Introduction To Algebraic Geometry: Igor V. Dolgachev August 19, 2013
Introduction To Algebraic Geometry: Igor V. Dolgachev August 19, 2013
Introduction To Algebraic Geometry: Igor V. Dolgachev August 19, 2013
Igor V. Dolgachev
August 19, 2013
ii
Contents
1 Systems of algebraic equations
13
21
31
6 B
ezout theorem and a group law on a plane cubic curve
45
57
69
77
83
11 Dimension
93
12 Lines on hypersurfaces
105
13 Tangent space
117
14 Local parameters
131
15 Projective embeddings
147
iii
iv
CONTENTS
159
17 Riemann-Roch Theorem
175
Index
191
Lecture 1
Systems of algebraic equations
The main objects of study in algebraic geometry are systems of algebraic equations and their sets of solutions. Let k be a field and k[T1 , . . . , Tn ] = k[T ] be
the algebra of polynomials in n variables over k. A system of algebraic equations
over k is an expression
{F = 0}F S ,
where S is a subset of k[T ]. We shall often identify it with the subset S.
Let K be a field extension of k. A solution of S in K is a vector (x1 , . . . , xn )
n
K such that, for all F S,
F (x1 , . . . , xn ) = 0.
Let Sol(S; K) denote the set of solutions of S in K. Letting K vary, we get
different sets of solutions, each a subset of K n . For example, let
S = {F (T1 , T2 ) = 0}
be a system consisting of one equation in two variables. Then
Sol(S; Q) is a subset of Q2 and its study belongs to number theory. For
example one of the most beautiful results of the theory is the Mordell Theorem
(until very recently the Mordell Conjecture) which gives conditions for finiteness
of the set Sol(S; Q).
Sol(S; R) is a subset of R2 studied in topology and analysis. It is a union of
a finite set and an algebraic curve, or the whole R2 , or empty.
Sol(S; C) is a Riemann surface or its degeneration studied in complex analysis
and topology.
1
All these sets are different incarnations of the same object, an affine algebraic
variety over k studied in algebraic geometry. One can generalize the notion of
a solution of a system of equations by allowing K to be any commutative kalgebra. Recall that this means that K is a commutative unitary ring equipped
with a structure of vector space over k so that the multiplication law in K is a
bilinear map K K K. The map k K defined by sending a k to a 1
is an isomorphism from k to a subfield of K isomorphic to k so we can and we
will identify k with a subfield of K.
The solution sets Sol(S; K) are related to each other in the following way.
Let : K L be a homomorphism of k-algebras, i.e a homomorphism of rings
which is identical on k. We can extend it to the homomorphism of the direct
products n : K n Ln . Then we obtain for any a = (a1 , . . . , an ) Sol(S; K),
n (a) := ((a1 ), . . . , (an )) Sol(S; L).
This immediately follows from the definition of a homomorphism of k-algebras
(check it!). Let
sol(S; ) : Sol(S; K) Sol(S; L)
be the corresponding map of the solution sets. The following properties are
immediate:
(i) sol(S; idK ) = idSol(S;K) , where idA denotes the identity map of a set A;
(ii) sol(S; ) = sol(S; ) sol(S; ), where : L M is another homomorphism of k-algebras.
One can rephrase the previous properties by saying that the correspondences
K 7 Sol(S; K), sol(S; )
define a functor from the category of k-algebras Algk to the category of sets
Sets.
Definition 1.1. Two systems of algebraic equations S, S 0 k[T ] are called
equivalent if Sol(S; K) = Sol(S 0 , K) for any k-algebra K. An equivalence class
is called an affine algebraic variety over k (or an affine algebraic k-variety). If X
denotes an affine algebraic k-variety containing a system of algebraic equations
S, then, for any k-algebra K, the set X(K) = Sol(S; K) is well-defined. It is
called the set of K-points of X.
3
Example 1.1. 1. The system S = {0} k[T1 , . . . , Tn ] defines an affine algebraic variety denoted by Ank . It is called the affine n-space over k. We have, for
any k-algebra K,
Sol({0}; K) = K n .
2. The system 1 = 0 defines the empty affine algebraic variety over k and is
denoted by k . We have, for any K-algebra K,
k (K) = .
We shall often use the following interpretation of a solution a = (a1 , . . . , an )
Sol(S; K). Let eva : k[T ] K be the homomorphism defined by sending each
variable Ti to ai . Then
a Sol(S; K) eva (S) = {0}.
In particular, eva factors through the factor ring k[T ]/(S), where (S) stands for
the ideal generated by the set S, and defines a homomorphism of k-algebras
evS,a : k[T ]/(S) K.
Conversely any homomorphism k[T ]/(S) K composed with the canonical
surjection k[T ] k[T ]/(S) defines a homomorphism k[T ] K. The images
ai of the variables Ti define a solution (a1 , . . . , an ) of S since for any F S the
image F (a) of F must be equal to zero. Thus we have a natural bijection
Sol(S; K) Homk (k[T ]/(S), K).
It follows from the previous interpretations of solutions that S and (S) define
the same affine algebraic variety.
The next result gives a simple criterion when two different systems of algebraic
equations define the same affine algebraic variety.
Proposition 1.2. Two systems of algebraic equations S, S 0 k[T ] define the
same affine algebraic variety if and only if the ideals (S) and (S 0 ) coincide.
Proof. The part if is obvious. Indeed, if (S) = (S 0 ), then for every F S
we can express F (T ) as a linear combination of the polynomials G S 0 with
coefficients in k[T ]. This shows that Sol(S 0 ; K) Sol(S; K). The opposite
inclusion is proven similarly. To prove the part only if we use the bijection
Sol(S; K) Homk (k[T ]/(S), K). Take K = k[T ]/(S) and a = (t1 , . . . , tn )
where ti is the residue of Ti mod (S). For each F S,
F (a) = F (t1 , . . . , tn ) F (T1 , . . . , Tn ) mod (S) = 0.
This shows that a Sol(S; K). Since Sol(S; K) = Sol(S 0 ; K), for any F (S 0 )
we have F (a) = F (T1 , . . . , Tn ) mod (S) = 0 in K, i.e., F (S). This gives
the inclusion (S 0 ) (S). The opposite inclusion is proven in the same way.
Example 1.3. Let n = 1, S = {T = 0}, S 0 = {T p = 0}. It follows immediately
from the Proposition 1.2 that S and S 0 define different algebraic varieties X and
Y . For every k-algebra K the set Sol(S; K) consists of one element, the zero
element 0 of K. The same is true for Sol(S 0 ; K) if K does not contain elements
a with ap = 0 (for example, K is a field, or more general, K does not have zero
divisors). Thus the difference between X and Y becomes noticeable only if we
admit solutions with values in rings with zero divisors.
Corollary-Definition 1.4. Let X be an affine algebraic variety defined by a
system of algebraic equations S k[T1 , . . . , Tn ]. The ideal (S) depends only on
X and is called the defining ideal of X. It is denoted by I(X). For any ideal
I k[T ] we denote by V (I) the affine algebraic k-variety corresponding to the
system of algebraic equations I (or, equivalently, any set of generators of I).
Clearly, the defining ideal of V (I) is I.
The next theorem is of fundamental importance. It shows that one can always
restrict oneself to finite systems of algebraic equations.
Theorem 1.5. (Hilberts Basis Theorem). Let I be an ideal in the polynomial
ring k[T ] = k[T1 , . . . , Tn ]. Then I is generated by finitely many elements.
Proof. The assertion is true if k[T ] is the polynomial ring in one variable. In fact,
we know that in this case k[T ] is a principal ideal ring, i.e., each ideal is generated
by one element. Let us use induction on the number n of variables. Every
polynomial F (T ) I can be written in the form F (T ) = b0 Tnr + . . . + br , where
bi are polynomials in the first n1 variables and b0 6= 0. We will say that r is the
degree of F (T ) with respect to Tn and b0 is its highest coefficient with respect to
Tn . Let Jr be the subset k[T1 , . . . , Tn1 ] formed by 0 and the highest coefficients
with respect to Tn of all polynomials from I of degree r in Tn . It is immediately
checked that Jr is an ideal in k[T1 , . . . , Tn1 ]. By induction, Jr is generated
by finitely many elements a1,r , . . . , am(r),r k[T1 , . . . , Tn1 ]. Let Fir (T ), i =
5
1, . . . , m(r), be the polynomials from I which have the highest coefficient equal
to ai,r . Next, we consider the union J of the ideals Jr . By multiplying a
polynomial F by a power of Tn we see that Jr Jr+1 . This immediately implies
that the union J is an ideal in k[T1 , . . . , Tn1 ]. Let a1 , . . . , at be generators
of this ideal (we use the induction again). We choose some polynomials Fi (T )
which have the highest coefficient with respect to Tn equal to ai . Let d(i) be
the degree of Fi (T ) with respect to Tn . Put N = max{d(1), . . . , d(t)}. Let us
show that the polynomials
Fir , i = 1, . . . , m(r), r < N, Fi , i = 1, . . . , t,
generate I.
Let F (T ) I be of degree r N in Tn . We can write F (T ) in the form
X
F (T ) = (c1 a1 + . . . + ct at )Tnr + . . . =
ci Tnrd(i) Fi (T ) + F 0 (T ),
1it
Exercises.
1. For which fields k do the systems
n
X
S = {i (T1 , . . . , Tn ) = 0}i=1,...,n , and S = {
Tji = 0}i=1,...,n
0
j=1
define the same affine algebraic varieties? Here i (T1 , . . . , Tn ) denotes the elementary symmetric polynomial of degree i in T1 , . . . , Tn .
2. Prove that the systems of algebraic equations over the field Q of rational
numbers
{T12 +T2 = 0, T1 = 0} and {T22 T12 +T12 +T23 +T2 +T1 T2 = 0, T2 T12 +T22 +T1 = 0}
define the same affine algebraic Q-varieties.
3. Let X Ank and X 0 Am
k be two affine algebraic k-varieties. Let us identify
n
the Cartesian product K K m with K n+m . Define an affine algebraic k-variety
such that its set of K-solutions is equal to X(K) X 0 (K) for any k-algebra K.
We will denote it by X Y and call it the Cartesian product of X and Y .
4. Let X and X 0 be two subvarieties of Ank . Define an affine algebraic variety
over k such that its set of K-solutions is equal to X(K) X 0 (K) for any kalgebra K. It is called the intersection of X and X 0 and is denoted by X X 0 .
Can you define in a similar way the union of two algebraic varieties?
5. Suppose that S and S 0 are two systems of linear equations over a field k.
Show that (S) = (S 0 ) if and only if Sol(S; k) = Sol(S 0 ; k).
6. A commutative ring A is called Noetherian if every ideal in A is finitely generated. Generalize Hilberts Basis Theorem by proving that the ring A[T1 , . . . , Tn ]
of polynomials with coefficients in a Noetherian ring A is Noetherian.
Lecture 2
Affine algebraic sets
Let X be an affine algebraic variety over k. For different k-algebras K the sets of
K-points X(K) could be quite different. For example it could be empty although
X 6= k . However if we choose K to be algebraically closed, X(K) is always
non-empty unless X = k . This follows from the celebrated Nullstellensatz of
Hilbert that we will prove in this Lecture.
Definition 2.1. Let K be an algebraically closed field containing the field k. A
subset V of K n is said to be an affine algebraic k-set if there exists an affine
algebraic variety X over k such that V = X(K).
The field k is called the ground field or the field of definition of V . Since
every polynomial with coefficients in k can be considered as a polynomial with
coefficients in a field extension of k, we may consider an affine algebraic k-set as
an affine algebraic K-set. This is often done when we do not want to specify to
which field the coefficients of the equations belong. In this case we call V simply
an affine algebraic set.
First we will see when two different systems of equations define the same
affine algebraic set. The answer is given in the next theorem. Before we state
it, let us recall that for every ideal I in a ring A its radical rad(I) is defined by
rad(I) = {a A : an I
Proof. Obviously, the set of zeroes of an ideal I and its radical rad(I) in K n
are the same. Here we only use the fact that K has no zero divisors so that
F n (a) = 0 F (a) = 0. This proves . Let V be an algebraic set in K n
given by a system of algebraic equations S. Let us show that the radical of the
ideal (S) can be defined in terms of V only:
rad((S)) = {F k[T ] : F (a) = 0 a V }.
This will obviously prove our assertion. Let us denote the right-hand side by I.
This is an ideal in k[T ] that contains the ideal (S). We have to show that for any
G I, Gr (S) for some r 0. Now observe that the system Z of algebraic
equations
{F (T ) = 0}F S , 1 Tn+1 G(T ) = 0
in variables T1 , . . . , Tn , Tn+1 defines the empty affine algebraic set in K n+1 .
In fact, if a = (a1 , . . . , an , an+1 ) Sol(Z; K), then F (a1 , . . . , an , an+1 ) =
F (a1 , . . . , an ) = 0 for all F S. This implies (a1 , . . . , an ) V and hence
G(a1 , . . . , an , an+1 ) = G(a1 , . . . , an ) = 0
and (1 Tn+1 G)(a1 , . . . , an , an+1 ) = 1 an+1 G(a1 , . . . , an , an+1 ) = 1 6= 0. We
will show that this implies that the ideal (Z) contains 1. Suppose this is true.
Then, we may write
X
1=
PF F + Q(1 Tn+1 G)
F S
9
k as a subfield. Note that A is finitely generated as a k-algebra (because k[T ]
is). Suppose we show that A is an algebraic extension of k. Then we will be
able to extend the inclusion k K to a homomorphism A K (since K is
algebraically closed), the composition k[T ]/I A K will give us a solution
of I in K n .
Thus Lemma 2.2 and hence our theorem follows from the following:
Lemma 2.3. Let A be a finitely generated algebra over a field k. Assume A is
a field. Then A is an algebraic extension of k.
Before proving this lemma, we have to remind one more definition from
commutative algebra. Let A be a commutative ring without zero divisors (an
integral domain) and B be another ring which contains A. An element x B is
said to be integral over A if it satisfies a monic equation : xn +a1 xn1 +. . .+an =
0 with coefficients ai A. If A is a field this notion coincides with the notion
of algebraicity of x over A. We will need the following property which will be
proved later in Corollary 10.2.
Fact: The subset of elements in B which are integral over A is a subring of
B.
We will prove Lemma 2.3 by induction on the minimal number r of generators
t1 , . . . , tr of A. If r = 1, the map k[T1 ] A defined by T1 7 t1 is surjective. It
is not injective since otherwise A
= k[T1 ] is not a field. Thus A
= k[T1 ]/(F ) for
some F (T1 ) 6= 0, hence A is a finite extension of k of degree equal to the degree
of F . Therefore A is an algebraic extension of k. Now let r > 1 and suppose
the assertion is not true for A. Then, one of the generators t1 , . . . , tr of A is
transcendental over k. Let it be t1 . Then A contains the field F = k(t1 ), the
minimal field containing t1 . It consists of all rational functions in t1 , i.e. ratios of
the form P (t1 )/Q(t1 ) where P, Q k[T1 ]. Clearly A is generated over F by r1
generators t2 , . . . , tr . By induction, all ti , i 6= 1, are algebraic over F . We know
d(i)
that each ti , i 6= 1, satisfies an equation of the form ai ti +. . . = 0, ai 6= 0, where
the coefficients belong to the field F . Reducing to the common denominator, we
may assume that the coefficients are polynomial in t1 , i.e., belong to the smallest
d(i)1
subring k[t1 ] of A containing t1 . Multiplying each equation by ai
, we see
that the elements ai ti are integral over k[t1 ]. At this point we can replace the
generators ti by ai ti to assume that each ti is integral over k[t1 ]. Now using the
Fact we obtain that every polynomial expression in t2 , . . . , tr with coefficients
in k[t1 ] is integral over k[t1 ]. Since t1 , . . . , tr are generators of A over k, every
element in A can be obtained as such polynomial expression. So every element
from A is integral over k[t1 ]. This is true also for every x k(t1 ). Since t1
10
11
Next we shall show that the set of algebraic k-subsets in K n can be used
to define a unique topology in K n for which these sets are closed subsets. This
follows from the following:
Proposition 2.7. (i) The intersection sS Vs of any family {Vs }sS of affine
algebraic k-sets is an affine algebraic k-set in K n .
(ii) The union sS Vs of any finite family of affine algebraic k-sets is an affine
algebraic k-set in K n .
(iii) and K n are affine algebraic k-sets.
Proof.
P (i) Let Is = I(Vs ) be the ideal of polynomials vanishing on Vs . Let
I = s Is be the sum of the ideals Is , i.e., the minimal ideal of k[T ] containing
the sets Is . Since Is I, we have V (I) VP(Is ) = Vs . Thus V (I) sS Vs .
Since each f I is equal to a finite sum
fs , where fs Is , we see that
f vanishes at each x from the intersection. Thus x V (I), and we have the
opposite inclusion.
Q
(ii) Let I be the ideal generated by products s fs , where fs Is . If
x s Vs , then x Vs for some s S. Hence all fs Is vanishes at x.
But then all products vanishes at x, and therefore x V (I). This shows that
s Vs V (I). Conversely, suppose that all products vanish at x but x 6 Vs for
any s. Then, for any
Q s S there exists some fs Is such that fs (x) 6= 0. But
then the product s fs I does not vanish at x. This contradiction proves the
opposite inclusion.
(iii) This is obvious, is defined by the system {1 = 0}, K n is defined by the
system {0 = 0}.
Using the previous Proposition we can define the topology on K n by declaring
that its closed subsets are affine algebraic k- subsets. The previous proposition
verifies the axioms. This topology on K n is called the Zariski k-topology (or
Zariski topology if k = K). The corresponding topological space K n is called
the n-dimensional affine space over k and is denoted by Ank (K). If k = K, we
drop the subscript k and call it the n-dimensional affine space.
Example 2.8. A proper subset in A1 (K) is closed if and only if it is finite.
In fact, every ideal I in k[T ] is principal, so that its set of solutions coincides
with the set of solutions of one polynomial. The latter set is finite unless the
polynomial is identical zero.
12
Remark 2.9. As the previous example easily shows the Zariski topology in K n is
not Hausdorff (=separated), however it satisfies a weaker property of separability.
This is the property
(T1 ): for any two points x 6= y in An (k), there exists an open subset U such
that x U but y 6 U (see Problem 4).
Any point x V = X(K) is defined by a homomorphism of k-algebras
evx : k[X]/I K. Let p = Ker(evx ). Since K is a field, p is a prime ideal. It
corresponds to a closed subset which is the closure of the set {x}. Thus, if x is
closed in the Zariski topology, the ideal p must be a maximal ideal. By Lemma
2.3, in this case the quotient ring (k[X]/I)/px is an algebraic extension of k.
Conversely, a finitely generated domain contained in an algebraic extension of k
is a field (we shall prove it later in Lecture 10). Points x with the same ideal
Ker(evx ) differ by a k-automorphism of K. Thus if we assume that K is an
algebraically closed algebraic extension of k then all points of V are closed.
Problems.
1. Let A = k[T1 , T2 ]/(T12 T23 ). Find an element in the field of fractions of A
which is integral over A but does not belong to A.
2. Let V and V 0 be two affine algebraic sets in K n . Prove that I(V V 0 ) =
I(V ) I(V 0 ). Give an example where I(V ) I(V 0 ) 6= I(V )I(V 0 ).
3. Find the radical of the ideal in k[T1 , T2 ] generated by the polynomials T12 T2
and T1 T23 .
4. Show that the Zariski topology in An (K), n 6= 0, is not Hausdorff but satisfies
property (T1 ). Is the same true for Ank (K) when k 6= K?
5. Find the ideal I(V ) of the algebraic subset of K n defined by the equations
T13 = 0, T23 = 0, T1 T2 (T1 + T2 ) = 0. Does T1 + T2 belong to I(V )?
6. What is the closure of the subset {(z1 , z2 ) C2 | |z1 |2 + |z2 |2 = 1} in the
Zariski topology?
Lecture 3
Morphisms of affine algebraic
varieties
In Lecture 1 we defined two systems of algebraic equations to be equivalent if
they have the same sets of solutions. This is very familiar from the theory of
linear equations. However this notion is too strong to work with. We can succeed
in solving one system of equation if we would be able to find a bijective map of
its set of solutions to the set of solutions of another system of equations which
can be solved explicitly. This idea is used for the following notion of a morphism
between affine algebraic varieties.
Definition 3.1. A morphism f : X Y of affine algebraic varieties over a field
k is a set of maps fK : X(K) Y (K) where K runs over the set of k-algebras
such that for every homomorphism of k-algebras : K K 0 the following
diagram is commutative:
X(K)
X()
fK
Y (K)
Y ()
/ X(K 0 )
(3.1)
fK 0
/ Y (K 0 )
(3.2)
14
From now on we will denote the factor algebra k[T ]/I(X) by O(X) and will
call it the coordinate algebra of X. We can view the elements of this algebra as
functions on the set of points of X. In fact, given a K-point a X(K) and an
element O(X) we find a polynomial P k[T ] representing and put
(a) = P (a).
Clearly this definition does not depend on the choice of the representative. Another way to see this is to view the point a as a homomorphism eva : O(X) K.
Then
(a) = eva ().
Note that the range of the function depends on the argument: if a is a K-point,
then (a) K.
Let : A B be a homomorphism of k-algebras. For every k-algebra K
we have a natural map of sets Homk (B, K) Homk (A, K), which is obtained
by composing a map B K with . Using the bijection (3.2) we see that any
homomorphism of k-algebras
: O(Y ) O(X)
defines a morphism f : X Y by setting, for any : O(X) K,
fK () = .
(3.3)
(3.4)
a
a
15
Proof. Let f : X Y be a morphism. It defines a homomorphism
fO(X) : Homk (O(X), O(X)) Homk (O(Y ), O(X)).
The image of the identity homomorphism idO(X) is a homomorphism : O(Y )
O(X). Let us show that () = f . Let X(K) = Homk (O(X), K). By
definition of a morphism of affine algebraic k-varieties we have the following
commutative diagram:
X(K) = HomO k (O(X), K)
fK
fO(X)
Take the identity map idO(X) in the left bottom set. It goes to the element
in the left top set. The bottom horizontal arrow sends idO(X) to . The right
vertical arrow sends it to . Now, because of the commutativity of the
diagram, this must coincide with the image of under the top arrow, which is
fK (). This proves the surjectivity. The injectivity is obvious.
As soon as we know what is a morphism of affine algebraic k-varieties we
know how to define an isomorphism. This will be an invertible morphism. We
leave to the reader to define the composition of morphisms and the identity
morphism to be able to say what is the inverse of a morphism. The following
proposition is clear.
Proposition 3.2. Two affine algebraic k-varieties X and Y are isomorphic if
and only if their coordinate k-algebras O(X) and O(Y ) are isomorphic.
Let : O(Y ) O(X) be a homomorphism of the coordinate algebras of
two affine algebraic varieties given by a system S in unknowns T1 , . . . , Tn and
a system S 0 in unknowns T10 , . . . , Tm0 . Since O(Y ) is a homomorphic image of
the polynomial algebra k[T ], is defined by assigning to each Ti0 an element
pi O(X). The latter is a coset of a polynomial Pi (T ) k[T ]. Thus is
defined by a collection of m polynomials (P1 (T ), . . . , Pm (T )) in unknowns Tj .
Since the homomorphism k[T ] O(X), Ti Pi (T ) + I(X) factors through
the ideal (Y ), we have
F (P1 (T ), . . . , Pm (T )) I(X), F (T10 , . . . , Tn0 ) I(Y ).
(3.5)
16
Note that it suffices to check the previous condition only for generators of the
ideal I(Y ), for example for the polynomials defining the system of equations Y .
In terms of the polynomials (P1 (T ), . . . , Pm (T )) satisfying (3.5), the morphism
f : X Y is given as follows:
fK (a) = (P1 (a), . . . , Pm (a)) Y (K), a X(K).
It follows from the definitions that a morphism given by polynomials
((P1 (T ), . . . , Pm (T )) satisfying (3.5) is an isomorphism if and only if there exist
polynomials (Q1 (T 0 ), . . . , Qn (T 0 )) such that
G(Q1 (T 0 ), . . . , Qn (T 0 )) I(Y ), G I(X),
Pi (Q1 (T 0 ), . . . , Qn (T 0 )) Ti0 mod I(Y ), i = 1, . . . , m,
Qj (P1 (T ), . . . , Pm (T )) Tj mod I(X), j = 1, . . . , n.
The main problem of (affine) algebraic geometry is to classify affine algebraic
varieties up to isomorphism. Of course, this is a hopelessly difficult problem.
Example 3.3. 1. Let Y be given by the equation T12 T23 = 0, and X = A1k
with O(X) = k[T ]. A morphism f : X Y is given by the pair of polynomials
(T 3 , T 2 ). For every k-algebra K,
fK (a) = (a3 , a2 ) Y (K), a X(K) = K.
The affine algebraic varieties X and Y are not isomorphic since their coordinate rings are not isomorphic. The quotient field of the algebra O(Y ) =
k[T1 , T2 ]/(T12 T23 ) contains an element T1 /T2 which does not belong to the
ring but whose square is an element of the ring (= T2 ). Here the bar denotes the
corresponding coset. As we remarked earlier in Lecture 2, the ring of polynomials
does not have such a property.
2. The circle X = {T12 + T22 1 = 0} is isomorphic to the hyperbola
Y = {T1 T2 1 = 0} provided that the field k contains a square root of 1 and
char(k) 6= 2.
3. Let k[T1 , . . . , Tm ] k[T1 , . . . , Tn ], m n, be the natural inclusion of the
polynomial algebras. It defines a morphism Ank Am
k . For any k-algebra K it
defines the projection map K n K m , (a1 , . . . , an ) 7 (a1 , . . . , am ).
Consider the special case of morphisms f : X Y , where Y = A1k (the affine
line). Then f is defined by a homomorphism of the corresponding coordinate
17
algebras: O(Y ) = k[T1 ] O(X). Every such homomorphism is determined by
its value at T1 , i.e. by an element of O(X). This gives us one more interpretation
of the elements of the coordinate algebra O(X). This time they are morphisms
from X to A1k and hence again can be thought as functions on X.
Let f : X Y be a morphism of affine algebraic varieties. We know that it
arises from a homomorphism of k-algebras f : O(Y ) O(X).
Proposition 3.4. For any O(Y ) = MorAff /k (Y, A1k ),
f () = f.
Proof. This follows immediately from the above definitions.
This justifies the notation f (the pull-back of a function).
By now you must feel comfortable with identifying the set X(K) of Ksolutions of an affine algebraic k-variety X with homomorphisms O(X)
K. The identification of this set with a subset of K n is achieved by choosing a set of generators of the k-algebra O(X). Forgetting about generators
gives a coordinate-free definition of the set X(K). The correspondence K
Hom( O(X), K) has the property of naturality, i.e. a homomorphism of kalgebras K K 0 defines a map Homk (O(X), K) Homk (O(X), K 0 ) such
that a natural diagram, which we wrote earlier, is commutative. This leads to a
generalization of the notion of an affine k-variety.
Definition 3.2. An (abstract) affine algebraic k-variety is the correspondence
which assigns to each k-algebra K a set X(K). This assignment must satisfy
the following properties:
(i) for each homomorphism of k-algebras : K K 0 there is a map X() :
X(K) X(K 0 );
(ii) X(idK ) = idX(K) ;
(iii) for any 1 : K K 0 and 2 : K 0 K 00 we have X(2 1 ) = X(2 )
X(1 );
(iv) there exists a finitely generated k-algebra A such that for each K there is
a bijection X(K) Homk (A, K) and the maps X() correspond to the
composition maps Homk (A, K) Homk (A, K 0 ).
18
We leave to the reader to define a morphism of abstract affine algebraic kvarieties and prove that they are defined by a homomorphism of the corresponding
algebras defined by property (iii). A choice of n generators f1 , . . . , fn of A defines
a bijection from X(K) to a subset Sol(I; K) K n , where I is the kernel of the
homomorphism k[T1 , . . . , Tn ] A, defined by Ti 7 fi . This bijection is natural
in the sense of the commutativity of the natural diagrams.
Example 3.5. 4. The correspondence K Sol(S; K) is an abstract affine
algebraic k-variety. The corresponding algebra A is k[T ]/(S).
5. The correspondence K K ( = invertible elements in K) is an abstract
affine algebraic k-variety. The corresponding algebra A is equal to k[T1 , T2 ]/(T1 T2
1). The cosets of T1 and T2 define a set of generators such that the corresponding
affine algebraic k-variety is a subvariety of A2 . It is denoted by Gm,k and is called
the multiplicative algebraic group over k. Note that the maps X(K) X(K 0 )
are homomorphisms of groups.
6. More generally we may consider the correspondence K GL(n, K) (=invertible n n matrices with entries in K). It is an abstract affine k-variety defined
by the quotient algebra k[T11 , . . . , Tnn , U ]/(det((Tij )U 1). It is denoted by
GLk (n) and is called the general linear group of degree n over k.
Remark 3.6. We may make one step further and get rid of the assumption in
(iv) that A is a finitely generated k-algebra. The corresponding generalization is
called an affine k-scheme. Note that, if k is algebraically closed, the algebraic set
X(k) defined by an affine algebraic k-variety X is in a natural bijection with the
set of maximal ideals in O(X). This follows from Corollary 2.6 of the Hilberts
Nullstellensatz. Thus the analog of the set X(k) for the affine scheme is the
set Spm(A) of maximal ideals in A. For example take an affine scheme defined
by the ring of integers Z. Each maximal ideal is a principal ideal generated by
a prime number p. Thus the set X(k) becomes the set of prime numbers. A
number m Z becomes a function on the set X(k). It assigns to a prime
number p the image of m in Z/(p) = Fp , i.e., the residue of m modulo p.
Now, we specialize the notion of a morphism of affine algebraic varieties to
define the notion of a regular map of affine algebraic sets.
Recall that an affine algebraic k-set is a subset V of K n of the form X(K),
where X is an affine algebraic variety over k and K is an algebraically closed
extension of k. We can always choose V to be equal V (I),where I is a radical
ideal. This ideal is determined uniquely by V and is equal to the ideal I(V )
of polynomials vanishing on V (with coefficients in k). Each morphism f :
19
X Y of algebraic varieties defines a map fK : X(K) = V Y (K) =
W of the algebraic sets. So it is natural to take for the definition of regular
maps of algebraic sets the maps arising in this way. We know that f is given
by a homomorphism of k-algebras f : O(Y ) = k[T 0 ]/I(W )) O(X) =
k[T ]/I(V ). Let Pi (T1 , . . . , Tn ), i = 1, . . . , m, be the representatives in k[T ] of
the images of Ti0 mod I(W ) under f . For any a = (a1 , . . . , an ) V viewed as
a homomorphism O(X) K its image fK (a) is a homomorphism O(Y ) K
given by sending Ti0 to Pi (a), i = 1, . . . , m. Thus the map fK is given by the
formula
fK (a) = (P1 (a1 , . . . , an ), . . . , Pm (a1 , . . . , an )).
Note that this map does not depend on the choice of the representatives Pi of
f (Ti0 mod I(W )) since any polynomial from I(W ) vanishes at a. All of this
motivates the following
Definition 3.3. A regular function on V is a map of sets f : V K such that
there exists a polynomial F (T1 , . . . , Tn ) k[T1 , . . . , Tn ] with the property
F (a1 , . . . , an ) = f (a1 , . . . , an ), a = (a1 , . . . , an ) V.
A regular map of affine algebraic sets f : V W K m is a map of sets such
that its composition with each projection map pri : K m K, (a1 , . . . , an ) 7 ai ,
is a regular function. An invertible regular map such that its inverse is also a
regular map is called a biregular map of algebraic sets.
Remark 3.7. Let k = Fp be a prime field. The map K K defined by x xp
is regular and bijective (it is surjective because K is algebraically closed and it is
injective because xp = y p implies x = y). However, the inverse is obviously not
regular.
Sometimes, a regular map is called a polynomial map. It is easy to see that it
is a continuous map of affine algebraic k-sets equipped with the induced Zariski
topology. However, the converse is false (Problem 7).
It follows from the definition that a regular function f : V K is given
by a polynomial F (T ) which is defined uniquely modulo the ideal I(V ) (of
polynomials vanishing identically on V ). Thus the set of all regular functions
on V is isomorphic to the factor-algebra O(V ) = k[T ]/I(V ). It is called the
algebra of regular functions on V , or the coordinate algebra of V . Clearly it is
isomorphic to the coordinate algebra of the affine algebraic variety X defined by
the ideal I(V ). Any regular map f : V W defines a homomorphism
f : O(W ) O(V ), 7 f,
20
Lecture 4
Irreducible algebraic sets and
rational functions
We know that two affine algebraic k-sets V and V 0 are isomorphic if and only if
their coordinate algebras O(V ) and O(V 0 ) are isomorphic. Assume that both of
these algebras are integral domains (i.e. do not contain zero divisors). Then their
fields of fractions R(V ) and R(V 0 ) are defined. We obtain a weaker equivalence
of varieties if we require that the fields R(V ) and R(V 0 ) are isomorphic. In
this lecture we will give a geometric interpretation of this equivalence relation by
means of the notion of a rational function on an affine algebraic set.
First let us explain the condition that O(V ) is an integral domain. We recall
that V K n is a topological space with respect to the induced Zariski ktopology of K n . Its closed subsets are affine algebraic k-subsets of V . From
now on we denote by V (I) the affine algebraic k-subset of K n defined by the
ideal I k[T ]. If I = (F ) is the principal ideal generated by a polynomial F , we
write V ((F )) = V (F ). An algebraic subset of this form, where (F ) 6= (0), (1),
is called a hypersurface.
Definition 4.1. A topological space V is said to be reducible if it is a union of
two proper non-empty closed subsets (equivalently, there are two open disjoint
proper subsets of V ). Otherwise V is said to be irreducible. By definition the
empty set is irreducible. An affine algebraic k-set V is said to be reducible (resp.
irreducible) if the corresponding topological space is reducible (resp. irreducible).
Remark 4.1. Note that a Hausdorff topological space is always reducible unless it
consists of at most one point. Thus the notion of irreducibility is relevant only for
non-Hausdorff spaces. Also one should compare it with the notion of a connected
21
23
Proof. Let us prove the first part. If V is irreducible, then the assertion is obvious.
Otherwise, V = V1 V2 , where Vi are proper closed subsets of V . If both of
them are irreducible, the assertion is true. Otherwise, one of them, say V1 is
reducible. Hence V1 = V11 V12 as above. Continuing in this way, we either stop
somewhere and get the assertion or obtain an infinite strictly decreasing sequence
of closed subsets of V . The latter is impossible because V is Noetherian. To
prove the second assertion, we assume that
V = V1 . . . Vk = W1 . . . Wt ,
where neither Vi (resp. Wj ) is contained in another Vi0 (resp. Wj 0 ). Obviously,
V1 = (V1 W1 ) . . . (V1 Wt ).
Since V1 is irreducible, one of the subsets V1 Wj is equal to V1 , i.e., V1 Wj .
We may assume that j = 1. Similarly, we show that W1 Vi for some i. Hence
V1 W1 Vi . This contradicts the assumption Vi 6 Vj for i 6= j unless
V1 = W1 . Now we replace V by V2 . . . Vk = W2 . . . Wt and repeat the
argument.
An irreducible closed subset Z of a topological space X is called an irreducible
component if it is not properly contained in any irreducible closed subset. Let V
be a Noetherian topological space and V = i Vi , where Vi are irreducible closed
subsets of V with Vi 6 Vj for i 6= j, then each Vi is an irreducible component.
Otherwise Vi is contained properly in some Z, and Z = i (Z Vi ) would imply
that Z Vi for some i hence Vi Vk . The same argument shows that every
irreducible component of X coincides with one of the Vi s.
Remark 4.5. Compare this proof with the proof of the theorem on factorization
of integers into prime factors. Irreducible components play the role of prime
factors.
In view of Proposition 4.3, we can apply the previous terminology to affine
algebraic sets V . Thus, we can speak about irreducible affine algebraic k-sets,
irreducible components of V and a decomposition of V into its irreducible components. Notice that our topology depends very much on the field k. For example,
an irreducible k-subset of K is the set of zeroes of an irreducible polynomial in
k[T ]. So a point a K is closed only if a k. We say that V is geometrically
irreducible if it is irreducible considered as a K-algebraic set.
Recall that a polynomial F (T ) k[T ] is said to be irreducible if F (T ) =
G(T )P (T ) implies that one of the factors is a constant (since k[T ] = k , this
a a0
aa0
0 = 0.
b b
bb
25
The homomorphism i : A Q(A) is defined by sending a A to a1 . Any
homomorphism : A K to a field K extends to a homomorphism :
Q(A) K by sending ab to (a)
. We will identify the ring A with the subring
(b)
i(A) of Q(A). In particular, the field R(V ) will be viewed as an extension
k O(V ) R(V ). We will denote the field of fractions of the polynomial ring
k[T1 , . . . , Tn ] by k(T1 , . . . , Tn ). It is called the field of rational functions in n
variables.
Definition 4.3. A dominant rational k-map from an irreducible affine algebraic
k-set V to an irreducible affine algebraic k-set W is a homomorphism of kalgebras f : R(W ) R(V ). A rational map from V to W is a dominant
rational map to a closed irreducible subset of W .
Let us interpret this notion geometrically. Restricting f to O(W ) and composing with the factor map k[T10 , . . . , Tm0 ] O(W ), we obtain a homomorphism k[T10 , . . . , Tm0 ] R(V ). It is given by rational functions R1 , . . . , Rm
R(V ), the images of the Ti s. Since every G I(W ) goes to zero, we have
G(R1 , . . . , Rm ) = 0. Now each Ri can be written as
Ri =
Pi (T1 , . . . , Tn ) + I(V )
,
Qi (T1 , . . . , Tn ) + I(V )
P (T1 , . . . , Tn )
.
Q(T1 , . . . , Tn )
2T
T2 1
, T2 = 2
.
2
T +1
T +1
27
Thus, our rational map is given by
T1 7
T2 1
2T
,
T
7
.
2
T2 + 1
T2 + 1
Next note that the obtained map is birational. The inverse map is given by
T 7
T2
.
T1 1
29
2. Same for the set defined by the equations T22 T1 T3 = 0, T12 T23 = 0.
Prove that all irreducible components of this set are birationally isomorphic to
the affine line.
3. Let f : X(K) Y (K) be the map defined by the formula from Problem 1
of Lecture 3. Show that f is a birational map.
4. Let F (T1 , . . . , Tn ) = G(T1 , . . . , Tn ) + H(T1 , . . . , Tn ), where G is a homogeneous polynomial of degree d 1 and H is a homogeneous polynomial of degree
d. Assuming that F is irreducible, prove that the algebraic set V (F ) is rational.
5. Prove that the affine algebraic sets given by the systems T13 + T23 1 = 0 and
T12 T23 /3 + 1/12 = 0 are birationally isomorphic.
Lecture 5
Projective algebraic varieties
Let A be a commutative ring and An+1 (n 0) be the Cartesian product
equipped with the natural structure of a free A-module of rank n + 1. A free
submodule M of An+1 of rank 1 is said to be a line in An+1 , if M = Ax for
some x = (a0 , . . . , an ) such that the ideal generated by a0 , . . . , an contains 1.
We denote the set of lines in An+1 by Pn (A)0 . One can define Pn (A)0 also as
follows. Let
C(A)n = {x = (a0 , . . . , an ) An+1 : (a0 , . . . , an ) = 1}.
Then each line is generated by an element of C(A)n . Two elements x, y C(A)n
define the same line if and only if x = y for some invertible A. Thus
Pn (A)0 = C(A)n /A ,
is the set of orbits of the group A of invertible elements of A acting on C(A)n
by the formula (a0 , . . . , an ) = (a0 , . . . , an ). Of course, when A is a field,
C(A)n = An+1 \ {0}, Pn (A)0 = (An+1 \ {0})/A .
If M = Ax, where x = (a0 , . . . , an ) C(A)n , then (a0 , . . . , an ) are called the
homogeneous coordinates of the line. In view of the above they are determined
uniquely up to an invertible scalar factor A .
Example 5.1. 1. Take A = R. Then P1 (R)0 is the set of lines in R2 passing
through the origin. By taking the intersection of the line with the unit circle
we establish a bijective correspondence between P1 (R) and the set of points
on the unit circle with the identification of the opposite points. Or choosing a
31
32
representative on the upper half circle we obtain a bijective map from P1 (R)0 to
the half circle with the two ends identified. This is bijective to a circle. Similarly
we can identify P2 (R)0 with the set of points in the upper unit hemisphere such
that the opposite points on the equator are identified. This is homeomorphic
to the unit disk where the opposite points on the boundary are identified. The
obtained topological space is called the real projective plane and is denoted by
RP2 .
2. Take A = C. Then P1 (C)0 is the set of one-dimensional linear subspaces
of C2 . We can choose a unique basis of x P1 (C)0 of the form (1, z) unless
x = (0, z), z C \ {0}, and Cx = C(0, 1). In this way we obtain a bijective map
from P1 (C)0 to c {}, the extended complex plane. Using the stereographic
projection, we can identify the latter set with a 2-dimensional sphere. The complex coordinates make it into a compact complex manifold of dimension 1, the
Riemann sphere CP1 .
Any homomorphism of rings : A B extends naturally to the map =
: An+1 B n+1 . If x = (a0 , . . . , an ) C(A)n , then one can write 1 =
a0 b0 + . . . + an bn for some bi A. Applying , we obtain 1 = (a0 )(b0 ) +
For our future needs we would like to enlarge the set Pn (A)0 a little further
to define the set Pn (A). We will not be adding anything if A is a field.
Let M = Ax An+1 ,P
x = (a0 , . . . , an ) Cn (A), be a line in An+1 . Choose
n+1
b0 , . . . , bn A such that i bP
M
i ai = 1. Then the homomorphism : A
defined by (0 , . . . , n ) 7 ( i i bi )x is surjective, and its restriction to M
is the identity. Since for any m An+1 we have m (m) Ker(), and
M Ker() = {0}, we see that
An+1
= M Ker().
So each line is a direct summand of An+1 . Not each direct summand of An+1 is
necessarily free. So we can enlarge the set Pn (A)0 by adding to it not necessarily
free direct summands of An+1 which become free of rank 1 after localizing the
ring. Let us explain the latter.
Let S be a non-empty multiplicatively closed subset of A containing 1. One
defines the localization MS of an A-module M in the similar way as one defines
33
the field of fractions: it is the set of equivalence classes of pairs (m, s) M
S with the equivalence relation: (m, s) (m0 , s0 ) s00 S such that
s00 (s0 m sm0 ) = 0. The equivalence class of a pair (m, s) is denoted by ms . The
equivalence classes can be added by the natural rule
m m0
s0 m + sm0
+ 0 =
s
s
ss0
(one verifies that this definition is independent of a choice of a representative).
If M = A, one can also multiply the fractions by the rule
aa0
a a0
0 =
.
s s
ss
Thus AS becomes a ring such that the natural map A AS , a 7 a1, is a
homomorphism of rings. The rule
am
a m
0 = 0.
s s
ss
equips MS with the structure of an AS -module. Note that MS = {0} if 0 S.
Observe also that there is a natural isomorphism of AS -modules
M A AS MS , m
a
am
7
,
s
s
where AS is equipped with the structure of an A-module by means of the canonical homomorphism A AS .
Example 5.2. 3. Take S to be the set of elements of A which are not zerodivisors. This is obviously a multiplicatively closed subset of A. The localized
ring AS is called the total ring of fractions. If A is a domain, S = A \ {0}, and
we get the field of fractions.
4. Let p be a prime ideal in A. By definition of a prime ideal, the set A \ p is
multiplicatively closed. The localized ring AA\p is denoted by Ap and is called the
localization of A at a prime ideal p. For example, take A = Z and p = (p), where p
is a prime number. The ring Z(p) is isomorphic to the subring of Q which consists of
fractions such that the denominator is not divisible by p.
34
35
Proof. Let Matn (A) be the ring of n n matrices with coefficients in a commutative ring A. For any ideal I in A we have a natural surjective homomorphism
which obtained by replacing each
of rings Matn (A) Matn (A/I), X 7 X,
entry of a matrix X with its residue modulo I. Now let A be a local ring, I = m
be its unique maximal ideal, and k = A/m (the residue field of A). Suppose
is an invertible matrix in Matn (k). I claim that
X Matn (A) is such that X
= In for some Y Matn (A).
X is invertible in Matn (A). In fact, let Y X
The matrix Y X has diagonal elements congruent to 1 modulo m and all offdiagonal elements belonging to m. By Lemma 5.3, the diagonal elements of
Y X are invertible in A. It is easy to see, that using elementary row transformations which do not involve switching the rows we can reduce Y X to the
identity matrix. This shows that there exists a matrix S Matn (A) such that
S(Y X) = (SY )X = In . Similarly, using elementary column transformations,
we show that X has the right inverse, and hence is invertible.
Let M be a A-module and I A an ideal. Let IM denote the submodule
of M generated by all products am, where a I. The quotient module M =
M/IM is a A/I-module via the scalar multiplication (a + I)(m + IM ) = am +
IM . There is an isomorphism of A/I-modules M/IM
= M M A (A/I),
where A/I is considered as an A-algebra via the natural homomorphism A
A/I. It is easy to check the following property.
(M N )/I(M N )
= (M/IM ) (N/IN ).
(5.1)
36
Lemma 5.6. Let P be a projective module over A. For any A-algebra B the
tensor product P A B is a projective B-module.
Definition 5.2. A projective module P over A has rank r if for each maximal
ideal m the module Pm is free of rank r.
Remark 5.7. Note that, in general, a projective module has no rank. For example,
let A = A1 A2 be the direct sum of rings. The module Ak1 An2 (with scalar
multiplication (a1 , a2 ) (m1 , m2 ) = (a1 m1 , a2 m2 )) is projective but has no rank
if k 6= n. If A is a domain, then the homomorphism A Am defines an
isomorphism of the fields of fractions Q(A)
= Q(Am ). This easily implies that
the rank of P can be defined as the dimension of the vector space P A Q(A).
We state without proof the converse of the previous Corollary (see, for example, N. Bourbaki, Commutative Algebra, Chapter 2, 5).
Proposition 5.8. Let M be a module over A such that for each maximal ideal
m the module Mm is free. Then M is a projective module.
Now we are ready to give the definition of Pn (A).
Definition 5.3. Let A be any commutative ring. The projective n-space over A
is the set Pn (A) of projective A-modules of rank 1 which are direct summands
of An+1 .
We have seen that
Pn (A)0 Pn (A).
The difference is the set of non-free projective modules of rank 1 which are direct
summands of An+1 .
A projective submodule of rank 1 of An+1 may not be a direct summand. For
example, a proper principal ideal (x) A is not a direct summand in A. A free
submodule M = A(a0 , . . . , an ) of An+1 of rank 1 is a direct summand if and
only if the ideal generated by a0 , . . . , an is equal to A, i.e. M Pn (A)0 .
This follows from the following characterization of direct summands of An+1 .
A submodule M of An+1 is a direct summand if and only if the corresponding
homomorphism of the dual modules
An+1
= HomA (An+1 , A) M = HomA (M, A)
is surjective. Sometimes Pn (A) is defined in dual terms as the set of projective
modules of rank 1 together with a surjective homomorphism An+1 M . When
37
A is a field this is a familiar duality between lines in a vector space V and
hyperplanes in the dual vector space V .
A set {fi }iI of elements from A is called a covering set if it generates
the unit
P ideal. Every covering set contains a finite covering subset. In fact if
1 = i ai fi for some ai A, we choose those fi which occur in this sum with
non-zero coefficient. For any f A we set Af = AS , where S consists of powers
of f .
Lemma 5.9. Let M be a projective module of rank r over a ring A. There
exists a finite covering set {fi }iI of elements in A such that for any i I the
localization Mfi is a free Afi -module of rank r.
Proof. We know that for any maximal ideal m in A the localization Mm is a
free module of rank r. Let x1 , . . . , xr be its generators. Multiplying them by
invertible elements in Am , we may assume that the generators belong to A. Let
: Ar M be the homomorphism defined by these generators. We know that
the corresponding homomorphism m : Arm Mm of localizations is bijective.
I claim that there exists fm 6 m such the homomorphism fm : Arfm Mfm is
bijective. Let K = Ker() and C = Coker(). Then Ker(m ) = Km = {0}.
Since K is a finitely generated module and Km = 0, there exists g 6 m such
gK = {0} and hence Kg = 0. Similarly, we find an element h 6 m such
that Ch = {0}. Now if we take fm = gh, then Kf and Cfm = {0}, hence
fm : Arfm Mfm is bijective. Since the set of elements fm is not contained in
any maximal ideal, it must generate the unit ideal, hence it is a covering set. It
remains to select a finite covering subset of the set {fm } .
Using Lemma 5.9 we may view every projective submodule M of An+1
of rank 1 as a local line : we can find a finite covering set {fi }iI such
that Mfi is a line in (Afi )n+1 . We call such a family a trivializing family for
M . If {gj }jJ is another trivializing family for M we may consider the family
{fi gjP
}(i,j)IJ . It P
is a covering family as one sees by multiplying the two relations
1 = i ai fi , 1 = j bj gj . Note that for any f, g A there is a natural homomorphism of rings Af Af g , a/f n ag n /(f g)n inducing an isomorphism of
Af g -modules Mf Af Af g
= Mf g . This shows that {fi gj }(i,j)IJ is a trivializing
family. Moreover, if Mfi = xi Afi , xi An+1
and Mgj = yj Agj , yj An+1
gj , then
fi
x0i = ij yj0
(5.2)
38
Now let us go back to algebraic equations. Fix a field k. For any k-algebra
K we have the set Pn (K). It can be viewed as a natural extension (in n + 1
different ways) of the set Ank (K) = K n . In fact, for every k-algebra K we have
the injective maps
i : Ank (K) = K n Pnk (K),
(a1 , . . . , an ) (a1 , . . . , ai , 1, ai+1 , . . . , an ), i = 0, . . . , n.
Assume that K is a field. Take, for example, i = 0. We see that
Pn (K) \ K n = {(a0 , a1 , . . . , an )A Pn (K) : a0 = 0}.
It is naturally bijectively equivalent to Pn1 (K). Thus we have
a
Pn (K) = Ank (K)
Pn1 (K).
By now, I am sure you understand what I mean when I say naturally. The
bijections we establish for different K are compatible with respect to the maps
Pn (K) Pn (K 0 ) and K n K 0n corresponding to homomorphisms K K 0
of k-algebras.
Example 5.10. The Riemann sphere
P1 (C) = C {P0 (C)}.
The real projective plane
P2 (R) = R2 P1 (R).
We want to extend the notion of an affine algebraic variety by considering
solutions of algebraic equations which are taken from Pn (K). Assume first that
L Pn (K) is a global line, i.e. a free submodule of K n+1 . Let (a0 , . . . , an ) be its
generator. For any F k[T0 , . . . , Tn ] it makes sense to say that F (a0 , . . . , an ) =
0. However, it does not make sense, in general, to say that F (L) = 0 because
a different choice of a generator may give F (a0 , . . . , an ) 6= 0. However, we can
solve this problem by restricting ourselves only with polynomials satisfying
F (T0 , . . . , Tn ) = d F (T0 , . . . , Tn ),
K .
39
Definition 5.4. A polynomial F (T0 , . . . , Tn ) k[T0 , . . . , Tn ] is called homogeneous of degree d if
X
X
ai T i
F (T0 , . . . , Tn ) =
ai0 0,...,in 0 T0i0 Tnin =
i0 ,...,in
with |i| = d for all i. Here we use the vector notation for polynomials:
i = (i0 , . . . , in ) Nn+1 , Ti = T0i0 Tnin , |i| = i0 + . . . + in .
By definition the constant polynomial 0 is homogeneous of any degree.
Equivalently, F is homogeneous of degree d if the following identity in the
ring k[T0 , . . . , Tn , t] holds:
F (tT0 , . . . , tTn ) = td F (T0 , . . . , Tn ).
Let k[T ]d denote the set of all homogeneous polynomials of degree d. This
is a vector subspace over k in k[T ] and
k[T ] = d0 k[T ]d .
Indeed every polynomial can be written uniquely as a linear combination of monomials Ti which are homogeneous of degree |i|. We write degF = d if F is of
degree d.
Let F be homogeneous polynomial in T0 , . . . , Tn . For any k-algebra K and
x K n+1
F (x) = 0 F (x) = 0 for any K .
Thus if M = Kx K n+1 is a line in K n+1 , we may say that F (M ) = 0 if
F (x) = 0, and this definition is independent of the choice of a generator of M .
Now if M is a local line and Mfi = xi Kfi Kfn+1
for some trivializing family
i
{fi }iI , we say that F (M ) = 0 if F (xi ) = 0 for all i I. This definition is
independent of the choice of a trivializing family follows from (2) above and the
following.
Lemma 5.11. Let {fi }iI be a covering family in a ring A and let a A.
Assume that the image of a in each Afi is equal to 0. Then a = 0.
Proof. By definition of Afi , we have a/1 = 0 in Afi fin ai = 0 P
for some
n 0. Obviously, we choose n to be the same for all i I. Since 1 = iI ai fi
for some
P ai A, after raising the both sides
Pin sufficient high power, we obtain
1 = iI bi fin for some bi A. Then a = iI bi fin a = 0.
40
41
Example 5.12. Let X be given by aT0 + bT1 + cT2 = 0, a projective subvariety
of the projective plane P2k . It is equal to the projective closure of the line L A2k
given by the equation bZ1 + cZ2 + a = 0. For every K the set X(K) has a
unique point P not in the image of L(K). Its homogeneous coordinates are
(0, c, b). Thus, X has to be viewed as L {P }. Of course, there are many
ways to obtain a projective variety as a projective closure of an affine variety. To
see this, it is sufficient to replace the map 0 in the above constructions by the
maps i , i 6= 0.
Let {F (T ) = 0}F S be a homogeneous system. We denote by (S) the ideal
in k[T ] generated by the polynomials F S. It is easy to see that this ideal has
the following property
(S) = d0 ((S) k[T ]d ).
In other words, each polynomial F (S) can be written uniquely as a linear
combination of homogeneous polynomials from (S).
Definition 5.6. An ideal I k[T ] is said to be homogeneous if one of the
following conditions is satisfied:
(i) I is generated by homogeneous polynomials;
(ii) I = d0 (I k[T ]d ).
Let us show the equivalence
of these two properties. If (i) holds, then every
P
F I can be written as i Qi Fi , where Fi is a set of homogeneous generators.
Writing each Qi as a sum of homogeneous polynomials, we see that F is a linear
combination of homogeneous polynomials from I. This proves (ii). Assume (ii)
holds. Let G1 , . . . , Gr be a system of generators of I. Writing each Gi as a
sum of homogeneous polynomials Gij from I, we verify that the set {Gij } is a
system of homogeneous generators of I. This shows (i).
We know that in the affine case the ideal I(X) determines uniquely an affine
algebraic variety X. This is not true anymore in the projective case.
Proposition 5.13. Let {F (T ) = 0}F S be a homogeneous system of algebraic
equations over a field k. Then the following properties are equivalent:
(i) PSol(S; K)0 = for some algebraically closed field K;
P
(ii) (S) k[T ]r := dr k[T ]d for some r 0;
(iii) for all k-algebras K, PSol(S; K) = .
42
43
Clearly I sat is a homogeneous ideal in k[T ] containing the ideal I (Check it !) .
Proposition 5.14. Two homogeneous systems S and S 0 define the same projective variety if and only if (S)sat = (S 0 )sat .
Proof. Let us show first that for any k-algebra K, the ideals (S) and (S)sat have
the same set of zeroes in Pnk (K). It suffices to show that they have the same set
of zeroes in every Pnk (K)0 . Clearly every zero of (S)sat is a zero of (S). Assume
that a = (a0 , . . . , an ) Pnk (K)0 is a zero of (S) but not of (S)sat . Then there
exists a polynomial F (S)sat which does not vanish at a. By definition, there
exists s 0 such that Ti F (S) for all monomials Ti of degree at least s. This
implies that Ti (a)F (a) = 0. By definition of homogeneous coordinates, one can
write 1 = a0 b0 + . . . + bn an for some bi . Raising this equality into the s-th
P power,
i
i
we obtain that T (a) generate the unit ideal. Thus we can write
P1 = i ci T for
some ci A, all zeros except finite many. This implies F (a) = ci T F (a) = 0.
Thus we may assume that (S) = (S)sat , (S 0 ) = (S 0 )sat . Take (t0 , . . . , tn )
Sol(S 0 , k[T ]/(S 0 )), where ti = Ti + (S 0 ). For every F = F (T0 , . . . , Tn ) (S 0 ),
we consider the polynomial F 0 = F (1, Z1 , . . . , Zn ) k[Z1 , . . . , Zn ], where
Zi = Ti /T0 . Let (S 0 )0 be the ideal in k[Z] generated by all polynomials
F 0 where F (S 0 ). Then (1, z1 , . . . , zn ) Sol(S 0 ; k[Z]/(S 0 )0 ) where zi =
Zi mod (S 0 )0 . By assumption, (1, z1 , . . . , zn ) Sol(S; k[Z]/(S 0 )0 ). This shows
that G(1, Z1 , . . . , Zn ) (S 0 )0 for each homogeneous generator of (S), i.e.
X
G(1, Z1 , . . . , Zn ) =
Qi Fi (1, Z1 , . . . , Zn )
i
T0
G(T0 , . . . , Tn ) (S 0 )
for some d(0). Similarly, we obtain that T d(i) G (S 0 ) for some d(i), i = 1, . . . , n.
This easily implies that ms+ G (S 0 ) for some large enough s (cf. the proof of
Proposition 5.3) . Hence, G (S 0 ) and (S) (S 0 ). Similarly, we obtain the
opposite inclusion.
Definition 5.7. A homogeneous ideal I k[T ] is said to be saturated if I = I sat .
Corollary 5.15. The map I P V (I) is a bijection between the set of saturated
homogeneous ideals in k[T] and the set of projective algebraic subvarieties of Pnk .
44
Lecture 6
B
ezout theorem and a group law
on a plane cubic curve
We begin with an example. Consider two concentric circles:
C : Z12 + Z22 = 1,
C 0 : Z12 + Z22 = 4.
Obviously, they have no common points in the affine plane A2 (K) no matter in
which algebra K we consider our points. However, they do have common points
at infinity. The precise meaning of this is the following. Let
C : T12 + T22 T02 = 0,
be the projective closures of these conics in the projective plane P2k , obtained
by
46LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
Theorem 6.1. There exists a homogeneous polynomial
Rn,m Z[A0 , . . . , An , B0 , . . . , Bm ]
of degree m + n satisfying the following property:
The system of algebraic equations in one unknown over a field k :
P (Z) = a0 Z n + . . . + an = 0, Q(Z) = b0 Z m + . . . + bm = 0
has a solution in a field extension K of k if and only if (a0 , . . . , an , b0 , . . . , bm )
is a k-solution of the equation
Rn,m = 0.
Proof. Define Rm,n to be equal to
A0 . . .
. . . . . .
0 ...
B0 . . .
. . . . . .
0 ...
An 0 . . . 0
... ... ... ...
0 A0 . . . An
Bm 0 . . . . . .
where the first m rows are occupied with the string (A0 , . . . , An ) and zeroes,
and the remaining n rows are occupied with the string (B0 , . . . , Bm ) and zeroes.
Assume K is a common solution of two polynomials P (Z) and Q(Z). Write
P (Z) = (Z )P1 (Z), Q(Z) = (Z )Q1 (Z)
where P1 (Z), Q1 (Z) K[Z] of degree n1 and m1, respectively. Multiplying
P1 (Z) by Q1 (Z), and Q(Z) by P1 (Z), we obtain
P (Z)Q1 (Z) Q(Z)P1 (Z) = 0.
(6.1)
This shows that the coefficients of Q1 (Z) and P1 (Z) (altogether we have n + m
of them) satisfy a system of n + m linear equations. The coefficient matrix of
this system can be easily computed, and we find it to be equal to the transpose
of the matrix
a0 . . . a n
0 ... ...
... ... ... ... ... ...
0 ...
0
a
.
.
.
a
0
n
b0 . . . bm 0 . . . . . . .
47
A solution can be found if and only if its determinant is equal to zero.
Obviously, this determinant is equal (up to a sign) to the value of Rn,m at
(a0 , . . . , an , b0 , . . . , bm ). Conversely, assume that the above determinant vanishes. Then we find a polynomial P1 (Z) of degree n 1 and a polynomial
Q1 (Z) of degree m1 satisfying (1). Both of them have coefficients in k. Let
be a root of P (Z) in some extension K of k. Then is a root of Q(Z)P1 (Z).
This implies that Z divides Q(Z) or P1 (Z). If it divides Q(Z), we found
a common root of P (Z) and Q(Z). If it divides P1 (Z), we replace P1 (Z) with
P1 (Z)/(Z ) and repeat the argument. Since P1 (Z) is of degree less than n,
we finally find a common root of P (Z) and Q(Z).
The polynomial Rn,m is called the resultant of order (n, m). For any two
polynomials P (Z) = a0 Z n + . . . + an and Q(Z) = b0 Z m + . . . + bm the value of
Rn,m at (a0 , . . . , an , b0 , . . . , bm ) is called the resultant of P (Z) and Q(Z), and
is denoted by Rn,m (P, Q).
A projective algebraic subvariety X of P2k given by an equation: F (T0 , T1 , T2 ) =
0, where F 6= 0 is a homogeneous polynomial of degree d will be called a plane
projective curve of degree d. If d = 1, we call it a line, if d = 2, we call it a plane
conic, plane projective cubic, plane quartic, plane quintic, plane sextic and so on.
We say that X is irreducible if its equation is given by an irreducible polynomial.
Theorem 6.2. (Bezout). Let
F (T0 , T1 , T2 ) = 0, G(T0 , T1 , T2 ) = 0
be two different plane irreducible projective curves of degree n and m, respectively, over a field k. For any algebraically closed field K containing k, the system
F = 0, G = 0 has exactly mn solutions in P2 (K) counted with appropriate multiplicities.
Proof. Since we are interested in solutions in an algebraically closed field K, we
may replace k by its algebraic closure to assume that k is algebraically closed. In
particular k is an infinite set. We shall deduce later from the theory of dimension
of algebraic varieties that there are only finitely many K-solutions of F = G = 0.
Thus we can always find a line T0 + bT1 + cT2 = 0 with coefficients in k that
has no K-solutions of F = G = 0. This is where we use the assumption that
k is infinite. Also choose a different line aT0 + T1 + dT2 = 0 with a 6= b such
that for any , K the line ( + a)T0 + (b + )T1 + (c + )T2 = 0 has
at most one solution of F = G = 0 in K. The set of triples (, , ) such that
48LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
the line T0 + T1 + T2 = 0 contains a given point (resp. two distinct points)
is a two-dimensional (resp. one-dimensional) linear subspace of k 3 . Thus the set
of lines T0 + T1 + T2 = 0 containing at least two solutions of F = G = 0
is a finite set. Thus we can always choose a line in k 3 containing (1, b, c) and
some other vector (a, 1, d) such that it does not belong to this set. Making the
invertible change of variables
T0 T0 + bT1 + cT2 , T1 aT0 + T1 + dT2 , T2 T2
we may assume that for every solution (a0 , a1 , a2 ) of F = G = 0 we have a0 6= 0,
and also that no line of the form T0 + T1 = 0 contains more than one solution
of F = G = 0 in K. Write
F = a0 T2n + . . . + an , G = b0 T2m + . . . + am ,
where ai , bi k[T0 , T1 ]i . Obviously, an , bm 6= 0, since otherwise T2 is a factor of
F or G. Let
R(A0 , . . . , An , B0 , . . . , Bm )
be the resultant of order (n, m). Plug ai in Ai , and bj in Bj , and let
= R(a0 , . . . , an , b0 , . . . , bm )
R
be the corresponding homogeneous polynomial in T0 , T1 . It is easy to see, using
is a homogeneous polynomial of degree
the definition of the determinant, that R
mn. It is not zero, since otherwise, by the previous Lemma, for every (0 , 1 )
the polynomials F (0 , 1 , T2 ) and G(0 , 1 , T2 ) have a common root in K. This
shows that P2 (K) contains infinitely many solutions of the equations F = G = 0,
which is impossible as we have explained earlier. Thus we may assume that
6= 0. Dehomogenizing it, we obtain:
R
= T nm R
0 (T1 /T0 )
R
0
0 is a polynomial of degree nm in the unknown Z = T1 /T0 . Assume
where R
0 is exactly mn. Let 1 , . . . , nm be its nm roots in the
first that the degree of R
) = 0,
algebraic closure k of k (some of them may be equal). Obviously, R(1,
hence
R(a0 (1, ), . . . , an (1, ), b0 (1, ), . . . , bm (1, )) = 0.
By Theorem 6.1, the polynomials in T2 F (1, , T2 ) and G(1, , T2 ) have a com It is also unique in view of our choice of the coordinate
mon root in k.
49
system. Thus (1, , ) is a solution of the homogeneous system F = G = 0 in
This shows that the system F = 0, G = 0 has nm solutions, the multiplicity
k.
0 = 0 has to be taken as the multiplicity of the correspondof a root of R
ing common solution. Conversely, every solution (0 , 1 , 2 ) of F = G = 0,
0 = 0. To complete the proof,
where 0 6= 0, defines a root = 1 /0 of R
0
is of degree d < nm. This happens
we have to consider the case where R
nmd
only if R(T0 , T1 ) = T0
P (T0 , T1 ), where P k[T0 , T1 ]d does not contain T0
1) = 0. Thus (0, 1, ) is a solution
as its irreducible factor. Obviously, R(0,
of F = G = 0 for some K. This contradicts our assumption from the
beginning of the proof.
Example 6.3. Fix an algebraically closed field K containing k. Assume that
m = 1, i.e.,
G = 0 T0 + 1 T1 + 2 T2 = 0
is a line. Without loss of generality, we may assume that 2 = 1. Computing
the resultant, we find that, in the notation of the previous proof,
0 , T1 ) = a0 (0 T0 + 1 T1 )n + . . . + an .
R(T
is obtained by eliminating the unknown T2 . We see that the line
Thus R
L : G = 0 intersects the curve X : F = 0 at n K-points corresponding to
0 , T1 ) = 0 in P1 (K). A solution is multiple, if
n solutions of the equation R(T
the corresponding root of the dehomogenized equation is multiple. Thus we can
speak about the multiplicity of a common K-point of L and F = 0 in P2 (K).
We say that a point x X(K) is a nonsingular point if there exists at most one
line L over K which intersects X at x with multiplicity > 1. A curve such that
all its points are nonsingular is called nonsingular. We say that L is tangent to
the curve X at a nonsingular point x P2 (K) if x L(K) X(K) and its
multiplicity 2. We say that a tangent line L is an inflection tangent line at x
if the multiplicity 3. If such a tangent line exists at a point x, we say that x
is an inflection point (or a flex point) of X.
Let P (Z1 , . . . , Zn ) k[Z1 , . . . , Zn ] be any polynomial in n variables with
P
of Z as follows.
coefficients in a field k. We define the partial derivatives Z
j
First we assume that P is a monomial Z1i1 Znin and set
(
i 1
ij Z1i1 Zjj Znin if ij > 0,
P
=
.
Zj
0
otherwise
50LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
Then we extend the definition to all polynomials by linearity over k requiring that
(aP + bQ)
P
Q
=a
+b
Zj
Zj
Zj
for all a, b k and any monomials P, Q. It is easy to check that the partial
derivatives enjoy the same properties as the partial derivatives of functions defined
P
is a derivation of the kby using the limits. For example, the map P 7 Z
j
algebra k[Z1 , . . . , Zn ], i.e. , it is a k-linear map satisfying the chain rule:
(P Q) = P (Q) + Q(P ).
The partial derivatives of higher order are defined by composing the operators of
partial derivatives.
Proposition 6.4. (i) X : F (T0 , T1 , T2 ) = 0 be a plane projective curve of degree d. A point (a0 , a1 , a2 ) X(K) is nonsingular if and only if (a0 , a1 , a2 )
is not a solution of the system of homogeneous equations
F
F
F
=
=
= 0.
T0
T1
T2
(ii) If (a0 , a1 , a2 ) is a nonsingular point, then the tangent line at this point is
given by the equation
2
X
F
(a0 , a1 , a2 )Ti = 0.
T
i
i=0
2F
T0 T1
2F
T12
2F
T2 T1
2F
T0 T2
2F
T1 T2 (a)
2F
T22
= 0.
Proof. We check these assertions only for the case (a0 , a1 , a2 ) = (1, 0, 0). The
general case is reduced to this case by using the variable change. The usual
formula for the variable change in partial derivatives are easily extended to our
51
algebraic partial derivatives. We leave the details of this reduction to the reader.
Write F as a polynomial in T0 with coefficients polynomials in T1 , T2 .
F (T0 , T1 , T2 ) = T0q Pdq (T1 , T2 )+T0q1 Pdq+1 (T1 , T2 )+ +Pd (T1 , T2 ),
q d.
Here the subscript indices coincides with the degree of the corresponding homogeneous polynomial if it is not zero and we assume that Pdq 6= 0. We assume
that F (1, 0, 0) = 0. This implies that q < d. A line through the point (1, 0, 0)
is defined by an equation T2 T1 = 0 for some k. Eliminating T2 we get
F (T0 , T1 , T1 ) = T0q T1dq Pdq (1, )+T0q1 T1dq+1 Pdq+1 (1, )+ +T1d Pd (1, )
dq
q
q1
q
= T1
T0 Pdq (1, ) + T0 T1 Pdq+1 (1, ) + + T1 Pd (1, ) .
It is clear that each line intersects the curve X at the point (1, 0, 0) with multiplicity > 1 if and only if d q > 1. Thus (1, 0, 0) is nonsingular if and only if
q = d 1. In this case we find that
F
F
F
(1, 0, 0) = 0,
(1, 0, 0) = a,
(1, 0, 0) = b,
T0
T1
T2
so both a and b cannot be zeros. On the other hand, if q < d 1, the same
computation shows that the partial derivatives vanish at (1, 0, 0). This proves
assertion (i). Assume that the point is nonsingular, i.e. d q = 1. The unique
tangent line satisfies the linear equation
P1 (1, ) = a + b = 0.
(6.2)
2F
T0 T1
2F
T12
2F
T2 T1
2F
T0 T2
2F
T1 T2 (1, 0, 0)
2F
T22
0
(d 1)a (d 1)b
2
= det (d 1)a
(d 1)b
52LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
Remark 6.5. The determinant
2F
T 2
2 F0
det T1 T0
2F
T1 T0
2F
T0 T1
2F
T12
2F
T2 T1
2F
T0 T2
2F
T1 T2
2F
T22
53
xy
o
xy
Figure 6.1:
line to X at e. If y = x, take for L1 the tangent line at y. We claim that this
construction defines the group law on X(K).
Clearly
y x = x y,
i.e., the binary law is commutative. The point e is the zero element of the law. If
x X(K), the opposite point x is the point of intersection of X(K) with the
line passing through x and the third point x1 at which the tangent at e intersects
the curve. The only non-trivial statement is the property of associativity.
Consider the eight points e, x, y, z, zy, xy, x y, y z. They lie on three
cubic curves. The first one is the original cubic X. The second one is the union
of three lines
< x, y > < yz, y z > < z, x y >
(6.3)
where for any two distinct points a, b P2 (K) we denote by < a, b > the unique
K-line L with a, b L(K). Also the union means that we are considering the
variety given by the product of the linear polynomials defining each line. The
third one is also the union of three lines
< y, z > < xy, x y > < x, y z > .
(6.4)
54LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
Proof. Let Y be given by an equation F = a0 T03 + a1 T02 T1 + . . . = 0 the
polynomial F . A point x = (0 , 1 , 2 ) X(K) if and only if the ten coefficients
of F satisfy a linear equation whose coefficients are the values of the monomials
of degree 3 at (0 , 1 , 2 ). The condition that a cubic curve passes through
8 points introduces 8 linear equations in 10 unknowns. The space of solutions
of this system is of dimension 2. Suppose that the dimension is exactly 2.
Then the equation of any cubic containing the points x1 , . . . , x8 can be written
in the form F1 + F2 , where F1 and F2 correspond to two linearly independent
solutions of the system. Let x9 be the ninth intersection point of F1 = 0
and F2 = 0 (Bezouts Theorem). Obviously, x9 is a solution of F = 0. It
remains to consider the case when the space of solutions of the system of linear
equation has dimension > 2. Let L be the line with x1 , x2 L(K). Choose two
points x, y L(K) \ {x1 , x2 } which are not in X(K). Since passing through
a point imposes one linear condition, we can find a cubic curve Y : G = 0
with x, y, x1 , . . . , x8 Y (K). But then L(K) Y (K) contains four points.
By Bezouts Theorem this could happen only if G is the product of a linear
polynomial defining L and a polynomial B of degree 2. By assumption L does
not contain any other point x3 , . . . , x8 . Then the conic C : B = 0 must contain
the points x3 , . . . , x8 . Repeating the argument for the points x1 , x3 , we find a
conic C 0 : B 0 = 0 which contains the points x2 , x4 , . . . , x8 . Clearly C 6= C 0
since otherwise C contains 7 common points with an irreducible cubic. Since
C(K) C 0 (K) contains 5 points in common, by Bezouts Theorem we obtain
that B and B 0 have a common linear factor. This easily implies that 4 points
among x4 , . . . , x8 must be on a line. But this line cannot intersect an irreducible
cubic at four points in P2k (K).
Here is an example of the configuration of 8 points which do not satisfy
the assumption of Lemma 6.6. Consider the cubic curve (over C) given by the
equation:
T03 + T13 + T23 + T0 T1 T2 = 0.
It is possible to choose the parameter such that the curve is irreducible. Let
x1 , . . . , x9 be the nine points on this curve with the coordinates:
(0, 1, ), (1, 0, ), (1, 1, )
where is one of three cube roots of 1. Each point xi lies on four lines
which contain two other points xj 6= xi . For example, (0, 1, 1) lies on the
line T0 = 0 which contains the points (0, 1, ), (0, 1, 2 ) and on the three lines
55
T0 T1 T2 = 0 which contains the points (1, 0, ), (1, , 0). The set x1 , . . . , x8
is the needed configuration. One easily checks that the nine points x1 , . . . , x9
are the inflection points of the cubic curve C (by Remark 6.5 we expect exactly 9
inflection points). The configuration of the 12 lines as above is called the Hesse
configuration of lines.
x1
x6
x2
x7
x8
x9
x3
x4
x5
x2
x1
Figure 6.2:
Nevertheless one can prove that the assertion of Lemma 6.6 is true without
additional assumption on the eight points.
To apply Lemma 6.6 we take the eight points e, x, y, z, zy, xy, x y, y z
in X(K). Obviously, they satisfy the assumptions of the lemma. Observe that
(x y)z lies in X(K) and also in the cubic (6.3), and x(y z) lies in X(K)
and in the cubic (6.4). By the Lemma (x y)z = x(y z) is the unique ninth
point. This immediately implies that (x y) z = x (y z).
Remark 6.7. Our proof is in fact not quite complete since we assumed that all the
points e, x, y, z, zy, xy, xy, yz are distinct. We shall complete it later but the
idea is simple. We will be able to consider the product X(K) X(K) X(K)
as a projective algebraic set with the Zariski topology. The subset of triples
(x, y, z) for which the associativity x (y z) = (x y) z holds is open
(since all degenerations are described by algebraic equations). On the other
hand it is also closed since the group law is defined by a polynomial map. Since
56LECTURE 6. BEZOUT
THEOREM AND A GROUP LAW ON A PLANE CUBIC CURVE
X(K) X(K) X(K) is an irreducible space, this open space must coincide
with the whole space.
Remark 6.8. Depending on K the structure of the group X(K) can be very
different. A famous theorem of Mordell-Weil says that this group is finitely
generated if K is a finite extension of Q. One of the most interesting problems
in number theory is to compute the rank of this group. On the other hand, the
group X(C) is isomorphic to the factor group C/Z2 . Obviously, it is not finitely
generated.
Problems.
1. Let P (Z) = a0 Z n + a1 Z n1 + . . . + an be a polynomial with coefficients
in a field k, and P 0 (Z) = na0 Z n1 + (n 1)a1 Z n2 + . . . + an be its derivative.
The resultant Rn,n1 (P, P 0 ) of P and P 0 is called the discriminant of P . Show
that the discriminant is equal to zero if and only if P (Z) has a multiple root in
the algebraic closure k of k. Compute the discriminant of quadratic and cubic
polynomials. Using computer compute the discriminant of a quartic polynomial.
2. Let P (Z) = a0 (Z 1 ) . . . (Z n ) and Q(x) = b0 (Z 1 ) . . . (Z m ) be
the factorizations of the two polynomials into linear factors (over an algebraic closure
of k). Show that
Rn,m (P, Q) =
n
am
0 b0
n Y
m
Y
i=1 i=1
(i j ) =
am
0
n
Y
i=1
Q(i ) =
(1)mn bn0
m
Y
P (j ).
j=1
3. Find explicit formulae for the group law on X(C), where X is a cubic curve
defined by the equation T12 T0 T23 T03 = 0. You may take for the zero element the
point (0, 1, 0).
4. In the notation of the previous problem, show that elements x X(C) of order
3 (i.e. 3x = 0 in the group law) correspond to inflection points of X. Show that
there are 9 of them. Show that the set of eight inflection points is an example of the
configuration which does not satisfy the assumption of Lemma 6.6.
5. Let X be given by the equation T12 T0 T23 = 0. Similarly to the case of a
nonsingular cubic, show that for any field K the set X(K)0 = X(K) \ {(1, 0, 0)} has
a group structure isomorphic to the additive group K + of the field K.
6. Let X be given by the equation T12 T0 T22 (T2 + T0 ) = 0. Similarly to the case
of a nonsingular cubic, show that for any field K the set X(K)0 = X(K) \ {(1, 0, 0)}
has a group structure isomorphic to the multiplicative group K of the field K.
Lecture 7
Morphisms of projective algebraic
varieties
Following the definition of a morphism of affine algebraic varieties we can define a
morphism f : X Y of two projective algebraic varieties as a set of maps fK :
X(K) Y (K) defined for each k-algebra K such that, for any homomorphism
: K L of k-algebras, the natural diagram
X(K)
X()
fK
Y (K)
Y ()
/ X(L)
(7.1)
fL
/ Y (L)
57
58
f : X Pnk and to check that fK (X(K)) Y (K) for each K. We know that
X(K) = Homkalg (O(X), K). Take K = O(X) and the identity homomorphism
idO(X) X(K). It is sent to an element M Pnk (O(X)). The projective O(X)module M completely determines f . In fact, let x X(K) and evx : O(X) K
be the corresponding homomorphism of k-algebras. Using the commutative diagram
(7.1) (where K = O(X), L = K, = evx ), we see that
fK (x) = M O(X) K,
(7.2)
(i)
(p0 , . . . , p(i)
n ) C(O(X)ai )n
(i)
Assume ai (x) = evx (ai ) 6= 0. Let xi be the image of x X(K) in X(Kai (x) )
under the natural homomorphism K Kai (x) . Let us consider Kai (x) as an O(X)evx
algebra by means of the composition of homomorphisms O(X)
K Kai (x) .
Then
fKai (x) (xi ) = M O(X) Kai (x)
= (M O(X) O(X)ai )O(X)ai Kai (x) = Mi O(X)ai Kai (x) ,
59
where Kai (x) is an O(X)ai -algebra by means of the homomorphism O(X)ai Kai (x)
defined by
a
ari
a(x)
ai (x)r .
(i)
(i)
n
fKai (x) (xi ) = (p0 (x), . . . , p(i)
n (x))Kai (x) P (Kai (x) ).
If K is a field, Kai (x) = K (because ai (x) 6= 0) and we see that, for any x X(K)
such that ai (x) 6= 0 we have
(i)
n
fK (x) = (p0 (x), . . . , p(i)
n (x)) P (K).
(7.3)
Thus we see that the morphism f : X Pnk is given by not a global polynomial
formula but by several local polynomial formulas (7.3). We take
P x X(K), find
i I such that ai (x) 6= 0 (we can always do it since 1 =
iI bi ai for some
bi O(X)) and then define fK (x) by formula (7.3).
The collection
(i)
{(p0 , . . . , pn(i) )}iI
(i)
(i)
(i)
(i)
(j)
(j)
(ii) for any i, j I, (p0 , . . . , pn ) = gij (p0 , . . . , pn ) in (O(X)ai aj )n+1 for some
invertible gij O(X)ai aj ;
(i)
(i)
(j)
pk = qk gij , k = 0, . . . , n,
where gij O(X)ai bj .
For each i I this collection defines a projective module Mi Pn (O(X)ai )
(i)
(i)
generated by (p0 , . . . , pn ). We shall prove in the next lemma that there exists
a projective module M Pn (O(X)) such that Mai
= Mi for each i I. This
module is defined uniquely up to isomorphism. Using M we can define f by sending
idO(X) X(O(X)) to M . If x X(K), where K is a field, the image fK (x) is
defined by formulae (7.3).
60
Let us now state and prove the lemma. Recall first that for any ring A a local line
M Pn (A) defines a collection {Mai }iI of lines in An+1
for some covering family
ai
{ai }iI of elements in A. Let us see how to reconstruct M from {Mai }iI . We know
that for any i, j I the images mi of m M in Mai satisfy the following condition
of compatibility:
ij (mi ) = ji (mj )
where ij : Mai Mai fj is the canonical homomorphism m/ari mfjr /(ai fj )r .
For any family {Mi }iI of Aai -modules let
Y
limindiI Mi = {(mi )I
Mi : ij (mi ) = ji (mj ) for any i, j I}.
iI
iI
Mi of
: M limindiI Mai
defined by m (mi = m)iI .
Lemma 7.1. The homomorphism
: M limindiI Mai
is an isomorphism.
Proof. We assume that the set of indices I is finite. This is enough for our applications
since we can always choose a finite covering subfamily. The proof of injectivity is similar
to the proof of Lemma 5.11 and is left to the reader. Let us show the surjectivity. Let
(
mi
)iI limindiI Mai
ani i
for some mi M and ni 0. Again we may assume that all ni are equal to some n.
Since for any i, j I
mj
mi
ij ( n ) = ji ( n ),
ai
aj
we have
(ai aj )r (anj mi ani mj ) = 0
for some r 0. Let pi = mi ari , k = r + n. Then
mi
pi
= k,
n
rai
ai
fjk pi = aki pj .
61
We can write 1 =
k
i bi ai .
Set m =
akj m =
i bi pi .
bi akj pi =
Then
bi aki pj = 1pj = pj .
This shows that the image of m in each Mai coincides with pi /aki = mi /ani for each
i I. This proves the surjectivity.
(i)
(i)
(i)
(i)
tj
= 0 because K0 = {0} and in the ring {0} one has 0 = 1. Thus it is enough
to define the maps X(K) Y (K) on the subsets X(K)00 of global K-lines with at
least one invertible projective coordinate.
Let X be defined by a homogeneous ideal I k[T0 , . . . , Tm ]. We denote by Ir the
ideal in the ring k[T0 /Tr , . . . , Tm /Tr ] obtained by dehomogenizations of polynomials
from I. Let Xr Am
k be the corresponding affine algebraic k-variety. We have
O(Xr ) = k[T0 /Tr , . . . , Tm /Tr ]/Ir . We have a natural map ir : Xr (K) X(K)00
obtained by the restriction of the natural inclusion map ir : K m Pm (K)00 (putting
1 at the rth spot). It is clear that each x X(K)00 belongs to the image of some
ir . Now to define the morphism X Y it suffices to define the morphisms fr :
Xr Y, r = 0, . . . , m. This we know how to do. Each fr is given by a collection
(s)
(s)
(s)
{(p0 , . . . , pn )}sS(r) , where each coordinate pj is an element of the ring O(X)r ),
(s)
(s)
62
(s)
(x)(r)
(r)
Fs
(s)
Lj Pj
+ I, plugging x in both
(r)
xr
Fs (x)(r)
is invertible, we obtain
X
(s)
=
Lj (x)Pj (x).
j
(s)
(s)
This shows that (P0 (x), . . . , Pn (x)) Cn (KFs (x) ) is satisfied. Note that this
definition is independent from the choice of projective coordinates of x. In fact, if we
(s)
(s)
multiply x by K , we get P0 (x) = k(s) P0 (x). Also Fs (x) will change to
d Fs (x) for some d 0, which gives the same localization KFs (x) .
Of course this representation is not defined uniquely in many ways. Also it must
be some compatibility condition, the result of our map is independent from which r we
take with the condition that xr K . As is easy to see this is achieved by requiring:
(s)
(s0 )
Pj Pk
(s)
(s0 )
Pk Pj
I
(s)
(s)
F (P0 , . . . , Pn(s) ) I
for any s S.
63
Then the formula:
a = (0 , . . . , m ) (F0 (a), . . . , Fn (a)), a X(K) Pm (K)0
defines a morphism f : X Y .
Proof. Observe that property (i) is needed in order the formula for the map is well
defined on orbits of K on C(K)n . We also have to check that (F0 (a), . . . , Fn (a))
C(K)n Y (K) for all k-algebras K. The functoriality (i.e. the commutativity of
the diagrams corresponding to homomorphisms K K 0 ) is clear. Let a] : k[T ]/I
K, Ti mod I i , be the homomorphisms defined by the point a. The composition
a] : k[T 0 ]/J K is defined by sending Tj0 mod J to Fj (a). Thus for any G J we
have G(F0 (a), . . . , Fn (a)) = 0. It remains to show that (F0 (a), . . . , Fn (a)) C(K)n .
Suppose the coordinates generate
a proper ideal a of K. By assumption, for some
P
s
s
s
s > 0, we can write
j Qj Fj + a, for some Qj k[T ]. Thus ai = Ti (a)
P Ti =
s
I. Writing 1 =
i bi ai , we obtain that 1 a. This contradiction shows that
(F0 (a), . . . , Fn (a)) C(K)n . This proves the assertion.
Example 7.3. Let : k[T0 , . . . , Tn ] k[T0 , . . . , Tn ] be an automorphism of the
polynomial algebra given by a linear homogeneous change of variables. More precisely:
(Ti ) =
n
X
aij Tj ,
i = 0, . . . , n
j=0
if a1 K ,
if a2 K
64
if a2 K .
Similarly, we check that the other composition of the functor morphisms is the identity.
Recall that the affine circle X is not isomorphic to the affine line A1k .
Example 7.5. A projective subvariety E of Pnk is said to be a projective subspace of
dimension d (or d-flat) if it is given by a system of linear homogeneous equations with
coefficients in k, whose set of solutions in k n+1 is a linear subspace E of k n+1 of
dimension d + 1. It follows from linear algebra that each such E can be given by a
homogeneous system of linear equations
L0 = 0, . . . , Lnd1 = 0.
Let X Pnk be such that
X(k) E(k) = .
Then the map
a 7 (L0 (a), . . . , Lnd1 (a)),
a X(K)0 ,
defines a morphism
pE : X Pnd1
k
which is said to be a linear projection from E. Let L be a projective subspace of
dimension n d 1 such that E(K) L(K) = . Then we can interpret the
composition pE : X Pnd1
Pnk as follows. Take a point x X(K)0 , find a
k
projective subspace E 0 Pnk of dimension d + 1 such that E 0 (K) contains E(K) and
x. Then
pE (x) = E 0 (K) L(K).
We leave this verification to the reader (this is a linear algebra exercise).
65
Example 7.6. We already know that P1k is isomorphic to a subvariety of P2k given
by
n+m
an equation of degree 2. This result can be generalized as follows. Let N = m 1.
Let us denote the projective coordinates in PN
k by
Ti = Ti0 ...in , i0 + . . . + in = |i| = m.
Choose some order in the set of multi-indices i with |i| = m. Consider the morphism
(the Veronese morphism of degree m)
vn,m : Pnk PN
k ,
defined by the collection of monomials (. . . , Ti , . . .) of degree m. Since Ti generate
an irrelevant ideal, we can apply Proposition 17.4, so this is indeed a morphism. For
any k-algebra K the corresponding map vn,m (K)0 : Pnk (K)0 PN
k (K) is defined by
i
the formula (a0 , . . . , an ) (. . . , T (a), . . .). The image of vn,m (K)0 is contained in
m
N
the set Verm
n (K), where Vern is the projective subvariety of Pk given by the following
system of homogeneous equations
{Ti Tj Tk Tt = 0}i+j=k+t .
It is called the m-fold Veronese variety of dimension n. We claim that the image
of vn,m (K) is equal to Verm
n (K) for all K, so that vn,m defines an isomorphism of
projective algebraic varieties:
vn,m : Pnk Verm
n.
00
To verify this it suffices to check that vn,m (K)(Pnk (K)00 ) = Verm
n (K) (compare with
00
the beginning of the lecture). It is easy to see that for every (. . . , ai , . . .) Verm
n (K)
at least one coordinate amei is invertible (ei is the i-th unit vector (0, . . . , 1, . . . 0)).
After reindexing, we may assume that ame1 6= 0. Then the inverse map is given by
the formula:
66
N = (n + 1)(m + 1) 1.
(7.4)
(7.5)
F ()
X(K)
X()
/ F(L)
/ X(L)
(7.6)
67
With this definition in mind we can say that the correspondence K Pn (K)
Pm (K) is a projective algebraic variety.
We leave to the reader to define the notions of a morphism and isomorphism
between projective algebraic k-varieties.
For example, one defines the projection morphisms:
n
p1 : Pnk Pm
k Pk ,
m
p2 : Pnk Pm
k Pk .
Tj
r(s0 )
Ti
Fs (T ) = 0,
Fs0 (T 0 ) = 0,
i, l = 0, . . . , n; j, k = 0, . . . , m,
0r(s)
r(s0 )
68
0i<j3
where the symmetric matrix (aij ) is nonsingular, is isomorphic to P1k P1k . Give an
explicit formula for the projection maps: pi : Q P1k .
3. Show that Vern1 is isomorphic to the projective closure of the affine curve given by
the equations {Zn Z1n = 0, . . . , Z2 Z12 = 0} (a rational normal curve of degree n).
Compare this with the Problem 6 of Lecture 5.
4. Show that the image of a linear projection of the twisted cubic curve in P3k from a
point not lying on this curve is isomorphic to a plane cubic curve. Find its equation
and show that this curve is singular in the sense of the previous lecture.
5. Show that the symmetric m-power S m (M ) of a projective module is a projective
module. Using this prove that the Veronese map vn,m is defined by the formula
M Symm (M ).
00
6. a) Show that Pn (K)00 Pm
k (K) is naturally bijectively equivalent to the set of
(n + 1) (n + 1) matrices of rank 1 with coefficients in K defined up to multiplication
by a nonzero scalar.
b) Show that Ver2n (K)00 is naturally bijectively equivalent to the set of symmetric rank
1 square matrices of size n + 1 with coefficients in K defined up to multiplication by
a nonzero scalar.
7. Construct a morphism from P1k to the curve X equal to the projective closure of
the affine curve (Z12 + Z22 )2 Z2 (3Z12 Z22 )) A2k . Is X isomorphic to P1k ?
Lecture 8
Quasi-projective algebraic sets
Let k be a field and K be an algebraically closed field containing k as a subfield.
Definition 8.1. A projective algebraic set over k (or a projective algebraic k-set) is a
subset V of Pn (K) such that there exists a projective algebraic variety X over k with
X(K) = V .
The variety X with X(K) = V 6= is not defined uniquely by V . However, as
follows from the Nullstellensatz
X(K) = Y (K) rad(I(X)) = rad(I(Y )).
Thus, if we require that X is given by a radical homogeneous ideal, the variety X is
determined uniquely by the set X(K). In the following we will always assume this.
Note that a radical homogeneous ideal I coincides with its saturation I sat . Indeed, if
ms F I for some s and F k[T ]d then all monomials entering into F belong to md .
In particular, F s mds ms and F s F = F s+1 I. Since I is radical this implies
that F I. In fact we have shown that, for any ideal I, we have
I I sat rad(I).
This, if I = rad(I), then I = I sat . Since a projective algebraic k-variety is uniquely
determined by a saturated homogeneous ideal, we see that there is a bijective correspondence between projective algebraic k-sets and projective algebraic k-varieties
defined by a radical homogeneous ideal (they are called reduced projective algebraic
k-varieties).
We can consider Pn (K) as a projective algebraic set over any subfield k of K.
Any projective algebraic k-subset of Pn (K) is called a closed subset of Pn (K). The
reason for this definition is explained by the following lemma.
69
70
Proposition 8.1. There exists a unique topology on the set Pn (K) whose closed
subsets are projective algebraic k-subsets of Pnk (K).
Proof. This is proven similarly to that in the affine case and we omit the proof.
The topology on Pn (K) whose closed sets are projective algebraic subsets is said
to be the Zariski k-topology. We will denote the corresponding topological space by
Pnk (K). As is in the affine case we will drop k from the definitions and the notations
if k = K. Every subset of Pnk (K) will be considered as a topological subspace with
respect to the induced Zariski k-topology.
Lemma-Definition 8.2. A subset V of a topological space X is said to be locally
closed if one the following equivalent properties holds:
(i) V = U Z, where U is open and Z is closed;
(ii) V is an open subset of a closed subset of X;
(iii) V = Z1 \ Z2 , where Z1 and Z2 are closed subsets of X.
Proof. Left to the reader.
Definition 8.2. A locally closed subset subset of Pnk (K) is called a quasi-projective
algebraic k-set.
In other words, a quasi-projective k-subset of Pn (K) is obtained by taking the set
of K-solutions of a homogeneous system of algebraic equations over k and throwing
away a subset of the solutions satisfying some additional equations.
An example of an open quasi-projective subset is the subset
Pn (K)i = {(a0 , . . . , an ) Pn (K) : ai 6= 0}.
Its complement is the coordinate hyperplane:
Hi = {(a0 , . . . , an ) Pn (K) : ai = 0}.
Every affine algebraic k-set V Ank (K) can be naturally considered as a quasiprojective algebraic set. We view An (K) = K n as the open subset Pn (K)0 , then note
that V = V Pn (K)0 , where V is the closure of V defined by the homogenization of
the ideal defining V . It is clear that, in general V is neither open nor closed subset
of Pn (K). Also observe that V equals the closure in the sense of topology, i.e., the
minimal closed subset of Pnk (K) which contains V .
Next, we want to define regular maps between quasi-projective algebraic sets.
71
Definition 8.3. A map f : V W Pm (K) of quasi-projective algebraic k-sets is
called regular if there exists a finite open cover V = i Ui such that the restriction of
f to each open subset Ui is given by a formula:
(i)
(i)
x (F0 (x), . . . , Fm
(x)),
(i)
(i)
72
with a regular function in the new definition given by polynomials (T0r , P (T0 , . . . , Tn )).
This defines a homomorphism O(V )0 O(V ). Its injectivity is obvious. Let us show
that this homomorphism is surjective. Let V be given by a system of equations
Fs (Z1 , . . . , Zn ) = 0, s S, and f O(V ) and {Ui }iI be an open cover of V such
that there exist homogeneous polynomials Pi (T0 , . . . , Tn ), Qi (T0 , . . . , Tn ) of the same
degree di for which
fi (x) = Pi (x)/Qi (x), Qi (x) 6= 0 for all x Ui .
Let Qi (Z)0 , Pi (Z)0 denote the dehomogenized polynomials. We have
Qi (x)0 f (x) = Pi (x)0 ,
i I, x Ui .
Ai Qi (Z)0 +
Bs Fs (Z)
(8.1)
X
i
X
i
X
Ai (x)Pi (x)0 = (
Ai Pi0 )(x).
i
73
Let V be a closed subset of Pn (K) defined by an irreducible homogeneous polynomial F of degree m > 1. The complement set U = Pn (K) \ V does not come from
any closed subset of Pn (K)i since V does not contain any hyperplane Ti = 0. So,
U is not affine in the way we consider any affine set as a quasi-projective algebraic
set. However, U is affine. In fact, let vn,m : Pn (K) PN (n,m) be the Veronese map
defined by monomials of degree m. Then vn,m (U ) is contained in the complement of
a hyperplane H in PN (n,m) defined by considering F as a linear combination of monomials. composing vn,m with a projective linear transformation we may assume that H
is a coordinate hyperplane. Thus vn,m defines an isomorphism from U to the open
subset of the Veronese projective algebraic set V ern,m (K) = vn,m (Pn (K)) whose
complement is the closed subset V ern,m (K) H. But this set is obviously affine, it is
defined in PN (n,m) (K)i = K N (n,m) by dehomogenizations of the polynomials defining
Vern,m .
Lemma 8.5. Let V be an affine algebraic k-set and f O(V ). Then the set
D(f ) = {x V : f (x) 6= 0}
is affine and
O(D(f ))
= O(V )f .
Proof. Replacing V by an isomorphic algebraic k-set, we may assume that V = X(K),
where X K n is an affine algebraic k-variety defined by an ideal I. Let F
k[Z1 , . . . , Zn ] be a polynomial representing f . Consider the closed subset of K n+1 =
K n K defined by the equation F Zn+1 1 = 0 and let V 0 be its intersection with
the closed subset V K. It is an affine algebraic k-set. We have
1
O(V 0 ) = k[Z1 , . . . , Zn , Zn+1 ]/(I, F Zn+1 1)
= k[Z1 , . . . , Zn ]/(I)[ ] = O(V )f .
f
Let p : K n+1 K n be the projection. I claim that the restriction of p to V 0 defines
an isomorphism p0 : V 0 D(f ). It is obviously a regular map, since it is defined
by the polynomials (Z1 , . . . , Zn ). The inverse map p1 : V V 0 is defined by the
1
map x 7 (x, f (x)
). Let us see that it is a regular map. Let P (T0 , . . . , Tn ) be a
homogenization of F , i.e., F =
P
T0d
of Pn+1 (K)0 and D(f ) as a locally closed subset of Pnk (K)0 . Obviously, the map p1
coincides with the map
x = (1, x1 , . . . , xn ) 7 (P T0 (x), P T1 (x), . . . , P Tn (x), T0d+1 (x)) = (1, x1 , . . . , xn ,
defined by homogeneous polynomials (P T0 , P T1 , . . . , P Tn , T0d+1 ) of degree d + 1.
1
).
f (x)
74
75
Corollary 8.10. Every algebraic set can be written uniquely as the union of finitely
many irreducible subspaces Zi , such that Zi 6 Zj for any i 6= j.
Lemma 8.11. Let V be a topological space and Z be its subspace. Then Z is
irreducible if and only if its closure Z is irreducible.
Proof. Obviously, follows from the definition.
Proposition 8.12. A subspace Z of Pnk (K) is irreducible if and only if the radical
homogeneous ideal defining the closure of Z is prime.
Proof. By the previous lemma, we may assume that Z is closed. Then Z is a projective
algebraic set defined by its radical homogeneous ideal. The assertion is proven similarly
to the analogous assertion for an affine algebraic set. We leave the proof to the
reader.
Problems.
1. Is the set {(a, b, c) P2 (K) : a 6= 0, b 6= 0} {(1, 0, 0)} quasi-projective?
2. Let V be a quasi-projective algebraic set in Pn (K), W be a quasi-projective algebraic
set in Pr (K). Prove that sn,m (K)(V W ) is a quasi-projective algebraic subset of
Segn,m (K) = sn,m (K)(Pn (K) Pr (K)) P(n+1)(m+1)1 (K).
3. Let us identify the product V W Pn (K) Pr (K) of two quasi-projective
algebraic k-sets with a quasi-projective algebraic k-subset of the Segre set Segn,m (K).
Let f : V V 0 and g : W W 0 be two regular maps. Show that the map
f g : V W V 0 W 0 is a regular map.
4. Is the union (resp. the intersection) of quasi-projective algebraic sets a quasiprojective algebraic set?
5. Find the irreducible components of the projective subset of P3 (K) given by the
equations: T2 T0 T12 = 0, T1 T3 T22 = 0.
6. Show that every irreducible component of a projective hypersurface V (F ) = {a
Pn (K) : F (a) = 0} is a hypersurface V (G), where G is an irreducible factor of the
homogeneous polynomial F (T ).
7. Describe explicitly (by equations) a closed subset of some K n which is isomorphic
to the complement to a conic T0 T1 + T22 = 0 in P2 (K).
8. Prove that a regular map is a continuous map and that the composition of regular
maps is a regular map.
76
Lecture 9
The image of a projective
algebraic set
Let f : V W be a regular map of quasi-projective k-sets. We are interested in its
image f (V ). Is it a quasi-projective algebraic set? For instance, let f : K 2 K 2 be
given by (x, y) 7 (x, xy). Then its image is the union of the set U = {(a, b) K 2 :
a 6= 0} and the closed subset Z = {(0, 0)}. The complement of a a locally closed
subset is equal to the union of an open a closed set. If the open part is not empty,
then closure must be equal to K 2 . The complement of f (K 2 ) is equal to the set of
points (0, y), y 6= 0. Obviously, its closure is the line y = 0 and it does not contain
any open subsets of K n . Thus f (A2k (K)) is not locally closed in A2k (K). Since K 2
is an open subset of P2k (K), f (A2k (K)) is not locally closed in P2k (K), i.e., it is not a
quasi-projective algebraic set.
However, the situation is much better in the case where V is a projective set. We
will prove the following result:
Theorem 9.1. The image of a projective algebraic k-set V under a regular map
f : V W is a closed subset of W in the Zariski k-topology.
To prove this theorem we note first that
f (V ) = pr2 (f )
where
f = {(x, y) V W : y = f (x)}
is the graph of f , and pr2 : V W W, (x, y) 7 y is the projection map. We will
always consider the product V W as a quasi-projective set by embedding it into a
projective space by the Segre map. In particular, V W is a topological space with
respect to the Zariski topology.
77
78
79
To prove Theorem 9.3 we need the following:
n
m
Lemma 9.5. Let V be a closed subset of Pnk (K)Pm
k (K) (resp. of Pk (K)Ak (K)).
0
0
0 ],
Then V is the set of zeroes of polynomials Ps (T, T ) k[T0 , . . . , Tn , T0 , . . . , Tm
s S, which are homogeneous of degree d(s) in variables T0 , . . . , Tn and homoge0 (resp. V is the set of zeroes of
neous of degree d(s)0 in the variables T00 , . . . , Tm
0
0
0 ], s S, which
polynomials Ps (T0 , . . . , Tn , Z1 , . . . , Zm ) k[T0 , , . . . , Tn , Z10 , . . . , Zm
are homogeneous of degree d(s) in variables T0 , . . . , Tn ). Conversely, every subset of
n
m
Pnk (K) Pm
k (K) (resp. of Pk (K) Ak (K)) defined in this way is a closed subset in
the Zariski k-topology of the product.
Proof. It is enough to prove the first statement. The second one will follow from
the first one by taking the closure of V in Pnk (K) Pm
k (K) and then applying the
0
dehomogenization process in the variable T0 . Now we know that V is given by a system
(n+1)(m+1)1
of homogeneous polynomials in variables Tij in the space Pk
and the system
of equations defining the Segre set Segn,m (K). Using the substitution Tij = Ti Tj0 ,
0 which
we see that V can be given by a system of equations in T0 , . . . , Tn , T00 , . . . , Tm
are homogeneous in each set of variables of the same degree. If we have a system
0 ) which are homogeneous of degree d(s) in
of polynomials Ps (T0 , . . . , Tn , T00 , . . . , Tm
0 , its set
variables T0 , . . . , Tn and homogeneous of degree d(s)0 in variables T00 , . . . , Tm
of solutions in Pnk (K) Pm
k (K) is also given by the system in which we replace each
0d(s)d(s)0
d(s)0 d(s)
Ps by Ti
Ps , i = 0, . . . , m, if d(s) > d(s)0 and by Ti
Ps , i = 0, . . . , n,
if d(s) < d(s)0 . Then the enlarged system arises from a system of polynomials in Tij
after substitution Tij = Ti Tj0 .
Now let us prove Theorem 9.3. Let V be a closed subset of Pnk (K). Then
Z V W is a closed subset of Pn (K) W and pr2 (Z) equals the image of Z under
the projection Pnk (K) W W . Thus we may assume that V = Pnk (K).
Let W = iI Ui be a finite affine covering of W (i.e. a covering by open affine
sets). Then V W = iI (V Ui ), Z = iI Z (V Ui ) and pr2 (Z) = iI pr2 (Z
(V Ui )). This shows that it suffices to check that pr2 (Z) (V Ui ) is closed in
Ui . Thus we may assume that W = Ui is affine. Then W is isomorphic to a closed
m
subset of some Am
k (K), V W is closed in V Ak (K) and pr2 (Z) is equal to the
m
image of Z under the second projection V Ak Am
k . Thus we may assume that
m
n
W = Ak (K) and V = Pk (K).
Let Z be a closed subset of Pnk (K) Am
k (K). By Lemma 9.5, Z can be given by
a system of equations
Fi (T0 , . . . , Tn , t1 , . . . , tm ) = 0, i = 1, . . . , N.
where Fi k[T0 , . . . , Tn , t1 , . . . , tm ] is a homogeneous of degree d(i) in variables
T0 , . . . , Tn . For every a = (a1 , . . . , am ) K m , we denote by Xa the projective
80
for any s 0}
= s0 {a K m : (T0 , . . . , Tn )s 6 Ia }.
Thus it suffices to show that each set Ys = {a K m : (T0 , . . . , Tn )s 6 Ia } is
s
closed in the Zariski k-topology. Note that (T0 , . . . , TP
n ) Ia means that every
homogeneous polynomial of degree s can be written as i , Fi (T, a)Qi (T )) for some
Qi (T ) k[T ]sd(i) , where d(i) = degFi (T, a). Consider the linear map of linear
k-spaces
:
N
M
i=1
(Q1 , . . . , QN ) 7
81
number of connected components of the corresponding K-set equals the number of
distinct roots of F (Z) in K.
Corollary 9.6. Assume k is a perfect field. Let V be a projective algebraic k-set,
n = #0 (V ). Then there is an isomorphism of k-algebras O(V )
= k1 . . . kn
where each ki is a finite field extension of k. Moreover
n
X
[ki : k] = #
0 (V ).
i
82
Lecture 10
Finite regular maps
The notion of a finite regular map of algebraic sets generalizes the notion of a finite
extension of fields. Recall that an extension of fields F E is called finite if E is
a finite-dimensional vector space over F . This is easy to generalize. We say that an
injective homomorphism : A B of commutative rings is finite if B considered as
a module over A via the homomorphism is finitely generated. What is the geometric
meaning of this definition? Recall that a finite extension of fields is an algebraic
extension. This means that any element in E satisfies an algebraic equation with
coefficients in F . The converse is also true provided E is finitely generated over F as
a field. We shall prove in the next lemma that a finite extension of rings has a similar
property: any element in B satisfies an algebraic equation with coefficients in (A).
Also the converse is true if we additionally require that B is a finitely generated algebra
over A and every element satisfies a monic equation (i.e. with the highest coefficient
equal to 1) with coefficients in (A).
Let us explain the geometric meaning of the additional assumption that the equations are monic. Recall that an algebraic extension E/F has the following property.
Let y : F K be a homomorphism of F to an algebraically closed field K. Then
y extends to a homomorphism of fields x : E K. Moreover, the number of these
extensions is finite and is equal to the separable degree [E : F ]s of the extension
E/F . An analog of this property for ring extensions must be the following. For any
algebraically closed field K which has a structure of a A-algebra via a homomorphism
y : A K (this is our analog of an extension K/F ) there a non-empty finite set of
homomorphisms xi : B K such that xi = y. Let us interpret this geometrically
in the case when is a homomorphism of finitely generated k-algebras. Let X and Y
be afiine algebraic k-varieties such that O(X)
= B, O(Y )
= A. The homomorphism
83
84
X such that fK (xi ) = y. Thus the analog of the extension property is the property
that the map X(K) Y (K) is surjective and has finite fibres. Let B is generated
over A by one element b satisfying an algebraic equation
a0 xn + a1 xn1 + . . . + an = 0
with coefficients in A. Assume the ideal I = (a0 , . . . , an1 ) is proper but an is
invertible in A. Let m be a maximal ideal in A containing I. Let K be an algebraically
closed field containing the residue field A/m. Consider the K-point of Y corresponding
to the homomorphism y : A A/m K. Since B
= A[x]/(a0 xn + a1 xn1 + . . . +
an ), any homomorphism extending y must send an to zero but this is impossible since
an is invertible. Other bad thing may happen if an I. Then we obtain infinitely
many extensions of y, they are defined by sending x to any element in K. It turns out
that requiring that a0 is invertible will guarantee that X(K) Y (K) is surjective
with finite fibres.
We start with reviewing some facts from commutative algebra.
Definition 10.1. A commutative algebra B over a commutative ring A is said to be
integral over A if every element x B is integral over A (i.e. satisfies an equation
xn + a1 xn1 + . . . + an = 0 with ai A).
Lemma 10.1. Assume that B is a finitely generated A-algebra. Then B is integral
over A if and only if B is a finitely generated module over A.
Proof. Assume B is integral over A. Let x1 , . . . , xn be generators of B as an A-algebra
(i.e., for any b B there exists F A[Z1 , . . . , Zn ] such that b = F (x1 , . . . , xn )).
n(i)
Since each xi is integral over A, there exists some integer n(i) such that xi can be
written as a linear combination of lower powers of xi with coefficients in A. Hence
every power of xi can be expressed as a linear combination of powers of xi of degree
less than n(i). Thus there exists a number N > 0 such that every b B can be
written as a polynomial in x1 , . . . , xn of degree < N . This shows that a finite set of
monomials in x1 , . . . , xn generate B as an A-module.
Conversely, assume that B is a finitely generated A-module. Then every b B
can be written as a linear combination b = a1 b1 + . . . + ar br , where b1 , . . . , br is a
fixed set of elements in B and ai A. Multiplying the both sides by bi and expressing
each product bi bj as a linear combination of bi s we get
X
bbi =
aij bi , aij A.
(10.1)
j
This shows that the vector b = (b1 , . . . , br ) satisfies the linear equation (M bIn )b =
0, where M = (aij ). Let D = det(M bIn ). Applying Cramers rule, we obtain that
85
Dbi = 0, i = 1, . . . , n. Using (10.1) we see that Dx = 0 for all x B. In particular,
D D = D2 = 0. It remains to use that the equation D2 = 0 is a monic equation for
b with coefficients in A.
This Lemma implies the following result which we promised to prove in Lecture 2:
Corollary 10.2. Let B be an A-algebra. The set of elements in B which are integral
over A is a subring of B (it is called the integral closure of A in B).
Proof. Let b, b0 B be integral over A. Consider the A-subalgebra A[b, b0 ] of B
generated by these elements. Since b is integral over A, it satisfies an equation bn +
a1 bn1 + . . . + an , ai A, hence A[b] is a finitely generated A-module generated by
1, . . . , bn1 . Similarly, since b0 is integral over A, hence over A[b], we get A[b, b0 ] =
A[b][b0 ] is a finitely generated A[b]-module. But then A[b, b0 ] is a finitely generated
A-module. By Lemma 10.1, A[b, b0 ] is integral over A. This checks that b + b0 , b b0
are integral over A.
Lemma 10.3. Let B be integral over its subring A. The following assertions are true:
(i) if A is a field and B is without zero divisors, then B is a field;
(ii) if I is an ideal of B such that I A = {0} and B is without zero divisors then
I = {0};
(iii) if P1 P2 are two ideals of B with P1 A = P2 A and P1 is prime, then
P1 = P2 ;
(iv) if S is a multiplicatively closed subset of A, then the natural homomorphism
AS BS makes BS an integral algebra over AS ;
(v) if I is a proper ideal of A then the ideal IB of B generated by I is proper;
(vi) for every prime ideal P in A there exists a prime ideal P 0 of B such that
P 0 A = P.
Proof. (i) Every x satisfies an equation xn + a1 xn1 + . . . + an = 0 with ai
A. Since B has no zero divisors, we may assume that an 6= 0 if x 6= 0. Then
x(xn1 + a1 xn2 + . . . + an1 )(an1 ) = 1. Hence x is invertible.
(ii) As in (i), we may assume that every nonzero x I satisfies an equation
xn + a1 xn1 + . . . + an = 0 with ai A and an 6= 0. Then an = x(xn1 + a1 xn2 +
. . . + an1 ) I A. Since I A = {0}, we obtain an = 0. Thus I has no nonzero
elements.
(iii) Let P0 = P1 A. Then we may identify A = A/P0 with a subring of
B = B/P1 with respect to the natural homomorphism A/P0 B/P1 . Let P20 be the
86
87
Example 10.4. 1. Let X = {(x, y) K 2 : y = x2 } A2 (K) and Y = A1 (K).
Consider the projection map f : X Y, (x, y) 7 y. Then f is finite. Indeed, O(X)
=
k[Z1 , Z2 ]/(Z2 Z12 ), O(Y )
= k[Z2 ] and f is the composition of the natural inclusion
k[Z2 ] k[Z1 , Z2 ] and the natural homomorphism k[Z1 , Z2 ] k[Z1 , Z2 ]/(Z2 Z12 ).
Obviously, it is injective. Let z1 , z2 be the images of Z1 and Z2 in the factor ring
k[Z1 , Z2 ]/(Z2 Z12 ). Then O(X) is generated over f (O(Y )) by one element z1 .
The latter satisfies a monic equation: z12 f (Z2 ) = 0 with coefficients in f (O(Y )).
As we saw in the proof of Lemma 10.1, this implies that O(X) is a finitely generated
f (O(Y ))-module and hence O(X) is integral over f (O(Y )). Therefore f is a finite
map.
2. Let x0 be a projective subspace of Pnk (K) of dimension 0, i.e., a point (a0 , . . . , an )
with coordinates in k. Let X be a projective algebraic k-set in Pnk (K) with x0 6 X
and let f = prx0 : X Pn1
(K) be the projection map. We know that Y = f (X)
k
is a projective set. Let us see that f : X Y is finite. First, by a variable change,
we may assume that x0 is given by a system of equations T0 = . . . = Tn1 = 0
where T0 , . . . , Tn are homogeneous coordinates. Then f is given by (x0 , . . . , xn ) 7
(x0 , . . . , xn1 ). We may assume that y Y lies in the open subset V = Y Pkn1 (K)0
where x0 6= 0. Its preimage U = f 1 (V ) = X Pnk (K)0 . Since f is surjective
f : O(V ) O(U ) is injective. Let us show that O(U ) is integral over f (O(V )).
Let I0 k[Z1 , . . . , Zn ] be the ideal of X Pnk (k)0 , where Zi = Ti /T0 , i = 1, . . . , n.
Then V is given by some ideal J0 in k[Z1 , . . . , Zn1 ], and the homomorphism f
is induced by the natural inclusion k[Z1 , . . . , Zn1 ] k[Z1 , . . . , Zn ]. Since O(U ) is
generated over k by the cosets zj of Zj modulo the ideal I0 we may take zn to be
a generator of O(U ) over f (O(V )). Let {Fs (T ) = 0}sS be the equations defining
X. Since x0 6 X, the ideal generated by the polynomials Fs and Ti , i n 1, must
contain k[T ]d for some d 0. Thus we can write
Tnd =
X
sS
As Fs +
n1
X
Bi Ti
i=0
88
f f ,
is bijective.
Proof. We know this already if X and Y are both affine. Let U be an affine open
subset of X. By restriction of maps (resp. functions), we obtain a commutative
diagram:
/ Homkalg (O(Y ), O(X))
Mapreg (X, Y )
Mapreg (U, Y )
/ Homkalg (O(Y ), O(U )).
Here the bottom horizontal arrow is a bijection. Thus we can invert the upper horizontal arrow as follows. Pick up an open affine cover {Ui }iI of X. Take a homomorphism
: O(Y ) O(X), its image in Homkalg (O(Y ), O(Ui )) is the composition with the
restriction map O(X) O(Ui ). It defines a regular map Ui Y . Since a regular
map is defined on its open cover, we can reconstruct a global map X Y . It is
easy to see that this is the needed inverse.
Lemma 10.7. Let X be a quasi-projective algebraic k-set. Then X is affine if and
only if O(X) is a finitely generated k-algebra which contains a finite set of elements
i which generate the unit ideal and such that each D(i ) is affine.
89
Proof. The part only if is obvious. Let 1 , . . . , n O(X) which generate the unit
ideal. Then X = i D(i ). Let k[Z1 , . . . , Zn ] O(X) be a surjective homomorphism
of k-algebras and I be its kernel. The set of zeroes of I in An (K) is an affine algebraic
set X 0 with O(X 0 )
= O(X). Let f : X X 0 be the regular map corresponding by
Lemma 10.6 to the previous isomorphism. Its restriction to D(i ) is an isomorphism
for each i (here we use that D(i ) is affine). Hence f is an isomorphism.
Proposition 10.8. Let f : X Y be a finite regular map of quasi-projective algebraic
k-sets. The following assertions are true:
(i) for every affine open subset U of Y, f 1 (U ) is affine and f : f 1 (U ) U is
finite;
(ii) if Z is a locally closed subset of Y , then f : f 1 (Z) Z is finite;
(iii) if f : X Y and g : Y Z are finite regular maps, then g f : X Z is a
finite regular map.
Proof. (i) Obviously, we may assume that Y = U is affine. For any y Y , there exists
an open affine neighborhood V of y such that f : f 1 (V ) V is a finite map of
affine k-sets. Let O(V ), then D() V is affine and f 1 (D()) = D(f ())
f 1 (V ) is affine. Moreover, the map f 1 (D()) D() is finite (this follows from
Lemma 10.3 (iv) and Lemma 10.5). Thus we may assume that Y is covered by affine
open sets of the form D() such that f 1 (D()) is affine and the restriction of the
map f to f 1 (D()) is finite.
Now let
Y = i Vi , Vi = D(i ), i O(Y ),
X = i Ui , Ui = f 1 (Vi ) = D(f (i )),
fi = f |Ui : Ui Vi
By Lemma 10.1, O(Ui ) is a finitely generated O(Vi )-module. Let {ij }j=1,...,n(i) be
a set of generators of this module. Since ij = a/f (i )n for some a O(X) and
n 0, and f (i ) is invertible in O(Ui ), we may assume that ij O(X). For every
O(X) we may write
|Ui =
n(i)
X
j=0
n(i)
90
and
=
X
i
f (hi )f (i )n(i) =
f (hi )bj ij .
i,j
This shows that = ij cij ij for some cij O(X), that is, {ij } is a generating
set of the f (O(Y ))-module O(X). In particular, O(X) is integral over f (O(Y ))
and O(X) is an algebra of finite type over k. Since the elements f (i )n(i) generate
the unit ideal in O(X), applying by Lemma 10.7, we obtain that X is an affine set .
(ii) Let Z be a locally closed subset of Y . Then Z = U Z 0 , where U is open and
0
Z is closed in Y . Taking an affine open cover of U and applying (i), we may assume
that Y = U is affine and Z is a closed subset of Y . Then f 1 (Z) is closed in X.
Since X is affine f 1 (Z) is affine. The restriction of f to f 1 (Z) is a regular map f :
f 1 (Z) Z of affine sets corresponding to the homomorphism of the factor-algebras
f : O(Y )/I(Z) O(X)/I(f 1 (Z)). Since I(f 1 (Z)) = f (I(Z))O(X), f is
injective. By Lemma 10.3, the corresponding extension of algebras is integral. Thus
f is finite.
(iii) Applying (i), we reduce the proof to the case where X, Y and Z are affine. By
Lemma 10.1, O(X) is finite over f (O(Y )) and f (O(Y )) is finite over f (g (O(Z))) =
(g f ) (O(Z)). Thus O(X) is finite over (g f ) (O(Z)), hence integral over
(g f ) (O(Z)).
Proposition 10.9. Let f : X Y be a finite regular map of algebraic k-sets. Then
(i) f is surjective;
(ii) for any y Y , the fibre f 1 (y) is a finite set.
Proof. Clearly, we may assume that X and Y are affine, B = O(X) is integral
over A = O(Y ) and = f is injective. A point y Y defines a homomorphism
evy : A K whose kernel is a prime ideal p. A point x f 1 (y) corresponds
to a homomorphism evx : B K of k-algebras such that its composition with
is equal to evy . By Lemma 10.3 (vi), there exists a prime ideal P in B such
that 1 (P) = p. Let Q(B/P) be the field of fractions of the quotient ring B/P
and Q(A/p) be the field of fractions of the ring A/p. Since B is integral over A,
the homomorphism defines an algebraic extension Q(B/P)/Q(A/p) (Lemma 10.3
(iv)). Since K is algebraically closed, there exists a homomorphism Q(B/P) K
which extends the natural homomorphism Q(A/p) K defined by the injective
homomorphism A/p K induced by evy . The composition of the restriction of the
homomorphism Q(B/P) K to B/P and the factor map B B/P defines a point
x f 1 (y). This proves the surjectivity of f .
Note that the field extension Q(B/P)/Q(A/p) is finite (since it is algebraic and
Q(B/P) is a finitely generated algebra over Q(A/p). It is known from the theory
91
of field extensions that the number of homomorphisms Q(B/P) K extending the
homomorphism A/p K is equal to the separable degree [Q(B/P) : Q(A/p)]s of
the extension Q(B/P)/Q(A/(p)). It follows from the previous arguments that the
number of points in f 1 (y) is equal to the sum
X
[Q(B/P) : Q(A/p)]s .
P:1 (P )=p
So it suffices to show that the number of prime ideals P B such that 1 (P) = p
is finite. It follows from Lemma 10.3 (iii) that the set of such prime ideals is equal
to the set of irreducible components of the closed subset of X defined by the proper
ideal pB. We know that the number of irreducible components of an affine k-set is
finite. This proves the second assertion.
Theorem 10.10. Let X be a projective (resp. affine) irreducible algebraic k-set.
Assume that k is an infinite field. Then there exists a finite regular map f : X
Pnk (K) (resp. Ank (K)).
Proof. Assume first that X is projective. Let X be a closed subset of some Prk (K).
If X = Prk (K), we take for f the identity map. So we may assume that X is a
proper closed subset. Since k is infinite, we can find a point x Prk (k) \ X. Let
px : X Pr1
k (K) be the linear projection from the point x. We know from the
previous examples that px : X px (X) is a finite k-map. If px (X) = Pr1
k (K), we
are done. Otherwise, we take a point outside px (X) and project from it. Finally, we
obtain a finite map (composition of finite maps) X Pnk (K) for some n.
Assume that X is affine. Then, we replace X by an isomorphic set lying as a closed
be the closure of X in Pr (K). Projecting
subset of Prk (K)0 of some Prk (K). Let X
k
r
r
Pr1 (K). Since
92
Problems.
1. Decide whether the following maps f : X Y are finite:
(a) Y = V (Z12 Z23 ) be the cuspidal cubic, X = A1 , f is defined by the formula
x (x3 , x2 );
(b) X = Y = A2 , f is defined by the formula (x, y) (xy, y).
2. Let f : X Y be a finite map. Show that the image of any closed subset of X is
closed in Y .
3. Let f : X Y and g : X 0 Y 0 be two finite regular maps. Prove that the
Cartesian product map f g : X X 0 Y Y 0 is a finite regular map.
4. Give an example of a surjective regular map with finite fibres which is not finite.
5. Let A be an integral domain, Q be its field of fractions. The integral closure A of
Lecture 11
Dimension
In this lecture we give a definition of the dimension of an algebraic (= quasi-projective
algebraic) k-set. Recall that the dimension of a linear space L can be defined by :
dimL = sup{r : a strictly decreasing chain of linear subspaces L0 L1 . . . Lr }.
The dimension of algebraic sets is defined in a very similar way:
Definition 11.1. Let X be a non-empty topological space. Its Krull dimension is
defined to be equal to
dimX = sup{r : a chain Z0 Z1 . . . Zr 6= of closed irreducible subsets of X}.
By definition the dimension of the empty set is equal to .
The dimension of an algebraic k-set X is the Krull dimension of the corresponding
topological space.
Example 11.1. dimA1k (K) = 1. Indeed, the only proper closed irreducible subset is
a finite set defined by an irreducible polynomial with coefficients in k. It does not
contain any proper closed irreducible subsets.
Proposition 11.2. (General properties of dimension). Let X be a topological space.
Then
(i) dim X = 0 if X is a non-empty Hausdorff space;
(ii) dim X = sup{dim Xi , i I}, where Xi , i I, are irreducible components of
X;
(iii) dim X dim Y if Y X, the strict inequality takes place if none of the
irreducible components of the closure of Y is an irreducible component of X;
93
94
95
Proof. Obviously, follows from the existence of the natural correspondence between
closed irreducible subsets of X and prime ideals in O(X)
= A.
Recall that a finite subset {x1 , . . . , xk } of a commutative algebra A over a field
k is said to be algebraically dependent (resp. independent) over k if there exists
(resp. does not exist) a non-zero polynomial F (Z1 , . . . , Zk ) k[Z1 , . . . , Zk ] such that
F (x1 , . . . , xk ) = 0. The algebraic dimension of A over k is the maximal number of
algebraically independent elements over k in A if it is defined and otherwise. We
will denote it by alg.dimk A.
Lemma 11.5. Let A be a k-algebra without zero divisors and Q(A) be the field of
fractions of A. Then
(i) alg.dimk Q(A) = alg.dimk A;
(ii) alg.dimk A dim A.
Proof. (i) Obviously, alg.dimk A alg.dimk Q(A). If x1 , . . . , xr are algebraically independent elements in Q(A) we can write them in the form ai /s, where ai A, i =
1, . . . , r, and s A. Consider the subfield Q0 of Q(A) generated by a1 , . . . , ar , s. Since
Q0 contains x1 , . . . , xr , s, alg.dimk Q0 r. If a1 , . . . , ar are algebraically dependent,
then Q0 is an algebraic extension of the subfield Q00 generated by s and a1 , . . . , ar with
some ai , say ar , omitted. Since alg.dimk Q0 = alg.dimk Q00 , we find r algebraically independent elements a1 , . . . , ar1 , s in A. This shows that alg.dimk Q(A) alg.dimk A.
(ii) Let P be a prime ideal in A. Let x
1 , . . . , x
r be algebraically independent
elements over k in the factor ring A/P and let x1 , . . . , xr be their representatives in A.
We claim that for every nonzero x P the set x1 , . . . , xr , x is algebraically independent
over k. This shows that alg.dimk A > alg.dimk A/P and clearly proves the statement.
Assume that x1 , . . . , xr , x are algebraically dependent. Then F (x1 , . . . , xr , x) = 0 for
some polynomial F k[Z1 , . . . , Zn+1 ] \ {0}. We can write F as a polynomial in Zn+1
with coefficients in k[Z1 , . . . , Zn ]. Then
F (x1 , . . . , xr , x) = a0 (x1 , . . . , xr )xn + . . . + an1 (x1 , . . . , xr )x + an (x1 , . . . , xr ) = 0,
where ai k[Z1 , . . . , Zn ]. Canceling by x, if needed, we may assume that an 6= 0
(here we use that A does not have zero divisors). Passing to the factor ring A/P, we
obtain the equality
F (
x1 , . . . , x
r , x
) =
n
X
ai (
x1 , . . . , x
r )
xni = an (
x1 , . . . , x
r ) = 0,
i=0
96
Proposition 11.6.
dim Ank (K) = dim Pn (K) = n.
Proof. Since Pn (K) is covered by affine sets isomorphic to Ank (K), the equality
dim Ank (K) = dim Pn (K) follows from Proposition 11.2. By Proposition 11.4, we
have to check that dim k[Z1 , . . . , Zn ] = n. Obviously,
(0) (Z1 ) (Z1 , Z2 ) . . . (Z1 , . . . , Zn )
is a strictly increasing chain of proper prime ideals of k[Z1 , . . . , Zn ]. This shows that
dim k[Z1 , . . . , Zn ] n.
By Lemma 11.5,
alg.dimk k[Z1 , . . . , Zn ] = alg.dimk k(Z1 , . . . , Zn ) = n dim k[Z1 , . . . , Zn ] n.
This proves the assertion.
Lemma 11.7. Let B a k-algebra which is integral over its subalgebra A. Then
dim A = dim B.
Proof. For every strictly increasing chain of proper prime ideals P0 . . . Pk in B,
we have a strictly increasing chain P0 A . . . Pk A of proper prime ideals in A
(Lemma 10.3 (iii) from Lecture 10). This shows that dim B dim A.
Now let P0 . . . Pk be a strictly increasing chain of prime ideals in A. By
Lemma 10.3 from Lecture 10, we can find a prime ideal Q0 in B with Q0 A = P0 .
= B/Q0 , the canonical injective homomorphism A B
is an
Let A = A/P0 , B
integral extension. Applying Lemma 11.5 again we find a prime ideal Q1 in B which
1 to a prime ideal Q1 in B we find Q1 Q0
cuts out in A the image of P1 . Lifting Q
and Q1 A = P1 . Continuing in this way we find a strictly increasing chain of prime
ideals Q0 Q1 . . . Qk in B. This checks that dim B dim A and proves the
assertion.
Theorem 11.8. Let A be a finitely generated k-algebra without zero divisors. Then
dim A = alg.dimk A = alg.dimk Q(A).
In particular, if X is an irreducible affine algebraic k-set and R(X) is its field of rational
functions, then
dim X = alg.dimk O(X) = alg.dimk R(X).
97
Proof. By Noethers Normalization Theorem from Lecture 10, A is integral over its
subalgebra isomorphic to k[Z1 , . . . , Zn ]. Passing to the localization with respect
to the multiplicative set S = k[Z1 , . . . , Zn ] \ {0}, we obtain an integral extension
k(Z1 , . . . , Zn ) AS . Since k(Z1 , . . . , Zn ) is a field, and A is a domain, AS must be
a field equal to its field of fractions Q(A). The field extension k(Z1 , . . . , Zn ) Q(A)
is algebraic. Applying Lemmas 11.5 and 11.7, we get
alg.dimk A dim A = dim k[Z1 , . . . , Zn ] = alg.dimk k(Z1 , . . . , Zn )
= alg.dimk Q(A) = alg.dimk A.
This proves the assertion.
So we see that for irreducible affine algebraic sets the following equalities hold:
dim X = dim O(X) = alg.dimk O(X) = alg.dimk R(X) = n
where n is defined by the existence of a finite map X Ank (K).
Note that, since algebraic dimension does not change under algebraic extensions,
we obtain
Corollary 11.9. Let X be an affine algebraic k-set and let X 0 be the same set considered as an algebraic k 0 -set for some algebraic extension k 0 of k. Then
dim X = dim X 0 .
To extend the previous results to arbitrary algebraic sets X, we will show that for
every dense open affine subset U X
dim U = dim X.
(11.1)
If X is an affine irreducible set, and U = D() for some O(X), then it is easy
to see. We have
dim D() = dim O(U )[ 1 ] = alg.dimk Q(O(U )[ 1 ])
(11.2)
98
Lemma 11.11. Let B be a domain which is integral over A = k[Z1 , . . . , Zr ], and let
x and y be coprime elements of A. Assume that x|uy for some u B. Then x|uj for
some j.
Proof. Let uy = xz for some z B. Since z is integral over Q(A) its minimal monic
polynomial over Q(A) has coefficients from A. This follows from the Gauss Lemma
(if F (T ) Q(A)[T ] divides a monic polynomial G(T ) A[T ] then F (T ) A[T ]).
Let
F (T ) = T n + a1 T n1 + . . . + an = 0, ai A,
be a minimal monic polynomial of z. Plugging z = uy/x into the equation, we obtain
that u satisfies a monic equation:
F (T )0 = T n + (a1 x/y)T n1 + . . . + (an xn /y n ) = 0
with coefficients in the field Q(A). If u satisfies an equation of smaller degree over
Q(A), after plugging in u = xz/y, we find that z satisfies an equation of degree
smaller than n. This is impossible by the choice of F (T ). Thus F (T )0 is a minimal
polynomial of u. Since u is integral over A, the coefficients of F (T )0 belong to A.
Therefore, y i |ai xi , and, since x and y are coprime, y i |ai . This implies that un +xt = 0
for some t A, and therefore x|un .
Lemma 11.12. Assume k is infinite. Let X be an irreducible affine k-set, and let be a
non-zero and non-invertible element in O(X). There exist 1 , . . . , n1 O(X) such
that the map X Ank (K) defined by the formula x ((x), 1 (x), . . . , n1 (x)) is
a regular finite map.
Proof. Replacing X by an isomorphic set, we may assume that X is a closed subset of
d
some Pm
k (K)0 , and = F (T0 , . . . , Tm )/T0 for some homogeneous polynomial F (T )
99
Pn (K) be the regular map given by the polynomials (T d , F, F1 , . . . , Fn1 ).
Let f : X
0
k
by its image vd (X)
under the Veronese
We claim that it is finite. Indeed, replacing X
N
map vd : Pm
k (K) Pk (k), we see that f is equal to the restriction of the linear
projection map
Pn (K)
prE : vd (X)
k
where E is the linear subspace defined by the linear forms in N + 1 unknowns corresponding to the homogeneous forms T0d , F, F1 , . . . , Fn1 . We know that the linear projection map is finite. Obviously, f (X) Pnk (K)0 , and the restriction map
f |X : X Pnk (K)0
= Ank (K), defined by the formula
x(
F
F1
Fn1
(x), d (x), . . . , d (x)) = ((x), 1 (x), . . . , n1 (x))
T0d
T0
T0
is finite.
Proof of Krulls Hauptidealsatz:
Let f : X An (K) be the finite map constructed in the previous lemma. It
suffices to show that the restrictions i of the functions i (i = 1, . . . , n 1) to any
irreducible component Y of V () are algebraically independent elements of the ring
O(Y ) (since dim Y = alg.dimk O(Y )). Let F k[Z1 , . . . , Zn1 ] \ {0} be such that
F (1 , . . . , n1 ) I(Y ). Choosing a function g 6 I(Y ) vanishing on the remaining
irreducible components of V (), we obtain that
V (F (1 , . . . , n1 )g) V ().
By the Nullstellensatz, |(F (1 , . . . , n1 )g)N for some N > 0. Now, we can apply
Lemma 11.11. Identifying k[Z1 , . . . , Zn1 , Zn ] with the subring of O(X) by means
of f , we see that = Zn , F (1 , . . . , n1 ) = F (Z1 , . . . , Zn ), and Zn+1 |F N g N in
O(X). From Lemma 11.11 we deduce that Zn |g jN for some j 0, i.e., g 0 on
V () contradicting the choice of g. This proves the assertion.
Theorem 11.13. Let X be an algebraic set and U be a dense open subset of X.
Then
dim X = dim U.
Proof. Obviously, we may assume that X is irreducible and U is its open subset. First
let us show that all affine open subsets of X have the same dimension. For this it is
enough to show that dim U = dim V if V U are affine open subsets. Indeed, we
know that for every pair U and U 0 of open affine subsets of X we can find an affine nonempty subset W U U 0 . Then the above will prove that dim W = dim U, dim W =
dim U 0 . Assume U is affine, we can find an open non-empty subset D() V U ,
where O(U ) \ O(U ) . Applying (11.2), we get dim U = dim D(). This shows
100
that all open non-empty affine subsets of X have the same dimension. Let Z0
Z1 . . . Zn be a maximal decreasing chain of closed irreducible subsets of X, i.e.,
n = dim X. Take x Zn and let U be any open affine neighborhood of x. Then
Z0 U Z1 U . . . Zn U 6=
is a decreasing chain of closed irreducible subsets of U (note that Zi U 6= Zj U for
i j since otherwise Zj = Zi (Zj (X U )) is the union of two closed subsets).
Thus dim U dim X, and Proposition 11.2 implies that dim U = dim X. This proves
that for every affine open subset U of X we have dim U = dim X. Finally, if U is any
open subset, we find an affine subset V U and observe that
n = dim V dim U dim X = n
which implies that dim U = dim X.
Corollary 11.14. Let f : X Y be a finite map of irreducible algebraic k-sets. Then
dim X = dim Y.
Proof. Take any open affine subset U of Y . Then V = f 1 (U ) is affine and
the restriction map V U is finite. By Lemma 11.7, dim U = dim V . Hence
dim f 1 (Y ) = dim V = dim U = dim Yi .
Theorem 11.15. Let F be a homogeneous polynomial not vanishing identically on
an irreducible quasi-projective set X in Pnk (K) and Y be an irreducible component of
X V (F ), then, either Y is empty, or
dim Y = dim X 1.
Proof. Assume Y 6= . Let y Y and U be an open affine subset of X containing
y. Then Y U is an open subset of Y , hence dim Y U = dim Y . Replacing U
with a smaller subset, we may assume that U Pn (k)i for some i. Then F defines a
regular function = F/Tir , r = deg(F ), on U , and Y U = V () U . By Krulls
Hauptidealsatz, dim Y U = dim U 1. Hence
dim Y = dim Y U = dim U 1 = dim X 1.
101
be the set of its common zeroes and Z be an irreducible component of this set. Then,
either Z is empty, or
dim Z dim X r.
The equality takes place if and only if for every i = 1, . . . , r the polynomial Fi does
not vanish identically on any irreducible component of X V (F1 ) . . . V (Fi1 ).
Corollary 11.17. Every r n homogeneous equations in n + 1 unknowns have a
common solution over an algebraically closed field. Moreover, if r < n, then the
number of solutions is infinite.
Proof. Apply the previous Corollary to X = Pnk and use that an algebraic set is finite
if and only if it is of dimension 0 (Proposition 11.3).
Example 11.18. Let C = v3 (P1 (K)) be a twisted cubic in P3 (K). We know that C
is given by three equations:
F1 = T0 T2 T12 = 0, F2 = T0 T3 T1 T2 = 0, F3 = T1 T3 T22 = 0.
We have V (F1 ) V (F2 ) = C L, where L is the line T0 = T1 = 0. At this point, we
see that each irreducible component of V (F1 )V (F2 ) has exactly dimension 1 = 32.
However, V (F3 ) contains C and cuts out L in a subset of C. Hence, every irreducible
component of V (F1 ) V (F2 ) V (F3 ) is of the same dimension 1.
Theorem 11.19. (On dimension of fibres). Let f : X Y be a regular surjective
map of irreducible algebraic sets, m = dim X, n = dim Y . Then
(i) for any y Y and an irreducible component F of the fibre f 1 (y), dim F
m n;
(ii) there exists a nonempty open subset V of Y such that, for any y V ,
dim f 1 (y) = m n.
Proof. Let x f 1 (y). Replacing X with an open affine neighborhood of x, and
same for y, we assume that X and Y are affine. Let : Y An (K) be a finite map
and f 0 = f . Applying Proposition 10.9 from Lecture 10 we obtain that, for any
z An (K), the fibre f 01 (z) is equal to a finite disjoint union of the fibres f 1 (y)
where y 1 (z). Since dim Y = n, we may assume that Y = An (K).
(i) Each point y = (a1 , . . . , an ) An (K) is given by n equations Zi ai = 0. The
fibre f 1 (y) is given by n equations f (Zi ai ) = 0. Applying Krulls Hauptidealsatz,
we obtain (i).
(ii) Since f is surjective, f : O(Y ) O(X) is injective, hence defines an extension of fields of rational functions f : R(Y ) R(X). By the theory of finitely
102
generated field extensions, L = R(X) is an algebraic extension of a purely transcendental extension K 0 = R(Y )(z1 , . . . , zr ) of K = R(Y ). Clearly,
m = alg.dimR(X) = alg.dimR(Y ) + r = n + r.
Let : X Y Ar (K) be a rational map of affine sets corresponding to the
extension L/K 0 . We may replace again X and Y by open affine subsets to assume
that is regular. Let O(X) be generated by u1 , . . . , uN as a k-algebra. We know
that every ui satisfies an algebraic equation a0 udi + . . . + ad = 0 with coefficients in
K 0 = R(Y Ar (K)). Replacing Y Ar (K) by an open subset Ui we may assume that
all ai O(U ) and a0 is invertible (throwing away the closed subset of zeroes of a0 ).
Taking the intersection U of all Ui s, we may assume that all ui satisfy monic equations
with coefficients in O(U ). Thus O(X) is integral over O(U ) hence : X U is
a finite map. Let p : Y Ar (K) Y be the first projection. The corresponding
extension of fields K 0 /K is defined by p . Since p is surjective, p(U ) is a dense subset
of Y . Let us show that p(U ) contains an open subset of Y . We may replace U by
r
a subset of the
P formi D(F ) where F = F (Y1 , . . . , Yn , Z1 , . . . , Zr ) O(Y A (K)).
Write F = i Fi Z as a sum of monomials in Z1 , . . . , Zr . For every y Y such
that not all Fi (y) = 0, we obtain non-zero polynomial in Z, hence we can find a
point z Ar (K) such that F (y, z) 6= 0. This shows that p(D(F )) D(Fi ), hence
the assertion follows. Let V be an open subset contained in p(U ). Replacing U by
an open subset contained in p1 (V ), we obtain a regular map p : U V and the
commutative triangle:
1 (U )
f
/U
p
The fibres of p are open subsets of fibres of the projection Y Ar (K) Ar (K) which
are affine n-spaces. The map : 1 (U ) U is finite as a restriction of a finite map
over an open subset. Its restriction over the closed subset p1 (y) is a finite map too.
Hence defines a finite map f 1 (y) p1 (y) and
dim f 1 (y) = dim p1 (y) = r = m n.
The theorem is proven.
Corollary 11.20. Let X and Y be irreducible algebraic sets. Then
dim X Y = dim X + dim Y.
Proof. Consider the projection X Y Y and apply the Theorem.
103
Theorem 11.21. Let X and Y be irreducible quasi-projective subsets of Pn (K). For
every irreducible component Z of X Y
dim Z dim X + dim Y n.
Proof. Replacing X and Y by its open affine subsets, we may assume that X and Y
are closed subsets of An (K). Let : An (K) An (K)An (K) be the diagonal map.
Then maps X Y isomorphically onto (X Y ) An (K) , where An (K) is the
diagonal of An (K). However, An (K) is the set of common zeroes of n polynomials
Zi Zi0 where Z1 , . . . , Zn are coordinates in the first factor and Z10 , . . . , Zn0 are the
same for the second factor. Thus we may apply Theorem 11.10 n times to obtain
dim Z dim X Y n.
It remains to apply the previous corollary.
We define the codimension codim Y (or codim (Y, X) to be precise) of a subspace
Y of a topological space X as dim X dim Y . The previous theorem can be stated
in these terms as
codim (X Y, Pn (K)) codim (X, Pn (K)) + codim (Y, Pn (K)).
In this way it can be stated for the intersection of any number of subsets.
Problems.
1. Give an example of
(a) a topological space X and its dense open subset U such that dim U < dim X;
(b) a surjective continuous map f : X Y of topological spaces with dim X <
dim Y ;
(c) a Noetherian topological space of infinite dimension.
2. Prove that every closed irreducible subset of Pn (K) or An (K) of codimension 1 is
the set of zeroes of one irreducible polynomial.
3. Let us identify the space K nm with the space of matrices of size m n with entries
in K. Let X 0 be the subset of matrices of rank m 1 where m n. Show that
the image of X 0 \ {0} in the projective space Pnm1 (K) is an irreducible projective
set of codimension n m + 1.
4. Show that for every irreducible closed subset Z of an irreducible algebraic set X
there exists a chain of n = dim X + 1 strictly decreasing closed irreducible subsets
104
Lecture 12
Lines on hypersurfaces
In this lecture we shall give an application of the theory of dimension. Consider the
following problem. Let X = V (F ) be a projective hypersurface of degree d = degF
in Pn (K). Does it contain a linear subspace of given dimension, and if it does,
how many? Consider the simplest case when d = 2 (the case d = 1 is obviously
trivial). Then F is a quadratic form in n + 1 variables. Let us assume for simplicity
that char(K) 6= 2. Then a linear m-dimensional subspace of dimension in V (F )
corresponds to a vector subspace L of dimension m + 1 in K n+1 contained in the set
of zeroes of F in K n+1 . This is an isotropic subspace of the quadratic form F . From
the theory of quadratic forms we know that each isotropic subspace is contained in
a maximal isotropic subspace of dimension n + 1 r + [r/2], where r is the rank of
F . Thus V (F ) contains linear subspaces of dimension n r + [r/2] but does not
contain linear subspaces of larger dimension. For example, if n = 3, and r = 4, X is
isomorphic to V (T0 T1 T2 T3 ). For every , K, we have a line L(, ) given by
the equations
T0 + T2 = 0, T1 + T3 = 0,
or a line M (, ) given by the equation
M (, ) : T0 + T3 = 0, T1 + T2 = 0.
It is clear that L(, ) L(0 , 0 ) = (resp. M (, ) M (0 , 0 ) 6= ) if and
only if (, ) 6= (0 , 0 ) as points in P1 (K). On the hand L(, ) M (0 , 0 ) is one
point always. Under an isomorphism V (F )
= P1 (K)P1 (K), the two families of lines
L(, ) and M (, ) correspond to the fibres of the two projections P1 (K)P1 (K)
P1 (K).
Another example is the Fermat hypersurface of V (F ) P3 (K) of degree d, where
F = T0d + T1d + T2d + T3d .
105
106
Since
Tid + Tjd =
d
Y
(Ti + s Tj ))
s=1
where is a primitive d-th root of 1, we see that V (F ) contains 3d2 lines. Each one
is defined by the equations of the type:
Ti + s Tj = 0, Tk + t Tl = 0,
where {i, j, k, l} = {0, 1, 2, 3}. In particular, when d = 3, we obtain 27 lines. As we
shall see in this Lecture, almost every cubic surface contains exactly 27 lines. On
the other hand if d 4, almost no surface contains a line.
To solve our problems, we first parametrize the set of linear r-dimension al subspaces of of Pn (K) by some projective algebraic set. This is based on the classic
construction of the Grassmann variety.
Let V be a vector space of dimension n + 1 over a fieldVK and let L be its linear
subspace of dimension r + 1. Then Vthe exterior product r+1 (L) can be identified
with a one-dimensional
subspace of r+1 (V ), i.e., with a point [L] of the projective
Vr+1
Vr+1
(V ) \ {0}/K .VIn coordinates, if e1 , . . . , en+1 is a basis of
(V )) =
space p(
V , and f1 , . . . , fr+1 is a basis of L, then r+1 (L) is spanned by one vector
f1 . . . fr+1 =
V
Theorem 12.1. The subset G(r + 1, n + 1) of lines in r+1 (V ) spanned
by decomVr+1
posable (r + 1)-vectors f1 . . . fr+1 is a projective algebraic set in P(
(V ))
=
n+1
V
1
r+1
(
)
r+1
P
(K). The map L [L] =
(L) is a bijective map from the set of linear
subspaces of V of dimension r + 1 to the set G(r + 1, n + 1).
107
V
Proof. We use the following fact from linear algebra. For every t r+1 (V ) let
L(t) = {x V : t x = 0}. This is a linear subspace of V . Then dim L(t) r + 1 if
and only if t is decomposable and equal to f1 . . . fr+1 for some linear independent
vectors f1 , . . . , fr+1 which have to form a basis of L(t). This assertion shows that the
subspace L can be reconstructed uniquely from [L] as the subspace L(t), where t is any
basis of [L]. Let us prove the assertion. The sufficiency is easy. If t = f1 . . . fr+1
for some basis {f1 , . . . , fr+1 } of a linear subspace of dimension r + 1, then, obviously,
f1 . . . fr+1 x = 0 for any x L
Kf1 + . . . + Kfr+1 so that L L(t). Since
V=
r+2
f1 . . . fr+1 x = 0 implies that
(Kf1 + . . . + Kfr+1 + Kx) = 0, we obtain
that dim(Kf1 + . . . + Kfr+1 + Kx) = r + 1, hence x Kf1 + . . . + fr+1 . This shows
that L = L(t). Conversely assume dim L(t) = r + 1. Let f1 , . . . fr+1 be a set of linear
independent vectors in L(t), and let {f1 , . . . , fr+1 , fr+1 , . . . , fn+1 } be an extension of
{f1 , . . . , fr+1 } to a basis of V . We can write
X
t=
ai1 ...ir+1 fi1 . . . fir+1 .
i1 <...<ir+1
It is easy to see that tfi = 0, i = 1, . . . , r+1, implies ai1 ...ir+1 = 0 for {i1 , . . . ir+1 } =
6
{1, . . . , r + 1}. Hence t is proportional to f1 . . . fr+1 .
To see whyVdecomposable non-zero (r + 1)-vectors define a closed subset G(r +
1, n + 1) of p( r+1 (V )) it suffices to observe that V
dim L(t) r + 1 if and only if
rk(Tt ) n r, where Tt is the linear map V r+2 (V ) defined by the formula
x 7 t x. The latter condition is equivalent to vanishing of (n r + 1)-minors of
the matrix of Tt with respect to some basis. By taking a basis e1 , . . . , en+1 of V ,
it is easy to see that the entries of the matrix of Tt are the Pl
ucker coordinates of
the space L(t). Thus we obtain that G(r + 1, n + 1) is the set of zeroes of a set of
homogeneous polynomials of degree n r + 1. Observe that these polynomials have
integer coefficients, so G(r + 1, n + 1) is a projective k-set for any k K.
More generally, we can define a projective algebraic variety Gk (r +1, n+1) defined
by:
Gk (r + 1, n + 1)(K) = {direct summands of K n+1 of rank r + 1}.
Note that a direct summand
V of a free module is a projective module. The operation
of exterior power, M r+1 (M ) defines a morphism of projective algebraic varieties
(n+1)1
p : Gk (r + 1, n + 1) Pkr+1 .
If r = 0 this morphism is an isomorphism.
Definition 12.1. The projective variety Gk (r + 1, n + 1) is called the Grassmann
variety over the field k. The morphism p is called the Pl
ucker embedding of Gk (r +
1, n + 1). For every algebraically closed field K containing k, we shall identify the set
(n+1)1
Gk (r + 1, n + 1)(K) with the projective algebraic subset G(r + 1, n + 1) of Pkr+1 .
108
Proposition 12.2. The projective algebraic set G(r + 1, n + 1) is an irreducible projective set of dimension (n r)(r + 1).
Proof. We shall give two different proofs of this result. Each one carries some additional information about G(r + 1, n + 1). In the first one we use the following obvious
fact: the general linear group GL(n + 1, K) acts transitively on the set of (r + 1)dimension al linear subspaces of K n+1 . Moreover, the stabilizer of each such subspace
L is isomorphic to the subgroup P of GL(n + 1, K) that consists of matrices of the
form:
A B
,
0 C
where A, B, C are matrices of size (r + 1) (r + 1), (r + 1) (n r), (n r) (n r),
2
respectively. Let us consider GL(n + 1, K) as a closed subset of K n +1 defined by the
equation T0 det((Tij ))1 = 0. then it is clear that P is a closed subset of GL(n+1, K)
defined by the additional equations Tij = 0, i = n + 2 r, . . . , n + 1, j = 1, . . . , r + 1.
The dimension of P is equal to (n + 1)2 (n r)(r + 1). Next we define a surjective
regular map of algebraic k-sets f : GL(n + 1, K) G(r + 1, n + 1) by the formula
M M (L0 ), where L0 = Ke1 + . . . + Ker+1 . If M = (aij ), then
M (L0 ) = span{a11 e1 + . . . + an+11 en+1 , . . . , a1r+1 e1 + . . . + an+1r+1 }
so that the Pl
ucker coordinates p[i1 , . . . , ir+1 ] of M (L0 ) are equal to the minor of
the matrix (aij ) formed by the first r + 1 columns and the rows indexed by the set
{i1 , . . . , ir+1 }. This shows that f is a regular map from the affine k-set GL(n + 1, K)
to the projective algebraic k-set G(r + 1, n + 1). Its fibres are isomorphic to P . By
the theorem on the dimension of fibres from Lecture 11, we obtain that
dim G(r+1, n+1) = dim GL(n+1, K)dim P = (n+1)2 ((n+1)2 (nr)(r+1))
= (n r)(r + 1).
Since GL(n + 1, K) is irreducible, G(r + 1, n + 1) is irreducible.
Now let us give another proof of this result. Choose the Pl
ucker coordinates
(n+1)1
p[j1 , . . . , jr+1 ] and consider the open subsets D(p[j1 , . . . , jr+1 ]) Pkr+1 (K). The
intersection D(p[j1 , . . . , jr+1 ]) G(r + 1, n + 1) is equal to the set of linear subspaces
L which admit a basis
f1 = a11 e1 + . . . + a1n+1 en , . . . , fr+1 = ar+11 e1 + . . . + ar+1n+1 en ,
such that p[j1 , . . . , jr+1 ] = det(Aj1 j2 ...jr+1 ) 6= 0,
a1j1 a1j2
...
...
Ai1 i2 ...ir+1 =
...
...
ar+1j1 a1j2
where
. . . a1jr+1
...
...
.
...
...
. . . ar+1jr+1
109
0
After we replace f1 , . . . , fr+1 with f10 , . . . , fr+1
such that
0
f10 = b11 f1 + . . . + b1r+1 fr+1 , . . . , fr+1
= br+11 f1 + . . . + br+1r+1 fr+1 ,
where (bij ) is the inverse of the matrix Aj1 j2 ...jr+1 , we may assume that Ai1 i2 ...ir+1 is
the identity matrix Ir+1 . Then we may take all (n r)(r + 1) other entries aij , j 6= jk
arbitrary, and obtain that D(p[i1 , . . . , ir+1 ]) G(r + 1, n + 1) is isomorphic to the
(nr)(r+1)
affine space Ak
(K). Thus G(r + 1, n + 1) is covered by n+1
r+1 open subsets
isomorphic to the affine space of dimension (n r)(r + 1). This obviously proves the
assertion.
Example 12.3. Let us consider the case r = 1, n = 3. Then G(2, 4) P5 parametrizes
lines Vin P3 (K). We have six Pl
ucker coordinates p[ij], i, j = 1, 2, 3, 4. An element
2 (V ) can be identified with a skew-symmetric bilinear form V V K. The
110
Remark 12.5. The equations for G(2, 4) given in the proof of Theorem 12.1 differ from
the equation Q = 0. They define a non-saturated ideal of the projective variety. Any
Grassmannian G(r + 1, n + 1) can be given by a system of equations of degree 2, so
called the Pl
ucker equations. They look as follows:
r+2
X
s=1
where {i1 , . . . , ir } and {j1 , . . . , jr+2 } are any two strictly increasing sequences of the
set {1, . . . , n + 1}.
n+d
We denote by Hyp(d; n) the projective space P( d )1 . If we use i0 ,...,in , 0
ij , i0 + . . . + in = d to denote projective coordinates in this space then each K-point
(. . . , ai0 ,...,in , . . .) of this space defines the projective K-subvariety F = 0 of PnK where
X
F =
ai0 ,...,in T0i0 Tnin = 0.
i0 ,...,in
n+d
Thus we can view K-points of the projective space P( d )1 as projective hypersurfaces
of degree d. This explains the notation. In the special case when d = 1, the space
n . Its K-points
Hyp(1, n) is called the dual projective space of Pnk and is denoted by P
k
are in a bijective correspondence with linear subspaces of Pnk (K) of dimension n 1
(hyperplanes) .
Now, everything is ready to solve our problem. Fix any algebraically closed field
K. Let H = Hyp(d; n)(K) and G = G(r + 1, n + 1)(K). Define
111
Pl
ucker coordinates of E 0 . Now plugging in (0 , . . . , n ) into the equation of X , we
see that I(r, d, n) is given by bi-homogeneous polynomials in the coefficients of F and
in the Pl
ucker coordinates of E. This proves that I(r, d, n) is a closed subset of the
product H G. Now consider the projection p : I(r, d, n) G. For each E G, the
fibre p1 (E) consists of all hypersurfaces V (F ) containing E. Choose a coordinate
system such that E is given by the equations Tr+1 = . . . = Tn = 0. Then E V (F )
if and only if each monomial entering into F with non-zero coefficient contains some
positive power of Ti with i r + 1. In other words F is defined by vanishing of all
coefficients at the monomials of degree d in the variables T0 , . . . , Tr . This gives r+d
r
r+d
r+d
.
r
Then the subset of Hyp(d; n)(K) which consists of hypersurfaces containing a linear
subspace of dimension r is a proper closed subset.
112
(ii) There exists an open subset U Hyp(3; 3)(K) such that any X U contains
exactly 27 lines.
Proof. (i) In the notation of the proof of Theorem 12.8, it suffices to show that the
projection map q : I(1, 3, 3) Hyp(3; 3)(K) is surjective. Suppose the image of q is
a proper closed subset Y of Hyp(3; 3)(K). Then dim Y < dim Hyp(3; 3)(K) = 19
and dim I(1, 3, 3) = 19. By the theorem on dimension of fibres, we obtain that all
fibres of q are of dimension at least one. In particular, every cubic surface containing
a line contains infinitely many of them. But let us consider the surface X given by
the equation
T1 T2 T3 T03 = 0.
Suppose a line ` lies on X. Let (a0 , a1 , a2 , a3 ) `. If a0 6= 0, then ai 6= 0, i 6= 0.
On the other hand, every line hits the planes Ti = 0. This shows that ` is contained
in the plane T0 = 0. But there are only three lines on X contained in this plane:
Ti = T0 = 0, i = 1, 2 and 3. Therefore X contains only 3 lines. This proves the first
assertion.
(ii) We already know that every cubic surface X = V (F ) has at least one line.
Pick up such a line `. Without loss of generality we may assume that it is given by
the equation:
T2 = T3 = 0.
113
As we saw in the proof of Lemma 12.6:
F = T2 Q0 (T0 , T1 , T2 , T3 ) + T3 Q1 (T0 , T1 , T2 , T3 ) = 0,
where Q0 and Q1 are quadratic homogeneous polynomials. Each plane containing
the line ` is given by the equation
T2 T3 = 0
for some scalars , K. The intersection V (F ) contains the line ` and a curve
of degree 2 in . More explicitly, choose coordinates t0 , t1 , t2 in the plane, related to
our coordinates T0 , T1 , T2 , T3 by the formulas:
T0 = t0 , T1 = t1 , T2 = t2 , T3 = t2 .
Plugging these expression into F , we obtain:
t2 Q0 (t0 , t1 , t2 , t2 ) + t2 Q1 (t0 , t1 , t2 , t2 ) = 0.
This shows that X consists of the line ` with the equation t2 = 0 and the
conic C(, ) with the equation:
Q0 (t0 , t1 , t2 , t2 ) + Q1 (t0 , t1 , t2 , t2 ) = 0.
We may also assume that the line enters with multiplicity one (since we take general
coeficients of F ). Let
X
X
Q0 =
aij Ti Tj , Q1 =
bij Ti Tj .
0ij3
0ij3
114
line case does not occur. Thus we found 11 lines on X: the line ` and 5 pairs of lines
`i , `0i lying each lying in the plane i . Pick up some plane, say 1 . We have 3 lines
`, `0 , and `00 in 1 . Replacing ` by `0 , and then by `00 , and repeating the construction,
we obtain 4 planes through `0 and 4 planes through `00 not containing `, and each
containing a pair of lines. Altogether we found 3 + 8 + 8 + 8 = 27 lines on X. To see
that all lines are accounted for, we observe that any line intersecting either `, or `0 , or
`00 lies in one of the planes we have considered before. So it has been accounted for.
Now let L be any line. We find a plane through L that contains three lines L, L0
and L00 on X. This plane intersects the lines `, `0 , and `00 at some points p, p0 and p00
respectively. We may assume that these points are distinct. Otherwise we find three
non-coplanar lines in X passing through one point. As we shall see later this implies
that X is singular at this point. Since neither L0 nor L00 can pass through two of these
points, one of these points lie on L. Hence L is coplanar with one of the lines `, `0 , `00 .
Therefore L has been accounted for.
Remark 12.11. Using more techniques one can show that every nonsingular (in the
sense of the next lectures) cubic surface contains exactly 27 lines. Let us define the
graph whose vertices are the lines and two vertices are joined by an edge if the lines
intersect. This graph is independent on the choice of a nonsingular cubic surface and
its group of symmetries is isomorphic to the group W (E6 ) of order 51840 (the Weyl
group of the root system of a simple Lie algebra of type E6 ).
Problems.
1. Show that the set x of lines in P3 (K) passing through a point x P3 (K) is a
closed subset of G(2, 4) isomorphic to P2 (K). Also show that the set P of lines in
P3 (K) contained in a plane P P3 (K) is a closed subset of G(2, 4) isomorphic to
P2 (K).
2. Prove that the subset of quartic surfaces in Hyp(4; 3) which contain a line is an
irreducible closed subset of Hyp(4; 3) of codimension 1.
3. Prove that every hypersurface of degree d 5 in P4 (K) contains a line, and, if
d 4, then it contains infinitely many lines.
4. Let X be a general cubic hypersurface in P4 (K) (general means that X belongs to
an open subset U of Hyp(3; 4)). Show that there exists an open subset V X such
that any x V lies on exactly six lines contained in X.
5*. Let 1 m1 < m2 < . . . < mr n + 1, a flag in K n+1 of type (m1 , . . . , mr ) is a
chain of linear subspaces L1 . . . Lr with dim Li = mi .
(a) Show that the set of flags is a closed subset in the product of the Grassmannians
G(m1 , n + 1) . . . G(mr , n + 1). This projective algebraic set is called the
flag variety of type (m1 , . . . , mr ) and is denoted by Fk (m1 , . . . , mr ; n + 1).
115
(b) Find the dimension of Fk (m1 , . . . , mr ; n + 1).
6. By analyzing the proof of Theorem 13.16 show the following:
(a) The set of 27 lines on a cubic surface X contains 45 triples of lines which lie in
a plane (called a tritangent plane).
(b) There exist 12 lines l1 , . . . , l6 , l10 , . . . , l60 such that li0 li = , i = 1, . . . , 6, li lj0 6=
if i 6= j. Such a set is called a double-six configuration.
(c)* Show that there are 36 different double-six configurations.
(d) Check all the previous assertions for the Fermat cubic.
7*. Prove that
(a) A general cubic surface V (F ) contains 9 lines `ij , i, j = 1, 2, 3 such that `ij
`km 6= if and only if i = k or j = m.
(b) Using (a) show that V (F ) can be given by the equation
L1
0 M1
0 = 0,
det M2 L2
0 M3 L3
where Li , Mi are linear forms.
(c) Show that the map T : V (F ) P2 which assigns to a point x V (F ) the set
of solutions of the equation (t0 , t1 , t2 ) A = 0 is a birational map. Here A is
the matrix of linear from from (b).
(d) Find an explicit formulas for the inverse birational map T 1 .
8. Using Problem 5 (b),(c) show that the group W of symmetries of 27 lines consists
of 51840 elements.
9*. Let C be a twisted cubic in P3 (the image of P1 under a Veronese map given
by monomials of degree 3). For any two distinct point x, y C consider the line
lx,y joining these points. Show that the set of such lines is a locally closed subset of
G(2, 4). Find the equations defining its closure.
10*. Let k = k0 (t), where k0 is an algebraically closed field and F (T0 , . . . , Tn )
k[T1 , . . . , Tn ] be a homogeneous polynomial of degree d < n. Show that V (F )(k) 6=
(Tsens Theorem).
116
Lecture 13
Tangent space
The notion of the tangent space is familiar from analytic geometry. For example, let
F (x, y) = 0 be a curve in R2 and let a = (x0 , y0 ) be a point lying on this surface.
The tangent line of X at the point a is defined by the equation:
F
F
(a)(x x0 ) +
(a)(y y0 ) = 0.
x
y
It is defined only if at least one partial derivative of F at a is not equal to zero. In
this case the point is called nonsingular. Otherwise it is said to be singular.
Another notion of the tangent space is familiar from the theory of differentiable
manifolds. Let X be a differentiable manifold and a be its point. By definition, a
tangent vector ta of X at a is a derivation (or differentiation) of the ring O(X) of
differentiable functions on X, that is, a R-linear map : O(X) R such that
(f g) = f (a)(g) + g(a)(f )
where (t1 , . . . , tn ) are the coordinates of ta and (x1 , . . . , xn ) are the local coordinates
of X at the point a.
In this lecture we introduce and study the notion of a tangent space and a nonsingular point for arbitrary algebraic sets or varieties.
For every k-algebra K let K[] = K[t]/(t2 ) be the K-algebra of dual numbers. If
is taken to be t mod(t2 ), then K[] consists of linear combinations a + b, a, b K,
which are added coordinate wise and multiplied by the rule
(a + b)(a0 + b0 ) = aa0 + (ab0 + a0 b).
117
118
We denote by
1 : K[] K
the natural homomorphism a + b a. Its kernel is the ideal () = {b, b K}.
Definition 13.1. Let X be an affine or a projective algebraic variety over a field k and
x X(K) be its K-point. A tangent vector tx of F at x is a K[]-point tx X(K[])
such that X(1 )(tx ) = x. The set of tangent vectors of X at x is denoted by T (X)x
and is called the tangent space of F at x.
Example 13.1. Assume X is an affine algebraic variety given by a system of equations
F1 (Z1 , . . . , Zn ) = 0, . . . , Fr (Z1 , . . . , Zn ) = 0.
A point x X(K) is a solution (a1 , . . . , an ) K n of this system. A tangent vector
tx is a solution (a1 + b1 , . . . , an + bn ) X(K[]) of the same system. Write down
the polynomials Fi (Z) in the form :
Fi (Z1 , . . . , Zn ) = Gi (Z1 a1 , . . . , Zn an )
=
n
X
(i)
j (Zj
aj ) +
j=1
n
X
(i)
jk (Zj aj )(Zk ak ) + . . . .
j,k=1
(Taylors expansion). Note that the Gi s do not contain the constant term because
(i)
(a1 , . . . , an ) X(K). By definition, the coefficient j is the partial derivative of Fi
Fi
with respect to Zj at the point x = (a1 , . . . , an ). It is denoted by Z
(x). Obviously,
j
it is an element of K. Now we plug the point (a1 +b1 , . . . , an +bn ) into the previous
equations to obtain
Fi (a1 +b1 , . . . , an +bn ) =
n
X
n
X
(i)
(i)
j bi +
jk bi bj 2 +(. . . .)3 +. . .
j=1
j,k=1
n
X
(i)
j bi = 0.
j=1
From this we deduce that (b1 , . . . , bn ) satisfies the system of linear homogeneous
equations:
n
X
Fi
(x)bj = 0, i = 1, . . . , r.
(13.1)
Zj
j=1
Thus the set of tangent vectors T (X)x is bijective to the submodule of K n which
consists of solutions of a homogeneous system of linear equations. In particular, we
have introduced the structure of a K-module on T (X)x .
119
Example 13.2. Assume X = Pnk is the n-dimensional projective space over k. Let
x = (a0 , . . . , an ) Pnk (K), where K is a field. A tangent vector at x is a local line M
over K[] such that M/M = (a0 , . . . , an )K. Since the ring K[] is obviously
Plocal,
M is a global line given by coordinates
(a0 + t0 , . . . , an +Ptn ). Note that 1 = i bi ai
P
for some bi K, and therefore i bi (ai + ti ) = 1 + ( i bi ti ) K[] . This shows
that K[](a0 + t0 , . . . , an + tn ) is a global line for any (t0 , . . . , tn ) and M T (Pnk )x
is determined by (t0 , . . . , tn ) up to the equivalence relation defined by
(t0 , . . . , tn ) (t00 , . . . , t0n ) iff (a0 + t0 , . . . , an + tn ) = (a0 + t00 , . . . , an + t0n )
in Pnk (K[]). The latter means that
(a00 + t00 , . . . , a0n + t0n ) = ( + )(a0 + t0 , . . . , an + tn )
for some + K[] (i.e. K , K). This implies that
(a00 , . . . , a0n ) = (a0 , . . . , an ),
120
Thus we can identify T (On,k )In with the vector space of skew-symmetric matrices
with entries in K. Note that the choice of K depends on identification of In with a
K-point.
The tangent space of an algebraic group at the identity point has a structure of a
Lie algebra defined by the Lie bracket.
Remark 13.4. For any functor F from the category of k-algebras to the category of
sets one can define the tangent space of F at a point x F (K) as the set of
elements t of the set F ()(t) = x.
Now, if we have a projective variety X given by a system of homogeneous equations
F1 = = Fk = 0, we obtain that
T (X)x = {a + b T (Pnk )x : F1 (a + b) = = Fk (a + b) = 0.}
By using the Taylor expansion, as in Example 13.1, we obtain that b = (b0 , . . . , bn )
satisfies a system of homogeneous linear equations:
n
X
Fi
j=0
Tj
(a)bj = 0,
i = 1, . . . , k.
(13.2)
n
X
j=0
This gives
0 = di Fi (a0 , . . . , an ) =
n
X
j=0
ai
Ti
F
.
Tj
F
(a),
Tj
(13.3)
i = 1, . . . , k,
121
The set of M -derivations is denoted by Derk (A, M ). It has a natural structure of a
A-module via
(a)(b) = a(b) for all a, b A.
We will be interested in a special case of this definition.
Lemma 13.5. If f : A B is a homomorphism of k-algebras, and : B M is a
M -derivation of B, then the composition f : A B M is a M[f ] -derivation
of A, where M[f ] is the A-module obtained from M by the operation of restriction of
scalars (i.e., a m = f (a)m for any a A, m M ).
Proof. Trivial verification of the definition.
Let us apply this to our situation. Note that the k-linear map:
2 : K[] K, a + b b
is a K-derivation of K[] considered as a K-algebra. Here K is considered as a K[]module by means of the homomorphism 1 : K[] K, a + b 7 a. We identify a
K-point x X(K) with a homomorphism of k-algebras evx : O(X) K, 7 (x).
A tangent vector tx T (X)x is identified with a homomorphism evtx : O(X) K[].
Its composition with the derivation 2 : K[] K, a + b 7 b, is a K-derivation
of k[X]. Here K is considered as a O(X)-module via the homomorphism evx . This
defines a map:
T (X)x Derk (O(X), K)x
where the subscript x stands to remind us about the structure of a O(X)-module on
K. By definition,
Derk (O(X), K)x = { Homk (O(X), K) : (pq) = p(x)(q)+q(x)(p), p, q O(X)}.
Lemma 13.6. Assume X is an affine algebraic k-set. The map
T (X)x Derk (O(X), K)x
is a bijection.
Proof. Let Derk (O(X), K)x . We define a map f : O(X) K[] by the formula:
f (p) = p(x) + (p).
It is easy to verify that f is a homomorphism, and its composition with 1 : K[] K
is equal to evx . Thus f defines a tangent vector at x and the formula 7 f makes
the inverse of our map.
122
Now Derk (O(X), K)x has a structure of a K-module, defined by the formula
(a)(p) = a(p) for any a K, p O(X). We transfer this structure to T (X)x by
means of the bijection from Lemma 13.6. This structure of a K-module on T (X)x
is obviously independent (up to isomorphism) on the choice of equations defining X.
We leave to the reader to verify that this structure agrees with the one defined in the
beginning of the lecture.
Let us specialize our definition to the case when x X(k) (a rational point of
X). Then the kernel of the homomorphism x : O(X) k is a maximal ideal mx of
O(X) and O(X)/mx
= k. Let Derk (O(X), k) be a k-derivation of O(X). For
any p, q mx , we have
(p q) = p(x)(q) + q(x)(p) = 0.
Thus the restriction of to m2x is identical zero.
Lemma 13.7. Assume x X(k). The restriction map |mx , defines an isomorphism of k-linear spaces
T (X)x Homk (mx /m2x , k).
Proof. Since O(X)/m2x has a natural structure of a k-algebra there is a canonical homomorphism k O(X)/m2x such that its composition with the factor map
O(X)/m2x O(X)/mx = k is the identity. We shall identify k with the subring of O(X)/m2x by means of this map so that the restriction of the factor map
O(X)/m2x O(X)/mx to k is the identity. For any p O(X) we denote by px
the residue of p mod m2x . Obviously, px p(x) mx /m2x , so that for every linear
function f Homk (mx /m2x , k) we can define the map : mx /m2x k by setting for
any p O(X)
(p) = f (px p(x)).
Since for any p, q O(X), (px p(x))(qx q(x)) m2x , we have
(pq) = f (px qx p(x)q(x)) = f ((px p(x))(qx q(x))+p(x)(qx q(x))+q(x)(px p(x)))
= f (p(x)(qx q(x)) + q(x)(px p(x))) = p(x)f (qx q(x)) + q(x)f (px p(x)))
= p(x)(q) + q(x)(p).
We leave to the reader to verify that the constructed map f 7 is the needed
inverse.
Let f : X Y be a morphism of algebraic k-varieties (affine or projective).
Let x X(K), and y = fK (x) Y (K). By definition of a morphism, the map
fK[] : X(K[]) Y (K[]) induces a natural map
(df )x : T (X)x T (Y )y .
123
It is called the differential of f at the point x. If f : X Y is a morphism of affine
k-varieties corresponding to a homomorphism f : O(Y ) O(X) of k-algebras,
x X(K), y = fK (x) Y (K), then, after we use the bijection from Lemma 13.6, it
is immediately verified that the differential (df )x coincides with the map
Derk (O(X), K)x Derk (O(Y ), K)y
defined in Lemma 13.5, where f is the homomorphism f : O(Y ) O(X). This is
obviously a K-linear map.
Proposition 13.8. (Chain Rule). Let f : X Y, g : Y Z be morphisms of
algebraic k-varieties, x X(K), y = f (x) Y (K). Then
d(g f )x = (dg)y (df )x .
Proof. Immediately follows from the definition of a morphism.
Now we can define the tangent space for any quasi-projective algebraic set V
Pn (K). Here K, as usual, is a fixed algebraically closed field containing k. First, we
assume that V is affine. Choose an affine algebraic K-variety X such that I(X) is
radical and X(K) = V . Then we define the the tangent space T (V )x of V at x by
setting
T (V )x = T (X)x .
By Lemma 13.7, for every x V we have an isomorphism of K-linear spaces.
T (X)x
= Derk (O(X), K).
Since an isomorphism of affine varieties is defined by an isomorphism of their coordinate
algebras, we see that this definition is independent (up to isomorphism of linear spaces)
of a choice of equations defining X.
Lemma 13.9. Let A be a commutative K-algebra, M an A-module, and S a multiplicatively close subset of A. There is an isomorphism of AS -modules
Derk (A, M )S
= Derk (AS , MS )
Proof. Let : A M be a derivation of A. We assign to it the derivation of AS
defined by the familiar rule:
a (a)s (s)a
=
.
s
s2
This definition does not depend on the choice of a representative of the fraction as . In
fact, assume s00 (s0 a sa0 ) = 0. Then
0 = s00 (s0 a sa0 ) (s0 a sa0 )(s00 ) = s00 [(s0 a) (sa0 )] + (as0 a0 s)(s00 ).
124
(13.4)
(13.5)
The previous isomorphism suggests a definition of the tangent space of any quasiprojective algebraic k-set X.
Definition 13.2. The local ring of X at x X is the factor set
[
OX,x =
O(U )/R
xU
where U runs through the set of all open affine neighborhoods of x and the equivalence
relation R is defined as follows:
Let f O(U ), g O(V ), then
f g f W = g W for some open affine neighborhood of x contained in U V .
125
We shall call the equivalence class of f O(U ), the germ of f at x. The structure of
a ring in OX,x is induced by the ring structure of any O(U ). We take two elements of
OX.x , represent them by regular functions on a common open affine subset, multiply
or add them, and take the germ of the result. Let mX,x be the ideal of germs of
functions f O(U ) which vanish at x.
It follows from the definition that, for any open affine neighborhood U of x, the
natural map O(U ) OX,x , 7 x defines an isomorphism
OU,x
= OX,x .
Lemma 13.10.
126
In the case when x X(k), choosing an open affine neighborhood of x and applying
Lemma 13.7, we obtain
T (X)x = Homk (mX,x /m2X,x , k).
(13.7)
For any rational point x X(k), the right-hand side of (13.7) is called the Zariski
tangent space of X at x.
Let f : X Y be a regular map of algebraic k-sets, x X and y = f (x). Let
V be an open affine neighborhood of y and U be an open affine neighborhood of x
contained in f 1 (V ). The restriction of f to U defines a regular map f : U V . For
any O(V ), the composition with f defines a regular function f () on U . Let
f ()x OU,x be its germ at x. The homomorphism f : O(V ) OU,x extends to a
homomorphism f : OV,y OU,x of the local rings. It is clear that f (mV,y ) mU,x .
Composing this homomorphism with the isomorphisms OX,x
= OU,x and OY,y
= OV,y
we get a homomorphism of local rings
fx,y
: OY,y OX,x .
(13.8)
(13.9)
127
Let us now show that dimK T (X)x dim X for any irreducible algebraic set X
and the equality takes place for almost all points x (i.e., for all points belonging to a
Zariski open subset of X). For this, we may obviously assume that X is affine.
Let V = X(K) for some affine variety defined by a radical ideal in k[Z1 , . . . , Zn ].
The set T (V ) = X(K[]) is a subset of K[]n which can be thought as the vector
space K 2n . It is easy to see that T (X) is a closed algebraic subset of K 2n and the
map p = X() : T (X) X, is a regular map (check it !). Note that the fibre p1 (x)
is equal to the tangent space T (X)x . Applying the theorem about the dimension of
fibres of a regular map, we obtain
Proposition 13.12. There exists a number d such that
dimK T (X)x d
and the equality takes place for all points x belonging to an open subset of X.
We will show that the number d from above is equal to dim X.
Lemma 13.13. Let K be an algebraically closed field of characteristic p. Let F
K[Z1 , . . . , Zn ] with all the partial derivatives F/Zi equal to zero. If p = 0, then F
is a constant polynomial. If p > 0, then F = Gp for some polynomial G.
Proof. Write F =
r ar Z
r.
Then
X
F
=
ar (r ei )Z rei
Zi
r
where denotes the dot product of vectors and ei is the i-th unit vector. If this
polynomial is equal to zero, then ar (r ei ) = 0 for all r. Assume that ar 6= 0. If
F
char(k) = 0 this implies that r ei = 0. In particular, if all Z
= 0, we get r = 0,
i
i.e., F is a constant polynomial. If char(k) = p > 0, we obtain that p divides r ei ,
i.e., r = pr0 for some vector r0 . Thus
F =
X
r
where G =
ar Z r =
X 1/p 0
0
ar (Z r )p = (
ar Z r )p = Gp ,
1/p r0
r ar Z .
Theorem 13.14. Let X be an irreducible algebraic set and d = min{T (X)x }. Then
d = dim X.
128
F1
Z1 (x) . . . Zn (x)
...
...
...
.
J(x) =
...
...
...
Fr
Z1 (x)
...
Fr
Zn (x)
129
Problems.
1. Assume k = K is algebraically closed field of characteristic 0. Show that, up to a
projective automorphism of P2 (K), there are only two irreducible singular plane cubic
curves.
2. Prove that T (X Y )(x, y)
= T (X)x T (Y )y . Using this show that if x is a
nonsingular point of X and y is a nonsingular point of Y , then (x, y) is a nonsingular
point of X Y .
3. Let X be a closed subset of An (K), x = (a1 , . . . , an ) X and f : A1 (K) An (K)
given by t 7 (b1 t + a1 , . . . , bn t + an ). Let (Z r ) be the ideal of O(A1 (K))
= k[Z]
generated by the functions f (), I(X). Show that (a1 + b1 , . . . , an + bn )
K[]n is a tangent vector of X if and only if r > 1. Note that r can be interpreted as
the intersection multiplicity of X and the line f (A1 (K)) at x.
4. Suppose a hypersurface X = V (F ) of degree > 1 in Pn (K) contains a linear
subspace E of dimension r n/2. Show that X has singular points contained in E.
5. Find singular points of the Steiner quartic V (T02 T12 + T12 T22 + T02 T22 T0 T1 T2 T3 )
in P3 (K).
6. Let X be a surface in P3 (K). Assume that X contains three non-coplanar lines
passing through a point x X. Show that this point is singular.
7. Let Gk (r + 1, n + 1) be the Grassmann variety over k. For every M Gk (r + 1, n +
1)(K) show that the tangent space of Gk (r + 1, n + 1) at M is naturally identified
with HomK (M, K n+1 /M ).
130
Lecture 14
Local parameters
In this lecture we will give some other properties of nonsingular points. As usual we
fix an algebraically closed field K containing k and consider quasi-projective algebraic
k-sets, i.e. locally closed subsets of projective spaces Pnk (K).
Recall that a point x X is called nonsingular if dimK T (X)x = dimx X. When
x X(k) is a rational point, we know that T (X)x
= Homk (mX,x /m2X,x , k). Thus a
rational point is nonsingular if and only if
dimk mX,x /m2X,x = dimx X.
Let us see first that dimx X = dim OX,x . The number dim OX,x is denoted often
by codimx X and is called the codimension of the point x in X. The reason is simple.
If X is affine and px = Ker(evx ), then we have
dim O(X)p = sup{r : a chain px = p0 ) . . . ) pr of prime ideals in O(X)}.
This follows from the following.
Lemma 14.1. Let p be a prime ideal in a ring A. Then
dim Ap = sup{r : a chain p = p0 ) . . . ) pr of prime ideals in A}.
Proof. Let qr . . . q0 be the largest increasing chain of prime ideals in Ap . We
may assume that q0 is the maximal ideal m of A. Let pi be the pre-image of qi in A
under the natural homomorphism A Ap . Since p0 = p, we get a chain of prime
ideals p = p0 . . . pr . Conversely, any chain of such ideals in A generates an
increasing chain of prime ideals in Ap . It is easy to see that pi Aq = pi+1 Ap implies
pi = pi+1 . This proves the assertion.
In commutative algebra the dimension of Ap is called the height of the prime ideal
p.
131
132
Proposition 14.2.
codimx X + algdimk k(x) = dimx X.
Proof. We use induction on dimx X. Let p = Ker(evx ). If dimx X = 0, an open
affine neighborhood of x consists of finitely many points, p is a maximal ideal, k(x) is
algebraic over k, and codimx X = 0. This checks the assertion in this case. Assume
the assertion is true for all pairs (Y, y) with dimy Y < dimx X. If p = {0}, then
k(x) = Q(O(X)) and algdimk k(x) = dim X. Obviously, codimx X = 0. This checks
the assertion in this case. Assume that p 6= {0}. Let X 0 be an irreducible component
of X of dimension dimx X which contains x. Take an nonzero element p which
does not vanish on X 0 and consider the closed subset V () of X 0 containing x. By
Krulls Theorem, the dimension of each irreducible component of V () is equal to
dim X 0 1 = dimx X 1. Let Y be an irreducible component of V () containing
x and let q be the prime ideal in O(X) of functions vanishing on Y . There exists a
strictly decreasing chain of length codimx Y of prime ideals in O(Y ) descending from
the image of p in O(Y ) = O(X)/q. Lifting these ideals to prime ideals in O(X) and
adding q as the last ideal we get a chain of length 1 + codimx Y of prime ideals in
O(X) descending from p. By induction,
codimx Y + algdimk k(x) = dimx Y = dim X 1.
Under the natural homomorphism OX,x OY,x , the maximal ideal mX,x generates
the maximal ideal mY,x . This easily implies that the residue field of x in X and in Y
are isomorphic. This gives
codimx X +algdimk k(x) 1+codimx Y +algdimk k(x) = 1+dimx X 1 = dimx X.
(14.1)
Recall that algdimk k(x) = dim O(X)/p. Any increasing chain of prime ideals in
O(X)/p can be lifted to an increasing chain of prime ideals in O(X) beginning at p,
and after adding a chain of prime ideals descending from p gives an increasing chain of
prime ideals in O(X). This shows that codimx X +algdimk k(x) dimx X. Together
with the inequality (14.1), we obtain the assertion.
Corollary 14.3. Assume that k(x) is an algebraic extension of k. Then
dim OX,x = dimx X.
Now we see that a rational point is nonsingular if and only if
dimk mX,x /m2X,x = dim OX,x .
Proposition 14.4. Let (A, m) be a Noetherian local ring. Then
dim m/m2 dim A.
133
Proof. We shall prove it only for geometric local rings, i.e., when A
= Bp , where B
is a finitely generated k-algebra B and p is a prime ideal in B. This will be enough
for our applications. Thus we may assume that B = O(X) for some affine algebraic
k-variety X and p corresponds to some irreducible subvariety Y of X. Let K be some
algebraically closed field containing the field of fractions Q(O(X)/p). The canonical
homomorphism O(X) O(X)/p Q(O(X)/p) K defines a point x of the
algebraic k-set X(K) with k(x) = Q(O(X)/p). Thus we see that any geometric local
ring is isomorphic to the local ring OX,x of some affine algebraic k-set and its point
x.
Let X1 be an irreducible component of X(K) of dimension equal to dim X which
contains x. Since alg.dimk O(X)/p = dim O(X)/p = dim Y , we see that
dim OX,x = dimx X dim Y = dim X1 dim Y.
Suppose a1 , . . . , an generate the maximal ideal of OX,x . Let U be an open affine neighborhood of x such that a1 , . . . , an are represented by regular functions 1 , . . . , n on U .
Clearly, Y U = V (1 , . . . , n ). Applying Krulls Hauptsatz, we obtain that dim Y =
dim V (1 , . . . , n ) dim X1 n. This implies dim OX,x = dim X1 dim Y n
which proves the assertion. In fact, this proof gives more. By choosing elements
from 1 , . . . , n such that each does not vanish on any irreducible component of
V (1 , . . . , i1 ) containing x, we obtain that V (1 , . . . , n ) = dim Y , where n =
codimx X. Thus, Y is an irreducible component of V (1 , . . . , n ). Let q1 , . . . , qr be
prime ideals corresponding to other irreducible components of V (1 , . . . , n ). Let U be
an open subset of X obtained by deleting the irreducible components of V (1 , . . . , n )
different from Y . Then, replacing X with U , we may assume that V (1 , . . . , n ) = Y .
Thus p = rad(1 , . . . , n ) and replacing i s with their germs ai in OX,x we obtain
that m = rad(a1 , . . . , an ).
Definition 14.1. A Noetherian local ring (A with maximal ideal m) and residue field
= A/m is called regular if dim (m/m2 ) = dim A.
Thus a rational point x is nonsingular if and only if the local ring OX,x is regular.
For any point x X (not necessary rational) we define the Zariski tangent space
to be
(X)x = Homk(x) (mX,x /m2X,x , k(x))
considered as a vector space over the residue field k(x) = OX,x /mX,x .
We define the embedding dimension of X at x by setting
embdimx X = dimk(x) (X)x .
Note that for a rational point we have
T (X)x = (X)x k K.
(14.2)
134
(14.3)
Definition 14.2. A point x X is called regular if OX,x is a regular local ring, i.e.
embdimx X = codimx X.
Remark 14.5. We know that a rational point is regular if and only if it is nonsingular.
In fact, any nonsingular point is regular (see next Remark) but the converse is not
true. Here is an example. Let k be a field of characteristic 2 and a k which is not
a square. Let X be defined in A2k (K) by the equation Z12 + Z23 + a = 0. Taking the
r
X
aij fj ,
i = 1, . . . , r,
j=1
for some aij in m. Let R = (aij ) be the matrix of coefficients. Since (f1 , . . . , fr ) is a
solution of the homogeneous system of equations R x = 0, by Cramers rule,
det(R)fi = 0, i = 1, . . . , r.
135
However, det(R) = (1)r + a for some a m (being the value of the characteristic
polynomial of R at 1). In particular det(R) is invertible in A. This implies that fi = 0
for all i, i.e., M = {0}.
Corollary 14.8. 1. Let A be a local Noetherian ring and m be its maximal ideal.
Elements a1 , . . . , ar generate m if and only if their residues modulo m2 span m/m2 as
a vector space over k = A/m. In particular, the minimal number of generators of the
maximal ideal m is equal to the dimension of the vector space m/m2 .
Proof. Let M = m, N = (a1 , . . . , ar ). Since A is Noetherian, M is a finitely generated
A-module and N its submodule. By the assumption, M = mM +N . By the Nakayama
lemma, M = N .
Corollary 14.9. The maximal ideal of a Noetherian local ring of dimension n cannot
be generated by less than n elements.
Proof. This follows from Proposition 14.4.
Definition 14.3. A system of parameters in a local ring A is a set of n = dim A
elements (a1 , . . . , an ) generating an ideal whose radical is the maximal ideal, i.e.,
ms (a1 , . . . , an ) m
for some s > 0).
It follows from the proof of Proposition 14.4 that local rings OX,x always contain
a system of parameters. A local ring is regular, if and only if it admits a system of
parameters generating the maximal ideal. Such system of parameters is called a regular
system of parameters.
Let a1 , . . . , an be a system of parameters in OX,x , Choose an U be an open affine
neighborhood of x such that a1 , . . . , an are represented by some regular functions
1 , . . . , n on U . Then V (1 , . . . , n ) U is equal to the closure of x in U corresponding to the prime ideal p O(U ) such that OX,x
= O(U )p ). In fact, the radical
of (1 , . . . , n ) must be equal to p.
Example 14.10. 1. Let X be given by the equation Z12 + Z23 = 0 and x = (0, 0).
The maximal ideal mX,x is generated by the residues of the two unknowns. It is easy
to see that this ideal is not principal. The reason is clear: x is a singular point of X
and embdimx X = 2 > dimx X = 1. On the other hand, if we replace X by the set
given by the equation Z12 + Z23 + Z2 = 0, then mX,x is principal. It is generated by
the germ of the function Z1 . Indeed, Z12 = Z2 (Z22 + 1) and the germ of Z22 + 1 at
the origin is obviously invertible. Note that the maximal ideal m(X)x of O(X) is not
principal.
136
n
X
F
(x)(Zi ai ) + G(Z1 , . . . , Zn ),
Zi
i=1
where G(Z1 , . . . , Zn ) m2x . Thus the cosets dZi of Zi ai mod m2X,x form a basis
of the linear space mX,x /m2X,x and the germ Fx F (x) = F (Z1 , . . . , Zn ) F (x)
mod m2X,x is a linear combination of dZ1 , . . . , dZn with the coefficients equal to the
This is P
also the value at F of the derivation of k[Z1 , . . . , Zn ] defined by the tangent
vector i i Z
.
i
Let f : X = An (K) Y = Am (K) be a regular map given by a homomorphism
k[T1 , . . . , Tm ] k[Z1 , . . . , Zn ], Ti Pi (Z1 , . . . , Zn ).
Let x =
i i Zi
T (X)x , then
n
X
j=1
X X Pi
Pi
(x) =
(x)
(Ti ).
j
Zj
Zj
Tk
k=1 j=1
From this we infer that the matrix of the differential (df )x with respect to the bases
Z1
...
...
Pm
Z1
...
...
...
...
P1
Zn
...
.
...
Pm
Zn
fx,y
: OY,y OX,x .
137
Since fx (mY,y ) mX,x , we can define a homomorphism OY,y /mY,y OX,x /mX,x
and passing to the fields of quotients we obtain an extension of fields k(x)/k(y).
induces a linear map m
2
2
Also, fx,y
Y,y /mY,y mX,x /mX,x , where the target space is
considered as a vector space over the subfield k(y) of k(x), or equivalently a linear map
of k(x)-spaces mY,y /m2Y,y k (y)k(x) mX,x /m2X,x The transpose map defines a
linear map of the Zariski tangent spaces
dfxzar : (X)x (Y )y k(y) k(x).
(14.4)
(14.5)
This shows that we can identify (Y )x with a linear subspace of (X)x . Let
codim((Y )x , (X)x ) = dim (X)x dim (Y )x ,
codimx (Y, X) = codimx X codimx Y,
x (Y, X) = codimx (Y, X) codim((Y )x , (X)x ).
(14.6)
Then
dim (Y )x codimx Y = dim (X)x codimx X + x (Y, X).
Thus we obtain
Proposition 14.11. Let Y be a closed subset of X and x Y . Assume x is a
regular point of X, then x (Y, X) 0 and x is a regular point of Y if and only if
x (Y, X) = 0.
In particular, x is a regular point of Y if and only if the cosets of the germs of
the functions defining X in an neighborhood of x modulo m2X,y span a linear subspace
of codimension equal to codimx X codimx Y . Applying Nakayamas Lemma, we see
that this is the same as saying that X can be locally defined by codimx X codimx Y
equations in an open neighborhood of x whose germs are linearly independent modulo
m2X,x .
For example, if Y is a hypersurface in X in a neighborhood of x, i.e. codimx Y =
codimx X 1, then x is a regular point of Y if and only if Y is defined by one equation
in an open neighborhood of x whose germ does not belong to m2X,x .
138
(14.7)
(14.8)
Corollary 14.12. Let Y and Z be closed subsets of an algebraic set X which intersect
transversally at x X. Then
(i) the linear subspaces (Y )x , (Z)x intersect transversally in (X)x , i.e.,
codim((Y )x (Z)x , (X)x ) = codim((Y )x , (X)x )+codim((Y )x , (X)x );
(ii) x is a regular point of Y Z;
(iii) Y and Z are nonsingular at x.
Proof. We have
x (Y, X) = codimx (Y, X) codim((Y )x , (X)x ) 0,
x (Z, X) = codimx (Z, X) codim((Z)x , (X)x ) 0.
Since Y and Z intersect transversally at x, we obtain from (14.8)
codimx (Y, X) + codimx (Z, X) = codim((Y )x (Z)x , (X)x )
codim((Y )x , (X)x ) + codim((Z)x , (X)x ) codimx (Y, X) + codimx (Z, X).
(14.9)
This shows that all the inequalities must be equalities. This gives
codim((Y )x (Z)x , (X)x ) = codim((Y )x , (X)x ) + codim((Z)x , (X)x )
proving (i), and x (Y, X) = x (Z, X) = 0 proving (iii). By Theorem 11.21 of Lecture
11, we have dimx (Y Z) dimx X dimx (Y ) dimx (Z). Applying Proposition
14.2, we get codimx (Y Z) codimx Y + codimx Z. Together with inequality (14.9)
we obtain x (Y Z, X) = 0 proving assertion (ii).
139
Next we will show that every function from OX,x can be expanded into a formal
power series in a set of local parameters at x.
Recall that the k-algebra of formal power series in n variables k[[Z]] = k[[Z1 , . . . , Zn ]]
consists of all formal (infinite) expressions
P =
ar Z r ,
P (i)Q(j).
i+j=r
X
i
gi (x)fi +
X
ij
hij fi fj ,
140
where
hij OX,x . Continuing in this way, we will find a formal power series P =
P
j Pj such that
()
for any r 0.
for any r 0.
n n0 .
141
We omit the proof which can be found in any text-book in commutative algebra.
Definition 14.5. Let : OX,x k[[Z1 , . . . , Zn ]] be the injective homomorphism
constructed in Theorem 14.1. The image (f ) of an element f OX,x is called the
Taylor expansion of f at x with respect to the local parameters f1 , . . . , fn .
Corollary 14.15. The local ring OX,x of a regular point does not have zero divisors.
Proof. OX,x is isomorphic to a subring of the ring k[[Z]] which, as is easy to see, does
not have zero divisors .
Corollary 14.16. A regular point of an algebraic set X is contained in a unique
irreducible component of X.
Proof. This immediately follows from Corollary 14.3. Indeed, assume x Y1 Y2
where Y1 and Y2 are irreducible components of X containing the point x. Replacing
X by an open affine neighborhood, we may find a regular function f1 vanishing on Y1
but not vanishing on the whole Y2 . Similarly, we can find a function f2 vanishing on
X \ Y1 and not vanishing on the whole Y1 . The product f = f1 f2 vanishes on the
whole X. Thus the germs of f1 and f2 are the zero divisors in OX,x . This contradicts
the previous corollary.
Remark 14.17. Note the analogy with the usual Taylor expansion which we learn
in Calculus. The local parameters are analogous to the differences xi = xi ai .
The condition f [P ]r (f1 , . . . , fn ) mr+1
X,x is the analog of the convergence: the
difference between the function and its truncated Taylor expansion vanishes at the
point x = (a1 , . . . , an ) with larger and larger order. The previous theorem shows that
a regular function on a nonsingular algebraic set is like an analytic function: its Taylor
expansion converges to the function.
For every commutative ring A and its proper ideal I, one can define the I-adic
formal completion of A as follows. Let pn,k : A/I n+1 A/I k+1 be the canonical
homomorphism of factor rings (n k). Set
Y
AI = {(. . . , ak , . . . , an . . .)
(A/I r+1 ) : pn,k (an ) = ak for all n k}.
r0
It is easy to see that AI is a commutative ring with respect to the addition and
multiplication defined coordinate-wise. We have a canonical homomorphism:
i : A AI , a 7 (a0 , a1 , . . . , an , . . .)
where an = residue of a modulo I n+1 . Note the analogy with the ring of p-adic
numbers which is nothing else as the formal completion of the local ring Z(p) of
rational numbers ab , p - b.
142
A , A.
is equal to the closure
Note that the ring A is local. Its unique maximal ideal m
m
is isoof m in A. It consists of elements (0, a1 , . . . , an , . . .). The quotient A/
m
is of course
morphic to A/m = . The canonical homomorphism (A) A/
(a0 , a1 , . . . , an , . . .) a0 .
(14.10)
143
X,x , K) Derk (OX,x , K) corresponding
We have a canonical linear map : Derk (O
to the inclusion map of the ring into its completion. Note that for any local ring (A, m)
which contains k, the canonical homomorphism of A-modules
A : Derk (A, K) Homk (m/m2 , K)
is injective. In fact, if M is its kernel, then, for any M we have (m) = 0. This
implies that for any a m and any x A, we have 0 = (ax) = a(x) + x(a) =
a(x). Thus a = 0. This shows that mM = 0, and by Nakayamas lemma we get
M = 0. Composing with OX,x we obviously get OX,x . Since the latter is injective,
is injective. Now we show that it is surjective. Let Derk (OX,x , K). Since its
restriction to m2X,x is zero, we can define (a + m2X,x ) for any a OX,x . For any
such that () = .
derivation of OX,x /m
So, we obtain an isomorphism of K-vector spaces:
X,x , K)
Derk (O
= Derk (OX,x , K).
X,x
By Cohens Theorem, O
= k(x)[[T1 , . . . , Tn ]], where the pre-image of the field of
X,x isomorphic to k(x) under the projection
constant formal series is a subfield L of O
to the residue field. It is clear that the pre-image of the maximal ideal (T1 , . . . , Tn )
X,x , K) be the subspace of Derk (O
X,x , K)
is the maximal ideal of OX,x . Let DerL (O
of derivation trivial on L. Using the same proof as in Lemma 13.3 of Lecture 13, we
X,x , K)
show that DerL (O
= (X)x . Now we have an exact sequence, obtained by
restrictions of derivations to the subfield L:
X,x , K) Derk (O
X,x , K) Derk (L, K).
0 DerL (O
(14.11)
It is easy to see that dimK Derk (L, K) = algdimk L = algdimk k(x). In fact,
Derk (k(t1 , . . . , tr ), K)
= Kr
(each derivation is determined by its value on each ti ). Also each derivation can be
uniquely extended to a separable extension. Thus exact sequence (14.11) gives
X,x , K) = dimK Derk (O
X,x , K) embdimx X + algdimk k(x).
dimK Derk (O
This implies that embdimx (X) = dim OX,x and hence OX,x is regular.
Let (X, x) be a pair that consists of an algebraic set X and its point x X.
Two such pairs are called locally isomorphic if the local rings OX,x and OY,y are
isomorphic. They are called formally isomorphic if the completions of the local rings
are isomorphic. Thus any pair (X, x) where x is a nonsingular point of X is isomorphic
to a pair (An (K), 0) where n = dimx X. Compare this with the definition of a smooth
(or complex manifold).
144
145
1 , . . . , n+1 on D(). Since OX,x is factorial, we may cancel the germs (i )x by
their common divisor to assume that not all of them are divisible by the germ x of
. The resulting functions define the same map to Pn (K). It is not defined at the
set of common zeroes of the functions i . Its intersection with Z cannot contain any
open neighborhood of x, hence is a proper closed subset of Z. This shows that we
can extend f to a larger open subset contradicting the maximality of U .
Corollary 14.21. Any rational map of a nonsingular curve to a projective set is a
regular map. In particular, two nonsingular projective curves are birationally isomorphic
if and only if they are isomorphic.
This corollary is of fundamental importance. Together with a theorem on resolution
of singularities of a projective curve it implies that the set of isomorphism classes of
field extensions of k of transcendence degree 1 is in a bijective correspondence with
the set of isomorphism classes of nonsingular projective algebraic curves over k.
Problems.
1. Using Nakayamas Lemma prove that a finitely generated projective module over a
local ring is free.
2. Problem 6 from Shafarevich, Chap. II, 3.
3. Let A be a ring with a decreasing sequence of ideals A = I0 I1 In
such that Ii Ij Ii+j for all i, j. Let GrF (A) =
i=0 Ii /Ii+1 with the obvious ring
structure making GrF (A) a graded ring. Show that a local ring (A, m) of dimension
n is regular if and only GrF (A)
= [T1 , . . . , Tn ], where Ii = mi .
4. Let X = V (F ) A2 (K) where F = Z13 Z2 (Z2 + 1). Find the Taylor expansion
at (0, 0) of the function Z2 mod (F ) with respect to the local parameter Z1 mod (F ).
5. Give an example of a singular point x X such that there exists an injective
homomorphism OX,x k[[Z1 ]]. Give an example of a curve X and a point x X
for which such homomorphism does not exist.
6. Let X = V (Z1 Z2 + Z32 ) K 4 . Show that the line V (Z1 , Z3 ) X cannot be
defined by one equation in any neighborhood of the origin.
7. Show that Theorem 15.10 is not true for singular projective algebraic curves.
8*. Let X = V (Z1 Z2 + F (Z1 , Z2 )) A2 (K) where F is a homogeneous polynomial
X,x
of degree 3. Show that O
= K[[T1 , T2 ]]/(T1 T2 ) and hence the singularity (X, 0)
and (V (Z1 Z2 ), 0) are formally isomorphic.
146
Lecture 15
Projective embeddings
Lecture we shall address the following question: Given a projective algebraic k-set X,
what is the minimal N such that X is isomorphic to a closed subset of PN
k (K)? We
shall prove that N 2 dim X + 1. For simplicity we shall assume here that k = K.
Thus all points are rational, the kernel of the evaluation maps is a maximal ideal, the
tangent space is equal to the Zariski tangent space, a regular point is the same as a
nonsingular point.
Definition 15.1. A regular map of projective algebraic sets f : X Pr (K) is called
an embedding if it is equal to the composition of an isomorphism f 0 : X Y and the
identity map i : Y Pr (K), where Y is a closed subset of Pr (K).
Theorem 15.1. A finite regular map f : X Y of algebraic sets is an isomorphism
if and only if it is bijective and for every point x X the differential map (df )x :
T (X)x T (Y )f (x) is injective.
Proof. To show that f is an isomorphism it suffices to find an open affine covering
of Y such that for any open affine subset V from this covering the homomorphism of
rings f : O(V ) O(f 1 (V )) is an isomorphism. The inverse map will be defined
by the maps of affine sets V f 1 (V ) corresponding to the inverse homomorphisms
(f )1 : O(f 1 (V )) O(V ). So we may assume that X and Y are affine and also
irreducible.
Let x X and y = f (x). Since f is bijective, f 1 (y) = {x}. The homomorphism
f induces the homomorphism of local rings fy : OY,x OX,x . Let us show that it
makes OX,x a finite OY,x -module. Let m O(Y ) be the maximal ideal corresponding
to the point y and let S = O(Y ) \ m. We know that OY,x = O(Y )S , and, since
finiteness is preserved under localizations, O(X)f (S) is a finite OY,x -module. I claim
that O(X)f (S) = OX,x . Any element in OX,x is represented by a fraction /
Q(O(X)) where (x) 6= 0. Since the map f is finite and bijective it induces a
147
148
bijection from the set (V ()) of zeroes of to the closed subset f (V ()) of Y . Since
y 6 f (V ()) we can find a function g S vanishing on f (V ()). By Nullstellensatz,
f (g)r = for some r > 0 and some O(X). Therefore we can rewrite the
fraction / in the form /f (g)r showing that it comes from O(X)f (S) . This
proves the claim.
By assumption fy : OY,x OX,x induces a linear surjective map:
t
where t stands for the transpose map of the dual vector spaces. Let h1 , . . . , hk be a
set of local parameters of Y at the point y. Their images fy (h1 ), . . . , fy (hk ) in mX,x
span mX,x /m2X,x . As follows from Lecture 14, this implies that fy (h1 ), . . . , fy (hk )
generate mX,x . Therefore,
fy (mY,y )OX,x = mX,x .
Since fy (OY,y ) contains constant functions, and OX,x = k + mX,x , we get
OX,x = fy (OY,y ) + mY,y OX,x .
Having proved that OX,x is a finitely generated OY,y -module we may apply Nakayamas
lemma to obtain that
OX,x = fy (OY,y ).
Therefore the map fy : OY,y OX,x is surjective. It is obviously injective. Let
1 , . . . , m be generators of the O(Y )-module O(X). The germs (i )x belong to
OX,x = f (OY,x ) allowing us to write (i )x = f ((i )y ), where i are regular functions on some affine open neighborhood V of f (x). This shows that the germs of i
and f (i ) at the point x are equal. Hence, after replacing V by a smaller set V 0 if
needed, we can assume that i = f (i ) for some open subset U of f 1 (V ). Since
X is irreducible we can further assume that U = f 1 (V ). If we replace again V by
a principal open subset D(h) Y , we get U = D(f (h)), O(V ) = O(Y )h , O(U ) =
O(X)f (h) , and the functions i |U generate O(U ) as a module over O(V ). This implies that f : O(V ) O(f 1 (V )) is surjective, hence an isomorphism. This proves
the assertion.
Remark 15.2. The assumption of finiteness is essential. To see this let us take X to be
the union of two disjoint copies of affine line with the origin in the second copy deleted,
and let Y = V (Z1 Z2 ) be the union of two coordinate lines in A2 (K). We map the
first copy isomorphically onto the lines Z1 = 0 and map the second component of X
isomorphically onto the line Z2 = 0 with the origin deleted. It is easy to see that all
the assumptions of Theorem 15.1 are satisfied except the finiteness. Obviously, the
map is not an isomorphism.
149
Definition 15.2. We say that a line ` in Pn (K) is tangent to an algebraic set X at a
point x X if T (`)x is contained in T (X)x (both are considered as linear subspaces
of T (Pn (K))x ).
of K n+1 . For
Let E be a linear subspace in Pn (K) defined by a linear subspace E
the tanany point x = (a0 , . . . , an ) E defined by the line Lx = K(a0 , . . . , an ) in E,
gent space T (E)x can be identified with the factor space HomK (Lx , E/K(a0 , . . . , an )
K n+1 identifies it naturally with
(see Example 13.2 of Lecture 13). The inclusion E
the subspace of T (Pn (K))x = HomK (Lx , K n+1 /Lx ). Now let X be a projective subset of Pn (K) defined by a system of homogeneous equations F1 (T0 , . . . , Tn ) = . . . =
Fm (T0 , . . . , Tn ) = 0 and let x X. Then the tangent space T (X)x can be identified
with the subspace of T (Pn (K))x defined by the equations
n
X
Fi
j=0
Tj
(x)bj = 0,
i = 1, . . . m.
(15.1)
is contained in
Now we see that a line E is tangent to X at the point x if and only if E
the space of solutions of (15.1) In particular we obtain that the union of lines tangent
to X at the point x is the linear subspace of Pn (K) defined by the system of linear
homogeneous equations
n
X
Fi
j=0
Tj
(x)Tj = 0,
i = 1, . . . m.
(15.2)
150
K n /K(x1 , . . . , xn1 ) induced by the projection K n+1 K n . It is clear that its kernel is spanned by Kx + K(0, . . . , 0, 1)/Kx. But this is exactly the tangent space of
the line ` spanned by the points x = (x0 , . . . , xn ) and a = (0, . . . , 0, 1). Thus the
differential of the restriction of pa to X is injective if and only if the tangent space of
the line ` is not contained in the tangent space T (X)x . This proves the assertion.
Lemma 15.4. Let X be a quasi-projective algebraic subset of Pn (K) and x X be
its nonsingular point. Then ET(X)x is a projective subspace in Pn (K) of dimension
equal to d = dimx X.
Proof. We know that ET(X)x is the subspace of Pn (K) defined by equation(15.2). So
it remains only to compute the dimension of this subspace. Since x is a nonsingular
point of X, the dimension of T (X)x is equal to d. Now the result follows from
comparing equations (15.1) and (15.2). The first one defines the tangent space T (X)x
and the second ET(X)x . The (linear) dimension of solutions of both is equal to
d + 1 = n + 1 rank(
Fi
)(x) = dimK T (X)x + 1 = dim ET(X)x + 1.
Tj
Note that the previous lemma shows that one can check whether a point of a
projective set X is nonsingular by looking at the Jacobian matrix of homogeneous
equations defining X.
Let
Z = {(x, y, z) Pn (K) Pn (K) Pn (K) : x, y, z `
This is a closed subset of Pn (K)Pn (K)Pn (K) defined by the equations expressing
the condition that three lines x = (x0 , . . . , xn ), y = (y0 , . . . , yn ), z = (z0 , . . . , zn ) are
linearly dependent. The tri-homogeneous polynomials defining Z are the 3 3-minors
of the matrix
T0 . . . Tn
T00 . . . Tn0 .
T000 . . . Tn00
Let p12 : Z Pn (K) Pn (K) be the projection map to the product of the first two
factors. For any (x, y) Pn (K) Pn (K)
(
< x, y > if x 6= y,
p3 (p1
,
12 ((x, y))) =
Pn (K)
if x = y
where < x, y > denotes the line spanned by the points x, y.
151
Let X be a closed subset of Pn (K). We set
SechX = p1
12 (X X \ X ),
SecX = closure of SechX in Z.
The projection p12 and the projection p3 : Z Pn (K) to the third factor define the
regular maps
p : SecX X X, q : SecX Pn (K).
For any (x, y) X X \ X the image of the fibre p1 (x, y) under the map q is
equal to the line < x, y >. Any such lines is called a secant of X. The union of all
honest secants of X is equal to the image of SechX under the map q. The closure of
this union is equal to q(secX ). It is denoted by Sec(X) and is called the secant variety
of X.
Lemma 15.5. Let X be an irreducible closed subset of Pn (K). The secant variety
Sec(X) is an irreducible projective algebraic set of dimension 2 dim X + 1.
Proof. It is enough to show that sechX is irreducible. This would imply that secX and
Sec(X) are irreducible, and by the theorem on dimension of fibres
dim SechX = dim(X X) + 1 = 2 dim X + 1.
This gives
dim Sec(X) dim Sec(X) = dim sechX = 2 dim X + 1.
To prove the irreducibility of sechX we modify a little the proof of Lemma 12.7 of
Lecture 12. We cannot apply it directly since sechX is not projective set. However,
the map ph : sechX X X \ X is the restriction of the projection sets (X X \
X ) Pn (K) X X \ X . By Chevalleys Theorem from Lecture 9, the image
of a closed subset of sechX is closed in X X \ X . Only this additional property of
the map f : X Y was used in the proof of Lemma 12.7 of Lecture 12.
Lemma 15.6. The tangential variety Tan(X) of an irreducible projective algebraic set
of Pn (K) is an irreducible projective set of dimension 2 dim X.
Proof. Let Z X Pn (K) Pn (K) Pn (K) be a closed subset defined by
equations (1), where x is considered as a variable point in X. Consider the projection
of Z to the first factor. Its fibres are the embedded tangent spaces. Since X is
nonsingular, all fibres are of dimension dim X. As in the case of the secant variety
we conclude that Z is irreducible and its dimension is equal to 2 dim X. Now the
projection of Z to Pn is a closed subset of dimension 2 dim X. It is equal to the
tangential variety Tan(X).
152
Now everything is ready to prove the following main result of this Lecture:
Theorem 15.7. Every nonsingular projective d-dimensional algebraic set X can be
embedded into P2d+1 .
Proof. The idea is very simple. Let X Pn (K), we shall try to project X into a
lower-dimensional projective space. Assume n > 2d + 1. Let a Pn (K) \ X. By
Lemma 15.3, the projection map
pa : X Y Pn1 (K)
is an isomorphism unless either x lies on a honest secant of X or in the tangential
variety of X. Since all honest secants are contained in the secant variety Sec(X) of
X, and
dim Sec(X) 2 dim X + 1 < n, dim Tan(X) 2 dim X < n,
we can always find a point a 6 X for which the map pa is an isomorphism. Continuing
in this way, we prove the theorem.
Corollary 15.8. Every projective algebraic curve (resp. surface) is isomorphic to a
curve (resp. a surface) in P3 (K) (resp. P5 (K)).
Remark 15.9. The result stated in the Theorem is the best possible for projective
sets. For example, the affine algebraic curve: V (T12 + Fn (T2 )) = 0, where Fn is a
polynomial of degree n > 4 without multiple roots, is not birationally isomorphic to
any nonsingular plane projective algebraic curve. Unfortunately, we have no sufficient
tools to prove this claim. Let me give one more unproven fact. To each nonsingular
projective curve X one may attach an integer g 0, called the genus of X. If K = C
is the field of complex numbers, the genus is equal to the genus of the Riemann surface
associated to X. Each compact Riemann surface is obtained in this way. Now for any
plane curve V (F ) P2 (K) of degree n one computes the genus by the formula
g=
(n 1)(n 2)
.
2
153
Theorem 15.10. Let X Pn (K) be a nonsingular irreducible closed subset of Pn (K).
Then
Sec(X) = Sec(X)h Tan(X).
Proof. Since Sec(X) is equal to the closure of an irreducible variety Sec(X)h and
Tan(X) is closed, it is enough to prove that Sec(X)h Tan(X) is a closed set.
Let Z ne the closed subset of X Pn (K) considered in the proof of Lemma 15.6.
Its image under the projection to X is X, and its fibre over a point x is isomorphic
to the embedded tangent space ET(X)x . Its image under the projection to Pn is the
variety Tan(X). We can view any point (x, y) = ((x0 , . . . , xn ), (y0 , . . . , yn )) ET(X)
as a pair x + y K[]n+1 satisfying the equations Fi (T ) = 0. Note that for X = Pn
we have ET(X) = Pn Pn . Consider a closed subset Z of ET(X)ET (X)ETPn (K)
defined by the equations
rank[x + y, x0 + y 0 , x00 + y 00 ] < 3,
(15.3)
where the matrix is of size 3(n+1) with entries in K[]. The equations are of course
the 3 3-minors of the matrix. By Chevalleys Theorem, the projection Z 0 of Z to
ET(X) ET(X) is closed. Applying again this theorem, we obtain that the projection
of Z 0 to Pn is closed. Let us show that it is equal to Sech (X) Tan(X).
It is clear that the image (x, x0 , x00 ) of z = (x + y, x0 + y 0 , x00 + y 0 ) in X X X
satisfies rank[x, x0 , x00 ] < 3. This condition is equivalent to the following. For any
subset I of three elements from the set {0, . . . , n} let |xI + yI , x0I + yI0 , x00I + yI00 |
be the corresponding minor. Then equation (15.3) is equivalent to the equations
|xI + yI , x0I + yI0 , x00I + yI00 | = 0.
Or, equivalently,
|xI , x0I , x00I | = 0,
|xI , yI0 , x00I | + |xI , x0I , yI00 | + |yI , x0I , x00I | = 0.
(15.4)
(15.5)
Suppose equations (15.4) and (15.5) are satisfied. Then (15.4)) means that the point
x00 Pn lies in the line spanned by the points x, x0 or rank[x, x0 ] = 1. In the first case
we obtain that x00 Sech (X). Assume x = x0 as points in Pn . Then (15.5) gives
|xI , x00I , yI0 yI | = 0. Since (x, y) and (x, y 0 ) lie in ET(X)x , we obtain that x00 lies
on the line spanned by a point x and a point in ET(X)x . Hence x00 ET(X)x . This
proves the assertion.
Remark 15.11. If X is singular, the right analog of the embedded tangent space ET(X)
is the tangent cone CT (X)x . It is defined as the union of limits of the lines < x, y >
where y X. See details in Shafarevichs book, Chapter II, 1, section 5.
154
3d
.
2
155
The following result of F. Zak classifies Severi varieties in characteristic 0:
Theorem 15.15. Assume char(K) = 0. Each Severi variety is isomorphic to one of
the following four varieties:
(n = 2) the Veronese surface v2 (P2 (K)) P5 (K);
(n = 4) the Segre variety s2,2 (P2 (K) P2 (K)) P8 (K);
(n = 8) the Grassmann variety G(2, 6) P14 (K) of lines in P5 (K);
(n = 16) the E6 -variety X in P2 (K).
The last variety (it was initially missing in Zaks classification and was added to
the list by R. Lazarsfeld) is defined as follows. Choose a bijection between the set of
27 lines on a nonsingular cubic surface and variables T0 , . . . , T26 . For each triple of
lines which span a tri-tangent plane form the corresponding monomial Ti Tj Tk . Let F
be the sum of such 45 monomials. Its set of zeroes in P26 (K) is a cubic hypersurface
Y = V (F ). It is called the Cartan cubic. Then X is equal to the set of singularities
of Y (it is the set of zeroes of 27 partial derivatives of F ) and Y equals Sec(X).
From the point of view of algebraic group theory, X = G/P , where G is a simply
connected simple algebraic linear group of exceptional type E6 , and P its maximal
parabolic subgroup corresponding to the dominant weight defined by the extreme
vertex of one of the long arms of the Dynkin diagram of the root system of G. The
space P26 (K) is the projectivization of the representation of G with highest weight .
We only check that all the four varieties from Theorem 15.15 are in fact Severi
varieties. Recall that the Veronese surface can be described as the space of 3 3
symmetric matrices of rank 1 (up to proportionality). Since a linear combination of
two rank 1 matrices is a matrix of rank 2, we obtain that the secant variety is
contained in the cubic hypersurface in P5 defining matrices of rank 2. Its equation
is the symmetric matrix determinant. It is easy to see that the determinant equation
defines an irreducible variety. Thus the dimension count gives that it coincides with
the determinant variety. Similarly, we see that the secant variety of the Segre variety
coincides with the determinant hypersurface of a general 3 3 matrix. The third
variety can be similarly described as the variety of skew-symmetric 6 6 matrices of
rank 2. Its secant variety is equal to the Pfaffian cubic hypersurface defining skewsymmetric matrices of rank < 6. Finally, the secant variety of the E6 -variety is equal
to the Cartan cubic. Since each point of the Severi variety is a singular point of the
cubic, the restriction of the cubic equation to a secant line has two multiple roots.
This easily implies that the line is contained in the cubic. To show that the secant
variety coincides with the cartan cubic is more involved, One looks at the projective
linear representation of the exceptional algebraic group G of type E6 in P26 defining
156
the group G. One analyzes its orbits and shows that there are only three orbits: the
E6 -variety X, the Cartan cubic with X deleted and P26 with Cartan cubic deleted.
Since the secant variety is obviously invariant under the action of G, it must coincide
with the Cartan cubic.
Note that in all four cases the secant variety is a cubic hypersurface and its set of
singular points is equal to the Severi variety. In fact, the previous argument shows that
the secant variety of the set of singular points of any cubic hypersurface is contained
in the cubic. Thus Theorem 15.15 gives a classification of cubic hypersurfaces in Pn
whose set of singular points is a smooth variety of dimension 2(n 2)/3.
There is a beautiful uniform description of the four Severi varieties. Recall that
a composition algebra is a finite-dimensional algebra A over a field K (not necessary
commutative or associative) such that there exists a non-degenerate quadratic form
: A K such that for any x, y A
(x y) = (x)(y).
According to a classical theorem of A. Hurwitz there are four isomorphism classes
of composition algebras over a field K of characteristic 0: K, Co, Ha and Oc of
dimension 1, 2, 4 and 8, respectively. Here
Co = K K, (a, b) (a0 , b0 ) = (aa0 bb0 , ab0 + a0 b),
Ha = Co Co, (x, y) (x0 , y 0 ) = (x x0 y y 0 , x y 0 + y x
0 ),
0 ),
Oc = Ha Ha, (h, g) (h0 , g 0 ) = (h h0 g g 0 , h g 0 + g h
where for any x = (a, b) Co we set x
= (a, b), and for any h = (x, y) Ha we
= (
set h
x, y). The quadratic form is given by
(x) = x x
,
h0 )
where x
is defined as above for A = Ca and H, x
= x for A = K, and x
= (h,
0
for any x = (h, h ) Oc.
For example, if K = R, then Co
= C (complex numbers), Ha
= H (quaternions),
Oc = O (octonions or Cayley numbers).
For every composition algebra A we can consider the set H3 (A) of Hermitian
3 3-matrices (aij ) with coefficients in A, where Hermitian means aij = a
ji . Its
dimension as a vector space over K equals 3 + 3r, where r = dimK A. There is
a natural definition of the rank of a matrix from H3 (A). Now Theorem 15.15 says
that the four Severi varieties are closed subsets of P3r+2 defined by rank 1 matrices in
H3 (A). The corresponding secant variety is defined by the homogeneous cubic form
representing the determinant of the matrix.
157
Let us define Pn (A) for any composition algebra as An+1 \{0}/A . Then one view
the four Severi varieties as the Veronese surfaces corresponding to the projective
planes over the four composition algebra.
As though it is not enough of these mysterious coincidences of the classifications,
we add one more. Using the stereographic projection one can show that
P1 (R) = S 1 ,
P1 (C) = S 2 ,
P1 (H) = S 4 ,
P1 (O) = S 8 ,
158
Lecture 16
Blowing up and resolution of
singularities
Let us consider the projection map pa : Pn (K) \ {a} Pn1 (K). If n > 1 it is
impossible to extend it to the point a. However, we may try to find another projective
set X which contains an open subset isomorphic to Pn (K) \ {a} such that the map
pa extends to a regular map pa : X Pn1 (K). The easiest way to do it is to
consider the graph Pn (K) \ {a} Pn1 (K) of the map pa and take for X its
closure in Pn (K)Pn1 (K). The second projection map X Pn1 (K) will solve our
problem. It is easy to find the bi-homogeneous equations defining X. For simplicity
we may assume that a = (1, 0, . . . , 0) so that the map pa is given by the formula
(x0 , x, . . . , xn ) (x1 , . . . , xn ). Let Z0 , . . . , Zn be projective coordinates in Pn (K)
and let T1 , . . . , Tn be projective coordinates in Pn1 (K). Obviously, the graph is
contained in the closed set X defined by the equations
Zi Tj Zj Ti = 0, i, j = 1, . . . , n.
(16.1)
The projection q : X Pn1 (K) has the fibre over a point t = (t1 , . . . , tn ) equal to
the linear subspace of Pn (K) defined by the equations
Zi tj Zj ti = 0, i, j = 1, . . . , n.
(16.2)
Assume that ti = 1. Then the matrix of coefficients of the system of linear equations
(16.2) contains n 1 unit columns so that its rank is equal to n 1. This shows
that the fibre q 1 (t) is isomorphic, under the first projection X Pn (K), to the line
spanned by the points (0, t1 , . . . , tn ) and (1, 0, . . . , 0). On the other hand the first
projection is an isomorphism over Pn (K) \ {0}. Since X is irreducible (all fibres of
q are of the same dimension), we obtain that X is equal to the closure of . By
plugging z1 = . . . zn in equations (16.2) we see that the fibre of p over the point
159
X = T1 /T0 ,
Y = T0 /T1 .
If L : Z2 tZ1 = 0 is the line in A2 (K) through the origin with slope t, then the
pre-image of this line under the projection : B A2 (K) consists of the union of
isomorphic to L
two curves, the fibre E
= P1 (K) over the origin, and the curve L
intersects E at the point ((0, 0), (1, t)) V0 . The pre-image
under . The curve L
of each line L with the equation tZ2 Z1 consists of E and the curve intersecting
E at the point ((0, 0), (t, 1)) V1 . Thus the points of E can be thought as the set
of slopes of the lines through (0, 0). The infinite slope corresponding to the line
Z1 = 0 is the point (0, 1) V1 E.
Let I be an ideal in a commutative ring A. Each power I n of I is a A-module and
I n I r I n+r for every n, r 0. This shows that the multiplication maps I n I r
I n+r define a ring structure on the direct sum of A-modules
A(I) = n0 I n .
Moreover, it makes this ring a graded algebra over A = A(I)0 = I 0 . Its homogeneous
elements of degree n are elements of I n .
Assume now that I is generated by a finite set f0 , . . . , fn of elements of A. Consider
the surjective homomorphism of graded A-algebras
: A[T0 , . . . , Tn ] A(I)
161
defined by sending Ti to fi . The kernel Ker() is a homogeneous ideal in A[T0 , . . . , Tn ].
If we additionally assume that A is a finitely generated algebra over a field k, we can
interpret Ker() as the ideal defining a closed subset in the product X Pnk where X
is an affine algebraic variety with O(X)
= A. Let Y be the subvariety of X defined
by the ideal I.
Definition 16.1. The subvariety of X Pnk defined by the ideal Ker() is denoted
by BY (X) and is called the blow-up of X along Y . The morphism : BY (X) X
defined by the projection X Pnk X is called the monoidal transformation or the
-process or the blowing up morphism along Y .
Let us fix an algebraically closed field K containing k and describe the algebraic
set BY (X)(K) as a subset of X(K)Pn (K). Let Ui = X (Pn (K))i and BY (X)i =
BY (X) Ui . This is an affine algebraic k-set with
O(BY (X)i )
= O(X)[T0 /Ti . . . . , Tn /Ti ]/Ker()i
where Ker()i is obtained from the ideal Ker() by dehomogenization with respect
to the variable Ti . The fact that the isomorphism class of BY (X) is independent of
the choice of generators f0 , . . . , fn follows from the following
Lemma 16.1. Let Y X Pnk (K) and Y 0 X Prk (K) be two closed subsets defined
by homogeneous ideals I O(X)[T0 , . . . , Tn ] and J O(X)[T00 , . . . , Tr0 ], respectively.
Let p : Y X and p0 : Y 0 X be the regular maps induced by the first projections
X Pnk (K) X and X Prk (K) X. Assume that there is an isomorphism of
graded O(X)-algebras : O(X)[T00 , . . . , Tr0 ]/I 0 O(X)[T0 , . . . , Tn ]/I. Then there
exists an isomorphism f : Y Y 0 such that p = p0 f .
Proof. Let t0i = Ti0 mod I 0 , ti = Ti mod I, and let
(t0i ) = Fi (t1 , . . . , tn ), i = 0, . . . , r,
for some polynomial Fi [T0 , . . . , Tn ]. Since is an isomorphism of graded O(X)algebras the polynomials Fi (T ) are linear and its coefficients are regular functions on
X. The value of Fi at a point (x, t) = (x, (t0 , . . . , tn )) in X Pnk (K) is defined
by plugging x into the coefficients and plugging t into the unknowns Tj . Define
f : Y Y 0 by the formula:
f (x, t) = (x, (F0 (x, t), . . . , Fn (x, t))).
Since is invertible, there exist linear polynomials Gj (T ) O(X)[T00 , . . . , Tr0 ], j =
0, . . . , n, such that
Fi (G0 (t00 , . . . , t0n ), . . . , Gn (t0 , . . . , t0n )) = t0i , i = 0, . . . , r,
To find explicitly the equations of the blow-up BY (X), we need to make some
assumptions on X and Y .
163
Definition 16.2. Let A be a commutative ring. A sequence of elements a1 , . . . , an A
is called a regular sequence if the ideal generated by a1 , . . . , an is a proper ideal of A
and, for any i = 1, . . . , n, the image of ai in A/(a1 , . . . , ai1 ) is a non-zero divisor
(we set a0 = 0).
Lemma 16.5. Let M be a module overa commutative ring A. Assume that for any
maximal ideal m of A, the localization Mm = {0}. Then M = {0}.
Proof. Let x M . For any maximal ideal m A, there exists am 6 m such that
am x P
= 0. The ideal of A generated by the elements am is the unit ideal. Hence
1 = m bm am for some bm A and
X
bm am x = 0.
x=1x=
m
A[a1
0 ] = Aa0 be the subring of the quotient field Q(A) of A, I0 = (a1 /a0 , . . . , an /a0 )
A0 . Define a homomorphism 0 : A[Z1 , . . . , Zn ] A0 [I0 ] via sending each Zi to
ai /a0 . We claim that J0 = Ker(0 ) is equal to the ideal J00 generated by the polynomials Li = a0 Zi ai . Assume this is so. Then for any F (T0 , . . . , Tn ) Ker(),
after dehomogenizing with respect to T0 , we obtain that F (1, Z1 , . . . , Zn ) belongs
to J00 . This would immediately imply that T0N F J 0 for some N 0. Replacing T0 with Ti , and f0 with fi , we will similarly prove that TiN F J 0 for any
i = 0, . . . , n. Now consider the A-submodule M of A[T0 , . . . , Tn ]/J 0 generated by F .
Since TiN F = 0, i = 0, . . . , n, it is a finitely generated A-module. For any maximal
ideal m A let Pij = (ai mod m)Tj (aj mod m)Ti . The ideal in (A/m)[T0 , . . . , Tn ]
generated by the linear polynomials Pij is obviously prime. Thus TiN F = 0 implies
M A/m = {0}. Applying Nakayamas Lemma we infer that, for any maximal ideal
m A, the localization Mm is equal to zero. By the previous lemma this gives M = 0
so that F J 0 .
It remains to show that Ker(0 ) is generated by by the polynomials Li = a0 Zi ai .
We use induction on n. Assume n = 1. Let F Ker(0 ), i.e., 0 (F (Z1 )) =
F (a1 /a0 ) = 0. Dividing by L1 = a0 Z1 a1 , we obtain for some G(Z1 ) A[Z1 ] and
r0
ar0 F (Z1 ) = G(Z1 )(a0 Z1 a1 ) = a0 G(Z1 )Z1 a1 G(Z1 ).
n
X
i=2
n
X
i=2
X
(1)j aij ei1 . . . eij1 eij+1 . . . eir .
i
165
Now the claim is that the complex of A-modules (called the Koszul complex)
{0}
n
^
n1
^
A ...
2
^
1
^
An A A/(a1 , . . . , an ) {0}
is
The previous proposition asserts only that this complex is exact at the term
V1exact.
An .
Proposition 16.8. Let X be an affine irreducible algebraic k-set, I be an ideal in
O(X) generated by a regular sequence (f0 , . . . , fn ), and let Y = V (I) be the set of
zeroes of this ideal. Let : BY (X) X be the blow-up of X along Y . Then for any
xY,
1 (x)
= Pn (K).
The pre-image of every irreducible component of Y is an irreducible subset of BY (X)
of codimension 1.
Proof. By Proposition 16.6, Z = BY (X) is a closed subset of X Pn (K) defined by
the equations
T0 fi Ti f0 = 0, i = 1, . . . , n.
For any point y Y we have f0 (y) = . . . = fn (y) = 0. Hence for any t Pn (K), the
point (y, t) is a zero of the above equations. This shows that 1 (y) is equal to the
fibre of the projection X Pn (K) X over y which is obviously equal to Pn (K). For
each irreducible component Yi of Y the restriction map : 1 (Yi ) Yi has fibres
isomorphic to n-dimensional projective spaces. By Lemma 12.7 of Lecture 12 (plus
the remark made in the proof of Lemma 15.5 in Lecture 15) we find that 1 (Yi ) is
irreducible of dimension equal to n + dim Yi . By Krulls Hauptidealsatz, dim Yi =
dim X n 1 (here we use again that (f0 , . . . , fn ) is a regular sequence).
Lemma 16.9. Let X be a nonsingular irreducible affine algebraic k-set, Y be a nonsingular closed subset of X. For any x Y with dimx Y = dimx X n there exists
an affine open neighborhood U of x in X such that Y U = V (f1 , . . . , fn ) for some
regular sequence (f1 , . . . , fn ) of elements in O(U ).
Proof. Induction on n. The case n = 1 has been proven in Lecture 13. Let f0 I(Y )
such that its germ (f0 )x in mX,x does not belong to m2X,x . Let Y 0 = V (f0 ). By Lemma
14.7 from lecture 14, T (Y 0 )x is of codimension 1 in T (X)x . By Krulls Hauptidealsatz,
dimx Y 0 = dim X 1, hence Y 0 is nonsingular at x. Replacing X with a smaller open
affine set U , we may assume that Y 0 U is nonsingular everywhere. By induction, for
some V Y 0 , Y V is given in V by an ideal (f1 , . . . , fn ) so that Y is given locally
by the ideal (f0 , . . . , fn ). Now the assertion follows from the following statement from
167
It is easy to compute the
G1
T1 (y, z)
...
...
Gm
(y, z)
F0T1
1
z1 T1 (y) F
T1 (y)
...
...
0
n
z1 F
(y)
F
T1
T1 (y)
(y)
F
n (y)
TN
N
...
...
...
...
0
...
...
0
...
...
...
...
...
...
...
...
0
...
...
0
0
...
...
0
169
Similarly, the restriction 2 of to V is a regular map V A2 (K) given by the
homomorphism of rings:
2 : k[Z1 , Z2 ] O(U ) = k[Z1 , Z2 , t]/(Z2 t0 Z1 )
= k[Z2 , t0 ].
The pre-image of Y in V is the set of zeroes of the function
2 (Z22 Z12 (Z1 + 1)) = Z22 (1 t02 (Z2 t0 + 1)).
Thus
1 (Y ) U = E1 C1 , 1 (Y ) V = E2 C2 ,
where
E1 = V (Z1 ), C1 = V (t2 Z1 1) U
= A2 (K),
E2 = V (Z2 ), C2 = V (1 t02 (Z2 t0 + 1)) V
= A2 (K).
It is clear that
E1 = 1 (0) U
= A1 (K), E2 = 1 (0) V
= A1 (K),
i.e., 1 (0) = E1 E2
= P1 (K). Thus the proper inverse transform of Y is equal to
the union C = C1 C2 . By differentiating we find that both C1 and C2 are nonsingular
curves, hence C is nonsingular. Moreover,
C1 1 (0) = V (Z1 , t2 1) = {(0, 1), (0, 1)},
C2 1 (0) = V (Z2 , t02 1) = {(0, 1), (0, 1)}.
Note that since t = t01 at U V , we obtain C1 1 (0) = C2 1 (0). Hence
1 (0) C consists of two points. Moreover, it is easy to see that the curve C
intersects the exceptional divisor E = 1 (0) transversally at the two points. So the
picture is as follows:
The restriction : C Y is a resolution of singularities of Y .
Example 16.14. 4 This time we take Y = V (Z12 Z23 ). We leave to the reader
to repeat everything we have done in Example 1 to verify that the proper transform
Figure 16.1:
Example 16.15. Let Y = V (F (Z1 , . . . , Zn )) An (K), where F is a homogeneous
polynomial of degree d. We say that Y is a cone over Y = V (F (Z1 , . . . , Zn ) in
Pn1 (K). If identify An (K) with Pn (K)0 , and Y with the closed subset V (Z0 , F )
V (Z0 )
= Pn1 (K), we find that Y is the union of the lines joining the point (1, 0, . . . , 0)
with points in Y . Let : B = B{0} (An (K)) An (K) be the blowing up of the
origin in An (K). Then
B = i Ui , Ui = B An (K) Pn1 (K)i ,
and
1 (Y ) Ui = V (F (Z1 , . . . Zn )) V ({Zj tj Zi }j6=i )
= V (Zid G(t1 , . . . , tn1 )),
where tj = Tj /T0 , and G is obtained from F via dehomogenization with respect to
Ti . This easily implies that
1 (Y ) =
1 (Y ) 1 (0),
1 (Y ) 1 (0)
= Y .
Example 16.16. Let X = V (Z12 + Z23 + Z34 ) A3 (K) and let Y1 = B{0} (A3 (K)) be
the blow-up. The full inverse transform of X in Y1 is the union of three affine open
subsets each isomorphic to a closed subset of A3 (K):
V1 : Z12 (1 + U 3 Z1 + V 4 Z12 ) = 0,
V2 : Z22 (U 2 + Z2 + V 4 Z22 ) = 0,
V3 : Z32 (U 2 + V 3 Z3 + Z32 ) = 0.
171
The equations of the proper inverse transform X1 are obtained by dropping the first
factors. In each piece Vi the equations Zi = 0 define the intersection of the proper
inverse transform X1 of X with the exceptional divisor E1
= P2 (K). It is empty set in
V1 , the affine line U = 0 in V2 and V3 . The fibre of the map X1 X over the origin is
R1
= P1 (K). It is easy to see (by differentiation) that V1 and V2 are nonsingular but
V3 is singular at the point (U, V, Z3 ) = (0, 0, 0). Now let us start again. Replace X by
V3
= V (Z12 + Z23 Z3 + Z32 ) A3 (K) and blow-up the origin. Then glue the blow-up
with V1 and V2 along V3 (V1 V2 ). We obtain that the proper inverse transform X2
of X1 is covered by V1 , V2 as above and three more pieces
V4 : 1 + U 3 V Z12 + V 2 = 0
V5 : U 2 + Z22 V + V 2 = 0,
V6 : U 2 + V 3 Z32 + 1 = 0.
The fibre over the origin is the union of two curves R2 , R3 each isomorphic to P1 (K).
The equation of R2 R3 in V5 is U 2 + V 2 = 0. The equation of R2 R3 in V3
is U 2 + 1 = 0. Since R1 V3 was given by the equation Z3 = 0 and we used the
substitution Z3 = V Z2 in V5 , we see that the pre-image of R1 intersects R1 and R2
at their common point (U, V, Z2 ) = (0, 0, 0) in V5 . This point is the unique singular
point of X2 . Let us blow-up the origin in V5 . We obtain X3 which is covered by open
sets isomorphic to V1 , V2 , V4 , V6 and three more pieces:
V7 : 1 + V U 2 Z1 + V 2 = 0,
V8 : U 2 + V 2 Z3 + 1.
V9 : U 2 + V Z2 + V 2 = 0,
The pre-image of the origin in the proper inverse transform X3 of X2 consists of
1
two curves R4 , R5 each isomorphic
to P (K). In the open set V9 they are given
by the equations V = 0, U = 1V . The inverse image of the curve R1 intersects R4 , R5 at their intersection
point. The inverse images of R2 intersects R4 at
point (1, 1, 0). Finally we blow up the origin at V9 and obtain that the properinverse transform X4 is nonsingular. It is covered by open affine subsets isomorphic
to V1 , . . . , V8 and three more open sets
V10 : 1 + U V + V 2 = 0,
V11 : U 2 + V + V 2 = 0,
V12 : U 2 + V + 1 = 0.
Figure 16.2:
Exercises.
1. Prove that BV (I) (X) is not affine unless I is (locally ) a principal ideal.
2. Resolve the singularities of the curve xn + y r = 0, (n, r) = 1, by a sequence
of blow-ups in the ambient space. How many blow-ups do you need to resolve the
singularity?
3. Resolve the singularity of the affine surface X : Z12 + Z23 + Z33 = 0 by a sequence
of blow-ups in the ambient space. Describe the exceptional curve of the resolution
X.
f :X
4. Describe A(I), where A = k[Z1 , Z2 , ], I = (Z1 , Z22 ). Find the closed subset BI (A)
of A2 (K) P1 (K) defined by the kernel of the homomorphism : A[T0 , T1 ]
A(I), T0 Z1 , T2 Z22 . Is it nonsingular?
5*. Resolve the singularities of the affine surface X : Z12 + Z23 + Z35 = 0 by a sequence
of blow-ups in the ambient space. Show that one can find a resolution of singularities
X such that the graph of irreducible components of f 1 (0) is the Dynkin
f :X
diagram of the root system of a simple Lie algebra of type E8 .
6*. Resolve the singularities of the affine surface X : Z1 Z23 + Z13 + Z32 = 0 by a
sequence of blow-ups in the ambient space. Show that one can find a resolution of
X such that the graph of irreducible components of f 1 (0) is
singularities f : X
the Dynkin diagram of the root system of a simple Lie algebra of type E7 .
7*. Resolve the singularities of the affine surface X : Z1 (Z22 + Z1n ) + Z32 = 0 by a
sequence of blow-ups in the ambient space. Show that one can find a resolution of
X such that the graph of irreducible components of f 1 (0) is
singularities f : X
the Dynkin diagram of the root system of a simple Lie algebra of type Dn .
173
8*. Resolve the singularities of the affine surface X : Z1 Z22 + Z3n+1 = 0 by a sequence
of blow-ups in the ambient space. Show that one can find a resolution of singularities
X such that the graph of irreducible components of f 1 (0) is the Dynkin
f :X
diagram of the root system of a simple Lie algebra of type An .
9*. Let f : P2 (K) P2 (K) be the rational map given by the formula T0
T1 T2 , T1 T2 T3 , T2 T0 T1 . Show that there exist two birational regular maps
1 , 2 : X P2 (K) with f 1 = 2 such that the restriction of each i over
P2 (K)j , j = 0, 1, 2 is isomorphic to the blow-up along one point.
Lecture 17
Riemann-Roch Theorem
Let k be an arbitrary field and K be its algebraic closure. Let X be a projective variety
over k such that X(K) is a connected nonsingular curve.
A divisor on X is an element of the free abelian group ZX generated by the set
X(K) (i.e. a set of maps X(K) Z with finite support). We can view a divisor as
a formal sum
X
D=
n(x)x,
xX(K)
where x X, n(x) Z and n(x) = 0 for all x except finitely many. The group law
is of course defined coefficient-wisely. We denote the group of divisors by Div(X).
A divisor D is called effective if all its coefficients are non-negative. Let Div(X)+
be the semi-group of effective divisors. It defines a partial order on the group Div(X):
D D0 D D0 0.
Any divisor D can be written in a unique way as the difference of effective divisors
D = D+ D .
We define the degree of a divisor D =
deg(D) =
n(x)x by
n(x)[k(x) : k].
xX(K)
Recall that k(x) is the residue field of the local ring OX,x . If k = K, then k(x) = k.
The local ring OX,x is a regular local ring of dimension 1. Its maximal ideal is
generated by one element t. We call it a local parameter. For any nonzero a OX,x ,
let x (a) be the smallest r such that a mrX,x .
175
176
This proves (i),(ii). Note that we have the equality in (ii) when x (a) 6= x (b).
Let f R(X) be a nonzero rational function on X. For any open affine neighborhood U of a point x X, f can be represented by an element in Q(O(X)). Since
Q(O(X)) = Q(OX,x ), we can write f as a fraction a/b, where a, b OX,x . We set
x (f ) = x (a) x (b).
It follows from Lemma 17.1 (i), that this definition does not depend on the way we
write f as a fraction a/b.
Lemma 17.2. Let f, g R(X) \ {0}. The following properties hold:
(i) x (f g) = x (f ) + x (g);
(ii) x (f + g) min{x (f ), x (g)} if f + g 6= 0;
(iii) x (f ) 0 f OX,x ;
(iv) x (f ) 6= 0 only for finitely many points x X(K).
Proof. (i), (ii) follow immediately from Lemma 17.1. Assertion (iii) is immediate.
Let U be an open Zariski set such that f, f 1 O(U ). Then, for any x U ,
x (f ) = x (f 1 ) 0 implies that x (f ) = 0. Since X(K) \ U is a finite set, we
get (iv).
Now we can define the divisor of a rational function f by setting
X
div(f ) =
x (f )x.
xX(K)
177
Proposition 17.3. For any nonzero f, g R(X),
div(f g) = div(f ) + div(g).
In particular, the map f 7 div(f ) defines a homomorphism of groups
div : R(X) Div(X).
If D = div(f ), we write D+ = div(f )0 , D = div(f ) . We call div(f )0 the divisor
of zeroes of f and div(f ) the divisor of poles of f . We say that x (f ) is the order
of pole (or zero) if x div(f ) (or div(f )0 ).
We define the divisor class group of X by
Cl(X) = Div(X)/div(R(X) ).
Two divisors in the sameP
coset are called linearly equivalent. We write this D D0 .
For any divisor D = n(x)x let
L(D) = {f R(X) : div(f ) + D 0} = {f R(X) : x (f ) n(x), x X(K)}.
By definition, L(D) contains the zero function f (thinking that x (f ) = for all
x X). It follows from Lemma 17.2 that L(D) is a vector space over k. The
Riemann-Roch formula is a formula for the dimension of the vector space L(D).
Proposition 17.4.
(ii) L(D)
= L(D + div(f )) for any f R(X);
(iii) L(0) = k.
Proof. (i) Let D = D+ D . Then D+ = D + D and for any f L(D), we have
div(f ) + D 0 div(f ) + D + D = div(f ) + D+ 0.
This shows that f L(D+ ). Thus it suffices to show that L(D) is finite-dimensional
for an effective divisor D. Let t be a local parameter at x X(K), since x (f )+n(x)
0, x (tn(x) f ) 0 and hence x (tn(x) f ) OX,x . Consider the inclusion OX,x K[[T ]]
given by the Taylor expansion. Then we can write
f =T
n(x)
X
(
ai T i ),
i=0
where the equality is taken in the field of fractions K((T )) of K[[T ]]. We call the
right-hans side, the Laurent series of f at x. Consider the linear map
L(D) xX(K) T n(x) K[[T ]]/K[[T ]]
= xX(K) K n(x) ,
178
which assigns to f the collection of cosets of the Laurent series of f modulo k[[T ]].
The kernel of this homomorphism consists of functions f such that x (f ) 0 for all
x X(K), i.e., regular function on X. Since X(K) is a connected projective set, any
regular function on X is a constant. This shows that L(D)k K is a finite-dimensional
vector space over K. This easily implies that L(D) is a finite-dimensional vector space
over k.
(ii) Let g L(D + div(f )), then
div(g) + div(f ) + D = div(f g) + D 0.
This shows that the injective homomorphism of the additive groups R(D) R(D), g 7
f g, restricting to the space L(D + div(f )) defines an an injective linear map L(D +
div(f ))
= L(D). The inverse map is defined by the multiplication by f 1 .
(iii) Clearly L(0) = O(X) = k.
It follows from the previous Proposition that dimk L(D) depends only on the
divisor class of D. Thus the function dim : Div(X) Z, D 7 dimk L(D) factors
through a function on Cl(X) which we will continue to denote by dim.
Theorem 17.5. (Riemann-Roch). There exists a unique divisor class KX on X such
that for any divisor class D
dimk L(D) = deg(D) + dimk L(KX D) + 1 g,
where g = dimk L(KX ) (called the genus of X),
Before we start proving this theorem, let us deduce some immediate corollaries.
Taking D from KX , we obtain
deg(D) = 2g 2.
Taking D = div(f ), we get L(D)
= L(0) and L(KX D)
= L(KX ). Hence
dimk L(D) = 1 and dimk L(KX D) = g. This gives
deg(div(f )) = 0.
This implies that the degrees of linearly equivalent divisors are equal. In particular, we
can define the degree of a divisor class.
Also observe that, for any divisor D of negative degree we have L(D) = {0}. In
fact, if div(f ) + D 0 for some f R(X) , then deg(div(f ) + D) = deg(D) 0.
Thus if take a divisor D of degree > 2g 2, we obtain dim L(KX D) = 0. Thus
the Riemann-Roch Theorem implies the following
179
Corollary 17.6. . Assume deg(D) > 2g 2, then
dim L(D) = deg(D) + 1 g.
Example 17.7. Assume X = P1k . Let U = P1 (K)0 = A1 (K) = K. Take D =
x1 +. . .+xn , where xi k. Then L(D) consists of rational functions f = P (Z)/Q(Z),
where P (Z), Q(Z) are polynomials with coefficients in k and Q(T ) has zeroes among
the points xi s. This easily implies that L(D) consists of functions
P (T0 , T1 )/(T1 a0 T0 ) (T1 xi T0 ),
where degP (T0 , T1 ) = n. The dimension of L(D) is equal to n + 1. Taking n
sufficiently large, and applying the Corollary, we find that g = 0.
The fact that deg(div(f )) = 0 is used for the proof of the Riemann-Roch formula.
We begin with proving this result which we will need for the proof. Another proof of
the formula, using the sheaf theory, does not depend on this result.
Lemma 17.8. (Approximation lemma). Let x1 , . . . , xn X, 1 , . . . , n R(X), and
N be a positive integer. There exists a rational function f R(X) such that
x (f i ) > N,
i = 1, . . . , n.
i = 1, . . . , n.
180
i = 1, . . . , n.
i=1
(i)
(i)
xj ((i)
s ) >> 0, j 6= i, s = 1, . . . , ei .
P
Let us show that ri=1 ei functions obtained in this way are linearly independent over
f (R(Y )). Assume
ei
r X
X
ais (i)
s =0
i=1 s=1
f (R(Y
181
Dividing by by a1s , we get
(i)
is cis s
= 0, where
y (cis ) 0,
c1s = 1,
We have
e1
X
(1)
s =
s=1
es
r X
X
cis (i)
s .
i=2 s=1
By Lemma 17.10,
(1)
x1 (c1s (1)
s ) = x1 (c1s ) + x1 (s ) s mod e1 .
This easily implies that no subset of summands in the left-hand side L.H.S. add up to
zero. Therefore,
x1 (L.H.S) = min{x1 (c1s (1)
s } e1 .
s
On the other hand, x1 (R.H.S.) can be made arbitrary large. This contradiction
proves the assertion.
Let be the direct product of the fraction fields R(X)x of the local rings OX,x ,
where x X. By using the Taylor expansion we can embed each R(X)x in the
fraction field K((T )) of K[[T ]]. Thus we may view as the subring of the ring of
functions
K((T ))X = Maps(X, K((T )).
The elements of will be denoted by (x )x . We consider the subring AX of formed
by (x )x such that x OX,xPexcept for finitely many xs. Such elements are called
adeles. For each divisor D =
n(x)x, we define the vector space over the field k:
(D) = {(x )x : x (x ) n(x)}.
Clearly,
(D) R(X) = L(D),
(D) AX .
182
0
(T n(x) K[[T ]]/T n(x) K[[T ])
=
xX
K n(x) n(x) ,
xX
183
Proof. Let f : X Y , g : Y Z be two regular maps. Let z Z and
g 1 (z) = {y1 , . . . , yr },
i = 1, . . . , n,
s = 0, . . . , m m0
184
Then
X
mdeg(D) = m(
ei ) = Nm + dim L(mD) 1 Nm + (m m0 + 1)n 1.
P
Dividing by m and letting m go to infinity, we obtain
ei n = [R(X) : R(Y )].
Together with Corollary 17.9, this proves the assertion.
Corollary 17.14. For any rational function R(X),
deg(div(f )) = 0.
Proof. Let f : X P1 be the regular map defined by . Then, as we saw in the
previous proof, deg(div(f ) ) = [R(X) : k()]. Similarly, we have deg(div(1 ) ) =
[R(X) : k()]. Since div() = div(f )0 div(f ) , we are done.
Corollary 17.15. Assume deg(D) < 0. Then L(D) = {0}.
Set
r(D) = deg(D) dim L(D).
By Corollary 17.6, this number depends only on the linear equivalence class of D. Note
that, assuming the Riemann-Roch Theorem, we have r(D) = g 1dim L(K D)
g 1. This shows that the function D 7 r(D) is bounded on the set of divisors. Let
us prove it.
Lemma 17.16. . The function D 7 r(D) is bounded on the set Div(X).
Proof. As we have already observed, it suffices to prove the boundness of this function
on Cl(X). By Proposition 17.13 (iii), for any two divisors D0 , D with D0 D,
r(D0 ) r(D) = dim((D0 ) + R(X))/((D) + R(X)) 0.
Take a non-zero rational function R(X). Let D = div() , n = degD. As we
saw in the proof of Proposition 17.13,
mn r(mD) r(0) + m(m m0 n) 1 = r(mD) + mn m0 n.
This implies
r(mD) m0 n n, hence r(mD) is bounded as a function of n. Let
P
D0 =
n(xi )xi be a divisor, yi = f (xi ) P1 , where f : X P1 is the regular map
defined by . Let P (t) be a polynomial vanishing at the points yi which belong to
the affine part (P1 )0 . Replacing P (t) by some power, if needed, we have f (P (t)) =
P () R(X) and div(P ()) + mD D0 for sufficiently large m. This implies that
r(D0 ) r(mD + div(P ())) = r(mD)).
This proves the assertion.
185
Corollary 17.17. For any divisor D
dim A/((D) + R(X)) < .
Proof. We know that
r(D0 ) r(D) = dim((D0 ) + R(X)/(D) + R(X))
is bounded on the set of pairs (D, D0 ) with D0 D. Since every adele belongs to
some space (D), the falsity of our assertion implies that we can make the spaces
((D0 )+R(X)/(D)+R(X)) of arbitrary dimension. This contradicts the boundness
of r(D0 ) r(D).
Let
H(D) = A/((D) + R(X)).
We have r(D0 ) r(D) = dimk H(D) dimk H(D0 ) if D0 D. In particular, setting
g = dimk H(0),
we obtain
r(D) = g 1 dimk H(D),
or, equivalently
dimk L(D) = deg(D) + dimk H(D) g + 1.
(17.1)
186
Lemma 17.18. Let (X). There exists a maximal divisor D (with respect to
the natural order on Div(X)) such that H(D) .
Proof. If H(D1 ) H(D2 ) , then H(D3 ) , where D3 = sup(D1 , D2 ). This
shows that it suffices to verify that the degrees of D such that H(D) is bounded.
Let D0 be any divisor, L(D0 ). Since D + div() D D0 , we have
(D D0 ) (D + div()).
Let 1 , . . . , n be linearly independent elements from L(D0 ). Since vanishes on
(D), the functions 1 , . . . , n vanish on (D D0 ) (D + div(i )) and
linearly independent over K. Thus
dimk H(D D0 ) dimk L(D0 ).
Applying equality (1) from above, we find
dimk L(D D0 ) = deg(D) + deg(D0 ) 1 + g dimk L(D0 )
deg(D0 ) + 1 g + dimk H(D0 ).
Taking D0 with deg(D0 ) > deg(D) to get L(D D0 ) = {0}, we obtain
deg(D) 2g 2.
Proposition 17.19.
dimR(X) (X) = 1.
Proof. Let , 0 be two linearly independent differentials. For any linearly independent
(over K) sets of functions {a1 , . . . , an }, {b1 , . . . , bn } in R(X), the differentials
a1 , . . . , an , b1 0 , . . . , bn 0
(17.2)
are linearly independent over K. Let D be such that , 0 (D). It is easy to see
that such D always exists. For any divisor D0 , we have
(D D0 ) (D + div()), L(D0 ).
Thus the 2n differentials from equation (2), where (a1 , . . . , an ) and (b1 , . . . , bn ) are
two bases of L(D0 ), vanish on (D D0 ). Therefore,
dimk H(D D0 ) 2 dimk L(D0 ).
187
Again, as in the proof of the previous lemma, we find
dimk L(D D0 ) 2deg(D0 ) + 2 2g.
taking D0 with deg(D0 ) > deg(D) + 2 2g, we obtain
0 2deg(D0 ) + 2 2g > 0.
This contradiction proves the assertion.
For any (X) we define the divisor of as the maximal divisor D such that
H(D) . We denote it by div().
Corollary 17.20. Let , 0 (X). Then div() is linearly equivalent to div( 0 ).
Proof. We know that H(D) implies H(D + div()). Thus the divisor
of is equal to div() + div(). But each 0 (X) is equal to for some
R(X).
The linear equivalence class of the divisor of any differential is denoted by KX . It
is called the canonical class of X. Any divisor from KX is called a canonical divisor
on X.
Theorem 17.21. (Riemann-Roch). Let D be any divisor on X, and K any canonical
divisor. Then
dimk L(D) = deg(D) + dimk L(K D) + 1 g,
where g = dimk L(K).
Proof. Using formula (2), it suffices to show that
dimk H(D) = dimk L(K D),
or, equivalently, dimk H(K D) = dimk L(D). We will construct a natural isomorphism of vector spaces
c : L(D) H(K D) .
Let L(D), K = div(). Then
div() = div() + div() K D.
Thus vanishes on (K D), and therefore H(K D) . This defines a
linear map c : L(D) H(KD ) . Let H(K D) and K 0 = div(). Since K 0
188
189
the ring k[Z1 , Z2 ]/(f (Z1 , Z2 )), where f (Z1 , Z2 ) = 0 is the affine equation of X. We
may represent it by a polynomial P (Z1 , Z2 ). Now it is easy to compute the dimension
of the space of polynomials P (X1 , X2 ) modulo (f ) which belong to the linear space
L(nD). We can write
n
X
P (Z1 , Z2 ) =
Gi (Z1 , Z2 ),
i=1
190
of order 6 at x, the coefficient a3 must be also zero. Then 1 2 is the only function
with pole of order 5 at x. This implies that a6 = 0. Now 21 is the only function with
pole of order 4, so we must have a2 = 0. If a4 6= 0, then 2 is a linear combination of
1 and 1 , and hence belongs to L(2 x). This contradicts the choice 2 . So, we get
a0 + a1 1 = 0. This implies that a0 = a1 = 0.
Consider the map f : X P1 given by the function 1 . Since 2 satisfies an
equation of degree 2 with coefficients from the field f (R(P1 )), we see that [R(X) :
R(P1 )] = 2. Thus, adding 2 to f (R(P1 )) we get R(X). Let
: X \ {x} A2
be the regular map defined by (Z1 ) = 1 , (Z2 ) = 2 . Its image is the affine
curve defined by the equation
a0 + a1 Z1 + a2 Z12 + a3 Z13 + a4 Z2 + a5 Z22 + a6 Z1 Z2 = 0.
Since k(X) = k( (Z1 ), (Z2 )) we see that X is birationally isomorphic to the
affine curve V (F ). Note that a3 6= 0, since otherwise, after homogenizing, we get a
conic which is isomorphic to P1 . So, homogenizing F we get a plane cubic curve with
equation
F (T0 , T1 , T2 ) = a0 T03 +a1 T02 T1 +a2 T0 T12 +a3 T13 +a4 T02 T2 +a5 T0 T22 +a6 T0 T1 T2 = 0.
(17.3)
It must be nonsingular, since a singular cubic is obviously rational (consider the pencil of
lines through the singular point to get a rational parameterization). Since a birational
isomorphism of nonsingular projective curves extends to an isomorphism we get the
assertion.
Note that we can simplify the equation of the plane cubic as follows. First we
may assume that a6 = a3 = 1. Suppose that char(k) 6= 2. Replacing Z2 with
Z20 = Z2 + 21 (a6 Z1 + a4 Z0 ), we may assume that a4 = a5 = 0. If char(k) 6= 2, 3, then
replacing Z1 with Z1 + 13 a2 Z0 , we may assume that a2 = 0. Thus, the equation is
reduced to the form
F (T0 , T1 , T2 ) = T0 T22 + T13 + a1 T02 T1 + a0 T03 ,
or, after dehomogenizing,
Z22 + Z13 + a1 Z1 + a0 = 0.
It is called the Weierstrass equation. Since the curve is nonsingular, the cubic polynomial Z13 + a1 Z1 + a0 does not have multiple roots. This occurs if and only if its
discriminant
= 4a31 + 27a20 6= 0.
(17.4)
191
Problems
1. Show that a regular map of nonsingular projective curves is always finite.
2. Prove that for any nonsingular projective curve X of genus g there exists a regular
map f : X P1 of degree (= [R(X) : f (R(P1 ))]) equal to g = 1.
3. Show that any nonsingular projective curve X of genus 0 with X(k) = is
isomorphic to a nonsingular conic on P2k [Hint: Use that dim L(KX ) > 0 to find a
point x with deg(1 x) = 2].
4. Let X be a nonsingular plane cubic with X(k) 6= . Fix a point x0 X(k).
For any x, y X let x y be the unique simple pole of a non-constant function
L(x + y x0 ). show that x y defines a group law on X. Let x0 = (0, 0, 1),
where we assume that X is given by equation (3). Show that x0 is the inflection point
of X and the group law coincides with the group law on X considered in Lecture 6.
5. Prove that two elliptic curves given by Weierstrass equations Z22 +Z32 +a1 Z1 +a0 = 0
and Z22 + Z32 + b1 Z1 + b0 = 0 are isomorphic if and only if a31 /a20 = b31 /b20 .
6. Let X be a nonsingular curve in P1 P1 given by a bihomogeneous equation of
degree (d1 , d2 ). Prove that its genus is equal to
g = (d1 1)(d2 1).
P
7. Let D = ri=1 ni xi be a positive divisor on a nonsingular projective curve X. For
any x X \ {x1 , . . . , xr } denote, let lx L(D) be defined by evaluating L(D)
at the point x. Show that this defines a rational map from X to P(L(D) ). Let
D : X P(L(D) ) be its unique extension to a regular map of projective varieties.
Assume X = P1 and deg(D) = d. Show that D (P1 ) is isomorphic to the Veronese
curve d (P1 ) Pd .
Index
E6 -variety, 155
K-point, 2
l-adic topology, 142
adele, 181
principal, 181
affine, 72
affine n-space, 3
affine algebraic variety, 2
affine line, 16
algebraic group
multiplicative, 18
tangent space, 120
algebraic set, 7
k-rational, 28
projective, 44, 69
quasi-projective, 70
unirational, 28
Bezout Theorem, 47, 52, 104
birational isomorphism, 26
birational map, 26
biregular map, 19, 72
blowing up, 161
canonical class, 187
canonical divisor, 187
Cartan cubic, 155
codimension, 103
Cohen Structure Theorem, 142
constructible set, 104
coordinate algebra, 14, 19
projective, 44
covering set, 37
cubic hypersurface, 114
rationality, 28
cubic surface
lines on it, 112
defining ideal, 4
degree
of a curve, 104
of a divisor, 175
of a homogeneous polynomial, 39
derivation, 50, 117
diagonal, 78
differentaial, 123
differential, 126, 185
discriminant, 56
divisor, 175
effective, 175
linear equivalence, 177
of a rational function, 176
principal, 176
divisor class group, 177
dominant map, 25
double-six configuration, 115
dual numbers, 117
dual projective space, 110
Dynkin diagram, 155
eliminant, 45
embedding, 147
embedding dimension, 133
Euler identity, 120
exceptional divisor, 167
192
INDEX
Fermat hypersurface, 105
field of definition, 7
field of rational functions, 24
finite map, 86
flag variety, 114
flex point, 49
general linear group, 18
genus, 152, 178, 188
geometrically connected, 80
geometrically irreducible, 23
germ, 125
Grassmann variety, 107, 155
Hartshornes Conjecture, 154
height, 131
Hesse configuration, 55
Hessian polynomial, 52
Hilberts Basis Theorem, 22
Hilberts Nullstellensatz, 7
homogeneous
coordinates, 31
ideal, 41
polynomial, 39
system of equations, 40
homogenization
of a polynomial, 40
of an ideal, 40
hyperplane, 110
hypersurface, 21
indeterminacy point, 144
index of ramification, 180
inflection point, 49
inflection tangent line, 49
integral, 84
integral closure, 85
integral element, 9
irreducible component, 23
irrelevant ideal, 42
isomorphism, 72
193
of affine varieties, 15
Jacobian criterion, 128
Jacobian matrix, 167
Koszul complex, 165
Krull dimension, 93
of a ring, 94
L
uroth Problem, 28
Laurent series, 177
Lie algebra, 120
line, 31
linear normal, 154
linear projection, 64
local line, 37
local parameter, 175
local ring, 34, 124
regular, 133
localization, 32, 33
monoidal transformation, 161
morphism, 13
Noetherian ring, 6
nondegenerate subset, 154
nonsingular point, 49
normal ring, 92
normalization, 92
Pfaffian hypersurface, 155
Pl
ucker coordinates, 106
Pl
ucker embedding, 107
Pl
ucker equations, 110
plane projective curve, 47
conic, 47
cubic, 47
quartic, 47
quintic, 47
sextic, 47
product
in a category, 67
194
projective algebraic variety, 40, 66
projective automorphism, 63
projective closure, 40
projective module, 33
proper map, 82
quadric, 68
radical, 7, 10
rational function, 24
rational map, 25
rational normal curve, 68
rational point, 122
real projective plane, 32
regular function, 19, 71
regular map, 18, 19, 71
regular point, 134
regular sequence, 163
residue field, 35
resolution of singularities, 168
resultant, 45, 47
Riemann sphere, 32
Riemann-Roch formula, 177
Riemann-Roch Theorem, 178
saturation, 42
secant line, 151
secant variety, 151
Segre map, 66
Segre variety, 66, 155
Severi variety, 154, 155
simple point, 128
singular point, 128
singularities
formal isomorphism, 143
local isomorphism, 143
smooth point, 128
subvariety, 5, 40
system of algebraic equations, 1
system of parameters, 135
regular, 135
INDEX
tangent line, 149
tangent space, 118
embedded, 149
tangent vector, 118
topological space
connected, 22, 80
irreducible, 21
locally closed, 70
Noetherian, 22
quasi-compact, 74
reducible, 21
total ring of fractions, 33
tritangent plane, 115
trivializing family, 37
truncation, 139
Veronese morphism, 65
Veronese surface, 154, 155
Veronese variety, 65
Weierstrass equation, 190
Zariski differential, 137
Zariski tangent space, 126
Zariski topology, 11, 44, 70