Algebraic Geometry, autumn term 2019

Christian Böhning
Mathematics Institute
University of Warwick
ii
Contents

1 Affine and projective algebraic sets 1

2 Algebraic varieties and regular maps 9

3 Hilbert’s Nullstellensatz, primary decomposition 17

4 Segre embeddings, Veronese maps and products 27

5 Grassmannians, flag manifolds, Schubert varieties 35

6 Images of projective varieties under morphisms 45

7 Finite morphisms, Noether normalization 51

8 Dimension theory 59

9 Lines on surfaces, degree 69

10 Regular and singular points, tangent space 81

11 Cubic surfaces and their lines 91

12 Local parameters, power series methods, divisors 101

iii
iv CONTENTS
Chapter 1

Affine and projective algebraic

sets; rational normal curves,
finite point sets

Below, k is an algebraically closed field of arbitrary characteristic, e.g. C, Q̄

or also F̄p , and An is the n-dimensional affine space over k, i.e. k n as a set
(though we will want to make use of its vector space structure occasionally
as well).

Definition 1.1. For a k-vector space, P(V ) denotes the set of 1-dimensional
subspaces of V , i.e. the projective space associated to V . For the vector
space k n+1 with componentwise addition and scalar multiplication, we also
write Pn = P(k n+1 ).
Equivalently, Pn is the quotient of k n+1 − {0} by the equivalence relation:

(X0 , . . . , Xn ) ∼ (X00 , . . . , Xn0 ), (X0 , . . . , Xn ), (X00 , . . . , Xn0 ) ∈ k n+1 − {0}

if there is a λ ∈ k ∗ with λ(X0 , . . . , Xn ) = (X00 , . . . , Xn0 ). We denote the

equivalence class [(X0 , . . . , Xn )] by (X0 : · · · : Xn ) in that case and call the
Xi ’s homogeneous coordinates of the point in Pn .
Define a subset Ui ⊂ Pn by

Ui := {(X0 : · · · : Xn ) ∈ Pn | Xi 6= 0} .

Below we will often identify Ui with An via the bijection (X0 : · · · : Xn ) 7→

(i) (i)
(xj )0≤j≤n = (Xj /Xi ) (so xi = 1).

1
2 CHAPTER 1. AFFINE AND PROJECTIVE ALGEBRAIC SETS

Definition 1.2. 1. An affine algebraic set X ⊂ An is the set of zeroes of

a family (fα )α∈A , of polynomials fα ∈ k[x1 , . . . , xn ]. Since k[x1 , . . . , xn ]
is Noetherian, we can assume |A| < ∞ without loss of generality.

2. A projective algebraic set Y ⊂ Pn is the set of zeroes of a family of

polynomials (Fα )α∈A with Fα ∈ k[X0 , . . . , Xn ].

We have to say a few words what it means to be a zero in (2) above,

i.e. what is meant by F (p) = 0 for a p = (P0 : · · · : Pn ) ∈ Pn . This is so
because the homogeneous coordinates Pi of p are not unique, and p being a
zero must be independent of the representative coordinate tuple. We do this
by defining

F (p) = 0 : ⇐⇒ F (P0 , . . . , Pn ) = 0
for all (P0 , . . . , Pn ) ∈ k n+1 − {0} with [(P0 , . . . , Pn )] = p.

This leads to the conclusion that we can assume, without loss of gen-
erality, that the Fα in (2) of Definition 1.2 are homogeneous, which means
the following: S = k[X0 , . . . , Xn ] is a graded ring, which means there is a
decomposition into k-vector subspaces
M
S= Sm
m≥0

where Sm := hX0α0 ·. . .·Xnαn ik , α0 +. . . αm = m, such that Sm1 ·Sm2 ⊂ Sm1 +m2 .

Polynomials in Sm are called homogeneous of degree m. Now, since k is
infinite, if F ∈ k[X0 , . . . , Xn ] vanishes in p as above, then all homogeneous
components Fm of F with respect to the preceding direct sum decomposition
vanish in p.
Remark 1.3. If Y ⊂ Pn is a projective algebraic set, then Yi = Y ∩ Ui ⊂ An is
an affine algebraic set. To see this, consider for simplicity Y0 ; the argument in
the other cases being the same. If Y is defined by homogeneous polynomials
Fα of degree dα , then Y0 is the set of zeroes of polynomials fα (x1 , . . . , xn ),
xi = Xi /X0 , where

fα (x1 , . . . , xn ) := Fα (X0 , . . . , Xn )/X0dα = Fα (1, x1 , . . . , xn ).

Remark 1.4. Every affine algebraic subset Xi ⊂ An ' Ui ⊂ Pn is the intersec-

tion of Ui with a projective algebraic subset X ⊂ Pn . Again we show that for
 3

U0 only since the other cases are only notationally different. If X0 is defined
by fα (x1 , . . . , xn ) of degree dα (of course not necessarily homogeneous now),
we can define a suitable X by
 
dα X1 Xn
Fα (X0 , . . . , Xn ) := X0 fα ,..., .
X0 X0

We can summarize the preceding two remarks by saying that X ⊂ Pn is

a projective algebraic subset if and only if each X ∩ Ui ⊂ Ui ' An is an affine
algebraic set.

Example 1.5. 1. If W ⊂ k n+1 is an (m+1)-dimensional sub-vector space,

then P(W ) ⊂ Pn is a projective algebraic set, called an m-dimensional
projective linear subspace (for m = 1: line, m = 2: plane, m = n − 1:
hyperplane).

2. Of course zeroes of a single homogeneous polynomial F ∈ k[X0 , . . . , Xn ]

(of degree d, say) are a projective algebraic set; this is called a hy-
persurface. We can assume F without multiple factors (note that
k[X0 , . . . , Xn ] is factorial). Then we call d the degree of the hyper-
surface.

3. For a more interesting example, let C be the image of the map

ν : P1 → P3
(X0 : X1 ) 7→ (X03 : X02 X1 : X0 X12 : X13 ) := (Z0 : Z1 : Z2 : Z3 ).

Then C is contained in the three quadrics Q0 , Q1 , Q2 defined as the

zero sets of

F0 (Z) = Z0 Z2 − Z12
F1 (Z) = Z0 Z3 − Z1 Z2
F2 (Z) = Z1 Z3 − Z22

and those define C, i.e. the zero set of them is exactly C (to see this,
note that if p ∈ P3 , p = (P0 : P1 : P2 : P3 ) lies on Q0 ∩ Q1 ∩ Q2 , then
P0 6= 0 or P3 6= 0; in the former case, P = ν((P0 : P1 )), in the latter
case, P = ν((P2 : P3 )).
4 CHAPTER 1. AFFINE AND PROJECTIVE ALGEBRAIC SETS

We continue the study of (3) in the above examples a little: although we

haven’t introduced any notion of “dimension” yet into our geometric study of
algebraic sets, it is intuitively plausible that C should be a one-dimensional
thing, a curve. It is called a twisted cubic curve. It is remarkable that two of
the above quadratic equations do not define C, more generally:
Proposition 1.6. For λ = (λ0 , λ1 , λ2 ) ∈ k 3 − {0}, write
Fλ = λ0 F0 + λ1 F1 + λ2 F2 ,
the Fi being as in Example 1.5, (3). Denote the projective algebraic set that
Fλ defines by Qλ . Then for [µ] 6= [ν] ∈ P2 , we have that Qµ ∩ Qν is equal to
the union of C and a line Lµν intersecting C in two points.
Proof. C is defined by the 2 × 2 minors of
 
Z0 Z1 Z2
Z1 Z2 Z3
and Qµ is the determinant of
 
Z0 Z1 Z2
Z1 Z2 Z3 
µ00 µ01 µ02
where the tuple (µ00 , µ01 , µ02 ) agrees with µ after a signed permutation. So the
locus outside of C where Fµ and Fν vanish, is the rank ≤ 2 locus of
 
Z0 Z1 Z2
Z1 Z2 Z3 
 0 
 µ0 µ01 µ02 
ν00 ν10 ν20
(where in addition the first two rows are independent). For [µ] 6= [ν], this
locus is the same as the one defined by
   
Z0 Z1 Z2 Z1 Z2 Z3
det  µ00 µ01 µ02  = det  µ00 µ01 µ02  = 0,
ν00 ν10 ν20 ν00 ν10 ν20
i.e. Qµ ∩ Qν = C ∪ Lµν where Lµν is the line defined by the last two deter-
minants. The intersection of Lµν with C is then given by Qλ and the two
linear equations above where hQλ , Qµ , Qν i = hF0 , F1 , F2 i.
 5

Proposition 1.7. There exists a homogeneous quadratic polynomial Q(Z0 , . . . , Z3 )

and a homogeneous cubic polynomial P (Z0 , . . . , Z3 ) whose common zeroes are
precisely C.

Proof. One can take

 
  Z0 Z1 Z2
Z0 Z1
Q(Z) = det , P (z) = det Z1 Z2 Z3  .
Z1 Z2
Z2 Z3 Z0

Namely, if the vector (Z0 , Z1 , Z2 ) and (Z1 , Z2 , Z3 ) are linearly dependent,

then these determinants vanish, and the converse holds as well: if the first
two rows of the matrix whose determinant defines P (z) are independent, and
the determinant vanishes, then the last row of the matrix must be a linear
combination of the first two rows. But then (Z2 , Z3 ) is dependent on (Z0 , Z1 )
and (Z1 , Z2 ), whence the rank of the submatrix consisting of the first two
rows would be 1 (taking into account Q(z) = 0), contradiction.

Thus we arrive at the curious fact that C, as a set, can be defined by

two polynomials, but if we look at the ideal I(C) ⊂ k[X0 , . . . , X3 ] of all
polynomials vanishing on C, this cannot be generated by 2 elements (since
dim I(C)2 ≥ 3, but dim I(C)1 = 0: clearly, C does not lie in a hyper-
plane since X03 , X02 X1 , X0 X12 , X13 are independent). One says that C is a set-
theoretic complete intersection, but not a complete intersection in P3 . There
are many open problems connected with these notions; e.g., one knows that
the union of two planes intersecting only in 0 is not a set-theoretic complete
intersection in A4 , but one does not know if every curve in P3 is a set-theoretic
complete intersection.
Let us consider the lines Lµν more closely:

Proposition 1.8. Every line L ⊂ P3 connecting two points two points P, Q ∈

C occurs among the Lµν .

Proof. Choose R ∈ L distinct from P, Q. The three-dimensional vector space

of Fλ ’s contains a two-dimensional subspace consisting of those vanishing in
R; suppose Fµ0 , Fν0 is a basis. But then the latter two polynomials vanish
on C, hence on P, Q, R. Since they are quadratic, they are then identically
zero on L. Whence Lµ0 ν0 = L.
6 CHAPTER 1. AFFINE AND PROJECTIVE ALGEBRAIC SETS

We call every algebraic set in P3 projectively equivalent to C a twisted

cubic curve. I.e., twisted cubic curves are precisely the images of maps P1 →
P3 given by [X] 7→ (A0 (X) : · · · : A3 (X)] where the Ai (X)’s are a basis of
k[X0 , . . . , X3 ]3 . Generalizing this, we make
Definition 1.9. Every curve in Pd projectively equivalent to the image of

νd : P1 → Pd ,
(X0 : X1 ) 7→ (X0d : X0d−1 X1 : · · · : X1d )

is called a rational normal curve in Pd .

Remark 1.10. 1. To make sense of the above definition, remark that the
image of νd is really a projective algebraic set. One can take Fij (Z) =
Zi Zj − Zi−1 Zj+1 , 1 ≤ i ≤ j ≤ d − 1, as a set od polynomials defining
the image. The word “curve” in the above definition so far has no
real mathematical meaning, but once we have introduced dimension, a
rational curve will indeed be 1-dimensional.
2. Every set of d + 1 points on a rational normal curve are linearly inde-
pendent (Van der Monde determinant).
Theorem 1.11. Through every set of d + 3 points in general position in Pd ,
there passes a unique rational normal curve.
Here “general position” means that the assertions holds for tuples of
points in a nonempty subset of (Pd )d+3 defined by some polynomial inequal-
ities.
Proof. We prove existence first. We can assume, after applying a projectivity,
are (1 : 0 : · · · : 0), . . . , (0 : · · · :
that the first d + 1 of the points, p1 , . . . , pd+1 , Q
1) (coordinate points). Putting G(X0 , X1 ) = d+1 k=1 (νk X0 − µk X1 ) and

G(X0 , X1 )
Hi := ,
(νi X0 − µi X1 )
we get that the image of νd given by

(X0 : X1 ) 7→ (H1 (X0 , X1 ) : · · · : Hd+1 (X0 , X1 ))

passes through the coordinate points, namely maps (µi : νi ) ∈ P1 to the i-th
coordinate point. We can also assume (µi : νi ) are different from (1 : 0) and
 7

(0 : 1); then, given two general additional points pd+2 , pd+3 in Pd , we can
always adjust the µi , νi so that (1 : 0) maps to pd+2 and (0 : 1) maps to pd+3 .
This proves existence.
For uniqueness, note that every rational normal curve passing through the
coordinate points p1 , . . . , pd+1 is the image of a map given by polynomials Hi
as above for certain (µi , νi ). Applying a projectivity in the source, we can also
assume that (1 : 0) and (0 : 1) map to pd+2 , pd+3 . Then such a rational normal
curve is given by polynomials Hi as above, and moreover, the µi are fixed up
to simultaneous rescaling by a constant nonzero factor α, and so are the νj
up to a factor β. Applying the projectivity (X0 : X1 ) 7→ (α−1 X0 : β −1 X1 ) in
the source, we see that the maps corresponding to different α, β all have the
same image. Hence this is the unique rational normal curve meeting all the
requirements.
Another example of projective algebraic sets are finite point sets Γ =
{p1 , . . . , pN } ⊂ Pn : indeed, if q ∈
/ Γ, there is a polynomial vanishing in Γ,
but not in q (take a product of N linear forms). Hence Γ is defined by
polynomials of degree ≤ N .
It is known from courses in linear algebra that

1. Two ordered point sets (p1 , . . . , pn+2 ), (q1 , . . . , qn+2 ) in general posi-
tion can be transformed into each other by a unique projectivity g ∈
PGLn+1 (k).
2. In P1 , one can transform 4 = 1 + 3 ordered points in general position
into another ordered four points in general position if and only if their
cross ratios
(z1 − z2 )(z3 − z4 )
(z1 − z3 )(z2 − z4 )
are the same.

So when can (n + 3) general ordered points (p1 , . . . , pn+3 ) in Pn be trans-

formed into another tuple (q1 , . . . , qn+3 ) of (n + 3)-points?By Theorem 1.11,
we can find rational normal curves through both sets of points. The maps
defining the rational normal curves allow us to interpret these point sets as
point sets
(p01 , . . . , p0n+3 ), (q10 , . . . , qn+3
0
)
on P1 . By the uniqueness property of rational normal curves, and since every
algebraic automorphism of P1 is a projectivity (cf. all biholomorphic maps
8 CHAPTER 1. AFFINE AND PROJECTIVE ALGEBRAIC SETS

of the Riemann sphere are fractional linear), one gets that (p1 , . . . , pn+3 ) and
(q1 , . . . , qn+3 ) are projectively equivalent on Pn if and only if (p01 , . . . , p0n+3 )
0
and (q10 , . . . , qn+3 ) are projectively equivalent on P1 . The latter means that

crossratio(p01 , p02 , p03 , p0i ) = crossratio(q10 , q20 , q30 , qi0 ), ∀ i = 4, . . . , n + 3.

Chapter 2

The sheaf of regular functions;

algebraic varieties and regular
maps

We can equip our algebraic sets with a topology.

Definition 2.1. The Zariski topology of Pn (resp. of An ) is the topology

whose closed sets are the projective (resp. affine) algebraic sets; we equip
every projective (resp. affine) algebraic set X in Pn (resp. An ) with the
induced topology, and call this the Zariski topology of X.

Remark 2.2. 1. If k = C or another field with an interesting topology,

e.g. the p-adic numbers Qp , then one can equip k n+1 with the corre-
sponding product topology, and Pn with the quotient topology via the
map k n+1 − {0} → Pn . Then all algebraic sets carry another “strong”
topology, which is topologically more relevant. The Zariski topology is
just convenient to talk about sets where polynomials vanish or do not
vanish, but does not carry too much information otherwise.

2. In Pn , the sets UF := {F 6= 0}, where F ranges over all homogeneous

polynomials, form a basis of the topology. In An , we have an analogous
basis Uf = {f 6= 0} with f ranging over all polynomials in k[x1 , . . . , xn ].

Definition 2.3. An open subset U ⊂ X, X a projective algebraic set in Pn ,

will be called a quasi-projective algebraic set.

9
10 CHAPTER 2. ALGEBRAIC VARIETIES AND REGULAR MAPS

In what follows, we will define a notion of local functions on affine, pro-

jective, quasi-projective algebraic sets, equipped with the Zariski topology.
Our objects of study in the sequel, called algebraic varieties, will then
(formally) be triples
(X, T, O)
where X is an algebraic set of some sort, T the Zariski topology on it, and
O the local functions.
One may justifiably ask: why such complicated contortions to define the
notion of a variety, which is meant to be a geometric object associated to
a bunch of polynomial equations? Why is O necessary? Why is even T
necessary?
The answer is that experience has shown that once one studies objects
of a certain type in mathematics, which are often sets with an additional
structure, one should at the same time study “maps” between those objects
that preserve the given structure. Thus one studies vector spaces along with
linear maps, groups along with group homomorphisms, rings with ring homo-
morphisms, or topological spaces with continuous maps. To get a meaningful
and interesting notion of maps in our theory of zero sets of polynomials, we
will need O, and to define it, it is convenient to have T first.
For example, suppose you would take algebraic sets together with their
Zariski topology as the fundamental structured objects of your theory, and
forget (or never learn about) O. The structure-preserving maps are then con-
tinuous maps between those topological spaces. Now look more specifically
at irreducible curves in A2 , i.e. zero sets of a single irreducible polynomial
f (x, y) where f ∈ Q̄[x, y] (we take Q̄ for simplicity). Then all of these turn
out to be homeomorphic! Namely, as sets, all such curves have the same
cardinality, they are countable since Q̄ is. Any bijection is even a homeo-
morphism since the nonempty proper Zariski open subsets are complements
of finite sets of points in this case (this uses the irreducibility of the f (x, y)).
But there are facts of an algebraic nature that strongly suggest that regard-
ing all irreducible curves in A2 as essentially the same (isomorphic) is much
too coarse and crude: for example, consider the question whether there exist
nonconstant rational functions

ϕ(t), ψ(t) ∈ Q(t)

with
f (ϕ(t), ψ(t)) = 0
 11

(i.e., a rational parametrization of the irreducible curve associated to f ).

Then it turns out that the general curve of degree deg(f ) > 2 does not admit
such a parametrization, but all degree 1 and 2 curves do. Moreover, there are
special irreducible curves of any degree that do admit such a parametrization.
We certainly don’t want to build our theory in such a way as to put all those
algebraically totally different curves in one bag! That is why we need O.
Definition 2.4. Let X be a topological space, k a field. For U ⊂ X open,
we denote by
Maps(U, k)
the set of all functions (continuous or not) of U to k. A sheaf of k-valued
functions on X is the datum, for any open U ⊂ X, of a subset

OX (U ) ⊂ Maps(U, k)
S
such that the following is true: if U ⊂ X is open and U = i∈I Vi a cover
of U by open subsets Vi ⊂ U ⊂ X, i ∈ I, I some index set, then a function
f ∈ Maps(U, k) belongs to OX (U ) if and only if all restrictions f |Vi belong
to OX (Vi ) (for all i ∈ I).
This may seem intimidating at first sight, but is totally simple really:
it just means that on any open set U in X we mark certain functions as
distinguished, by painting them red, say; we call these OX (U ). The condition
then means that being “red” is a local property: if we restrict one of our
functions to the open sets of a cover, and if it is red on every one of them,
then it is red globally.
Example 2.5. 1. Let X = R1 with the Euclidean topology, and put for
U ⊂ R open, OX (U ) = {f : U → R | f continuous}. Then OX is a
sheaf of R-valued functions. The property of being continuous is local.

2. For X as in (1), let OX (U ) = {f : U → R | f differentiable}. Then OX

is a sheaf of R-valued functions. The property of being differentiable is
local.

3. Let X = ∆1 t∆2 be the disjoint union of two open discs in R2 , with the
induced topology from R2 . For U ⊂ X open, let OX (U ) = {f : U → R |
f constant}. Then OX is not a sheaf: the property of being constant
is not local. For example, the function which is 1 on ∆1 and 0 on ∆2
is constant locally, but not constant globally.
12 CHAPTER 2. ALGEBRAIC VARIETIES AND REGULAR MAPS

4. Let X = R1 again, OX (U ) = {f : U → R | f bounded}. Then OX is

not a sheaf: the property of being bounded is not local. For example,
f (x) = exp(x) is bounded locally, but not globally.

Definition 2.6. Let X be a topological space with a sheaf of functions OX ,

and Y ⊂ X a subspace. Then put for V ⊂ Y open:

(OX ) |Y (V ) := {f : V → k | ∀ v ∈ V ∃ U (v) ⊂ X open, U (v) 3 v,

o
˜ ˜
∃ fv ∈ OX (U (v)) : fv |U (v)∩V = f |U (v)∩V .

Then (OX ) |Y is a sheaf of functions on Y , the sheaf induced by OX on Y

by restriction.

Definition 2.7. Let (X, OX ) and (Y, OY ) be spaces with sheaves of func-
tions. A morphism
ϕ : (X, OX ) → (Y, OY )
is a continuous map ϕ : X → Y with the property that for all U ⊂ Y open,
and for all f ∈ OY (U ), one has ϕ∗ (f ) := f ◦ ϕ ∈ OX (ϕ−1 (U )).

We now define sheaves of functions on An and Pn ; by the construction

of Definition 2.6, we can turn all quasi-projective algebraic sets (with their
Zariski topology) into spaces with sheaves of functions.

Definition 2.8. For U ⊂ An (Zariski-)open, put

OAn (U ) := {f : U → k | ∀ x ∈ U ∃ V (x) ⊂ U open,

x ∈ V (x), ∃ polynomials p, q ∈ k[x1 , . . . , xn ], q(y) 6= 0 ∀ y ∈ V (x)

p
such that : f |V (x) = |V (x) .
q

For U ⊂ Pn open, put

OPn (U ) := {f : U → k | ∀ x ∈ U ∃ V (x) ⊂ U open,

x ∈ V (x), ∃ homogeneous polynomials of the same degree p, q ∈ k[X0 , . . . , Xn ],

p
q(y) 6= 0 ∀ y ∈ V (x) such that : f |V (x) = |V (x) .
q

These define sheaves of functions on the respective spaces.

 13

This also looks complicated, but means only that the local functions are
those that can be written, locally, as quotients of two polynomials (homo-
geneous of the same degree in the case of Pn ), with the denominator non-
vanishing on the open under consideration.

Definition 2.9. Suppose X ⊂ An is an affine algebraic subset, or X ⊂ Pn

is a projective or quasi-projective algebraic subset together with its Zariski
topology T. We define OX as (OAn ) |X resp. (OPn ) |X . Then the triple
(X, T, OX ) is called an affine/projective/quasi-projective algebraic vari-
ety. Moreover, slightly more generally, any topological space with a sheaf of
k-valued functions that is isomorphic, as a space with a sheaf of functions,
to an affine/projective/quasi-projective algebraic variety, will be called an
affine/projective/quasi-projective algebraic variety itself. If we use algebraic
variety without further qualification, we mean the most general class intro-
duced, a quasi-projective variety.
The elements in OX (U ) for U ⊂ X open, are called regular functions on U .
For a variety X, an open subset U ⊂ X as well as a closed subset Y ⊂ X are
varieties in their own right, the inclusions are morphisms. We call U an open
subvariety in this case, and Y a closed subvariety of X. Subvariety without
qualification will mean closed subvariety.

Remark 2.10. With these definitions, the natural projection An+1 −{0} → Pn
becomes a morphism, and the natural bijections of the subsets Ui = {Xi 6=
0} ⊂ Pn with An become isomorphisms, as it should be.

Definition 2.11. If (X, OX ) is a space with a sheaf of functions, x ∈ X, put

OX,x := {equivalence classes of pairs (f, U )

where U 3 x is open, f ∈ OX (U ), and
(f, U ) ∼ (g, V ) : ⇐⇒ ∃W ⊂ U ∩ V open, x ∈ W such that f |W = g |W } .

We call OX,x the stalk of OX in x; elements of OX,x are called germs (of
functions around x).

These definitions are very slick and smooth, and leave us with a beautiful
category of algebraic varieties, but the drawback is that it takes quite a
while to develop an intuition for what they mean and to work with them.
For example, what is OAn (An ) or OPn (Pn )? Here is a useful example of a
morphism.
14 CHAPTER 2. ALGEBRAIC VARIETIES AND REGULAR MAPS

Example 2.12. Let X, Y, Z be homogeneous coordinates in P2 , and C the

plane cubic curve given by

C : ZY 2 = X(X 2 − Z 2 ).

Let P0 = (0 : 0 : 1) ∈ C. For a (variable) point P ∈ C, let l be the line

through P0 and P and α(P ) the third intersection point of l with C. Then
the assignment P 7→ α(P ) defines an automorphism of C of order 2, α2 = id.
To see this, first note that α is well-defined as a map of point sets: the
polynomial defining C restricted to l ' P1 factors, and will have one zero ζP0
corresponding to P0 , one zero ζP corresponding to P , and a third zero ζα(P )
determining α(P ) uniquely, when we take into account multiplicities.
Consider CZ := C ∩ {Z 6= 0} and take coordinates x = X/Z, y = Y /Z
on A2 ' UZ := {(X : Y : Z) ∈ P2 | Z 6= 0}. Then CZ has an equation

y 2 = x(x2 − 1).

If P = (a, b) ∈ CZ , then the line l through P0 and P is given by x = at,

y = bt. Substituting yields

b2 t2 = at(a2 t2 − 1)
0 = at(t − 1)(a2 t + 1)

So we get the third intersection point for t = −1/a2 , whence

 
1 b
α : (a, b) 7→ − , − 2 .
a a

However, this is not a well-defined map of UZ ' A2 into itself because the
formula makes no sense for (a, b) = (0, 0). However, (draw a picture!) the
point (0, 0) should map to P1 = (0 : 1 : 0), the unique point at ∞ on C with
respect to the coordinates x, y! To prove that α is a morphism, we must
rather cover C with various open sets as follows: put

P1 = (0 : 1 : 0),
P2 = (0 : 0 : 1)(= P0 ),
Q1 = (1 : 0 : 1),
Q2 = (−1 : 0 : 1)
 15

From the geometric definition, α exchanges P1 , P2 and Q1 , Q2 . Put

U1 := C − {P1 , P2 }
U2 := C − {P1 , Q1 , Q2 }
U3 = C − {P2 , Q1 , Q2 }
V1 := C − {P1 } = C ∩ UZ = CZ
V2 = C − {P2 , Q1 , Q2 } = C ∩ {Y 6= 0} = C ∩ UY = CY .

Then
1. U1 , U2 , U3 is an open cover of C.
2. V1 , V2 is an affine open cover of C.
3. We have
α(U1 ) ⊂ V1 , α(U2 ) ⊂ V2 , α(U3 ) ⊂ V1 .

Let x, y be coordinates in UZ (resp. V1 ) as above, and s = X/Y , t = Z/Y

coordinates in UY (resp. V2 ). Then
(A) The map α |U1 : U1 → V1 is given by
 
1 b
(a, b) 7→ − , − 2 .
a a
Note that the right hand side are polynomials in a, b, 1/a and that
x, y, 1/x ∈ OC (U1 ), so α |U1 is a morphism.
(B) To describe the map α |U2 : U2 → V2 , let us compute the image under
α of a point (x, y) = (a, b) ∈ U2 (x, y are also coordinates on U2 ⊂ V1 )
in terms of the coordinates s, t on V2 ; since
x
s= , t = y −1
y
we get
a a2
s(α(a, b)) = t(α(a, b)) = −
b b
2 2
and since b = a(a − 1), this can be rewritten as
 
b ab
α |U2 : (a, b) 7→ (s, t) = ,− .
a2 − 1 a2 − 1
16 CHAPTER 2. ALGEBRAIC VARIETIES AND REGULAR MAPS

Note that x, y, 1/(x2 − 1) ∈ OC (U2 ) and the right hand side of the last
displayed equation is a polynomial in a, b, 1/(a2 − 1), so α |U2 : U2 → V2
is a morphism.

(C) Finally, consider α |U3 : U3 → V1 . We have to compute the image of a

point with coordinates (s, t) = (c, d) in U3 in terms of the coordinates
x, y on V1 : since
s
x = , y = t−1
t
we get
d d
x(α(s, t)) = − y(α(s, t)) = − 2
c c
2 2
and since d = c(c − d )

α |U3 : (s, t) = (c, d) 7→ (x, y) = d2 − c2 , d(c2 − d2 ) − c .

The right hand side are polynomials in c, d and s, t ∈ OC (U3 ), so

α |U3 : U3 → V1 is a morphism.

Hence α is a morphism itself.

This example shows that, because the requirement of being a morphism is
local, checking this in practice may require passing to suitable covers by open
sets and finding formulas for the map locally in terms of rational functions
of the local coordinates that are suitably regular in the open sets under
consideration.
We will see later that there are instances where it is easier to check that
something is a morphism differently.
Chapter 3

Hilbert’s Nullstellensatz,
primary decomposition and
geometric applications

To understand varieties, their morphisms and regular functions better, we

need a little more background in commutative algebra.
An affine subvariety X ⊂ An in An is given by the zero set of a family
of polynomials (fα )α∈A , f ∈ A := k[x1 , . . . , xn ]. We write I for the ideal
generated by the fα in A and X = V (I), V for “Verschwindungsmenge”, the
German for zero set, or vanishing set. For any ideal J ⊂ A, we thus use the
notation

V (J) := {(x1 , . . . , xn ) ∈ An | f (x1 , . . . , xn ) = 0 ∀ f ∈ I}.

Definition 3.1. We write

I(X) = {g ∈ A | g(x1 , . . . , xn ) = 0 ∀(x1 , . . . , xn ) ∈ X}

for the ideal of all polynomials vanishing on X. We call it the ideal of X for
short.

Theorem 3.2 (Hilbert’s Nullstellensatz). We have

√
I(V (I)) = I := {f ∈ A | ∃ n : f n ∈ I}.
√
Here I is called the radical of the ideal I.

17
18CHAPTER 3. HILBERT’S NULLSTELLENSATZ, PRIMARY DECOMPOSITION
√
Proof. It is clear that I ⊂ I(V (I)), so it suffices to prove the opposite
inclusion.
Step 1. We prove that if V (I) = ∅, then I = A. Suppose by contradiction
that I ( A. Then I ⊂ m for some maximal ideal in A. Now Step 1 will be
complete once we can show that every maximal ideal of A is of the form
m = (x1 − a1 , . . . , xn − an )
for some ai ∈ k. Indeed, then V (I) will contain the point (a1 , . . . , an ), hence
not be empty. Now the assertion that every maximal ideal is of the form
above is equivalent to the claim that the quotient field
L = k[x1 , . . . , xn ]/m
is isomorphic to k via the inclusion k ⊂ L, and since k is algebraically closed,
this is in turn equivalent to showing that L is algebraic over k. We can
assume that x1 , . . . , xl ∈ L are algebraically independent over k and that
xl+1 , . . . , xn are algebraic over k(x1 , . . . , xl ) ⊂ L.
Lemma 3.3. Let R be a Noetherian ring, S ⊃ R a finitely generated R-
algebra. If T ⊂ S is an R-algebra such that S is a finitely T -module, then T
is a finitely generated R-algebra.
Let us first indicate how the Lemma allows us to finish the proof of Step
1. Apply it with R = k, S = L, T = k(x1 , . . . , xl ). The hypotheses are
then satisfies, hence we would get that k(x1 , . . . , xl ) is a finitely generated
k-algebra, which is absurd unless l = 0: if rational functions z1 , . . . , zN
Pi (x1 , . . . , xn )
zi =
Q(x1 , . . . , xn )
were generators, and if f ∈ k[x1 , . . . , xn ] is irreducible, then 1/f must be a
polynomial function of the zi . But then f would have to divide one of the
Qi since A = k[x1 , . . . , xn ] is a UFD, leading to the absurd conclusion that
there are only finitely many irreducible polynomials in A.
Proof. (of Lemma 3.3) Let ξ1 , . . . , ξp be R-algebra generators of S and let
s1 , . . . , sq be T -module generators of S. Thus ∃ tικ , t0ικλ ∈ T such that
X
ξι = tικ sκ (3.1)
κ
X
sι · sκ = t0ικλ sλ . (3.2)
λ
 19

Let T0 be the R-sub-algebra of T which is generated by the tικ , t0ικλ over R.

Then S is finitely generated as a T0 -module (clear because of formulas (3.1)
and (3.2)), and then T is finitely generated as a T0 -module: indeed, T0 is
Noetherian since R is, and a sub-module of a finitely generated module over
a Noetherian ring is finitely generated. Hence the claim.
Step 2. Having completed Step 1, let us show how it quickly implies the
entire assertion of the Nullstellensatz. We have to show that if f ∈ I(V (I)),
then for some integer m > 0: f m ∈ I. Put

J := hI, xn+1 · f (x1 , . . . , xn ) − 1i ⊂ k[x1 , . . . , xn ]

(the brackets indicate to take the ideal generated). Then V (J) = ∅, thus by
Step 1, J = k[x1 , . . . , xn ]. Thus if

B := k[x1 , . . . , xn+1 ]/(xn+1 · f − 1),

P
then I · B = (1), so there is an equation 1 = gi ai with gi ∈ I, ai ∈ B in
B. Hence in the ring B we get an equation

1 = h0 + h1 xn+1 + · · · + hm xm
n+1 , hi ∈ I.

Thus in the ring A = k[x1 , . . . , xn ] we have

f m = f m h0 + · · · + hm

and this is in I.
We can use the above to describe rings of regular functions in several
cases more simply:

Corollary 3.4. We have

OAn (An ) ' k[x1 , . . . , xn ]

and more generally: if X ⊂ An is a closed subvariety and I(X) is prime (we

will shortly interpret this simplifying hypothesis geometrically), then

OX (X) ' k[x1 , . . . , xn ]/I(X).

One customarily calls k[x1 , . . . , xn ]/I(X) the affine coordinate ring of X.

20CHAPTER 3. HILBERT’S NULLSTELLENSATZ, PRIMARY DECOMPOSITION

Proof. Suppose f ∈ OX (X). By definition, this means that there is an open

cover {Uα } of X such that
hα
f |Uα = |U , kα 6= 0 on Uα , hα , kα ∈ k[x1 , . . . , xn ].
kα α
Since every Zariski open set on X is a union of the sets Ug = {x ∈ X |
g(x) 6= 0}, g ∈ k[x1 , . . . , xn ], and since the fact that a set of Ugα covers is
equivalent, by the Nullstellensatz, to the ideal generated by the gα being the
unit ideal in k[x1 , . . . , xn ]/I(X), we can assume that Uα = Ugα , some finitely
many gα ∈ k[x1 , . . . , xn ]. So α ∈ A, A an index set with |A| < ∞.
The ideal (kα ) has no zero on X, hence, by the Nullstellensatz again,
(kα , I(X)) = (1). Thus there is an equation (which can be viewed as an
algebraic partition of unity, if you are familiar with partitions of unity from
Differential Topology)
X
1= lα kα in k[x1 , . . . , xn ]/I(X).
α∈A
P
Put Pf := α lα hα . Then
X X
k β Pf = lα hα kβ = lα hβ kα = hβ
α∈A α∈A

on Uβ , since
hα hβ
= in Quot(k[x1 , . . . , xn ]/I(X)),
kα kβ
where Quot(k[x1 , . . . , xn ]/I(X)) is the quotient field of k[x1 , . . . , xn ]/I(X).
Namely, hα kβ − hβ kα vanishes on the open Uα ∩ Uβ in X, hence is in I(X),
since I(X) is prime.
Remark 3.5. The statement of Corollary 3.4 also holds without the assump-
tion that I(X) is prime, but the proof is a bit messier then and we omit
it.
Corollary 3.6. We have OPn (Pn ) = k.
Proof. Because of Corollary 3.4, we have in An ' U0 = {X0 6= 0} ⊂ Pn that
an f ∈ OPn (Pn ) can be written as
 
X1 Xn
f =p ,..., , p ∈ k[x1 , . . . , xn ].
X0 X0
 21

As a rational function in k(X0 , . . . , Xn ), f is the quotient r/s of two ho-

mogeneous polynomials r, s of the same degree with the denominator not
vanishing identically on X0 = 0; hence
 
deg(p) deg(p) X1 Xn
rX0 = X0 p ,..., s,
X0 X0
deg(p)
but X0 does not divide sand X0 divides the polynomial
 
deg(p) X1 Xn
X0 p ,...,
X0 X0

only when p is constant.

The same method of proof even shows that OPn (Pn − H1 ∩ H2 ) = k where
H1 , H2 are two different hyperplanes in Pn .

Corollary 3.7. Let X ⊂ An , Y ⊂ Am be closed affine subvarieties. Every

morphism f : X → Y is of the form

f = (f1 (x1 , . . . , xn ), . . . , fm (x1 , . . . , xn ))

with fi ∈ k[x1 , . . . , xn ].

Proof. This follows from Corollary 3.4 (and Remark 3.5) since Tj ◦ f , Tj the
j-th coordinate function on Am , is regular on X.

Theorem 3.8. Every radical ideal I ⊂ A = k[x1 , . . . , xn ] is a finite intersec-

tion of prime ideals pi with pi 6⊂ pj for i 6= j.

Proof. Let I0 be a maximal element in the family of all radical ideals I ⊂ A

which are not a finite intersection of prime ideals.Then I0 is clearly not prime
itself. Pick a, b ∈ A with a, b 6∈ I0 such that ab ∈ I0 . Put
p p
I1 = (I0 , a), I2 = (I0 , b).

By construction, I1 , I2 are finite intersections of prime ideals. But I0 = I1 ∩I2 ,

contradiction. (To see that really I0 = I1 ∩ I2 , take f ∈ I1 ∩ I2 ; then there
are integers m, n > 0 with f m ∈ (I0 , a) and f n ∈ (I0 , b) whence f m+n ∈ I0
because ab ∈ I0 . Consequently, f ∈ I0 ).
22CHAPTER 3. HILBERT’S NULLSTELLENSATZ, PRIMARY DECOMPOSITION

Thus we have proved existence of a representation of any radical ideal as

an intersection of prime ideals; uniqueness is easy under the above irredun-
dancy hypothesis: if \ \
I= pi = qj ,
i j
T
then clearly for all i we have pi ⊃ j qj . Then qk ⊂ pi for some k. Vice
versa, we also have pl ⊂ qk for some l. By irredundancy, pl = pi = qk . So
there is a one-one correspondence between the p’s and q’s.

Definition 3.9. A variety X is called irreducible if for all closed subvari-

eties Y, Z ⊂ X with X = Y ∪ Z, we have Y = X or Z = X. Otherwise, X
is called reducible.

Remark 3.10. An affine subvariety X ⊂ An is irreducible if and only if I(X)

is prime. Indeed, if Y ( X and Z ( X are proper subvarieties, there is an
f ∈ I(Y ), f 6∈ I(X) and there is a g ∈ I(Z), g 6∈ I(X). If X = Y ∪ Z, then
f g ∈ I(X), so I(X) is not prime. And conversely, if f , g are some elements
in k[x1 , . . . , xn ] with f g ∈ I(X), then X = V (I(X), f ) ∪ V (I(X), g). So if
I(X) is not prime, then X is reducible.
The preceding remark holds also for projective subvarieties X ⊂ Pn with
the following adjustments: first, putting S = k[x0 , . . . , xn ], a graded ring, we
can consider
I(X) := {F ∈ S | F ≡ 0 on X}.
L
This is a homogeneous ideal in S = m≥0 Sm , which means that if a is in I,
all its homogeneous components with respect to the direct sum decomposition
of S are. We call I(X) the homogeneous ideal of X. Then

S/I(X)

is another graded ring, the so-called homogeneous coordinate ring of X. It is

easy to see that X is irreducible if and only if I(X) is prime (with the same
proof).
Because of Theorem 3.8 we get immediately

Theorem 3.11. Every (possibly reducible) variety X is a finite union of

irreducible subvarieties Xi with Xi ( Xj for i 6= j, and in a unique way. The
Xi are called the irreducible components of X.
 23

This is a simply the geometric translation of Theorem 3.8, but to be sure

that everything works in the projective case as well one has to remark: if
I ⊂ S is homogeneous and radical, and

I = p1 ∩ · · · ∩ ps

a representation as an intersection of primes, then all the pi are homogeneous

ideals as well. For, I being homogeneous is equivalent to

I = I λ := {f (λx0 , . . . , λxn ) | f ∈ I}

for infinitely many λ ∈ k ∗ . This implies that for every i there is a j with
pi = pλj for infinitely many λ = λ1 , λ2 , . . . since applying (−)λ gives another
representation of I as an irredundant intersection of primes, and it is unique
λ λ−1
up to reordering. But then pi j 1 = pi for all j, hence pi is homogeneous.
Remark 3.12. As an exercise, one should formulate and prove an extension of
the Nullstellensatz for projective varieties now. Moreover, one should verify
that

OAn (Uf ) = k[x1 , . . . , xn ]f (localization),

OPn (UF ) = k[x0 , . . . , xn ](F ) (homogeneouslocalization)

where f ∈ k[x1 , . . . , xn ] is a polynomial, F ∈ k[x0 , . . . , xn ], and Uf ⊂ An

resp. UF ⊂ Pn are the principal open subsets where f resp. F do not vanish.
The same holds for any affine resp. projective subvarieties.
Corollary 3.7 gives a nice characterization of morphisms between affine
subvarieties. We can do something similar for projective subvarieties.

Theorem 3.13. Let X ⊂ Pn , Y ⊂ Pm be projective subvarieties, and

f : X → Y a morphism. Then there is a N × (m + 1) matrix A = (Fij ) of
homogeneous polynomials Fij ∈ k[X0 , . . . , Xn ] such that deg Fij is constant
in each row,
rk(A(x)) = 1 ∀ [x] ∈ X,
P(Im(A(x)t )) ∈ Y for all [x] ∈ X, and f ([x]) = P(Im(A(x)t )) for all [x] ∈
X. Conversely, every such matrix determines a morphism X → Y in the
preceding manner.
24CHAPTER 3. HILBERT’S NULLSTELLENSATZ, PRIMARY DECOMPOSITION

Proof. Denote homogeneous coordinates in the target Pm by (T0 : · · · : Tm )

and Ui = {Ti 6= 0} ' Am ⊂ Pm . The sets Vi := f −1 (Ui ) are open and cover
X. We can write r
[
Vi = XGij
j=1

Gij ∈ k[X0 , . . . , Xn ] homogeneous, XGij principal open. Then f (XGij ) ⊂

Ui ∩ Y ⊂ Ui ' Am , whence
 
P0 (X0 , . . . , Xn ) Pm (X0 , . . . , Xn )
f |XGij = : ··· : 1 : ··· : (3.3)
Gµij0 Gµijm

with deg Pk = deg(Gij ) · µk and 1 is in the j-th position. This is so because

OX (XGij ) = (k[X0 , . . . , Xn ]/I(X))(Gij ) .

Clearing denominators in (3.3), we get one matrix row for A. Doing this for
all i, j, we get A.
Example 3.14. Let us illustrate the preceding Theorem 3.13 in one geomet-
ric situation. Suppose C ⊂ P2 is given by
X 2 + Y 2 − Z 2 = 0.
Then: the line Y − Z = 0 intersects C in one point, namely P = (0 : 1 : 1),
the line Y + Z = 0 intersects C in one point (0 : −1 : 1), and we want
to consider the stereographic projection of that conic from P onto the line
Y = 0:
f : C − {(0 : 1 : 1)} → P1
(X : Y : Z) 7→ (X : Z − Y ).
Thus, geometrically, we construct the line through P and a point R ∈ C−{P }
and let f (R) be the intersection of that line with the line Y = 0. Now
define f˜ by f on C − {P } and f˜(P ) = (1 : 0) (the point at infinity on
P1 ' {Y = 0}. Then f˜ is a morphism, which can be seen as follows: let
(S : T ) be homogeneous coordinates on P1 , with US = {S 6= 0} ' A1 ,
UT = {T 6= 0} ' A1 . Then
f˜−1 (US ) = C − {(0 : 1 : −1)},
f˜−1 (UT ) = C − {(0 : 1 : 1)}.
 25

Then f˜ is regular on f˜−1 (UT ), and on f˜−1 (US ) f˜ can be given by

(X : Y : Z) 7→ (Y + Z : X).

The matrix  
X Z −Y
A=
Y +Z X

has rank 1 everywhere on C, and is a matrix defining f˜ in the sense of

Theorem 3.13.
Now one can show that in this case

f˜: C → P1

cannot be defined by homogeneous polynomials F0 , F1 ∈ k[X0 , X1 , X2 ] of the

same degree without common zeroes on C. I.e., we cannot find a 1 × 2 matrix
as in Theorem 3.13 defining f˜. To see this, assume by contradiction that
such F0 , F1 would exist. Then note that

τ : A1 ' C − {(0 : −1 : 1)}

via the inverse of the projection onto {Y = 0} from (0 : −1 : 1). We have

τ (t) = (2t : −t2 + 1 : t2 + 1). Now

X F0
and
Z −Y F1

define the same rational function in k(t) on A1 via τ since XF1 −F0 (Z −Y ) =
0 on C. Now F1 vanishes in (0 : 1 : 1), but F0 does not, since (0 : 1 : 1) maps
to (1 : 0). Now  
∗ X 1
τ =
Z −Y t
has a simple pole in 0 ∈ A1 . Hence the same must hold for

F0 (2t, −t2 + 1, t2 + 1)
 
∗ F0
τ = .
F1 F1 (2t, −t2 + 1, t2 + 1)

Consequently, F1 (2t, −t2 + 1, t2 + 1) is a polynomial in k[t] with a simple zero

at 0 and no further zero, since zeroes of F1 (2t, −t2 + 1, t2 + 1) correspond
26CHAPTER 3. HILBERT’S NULLSTELLENSATZ, PRIMARY DECOMPOSITION

bijectively to zeroes of F1 (X, Y, Z) on C − {(0 : −1 : 1)} and (1 : 0) has only

one preimage under f˜ there. Thus

F1 (2t, −t2 + 1, t2 + 1) = ct, c ∈ k∗

and then F1 (X, Y, Z) − (1/2)X is a polynomial vanishing identically on C

since τ ∗ (F1 (X, Y, Z) − (1/2)X) = 0 in k[t]. So F1 (X, Y, Z) has the further
zero (0 : −1 : 1) on C, but this must also a zero of F0 (X, Y, Z) since (0 :
−1 : 1) 7→ (0 : 1) under f˜. Hence F0 , F1 do have a common zero on C,
contradicting our assumption.

Besides morphism and isomorphisms, there is another type of “map” and

equivalence in algebraic geometry that is extremely important:

Definition 3.15. Let X and Y be irreducible varieties. A rational map

f : X 99K Y is given by an equivalence class of pairs (fU , U ) where U ⊂ X
is Zariski open and nonempty (hence dense) and fU : U → Y is a mor-
phism. Two pairs (fU , U ) and (fV , V ) are considered equivalent if (fU ) |U ∩V =
(fV ) |U ∩V . A rational function on X is a rational map to A1 . A rational
map f is called dominant if the image of fU is dense in Y for any represen-
tative fU of f . Varieties X, Y are called birational if there exist dominant
rational maps
f : X 99K Y, g : Y 99K X
such that g ◦ f = idX and f ◦ g = idY as rational maps.

Thus a rational map f as above need not be everywhere defined on X,

which we indicate by the dotted arrow above. Hence it is not a “map”, but
this is consistent with the “red herring principle” in mathematical termi-
nology, which means that a red herring need neither be red nor a herring
in mathematics. Anyway, varieties with dominant rational maps do form a
category, so everything is perfectly fine.
For example, A2 and P2 are birational, but not isomorphic (for example
because P2 has no nonconstant global regular functions whereas A2 has lots
of them).
Chapter 4

Segre embeddings, Veronese

maps and products

If X ⊂ An and Y ⊂ Am are affine subvarieties, clearly X × Y ⊂ An × Am '

An+m is an affine subvariety. (But note that the product topology on X × Y
is in general strictly weaker than the Zariski topology!) What about
the same question for projective subvarieties X ⊂ Pn , Y ⊂ Pm ? The first
question is how to turn the set Pn × Pm into a projective variety.
Definition 4.1. We define a topology on the set Pn × Pm , which we call
the Zariski topology, in the following way: its closed sets are zero sets
V (f1 , . . . , fN ) in Pn ×Pm of polynomials fi ∈ k[X0 , . . . , Xn , Y0 , . . . , Ym ] which
are homogeneous in the X’s and Y ’s separately, i.e.
X
fi = ca0 ,...,an ;b0 ,...,bm X0a0 . . . Xnan Y0b0 . . . Ynbn .
a0 +···+an =d,b0 +···+bm =e

Thus fi is “bi-homogeneous” of bi-degree (d, e).

Theorem 4.2. Let (Zij )0≤i≤n,0≤j≤m be homogeneous coordinates in Pnm+n+m .
Let a be the homogeneous ideal in k[Zij ] generated by

Zij Zkl − Zil Zkj ∀ i, j, kl

(so all 2 × 2-minors of the matrix (Zij )). Then the map

sn,m : Pn × Pm → V (a) ⊂ Pnm+n+m

((X0 : · · · : Xn ), (Y0 : · · · : Ym )) 7→ (Zij ) := (Xi Yj )

27
28CHAPTER 4. SEGRE EMBEDDINGS, VERONESE MAPS AND PRODUCTS

is a homeomorphism and V (a) is an irreducible projective subvariety. One

calls sn,m the Segre embedding and V (a) =: Σn,m the Segre variety.
Proof. Step 1. We show that sn,m is injective. Thus suppose that sn,m (X, Y ) =
sn,m (X 0 , Y 0 ). It means that there is a nonzero constant λ such that for all i, j
we have Xi Yj = λXi0 Yj0 . Moreover, X, Y representing points in projective
spaces, there is a tuple of indices i0 , j0 with Xi0 6= 0, Yj0 6= 0. Thus both Xi00
and Yi00 have to be nonzero as well. Putting
Xi 0 Yj0
µ := , ν :=
Xi00 Yj00

(whence λ = µν), we calculate

Xi Yj0 = λXi0 Yj00 = µνXi0 Yj00 = µXi0 Yj0

and thus Xi = µXi0 for all i. Thus X = X 0 in projective space, and analo-
gously we get Y = Y 0 .
Step 2. The map sn,m is surjective. Suppose Zij satisfy Zij Zkl = Zil Zkj
and not all Z’s are zero, Zi0 j0 6= 0, say. Put
Zij0 Zi0 j
Xi := , Yj = .
Zi0 j0 Zi0 j0
Then
(Zi0 j0 )2 Xi Yj = Zij0 Zi0 j = Zij Zi0 j0
which means s(X, Y ) = [(Zij )].
Step 3. The map sn,m is a homeomorphism. Indeed, the topology we
put on Pn × Pm has a basis consisting of the sets

{(X, Y ) | f (X, Y ) 6= 0, f bihomogeneous of bidegree (d, d)} .

For, by definition, the topology has a basis consisting of

{(X, Y ) | f (X, Y ) 6= 0, f bihomogeneous of bidegree (d, e)} .

Suppose d ≥ e (the opposite case being similar). Then

m
[
(X, Y ) | (Yjd−e f )(X, Y ) 6= 0

{(X, Y ) | f (X, Y ) 6= 0} =
j=0
 29

and Yjd−e f has bidegree (d, d).

Now every polynomial f (X, Y ) which is bihomogeneous of bidegree (d, d)
can be written as F (. . . , Xi Yj , . . . ) where F is a homogeneous polynomial in
the Zij of degree d. That means that the open set f 6= 0 in Pn × Pm gets
mapped to the open set V (a) ∩ (F 6= 0). Thus sn,m is a homeomorphism.
Step 4. The image V (a) is irreducible. This follows from a general
topological Lemma.

Lemma 4.3. Let X, Y be irreducible topological spaces. If on X × Y a

topology is given which induces the given topology on {x} × Y ' Y and
X × {y} ' X for all x ∈ X and all y ∈ Y , then X × Y is irreducible.

Proof. Let X × Y = S ∪ T be a decomposition into closed subsets. For all

x ∈ X we have

{x} × Y = [S ∩ ({x} × Y )] ∪ [T ∩ ({x} × Y )] .

Since Y is irreducible, it follows that for all x we have {x} × Y ⊂ S or

{x} × Y ⊂ T . Let sy be the map sy : X → X × Y given by sy (x) = (x, y).
Then
\
S 0 := s−1
y (S) = {x | (x, y) ∈ S ∀ y} = {x | {x} × Y ⊂ S}.
y∈Y

Clearly, S 0 is also closed. Similarly,

\
T 0 := s−1
y (T ) = {x | {x} × Y ⊂ T }
y∈Y

is closed. Since we saw that X = S 0 ∪ T 0 , the irreducibility of X implies

X = S 0 or X = T 0 . Thus X × Y = S or X × Y = T .

Thus Σn,m = V (a) is irreducible and this concludes the proof of Theorem
4.2.

We now give Pn ×Pm the structure of a projective variety, and every prod-
uct X × Y of (quasi-)projective varieties the structure of (quasi-)projective
variety that comes from identifying it with the image under sn,m .
30CHAPTER 4. SEGRE EMBEDDINGS, VERONESE MAPS AND PRODUCTS

Example 4.4. The variety Σ1,1 = s1,1 (P1 × P1 ) ⊂ P3 is the image of the map

((X0 : X1 ), (Y0 : Y1 )) 7→ (X0 Y0 : X0 Y1 : X1 Y0 : X1 Y1 ) =: (Z0 : Z1 : Z2 : Z3 )

i.e., the quadric  

Z0 Z1
det =0
Z2 Z3
in P3 . In general, sn,m maps each Pn × {y} and {x} × Pm onto projective
linear subspaces of P(n+1)(m+1)−1 ; in this case this gives two families of lines
on Σ1,1 , namely

V (Z1 = λZ0 , Z3 = λZ2 )λ ∈ P1 ,

V (Z2 = λ0 Z0 , Z3 = λ0 Z1 )λ0 ∈P1 .

These are the loci where either the rows or columns of the above matrix
satisfy a linear dependency relation.
In fact, any line L ⊂ P3 on Σ1,1 belongs to one of these two families:
suppose we write L as the image of a map P1 → P3

(λ : µ) 7→ (l1 (λ, µ) : · · · : l4 (λ, µ))

where the li (λ, µ) are linear forms. The condition that L lie on Σ1,1 means
that  
l1 (λ, µ) l2 (λ, µ)
det ≡ 0.
l3 (λ, µ) l4 (λ, µ)
Without loss of generality, after possibly interchanging rows or columns and
transposing, we may assume l1 6= 0 and l1 , l2 linearly independent. Then
l1 l4 = l2 l3 implies that l1 | l3 and l2 | l4 , hence l3 = cl1 and l4 = cl2 , some
c ∈ k.

Remark 4.5. In coordinate free form, the Segre map can be given as

s : P(V ) × P(W ) → P(V ⊗ W )

[v] × [w] 7→ [v ⊗ w].

Remark 4.6. The product X × Y of varieties X, Y as we defined it is the

categorical product of varieties. This means that the projections prX : X ×
Y → X and prY : X × Y → Y are morphisms, and for all varieties Z and
 31

morphisms α : Z → X, β : Z → Y there is a unique morphism α × β : Z →

X × Y such that

Z
α×β
 β
α
X ×Y
prY
prX
{ # 
X Y
commutes. To see this note that purely set-theoretically it is clear how we
have to define α × β if it is to exist at all. Then we just have to check that
α × β is a morphism. To see this suppose r0 ∈ Z is a point with α(r0 ) = p,
β(r0 ) = q. Suppose p is in X0 6= 0 in Pn . Then, in a neighborhood of r0 the
map α can be given by

r 7→ (1 : f1 (r) : · · · : fn (r))

with the fi regular in a neighborhood of r0 . After possibly renumbering

coordinates Yi in the target Pm we can also assume that β is locally around
r0 given by
β(r) = (1 : g1 (r) : · · · : gm (r))
with the g’s regular in a neighborhood of r0 . Then α × β is, locally around
r0 , given by

r 7→ (1 : · · · : fi (r) : · · · : gj (r) : · · · : fi (r)gj (r) : . . . ) ∈ Pnm+n+m ,

which is regular.
Remark 4.7. If f : X → Y is a morphism between projective varieties X ⊂
Pn , Y ⊂ Pm , then

Γf := {(x, f (x)) | x ∈ X} ⊂ X × Y ⊂ Pn × Pm

is a (closed) subvariety. Indeed, by the preceding Remark 4.6, the map

f × idY : X × Y → Y × Y is a morphism, and

Γf = (f × idY )−1 (∆Y )

where ∆Y ⊂ Y × Y is the diagonal. But ∆Y = ∆Pm ∩ (Y × Y ) is closed

because ∆Pm can be defined by Xi Yj − Xj Yi = 0.
32CHAPTER 4. SEGRE EMBEDDINGS, VERONESE MAPS AND PRODUCTS

We turn to the subject of Veronese embeddings. Note that the homoge-

neous polynomials F of degree m in the variables X0 , . . . , Xn form a vector
space of dimension  
m+n
.
m
This can be seen by the following count: decompositions

α0 + · · · + αn = m

of m into n + 1 nonnegative integers αi correspond bijectively to subsets of

n elements in {1, . . . , m + n} via

1 ≤ α0 + 1 < α0 + α1 + 2 < · · · < α0 + · · · + αn−1 + n ≤ m + n.

Now put  
m+n
N := −1
m
and for each decomposition into nonnegative integers i0 + · · · + in = m,
introduce a symbol vi0 ...in , and consider these as homogeneous coordinates in
PN . Then the m-th Veronese map is the map

v m : Pn → PN
vi0 ...in := ui00 . . . uinn

where u0 , . . . , un are homogeneous coordinates in Pn . This is visibly a mor-

phism. The image vm (Pn ) ⊂ PN is called Veronese variety. Indeed, this
image is a projective subvariety of PN , which we check as follows:
For all points in vm (Pn ) we have the relations

vi0 ...in vj0 ...jn = vk0 ...kn vl0 ...ln ∀ i0 + j0 = k0 + l0 , . . . , in + jn = kn + ln . (4.1)

If (4.1) holds, then we have v0,...,0,m,0...,0 6= 0 for m sitting in some position

in the multi-index. If this were not so, we could pick a vl0 ...ln such that the
maximum of the indices l0 , . . . , ln , call it lmax , is maximal among all indices
that occur in the coordinates vj0 ...jn with vj0 ...jn 6= 0, and such that 1 ≤ lmax ≤
m − 1. But this would lead to a contradiction using the equations (4.1), for
 33

then there will be another index ˜l among the l0 , . . . , ln with 1 ≤ ˜l ≤ m − 1

and we can write
(vl0 ...ln )2 = v...,lmax +1,...,l̃−1,... v...,lmax −1,...,l̃+1,... .
But then v...,lmax +1,...,l̃−1,... is nonzero and has greater maximum index!
So we have get a v0,...,0,m,0...,0 6= 0. Suppose for simplicity of notation that
the m is in the first place: vm,0,0,... 6= 0. Then we can set
u0 := vm,0,0,... , ui := vm−1,0,...,0,1,0,..., , i ≥ 2
where the 1 in the formula for ui is in the i-th position. This defines a regular
inverse map to vm on the open set vm,0,... 6= 0. Thus we conclude that (1)
vm (Pn ) is defined by the equations (4.1) and the map vm : Pn → vm (Pn ) ⊂ PN
is an isomorphism onto the image.
For n = 1, we get rational normal curves vm (P1 ) ⊂ PN . The image of
v2 : P2 → P5 is classically called the Veronese surface.
Of course we can also aplly vm to any subvariety X ⊂ Pn and get a
subvariety vm (X) ⊂ PN . X and vm (X) are then isomorphic as varieties.
This leads to the following somewhat surprising result.
Theorem 4.8. Every projective variety X is isomorphic to an intersection of
a Veronese variety with a linear space. In particular, every projective variety
is isomorphic to an intersection of quadrics.
Proof. The zero set of a homogeneous polynomial F (u0 , . . . , un ) of degree d is
the same as the zero set of all polynomials {ui F }i=0,...,n . Hence we can assume
that X ⊂ Pn is defined by homogeneous polynomials Fj (u0 , . . . , un ) = 0, all
of which have the same degree D. This means that Fj is a linear polynomial
in the monomials ui00 . . . uinn , i0 + · · · + in = D, hence vD (X) ⊂ PN is an
intersection of vD (Pn ) with a linear subspace.
Of course not every X is isomorphic to a complete intersection of quadrics,
the homogeneous ideal of X in any embedding need not be generated by
quadrics. So the algebraic consequences of Theorem 4.8 are not so strong.
Remark 4.9. We can also describe the Veronese map in a coordinate free
manner (for char(k) = 0) as
vd : P(V ) → P(Symd V )
[v] 7→ [v d ].
If char(k) = p, the p-powers of linear forms in P(Symp V ) are not a rational
normal curve, but lie on a line.
34CHAPTER 4. SEGRE EMBEDDINGS, VERONESE MAPS AND PRODUCTS
Chapter 5

Grassmannians, flag manifolds,

Schubert varieties

We want to make the set Grass(p, V ) of all p-dimensional sub-vector spaces

of an n-dimensional k-vector space V into a projective variety (equivalently,
the set of all (p − 1)-dimensional projective linear subspaces of P(V )).
Definition 5.1. We call elements ω ∈ Λp V (the p-th graded piece of the exte-
rior algebra) p-forms. We call ω completely reducible if there are v1 , . . . , vp ∈
V such that ω = v1 ∧ · · · ∧ vp .
The k-sub-vector space of V
Ann(ω) = {v ∈ V | v ∧ ω = 0}
is called the annihilator of ω.
Theorem 5.2. Let ω1 resp. ω2 be completely reducible p− resp. q−forms.
Then:
1.
Ann(ω1 ) ⊃ Ann(ω2 ) ⇐⇒ ∃ ω ∈ Λp−q V : ω1 = ω ∧ ω2
unless ω2 = 0, ω1 = e1 ∧ · · · ∧ en for some basis e1 , . . . , en of V .
2.
Ann(ω1 ) ∩ Ann(ω2 ) = {0} ⇐⇒ ω1 ∧ ω2 6= 0.

3.
Ann(ω1 ) ∩ Ann(ω2 ) = {0} =⇒ Ann(ω1 ) + Ann(ω2 ) = Ann(ω1 ∧ ω2 ).

35
36CHAPTER 5. GRASSMANNIANS, FLAG MANIFOLDS, SCHUBERT VARIETIES

Proof. For (1) we note that if v ∧ ω2 = 0, then v ∧ ω ∧ ω2 = ±ω ∧ v ∧ ω2 = 0,

thus ⇐= holds. To prove the converse, consider some completely reducible

ω̃ = v1 ∧ · · · ∧ vp .

If v1 , . . . , vp are linearly dependent then ω̃ = 0. Suppose that they are linearly

independent, v1 ∧ · · · ∧ vp 6= 0. Then we will show that

Ann(v1 ∧ · · · ∧ vp ) = hv1 , . . . , vp i. (5.1)

What is clear is that ⊃ holds in the preceding formula. Let v1 , . . . , vp , vp+1 , . . . , vn

be a basis of V . Then to show (5.1), it suffices to show
n
X
ai vi ∈ Ann(v1 ∧ · · · ∧ vp ) =⇒ ai = 0 ∀ i > p.
i=1

This is clear since

n
! n
X X
0= ai vi ∧ (v1 ∧ · · · ∧ vp ) = ai v i ∧ v 2 ∧ · · · ∧ v p
i=1 i=p+1

and vi ∧ v1 ∧ · · · ∧ vp , p + 1 ≤ i ≤ n, are linearly independent. Suppose now

Ann(ω1 ) ⊃ Ann(ω2 ), ω1 = v1 ∧ · · · ∧ vp , ω2 = v10 ∧ · · · ∧ vq0 . If ω1 = 0, the
conclusion of (1) is trivial, and if ω2 is zero, the annihilator of ω1 is the entire
space, hence ω1 is zero or ω1 = e1 ∧ · · · ∧ en for a basis e1 , . . . , en of V ; in the
first case, the conclusion of (1) holds trivially, in the second case it holds,
too. If ω1 6= 0, ω2 6= 0, we get by the above considerations

hv1 , . . . , vp i ⊃ hv10 , . . . , vq0 i.

Thus we can choose a basis of Ann(ω1 ) of the form (v10 , . . . , vq0 , vq+1
0
, . . . , vp0 )
whence
ω1 = det(A)v10 ∧ · · · ∧ vq0 ∧ vq+1
0
∧ . . . vp0
where A is the base change matrix from (v10 , . . . , vp0 ) to (v1 , . . . vp ). Thus there
0
exists ω = ± det(A)vq+1 ∧ . . . vp0 with ω1 = ω ∧ ω2 .
For (2) and (3), first note that

(v1 ∧ · · · ∧ vp ) ∧ (v10 ∧ · · · ∧ vq0 ) 6= 0

 37

if and only if v1 , . . . , vp , v10 , . . . , vq0 are linearly independent if and only if

hv1 , . . . , vp i ∩ hv10 , . . . , vq0 i = {0}.

This shows (2), and (3) follows from what has been proven so far (in partic-
ular (5.1)).

Corollary 5.3. The map

Ann : {[ω] ∈ P(Λp V ) | ω completely reducible} → Grass(p, V )

is bijective.

Proof. Every p-dimensional sub-vector space W ⊂ V is in the image since

if W = hv1 , . . . , vp i, then W = Ann(v1 ∧ · · · ∧ vp ). Theorem 5.2, (1) gives:
Ann(ω1 ) = Ann(ω1 ) =⇒ ω1 = ω ∧ ω2 with ω ∈ Λ0 V = k.

Hence the map

iPl := Ann−1 : Grass(p, V ) → P(Λp V )

is a bijection onto its image. This map iPl is called the Plücker embedding.
If W = hv1 , . . . , vp i and A the p × n-matrix with rows the coordinates of
v1 , . . . , vp (with respect to some basis e1 , . . . , en of V ), then W gets mapped
under iPl to X
M i1 ,...,ip ei1 ∧ · · · ∧ eip
1≤i1 <···<ip ≤n

whereM i1 ,...,ip is the minor of A belonging to the columns i1 , . . . , ip . One calls

the (M i0 ,...,ip ) the Plücker coordinates of W .
To check that Grass(p, V ) ⊂ P(Λp V ) is a projective subvariety, we have
to exhibit polynomial equations which characterize the completely reducible
p-forms.

Theorem 5.4. A p-form ω ∈ Λp V , ω 6= 0, is completely reducible if and only

if dim Ann(ω) = p. Otherwise we always have dim Ann(ω) < p. Moreover,
Grass(p, V ) ⊂ P(Λp V ) is a projective variety.

Proof. From the proof of Theorem 5.2 we know that if ω 6= 0 and ω is com-
pletely reducible, then dim Ann(ω) = p. Suppose conversely that dim Ann(ω) =
38CHAPTER 5. GRASSMANNIANS, FLAG MANIFOLDS, SCHUBERT VARIETIES

r (we will need the case r = p). Suppose that Ann(ω) = hv1 , . . . , vr i and
that (v1 , . . . , vr , vr+1 , . . . , vn ) is a basis of V . Write
X
ω= ωi1 ...ip vi1 ∧ · · · ∧ vip .
1≤i1 <···<ip ≤n

The condition vi ∧ ω = 0 for all i = 1, . . . , r translates into ωi1 ...ip = 0 for

{1, . . . , r} 6⊂ {i1 , . . . , ip }. This means that for ω 6= 0, we get r ≤ p and
ω = (v1 ∧ · · · ∧ vr )∧(something). For r = p, we obtain the claim.
Now this implies that [ω] ∈ Grass(p, V ) is equivalent to the rank of the
map
ϕ(ω) : V → Λp+1 V, v 7→ ω ∧ v
being n − p (namely, ker(ϕ(ω)) = Ann(ω)). Now rk(ϕ(ω)) < n − p cannot
happen since always dim Ann(ω) ≤ p. Hence
[ω] ∈ Grass(p, V ) ⇐⇒ rk(ϕ(ω)) ≤ n − k. (5.2)
The map
Φ : Λp V → Hom(V, Λp+1 V ), ω → ϕ(ω)
is k-linear, i.e. the entries of a matrix for ϕ(ω) are linear in the homogeneous
coordinates of P(Λp V ). The (n − p + 1) × (n − p + 1)-minors of this matrix
define Grass(p, V ).
Remark 5.5. Actually the above equations are easy to produce but they do
not generate the homogeneous ideal of Grass(p, V ).
Now consider the special case p = 2. We will study this in a little more
detail.
Theorem 5.6. A vector 0 6= ω ∈ Λ2 V is completely reducible if and only if
ω ∧ ω = 0.
Proof. The direction =⇒ is clear, and for ⇐= we do induction over
n = dim V , n = 2 being trivial. For the induction step, let (v1 , . . . , vn+1 ) be
a basis of V . Then we can write
ω = vn+1 ∧ ω1 + ω2
where ω1 is a linear combination of vi , 1 ≤ i ≤ n, and ω2 is a linear combi-
nation of vi ∧ vj , 1 ≤ i < j ≤ n. Now ω ∧ ω = 0 translates into
ω2 ∧ ω2 + 2vn+1 ∧ ω1 ∧ ω2 = 0
 39

because (vn+1 ∧ ω1 ) ∧ (vn+1 ∧ ω1 ) = 0 and ω2 is in the center of Λ• V . Since

ω2 ∧ ω2 does not contain vn+1 in its expansion with respect to basis elements
in V , we have separately
ω2 ∧ ω2 = 0 vn+1 ∧ ω1 ∧ ω2 = 0.
By the induction hypothesis, this implies that ω2 is completely reducible.
Since ω1 ∧ ω2 does not contain vn+1 in its basis expansion, we conclude that
even ω1 ∧ ω2 = 0, i.e., ω1 ∈ Ann(ω2 ). Thus dim Ann(ω2 ) = 2 (by Theorem
5.4 and because we just proved that it is completely reducible), hence
ω2 = ω10 ∧ ω1
for some ω10 , unless ω1 = 0 in which the case the proof is already complete
anyway. Then
ω = vn+1 ∧ ω1 + ω10 ∧ ω1 = (vn+1 + ω10 ) ∧ ω1
is completely reducible.
Corollary 5.7. The variety Grass(2, V ) ⊂ P(Λ2 V ), n = dim V ≥ 3, is an
intersection of quadrics.
Proof. Indeed, in a basis e1 , . . . , en of V , the condition that
X
ω= ωi1 i2 ei1 ∧ ei2
1≤i1 <i2 ≤n

be completely reducible is, by the Theorem 5.6,

! !
X X
ωi1 i2 ei1 ∧ ei2 ∧ ωj1 j2 ej1 ∧ ej2 = 0.
1≤i1 <i2 ≤n 1≤j1 <j2 ≤n

This is equivalent to
 
X i i i i
ωi1 i2 ωj1 j2 sgn 1 2 3 4 = 0.
k1 k2 k3 k4
Thus we get one equation for each quadruple of indices 1 ≤ k1 < k2 < k3 <
k4 ≤ n and the sum above then runs over those indices 1 ≤ i1 < i2 ≤ n and
1 ≤ j1 < j2 ≤ n with {i1 , i2 , i3 , i4 } = {k1 , k2 , k3 , k4 }. With sgn we mean the
sign of the permutation  
i1 i2 i3 i4
.
k1 k2 k3 k4
40CHAPTER 5. GRASSMANNIANS, FLAG MANIFOLDS, SCHUBERT VARIETIES

In particular, if n = 4, we can consider Grass(2, V ) ⊂ P(Λ2 V ) ' P5 given

by
ω12 ω34 − ω13 ω24 + ω14 ω23 = 0.
This quadric is called the Plücker quadric. It parametrizes projective lines
in P3 .
We can in general consider the subset

Σl (M ) ⊂ Grass(p, V )
where M ⊂ V is an m-dimensional subspace, and Σl (M ) consists of those
p-dimensional W ⊂ V with dim(W ∩ M ) ≥ l.

Proposition 5.8. The subset Σl (M ) is a subvariety of Grass(p, V ) ⊂ P(Λp V )

which is a section of Grass(p, V ) by a linear subspace of P(Λp V ).

Proof. Indeed,

Σl (M ) = {[ω] ∈ Grass(p, V ) | ω ∧ v1 ∧ · · · ∧ vm−l+1 = 0 ∀ v1 , . . . , vm−l+1 ∈ M } .

This means that the span of W and M is at most p + m − l-dimensional.

Customarily, the Σl (M ) are called (special) Schubert varieties in Grass(p, V ).

For 1 ≤ a1 < a2 < · · · < at ≤ n consider the subset

Flag(a1 , . . . , at ; V ) = {(W1 , . . . , Wt ) | W1 ⊂ · · · ⊂ Wt }
⊂ Grass(a1 , V ) × · · · × Grass(at , V )

where the latter product is a projective variety via the Segre embedding.
Elements in Flag(a1 , . . . , at ; V ) are called flags of type (a1 , . . . , at ) in V .

Proposition 5.9. The subset Flag(a1 , . . . , at ; V ) is a subvariety of Grass(a1 , V )×

· · · × Grass(at , V ).

We call it the flag variety of flags of type (a1 , . . . , at ) in V .

Proof. If

prij : Grass(a1 , V ) × · · · × Grass(at , V ) → Grass(ai , V ) × Grass(aj , V )

 41

is the projection, then

\
Flag(a1 , . . . , at ; V ) = pr−1
ij (Flag(ai , aj , V )) ,
1≤i<j≤t

so it suffices to show that some Flag(a1 , a2 , V ) ⊂ Grass(a1 , V ) × Grass(a2 , V )

is closed.
Step 1. We first describe local charts on Grassmannians. For Grass(p, V ),
V ' k n , consider the sub-vector space Γ = hep+1 , . . . , en i where e1 , . . . , en is
the standard basis of V = k n . Consider
U := {W ∈ Grass(p, V ) | W ∩ Γ = {0}}.
If W is spanned by the rows of
 
w11 . . . w1n
 .. ... ..  ,
 . . 
wp,1 . . . wpn
then W ∈ U if the minor (=Plücker coordinate of W )
 
w11 . . . w1p
det  ... ... .. 

. 
wp,1 . . . wpp
is nonzero. Thus U ⊂ Grass(p, V ) is open in Grass(p, V ) and isomorphic to
Ap(n−p) ; to see the last statement it suffices to consider the map
Matk (p × (n − p)) ' Ap(n−p) → U
which associates to a matrix
 
a1,1 . . . a1,n−p
A =  ... ... .. 

. 
ap,1 . . . ap,n−p
the Plücker coordinates of the subspace W of k n which is spanned by the
rows of the p × n matrix
 
1 ... 0 a1,1 . . . a1,n−p
 .. . . .. .. .. ..  .
. . . . . . 
0 ... 1 ap,1 . . . ap,n−p
42CHAPTER 5. GRASSMANNIANS, FLAG MANIFOLDS, SCHUBERT VARIETIES

If instead of Γ above we take

Γi1 ...in−p = hei1 , . . . , ein−p i, 1 ≤ i1 < · · · < in−p ≤ n,

then the construction runs analogously: the subspaces W with W ∩Γi1 ...in−p =
{0} form an open subset Uj1 ,...,jp ' Ap(n−p) of Grass(p, V ), where (j1 , . . . , jp )
are the indices complementary to (i1 , . . . , in−p ) in {1, . . . , n}, and Uj1 ,...,jp is
then given by  
w1j1 . . . w1jp
det  ... .. ..  6= 0.

. . 
wp,j1 . . . wpjp
The Uj1 ,...,jp form a cover of Grass(p, V ) by affine spaces.
Step 2. Consider now

Flag(a1 , a2 , V ) ⊂ Grass(a1 , V ) × Grass(a2 , V ) ⊂ P(Λa1 V ) × P(Λa2 V ).

The open sets

Uj1 ...ja1 × Ul1 ...la2 ' Aa1 (n−a1 ) × Aa2 (n−a2 )

1 ≤ j1 < · · · < ja1 ≤ n, 1 ≤ l1 < · · · < la2 ≤ n, cover Grass(a1 , V ) ×

Grass(a2 , V ), and it suffices to show that

Flag(a1 , a2 , V ) ∩ (Uj1 ...ja1 × Ul1 ...la2 )

is closed in Aa1 (n−a1 ) × Aa2 (n−a2 ) . Without loss of generality we can assume
{l1 , . . . , la2 } = {1, . . . , a2 }. An M ∈ Ul1 ...la2 is given by a matrix
 
1 . . . 0 m1,1 . . . m1,n−a2
 .. . . .. .. .. ..
.

. . . . . .
0 . . . 1 ma2 ,1 . . . ma2 ,n−a2

Moreover, S ∈ Uj1 ,...,ja1 is given by a matrix

 
s1,1 . . . s1,n
 .. .. .. 
 . . . 
sa1 ,1 . . . sa1 ,n
 43

and the columns corresponding to j1 , . . . , ja1 form the unit matrix. Then
S ⊂ M means that the matrix
 
s1,1 . . . . . . ... ... s1,n
 .. ... ... ... ..
 . ... .


sa1 ,1 . . . . . . ... ... sa1 ,n 
 
 1 ... 0 m1,1 ... m1,n−a2 
 
 . .. . .. .. ..
 .. . ..

. . . 
0 ... 1 ma2 ,1 . . . ma2 ,n−a2

has rank ≤ a2 , i.e., all (a2 + 1) × (a2 + 1)-minors vanish.This finishes the
proof.
2
Remark 5.10. The subset G = GLn (k) ⊂ An is open, hence a variety. What
is more, the group composition and inverse map are morphisms. The varieties

Grass(p, V ) and Flag(a1 , . . . , at , V )

are homogeneous spaces for G, which means that G acts transitively on them
and the obvious group actions

G × Grass(p, V ) → Grass(p, V ),
G × Flag(a1 , . . . , at , V ) → Flag(a1 , . . . , at , V )

are morphisms.
44CHAPTER 5. GRASSMANNIANS, FLAG MANIFOLDS, SCHUBERT VARIETIES
Chapter 6

Images of projective varieties

under morphisms

If f : X → Y is a morphism of varieties, is it true that the image f (X) ⊂ Y

is a quasi-projective subvariety? I.e., is it locally closed? Unfortunately (or
fortunately), things aren’t quite as simple:

Example 6.1. Look at the morphism

f : A2 → A2
(x, y) 7→ (x, xy).

Then Im(f ) = {(a, b) ∈ A2 | a 6= 0} ∪ {(0, 0}. This is not open in its closure
A2 .

However, everything is nicer when the source is projective.

Theorem 6.2. Let X be a projective variety and Y any variety, f : X → Y

a morphism. The f (X) ⊂ Y is closed.

Proof. The graph Γf ⊂ X×Y is closed by by Remark 4.7 and f (X) = pr2 (Γf )
where pr2 : X × Y → Y is the projection onto the second factor. So it suffices
to show:
If Z ⊂ X × Y is closed and X is projective, then pr2 (Z) ⊂ Y is closed.
Now if X ⊂ Pn then Z is also closed in Pn × Y and pr2 (Z) is equal to the
image of the projection Pn × Y → Y . Thus we can also assume X = Pn .

45
46CHAPTER 6. IMAGES OF PROJECTIVE VARIETIES UNDER MORPHISMS
S
Now let Y = Ui be a finite cover of Y by affine open subsets. Then
i∈I
[ [
X ×Y = X × Ui , Z = Z ∩ (X × Ui ),
i∈I
[
pr2 (Z) = pr2 (Z ∩ (X × Ui )) ,
i∈I

so it suffices to show that pr2 (Z ∩ (X × Ui )) is closed in Ui . Hence we can

furthermore assume that Y is affine, Y ⊂ Am . Moreover, it is then X × Y
closed in X × Am and pr2 (Z) equal to the image of Z under the projection
X × Am → Am , so we can even assume Y = Am .
Thus let Z ⊂ Pn × Am be closed. Because of Theorem 4.2, Z can be
defined by equations

Fi (T0 , . . . , Tn , t1 , . . . , tm ) = 0, i = 1, . . . , N,

where Fi ∈ k[T0 , . . . , Tn , t1 , . . . , tm ] is a polynomial which is homogeneous,

of degree di , say, in the coordinates T0 , . . . , Tn on Pn (but of course not
necessarily in the coordinates t1 , . . . , tm on Am ). For all a = (a1 , . . . , am ) ∈
k m ' Am put

Za := {(T0 : · · · : Tn ) ∈ Pn | Fi (T0 , . . . , Tn , a1 , . . . , am ) = 0 ∀ i = . . . , N }

(which can be thought of as the fiber of Z over a point a ∈ Am ). Then Za is

empty if and only if (0, 0, . . . , 0) is the only solution (in An+1 ) of the equations
Fi (T0 , . . . , Tn , a1 , . . . , am ) = 0 ∀ i = . . . , N . This√in turn is equivalent, by the
Nullstellensatz 3.2, to the fact that the radical Ia of the ideal Ia generated
by the Fi (T0 , . . . , Tn , a1 , . . . , am ) in k[T0 , . . . , Tn ] is equal to (T0 , . . . , Tn ). We
can also phrase this as

Za = ∅ ⇐⇒ (T0 , . . . , Tn )s ⊂ Ia for some s ≥ 0.

Thus we get the characterization

pr2 (Z) = {a ∈ Am | Za 6= ∅}
\
= {a ∈ Am | (T0 , . . . , Tn )s 6⊂ Ia }.
s≥0

Thus it is sufficient to show that each set

Ys := {a ∈ Am | (T0 , . . . , Tn )s 6⊂ Ia }
 47

is closed in Am . Now remark that (T0 , . . . , Tn )s ⊂ Ia if and only if every

homogeneous polynomial of degree s in k[T0 , . . . , Tn ] can be written as
N
X
Fi (T0 , . . . , Tn , a1 , . . . , am )Qi (T0 , . . . , Tn )
i=1

for some Qi (T0 , . . . , Tn ) ∈ k[T0 , . . . , Tn ]s−di . This leads us to consider the

k-linear map
N
M
ϕa : k[T0 , . . . , Tn ]s−di → k[T0 , . . . , Tn ]s
i=1
N
X
(Q1 , . . . , QN ) 7→ Fi (T0 , . . . , Tn , a1 , . . . , am )Qi (T0 , . . . , Tn ),
i=1

and to reformulate once more: a ∈ Am \Ys if and only if ϕa is surjective,

thus a ∈ Ys if and only if rk(ϕa ) < dim k[T0 , . . . , Tn ]s =: d. But then the
d × d-minors of any matrix representing ϕa are polynomials, with coefficients
in k, in the a1 , . . . , am , which then define Ys .
Corollary 6.3. If f : X → Y is a morphism, and X is projective and irre-
ducible, and Y is affine, then f is constant.
Proof. Suppose Y ⊂ Am and let xi : Am → A1 be the i-th coordinate func-
tion. Then the composition
 
Y A1 
f xi
X / / Am / / P1
is a morphism from X to P1 whose image is closed in P1 by Theorem 6.2.
A closed subset of P1 is either the whole of P1 , which is not the case in the
present situation since f maps into Am , or a finite point set. Since X is
irreducible, this has to consist of exactly one point. Since this holds for all
i = 1, . . . , m, the assertion follows.
Let f : X → Z, g : Y → Z morphisms of (quasi-)projective varieties. Let
X ×Z Y := {(x, y) ∈ X × Y | f (x) = g(y)}.
Since this is the preimage of the diagonal ∆Z ⊂ Z × Z under the morphism
X × Y → Z × Z, (x, y) 7→ (f (x), g(y)), this is a quasi-projective variety. We
call it the fiber product of f : X → Z and g : Y → Z.
48CHAPTER 6. IMAGES OF PROJECTIVE VARIETIES UNDER MORPHISMS

Definition 6.4. A morphism f : X → Z of varieties is called proper if for

all varieties Y and for all morphisms g : Y → Z and for all closed subsets
W ⊂ X ×Z Y , the image of W under the projection X ×Z Y → Y is closed
in Y .

Theorem 6.5. If X is projective, then every morphism f : X → Z to another

variety Z is proper.

Proof. In the proof of Theorem 6.2 we saw that, in the situation and notation
of Definition 6.4, the projection X × Y → Y maps closed sets to closed sets.
But X ×Z Y is closed in X × Y , and W is closed in X ×Z Y , hence also in
X ×Y.

How does the image of a morphism of general varieties f : X → Y look

like?
Remark 6.6. If X is not necessarily projective, one can still show that the
image of f : X → Y in Y is always a constructible set (this is a theorem
due to Chevalley). Here the constructible sets in Y are the smallest family
of subsets in Y which (1) contains all the open subsets, (2) is stable under
finite intersections, and (3) stable under taking complements.
As applications of the fore-going theory, the reader may try to prove the
following assertions for herself or himself:

1. Consider k[X0 , . . . , Xn ]d as an affine space

 
N n+d
A , N= .
d

Prove that the subset of reducible polynomials is closed.

2. Prove that for n > 1, the varieties

An − {point}, Pn − {point}

are isomorphic neither to affine nor to projective varieties. (To show

that An − {(0, . . . , 0)} is not affine, compute the global regular func-
tions OAn −{(0,...,0)} (An − {(0, . . . , 0)}) and remark that for any affine
variety Z, a proper ideal I ( OZ (Z) defines a non-empty subset by the
Nullstellensatz).
 49

3. If X ⊂ Pn is a projective subvariety which is not a finite point set, and

Y ⊂ Pn is a projective subvariety defined by a single (non-constant)
homogeneous polynomial F ⊂ k[X0 , . . . , Xn ], then X ∩ Y 6= ∅.

4. Let n ≥ 2 and f : Pn → P1 a morphism. Show that f is constant.

(Note that the projective closure of the zero set of a polynomial p in
An is defined by the homogenization of the polynomial).
50CHAPTER 6. IMAGES OF PROJECTIVE VARIETIES UNDER MORPHISMS
Chapter 7

Finite morphisms, Noether

normalization

In this chapter, all varieties are assumed to be irreducible.

Let B ⊃ A a ring extension. Then b ∈ B is called integral over A if there
is an equation

bk + a1 bk−1 + · · · + ak = 0, ai ∈ A, some k > 1.

Moreover, B is called integral over A if every element in B is integral over

A.
Remark 7.1. If B is a finitely generated A-algebra, then B is integral over A
if and only if B is a finite A-module.
To see this, it is sufficient to prove that: b ∈ B integral over A ⇐⇒ A[b]
is a finitely generated A-module. Now the direction =⇒ is clear. To show
⇐= , assume that A[b] is a finite A-module, and let ω1 , . . . , ωr be A-module
generators. Then if c ∈ A[b] is arbitrary, we get equations
r
X
cωi = aij ωj , I = 1, . . . , r, aij ∈ A.
j=1

We can rewrite this as

 
ω1
(cId − M )  ...  = 0, M = (aij ).
 
ωr

51
52 CHAPTER 7. FINITE MORPHISMS, NOETHER NORMALIZATION

Multiplying through with the adjoint of cId − M we obtain

det(cId − M )ωi = 0, i = 1, . . . , r.

Since 1A = 1B can be linearly combined from the ωi , we see that det(cId −

M ) = 0 is an integrality equation (monic equation) for c over A.
Definition 7.2. Let X, Y be affine varieties, and f : X → Y a morphism
with f (X) = Y . Then f is called finite if

f ∗ : OY (Y ) ,→ OX (X)

is an integral ring extension.

Remark 7.3. If f is finite as in Definition 7.2, then f has finite fibers: indeed,
suppose X ⊂ An and let t1 , . . . , tn be the coordinate functions on An . Then
ti assumes only finitely many values on the fiber f −1 (y), y ∈ Y , since we
have a monic equation

tki + a1 tik−1 + · · · + ak = 0 ai ∈ OY

and if x ∈ f −1 (y)

ti (x)k + a1 (y)ti (x)k−1 + · · · + ak (y) = 0

has only finitely many roots ti (x).

Theorem 7.4. If f : X → Y is a finite morphism between affine varieties,
f (X) = Y , then f is surjective.
Proof. Let y = (y1 , . . . , yn ) ∈ Y ⊂ An with my ⊂ OY (Y ) the maximal ideal
of y, my = (t1 −y1 , . . . , yn −tn ). The variety f −1 (y) has equations f ∗ (ti ) = yi ,
i = 1, . . . , n. Thus by the Nullstellensatz 3.2

f −1 (y) = ∅ ⇐⇒ (f ∗ (t1 ) − y1 , . . . , f ∗ (tn ) − yn ) = OX (X).

In other words, the fiber is empty if and only if my · OX (X) = OX (X). Hence
the assertion follows from Lemma 7.5 below.
Lemma 7.5. Suppose B ⊃ A is finite. Then, if a ⊂ A is an ideal, we have
the implication
a ( A =⇒ aB ( B.
 53

Proof. B contains 1A , thus for every a ∈ A with aB = 0, we have a = 0. On

the other hand, if a 6= A, then 0 ∈/ 1 + a. Thus Lemma 7.5 in turn follows
from Lemma 7.6 below.
Lemma 7.6 (Nakayama’s Lemma). Suppose M is a finite A-module, a ⊂ A
an ideal. Suppose that for every a ∈ 1 + a, the equation aM = 0 can only
hold if M = 0. Then aM = M implies M = 0.
Proof. Suppose M is generated by ω1 , . . . , ωr as an A-module. Then the
hypothesis aM = M translates into the existence of a matrix M with entries
in A such that  
ω1
 .. 
(Id − M )  .  = 0.
ωr
Multiplying by the adjoint matrix again, we obtain that det(Id − M ) ∈ 1 + a
annihilates M , which by hypothesis implies M = 0.
Corollary 7.7. Every finite morphism as in Theorem 7.4 maps closed sets
to closed sets.
Proof. If Z ⊂ X is closed (and, without loss of generality, irreducible), the
map f |Z : Z → f (Z) is finite. Hence Theorem 7.4 yields f (Z) = f (Z).
Finiteness is a local property:
Theorem 7.8. If f : X → Y is a morphism of affine varieties, and if every
y ∈ Y has an affine open neighborhood U 3 y such that f |f −1 (U ) : V :=
f −1 (U ) → U is finite, then f is finite.
Proof. We can assume that U is principal open and cover Y with finitely
many Ugα with gα ∈ OY (Y ). That those cover means
(gα ) = OY (Y )
by the Nullstellensatz. Then
Vα := f −1 (Ugα ) = Uf ∗ (gα )
is principal open, and
OX (Vα ) = Bgα , B := OX (X),
OY (Ugα ) = Agα , A := OY (Y ).
54 CHAPTER 7. FINITE MORPHISMS, NOETHER NORMALIZATION

The hypothesis says that B[1/gα ] is a finite A[1/gα ]-module, generated by

elements ωi,α , say. We can even assume ωi,α ∈ B after clearing denominators.
We claim that all the ωi,α taken together form a generating system for B/A:
indeed, every b ∈ B has a representation for all α as
X ai,α
b= ω , some nα ∈ N.
nα i,α
i
g α

But (gαnα ) = P
1 since the gαnα have no common zeroes on Y , hence there are
hα ∈ A with α gαnα hα = 1. Thus
X XX
b=b gαnα hα = ai,α hα ωi,α .
α i α

Thus we can define:

Definition 7.9. A morphism f : X → Y of (quasi-projective) varieties is

called finite if for all y ∈ Y there is an open affine neighborhood V 3 y such
that f −1 (V ) =: U ⊂ X is affine and f |U : U → V is finite.

Theorem 7.8 tells us that “finite” in this new sense coincides with the
notion of “finite” introduced in Definition 7.2.
Remark that we immediately obtain that for finite f in the sense of Def-
inition 7.9, it continues to be true that all fibers f −1 (y) are finite, and every
such f is surjective.

Theorem 7.10. If f : X → Y is a morphism of varieties and f (X) ⊂ Y is

dense, then f (X) contains an open subset of Y .

Proof. We can assume X, Y affine (and irreducible). Via f ∗ , we have an

inclusion of coordinate rings OY (Y ) ⊂ OX (X). We introduce notation for
quotient fields:

k(X) = Quot(OX (X)), k(Y ) = Quot(OY (Y ))

and put r := tr.deg(k(X)/k(Y )). Suppose that u1 , . . . , ur ∈ OX (X) are

algebraically independent over k(Y ). Then we have the inclusions

OX (X) ⊃ OY (Y )[u1 , . . . , ur ] = O(Y × An ) ⊃ OY (Y )

 55

which means we can factor f as f = g ◦ h where

h : X → Y × Ar , g = pr1 : Y × Ar → Y.

By construction, every element v ∈ OX (X) is algebraic over O(Y × Ar ),

thus av is integral for a suitable a ∈ O(Y × Ar ). Let v1 , . . . , vm be k-algebra
generators of OX (X) and ai ∈ O(Y × Ar ) such that ai vi is integral over
O(Y × Ar ). Put F = a1 · · · · · am . Now, by construction, the functions vi
restricted to {h∗ (F ) 6= 0} ⊂ X are integral over O(Y × Ar )[1/F ], i.e.,

h |{h∗ (F )6=0} : {h∗ (F ) 6= 0} → UF ⊂ Y × Ar

is finite. Thus by Theorem 7.4, h({h∗ (F ) 6= 0}) = {F 6= 0}. In other words,

UF ⊂ h(X). It remains to show that P g(UF ) contains an open subset of Y .
But g is just a projection! So if F S
= Fα (y)T α , T a tuple of coordinates
on Ar , Fα ∈ OY (Y ), then g(UF ) ⊃ UFα .
Theorem 7.10 may look unspectacular at first sight, but it really expresses
a very fundamental and remarkable property of morphisms between varieties.
For example, an analogue of it does not hold in the√differentiable category
(think of a “dense wind” R1 → T = R2 /Z2 , x 7→ (x, 2x) mod Z2 ).
Theorem 7.11. Let X ⊂ Pn be a projective variety. Let L ⊂ Pn be a
d-dimensional (projective) linear subspace with L ∩ X = ∅. Let

πL : X → Pn−d−1

be the projection with center L; this means that if L is given by linear equa-
tions L0 = · · · = Ln−d−1 = 0, then

πL (x) := (L0 (x) : · · · : Ln−d−1 (x)).

Then π := πL : X → πL (X) is finite.

Before embarking on the proof we state and prove a few corollaries.
Corollary 7.12 (Noether normalization). For every irreducible projective
variety X, there is a finite morphism f : X → Pn for some n ∈ N.
Proof. Suppose X ⊂ PN . If there is a point x ∈ PN \X, then we project to
PN −1 from x, and by Theorem 6.2, and the projection is finite onto its image.
If the image is not yet the entire PN −1 , we continue in the same manner.
56 CHAPTER 7. FINITE MORPHISMS, NOETHER NORMALIZATION

Corollary 7.13. If X is affine, then there is a finite morphism f : X → An

for some n ∈ N.

Proof. Suppose X ⊂ AN ⊂ PN , and let X ⊂ PN its closure. If X 6= AN , we

can project from x ∈ PN \AN , x ∈ X. Then X gets mapped into AN −1 ⊂
PN −1 and we conclude by the same pattern as in the proof of Corollary
7.12.

Corollary 7.14. If F0 , . . . , Fs ∈ k[X0 , . . . , Xn ]m are homogeneous polynomi-

als of degree m > 0 without common zeroes on a projective variety X ⊂ Pn ,
then
f = (F0 : · · · : Fs ) : X → f (X)
is finite. Thus morphisms which can be defined by “one row” in the sense of
Theorem 3.13 are finite.

Proof. This follows from Theorem 7.11 since f is the composition of the m-th
Veronese embedding of X and a linear projection.

Proof. (of Theorem 7.11) Let y0 , . . . , yn−d−1 be homogeneous coordinates in

Pn−d−1 so that π is given by yj = Lj (x), j = 0, . . . , n − d − 1, x = (x0 : · · · :
xn ) ∈ X. Put
Ui := π −1 An−d−1


yi 6=0 ∩X
the affine open subset of X given by Li 6= 0. We show that

π : Ui → An−d−1
yi 6=0 ∩ π(X)

is finite. Every function g ∈ OX (Ui ) is of the form

Gi (x0 , . . . , xn )
g=
Lm i

where Gi is homogeneous of degree deg G = m. Consider the map

π̃ : X → Pn−d
x 7→ (z0 : · · · : zn−d ), zj := (Lj (x))m , j ≤ n − d − 1, zn−d := Gi (x).

It is a morphism since the Lj have no common zeroes on X. So by Thorem

6.2, π̃(X) ⊂ Pn−d is closed. Suppose F1 , . . . , Fs are homogeneous equations
 57

for π̃(X). Since the Lj have no common zeroes on X, the point (0 : · · · : 0 :

1) ∈ Pn−d is not in π̃(X), which means

z0 = · · · = zn−d−1 = F1 = · · · = Fs = 0

has no solution in Pn−d . By the Nullstellensatz 3.2, we get

(z0 , . . . , zn−d−1 , F1 , . . . , Fs ) ⊃ (z0 , . . . , zn−d )k some k > 0.

In particular, we have an equation

n−d−1
X s
X
k
zn−d = Hj zj + Pj Fj
j=0 j=1

for some polynomials Hj , Pj . Then

n−d−1
X
k (k−1)
Φ(z0 , . . . , zn−d ) := zn−d − Hj ≡ 0 on π̃(X)
j=0

(k−1)
where Hj is the homogeneous component of degree k − 1 of Hj . This can
be rephrased by saying that

Φ(Lm m
0 , . . . , Ln−d−1 , Gi ) ≡ 0 on X.

Then dividing by Lmk

i we get an integrality equation for g over O(An−d−1 yi 6=0 ∩
π(X)):
k−1
X
k
g − H̃k−j (ξ0m , . . . , 1, . . . , ξn−d−1
m
)g j = 0
j=0

where ξr = yr /yi are coordinates on An−d−1

yi 6=0 .
58 CHAPTER 7. FINITE MORPHISMS, NOETHER NORMALIZATION
Chapter 8

Dimension theory

In Definition 3.15, we defined a rational function on an irreducible variety X

as a rational map to A1 . This means that the totality of all rational functions
can be described as
k(X) = lim OX (U )
−→
U ⊂X open,U 6=∅

and k(X) is a ring. In the following Remark we summarize some properties

that are immediate from the definition.
Remark 8.1. 1. Not only is k(X) a ring, it is a field : if [(f, U )] ∈ k(X),
and f 6= 0, then [(1/f, U − {f = 0})] is its inverse.

2. If X is affine, then k(X) = Quot(OX (X)).

3. If X is projective, then
 
f
k(X) = | f, g ∈ k[X0 , . . . , Xn ]/I(X), f, g homogeneous of the same degree, g 6≡ 0 .
g

4. If U ⊂ X is open and nonempty, then k(X) ' k(U ).

Definition 8.2. Let X be an irreducible variety. Then we define its dimen-
sion by putting
dim X := tr.degk k(X).
If X is reducible, we set dim X := maxi dim Xi where Xi is an irreducible
component of X. If Y ⊂ X is a (closed) subvariety

codimX Y := dim X − dim Y

59
60 CHAPTER 8. DIMENSION THEORY

is called the codimension of Y in X. Irreducible varieties of dimension 1,

2, 3 . . . are called curves, surfaces, threefolds, . . .
In the sequel all varieties will be irreducible unless otherwise stated.
Remark 8.3. 1. If X is projective resp. affine, and f : X → Pn resp. An a
finite morphism (which always exists by Corollaries 7.12 and 7.13), then
n = dim X. Namely, in those cases k(X) is a finite algebraic extension
of k(T1 , . . . , Tn ). This also shows that our definition of dimension is
intuitively reasonable.
2. If U ⊂ X is open and nonempty, then dim U = dim X.
3. dim X = 0 (and X possibly reducible) ⇐⇒ X is a finite point set.
Namely, ⇐= is clear. For the direction =⇒ assume without loss of
generality that X is irreducible, affine, X ⊂ An . Then the coordinate
functions ti on X are algebraic over k, hence can take only finitely
many values.
4. For varieties X, Y , dim(X × Y ) = dim X + dim Y . Indeed, it suffices
to show this for X ⊂ Am , Y ⊂ An affine. Let dim X = d, dim Y = e,
x1 , . . . , xm , y1 , . . . , yn affine coordinates on Am , An such that x1 , . . . , xd
are algebraically independent in C(X), y1 , . . . , ye are algebraically in-
dependent in C(Y ).
Now OX×Y (X ×Y ) is generated by x1 , . . . , xm , y1 , . . . , yn which are alge-
braically dependent on x1 , . . . , xd , y1 , . . . , ye . Thus dim(X ×Y ) ≤ d+e.
Let us show that x1 , . . . , xd , y1 , . . . , ye are algebraically independent on
X × Y , which will prove dim(X × Y ) ≥ d + e. Suppose
F (x1 , . . . , xd , y1 , . . . , ye ) = 0 on X × Y.
Since this means that for all x̃ ∈ X we have F (x̃, y1 , . . . , ye ) = 0 on
Y , the algebraic independence of the y’s implies that every coefficient
ai (x1 , . . . , xd ) of F (viewed as a polynomial in y1 , . . . , ye ) is zero on X.
Since x1 , . . . , xd are also algebraically independent, it follows ai ≡ 0 on
X.
5. It is dim Grass(r, n) = r(n − r) since Grass(r, n) is irreducible (because
GLr (k) is irreducible and acts transitively on Grass(r, n)) and since
Grass(r, n) contains open subsets isomorphic to Ar(n−r) as we saw in
the proof of Proposition 5.9.
 61

A good dimension function should decrease on proper subvarieties:

Theorem 8.4. If X, Y are not necessarily irreducible varieties, X ⊂ Y
closed, then dim X ≤ dim Y . If Y is irreducible, then dim X = dim Y hap-
pens only if X = Y .
Proof. It suffices to prove that for X, Y affine and irreducible, X ⊂ Y ⊂ AN
closed. Suppose t1 , . . . , tN are coordinates on AN and dim Y = n. Thus each
n+1 out of the t1 , . . . , tN are algebraically dependent on Y , whence the same
holds for X and
tr.degk k(X) ≤ tr.degk k(Y ).
If dim X = dim Y = n, then after reordering we can assume that t1 , . . . , tn
are algebraically independent on X. These are then also algebraically in-
dependent on Y . Let u ∈ OY (Y ), u 6= 0. Then there is a polynomial
p(t1 , . . . , tn , u) ∈ k[t1 , . . . , tn , u] with

p = a0 (t1 , . . . , tn )uk + · · · + ak (t1 , . . . , tn ) ≡ 0 on Y.

We can assume ak (t1 , . . . , tn ) 6= 0 by assuming p(t1 , . . . , tn , u) of smallest

degree in u. Suppose that u vanishes on X. Then ak (t1 , . . . , kn ) ≡ 0
on X, and by the algebraic independence of the t1 , . . . , tn on X, we have
ak (t1 , . . . , tn ) = 0 on all of AN , hence on Y , a contradiction. So u does not
vanish on X if it does not vanish on Y . This means X = Y .
Theorem 8.5. Every irreducible component of a hypersurface (=the zero
locus of a single polynomial, homogeneous in the projective case) in An (or
Pn ) has codimension 1.
Proof. Suppose without loss of generality that X ⊂ An is given by an irre-
ducible polynomial F 6= 0. After reordering the coordinates t1 , . . . , tn we may
assume that tn occurs in F . Then t1 , . . . , tn−1 are algebraically independent
on X, for G(t1 , . . . , tn−1 ) ≡ 0 on X would imply, by the Nullstellensatz that
F would divide a power Gl , some l > 0, and this is only possible if G is the
zero polynomial. Hence, dim X = n − 1, using Theorem 8.4 (dim X = n is
impossible because then X = An .
In fact, we have a converse to the latter result.
Theorem 8.6. If X ⊂ An is a closed subvariety all of whose irreducible com-
ponents have codimension 1, then X is a hypersurface and IX is a principal
ideal. The same holds in the case of a projective X ⊂ Pn .
62 CHAPTER 8. DIMENSION THEORY

Proof. Again it suffices to do the proof for X ⊂ An affine and irreducible.

Since X 6= An , there is a polynomial F 6= 0 in k[x1 , . . . , xn ] with F ≡ 0 on
X. Since X is irreducible, one irreducible factor of F , F 0 say, also vanishes
identically on X. Let Y := {F 0 = 0} ⊂ An . Then Y is irreducible: indeed,
GH ≡ 0 on Y implies, by the Nullstellensatz, F 0 | (GH)l , some l > 0, thus
G ≡ 0 or H ≡ 0 on Y . But X ⊂ Y . Hence Theorems 8.4 and 8.5 p imply
0
X = Y . Now IX is generated by F since by the Nullstellensatz IX = (F 0 )
and (F 0 ) is a prime ideal, hence radical.
The next result says that the dimension function does not drop too much.
Theorem 8.7. If X ⊂ PN is an irreducible projective variety, F ∈ k[X0 , . . . , XN ]m
with m > 0 and F 6≡ 0 on X, then for the intersection dim(X ∩ {F = 0}) =
dim X − 1, i.e. (X ∩ {F = 0} contains at least one component of dimension
dim X − 1.
In particular, a projective variety contains subvarieties of any dimension
≤ dim X. Also we immediately get the following “combinatorial” character-
ization of dimension:
Corollary 8.8. One can define the dimension of a projective variety as the
maximal integer n such that there is a chain
Y0 ) Y1 ) · · · ) Yn ) 0
of irreducible subvarieties Yi ⊂ X.
Proof. (of Theorem 8.7) For every projective variety X ⊂ PN one can find a
homogeneous polynomial G(X0 , . . . , XN ) (of every degree m > 0) which does
not vanish identically on any component of X (for example, pick a point in
every component and take a linear form, or a product of such, which does
not vanish in any of those points).
Now suppose X ⊂ PN is not necessarily irreducible and F 6≡ 0 on every
component of X. Then Theorem 8.4 implies
X (1) := X ∩ {F = 0}
has dimension strictly smaller than dim X. Now pick a homogeneous poly-
nomial F1 , of degree deg F , which does not vanish on any component of X (1) .
Continuing in this fashion we get a chain of subvarieties (possibly reducible)
X = X (0) ⊃ X (1) ⊃ · · · ⊃ X (i+1) := X (i) ∩ {Fi = 0},
 63

F0 := F , dim X (i+1) < dim X (i) . If dim X = n, then X (n+1) is empty. This
means that F0 = F, . . . , Fn have no common zeroes on X. Now, if X is
irreducible, we get by Corollary 7.14 that

f : X → Pn
f (X) = (F0 (X) : · · · : Fn (X))

is finite onto its image f (X) ⊂ Pn . Hence dim X = dim f (X) = n, and
by Theorem 8.4, f (X) = Pn . Now, if, arguing by contradiction, we had
dim X (1) < n−1, the already X (n) would be empty. I.e., already F0 , . . . , Fn−1
would have no common zeroes on X. But then (0 : · · · : 0 : 1) wouldn’t be
in the image of f , a contradiction.

This leads to a third way of characterizing dimension which is more geo-

metric:

Corollary 8.9. The dimension n of a projective variety X ⊂ PN is equal to

N − s − 1 where s is the maximal dimension of a linear subspace L ⊂ PN
with X ∩ L = ∅.

Proof. If L is of dimension s ≥ N − n, then L can be defined by ≤ n

equations, so Theorem 8.7 gives dim X ∩ L ≥ 0, so X ∩ L 6= ∅. On the other
hand, if in the proof of Theorem 8.7 we take the Fi equal to L0 , . . . , Ln , then
(L0 = · · · = Ln = 0) = L satisfies dim L = N − n − 1 and X ∩ L = ∅.

We can also regard Theorem 8.7 as giving a strong existence result for
solution sets of polynomial equations:

Corollary 8.10. The zero set of r homogeneous polynomials F1 , . . . , Fr on

an n-dimensional projective variety has dimension ≥ n − r; so in particular,
if r ≤ n, then solutions exist.

This implies for example that in P2 , any two curves intersect since by
Theorem 8.6, these are defined by one equation; this is false on P1 × P1 .
Lines of the same ruling do not intersect. This proves also that P2 is not
isomorphic to P1 × P1 although the two are birational.
We can also use Corollary 8.10 to see that any curve of degree ≥ 3 in P2
has an inflection point, and for many similar geometric existence results.
We now prove a strengthening of Theorem 8.7.
64 CHAPTER 8. DIMENSION THEORY

Theorem 8.11. Let X be an irreducible projective variety X in PN , and F a

homogeneous polynomial of positive degree which does not vanish identically
on X. Then every irreducible component of X ∩ {F = 0} has dimension
dim X − 1 =: n − 1.

Proof. We go back to the set-up we produced in the course of the proof of

Theorem 8.7: there we produced a chain of (possibly reducible) subvarieties

X = X (0) ) X (1) ) · · · ) X (i+1) = X (i) ∩ {Fi = 0}

with F0 , . . . , Fn homogeneous polynomials of the same degree m without

common zeroes on X, F0 = F . Then we saw that

f : X → Pn , x 7→ (F0 (x) : · · · : Fn (x))

is finite onto its image f (X). Let An(xi 6=0) ⊂ Pn be the standard affine chart,
and Ui ⊂ X its preimage under f . This is open and affine (the latter since
we can realize f as the composition of a Veronese embedding with a linear
projection).
Now it suffices to prove that every component of (X ∩ {F = 0}) ∩ Ui has
dimension n−1 (for all i). The latter set is the zero set, in Ui , of the function

F
ϕ := ∈ OX (Ui ).
Fi
Put U := Ui . Then f restricted to U gives a map

f : U → An , x 7→ (f1 (x), . . . , fn (x))

which is finite onto its image, and f1 = f is the dehomogenization of F . We

have to show that every component of {ϕ = 0} in U has dimension ≥ n − 1.
To do this we will prove that f2 , . . . , fn are algebraically independent on every
component of ϕ = f1 = 0 in U . Let P ∈ k[t2 , . . . , tn ]. To prove that if P is
not the zero polynomial, then R := P (f2 , . . . , fn ) 6≡ 0 on every component
of ϕ = 0, it suffices to show

∀ Q ∈ OU (U ) : RQ ≡ 0 on (ϕ = 0) =⇒ Q ≡ 0 on (ϕ = 0). (8.1)

For, if
{u ∈ U | ϕ(u) = 0} = U (1) ∪ · · · ∪ U (t)
 65

is the decomposition into irreducible components and, for example, R ≡ 0

on U (1) , then we could take a Q which vanishes on U (2) ∪ · · · ∪ U (t) , but not
on U (1) . Then we would have RQ ≡ 0 on (ϕ = 0), but Q 6≡ 0 on (ϕ = 0), a
contradiction.
By the Nullstellensatz, condition (8.1) is equivalent to

ϕ | (RQ)l some l > 0 =⇒ ϕ | Qk some k > 0. (8.2)

Thus Theorem 8.11 follows from the next Lemma.

Lemma 8.12. Let B = k[T1 , . . . , Tn ] and A ⊃ B an integral domain which
is integral over B. Let x := T1 , and suppose B 3 y = p(T2 , . . . , Tn ) 6= 0.
Then for all u ∈ A we have that

x | (yu)l in A for some l > 0 =⇒ x | uk for some k > 0.

Proof. This uses that x, y are relatively prime in B. In fact, replacing y l by

z and ul by v, it suffices to show: if x, z are relatively prime in k[T1 , . . . , Tn ],
then
x | zv in A =⇒ x | v k some k > 0.
This means, intuitively, that the property that x, z are relatively prime is
“inherited” in a certain sense by the overring A ⊃ B, which is integral over
B.
Put K = Quot(B) = k(T1 , . . . , Tn ). If t ∈ A is integral over B, then t is
algebraic over K. Let F (T ), T = (T1 , . . . , Tn ), be the minimal polynomial
(with leading coefficient 1) of t over K. Then

t integral over B ⇐⇒ F (T ) ∈ B[T ].

Indeed, to see =⇒ , ⇐= being trivial, remark that if G(t) = 0 for a

G ∈ B[T ] with leading coefficient 1, then G(T ) = F (T )H(T ) in K[T ]. But
B is factorial, whence by Gauss’s Lemma, F (T ), H(T ) ∈ B[T ].
Now if zv = xw, v, w ∈ A, let

F (T ) = T k + b1 T k−1 + · · · + bk

be the minimal polynomial of w, bi ∈ B. Then the minimal G(T ) of v =

(xw)/z is
 x k  zT 
F ,
z x
66 CHAPTER 8. DIMENSION THEORY

namely,
xb1 k−1 x k bk
G(T ) = T k + T + ··· + k . (8.3)
z z
Since v is integral over B,
x i bi
∈ B ∀ i.
zi
But then, since x, z are relatively prime, z i | bi , and substituting v for T in
(8.3) yields x | v k .
Corollary 8.13. Let X ⊂ PN be an irreducible quasi-projective variety, and
F 6= 0 a homogeneous polynomial of positive degree such that F 6≡ 0 on X.
Then every irreducible component of X ∩ {F = 0} has codimension 1.
Proof. It suffices to apply Theorem 8.11 to the closure X ⊂ PN , and remark
that X ∩ {F = 0} = (X ∩ {F = 0}) ∩ X.
The next result also follows directly from Theorem 8.11 now.
Corollary 8.14. If X ⊂ PN is irreducible and quasi-projective, dim X = n,
and Y ⊂ X is the zero set of homogeneous polynomials of positive degrees
F1 , . . . , Fm , then every component of Y has dimension ≥ n − m.
Theorem 8.15. Let X, Y ⊂ PN be irreducible, quasi-projective varieties,
dim X = n, dim Y = m. Then every component Z of X ∩ Y has dimension
≥ n + m − N . If X and Y are projective and n + m ≥ N , then X ∩ Y 6= ∅.
Proof. To prove the first assertion, it suffices to consider the case when
X, Y ⊂ AN are affine. Then
X ∩ Y ' (X × Y ) ∩ ∆AN
and the diagonal ∆AN ⊂ A2N is defined by N equations. Then the first asser-
tion follows from Corollary 8.14. The second assertion follows by considering
the affine cones over X, Y and applying the first assertion.
We can also say that
r
\ r
X
codim Pn Xi ≤ codimPn (Xi )
I=1 i=1
for any finite number r of irreducible quasi-projective subvarieties Xi ⊂ Pn .
The next theorem describes how the dimensions of the fibers of a mor-
phism can vary and has very many applications.
 67

Theorem 8.16 (Theorem about the (Upper-Semicontinuity of) Fiber Di-

mension). Let f : X → Y be a morphism of irreducible varieties, and suppose
that f (X) = Y , dim X = n, dim Y = m. Then we have m ≤ n and

1. for every component F of a fiber f −1 (y), y ∈ Y , we have dim F ≥ n−m,

2. and there is a nonempty open subset U ⊂ Y with dim f −1 (y) = n − m

for all y ∈ U .

Proof. The assertions are local on Y , so we can assume Y ⊂ AN closed affine.

Since dim Y = m, we can find a chain of subvarieties

Y (0) ) Y (1) ) · · · ) Y (m)

where Y (i) is purely i-codimensional in Y so that Y (m) is a finite point set,

and Y (m) = Y ∩ Z where Z ⊂ AN is defined by m equations. We can assume
that y ∈ Z. After shrinking the affine open which we work on we can assume
Z ∩Y = {y}. Suppose that Z is defined by g1 = · · · = gm = 0; then f −1 (y) in
X is defined by f ∗ (g1 ) = · · · = f ∗ (gn ) = 0. Thus (1) follows from Corollary
8.14.
For (2) we can again replace Y by an affine open W and X by an affine
open V ⊂ f −1 (W ). Since V is dense on f −1 (W ) and f is surjective, we have
that f (V ) is dense in W . Hence there is an inclusion

f ∗ : OW (W ) ,→ OV (V )

of coordinate rings. Let OW (W ) = k[w1 , . . . , wM ], OV (V ) = k[v1 , . . . , vN ]

where the wi resp. vj are k-algebra generators. Now k(V ) has transcendence
degree n − m over k(W ). We can assume after reordering that v1 , . . . , vn−m
are algebraically independent over k(W ), and the other vi are algebraic over
k(W )[v1 , . . . , vn−m ]. Thus we have equations

Fi (vi ; v1 , . . . , vn−m ; w1 , . . . , wM ) = 0, i = n − m + 1, . . . , N

where Fi is a polynomial in the variables vi , v1 , . . . , vn−m with coefficients in

k[w1 , . . . , wM ]. Let Yi the subvariety
S of W defined by the vanishing of the
leading term of Fi . Then U = W \ i Yi is open and nonempty (the leading
terms are nonzero). If y ∈ U , then none of the polynomials

Fi (Ti ; T1 , . . . , Tn−m ; w1 (y), . . . , wM (y))

68 CHAPTER 8. DIMENSION THEORY

is identically zero, whence the functions vi |f −1 (y)∩V are algebraically depen-

dent on the functions v1 |f −1 (y)∩V , . . . , vn−m |f −1 (y)∩V . Since the v1 |f −1 (y)∩V
, . . . , vN |f −1 (y)∩V generate O(f −1 (y) ∩ V ), we obtain dim f −1 (y) ≤ n − m,
and with part (1), equality.
Corollary 8.17. In the situation of Theorem 8.16,the sets Yk := {y ∈ Y |
dim f −1 (y) ≥ k} are closed in Y .
Proof. By (1) of Theorem 8.16, Yn−m = Y , and there is a closed subset
Z ( Y such that Yk ⊂ Z for k > n − m by (2) of that Theorem. If Zi are
the irreducible components of Z, then dim Zi < dim Y , and the claim follows
considering f |f −1 (Zi ) : f −1 (Zi ) → Zi by induction on dim Y .
The following is an often useful criterion to prove irreducibility.
Theorem 8.18. Let f : X → Y be a surjective morphism of projective vari-
eties where we assume that Y is irreducible, but a priori we do not assume
irreducibility of X; if then for all y ∈ Y the fibers f −1 (y) are irreducible and
of the same dimension, then X is irreducible.
S
Proof.S Let X = Xi the decomposition into irreducible components; then
Y = f (Xi ); note that f (Xi ) is closed in Y since Xi is projective, and since
Y is irreducible, Y = f (Xi ) for some i. Put n := dim f −1 (y). By Theorem
8.16, for all i with f (Xi ) = Y there is Ui ⊂ Y open and dense such that
dim((f |Xi )−1 ) = ni for some ni ∈ N and T all y ∈ Ui . −1For indices j with
f (Xj ) 6= Y , put Uj = Y \f (Xj ). Let y ∈ i Ui . Since f (y) is irreducible,
we have f −1 (y) ⊂ Xi0 for some i0 . Then

−1
f −1 (y) ⊂ f |Xi0 (y)
and the reverse inclusion being trivial we get equality

−1
f −1 (y) = f |Xi0 (y)
and n = ni0 . By construction, f |Xi0 is surjective, hence for all y ∈ Y , we

−1
have that the subset f |Xi0 (y) ⊂ f −1 (y) is nonempty and of dimension
≥ ni0 = n by Theorem 8.16. Thus, since f −1 (y) is irreducible and always of
dimension n = ni0 , we have

−1
f −1 (y) = f |Xi0 (y) ∀y ∈ Y.
This means X = Xi0 .
It is easy to construct examples (exercise!) that show that the hypothesis
of equidimensionality of the fibers in Theorem 8.18 cannot be dropped.
Chapter 9

Lines on surfaces, the

associated form of Chow and
van der Waerden, and degree

Let X ⊂ Pn be a hypersurface given by the vanishing of a homogeneous poly-

nomial F of degree d. We consider l-dimensional projective linear subspaces
Λ = P(L) ⊂ Pn which are contained in X, i.e. L ∈ Grass(l + 1, n + 1) with
P(L) ⊂ X.
Definition 9.1. Put PN = P(k[x0 , . . . , xn ]d ), N = n+d


d
, and define
Φn,d,k = Φ = {([F ], Λ) ∈ PN × Grass(k + 1, n + 1) | Λ ⊂ X = {F = 0}}.
Lemma 9.2. The subset Φ ⊂ PN × Grass(k + 1, n + 1) is a subvariety.
Proof. Let U = Ui1 ...ik+1 ' A(k+1)(n−k) be the open subset of Grass(k+1, n+1)
where the Plücker coordinate ωi1 ...ik is not equal to zero, some 1 ≤ i1 < · · · <
ik+1 ≤ n. We show that Φ ∩ (PN × U ) is an algebraic subvariety.
Without loss of generality we can assume that i1 = 1, . . . , ik+1 = k + 1.
Elements in U are matrices
 
1 . . . 0 a11 . . . a1 n−k
A =  ... . . . ... .. ... ..
 
. . 
0 ... 1 ak+1 1 . . . ak+1 n−k
whose rows are a basis of L ∈ Grass(k + 1, n + 1). If
X
F = ci0 ...in X0i0 . . . Xnin
i0 +···+in =d

69
70 CHAPTER 9. LINES ON SURFACES, DEGREE

is an equation for X, then Λ = P(L) ⊂ X = {F = 0} is equivalent to the

following condition: if ai denotes the i-th row vector of A and λi ∈ k, then
F (λ1 a1 + · · · + λk+1 ak+1 )
vanishes identically as polynomial in the λ’s. This means that the coefficients
of all monomials in the λ’s must vanish, and those are polynomials in the aij
and ci0 ...in .
We will now consider more closely the case of lines on surfaces in P3 , i.e.
Φ = Φ3,d,1 . We have two projections
Φ
p q
~ %
PN Grass(2, 4)

where N = d+3


3
and Grass(2, 4) is the Plücker quadric in P5 . We have
q(Φ) = Grass(2, 4) since every line l ∈ P3 lies on some (possibly reducible)
surface of degree d, for example we could take a union of d planes through l.
The next question is: what is dim q −1 (l)? After applying a projectivity, we
may assume l is given by X0 = X1 = 0 in P3 . Then any surface of degree d
containing l is given by an equation of the form
F = X0 G + X1 H, G, H ∈ k[X0 , X1 , X2 , X3 ]d−1
and any such equation defines a surface containing l; thus the fibers of q are
all projective linear subspaces of PN of dimension (the same for all l)
d(d + 1)(d + 5)
dim q −1 (l) = − 1.
6
To see that this is the dimension of q −1 (l), use the exact sequence (where
S = k[X0 , X1 , X2 , X3 ])
 
X1 
β=  
−X0 α= X0 X1
0 / Sd−2 / Sd−1 ⊕ Sd−1 / Sd
which gives
dim q −1 (l) + 1 = dim im(α) = 2 dim Sd−1 − dim Sd−2

2(d+2)(d+1)d−(d+1)d(d−1)
= 2 d−1+3 − d−2+3


3 3
= 6
(d+5)d(d+1)
= 6
.
 71

By Theorem 8.18 it follows that Φ is irreducible. By Theorem 8.16

dim Φ = dim Grass(2, 4) + dim q −1 (l)

d(d+1)(d+5)
= 4+ 6
−1
(d+3)(d+2)(d+1)
= 6
−1 −d+3
= N + 3 − d.

Now by Theorem 6.2, p(Φ) ⊂ PN is closed. Also by Theorem 8.16, clearly

dim p(Φ) ≤ dim Φ. Thus, if dim Φ < N , then dim p(Φ) < N and by Theorem
8.4, p(Φ) 6= PN . This conclusion means that there is a nonempty Zariski
open subset in PN such that the corresponding surfaces do not contain any
lines (in other words, a “general” such surface does not contain lines).
The numerical condition for this, dim Φ < N translates into d > 3. Hence
we have proven:

Theorem 9.3. If d > 3, then a general surface of degree d in P3 does not

contain any lines.

There remain the cases d = 1, 2, 3. The case d = 1 corresponds to planes

and is trivial, these contain an abundance of lines. For d = 2 we get N = 9
and dim Φ = 10. By Theorem 8.16, we have dim p−1 ([Q]) ≥ 1 where Q is
a homogeneous quadratic polynomial defining a quadric. This corresponds
to the fact that every quadric contains infinitely many lines. However, this
case illustrates nicely the jump phenomenon described in Corollary 8.17: the
fiber dimension can jump up on closed subsets of the base. Indeed, if the
quadric is irreducible, dim p−1 ([Q]) = 1: the fiber consists of two disjoint
P1 ’s corresponding to the two rulings if Q is not a cone, and is one P1 if Q is
a cone. But if Q is reducible, it splits as two planes or a double plane, and
then dim p−1 ([Q]) = 2.
The case d = 3 is that of cubic surfaces. Here dim Φ = N = 19. Moreover,
there exist cubic surfaces with finitely many lines on them, for example it is
not hard to find all lines on

X03 + X13 + X23 + X33 = 0.

Thus there is a point in P19 over which the fiber of p has dimension 0. By
Theorem 8.16 we get dim p(Φ) = 19, whence p(Φ) = P19 . Thus we have
proven:
72 CHAPTER 9. LINES ON SURFACES, DEGREE

Theorem 9.4. Every cubic surface in P3 contains at least one line. There is
a nonempty Zariski open subset in P19 such that the corresponding surfaces
contain finitely many lines.
However, there are cubic surfaces with infinitely many lines, e.g. cones
over cubics in P2 . Hence here we observe again jumps in the fiber dimension.
Moreover, one can show that for “most” cubic surfaces with finitely many
lines, the number and configuration of the lines (i.e., their incidence graph)
are independent of the surface. The number is 27. More concerning this in
Chapter 11.
We start discussing the associated form of Chow and van der Waerden.
Let X ⊂ Pn = P(V ), V an (n + 1)-dimensional k-vector space, be a
projective variety all of whose irreducible components have dimension k (X
is purely k-dimensional). Intuitively, we want to describe such X by “coor-
dinates”, i.e. parametrize them by the points of another variety (we have
already accomplished this in cases of hypersurfaces or if X is a linear sub-
space). Write (Pn )∗ = P(V ∗ ). Look at the incidence correspondence
Γ := {(p, H1 , . . . , Hk+1 ) | p ∈ Hi ∀ i} ⊂ X × (Pn )∗ × · · · × (Pn )∗
where the Hi ⊂ Pn are hyperplanes; here we identify a hyperplane with its
defining equation in (Pn )∗ . Clearly, Γ is a closed subvariety of X × (Pn )∗ ×
· · · × (Pn )∗ equipped with two projections
Γ
ψ ϕ
$
X ((Pn )∗ )k+1 .
Clearly, ψ(Γ) = X, and for p ∈ X the set of hyperplanes containing p
forms an (n − 1)-dimensional projective linear subspace Hp ' Pn−1 ⊂ (Pn )∗ .
Thus ψ −1 (p) is irreducible of dimension (k + 1)(n − 1), and isomorphic to
Pn−1 × · · · × Pn−1 (k + 1 factors). By Theorems Theorem 8.16 and 8.18 Γ is
purely k + (k + 1)(n − 1) = (k + 1)n − 1 dimensional, with one irreducible
component lying over each irreducible component of X. Moreover, there
exist points
y = (H1 , . . . , Hk+1 ) ∈ ((Pn )∗ )k+1
such that ϕ(Γ) 3 y and ϕ−1 (y) is a single point: by the same construction
as the one employed in the proof of Corollary 8.8, we can find a chain
X (0) = X ) X (1) = X ∩ H1 ) · · · ) X (k+1) = X ∩ Hk+1 = {p}
 73

where p ∈ X is a point, and the Hi ’s are hyperplanes; namely we can choose

the Hi ’s such that all of them pass through p, Hi+1 does not contain any
irreducible component of X (i) for i < k, and Hk+1 contains only p ∈ X (k)
and none of the other finitely many points of X (k) . Then, by Theorem 8.16,
we obtain
dim ϕ(Γ) = dim Γ = (k + 1)n − 1,
i.e., ϕ(Γ) ⊂ ((Pn )∗ )k+1 is purely 1-codimensional. Now Theorem 8.6 remains
valid, with the same proof, in the multi-graded set-up:
Theorem 9.5. If Y ⊂ Pn1 × · · · × Pnk is an algebraic subvariety and if
all components of Y have dimension n1 + · · · + nk − 1, then Y is defined
by a single equation F = 0 where F is homogeneous in each of the k set
of variables separately; F is unique up to a constant factor if we choose it
without multiple factors.
Thus we can make
Definition 9.6. The multi-homogeneous polynomial FX without multiple
factors (unique up to a constant) which defines ϕ(Γ) ⊂ ((Pn )∗ )k+1 is called
the Cayley form of X or the associated form of Chow and van der Waerden
of X.
Remark 9.7. The variety X is determined by the datum of FX :
∀ p ∈ Pn : (p ∈ X) ⇐⇒
(∀ H1 , . . . , Hk+1 ∈ (Pn )∗ with p ∈ H1 ∩ · · · ∩ Hk+1 : FX (H1 , . . . , Hk+1 ) = 0) .
Indeed, =⇒ is nothing but the definition of FX . To see ⇐= , notice
that if p ∈ X, then, by the by now standard construction employed T in the
Tof Corollary 8.8, there are hyperplanes H1 , . . . , Hk+1 with p ∈ i Hi and
proof
X ∩ i Hi = ∅. Such Hi are not in the image of ϕ, i.e. FX (H1 , . . . , Hk+1 ) 6= 0.
If u0 , . . . , un are homogeneous coordinates in Pn and
n
X (i)
Hi = vj uj
j=0

the equation of a hyperplane in the i + 1-th copy of (Pn )∗ in ((Pn )∗ ), then we

can take
(0) (k)
(v0 : · · · : vn(0) ; . . . ; v0 : · · · : vn(k) )
as coordinates in ((Pn )∗ ).
74 CHAPTER 9. LINES ON SURFACES, DEGREE

Remark 9.8. The Cayley form

(0) (k)
FX = FX (v0 , . . . , vn(0) ; . . . ; v0 , . . . , vn(k) )

is homogeneous of the same degree, d say, in each of the sets of variables

(i) (i)
v0 , . . . , vn , i = 0, . . . , k. This follows because ϕ(Γ) ∈ (Pn )∗ × · · · × (Pn )∗ is
invariant under permutations of the factors.
Definition 9.9. The coordinates of FX with respect to the basis given by
(0) (0) (k) (k)
monomials in k[v0 , . . . , vn ; . . . ; v0 , . . . , vn ] are called the Chow coordi-
nates of X.
Thus attached to X we get a discreet parameter d and continuous pa-
rameters (the coefficients of monomials in FX ) which determine X ⊂ Pn .
In the rest of this Chapter we assume char(k) = 0 for simplicity, but the
results below continue to hold without this assumption.
Theorem 9.10. Let X be a projective subvariety of Pn as above, and FX
its Cayley form. Suppose FX is of degree d in each of the sets of variables
(i) (i)
v0 , . . . , vn , i = 0, . . . , k. Then this d is equal to the maximum number of
intersection points X ∩ L of X with a linear subspace L ⊂ Pn of dimension
dim L = n − k whenever this number is finite.
We call this d the degree deg X of the projective subvariety X ⊂ Pn .
Proof. Let H1 , . . . , Hk ∈ (Pn )∗ be hyperplanes with |X ∩ i Hi | < ∞ and
T

\ (l)
X ∩ Hi = {x(1) , . . . , x(c) }, x(l) = (u0 : · · · : u(l)
n ), l = 1, . . . , c
i

(such hyperplanes exist by the standard construction in the proof of Corollary

8.8). Let H0 be a further hyperplane and write
n
!
X (i)
Hi = vj uj = 0 .
j=0

Now we want to think of the H1 , . . . , Hk as fixed and of H0 as variable. Then

(0) (0)
FX is a polynomial of degree d in v0 , . . . , vn which vanishes if and only if
n
X (0) (l)
vj uj = 0
j=0
 75

for at least one l. This means that we get a factorization

c X
Y n
(0) (0) (l)
FX (v0 , . . . , vn(0) ; H1 , . . . , Hk ) =α ( vj uj )rl .
l=1 j=0

Here α is a constant and rj ≥ 1 some integers. In other words, c ≤ d and if

(0) (0)
FX (v0 , . . . , vn ; H1 , . . . , Hk ) does not have multiple factors, then c = d.
Thus, to conclude the proof, it suffices to show that for suitable choice
(0) (0)
of H1 , . . . , Hk , the polynomial FX (v0 , . . . , vn ; H1 , . . . , Hk ) does not have
multiple factors. We apply Lemma 9.11 below which we leave as an exercise.

Lemma 9.11. Suppose char(k) = 0. If a polynomial

(0) (k)
F (v0 , . . . , vn(0) ; . . . ; v0 , . . . vn(k) )
(0)
has no multiple factors (and at least some variable vi occurs in F ), then
there are special values
(1)
V0 , . . . , Vn(k) ∈ k
(1) (k)
for the v0 , . . . , vn such that
(0) (1)
F (v0 , . . . , vn(0) ; V0 , . . . , Vn(k) )

has no multiple factors.

Corollary 9.12. Under the hypotheses of Theorem 9.10 there exists a nonempty
Zariski open subset of (H1 , . . . , Hk ) in (Pn )k such that

|X ∩ H1 ∩ · · · ∩ Hk | = d.
(0) (0)
Proof. The sets of (H1 , . . . , Hk ) for which FX (v0 , . . . , vn ; H1 , . . . , Hk )
1. does not vanish identically

2. has no multiple factor

are both Zariski open and nonempty. Namely, let us first consider the set
in (1). It is nonempty since there are hyperplanes H0 , . . . , Hk with X ∩
H0 ∩ · · · ∩ Hk = ∅, thus FX (H0 , . . . , Hk ) 6= 0. It is clear that it is open
(0) (0)
since otherwise the coefficients of all monomials in v0 , . . . , vn occurring in
76 CHAPTER 9. LINES ON SURFACES, DEGREE

(0) (0)
FX (v0 , . . . , vn ; H1 , . . . , Hk ) must vanish and this defines a closed subvariety
of (Pn )k .
Now in the course of the proof of Theorem 9.10 we saw that the set in (2)
is nonempty. It is Zariski open since the subset of homogeneous polynomials
F ∈ k[t0 , . . . , tn ]d with a multiple factor is closed since the multiplication
map

P(k[t0 , . . . , tn ]d−2k ) × P(k[t0 , . . . , tn ]k ) → P(k[t0 , . . . , tn ]d )

(G, H) 7→ GH 2

is a morphism and we can apply Theorem 6.2 to conclude that the image is
closed.
Now being in set (1) means that H1 , . . . , Hk intersect X in finitely many
points, and being in (2) means that their number is d.
0) (0) (k) (k)
The set of all forms F (v0 , . . . , vn ; . . . ; v0 , . . . , vn ) of degree d in each
set of variables forms a projective space PN and by Remark 9.7 we get an
injection

c : {X ⊂ Pn | X purely k−dimensional, deg X = d} → PN

X 7→ [FX ].

Remark 9.13. We summarize some properties of the image of c without proof.

It turns out that im(c) =: Cn,k,d ⊂ PN is a quasi-projective variety. Hence in
this way one can parametrize purely k-dimensional subvarieties X of Pn of
degree d by the points of another variety, similar to the situation for linear
subspaces and Grassmannians. We get a closed (projective) subvariety in
PN if we consider, instead of X’s as before, k-dimensional cycles of degree d
which are formal linear combinations

Zk = m1 X1 + · · · + mp Xp

where Xi ⊂ Pn is irreducible of dimension k and d = m1 deg X1 + · · · +

mp deg Xp . For those one extends the definition of the Cayley form by

FZk := FXm11 . . . FXmll .

In this way we obtain exactly the closure C n,k,d of Cn,k,d in PN . Then C n,k,d
is called the Chow variety of algebraic cycles of dimension k and degree d
 77

in Pn , and Cn,k,d is called the open Chow variety of purely k-dimensional

subvarieties of Pn of degree d.
Both Cn,k,d and C n,k,d are thus parameter spaces for certain algebro-
geometric objects. In general it is unknown how many irreducible compo-
nents these Chow varieties have, even for curves in P3 . Here are some known
results for C 3,1,d , d = 1, 2, 3:
1. For d = 1, C 3,1,1 is the Plücker quadric in P5 , which is irreducible of
dimension 4.
2. For d = 2, C 3,1,2 is reducible with two irreducible components, C 3,1,2 =
C 0 ∪ C 00 , with dim C 0 = dim C 00 = 8. Here C 0 parametrizes plane
conics, and C 00 parametrizes 2 lines in P3 . They intersect in the locus
corresponding to two intersecting lines.
3. For d = 3 there is a decomposition into irreducible components
C 3,1,3 = C (1) ∪ C (2) ∪ C (3) ∪ C (4)
where all C (i) have dimension 12 and C (1) corresponds to unions of three
lines; points in C (2) parametrize unions of a plane conic and a line; C (3)
parametrizes plane cubics; and C (4) parametrizes twisted cubics in P3 .
A funny loose end in our approach to degree is that we still have to show
that the answer for hypersurfaces is as expected:
Theorem 9.14. If X is an irreducible hypersurface in Pn defined by
G(u0 , . . . , un ) = 0
where G is a homogeneous irreducible polynomial of degree d, then deg X = d.
Proof. It is clear that deg X ≤ d. Thus it suffices to show (after dehomoge-
nization): if
X ∩ Pnu0 6=0 =: H ⊂ An
is defined by an irreducible polynomial g ∈ k[X1 , . . . , Xn ] of degree d, then
there is a line l ⊂ An such that H ∩ l consists of exactly d points. We show
this directly as follows.
We search for such an l in the parameter form
l = {(x1 , . . . , xn ) + λ(y1 , . . . , yn ) | λ ∈ k},
x = (x1 , . . . , xn ) ∈ An , y = (y1 , . . . , yn ) ∈ An .
78 CHAPTER 9. LINES ON SURFACES, DEGREE

Let
g = gd + gd−1 + · · · + g0
be the decomposition into homogeneous components, gd 6= 0. Then the
condition g(x + λy) = 0 can be written as
n
!
d
X ∂gd
gd (y)λ + (y)xi + gd−1 (y) λd−1 + · · · = 0. (9.1)
i=1
∂Xi

Now we want to determine x and y such that

1. gd (y) 6= 0 (this is O.K. since k is an infinite field)

2. and such that G̃(λ) = g(x + λy) has degree d (O.K. by (1)) and is
coprime to
n
0
X ∂g
G̃ (λ) = (x + λy)yi .
i=1
∂X i

If we accomplish (2) the proof is complete since then G̃(λ) has d simple roots.
Let Y1 , . . . , Yn , U be further indeterminates. Expand

ϕ := g(X1 + U Y1 , . . . , Xn + U Yn )

in powers of U :
n
!
X ∂g
ϕ = g(X1 , . . . , Xn )+ (X1 , . . . , Xn ) · Yi ·U +· · ·+gd (Y1 , . . . , Yn )U d .
i=1
∂Xi

We cannot have ∂g/∂Xi = 0 for all i = 1, . . . , n (g is nonzero irreducible).

Thus choose x ∈ An such that
n
X ∂g
g(x) 6= 0, (x)Yi 6= 0. (9.2)
i=1
∂Xi

Then
ψ(Y1 , . . . , Yn , U ) := g(x1 + U Y1 , . . . , xn + U Yn )
is irreducible in k[Y1 , . . . , Yn , U ]. To see this, assume by contradiction ψ =
h1 h2 is a nontrivial decomposition. There are two possibilities now:
 79

(I) At least one of the variables Y1 , . . . , Yn occurs in h1 and h2 (it need not
be the same variable though). Then, for suitable u ∈ k ∗ , there would
be a nontrivial decomposition of

ψ(Y1 , . . . , Yn , u) = g(x1 + uY1 , . . . , xn + uYn )

in k[Y1 , . . . , Yn ]. This is a contradiction to the irreducibility of g in

k[X1 , . . . , Xn ] since it will remain irreducible after the substitution
Xi 7→ xi + uYi .

(II) Either h1 or h2 (or both) are polynomials in which none of the Y ’s

occurs. Suppose that it is h1 without loss of generality. But then, since
g(x) 6= 0, by our choice in (9.2), h1 is not divisible by U and there exists
a u ∈ k ∗ such that h1 (u) = 0. But then also g(x1 +uY1 , . . . , xn +uYn ) ≡
0, contradicting (9.2).

Thus ψ is irreducible and ∂ψ/∂U 6= 0 because of the second inequality in

(9.2). Hence
∂ψ
ψ,
∂U
are coprime. Thus there exist polynomials δ ∈ k[Y1 , . . . , Yn ]\{0}, a, b ∈
k[Y1 , . . . , Yn , U ] such that
∂ψ
δ = aψ + b
∂U
since ψ and ∂ψ/∂U are coprime also in k(Y1 , . . . , Yn )[u] by Gauss’s Lemma,
and then we can use the Euclidean algorithm solved backwards to find a
representation as before. Now choose y ∈ An such that

gd (y) 6= 0, δ(y) 6= 0. (9.3)

Then
G̃(U ) = ψ(y1 , . . . , yn , U ) = g(x1 + y1 U, . . . , xn + yn U )
is of degree d since gd (y) 6= 0 and coprime to

∂ψ
G̃0 (U ) = (y1 , . . . , yn , U )
∂U
since δ(y) 6= 0. Hence l = {x + λy | λ ∈ k} is a line with the desired
properties.
80 CHAPTER 9. LINES ON SURFACES, DEGREE

One vexing point in the theory of Cayley forms as we have been developing
it so far is that we do not know how to compute them explicitly! We can
now say what happens in the case of a hypersurface.
Corollary 9.15. Under the hypotheses of Theorem 9.14 the Cayley form of

X = {G(u0 , . . . , un ) = 0}

is
(0) (n−1)
FX (v0 , . . . , vn(0) ; . . . ; v0 , . . . , vn(n−1) )
= G(∆0 , . . . , ∆n )

where (−1)i ∆i is the minor of the matrix

 (0) (0)

v0 ... vn
 .. .. ..
V = .

. . 
(n−1) (n−1)
v0 ... vn
obtained by deleting the i-th column.
Proof. If H1 , . . . , Hn intersect in a point, then it is given by

(∆0 : · · · : ∆n ).

Hence G(∆0 , . . . , ∆n ) vanishes on a Zariski open dense subset of FX = 0.

Thus FX divides G(∆0 , . . . , ∆n ), but according to Theorem 9.14, FX has
multi-degree d, the same as G.
One can make Cayley forms more explicit using resultants, but we don’t
go into this.
We state the following important result about degree without proof.
Theorem 9.16 ((Weak) Bezout’s Theorem). Let X and Y be two closed sub-
varieties in Pn which are pure dimensional. Suppose that X and Y intersect
properly, i.e. dim(X ∩ Y ) = dim X + dim Y − n. Then

deg(X ∩ Y ) ≤ deg X · deg Y.

In fact, by assigning appropriate integers, called multiplicities, to the

irreducible components of X ∩Y , one can even turn the inequality in Theorem
9.16 into an equality.
Chapter 10

Regular and singular points,

tangent space

Let X ⊂ Pn be a quasi-projective variety. For x ∈ X, we want to define

the tangent space Tx X of X in x. Clearly, X ⊂ X ⊂ Pn , and if x ∈ X, we
will want to define Tx X = Tx X, thus we start with a projective X ⊂ Pn .
After reordering the coordinates we can assume that x is in the affine chart
U0 = {X0 6= 0}. Consider X0 = X ∩ U0 ⊂ An and let I(X0 ) = (F1 , . . . , Fm ),
Fj ∈ k[x1 , . . . , xn ] be the ideal of X0 . We can also assume, after an affine-
linear coordinate transformation, that x = (0, . . . , 0) = 0.
For a line L ⊂ Pn through x we want to define the intersection multiplicity
multx (L ∩ X) of L and X in x. Then L0 = L ∩ U0 is given by

L0 = {λa | λ ∈ k}, a ∈ U0 \{0}.

Put Y
f (λ) := gcd(F1 (λa), . . . , Fm (λa)) = (λ − αi )ki .
i

Here the λ = αi correspond to the intersection points of X0 with L0 if these

are finitely many. Then we put multx (X ∩L) := multiplicity of the zero λ = 0
of f (λ), i.e. the highest power of λ which divides all Fi (λa). If Fi (λa) ≡ 0
for all i, we put multx (X ∩ L) = +∞.
This is independent of the choice of the Fi since f (λ) = gcd{F (λa) | F ∈
I(X0 )}, and it is also independent of a choice of affine chart containing x
since we can alternatively define multx (X ∩ L) as follows: if

L = {[µ0 p0 + µ1 p1 ] | (µ0 : µ1 ) ∈ P1 , [p0 ], [p1 ] ∈ Pn }

81
82CHAPTER 10. REGULAR AND SINGULAR POINTS, TANGENT SPACE

and if the homogeneous ideal of X is

I(X) = (Φ1 , . . . , ΦM )
(0) (0) (0)
and p = [µ0 p0 + µ1 p1 ], then multx (X ∩ L) is the highest power of (µ1 µ0 −
(0)
µ0 µ1 ) which divides all Φ’s.

Definition 10.1. A line L is called a tangent to X in x if

multx (X ∩ L) ≥ 2.

Since X0 3 0, in the notation above no Fi has a constant term. Let its

linear term be Li and write Fi = Li + Gi , i = 1, . . . , m. Then

Fi (λa) = λLi (a) + Gi (λa), λ2 | Gi (λa).

Thus λ2 divides Fi (λa) if and only if Li (a) = 0 for all i; thus L0 is tangent
to X0 in 0 if and only

L1 (a) = · · · = Lm (a) = 0.

Definition 10.2. Let us make An ' U0 into a vector space with origin x = 0
by componentwise addition and scalar multiplication. The sub-vector space
of all points a ∈ An which lie on tangents to X in x is called the (embedded,
affine) tangent space of X in x, denoted by Tx X. Its closure in Pn (i.e. the
locus of points in Pn which lie on tangents L to X in x) is called (embedded,
projective) tangent space, denoted by Tx X.

Definition 10.3. The local dimension of X in x is defined by

dimx X := max{dim Z | Z irreducible component of X through x}.

Definition 10.4. A point x ∈ X is called regular if dimx X = dimk Tx X

(= dim Tx X). A point x ∈ X is called singular if dimk Tx X > dimx X. X is
called regular or nonsingular if all points of X are regular.

At the moment it is not a priori clear that X is equal to the union of its
regular and singular points, or that the singular points form a proper closed
subvariety. We will see this later.
 83

Theorem 10.5. If X ⊂ Pn is a projective variety, I(X) = (Fα )α∈A , Fα

homogeneous, then Tx X ⊂ Pn is defined by
n
X ∂Fα
(x)Xi = 0, α ∈ A,
i=0
∂Xi

where (X0 : · · · : Xn ) are homogeneous coordinates in Pn .

If X ⊂ An is affine, I(X) = (Gβ )β∈B , then for x ∈ X, the tangent space
Tx X ⊂ An is defined by
n
X ∂Gβ
(x)(ti − xi ) = 0, β ∈ B,
i=1
∂ti

where (t1 , . . . , tn ) are coordinates in An , and Tx X is an affine linear space

through x ∈ An , which we make into a vector space by choosing x = 0.
Proof. Let us prove the affine case first. Put yi = ti − xi ,

G̃β (y1 , . . . , yn ) := Gβ (y1 + x1 , . . . , yn + xn ).

In the special case when X ⊂ An is a hypersurface with I(X) = (F̃ ), 0 ∈ X,

and if F̃ = L + G with L the linear part, then T0 X = {L = 0}. But
n
X ∂ F̃
L= (0)yi
i=1
∂yi

and
∂ F̃ ∂F (y1 + x1 , . . . , yn + xn ) ∂F (t1 , . . . , tn )
(0) = (0) = (x)
∂yi ∂yi ∂tn
whence n
X ∂F
L= (x)(ti − xi ),
i=1
∂ti
so the formula is valid in the hypersurface case.
In the general case, according to Definition 10.1, we have
\
Tx X = Tx H
H⊃X hypersurface

which concludes the proof in the affine case.

84CHAPTER 10. REGULAR AND SINGULAR POINTS, TANGENT SPACE

If X ⊂ Pn is projective, we can assume without loss of generality that x ∈

{X0 6= 0} ' An ⊂ Pn where t1 = X1 /X0 , . . . , tn = Xn /X0 are coordinates in
An . If X is a hypersurface with IX = (F ), then

f (t1 , . . . , tn ) = F (1, t1 , . . . , tn )

generates the ideal of X ∩ An , thus by the the first part,

n
X ∂f
(x1 , . . . , xn )(ti − xi ) = 0,
i=1
∂ti

where xi are affine coordinates of x, defines the affine tangent space Tx X

and, by definition, Tx X = Tx X. Thus
( n
)
X ∂F
Tx X = (X0 : · · · : Xn ) | (1, x1 , . . . , xn )(Xi − xi X0 ) = 0 .
i=1
∂Xi

But since F is homogeneous, of degree d say, we have the Euler relation

n
X ∂F
Xi = d · F
i=0
∂Xi

and F (1, x1 , . . . , xn ) = 0, whence

n
X ∂F ∂F
(1, x1 , . . . , xn ) · (−xi X0 ) = (1, x1 , . . . , xn )X0 .
i=1
∂Xi ∂X0

In other words, if x = (P0 : · · · : Pn ), xi = Pi /P0 , then

( n
)
X ∂F
Tx X = (X0 : · · · : Xn ) | (P0 , . . . , Pn )Xi = 0 .
i=0
∂X i

This settles the hypersurface case, and the general case follows again by
remarking that \
Tx X = Tx H.
H⊃X hypersurface
 85

Remark 10.6. One must be careful that in Theorem 10.5 one has to take the
Fα (or Gβ ) as ideal generators, i.e. defining equations that do not generate
T the
respective ideals may lead to a different answer. In other words, ifTX = i Hi
where the Hi ⊂ An are hypersurfaces, then in general Tx X 6= Tx Hi . To
see this, take for example

H1 = {y − x2 = 0}, H2 = {y = 0} in A2 .

Then T0 H1 ∩ T0 H2 = x-axis, but T0 (H1 ∩ H2 ) = {0}.

Now consider X ⊂ An affine again, t1 , . . . , tn coordinates on An . Any
polynomial F ∈ k[t1 , . . . , tn ] has a formal “Taylor expansion” around a point
x = (x1 , . . . , xn ) ∈ An , i.e., we can write

F (t1 , . . . , tn ) = F (x) + F1 (t1 − x1 , . . . , tn − xn ) + · · · + Fk (t1 − x1 , . . . tn − xn )

where Fi is homogeneous of degree i. Then we call

n
X ∂F
F1 =: dx F = (ti − xi )
i=1
∂ti

the differential of F in x. For the differential we have the rules

dx (F + G) = dx F + dx G, (10.1)
dx (F G) = F (x)dx G + G(x)dx F. (10.2)

We can view dx F as a linear form on An if we give it a linear structure with

x as origin. Moreover,

Tx X = {dx F = 0 | F ∈ I(X)}.

If g ∈ OX (X) = k[t1 , . . . , tn ]/I(X), G ∈ k[t1 , . . . , tn ] such that G |X = g, then

clearly dx G |Tx X is independent of the choice of G.
Definition 10.7. The linear form

dx g := dx G |Tx X ∈ Tx X ∗

is called the differential of g ∈ OX (X) in x.

The rules (10.1) and (10.2) continue to hold. Moreover, dx c = 0 for
c ∈ k ⊂ OX (X). Let Mx ⊂ OX (X) be the ideal of x ∈ X. Then dx kills
elements in M2x by (10.2).
86CHAPTER 10. REGULAR AND SINGULAR POINTS, TANGENT SPACE

Theorem 10.8. The map

∗
dx : Mx /M2x → TX,x

is an isomorphism of k-vector spaces.

Proof. Clearly, the map dx is well-defined and k-linear. We have im(dx ) =

∗
TX,x ∗
since every linear form l ∈ TX,x extends to a linear form ˜l ∈ (An )∗ and
dx (˜l |X ) = l.
Let us show that ker(dx ) = M2x : we choose coordinates such that x =
(0, . . . , 0); suppose g ∈ Mx is such that dx (g) = 0 and suppose g is repre-
sented by some polynomial G ∈ k[t1 , . . . , tn ], g = G |X . Thus we assume
dx G |Tx X = 0, in other words,

dx G = λ1 dx F1 + · · · + λm dx Fm , where I(X) = (F1 , . . . , Fm ).

Put G̃ := G−λ1 F1 −· · ·−λm Fm , thus dx G̃ = 0. This means G̃ ∈ (t1 , . . . , tn )2 ,

whence
G̃ |X = g ∈ (t1 |X , . . . , tn |X )2 .
But Mx = (t1 |X , . . . , tn |X ).

Theorem 10.9. If mx ⊂ OX,x is the maximal ideal of x in the local ring

OX,x of x ∈ X, then there is an isomorphism

TX,x ' (mx /m2x )∗ .

Proof. For f ∈ OX,x we can find a representation F/G, with F, G ∈ k[t1 , . . . , tn ],

G(x) 6= 0, and define the differential
 
F G(x)dx F − F (x)dx G
dx f := dx |TX,x = |TX,x .
G G(x)2

In just the same way as above one can show that this is well-defined, sat-
isfies the computational rules (10.1) and (10.2), and gives a k-linear map
∗
dx : mx /m2x → TX,x . The same proof that we used for Theorem 10.8 then also
shows that it is an isomorphism.

Corollary 10.10. The tangent space Tx X to a quasi-projective variety X

in a point x ∈ X is an isomorphism invariant and isomorphic to (mx /m2x )∗
where mx ⊂ OX,x is the maximal ideal.
 87

Proof. If ϕ : X → Y is a morphism, ϕ(x) = y, then it induces a pull-back

map ϕ∗ : OY,y → OX,x such that

ϕ∗ (mY,y ) ⊂ mX,x , ϕ∗ (m2Y,y ) ⊂ m2X,x .

Hence it induces a k-linear map

∗
∗
dx ϕ : mX,x /m2X,x → mY,y /m2Y,y .

If ϕ has an inverse ψ, this linear map is an isomorphism.

Definition 10.11. The vector space (mX,x /m2X,x )∗ is called the Zariski tan-
gent space to X in x. Moreover, for ϕ : X → Y a morphism as in the proof
of Corollary 10.10, the linear map dx ϕ is called the differential of ϕ in x.

Note that for morphisms ϕ : X → Y , ψ : Y → Z we have

dx (ψ ◦ ϕ) = dx ψ ◦ dx ϕ, dx (idX ) = idTx X .

Corollary 10.10 sometimes gives an answer to certain embedding prob-

lems:

Corollary 10.12. The union X of the three coordinate axes in A3

X = {x = y = 0} ∪ {x = z = 0} ∪ {y = z = 0}

with I(X) = (xy, yz, xz) is not isomorphic to any subvariety of A2 (in par-
ticular not isomorphic to the union of three lines through a point in A2 ).

Proof. Indeed,
dim T0 X = 3.

We now want to study the loci of regular and singular points in a variety
X.

Theorem 10.13. Let X be an irreducible variety. The set of regular points

x ∈ X, defined by dim Tx X = dimx X, is open and Zariski dense in X.
The set of singular points, by definition those where dim Tx X > dimx X, is
exactly the complement of the regular points. In particular, we always have
dim Tx X ≥ dimx X.
88CHAPTER 10. REGULAR AND SINGULAR POINTS, TANGENT SPACE

Proof. The proof will be divided into several steps.

Step 1. Without loss of generality we can assume X ⊂ An affine and
irreducible. Look at the incidence correspondence

T := {(x, a) ∈ X × An | a ∈ Tx X} ⊂ X × An .
π

X

This is an algebraic subvariety of X × An by Theorem 10.5, and π(T ) = X,

π −1 (x) = Tx X. For this reason one calls T the (total space of the) tangent
bundle in the case when X consists entirely of regular points. In general one
may call it the tangent fiber space.
By Theorem 8.16 and Corollary 8.17, there is an s ∈ N such that dim Tx X ≥
s for all x ∈ X and
{y ∈ X | dim Ty X > s} ⊂ X
is a proper closed algebraic subvariety; here s = minx∈X dim Tx X. Thus it is
sufficient to show s = dim X.
Step 2. We show that s = dim X holds for a hypersurface X ⊂ An with
ideal I(X) = (F ). In this case,
n
X ∂F
(x)(ti − xi ) = 0
i=1
∂ti

defines Tx X ⊂ An , and s = dim X = n − 1 if and only if not all the partials

∂F/∂ti vanish identically on X. Now if char(k) = 0, the contrary means that
F is constant, a contradiction, or if char(k) = p > 0, this means that we can
write F = G(tp1 , . . . , tpn ); since k is algebraically closed and in char(k) = p
the binomial theorem takes the form (a + b)p = ap + bp , we conclude that
F is then a p-th power, a contradiction because then I(X) would not be
generated by F .
Step 3. We reduce to the case of a hypersurface. We claim: for X ⊂ An
irreducible, there exists a hypersurface Y ⊂ Am and open, nonempty subsets
U ⊂ X, V ⊂ Y and an isomorphism ϕ : U → V . Indeed, this claim follows
from Theorem 10.14 and Theorem10.15 below. Now the subset Yreg ⊂ Y of
regular points is open by Step 2, and

dim Ty Y = dim Y = dim X ∀ y ∈ Yreg

 89

whence, by Corollary 10.10, we have for all points x ∈ ϕ−1 (Yreg ∩ V ) ⊂ X

(which is open) that dim Tx X = dim X.
Theorem 10.14. Every irreducible subvariety X ⊂ An is birational to a
hypersurface.
Proof. We can find a transcendence basis x1 , . . . , xr of k(X) such that k(X) ⊃
k(x1 , . . . , xr ) is a separable algebraic extension (this needs k perfect, which is
O.K. since our k is algebraically closed). Then the Theorem of the Primitive
Element says that there is a y ∈ k(X) such that k(X) = k(y, x1 , . . . , xr ).
Now y satisfies a polynomial equation over k(x1 , . . . , xr ). Hence there is an
irreducible nonzero polynomial
f (x1 , . . . , xr , y) = 0
which defines a hypersurface in Ar+1 with function field k(X).
Theorem 10.15. Let X, Y be irreducible and affine. Suppose that X and Y
are birational to each other. Then there exist a nonempty open U ⊂ X and
a nonempty open V ⊂ Y such that U and V are isomorphic.
Proof. Suppose that X ⊂ Am with coordinates xi and Y ⊂ An with coordi-
nates yj . Without loss of generality, we can assume that X and Y are not
contained in any hyperplane. Let
i : k(Y ) ' k(X)
be an isomorphism. Viewing yj as an element of OY (Y ) ⊂ k(Y ), we can con-
sider i(yj ), j = 1, . . . , n. These are not identically zero on X, and there is an
affine open subset W1 ⊂ X on which all the i(yj ) are regular. Then i defines
an inclusion OY (Y ) ,→ OW1 (W1 ) corresponding to a dominant morphism
ϕ : W1 → Y.
Applying the same reasoning to i−1 , we also get a dominant morphism
ψ : W2 → X,
W2 ⊂ Y affine open, such that ψ ◦ ϕ = id, ϕ ◦ ψ = id (as rational maps).
Now consider
ϕ−1 (ψ −1 (W1 )) ⊂ ϕ−1 (W2 ) ⊂ W1 ⊂ X,
ψ −1 (ϕ−1 (W2 )) ⊂ ψ −1 (W1 ) ⊂ W2 .
90CHAPTER 10. REGULAR AND SINGULAR POINTS, TANGENT SPACE

Then ψ ◦ ϕ is defined on ϕ−1 (W2 ) and ϕ ◦ ψ is defined on ψ −1 (W1 ),

ψ ◦ ϕ |ϕ−1 (W2 ) = id, ϕ ◦ ψ |ψ−1 (W1 ) = id.

If we put U := ϕ−1 (ψ −1 (W1 )), V := ψ −1 (ϕ−1 (W2 )), then U ' V via ϕ and
ψ.
As a corollary of Theorem 10.5 and Theorem 10.13 we get

Corollary 10.16. Let X ⊂ Pn be an irreducible projective variety. Then it

singular locus Sing(X) can be described as
   
∂Fi
Sing(X) = rk ≤ codim(X) − 1 ∩ X.
∂Xj
Chapter 11

Cubic surfaces and their lines

Let S ⊂ P3 be a nonsingular cubic surface given by f (x, y, z, w) = 0 where

f = f3 is an irreducible homogeneous polynomial of degree 3, and x, y, z, w
are homogeneous coordinates in P3 .
By Theorem 9.4 we know that S contains at least one line. The next
Proposition states some facts about lines on S which use that S is nonsin-
gular.

Proposition 11.1. Let S ⊂ P3 be a nonsingular cubic surface as before.

1. There are at most three lines passing through a point p ∈ S. If they are
three, they lie in a plane.

2. Every plane E ' P2 ⊂ P3 intersects S in one of the following:

(a) an irreducible cubic curve in E;

(b) the union of an irreducible conic and a line in E;
(c) three distinct lines in E.

In other words, the intersection E ∩ S does not contain a line counted

with multiplicity > 1.

Proof. To show (1) notice that if l ⊂ S is a line, then Tp l = l ⊂ Tp S, hence

all lines through p lie in the plane Tp S (this is a plane since S is nonsingular).
It will follow from the proof of (2) that there are at most three such lines.
For (2) we can assume without loss of generality that E = {w = 0} and
l = {z = w = 0} ⊂ E. We have to show: f |E does not have the equation of

91
92 CHAPTER 11. CUBIC SURFACES AND THEIR LINES

l as a double or triple factor. If so, we would have

f = z 2 L(x, y, z, w) + wQ(x, y, z, w)

with L linear and Q quadratic. But then S would be singular in a point

where z = w = Q = 0, i.e. in the zeroes of Q on the line l.
The following Theorem gives us some first picture of the configuration of
the lines on S.
Theorem 11.2. Given a line l ⊂ S, there are exactly five pairs of lines
(li , li0 ), i = 1, . . . , 5, on S which intersect l and such that
1. for all i = 1, . . . , 5, l ∪ li ∪ li0 lie in a plane;
2. for all i 6= j one has (li ∪ li0 ) ∩ (lj ∪ lj0 ) = ∅.
Proof. Let E ⊃ l be a plane in P3 . Then by Proposition 11.1 (2), E ∩ S is
one of the following:
(A) a nonsingular conic union a line;
(B) a triangle of lines;
(C) three planar lines through a point.
Now Theorem 11.2 (2) follows from Proposition 11.1 (1) once the remaining
assertions have been proven. Thus we only have to show: there are exactly
five planes Ei ⊃ l in which the intersection with S looks like (B) or (C).
After a coordinate change we can assume l = {z = w = 0} whence

f = Ax2 + Bxy + Cy 2 + Dx + Ey + F

where A, B, C ∈ k[z, w]1 , D, E ∈ k[z, w]2 , F ∈ k[z, w]3 . This defines a conic
with coefficients in k[z, w]. It is singular if and only if the discriminant

∆(z, w) = 4ACF + BDE − AE 2 − B 2 F − CD2 = 0.

In other words: every plane E through l is of the form E = {µz = λw},

(µ : λ) ∈ P1 . If µ 6= 0, then we can scale µ to be 1, and the equation for E
becomes z = λw. We can take (x : y : w) as homogeneous coordinates on E
and write

f |E = w(A(λ, 1)x2 +B(λ, 1)xy+C(λ, 1)y 2 +D(λ, 1)wx+E(λ, 1)wy+F (λ, 1)w2 ).
 93

In the same way, if λ 6= 0, then

f |E = z(A(1, µ)x2 +B(1, µ)xy+C(1, µ)y 2 +D(1, µ)wx+E(1, µ)wy+F (1, µ)w2 ).
Thus we see that E ∩ S splits as a union of three lines exactly for zeroes of ∆,
which is a homogeneous quintic; we have to prove (using the nonsingularity
of S) that ∆(z, w) has no multiple zeroes.
Suppose, after a change of coordinates, that z = 0 is a zero of ∆, and
E = {z = 0} the corresponding plane. We have to show that z 2 does not
divide ∆. Now E ∩ S consists of three lines such that (i) they do not all pass
through one point, (ii) or they do pass through one point.
Then we can choose coordinates (x : y : w) in E such that in case (i) the
lines are
l = {w = 0}, l1 = {x = 0}, l10 = {y = 0}
and in case (ii) the lines are
l = {w = 0}, l1 = {x = 0}, l10 = {x = w}.
In case (i) the equation f of S takes the form f = xyw + zg, g quadratic.
Since
f = Ax2 + Bxy + Cy 2 + Dx + Ey + F
we get B = w + az, a ∈ k, and z divides A, C, D, E, F . Hence looking back
at the equation for ∆ we find
∆ ≡ −w2 F (mod z 2 ).
But p = (0 : 0 : 0 : 1) ∈ S and the nonsingularity of S in p means that F
contains zw2 with coefficient 6= 0. Thus z 2 - F and z 2 - ∆.
In case (ii) we can write
f = x(x − w)w + zg̃
with g̃ quadratic. Then it follows that
A = w + ãz, D = −w2 + z ˜l
where ã ∈ k, ˜l is linear. Thus z | B, C, E, F , z - D. The nonsingularity of S
in (0 : 1 : 0 : 0) implies C = c̃z, c̃ 6= 0. But
∆(z, w) ≡ −CD2 (mod z 2 )
in this case, so we are done since z 2 - ∆ here as well.
94 CHAPTER 11. CUBIC SURFACES AND THEIR LINES

Corollary 11.3. There are two disjoint lines on S and S is birational to P2 .

Proof. The first assertion follows from Theorem 11.2 (2).
For the second assertion take two disjoint lines l, m ⊂ S and define

ϕ : S 99K l × m ' P1 × P1 ,
ψ : l × m ' P1 × P1 99K S

as follows: if p ∈ P3 \(l ∪ m), then there is a unique line n through p which

intersects l, m. This defines ϕ. If conversely (Q, R) ∈ l × m, n = QR ⊂ P3
the line passing through them, then Theorem 11.2 (1) implies that there are
only finitely many lines on S intersecting l, and the same holds for m, in
other words, in general, n intersects S in three points {P, Q, R}. One then
defines ψ(Q, R) = P . Then ϕ and ψ are clearly dominant rational maps
inverse to each other.
Let us now try to describe the number of lines on S and their configuration
more precisely. We need
Lemma 11.4. Let l1 , l2 , l3 , l4 ⊂ P3 be disjoint lines. Then
1. either all of the li lie on a nonsingular quadric Q and then there exist
infinitely many lines intersecting l1 , . . . , l4

2. or the li do not all lie on a quadric and there are exactly one or two
lines which intersect l1 , . . . , l4 .
Proof. It is easy to see as an exercise that l1 , l2 , l3 always lie on a nonsingular
quadric Q ⊂ P3 . In suitable coordinates, Q = {xw − yz =} ' P1 × P1 . Every
line which intersects all of l1 , l2 , l3 lies on Q. If l4 does not lie on Q, then it
has 1 or 2 intersection points with Q. The lines intersecting l1 , . . . , l4 in this
case are those of the ruling on Q to which l1 , l2 , l3 do not belong and which
pass through l4 ∩ Q.
If l4 ⊂ Q, then all of l1 , . . . , l4 belong to one ruling since they are disjoint,
and then the lines of the opposite ruling give infinitely many lines intersecting
all four.
Now let l be a line on S, and let (li , li0 ), i = 1, . . . , 5, be as in Theorem
11.2. Every other line n ⊂ S intersects exactly one of the lines li , li0 for all
i = 1, . . . , 5: namely, n intersects Ei = hli , li0 i, and Ei ∩ S = l ∪ li ∪ li0 . Because
of Proposition 11.1 (1), n cannot intersect both li and li0 .
 95

Now let l, m be disjoint lines on S. Then by the preceding m intersects

exactly one of the lines li , li0 for all i = 1, . . . , 5. Let us number the lines such
that m always intersects li , i = 1, . . . , 5. We denote the five pair of lines
intersecting m by (li , li00 ), i = 1, . . . , 5. Remark that li and lj do not intersect
for i 6= j so that in each of the pairs of lines which meet m, exactly one li
occurs.
Now Theorem 11.2 (2) applied to to m gives: for i 6= j, the line li00 does
not intersect lj . But as we just saw, every line on S intersects l, lj or lj0 . Thus
li00 intersects lj0 for i 6= j.
Thus we now have 17 distinct lines on S: li , li0 , li00 , l, m.
Lemma 11.5. 1. If n ⊂ S is a line which is distinct from these 17 lines,
then l meets exactly three out of the lines l1 , . . . , l5 .

2. Conversely, for every choice of 3 indices {i, j, k} ⊂ {1, 2, 3, 4, 5} there

is a unique line lijk ⊂ S which meets li , lj , lk .
Proof. We prove (1) first. To begin with, four disjoint lines on S do not all lie
on a quadric Q since otherwise Q ⊂ S, contradicting the irreducibility of S:
indeed, the equation of S gives a polynomial of bidegree (3, 3) on P1 ×P1 ' Q
and if this vanishes identically for four values of one of the sets of variables,
it is identically zero. Now if n intersects ≥ 4 of the li , then we can conclude
by Lemma 11.4 that n = l or n = m, a contradiction.
If, on the other hand, n intersected ≤ 2 of the li , then it would intersect
≥ 3 of the li0 since each line 6= l intersects one of li or li0 . That is, we could
assume without loss of generality that l20 , l30 , l40 , l50 or l1 , l30 , l40 , l50 . But l and l100
are then two lines which intersect the disjoint lines l20 , l30 , l40 , l50 and l1 . Thus
under our assumption that n intersects ≥ 4 of those, Lemma 11.4 implies
n = l or n = l100 , a contradiction.
To prove (2), we note that by Theorem 11.2 there are 10 lines on S
which intersect l1 only four of which we have given names to so far, namely
l, l10 , m, l100 . By part (1), each of the remaining 6 lines must intersect two
out of l2 , . . . , l5 . Moreover, every line n (not equal to l, li or lj0 ) is uniquely
determined by which of the lines li or lj0 it intersects by Lemma 11.4 (2).
Thus there are exactly 6 = 42 choices, and all of them occur.

Thus we now know all the lines on S (even by name!):

{l, m, li , li0 , li00 , lijk },

96 CHAPTER 11. CUBIC SURFACES AND THEIR LINES

in total  
5
1+1+5+5+5+ = 27.
3
We can summarize what we have seen so far in
Theorem 11.6. Every nonsingular cubic surface S ⊂ P3 contains exactly
27 lines {l, m, li , li0 , li00 , lijk }, i ∈ {1, . . . , 5}, i, j, k ∈ {1, . . . , 5} with i < j < k,
as above. The incidence graph Γ of those, whose vertices are lines and edges
denote intersections, can be described as follows:
1. l intersects l1 , . . . , l5 , l10 , . . . , l50 ;

2. l1 intersects l, m, l10 , l100 and l1jk for 6 choices of {j, k} ⊂ {2, 3, 4, 5}.

3. l10 intersects l, l1 , lj00 (for 4 choices of j 6= 1) and lijk (for 4 choices of

{i, j, k} ⊂ {2, 3, 4, 5}).

4. l100 intersects m, l1 , lj0 (for 4 choices of j 6= 1) and lijk (for 4 choices of

{2, 3, 4, 5}).

5. l123 intersects l1 , l2 , l3 , l145 , l245 , l345 , l40 , l50 , l400 , l500 .

The other intersection relations follows by symmetry.
One can treat the lines on S a bit more symmetrically, and we explain
this now.
Definition 11.7. Define a Z-module A(S) by picking as generators the 27
lines on S and relations

l0 + l00 + l000 = m0 + m00 + m000

whenever l0 , l00 , l000 and m0 , m00 , m000 are coplanar lines, i.e. they form triangles
on S.
Proposition 11.8. The module A(S) is free of rank 7, i.e., a lattice A(S) '
Z7 . As a basis we can choose l1 , . . . , l4 , l50 , l500 , l5 .
Proof. The triangles l+li +li0 , i = 1, . . . , 5, contain l, thus l+li +li0 = l+l5 +l50
for i = 1, . . . , 5, and we get

li0 = l5 + l50 − li , i = 1, . . . , 5. (11.1)

 97

Considering the triangle which contain m yields in the same way

li00 = l5 + l500 − li , i = 1, . . . , 5. (11.2)

Now li0 + lj00 + lkop is a triangle if {ij, k, o, p} is a permutation of {1, 2, 3, 4, 5},

hence

lkop = l + li − lj00 . (11.3)

But l1 + l123 + l145 is also a triangle; thus

l + l1 + l10 = l1 + l123 + l145

= l1 + 2l + l4 − l500 + l2 − l300

and thus

l = l10 − l4 + l500 − l2 + l300 = 2(l5 + l50 + l500 ) − l1 − l2 − l3 − l4 − l50 . (11.4)

Then (11.1), (11.2) and (11.3) show that the elements in the statement of
Proposition 11.8 generate. That they are independent will be shown in
Proposition 11.10 below.
Before we can finish the proof of Proposition 11.8 we need a further
definition.
Definition 11.9. Define a bilinear pairing

A(S) × A(S) → Z

in the following way:

1. for two different lines l, l0 we put ll0 = 0 or ll0 = 1 depending on whether
they are disjoint or intersect;

2. for all lines l we put l2 = −1;

3. we put l(m+m0 +m00 ) = 1 for every line l and every triangle m+m0 +m00 .
This is well-defined by (1) and (2) alone: A(S) is generated by lines,
and we know that if l is distinct from m, m0 , m00 , it intersects exactly one of
m, m0 , m00 , hence (3) follows; if on the other hand l = m, say, then lm0 =
lm00 = 1 and l2 = lm = −1 by (2), hence (3) follows as well. So the bilinear
pairing descends to the quotient and is well-defined.
98 CHAPTER 11. CUBIC SURFACES AND THEIR LINES

Proposition 11.10. If

e0 = l5 + l50 + l500 , e1 = l1 , e2 = l2 , e3 = l3 , e4 = l4 ,
e5 = l50 , e6 = l500

then
e20 = 1, e2i = −1, i = 1, . . . , 6, ei ej = 0 i 6= j.
In particular, e0 , . . . , e6 are a basis of A(S).
Proof. Remark that e1 = l1 , . . . , e4 = l4 , e5 = l50 , e6 = l500 are six disjoint lines,
and the first four are also disjoint from l5 ; this implies everything except
e20 = 1. To see this remark that

e0 e5 = e0 l50 = (l5 + l50 + l500 )l50 = 1 − 1 + 0 = 0

and similarly e0 e6 = e0 l500 = 0. Thus

e20 = e0 (l5 + l50 + l500 ) = e0 l5

= (l5 + l50 + l500 )l5 = −1 + 1 + 1 = 1.

Proposition 11.11. If h is the class of a triangle in A(S), then h2 = 3 and

hx ≡ x2 (mod 2) for all x ∈ A(S)
Proof. If h = l + l0 + l00 is a triangle, then h2 = h(l + l0 + l00 ) = 1 + 1 + 1 = 3.
Now for all lattices the function A → F2 , x 7→ x2 (mod 2) is linear because
(x + y)2 ≡ x2 + y 2 (mod 2). Thus it suffices to prove the second assertion for
generators of A(S). But A(S) is generated by lines l, and for those hl = 1,
l2 = −1.
To summarize: S has an associated lattice A(S) ' Z7 with a nondegen-
erate bilinear pairing which we can diagonalize (in the basis e0 , . . . , e6 ) to
(1, −1, . . . , −1). Moreover there is a class h ∈ A(S) with h2 = 3 and hx ≡ x2
(mod 2) for all x ∈ A(S). Indeed, one can show that these conditions char-
acterize the pair A, h uniquely up to isomorphism.
One also calls A(S) the Picard group of S and the bilinear pairing the
intersection pairing for reasons that will become clearer in the next Chapter.
The advantage of A(S) is that one can show that the lines on S are exactly
the solutions of the equations hl = 1, l2 = −1 in A(S). This is a bit messy
 99

to do directly by hand; it is easy to see once one knows adjunction and the
genus formula. This gives a symmetric description of the lines.
We also mention without proof that there is a striking connection to the
root system of type E6 : namely, h⊥ ⊂ A(S) is the (negative of the) root
lattice, (−E6 ).
100 CHAPTER 11. CUBIC SURFACES AND THEIR LINES
Chapter 12

Local parameters, power series

methods, divisors

Let X be a (not necessarily irreducible) variety, x ∈ X a regular point,

dimx X = n.
Definition 12.1. Elements p1 , . . . , pn ∈ OX,x are called local parameters (or
local coordinates) in x if pi ∈ mx ⊂ OX,x and the residue classes p̄1 , . . . , p̄n
∗
form a basis of the k-vector space mx /m2x = TX,x . An equivalent condition is
that the differentials dx p1 , . . . , dx pn are independent linear forms on Tx X.
Choose U 3 open affine such that p1 , . . . , pn ∈ OU (U ). Let Hi := {u ∈
U | pi (u) = 0} and let I(Hi ) ⊂ OU (U ) its ideal. Assume that U ⊂ AN closed
and let Pi (t1 , . . . , tN ) be a polynomial in the coordinates on AN such that
Pi |U = pi . Then (Pi , I(U )) ⊂ I(Hi ), thus

Tx Hi ⊂ {v ∈ Tx U | dx Pi (v) = 0}.

Since the dx p1 , . . . , dx pn are linearly independent, we get that dim Tx Hi ≤

n − 1. But because of Theorem 8.11 and Theorem 10.13 we have

dim Tx Hi ≥ dimx Hi ≥ n − 1,

hence dim Tx Hi = n − 1 and x is regular on Hi . In a neighborhood of x the

intersection of the Hi consists of x alone, for otherwise dx p1 = · · · = dx pn = 0
wouldn’t have 0 as its only solution (but would vanish on the tangent space
of a component of the intersection of the Hi ’s passing through x). Thus we
obtain

101
102CHAPTER 12. LOCAL PARAMETERS, POWER SERIES METHODS, DIVISORS

Theorem 12.2. Let p1 , . . . , pn be local parameters

T around x. Then x is a
regular point on every Hi = {pi = 0} and i Tx Hi = {0}

One also says that the Hi ’s intersect transversally at x.

Theorem 12.3. Local parameters generate the maximal ideal mx ⊂ OX,x .

Proof. The module M = mx over OX,x is finite. Now the residue classes of
the pi in mx /m2x generate that vector space, hence the submodule M 0 ⊂ M
generated by the pi ’s satisfies M 0 + mx M = M , in other words

m · (M/M 0 ) = (M/M 0 ).

By Nakayama’s Lemma 7.6 we have M 0 = M .

We will now consider power series expansions, also called Taylor expan-
sions of elements f ∈ OX,x , given local parameters p1 , . . . , pn . AMong other
things, this will allow us to understand regular and singular points on X bet-
ter, in particular prove that the intersection of two irreducible components
must consist entirely of singular points.
Put
f (x) =: x0 .
Then
f1 := f − x0 ∈ mx .
Since p̄1 , . . . , p̄n generate mx , there are elements c1 , . . . , cn ∈ k such that
n
X n
X
f2 := f1 − ci p i = f − x 0 − ci pi ∈ m2x .
i=1 i=1

Now f2 ∈ m2x , thus we can write f2 = j gj hj with gj , hj ∈ mx , thus there

P
2 2
P P
exist elements dij , eij ∈ k such
P that gj − i d
P P ij p i ∈ mx , h
Pj − i eij pi ∈ mx .
Defining elements flk in k by lk flk pl pk := j ( i dij pi )( i eij pi ) we obtain
X X X
f3 = f2 − flk pl pk = f − x0 − ci p i − flk pl pk ∈ m3x .
i l,k

Continuing in this way we inductively obtain polynomials

Fi , i = 0, 1, 2, . . . , Fi ∈ k[t1 , . . . , tn ]i
 103

with

k
X
f− Fi (p1 , . . . , pn ) ∈ mk+1
x , ∀ k ≥ 0.
i=0

Definition 12.4. The ring of formal powers series in variables t1 , . . . , tn ,

denoted by k[[t]] = k[[t1 , . . . , tn ]], is the ring whose elements are formal sums

F = F0 + F1 + F2 + . . .

(in other words, simply sequences (F0 , F1 , F2 , . . . )) where Fi is a homogeneous

polynomial of degree i in the t1 , . . . , tn ; and addition is defined component-
wise:
F + G := (F0 + G0 ) + (F1 + G1 ) + . . .
and multiplication
F · G := H0 + H1 + H2 + . . .
is defined by formally imitating the Cauchy product rule:
X
Hq := Fi · Gj .
i+j=q

The leading term of a powers series F , lt(F ) is Fi0 where i0 = min{i | Fi 6= 0}.

It then follows that lt(F · G) = lt(F ) · lt(G), whence k[[t]] is an integral

domain.

Definition 12.5. An element F ∈ k[[t]] is called a Taylor series for f ∈ OX,x

(with respect to local parameters p1 , . . . , pn if

k
X
f− Fi (p1 , . . . , pn ) ∈ mk+1
x , ∀ k ≥ 0.
i=0

We have seen above that every f ∈ OX,x has at least one Taylor series.

Theorem 12.6. If x ∈ X is a regular point and p1 , . . . , pn are local param-

eters around x, then every f ∈ OX,x has a unique Taylor series.
104CHAPTER 12. LOCAL PARAMETERS, POWER SERIES METHODS, DIVISORS

Proof. It suffices to show that if f = 0 in OX,x , then F = 0 in k[[t]] for

every Taylor series F for f ; in other words, it suffices to show that for Fk ∈
k[t1 , . . . , tn ]k
Fk (p1 , . . . , pn ) ∈ mk+1
x =⇒ Fk (t1 , . . . , tn ) = 0 in k[t1 , . . . , tn ]k .
Suppose by contradiction that Fk (t1 , . . . , tn ) 6= 0. The coefficient of tkn in Fk
is Fk (0, . . . , 0, 1) and there is a (c1 , . . . , cn ) ∈ k n such that Fk (c1 , . . . , cn ) 6= 0.
Thus after a linear change of coordinates with (c1 , . . . , cn ) 7→ (0, . . . , 0, 1),
we can assume that the coefficient of tkn is nonzero. Thus without loss of
generality
Fk (t1 , . . . , tn ) = αtkn + G1 (t1 , . . . , tn−1 )tkn + · · · + Gk (t1 , . . . , tn−1 )
with α 6= 0 and Gi ∈ k[t1 , . . . , tn−1 ]i . Since the p1 , . . . , pn generate mx by
Theorem 12.3 we can write Fk (p1 , . . . , pn ) ∈ mk+1x as a homogeneous poly-
nomial of degree k in p1 , . . . , pn with coefficients in mx whence we get an
equality
αpkn + Gn (p1 , . . . , pn−1 )pk−1
n + · · · + Gk (p1 , . . . , pn−1 )
= mpkn + G̃1 (p1 , . . . , pn−1 )pk−1
n + · · · + G̃k (p1 , . . . , pn−1 )
with m ∈ mx and G̃i a homogeneous polynomial of degree i in n − 1 variables
with coefficients in mx . Then (α − m)pkn ∈ (p1 , . . . , pn−1 ) and α − m is
invertible in OX,x whence pkn ∈ (p1 , . . . , pn−1 ). But this means, putting Hi =
(pi = 0), that
Hn ⊃ H1 ∩ · · · ∩ Hn−1
and thus
Tx Hn ⊃ Tx H1 ∩ · · · ∩ Tx Hn−1 .
Then
Tx H1 ∩ · · · ∩ Tx Hn 6= {0}
contradicting Theorem 12.2.
Thus if x ∈ X is regular, then we get a well-defined map
T : OX,x → k[[t1 , . . . , tn ]]
by associating to an f its unique Taylor series. Moreover, this is clearly a
k
homomorphism. We T wantk to determine its kernel: T (f ) = 0 means f ∈ mx
∀ k ≥ 0, i.e. f ∈ k≥0 mx . We then use the following commutative algebra
result.
 105

Theorem 12.7 (Krull’s Intersection Theorem). Let A be a Noetherian local

ring with maximal ideal m. Then
\
mk = (0).
k≥0

The proof is not overly complicated, but we omit it to continue with

the geometric consequences. Notice that Krull’s Intersection Theorem can
become false without the Noetherian hypothesis: for example it does not
hold for germs of infinitely differentiable functions around 0 ∈ R1 since there
are functions all of whose derivatives in 0 vanish, but the germ is nonzero.
Theorem 12.8. If x ∈ X is a regular point on a variety, then we have an
inclusion
OX,x ,→ k[[t1 , . . . , tn ]]
defined by associating to a germ its Taylor series with respect to some set
p1 , . . . , pn of local parameters.
As a consequence we can now prove a result that complements Theorem
10.13.
Theorem 12.9. Let x ∈ X be a regular point on a variety X. Then x lies
on a unique irreducible component of X. In other words, intersections of two
irreducible components consist entirely of singular points.
Proof. Replacing X by some affine open subset containing x we can assume
that X is affine and all irreducible components of X pass through x. Then
OX (X) ⊂ OX,x ⊂ k[[t1 , . . . , tn ]]. Since k[[t1 , . . . , tn ]] is an integral domain,
OX (X) cannot have any zero divisors, hence X is irreducible.

We will assume the following result without proof to proceed further.

Theorem 12.10. Let x ∈ X be a regular point on a variety X. Then the
local ring OX,x is factorial.
A proof can be found in almost any textbook on commutative algebra.
One approach is to use power series methods and the Weierstrass Preparation
Theorem to prove that k[[t1 , . . . , tn ]] is a UFD, and then show that the UFD
property is inherited from OX,x ⊂ k[[t1 , . . . , tn ]]. This would be perfectly
feasible with our current background, but is a bit lengthy and messy, so we
omit it.
We will assume that varieties X are irreducible in the sequel.
106CHAPTER 12. LOCAL PARAMETERS, POWER SERIES METHODS, DIVISORS

Definition 12.11. Germs of functions f1 , . . . , fm ∈ OX,x are called local

equations for a subvariety Y ⊂ X if there is an affine open neighborhood
U 3 x with fi ∈ OX (U ) for all i and I(Y ∩ U ) = (f1 , . . . , fm ) ⊂ OX (U ).

Let IY,x ⊂ OX,x be the ideal of germs of functions in OX,x which vanish
on Y in an open neighborhood of x; if X is affine we have
n u o
IY,x = f = | u, v ∈ OX (X), u ∈ I(Y ), v(x) 6= 0 .
v
Theorem 12.12. The elements f1 , . . . , fm ∈ OX,x are called local equations
for a subvariety Y ⊂ X ⇐⇒ IY,x = (f1 , . . . , fm ).

Proof. Clearly =⇒ holds since if in an affine open neighborhood U of x we

have I(Y ∩ U ) = (f1 , . . . , fm ), then IY,x = (f1 , . . . , fm ) by what we remarked
immediately before the statement of the Theorem.
For ⇐= assume IY,x = (f1 , . . . , fm ), and write I(Y ∩ U ) = (g1 , . . . , gs ),
gi ∈ OU (U ) for some affine open U containing x. Since gi ∈ IY,x we have
equations
X m
gi = hij fj , hij ∈ OX,x , i = 1, . . . , s.
j=1

All fi , hij are regular on some small principal open subset V 3 x, V =

U − {g = 0}, g ∈ OU (U ). Thus in OV (V ) we have

(g1 , . . . , gs ) = I(Y ∩ U ) · OV (V ) ⊂ (f1 , . . . , fm ).

Let us show that I(Y ∩ U ) · OV (V ) = I(V ∩ Y ). This will imply I(V ∩ Y ) ⊂

(f1 , . . . , fm ), hence the Theorem since fi ∈ I(V ∩ Y ) so that the reverse
inclusion is obvious.
Now it is clear that I(Y ∩ U ) · OV (V ) ⊂ I(V ∩ Y ) Now let v = u/g l ∈
I(V ∩Y ), u ∈ OU (U ), so u = vg l . Then u ∈ I(Y ∩U ) and since 1/g l ∈ OV (V )
it follows that v ∈ I(Y ∩ U ) · OV (V ).

Theorem 12.13. An irreducible subvariety Y ⊂ X of codimension 1 has a

local equation around any nonsingular point x of X.

Proof. We can assume that X is affine. Let f ∈ OX,x be a germ with

f ∈ IY,x . Note that OX,x is factorial by Theorem 12.10. Decompose f into
prime factors in OX,x . Since Y is irreducible, one of the prime factors g must
vanish on Y . We show that IY,x = (g). After shrinking X we can assume
 107

that g is regular on X. Then V (g) ⊃ Y , V (g) = Y ∪ Y 0 . If x ∈ Y 0 , then

there exist regular functions h, h0 ∈ OX (X) with hh0 = 0 on (g = 0), but h
and h0 are both not identically zero on (g = 0). By the Nullstellensatz 3.2,
g divides (hh0 )r for some r where the divisibility holds in OX (X); thus also
g | (hh0 )r in OX,x . Now since OX,x is factorial, we get g | h or g | h0 since g
is prime. Hence h or h0 is zero on V (g) in a neighborhood of x. Shrinking X
we can assume from the beginning that all irreducible components of V (g)
pass through x. Thus we have a contradiction because we obtained that h
or h0 is identically zero on V (g) but assumed the contrary at the beginning.
Thus V (g) = Y . If u ∈ OX (X) vanishes on Y , then the Nullstellensatz
implies that g | us , some s, in OX (X). Hence g | us in OX,x and since the
latter is factorial, g | u. Thus IY,x = (g).
Recall that a rational map f : X 99K Y is an equivalence class of mor-
phisms fU : U → Y , U ⊂ X Zariski open dense; two being considered equiv-
alent if they coincide when they are both defined. It follows that there is
a largest open set dom(f ) ⊂ X on which f is a morphism, namely the
union of all open sets U for representatives (fU , U ) of f . The complement
Zf = X\dom(f ) is called the indeterminacy locus of f . It is a closed subset
of X.

Theorem 12.14. Let X be a nonsingular variety, and f : X 99K Pn a ratio-

nal map. Then the indeterminacy locus Zf ⊂ X has codimension ≥ 2.

Proof. It suffices to prove the assertion in an affine neighborhood of a non-

singular point x ∈ X. We can then find a representative of f in the form

f = (f0 : · · · : fn )

where fi ∈ k(X), fi ∈ OX,x for all i (we can clear denominators because the
target is a projective space) and the fi without common factor in OX,x (we
can divide by any common factor since the target is projective). Then no
irreducible codimension 1 subvariety Y can be contained in f0 = · · · = fn = 0
since by Theorem 12.13 we would have IY,x = (g) and if fi vanishes on (g = 0),
then g | fi for all i, a contradiction.
The following are immediate consequences of Theorem 12.14.

Corollary 12.15. Every rational map f : C 99K Pn from a nonsingular curve

C to Pn is a morphism.
108CHAPTER 12. LOCAL PARAMETERS, POWER SERIES METHODS, DIVISORS

Corollary 12.16. Two birationally equivalent nonsingular projective curves

are isomorphic.
We proceed to discuss divisors of zeroes and poles of rational functions.
Let X be an irreducible nonsingular projective variety, f ∈ k(X) a rational
function, Y ⊂ X an irreducible subvariety of codimension 1. We want to
define the order of zero or pole of f along Y . Let U ⊂ X be affine open,
U ∩ Y 6= ∅ such that I(Y ∩ U ) = (π) with π ∈ OX (U ), which is possible by
Theorem 12.13.
For g ∈ OX (U ), g 6=T 0, there is a k ≥ 0 with g ∈ (π k ),Tbut g ∈/ (π k+1 );
namely, if we had g ∈ k≥0 (π k ), then we would have g ∈ k≥0 mk where m
is the maximal ideal of the local ring of Y ∩ U , i.e. the localization OU (U )p
where p = I(Y ∩ U ). But then Krull’s intersection theorem would imply
g = 0 in the local ring of Y ∩ U , hence g would be identically zero on U .
Now we put k := vY (g). This is independent of the choice of an affine U
above: if V ⊂ U is affine open, V ∩ Y 6= ∅, then π gives a local equation of
Y ∩ V and vYU (g) = vYV (g). In general, if U, V are open affine with U ∩ Y 6= ∅,
V ∩ Y 6= ∅, then there is an affine open W ⊂ U ∩ V with W ∩ Y 6= ∅.
We call vY (g) the valuation of g in Y . We have

vY (g1 g2 ) = vY (g1 ) + vY (g2 ), (12.1)

vY (g1 + g2 ) ≥ min{vY (g1 ), vY (g2 )} if g1 + g2 6= 0. (12.2)

If f ∈ k(X), we can write

h1
f= , h1 , h2 ∈ OU (U ).
h2
Now (12.1) and (12.2) imply that if f 6= 0, then

vY (f ) := vY (h1 ) − vY (h2 )

is independent of the choice of representative h1 /h2 and (12.1), (12.2) remain

valid for all g1 , g2 ∈ k(X)\{0}.
If vY (f ) = k > 0, then we say f has a zero of order k along Y . If
vY (f ) = k < 0, then we say that f has a pole of order −k along Y .
There are only finitely many Y ’s with vY (f ) 6= 0: indeed, if U is affine,
f ∈ OU (U ), then vY (f ) = 0 if Y is not component of (f = 0). If f = g1 /g2 ,
then vY (f ) = 0 if Y is not a component of (g1 = 0) or (g2 = 0). In general,
X is a union of finitely many such Ui .
 109

Definition 12.17. For smooth irreducible projective X let Div(X) the free
abelian group on all codimension 1 irreducible subvarieties. We call Div(X)
the group of (Weil) divisors on X. Any element in it is called a (Weil)
divisor. For f ∈ k(X)∗ we put
X
div(f ) := vY (f )Y ∈ Div(X)
Y ⊂X

where the sum runs over all irreducible codimension 1 subvarieties. A divisor
of the form div(f ) (or the zero divisor 0) is called a principal divisor. The
principal divisors form a subgroup PrincDiv(X) ⊂ Div(X). The quotient

Pic(X) := Div(X)/PrincDiv(X)

is called the Picard group (or more accurately, Weil divisor class group) of
X.

The group Pic(X) is an important invariant of X.

Example 12.18. We have Pic(Pn ) ' Z. Every irreducible codimension

1 subvariety is defined by an irreducible homogeneous polynomial of some
degree k by Theorem 8.6. We have for Ui = {Xi 6= 0} ⊂ Pn that I(Y ∩ U ) =
(F/Xik ). If
F
f= ∈ k(Pn )∗
G
Q ki
is a rational
Q m function, F , G homogeneous of the same degree k, F = i Hi ,
G = j Lj j with Hi and Lj irreducible, then
X X
div(f ) = ki (Hi = 0) − mj (Lj = 0)
i j
P
which has degree 0 as an n − 1 cycle. If conversely, D = ki YiQwith
Hiki
P
ki deg Yi = 0, and Yi is defined by an irreducible Hi , then f =
is an element in k(X) with div(f ) = D.

Example 12.19. Using the Segre embedding, one can show that

Pic(Pn × Pm ) = Z ⊕ Z.

The proof is similar to the one in Example 12.18.

110CHAPTER 12. LOCAL PARAMETERS, POWER SERIES METHODS, DIVISORS

By these simple examples one should not be tempted to think that Picard
groups are always finitely generated or torsion-free: this fails in very simple
examples already, like that of plane cubic curves.
Instead of with codimension 1 subvarieties, leading to Pic(X), one can
work with higher-codimensional subvarieties as well, leading to the so-called
Chow groups of X. But these are generally much harder to compute.
We also point out that the power series methods we have started to de-
velop in this Chapter have much more far reaching consequences and develop-
ments, in particular in Zariski’s theory of formal functions and deformation
theory.
Bibliography

[AM69] Atiyah M.& Macdonald I. G., Introduction to commutative algebra,

Addison-Wesley, Reading MA (1969)

[Harr92] Harris, J., Algebraic Geometry, A First Course, Graduate Texts

in Mathematics 133, Springer-Verlag (1992)

[Mum95] Mumford, D., Algebraic Geometry I: Complex Projective Varieties,

Classics in Mathematics, reprint of the 1st ed. (1976); Springer-
Verlag (1995)

[Reid10] Reid, M., Undergraduate Algebraic Geometry, London Math. Soc.

Student Texts 12, Cambridge University Press (2010)

[Shaf94] Shafarevich, I.R., Basic Algebraic Geometry 1, second edition,

Springer-Verlag (1994)

[ZS60] Zariski, O. & Samuel, P., Commutative algebra, Vol. II, Van Nos-
trand, New York (1960)

111