.
D X) W
D X)
D Y g e X) D
D X) g
DD ! v) g
t D g
Pol
D ) W
D w
D )
" D X) W
Pol
SL
is very simple.
DD 5 p v) g
DD 5 vT)
( x |X9G1
D) (
DD IvT)
D w
D hw
D w e
D 9 x U"
D Y g e T) D
j o) f D x 3 1 1
Clearly,
( D e Xx
j Pt D v666 f ) 555 ( 5
0 1 0 0 0 0
x( ( $|u 3 1
D z666 ( $ 555
) ) ) ) )
x( D T)
( p1
x(
67
On can easily dene the symbolic expression of covariants. By polarizing, SL an element of Cov Pol Pol Pol becomes an SL invariant polynomial function on the space of matrices Mat which is homogeneous of degree in each column different from the last one, and is homogeneous of degree in the last column. Observe that each of the rst columns corresponds 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 rst columns and dot-products of the last column with one of the rst 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
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 Pol . Let
5 U 666 ( z g 555 t
DT) W
"DT) W `
is a covariant of degree
and of order .
c #`a
D T)
D 5 )
3p e DD ! v) g
t iw
5 zq Y ` Qw ( 5v
c a #`t zs
D D ( Dx gx g A w Xw w
3 pv
gu
DD X)
D 55 3 665
r s r (
t 555 D 7667
D 7667 555
3 p
e ( D Y g
Y 2(
D v7676 555
where
in the coefcients
g W ( g F W D
In coordinates:
68
is a covariant of degree
and order
where is the omega-operator. The last subscript means that we have to replace each unknown with . The map is a covariant of degree and order . For example, 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. The result must be an invariant of degree 4; let us compute it. If 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.
r gg g 3 4 g g g zg r g g3 g 5 D g 3 g4 f4 DD gy gg g g 3 g3 g g D 3 zg D zg v r D D D 3 v3 g 3g g 3 g v3 g g r g 3 g 3 g g 3 3 g g 3 3g 3 g 3 g 3 r g
D 555 3 v666
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 . Dene the th transvectant as
considered as as a
D 4 TY
DD ( D 5 W D 665 3 # D '665 Y Y 5 5 5 5
D 5
. . .
..
. . .
0 1
7 665 55
55 665 ( D X
" p
)
( D X 3
( D X D V D
g 4(
( D v
( D X
( ( ( DD v
3 r vB 3 g g (
The Hessian of
(5.2)
. Its
69 Pol Pol
5.2
Covariants of an action
D w " D y1
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 afne algebraic group on an afne variety Specm . Let GL be a linear representation of in a nite-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 afne space. Equivalently, it is a -equivariant homomorphism of algebras Pol . Since any such homomorphism is determined by the images of the unknowns, it is dened by a linear map . Let Hom be the set of such maps. The group acts 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 . Another special case is when contravariant of order on the space Pol Pol Pol and Pol . In this case a covariant is called a concomitant of order . A concomitant of order is called a combinant. For example, the resultant of homogeneous polynomials is a combinant. be the set of covariants with values in a Let Hom module . It has an obvious structure of an -module. It is called the module of
D) l
Do ) l
D D ( x 5 vD x 3 w w D 9Tw " R
D T) ( g DX) W
A ( D T) W
" o
"
w
5 D w D
D 555 3 v@ 7676
d 1
D g
w ( o D
D)
d 1 D
T) D
DX) W
qg
x Tw
" Dhw
D ( d y1 o
(
x x x D ET)
w 1 1 ( D
(
zW
" B
Then
Hess
D 4qY
5
.
0 1 0 0
3 r 3 r
r r
) ) )
D V W
t D V W
dene their
( D
X
DD v
( 1
70
w
has a natural structure of a -algebra. It is called the algebra of covariants. Applying Nagatas Theorem we obtain
Corollary 5.1. Suppose is a geometrically reductive algebraic group acting rationally on Specm . Then the module of covariants Hom is nitely generated. Proof. The algebra Cov is a graded nitely generated -algebra. We identify with the subalgebra of covariants Cov , where is the trivial -module. Obviously Cov is a nitely generated -algebra. We may assume that it is generated by a nite set of homogeneous elements of positive degrees . Thus there is a surjective homomorphism of graded -algebras Cov , where . Since each is a nite free -module, its image Cov is a nitely generated -module; hence Hom Cov is nitely 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 nite-dimensional linear space is completely reducible in the following sense.
H )
Proof. Without loss of generality, we may assume that and is an irredoes not contain any nontrivial proper submodules. ducible submodule, i.e., Consider the natural map of -modules Hom Hom . By
D ) ) l
j x ( ) f
"PX) l D
h
of
d D1 hw o
666 555 3
3D w 1 e hw 1 2 2 D (e h t
d 1
D V 1 2
Dhw 1 2 D hw 1 2
D1
D hw 1 2
dD 1 D o
Pol
d D1D o w e
Cov
Hom Pol
Pol
Do
w
covariants with values in . If char and Pol so that the direct sum
dD D y1 hw e
( D V
( D w 1 2
(
o hw D d 1
) ( p
71
Schurs Lemma, the subspace Hom is one-dimensional. Its inverse is a submodule of Hom , and the restriction of image to is a nonzero linear -invariant function. By denition 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 nite rank, is a nitely generated -module. Let be its spanning set. For any we can write (5.3)
be the projection operator (called the Reynolds operator). It has the property
Let GL be a nite-dimensional linear rational representation of a linearly reductive group. By Theorem 5.2, can be decomposed into a direct sum of irreducible representations . When is nite, there are only nitely many irreducible representations (up to isomorphism); in general has innitely many nonisomorphic irreducible representations. Let be a decomposition of into a direct sum of irreducible representations. We have an isomorphism of -modules:
w w
Hom
Irr
` Dg
3 w p
d 1 "!h ` U g
dD w
uw
" R1 b3
D hw
d
( {V
"Iw 3 #` d 1 p d
the averaging operator over the compact form be the map dened by . is equal to the image under of the nitely generated -module and hence it is nitely generated.
we take for
d 1
d 1 t 1 t 5 eut ! u#g
for some . Since is linearly reductive the complementary invariant submodule, i.e.,
-submodule . Let
of
has a
(5.4)
x( H pD
d {t ` 666 ` 555 3 1 1 uw
D ) l d T) ) l D ) d v1 1 (
w dt
xx 3 `3 g ( ` g x `
d y1 " 1
D ( g t D t g
H
D 3 ( 1
1 4g t
) hf "
w 3(
d ` t
72
where Irr is the set of isomorphism classes of nite-dimensional irreducible -modules, is a representative of the class , and acts trivially on the space of linear maps Hom . This isomorphism is dened by the map
w t w
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 nite-dimensional -invariant subspace of generated by its -translates (see Lemma 3.4). This allows us to apply (5.4) to the -module . We have
w dD 1 w
Hom
Irr
We consider both sides as -modules. By Corollary 5.1 each summand is a nitely generated -module. Thus we see that any module of covariants for is contained in as a direct summand. Example 5.6. Let be a nite abelian group of order prime to char . Then any irreducible representation of is one-dimensional, and hence is dened by a GL . For each , let character
5.3
D hw
Let be a linearly reductive connected afne 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
d 1
V ( d D
D V
d Dhw w w
dD hw 5
5 jHutiw w P (EgxTw$1tg9fh(X1 gD ( e
D w
5 1
t (
2 d
D x
( p1
D l
d 1
d Dhw
( 1
U"
w
d 1
w
" hf
jf
1
P
D
(5.5)
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 SL the algebra Pol Pol is have seen in section 5.1 that in the case 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 nitely generated -algebra . For this we have to recall some basic facts about nite-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
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
w
(5.6)
Since is nite-dimensional, Wt is a nite 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 such that, for any and , any
T t
}xx r x 5 c 3
j f x( w 5 j hI
D 5 G ( e
"
r S
d 0
D f ( D x
oe
" D e
t 5 j u'
, i.e., homomorphisms of
( D iV
D vD V W
Dhw
"v F
1 D VD V W 2 ( 2
F D ( D X P D !
Hw
"
( 0 1
D w
( c 3 } } V a
D e
( " ! V t
74
For example, there are roots for GL . Each is dened by the to the matrix , where homomorphism which sends . Let be the set of roots. There is the notion of a positive root. We x 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 nite set of roots such that any root can be written as a linear combination of the with nonnegative integer coefcients. 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 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
Wt
, we have
d %
' D
( D d 5 D x vD
( D D d D x vD '
w c a l V
wt
"
be the homomorphism dening the action of on its image is equal to . This means that for any (5.7)
h v h % n
w (
d %
(
5 edHw
(
u
5 w
( %
B D 4f w
d %
Dd %
V ht D G |
( %
t
x (
(resp.
V t
75
we get
The set denes an order on the set of characters. We say that if is equal to a linear combination of positive roots with nonnegative coefcients. Let Wt be a maximal element (not necessary unique) with respect to this order. Then, for any , we have if It follows from (5.7) that acts identically on . Thus the whole group acts identically on . On the other hand, by Lemma 5.1, we get
invariant. Since the subset is Zariski dense in (check this for SL or GL , where this set consists of matrices with nonzero pivots), all elements of leave invariant. Thus is a -submodule. Let . Consider the -submodule generated by . Obviously it is contained in and
Hwt0
t E
D u
( B
j f g w t
V ( g w 5 vHQ D
g h
t w
Since
of the form
, where
H P
9x x
j R( edHw f g
5 fdw
5 w
g h
B D 49 w
B 4CD
( 3D
( DD D d D x v D ' ( ( D d D D x vD (
g D H !q%
g Hw
Dd x % D w q x % %x
w w
t
t u E % g ( Dg D Hw 9w e
0 g t %
Dd %
t
w
g w
H P
76
w w
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 , (ii) (a nonzero vector in is called a highest weight vector), (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 nite-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 afne algebraic variety Specm . For every dominant weight a homomorphism of -modules is determined by the image of a xed 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 irreducible 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
to
Conversely, if , by (5.4) can be written uniquely as a sum , where each belongs to an irreducible -submodule of . This implies that each
DH
g 1
d 1
o D u
d 1 1 D H BD u H h D u 5 ig 1 D o l i 1 ( dD
d 1
)g b$h
5
1 B g 1
g ') ( )
D 1
1 D " (
DH u (D1D u d
D o u I( o D
D u
1 t ug
g T) D % ( g y') D u
g Hw
g
D w g 1 1
77
Since every irreducible representation is isomorphic to some highest weight representation , we can apply (5.4) to obtain an isomorphism of -modules
Hom
It follows from the denition of that is spanned by a highest weight vector, and hence is one-dimensional. This gives
5.4
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 elsewhere, the dual root is given by . Note that the composition of a homomorphism (called a oneparameter subgroup) of and a rational character can be identied 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
o Dy
e a 5 c 3 } } V ( D
" D e
D D " e E
U"
( c 3 } 7676 3 k} V E 5 5 5 E3 a
g 4D (
5 g 1
Do
Hom
D o
d Dy1 D o u l
( V l
g 1
D u
This gives
D d 1
5 D o
dD y1 D o u l
1 ( g 1
( 1
D u
o y D
78
It is dened by the images of . Since it denes a homomorphism of groups it is for some . Thus a easy to see that the image of each is a monomial 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 denes a fundamental weight as a dominant weight with the property (the Kronecker symbol). Of course, one has to prove rst that such vectors, which obviously exist in , are really in . 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 subgroup of which consists of characters such that for all roots . 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
where . Any dominant weight from denes a one-dimensional representation . We have rank fundamental representations corresponding 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 dened 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 nitely generated -algebra . Then the subalgebra of -invariant elements is nitely 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
D P G D o e G
De
(5.9)
D e
l 56566 l 5 3 (D j P e D (
o D G G G ( D
m D cy
D e
D l 7766 l l 555 3
!l j ( D D j e & " 5 D GD P ( P
C}
D e
% ( 0%
t
79
is nitely 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 nitely generated by Nagatas Theorem, must be nitely generated too.
5.5
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 roots. Since if and only if , we see that consists of roots with . Simple roots are
The dual roots are the homomorphisms dened by Thus all dual roots can be identied with linear functions dened by where is the dual basis to the standard basis . A dominant weight must satisfy
and
766 &( 5 5 5
xx 3 ( 5 7 x $4l
3 ` p##`
"
xx 3 ( 7 x $l!l
1 d 1 d 1 1 r s l 5 5 7566 ( p l 666 ( o( 5 5 5 l d 1 r
diag . There are
fundamental
U"
3 ( 5555 666 ( p ! i D | n t i z!7 D G ( y U" c 5 5 5 3 3 7766 a D 555 3 D 7 666 D 555 3 5 Dx9x ( 3 666 x ( v9 h v h
" e
{
d y&D D 1 D e
( A A W e3 D
as a monomial of degree
5 e (
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
Here the highest weight is the monomial , where is the standard basis of . All other weights are with . The corresponding subspace is spanned by the monomial . We can write
We have
( o vD V 3 D ( D rxxx r W s'3 5 U 766 G( D o D V W 555 D xx xxx D 3 l v3 4x 3 Y $C3 3 Y Y 3 pu D3 pudd D x x x 3 3 Y $(Y 3 D3 l Y Y2P x ( D z xx 3 555 ( 663 6 3 x eq D 5 5 z v6566 3 W ( D3 l Y 5 D o D V v W
Pol . When we get
Let us consider the representation Pol Pol . The space has a basis formed by monomials in coefcients of a general binary -form
80
det
Pol
in the
Y `
x xx 3 ( 1 W3 xxx!D W3 ( mvxxxp1 D 3 ` )
(Y D l ( ( q V
where
D
det
det
Pol
det
5 Y
Y u ` 'D
A W
e 8 D W e
A
A DD ( vX) W YW e x D T)
Y
is
Y u D ` v( Y `
DY ` W e 3 W e
The cardinality of this set is equal to the number of monomials with weight . Let be a dominant weight. Suppose is a direct summand of Pol Pol . Then for some with . The weights of are the vectors This shows that Pol Pol contains
( D
5j (
belongs to
and so on. It is known that the generating function for the numbers equal to the Gaussian polynomial
where
D s !D D xxx w 5 D !D w D ( t g u 3 ps r x x x 3 r r
( I)
Pol
Pol
Pol
D X)
W e
where
coefcient of
( D Y `
f
82
Pol
As a corollary we obtain the Cayley-Sylvester formula for the dimension of the space of covariants:
and it is zero if
is odd.
This can be checked by dening the bijection ing a vector from to the vector
Corollary 5.3.
Pol
Pol
SL
Pol Pol
DD e 5 v)
5 Y
u ` y( Y
( p vX) W DD
w Y
`u
Pol
Pol
Pol Pol
DD e 5 T)
W D T)
( T) W DD
f ( D Y
Cov
D 4vD qY ` Y `
e g
Corollary 5.2.
-modules
SL
A W
D e 8 X) A W e
D T)
we have an isomorphism
A ( vX) W DD YW e x
qY `
D X)
by send, where
83
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 given by , where is a linear function on . It is easy to see that this map is SL -equivariant, where SL acts naturally on and on Pol . The inverse image under denes an equivariant linear map
Pol Pol
Pol
Thus, if the kernel of the map (5.10) is a SL sion given by the above formula.
Pol Pol
We use the symbolic expression to identify elements of Pol Pol with multihomogeneous functions on of multi-degree (see Lemma 1.1). The product of functions denes bilinear maps Pol Pol Pol Pol Pol Pol
DD T) e
DD v) e
DD e 5 X) DD T) e
D W `
" DD vX)
Pol Pol
Pol
Pol
. Let (5.11)
Remark 5.2. One can strengthen Theorem 5.6 as follows (see [49]). Let vector space of dimension , and let
D X) ( D 5 F W D D F W D F W D 3 T) W
DD vX) W
) ( ) o
( DD v) e
( W
W )
DD 'vX) e
D D o X) W
( W
B` (
( 2U
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 the Veronese variety) is always dened by equations of degree 2 (see [102]). The number of linearly independent equations is equal to
Pol
Pol
Pol
Do )
D T) e 5Do ) W
D vDX) W
D T)
" DD vX) W
o ) t " D o ) yW
W 9
Pol
Pol
Pol
D ) W e
" DD X) W
(5.10)
o b` $ D W
D T)
U"
DD vT) W
o b` $ D W
84
which endow with a structure of a graded algebra. The natural action of on denes an action of GL on both algebras and by auGL SL tomorphisms of graded algebras. Notice that is isomorphic to the algebra (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 Thus we have the following isomorphism of SL 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.
r Y(
Thus quadrics in Pol can be canonically identied with cubics in Pol . The Veronese curve is a rational space curve of degree 3. It is dened 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
Pol
Pol
5.6
Finally let us explain the tableau functions from the point of view of represencan be embedded (as a representation) into tation theory. Note that any some tensor power of some copies of . So when we take their symmetric
we can identify
, SL
(
D )
D T)
Pol Pol
Pol Pol
DD 5 vT)
D T) " X) DD D 3 ( "
( T) DD
( v) DD
DT)
D T)
( q
D DD ( X) 5 V X) D X) ( D ( D ( D 2( 2( ` Y 5 2q4 4 3 q4
-representations:
D )
( DD PvX) W
DX) 3
zW
o D
W DX) # " 5 W | W { W #
D X)
Do zX) W
Y I(
( DD vX) 3
D T)
"
oD X) c a W V
D T)
D X)
85
products and their tensor products we can embed each again into some . So for each irreducible representation is realized as an irreducible submodule of some . Let us nd 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
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., must be proportional to . If we must have
or or
. We will be taking
. . .
. . .
..
. . .
0 1
55 8 3 v665 3 3 3
)
( 3 (#3 3 D 3
c a 5
8 ) 8 )
xx x
c% b 5fa
r ( 4kf
3 3 3 y
3 `
53
fsx &( xx
3 3
` x x x #EB
8 )
D 3 p7r7
~s7 xxx 7(
yC3 ` 3 ` (
@3 ( m 3
8 )
86
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 rst 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 as a partition of . It is described by putting dominant vector boxes in the th row. It has columns of length ( ). We ll the boxes with different numbers . Each indicates which column enters into the minor of the matrix of the corresponding size. A lled 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 2 7 3 5 4 6 8 9
It turns out that the multiplicity of each in is equal to the number of standard lled Young diagrams of the shape given by the vector . It is given by the hook formula
(see [66]).
by embedding this space Example 5.9. We described invariants in Pol Pol into via the polarization map. Since the space of invariants is contained
D w
DD X) W
#` e $3ff3
mult
D r i
5 5 5 667
s( c
c 7fyy a t
8 )
55 665
c f
55 665 e
55 665
` ( #p
55 665
( D 8 F) g
y ( ` ` W e ) ( 3 p#` #`
` ( qkf
BIBLIOGRAPHICAL NOTES
87
in the representation of GL where , the corresponding dom. The representation is of course inant vector is one-dimensional. The Young diagram is of rectangular shape with rows and columns. The number of such diagrams is equal to the dimension SL of the space . It is not difcult to see that the hook formula gives the formula
SL
The standard tableaux on the set of size dened in Chapter 1 correspond to standard Young diagrams which are lled in such a way that if we as the disjoint union of subsets write the set , 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 forSL mula for the dimension of the space Pol Pol is not known for .
Bibliographical notes
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 rst 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 nitely generated was rst proved by P. Gordan [38] (see also classical proofs in [28], [39]). A modern proof SL on the can be found in [113]. Theorem 5.4 applied to the action of algebra Pol Pol is a generalization of Gordans Theorem. The rst 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 in [81]). Also known are combinants of two account of the case ternary forms of degrees ([28]).
Y j 555 9f Y 667 ff
GR U
D X)
(
h Y Y 3
DD v) W
` Y I( h Y
Dr D D Y Y ( 3 r ( YRPY ( 3 D Y Y 3
Gt D 565657Xx (
D 8 ) DD vT) W
( y Y Y
` Y 0(
( xY
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 rst 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 nd 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
`
Pol 5.1 Let SL Pol Pol its degree and weight. Pol be a covariant of degree and order and be an invariant. Consider the composition and compute
5.2 Let Hess Pol Pol be the Hessian covariant. Show that it denes 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 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.
. Describe the .
5.7 Dene the th transvectant of homogeneous forms in variables by generalizing the denition of the covariant . Prove that it is a concomitant and nd its multi-degree and order. 5.8 Consider the operation of taking the dual hypersurface in projective space. Show that it denes 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 . Find and show that this construction denes a contravariant on the space Pol its degree.
D zV
I " D V
D ) W
D 555 3 z 666 4
D V
D X) g
D V ( 3 g 3
EXERCISES
89
5.10 Let be a nite group which acts on a nitely generated domain . Asacts identically). Show that sume that the action is faithful (i.e., only for any irreducible representation of the rank of the module of covariants Hom is equal to . [Hint: Use the fact that each irreducible representation is contained in the regular representation (realized in the group algebra of ) with multiplicity equal to its dimension.] 5.11 Let be a nitely generated abelian group and let be its group algebra over a eld . Show (i) Specm is an afne algebraic group. (ii) if and only if is free. is naturally isomor(iii) The group of rational homomorphisms phic 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 .
subgroup). (ii) Show that the map denes a projective embedding of the homogeneous 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 dened in (ii) is the Pl cker embedding. u
(
Pol
Pol
( h
5.14 In the notation of section 5.1 show that Show that there is an isomorphism of Pol
p
t w f (
SL
of upper
D )
5 yg hw D
d hw D D3 ( l vh)
l (
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
D G
H
DH e DG e "
c a V
"DG
( w
Dv
D hw
Dc a V
D wG
j ( '4 x w V( D u ( h
( G D ( D v V
d 1 o D X g
( o
c V a
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
Pol
5.17 Let be as in the previous exercise. Find the decomposition of the GL module Pol Pol into irreducible summands (the ClebschGordan decomposition). 5.18 Find an irreducible representation of GL with highest weight equal to .
D )
3 l
D 5 )
( D bV g
( vX) W DD
(iii) Consider the example of 5.16 Let and char SL -modules Pol Pol
2h
Dg
D u l d 1
D u
D 1
D)
I( ) 5 g 1 p i g D ug
DT)
D )
Chapter 6 Quotients
6.1
Let be an afne algebraic group acting (rationally, as always) on an algebraic variety over an algebraically closed eld . We would like to dene 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 denition of ; for this we look rst at the categorical notion of a quotient object with respect to an equivalence relation. be a set together with an equivalence relation . The Let canonical map has the universal property with respect to all maps such that . Also we have This equality expresses the property that the bers of the map are the equivalence classes. Let us express this in categorical language. Let be any category with bered products. We dene an equivalence relation on an object as a subobject (or more generally just a morphism ) satisfying the obvious axioms (expressed by means of commutative diagrams). Then we dene a quotient as an object in for having the universal property with which there is a morphism such that factors through a morphism respect to morphisms . By denition 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
(
B
e 5 Tnx " P" I" " @ " " ' " y kB e e 5 D nx 3 D ( nx @ " h D 3 ( u" kB D "
D
92
CHAPTER 6. QUOTIENTS
Let be an algebraic action. We say that the pair is be a -variety and often drop from the notation. Let the morphism pr . This morphism should be thought of as an equivalence relation on dened by the action. A -equivariant morphism of -varieties corresponds to a morphism of sets with an equivalence relation. The denition of a -equivariant morphism can be rephrased by saying that the map factors through the natural morphism ; this corresponds to the property . This suggests the following denition. Denition. A categorical quotient of a -variety is a -invariant morphism such that for any -invariant morphism there exists a unique morphism satisfying . 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 dened uniquely up to isomorphism. A different approach to dening a geometric quotient is as follows. We know how to dene 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 this will be true is the topology in which a subset is open if and only if is open. Then we examine ringed spaces, whose denition is given in terms of choosing a class of functions on (e.g. regular functions, smooth functions, analytic functions). If is a function on , then the composition must be a function on . It is obviously a -invariant function. Using this remark we can dene the structure of a ringed space on by setting . This makes a categorical quotient in the category of ringed spaces. Finally, we want the bers of to be orbits; this is the condition that the morphism (6.1) is an isomorphism.
(i)
is surjective;
(iii) for any open subset of , the natural homomorphism is an isomorphism onto the subring of functions;
-invariant
" D Sy%
d vy% 3 DD
of
D y% 3
Denition. A good geometric quotient of a -variety phism satisfying the following properties:
is a
-invariant mor-
S D q
# "
RB %
" S " D y% 3 w
B #%
RI "
w s( EF w
d Dvy% 3 (TDy% D
D y%
" r
" @ F w
( D o
BCD D
" I4
@ 4 "
"
D y% 3
"
93 .
Proposition 6.1. A good geometric quotient is a categorical quotient. Proof. Let be a -invariant morphism. Pick any afne open cover of . For any the inverse image will be an open -invariant subset of . Then we have the obvious inclusion , where . Comparing the bers over points and using property (iv) (which says that the bers 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 dened by a homomorphism
-invariant morphism, the image of is contained in the subring of . This denes a unique homomorphism , and hence a unique morphism (because is afne). It is immediately checked that the maps agree on the intersections , and hence dene a unique morphism satisfying .
(i) for any open subset of , the homomorphism of rings is an isomorphism onto the subring of -invariant functions;
Conversely, a good geometric quotient satises properties (i)(iv). Proof. This is similar to the previous proof. With the same notation, let
P"
D 5 ) 3
( !uw
If& "
is equal to
w Q 3 {w
(iii) if
with
, then
QD3 w
(D w 3 w
(ii) if
is a closed
-invariant subset of
then
is a closed subset of
" D y%
d vDy% 3 D
D hw
be a
D %
"y) D
!%{Qp% )
F w (
" % ) !F
" } EF " @ F
D x%
DD vy% 3
d Dv% 3 D
Since
is a
( D ) 3
D % 3
@&t 9 Bd) D 3
D ) 3
fh " )
I
is equal to
"
@ j % 4f D % 3 D) 3 q% D ( j ) u$f }
94
CHAPTER 6. QUOTIENTS
This is a closed -invariant subset of , hence, by (ii), is an . Since , by (iii) we open subset of . Clearly, have , hence . Now composing the homomorphisms with the restriction homomorphism we get a homomorphism . Since is afne this denes 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 definition 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 denition. Conversely, assume is a good geometric quotient. Properties (i) and (iv) follow from the denition. Let us check properties (ii) and (iii). The set is open and invariant. Since the bers 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 satises the following properties:
x( H
the ber
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 bers. Since the bers are closed subsets, and lie in different bers, and hence they are disjoint. This proves (i). To prove (ii) we notice that by (i) two closed orbits in the same ber must have nonempty intersection, but this is absurd. Since each ber contains at least one closed orbit, we are done. Denition. A categorical quotient satisfying properties (i), (ii) and (iii) from Proposition 6.2 is called a good categorical quotient.
H
x |
xv x
x v
if and only if
D w
( 3 (
( w @ ( y% @
ww Q 3 w
" D ) @
Dy% 3
" !
D ) 3
DH x( D D H ( D ( x Q x bHD v D v ( H v Q v x x
% ( @ B% D 3 D 3 9 @
@ 4 "
6.2. EXAMPLES
95
Remarks 6.1. 1. Note that condition (ii) in the denition of a good geometric quotient is satised 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 eld of characteristic , and is a surjective -invariant morphism such that its ber 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 denitions of categorical and geometric quotients are obviously local is a -equivariant morphism, and is in the following sense: If 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 nite group considered as an algebraic group over a eld . Assume that is quasi-projective. Then the geometric quotient always exists. In fact, assume rst that is afne. By Theorem 3.1, the algebra is nitely generated over . Let be an afne algebraic variety with . By the theorems on lifting of ideals in integral extensions, the map satises properties (ii) and (iii) from Proposition 6.2. Also, the group acts transitively on the set of prime ideals in which lie over a xed prime ideal of (see, for example, [9], Chapter V, 2, Theorem 2). This shows that is an isomorphism. Now let be quasi-projective but not necessarily afne. Let be the closure of . Let be an orbit and let be a homogeneous polynomial vanishing on but not vanishing at any point of . Thus is contained in an afne subset . Recall that the complement of a hypersurface in a projective space is an open afne subset. This implies that , being closed in an afne set, is afne. Let . This is an open -invariant afne subset of containing . By letting vary, we get an open afne -invariant of . We already know that each quotient covering exists. We will glue the together to obtain the geometric quotient
j % 4f
F
D
% "x% 3 D
D e} @
"
fh " D E@
@ t 9 @ "R
"
|B
BB
96
CHAPTER 6. QUOTIENTS
(we refer to the gluing construction of algebric varieties in section 8.2). To do this is afne and is open in and ; this we observe rst that follows from considering the afne case. Thus we can glue and together along the open subset ; we do this for all and . The resulting algebraic variety is separated. In fact we use that in the afne situation
where acts on by the Cartesian product of the actions. Thus the image of in is closed, and, as is easy to see, coincides with . This shows that is is quasi-projective; we shall do this later. closed. It remains to prove that 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 rst 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 afne subset, the previous construction works and exists. Example 6.2. Let be a nitely generated -algebra with a geometric grading (see Example 3.1). Consider the corresponding action of on Specm . Let be the vertex of dened 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 afne 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 denition are veried. Let us show now that the categorical quotient of an afne variety always exists. We will need the following lemma. Lemma 6.1. Let be an afne -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 This is easy: since the sum of the ideals dening and is the unit ideal, we can nd a function and a function
D } t 3 } x( 5 D } } ( D3 }
e 3 m1 p e (
g
g X
f)
D Q 7f) w) S !%$% D D % k D % ( v!% !C% Q e k yi 3 3 yeT 3 3 3 3 'Ti ( D T D wi
" 2
Dy1
3 '}
D3 } t
% Q 7!{%
dD
D
Q % !% r( )
t ! ( '} } Q 3
( 1 p1
% Q !{% ( H
} ( 5 wD } w'} ( D3
D y1
6.2. EXAMPLES
97
such that . Then we take . Let be the linear subspace of spanned by the translates . We know that it is nite-dimensional (Lemma 3.4); let be a basis. Consider the map dened by these functions. Clearly, . The group acts linearly on the afne space, dening a linear representation. By definition of geometrically reductive groups, we can nd a nonconstant -invariant homogeneous polynomial such that . Then satises the assertion of the lemma. Now we are ready to prove the following main result of this chapter:
w
Theorem 6.1. Let be a geometrically reductive group acting on an afne variety . Then the subalgebra is nitely generated over , and the canonical morphism Specm is a good categorical quotient. Proof. The rst 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 veries condition (ii). Now let and be two disjoint -invariant closed subsets of . As above we nd a function with . This obviously implies that . This veries (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
'D
-equivariant map GL by the formula 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 ber of does not consist of one orbit, so the quotient is not a geometric quotient.
Dene a regular
D
( D w
D 7766 pX !v( 555 3 ( D o x( D 5 5 5 c } 7667 '} Gt 666 555 3 a V D 765756 qD } D z565656u (XDd'} ( 3 5 766 555 3 A " vw D t C o w ( S ( m V D d D
dD
( @ 4 "
98
CHAPTER 6. QUOTIENTS
6.3
Rational quotients
$
We know that neither nor exists in general. So a natural problem is to nd 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 rst 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 nite group. It is easy to see (see Exercise 6.11) that we can divide by in two steps: rst divide by , and then divide the quotient by the nite 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 ber 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 an afne algebraic group. If char , the set theoretical ber of coincides with the scheme theoretical ber (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 eld of characteristic zero is reduced. Since all bers of over points in O are isomorphic (they are conjugate subgroups of ), the theorem on the dimension of bers (see [102]) gives O
(6.2)
" bp
t ( t f ( e 5 j D v w 4 D v w D k 3
Let
. . By the
9 D
t 9
O , the complement O If O hence its dimension is strictly less than consider its orbit O . Since O that
is a proper closed subset of O , . Take any O O and O , applying (6.2) to we see (6.3)
Q
'%
( D V D 5 x w T( D v w U" w
i 5 p D i 9 D D D D ( 5 v D
Q
I"hfXE t
D 9
X ( H
Q
x(
99
theorem on the dimension of bers applied to pr , there exists an open subset reg for all for all of such that reg and reg . Applying (6.2) we obtain that for any is closed in reg reg the orbit O and has dimension equal to . Also, any other orbit in has dimension strictly less than . Let be any -invariant open subset of for which a geometric quotient exists. We assume that is irreducible. So and hence some of the orbits in must be of dimension reg . By the theorem on dimension of bers all bers of have dimension greater than or equal to and hence all bers 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 . is irreducible. Then reg contains an Theorem 6.2. (M. Rosenlicht) Assume open subset such that a good geometric quotient exists with quasiprojective . The eld of rational functions on is isomorphic to the subeld of -invariant rational functions on .
Proof. The proof is easy if we assume additionally that is geometrically reductive and is afne. Let be an algebraic variety with eld of rational functions isomorphic to ; such a always exists since is of nite transcendence degree over . Consider the rational dominant map reg dened by the inclusion of the elds . By deleting some subset from reg we nd . Replaca -invariant open subset reg and a regular map from ing by an open subset we may assume that is surjective. This is condition (i) from the denition 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 bers 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 bers 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 afne, even if is geometrically reductive we cannot separate . Consider the the closed orbits contained in the bers of the map generic ber of as an algebraic variety over the eld
x t dD 3 )
@ G " %
V hB d vX) 3 DD B DD T)
B %
@ " dD V
'%
'%
R p
'% " %
o
t
B%
t
% Y v
B % D B d D V V
Y
'% I% " Y y @ @
Y ( ve
V dD V
x (
$w% %
Yyp Y Q % y
dD V
100
m
CHAPTER 6. QUOTIENTS
Let be the algebraic closure of . The group acts on and the eld have the of invariant rational functions is isomorphic to . All orbits of same dimension. Suppose that a group acts on an irreducible quasi-projective variety such that all orbits are of the same dimension and closed. We dene 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 ber 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 veries condition (ii) of the denition of a good geometric quotient. The remaining conditions have been checked already.
Bibliographical notes
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 bers of the quotient maps have been omitted; we refer to [89] for a survey of these results.
Exercises
D3I v$4 3$I U"CD I 3 $I 3 A " A A r
6.1 Let map quotient? act on by the formula . Consider the . Is it a categorical quotient? If so, is it a geometric
'%
Any model of is called a rational quotient of by contains an open subset such that a good geometric quotient coincides with a rational quotient.
is equal to
ut
Dm F D F m %
I $I ED I $I ' 3 ( 3 x
D F m B%
Fm D F m
dD V
j 4
d V D
xG Y
f e ( W EY
8 B
Fm
EXERCISES
101
6.2 Let act on by the formula for some coprime to char . Let with the positive integers corresponding geometric grading dened by the action. Show that the geometric quotient (see Example 6.2) is isomorphic to a quotient of by a nite group. 6.3 Let be a graded nitely generated -algebra, and . Show that, if is coprime to char , , where is a cyclic group of order . 6.4 Construct a counterexample to Lemma 6.1 when is the additive group. 6.5 In the notation of Nagatas Theorem show that for any open subset of , the restriction map is a categorical quotient with respect to the induced action of . 6.6 Describe the orbits and the bers of the categorical quotient from Example 6.4 when . 6.7 Show that the categorical quotient of Pol Pol by SL is isomorphic to . Describe the orbits and the bers of the categorical quotient. 6.8 Let act on an irreducible afne variety and let be a invariant morphism to a normal afne variety. Assume that codim and that there exists an open subset of such that for all the ber contains a dense orbit. Show that . 6.9 Let be a nite group of automorphisms of an irreducible algebraic variety. Prove that . 6.10 Show by example that in general the eld of fractions of the ring of invariants is not equal to . Prove that if is a UFD and any rational homomorphism is trivial. 6.11 Let be an algebraic group acting regularly on an algebraic variety and let be a closed invariant subgroup of nite index. Suppose that a geometric quotient exists. Show that geometric quotients and exist and . Is the same true without assuming that is of nite index?
g Dg D g
3
% t 9 @ D D y@ @ " '
( 1
d 1&( 1 V D
g
D D V
( @
"h d y1 D
d D pX$# V V ( D
% y% 3 " D
3 565656 9 A
m1
j A f
d 1
( pU
1 7 1
D 9 3 g
3 A
We have seen already in the proof of Lemma 3.5 that a rational action of an afne algebraic group on an afne variety can be linearized. This means that we can -equivariantly embed in afne 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 dened 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 dened by sending to the point . This point is well-dened 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. Let be a map dened by a base-point-free . Of course, it depends on the choice of a basis; different choices dene maps which are the same up to composition with a projective transformation of . If is a closed embedding, is called very ample. If is very ample for some , then is called ample. , which comes with a proWe will often identify with its total space jection ; locally is the product of and the afne line 103
D !
D
D !
8 ( 8
" D ! y 2
" t D s
104
id
is commutative,
-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 denition that for any and any the induced map of the bers is a linear isomorphism. We can view the set of such isomorphisms as an isomorphism of line bundles
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 denition of linearization by means of an isomorphism it is easy pr to dene an abelian group structure on the set of line -bundles. If and pr are two line -bundles, we dene their tensor
"D o
where we consider as an automorphism of actions translate into the following 1-cocycle condition:
{t
t w
" D yfw EF
DH o
" D mH o
t w
H
is
-invariant.
" 7
" 7
yfF "
Denition. A
D qF
D o
3 A
(7.1)
The zero element in this group is the trivial line bundle whose linearization is given by the product id . This is called the trivial linearization. The inverse is equal to with dened as the inverse of the transpose of . One checks that this again satises the cocycle condition. The structure of an abelian group which we have just dened 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 dened by forgetting the linearization. Let us now describe the kernel of the homomorphism . Observe rst that if is an isomorphism of line bundles and pr is a -linearization of , then we can dene a -linearization of by setting pr . Thus, if is isomorphic to the trivial bundle, 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
dene isomorphic linearizations. Let Let us see when two functions be an automorphism of the trivial bundle. It is dened by
pg
D o
o D
H " D o
DH 3 A 3 y A 3 'B
H q ( w 5 D H w D x H w w D v H w
D w D v w & U" D ( D D F dD v w Xw w x t #w D A 3
D ( 5 D H o o vD H o
D
D " 3 { y2 A
DD qF v'
" D d
D d
D o
( D H
( 3 A y D #
A 3
A 3 o V D v w t
3 ! o H D ( "
pr
pr
pr
5 DH o
(7.2)
( D " o p o D H o
vo o
with the
" 3 A
Let alg denote the group of functions satisfying (7.2) considered as a subgroup of the group and let alg be its subgroup consisting of functions of the form for some . It follows from the denition 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.
pr
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 . Thus we obtain alg
pr
. Then
Ker
( D w o
D e ('D
D o X 3o D
D 5 ( D D o 3o ( o D D
( o D
( D o D 3 }
D n } 5 D o 3 D o D 3 G( D o D 3 g
alg
alg
alg
Theorem 7.1. The kernel of the forgetful homomorphism is isomorphic to the group
Pic
is connected and
DoD 3 }
Pic
D
" D d
oD
DoD 3 n
x ( 5 D vD qw D w D v w H t u#w
5 D v w H D
t
5 p o
m 4(
D d
D o m l
( D v w YD w x
vD o w o ih D
( DoD z 3 n
DoD 3 }
and
oD
D dD
U" D
107
Remark 7.1. According to a theorem of Rosenlicht ([93]), for any two irreducible and over an algebraically closed eld , the natural hoalgebraic varieties momorphism is surjective. Let us give a sketch of the proof. First we use that for any irreducible algebraic variety the group is nitely generated. (This is not difcult to prove by reducing to the case of a normal variety and then nding a complete normal variety containing such that is a divisor. Then for any the divisor of is contained in the support of and hence is equal to a linear combination of irreducible components of . This denes an to a nitely generated abelian injective homomorphism from the group group.) Now assume we have an invertible function on . For a xed we have a function . Since is a nitely 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
To nd 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 afne 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 sufcient. 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 isomorphisms . When , the unity element, we get an automorphism . 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
3 D # o H w 6
D @ 9
oV oV4DX@ o
@ Xs
D 9 U" v oD E@ D @ 9
D o o " D
D @ o T
D w ( 5 q7 # o H D H w w DH w w oD !t ( 4Ew
X F (
" o T@ D
o o D V
D o
D 9
D 4D D @ 9 ( 9
oD F
t Sk
t
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 This proves the assertion. 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 nite 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
H D (w w
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 Pic Pic is an isomorphism (see [46], Chapter 2, Proposition 6.6). These two facts imply that pr for some line bundle on . Let be a Cartier divisor on representing (i.e., ). 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
o
H D#
( D
|
Lemma 7.2. Assume that is normal (for example, nonsingular) and . For any line bundle on connected afne algebraic group. Let we have pr pr
D 8 $j f 3 A 3 ( tE3 D o v 3o ( D3 D 5 D q o 3o
5 Do V g " dD
% ( wv}
D o
tX
( B% D o
" D d
u&%
S
" D o V
D X'j 3 A f
( H
D
"
" D
#y t
H
o V
(7.3) is a
109
on the dimension of bers, the bers 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 . Dene now a homomorphism pr Pic Pic by
where is a chosen point in pr applied to restriction of and pr to there exists an isomorphism -linearization. This proves
. Suppose is trivial. By the preceding lemma we obtain that pr . But the are equal. This implies that is trivial, hence pr . By Lemma 7.1, admits a
Theorem 7.2. Let be a connected afne 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 nite index. In particular, for any line bundle on there exists a number such that admits a -linearization.
Remark 7.3. The assertion that Pic is nite can be checked directly for many groups. For example, the group is trivial for GL since these groups are open subsets of afne space. To compute Pic for PGL SL we use the following facts. Let be an irreducible hypersurface of degree in . Then Pic
This isomorphism is dened by restricting a sheaf to an open subset. Another fact, which is not trivial, is that Pic (7.5) . This is called where is the class of a hyperplane section of , provided the Lefschetz Theorem on hyperplane sections (see [40], p. 169).
f
8
Proof. Use the fact that for any afne algebraic -group Pic is nite (see [65], p.74).
( H
D d
" HD
GY D ) 5 $G ( Tm 8
) G ( D # qT)
d " D d
D X
" D
"
D 3# 3o
D
D X
(7.4)
110
Pic PGL
On the other hand, SL is isomorphic to the complement of a hyperplane section of the hypersurface
Pic SL
There is a notion of a simply connected algebraic group (which makes sense over an arbitrary algebraically closed eld). For all such groups Pic is trivial. Any is isomorphic to a quotient , where is simply connected and is a nite abelian group whose dual abelian group is isomorphic to Pic . For example, SL for PGL . For simple algebraic groups Pic is a subgroup of the abelian group dened by the Cartan matrix of the root system of the Lie algebra of . Here are the values of for different types of simple Lie algebras:
We refer to [86] for a description of the Picard group of any homogeneous space .
7.3
Linearization of an action
D ) (
Now we are ready to prove that any algebraic action on a normal quasi-projective variety can be linearized. Let be a -linearized line bundle, let be its space of sections, and let be an afne algebraic group. The group acts naturally and linearly on by the formula
(7.6)
G4wG G Gr
D
D D
F ( D D DvD v 3 w s w D s Dw )
3 p
5 ( D
GG G 4G D l l
D 1 T
1
G 4G
"
D 1
in
. So when
3 D ( D f
5 GU$G ( D ( D (
T
Now notice that PGL is isomorphic to . This gives the determinant equation
) y 3
, where
is given by
G 4G n
( k
GD 4
( w U G
g
( 2
LINEARIZATION OF AN ACTION
111
We know that any nite-dimensional subspace of is contained in a generated by the translates of a basis invariant nite-dimensional subspace of . Thus we obtain a linear representation GL
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-dened. The representation (7.6) in denes a representation in and the induced projective representation in . It is given by the formula
w
is
-equivariant. in we obtain a
If the rational map dened by a basis of is an embedding, then this map is an embedding too. Now let be an embedding of as a locally 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 nite-dimensional base-point-free linear system. It denes 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 as above, we obtain a linearization of the action of invariant linear system on .
"vDqU D s 8 D 4w B H U
5 D w x j w t s (D 9x 3 w q$f Ds j D w q$f ( x s w t s
H 4w
D D s k o (
3 D s
H 4w
8
"~
D 3p8 V " 8
D H w DD vU s 8 DD qU s o (
" 3 bD p8 V $ W
t $( sf
where
is a hyperplane in
. Now
"
Do
( s 5 j D
D w ( gx g 3 kXw
Do Bw t sf ( q$D
x( D s
Assume that the linear system exists such that by the formula
w t 3s
there
t
H 4w
D 5 hw
" Bf o w ( wD s Do
D s
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 way. Let us see that the line bundle is not -linearizable but is. We view as an open subset of the projective space whose complement is the determinant hypersurface given by the equation . The action is the restriction to of the rational map given by the formula
Note that this map is not dened 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 subset of matrices of corank equal to 1). Since is of codimension in the line bundle is the restriction of a line bundle on . The formula for the action shows that this bundle must be pr pr . Thus restricted to is isomorphic to the restriction 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 .
Bibliographical notes
The existence of a linearization of some power of a line bundle on a normal complete algebraic variety was rst 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.
( U G$ D p 3
( v q DD o DD v o v s 3o DD 8 8 } ( x 1 1 ( D
#B
3 5 p
U( f k (D 8 U
5 g
" h
7667 g 5 5 5 3
GD U wG ( D
" 8 I 8 D 1
k( }
j f 6 D
( D 5 5 5 D vD v7766 4g H
8 D 8
3 p
DD o
D (
DD o } 8 s k D DD v o
~ "
EXERCISES
113
Exercises
7.1 Let be a line bundle over a connected afne 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 afne algebraic group. Show that alg is a . In particular it is trivial if is connected homomorphic image of the group and complete. 7.3 Use Rosenlichts Theorem from Remark 7.1 to show that any invertible regular function on a connected afne algebraic group with value 1 at the unity denes a rational character of . 7.4 Let be a nonsingular algebraic variety and let be its nite group of automorphisms. Show that the group Pic is isomorphic to the group of invariant Weil divisors modulo linear equivalence dened by -invariant rational functions. [Hint: Use Hilberts Theorem 90 which asserts that ] 7.5 Let act on an afne algebraic variety dening the corresponding grading 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.
(yDD V 3 g o
D o D 3 g D H)
D
3 p D
D
( D D ! o (
D d
DX
" o '| 3 p
3 p
D H)
oD X
"
G
t t {
D
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 afne -invariant sets and then to construct the categorical quotient by gluing together the quotients . The latter quotients are dened by Nagatas Theorem. Unfortunately, such a cover does not exist in general. Instead we nd such a . cover of some open subset of . So we can dene only a partial quotient The construction of will depend on a parameter, a choice of a -linearized line bundle .
115
ss
us
D 5
D
(iii)
(ii)
is called stable (with respect to ) if there exists as in (i) and additionally is nite and all orbits of in are closed.
t Ss
j qD t 9 h x( 9 s f (
(i)
is called semi-stable (with respect to ) if there exists and such that is afne and contains .
Denition. Let
be a
and
w%
t
$ %
dD e
116
CHAPTER 8. STABILITY
Remark 8.1. 1. Obviously ss and s are open -invariant subsets (but could be empty). is projective, the sets are always afne, so this 2. If is ample and condition in the denition of semi-stable points can be dropped. In fact, for any , so we may assume that is very ample. Let be a closed embedding dened by some complete linear system associated to . Then is equal to the inverse image of an afne open subset in which is the complement to some hyperplane. Since a closed subset of an afne set is afne 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 afne 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. s us 4. The denitions of the sets ss do not change if we replace by a positive tensor power (as a -linearized line bundle). ss 5. Assume is ample. Let be a point whose orbit is closed s and whose isotropy subgroup is nite. I claim that . In fact let be as in the denition of semi-stable points. Then the set is 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 denition of stais omitted. bility the niteness of Let us explain the denition 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
Pol
"$
% % q $ y
He s" s"
"$
$ py
"$
$ c w
w Wb 7
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 afne variety given by a system of homogeneous equations (an afne 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 nd an invariant polynomial such that but . If we write as a sum of homogeneous polynomials of positive degree, we nd 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
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 denition of the gluing construction of algebraic varieties. Let be a nite set of afne algebraic varieties. The gluing data is a choice of an open afne subset for each , and an isomorphism for each pair . It is required that (i) , and is the identity for each , (ii) for any , and
Let be an equivalence relation on the set dened by if and only if there exists a pair such that and . 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 denes a homeomorphism of with an open subset of so that and We also introduce the notion of a regular function on an
'
6 7
w
0
F$
h 4 w0 w
$ %
! #"
4 eT e 4 p
' 1
'
$ %
! #"
8
4 2 530
& 0
118
CHAPTER 8. STABILITY
open subset . By denition, this is a collection of regular functions on such that for any . Let be the -algebra of regular functions on . The assignment is 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. obtained by gluing It follows from the denition that the ringed space of afne varieties is locally isomorphic to an afne variety, i.e., it admits an open cover by subsets which are isomorphic to afne 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 afne algebraic variety. One usually adds a separatedness property which ensures that the intersection of two open afne subsets is an afne 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 dene 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
There is an open subset in ss such that s and the s s is a geometric quotient of by . Moreover there restriction of to exists an ample line bundle on ss such that , restricted ss ss , for some . In particular, is a quasi-projective variety. to Proof. Since any open subset of is quasi-compact in the Zariski topology we can nd a nite 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 sufciently large . Let . For every , we consider the ring of invariant regular functions and let with as constructed in
X W D ( D 4 Q Y S T 4 52 Q R Q R U V Q
F$ $ 0
D I P B C B C B C B C
B H
4 52
U
p4 F $
B H D G
A@
B C
x %
E F 4 52
%
E F
D A
119
Nagatas Theorem. For each we can consider as a regular -invariant be the corresponding regular function on the function on . Let quotient. Consider the open subset . Obviously
This easily implies that both sets and are categorical quotients of . By the uniqueness of categorical quotient there is an isomorphism . It is easy to see that the set of isomorphisms satises the conditions of gluing. So we can glue together the quotients and ss ss the maps to obtain a morphism , where . To show that is separated it is enough to observe that it admits an afne open cover by the sets which satises the following properties: are afne and is generated by restrictions of functions from and . The latter property follows from the fact that is generated by 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 ss . The section of is equal to the section where is a section of . Thus the open subset is equal to where is an open subset of isomorphic to afne space. This shows that restricted to is equal to . However, is isomorphic to the trivial line bundle since any line bundle over afne space is isomorphic to the trivial bundle. By xing 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 dene a line bundle on . Obviously . Let us show that is ample. First we dene some sections by setting for a xed 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 afne, we ss is ample. Since is obtained by gluing together obtain that good categorical quotients, the morphism is a good categorical quotient.
D ( R D
D A
5
% '
s t
4 2 5v
5
q T
~ ~ "$ "$
d
` D
i
Y
"$ ' Y T T
D r 1 Y a
'
$ Ti ' '
D A
r u
"$
tX
4 52
q T
h i4 p p p h
' Y
D D Q D A q T
f g
w
4 52
r u
e
q T
T ' i i
'
r u
D A
S e
9#yw !
"$
D
a C
b
p
` D ( R
120
Q
CHAPTER 8. STABILITY
It remains to show that the restriction of to s is a geometric quotient. s is covered by afne open -invariant sets where acts with By denition s closed orbits. Since is a good categorical quotient, for any the ber s consists of one orbit. Thus 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
ss
is a projective variety.
Proof. First of all, we observe that by Nagatas Theorem the algebra is nitely 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 denition of Projm for any nitely generated graded -algebra .) The generated by homogeneous elements of elements generate the ideal positive degree. Thus the afne open sets cover ss . On the other hand the open sets form an open cover of with the property that . The maps dene a morphism ss which coincides with the categorical quotient dened 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 dene a -equivariant rational map . The closed subset of where this map is not dened is exactly the closed subvariety of equal to , where the bar denotes the image of the null-cone in . So deless we obtain the set and the quotient map ting this closed subset from ss ss .
S D
"
&
S e
j
&
"$
&
Proj
He
"$
S T
Y i s i % "y
&
" "$
7 r
& g
#S
&
"
w f &
e (
S 3
"$
"$
D & d
D (
Y
Q
4 52
8.3. EXAMPLES
121
ss Remark 8.3. Note that the morphism ss is afne, i.e., inverse image of an afne open set is afne. There is also the following converse of the such that the preceding proposition. Let be a -invariant open subset of geometric quotient exists and is an afne map. Assume is quasi-projective. Then there exists a -linearized line bundle such that . For the proof we refer to [73], p. 41.
k l
8.3
Examples
4 h S
and SL acting on naturally via its linear Example 8.1. Let 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 ,
Here are some integers. We assume that . Since Pic and we have Pic . A -linearized bundle must be of the form ; it denes a -equivariant Veronese embedding , 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 corresponding projective space. Obviously it is dened by the formula
Y S y t p p p } ~ u h r r h h S z y { S | Y u v 4 t
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 . does not change if we replace by . So we may We know that ss assume that , where by denition is dened only for
m y
y 8
r 9 i4 h
p p p
r s
p p p
h h
r s
r s
2
p q
p p p
r u
n o
ss
Pol
T T T T T Wb i i Y
"$ "
S
Y S
r u
r u
"
r @ i4 h
r u
v w
"$
t S p p p m
(8.1)
122
CHAPTER 8. STABILITY
divisible by and In other words we permit 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 , Assume now that . It is obvious that or . This implies that ss When , we have
i4 h
ss
and
ss
In particular, if , the quotient is the point. Next, we increase the parameter . If , we have further invariant polynomials. For example, if , the monomial belongs to . So the set ss 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 and do change otherwise. and Example 8.3. Consider the special case of the previous example where . The restriction of the action to is given by the formula
t
This shows that us . In other words, the set of semi-stable points is equal to the complement of the hyperplane at innity and the point . So it can be identied with . The quotient
i y
p w "$
i4 h
If we take
for
i4 h
we get
hr
5
i4 h
Projm
7 q
uS
Y y
i4 h
i4 h
i4 h
t o
if
. Hence
t
if
for all
y q
S i4 h S n 7 e S r h i4 h 2 S r
r @ h
y
S
y
r @ i4 h
2
8H
t u 4 t
if
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. . If , we have Similar conclusions can be made for any rational
Y
where is a homomorphism of algebraic groups. It is easy to see that any such homomorphism is given by the formula for some integer . In fact is dened 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
s
A section
of
, where
(
Thus
if and only if
Specm
for all
Y
4 52
0 y
The group
0 8 p 8
s 8 7
8 8
4 52
4 52
and . Every line bundle is isomorphic to the . As we saw in Chapter 7, its -linearization is dened
us
Thus
Y S
q w
y
S 4 h i4 4 S
4 52
for
124
CHAPTER 8. STABILITY
v
Recall that a -action on an afne 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 difcult one) whether any action of on afne 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
where Specm (see Example 3.1). . Similar conclusion can be reached in the case
ss
Example 8.5. A special case of the previous example is when the formula
7 r 7 i a t s i
ss
ss
s &
0 Y $
. Let
be its homogeneous
acts on
i h h i y
by
i
S p 4 S 4 e 4
h a
8.3. EXAMPLES
125
and the construction of the categorical quotient coincides with the construction (see Example 3.1). So we see two of the weighted projective space different ways to dene : as a quotient of and as a quotient of .
o p
As in the previous example, each -linearized line bundle is isomorphic to the trivial line bundle with the -linearization dened by an integer . We have
c e
Thus
a q
a 7
and
ss
. We have
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
y & &
h h
&
ac
Thus ss equation
of
given by the
. It is
We have a surjection
4
, dened by
However, this time the grading is weighted; the weights are Assume . Then for any ss , and
4 Y 4 e e c
" b y q "$ 4 @ Y 8 y 8 y
be again
` 4 (
and
. Hence
CHAPTER 8. STABILITY
4 h x D
Here we use for homogeneous coordinates in . In fact, this variety is covered by the two afne open sets given by . It is easy to see that . We also verify that these two sets are glued together as they should be according to our construction of the categorical quotient. ss Thus we obtain an isomorphism . 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 we see that induces an isomorphism . Similarly we treat the other pieces . Over the origin, the ber of is isomorphic to . Also, we immediately check that is a nonsingular variety. Thus is a resolution of singularities of . It is called a small resolution because the exceptional set is of codimension . The reader familiar with the notion of the blowing up will recognize as the variety obtained by blowing up the closed subvariety of dened by the equations . Assume . Similar arguments show that is isomorphic to the closed subvariety of given by the equations
h D
& s o i & 7 7
is isomorphic to
p Y 4
D 4 52 D
of
4 52 h
$ ve
2 e
4
4
D D
&
h 4 D h Y 4 4
4 X y
given
D A 4
a
D D
BIBLIOGRAPHICAL NOTES
127
represents a type of birational transformations between algebraic varieties which is not isomorphic to , but they are nowadays is called a ip. Note that isomorphic outside the bers .
Bibliographical notes
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 niteness 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 afne algebraic group . Assume is not afne. Show that for any Pic the set ss is empty.
n l
act on an afne algebraic variety and let 8.3 Let be the corresponding grading. Dene and similarly . Let Pic be trivial as a line bundle. Show that there are only three possibilities (up to isomorphism): ss , where (resp. ) is the ideal in generated by ss (resp. ). Show that in the rst case is isomorphic to Specm , in the second (resp. the third) case ss is isomorphic to Projm (resp. Projm ). 8.4 In Example 8.6 show that the bered product is a nonsingular 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 .
h e e ( e ( 2 D
"$ W
4
4
Y "
e X
4
4
e X
w h
e X
Pe
8.2 A
-linearized line bundle is called -effective if is -effective if both and are -effective.
ss
. Show that
"$
4
4 52
F( e
4
128
CHAPTER 8. STABILITY
be a nite group acting regularly on . Show that for any Pic , s s . Also 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 afne space of matrices. Consider the corresponding action of on the projective space . Find s the sets ss 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 ,
ss
(
s
D D
and
. Find all
7 7 y s"$ y
"$ "$
"$
"$ &"$
"$
D
8.5 Let
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 via a linear represenafne algebraic group acts on a projective variety tation GL . This can be achieved by taking a very ample -linearized line bundle on . As in Chapter 8, we denote by a representative in us of a point . We know that 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
4 h S
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 dened by the formula to reach the same conclusion. Let us set
4
h TB
129
4 52
m
Suppose all
for which
b 7 T T
U "$ p
4 h S
4 52
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 semistability: ss (9.1) Assume the preceding condition is satised 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 denition of stability, must be in this orbit. However, obviously xes , so that cannot be stable. Thus we obtain a necessary condition for stable points:
s
p m
We have to show rst that the numbers are independent of a choice of coordinates in , and also that the previous condition is sufcient 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 ; for any the corresponding point is equal to the point with projective coordinates where 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
So our point
is equal to
4 52
Obviously
b e
i t W s T e W
We set
Fb
Since
Tc i 7 T
y m
(9.2)
4 2 5e
x V
x V
i & 7a
"$
S ~
Qb
(9.3)
This can be interpreted as follows. Restrict the action of on to the action of dened by . Then has a natural -linearization and, since is a xed point, acts on the ber ; this denes a linear representation GL We know the geometric interpretation of the total space of the line bundle . It follows from this that the ber of the canonical projection over a point can be identied with . Thus from (9.3) we get that acts on the ber by the character . Hence it acts on the ber by the rational character . This gives us a coordinate-free denition of . In fact, this allows one to dene the number for any -linearized line bundle as follows. Let . Then and, as above, there is a representation of on the ber . It is given by an integer which is taken to be . In the case when , we can give another coordinate-free geometric interpretation of . Let be the homogeneous ideal dening in and be the homogeneous coordinate ring of . We have Projm . Let Specm be the afne cone over . Let and be as above. A one-parameter subgroup as above denes a morphism Let be the corresponding homomorphism of the rings of regular functions. The image of the maximal ideal dening the vertex of generates a principal ideal . I claim that
B 4 h S o 4 52
(9.4)
Since is generated by the cosets of the , we see that is generated by the such that . Now the assertion follows from the denition of monomials .
f f
i & b i
HH
i i
that is, is a xed point for the subgroup coordinate-free. Furthermore, for any vector
p i
7
p S e (
S 3
r u
e (
of . Also the denition of over ,
4 h
h q S e p p
131 is
2 X
p 8
r u r u
132
9.2
Now we are ready to prove the sufciency 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
x V
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. Denition. A regular map of algebraic varieties over an algebraically closed eld 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 afne, Hom as was dened in section 3.3. If is glued together from afne varieties , and is a eld, 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 eld of fractions. If is glued together from afne 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 of algebraic varieties, with which we are dealing. A regular map varieties over denes a map of -points. In particular, the residue homomorphism induces a map , which is 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. is proper over . First of all is Example 9.1. Any closed subvariety of proper over . Any -point of comes from a unique -point after multiplying
S
w w
x V
&
&
&
&
s &
&
for all
ss
for all
"$ yh
D 4 52 e
&
&
4 52
133
its projective coordinates by some power of a generator of the maximal ideal of . Now, it follows immediately from the denition of properness that a closed subvariety of a proper variety is proper. On the other hand, an afne variety is obviously not proper. Let us show that is not complete. First notice that the point is a -point of . In fact, it belongs to any open subset since it corresponds to a homomorphism dened by . However, this point does not come from any -point of . In fact for any . Now and but .
be the ring of formal Lemma 9.1. (Cartan-Iwahori-Matsumoto) Let power series with coefcients in and let be its eld 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 identies with a subeld 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 dene a one-parameter subgroup of by diag
e
Then
9.3
Let us prove Theorem 9.1. We have already proved the necessity of the conditions. First of all, by replacing with a sufciently high tensor power, we can place
' i 7 i 7 i T 8B' T 7
We will need the following fact. represents the double coset of the point as asserted.
The proof
p 8
&
Y S
S e
H i
&
Y v
S
&
&
i
Y & S
&
&
4 52
&
&
a q e
4 52
H
&
134
ourselves in the following situation: acts on a projective space by means of GL , is a -invariant closed subvariety of a linear representation . We have to prove the following. s Let and . Then there exists such that us . Moreover, if then there exists such that . From now on we drop from the notation , remembering that
s for all . We have to show that . s . Choose a point over . Then the map is not proper. In fact, if it is proper, is closed and the ber of over is proper over (Exercise 9.4). Since the ber is a closed subvariety of an afne variety, it must consist of nitely many points (Exercise is nite and is closed, so that is a stable 9.3). This easily implies that 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 which comes from a one-parameter subgroup . Let be the image of under the reduction homomorphism corresponding to the natural homomorphism . We can write
The expression in the parentheses is a -point of dened by a one-parameter subgroup of . Choose a basis in such that the action of is diagonalized. That is, we may assume that
4 h S S
This is equivalent to
&
52 4
52 4
i
52 4
&
Since
52 4
52 4
Thus, if we write
4 52 4
, we obtain
o w ed
&
&
&
&
52 4
&
&
4 52
52 4
52 4
F
&
52 4
52 4
Assume Suppose
(9.5)
x o
x w
a p "$
Y p
& U
r u
&
135
This implies that if . In fact, the element is reduced to the , hence modulo are constants equal to . identity modulo On the other hand, they are equal to modulo for some . This of course implies that if . Recalling our denition of we see that . This contradics tion shows that for all . if ss Assume now that for all . We have to show that . 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 and hence we get if . Thus . (9.5) belongs to This contradiction proves the theorem.
9.4
where
Also recall from Chapter 5 that there is a natural identication between the sets and which preserves the natural structures of abelian groups on both sets. We dene the weight set of the representation space by setting
x V
wt
This is a nite 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 dened by a homomorphism GL given by a formula
. . .
. . .
. . .
..
. . .
Recall from Chapter 5 that a linear representation of a torus space splits into the direct sum of eigensubspaces
(9.6)
&
&
q 0
i i
T
2 F
F$
52 4
w %
& "4
h
2 C
S u &
136
by setting
(Recall that the natural bilinear pairing between and is dened by the composition . When we identify and 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
x S x
The weight set of a point with projective coordinates set Its weight polytope is the subsimplex given by corresponding to diag
i h T
TY
T
7 i 7 B
wt
v q
m
x V
4 52
S v
Since
we obtain that
wt
If we choose coordinates in
x V
wt
Let setting
with
. . .
. . .
. . .
..
. . .
v ~
where we use the vector notation for a monomial be a one-parameter subgroup of Now let formula for some representation with we have a representation formula
4 p 1H 4 i i s p s
n i F% 7 e s 7 i 7 F $ pp i i i i c
(9.7)
by
(9.8) (9.9)
then
is the . If is , then
137
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 afne function such that (resp. ), and Moreover, the proof of this fact shows that one can choose with integral coefcients 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
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 x one maximal torus . Then for any other maximal torus , we can nd 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 nite). It immediis semi-stable (resp. ately follows that this property is satised if and only if . This implies that stable) with respect to
d
ss
ss
0F
& $ F
maximal tori
r 1
s $
S ~
ss
wt interior wt
"$
"$ "$
p
4 2 55
4 52
S
dT
is a torus we can restate Theorem 9.1 in the following way. be an ample -linearized line bundle
138
9.5
Kempf-stability
To nish 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 destabilizing one-parameter subgroups of we want to consider those for which is maximal. Since , we should rst normalize by dividing it by and show that the maximum is dened. 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 x one maximal torus . For any we can nd such that belongs to . Then we can set However, we have to check that this denition 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 nite 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 nite. This solves our problem of dening for any . So we set
exists in
4 52
W Hw
For any
we dene
4 2 55
t
p
ss
s $
ss
be a maximal torus in
. Then
% q
"$ "$ k7
4 2 55o
9.5. KEMPF-STABILITY
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
2 S 4 52 T S m u u 4 m
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 . Now the limits form a set of matrices such that if . It is immediately checked that this is the subgroup .
, we obtain
Now
4 2 5s55
and
4 52
F '
5 52 4
4 52
4 52
52 4
52 4
4 52
s x
s A
52 4
4 52
s w
s e
4 52
4 52
d 52 4
o m
d 52 4
4 52
4 52
W W i W Hw H p i c H e W H
is a subgroup of ,
m s 4 52 q q
140
Denition. The ag complex of is the set modulo the following equivalence relation:
It follows from Lemma 9.2 that the function is well-dened as a function on . Also the function is well-dened on . Now the idea is to nd 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.
x V
is called adapted for the point Denition. A one-parameter subgroup us if it satises the assertion of the preceding theorem. Let be the set of adapted one-parameter subgroups of . It is an equivalence class representing one point . We can assign to it the unique parabolic subgroup which we denote by . Of course we have to remember that all of these objects depend on the linearization of the action.
By Theorem 9.4, we must have . It follows from the denition that . However, it is known that the normalizer of a parabolic subgroup is equal to the subgroup.
Corollary 9.2. Assume is semisimple (e.g. in any proper parabolic subgroup of . Then linearization.
5 52 4
4 2 d55
4 2 d"55
4 2 55w
4 2 ds55
4 ! ! w $ # F
and
is unstable. Then
we have
. Indeed
such that
52 4
such that
of one-parameter subgroups of
H h
4
"
b
"
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 denes the set of adapted subgroups with this property for which the limit is reached the fastest. These subgroups dene a unique proper parabolic subgroup and is contained in this subgroup.
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 nd 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 eld and let be an ample divisor on . One denes 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 denes 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 representain is irreducible. As an abstract group is isomortion of phic to , where diag
&
" " 4
"$
&
"$ %
& (
v a
"
& (
"$
stability
Kempf-stability
semi-stability
Denition.
is called Kempf-stable if
74n W 7
is closed in
representing
v e
"$
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 nite abelian group , and the representation of on this space is called the Schr dinger represeno tation. If we assume that , then is very ample and can be used to dene 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 denes 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 denition 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 nite. 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 consider the geometric quotient GL which is a ne moduli scheme for abelian varieties with polarization of type and a level structure.
Bibliographical notes
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 dened by the formula d , where, for any , . Here we xed a -invariant hermitian norm in . The criterion states that is semi-stable if and only if belongs to the closure of the
@F
& (
C 8 "9 $
& (
1 7
e (
& (
1 2
v a
h R v e s a a
e (
"yF
Y
@ A
e S S
"
v a
v 3
"
m 4 e (
v a e
B
v e e
py
4 6
e (
@ A
e (
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-dened and can be extended to a function on the vector space Pic . These functions are used in [23] to dene 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.
9.4 Prove that a ber of a proper map is a proper variety. Give an example of a nonproper map such that all its bers are proper varieties.
8
p 7
SL and
"$
"$
(iv)
, then
D A
(iii) if Pic
is a
-equivariant morphism of ;
-varieties, and
x j
x V
(i)
for any
I Q QP
H FD AGE
S T
t Hw
& y ee F
"$
144
9.7 Let be an -dimensional torus acting linearly on a projective space . Show ss and the set of Pic such that is a that Pic nitely generated semigroup of . 9.8 In the notation of Exercise 8.6 from Chapter 8, nd 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 afne 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 matrices of size . Consider the action of the group SL on dened by conjugation of matrices. Using the numerical criterion of stability nd the sets of unstable and stable points. 9.12 Let and let act on via its linear representation. Consider the ag complex . For any point let . Show that this set is convex.
"
Let SL act linearly on in the natural way. This action denes an action of on the subspace of homogeneous polynomi. We view the latter as the afne space , where . als of degree A point of the projective space
4 2 5 S e
Hyp
is called a hypersurface of degree in . For each nonzero we denote the corresponding hypersurface by . When is an irreducible polynomial, it can be identied 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 multiplicities. 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 nite (see a modern proof in [85]). This implies that SL acts on an open nonempty subset with nite stabilizer groups. By Corollary 6.2,
y } { S S e
145
Hyp
SL
(10.1)
S {
S e
V W
Wb
S 3
3 x
4 2 5
Wb
146
Let be arbitrary. Recall that a hypersurface gular variety if and only if the equations
So if char does not divide , the rst equation can be eliminated. Let be the resultant of the polynomials . It is a homogeneous polynomial of degree in the coefcients 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 . Since the latter property is independent of the choice of coordinates, the hypersurface Hyp is invariant with respect to the action of SL . This means that for any we have for some . One immediately veries 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
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 nite.
Y Y p 9
`
Y e a
S h h p p p
` Y
and char . Then Hyp is the space of is the space of quadratic forms
X
w
X
X X
S f g
X
denes a nonsin-
147
A quadric is nonsingular if and only if the rank of the corresponding matrix . The determinant function on is the resultant is equal to 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 , which must consist of on an invariant subvariety of codimension 1 in Hyp 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
4
Let us consider the case . The elements of the space are binary forms of degree . The corresponding hypersurfaces can be viewed as nite subsets of points in taken with multiplicities (or, equivalently, as effective divisors on ). Let
e
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 rst 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
&
c d
4 2 5X
b E
f g
4
4
a
148
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
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.
Hyp
ss Now assume is even and let Hyp Hyp . This means that has a root of multiplicity but no roots of multiplicity greater than . ss ss Hyp containing . Consider the ber of the projection Hyp Since our categorical quotient is good, the ber contains a unique closed orbit.
ss
4 2 55
4 2 55
7 d q
(resp.
(10.2)
B H
4 2 55
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 is semi-stable (resp. stable) with respect to if and only if
x 7
7
v z ` y
i 7
T e
y y
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 nite 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
Hyp
Hyp
4
where the single point represents the orbit of . ss The variety Hyp is an irreducible normal projective variety of dimension : by construction of the categorical quotient, Projm Pol Pol
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 . ss s If we have Hyp . If we have Hyp and ss consists of subsets of two distinct points in . There is only one orbit Hyp of such subsets. ss The set Hyp consists of three distinct points in . By a projective transformation they can be reduced to the points . So the variety is SL again one point. This also agrees with the fact that Pol Pol , where is the discriminant invariant (see Exercise 2.6). s The set Hyp and the set consists of subsets of four distinct points in ss Hyp where has at most double roots. consists of closed subsets Since Hyp is an open Zariski subset of the projective space (see Exercise s s 10.1), and the bers of the projection Hyp Hyp 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
4
Hyp
s as a rational function on Hyp then we can nd its explicit expression as a in the coordinates of a binary form. To do this we rational function
Y
ss
a e
4
Hyp
Hyp
ss
&
where
ss
. In particular,
SL
e c
4
4
150
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 coefcients and we have
is the categorical quotient map. Its ber over 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 ber 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
&
Y w
Y f hg
Hyp
f g
ss
is invariant with respect to SL and denes a regular map from Hyp This is the geometric quotient map. The map
(10.3)
to
#t
f g
. Another invariant is
have rst to nd the algebra of invariants Pol Pol invariant, the catalecticant
4
SL
f g
a y
Y &
4
7 Y 4
151
This coincidence is not accidental. The equation above describes an elliptic curve branched over four points: the innity point and the three as a double cover of roots of the equation . In other words they are the zeros of the 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 .
Pol Pol
can be computed explicitly (see [28]). Let us write a general binary quintic in the form
e
). Then
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 dened by the natural grading of Pol Pol with the degree divided by 2. It follows that there is an isomorphism of graded algebras
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
x X t e e
Projm
Projm
7 b h
where
is graded by setting
hr
Y Y f vg
&s
qY Y x w
Y q
Y f hg
u @
Y q qY x t0 Y i h q Y q q qW q Y Y
qv Y
g r
ti Y
i s
u & P@
Y i
Y W
y B
u Y P u
i p
Let
4
4
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 SL Pol Pol 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
4
for some polynomial . We will consider as a graded algebra whose grading is dened by the natural grading of Pol Pol . It follows that there is an isomorphism of graded algebras
and is a weighted homogeneous polynomial. Arguing as in the preceding example, we see that
x 4 e
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 dened 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 in with , it follows from the denition of weighted projective space that
i 7
i 7
s i b
PA
where
is graded by setting
sYs
s @ @i q
u P
@ @ @
PA 7 PA PA 8
Y 4
q w
4
e e
153 acts on
i
Q
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 . It admits an automorphism of order 5. Finally observe that the locus of binary sextics with a multiple root and are both isomorphic to .
10.3
Plane cubics
and . Every homogeneous form of degree 3 in three variables (a Let ternary cubic) can be written in the form:
4
Now let us recall the classication 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
4 4
(1)
q W
4
&&
i
4
v 4
by the formula
x 4 4
t y
Y
q
4
q
Y
154
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
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 inection point and is the equation of the tangent line at this point. It is known that any plane curve contains at least one inection 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 coefcient of at is equal to zero. Thus in afne coordinates , the equation takes the form (10.4)
4
Obviously , so after scaling we may assume . Assume char . Replacing with , we may assume that . If char , by a change of variables , we may assume that . Thus, we obtain the Weierstrass equation of a nonsingular plane cubic:
, where ,
, .
(10.8)
7 7
&
D `
&h
rY
D x
D x
Y Y f hg
Y `
(6)
(5)
(4)
(3)
(2)
155
we reduce to one of two forms: By a linear transformation of variables or . Consider the rst 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
4 4
All nonsingular points of the rst curve are inection points. The second curve does not have nonsingular inection points. If char , then the curve has only one inection point with tangent line given by . Now change the coordinates in such a way that is the unique nonsingular inection 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
and
are equal. Now suppose is singular. We may choose Then the equation is of the form
7
.
4 4
q Y
(10.9)
in
156
`
Clearly,
We leave it to the reader to nd 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
where
or
Y t
rY
char
Y f vg
4 4
7
Y
157
Thus each monomial belongs to the eigensubspace , where is the character of dened by the vector . It is easy to see that is one-dimensional and is spanned by the monomial . Thus
u U v y x y A
wt
T1 T2
T2
is unstable with respect to . Then the origin lies outside of the Suppose convex hull of wt . It is easy to see that this is possible only if wt consists
:
T1
3
T0 T1
T1 T2 T0 T1 T2
T0 T1
T0
T0 T2
2
T0 T2
h `
Y v
Y f GP ed
9 Y
Y 9 r
2 ` o
4 52
Y h
&
f Gd e
acts on diag
Pol
4 4 52 52 4
& 0 0
F$
r
4
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
From this we see that the singular point is not an ordinary double point. It follows from the above classication of plane cubic curves that the following curves are unstable:
);
(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 xed 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.
It is clear that
. In afne coordinates
4 D
g g
D 4
c
4 7 Y q
D `
159
The dimension of its bres 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 positive 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. ss Since Hyp is of dimension 9, we obtain Hyp SL . It is a normal projective unirational curve, hence we nd that
4 t
Hyp
ss
SL
Since there is only one closed semi-stable but not stable orbit, namely the set of three non-concurrent lines, we obtain
It is easy to see that the orbit of the curve is of dimension . In the same bre we nd 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 rst one, and the closed orbit lies in the closure of the second one. , we have ve unstable orbits: irreducible cuspidal cubics (of If char 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 coefcients (see [98], p. 189192):
e x
Hyp
SL
b
Y
4 2 5X
Wb
Hyp
ss
Hyp
ss
SL
f v
f g Y W
SL In fact the algebra Pol Pol is freely generated by and . If one evaluates and on the curve given in the Weierstrass form from above, we obtain
Y Y 0
`
j 0 q W u g W W
4 4
where
160
Here we use the following dictionary between our notation of coefcients and Salmons:
4 4
y 4
y Y 4 4
y y y y y y y
Y Y Y
Y r rY Y w r Y r ` i Y r r ` Y r r `b
` m m m # 4
y 4
4 4
` y
` 4
Y s9
`
r Y f r Y bg Y Y r r t Y r Y Y r Y Y r r ` w s r Y Y r 9 9 rr r Y Y r Yr Y r Y Y r r Y Y r Y r r r Y sY Y9 r Yr rY r r Y Y Y r Y Y r Y Y r Y r Y r r Y Y r ` r Y
4 4 m
&r
Y r Y t w Y r b r r Y Y ' rY r Y r Y rY Y r rY Y r Y r `b Y Y r Y` rY T
Yr
Y r
Y r `
y
rY
r Y
`
rY
4
r Y
4 4
y y m
ss
4
r r Y
4
Y Y r
` m
Y 9
r
m y
Y r
4
Yr
m 4
rY
Y r
Y r t r Y
m
Y r
sr
4
Y r 9
`
Y s
`
Y r Y r `
rY
r Y
x
sr
rY
Y Y r
Y r `
rY
161 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 algebra 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 rst ve invariants. In analogy with the case , we nd 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:
v v
cubic
where . The automorphism group of this surface is isomorphic to the dihedral group of order 8. dened by the equation is isomorphic to The subvariety of . Recall that the latter is isomorphic to ; this is not an accident. represents six distinct points in , we consider the Veronese If a point of map to identify them with six points on a nonsingular conic in . Then the linear
4
cubic
i
Q
acts on
by the formula
i 8 7 T u n @ @ @ @
i9
&
Y f hg
h
m k h l
0 0
o T
i 7
162
system of cubics through these points denes a rational map from to . Its . The singular point of image is a singular cubic representing a 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:
t
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 (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 explicitly. probably the only examples where one can compute the spaces 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]). are rational varieties. In the case of It is conjectured that all the spaces 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
10.2 Let , be four distinct roots of a binary quartic . Let denote the determinant of the matrix with columns . The expression is called the cross-ratio of the four points. Prove that if and only if the correspondtwo binary quartics dene the same orbit in Hyp ing cross-ratios coincide after we make some permutations of the roots.
w5s
9 s 9 Y Y
w s h f
104
RY
'
73 2
103
q t i 0s p t i h 0& i s i h 0& 7
'
(
X X
A Y X
s Y
EXERCISES
a
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 SL , where is a subgroup of order 12. s with non-trivial stabilizer. 10.4 Show that there are exactly two orbits in Hyp Show that the closures of these orbits in Hyp are given by the equations and , where are the polynomials of degree 2 and 3 dened in section 10.2. us 10.5 Show that Hyp 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 .
4 a h
Stable congurations
h Sh
In the last two chapters, for typographical reasons, we denote the Grassmannian of -dimensional linear projective subspaces in by Gr . The Gr 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
Lemma 11.1. A generator of this group is the line bundle Gr plane section in the Pl cker embedding of Gr in u .
Sh v
Pic Gr
Pic Gr
165
S t
. . .
. . .
4
..
. . .
Gr
as a matrix
S
"! e
corresponding to a hyper4 h
Sh
Sh
Sh
Sh
Gr
4
f g
Sh
Sh
r s
i du
a
Y y } 4 4
S {
166
Its rows form a basis of . The Pl cker coordinates u of are the maximal . The open subset of minors of this matrix formed by the columns Gr with is the afne space . The restriction of any 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 Pl cker coordinate u of . Since Gr is not contained in a linear subspace of , the restriction map
Gr
Gr
where pr 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 Also is ample if and only if all are positive. In fact, if some tensor power of denes a closed embedding , then the restriction of to any subvariety isomorphic to a factor is an ample line bundle. But it is obvious that . The latter is ample if and only if this restriction is isomorphic to Gr . Conversely, any with positive (meaning that all s are positive) is very ample. It denes a projective embedding of which is equal to the composition
where the rst map is the product of the Pl cker embeddings, the second map is u the product of the Veronese embeddings, and the last map is the Segre map.
w z
4 52
Sh
r
~|}
4 52
{ 6
Gr
pr
Sh
r s
is injective. One can also show that it is surjective. For any vector we dene a line bundle on
Sh
Y e
Sh
b
4
e p p p h S
s e
0
Sh
Sh
r s
x w y
wz w z
p p p
Sh
r
f g
f g
V g
h h p p p
w z
Sh
Sh
r s
p p p Sh
r s
Sh
w z
w x
Sh
167
Now we are ready to describe semi-stable and stable congurations of linear projective subspaces
Proof. Let be the maximal diagonal torus in SL . Each one-parameter subgroup of is dened by diag , where . By permuting coordinates we may assume that
Suppose is semi-stable. Let be the linear space spanned by the unit vectors and let be the corresponding projective subspace. For any Gr and any integer , there is a unique integer for which
u
To see this we list the numbers and observe that since each is a hyperplane in and . Then we see that each occurs among these numbers and we dene to be the rst with . With this notation we can represent by a matrix of the form
S
(11.2)
where for all . It is clear from viewing the maximal minors of this matrix that if for any value of and . in the Now we notice that the projective coordinates of embedding dened by the line bundle are equal to the product of monomials
i i a i a i a
k & 8 Wb s b
of
(11.1)
Sh
r
Y 4
p p p
a
4 y u
Theorem 11.1. Let . Then s ) if and only if for any proper linear subspace
ss
z w
4 52
Sh
v rs
Wb
i i
4 52
w z
Sh
Wb
(resp.
b d n
#w
p p p
4 52
z w
y e
Sh
r s
u
168
Since we want this number to be non-positive (resp. negative) for all , we can take the special one-parameter subgroup given by
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 , we nd
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. It is also sufcient. In fact, if it is satised but is not semi-stable, we can nd some SL such that . By choosing appropriate coordinates, we may assume that and satises (11.1). Then we write as a positive linear combination of s to obtain that for some . Then the above computations show that (11.2) does not hold, contradicting our assumption.
4 S x 4 4 h S
' ' H i c Wb b
Xt
5d
d
Wb
Wb Wb
Xt
if
t o
'
d
4 52
4 52
Wb
f g
4 52
y
Xt
t P
f g
f g
t q
s b
as in the above. Using , we can rewrite the previous
p p p
h h
b h
r s
f g
b
b Wb i
p p p
of degree have
as in (11.1) we
f g
169 and
Y
ss
Sh
Corollary 11.2.
m Y Y
ss
Sh
Let us consider some examples. Example 11.1. Let . Taking to be a point, we get that can be equal to at most points among if is semistable 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.
. Then
is semistable no point is repeated more than times and no more than points are on a line .
m
4Y
v i
4 52
v i 7
w x
of
f g
Sh
Wb
of
. Also,
Wb
ss
SL
Sh
Sh
Sh
Sh
Wb
f g
Sh
Sh
ss
ss
s $
Let us rewrite Theorem 11.1 in the case where all case the linearization is called democratic). We set
Then
and
w w z r z$ i
Sh f g
are coprime.
&
0 4 4 0 m 0
4 0
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 , where for some points are represented by sextuples with the remaining four points on a line. Among them there are special orbits O corresponding to the sextuples with , where ss . So is a four-dimensional variss s ety, and is isomorphic to the union of 15 curves each ss isomorphic to SL . Each curve contains three points represented 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 s orbit space SL of sextuples of distinct points on . However, as we ss will see later, its closure in SL is not isomorphic to ss SL . Example 11.3. Let us take . Then we are dealing with sequences of lines in . Let us apply the criterion of semi-stabilty, taking to be rst a point, then a line, and nally a plane. In the rst case we obtain that is, no more than lines intersect at one point. Taking to be a line, we obtain
that is, no more than lines are coplanar. For example, there are no stable points if . This follows from the fact there is a line intersecting all of them. There are no that for any four lines in semi-stable points for . If , a pair of lines is semi-stable if and only
u m
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
S S B
lines
4 h
h g
'S
t i
n o
ou
h 4
v Y
h 4
A i
h 4
S m t t i S
R
4 S
h g
h 4
R R
m m
4 h
4
h 4
4
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 semi-stable orbit represented by the lines , and . If , the formula for the dimension of the quotient ss space gives us that , where is the stabilizer of a ss generic point in since there are no stable orbits. It is . In our case easy to see that (use that there is a unique quadric through the rst 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 x two points and three lines in one ruling is isomorphic to ). We will show later, by explicit computation of invariants, that
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 rst three lines . They belong to one ruling of lines on . The fourth line intersects at two points . The two transversals are the lines from the other ruling of which pass through . If the fourth line happens to be tangent to , so that , we get only one transversal. Now let be the two transversals. Then we have two ordered sets of four points on :
t
The proof that this map extends to an isomorphism consists of the study of how this construction can be extended to degenerate congurations.
11.2
Wb Wb 8 b y
Points in
points in . We have
p gY
gY
ss
h
SL
4 h
h i4
4 h
S 4 4
11.2. POINTS IN
171
(11.4)
172
ss
Corollary 11.3.
Proof. If , the left-hand side is empty and the assertion is obviously true in this case. We assume that . Let
S
ss This is an open nonempty subset of . We know that is an gen open subset. So if it is not empty it has nonempty intersection with . in the intersection, we obtain, since If we take a set of points
m u
This proves the assertion about the semi-stability. We prove the second assertion similarly.
S e
f g
if this condition is satised then each point semi-stable with respect to . In fact, each subspace points . Hence at most
0
w z
no two points
coincide,
for each
is of dimension contains
w x
i
. Conversely,
gen
S e
each subset of
S
points spans
gen
f g
f g
r i
ss
(resp.
m
r i
n q
n q
Bb
In particular, if all
of
(resp.
f g
i b wz$ Y w z$ Y
. Then
s
h S 0
z$ Y w
z$ Y w
Let
This is called an -dimensional hypersimplex of type . One can restate the preceding corollary in the following form. Consider the cone over in
4 h
In fact, if the rst coordinates of a point from the left-hand side are all positive, this follows immediately from Corollary 11.3. Suppose some of the rst coordinates of are equal to zero, say the rst coordinates. Then pr , where pr is the projection to the last factors, and . By applying Corollary 11.3 to , we obtain ss . It is easy to see that that
ss
q 2 S 4 52
pr
ss
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 coordinates equal to zero, and of points with one of the rst satisfying for some . The intersection of the latter
4 h
y yy
ss
SL
S
Y w $ Y
pr
ss
ss
w $ Y
pr
y y yy
ss
SL
n 7
Y w $ Y
wz w v w w z
z$ w
z w
n q
pT
PicSL
PicSL
ss
with a subset of
"$ Y
. We have
f g
4 52
PicSL
w 0 i zy
T i d
f g
11.2. POINTS IN
173
174
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 . Then there are two chambers with common boundary . We have a commutative diagram
4 h S 2 S e
s Here means that we dene the stability with respect to any from . The corner maps are birational morphisms, and the upper arrow is a birational map (a ip). 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
d S
Projm
Projm
Pol
SL
& %
SL
w z
ss
SL
S
qe
Y z$ Y w
S e
Sh
SL
SL
w z
T
f g
T w z$ Y z w
semi-stable but not stable (with respect to be open, it must be empty. ss Observe that subspace of dimension
y u u
). Since the set of stable points must if and only if there exists a such that
s
w z
b
w z
w z
such that
gen
S e
are
SL where . Let us denote the graded algebra Pol by . The First Fundamental Theorem tells us how to compute generators of the graded algebra . We have
h
Pol Mat
S
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 permuting the factors. It acts on the graded algebra via its action on the columns of matrices of size . The quotient is the moduli space of (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
h h i4 h
Pol
Pol
Pol Mat
S
Now, if we use Hermite Reciprocity (Theorem 5.6), we get an isomorphism Pol Pol
I
SL
Pol Pol
The projective spectrum of the left-hand side is the variety . The projective spectrum of the right-hand side is the variety Hyp SL Thus
rr YY
Example 11.4. Let us start with the case is spanned by the two functions on the set
0 Y 0
. Then the degree 1 piece of and . The value of the ratio dened by the coordinate matrix
SL
It can be shown (see Remark 5.2) that the isomorphisms phism of graded algebras
F$
SL
dene an isomor-
Y%
i x v
SL
SL
(11.6)
SL
(11.5)
&
h i4
&
&
&
&
S S m
11.2. POINTS IN
175
$ q
S &
&
&
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 cross-ratio. If we choose coordinates in the form , assuming that none of the points is the innity point, we obtain
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 ber over consists of congurations with . Similarly, one describes the bers 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:
4 u& x
Pol Pol
Using (11.6) we can identify (up to a constant factor) these invariants with the invariants and from section 10.2 of Chapter 10. Example 11.5. Let . The computations here are more involved than in the case which we will discuss in the next example. Here we is isomorphic to a Del Pezzo surface of only sketch a proof that the space degree 5 isomorphic to the blow-up of with center at four points
0
'
where
SL
'
If
we get
4 4
s T
s s
Y
y
y 4
`rY i Yr i
Y 4 4
0& 0s s i i
4 4
y y Y y y
Y r
Y
i
0 Y Y 0 0 0 0 4 Y 0 0 4 0
no three of which are on a line. The linear system of conics denes a morphism . Its bers are conics through the four points . There are three singular bers 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 ber , we consider the ber as and assign to the orbit of the ve points in . If lies on a singular ber, 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 bration map corresponds to the natural map dened by the projection . The three points in over which the ber is singular are the three special orbits of , and . The section corresponds to the set of orbits of , where .
The last two inequalities say that each row consists of two different numbers, so that
4 T
4 T
4 T
4 T
(11.7)
q T p 99 YY
R #
b
0
q Y s9
4
p i Y9 s99 Y Y q i q R a R R R 5 q
4
S 4 4 4
11.2. POINTS IN
177
4 4 U
q
4
178
For every , the product is a standard tableau function from . Applying the straightening algorithm, we nd
can be expressed as polynomials of degree 2 in the . Counting the number of standard tableau functions of size , we nd that In fact, we
This suggests that is isomorphic to a cubic hypersurface in First of all we have the following generators of :
&
"
&
&
i ~
&
is equal to
. This is true.
i ~
i ~
s Y
&
i ~
#A
f g
f g
When
'
u
T
u
T
X
` u
u u
R p
"
4 &
y y y 4 y y 4 d y
f g
y
f g
y 4
y 4
y y
q x
have for any . If we take a tableau function , we can write it as to tableau (11.7) with
and comparing the Hilbert functions we see that it is bijective. Thus Projm is isomorphic to the cubic hypersurface If we change the variables,
4 4 &
Y i 8
4 4
4 e
4 2
u 4 4 2 4 2 4 2 u
u u U
2 2 h 2
2 2 2
h h
4 2 `
Y
h
Y
U
2
2 2
2 `
h `
Y
h
h `
11.2. POINTS IN
if if
2 `
&
R Y R R r
7 4
4 &
whenever
, and similarly
if if if
U u
h x
whenever
we obtain that
Let
&
180
in which manifest the -symmetry. The cubic hypersurface dened 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). and . Again we take and try to Example 11.7. Let 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 ve 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:
pv
B R
&
&
&
The equation is again symmetric with respect to a linear representation of in the variables (but not with respect to the standard permutation representation 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 x the points and vary we see that this is of degree 2 in the coordinates of and vanishes when for some . Thus it 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 xed points are the sextuples lying on a conic. This is the self-association involution. We have a remarkable isomorphism, the association isomorphism: It is dened by the isomorphism of the graded algebras dened on tableau functions by replacing each determinant with the determinant , where . In the case , we get an involutive automorphism of the algebra which denes 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.
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
11.3
h i4
Lines in
&
i i
&
&
4 Y
4 52
i b A
q A
4 52
If char
11.3. LINES IN
181
4 m
182
is the complex projective orthogonal group PO O . The natural acto PO . tion of SL on Gr denes an injective homomorphism from PSL 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 stability for lines in shows that a semi-stable conguration 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 dened 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 functions dened by (see Exercise 2.9, or [121]). 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 functions. For , there are no relations between the basic invariants. Now
Pol
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 O Pol is spanned by monomials in such that each index appears among exactly times. In our case we have 10 basic invariants . For we have three monomials , 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 lying on the quadric is obviously zero. the subset of points in Thus, the restriction of the algebra
I
to
is freely generated by
7 y R Y i i
ss
Wb
( m
F%
`b
F$
9 4 Y 4
m y %
y $
f Y
F%
h i4 m
h i4
R %
m
BIBLIOGRAPHICAL NOTES
183
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 dened by the equation . The locus of sextuples of lines dened by this hypersurface coincides with the locus of self-polar for which there exists a nondegenerate sextuples, i.e., the sextuples quadric in such that the set of the polar lines is projectively equivalent to . Note the remarkable analogy with the structure of the variety , where the analog of the polarity involution is the association involution.
Bibliographical notes
The stability criterion for congurations of linear spaces (with respect to the democratic linearization) was rst given by Mumford ([73], Chapter 3). He also is a principroved that the quotient map for stable congurations of points in pal bration 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 coordinate algebras of four Grassmannians Gr was studied by R. Howe and R. Huang ([51], [50]). They show that this ring is isomorphic to a polynomial algebra. In the case when this was rst proved by H. W. Turnbull ([116]). Note that the GIT quotient 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
4 h S S
B R
S b
h g
Yz$ w
Sh
r s
nR i nR
m
Sh
S b
h i4
Gr
h i4
i
m
Note that a similar computation can be made in the case (see [117]). In the case , the algebra
and
@ s
R i R
h 0 4 4
h S
S S
S b
h g h 0
R i R
ib 7
4 X D x
184
algebras of the Grassmannians; so one needs additional work to compute the quo(see [22]). The moduli tients. One can also describe all orbits of four lines in spaces of ve 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 conguration spaces of points is obvious. It is not ss known whether the spaces SL are rational in general. This is known for lines in ([122]) and, more generally, in the case when (see [100]).
Exercises
in is closed but not 11.1 Prove that the orbit of stable if and only if there exists a partition of into subsets such that for any one can nd a proper subspace of such that .
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. can be generated by six elements of degree 5 satis11.5 Prove that the algebra fying ve linearly independent quadric relations.
4 52 S 4 C
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 .
4 52
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
Y S Q S Q Y
. of .
i4 h
&
4 52
is the quotient
ss
isomorphic to
r h
w z$ % Y i
S e
wz$ q Y bY t
f g
Sh
b
4 52
u S
dx
EXERCISES
185
11.9 Prove that the closure of the locus of Gr which admit 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 ss SL . Theory show that 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 ve points in general position; (ii) the nodes of are the images of the lines 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 , contains 15 double lines, each line is intersected by three other dou(ii) ble lines (nd the meaning of the double lines and the corresponding points of ss SL ), intersection in terms of the quotient (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 is a resolution of singularities of , (v) is isomorphic to the dual hypersurface of the Segre cubic primal . 11.15 Describe the orbits of SL in its diagonal action on Gr . Match the minimal orbits of semi-stable points with points in .
o
0
qA i
h i4
h i4
q i
h i4
4 Q
qA i
4 h i4
Sh
In this chapter we consider an interesting class of algebraic varieties which arise as categorical quotients of some open subsets of afne 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. act linearly on by the formula Let
where
As always we will identify the group with so that we consider the vectors as characters of . Since Pic is trivial and , we have a natural isomorphism (see Chapter 5)
v x S
Pic
v
Let us x and denote by the corresponding linearized line bundle. It is the trivial line bundle with the linearization dened by the formula
187
&
S S
8 i
4
7 7
p o
188
Let
where the matrix of coefcients 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 given by a system of convex polyhedral cone. By denition, this is a subset 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 ). For every polyhedral cone one can dene the dual cone:
S
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 dened as a convex hull of a nite set of positive rays spanned by vectors in . So we see that the set of vectors satisfying the system of linear for some rational convex equations (12.1) is equal to a set of the form polyhedral cone in . Now we use
4 h S v 4 S 4
#e 4 q i g
Hl i P
xc
such that
(12.1)
Here
are independent variables. It is clear that belongs to if and only if is equal to a linear combination of monomials , or, equivalently,
We identify its sections with polynomials . A polynomial denes an invariant section of some nonnegative tensor power if
d
S 4
cy
4 e
i o
y y
S 4 S
189 .
S S v
Proof. Let
is compact and hence its intersection with is nite. Let be this intersection. This obviously includes the vectors . We claim that this set generates the monoid . In fact we can write each in the form , 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 we can idengiven on the generators by the monoid multiplication. If tify 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
4 S S v w Y S u
where is the monoid of nonnegative vectors which satisfy (12.1) for some , and is the linear span of the set of monomials with . By Gordans Lemma, is a nitely generated graded algebra. Its homogeneous part of degree is . Let be the ideal . It can be generated by monomials and we choose a minimal set of monomial generators . For each let . For each subset of let Obviously, the open sets and coincide. By denition of semi-stability
ss
t 7 B i i
TY
T B0 T
7 h
v o
m
0 0
T T
7 i
v z
. The set
m m
(12.3)
190
where We know that the categorical quotient is obtained by gluing together the afne algebraic varieties with . We will now describe these rings and their gluing in terms of certain combinatorial structures.
12.2 Fans
R
be the map given by the matrix , then Let is a free abelian group of rank rank . Let
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
R e
&
&
be as in (12.4). Then
W 7
is spanned in
by the columns
of
with
4Y
spanned by the vectors More explicitly, let formed by a basis of the dual basis
4 4
d
4
R e
t t
R
S v
R 7
&
W o
S 7 h S i ` Y
R R
Y& p
For any
&
let (12.4)
(12.5)
(12.6)
12.2. FANS
191
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 dened 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 Lemma 12.2 that is nonnegative on and on . This proves the assertion. Denition. A nite 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 nite dimension, then chooses a lattice in , i.e., a nitely generated abelian subgroup of the additive group of with , and considers rational convex polyhedral cones, i.e., cones spanned by a nite subset of . Then an -fan is a nite collection of -rational polyhedral cones in satisfying the property from the above denition. A version of this denition includes in the fan all faces of all cones . be the dual lattice in the dual space . By Gordans Lemma, Let for each the algebra is nitely generated. Let
S
Lemma 12.3. Let be the set of convex cones is a face of both and .
. For any
i b
3
h r i r r &
m o
T
For each
for some
R V
&
Proof. Obviously
is isomorphic to
, where
m 3
S W
192
Specm be the afne 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. By denition has a cover by open afne subsets isomorphic to . 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
S
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 signicance 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 afne open subset is isomorphic to the product of a torus and the quotient of an afne space by a nite abelian group.
ss
ss Proposition 12.1. Let be the toric variety of the action homomorphism Aut is nite. Then only if
Y $
Y $
be the transpose of the inclusion map be the -fan formed by the cones
R
R j
R V 4
R %e
bS
e (
12.2. FANS
193
Proof. Assume some is not simplicial. We have to show that there exists a be the spanning vectors of Since semi-stable but not stable point. Let 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
Take a point , where if and otherwise. Since ss , we see that . 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 nd a one-parameter subgroup such that for all where . Then has not all coordinates equal to zero for and for all . This gives , hence is not simplicial. Since every line bundle on an afne variety is ample, we obtain that the toric ss varieties are always quasi-projective. Let us nd out when they are projective. Denition. A fan in a linear space
is called complete if
is complete.
r i r r
QT
8H
For any
and
we have
for some . This implies that Let us consider the one-parameter subgroup vector . It is dened by
v 4
S v
p Y
'
X
r i r
Y
r
9 i ` e Y Y
S v
(12.7)
194
Theorem 12.2. Assume that is not the trivial linearized bundle (i.e., ) ss ss . The toric variety is projective if and only if and is not contained in the convex hull of the character vectors
ss Proof. It follows from the construction of that it is equal to the projective spectrum Projm , where is the monoid of solutions of the system (12.1). We have and the inclusion denes 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 . This proves the assertion. ss Assume 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., dened by for some . Assume that , i.e., there exists a solution of for some . Then . Let
We can choose
such that
. For any
we have (12.8)
(12.9)
S v
i i 8 H s i v e i
$
T 4
8 8
Y $
v
7 e
e R %3
i 8
y 9 ` Y Y
n m
$
e
E F
D ( t
r u
S t
Y " Y $ Y
t S t 4 Y t
t Y
S 4
Y
S
"
A E F D ( t Y
Y Y
"
A E C D ( 4 m S m 4 t t 4
f g
y S t 4
R
y
t Y 4 y m
This action contains in its kernel the nite subgroup of equal to the group of points such that . The induced action of the torus is isomorphic to our old action. Clearly each is a linear combination of monomials such that
R Y R 4
Y S 4 h r 4
b
z 4 4
S Y
r h
i
S 4 S
r h
t 4 t S t 4 t Y
i
S 2 D
195
The restriction of this action to the open subset coincides with the action
r h
Thus we obtain
Comparing this with equations (12.8) and (12.9) we nd an isomorphism of vector spaces
12.2. FANS
ss
$ i
S 4 t Y
ss
ss
of
(12.11) (12.10)
m m
196
Applying the numerical criterion of stability we can nd the set of unstable . It follows from Chapter 9 (up to some modications points in using a weighted projective linearization, i.e. a -equivariant embedding of a is unstable variety into a weighted projective space) that a point if and only if the set such that satises the property that does not belong to the convex hull of the vectors .
12.3 Examples
Let us give some examples.
4 S
We have
the vectors
. Then
4 52
of
h
&
f g
i i
act on
by the formula
0 ' T i d
for some
and
acts on
ss
ss
&
r 8 pt Q i q
. Assume that
12.3. EXAMPLES
197
e3 1 2 3 e2
e1
We easily nd
4 52
4 52
on
by the formula
S 3
R
S e
This is the fan dening the projective space responding gluing. We can take for a basis of which is the set of vectors
y R 4 h S X
i i
4 52
span
Obviously the minimal generators of the ideal the cones of our fan are
Let us linearize the action by taking the line bundle have an isomorphism of graded rings
e
, where
. Then we
y
d 4 4
h S
R
. Thus
198 We have
. If or we can subtract 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 . Thus the fan consists of two cones
The quotient is obtained by gluing together two nonsingular algebraic varieties with the coordinate algebras
Similarly if we take
4
span
4 52
R
R
span
span
span
span
h e t
or
4
For any
Y
we have
Choose tion
7 h
Y y 4
Y m y
Y v
Y {
e m
4 m y 4
R Y
of
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 afne 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 modication. We refer the interested reader to [90] for more details.
-e 1 3
- e*2 - 2
span span
span span
R
4 4
R 4
R 3
e U
200 We have
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 . The torus of digaonal matrices in GL acts naturally 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 , where Consider the cone Gr over Gr character 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 pears exactly times in the sets . In other words, correspondence with the set of tableaux of degree and size . Let be the restriction of to Gr . Then
h d 4 52
4 52
Gr
h S
ss
SL
h h p p p
h i4
4 52
Gr
h S
Pol Mat
Pol
SL
SL
4 52
# e
q h
4 52
4
4
4 52
4 52
4 52
@ i t y
4 52
r Y 7
d T i& T
4 52
h i4
h i4
The variety
c 7
b
4
S m
m o
12.3. EXAMPLES
Also, we see that there is a natural closed embedding
201
It is easy to see that the monoid of nonnegative integer solutions of the equation consists of vectors with . Thus and . Thus
One can go in the opposite direction by identifying any toric variety with a categorical quotient of some open subset of an afne 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 generated by these variables. For each cone 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 . Let be the matrix whose columns are the vectors , and let be an matrix whose rows form a basis of the module . Assume that the vectors span . Then (i) with the action of
where (ii)
v
F B i
The embedding
e i
Y Y
i4 h
Y $
ss
depends .
U m 4
e m
202
Remark 12.1. Note that applying this construction to the toric varieties obss ss we obtain and the action tained as the quotients ss is isomorphic to the one we started with. However, in general, 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 nite abelian group.
Bibliographical notes
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 afne space was rst observed by M. Audin ([3]) and D. Cox ([16]). The relationship between solutions of systems of linear integral equations, Gr bner bases and toric varieties is a subject of the book o [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
4
12.1 Consider the action that the quotient is isomorphic to the blow-up of corresponding fan.
p
12.2 Let
act on
4
by the formula
Take , where . Show that the quotient is isomorphic to the blow-up of the projective plane at three points. Draw the picture of the fan.
$ $
4 52
&
4 52
S (
Y ` 4 Y 4
52 4
EXERCISES
203
12.3 Take a fan in formed by three one-dimensional cones spanned by the . Using Coxs Theorem represent the toric variety as a unit vectors 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 . 12.7 Consider the isomorphism Gr Gr dened by assigning to a 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 dened in Chapter 11.
h i4
4 52
q
4 52 S 2
4 52
h i4
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, Birkh user, 1991. a [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 vari t projective, in Operee ator Algebras, Unitary Representations, Enveloping Algebras, and Invariant Theory, Progress in Math. vol. 192, Birkh user, 1990, pp. 509539. a [11] C. Chevalley, Invariants of nite groups generated by reections, Amer. J. Math. 77, 1955, 778782. 205
206
BIBLIOGRAPHY
[12] A. Clebsch, Theorie der binaren algebr ischen Formen, Teubner, 1872. a [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, Birkh user, 1985. a [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. Dieudonn , J. B. Carrell, Invariant Theory, Old and New, Acad. Press, e 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 elds 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, Ast risque, vol. 165, 1989. e [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. Fa di Bruno, Th orie des formes binaire, Librairie Brero Succ. de P. a e 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 vollst ndig Formensystem der bin ren Form achter Orda a nung, Math. Ann. 17, 1880, 3152, 139152. [36] F. von Gall, Das vollst ndig Formensystem der bin ren Form a a Math. Ann. 31, 1888, 318336.
ter
[37] P. Gordan, Uber tern re Formen dritter Grades, Math. Ann. 1, 1869, 90 a 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. Grifths, 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.
Ordnung,
208
BIBLIOGRAPHY
[42] S. Gundelnger, Zur Theorie der tern ren cubischen Formen, Math. Ann. a 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, Birkh user Verlag, 1989. a [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 uber den gegenw rtigen Stand der Invariantentheorie, a Jahresbericht Deut.Math.Ver. 1, 1892, 79292. [69] S. Mori, Classication 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 nite 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 afne 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, Birkh user, 1983, pp. 395418. a [91] M. Roberts, The covariants of a binary quantic of the n-th degree, Quarterly J. Math. 4, 1861, 168178. [92] P. Roberts, An innitely 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 sph riques et th orie des invariants, Comptes e e 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. Schoeld, Birational classication of moduli spaces, in Innite Length Modules (Bielefeld, 1998), Trends Math., Birkh user, 2000, pp.297309. a [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 reection 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 uber 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, Birkh auser, 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
[113] B. Sturmfels, Algorithms in Invariant Theory, Springer-Verlag, 1993.
213
[114] B. Sturmfels, Gr bner Bases and Convex Polytopes, University Lecture o Series, vol. 8, Amer. Math. Soc., 1996. [115] M. Thaddeus, Geometric invariant theory and ips, 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. Weitzenb ck, Invariantentheorie, Noordhoff, 1923. o [120] R. Weitzenb ck, Uber die Invarianten von linearen Gruppen, Acta Math. o 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, Congurations of linear subspaces and rational invariants, Michigan Math. J. 46, 1999, 187202.
Sh
Sh
e (
"
&"$ s"$
e ( 2
7
S
t 8 %
S m
&
F
h & a
Sh
! ! "y h %
h S h S S e
Index of Notation
,2 ,4
F
a
ss
, 147 , 36
us
, 117
ss
4 Sh
alg
alg
4 G
,5
215
m 4 F
Sh
r s
, 142 ,2 SL , 37 , 105 Gr , 21 Gr , 167 , 147 Hyp , 21 , 142 Mat , 2 , 17 O , 28 Pic , 107 Pol ,4 Pol Pol ,4 Pol , 3, 5 Projm , 39 , 32 SL SL , 13 , 203 , 20 , 17 Spec , 2 Specm , 35 Sym ,6 Wt , 75 av, 33 , vii Cov Cov GL GL , 72 , 68
e e
p p p
0 0
b 7
m
e e e
h h
n n
s s$
x x x
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 wt , 137 wt wt , 138 symb ,9 , 13 Tab , 20 Tab hom , 20 , 17 ,4 , 49 , 35 , 21 , 191 , 83 , 21
h p h @ p
8 t @ i "$ $
INDEX
Index
absolute invariant, 150, 155, 161 action faithful, 89 linearizable, 124 rational, 37 regular, 37 additive group, 37, 56, 62, 101 adjugate matrix, 18 afne algebraic group, 35 afne 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 afne, 35 complete, 132 projective, 40 proper, 132 quasi-afne, 50 217 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
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 ag complex, 140 ip, 127, 174 Fundamental Theorem First, 20, 67 Second, 24 G-variety, 92 general points, 61 geometric grading, 38 gluing construction, 117 gluing data, 117 -action, 38 good GordanHilbert Theorem, 30 group scheme, 98 Haboushs Theorem, 42 Halphen pencil, 59
INDEX
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 inection 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 Pl cker equations, 24 u 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 Schr dinger representation, 142 o 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
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
INDEX
of a point, 136 of the representation, 135 weight set, 136 weight space, 135 weighted projective linearization, 196 space, 39, 180 Weitzenb cks Theorem, 51 o Weyl group, 138 Young diagram, 86 Zariskis problem, 49
Sh
Sh
"
&"$ s"$
e ( 2
7
S
t 8 %
m S m
Sh
S S
! ! "y h %
A e m S
Index
,2 ,4
e ( e
&
F
h & a
F
a
ss
ss
, 145 , 36
Sh
alg
alg
, 12
, 115
221
4 G
4 F
Sh
r s
e e
,5 , 70 Cov , 66 Cov , 140 GL ,2 GL SL , 36 , 103 , 21 Gr Gr , 165 Hyp , 145 , 21 , 140 Mat , 2 , 17 O , 28 Pic , 105 Pol ,4 ,4 Pol Pol Pol , 3, 5 Projm , 38 SL , 32 SL , 13 , 201 ,6 , 17 Spec , 2 Specm , 35 Sym ,6 Wt , 73 , 12
p p p
0 0
b 7
e e e m
h h
n n
s s$
8 t @ i
4
x x x
A A "$
S 4 t h @
, 37 , vii, 14 , 125 , 40 , 103 , 117 , 35 , 106 , 130 , 77 , 79 , 12 , 12 , 19 , 18 , 33 , 143 , 13 , 39 O , 98 pol ,5 res ,5 wt wt , 135 wt wt , 136 symb ,9 , 13 Tab , 20 Tab hom , 20 , 17 ,4 , 47 , 35 , 21 , 189 , 81 , 21
h p p
, 83
INDEX