K¨ahlerian structures of coadjoint orbits of semisimple Lie groups and their orbihedra
DISSERTATION
zur Erlangung des Doktorgrades der Naturwissenschaften an der Fakult¨atf¨urMathematik der Ruhr-Universit¨atBochum
vorgelegt von Patrick Bj¨ornVilla aus Bochum im Januar 2015 Contents
List of Figures ii
Introduction iii
1 Preliminaries 1 1.1 Convex geometry ...... 1 1.1.1 Polyhedra, Polytopes and Cones ...... 4 1.2 Complexification of irreducible real representations ...... 8 1.3 Almost complex structures ...... 10 1.3.1 Integrability ...... 10 1.3.2 Almost complex structures on homogeneous spaces ...... 10 1.4 Compatible subgroups ...... 12 1.5 Compact Centralizer ...... 13 1.6 Compact Cartan subalgebras ...... 15 1.6.1 Root decomposition ...... 15 1.6.2 Quasihermitian Lie algebras ...... 22 1.6.3 Hermitian Lie algebras ...... 24
2 Coadjoint orbits as K¨ahlermanifolds 26 2.1 Invariant almost complex structures ...... 27 2.1.1 The special case H = T ...... 27 2.1.2 The case T ⊂ H ⊂ K ...... 28 2.2 Integrable invariant almost complex structures ...... 29 2.3 Parabolics q− and Q− ...... 33 2.4 The K¨ahler structure ...... 35 2.5 Open embedding ...... 39
3 Coadjoint orbihedra 42 3.1 Momentum map ...... 42 3.1.1 Connected Fibers ...... 42 3.2 Face structure ...... 46 3.2.1 Faces as orbihedra ...... 46 3.2.2 All faces are exposed ...... 53 3.3 Correspondences of faces ...... 60 3.3.1 The correspondence of the faces of the polyhedron and the orbihedron 60 3.3.2 Complex geometry related to the faces ...... 63
i List of Figures
1 Nonexposed faces of a convex set E ...... 2 2 Root systems ...... 17 + 3 Choices of ∆p ...... 17 4 The maximal and minimal cone ...... 19 ∼ 5 The adjoint SL(2, R) = Sp(2, R)-orbits ...... 21 6 Sketch of the connecting curve c ...... 45 7 The orbit O and its projection ...... 56 8 The projection in the Sp(4, R)-case ...... 57 9 The projection in the SU(2, 1)-case ...... 57 10 Faces of the polyhedron up to equivalency ...... 62 11 Sketch of the general Sp(4, R)-orbihedron and its faces ...... 63 12 One-dimensional projections PG and PU ...... 69 13 Two-dimensional projections PG and PU ...... 70 14 Three-dimensional projections PG and PU ...... 71
ii Introduction
In 1923, Schur ([Schu]) proved that the diagonal entries of a Hermitian n × n-matrix with eigenvalues λ = (λ1, . . . , λn) are contained in the convex hull of
Sn · λ = {(λi1 , . . . , λin )|ij ∈ {1, . . . , n}, ij 6= ik for j 6= k} ,
n where Sn denotes the symmetric group acting on C by permuting the coordinates and is in a more general context to be known as the Weyl group. In 1954, Horn ([Hor]) proved that each point in the convex hull of Sn ·λ is the diagonal of a Hermitian matrix whose eigenvalues are λ. Kostant ([Ko]) generalized in 1973 the result of Schur and Horn to arbitrary compact Lie groups. In particular he formulated their result on Hermitian matrices as a property of an adjoint orbit O of the unitary group. More precisely, he proved that for an element x ∈ t, where t is a maximal abelian subspace in a compact Lie algebra k, one has a convex polytope given by
µ(Ad K · x) = conv(W · x),
where the group K is the connected analytic Lie group associated to k, the map µ : k → t is the orthogonal projection with respect to the Killing form and W is the Weyl group associated to (kC, tC). In 1982 Atiyah ([At]) and Guillemin and Sternberg ([GS]) gave independently a far reaching interpretation of Kostant’s result in terms of images of a momentum map with respect to a Hamiltonian torus action. A momentum map for a symplectic action of a Lie group G on a symplectic manifold (X, ω) is given by a G-equivariant map µ : X → g∗ from X to the dual of the Lie algebra g such that
ξ dµ = ιξX ω holds for all ξ ∈ g. Here the function µξ : X → R with µξ(x) := hµ(x), ξi is the component X of µ along ξ and ιξX ω is the contraction of the symplectic form ω with the vector field ξ on X induced by ξ ∈ g. If such a map exists the action of G on X is said to be Hamiltonian. Momentum maps admit considerable convexity properties of its images which have been studied extensively. The general result of Atiyah and Guillemin and Sternberg is that the set µ(X) is a convex polytope in t∗ where X is a compact connected symplectic manifold with a Hamiltonian torus action of a compact torus and µ : X → t∗ is the momentum map. Furthermore its vertices are given by images of T -fixed points in X. This symplectic convexity theorem was proven applying Morse theory on components of the momentum map. Furthermore they used the necessary fact that the symplectic manifold is compact. In the case of noncompact symplectic manifolds with a Hamiltonian torus action one can show similar results under the assumption that the momentum map is proper. In 1994, Hilgert, Neeb and Plank ([HNP]) used this to prove a convexity theorem for noncompact coadjoint orbits.
iii In this case one needs a Cartan subalgebra t ⊂ k in the Lie algebra g, where k is a maximal ∗ ∗ compact subalgebra in g. For any element α ∈ t the projection µ : Oα → t is proper if the convex envelope Obα := conv(Oα) contains no affine lines. Then one has
P := µ(Oα) = conv Wk · α + Cmin,
where Wk := NK (t)/ZK (t) is the Weyl group of t with respect to k and Cmin is a closed convex cone which is computed by the roots with respect to (gC, tC). In the same year Prato [Pr] proved a similar result using the properness of components of µ rather the properness of µ. These results show that the image P of the momentum map is a polyhedron, which generalizes the theorem in the compact setting In this work we are interested in the geometry of the convex hull of coadjoint orbits and its relation to the geometry of their momentum polyhedra P . The momentum map will play a major role in describing the geometry of the convex hull of coadjoint orbits. We will consider coadjoint orbits O of a reductive Lie group G such that its Lie algebra is quasihermitian and their momentum map µ is proper. Their convex hulls Ob := conv(O) will be studied extensively by means of µ. We will see that this convex body has a very rich boundary structure. The class of quasihermitian Lie algebras generalizes the class of compact Lie algebras in a natural way. In the case where G is a reductive Lie group its Lie algebra g is quasihermitian if and only if every simple ideal in g is either compact or hermitian. These types of Lie algebras admit a compact Cartan subalgebra t which is an essential ingredient in this work. Considering special elliptic coadjoint orbits (see Section 1.6.2) with a proper momentum map we prove the following theorem.
Theorem (3.35). All proper faces of Ob are exposed.
This theorem shows that every face of the orbihedron Ob is given by a supporting hyperplane. Moreover we will study the faces by means of the momentum map and we will show that every face has the same structure as Ob. More precisely the extremal points of any face are given as an orbit of a reductive and quasihermitian subgroup of G. This theorem generalizes the result of Biliotti, Ghigi and Heinzner in [BGH, Theorem 38] for compact Lie groups and their orbitopes, i.e. compact orbihedra. In order to prove this theorem one has to study components of the momentum map and their maxima. It becomes apparent that this property is more difficult in the noncompact setting than in the compact setting since not every continuous function on a noncompact set has a maximum. Using the result of [HNP] and [Pr] the image P of the t-momentum map of the orbit O is a polyhedron, which has only exposed proper faces. We will verify a strong connection between the faces of the orbihedron and its polyhedron. For this let F(P ) denote the set of faces of the polyhedron P and similarly let F(Ob) denote the set of faces of Ob. The Weyl group Wk acts on the set of faces F(P ) and the Lie group G acts on F(Ob). With respect to the actions on the set of faces we get the following theorem on the quotient spaces.
iv Theorem (3.37). The set of equivalence classes of faces F(P )/Wk and F(Ob)/G are in bijective correspondence.
This theorem draws a connection between the faces of the orbihedron Ob and the faces of its momentum image P . More precisely up to conjugation every face of Ob is completely determined by the faces of its momentum polyhedron P . In the special case where G is a compact Lie group, we obtain the result of Biliotti, Ghigi and Heinzner in [BGH, Theorem 49]. The coadjoint orbits we consider admit another property. In Chapter 2 we will study invariant almost complex structures to a high extent. In particular we will investigate which coadjoint orbits admit a homogeneous complex structure J such that their natu- ral symplectic structure ω is a K¨ahlerform. These complex structures relate to specific parabolic subalgebras in gC using the results of Fr¨olicher ([F]). Then following similar ideas of Bordemann, Forger and R¨omer([BFR]) any complex structure J relates to a closed set + C C of roots ∆J ⊂ ∆(g , t ) which corresponds to Weyl chambers in a subalgebra of t. + If the coadjoint G-orbit is K¨ahlerianthen the set of roots ∆J contains a set of positive + noncompact roots ∆p which is invariant under the Weyl group W . These sets of roots play a major role in describing the coadjoint orbits which are K¨ahlerian.In the case where G is semisimple, a coadjoint G-orbit may be identified with an adjoint G-orbit O through a point x ∈ t. Then the orbit O admits a K¨ahlerstructure (O, ω, J) if and only if there exist + + positive, Wk-invariant noncompact roots ∆p such that −iα(x) is positive for any α ∈ ∆p . It should be underlined that the complex structure J is the unique complex structure such that the orbit O can be embedded holomorphically into a generalized flag manifolds of the compact form U. This generalizes the case where the coadjoint orbit is a hermitian symmetric space and thus can be embedded into its compact dual (see [He, Proposition 7.14]). Moreover the K¨ahlerianorbits have a proper momentum map such that the previous theorems hold. In the compact case, i.e. G = K, every vector in g defines an exposed face to the orbitope and one can find a well defined representative by averaging over a compact group. Moreover, the complexification KC of the compact group acts on O as well. A further result in [BGH] is the correspondence of the faces to closed orbits of parabolic subgroups in O. These facts do not hold in the case where the Lie group G is noncompact. In particular in the noncompact setting the complexification does not act on O. Hence one can not expect to have a similar result. Nevertheless one may use the above open embedding of the orbit OG of G into an orbit OU of the compact form U. Using the complex structure J of Chapter 2 we may embed OG holomorphically into OU . Since the natural symplectic forms ωG and ωU are quite different due to the G- resp. U-invariance the induced momentum maps and their images also differ. Nonetheless the face structure of the momentum polyhedron PG and the momentum polytope PU are quite similar. Moreover we may describe the exposed faces of ObG and ObU in relation to each other with components of their momentum maps. The last goal of this monograph is to draw a connection between the faces of ObG and ObU . This is described in the following theorem.
v Theorem (3.46). There is a one to one correspondence between the set of faces of ObG and ObU with respect to the corresponding actions of G and U. Hence we get the isomorphisms on the quotients: ∼ ∼ ∼ F(ObG)/G = F(PG)/Wk = F(PU )/Wu = F(ObU )/U.
In this theorem the isomorphisms on the left and right hand side are given by the previous theorem or the respective result in the compact case. The isomorphism in the middle exposes an entirely new connection between the orbits OG and OU which shows that they behave similarly with respect to their convex hulls and their face structure.
Acknowledgements
First of all, I would like to thank my supervisor Prof. Dr. Peter Heinzner for introducing me to the research field, for many helpful and encouraging discussions during the last years and for critically reading the manuscript. Next I would like to thank my friends and members of the chair of complex analysis in Bochum, in particular Fabian Heising, Hannah Bergner, Dr. Valdemar Tsanov and Lisa Knauss. From each of them I have learned a lot. Moreover I thank Prof. Dr. Leonardo Biliotti and Dr. Alessandro Ghigi for several fruitful discussions while they were here in Bochum or I was visiting them in Parma. I am very grateful for the support of the Sonderforschungsbereich SFB/TR 12 of the Deutsche Forschungsgemeinschaft and the Faculty of Mathematics at Bochum. Last but not least I would like to thank my family for their invaluable help and all those who supported me personally while I worked on this thesis.
vi 1 Preliminaries
In this chapter we will review the most important facts and basic notations concerning the notions of convex geometry, almost complex structures and Lie groups which admit a compact Cartan subalgebra t in their Lie algebra. Most of the results in convex geometry are taken from [Schn]. I will assume the reader is familiar with the concepts of general Lie theory. Most of these facts are taken from [Kn] and [He]. For more details concerning Lie algebras admitting compact Cartan subalgebras the reader is referred to [Ne].
1.1 Convex geometry First we will recall some definitions and results of convex sets.Most of which can be found in [Schn]. In this section we denote by V a real vector space equipped with a fixed scalar product h , i and by E ⊂ V a convex set, i.e. a set E such that for any two points x, y ∈ E the closed segment [x, y] := {(1 − t)x + ty|0 ≤ t ≤ 1} is also contained in E. The relative interior of E, denoted relint(E), is the interior of E in its affine hull.
Definition 1.1 (Faces). A face F of E is a convex subset of E with the following property: if x, y ∈ E and relint([x, y]) ∩ F 6= ∅, then [x, y] ⊂ F . The extreme points of E are the points x ∈ E such that {x} is a face. The set of all extreme points of E is denoted by ext(E). An extreme ray of E is a ray that is a face of E. By extr(E) we denote the union of the extreme rays of E. A face distinct from E and ∅ will be called a proper face.
If E is a nonempty, closed and convex set then the faces are closed [Schn, p. 62]. For the description of a certain type of faces one needs to discuss hyperplanes in the vector space V .A hyperplane of V can be written in the form
Hu,α := {x ∈ V |hx, ui = α} with u ∈ V \{0} and α ∈ R. We say that u is a normal vector of Hu,α. The hyperplane Hu,α bounds the two closed halfspaces
− Hu,α := {x ∈ V |hx, ui ≤ α} ,
+ Hu,α := {x ∈ V |hx, ui ≥ α} .
Definition 1.2 (Hyperplanes/exposed faces). The hyperplane Hu,α supports E at x if − x ∈ E ∩ Hu,α and E ⊂ Hu,α. The hyperplane Hu,α is called the supporting hyperplane of E for u. If Hu,α is a supporting hyperplane then the set
Fu(E) := E ∩ Hu,α
is a face and is called the exposed face of E defined by u. Extreme points of E which are also exposed will be called exposed points and are denoted by exp E.
1 Figure 1: Nonexposed faces of a convex set E
Example 1.3. Figure 1 shows a convex set E which has faces that are not exposed. The set E is the convex hull of a circle and a square. The two intersection points are extreme points of E but are not exposed.
Lemma 1.4 ([BGH, Lemma 5]). If F is a face of a convex set E, then ext(F ) = F ∩ext(E).
Proposition 1.5 ([BGH, Proposition 7]). If F is an exposed face of a convex set E, the set CF := {u ∈ V : F = Fu(E)} is a convex cone. If G is a compact subgroup of O(V ) that preserves both E and F , then CF contains a fixed point of G.
Theorem 1.6 ([Schn, Theorem 2.1.2.]). If E is a nonempty closed convex set and F1,F2 are distinct faces of E then relint(F1)∩relint(F2) = ∅. To each nonemtpy relatively open convex subset A of E there is a unique face F of E with A ⊂ relint(F ). Therefore E is the disjoint union of the relative interiors of its faces.
The following two Lemmas taken from [BGH] refer to compact convex sets but the previous theorem only uses that E is nonempty and closed. Therefore these lemmas, which proofs base on the last theorem, extend to the case of closed convex sets.
Lemma 1.7 ([BGH, Lemma 9]). If E is a nonempty closed convex set and F ( E is a face, then dim F < dim E.
Lemma 1.8 ([BGH, Lemma10]). If E is a nonempty closed convex set and F ⊂ E is a face, then there is a chain of faces
F0 = F ( F1 ( ... ( Fk = E which is maximal, in the sense that for any i there is no face of E strictly contained between Fi−1 and Fi.
Lemma 1.9 ([BGH, Lemma 11]). If E is a convex subset of Rn, M ⊂ Rn is an affine subspace and F ⊂ E is a face, then F ∩ M is a face of E ∩ M.
The importance of the extreme points and extreme rays of a closed convex set becomes evident with the following theorem. There the closed convex set needs to be line-free, i.e. does not contain any line. Otherwise the set does not simply have any extreme points or rays.
2 Theorem 1.10 ([Schn, Theorem 1.4.3.]). Each line-free closed convex set A ⊂ Rn is the convex hull of its extreme points and extreme rays;
A = conv(ext(A) ∪ extr(A)).
Recall that for two sets A ⊂ V1,B ⊂ V2 the cartesian product A × B is convex if and only if A and B are convex. The following lemma shows how to compute the exposed points and the extreme points of the cartesian product.
Lemma 1.11. For two convex sets A, B we have
i) ext(A × B) = ext(A) × ext(B), ii) exp(A × B) = exp A × exp B.
Proof. i) For x = (x1, x2), y = (y1, y2) and z = (z1, z2) ∈ A × B it is straight forward that z ∈ relint([x, y]) if and only if zi ∈ relint([xi, yi]) for i = 1, 2. Hence i) follows. ii) We may assume that the exposed points will be zero in the respective vector space V1,V2 or V = V1 × V2. The scalar product on the product space V1 × V2 is the sum of the scalar products on each space V1 and V2. Furthermore let u1 ∈ V1, u2 ∈ V2 and u ∈ V1 × V2 be the respective normal vectors defining the exposed point. With the relation u = (u1, u2) we conclude ii) since
hu, (x1, x2)i = hu, (x1, 0)i + hu, (0, x2)i = hu1, x1i + hu2, x2i
for all x1 ∈ A, x2 ∈ B holds. If each summand on the right hand side is bigger than zero then the same holds for the left hand side. Otherwise if the left hand side is bigger than zero and u is a defining normal vector to the exposed point 0 then each summand is bigger than zero in the above equation.
3 1.1.1 Polyhedra, Polytopes and Cones In this section we will consider special types of convex sets, namely polyhedra, polytopes and cones. We will see that if we impose the finiteness of the sets ext(A) and extr(A) in Theorem 1.10 we get exactly these sets which are line-free.
Definition 1.12 (Cones). A cone C in Rn is a subset such that for any positive scalar λ the equation λC = C holds. In the following we distinguish some special types of cones.
a) A polyhedral cone is a subset C ⊂ Rn which is the intersection of finitely many closed linear half-spaces.
n b) For a finite number of vectors b1, . . . , bk ∈ R a set which is given by
( k )
X C := cone(b1, . . . , bk) := λibi λi ≥ 0 i=1 is called a finitely generated cone C.
c) The dual cone C∗ of a subset C in Rn is the set C∗ = {y| hy, xi ≥ 0 for all x ∈ C}. d) A cone C is said to be generating if C contains a basis of Rn. Remark 1.13. In fact C∗ is a closed convex cone for any subset C. If C is indeed a cone itself then C∗ is given by the set of elements which define supporting hyperplanes to C.
The concept of a polyhedral cone and being finitely generated are equivalent and this statement goes back to Minkowski ([Min]) and Weyl ([We]).
Theorem 1.14. A convex cone is polyhedral if and only if it is finitely generated.
Definition 1.15 (Polytopes and Polyhedra). a) A polyhedron is a subset P ⊂ Rn which is the intersection of finitely many closed affine half-spaces. b) A polytope is the convex hull of a finite set of points in Rn. In convex geometry it is a well known fact that the concepts of polyhedra and of polytopes are related. Again this statement goes back to Minkowski ([Min]) and Weyl ([We]).
Theorem 1.16. A set P is a polytope if and only if P is a bounded polyhedron.
Definition 1.17. For two subsets P,Q ⊂ Rn the set P + Q := {p + q| p ∈ P, q ∈ Q} is the Minkowski sum of P and Q.
4 Definition 1.18. a) The recession cone of a convex set P ⊂ Rn is the cone n rec(P ) := {y ∈ R | x + λy ∈ P for all x ∈ P, λ ≥ 0} . b) The lineality space of a polyhedron is the linear subspace
n lin(P ) := rec(P ) ∩ (− rec(P )) = {y ∈ R | x + λy ∈ P for all x ∈ P, λ ∈ R} . c) A polyhedron is pointed if lin(P ) = {0}. Lemma 1.19 ([Schr, p. 100-101]). Let P be a polyhedron. a) Any polyhedron P is the Minkowski sum of its lineality space and a pointed polyhedron, i.e. P = lin(P ) + ((lin(P ))⊥ ∩ P ). b) P + rec(P ) = P . c) P is bounded, i.e. a polytope, if and only if rec(P ) = {0}. d) If P = Q + C, with Q a polytope and C a polyhedral cone, then C = rec(P ). e) P has only finitely many faces. f) Each face of P is a polyhedron. g) If F is a face of P and F 0 ⊂ F , then F 0 is a face of P if and only if F 0 is a face of F . With the previous lemma and definitions the following theorem by Motzkin ([Mo]) states that any polyhedron can be decomposed into a sum of a polytope, a cone and a linear subspace. Theorem 1.20. A set P is a polyhedron if and only if P = Q+C for some polytope Q and some polyhedral cone C. More precisely any polyhedron P has a minimal representation
P = conv{x1, . . . xk} + cone{y1, . . . , yl} + lin(P )
where x1, . . . xk, y1, . . . , yl are orthogonal to lin(P ) and k, l are minimal. If P is pointed then x1, . . . , xk are the vertices of P and y1, . . . , yl are the generators of the extreme rays of rec(P ), which are unique up to scaling. Lemma 1.21. Let P = Q + C be a pointed polyhedron where Q is a polytope and C is a convex polyhedral cone. An element u defines a supporting hyperplane to P if and only if it supports C. Proof. Let u define a supporting hyperplane to P . Hence the number α := max hx, ui is x∈P defined and Hu,α supports the polyhedron P with normal vector u. After a translation the cone C lies in P and hence β := max hx, ui is also well defined. Thus the hyperplane x∈C Hu,β supports C with normal vector u. Conversely let u define a supporting hyperplane to C. Hence max hx, ui ≤ max hx, ui + max hx, ui is defined and induces a supporting x∈P x∈Q x∈C hyperplane.
5 The previous Lemma shows that it suffices to know which vectors support the recession cone of a polyhedron in order to know which vectors support the polyhedron. Now we want to show that every face of a polyhedron is exposed. Although this is a well known property of polyhedra the following proofs are based on [Schn, Chapter 2.4], which discusses polytopes, i.e. compact polyhedra.
n Proposition 1.22. Let P ⊂ R be a polyhedron, F1 an exposed face of P and F an exposed face of F1. Then F is an exposed face of P .
Proof. We may assume that P is pointed. Otherwise we may quotient Rn by the lineality space lin P . Later the hyperplanes in the quotient can be extended by lin P to acquire a n hyperplane in R . Furthermore we may assume that 0 ∈ F ⊂ F1. Since F1 is exposed there exists a support plane Hu,0 to P with normal vector u such that Hu,0 ∩ P = F1 and − P ⊂ Hu,0 hold. Considering F as an exposed face of F1 there exists a support plane H to F1 with H ∩ F1 = F inside the hyperplane Hu,0. Furthermore there is a vector v ∈ Hu,0 − with H = {x ∈ Hu,0| hx, vi = 0} and F1 ⊂ H = {x ∈ Hu,0| hx, vi ≤ 0}. Since P is a pointed polyhedron, i.e. it is line-free, we have P = conv(ext(P ) ∪ extr(P )) due to Theorem 1.10 where ext(P ) and extr(P ) are finite sets. For every extreme ray Ri ∈ extr(P ) we choose a point xi in Ri which is not an extreme point of P . In order to construct the supporting hyperplane to P and the face F we need the following numbers. Define hx, vi η0 := max − x ∈ ext(P ) \ ext(F ) hx, ui and hxi − x, vi η1 := max − xi ∈ Ri ∈ extr(P ) \ extr(F ), x extreme point of Ri and hxi − x, ui= 6 0 . hxi − x, ui
Note that hxi − x, ui ≤ 0 since Hu,0 supports P . Now we will show that the hyperplane Hηu+v,0 with η > max{η0, η1} supports P in F . In order to prove that we will show that every extreme point and every extreme ray of F lies − in Hηu+v,0. Furthermore any other extreme point or extreme ray of P will be in Hηu+v,0 but not in Hηu+v,0. • The extreme points:
– If x ∈ ext(P ) \ ext(F1) we have hx, ui < 0 hence
hx, ηu + vi < η0 hx, ui + hx, vi < 0.
– If x ∈ ext(F1) \ ext(F ) we have hx, ui = 0 and hx, vi < 0 hence
hx, ηu + vi = hx, vi < 0.
6 – If x ∈ ext(F ) we have hx, ui = 0 and hx, vi = 0 hence
hx, ηu + vi = 0.
• The extreme rays:
– If xi ∈ Ri ∈ extr(P ) \ extr(F1), x the extreme point to Ri with hxi − x, ui < 0 and λ > 0 we have
hx + λ(xi − x), ηu + vi < hx, ηu + vi + λ(η1 hxi − x, ui + hxi − x, vi) < 0.
– If xi ∈ Ri ∈ extr(P ) \ extr(F1), x the extreme point to Ri with hxi − x, ui = 0 and λ > 0 we have
hx + λ(xi − x), ηu + vi < hx, ηu + vi + λ hxi − x, vi < 0.
– If xi ∈ Ri ∈ extr(F1) \ extr(F ), x the extreme point to Ri and λ > 0 we have
hx + λ(xi − x), ηu + vi = hx + λ(xi − x), vi < 0.
– If xi ∈ Ri ∈ extr(F ), x the extreme point to Ri and λ > 0 we have
hx + λ(xi − x), ηu + vi = 0.
− Thus ext(F ), extr(F ) ⊂ Hηu+v,0 whereas ext(P ) \ ext(F ), extr(P ) \ extr(F ) ∈ int Hηu+v,0.
Corollary 1.23. Each proper face of a polyhedron is exposed.
Proof. We will prove this by induction on the dimension of the polyhedron. For zero- dimensional polyhedra there is nothing to prove. Assume that P is a polyhedron of dimen- sion n and that the assertion has been proved for polyhedra of smaller dimension. Let F be a proper face of P . Since every face is contained in some exposed face of P , there exists an exposed face F1 which contains F . Furthermore F is a face of F1. Now either F = F1 and then F is an exposed face of P , or F is a face of F1 and then an exposed face of F1, by the induction hypothesis. By Proposition 1.22, F is also an exposed face of P .
7 1.2 Complexification of irreducible real representations Let G be a Lie group and ρ : G → GL(V ) be an irreducible real representation on the vector space V . Moreover let J be a G-invariant complex structure on V i.e. g ◦ J = J ◦ g for all g ∈ G and J is a linear map on V with J 2 = −id. The aim of this section is to understand the representation of G on the complexified space V C. The space V C decomposes into two eigenspaces of J. The eigenspace to the eigenvalue i is V i = {X − iJX| X ∈ V } and the eigenspace to the eigenvalue −i is
V −i = {X + iJX| X ∈ V } .
The eigenspaces V i and V −i are complex, G-invariant subspaces of V C with V C = V i ⊕V −i.
Lemma 1.24. The induced representations of G on V i and V −i are irreducible.
Proof. The invariance of the eigenspaces with respect to G follows from the linearity of the representation and the invariance of the complex structure J. The map V → V i,X 7→ X − JX is R-linear, bijective and G-equivariant. Hence these vector spaces are isomorphic as G-vector spaces. Since the representation on V is irreducible the vector space V i resp. V −i has no real proper subspaces which are invariant under G. In particular it has no complex invariant proper subspaces.
Remark 1.25. The map V i → V −i,X − iJX 7→ X + iJX is C-linear, bijective and G-equivariant.
Now we will be interested in the G-invariant real resp. complex endomorphisms on V C G C G resp. V . Thus we will consider the map EndR(V ) → EndC(V ) . Under the assumption that the subspaces V i and V −i are non-isomorphic as complex representations of G we get the following theorem.
C G C Theorem 1.26. Let f ∈ EndC(V ) be a G-equivariant endomorphism of V . Then there exist complex numbers λ, µ ∈ C such that f(v + w) = λv + µw, for all v ∈ V i, w ∈ V −i.
Proof. The image f(V i) (resp. f(V −i)) is a G-invariant subspace of V C. Since the repre- sentation of G on V i and V −i is irreducible and V C = V i ⊕ V −i holds we conclude that the image f(V i) resp. f(V −i) lies in {0,V i,V −i}. The assumption V i V −i obstructs the case f(V i) = V −i resp. f(V −i) = V i. Hence we have f(V i) ⊂ V i and f(V −i) ⊂ V −i. Since we consider irreducible complex representations we can use the Lemma of Schur and get
f|V i = λIdV i und f|V −i = µIdV −i for complex numbers λ, µ ∈ C.
8 C G C Corollary 1.27. The ring EndC(V ) of G-equivariant endomorphisms on V is commu- tative.
Remark 1.28. The complex structure J is of the form J = iIdV i − iIdV −i . An important example where we complexify an irreducible representation will be de- scribed in the following example. The majority of these cases will be of this type. Example 1.29. Let V be the real vector space R2 with the usual representation of S1 given cos φ − sin φ by multiplication of 2×2 matrices of the form A = . This representation φ sin φ cos φ 1 0 −1 is irreducible of complex type, i.e. End ( 2)S = holds. Here J = is a S1- R R C 1 0 invariant complex structure on R2. The complexification of R2 is C2 with the decomposition into the following subspaces i x −y x 2 1 V = − i ∈ R = z z ∈ C y x y −i and −i x −y x 2 z 1 V = + i ∈ R = z ∈ C = z z ∈ C . y x y iz i
Moreover we have the following equations V i = V −i and V C = V i⊕V −i. The representation of S1 on V i resp. V −i is given as follows: z z A · = eiφ φ −iz −iz and z z A · = e−iφ . φ iz iz Here we see that V i and V −i are not isomorphic as complex representations.
C G Corollary 1.30. Let f ∈ EndC(V ) be a G-equivariant endomorphism, i.e. it is of the form f = λIdV i + µIdV −i , where λ, µ ∈ C. Then the restriction of f on V is G-equivariant if and only if λ = µ holds. Proof. For all X ∈ V we have the following continued equation. 1 f(X) = f(X − iJX + X + iJX) 2 1 = (λX − iJλX + µX + iJµX) 2 1 = ((λ + µ) X − iJ (λ − µ) X) . 2 Then V is invariant under the map f if and only if λ + µ ∈ R und λ − µ ∈ iR. This is exactly the case when λ = µ.
9 G Corollary 1.31. The algebra EndR(V ) of G-equivariant endomorphisms on V is isomor- phic to C.
G Proof. Every endomorphism f ∈ EndR(V ) may be complexified to an endomorphism on C V . Then f is of the form f = λIdV i + λIdV −i and it is completely determined by the complex number λ.
1.3 Almost complex structures An almost complex structure on a manifold can be seen as a complex structure on every tangent space of that manifold which varies smoothly. In this chapter we will consider the integrability of almost complex structures, i.e. if the manifold is a complex manifold. Fur- thermore we will discuss homogeneous almost complex structures on homogeneous spaces.
1.3.1 Integrability Definition 1.32. A smooth manifold M is called an almost complex manifold if it admits an almost complex structure J : TM → TM which is a smooth vector bundle isomorphism on the tangent bundle TM with J 2 = −id.
In general an almost complex manifold M does not need to be a complex manifold. But the following theorem (cf. [NN]) by Newlander and Nirenberg states that an almost complex manifold is indeed a complex manifold if the almost complex structure satisfies a certain torsion condition.
Theorem 1.33 (Newlander-Nirenberg). An almost complex structure J is integrable, i.e. is a complex structure, if and only if the torsion of J which is given by
NJ (X,Y ) = [X,Y ] + J [JX,Y ] + J [X,JY ] − [JX,JY ]
vanishes for all vector fields X,Y ∈ X(M) on M.
1.3.2 Almost complex structures on homogeneous spaces Let M be a manifold and G be a Lie group that acts transitively on M. The goal of this section is to understand almost complex structures on M which are compatible with the action of the Lie group G. These almost complex structures are called homogeneous and are completely determined in one base point. Let X0 ∈ M be a fixed point in M. The X0 stabilizer H = G of X0 is a Lie subgroup of G such that we can identify M with the quotient space G/H.
Theorem 1.34 ([F, Chapter 20]). The existence of a homogeneous complex structure on M = G/H is necessary and, if Ad H is connected, also sufficient if there exist in the complexified Lie algebra gC two complex conjugate Lie subalgebras which span gC and intersect each other in hC.
10 Here the complexified Lie algebra gC is equal to g ⊕ ig, where g (resp. ig) is the 1- eigenspace (resp. −1-eigenspace) of the conjugation. To get an idea of the proof of this theorem we start by identifying the manifold M with the homogeneous space G/H. Since the almost complex structure is assumed to be invariant under the action of the Lie group G it suffices to analyze the tangent space at eH. This tangent space can be identified with a subspace m of g which is complementary to h. So the existence of an integrable almost complex structure on G/H reduces to the existence of a linear map J on g with
J = 0 on h, (1) 2 m is J-invariant with J|m = − Idm, (2) [X,Y ] + J [JX,Y ] + J [X,JY ] − [JX,JY ] ∈ h for all X,Y ∈ g. (3)
(cf. [F, 19.15]). Here condition (3) is the torsion of J, i.e. corresponds to the integrability condition of Theorem 1.33, in terms of the Lie algebra g. Given two Lie subalgebras q, q0 ⊂ gC with q + q0 = gC, q ∩ q0 = hC and ¯q = q0 the complex structure is obtained in the following way:
J(X) = 0 for X ∈ hC, (4) J(X) = iX for X ∈ mC ∩ q (5) J(X) = −iX for X ∈ mC ∩ q0. (6)
The linear map J satisfies conditions (1) - (3) which is proven in [F, Chapter 20].
11 1.4 Compatible subgroups In general and for setting the notation for further chapters if G is a Lie group with Lie algebra g and E,F ⊂ g we set
EF := {η ∈ E| [η, ξ] = 0, for all ξ ∈ F }
and GF := {g ∈ G| Ad gξ = ξ, for all ξ ∈ F } . If F = {X} consists of only one element we simply write EX and GX instead of E{X} and G{X}. Let U be a compact Lie group and U C be its universal complexification in the sense of [Ho] which is a linear reductive complex algebraic group. Furthermore we denote by θ the conjugation map θ : uC → uC with respect to u and the corresponding group isomorphism θ : U C → U C. The map θ is then the Cartan involution in the sense of [Kn, Proposition VI.6.14]. Let φ : U × iu → U C be the diffeomorphism φ(g, ξ) = g · exp ξ.
Definition 1.35. A closed subgroup G ⊂ U C is said to be compatible with respect to U if for K := G ∩ U and p := g ∩ iu the equation φ(K × p) = G holds.
If G is a compatible subgroup then the restriction of φ to K × p is a diffeomorphism onto G. Furthmore the statements in this section apply to any compatible subgroups but in this work we are mostly concerned with a compatible Lie group G which is also a real form of U C = GC. Especially the compact subgroup K is the maximal compact subgroup of G such that the relations g = k ⊕ p,[k, p] ⊂ p and [p, p] ⊂ k hold. Note that G has finitely many connected components, is a closed linear group and is given as a set of fixed points of an involution on GC denoted by σ cf. Chapter 1.6. Moreover g is a real reductive Lie algebra, hence g = z(g) ⊕ [g, g]. Let Gss be the analytic subgroup of G with Lie algebra [g, g] then Gss is 0 0 closed and G = Z(G) ·Gss ([Kn, Corollary 7.11]), where Z(G) is the connected component of the centralizer of G containing the identity. Furthermore let B be the Killing form of g. Then by Bθ(·, ·) := −B(·, θ·) we define a positive definite, symmetric, bilinear form on g (see [Kn, 6.13, p. 355]).
Lemma 1.36 ([BGH2, Lemma 2.7]). Let G ⊂ U C be a compatible subgroup then the following assertions hold. a) If H ⊂ G is closed and θ-invariant, then H is compatible if and only if it has only finitely many connected components. b) The subgroup Gss is compatible. c) If E ⊂ g is a θ-invariant subalgebra, then GE is compatible.
12 1.5 Compact Centralizer In this section we will describe properties of elements X ∈ g which admit a compact centralizer GX . This leads for example to the fact that the induced torus exp(RX) is also compact. Later in this thesis we will see that the compactness of the centralizer does impose a restriction to the existence of an almost complex structure on the homogeneous space G/GX . In this section G denotes a connected semisimple real Lie group with finite center.
Definition 1.37. Let G, resp., H denote connected real Lie groups with Lie algebras g, resp., h and ρ : G → H a Lie group homomorphism. An element X ∈ g is called ρ-compact if exp dρ(RX) ⊂ H is compact. We are mostly interested in the case where ρ = Ad but the following proposition shows that it suffices to look at ρ = Id. Hence we drop the notion and call elements compact if they are Ad-compact or Id-compact.
Proposition 1.38. An element X ∈ g is Ad-compact if and only if it is Id-compact.
Proof. Since G is semisimple with finite center the map Ad : G → Ad(G) is a finite covering. This shows that Ad(M) = Ad(M) holds and the claim follows.
Definition 1.39. The set comp(g) := {X ∈ g|GX is compact} is the set of elements with compact centralizers.
Since GX is a closed subgroup of G with exp(RX) ⊂ GX , an element with compact centralizer is compact in the sense of Definition 1.37. Now let g = k ⊕ p be the Cartan decomposition of g and K be the maximal compact subgroup of G with Lie algebra k.
Lemma 1.40. An element X ∈ g is compact, if and only if there exists an element g ∈ G with Ad g X ∈ k.
Proof. Suppose that the Torus T = exp RX is compact. By Theorem 2.1, Chapter VI in [He], there exists a g ∈ G with gT g−1 ⊂ K. So the inclusion Ad g RX ⊂ k follows on the level of Lie algebras and we have proven the claim. Conversely we can assume that X lies in k. Using Proposition 1.38 we see that the inclusion Ad(exp RX) ⊂ Ad(exp RX) ⊂ Ad K holds. Since K is compact, every set in this chain of inclusion is compact. It shows that X is Ad-compact and with Proposition 1.38 the proof follows.
Lemma 1.41. If X ∈ g is compact, then the centralizer GX is connected and is the subgroup generated by exp gX .
Proof. A more general statement is proven in [Ne, Theorem VII.1.10].
13 Lemma 1.42. Let X ∈ k. Then the following conditions are equivalent.
1) X ∈ comp(g).
2) gX ∩ p = {0}.
3) ad X|p : p → p is an isomorphism. Proof. First we show the equivalence of 1) and 2). Let X ∈ comp(g), i.e. GX is compact. By Theorem 2.1, Chapter VI in [He], there exists a g ∈ G with GX ⊂ gKg−1. Hence gX ⊂ Ad g k. Since Ad g k is a compact Lie subalgebra of g, i.e. the Killing form is negative definite on Ad g k, and the Killing form is positive definite on p the identity p ∩ Ad g k = {0} holds. Therefore gX ∩ p = {0}. X X Conversely suppose g ∩ p = {0}. Let Y = Yk + Yp ∈ g with Yk ∈ k and Yp ∈ p. Since 0 = [X,Y ] = [X,Yk] + [X,Yp] with [X,Yk] ∈ k and [X,Yp] ∈ p we have 0 = [X,Yk] = [X,Yp]. X X Since g ∩ p = {0} the element Yp must be 0. Therefore g ⊂ k and X is compact. Thus GX ⊂ K because of Lemma 1.41 and we have shown that GX is compact. X The properties 2) and 3) are equivalent because of g ∩ p = ker(ad X|p).
14 1.6 Compact Cartan subalgebras Let G be a connected real semisimple Lie group with Lie algebra g. We say that the Lie algebra g contains a compact Cartan subalgebra t if there exists a Cartan decomposition g = k ⊕ p such that t is a Cartan subalgebra which lies in k. From now on we will assume that the Lie algebra g is always equipped with such a compact Cartan subalgebra t.
1.6.1 Root decomposition In this chapter we will state some results regarding roots of Lie algebras which will be needed for the construction of almost complex structures on homogeneous spaces. These roots will give us a tool to describe the Lie subalgebras discussed in Chapter 1.3.2. For a basic introduction into root systems we refer the reader to [Hu2].
Example 1.43. Our main example will be the Lie algebras sp(2n, R) and su(p, q) where we have compact Cartan subalgebras. The description of their Cartan subalgebras t ⊂ k of g are as follows. 0 D AB t = : D diagonal ⊂ k = : A = −At,B = Bt −D 0 −BA AB ⊂ sp(2n, ) = : B = Bt,C = Ct . R C −At t = {iD : D diagonal, T r(D) = 0}⊂ k A 0 = : A ∈ u ,B ∈ u , T r(A) + T r(B) = 0 0 B p q ⊂ su(p, q) AC = t : A ∈ u ,B ∈ u , T r(A) + T r(B) = 0 . C B p q
Let gC be the complexification of g. Then tC is a Cartan algebra in gC. We consider the involution σ of gC with Fixσ(gC) = g. Here σ(X + iY ) = X − iY , where X,Y ∈ g. ∗ For a linear functional α ∈ tC , we define the root space
C C C gα = X ∈ g |[H,X] = α(H)X for all H ∈ t
∗ C C C C C and write ∆ = ∆(g , t ) := α ∈ t \{0}|gα 6= 0 for the set of roots of g with respect C C C to t . We further put g[α] := gα ⊕ g−α ∩ g, where [α] := {α, −α}.
15 Lemma 1.44. The following assertions hold:
C C L C (i) g = t ⊕ α∈∆ gα.
C (ii) dim gα = 1 and Rα ∩ ∆ = {±α}.
(iii) α(t) ⊂ iR for all α ∈ ∆.
C C (iv) If Z ∈ gα, then σ(Z) ∈ g−α and [Z, σ(Z)] ∈ it.
C C C (v) g[α] = gα ⊕ g−α. For the proof of the previous lemma and more detail on the topic of roots we refer the reader to [Kn, Chapter II.4, II.5 and VI.7] and [Ne, Theorem VII.2.2, Lemma VII.2.3.]. We can show the first property of Lemma 1.44 (iv) in a more general sense. With the following lemma it follows that also the involution θ, given by the Cartan decomposition g = k ⊕ p, acts on the root spaces.
Lemma 1.45. Let τ be an involution, i.e. a Lie algebra homomorphism with τ 2 = Id, on C ¯ C C g which is C-linear and fixes t. Then τ(gα) = g−α for all α ∈ ∆.
Proof. Since τ fixes t and is C¯-linear it follows directly that τ(t) = t¯ for all t ∈ tC. C Now let ξ ∈ gα with [t, ξ] = α(t)ξ. Using τ on that equation we get [τ(t), τ(ξ)] hom.= τ[t, ξ] = τ (α(t)ξ) C−=lin. α(t)τ(ξ).
Inserting τ(t) instead of t = t1 + it2 we get [t, τ(ξ)] = α(τ(t))τ(ξ)
= α(t1 − it2)τ(ξ) α,C−lin. = (α(t1) − iα(t2))τ(ξ) 1.44(iii) = −α(t1) − iα(t2)τ(ξ) = −α(t)τ(ξ).
C C Definition 1.46. A root α in ∆ is called compact (resp. noncompact) if gα ⊂ k (resp. C C gα ⊂ p ) holds. The set of compact roots is denotet by ∆k and the set of noncompact roots by ∆p. Lemma 1.47 ([Ne, p. 234-236]). A root is compact (resp. noncompact) if and only if C α([Zα, σ(Zα)]) < 0 (resp. α([Zα, σ(Zα)]) > 0) holds for all Z ∈ gα. Furthermore the subspace k = t ⊕ L g is the unique maximal compact subalgebra of α∈∆k [α] g containing t. The corresponding Cartan decomposition is then given by g = k ⊕ p with p = L g . α∈∆p [α]
16 (a) The root system of Sp(4, R) (b) The root system of SU(2, 1) Figure 2: Root systems
Remark 1.48. Let α in ∆ be a root. Then we denote by tα ∈ t with −iα(X) = Bθ(tα,X) ∗ for all X ∈ t the coroot of −iα ∈ t with respect to Bθ. By abuse of notation we may call coroots just roots when in t.
Example 1.49. Figure 2a and 2b show the compact and noncompact roots of Sp(4, R) resp. SU(2, 1). The red points describe the noncompact roots and the blue points describe the compact roots. The lines are hyperplanes through zero which are orthogonal to some root in ∆. Definition 1.50. a) A subset ∆+ ⊂ ∆ is called a positive system if there exists an element X0 ∈ t such that + ∆ = {α ∈ ∆ : −iα(X0) > 0}
and α(X0) 6= 0 for all α ∈ ∆.
+ + + b) A positive system ∆ is called adapted if ∆p := ∆ ∩ ∆p is Wk-invariant, where Wk is the Weyl group with respect to t and k.
+ + (a) A set ∆p for Sp(4, R) (b) A set ∆p for SU(2, 1) + Figure 3: Choices of ∆p
+ Example 1.51. Figure 3a and 3b show possible choices of ∆p for an adapted positive + + root system in Sp(4, R) resp. SU(2,1). The only choices for ∆p in these cases are ∆p and + − −∆p =: ∆p . Any other choice of defining noncompact positive roots are not invariant under the Weyl group Wk. In these cases the group Wk is just given by a single reflection with respect to the line orthogonal to the compact root.
17 Recall that the Weyl chamber with respect to a positive root system ∆+ is defined as the set C := {X ∈ t| − iα(X) ≥ 0 for all α ∈ ∆+}. These Weyl chambers are in a natural one to one correspondence to positive root systems. Furthermore the respective Weyl group permutes the Weyl chambers and acts simply transitive on the set of Weyl chambers.
C Lemma 1.52. A root α is compact if and only if hZα, σ(Zα)i < 0 for all Zα ∈ gα. C Analogously α is noncompact if and only if hZα, σ(Zα)i > 0 for all Zα ∈ gα.
C Proof. Let α be a compact root, i.e. 0 > α([Zα, σ(Zα)]) holds for all Zα ∈ gα. Furthermore let tα be the coroot to −iα with respect to Bθ, i.e. −iα(X) = Bθ(tα,X) holds for all X in t, where B is the Killing form. Thus we get the following equation:
0 > α([Zα, σ(Zα)]) = −iα(i[Zα, σ(Zα)]) = Bθ (tα, i[Zα, σ(Zα)]) = − htα, i[Zα, σ(Zα)]i = −i h[tα,Zα], σ(Zα)i = −i hα(tα)Zα, σ(Zα)i = −iα(tα) hZα, σ(Zα)i = Bθ(tα, tα) hZα, σ(Zα)i . | {z } >0
Since Bθ is positive definite on g we have proven the lemma. For a noncompact root α the prove is analogous with 0 < α([Zα, σ(Zα)]). Lemma 1.53. Let t ⊂ k be a compact Cartan subalgebra of g. Then [ comp(g) ∩ t = t \ ker α.
α∈∆p Proof. Using Lemma 1.42 and Lemma 1.47 it follows that X ∈ comp(g) ∩ t if and only if X ∈ t and ad X is injective on p = L g . α∈∆p [α] X L X Remark 1.54. If X ∈ t we get g = t⊕ g[α]. Hence the centralizer g is completely α∈∆ α(X)=0 determined by the roots which vanish on X. The special case of a compact element X is described in the following lemma. Lemma 1.55. If X ∈ comp(g) ∩ t then
X M g = t ⊕ g[α].
α∈∆k α(X)=0 Proof. The inclusion “⊃” follows from X ∈ t and from the definition of roots. Conversely X P C suppose Y ∈ g with the root decomposition Y = Z + Zα with Z ∈ t and Zα ∈ gα. P α∈∆ Then 0 = [X,Y ] = α∈∆ α(X)Zα. By Lemma 1.53 we have α(X) 6= 0 for all α ∈ ∆p and thus we conclude Zα = 0 for all ∆p.
18 Definition 1.56. A Lie algebra g with compact Cartan subalgebra t and its respective roots ∆ is said to have cone potential if for any noncompact root α ∈ ∆p and any nonzero C vector Zα ∈ gα the inequality [Zα, Zα] 6= 0 holds.
The following cones Cmin and Cmax will play an important part in describing the mo- mentum polyhedra in the later chapters.
Definition 1.57. For an adapted positive system ∆+ of roots we define the cones
+ + Cmax := Cmax(∆p ) := {X ∈ t : −iα(X) ≥ 0 for all α ∈ ∆p } and + + gα Cmin := Cmin(∆p ) := cone{i[Zα, Zα]: α ∈ ∆p ,Zα ∈ C}, where for a subset M of a vector space, the set cone(M) denotes the smallest closed convex cone containing M.
Example 1.58. See Figure 4a for an example of Cmax and Cmin in the case of Sp(4, R) and see Figure 4b for an example of Cmax and Cmin in the case of SU(2, 1). The choices of + the noncompact positive roots ∆p are shown in Figure 3. Note that the negatives of these cones denote other maximal and minimal cones with respect to the negative set of roots − ∆p .
(a) Cones for Sp(4, R) (b) Cones for SU(2, 1) Figure 4: The maximal and minimal cone
0 D Example 1.59 (Sp(2n, ), see also Example 1.43). Given X ∈ n, let X = ∈ R R t −D 0 t be that element where diag D = X. Identifying t with Rn in this way, the Weyl group is given by all permutations of the entries of X ∈ Rn. The center of k is spanned by the vector (1,..., 1)t. The noncompact root vectors generate the convex cone Cmin = {Xt : Xj ≥ 0, 1 ≤ j ≤ n}, which is a positive orthant. Since an orthant is self-dual we have Cmax = Cmin using Theorem 1.76. Figure 4a describes the case of n = 2.
19 Example 1.60 (SU(p, q), see also Example 1.43). Let
p+q X X E = {(λ1, . . . , λp, µ1, . . . , µq) ∈ R : λi = µj}, 1 E = {(λ, µ) ∈ E : λi + µj ≥ 0 for all i, j}, 0 E = {(λ, µ) ∈ E : λi, µj ≥ 0}.
A 0 For X = (λ, µ) ∈ E denote X = ∈ t that element where diag A = −iλ t 0 B and diag B = iµ holds. The Weyl group acting on t consists of all permutations of the λi, together with all permutations of the µj separately. The center of k is spanned 0 by (1/p, . . . , 1/p, 1/q, . . . , 1/q)t. Then the noncompact root vectors generate Cmin = Et , 1 whose dual cone is Cmax = Et . Figure 4b describes the case of p = 2 and q = 1. For more details to these examples and for further classical examples the reader is referred to [Pa].
C Remark 1.61. For any X ∈ t and Zα ∈ gα we get the following continued equation:
Bθ(X, i[Zα, Zα]) = −i α(X)Zα, Zα = Bθ(X, tα) Zα, Zα = Bθ(X, tα Zα, Zα ).
Since Bθ is nondegenerate the equality i[Zα, Zα] = tα Zα, Zα holds. Hence for any non- compact root α the vector i[Zα, Zα] and the coroot tα differ only by the positive scalar
Zα, Zα using Lemma 1.52. We have shown that Cmin is just the closed convex cone spanned by the coroots of the positive noncompact roots.
Definition 1.62. For an adapted positive system ∆+ of roots we define the cones
Wmax := {X ∈ g|πt(OX ) ⊂ Cmax}
and Wmin := {X ∈ g|πt(OX ) ⊂ Cmin},
where OX is the adjoint G-orbit of X and πt is the projection onto the torus.
Example 1.63. Figure 5 shows the different types of adjoint orbits of the group Sp(2, R) in its Lie algebra sp(2, R). There are the trivial orbit which is the point zero, the two nilpotent orbits which are the upper and lower part of the cone, the one-sheeted hyperboloids and the connected components of the two-sheeted hyperboloids. The compact torus t is one-dimensional, equals the maximal compact subalgebra k of sp(2, R) and is represented by the vertical axis. As we have seen in Example 1.59 Cmin = Cmax is an orthant in t, i.e. in this case a half ray. Hence in the symplectic case we get Wmax = Wmin. In the case of Sp(2, R) the cone Wmax = Wmin is the convex hull of the closure of either the upper or the lower nilpotent orbit.
20 ∼ Figure 5: The adjoint SL(2, R) = Sp(2, R)-orbits
Theorem 1.64 ([HLV, p. 136, Theorem 5.3]). Let g be a finite dimensional real Lie algebra which contains a pointed generated invariant cone C0. Then there exists a compact Cartan subalgebra t ⊂ g and an adapted positive root system ∆+ satisfying the following conditions.
1. The chain of inclusions Cmin ⊆ C0 ∩ t ⊆ Cmax hold.
2. The invariant pointed generating cones in g containing Cmin are in one-to-one cor- respondence via C ↔ C ∩ t with those cones in t which are pointed and generate t, contain Cmin, are contained in Cmax and are invariant under the action of the Weyl group Wk. 3. If g is semisimple, then the largest and smallest pointed generating invariant cones Wmax and Wmin contain Cmin and they satisfy Wmax ∩t = Cmax and Wmin ∩t = Cmin.
21 1.6.2 Quasihermitian Lie algebras Let g be a real Lie algebra, k ⊂ g a maximally compact subalgebra and t ⊂ k a compact Cartan subalgebra of g.
Proposition 1.65 ([Ne, Proposition VII.2.14]). The following are equivalent:
(1) There exists an adapted positive root system.
(2) zg(z(k)) = k. Definition 1.66. A Lie algebra which satisfies one of the properties in Proposition 1.65 is called quasihermitian.
Since any two maximal compact subalgebras are conjugate under inner automorphisms, the condition zg(z(k)) = k is satisfied by one maximal compact subalgebra if and only if it is satisfied by all of them.
Lemma 1.67. Let g be a quasihermitian Lie algebra and Y ∈ k. Then gY is quasihermitian. More generally for any vector space V ⊂ k we get that gV is quasihermitian.
Y Y Proof. We have to show that zgY (zkY (k )) = k . The inclusion ”⊃ ” follows from the Y Y definition of the centralizer. On the other hand we have g ⊃ zgY (zkY (k )) ⊂ zgY (zk(k)) ⊂ Y Y zg(zk(k)) = k where we have used that zk(k) ⊂ zkY (k ) since Y ∈ k. Now zgY (zkY (k )) ⊂ gY ∩ k = kY . The second statement follows analogously. Later we will examine orbits of centralizers inside a G-orbit, where the the Lie algebra g is quasihermitian. Thus the lemmas in this chapter allow us to make similar conclusion on the orbits of centralizers inside the G-orbit.
Lemma 1.68. Let L be a Lie group, g ⊂ l a quasihermitian Lie algebra and g0 ∈ L. Then Ad g0 g is quasihermitian. Proof. The following calculation shows the lemma:
zAd g0 g (zAd g0 k(Ad g0 k)) = {Ad g0 ξ : ξ ∈ g, [Ad g0 ξ, zAd g0 k(Ad g0 k)] = 0} = {Ad g0 ξ : ξ ∈ g, [Ad g0 ξ, Ad g0 zk(k)] = 0} = {Ad g0 ξ : ξ ∈ g, [ξ, zk(k)] = 0} = Ad g0 {ξ : ξ ∈ g, [ξ, zk(k)] = 0} = Ad g0 zg(zk(k)) = Ad g0 k.
22 Lemma 1.69. Let g be a reductive quasihermitian Lie algebra with the decomposition z⊕gss into its center and its semisimple part and t a compact torus. Then the Lie algebra gss is quasihermitian.
Proof. We know that the Lie algebra g satisfies the condition zg(z(k)) = k. We want to show that the semisimple part gss has the same property. Since z ⊂ t we also get the 0 0 0 decomposition k = z ⊕ k . Furthermore the inclusion k ⊂ zgss (z(k )) together with the following continued equality prove the lemma:
0 0 0 0 z ⊕ zgss (z(k )) = zg(z(k )) = zg(z ⊕ z(k )) = zg(z(k)) = k = z ⊕ k .
0 0 Thus we get zgss (z(k )) = k . Remark 1.70. Among the simple real Lie algebras the compact Lie algebras, i.e. g = k, are trivially quasihermitian. The non-compact simple Lie algebras which are quasihermitian are those where z(k) is not trivial which then turns out to be one-dimensional. These kind of Lie algebras are called hermitian Lie algebras. They will be studied more in the next chapter. In the reductive case a Lie algebra is quasihermitian if and only if all its simple ideals are either compact or hermitian (see [Ne, p. 241]). Hence we get an ideal direct sum decomposition
n M g = z(g) ⊕ gj, j=1
where gj are either compact or hermitian simple ideals. Thus the classification of reductive quasihermitian Lie algebras corresponds to the classification of compact Lie algebras (which is the same as simple complex Lie algebras) and the classification of hermitian simple Lie algebras. Proposition 1.71 ([Ne, Propositions VIII.1.18, VIII.3.7]). Let g be a reductive quasi- hermitian Lie algebra, t ⊂ g a compact Cartan algebra and ∆+ an adapted positive root system. Then the following assertions hold:
i) The cone Cmin is pointed.
ii) Cmin ⊆ Cmax.
Fact ii) is equivalent to the existence of a closed convex generating invariant cone Wmax ⊂ g such that Wmax ∩ t = Cmax. Definition 1.72. A generating invariant convex subset C ∈ g is called elliptic if C0 ⊂ comp(g).
Example 1.73. In Example 1.63 where G = Sp(2, R) we have seen that the cone Wmax contains exactly one connected component of every two-sheeted hyperboloid. These are exactly the orbits which have T = K or conjugates of this group as their isotropy. Hence 0 Wmax ⊂ comp(sp(2, R)) holds and Wmax = Wmin is an elliptic cone in sp(2, R).
23 Theorem 1.74 ([Ne, Theorem VIII.3.10]). A Lie algebra g contains an invariant elliptic cone if and only if it contains a compact Cartan subalgebra t and there exists an adapted + positive root system ∆ with Cmin ⊆ Cmax. If this condition is satisfied and W ⊂ g is an invariant elliptic cone, then there exists a unique adapted positive system ∆+ such that W ⊆ Wmax. Then we also have Cmin ⊆ Cmax. In particular, the cones Wmax are maximally elliptic.
1.6.3 Hermitian Lie algebras A semisimple Lie algebra is hermitian if each of its simple ideals is hermitian, i.e. semisim- ple quasihermitian with no compact ideals. A hermitian Lie algebra g with Lie group G admits a hermitian symmetric space G/K. The reader is referred to [Se], [Vin] for the following result.
Theorem 1.75 (Kostant-Vinberg). A semisimple Lie algebra g contains a pointed, gen- erating invariant cone if and only if it is hermitian.
Moreover it is clear that any invariant convex cone in a semisimple Lie algebra g is contained in the direct sum of invariant cones in the simple summands. That is why we will mostly be concerned about the simple case. The next theorem describes more properties of generating invariant cones in hermitian simple Lie algebras. Therefore let Z0 ∈ k be an element which spans the one-dimensional center of k. Then Z0 ∈ t, by maximal commutativity of t, and there exists an adapted + + positive root system ∆ such that −iα(Z0) ≥ 0 for all α in ∆ (see [He, Corollary 7.3, 7.13]).
Theorem 1.76 ([Pa] or [HLV, p. 137, Theorem 6.2]). Let g be a hermitian simple Lie algebra. Then
∗ i) Cmin contains Z0 in its algebraic interior, and Cmax = (Cmin) , where the dual is taken in t.
ii) Each pointed generating invariant cone in g contains either Z0 or −Z0 in its interior. iii) For each pointed generating invariant cone C, we have (C ∩ t)∗ = C∗ ∩ t, where the first dual cone is computed in t and the second in g, both with respect to Bθ(·, ·).
iv) The cone Wmin is the smallest pointed generating invariant cone containing Z0 and ∗ Wmax = (Wmin) . Theorem 1.77 ([Ne, Theorem A.V.1]). If Π is a base of the root system ∆ of a complex simple Lie algebra g and β ∈ Π is a base root for which the coefficient of the highest root is 1, then there exists an involution σ : X 7→ X∗ on g turning g into an involutive Lie algebra with root decomposition such that for Πk = Π \{β} we have
+ ∆k = (spanZΠk) ∩ ∆ and ∆p = (β + spanZΠk) ∩ ∆
24 is contained in an adapted positive root system. Each real simple hermitian Lie algebra is isomorphic to one of the following:
(An) gR = su(p, q), g = sl(p + q, C), n = p + q − 1
(Bn) gR = so(2n − 1, 2), g = so(2n + 1, C), n ≥ 2
(Cn) gR = sp(2n, R), g = sp(2n, C), n ≥ 3
(Dn) gR = so(2n − 2, 2), g = so(2n, C), n ≥ 4 ∗ (Dn) gR = so (2n), g = so(2n, C), n ≥ 4
(E6) gR = e(6,−14),
(E7) gR = e(7,−25). Furthermore, we have the exceptional isomorphisms ∼ ∼ ∼ ∼ sl(2, R) = so(1, 2) = su(1, 1) = sp(2, R), so(2, 2) = sl(2, R) ⊕ sl(2, R), ∗ ∼ ∼ so (4) = sl(2, R) ⊕ so(3, R), so(3, 2) = sp(4, R), so(4, 2) ∼= su(2, 2), so∗(6) ∼= su(1, 3) and so∗(8) ∼= su(6, 2). Theorem 1.78 ([He, Chapter X §6.3]). The classification of all irreducible Hermitian symmetric spaces G/K are given in the second column in the following table. Their compact dual U/K is given in the third column.
Type G/K U/K Dimension A III SU(p, q)/S(Up × Uq) SU(p + q)/S(Up × Uq) 2pq BD I SO0(p, 2)/SO(p) × SO(2) SO(p + 2)/SO(p) × SO(2) 2p D III SO∗(2n)/U(n) SO(2n)/U(n) n(n − 1) CI Sp(2n, R)/U(n) Sp(2n)/U(n) n(n + 1) E III (e(6(−14)), so(10) + R)(e(6(−78)), so(10) + R) 32 E VII (e(7(−25)), e6 + R)(e(7(−133)), e6 + R) 54
In this work most examples we will consider are the groups SU(p, q) and Sp(2n, R).
25 2 Coadjoint orbits as K¨ahlermanifolds
∗ Let α ∈ g then the coadjoint orbit O := Oα := G · α may be identified with the corre- sponding homogeneous space G/Gα. For ξ ∈ g we consider the fundamental vector field O O d ξ defined by ξ (x) = dt 0 exp(tξ) · x for every x ∈ O. Note that the tangent space to the orbit at any point is given by evaluation of the fundamental vector fields in that point, in O particular we have Tβ(Oα) = ξ (β)|ξ ∈ g . It is a well known theorem that for any Lie group G every coadjoint orbit is a symplectic manifold. This is described in the following theorem. For a more detailed introduction into the topic of coadjoint orbits the reader is referred to [Be, Chapter 2.5] or [Sil, Part VIII]. A proof of the well known theorem is given for example in [Mi, Chapter VII.31.14].
Theorem 2.1 (Kirillov, Kostant, Souriau). If G is a Lie group, then a coadjoint orbit Oα carries a canonical symplectic structure ω which is invariant under the coadjoint action of G and is given in the point α by