Notes on basic algebraic geometry

June 16, 2008

These are my notes for an introductory course in algebraic geometry. I have
trodden lightly through the theory and concentrated more on examples. Some
examples are handled on the computer using Macaulay2, although I use this as
only a tool and wont really dwell on the computational issues.
Of course, any serious student of the subject should go on to learn about
schemes and cohomology, and (at least from my point of view) some of the
analytic theory as well. Hartshorne [Ht] has become the canonical introduction
to the rst topic, and Griths-Harris [GH] the second.
1
Contents
1 Ane Geometry 3
1.1 Algebraic sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Weak Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Zariski topology . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 The Cayley-Hamilton theorem . . . . . . . . . . . . . . . . . . . 9
1.5 Ane Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.6 Hilberts Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Nilpotent matrices . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2 Projective Geometry 15
2.1 Projective space . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Projective varieties . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Projective closure . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.4 Miscellaneous examples . . . . . . . . . . . . . . . . . . . . . . . 18
2.5 Grassmanians . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 Elimination theory . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.7 Simultaneous eigenvectors . . . . . . . . . . . . . . . . . . . . . . 23
3 The category of varieties 26
3.1 Rational functions . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Quasi-projective varieties . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4 Dimension theory 30
4.1 Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2 Dimension of bres . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.3 Simultaneous eigenvectors (continued) . . . . . . . . . . . . . . . 32
5 Dierential calculus 34
5.1 Tangent spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5.2 Singular points . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.3 Singularities of nilpotent matrices . . . . . . . . . . . . . . . . . . 36
5.4 Bertini-Sard theorem . . . . . . . . . . . . . . . . . . . . . . . . . 38
2
Chapter 1
Ane Geometry
1.1 Algebraic sets
Let k be a eld. We write A
n
k
= k
n
, and call this n dimensional ane space
over k. Let
k[x
1
, . . . x
n
] =

c
i1...in
x
i1
1
. . . x
in
n
[ c
i1...in
k
be the polynomial ring. Given a = (a
i
) A
n
, we can substitute x
i
by a
i
k in
f to obtain an element denoted by f(a) or ev
a
(f), depending on our mood. A
polynomial f gives a function ev(f) : A
n
k
k dened by a ev
a
(f).
Given f k[x
1
, . . . x
n
], dene it zero set by
V (f) = a A
n
k
[ f(a) = 0
At this point, we are going to need to assume something about our eld. The
following is easy to prove by induction on the number of variables. We leave
this as an exercise.
Lemma 1.1.1. If k is algebraically closed and f nonconstant, then V (f) is
nonempty.
If S k[x
1
, . . . x
n
], then let V (S) be the set of common zeros
V (S) =

fS
V (f)
A set of this form is called algebraic. I want to convince you that algebraic sets
abound in nature.
Example 1.1.2. The Fermat curve of degree d is V (x
d
1
+x
d
2
1) A
2
. More
generally, a Fermat hypersurface is given by V (x
d
1
+x
d
2
+. . . x
d
n
1).
Example 1.1.3. Let us identify A
n
2
k
with the set Mat
nn
(k) of nn matrices.
The set of singular matrices is algebraic since it is dened by the vanishing of
the determinant det which is a polynomial.
3
Example 1.1.4. Then the set SL
n
(k) A
n
2
of matrices with determinant 1
is algebraic since its just V (det 1).
The set of nonsingular matrices GL
n
(k) is not an algebraic subset of Mat
nn
(k).
However, there is useful trick for identifying it with an algebraic subset of
A
n
2
+1
= A
n
2
A
1
.
Example 1.1.5. The image of GL
n
(k) under the map A (A, 1/ det(A))
identies it with the algebraic set
(A, a) A
n
2
+1
[ det(A)a = 1
Example 1.1.6. Identify A
mn
k
with the set of m n matrices Mat
mn
(k).
Then the set of matrices of rank r is algebraic. This is because it is dened
by the vanishing of the (r + 1) (r + 1) minors, and these are polynomials in
the entries. Notice that the set of matrices with rank equal r is not algebraic.
Example 1.1.7. The set of pairs (A, v) Mat
nn
(k) k
n
such that v is an
eigenvector of A is algebraic, since the condition is equivalent to rank(A, v) 2.
Example 1.1.8. Let N
i
A
n
2
k
be the set of matrices which are nilpotent of
order i, i.e matrices A such that A
i
= 0. These are algebraic.
Before doing the next example, let me remind you about resultants. Given
two polynomials
f = a
n
x
n
+. . . a
0
and
g = b
m
x
m
+. . . b
0
Suppose, we wanted to test whether they had a common zero, say . Then
multiplying f() = g() = 0 by powers of yields
a
n

n
+ a
n1

n1
+. . . a
0
= 0
. . .
a
n

n+m
+ a
n1

n+m1
+. . . = 0
b
m

m
+. . . b
0
= 0
. . .
b
m

n+m
+. . . = 0
We can treat this as a matrix equation, with unknown vector (
n+m
,
n+m1
, . . . , 1)
T
.
For the a solution to exist, we would need the determinant of the coecient ma-
trix, called the resultant of f and g, to be zero. The converse, is also true (for
k =

k) and can be found in most standard algebra texts. Thus:
Example 1.1.9. Identify the set of pairs (f, g) with A
(n+1)+(m+1)
. The set of
pairs with common zeros is algebraic.
4
We can use this to test whether a monic polynomial (i.e. a polynomial with
leading coecient 1) f has repeated root, by computing the resultant of f and
its derivative f

. This called the discriminant of f. Alternatively, if we write

f(x) =

(x r
i
), the discriminant disc(f) =

i<j
(r
i
r
j
)
2
. This can be
written as a polynomial in the coecients of f by the theorem on elementary
symmetric functions.
Example 1.1.10. The set of monic polynomials with repeated roots is a an
algebraic subset of A
n
.
We call a map F : A
n
A
m
a morphism if it is given by
F(a) = (f
1
(a), . . . f
m
(a))
for polynomials f
i
k[x
1
, . . . x
n
]. Clearly the preimage under a regular map of
an algebraic set is algebraic. Let us identify A
n
2
with the set of n n matrices
once again. To every matrix, A we can associate its characteristic polynomial
det(tI A). We thus get a morphism ch : A
n
2
A
n
given by taking the
coecients of this polynomial other than the leading coecient which is just
one. Therefore
Example 1.1.11. The set of matrices in A
n
2
with repeated eigenvalues is an
algebraic set. More explicitly it is the zero set of the discriminant of the char-
acteristic polynomial.
Exercise 1.1.12.
1. Identify A
6
= (A
2
)
3
with the set of triples of points in the plane. Which
of the following is algebraic:
a) The set of triples of distinct points.
b) The set of triples (p
1
, p
2
, p
3
) with p
3
= p
1
+p
2
.
c) The set of triples of colinear points.
2. Check that V (S) = V (S)), where
S) =

h
i
f
i
[ h
i
k[x
1
, . . . x
n
], f
i
S
is the ideal generated by S. Therefore by the Hilbert basis theorem, which
says that k[x
1
, . . . x
n
] is Noetherian, we nd that any algebraic set is de-
ned by a nite number of polynomials.
1.2 Weak Nullstellensatz
Recall that the a (commutative) k-algebra is a commutative ring R with a ring
homomorphism k R. For example, k[x
1
, . . . x
n
] is a k-algebra. A homomor-
5
phism of k-algebras is a ring homomorphism R S such that
k

?
?
?
?
?
?
?
?
R
//
S
commutes. Lets make a simple observation at this point:
Lemma 1.2.1. The map f ev(f) is a homomorphism of k-algebras from
k[x
1
, . . . x
n
] to the algebra of k-valued functions on A
n
.
Exercise 1.2.2. Show that this homomorphism is injective if k is innite, but
not in general.(In view of this, we will eventually stop distinguishing between f
and ev(f) when k is innite.)
Lets suppose that S is explicity given to us as a nite set of polynomials.
We can now ask is there an algorithm to decide when V (S) this nonempty?
Here are some answers:
1. Algebraically closed elds: Yes by Hilberts Nullstellensatz (see below).
2. Finite elds: Yes, since there are only a nite number of points to check.
3. R: Yes, by Tarski.
4. Q: Unknown! (However, Matjisevich proved that there is no algorithm
for Z, or equivalently Hilberts 10th problem has a negative solution. So
its reasonable to expect that it would be no for Q as well.)
Theorem 1.2.3 (Weak Hilbert Nullstellensatz). If k is algebraically closed,
then V (S) = i there exists f
1
. . . f
N
S and g
1
. . . g
N
k[x
1
, . . . x
n
] such
that

f
i
g
i
= 1
The German word nullstellensatz could be translated as zero set theorem.
The Weak Nullstellensatz can be rephrased as V (S) = i S) = (1). Since
this result is central to much of what follows, we will assume that k is alge-
braically closed from now on unless stated otherwise. To get an algorithm
as claimed above, we need an eective form of the nullstellensatz:
Theorem 1.2.4 (Hermann). If (f
1
, . . . f
N
) = (1), then there exists g
i
, with
degree bounded by a computable constant depending on maxdeg f
i
, such that

f
i
g
i
= 1.
Dene the ring
R = k[x
1
, . . . x
n
]/S)
Lemma 1.2.5. (k any eld.) ev
a
: k[x
1
, . . . x
n
] k factors throught the canon-
ical map k[x
1
, . . . x
n
] R i a V (S).
6
Proof. ev
a
factors through R i ev
a
(S)) = 0 i f(a) = 0, f S) i a
V (S).
In view of the lemma, we can view ev
a
as a homomorphism of R k when
a V (S).
Corollary 1.2.6. The map a ev
a
gives a bijection
V (S)

= Hom
kalgebras
(R, k)
Proof. Given h Hom(R, k), let a(h) be the vector whose ith component is
ev
a
( x
i
), where x
i
= image(x
i
) in R. Then h a(h) gives the inverse.
We are now, almost already to prove WN, we need the following which is
really an algebraic form of the Nullstellensatz:
Theorem 1.2.7 (Algebraic Nullstellensatz). Let k
1
k
2
be a eld extension,
such that k
2
is nitely generated as a k
1
-algebra, then k
2
is a nite eld extension
of k
1
.
Proof. See Atiyah-MacDonald [AM, prop. 7.9 ].
Proof of W. Nullstellensatz. If

f
i
g
i
= 1 for f
i
S, then clearly these poly-
nomials have no common zeros.
Conversely, suppose that S) is a proper ideal. Therefore R = k[x
1
, . . . x
n
]/S)
is nonzero, so it has a maximal ideal m. R/m is a eld containing k which is
nitely generated as a k-algebra. By the previous, theoremR/mis a nite, hence
algebraic, extension of k. Thus k = R/m. The homomorphism, R R/m = k
corresponds to by above, to a point of V (S).
1.3 Zariski topology
One nice feature of working over R or C is that ane carries natural topology
dened by the Euclidean metric. It turns out that one can dene a topology
over any eld which is sometimes just as good. Given a subset X A
n
k
, let
1(X) = f k[x
1
, . . . x
n
] [ f(a), a X
This is an ideal of k[x
1
, . . . x
n
]. We now have two operations
subsets of A
n

I
//
subsets of k[x
1
, . . . x
n
]
V
oo
which we need to compare.
Proposition 1.3.1. 1. X Y 1(X) 1(Y ).
2. S T V (S) V (T).
3. V (1(X)) X.
7
4. 1(V (S)) S.
5. If I and J are ideals, V (I J) = V (I) V (J).
6. If I
a
is a family of ideals, then
V (

I
a
) =

V (I
a
)
Proof. We prove (5) assuming the previous parts. We have I J I, J which
implies V (IJ) V (I), V (J). Therefore V (IJ) V (I)V (J). If a V (IJ)
is not contained in the union, then f(a) ,= 0 ,= g(a). Therefore fg(a) ,= 0 which
contradicts the fact that fg I J.
Corollary 1.3.2. V (1(V (S))) = V (S).
Proof. follows by (3). (4) gives S 1(V (S)) which implies the opposite
inclusion.
Corollary 1.3.3. The complements of algebraic sets forms a topology on A
n
called the Zariski topology. In other words the algebraic sets are the precisely
the closed sets for this topology.
Proof. A collection of subsets is a topology if it is closed under arbitrary unions,
nite intersections, and contains and A
n
. The collection of complements of
algebraic sets satises all of the above.
While its nice to have a topology, you have to be careful about your intuition.
Its much coarser than the ones you may encounter in an analysis class. For
example, the nonempty open sets of A
1
are the complements of nte sets. This
is often called the conite topology.
A function f : A
n
A
1
is called regular if it is a morphism, i.e. if it is
dened by a polynomial.
Lemma 1.3.4. All regular functions f : A
n
k
A
1
k
are continuous with respect to
the Zariski topology on A
n
k
and the conite topology on A
1
k
. This is the weakest
topology with this property.
Proof. Continuity means that the preimage of open sets are open, or equivalently
the preimage of closed sets are closed. The preimage of a
1
, . . . a
N
A
1
k
is
V (

(f a
i
)) which is Zariski closed by denition.
Given any other topology with this property, V (f) would be closed for it.
Therefore V (S) is closed in this other topology for any S.
More generally, we see that morphisms F : A
n
A
m
are continuous when
both spaces are given their Zariski topologies.
When k = C, we have two choices of topologies, the Zariski topology and
the Euclidean metric space topology that we will call the classical or strong
topology. The strong topology on A
n
C
is Hausdor, noncompact, and coincides
with the product topology A
1
C
A
1
C
. . . A
1
C
. All of these fail for Zariski.
8
Exercise 1.3.5.
1. Finish the proof of proposition 1.3.1.
2. Let D(f) = A
n
V (f). Show that any open set is a union of D(f)s, in
other words, these sets form a basis for the topology.
3. Show that A
n
is not Hausdor if k is innite.
4. Prove that any nonempty open subset of A
n
is dense.
5. Prove that A
n
is compact.
6. Show that the Zariski topology of A
2
is ner than the product topology of
A
1
A
1
.
1.4 The Cayley-Hamilton theorem
Given a square matrix A, recall that its characteristic polynomial is det(tIA)
k[t]. Its roots are precisely the eigenvalues of A.
Theorem 1.4.1 (Cayley-Hamilton). If p
A
(t) = det(tI A), then p
A
(A) = 0.
We give a proof which makes use of the Zariski topology.
Lemma 1.4.2. The subset of matrices D
n
in Mat
nn
with distinct eigenvalues
is dense in the Zariski topology.
Proof. The set of these matrices is certainly nonempty and open since comple-
ment is an algebraic set by example 1.1.11. As observed in the exercises of the
last section, nonempty open sets are dense.
We recall the following standard fact from linear algebra.
Theorem 1.4.3. Any matrix A D
n
can be diagonalized, that this there exist
T Gl
n
(k) such that T
1
AT is diagonal.
Proof. Take D to be the diagonal matrix of eigenvalues
i
of A and T a matrix
with eigenvectors v
i
as columns. Then AT = TD. We will be done if we can
show that T is invertible. For this it is enough to prove that v
i
are linearly
independent. Suppose that

a
i
v
i
= 0
where not all the a
i
= 0. We can assume the number of a
i
,= 0 is chosen as
small as possible and that i = 1 is among these indices. Then multiplying the
equation by T
1
I yields a shorter relation

a
i
(
i

1
)v
i
= 0
which is a contradiction.
9
Proof of Cayley-Hamilton. We have to show that the morphism A
n
2
A given
by A p
A
(A) vanishes identically. It suces to check this for A D
n
since
it is dense. By the previous theorem, any matrix A D
n
can be diagonalized.
Since p
T
1
AT
(T
1
AT) = T
1
p
A
(A)T, we can reduce to the case where
A =

1
0 . . .
0
2
0 . . .
. . .

The result is now straightforward calculation using p

A
(t) =

(t
i
).
1.5 Ane Varieties
The Noetherian property of k[x
1
, . . . x
n
] has the following topological interpre-
tation.
Lemma 1.5.1. Any descending chain X
1
X
2
. . . stabilizes (X
N
= X
N+1
=
. . . for some N).
Proof. The chain of ideals 1(X
1
) 1(X
2
) . . . has to stabilize by the Noethe-
rian property.
A space satisfying this lemma is called Noetherian. Let X = V (I) A
n
k
.
We give X the induced topology, which means that a subset of X is closed if it
is closed as a subset of A
n
k
. X is again Noetherian. We say that X is reducible
if it is a union of two proper closed sets. Otherwise X is called irreducible.
Exercise 1.5.2. Show that V (f) is irreducible if f is an irreducible polynomial.
The unique factorization property for polynomials can be generalized as
follows.
Theorem 1.5.3. Any Noetherian space X can be expressed as a union of X =
X
1
X
2
. . . X
n
of irreducible closed sets, where no X
i
is contained in an X
j
.
This is unique up to reordering.
Proof. If X is irreducible, there is nothing to prove. Suppose X is reducible,
then we can write X = X
(0)
X
(1)
where X
(i)
are proper and closed. Repeat
this for each X
(i)
, and continue doing this. Lets represent this as a tree:
X
z
z
z
z
z
z
z
z
D
D
D
D
D
D
D
D
X
(0)
w
w
w
w
w
w
w
w
X
(1)
G
G
G
G
G
G
G
G
X
(00)
X
(01)
X
(10)
X
(11)
10
By the Noetherian property, we cant continue this forever. Thus the tree
must be nite (were using Konigs lemma that an innite binary tree contains
an innite path). The leaves, i.e. the lowest elements give the X
i
.
Suppose we have another decomposition, X = X

1
X

2
. . . X

m
. Then
X

i
= (X
1
X

i
) (X
2
X

i
) . . .
Since the left side is irreducible, we must have X

i
= X
j
X

i
for some j. So
that X

i
X
j
. By symmetry, X
j
X

. Therefore X

i
= X

by assumption, and
this forces X

i
= X
j
. This proves
X

1
, X

2
, . . . X
1
, X
2
, . . .
We get the opposite inclusion by symmetry.
The X
i
in the theorem are called the irreducible components. An irreducible
closed subset of some A
n
k
is called an ane algebraic variety.
Lemma 1.5.4. X A
n
k
is irreducible i 1(X) is a prime ideal.
In terms of ideal theory, the irreducible components of X correspond to the
minimal primes of I = 1(X). That is primes ideals p containing I and minimal
with respect to this property.
1.6 Hilberts Nullstellensatz
We havent completely nished our analysis of V and I. We need to understand
what happens if we follow one by the other. One direction is easy.
Lemma 1.6.1. V (1(X)) is the closure of X (in the Zariski topology).
Exercise 1.6.2. Prove this.
The other direction is harder. The radical of an ideal I is dened by

I = f [ n N, f
n
I
This is an ideal containing I. We dene the localization of a ring R at f R
by
R[1/f] = R[T]/(Tf 1)
The image of T will be denoted by 1/f. Notice that R[1/0] makes sense, but
the whole ring is 0. More generally,
Lemma 1.6.3. R[1/f] = 0 i f is nilpotent.
Corollary 1.6.4. Let R = k[x
1
, . . . x
n
]/I, f k[x
1
, . . . x
n
] and

f its image.
Then R[1/

f] = 0 i f

I.
Theorem 1.6.5 (Hilberts Nullstellensatz). If k is algebraically closed, 1(V (I)) =

1.
11
Proof. The inclusion

1 1(V (I)) is obvious: if f
n
vanishes on V (I) then so
does f.
Suppose that f

I. Let R = k[x
1
, . . . x
n
]/I. Then R[1/

f] ,= 0. Choose
a maximal ideal m R[1/

f]. Then R[1/

f]/m = k by the Algebraic Nullstel-

lensatz. Thus we have a homomorphism h : R[1/

f] k. The composition
R R[1/

f] k is necessarily of the form ev

a
with a V (I). The fact that
ev
a
factors through R[1/

f] means that f(a) has an inverse i.e. f(a) ,= 0. Which

shows that f / 1(V (I)).
1.7 Nilpotent matrices
Let A be 22 matrix over a eld k. The Cayley-Hamilton theorem tells us that
A
2
trace(A)A+det(A)I = 0
Therefore det(A) = trace(A) = 0 implies that A is nilpotent of order 2. Con-
versely, these vanish for a nilpotent matrix since it has zero eigenvalues. Lets
try and understand this using the Nullstellensatz. Let
A =

x
1
x
2
x
3
x
4

be the generic matrix. The polynomials det(A), trace(A) generate an ideal I

k[x
1
, . . . x
4
]. The entries of A
2
generate another ideal J. We need to check that

I =

J.
We have already hinted that many of these computations can be carried
algorithmically. The name of the game here is Grobner basis theory, and the
book by Cox, Little, Oshea [CLO] gives an introduction to this. These algo-
rithms have been implemented in several computer packages. We are going to
use Grayson and Stillmans Macaulay2 program
http://www.math.uiuc.edu/Macaulay2/
which is particularly convenient for algebraic geometry/commutative algebra.
We will check this in characteristic 0, however we need to work over a eld
where the elements and operations can be represented precisely on a machine.
We will use the prime eld k = Q even though we are interested in algebraically
closed elds containing it. This is justied by the following:
Lemma 1.7.1. Let k
1
k
2
be a eld extension. Suppose that I k
1
[x
0
, . . . x
n
]
is an ideal, and let I

k
2
[x
0
, . . . x
n
] be the ideal generated by I. Then I

k
1
[x
0
, . . . x
n
] = I and

I

is generated by

I.
Proof. We prove this using tensor products (see Atiyah-MacDonald for a re-
view). We have k
2
[x
0
, . . . x
n
] = k
2

k1
k
1
[x
0
, . . . x
n
] as algebras. Furthermore
I

= k
2

k1
I and the rst statement follows easily.
12
Let J = k
2

I. We certainly have I

, we just have to check

that J is radical. This is equivalent to condition that the ring
k
2
[x
1
, . . . x
n
]/J

= k
2

k1
(k
1
[x
1
, . . . x
n
]/

I)
has no nonzero nilpotents. This is clear, since (a f)
n
= a
n
f
n
= 0 forces a
or f to be 0.
Beware that for some questions the eld does matter. For example, the ideal
(x
2
2) is prime over Q but not over C.
Below is a transcript of a Macaulay 2 session which shows that

I =

J
and that J I. It shouldnt be too hard to understand what the commands
are doing. The ; is used to suppress output, = is used to assign values, and ==
for testing equality.
i1 : R = QQ[x_1..x_4];
i2 : A = matrix{{x_1,x_2},{x_3,x_4}};
i3 : D = det A;
i4 : T = trace A;
i5 : I =ideal {D,T};
i6 : J = ideal A^2;
i7 : I == J
o7 = false
i8 : isSubset(J,I)
o8 = true
i9 : radical I == radical J
o9 = true
We will do a similar calculation for 3 3 matrices over the nite eld k =
Z/101Z, since Macaulay 2 is more ecient in nite characteristic. We let I
denote the ideal dened by the coecients of the characteristic polynomial of a
generic 3 3 matrix which in addition to the det and trace includes the trace
of the matrix of 2 2 minors generated by the exteriorPower command. We
show that V (I) is the set of nilpotent matrices of order 3.
i1 : R = ZZ/101[x_1..x_9];
13
i2 : A = genericMatrix(R,x_1,3,3)
o2 = | x_1 x_4 x_7 |
| x_2 x_5 x_8 |
| x_3 x_6 x_9 |
3 3
o2 : Matrix R <--- R
i3 : I = ideal { det(A), trace(A), trace(exteriorPower(2,A))};
o3 : Ideal of R
i4 : J = ideal A^3;
o4 : Ideal of R
i5 : radical I == radical J
o5 = true
Exercise 1.7.2. Let K be the ideal dening nilpotent 2 2 matrices of order
3. Show that

J coincides with

K. Does J = K?
14
Chapter 2
Projective Geometry
2.1 Projective space
In Euclidean plane geometry, we need seperate the cases of pairs of lines which
meet and parallel lines which dont. Geometry becomes a lot simpler if any
two lines met possibly at innity. There are various ways of arranging this,
the most convenient method is to embed the A
2
into 3 dimensional space as
depicted. To each point P A
2
, we can associate the line OP. The lines
parallel to the plane correspond to the points at innity.
P
O
We now make this precise. n dimensional projective space P
n
k
over a (not
necessarily algebraically closed) eld k consists of the set of lines through 0, or
equivalently one dimensional subspaces, in A
n+1
. There is a map : A
n+1
k

0 P
n
k
which sends v to its span. We will usually write [a
0
, . . . a
n
] for
((a
0
, . . . a
n
)). We identify (a
1
, . . . a
n
) A
n
with the point [1, a
1
, . . . a
n
] P
n
.
The complement of A
n
is called the hyperplane at innity. It can be identied
with P
n1
.
15
2.2 Projective varieties
We want to do algebraic geometry on projective space. Given X P
n
, dene
the cone over X to be Cone(X) =
1
X 0 A
n+1
. A subset of A
n+1
of
this form is called a cone. We dene X P
n
to be algebraic i Cone(X) is
algebraic in A
n+1
.
Lemma 2.2.1. The collection of algebraic subsets are the closed for a Noethe-
rian topology on P
n
also called the Zariski topology. A
n
P
n
is an open subset.
Proof. Exercise!
There are natural embeddings A
n
P
n
given by
(a
1
, . . . a
n
) [a
0
, . . . a
i1
, 1, a
i
. . . a
n
]
This identies the image with the open set U
i
= x
i
,= 0. This gives an open
cover of P
n
which allows many problems to be reduced to ane geometry.
We can dene the strong topology of P
n
C
in the same way as the Zariski
topology.
Lemma 2.2.2. P
n
C
is a compact Hausdor space which contains A
n
C
as a dense
open set. In other words, it is a compactication of ane space.
Proof. P
n
C
is compact Hausdor since it is the image of the unit sphere z
C
n+1
[ [z[ = 1.
Lets make the notion of algebraic set more explicit. We will use variables
x
0
, . . . x
n
. Thus X is algebraic i Cone(X) = V (S) for some set of polynomials
in k[x
0
, . . . x
n
]. Lets characterize the corresponding ideals. Given a polynomial
f, we can write it as a sum of homogeneous polynomials f = f
0
+f
1
+. . .. The
f
i
will be called the homogeneous components of f.
Lemma 2.2.3. I k[x
0
, . . . x
n
] is generated by homogeneous polynomials i I
contains all the homogeneous components of its members.
Proof. Exercise!
An I k[x
0
, . . . x
n
] is called homogeneous if it satises the above conditions
Lemma 2.2.4. If I is homogeneous then V (I) is a cone. If X is a cone, then
1(X) is homogeneous .
Proof. We will only prove the second statement. Suppose that X is a cone.
Suppose that f 1(X), and let f
n
be its homogenous components. Then for
a X,

t
n
f
n
(a) = f(ta) = 0
which implies f
n
1(X).
16
We let PV (I) denote the image of V (I) 0 in P
n
. Once again, we will
revert to assuming k is algebraically closed. Then as a corollary of the weak
Nullstellensatz, we obtain
Theorem 2.2.5. If I is homogeneous , then PV (I) = i (x
0
, . . . x
n
)

I.
A projective variety is an irreducible algebraic subset of some P
n
.
2.3 Projective closure
Given a closed subset X A
n
, we let X P
n
denote its closure. Let us describe
this algebraically. Given a polynomial f k[x
1
, . . . x
n
], it homogenization (with
respect to x
0
) is
f
H
= x
deg f
0
f(x
1
/x
0
, . . . x
n
/x
0
)
The inverse operation is f
D
= f(1, x
1
, . . . x
n
). The second operation is a homo-
morphism of rings, but the rst isnt. We have (fg)
H
= f
H
g
H
for any f, g, but
(f +g)
H
= f
H
+g
H
only holds if f, g have the same degree.
Lemma 2.3.1. PV (f
H
) is the closure of V (f).
Proof. Obviously, f(a) = 0 implies f
H
([1, a]) = 0. Thus PV (f
H
) contains V (f)
and hence its closure. Conversely, its enough to check that
1
P
(PV (f
H
)) 1
P
(V (f))
For simplicity assume that f is irreducible. Then the left hand ideal is (f
H
).
Suppose that g 1
P
(V (f)), then g
D
1(V (f)). This implies f[g
D
which shows
that f
H
[g.
We extend this to ideals
I
H
= f
H
[ f k[x
1
, . . . x
n
]
I
D
= f
D
[ f k[x
0
, . . . x
n
]
Lemma 2.3.2. I
H
is a homogenous ideal such that (I
H
)
D
= I.
Theorem 2.3.3. V (I) = PV (I
H
).
Proof.
V (I) =

fI
V (f)

V (f)
Conversely, proceed as above. Let g 1
P
(V (I)) then g
D
1(V (I)) =

I.
Thus
(g
D
)
N
= (g
D
)
N
I
for some N. Therefore g
N
I
H
. So that g 1
P
(PV (I
H
)) =

I
H
.
17
For principal ideals, homogenization is very simple. (f)
H
= (f
H
). In gen-
eral, homogenizing the generators need not give the right answer. For ex-
ample, the ane twisted cubic is the curve in A
3
dened by the ideal I =
(x
2
x
2
1
, x
3
x
3
1
). Then
(x
2
x
0
x
2
1
, x
3
x
2
0
x
3
1
) I
We will check on the computer for k = Q. But rst, let us give the correct
answer,.
Lemma 2.3.4. Let I = (f
1
, . . . f
N
) and J = (f
H
1
, . . . f
H
N
). Then
I
H
= f k[x
0
, . . . x
n
] [ m, x
m
0
f J
The process of going from J to I
H
above is called saturation with respect
to x
0
. It can be computed with the saturate command in Macaulay2.
i1 : R = QQ[x_0..x_3];
i2 : I = ideal {x_2-x_1^2, x_3 -x_1^3};
o2 : Ideal of R
i3 : J = homogenize(I, x_0)
2 3 2
o3 = ideal (- x + x x , - x + x x )
1 0 2 1 0 3
o3 : Ideal of R
i4 : IH = saturate(J, x_0)
2 2
o4 = ideal (x - x x , x x - x x , x - x x )
2 1 3 1 2 0 3 1 0 2
Exercise 2.3.5. Prove that I
H
is generated by the polynomials given in the
computer calculation, and conclude I
H
,= J (for arbitary k).
2.4 Miscellaneous examples
Consider the quadric Q given by x
0
x
3
x
1
x
2
= 0 in P
3
.
18
This is a doubly ruled surface which means that it has two families of lines.
This can be see explicitly by setting up a bijection P
1
P
1
= Q by
([s
0
, s
1
], [t
0
, t
1
]) [s
0
t
0
, s
0
t
1
, s
1
t
0
, s
1
t
1
]
More generally, the Segre embedding of
P
m
P
n
P
(m+1)(n+1)1
is given by sending ([s
i
], [t
j
]) to [s
i
t
j
] ordered appropriately.
Exercise 2.4.1. Check that the image of the Segre embedding is a projective
variety.
The rational normal curve of degree n is the image of P
1
in P
n
under
[x
0
, x
1
] [x
n
0
, x
n1
0
x
1
, . . . x
n
1
]
Exercise 2.4.2. Check that the rational normal curve is a projective variety.
2.5 Grassmanians
Lets turn to a fairly important but subtle example of a projective variety. Let
r n. As a set the Grassmanian Gr(r, n) is the set of r dimensional subspaces
of k
n
. For example, Gr(1, n) = P
n1
. Let Mat = Mat
rn

= A
nr
denote the
space of r n matrices, and let R(r, n) Mat denote the subset of matrices
of rank r. This is a dense open subset of Mat. Choose the standard basis of
k
n
, and represent the elements by row vectors. Then we have a surjective map
R(r, n) Gr(r, n) which sends A to the span of its rows. This is not a very
good parameterization since it is very far from one to one. In fact, A and A

represent the same subspace i A

= MA for some invertible r r matrix. Let

N =

n
r

and label the minors of A by integers 1, . . . N. Let us consider the map

pl : Mat A
N
, which sends A to its vector of r r minors. We call pl(A) the
Pl ucker vector of A. Note that pl
1
A
N
0 = R(r, n). If A and A

dene the
same point, so that A

= AM, then pl(A

) = det(M)pl(A). Therefore, we have

proven
19
Lemma 2.5.1. The map A [pl(A)] is an injection from Gr(r, n) P
N1
.
We are going to prove that
Theorem 2.5.2. The image of Gr(r, n) in P
N1
is a projective variety.
Lets start by trying to discover the equations for Gr(2, 4). Identify (x
1
, . . . x
8
)
A
8
with

x
1
x
3
x
5
x
7
x
2
x
4
x
6
x
8

Order the minors by

1 2, 1 3, 2 3, 1, 2 4, 3 4
Then pl is associated to the map of polynomials rings:
k[y
1
, . . . y
6
] k[x
1
, . . . x
8
]
sending y
i
to the ith minor. We can discover the relations among the minors
by looking at the kernel of this map. We do this using Macaulay 2.
i1 : R = QQ[x_1..x_8];
i2 : S = QQ[y_1..y_6]
i3 : A = genericMatrix(R,x_1,2,4)
o3 = | x_1 x_3 x_5 x_7 |
| x_2 x_4 x_6 x_8 |
2 4
o3 : Matrix R <--- R
i4 : M2 = exteriorPower(2,A);
1 6
o4 : Matrix R <--- R;
i5 : pl = map(R,S,M2);
i6 : ker pl
o6 = ideal(y y - y y + y y )
3 4 2 5 1 6
20
Having discovered the basic equation
y
3
y
4
y
2
y
5
+y
1
y
6
= 0 (2.1)
by machine over Q, lets check this by hand for any k. Let R(2, 4)
i
be the set of
matrices where the ith minor is nonzero. This gives an open covering of R(2, 4).
The matrices in R(2, 4) are of the form

x
1
x
3
1 0
x
2
x
4
0 1

times a nonsingular 2 2 matrix. Thus its Pl ucker vector is nonzero multiple

of
(x
1
x
4
x
2
x
3
, x
2
, x
4
, x
1
, x
3
, 1)
It follows easily that (2.1) holds for this. Moreover, the process is reversable,
any vector satisfying (2.1) with y
6
= 1 is the Pl ucker vector of

y
4
y
5
1 0
y
2
y
3
0 1

By a similar analysis for the other R(2, 4)

i
, we see that Gr(2, 4) is determined
by (2.1).
In order to analyze the general case, it is convenient to introduce the appro-
priate tool. Given a vector space V , the exterior algebra

V is the free asso-

ciative (noncommutative) algebra generated by V modulo the relation v v = 0
for all v V . This relation forces anticommutivity:
(v +w) (v +w) = v w +w v = 0
If v
1
, . . . v
n
is a basis for V , then
v
i1
. . . v
in
[ i
1
< . . . i
n

forms a basis for

V .
r
V is the space of

V spanned by r-fold products of

vectors. dim
r
V =

n
r

.
Let V = k
n
, with v
i
the standard basis. The exterior algebra has a close
connection with determinants.
Lemma 2.5.3. Let A be an r n matrix and let w
i
=

a
ij
v
j
be the ith row,
then
w
1
. . . w
r
=

(i
1
. . . i
r
th minor )v
i1
. . . v
ir
This says that the w
1
. . . w
r
is just the Pl ucker vector pl(A). The condition
for an element of
r
V to be of the form pl(A) is that it is decomposable, this
means that it is a wedge product of r elements of V .
Lemma 2.5.4. Given
s
V and linearly independent vectors v
1
, . . . v
r
V ,
i, v
i
= 0 i =

v
1
. . . v
r
.
21
Proof. One direction is clear. For the other, we can assume that v
i
is part of
the basis. Writing
=

i1<...<is
a
i1...is
v
i1
. . . v
is
Then
v
1
=

i1<...<is;i=i1,...i=is
a
i1...is
v
i1
. . . v
is
v
1
= 0
precisely if 1 is included in the set of indices for which a
i1...is
,= 0. This means
that v
1
can be factored out of . Continuing this way gives the whole lemma.
Corollary 2.5.5. A nonzero is decomposable i the dimension of the kernel
of v v on V is at least r, or equivalently if the rank of this map is at most
n r.
Corollary 2.5.6. The image of pl is algebraic.
Proof. The conditions on the rank of v v are expressible as the vanishing
of (n r + 1) (n r + 1) minors of its matrix.
2.6 Elimination theory
Although A
n
is compact with its Zariski topology, there is a sense in which it
isnt. To motivate the precise statement, we start with a result in point set
topology. Recall that a map of topological spaces is closed if it takes closed sets
to closed sets.
Theorem 2.6.1. If X is a compact metric space then for any metric space Y ,
the projection p : X Y Y is closed
Sketch. Given a closed set Z XY and a convergent sequence y
i
p(Z), we
have to show that the limit y lies in p(Z). By assumption, we have a sequence
x
i
X such that (x
i
, y
i
) Z. Since X is compact, we can assume that x
i
converges to say x X after passing to a subsequence. Then we see that
(x, y) is the limit of (x
i
, y
i
) so it must lie in Z because it is closed. Therefore
y p(Z).
The analogous property for algebraic varieties is called completeness or
properness. To state it precisely, we need to dene products. For ane va-
rieties X A
n
and Y A
m
, their set theoretic product X Y A
nm
is
again algebraic and irreducible.
Exercise 2.6.2. Prove this.
Given a polynomial f k[x
0
, . . . x
n
, y
1
, . . . y
m
] which is homogeneous in the
x
i
, we can denes its zeros in P
n
A
m
as before. Similarly if f is homogeneous
seperately in the xs and ys or bihomogeneous, then we can dene its zeros in
P
n
P
m1
. A subset of either product is called algebraic if it is the intersection
of the zero sets of such polynomials. We can dene a Zariski topology whose
closed sets are exactly the algebraic sets.
22
Exercise 2.6.3. Show that the Zariksi topology on P
n
P
m
coincides with
induced topology under the Segre embedding P
n
P
m
P
(n+1)(m+1)1
.
With this it is easy to see that:
Lemma 2.6.4. The projections P
n
A
m
P
n
etcetera are continuous.
We say that a variety X is complete if the projections X A
m
A
m
are
all closed. Then A
n
is not complete if n > 0. To see this, observe that the
projection A
n
A
1
A
1
is not closed since V (x
1
y 1) maps onto A
1
0.
It is possible to describe the closure of the images explicitly. Given coordinate
x
1
, . . . x
n
on A
n
and y
1
, . . . y
m
on A
m
. A subvariety of the product is given
by an ideal I k[x
1
, . . . x
n
, y
1
, . . . y
m
]. The intersection I

= I k[y
1
, . . . y
m
]
is called the elimination ideal (with respect to the chosen variables). Then it
known V (I

) = V (I) [CLO, chap 3].

Exercise 2.6.5. Prove that V (I

) V (I).
Theorem 2.6.6 (Main theorem of elimination theory). P
n
is complete. That
is given a collection of polynomials f
i
(x, y) homogeneous in x, there exists poly-
nomials g
j
(y) such that
x ,= 0, if
i
(x, y) = 0 jg
j
(y) = 0
Proofs can be found in the references at the end. Note that the images can
be computed explicitly by using elimination ideals.
Corollary 2.6.7. The projections P
n
P
m
P
m
are closed.
Proof. Let Z P
n
P
m
be closed. P
m
has a nite open cover U
i
dened
earlier. Since |
i
can be identied with A
m
, it follows that Z U
i
is closed in
U
i
. Therefore P
m
= (U
i
Z) is open.
2.7 Simultaneous eigenvectors
Given two n n matrices A, B, a vector v is a simultaneous eigenvector if it is
an eigenvector for both A and B with possibly dierent eigenvalues. It should
be clear that the pairs admiting simultaneous eigenvectors are special. The
question is how special.
Proposition 2.7.1. The set S of (A, B) (Mat
nn
(k))
2
= A
n
2
A
n
2
such
that A and B admit a simultaneous eigenvector is a proper Zariski closed set.
Proof. We rst introduce an auxillary set
E
2
= (A, B, v) A
n
2
A
n
2
k
n
[ v is an eigenvector for A&B
This is dened by the conditions rank(Av, v) 1 and rank(Bv, v) 1 which
is expressable as the vanishing of the 2 2 minors of these matrices. These
23
are multihomogeneous equations, therefore they dene a closed set in P
n
2
1

P
n
2
1
P
n1
. Let PS be the image of this set in P
n
2
1
P
n
2
1
under projection.
This is closed by the elimination theorem and therefore the preimage of PS in
(A
n
2
0)
2
is also closed. It is now easy to see that our set S, which is the
union of the preimage with the axes 0 A
n
2
A
n
2
0, is also closed.
The proposition implies that the set S above is dened by polynomials. Lets
work out the explicit equations when n = 2 and k = Q. We write our matrices
as
A =

a b
c d

B =

e f
g h

Now we work out the ideal I dening E

2
:
i1 : R = QQ[x,y,a,b,c,d,e,f,g,h];
i2 : Avv = matrix {{a*x+ b*y, x},{c*x+d*y, y}}
o2 = | xa+yb x |
| xc+yd y |
2 2
o2 : Matrix R <--- R
i3 : Bvv = matrix {{e*x+f*y,x}, {g*x+h*y, y}}
o3 = | xe+yf x |
| xg+yh y |
2 2
o3 : Matrix R <--- R
i4 : I= ideal( determinant Avv, determinant Bvv)
2 2 2 2
o4 = ideal (x*y*a + y b - x c - x*y*d, x*y*e + y f - x g - x*y*h)
Now try to eliminate x, y:
i5 : eliminate(I, {x,y})
o5 = 0
24
And we get 0? The problem is that ane algebraic set V (I) has two irreducible
components which we can see by calculating minimal primes
i6 : minimalPrimes I
o6 = {ideal (- x*y*a - ... ),ideal (y, x)}
(we are suppressing part of the output). The uninteresting component x = y = 0
corresponding to the second ideal is throwing o our answer. Setting J to the
rst ideal, and then eliminating gives us the right answer.
i7 : J= o6#0;
o7 : Ideal of R
i8 : eliminate(J,{x,y})
After cleaning this up, we see that the locus of pairs of matrices with a simul-
taneous eigenvector is dened by
bce
2
acef +cdef c
2
f
2
abeg+bdeg+a
2
fg+2bcfg2adfg+d
2
fgb
2
g
2
2bceh
+acfh cdfh +abgh bdgh +bch
2
= 0
The corresponding computation for n = 3 seems painfully slow (at least on
my computer). So we ask a more qualitative question: can S be dened by a
single equation for n > 2? We will see that the answer is no in the next chapter.
25
Chapter 3
The category of varieties
3.1 Rational functions
Given an ane variety X A
n
, its coordinate ring
O(X) = k[x
1
, . . . x
n
]/1(X)
An f O(X) is an equivalence class of polynomials. Any two polynomial
representatives

f
i
of f give the same value

f
1
(a) =

f
2
(a) for a X. Thus the
elements can be identied with functions X k called regular functions. Since
1(X) is a prime ideal, it follows that O(X) is an integral domain. Its eld of
fractions is called the function eld k(X) of X. An element f/g k(X) is called
a rational function. It is regular on the open set D(g) = g(x) ,= 0. Note that
D(g) can be realized as the ane variety
(a, b) A
n+1
[ g(a)b = 1
Its coordinate ring is the localization O(X)[1/g]. This has the same function
eld as X. Thanks to this, we can dene the function eld of a projective variety
X P
n
as follows. Choose i so that U
i
has a nonempty intersection with X,
then set k(X) = k(X U
i
). If X U
j
,= then
k(X U
i
) = k(X U
i
U
j
) = k(X U
j
)
so this is well dened.
Example 3.1.1. k(P
n
) = k(A
n
) = k(x
1
. . . x
n
).
It is convenient to enlarge the class of varieties to a class where ane and
projective varieties can be treated together. A quasi-projective variety is an
open subset of a projective variety. This includes both ane and projective
varieties and some examples which are neither such as P
n
A
m
. The local
study of quasi-projective varieties can be reduced to ane varieties because of
26
Lemma 3.1.2. Any quasi-projective variety X has a nite open cover by ane
varieties.
Proof. Suppose that X is an open subset of a projective variety Y P
n
. Then
X U
i
is open in Y U
i
. Let f
ij
be a nite set of generators of the ideal of
(Y X) U
i
Then Y U
i
V (f
ij
) is our desired open cover. These sets are
ane since they can be embedded as closed sets into A
n+1
.
When X is quasi-projective with a cover as above, we dene k(X) = k(U
i
)
for some (and therefore any) i.
3.2 Quasi-projective varieties
We have dened morphisms of ane varieties. These are simply maps dened
by polynomials. We can compose morphisms to get new morphisms. Thus the
collection of ane varieties and morphisms between them forms a category. The
denition of categories and functors can be found, for example, in [L].
A morphism F : X Y of ane varieties given by
(x
1
, . . . x
n
) (F
1
(x
1
. . . x
n
), . . . F
m
(x
1
, . . . x
n
))
induces an algebra homomorphism F

: O(Y ) O(X) called pull back, given

by
F

(p(y
1
, . . . y
m
)) = p(F
1
(x), . . . F
m
(x))
This can be identied with the composition p pF of functions. If F : X Y
and G : Y Z, then we have (G F)

= F

. Therefore the assignment

X O(X), F F

is a contravariant functor from the category of ane

varieties to the category of nitely generated k-algebra which happen to be
domains. In fact:
Theorem 3.2.1. The category of ane varieties is antiequivalent to this cate-
gory of algebras.
Exercise 3.2.2. The theorem amounts to the following two assertions.
1. Show that any algebra of the above type is isomorphic to O(X) for some
X.
2. Show that there is a one to one correspondence between the set of mor-
phisms from X to Y and k-algebra homomorphisms O(Y ) O(X).
We want to have a category of quasi-projective varieties. The denition of
morphism is a little trickier than before, since we can allow rational expressions
provided that the singularities are removable. We start with a weaker notion.
A rational map F : X Y of projective varieties is given by a morphism
from an ane open set of X to an ane open set of Y . Two such morphisms
dene the same rational map if they agree on the intersection of their domains.
So a rational map is really an equivalence class. A morphism F : X Y of
27
quasi-projective varieties is a rational map such that for an ane open cover
U
i
F : X U
i
Y extends to a morphism for all i.
Here are two examples.
Example 3.2.3. The map P
1
P
3
given by [x
0
, x
1
] [x
3
0
, x
2
0
x
1
, x
0
x
2
1
, x
3
1
] is a
morphism.
Example 3.2.4. Let E be PV (zy
2
x(x z)(x 2z)) P
2
. Then E P
1
given by [x, y, z] [x, z] is a morphism.
The rst should be obvious, but the second isnt. The second map appears
to be undened at [0, 1, 0] E. To check, we can work on U
2
= y ,= 0.
Normalizing y = 1 and substituting into the dening equation gives
[x, y, z] [x, z] = [x, x(x z)(x 2z)] = [1, (x z)(x 2z)]
The last expression is well dened even on [0, 1, 0].
A regular function on a quasi-projective variety is morphism from it to A
1
.
For ane varieties, this coincides with the denition given earlier. The following
exercise gives a characterization of morphism which is taken as the denition
in more advanced presentations. (Of course one would need to dene regular
functions rst.)
Exercise 3.2.5. Show that a continuous map F : X Y of quasi-projective
varieties is a morphism if and only if f F :
1
U k is regular whenever
f : U k is regular.
3.3 Graphs
Proposition 3.3.1. If f : X Y is a morphism of quasi-projective varieties,
then the graph
f
= (x, f(x)) [ x X is closed.
Proof. We rst treat the case of the identity map id : X X. The graph is
just the diagonal
X
= (x, x) [ x X. This is closed when X = P
n
since
it is dened by the equations x
i
y
i
= 0 . For X P
n
,
X
=
P
n X X
is therefore also closed. Dene the morphism f id : X Y Y Y by
(x, y) (f(x), y), then
f
= (f id)
1

Y
. So it is closed.
Exercise 3.3.2. Let C = V (y
2
x
3
). Show that f : C A
1
given by (x, y)
y/x if x ,= 0, and f(0, 0) = 0 has closed graph but is not a morphism.
Proposition 3.3.3. If X is projective and Y quasi-projective, then any mor-
phism f : X Y is closed
Proof. Let p : XY X and q : XY Y be the projections. q is closed by
theorem 2.6.6. If Z X is closed then f(Z) = q(p
1
Z
f
) is also closed.
The next theorem can be viewed as an analogue of Liovilles theorem in
complex analysis, that bounded entire functions are constant.
28
Theorem 3.3.4. Any regular function on a projective variety is constant.
Proof. Let X be a projective variety. A regular function f : X A
1
is closed,
so f(X) is either a point or A
1
. The second case is impossible since f composed
with the inclusion A
1
P
1
is also closed.
Exercise 3.3.5. Show that a point is the only variety which is both projective
and ane.
29
Chapter 4
Dimension theory
4.1 Dimension
We dene the dimension dimX of an ane or projective variety to be the tran-
cendence degree of k(X) over k. This is the maximum number of algebraically
independent elements of k(X). A point is the only zero dimensional variety.
Varieties of dimension 1, 2, 3 . . . are called curves, surfaces, threefolds...
Example 4.1.1. The function eld k(A
n
) = k(P
n
) = k(x
1
, . . . x
n
). Therefore
dimA
n
= dimP
n
= n as we would expect.
A morphismF : X Y is called dominant if F(X) is dense. If F is dominant
then the domain of any rational function on Y meets F(X), so it can be pulled
back to a rational function on X. Thus we get a nonzero homomorphism k(Y )
K(X) which is necessarily injective. A rational map is called dominant if it can
be realized by a dominant morphism. Putting all of this together yields:
Lemma 4.1.2. If F : X Y is a dominant rational map, then we have an
inclusion k(Y ) k(X). Therefore dimX dimY .
A dominant morphismF : X Y is called generically nite if k(Y ) k(X)
is a nite eld extension.
Lemma 4.1.3. If F : X Y is a generically nite rational map, then
dimX = dimY .
Proof. The extension k(Y ) k(X) is algebraic, so the transcendence degree is
unchanged.
Example 4.1.4. Suppose that f(x
1
, . . . x
n
) =

f
i
(x
1
, . . . x
n1
)x
i
n
is a polyno-
mial nonconstant in the last variable. Then projection V (f) A
n1
onto the
rst n 1 coordinates is generically nite.
Lemma 4.1.5. If f is a nonconstant (homogeneous) polynomial in n (respec-
tively n+1) variables, then dimV (f) = n1 (respectively dimPV (f) = n1).
30
Proof. By reordering the variables, we can assume that the conditions of the
previous example hold. Then dimV (f) = dimA
n1
. The projective version
follows from this.
In fact, much more is true:
Theorem 4.1.6 (Krulls principal ideal theorem). If X A
n
is a closed set
and f k[x
1
, . . . x
n
] gives a nonzero divisor in k[x
1
, . . . x
n
]/1(X), then the
dimension of any irreducible component of V (f) X is dimX 1.
Although we have stated it geometrically, it is a standard result in commu-
tative algebra [AM, E]. By induction, we get:
Corollary 4.1.7. Suppose that f
1
, f
2
, . . . f
r
gives a regular sequence in k[x
1
, . . . x
n
]/1(X)
which means f
1
is a nonzero divisor in k[x
1
, . . . x
n
]/1(X), f
2
a nonzero divisor
in k[x
1
, . . . x
n
]/1(X)+(f
1
) and so on. Then any component of V (f
1
, . . . f
r
)X
has dimension dimX r.
Regularity can be understood as a kind of nondegeneracy condition. Without
it, dimX r only gives a lower bound for the dimension of the components.
4.2 Dimension of bres
We want to state a generalization of the rank-nullity theorem from elementary
linear algebra. The role of the kernel of a linear map is played by the bre. Given
a morphism f : X Y of quasi-projective varieties. The bre f
1
(y) X is a
closed set. So it is a subvariety if it is irreducible.
Theorem 4.2.1. Let f : X Y be a dominant morphism of quasi-projective
varieties and let r = dimX dimY . Then for every y f(X) the irreducible
components of f
1
(y) have dimension at least r. There exists a nonempty open
set U f(X) such that for y U the components of f
1
(y) have dimension
exactly r.
A complete proof can be found in [M, I8]. We indicate a proof of the rst
statement. We can reduce to the case where X and Y are ane, and then
to the case Y = A
m
using Noethers normalization theorem. Let y
i
be the
coordinates on A
m
. Then the bre over a f(X) is dened by m equations
y
1
a
1
= 0, . . . y
m
a
m
= 0. So the inequality follows from the remark at the
end of the last section. For the second part, it be enough to nd an open set U
such that y
1
a
1
, . . . y
m
a
m
gives a regular sequence whenever a U.
Corollary 4.2.2. With the same notation as above, dimX = dimY if and only
if there is nonempty open set U f(X) such that f
1
(y) is nite for all y U.
Corollary 4.2.3. dimX Y = dimX + dimY .
The dimensions of the bres can indeed jump.
31
Example 4.2.4. Let B = V (y xz) A
3
. Consider the projection to the
xy-plane. The bre over (0, 0) is the z-axis, while the bre over any other point
in the range is (x, y, y/x).
We can apply this theorem to calculate the dimension of the Grassmanian
Gr(r, n). Recall there is a surjective map from the space R(r, n) of r n
matrices of rank r to Gr(r, n). This is a morphism of quasi-projective varieties.
An element W Gr(r, n) is just an r dimensional subspace of k
n
. The bre

1
(W) is the set of all matrices A R(r, n) whose columns span W. Fixing
A
0

1
(W). Any other element of the bre is given by MA
0
for a unique
matrix Gl
r
. Thus we can identify
1
(W) with Gl
r
. Since Gl
r
Mat
rr
and R(r, n) Mat
rn
are open, they have dimension r
2
and rn respectively.
Therefore
dimGr(r, n) = rn r
2
= r(n r)
4.3 Simultaneous eigenvectors (continued)
Our goal is to estimate the dimension of the set S of pairs (A, B) Mat
2
nn
having a simultaneous eigenvector. As a rst step we compute the dimension of
E
1
= (A, v) Mat
nn
k
n
[ v is an eigenvector of A
We have a projection to p
2
: E
1
k
n
= A
n
which is obviously surjective.
The bre over 0 is all of Mat
nn
which has dimension n
2
. This however is the
exception and will not help us. If M Gl
n
, then (A, v) (MAM
1
, Mv)
denes an automorphism of E
1
, that is a morphism from E
1
to itself whose
inverse exists and is also a morphism. Suppose v k
n
is nonzero. Then there
is an invertible matrix M such that Mv = (1, 0 . . . 0)
T
. The automorphism just
dened takes the bre over v to the bre over (1, 0 . . . 0)
T
, so they have the
same dimension. p
1
2
((1, 0 . . . 0)
T
) is the set of matrices A = (a
ij
) satisfying
a
21
= a
31
= . . . a
n1
= 0. This is an ane space of dimension n
2
(n1). Since
this is the dimension of the bres over an open set. Therefore
dimE
1
= n
2
n + 1 +n = n
2
+ 1
This can be seen in another way. The rst projection p
1
: E
1
Mat
nn
is also
surjective, and the bre over A is the union of its eigenspaces. These should be
one dimensional for a generic matrix.
Consider
E
2
= (A, B, v) [ v is a simultaneous eigenvector
This is an example of bre product. The key point is that bre of (A, B, v) v
over v is the product of the bres p
1
2
(v) p
1
2
(v) considered above. So this has
dimension 2(n
2
n + 1). Therefore
dimE
2
= 2(n
2
n + 1) +n = 2n
2
n + 2
32
We have a surjective morphism E
2
S given by projection. The bre is the
union of simultaneous eigenspaces, which is at least one dimensional. Therefore
dimS 2n
2
n + 1
To put this another way, the codimension of S in Mat
2
nn
is at least n 1. So
it cannot be dened by single equation if n > 2.
33
Chapter 5
Dierential calculus
5.1 Tangent spaces
In calculus, one can dene the tangent plane to a surface by taking the span of
tangent vectors to arcs lying on it. In translating this into algebraic geometry, we
are going to use innitesimally short arcs. Such a notion can be made precise
using scheme theory, but this goes beyond what we want to cover. Instead we
take the dual point of view. To give a map of ane varieties Y X is equivalent
to giving a map of their coordinate algebras O(X) O(Y ). In particular,
the points of X correspond to homomorphisms O(X) k (cor. 1.2.6). Our
innitesimal arc should have coordinate ring k[]/(
2
). The intuition is that the
parameter is chosen so small so that
2
= 0. We dene a tangent vector to be
an algebra homomorphism
v : O(X) k[]/(
2
)
We can compose this with the map k[]/(
2
) k (a +b 0) to get homomor-
phism to k, and hence a point of X. This will be called the base of the vector.
Dene the tangent space T
a
X to X at a X to be the set of all tangent vectors
with base a. A tangent vector on A
n
amounts to an assignment x
i
a
i
+ b
i
,
where a
i
are coordinates of the base. Therefore T
a
A
n
= k
n
.
Exercise 5.1.1. Show that the f k[x
1
, . . . x
n
] goes to
f(a
1
+b
1
, . . .) = f(a) +

b
i
f
x
i
(a) k[]/(
2
)
In general, a vector v T
a
X is determined by the coecient of . These
coecients can be added, and multiplied by scalars. Thus T
a
X is a k-vector
space.
Theorem 5.1.2. Given a variety X A
n
and a X. Let 1(X) = (f
1
, . . . f
N
).
34
Then T
a
X is isomorphic to the kernel of the Jacobian matrix
J(a) =

f1
x1
(a) . . .
f1
xn
(a)
. . .
f
N
x1
(a) . . .

Proof. The homomorphism x

i
a
i
+ b
i
factors through O(X) if and only if
each f
j
maps to 0. From the exercise this holds precisely when J(a)b = 0.
Example 5.1.3. The locus of noninvertible matrices

x y
z t

is given by xt
yz = 0. The Jacobian J = (t, z, y, x) so the tangent space at the zero matrix
is 4 dimensional while the tangent space at any other point is 3 dimensional.
Exercise 5.1.4. Show that det(I + B) = 1 + trace(B), and conclude that
T
I
Sl
n
(k) = B Mat
nn
[ trace(B) = 0.
Suppose that U = D(f) X is a basic open set. This is also ane. Suppose
that a U.
Lemma 5.1.5. T
a
U

= T
a
X canonically.
Proof. Any tangent vector v : O(U) = O(X)[1/f] k[]/(
2
) with base a
induces a tangent vector on X. Conversely any vector v : O(X) k[]/(
2
)
with base a must factor uniquely through O(X)[1/f] since v(f) = f(a) + b is
a unit.
Thanks to this, we can dene the tangent space of a quasi-projective variety
X as T
a
U where U is an ane neighbourhood of a.
5.2 Singular points
A point a of a variety X is called nonsingular if dimT
a
X = dimX, otherwise it
is called singular. For example, the hypersurface xtyz = 0 has dimension equal
to 3. Therefore the origin is the unique singular point. X is called nonsingular
if it has no singular points. For example A
n
and P
n
are nonsingular. Over C
nonsingular varieties are manifolds.
Theorem 5.2.1. The set of nonsingular points of a quasi-projective variety
forms a nonempty open set.
Details can be found in [Ht, II 8.16] or [M, III.4]. Let us at least explains why
the set of singular points is closed. It is enough to do this for ane varieties.
Lemma 5.2.2. Let m
a
O(X) be the maximal ideal of regular functions van-
ishing at a X. Then T
a

= (m
a
/m
2
a
)

denotes the k-vector space dual).

Exercise 5.2.3. Given h (m
a
/m
2
a
)

, we can interpret it has linear map

m
a
k killing m
2
a
. Show that v(f) = f(a) + h(f f(a)) denes a tangent
vector. Show that this gives the isomorphism T
a

= (m
a
/m
2
a
)

.
35
We need a couple of facts from commutative algebra [AM, E]
The dimension of X as dened earlier coincide with the Krull dimension
of both O(X) and its localization O(X)
ma
.

dimm
a
/m
2
a
= dimm
a
O(X)
ma
/(m
a
O(X)
ma
)
2
dimO(X)
ma
It follows that if X A
n
, then
rankJ(a) n dimX
for all a X. The set of singular points of is the set where this inequality is
strict. This set is dened by the vanishing of the (n dimX)
2
minors of J.
Therefore it is closed.
A variety is called homogenous if its group of automorphisms acts transi-
tively. That is for any two points there is an automorphism which one to the
other.
Corollary 5.2.4. A homogenous quasi-projective variety is nonsingular.
Exercise 5.2.5.
1. Show that the space of matrices in Mat
nm
of rank r is homogeneous.
2. Show that any Grassmanian is homogeneous.
5.3 Singularities of nilpotent matrices
In this section, we again return to the computer, and work out the singularities
of the variety Nilp of nilpotent 3 3 matrices. Since these computations are
a bit slow, we work over a eld of nite (but not too small) characteristic. We
will see that in this case the singular locus of Nilp coincides with the set of
matrices A [ A
2
= 0 which is about as nice as one could hope for. Certainly
this calls for proper theoretical explanation and generalization (exercise**)!
Using Cayley-Hamilton, we can see that the ideal I of this variety is gener-
ated by the coecients of the characteristic polynomial (see 1.7). We compute
the dimension of the variety, and nd that its 6. (Actually we are comput-
ing the Krull dimension of the quotient ring k[x
1
. . . x
9
]/I which would be the
same.)
R = ZZ/101[x_1..x_9];
i2 : A = genericMatrix(R, x_1, 3,3);
3 3
o2 : Matrix R <--- R
36
i3 : I = ideal {det(A), trace(A), trace(exteriorPower(2,A))};
o3 : Ideal of R
i4 : Nilp = R/I;
i5 : dim Nilp
o5 = 6
Next we do a bunch of things. We dene the ideal K of matrices satisfying
A
2
= 0, and them compute its radical. On the other side we compute the ideal
Sing of the singular locus by running through the procedure of the previous
section. Sing is the sum of I with the 3 3 minors of the Jacobian of the
generators of I. The nal step is to see that

K =

Sing.
i6 : K = ideal A^2;
o6 : Ideal of R
i7 : rK = radical K;
o7 : Ideal of R
i8 : Jac = jacobian I;
9 3
o8 : Matrix R <--- R
i9 : J = minors(3, Jac);
o9 : Ideal of R
i10 : Sing = J + I;
o10 : Ideal of R
i11 : rSing = radical Sing;
o11 : Ideal of R
i12 : rK == rSing
o12 = true
37
5.4 Bertini-Sard theorem
Sards theorem states that the set of critical points of a C

map is small in the

sense of Baire category or measure theory. We state an analogue in algebraic
geometry whose sister theorem due to Bertini is actually much older. Given
a morphism of ane varieties f : X Y and a X, we get a linear map
df
a
: T
a
X T
f(a)
Y dened by v v f

.
Exercise 5.4.1. Show that for f : A
n
A
m
, df
a
can be represented by the
Jacobian of the components evaluated at a.
Theorem 5.4.2. Let f : X Y be a dominant morphism of nonsingular
varieties dened over a eld of characteristic 0, then there exists a nonempty
open set U Y such that for any y U, f
1
(y) is nonsingular and for any
x f
1
(y), df
x
is surjective.
A proof can be found in [Ht, III 10.7]. This is the rst time that we have
encountered the characteristic zero assumption. The assumption is necessary:
Example 5.4.3. Suppose k has characteristic p > 0. The Frobenius morphism
F : A
1
A
1
is given by x x
p
. Then dF = px
p1
= 0 everywhere.
A hyperplane of P
n
is a subvariety of the form H = PV (f) where f is linear
polynomial. It is clear that H depends only on f up to a nonzero scalar multiple.
Thus the set of hyperplanes forms an n dimensional projective space in its own
right called the dual projective space

P
n
.
Lemma 5.4.4. If X P
n
is a subvariety of positive dimension then XH ,=
for any hyperplane.
Proof. If XH = then X would be contained in the ane space P
n
H.
Theorem 5.4.5 (Bertini). If X P
n
is a nonsingular variety, there exists a
nonempty open set U

P
n
such that X H is nonsingular for any H U.
Although this theorem is valid in any characteristic, the proof we give only
works in characteristic 0.
Proof. Let
I = (x, H) X

P
n
[ x H
be the so called incidence variety with projections p : I X and q : I

P
n
.
In more explicit, terms
I = ([x
0
, . . . x
n
], [y
0
, . . . y n]) [

x
i
y
i
= 0
So the preimage
p
1
(x
0
,= 0X) = ([1, x
1
, . . . x
n
], [x
1
y
1
x
2
y
2
. . . , y
1
, . . . y
n
])

= A
n
P
n1
38
A similar isomorphism holds for p
1
(x
i
,= 0 X) for any i. Thus I is nonsin-
gular. The map q is surjective by the previous lemma. Therefore by theorem
5.4.2, there exists a nonempty open U

P
n
such that X H = q
1
(H) is
nonsingular for every H U.
39
Bibliography
[AM] M. Atiyah, I. Macdonald, Commutative algebra
[CLO] D. Cox, J. Little, D. OShea, Ideals, Varieties, and Algorithms
[E] D. Eisenbud, Commutative algebra, with a view toward algebraic geometry
[GH] P. Griths, J. Harris, Principles of algebraic geometry.
[H] J. Harris, Algebraic geometry: a rst course
[Ht] R. Hartshorne, Algebraic geometry
[L] S. Lang, Algebra
[M] D. Mumford, Red book of varieties and schemes
[S] I. Shafarevich, Basic algebraic geometry
[Sk] H. Schenck, Computational algebraic geometry
40