SYMPLECTIC CONVEXITY THEOREMS AND APPLICATIONS TO THE STRUCTURE THEORY OF SEMISIMPLE LIE GROUPS
DISSERTATION
Presented in Partial Fulfillment of the Requirement for
the Degree Doctor of Philosophy in the Graduate
School of The Ohio State University
By
Michael Otto
*****
The Ohio State University 2004
Dissertation Committee: Approved by
Professor T. Kerler
Professor B. Kr¨otz,Co-Adviser Adviser
Professor R. Stanton, Co-Adviser Department of Mathematics
ABSTRACT
Atiyah’s well known convexity theorem states that for a Hamiltonian torus ac- tion T × M → M on a compact connected symplectic manifold M the image Φ(M) under the associated moment map Φ : M → t∗ is convex. Duistermaat in addition considered antisymplectic involutions τ on M satisfying Φ ◦ τ = Φ. He showed that
Φ(M) = Φ(Q) for Lagrangian submanifolds Q that arise as fixed point sets of such involutions.
We prove a generalization of Duistermaat’s symplectic convexity theorem for invo- lutions τ which satisfy several compatibility properties with the torus action, but which are not necessarily antisymplectic. By the same method we can also extend a
Duistermaat-type theorem for non-compact M with proper Φ.
All the symplectic convexity theorems mentioned have applications to the structure theory of semisimple Lie groups. With the generalization of Duistermaat’s theorem we are able to extend existing symplectic proofs for Kostant’s and Neeb’s convexity theorems to all semisimple groups. In addition we develop the symplectic framework to refine a recently discovered complex convexity theorem.
ii ACKNOWLEDGMENTS
First and foremost, I wish to thank Bernhard Kr¨otz for his role as adviser through- out my graduate studies. With his energy and enthusiasm he provided a great working environment.
I would also like to thank Bob Stanton for his support, especially during the process of writing this dissertation, and for his frequent valuable advice.
iii VITA
April 19, 1974 ...... Born — Braunschweig, Germany
2001 ...... Diplom, TU Clausthal/Germany 2000–present ...... Graduate Teaching Associate, The Ohio State University
PUBLICATION
B. Kr¨otzand M. Otto, Vanishing properties of analytically extended matrix coeffi- cients, Journal of Lie Theory 12, 2002, 409-421.
FIELDS OF STUDY
Major Field: Mathematics
iv TABLE OF CONTENTS
Page Abstract ...... ii Acknowledgments ...... iii Vita ...... iv
Chapters:
1. Introduction ...... 1
2. Symplectic Geometry ...... 7
2.1 Background ...... 8 2.2 Local normal forms ...... 18 2.3 Symplectic convexity theorems ...... 28
3. Lie theory ...... 41
3.1 Kostant’s convexity theorems ...... 42 3.1.1 Kostant’s theorems ...... 42 3.1.2 The linear version ...... 44 3.1.3 The nonlinear version ...... 47 3.2 Neeb’s convexity theorem for semisimple symmetric spaces . . 56 3.2.1 Statement of the theorem and reduction of the problem 57 3.2.2 A symplectic proof ...... 60 3.3 A refinement of the complex convexity theorem of Gindikin- Kr¨otz ...... 65 3.3.1 Notation and basic facts ...... 66 3.3.2 Complex groups ...... 69 3.3.3 The complex convexity theorem for non-complex groups 76 Appendix: Poisson Lie groups and Manin triples ...... 81 Bibliography ...... 86
v CHAPTER 1
INTRODUCTION
A complex (n × n)-matrix A is called Hermitian if A = A¯T . Here the bar sym- bolizes the matrix operation where each matrix entry is replaced with its complex conjugate. The exponent T denotes the transposed matrix. It is a well known fact that a given Hermitian matrix A is diagonalizable. This means one can find (com- plex) numbers λ1, . . . , λn, not necessarily distinct, and a complex matrix V with the property V −1 = V¯ T (such a V is called unitary), such that λ 1 −1 .. V AV = . . λn
The λi’s are the eigenvalues of A and can easily shown to be real numbers.
How do the entries of A depend on the λi’s, or in other words, how much infor- mation on the matrix A is encoded in its eigenvalues ?
A partial answer to this question was provided by Schur in 1923 [21]. He showed that the range of the diagonal elements of A was restricted by the eigenvalues. Let
1 us be more specific. We denote the n-dimensional vector of diagonal elements of A by diag(A), and we write the eigenvalues in vector form (λ1, . . . , λn). Note that both
n diag(A) and (λ1, . . . , λn) are points in R . For the finite set of points which are ob- tained from (λ1, . . . , λn) by permuting the entries we write Sn.(λ1, . . . , λn) (the orbit of (λ1, . . . , λn) under the symmetric group Sn). The convex hull conv(Sn.(λ1, . . . , λn))
n of the finite set Sn.(λ1, . . . , λn) is a compact region in R which is determined uniquely by the eigenvalues λ1, . . . , λn.
Schur’s result asserts that the diagonal vector must lie in this region, i.e.
diag(A) ∈ conv(Sn.(λ1, . . . , λn)).
It turns out that diagonal vectors of Hermitian matrices with eigenvalues λ1, . . . , λn not only lie in conv(Sn.(λ1, . . . , λn)) but actually make up all that region, i.e.
{diag(A): A Hermitian with eigenvalues λ1, . . . , λn} = conv(Sn.(λ1, . . . , λn)).
This was proven by Horn in 1954 [12].
Statements on square matrices can often be extended to semisimple Lie groups.
But it was not until 1973 that Kostant [14] proved such a generalization for the
Schur-Horn result.
Let G be a connected semisimple Lie group with Lie algebra g. The Iwasawa de- composition of G generalizes the Gram-Schmidt decomposition of a complex matrix g with determinant one into a unique product g = nak of an upper triangular ma- trix n with ones on the diagonal, a real diagonal matrix a with determinant one and a unitary matrix k with determinant one. The usual notation is G = NAK for
2 certain subgroups N, A, K of G that are unipotent, abelian, compact, respectively.
On the Lie algebra level the Iwasawa decomposition states g = n ⊕ a ⊕ k, where n = Lie(N), a = Lie(A), k = Lie(K). It is a generalization of the fact that any com- plex matrix with trace zero can be uniquely written as the sum of a strictly upper triangular matrix, a real diagonal matrix with trace zero and a skew-Hermitian ma- trix.
The group Sn from above becomes the Weyl group W = NK (a)/ZK (a). We denote bya ˜ : G = NAK → A, nak 7→ a, and pra : g = n + a + k → a,Xn + Xa + Xk 7→ Xa the middle projections in the Iwasawa decompositions. Then Kostant’s result is the following.
Kostant’s convexity theorems. Let Y ∈ a. Then
1. pra(Ad(K).Y ) = conv(W.Y ) linear version
2. loga ˜(K exp(Y )) = conv(W.Y ) nonlinear version where conv(·) denotes the convex hull of (·).
The Schur-Horn theorem is a special case of the linear version in case G =
SL(n, C).
Atiyah [2] took this subject into a new and unexpected direction. He related the
Lie theoretic results of Kostant with symplectic geometry. He, and independently,
Guillemin and Sternberg [6], discovered a symplectic convexity theorem from which
Kostant’s linear result could be derived.
3 AGS-Theorem. Let (M, ω) be a compact connected symplectic manifold. Sup- pose a torus T acts on M in a Hamiltonian way with moment map Φ: M → t∗.
Then Φ(M) is convex. More precisely, Φ(M) is the convex polytope spanned by the
finite set Φ(Fix(M)), where Fix(M) denotes the T -fixed points of M.
Subsequently, the AGS-theorem has been generalized in different directions by numerous authors. Some consider not only torus actions but more general Hamil- tonian actions. Others study non-compact symplectic manifolds. In many instances additional conditions on the moment map are required.
One of the early extensions of the AGS-theorems will be of particular interest to us.
It was given by Duistermaat [4] who considered certain involutions on the underlying symplectic manifold.
Duistermaat’s Theorem. Let (M, ω) be a compact connected symplectic man- ifold. Suppose a torus T acts on M in a Hamiltonian fashion with moment map
Φ: M → t∗. In addition, let τ : M → M be an antisymplectic involution which satisfies Φ ◦ τ = Φ.
If Q denotes the set of τ-fixed points of M, then Φ(Q) = Φ(M). In particular, Φ(Q) is convex. The statement remains true if one replaces Q with any of its connected components.
It turns out that antisymplecticity is a rather strong condition on the involution
τ that is not always satisfied in applications. One of the main results of this thesis is
4 the following theorem which states how the conditions on τ in Duistermaat’s theorem can be weakened.
Theorem A. [15] Let M be a compact connected symplectic manifold with Hamil- tonian torus action T × M → M and moment map Φ: M → t∗. Furthermore, let
τ : M → M be an involutive diffeomorphism with fixed point set Q such that
1. t ◦ τ = τ ◦ t−1 for all t ∈ T .
2. Φ ◦ τ = Φ.
3. Q is a Lagrangian submanifold of M.
Then Φ(Q) = Φ(M) = conv Φ(Fix(M)) = conv Φ(Fix(Q)). In particular Φ(Q) is a convex subset of t∗. Moreover, the same assertions hold if Q is replaced with any of its connected components.
Note that this theorem extends the class of Lagrangian submanifolds Q in Duis- termaat’s theorem for which the image Φ(Q) is still the full convex set Φ(M).
We will also consider the convexity theorem by Hilgert, Neeb and Plank [11] which deals with the situation that M is non-compact and the moment map Φ is proper.
We will show that, again, the condition of Φ being antisymplectic can be weakened as in Theorem A.
All the symplectic convexity theorems listed have interesting applications to the structure theory of semisimple Lie groups. The AGS- and Duistermaat’s theorem
5 can be used to give a symplectic proof of Kostant’s linear convexity theorem.
Lu and Ratiu found a way to put even Kostant’ s nonlinear theorem into a symplectic framework [18]. The use of Duistermaat’s theorem, however, restricts their method to special cases of groups G. We will show that by using Theorem A these restrictions can be overcome.
Neeb’s convexity theorem for semisimple symmetric spaces is a generalization of
Kostant’s (nonlinear) theorem. In Kostant’s theorem one considers orbits of the compact group K. This K comes about as fixed point group of the Cartan involution
θ on G. For arbitrary involutions σ on G the fixed point group H = Gσ is not necessarily compact anymore. Neeb’s theorem describes the image of certain H- orbits under a projection map which is the middle projection of an Iwasawa-type decomposition of G adapted to the involution σ. A symplectic proof has been given in [10], but it has the same constraints as the method of Lu and Ratiu for Kostant’s theorem. The proof can be finished by applying our version of Theorem A for non- compact symplectic manifolds.
Finally, we will use symplectic techniques to refine a complex convexity theorem recently discovered by Gindikin and Kr¨otz[5].
6 CHAPTER 2
SYMPLECTIC GEOMETRY
In this chapter we develop the symplectic convexity results mentioned in the introduction. These will serve as the main ingredients for the proofs in our Lie theoretic applications in the following chapter. However, this chapter is completely independent of the following one, being formulated entirely in symplectic geometric terms.
The first section introduces the fundamental concepts, in particular the notion of a Hamiltonian torus action with its corresponding moment map on a symplectic manifold. We are especially interested in involutions on this manifold which satisfy certain compatibility properties with the torus action. The second section is of a rather technical nature. Here, we consider local properties of the moment map and
find local normal forms. This local information will be used in the third section to derive the convexity theorems which state global properties of the moment map
(provided this map is proper). Major parts of this chapter appear in [15].
7 2.1 Background
We introduce notation and basic concepts in symplectic geometry. Most of the material can be found in standard text books such as [7].
Definition 2.1. A real vector space V equipped with a bilinear form Ω is called symplectic if Ω is antisymmetric and non-degenerate.
One can show by induction that for a symplectic vector space (V, Ω) one can always find a basis for V with respect to which Ω takes the matrix form 0 1 −1 0 . Ω = .. . 0 1 −1 0
In particular, a symplectic vector space V has to be of even dimension.
For a subspace W ⊂ V denote by W ⊥ its orthogonal complement with respect to
Ω. One calls W
isotropic, if W ⊂ W ⊥,
coisotropic, if W ⊥ ⊂ W,
Lagrangian, if W = W ⊥.
A smooth map φ :(V, Ω) → (V, Ω) is called symplectic if it preserves the symplec- tic form Ω, i.e.
Ω(x, y) = Ω(Dφ(v).x, Dφ(v).y) ∀ v, x, y ∈ V,
8 where Dφ(v) denotes the differential of φ at v ∈ V .
This notion for vector spaces can be generalized to manifolds in the following way.
Let M be a connected smooth manifold of dimension 2n. Let us denote the tangent
∗ space at a point m ∈ M by TmM, the cotangent space by TmM . A 2-form ω on M defines a smooth family of bilinear forms ωm on TmM × TmM.
Definition 2.2. A 2-form ω on M is called symplectic if it satisfies
(1) ωm(X,Y ) = −ωm(Y,X) ∀ m ∈ M,X,Y ∈ TmM.
∗ ∗ (2) ωm ∈ TmM ⊗ TmM is non-degenerate for all m ∈ M.
(3) dω = 0, i.e. ω is a closed form.
The pair (M, ω) is called symplectic manifold.
Note that the first two conditions define a family {(TmM, ωm): m ∈ M} of symplectic vector spaces, for which the symplectic forms ωm are unrelated. The closedness condition on ω forces some relation on the ωm’s. In particular, it implies that a symplectic manifold has local charts ϕi : Ui → V into a fixed symplectic vector
−1 space (V, Ω) such that the coordinate changes ϕi ◦ ϕj preserve Ω. A submanifold P ⊂ M is called
isotropic, if TpP ⊂ TpM is isotropic for all p ∈ P,
coisotropic, if TpP ⊂ TpM is coisotropic for all p ∈ P,
Lagrangian, if TpP ⊂ TpM is Lagrangian for all p ∈ P.
The symplectic form ω, as every 2-form on M, induces a mapping from the vector
fields on M to the set of differential forms: A vector field X gives a differential form
9 ι(X )ω = ω(X , ·).
Since ω is non-degenerate this mapping is invertible. In particular, for every smooth
∞ function f ∈ C (M) there is a unique vector field Xf such that df = ι(Xf )ω. This vector field Xf is called Hamiltonian, and f is its Hamiltonian function.
Definition 2.3. A smooth manifold M is called a Poisson manifold if there is a bracket operation {, } : C∞(M) × C∞(M) → C∞(M) such that
1. {, } is bilinear and antisymmetric,
2. For f, g, h ∈ C∞(M),
{f, {g, h}} + {g, {h, f}} + {h, {f, g}} = 0 (Jacobi identity),
3. For f, g, h ∈ C∞(M),
{f, gh} = {f, g}h + g{f, h} (Leibniz rule).
A symplectic manifold (M, ω) is automatically a Poisson manifold. Its Poisson bracket is defined by
∞ ∞ ∞ {, } : C (M) × C (M) → C (M), (f, g) 7→ {f, g} = ω(Xf , Xg).
For basic properties of general and some special Poisson manifolds, see the Appendix.
A map φ : M1 → M2 between two symplectic manifolds (M1, ω1) and (M2, ω2) is
∗ ∗ called symplectic if it preserves the symplectic forms, i.e. φ ω2 = ω1. Here, φ stands for the pull back. If Dφ denotes the differential of φ (at some point), then φ∗ is the extension of the dual map (Dφ)∗ to the exterior algebra of differential forms.
10 Let G be a Lie group that acts on the manifold M via diffeomorphisms,
G × M → M, (g, m) 7→ g.m =: φg(m).
The G-action is called symplectic if each g ∈ G acts as a symplectic map on (M, ω),
∗ i.e. φgω = ω ∀ g ∈ G. Let g denote the Lie algebra of G. Each X ∈ g defines a vector field X˜ on M via
˜ d Xm = exp(tX).m ∀ m ∈ M. dt t=0
Moreover, since ω is symplectic, X ∈ g gives rise to a differential form ι(X˜)ω. The form ι(X˜)ω is closed as the following argument shows. Recall ”Cartan’s magic for- mula”
LX = ι(X )d + dι(X )
which, for a vector field X , relates the Lie derivative LX , the inner product ι and the exterior differential d.
In our case ι(X˜)dω = 0, because ω is closed by assumption. Since X ∈ g and G was assumed to act symplectically,
d d ∗ LX˜ ω = φexp(tX)ω = ω = 0. dt t=0 dt t=0
It follows that dι(X˜)ω = 0.
If ι(X˜) turns out to be even exact with
˜ i(X)ω = dΦX , (2.1)
∞ we call ΦX ∈ C (M) a Hamiltonian function (for X ∈ g).
11 Definition 2.4. Suppose G acts on M in a way such that there exists a Hamiltonian function ΦX for every X ∈ g and such that the mapping
∞ g → (C (M), {, }),X 7→ ΦX is a G-equivariant Lie algebra homomorphism. Then we call the action of G on M
Hamiltonian. In this case the mapping Φ : M → g∗ of M into the dual g∗ of g defined by
hΦ(m),Xi = ΦX (m) ∀ m ∈ M,X ∈ g, is called the moment map for the Hamiltonian action of G on M.
We will be interested in Hamiltonian torus actions, i.e. in the case when G = T is a torus. Then we have a moment map Φ : M → t∗ where t denotes the abelian Lie algebra of T and t∗ its dual. Write Fix(M) for the set of T -fixed points in M. Note that
˜ d t.m = m ∀ t ∈ T ⇔ Xm = exp(tX).m = 0 ∀ X ∈ t dt t=0 ˜ ⇔ dΦX (m) = ι(Xm)ωm = 0 ∀ X ∈ t
⇔ dΦ(m) = 0.
Hence, Fix(M) is the critical set of the moment map Φ : M → t∗.
Let τ : M → M be an involutive diffeomorphism. We will denote by Q its fixed point set, i.e. Q = {m ∈ M : τ(m) = m}, and assume Q to be non-empty. Notice that Q is a closed submanifold of M.
We will be interested in involutions τ satisfying special conditions.
12 T -τ-compatibility: t ◦ τ = τ ◦ t−1 for all t ∈ T . (2.2)
Φ ◦ τ = Φ. (2.3)
Q is a Lagrangian submanifold of M. (2.4)
Remark 2.5. 1. Taking derivatives one sees that condition (2.2) is equivalent to
˜ ˜ dτ(m)Xm = −Xτ(m) ∀ m ∈ M,X ∈ t. (2.5)
In particular we have
˜ ˜ dτ(m)Xm = −Xm ∀ m ∈ Q, X ∈ t. (2.6)
2. Suppose τ is antisymplectic, i.e. τ ∗ω = −ω. Then (2.2) and (2.3) are equivalent,
and (2.4) is automatically satisfied.
3. Considering involutions τ with properties (2.2), (2.3) and (2.4) seems to be
a rather unmotivated generalization of the more natural setup of an antisym-
plectic involution satisfying (2.2). However, in the Lie theoretic applications of
the following chapters we will encounter quite natural involutions which satisfy
(2.2), (2.3) and (2.4) but which are not antisymplectic.
To illustrate the concepts introduced we want to give an example. The setup is that of a symplectic vector space. However, it is central to our treatment of more general symplectic manifolds in the next section where we linearize torus action and involution.
13 Example 2.6. Consider the symplectic vector space (R2n, Ω) where Ω has the matrix form 0 1 −1 0 . Ω = .. . 0 1 −1 0
The symplectic group Sp(2n) (with respect to Ω) is the group of invertible real
(2n) × (2n)-matrices A that preserve Ω, i.e. AΩAT = Ω. A linear symplectic action by the torus T is a representation ρ : T → Sp(2n). Since T is compact and abelian the elements of ρ(T ), viewed as operators on C2n, are simultaneously diagonalizable.
Therefore, on R2n one can choose a basis such that ρ : T → Sp(2n) differentiates to the Lie algebra homomorphism 0 λ (X) 1 −λ1(X) 0 . ρ : t → sp(2n),X 7→ .. , (2.7) 0 λn(X) −λn(X) 0
∗ for certain weights λ1, . . . , λn ∈ t .
The action of T is Hamiltonian with moment map
n X 1 Φ: 2n → t∗, (q , p , . . . , q , p ) 7→ λ (q2 + p2) (2.8) R 1 1 n n 2 j j j j=1
14 The linear involution τ given by 1 0 0 −1 . τ = .. 1 0 0 −1 is antisymplectic. It is immediate from (2.8) that Φ ◦ τ = Φ.
We shall investigate the condition of T -τ-compatibility in the linear case, i.e. when (M, ω) is a symplectic vector space (V, Ω) with linear torus action and linear involution τ.
Let (V, Ω) denote a finite dimensional symplectic vector space and τ : V → V a linear involution. As τ ◦ τ = idV , the linear operator τ is semisimple with eigenvalues +1 and −1. Accordingly we have an eigenspace decomposition V = V1 ⊕ V−1.
Next we endow (V, Ω) with a linear symplectic torus action T × V → V . Then V decomposes into fixed and effective part
V = Vfix ⊕ Veff , (2.9) where
Vfix = {v ∈ V : X.v = 0 ∀ X ∈ t}
= {v ∈ V : t.v = v ∀ t ∈ T }, and
Veff = t.V.
15 Notice that Ω remains non-degenerate when restricted to Vfix or Veff . Hence both Vfix and Veff become symplectic subspaces of V .
The following lemma gives equivalent characterizations of τ being antisymplectic.
Note that the first three statements do not rely on the existence of the torus action at all.
Lemma 2.7. Let (V, Ω) be a symplectic vector space and τ : V → V be a linear involution with eigenspace decomposition V = V1⊕V−1. Then the following statements are equivalent:
1. τ is antisymplectic.
2. V1 and V−1 are Lagrangian.
3. V1 is Lagrangian and there exists a symplectic isomorphism ϕ : V → V with
ϕ(V1) = V−1 and ϕ(V−1) = V1.
4. V1 is Lagrangian and there exists a linear symplectic torus action T × V → V
with the following properties:
(a) t ◦ τ = τ ◦ t−1 ∀ t ∈ T.
(b) Vfix = {0}.
Proof. For v ∈ V let v1 ∈ V1, v−1 ∈ V−1 be such that v = v1 + v−1.
(1) =⇒ (2): Assume that τ is antisymplectic. Then, for v, w ∈ V ,
Ω(v1, w1) = Ω (τ(v1), τ(w1)) = −Ω(v1, w1), and similarly,
Ω(v−1, w−1) = Ω (−τ(v−1), −τ(w−1)) = −Ω(v−1, w−1).
16 This implies that V1 and V−1 are isotropic. As V = V1 ⊕ V−1, it follows that both V1 and V−1 are maximally isotropic, i.e. Lagrangian subspaces of V .
(2) =⇒ (1): Assume that V1 and V−1 are Lagrangian. We compute
Ω(v1 + v−1, w1 + w−1) = Ω(v1, w1) + Ω(v−1, w1) + Ω(v1, w−1) + Ω(v−1, w−1)
= Ω(v−1, w1) + Ω(v1, w−1), and
Ω(τ(v1 + v−1), τ(w1 + w−1)) = Ω(v1 − v−1, w1 − w−1)
= Ω(v1, w1) − Ω(v−1, w1) − Ω(v1, w−1) + Ω(v−1, w−1)
= −Ω(v1 + v−1, w1 + w−1).
Hence, τ is antisymplectic.
(2) =⇒ (3): As Lagrangian subspaces both V1 and V−1 have dimension n =
1 2 dimV . Let {e1, . . . en} be a basis of V1. As Ω is non-degenerate, assumption (2) implies that there is a basis {f1, . . . fn} of V−1 such that Ω(ei, fj) = δij for all 1 ≤ i, j ≤ n.
Define ϕ by
ϕ(ei) = fi, ϕ(fi) = −ei ∀ i.
Then for all 1 ≤ i, j ≤ n,
Ω(ei, fj) = δij = Ω(ej, fi) = −Ω(fi, ej) = Ω(ϕ(ei), ϕ(fj)), completing the proof of (2) =⇒ (3).
(3) =⇒ (2): It suffices to show that V−1 is isotropic. But this follows from the fact that ϕ is surjective and that for v, w ∈ V1,
Ω(ϕ(v), ϕ(w)) = Ω(v, w) = 0.
17 (3) =⇒ (4): For t = Rϕ, the corresponding torus action by T = exp t clearly has the desired properties.
(4) =⇒ (3): Let T × V → V be a torus action which satisfies (4a) and (4b). We notice that (4a) is equivalent to its infinitesimal version
τ ◦ X = −X ◦ τ ∀ X ∈ t.
For any X ∈ t this implies
X(V1) ⊆ V−1 and X(V−1) ⊆ V1 .
If we set Y = exp X ∈ T , then
Y (V1) = V−1 and Y (V−1) = V1 .
Since Y ∈ T acts as a symplectic isomorphism on V we have found the isomorphism
ϕ in (3).
2.2 Local normal forms
This section is essentially Section 2.2 in [15]. We will assume that (M, ω) is a connected symplectic manifold endowed with a Hamiltonian torus action with mo- mentum map Φ. Also, we have an involution τ such that the T -τ-compatibility is satisfied. We denote the set of τ-fixed points of M by Q. Our objective is to pro- vide a local normal form for Φ Q near a point m ∈ Q. To that end we first recall a method of finding suitable local descriptions for ω and Φ in the neighborhood of a generic point m ∈ M. We then consider points m ∈ Q and obtain a refined form of
Φ Q which is adapted to the involution τ. We start with a simple observation (cf.[4], Lemma 2.1):
18
Lemma 2.8. Let m ∈ Q and X ∈ t. Then d(ΦX Q)(m) = 0 implies dΦX (m) = 0.
In particular, m is fixed under the action of the one parameter subgroup exp(RX).
Proof. Write E = TmM for the tangent space at m. Let E = E1 ⊕ E−1 be the decomposition of E into ±1-eigenspaces of the involution dτ(m). Notice that TmQ =
E1. In order to show that dΦX (m) = 0 it hence suffices to prove dΦX (m)(v) = 0 for all v ∈ E−1.
Let v ∈ E−1. Then it follows from (2.3) that
dΦX (m)(v) = −dΦX (m)(dτ(m)v) = −d(ΦX ◦ τ)(m)(v) = −dΦX (m)(v),
and so dΦX (m)(v) = 0. The last assertion in the Lemma follows from (2.1). This concludes the proof of the lemma.
For m ∈ M we write Tm for the stabilizer of T in m, i.e.
Tm = {t ∈ T : t.m = m}.
The Lie algebra tm of Tm is then given by
˜ tm = {X ∈ t : Xm = 0}.
If in addition m ∈ Q, then it follows from Lemma 2.8 that we can equally characterize tm by
tm = {X ∈ t : d(ΦX Q)(m) = 0}. (2.10)
Fix now m ∈ M. Next we provide local normal forms for ω and Φ near m. We will recall the procedure of momentum reconstruction (cf.[6], Chapter 41): The moment map near m is uniquely characterized by Φ(m), the stabilizer Tm and the linear representation of Tm on the tangent space TmM. This is even true for a general
19 compact Lie group T and in case of a torus was further exploited in [11], Sect. 2.
Notice that Tm(T.m) is an isotropic subspace of TmM. Thus ωm induces on the quotient
⊥ V = Tm(T.m) /Tm(T.m) a symplectic form Ω. We write Sp(V, Ω) for the corresponding symplectic group.
Notice that the isotropy subgroup Tm acts on TmM symplectically. Clearly this action
⊥ leaves Tm(T.m) and hence Tm(T.m) invariant, thus giving rise to a representation on V , say
π : Tm → Sp(V, Ω).
Write (Tm)0 for the connected component of Tm containing 1. Notice that (Tm)0 < T is a subtorus and so we can find a torus complement Sm to (Tm)0 in T , i.e.
T = Sm × (Tm)0.
We denote the Lie algebra of Sm by sm. Then t = tm ⊕ sm and we have a canonical
∗ ∗ ∗ ∗ identification t = sm × tm. Denote by T Sm the cotangent bundle of Sm with its
∗ ∗ canonical symplectic structure. We will use the identification T Sm = sm × Sm.
∗ Hence T Sm × V carries a natural symplectic structure. Further T = Sm × (Tm)0
∗ acts symplectically on T Sm × V via
(s, t).(β, s0, v) = (β, ss0, π(t)v), (2.11)
0 ∗ for s, s ∈ Sm, t ∈ (Tm)0, β ∈ sm and v ∈ V . It follows from [11], Lemma 2.1 that there is a symplectic diffeomorphism
∗ φ : T Sm × V ⊃ U → U ⊂ M
20 ∗ from an open neighborhood U of (0, 1, 0) ∈ T Sm × V to an open neighborhood U of m such that φ is locally T -equivariant and satisfies φ(0, 1, 0) = m.
∗ In the following we will identify U with U ⊂ T Sm × V via our symplectic, locally
T -equivariant chart φ : U → U. Write V = Vfix ⊕ Veff for the decomposition of V into effective and fixed part for the linear action of Tm on V (cf. (2.9)). Furthermore we have the tm-weight space decomposition V = ⊕λ∈ΛVλ. We decompose elements v ∈ V P as v = λ∈Λ vλ with vλ ∈ Vλ. Recall that there is a Tm-invariant complex structure
J on Veff such that hv, wi = Ω(v, Jw) defines a positive definite scalar product on
Veff . ( Choose a basis for Veff with respect to which an element X ∈ tm acts as in
(2.7) (the λi are not necessarily distinct). For some Z ∈ tm with λi(Z) 6= 0 ∀i, let J be the endomorphism of Veff with matrix representation ρ(Z). )
Then it follows from [11], Lemma 2.2 that the local normal form of Φ near a generic point m ∈ M is given on U by
1 X Φ: U → t∗ = s∗ × t∗ , (β, s, v) 7→ Φ(0, 1, 0) + β, kv k2λ . (2.12) m m 2 λ λ∈Λ λ6=0
Assume now that m ∈ Q. So far we have not adressed the question of the nature
∗ of Q and τ within our new coordinates in T Sm × V . And we would like to find a normal form for Φ Q. In case τ is antisymplectic on M, there is a beautiful answer (cf. Remark 2.10 below).
However, with our restricted assumption of T -τ-compatibility we cannot hope for such a nice form.
Near (0, 1, 0) the shape of Q is essentially determined by the linear involution σ = dτ(0, 1, 0) on E = T(0,1,0)U'TmM.
∗ ∗ We will use the natural identification E = sm × sm × V . Define W = sm × sm × Vfix.
21 Then it follows from (2.11) that E = W ⊕ Veff is the decomposition of E into fixed and effective part of the isotropy action of (Tm)0 on E 'TmM. Then (2.2) implies that the involution σ leaves the decomposition E = W ⊕ Veff invariant. Hence
σ 0 σ = W (2.13) 0 σ Veff
Accordingly we have a splitting