See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/226970816

Convexity properties of the moment mapping. II

Article in Inventiones mathematicae · October 1984

DOI: 10.1007/BF01388837

CITATIONS READS
211 238

2 authors, including:

Victor Guillemin
Massachusetts Institute of Technology
75 PUBLICATIONS 5,905 CITATIONS

SEE PROFILE

All content following this page was uploaded by Victor Guillemin on 17 February 2015.

The user has requested enhancement of the downloaded file.

 CONVEXITY PROPERTIES OF THE MOMENT MAPPING
RE-EXAMINED

REYER SJAMAAR
arXiv:dg-ga/9408001v2 5 Aug 1997

Abstract. Consider a Hamiltonian action of a compact Lie group on a com-

pact symplectic manifold. A theorem of Kirwan’s says that the image of the
momentum mapping intersects the positive Weyl chamber in a convex poly-
tope. I present a new proof of Kirwan’s theorem, which gives explicit infor-
mation on how the vertices of the polytope come about and on how the shape
of the polytope near any point can be read off from infinitesimal data on the
manifold. It also applies to some interesting classes of noncompact or singular
Hamiltonian spaces, such as cotangent bundles and complex affine varieties.

Contents
1. Introduction 1
2. Preliminaries 3
3. Semistability and convexity 6
4. Affine varieties 10
5. Stein varieties 20
6. Hamiltonian actions and convexity 24
7. Examples 29
References 35

1. Introduction
Let K be a compact Lie group acting smoothly on a compact symplectic mani-
fold M and suppose there exists a moment(um) map for the action. This map has
a host of interesting properties, one of the most important of which is the fact that
the intersection of its image with any Weyl chamber is a convex polytope, referred
to as the momentum polytope of M . This theorem, which is due to Kirwan, has
a long history, part of which I now summarize. Kostant proved a convexity the-
orem for torus actions on conjugacy classes and flag manifolds in [17]. Atiyah in
[2] and Guillemin and Sternberg in [5] dealt with the case of general Hamiltonian
torus actions. In their paper, Guillemin and Sternberg further proved a convex-
ity theorem for Hamiltonian actions of arbitrary compact Lie groups on integral
Kähler manifolds (or projective manifolds), which was also proved by Mumford in
[27]. Kirwan subsequently extended this result to Hamiltonian actions on arbitrary

Date: December 1994. Revised July 1997. To appear in Adv. in Math.

1991 Mathematics Subject Classification. Primary 58F06; Secondary 14L30, 19L10.
Key words and phrases. Momentum mappings, geometric quantization, geometric invariant
theory.
1
2 REYER SJAMAAR

compact symplectic manifolds in [14]. Many useful refinements in the projective-

algebraic case were made later by Brion in [3]. See [4], [11] and [23] for other results
and more references. See [9], [15] and [20] for some developments subsequent to the
present paper.
A striking difference between Kirwan’s general convexity theorem and the abelian
convexity theorem of Atiyah-Guillemin-Sternberg lies in the fact that the latter
offers far more quantitative information on the shape of the momentum polytope.
For example, in the abelian case one knows that the vertices of the polytope are
images of fixed points in M , and that the shape of the polytope near a vertex can
be read off from the isotropy action on the tangent space at a corresponding fixed
point. This follows from a combination of the equivariant Darboux Theorem and
Morse theory applied to the components of the momentum map.
The goal of this paper is to obtain such information in the nonabelian case as
well. The main result is Theorem 6.7, which is a sharpened version of Kirwan’s
convexity theorem. Given a point m in M mapping to a point µ in the momentum
polytope, it provides a description of the shape of the polytope near µ in terms
of the action of the stabilizer of m on polynomials on the tangent space at m. It
also states a necessary criterion for µ to be a vertex, which generalizes the criterion
for the abelian case referred to above. Other results include convexity theorems
for actions on affine varieties, Theorem 4.9, and cotangent bundles, Theorem 7.6.
Theorem 4.8 describes the relation between the momentum cone of an affine variety
and the momentum polytopes of its projective closure and the divisor at infinity.
These results are inspired by Brion’s treatment of Kirwan’s theorem for pro-
jective varieties. It came as a surprise to me how well Brion’s algebro-geometric
techniques can be adapted to a C ∞ setting essentially without sacrificing any of
their power. The main reason why this is possible is that every point in M possesses
an invariant neighbourhood that is isomorphic as a Hamiltonian K-manifold to (a
germ of) a complex quasi-projective variety.
In the language of the orbit method, the momentum polytope of M is the “classi-
cal” analogue of the set of highest weights of the unitary irreducible representations
occurring in the “quantization” of M . There is also a classical analogue of the space
of highest-weight vectors. This will be the subject of a forthcoming paper.
The paper is organized as follows. Section 2 is a review of some basic facts con-
cerning representations and momentum maps. Section 3 is a review of the convexity
theorem for complex projective varieties, where I have presented the argument in
such a manner that it can be applied to noncompact varieties. In Sections 4 and 5 I
prove convexity theorems for complex affine and Stein varieties. In Section 6 I ap-
ply these results to prove local convexity properties of arbitrary momentum maps,
whence I derive the main result, Theorem 6.7. The local description of the mo-
mentum polytope given by this theorem, although explicit, is unwieldy in practice,
and one often has to revert to ad hoc methods to calculate momentum polytopes.
In Section 7 I illustrate this in a number of examples, such as actions on cotangent
bundles and projective spaces.
I thank Sheldon Xu-Dong Chang, Yael Karshon and Eugene Lerman for their
help and encouragement. I am grateful to Laurent Laeng, Domingo Luna and the
referee for correcting a number of errors.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 3

2. Preliminaries
In this section I introduce notation and review basic material to be used later.
2.1. Groups, representations. Throughout this paper K will be a compact con-
nected Lie group with a fixed maximal torus T . The complexification of K is
denoted by G and the complexification of T by H. Let us fix a Borel subgroup B
of G containing H. Its unipotent radical (B, B) is denoted by N , and the corre-
sponding positive Weyl chamber in t∗ by t∗+ . The lattice ker(exp |t ) is denoted by
Λ. Its dual lattice
Λ∗ = HomZ (Λ, Z) ⊂ t∗
is the lattice of (real) weights and Λ∗+ = Λ∗ ∩ t∗+ is the monoid of dominant
weights.
√ To a real weight λ corresponds a character ζλ of T defined by ζλ (exp ξ) =
exp(2π −1 hλ, ξi) for ξ ∈ t.
The complex reductive group G = K C has a unique complex affine structure.
Let R = C[G]N be the algebra of polynomial functions on G which are N -invariant
on the right, that is to say, f ∈ R if f (gn) = f (g) for all g ∈ G and n ∈ N . Then G
acts on R by left multiplication and, since H normalizes N , H acts on R by right
multiplication. Under the right H-action, R has a weight space decomposition
M
R= Rλ , (2.1)
λ∈Λ∗
+

and it follows from the Borel-Weil Theorem that Rλ is an irreducible G-module

with highest weight λ. (See [18], Ch. III.) This implies that the algebra R is of
finite type and, hence, that the scheme G//N = Spec R “is” an affine variety. (I
shall not distinguish between an affine variety and the scheme associated to it.)
Let W = N (T )/T be the Weyl group of (K, T ) and let w0 be the longest Weyl
group element. Define an involution ∗ : t → t by µ∗ = −w0 µ. The complex-linear
extension of ∗ to tC and the dual map on (t∗ )C will also be denoted by ∗. It is well-
known that ∗ leaves the set of dominant weights invariant and that for all λ ∈ Λ∗+
the representation Rλ∗ is isomorphic to Rλ∗ , the contragredient representation of
Rλ .
2.2. Cones, polytopes. Let E be a finite-dimensional vector space over Q and
let S be a subset of E. The Q-convex hull of S is the smallest convex subset of E
containing S and is denoted by hullQ S. The convex hull of S is the smallest convex
subset of E ⊗ R containing S and is denoted by hull S. A subset of E (resp. E ⊗ R)
is called a cone if it is invariant under multiplication by nonnegative rational (resp.
real) scalars. The convex Q-cone spanned by S is the smallest convex cone in E
containing S and is denoted by coneQ S. In other words, coneQ S = Q≥0 · hullQ S.
The convex cone spanned by S is the smallest convex cone in E ⊗ R containing S
and is denoted by cone S. That is, cone S = R≥0 · hull S. If S is finite, hull S and
cone S are called a rational convex polytope, resp. a rational convex polyhedral cone
in E ⊗ R. A cone is called proper if it does not contain any linear subspaces (apart
from {0}).
2.3. Hamiltonian actions. Let (M, ω) be a symplectic manifold with a K-action
defined by a smooth map τ : K × M → M . The action τ is called Hamiltonian if
there exists a momentum map, that is, a map Φ : M → k∗ with the property that
dΦξ = ι(ξM )ω for all ξ ∈ k. Here ξM denotes the vector field on M induced by ξ,
and Φξ is the function defined by Φξ (m) = Φ(m) (ξ). We may, and will, assume


4 REYER SJAMAAR

Φ to be K-equivariant with respect to the coadjoint action on k∗ . The quadruple

(M, ω, τ, Φ) is called a Hamiltonian K-manifold. (See e.g. [7].)
If Y is any subset of M , we denote the restriction of Φ to Y by ΦY . The
momentum set ∆(Y ) of Y is defined by
∆(Y ) = Φ(Y ) ∩ t∗+ .
In “good” cases, ∆(Y ) is known to be a convex cone or polytope ([5], [14]), and is
then called the momentum cone, resp. polytope of Y .
On every Hamiltonian K-manifold (M, ω, τ, Φ) there exists an almost-complex
structure J that is compatible with the Hamiltonian action, that is to say, J : T M →
T M is a symplectic map, the symmetric bilinear form ω(·, J·) is positive definite,
and J is K-equivariant. If in addition J is√integrable, then M is a Kähler manifold
with K-invariant metric ds2 = ω(·, J·) − −1 ω(·, ·), and K acts holomorphically.
Example 2.1 ([5]). Let (V, ωV ) be a symplectic vector space and assume K acts on
V by linear symplectic transformations. This action is Hamiltonian; a momentum
map is given by the quadratic map
1
ΦξV (v) = ωV ξv, v ,


(2.2)
2
where ξv denotes the image of v ∈ V under ξ ∈ k (viewed as a linear operator
on V ). Choose a K-invariant ωV -compatible complex structure J on V . Let h·, ·i
be the Hermitian inner product whose imaginary part is equal to −ωV . Then the
momentum map can also be written as
√
−1
ΦξV (v) = hξv, vi. (2.3)
2
Now suppose L K = T is a torus. Then V is an orthogonal direct sum of weight
spaces, V = ν∈Λ∗ Vν . If Vν 6= 0, then ν is called a weight of the symplectic
action of T on V . The weight space decomposition depends on the choice of the
complex structure, but the weights do not. (This follows from the fact that any two
K-invariant compatible complex structures on V are conjugate by a √ K-equivariant
linear symplectic map.) If v is a vector of weight ν, then ξv = 2π −1 ν(ξ)v, so
Φ(v) = −πkvk2 ν by (2.3), and therefore
∆(V ) = − cone{ν1 , . . . , νl }, (2.4)
where ν1 , . . . , νl are the (real) weights of V .

Example 2.2 ([13],[27],[1]). Let V , J and h·, ·i be as in the previous example, and
let PV be the space of complex lines in V . The natural K-action on PV leaves the
Fubini-Study symplectic form invariant, and with the volume of PV normalized to
1, a momentum map is given by
√
ΦV (v) −1 hξv, vi
ΦPV ([v]) = = , (2.5)
πkvk2 2π kvk2
where [v] denotes the line through v. Consequently, if K is a torus and v is a weight
vector in V with weight ν, then ΦPV ([v]) = −ν. Hence,
∆(PV ) = − hull{ν1 , . . . , νl }, (2.6)
where ν1 , . . . , νl are the weights of V . See further Section 7.1.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 5

2.4. Coadjoint orbits. For every µ in t∗+ the coadjoint orbit Kµ with its Kirillov-
Kostant symplectic form ωµ is a Hamiltonian K-manifold. The momentum map is
simply the inclusion ιµ : Kµ → k∗ . (See [16], [7].) Let Pµ be the parabolic subgroup
(Kµ )C N of G. The K-equivariant diffeomorphism G/Pµ → Kµ sending the coset
1Pµ to the vector µ provides Kµ with a complex structure with respect to which
ωµ is Kähler.
Now let µ be a dominant weight. Then the cohomology class of the form ωµ on
Kµ is integral and because Kµ is compact, √ there exists a Hermitian holomorphic
line bundle Oµ on Kµ with curvature −2π −1 ωµ . The pullback of Oµ to G/Pµ is
just the homogeneous line bundle G ×Pµ C, where Pµ acts on C by the character µ.
Let µ1 and µ2 be two points in t∗+ and let µ = µ1 + µ2 . Then Kµ = Kµ1 ∩ Kµ2
and Pµ = Pµ1 ∩ Pµ2 , so we have canonical holomorphically locally trivial fibrations
πi : Kµ → Kµi . It is not hard to see that ωµ = π1∗ ωµ1 + π2∗ ωµ2 . If µ, µ1 and µ2 are
dominant, the holomorphic line bundle Oµ is isomorphic to π1∗ Oµ1 ⊗ π2∗ Oµ2 . Let
me summarize this in a commutative diagram:
π1∗ Oµ1 ⊕ π2∗ Oµ2
NNN
p ppp NNN
ppp ⊗ NNN
p NNN
x pp
p  '
Oµ1 Oµ Oµ2

 π1  π2 
Kµ1 o Kµ / Kµ2 .

2.5. Gradient flows, semistability. Let X be a smooth connected Riemannian

manifold and let f be a function on X with the property that at every point
of X there exists a system of local coordinates in which f is real-analytic. Let
F(t, ·) be the gradient flow of −f . Assume that the path of steepest descent F(t, x)
through every point x is contained in a compact set. Then the flow is defined
for all t ≥ 0. Moreover, by results of Lojasiewicz [21] and Simon [29], the limit
x∞ = limt→∞ F(t, x) exists for all x. Let a be a critical level of f and let Sa = {x ∈
X : f (x∞ ) = a} be the stable set of a. Then Sa is a locally closed subset of X and
that x`7→ x∞ is a continuous retraction from Sa onto f −1 (a). The decomposition
X = a Sa is called the Morse decomposition of X with respect to f (even if f is
not a Morse function).
(A note on the literature: Lojasiewicz’ paper [21] does not contain a complete
proof of these assertions. He explains part of the requisite estimates in [22]. Simon
gives a fuller account of the retraction argument in [29], while also generalizing it to
an infinite-dimensional situation. Apparently independently of both [21] and [29],
Neeman [26] and Schwarz [28] rederive the retraction theorem for certain flows on
vector spaces from the inequalities in [22]. Their arguments can easily be generalized
to prove the above statements.)
As an example, let (X, ω, τ, Φ) be a connected Hamiltonian K-manifold equipped
with a compatible almost-complex structure J. Let (·, ·) be a K-invariant inner
product on k and let |·| be the associated norm. I use the same symbols to denote
the corresponding inner product and norm on the dual k∗ . Put f = |Φ|2 . It follows
from the local model for Hamiltonian actions (see Section 6) that this function is
real-analytic in suitable local coordinates. Note that since f is K-invariant, the
flow F(t, ·) is K-equivariant. Let us assume the momentum map to be admissible
6 REYER SJAMAAR

in the sense that for every x ∈ X the path of steepest descent F(t, x) is contained in
a compact set. The set S0 is called the set of (analytically) semistable points and
is denoted by X ss (Φ). So X ss (Φ) is nonempty if and only if 0 ∈ Φ(X). Kirwan has
shown in [13] that Sa is a submanifold of even codimension for every critical level
a (and if J is integrable, then
` Sa is a complex submanifold). Now assume that the
Morse decomposition X = a Sa is locally finite. (This is for instance the case if
for every a there are only finitely many critical levels below a). Then, if nonempty,
X ss (Φ) is open, connected and dense.

3. Semistability and convexity

In this section I review the convexity theorem for Kähler manifolds due to
Guillemin and Sternberg [5] and Mumford [27]. Guillemin and Sternberg have
pointed out in [6] that semistability and convexity are closely related. This idea
goes back to Heckman’s paper [8], and can be formulated as follows.
Proposition 3.1. Let (Yi , σi , Ψi ) be compact Hamiltonian K-manifolds with com-
patible (integrable) complex structures Ji , where i = 1, 2, . . . , k. Assume that the
cohomology classes of the σi are integral. Let Y be a compact complex K-manifold
and let pi : Y → Yi be K-equivariant surjective holomorphic Pmaps. Let a1 , a2 , . . . ,
ak be nonnegative numbers, and let σ = i ai p
∗i σi and Ψ = i ai p∗i Ψi . Assume σ
P

is a Kähler form on Y . Then i p−1 Y ss (Ψi ) is contained in Y ss (Ψ). Hence, if

T
i
0 ∈ Ψi (Yi ) for every i, then 0 ∈ Ψ(Y ).
By the equivariant version of Kodaira’s Embedding Theorem, the manifolds Yi
and Y are of course biholomorphically equivalent (but not necessarily isometric) to
projective manifolds with linear G-actions. The proof is a straightforward adapta-
tion of the techniques of [5] and [27]. With a view to later applications I supply an
argument which can easily be made to work for noncompact manifolds.

Proof. Note that Ψ is a momentum map for the K-action on Y with respect to the
symplectic form σ. Also, since Y is compact, Ψ is admissible in the sense of Section
2.5. For clarity I will first handle the case where Y Ti = Y and pi = idY . So we are
given Kähler forms σi on Y and we wish to show i Y ss (Ψi ) ⊂ Y ss (Ψ).
For i = 1, 2, . . . , √
k, let Li be a Hermitian holomorphic line bundle on Y with
curvature form −2π −1 σi . (These exist because Y is compact.) Let ni be a
positive integer and let si ∈ Γ(Y, Lni i )K , where Γ stands for holomorphic sections.
Let hsi , si i denote the length squared of si with respect to the fibre metric on Lni i .
It follows from the invariance of the si that for every ξ ∈ k we have LJξY hsi , si i =
−4πni Ψξi hsi , si i, and hence
LJξY hsi , si iai /ni = −4πai Ψξi hsi , si iai /ni . (3.1)
N
Here L stands for the Lie derivative. Let L be the line bundle i Li with the
product Hermitian metric and let F(t, ·) be the flow of the vector field − grad |Ψ|2 .
From the elementary fact that
JξY = grad Ψξ (3.2)
one easily deduces that
grad |Ψ(y)|2 = 2JΨ(y)♭Y,y , (3.3)
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 7

where ♭ : k∗ → k is the linear isomorphism defined by the inner product, and Ψ(y)♭Y
n/n n/n
is the vector field on Y induced by Ψ(y)♭ . (See [13].) Put s = s1 1 ⊗ s2 2 ⊗
n/nk n K
· · · ⊗ sk , where n = n1 n2 · · · nk . Then s ∈ Γ(Y, L ) . Consider the function
u = hs1 , s1 ia1 /n1 hs2 , s2 ia2 /n2 · · · hsk , sk iak /nk . By (3.1),
LJξY u = −4π(a1 Ψξ1 + a2 Ψξ2 + · · · + ak Ψξk )u = −4πΨξ u.
Using this and (3.3) we find that the derivative of u along a trajectory F(t, y) is
equal to
d 
♭ 
u F(t, y) = −du grad |Ψ(F(t, y))|2 = −2du JΨ F(t, y) Y,F(t,y)



dt




2

= 8π Ψ F(t, y) , Ψ F(t, y) u F(t, y) = 8π Ψ F(t, y) u F(t, y) ≥ 0. (3.4)
Now suppose y ∈ Y is semistable with respect to all σi . Kirwan [13] and Ness
[27] observed that for a projective manifold with the Fubini-Study metric analytic
semistability is equivalent to semistability in Mumford’s sense. This is true in gen-
eral for a compact complex manifold with an integral Kähler metric; see [30]. This
means we can find positive integers ni and invariant global holomorphic sections si
n/n n/n
of Lni i such that si (y) 6= 0. Then s(y) = s1 1
(y) ⊗ · · · ⊗ sk k (y) 6= 0, so u(y) > 0.
Put y∞ = limt→∞ F(t, y). By (3.4), u F(t, y)
is increasing along the path F(t, y),
so u(y∞ ) > 0. On the other hand, du F(t, y) dt tends to zero as t tends to infin-
ity, so from (3.4) we get |Ψ(y∞ )|2 u(y∞ ) = 0, T and therefore |Ψ(y∞ )|2 = 0. In other
words, y is semistable for σ. We have shown i Y ss (Ψi ) ⊂ Y ss (Ψ).
Suppose now that 0 ∈ Ψi (Yi ) for all i. Then the sets Y ss (Ψi ) are nonempty
for all i, and are T therefore open and dense. It follows that their intersection is
nonempty. If y ∈ i Y ss (Ψi ), then, by the first part of the proof, Ψ(y∞ ) = 0, that
is, 0 ∈ Ψ(Y ).
In the general case the argument is almost N exactly the same. The difference
∗
is that one considers the line bundle L = p
i i iL on Y and the section s =
∗ n/n1 ∗ n/nk n
p 1 s1 ⊗ · · · ⊗ p k sk of L . Further, the assumptions on the pi guarantee that
the pre-image p−1 i (S) of a complex-analytic subset S ⊂ Yi of positive codimension
is a complex-analytic T subset of positive
codimension of Y . This implies that if
Yiss 6= ∅ for all i, then i p−1 i Y ss
(Ψ i ) 6= ∅.

For noncompact Y the flow of − grad |Ψ|2 may not be defined for all time or
its trajectories may fail to converge, and the equivalence between analytic and
algebraic semistability can break down. But the following qualified statement is
still true. The proof is almost word for word the same.
Proposition 3.2. Let (Yi , σi , Ψi ) be Hamiltonian K-manifolds endowed with com-
patible complex structures Ji , where i = 1, 2, . . . , k. Assume there exist K-
equivariant
√ Hermitian holomorphic line bundles Li on Yi with curvature forms
−2π −1 σi for all i. Let Y be a complex K-manifold and let pi : Y → Yi be K-
equivariant surjectiveP holomorphic maps. P Let a1 , a2 , . . . , ak be nonnegative num-
bers, and let σ = i ai p∗i σi and Ψ = i ai p∗i Ψi . Assume σ is a Kähler form on
Y and that the momentum map Ψ is admissible in the sense of Section 2.5. If
for every i there exist a positive integer ni and a nonzero K-invariant holomorphic
section of Lni i , then 0 ∈ Ψ(Y ).
8 REYER SJAMAAR

Remark 3.3. Suppose that Z ⊂ Y and Zi ⊂ Yi are irreducible K-stable locally

closed analytic subvarieties, and that for all i the restriction of the map pi to Z is
a surjective map Z → Zi . Assume that for every z in Z the path F(t, z) and its
limit z∞ are contained in Z. Also assume that the sections si restrict to nonzero
sections on Zi . Then 0 ∈ Ψi (Zi ) for all i implies 0 ∈ Ψ(Z). Exactly the same proof
works.
Here is an application of Proposition 3.2, where the notation is as in Section
2.4. Let (X, ω, Φ) be a Hamiltonian K-manifold, not necessarily compact, with a
compatible complex structure J. Suppose that there exists a K-equivariant
√ Her-
mitian holomorphic line bundle L on X with curvature form −2π −1 ω. For i = 1,
2, . . . , k, let µi be a dominant weight and let Yi be the manifold X × Kµ∗i with
symplectic form σi = ω + ωµ∗i . (Recall that ∗ is defined by ν ∗ = −w0 ν.) Then the
K-action on (Yi , σi ) is Hamiltonian with momentum map Ψi = Φ + ιµ∗i . Let Li be
the Hermitian line bundle L ⊗ Oµ∗i on Yi . Let ai be arbitrary positive numbers,
let µ = i ai µi , and let Y be the K-manifold X × Kµ∗ . Consider the fibrations
P
q
pi : Y → Yi induced by the fibrations of coadjoint orbits Kµ∗ → K(ai µi ) − → Kµ∗i ,
where q is the equivariant P diffeomorphism sending ai µi to µi . Since the ai are
positive, the form σ = i ai p∗i σi is a Kähler form on Y , and the action on Y is
Hamiltonian with momentum map Ψ = i ai p∗i Ψi . Let us assume that
P

for all i there exist ni > 0 and nonvanishing sections si ∈ Γ(Yi , Lni i )K ; (3.5)
for all ai ≥ 0 the momentum map Ψ is admissible. (3.6)
(Assumption 3.6 holds e.g. when Φ is proper.) Then by Proposition 3.2 there exists
a point (x, kµ∗ ) in Y = X × Kµ∗ with Ψ(x, kµ∗ ) = 0, that is, (a1 + · · · + ak )Φ(x) =
kw0 (a1 µ1 + · · · + ak µk ). This shows that
a1 µ1 + a2 µ2 + · · · + ak µk
∈ Φ(X) ∩ t∗+ = ∆(X)
a1 + a2 + · · · + ak
for all ai > 0. The same is obviously true if some of the ai are 0 (just replace the
λi by the subset consisting of those µi for which ai 6= 0), so we have proved:
Proposition 3.4. Under the assumptions (3.5) and (3.6) the convex hull of µ1 ,
µ2 , . . . , µk is contained in ∆(X).

Remark 3.5. By Remark 3.3, a similar result holds for an irreducible K-stable
locally closed analytic subvariety Z of X. Namely, assume that the gradient flow
of the function −|Ψ|2 preserves the subvariety Z × Kµ∗ ⊂ X × Kµ∗ and that the
forward trajectories converge to points in Z × Kµ∗ . (This assumption is satisfied
e.g. if the K-action on X extends to a holomorphic G-action and if Z has the
property that Gz ⊂ Z whenever z ∈ Z.) Assume further that for all i the sections
si restrict to nonzero sections on Z × Kµi . Then hull{µ1 , µ2 , . . . , µk } ⊂ ∆(Z).
Now consider the graded G-algebra A = n≥0 Γ(X, Ln ).
L

Definition 3.6 (Brion). The highest-weight set of X is the subset C(X) of the Q-
vector space Λ∗ ⊗ Q consisting of all λ/n with the property that λ is a dominant
weight and n a positive integer such that the irreducible representation Rλ occurs
in the degree-n piece An . In other words, λ/n ∈ C(X) if and only if n > 0 and λ
occurs as a weight of the T -action on the degree-n piece of the ring AN .
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 9

Let me recall a few basic facts concerning highest-weight sets. See [3] for details.
Note first that C(X) is contained in t∗+ . The fact that A has no zero divisors
implies that C(X) is a convex subset of the Q-vector space Λ∗ ⊗ Q. Now suppose
X is compact. Then A is of finite type. A result of Luna and Vust says that for
every G-algebra A of finite type (not necessarily graded)
AN ∼= (R ⊗ A)G , (3.7)
where R is as in (2.1). This implies that AN is of finite type. (See e.g. [18].)
It follows from this that C(X) is the convex hull of a finite number of points in
Λ∗ ⊗ Q. Moreover, ∆(X) is closed. It is now easy to deduce the following result
from Proposition 3.4.
Theorem 3.7 ([5],[27],[3]). If X is a compact integral Hamiltonian K-manifold
with a compatible Kähler structure, ∆(X) is equal to the closure of C(X) in t∗+ and
is therefore a rational convex polytope.
By Remark 3.5, Theorem 3.7 also holds if we replace X with a G-stable irre-
ducible closed analytic subvariety.
This result applies in particular to a G-stable irreducible closed subvariety X
of PV , the projective space of a G-module V , equipped with the Fubini-Study
symplectic form. But now consider a subvariety X of PV that is not necessarily
irreducible or even reduced. What is the connection between ∆(X) and C(X)? Let
X red denote the reduced variety associated to X and let X1 , X2 , . . . , Xl be its
irreducible components (each endowed with the reduced induced structure). Let Z
red
and Zi be the affineT cones of X , resp. Xi , and let I, resp. Ii , be their homogeneous
ideals. Then I = i Ii and for each i we have an exact sequence
Ii /I ֒ / C[Z] / / C[Zi ] .
red red
S
S follows easily from this that C(X ) = i C(Xi ). It is evident that ∆(X ) =
It
i ∆(Xi ), so applying Theorem 3.7 to each of the Xi we obtain the following result.

Addendum 3.8 ([19]). If X is a (not necessarily irreducible or reduced) subvari-

ety of PV , then ∆(X red ) is equal to the closure of C(X red) in t∗+ and is therefore a
union of finitely many rational convex polytopes.
The following example shows that C(X) is in general not a union of finitely many
Q-convex sets and that it is not necessarily equal to C(X red ).
Example 3.9. Fix a positive integer n. Let A be the algebra C[x, y]/(xn ) and let
X = Proj A, considered as a closed subscheme of P1 . For any weight λ of K = S 1
define an action of K on C2 by g(x, √ y) = (ζ−λ
(g)x, y), where ζ−λ is the character
defined by ζ−λ (exp ξ) = exp −2π −1hλ, ξi . According to (2.6), ∆(P1 ) is equal
to the interval between 0 and λ. Note that X red is the point with homogeneous
coordinates [0, 1], so
C(X red ) = ∆(X red ) = {0}.
Write f¯ for f + (xn ) ∈ A; then x̄k ȳ l has degree k + l and weight kλ for k = 0, 1, . . . ,
n − 1 and l ≥ 0, so that
 
kλ
C(X) = : k, l ∈ N, k < n, kl 6= 0 ,
k+l
which cannot be written as a union of finitely many Q-convex sets.
10 REYER SJAMAAR

Nevertheless, it is always true that the intersection of C(X) with a rational line
in t∗ is a bounded set which contains both its endpoints.
Addendum 3.10. For every (not necessarily irreducible or reduced) subvariety X
of PV and every ν in C(X), let Iν = { q ∈ Q : qν ∈ C(X) }. Then inf Iν and sup Iν
are in Iν .
Proof. This is similar to the proof of Theorem 3.7. The assertion is trivial for ν = 0,
so let me assume that ν 6= 0. Let t1 be the subalgebra of t annihilated by ν; then
T1 = exp t1 is a subtorus of T of codimension one. Denote the quotient circle T /T1
by T2 and identify t∗2 , the kernel of the canonical projection t∗ → t∗1 , with the line
Rν ⊂ t∗ . Put Cν = C(X) ∩ Rν.
Let Z be the affine cone on X and let A be the graded algebra C[Z]N , which as
noted above is of finite type. Since the maximal torus T normalizes N , it acts in
a natural way on A. The algebra B = AT1 is likewise of finite type and it carries
a representation of T2 . Let λ0 be the (unique) primitive element of Λ∗ such that
λ0 = pν for some positive rational p. Note that f ∈ A is a weight vector of weight
proportional to λ0 if and only if f is in B and is a weight vector for T2 . It follows
that Cν is equal to the set of mλ0 /n such that there exists f ∈ B of weight mλ0
and degree n. Choose (nonzero) generators f1 , f2 , . . . , fl of B with weights m1 λ0 ,
m2 λ0 , . . . , ml λ0 and degrees n1 , n2 , . . . , nl . Then mj λ0 /nj ∈ Cν , so mj /nj ∈ Iν
for 1 ≤ j ≤ l. Q Moreover, qν ∈ Cν if and only if there exist a1 , a2 , . . . , al such that
a
the monomial j fj j is nonzero and
P
j aj mj
qν = P λ0 .
j aj n j

In other words, every element of Cν is a convex combination of the mj λ0 /nj (but

not every such combination need be in Cν , because B may have zero divisors). Now
let r be the minimum of the mj /nj and s their maximum. Then r and s are in Iν
and r = inf Iν and s = sup Iν .

4. Affine varieties
In this section X denotes an affine algebraic variety (not necessarily reduced or
irreducible) on which G acts algebraically. The main results are theorems describing
the momentum map image of X with respect to suitable K-invariant symplectic
forms, Theorems 4.9 and 4.23. My main interest is in smooth varieties, but the
proofs turn out to be no harder for general varieties. There are some examples at
the end of Section 4.1.
4.1. Highest weights and the momentum cone. The natural analogue of Def-
inition 3.6 is the following.
Definition 4.1 (Brion). The set of highest weights of X is the subset C(X) of Λ∗+
consisting of all dominant weights λ such that the irreducible G-representation Rλ
occurs in the coordinate ring C[X]. In other words, λ ∈ C(X) if and only if λ is a
weight of the T -action on the ring C[X]N . If G is a torus, we refer to C(X) as the
weight set of X.
Note that if Y is another affine G-variety, then C(X) = C(Y ) if and only if the
coordinate rings of X and Y contain the same irreducible G-representations (up to
multiplicities). It is easy to see that if X is irreducible, then C(X) is a submonoid
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 11

of Λ∗+ (that is, it contains 0 and is invariant under addition). It follows from (3.7)
that C(X) is finitely generated as a monoid.
Let me first discuss a few simple tricks for computing highest-weight sets. Let
X//G = Spec C[X]G denote the categorical quotient of X, that is, the variety of
closed G-orbits in X, and let π : X → X//G be the quotient mapping. (See [18]
or [24].) A subset U of X is called saturated (with respect to π) if π −1 π(U ) = U ,
that is, Gx ⊂ U whenever x ∈ U . Saturated subsets are evidently G-stable. The
following result says that C(X) is determined locally in the Zariski topology, or,
more precisely, that it does not change when we remove from X a divisor defined
by an invariant polynomial.
Lemma 4.2. Assume X is irreducible. Let Y be any saturated affine Zariski-open
subvariety of X. Then C(X) = C(Y ).
Proof. The coordinate ring of X embeds equivariantly into the coordinate ring of
Y . This implies C(X) ⊂ C(Y ). For the reverse inclusion, note that the assumption
that Y is saturated implies that the quotient Y //G is an affine open subvariety
of X//G. Let D be the complement of Y //G in X//G, let D1 , D2 , . . . , Dk be the
irreducible components of D, and let fi ∈ C[X//G] be the polynomial defining Di
for i = 1, 2, . . . , k. Put f = f1 f2 · · · fk ; then the divisor X − Y = π −1 (D) is the
zero set of f (viewed as an element of C[X]), and the coordinate ring of Y is just
the localization of C[X] at f . Define for every p the linear map ψp : C[X]p → C[Y ]
by ψp (a) = a/f p . Since X is irreducible, ψp is injective. Since f is invariant, ψp
is equivariant. The direct sum of the maps ψp is an equivariant map from C[X] to
C[Y ], which is surjective, because C[Y ] is the localization of C[X] at f . It follows
from this that if an irreducible G-representation occurs in C[Y ], then it occurs in
C[X]. This proves C(Y ) ⊂ C(X).
The problem of computing highest-weight sets can in principle be reduced to
torus actions. The variety Spec C[X]N is a categorical quotient of X by N in the
category of affine varieties, and will be denoted by X//N . (It is not always the
same as the set-theoretical quotient of X by N and the natural map X → X//N is
not always surjective. For instance, the homogeneous space G/N is a quasi-affine
variety and G//N is its affine closure.) Let C(X//N ) be the weight set of the H-
action on X//N . By the theorem of the highest weight, a weight occurs in C[X]N
if and only if it occurs as the highest weight of an irreducible component of C[X].
This proves the following lemma.
Lemma 4.3. C(X) = C(X//N ).
Furthermore, highest-weight sets are invariant up to denominators under finite
morphisms.
Lemma 4.4. Let X and Y be affine G-varieties and let φ : X → Y be a finite
surjective G-morphism. Then C(Y ) is contained in C(X) and n! C(X) is contained
in C(Y ), where n is the cardinality of the generic fibre of φ.
Proof. Let A and B be the coordinate rings of X, resp. Y . Then B can be regarded
as a subring of A via the pull-back map φ∗ . This implies C(Y ) ⊂ C(X). Now for the
second inclusion. By Lemma 4.3, it suffices to show that n! C(X//N ) ⊂ C(Y //N ).
Recall that the finiteness of φ means that A is a B-module of rank n. By Satz 1 on
p. 192 of [18], AN is a BN -module of rank ≤ n. This implies that every element a
12 REYER SJAMAAR

of AN satisfies an equation P (a) = 0, where P (t) is a monic polynomial of degree

n in BN [t]. Let a ∈ AN be an element of weight λ ∈ Λ∗ . I will show that BN
contains an element of weight kλ for some k ≤ n. There exist b0 , b1 , . . . , bn−1 in
BN such that an + bn−1 an−1 + · · · + b1 a + b0 = 0. Let k be the largest number l
such that bn−l 6= 0. Then
ak + bn−1 ak−1 + · · · + bn−k+1 a + bn−k = 0.
Because the action of H on BN is completely reducible, we may assume all terms
in this equation have the same weight. Then the weight of bn−k is equal to the
weight of ak , which is kλ.

Example 4.5. Let Γ be a finite group acting on X and suppose that the actions of
G and Γ commute. Let Y be the affine G-variety X/Γ. Then the lemma shows
that C(Y ) ⊂ C(X) and n! C(X) ⊂ C(Y ), where n is the cardinality of Γ.

Remark 4.6. If φ : X → Y is any surjective G-morphism of affine G-varieties, then

C(Y ) is a subset of C(X).

Remark 4.7. Suppose that G is the direct product of two reductive subgroups G1
and G2 . Then the monoid of dominant weights of G is simply the product of the
monoids of dominant weights of G1 and G2 , and the positive Weyl chamber of
G is the product of the positive Weyl chambers of G1 and G2 . Moreover, every
irreducible representation of G is a tensor product of an irreducible representation
of G1 and an irreducible representation of G2 . It follows that for every affine G1 -
variety X1 and for every affine G2 -variety X2 , C(X1 × X2 ) is the product of C(X1 )
and C(X2 ).
Now take any equivariant algebraic closed embedding of X into some representa-
tion space V . Such embeddings always exist; see e.g. [18]. Let h·, ·i be a K-invariant
Hermitian inner product on V and let ωV be the symplectic form − Imh·, ·i. De-
note by k·k the corresponding norm, and by ΦV the momentum map given by (2.2).
Now attach a copy of the one-dimensional trivial representation C to V . Then the
projective space P(V ⊕ C) carries a natural G-action, and the projective space PV
can be identified equivariantly with the hyperplane at infinity in P(V ⊕ C). By
(2.5), the momentum map on P(V ⊕ C) is given by
√
ΦV (v) −1 hξv, vi
ΦP(V ⊕C) ([v, 1]) = 2
= , (4.1)
π(1 + kvk ) 2π 1 + kvk2
where [v, 1] denotes the line through (v, 1). Denote by X̄ the closure of X in
P(V ⊕ C). Let X∞ be the divisor at infinity X̄ ∩ PV in X̄. The highest-weight sets
of X, X̄ and X∞ are closely related. If X is irreducible, then C(X) is a submonoid
of Λ∗+ , so we have the equalities
coneQ C(X) = hullQ C(X) = Q≥0 · C(X).
(See Section 2.2 for the notation.) Also, Q≥0 · C(X̄) = coneQ C(X̄), because C(X̄)
red
is Q-convex. As before, let X∞ denote the reduced variety associated to X∞ . By
red
Addendum 3.8, ∆(X∞ ) is a union of convex polytopes, one for each irreducible
red
component of X∞ .
Theorem 4.8. Let X be an irreducible affine G-variety.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 13

1. The highest-weight set of X̄ is the Q-convex hull of the
highest-weight set of

red
X∞ and the origin in t∗+ : C(X̄) = hullQ C(X∞ red
) ∪ {0} ;
red
2. the momentum polytope of X̄ is the convex hull of
the momentum set of X∞
and the origin in t∗+ : ∆(X̄) = hull ∆(X∞ red
) ∪ {0} ;
3. the highest-weight sets of X and X̄ span the same cone: coneQ C(X) =
coneQ C(X̄).
Proof. 1. First I show that


C(X̄) = hullQ C(X∞ ) ∪ {0} . (4.2)
Let Z be the affine cone on X̄ and let Y be the affine cone on X∞ . Let z : V ×C → C
be the projection onto the second factor and let f be the restriction of z to Z. Then
f can be regarded as an element of C[Z]1 and as such it is an invariant, because G
acts trivially on the second factor. The coordinate ring of Y is equal to C[Z]/(f ).
From the G-equivariant exact sequence
(f ) ֒ / C[Z] / / C[Y ] (4.3)
it is clear that every irreducible representation occurring in C[Y ]n also occurs in
C[Z]n . Therefore, C(X∞ ) ⊂ C(X̄). Furthermore, C[Z]1 contains the copy C[Z]0 f

representation, and so 0 ∈ C(X̄). Consequently
of the one-dimensional trivial
C(X̄) ⊃ hullQ C(X∞ ) ∪ {0} . Conversely, from (4.3) we see that if Rλ∗ occurs in
C[Z]n , then it occurs in either C[Z]n−1 or C[Y ]n . In other words, if λ/n ∈ C(X̄),
then λ/(n − 1) ∈ C(X̄) or λ/n ∈ C(X∞ ). This implies that every element of C(X̄)
lies on the segment
joining an element of C(X∞ ) to the origin. Thus, C(X̄) ⊂
hullQ (C X∞ ) ∪ {0} . This proves (4.2).
To finish the proof of 1 it is enough to show that
red


C(X∞ ) ⊂ hullQ C(X∞ ) ∪ {0} .
I do this by showing that for every ν ∈ C(X∞ ) there exists a rational q ≥ 1 such
red
that qν ∈ C(X∞ ). Let q be the largest rational number such that µ = qν ∈ C(X∞ ).
Such a q exists by Addendum 3.10 and is clearly ≥ 1. There exist g ∈ C[Z]N n and
λ ∈ Λ∗+ such that µ = λ/n, g transforms according to the weight λ∗ under the
action of T , and g is not in (f ). I assert that g does not vanish identically on Y .
For if it did, then by the Nullstellensatz there would exist l such that g l ∈ (f ).
Write g l = hf m with h ∈ C[Z] and m as large as possible; then h 6∈ (f ), so h + (f )
is a nonzero element of C[Y ] = C[Z]/(f ). Since f is an invariant of degree one, the
weight of h + (f ) is equal to lλ and its degree is nl − m. Therefore,
nlqν nlµ lλ
= = ∈ C(X∞ ),
nl − m nl − m nl − m
which contradicts the maximality of q. We conclude that g(y) 6= 0 for some y in Y .
This means that g represents a nonzero element of C[Y red ], and hence µ ∈ C(X∞ red
).
2. This follows immediately from 1 and Theorem 3.7.
3. Note first that cone C(X̄) = cone C(Z), because Z is the affine cone on X̄.
Furthermore, the affine G-variety Z − Y is saturated in Z, because Y is defined as
the zero set of the invariant function f . Therefore, C(Z − Y ) = C(Z) by Lemma
4.2. Moreover, the G-equivariant map from V ⊕ C to itself sending (x, t) to (tx, t)
maps X × C× isomorphically onto Z − Y . It follows that C[Z − Y ] is isomorphic to
C[X] ⊗ C[f, f −1 ] as a G-algebra, where G acts trivially on C[f, f −1 ]. This implies
C(Z) = C(X). In sum, we have shown that cone C(X) = cone C(X̄).
14 REYER SJAMAAR

red
The proof shows that ∆(X̄) is in fact equal to the join of ∆(X∞ ) with the origin
∗ red
in t+ , that is, the union of all intervals joining points in ∆(X∞ ) to the origin.
Here is the main result of this section.
Theorem 4.9. For every G-stable irreducible closed affine subvariety X of V the
set ∆(X) is equal to cone C(X). In particular, it is a rational convex polyhedral
cone.
Proof. Because the monoid C(X) is finitely generated, it spans a rational convex
polyhedral cone in t∗ . To prove that ∆(X) = cone C(X) it suffices to prove that
∆(X) ⊂ cone C(X); (4.4)
hull{λ1 , . . . , λk } ⊂ b ∆(X) for all b > 0 and λ1 , . . . , λk in C(X). (4.5)
Proof of (4.4). By 4.1, ∆(X) is a subset of the cone on ∆(X̄). By Theorem 3.7,
∆(X̄) is equal to the closure of C(X̄) in t∗+ , so by 3 of Theorem 4.8, the cone on
∆(X̄) is equal to the convex hull of C(X). Consequently, ∆(X) is a subset of the
convex hull of C(X).
Proof of (4.5). Let b be any positive number, let µ ∈ t∗+ and let Y be the product
V × Kµ with symplectic form σ = bωV + ωµ and momentum map Ψ = bΦV + ιµ .
I assert that
the momentum map Ψ is admissible for all b > 0 and µ ∈ t∗+ . (4.6)
Assuming this for the moment, let us consider the trivial line bundle OV = V × C
2
on V with the Hermitian metric defined √ by the Gaussian h(v) = exp(−πbkvk ).
The curvature form of (OV , h) is −2π −1 bωV . Lift the K-action on V to OV by
letting K act trivially on the fibre C. Then the fibre metric is K-invariant, the
K-invariant (or G-invariant) holomorphic sections of OV are just the G-invariant
holomorphic functions on V , and the associated
P momentum map is bΦV .
Let us apply this to the special case µ = i ai µi , where the ai are nonnegative
numbers and the µi are in C(X). Consider the varieties Yi = V × Kµ∗i , on which
we have the line bundles Li = OV ⊗ Oµ∗i , symplectic forms σi = bωV + Pωµi and
momentum maps Ψi = bΦV + ιµi . By (4.6), the momentum map Ψ = i ai p∗i Ψi
on Y = V × Kµ∗ is admissible for all b > 0. Moreover, by the definition of C(X),
µi ∈ C(X) ⇐⇒ there exists a G-equivariant linear surjection C[X] → Rµi
⇐⇒ there exists a nonzero G-invariant vector in C[X] ⊗ Rµ∗i
⇐⇒ there exists a nonzero G-invariant algebraic section of OX ⊗ Oµ∗i .
Remark 3.5 now implies that for all b > 0 the polytope hull{µ1 , µ2 , . . . , µk } is
contained in bΨ(X).
Proof of (4.6). After rescaling Ψ we may assume that b = 1. Let F(t, ·) be the
flow of − grad |Ψ|2 . We have to show that for every y in Y the trajectory F(t, y)
is contained in a compact subset of Y = V × Kµ. Since Kµ is compact, we need
only show that the projection of F(t, y) onto V is contained in a compact subset of
V . For any function u on Y , let gradV u denote the component of grad u along V .
Then using (3.2) we find for every pair (v, β) in Y

gradV |Ψ(v, β)|2 = gradV |ΦV (v) + β|2

= gradV |ΦV (v)|2 + 2(ΦV (v), β) + |µ|2 = gradV |ΦV (v)|2 + 2JβV,v
♭


.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 15

This implies
♭
hgradV |Ψ(v, β)|2 , vi = 4|ΦV (v)|2 + 2hv, JβV,v
♭
i = 4|ΦV (v)|2 + 2 v, grad ΦβV (v)
2
♭ β
= 4|ΦV (v)|2 + 4ΦβV (v) = 4|ΦV (v)|2 + 4 ΦV (v), β = 4 ΦV (v) + − |µ|2 ,


2
(4.7)
ξ
where I have used the fact that Φ is homogeneous of degree two for all ξ in k,
and that |ΦV |2 is homogeneous of degree four. Now suppose (v, β) is a point where
hgradV |Ψ(v, β)|2 , vi ≤ 0. Then it follows from (4.7) that ΦV (v) is contained in the
ball of radius |µ|/2 about the point −β/2 ∈ k∗ . Therefore, ΦV (v) is contained in
2
the ball of radius |µ| about the point −β. In other words, ΦV (v) + β ≤ |µ|2 , that
is, |Ψ(v, β)|2 ≤ |µ|2 . In short,
hgradV |Ψ(v, β)|2 , vi ≤ 0 =⇒ |Ψ(v, β)|2 ≤ |µ|2 . (4.8)


Now let γ(t) be the projection onto V of the trajectory F t, (v, β) through any point
(v, β) ∈ V ×Kµ. It follows from (4.8) that we have the following two (non-exclusive)

2
possibilities: gradV Ψ F(t, (v, β)) , v > 0 for all t > 0, or |Ψ F(s, (v, β)) |2 ≤

|µ|2 for some s > 0. In the first case, the curve γ(t) is trapped inside
2 the2 ball of
radius kvk about the origin in V . In the second case, |Ψ F(t, (v, β)) | ≤ |µ| for
all
t ≥ s, because |Ψ|2 is decreasing along F t, (v, β) . This implies that |ΦV |2 γ(t) ≤

4|µ|2 for all t ≥ s. Moreover, γ(t) is contained in the G-orbit through v. It now
follows from Lemma 4.10 below that γ(t) is contained in a compact subset of V .
The following lemma implies that for every point v in V the restriction of ΦV to
the affine variety Gv is a proper map.
Lemma 4.10. For every bounded subset D of k∗ and for every bounded subset B
of V the intersection Φ−1
V (D) ∩ GB is a bounded subset of V .

Proof. Let B be any bounded subset of V . Let {g(n)}n≥0 be a sequence of

elements of G, let {v(n)}n≥0 be a sequence of vectors in B, and put f (n) =
2
kg(n)v(n)k
 . Suppose
that limn→∞ f (n) = ∞. We need to show that the se-
quence ΦV g(n)v(n) n≥0 is unbounded. By the Cartan decomposition, G =
√ √

K exp( −1 t)K, so we can write g(n) = k(n) exp −1 ξ(n) h(n), where k(n),
h(n) ∈ K and ξ(n) ∈ t. Choose an orthonormal basis {ei } of V with√respect to
∗
P there exist βi ∈ t such that ξei = −1 βi (ξ)ei
which the T -action is diagonal. Then
for all ξ ∈ t. Write h(n)v(n) = i vi (n)ei and ρ = sup{ kvk : v ∈ B }; then
|vi (n)| ≤ ρfor all i and n,
X
g(n)v = k(n) eβi (ξ(n)) vi (n)ei ,
i
X
2βi (ξ(n))
f (n) = e |vi (n)|2 .
i

Consider the set I consisting of all i such that the sequence exp 2βi (ξ(n)) |vi (n)|2

is unbounded. Then I is nonempty, because limn→∞ f (n) = ∞. After replacing

{g(n)} and {v(n)} by suitable subsequences we may assume that
lim e2βi (ξ(n)) |vi (n)|2 = ∞
n→∞
16 REYER SJAMAAR

for all i ∈ I. Then limn→∞ βi ξ(n) = ∞ for all i ∈ I, because |vi (n)| is bounded.
This implies there exists an η ∈ t with βi (η) > 0 for all i ∈ I. We may assume η
has length 1. Then k(n)−1 η has length 1, so

2 −1
2 √
2
ΦV g(n)v ≥ Φk(n) η g(n)v = Φη exp( −1 ξ(n))h(n)v . (4.9)
A straightforward computation using (2.3) shows that for all η ∈ t
√
1X
Φη exp( −1 ξ(n))h(n)v = βi (η)e2βi (ξ(n)) |vi (n)|2 . (4.10)
2 i
The vector η was chosen in such a way that all unbounded terms in the right-hand

2
side of (4.10) tend to ∞ for n → ∞. It now follows from (4.9) that ΦV g(n)v
tends to ∞ for n → ∞.

Corollary 4.11. The momentum cone of X is the cone over the momentum poly-
tope of the projectivization of X: ∆(X) = cone ∆(X̄) = Q≥0 · ∆(X̄).
Proof. Combine Theorem 4.8.3 with Theorem 4.9.

Corollary 4.12. The momentum cone of X is a closed subset of t∗+ , and it does
not depend on the embedding of X into the unitary K-module V .

Corollary 4.13. If X is normal, all fibres of the momentum map ΦX are con-
nected.
Proof. The function |Ψ|2 = |ΦV + ιµ |2 is real-algebraic on Y = V × Kµ and has
therefore only finitely many critical levels. This implies that the Morse decomposi-
tion of Y with respect to |Ψ|2 is finite. By the proof of Theorem 4.9, the momentum
map Ψ is admissible. It now follows from the results quoted in Section 2.5 that
its zero level, which is KΦ−1
V (−µ), is a deformation retract of an open subset of V
the complement of which is a finite union of complex-analytic subsets of positive
codimension. The same holds with V replaced by X, because the flow of |Ψ|2 leaves
X ×Kµ invariant. Since X is normal, the complement of a finite number of analytic
subsets is always connected. This implies that KΦ−1 −1
X (−µ), and hence ΦX (−µ),
are connected.

Remark 4.14. The zero fibre Φ−1X (0) is connected regardless of whether X is normal.
The reason is that the function |ΦV |2 has only one critical level, namely 0.

Corollary 4.15. Let Y be a saturated Zariski-open subvariety of X. Then ∆(Y ) =

∆(X).
Proof. Evidently, ∆(Y ) is contained in ∆(X). For the reverse inclusion it suffices
to show that hull{λ1 , . . . , λk } ⊂ b ∆(Y ) for all b > 0 and λ1 , . . . , λk in C(X). The
proof of this fact is a straightforward generalization of the proof of (4.5). (Cf.
Remark 3.5.)

Example 4.16. Suppose the affine G-variety X is defined over the real numbers in
the sense that the complex G-algebra C[X] is the complexification of a real K-
algebra of finite type. Then complex conjugation defines an antilinear involution
on the G-module C[X], so whenever an irreducible representation Rλ occurs in
C[X], its contragredient representation Rλ∗ also occurs. It follows from this that
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 17

the monoid C(X) is invariant under the involution ∗ : Λ∗+ → Λ∗+ . Therefore, by
Theorem 4.9, the cone ∆(X) is invariant under the involution ∗ : t∗+ → t∗+ . This
can also be shown directly as follows. We may assume the embedding of X into the
G-module V to be defined over the real numbers (in the sense that both V and the
G-morphism X → V are defined over the reals). Then X is invariant under complex
conjugation on V . From (2.2) one deduces immediately that ΦξV (v̄) = −ΦξV (v).
Hence ΦV (X) = −ΦV (X), and therefore ∆(X)∗ = −w0 ∆(X) = ∆(X).

Example 4.17 (Peter-Weyl). The group G is an affine variety in its own right, and
it acts on itself by left multiplication: Lg h = gh, and by right multiplication:
Rg h = hg −1 . Consider the L × R-action of G × G on G. Let us denote the highest-
weight set of G for this action by C(G, L × R) and the momentum cone (with
respect to any algebraic G × G-equivariant embedding of G into a unitary K × K-
module) by ∆(G, L × R). The monoid of dominant weights of G × G is simply the
product Λ∗+ × Λ∗+ and its positive Weyl chamber is t∗+ × t∗+ . By the Peter-Weyl
Theorem Lthe coordinate ring of G is a direct sum of irreducible G × G-modules:
C[G] = λ∈Λ∗ Rλ ⊗ Rλ∗ . This implies that the highest-weight monoid C(G, L × R)
+
is equal to the subset { (λ, λ∗ ) : λ ∈ Λ∗+ } of Λ∗+ × Λ∗+ . By Theorem 4.9, the
momentum cone ∆(G, L × R) is therefore the “anti-diagonal” { (µ, µ∗ ) : µ ∈ t∗+ }
inside t∗+ × t∗+ . Notice that this set is ∗-invariant as it should be, because G is
defined over the real numbers.

Example 4.18. We use the notation of the previous example. There are three dif-
ferent embeddings of G into G × G: the maps i1 (g) = (g, 1), i2 (g) = (1, g) and
d(g) = (g, g). Pulling back the L × R-action via these three embeddings yields
three actions of G on itself: the actions L (left multiplication), R (right multipli-
cation) and C (conjugation). The momentum maps for these actions are obtained
by composing the L × R-momentum map with the maps i∗1 , i∗2 and d∗ , respec-
tively. Thus we find that ∆(G, L) and ∆(G, R) are equal to the positive Weyl
chamber, t∗+ , and ∆(G, C) is the positive Weyl chamber of the semisimple part of
K: ∆(G, C) = t∗+ ∩ [k, k].

Example 4.19 (Gelfand’s variety G//N ). The action L of G on itself descends to

an action of G on G//N . Every irreducible G-module occurs exactly once in the
coordinate ring R = C[G]N , so, once again, C(G//N ) = Λ∗+ and ∆(G//N ) = t∗+ .
We can embed G//N into affine space and compute the momentum polytope of
its projectivization. First assume G is semisimple
Lr and simply connected. Then the
algebra R is generated by the subspace E = i=1 Rπi , where π1 , π2 , . . . , πr are the
fundamental weights of G and r is the rank of G. Choose a highest-weight vector
vi in each of the Rπi . Consider the left-G-equivariant map from G to E defined by
sending the identity of G to the vector v1 ⊕ v2 ⊕ · · · ⊕ vr . This map is right-N -
equivariant, so it descends to a map from G//N to E, which is by construction an
embedding. Let us identify G//N with its image in E. It is not hard to show that
the subvariety G//N is invariant under the standard C× -action on E, and the divisor
at infinity (G//N )∞ is therefore the quotient (G//N − {0})/C×. In other words,
G//N is the affine cone on (G//N )∞ . It now follows immediately from Theorem
3.7 that the momentum polytope of (G//N )∞ (with respect to any K-invariant
inner product on E) is the r − 1-dimensional simplex spanned by the fundamental
weights. By Theorem 4.8, the momentum polytope of the projective closure of
18 REYER SJAMAAR

G//N is therefore the r-dimensional simplex spanned by the fundamental weights

and the origin in t∗+ .
Now assume G is a torus of dimension k. Then the subgroup N is trivial and
so R = C[G] and G//N = G. Let ζ1 , ζ2 , . . . , ζk be a basis over Z of the weight
lattice Λ∗ , and identify t∗ with Rk by sending this basis to the standard basis in
Rk . This choice of basis gives an identification of G with the product (C× )k =
{ (t1 , t2 , . . . , tk ) : ti ∈ C× }. A closed affine embedding of G is given by sending
(t1 , t2 , . . . , tk ) to (t1 , t−1 −1 −1 2k
1 , t2 , t2 , . . . , tk , tk ) ∈ C . The projective closure of G in
P is a product of k copies of P . The divisor at infinity G∞ contains 2k fixed
2k 1

points for the action of G, whose images under the momentum map are the points
±ζ1 ± ζ2 ± · · · ± ζk . Theorem 4.8 now implies that the momentum polytope of the
projective closure of G is the parallelepiped spanned by these 2k points.
For an arbitrary connected reductive group G, the variety G//N can be embedded
into affine space in a similar way, by choosing a basis of the monoid of highest
weights. One can show that the momentum polytope of the projective closure of
G//N under such an embedding is the product of the simplex spanned by the origin
and the fundamental weights of [k, k], and the parallelepiped spanned by the points
±ζ1 ± ζ2 ± . . . , where the ζi are a basis of the weight lattice of z(k), the centre of k.

Example 4.20 (associated bundles). Let F be a reductive subgroup of G and let Y

be an affine F -variety. Consider the bundle X = G×F Y associated to the principal
fibration F → G → G/F . The action L of G on itself induces a G-action on X.
Also, X is an affine variety with coordinate ring C[X] = (C[G] ⊗ C[Y ])F . Note
0
that if F 0 is the identity component of F , there is a finite map G ×F Y → X, so
0
∆(X) = ∆(G×F Y ) by Lemma 4.4. This means we may assume F to be connected.
The categorical quotient of X by N has coordinate ring C[X]N = (C[G]⊗C[Y ])F ×N ,
which is isomorphic to (R ⊗ C[Y ])F , where R is the ring C[G]N , on which F acts by
left multiplication. By Lemma 4.3 and Theorem 4.9, the momentum cone of X is
therefore the convex cone spanned by the weights of the action of the maximal torus
H on the algebra (R ⊗ C[Y ])F defined by right multiplication on R. In general,
this is hard to calculate explicitly.

Example 4.21 (tori). In the setting of the previous example, let us assume that
G = H is a torus, and let us write F = H1 . As noted above, we may assume
H1 to be connected. Then H1 is the complexification of a subtorus T1 of T . We
let T2 be the quotient T /T1 and identify it with a complement of T1 in T , so that
T ∼= T1 × T2 . Put H2 = (T2 )C . Then H ∼= H1 × H2 and
H1 ∼
X = H × Y = (H1 × H2 ) × Y = H2 × (H1 ×H1 Y ) = H2 × Y.
H1

Therefore, by Remark 4.7,

C(X) ∼
= C(H2 ) × C(Y ) = Λ∗2 × C(Y ),
where Λ∗2 is the weight lattice of H2 . Consequently, ∆(X) ∼ = t∗2 × ∆(Y ). (These
∼
identifications depend on the splitting H = H1 × H2 . An invariant way of stating
these facts is: X is a trivial principal H2 -bundle over Y ; C(X) is equal to the
preimage of C(Y ) under the canonical projection Λ∗ → Λ∗1 ; and ∆(X) is equal to
the preimage of ∆(Y ) under the canonical projection t∗ → t∗1 .) If Y is a vector
space, then by (2.4), ∆(X) ∼ = t∗2 × − cone{ν1 , . . . , νl }, where ν1 , . . . , νl are the
weights of the H1 -action on Y .
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 19

4.2. The momentum cone and étale slices. Remarkably, the momentum cone
∆(X) of an affine G-variety X turns out to be entirely determined by infinitesimal
data at any point on a closed G-orbit. I shall deduce this from Luna’s étale slice
theorem. First I discuss a variation on Lemma 4.4.
Proposition 4.22. Let X and Y be affine G-varieties, let φ : X → Y be a G-
morphism, let x be a point in X, and let y = φ(x). Suppose that φ has finite fibres,
that the image of φ is open in Y , and that the orbits Gx and Gy are closed. Then
C(Y ) is contained in C(X), and C(X) is contained in the cone on C(Y ).
Proof. The orbit Gy and the complement of φ(X) are G-stable Zariski-closed sub-
sets of Y . Because G-invariant polynomial functions separate G-stable Zariski-
closed subsets, there exists an f ∈ C[Y ]G that vanishes outside φ(X) and satisfies
f (y) = 1. Let Y ′ = Yf and X ′ = φ−1 (Y ). Then X ′ and Y ′ are saturated affine open
subsets of X, resp. Y , containing the orbits Gx, resp. Gy, and the restriction of φ
to X ′ is surjective onto Y ′ . Then C(X ′ ) = C(X) and C(Y ′ ) = C(Y ) by Lemma 4.2.
We are therefore reduced to proving that C(Y ′ ) ⊂ C(X ′ ) and C(X ′ ) ⊂ cone C(Y ′ ).
Let C be the integral closure of C[Y ′ ] in C[X ′ ], and let Z = Spec C. By Luna’s
equivariant version of Zariski’s Main Theorem ([24], part I), the natural maps
ι : X ′ → Z and ψ : Z → Y ′ have the following properties: ι is an open immersion,
ψ is a finite morphism, and φ = ψ ◦ ι. Also, ψ is surjective, because φ : X ′ → Y ′
is. Consequently, C(Y ′ ) ⊂ C(Z) and C(Z) ⊂ cone C(Y ′ ) by Lemma 4.4. So if we
can show that C(X ′ ) = C(Z), we are done. Let us identify X ′ with its image ι(X ′ )
in Z. The orbit Gx ⊂ X ′ is closed in Z (cf. [24], p. 94): since Gy is closed in Y ′
and ψ is finite, ψ −1 (Gy) is closed in Z and consists of a finite number of orbits,
one of which is Gx. The conclusion is that Gx and the complement of X ′ in Z are
G-stable closed subsets of Z. This implies the existence of a G-invariant h ∈ C[Z]
that vanishes outside X ′ and satisfies h(x) = 1. Then Xh′ is a G-stable affine open
subset of X ′ , and it is saturated as a subset of both X ′ and Z. Hence, by Lemma
4.2, C(X ′ ) = C(Xh′ ) = C(Z).

Theorem 4.23. Let X be an affine G-variety, let x be a point on a closed G-

orbit, and let Sx be an étale slice at x. Then the momentum cone of X is equal
to the momentum cone of G ×Gx Sx . If x is a smooth point of X, then ∆(X) =
Gx


∆ G × Vx , where Vx is the tangent space to Sx at x.

Proof. By Luna’s Etale Slice Theorem the natural map from the bundle G ×Gx Sx
into X is étale and its image is Zariski-open. Furthermore the G-orbits through
the point [1, x] in G ×Gx Sx and the point x in X
are closed. It now follows
from Proposition 4.22
that C(X) and C G ×Gx Sx span the same cone. Hence
∆(X) = ∆ G ×Gx Sx by Theorem 4.9.
If x is a smooth point of X, we may assume the étale slice Sx to be smooth, and
there exists a Gx -equivariant étale morphism ψ : Sx → Vx with Zariski-open image.
The map ψ extends to a G-equivariant map G ×Gx Sx → G ×Gx Vx , which is
étale
and has Zariski-open image as well. Again by Proposition 4.22, C G ×Gx S
x and
C G ×Gx Sx span the same cone. We conclude that ∆(X) = ∆ G ×Gx Vx .

Corollary 4.24. The cone on C G ×Gx Sx is independent of the point x. Here x

ranges over the set of all points in X through which the G-orbit is closed.
20 REYER SJAMAAR

The following result is a necessary condition for the origin to be an extreme point
of ∆(X). Here [G, G] denotes the commutator subgroup of G. Note that for every
subgroup F of G, [G, G]F is a subgroup of G, because [G, G] is normal. If F is a
closed reductive subgroup, then so is [G, G]F .
Theorem 4.25. Assume that ∆(X) is a proper cone. Then for every point x such
that Gx is closed the following condition holds: G = [G, G]Gx .
Proof. Let x be any point such that Gx is closed. Let Y denote the homogeneous
space G/Gx and Z the homogeneous space G/[G, G]Gx . Consider the maps
ι τ / Z,
Xo Y
where ι is the G-map sending the coset 1Gx to x and τ is the canonical projection.
Clearly, ∆(Y ) is a subset of ∆(X) and, by Remark 4.6 and Theorem 4.9, ∆(Z) is a
subset of ∆(Y ). Therefore, since ∆(X) is a proper cone, so is ∆(Z). On the other
hand, the torus G/[G, G] acts transitively on Z, so ∆(Z) is a vector space. (Cf.
Example 4.21.) We conclude that Z is a point, in other words, G = [G, G]Gx .
Note that if G is semisimple, the condition G = [G, G]Gx is void. This is as it
should be, because in this case the positive Weyl chamber t∗+ is a proper cone, so
every cone contained in it is a proper cone.

5. Stein varieties
In this section I prove a convexity theorem for certain Stein K-varieties, Theorem
5.4. It can be regarded as a local version of Theorem 4.9. The results are far from
optimal, but will be sufficient for our purposes. Let me start with a number of
elementary observations on Kähler potentials and momentum maps.
Lemma 5.1. Suppose Y is a connected complex manifold
√ and ρ a strictly plurisub-
¯ with associated Rie-
harmonic function on Y . Let σ be the Kähler form −1 ∂ ∂ρ
mannian metric h·, ·i, and let ϑ be the Hamiltonian vector field of ρ. Then the
vector field Jϑ = grad ρ is expanding: LJϑ σ = 2σ.
Proof. Let J : T Y → T Y denote the complex structure on Y and also the transpose
operator T ∗ Y → T ∗ Y . For all functions f ,
√ √ √
¯ = 1 (d + −1 Jd)f and dJdf = −2 −1 d∂f
∂f ¯ = −2 −1 ∂ ∂f.
¯ (5.1)
2
Moreover, for all functions f and all tangent vectors η,
σ(grad f, η) = −hgrad f, Jηi = −df (Jη) = −Jdf (η),
and therefore Jdf = −ι√ grad f σ. Together with (5.1) this implies that Lgrad ρ σ =
dιgrad ρ σ = −dJdρ = 2 −1 ∂ ∂ρ¯ = 2σ.

Let me add to this that the vector fields ϑ and Jϑ are usually not holomorphic.
The symplectic form on Y being exact, every symplectic action of K on Y has
a momentum map Ψ. It turns out that the flow of the vector field − grad ρ has the
peculiar property that it pushes forward under Ψ to a flow on k∗ , which retracts
the image of Ψ exponentially to a single point in z(k)∗ , where z(k) is the centre of k.
Proposition 5.2. Let Y and ρ be as in Lemma 5.1. √ Suppose that K acts holo-
morphically on Y , leaving ρ invariant. Put α = − −1 ∂ρ ¯ and Ψξ = ιξ α for all
Y
ξ ∈ k. Let G(t, ·) = Gt (·) denote the flow of −Jϑ = − grad ρ. Then
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 21

1. the functions Ψξ are the components of an equivariant momentum map Ψ for

the K-action on Y with respect to the symplectic form σ;
2. there exists a character c of k such that LJϑ Ψ = 2(Ψ + c) (cf. [10], § 3).
Therefore
G∗t Ψ = e−2t Ψ + (e−2t − 1)c (5.2)
for all t such that Gt is defined. It follows that Gt maps fibres of Ψ to fibres of
Ψ. Moreover, if Gt is defined for all t ≥ 0, then limt→∞ G∗t Ψ is the constant
map −c;
3. if c = 0, then the critical set of ρ is contained in the fibre Ψ−1 (0). The
converse holds if ρ has at least one critical point.
Proof. Note first that σ = −dα. Further, since ρ is K-invariant and K acts holo-
morphically, α is K-invariant. Consequently,
dΨξ = dιξY α = LξY α − ιξY dα = −ιξY dα = ιξY σ,
so ξY is the Hamiltonian vector field of the function Ψξ . An easy calculation shows
that {Ψξ , Ψη } = Ψ[ξ,η] , so Ψ is K-equivariant. This proves 1.
Note that since ϑ is the Hamiltonian vector field of the K-invariant function ρ,
the induced vector field ξY commutes with ϑ for all ξ ∈ k. Being holomorphic, ξY
therefore commutes with Jϑ as well, so by 1
dLJϑ Ψξ = LJϑ dΨξ = LJϑ ιξY σ = (ι[Jϑ,ξY ] + ιξY LJϑ )σ = ιξY LJϑ σ = ιξY 2σ = 2dΨξ .
This implies the function LJϑ Ψξ − 2Ψξ is a constant, say 2c(ξ), for all ξ ∈ k. It is
evidently linear in ξ. From the equivariance of Ψ and the fact that [Jϑ, ξY ] = 0 it
is now easy to deduce that c([ξ, η]) = 0 for all ξ and η in k. This proves the first
assertion in 2. Integrating the equation LJϑ Ψ = 2(Ψ + c) yields (5.2). The last two
assertions are obvious.
Assume c
= 0. Let y be a critical point of ρ. Then 2 implies that 2Ψ(y) =
Lgrad ρ(y) Ψ (y) = (L0 Ψ)(y) = 0, so y ∈ Ψ−1 (0). Conversely, assume the critical
set of ρ is nonempty and is contained in the fibre Ψ−1 (0). Let y be a critical point
of ρ. Then from (5.2) we obtain
0 = Ψ(y) = Ψ G(t, y) = e−2t Ψ(y) + (e−2t − 1)c = (e−2t − 1)c

for t ≥ 0, and so c = 0.
The set-up of this proposition is functorial in the following sense. Let Z be a
K-invariant closed complex submanifold of Y and let ρZ = ρ|Z be the restriction
√
of ρ to Z. Put αZ = − −1 ∂ρ, ¯ σZ = −dα and Ψξ = ιξ αZ for ξ ∈ k. Then
Z Z
αZ = α|Z , σZ = σ|Z and ΨZ = Ψ|Z . Of course, the Hamiltonian vector field ϑZ of
ρZ is not the restriction of ϑ to Z, unless ϑ happens to be tangent to Z.
These observations apply to the pair of manifolds Y = V and Z = X, where V
is a G-representation space with a K-invariant inner product as in Section 4, and
X a G-stable closed nonsingular algebraic√ subvariety of V . We take ρ to be the
function ρ(v) = kvk2 /2. Clearly, σ = −1 ∂ ∂ρ ¯ is the standard symplectic form
ωV , Ψ is the quadratic momentum map ΦV given by (2.3), and Jϑ = grad ρ is the
radial vector field v ∂/∂v on V . The idea to use the length function as a tool in
invariant theory is due to Kempf and Ness [12]. I shall frequently refer to their
main result (see also [28]):
Theorem 5.3. For all v in V the following conditions are equivalent :
22 REYER SJAMAAR

1. the orbit Gv is closed ;

2. the restriction of ρ to Gv has a stationary point ;
3. Gv intersects the zero level set of the momentum map ΦV .
If v is a stationary point of ρ|Gv , then: ρ|Gv takes on its minimum at v; for all
w ∈ Gv, ρ(w) = ρ(v) implies w ∈ Kv; and Gv = (Kv )C .
Note that grad ρ is tangent to the subvariety X only if X is invariant under the
standard C× -action on V . Because X is closed, the restriction of ρ to X, ρX , is a
proper function, so the forward trajectories of −JϑX = − grad ρX are bounded and
the flow GX (t, ·) is defined for all t ≥ 0. Since ρX is real-analytic, limt→∞ GX (t, x)
exists for all x in X. The properness of ρX implies that the the flow retracts the
stable set of every critical level continuously onto the critical set. (See Section 2.5.)
Furthermore, ρX always has critical points, for example minima. If x is a critical
point of ρX , it is a critical point of the restriction of ρX to the orbit Gx, and
therefore ΦX (x) = 0 by Theorem 5.3. Hence, the character c in Proposition 5.2 is
0, so that
LJϑX ΦX = 2ΦX and (GX )∗t ΦX = e−2t ΦX . (5.3)

Theorem 5.4. Let X be a G-stable closed nonsingular algebraic subvariety of V .

Suppose that ρX has a unique critical level. Let U be a basis of neighbourhoods (in
the classical topology on X) of the critical set of ρX . Then the sets ∆(U ), where
U ∈ U, form a basis of neighbourhoods of the vertex 0 of the momentum cone ∆(X).
In particular, the cone spanned by ∆(U ) is equal to ∆(X) for every U ∈ U.
Proof. Let Bε denote the closed ball of radius ε about the origin in V and let
δ = min{ kxk : x ∈ X } be the distance from X to the origin. The assumption on
ρX implies that its only critical level is the global minimum, δ 2 /2. Therefore, the
sets X ∩ Bε , where ε > δ, are a basis of neighbourhoods of the critical set X ∩ Bδ
of ρX . So it suffices to prove that the sets ∆(X ∩ Bε ), where ε > δ, form a basis of
neighbourhoods of the vertex 0 of the momentum cone ∆(X). The proof has three
parts: first I show that for some ε > δ the set ∆(X ∩ Bε ) is a neighbourhood of
the vertex in ∆(X). Then I show that the same is true for every ε > δ. Lastly, I
prove that for every ball D about the origin in k∗ there exists an ε > δ such that
∆(X ∩ Bε ) ⊂ D.
Part 1. The flow GX extends to a deformation retraction
ḠX : X × [0, ∞] −→ X (5.4)
of X onto the set X ∩ Bδ ⊂ Φ−1 X (0). Now take any η > δ. Then for every x in
X the trajectory G
X (t, x) is contained in Bη for sufficiently large t. Moreover, by
(5.3), ΦX GX (t, x) = e−2t ΦX (x). This implies that
the cone spanned by ∆(X ∩ Bη ) is the whole of ∆(X). (5.5)
Now consider the subset S = G · (X ∩ Bη ) of X. I assert that
for every x ∈ S the affine variety Gx is contained in S. (5.6)
Indeed, take any x in S and any y in Gx. We have to show that y is in S. Let
Ft (·) be the gradient flow of the function −|ΦV |2 . The limit map F∞ = limt→∞
is continuous, it retracts S onto Φ−1
X (0) ∩ Bη , and it retracts Gx onto the K-orbit
K(F∞ x) ⊂ Bη . Moreover, ρX is decreasing along the flow lines, so the restriction of
ρX to Gx takes on its minimum at F∞ (x). (See [26] and [28].) This implies F∞ (y)
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 23

is in the K-orbit through F∞ (x). There are two possibilities: either F∞ (x) is in
the interior of the ball Bη , or it is on the boundary. In the first case, the G-orbit
Gy intersects the interior of Bη , so y ∈ G · (X ∩ Bη ) = S. In the second case,
since F∞ (x) is the point closest to the origin on Gx and by assumption the orbit
Gx intersects Bη , we see that F∞ (x) lies on Gx. By Theorem 5.3 every G-orbit
intersecting Φ−1
V (0) is closed, and therefore y ∈ Gx = Gx ⊂ S. This proves 5.6.
Next, I assert that
∆(S) = ∆(X). (5.7)
To see this, let λ be any point in ∆(X). Then bλ ∈ ∆(X ∩ B
η ) for some b > 0
by 5.5. Take x ∈ X ∩ Bη such that ΦX (x) = bλ. Then ∆ Gx contains the ray


through bλ by Theorem 4.9. But Gx ⊂ S by 5.6, so λ ∈ ∆ Gx ⊂ ∆(S). This
proves 5.7.
Now let D be any ball about the origin in k∗ . Then Φ−1 V (D) ∩ S is a bounded
subset of V by Lemma 4.10. This means we can find ε > δ such that the ball
Bε
contains Φ−1V (D) ∩ S = Φ −1
X (D) ∩ S. Then Φ X (X ∩ B ε ) ⊃ Φ X Φ −1
X (D) ∩ S . By
5.7 above, ΦX maps S surjectively onto ΦX (X), and therefore ΦX Φ−1


X (D) ∩ S =
D ∩ ΦX (X). Consequently, ∆(X ∩ Bε ) contains ∆(X) ∩ D and is therefore a
neighbourhood of the vertex in ∆(X).
Part 2. Suppose ∆(X ∩ Bε ) is a neighbourhood of the vertex in ∆(X) for a
certain ε > δ. Then e−2t ∆(X ∩ Bε ) is a neighbourhood of the vertex for all t.
Choose an arbitrary ε′ with δ < ε′ < ε. By the continuity of the retraction (5.4)
and the compactness of X ∩ Bε there exists a t such that

Gt (X ∩ Bε ) is a subset of
X ∩ Bε′ . So by (5.3), e−2t ∆(X ∩ Bε ) = ∆ Gt (X ∩ Bε ) is a subset of ∆(X ∩ Bε′ ),
so ∆(X ∩ Bε′ ) is a neighbourhood of the vertex in ∆(X).
Part 3. Let D be any ball about the origin in k∗ . Take
any ε > δ; then
there exists a t such that e−2t ∆(X ∩ Bε ) = ∆ Gt (X ∩ Bε ) is contained in D.
Again by continuity and compactness, there exists an ε′ with δ < ε′ < ε such that
X ∩ Bε′ ⊂ Gt (X ∩ Bε ). But then ∆(X ∩ Bε′ ) is contained in D.
This proof gives no information on the shape of the set ∆(X ∩Bε ) away from the
vertex. It seems not unlikely that ∆(X ∩ Bε ) is convex. I am tempted to speculate
that it is equal up to a dilation to the momentum polytope of the projective closure
of X, ∆(X̄).
Example 5.5 (homogeneous vector bundles). Let L be a closed subgroup of K and
let F be the reductive subgroup LC of G. Let W be a unitary L-module and let
X = G ×F W . There exists an orthogonal (real) representation V1 of K containing
a vector v0 with stabilizer Kv0 = L. The map k 7→ kv0 therefore induces an
embedding K/L → V1 . The complexification of this map is an embedding of
K C /LC = G/F into the unitary K-module V1C . There also exists an L-equivariant
isometric embedding of W into a unitary K-module V2 . Then the map X →
V1C ⊕ V2 defined by [g, w] 7→ gv0 + gw is a G-equivariant closed embedding of X
into V = V1C ⊕ V2 . (See Lemmas 1.16 and 1.18 of [30] for a proof of these facts.) Let
ρ(v) = kvk2 /2, where k·k denotes the length function with respect to the direct sum
metric on V . Let us identify X with its image in V . Using Theorem 5.3 one can
easily show that ρX has a unique critical level, which is a minimum, and that the
critical set is the compact orbit Kv0 . Hence, by Theorem 5.4, the sets ∆(U ), where
U ranges over the neighbourhoods of Kv0 in X, form a basis of neighbourhoods of
the vertex of ∆(X).
24 REYER SJAMAAR

Here is an example of a singular variety for which the conclusion of Theorem 5.4
holds.
Example 5.6. Let F , W and X be as in the previous example, and let Y be an
F -invariant affine cone in W . Let X ′ be the affine subvariety G ×F Y of the vector
bundle X = G ×F W and embed X into a G-module V as in the previous example.
Because Y ⊂ W is invariant under dilations, the subvariety X ′ is invariant under
the gradient flow of ρX . It follows that the restriction of the flow GX to X ′ retracts
X ′ onto the compact orbit Kv0 . Exactly the same proof as that of Theorem 5.4
now shows that the sets ∆(U ′ ), where U ′ ranges over the neighbourhoods of Kv0
in X ′ , form a basis of neighbourhoods of the vertex of the cone ∆(X ′ ).

6. Hamiltonian actions and convexity

In this section I explain how the previous, mainly algebro-geometric, results
can be generalized to arbitrary Hamiltonian actions. The basic idea is that every
symplectic manifold with a Hamiltonian K-action can locally near every orbit in
the zero fibre of the momentum map be identified with a germ of a complex affine
G-variety. I then state the main result of the paper, Theorem 6.7. First recall the
following standard definition.
Definition 6.1. Let M be a Hamiltonian K-manifold with momentum map Φ. For
every µ ∈ k∗ the (Meyer-Marsden-Weinstein) reduced space or symplectic quotient
at level µ is the space Φ−1 (Kµ)/K. It is denoted by Mµ,K , or by Mµ , if the group
K is clear from the context.
By the results of [31], the symplectic quotient is a stratified space carrying natural
symplectic forms on the strata satisfying certain compatibility conditions. For most
µ in k∗ , Mµ is actually a symplectic V-manifold.
One application of symplectic reduction is the construction of “local models”,
which I now briefly explain. See [25] or [7] for details. Let µ be any vector in k∗
and let L be any closed subgroup of Kµ . Let W be a symplectic representation
of L and let ΦW be the standard quadratic L-momentum map on W . Let zµ be
the centre of kµ . Then the zero-weight space in k under the adjoint action of zµ is
exactly kµ , so kµ has a natural kµ -invariant complement in k. This means that the
principal fibre bundle Kµ → K → Kµ comes equipped with a natural connection.
Furthermore, the Levi decomposition kµ = zµ ⊕ [kµ , kµ ] shows that zµ is a direct
summand of kµ . We can therefore view z∗µ as a subspace of k∗µ and k∗µ as a subspace
of k∗ . Under these natural identifications, µ is an element of z∗µ ⊂ k∗µ .
The manifold K × k∗µ carries a natural closed two-form (the minimal-coupling
form defined by the symplectic form on Kµ and the connection on K → Kµ),
which is nondegenerate in a K-invariant neighbourhood of K × {0}. Now consider
the manifold X = K × k∗µ × T ∗ Kµ × W and identify T ∗ Kµ with Kµ × k∗µ by means
of left-translations. The action of Kµ × L on X defined by
(k, l) · (g, κ, h, ν, w) = (gk −1 , kκ, khl−1, lν, lw)
is Hamiltonian with momentum map Ψ : X → k∗µ × l∗ given by


Ψ(g, κ, h, ν, w) = −κ + hν, −ν|l + ΦW (w) .

Definition 6.2. X(µ, L, W ) = Ψ−1 (−µ, 0)/(Kµ × L) is the symplectic quotient of

X by the Kµ × L-action at the value (−µ, 0) ∈ k∗µ × l∗ .
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 25

It turns out that X(µ, L, W ) is smooth. Consider the K-action on X defined by

left-multiplication on the first factor. It is Hamiltonian as well and, moreover, it
commutes with the Kµ × L-action. It descends therefore to a Hamiltonian K-action
on X(µ, L, W ). The easiest way to write the K-momentum map on X(µ, L, W ) is
as follows. Put m = kµ /l. Choose an L-invariant complement of l in kµ and identify
it with m. Then m∗ is a subspace of k∗µ . The map ϕ from the product K × m∗ × W
to Ψ−1 (−µ, 0) defined by


ϕ(g, ν, w) = g, ν + ΦW (w) + µ, 1, ν + ΦW (w), w
descends to a K-equivariant diffeomorphism
ϕ̄ : K ×L (m∗ × W ) −→ X(µ, L, W ).
Via this diffeomorphism, the associated bundle K ×L (m∗ × W ) acquires a closed
two-form that is symplectic in a neighbourhood of the zero section. Note that the
definition of ϕ̄ depends only on the choice of a complement of l in kµ (which is
equivalent to the choice of a connection on the principal fibre bundle L → K →
K/L). We shall henceforth identify X(µ, L, W ) with K ×L (m∗ × W ) through ϕ̄.
The K-momentum map on K ×L (m∗ × W ) is given by


Φ([g, ν, w]) = g ν + ΦW (w) + µ . (6.1)
It is useful to consider the restriction of Φ to a K-invariant neighbourhood of
the point [1, 0, 0]. If we perform the reduction of X by the Kµ × L-action in stages,
first with respect to L and then with respect to Kµ , we see that X(µ, L, W ) is an
“iterated” associated bundle: it is a bundle
X(µ, L, W ) ∼ = K ×Kµ Kµ ×L (m∗ × W ) = K ×Kµ Y,


(6.2)
over the coadjoint orbit Kµ ∼
= K/Kµ with fibre the Hamiltonian Kµ -space
Y = Kµ ×L (m∗ × W ). (6.3)
The Kµ -momentum map Y → k∗µ is the restriction of Φ to Y . We can now write Φ
as the composition of two maps,
ι
K ×Kµ Y −→ K ×Kµ k∗µ −→ k∗ , (6.4)
the first of which is the unique bundle map extending the Kµ -momentum map on
the fibre Y , and the second of which is defined by ι([g, ν]) = gν.
Recall that k∗µ is a slice at µ for the coadjoint action on k∗ : the restriction of ι
to a sufficiently small K-invariant open neighbourhood of [1, µ] is a K-equivariant
embedding onto an open neighbourhood of µ. This implies that the restriction of
Φ to a sufficiently small K-invariant neighbourhood U of [1, 0, 0] is a bundle map
of associated bundles over Kµ. Consequently, U ∩ Φ−1 (k∗µ ) = U ∩ Y and the image
Φ(U ) is a bundle over Kµ with fibre Φ(U ∩ Y ).
If µ happens to be in t∗+ , then t∗ is a subset of k∗µ , and t∗+ is contained in t∗+,µ ,
the positive Weyl chamber of k∗µ . In fact, t∗+,µ ∩ D = t∗+ ∩ D for a sufficiently small
neighbourhood D of µ in k∗ . It follows from this that if U is small enough
∆(U ) = Φ(U ) ∩ t∗+ = Φ(U ∩ Y ) ∩ t∗+ = Φ(U ∩ Y ) ∩ t∗+,µ = ∆(U ∩ Y ), (6.5)
where ∆(U ∩Y ) stands for the momentum set of U ∩Y considered as a Hamiltonian
Kµ -space.
26 REYER SJAMAAR

Now let M be an arbitrary Hamiltonian K-manifold with momentum map Φ.

Marle [25] and Guillemin and Sternberg [7] have shown that locally at any orbit M
is isomorphic to some X(µ, L, W ).
Theorem 6.3 (symplectic slices). Let m be any point in M . Let L = Km be the
stabilizer of m, let µ = Φ(m), and let
W = Tm (Km)ω Tm (Km) ∩ Tm (Km)ω .


Then there exist a K-invariant neighbourhood U1 of m in M , a K-invariant neigh-

bourhood U2 of [1, 0, 0] in X(µ, L, W ), and a map f : U1 → U2 with the following
properties: f is a K-equivariant symplectomorphism; f intertwines the momentum
maps on U1 and U2 ; and f (m) = [1, 0, 0].
The symplectic vector space W is called the symplectic slice at m. Note that
in the situation of the theorem m is simply the tangent space to Kµ m at m. An
immediate consequence of the theorem is that the germ at m of Φ−1 (k∗µ ), called a
local cross-section of M , is a smooth Kµ -invariant symplectic submanifold, because
in the local model X(µ, L, W ) it is equal to the germ of Y = Kµ ×L (m∗ × W )
at [1, 0, 0]. Furthermore, if m has the property that µ = Φ(m) ∈ t∗+ , then for any
small K-invariant neighbourhood U of m we have an equality of momentum sets
∆(U ) = ∆ U ∩ Φ−1 (k∗µ ) ,


(6.6)
because by (6.5) the same is true in the local model. An analogue of (6.6) for
projective varieties was proved by Brion (Proposition 4.1 in [3]).
Definition 6.4. Let m be any point in M . The local momentum cone at m is the
C
set ∆m = µ+∆(Ym ). Here Ym is the complex affine (Kµ )C -variety (Kµ )C ×(Km ) W ,
with µ = Φ(m) and with W being the symplectic slice at m furnished with a
compatible Km -invariant complex structure.
So ∆m is a convex polyhedral cone with vertex µ and is contained in t∗+,µ , the
positive Weyl chamber of Kµ . Up to a translation by µ it is a rational cone.
From the symplectic slice theorem we deduce the following “local” convexity
theorem.
Theorem 6.5. 1. For every m in M such that µ = Φ(m) is contained in t∗+ and
for every sufficiently small K-invariant neighbourhood U of m the set ∆(U )
is a neighbourhood of the vertex of the local momentum cone ∆m ⊂ t∗+,µ . In
particular, the cone with vertex µ spanned by ∆(U ) is equal to ∆m ;
2. for every µ ∈ k∗ and for every connected component C of the fibre Φ−1 (µ),
the local momentum cone ∆m is independent of the point m ∈ C.
Proof. 1. First we treat the case µ = 0. At points in the zero fibre of Φ there is an
alternative local model for M as follows. Put F = LC = (Km )C and choose an F -
invariant compatible complex structure on the symplectic slice W at m. Consider
the affine G-variety Z = G ×F W . Embed it into a unitary K-module V as in
Example 5.5, and regard it as a Hamiltonian K-space with the symplectic structure
and momentum map inherited from V . The stabilizer of the point [1, 0] ∈ Z under
the K-action is equal to L. The G-orbit through [1, 0] is closed, so by Theorem 5.3,
[1, 0] is contained in Φ−1
Z (0). The symplectic slice at [1, 0] is simply the (Hermitian)
orthogonal complement to the G-orbit through [1, 0] and is therefore equal to W .
Theorem 6.3 now allows us to conclude that the K-invariant germ of M at m is
isomorphic as a Hamiltonian K-manifold to the K-invariant germ of Z at [1, 0].
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 27

Putting this information together with Theorems 4.9 and 5.4 and Example 5.5,
we see that for every small K-invariant neighbourhood U of m the set ∆(U ) =
Φ(U ) ∩ t∗+ is a neighbourhood of the vertex of the cone ∆m = ∆(G ×F W ).
If µ 6= 0, we may without loss of generality assume that µ ∈ t∗+ , because every
K-orbit in k∗ intersects t∗+ . This case can then be reduced to the case µ = 0 by
using (6.6) and shifting the Kµ -momentum map on the space Y in (6.3) by the
vector −µ. This shifted map is still an equivariant Kµ -momentum map, because µ
is in z∗µ .
2. Again, it suffices to prove this for µ = 0. If M is an affine G-variety, then
by Theorem 5.3 the points in Φ−1 (0) are exactly those whose G-orbits are closed.
Corollary 4.24 then says that the local momentum cones ∆m are constant along
the fibre Φ−1 (0). For general M , this argument combined with the symplectic slice
theorem shows that ∆m is locally constant along the fibre Φ−1 (0).

Remark 6.6. It follows from the Cartan decomposition of G that there exists a
global K-equivariant diffeomorphism between the two local models K ×L (m∗ × W )
and G ×F W , but I don’t know if there is a global symplectomorphism.
The following generalization of Kirwan’s convexity theorem [14] is the main result
of this paper. The first part describes ∆(M ) as a locally finite intersection of
polyhedral cones, each of which is determined by local data on M . (The fact that
∆(M ) is a convex set when Φ is proper was also proved in [11].) The second part
is a necessary condition for a point to be a vertex of ∆(M ). It generalizes the
well-known fact that for torus actions vertices arise as images of fixed points. The
third part states that the points where this necessary condition is fulfilled form a
discrete subset of t∗+ .
Theorem 6.7. Assume that the momentum map Φ : M → k∗ is proper.
1. ∆(M ) is the intersection of local momentum cones:
\
∆(M ) = ∆m . (6.7)
m∈Φ−1 (t∗
+)

This intersection is locally finite and therefore ∆(M ) is a closed convex poly-
hedral subset of t∗+ ;
2. if µ is a vertex of ∆(M ) and m is any point in the fibre Φ−1 (µ), then kµ =
[kµ , kµ ] + km , or, equivalently, Kµ = [Kµ , Kµ ]Km . In particular, if µ is a
vertex of ∆(M ) lying in the interior of t∗+ , then T fixes m;
3. let E be the subset of M consisting of all points m such that µ ∈ t∗+ and
kµ = [kµ , kµ ] + km , where µ = Φ(m). The image Φ(E) is a discrete subset of
t∗+ . If M is compact, then ∆(M ) is the convex hull of Φ(E).
Proof. 1. The assumption that Φ is proper implies that its image Φ(M ) is closed
and, by the argument outlined in Section 2.5, that its fibres are connected. Con-
sequently, by Theorem 6.5.2, for every µ ∈ ∆(M ) the cone ∆m is the same for all
points m ∈ Φ−1 (µ). It is now easy to deduce from Theorem 6.5.1 plus the fact that
Φ is proper that for every µ ∈ ∆(M ) there exists an open subset D of t∗ containing
µ such that
∆(M ) ∩ D = ∆m ∩ D, (6.8)
−1
where m is any point in the fibre Φ (µ). This means that ∆(M ) is locally convex.
But every closed locally convex set is convex, so ∆(M ) is convex. Since every
28 REYER SJAMAAR

closed convex set is the intersection of all closed cones containing it, the equality
(6.8) also implies (6.7). Furthermore, applying Theorem 6.5 to the Hamiltonian
K-manifold Φ−1 (D), we find that for every µ′ ∈ D and every m′ ∈ Φ−1 (µ′ ) the
local momentum cone ∆m′ is equal to the cone with vertex µ′ spanned by ∆m .
Since ∆m is a polyhedral cone, it follows from this that as µ′ ranges over D, only
finitely many different cones ∆m′ can occur. In other words, the collection of cones
appearing in the intersection (6.7) is locally finite on t∗+ . This means that ∆(M ) is
a polyhedron.
2. This follows from (6.8) and Theorem 4.25 (applied to the group G = (Kµ )C
C
and the variety X = (Kµ )C ×(Km ) W ).
3. First I prove that Φ(E) is discrete. Since Φ is proper, it suffices to show that
every point m in M possesses a neighbourhood U such that Φ(E ∩ U ) is discrete.
By the symplectic slice theorem, we may therefore assume M is an affine variety.
But it follows from Lemma 6.8 below that for an affine G-variety the set Φ(E)
consists of the origin in t∗+ only.
Finally, the second statement in 3 follows immediately from 2 and the fact that
a compact convex set is the convex hull of its extreme points.

Lemma 6.8. Let V be a unitary K-module and suppose that v ∈ V satisfies the
condition Kµ = [Kµ , Kµ ]Kv , where µ = ΦV (v). Then µ = 0.
√
Proof. We want to prove that µ(ξ) = ΦξV (v) = ( −1/2)hξv, vi = 0 for all ξ ∈ k.
It suffices to show this for all ξ ∈ kµ . The condition on v says that ξ = [χ, ζ] + η
for some χ, ζ ∈ [kµ , kµ ] and η ∈ kv . Now ηv = 0, so µ(η) = 0; and µ(χ, ζ) =
µ(ad χ(ζ)) = (ad∗ χ)µ(ζ) = 0. We conclude that µ(ξ) = 0.

Theorem 6.7 is clearly not optimal. As Theorem 4.9 shows, it is not always
necessary to assume that Φ is proper, nor even that M is nonsingular. See Section
7 for further examples of convexity for noncompact or singular spaces.
On the other hand, the necessary condition 2 for a point to be mapped to a
vertex of ∆(M ) is optimal in the following sense.
Proposition 6.9. For every µ ∈ t∗+ and for every closed subgroup L of Kµ such
that Kµ = [Kµ , Kµ ]L, there exists a Hamiltonian K-manifold (M, ω, K, Φ) with a
point m ∈ M satisfying the following properties: Km = L, Φ(m) = µ, and λ is a
vertex of ∆(M ).
The proof will be given in Section 7.3.
Theorem 6.7.1 implies that for every µ in ∆(M ) the cone with vertex µ spanned
by ∆(M ) is equal to the local momentum cone ∆m , where m is any point in Φ−1 (µ).
In other words, the local shape of ∆(M ) near µ is determined by the representation
of the isotropy subgroup Km on the symplectic slice W at m. Calculating ∆m boils
down to finding generators for the monoid of highest weights of the homogeneous
C
vector bundle (Kµ )C ×(Km ) W . If µ is contained in the boundary of the positive
Weyl chamber, this is usually an arduous task. However, the situation is more
manageable if the fibre Φ−1 (µ) contains a point m that is fixed under Kµ . This
means that Km = Kµ , or, equivalently, that the restriction of Φ to the orbit Km is
a symplectic isomorphism onto the coadjoint orbit Kµ. If this is the case, the vector
space m in the local model (6.2) is 0, so the tangent space at m to the symplectic
cross-section Φ−1 (k∗µ ) is equal to the symplectic slice W at m, and W is simply the
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 29

symplectic orthogonal complement of Tm (Km) inside Tm M . In other words,

ω
Tm Φ−1 (k∗µ ) = W = Tm M/Tm (Km) ∼


= Tm (Km) . (6.9)
It follows that ∆m = ∆(W ), where W is regarded as a Kµ -module. In some
examples this enables one to determine the entire momentum set ∆(M ); see for
instance Section 7.1.

7. Examples
7.1. Actions on projective space. Let V be a finite-dimensional unitary K-
module. If V is irreducible and has highest weight λ, it follows from (2.6) that the
momentum polytope of the projective space PV for the T -action is the convex hull
of the Weyl group orbit through λ∗ . Similarly, if λ1 , λ2 , . . . , λk are the highest
weights of the irreducible submodules of V , then the T -momentum map image of
PV is the convex hull of the union of the W-orbits through λ∗1 , λ∗2 , . . . , λ∗k . This
implies ∆(PV ) is a subset of
t∗+ ∩ hull(Wλ∗1 ∪ Wλ∗2 ∪ · · · ∪ Wλ∗k ).
Arnal and Ludwig [1] and Wildberger [32] determined this subset for “most” V
that are irreducible. I shall calculate ∆(PV ) in some (but not all) of the remaining
cases.
For simplicity I assume K to be semisimple, although the results can easily be
generalized to arbitrary compact groups. Let Ψ be the root system of (k, t) and let
g = α∈Ψ CEα be the root space decomposition of g = kC . The following result
L
can be used to find a “lower bound” for the polytope ∆(PV ).
Lemma 7.1. 1. Let v be any vector in V and let [v] be the ray through v. Then
ΦPV ([v]) ∈ t∗ if and only if hEα v, vi = 0 for all roots α.
2. Let v1 , . . . , vl be weight vectors in V with weights ν1 , . . . , νl . Assume that
hEα vi , vj i = 0 for all roots α and for all i and j with 1 ≤ i < j ≤ l. (This
is for instance the case if for all i and j the difference νi − νj is not a root.)
Then the subspace spanned by v1 , . . . , vl is contained in Φ−1 ∗
V (t ), and the
∗ ∗ ∗
intersection t+ ∩ hull{ν1 , . . . , νl } is contained in ∆(PV ).
√
Proof. 1. Let ξα = Eα − E−α and ηα = −1(Eα + E−α ). By (2.5), ΦPV ([v]) ∈ t∗ if
and only if hξα v, vi = hηα v, vi = 0 for all roots α. Since E−α (viewed as an operator
on V ) is the adjoint of Eα , this is equivalent to hEα v, vi = 0 for all α.
2. Let W denote the linear span of the vi . Let α be any root. The assumption
implies hEα v, vi = 0 whenever v is in W . It now follows from 1 that ΦV (W ) is
contained in t∗ .
From this fact and from (2.5) we infer that
|c1 |2 ν1∗ + · · · + |cl |2 νl∗
ΦPV (g[c1 v1 + · · · + cl vl ]) = ,
|c1 |2 + · · · + |cl |2
where g is any element of the normalizer of T representing w0 ∈ W. Hence ∆(PV ) =
ΦPV (PV ) ∩ t∗+ contains the set t∗+ ∩ hull{ν1∗ , . . . , νl∗ }.
Put µ̌ = 2µ/(µ, µ) for any µ in (tC )∗ . Let α1 , . . . , αr be the simple roots
of k and π1 , . . . , πr the corresponding
P fundamental weights, that is, the basis of
t∗ dual to α̌1 , . . . , α̌r . Then λ = ri=1 (λ, α̌i )πi , where the coefficients (λ, α̌i ) are
nonnegative integers. The next result gives an upper bound for the polytope ∆(PV )
for irreducible V .
30 REYER SJAMAAR

Proposition 7.2. Assume V is irreducible and has highest weight λ. Let Πλ be

the set of weights of V that are not of the form λ − α for any positive root α such
that (λ, α̌) = 1. Then ∆(PV ) is contained in t∗+ ∩ hull Π∗λ .
Proof. Let vλ be a highest-weight vector in V , so that ΦPV ([vλ ]) = −λ. It suffices
to show that the local momentum cone ∆[vλ ] of PV at the point [vλ ] (see Definition
6.4) is contained in the cone with vertex −λ spanned by the set −Πλ . To this end
let us compute the tangent space to the symplectic cross-section at [vλ ] and the
weights of the T -action on it. The vector vλ is an eigenvector for Kλ , so K[vλ ] = Kλ .
In view of (6.9) this implies that we have an isomorphism of Kλ -modules
T[vλ ] Φ−1 ∗ ∼


PV (kλ ) = W = T[vλ ] (PV )/T[vλ ] (K[vλ ]), (7.1)
W being the symplectic slice at [vλ ]. Define Π to be the subset of the weight lattice
Λ∗ consisting of 0 and of all weights occurring in W under the action of the maximal
torus T of Kλ . From (7.1) we get:
∆[vλ ] ⊂ −(λ + cone Π). (7.2)
(If λ is strictly dominant, so that Kλ = T , then this inclusion is an equality.) I
assert that
cone Π = cone(−λ + Πλ ). (7.3)
L
Because of (7.2), this will finish the proof. Let V = ν∈Λ∗ Vν be the weight
space decomposition of V and let C−λ be the one-dimensional representation of Kλ
defined by the character −λ ∈ z∗λ . Then the quotient map V − {0} → PV induces
an isomorphism of Kλ -modules
T[vλ ] (PV ) ∼
M
= Vν ⊗ C−λ . (7.4)
ν∈Λ∗ −{λ}

Since the maximal unipotent subgroup N fixes vλ , the complex stabilizer G[vλ ] is
the parabolic subgroup Pλ = (Kλ )C N . This implies that the real orbit K[vλ ] is
equal to the complex orbit G[vλ ] and therefore we have natural isomorphisms of
complex Kλ -representations
= p◦λ ∼
T[vλ ] (K[vλ ]) ∼
M
= CEα , (7.5)
α∈Ψ−
(λ,α)6=0

where p◦λ denotes the annihilator of pλ in g∗ . (The second isomorphism is induced

by the Killing form on g.) Let Λ∗r ⊂ Λ∗ denote the root lattice of the pair (k, t).
There are two inclusions,
(Λ∗r \Ψ− ) ∩ (−λ + hull Wλ) ⊂ Π ⊂ Λ∗r ∩ (−λ + hull Wλ),
the second of which follows from (7.4) and the first of which follows from (7.1) and
(7.5). Moreover, by the definition of Πλ , the same inclusions hold with Π replaced
by −λ+Πλ . Therefore, to establish (7.3), it suffices to prove the following statement
for every positive root α: if −α ∈ Π, then −α ∈ −λ + Πλ ; and if −α ∈ −λ + Πλ ,
then some multiple of −α is in Π. There are three possibilities:
1. (λ, α̌) = 0. Then λ − α is not a weight of V , so −α is contained in neither Π
nor −λ + Πλ .
2. (λ, α̌) = 1. Then λ − α 6∈ Πλ by definition. Also, λ − α is a weight of V with
multiplicity one, so by (7.5) −α is not a weight of W , so −α 6∈ Π.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 31

λ∗ = π1 + 2π2

π1 π2

Diagram 1. K = SU(3) and λ = 2π1 + π2

3. (λ, α̌) = l > 1. Then the weights λ, λ − α, . . . , λ − lα occur in V . Since

(λ, α̌) 6= 1, all these weights are also in Πλ . The weight −α may or may not occur
in W (depending on whether the multiplicity of λ − α in V is greater than one), but
at any rate −lα is a weight of W . We conclude that −α ∈ −λ + Πλ and −lα ∈ Π.
In sum, we have shown that Π ⊂ −λ + Πλ and −λ + Πλ ⊂ cone Π. This proves
(7.3).
The following result follows easily from Lemma 7.1 and Proposition 7.2.
Proposition 7.3 ([1],[32]). Assume V is irreducible and has highest weight λ.
1. Suppose that (λ, α̌i ) 6= 1 for i = 1, 2, . . . , r. Then ∆(PV ) = t∗+ ∩ hull Wλ∗ ;
2. if (λ, α̌) = 1 for some positive root α, then ∆(PV ) is not equal to t∗+ ∩hull Wλ∗ .
3. ΦPV (PV ) ∩ t∗ is convex if and only if (λ, α̌i ) 6= 1 for i = 1, 2, . . . , r.
If (λ, α̌) = 1 for some positive root α, Proposition 7.3 does not give an explicit
description of the polytope ∆(PV ). Using Lemma 7.1 one can easily check that the
upper bound given by Proposition 7.2 is sharp e.g. in the following cases.
Proposition 7.4. The equality ∆(PV ) = t∗+ ∩ hull Π∗λ holds if
1. K = SU(4) and λ = π1 , π2 , π3 , or π1 + π2 + π3 ;
2. K has rank two and λ ∈ t∗+ is arbitrary.
Diagrams 1–3 illustrate Proposition 7.4. The figures show the convex hull of Wλ∗
and the polytope ∆(PV ) (shaded). The intersection t∗+ ∩ hull Wλ∗ is indicated in
light shading. The dominant weights occurring in V are denoted by black circles.
These are exactly the images of the T -fixed points in PV intersected with t∗+ . Notice
that few of the vertices on the walls of t∗+ arise as images of T -fixed points. In those
cases where they are not weights of V , the fundamental weights of k are indicated
by black squares.
Finally, here is a generalization of Proposition 7.3 to reducible representations.
Proposition 7.5. 1. Let λ1 , λ2 , . . . , λk be the highest weights of the irreducible
submodules of V . Suppose that (λj , α̌i ) 6= 1 for all i and j. Then ∆(PV ) =
t∗+ ∩ hull(Wλ∗1 ∪ Wλ∗2 ∪ · · · ∪ Wλ∗k ).
32 REYER SJAMAAR

π2

π1

Diagram 2. K = G2 and λ = π2 (highest weight of complexified

adjoint representation, gC
2)

π2

π3
π1
0

Diagram 3. K = SU(4) and λ = π1 + π2 + π3 = 12

P
α∈Ψ+ α.
4
Vertices of ∆(PV ) are π1 + π2 + π3 , 0, 2π1 , 2π2 , 2π3 , 3 1 + π2 ,
π
4 5 5
3 π3 + π2 , π1 + 3 π3 and 3 π1 + π3

2. Suppose V is the direct sum of at least r + 1 copies of a unitary irreducible

representation with highest weight λ, where r is the rank of K. Then ∆(PV ) =
t∗+ ∩ hull Wλ∗ .
Proof. 1. This is easy to deduce from Lemma 7.1 and Proposition 7.3.
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 33

Lk
2. Write V = 1 V ′ , where k > r and V ′ is an irreducible module with highest
weight λ. Clearly, ∆(PV ) is a subset of t∗+ ∩hull Wλ∗ . The weight polytope hull Wλ
has exactly r edges containing the vertex λ. Let ν1 , ν2 , . . . , νr be the opposite
endpoints of these edges. Choose weight vectors v0 , v1 , . . . , vr , each coming from a
different copy of V ′ , and having weights λ, ν1 , ν2 , . . . , νr , respectively. Since each
copy of V ′ is K-invariant and they are all mutually orthogonal, hEα vi , vj i = 0 for
all roots α and for all i and j with 0 ≤ i < j ≤ r. Lemma 7.1 therefore tells us that
t∗+ ∩hull{λ∗ , ν1∗ , . . . , νr∗ } is contained in ∆(PV ). Hence, ∆(PV ) = t∗+ ∩hull Wλ∗ .
7.2. Cotangent bundles. Let Q be a connected K-manifold and let M be the
cotangent bundle of Q. Points in M will be written as pairs (q, p), where q ∈ Q
and p ∈ Tq∗ Q, and tangent vectors to M as pairs (δq, δp), where δq ∈ Tq Q and
δp ∈ Tp (Tq∗ Q). The standard one-form α on M is the K-invariant form defined
by α(q,p) (δq, δp) = p(δq). The two-form ω = −dα is symplectic, and the lifted
K-action on M is Hamiltonian with momentum map defined by Φξ = ιξM α, that
is, Φξ (q, p) = p(ξQ,q ). Clearly, Φ is homogeneous of degree one in p, so Φ−1 (0) is a
conical subset of M . In particular, Φ is not proper (not even if Q is compact) and
Theorem 6.7 does not apply. But the homogeneity of Φ also implies that ∆(M )
is equal to the cone on ∆(U ) for any neighbourhood U of the zero section. In
view of Theorem 6.5 this means that ∆(M ) = ∆(q,0) , where q is any point in Q.
Furthermore, Φ(M ) = −Φ(M ), so ∆(M ) = ∆(M )∗ . The symplectic slice W at
(q, 0) to the K-action on M is equal to T ∗ V , where V is the slice Tq Q/Tq (Kq)
at q to the K-action on Q. This implies that W = V + JV = V C for a suitable
C
Kq -invariant complex structure J on W , and so the variety G ×(Kq ) W is the
complexification of the real-algebraic variety K ×Kq V . We have proved:
Theorem 7.6. For every connected K-manifold Q, the set ∆(T ∗ Q) is a rational
convex polyhedral cone. It is invariant under the involution ∗ and equal to the
momentum cone of the complexification of the K-variety K ×Kq V . Here V =
Tq Q/Tq (Kq) is the slice at an arbitrary point q ∈ Q.
For instance, let L be a closed subgroup of K and

let Q be the homogeneous
space K/L. Then the theorem says that ∆ T ∗ (K/L) = ∆(G/LC ).
7.3. Symplectic quotients. Let M be a Hamiltonian K-manifold with momen-
tum map Φ : M → k∗ . Let L be a closed normal subgroup of K. Then M is a
Hamiltonian L-space with momentum map Φ(L) = ι∗ ◦ Φ : M → l∗ , where ι is the
inclusion of l into k. Let K̄ be the Lie group K/L. The kernel of ι∗ can be identified
in a natural way with k̄∗ , the dual of the Lie algebra of K̄. Let µ be any point in
z(k)∗ , where z(k) denotes the centre of k. The symplectic quotient of M at the level
ι∗ µ ∈ z(l)∗ with respect to the L-action,
Mι∗ µ = Mι∗ µ,L = Φ−1 ∗ −1
µ + k̄∗


(L) (ι µ)/L = Φ L,
is a stratified Hamiltonian K̄-space. (Cf. [31].) A momentum map Φ(K̄) : Mι∗µ →
k̄∗ ⊂ k∗ for the K̄-action on Mι∗ µ is induced by the map Φ−1 µ + k̄∗ → k̄∗ sending

m to Φ(m) − µ. (Up to a shift by an element of z(k̄)∗ , the map Φ(K̄) only depends
on the point ι∗ µ.) It is easy to calculate ∆(Mι∗ µ ) in terms of ∆(M ). Let T̄ be the
maximal torus T /(T ∩ L) of K̄. Then t̄∗ is naturally isomorphic to t∗ ∩ k̄∗ , and the
intersection t̄∗+ = t∗+ ∩ k̄∗ is a Weyl chamber of the pair (K̄, T̄ ). The following result
is now obvious (regardless of whether Mι∗ µ is smooth or not).
34 REYER SJAMAAR

Proposition 7.7. ∆(Mι∗ µ ) = −µ + ∆(M ) ∩ k̄∗ . Therefore, if ∆(M ) is a closed

convex polyhedral subset of t∗+ , then ∆(Mι∗ µ ) is a closed convex polyhedral subset
of t̄∗+ .
For example, consider the Hamiltonian K ×K-space T ∗ K ∼ = K ×k∗ with momen-
tum map (k, ν) 7→ (kν, −ν). Let L be any closed subgroup of K. The momentum
map for the restriction of the action to K × L is (k, ν) 7→ (kν, −ν|l ). The symplectic
quotient of T ∗ K at level 0 with respect to the normal subgroup {1} × L ⊂ K × L is
isomorphic as a Hamiltonian K-space to the cotangent bundle of the homogeneous
space K/L. Proposition 7.7 tells us that the momentum map image of T ∗ (K/L) is
the set K{ ν : ν|l = 0 }, and hence
∆ T ∗ (K/L) = Kl◦ ∩ t∗+ ,


(7.6)
where l◦ is
the annihilator of l in k∗ . (This can also be seen from the equality
∆ T ∗ (K/L) = ∆(G/LC ) proven in Section 7.2 and the Peter-Weyl Theorem.)
Consequently, the set Kl◦ ∩ t∗+ is a rational convex polyhedral cone.
Now assume that K is semisimple and take L to be the maximal torus T .
Kostant’s convexity theorem [17] implies that Kt◦ ∩ t∗+ = t∗+ . (Consider the natural
projection ι∗ : k∗ → t∗ . Take any µ ∈ t∗+ . By Kostant’s theorem the set ι∗ (Kµ) ⊂ t∗
is equal to the convex hull of the Weyl group orbit through µ, which contains the
origin in t∗ . Therefore Kµ ∩ t◦ = Kµ ∩ ker ι∗ is not empty.) We conclude from (7.6)
∗ ∗


that ∆ T (K/T ) = t+ .
More generally, take L to be the centralizer Kσ of a wall σ of the Weyl chamber
t∗+ . It is not difficult to see from the root space decomposition of the pair (k, t) that
Kk◦σ ∩ t∗+ contains the ray through every dominant root that is not perpendicular to
the wall σ. This is insufficient information to determine ∆ T ∗ (K/Kσ
) in general,

but if K is e.g. of type B2 or G2 , then this implies that ∆ T ∗ (K/Kσ ) = t∗+ for any
wall σ 6= {0}. If K = SU(n) and σ is the one-dimensional wall spanned
by either
π1 or πn−1 = π1∗ , one can easily calculate by hand that ∆ T ∗ (K/Kσ ) = Kk◦σ ∩ t∗+
is equal to the ray spanned by the maximal root α1 + · · · + αn−1 = π1 + πn−1 .
As another application of Proposition 7.7, I now give a proof of Proposition
6.9. For any µ ∈ t∗+ and for any closed subgroup L of Kµ , let M be the space
X(µ, L, {0}), the local model of Definition 6.2 with trivial symplectic slice W . By
(6.2), M is the bundle over the coadjoint orbit Kµ with fibre T ∗ (Kµ /L) furnished
with the minimal-coupling form. Put m = kµ /l. Then m∗ is canonically isomorphic
to the annihilator of l inside k∗µ . Clearly, the point m = (1, 0, 0) in K ×L m∗ ∼ =M
has the property that Km = L and Φ(m) = µ. Also, the symplectic cross-section
of M at m is just the cotangent bundle T ∗ (Kµ /L) with its standard momentum
map shifted by µ ∈ z∗µ . By (7.6), the local momentum cone of M at m is therefore
equal to
∆m = µ + Kµ m∗ ∩ t∗+,µ ,


(7.7)
where t∗+,µ denotes the positive Weyl chamber of kµ . Now assume that Kµ =
[Kµ , Kµ ]L, or, in other words, kµ = [kµ , kµ ] + l. Because of the decomposition
k∗µ = [kµ , kµ ]∗ ⊕ z∗µ , this is equivalent to m∗ ∩ z∗µ = {0}, or:
Kµ m∗ ∩ z∗µ = {0}. (7.8)
Now the Weyl chamber t∗+,µ is the product ofz∗µ and the Weyl chamber of the
semisimple part, t∗+,µ ∩ [kµ , kµ ], which is a proper cone. So the cone Kµ m∗ ∩ t∗+,µ
could only fail to be proper if Kµ m∗ contained a nontrivial linear subspace of z∗µ .
 CONVEXITY PROPERTIES OF THE MOMENT MAPPING RE-EXAMINED 35

But this is impossible because of (7.8). By (7.7), the point Φ(m) is therefore a
vertex of ∆(M ). This completes the proof of Proposition 6.9.

References
[1] D. Arnal and J. Ludwig, La convexité de l’application moment d’un groupe de Lie, J. Funct.
Anal. 105 (1992), 256–300.
[2] M. Atiyah, Convexity and commuting Hamiltonians, Bull. London Math. Soc. 14 (1982),
1–15.
[3] M. Brion, Sur l’image de l’application moment, Séminaire d’algèbre Paul Dubreuil et Marie-
Paule Malliavin (Paris, 1986) (M.-P. Malliavin, ed.), Lecture Notes in Mathematics, vol. 1296,
Springer-Verlag, Berlin-Heidelberg-New York, 1987, pp. 177–192.
[4] J. J. Duistermaat, Convexity and tightness for restrictions of Hamiltonian functions to fixed
point sets of an antisymplectic involution, Trans. Amer. Math. Soc. 275 (1983), 412–429.
[5] V. Guillemin and S. Sternberg, Convexity properties of the moment mapping, Invent. Math.
67 (1982), 491–513.
[6] , Convexity properties of the moment mapping, II, Invent. Math. 77 (1984), 533–546.
[7] , Symplectic techniques in physics, Cambridge Univ. Press, Cambridge, 1990, second
reprint with corrections.
[8] G. J. Heckman, Projections of orbits and asymptotic behavior of multiplicities for compact
Lie groups, Invent. Math. 67 (1982), 333–356.
[9] P. Heinzner and A. Huckleberry, Kählerian potentials and convexity properties of the moment
map, Invent. Math. 126 (1996), 65–84.
[10] P. Heinzner, A. Huckleberry, and F. Loose, Kählerian extensions of the symplectic reduction,
preprint, Ruhr-Universität Bochum, 1993.
[11] J. Hilgert and K.-H. Neeb, Symplectic convexity theorems and coadjoint orbits, Compositio
Math. 94 (1994), 129–180.
[12] G. Kempf and L. Ness, The length of vectors in representation spaces, Algebraic Geometry
(Copenhagen, 1978) (K. Lønsted, ed.), Lecture Notes in Mathematics, vol. 732, Springer-
Verlag, Berlin-Heidelberg-New York, 1979, pp. 233–244.
[13] F. C. Kirwan, Cohomology of quotients in symplectic and algebraic geometry, Mathematical
Notes, vol. 31, Princeton Univ. Press, Princeton, 1984.
[14] , Convexity properties of the moment mapping, III, Invent. Math. 77 (1984), 547–552.
[15] A. Klyachko, Stable bundles, representation theory and Hermitian operators, preprint,
Bilkent University, Bilkent, Turkey, 1996.
[16] B. Kostant, Quantization and unitary representations, Lectures in Modern Analysis and
Applications III (Washington, D.C.) (C. T. Taam, ed.), Lecture Notes in Mathematics, vol.
170, Springer-Verlag, Berlin-Heidelberg-New York, 1970, pp. 87–208.
[17] , On convexity, the Weyl group and the Iwasawa decomposition, Ann. Sci. École Norm.
Sup. (4) 6 (1973), no. 4, 413–455.
[18] H. Kraft, Geometrische Methoden in der Invariantentheorie, second revised ed., Aspekte der
Mathematik, vol. D1, Vieweg, Braunschweig/Wiesbaden, 1985.
[19] L. Laeng, Highest weight sets for actions of reductive groups, Rapport de stage, Ecole normale
supérieure, Lyon, unpublished.
[20] E. Lerman, E. Meinrenken, S. Tolman, and C. Woodward, Non-abelian convexity by sym-
plectic cuts, Topology, to appear, dg-ga/9603015.
[21] S. Lojasiewicz, Une propriété topologique des sous-ensembles analytiques-réels, Les équations
aux dérivées partielles (Paris, 1962) (B. Malgrange, ed.), Colloques internationaux du CNRS,
vol. 117, Editions du CNRS, Paris, 1963, pp. 87–89.
[22] , Ensembles semi-analytiques, Tech. report, Inst. Hautes Études Sci. Publ. Math.,
1965.
[23] J.-H. Lu and T. Ratiu, On the nonlinear convexity theorem of Kostant, J. Amer. Math. Soc.
4 (1991), no. 2, 349–363.
[24] D. Luna, Slices étales, Sur les groupes algébriques, Mém. Soc. Math. France 33 (1973),
81–105.
 36 REYER SJAMAAR

[25] C.-M. Marle, Constructions d’actions hamiltoniennes de groupes de Lie, Actions hamiltoni-
ennes de groupes—Troisième théorème de Lie (Journées lyonnaises de la SMF, 1986) (P. Da-
zord et al., eds.), Travaux en cours, vol. 27, Séminaire sud-rhodanien de géométrie VIII,
Hermann, Paris, 1988, pp. 57–70.
[26] A. Neeman, The topology of quotient varieties, Ann. of Math. (2) 122 (1985), 419–459.
[27] L. Ness, A stratification of the null cone via the moment map, Amer. J. Math. 106 (1984),
no. 6, 1281–1329, with an appendix by D. Mumford.
[28] G. W. Schwarz, The topology of algebraic quotients, Topological Methods in Algebraic Trans-
formation Groups (Rutgers Univ., 1988) (Hanspeter Kraft et al., eds.), Progress in Mathe-
matics, vol. 80, Birkhäuser, Boston, 1989, pp. 135–151.
[29] L. Simon, Asymptotics for a class of non-linear evolution equations, with applications to
geometric problems, Ann. of Math. (2) 118 (1983), 525–571.
[30] R. Sjamaar, Holomorphic slices, symplectic reduction and multiplicities of representations,
Ann. of Math. (2) 141 (1995), 87–129.
[31] R. Sjamaar and E. Lerman, Stratified symplectic spaces and reduction, Ann. of Math. (2)
134 (1991), 375–422.
[32] N. Wildberger, The moment map of a Lie group representation, Trans. Amer. Math. Soc.
330 (1992), 257–268.

School of Mathematics, Institute for Advanced Study, Princeton, New Jersey 08540
Current address: Department of Mathematics, Cornell University, Ithaca, New York 14853-
7901
E-mail address: [email protected]

View publication stats