Nilpotent Orbits in the Symplectic and

Orthogonal Groups

A dissertation

submitted by

Ellen Goldstein

In partial fulfillment of the requirements

for the degree of

Doctor of Philosophy

in

Mathematics

TUFTS UNIVERSITY

May 2011

Adviser: George McNinch

Abstract

Let K be an algebraically closed field of arbitrary characteristic and consider a

linear algebraic group G over K and its Lie algebra g. For X ∈ g, denote by OX the

orbit of X under the action of G on g defined by the adjoint representation. The

Zariski closure OX is then a subvariety of g. For G = On or Spn , the orthogonal

and symplectic groups, Kraft and Procesi showed that OX is a normal variety for

certain nilpotent X ∈ g when char K = 0. We begin to generalize their result for

char K = p 6= 2, concluding that an orbit closure OX of a nilpotent element X ∈ g is

normal if and only if it is normal in the union of OX with all orbits O of codimension

2 contained in the boundary of OX . In particular, if OX \ OX does not contain any

orbits of codimension 2, then OX is normal.

ii
Acknowledgements

First and foremost, I would like to thank my adviser, George McNinch, for sug-

gesting this problem and subsequently spending hour upon hour patiently answering

questions, explaining concepts, and giving me a shove when I would get stuck. I

would also like to thank the members of my committee — Mark Reeder, Montserrat

Teixidor i Bigas, and Richard Weiss — for their helpful comments and edits. Thanks

as well to Chuck Hague for the answering of yet further questions.

I owe many thanks and doubtless a few apologies to my officemates over the

years who have endured muttering, elated outbursts, and despondent venting alike

with humor, sympathy, and support. In particular, thanks to Meredith Burr who

continues to keep my spirits up, Andy Eisenberg and Charlie Cunningham who are

always willing to be distracted by a diagram chase, and Christine Offerman who

always has an open ear and a helpful perspective.

I am grateful for the mentoring I have received at Tufts from the various faculty

with whom I have taught, particularly Kim Ruane and Mary Glaser whose examples

have served to make me a better teacher. Thanks to Genevieve Walsh for inviting

me to be her TA at the Program for Women and Mathematics at the IAS and

thereby introducing me to the wider world of women mathematicians.

I would like to thank the recent chairs of the Mathematics Department, Bruce

Boghosian and Boris Hasselblatt, for encouraging all of the graduate students in

the department to become more involved. Thanks as well to the Tufts Center for

STEM Diversity for funding my trip to the MSRI forum Promoting Diversity at the

Graduate Level, which opened my eyes to the many ways that individuals can make

iii
a difference.

Thanks to the Tufts ARC and the Graduate Writing Exchange for creating a

structured environment for writing and a support system during this final period

of stress. Particular gratitude goes to Nicole Flynn for organizing the Disserta-

tion Writing Retreats and to Elizabeth Pufall Jones and Carolyn Salvi, my writing

buddies.

Thanks to my best friends, Jen Finn, Sarah Worrest and my twin in academia,

Karen Smyth, for keeping me sane. Thanks to my parents for their implicit support

over the many years. Finally, an infinitude of thanks to my fiancé, Kraig Theriault,

who has kept me fed, functional, and happy.

iv
Contents

Abstract ii

Acknowledgements iii

1 Introduction 2

1.1 Historical context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.2 Symplectic and orthogonal definitions . . . . . . . . . . . . . . . . . 3

1.3 Reducing to the Nilpotent Case . . . . . . . . . . . . . . . . . . . . . 4

1.4 Some Key Theorems and Definitions . . . . . . . . . . . . . . . . . . 5

1.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Foundational Material 9

2.1 Nilpotent Orbits and their Classification . . . . . . . . . . . . . . . . 9

2.1.1 Quadratic Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1.2 Classification of Orbits . . . . . . . . . . . . . . . . . . . . . . 10

2.1.3 Young Diagrams . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.1.4 ε-Degenerations . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Uniting the Orthogonal and Symplectic Cases . . . . . . . . . . . . . 15

2.2.1 Hom(V,U) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.2.2 Decomposing a Nilpotent Element . . . . . . . . . . . . . . . 17

2.2.3 Orbits in Hom(V,U) . . . . . . . . . . . . . . . . . . . . . . . 18

2.3 The Variety Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.3.1 Defining the Vi . . . . . . . . . . . . . . . . . . . . . . . . . . 21

v
2.3.2 Definition of Z . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.3.3 About Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3.4 Statement of the Main Theorem . . . . . . . . . . . . . . . . 24

3 Part A 25

3.1 Orthosymplectic Orbits . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1.1 Nilpotent Pairs . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.1.2 Young Diagrams of Nilpotent Pairs . . . . . . . . . . . . . . . 27

3.1.3 Orthosymplectic Nilpotent Pairs . . . . . . . . . . . . . . . . 28

3.1.4 Type αn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.1.5 Type βn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.1.6 Type γn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.1.7 Type δn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.1.8 Type ǫn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.2 Associated Cocharacters . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.2.1 Definition and Properties . . . . . . . . . . . . . . . . . . . . 40

3.2.2 Results in the Orthosymplectic Setting . . . . . . . . . . . . . 42

3.3 Dimensions of Orthosymplectic Orbits . . . . . . . . . . . . . . . . . 48

3.3.1 e as a Θ-group . . . . . . . . . . . . . . . . . . . . . . . . . .
G 49

3.3.2 e via Cocharacters . . . . . . . . . . . . . . . . .

Grading of G 51

3.3.3 More on Orbit Dimensions . . . . . . . . . . . . . . . . . . . 53

3.3.4 Proof of Proposition 3.10 . . . . . . . . . . . . . . . . . . . . 54

3.4 Dimensions in Z . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

3.4.1 Z as a Fiber Product . . . . . . . . . . . . . . . . . . . . . . . 57

3.4.2 From Orbits to Z . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.4.3 Proof of Theorem 2.13(a) . . . . . . . . . . . . . . . . . . . . 61

4 Part B 64

4.1 General Results on Good Pairs . . . . . . . . . . . . . . . . . . . . . 64

4.1.1 Good Filtrations . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.1.2 Good Pairs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

vi
4.1.3 Quotient Maps . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.2 Orthosymplectic Good Pairs . . . . . . . . . . . . . . . . . . . . . . . 71

4.3 Proof of Theorem 2.13(b) . . . . . . . . . . . . . . . . . . . . . . . . 74

5 Conclusion 80

5.1 Final Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

5.2 Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

5.3 More Known Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

Appendix A 85

Appendix B 87

Bibliography 90

vii
List of Tables

3.1 Indecomposable Orthosymplectic ab-Diagrams . . . . . . . . . . . . . 29

5.1 Irreducible Minimal ε-degenerations . . . . . . . . . . . . . . . . . . 83

viii
Nilpotent Orbits in the Symplectic and
Orthogonal Groups

Ellen Goldstein
Chapter 1

Introduction

1.1 Historical context

Over the past 50 years, many people have studied the question of which elements

of a Lie algebra have normal orbit closure. Kostant first proved in [Kos63] that when

char K = 0, regular nilpotent elements in a semisimple Lie algebra have normal orbit

closure, i.e. that the full nilpotent variety is normal. Veldkamp extended this result

to most positive characteristics, Demazure later proving this result for semisimple

and simply connected Lie algebras when K has characteristic p that is ”good”

[Jan04, 8.5]. Additionally, the closure of the ”minimal” nilpotent orbit is normal,

and is equal to the union of the orbit and the zero element, for simply connected,

almost simple algebraic groups [Jan04, 8.6]. This was first proved by Vinberg and

Popov in [VP72] in characteristic 0, and is implied by results of Ramanan and

Ramanathan in [RR85] in prime characteristic.

There are also many results for orbits of arbitrary nilpotent elements of simple

Lie algebras that make up a growing classification of normal nilpotent orbit closures.

Kraft and Procesi proved that for K with characteristic 0, closures of orbits are nor-

mal in the case of G = GLn acting on its Lie algebra g = gln [KP79]. Generalizing

their method for the general linear case, they also obtained a partial classification

of normal nilpotent orbit closures for G of type B, C, or D, the symplectic and

orthogonal groups [KP82]. Sommers then completed this classification in the sym-

2
plectic and orthogonal cases, over K with characteristic still zero [Som05]. Other

results in some of the exceptional groups when char K = 0 include those of Broer

in F4 [Bro98], Kraft in G2 [Kra89], and Sommers in E6 [Som03].

Some of these results have been generalized to positive characteristic. Donkin

generalized the general linear results of Kraft and Procesi to positive characteris-

tic in [Don90], Thomsen generalized in [Tho00] those results of Broer in F4 , and

Christophersen identified the normal orbit closures of E6 in [Chr06], generalizing

the results of Sommers. The generalization of Kraft and Procesi’s results, and those

of Sommers, in symplectic and orthogonal cases would complete the classification

of normal nilpotent orbit closures in the classical cases. The majority of Kraft and

Procesi’s proof is still valid in positive characteristic, and we follow their method

closely, adjusting the proofs of those results that no longer hold for char K > 0.

As we will see in section 1.3, the classification of normal nilpotent orbit closures

extends to arbitrary orbit closures, and thus can answer the question of which orbits

have normal closure.

1.2 Symplectic and orthogonal definitions

To examine the symplectic and orthogonal cases, we begin with a finite dimen-

sional vector space V over an algebraically closed field K of characteristic p 6= 2.

Endow V with a non-degenerate bilinear form h , iV such that hu, viV = εhv, uiV

for all u, v ∈ V , ε = ±1.

Denote by GL(V ) the linear algebraic group of all invertible linear transforma-

tions of V and let G(V ) denote the subgroup of GL(V ) that leaves the form h , iV

invariant:

G(V ) = {g ∈ GL(V ) | hgu, gviV = hu, viV }.

When ε = +1, G(V ) is the orthogonal group On , n = dim V . When ε = −1,

G(V ) is the symplectic group Spn .

The Lie algebra g(V ) of G(V ) consists of those elements of gl(V ) that are skew

3
with respect to the form:

g(V ) = {X ∈ gl(V ) | hXu, viV = −hu, XviV }.

The adjoint representation of G(V ) is given by conjugation in End(V ) of elements

in g(V ), so that the adjoint orbit OX of an element X ∈ g(V ) is the conjugacy class

Ad(G(V ))X = {gXg−1 | g ∈ G(V )}.

Our question then becomes: is the closure of a conjugacy class normal?

It is worth noting that OX is a locally closed subvariety of g(V ) isomorphic to

the quotient space G(V )/ G(V )X , where G(V )X = {g ∈ G(V ) | gX = X} is the

centralizer of X in G(V ) [Jan04] section 2.1. This relies on our restriction p 6= 2, so

that the character of K is good for G(V ), the symplectic or orthogonal group. This

then implies that OX is separable, meaning that g(V )X = Lie(G(V )X ) [Jan04, 2.5].

(For K with positive characteristic, the centralizer of X in the Lie algebra g(V )

is defined as gX = {Z ∈ g | [Z, X] = 0}.) For char K = 0, one always has that

Lie(G(V )X ) = g(V )X .

1.3 Reducing to the Nilpotent Case

In this paper, we consider only those orbits arising as conjugacy classes of nilpo-

tent elements X ∈ g(V ). These are, in some instances, sufficient to treat the general

case. For arbitrary X ∈ g(V ), OX is open in its closure (locally closed) and OX is

the union of finitely many conjugacy classes [Spr98], Lemma 2.3.3. There is then a

unique closed class O′ contained in OX , which is necessarily the conjugacy class of

a semisimple element. By [Lun73], one can fiber OX over O′ , associating to each

element Y of OX its semisimple part S in O′ , where Y = SU is the Jordan de-

composition of Y , and U is a unipotent element of G(V ). If Y ∈ OX has Jordan

decomposition Y = SU with S ∈ O′ , the fiber over S is isomorphic to the closure

of the conjugacy class of U in the centralizer ZG(V ) (S) of S in G(V ). The group

4
ZG(V ) (S) is reductive, and since G(V ) is one of the classical groups, so is ZG(V ) (S).

Hence we restrict our study to unipotent elements. Since char K is good as long as

p 6= 2, the unipotent variety of G(V ) is isomorphic to the nilpotent variety of g(V )

[Jan04] Remark 6.1. Thanks to the above filtration, one can study the normality of

the closure of the original orbit by studying the closures of nilpotent orbits in the

various centralizers. In the case G = GL(V ) where all orbit closures OX of nilpo-

tent elements are found to be normal, this is enough to show that orbit closures are

normal for arbitrary X ∈ g(V ) [Don90] Theorem 2.2(a).

1.4 Some Key Theorems and Definitions

Given an affine variety V, it is irreducible when its coordinate ring K[V] is an

integral domain. In this case, we denote the field of fractions of K[V] by K(V). An

irreducible affine variety V is then normal if K[V] is integrally closed in K(V). It

is smooth if, for each x ∈ V, the dimension of m/m2 as a K-vector space is equal

to the dimension of the local ring K[V]m of regular functions at x, where m is the

maximal ideal of functions vanishing at x. Equivalently, V is nonsingular at x if

K[V]m is regular [Har77, page 32], [Ser00, page 76] Theorem 9. As a consequence of

Corollary 3 of [Ser00, page 77], a smooth affine variety is necessarily normal.

The orbits OX for X ∈ g(V ) are themselves smooth, and in general it is the

boundary of the orbit closure, OX \OX , that may contain singularities. The following

consequence of Serre’s Criterion for Normality, [Mat80] Theorem 116, will be our

main tool in proving the normality of orbit closures.

Theorem 1.1. The affine variety V is normal under both of the following two

assumptions:

(a) V is a complete intersection in some affine space An .

(b) The singular locus of V has codimension at least two in V.

The concept of a (local) complete intersection is defined for subschemes of non-

singular varieties: If Y is a closed subscheme of a nonsingular variety X over K,

5
then Y is a local complete intersection in X if the ideal sheaf IY of Y in X can be

locally generated by r = codimX Y elements at every point. A complete intersection

subscheme is Cohen-Macaulay [Har77] Proposition 8.23. Applying this to V as in

Theorem 1.1, condition (a) implies that V is Cohen-Macaulay, i.e. satisfies (Sk ) for

all k ≥ 0 [Mat80] Proposition 113. Theorem 1.1(b) is equivalent to condition (R1 )

in Serre’s Criterion.

To use the above theorem, we also need the concept of a quotient for affine

varieties, referred to as a categorical quotient in [Bor91, 6.16], where it is defined in

more generality. If V and W are affine G-varieties and π : V → W is a morphism of

varieties, then π is a G-quotient if it is closed, surjective and

π ∗ (K[W]) = K[V]G .

We will also refer (somewhat imprecisely) to W itself as a G-quotient of V. The

upshot of this is that if V is a normal variety and W is a quotient of V for some

linear algebraic group G, then W is also a normal variety.1

Following the method of Kraft and Procesi, our over-arching strategy will be to

define an affine variety Z for each nilpotent X ∈ g(V ) such that OX is a quotient

of Z under an appropriate linear algebraic group. If we can then show that Z

satisfies (a) and (b) of Theorem 1.1, we have that OX is a normal variety. Initially,

Z is defined as a scheme, and we must show that it is indeed an affine variety by

showing that its associated coordinate ring is reduced, i.e. that it satisfies (R0 )

and (S1 ) [Mat80] Proposition 115. This is done by showing that Z is a complete

intersection and nonsingular in codimension 0. Kraft and Procesi remark that, in

general, Z is not normal and can have singular locus Zsing with codimension two. If

O′ is an orbit contained in the boundary OX \ OX , we shall see that codimZ Zsing ≥

1
2 codimOX O′ . Our approach then leads rather to a reduction of our original question

to one involving the analysis of certain orbits contained in the boundary of OX .

1
Suppose that f ∈ K(W) is integral over K[W]. Then f viewed as an element of K(V) is integral
over K[V] since the comorphism π ∗ : K[W] → K[V] is injective. This implies that f ∈ K[V] since
V is normal. Since K[W ] = K[V ]G , f ∈ K(W) is fixed by G, hence f ∈ K[V]G = K[W].

6
1.5 Summary

In Chapter 2, we give a rigorous treatment of the background needed for stating

the main result of the paper in Theorem 2.13. Section 2.1 contains foundational

results about orthogonal and symplectic orbits, and section 2.2 introduces spaces of

linear homomorphisms from an ε-space V to a −ε-space U , from which Z is built

and which serve to unite the orthogonal and symplectic cases. In particular, section

2.3 contains the explicit construction of Z, which is not a priori an affine variety.

Chapter 3 contains the background and proof of part (a) of Theorem 2.13. It

generalizes the proof given in [KP82] for K with characteristic zero. The key ad-

ditions here are an explicit treatment of the classification of orthosymplectic

 ab-
 0 X
diagrams, and the existence of a cocharacter associated to the element  ∈
X ∗ 0
End(U ⊕ V ) that takes values in O(U ) × Sp(V ). The associated cocharacter replaces

the use of sl2 -triples in [KP82] in the proof of a dimension formula relating the

dimension of an orthosymplectic orbit to its associated orthogonal and symplectic

orbits (sections 3.3.2, 3.3.3, and 3.3.4 ).

In section 3.1, we introduce nilpotent pairs and go on to classify them via ab-

diagrams. These serve as a classification of the orbits of elements in Hom(V, U )

under O(U ) × Sp(V ), introduced in Chapter 2. Section 3.2 uses this classification to

prove the existence of cocharacters associated to elements of Hom(V, U ). Section 3.3

contains results on the dimension of such “orthosymplectic” orbits, relating them to

the dimensions of symplectic and orthogonal groups. The main result of this section

is Proposition 3.10, which generalizes the characteristic zero result in [KP82]. In

section 3.4, we first describe Z as a fiber product, and then use this to relate the

dimension of Zsing to the dimension of orbits in our nilpotent orbit closure. Section

3.4.3 contains the proof of part (a) of Theorem 2.13, and shows that Z is a complete

intersection in an affine space and is nonsingular in codimension 1, showing that Z

is an affine variety, as well as a complete intersection.

Chapter 4 introduces the concept of good pairs of varieties and gives the proof of

7
part (b) of Theorem 2.13. In [KP82], this analysis was not necessary since quotient

maps necessarily restrict to quotient maps on a stable subvariety when working over

characteristic zero. We follow the method of [Don90], which generalizes [KP79] to

arbitrary characteristic, and utilize a property of good pairs of varieties to restrict

a quotient map to Z → OX .

Section 4.1 introduces the concept of good filtrations, as well as good pairs of

varieties, and presents general results, mainly taken directly from [Don90]. Section

4.2 contains results specific to the orthosymplectic case, in particular that V is a

good G0 (V )-variety [AJ84] section 4.9, and that (g(V ), g(V )e ) is a good pair of

G0 (V )-varieties, where g(V )e is the subvariety of matrices of rank less than or equal

to e, and G0 (V ) is the connected identity component of G(V ), which differs from

G(V ) only in the orthogonal case. Section 4.3 completes the proof of Theorem 2.13

where we prove that the map Z → OD is indeed a quotient for a suitable algebraic

group.

In Chapter 5 we restate our results and their implications. In particular, we

prove the final result of the paper: that OX is normal if and only if it is normal

in classes of codimension 2. We also discuss further results in [KP82] and [Som05]

that complete the classification of normal nilpotent orbit closures in characteristic

zero for the symplectic and orthogonal groups.

8
Chapter 2

Foundational Material

We begin by introducing the background necessary for stating the main result

of the paper, Theorem 2.13. We also give the classification of nilpotent orbits in the

orthogonal and symplectic groups.

2.1 Nilpotent Orbits and their Classification

2.1.1 Quadratic Spaces

Let K be an algebraically closed field of characteristic p 6= 2 and V a finite

dimensional vector space over K, where dim V = n for n ∈ Z>0 . For ε either +1 or

−1, a quadratic space of type ε is such a space V endowed with a non-degenerate

bilinear form, denoted by h , i, such that hu, vi = εhv, ui for all u, v ∈ V . In the

case that ε = +1, V is called an orthogonal space, and if ε = −1, V is called a

symplectic space. We will often refer to such a quadratic space by V alone when

the specific form is understood, though the notation (V, h , i) is more precise and

will be used when necessary. There is a more general definition of orthogonal spaces

when char K = 2, but we will not be concerning ourselves with it here.

Denote by G(V ) the subgroup of the linear algebraic group GL(V ) leaving the

form invariant:

G(V ) = {g ∈ GL(V ) | hgu, gvi = hu, vi for all u, v ∈ V }.

9
G(V ) ∼
= On or G(V ) ∼
= Spn when ε = +1 or −1 respectively. We will often use the
notation O(V ) and Sp(V ) when G(V ) is ambiguous. When ε = +1, G(V ) = O(V )

is not connected, and has connected component O(V )0 = SO(V ), the subgroup of

elements of O(V ) with determinant 1. (Elements of O(V )\SO(V ) have determinant

−1. When ε = −1, Sp(V ) is connected and all elements have determinant 1.) SO(V )

will play a role in what follows, though we will largely work with the full orthogonal

group.

The Lie algebra g(V ) of G(V ) consists of the elements of gl(V ) = Lie GL(V )

that are skew with respect to the form:

g(V ) := Lie G(V ) = {X ∈ gl(V ) | hXu, vi = −hu, Xvi for all u, v ∈ V }

and we have that g(V ) = so(V ) ∼

= son when ε = +1, and g(V ) = sp(V ) ∼
= spn when
ε = −1.

For any X ∈ End V , there exists a unique X ∗ ∈ End V called the adjoint of X,

defined by

hXu, vi = hu, X ∗ vi for u, v ∈ V.

We have that (XY )∗ = Y ∗ X ∗ and that (X ∗ )∗ = X, hence the map taking X ∈

End V to its adjoint is an involution. From the above definitions, G(V ) is the set of

all g ∈ GL(V ) such that g∗ = g−1 , and g(V ) is the set of all X ∈ gl(V ) = End(V )

such that X ∗ = −X. Choosing an appropriate basis for V , one can compute that

n2 − εn
dim G(V ) = dim g(V ) = .
2

2.1.2 Classification of Orbits

G(V ) acts on g(V ) via the adjoint representation, i.e. by conjugation in End V :

Ad(g)X = gXg −1 for g ∈ G(V ), X ∈ g(V ). Denote by OX the adjoint orbit of

X ∈ g(V ):

OX = Ad(G)X = {gXg −1 | g ∈ G(V )}.

10
If X ∈ g(V ) is nilpotent, the theory of Jordan canonical form says that there exists

a basis for V such that the matrix of X with respect to this basis has block form

 
Jλ1 0 ... 0
 
 
 0 Jλ2 ... 0 
 
 .. .. .. .. 
 . 
 . . . 
 
0 0 . . . Jλr

for suitable integers λi > 0. For m ∈ Z≥1 , Jm is the (m × m)-matrix with the

(i, i + 1)-entries equal to one (1 ≤ i < m) and all remaining entries equal to zero

(the m × m Jordan block).

We may insist that λ1 ≥ λ2 ≥ . . . ≥ λr , in which case the Jordan canonical form

of X is unique. The λi form a partition (λ1 , λ2 , . . . , λr ) of n, which we can also

write λ = [1r1 2r2 3r3 . . .] where r1 is the number of i with λi = 1, r2 is the number

of i with λi = 2, etc. Since the ri are uniquely determined by X, we may refer to

λ as the partition of X. It is a basic result of the theory of the Jordan canonical

form that two nilpotent elements in End V belong to the same adjoint GL(V )-orbit

if and only if they have the same partition. The following theorems from [Jan04, 1.4

and 1.6] classify the orbits of nilpotent elements of g(V ) under the adjoint action of

G(V ):

Theorem 2.1. Two elements in g(V ) belong to the same G(V )-orbit if and only if

they belong to the same GL(V )-orbit.

Theorem 2.2. Let λ = [1r1 2r2 3r3 . . .] be a partition of n = dim V , V a quadratic

space of type ε. Then there exists a nilpotent element in g(V ) with partition λ if
1−ε
and only if ri is even for i ≡ 2 mod 2, i.e.

(a) if V is an orthogonal space, then there exists a nilpotent element in g(V ) with

partition λ if and only if ri is even for all even i (even parts occur an even

number of times).

(b) if V is a symplectic space, then there exists a nilpotent element in g(V ) with

11
partition λ if and only if ri is even for all odd i (odd parts occur an even

number of times).

A partition corresponding to a nilpotent element in g(V ) as described in Theorem

2.2 is called an ε-partition or an ε-diagram (the terminology will become clear in

section 2.1.3).

In addition to classifying the open nilpotent orbits, ε-diagrams provide a classi-

fication for the orbit closures, since the closure of the G(V )-orbit of an element of

End V is equal to the intersection of g(V ) with the closure of the GL(V )-orbit of that

element. This follows from Theorems 2.1 and 2.2, which imply that Ad(G(V ))X =

Ad(GL(V ))X ∩ g(V ) for X ∈ g(V ) (the right-hand side is empty otherwise), and

the characterization of an orbit closure in Proposition 2.6 as a finite union of orbits.

In other words,

OX := Ad(G(V ))X = Ad(GL(V ))X ∩ g(V ) for X ∈ End V.

Remark 2.3. A conjugacy class OX under the orthogonal group is connected if and

only if OX is also a conjugacy class under the special orthogonal group SO(V ) =

O(V )0 . Jantzen discusses this distinction in [Jan04, 1.12], and concludes that a

conjugacy class OX under G(V ) is disconnected if and only if V is an orthogonal

space (ε = +1), and the partition of X is very even, i.e. if ri = 0 for all odd i (all

rows are of even length) and ri is even for all even i (because we are considering the

partition of an element in the orthogonal group). In this case, n = dim V ≡ 0 mod 4,

and OX splits into two conjugacy classes, O1 and O2 , with respect to SO(V ). The

normality of orbit closures under the action of the special orthogonal group was

established for char K = 0 in [Som05].

2.1.3 Young Diagrams

A useful way of representing partitions visually is as Young diagrams or Young

tableaux. Given a partition λ = (λ1 , λ2 , . . . , λr ) of n with λ1 ≥ λ2 ≥ . . . ≥ λr , the

Young diagram of λ is an array with l(λ) = r rows of boxes and λi boxes in the ith

12
row. For example, the partition (5, 4, 4, 3, 2, 1) = [1 2 3 42 5] has Young diagram

The dual partition λ̂ of λ is the partition (λ̂1 , λ̂2 , . . . , λ̂s ) of |λ| := n = dim V

with λ̂i equal to the number of boxes in the ith column of the Young diagram of λ
X
(alternately, λ̂i = rj ). The dual partition provides a concise way of describing
j≥i
j
X
the nullspace of a nilpotent endomorphism: dim Ker X j = λ̂i for X ∈ End V
X i=1
with partition λ. Equivalently, rk X j = λ̂i .
i>j
The following proposition describes the dimensions of a conjugacy class OX in

terms of the dimension of the centralizer G(V )X = {g ∈ G(V ) | gX = Xg} of X.

These are both given explicitly in terms of the Young diagram λ of X.

Proposition 2.4. Let V be a quadratic space of type ε and X ∈ g(V ) a nilpotent

element with associated Young diagram λ. Then

1 X 2 X
dim G(V )X = ( λ̂i − ε ri )
2
i i odd

and
1 X X
dim OX = (|λ|2 − ε|λ| − λ̂2i + ε ri )
2
i i odd

Proof. The first equation follows from [Jan04] Theorem 2.5 and equation 3.2(4)

rewritten in terms of the dual partition. The second equation follows from the first

since dim OX = dim G(V ) − dim G(V )X , where |λ| = n = dim V .

X
Remark 2.5. We also have from [Jan04, 2.5, 3.1(3)] that dim GL(V )X = λ̂2i ,
i
hence
X
dim GL(V )X = 2 dim G(V )X + ε ri
i odd

13
and
X
dim(Ad(GL(V ))X) = 2 dim OX + ε(|λ| − ri )
i odd

2.1.4 ε-Degenerations

Let λ be an ε-diagram, i.e. the partition of a nilpotent element X ∈ g(V ), V

a space of type ε. Denote by Oλ the orbit of such an element. Note that this is

well defined, since two elements X, Y ∈ g(V ) have the same partition λ if and only

if they belong to the same conjugacy class OX = OY = Oλ under conjugation by

G(V ).

Given an ε-diagram, we define an ε-degeneration of λ to be an ε-diagram µ such

that |µ| = |λ| and Oµ ⊆ Oλ . We write µ ≤ λ.

The following proposition gives a description of those conjugacy classes contained

in the closure of another in terms of their ε-diagrams.

Proposition 2.6. Given two ε-diagrams µ and λ with |µ| = |λ|, we have Oµ ⊆ Oλ if
j
X j
X X X
and only if µi ≤ λi for all j. Equivalently, µ̂k ≤ λ̂k or rk Y j ≤ rk X j
i=1 i=1 k>j k>j
for all j, Y with partition µ, and X with partition λ.

Proof. This is Theorem 3.10 of [Hes76], which holds for char K 6= 2.

An ε-degeneration µ ≤ λ is called minimal if µ 6= λ and there is no ε-diagram

ν such that µ < ν < λ. The conjugacy class Oµ is then open in the complement

Oλ \ Oλ ([KP82] section 3.1).

In terms of Young tableaux, an ε-degeneration of λ is obtained by moving boxes

down to another row so that the result is another ε-diagram. For example, when

ε = 1, let λ = (4, 4, 2, 2, 1):

Two ε-degenerations are µ = (3, 3, 3, 3, 1) and ν = (3, 3, 3, 2, 2):

14
µ is obtained from λ by moving the boxes on the ends of the top two rows down

to the third and fourth rows. It is a minimal ε-degeneration since the diagram

(4, 3, 3, 2, 1) obtained by moving down one box from row two to row three is not an

ε-diagram:

ν is obtained from λ by moving the boxes on the ends of the top two rows down to

the third and last rows. ν is a minimal ε-degeneration of µ.

Kraft and Procesi classify the irreducible minimal ε-degenerations in [KP82]

Section 3. They are used to classify those conjugacy classes Oµ ⊆ Oλ \ Oλ of

codimension two, the importance of which will become clear later.

2.2 Uniting the Orthogonal and Symplectic Cases

2.2.1 Hom(V,U)

Broadening our setting slightly, if V and U are quadratic spaces of type ε and

ε′ respectively, denote by L(V, U ) the set of linear maps from V to U . L(V, U )

identifies in a natural way (by choosing bases) with the set of (m × n) matrices,

where n := dim V, m := dim U . Here, we define the adjoint of X ∈ L(V, U ) to be

the (linear) map X ∗ : U → V such that

hXv, uiU = hv, X ∗ uiV for v ∈ V, u ∈ U

where h , iU and h , iV are the bilinear forms on U and V respectively. (X ∗ )∗ ∈

L(V, U ) again, and we can easily compute that (X ∗ )∗ = ε · ε′ · X:

hXv, uiU = hv, X ∗ uiV = εhX ∗ u, viV = εhu, (X ∗ )∗ viU = ε · ε′ h(X ∗ )∗ v, uiU .

Let ε′ = −ε, i.e. suppose that V and U are of opposite types, and let X ∈

L(V, U ). Using the characterization of elements of g(V ) and g(U ) from section

15
2.1.1, we have that XX ∗ ∈ g(U ) and X ∗ X ∈ g(V ):

(XX ∗ )∗ = (X ∗ )∗ X ∗ = −XX ∗ and (X ∗ X)∗ = X ∗ (X ∗ )∗ = −X ∗ X

since here ε · ε′ = ε · −ε = −1. We can therefore define maps π : L(V, U ) → g(U )

and ρ : L(V, U ) → g(V ) by π(X) := XX ∗ and ρ(X) := X ∗ X.

π
L(V, U ) / g(U )

ρ

g(V )

π and ρ are equivariant with respect to the action of G(U ) × G(V ) on L(V, U ),

given by (g, h)X := gXh−1 for g ∈ G(U ), h ∈ G(V ), X ∈ L(V, U ), and the adjoint

operation of G(U ) and G(V ) on g(U ) and g(V ) respectively:

π((g, h)X) = π(gXh−1 ) = gXh−1 (gXh−1 )∗

The elements g and h are in G(U ) and G(V ), respectively, and so h∗ = h−1 and

g∗ = g −1 as in section 2.1.1. Hence

gXh−1 (gXh−1 )∗ = gXh−1 hX ∗ g−1 = gXX ∗ g −1 = g.π(X)

where the action of G(U ) on g(U ) is conjugation. Likewise for the map ρ we have

ρ((g, h)X) = (gXh−1 )∗ gXh−1 = hX ∗ g−1 gXh−1 = h.ρ(X).

The space L(V, U ) is essential in what follows since it provides a way to unify

the symplectic and orthogonal cases. The G(V ) × G(U )-equivariant maps π and ρ

will help us translate back “down” to orbits in g(V ) and g(U ). We will always use

the setting above, where V and U are of opposite types. For our purposes, it is

only necessary to start with a space V of type ε and build an appropriate U . Also

essential is the following observation:

16
Theorem 2.7. Let V be an ε-space and U a −ε-space with m = dim(U ) ≤ dim(V ).

The map ρ : L(V, U ) → g(V )m given by ρ(X) = X ∗ X is a G(U )-quotient for the

action g.X = gX of G(U ) on L(V, U ), where g(V )m is the subvariety of matrices

of rank at most m. The map π : L(V, U ) → g(U ) given by π(X) = XX ∗ is a

G(V )-quotient map for the action h.X = Xh−1 of G(V ) on L(V, U ).

Proof. This is Theorem 1.2 in [KP82], which is reformulated from the characteristic

free proof due to De Concini-Procesi, [dCP76] Theorem 5.6(i) and Theorem 6.6.

2.2.2 Decomposing a Nilpotent Element

We begin with a quadratic space V of type ε and a nilpotent element D ∈ g(V )

with conjugacy class OD as in section 2.1.2. Define |u, v| := hu, Dvi a bilinear form

on V , which is degenerate with kernel ker D. If we define U := Im D ∼

= V / ker D,
| , | defines a non-degenerate form on U of type −ε:

|v, u| = hv, Dui = εhDu, vi = εhu, D ∗ vi = εhu, −Dvi = −εhu, Dvi = −ε|u, v|.

Decompose D ∈ g(V ) via D = I ◦ X where X ∈ L(V, U ) is the composition

proj D
V / / V / ker D / Im D = U

of D with the projection, the second map being an isomorphism. In other words, X

is just the map D with the codomain replaced with the image U of D. I is then the

inclusion U ∼
= V / ker D ֒→ V . I and X are adjoint as in section 2.2.1, i.e. X ∗ = I:

hXv, uiU = h[v], [w]iV / ker D = |v, w| = hv, IuiV

where v, w ∈ V, u ∈ U = Im D, w such that Dw = Iu ∈ V , and [v] and [w] are

the class of v, w in V / ker D respectively. Then D = IX = X ∗ X and D ′ := D|U =

XI = XX ∗ ∈ g(U ).

If D has partition λ of n = dim V , D ′ has partition λ′ of m := n − l(λ) where

17
X
l(λ) is the length of the partition, i.e. ri if we write λ = [1r1 2r2 . . .]. We have

m = dim U since D has kernel with dimension l(λ). In fact, λ′ can be obtained

from λ by erasing the first column of the Young diagram of λ. To see this, write

λ = (λ1 , λ2 , . . . , λr ). Then there exists a basis for V of the form

{D j vi | 1 ≤ i ≤ r, 0 ≤ j ≤ λi − 1} = {v1 , Dv1 , . . . , D λ1 −1 v1 , . . . , vr , . . . , D λr −1 vr }.

By definition, D λi vi = 0. After applying D, U then has basis

{Dv1 , D 2 v1 , . . . , D λ1 −1 v1 , . . . , Dvr , . . . , D λr −1 vr },

with respect to which D ′ = D|U has matrix in Jordan canonical form described by

the partition (λ1 − 1, λ2 − 1, . . . , λr − 1) =: λ′ .

2.2.3 Orbits in Hom(V,U)

Recall from section 2.2.1 the maps π and ρ:

π
L(V, U ) / g(U )

ρ

g(V )

We are interested in orbits and orbit closures, so we examine the behavior of

orbits under π and ρ.

Using the notation from section 2.1.4, let Nλ := π −1 (Oλ′ ), where Oλ′ ⊂ g(U ).

Here U a space of type −ε constructed from a space V of type ε and a nilpotent

element D ∈ g(V ) as in section 2.2.2, so that λ′ is the −ε-diagram obtained from

the ε-diagram λ by removing the first column, where λ is the ε-diagram associated

to the orbit of the element D.

D ∈ g(V ) determines the surjective map X from V to U by the construction in

section 2.2.2. Define L′ (V, U ) := {Y ∈ L(V, U ) | Y is surjective}.

Proposition 2.8. (a) ρ(Nλ ) = Oλ .

18
(b) ρ−1 (Oλ ) is a single orbit under G(U ) × G(V ) contained in Nλ ∩ L′ (V, U ).

(c) π(ρ−1 (Oλ )) = Oλ′ .

i.e. Nλ π /O ′
λ
ρ

Oλ

We begin first with a lemma regarding surjective maps V → U .

Lemma 2.9. For any map Y ∈ L′ (V, U ), the stabilizer of Y in G(U ) is trivial and

ρ−1 (ρ(Y )) is an orbit under G(U ).

Proof. Recall the action of G(U ) × G(V ) on L(V, U ): (g, h)X = gXh−1 for g ∈

G(U ), h ∈ G(V ), X ∈ L(V, U ). So the action of G(U ) on L(V, U ) is left multiplication

and StabG(U ) (X) = {g ∈ G(U ) | gX = X}. For such g we have a commutative

diagram
X /U
V @ O
@ @@
@@ g
X @
U

and for Y surjective, this necessitates g = id for g ∈ StabG(U ) (Y ). Hence the

stabilizer of y ∈ L′ (V, U ) is trivial.

For the second claim, let X ∈ ρ−1 (ρ(Y )), meaning that X ∗ X = Y ∗ Y . Y is

surjective and so Y ∗ Y has maximal rank m = dim U . This forces X to be surjective

and X ∗ injective. Hence ker X = ker X ∗ X = ker Y ∗ Y = ker Y and we can find

g ∈ GL(U ) such that gX = Y . We have then X ∗ X = Y ∗ Y = X ∗ g∗ gX. The

injectivity and surjectivity of X ∗ and X respectively implies that g∗ g = 1, hence

g ∈ G(U ) and X is in the G(U )-orbit of Y .

Remark 2.10. It is worth noting that the preimage ρ−1 (Oµ ) for µ ≤ λ need not be

a single orbit under the action of G(U ) × G(V ). These will be described in more

detail in Section 3.4.2 and Appendix B in terms of ab-diagrams.

19
Proof of Proposition 2.8. Let (g, h) ∈ G(U ) × G(V ) and X ∈ Nλ , i.e. X ∈ L(V, U )

such that XX ∗ ∈ Oλ′ . Then

gXh−1 (gXh−1 )∗ = gXh−1 (h−1 )∗ X ∗ g∗ = gXX ∗ g∗ = gXX ∗ g−1

since h ∈ G(V ), g ∈ G(V ) and so h∗ = h−1 , g∗ = g−1 . Now Oλ′ is a union of

conjugacy classes under G(U ), and so XX ∗ ∈ Oλ′ implies gXX ∗ g−1 ∈ Oλ′ . Hence

Nλ is stable under G(U ) × G(V ). Nλ is closed since π is open.

If we fix X ∈ L(V, U ) as constructed in section 2.2.2 from a nilpotent element

D ∈ g(V ) with partition λ, we have that ρ(X) = D ∈ Oλ and π(X) = D|U ∈ Oλ′ .

Thus Oλ ⊆ ρ(Nλ ) and Oλ ⊆ ρ(Nλ ), the image ρ(Nλ ) being closed as ρ is a quotient

map. Conversely, for each X ∈ Nλ , π(X) = XX ∗ ∈ Oλ′ , and combining sections

2.1.3 and 2.1.4, we have

X X
rk(XX ∗ )h−1 ≤ λ̂′j = λ̂j ,
j≥h j>h

λ̂′i = λ̂i+1 by construction. We then have

X
rk(X ∗ X)h = rk X ∗ (XX ∗ )h−1 X ≤ rk(XX ∗ )h−1 ≤ λ̂j .
j>h

Thus ρ(X) = X ∗ X ∈ Oλ , hence (a).

As stated, X as determined by D is surjective and by Lemma 2.9, ρ−1 (ρ(X))

is the orbit of X under G(U ). It follows that ρ−1 (Oλ ) is the orbit of X under

G(U ) × G(V ), the elements of Oλ having the form hX ∗ Xh−1 = (Xh−1 )∗ Xh−1 for

h ∈ G(V ), Xh−1 surjective. This then implies (b) and (c).

20
2.3 The Variety Z

2.3.1 Defining the Vi

Iterating the process in section 2.2.2, we now construct a variety Z sitting

“above” the closure of a given nilpotent orbit. The goal is to find an affine va-

riety that is normal and such that there is a quotient map from it to OD .

Start with a nilpotent element D ∈ g(V ) with conjugacy class OD = Oλ , i.e.

λ the partition of n = dim V associated to D, V a quadratic space of type ε. D

decomposes (canonically) as D = I ◦ X : V → D(V ) ֒→ V , and there exists a

non-degenerate form of type −ε on D(V ) such that X and I are adjoint, and such

that D|D(V ) = X ◦ I = XX ∗ is skew (see section 2.2.2 for further details).

Let V0 = V and X1 = X. Define the space Vi to be the image D(Vi−1 ), so that

V0 := V, V1 := D(V ), . . . , Vi := D i (V ), . . . , Vt := D t (V ).

Each Vi is a quadratic space of type (−1)i ε. We have Vt+1 = 0 for some minimal t ∈

N since D is nilpotent. In the previous paragraph, D = D|V0 = X1∗ X1 and D|V1 =

X1 X1∗ . We then decompose each D|Vi = Xi Xi∗ , Xi ∈ Hom(Vi−1 , Vi ), as above into

∗ X
the composition Xi+1 i+1 , Xi+1 ∈ Hom(Vi , Vi+1 ). Note that we naturally have

Xi Xi∗ = Xi+1
∗ X
i+1 , both equal to the skew endomorphism D|Vi belonging to the

conjugacy class Oλi , where

λ0 := λ, λ1 := λ′ , . . . , λi := (λi−1 )′ , . . . , λt = [1dim Vt ].

If we denote by ni the dimension of Vi , it is easy to see that ni = ni−1 −l(λi−1 ) = |λi |,

where l(λi ) is the length (number of rows) of λi .

2.3.2 Definition of Z

Define L(Vi−1 , Vi ) = Hom(Vi−1 , Vi ), the variety of (ni × ni−1 )-matrices for 1 ≤

i ≤ t. Let M := L(V0 , V1 ) × L(V1 , V2 ) × . . . × L(Vt−1 , Vt ) be their direct product.

21
Consider the affine subscheme Z ⊆ M of tuples (X1 , . . . , Xt ) satisfying the equations

(∗) below. We will prove that the equations (∗) define a reduced ideal, so that Z

can be viewed as an affine variety.

X1 X1∗ = X2∗ X2

X2 X2∗ = X3∗ X3
..
(∗) .
∗
Xt−1 Xt−1 = Xt∗ Xt

Xt Xt∗ = 0

Here Xi ∈ L(Vi−1 , Vi ), π(Xi ) = Xi Xi∗ ∈ g(Vi ), ρ(Xi ) = Xi∗ Xi ∈ g(Vi−1 ). The

following diagram helps to understand the equations in (∗):

Xi Xi+1
/ /
Vi−1 o Vi o Vi+1
Xi∗ ∗
Xi+1

where the two compositions Xi Xi∗ and Xi+1

∗ X
i+1 from Vi to Vi are both equal as in

eV
the previous section. We will see below that they are equal to the restriction D| i

e ∈ g(V ) with rk D
of some nilpotent element D e h ≤ rk D h for all h ∈ Z≥0 , i.e. some

e ∈ OD .
D

2.3.3 About Z

As stated above, each Vi is a quadratic space of type (−1)i ε, so we have G(V ) =

G(V0 ), G(V1 ), G(V2 ) . . . , G(Vt ), symmetric and orthogonal groups of alternating type.

Define the action of G(V0 ) × G(V1 ) × . . . × G(Vt ) on M via

(g0 , g1 , . . . , gt )(X1 , . . . , Xt ) = (g1 X1 g0−1 , g2 X2 g1−1 , . . . , gt Xt gt−1

−1
).

Z is then a stable subset of M . Indeed, if (X1 , X2 , . . . , Xt ) ∈ Z,

−1 −1 ∗ −1 −1 ∗ ∗ ∗
gi Xi gi−1 (gi Xi gi−1 ) = gi Xi gi−1 (gi−1 ) Xi gi = gi Xi Xi∗ gi−1 = gi Xi+1
∗ X −1
i+1 gi =

∗ (g −1 g
gi Xi+1 −1
= (gi+1 Xi+1 gi−1 )∗ gi+1 Xi+1 gi−1 .
i+1 i+1 )Xi+1 gi

22
Once we know that the scheme Z is reduced, it follows that Z is G(V0 )× . . . × G(Vt )-

stable.

Moreover, we can define a map ϑ : Z → OD by ϑ(X1 , X2 , . . . , Xt ) = X1∗ X1 .

Indeed, for any (X1 , X2 , . . . , Xt ) ∈ Z,

rk(X1∗ X1 )h = rk X1∗ (X1 X1∗ )h−1 X1 ≤ rk(X1 X1∗ )h−1 = rk(X2∗ X2 )h−1 =

rk X2∗ (X2 X2∗ )h−2 X2 ≤ . . . ≤ rk(Xh+1

∗ X 0
h+1 ) = rk(id|Vh )

i.e.

rk(X1∗ X1 )h ≤ dim Vh = rk D h ,

implying by section 2.1.4 that X1∗ X1 ∈ OD . Heuristically, the elements in Z are

the tuples that arise from the Xi as in the earlier construction (without altering the

Vi ’s), beginning with a nilpotent element with ε-diagram less than or equal to that

of the original D, meaning those elements contained in the closure OD (see section

2.1.4).

To see that OD is contained in the image of ϑ, consider the element (X1 , X2 , . . . , Xt )

of M with Xi the linear map Vi−1 → Vi defined in the decomposition of D|Vi−1 . It is

in Z by construction and D = X1∗ X1 . ϑ is G(V0 )-equivariant by construction, and

so OD ⊆ ϑ(Z). We will see in the proof of Theorem 2.13(b) that ϑ : M → g(V )

restricts to a quotient map on Z. Then ϑ(Z) is a closed subvariety of g(V ) and since

OD ⊆ ϑ(Z), OD = ϑ(Z).

Remark 2.11. An alternate method of defining Z as a scheme is as the fiber over 0

of the map ζ : M → N := g(V1 ) × g(V2 ) × . . . × g(Vt ) given by

ζ(X1 , X2 , . . . , Xt ) = (X1 X1∗ − X2∗ X2 , X2 X2∗ − X3∗ X3 , . . . , Xt Xt∗ ).

Remark 2.12. Note that the ith equation in (∗) generates dim g(Vi ) equations in

ni−1 × ni affine coordinates. This can be seen inductively, starting with the last (the

tth ) equation

Xt Xt∗ = 0.

We have Xt Xt∗ ∈ g(Vt ), and the map π : L(Vt−1 , Vt ) → g(Vt ) given by X 7→

23
XX ∗ is onto by Theorem 2.7, so Xt Xt∗ gives dim g(Vt ) quadratic expressions in the

nt−1 × nt affine coordinates corresponding to the entries in an arbitrary Xt , each

equal to 0.

In the ith equation, Xi Xi∗ ∈ g(Vi ) and π : L(Vi−1 , Vi ) → g(Vi ) is onto, so there

are dim g(Vi ) expressions in the ni−1 × ni affine coordinates of the entries of Xi ,
∗ X
each equal to the corresponding entry in Xi+1 i+1 ∈ g(Vi ).

2.3.4 Statement of the Main Theorem

We can now formulate our key theorem, which implies that the variety OD is

normal provided that Z is nonsingular in codimension 1, which is not true for all

nilpotent D ∈ g(V ). Specific instances where Z is normal will be discussed in

Chapter 5.

Theorem 2.13. Let G(V ) be a symplectic or orthogonal group for a finite dimen-

sional vector space V over a field K of prime characteristic p 6= 2. For a nilpotent

element D ∈ g(V ), construct M and Z as in sections 2.3.1 and 2.3.2 above.

(a) The affine scheme Z is reduced, hence Z is an affine variety. Moreover, Z is

a complete intersection in M with respect to the equations (∗).

(b) The map ϑ : Z → OD given by ϑ(X1 , . . . , Xt ) = X1∗ X1 is G(V0 ) × . . . × G(Vt )-

equivariant and a G(V1 ) × . . . × G(Vt )-quotient map.

The proof of part (a) is in section 3.4.3. It follows the first half of the proof

of the analogous result in section 5.5 of [KP82], which relies on the characteristic

of K insofar as it relies on Proposition 7.1 of [KP82], proven there for char K = 0

only. We generalize this result in Proposition 3.10, which is proven in section 3.3

for char K 6= 2 after a lengthy discussion on orthosymplectic nilpotent pairs and

associated cocharacters.

Part (b) follows from Theorem 4.22(b) in section 4.3.

24
Chapter 3

Part A

We begin by stating the classification of orbits of nilpotent pairs of linear maps

under the general linear group, followed by a careful treatment of their classifica-

tion under the action of a symplectic and an orthogonal groups. The explicit de-

scription of these orthosymplectic nilpotent pairs is then used to show that cochar-

acters
 : K × → GL(U ⊕ V ) exist associated to nilpotent elements of the form
φ 

 0 X
  ∈ End(U ⊕ V ), which naturally identify with the nilpotent orthosym-
X∗ 0
plectic pairs (X ∗ , X), where V a ε-space and U a −ε-space. These associated cochar-

acters are then used in the generalized proof of Proposition 3.10 relating the dimen-

sion of the orthosymplectic orbits to the dimension of the associated orthogonal and

symplectic orbits. We finish by analyzing the dimension of the preimage of O ⊂ OD

in Z, concluding that Zsing has codimension at least 1, and that Z is a complete

intersection, hence an affine variety, proving part (a) of Theorem 2.13.

3.1 Orthosymplectic Orbits

We digress for a moment in order to introduce the concept of a nilpotent pair.

After establishing the basic notation, we will limit our focus to orthosymplectic

nilpotent pairs.

25
3.1.1 Nilpotent Pairs

Let U and V be two finite dimensional vector spaces over K. Denote as before

L(V, U ) = Hom(V, U ) and consider the product L := L(U, V ) × L(V, U ) of pairs

A /
of maps U o V . Define the action of GL(U ) × GL(V ) on L by (g, h)(A, B) =
B
−1 −1
(hAg , gBh ), g ∈ GL(U ), h ∈ GL(V ). The theory of orbits and the invariant

theory for this representation are known [DF73, Naz73, Gab75, KR71, Hes76]. Such

pairs can be thought of as objects in a category of modules over a suitable K-

algebra, with objects P = {(A, B), (U, V )} and morphisms P → P ′ given by a pair

of linear maps φU : U → U ′ and φV : V → V ′ such that A′ ◦ φU = φV ◦ A and

B ′ ◦ φV = φU ◦ B, where P ′ = {(A′ , B ′ ), (U ′ , V ′ )}. The classification is through

indecomposable modules, in much the same way as one classifies endomorphisms

of vector spaces via Jordan blocks. This is explored in more detail in Appendix

A. We are specifically interested in those pairs (A, B) for which the composition

BA ∈ End U (or equivalently AB ∈ End V ) is nilpotent, calling such a pair a

nilpotent pair.

We classify the indecomposable nilpotent pairs by indecomposable ab-diagrams.

One such is

) & & )
• • • • • • • • •
M2n+1 : an+1 g an
...... a3
d a2 e a1 b1 b2 b3
...... bn

meaning that U has basis {a1 , a2 , . . . , an+1 } and V has basis {b1 , b2 , . . . , bn } and

Aai = bi , Bbj = aj+1 . The diagrams are referred to by a string of a’s and b’s, hence

the name ab-diagram. Type M2n+1 above is indicated by the string ababab . . . ba

where the number of a’s is n+1 and the number of b’s is n. Another indecomposable

diagram is

26
' ( *
• h ...... • • • • • • •
N2n+1 : an a2 b a1
c
b1 b2 b3
...... bn bn+1

where U has basis {a1 , a2 , . . . , an }, V has basis {b1 , b1 , . . . , bn+1 }, and Aai = bj+1 ,

Bbj = aj . N2n+1 is denoted by the string bababa . . . ab with n a’s and n + 1 b’s.

For indecomposable nilpotent pairs (A, B), there are two other indecomposable

ab-diagrams under the action of GL(U ) × GL(V ), M2n = ababab . . . ab and N2n =

bababa . . . ba, defined similarly to M2n+1 and N2n+1 .

An arbitrary nilpotent pair (A, B) is a direct sum of indecomposables, and so it

is determined by a finite set of indecomposable ab-diagrams to which we associate

an array with each row an ab-string. We call this array the ab-diagram of the pair.

For example a pair may have ab-diagram

ababababa
τ = ababa

bab

consisting of three indecomposable ab-diagrams of types M9 , M5 , N3 .

3.1.2 Young Diagrams of Nilpotent Pairs

The nilpotent endomorphisms BA ∈ End U and AB ∈ End V necessarily have

associated Young diagrams. Given a nilpotent pair (A, B) and its ab-diagram τ , it

is simple to recover the Young diagrams of BA and AB as follows: for the diagram

of BA suppress all the b’s in the ab-strings of τ , and replace the a’s by boxes. For

the diagram of AB, suppress the a’s in the ab-strings of τ . For example, if the pair

(A, B) has ab-diagram τ from above, then BA has Young diagram

aaaaa

aaa =

and AB has Young diagram

27
bbbb

bb =

bb

Remark 3.1. Not all pairs of Young diagrams describing a nilpotent orbit in End U

and one in End V are associated to some nilpotent pair. Furthermore, there can be

different nilpotent pairs giving rise to the same pair of Young diagrams. In other

words, the “map” taking ab-diagrams to Young diagrams is neither one-to-one, nor

onto.

Remark 3.2. For a nilpotent pair (A, B) with ab-diagram τ , we have that

(i) A is injective if and only if every ab-string in τ ends with b.

(ii) A is surjective if and only if every ab-string in τ starts with a.

(iii) B is injective if and only if every ab-string in τ ends with a.

(iv) B is surjective if and only if every ab-string in τ starts with b.

Indeed, if A is injective, ker A = 0, so Aai 6= 0 for all basis elements ai ∈ U , and

so no string can end with an a. Conversely, if an ab-string ends with a, then there

exists some ai ∈ U such that Aai = 0, and A is not injective. If A is surjective,

each b in the ab-string corresponds to a basis element bi ∈ V , which must be the

image under A of some ai ∈ U , which can be taken to be an element of a basis for

U . Hence there must be an a preceding each b, and each ab-string in τ starts with

an a. If each ab-string in τ starts with a, each basis element bi of V is the image of

some basis element ai ∈ U , and so each v ∈ V is the image of a linear combination

of the ai , and A is surjective. The same argument applies to B reversing a’s and

b’s.

3.1.3 Orthosymplectic Nilpotent Pairs

Assume for the remainder of this section that U is an orthogonal space, V is a

symplectic space and let X ∈ L(V, U ). From now on we restrict our attention to

28
Table 3.1: Indecomposable Orthosymplectic ab-Diagrams
Type αn βn γn δn ǫn
aba . . . ba bab . . . ab aba . . . ab
ab-diagram aba . . . ba bab . . . ab aba . . . ba bab . . . ab bab . . . ba
n — — odd even —
#a 2n + 1 2n − 1 2(n + 1) 2n 2n
#b 2n 2n 2n 2(n + 1) 2n

nilpotent pairs of the form (A, B) = (X ∗ , X) where X ∗ is the adjoint of X as in

section 2.2.1. The action of GL(U ) × GL(V ) above restricts to an action of G(U ) ×

G(V ) = O(U ) × Sp(V ) on (X ∗ , X) ∈ L = L(U, V ) × L(V, U ). So (g, h)(X ∗ , X) =

(hX ∗ g−1 , gXh−1 ). We have then that if OX denotes the O(U ) × Sp(V )-orbit of

(X ∗ , X) and PX denotes the GL(U ) × GL(V )-orbit of (X ∗ , X), then

PX ∩ L = OX .

Hence the orbit OX is determined by the ab-diagram of the pair as defined above.

We call nilpotent pairs of the form (A, B) = (X ∗ , X) orthosymplectic nilpotent

pairs and the O(U ) × Sp(V )-orbit OX of such a pair an orthosymplectic orbit. We

saw in section 2.2.1 that π(X) = XX ∗ ∈ g(U ) = so(U ) and ρ(X) = X ∗ X ∈ g(V ) =

sp(V ), hence the Mr and Nr no longer classify the orthosymplectic orbits. In partic-

ular, the Young diagrams of the compositions π(X) and ρ(X), denoted by π(τ ) and

ρ(τ ) respectively, must satisfy Theorem 2.2. Such an ab-diagram is then called an

orthosymplectic ab-diagram. The remainder of this section gives an explanation and

proof of existence of the five types of indecomposable orthosymplectic ab-diagrams,

summarized in Table 3.1. The proof is inspired by the treatment in [Jan04] of the

classification of nilpotent orbits in the symplectic and orthogonal groups.

3.1.4 Type αn

The irreducible string M2r+1 = ababab . . . ba is an orthosymplectic ab-diagram

only if r is even. Otherwise, for r odd, π(M2r+1 ) = aaa . . . a and ρ(M2r+1 ) = bbb . . . b

(an even number r + 1 of a’s and an odd number r of b’s) do not satisfy Theorem

29
2.2. We define then the indecomposable orthosymplectic ab-diagram of type αn for

n ∈ Z+ to be the string ababa . . . ba of 2n + 1 a’s and 2n b’s. The αn are those

M2r+1 where r = 2n is even.

We would like to see that such a string exists for each n, i.e. that there exists

a pair (X ∗ , X) with ab-diagram αn for each n. Let U be a 2n + 1 dimensional

vector space and V a 2n dimensional vector space, both over K, and choose bases

{e1 , e2 , . . . , e2n+1 } and {f1 , f2 , . . . f2n } of U and V respectively. Define a linear map

X : V → U by X(fi ) = ei+1 and a map X ′ : U → V by X ′ (ej ) = fj . The pair

(X ′ , X) ∈ L is clearly nilpotent and has ab-diagram αn . We would like to see that

(X ′ , X) is an orthosymplectic nilpotent pair.  

0 X
e the endomorphism of U ⊕ V given by 
Denote by X  , i.e. correspond-
X′ 0
ing to the pair (X ′ , X). The 4n + 1 dimensional vector space U ⊕ V has basis
e nota-
{e1 , X ′ (e1 ), XX ′ (e1 ), . . . , X ′ (XX ′ )2n (e1 )} which we will rewrite using the X
e X
tion as {e, Xe, e 2 e, . . . , X
e 4n e} where e = e1 viewed as an element of U ⊕ V . On

U ⊕ V , define a nondegenerate bilinear form β by






 (−1)j/2 j, h even, j + h = 4n,


e j e, X
β(X e h e) = (−1)(j+1)/2 j, h odd, j + h = 4n,






0 otherwise.

On U , i.e. for j and h even, β restricts to a nondegenerate form, which is

e j e, X
symmetric. We have β(X e h e) = 0 unless j + h = 4n. Assuming j + h = 4n with

j and h both even, we have

e j e, X
β(X e h e) = (−1)j/2 = (−1)(4n−h)/2 = (−1)h/2 = β(X
e h e, X
e j e).

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U .

On V , i.e. for j and h odd, β restricts to a nondegenerate form, which is

e j e, X
alternating. We have β(X e h e) = 0 unless j + h = 4n. Therefore assuming that

30
j + h = 4n with j, h odd,

e j e, X
β(X e h e) = (−1)(j+1)/2 = (−1)(4n−h+1)/2 = (−1)(h−1)/2

e h eX
= −(−1)(h+1)/2 = −β(X e j e).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .

It remains to be seen that X ′ is adjoint to X. This is a straightforward calcu-

eX
lation once we recall that for odd powers i, X e ie = X X
e i e and for even powers i,

eX
X e ie = X ′X
e i e. We have then for j odd, h even and j + h = 4n − 1,

e j e, X
β(X X e h e)|U = β(X
eXe j e, X
e h e)|U = (−1)(j+1)/2

e j e, X
= β(X eXe h e)|V = β(X
e j e, X ′ X
e h e)|V

with the left- and right-most terms equal to zero otherwise. Hence X ′ = X ∗ and

(X ∗ , X) is an orthosymplectic nilpotent pair with ab-diagram αn .

3.1.5 Type βn

The irreducible string N2r+1 = bababa . . . ab is an orthosymplectic ab-diagram

only if r is odd, with π(N2r+1 ) = aaa . . . a and ρ(N2r+1 ) = bbb . . . b of odd and even

lengths respectively. This leads us to define the indecomposable orthosymplectic

ab-diagram of type βn to be the string babab . . . ab of 2n − 1 a’s and 2n b’s (N2r+1

where r = 2n − 1).

We now show that such strings exist. Let U be a 2n − 1 dimensional vector space

and let V be a 2n dimensional vector space over K, {e1 , e2 , . . . , e2n−1 } a basis for U

and {f1 , f2 , . . . , f2n } a basis for V . Define a linear map X : V → U by X(fi ) = ei

and X ′ : U → V by X ′ (ej ) = fj+1 . The pair (X ′ , X) is nilpotent and has ab-diagram

βn . As in section 3.1.4, we show that U is an orthogonal space, V is a symplectic

space, and that X ′ is the adjoint of X, hence that we have an orthosymplectic pair

with ab-diagram βn .

The 4n− 1 dimensional e X

space U ⊕ V has basis {f = f1 , Xf, e 2 f, . . . , X
e 4n−2 f }

0 X
e=
with X   as in section 3.1.4. Define the nondegenerate bilinear form β
X′ 0

31
by 




 (−1)(j−1)/2 j, h odd, j + h = 4n − 2


e j v, X
β(X e h v) = (−1)j/2 j, h even, j + h = 4n − 2






0 otherwise,

On U , where j and h are odd, β restricts to a nondegenerate form, which is

e j f, X
symmetric. Indeed, β(X e h f ) = 0 unless j +h = 4n−2. Assuming j +h = 4n−2

with j and h both odd, we have

e j f, X
β(X e h f ) = (−1)(j−1)/2 = (−1)(4n−2−h−1)/2 = (−1)(−h−3)/2 = (−1)(h+3)/2−2 =

e h f, X
(−1)(h−1)/2 = β(X e j f ).

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U.

On V , where j and h are even, β restricts to a nondegenerate form, which is

e j f, X
alternating. Indeed, β(X e h f ) = 0 unless j + h = 4n − 2. Therefore assuming

that j + h = 4n − 2 with j, h even,

e j f, X
β(X e h f ) = (−1)j/2 = (−1)(4n−2−h)/2 = (−1)(h+2)/2

e h f, X
= −(−1)h/2 = −β(X e j f ).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .
e applied to X
Here we have that for even powers i, X e i f is the map X ∈ L(V, U ),

e applied to X
and for odd powers i, X e i f is X ′ ∈ L(U, V ). Then for j even and h

odd and j + h = 4n − 3,

e j f, X
β(X X e h f )|U = β(X
eXe j f, X
e h f )|U = (−1)(j+1−1)/2 = (−1)j/2 =

e j f, X
β(X eXe h f )|V = β(X
e j f, X ′ X
e h f )|V

with the right- and left-most terms equal to zero otherwise. Hence X ′ = X ∗ and

(X ∗ , X) is an orthosymplectic nilpotent pair with ab-diagram βn .

32
3.1.6 Type γn

If n is odd, the ab-diagram consisting of two strings of type M2n+1 is an or-

aba...ba
thosymplectic diagram. Indeed, let τ = aba...ba , each row having n + 1 a’s and n b’s.
aa...a
Then π(τ ) = aa...a has all odd-length rows occurring an even number of times and
bb...b
ρ(τ ) = bb...b has all even-length rows occurring an even number of times. Since the

ab-diagram consisting of just one M2n+1 is not orthosymplectic for n odd as shown
aba...ba
in section 3.1.4, γn = aba...ba where the number of a’s is 2(n + 1) and the number

of b’s is 2n is an indecomposable orthosymplectic ab-diagram for n an odd, positive

integer.

To see that γn exists as the ab-diagram of an orthosymplectic nilpotent pair for

n odd, let U be a 2(n + 1) dimensional vector space over K with basis

{e1 , e2 , . . . , en+1 , e′1 , e′2 , . . . , e′n+1 } and let V be a 2n dimensional vector space over

K with basis {f1 , f2 , . . . , fn , f1′ , f2′ , . . . , fn′ }. Define a linear map X : V → U by

X(fi ) = ei+1 and X(fi′ ) = e′i+1 and define X ′ : U → V by X ′ (ej ) = fj and

X ′ (e′j ) = fj′ . The pair (X ′ , X) is nilpotent and has ab-diagram γn . We show that

(X ′ , X) is orthosymplectic.

U ⊕ V is a 4n + 2 dimensional vector space with basis

{e1 , X ′ (e1 ), XX ′ (e1 ), . . . , (XX ′ )n (e1 ), e′1 , X ′ (e′1 ), XX ′ (e′1 ), . . . , (XX ′ )n (e′1 )}

 
0 X
e =
which we rewrite letting X  , u1 = e1 and u2 = e′1 as
X′ 0

e 1, X
{u1 , Xu e 2 u1 , . . . , X
e 2n u1 , u2 , Xu
e 2, X
e 2 u2 , . . . , X
e 2n u2 }.

Define the nondegenerate bilinear form β by






 (−1)j/2+i j, h even, j + h = 2n, i 6= i∗ ∈ {1, 2},


e j ui , X
β(X e h ui∗ ) = (−1)(j+1)/2+i j, h odd, j + h = 2n, i 6= i∗ ∈ {1, 2},






0 otherwise.

33
e j u1 , X
In particular, β(X e h u1 ) = 0 = β(X
e j u2 , X
e h u2 ) for all j and h. The condition

i 6= i∗ is equivalent to requiring i − i∗ ≡ 1 mod 2 since i, i∗ ∈ {1, 2}.

On U , i.e. for j and h even, β restricts to a nondegenerate form, which is

e j ui , X
symmetric. We have β(X e h ui∗ ) = 0 unless j + h = 2n and i 6= i∗ . Assuming

j + h = 2n with j and h both even, i − i∗ ≡ 1 mod 2 and recalling that n is odd, we

have

e j ui , X
β(X e h ui∗ ) = (−1)j/2+i = (−1)(2n−h)/2+1+i∗ = (−1)h/2+i∗ +n+1 = (−1)h/2+i∗ =

e h ui ∗ , X
β(X e j ui ).

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U .

On V , i.e. for j and h odd, β restricts to a nondegenerate form, which is

e j ui , X
alternating. We have β(X e h ui∗ ) = 0 unless j + h = 2n and i 6= i∗ . Therefore

assuming that j + h = 2n with j, h odd and i − i∗ ≡ 1 mod 2,

e j ui , X
β(X e h ui∗ ) = (−1)(j−1)/2+i = (−1)(2n−h−1)/2+1+i∗ = (−1)(h+1)/2+n+1+i∗ =

e h ui∗ , X
−(−1)(h−1)/2+i∗ = −β(X e j ui ).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .

We need now that X ′ is adjoint to X. In order to see this we unpack some

e k ui is an element of V and we have X
of our identifications. For k odd, X e k u1 =

e k u2 = X ′ (XX ′ )(k−1)/2 e′ . Thus X

X ′ (XX ′ )(k−1)/2 e1 and X eXe k ui = X X
e k ui since X
e
1

e k ui is an element of U with X
acts on V by X. For k even, X e k u1 = (XX ′ )k/2 f1 and

X eX
e k u2 = (XX ′ )k/2 f ′ , hence X e k ui = X ′ X
e k ui .
1

For j odd, h even and j + h = 2n − 1,

e j ui , X
β(X X e h ui∗ )|U = β(X
eXe j ui , X
e h ui∗ )|U = (−1)(j+1)/2+i

e j ui , X
= β(X eXe h ui∗ )|V = β(X
e j ui , X ′ X
e h ui∗ )|V .

The left- and right-most terms are equal to zero otherwise, and so X ′ = X ∗ .

34
3.1.7 Type δn

bab...ab
If n is even, the ab-diagram δn = bab...ab consisting of two strings of type N2n+1

(2n a’s and 2(n + 1) b’s) is an indecomposable orthosymplectic ab-diagram. Indeed,

aa...a bb...b
π(δn ) = aa...a has an even number of rows of even length n, and ρ(δn ) = bb...b has

an even number of rows of odd length n + 1. Since the ab-diagram consisting of

just one N2n+1 is not orthosymplectic for n odd as shown in section 3.1.5, δn is an

indecomposable orthosymplectic ab-diagram for n an even, positive integer.

We construct an orthosymplectic nilpotent pair with associated ab-diagram δn for

n even. Let U be a 2n dimensional and V a 2(n + 1) dimensional vector space over

K, with bases {e1 , e2 , . . . , en , e′1 , e′2 , . . . , e′n } and {f1 , f2 , . . . , fn+1 , f1′ , f2′ , . . . , fn+1
′ }

respectively. Define a linear map X : V → U by X(fi ) = ei and X(fi′ ) = e′i ,

′ . The pair (X ′ , X) is
and define X ′ : U → V by X ′ (ej ) = fj+1 and X ′ (e′j ) = fj+1

nilpotent and has ab-diagram δn . We show that (X ′ , X) is orthosymplectic.

U ⊕ V is a 4n + 2 dimensional vector space with basis

{f1 , X(f1 ), X ′ X(f1 ), . . . , (X ′ X)n (f1 ), f1′ , X(f1′ ), X ′ X(f1′ ), . . . , (X ′ X)n (f1′ )}.

 
0 X
e =
Letting v1 = f1 and v2 = f1′ and writing X  , this basis becomes
X′ 0

e 1, X
{v1 , Xv e 2 v1 , . . . , X
e 2n v1 , v2 , Xv
e 2, X
e 2 v2 , . . . , X
e 2n v2 }.

Define the nondegenerate bilinear form β by




(−1)(j−1)/2+i
 j, h odd, j + h = 2n, i 6= i∗ ∈ {1, 2},



e j e h
β(X vi , X vi∗ ) = (−1)j/2+i j, h even, j + h = 2n, i 6= i∗ ∈ {1, 2},






0 otherwise.

On U , where j and h are odd, β restricts to a nondegenerate form, which is

e j vi , X
symmetric. Indeed, β(X e h vi∗ ) = 0 unless j + h = 2n and i 6= i∗ (equivalently

35
i − i∗ ≡ 1 mod 2). Assuming j + h = 2n with j and h both odd, i − i∗ ≡ 1 mod 2,

and recalling that n is even, we have

e j vi , X
β(X e h vi∗ ) = (−1)(j−1)/2+i = (−1)(2n−h−1)/2+i∗ +1

e h vi∗ , X
= (−1)(h+1)/2+i∗ +1 = (−1)(h−1)/2+i∗ = β(X e j vi )

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U.

On V , where j and h are even, β restricts to a nondegenerate form, which is

e j vi , X
alternating. Indeed, β(X e h vi∗ ) = 0 unless j + h = 2n and i − i∗ ≡ 1 mod 2.

Therefore assuming these,

e j vi , X
β(X e h vi∗ ) = (−1)j/2+i = (−1)(2n−h)/2+1+i∗

e h vi ∗ , X
= (−1)h/2+n+1+i∗ = −(−1)h/2+i∗ = −β(X e j vi ).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .
eX
We show now that X ′ is adjoint to X. For k even, X e k vi = X X
e k vi since

e k vi ∈ V . For k odd, X
X eXe k vi = X ′ X
e k vi since X
e k vi ∈ U . Thus for j even and h

odd and j + h = 2n − 1,

e j ui , X
β(X X e h ui∗ )|U = β(X
eXe j ui , X
e h ui∗ )|U = (−1)(j−1+1)/2+i = (−1)j/2+i =

e j ui , X
β(X eXe h ui∗ )|V = β(X
e j ui , X ′ X
e h ui∗ )|V .

The left- and right-most terms are equal to zero otherwise, and so X ′ = X ∗ and

(X ′ , X) = (X ∗ , X) is orthosymplectic.

3.1.8 Type ǫn

The ab-strings of type M2n and N2n present more of a challenge. Neither is

orthosymplectic if taken by itself as an ab-diagram τ , since then either π(τ ) or ρ(τ )

will violate Theorem 2.2. Indeed,

- if n is even and τ = M2n or τ = N2n , then π(τ ) has one row of even length.

- if n is odd and τ = M2n or τ = N2n , then ρ(τ ) has one row of odd lenth.

36
While they do satisfy the criteria for being orthosymplectic, ab-diagrams with two

M2n ’s or two N2n ’s do not occur as the ab-diagrams of orthosymplectic nilpotent

pairs. To see this, assume the opposite and let (X ∗ , X) be an orthosymplectic

abab...ab
nilpotent pair with ab-diagram τ = abab...ab (consisting of two M2n ’s). By remark

3.2, X ∗ : U → V is both injective and surjective. By the definition of M2n , there

exist a basis

{u1 , XX ∗ u1 , . . . , (XX ∗ )n−1 u1 , u2 , XX ∗ u2 , . . . , (XX ∗ )n−1 u2 }

of U and a basis

{X ∗ u1 , X ∗ (XX ∗ )u1 , . . . , X ∗ (XX ∗ )n−1 u1 , X ∗ u2 , X ∗ (XX ∗ )u2 , . . . , X ∗ (XX ∗ )n−1 u2 }

of V . Then

hX ∗ (XX ∗ )n−1 ui , X ∗ (u)i|V = hX(X ∗ (XX ∗ )n−1 ui ), ui|U = h0, ui|U = 0

for all u ∈ U , since by the definition of M2n , (XX ∗ )n ui = 0 for i = 1, 2. Hence

hX ∗ (XX ∗ )n−1 ui , X ∗ (u)i|V = 0

for all u ∈ U and

hX ∗ (XX ∗ )n−1 ui , vi|V = 0

for all v ∈ V since X ∗ is surjective. The form is nondegenerate, which forces

X ∗ (XX ∗ )n−1 ui = 0. This contradicts the existence of the above basis for V , and so

(X ∗ , X) cannot have ab-diagram τ . The same argument can be used to show that

there is no orthosymplectic pair with ab-diagram consisting of two N2n ’s, reversing

the roles of U and V . (In this case, X : V → U is injective and surjective.)

Finally, the ab-diagrams consisting of one M2n and one N2n are orthosymplectic

and indecomposable, and also exist as the ab-diagrams of orthosymplectic nilpotent

aba...ab
pairs. Let ǫn = bab...ba for n ∈ Z+ with the number of a’s and b’s both equal to 2n.

37
Then π(ǫn ) and ρ(ǫn ) both satisfy Theorem 2.2 for n even and odd. However, we

must prove existence for the cases n even and n odd separately.

For arbitrary n ∈ Z+ take U and V to be 2n dimensional vector spaces over K

with respective bases {e1 , e2 , . . . , en , e′1 , e′2 , . . . , e′n } and {f1 , f2 , . . . , fn , e′1 , f2′ , . . . , fn′ }.

Define X : V → U by X(fi ) = ei+1 and X(fi′ ) = e′i and define X ′ : U → V by

X ′ (ei ) = fi and X ′ (e′i ) = fi+1

′ (note the antisymmetry not present in previous

cases). Letting w1 = f1 ∈ U and w2 = e′1 ∈ V , U ⊕ V has basis

e 1, X
{w1 , Xw e 2 w1 , . . . , X
e 2n−1 w1 , w2 , Xw
e 2, X
e 2 w2 , . . . , X
e 2n−1 w2 }

 
0 X
where Xe = 

 ∈ End(U ⊕ V ).
X′ 0
First, let n be even. We define nondegenerate bilinear form β by





 (−1)(j+i−1)/2 j + i, h + i∗ odd, j + h = 2n − 1, i 6= i∗


e j wi , X
β(X e h wi∗ ) = (−1) (j+i)/2 j + i, h + i∗ even, j + h = 2n − 1, i 6= i∗






0 otherwise.

On U , i.e. for j + i and h + i∗ odd, β restricts to a nondegenerate form, which

e j wi , X
is symmetric. We have β(X e h wi∗ ) = 0 unless j + h = 2n − 1 and i 6= i∗

(equivalently i − i∗ ≡ 1 mod 2). We first observe that if i − i∗ ≡ 1 mod 2 with

i, i∗ ∈ {1, 2}, then i ≡ −1 − i∗ mod 4 (this is easily checked explicitly for all possible

combinations of i and i∗ ). Assuming then j + h = 2n − 1 with j + i and h + i∗ both

odd, i ≡ −1 − i∗ mod 4, and recalling that n is even, we have

e j wi , X
β(X e h wi∗ ) = (−1)(j+i−1)/2 = (−1)(2n−1−h−i∗ −1−1)/2 = (−1)(2n−3−h−i∗ )/2 =

e h wi∗ , X
(−1)(h+i∗ +3)/2−2 = (−1)(h+i∗ −1)/2 = β(X e j wi ).

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U.

On V , i.e. for j + i and h + i∗ even, β restricts to a nondegenerate form, which is

e j wi , X
alternating. We have β(X e h wi∗ ) = 0 unless j +h = 2n−1 and i−i∗ ≡ 1 mod 2.

38
Therefore assuming these and i ≡ −1 − i∗ mod 4 as above,

e j wi , X
β(X e h wi∗ ) = (−1)(j+i)/2 = (−1)(2n−1−h−1−i∗ )/2 = (−1)(−h−i∗ −2)/2 =

e h wi∗ , X
−(−1)(h+i∗ )/2 = −β(X e j wi ).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .
e acts by X ′ , while on V (j + i and h + i∗ even),
On U (j + i and h + i∗ odd), X
e acts by X. Thus for j + i even and h + i∗ odd and j + h = 2n − 2,
X

e j wi , X
β(X X e h wi∗ )|U = β(X
eXe j wi , X
e h wi∗ )|U = (−1)(j+i−1+1)/2 = (−1)(j+i)/2 =

e j wi , X
β(X eXe h wi∗ )|V = β(X
e j wi , X ′ X
e h wi∗ )|V .

The left- and right-most terms are equal to zero otherwise, and so X ′ = X ∗ and

(X ′ , X) = (X ∗ , X) is orthosymplectic.

Lastly, let n be odd, and define nondegenerate bilinear form β by




(−1)(j−i−1)/2
 j + i, h + i∗ odd, j + h = 2n − 1, i 6= i∗



e j e h
β(X wi , X wi∗ ) = (−1)(j−i)/2 j + i, h + i∗ even, j + h = 2n − 1, i 6= i∗






0 otherwise.

On U , i.e. for j + i and h + i∗ odd, β restricts to a nondegenerate form, which

e j wi , X
is symmetric. We have β(X e h wi∗ ) = 0 unless j + h = 2n − 1 and i 6= i∗

(equivalently i − i∗ ≡ 1 mod 2). As in the previous case, i − i∗ ≡ 1 mod 2 with

i, i∗ ∈ {1, 2} implies i ≡ −1 − i∗ mod 4. Assuming then j + h = 2n − 1 with j + i

and h + i∗ both odd, i ≡ −1 − i∗ mod 4, and recalling that n is odd, we have

e j wi , X
β(X e h wi∗ ) = (−1)(j−i−1)/2 = (−1)(2n−1−h+1+i∗ −1)/2 = (−1)(−h+i∗ −1)/2+n =

e h wi∗ , X
−(−1)(h−i∗ +1)/2 = (−1)(h−i∗ −1)/2 = β(X e j wi ).

Hence U is an orthogonal space with nondegenerate bilinear form given by the

restriction of β to U.

On V , i.e. for j + i and h + i∗ even, β restricts to a nondegenerate form, which is

e j wi , X
alternating. We have β(X e h wi∗ ) = 0 unless j +h = 2n−1 and i−i∗ ≡ 1 mod 2.

Therefore assuming these and i ≡ −1 − i∗ mod 4 as above,

39
e j wi , X
β(X e h wi∗ ) = (−1)(j−i)/2 = (−1)(2n−1−h+1+i∗ )/2 = (−1)(−h+i∗ )/2+n =

e h wi∗ , X
−(−1)(h−i∗ )/2 = −β(X e j wi ).

Hence V is a symplectic space with nondegenerate bilinear form given by the re-

striction of β to V .
e acts by X ′ , while on V (j + i and h + i∗ even),
On U (j + i and h + i∗ odd), X
e acts by X. Thus for j + i even and h + i∗ odd and j + h = 2n − 2,
X

e j wi , X
β(X X e h wi∗ )|U = β(X
eXe j wi , X
e h wi∗ )|U = (−1)(j−i−1+1)/2 = (−1)(j−i)/2 =

e j wi , X
β(X eXe h wi∗ )|V = β(X
e j wi , X ′ X
e h wi∗ )|V .

The left- and right-most terms are equal to zero otherwise, and so X ′ = X ∗ and

(X ′ , X) = (X ∗ , X) is orthosymplectic.

3.2 Associated Cocharacters

If char K = 0, one can use a semisimple element H ∈ g to define a grading of g

by letting gi be the subset of all Z ∈ G such that [H, Z] = iZ. When char K > 0,

this method of defining a grading clearly breaks down. One can instead use a

cocharacter, i.e. a homomorphism of algebraic groups φ : K × → G, and define

gi = g(i; φ) = {Z ∈ g | Ad(φ(t))Z = ti Z for all t ∈ K × }. Just as the H arising as

semisimple elements of sl2 -triples have certain properties that make them “nice” to

work with, some cocharacters have analogous properties that make them preferable

to others. Such cocharacters are called associated cocharacters and will be explored

in this section.

3.2.1 Definition and Properties

Let G be a connected reductive algebraic group with Lie algebra g. In order

to define an associated cocharacter, we first need to define what it means for a

nilpotent element in g to be distinguished. Let GX = {Z ∈ G | ZX = XZ} be the

centralizer of X in G. By a torus we mean an algebraic subgroup isomorphic to a

direct product of copies of K × . A nilpotent element X ∈ g is called distinguished

40
if each torus contained in GX is contained in the center of G. Equivalently, X is

distinguished if each homomorphism K × → GX of algebraic groups takes values in

the center of G. We will need the following result from [Jan04, 4.1] classifying the

distinguished nilpotent elements of gl(V ).

Lemma 3.3. Let X ∈ gl(V ) be nilpotent. Then X is distinguished if and only if X

has partition [dim V ].

For a nilpotent element X ∈ g, a cocharacter φ : K × → G is associated to X if

X ∈ g(2; φ) = {Z ∈ g | Ad(φ(t))(Z) = t2 Z for all t ∈ K × }

and if there exists a Levi subgroup L in G such that X is a distinguished nilpotent

element in Lie L and such that φ(K × ) ⊂ DL, where DL is the derived group of

L [Jan04, 5.3]. By a Levi subgroup of G we mean a Levi factor of a parabolic

subgroup of G as in [Hum75, 30.2], i.e. the centralizer in G of a maximal torus

of the radical of a parabolic subgroup P ⊆ G. For G = GL(V ), the subgroups

L = GL(V1 ) × GL(V2 ) × . . . × GL(Vr ) are the Levi subgroups in G, with V =

V1 ⊕V2 ⊕. . .⊕Vr a direct sum decomposition and dim Vi = di , d1 ≥ d2 ≥ . . . ≥ dr > 0

a partition of dim V [Jan04, 4.4].

Let X ∈ g be nilpotent and let τ be a cocharacter associated to X. Associate to

τ a parabolic subgoup P = Pτ such that

M M
p := Lie(Pτ ) = gi = g(i; τ ).
i≥0 i≥0

One possible definition of Pτ is as the set of all g ∈ G such that limt→0 τ (t)gτ (t)−1

exists [Spr98, 8.4.5]. p ⊆ g is a parabolic subalgebra with nilpotent radical n :=

M M
gi and Levi-decomposition p = g0 ⊕n [Jan04, 5.1]. Define n2 := gi .
i>0 i≥2

Proposition 3.4. Suppose that char K is good for G. Then

(a) The group P depends only on X, not on the choice of τ .

(b) GX = PX .

41
(c) Ad(P )(X) = n2 .

Proof. This is Proposition 5.9 in [Jan04].

Proposition 3.5. Two cocharacters associated to X are conjugate under G0X , the

identity component of the centralizer of X.

Proof. This is Lemma 5.3b in [Jan04].

L
Proposition 3.6. X ∈ n2 = i≥2 gi and the map ad X : p → n2 is surjective.

Proof. The first part follows from the definition of an associated cocharacter, and

the second assertion from the proof of Prop. 5.9c in [Jan04].

Let OX be the conjugagy class of X in g. Propositions 3.4 and 3.6 imply that

G ×P n2 → OX is a resolution of singularities [Jan04, 6.10 and 8.7], and that

dim OX = dim n + dim n2 .

3.2.2 Results in the Orthosymplectic Setting

Return to the setting of section 3.1.3 where U is an orthogonal space, V is

a symplectic space and X ∈ L(V, U ). Let G e denote the group of invertible linear
 
0 X
e be the element 
transformations of the sum U ⊕V and let X   ∈ End(U ⊕V )
X∗ 0
for X ∈ L(V, U ) such that (X ∗ , X) is a nilpotent
  pair.
a b
e is nilpotent. Indeed, if 
Note that X   ∈ End(U ⊕ V ),
c d
    
 0 X  a b   Xc Xd 
  = 
X∗ 0 c d X ∗a X ∗b

and  2    
∗ ∗
 0 X  a b  XX a XX b 
   = .
X∗ 0 c d X ∗ Xc X ∗ Xd
 2  
 0 X  XX ∗ 0 
Hence   =  . Since we are assuming that (X ∗ , X) is a
X∗ 0 0 ∗
X X

42
 2l    
 0 X  (XX ∗ )l 0  0 0
nilpotent pair, we have   = =  for some
X∗ 0 0 ∗
(X X)l 0 0
suitably large l ∈ Z+ .

Theorem 3.7. There is a cocharacter φ : K × → Ge associated to nilpotent X

e =

 0 X e = GL(U ⊕ V ) with φ(t) ∈ O(U ) × Sp(V ) for all t ∈ K × .
  in G
X ∗ 0

To prove this, we utilize the classification via ab-diagram of the pairs (X ∗ , X) ∈

e We begin by showing that
L(U, V ) × L(V, U ), which naturally identify with X.
e = (X ∗ , X) with indecomposable orthosymplectic
such a cocharacter exists for X

ab-diagram as given in Table 3.1. We go on to show that U ⊕ V can be decomposed

e restricted to each component corresponds to a pair with
in such a way that X
e exists
indecomposable ab-diagram, and conclude that a cocharacter associated to X

by combining the cocharacters associated to the restrictions.

e = (X ∗ , X) with inde-
Lemma 3.8. A cocharacter as in Theorem 3.7 exists for X

composable orthosymplectic ab-diagram.

Proof. We examine each indecomposable diagram in turn.

e have ab-diagram αn = abab . . . ba where the number of a’s is
Case 1. Let X

2n + 1 and the number of b’s is 2n. Then there exists u ∈ U such that

e X
{u, Xu, e 2 u, . . . , X
e 4n u}

is a basis for U ⊕ V (see proof in section 3.1.4).

e = GL(U ⊕ V ) by
Define the cocharacter φ : K × → G

e k u) = t4n−2k X
φ(t)(X e k u.

Let T = diag(t4n , t4n−2 , . . . , t2−4n , t−4n ) be the matrix associated to φ(t) with re-
e ∈ g(2; φ) since
spect to the given basis for U ⊕ V . Then X

e = T J4n+1 T −1 = t2 J4n+1 = t2 X,
Ad(φ(t))X e

43
e
where J4n+1 is the (4n + 1) × (4n + 1) Jordan matrix associated to X.
e has partition [4n + 1] in G,
Let L = GL(U ⊕ V ) = GL4n+1 (K). Since X e Xe is

distinguished in Lie L = gl(U ⊕ V ) by Lemma 3.3. We have DL = SL(U ⊕ V ). The

image φ(t) has matrix representation diag(t4n , t4n−2 , . . . , t0 , . . . , t2−4n , t−4n ), hence

Det(φ(t)) = 1 and it is immediate that φ(t) ∈ DL for all t ∈ K × . Thus φ is

e in G.
associated to X e

We would like to see also that φ(t) ∈ O(U ) × Sp(V ) for all t ∈ K × . To do this,

we may use the bilinear form β defined in section 3.1.4, since β defines symplectic

and orthogonal groups Sp(V, β) and O(U, β) conjugate to Sp(V ) and O(V ). In

particular, there exist g ∈ GL(U ) such that O(U ) = g O(U, β)g −1 with

β(h(u), h(u′ ))|U = hu, u′ iU

for u, u′ ∈ U and h ∈ GL(V ) such that Sp(V ) = h Sp(V, β)h−1 with

β(h(v), h(v ′ ))|V = hv, v ′ iV

for v, v ′ ∈ V [Jan04, 1.3].

φ(t) ∈ GL(U ) × GL(V ) since φ(t) leaves U and V invariant and det(φ(t)|U ) =
e j u, X
det(φ(t)|V ) = 1. For t ∈ K × , β(φ(t)X e h u) = t4n−2j β(X
e j u, X
e h u) = 0 unless

j + h = 4n. In this case, t4n−2j = t−(2(4n−h)−4n) = (t−1 )4n−2h and so

e j u, X
β(φ(t)X e h u) = t4n−2j β(X
e j u, X
e h u)

e j u, X
= (t−1 )4n−2h β(X e h u) = β(X
e j u, φ(t−1 )X
e h u).

Thus φ(t)∗ = φ(t)−1 with respect to β on either U or V , so φ(t) ∈ O(U ) × Sp(V ).

(In fact, we showed that φ(t) ∈ SO(U ) × Sp(V ).)

e have ab-diagram βn = bab . . . ab where the number of a’s is 2n − 1
Case 2. Let X
e ...,X
and the number of b’s is 2n. Then there exists v ∈ V such that {v, Xv, e 4n−2 v}

is a basis for U ⊕ V .
e = GL(U ⊕ V ) by φ(t)X
Define φ : K × → G e k v = t4n−2k−2 X
e k v. We have then

e ∈ g(2; φ). Indeed, φ(t) has matrix T = diag(t4n−2 , t4n−4 , . . . , t4−4n , t2−4n )
that X

44
e = T J4n−1 T −1 = t2 X.
with respect to the above basis, thus Ad(φ(t))X e

e The element
Consider the Levi subgroup L = GL(U ⊕ V ) = GL4n−1 (K) of G.
e has partition [4n − 1] in G,
X e and so X
e is a distinguished nilpotent element in

Lie(L) = gl(U ⊕ V ) by Lemma 3.3. The derived group DL is equal to SL(U ⊕ V ),

and thus φ(t) ∈ DL since Det(T ) = 1.

Using the bilinear form defined in section 3.1.5, we compute that φ(t)∗ = φ(t)−1

for t ∈ K × , and thus φ(t) ∈ O(U ) × Sp(V ). Assuming that j + h = 4n − 2,

e j v, X
β(φ(t)X e h v) = t4n−2j−2 β(X
e j v, X
e h v) = t4n−2(4n−2−h)−2 β(X
e j v, X
e h v) =

e j v, X
t−4n+2+2h β(X e h v) = (t−1 )4n−2h−2 β(X
e j v, X
e h v) = β(X
e j v, φ(t−1 )X
e h v),

the right and left-hand sides equal to zero otherwise. Hence φ(t)∗ = φ(t)−1 with

respect to β and φ(t) ∈ O(U ) × Sp(V ).

φ is therefore an associated cocharacter of X with φ(t) ∈ O(U ) × Sp(V ).

e have ab-diagram γn = aba...b...ba
Case 3. Let X aba...b...ba with n odd, and #a = 2(n + 1)

and #b = 2n. Then there exist ui ∈ U, i ∈ {1, 2} such that

e 1, X
{u1 , Xu e 2 u1 , . . . , X
e 2n u1 , u2 , Xu
e 2, X
e 2 u2 , . . . , X
e 2n u2 }

is a basis for U ⊕ V .

Define φ : K × → G e = GL(U ⊕ V ) by φ(t)X e k ui = t2n−2k ui . Then X

e
  ∈ g(2; φ).
T1 0 
Indeed, φ(t) has matrix T equal to the block-diagonal matrix   with Ti =
0 T2
diag(t2n , t2n−2 , . . . , t2−2n , t−2n ). Hence

   
J2n+1 0  −1 J2n+1 0 
e =T
Ad(φ(t))X  T = t2  2 e
=t X
0 J2n+1 0 J2n+1

Consider Levi subgroup L = GL(W1 ) × GL(W2 ) = GL2n+1 (K) × GL2n+1 (K)

e j ui | 0 ≤ j ≤ 2n}, i = 1, 2. We have U ⊕ V = W1 ⊕ W2 and
where Wi = span{X
e = X1 + X2 where Xi = X|
X e W has partition [2n + 1]. Each Xi a distinguished
i

e is a distinguished nilpotent element of Lie(L)

nilpotent element of Lie(Wi ), and so X

45
by Lemma 3.3 and [Jan04, 4.3]. The derived group DL is equal to SL(W1
) × SL(W
2)

T1 0 
and φ(t) ∈ DL for t ∈ K × . Indeed, φ(t) has associated matrix T =  ,
0 T2
which has det(Ti ) = 1 and so T ∈ SL(W1 ) × SL(W2 ).

Using the bilinear form β defined in section 3.1.6, we assume j + h = 2n. Then

e j ui , X
β(φ(t)X e j ui∗ ) = t2n−2j β(X
e j ui , X
e h ui∗ ) = t−2n+2h β(X
e j ui , X
e h ui ∗ ) =

e j ui , X
(t−1 )2n−2h β(X e h ui∗ ) = β(X
e j ui , φ(t−1 )X
e h ui ∗ )

and thus φ(t)∗ = φ(t)−1 and φ(t) ∈ O(U ) × Sp(V ). Hence φ is a cocharacter
e with φ(t) ∈ O(U ) × Sp(V ).
associated to X
bab...b...ab
Case 4. Let τ = δn = bab...b...ab with n even, #a = 2n and #b = 2(n + 1). Then
e i v1 , X
{X e j v2 | 0 ≤ i, j ≤ 2n} is a basis for U ⊕ V with vi ∈ V, i ∈ {1, 2}.

e = GL(U ⊕ V ) by φ(t)X
Define φ : K × → G e k vi = t2n−2k vi . Then X
e ∈ g(2; φ):
 
T1 0 
φ(t) has matrix T equal to the block-diagonal matrix   where
0 T2
Ti = diag(t2n , t2n−2 , . . . , t2−2n , t−2n ) with respect to the above basis. Hence

   
J2n+1 0  −1 J2n+1 0 
e =T
Ad(φ(t))X  T = t2  e
 = t2 X
0 J2n+1 0 J2n+1

Consider Levi subgroup L = GL(W1 ) × GL(W2 ) = GL2n+1 (K) × GL2n+1 (K)

e j vi | 0 ≤ j ≤ 2n}, i = 1, 2. We have U ⊕ V = W1 ⊕ W2 and
where Wi = span{X
e = X1 + X2 where Xi = X|
X e W has partition [2n + 1]. Each Xi is a distinguished
i

e is a distinguished nilpotent element of Lie(L)

nilpotent element of Lie(Wi ), and so X

by Lemma 3.3 and [Jan04,  4.3]. DL= SL(W1 ) × SL(W2 ) and φ(t) ∈ DL. Indeed,

T1 0 
for t ∈ K × , the matrix T =   of φ(t) has det(Ti ) = 1 and so T ∈ SL(W1 ) ×
0 T2
SL(W2 ).

Using the bilinear form β defined in section 3.1.7, we assume j + h = 2n. Then

e j vi , X
β(φ(t)X e j vi∗ ) = t2n−2j β(X
e j vi , X
e h vi∗ ) = t−2n+2h β(X
e j vi , X
e h vi ∗ ) =

e j vi , X
(t−1 )2n−2h β(X e h vi∗ ) = β(X
e j vi , φ(t−1 )X
e h vi∗ )

46
Thus φ(t)∗ = φ(t)−1 and φ(t) ∈ O(U ) × Sp(V ). Hence φ is a cocharacter
e with φ(t) ∈ O(U ) × Sp(V ).
associated to X
abab...bab
Case 5. Let τ = ǫn = baba...aba , where there are 2n a’s, and 2n b’s. U ⊕ V
e 1, X
has basis {w1 , Xw e 2 w1 , . . . , X
e 2n−1 w1 , w2 , X
e 2 w2 , . . . , X
e 2n−1 w2 } with w1 ∈ U and

w2 ∈ V .
e = GL(U ⊕ V ) by φ(t)X
Define φ : K × → G e k wi = t2n−2k−1 wi . Then X
e ∈ g(2; φ).
 
T1 0 
Indeed, φ(t) has matrix T equal to the block-diagonal matrix   where
0 T2
Ti = diag(t2n−1 , t2n−3 , . . . , t3−2n , t1−2n ) with respect to the above basis. Hence

   
J2n 0  −1 J2n 0 
e =T
Ad(φ(t))X  T = t2  e
 = t2 X
0 J2n 0 J2n

Consider Levi subgroup L = GL(W1 ) × GL(W2 ) = GL2n (K) × GL2n (K) where
e j wi | 0 ≤ j ≤ 2n − 1}, i = 1, 2. We have U ⊕ V = W1 ⊕ W2
Wi = span{X
e = X1 + X2 where Xi = X|
and X e W has partition [2n]. Each Xi is a distinguished
i

e is a distinguished nilpotent element of Lie(L)

nilpotent element of Lie(Wi ), and so X

by Lemma 3.3 and [Jan04,  4.3]. DL= SL(W1 ) × SL(W2 ) and φ(t) ∈ DL. Indeed,

T1 0 
for t ∈ K × , the matrix T =   of φ(t) has det(Ti ) = 1 and so T ∈ SL(W1 ) ×
0 T2
SL(W2 ).

In section 3.1.8, we defined two possible bilinear forms, one when n is even and

the other when n is odd. The forms are nonzero for the same pairing of basis

elements, and so we assume j + h = 2n − 1 and proceed identically for either n even

or odd. We have then

e j wi , X
β(φ(t)X e j wi∗ ) = t2n−2j−1 β(X
e j wi , X
e h wi∗ ) =

e j wi , X
t2n−2(2n−1−h)−1 β(X e h wi∗ ) = (t−1 )2n−2h−1 β(X
e j wi , X
e h wi∗ ) =

e j wi , φ(t−1 )X
β(X e h wi∗ )

Thus φ(t)∗ = φ(t)−1 and φ(t) ∈ O(U ) × Sp(V ). Hence φ is a cocharacter

e with φ(t) ∈ O(U ) × Sp(V ).
associated to X

47
Lemma 3.9. Given X ∈ L(V, U ), U and V quadratic spaces of types +1 and -
m
M m
M
1 respectively, there exist decompositions V = Vi and U = Ui such that
i=1 i=1
X|Vi : Vi → Ui and X ∗ |Ui : Ui → Vi , with each pair (Xi∗ , Xi ) ∈ L(Ui , Vi ) × L(Vi , Ui )

having indecomposable orthosymplectic ab-diagram τi . Furthermore, the restriction

to Ui or Vi of the bilinear form on U or V is nondegenerate.

Proof. This is the content of the ab-diagrams.

Proof of Theorem 3.7. For (X ∗ , X) with arbitrary ab-diagram τ , construct βi , φi ,

and Li as in Lemma 3.8 for each pair (Xi∗ , Xi ) with indecomposable ab-diagram

τi as in Lemma 3.9. Taking φ such that φ|Ui ⊕Vi = φi and Levi subgroup L =
Y
Li , we have that φ : K × → G e is an associated cocharacter to X and φ(t) ∈
Yi
O(Ui ) × Sp(Vi ) ⊆ O(U ) × Sp(V ) for all t ∈ K × .
i

3.3 Dimensions of Orthosymplectic Orbits

The content of this section is the proof of the following result relating the dimen-

sion of such an orbit to the dimensions of its images under the maps ρ and π. This

result was originally proven in [KP82] for char K = 0, the argument from which is

generalized here for char K 6= 2. Our proof differs from that in [KP82] in that we use

g(0) from [KP82] Lemma

the associated cocharacter in Theorem 3.7 to replace H ∈ e

7.3, the semisimple element of an sl2 -triple. We reproduce the remainder of the

proof here to demonstrate its lack of dependence on the character of K, giving ref-

erences to needed characteristic free results from [Jan04]. Additionally, the method

for computing the dimension of the weight spaces of the associated cocharacter φ

follows from the explicit construction of φ in the proof of Lemma 3.8.

Once  and V are quadratic spaces of types +1 and −1, respectively.

again, U 
0 X
e =
Let X   ∈ L(U, V ) × L(V, U ) such that (X ∗ , X) is an orthosymplectic
X ∗ 0
nilpotent pair with associated ab-diagram τ and let OX be the O(U ) × Sp(V )-orbit

48
of X ∈ L(V, U ) with action as in section 3.1.1. Denote by π and ρ the maps

π
L(V, U ) / g(U )

ρ

g(V )

as in section 2.2.1.

Proposition 3.10.

1
dim OX = (dim π(OX ) + dim ρ(OX ) + dim U · dim V − ∆τ )
2

X
where ∆τ = ai bi , ai the number of rows of τ of length i starting with a and bi
iodd
the number of rows of τ of length i starting with b.

Note that ∆τ = 0 when X : V → U is either injective or surjective.

3.3.1 e as a Θ-group
G

We begin by describing L(U, V )× L(V, U ) as a Θ-group as in [Vin76] (see [Lev09]

e := GL(U ⊕V ) and the automorphism
for characteristic p > 2). Consider the group G
e→G
Θ:G e given by
   −1
A B  A∗ C∗
 
Θ /


 .
C D B∗ D∗

 
Id 0 
We observe that Θ2 is conjugation with the matrix J :=  . Indeed,
0 −Id

   −1    
A B  A ∗ C∗ (A∗ )∗ (B ∗ )∗   A −B 
 
Θ /
 
Θ /
 = 
C D B∗ D∗ (C ∗ )∗ (D ∗ )∗ −C D

since A ∈ L(U, U ), D ∈ L(V, V ), C ∈ L(U, V ) and D ∈ L(V, U ), with U and V of

type 1 and −1, respectively (see section 2.2.1). This then implies that Θ4 is the

identity.

49
eΘ2 , the fixed point group
eΘ , the fixed point group of Θ, is O(U ) × Sp(V ) and G
G
eΘ2 since G
of Θ2 , isGL(U ) × GL(V ). To see this, we first compute G eΘ2 . If
eΘ ⊆ G

 A B  e Θ2
 ∈G ,
C D

     
A B  2
A B   A −B 
 =Θ  = ,
C D C D −C D

 
A 0
which implies B = C = 0. Hence any element of G eΘ2 has block form 
 .
0 D
   
A 0  A B  e Θ
Since   is invertible, A ∈ GL(U ) and D ∈ GL(V ). Now if  ∈G ,
0 D C D
we have B = C = 0 from before, and

   −1  
∗
A 0  A 0 A 0 
Θ =  = ,
0 D 0 D∗ 0 D

   
∗
A 0 A 0 
i.e.   is the inverse of  . Multiplying
0 D ∗ 0 D

       
 A∗ 0  A 0  IU 0  A 0   A∗ 0 
  = =  ,
0 D∗ 0 D 0 IV 0 D 0 D∗

we get that AA∗ = IU = ∗ ∗ ∗

 A A and DD = IV = D D, hence A = A
∗ −1 and

A 0 
D ∗ = D −1 . Thus   ∈ O(U ) × Sp(V ) (section 2.1.1). The reverse inclusions
0 D
eΘ2 = GL(U ) × GL(V ).
eΘ = O(U ) × Sp(V ) and G′ := G
are clear. Define G := G

We denote by dΘ the automorphism of order four induced by Θ on the Lie

50
algebra e e = End(U ⊕ V ). It is given by
g := Lie G

   
A B  A∗ C∗
 
dΘ / −


.
C D B ∗ D∗

Since we have been assuming p 6= 2, Θ is a semisimple automorphism automorphism

e Hence the Lie algebra of the centralizer G, above, of Θ is the centralizer in
of G.

Lie(G) of dΘ [Jan04, 2.5, 2.6].

g(i) := {X ∈ e
Fixing a primitive fourth root of unity, ζ4 , define e g | dΘX = ζ4i X}.

Then
e g(0) ⊕ e
g=e g(1) ⊕ e
g(2) ⊕ e
g(3)

is a G-stable Z/4Z-graduation of e
g where, by definition,

g(0) = g := Lie G
e and g(0) ⊕ e
e g(2) = g′ := Lie G′ .

g(1) ∼
Additionally, we can identify e = L(V, U ) as G-modules. If
     
A B    A∗ C∗
A B 
dΘ   = −  = ζ4  
C D B ∗ D∗ C D

Then −A∗ = ζ4 A, −D ∗ = ζ4 D, −C ∗ = ζ4 B, and −B ∗ = ζ4 C. The last two

equations yield C = ζ4 B ∗ , recalling that (C ∗ )∗ = −C and (B ∗ )∗ = −B. If we apply

∗ to the first two equations, we get A = (A∗ )∗ = (−ζ4 A)∗ = ζ42 A = −A, likewise

D = −D, implying that A = D = 0. Hence

  

 

 0 B
g(1)
e =   | B ∈ L(V, U ) .

 ζ4 B ∗ 0 


3.3.2 e via Cocharacters

Grading of G

Recall from section 3.2.1 the means of grading the Lie algebra of a reductive
e be a cocharacter associated to X
group via a cocharacter. Let φ : K × → G e in G
e

51
M
as in Theorem 3.7 and define e
g= e
gi where e g(i; φ) = {Z ∈ e
g=e g | Ad(φ(t))Z =
i∈Z
ti Z for all t ∈ K × }.

It is clear from the proof of Lemma 3.8 how to calculate the dimensions of

gi of φ in terms of the the ab-diagram, τ , associated to X ∈

the weight spaces e
e k w, w ∈ U or V , such that
L(V, U ). Namely, count the number of basis vectors X
e k w = ti X
Ad(φ(t))X e k . This is most easily done by labeling each row of the diagram

of length r by (r − 1, r − 3, . . . , 2, 0, −2, . . . , 3 − r, 1 − r) when r is odd, and by

(r − 1, r − 3, . . . , 1, −1, . . . , 3 − r, 1 − r) when r is even. These dimensions depend

e thus only on X, and not on the choice of cocharacter since we have
only on X,
e and any associated cocharacter will result in the
defined φ to be associated to X

same weight space decomposition by Proposition 3.5.

The cocharacter φ as in Theorem 3.7 takes values in O(U ) × Sp(V ) ⊂ GL(U ) ×

gi has the form Ui ⊕ Vi . The dimensions of each Ui and Vi

GL(V ), and so each e

can be calculated by keeping track of the labeling of the a’s (corresponding to basis

vectors belonging to U ) and b’s (corresponding to basis vectors belonging to V ) in

the ab-diagram of X. In particular,

dim U0 = (#ατ ) + (#βτ )

and

dim V0 = 2(#δτ ) + 2(#γτ )

where #ατ denotes the number of irreducible diagrams of type αk in the ab-diagram

τ of (X ∗ , X), k ∈ Z, etc.

52
For example, by labeling the ab-diagram

abababa a6 b4 a2 b0 a−2 b−4 a−6

bababab b6 a4 b2 a0 b−2 a−4 b−6

ababa a4 b2 a0 b−2 a−4

τ= =
ab a1 b−1

ba b1 a−1

a a0

and we can see that dim(U0 ) = 3 (there are three a0 ’s) and dim(V0 ) = 1 (there is

one b0 ).

3.3.3 More on Orbit Dimensions

g(1) ∼
For X ∈ e = L(V, U ) a nilpotent element with associated orthosymplectic ab-
e associated to X
diagram τ , choose a cocharacter φ : K × → G e with image in G
e(0) =

O(U ) × Sp(V ) as in Theorem 3.7. Let Pe = Peφ be the associated parabolic subgroup
M
e with Lie(Pe) = e
of φ in G p := gi . The Lie algebra e
e p is a parabolic subalgebra of
i≥0
M
e
g and has nilpotent radical e
n := gi and Levi-decomposition e
e g0 ⊕ e
p=e n. We have
i>0
from Proposition 3.6 that ad X : e
p→e
n2 is surjective.

φ defines a Z-grading of g′ = gl(U ) × gl(V ) and g = so(U ) × sp(V ) induced by

g in section 3.3.2. p′ := e
the grading of e p ∩ g′ and p := e
p ∩ g are parabolic subalgebras

of g′ and g with Levi decompositions

p′ = g′0 ⊕e
n′ , where g′0 := e
g0 ∩ g′ , n′ := e
e n ∩ g′

and

p = g0 ⊕n, g0 ∩ g,
where g0 := e n ∩ g.
n := e

Let P ′ and P be the parabolic subgroups of G′ = GL(U ) × GL(V ) and G =

O(U ) × Sp(V ), respectively, with Lie(P ′ ) = p′ and Lie(P ) = p. By Proposition

53
3.4(b), it follows that G′X ⊂ P ′ and GX ⊂ P . From the surjectivity of the map

ad X : e
p→e
n2 , we have that the maps

ad X : p′ → n′2 and ad X : p → n2

are surjective, where n′2 := e g(1) ⊕ e

n2 ∩ (e g(3) ) and n2 := e g(1) .
n2 ∩ e

′ , O and O 0 denote the orbits of X under G′ , G and G0 =

Lemma 3.11. Let OX X X

SO(U ) × Sp(V ) respectively.

′
(a) The canonical maps G′ ×P n′2 → OX
′ and G0 ×P n → O 0 are resolutions of
2 X

singularities.

′ = dim n′ + dim n′ and dim O = dim O 0 = dim n + dim n .

(b) dim OX 2 X X 2

(c) dim n′2 = 2 dim n2 .

Proof. (a) and (b) follow from the above discussion and that in section 3.2.1. For
2 e e
(c), we  earlier that the automorphism Θ : G → G is conjugation with
 recall from
Id 0 
eΘ , we have Θ2 φ(t) = φ(t), and so J and
J =  ∈ G. Then since φ(t) ∈ G
0 −Id
gi = e
φ(t) commute. Thus Je g(1) = e
gi for all i. Additionally, we observe that Je g(3) ,

g(1) ∩ e
so that J(e g(3) ∩ e
gi ) = e g(1) ∩ e
gi . This implies that J(e g(3) ∩ e
n2 ) = e n2 , and in

g(1) ∩ e
particular, dim(e g(3) ∩ e
n2 ) = dim(e g(1) ∩ e
n2 ). Since n′2 = (e g(3) ∩ e
n2 ) ⊕ (e n2 ) and

g(1) ∩ e
n2 = e n2 , dim n2 = 2 dim n′2 .

3.3.4 Proof of Proposition 3.10

Proof. We begin by describing the dimension of the orbit OX of a nilpotent element

g(1) ∼
X ∈ e = L(V, U ) under G = O(U ) × Sp(V ) in terms of the dimension of the
′ of X under G′ = GL(U ) × GL(V ). Choose an associated cocharacter φ
orbit OX
e(0) (Theorem 3.7) and consider the associated parabolic
of X with φ(t) ∈ G = G

subalgebras

p′ = g′0 ⊕n′ ⊆ g′ and p = g0 ⊕n ⊆ g

54
as in section 3.3.3. By definition, the Levi factors g′0 and g0 are the stabilizers of
M M
φ(t), t ∈ K × , in g′ and g respectively. If U = Ui and V = Vj are the weight
i j
space decompositions of U and V with respect to φ (i.e. Ui := {u ∈ U | φ(t)u =

ti u for all t ∈ K × } and Vj := {v ∈ V | φ(t)v = ti v for all t ∈ K × }), we have from

[Jan04, 4.4] that

M M
g′0 = gl(Ui ) ⊕ ( gl(Vj )).
i j

We can see from the construction in Lemma 3.8 that the subspaces Ui + U−i are

nondegenerate orthogonal subspaces of U , and Vj +V−j are nondegenerate symplectic

subspaces of V . So by section 4.5 in [Jan04]

M M
g0 ∼
=( gl(Ui )) ⊕ so(U0 ) ⊕ ( gl(Vj )) ⊕ sp(V0 ).
i>0 i>0

Let m0 := dim U0 and n0 := dim V0 . Then by the above and equation (#) in section

2.1.1, we get

2 dim g0 − dim g′0 = (2 dim so(U0 ) − m20 ) + (2 dim sp(V0 ) − n20 )

= (m0 (m0 − 1) − m20 ) + (n0 (n0 + 1) − n20 )

= n 0 − m0 .

Let m = dim U and n = dim V . By lemma 3.11(b),(c), and equation (#), we have

′
4 dim OX − 2 dim OX = 4 dim n − 2 dim n′

= 2(dim g − dim g0 ) − (dim g′ − dim g′0 )

= (2 dim g − dim g′ ) − (n0 − m0 )

= (m(m − 1) + n(n + 1) − m2 − n2 ) − (n0 − m0 ).

For the second equality, note that 2 dim n = dim g − dim g0 and 2 dim n′ = dim g′ − dim g′0

by definition since dim e g−i for all i ∈ Z>0 . Combining these,

gi = dim e

′
4 dim OX = 2 dim OX + (n − m) − (n0 − m0 ).

55
Next, consider the conjugacy classes Ad(O(U ))X = π(OX ) ⊆ so(U ) and

Ad(Sp(V ))X = ρ(OX ) ⊆ sp(V ) and denote by π(OX )′ and ρ(OX )′ the conjugacy

classes Ad(GL(U ))X ⊆ gl(U ) and Ad(GL(V ))X ⊆ gl(V ) generated by π(OX ) and

ρ(OX ) under the action of GL(U ) and GL(V ), respectively. We recall the dimen-

sion formula in the linear case from proposition 5.3 in [KP79], which is given in

characteristic 0 but works in any characteristic ([Don90] proof of Corollary 2.1(d)):

′ 1
dim OX = (dim π(OX )′ + dim ρ(OX )′ ) + nm − ∆τ .
2

From Remark 2.5 in section 2.1.3, we have

dim π(OX )′ = 2 dim π(OX ) + m − ra

and

dim ρ(OX )′ = 2 dim ρ(OX ) − n + rb ,

where ra , rb are the number of odd rows in the Young diagrams of π(OX ) and ρ(OX )

respectively.

Combining all of these,

4 dim OX = dim π(OX )′ + dim ρ(OX )′ + 2nm − 2∆τ + (n − m) − (n0 − m0 )

= 2 dim π(OX ) + 2 dim ρ(OX ) + 2nm − 2∆τ + (rb − ra ) − (n0 − m0 ).

We use our previous analysis of ab-diagrams to show that rb − ra = n0 − m0 .

Denote by τi,a the number of a’s in the ith row of τ and by τi,b the number of b’s.

From the discussion in section 3.3.2, we have m0 = #{i | τi,a odd and τi,b even}

and n0 = #{i | τi,b odd and τi,a even}. Hence n0 − m0 = #{i | τi,b odd} − #{j |

τj,a odd} = rb − ra . Thus

2 dim OX = dim π(OX ) + dim ρ(OX ) + nm − ∆τ .

56
3.4 Dimensions in Z

Recall the definition of Z from section 2.2.1. It is the subset of tuples (X1 , . . . , Xt )

of M = L(V0 , V1 ) × . . . × L(Vt−1 , Vt ) defined by the equations

X1 X1∗ = X2∗ X2

X2 X2∗ = X3∗ X3
..
(∗) .
∗
Xt−1 Xt−1 = Xt∗ Xt

Xt Xt∗ = 0

with each Xi ∈ L(Vi−1 , Vi ). We have n0 = dim(V0 ) ≥ n1 = dim(V1 ) ≥ . . . ≥ nt =

dim(Vt ) > 0 = dim(Vt+1 ), the Vi quadratic spaces of alternating types, constructed

as the images of Di−1 = D|Vi−1 ∈ g(Vi−1 ) for some nilpotent D ∈ g(V ).

Before proving part (a) of Theorem 2.13, we first need a lemma relating the

dimension of the singular locus in Z to the dimension of the orbits in OD . This

will use the formula in Proposition 3.10, as well as a new realization of Z as a fiber

product of the orbit closures ODi . The proof is identical to that in [KP79] 8.1,8.2,

which depends on the character of K only insofar as it relies on the dimension result

from Proposition 3.10, proven in [KP79] for char K = 0 only.

3.4.1 Z as a Fiber Product

Z can be realized as the iterated fiber product:

57
Z / Z1,t / Z2,t / ... / Zt−1,t /O =0
Dt

Zt−1 / Z1,t−1 / Z2,t−1 / ... / OD
t−1
·
·· ···
··
···

..
.
..
.
..
. · · ···
·
· · ···
· ···
Z2 / Z1,2 /O
D2

Z1 /O
D1

OD

We have for 0 ≤ i < j ≤ t, Zi,j = {(Xi+1 , . . . , Xj ) ∈ L(Vi , Vi+1 ) × . . . ×

L(Vj−1 , Vj ) | ρ(Xk ) ∈ ODk−1 , π(Xk ) ∈ ODk , ρ(Xk ) = π(Xk−1 ), i + 1 ≤ k ≤ j}.

For 1 ≤ j ≤ t, Zj = Z0,j (these will reappear in section 4.3). Note that Zt = Z.

The last condition, ρ(Xk ) = π(Xk−1 ), is the same as the equations in (∗).

Each coordinate Xk of the element (Xi+1 , . . . , Xj ) ∈ Zi,j as in the previous

paragraph, 0 ≤ i < j ≤ t has an ab-diagram τk satisfying π(τk ) = ρ(τk+1 ) =: σk ,

where σk is an ε-diagram of a nilpotent element in g(Vk ) (i.e. π(Oτk ) = ρ(Oτk+1 ) =

Oσk ) such that σk ≤ λk where λk is the Young diagram of Dk = D|Vk as in section

2.3.1. In other words, Oσk ⊆ ODk . This follows from the condition that π(Xk ) =

ρ(Xk+1 ) ∈ ODk .

Thus we have a finite set Ω of strings ω = (τ1 , τ2 , . . . , τt ) of orthosymplectic

ab-diagrams with Oτi ⊆ L(Vi−1 , Vi ) satisfying

(i) π(τi ) = ρ(τi+1 ) = σi for 1 ≤ i ≤ t − 1, with σi as above an ε-diagram with

σi ≤ λk , and

(ii) σt = 0, i.e. Oσt = 0.

We define σ0 := ρ(τ1 ). For ω = (τ1 , . . . , τt ) ∈ Ω, define the locally closed subset

Zω ⊆ Z by Zω = {(X1 , . . . , Xt ) ∈ Z | Xi ∈ Oτi }.

58
The conditions on the ω ∈ Ω listed above imply that we have a fiber product

diagram subordinate to the previous diagram defining Z:

Zω / Zω / Zω / ... / Oτt / Oσt = 0

1 2

· /· /· / ... / Oσt−1
·
··
··
··
··
.. .. .. · ··
·
. . . ··
· ··
··
···
· ·
· / Oτ / Oσ
2 2

Oτ1 / Oσ
1

Oσ0

where ωi denotes the string (τi+1 , . . . , τt ) and ω0 = ω. All of the maps in this

diagram are smooth, being either projections or the maps ρ or π at the bottom of

the columns and right-hand ends of the rows. Thus each Zω is a smooth variety,

and we get from Proposition 3.10

dim Zω = dim Oτ1 − dim Oσ1 + dim Zω1

1
= (dim Oσ0 + dim Oσ1 + n0 n1 − ∆τ1 ) − dim Oσ1 + dim Zω1
2
1 1
= (dim Oσ0 + n0 n1 − ∆τ1 ) − dim Oσ1 + dim Zω1
2 2

and for any Zωi , 0 ≤ i ≤ t − 3, we have the analagous

dim Zωi = dim Oτi+1 − dim Oσi+1 + dim Zωi+1

1
= (dim Oσi + dim Oσi+1 + ni ni+1 − ∆τi+1 ) − dim Oσi+1 + dim Zωi+1
2
1 1 1
= dim Oσi + (ni ni+1 − ∆τi+1 ) − dim Oσi+1 + dim Zωi+1
2 2 2

59
For t = t − 2, we have the fiber product diagram Zωt−2 / Oτ
t which implies

Oτt−1 / Oσt−1

dim Zωt−2 = dim Oτt−1 − dim Oσt−1 + dim Oτt

1
= (dim Oσt−2 + dim Oσt−1 + nt−2 nt−1 − ∆τt−1 ) − dim Oσt−1 + dim Oτt
2
1 1 1
= dim Oσt−2 + (nt−2 nt−1 − ∆τt−1 ) − dim Oσt−1 + dim Oτt
2 2 2

Proposition 3.10 implies dim Oτt = 12 (dim Oσt−1 + dim Oσt + nt−1 nt − ∆τt ). Recall

that Oσt = 0, so we have

1 1
dim Zωt−2 = dim Oσt−2 + (nt−2 nt−1 + nt−1 nt − ∆τt−1 − ∆τt ).
2 2

Thus inductively,

t
1 1X
dim Zω = dim Oσ0 + (ni−1 ni − ∆τi ).
2 2
i=1

3.4.2 From Orbits to Z

Recall from section 2.3.3 the map ϑ : Z → OD given by ϑ(X1 , . . . , Xt ) = X1∗ X1 .

Lemma 3.12. For every conjugacy class O ⊆ OD we have

1
codimZ ϑ−1 (O) ≥ codimOD O.
2

Proof. Recall that the orbit OD is open in its closure OD . As a consequence of

Lemma 2.8 and the discussion in section 2.3.3, there is a unique open stratum Zλ0

that is the iterated fiber product of the open orbits Oλi = ODi ⊆ ODi , i.e. sitting

“above” OD . We have , λ0 = (τ10 , . . . , τt0 ) ∈ Ω where τi0 is the ab-diagram of

D|Vi−1 : Vi−1 → Vi .
Qt ′
Define M ′ := {(X1 , . . . , Xt ) ∈ M | all Xi surjective} = i=1 L (Vi−1 , Vi ) and

let Z ′ = Z ∩ M ′ . We have by Lemma 2.8 that Zλ0 ⊆ Z ′ , and ∆τ 0 = 0 by the

i

remark after Proposition 3.10 since the D|Vi are by definition surjective. Hence by

60
the dimension result of the previous section

t
1 1X
dim Zλ0 = dim OD + ni−1 ni .
2 2
i=1

The classes O ∈ g(V ) have even dimension, so for O ⊂ OD , dim O ≤ dim OD −2.

Thus for all other ω ∈ Ω, dim Zω ≤ dim Zλ0 − 1. Z is a finite union of strata Zω ,

and so we have that dim Z = dim Zλ0 = dim Z ′ . So for ω = (τ1 , . . . , τt ) 6= λ0 above

O=
6 OD , we have

codimZ Zω = dim Zλ0 − dim Zω

X t X t
1
= (dim OD + ni−1 ni − dim O − (ni−1 ni − ∆τi ))
2
i=1 i=1
t
X
1
= (codimOD O + ∆τi ).
2
i=1

Since ϑ−1 (O) is a finite union of strata Zω satisfying the above1 , we have that

1
codimZ ϑ−1 (O) ≥ codimOD O.
2

Remark 3.13. As a result of the above proof, those strata Zω , ω = (τ1 , . . . , τt ), of

codimension 1 are iterated fiber products above those conjugacy classes O ⊂ OD of

codimension 2 satisfying ∆τi = 0.

3.4.3 Proof of Theorem 2.13(a)

Theorem (2.13a). Let G(V ) be a symplectic or orthogonal group for a finite dimen-

6 2. For a nilpotent
sional vector space V over a field K of prime characteristic p =
Y t
element D ∈ g(V ), construct M = L(Vi−1 , Vi ) and Z ⊆ M the set of tuples
i=1
(X1 , . . . , Xt ) satisfying equations (∗) below, the Vi as in section 2.3.1. Then the
1
See Appendix B for a detailed discussion.

61
affine scheme Z is reduced, so that Z isan affine variety. Moreover, Z is a complete

intersection in M with respect to the equations (∗).

X1 X1∗ = X2∗ X2

X2 X2∗ = X3∗ X3
..
(∗) .
∗
Xt−1 Xt−1 = Xt∗ Xt

Xt Xt∗ = 0

Proof. Consider the map

t
Y
ζ :M → g(Vi ) =: N
i=1

given by (X1 , X2 , . . . , Xt ) 7→ (X1 X1∗ −X2∗ X2 , X2 X2∗ −X3∗ X3 , . . . , Xt Xt∗ ) as in Remark

2.11. As a scheme, Z is the fiber ζ −1 (0). In order to see that Z is an affine variety,

we show that it is a complete intersection and nonsingular in codimension zero,

which implies by the discussion in section 1.4 that the ideal defining Z is reduced.

First, we show that ζ is smooth in

t
Y
′
M := {(X1 , . . . , Xt ) | all Xi surjective} = L′ (Vi−1 , Vi ).
i=1

To see this we compute the differential dζ at a point α = (X1 , . . . , Xt ) ∈ M ′ . Taking

a tangent vector (T1 , . . . , Tt ) ∈ M we get:

(dζ)α (T1 , . . . , Tt ) = (T1 X1∗ + X1 T1∗ − T2∗ X2 − X2∗ T2 , . . . ,

∗ ∗
Tt−1 Xt−1 + Xt−1 Tt−1 − Tt∗ Xt − Xt∗ Tt , Tt Xt∗ + Xt Tt∗ ).

Since each Xi is surjective, we can solve the equation

(dζ)α (T1 , . . . , Tt ) = (S1 , . . . , St )

62
inductively: If Tt , Tt−1 , . . . , Tj+1 have been determined, one has to solve an equation

Tj Xj∗ + Xj Tj∗ = Sj′

for some Sj′ satisfying (Sj′ )∗ = −Sj′ . This can be done setting Sj′ = Rj − Rj∗ and then

solving Xj Tj∗ = Rj using the fact that Xj is surjective. Thus (dζ)α is surjective for

α ∈ M ′ , and ζ is smooth in M ′ .

In particular, Z, as a scheme, is smooth in Z ′ := Z ∩ M ′ . Applying Lemma 2.8

(b) and (c) inductively, we see that ϑ−1 (OD ) ⊆ Z ′ , where ϑ : Z → OD is the map

(X1 , . . . , Xt ) = X1∗ X1 . Hence Z ′ 6= ∅.

ζ is surjective. Indeed, since π : L(Vt−1 , Vt ) → g(Vt ) given by π(X) = XX ∗ is

surjective, ζ is surjective in the last coordinate. Proceeding inductively, assume that

we have chosen (Xi , . . . , Xt ) such that Xi Xi∗ = Xi+1

∗ X ∗
i+1 + Si , . . . , Xt Xt = St for

some (S1 , . . . , St ) ∈ N . Then, again, the surjectivity of π : L(Vi−2 , Vi−1 ) → g(Vi−1 )

∗
implies that we can find Xi−1 ∈ L(Vi−2 , Vi−1 ) such that π(Xi−1 ) = Xi−1 Xi−1 =

Xi∗ Xi + Si−1 .

Thus we have that the dimension of the generic fiber Z ′ is dim M − dim N .

Hence codimM Z = codimM Z ′ = dim N , since we showed dim Z = dim Z ′ during

the proof of Lemma 3.12. As noted in Remark 2.12, equations (∗) give exactly
t
Y
dim g(V1 ) × . . . × dim g(Vt ) = dim N equations in Ani ni−1 ∼
= M . Hence Z is a
i=1
complete intersection.

OD \ OD consists of finitely many conjugacy classes Oi and OD is a finite union

S S
OD ∪ ( Oi ). ϑ−1 (OD ) ⊆ Z ′ so we must have that Zsing ⊂ ϑ−1 ( Oi ). For each we
1
have codimZ ϑ−1 (Oi ) ≥ 2 codimOD Oi ≥ 1 by Lemma 3.12, which shows that Z is

smooth in codimension 0.

Thus Z is an affine variety which is a complete intersection in M , proving (a).

63
Chapter 4

Part B

In this chapter we introduce the concept of a good pair of varieties as well as

some specific results in the symplectic and orthogonal groups. These will be used

to prove part (b) of Theorem 2.13, showing that there is a quotient map Z → OD

for a suitable algebraic group.

4.1 General Results on Good Pairs

We begin by stating some definitions and results needed from [Don90] and [AJ84].

In the following section, let G be an arbitrary reductive linear algebraic group over

K. A G-module connotes a rational, possibly infinite dimensional, G-module. De-

note by B a Borel subgroup, T ⊂ B a maximal torus, and X(T ) the character group

of T .

We choose a system of positive roots such that the weights of T in Lie(B) are

zero and the negative roots. This determines a partial order on X(T ), denoted by

≤. Let X+ (T ) be the set of dominant weights. For ξ ∈ X(T ), let Kξ be the one-

dimensional B-module on which T acts via ξ and let Y(ξ) = IndG

B Kξ be the induced

module as in [Jan03].

64
4.1.1 Good Filtrations

By a good filtration of a G-module M we mean an ascending filtration of M

0 = M0 ⊂ M1 ⊂ M2 ⊂ . . .

[
with M = Mi such that for each i > 0, Mi /Mi−1 is either 0 or isomorphic to
i
Y(ξi ) for some ξi ∈ X+ (T ).

The following results, which we will need later, are proven for general reductive

G in [Don90] where H ∗ is Hochschild cohomology.

Proposition 4.1. Let M be a G-module.

(a) If M has a good filtration then H i (G, M ⊗ Y(ξ)) = 0 for all i > 0 and ξ ∈

X+ (T ).

(b) If M has countable dimension and H 1 (G, M ⊗ Y(ξ)) = 0 for all ξ ∈ X+ (T )

then M has a good filtration.

(c) If M has a good filtration and ξ ∈ X+ (T ), then the filtration multiplicity

(M : Y(ξ)) of Y(ξ) in M is independent of the choice of good filtration and

equal to dim(M ⊗ Y(ξ ∗ ))G .

Proof. This is Proposition 1.2a(i-iii) in [Don90]. Part (a) is Corollary 3.3 of [CPSvdK77]

and part (b) is due to Friedlander in [Fri85], which generalizes the finite dimensional

case in [Don81]. Part (c) follows from [Don85, 12.1.1].

Define as in [FP86] the good filtration dimension d = d(M ) of a G-module M by

the two conditions

1. H i (G, M ⊗ Y ) = 0 for all i > d and G-modules Y admitting a good filtration,

and

2. H d (G, M ⊗ Y ) 6= 0 for some G-module Y admitting a good filtration.

If M 6= 0 and there is no non-negative integer d satisfying 1 and 2, we put d(M ) =

+∞. We follow the convention d(0) = −∞.

65
Lemma 4.2. (a) If 0 → M → N0 → N1 → . . . → Nn → 0 is an exact sequence

of G-modules then d(M ) ≤ max{d(Ni ) + i | 0 ≤ i ≤ n}.

(b) If 0 → Nn → Nn−1 → . . . → N0 → M → 0 is an exact sequence of G-modules

then d(M ) ≤ max{d(Ni ) − i | 0 ≤ i ≤ n}.

Proof. This is Lemma 1.2c in [Don90]. The result follows from the long exact

sequence of cohomology for n = 1, and by induction, splicing exact sequences and

using the case n = 1. When n = 0, there is nothing to prove.

Corollary 4.3. Let M be a finite dimensional G-module such that every symmetric
V
power of M admits a good filtration. Then d( r M ) ≤ r−1 for every positive integer

r.

Proof. This is Corollary 1.2d in [Don90].

We will often be concerned with a product G = G1 × G2 . In this case, we

take B = B1 × B2 and T = T1 × T2 , where Ti is a maximal torus in the Borel

subgroup Bi ⊂ Gi , i = 1, 2. Then we have X+ (T ) = X+ (T1 ) × X+ (T2 ) and Y(ξ, η) =

G1
IndG + +
B K(ξ,η) = Y 1 (ξ)⊗Y 2 (η) for ξ ∈ X (T1 ) and η ∈ X (T2 ) where Y1 (ξ) = IndB1 Kξ
G2
and Y2 (η) = IndB2
Kη [Don90, 1.1].

Proposition 4.4. Suppose that G = G1 ×G2 and M is a G-module.

(a) If M has a good G-module filtration then M has a good G1 -module filtration.

(b) If G2 acts trivially on M then M has a good G1 -module filtration if and only

if M has a good G-module filtration.

(c) If M has a good G-module filtration then M G1 has a good G2 -module filtration

and (M G1 : Y2 (η)) = (M : Y(0, η)) for η ∈ X+ (T2 ).

Proof. This is Proposition 1.2e in [Don90].

66
4.1.2 Good Pairs

Let V be a reduced affine algebraic variety over K with coordinate ring K[V].

If φ : U → V is a morphism of varieties, then we also have the comorphism φ∗ :

K[V] → K[U ], given by φ∗ f = f ◦ φ ∈ K[U ] for f ∈ K[V]. Let V be a G-variety,

i.e. a variety V equipped with a left action G × V → V, which is also a morphism of

varieties. Then K[V] is a left G-module with action given by (gf )(v) = f (g −1 v) for

g ∈ G, f ∈ K[V], and v ∈ V.

A good G-variety is a G-variety V such that K[V] admits a good G-module

filtration. A good pair of G-varieties is a pair (V, A) where V is a good G-variety

and A ⊂ V is a G-stable closed subset such that the defining ideal IA ⊂ K[V] of A

admits a good G-module filtration. If (V, A) is a good pair of G-varieties, then A is

a good G-variety since K[A] = K[V]/IA has a good filtration by Lemma 4.1(a) and

(b) and the exact sequence 0 → IA → K[V] → K[A] → 0.

The following are a few useful results about good pairs.

Lemma 4.5. Let V be a G-variety and let A ⊂ A′ be G-stable closed subsets of V.

(a) If (V, A′ ) and (A′ , A) are good pairs then (V, A) is a good pair.

(b) If (V, A) and (V, A′ ) are good pairs then (A′ , A) is a good pair.

Proof. This is Lemma 1.3a in [Don90].

Proposition 4.6. (a) If (V1 , A1 ) and (V2 , A2 ) are good pairs of G-varieties, then

(V1 × V2 , A1 × A2 ) is a good pair of G-varieties for the action g(v1 , v2 ) =

(gv1 , gv2 ) (g ∈ G, vi ∈ Vi ) of G on V1 × V2 .

(b) If G = G1 × G2 , (V1 , A1 ) is a good pair of G1 -varieties, and (V2 , A2 ) is a good

pair of G2 -varieties, then (V1 × V2 , A1 × A2 ) is a good pair of G-varieties for

the action (g1 , g2 )(v1 , v2 ) = (g1 v1 , g2 v2 ) (gi ∈ Gi , vi ∈ Vi ) of G on V1 × V2 .

Proof. This is Proposition 1.3e in [Don90].

67
The concept of a good complete intersection is a useful tool for finding good

pairs of varieties. Let V be a good, irreducible and smooth G-variety with a G-

stable closed subvariety A of codimension r, and suppose that E is a G-stable

subspace of K[V] of dimension r that generates the ideal of A. Then A is a good

complete intersection in V with respect to E if the symmetric algebra S(E) has a

good G-module filtration.

Proposition 4.7. Suppose V is a good, irreducible, smooth G-variety and A is a

good complete intersection with respect to some G-stable subspace E of K[V]. Then

(V, A) is a good pair of G-varieties.

Proof. This is Proposition 1.3b(i) in [Don90]. We include the proof for flavor.

Consider the Koszul resolution

/ K[V] ⊗
Vr / K[V] ⊗ Vr−1 (E) / ... / K[V] ⊗ E
0 (E) / IA /0

as in [Mat90, page 135]. The maps are G-equivariant. If Y is a G-module with a

good filtration, then so is Y ′ = Y ⊗ K[V] by [Mat90] and [Don85, 3.1.1]. Therefore,

V V
we have d(K[V] ⊗ i (E)) ≤ d( i (E)) ≤ i − 1 for i > 0 by Corollary 4.3. Hence we

have from the Koszul resolution and Lemma 4.2(b),

Vi+1
d(IA ) ≤ max{d(K[V] ⊗ (E)) − i) | 0 ≤ i ≤ r − 1}
V
≤ max{d( i+1 (E)) − i) | 0 ≤ i ≤ r − 1}

≤ 0.

Thus IA has a good filtration and (V, A) is a good pair.

Given a finite dimensional G-module, M, over K, one can view M as the affine

G-variety Adim M . Then M is a good G-variety if and only if all symmetric powers

Sn (M∨ ) of the dual of M have good filtrations as G-modules, hence S(M∨ ) has

a good filtration. The following Lemma implies that one must only examine the

exterior powers of M, of which there are finitely many, in order to know about the

symmetric powers.

68
V
Lemma 4.8. If M is a finite dimensional G-module such that (M∨ ) has a good

filtration, then S(M∨ ) has a good filtration.

Proof. This is [AJ84, 4.3(1)].

V
Lemma 4.9. If M and N are finite dimensional G-modules such that (M∨ ) and
V ∨
(N ) have good filtrations, then M ⊗ N ∨ is a good G-variety.

V V
Proof. If (M∨ ) and (N ∨ ) have good filtrations, [AJ84, 4.3(2)] states that S(M∨ ⊗

N ∨ ) has a good filtration. This then implies that M⊗ N ∨ is a good G-variety, iden-

tifying N with its dual N ∨ .

The following Corollary gives another means of determining good pairs of vari-

eties.

Corollary 4.10. Let M be a finite dimensional G-module with submodule A. If the

symmetric algebras S(M∨ ) and S((M/A)∨ ) admit good filtrations, then (M, A) is

a good pair when regarded as G-varieties.

Proof. This is Corollary 1.3c in [Don90].

4.1.3 Quotient Maps

As advertised, good pairs of varieties are the key to generalizing the proof of

Theorem 5.3 in [KP82]. Over a field of characteristic 0, quotient maps restrict to

quotient maps on stable subvarieties. This is not the case in positive characteristic,

where the restriction of a quotient map to a stable subvariety may or may not be a

quotient map. However, the following proposition shows that a quotient map of a

good variety V does restrict to a quotient map on a subvariety A if (V, A) is a good

pair of varieties.

Recall that if V is a G-variety and q : V → U a morphism of varieties. We call

q, or less precisely U , a G-quotient if q is closed, surjective, and q∗ K[U ] = K[V]G .

Proposition 4.11. Suppose G = G1 × G2 and (V, A) is a good pair of G-varieties.

If q : V → U is a G-equivariant morphism of varieties which is a G1 -quotient map,

69
then (U , q(A)) is a good pair of G2 -varieties and the restriction A → q(A) is a

G1 -quotient.

Proof. This is Proposition 1.4a in [Don90]. We include the proof as reference for

the remark following.

q(A) is closed in U by [Fog69] Lemma 5.4 and [Hab75]. Now, let I be the ideal

of A in V and let J be the ideal of q(A) in U . I has good G-module filtration by

definition, and hence has good G1 -module filtration by Proposition 4.4(a). Thus

H 1 (G1 , I) = 0 by Proposition 4.1(a) (with ξ = 0). Thus we have a commutative

diagram of G2 -modules:

0 /J / K[U ] / K[q(A)] /0

0 / I G1 / K[V]G1 / K[A]G1 /0

The rows are exact, and the middle vertical map, which is the restriction of the

comorphism q∗ , is an isomorphism. We have that q : A → q(A) is a surjection, so

the right hand map is injective, and therefore an isomorphism. Therefore q restricts

to a G1 -quotient A → q(A). We also have that J is isomorphic to I G1 , which

has good G2 -filtration by Proposition 4.4(c). Hence (U , q(A)) is a good pair of

G2 -varieties.

Remark 4.12. If (V, A) is a good pair of SO(V )-varieties such that A is a O(V )-

stable subvariety, then a O(V )-quotient map V → U restricts to a O(V )-quotient

A → q(A).

Indeed, we have H 1 (SO(V ), I) = 0 when G1 = SO(V ). Since H i (O(V ), I) =

H i (SO(V ), I)Z/2Z by I 6.9(3) of [Jan03], we have that H 1 (O(V ), I) = 0 when V is

a good SO(V )-variety. The argument above in the proof of Proposition 4.11 (from

the commutative diagram on) then shows that a O(V )-quotient q : V → U restricts

to a O(V )-quotient A → q(A) letting G1 = O(V ) and G = G01 × G2 .

70
4.2 Orthosymplectic Good Pairs

The following are results on good filtrations and good pairs in the orthosym-

plectic setting. One notes that it does not make sense to ask if a variety is a good

On -variety, since On is not connected. Instead, we identify certain varieties that are

good SOn -varieties. In other words, those that are good G0 (V )-varieties, where V

is a space of type ε = +1. In the symplectic setting, i.e. when V is a space of type

ε = −1, we have G0 (V ) = G(V ) in the results below.

Lemma 4.13. V is a good G0 (V )-variety.

Vi
Proof. [AJ84] proves in section 4.9 that each (V ∨ ) has a good filtration when

G(V ) = Sp(V ), proving by Lemma 4.8 that V is a good Sp(V )-variety. When
V
G0 (V ) = SO(V ), [AJ84] remark that when char K = p 6= 2, the i (V ∨ ) are of the

form H 0 (G/B, λi ) for some λi ∈ X(T )+ . This is sufficient for our purposes, but

it is worth remarking that they go on to prove directly that all Si (V ∨ ) have good

filtrations when p = 2.

Proposition 4.14. L(V, U ) is a good G0 (V )×G0 (U )-variety for the action (g, h)X =

hXg−1 , V an ε-space and U a −ε-space.

Proof. We first observe that we may identify L(V, U ) = Hom(V, U ) with V ∨ ⊗

U where G0 (V ) acts on V ∨ and G0 (U ) acts on U . We have already seen that

V V
the exterior powers i (V ) and j (U ∨ ) have good G0 (V )- and G0 (U )-filtrations,

respectively. Hence by considering the trivial action of G0 (V ) on U and of G0 (U )

on V , we have by Proposition 4.4(b) and Lemma 4.9 that V ∨ ⊗ U is a good G0 (V ) ×

G0 (U )-variety.

Remark 4.15. We have by the same argument that L(V, U ) is a good G0 (V )×GL(U )-

variety, or a good GL(V ) × G0 (U )-variety, since by [AJ84] sections 4.3 and 4.9, a

finite dimensional vector space V is also a good GL(V )-variety and the exterior
V
powers i (V ) have good GL(V )-filtrations.

Proposition 4.16. g(V ) is a good G0 (V )-variety, with action given by conjugation:

g.X = gXg −1 .

71
Proof. This is implied by [AJ84] Proposition 4.4 since we are assuming p 6= 2.

Proposition 4.17. Let V be an ε-space and let e ≤ dim(V ). (g(V ), g(V )e ) is a good

pair of G0 (V )-varieties, where g(V )e is the determinantal subvariety of matrices of

rank less than or equal to e.

First, we prove two intermediate lemmas.

Lemma 4.18. (L(V, W ), L(V, W )e ) is a good pair of G0 (V ) × GL(W )-varieties,

where W is a finite dimensional vector space over K with e ≤ dim(W ) ≤ dim(V )

and action given by (g, h)X = hXg −1 g ∈ G0 (V ), h ∈ GL(W ), X ∈ L(V, W ).

Proof. Let W ′ be a K-vector space with dim W ′ = dim W . The map

ψ
L(V, W ′ ) × L(W ′ , W ) / L(V, W )

given by matrix multiplication (X, Y ) 7→ Y X is a GL(W ′ )-quotient by [Don90,

1.4c(ii)]. We also have that L(V, W ) is a good G0 (V ) × GL(W ′ )-variety by the re-

mark following Proposition 4.14 and that (L(W ′ , W ), L(W ′ , W )e ) is a good pair of

GL(W ′ )×GL(W )-varieties by [Don90, 1.4c(i)]. Thus (L(V, W ′ )×L(W ′ , W ), L(V, W ′ )×

L(W ′ , W )e ) is a good pair of G0 (V ) × GL(W ′ ) × GL(W )-varieties by 4.6(b).

Hence ψ restricts to a GL(W ′ )-quotient L(V, W ′ ) × L(W ′ , W )e → L(V, W )e and

(L(V, W ), L(V, W )e ) is a good pair of G0 (V ) × GL(W )-varieties by 4.11.

Lemma 4.19. (L(V, U ), L(V, U )e ) is a good pair of G0 (V ) × G0 (U )-varieties, where

e < dim(V ) ≤ dim(U ) ≤ dim(V ) + 1 and U is a quadratic space of opposite type

from V .

Proof. We begin by downward induction on e.

First, let d = dim(V ) = dim(U ) and let e = d − 1. Then L(V, U )e is a good

complete intersection in L(V, U ). To see this, note that the ideal defining L(V, U )e

is generated by the determinant function, det. The rank 1 K-module E = K · det is

the trivial representation for the action of G0 (V ) × G0 (U ) , and so is isomorphic to

72
the induced module Y (0). Thus E has a good G0 (V ) × G0 (U )-module filtration, as

do the symmetric powers S p (E).

Now, let d = dim(V ) and dim(U ) = d + 1 with e = d − 1. Here, the ideal of

L(V, U )e is generated by det as well as the polynomials determined by the vanishing

of all d × d-minors, which can be indexed u1 , . . . , ud+1 by choosing a basis of U .

Letting E be the K-module generated by these, E = K ⊗ U , which also has a good

V
G0 (V ) × G0 (U )-filtration by Lemma 4.9 (we showed already that the i (U ) have

good G0 (V ) × G0 (U )-filtrations).

Finally, let d = dim(V ) ≤ dim(U ) ≤ dim(V ) + 1 and let e < d − 1. Assume

by induction that for m > e, (L(V, U ), L(V, U )m ) is a good pair of G0 (V ) × G0 (U )-

varieties. Let W be a vector space of dimension e. Then L(V, W )×L(W, U ) is a good

G0 (V )× GL(W )× G0 (U )-variety with action (g, h, f )(X, Y ) = (hXg −1 , f Y h−1 ). By

[Don90, 1.4c(ii)], the map

ψ
L(V, W ) × L(W, U ) / L(V, U )e

given by matrix multiplication is a GL(W )-quotient.

Consider the factorization of ψ

L(V, W ) × L(W, W ′ ) × L(W ′ , U ) / L(V, W ′ )e × L(W ′ , U ) / L(V, U )e

where W ′ is a vector space of dimension e + 1, both maps given by matrix mul-

tiplication. The first map is a GL(W )-quotient. We know that the multiplica-

tion in the second map, L(V, W ′ ) × L(W ′ , U ) → L(V, U )e+1 is a GL(W ′ )-quotient

([Don90, 1.4c(ii)] again). By the previous lemma, (L(V, W ′ ), L(V, W ′ )e ) is a good

pair of G0 (V ) × GL(W ′ )-varieties, so by 4.11, (L(V, U )e+1 , L(V, U )e ) is a good pair

of G0 (V )×G(U )-varieties, and L(V, W ′ )e ×L(W ′ , U ) → L(V, U )e is again a GL(W ′ )-

quotient (here, G1 = GL(W ′ ), G2 = G0 (V ) × G0 (U )).

By induction, (L(V, U ), L(V, U )e+1 ) is a good pair of G0 (V ) × G0 (U )-varieties,

hence (L(V, U ), L(V, U )e ) is a good pair of G0 (V ) × G0 (U )-varieties by 4.5(a).

73
We are now able to prove Proposition 4.17.

Proof of Proposition 4.17. The case e = dim(V ) is trivial, since by Proposition 4.16

g(V ) is a good G0 (V )-variety.

Now let e < dim(V ) and let U be a −ε-space with either dim(U ) = dim(V ) or

dim(U ) = dim(V ) + 1. Then the map

ρ
L(V, U ) / g(V )

given by X 7→ X ∗ X is onto and is a G(U )-quotient by Theorem 2.7. By the

previous lemma, (L(V, U ), L(V, U )e ) is a good pair of G0 (V )×G0 (U )-varieties, hence

ρ restricts to a G(U )-quotient L(V, U )e → g(V )e and (g(V ), g(V )e ) is a good pair of

G0 (V )-varieites (4.11 with G1 = G0 (U ) and G2 = G0 (V ), and the remark following).

4.3 Proof of Theorem 2.13(b)

Recall the definition of M := L(V0 , V1 )×L(V1 , V2 )×. . .×L(Vt−1 , Vt ) from section

2.3 and the subset Z ⊆ M defined by

X1 X1∗ = X2∗ X2

X2 X2∗ = X3∗ X3
..
(∗) .
∗
Xt−1 Xt−1 = Xt∗ Xt

Xt Xt∗ = 0

with Xi ∈ L(Vi−1 , Vi ). Recall as well that π(Xi ) = Xi Xi∗ ∈ g(Vi ) and ρ(Xi ) =

Xi∗ Xi ∈ g(Vi−1 ). We had n0 = dim(V0 ) ≥ n1 = dim(V1 ) ≥ . . . ≥ nt = dim(Vt ) >

0 = dim(Vt+1 ).

74
For 1 ≤ i ≤ t define

Mi = L(V0 , V1 ) × L(V1 , V2 ) × . . . × L(Vi−1 , Vi )

and define Gi := G(V0 ) × G(V1 ) × . . . × G(Vi ) which acts on Mi by

(g0 , g1 , . . . , gi )(X1 , . . . , Xi ) = (g1 X1 g0−1 , g2 X2 g1−1 , . . . , gi Xi gi−1

−1
).

In Mi , define Zi by the first i equations of (∗), except replacing the last Xi Xi∗ =
∗ X ∗
Xi+1 i+1 by Xi Xi ∈ Cλi ⊂ g(Vi ), i.e. is contained in the closure of the conjugacy

class of the nilpotent element D i+1 . Note that for all of the above, setting i = t

gives the original definitions of M and Z since in that case, D t+1 = 0 and so we have

Xt Xt∗ = 0 as in (∗). These Zi are in fact the same as the Zi in the fiber product

diagram of section 3.4.1. The action of each Gi is the same as the original action of

G(V0 ) × G(V1 ) × . . . × G(Vt ) restricted to the first i + 1 coordinates, hence we can

view Mi and Zi as G0t := G0 (V0 ) × G0 (V1 ) × . . . × G0 (Vt )-varieties via the projection

G0t = G0 (V0 ) × G0 (V1 ) × . . . × G0 (Vt ) → G0i := G0 (V0 ) × G0 (V1 ) × . . . × G0 (Vi ),

meaning that

(g0 , g1 , . . . , gt )(X1 , . . . , Xi ) = (g1 X1 g0−1 , g2 X2 g1−1 , . . . , gi Xi gi−1

−1
).

As before, Zi is a Gi -stable subvariety of Mi , thus also a Gt -stable subvariety.

Lemma 4.20. (M, Z) is a good pair of G0t -varieties for the action described above.

Proof. By Proposition 4.14, each L(Vi−1 , Vi ) is a good G0 (Vi−1 ) × G0 (Vi )-variety,

1 ≤ i ≤ t. Regarding L(Vi−1 , Vi ) as a Gt = G(V0 ) × G(V1 ) × . . . × G(Vt )-variety via

the action
−1
(g0 , g1 , . . . , gt )Xi = gi Xi gi−1 ,

i.e. letting G(Vj ), j 6= i − 1, i, act trivially, we get that L(Vi−1 , Vi ) is a good G0t -

variety for each 1 ≤ i ≤ t by Proposition 4.4(b). Hence each Mi is a good G0t -variety

by Proposition 4.6(a).

75
Let N = g(V1 ) × . . . × g(Vt ) and view N as a Gt -variety via the action

(g0 , g1 , . . . , gt )(Z1 , . . . , Zt ) = (g1 Z1 g1−1 , . . . , gt Zt gt−1 ).

N is a good G0t -variety by Propositions 4.16, 4.6(b) and 4.4(b).

Recall the map ζ : M → N given by

ζ(X1 , . . . , Xt ) = (X1 X1∗ − X2∗ X2 , X2 X2∗ − X3∗ X3 , . . . , Xt Xt∗ ).

ζ is Gt -equivariant and is surjective, as shown in section 3.4.3, and Z is the fiber

ζ −1 (0). Thus the comorphism ζ ∗ : K[N ] → K[M ] is injective, and identifying K[N ]

with S(N ∨ ), the symmetric algebra of the dual (as a Gt -module) of N , we have

that ζ ∗ (N ∨ ) is the span of the defining relations (∗) above for Z in M . We proved

in section 3.4.3 that Z is a complete intersection in M given by (∗). Since N is a

good G0t -variety, S(N ∨ ) ∼

= S(ζ ∗ (N ∨ )) has a good G0t -module filtration. Thus Z is a
good complete intersection and (M, Z) is a good pair of G0t -varieties by Proposition

4.7(a).

Lemma 4.21. For 2 ≤ i ≤ t, the projection ϕ : Mi → Mi−1 given by ϕ(X1 , . . . , Xi ) =

(X1 , . . . , Xi−1 ) induces a surjection Zi → Zi−1 .

∗
Proof. If X = (X1 , X2 , . . . , Xi ) ∈ Zi , then Xi Xi∗ ∈ Oλi and therefore Xi−1 Xi−1 =

Xi∗ Xi ∈ Oλi−1 by Proposition 2.8(a). Thus ϕ(X) ∈ Zi−1 .

∗
Conversely let Y = (X1 , X2 , . . . , Xi−1 ) ∈ Zi−1 . Then Xi−1 Xi−1 ∈ Oλi−1 so that
∗
Xi−1 Xi−1 = Xi∗ Xi for some Xi ∈ L(Vi−1 , Vi ) (Proposition 2.8(a)) with Xi Xi∗ ∈ Oλi

(Proposition 2.8(c)). Hence Y = ϕ(X1 , X2 , . . . , Xi−1 , Xi ) where (X1 , X2 , . . . , Xi−1 , Xi ) ∈

Zi .

Theorem 4.22. (a) (Mi , Zi ) is a good pair of G0i -varieties for 1 ≤ i ≤ t and

furthermore, (g(V ), OD ) is a good pair of G0 (V )-varieties.

(b) There is a map Zi → Zi−1 which is G(V0 )×G(V1 )×. . .×G(Vt )-equivarient and

76
a G(Vi )-quotient for 2 ≤ i ≤ t. Furthermore, there is a map ϑ : Z → OD which

is G(V0 ) × G(V1 ) × . . . × G(Vt )-equivarient and a G(V1 ) × . . . × G(Vt )-quotient.

Proof. For part (a), we proceed by downward induction on i. Lemma 4.20 is the case

i = t. Suppose now that 1 < i ≤ t and that (Mi , Zi ) is a good pair of G0i -varieties.

Consider the map ρi : Mi → Mi−1 × g(Vi−1 )ni given by ρi (X1 , X2 , . . . , Xi ) =

(X1 , X2 , . . . , Xi−1 , Xi∗ Xi ). ρi is a G(Vi )-quotient map by Theorem 2.7. Since

(Mi , Zi ) is a good pair of G0i -varieties, by Proposition 4.11, we have that (Mi−1 ×

g(Vi−1 )ni , ρi (Zi )) is a good pair of G0i−1 -varieties and the restriction Zi → ρi (Zi ) is a

G(Vi )-quotient. We have that (g(Vi−1 ), g(Vi−1 )ni ) is a good pair of G0 (Vi−1 )-varieties

by Proposition 4.17, and hence by Lemma 4.4(b) a good pair of G0i−1 -varieties let-
−1
ting G0 (V0 ) × . . . × G0 (Vi−2 ) act trivially (so that (g0 , g1 , . . . , gi−1 )X = gi−1 Xgi−1 ).

Then by Proposition 4.6(a), (Mi−1 × g(Vi−1 ), Mi−1 × g(Vi−1 )ni ) is a good pair of

G0i−1 -varieties, and hence (Mi−1 × g(Vi−1 ), ρi (Zi )) is a good pair of G0i−1 -varieties

by Lemma 4.5(a) (and also G0 (Vi−1 )-varieties by Proposition 4.4(b)).

Consider now the Gi−1 -equivarient map πi : Mi−1 → Mi−1 × g(Vi−1 ) given
∗ ). π is injective, and so gives an
by πi (X1 , . . . , Xi−1 ) = (X1 , . . . , Xi−1 , Xi−1 Xi−1 i

isomorphism of varieties Mi−1 ∼ ∗ ) ∈ M

= πi (Mi−1 ) = {(X1 , . . . , Xi−1 , Xi−1 Xi−1 i−1 ×

g(Vi−1 )}, as well as an isomorphism Zi−1 ∼

= πi (Zi−1 ). Indeed, the map πi is the
∗
identity in all but the last coordinate. We note that Xi−1 → Xi−1 Xi−1 is the

quotient map π : L(Vi−2 , Vi−1 ) → g(Vi−1 ), and so dπi is a surjection in the last

coordinate, hence in all coordinates. Since Zi−1 is a closed subvariety of Mi−1 ,

πi : Zi−1 → πi (Zi−1 ) is also an isomorphism of varieties.

Note that ρi (Zi ) ⊆ πi (Zi−1 ) ⊆ πi (Mi−1 ). Indeed, given (X1 , X2 , . . . , Xi−1 , Xi∗ Xi ) ∈

ρi (Zi ), we have that Xi∗ Xi = Xi−1 Xi−1

∗ since (X1 , X2 , . . . , Xi−1 , Xi ) ∈ Zi , hence

(X1 , X2 , . . . , Xi−1 , Xi∗ Xi ) = (X1 , X2 , . . . , Xi−1 , Xi−1 Xi−1

∗ ) ∈ π (Z
i i−1 ). Thus we

have a commutative diagram

ι
Zi / / ρi (Zi )  / πi (Zi−1 ) ∼
= 3 3 Zi−1
ϕ

where the projection ϕ from Zi to Zi−1 is a surjection by Lemma 4.21. Hence the

77
∗ )∈
inclusion from ρi (Zi ) to πi (Zi−1 ) is surjective. Indeed, given (X1 , . . . , Xi−1 , Xi−1 Xi−1

πi (Zi−1 ), (X1 , . . . , Xi−1 ) ∈ Zi−1 is the image of some (X1 , . . . , Xi−1 , Y ) ∈ Zi . Then

ρi (X1 , . . . , Xi−1 , Y ) = (X1 , . . . , Xi−1 , Y ∗ Y ) = (X1 , . . . , Xi−1 , Xi−1 Xi−1

∗ ) since by

the definition of Zi , Y ∗ Y = Xi−1 Xi−1

∗ . Thus (π ◦ ι)−1 : Z
i i−1 → ρi (Zi ) is an isomor-

phism, having proven earlier that πi is an isomorphism and ι is evidently just the

identity.

We have from Proposition 4.16 that g(Vi−1 ) is a good G0 (Vi−1 )-variety, hence

(g(Vi−1 ), {0}) is a good pair of G0 (Vi−1 )-varieties, and so (Mi−1 × g(Vi−1 ), Mi−1 ×

{0}) is a good pair of G0i−1 -varieties by Proposition 4.6. The map θ : Mi−1 ×
∗ −
g(Vi−1 ) → Mi−1 ×g(Vi−1 ) given by θ(X1 , . . . , Xi−1 , Y ) = (X1 , . . . , Xi−1 , Xi−1 Xi−1

Y ) is Gi−1 -equivariant and satisfies θ 2 = 1, hence is an isomorphism. We have that

θ(Mi−1 ×{0}) = πi (Mi−1 ), hence Mi−1 ∼

= πi (Mi−1 ). Thus (Mi−1 ×g(Vi−1 ), πi (Mi−1 ))
is a good pair of G0 (Vi−1 )-varieties by above (hence also a good pair of G0i−1 -varieties

by Proposition 4.4(b)). We already showed that (Mi−1 × g(Vi−1 ), ρi (Zi )) is a good

pair of G0i−1 -varieties, so applying Lemma 4.5(b) to ρi (Zi ) ⊆ πi (Mi−1 ) ⊆ Mi−1 ×

g(Vi−1 ), (πi (Mi−1 ), ρi (Zi )) is a good pair of G0i−1 -varieties, and by the preceding

paragraph, Zi−1 ∼
= ρi (Zi ), so (Mi−1 , Zi−1 ) is a good pair of G0i−1 -varieties.
For the last assertion of part (a), we let i = 1. We have by induction that

(M1 = L(V0 , V1 ), Z1 ) is a good pair of G01 = G0 (V0 ) × G0 (V1 )-varieties. The map

π1 = π : L(V0 , V1 ) → g(V0 )n1 = g(V )n1 given by π(X1 ) = X1 X1∗ is a G(V1 )-

quotient by Theorem 2.7. Recall from the discussion in section 2.3 that ϑ(Z) = OD

where ϑ(X1 , X2 , . . . , Xt ) = X1∗ X1 . Thus Z1 → π(Z1 ) = OD is a G(V1 )-quotient

and (g(V )n1 , OD ) is a good pair of G0 (V0 ) = G0 (V )-varieties by Proposition 4.11.

(g(V ), g(V )n1 ) is a good pair of G0 (V )-varieties by Proposition 4.17, implying that

(g(V ), OD ) is a good pair of G0 (V )-varieties by Lemma 4.5(a).

For part (b), we note from above that for 1 < i ≤ t, ρi restricts to a G(Vi )-

quotient Zi → ρi (Zi ). Thus (πi ◦ ι)−1 ◦ ρi : Zi → Zi−1 (given by the restriction of

the projection Mi → Mi−1 ) is a G(Vi )-quotient, since πi ◦ ι is an isomorphism of

varieties.

Composing these G(Vi )-quotient maps with the G(V1 )-quotient map π : Zi →

78
OD ,
/ Zt−1 / ... π
Z = Zt / Z1 /O
D

is a G(V0 ) × . . . × G(Vt )-equivariant, G(V1 ) × . . . × G(Vt )-quotient map Z → OD

given by (X1 , . . . , Xt ) 7→ (X1∗ X1 ). This is the map ϑ originally introduced in section

2.3.3.

79
Chapter 5

Conclusion

5.1 Final Results

The astute (determined? punctilious?) reader will notice that we have not

proven the promised result that an orbit closure is normal if and only if it is normal

in classes of codimension 2. We will prove this below by applying our generalization

of Kraft and Procesi’s result (Theorem 2.13) to Theorem 9.2 of [KP82] which shows
eD where
that OD is normal if and only if it is normal in O

[
eD = OD ∪
O Oi , codimOD Oi = 2.
i

First we state the following lemma, proven directly in [KP82, 9.1] for lack of a

suitable source. The proof given therein is independent of the characteristic of K.

Lemma 5.1. Let V be an affine Cohen-Macaulay variety, W ⊂ V a closed subset

of codimension ≥ 2. Then every regular function on V \ W extends to a regular

function on V, i.e. K[V \ W] = K[V] where K[V] denotes the ring of global regular

functions on a variety V.

eD denote the union of OD with those conjugacy classes Oi of

As above, let O
eD is then the complement of the union of those classes
codimension 2. The variety O

of codimension at least 4.

80
Proposition 5.2. eD extends to OD , i.e. K[O
(a) Every regular function on O eD ] =

K[OD ].

eD is normal, i.e. if OD is normal at all points

(b) OD is normal if and only if O

contained in orbits Oi of codimension 2.

eD ] and consider the quotient map ϑ : Z → OD . We have

Proof. For (a), let f ∈ K[O

from Proposition 3.10 that

eD ) ≥ 2
codimZ ϑ−1 (OD \ O

[
eD =
since OD \ O O. We can then define the regular function F := f ◦ ϑ
codimO O≥4
D
eD ).
on ϑ−1 (O

Because Z is a complete intersection in M , it is Cohen-Macaulay and thus by

eD ) is a closed
Lemma 5.1, F extends to a regular function on Z, since ϑ−1 (OD \ O

subset of codimension ≥ 2. Since ϑ is G := G(V0 )×. . .×G(Vt )-equivariant, g.F (x) =

eD ) and
F (g−1 x) = f ◦ ϑ(g −1 x) = f ◦ ϑ(x) = F (x). Hence F is invariant on ϑ−1 (O

thus also on Z, so we have F ∈ K[Z]G = ϑ∗ (K[OD ]). Since ϑ∗ is injective, F

eD extends
identifies with an extension of f to OD . Thus every regular function on O

to OD .
eD \ OD ) ≥ 2, so if O
For (b), we note that codimOeD (O eD is normal, then every

eD ([Eis95] Corollary
regular function on OD must extend to a regular function on O

11.4). By (a), they extend to OD as well. We note that OD is normal if and only if

every regular function on OD extends to OD , since codimOD (OD \ OD ) ≥ 2.

5.2 Summary of Results

In sum, we have shown that an affine variety Z exists for each nilpotent D ∈ g(V )

such that OD is a G-quotient of Z for some algebraic group G (Theorem 2.13). Z

is not necessarily normal, but it is Cohen-Macaulay. We use this fact in the proof

81
of Proposition 5.2 to show that OD is normal if and only if it is normal at all points

contained in codimension two orbits O ⊂ OD .

We summarize some additional intermediate results that may be of general in-

terest:

Theorem (3.7). If U is an orthogonal space, V is a   space and X ∈

symplectic
0 X
L(V, U ), let G e be the element 
e := GL(U ⊕ V ) and let X   ∈ End(U ⊕ V ).
X ∗ 0
e associated to nilpotent X
Then there is a cocharacter φ : K × → G e in G
e = GL(U ⊕V )

with φ(t) ∈ O(U ) × Sp(V ) for all t ∈ K × .

Proposition (3.10, generalized from char K = 0). Let U and V be quadratic spaces

of opposite types and X ∈ L(V, U ). If (X ∗ , X) is the orthosymplectic nilpotent pair

with associated ab-diagram τ , let OX denote the O(U )×Sp(V )-orbit of X ∈ L(V, U ).

Let π(OX ) denote the G(U )-orbit of XX ∗ ∈ g(U ) and let ρ(OX ) denote the G(V )-

orbit of X ∗ X ∈ g(V ). Then we have that

1
dim OX = (dim π(OX ) + dim ρ(OX ) + dim U · dim V − ∆τ )
2

X
where ∆τ = ai bi , ai the number of rows of τ of length i starting with a and bi
iodd
the number of rows of τ of length i starting with b.

Proposition (4.17). Let V be an ε-space and let e ≤ dim(V ). (g(V ), g(V )e ) is a

good pair of G0 (V )-varieties, where g(V )e is the determinantal subvariety of matrices

of rank less than or equal to e.

5.3 More Known Results

Kraft and Procesi go on to describe in [KP82] exactly which nilpotent orbits

under the action of the symplectic and orthogonal groups have normal closure when

char K = 0. This is done in terms of Young diagrams and irriducible minimal ε-

degenerations, of which there are finitely many. They also explicitly describe the

singularities occurring in these cases.

82
Table 5.1: Irreducible Minimal ε-degenerations
Type a b c d
Lie algebra sp2 sp2n so2n+1 sp4n+2
n>1 n>0 n>0
ε −1 −1 1 −1
η (2) (2n) (2n + 1) (2n + 1, 2n + 1)
σ (1, 1) (2n − 1, 2) (2n − 1, 1, 1) (2n, 2n, 2)
codimOε,η Oε,σ 2 2 2 2
Type e f g h
Lie algebra so4n so2n+1 sp2n so2n
n>0 n>1 n>1 n>2
ε 1 1 −1 1
η (2n, 2n) (2, 2, 12n−1 ) (2, 12n−2 ) (2, 2, 12n−4 )
σ (2n − 1, 2n − 1, 1, 1) (12n+1 ) (12n ) (12n )
codimOε,η Oε,σ 2 4n − 4 2n 4n − 6

Recall that an ε-degeneration σ ≤ η is minimal if σ 6= η and there is no ε-diagram

ν such that σ < ν < η.

Proposition ([KP82] Proposition 3.2). Let σ ≤ η be an ε-degeneration. Assume

that for two integers r and s the first r rows and the first s columns of η and σ

coincide and that (η1 , . . . , ηr ) is an ε-diagram. Denote by η ′ and σ ′ the diagrams

obtained by erasing these rows and columns of η and σ respectively, and put ε′ :=

(−1)s ε. Then σ ′ ≤ η ′ is an ε′ -degeneration and

codimO Oε′ ,σ′ = codimOε,η Oε,σ .

ε′ ,η ′

Though we do not give the proof of this here, it relies on the combinatorics of

ε-diagrams in sections 2.1.2 through 2.1.4 and does not depend on the character of

K.

In the setting of the proposition, an ε-degeneration σ ≤ η is obtained from the

ε′ -degeneration σ ′ ≤ η ′ by adding rows and columns. An ε-degeneration σ ≤ η is

then irreducible if it cannot be obtained from another ε′ -degeneration σ ′ ≤ η ′ by

adding rows and columns in a non-trivial way. We also have that any ε-degeneration

is obtained in a unique way from an irreducible ε′ -degeneration by adding rows and

columns, and that σ ′ ≤ η ′ is minimal if and only if σ ≤ η is minimal.

83
Table 5.1 classifies the irreducible minimal ε-degenerations. For t equal to one

of a, b, c, d, e, f, g, or h, an ε-degeneration is said to be of type t if it is obtained from

the corresponding minimal irreducible degeneration by adding rows and columns.

Types a, b, c, d, and e are of particular interest since they correspond to those classes

of codimension 2.

Kraft and Procesi show that Sing(Oε,η , Oε,σ ), the equivalence class of singular-

ities of Oε,η contained in Oε,σ (see [KP82, 12.1] for more detail), is equal to the

equivalence class of singularities of Oε′ ,η′ in Oε′ ,σ′ :

Theorem ([KP82] Theorem 12.3). Let the ε-degeneration σ ≤ η be obtained from

the ε′ -degeneration σ ′ ≤ η ′ by adding rows and columns. Then

Sing(Oε,η , Oε,σ ) = Sing(Oε′ ,η′ , Oε′ ,σ′ ).

Kraft and Procesi conclude in Theorem 16.2 that an orthogonal or symplectic

conjugacy class has normal closure if and only if its partition has no degenerations

of type e by computing Sing(Oε,η , Oε,σ ) for the irreducible minimal ε-degenerations.

These results do not extend to the connected components of the disconnected orbits

in the orthogonal case, the so-called “very even” orbits. These are addressed by

Sommers in [Som05] where he concludes that orbits of nilpotent elements in so(V )

under the action of the special orthogonal group have normal closure, again for

char K = 0. Whether these results generalize to positive characteristic without

alteration has yet to be seen. We hope to make progress towards this in the future.

84
Appendix A

Let U and V be finite dimensional K-vector spaces, K algebraically closed of

characteristic not 2, and let A : U → V and B : V → U be linear maps such that the

compositions BA and AB are nilpotent endomorphisms on U and V , respectively.

We describe here K-algebras A such that pairs P = {(U, V ), (A, B)} of vector

spaces and maps between them form a category of A-modules.

Let W be a K-vector space with basis A, B, and T . Then let S be the free

algebra
i
S := KhA, B, T i = K ⊕ W ⊕ (W ⊗ W ) ⊕ . . . ⊕ W ⊗ ⊕ . . .

and let I = hT 2 − 1, A(1 − T ), B(1 + T ), AT + T A, BT + T Bi be the (two-sided)

ideal. Denote by A the quotient S/I.

Consider an A-module M which is also finite dimensional as a K-vector space.

Multiplication by T defines an order two linear automorphism of M. Let

M+ = (1 + T )M and M− = (1 − T )M

be the +1 and −1 T -eigenspaces. Then M = M+ ⊕ M− is their direct sum.

Since A(1−T ) = 0 in A, we have AM− = 0. Likewise BM+ = 0 since B(1+T ) =

0 in A. Additionally, we have AM+ ⊆ M− and BM− ⊆ M+ since AT = −T A and

BT = −T B. Indeed, we have for m ∈ M+ , AT m = −T Xm = Xm, so Xm is in the

−1 T -eigenspace, namely M− . Similarly for m ∈ M− , BT m = −T Bm = −Bm.

Hence one can view A as a linear mapping M+ → M− and B as a linear

mapping M− → M+ .

85
If φ : M → N is an isomorphism of A-modules, then φ(M+ ) = N + and

φ(M− ) = N − . Thus such an isomorphism is classified by the action of GL(M+ ) ×

GL(M− ) on the pair (A, B) in Hom(M+ , M− ) × Hom(M− , M+ ).

Now, for n ≥ 1 let An = A/h(AB)n i. Since (AB)n = 0 in An , we must have

that (BA)n+1 = 0 as well. Thus for any An module M, both AB and BA act

nilpotently. Moreover, any A-module on which AB and BA act nilpotently can be

seen as a module for An for some n.

Thus given vector spaces U and V and a pair of linear maps A : U → V and

B : V → U such that AB and BA are nilpotent, U ⊕ V can be viewed as a An -

module for some n with U the +1 T -eigenspace and V the −1 T -eigenspace. This

module is then a direct sum of indecomposable modules, uniquely determined up to

isomorphism.

Given (AB)n = 0 with n minimal, we may have (AB)n A = 0 or (AB)n A 6= 0, but

as stated above, we must have B(AB)n A = (BA)n+1 = 0. Thus up to interchanging

U and V (and thus A and B), there are two indecomposable An -modules which are

not An−1 -modules, and which we can describe by abab . . . abab and abab . . . ababa,

using the notation in section 3.1.1.

86
Appendix B

Here we describe the preimage ϑ−1 (O) of an orbit O ∈ OD where ϑ : Z → OD

is the map (X1 , . . . , Xt ) = X1∗ X1 . Let λ be the ε-diagram of D.

Let (X1 , . . . , Xt ) be an element of Z. If τi is the ab-diagram of the pair (Xi∗ , Xi ),

we associate to (X1 , . . . , Xt ) the string of ab-diagrams (τ1 , . . . , τt ). Because of the

fiber product diagram describing Z in section 3.4.1, the τi must satisfy the following

conditions:

(i) π(τi ) = ρ(τi+1 ) = σi for 1 ≤ i ≤ t − 1, with σk ≤ λk where λk is the Young

diagram of Dk = D|Vk as in section 2.3.1, and

(ii) σt = 0, i.e. Oσt = 0.

We define σ0 := ρ(τ1 ).

ϑ−1 (O) is the set of tuples (X1 , . . . , Xt ) of Z such that σ0 = µ where µ is the

ε-diagram associated with the orbit O. The task then is to describe those strings of

ab-diagrams (τ1 , . . . , τt ) such that ρ(τ1 ) = µ for ε-diagrams µ ≤ λ.

First, we consider the case O = OD , i.e. λ = µ. By definition, we have |λ| =

dim(V0 ), |λ′ | = dim(V1 ), and in general |λi | = dim(Vi ) where λi = (λi−1 )′ . If we

draw the Young diagram of λ using b’s rather than boxes, we obtain an ab-diagram

τ1 with ρ(τ1 ) = λ by inserting a’s before, between and after the b’s, or by adding a

row with a single a. The number of a’s needed to “fill” the b-diagram of λ (to put a’s

between every two b’s) is exactly |λ′ |. This is the minimum number of a’s needed to

make an ab-diagram from the b-diagram. As long as the b-diagram we begin with is

indeed an ε-diagram, the ab-diagram obtained from filling the b-diagram is indeed an

87
orthosymplectic ab-diagram. Since we must have π(τ1 ) ≤ λ′ , there must be exactly

|λ′ | a’s in the ab-diagram τ1 , since the partial order defined on Young diagrams

compares partitions of the same number. Thus there is only one ab-diagram τ1 such

that ρ(τ1 ) = λ and π(τ1 ) ≤ λ′ . In fact, we have π(τ1 ) = λ′ , since λ′i = λi − 1 = #

of a’s needed to fill a row of λi b’s.

For example, if λ = (4, 3, 3, 2) then

bbbb bababab aaa

bbb babab aa
λ= τ1 = π(τ1 ) = = λ′
bbb babab aa

bb bab a

Now we need τ2 , and iteratively τi , such that ρ(τi ) = λi−1 and π(τi ) ≤ λi . Once

again, since we have exactly |(λi−1 )′ | a’s with which to fill the b-diagram of λi−1 ,

there is a unique τi .

In the above example,

bbb babab
aa
bb bab
λ′ = τ2 = π(τ2 ) = a = λ2
bb bab
a
b b

and finally

bb bab
λ2 = b τ3 = b π(τ3 ) = a = λ3

b b

Hence we have shown

Lemma. There is a unique stratum Zλ0 with λ0 = (τ10 , . . . , τt0 ) such that ρ(τ1 ) = λ.

In particular, τi0 is the ab-diagram of Di−1 = D|Vi−1 .

Proof. By definition, the ab-diagrams of the Di form such a string. Its unicity follows

88
from the above discussion and the τi0 are built by “filling” with a’s the ε-diagram
0 ), written as a b-diagram, and τ 0 is the filling of λ. This is also Lemma 5.4
of π(τi−1 1

of [KP79], though their discussion refers to ab-diagrams of general nilpotent pairs,

rather than orthosymplectic pairs.

Now we consider the case µ < λ. The key observation is that |µi | ≤ |λi | for

all i ∈ Z+ and for some (at least one) j ∈ Z+ , we must have |µj | < |λj |. At

this stage, when we go to construct τj from π(τj−1 ), we first fill the b-diagram of

π(τj−1 ) with |µj | a’s (in a unique way), but we must have exactly |λj | a’s in the

ab-diagram τj since π(τj ) = σj ≤ λj (remember, comparable partitions are of the

same number). So now we have an additional |λj | − |µj | a’s to add to our b-diagram,

which can be placed in many different positions, as long as the resulting ab-diagram

is orthosymplectic (some union of the indecomposable types of Table 3.1). This

restricts our choice for where to place the a’s, but does not guarantee a unique

ab-diagram.

In particular, we will have σj 6= µj (they are not even partitions of the same

number). While the stratum Zλ0 is the iterated fiber product above the ODi , the

other strata Zω with ω = (τ1 , . . . , τt ) and ρ(τ1 ) = µ 6= λ are iterated fiber prod-

ucts above the Oσi and not the Oµi . There are, however, finitely many ab-strings

(τ1 , . . . , τt ) with ρ(τ1 ) = µ, each corresponding to a choice of where to place a’s in

the diagrams τj for j such that |µj | =

6 |λj |. Hence there are finitely many Zω sitting

above each orbit O contained in the boundary OD \ OD .

89
Bibliography

[AJ84] Henning Haahr Andersen and Jens Carsten Jantzen, Cohomology of induced represen-
tations for algebraic groups, Math. Ann. 269 (1984), no. 4, 487–525.
[Bor91] Armand Borel, Linear algebraic groups, 2nd ed., Graduate Texts in Mathematics,
vol. 126, Springer-Verlag, New York, 1991.
[Bro98] A. Broer, Normal nilpotent varieties in F4 , J. Algebra 207 (1998), no. 2, 427–448.
[Chr06] A. L. Christophersen, A Classification of the Normal Nilpotent Varieties for groups of
Type E6 , Ph.D. Thesis, University of Aarhus, 2006.
[CPSvdK77] E. Cline, B. Parshall, L. Scott, and Wilberd van der Kallen, Rational and generic
cohomology, Invent. Math. 39 (1977), no. 2, 143–163.
[dCP76] C. de Concini and C. Procesi, A characteristic free approach to invariant theory,
Advances in Math. 21 (1976), no. 3, 330–354.
[Don81] Stephen Donkin, A filtration for rational modules, Math. Z. 177 (1981), no. 1, 1–8.
[Don85] , Rational representations of algebraic groups, Lecture Notes in Mathematics,
vol. 1140, Springer-Verlag, Berlin, 1985. Tensor products and filtration.
[Don90] S. Donkin, The normality of closures of conjugacy classes of matrices, Invent. Math.
101 (1990), no. 3, 717–736.
[DF73] Peter Donovan and Mary Ruth Freislich, The representation theory of finite graphs and
associated algebras, Carleton University, Ottawa, Ont., 1973. Carleton Mathematical
Lecture Notes, No. 5.
[Eis95] David Eisenbud, Commutative algebra, Graduate Texts in Mathematics, vol. 150,
Springer-Verlag, New York, 1995. With a view toward algebraic geometry.
[Fog69] John Fogarty, Invariant theory, W. A. Benjamin, Inc., New York-Amsterdam, 1969.
[Fri85] Eric M. Friedlander, A canonical filtration for certain rational modules, Math. Z. 188
(1985), no. 3, 433–438.
[FP86] Eric M. Friedlander and Brian J. Parshall, Cohomology of Lie algebras and algebraic
groups, Amer. J. Math. 108 (1986), no. 1, 235–253 (1986).
[Gab75] Pierre Gabriel, Représentations indécomposables, Séminaire Bourbaki, 26e année
(1973/1974), Exp. No. 444, Springer, Berlin, 1975, pp. 143–169. Lecture Notes in
Math., Vol. 431.
[Hab75] W. J. Haboush, Reductive groups are geometrically reductive, Ann. of Math. (2) 102
(1975), no. 1, 67–83.
[Har77] Robin Hartshorne, Algebraic geometry, Springer-Verlag, New York, 1977. Graduate
Texts in Mathematics, No. 52.
[Hes76] Wim Hesselink, Singularities in the nilpotent scheme of a classical group, Trans. Amer.
Math. Soc. 222 (1976), 1–32.
[Hum75] James E. Humphreys, Linear algebraic groups, Springer-Verlag, New York, 1975.
Graduate Texts in Mathematics, No. 21.
[Jan03] J. C. Jantzen, Representations of algebraic groups, 2nd ed., Mathematical Surveys and
Monographs, vol. 107, American Mathematical Society, Providence, RI, 2003.
[Jan04] , Nilpotent orbits in representation theory, Lie theory, Progr. Math., vol. 228,
Birkhäuser Boston, Boston, MA, 2004, pp. 1–211.

90
[Kos63] B. Kostant, Lie group representations on polynomial rings, Amer. J. Math. 85 (1963),
327–404.
[KR71] B. Kostant and S. Rallis, Orbits and representations associated with symmetric spaces,
Amer. J. Math. 93 (1971), 753–809.
[Kra89] H. Kraft, Closures of conjugacy classes in G2 , J. Algebra 126 (1989), no. 2, 454–465.
[KP79] H. Kraft and C. Procesi, Closures of conjugacy classes of matrices are normal, Invent.
Math. 53 (1979), no. 3, 227–247.
[KP81] Hanspeter Kraft and Claudio Procesi, Minimal singularities in GLn , Invent. Math.
62 (1981), no. 3, 503–515.
[KP82] H. Kraft and C. Procesi, On the geometry of conjugacy classes in classical groups,
Comment. Math. Helv. 57 (1982), no. 4, 539–602.
[LS88] T. Levasseur and S. P. Smith, Primitive ideals and nilpotent orbits in type G2 , J.
Algebra 114 (1988), no. 1, 81–105.
[Lev09] Paul Levy, Vinberg’s θ-groups in positive characteristic and Kostant-Weierstrass
slices, Transform. Groups 14 (2009), no. 2, 417–461.
[Lun73] Domingo Luna, Slices étales, Sur les groupes algébriques, Soc. Math. France, Paris,
1973, pp. 81–105. Bull. Soc. Math. France, Paris, Mémoire 33.
[Mat90] Olivier Mathieu, Filtrations of G-modules, Ann. Sci. École Norm. Sup. (4) 23 (1990),
no. 4, 625–644.
[Mat80] Hideyuki Matsumura, Commutative algebra, 2nd ed., Mathematics Lecture Note Se-
ries, vol. 56, Benjamin/Cummings Publishing Co., Inc., Reading, Mass., 1980.
[Naz73] L. A. Nazarova, Representations of quivers of infinite type, Izv. Akad. Nauk SSSR Ser.
Mat. 37 (1973), 752–791.
[RR85] S. Ramanan and A. Ramanathan, Projective normality of flag varieties and Schubert
varieties, Invent. Math. 79 (1985), no. 2, 217–224.
[Ser00] Jean-Pierre Serre, Local algebra, Springer Monographs in Mathematics, Springer-
Verlag, Berlin, 2000. Translated from the French by CheeWhye Chin and revised
by the author.
[Som03] E. Sommers, Normality of nilpotent varieties in E6 , J. Algebra 270 (2003), no. 1,
288–306.
[Som05] Eric Sommers, Normality of very even nilpotent varieties in D2l , Bull. London Math.
Soc. 37 (2005), no. 3, 351–360.
[Spr98] T. A. Springer, Linear algebraic groups, 2nd ed., Progress in Mathematics, vol. 9,
Birkhäuser Boston Inc., Boston, MA, 1998.
[Tho00] J. Thomsen, Normality of certain nilpotent varieties in positive characteristic, J. Al-
gebra 227 (2000), no. 2, 595–613.
[Vin76] È. B. Vinberg, The Weyl group of a graded Lie algebra, Izv. Akad. Nauk SSSR Ser.
Mat. 40 (1976), no. 3, 488–526, 709.
[VP72] È. B. Vinberg and V. L. Popov, A certain class of quasihomogeneous affine varieties,
Izv. Akad. Nauk SSSR Ser. Mat. 36 (1972), 749–764 (Russian).

91