Introduction to Basic Toric Geometry

Gottfried Barthel, Karl-Heinz Fieseler, Ludger Kaup

Contents
1 Fundamental Notions 2
1.1 Group embeddings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Toric varieties: Basic examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Characters and one-parameter subgroups of tori . . . . . . . . . . . . . . . . . . . 7

2 Affine Toric Varieties 11

2.1 Algebraic description: The coordinate ring . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Geometric description: Polyhedral lattice cones . . . . . . . . . . . . . . . . . . . 13
2.3 Cones and orbit structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Toric Singularities 28
3.1 Local structure at a fixed point . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 General toric varieties and fans . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Resolution of toric singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

Introduction
The aim of these notes is to give a concise introduction to some fundamental notions of toric
geometry, with applications to singularity theory in mind. Toric varieties and their singularities
provide a lot of particularly interesting examples: Though belonging to a restricted class, they
illustrate many central concepts for the general study of algebraic varieties and singularities.
Quoting from the introduction of [Ful], one may say that “toric varieties have provided a re-
markably fertile testing ground for general theories”. Whereas a singular variety may not be
“globally” toric, singularities often are “toroidal”, i.e., locally analytically equivalent to toric
ones, so toric geometry can help for a better understanding even of non-toric singular varieties.
In addition to that, for studying certain classes of non-toroidal singularities, methods of toric
geometry turn out to be most useful, e.g., for the resolution of “non-degenerate complete inter-
section singularities”. As a key feature, toric varieties admit a surprisingly simple, yet elegant
and powerful description that prominently uses objects from elementary convex and combinato-
rial geometry. These objects are “rational” convex polyhedral cones and compatible collections
thereof, called “fans”, in a real vector space of dimension equal to the complex dimension of the
variety.

1
2 G.Barthel, K.-H.Fieseler, L.Kaup

The attribute “toric” refers to the algebraic torus of algebraic group theory. In the complex
setting we are dealing with exclusively, the complex algebraic n-torus is an n-fold product
Tn := (C∗ )n , endowed both with its group structure and its structure as an affine algebraic
variety. (It is the complexification of the familiar real n-torus (S 1 )n and includes the latter as
an equivariant deformation retract.) A toric variety is an algebraic variety including T n as an
open dense subset such that the group structure extends to an action on the variety. It turns
out that many familiar algebraic varieties actually are toric; basic singular ones are the quadric
cones V (C3 ; xy−z 2 ) and V (C4 ; xy−zw).
In these notes, we focus on fundamental parts of the theory that are indispensable if one
wants to apply toric methods as a tool for singularity theory. The picture presented here is by
no means complete since important applications to singularity theory, let alone to other parts
of mathematics, had to be left out. As examples, we just mention the role of toric geometry
in studying non-degenerate complete intersection singularities or in the general resolution of
singularities.
We assume that the reader is familiar with elementary concepts of algebraic geometry. Affine
complex algebraic varieties and their morphisms are in one-to-one contravariant correspondence
to finitely generated reduced C-algebras and their homomorphisms: The elements of the algebra
yield the regular functions on the variety, and the points of the variety correspond to the maximal
ideals of the algebra. Ideals determine closed subvarieties; conversely, to any closed subvariety
corresponds its vanishing ideal. General varieties are obtained from affine ones by a natural
gluing procedure that respects the separation condition. All varieties to be considered here are
of finite type, i.e., they admit a finite covering by open affine subspaces. Moreover, we exclusively
deal with (connected) normal varieties, i.e., the “coordinate algebras” corresponding to their
affine open subsets are integral domains and integrally closed in their field of fractions.
Besides these fundamental notions of algebraic geometry, we use some basic concepts of
group actions like orbits, invariant subsets, isotropy subgroups, and fixed points.

1 Fundamental Notions
1.1 Group embeddings
Let G be a complex algebraic group, which means that G is both, a group and a
complex algebraic variety, and these structures are compatible: The group multiplication
G × G −→ G , (g, h) 7−→ gh
and the inversion
G −→ G , g 7−→ g −1
are morphisms of algebraic varieties. A (homo-)morphism between algebraic groups is a
homomorphism of groups which at the same time is a morphism of varieties.
Standard examples are the general linear group GLn (C) and its closed subgroups like
SLn (C), regular upper and lower triangular matrices, and regular diagonal matrices. The
latter family of commutative connected complex algebraic groups plays the key role in
these notes:
Basic Toric Geometry 3

1.1.1 Definition. A (complex) algebraic (n-) torus, usually denoted by T or T n , is

an algebraic group isomorphic to the n-fold cartesian product of the multiplicative group
C∗ of nonzero complex numbers:

T = Tn ∼
= (C∗ )n .

With such a torus, we usually associate a fixed isomorphism Tn ∼ = (C∗ )n . – The toric
varieties, to be considered in the sequel, are embeddings of algebraic tori. We first define
that notion for an arbitrary algebraic group G:
1.1.2 Definition. A G-embedding is an algebraic variety X together with
(1) an algebraic action
G × X −→ X , (g, x) 7−→ g·x = gx
of the group G on X, i.e., that mapping is both, a G-action and a morphism of
algebraic varieties;

(2) an open embedding j : G → X with dense image such that j(gh) = g·j(h) holds for
arbitrary elements g, h ∈ G, i.e., the action of G on X extends the G-action on
G∼= j(G) by left translation.
An immediate example for the group G = GLn (C) is its embedding into the vector
space Cn×n of square matrices.
1.1.3 Remark. Condition (2) may be equivalently restated as follows: There is a “big”
(i.e., open and dense) G-orbit O = G·x0 in X such that the isotropy subgroup Gx0 is
trivial. The embedding j then is just the orbit map g 7→ g·x0 , where x0 is the j-image of
the unit element of G. – The point x0 is often called the base point.
We usually identify G with its image j(G) in X. – Next, we consider morphisms of
group embeddings:
1.1.4 Definition. Given a homomorphism q : G → H of algebraic groups, a morphism
ϕ : X → Y from a G-embedding X to an H-embedding Y is called
a) a q-extension if ϕ ◦ jX = jY ◦ q ;
b) q-equivariant if ϕ(gx) = q(g)ϕ(x) for arbitrary g ∈ G and x ∈ X.
Every q-extension ϕ is q-equivariant, since the equality ϕ(g ·x) = q(g)·ϕ(x) holds on
the dense open subset G of X and thus on all of X. Conversely, for an abelian group G,
if ϕ : X → Y is q-equivariant and the image ϕ(x0 ) of the base point x0 of X lies in the
big orbit H ∼= H ·y0 of Y – say ϕ(x0 ) = h0 ·y0 –, then ψ := h−10 ·ϕ is a q-extension. So
a q-equivariant morphism ϕ : X → Y is a q-extension if (and only if) it maps the base
point of X to the base point of Y .
Referring to the subsequent remarks for the notion of a “normal” variety, we introduce
the main object of the present course:
4 G.Barthel, K.-H.Fieseler, L.Kaup

1.1.5 Definition. For a torus T, a T-embedding into a normal 1 algebraic variety X is

called a (T-) toric variety.

Sometimes it is useful to use a more precise notation for a toric variety, writing a pair
(X, T) or even a triplet (X, T, xo ) instead of X.

Normal varieties
We briefly recall that an algebraic variety X is called normal if all its local rings O X,x are normal
integral domains, i.e., they are integrally closed in their respective field of fractions Q(O X,x ). A
connected normal variety is irreducible.
If X is affine, then normality is equivalent to the fact that the restriction of functions

O(X) −→ O(Xreg )

from all of X to the regular locus Xreg := X \ S(X) is an isomorphism of rings. (This is a
strong “Riemann removable singularity” property.) If furthermore X is irreducible, then X is
normal if (and only if) the ring O(X) of globally regular functions is integrally closed in its field


of fractions Q O(X) = C(X), the function field of X.
Smooth varieties are normal, since their local rings are factorial, and normal varieties are
“not too singular”: If an n-dimensional irreducible variety X is normal, then its singular locus
satisfies dim S(X) 5 n−2. For hypersurfaces, the converse holds.
A standard example of an irreducible variety that is not normal is provided by “Neil’s
parabola” X = V (C2 ; y 2 − x3 ): The rational function h = y/x ∈ C(X) satisfies the integral
equation h3 = y, but it is not regular. (The T1 -action t·(x, y) := (t2 x, t3 y) with base point (1, 1)
actually would make this singular curve a non-normal T 1 -embedding.)

1.2 Toric varieties: Basic examples

After these preparations, we proceed to discuss our main object of interest, namely, the
toric varieties. According to Remark 1.1.3, their definition sums up to the following: A
normal algebraic T-variety X is toric if and only if X has a base point x0 with trivial
isotropy and dense orbit. The embedding then is provided by the orbit map
∼
=
j = jX : T −→ T·x0 ⊂
◦ X , t 7→ t · x0 ,

where “◦⊂” means an open inclusion.

We present a few fundamental examples, most of which will be considered repeatedly
in these notes.

1.2.1 Example. The following varieties, endowed with the torus action and base point
as indicated, are toric:
1
For some problems in algebraic geometry, the normality condition is unnecessarily restrictive. Since
those problems lie outside the scope of these notes, we stick here to the “classical” definition.
Basic Toric Geometry 5

(0) The torus T, acting on itself by translation, with the natural base point (1, . . . , 1).

(1) The linear space Cn , with Tn = (C∗ )n ⊂ ◦ Cn acting by componentwise multiplication,

and the natural base point (1, . . . , 1).

(2) The (singular) two-dimensional affine quadric cone Y := V (C3 ; xz − y 2 ) with the
T2 -action (s, t)·(x, y, z) := (sx, sty, st2 z) and base point (1, 1, 1). – The T2 -action
(s, t)·(x, y, z) := (sx, ty, s−1 t2 z) on that variety yields another toric structure, de-
noted Y 0 for distinction, which is q-isomorphic to the first one (for which q ?), but
not isomorphic in the sense of Def. 1.2.4. – Normality is assured by the fact that Y
is a hypersurface with an isolated singularity.
y

Figure 1: The set of real points of Y

With regard to Figure 1, the reader should keep in mind that it does not faithfully
reflect the situation in the complex case: Whereas the punctured real part YR \ {0}
is disconnected, its complex counterpart Y \ {0} is connected.

(3) The (singular) three-dimensional “determinantal variety” Z ,→ C 2×2 consisting of

all singular 2×2 -matrices
   
x y 2×2
Z := A := ∈C ; det A = 0
z w

with the T3 -action    

x y sx ty
(s, t, u) · :=
z w suz tuw
„ «
and base point 11 11 . Normality is seen as above. – This variety Z will also
be interpreted as the three-dimensional “Segre cone”: The obvious identification
C2×2 ∼= C4 yields Z ∼= V (C4 ; xw−yz), the affine cone over the smooth projective
quadric surface in P3 that is the image of the Segre embedding of P1 × P1 .

(4) The projective n-space Pn with Tn -action t · [x] := [x0 , t1 x1 , . . . , tn xn ] and base point
[1, . . . , 1]. This is the most basic example of a compact toric variety.
We remark that compact toric surfaces are always projective, whereas higher-dimen-
sional compact toric varieties in general are not.
6 G.Barthel, K.-H.Fieseler, L.Kaup

The two-dimensional quadric cone Y actually is the lowest-degree member in an infinite fam-
ily of singular two-dimensional hypersurfaces in C 3 that are toric varieties, playing an important
role in the theory of surface singularities:

1.2.2 Remark. For every integer k = 2, the variety Y k := V (C3 ; xz − y k ), with the T2 -action
(s, t)·(x, y, z) := (sx, sty, sk−1 tk z) and the natural base point (1, 1, 1), is toric. In the literature,
the singularity at the origin of Yk is called a “rational double point of type A k−1 ”.
With respect to the projection onto the (x, z) plane, the surface Y k is a finite covering
branched along the coordinate axes – such coverings occur during the resolution of arbitrary
singular surfaces. We note that Yk is a “cyclic quotient singularity”: The cyclic group C k of
k-th roots of unity acts on the plane C 2 as a subgroup of SL2 (C) via ζ·(u, v) := (ζu, ζ k−1 v). The
quotient variety C2 /Ck is a normal surface. The map C2 → C3 , (u, v) 7→ (uk , uv, v k ) given by
∼
=
invariant polynomials induces an isomorphism C 2 /Ck −→ Yk (see also Example 3.1.10). – The
restriction of this map to R2 yields a parametrization of (Yk )R = Yk ∩ R3 if k is odd.

Figure 2: Three views of the set (Y3 )R with the real A2 surface singularity

1.2.3 Remark. There are some natural ways of constructing new toric varieties from
given ones:
(1) Every nonempty open T-invariant subset of a toric variety is itself toric.
(2) A finite product of toric varieties is again toric (with respect to the direct product
of the involved tori).
(3) Let X be a T-toric variety and G, a closed subgroup of the torus T. The residue
class group T/G is again a torus of dimension n − dim G, see Remark 1.3.4 (4). This
quotient torus acts on the topological orbit space X/G of the induced G-action on X.
The embedding T ⊂ ◦ X induces an open T/G-equivariant inclusion T/G ⊂ ◦ X/G.
If G is finite, then X/G has a natural T/G-toric structure that makes the projection
X  X/G a toric morphism. This is discussed at the end of subsection 2.3 when X
is affine; using Sumihiro’s theorem 1.2.5, the general case then follows by a natural
guing procedure.
If G is not finite, then the G-orbit G·t of a point t ∈ T is closed in T, but it may fail
to be closed in X; see Example 2.3.12 for three typical subgroups G ∼ = C∗ of T2 acting
on X = C2 . In that case, the topological orbit space is not separated; in particular, it is
Basic Toric Geometry 7

not an algebraic variety! To obtain a “categorical quotient”, a more involved approach is

needed, since such a quotient morphism identifies different orbits if their closures intersect.
For the affine case, we discuss some aspects of the “algebraic quotient” in the paragraph
on quotients at the end of subsection 2.3. The general situation lies outside the scope of
these notes.

Morphisms of toric varieties are defined as in 1.1.4:

1.2.4 Definition. Let q : T → T 0 be a homomorphism of algebraic tori, and (X, T, x0 )

and (X 0 , T 0 , x00 ) be toric varieties. Then a base point preserving q-equivariant morphism
(i.e., a q-extension) is called a q-toric morphism. – In the case T = T 0 and q = idT , we
simply speak of a toric morphism.

The theory of toric varieties heavily relies on the following result:

1.2.5 Theorem (Sumihiro). Every point in a (normal) toric variety admits an affine
open T-invariant neighbourhood.

Thus, in order to analyse arbitrary T-toric varieties, it suffices to consider affine T-

toric varieties, what we shall do in section 2, and then to study how they can be patched
together, see section 3.2.
Without assuming normality, the conclusion of Sumihoro’s theorem is no longer valid:


1.2.6 Example. The binary cubic forms −4tu(t + u), −4tu(t − u), (t + u)3 define a morphism


P1  C ,→ P2 onto the projective nodal cubic curve C = V P2 ; y 2 z − x2 (x + z) . The map
is injective except for identifying the points 0 := [0, 1] and ∞ := [1, 0], and it induces an
isomorphism P1 /(0 ∼ ∞) ∼ = C. The C∗ -action s · [t, u] := [st, u] on P1 thus defines a almost
transitive algebraic action of the 1-torus on that (non-normal) projective curve with one big
orbit and the single fixed point [0, 0, 1], so the fixed point does not have an invariant affine
neighbourhood.

1.3 Characters and one-parameter subgroups of tori

In the study of toric varieties, algebraic group homomorphisms between the acting torus
and C∗ play a key role. This starts with the following easy but crucial fact: Every algebraic
group endomorphism of the algebraic 1-torus T1 = C∗ is of the form s 7→ sk with a unique
integer k ∈ Z. The resulting canonical group isomorphism Hom(C∗ , C∗ ) ∼ = Z sending idC∗
to 1 can be generalized in two ways:

1.3.1 Definition. Let Tn ∼= (C∗ )n be an algebraic n-torus. A homomorphism of algebraic

groups χ : Tn → C∗ is called a character of Tn , and a homomorphism λ : C∗ → Tn is
called a one-parameter subgroup.

With respect to the argumentwise multiplication, the sets

(1.3.1.1) X(Tn ) := Hom(Tn , C∗ ) and Y(Tn ) := Hom(C∗ , Tn )

8 G.Barthel, K.-H.Fieseler, L.Kaup

are abelian groups. The canonical group isomorphism X(C∗ ) = Y(C∗ ) ∼

= Z is generalized
as follows:

1.3.2 Remark. The set X(Tn ) is a lattice (i.e., a free abelian group) of rank n : In fact,
the mapping

 Qn 
M := (Zn , +) −→ X(Tn ), · , µ 7−→ χµ : t 7→ tµ := tµi i
i=1

(with coordinates µ = (µ1 , . . . , µn ) ∈ Zn and t = (t1 , . . . , tn ) ∈ (C∗ )n ) is an isomorphism

of abelian groups. Hence, every character of T is a Laurent monomial in the coordinate
functions (i.e., the basis characters) t1 , . . . , tn on Tn . The Laurent algebra generated by
these monomials is the coordinate ring of the torus as an affine algebraic variety, i.e.,
M
(1.3.2.1) O(Tn ) = C[t1 , t−1 1 , . . . , t ,
n nt −1
] = C·χ .
χ∈X(Tn )

Dually, the set Y(Tn ) is a lattice of rank n, too: There is an isomorphism

N := Zn −→ Y(Tn ) , ν = (ν1 , . . . , νn ) 7−→ λν : s 7→ (sν1 , . . . , sνn ) .

By a slight abuse of terminology, we occasionally call M and N the lattice of char-

acters and of one-parameter subgroups, respectively. – Each of these lattices M and N
determines the torus T: Using the canonical Z-module structure on the abelian group C∗ ,
there are functorial isomorphisms

(1.3.2.2) = HomZ (M, C∗ ) and T ∼

T ∼ = N ⊗Z C∗ =: TN .
∼
= ∼
=
1.3.3 Remark. Via the isomorphisms M = Zn −→ X(T) and N = Zn −→ Y(T) provided
by the fixed identification Tn ∼
= (C∗ )n , the composition pairing

(1.3.3.1) X(T) × Y(T) −→ Hom(C∗ , C∗ ) , (χµ , λν ) 7−→ hχµ , λν i := χµ ◦ λν

corresponds to the usual inner product

n
X
(1.3.3.2) h , i : M × N −→ Z , (µ, ν) 7−→ hµ, νi := µi ν i ,
i=1
i.e.,
(1.3.3.3) (χµ ◦ λν )(s) = shµ,νi holds for every s ∈ C∗ .

We use the same symbol for the extended dual pairing

h , i : MR × NR −→ R

of real vector spaces, where for a lattice L ∼

= Zn , we set

(1.3.3.4) LR := L ⊗Z R ∼
= Rn .
Basic Toric Geometry 9

Moreover, for the dual pair of standard lattice bases in M and in N and thus, dual
vector space bases of MR and NR , we shall use this notation:
(1.3.3.5) (e1 , . . . , en ) in M , and (f1 , . . . , fn ) in N .
To better understand the mutual relations between the torus T and the lattice N expressed
in formulae (1.3.1.1) and (1.3.2.2) as well as the structure of closed subgroups of the torus, it is
helpful to use an intermediate “analytic” object. 2
exp
1.3.4 Remark. (1) We use the exact “exponential sequence” 0 → Z → C −→ C∗ → 1 (where
exp(z) := e2πiz ). “Tensoring” with the lattice N yields a new exact sequence
exp
(1.3.4.1) N −→
0 −→ |{z} N −→ TN −→ 1T ,
| {zC} |{z}
∼
=Zn :=N ⊗Z C∼ ∼
=C n =(C∗ )n

where the function exp is applied componentwise, and 1 T is the unit element of T := TN .
Occasionally, we interpret NC as tangent space T1 (T) of the torus at the neutral element.
The exact sequence immediately provides identifications
N = ker (exp : NC  T ) and T ∼
= NC /N .

In this setting, the isomorphism N ∼

= Y(T) can be seen as follows: There is a canonical identification
∼
N = Hom(Z, N ), with ν ∈ N corresponding to the map Z → N , 1Z 7→ ν and conversely. Scalar
extension of lattice homomorphisms then yields a natural identification of N with the group


Hom (C, Z), (NC , N ) of vector space homomorphisms respecting the given lattices. Passing to
the quotient modulo these sublattices, such a linear homomorphim then uniquely “descends” to a
homomorphism of tori C∗ → T. The inverse homomorphism Y(T) → N is obtained by “lifting”
a one-parameter subgroup λ ∈ Y(T) to such a vector space homomorphism λ e = dλ : (C, Z) →
(NC , N ).

(2) Any homomorphism q : T0 → T of tori lifts to a vector space homomorphism qe =:

dq : NC0 → NC that respects the lattices and hence induces a lattice homomorphism
qe: N 0 → N . We thus have a commutative “ladder”
exp
0 −−−−→ N 0 −−−−→ NC0 −−−−→ T0 −−−−→ 1
  

qe|N 0 y

qey=dq

qy
(1.3.4.2)
0 −−−−→ N −−−−→ NC −−−−→ T −−−−→ 1 .
exp

In the “differential” interpretation, the map dq actually is the derivative of q at 1 T0 .

Conversely, to any lattice homomorphism ϕ : N 0 → N corresponds a homomorphism
T(ϕ) := ϕ ⊗ idC∗ : TN 0 → TN of tori. After a choice of bases, the map ϕ is explicitly
represented by a matrix A = (aij ) ∈ Zn×m . In the corresponding coordinates for the tori,
the homomorphism T(ϕ) := ϕ ⊗ idC∗ is given by
Y
m 
∗ m ∗ n aij
T(ϕ) : (C ) −→ (C ) , (t1 , . . . , tm ) 7−→ tj .
j=1 i=1,...,n

2
From the “categorical” point of view, this is quite natural since a lattice of positive rank, being an
infinite discrete group, is not an algebraic group, but rather an analytic group.
10 G.Barthel, K.-H.Fieseler, L.Kaup

(3) The pair of covariant functors N T N and T Y(T) ∼= N actually establishes an

equivalence of categories. One should note, however, that the behaviour of morphisms is
not quite straightforward since the functor N T N is right-exact only: For the inclusion
0
ι : N ,→ N of a sublattice, the resulting homomorphism T(ι) : T N 0 → TN is injective if
and only if N 0 is a saturated sublattice (i.e., if N/N 0 is torsion-free and thus, free: In that
case, there is a one-sided inverse to ι that, upon tensoring with C ∗ , provides a one-sided
inverse on the level of tori). In particular, if rank(N 0 ) = rank(N ), then the homomorphism
T(ι) is surjective with the finite abelian group N/N 0 as kernel. Analogously, the inverse
T N is left-exact only.

Readers with an interest in “categorical” aspects might wonder why this behaviour does not
contradict the equivalence property. Looking closer, one notes a subtle difference between the two
categories: both are additive, neither of them is abelian, but they fail “on different sides”. In fact,
in the category of finitely generated lattices, morphisms do not always have cokernels, whereas in
the category of tori, morphisms do not always have kernels.

(4) Any closed subgroup G of T can be diagonalised: There is an isomorphism of tori trans-
Q Q

forming the pair (T, G) into a product ni=1 (C∗ , Gi ) := (C∗ )n , ni=1 Gi , where each Gi , a
closed subgroup of the one-torus C∗ , is either C∗ or a finite (cyclic) group of roots of unity.
To obtain such a diagonalization, consider the inverse image exp −1 (G) in NC . It splits
(non-canonically) into a direct sum V ⊕ L of the vector subspace V := exp −1 (G◦ ) ∼ = T1 G
◦
and a “transversal” lattice L, where G denotes the connected component of the identity
in G. The lattice L spans a vector subspace L C of NC that is complementary to V , and
L0 := N ∩ LC is included in L as a sublattice of finite index. Splitting N ∩ V into sub-
lattices of rank one, applying the structure theorem for subgroups of finitely generated
free abelian groups to the pair of lattices (L, L 0 ), and then passing to the image in T then
yields a diagonalization. This immediately implies that the residue class group T/G is a
torus, as stated in Remark 1.2.3 (3).
If G is finite, then its inverse image exp −1 (G) is discrete, so it consists only of the lattice L.
The latter includes N , and the exponential mapping induces an isomorphism L/N ∼ = G.
0
Conversely, given the inclusion ι : N ,→ N of a sublattice of finite index, the corresponding
surjective homomorphism T(ι) : TN 0 → TN of tori identifies TN with the quotient TN /G
by a finite subgroup G ∼ = N/N 0 (see also (2) and (3) above).
For the study of quotient structures, it is useful to have an alternative approach to the
closed subgroups of a torus T: Associating to G the sublattice K := {µ ∈ M ; χ µ |G = 1},
T
and to such a sublattice K ⊆ M the subgroup G := µ∈K ker(χµ ), establishes a one-to-one
correspondence.

We illustrate the correspondence N ←→ T N on the level of morphisms with an example.

The idea will be used again several times (see 2.2.13, 2.2.17, and 2.2.19) since it is essential for
one of our standard examples:

1.3.5 Example. We consider the standard lattice N = Z 2 with the standard basis f1 , f2 , and
e spanned by v1 = 2f1 − f2 and v2 = f2 . With respect to this basis, the inclusion
the sublattice N


e ,→ N is given by the matrix 2 0 . Under the basis isomorphism N
ι: N e ∼
= Z2 , the torus
−1 1
Basic Toric Geometry 11

TNe = N e ⊗Z C∗ is identified with (C∗ )2 by sending v1 ⊗ s to (s, 1), and v2 ⊗ t to (1, t). Thus,
with the standard identification TN ∼ = (C∗ )2 , the associated morphism of tori takes the form
T(ι) : (C∗ )2 → (C∗ )2 , (s, t) 7→ (s2 , t/s), with kernel ±(1, 1).

2 Affine Toric Varieties

2.1 Algebraic description: The coordinate ring
An affine variety X is completely determined by its ring O(X) of regular functions. If X
is toric, then the restriction of global regular functions to the (open dense) embedded
torus T provides an injective algebra homomorphism from O(X) into O(T), the Laurent
monomial algebra of (1.3.2.1). We may thus identify O(X) with a subalgebra of the latter:
M
(2.1.0.1) O(X) ∼ = O(X)|T ⊂ O(T) = C·χ .
χ∈X(T)

The characters of T that extend to a regular function on X – and thus, are elements
of O(X) – play a key role in the study of the coordinate ring. Evidently, the set

(2.1.0.2) S = SX := O(X) ∩ X(T)

of these characters is a (multiplicative) submonoid of the character group. We first study

its role for the vector space structure of O(X):

2.1.1 Lemma. The set S = SX provides a vector space basis of the coordinate ring:
M
(2.1.1.1) O(X) = C·χ .
χ∈S

Proof. The torus action on X induces an action on the coordinate ring:

T × O(X) −→ O(X) , (t, f ) 7−→ f t where f t (x) := f (t·x).

P
By (2.1.0.1), each non-zero function f ∈ O(X) can uniquely be written as f = ri=1 λi χi
with distinct characters χi ∈ X(T) and non-zero complex coefficients λi . Applying the
torus action yields
Xr
ft = λi ·χi (t)·χi ∈ O(X) for every t ∈ T .
i=1
Since the characters χ1 , . . . , χr ∈ O(T) are linearly independent, we find points u1 , . . . , ur


in T such that the matrix χi (uj ) is nonsingular. Hence, each λi χi and thus, each χi lies
in the span of the functions f uj ; consequently, all χi belong to O(X).

We now study an additive description of O(X) using the corresponding subset of

“exponents”

(2.1.1.2) E = EX := {µ ∈ M ; χµ ∈ SX }
12 G.Barthel, K.-H.Fieseler, L.Kaup

in M = Zn . Since SX is a monoid, this set EX is a sub-semigroup of M , so we can describe

O(X) as the semigroup algebra:
M
(2.1.1.3) C[EX ] := C·χµ = O(X) .
µ∈E

We list some essential properties:

2.1.2 Remark. (1) 0 ∈ E;

(2) E is finitely generated;

(3) E generates M as a group;

(4) E is a saturated sub-semigroup of M : If kµ ∈ E holds for some k ∈ N>1 and µ ∈ M ,

then µ ∈ E.

Proof. (1) holds, since 1 = χ0 ∈ O(X).

(2) This assertion is true since O(X) is a finitely generated C-algebra that is spanned
P
by characters. Hence, we find elements µ1 , . . . , µr ∈ M with3 E = ri=1 N·µi .
P
(3) We have to verify that E + (−E) = M . To that end, let χ := χµ with µ := ri=1 µi
1 r
as in the proof of (2). Since none of the generators χµ , . . . , χµ of SX = X(T)∩O(X)
has a zero on the principal open subset U := Xχ of X, every character in SX even
yields an (invertible) function on U . In particular, this implies that the character
1 r
χ−1 = (χµ · · · χµ )−1 belongs to O(U )× , and since O(U ) = O(X)[χ−1 ], this yields
SU := O(U ) ∩ X(T) ⊂ O(U )× . It now suffices to show that U is just the embedded
torus T of X, since then O(U ) = C [E + N·(−µ)] agrees with O(T) = C [M ], so

M = E + N·(−µ) ⊆ E + (−E) ⊆ M .
The (open) inclusion T ⊂ U being obvious, we have to verify that the complementary
subset Z := U \T of U is empty. This complement being a proper closed T-invariant
subset, its vanishing ideal I(Z) is non-zero and spanned by characters in SU . We
have seen that such characters are invertible functions on U . Hence, the ideal I(Z)
is the unit ideal in O(U ), and thus Z is empty.


(4) If some power (χµ )k = χkµ of a character χµ ∈ X(T) ⊂ Q O(X) lies in O(X), then
χµ is integral over this normal ring and thus lies in it.

For any affine variety X, the generators of its coordinate ring O(X) are just the
components of a closed embedding X ,→ Cr and vice versa. In the toric case, we may
thus characterize the generators of the semigroup of exponents:
3
We adopt the convention that N := N=0 = Z=0 .
Basic Toric Geometry 13

2.1.3 Remark. Let X be an affine toric variety, and fix µ1 , . . . , µr ∈ EX with corre-
i
sponding character functions χi := χµ ∈ SX . Then
Xr
EX = N·µi ,
i=1
1 r
i.e., µ , . . . , µ generate the semigroup EX , if and only if the morphism


(2.1.3.1) X −→ Cr , x 7−→ χ1 (x), . . . , χr (x)

is a closed embedding (see also Lemma 2.3.1). In that case, the T-action extends to the
ambient space Cr in the form


(2.1.3.2) t·(z1 , . . . , zr ) = χ1 (t)·z1 , . . . , χr (t)·zr .

On the other hand, if the equivariant morphism (2.1.3.1) given by the vectors µ1 , . . . , µr
is a closed embedding, then these vectors generate EX as a semigroup.

We apply this remark to our standard examples (see also Remark 2.2.6):

2.1.4 Example. Using the numbering of 1.2.1, we obtain:

(1) The semigroup EX for X = Cn is generated by e1 , . . . , en .
(2) The semigroup EY for the quadric cone Y is generated by e1 , e1 +e2 , e1 +2e2 .
More generally, for the two-dimensional toric hypersurfaces Y k (with k = 2) discussed in
Remark 1.2.2, the semigroup EYk is generated by e1 , e1 +e2 , (k−1)e1 +ke2 .

p p
q q pppqu u q u q u q u q u q u q q q pppq q q qu u q u q u q
p p
q q pppqu u q u q u q u q u q u q q q pppq q q qu u q u q u q
p p
q q pppqu u q u q u q u q u q u q q q pppq q qu u q u q u q u q
p p
q q pppqu u q u q u q u q u q u q q q pppq q qu u q u q u q u q
p p
q q pppqu u q u q u q u q u q u q q q pppq i qu u q u q u q u q u q
p p
q q pi pppqu u q u q u q u q u q u q q q pppq i qu u q u q u q u q u q
p
p pq p p p pq p p p ppu ppq p p p pu
q p p p pu
i q p p p pu
q p p p pu
q p p p pu
q p p p pu
qp p p pq p p p pq p p p ppu
ppq p p p pu
q p p p pu
i q p p p pu
q p p p pu
q p p p pu
q p p p pu
qp p
q q pppq q q q q q q q q pppq q q q q q q
p p
q q pppq q q q q q q q q pppq q q q q q q

Figure 3: Generators and points of EC2 and of EY

(3) The semigroup EZ for the Segre cone Z is generated by e1 , e2 , e1 + e3 , e2 + e3 .

(See Figure 7 in Example 2.2.7: these generators are the four points on the first
cross-section.)

2.2 Geometric description: Polyhedral lattice cones

We now prepare for the geometric description of toric varieties in terms of cones and fans
as announced in the introduction.
14 G.Barthel, K.-H.Fieseler, L.Kaup

(I) Recollection: Polyhedral cones

The semigroup of exponents E is the set of lattice points E = γ ∩ M in a suitable n-
dimensional “lattice cone” γ in MR . We briefly recall the general notion of a (lattice)
cone:
2.2.1 Definition. For a lattice L ∼
= Zn , let γ be a subset of V := LR .

(1) The set γ is called a polyhedral cone if there are finitely many vectors v 1 , . . . , vr
in V such that r
X
γ = R≥0 ·vi =: cone(v1 , . . . , vr ) .
i=1
These vectors are called spanning vectors or generators of the cone. A cone spanned
by a single non-zero vector v is called a ray, denoted by
ray(v) := cone(v) .
(2) A polyhedral cone γ is called an L-cone (or lattice cone) if its spanning vectors
v1 , . . . , vr can be chosen in the lattice L.
(3) A polyhedral cone γ is called strongly convex (or pointed) if it does not include
a line through the origin.
(4) A subset δ of a polyhedral cone γ is called a face, denoted δ  γ, if it is of the form
δ = γ ∩ {v ∗ = 0} for some linear form v ∗ ∈ V ∗ with v ∗ |γ ≥ 0 .
























































 




 


 

 

 

 

 

 

 

 

 

 

 

 

 













































  





  









 


 


 


 


 


 


 


 


 


 


 


 






















































 

 



  

 



   



 

 

 

 

 

 

 

 

 

 

 

 

 





































    

       

       

   

 

 

 

 

 

 

 

 

 

 

 


 








































 

       

 

       

 

       

 

 

 

 

 

 

 

 

 

 

 

 








































   

      

   

   

      

   

   

      

   



 


 


 


 


 


 


 


 


 


 


 

 










































           






           






           




 

 

 

 

 

 

 

 

 

 

 

 
 
 



































                                 


 


 


 


 


 


 


 


 


 


 


 


 

 

 







































      

        

  

      

        

  

      

        

  


 

 

 

 

 

 

 

 

 

 

 

 


 


 




































 





















   

           






   

           






   

           








 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
 
 

 
  

 
 
 
 

 






 
 




 


 
  
 





  
  
 





  
  


  


 

  


 

  


 

  


 

  


 

  


 

  


 

  


 

  


 

  


 

  


   


  
 
 
 





 

 



 

 

 



 

 

 

 

 






 


 





 


 





 


 





 


 





 


 





 


 





 


 





 


 





 


 





 


 





  





  
 
 
 


 
 

 






 







 







 
















 








 








 








 








 








 








 








 








 








 





 





 
   
 

 




 


  






 




 


  




 




 


  




 




  


 










 








 








 








 








 








 








 








 








 








 





 





   




 

 









 







 













 


 

 


  
 


  
 


  
 


  
 


  
 


  
 


  
 


  
 


  
 


  
 


   


    
  

 
 






 


 


 


 


 


 


 


 


 


 


 


 


 


 
   
 

Figure 4: A strongly convex polyhedral cone and its faces
Since v ∗ = 0 is admissible, each cone is a face of itself. We use the notation δ  γ
if we want to emphasize that a face δ is proper. – For future use, we introduce some
conventions:
2.2.2 Convention. All cones to be considered in the sequel will be polyhedral, and
usually, they are assumed to be lattice cones. A system of spanning vectors v1 , . . . , vr for
a cone γ is usually assumed to be irredundant. Moreover, if γ is a lattice cone, then each
such vector vi is usually assumed to be a primitive lattice vector, i.e., not a non-trivial
positive integer multiple of another lattice vector.
Basic Toric Geometry 15

We add a few remarks and introduce some additional notions and notations:
2.2.3 Remark. A polyhedral cone γ as in 2.2.1 (1) can be equivalently described as finite
intersection of closed half spaces
s
\
γ= {vj∗ ≥ 0} with suitable linear forms vj∗ ∈ V ∗ .
j=1

The boundary hyperplanes {vj∗ = 0} with vj∗ 6= 0 are called supporting hyperplanes of γ.
– In the case of a lattice cone, the linear forms vj∗ can be chosen as vectors of the dual
lattice Hom(L, Z). Hence, each face of a lattice cone is again a lattice cone.
The intersection γ ∩ (−γ) is the largest linear subspace included in γ. In particular,
γ ∩ (−γ) is the zero cone
o := {0}
if and only if γ is strongly convex.
A proper face δ of γ is cut out by a supporting hyperplane. The relative interior γ ◦ is
the set of all points in γ not included in a proper face. – The dimension of γ is defined as

dim γ := dim lin(γ) , where lin(γ) = γ + (−γ)

is the linear subspace of V spanned by gamma. A cone of dimension d is usually called

a d-cone; a (d−1) -face of γ is called a facet, and and a 1-face, an edge. As a notational
convention, we mostly use symbols like σ, τ, % for N -cones, and γ, δ etc. for M -cones.

(II) “Contravariant” description

For the next results, we need some more notation:
2.2.4 Notation. For a ring R, we denote by Sp(R) its maximal spectrum. Furthermore,
we denote by
ATVT the category of affine T-toric varieties with (idT -)toric morphisms,
CL for a lattice L, the category of L-cones (in LR ) and their inclusions,
CL,d the full subcategory of d-cones,
SCL and SCL,d the respective full subcategories of strongly convex cones.
We may now start the construction of the bridge between the algebraic geometry of
toric varieties and the elementary real convex geometry of cones that will be provided by
the Equivalence Theorem 2.2.11.
2.2.5 Anti-Equivalence Theorem. For an n-torus T with corresponding group M of
exponents for X(T), the assignment


CM,n −→ ATVT , γ 7→ X γ := Sp C[γ ∩ M ]

is an anti-equivalence of categories.
16 G.Barthel, K.-H.Fieseler, L.Kaup

Indication of proof. On the one hand, for an n-cone γ = cone(µ1 , . . . , µr ) with primitive
generators µi , one verifies that E := γ ∩ M shares the properties of Remark 2.1.2 and that
this fact guarantees that X γ actually is a toric variety: The semigroup E is generated
P r i

by the finite set P ∩ M , where P := i=1 t i µ ; 0 5 t i 5 1 for i = 1, . . . , r is the
“fundamental polytope” of the cone γ (see the left-hand side of Figure 5 in 2.2.6).
Furthermore, E + (−E) = M (this guarantees that T → X γ is an open embedding),
since for every sufficiently “long” vector µ ∈ M ∩ γ ◦ we have µ+ei ∈ E for i = 1, . . . , n.
T
Finally, we may write γ = sj=1 Hj , an intersection of closed half spaces Hj = {ν j ≥ 0}
in MR with suitable ν j ∈ N ∼ = M ∗ . To this corresponds a description
s
\
C [γ ∩ M ] = C [Hj ∩ M ] .
j=1


Since the subrings C [Hj ∩ M ] ∼ = O C × (C∗ )n−1 of the Laurent algebra C[M ] are normal,
so is their intersection.
On the other hand, to an affine toric variety X corresponds a semigroup of exponents
EX as in (2.1.1.2) with finitely many generators, say µ1 , . . . , µr , in M . Then, for the
cone γ := cone(µ1 , . . . , µr ), the variety X is isomorphic to X γ . For more details, see
Lemma 2.3.1.

Whereas the above elements µ1 , . . . , µr of E also generate the corresponding cone γ,

the converse need not be true, even if the generating vectors of the cone are primitive; cf.
example 2.2.7 (2).
2.2.6 Remark. For the semigroup Eγ := γ ∩ M cut out by a strongly convex cone γ, there is a
canonical minimal system of generators, sometimes called a Hilbert basis: It is the set Ė \(Ė + Ė)
of indecomposable elements in E, with Ė := E \ {0}.
In the two-dimensional case, it consists of those primitive lattice vectors µ i which lie on
the boundary of the (unbounded) “polyhedron” K := conv( Ė). This can be seen as follows:
According to the regularity criterion in Corollary 3.1.6, two neighbouring points µ i , µi+1 generate
a regular cone, so each lattice point in this cone is a linear combination of µ i and µi+1 .

b
b
sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp  sp b
b
sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp  sp
p bp ps ps ps ps ps ps spg sp sp sp sp sp sp sp sp sppp p bp ps ps ps ps ps sp sp sp sp sp sp sp sp sp sp sppp
p p b psgbpsg ps ps ps ps ps ps spp p p p p b p ps ps ps ps ps ps ps ps ps ps ps ps spp p p
p bp ps ps pspsg b p b
p p p p b spsg psg psg psg b psg ps ps pspp p p p
psg b p p p p pbpg sp p p ps ps ps ps ps ps ps ps pspp p p p
p p p p p pHppH
p pg
p p p p p p b b s psg psg psg psg b sp p p p p p p p p pg
sp p ps ps ps ps  
p p p p p p p pH
p pg psg pg ps p p pg
s p p p p p p
p p p p p p p b p b p pg s pgs pgspp p p p p p p p pH
p p p pg s 
sp pg sp p p p pp p p p p p p p p
pg
p p p p p p p p p b p pg 
s p p p p p p p p p p p p p p p p p p ppppppppp p p p p p p p p
p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p

Figure 5: Fundamental cell Pγ and polyhedron Kγ for γ = cone(−5e1 +3e2 , 3e1 +2e2 )

We illustrate the correspondence X = X γ ←→ γ between affine toric varieties and

M -cones with our basic examples from 2.1.4:

2.2.7 Example. (1) The linear space X = Cn corresponds to γ = cone(e1 , . . . , en ).

Basic Toric Geometry 17

(2) The affine quadric cone Y corresponds to γ = cone(e1 , e1 +2e2 ) .

More generally, the toric surface Y k (with k = 2) discussed in Remark 1.2.2 corresponds


to γ = cone e1 , (k−1)e1 +ke2 .

p
q q s s s s s s s q q pppq q q
sq sq sq sq
q q pppq q q
sq
p
q q s s s s s s s sq sq sq
q q pppq q sq
sq
p
q q s s s s s s s sq sq sq
p

q q s s s s s s s q q pppq q
sq sq sq sq sq
pp
q q s s s s s s s q q ppq
sg q sq sq sq sq sq
p
q q sg s s s s s s q q pppq
sg q sq sq sq sq sq
p
p qp p p p pq p p p psp sg s s s s s p pq p p p pq p p p ppspq
g
sq sq sq sq sq sq
p p
q q qppp q q q q q q q q pppq q q q q q q
p p
q q qppp q q q q q q q q pppq q q q q q q

Figure 6: The M -cones cutting out EC2 and EY (cf. Figure 3)

In Figure 6, the encircled dots represent a Hilbert basis of the pertinent semigroups
EC2 and EY .
(3) The Segre cone Z corresponds to γ = cone(e1 , e2 , e1 + e3 , e2 + e3 ) with generators
forming a Hilbert basis (see Figure 7).

Figure 7: The M -cone γ = cone(e1 , e2 , e1 +e3 , e2 +e3 ) cutting out EZ

(III) Dual cones and the “covariant” description

For a truly elegant and powerful geometric description of toric varieties, it is essential to
complement the correspondence of the Anti-Equivalence Theorem 2.2.5 with a “covariant”
version, which is obtained by dualization of cones.
For a lattice L, its dual lattice is defined as L∗ := HomZ (L, Z):

2.2.8 Definition. For an L-cone γ, its dual is the cone


γ ∨ := u ∈ (L∗ )R ; hu, γi = 0 ,

where hu, γi ≥ 0 means that hu, vi ≥ 0 holds for every v ∈ γ.

18 G.Barthel, K.-H.Fieseler, L.Kaup

p p ppp p p ps sp sp sp p p cppp cp cp cp cp cp cp
p p pppp p p

ps sp sp sp p p cpppp cp cp cp cp cp cp
p p pppp p ps
sp sp sp sp p p cpppp cp cp cp cp cp cp
p p pppp p
ps sp sp sp sp p p cpppp cp cp cp cp cp cp
p p pppp pf s
sp sp sp sp sp p p cpppp cp cp cp cp cp cp
p p pppp
psf sp sp sp sp sp p p f cpppp cp cp cp cp cp cp
p p p p p p p p p ppspp
p p p pf p p p p p p p p p ppcpp p p p pcfp p p pcp p p pcp p p pcp p p pcp p p pcp p
p p ppppHH
sp p p psp p p psp p p psp p p psp p p psp p
p p pppp p p p p p p
p p pppp p H
p cf p cp cp cp cp
p p pppp p p p p p p p H
H
p cp cp cp

Figure 8: A pair of dual cones γ, γ ∨ for dual lattices L, L∗

2.2.9 Remark. The dualization of cones enjoys the following properties:

(1) The dual of an L-cone is an L∗ -cone, so dualization defines a map CL → CL∗ .
(2) Dualization is inclusion-reversing, i.e., δ ⊆ γ implies δ ∨ ⊇ γ ∨ .
(3) Dualization is involutive, i.e., (γ ∨ )∨ = γ.
(4) If γ is a linear subspace, then γ ∨ = γ ⊥ holds.
(5) (δ+γ)∨ = δ ∨ ∩ γ ∨ and (δ ∩ γ)∨ = δ ∨ + γ ∨ .


(6) dim σ ∨ = n − dim σ ∩ (−σ) .

In particular, according to properties (3) and (6), there is a one-to-one correspondence

between the objects of the category SCL of strongly convex L-cones, and of the category
CL∗, n of full-dimensional L∗ -cones. Applying this observation to the dual lattices M
and N , we obtain:

2.2.10 Corollary. The dualization of cones

SCN ←→ CM,n , σ ←→ σ ∨

is an anti-equivalence of categories.

Combining this result with the Anti-Equivalence Theorem 2.2.5, we achieve the con-
struction of the bridge, which is fundamental for the toric geometry:

2.2.11 Corollary (Equivalence Theorem). The functor

∨

SCN −→ ATVT , σ 7−→ Xσ := X σ = Sp C[σ ∨ ∩M ]

is a (covariant) equivalence of categories,

2.2.12 Convention. In view of the above correspondence, all N -cones considered in the
sequel will usually be assumed to be strongly convex. The symbol σ will denote such
a cone.

We illustrate the correspondence X = Xσ ←→ σ between affine toric varieties and

N -cones again with our basic examples of 2.1.4.
Basic Toric Geometry 19

2.2.13 Example. (1) The linear space X = Cn corresponds to σ = cone(f1 , . . . , fn ).

(2) The affine quadric cone Y corresponds to σ = cone(2f1 −f2 , f2 ) (see Figure 9).
More generally, the toric surface Y k (with k = 2) discussed in Remark 1.2.2 corresponds


to σ = cone kf1 −(k−1)f2 , f2 .

q q s s s s s s s q q spppq sq sq sq sq sq sq
p
q q s s s s s s s q q spppq sq sq sq sq sq sq
p
q q s s s s s s s q q spppq sq sq sq sq sq sq
p
q q s s s s s s s q q spppq sq sq sq sq sq sq
p
q q s s s s s s s q q spppq sq sq sq sq sq sq
p
q q sg s s s s s s q q g spppq sq sq sq sq sq sq
p
p qp p p p pq p p p psp sg s s s s s p pq p p p pq p p p ppspHq p p p psg
q p p p psq p p p psq p p p psq p p p psq p p p psq p p
q q pppq H
p p
q q qppp q q q q q q q H g
sq sq sq sq sq
q q pppq q q H
p p
q q qppp q q q q q q q Hsq sq sq
H

Figure 9: The N -cones for C2 and Y

(3) The Segre cone Z corresponds to σ = cone(f1 , f2 , f3 , f1 +f2 −f3 ) .

Let us also give an example for the correspondence of morphisms in Corollary 2.2.11
that at the same time exhibits the important role played by faces:

2.2.14 Example. To the face relation τ  σ of N -cones corresponds an open embedding

Xτ → Xσ of affine toric varieties.

Proof. According to Remark 2.2.9, the “face equality” τ = σ ∩µ⊥ (with some µ ∈ σ ∨ ∩M )
translates into τ ∨ = σ ∨ + R · µ = σ ∨ + ray(−µ). This in turn implies that O(Xτ ) =
O(Xσ )[χ−1 ], where χ := χµ . As a consequence, the morphism Xτ → Xσ decomposes into
an isomorphism Xτ ∼ = (Xσ )χ and the open inclusion of the principal open subset (Xσ )χ
into its ambient variety Xσ .

So far, we mainly have been interested in full-dimensional N -cones. Fortunately, their

investigation is essentially sufficient for the general theory. This fact, to be applied time
and again, is a consequence of the following result, where we use this notation:

(2.2.15.0) Nσ := N ∩ lin σ ,→ N and Tσ := Nσ ⊗Z C∗ ,→ T = N ⊗Z C∗ .

2.2.15 Proposition (Product decomposition). For a d-dimensional N -cone σ, there

exists a complementary subtorus Tn−d of Tσ in Tn and a Tσ -toric variety Zσ , such that

(2.2.15.1) T ∼
= Tσ × Tn−d and Xσ ∼
= Zσ × Tn−d ,

endowed with the product action, where Tn−d acts on itself by translation.
20 G.Barthel, K.-H.Fieseler, L.Kaup

Proof. Since σ is an N -cone, the lattice Nσ is a saturated sublattice of rank d in N .

Hence, the residue class group N/Nσ is free of rank n−d, and hence, the sublattice Nσ
admits a complement N 0 in N . We then choose Tn−d := N 0 ⊗Z C∗ . Eventually, we define
Zσ as the Tσ -toric variety associated to the cone σ, considered as a (full-dimensional)
cone in (Nσ )R .
The factors Zσ and Tn−d actually are naturally embedded in Xσ as closed subvarieties;
see Remark 2.3.8 for details.
A word of warning may be in order concerning the role of the lattice in the correspondence
between cones and affine toric varieties, a role that is not apparent from the usual notation:

2.2.16 Remark. Given a lattice N ∼ = Zn , then every finite collection v1 , . . . , vr in the rational
vector space NQ := N ⊗Z Q that includes a vector space basis generates a new lattice N e :=
Pr
i=1 Z·vi (in general, this sum is not direct). The two lattices intersect in a sublattice of finite
index (such lattices are called commensurable in the associated real vector space N R ).
An N -cone σ then also is an N e -cone, here denoted by σ e for distinction. Disregarding the
primitivity condition, one may even choose the same set of generators for σ and σ e. Obviously,
the resulting affine toric varieties X and X e are not isomorphic if N 6= N e , since then T 6= T. e
Isomorphy may even fail when disregarding the torus action and only considering the underlying
structures of affine algebraic varieties:

2.2.17 Example. Let N = Z2 be the standard lattice in R2 and σ, the cone spanned by
v1 = 2f1−f2 and v2 = f2 as in Example 2.2.13 (2). Hence, X(N,σ) is the singular two-dimensional
quadric cone Y . With respect to the sublattice N e := Z·v1 ⊕ Z·v2 , however, the cone is spanned
by a lattice basis, and the resulting variety X (Ne ,eσ) is the affine plane C2 .
These two surfaces are even topologically distinct since the local fundamental group of Y at
the origin is the cyclic group C2 .

This example fits into an important generalization of the equivalence of categories described
in Remark 1.3.4:

2.2.18 Remark. Given two lattices N 1 , N2 and Ni -cones σi , a morphism of lattice cones
(N1 , σ1 ) → (N2 , σ2 ) is a lattice homomorphism ϕ : N1 → N2 such that ϕR (σ1 ) j σ2 holds. Then
the homomorphism of tori T(ϕ) : TN1 → TN2 extends to a toric morphism X(ϕ) : X(N1 ,σ1 ) →
X(N2 ,σ2 ) . The correspondence

(2.2.18.1) X : (N, σ) 7→ X(N,σ) , ϕ 7→ X(ϕ)

is an equivalence beween the categories of (strictly convex) lattice cones and of affine toric
varieties.

We come back to the last example to illustrate this correspondence:

2.2.19 Example (2.2.17 continued). The inclusion of lattices ι : N e ,→ N defines a morphism

e, σ
of lattice cones (N e) → (N, σ). To describe the corresponding morphism of toric varieties, we
identify TN and TNe with (C∗ )2 by the choice of the above bases. According to Example 1.3.5,
the morphism of tori T(ι) takes the form (s, t) 7→ (s 2 , t/s). Composed with the embedding of T N
Basic Toric Geometry 21

into X(N,σ) = Y given by the orbit map (u, v) · (1, 1, 1) = (u, uv, uv 2 ) of Example 1.2.1 (2), we
obtain the map (C∗ )2 ,→ Y , (s, t) 7→ (s2 , st, t2 ) that clearly extends to a map from C 2 = X(Ne ,eσ)
to X(N,σ) .

2.3 Cones and orbit structure

We now analyse how the orbit structure of an affine toric variety Xσ can be read off from
the (strongly convex) N -cone σ. The following facts turn out to be easy, but crucial:

2.3.1 Lemma. Let σ be an N -cone, moreover, let ν ∈ N and µ ∈ M . Then:

(1) µ ∈ σ ∨ ⇐⇒ χµ extends to a regular function on Xσ .
(2) ν ∈ σ ⇐⇒ λν : C∗ −→ T ⊂
◦ Xσ extends to a morphism C −→ Xσ .

Proof. (1) is true by definition.

(2) Using (1), a system of semigroup generators µi ∈ σ ∨ ∩ M , i = 1, . . . , r, with corre-
i
sponding characters χi := χµ defines a closed embedding



(2.3.1.1 X ,→ Cr , x 7→ χ1 (x), . . . , χr (x) ,


which is equivariant with respect to the torus action t·z = χ1 (t)z1 , . . . , χr (t)zr of (2.1.3.2)
on Cr . The proof now follows from the equivalence of these five properties:
(a) The one-parameter subgroup λν extends to a morphism C → Xσ ;
(b) The group homomorphisms
i i
χµ ◦ λν : C∗ −→ C∗ ⊂
◦ C , s 7→ shµ , νi

extend to regular functions C → C,

(c) hµi , νi ≥ 0 for i = 1, . . . , r ; (d) hσ ∨ , νi ≥ 0 ; (e) ν ∈ σ = (σ ∨ )∨ .

By the above equivalence (2), given an affine toric variety, we may recover the set of
lattice points in the defining cone and thus, recover the cone:

2.3.2 Remark. For an affine toric variety X, the set of lattice points in its defining
N -cone σX is explicitly given as follows:

σX ∩ N = {ν ∈ N ; λν (0) exists in X} .

In the preceding formula, we have adopted the following convenient notation:

2.3.3 Convention. Let X be a (not necessarily affine) toric variety and λ ∈ Y(T), a
one-parameter subgroup. If λ extends to a morphism from C to X, then the limit

(2.3.3.1) λ(0) := lim λ(s)

s→0

exists in X. We only use the symbol λ(0) in this situation.

22 G.Barthel, K.-H.Fieseler, L.Kaup

In Example 2.2.14, we have seen that the inclusion τ  σ of a face induces an open
embedding Xτ ⊂ ◦ Xσ of affine toric varieties. We are now ready to prove the converse, a
result that plays an essential role for the gluing of affine toric varieties to general ones:
2.3.4 Proposition. Let σ 0 ⊆ σ be an inclusion of N -cones. Then the induced morphism
Xσ0 → Xσ is an open embedding (if and) only if σ 0  σ.
Proof. We may interpret U := Xσ0 as an open subset of X := Xσ . Since X \ U is a
closed T-invariant subvariety, its vanishing ideal in O(X) is generated by finitely many
i
characters χi = χµ ∈ SX . As a consequence, U is the union of the principal open toric
subvarieties Xχi = Xτi corresponding to the faces τi = σ ∩ (µi )⊥ (cf. Example 2.2.14).
S
According to Remark 2.3.2, the analoguous relation σ 0 = τi holds. Hence, there is a
face τi  σ satisfying dim τi = dim σ 0 . Together with τi ⊆ σ 0 ⊆ σ, this implies that σ 0 is
included in lin(τi ) ∩ σ = τi , thus proving σ 0 = τi .

The one-to-one correspondence between faces and affine open toric subvarieties thus
established actually is only one aspect of a larger picture: The combinatorial face structure
of a cone also corresponds to the orbit structure and the structure of invariant irreducible
closed subvarieties.
2.3.5 Theorem. Let σ be an N -cone.
(1) There is a one-to-one correspondence between the following sets:

(a) The set ∆(σ) := {τ ⊂ NR ; τ  σ} of faces of σ,

(b) the set Xσ /T := {T·x ; x ∈ Xσ } of T-orbits in Xσ ,
(c) the set of non-empty closed irreducible T-invariant subvarieties,
(d) the set of open affine toric subvarieties.

(2) This correspondence is explicitly given as follows: To a face τ  σ, we first associate

a “base point”
xτ := λν (0) ∈ Xσ ,
where ν is an arbitrary lattice vector in the relative interior τ ◦ . Then the orbit,
the closed irreducible T-invariant subvariety, and the affine open toric subvariety
corresponding to τ are

Oτ := T·xτ , Vτ := Oτ , and Xτ ,

respectively. Here the closure of Oτ is taken with respect to Xσ , and Xτ is identified

with the image of the open embedding into Xσ .

(3) The correspondence τ ←→ Vτ is inclusion-reversing. Each orbit Oτ is a locally

closed subvariety in Xσ , and it is isomorphic to a torus of the complementary di-
mension
dim Oτ = dim Vτ = dim σ − dim τ .
Basic Toric Geometry 23

Moreover, this orbit is the unique closed orbit in Xτ ; on the other hand, it is open
in its closure [
Vτ = Oτ 0 ,
τ τ 0 σ

the union of Oτ and finitely many orbits of strictly smaller dimension.

(4) Dually, the correspondence τ ←→ Xτ is inclusion-preserving, and the affine open

toric subvariety [
Xτ = Oτ 0
τ 0 τ

is the union of Oτ and finitely many orbits of strictly larger dimension.

The reader might wish to keep in mind that the notation Vτ has no “absolute” mean-
ing, since the closure depends on the ambient toric variety Xσ . – Before proving the
theorem, we first add a few comments: The zero cone o corresponds to the “big” orbit
Oo = T (the embedded torus), whereas the orbit Oσ corresponding to a full-dimensional
cone σ consists of precisely one point xσ , which is then the unique fixed point of the affine
toric variety Xσ .
Next, we look at a basic example (see Figure 10 for n = 2):

2.3.6 Example. Assume that σ = cone(ν 1 , . . . , ν n ) is spanned by a lattice basis of N .

Denote by µ1 , . . . , µn the corresponding dual basis of M . Then Xσ is isomorphic to the
linear space Cn , endowed with the action (2.1.3.2). The base point is (1, . . . , 1). The faces
of σ are of the form σJ := cone(ν j ; j ∈ J) given by the subsets J of {1, . . . , n}. To σJ
corresponds the orbit

OσJ = (z1 , . . . , zn ) ; zi = 0 ⇐⇒ i ∈ J ,

the closure of which is a coordinate subspace. The pertinent orbit base point xσJ is the
unique point of OσJ with zi = 1 for i 6∈ J.

2
τ2N R = R

Xσ = C 2



























































xτ1   xo








































o τ1 xσ xτ 2

Figure 10: Faces of cone(f1 , f2 ) and corresponding orbit base points in C2

If σ is a general d-cone, then the orbit Oσ is isomorphic to a torus of the complementary

dimension n − d :
24 G.Barthel, K.-H.Fieseler, L.Kaup

2.3.7 Corollary. In the product decomposition (2.2.15.1), the variety Zσ has a (unique)
fixed point zσ , and the orbit Oσ corresponds to {zσ } × Tn−d (with isotropy group Tσ at
each point).

In order to reduce the proof of Theorem 2.3.5 to the case of full-dimensional cones,
we add the following complement to the product decomposition of Proposition 2.2.15:

2.3.8 Remark. In the affine toric variety Xσ defined by a d-cone σ, let xo and xσ denote
the base points of the orbits Oo ∼ = T and Oσ , respectively. In the product decomposition
(Xσ , T) ≡ (Zσ , Tσ ) × Tn−d of (2.2.15.1), the two factors admit a natural closed embedding
into Xσ as follows:
= Tσ ·x0 and Tn−d ∼
Zσ ∼ = T·xσ = Oσ .

Proof of Theorem 2.3.5. We proceed by induction on d := dim σ. For d = 0, there

is nothing to prove since then σ = o, so Xo = Oo = Vo = T because of o∨ = MR .
For d > 0, we may assume that d = n: By the product decomposition Xσ ∼ = Zσ ×
Tn−d of Proposition 2.2.15 together with Remark 2.3.8, T-objects like orbits, invariant
irreducible closed subvarieties, and affine open invariant subvarieties in Xσ correspond to
the respective Tσ -objects in Zσ via Y 7→ Y ∩ Zσ and vice versa.
For convenience, we fix a closed equivariant embedding X ,→ Cr given by the character
i
functions χi = χµ ∈ SX corresponding to a system of non-zero generators µ1 , . . . , µr
for EX as in Remark 2.1.3. These characters also describe the torus action on the ambient
space Cr . Since they are non-trivial, the origin is the only fixed point on Cr . We have to
show that 0 ∈ Xσ .
By induction hypothesis, the theorem holds if we replace σ with any proper face τ .
From Example 2.2.14, we know that each natural morphism Xτ → Xσ is an open T-
equivariant embedding. We now consider the open “quasi-affine” toric subvariety X∂σ :=
S
τ σ Xτ of Xσ and its complement, the closed invariant subvariety F := Xσ \X∂σ . The
proof of the theorem will essentially be deduced from the following three properties:

(i) F = {0}, which thus is the unique fixed point on Xσ ,

(ii) F is included in each orbit closure Oτ (taken with respect to Xσ ),

(iii) the equality λν (0) = 0 holds for any ν ∈ σ ◦ .

In fact, properties (iii) and (i) imply that the base point xσ = 0 is well defined and
that its orbit Oσ = T·0 = F is closed, so Oσ equals Vσ , and it has the asserted dimension
n − dim σ = 0. This proves part (2) for τ = σ.
To establish the correspondence between faces and orbits, we fix an arbitrary non-zero
orbit O = T · x through some point x ∈ X \ {0} = X∂σ . We thus have x ∈ Xτ for a
suitable proper face τ of σ, so by induction hypothesis, there exists a face τ0  τ with
O = Oτ0 = T·xτ0 . We may thus replace x with the orbit base point xτ0 . By hypothesis,
Basic Toric Geometry 25

the unicity of τ0 is valid in every affine open subvariety Xτ 0 that includes O, so it is valid
in Xσ .
For the asserted correspondence between faces and invariant irreducible closed subvari-
S
eties, we consider such subvariety A ,→ Xσ with Ȧ := A\F 6= ∅. Then Ȧ = τ ∈∂σ (A∩Xτ )
implies the equality dim(A ∩ Xτ ) = dim A for some face τ ∈ ∂σ. Again by induction hy-
pothesis, there is a face τ0  τ with A ∩ Xτ = V(τ0 ) = O(τ0 ), the closure being taken
in Xτ ; moreover, O(τ0 ) is open in A ∩ Xτ and hence, has the same dimension. Since a
T-orbit is irreducible, this readily implies that A is the closure of O(τ0 ) in Xσ . The unicity
of τ0 is seen as above.
We still have to prove the properties (i)–(iii): The coordinate functions χi ∈ SX
generate O(Xσ ) and thus have at most one common zero. On the other hand, for any
lattice vector ν ∈ N ∩ σ ◦ and for each µi , the inner product satisfies hµi , νi > 0, and this
implies that the limit λν (0) = 0 exists in Xσ .

From the proof of Theorem 2.3.5 and Proposition 2.2.15, we deduce the following
consequence:

2.3.9 Corollary. In an affine toric variety Xσ , the unique closed orbit Oσ is a T-

equivariant deformation retract of Xσ .

Proof. It clearly suffices to consider the case of a full-dimensional cone σ, or, in other
words, the case where Xσ has a (unique) fixed point xσ . We consider an equivariant

i
embedding Xσ ,→ Cr , x 7→ χ1 (x), . . . , χr (x) as in (2.3.1.1) with χi := χµ . To each fixed


lattice vector ν ∈ N corresponds a set of exponents ki := hµi , νi with χi λν (s) = ski and
thus, an induced C∗ -action



C∗ × Cr −→ Cr , (s, z) 7−→ χi λν (s) ·zi i=1,...,r = (sk1 z1 , . . . , skr zr )

on Cr that respects Xσ . For a lattice vector ν in the relative interior σ ◦ (which here
coincides with the topological interior), these exponents satisfy the strict inequality k i > 0.
It follows that the corresponding C∗ -action extends to s = 0. Restricting to scalars
s ∈ [0, 1] yields a T-equivariant (why?) homotopy, providing an equivariant retraction by
deformation to the fixed point.

To study the geometry of orbit closures in an affine toric variety, let τ be a face of a
cone σ. We recall from Corollary 2.3.7 that the subtorus Tτ of T is the isotropy group at
each point of Oτ . Hence, restricting the T-action to Oτ provides an action of the quotient
torus T/Tτ on Oτ .

2.3.10 Proposition. Every orbit closure O τ in Xσ is an affine T/Tτ -toric variety:

Oτ ∼
= Xσ/τ ,

where σ/τ := π(σ) denotes the image cone for the quotient map π : NR → (N/Nτ )R .
26 G.Barthel, K.-H.Fieseler, L.Kaup

Proof. We may regroup the “weight subspaces” C · χµ occuring in formula (2.1.1.3) ac-
cording to the decomposition E = F ∪˙ J of the exponent semigroup E := M ∩ σ ∨ into
the sub-semigroups
F := E ∩ τ ⊥ and J := E \ F
(where J actually is an “ideal”, i.e., J + E j J). We claim that this regrouping yields a
direct sum decomposition

(2.3.10.1) O(Xσ ) = O(Xσ )Tτ ⊕ I(Oτ )

of the coordinate ring into the subalgebra of all Tτ -invariant functions

M
O(Xσ )Tτ = C·χµ
µ∈F

and the ideal of Oτ in Xσ . To that end, it suffices to verify that

M
(2.3.10.2) I(Oτ ) = C·χµ .
µ∈J

We choose an arbitrary lattice vector ν ∈ τ ◦ ∩ N . For a character χ = χµ ∈ O(Xσ ) ∩ X(T)

(i.e., with exponent µ ∈ E), we have the following chain of equivalences since xτ = λν (0) :

χ ∈ I(Oτ ) ⇐⇒ χ(xτ ) = 0 ⇐⇒ hµ, νi > 0 ⇐⇒ µ 6∈ τ ⊥ .

As a consequence of formulae (2.3.10.1) and (2.3.10.2), factoring out the ideal yields an
isomorphism M
O(Oτ ) ∼ = O(Xσ )Tτ = C·χµ .
µ∈F

Under the natural dual pairing of M and N , an arbitrary exponent vector µ ∈ F

vanishes on the sublattice Nτ . It thus corresponds to a unique linear form µ on the
quotient lattice N/Nτ . Such a linear form is the exponent of a character for the quotient
torus T/Tτ , and the condition µ ∈ σ ∨ then is equivalent to µ belonging to the dual cone
of the image σ/τ = π(σ).

Quotients of affine toric varieties

To finish the discussion of affine toric varieties, we briefly come back to quotients, continuing
Remark 1.2.3 (3). We use this notation: Let G be a closed subgroup of a torus T, and let q : T 
T0 := T/G be the quotient map. We denote by N 0 ∼ = Hom(C∗ , T0 ) the lattice of one-parameter
subgroups of T0 , and by qe = dq : N → N 0 , the lattice homomorphism corresponding to q.
Its kernel N0 := ker(dq) is a saturated sublattice of N . Finally, we consider the sublattice
K = N0⊥ ⊂ M corresponding to G introduced in Remark 1.3.4 (4). We note that for each
µ ∈ K, the character χµ passes to the quotient torus T0 ; in fact, the lattice K is naturally
identified with the lattice M 0 = (N 0 )∗ of characters of T0 .
Basic Toric Geometry 27

2.3.11 Remark. Let σ be a full-dimensional (strongly convex) N -cone with dual M -cone σ ∨ .
The image σ 0 := qR (σ) is an N 0 -cone, but it may fail to be strongly convex (this is illustrated
by the “elliptic” case of the following example). The strong convexity of σ 0 is equivalent to the
condition that N0 does not intersect the relative interior σ ◦ .
If σ 0 is strongly convex, then dq induces a map of cones (N, σ) → (N 0 , σ 0 ), so the surjective
homomorphism q : T  T0 extends to a q-toric map

ϕ : X := X(N,σ) −→ X(N 0 ,σ0 ) =: X 0 .

This map is surjective, since each orbit base point x τ 0 of X 0 lies in the image: For a face τ 0  σ 0 ,
there is a face τ  σ and a vector ν ∈ τ ◦ such that dq(ν) lies in the relative interior of τ 0 .
Then ϕ maps xτ = λν (0) to λdq(ν) (0) = xτ 0 .
The comorphism ϕ∗ : O(X 0 ) → O(X) induces an isomorphism between O(X 0 ) and the ring

O(X)G := {f ∈ O(X) ; ∀ t ∈ G : f t = f }

of G-invariant functions on X. This provides an identification

X0 ∼ = X//G := Sp O(X)G (2.3.11.1)

with the algebraic quotient of X by G. The quotient morphism X  X//G admits a factorisation

X  X/G  X//G

through the topological orbit space.

As stated in Remark 1.2.3 (3), this orbit space need not be separated if G is not finite:

2.3.12 Example. We consider three closed embeddings ι∗ : C∗ ,→ T2 with the index ∗ = e, h, p indicating
“elliptic, hyperbolic, parabolic”, given by ιh : s 7→ (s, s−1 ), ιp : s 7→ (s, 1), and ιe : s 7→ (s, s). We denote
by G = Gh := ιh (C∗ ), Gp , and Ge the respective image group. The quotient map q : T2  T2 /G ∼ = C∗ is
±1 −1
then given by a character χ , namely χh (t, u) = tu, χp (t, u) = u, and χe (t, u) = tu .
The map of lattices dq : Z2 → Z is represented by the (2 × 1)-matrices (1, 1) for Gh , (0, 1) for Gp ,
and (1, −1) for Ge . The respective sublattices ker(dq) of one-parameter subgroups are Nh = Z·(1, −1),
Np := Z·(1, 0), and Ne = Z·(1, 1).
Let σ = cone(f1 , f2 ) be the standard lattice cone defining X := C2 . The respective image cone dq(σ)
in NR0 ∼
= R is σ 0 = ray(1) for Gh and Gp . For Ge , we obtain σ 0 = cone(1, −1) = NR0 , so it is not strongly
convex; accordingly, the lattice point (1, 1) ∈ Ne lies in the relative interior of σ.
The C∗ -actions on X given by these groups are as follows:
(1) For Gh , we have the hyperbolic action s·(x, y) = (sx, s−1 y). Each “generic” orbit is a fibre of the
quotient map ϕ : C2 → C2 //Gh ∼ = C, (x, y) 7→ xy, so it is closed. The fibre (xy = 0) consists of
three orbits: the fixed point (0, 0) and the punctured coordinate axes C∗ ×0 and 0×C∗. In the
topological orbit space, these punctured coordinate axes are non-closed points having the fixed
point in their closure. The map C2 /Gh  C2 //Gh identifies these three points.
(2) For Gp , we get the parabolic action s·(x, y) = (sx, y). The quotient map ϕ : C2  C2 //Gp ∼ =C
−1
is given by (x, y) 7→ y. Each fibre ϕ (y) = C×y consists of two orbits, namely the (non-closed)
“pointed” horizontal line C∗ ×{y}, and the (closed) fixed point (0, y). The topological orbit space
consists of two copies of C, say 0×C and 1×C. The points in the first copy are closed, whereas
the closure of any point (1, y) in the second copy consists of the two points (1, y), (0, y). The map
C2 /Gp  C2 //Gp identifies these two points.
28 G.Barthel, K.-H.Fieseler, L.Kaup

(3) For Ge , we obtain the elliptic action s·(x, y) = (sx, sy). The fixed point (0, 0) is the only closed
orbit; any other orbit is a “pointed” line. The topological orbit space consists of the (closed)
unique fixed point and a projective line, with the closure of any point consisting of that point and
the fixed point. The only invariant functions are the constants, so the algebraic quotient C 2 //Ge
consists of a single point, with the obvious quotient map.

To end this discussion of quotients, we add a few further remarks that might help for a
better understanding of the situation. Using the notation introduced above, we again assume
that σ 0 is strongly convex.

2.3.13 Remark (2.3.11 continued). For each T 0 -orbit O(τ 0 ) in X 0 , given by a face τ 0  σ 0 ,
there is a unique face τ  σ such that the T-orbit O(τ ) in X is relatively closed in the preimage


ϕ−1 O(τ 0 ) : We write τ 0 = σ 0 ∩ (µ0 )⊥ with a suitably chosen lattice vector µ 0 ∈ M 0 ∩ (σ 0 )∨ . The
0
map χµ ◦ q : T → C∗ is a character of T; so it is of the form χ µ , and its exponent µ = q ∗ (µ0 )
clearly lies in σ ∨ . It is not difficult to see that in fact, the T 0 -orbit O(τ 0 ) is the G-orbit space
O(τ )/G.
If G is finite, then qe: N → N 0 is the inclusion of a sublattice of finite index |G| (see item (4)
in Remark 1.3.4). Any (strongly convex) N -cone also is an N 0 -cone; the face structure is of
course independent of the lattice, so there is a one-to-one correspondence between T-orbits and
T0 -orbits. In that case, the induced map X/G  X//G is a homeomorphism, so the algebraic
quotient is just the topological orbit space. In such a situation, one calls X//G a geometric
quotient.


We finally note that the algebraic quotient X//G := Sp O(X)G is defined for any closed
subgroup G of T, and it always admits a natural T/G-action. The strong convexity condition
for the image cone σ 0 = qe(σ) implies that the canonical morphism T/G → X//G is an open
embedding, and conversely.

3 Toric Singularities
3.1 Local structure at a fixed point
The common theme of the notes collected in the present volume is the geometry of singu-
larities. In our context, we have to study the singularities occuring on affine toric varieties.
As a consequence of the product decomposition obtained in Proposition 2.2.15, it suffices
to consider the variety defined by a full-dimensional N -cone. Here, the fixed point is the
natural candidate for a singularity.
So first of all we have to find out under which conditions a fixed point is or is not
singular:

3.1.1 Proposition. Let σ be a full-dimensional (strongly convex) N -cone. Then the

following statements are equivalent:

(1) The affine variety Xσ is smooth.

Basic Toric Geometry 29

(2) The unique fixed point xσ of Xσ is a regular point.

(3) The M -cone σ ∨ is spanned by a lattice basis µ1 , . . . , µn , i.e., σ ∨ = cone(µ1 , . . . , µn ).

(4) The N -cone σ is spanned by a lattice basis ν 1 , . . . , ν n , i.e., σ = cone(ν 1 , . . . , ν n ).

(5) Xσ ∼
= Cn with the base point (1, . . . , 1) and the T-action

t · z = (χ1 (t)z1 , . . . , χn (t)zn ),

i
where χi = χµ for a lattice basis µ1 , . . . , µn of M .

A cone satisfying these properties is called regular (see Definition 3.1.2 for the case
of not full-dimensional cones).

Proof. Since the implication “(4) =⇒ (5)” has been discussed in Example 2.2.13, the
only non-trivial implication is “(2) =⇒ (3)”: The maximal ideal of all regular functions
vanishing at xσ is of the form M
m = C·χµ
µ∈Ė
L
with Ė := E \ {0} for E = σ ∨ ∩ M , so m2 = µ∈Ė+Ė C·χµ . Hence, the Zariski tangent
space Txσ Xσ = (m/m2 )∗ has the dual
M
m/m2 ∼ = C·χµ ,
µ∈B

where B := Ė \ (Ė+ Ė) denotes the Hilbert basis of the semigroup E (cf. Remark 2.2.6).
Now
 
B = dimC (m/m2 ) = dim Xσ = n

since xσ is a regular point. As a semigroup, E is generated by B (why?); furthermore, we

know that M = E + (−E) by Remark 2.1.2 (2). Hence, the elements in B generate M
as a lattice and thus form a lattice basis, since there are only n of them.

Somewhat more general are the cones spanned by linearly independent lattice vectors:

3.1.2 Definition. A d-dimensional L-cone is called simplicial if it has exactly d edges.

It is called regular if it is simplicial and if the primitive spanning vectors of the edges
are part of a lattice basis of L.

The name “simplicial” is to indicate that transversal sections of such a cone are (d−1)-
simplices. (In [Ew], they are called “simplex cones”, whereas “simplicial cone” there
means a cone that has simplicial polytopes as cross-sections.) Two-cones are always
simplicial, whereas cones of dimension d = 3 in general are not. Our standard example
is provided by σ = cone(f1 , f2 , f3 , f1 +f2 −f3 ) defining the Segre cone Z of 2.2.13 (3) (see
also 2.2.7 (3) and Figure 7 for its dual).
30 G.Barthel, K.-H.Fieseler, L.Kaup

To a simplicial d-cone σ = cone(v1 , . . . , vd ), we associate the sublattice

d
X
Γσ := Li j L σ with Li := L ∩ lin(vi )
i=1

(recalling Lσ := L ∩ lin(σ), cf. Proposition 2.2.15), generated by the lattice points on

the edges (see Figure 11 for an example with d = 2). Then the regularity of σ can be

b
qs q q q q q q q q q q q q q q q q q q 
f
b
qf
s
q bq q q q q q q qfs q q q q q q q q qqq
q q bq b
q q q q q q q q q q q q q qfsq q q
q q q q b s q q q q q q q q qqq q q q
q qf
q q q q q bq b sq q q q q q
q q q q q q q qf
s q q q q b
q q qf q qbq q qqq q q q q q q
q q q q q q q q q qbqf
s q q q q q q q q q
q q q q q q q q q q q q q q q q q q qfs q
q q q q q q q qf
s q q q q q q q q q q q q
q q q q q q q q q q q q q q q qfs q q q q

Figure 11: Sublattice Γσ for σ = cone(−5f1 +3f2 , 3f1 +2f2 )

characterized by the equality Γσ = Lσ . More precisely, the deviation of a simplicial L-cone

from regularity is measured by the index of this sublattice:

3.1.3 Definition (Multiplicity of a simplicial cone). For a d-dimensional simplicial

L-cone σ, the positive integer

mult(σ) := mσ := [Lσ : Γσ ] = |Lσ /Γσ |

is called its multiplicity.

It follows from the structure theorem for finitely generated abelian groups that the
multiplicity of a full-dimensional simplicial L-cone is readily computed in terms of a
primitive generating system:

3.1.4 Remark. If the generators v1 , . . . , vn ∈ L of a simplicial n-cone are primitive and

n = rank(L), then, after fixing an isomorphism L ∼ = Zn ,

 
mult cone(v1 , . . . , vn ) = det(v1 , . . . , vn ) .

A topological interpretation of the multiplicity for full-dimensional cones is given in

Corollary 3.1.9.
The next observation follows immediately from the definition of the multiplicity.

3.1.5 Remark (Geometric interpretation of the multiplicity). For a simplicial

d-cone σ spanned by primitive lattice vectors v1 , . . . , vd ∈ L, let
X
d 
P = P (v1 , . . . , vd ) = ti v i ; 0 5 t i < 1
i=1
Basic Toric Geometry 31

be the “half open” fundamental parallelotope of the sublattice Γσ . Then the multiplicity
is given by the number of N -lattice points in P :

mult(σ) = #(P ∩ L) .

As a useful consequence, one may check for regularity by counting lattice points:

3.1.6 Corollary (Geometric regularity criterion). A simplicial d-cone σ as above is

regular if its fundamental parallelotope P intersects the lattice L σ only at the origin.
In dimension d = 2, the same conclusion holds if we replace P with the d-simplex
conv(0, v1 , . . . , vd ).

Toric varieties given by a simplicial N -cone have a remarkably simple structure that
we are now going to describe. In the two-dimensional case (where cones are always
simplicial), we have already seen that the affine toric surface Yk (with k = 2) introduced
in Remark 1.2.2 is a quotient C2 /Ck of the affine plane by a suitable action of a finite
group, namely the cyclic group of order k. We want to show that an analoguous result,
except for the cyclic structure of the group, holds for arbitrary affine toric varieties in the
“simplicial class”.
Using the product decomposition of Proposition 2.2.15, we again may assume that
dim σ = n. We then consider on Cn the action of T ∼ = (C∗ )n by componentwise multipli-
cation, thus identifying the elements in T with diagonal matrices in GLn (C):

3.1.7 Proposition. For a full-dimensional simplicial N -cone σ, the toric variety X σ

is the algebraic quotient Cn /G := Sp O(Cn )G of Cn by the induced action of a finite
subgroup G ∼= N/Γσ of the torus T, so Xσ is just the usual orbit space of the action of G
n
on C .

In the literature, varieties that locally are orbit spaces of a smooth manifold with respect to
an action of a finite group often are called “orbifolds”.

Proof. With respect to the sublattice N 0 := Γσ of the lattice N , the cone σ =: σ 0 is

regular. By Proposition 3.1.1, the associated toric variety X(N 0 ,σ0 ) is the affine space Cn .
The assertion now follows from our discussion of quotients in Remark 1.3.4 and 2.3.13.

The group G ∼ = N/Γσ actually can be recovered from the toric variety by topological
means, namely as fundamental group of the open invariant subvariety of regular points:

3.1.8 Remark. There is an isomorphism

G ∼
= π1 (Xσ )reg .

Indication of Proof. We let ν 1 , . . . , ν n ∈ N denote the primitive generators of σ, and

ι : fi0 7→ ν i , the inclusion of the sublattice Zn ∼= N 0 := Γσ into N . On the side of tori, we
have the finite morphism
T(ι) =: q : T0  T ∼
= T0 /G ,
32 G.Barthel, K.-H.Fieseler, L.Kaup

with T0 := TN 0 and T := TN for ease of notation. Then q extends to the quotient

morphism
ϕ = X(ι) : X 0 = Cn  Cn /G ∼
=X,

with X 0 := X(N 0 ,σ0 ) and X = X(N,σ) . Furthermore, let Z ,→ Cn be the set of points where
the G-action has non-trivial isotropy. It suffices to show thatϕ induces an unramified
covering Cn \ Z → (Xσ )reg which is the universal covering.
First of all, we verify that Cn \ Z is simply connected: We use the fact that a lat-
tice vector ν ∈ N is primitive if and only if the corresponding one-parameter subgroup
λν : C∗ → T is an injective homomorphism. Since ι(fi0 ) = ν i , this implies for each coor-
dinate subtorus Ti := λfi0 (C∗ ) of TN 0 that the group G ∩ Ti is trivial. Restated in other
words: If a matrix in G ⊂ T ⊂ GLn (C) has the eigenvalue 1 with multiplicity at least
n−1, then it is the identity; such a group G is called a small subgroup of GLn (C). So the
subset Z ,→ Cn is a union of coordinate subspaces of dimension at most n−2. Using the
fact that spheres of dimension at least 2 are simply connected, the complement Cn \ Z
can be seen to be simply connected as well.
Eventually, we have to show that the preimage of the singular locus S(Xσ ) coincides


with Z. Since Z = ϕ−1 ϕ(Z) , this follows from the equality S(Xσ ) = ϕ(Z). The
inclusion “⊆” is clear, while “⊇” follows from the fact that the set of points where a
dominant morphism between equidimensional smooth varieties is not a local analytic
isomorphism has codimension one.

3.1.9 Corollary. The multiplicity of a full-dimensional simplicial N -cone σ satisfies



mult(σ) = π1 (Xσ )reg  .

3.1.10 Example. (1) Let n = 2. For coprime integers k, ` with ` > 0, we consider σ =
cone(`f1 −kf2 , f2 ) in R2 . Then q : T → T of (3.1.7.1) satisfies q(t 1 , t2 ) = (t`1 , t−k
1 t2 ); thus
G is the cyclic subgroup of order ` in GL 2 (C) generated by (ζ, ζ k ) with ζ := e2πi/` . The
singularity C2 /G thus obtained is a cyclic quotient singularity, and mult(σ) = |G| = ` .
The group G lies in SL2 (C) if and only if k = ` − 1. In that case, for ` = 2, the semigroup
σ ∨ ∩ M is generated by e1 , e1 + e2 , (` − 1)e1 + `e2 . The closed equivariant embedding of
(2.3.1.1) shows that Xσ is a toric hypersurface in C3 , namely the surface Y` of Remark
1.2.2. In particular, for ` = 2, we thus obtain one of our standard examples, the quadric
cone Y of Example 1.2.1 (2).

(2) Under the primitivity hypothesis of Remark 3.1.8, for any dimension n, the group G has a
system of at most n−1 generators. Non-cyclic groups actually occur already in dimension
n = 3. An example is furnished by σ = cone(v 1 , v2 , v3 ) with v1 = f1 , v2 = f1 + 2f2 , and
v3 = f1 + 2f3 .

To close this subsection, we have to add at least some remarks on the non-simplicial case.
As usual, we may restrict to full-dimensional cones.
Basic Toric Geometry 33

3.1.11 Remark. The structure of general toric singularities is considerably more complicated.
There is a close relation with polytopes spanned by lattice vectors (called lattice polytopes for

short): Intersecting a given n-cone σ with the affine hyperplanes H µ := v ∈ NR ; hv, µi = l
for any µ ∈ (σ ∨ )◦ ∩ M (on a suitable integral “level” l > 0) associates to σ a family of (n−1)-
dimensional lattice polytopes with fixed combinatorial type. Conversely, any lattice polytope
in Rn−1 , placed in the affine hyperplanes (xn = l) on different levels l ∈ N>0 , spans a family of
n-cones.
For n = 3, the associated polytopes are plane polygons. Their combinatorial type is just
given by the number of vertices (or edges). For any fixed number k, however, there are countably
many non-equivalent realizations of a k-gon as a lattice polygon. For n = 4, we have to look at
three-dimensional polytopes (classically also called “polyhedra”). Here the situation gets much
worse already on the side of combinatorial types: The enumeration results known so far show
that the number of such types, considered as a function of the number f i of i-faces (vertices,
edges, or facets for i = 0, 1, or 2), grows rather rapidly. (An empirical formula gives (k−6) (k−8)/3
as approximate number of types for f 1 = k.)
For special types of toric singularities, there are satisfactory classification results: Toric
complete intersection singularities, for example, correspond to the so-called Nakajima polytopes
placed on the level l = 1. Starting from points and line segments, these polytopes are inductively
constructed as follows: One takes a sufficiently high prism over a Nakajima base polytope and
then makes a “skew” truncation by a linear height function that is strictly positive on the relative
interior of the base and integer-valued at lattice points. – For a survey of results about toric
singularities, we refer to section 2 in [Cox]. A more detailed exposition is outside the scope of
the present notes.

3.2 General toric varieties and fans

Except for the general definition, the example of the projective space (see 1.2.1), and
Sumihiro’s Theorem 1.2.5, we so far only have studied affine toric varieties. This is
sufficient for the local investigation of toric singularities, but it does not allow to deal
with problems like resolution of such singularities. Since we intend to address this topic
in the final subsection, we have to provide the necessary tools.
As a consequence of Sumihiro’s Theorem, every toric variety, say X, can be covered
by open affine toric subvarieties. From Remark 2.3.2, we know that to each affine toric
variety, say U , corresponds a unique N -cone σ = σU such that U = Xσ .
To the general toric variety X, we may thus associate the following collection of N -
cones:
∆ := ∆(X) := {σ = σU ∈ Ob(SCN ) ; U ⊂
◦ X} ,

where U runs through the affine open toric subvarieties of X. For any two cones σ, σ 0 ∈ ∆,
the intersection Xσ ∩ Xσ0 is a T-invariant affine open subspace of X, and thus Xσ ∩ Xσ0 =
Xτ with a cone τ ∈ ∆. Since X is separated, a one-parameter subgroup λ ∈ Y(T) has at
34 G.Barthel, K.-H.Fieseler, L.Kaup

most one limit λ(0) ∈ X. Hence, according to Remark 2.3.2 (2), the following holds:

τ ∩ N = {ν ∈ N ; λν (0) ∈ Xσ ∩ Xσ0 }
= {ν ∈ N ; λν (0) ∈ Xσ } ∩ {ν ∈ N ; λν (0) ∈ Xσ0 }
= (σ ∩ N ) ∩ (σ 0 ∩ N ) .

This readily implies that τ = σ ∩ σ 0 . Furthermore, it follows from Proposition 2.3.4 that
σ ∩ σ 0 is a common face of both, σ and σ 0 . That leads to the following notion:
3.2.1 Definition. An (N -lattice) fan in NR is a finite non-empty set ∆ of (strongly
convex) N -cones satisfying
(1) τ  σ ∈ ∆ =⇒ τ ∈ ∆ ;
(2) σ, σ 0 ∈ ∆ =⇒ σ ∩ σ 0  σ, σ 0 ; in particular, σ ∩ σ 0 ∈ ∆.
A fan is called simplicial or regular, respectively, if each of its cones has that property.
There are two fans that naturally correspond to a cone:
3.2.2 Remark. A single (strongly convex) N -cone σ generates its full face fan

∆(σ) := {τ ; τ  σ} ,

also called an affine fan. If σ 6= o, then

∂σ := {τ ; τ  σ}

is a subfan, called the boundary fan of σ.

We have just seen that a toric variety determines a fan. On the other hand, given a
fan, there is a unique corresponding toric variety:
3.2.3 Proposition. To every N -fan ∆, one associates the toric variety
[
X∆ := Xσ
σ∈∆

by gluing the family of affine toric varieties (Xσ )σ∈∆ , where Xσ and Xσe are glued along
the common open affine invariant subvariety Xσ∩eσ .
The fact that the prevariety X∆ is separated and thus, a variety, can be seen as follows:
For cones σ and σ e in ∆, the cone τ := σ ∩ σ e lies in ∆, too. Hence, the intersection
Xσ ∩ Xσe ∼
= Xτ again is affine. We have to verify that the corresponding comorphism

d∗ : O(Xσ × Xσe ) = C[M ∩ σ ∨ ] ⊗C C[M ∩ σ

e∨ ] −→ O(Xτ ) = C[M ∩ τ ∨ ]

of affine algebras, given by χµ ⊗ χµe 7→ χµ+eµ , is surjective. This is an immediate conse-

quence of the equality τ ∨ = σ ∨ + σ
e∨ , see Remark 2.2.9 (5).
Obviously, Proposition 3.1.1 implies that the variety X∆ is smooth if and only if the
fan ∆ is regular.
Basic Toric Geometry 35

Pn
3.2.4 Example. In NR ∼
= Rn , we set f0 := − i=1 fi and

(3.2.4.1) ∆ := σJ := cone(fi ; i ∈ J) ; J $ {0, . . . , n} .

Then X∆ is isomorphic to the projective space Pn with the toric structure of Example
1.2.1 (4). – In Figure 12, we depict the fan ∆ in the two-dimensional case.


 
 
 
 
 
 
 
 
 
 
 
 
 



























 






























 

 

 

 

 

 

 

 

 

 

 

 




























 


















































































 

 

 

 

 

 

 

 

 

 

 

 

 




























 


















































































 

 

 

 

 

 

 

 

 

 

 

 

 




























 


















































































 

 

 

 

 

 

 

 

 

 

 

 

 























































 
























































 

 

 

 

 

 

 

 

 

 

 

 























































 






















































 

 

 

 

 

 

 

 

 

 

 

 

 























































 
























































 

 

 

 

 

 

 

 

 

 

 

 























































 






















































 

 

 

 

 

 

 

 

 

 

 

 

 























































 






















































 

 

 

 

 

 

 

 

 

 

 

 

 























































 
























































 

 

 

 

 

 

 

 

 

 

 

 























































 











































































































 























































 


 


 


 


 


 


 


 


 


 


 


 


 































 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 




 

 

 

 

 

 

 

 

 

 

 

 

 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 




 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 






 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 








 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 



 


 

 

 

 

 

 

 

 

 

 

 

 

 

 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 






 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 






 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 



 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 


 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Figure 12: The fan for the projective plane P2

Proof. We write Nr in order to indicate the rank of the lattice under consideration; thus
∆ is a fan in (Nn )R . In Nn+1 , we exceptionally denote the standard lattice basis by
(g0 , . . . , gn ), and consider the regular cone σ := cone(g0 , . . . , gn ). The homomorphism

p : Tn+1 −→ Tn , u := (u0 , . . . , un ) 7−→ (u1 u−1 −1

0 , . . . , u n u0 )

induces the linear map

n
X
dp : (Nn+1 )R −→ (Nn )R , g0 7−→ − fi , gi 7−→ fi for i = 1, . . . , n ,
i=1

which maps the cones in ∂σ onto the cones in ∆. Hence, p extends to a morphism

Cn+1 \{0} = X∂σ −→ X∆ .


The kernel of p is the diagonal D := (s, . . . , s) ∈ (C∗ )n+1 . As a consequence, the


map p is D-invariant and factors through Pn = Cn+1 \ {0} D. In order to establish the
isomorphism Pn ∼ = X∆ , we consider a cone σJ ∈ ∆ and denote by τJ the unique cone in
∂σ with dp(τJ ) = σJ . Then

p−1 (XσJ ) = XτJ ∼

= ZτJ ×D −→ XσJ ∼
= ZτJ

is the projection onto the first factor. That proves the claim.
36 G.Barthel, K.-H.Fieseler, L.Kaup

Many of the general remarks already made in the affine case remain valid in this
more general setting. Firstly, the dependence on the lattice N has to be kept in mind
(cf. Remark 2.2.16): If N and N e are commensurable lattices, then any N -fan ∆ can be
e -fan ∆.
considered as an N e The resulting toric varieties need not be isomorphic as abstract
varieties:

3.2.5 Example (Weighted projective spaces). For an (n + 1)-vector a := (a 0 , . . . , an ) of

integers ai = 1 with gcd(a0 , . . . , an ) = 1, we consider the fan ∆ of (3.2.4.1), but replace the
standard lattice N ∼ = Zn with the finer lattice Ne generated by the rational vectors (1/a i )fi for
i = 0, . . . , n. Then ∆ of course remains simplicial, but in general, it is no longer regular. The
resulting toric variety is called the weighted projective space P(a).
The open affine “charts” given by the n-cones σJ (for J $ {0, . . . , n} as above) are cyclic
Qn
quotients Cn /GJ . There is an isomorphism P(a) ∼ = Pn /G(a) with G(a) := i=0 Cai acting
coordinatewise on Pn , so in particular, P(1, . . . , 1) ∼= Pn . Moreover, the description Pn ∼ =
(C n+1 )/D given above generalizes to the weighted projective space if one replaces the diagonal
1-subtorus D ⊂ (C∗ )n+1 with D(a) := {ta = (ta0 , . . . , tan ) ; t ∈ C∗ }.

Secondly, there is a similar equivalence of suitable categories (cf. Remark 2.2.18);

see (3) below. Thirdly, the results pertaining to orbits and orbit closures carry over to
the general case:

3.2.6 Remark. (1) For each cone σ ∈ ∆, there exists an associated orbit

Oσ := T·λν (0) ,

where ν ∈ σ ◦ . Again, Theorem 2.3.5 and formula (2.3.10) hold. In fact, the orbit
Oσ is the unique closed orbit in the open subvariety Xσ of X∆ , and the T/Tσ -toric
variety O σ ,→ X∆ satisfies
Oσ ∼
= X∆/σ
with the quotient fan ∆/σ := {τ /σ ; τ ∈ ∆, σ  τ } in (N/Nσ )R .
In Figure 13, we indicate schematically the correspondence between cones and orbits.

(2) A toric variety X∆ is complete if and only if the fan ∆ is complete (that is, its
S
support |∆| := σ∈∆ σ is the entire linear space NR ). – In the strong topology of
complex varieties, “complete” means compact.

(3) Let ∆, Λ be fans in NR . Then there is a toric morphism X∆ → XΛ if and only if

each cone of ∆ is included in some cone of Λ. – More generally, the equivalence
of categories stated for the affine case as in (2.2.18.1) easily carries over to lattice
fans and general toric varieties. (For a lattice homomorphism ϕ : N1 → N2 and
Ni -lattice fans ∆i , the condition for a morphism of lattice fans is that the ϕR -image
of each cone of ∆1 be included in some cone of ∆2 .)
Basic Toric Geometry 37

O σ1
%1
%
 
 2

 
 

 
 

 
 

 
 

 
 

 
 

 
 σ

  

 1

 

 

 

 

  O σ2
 


  
  
  
  
  
  
  
  
 
 
 
 
 
  O %2


 
 
  


  
  
  
  
  
  
  


 


 


 


 


 


 































 



 





  
  


  

   


  

   


  

   


  

   


  

   


  

   


  

 

 
   

 

  

  

  

  

   

O %1








 

σ2

 

 

  

  

  

  

  

  

  

  

 

 

 

 

 

  O %3






 


  


  


  


  


  


  


   

  

 

 

 

  




 
  
  
  
  
  
  
  
 
  

























 




  




  




  




  




 



 











 Oo = T
 


 
 
 
 
 
 
 
  







































 


 


 


 


 

































  



































































 


 
 
 
 
 
 
 
  





































































































































 %5
%3










































 



 



 



 



 



 



 



 



 



 O %5

























































































 











































 



 



 



 



 



 



 



 



 



 































 





 





 





 





 





 





 





 





 





 



  O %4









































 


 


 


 


 


 
σ

 


 


 


 




































 





 





 





 





 





 





 






3 





 























































%4 O σ3

Figure 13: Correspondence between fan cones and orbits

(4) For fans ∆ and Λ in NR , we assume that each cone of ∆ is included in some cone
of Λ. The resulting morphism X∆ → XΛ is proper if and only if |Λ| = |∆|; and in
that case ∆ is called a subdivision or refinement of Λ. – In the strong topology
of complex varieties, “proper” means that the inverse image of a compact subset is
again compact.

For a proof of (2) and (4), we refer to the standard literature.

3.2.7 Exercise. In a two-dimensional fan ∆, let a ray % be the common edge of two 2-cones.
Describe ∆/% and show that O % is a projective line.

3.2.8 Exercise. Given the lattice vectors v 1 = f2 , v2 = 2f1 − f2 , and v3 = −f1 , the 2-cones
σi,j := cone(vi , vj ) determine a fan ∆, indicated in Figure 14. Show that two of the cones are
regular and one is singular, defining the affine quadric cone of Example 1.2.1 (2).




  
 











  



 
 
 
 
 



   














































 
 
  
 
 
 
 
 
   




























 







  
 
 
 
 
 
 
   





























































 
  
 
 
 
 
 
 
   




























 


  




  
 
 
 
 
 
  

































































 
   
 
 
 
 
 
 
 
 
 
 
 
   




























 

      

   
  

  

































                  



 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
 

  

 
  
 
 
             



 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 

 


 

 
 


 
             
    





    



    



    



    



    



    



    



    



    



    



    



    



    



    



     




 
             
 

  

   

   

   

   

   

   

   

   

   

   

   

   

   

   

    
 
             
   

     
 
 
 
 
 
 
 
 
 
 
 
 
 
 







  


  


  


  


  


  


  


  


  


  


  


  


  


  


  
 



             
    


        

   

   

   

   

   

   

   

   

   

   

   

   

   



 
             
 

  
    
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   
 

   


 
               

     
   
   
   
   
   
   
   
   
   
   
   
   
   
   
  

 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 
  
 


                                

Figure 14: The fan for the projective quadric cone

The fan ∆ is complete, and X∆ is the projective quadric cone. – Figure 15 shows the
set (X∆ )R of real points, also called the “pinched torus” (tore pincé). The caveat about inter-
preting “real” pictures of complex varieties stated in Example 1.2.1 (2) also applies here.
38 G.Barthel, K.-H.Fieseler, L.Kaup

Figure 15: The set of real points of the projective quadric cone

The projective quadric cone is the closure in P 3 of the affine quadric cone Y in C3 . In
general, the projective closure of an affine toric variety needs not be normal:
3.2.9 Example. Among the singular two-dimensional toric hypersurfaces Y k = V (C3 ; xz −y k )
considered in Remark 1.2.2, only this cone Y = Y 2 and the cubic Y3 with the A2 singularity have
projective closures in P3 with isolated (and thus, normal) singularities. Since the torus action
on the affine part clearly extends, the closure then is a (normal) toric variety.
We add a few remarks on the closure of Y 3 in P3 : In homogeneous coordinates [x, y, z, w], it
is given by the homogenized equation wxz = y 3 . Hence, there are two additional A2 singularities
at infinity, namely at the origins of the affine charts (x = 1) and (z = 1), respectively. The affine
quotient representation Y3 ∼ = C2 /C3 explained in Remark 1.2.2 also carries over to the closure:
In homogeneous coordinates [r, s, t] for P 2 , the action takes the form ζ · [r, s, t] := [r, ζs, ζ 2 t],
and the quotient P2 /C3 is embedded in P3 via [r, s, t] 7→ [r 3 , s3 , rst, t3 ]. – The defining fan is
spanned by v2 = f2 , v1 = 3f1 −2f2 , and v0 = −3f1 +f2 , so it consists of three 2-cones σi with
mult(σi ) = 3.

Moreover, the projective quadric cone is isomorphic to the weighted projective plane P(1, 1, 2).
This is a particular case of the following exercise (whereas the cubic is not of this form):
3.2.10 Exercise. Let ∆(v0 , v1 , v2 ) be the complete two-dimensional lattice fan given by prim-
itive lattice vectors vi , and let mi be the multiplicity of cone(vj , vk ). Then the toric surface
X∆ is a weighted projective plane – and then isomorphic to P(m 0 , m1 , m2 ) – if and only if
gcd(m0 , m1 , m2 ) = 1. That condition in turn is equivalent to the fact that N coincides with its
P
sublattice 2i=0 Z·vi .

Toric divisors
For applications in Shihoko Ishii’s course, we introduce the concept of toric divisors. First,
we briefly recall the general notion: In order to study the zeros and poles of a rational
function f on a normal variety Z , one first notes that the locus of zeros and poles of f has
only finitely many irreducible components, all of codimension one. Hence, one associates
to every 1-codimensional irreducible subvariety A ,→ Z a “multiplicity” vA (f ) ∈ Z, the
vanishing order of f along V . Now the pertinent information may be encoded in the
“divisor” (f ) of f : By definition, a divisor on Z is a formal sum
X
D = vA · A,
A,→Z
Basic Toric Geometry 39

over all one-codimensional irreducible subvarieties, where only finitely many of the integer
coefficients vA are allowed to be non-zero. Then the divisor of the rational function f is
X
(f ) := vA (f ) · A .
A,→Z

Returning to the toric situation, the T-invariant irreducible subvarieties of codimension

one in a toric variety X∆ are just the orbit closures D%i := O%i , where the %i denote the rays
of ∆. The characters of the torus are regular functions on T without zeros. Considering
them as rational functions on X∆ , it follows that non-trivial multiplicities only can occur
S
on the “boundary” X∆ \ T = ki=1 D%i . – Let ν i ∈ N be the primitive generator of
%i = ray(ν i ).

3.2.11 Remark. The divisor in X∆ of the character χ is

k
X
(3.2.11.1) (χ) = hχ, λν i i D%i .
i=1

Proof. For a fixed index i, we set % := %i and λ = λν i . Then λ maps C∗ isomorphically

onto the subtorus T% of T. As in 2.2.15 and 2.3.7, we write X% = Z% × Tn−1 = C × Tn−1
with O% = {0} × Tn−1 and closure Y := O% . For f ∈ C(X% ), the multiplicity vY (f ) in
the divisor (f ) is just the multiplicity of the function s 7→ f (s, t) at s = 0, for generic


t ∈ Tn−1 . Applying this to the special case f = χ, the equation χ(s, 1) = χ λ(s) = shχ,λi
implies that χ(s, t) = χ(1, t)·shχ,λi has multiplicity hχ, λi at s = 0 for all t ∈ Tn−1 .

A divisor on X∆ of the form

k
X
D = n i D %i
i=1

is called a toric divisor. The special case where all coefficients ni = 1 is of particular
interest. Its negative,
Xk
KX := − D%i ,
i=1

is the famous canonical divisor.

3.3 Resolution of toric singularities

In general, smooth (i.e., non-singular) varieties are much better understood and usually
enjoy much nicer formal properties than singular ones. In studying singular varieties,
it is thus a natural attempt to “resolve” the singularities. This means to find a non-
singular “model” X e of the given singular variety X, i.e., a smooth variety X e together
with a proper morphism X e → X that is an isomorphism over the regular locus Xreg . The
general resolution of singularities is rather involved. For complex varieties, it has been
achieved by a celebrated result of Hironaka.
40 G.Barthel, K.-H.Fieseler, L.Kaup

In the toric case, resolution of singularities is much more accessible: We recall that
such a variety X∆ is smooth if and only if the defining fan ∆ is regular. According to
Remark 3.2.6, a subdivision ∆0 of a (general) fan ∆ corresponds to a proper toric morphism
X∆0 → X∆ that induces an isomorphism on the common open invariant subvariety X∆∩∆0 .
Hence, an equivariant resolution of singularities is given by a regular subdivision ∆0 of ∆,
such that the subdividing fan ∆0 contains ∆reg as a subfan. In that case we call ∆0 a


resolution of ∆ or of the cone σ if ∆ = ∆(σ) .

3.3.1 Theorem (Equivariant resolution of toric singularities). For every toric

variety, there exists a resolution of the defining fan and thus, an equivariant resolution of
singularities.

Since a one-dimensional normal variety is smooth, the case of (normal!) toric curves is
without relevance for singularities: In fact, the only such curves are C∗ , C, and P1 . Thus,
we first discuss the resolution of singular toric surfaces.

(I) The surface case

This situation can be dealt with most explicitly: A two-dimensional fan necessarily is sim-
plicial, so the singularities are of the nice “quotient” type discussed in Proposition 3.1.7,
and they are necessarily isolated.

3.3.2 Theorem. For every toric surface, there exists a unique minimal resolution of the
defining fan and thus, a canonical equivariant resolution of singularities.

It obviously suffices to prove this statement in the affine case:

3.3.3 Lemma. A two-dimensional N -cone admits a unique minimal resolution; in par-

ticular, every resolution is a refinement of the minimal one.

Proof. We may assume that N = Nσ . To construct a minimal resolution, we consider the

convex hull K of the set N ∩ σ \ {0} as in Remark 2.2.6. The boundary of this polyhedron
consists of two unbounded half-lines, each one included in an edge of σ, and finitely many
bounded line segments. It thus contains only finitely many primitive lattice points, say
ν 0 , . . . , ν r+1 ∈ N in clockwise order. We set %i := ray(ν i ), σi := %i +%i+1 , and

∆ := {τ ; τ  σi for some i , 0 ≤ i ≤ r}

(see Figure 16). According to Remark 3.1.6, each cone σi is regular, since the triangle
with vertices 0, ν i , ν i+1 contains no further lattice vector.
Now let ∆0 be an arbitrary regular subdivision of σ. To verify that ∆0 is a refinement
of ∆ – thus also proving minimality and unicity of ∆ –, it suffices to show that each
ray %i of ∆ is a ray of ∆0 : The ray %i is included in a 2-cone τ := cone(b1 , b2 ) of ∆0 .
Since τ is regular, we may assume that (b1 , b2 ) is a basis of N . Hence, by the regularity
criterion of Remark 3.1.6, there is no further lattice point in the triangle spanned by 0,
Basic Toric Geometry 41

bsp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp  b sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp sp 
sp Q
ps ps @
sp sp
b p bp ps ps ps ps ps ps sp sp sp sp sp sp sp sp sp sppp b p Q ps ps ps sp sp sp sp sp sp sp sp sp sp sppp
@
bp Q
p p b p bp ps ps ps ps ps ps ps ps ps ps ps ps spp p p p p b p Qp ps @
b ps ps ps ps ps ps ps ps ps ps ps spp p p
p p p p pbpg p p p ps ps ps ps ps ps ps ps pspp p p p
sH p p p p Q b
p Qpg ps ps ps ps ps ps ps ps pspp p p p
s @
p p p p p p ppH p p p pg  p p p p p pbQ pb@ 
p p p p p p p pH
sp p ps ps ps ps  ps p p pg
s p p p p p p s ps ps ps ps ps pg
pg s p p p p p p
p pH
p p p pg s 
sp pg sp p p p pp p p p p p p p p
pg p p p p p p p Q pb@Qpbpg s 
s pg s 
pg p p p p p p p p
p p p p p p p p p ppppppppp p p p p p p p p p p p p p p p p p @ Q
bp pp p p p p p p p p
p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p

Figure 16: Polyhedron Kσ and minimal resolution for σ = cone(−5f1 +3f2 , 3f1 +2f2 )

b1 , b2 . Moreover, the lattice points b1 , b2 , ν i lie in K, so the line segments [b1 , b2 ] and
[0, ν i ] intersect in a point of K. Since ν i is an element of ∂K, it lies on the line segment
[b1 , b2 ], so it is one of the endpoints.
We indicate how to construct the first subdividing vector ν 1 ; iterating that step then
enables a recursive computation of all vectors ν i : For the primitive spanning vectors
ν 0 , ν ∈ N of σ, there is a lattice basis b1 , b2 of N ∼= Z2 such that

(3.3.3.1) ν 0 = b2 and ν = mσ b1 − kb2 with an integer 1 5 k < mσ

(see the following exercise). Then ν 1 is the vector b1 .

3.3.4 Exercise. (1) Given two primitive lattice vectors v 1 , v2 ∈ Z2 with det(v1 , v2 ) = m > 1,
prove that there exists a (positively oriented) lattice basis (b 1 , b2 ) and an integer k with
1 5 k < m, gcd(m, k) = 1 such that v1 = mb1 − kb2 and v2 = b2 .

(2) In the proof of Lemma 3.3.3, show that the first subdivision yields a resolution if and only
if k = 1.

(3) Show that the maximal number of necessary subdivisions equals m σ −1, and characterize
the case when this occurs.

The theory of toric surface singularities and their resolution is particularly rich and
fascinating. Within the scope of these notes, we have to content ourselves to indicating
some key results:

3.3.5 Remark. In the situation of Lemma 3.3.3, let X∆ → Xσ be the minimal resolution
of the singular affine surface Xσ , and π : X∆0 → Xσ , an arbitrary resolution. We denote
by %0 , . . . , %r+1 the rays of ∆0 in clockwise order.

(1) The “exceptional fiber” E := π −1 (xσ ) over the singular point of Xσ consists of
the “new” orbit closures Ei := O%i ∼ = P1 (cf. Exercise 3.2.7) corresponding to the
subdividing rays %1 , . . . , %r . They form a “chain” E1 , . . . , Er of curves as depicted
in Figure 17: Each Ei only intersects its neighbours; the intersection is transverse
and consists of one point (transversality means that the curves meet like coordinate
axes).
42 G.Barthel, K.-H.Fieseler, L.Kaup

@ @ @
E1 @ E3 @ @ Er
E2 @ @ ... Er−1@
@
@ @
@ @
@

Figure 17: System of exceptional curves

(2) To curves Ei and Ej , one attaches their “intersection number” Ei·Ej . For i 6= j, this
is just the number of intersection points since a non-empty intersection is transverse.

For i = j, the “self intersection number” Ei·Ei = Ei2 expresses the “twisting” of the
ambient smooth surface along the curve. In the present case, this number equals
−ai , where ai = 1 is the multiplicity of the cone %i−1 + %i+1 spanned by the two
neighbouring rays of %i . For a minimal resolution, we determine the ai in (3).


The resulting “intersection matrix” Ei · E j is an integral symmetric tridiagonal


(r×r)-matrix. It is negative definite with det Ei ·Ej  = mult(σ).

Instead of schematically indicating the chain of curves as in Figure 17, it is customary

to depict its dual graph (see Figure 18): The dual graph has one node for each of

t t t t ... t t
−a1 −a2 −a3 −ar−1 −ar

Figure 18: Weighted dual graph of exceptional curves

the curves; two nodes are joined by a simple edge if the corresponding curves meet.
Moreover, the nodes are weighted by their self-intersection numbers.

(3) Since ∆0 is a refinement of ∆, there is a factorization X∆0 → X∆ → Xσ of π.

The first morphism X∆0 → X∆ is a finite composition of “blow ups” in (regular)
fixed points: Such a blow up of X∆ in the fixed point xτ corresponding to a 2-cone
τ = cone(b1 , b2 ) ∈ ∆ spanned by a lattice basis, corresponds to subdividing τ with
the diagonal % := ray(b1 +b2 ).

In the exceptional fibre, the blow up introduces a new chain-link Eτ ∼ = P1 of self-

intersection −1. (Such a curve that may be “contracted” to a smooth point is called
an “exceptional curve of the first kind”). The self-intersection number of each “old”
curve Ei passing through xτ drops by 1. In particular, if all exceptional curves Ei
of the resolution have self-intersection number −ai 5 −2, then the resolution is
minimal.

(4) For the minimal resolution ∆ of σ, each multiplicity ai is at least two. Furthermore,
in the situation of (3.3.3.1), the numbers ai are the integers occurring in the following
Basic Toric Geometry 43

“Hirzebruch-Jung continued fraction”

mσ 1
(3.3.5.1) = a1 − .
k 1
a2 −
.. 1
.−
1
ar−1 − ar

Proof. (3) It suffices to verify the description of blow ups for the “affine” fan ∆ generated
by the regular 2-cone τ := cone(f1 , f2 ). We have to show that the diagonal ray %
spanned by f := f1 +f2 belongs to every non-trivial regular subdivision ∆0 of τ , or,
equivalently, that a regular subdivision ∆0 of τ not containing % as an edge coincides
with ∆.
In fact, assuming % 6∈ ∆0 , then the primitive lattice vector f lies in the interior of a
regular cone τ 0 := cone(b1 , b2 ) j τ spanned by a lattice basis. We thus may write
f = n1 b1 +n2 b2 with integers n1 , n2 > 0. Since both, b1 and b2 , are non-zero, non-
negative integral linear combinations of f1 , f2 and the decomposition f = f1 + f2 is
unique, we necessarily have τ 0 = cone(b1 , b2 ) = cone(f1 , f2 ) = τ .

(4) We verify formula (3.3.5.1) by induction on r. For r = 1, we find k = 1 and

a1 = mσ . In the induction step, we apply the induction hypothesis to the cone σ 0 :=
cone(b1 , mσ b1 − kb2 ). It has multiplicity m0 = k. We write mσ = qm0 − k 0 = qk − k 0
with 0 ≤ k 0 < k and consider the lattice basis b02 := b1 , b01 := qb1 − b2 . Then a1 = q
and
m0 1
0
= a2 −
k 1
a3 −
.. 1
.−
1
ar−1 − ar

by induction hypothesis. Formula (3.3.5.1) now is an immediate consequence.

We describe the blowing up of the origin in local coordinates:

3.3.6 Example. Let ∆ be the regular fan obtained by subdividing σ = cone(f 1 , f2 ) with
% = ray(v) for v := f1 +f2 . The basic characters e1 , e2 yield coordinates (x, y) on Xσ = C2 .
For σ1 = cone(f1 , v) and σ2 = cone(v, f2 ), the dual cones are σ1∨ = cone(e1 −e2 , e2 ) and σ2∨ =
cone(e2 −e1 , e1 ), thus providing coordinates (ui , vi ) for Xσi ∼ = C2 . The inclusion σi ,→ σ then
corresponds to (u1 , v1 ) 7→ (x, y) = (u1 v1 , v1 ) and (u2 , v2 ) 7→ (x, y) = (v2 , u2 v2 ). Hence, the ui -
axes vi = 0 get collapsed to the origin. The two coordinate charts for X ∆ are glued along C∗ × C
by the transition functions u2 = 1/u1 – thus gluing together the two ui -axes to the exceptional
fibre of the blow-up map X∆ → C2 over the origin, a projective line – and v 2 = u1 v1 = v1 /u2 .
Hence, the constant function v1 = 1 is transformed into the rational function v 2 = 1/u2 with
a simple pole at the origin. This vanishing order −1 is the self-intersection number of the
exceptional curve.
44 G.Barthel, K.-H.Fieseler, L.Kaup

(II) The general case

The basic idea for the toric resolution in higher dimensions is to proceed in two steps:
Firstly, the fan is made simplicial, secondly, it is regularized. Both steps rely on the
process of “stellar subdivision”. In the following definition, we do not assume that the
ray % is an edge of the cone σ; we even allow % not to be included as a subset.

3.3.7 Definition (Stellar subdivision). Let σ be a cone and %, an arbitrary ray. Then
the union of face fans
 S
 ∆(%+τ ) if % ⊂ σ
Σ% (σ) := τ ∈∂σ , %6⊆τ

∆(σ) if % 6⊂ σ

is a fan subdividing σ, called the stellar subdivision of σ with center %.

If ∆ is a fan and % an arbitrary ray included in the support of ∆, then the stellar
subdivision of ∆ with center % is the fan
[
(3.3.7.1) Σ% (∆) := Σ% (σ) .
σ∈∆

3.3.8 Remark. (1) If the ray % even is an element of ∆, then no simplicial cone of ∆
gets subdivided.

(2) The resolution of toric surface singularities described in the proof of Lemma 3.3.3
is obtained by an iterated stellar subdivision. Similarly, the blowing up of regular
fixed points in a toric surface discussed in Remark 3.3.5 (4) is nothing but the stellar
subdivision of cone(b1 , b2 ) with respect to ray(b1 + b2 ). This procedure generalizes
from n = 2 to regular cones of arbitrary dimension n = 3, thus describing the
higher-dimensional “blowing up” of regular fixed points.

(3) For a reader familiar with the notion of “blowing up an ideal”, we add the following:
On the level of toric varieties, a stellar subdivision Σ % (σ) of a cone σ with respect to
% := ray(ν) corresponds to the blow up of a T-invariant ideal I in O(X σ ): There is a
positive integer k such that the affine hyperplane h. . . , νi = k in M R intersects each edge
of the dual cone σ ∨ not contained in %⊥ in a lattice point. Then I may be chosen as the
ideal generated by the characters corresponding to those lattice points.

We now apply the stellar subdivision to achieve the first step of a resolution.

3.3.9 Lemma (“Simplicialization”). Every fan admits a simplicial subdivision.

Proof. We introduce the following terminology: An edge % of a cone σ is said to split σ

if there is a “complementary” facet τ of σ, i.e., such that σ = τ +% ; a cone is called stout
if it has no splitting edges. A cone is simplicial if (and only if) it does not include any
stout face. If a ray % is included in a cone of a fan ∆, then Σ% (∆) \ ∆ obviously does not
Basic Toric Geometry 45

contain any stout cone. As a consequence, a stellar subdivision with center included in a
stout cone lowers the number of such cones. Hence, after finitely many subdivisions with
centers in stout cones, one arrives at a simplicial fan.

3.3.10 Remark. For such an iterated stellar subdivision ∆0 of ∆ , there exist two extreme
possibilities for the choice of the centers:

(1) We call the subdivision ∆0 thin if each center % is an edge in ∆; in other words, there
are no “new” rays. We denote by E the “exceptional locus” of the associated toric
morphism X∆0 → X∆ , that is, the union of all infinite fibres. Then each irreducible
component Ei of E is of the form V(τi0 ) where τi0 is a minimal new cone of ∆0 , so Ei
has codimension at least 2.

(2) The subdivision ∆0 is called fat if each center % is generated by a lattice vector in
the relative interior of a stout cone. Then each Ei is of codimension 1.

This follows from the fact that the irreducible components of E are the orbit closures
corresponding to cones σ ∈ ∆0 with boundary ∂σ included in ∆.

As an example, we discuss the Segre cone:

3.3.11 Example. For the three-dimensional toric variety Xσ = N (C4 ; z1 z4 − z2 z3 ) of

example 3.1.10 (2), one can show that there are exactly two different thin simplicial
subdivisions ∆1 , ∆2 of the cone σ. The fans ∆i actually are regular, and in both cases,
the corresponding exceptional set E := π −1 (xσ ) is a projective line. – We indicate such a
dividing cone τi in the next figure.
τi

Figure 19: A thin simplicial subdivision

46 G.Barthel, K.-H.Fieseler, L.Kaup

A fat simplicial subdivision for the pertinent cone σ is provided by the stellar subdi-
vision ∆0 := Σ% (σ) with center % := cone(f1 +f2 ); the exceptional fiber E is isomorphic
to the surface P1 ×P1 . We remark that X∆0 → Xσ factors through the two thin simplicial
resolutions ∆i of the preceding exercise: The fan ∆0 consists of all cones which are the
intersection of a cone in ∆1 with a cone in ∆2 . In fact, ∆0 also is a resolution.

Finally we show
3.3.12 Lemma (“Regularization”). Every simplicial fan admits a regular subdivision.
Proof. We first remark that a stellar subdivision Σ% (∆) of a simplicial fan again is sim-
plicial, and that a ray % not included in a regular cone of the fan ∆ provides an inclusion

∆reg ⊂ Σ% (∆)reg .

Obviously a fan ∆ is regular if all its maximal cones are regular.

One now successively lowers the multiplicities of maximal cones: We fix such a cone
σ of maximal multiplicity mσ = m > 1 and a minimal face τ  σ of multiplicity mτ > 1.
It is of the form τ = cone(b1 , . . . , bd ) with linearly independent primitive lattice vectors
b1 , . . . , bd . According to the regularity criterion in Remark 3.1.6, there is a (w.l.o.g.
primitive) lattice vector ν ∈ N in the parallelotope spanned by b1 , . . . , bd which is not
P
a vertex. Then necessarily ν = di=1 αi bi with rational coefficients αi strictly between 0
and 1. For % := ray(ν), we now consider the stellar subdivision Σ% (∆) and show that the
multiplicity of each new maximal cone σ̂ in Σ% (∆) is strictly less than mσ . Such a cone σ̂
includes a face of the form τ̂ = γ + % with a facet γ of τ , say γ = cone(b1 , . . . , bd−1 ). Now
an easy computation, using the fact that b1 , . . . , bd−1 are part of a basis of Nτ (∂τ being
a regular fan) shows that mτb = αd mτ < mτ . Finally there are exact sequences

0 −→ Nτ /Γτ −→ Nσ /Γσ −→ Nσ (Γσ +Nτ ) −→ 0
and

0 −→ Nτb/Γτb −→ Nσb /Γσb −→ Nσb (Γσb +Nτb) −→ 0
of finite abelian groups. Since the third terms are isomorphic, counting elements yields
that mσb = αd ·mσ < mσ .
Since each maximal cone of ∆ that includes τ may take the role of Σ, the fan Σ% (∆)
has less maximal cones of multiplicity m than the original fan ∆.
3.3.13 Remark. If the simplicialization step only consists of fat stellar subdivisions, the
subsequent regularization yields a resolution where the exceptional set

E = π −1 (S(X)) ,→ X 0

is a divisor, as it only has irreducible components of codimension 1. Being a smooth

toric variety, X 0 can be covered by invariant coordinate patches of the type Cd × (C∗ )n−d .
Hence, the invariant subvariety E intersects them in unions of coordinate hyperplanes.
Such a divisor E is called a “divisor with normal crossings”.
Basic Toric Geometry 47

P3
3.3.14 Exercise. Consider the singular simplicial cone σ = cone(f 1 , f2 , j=1 j·fj ) of dimension
three.

(1) Prove that its boundary subfan ∂σ is regular.

(2) Prove that ray f1+2(f2 +f3 ) passes through the relative interior of σ and that the stellar
subdivision with respect to it yields cones of smaller multiplicity.

References
[Bry] J.-L. Brylinski, Eventails et variétés toriques, in: Séminaire sur les Singularités des
Surfaces, Lecture Notes in Mathematics 777, Springer-Verlag, 1980.

[Cox] D. Cox, Update on toric geometry, pp. 1–41 in Geometry of Toric Varieties, L. Bonavero,
M. Brion (ed.), Séminaires et Congrès 6, Société Math. de France, 2002.

[Ew] G. Ewald, Combinatorial Convexity and Algebraic Geometry, Grad. Texts in Math. 168,
Springer-Verlag, 1996.

[FiKp] K.-H. Fieseler, L. Kaup, Algebraische Geometrie – Grundlagen, Berliner Studienreihe

zur Mathematik 13, Heldermann-Verlag, 2005.

[Ful] W. Fulton, Introduction to Toric Varieties, Annals of Math. Studies 131, Princeton
Univ. Press, 1993.

[Kp] L. Kaup, Vorlesungen über torische Varietäten, Konstanzer Schriften in Mathematik und
Informatik 130, 2001, web address: http://www.inf.uni-konstanz.de/Schriften.

[Oda] T. Oda, Convex Bodies and Algebraic Geometry, Ergebnisse Math. Grenzgeb. (n.S.) 15,
Springer-Verlag, 1988.

[Oka] M. Oka, Non-Degenerate Complete Intersection Singularity, Actualités Mathématiques,

Hermann, 1997.

Adresses
Gottfried Barthel, Ludger Kaup Karl-Heinz Fieseler
Fachbereich Mathematik und Statistik Matematiska Institutionen
Universität Konstanz, Fach D 203 Box 480
D-78 457 Konstanz SE-751 06 Uppsala
[email protected] [email protected]
[email protected]