The Kähler geometry of toric manifolds

Vestislav Apostolov
E-mail address: apostolov.vestislav@uqam.ca
 Contents

Introduction 5
Chapter 1. Delzant Theory 7
1. Hamiltonian group actions on symplectic manifolds 7
2. Hamiltonian actions of tori 10
3. The complex projective space 12
4. Toric symplectic manifolds from Delzant polytopes 17
5. Toric complex varieties from Delzant polytopes. Fans 19
6. Equivariant blow-up 20
7. Polarized projective toric varieties 22
8. Toric orbifolds 24
Chapter 2. Abreu–Guillemin Theory 27
1. The orbit space of a toric manifold 27
2. Toric Kähler metrics: local theory 27
3. The scalar curvature 30
4. Symplectic versus complex: the Legendre transform 32
5. The canonical toric Kähler metric 33
6. Toric Kähler metrics: compactification 34
Chapter 3. The Calabi Problem and Donaldson’s theory 37
1. The Calabi problem on a toric manifold 37
2. Donaldson–Futaki invariant 39
3. Uniqueness 40
4. K-stability 41
5. Toric test configurations 42
6. Uniform K-stability 45
7. Existence: an overview 48
Bibliography 59

3
 Introduction

These lecture notes grew from the mini-courses I gave at the University
of Nantes in 2017 and at the CIRM in 2019. Their purpose is to present a
relatively self-contained and fast introduction to the theory of extremal Kähler
metrics, pioneered by E. Calabi in the 1950’s and developed over the years. The
framework of toric varieties, well-studied and used in both symplectic geometry
and algebraic geometry, offers a good testing ground and has been instrumental
to the general theory in many ways.
I am grateful to Alexandro Lepé, Lars Sektnan, Isaque Viza de Souza, and
Yicao Wang for their careful reading of previous versions of the manuscript and
for spotting and correcting errors, and suggesting many improvements. I have
benefited from numerous discussions with David Calderbank, Paul Gauduchon,
Éveline Legendre and Lars Sektnan on toric geometry.

5
 CHAPTER 1

Delzant Theory

1. Hamiltonian group actions on symplectic manifolds

We start with some basic definitions in symplectic geometry.
Let (M 2m , ω) be a symplectic manifold of real dimension 2m, i.e. a smooth
manifold endowed with a closed 2-form ω which is non-degenerate at each point
p ∈ M meaning that ωp : Tp M → Tp∗ M is an isomorphism. A simple linear alge-
bra shows that the latter condition is also equivalent to ωp∧m 6= 0 ∈ ∧2m (T ∗p M ).
Example 1.1. The basic example of a symplectic manifold is (R2m , ω0 ) where, using the identification
2m ∼ m √
R = C = {z = (z1 , · · · , zm ) : zi = xi + −1yi , i = 1, . . . m},
the symplectic form is
m √ n
X −1 X
ω0 := dxi ∧ dyi = dzi ∧ dz̄i .
i=1
2 i=1

Example 1.2. Let M = S2 ⊂ R3 be the unit sphere endowed with the atlas of S2 given by the
stereographic projections from the north and the south poles, N = (0, 0, 1) and S = (0, 0, −1), respectively.
We denote by
2u 2v 1 − u2 − v 2 2
x= ,y = ,z = − , (u, v) ∈ R ,
1 + u2 + v 2 1 + u2 + v 2 1 + u2 + v 2
2ũ 2ṽ 1 − ũ2 − ṽ 2 2
x= 2 2
,y = 2 2
,z = , (ũ, ṽ) ∈ R
1 + ũ + ṽ 1 + ũ + ṽ 1 + ũ2 + ṽ 2
the corresponding chart patches on S2 . Recall that on S2 \ ({N } ∪ {S}), the transition between the coordiantes
u v 2
(u, v) and (ũ, ṽ) is given by the diffeomorphism ũ = u2 +v 2 , ṽ = u2 +v 2 of R \ {(0, 0)}. Then the 2-form

4du ∧ dv 4dũ ∧ dṽ

ωS2 := =−
(1 + u2 + v 2 )2 (1 + ũ2 + ṽ 2 )2

introduces a symplectic structure on S2 .

Definition 1.1. Let M be a smooth manifold and G a Lie group. An action

of G on M is a group homomorphism
ρ : G → Diff(M ),
where Diff(M ) stands for the diffeomorphism group of M . The action of G is
smooth if the evaluation map
ev : G × M → M, ev(g, p) := ρ(g)(p),
is a smooth map between manifolds.
Let (M, ω) be a symplectic manifold and G a Lie group acting smoothly on
M . We say that G acts symplectically on (M, ω) if
ρ(g)∗ (ω) = ω.
Example 1.3. Consider (M, ω) = (R2m , ω0 ) as in Example 1.1. The action of the m-dimensional torus
√ √
m −1t1 −1tm
T = (e ,...,e ),
given by √ √ √ √
−1t1 −1tm −1t1 −1tm
ρ(e ,...,e )(z) := (e z1 , . . . , e zm )
is symplectic.

7
8 1. DELZANT THEORY

Example 1.4 (Hamiltonian flows). Let f be a smooth function on (M, ω). Using the non-degeneracy
of ω, we define a vector field by
−1
Xf := −ω (df ),
called the Hamiltonian vector field of f . Suppose that Xf is complete, i.e. its flow ϕt is defined for all t ∈ R
(this always holds if M is compact). Then, ρ(t) := ϕt defines a symplectic action of R on (M, ω). Indeed, we
have
d ∗ ∗
d
∗

(ϕ ω) = ϕs (ϕ ω)
dt |t=s t dt |t=0 t
 
∗
= ϕs LXf ω
 
∗
= ϕs dıXf + ıXf d)(ω)
 
∗
= ϕs d(−df ) = 0,
∗
ϕ0 (ω) = ω.

The previous two examples show that different groups (in particular Tm and
Rm ) can possibly give rise to “equivalent” symplectic actions, in the sense that
their images in Diff(M ) are the same. A way to normalize the situation is to
consider
Definition 1.2. A symplectic action of a Lie group G on (M, ω) is effective
if the homomorphism ρ : G → Diff(M ) has trivial kernel.
Thus, the symplectic action of Rm = {(t1 , . . . , tm )} on (R2m , ω0 ) defined in
Example 1.3 is not effective, whereas the action of Tm on the same manifold is.
In the sequel, we shall be interested in effective symplectic actions.
Let G be a Lie group acting smoothly on M . Denote by g = Lie(G) its Lie
algebra, i.e. g is the vector space of left-invariant vector fields on G. For any
ξ ∈ g, we denote by exp(tξ) the corresponding 1-parameter subgroup of G, and
by Xξ the vector field on M induced by the one parameter subgroup ρ(exp(tξ))
in Diff(M ), i.e.
d  
Xξ (p) := ρ(exp(tξ))(p) .
dt |t=0
The vector field Xξ is called fundamental vector field of ξ ∈ g.
We recall that G also acts on itself by conjugation
ρ(g)(h) := ghg −1 , g ∈ G, h ∈ G,
fixing the unitary element e ∈ G and thus inducing a linear action on the vector
space Te G ∼= g, called the adjoint action
Ad : G → GL(g).
The induced linear action on the dual vector space Te∗ G ∼ = g∗ is given by
Ad∗ (g)(α) := α ◦ Ad(g −1 ) and is called the co-adjoint action. We are now
ready to give the definition of a hamiltonian action on (M, ω).
Definition 1.3. Let ρ : G → Diff(M ) be a smooth action of a Lie group G
on (M, ω). It is said to be hamiltonian if there exists a smooth map
µ : M → g∗ ,
called a momentum map, which satisfies the following two conditions.
(i) For any ξ ∈ g, the fundamental vector field Xξ satisfies
ω(Xξ , ·) = −dhµ, ξi,
where h·, ·i denotes the natural pairing between g and its dual g∗ .
 1. HAMILTONIAN GROUP ACTIONS ON SYMPLECTIC MANIFOLDS 9

(ii) µ is equivariant with respect to the action ρ of G on M and the co-

adjoint action Ad∗ of G on g∗ , i.e.
µ ρ(g)(p) = Ad∗ (g) µ(p) .

Remark 1.1. (a) Notice that the condition (i) of Definition 1.3 implies that
any fundamental vector field Xξ is hamiltonian (see Example 1.4) with respect
to the smooth function
(1) µξ (p) := hµ(p), ξi.
In particular, hamiltonian actions are symplectic.
(b) By a standard argument in Lie theory, the condition (ii) of Definition 1.3
can be equivalently expressed as
dµξ (Xη ) = −µ[η,ξ] .
For any two smooth functions f, h on (M, ω)
{f, h}ω := ω(Xf , Xh )
is the so-called Poisson bracket. Thus, the condition (ii) of Definition 1.3 is
equivalent to
(2) {µξ , µη }ω = −µ[ξ,η] .
An important example is the following
Example 1.5. Suppose G is a semi-simple compact Lie group with Lie algebra g, and T ⊂ G is a
maximal torus with Lie algebra t ⊂ g. Complexifying, we also have a semi-simple complex Lie group GC with
Lie algebra gc C := g ⊗ C, and a√Cartan sub-algebra
√
h := t ⊗ C. A specific example which one can bear in
mind is G = SU(n), T = {diag(e −1t1 , . . . , e −1tm ) : t1 + · · · + tm = 0}. Then GC = SL(n, C).
The general theory of semi-simple Lie algebras yields that gC admits a root decomposition
C
M C C
g =h⊕ (g−α ⊕ gα ),
α∈R+

∗
where R ⊂ h denotes the finite subspace of roots of gC and R = R+ ∪ (−R+ ) with R+ ∩ (−R+ ) = ∅ is the
choice of a subset R+ of positive roots. For α ∈ R, we have set
C C
gα := {u ∈ g : adx (u) = α(x)u, ∀x ∈ h}.

We denote by M an orbit in g for the adjoint action Ad : G → GL(g) and assume that M is not a point.
The general theory tells us that M is aways the orbit of a non-zero element x ∈ t, and that such x is unique
if we require moreover that −iα(x) ≥ 0 for all α ∈ R+ . If the unique x ∈ t determined as above also satisfies
−iα(x) > 0 for all α ∈ R+ , then the orbit M ∼ = G/T is principal for the adjoint action of G, and M is called
a flag manifold. In general, M ∼ = G/Gx where Gx is the subgroup of G fixing x (and whose Lie algebra is
gx = Ker(adx )). √
In the case G = SU(n), this corresponds to the basic fact that any hermitian matrix a ∈ −1su(n, C)
can be diagonalized by a conjugation with an element of SU(n). The diagonal matrix is uniquely determined
up to a permutation of its (real) eigenvalues, so we can normalize it by ordering the eigenvalues in decreasing
order. A hermitian matrix a determines a regular orbit for the action of G = SU(n) by conjugation precisely
when a has simple spectrum.
Each orbit M admits a natural symplectic form ω, called the Kirillov–Kostant–Souriau form, whose
definition we now recall. As the tangent space Ty M for any y ∈ M is generated by the fundamental vector
fields for the adjoint action of G, Ty M can be identified with the image my ⊂ g of the map
d
(3) u→ Adexp(tu) (y) = −ady (u).
dt |t=0
For u, v ∈ g, we denote by
hu, vi := tr(adu ◦ adv )
the Killing form of g. By assumption h·, ·i is negative definite (G being compact and semi-simple). We then
set
ωy (u, v) := −hy, [u, v]i,
where u, v ∈ Ty M ∼= Im(ady ). Then ω gives rise to a symplectic structure on M = G/Gx . The main fact is
Proposition 1.1 (Kirillov–Kostant–Souriau). The 2-form ω defines a symplectic structure on M =
G/Gx , the adjoint action of G on (M, ω) is hamiltonian with momentum map µ : M → g∗ identified with
the inclusion ı : M ⊂ g composed with the Killing form h·, ·i : g → g∗ .
10 1. DELZANT THEORY

2. Hamiltonian actions of tori

We now specialize to the case when the Lie group G = Tk is a k-dimensional
torus.
Lemma 1.1. Suppose G = Tk acts symplectically on (M, ω). Then, the action
is hamiltonian if and only if for any ξ ∈ g, there exists a smooth function µξ on
M , such that ıXξ ω = −dµξ . In this case, the momentum map µ : M → g∗ is
determined up to the addition of a vector in g∗ .
Proof. If the action of G is hamiltonian, the existence of µξ follows from
Remark 1.1(i).
Conversely, suppose that each fundamental vector field Xξ is hamiltonian
with respect to a smooth function µξ on M . Choose a basis {ξ1 , . . . , ξk } of g and
let X1 , . . . , Xk and µ1 , . . . , µk be the corresponding fundamental vector fields
and hamiltonian function, respectively. We claim that for ξ = ki=1 ai ξi , the
P
function
Xk
(4) µξ := a i µi
i=1
is a hamiltonian of the induced fundamental vector field Xξ . Indeed, since G
is abelian, exp(tξ) = exp(ta1 ξ1 ) ◦ · · · ◦ exp(tak ξk ) and, therefore, the induced
fundamental vector field is Xξ = ki=1 ai Xξi . The claim then follows trivially.
P
Thus, we can define µ : M → g∗ by letting
hµ(p), ξi := µξ .
It remains to show that the condition (2) holds. As the co-adjoint action
of an abelian group G is trivial, this condition reduces to show that µξ is an
invariant function under the action of G. Equivalently, by Remark 1.1(ii), we
have to show that for any fundamental vector fields Xξ , Xη , the smooth function
{µξ , µη }ω = ω(Xξ , Xη ) identically vanishes on M . Since the action of G is
symplectic, for any fundamental vector field Xζ , LXζ ω = 0. It follows (by using
that G is abelian again) that LXζ (ω(Xξ , Xη )) = 0. Thus, ω(Xξ , Xη ) is a constant
function on each orbit O ⊂ M for the action of G on M . It is a standard fact
that O is a homogeneous manifold G/Gp , where p ∈ O and Gp is the stabilizer
of p in G, see e.g. [5]. As G is compact, O is a compact manifold (see [5],
Chapter 1, Corollary 1.3), and therefore the restriction of µξ to O has a critical
point. At this point, ω(Xξ , Xη ) = −dµξ (Xη ) = 0, thus showing that the function
ω(Xξ , Xη ) is identically zero on O, and hence on M .
The last statement trivially follows as we can take instead of µi the hamil-
tonian µi + λi with λi ∈ R or, equivalently, we can replace in (4) µξ by µξ + hξ, λi
for λ ∈ g∗ . 
For the next result, we use the facts that if a compact Lie group G acts
effectively on M , then each orbit O = G(p) is a compact homogeneous manifold
of dimension ≤ dim G and, furthermore, that there exist an open dense subset
M 0 ⊂ M of points whose orbits are of dimension = dim G (called principal
orbits of G). We refer to [5] for a proof of these facts.
Lemma 1.2. Suppose that (M 2m , ω) admits an effective hamiltonian action
of G = Tk . Then k ≤ m.
 2. HAMILTONIAN ACTIONS OF TORI 11

Proof. By the argument in the proof of Lemma 1.1, the tangent space Tp O
of a principal orbit O = G(p) is ωp isotropic subspace of Tp M . Its dimension
(= dim G), therefore, is ≤ m. 
Example 1.6. We return again to Example 1.3 and will now show that the symplectic action of Tm
on (R2m , ω0 ) is hamiltonian. We shall work on the dense open subset (C √
m 0
) = {(z1 , . . . , zm ) : zi 6= 0, ∀i =
1, . . . m} of Cm ∼
=R
2m
on which we introduce polar coordinates zi = ri e −1ϕi , i = 1, . . . , m. The symplectic
2-form ω0 then becomes
Xm
ω0 = ri dri ∧ dϕi ,
i=1
whereas the fundamental vector fields {X1 , . . . , Xm } associated to the standard basis of Rm = {(t1 , . . . , tm )} ∼
=
Lie(Tm ) are Xi = ∂ϕ∂
. It follows that ıXi ω0 = −ri dri = −d( 12 ri2 ). In other words, the smooth map
 i ∗
µ : Cm → Rm ∼ m
= Lie(T ) defined by

1 2 2

(5) µ(z) := |z1 | , . . . , |zm |
2
is a momentum map for the action of Tm on (Cm )0 , and hence on Cm (by continuity). We also notice that
m
(6) Im(µ) = Cm := {(x1 , . . . , xm ) ∈ R : xi ≥ 0}.
1 2
Exercise 1.1. Consider the S -action on (S , ωS2 ) corresponding to the rotation around the z-axis of
R3 , see Example 1.2. Show that this is hamiltonian with momentum map given by the z-ccordinate.

The central result in the theory is the following convexity result

Theorem 1.1 (Atiyah [4], Guillemin–Stenberg [30]). Suppose Tk acts
in a hamiltonian way on a connected compact symplectic manifold (M, ω), with
momentum map µ : M 7→ Rk . Then,
(i) The image of µ is the convex hull ∆ ⊂ Rk of the images of the fixed
points for the Tk -action on M .
(ii) The pre-image of any point of ∆ is connected.
Example 1.7 (Schür’s Theorem). This is the main application of Theorem 1.1 given in [4]. In the
general setting of Example 1.5, let us take G = SU(n) and consider the orbit M = Ma of all conjugated
hermitian matrices to a given hermitian matrix a by elements of G. We thus have a hamiltonian action on
(M, ω) of the (n − 1)-dimensional torus T ⊂ SU(n) of diagonal matrices in SU(n). Furthermore, it easily
follows by Proposition 1.1 that the momentum map for this action is
µ(b) = diag(b11 , . . . , bnn ),
where b is any hermitian matrix in Ma and (b11 , . . . , bnn ) are the diagonal elements of b (a vector in Rn ).
Furthermore, the fixed points for the action of T on Ma are precisely the diagonal hermitian matrices in Ma .
We thus obtain from Theorem 1.1 the following classical result due to Schür.
Corollary 2.1. Let a be a hermitian matrix in Cn with spectrum (λ1 , . . . , λn ) ∈ Rn . Then the
diagonal elements of the hermitian matrices in the conjugacy class of a by elements of U(n) consisting of
all points in the convex hull in Rn of {(λσ(1) , . . . , λσ(n) ), σ ∈ Sn }, where Sn denotes the symmetric group
of permutation of n-elements.

2.1. Toric symplectic manifolds and Delzant Theorem. In view of

Lemma 1.2, we give the following
Definition 1.4. A symplectic toric manifold is a compact connected sym-
plectic manifold (M 2m , ω) endowed with an effective hamiltonian action ρ of a
torus T with
1
dim T = m = dim M.
2
Two toric symplectic manifolds (Mi , ωi , Ti , ρi ), i = 1, 2 are considered equivalent
if there exist an isomorphism of Lie groups φ : T1 → T2 and a diffeomorphism
Φ : M1 → M2 with Φ∗ ω2 = ω1 , satisfying
Φ(ρ1 (g)(p)) = ρ2 (φ(g))(Φ(p)), ∀g ∈ T1 , ∀ p ∈ M1 .
By Lemma 1.1, we can always assume that in this case the momentum maps µi
of (Mi , ωi , Ti , ρi ) are linked by µ1 = µ2 ◦ Φ.
12 1. DELZANT THEORY

By virtue of Theorem 1.1, the image of a toric symplectic manifold of di-

mension 2m is a compact convex polytope ∆ in the m-dimensional vector space
t∗ = Lie(T). The Delzant theorem provides a classification (up to the equiva-
lence of Definition 1.4) of symplectic toric manifolds (M, ω, T) in terms of the
corresponding polytopes ∆ ⊂ t∗ . To state it properly, we notice that being Lie
algebra of a torus T, the vector space t comes equipped with a lattice Λ ⊂ t such
that 2πΛ = exp−1 (e) (where e stands for the identity element of T). In other
words,
exp : t/2πΛ ∼
= T.
Definition 1.5. Let V be an m-dimensional real vector space whose dual
space is V ∗ , and let Λ ⊂ V be a lattice. We denote also by V ∗ the affine space
determined by V ∗ . Let ∆ ⊂ V ∗ be a convex polytope written as the intersection
of a minimal number of d linear inequalities
∆ = {x ∈ V ∗ : Lj (x) = huj , xi + λj ≥ 0, j = 1, . . . , d},
where uj ∈ V are called (labelled) normals of ∆ and λj ∈ R. We shall refer to
the collection L = {L1 , . . . , Ld } of affine-linear functions defining ∆ as a labelling
of ∆, and to the couple (∆, L) as a labelled polytope in V . We say that the data
(∆, L, Λ) define a Delzant polytope if the following conditions are satisfied:
(i) ∆ is compact;
(ii) ∆ is simple, meaning that each vertex x0 of ∆ annihilates precisely m
of the affine functions in L and the corresponding normals form a basis
of V ;
(iii) ∆ is integral, meaning that for each vertex x0 of ∆,
spanZ {uj ∈ V : Lj (x0 ) = huj , x0 i + λj = 0} = Λ.
We say that (∆, L, Λ) defines a rational Delzant polytope if instead of (iii) we
require the weaker assumption
(iii)’ ∆ is rational, meaning that for each Lj ∈ L, the normal uj ∈ Λ.
Exercise 1.2. Show that if the triple (∆, L, Λ) is a Delzant polytope, then (∆, Λ) determine the
labelling L. Is this true for a rational Delzant polytope (∆, L, Λ)?

Theorem 1.2 (Delzant [16]). There exists a bijective correspondence be-

tween the equivalence classes of 2m-dimensional toric symplectic manifolds and
equivalent Delzant polytopes (∆, L, Λ) in an m-dimensional vector space V ∗ , up
to the natural action of the affine group Aff(V ∗ ) on the triples (∆, L, Λ).
We shall not develop in these notes a detailed proof of this key result, but in
the next section we are going to sketch one direction of it, namely the Delzant
construction which associates to a Delzant triple (∆, L, Λ) a toric symplectic
manifold (M, ω, T).

3. The complex projective space

We start with the following basic example of a toric symplectic manifold.
Recall that the (complex) m-dimensional projective space
CP m = Cm+1 \ {0}/C∗
is the quotient of Cm+1 \ {0} by the diagonal (holomorphic) action of C∗ given
by
(λ, z) → λz, λ ∈ C∗ , z ∈ Cm+1 \ {0}.
 3. THE COMPLEX PROJECTIVE SPACE 13

This description naturally realizes M = CP m as a complex manifold, by passing

from the homogenous coordinates [z0 , . . . , zm ] (where (z0 , . . . , zm ) ∈ Cm+1 \ {0}
and [z0 , . . . , zm ] stands for the equivalence under the diagonal action of C∗ ) to
the affine charts Ui ∼ = Cm on the open subsets of CP m where zi 6= 0. In order
to describe the Fubini–Study metric on CP m , or its symplectic form, it is more
convenient to consider the identification
CP m = S2m+1 /S1 ,
where S2m+1 ⊂ Cm+1 is the unit sphere with respect to the standard euclidean
product of R2m+2 ∼
= Cm+1 , and S1 denotes the diagonal circle action on S2m+1 ⊂
Cm+1 by
√ √
−1t −1t
(7) ρ(e )(z0 , . . . , zm ) = e (z0 , . . . , zm ).
Definition 1.6. Let (M, ω) be a symplectic manifold. An ω-compatible
riemannian metric g on M is a riemannian metric such that the field of endo-
morphisms J defined by
gp (Jp u, v) = ωp (u, v), ∀p ∈ M, ∀u, v ∈ Tp M,
is an almost-complex structure on M , i.e. satisfies
Jp2 = −Id|Tp M .
We say that (g, J) is an ω-compatible almost-Kähler structure on M .
If, furthermore, J is an integrable almost-complex structure, i.e. satisfies
(8) N J (X, Y ) = [X, Y ] − [JX, JY ] + J[JX, Y ] + J[X, JY ] = 0,
then (g, J) defines a compatible Kähler structure on (M, ω).
Remark 1.2. (a) It is well-known (see e.g. [42]), and easy to show, that
any symplectic manifold admits infinitely many compatible almost-Kähler struc-
tures. Actually, the space of such structures, denoted AK(ω), is a contractible
Fréchet manifold. The group Symp(M, ω) of symplectomorphisms of M natu-
rally acts on AK(ω) by pull-backs of the riemannian metrics g. It thus follows
that for any compact subgroup G ⊂ Symp(M, ω) one can take an average over
G, and thus produce a G-invariant ω-compatible almost-Kähler structure (g, J)
on (M, ω).
(b) By a result of Newlander–Nirenberg [44], the integrability condition (8)
is equivalent to the existence of a holomorphic atlas on M , compatible with the
almost-complex structure J, i.e. to (M, J) being a complex manifold. There are
many examples of symplectic manifolds which do not admit compatible Kähler
structures.
In our situation, the standard flat metric of Cm+1 = R2m+2
m
X
(9) g0 = (dx2i + dyi2 )
i=0

is ω0 -compatible and the corresponding almost complex structure J0 is simply

the standard complex structure on Cm+1 . As both ω0 and g0 are preserved by
the Tm+1 -action, we thus have a Tm+1 -invariant flat Kähler structure (g0 , J0 )
on Cm+1 . The restriction of the flat metric g0 on Cm+1 to S2m+1 induces the
14 1. DELZANT THEORY

2m+1
canonical round metric g S and the projection map (known as the Hopf fibra-
tion)
π : S2m+1 → CP m ,
2m+1
provides an example of a riemannian submersion between (S2m+1 , g S ) and
the (uniquely determined) riemannian metric g F S on CP m , introduced by the
property   2m+1
π ∗ gπ(z)
FS
= gzS |Hz ,
 ⊥
where at each point z ∈ S2m+1 , Hz = Ker(π∗ )z . We use a similar con-
struction in order to endow CP m with a symplectic form ω, using the pointwise
isomorphism (π∗ )z : Hz ∼ = Tπ(z) CP m and the restriction of ω0 to the subspace
Hz . As all tensor fields used in this construction are Tm+1 -invariant, and the
S1 -action√
(7) is obtained
√
by restricting the action of Tm+1 to its S1 subgroup
N = (e −1t ,...,e −1t ) ⊂ Tm+1 , there is a natural induced Tm = Tm+1 /N -
action on CP m = S2m+1 /N . With respect to this data, (CP m , ω F S , Tm ) is an
example of a symplectic toric manifold. In fact, this is just a very particular case
of a general fact in symplectic geometry, known as symplectic reduction, which
we shall present below.
To this end, let us summarize the construction of CP m : by Example 1.6,
(C m+1 , ω0 ) is a symplectic manifold endowed with the hamiltonian action on
Tm+1 with momentum map
1
µTm+1 (z) = (|z0 |2 , . . . , |zm |2 ).
2
N ⊂ Tm+1 is a circle subgroup and the momentum map for the action of N is
therefore
m
1X 2
µN (z) = |zi | .
2
i=0
Thus, 1
S2m+1 = µ−1
N
2
is the level set of the momentum map µN , and furthermore, N acts freely on
this level set. We are thus in the situation of the following fundamental result.
Proposition 1.2 (Marsden–Weinstein [41], Meyers [43]). Let N ⊂ G
be a closed normal subgroup of a compact group G acting in a hamiltonian way
on a symplectic manifold (M̃ , ω̃). Let µG : M̃ → g∗ be a momentum map for
the G-action, and µN : M̃ → n∗ the corresponding momentum map for the N -
action, obtained from µG by composing with the natural projection ı∗ : g∗ → n∗
(which is adjoint to the inclusion of the Lie algebras ı : n ⊂ g).
Suppose further that c̃ ∈ g∗ is a fixed point for the co-adjoint action of G on
g , such that ı∗ (c̃) = c is a regular point for µN , and that the action of N on
∗
−1
µN (c) is free.
Then, the N -invariant symplectic 2-form
 ω̃, restricted to µ−1
N (c) defines a
symplectic form on the manifold M := µ−1N (c) /N , and the natural action of
G/N on M is hamiltonian with moment map, viewed as an N -invariant function
on µ−1
N (c), given by
µ = µG − c̃.
 3. THE COMPLEX PROJECTIVE SPACE 15

Proof. By construction, S := µ−1

N (c) ⊂ M̃ is a closed submanifold of M̃ on
which G acts smoothly. To see that G acts on S, notice that, by construction,
p ∈ S iff
(10) hµG (p) − c̃, ξi = 0, ∀ξ ∈ n.
Using the equivariance of the momentum map, we have for any g ∈ G
(µG (g · p) − c̃), ξ = Ad∗g (µG (p)) − c̃, ξ




= Ad∗g (µG (p) − c̃), ξ




= (µG (p) − c̃), Adg−1 (ξ)


= (µG (p) − c̃), ξ ,
where for passing from the second line to the third we have used that c̃ is a fixed
point for the co-adjoint action, and for passing from the third line to the forth
the fact that N is a normal subgroup of G.
We denote by
i : S ,→ M̃
the inclusion map and by
π:S→M
the projection map, which is a smooth submersion by the regularity assumption
for c. It follows from (10) that µG − c̃ is a smooth function on S with values in
the annihilator

∗ Ann(n) ⊂ g∗ . The latter subspace is canonically identified with
g/n . We summarize the construction in the following diagram:
µ
M̃ −−−G−→ g∗
x

i
(µG )|
(11) S
S −−−−→ g∗
 x
π `:=j+c̃
y 
µ
M −−−−→ (g/n)∗
where j denotes the natural inclusion of (g/n)∗ in g∗ , and ` = j + c̃ denotes the
affine map from (g/n)∗ to g∗ obtained by composing j with the translation by
c̃, i.e. `(x) = j(x) + c̃, for any x in (g/n)∗ .
As π : S → M is a smooth submersion, at each p ∈ S we let Vp :=
Ker(π∗ )p be the vertical distribution. By Remark 1.2-(a), we can introduce
an ω̃-compatible, G-invariant riemannian metric g̃ on M̃ , and use it to define
the horizontal space
⊥
Hp := Vp g̃ ⊂ Tp S.
As g̃ is G-invariant, π : (S, g̃) → M defines a riemannian submersion, thus giving
rise to a riemannian quotient metric g on M with the property
(π ∗ g)|H = g̃|H .
Similarly, there exists a 2-form ω on M , determined by the property
(π ∗ ω)|H = ω̃|H .
We claim that ω is non-degenerate and closed.
16 1. DELZANT THEORY

To see the non-degeneracy of ω, we observe that g is an ω-compatible rie-

mannian metric. As g̃ is ω̃-compatible on M̃ , it is enough to show that the corre-
sponding almost complex structure J˜ preserves the subspace Hp ⊂ Tp S ⊂ Tp M̃ .
This will follow if we show that for each p ∈ S,
˜ p ⊕ Hp
Tp M̃ = Vp ⊕ JV
is a g̃p -orthogonal decomposition. For any ξ ∈ n, the corresponding fundamental
vector field Xξ (p) belongs to Vp and, furthermore, we have


(12) (ıXξ ω̃)|Tp S = −d hµG − c̃, ξi|S = 0.
Notice that, as the action of N on S is free, the fundamental vector fields
˜ p ⊥g̃ (Vp ⊕ Hp ). As Vp ⊥g̃ Hp by
{Xξ (p), ξ ∈ n} span Vp . Thus, (12) reads as JV
construction, the claim follows.
We now discuss the closedness of ω. For any vector field X on M , we denote
by X̃ the corresponding horizontal lift to S, i.e. the unique section of H ⊂ T S
such that π∗ (X̃) = X. For any vector fields X, Y on M , we thus have the
orthogonal decomposition
(13) ^
[X, Y ]p = [X̃, Ỹ ]p + Πp ([X̃, Ỹ ]p ),
where Πp (·) denotes the orthogonal projection of Tp S to Vp . Using Cartan’s
formula and the fact that Vp ⊂ Ker(ω̃p ), we compute
 
dω(X, Y, Z) = σX,Y,Z d(ω(Y, Z))(X) − ω([X, Y ], Z)
 
^
= σX̃,Ỹ ,Z̃ d(ω̃(Ỹ , Z̃))(X̃) − ω̃([X, Y ], Z̃)
 
= σX̃,Ỹ ,Z̃ d(ω̃(Ỹ , Z̃))(X̃) − ω̃([X̃, Ỹ ], Z̃)
= dω̃(X̃, Ỹ , Z̃) = 0.
We finally discuss the induced action of G/N on M . The action of [g] ∈ G/N
on a point π(p) ∈ S/N is defined by π(g·p). (This is well-defined as S is invariant
under the action of G, as we have already shown.) We have also observed that
when restricted to S, the function ν := µG − c̃ takes values in the annihilator
n0 ∼= (g/n)∗ of n in g∗ . Using that N is normal, and that µG is G-equivariant,
it follows that µG is N -invariant. It then descends to define a smooth, G/N
equivariant function µ : M → (g/n)∗ on M = S/N . By construction, for any
[ξ] ∈ g/n,
dhµ, [ξ]iπ(p) = dhν, ξip
= dhµG , ξip
= −(ıXξ (p) ω̃)|Tp S
= −(ıXξ (p) ω̃)|Hp
= −(ıX[ξ] ω)π(p) ,
where we have used that Vp ⊂ Ker(ω̃p ) and π∗ (Xξ ) = X[ξ] . 
Exercise 1.3. (1) Show that the induced 2-form ω on M in the proof of Proposition 1.2 is independent
of the choice of a G-invariant ω̃-compatible riemannian metric on M̃ .
(2) Show that if g̃ is a G-invariant, ω̃-compatible riemannian metric on M̃ , which defines a Kähler
structure on M̃ , then the almost-Kähler structure (g, ω, J) on M defined in the proof of Proposition 1.2 is
Kähler.
 4. TORIC SYMPLECTIC MANIFOLDS FROM DELZANT POLYTOPES 17

(3) Show that the complex structure J induced on M = S2m+1 /S1 makes (M, J) biholomorphic to
CP m = (C \ {0})/C∗ endowed with its atlas of affine charts.

Now we can apply Proposition 1.2 in order to conclude that (CP m , ω F S ) is a

toric symplectic manifold under the action of the torus T = Tm+1 /S1 , and that
the induced Fubini–Study metric g F S gives rise to a T-invariant Kähler structure
on (CP m , ω F S ). Furthermore, the image of the induced momentum map
µ : CP m → (tm+1 /t1 )∗ ⊂ (tm+1 )∗
is identified with the intersection of Im(µTm+1 ) = Cm+1 = {xi ≥ 0, i = 0, . . . , m} ⊂
Rm+1 with the hyperplane x0 + · · · + xm = 12 , which is a simplex in this hyper-
plane. Alternatively, we can consider the subtorus Tm ⊂ Tm+1 defined by
√ √
−1t1 −1tm
Tm = (1, e ,··· ,e )
so that we have an exact sequence of Lie groups
{e} −−−−→ Tm −−−−→ Tm+1 −−−−→ Tm+1 /S1 −−−−→ {e},
giving rise to an isomorphism between T = Tm+1 /S1 and Tm . Using the induced
projection of the dual Lie algebras (tm+1 )∗ → (tm )∗ ∼
= Rm , the moment map µ
m
sends CP onto the simplex
n 1 X m  o
∆m = (x1 , . . . , xm ) : − xi ≥ 0, xi ≥ 0, i = 1, . . . , m, .
2
i=1
Finally, the lattice Λ of the dual space tm is just the standard lattice Zm ⊂
Rm ∼
= (Rm )∗ and we thus conclude
Lemma 1.3. (CP m , ω F S ) is a toric symplectic manifold under the induced
action of Tm , classified by the standard simplex in ∆m ⊂ Rm , labelled by
n 1 X m o
L = L1 (x) = x1 , . . . , Lm (x) = xm , Lm+1 (x) = − xi ,
2
i=1
and the standard lattice Zm ⊂ (Rm )∗ .

4. Toric symplectic manifolds from Delzant polytopes

The discussion in the previous subsection is the main tool of the explicit
construction, proposed by Delzant in [16], which associates a toric symplectic
manifold (M∆ , ω, T) to any Delzant triple (∆, L, Λ), as in Definition 1.5. We first
notice that by the condition (iii) of Definition 1.5, the lattice Λ is determined
by the labelling L = {L1 (x), . . . , Ld (x)}. Indeed, Λ is the span over Z of the
linearizations dLi := ui ∈ V ∗ of the affine-linear functions Lj (x) = huj , xi + λi ∈
L. We denote by T = V /2πΛ the corresponding torus, and by t = V its Lie
algebra. Thus, ∆ ⊂ t∗ . With this in mind, we consider the linear map τ : Rd → t,
defined by
d
X
τ (ξ1 , . . . , ξd ) := ξj uj .
j=1
Using the Delzant condition (i)-(ii)-(iii), one checks that
Claim 1. τ sends the standard lattice Zd of Rd onto Λ and thus defines a
homomorphism of tori
τ : Rd /2πZd = Td → T.
18 1. DELZANT THEORY

Claim 2. The kernel N of τ : Td → T is a connected subgroup of Td , i.e. it is

a (d − m)-dimensional torus. (This is no longer true if we consider the weaker
conditions (i)-(ii)-(iii)’ : then N can be the product of a torus with a finite
abelian group.)
We denote by n (resp. n∗ ) the Lie algebra (resp. its dual) of N . We thus
have an exact sequence of Lie groups
ı τ
{e} −−−−→ N −−−−→ Td −−−−→ T −−−−→ {e}
and the corresponding exact sequence of Lie algebras
ı τ
{0} −−−−→ n −−−−→ Rd −−−−→ t −−−−→ {0},
and its dual sequence
τ∗ ı∗
(14) {0} −−−−→ t∗ −−−−→ (Rd )∗ −−−−→ n∗ −−−−→ {0}.
We now consider the hamiltonian action of Td on (Cd , ω0 ) with momentum
map µTd (z) = 21 (|z1 |2 , . . . , |zd |2 ) ∈ (Rd )∗ . We denote by µN (z) = ı∗ ◦ µTd (z) the
momentum map for the action of N ⊂ Td .
By acting with a translation on ∆, we can assume without loss of generality
that the origin of t∗ is in the interior of ∆. Then, letting λ := (λ1 , . . . , λd ) =
(L1 (0), . . . , Ld (0)) ∈ (Rd )∗ , we have
Claim 3. S := µ−1 ∗
N (ı (λ)) is a compact submanifold of C .
d

As λi > 0, λ is in the interior of the momentum image Cd = {(x̃1 , . . . , x̃d ) ∈

(Rd )∗ : x̃i ≥ 0} of Cd , showing that λ is a regular point of µTd . It follows that
ı∗ (λ) is a regular point of µN . Thus, S is a closed submanifold of Cd . We still
need to show that S is compact. As the momentum map µTd : Cd → (Rd )∗ is
manifestly proper, it is enough to show that µTd (S) is bounded.
By the very definition of S, z ∈ S iff h(µTd (z) − λ), ξi = 0, ∀ξ ∈ n. Thus,
µTd (S) = Cd ∩ {x̃ ∈ (Rd )∗ : h(x̃ − λ), ξi = 0, ∀ξ ∈ n}.
Let ∆0 = τ ∗ (∆)+λ be the (compact) image of ∆ under the inclusion τ ∗ composed
with the translation λ, see (14). We claim that µTd (S) = ∆0 . Indeed, for
x̃ = τ ∗ (x) + λ with x ∈ ∆, we have
x̃i = hτ ∗ (x), ei i + λi
(15) = hx, τ (ei )i + λi = hx, ui i + λi
= Li (x) ≥ 0.
As x̃ − λ = τ ∗ (x), we have by the exact sequence (14),
(16) h(x̃ − λ), ξi = 0, ∀ξ ∈ n.
Thus, ∆0 ⊂ µTd (S). For the other inclusion, observe that, again using (14), the
equality (16) tell us that x̃ − λ = τ ∗ (x) for some x ∈ t∗ . Using x̃i ≥ 0 and the
computation in (15), we conclude that x ∈ ∆.
Claim 4. N acts freely on S.
We first determine the stabilizer group of a point z ∈ S, under the action
of Td . It is a subtorus Tz ⊂ Td of dimension equal to the number of vanishing
coordinates z = (z1 , ..., zd ), or equivalently, the number of vanishing coordinates
x̃ = (x̃1 , . . . , x̃d ) where x̃ = µTd (z) ∈ (Rd )∗ is the momentum image of z. By the
argument in the proof of Claim 3, µTd (S) = ∆0 = τ ∗ (∆) + λ and the number
of vanishing coordinates of x̃ = τ ∗ (x) + λ equals the number of vanishing labels
 5. TORIC COMPLEX VARIETIES FROM DELZANT POLYTOPES. FANS 19

Lj at x. Thus, the maximum number is m and it is achieved at the images of

the vertices of ∆. Suppose that x̃ = µTd (z) ∈ ∆0 is a vertex point. Up to a
coordinate permutation, it can be written as√ x̃ = (0, ...0, x̃
√ m+1
, . . . , x̃d ). Then,
the stabilizer of z is the torus Tz = Tm = (e −1t1 , . . . , e −1tm , 1, . . . , 1). Notice
that τ : Tz → T is an isomorphism because of the condition (iii) in Definition 1.5.
Thus,
StabN (z) = StabTd (z) ∩ N = Ker(τ ) ∩ N = {e} ∩ N = {e},
showing that N acts freely as the stabilizer of any point of S is contained in the
stabilizer of some vertex point.
As a final step, we let (M, ω) be the Kähler quotient of (Cd , ω0 , g0 ) at the
momentum value λ, associated to G = Td and N , see Proposition 1.2. The
verification that the momentum image of M under the induced momentum map
is ∆ follows the above discussion.
Corollary 4.1 (Delzant [16]). Any symplectic toric manifold (M, ω, T)
admits an ω-compatible T-invariant Kähler structure (g0 , J0 ).
Exercise 1.4. Show that any toric symplectic manifold (M 2m , ω, T) equipped with a T-invariant ω-
compatible Kähler structure J is a projective variety, i.e. (M 2m , J) admits an holomorphic embedding into a
complex projective space CP N .

Hint. Show that the Dolbeault cohomology H 2,0 (M, J) = {0} and use Hodge decomposition theorem to
conclude that M admits a Kähler structure (ω 0 , J) with [ω 0 ] ∈ H 2 (M, Q). The conclusion then follows from
the Kodaira embedding theorem, see e.g. [26].

5. Toric complex varieties from Delzant polytopes. Fans

Given a symplectic toric manifold (M, ω, T) classified by the Delzant triple
(∆, L, Λ), we can associate a complex manifold M∆ C of dimension m as follows: to
m v v )} and consider the iden-
each vertex v ∈ ∆, we take a copy of Cv = {(z1 , . . . , zm
tification T ∼ m 1 m
= T = (S ) with respect to the lattice base {uv1 , . . . , uvm } given
by the normals of the v-adjacent facets of ∆; we denote this identification by
Tm m m
v and endow the chart Cv with the standard action of Tv , as in Example 1.3.
If w ∈ ∆ is another vertex, we consider the respective bases {uv1 , . . . , uvm } and
{uw1 , · · · , uwm } of Λ (corresponding to the normals of the v-adjacent and w-
adjacent facets of ∆), and let A = (aij ) ∈ SL(m, Z) be the coordinate transition
matrix. We then identify the subset (C∗ )m m ∗ m
v ⊂ Cv with the subset (C )w ⊂ Cw
m

through the identification

(17) zjw = (z1v )aj1 · · · (zm
v ajm
) , j = 1, . . . , m.
It is easily seen that (17) is an equivariant map with respect to the action of Tm
v
on Cm m n C
v and Tw on Cw . This way, one obtains a complex space M∆ endowed with
an affective holomorphic T-action, covered by the affine charts Cmv (parametrized
by the vertices v of ∆) and identified at the intersections Cm ∩ Cm ∼ ∗ m
v w = (C ) as
C ∗ m
explained above. We shal denote by T the induced (C ) -action.
To construct M∆ C explicitly, one can use the Geometric Invariant Theory

(GIT), see [35]: Accordingly, M∆ C := (Cd ) /N C is the space of orbits for

ss
the holomorphic action of the complexified (d − n)-dimensional torus N C ∼ =
∗
(C ) d−n ∗ d d d d
⊂ (C ) (corresponding to N ⊂ T ) on the subset (C )ss ⊂ C of semi-
stable points for the action of N C on Cd , i.e. the points such that the closure
of the N C -orbit does not contain 0 ∈ Cd . Considerations similar to the one in
20 1. DELZANT THEORY

the proof of Claim 4 of Section 4 lead to the introduction of (C∗ )d /N C = ∼ (C∗ )m

equivariant affine charts on (Cd )ss /N C , parametrized by the vertices of ∆ (which
in turn determine sets of (d − n) non-vanishing affine coordinates of Cd on which
N C acts transitively).
Exercise 1.5. Show that if we start with the standard Delzant simplex ∆m ⊂ Rm and the standard
lattice Zm ⊂ (Rm )∗ as in Section 3, the resulting complex manifold M∆C
m
constructed as above is CP m ,
endowed with its atlas of affine charts.

We notice that in order to construct M∆C we did not use the whole data from

(∆, L, Λ). The relevant information is captured by the normals uj = dLj ∈ t

adjacent to the set of vertices, and the combinatorial object which describes it
is the so called Fan associated to ∆.
Definition 1.7 (Fan). Let (∆, L, Λ) be a Delzant triple and P = {F ⊂ ∆}
the poset of closed facets of ∆, partially ordered by the inclusion. The fan
F(∆, L) of (∆, L) is the union of polyhedral cones {CF , F ∈ P} in V ∗ , defined
by
CF = {dL : L(x) ≥ 0 ∀x ∈ ∆, L(x) = 0 ∀x ∈ F }.
The central fact is
Theorem 1.3 (Lerman–Tolman [39]). Suppose J is an ω-compatible, T-
invariant complex structure on the toric manifold (M, ω, T). Then, (M, J) is
C associated to the fan
T-equivariantly biholomorphic to the complex manifold M∆
F(∆, L) of the corresponding Delzant triple (∆, L, Λ).
Proof. The idea of the proof is the following. The effective action of T on
(M, J) complexifies to an effective holomorphic action of a complex algebraic
torus TC = (C∗ )m . The pre-images of the vertices of ∆ are precisely the fixed
points for Tc . At each such fixed point p ∈ M , TC induces a linear action on the
complex space (Tp M, Jp ). The holomorphic slice theorem (see e.g. [47, 33]) tells
us that there exist a TC -invariant neighbourhood 0 ⊂ V ⊂ Tp M , a TC -invariant
neighbourhood p ⊂ U ⊂ M and a Tc -equivariant biholomorphism f : V → U .
Since the action of TC is effective, U must be the whole Tp M , so we obtain
an embedding Tp M into M . This is a TC -invariant affine chart Cm p around p.
The fan associated to this affine chart is a the simplicial cone Cm ∗ dual to the

Delzant image of (Cm , ω0 ). Moreover, if q ∈ M is another fixed point, and Cm q

the corresponding affine chart, then both Cm p and C m must contain the dense
q
principal orbit TC (p0 ) ∼
= (C∗ )m of a pont p0 ∈ M . It is not difficult to see that on
TC (p0 ), the transition map between (C∗ )m ∗ m
p and (C )q is precisely as described
in the construction of M∆ C associated to F = (C ∗ (∆), Λ). 

6. Equivariant blow-up
We explain now how to blow-up equivariantly a fixed point of the action
of T on the complex manifold M∆ C constructed in Section 5. Recall that M C
∆
is endowed with a T-equivariant atlas of affine charts Cm v , parametrized by the
the vertices v of ∆, such that the action T on each chart Cm v is the standard
action of Tm on Cm , as described in Example 1.3. We notice that Cm v = C
m

is the complex manifold associated to (the fan of) the (unbounded) standard
cone Cm ⊂ Rm , and the standard lattice Zm ⊂ (Rm )∗ , via the construction in
Section 5: indeed, Cm has a unique vertex at the origin, and the inward primitive
 6. EQUIVARIANT BLOW-UP 21

normals of the adjacent facets form the standard basis of (Rm )∗ , which define a
single chart Cm
0 .
We now blow-up the origin of Cm and obtain as a resulting manifold C m =
d
0
OCP m−1 (−1) → CP m−1 , the total space of the tautological bundle over CP m−1 .
The blow-down map b : OCP m−1 (−1) → Cm 0 is explicitly given by


b [w1 , . . . , wm ], ζ(w1 , . . . , wm ) = (ζw1 , . . . , ζwm ),
where [w1 , . . . , wm ] are homogeneous coordinates on CP m−1 , and ζ is a fibre-wise
coordinate of the tautological bundle (with respect to the generator (w1 , . . . , wm )).
The inverse map, defined on Cm 0 \ {0}, is
b−1 (z1 , . . . , zm ) = ([z0 , . . . , zm ], (z1 , . . . , zm )),
showing that the action of Tm on Cm 0 lifts to an action of T
m on C
d m , given in
0
the above coordinates by
√ √
(e −1t1 , . . . , e −1tm ) · [w1 , . . . , wm ]; ζ(w1 , . . . , wm ) =


(18) √ √ √ √
[e −1t1 w1 , . . . , e −1tm wm ], ζ(e −1t1 w1 , . . . , e −1tm wm ) ,

thus making b manifestly Tm -equivariant.

Definition 1.8. The (non-compact) manifold C m = O(−1)
CP m−1 endowed
d
0
m
the action of T defined in (18) will be referred to as the equivariant blow up of
(Cm , Tm ).
Introducing affine charts on CP m−1 gives rise to m affine charts Cm m
v1 , . . . , C vm
on C
d m , such that the Tm action (18) becomes the standard action of Tm on each
0
such chart. Furthermore, by writing down the transition between these charts
one sees that C m = O(−1)
CP m−1 becomes the complex manifold associated to
d
0
(the fan of) the unbounded polytope
Ĉm := {L0 (x) = x1 + · · · xm − 1 ≥ 0, Lj (x) = xj ≥ 0, j = 1, . . . , m},
and standard lattice Zm ⊂ (Rm )∗ .

b 2 and C2 .
Figure 1. The polytopes of C 0

We thus get the following

22 1. DELZANT THEORY

Theorem 1.4. Let M∆ C be the compact complex manifold associated to a

Delzant polytope ∆ ⊂ Rm with respect to the standard lattice Zm ⊂ (Rm )∗ . Let

w1 , w2 , . . . , wm be primitive inward-pointing edge vectors at a vertex v of ∆ and
∆ˆ ε the polytope obtained from ∆ by replacing v with the m-vertices v + εwi , i =
1, . . . , m for some ε > 0. Then, ∆ ˆ ε is Delzant polytope too, and the corresponding
complex manifold M∆ C C
ˆ ε is obtained from M∆ by blowing up a fixed point for the
torus action.
Exercise 1.6. Identify the complex manifolds and the equivariant blowdown maps corresponding to the
Delzant polytopes of Figure 2.

Figure 2
.

7. Polarized projective toric varieties

There is a third facet of the story, which relates the smooth compact toric
symplectic manifolds to the theory of polarized projective varieties. Recall that a
smooth polarized projective complex variety is a compact complex manifold M
of complex dimension m, endowed with a very ample holomorphic line bundle
L → M , i.e. a holomorphic line bundle such that the Kodaira map [26]
κ : M −→ P(W ∗ ),
where W = H 0 (M, L) is the (N + 1)-dimensional complex vector space of holo-
morphic sections of L, is a holomorphic embedding and L = κ∗ O(1)W ∗ . In this
case, one can identify (M, L) with its embedded image M̃ = κ(M ) ⊂ P(W ∗ ) ∼ =
CP N , by considering the polarization of M̃ defined by the the restriction of the
anti-tautological line bundle O(1) of CP N . In this setting, we have
Definition 1.9. A toric polarized projective variety M̃ ⊂ CP N is an m-
dimensional complex submanifold of CP N which is the (Zariski) closure in CP N
of a principal orbit for the action of a complex m-dimensional complex torus
TC ⊂ SL(N + 1, C) on CP N .
 7. POLARIZED PROJECTIVE TORIC VARIETIES 23

Denote by T the real m-dimensional torus corresponding to TC , and let

TN ⊂ SL(N + 1, C) be a maximal real torus containing T. As discussed in
Section 3, CP N admits a TN -invariant Fubini–Study Kähler metric ω F S and it
is not difficult to show that 2ω F S is the curvature of a TN -invariant hermitian
metric on O(1). Restricting to M̃ , we obtain a T-invariant symplectic form ω
on M̃ , belonging to the deRham class 2πc1 (L). The T-action is hamiltonian
with respect to this symplectic structure, since T ⊂ TN ⊂ SL(N + 1, C) and the
action of TN is hamiltonian with respect to 2ω F S . We thus have
Proposition 1.3. Any smooth polarized toric complex variety is a symplectic
toric manifold.
From the considerations in Section 5, we get a correspondence between
smooth toric projective varieties and complex toric varieties defined via the
Delzant construction. It is however not entirely clear how this correspondence
translates in terms of Delzant polytopes or, in other words, how to construct a
polarized toric variety from a given Delzant polytope. To explain this, we give
the following
Definition 1.10 (Lattice polytopes). Suppose (∆, L, Λ) is a Delzant poly-
tope as in Definition 1.5. We say that (∆, L, Λ) is a lattice Delzant polytope if,
moreover, all vertices of ∆ belong to the dual lattice Λ∗ ⊂ V .
For any lattice Delzant polytope, we can take a basis {e1 , . . . , em } of Λ and
consider ∆ ⊂ Rm with the standard lattice Λ∗ = Zm ⊂ Rm ; the fact that
(∆, L, Λ) is a lattice Delzant polytope translates to the fact that the vertices of
∆ are in Zm (see Exercise 1.2). We denote by
(j)
A(∆) := {λ(j) = (λ1 , . . . , λ(j)
m ), j = 0, . . . , N }
all the lattice points in ∆. We then consider the (C∗ )m -action on CP N , given
in homogeneous coordinates by
(0) (0) (N ) (N )
λ λm λ λm
(z1 , . . . , zm ) · [w0 , . . . , wN ] := [(z1 1 · · · zm )w0 , . . . , (z1 1 · · · zm )wN ].
We associate to such a data the toric polarized variety MA(∆) ⊂ CP N which is
the Zariski closure in CP N of the (C∗ )m -orbit of the point [1, . . . , 1] ∈ CP N for
this action. The basic fact is the following
Theorem 1.5 (Section 6.6 in [9]). For any lattice Delzant polytope ∆,
MA(∆) ⊂ CP N is a smooth polarized toric projective variety whose Delzant poly-
tope with respect to a Tm -invariant Fubini–Study symplectic form in 2πc1 (O(1))
is ∆ ⊂ Rm with lattice Λ∗ = Zm . In particular, MA(∆) is biholomorphic to the
complex manifold M∆ C defined in Section 5.

Example 1.8. Let ∆ ⊂ Rm be the lattice Delzant simplex, defined by

X
∆ = {L0 = (1 − xj ) ≥ 0, Lj (x) = xj ≥ 0, j = 1, . . . m}
j=1n

and the standard lattice Zm ⊂ (Rm ). ∆. The set of lattice points of ∆ is

(0) (j)
A(∆) = {λ = (0, . . . , 0), λ = ej , j = 1, . . . , m},
m
where ej is the standard basis of R . The Delzant construction identifies the corresponding symplectic
toric manifold with (CP m , 2ω F S ), whereas Theorem 1.5 realizes MA(∆) as the closure in CP m of a principal
(C∗ )m -orbit for the action
(z1 , . . . , zm ) · [w0 , . . . , wm+1 ] = [w0 , z1 w1 , . . . , zm wm ],
which clearly is CP m . The induced symplectic structure on this polarized variety is again 2ω F S .
24 1. DELZANT THEORY

Exercise 1.7. Let MA(∆) ⊂ CP N be a toric polarized projective variety corresponding to a lattice
Delzant polytope ∆ ⊂ Rm with respect to the lattice Zm ⊂ Rm . Using that the sections of O(1) are identified
with linear functions in homogeneous coordinates [w0 , . . . , wN ] on CP N , show that the (C∗ )m -action on
MA(∆) defines a (C∗ )m -action on the vector space H 0 (MA(∆) , L) of holomorphic sections of L. Furthermore,
show that here is a basis {s0 , . . . , sN } of H 0 (MA(∆) , L), parametrized by the lattice points {λ(0) , . . . , λ(N ) }
in A(∆), such that (C∗ )m acts on sj by
(j) (j)
λ λ
(z1 , . . . , zm ) · sj = (z1 1 · · · zmm )sj .

8. Toric orbifolds
We briefly discuss here what happens with the Delzant construction reviewed
in Section 4 when one starts with a rational Delzant triple (∆, L, Λ), i.e satisfying
the weaker conditions (i)-(ii)-(iii)’ of Definition 1.5. There are two points in the
construction which deserve a special attention.
The first point is Claim 1. Consider for example the labelledPm Delzant simplex
1
(∆m , L) with L = {Lj (x) = xj , j = 1, . . . m, Lm+1 (x) = − j=1 xj + 2 }, corre-
sponding to (CP m , ω F S ), and change the standard lattice Λ = Zm ⊂ (Rm )∗ with
the lattice Λ0 = 12 Λ. The condition (iii)’ then holds for the triple (∆m , L, Λ0 ).
However, in this case,
N 0 = Ker(τ ) = S1 × Zm2 ,
m
where Z2 := (±1, ±1, . . . , ±1) ⊂ T m+1 . In particular, N 0 is no longer connected.
Following the construction further, the quotient space will be
M 0 = S2m+1 /N 0 = CP m /Zm
2 .
Such a space is an example of an orbifold.
Definition 1.11. An orbifold chart on a topological space M is a triple
(Û , Γ, ϕ) where:
• Û ⊂ Rn is an open subset;
• Γ ⊂ GL(n) is a finite subgroup acting on Û
• ϕ : Û /Γ → M is a homeomorphism between Û /Γ and an open subset
V ⊂ M.
We denote by ϕ̂ : Û → M the induced Γ-invariant map.
An embedding between two orbifold charts (Û1 , Γ1 , ϕ̂1 ) and (Û2 , Γ2 , ϕ̂2 ) is a
smooth embedding λ : Û1 → Û2 such that ϕ̂1 = ϕ̂2 ◦ λ.
Two orbifold charts (Û1 , Γ1 , ϕ̂1 ) and (Û2 , Γ2 , ϕ̂2 ) are compatible if there exists
an open subset V ⊂ V1 ∩ V2 (where we have set Vi = ϕ̂i (Ûi )) and an orbifold
chart (Û , Γ, ϕ̂) with ϕ̂(Û ) = V , and two embeddings λi : Û → Ûi , i = 1, 2.
An orbifold M of (real) dimension n is a Hausdorff paracompact topo-
logical space, endowed with an open covering of compatible orbifold charts
{(Ûi , Γi , ϕ̂i )}i∈I . A smooth function f : M → R on an orbifold M is defined
by the property that on each orbifold chart (Û , Γ, ϕ̂), it pulls-back by ϕ̂ to a
Γ-invariant smooth function on Û . One can define smooth tensors on M is a
similar way.
The above phenomenon can be remedy by taking Λmin = spanZ {u1 , . . . ud },
which is the minimal (under the inclusion) lattice for which the condition (iii)’
is satisfied for the given labelling L. The subgroup Nmin will be then a torus
and for any other choice of lattice Λ, the produced quotient space S/N will be
a quotient of S/Nmin by the finite abelian group Λ/Λmin . Geometrically, this
 8. TORIC ORBIFOLDS 25

is translated to taking orbifold coverings, S/Nmin being the simply connected

orbifold covering all other quotients, see Thurston [50].
The second point is Claim 4. When we assume (iii)’ instead of (iii), the
stabilizers for the Td -action on S will be finite abelian groups, and in general
so be the stabilizers of the action of N . However, once again the quotient space
S/N will be an orbifold in the sense of the above definition.
To summarize, we observed that
Proposition 1.4. Suppose (∆, L, Λ) is a rational Delzant triple, i.e. satis-
fies the conditions (i)-(ii)-(iii)’ of Definition 1.5. Then the Delzant construction
associates to (∆, L, Λ) a compact symplectic orbifold (M, ω) endowed with a
Hamiltonian action of the torus T = V /2πΛ and momentum image ∆.
The converse is also true, due to the following extension of Delzant’s corre-
spondence to toric orbifolds
Theorem 1.6 (Lerman–Tolman [39]). Compact symplectic orbifolds mod-
ulo equivalence are in bijective correspondence with rational Delzant triples mod-
ulo the action of the affine group.
We give below one specific example
Example 1.9 (The weighted projective spaces). Similarly to the way we introduced the com-
plex projective space, we can consider for any (m + 1)-tuple of positive integers a = (a0 , . . . , am ) with
g.c.d(a0 , . . . , am ) = 1 the quotient space
m m+1 ∗
CPa = C \ {0}/Ca ,
where C∗
a
∗
denotes the C action ρa on C m+1
= (z0 , . . . , zm ) by
a0 a
ρa (λ)(z) := (λ z0 , . . . , λ m zm ).
Lemma 1.4. CPam is an orbifold, which is diffeomorphic (as an orbifold) to the quotient space
2m+1 1
S /Sa
where S1a stands for the circle action ρa on S2m+1 , given by
√ √ √
−1t −1a0 t −1am t
ρa (e )(z) = (e ,...,e ).
The proof of this result is left as an exercise. We can now easily modify the construction of the toric
structure on CP m and get the following
Lemma 1.5. The orbifold CPam admits a toric Kähler structure (g, ω, J) obtained by the Delzant con-
struction starting with the rational Delzant triple (∆m , La , Λa ) where ∆m ⊂ Rm is the standard simplex
labelled as
n a ···a 
0 m
∆m := (x1 , . . . , xm ) : Lj (x) = xj ≥ 0, j = 1, . . . , m,
aj
 a · · · a  1 m  o
0 m
X
Lm+1 (x) = − xj ≥ 0 ,
a0 2 j=1

and the lattice Λa is defined by

n a · · · a  a ···a  o
0 m ∗ 0 m ∗ ∗
Λa := spanZ ej , j = 1, . . . , m, (e1 + · · · + em )
aj a0
where {e∗ m m ∗
j }j=1 stands for the standard basis of (R ) .
 CHAPTER 2

Abreu–Guillemin Theory

1. The orbit space of a toric manifold

Let (M, ω, T) be a toric symplectic manifold and µ : M → ∆ the correspond-
ing Delzant polytope. We denote by
Mred := M/T
the space of orbits for the T action on M , which is a compact Hausdorff space
with respect to the quotient topology, see e.g. [5]. The momentum map µ
being T-invariant descends to a continuous map µ̌ : Mred → ∆. Some important
ingredients in Delzant’s theorem are the following
Fact 1. µ̌ is a bijection.
Fact 2. Delzant’s proof [16] shows that the pre-image p = µ−1 (x) of a point
x ∈ ∆ situated on an open face F 0 ⊂ ∆ of co-dimension ` has a stabilizer Tp
which is a torus of dimension `. In particular, the pre-image of a point x ∈ ∆0
is a principal orbit isomorphic to T. We shall denote M 0 := µ−1 (∆0 ) the dense
open subset of M consisting of points having principal orbits. As T acts freely
on M 0 , µ : M 0 → ∆0 is a principal T bundle over ∆0 .
Another fact coming from [16] is
Fact 3. Mred and ∆ both admit “differentiable structure” of manifolds with
corners induced from the smooth structure of M , and µ̌ is a diffeomorphism in
this category, ([34], Proposition C7).
Exercise 2.1. Check Facts 1–3 for the symplectic manifold (M, ω) associated to a Delzant polytope
(∆, L, Λ) via the construction in Section 4.

The differentiable structure on ∆ ⊂ t∗ is the one naturally induced by re-

stricting to ∆ the smooth functions on t∗ . To see how it is related to the smooth
structure on M , we notice the following
Lemma 2.1. (Schwarz [48]) A T-invariant function f (p) on M is smooth
iff
f (p) = ϕ(µ(p))
for some smooth function ϕ(x) on t∗ .
In the case when (M, ω, T) = (R2 , ω0 , S1 ), this is a well-know result by
Whitney; in general, Lemma 2.1 can be easily derived from ([16], Lemme 5).
Because of this basic fact, we shall often write f (x) for the composition f (p) =
ϕ(µ(p)).

2. Toric Kähler metrics: local theory

To simplify the presentation, we fix a basis {e1 , . . . , em } of t and denote by
Kj = Xej the induced fundamental vector fields; we shall also identify t∗ ∼ =
27
28 2. ABREU–GUILLEMIN THEORY

(Rm )∗ by using the dual basis {e∗1 , . . . , e∗m } and write x = (x1 , . . . , xm ) for the
elements of t∗ . As explained in the previous section, on M 0 , K1 , . . . , Km are
functionally independent, i.e. for each point p ∈ M 0 , (K1 ∧ · · · ∧ Km )(p) 6= 0.
We shall also identify the coordinate function xi = hx, ei i with the momentum
function hµ, ei i on M , i.e. we write
(19) ıKi ω = −dxi .
Let (g, J) be a T-invariant ω-compatible Kähler structure on M . Letting
Hij = g(Ki , Kj )
we get a T-invariant smooth function on M , which will tacitly identify with a
smooth function Hij (x) on ∆, see Fact 2 above. We denote by Hx = (Hij (x))
the corresponding symmetric matrix with smooth entries over ∆. In a more
intrinsic language, we regard H as a smooth function over ∆ with values in S 2 t∗
by letting
Hx (ξ1 , ξ2 ) := gp (Xξ1 , Xξ2 )
for any ξ1 , ξ2 ∈ t and any p ∈ µ−1 (x).
On ∆0 , H is positive definite and we denote by G := H−1 the inverse matrix;
equivalently, G is a smooth S 2 t-valued function on ∆0 .
Notice that, by (19),
(20) dxi (JKj ) = −ω(Ki , JKj ) = − − g(JKi , JKj ) = −g(Ki , Kj ) = −Hij (x).
We now consider the vector fields {K1 , . . . , Km , JK1 , . . . , JKm }. They form
a basis of Tp M at each p ∈ M 0 (because {K1 , . . . , Km } span an m-dimensional
space and {JK1 , . . . , JKm } span its g-orthogonal complement) and satisfy
(21) [Ki , Kj ] = [Ki , JKj ] = [JKi , JKj ] = 0,
where for the second identity we used that T preserves J, whereas for the third
identity we used the integrability of J, see (8). We now denote by
{θ1 , . . . , θm , Jθ1 , . . . , Jθm }
the dual basis of T ∗ M 0 , corresponding to
{K1 , . . . , Km , JK1 , . . . , JKm },
where for a 1-form θ we set Jθ(X) = −θ(JX), for any vector field X. The
commuting relation (21) is equivalent to
(22) dθi = 0 = d(Jθi ), i = 1, . . . m.
As the 1-forms Jθi satisfy
ıKj Jθi = 0, LKj Jθi = 0,
in terms of the T bundle structure µ : M 0 → ∆0, each Jθi is basic, i.e. Jθi =
µ∗ (αi ) of a closed 1-form αi on ∆0 . As ∆0 is contractible, we can write
(23) Jθi = −dyi
for some smooth function yi (x), defined on ∆0 up to an additive constant (as
usual, we omit the pull-back by µ in the notation). Furthermore, by (20), we
 2. TORIC KÄHLER METRICS: LOCAL THEORY 29

find
m
X
−Jθi = dyi = Gij (x)dxj ,
j=1
(24) m
X
Jdxi = Hij (x)θj .
j=1

The 1-forms {θ1 , . . . , θm } on the other hand define a 1-form with values
t = Lie(T) by letting
m
X
(25) θ= θi ⊗ ei ,
i=1

whose curvature is identically zero. The symplectic 2-form ω then becomes

m
X
(26) ω= dxi ∧ θi = hdµ ∧ θi
i=1

whereas the Kähler metric is

m
X  
g= g(Ki , Kj ) θi ⊗ θj + Jθi ⊗ Jθj
i,j=1
Xm  
= Hij θi ⊗ θj + dyi ⊗ dyj
(27) i,j=1
Xm  
= Gij dxi ⊗ dxj + Hij θi ⊗ θj
i,j=1
= hdµ, G, dµi + hθ, H, θi.

Lemma 2.2 (Guillemin [28]). Let (M, ω, T) be a symplectic toric manifold

with Delzant polytope ∆ and (g, J) an ω-compatible T-invariant Kähler structure.
2u
Then, on M 0 , (g, ω) can be written in the form (26)–(27), where Gij = ∂x∂i ∂x j
for a smooth, strictly convex function u(x) on ∆0 , called symplectic potential
of g. Conversely, for any strictly-convex smooth function u on ∆0 , the riemann-
ian metric on M 0 defined by (27) with G = Hess(u), H = G−1 and the flat
connection 1-form θ defines an ω-compatible, T-invariant Kähler structure on
M 0.
∂f
Proof. We shall denote with subscript f,k the partial derivative ∂xk of a
smooth function f on ∆. Letting
m
X
β := yi dxi ,
i=1

we have by (24)
m
X m
X
dβ = dyi ∧ dxi = Gij dxi ∧ dxj = 0.
i=1 i,j=1
30 2. ABREU–GUILLEMIN THEORY

By the Poincaré Lemma on ∆0 , β = du for some smooth function u on ∆0 , i.e.

yi = u,i . It follows that
(28) Gij = yi,j = u,ij = (Hess(u))ij
Conversely, for the metric (27), we let Gij = g(Ki , Kj ) and suppose that
(28) holds true. It follows that the 1-forms
m
X m
X
−Jθi = Gij dxj = u,ij dxj
j=1 j=1

are closed. As θi are closed too (by the assumption √ that θ is flat), we get
a basis of Λ 1,0 (M 0 , J) of closed (1, 0)-forms −Jθ + −1θi , so writing
√ √ i √ locally
−Jθi + −1θi = dyi + −1dti we get holomorphic coordinates yi + −1ti for
J, i.e. J is integrable. 
We have seen in Chapter 1 that the action of T ∼ = (S1 )m on (M, J) extends
to an (effective) holomorphic action of the complex torus TC ∼ = (C∗ )m . Fixing
a point p0 ∈ M , we can identify M with the orbit T (p0 ) ∼
0 0 C
= (C∗ )m . Using the
polar coordinates (ri , t0i ) on each C∗ , this identification gives rise the so-called
angular coordinates
t = (t01 , . . . , t0m ) : M 0 → t/2πΛ.
If another reference point is chosen, t varies by an additive constant in t. Writing
θ = dt, we have
Definition 2.1. For a fixed ω-compatible, T-invariant complex structure J
on (M, ω, T) (corresponding to a symplectic potential P u(x) on ∆0 ), and a base
point p0 ∈ M (giving rise to angular coordinates t = m
0
i=1 ti ei with respect to
a basis e = {e1 , . . . , em } of t), the functions {x1 , . . . , xm ; t1 , . . . , tm } on ∆0 × T
are called momentum-angle coordinates associated to (g, J).

3. The scalar curvature

Recall that if (M 2m , J, g, ω) is a Kähler manifold, the riemannian metric g
induces a hermitian metric (still denoted by g) on the anti-canonical complex
line bundle K −1 (M ) = ∧m (T 1,0 M ), where T M ⊗ C = T 1,0 M √ ⊕ T 0,1 M is√the
type decomposition of the complexified tangent bundle of M into −1 and − −1
eigen-spaces of J. Furthermore, K −1 (M ) has a canonical holomorphic structure,
induced by the holomorphic structure of T 1,0 M (for which the holomorphic
vector fields are the holomorphic sections). The Ricci-form ρg of g is defined in
terms of the curvature of the Chern connection ∇g of (K −1 (M ), g), by
∇g
√
RX,Y = −1ρg (X, Y ).
In the Kähler case, ∇g coincides with the induced connection on K −1 (M ) by
the Levi–Civita connection of g, and the 2-form ρg is essentially the Ricci tensor
Ricg of g, i.e. we have
ρg (X, Y ) = Ricg (JX, Y ).
It thus follows that the scalar curvature sg := trg (Ricg ) of g is equivalently given
by
 
(29) sg = trω (ρg ) := 2m (ρg ∧ ω m−1 )/ω m .
 3. THE SCALAR CURVATURE 31

We now consider a T-invariant, ω-compatible Käher metric on a symplectic

toric manifold (M, ω, T), written on M 0 in the form (27) where G = Hess(u) for
a symplectic potential u ∈ S(∆, L).

Lemma 2.3 (Abreu [1]). The Ricci form of (g, J) is given by

m
1 X
ρg = − Hij,ik dxk ∧ θj
2
i,j,k=1

whereas the scalar curvature is

m
X
sg = − Hij,ij ,
i,j=1

where we recall that for u ∈ S(∆, L), Hij = (Hess(u))−1 ,ij

ij = u .

Proof. As the fundamental vector fields Ki preserve J,

√ √
σ := (K1 − −1JK1 ) ∧ · · · ∧ (Km − −1JKm )

is a holomorphic section on K −1 (M ) which does not vanish on M 0 . It is a

well-known fact (see e.g. [56]) that the (Chern) Ricci form is then given on M 0
by
1
ρg = − ddc log |σ|2g .
2
In our case, |σ|2g = 2m det(g(Ki , Kj )) = det H, so we obtain
  X
c −1 c
d log det H = tr H d H = Gij Hij,k Jdxk
i,j,k
X
= Gij Hij,k Hk` θ`
i,j,k,`
X
=− Gij,k Hij Hk` θ`
i,j,k,`
X
=− Gik,j Hij Hk` θ`
i,j,k,`
X
= Gik Hij Hk`,j θ`
i,j,k,`
X
= Hj`,j θ` .
j,`

We then compute
1
ρg = − ddc log det H
2
(30) 1X
=− Hij,ik dxk ∧ θj .
2
i,j,k
P
The formula for sg follows from (29), by using that ω = i dxi ∧ θi and (30). 
32 2. ABREU–GUILLEMIN THEORY

4. Symplectic versus complex: the Legendre transform

We now turn our attention to the function yj = u,j on M 0 introduced √ in
(24). √
For a choice of angular √ coordinates θj = dtj we have −Jθj + −1θj =
dyj + −1dtj and yj + −1tj define J-holomorphic coordinates on M√0 ; expo-
nentiating, we obtain another set of holomorphic coordinates zj := eyj e −1tj on
M 0.
Now, choosing {e1 , . . . , em } to be a basis of Λ, and considering the action
of the flows of {K1 , . . . , Km ; JK1 , . . . , JKm } around a reference point p0 ∈ M 0
corresponding to x0 ∈ ∆0 under the momentum map, we see that (z1 , . . . , zm )
provide an TC equivariant identification

(31) Φp0 ,u,e : (C∗ )m ∼

= (M 0 , J),

provided that we have arranged, by adding an affine linear function to u, that u

atteints its minimum at x0 . A subtle point in the construction, however, is that
Φp0 ,u,e depends upon the choice of u, x0 and the basis e = {e1 , . . . , em } of t.

Definition 2.2 (Legendre transform). Let u be a strictly convex smooth

function on ∆0 . Letting yj (x) := u,j (x) and
m
yi (x)ei = (du)x ∈ (Tx ∆0 )∗ ∼
X
y(x) := = (t∗ )∗ = t
i=1

be the map from ∆0 to t, we define the Legendre transform of u(x) to be the

function φ(y) = φ(y1 , . . . , ym ) satisfying

(32) φ(y(x)) + u(x) = hy(x), xi,

Pm ∗
where x = i=1 xi ei is seen as a smooth function from ∆0 to t∗ .

Lemma 2.4 (Guillemin [28]). Let (g, J) be an ω-compatible, T-invariant

Kähler structure on (M, ω, T) with symplectic potential u(x) defined on ∆0 .
Then, the Legendre transform φ(y) of u(x) is a Kähler potential of the sym-
plectic form ω on M 0 , i.e. satisfies

ω = ddc φ,

where dc φ = Jdφ.

Proof. By its very definition,

m
X m
X
φ(y(x)) = yi xi − u(x) = xi u,i − u(x),
i=1 i=1

so that
m 
X  m
X m
X
dφ = d(xi u,i ) − u,i dxi = xi u,ij dxj = xi Gij dxj ,
i=1 i,j=1 i,j=1
 5. THE CANONICAL TORIC KÄHLER METRIC 33
m
X
c
dd φ = dJdφ = d xi Gij Jdxj
i,j=1
Xm
=d xi Gij Hjk θk
i,j,k=1
m
X
= dxi ∧ θi = ω
i=1


5. The canonical toric Kähler metric

We now compute the symplectic potential of the Kähler metric (g0 , J0 ) in-
duced by the flat Kähler structure (g̃0 , J˜0 ) on Cd via the Delzant construction,
see Section 4. We use the general form (27) of a T-invariant, ω-compatible
Kähler structure and adopt the following
Definition 2.3. The reduced metric gred associated to g written as (27) on
the fibration µ : M 0 → ∆0 is
m
X
gred = Gij dxi ⊗ dxj .
i,j=1

Geometrically, gred is the unique metric on ∆0 such that

µ : (M 0 , g) → ∆0
is a riemannian submersion.
Lemma 2.5 (Calderbank–David–Gauduchon [7]). In the setting of Propo-
sition 1.2 suppose, moreover, that G acts freely on M̃ and G/N acts freely on
M . Denote respectively by g̃red the reduced metric on M̃ ∼
= Imµg ⊂ g∗ and by
gred the reduced metric on M/(G/N ) ∼ ∗
= Imµ ⊂ (g/n) . Then,
gred = `∗ g̃red ,
see (11).
Proof. Let q ∈ M and p ∈ S with q = π(p). We decompose, orthogonally
with respect to g̃p , Tp S = Tp G(p)⊕(Tp G(p))⊥ and observe that, by the definition
of reduced metric, for any X̃ ∈ (Tp G(p))⊥ , |X̃|g̃ = |(µG )∗ X̃|g̃red . Let X :=
π∗ (X̃). We have, by the definition of g, |X|g = |X̃|g̃ and as X is orthogonal to
the tangent space at q of G/N (q), we deduce |µ∗ (X)|gred = |(µG )∗ X̃|g̃red . The
claim follows. 
We turn again to Example 1.6.
Example 2.1 (The symplectic potential of the flat Kähler struc-
ture). Let (Cd , g̃0 , J˜0 , ω̃0 ) be the flat Kähler structure of Cd . Then, the reduced
metric is written on the interior of the cone Cd0 = {x̃i > 0} as
d
1 X dx̃i ⊗ dx̃i
(g̃0 )red = .
2 x̃i
i=1
34 2. ABREU–GUILLEMIN THEORY

To see this, recall that the flat metric g̃0 on R2d can be written in polar coordi-
nates (ri , ti ), i = 1, . . . , d as
d
X
g̃0 = dri2 + ri2 dt2i .
i=1

We have already observed in Example 1.6 that the momentum coordinates are
x̃i = ri2 /2.
Theorem 2.1 (Guillemin [28]). The symplectic potential of the induced
Kähler structure (g0 , J0 ) via the Delzant construction is
d
1X
u0 (x) = Lj log Lj .
2
j=1

Proof. We have already observed in Section 4 that the map ` = j + c̃ in

(11) is given by yj = Lj (x). It follows from Lemma 2.5 and Example 2.1 above
that
d
1 X dLj ⊗ dLj
gred = .
2 Lj
j=1
The claimed expression for the symplectic potential follows easily. 

6. Toric Kähler metrics: compactification

We now turn to global questions. To this end we suppose that (g0 , J0 ) is
a globally defined, T-invariant, ω-compatible Kähler metric on a compact toric
symplectic manifold (M, ω). For instance, we can take the metric provided by
Corollary 4.1. We denote by G0 = Hess(u0 ) and θ 0 the hessian of the symplectic
potential and the angular coordinates of this fixed reference metric. We can take
for instance u0 and θ 0 to be symplectic potential and connection 1-form of the
Guillemin Kähler metric, see Theorem 2.1.
Lemma 2.6. Suppose g is an invariant Kähler structure, given on M 0 by
(27), where θ = θ 0 are the angular coordinates of g0 . Then g extends to a
Kähler structure on M provided that
(33) G − G0 is smooth on ∆,

(34) G0 G−1 G0 − G0 is smooth on ∆

Proof. The key point is that it suffices to show that g extends as smooth
tensor on M : Indeed, g will be then non-degenerate as the endomorphism it
defines with respect to ω will be, by continuity, an almost-complex structure J
everywhere; the latter will be integrable everywhere by continuity too. For the
smoothness of g, we compute (on M 0 ):
g − g0 = hdµ, G − G0 , dµi + hθ 0 , H − H0 , θ 0 i
= hdµ, G − G0 , dµi + hJ0 dµ, G0 (H − H0 )G0 , J0 dµi
= hdµ, G − G0 , dµi + hJ0 dµ, G0 HG0 − G0 , J0 dµi.
The claim follows. 
 6. TORIC KÄHLER METRICS: COMPACTIFICATION 35

Exercise 2.2 (Abreu’s boundary conditions [2]). Show that (33) and (34) are equivalent to
(35) G − G0 is smooth on ∆,

−1
(36) G0 G is smooth and nondegenerate on ∆.
Exercise 2.3. [2] Show that the conditions (35)-(36) are equivalent to
d
1X
(37) u− Lj log Lj is smooth on ∆.
2 j=1

d
Y 
(38) det(Hess(u)) × Lj (x) is positive and smooth on ∆.
j=1

There is, however, a subtle point in the above theory (which I believe is often
neglected in the literature): in order to apply the sufficient conditions (33)-(34)
or (35)-(36), we need to use the angular coordinates defined by the initial metric
g0 . This is the main difficulty to show that the conditions are also necessary.
The following criterion is established in [3].
Proposition 2.1. A smooth, positive definite S 2 (t∗ )-valued function H on
∆0 corresponds via (27) to a T-invariant ω-compatible almost-Kähler structure
on M if and only if H satisfies the following conditions.
• [smoothness] H has a smooth extension as a S 2 (t∗ )-valued function on
∆;
• [boundary conditions] If x belongs to a co-dimension one face Fj ⊂ ∆
with normal uj ∈ t, then
Hx (uj , ·) = 0, dHx (uj , uj ) = 2uj .
• [positivity] Let F 0 ⊂ ∆ be the interior of a face of ∆ and denote by
tF ⊂ t the subspace spanned by the normals of all labels vanishing on F 0 .
Then, the restriction of H to F 0 , viewed as a S 2 (t0F )-valued function
for t0F = ann(tF ) ⊂ t∗ is positive definite.
Exercise 2.4. State and prove Proposition 2.1 in the case m = 1.

Lemma 2.7. Suppose g, g 0 are two T-invariant, ω-compatible Kähler struc-

tures on a compact toric symplectic manifold (M, ω, T) which induce the same
S 2 (t)-valued function G on ∆0 . Then g and g 0 are isometric under a T-equivariant
symplectomorphism.
Proof. Fix p0 ∈ M 0 and let t (resp. t0 ) be the angular coordinates de-
termined by g (resp. g 0 ). The map Ψ0 which sends t to t0 and leaves x un-
changed defines a T-equivariant symplectomorphism on M 0 , which sends g to
g 0 . As (M, g) is complete, we can extend Ψ0 to an isometry of the metric spaces
(M, g) and (M, g 0 ), so to a riemannian isometry by a well-known result (see e.g.
[36]). 
Definition 2.4. For any compact convex labelled polytope (∆, L), we de-
note by S(∆, L) the space of smooth strictly convex functions u defined on the
interior of ∆0 , such that H = Hess(u)−1 satisfies the conditions of Proposi-
tion 2.1, or equivalently, the conditions (35)-(36).
Combining Lemma 2.6, Exercise 2.2, Proposition 2.1 and Lemma 2.7, we
obtain the following key result
36 2. ABREU–GUILLEMIN THEORY

Theorem 2.2 (Abreu [1, 2]). The space of T-invariant, ω-compatible Kähler
structures (g, J) on (M, ω, T), modulo the action of the group of T-equivariant
symplectomorphisms, is parametrized by the space S(∆, L).
One can further refine Theorem 2.2.
Proposition 2.2 (Donaldson [22]). The functional space S(∆, L) of sym-
plectic potentials of globally defined T-invariant, ω-compatible Kähler metrics on
(M, ω, T) can be equivalently defined as the sub-space of the space C(∆) of convex
continious functions on ∆, such that
• [convexity] The restriction of u to the interior of any face of ∆ (includ-
ing ∆0 ) is a smooth strictly convex function.
• [asymptotic behaviour] u − 21 dj=1 Lj log Lj is smooth on ∆.
P
 CHAPTER 3

The Calabi Problem and Donaldson’s theory

1. The Calabi problem on a toric manifold

In [6], Calabi introduced the notion of an extremal Kähler metric (g, ω) on
a complex manifold (M, J)
Definition 3.1 (Calabi [6]). A Kähler metric (g, ω) on a complex mani-
fold (M, J) is called extremal if the ω hamiltonian vector field Xg := ω −1 (dsg )
preserves the complex structure J, i.e. LXg J = 0.
Constant scalar curvature (CSC) Kähler metrics are examples, and, moti-
vated by the Uniformization Theorem for complex curves, Calabi proposed the
problem of finding an extremal Kähler metric (g, ω) in a given deRham class
[ω] ∈ H 2 (M, R). This is known as the Calabi Problem, and is of greatest interest
in current research in Kähler geometry.
We now turn to the toric case, and observe the following ramification of the
Calabi problem.
Lemma 3.1. Suppose (g, J) is a T-invariant, ω-compatible Kähler metric
on (M, ω, T), corresponding to a symplectic potential u ∈ S(∆, L). Then g is
extremal iff sg is the pull back by the momentum map of an affine function
s(x) = hξ, xi + λ on ∆.
Proof. As g is T invariant, its scalar curvature is a T-invariant function on
M , whence is the pull back by the momentum map of a smooth P function s(x)
−1
on ∆. It thus follows from (26) that on M , Xg = ω (ds) = i s,i Ki . As each
Ki preserves J, the condition LXg J = 0 reads as
X X
0 = JKj · s,i = (ds,i )(JKj ) = − s,ik Jdxk (Kj ) = − s,ik Hk` θ` (Kj )
i,k i,k,`
X
=− s,ik Hkj
i,k

As H is non-degenerate on ∆0 , it follows that s,ik = 0, i.e. s(x) must be an

affine-linear function on ∆0 , and hence on ∆.
Conversely, for any affine linear function s(x) = hξ, xi + λ, ω −1 ds = Xξ ,
which preserves J. 
A key observation of Donaldson [21] is that the affine-linear functions s(x)
in Lemma 3.1 is in fact predetermined from the labelled polytope (∆, L). To
state the precise result, we need to introduce measures on ∆ and ∂∆. The
standard Lebesgue measure dv = dx1 ∧ . . . ∧ dxm on t∗ ∼ = Rm and the labels
L = (L1 , . . . , Ld ) of ∆ induce a measure dσ on each facet Fi ⊂ ∂∆ by letting
(39) dLi ∧ dσ = ui ∧ dσ = −dv.
37
38 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

Proposition 3.1 (Donaldson [21]). Suppose (∆, L) is a compact convex

simple labelled polytope in t∗ . Then, there exists a unique affine-linear function
s(∆,L) on t∗ , called the extremal affine function of (∆, L) such that for any affine
linear function f
Z Z
(40) 2 f dσ − s(∆,L) f dv = 0.
∂∆ ∆
Furthermore, if for u ∈ S(∆, L) the metric (27) is extremal, i.e. satisfies
Xm
(Hess(u))−1 ij,ij = s(x) = hξ, xi + λ,


−
i,j=1

then the affine-linear function s(x) must be equal to s(∆,L) .

Before we give the proof of this important result, we start with a technical
observation.
Lemma 3.2. Let H be any smooth S 2 t∗ -valued function on ∆ which satisfies
the boundary conditions of Proposition 2.1, but not necessarily the positivity
condition. Then, for any smooth function ϕ on ∆
Z X m  Z X m  Z
(41) Hij,ij ϕ dv = Hij ϕ,ij dv − 2 ϕdσ.
∆ i,j=1 ∆ i,j=1 ∂∆

Proof. The proof is elementary and uses integration by parts: recall that
for any smooth t∗ -valued function V = (V1 , . . . , Vm ) on t∗ , Stokes theorem gives
Z X m d Z
X
(42) Vj,j dv = − hV, dLk idσ,
∆ j=1 k=1 Fk

where we have used the convention (39) for dσ. We shall use the identity
Xm m
X m
X
(43) ϕ,ij Hij = ϕHij,ij − Vj,j ,
i,j=1 i,j=1 j=1
where
m
X m
X
(44) Vj := ϕ Hij,i − ϕ,i Hij .
i=1 i=1
It follows by (43) and (42) that
Z X m Z m
X d Z
X
(45) ϕ,ij Hij dv = ϕHij,ij + hV, dLk idσ.
∆ i,j=1 ∆ i,j=1 k=1 Fk

On each facet Fk we have, using a basis {e1 , . . . em } of t,

m
X
hdLk , V i = hdLk , e∗j iVj
j=1
Xm m
X m
X
(46) = hdLk , e∗j i(ϕ Hij,i − ϕ,i Hij )
j=1 i=1 i=1
m
X m
X
=ϕ hdH(dLk , ei ), e∗i i − H(dLk , dϕ).
i=1 i=1
 2. DONALDSON–FUTAKI INVARIANT 39

Using the boundary conditions of Proposition 2.1, we have H(dLk , ·) = 0 on Fk

and (by choosing a basis e1 = dLk , e2 , . . . , em with e∗j tangent to Fk for j > 1)
m
X
hdH(dLk , ei ), e∗i i = 2.
i=1

Substituting back in (46) and (45) completes the proof. 

Proof of Proposition 3.1. Writing

m
X
s(∆,L) = a0 + aj xj ,
j=1

the condition (40) gives rise to a linear system with positive-definite symmetric
matrix
Z Xm Z Z
a0 xi dv + aj xj xi dv = 2 xi dσ
∆ j=1 ∆ ∂∆
(47) Z m Z Z
X
a0 dv + aj xj dv =2 dσ,
∆ j=1 ∆ ∂∆

which therefore determines (a0 , . . . , am ) uniquely.

We now suppose H = (Hij ) is a smooth S 2 (t∗ )-valued function on ∆ which
satisfies the boundary conditions of Proposition 2.1 and
m
X
(48) − Hij,ij = s(x) = hξ, xi + λ.
i,j=1

We are going to show that s(x) = s(∆,L) . Notice that to this end we do not
assume that H satisfies the positivity conditions of Proposition 2.1 nor that
H = Hess(u)−1 for some u ∈ S(∆, L). Indeed, by Lemma 3.2 applied to an
affine-linear function ϕ = f , we get that s(x) satisfies the defining property
(40). 
Definition 3.2. Let (∆, L) be a labelled compact convex simple polytope
in Rm , S(∆, L) the space of strictly convex smooth function on ∆0 satisfying the
conditions of Proposition 2.1 and s(∆,L) the extremal affine function of (∆, L).
Then the non-linear PDE
m
X
(49) s(u) := − (u,ij ),ij = s(∆,L)
i,j=1

is called the Abreu equation. If (∆, L, Λ) is a Delzant triple, solutions of (49)

correspond to extremal T-invariant, ω-compatible Kähler metrics on the toric
symplectic manifold (M, ω, T).

2. Donaldson–Futaki invariant
We now introduce an obstruction to the existence of solutions of (49), due
to Donaldson [21].
40 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

Definition 3.3 (Donaldson-Futaki invariant). Given a labelled compact

convex simple polytope (∆, L) in Rm , we define the functional
Z Z
(50) F(∆,L) (ϕ) := 2 ϕdσ − s(∆,L) ϕdv
∂∆ ∆
acting on the space of continuous functions on ∆. F(∆,L) is called the Donaldson–
Futaki invariant associated to (∆, L)
Proposition 3.2 (Donaldson [21]). If (∆, L) admits a solution of (49),
then
F(∆,L) (ϕ) > 0
for any smooth convex function ϕ on ∆ which is not affine-linear.
Proof. Using Lemma 3.2 we find
Z X
F(∆,L) (ϕ) = Hij ϕ,ij dv ≥ 0
∆ ij

where we have used the convexity of ϕ for the inequality. Furthermore, as H > 0
on ∆0 , the inequality is strict unless ϕ,ij = 0, i.e. ϕ is affine-linear. 
Exercise 3.1 (Donaldson [21]). Show that the statement of Proposition 3.2 holds true for continuous
convex functions ϕ which are smooth on the interior ∆0 and are not affine-linear.

3. Uniqueness
We show now that the solution of (49) is unique up to the addition of affine-
linear functions.
Theorem 3.1 (D. Guan [27]). Any two solutions u1 , u2 ∈ S(∆, L) of (49)
differ by an affine-linear function. In particular, on a compact toric Kähler
manifold or orbifold (M, ω, T), there exists at most one, up to a T-equivariant
isometry, ω-compatible T-invariant extremal Kähler metric (g, J).
Proof. Consider the functional
Z  
(51) E(∆,L) (u) := F(∆,L) (u) − log det Hess(u) − log det Hess(u0 ) dv,
∆
referred to as the relative K-energy of (∆, L) with respect to T. It is well-defined
for elements u ∈ S(∆, L) by virtue of the equivalent boundary conditions (35)-
(36). Using the formula d log det A = tr A−1 dA for any non-degenerate matrix
A and Lemma 3.2, one computes the first variation of E(∆,L) at u in the direction
of u̇
  Z X m
dE(∆,L) (u̇) =F(∆,L) (u̇) − Hiju u̇,ij dv
u ∆ i,j=1
Z h m
X   i
= − Hiju − s(∆,L) u̇ dv,
∆ ,ij
i,j=1

showing that the critical points of E(∆,L) are precisely the solutions of (49).
Furthermore, using dA−1 = −A−1 dAA−1 , the second variation of E(∆,L) at u
in the directions of u̇ and v̇ is computed to be
  Z  
2
d E(∆,L) (u̇, v̇) = tr Hess(u) Hess(u̇) Hess(u) Hess(v̇) dv,
u ∆
 4. K-STABILITY 41

showing that E(∆,L) is convex. In fact, as Hess(u) is positive definite and Hess(u̇)
 
is symmetric, the vanishing d2 E∆,L,f (u̇, u̇) = 0 is equivalent to u̇ being affine.
u
It follows from (35)-(36) that for any two elements u1 , u2 ∈ S(∆, L), u(t) =
tu1 + (1 − t)u2 , t ∈ [0, 1] is a curve in S(∆, L) with tangent vector u̇ = u1 − u2 .
Using the convexity of E(∆,L) , it follows that that if u1 and u2 are two solutions
of (49) (equivalently, u1 and u2 are critical points of E(∆,L) ), then u1 − u2 must
be affine. 

Corollary 3.1. If (49) admits a solution u∗ ∈ S(∆, L), then the relative
K-energy E(∆,L) atteints its minimum at u∗ .

Proof. The arguments in the proof of Theorem 3.1 show that E(∆,L) is
convex on S(∆, L). The solution u∗ being a critical point of E(∆,L) , it is therefore
a global minima. 

4. K-stability
Definition 3.4 (Toric K-stability). We say that a labelled compact con-
vex simple polytope (∆, L) in Rm is K-semistable if

(52) F(∆,L) (ϕ) ≥ 0

for any convex piecewise affine linear (PL) function ϕ = max(f1 , . . . , fk ) on ∆.

The labelled polytope (∆, L) is K-stable, if moreover, equality in (52) is achieved
only for the affine-linear functions ϕ. If (∆, L, Λ) is a Delzant triple, we say that
the corresponding toric symplectic manifold(M, ω, T) is K-stable iff (∆, L) is.

An immediate corollary of Proposition 3.2 is the following

Corollary 4.1 (Donaldson [21]). If (49) admits solution, then (∆, L) is

K-semistable.

This can be improved to

Theorem 3.2 (Zhou–Zhu [58]). If (49) admits a solution u ∈ S(∆, L)

then (∆, L) is K-stable.

Proof. The proof is elementary and uses integration by parts (similar to

the proof of Lemma 3.2) in order to obtain a distribution analogue of (41) in the
case when ϕ is a PL convex function. For simplicity we will check that if (49)
admits a solution u ∈ S(∆, L), then F(∆,L) is strictly positive for a simple convex
PL function ϕ, i.e. ϕ = max(L, L̃) where L and L̃ are affine linear functions on
t∗ with L − L̃ vanishing in the interior of ∆; as F(∆,L) (L̃) = 0 by (40) we can
assume without loss of generality that L̃ ≡ 0.
Denote by F = ∆ ∩ {L = 0} the ‘crease’ of ϕ. This introduces a partition
∆ = ∆0 ∪ ∆00 of ∆ into 2 sub-polytopes with F being a common facet of the
two. Notice that ϕ is affine linear over each ∆0 and ∆00 , being zero over ∆00 , say.
Furthermore, dL defines an inward normal for ∆0 (by convexity) and we use (39)
to define a measure dσ on ∂∆0 . Let us write ∂∆0 = F ∪ ∂∆0 , where ∂∆0 is the
union of facets of ∆0 which belong to ∂∆. We then have (using that ϕ ≡ 0 on
42 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

∆00 )
Z Z
F(∆,L) (ϕ) = 2 ϕdσ − s(∆,L) ϕdv
∂∆ ∆
Z Z
=2 ϕdσ − s(∆,L) ϕdv.
∂∆0 ∆0
We now use (45) and (46) 0
P over ∆ with Hij satisfying the conditions of Propo-
sition 2.1 on ∆ and − ij Hij,ij = s(∆,L) . Noting that ϕ is affine over ∆0 , i.e.
ϕ,ij = 0 we have
Z Z X m 
− s(∆,L) ϕdv = ϕ Hij,ij dv
∆0 ∆0 i,j=1
Z
=−2 ϕdσ
∂∆0
Z  Xm 
− ϕ hdH(dL, ei ), e∗i i dσ
F i=1
Z
+ H(dL, dL)dσ.
F
As ϕ vanishes on F , the term at the third line is zero, so that
Z
(53) F(∆,L) (ϕ) = H(dL, dL)dσ > 0,
F
as H is positive definite over ∆0 . 
Thus motivated, the central conjecture is
Conjecture 3.1 (Donaldson [21]). (49) admits a solution in S(∆, L) if
and only if (∆, L) is K-stable.

5. Toric test configurations

In this section we shall explain the notion of toric test configuration which
gives a geometrical meaning of the convex PL functions appearing in the defini-
tion of K-stability.
We suppose that (∆, L) ⊂ Rm is a Delzant polytope with respect to (the dual
of) the standard lattice Zm ⊂ Rm , corresponding to a symplectic toric manifold
(M, ω, Tm ), and f = max(f1 , . . . , fp ) is a convex PL function over ∆, with
f1 , . . . , fp being a minimal set of affine linear functions with rational coefficients
defining f . Multiplying f by a suitable denominator, we can assume without
loss that all fi ’s have integer coefficients. We choose R > 0 such that (R − f ) is
strictly positive on ∆, and consider the labelled polytope in P ⊂ Rm+1 = Rm ×R,
defined by
n o
(54) P = (x, xm+1 ) ∈ Rm+1 : x ∈ ∆, 0 ≤ xm+1 ≤ (R − f (x)) .
The polytope P is rational Delzant with respect to the labels
n o
Lj (x) ≥ 0, xm+1 ≥ 0, (R − xm+1 − fi (x)) ≥ 0, j = 1, . . . , d, i = 1, . . . , p ,
where L = (L1 (x), . . . , Ld (x)) are the labels of ∆, and (the dual of) the lattice
Zm+1 ⊂ Rm+1 . For simplicity, we shall assume that P is Delzant, i.e. that it
 5. TORIC TEST CONFIGURATIONS 43

gives rise to a smooth toric symplectic 2(m + 1)-dimensional manifold (M, Ω),
but the discussion below holds in the general orbifold case.
We denote by Tm+1 = Tm × S1(m+1) the corresponding torus, with Tm identi-
fied with the torus acting on M . Notice that ∆ is a facet of P, whose pre-image
is a smooth submanifold M̃ ⊂ M. The Delzant construction identifies the sta-
bilizer of points in M̃ 0 with the circle subgroup S1(m+1) ⊂ Tm+1 , so that with
respect to the induced action of Tm+1 /S1 ∼
= Tm , (M̃ , Ω| , is equivariantly
(m+1) M̃
isomorphic to (M, ω) by Delzant’s theorem. We shall thus assume without loss
of generality that M = M̃ and ω = Ω|M̃ .
Let us now choose an Ω-compatible Tm+1 -invariant complex structure J on
M, which induces a Tm -invariant ω-compatible complex structure J on (M, ω).
Donaldson [21, Proposition 4.2.1] shows that with respect to the C∗ -action ρ :
C∗ → Aut(M) induced by S1(m+1) , the complex (m + 1)-dimensional manifold
(M, Ω, ρ) is an example of a Kähler test configuration associated to the Kähler
manifold (M, ω), meaning that the following are satisfied:
Definition 3.5 (Kähler test configurations). A smooth Kȧhler test
configuration associated to a compact complex m-dimensional Kähler manifold
(M, ω) is a compact complex (m + 1)-dimensional Kȧhler manifold (M, Ω) en-
dowed with a C∗ -action ρ, such that
• there is a surjective holomorphic map π : M → CP 1 such that each
fibre Mτ := π −1 (τ ) ∼
= M for τ ∈ CP 1 \ {0};
• the C -action ρ on M is equivariant with respect to the standard C∗ -
∗

action on CP 1 fixing 0 and ∞;

• there is a C∗ -equivariant biholomorphism α : M\M0 → M ×(CP 1 \{0})
with respect to the C∗ -action ρ and the standard action of C∗ on the
second factor of the product.
• M is endowed with a Kähler metric Ω, which restricts to the Kähler
metric ω on M ∼ = M1 .
To see the existence of such equivariant π in our toric case, it is instructive
to think of the blow down map b1 plotted in Figure 2. The Delzant polytope
at the lhs is associated to a PL convex over the interval whereas the Delzant
polytope at the rhs represents the product CP 1 × CP 1 . It this case, M = CP 1
is one of the factors of the product, and the projection to the other factor CP 1
defines the map π. Notice that for this example, π −1 (0) = CP 1 ∪ E is the union
of a copy of M with the exceptional divisor of the blow-up, which is also the
pre-image of the facets of the polytope satisfying (R − f ) = 0.
Definition 3.6 (Toric test configuration). Let (M, ω) be a toric Kähler
manifold with labelled Delzant polytope (∆, L) in Rm with respect to the lattice
Zm . A Kähler test configuration (M, Ω, ρ) for (M, ω) obtained from a rational
PL convex function f as above is called toric test configuration.
We shall now express the Donaldson–Futaki invariant (50) of the PL function
f defining a toric test configuration in terms of a differential-geometric quantity
on M. To simplify notation, we shall denote by Scal(Ω) the scalar curvature of
the corresponding Kähler metric on M (defined by (Ω, J)). We also notice that
the momentum map of the sub-torus Tm ⊂ Tm+1 sends M onto ∆ (which is the
projection of P to Rm ) and, with a slight abuse of notation, we shall denote by
44 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

s(∆,L) the smooth function on M obtained by pulling back the extremal affine
linear function s(∆,L) associated to (M, ω). We then have
Lemma 3.3. In the above setting, the Donaldson–Futaki invariant (50) as-
sociated to the convex PL function f is given by
Z   Ωm+1
m+1
(2π) F(∆,L) (f ) = − Scal(Ω) − s(∆,L)
M (m + 1)!
(55)
ωm
Z
+ (8π) .
M m!

Proof. We write (x, x0 ) ∈ Rm × R for the linear coordinates on Rm+1

and let dx and dx ∧ dx0 denote the standard Lebesgue measures on Rm and
Rm+1 , respectively. Using the description (26) of the Kähler forms ω and Ω in
momentum/angular coordinates, we get
ωm
Z Z
(56) (8π) = 4(2π)m+1 dx
M m! ∆

Ωm+1
Z Z
s(∆,L) = (2π)m+1 s(∆,L) (x)dx ∧ dx0
M (m + 1)!
(57) ZP
= (2π)m+1 s(∆,L) (x)(R − f (x)dx.
∆
Using Lemma 2.3 for Scal(Ω) and (41) with ϕ = 1 on P, we compute
Ωm+1
Z Z
1
Scal(Ω) =2 dσP
(2π)m+1 M (m + 1)! ∂P
 Z Z Z 
(58) =2 dx + dσ(R−f )(∆) + (R − f )dσ∂∆
∆ (R−f )(∆) ∂∆
 Z Z 
=2 2 dx + (R − f )dσ∂∆ ,
∆ ∂∆
where for passing from the second line to the third we have used that dσ(R−f )(∆)
is defined by the equality
df ∧ dσ(R−f )(∆) = dx ∧ dx0 .
Lemma 3.3 now follows easily by combining (56), (57) and (58). 
Remark 3.1. (a) In the special case when s(∆,L) = λ is a constant, which
means that (M, ω, J) has vanishing Futaki invariant, one can further specify
R the
dσ
expression in the rhs of (55) as follows. By (40) with f = 1 we obtain λ = 2 R∂∆
dv
R R ∆
whereas (41) yields ∆ s(u) = 2 ∂∆ dσ. It thus follows that
R
s(u)dv
λ = ∆R
dv
R ∆
Scal(ω)ω m
= MR m
Mω
c1 (M ) · [ω]m−1 [M ]
= 4πm ,
[ω]m [M ]
 6. UNIFORM K-STABILITY 45

where Scal(ω) stands for the induced toric Kähler metric on M and u for its sym-
plectic potential. Substituting in (55), we obtain the following co-homological
expression for the Donaldson–Futaki invariant
F(M, Ω) := (2π)m F(∆,L) (f )
h c (M) · [Ω]m [M]   c (M ) · [ω]m−1 (M )  Vol(M, Ω) i
1 1
= −2 −
m! (m − 1)! Vol(M, ω)
+ 4Vol(M, ω).
This formula makes sense for any Kähler test configuration (see Definition 3.5)
and only depends upon the deRham classes [Ω] on M and [ω] on M . By Corol-
lary 4.1, F(M, Ω) ≥ 0 on any toric Kähler test configuration (M, Ω) associ-
ated to (M, ω), with equality if and only if M corresponds to a single rational
affine linear function f . This is the original notion of K-stability going back
to Tian [53]. In this integral form, the Futaki invariant of a test configuration
was first used by Odaka [45, 46] and Wang [55] to study (possibly singular)
polarized projective test configurations.
(b) It turns out that the expression at the rhs of (55) makes sense for any
T-invariant Kähler test configuration associated to a Kähler manifold (M, ω)
endowed with a maximal compact torus T in its reduced group of complex au-
tomorphisms, and it merely depends upon the deRham classes [ω] and [Ω] and
the momentum image ∆ of M for that action of T. This leads to the notion of a
T-relative Donaldson–Futaki invariant F T (M, [Ω]) of a compatible test configu-
ration. This point of view has been taken and developed in [18, 17, 19, 20, 37]
for a general Kähler manifold, where an extension of Corollary 4.1 is also ob-
tained.

6. Uniform K-stability
Let C(∆) denote the set of continuous convex functions on ∆ (continuity
follows from convexity on the interior of ∆), C∞ (∆) ⊂ C(∆) the subset of those
functions which are smooth on the interior ∆0 , and S(∆, L) ⊂ C∞ (∆) the set
of symplectic potentials. Note that, by virtue of Proposition 2.2, if u ∈ S(∆, L)
and f ∈ C∞ (∆), with f smooth on all of ∆, then u + f ∈ S(∆, L). Conversely,
the difference of any two functions in S(∆, L) is a function in C∞ (∆) which is
smooth on ∆.
The affine linear functions act on C(∆) and C∞ (∆) by translation. Let
C ∗ (∆) be a slice for the action on C(∆) which is closed under positive linear
combinations, and C∞ ∗ (∆) the induced slice in C (∆). Then any f in C(∆) can
∞
be written uniquely as f = π(f )+g, where g is affine linear and π(f ) ∈ C ∗ (∆) for
a linear projection π. Functions in C ∗ (∆) are sometimes said to be normalized.
Example 3.1. [21] If x0 ∈ ∆0 is a fixed interior point, a slice as above is given by
n o
∗
C (∆) := f ∈ C(∆) : f (x) ≥ f (x0 ) = 0 .

Example 3.2. [49] Another natural choice for the slice is given by
n Z o
∗
C (∆) := f ∈ C(∆) : f (x)g(x) dv = 0 ∀ g affine linear .
∆

Definition 3.7. Let || · || be a semi-norm on C(∆) which indices a norm (in

the obvious sense) on C ∗ (∆). We say that || · || is tamed if there exists C > 0
46 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

such that on C ∗ (∆)

1
|| · ||1 ≤ || · || ≤ C|| · ||∞ ,
C
· | dv is the L1 -norm and || · ||∞ is the C 0 -norm on C(∆).
R
where || · ||1 := ∆|
R 1/p
Example 3.3. The Lp norm ||f ||p := ∆
|f |p dv defines a tamed norm on the slices defined in
Examples 3.1 and 3.2 for any p ≥ 1.

R Exercise 3.2. Donaldson considers in [21] the slice C ∗ (∆) of Example 3.1 with the norm ||f ||∗ :=
∂∆
f dσ. Show that || · ||∗ is tamed. Hint. Use that f is a positive convex function.

Remark 3.2. (a) For any tamed norm || · ||, both the spaces of PL convex
functions and smooth convex functions on the whole of ∆ are dense in C ∗ (∆).
(b) With respect to a tamed norm || · ||, F(∆,L) is well defined and continuous
on C ∗ (∆).

In this general situation arguments of Donaldson [21] together with an en-

hancement by Zhou–Zhu [59] can be used to prove the following key result.

Proposition 3.3 (Calderbank [8], Donaldson [21], Zhou–Zhu [59]).

For any λ > 0 the following are equivalent:
(i) F(∆,L) (f ) ≥ λ||π(f )|| for all f ∈ C(∆), where F(∆,L) is the Donaldson–
Futaki linear functional (50);
(ii) for all 0 ≤ δ < λ there exists Cδ such that E(∆,L) (u) ≥ δ||π(u)|| + Cδ for
all u ∈ S(∆, L), where E(∆,L) is the relative K-energy introduced in (51).

Proof. As F(∆,L) (f ) and and E(∆,L) (u) are unchanged by the addition of
an affine linear function, it suffices to prove the equivalence for normalized f
and u.
(i)⇒(ii) For any bounded function a on ∆, one can define a modified Futaki
invariant
R Fa by replacing the second integral (over ∆) in the formula (50) by
− ∆ a(x)f (x) dv. Similarly, one can define a modified K-energy Ea using Fa
instead of F(∆,L) in the formula (51). Donaldson [21] shows that Ea (which is
introduced on the space S(∆, L)) can be extended to C∞ (∆) (in fact on a slightly
larger space) taking values in (−∞, +∞].
For any bounded functions a, b, there is a constant C = Ca,b > 0 with
|Fa (f ) − Fb (f )| ≤ C||f || for all f ∈ C ∗ (∆), because || · || bounds the L1 norm on
C ∗ (∆) by assumption. Let us write C = (1+k)C −kC for an arbitrary k ≥ 0 and
take b to be the extremal affine-linear function b = s(∆,L) , so that Fb = F(∆,L) ,
whereas we can take a = s(u0 ) be the scalar curvature of the canonical potential
u0 ∈ S(∆, L). Thus, u0 trivially solves the equation s(u0 ) = a.
By assumption, |Fa (f ) − F(∆,L) (f )| ≤ Cλ−1 (1 + k)F(∆,L) (f ) − kC||f || for all
f ∈ C∞ ∗ (∆) and so F (f ) ≤ (1 + Cλ−1 (1 + k))F
a (∆,L) (f ) − kC||f ||. Turning this
around,
F(∆,L) (f ) ≥ εFa (f ) + δ||f ||,

where 0 < ε := (1 + Cλ−1 (1 + k))−1 < 1 and δ := kCλ(λ + C(1 + k))−1 . Notice
that δ is an injective function of k ∈ [0, ∞) with range [0, λ). For any normalized
 6. UNIFORM K-STABILITY 47

u ∈ S(∆, L) now we estimate

Z  
E(∆,L) (u) = − log det(Hess u) − log det(Hess u0 ) dv + F(∆,L) (u)
Z∆  
≥− log det(Hess u) − log det(Hess u0 ) dv + εFa (u) + δ||u||
∆
= Ea (εu) − m log ε + δ||u|| ≥ Ea (εu) + δ||π(u)||.
It is shown in [21], Proposition 3.3.4 that Ea is bounded below on the space
C∞ (∆). 1 Letting Cδ be a lower bound of Ea , the claim follows.
(ii)⇒(i) Suppose E(∆,L) (u) ≥ δ||u|| + Cδ for all normalized u ∈ S(∆, L). We
shall fix one such u. By density and continuity (see Remark 3.2), it suffices
to prove (i) for f ∈ C∞ ∗ (∆) which are smooth on ∆. Then for all k > 0,

u + kf ∈ S(∆, L) and so E(∆,L) (u + kf ) ≥ δ||u + kf || + Cδ . We thus find

Z  det Hess (u + kf ) 
kF(∆,L) (f ) = E(∆,L) (u + kf ) − E(∆,L) (u) + log dv
∆ det Hess u
Z  det Hess (u + kf ) 
≥ δ||u + kf || + Cδ + log dv − E(∆,L) (u)
∆ det Hess u
≥ δ||u + kf || + C̃δ
with C̃δ = Cδ − E(∆,L) (u) (for the fixed u ∈ S(∆, L)), since the ratio of the
determinants is at least one for k sufficiently large. Dividing by k and letting
k → ∞ we obtain F(∆,L) (f ) ≥ δ||f ||. Since this is true for all 0 ≤ δ < λ and all
smooth functions in C∞ (∆), we have (see Remark 3.2) that F(∆,L) (f ) ≥ λ||f ||
for all f ∈ C(∆). 
Remark 3.3. In the case || · || = || · ||∗ of Example 3.1 and Exercise 3.2,
the implication (i)⇒(ii) of Proposition 3.3 is due to Donaldson [21] when δ = 0,
and to Zhou–Zhu [59] for some δ > 0. The proof presented here is due to David
Calderbank [8].
Definition 3.8 (Uniform K-stability [49]). A compact convex simple
labelled polytope (∆, L) for which the condition (i) of Proposition 3.3 is satisfied
for some constant λ > 0 is called uniformly K-stable with respect to the chosen
slice C∞∗ (∆) and norm || · ||. We say that (∆, L) is L uniformly K-stable if it is
p
uniformly K-stable with respect to the slice introduced in Example 3.2 and the
Lp norm with p ≥ 1, see Example 3.3. We say that (∆, L) is ∗-uniformly K-stable
if it is uniformly K-stable with respect to the slice introduced in Example 3.1
and norm in Exercise 3.2.
Using that convex PL functions are dense in C(∆) (see Remark 3.2), uniform
K-stability can be equivalently introduced by requiring that the condition (i) of
Proposition 3.3 is satisfied on convex PL functions. Thus, uniform K-stability
(with respect to any tamed norm) is a stronger condition on (∆, L) than the
K-stability.

A key result in [21] is the following

1It follows along the lines of Corollary 3.1 that E is convex on S(∆, L), and that u being
a 0
a critical point is a global minima. Donaldson’s argument extends this property to the larger
domain C∞ (∆)
48 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

Theorem 3.3 (Donaldson [21]). If (∆, L) is a compact convex labelled

polygone in R2 such that the corresponding extremal affine linear function s(∆,L)
is strictly positive on ∆, then (∆, L) is K-stable if and only if it is ∗-uniformly
K-stable.
Notice that the above result applies in particular to labelled polygons (∆, L)
for which s(∆,L) is constant. In this case, we also have
Theorem 3.4 (Székelyhidi [49]). If (∆, L) is a compact convex labelled
polygone in R2 such that the corresponding extremal affine linear function s(∆,L)
is constant, then (∆, L) is K-stable if and only if it is L2 -uniformly K-stable.
We end this section by mentioning the following generalization of Theo-
rem 3.2.
Theorem 3.5 (Chen–Li–Sheng [10]). If (∆, L) is a compact convex simple
labelled polytope in Rm such that the Abreu equation (49) admits a solution in
S(∆, L), then (∆, L) is ∗-uniformly K-stable.

7. Existence: an overview
7.1. The Chen–Cheng and He results. We review here a recent break-
through in the general existence theory of extremal Kähler metrics, due to Chen–
Cheng [12, 13, 14] in the constant scalar curvature case, with an enhancement
by He [32] to cover the extremal Kähler metric case. The key notion for these
results to hold is the properness of the relative K-energy with respect to a re-
ductive complex Lie group, first introduced by Tian [52] in the Fano case, and
adapted by Zhou–Zhu [59] to the general Kähler case. We start by explaining
the general setting.

Let (M, J) be a compact m-dimensional complex manifold admitting a Kähler

metric g0 with Kähler form, ω0 . It is well known (see e.g. [25]) that the group
of complex automorphisms Aut(M ) of (M, J) is a complex Lie group with Lie
algebra identified with the real smooth vector fields on M whose flow preserves
J. We denote by Aut0 (M ) its connected component to the identity.
We let
H(M ) = {φ ∈ C ∞ (M ) : ωφ = ω0 + ddc φ > 0}
be the space of smooth Kähler potentials relative to ω0 . As R acts on H(M ) by
translation, preserving the Kähler metric corresponding to the Kähler ωφ , it is
convenient to choose a slice (or normalization) H0 (M ), establishing a bijection
between Kähler metrics with Kähler forms in the deRham class [ω0 ]Rand elements
of H0 (M ). One popular normalization is H0 (M ) = φ ∈ H(M ) : M φ ωφm = 0


but we shall use below another choice for H0 (M ). For any chosen normalization
H0 (M ), we will write
φ = φ0 + const,
where φ ∈ H(M ) and φ0 ∈ H0 (M ). Furthermore, for any σ ∈ Aut0 (M ), we
denote by σ[φ] ∈ H0 (M ) the unique ω0 -relative Kähler potential in H0 (M )
associated to the Kähler form σ ∗ (ωφ ).
We now fix a (real) connected compact subgroup K ⊂ Aut0 (M ) and denote
by G = K C ⊂ Aut0 (M ) its complexification, i.e. the smallest closed complex
subgroup in Aut(M ) containing K. By a standard averaging argument, we can
 7. EXISTENCE: AN OVERVIEW 49

assume that the initial Kähler structure (g0 , J, ω0 ) is invariant under the action of
K, and consider the subspace H(M )K ⊂ H(M ) of K-invariant Kähler potentials
in H(M ). Thus, with a chosen normalisation H0 (M ), the space H0 (M )K :=
H0 (M ) ∩ H(M )K parametrizes the K-invariant Kähler metrics on (M, J) whose
Kähler forms belong to [ω0 ].
Let d be a distance on H(M ). For φ1 , φ2 ∈ H(M ), we let
(59) dG (φ1 , φ2 ) := inf d(φ1 , σ[φ2 ]),
σ∈G

where, we recall, φ0
∈ H0 (M )
denotes the normalized Kähler potential of ωφ for
0
the chosen normalization H (M ).
Definition 3.9 (Tian [52]). Let E K be a functional defined on the space
H(M )K and d a distance on H(M ). We say that E is G-proper with respect to
d if
• E K is bounded on H(M )K ;
• for any sequence φi ∈ H0 (M )K with dG (φ0 , φi ) → ∞, we have E K (φi ) →
∞.
It is useful to notice that if K0 ⊂ K is a compact subgroup with complex-
ification G0 ⊂ G, and E K is the restriction to H(M )K of a functional E K0 on
H(M )K0 which is G0 -proper, then E K is also G-proper. We also notice that the
above notion of properness depends upon the choice of normalization H0 (M ).

The normalization we shall use is given by

(60) H0 (M ) := {φ ∈ H(M ) : I(φ) = 0},
where the functional I : H(M ) → R is introduced by the formula
Z Xm 
(61) I(φ) = φ ω0m−j ∧ ωφj = 0.
M j=0

It is known that (see [25], Chapter 4) a smooth segment φ(t) starting at φ0 = 0

belongs to H0 (M ) iff
Z
m
(62) φ̇(t) ωφ(t) = 0, ∀t.
M
Furthermore, the distance relevant to us will be the one considered by Darvas in
[15]: For any smooth segment φ(t) ∈ H(M ), t ∈ [0, 1], its length can be defined
by
 ωm 
Z 1Z
φ̇(t) φ(t) dt.

0 M m!
We then define d1 (φ1 , φ2 ) to be the infimum of the lengths of all segments with
endpoints φ1 and φ2 .

It is known (see e.g. [25]) that the group of complex automorphisms Aut0 (M )
admits a closed connected subgroup Autr (M ), called reduced group of automor-
phisms, whose Lie algebra is the space of real vector fields whose flow preserves
J, and which vanish somewhere on M . Furthermore, we let T ⊂ Autr (M ) be
a maximal (real) subtorus of Autr (M ), and denote by G = T C ⊂ Autr (M )
its complexification (which is a maximal complex subtorus of Autr (M )). By a
result of Calabi (see [25], Chapter 3), if there exists an extremal Kähler metric
50 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

ωφ for some φ ∈ H(M ), then there is also an isometric extremal Kähler metric
ωφ̃ with φ̃ ∈ H(M )T . Thus, without loss, one can reduce the problem of finding
extremal Kähler metric in the de Rham class of ω0 to the related problem on
H0 (M )T . Furthermore, there is a natural functional E T : H0 (M )T → R, called
the T-relative K-energy and introduced by Mabuchi [40] and Guan [27], whose
critical points are precisely the Kähler potentials in H∗ (M )T , corresponding to
T-invariant extremal Kähler metrics in [ω0 ] (see [25], Chapter 4). Classically,
the relative K-energy is introduced with respect to a fixed maximal subgroup
K ⊂ Autr (M ), and acts on the space of Kähler potentials in H∗ (M )K , but it
is not difficult to see (using that the extremal vector field is central in the Lie
algebra of K) that its definition actually extends to H(M )T for any maximal
torus T ⊂ K.
The main result of [14, 32] can be then stated as follows.
Theorem 3.6 (Chen–Cheng [13], He [32]). Suppose T ⊂ Autr (M ) is
a maximal real torus and G = TC its complexification inside Aut(M ). If the
relative K-energy E T acting on H(M )T is G-proper with respect to the distance
d1 and the normalization (60), then there exists φ ∈ H(M )T such that ωφ is an
extremal Kähler metric.

We now turn to the toric case. Let (M, ω) be a smooth toric symplectic
manifold, classified by its Delzant polytope (∆, L, Λ). In this case T is a maximal
torus in Autr (M ) for any T-invariant compatible complex structure J on M .
We shall fix a point p0 on M 0 , corresponding to x0 ∈ ∆0 , a basis e of Λ, and
consider the slice
n o
∗ ∗
C∞ (∆) := f ∈ C∞ (∆) : f (x0 ) = 0 and dx0 f = 0
introduced in Example 3.1. We shall denote S ∗ (∆, L) := S(∆, L) ∩ C∞ ∗ (∆) the

corresponding slice in the space of symplectic potentials.

Using the identification (31), for any u ∈ S ∗ (∆, L) we associate a function
Fu (z) on (Cm )∗ , defined by
1 1
Fu (z) = φ(y) = φ( log |z1 |2 , . . . , log |zm |2 ),
2 2
where φ(y) is given by (32). Furthermore, Lemma 2.4 tels us that the Tm -
invariant function Fu (z) introduces a Kähler form ωu := ddc Fu (z) on (C∗ )m ,
which extends to a smooth Kähler metric (gu , ωu , Ju ) on M by identifying
(M 0 , Ju ) to (C∗ )m via (31). The fact that u ∈ S ∗ (∆, L) leads to the normal-
ization Fu (1, . . . , 1) = Fu (p0 ) = 0 for Fu (z). Let u0 be the canonical symplectic
potential given in Theorem 2.1 and u∗0 := π(u0 ) ∈ S ∗ (∆, L) its normalization.
Then, on (C∗ )m , the corresponding Kähler forms ω0 := ddc Fu∗0 and ωu = ddc Fu
are related by
ωu = ω0 + ddc (Fu (z) − Fu∗0 (z)) = ω0 + ddc φu
for the Tm -invariant smooth function φu (z) := Fu (z) − Fu∗0 (z) on (C∗ )m , satis-
fying
(63) φu (p0 ) = 0.
The point is that now ωu and ω0 define two different Kähler metrics on the
same complex manifold M∆C , and φ extends to a T-invariant Kähler potential
u
 7. EXISTENCE: AN OVERVIEW 51

with respect to ω0 , i.e. φu ∈ H(M )T . Conversely, using the dual Legendre

transform, one can show that any T-invariant Kähler potential φ ∈ H(M )T with
respect to ω0 which satisfies (63) gives rise to a normalized symplectic potential
u ∈ S ∗ (∆, L). A key feature of this correspondance is that if we take a path
u(t) ∈ S ∗ (∆, L) and consider the corresponding path φ(t) := φut ∈ H(M )T ,
formula (32) shows that
(64) u̇(t) = −φ̇(t).
However, the normalization (63) for the corresponding relative Kähler potentials
φ(t) is not consistent with (60), so we further consider the action of R on S(∆, L)
by translations. If ũ = u+c for u ∈ S ∗ (∆, L), then we have ỹi = yi and φ̃ = φ−c
in (32), showing that (64) holds true for paths ũ(t) ∈ S(∆, L) satisfying the
weaker normalization condition dũx0 = 0. We now introduce a different slice on
C∞ (∆):
n Z Z o
(65) 0
C∞ (∆) := f ∈ C∞ (∆) : dfx0 = 0 and f dv = u∗0 dv ,
∆ ∆
where, we recall, u∗0 = π(u0 ) is the ∗ (∆)
C∞ normalization of the canonical sym-
plectic potential. Letting S 0 (∆, L = S(∆, L) ∩ C∞
0 (∆) be the corresponding slice

in the space of symplectic potentials, we have

Lemma 3.4. For any path ũ(t) in S 0 (∆, L), the corresponding ω0 -relative
Kähler potentials φ(t) = φũ(t) obtained by (32) belong to H0 (M )T and satisfy
d d
ũ(t) = − φ(t).
dt dt
Conversely, any path in H (M ) comes from a path ũ(t) in S 0 (∆, L).
0 T

Proof. We have already observed that the second part of the claim holds
for any path satisfying the weaker normalization condition dx0 ũ(t) = 0. Using
this and (62), it follows that I(φ(t)) = const. As S 0 (∆, L) is convex and u∗0 ∈
S 0 (∆, L), it follows that I(φ(t)) = 0. 
Through the above correspondence, one can identify as in [21] the the relative
K-energy E T acting on H0 (M )T to the functional E(∆,L) defined in (51), acting
on S 0 (∆, L). We are now ready to state and prove our main observation in this
section.
Proposition 3.4. Suppose (∆, L) is a ∗-uniformly K-stable Delzant labelled
polytope, corresponding to toric Kähler manifold (M, ω0 , J). Then the relative
K-energy E T is TC -proper on H(M )T with respect to the distance d1 and the
normalization (60).
Proof. We use the bijection between the ω0 -relative Kähler potentials in
H0 (M ) and symplectic potentials ũ ∈ S 0 (∆, L) established in Lemma 3.4 in
order to deduce the G = TC properness of E T from the property (ii) in Proposi-
tion 3.3 of the functional E(∆,L) .
Let φj , j = 1, . . . , ∞ be a sequence in H0 (M )T with d1,G (0, φj ) → ∞. De-
note by ũj ∈ S 0 (∆, L) the corresponding sequence of normalized symplectic
potentials. We write
Z
∗ 1
ũj = uj + (u∗ − ũ∗j ) dv, j = 0, 1, . . . , ∞,
Vol(∆) ∆ 0
52 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

where u∗j := π(ũj ) is the projection of ũj to the slice S ∗ (∆, L). Let us consider
the path ũj (t) =R (1 − t)u∗0 + tũj ∈ S 0 (∆, L). Using Lemma 3.4, the d1 -length of
this path is Cm ∆ |ũj − u∗0 | dv where Cm = (2π)m . Thus, we have
d1,G (0, φj ) ≤ d1 (0, φj )
Z
≤ Cm |ũj − u∗0 | dv
∆
(66)
Z  Z
 ∗ ∗ 1 
= Cm (uj − u0 ) + (u∗0 − ũ∗j ) dv  dv

∆ Vol(∆) ∆
Z
≤ (Cm + 1) |u∗j − u∗0 | dv.
∆
∗
R
Thus, d1,G (0, φj ) → ∞ yields ∆ |uj | dv
→ ∞.
Now, by assumption, (∆, L) is ∗-uniform K-stable and by Proposition 3.3
(ii) we have
E T (φj ) = E(∆,L) (ũj ) ≥ δ||u∗j ||∗ + Cδ
Z
0
≥δ |u∗j | dv + Cδ → ∞,
∆
where for passing from the first line to the second we have used that ||·||∗ bounds
the L1 -norm, see Exercise 3.2. 
Corollary 7.1. Suppose (∆, L) is a ∗-uniformly K-stable Delzant labelled
polytope, corresponding to a toric Kähler manifold (M, ω0 , J). Then (M, J) ad-
mits a T-invariant extremal Kähler metric whose Kähler form is in the deRham
class [ω0 ]. Equivalently, the Abreu equation (49) admits a solution in S(∆, L).
Proof. This follows from Theorem 3.6 and Proposition 3.4, using the bi-
jection between the ω0 relative Kähler potentials in H0 (M ) and symplectic po-
tentials ũ ∈ S 0 (∆, L) established in Lemma 3.4. 

7.2. Donaldson’s result. We now discuss the Donaldson existence result

in dimension m = 2.
Theorem 3.7 (Donaldson [24]). Suppose (∆, L) is a labelled compact con-
vex simple polytope in R2 for which the affine extremal function s(∆,L) is constant
and (∆, L) is K-stable. Then (49) admits a solution. In particular, a compact
symplectic toric 4-manifold admits a CSC Kähler metric iff it is K-stable and
its extremal affine-linear function is constant.
An extension to the extremal case, under the ∗-uniform stability assumption
and positivity of s(∆,L) , appears in [11].

7.3. The continuity method. The main idea of the approach in [24] is to
use the continuity method over paths (∆, L(t)), t ∈ [0, 1] of labelled, compact,
convex, simple Delzant polytopes in Rm , where the labels Ltj (x) = hutj , xi + λtj
vary in a continuous way. We thus introduce
Definition 3.10. Let ∆ ⊂ (Rm )∗ be a given compact, convex, simple poly-
tope which has d codimension-one faces and
L = {L1 (x), · · · , Ld (x)}
 7. EXISTENCE: AN OVERVIEW 53

is a chosen labelling. The cone N (r) ⊂ Rd of admissible labels of ∆ is

n 1 1 o
N (r) := L(r) = { L1 (x), . . . , Ld (x)} : ri > 0 .
r1 rd
Thus, L(1, . . . , 1) = L. We further introduce the subspaces
• SN (r) ⊂ N (r) the subset of normals for which (∆, L(r)) is K-stable;
• CN (r) ⊂ N (r) the subset of normals for which the extremal affine
function of (∆, L(r)) is constant;
• SCN (r) = CN (r) ∩ SN (r).
The basic observation is
Lemma 3.5. SN (r) is a convex subset of N (r) ∼ = {ri > 0} ⊂ Rd whereas
CN (r) is the (convex) cone obtained by intersecting N (r) with a co-dimension
m hyperplane of Rd . In particular, SCN (r) is convex too.
Proof. By the definition (39) of the measure on ∂∆, dσ depends linearly on
r = (r1 , . . . , rd ). Thus, the RHS of (47) depends linearly on r, and so does the
extremal affine function s(∆,L(r)) . This shows that CN (r) is a the intersection of
N (r) with a co-dimension m hyperplane in Rd . Finally, by the above arguments,
the Donaldson–Futaki invariant F(∆,L(r)) (ϕ) (see (50)) depends linearly on r,
showing the first claim too. 
Thus, given a K-stable labelled polytope polytope (∆, L), the continuity
method of [24] consists of establishing the following
Step 1. There exists a canonical choice r0 ∈ SCN (r) such that on (∆, L(r0 ))
(49) admits a solution. Connecting r0 and r1 := (1, . . . , 1) by a linear
segment, we obtain a path of labelled polytopes (∆, L(t)), t ∈ [0, 1]
such that: (a) for any t, (∆, L(t)) is K-stable and its extremal affine
function is a constant; (b) for t = 0, (49) admits solution.
Step 2. If (49) admits a solution in S(∆, L(t0 )) for t0 ∈ [0, 1], then there exists
ε0 > 0 such that (49) admits a solution in S(∆, L(t)) for any |t−t0 | < ε0 .
Step 3. The subset {t ∈ [0, 1] : (49) admits a solution in S(∆, L(t))} is closed.
The hardest part is Step 3. We discuss below how to obtain Steps 1 and 2.
7.4. The canonical solution. This concerns Step 1. The original method
in [23] uses a result of Arezzo–Pacard. However, in [23], an alternative, simpler
approach is suggested, modulo a suitable generalization of a result of [51] to
arbitrary labelled polytopes, which is now available due to E. Legendre [38].
We follow this second approach.
Suppose (∆, L, Λ) is a Delzant triple. It is well-known that the corresponding
Kähler manifold (M, J0 , ω) is Fano, i.e. satisfies
1
c1 (M, J0 ) = [ω]
2π
if and only if (∆, L, Λ) is a reflexive lattice polytope in (V ∗ , Λ∗ ), where Λ∗ ⊂ V ∗
denotes the dual lattice of Λ ⊂ V . Recall that a lattice polytope in (V ∗ , Λ∗ ) is a
polytope whose vertices belong to Λ∗ . The dual polytope ∆∗ ⊂ V is defined by
∆∗ = {v ∈ V : (v, x) + 1 ≥ 0, ∀x ∈ ∆},
and ∆ is reflexive if ∆∗ is also a lattice Delzant polytope in (V, Λ). It is easy to
see that for such reflexive Delzant polytopes (∆, L, λ), Lj (0) = λ is independent
54 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

of j, i.e., in a more affine-invariant way, there is an interior point x0 ∈ ∆ and a

positive constant λ > 0 such that
(67) Lj (x0 ) = λ, ∀Lj ∈ L
Of course, this condition makes sense for an arbitrary labelled convex poly-
tope (∆, L)
Definition 3.11. We say that a convex, simple labelled polytope (∆, L) is
monotone if there exists an interior point x0 ∈ ∆ for which (67) is satisfied.
The reason for the name is the following
Lemma 3.6. Suppose (∆, L) is a monotone labelled simple convex polytope
and and u ∈ S(∆, L) a symplectic potential defining a Kähler metric (27) on
∆0 × Tm . Then,
1 1
ρg − ω = ddc f,
λ 2
for a smooth function on ∆, called a Ricci potential of u.
Proof. We can translate (∆, L) and assume without loss of generality that
x0 is the origin in V . We now use (30) and Lemma 2.4 to write
m
1 c
2 X 
ρg − λω = dd log det Hess(u) − (−u(x) + xi u,i ) .
2 λ
i=1
Using the boundary conditions restated in Exercise 2.3, it is easily seen that
m

2 X
f (u) := log det Hess(u) − (−u(x) + xi u,i )
λ
i=1
d  m
X 1 X 
= − log Lj − (−Lj log Lj + xi (Lj log Lj ),i ) + smooth
(68) λ
j=1 i=1
d 
X 1 
= − log Lj − (−Lj log Lj + (Lj − λ) log Lj ) + smooth
λ
j=1
= smooth.

1
If (∆, L, Λ) is a Delzant triple satisfying (67), using that 2π ρg represents
c1 (M, J), Lemma 3.6 explicitly shows the Fano property of (M, J, ω). In this
case, Wang–Zhu [51] proved the following important result
Theorem 3.8 (Wang–Zhu [51]). Suppose (∆, L) is a monotone Delzant
polytope for which the extremal affine function s(∆,L) is constant. Then there
exists u ∈ S(∆, L) such that ρg = λ1 ω, i.e. the corresponding T-invariant, ω-
compatible metric on (M, ω, T) is Kähler–Einstein. In particular, u is a solution
to (49).
The proof of this result itself uses the continuity method. By Lemma 3.6,
we want to solve for each t ∈ [0, 1]

m
det Hess(u0 + ϕ)  2t X 
(69)
= exp (−ϕ(x) + xi ϕ,i ) − f0 (x) ,
det Hess(u0 ) λ
i=1
 7. EXISTENCE: AN OVERVIEW 55

where (u0 , f0 ) are the symplectic and Ricci potentials of the canonical metric
(g0 , J0 ). For t = 0, the existence of a solution u0 + ϕ ∈ S(∆, L) follows from
Yau’s theorem [57] of prescribing the Ricci form in a given Kähler class on
M∆ C = (M, J ). The openness follows too from a standard argument on M C .
0 ∆
The closedness reduces, via the method of [57], to a uniform C 0 (M ) bound for
the solutions ϕt of (69), see e.g. [54, 60]. The latter is derived in [51] by using
only the convexity of ∆ ⊂ (Rm )∗ .

An extension of Theorem 3.8 to arbitrary compact convex simple labelled

polytopes was obtain by E. Legendre [38], building on the general strategy sug-
gested in [23] to apply the machinery of global analysis to compact simple la-
belled polytopes which are not necessarily Delzant. Recall that by Theorem 1.3,
if (∆, L) is Delzant, there are natural holomorphic chart Cm v associated to a
vertex v of ∆, and two different charts are identified using the relation (17). We
can match this with the discussion in Section 4. Indeed, fixing a reference point
p0 ∈ M 0 with x0 = µ(p0 ) ∈ ∆0 and, by adding an affine-linear function to u,
we assume yi (x0 ) = u,i (x0 ) = 0. Furthermore, at each vertex v ∈ ∆ we use
the identification (31) with ev = {uv1 , . . . , uvm } being the basis obtained from the
normals of facets containing v. We can also translate the vertex v at the origin
so that, by the boundary conditions, u = 21 m
P
j=1 (xj log xj − xj ) + ϕ for some
smooth function ϕ(x) on ∆. Then, the J holomorphic coordinates
√
−1tj
zju = eu,j e = zj0 eϕ,j
allow to extend the identification Φu,p0 ,ev : (C∗ )m 0
v → M to an equivariant
m
embedding Φu,p0 ,ev : Cv → (M, J). Choosing another vertex w, we get that
Φu,p0 ,ev and Φu,p0 ,ew are related via (17). Now, for any u ∈ S(∆, L), Lemma 2.4
shows that the function on Cm v defined by
1 1
Fv (z) = φ(y) = φ( log |z1 |2 , . . . , log |zm |2 )
2 2
satisfies
¯ v > 0,
• 2i∂ ∂F
• Fv (z) is invariant m m
Pm under the linear action of Tv on Cv ;
• Fw = Fv + i=1 (v − w)i log |zi |,
where the third relation reflects the fact that each time we translate the poly-
tope to ensure the corresponding vertex is at the origine. Conversely, any such
collection of functions gives rise to u ∈ S(∆, L).
When (∆, L) is not Delzant, we cannot use (17) in order to construct a
complex manifold M∆ C . But we still have “charts” for the “quotient space”

|M | = M∆ /T ∼
C
= ∆. More precisely, for each vertex v we have maps Φu,v : Cm v →
m
∆ where ri = |zi | are the radial coordinates on Cv . We can thus define an
embedding of Φu,v : [0, ∞)m v → ∆ by letting ri = yi = u,i . The identification
(17) has now sense if we replace zi with ri , and thus we associate to ∆ a Hausdorff
topological space |M | which admits “uniformizing” charts Φv : Cm m
v /Tv → |M |,
precisely as in the definition (1.11) for orbifolds, but now the groups Γi ∼ = Tmv
are tori. Furthermore, we can define the sheaf of smooth functions of |M | by
pull back to each Cm v , etc. Using the relevant analysis in such charts, Legendre
has proved
56 3. THE CALABI PROBLEM AND DONALDSON’S THEORY

Theorem 3.9 (Legendre [38]). Theorem 3.8 holds true for any compact
convex simple labelled polytope (∆, L).

The main advantage of this is the following

Theorem 3.10 (Donaldson [23], Legendre [38]). Let (∆, L) be a labelled

compact convex simple polytope in (Rm )∗ . Then there exits a unique up to an
overall positive scale labelling L∗ such that (∆, L∗ ) is monotone and s(∆,L∗ ) is
constant.

Proof. For any interior point x∗ ∈ ∆0 we let r(x∗ ) = (L1 (x∗ ), . . . , Ld (x∗ )) ∈
N (r) and
 L (x) Ld (x) 
1
L∗ := , . . . , ,
L1 (x∗ ) Ld (x∗ )

be the corresponding label, so that (∆, L∗ ) is monotone. We want to show that

one can choose x∗ so that s(∆,L∗ ) ≡ a∗0 is a constant function. By the last relation
in (47),
R
∗ dσL∗
a0 = 2 ∂∆
Vol(∆)

whereas the other m relations give

R
∂∆ dσL
Z Z
∗
xi dσL∗ = xi dv.
∂∆ Vol(∆) ∆

We thus obtain a linear system for the unknown x∗ :

d Z d
 R x dv  X Z 
∆ i
X
(70) Lj (x∗ ) xi dσLj = R ∗
Lj (x ) dσLj , i = 1, . . . , m.
j=1 Fj ∆ dv j=1Fj

In order to show that (70) has a unique solution we shall use the following
elementary

Lemma 3.7. Let f (x) = hu, xi + λ be an affine linear function on Rm . Then

d Z
X  Z 
f dσLj uj = − dv u ∈ (Rm )∗ .
j=1 Fj ∆

The proof of Lemma 3.7 is left as an exercise, and can be obtained easily
starting from a triangulation of ∆ as the union of simplexes with a common
vertex at the interior of ∆.
We now suppose, without loss of generality, that 0 ∈ ∆0 , and the initial
labelling L satisfies Lj (0) = 1, i.e.

Lj (x) = huj , xi + 1, j = 1, . . . , d.
 7. EXISTENCE: AN OVERVIEW 57

By Lemma 3.7,
d Z
1 X
x∗i = − huj , x∗ i xi dσLj
Vol(∆) Fj
j=1
d Z Z
1 X 1
=− Lj (x∗ ) xi dσLj + xi dσL
Vol(∆) Fj
Vol(∆) ∂∆
j=1
Z Z
(71) 1  
= xi dσL − xi dσL∗ .
Vol(∆) ∂∆ ∂∆
Z X d Z Z
∗ ∗
0 = huj , x i dσLj = Lj (x ) dσLj − dσL
Fj j=1 Fj ∂∆
Z Z
= dσL − dσL∗
∂∆ ∂∆
and (70) becomes
Z Z Z Z
1  
x∗i = dv xi dσL − xi dv dσL .
Vol(∆)2 ∆ ∂∆ ∆ ∂∆

 Bibliography

[1] M. Abreu, Kähler geometry of toric varieties and extremal metrics, Internat. J. Math 9
(1998), 641–651.
[2] M. Abreu, Kähler metrics on toric orbifolds, J. Differential Geom. 58 (2001), 151–187.
[3] V. Apostolov, D. M. J. Calderbank, P. Gauduchon and C. Tønnesen-Friedman, Hamilton-
ian 2-forms in Kähler geometry II Global classification, J. Differential Geom. 68 (2004),
277–345.
[4] M. F. Atiyah, Convexity and commuting Hamiltonians, Bull. London Math. Soc. 14 (1982),
1–15.
[5] G. Bredon, Introduction to Compact Transformation Groups, Pure and Applied Mathe-
matics, 46, Academic Press, New York-London, 1972.
[6] E. Calabi, Extremal Kähler metrics, in Seminar on Differential Geometry, pp. 259–290,
Ann. of Math. Stud. 102, Princeton Univ. Press, Princeton, N.J., 1982.
[7] D.M.J. Calderbank, L. David and P. Gauduchon, The Guillemin formula and Kähler
metrics on toric symplectic manifolds, J. Symplectic Geom. 1 (2003), 767–784.
[8] D.M.J. Calderbank, K-energy and uniform stability for toric varieties, unpublished man-
uscript, 2006.
[9] A. Cannas da Silva, Symplectic toric manifolds, in “Symplectic geometry of integrable
Hamiltonian systems ”(Barcelona, 2001)’, 85–173, Adv. Courses Math. CRM Barcelona,
Birkhuser, Basel, 2003.
[10] B. Chen, A.-M. Li and L. Sheng, Uniform K-stability for extremal metrics on toric vari-
eties, Journal of Differential Equations 257 (2014), 1487–1500.
[11] B. Chen, A.-M. Li and L. Sheng, Extremal metrics on toric surfaces, preprint
arXiv:1008.2607.
[12] X. Chen, J. Cheng, On the constant scalar curvature Kähler metrics: apriori estimates,
arXiv:1712.06697.
[13] X. Chen, J. Cheng, On the constant scalar curvature Kähler metrics: existence results,
arXiv:1801.00656.
[14] X. Chen, J. Cheng, On the constant scalar curvature Kähler metrics: general automor-
phism group, arXiv:1801.05907.
[15] T. Darvas, The Mabuchi geometry of finite energy classes, Amer. J. Math. 139 (2017),
1275–1313.
[16] T. Delzant, Hamiltoniens périodiques et image convexe de l’application moment, Bull. Soc.
Math. France 116 (1988), 315–339.
[17] R. Dervan, Relative K-stability for Kähler manifolds, Math. Ann. 372 (2018), 859–889.
[18] R. Dervan and J. Ross, K-stability for Kähler manifolds, Math. res. Lett. 24 (2017), 689–
739.
[19] Z. Sjöström Dyrefelt, K-semistability of cscK manifolds with transcendental cohomology
class, to appear in J. Geom. Anal.
[20] Z. Sjöström Dyrefelt, On K-polystability of cscK manifolds with transcendental cohomology
class (with Appendix written by R. Dervan), to appear in Int. Math. Res. Not. (IMRN).
[21] S. K. Donaldson, Scalar curvature and stability of toric varieties, J. Differential Geom. 62
(2002), 289–349.
[22] S. K. Donaldson, Interior estimates for solutions of Abreu’s equation, Collect. Math. 56
(2005) 103–142.
[23] S.K. Donaldson, Extremal metrics on toric surfaces: a continuity method, J. Differential
Geom. 79 (2008), 389–432.

59
60 BIBLIOGRAPHY

[24] S.K. Donaldson, Constant scalar curvature metrics on toric surfaces, Geom. Funct. Anal.
19 (2009), 83–136.
[25] P. Gauduchon, Calabi’s extremal Kähler metrics: An elementary introduction, Lecture
Notes available upon request.
[26] P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley Classics Library, John
Wiley and Sons, Inc., New York, 1994.
[27] D. Guan, On modified Mabuchi functional and Mabuchi moduli space of Kähler metrics
on toric bundles, Math. Res. Let. 6 (1999), 547–555.
[28] V. Guillemin, Kähler structures on toric varieties, J. Differential Geom. 40 (1994), 285–
309.
[29] V. Guillemin, Moment Maps and Combinatorial Invariants of Hamiltonian T n -spaces,
Progress in Mathematics 122, Birkhauser, Boston, 1994.
[30] V. Guillemin and S. Sternberg, Convexity properties of the moment mapping, Invent. Math.
67 (1982) 491–513.
[31] V. Guillemin and S. Sternberg, Riemann sums over polytopes, Ann. Inst. Fourier (Greno-
ble) 57 (2007), no. 7, 2183–2195.
[32] W. Hu, on Calabi’s extremal Kähler metrics and properness, arXiv:1801.07636.
[33] P. Heinzner and F. Loose, Reduction of complex Hamiltonian G-spaces, Geom. Funct.
Anal. 4 (1994) 288–297.
[34] Y. Karshon and E. Lerman, Non-compact symplectic toric manifolds, preprint (2009),
arXiv: 0907.2891.
[35] F. Kirwan, Cohomology of quotients in symplectic and algebraic geometry. Mathematical
Notes, 31. Princeton University Press, Princeton, NJ, 1984.
[36] S. Kobayashi and K. Nomizu, Foundations of Differential Geometry, I,II, Interscience
Publishers (1963).
[37] A. Lahdili, Kähler metrics with constant weighted scalar curvature and weighted K-stability,
arXiv:1808.07811.
[38] E. Legendre, Toric Kähler–Einstein metrics and convex compact polytopes, J. Geom. Anal.
26 (2016), 399–427.
[39] E. Lerman and S. Tolman, Hamiltonian torus actions on symplectic orbifolds and toric
varieties, Trans. Amer. Math. Soc. 349 (1997), 4201–423.
[40] T. Mabuchi, K-energy maps integrating Futaki invariants, Tohoku Math. J. (2) 38 (1986),
575–593.
[41] J. Marsden and A. Weinstein, Reduction of symplectic manifolds with symmetry, Rep.
Mathematical Phys. 5 (1974), 121–130.
[42] D. McDuff and D. Salamon, Introduction to Symplectic Topology, Oxford Mathematical
Monographs, Oxford University Press, New York, 1995.
[43] K. Meyer, Symmetries and integrals in mechanics, Dynamical Systems (Proc. Sympos.,
Univ. Bahia, Salvador, 1971), 259–272. Academic Press, New York, 1973.
[44] A. Newlander and L. Nirenberg, Complex analytic coordinates in almost complex manifolds,
Ann. of Math. 65 (1957), 391–404.
[45] Y. Odaka, The GIT stability stability of polarixed varieties via discrepency, Ann. Math.
(2) 177 (2013), 645–661.
[46] Y. Odaka, A generalization of the Ross-Thomas slope theory, Osaka J. Math. 50 (2013),
171–185.
[47] R. Sjamaar, Holomorphic slices, symplectic reduction and multiplicities of representations,
Ann. of Math. (2) 141 (1995), 87–129,
[48] G.W. Schwarz, Smooth functions invariant under the action of a compact Lie group, Topol-
ogy 14 (1975) 63–68.
[49] G. Székelyhidi, Extremal metrics and K-stability, PhD. Thesis, Imperial College London
(2006).
[50] W. Thurston, The Geometry and Topology of Three-Manifolds, New Jersey: Princeton
University Press, 1997.
[51] X.-J. Wang and X. Zhu, Kähler–Ricci solitons on toric manifolds with positive first Chern
class, Adv. Math. 188 (2004), 87–103.
[52] G. Tian, The K-energy on hypersurfaces and stability, Comm. Anal. Geom. 2 (1994),
239–265.
 BIBLIOGRAPHY 61

[53] G. Tian, Kähler-Einstein metrics with positive scalar curvature, Inv. Math. 130 (1997),
1–37.
[54] G. Tian and X. Zhu, Uniqueness of Kȧhler–Ricci soliton, Acta Math. 184 (2000), 271–305.
[55] X. Wang Heigh and GIT weight, Math. res. Letter. 19 (2012), 909–926.
[56] R. O. Wells, Differential Analysis on Complex Manifolds, second edition, Graduate Texts
in Mathematics 65, Springer-Verlag, New York-Berlin, 1980.
[57] S.T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge-
Ampère equation, Comm. Pure Appl. Math. 31 (1978), 339–411.
[58] B. Zhou and X. Zhu, K-stability on toric manifolds, Proc. Amer. Math. Soc. 136 (2008),
3301–3307.
[59] B. Zhou and X. Zhu, Relative K-stability and modified K-energy on toric manifolds, Adv.
Math. 2019 (2008), 1327–1362.
[60] X. Zhu, Kähler–Ricci soliton type equations on compact complex manifolds with C1 (M ) >
0. J. Geom. Anal. 10 (2000), 759–774.