Yuri Manin

Introduction into theory of

schemes
Translated from the Russian and edited
by Dimitry Leites

Abdus Salam School

of Mathematical Sciences
Lahore, Pakistan
 .

Summary. This book provides one with a concise but extremely lucid exposition
of basics of algebraic geometry seasoned by illuminating examples.
The preprints of this book proved useful to students majoring in mathematics
and modern mathematical physics, as well as professionals in these fields.

c Yu. Manin, 2009

°
c D. Leites (translation from the Russian, editing), 2009
°
Contents

Editor’s preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Author’s preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1 Affine schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1 Equations and rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Geometric language: Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 Geometric language, continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4 The Zariski topology on Spec A . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 The affine schemes (a preliminary definition) . . . . . . . . . . . . . . . . 29
1.6 Topological properties of morphisms and Spm . . . . . . . . . . . . . . . 33
1.7 The closed subschemes and the primary decomposition . . . . . . . 42
1.8 Hilbert’s Nullstellensatz (Theorem on zeroes) . . . . . . . . . . . . . . . 49
1.9 The fiber products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1.10 The vector bundles and projective modules . . . . . . . . . . . . . . . . . 55
1.11 The normal bundle and regular embeddings . . . . . . . . . . . . . . . . . 63
1.12 The differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
1.13 Digression: Serre’s problem and Seshadri’s theorem . . . . . . . . . . 71
1.14 Digression: ζ-function of a ring . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
1.15 The affine group schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
1.16 Appendix: The language of categories. Representing functors . . 87
1.17 Solutions of selected problems of Chapter 1 . . . . . . . . . . . . . . . . . 96

2 Sheaves, schemes, and projective spaces . . . . . . . . . . . . . . . . . . . 97

2.1 Basics on sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
2.2 The structure sheaf on Spec A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
2.3 The ringed spaces. Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
2.4 The projective spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
2.5 Algebraic invariants of graded rings . . . . . . . . . . . . . . . . . . . . . . . . 116
2.6 The presheaves and sheaves of modules . . . . . . . . . . . . . . . . . . . . . 125
2.7 The invertible sheaves and the Picard group . . . . . . . . . . . . . . . . 131
4 Contents

2.8 The Čech cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

2.9 Cohomology of the projective space . . . . . . . . . . . . . . . . . . . . . . . . 145
2.10 Serre’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
2.11 Sheaves on Proj R and graded modules . . . . . . . . . . . . . . . . . . . . . 153
2.12 Applications to the theory of Hilbert polynomials . . . . . . . . . . . . 156
2.13 The Grothendieck group: First notions . . . . . . . . . . . . . . . . . . . . . 161
2.14 Resolutions and smoothness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Prefaces 5

Editor’s preface
It is for more than 40 years now that I wanted to make these lectures — my first
love — available to the reader: The preprints of these lectures [Ma1, Ma2] of cir-
culation of mere 500+200 copies became bibliographic rarities almost immediately.
Meanwhile the elements of algebraic geometry became everyday language of work-
ing theoretical physicists and the need in a concise manual only increased. Various
(nice) text-books are usually too thick for anybody who does not want to become a
professional algebraic geometer, which makes Manin’s lectures even more appealing.
The methods described in these lectures became working tools of theoretical
physicists whose subject ranges from high energy physics to solid body physics (see
[Del] and [Ef], respectively) — in all questions where supersymmetry naturally arises.
In mathematics, supersymmetry in in-build in everything related to homotopy and
exterior products of differential forms, hence to (co)homology. In certain places,
supersymmetric point of view is inevitable, as in the study of integrability of certain
equations of mathematical physics ([MaG]). Therefore, preparing proceedings of
my Seminar on Supermanifolds (see [SoS]) I urgently needed a concise and clear
introduction into basics of algebraic geometry.
In 1986 Manin wrote me a letter allowing me to include a draft of this trans-
lation as a Chapter in [SoS], and it was preprinted in Reports of Department of
Mathematics of Stockholm University. My 1972 definition of superschemes and su-
permanifolds [L0] was based on these lectures; they are the briefest and clearest
source of the background needed for studying those aspects of supersymmetries
that can not be reduced to linear algebra.
Written at the same time as Macdonald’s lectures [M] and Mumford’s lectures
[M1, M2, M3] Manin’s lectures are more lucid and easier to come to grasps. Later
on there appeared several books illustrating the topic from different positions, e.g.,
[AM], [Sh0], [E1, E2, E3], [H, Kz] and I particularly recommend [Reid] for the first
reading complementary to this book.
For preliminaries on algebra, see [vdW, Lang]; on sheaves, see [God, KaS]; on
topology, see [K, FFG, RF, Bb3]; on number theory, see [Sh1, Sh2].
In this book, N := {1, 2, . . . } and Z+ := {0, 1, 2, . . . }; Fm := Z/m; A× is the
group of invertible elements of the ring A. The term “identity” is only applied to
relations and maps, the element 1 (sometimes denoted by e) such that 1a = a = a1
is called unit (sometimes unity). Other notation is standard.
I advise the reader to digress from the main text and skim the section on cate-
gories as soon as the word “functor” or “category” appears for the first time.

The responsibility for footnotes is mine.

D. Leites, April 2, 2009.

6 Prefaces

Author’s preface
In 1966–68, at the Department of Mechanics and Mathematics of Moscow State
University, I read a two-year-long course in algebraic geometry. The transcripts of
the lectures of the first year were preprinted [Ma1, Ma2] (they constitute this book),
that of the second year was published in the Russian Mathematical Surveys [Ma3].
These publications bear the remnants of the lecture style with its pros and cons.
Our goal is to teach the reader to practice the geometric language of commutative
algebra. The necessity to separately present algebraic material and later “apply” it
to algebraic geometry constantly discouraged geometers, see a moving account by
O. Zariski and P. Samuel in their preface to [ZS].
The appearance of Grothendieck’s scheme theory opened a lucky possibility not
to draw any line whatever between “geometry” and “algebra”: They appear now as
complimentary aspects of a whole, like varieties and the spaces of functions on these
spaces in other geometric theories. From this point of view,
the commutative algebra coincides with (more pre-
c i s e l y, i s f u n c t o r i a l l y d u a l t o ) t h e t h e o r y o f l o c a l
g e o m e t r i c o b j e c t s — a f f i n e s c h e m e s.
This book is devoted to deciphering the above claim. I tried to consecutively
explain what type of geometric images should be related with, say, primary de-
composition, modules, and nilpotents. In A. Weil’s words, the spacial intuition is
“i n v a l u a b l e i f i t s r e s t r i c t e d n e s s i s t a k e n i n t o a c c o u n t”. I
strived to take into account both terms of this neat formulation.
Certainly, geometric accent considerably influenced the choice of material. In
particular, this chapter should prepare the ground for introducing global objects.
Therefore, the section on vector bundles gives, on “naive” level, constructions be-
longing, essentially, to the sheaf theory.
Finally, I wanted to introduce the categoric notions as soon as possible; they are
not so important in local questions but play ever increasing role in what follows. I
advise the reader to skim through the section “Language of categories” and return
to it as needed. 1)
I am absolutely incapable to edit my old texts; if I start doing it, an irresistible
desire to throw everything away and rewrite completely grips me. But to do some-
thing new is more interesting. Therefore I wish to heartily thank D. Leites who saved
my time.
The following list of sources is by no means exhaustive. It may help the reader to
come to grasps with the working aspects of the theory: [Sh0], [Bb1] (general courses);
[M1], [M2], [Ma3], [MaG], [S1], [S2] (more special questions).
The approach of this book can be extended, to an extent, to non-commutative
geometry [Kas]. My approach was gradually developing in the direction along which
the same guiding principle — construct a matrix with commuting elements satisfying
only the absolutely necessary commutation relations — turned out applicable in ever
wider context of n o n - c o m m u t a t i v e geometries, see [MaT], see also [GK],
[BM] (extension to operads and further).

Yu. Manin, March 22, 2009.

1
See also [McL, GM].
Chapter 1

Affine schemes

1.1. Equations and rings

Study of algebraic equations is an ancient mathematical science. In novel
times, vogue and convenience dictate us to turn to rings.
1.1.1. Systems of equations. Let I, J be some sets of indices, let
T = {Tj }j∈J be indeterminates and F = {Fi ∈ K[T ]}i∈I a set of polyno-
mials.
A system X of equations for unknowns T is the triple (the ring K, the
unknowns T , the functions F ), more exactly and conventionally expressed as

Fi (T ) = 0, where i ∈ I. (1.1)

Why the ground ring or the ring of constants K enters the definition is clear:
The coefficients of the Fi belong to a fixed ring K. We say that the system X
is defined over K.
What should we take for a solution of (1.1)?
To say, “a solution of (1.1) is a set t = (tj )j∈J of elements of K such that
Fi (t) = 0 for all i” is too restrictive: We wish to consider, say, complex roots
of equations with real coefficients. The radical resolution of this predicament
is to consider solutions in any ring, and, “for simplicity”, in all of the rings
simultaneously.
To consider solutions of X belonging to a ring L, we should be able to
substitute the elements of L into Fi , the polynomials with coefficients from K,
i.e., we should be able to multiply the elements from L by the elements from
K and add the results, that is why L must be a K-algebra.
Recall that the set L is said to be a K-algebra if L is endowed with the
structures of a ring and a K-module, interrelated by the following properties:
1. The multiplication K × L −→ L is right and left distributive with respect
to addition;
2. k(l1 l2 ) = (kl1 )l2 for any k ∈ K and l1 , l2 ∈ L.
8 Ch. 1. Affine schemes

A map f : L1 −→ L2 is a a K-algebra homomorphism if f is simultaneously

a map of rings and K-modules.
Let 1L denote the unity of L. To define a K-algebra structure on L, we need
a ring homomorphism (embedding of K) i : K −→ L which sends k · 1K ∈ K
into i(k)1L .
1.1.1a. Examples. 1) Every ring L is a Z-algebra with respect to the ho-
momorphism n 7→ n · 1L for any n ∈ Z.
2) If K = Fp or Q and L = Z or Fp2 , then there are no ring homomorphisms
K −→ L.
1.1.1b. Exercise. Prove 2).
1.1.2. Solutions of systems of equations. A solution of a system (1.1)
with values in a K-algebra L is a set t = (tj )j∈J of elements of L such that
Fi (t) = 0 for all i ∈ I. The set of all such solutions is denoted by X(L) and
each solution is also called an L-point of the system of equations X.
As is clear from Example 1.1.1a 1), for any system of equations with integer
coefficients, we may consider its solution in any commutative ring.
Two systems X and Y for the same unknowns given over a ring K are
said to be equivalent, more precisely equivalent over K (and we write X ∼ Y )
if X(L) = Y (L) for any K-algebra L.
Among the system of equations equivalent to a given one, X, there exists
the “biggest” one: For the left hand sides of this “biggest” system take the
left hand sides of (1.1) that generate the ideal (F ) in K[T ]j∈J .
The coordinate ring of the variety X = V (F ) is

K(X) = K[T ]/(F ). (1.2)

1.1.2a. Proposition. 1) The system whose left hand sides are elements of
the ideal (F ) is the largest system equivalent to (F ).
2) There is a one-to-one correspondence X(L) ←→ HomK (K(X), L),
where HomK denotes the set of K-algebra homomorphisms.
Proof. 1) Let P be the ideal in K[T ], where T = (T )j∈J , generated by the
left sides of the system of equations (1.1). It is easy to see that the system
obtained by equating all elements of P to zero is equivalent to our system, call
it (X). At the same time, the larger system is the maximal one since if we add
to it any equation F = 0 not contained in it, we get a system not equivalent
to (X). Indeed, take L = K[T ]/P . In L, the element t, where tj ∼= Tj0 mod P ,
is a solution of the initial system (X) whereas F (t) 6= 0 since F 6∈ P .
2) Let t = (tj )j∈J ∈ X(L). There exists a K-algebra homomorphism
K[T ] −→ L which coincides on K with the structure homomorphism K −→ L
and sends Tj to tj . By definition of X(L), the ideal P lies in the ker-
nel of this homomorphism, so we can consider a through homomorphism
A = K[T ]/P −→ L.
Conversely, any K-algebra homomorphism A −→ L uniquely determines
a through homomorphism K[T ] −→ A −→ L. Let tj be the image of Tj under
1.1 Equations and rings 9

this through homomorphism; then t = (tj )j∈J ∈ X(L) since all elements of P
vanish under this homomorphism.
It is easy to check that the constructed maps X(L) ←→ HomK (K(X), L)
are inverse to each other. u
t
For a non-zero K-algebra L, a system X over a ring K is said to be
consistent over L if X(L) 6= ∅ and inconsistent otherwise. Proposition 1.1.2a
shows that X is inconsistent only if its coordinate ring, or rather, algebra,
K(X), is zero, in other words, if 1 ∈ (X). 1)
1.1.3. Examples from arithmetics.
1.1.3a. The language of congruences. Let n be an integer of the form
4m + 3. The classical proof of the fact that n is not representable as a sum
of two perfect squares is as follows: If it were, the congruence
T12 + T22 ≡ 3 (mod 4) (1.3)
would have been solvable, whereas a simple case-by-case checking (set
T1 = 4a + r1 , T2 = 4b + r2 and consider the eight distinct values of (r1 , r2 ))
establishes that this is not the case.
From our point of view this argument reads as follows: Consider the system
T12 + T22 = n, where K = Z. (1.4)
The reduction modulo 4, i.e., the map Z −→ Z/4, determines a map
X(Z) −→ X(Z/4) and, if X(Z) 6= ∅, then X(Z/4) 6= ∅, which is not the
case.
More generally, in order to study X(Z) for any system X with integer
coefficients, we can consider sets X(Z/m) for any m and try to deduce from
this consideration some information on X(Z).
Usually,
if X(L) = ∅ for a nontrivial (i.e., 1 6= 0) field L, then X(Z) = ∅. (1.5)
In practice one usually tests L = R and the fields L = Fm (:= Z/m) for all
prime m’s.
A number of deepest results of the theory of Diophantine equations are
related with the problem: When is the converse statement true? A prototype
of these results is
Legendre’s theorem ([BSh]). Let
a1 T12 + a2 T22 + a3 T32 = 0, K = Z. (1.6)
If X(Z) = {(0, 0, 0)}, then X(L) = {(0, 0, 0)} for at least one of the rings
L = R or L = Z/mZ, where m > 1.
1
In school, and even in university, one often omits “over L” in the definition
of consistency thus declaring systems inconsistent only partly (over some classes
of rings) as totally inconsistent.
10 Ch. 1. Affine schemes

1.1.3b. Equations in prime characteristic. Consider the equation

0 · T + 2 = 0, K = Z. (1.7)

Clearly, X(L) = ∅ if 2 · 1L 6= 0 and X(L) = L if 2 · 1L = 0.

This example seems manifestly artificial; still, we often encounter the like
of it in “arithmetical geometry”.
1.1.3c. On usefulness of complexification. When studying X(R) ⊂ Rn
(for the case where K = R, and the number of unknowns is equal to n),
it is expedient to pass to the complexification of X(R), i.e., to X(C). Since
C is algebraically closed, it is usually easier to study X(C) than X(R); this
often constitutes the first stage of the investigation even if we are primarily
interested in purely real problems. The following example is illuminating.

Fig. 1 Fig. 2

Harnak’s theorem. Let F (T0 , T1 , T2 ) be a form (i.e., a polynomial homoge-

neous with respect to the degree) of degree d with real coefficients. Let X(R)
be the curve in RP 2 , the real projective plane, singled out by the equation
F = 0. Then the number of connected components of X(R) does not exceed
1
(d − 1)(d − 2) + 1.
2
The method of the proof uses the embedding X(R) −→ X(C). For sim-
plicity, let us confine ourselves to the case of a nonsingular X(C), i.e., to the
case where X(C) is a compact orientable 2-dimensional manifold. Its genus —
1
the number of handles — is, then, equal to (d − 1)(d − 2). On Fig. 1, d = 3
2
and X(C) is a torus. The proof is based on the two statements:
First, the complex conjugation can be extended to a continuous action on
X(C) so that X(R) is precisely the set of its fixed points.
Second, X(C) being cut along X(R) splits into precisely 2 pieces, cf. Fig. 2
(as, for d = 1, the Riemannian sphere — the compactified real plane — does
1.1 Equations and rings 11

being cut along the real axis). Now, routine topological considerations give
Harnak’s estimate, see, e.g., [Ch].
1.1.3d. The algebra of mathematical logic in geometric terms. 2) A
Boolean ring is any ring R (with 1) such that P 2 = P for any P ∈ R. Clearly,

P + Q = (P + Q)2 = P 2 + P Q + QP + Q2 = P + P Q + QP + Q (1.8)

implies P Q + QP = 0. Since R is commutative by definition, 2P Q = 0;

moreover,
2P = P + P = P 2 + P 2 = 0 (1.9)
implies P = −P . Therefore every Boolean ring is a commutative ring over F2 .
It is not difficult to show that every prime ideal of a Boolean algebra
R is a maximal one, and therefore every element P ∈ R can be viewed as
an F2 -valued function on Spec R.
Given two statements, P and Q, each either true of false, define their sum
and product by setting

P + Q = (P ∨ Q) ∧ (P̄ ∨ Q̄), P Q = P ∧ Q, (1.10)

where the bar stands for the negation, ∧ for the conjunction and ∨ for the
disjunction. With respect to the above operations the empty statement ∅ is
the zero, and ¯∅ is the unit. Clearly, P 2 = P and 2P = P + P = 0 for any P .
1.1.4. Summary. We have established the equivalence of the two languages:
That of the systems of equations (which is used in the concrete calculations)
and that of the rings and their morphisms. More exactly, we have established
the following equivalences:
   
 A system of equations X   A K-algebra K(X) with a 
over a ring K ⇐⇒ system of generators
   
for unknowns {Tj | j ∈ J}. {tj | j ∈ J}.
 
 A solution of the system  ½ ¾
A K-algebra homomorphism
of equations X ⇐⇒
  K(X) −→ L.
in a K-algebra L.
Finally, notice that using the language of rings we have no need to consider
a fixed system of generators t = (tj )j∈J of K(X) = K[T ]/(X); we should
rather identify the systems of equations obtained from each other by any
invertible change of unknowns. Every generator of K(X) plays the role of one
of “unknowns”, and the value this unknown takes at a given solution of the
system coincides with its image in L under the corresponding homomorphism.
1.1.5. Exercises. 1) The equation 2T − 4 = 0 is equivalent to the equation
T − 2 = 0, if and only if 2 is invertible in K.
2
For mathematical logic from an algebraist’s point of view, see [Ma4].
12 Ch. 1. Affine schemes

2) The equation (T − 1)2 = 0 is not equivalent to the equation T − 1 = 0.

3) Let the system of equations {Fi (X) = 0 | i ∈ I}, where X = (Xj )j∈J ,
be incompatible. Then it has a finite incompatible subsystem.
4) Let T1 , . . . , Tn be indeterminates; si (T ) be the i-th elementary sym-
metric polynomial in them. Determine the rings of constants over which the
following systems of equations are equivalent:

X1 : si (T ) = 0, i = 1, . . . , k ≤ n,
P
X2 : Tji = 0, i = 1, . . . , k ≤ n.
1≤j≤n

Hint. Use Newton’s formulas.

5) Any system of equations over a ring K in a finite number of unknowns
is equivalent to a finite system of equations if and only if the ring is Noethe-
rian. 3)
6) Let X be a system of equations over K and A the ring correspond-
ing to X. The maps L 7→ X(L) and L 7→ HomK (A, L) determine covariant
functors on the category AlgsK of K-algebras with values in Sets. Verify that
Proposition 1.1.2a determines an isomorphism of these functors.

1.2. Geometric language: Points

Let, as earlier, K be the main ring, X a system of equations over K for
unknowns T1 , . . . , Tn .
For any K-algebra L, we realize the set X(L) as a “graph” in Ln , the coor-
dinate space over L. The points of this graph are solutions of the system (1.1).
Taking into account the results of the preceding section let us introduce the
following definition.
1.2.1. The points of a K-algebra A. The points of a K-algebra A with
values in a K-algebra B (or just B-points of A) are the K-homomorphisms
A −→ B. Any B-point of A is called geometric if B is a field.
Example. (This example shows where the idea to apply the term “point” to
a homomorphism comes from.) Let K be a field, V an n-dimensional vector
space over K. Let us show that the set of point of V is in on-to-one correspon-
dence with the set of maximal ideals of the ring K[x], where x = (x1 , . . . , xn ),
of polynomial functions on V .
Every point (vector) of V is a linear functional on V ∗ , the space of linear
functionals on V with values in K and, as is easy to show ([Lang]), this linear
functional can be extended to an algebra homomorphism:
3
A ring is said to be Noetherian if it satisfies any of the following equivalent
conditions ([Lang]):
1. Any set of its generators contains a finite subset.
2. Any ascending chain of its (left for non-commutative rings) ideals stabilizes.
1.2 Geometric language: Points 13

.
S (V ∗ ) = K[x1 , . . . , xn ] −→ K, (1.11)

where V ∗ = Span(x1 , . . . , xn ). This homomorphism can be canonically iden-

tified with a point (vector) of V .
The next step is to separate the properties of a fixed algebra A from whims
of the variable algebra B, namely, instead of homomorphisms h : A −→ B we
consider their kernels, the ideals in A.
1.2.2. The spectrum. The kernel of a homomorphism A −→ B corre-
sponding to a geometric point is, clearly, a prime ideal (to be defined shortly).
There are many reasons why one should confine oneself to prime ideals instead
of seemingly more natural maximal ones and in the following sections we will
give these reasons.
An ideal p of a commutative ring A is said to be prime if A/p is an integral
domain, i.e., has no zerodivisors (and if we do not forbid 1 = 0, then the zero
ring may not be an integral domain). Equivalently, p is prime if p 6= A and

a ∈ A, b ∈ A, a b ∈ p =⇒ either a ∈ p or b ∈ p. (1.12)

The set of all the prime ideals of A is said to be the (prime) spectrum of A
and is denoted by Spec A. The elements of Spec A are called its points.
In what follows, we enrich the set Spec A with additional structures mak-
ing it into a topological space rigged with a sheaf of rings: This will lead to the
definition of an affine scheme. Schemes, i.e., topological spaces with sheaves,
locally isomorphic to affine schemes, are the main characters of algebraic ge-
ometry.
Starting the study of spectra we have to verify, first of all, that there exists
indeed what to study.
1.2.3. Theorem. If A 6= 0, then Spec A 6= ∅.
In the proof of this theorem we need the following:
1.2.4. Lemma (Zorn’s lemma). In a partially ordered set M , let every linearly
ordered subset N ⊂ M contain a maximal element. Then M contains a maximal
element.
For proofs of Zorn’s lemma, see, e.g., [K] or [Hs], where it is proved together with
its equivalence to the choice axiom, the complete order principle, and several other
statements. For an interesting new additions to the list of equivalent statements,
see, e.g., [Bla]. u
t
An ordered set satisfying the condition of Zorn’s lemma is called an inductive
set.
Proof of Theorem 1.2.3. Denote by M the set of all the ideals of A different
from A. Since M contains (0), it follows that M 6= ∅. The set M is partially ordered
with respect
S
to inclusion. In M , take an arbitrary linearly ordered set {Pα }α∈Λ .
Then Pα is also an ideal of A (mind the linear order) and this ideal is different
α S
from A (since the unit element does not belong to Pα ). Therefore M is an inductive
α
14 Ch. 1. Affine schemes

set. Denote by p its maximal element; it is a maximal ideal, and therefore a prime
one: In A/p, every non-zero ideal, in particular, all the principal ones, coincide with
A/p. Therefore every non-zero element of A/p is invertible and A/p is a field. u
t

Corollary. Every prime ideal is contained in a maximal ideal.

This theorem implies, in particular, that the spectrum of every non-zero
ring A possesses a geometric point (e.g., any homomorphism A −→ A/p,
where p ⊂ A is a maximal ideal, is one of them).
1.2.5. The center of a geometric point. The center of a geometric point
A −→ L is its kernel considered as an element of Spec A. Let k(x) be the field
of quotients 4) of the ring A/px , where px ⊂ A is the ideal corresponding to
x ∈ Spec A.
Proposition. The geometric L-points of a K-algebra A with center in
x ∈ Spec A are in one-to-one correspondence with the K-homomorphism
k(x) −→ L.
Proof. Indeed, since L is a field, any homomorphism A −→ L factorizes as

A −→ A/px −→ k(x) −→ L. (1.13)

The first two arrows of this sequence are rigidly fixed. u

t
5)
1.2.6. Examples. a) Let K be a perfect field, L an algebraically closed
field containing K, let A be a K-algebra, x ∈ Spec A, and px the corresponding
ideal.
If deg x := [k(x) : K] < ∞, then, by Galois theory, there are exactly deg x
geometric L-points with center in x.
If the field k(x) is not algebraic over K, and L contains sufficiently many
transcendences over K, then there may be infinitely many geometric L-points
with center in x.
Here is a very particular case.
b) The set of geometric C-points of R[T ] is the complex line C. The set
Spec R[T ] is the union of the zero ideal (0) and the set of all monic polynomials
irreducible over R.
Every such polynomial of degree 2 has two complex conjugate roots cor-
responding to two distinct geometric points, see Fig. 3.
In general, for any perfect field K, the geometric points of the K-algebra
K[T ] with values in the algebraic closure K are just the elements of K, and
their centers are irreducible polynomials over K, i.e., the sets consisting of all
the elements of K conjugate over K to one of them. u
t
4
The field of fractions or field of quotients of a ring is the smallest field in which
the ring can be embedded. It will be explicitly constructed in what follows.
5
Recall that a field K of characteristic p is said to be perfect if K p := {xp | x ∈ K}
coincides with K. (The symbol K p is also used to designate K × · · · × K .)
| {z }
p times
1.2 Geometric language: Points 15

Spec R [T] \ {0}

C

Fig. 3

1.2.7. A duality: Unknowns ←→ Coefficients. By considering Spec A

we may forget, if needed, that A is a K-algebra; every ideal of A is stable
under multiplication by the elements of K. When we are interested in geo-
metric points (or, more generally, in arbitrary L-points) the reference to K
is essential, since we have to consider K-homomorphisms A −→ L. Arbitrary
ring homomorphisms are, clearly, Z-homomorphisms; therefore this “abso-
lute” case may be considered as a specialization of a “relative” one, the one
over K.
For the systems of equations, the passage to the absolute case means that
we forget about the difference between “unknowns” and “coefficients” and
may vary the values of both. More precisely, consider a system of equations
in which the jth equation is
X (j)
ak xk = 0, where xk runs over monomials in our indeterminates.
k

(j)
Generally, ak are fixed elements of a ring of constants K whereas “pas-
sage to the absolute case” means that we now w r i t e a g e n e r a t -
(j)
i n g s y s t e m o f a l l r e l a t i o n s b e t w e e n t h e ak O V E R Z ,
and add it to our initial system of equations, bind-
(j)
i n g t h e xi a n d t h e ak t o g e t h e r . A n d a f t e r t h a t w e
may specialize coefficients as well, retaining only
the relations between them.
1.2.8. Exercise. A weak form of Hilbert’s Nullstellensatz (theorem
on zeros). Consider a system of equations {Fi (T ) = 0}, where T = (Tj )j∈J ,
over the ring K. Then either this system has a solution with values in a field,
or there exist polynomials Gi ∈ K[T ] (finitely many of them are 6= 0) such
that X
Gi Fi = 1.
i

Hint. Apply Theorem 1.2.3 to the ring corresponding to the system.

16 Ch. 1. Affine schemes

1.3. Geometric language, cont.: Functions on spectra

1.3.1. Functions on spectra. Let X be a system of equations over K in
unknowns T = (Tj )j∈J . Every solution of X in a K-algebra L, i.e., an element
of X(L), evaluates the Tj ; let {tj ∈ L}j∈J be the values. Therefore it is natural
to consider Tj as a function on X(L) with values in L. Clearly, this function
only depends on the class of Ti modulo the ideal generated by the left-hand
sides of X. This class is an element of the K-algebra K(X) := K[T ]/(X)
related with X, and generally all the elements of K(X) are functions on
X(L) = HomK (K(X), L): Indeed, for every ϕ : K(X) −→ L and f ∈ K(X),
“the value of f at ϕ” is by definition ϕ(f ).
The classical notation of functions is not well adjusted to reflect the funda-
mental duality that manifests itself in modern mathematics more and more:

“A space ←→ the ring of functions on the space” (1.14)

or, symmetrically, with a different emphasis:

“A ring ←→ the spectrum of its ideals of certain type”. (1.15)

When applied to Spec A, this duality leads to consideration of any element

f ∈ A as a function on Spec A. Let x ∈ Spec A and px the corresponding ideal.
Then, by definition, f (x) = f (mod px ) and we assume that f (x) belongs to
the field of quotients k(x) of the ring A/px .
1.3.1a. Convention. In what follows speaking about functions on Spec A
we will always mean the elements of A.
Thus, to every point x ∈ Spec A, its own field k(x) is assigned and the
values of the functions on Spec A belong to these fields.

Q
Z/5Z
4

3
Z/3Z
2
Z/2Z
1
0
(0) (2) (3) (5) Spec Z

Fig. 4

I tried to plot the first five integers (0, 1, 2, 3, 4) considered as functions

on Spec Z. The picture is not too convincing; besides, for various reasons that
1.3 Geometric language, continued 17

lie beyond the scope of these lectures, the “line” over the field Z/p — the
“vertical axis” over the point (p) — should be drawn “coiled into a ring”, i.e.,
the points of this “line” should form vertices of a regular p-gon. This does not
simplify the task of an artist.
To distinct elements of A the same functions on the spectrum might T corre-
spond; their difference represents the zero function, i.e., belongs to px .
x∈Spec A
Clearly, all the nilpotents are contained in this intersection. Let us prove the
opposite inclusion. For this, we need a new notion.
The set N of all nilpotent elements of a ring R is an ideal (as is easy to
see), it is called the nilradical of R.
1.3.2. Theorem. A function that vanishes at all the points of Spec A is
represented by a nilpotent element of A. In other words, the nilradical is the
intersection \
p. (1.16)
p∈Spec A

Proof. It suffices to establish that, for every non-nilpotent element, there

exists a prime ideal which does not contain it.
Let h ∈ A and hn 6= 0 for any positive integer n. Let M be the set of all
the ideals of A that do not contain hm for any m ∈ M ; then M 6= ∅ since M
contains (0). The inductive property of M can be proved as in Theorem 1.2.3.
Let p be a maximal element of M . Let us prove that p is prime.
Let f, g ∈ A and f, g ∈
/ p. Let us prove that f g ∈/ p. Indeed, 6) p + (f ) ⊃ p
and p + (g) ⊃ p (strict inclusions). Since p is maximal in M , we see that
hn ∈ p + (f ) and hm ∈ p + (g) for some m, n ∈ N. Hence, hn+m ∈ p + (f g);
but hn+m ∈ / p, and hence f g ∈
/ p. Thus, p is prime. u
t
This result might give an impression that there is no room for nilpotents in
a geometric picture. This would be a very false impression: 7) On the contrary,
nilpotents provide with an adequate language for description of differential-
geometric notions like “tangency”, “the multiplicity of an intersection”, “in-
finitesimal deformation”, “the fiber of a map at the points where regularity is
violated”.
1.3.3. Examples. 1) A multiple intersection point. In the affine plane
over R, consider the parabola T1 − T22 = 0 and the straight line T1 − t = 0,
where t ∈ R is a parameter.
Their intersection is given by the system of equations
6
Recall that (f ) denotes the ideal generated by an element f . The notation
(f1 , . . . , fn ) means — in this context — the finitely generated ideal, not the
vector (f1 , ..., fn ) ∈ An .
7
The instances of this sort will be illustrated by the following examples 3.2–3.5
and in the next section to say nothing of the main body of the book, where odd
indeterminates, especially odd parameters of representations of Lie supergroups
and Lie superalgebras, are the heart (and the soul, as B. de Witt might had put
it) of the matter.
18 Ch. 1. Affine schemes
(
T1 − T22 = 0,
(1.17)
T1 − t = 0.

to which the ring At = R[T1 , T2 ]/(T1 − T22 , T1 − t) corresponds. An easy

calculation shows that

R × R if t > 0,
At ∼
= R[T ]/(T 2 ) if t = 0, (1.18)

C if t < 0.

The geometric R-points of At are the following ones: There are two of them
for t > 0, there is one of them for t = 0, there are none for t < 0.
The geometric C-points: There are always two of them except for t = 0
(the case of tangency).
In order to be able to state that over C there are always two intersection
points if proper multiplicities are ascribed to them, we have to assume that
at t = 0 the multiplicity of the intersection point is equal to 2.
Observe that dimR At = 2 regardless of the value of t. The equality:

dimR At = the number of intersection points (multiplicity counted) (1.19)

is not accidental, and we will prove the corresponding theorem when we intro-
duce the projective space that will enable us to take into account the points
that escaped to infinity.
A singularity like the coincidence of the intersection points corresponding
to tangency creates nilpotents in A0 .
2) One-point spectra. Let Spec A consist of one point corresponding to
an ideal p ⊂ A. Then A/p is a field and p consists of nilpotents. The ring A
is Artinian 8) , hence Noetherian, and the standard arguments show that p is
a nilpotent ideal.
Indeed, let f1 , . . . , fn be its generators and fim = 0 for 1 ≤ i ≤ n. Then
mn X
à n !
Y
aij fi = 0 (1.20)
j=1 i=1

for any aij ∈ A, where 1 ≤ i ≤ n and 1 ≤ j ≤ mn, since, in every monomial

of the product, there enters at least one of the fi raised to the power ≥ m.
Recall that the length of a module M is the length r of the filtration

M = M1 ⊃ M2 ⊃ · · · ⊃ Mr = 0. (1.21)

The filtration (1.21) of a module M is said to be simple if each module Mi /Mi+1 is

simple, whereas as module is said to be simple if it contains no proper submodules
(different from {0} and itself). A module is said to be of finite length if either it is
{0} or admits a simple finite filtration.
8
Recall that the ring is Artinian if the descending chain condition (DCC) on ideals
holds.
1.4 The Zariski topology on Spec A 19

Therefore pmn = 0. In the series A ⊃ p ⊃ p2 ⊃ . . . ⊃ pmn = (0), the quo-

tients pi /pi+1 are finite dimensional vector spaces over the field A/p. There-
fore, as a module over itself, A is of finite length. In the intersection theory,
the length of a local ring A plays the role of the multiplicity of the only point
of Spec A, as we have just seen. The multiplicity of the only point of Spec A
is equal to 1 if and only if A is a field.
3) Differential neighborhoods. Jets. Let x ∈ Spec A be a point, px the
corresponding ideal. In differential geometry, and even in freshmen’s calculus
courses, we often have to consider the m-th differential neighborhood of x, i.e.,
take into account not only the values of functions but also the values of its
derivatives to the m-th inclusive, in other words, consider the m-jet of the
function at x. This is equivalent to considering the function’s Taylor series
expansion in which the infinitesimals of order greater than m are neglected,
i.e., forcefully equated to 0.
Algebraically, this means that we consider the class f (mod pm+1 x ). The
elements from px are infinitesimals of order ≥ 1, and, in the ring A/pm+1 x ,
they turn into nilpotents.
In what follows we will see that the interpretation of Spec A/pm+1x as the
differential neighborhood of x is only natural when px is maximal. In the
general case of a prime but not maximal px , this intuitive interpretation can
not guide us but still is useful.
4) Reduction modulo pN . Considering Diophantine equations or, equiv-
alently, the quotients of the ring Z[T1 , . . . , Tn ], one often makes use of the
reduction modulo powers of a prime, see sec. 1.1.3. This immediately leads to
nilpotents and we see that from the algebraic point of view this process does
not differ from the consideration of differential neighborhoods in the above
example.
(The congruence 35 ∼ = 7 mod 53 means that “at point (5) the functions
35 and 7 coincide up to the second derivative inclusive”. This language does
not look too extravagant in the number theory after Hensel had introduced
p-adic numbers.)
1.3.4. Exercises. 1) Let a1 , . . . , an ⊂ A be ideals. Prove that

V (a1 . . . an ) = V (a1 ∩ · · · ∩ an ).

2) Let f1 , . . . , fn ∈ A, and m1 , . . . , mn positive integers. If (f1 , . . . , fn ) = A,

then (f1m1 , . . . , fnmn ) = A.
3) The elements f ∈ A not vanishing in any point of Spec A are invertible.

1.4. The Zariski topology on Spec A

The minimal natural condition for compatibility of the topology with
“functions” is that the set of zeroes of any function should be closed. The
topology on Spec A that satisfies this criterion is called Zariski topology. To
describe it, for any subset E ⊂ A, denote the variety singled out by E to be
20 Ch. 1. Affine schemes

V (E) = {x ∈ Spec A | f (x) = 0 for any f ∈ E}. (1.22)

1.4.1. Lemma. The sets V (E), where E ⊂ A, constitute the system

of closed sets in a topology of Spec A.
The topology defined in Lemma 1.4.1 is called the Zariski topology.
Proof. Since ∅ = V (1) and Spec A = V (∅), it suffices to verify that V (E) is
closed with respect to finite unions and arbitrary intersections. This follows
from the next statement.
1.4.1a. Exercise. Set E1 E2 = {f g | ³ f ∈ ´ E1 , g ∈ E2 }. Then
T T
V (E1 ) ∪ V (E2 ) = V (E1 E2 ) and V (Ei ) = V Ei for any I. u
t
i∈I i∈I

Using Theorem 1.3.1, we can describe the set of functions that vanish on
V (E). Obviously, it contains all the elements of the ideal (E) generated by E
and all the elements f ∈ A such that f n ∈ (E) for some n. It turns out that
this is all.
The radical r(I) of the ideal I ⊂ A is the set (actually, an ideal in the ring
A) defined to be

r((E)) = {f ∈ A | there exists n ∈ N such that f n ∈ I}. (1.23)

An ideal coinciding with its own radical is said to be a radical ideal.

1.4.2. Theorem. If f (x) = 0 for any x ∈ V (E), then f ∈ r((E)).
Proof. The condition

f (x) = 0 for all x ∈ V (E)

T
means that f ∈ px , i.e., every element f (mod (E)) from
px ⊂Spec A/(E)
A/(E) belongs to the intersection of all prime ideals of A/(E). Therefore
f n (mod (E)) = 0 for some n, thanks to Theorem 1.3.1. This proves the
statement desired. u
t
1.4.2a. Corollary. The map I 7−→ V (I) establishes a one-to-one correspon-
dence between the radical ideals of A and the closed subsets of its spectrum.
Proof immediately follows from Theorem 1.3.1. u
t
The topology of the spaces Spec A is very non-classical, in the sense that is
very non-Hausdorff (non-separable). We consider typical phenomena specific
for algebraic geometry.
1.4 The Zariski topology on Spec A 21

1.4.3. Non-closed points. Let us find the closure of a given point

x ∈ Spec A. We have
\ ³ [ ´
{x} = V (E) = V E = V (px ) = {y ∈ Spec A | py ⊃ px }.
E⊂px E⊂px

In other words,

{x} ∼
= Spec(A/px ), and
only the points corresponding to the maximal ideals are closed.

A specific relation among points, y ∈ {x}, is sometimes expressed by saying

that y is a specialization of x; this is equivalent to the inclusion px ⊂ py . If A
has no zero divisors, then {0} ∈ Spec A is the point whose closure coincides
with the whole spectrum.
Therefore Spec A is stratified: The closed points are on the highest level,
the preceding level is occupied by the points whose specializations are closed,
and so on, the i-th level (from above) is manned by the points whose special-
izations belong to the level < i. The apex of this inverted pyramid is either
the generic point, (0), if A has no zero divisors, or a finite number of points
if A is a Noetherian ring (for proof, see subsec. 1.4.7d).

Fig. 5

On Fig. 5 there are plotted two spectra: The spectrum of the ring of integer
p-adic numbers, Zp , and Spec C[T1 , T2 ]. The arrows indicate the specialization
relation. The picture of Spec Zp does not require comments; note only that
Spec A may be a finite but not discrete topological space. The other picture
is justified by the following statement.
Proposition. Let K be algebraically closed. The following list exhausts the
prime ideals of the ring K[T1 , T2 ]:
a) the maximal ideals (T1 − t1 , T2 − t2 ), where t1 , t2 ∈ K are arbitrary;
22 Ch. 1. Affine schemes

b) the principal ideals (F (T1 , T2 )), where F runs over all irreducible poly-
nomials;
c) (0).
For proof, see [Reid].
The images influenced by this picture can serve as a base for a working
dimension theory of algebraic geometry. We will show this later; at the moment
we will confine ourselves to preliminary definitions and two simple examples.
We say that a sequence of points x0 , x1 , . . . , xn of a topological space X
is a chain of length n with the beginning at x0 and the end at xn if xi 6= xi+1
and xi+1 is a specialization of xi for all 0 ≤ i ≤ n − 1.
The height ht x of x ∈ X is the upper bound of the lengths of the chains
with the beginning at x. The dimension dim X of X is the upper bound of the
heights of its points.
Example. In the space X = Spec K[T1 , . . . , Tn ], where K is a field, there is
a chain of length n corresponding to the chain of prime ideals

(0) ⊂ (T1 ) ⊂ . . . ⊂ (T1 , . . . , Tn ), (1.24)

and therefore dim X ≥ n.

Similarly, since there is a chain

(p) ⊂ (p, T1 ) ⊂ . . . ⊂ (p, T1 , . . . , Tn ), (1.25)

it follows that dim Spec Z[T1 , . . . , Tn ] ≥ n + 1.

As we will see later, in both cases the inequalities are, actually, equalities.
This definition of dimension can be traced back to Euclid: (Closed) points
“border” lines that “border” surfaces, and so on. 9)
1.4.4. Big open sets. For any f ∈ A, set

D(f ) = Spec A \ V (f ) = {x | f (x) 6= 0}. (1.26)

The sets D(f ) are called big open sets; they constitute a basis of the Zariski
topology of Spec A, since
\
Spec A \ V (E) = D(f ) for any E ⊂ A. (1.27)
f ∈E

Consider, for example, Spec C[T ]. Its closed points correspond to the ideals
(T − t), where t ∈ C, and constitute, therefore, the “complex line”; the non-
empty open sets are {0} and the sets of all the points of the complex line
9
For discussion of non-integer values of dimension, with examples, see [Mdim]. As
Shander and Palamodov showed, in the super setting, the relation of “border”ing
is partly reversed: The odd codimension of the boundary of a supermanifold might
be negative, see [SoS].
1.4 The Zariski topology on Spec A 23

except a finite number of them. T h e c l o s u r e o f a n y o p e n s e t c o -

incides with the whole space!
More generally, if A has no zero divisors and f 6= 0, then D(f ) is dense
in Spec A. Indeed, D(f ) ⊂ (0); hence, D(f ) = (0) = Spec A. Therefore
any non-empty open set of the spectrum of any ring
w i t h o u t z e r o d i v i s o r s i s d e n s e.
In the analysis of the type of (non-)separability of topological spaces,
an important class was distinguished:
1.4.5. Lemma. The following conditions are equivalent:
a) Any non-empty open subset of X is dense.
b) Any two non-empty open subsets of X have a non-empty intersection.
c) If X = X1 ∪ X2 , where X1 , X2 are closed, then either X1 = X or
X2 = X.
Proof. a) ⇐⇒ b) is obvious. If c) is false, then there exists a representa-
tion X = X1 ∪ X2 , where X1 , X2 are proper closed subsets of X; then
X \ X2 = X1 \ (X1 ∩ X2 ) is non-dense open set; therefore a) fails. This is
a contradiction.
Conversely, if a) fails and U ⊂ X is a non-dense open set, then

X = U ∪ (X \ U ). u
t

A topological space X satisfying any of the above conditions is called

irreducible. Notice that no Hausdorff space with more than one point can be
irreducible.
1.4.6. Theorem. Let A be a ring, N its nilradical. The space Spec A is
irreducible if and only if N is prime.
Proof. Let N be prime. Since N is contained in any prime ideal, then Spec A
is homeomorphic to Spec A/N , and A/N has no zero divisors.
Conversely, let N be non-prime. It suffices to verify that Spec A/N is
reducible, i.e., confine ourselves to the case where A has no nilpotents but
contains zero divisors.
Let f, g ∈ A be such that g 6= 0, f 6= 0 but f g = 0. Obviously,

Spec A = V (f ) ∪ V (g) = V (f g). (1.28)

Therefore f and g vanish on the closed subsets of the whole spectrum that
together cover the whole space (this is a natural way for zero divisors to appear
in the rings of functions).
We only have to verify that V (f ), V (g) 6= Spec A. But this is obvious since
f and g are not nilpotents. ut
Corollary. Let I ⊂ A be an ideal. The closed set V (I) is irreducible if and
only if the radical r(I) is prime.
24 Ch. 1. Affine schemes

Thus, we obtain a one-to-one correspondence (cf. Corollary 1.4.2a):

The points of Spec A ←→ the irreducible closed subsets of Spec A. (1.29)

To every point x ∈ Spec A the closed set {x} corresponds; x is called a generic
point of this closed set. Every irreducible closed subset has only one generic
point, obviously.
1.4.7. Decomposition into irreducible components.
1.4.7a. Theorem. Let A be a Noetherian
S
ring. Then Spec A can be uniquely repre-
sented in the form of a finite union Xi , where the Xi are maximal closed irreducible
i
subsets.
Proof. The theorem is a geometric reformulation of the ascending chain condition
on the ideals of A: Every descending chain of closed subsets of Spec A stabilizes. u
t
1.4.7b. Noetherian topological space. Since we will encounter spaces with
the property described in Theorem 1.4.7 not homeomorphic to the spectrum of any
ring, let us introduce a special definition: A given topological space X will be called
Noetherian if any descending chain of its closed subsets stabilizes (we say: DCC
holds for X).
Theorem. Let X be a Noetherian topological space. Then X is finite union of its
maximal closed irreducible subsets.
The maximal closed irreducible subsets of a Noetherian topological space X are
called irreducible components of X.
Proof. Consider the set of irreducible closed subsets of X ordered with respect to
inclusion. Let us prove that the set is an inductive one, i.e., if {Xα | α ∈ J} is
a linearly ordered family of irreducible closed subsets of X, then for its maximal
T  T 
element we may take Xα . The irreducibility of the set Xα follows, for
α α
example, from the fact that if U1 , U2 are non-empty open subsets, then U1 ∩ Xα
and U2 ∩ Xα are non-empty for some α, and therefore U1 ∩ U2 is non-empty since
the Xα are irreducible. S
It follows that X is the union of its irreducible components: X = Xi .
i∈I
So far, we have not used the Noetherian property.
Now, let X be Noetherian and X = X1 ∪ X2 , where X1 , X2 are closed. If one
of the Xi is reducible, we can represent it in the form of the union of two closed
sets, and so on. This process terminates, otherwise we would have got an infinite
descending chain of closed sets (the “Noetherian induction”).
S
In the obtained finite
union, let us leave only the maximal elements: X = Xi . This decomposition
1≤i≤n
coincides with
S
the above one: If YS is an (absolutely) maximal closed subset of X,
then Y ⊂ Xi implies Y = (Xi ∩ Y ), and therefore Xi ∩ Y = Y for some i;
1≤i≤n 1≤i≤n
hence Y = Xi . S
If I 0 is a proper subset of I, then Xi does not coincide with X: Let Xj be
i∈I 0 S S
a discarded component, i.e., j 6∈ I 0 . If Xj ⊂ Xi then Xj = (Xi ∩ Xj ) and,
i∈I 0 i∈I 0
1.4 The Zariski topology on Spec A 25

due to the irreducibility of Xj , we would have Xi ∩ Xj = Xj for some i ∈ I 0 . This

is a contradiction. u
t
1.4.7c. Corollary. For any Noetherian ring A, the number of its minimal prime
ideals is finite.
Proof. Indeed, the minimal prime ideals of Spec A are the generic points of maximal
closed subsets, i.e., the irreducible components of Spec A. u
t
1.4.7d. Corollary. Let A be a Noetherian ring. If all the points of Spec A are
closed, then Spec A is a finite and discrete space.
The rings A with this property (all the points of Spec A are closed) are called
Artinian ones. We recall that a common definition of an Artinian ring is as a one for
which DCC on ideals holds. By a theorem ([ZS]) DCC on ideals implies that every
prime ideal is a maximal one we arrive at another formulation:
An Artinian ring is a Noetherian ring
all prime ideals of which are maximal.
The spectra of Artinian rings resemble very much finite sets in the usual topology.
As noted in sec. 1.3.3 (and in what follows), every point of the spectrum of an
Artinian ring is additionally endowed with a multiplicity.
1.4.8. An interpretation of zero divisors. The following theorem will
be refined later.
Theorem. 1) Any element f ∈ A that vanishes (being considered as a func-
tion) on one of irreducible components of Spec A is a zero divisor of A.
2) Conversely, if f (mod N ) is a zero divisor in A/N , where N is the
nilradical of A, then f vanishes on one of irreducible components of Spec A.
Remark. The nilpotents cannot be excluded from the second heading of the
theorem: If f is a zero divisor only in A, not in A/N , then it is possible for f
not to vanish on an irreducible component. Here is an example: Let the group
structures of the rings A and B ⊕ I, where B is the subring of A without zero
divisors, be isomorphic, I ⊂ A an ideal with zero multiplication. Let I ∼ = B/p
as B-modules, where p ⊂ B is a non-zero prime ideal. Then the elements from
p are zero divisors in A, since they are annihilated under multiplication by
the elements from I.
On the other hand, clearly, Spec A ∼ = Spec B, the spectra are irreducible,
and the nonzero elements from p cannot vanish on the whole Spec A.
Proof of Theorem. Let Spec A = X ∪ Y , where X is an irreducible com-
ponent on which f ∈ A vanishes, Y the union of the other irreducible com-
ponents. Since Y is closed and X 6⊂ Y , there exists g ∈ A that vanishes on
Y but is not identically zero when restricted to X. Then f g vanishes at all
the points of Spec A; hence, (f g)n = 0 for some n. It follows, f (f n−1 g n ) = 0.
This does not prove yet that f is a zero divisor since it might happen that
f n−1 g n = 0; but then we may again separate f and continue in this way until
we obtain f m g n = 0 and f m−1 g n 6= 0. This will always be the case eventually
because g n 6= 0; otherwise, g would have also vanished on X.
26 Ch. 1. Affine schemes

Now, let f = f (mod N ) be a zero divisor in N/M , i.e., f g = 0 for some

g = g (mod N ). Then Spec A = Spec A/N = V (f ) ∪ V (g). Splitting V (f ) and
V (g) into irreducible components we see that at least one of the irreducible
components of V (f ) is also irreducible in Spec A. Otherwise, all the irreducible
components of Spec A would have been contained in V (g) contradicting the
fact that g 6= 0, i.e., g 6∈ N . Therefore f vanishes on one of the irreducible
components of Spec A, as required. u
t
Examples. 1) Let A be a unique factorization ring, f ∈ A. The space
Spec A/(f ) ∼= V (f ) is irreducible if and only if f = epn , where p is an in-
decomposable element and e is invertible. This follows directly from Theo-
rem 1.4.6. In particular, let A = K[T1 , . . . , Tn ], where K is a field. Then V (f )
corresponds to the hypersurface (in the affine space) singled out by one equa-
tion: f = 0. We have just obtained a natural criterion for irreducibility of such
a hypersurface.
2) Let K be a field, Char K 6= 2, and f ∈ K[T1 , . . . , Tn ] a quadratic form.
The equation f = 0 determines a reducible set if and only if rk f = 2. Indeed,
the reducibility is equivalent to the fact that f = l1 · l2 , where l1 and l2 are
non-proportional linear forms.
1.4.9. Connected spaces. The definition of connectedness from general
topology is quite suitable for us: A space X is said to be connected if it can
not be represented as the union of two non-intersecting non-empty closed
subsets. Clearly, any irreducible space is connected.
Any space X can be uniquely decomposed into the union of its maximal
connected subspaces which do not intersect ([K]) and are called the connected
components of X. Every irreducible component of X belongs entirely to one
of its connected components. Theorem 1.4.7b implies, in particular, that the
number of connected components of a Noetherian space is finite.
The space Spec A might be disconnected. For the spectral, as well as for
the usual (Hausdorff), topology, we, clearly, have:
`
The ring of continuous function on the disjoint union X1 X2
naturally factorizes into the direct product of the rings (1.30)
of functions on X1 and X2 .

Same happens with the spectra.

1.4.10. The decomposition of Spec A corresponding to a Q factoriza-
tion of A. Let A1 , . . . , An be some rings, and let the product Ai = A
1≤i≤n
be endowed with the structure of a ring with coordinate-wise addition and
multiplication. The set

ai = {x ∈ A | all coordinates of x except the i-th one are 0} (1.31)

is an ideal of A, and ai aj = 0 for i 6= j. Set

1.4 The Zariski topology on Spec A 27
X
bi = ar and Xi = V (bi ) ⊂ Spec A. (1.32)
r6=i

Then
[
Xi = V (b1 . . . bn ) = V (0) = Spec A,
1≤i≤n

Xi ∩ Xj = V (bi ∪ bj ) = V (A) = ∅ if i 6= j.
Therefore
Q
Spec Ai splits into the disjoint union of its closed subsets
1≤i≤n (1.33)
V (bi ) ∼
= Spec A/bi = Spec Ai .
Note that, for infinite products, the statement (1.33) is false, see Exercise
1.4.14 7e).
1.4.11. The statement converse to (1.30).
`
Proposition. Let X = Spec A = Xi , where the Xi are closed non-
1≤i≤n
intersecting subsets. Then there exists an isomorphism A = ∼ Q Ai such
1≤i≤n
that in the notation of the above subsection Xi = V (bi ).
Proof. We will consider in detail the case n = 2. Let Xi = V (bi ). By Corol-
lary 2.4.12 we have

X1 ∪ X2 = X ⇐⇒ V (b1 b2 ) = X ⇐⇒ b1 b2 ⊂ N,
X1 ∩ X2 = ∅ ⇐⇒ V (b1 + b2 ) = ∅ ⇐⇒ b1 + b2 = A.

Therefore there exist elements fi ∈ bi and an integer r > 0 such that

f1 + f2 = 1 and (f1 f2 )r = 0.
Lemma. Let f1 , . . . , fn ∈ A. If (f1 , . . . , fn ) = A, then (f1m1 , . . . , fnmn ) = A
for any positive integers m1 , . . . , mn .
T
Proof. By sec. 1.4 we have (g1 , . . . , gn ) = A if and only if V (gi ) = ∅. Since
i
V (g m ) = V (g) for m > 0 we obtain the statement desired. u
t
Thanks to Lemma we have g1 f1r + g2 f2r = 1 for some gi ∈ A. Set ei = gi fir
for i = 1, 2. Then e1 +e2 = 1, e1 e2 = 0, and therefore the ei ∈ bi are orthogonal
idempotents which determine a factorization of A:
∼
A −→ A1 × A2 , g 7→ (ge1 , ge2 ). (1.34)
It only remains to show that V (Aei ) = Xi . But, clearly, the V (Aei ) do not
intersect, their union is X, and Aei ⊂ bi ; hence Xi ⊂ V (Aei ) yielding the
statement desired.
1.4.11a. Exercise. Complete the proof by induction on n.
u
t
28 Ch. 1. Affine schemes

1.4.12. Example. Let A be an Artinian ring. Since Spec A is the union

of a finite number of closed points, A is isomorphic to the product of the
finite number of local Artinian rings. In particular, any Artinian ring is of finite
length (cf. Example 1.3.3).
1.4.13. Quasi-compactness. The usual term “compact” is prefixed with
a quasi- to indicate that we are speaking about non-Hausdorff spaces. A to-
pological space X is said to be quasi-compact if every of its open coverings
contains a finite subcovering.
The following simple result is somewhat unexpected, since it does not
impose any finiteness conditions onto A:
Proposition. Spec A is quasi-compact for any A.
Proof. Any S cover of SpecTA can be refined to a cover with big open sets:
Spec A = D(fi ). Then V (fi ) = ∅, so that (fi )i∈I = A. Therefore there
i∈I i P
exists a partition of unity 1 = gi fi , where gi 6= 0 for a finite number
i∈I S
of indices i ∈ J ⊂ I. Therefore Spec A = D(fi ), as desired. u
t
i∈J

1.4.14. Exercises. 1) Let S ⊂ A be a multiplicative system 10) . A given

multiplicative system S is said to be complete, if f g ∈ S =⇒ f ∈ S and g ∈ S.
Every multiplicative system S has a uniquely determined completion S: e It is
the minimal complete multiplicative system containing S.
Show that D(f ) = D(g) ⇐⇒ (f gn) g n
n≥0 = (g )n≥0 .
2) Show that the spaces D(f ) are quasi-compact.
3) Are the following spaces connected:
a) Spec K[T ]/(T 2 − 1), where K is a field;
b) Spec Z[T ]/(T 2 − l)?
4) The irreducible components of each of the curves

T1 (T1 − T22 ) = 0,
T2 (T1 − T22 ) = 0,

in Spec C[T1 , T2 ] consists of the line and a parabola, and hence are pair-
wise isomorphic. The intersection point in each case is the “apex” of the
parabola. Prove that, nevertheless, the rings of functions of these curves are
non-isomorphic.
5) Let A be a Noetherian ring. Construct a graph whose vertices are in one-
to-one correspondence with the irreducible components of the space Spec A,
and any two vertices are connected by an edge if and only if the corresponding
components have a non-empty intersection. Prove that the connected compo-
nents of Spec A are in one-to-one correspondence with the linearly connected
components of the graph.
10
I.e., 1 ∈ S and f, g ∈ S =⇒ f g ∈ S.
1.5 The affine schemes (a preliminary definition) 29

Q
n
6) Finish the proof of Proposition 1.4.11. Is the decomposition A = Ai ,
i=1
whose existence is claimed, uniquely defined? Q
7) Let (Ki )i∈I be a family of fields. Set A = Ki and let πi : A −→ Ki
be the projection homomorphisms. i∈I

a) Let a ⊂ A be a proper ideal. Determine from it a family Φa of subsets

of I by setting:

K ∈ Φa ⇐⇒ there exists f ∈ a such that πi (f ) = 0,

if and only if i ∈ K.

Show that the subsets K are non-empty and the family Φa possesses the
following two properties:

α) K1 ∈ Φa & K2 ∈ Φa =⇒ K1 ∩ K2 ∈ Φa ,
β) K1 ∈ Φa & K2 ⊃ K1 =⇒ K2 ∈ Φa .

b) The family Φ of non-empty subsets of a set I with properties α) and

β) is called a filter on the set I.
Let Φ be a filter on I; assign to it the set aΦ ⊂ A by setting:

f ∈ aΦ ⇐⇒ {i | πi (f ) = 0} ∈ Φ.

Show that aΦ is an ideal in A.

c) Show that the maps a 7→ Φa and Φ 7→ aΦ determine a one-to-
one correspondence between the ideals of A and filters on I. Further,
a1 ⊂ a2 ⇐⇒ Φa1 ⊂ Φa2 . In particular, to the maximal ideals maximal fil-
ters correspond; they are called ultrafilters.
d) Let i ∈ I, and Φ(i) = {K ⊂ I | i ∈ K}. Show that Φ(i) is an ultrafilter.
Show that if I is a finite set, then any ultrafilter is of the form Φ(i) for some
i. Which ideals in A correspond to filters Φ(i) ? What are the quotients of A
modulo these ideals?
e) Show that if the set I is infinite, then there exists an ultrafilter on I
distinct from Φ(i) . (
Hint. Let Φ = {K ⊂ I | I \ K} be finite; let Φ be a maximal filter containing
6= Φ(i) for all i ∈ I.)
Φ. Verify that ΦQ
f) Let A = Z/qZ, where I is the set of all primes. Let p ⊂ A be the
q∈I
prime ideal corresponding to an ultrafilter distinct from all the Φ(q) . Show
that A/p is a field of characteristic 0.

1.5. The affine schemes (a preliminary definition)

To any map of sets f : X −→ Y , there corresponds a homomorphism of
the rings of functions on these sets f ∗ : F (Y ) −→ F (X) given by the formula
30 Ch. 1. Affine schemes

f ∗ (ϕ)(x) = ϕ(f (x)). (1.35)

If X, Y are topological spaces and F (X), F (Y ) are rings not of arbitrary, but
of continuous functions, then the homomorphism f ∗ is uniquely recovered
from a continuous f . Without certain conditions this correspondence does
not have to take place, e.g., the homomorphism f ∗ of rings of, say, continuous
functions is not uniquely recovered from an arbitrary f .
The prime objects of our study are the rings (of “functions”); therefore
important maps of the spaces are — for us — only those obtained from ring
homomorphisms.
Let ϕ : A −→ B be a ring homomorphism. To every prime ideal p ⊂ B,
we assign its pre-image ϕ−1 (p). The ideal ϕ−1 (p) is prime because ϕ induces
the embedding A/ϕ−1 (p) ,→ B/p and, since B/p has no zero divisors, neither
has A/ϕ−1 (p). We have determined a map aϕ : Spec B −→ Spec A, where the
superscript a is for “affine”.
1.5.1. Theorem. 1) aϕ is continuous as a map of topological spaces (with
respect to the Zariski topologies of these spaces).
a
2) (ϕψ) = aψ ◦ aϕ.
Proof. 2) is obvious. To prove 1), it suffices to verify that the pre-image of
a closed set is closed. Indeed,

y ∈ V (ϕ(E)) ⇐⇒ ϕ(E) ⊂ py ⇐⇒ E ⊂ ϕ−1 (py ) = p aϕ(y)

a
⇐⇒ ϕ(y) ∈ V (E) ⇐⇒ y ∈ ( aϕ)−1 (V (E)).

We have actually proved that ( aϕ)−1 (V (E)) = V (ϕ(E)).

u
t
Therefore
Spec : Rings◦ −→ Top is a functor.
1.5.2. Affine schemes as spaces. An affine scheme is a triple (X, α, A),
∼
where X is a topological space, A a ring and α : X −→ Spec A an isomorphism
of spaces.
A morphism of affine schemes (Y, β, B) −→ (X, α, A) is a pair (f, ϕ), where
f : Y −→ X is a continuous map and ϕ : A −→ B is a ring homomorphism,
such that the diagram
β
Y / Spec B

f a
ϕ (1.36)
² α
²
X / Spec A

commutes. The composition of morphisms is obviously defined.

To every ring A, there corresponds the affine scheme (Spec A, id, A), where
id is the identity map; for brevity, we often shorten its name to Spec A. Clearly,
1.5 The affine schemes (a preliminary definition) 31

any affine scheme is isomorphic to such a scheme; affine schemes constitute

a category; the dual category is equivalent to the category of rings.
Our definition is not final since it is ill-adjusted to globalizations, the
glueings of general schemes from affine ones. In what follows we will modify
it: An additional element of the structure making the space Spec A into the
scheme Spec A is a sheaf associated with the ring A. But the above definition
will do for a while since
the ring A and the corresponding sheaf of functions
(1.37)
on Spec A can be uniquely recovered from each other.

1.5.3. Examples. To appreciate the difference between the set Hom(A, B),
which only is of importance for us, and the set of all the continuous maps
Spec B −→ Spec A consider several simple examples.
1.5.3a. A = B = Z. As we have already observed, Spec Z consists of the
closed points (p), where p runs over the primes, and (0). The closure of (0) is
the whole space; the remaining closed sets consist of a finite number of closed
points. There are lots of automorphisms of the space Spec Z: We may permute
the closed points at random; contrariwise Hom(Z, Z) contains only the identity
map.
1.5.3b. A = K[T ], where K is a finite field, B = Z. Obviously, Spec A and
Spec B are isomorphic as topological spaces, whereas Hom(A, B) = ∅.
Examples 1.5.3a and 1.5.3b might make one think that there are much
less homomorphisms of rings than there are continuous maps of their spectra.
The opposite effect is, however, also possible.
1.5.3c. Let K be a field. Then Spec K consists of one point, and therefore
the set of its automorphism consists of one point, whereas the group of auto-
morpisms of K may be even infinite (a Lie group). Therefore

one-point spectra may have “inner degrees of freedom”,

(1.38)
like elementary particles.
1.5.3d. The spectrum that personifies an “idea” of a vector.
(“Combing nilpotents”). Let A be a ring, B = A[T ]/(T 2 ). The natu-
ral homomorphism

ε : B −→ A, a + bt 7→ a, where t = T (mod (T 2 )),

induces an isomorphism of topological spaces aε : Spec A −→ Spec B but by

no means that of the schemes.
The scheme (Spec B, id, B) is “richer” than (Spec A, id, A) by the nilpo-
tents tA. To see how this richness manifests itself, consider arbitrary “projec-
tions” aπ : Spec B −→ Spec A, i.e., scheme morphisms corresponding to the
ring homomorphisms π : A −→ B such that επ = id. Then π(f ) − f ∈ At. For
any such π, define the map ∂π : A −→ A by setting
32 Ch. 1. Affine schemes

π(f ) − f = ∂π (f )t. (1.39)

Since π is a ring homomorphism, we see that ∂π (f ) is a derivation of A, i.e.,
is linear and satisfies the Leibniz rule:
∂π (f + g) = ∂π (f ) + ∂π (g),
(1.40)
∂π (f g) = ∂π (f ) · g + f · ∂π (g).
Indeed, the linearity is obvious while the Leibniz rule follows from the identity
π(f g) = f g + ∂π (f g)t = π(f )π(g) = (f + ∂π (f )t)(g + ∂π (g)t) (1.41)
true thank to the property t2 = 0.
It is easy to see that, the other way round, for any derivation ∂ : A −→ A
the map π : A −→ B determined by the formula π(f ) = f + (∂f )t is a ring
homomorphism and determines a projection aπ.

Fig. 6

In Differential Geometry, every derivation of the ring of functions on

a manifold is interpreted as a “vector field” on the manifold. One can imagine
the scheme (Spec B, id) endowed with a field of vectors “looking outwards”,
as on a hedgehog, as compared with the scheme (Spec A, id). The morphism
a
π “combs” these vectors turning them into vector fields on Spec A.
In particular, if K is a field, then Spec K is a point and
the scheme (Spec K[T ]/(T 2 ), id) is an “idea of the vector” (1.42)
with the point Spec K as the source of the vector.
In what follows we will sometimes plot nilpotents in the form of arrows
though it is obvious that even for the simplest scheme like

(Spec K[T1 , T2 ]/(T12 , T1 T2 , T22 ), id);

(Spec K[T ]/(T n ), id) for n ≥ 3, (1.43)
2
Spec(Z/(p ), id) for p prime
such pictures are of limited informative value.
1.6 Topological properties of morphisms and Spm 33

1.5.3e. “Looseness” of affine spaces. Let K be a field (for simplicity’s

.
sake), V a linear space over K and A = SK (V ). Consider the group G of au-
tomorphisms of the K-scheme (Spec A, id). This group consists of transforma-
tions induced by those that constitute the group of K-automorphisms of the
ring K[T1 , . . . , Tn ], where n = dim V . The group G contains as a subgroup
the usual Lie group G0 of invertible affine linear transformations
X j
Ti 7→ ci Tj + di , where cji , di ∈ K. (1.44)
j

For n = 1, it is easy to see that G0 = G. This is far from so for n ≥ 2. Indeed,

in this case any “triangular” substitution of the form

T1 7→ T1 + F1 ,
T2 7→ T2 + F2 (T1 ),
(1.45)
..........................
Ti 7→ Ti + Fi (T1 , . . . , Ti−1 ),

where Fi ∈ K[T1 , . . . , Ti−1 ] ⊂ K[T1 , . . . , Tn ], clearly, belongs to G. Therefore

the group of automorphisms of the scheme corresponding to the affine space
of dimension ≥ 2 contains non-linear substitutions of however high degree.
Their existence is used in the proof of Noether’s normalization Theorem. 11)
Remark. For n = 2, the group G is generated by linear and triangular sub-
stitutions (Engel, Shafarevich); for n ≥ 3, this is not so (Nagata’s conjecture
on automorphisms proved by I. Shestakov and U. Umirbaev 12) ). The explicit
description of G is an open (and very tough) problem 13) .
1.5.3f. Linear projections. Let V1 ⊂ V2 be two linear spaces over
.
a field K, and let Xi = Spec SK (Vi ). The morphism X2 −→ X1 , in-
. .
duced by the embedding SK (V1 ) ,→ SK (V2 ), is called the projection of the
scheme X2 onto X1 . On the sets of K-points it induces the natural map
X2 (K) = V2∗ −→ V1∗ = X1 (K) which restricts every linear functional on V2
to V1 .

1.6. Topological properties of certain morphisms and

the maximal spectrum
In this section we study the most elementary properties of the morphisms
11
Noether’s normalization Theorem: Let k[x], where x = (x1 , . . . , xn ), be a
finitely generated entire ring over a filed k, and assume that k(x) has transcen-
dence degree r. Then there exist elements y1 , . . . , yr in k[x] such that k[x] is
integral over k[y]. ([Lang])
12
For a generalization of the result of Shestakov and Umirbaev, see [Ku].
13
This problem is close to a famous and still open Jacobian problem, see [EM].
34 Ch. 1. Affine schemes

a
ϕ : Spec B −→ Spec A, px 7→ ϕ−1 (px ). (1.46)

This study gives a partial answer to the question: What is the structure of the
topological space aϕ(Spec B)?
As is known (see, e.g., [Lang]), any homomorphism ϕ : A −→ B factorizes
into the product of the surjective ring homomorphism A −→ A/ Ker ϕ and an
embedding A/ Ker ϕ −→ B. Let us find out the properties of aϕ in these two
cases.
a
1.6.1. The properties of ϕ. The first of these cases is very simple.
Proposition. Let ϕ : A −→ B be a ring epimorphism. Then aϕ is a homeo-
morphism of Spec B onto the closed subset V (Ker ϕ) ⊂ Spec A.
1.6.1a. Exercise. Verify that this is a direct consequence of the definitions.
Hint. Prove the continuity of the inverse map

( aϕ)−1 : V (Ker ϕ) −→ Spec B. u

t

In particular, let A be a ring of finite type over a field K or over Z, i.e., let
A be a quotient of either K[T1 , . . . , Tn ] or Z[T1 , . . . , Tn ] for n < ∞.
The spectrum of the polynomial ring plays the role of an affine space
(over K or Z, respectively, cf. Example 1.2.1). Therefore s p e c t r a
of the rings of finite typ e corresp ond to affine vari-
e t i e s (“arithmetic affine varieties” if considered over Z): They are embedded
into finite dimensional affine spaces.
Thus, ring epimorphisms correspond to embeddings of spaces. Ring
monomorphisms do not necessarily induce surjective maps of spectra: Only
the closure of aϕ(Spec B) coincides with Spec A. This follows from a trifle
more general result.
1.6.2. Proposition. For any ring homomorphism ϕ : A −→ B and an ideal
b ⊂ B, we have
aϕ(V (b)) = V (ϕ−1 (b)).

In particular, if Ker ϕ = 0, then aϕ(V (0)) = V (0), i.e., the image of Spec B
is dense in Spec A.
Proof. We may assume that b is a radical ideal, since V (r(b)) = V (b) and
ϕ−1 (r(b)) = r(ϕ−1 (b)). The set aϕ(V (b)) is the intersection of all closed sub-
sets containing aϕ(V (b)), i.e., the set of common zeroes of all the functions
f ∈ A which vanish on aϕ(V (b)). But f vanishes on aϕ(V (b)) if and only if
ϕ(f ) vanishes on V (b), i.e., if and only if ϕ(f ) ∈ b (since b is a radical ideal)
or, finally, if and only if f ∈ ϕ−1 (b). Therefore the closure we are interested
in is equal to V (ϕ−1 (b)). u
t
Now, let us give examples of ring monomorphisms for which aϕ(Spec B)
does not actually coincide with Spec A.
1.6 Topological properties of morphisms and Spm 35

1.6.2a. Examples. 1)The projection of a hyperbola onto a coordinate axis.

For a field K, consider an embedding

ϕ : A = K[T1 ] ,→ K[T1 , T2 ]/(T1 T2 − 1) = B. (1.47)

Then aϕ(Spec B) = D(T1 ), in accordance with the picture, see Fig. 7.

Fig. 7

Indeed, aϕ maps a generic point into a generic one. The prime ideal
(f (T1 )) ⊂ A, where f 6= cT1 is an irreducible polynomial, is the pre-image
of the prime ideal (f (T1 )) (mod T1 T2 − 1) ⊂ B. Finally, T1 and T1 T2 − 1
/ aϕ(Spec B).
generate the ideal (1) = K[T1 , T2 ], and therefore (T1 ) ∈ u
t
a
In this example ϕ(Spec B) is open; but it may be neither open nor closed:
2) The projection of a hyperbolic paraboloid onto the plane. Consider the
homomorphism

ϕ : A = K[M, N ] ,→ B = K[M, N, T ]/(M T − N ).

1.6.2b. Exercise. Verify that aϕ(Spec B) = D(M ) ∪ V (M, N ), and that

this set is actually not open (it is obvious that it is not closed).
u
t
This example, see also Fig. 8, illustrates the phenomenon noted long ago
in the study of equations:

The set aϕ(Spec B) is the set of values of the coefficients M , N

in a K-algebra for which the equation M T − N = 0 is solvable for (1.48)
T.
In general, the condition for solvability is the inequality M 6= 0, but even for
M = 0 the solvability is guaranteed if, in addition, N = 0.
We may prove that if A is Noetherian and an A-algebra B has a finite num-
ber of generators, then aϕ(Spec B) is the union of a finite number of locally
36 Ch. 1. Affine schemes

Fig. 8

closed sets 14) . Such unions are called constructible sets; the image of a con-
structible set under aϕ is always constructible under the described conditions
(Chevalley’s theorem [AM]).
In terms of undetermined coefficients of a (finite) system of equations this
means that the condition for the system’s compatibility is the following:

The coefficients should satisfy one of a finite number of statements,

each statement is a collection of a finite number of polynomial (1.49)
equalities and inequalities (with the zero right-hand sides).

For example, for the equation M T − N = 0, there are the two statements:
1. M 6= 0;
2. M = 0, N = 0.
1.6.3. Analogs of “finite-sheeted coverings” of Riemannian surfaces.
In the above examples something “escaped to infinity”. Let us describe an im-
portant class of morphisms aϕ for which this does not happen.
Let B be an A-algebra. An element x ∈ B is called integer over A if it satisfies
an equation of the form

xn + an−1 xn−1 + . . . + a0 = 0, a0 , . . . , an−1 ∈ A (1.50)

and B is called integer over A if all its element are integer over A.
There are two important cases when it is easy to establish whether B is integer
over A.
Case 1. If B has a finite number of generators as an A-module, then B is integer
over A.
Indeed, if A is Noetherian ring, then, for any g ∈ B, the ascending sequence
of A-modules X
Bk = Ag i ⊂ B (1.51)
0≤i≤k
P
stabilizes. Therefore, for some k, we have g k ∈ Ag i , which provides us with
0≤i≤k−1
an equation of integer dependence.
14
A locally closed set is the intersections of a closed and an open set.
1.6 Topological properties of morphisms and Spm 37

The general
P
case reduces to the above one with the help of the following trick.
Let B = Afi . Let
1≤i≤n
X X
fi fj = akij fk , where akij ∈ A, and g = gi fi , where gi ∈ A; (1.52)
1≤k≤n 1≤i≤n

by A0 ⊂ A the minimal subring containing all the akij and gi and set
denote P
B0 = A0 fi . Obviously, A0 is a Noetherian ring, B is an A0 -algebra and g ∈ B0 .
1≤i≤n
Therefore g satisfies the equation of integer dependence with coefficients from A.
Case 2. Let G be a finite subgroup of the group of automorphisms of B and
A = B G the subring of G-invariant elements. Then B is integer over A.
Indeed, all the elementary symmetric
Q
polynomials in s(g), where s ∈ G, belong
to A for any g ∈ B, and g satisfies (g − s(g)) = 0. u
t
s∈G

1.6.4. Localization with respect to a multiplicative system. We

would like to define the ring of fractions f /g, where g runs over a set S
of elements of an arbitrary ring A. When we add or multiply fractions their
denominators are multiplied; hence S should be closed with respect to multi-
plication. Additionally we require S to have the unit element, i.e., to be what
is called a multiplicative system.
Examples of multiplicative systems S:
(1) Sf = {f n | n ∈ Z+ } for any f ∈ A,
(2) Sp := A \ p for any prime ideal p.
The set Sp is indeed a multiplicative system because Sp consists of “functions”
f ∈ A that do not vanish at {p} ∈ Spec A; now recall the definition of the
prime ideal.
For any multiplicative system S of a ring A, we define the ring of fractions
AS (also denoted S −1 A or A[S −1 ]), or the localization with respect to S, as
follows. As a set, AS is the quotient of A × S modulo the following equivalence
relation:
(f1 , s1 ) ∼ (f2 , s2 ) ⇐⇒ there exists t ∈ S such that
(1.53)
t(f1 s2 − f2 s1 ) = 0.

Denote by f /s the class of an element (f, s) and define the composition

law in AS by the usual formulas:
f /s + g/t = (f t + gs)/st; (f /s) · (g/t) = f g/st. (1.54)
Exercise. Verify that the above equivalence relation is well-defined. The unit
of AS is the element 1/1, and the zero is 0/1.
Quite often the localizations with respect to Sf and Sp are shortly (and
somewhat self-contradictory) denoted by Af and Ap , respectively.
For the rings without zero divisors, the map a 7→ a/1 embeds A into AS .
In general, however, a nontrivial kernel may appear as is described in the
following obvious Lemma.
38 Ch. 1. Affine schemes

1.6.4a. Lemma. Let j : A −→ AS be the map a 7→ a/1. Then

1) j is a ring homomorphism and

Ker j = {f ∈ A | there exists s ∈ S such that sf = 0};

2) if 0 ∈
/ S, then all the elements from j(S) are invertible in AS , otherwise
AS = 0;
3) every element from AS can be represented in the form j(f )/j(s) with
s ∈ S, f ∈ A.
Notice how drastically AS shrinks when we introduce zero divisors in S.
The following theorem is a main fact on rings of fractions, it describes a
universal character of localization.
1.6.4b. Theorem. Let S be a multiplicative system in a ring A and
j : a 7→ a/1 the canonical homomorphism A −→ AS . For any ring homomor-
phism f : A −→ B such that every element from f (S) is invertible, there exists
a unique homomorphism f 0 : AS −→ B such that f = f 0 ◦ j.
1.6.4c. Exercise. Prove the theorem.
Corollary. Let A be a ring, T and S its multiplicative systems such that
T ⊃ S. Then
1) the following diagram commutes:
jS
A; / AS 3 jS (a)/jS (t)
;; ¡
;; ¡¡
jT ;;
; ¡¡¡ (1.55)
À ¡¡¡ jT (S )
−1
²
AT 3 jT (a)/jT (t)

2) A[T −1 ] ∼
= A[S −1 ][jS (T )−1 ].
1.6.4d. Theorem. Let j : A −→ AS be the canonical homomorphism. Then
the induced map aj : Spec AS −→ Spec A homeomorphically maps Spec AS
onto the subset {x ∈ Spec A | px ∩ S = ∅}.
Proof. We may confine ourselves to the case 0 ∈
/ S. First, let us prove that
there is a one-to-one correspondence

Spec AS ←→ {x ∈ Spec A | px ∩ S = ∅}. (1.56)

Let y ∈ Spec AS and x = aj(y). Then px ∩ S = ∅, since otherwise py would

have contained the image under j of an element from S, which is invertible,
so py would have contained the unit.
Second, let x ∈ Spec A and px ∩ S = ∅. Set py = px [S −1 ]. The ideal py is
prime. Indeed, let (f s−1 )(gt−1 ) ∈ py . We have f g ∈ px and, since px is prime,
then either f ∈ px or g ∈ px , and therefore either f s−1 ∈ py or gt−1 ∈ py .
The fact that the maps py 7→ j −1 (py ) and px 7→ px [S −1 ] are mutually
inverse may be established as follows.
1.6 Topological properties of morphisms and Spm 39

Consider the set {f ∈ A | f /1 ∈ px [S −1 ]}. Let f /1 = f 0 /s, where f 0 ∈ px .

Multiplying by some t ∈ S we obtain tf ∈ px implying f ∈ px .
Now, it remains to show that this one-to-one correspondence is a homeo-
morphism. As had been proved earlier, the map aj is continuous. Therefore it
suffices to prove that under this map the image of every closed set is closed,
i.e., j(V (E)) = V (E 0 ) for some E 0 . Indeed, take

E 0 = {denominators of the elements from E}. t

u

1.6.4e. Example. The set D(f ) is homeomorphic to Spec Af . Indeed,

px ∩ {f n | n ∈ Z} = ∅ ⇐⇒ f ∈
/ px . (1.57)

Therefore Spec A splits into the union of an open and a closed set as follows:

Spec A = Spec Af ∪ Spec A/(f ). (1.58)

Here a certain “duality” between the localization and the passage to the
quotient ring is reflected. Considering D(f ) as Spec Af we “send V (f ) to
infinity”.
If S is generated by a finite number of elements f1 , . . . , fn , then
\
px ∩ S = ∅ ⇐⇒ x ∈ D(fi ) (1.59)
1≤i≤n

and in this case the image of Spec AS is open in Spec A. However, this is not
always the case; cf. Examples 1.6.2.
If S = A\px , then the image of Spec AS in Spec A consists of all the points
y ∈ Spec A whose specialization is x. It is not difficult to see that in general
this set is neither open nor closed in Spec A: It is the intersection of all the
open sets containing x.
The ring Apx = Ox contains the unique maximal ideal px whose spectrum
describes geometrically the “neighborhood of x” in the following sense: We
may trace the behavior of all the irreducible subsets of Spec A through x in
a “vicinity” of x. This is the algebraic version of the germ of neighborhoods
of x in the following sense: We can trace the behavior of the irreducible subsets
of Spec A passing through x in a “vicinity” of x. u
t
1.6.5. Theorem. Let ϕ : A ,→ B be a monomorphism and let B be integer
over A. Then aϕ(Spec B) = Spec A.
Proof. We first prove two particular cases; next we reduce the general state-
ment to these cases.
Case 1. B is a field. Then aϕ(Spec B) = {(0)} ⊂ Spec A and aϕ is
an epimorphism if and only if A has no other prime ideals, i.e., A is a field.
Let us verify this.
40 Ch. 1. Affine schemes

Fig. 9

Let f ∈ A, f 6= 0. Let us show that if f is invertible in B, then so it is

in A. Since f −1 is integer over A, i.e., satisfies an equation
X
f −n + ai f −i = 0, ai ∈ A. (1.60)
1≤i≤n−1

Then multiplying by f n−1 we obtain

X
f −1 = − ai f n−i−1 ∈ A, (1.61)
0≤i≤n−1

as desired.
Case 2. A is a local ring. Then, under the conditions of the theorem, the
unique closed point of Spec A belongs to aϕ(Spec B) and, moreover, it is the
image under aϕ of any other closed point of Spec B.
Indeed, let p be a maximal ideal of A, q any maximal ideal of B. Then
B/q is a field integer over the subring A/(A ∩ q) which by the proved above
case 1, should also be a field. This means that A ∩ q is a maximal ideal in A,
and therefore A ∩ q = ϕ−1 (q) = p.
The general case. Let p ⊂ A be a prime ideal; we wish to show that
there exists an ideal q ⊂ B such that q ∩ A = p. Set S = A \ p.
Considering S as a subset of both A and B, we may construct the rings
of quotients AS ⊂ BS . Set pS = {f /s | f ∈ p, s ∈ S}. It is easy to see that
pS ⊂ AS is a prime ideal. It is maximal since AS \ pS consists of invertible
elements s/1. P
The ring BS is integer over AS since if f ∈ B satisfies f n + ai f i = 0,
0≤i≤n−1
then f /s ∈ BS satisfies
X
(f /s)n + ai /sn−i · (f /s)i = 0. (1.62)
0≤i≤n−1
1.6 Topological properties of morphisms and Spm 41

Therefore, by case 2, there exists a prime ideal qS ⊂ BS such that

AS ∩ qS = pS .
The pre-image of qS in B with respect to the natural homomorphism
B −→ BS is a prime ideal. It remains to verify that A ∩ q = p. The inclusion
p ⊂ A ∩ q is obvious.
Let f ∈ A ∩ q. There exist n ∈ N and s ∈ S such that f /sn ∈ qS . Therefore
f /sn ∈ AS ∩ qS = pS , so that sm f ∈ p for some m ≥ 0; hence f ∈ p. u
t
In this proof the quotient ring AS appeared as a technical trick which
enabled us to “isolate” — localize — the prime ideal p ⊂ A making from it a
maximal ideal pS in AS . The term “localization” is applied to the construction
of quotient rings with precisely this geometric meaning.
1.6.6. Addendum to Theorem 1.6.4a. Denote by Spm A the set of max-
imal ideals of A (the “maximal spectrum”).
Proposition. Under conditions of Theorem 1.6.4a we have
1) aϕ(Spm B) = Spm A;
2) ( aϕ)−1 (Spm A) = Spm B.
Proof. Let p ∈ Spm B; then B/p is a field integer over A/(A∩p) = A/ϕ−1 (p).
Thanks to Theorem 1.6.4a (case 1) A/ϕ−1 (p) is also a field; therefore ϕ−1 (p)
is maximal in A. This demonstrates that a ϕ(Spm B) ⊂ Spm A.
To prove heading 2), consider a prime ideal q ⊂ B such that p = A ∩ q ⊂ A
is maximal. The ring without zero divisors B/q is integer over the field A/p
and we ought to verify that it is also a field. Indeed, any f ∈ B/q, being
integer over A/p, belongs to a finite dimensional A/p-algebra generated by
the powers of f . Multiplication by f in this algebra is a linear map without
kernel, and therefore an epimorphism. In particular, the equation f u = 1 is
solvable, proving the statement desired. u
t
1.6.6a. Warning. Let ϕ : A −→ B be a ring homomorphism, and
x ∈ Spm B, y ∈ Spm A. In general, the point aϕ(x) is nonclosed, and
( aϕ)−1 (y) also contains nonclosed points; so Proposition 2.7.6 describes
a rather particular situation.
Example. Let ϕ : Zp ,→ Zp [T ] be the natural embedding. Then px = (1−pT )
is, clearly, a maximal ideal in Zp [T ] (the quotient modulo it is isomorphic
to Qp ). Obviously,

ϕ−1 (px ) = Zp ∩ (1 − pT ) = (0), (1.63)

and therefore aϕ(x) ∈ / Spm Zp , where x ∈ Spec Zp [T ] is the point corre-

sponding to px . Moreover, the image of the closed point x is a generic point
of Spec Zp ; this generic point is an open set being the complement to (p).
Now, let py = (p) ⊂ Zp [T ] and px = (p) ⊂ Zp . Then y ∈ ( aϕ)−1 (x), and x
is closed whereas y is not.
42 Ch. 1. Affine schemes

This, however, is not unexpected. Even more transparent is the example

of the projection of the plane onto the straight line given by the following
embedding of the corresponding rings

K[T1 ] ,→ K[T1 , T2 ], T1 7→ T1 . (1.64)

The pre-image of the point T1 = 0 of the line contains, clearly, the generic
point of the T2 -axis which is not closed in the plane.

In particular, Spm A is not a functor in A, unlike Spec A.

1.6.7. Exercises. 1) Let B be an A-algebra. Prove that the elements of B,

integer over A, constitute an A-subalgebra of B.
2) Let A ⊂ B ⊂ C be rings; let B be integer over A, and C integer over
B. Prove that C is integer over A.
3) Let A be a unique factorization ring. Then A is an integrally closed ring
in its ring of quotients, i.e., any f /g integer over A belongs to A.

1.7. The closed subschemes and the primary

decomposition
In this section we will often denote the subscheme (V (a), α, S) by Spec A/a
and omit the word “closed” because no other subschemes will be considered
here.
1.7.1. Reduced schemes. Closed embeddings. Let X = Spec A be
an affine scheme, a ⊂ A an ideal. The scheme (V (a), α, A/a), where
∼
α : V (a) −→ Spec A/a is the canonical isomorphism (see sec. 1.6.1) is said
to be a closed subscheme of X corresponding to a. Therefore

the closed subschemes of the scheme X = Spec A are

(1.65)
in 1–1 correspondence with all the ideals of A,

unlike the closed subsets of the space Spec A, which only correspond to the
radical ideals.
The support of Y = Spec A/a ⊂ X is the space V (a); it is denoted by
supp Y .
To the projection A −→ A/a a scheme embedding Y −→ X corresponds;
it is called a closed embedding of a subscheme.
For any ring L, we had set

X(L) := Hom(Spec L, X) := Hom(A, L), (1.66)

and called it the set of L-points of X (cf. definition 1.2.1). Clearly, the
L-points of a subscheme Y constitute a subset Y (L) ⊂ X(L) and the functor
L −→ Y (L) is a subfunctor of the functor L −→ X(L).
1.7 The closed subschemes and the primary decomposition 43

There is a natural order on the set of closed subschemes of a scheme X:

We say that Y1 ⊂ Y2 if a1 ⊃ a2 , where ai is the ideal that determines Yi . The
use of the inclusion sign is justified by the fact that

Y1 ⊂ Y2 ⇐⇒ Y1 (L) ⊂ Y2 (L) for every ring L. (1.67)

The relation “Y is a closed subscheme of X” is transitive in the obvious

sense.
For any closed subset V (E) ⊂ X, there exists a unique minimal closed
subscheme with V (E) as the support: It is determined by the ideal r((E))
and its ring has no nilpotents. Such schemes are called reduced ones.
In particular, the subscheme Spec A/N is the minimal closed subscheme
whose support is the whole space Spec A. If X = Spec A, then its reduction
Spec A/N is denoted by Xred . Thus, a scheme X is reduced if X = Xred .
T
1.7.2. Intersections. The intersection Yi of a family of subschemes Yi , where
i P
Yi = Spec A/ai is the subscheme determined by the ideal ai .
i

T
The

notation is justified by the fact that, for any T ring L, the set of L-points
Yi (L) of the intersection is naturally identified with Yi (L). Indeed, an L-point
i T i
ϕ : A −→ L belongs to Yi (L) if and only if ai ⊂ Ker ϕ for all i, which is equivalent
P i
to the inclusion ai ⊂ Ker ϕ. This argument shows that
i
\  \
supp Yi = supp Yi . (1.68)
i i

1.7.3. Quasiunions. The notion of the union of a family of subschemes is not

defined in a similarSway. In general, for given Yi ’s, there is no closed subscheme Y
such that Y (L) = Yi (L) for all L. However, there exists a minimal subscheme Y
i
such that [
Y (L) ⊃ Yi (L) for all L. (1.69)
i
T S
This Y is determined by the ideal ai . Indeed, if Y (L) ⊃ Yi (L), then the ideal a
i i
that determines Y satisfies the condition

“any ideal containing one of the ai contains a.” (1.70)

The sum a of all such ideals also satisfies (1.70) and is the unique maximal element
of the set of ideal satisfying (1.70); on
T
the other hand, all the elements of this set
are contained in ai , and therefore in ai .
i
Having failed to define the union
W
of subschemes, let us introduce a notion corre-
sponding to ∩ai : The quasiunion Yi of a family of closed subschemes Yi of a scheme
i
X is the subscheme corresponding to the intersection of all the ideals defining the
subschemes Yi .
It is important to notice that the quasiunion of the subschemes Yi does not
depend on the choice of the closed scheme containing all the Yi , inside of which we
construct this quasiunion.
44 Ch. 1. Affine schemes

The main aim of this section is to construct for the Noetherian affine schemes
the decomposition theory into “irreducible” in some sense components similar to
the one constructed above for Noetherian topological spaces. To this end, we will
use the quasi-union; on supports it coincides with the union (for finite families
of subschemes).
 W  S
1.7.3a. Lemma. supp Yi = supp Yi .
1≤i≤n 1≤i≤n
S
Proof. The inclusion ⊃ is already proved. Conversely, if x ∈
/ supp Yi , then,
1≤i≤n
for every i, there exists an element fi ∈ ai such that fi (x) 6= 0, therefore
 Q
fi (x) 6= 0. Hence, x does not belong to the set of zeroes of all the functions
1≤i≤n
T  W 
from ai , which is supp Yi . Now, apply Exercise 1.3.4 1).
1≤i≤n 1≤i≤n
u
t
1.7.4. Irreducible schemes. Now, we have to transport to the subschemes the
notion of irreducibility. The first that comes to one’s mind, is to try to imitate the
definition of irreducibility for spaces.
An affine scheme X is said to be reducible if there exists a representation of the
form X = X1 ∨ X2 , where X1 , X2 are proper closed subschemes of X; it is said to
be irreducible otherwise.
An affine scheme X is said to be a Noetherian one if its ring of global functions
is Noetherian or, equivalently, if X satisfies the descending chain condition on closed
subschemes.
1.7.4a. Theorem. Any Noetherian affine scheme X decomposes into the quasiu-
nion of a finite number of closed irreducible subschemes.
Proof. The same arguments as at the end of sec. 1.4.7b lead to the result desired.
If X is reducible, we write X = X1 ∨ X2 and then decompose, if necessary, X1 and
X2 , and so on. Thanks to the Noetherian property, the process terminates. u
t
1.7.5. Primary schemes and primary ideals. The above notion of irre-
ducibility turns out to be too subtle. The following notion of primary affine
schemes is more useful: An ideal q ⊂ A is called primary if any zero divisor in
A/q is nilpotent. A closed subscheme is called primary if it is determined by
a primary ideal.
Proposition. Any irreducible Noetherian scheme is primary.
Remark. The converse statement is false. Indeed, let K be an infinite field.
Consider the ring A = K × V , where V is an ideal with the zero product. The
ideal (0) is primary and, for any subspace V 0 ⊂ V , the ideal (0, V 0 ) is primary.
At the same time, if dimK V > 1, there exists infinitely many representations
of the form (0) = V1 ∩ V2 , where Vi ⊂ V are proper subspaces, i.e., represen-
tations of the form X = Y1 ∨ Y2 , where X = Spec A. This deprives us of any
hope for the uniqueness of the decomposition into quasiunion of irreducible
subschemes. Primary subschemes behave a sight more nicely than irreducible
subschemes as we will see shortly.
1.7 The closed subschemes and the primary decomposition 45

Proof of Proposition. Let us show that a non-primary Noetherian scheme

X is reducible. Indeed, in the ring of its global functions A there are two
elements f, g such that f g = 0, g 6= 0 and f is not nilpotent.
Set ak = Ann f k := {h ∈ A | hf k = 0}. The set of ideals ak ascends, and
therefore stabilizes. Let an = an+1 . Then (0) = (f n ) ∩ (g).
Indeed, h ∈ (f n ) ∩ (g); hence, h = h1 f n = h2 g; but h1 f n+1 = h2 g f = 0,
implying h1 f n = 0, since an+1 = an . Therefore Y = Y1 ∨ Y2 , where Y1 is
determined by (f n ) and Y2 by (g). u
t
1.7.6. Remarks. a) The support of a primary Noetherian scheme is irre-
ducible. Indeed: The radical of a primary ideal is a prime ideal.
b) The results of sec. 1.7.4 and 1.7.5 show thatW
an affine Noetherian scheme
is a quasi-union of its primary subschemes X = Xi . We could have left in
1≤i≤n
this decomposition only maximal elements and then try to prove its uniqueness
in the same way as we have done for the spaces at the end of sec. 1.4.7b. But
these arguments fail twice: First, the formula
³ _ ´ _
X∩ Yi = (X ∩ Yi ) (1.71)
1≤i≤n 1≤i≤n

is generally false, and, second, as we have established, our primary subschemes

Xi can be reducible themselves. W
Therefore, instead of striking out non-maximal components from Yi ,
1≤i≤n
we should apply less trivial process, and even after that the uniqueness theo-
rem will be harder to formulate and prove.
1.7.7. Incompressible
W primary decompositions. A primary decompo-
sition X = Xi is called incompressible if the following two conditions
1≤i≤n
are satisfied:
(1) supp XW
i 6= supp Xj if i 6= j,
(2) Xk 6⊂ Xi for every k, 1 ≤ k ≤ n.
i6=k

Theorem. Every Noetherian affine scheme X decomposes into an incom-

pressible quasiunion of a finite number of its primary closed subschemes.
W
Proof. Let us start with a primary decomposition X = Xi , see Re-
1≤i≤n
mark 1.7.6 b. Let Yj , where j = 1, . . . , m, be the
W quasiunions ofW the sub-
schemes Xi with common support. Then X = Yj . If Y1 ⊂ Yj , let
1≤i≤n 2≤j≤n
us delete Y1 from the representation of X. Repeating the process we obtain
after a finite number of steps a quasiunion X = ∨Yj which satisfies the second
condition in the definition of incompressibility. It only remains to verify that
the subschemes Yj are primary.
Lemma. The quasiunion of a finite number of primary subschemes with com-
mon support is primary and has the same support.
46 Ch. 1. Affine schemes

W
Proof. Let Y = Yi , supp Yi = supp Yj for all i, j. Let Yi correspond
1≤i≤n T
to an ideal ai in the ring of global functions A on Y . Then ai = (0).
1≤i≤n
Consider a zero divisor f ∈ A. Let f g = 0, where g 6= 0 and g ∈ ai for some
i. Since ai is primary, f n ∈ ai for some n. But since V (ai ) = V (0), it follows
that ai consists of nilpotents and f is nilpotent; moreover, supp Yi = supp Y .
u
t

W
1.7.8. Theorem (A uniqueness theorem). Let X = Xi be an incom-
1≤i≤n
pressible primary decomposition of a Noetherian affine scheme X. The system
of generic points of irreducible closed sets supp Xi does not depend on such
a decomposition.
This system of generic points is denoted by Prime X (or Prime A if
X = Spec A) and is called the set of prime ideals associated with X (or A).
We will establish a more precise result giving an invariant characterization
of Prime X. Let X = Spec A and Xi = Spec(A/ai ) for an ideal ai ⊂ A.
Proposition. The following two statements are equivalent for any reduced
scheme X (for the general scheme, T the statement is only true when Ann f
in b) is replaced by {g ∈ A | gf ∈ ai }:
a) Any prime ideal p ⊂ A corresponds to a generic point of one of the sets
supp Xi .
b) There exists an element f ∈ A such that Ann f := {g ∈ A | f g = 0} is
primary and p is its radical.
Proof. a) =⇒ b). Let pj be the ideal of W a generic point of supp Xj . Clearly,
T
pj = r(aj ). Since the representation X = Xj is incompressible, ai 6⊃ aj ,
1≤j≤n j6=i
T
where 1 ≤ i ≤ n. Let us select an element f ∈ aj \ ai and show that Ann f is
j6=i
primary and pi its radical.
First of all, Ann(f mod ai ) consists only of nilpotents in A/ai ; therefore
Ann f ⊂ pi (since pi is the pre-image of the nilradical of A/aj with respect to the
naturalThomomorphism A −→ A/ai ). Besides, ai ⊂ Ann f because, by construction,
f ai ⊂ aj = (0); therefore r(Ann f ) = pi .
j
Now let us verify that all the zero divisors in A/ Ann f are nilpotents. Assume
the contrary; then there exist elements g, h ∈ A such that gh ∈ Ann f , h 6∈ Ann f
and g is not nilpotent modulo Ann f ; therefore g is not nilpotent modulo ai as well.
On the other hand, f gh = 0; and,
 T since
 g mod ai is not nilpotent, f h mod ai = 0
because ai is primary, i.e., f h ∈ aj ∩ ai = 0, implying h ∈ Ann f contrary to
j6=i
the choice of h. This contradiction shows that Ann f is primary.
b) =⇒ a). Let f ∈ A be an element such that T
Ann f is primary, p its radical.
Set si = (ai : f ) = {g ∈ A | gf ∈ ai }. Since ai = (0), it is easy to see that
T T i
Ann f = si and p = r(Ann f ) = r(si ). If f ∈ ai , then si = r(si ) = A. If, on
i i
1.7 The closed subschemes and the primary decomposition 47

the contrary, f ∈ / ai , then Ann(f mod ai ) consists of nilpotents in A/ai , so that

r(si ) = pi . T
Therefore, p = pi , where I = {i | f 6∈ ai }, implying that p coincides with one
i∈I S
of the pi . Indeed, V (p) = V (pi ) and V (p) is irreducible. Proposition is proved
i∈I
together with the uniqueness theorem 1.7.8. u
t
1.7.9. Incompressible primary W decompositions, cont. An incompress-
ible primary decomposition X = Xi drastically differs from a decomposition
i
of supp X into the union of maximal irreducible components: Though the sup-
ports of the subschemes Xi contain all the irreducible components of supp X
only once, the supports may also have another property: supp Xi ⊂ supp Xj
for some i, j.
A simple example is given by the ring described in Remark 1.4.8. In
notation of Remark 1.4.8, we have (0) = (0, I) ∩ (p, 0) in A; so that
X = X1 ∨ X2 , where supp Xi = supp X and supp X2 = V ((p, 0)). The
space supp X2 is entirely contained in the space supp X1 , whereas the sub-
scheme X2 is distinguished from the “background” by its nilpotents (see
Fig. 10), where X1 = Spec A and X2 = Spec B[T ]/(T 2 ), where B = A/p, and
X = Spec R[T1 , T2 ] ⊕ a, where a = R[T1 , T2 ]/(T1 ). Clearly, a2 = (0).

X2

X1

Fig. 10

This remark on nilpotents is of general character. Indeed, let in the incom-

pressible decomposition, supp Xi ⊂ supp Xj and Xi 6⊂ Xj . Then
supp(Xi ∩ Xj ) = supp Xi , but Xi ∩ Xj is a proper subscheme in Xi .
This may only happen when there are more nilpotents in the ring of Xi then
in the ring of Xi ∩ Xj where they are induced by nilpotents from the greater
space Xj .
Among the components Xi of the incompressible primary decomposition,
those for which supp Xi is maximal are called isolated,, the other ones are
called embedded. The same terminology is applied to the sets supp Xi them-
selves and their generic points, which constitute the set Prime X.
The space of an embedded component may belong simultaneously to sev-
eral (isolated or embedded) components. Besides, the chain of components
subsequently embedded into each other may be however long.
48 Ch. 1. Affine schemes

Therefore, innocent at the first glance, the space of an affine scheme may
hide in its depth a complicated structure of embedded primary subschemes
like the one illustrated on Fig. 11. The reader should become used to the
geometric reality of such a structure.

Fig. 11

Clearly, depicting nilpotents by arrows does not make possible to reflect the
details however precisely. It is only obvious that on the embedded components
the nilpotents grow thicker, thus giving away their presence.
1.7.10. Incompressible primary decompositions, cont. 2. The finite set
of prime ideals Spec A which is invariantly connected with every Noetherian ring
A has a number of important properties. In particular, it enables us to refine The-
orem 1.4.8
Theorem. An element f ∈ A is a zero divisor if and only if it vanishes
(as a function) on one of the components of the incompressible primary decom-
position of Spec A. T
In other words, the set of zero divisors of A coincides with p.
p∈Prime A
T
Proof. First, let us show that if f ∈
/ p , then Ann f = (0).
p∈Prime A
S
Let (0) = ai be the incompressible primary decomposition, where pi = r(ai ).
i
Let f g = 0. Since f ∈ / pi , it follows that ai being primary implies that g ∈ ai . This
is true for all i; therefore g = 0.
Conversely, let Ann(f ) = (0). Suppose that f ∈ pi ; since A is Noetherian, it
follows that f k ∈ pki ⊂ ai for some k ≥ 1; i.e., (ai : (f k )) = A. 15) On the other hand,
Ann(f k ) = (0) implying
\ \ \
(0) = Ann(f k ) = (aj : (f k )) = (aj : (f k )) ⊃ aj . (1.72)
j j6=i j6=i

15
The quotient (a : b) of two ideals in A is defined to be the set {x ∈ A | xb ⊂ a}. It
is easy to see that (a : b) is an ideal in A. In particular, the ideal (0 : b) is called
the annihilator of b and is denoted Ann(b).
1.8 Hilbert’s Nullstellensatz (Theorem on zeroes) 49

This contradicts to the fact of incompressibility of the decomposition and completes

the proof. u
t
1.7.11. Incompressible primary decompositions for Noetherian schemes.
Finally, notice that the uniqueness theorem 1.7.8 only concerns the supports of pri-
mary components of incompressible decompositions, not the components themselves.
About them one may only claim the following.
Theorem. The set of isolated components of an incompressible primary decompo-
sition of a Noetherian scheme Spec A does not depend on the choice of the decom-
position.
For embedded components, this statement is false.
We skip the proof; the reader may find it in [ZS] (v. 1, Theorem 4.5.8) or [Lang].

1.7.12. Exercises. 1) Let K be an algebraically closed field.

a) Describe all primary closed subschemes of the line Spec K[T ].
b) Same for non-closed fields.
c) Same for Spec Z.
2) Describe up to an isomorphism primary closed subschemes of the plane
Spec K[T1 , T2 ] supported at V (T1 , T2 ) and whose local rings are of length ≤ 3.

1.8. Hilbert’s Nullstellensatz (Theorem on zeroes)

In this section we will establish that the closed subschemes of finite di-
mensional affine spaces over a field or a ring Z have many closed points.
1.8.1. Theorem. Let A be a ring of finite type. Then Spm A is dense in
Spec A.
Corollary. The intersection X ∩ Spm A is dense in X for any open or closed
subset X ⊂ Spec A.
Indeed, if X = V (E) and we identify X with Spec A/(E), then
Spm A ∩ V (E) coincides with Spm A/V (E) and A/(E) is a ring of finite type
together with A. This easily implies the statement for open sets, too. u
t
The space Spm A is easier to visualize since it has no nonclosed points (the
“big open sets” still remain nevertheless). On the other hand, Corollary 1.8.1
implies that for rings of finite type the space of Spec A is uniquely recovered
from Spm A (assuming that the induced topology in Spm A is given).
The recipe is to use the following statements:
1) The points of Spec A are in a one-to-one correspondence with the irre-
ducible closed subsets of Spm A. (Therefore to every irreducible closed subset
of Spm A, its “generic point” in Spec A corresponds.)
2) Every closed subset of Spec A consists of generic points of all the irre-
ducible closed subsets of a closed subset of Spm A.
We advise the reader to prove these statements in order to understand
them :-)
50 Ch. 1. Affine schemes

Proof of Theorem. We will successively widen the class of rings for which this
theorem is true.
a) Let K be algebraically closed. Then the set of closed points of Spec K[T1 , . . . , Tn ]
is dense.
The closure of the set of closed points coincides with the space of zeroes of all the
functions which vanish at all the closed points, and therefore it suffices to prove that
a polynomial F , which belongs to all the maximal ideals of K[T1 , . . . , Tn ], is iden-
tically zero. But such a polynomial satisfies F (t1 , . . . , tn ) = 0 for all t1 , . . . , tn ∈ K,
and an easy induction on n shows that F = 0 (here we actually use not closedness
even, but only the fact that K is infinite).
b) The same as in a) but K is not supposed to be algebraically closed.
Denote by K the algebraic closure of K. We have a natural morphism

i : A = K[T1 , . . . , Tn ] ,→ K[T1 , . . . , Tn ] = B. (1.73)

The ring B is integer over K, and therefore thanks to the results of 7.4–7.5, we have

Spm A = ai(Spm B) = ai(Spec B) = Spec A. (1.74)

c) Theorem 1.8.1 holds for the rings A of finite type without zero divisors over K.
Indeed, by Noether’s normalization theorem there exists a polynomial subalgebra
B ⊂ A such that A is integer over B. By the already proved Spm B = Spec B, and
the literally same argument as in b) shows that Spm A = Spec A.
d) Theorem 1.8.1 holds for any rings A of finite type over a field.
Indeed, any irreducible component of Spec A is homomorphic to Spec A/p, where
p is a prime ideal. The ring A/p satisfies the conditions of c), hence, the closed points
are dense in all the irreducible components of Spec A, and therefore in the whole
space.
e) The same as in d) for the rings A of finite type over Z.
1.8.2. Lemma. No field of characteristic 0 can be a finite type algebra over Z.
Proof. If a field of characteristic 0 is a finite type algebra over Z, then thanks
to Noether’s normalization theorem there exist algebraically independent over Q
element t1 , . . . , tr ∈ K such that K is integral over R = Q[t1 , . . . , tr ]. By Proposi-
tion 1.6.6 the natural map Spm K −→ Spm Q[t1 , . . . , tr ] is surjective; since Spec K
contains only one closed point, so does Spec R, which is only possible if R = Q.
Therefore, K is integral over Q, and hence is a finite extension of the field Q.
Let x1 , . . . , xn generate K as a Z-algebra. Each of the xj is a root of a polynomial
with rational coefficients. If N is the LCM of the denominators of these coefficients,
then, as is easy to see, all the N xj are integral over Z; if y is the product of m of
the generators x1 , . . . , xn (perhaps, with multiplicities), then N m y is integral over
Z. Since all elements of K are linear combinations of such products with integer
coefficients, it follows that for any y ∈ K, there exists a natural m such that N m y
is integral over Z.
Now, let p be a prime not divisor of N . Since 1/p ∈ Q ⊂ K, we see that a
non-integer rational number N m /p is integral over Z; Exercise 1.6.7 shows that this
is impossible. u
t
To complete the proof in case e), denote by ϕ : Z → A a natural homomor-
phism and show that a ϕ(Spm A) ⊂ Spm Z. Indeed, otherwise there exists a maximal
idealp ⊂ A such that ϕ−1 (p) = (0); so Z can be embedded into the field A/p (and
1.8 Hilbert’s Nullstellensatz (Theorem on zeroes) 51

hence this field is of characteristic 0) which leads to a contradiction with Lemma

1.8.2. S
Thus, Spm A = V (p), where p runs over primes. Since V (p) is homeomorphic
p
to Spec Z/p which is an algebra of finite type over A/pA, the set of closed points
of V (p) is dense in Spec A/pA; hence, so is it in Spec A. u
t
1.8.3. Proposition. Let p ⊂ A be a maximal ideal of a ring of finite type over Z
(resp. a field K). Then A/p is a finite field (resp. a finite algebraic extension of K).
Proof. It follows from step e) of the proof of Theorem 1.8.1 that it suffices to
confine ourselves to the case of a ring A over a field K. The quotient ring modulo
the maximal ideal A/p, being a field, contains the unique maximal ideal. On the
other hand, by Noether’s normalization theorem A is an integer extension of the
polynomial ring B in n indeterminates over K. The case n ≥ 1 is impossible, since
then B, and therefore A, would have had infinitely many maximal ideals. Therefore
n = 0, and A is an integer extension of finite type of K. This proves the statement
desired. u
t
1.8.4. Hilbert’s Nullstellensatz. Now consider the case of K algebraically
closed. By Proposition 1.8.3 in this case the closed points of Spec A are in
one-to-one correspondence with the K-points of the scheme Spec A; the space
of the latter ones is called an affine algebraic variety over K in the classical
sense of the word. The discussion of sec. 1.8.1 shows that in this case Spec A
with the spectral topology and the space of geometric K-points of the scheme
Spec A with Zariski topology are essentially equivalent notions: The passage
from one to another does not require any additional data.
Finally, let us give the classical formulation of Hilbert’s Nullstellensatz in
the language of systems of equations.
Theorem (Hilbert’s Nullstellensatz). Let K be algebraically closed and
F = (Fi )i∈I , where Fi ∈ K[T ] for T = (T1 , . . . , Tn ] is a family of polyno-
mials.
a) The system
P Fi = 0 for i ∈ I, has a solution in K if and only if the
equation 1 = Fi Xi has no solutions in K[T1 , . . . , Tn ], i.e., the ideal (F )
i
does not coincide with the whole ring.
b) If a polynomial G ∈ K[T1 , . . . , P
Tn ] vanishes on all the solutions of the
system Fi = 0 for i ∈ I, then Gn = Fi Gi , where Gi ∈ K[T1 , . . . , Tn ], for
i
some positive integer n.
Proof. a) If (F ) does not coincide with the whole ring, then by Theorem 1.2.3
we have Spec K[T1 , . . . , Tn ]/(F ) 6= ∅. Therefore the spectrum contains a maxi-
mal ideal the residue field modulo which thanks to Proposition 1.8.3 coincides
with K, and the images of Ti in this field give a solution of the system Fi = 0
for i ∈ I.
b) If G vanishes at all the solutions of the system, then the image of G in
K[T1 , . . . , Tn ]/(F ) belongs to the intersection of all the maximal ideals of this
ring, hence, by Theorem 1.7.10, to the intersection of all the prime ideals.
Therefore G is nilpotent thanks to Theorem 1.4.7. u
t
52 Ch. 1. Affine schemes

1.9. The fiber products

This section does not contain serious theorems. We only give a construction
of the fiber product of affine schemes. This notion, though a simple one, is one
of the most fundamental and explains the popularity of the tensor product
in modern commutative algebra. Our main aim is to connect with the fiber
products geometrically intuitive images.
We advise the reader to refresh the knowledge of categories (see the last
section of this Chapter) before reading this section.
1.9.1. Fiber product. Let C be a category, S ∈ Ob C, and CS the cate-
gory of “objects over S”. The fiber product of two objects ψ : Y −→ S and
ϕ : X −→ S over S is their product in CS .
In other words, the fiber product is a triple (Z, π1 , π2 ), where Z ∈ Ob C,
π1 : Z −→ X, π2 : Z −→ Y , such that
a) the diagram
π1
Z /X
π2 ϕ (1.75)
² ψ ²
Y /S

where (Z, ϕ π1 ) = (Z, ψπ2 ) ∈ Ob CS and π1 , π2 ∈ Mor CS ), commutes;

b) for any object X : Z 0 −→ S (in what follows it will be denoted just
by Z 0 ) of CS , the maps induced by the morphisms π1 , π2 identify the set
HomCS (Z 0 , Z)with HomCS (Z 0 , X) × HomCS (Z 0 , Y ).
In still other words, (Z, π1 , π2 ) is a universal object in the class of all the
triples that make the above diagram commuting.
The diagram (1.75) with the properties a) and b) is sometimes called
a Cartesian square. An object Z in it is usually denoted by X × Y and this
S
object is said to be the fiber product of X and Y over S. Using this short
notation one should not forget that in it the indication to the four morphisms
X −→ S, Y −→ S, X × Y −→ X, and X × Y −→ Y , of which the first two
S S
are vital, is omitted.
Speaking formally, the usual — set-theoretical — direct product is not
a particular case of a fiber one; but this is so if the category C has a final
object F . Then X × Y is actually the same as X × Y .
F
The fiber product exists in Sets; we will elucidate its meaning with several
examples.
Lemma. Let ϕ : X −→ S and ψ : Y −→ S be some maps of sets; put
Z = {(x, y) ∈ X × Y | ϕ(x) = ψ(y)} ⊂ X × Y (1.76)
and define π1 : Z −→ X and π2 : Z −→ Y as maps induced by the projections
X × Y −→ X, X × Y −→ Y . Then the triple (Z, π1 , π2 ) forms a fiber product
of X and Y over S.
1.9 The fiber products 53

Proof is absolutely trivial. u

t
This construction explains wherefrom stems the name of our operation:
Over every point of S the fiber of the map Z −→ S is the direct product
of the fibers X and Y .
1.9.2. Examples. The following notions are encountered everywhere.
Product. In Sets, the one-point set ∗ is a final object, and so

X × Y = X × Y for any X, Y ∈ Sets.

∗

Intersection. Let ϕ, ψ be embeddings of X and Y , as subsets, into S.

Then identifying Z = X × Y with a subset of S we see that Z = X ∩ Y .
S
The fiber of a map. Let Y = F be a final object, ψ(F ) = s ∈ S. Then
Z = ϕ−1 (s). More generally, if ψ is an embedding, then having identified Y
with ψ(Y ) ⊂ S we obtain Z = ϕ−1 (Y ).
The change of base. This terminology is induced by topology: If
ϕ : X −→ S is a bundle (in any sense) and ψ : S 0 −→ S a morphism of to-
pological spaces, then the bundle X 0 = X × S 0 is said to be obtained from
S
X −→ S by the change of base S to S 0 . The other name for it is the induced
bundle. u
t
1.9.3. Fiber products exist. The above examples will serve us as a model
for the corresponding notions in the category of schemes (affine ones, for a
time being). First of all, let us establish their existence.
Theorem. Let X = Spec A, Y = Spec B, S = Spec C, where A and B are
C-algebras. The fiber
N product of X and Y over S exists and is represented by
the triple (Spec A B, π1 , π2 ), where π1 (resp. π2 ) is the map induced by the
C N N
C-algebra homomorphism A −→ B B, f 7→ f ⊗ 1 (resp. B −→ A B,
C C
g 7→ 1 ⊗ g).
For proof, see [Lang], where the fact that the fiber coproducts in the cate-
gory Rings exist, and are described just as stated, is established. The inversion
of arrows gives the statement desired. u
t
Observe that the category of affine schemes has a final N object, Spec Z. So
we may speak about the absolute product X × Y = Spec A B.
Z
1.9.4. Warning. The statement “the set of points |X × Y | of the scheme
S
X × Y is the fiber product |X| × |Y | of the sets of points |X| and |Y | over
S S
|S|” (where |X| is just the set of points of X, not the cardinality of this set) is
only true for the points with values in C-algebras; i.e., when S = Spec C or,
which is the same, for the sets of morphisms over S. Here are typical examples
showing what might happen otherwise.
54 Ch. 1. Affine schemes

Examples. 1) Let K be a field, S = Spec K, let X = Spec K[T1 ], and

Y = Spec K[T2 ] with obvious morphisms between these spectra. Then X × Y
S
is the plane over K; it has a lot of non-closed points: Generic points of ir-
reducible curves non-parallel to the axes, which are not representable by the
pairs (x, y), where x ∈ X, y ∈ Y .
2) Let L ⊃ K be fields.
2a) Let L B K be a finite Galois field extension.
N Let X = Spec L,
S = Spec K. Let us describe X × X or, dually, L L.
S K
Let us represent the second factor L in the form L = K[T ]/(F (T )), where
F (T ) is an irreducible polynomial. In other words, in L, take a primitive
element t = T mod (F ) over K.
It follows from the definition of tensor product that in this case, as L-al-
gebras, O
L L ' L[T ]/(F (T )) (1.77)
K
N
if we assume that the L-algebra structure on L L is determined by the map
K
l 7→ l ⊗ 1.QBut, by assumption, F (T ) factorizes in L[T ] into linear factors
F (T ) = (T − ti ), where ti are all the elements conjugate to t over K and
1≤i≤n
n = [L : K].
By the general theorem on the structure of modules over principal ideal
rings [Lang], we obtain
O Y Y
L L ' L[T ]/( (T − ti )) ' L[T ]/(T − ti ) ' Ln . (1.78)
K
N `
In particular, Spec L L' Spec L: Though Spec L consists of only one
K 1≤i≤n N
point, there are, miraculously, n of them in Spec L L.
K
2b) The trouble of another nature may happen if for L we take a purely
inseparable extension of K. Let, e.g., F (T ) = T p − g, where g ∈ K \ K p and
p = Char K (recall that here: K p := {xp | x ∈ K}). Then in L[T ] we have
T p − g = (T − t)p , where t = g 1/p , so that
O
L L ' L[T ]/((T − t)p ) ' L[T ]/(T p ); (1.79)
K

therefore we
N have acquired nilpotents which were previously lacking. The space
of Spec L L consists, however, of one point. ut
K

1.9.5. Examples. Let us give examples which are absolutely parallel to the
set-theoretical construction.
i1 i2
1) Let X = Spec A, let Y1 −→ X and Y2 −→ X be two closed subschemes
of X determined by ideals a1 , a2 ⊂ A. Thanks to results of sec. 1.7.2 their
1.10 The vector bundles and projective modules 55

intersection Y1 ∩ Y2 represents the functor Y1 (Z) ∩ Y2 (Z), i.e., should coincide

with their fiber product over X. This is indeed the case: The corresponding
statement on rings is
O
A/(a1 + a2 ) ' A/a1 A/a2 (1.80)
A

and is easy to verify directly.

2) Let Spec B = Y −→ X = Spec A be an affine scheme morphism, k(x)
the algebraic closure of the field of fractions corresponding to a given point
x ∈ X. The natural homomorphism A −→ k(x) represents a geometric point
with center at x. The fiber product Yx = Y × Spec k(x) is said to be the
X
geometric fiber of Y over x, and Y × Spec k(x) the usual fiber.
X
2a) A particular case: Spec A/pA is the fiber of Spec A over (p) ∈ Spec Z
for any prime ideal p in any ring A.
1.9.6. The diagonal. Let Spec B = X −→ S = Spec A be an affine scheme
morphism. The commuting diagram

id /X
X
id (1.81)
² ²
X /S

defines a morphism δ : X −→ X × X (see sec. 1.9.1b) which is said to be the

S
diagonal one.
Proposition. δ identifies X with the closed subscheme ∆X of X × X singled
S
out by the ideal
O
I∆X = Ker(µ : B B −→ B), where µ(b1 ⊗ b2 ) = b1 b2 . (1.82)
A

Proof. Writing down all the necessary diagrams we see that δ = aµ. Since
µ is surjective, its kernel determines a closed subscheme isomorphic to the
image of δ. u
t
The scheme ∆X is said to be the (relative over S) diagonal.

1.10. The vector bundles and projective modules

1.10.1. Families of vector spaces. Let ψ : Y −→ X be an affine scheme
morphism, X = Spec A, Y = Spec B; let ϕ : A −→ B be the corresponding
ring homomorphism. We would like to single out a class of morphisms which
is similar to locally trivial vector bundles in topology.
56 Ch. 1. Affine schemes

It is convenient to start from a wider notion of “families of vector spaces”.

Example 2.1 shows that an analogue of a vector space V over a field K is
. .
given by the scheme Spec SK (V ∗ ), where V ∗ = HomK (V, K) and SK (V ∗ ) is
the symmetric algebra of V ∗ . Replacing here K by an arbitrary ring A and V
by an A-module M we get the following definition.
Under the above notation, let χ : M −→ B an A-module morphism. Sup-
. ∼
pose that χ induces an A-algebra isomorphism SA (M )−→B. Then the pair
(χ, ψ) is said to be a family of vector spaces over X = Spec A and M is said
to be the module that defines the family.
In other words, an explicit structure of the symmetric algebra over A on B
determines a fiber-wise linearization of the morphism χ.
The morphisms of families of vector spaces over a fixed base are obviously
defined. The category of such families is dual to the category of A-modules;
so, in particular, every family of vector spaces is determined by its module M
up to an isomorphism.
The following notion is more important for us for the time being.
1.10.2. Families of vector spaces exist. Under the above notation, let
there be given a ring homomorphism A −→ A0 which defines a scheme mor-
phism
X 0 = Spec A0 −→ X = Spec A. (1.83)
Consider a family of vector spaces (χ0 , ψ 0 ), where

χ0 = id ⊗χ : M 0 = A0 ⊗A M −→ A0 ⊗A B (1.84)

and ψ 0 is a morphism Y 0 = Spec A0 ⊗A B −→ X 0 . This family is said to be

induced by the change of base X 0 by X.
We see that X 0 is indeed a family of vector spaces, since there is a canonical
isomorphism
. ∼ .
SA0 (A0 ⊗A M )−→A0 ⊗A SA (M ). (1.85)
In particular, if A0 is a field, Y 0 is the scheme of the vector space (A0 ⊗A M )∗
over A0 . This means that all the fibers of the family ψ : Y −→ X over geometric
points are vector spaces, which justifies the name “the scheme of the vector
space”. The dimensions of fibers can, clearly, jump.
Note also that thanks to the isomorphism (1.85) the scheme Y 0 is identified
with the fiber product X 0 × Y , so that our operation of the change of base is
X
an exact analogue of the topological one.
1.10.3. Vector bundles. A family of vector spaces χ : Y −→ X is said to
be trivial if its defining A-module is free.
The families of vector spaces trivial in a neighborhood of any point are said
to be vector bundles. It is not quite clear though how to define the property
of local triviality: The neighborhoods of points in Spec A are just topological
spaces, not schemes. Here we first encounter the problem that will be system-
atically investigated in the next Chapter. For a time being, it is natural to
adopt the following preliminary definition.
1.10 The vector bundles and projective modules 57

A family of vector spaces χ : Y −→ X is said to be trivial at x ∈ X if there

exists an open neighborhood U 3 x such that, for any morphism ψ : X 0 −→ X
with ψ(X 0 ) ⊂ U , the induced family Y 0 −→ X 0 is trivial.
We will now replace this condition of triviality by another one which is
easier to verify. First of all, since the big open sets D(f ) constitute a basis
of the topology of Spec A, it suffices to consider neighborhoods of the form
D(f ). They possess the following remarkable property.
Proposition. Let A be a ring, f ∈ A not a nilpotent. Set Xf := Spec Af
and denote by i∗ : Xf −→ X the morphism induced by the homomorphism
i : A −→ Af , g 7→ g/1. Then
a) i∗ determines a homeomorphism Xf ' D(f )
b) For any morphism ψ : X 0 −→ X such that ψ(X 0 ) ⊂ D(f ), there exists
a unique morphism χ : X 0 −→ Xf such that the diagram

ψ
X0 B /X
BB }}>
BB }} (1.86)
B
χ BB }} ∗
! }} i
Xf

commutes.
This implies that the family of vector spaces Y −→ X is trivial at
x ∈ Spec A = X if and only if, for any element f ∈ A such that f (x) 6= 0, the
family induced over Xf is trivial. Translating this into the language of modules
we find a simple condition which will be used in what follows.
Corollary. An A-module M determines a family of vector spaces trivial at
x ∈ Spec A if and only if there exists f ∈ A such that f (x) 6= 0 and such that
the Af -module Mf := Af ⊗A M is free.
A module M satisfying conditions of Corollary for all the points x ∈ Spec A
is said to be locally free.
Proof of Proposition. a) This is a particular case of Theorem 1.6.4d
b) This statement expresses the known universal property of the rings
of quotients. Indeed, let ψ : A −→ A0 be a ring homomorphism such that
a
ψ(Spec A0 ) ⊂ D(f ). This means that f does not belong to any of the ideals
ψ −1 (p), where p ∈ Spec A0 , i.e., ψ(f ) does not vanish on Spec A0 . Therefore
ψ(f ) is invertible in A0 .
In the category of such A-algebras, the morphism A −→ Af is universal
object (see 1.6.4b) which proves the statement desired.
In particular, if D(f ) = D(g), then the ring Af is canonically isomorphic
to Ag . u
t
58 Ch. 1. Affine schemes

1.10.4. Main definition and main result of this section. A vector

bundle over a scheme X = Spec A is a family of vector spaces locally trivial
at each point of Spec A.
Unless otherwise stated, until the end of this section we will only consider
Noetherian rings and modules.
Recall that a module M is said to be projective if it is isomorphic to a direct
summand of a free module.
Theorem. An A-module M determines a vector bundle over Spec A if and
only if it is projective.
The theorem claims that the class of locally free modules coincides with
the class of projective modules. This is just the statement we will prove; first
the inclusion in one direction and then into the opposite one. We will have
to perform rather hard job and we will use the opportunity and establish
meanwhile more auxiliary results than is actually needed: They will serve us
later.
1.10.5. Localizations of modules. Let S ⊂ A be a multiplicative system
not containing 0, and M an A-module. Set MS = AS ⊗A M . Though we
only need here the information on Af := AS for S = {f n | n ∈ Z+ } and
Mf := MS for the same S, it is not a problem to extend the result to general
multiplicative system S.
1.10.5a. Lemma. The equality m/s = 0 holds if and only if there exists
t ∈ S such that tm = 0. In particular, the kernel of the natural homomorphism

M −→ MS , m 7→ m/1, (1.87)

consists of the elements m such that (Ann M ) ∩ S 6= ∅.

Proof. Clearly, tm = 0 =⇒ tm/ts = 0 = m/s. To prove the converse impli-
cation, consider first a particular case.
a) M is free. Let {mi }i∈I be aPfree A-basis of M . Then {mi = mi /1}i∈I is
a free AS -basis of MS . Let m = fi mi , where fi ∈ A, be an element of M .
If m/s = 0, thenQfi /s = 0 for all i, and hence there exist ti ∈ S such that
ti fi = 0. Set t = ti , where i runs over a finite set of indices I0⊂I for which
fi 6= 0. Clearly, tm = 0 since tfi = 0 for all i ∈ I0 .
b) The general case. There exists an exact sequence
ϕ ψ
F1 −→F0 −→M −→ 0, (1.88)

where F0 , F1 are free modules. Tensoring the sequence by AS we get the exact
sequence
ϕS ψS
(F1 )S −→(F0 )S −→MS −→ 0 (1.89)
(see [Lang]). Here we set ϕS = idAS ⊗A ϕ, and so on.
Let m/s = 0 for m = ψ(n), where n ∈ F0 . Then ψS (n/s) = 0; this implies
1.10 The vector bundles and projective modules 59

n/s = ϕS (l/t) = ϕ(l)/t, where l ∈ F1 and t ∈ S. (1.90)

In other words, (tn − sϕ(l))/st = 0 in (F0 )S . Since F0 is free, there exists

r ∈ S such that rtn = rsϕ(l) in F0 . Applying ψ to this relation we get

rtm = ψ(rtn) = rsψ ◦ ϕ(l) = 0, (1.91)

as desired. u
t
Observe that we never used the Noetherian property.
1.10.5b. Corollary. Let M be a Noetherian A-module, f ∈ A. There exists
an integer q > 0 such that f q m = 0 for all m ∈ Ker(M −→ Mf ).
Proof. Select the needed value qi for every of a finite number of generators
of the kernel, and set q = max qi . u
t
i
1.10.6. Tensoring exact sequences. In the proof of Lemma 1.10.5 we
have used the following general property of the tensor product:
Tensoring sends short exact sequences into the sequences exact
(1.92)
everywhere except the leftmost term.
Tensoring by AS , however, possesses a stronger property: It completely pre-
serves exactness; this means AS is what is called a flat A-algebra.
ϕS ψS
1.10.6a. Proposition. The sequence MS −→NS −→PS of AS -modules is ex-
ϕ ψ
act for any exact sequence M −→N −→P of A-modules.
Proof. ψ ◦ ϕ = 0 =⇒ ψS ◦ ϕS = 0 =⇒ Ker ϕS ⊃ Im ϕS .
Conversely, let n/s ∈ Ker ψS ; then ψ(n/s) = 0 implying, thanks to the
above, tψ(n) = 0 for some t ∈ S. Therefore tn = ϕ(m) implying

n/s = tn/ts = ϕ(m)/ts = ϕS (m/ts), (1.93)

as desired. u
t
1.10.7. Lifts of Af -module homomorphisms. Let ϕ : M −→ N be
an A-module homomorphism. For any f ∈ A, we have an induced Af -module
homomorphism ϕf : Mf −→ Nf . We will say that ψ : Mf −→ Nf can be lifted
to ϕ : M −→ N if ϕf = ψ .
Lemma. Let F be a free Noetherian A-module, M a Noetherian A-module,
f ∈ A, Mf a free A-module. Then, for any homomorphism ϕ : Mf −→ Ff ,
there exists an integer q such that the homomorphism f q ϕ : Mf −→ Ff can
be lifted to a homomorphism M −→ F .
Proof. First of all, Ff is free and has a finite number of generators, so that
ϕ is given by a finite number of coordinate Af -morphisms Mf −→ Af . If
being multiplied by an appropriate power of f each of them can be lifted to
60 Ch. 1. Affine schemes

a morphism M −→ A, then so does ϕ. Therefore we can and will assume that

F = A.
Let mi , where i = 1, . . . , n, be a system of generators of M . Multiplying
ϕ by an appropriate power of f , we may assume that ϕ(mi ) = gi /1, where
gi ∈ A for all i.
It is tempting to lift ϕ : Mf −→ Af to a homomorphism ψ : M −→ A by
setting ψ(mi )P = gi . This, however, might P prove to be impossible since there
are
P relations fi m i = 0 such that fi gi 6= 0. But we have the equality
fi (gi /1) = 0, and therefore the set
½X X ¾
fi g i | fi mi = 0 (1.94)

constitutes, thanks to Lemma 1.10.5, a Noetherian A-submodule, which be-

longs to Ker(A −→ Af ). By Corollary 1.10.5 this submodule is annihilated by
f q for some q. This implies that there exists a homomorphism f q ψ : M −→ A
such that f q ψ(mi ) = f q gi since (f q ψ)f = f q ϕ. u
t
1.10.8. Locally free modules are projective. Now, we can establish the
“if” part of Theorem 1.10.4.
Proposition. Locally free modules are projective.
Proof. Let M be a Noetherian locally free A-module, ψ : F −→ M an epi-
morphism, where F is a Noetherian free module. In order to prove that M is
a direct summand of F , we have to find a homomorphism ϕ : M −→ F such
that ψ ◦ ϕ = idM . More generally, let
P = {χ ∈ HomA (M, M ) | χ = ψ ◦ ϕ; for some ϕ ∈ HomA (M, F )}. (1.95)
First, let us show that, for every point x ∈ Spec A, there exists f ∈ A such
that f (x) 6= 0 and f q idM ∈ P for some q ≥ 0.
Select f so that Mf is Af -free. Then the epimorphism ψf : Ff −→ Mf
has a section ϕ : Mf −→ Ff . By Lemma 1.10.7 we can lift f r ϕ to a homo-
morphism χ : M −→ F for some r ≥ 0. Since ψf ◦ ϕ = idMf , this implies
(ψ ◦ χ)f = f r idMf ; in particular,
(ψ ◦ χ − f r idM )f (mi ) = 0 (1.96)
t r
for a finite number of generators mi of M . Therefore f (ψ ◦ χ − f idM ) = 0
for some t ≥ 0; hence f r+t idM = ψ ◦ f t χ ∈ P .
SNow, from a cover of Spec A with big open sets, select a finite subcovering
D(fi ) which is possible because Spec A is quasi-compact. Find q for
1≤i≤k
which fiq idM ∈ P for all i. Since D(fiq ) P
= D(fi ), it follows that {fiq }1≤i≤k
generates (1.75). The partition of unity gi fiq = 1 implies that
1≤i≤k
 
X
idM =  gi fiq  idM ∈ P. t
u (1.97)
1≤i≤k
1.10 The vector bundles and projective modules 61

1.10.9. Nakayama’s lemma. Now, we would like to establish that projec-

tive modules are locally free. First, we verify this for a stronger localization
procedure.
The following simple but fundamental result is called Nakayama’s lemma.
Lemma (Nakayama’s lemma). Let A be a local ring, a ⊂ A an ideal not equal
to A, and M an A-module of finite type. If M = aM , then M = {0}.
Examples illustrating the necessity of finiteness condition.
a) Let A be a ring without zero divisors, M the field of quotients. Obvi-
ously, if a 6= {0}, then aM = M , but M 6= {0}.
b) Let A be the ring of germs of C ∞ -functions in a vicinity of the T
origin
of R and a the ideal of functions that vanish at the origin. Let M = an
n∈Z+
be the ideal of flat functions, i.e., the functions that vanish at the origin
together with all the derivatives. It is not difficult to establish that aM = M :
This follows from the fact that, for any flat function f and the coordinate
function x, the quotient f /x whose value at the origin is defined as zero is a
flat function.
Proof of Nakayama’s lemma. Let M 6= {0}. Select a minimal finite
system of
P generators m1 , . . . , m2 of M . Since M P= aM , it follows that
m1 = fi mi for fi ∈ a, i.e., (1 − f1 )m1 = fi mi . Since f1 lies in
1≤i≤r i≥2
a maximal ideal of A, it follows that 1 − f1 is invertible; therefore m1 can
be linearly expressed in terms of m2 , . . . , mr . This contradicts to minimality
of the system of generators. u
t
1.10.9a. Corollary. Let M be the module of finite type over a local ring
A with a maximal ideal p. Let the elements mi = mi (mod pM ), where
i = 1, . . . , r, generate M/pM as a linear space over the field A/p, then the mi
generate the A-module M . In particular, if A is Noetherian, the generators
of the A/p-space p/p2 generate the ideal p.
Proof. Let M 0 = M/(Am1 + . . . + Amr ). Since M = pM + Am1 + . . . + Amr ,
we see that M 0 = pM 0 , implying M 0 = 0. u
t
1.10.10. Proposition. A projective module M of finite type over a local
ring A is free.
Proof. Let p be a maximal ideal in A. Then M/pM is a finite-dimensional
space over A/p; let mi = mi (mod pM ), where i = 1, . . . , r, be its basis. By
the above, the mi constitute a system of generators of M . Let us show that
M is free. Consider an epimorphism F −→ M −→ 0, where F = Ar is a free
module of rank r whose free generators are mapped into {mi }ri=1 . Since M is
projective, there exists a section ϕ : M −→ F which induces an isomorphism
ϕ : M/pM −→ F/pF because the dimensions of both the spaces are equal to r.
Therefore either F = ϕ(M ) + pF or F/ϕ(M ) = p(F/ϕ(M )). By Nakayama’s
lemma F = ϕ(M ); hence ϕ is an isomorphism. u
t
62 Ch. 1. Affine schemes

1.10.11. Proof of Theorem 1.10.4, completion.

Proposition. Any Noetherian projective module M over any Noetherian ring
A is locally free.
Proof. Let x ∈ Spec A and p ⊂ A the corresponding prime ideal, the module
Mp = Ap ⊗ M is projective, and therefore free thanks to Proposition 1.10.10.
Take its Ap -basis. Reducing the elements of the basis to the common denom-
inator we may assume that they are of the form mi /g, where mi ∈ M for
i = 1, . . . , n and g ∈ A. Consider a homomorphism ϕ : Ang −→ Mg sending
the elements of a free basis of An(g) into mi /g; set K = Ker ϕ and C = Coker ϕ.
Tensoring the exact sequence of A(g) -modules

0 −→ K −→ Ang −→ Mg −→ C −→ 0 (1.98)

by Ap , which is also the localization of Ag modulo Ag \ pg , we get thanks

to Theorem 1.10.6 an exact sequence of Ap -modules. Its middle arrow is an iso-
morphism, and therefore Ap ⊗Ag K = 0 and Ap ⊗Ag C = 0.
Let k2 , . . . , ks and c1 , . . . , cr be bases of the Ag -modules K and C, respec-
tively. By Lemma 1.10.9 there exist hi , h0j ∈ Ag \ pg such that hi ki = 0 and
h0j cj = 0 for all i, j. In particular,
Y Y
h= hi h0j ∈ Ag \ pg (1.99)
1≤I≤r 1≤j≤s

and h annihilates K and C. Let h = f /g k , where f ∈ A \ p. Then f /1

annihilates both K and C. Tensoring (1.98) by (Ag )f /1 over Ag , and with
the Af g -module isomorphism (Mg )f /1 ' Mf g taken into account, we see that
there exists an isomorphism Anfg ' Mf g because Kf /1 = {0} and Cf /1 = {0}.
Since f g(x) 6= 0, we see that M is locally free at x. u
t
1.10.12. An example of a non-free projective module. Let A be a ring
of real-valued continuous functions on [0, 1] such that f (0) = f (1), i.e., A may
be viewed as a ring of functions on the circle S 1 . The module of sections of the
Möbius bundle over S 1 may be described as the A-module M of functions on
[0, 1] such that f (0) = −f (1).
Theorem. M is not free, but M ⊕ M ∼
= A ⊕ A.
Proof. 1) For any f1 , f2 ∈ M , we have f1 f2 (f2 )−f22 (f1 ) = 0 and f1 f2 , f22 ∈ A;
hence, any two elements from M are linearly dependent over A. This means
that if M is free, M ∼
= A.
But M 6∼ = Am for any m ∈ M since m vanishes somewhere on [0, 1]
thanks to continuity and since M possesses elements that do not vanish at
any prescribed point.
2) The elements f = (sin πt, cos πt) and g = (− cos πt, sin dπt) constitute
a free basis of M ⊕ M since, for any (m1 , m2 ) ∈ M ⊕ M , the system
1.11 The normal bundle and regular embeddings 63
(
x sin πt − y cos πt = m1 ,
(1.100)
x cos πt + y sin πt = m2

is uniquely solvable in A. u
t

1.11. The normal bundle and regular embeddings

1.11.1. Conormal module. Let Y be a closed subscheme of an affine
scheme X = Spec A determined by an ideal a. Then the A/a-module a/a2
is said to be the conormal module to Y with respect to the embedding
Y ,→ X or just the conormal module to Y and the family of vector spaces
N = Spec SA/a (a/a2 ) is said to be the normal family.
The following geometric picture illustrates this definition: a is the ideal
of functions on X that vanish on Y , a2 the ideal of functions whose zeroes on
Y are of order ≥ 2, and a/a2 is the module of linear parts of these functions
in a neighborhood of Y . A tangent vector to X at a point y ∈ Y determines
a linear function on such linear parts. A normal vector to Y at y (in the
absence of a natural metric) is a class of tangent vectors to X at y ∈ Y
modulo those that are tangent to Y , i.e., the ones which vanish on the linear
parts of the functions from a. Therefore, in “sufficiently regular” cases, a/a2
is (locally) the space dual to the space of vectors normal to Y . This explains
the term.
1.11.2. Regular embeddings. The conormal module is, in general, neither
free nor projective, but it is both free and projective for an important class
of subschemes.
A sequence of elements (f1 , . . . , fn ) of a ring A is said to be regular
(of length n) if, for all i ≤ n, the element fi mod (f1 , . . . , fi−1 ) is not zero
divisor in A/(f1 , . . . , fi−1 ); it is convenient to assume that the empty sequence
is regular of length 0 and generates the zero ideal.
A closed subscheme Y ⊂ X = Spec A is said to be regularly embedded or,
more often, a complete intersection (of co-dimension n), if A contains a regular
sequence of length n generating the ideal which singles out Y .
The geometric meaning of complete intersection becomes totally trans-
parent when we recall that we define Y adding one of the equations
fi = 0 at a time. Thus, we get a decreasing sequence of subschemes
X ⊃ Y1 ⊃ Y2 ⊃ . . . ⊃ Yn = Y . The complete intersection condition means
that Yi does not contain the whole support of any of the components of the
incompressible primary decomposition of Yi−1 . In other words, each equation
fi = 0 should be “transversal” (in a very weak sense) to all these supports.
Proposition. Let Y ,→ X be a complete intersection. Then its conormal
module is free.
In particular, the rank n of the conormal module does not depend on the
choice of a regular system of generators of the ideal.
64 Ch. 1. Affine schemes

The rank n of the conormal module is said to be the codimension of Y in

X.
Proof. Let a = (f1 , . . . , fn ) ⊂ A, where f1 , . . . , fn is a regular sequence. Ob-
viously, the elements f i = fi (mod a) generate the A/a-module a/a2 . There-
fore it suffices to verify that they are linearly independent. This is done by
induction on n:
First, let n = 1. Then f1 = f , g = g (mod Af ). If gf = 0, then gf = hf 2
for some h ∈ A, and therefore f (g − hf ) = 0; hence g = hf , since f is not
a zero divisor in A. Therefore g = 0.
Let the resultP be already proved for a regular sequence {f1 , . . . , fn−1 }.
Assume that g i f i = 0 in a/a2 , where g i = gi (mod a). We may assume
P 1≤i≤n P P
that gi fi = 0 in A: Otherwise gi fi = ui fi , where ui ∈ a,
1≤i≤n 1≤i≤n 1≤i≤n
and we may replace gi by gi − ui without affecting g i .
Since the class f n is not a zero divisor in A/(f1 , . . . , fn−1 ), it follows that
X
gn fn + gi fi = 0 =⇒ gn ∈ (f1 , . . . , fn−1 ) (1.101)
1≤i≤n−1
P P
i.e., gn = hi fi , implying (gi + hi fn )fi = 0. By the induction
1≤i≤n−1 1≤i≤n−1
hypothesis this means that gi + hi fn ∈ (f1 , . . . , fn−1 ) for i = 1, . . . , n − 1;
hence gi ∈ a for all i, i.e., g i = 0. u
t
1.11.3. Locally regularly embedded subscheme. A subscheme Y ,→ X
is said to be locally regularly embedded at y ∈ Y if there exists a neighborhood
D(f ) 3 y such that Y ∩ D(f ) is regularly embedded into D(f ). Obviously,
Y ∩ D(f ) is determined by an ideal af ⊂ Af and coincides with the fiber
product Y × Xf .
X

Statement. The normal family to a locally regularly embedded subscheme is

a vector bundle.
Indeed, (a/a2 )f = af /a2f so that A/a-module a/a2 is locally free for such
a subscheme. u
t
Remark. It well may happen that a subscheme is regularly embedded locally
but not globally. The first example of such a happening was encountered in
number theory.
Let A ⊃ Z be a ring of integer algebraic numbers of a field K. If the
number of classes 16) of K is greater than 1, then A possesses non-principal
ideals a ⊂ A (which are even prime). However, any such ideal, as is known,
is “locally” principal. Therefore a determines a locally regularly embedded
subscheme of co-dimension 1.
16
For definition, see [BSh].
1.11 The normal bundle and regular embeddings 65

1.11.4. The tangent and cotangent spaces. Let x ∈ X be a closed

point: For brevity, we will also denote by x the unique reduced subscheme
with support at this point. Let mx be the maximal ideal corresponding to x.
The discussion from sec. 1.11.1 shows that mx /m2x , the co-normal module to x,
is an analogue of the cotangent space to X at x. This is the Zariski cotangent
space. Its dual, (mx /m2x )∗ , is called the tangent space at x.
The closed points may be and may not be locally regularly embedded.
For instance, all the closed points of An = Spec K[T1 , . . . , Tn ], where, for
simplicity, K is algebraically closed, correspond to the ideals (T −t1 , . . . , T −tn ),
where ti ∈ K. The written system of generators of such an ideal (i.e.,
T − t1 , . . . , T − tn ) is, obviously, a regular sequence.
To get examples of non-locally regularly embedded points, it suffices to
consider the spectrum of a local Artinian ring which is not a field: All the
elements of its maximal ideal are nilpotents, and therefore there is no system
of generators whose first element is not a zero divisor. More meaningful ex-
amples are provided by hypersurfaces, i.e., subschemes of the affine space An
given by one equation.
1.11.5. Example. Let X ⊂ An be a closed subscheme of an affine space
over the field K through the origin 0; let the equation of X be F = 0, where
F = F1 + F2 + . . . and Fi is a form of degree i in T1 , . . . , Tn .
1.11.5a. Statement. The point x is locally regularly embedded into X if
and only if F1 (x) = 0.
1.11.5b. Corollary. Let F (t1 , . . . , tn ) = 0, where ti ∈ K. The point x sin-
gled out by the ideal (. . . , T − ti , . . .) is locally regularly embedded into X if
and only if there exists an i such that
∂F
(t , . . . , tn ) 6= 0. (1.102)
∂Ti i
Proof. Indeed, translate the origin to (t1 , . . . , tn ); then the linear part of F
in a vicinity of the new origin is equal to
X ∂F
(t1 , . . . , tn )(Ti − ti ) (1.103)
∂Ti
and it suffices to apply the above statement. u
t
This differential criterion shows that locally regularly embedded points are
exactly the ones which in the classical algebraic geometry are called nonsin-
gular ones.
Leaving a systematic theory of such points for future, we confine ourselves
here with the general facts needed for studying Example 1.11.5.
Note, first of all, that it suffices to consider the localized rings. More pre-
cisely, let p = (T1 , . . . , Tn ) ⊂ A, and p = p mod (F ) ⊂ B. Proof of Propo-
sition 1.11.2 shows that the point x is locally regularly embedded into X if
and only if the maximal ideal of the local ring Bp is generated by a regular
sequence. Observe that Bp = Ap /(F/1).
66 Ch. 1. Affine schemes

1.11.5c. Lemma. Under conditions of Example 1.11.5, if F1 6= 0, then the

maximal ideal in Bp is generated by a regular sequence.
Proof. Indeed, making a non-degenerate linear change of indeterminates we
may assume that F1 = T1 . Now, for any G ∈ K[T1 , . . . , Tn ] let G denote the
class G/1 mod (F/1) in the ring Bp .
The elements T2 /1, . . . , Tn /1 and F/1 form a regular system in the ring
Ap since F ≡ T1 + a2 T12 + . . . mod (T2 , . . . , Tn ), where ai ∈ K. Corollary 2.9.8
(to be proved in Chapter 2) implies that F/1, T2 /1, . . . , Tn /1 is also a regular
sequence. Therefore, T 2 /1, . . . , T n /1 is a regular sequence in Bp = Ap /(F/1);
clearly this regular sequence generates a maximal ideal in Bp . u
t
1.11.6. The statement converse to that of Lemma 1.11.5c. To prove
this statement, observe that if the origin is locally regularly embedded in X,
then the maximal ideal of the local ring Bp should be generated by a regular
sequence. The condition F1 = 0 means that F/1 belongs to the square of the
maximal ideal in Ap . Therefore it suffices to establish the following Lemma.
1.11.6a. Lemma. Let A be a local Noetherian ring, p ⊂ A its maximal ideal
generated by a regular sequence of length n. If f ∈ p2 and f is regular, then
the maximal ideal in the local ring A/(f ) can not be generated by a regular
sequence.
Proof. Let the maximal ideal in A/(f ) be generated by a maximal sequence
g 1 , . . . , g k , where g i = gi mod (f ), and gi ∈ A. Then (f, g1 , . . . , gk ) is a
regular sequence in A that generates p. Since the length of any such sequence
is equal to n (Prop. 1.11.2), we should have k = n − 1. But the elements
(f, g1 , . . . , gk ) generate in the n-dimensional A/p-space p/p2 a subspace of
dimension ≤ k = n − 1, since f ∈ p2 . The contradiction obtained proves the
lemma and completes the study of the example 1.11.5c. u
t

1.12. The differentials

1.12.1. The module of universal differentials. Let A and B be com-
mutative rings and B an A-algebra. Set

I = IB/A = Ker µ, where µ : B ⊗A B −→ B,

(1.104)
µ(b1 ⊗ b2 ) = b1 b2 is the multiplication.

Clearly, I is an ideal in B ⊗A B and (B ⊗A B)/I ∼

= B.
Consider the B-module:

CovectB/A = I/I 2 (1.105)

called the modules of (relative) differentials of the A-algebra B.

1.12 The differentials 67

Fig. 12

2
The B-module CovectB/A = IB/A /IB/A is called the module of (relative)
17)
differentials of the A-algebra B.
By Proposition 1.9.6 the ideal I determines the diagonal subscheme
∆X ⊂ X × X, where X = Spec B, S = Spec A. Due to the interpretation
S
from sec. 1.11.1 the module CovectB/A represents the co-normal to the diag-
onal.
In Differential Geometry, the normal bundle to the diagonal ∆ is isomor-
phic to the tangent bundle to the manifold X itself. Indeed, transport a vector
field along one of the fibers of the product X ×X “parallel” to the diagonal; we
get a vector field everywhere transversal to the diagonal, see Fig. 12. There-
fore CovectB/A is a candidate for the role of the module “cotangent ” to X
along the fibers of the morphisms X −→ S.
On the other hand, in the interpretation of nilpotents given in §§ 1.5 as
an analog of the elements of the “tangent module” to X (over S), we have
already considered the B-module DB/A of derivations of the A-algebra B
(a vector field on X, i.e., a section of the tangent bundle can be naturally
interpreted as a derivation of the ring of functions on X).
Remark. The higher order “differential neighborhoods of the diagonal” are
n
represented by the schemes Spec(B ⊗A B)/IB/A . They replace the spaces
of jets of the germs of the diagonal considered in differential geometry.
Define a map d = dB/A : B −→ CovectB/A by setting
2
d(b) = (b ⊗ 1 − 1 ⊗ b) (mod IB/A ). (1.106)

In the differential geometry, the tangent and the co-tangent bundles are
dual to each other. In the algebraic setting (over finite fields), this is not the
case, generally: Only “a half” of the duality is preserved:

DB/A = HomB (CovectB/A , B). (1.107)

17
In the differential and algebraic geometries we also need the exterior powers of
differential forms, and in this context the module of differentials is denoted by
1
ΩB/A . For details, see [MaG, SoS].
68 Ch. 1. Affine schemes

Therefore, DB/A is recovered from CovectB/A but not the other way round.
This explains the advantage of differentials as compared with derivatives. 18)
Lemma. 1) d is an even A-derivation, and d(ϕ(a)) = 0 for any a ∈ A, where
ϕ : A −→ B is the morphism that defines the A-algebra structure.
2) Let {bi | i ∈ I} be a system of generators of the A-algebra B. Then
{dbi | i ∈ I} is a system of generators of the B-module CovectB/A .
Proof. 1) is subject to a straightforward verification.
2) Notice that
X X
bi ⊗ b0i ∈ IB/A ⇐⇒ bi b0i = 0 ⇐⇒
i i
X X X X (1.112)
bi ⊗ b0i = bi ⊗ b0i − bi b0i ⊗ 1 = bi ⊗ 1(1 ⊗ b0i − b0i ⊗ 1).
i i i i

This implies that, as a B-module, CovectB/A is generated by the elements dB

for all b ∈ B. Since d is a derivation that vanishes on the image of A, this
easily implies the desired. u
t
Example. Let B = A[T1 , . . . , Tn ]. Then CovectB/A is the free B-module
freely generated by dTi .
1.12.2. Proposition. For any derivative d0 : B → M of B into a B-mod-
ule M that vanishes on the image of A, there exists a unique B-module ho-
momorphism ψ : CovectB/A −→ M such that d0 = ψ ◦ dB/A .
18
Grothendieck showed that one can, however, define differential operators of or-
der ≤ k for any ring R over K or Z as the K- or Z-linear maps of R-modules
D : M −→ N such that

[lr0 , [lr1 , . . . [lrk , D] . . .]] = 0 for any r0 , r1 , . . . rk ∈ R (1.108)

where lr denotes the operator of left multiplication by r in M and in N .

Denote by Diff k (M, N ) the R-module of differential operators of degree ≤ k;
set
Diff(M, N ) = lim Diff k (M, N ). (1.109)
−→
Define the R-module of symbols of differential operators to be the graded space
associated with Diff(M, N )

Smbl(M, N ) = ⊕ Smblk (M, N ), (1.110)

where
Smblk (M, N ) = Diff k (M, N )/ Diff k−1 (M, N ). (1.111)
If M = N , then, clearly, Diff (M ) = Diff (M, M ) is an associative algebra with
respect to the product and a Lie algebra — denoted by diff(M ) — with respect
to the bracket of operators. Smbl(M ) is a commutative R-algebra with respect to
the product. The bracket in diff(M ) induces a Lie algebra structure in Smbl(M )
(the Poisson bracket), as is not difficult to see, this Lie algebra is isomorphic to
the Poisson Lie algebra in dim M variables over R.
1.12 The differentials 69

(Applying this result to M = B we get (1.107)).

Proof. The uniqueness of ψ follows immediately from the fact that
d0 b = ψ(db) for all b ∈ B, so that ψ is uniquely determined on the system
of generators of CovectB/A .
To prove the existence, let us first define a group homomorphism

χ : B ⊗A B −→ M χ(b ⊗ b0 ) = bd0 b0 . (1.113)

2
This homomorphism vanishes on IB/A . Indeed, first notice that χ is a B-mod-
ule homomorphism with respect to the B-action on B ⊗A B given by b 7→ b⊗1.
Furthermore, as shown above, the elements b ⊗ 1 − 1 ⊗ b generate the B-mod-
ule IB/A , and therefore the products

(b1 ⊗ 1 − 1 ⊗ b1 )(b2 ⊗ 1 − 1 ⊗ b2 ) (1.114)

2
generate the B-module IB/A . Therefore, it suffices to verify that χ vanishes
on such products. This is straightforward; hence, we see that χ induces a map
ϕ : I/I 2 −→ M . We get

ϕ(d b) = ϕ(b ⊗ 1 − 1 ⊗ b) = d0 b, (1.115)

completing the proof. u

t
1.12.3. Conormal bundles. Now, let i : Y ,→ X be a closed embedding
of schemes. In models of differential geometry the restriction of the tangent
bundle to X onto Y contains, under certain regularity conditions, the tangent
bundle to Y , and the quotient bundle is the normal bundle to Y . We would
like to find out to what extent similar statement is true for the schemes.
Let us translate the problem into the algebraic language.
Let B be an A-algebra, b ⊂ B an ideal. Then B = B/b is also an A-algebra,
and the relative (over Spec A) cotangent sheaves on SpecB and Spec B are
represented by CovectB/A and CovectB/A , respectively. On the other hand,
the conormal bundle to the embedding Spec B −→ Spec B is represented by
the B/b-module b/b2 .
An analogue of the classical situation is given by the following Propo-
sition. Let δe = 1 ⊗B dB/A (e) for e ∈ b/b2 be represented by an element
e ∈ b. Since a map d0 : B −→ CovectB/A such that d0 f = dB/A (f mod b) is
an A-derivation, it factorizes through a uniquely determined B-module ho-
momorphism CovectB/A −→ CovectB/A ; since the target is annihilated by
multiplication by b, this homomorphism determines a B/b-module morphism
u : B/b ⊗B CovectB/A −→ CovectB/A .
Proposition. There exists an exact sequence of B/b-modules:
δ u
b/b2 −→B/b ⊗B CovectB/A −→ CovectB/A −→ 0, (1.116)

where δe = 1 ⊗B dB/A (e) for e ∈ b/b2 is represented by an element e ∈ b.

70 Ch. 1. Affine schemes

Proof. If e = 0, i.e., e ∈ b2 , then de ∈ bdB/A b, so that 1B ⊗ dB/A (e) = 0.

Hence, δ(e) does not depend on the choice of e. It is obvious that δ is a group
homomorphism, and the compatibility with the B/b-action follows from the
fact that
δ(f e) = 1 ⊗B d(f e) = 1 ⊗B (edf + f de) = f ⊗B de = f δ(e)
(1.117)
for any f = f mod b.

It is easy to see that u(1⊗B dB/A f ) = dB/A (f mod b) and, therefore, u is

an epimorphism.
It is easy to see that u ◦ δ = 0:

u ◦ δ(e) = u(1 ⊗ de) = d(e mod b) = 0. (1.118)

Let us verify the exactness in the middle term. Construct a homomorphism

v : CovectB/A −→ B ⊗B CovectB/A / Im δ (1.119)

such that u and v are mutually inverse. For this, first define a derivation

d0 : (B → B ⊗B CovectB/A )/ Im δ (1.120)

by setting

d0 (f ) = 1 ⊗B dB/A (f ) mod Im δ, for any f = f mod b. (1.121)

The result does not depend on the choice of a representative of f since

1 ⊗ dB/A (e) ∈ Im δ for any e ∈ b. (1.122)

This derivation determines the homomorphism v. Since

u ◦ v(df ) = df and (v ◦ u)(1 ⊗ df mod Im δ) = 1 ⊗ df mod Im δ, (1.123)

it follows that v and u are mutually inverse in some cases of our modules,
proving the desired. u
t
1.12.3a. Remark. The difference of the above constructions from similar
ones in the differential geometry is crucial: It well may happen that Ker δ 6= 0
even if the subscheme Y ,→ X is regularly embedded. For example, let
X = Spec Z and Y = Spec Z/p, where p is a prime, S = Spec Z. Then
CovectX/S = 0 and CovectY /S = 0; whereas (p)/(p)2 is a one-dimensional
linear space over Z/p.
Informally speaking, 19)

it is impossible to differentiate in the “arithmetic direction”.

19
See also the previous footnote.
1.13 Digression: Serre’s problem and Seshadri’s theorem 71

1.13. Digression: Serre’s problem and Seshadri’s

theorem
Serre posed the following problem: Over n-dimensional affine space, are
there non-trivial vector bundles?
In other words, is the following statement true?
Any projective Noetherian module over K[T1 , . . . , Tn ], where K is a field,
is free.
For n = 1, the ring K[T ] is an integral principal ideal ring. Therefore, any
Noetherian torsion-free K[T ]-module (in particular, any projective Noetherian
module) is free ([Lang]).
For n = 2, there are no non-trivial bundles, either. This theorem is due to
Seshadri; this section is devoted to its proof.
For n ≥ 3, the answer to Serre’s question remained unknown for some
time 20) . The problem is vary attractive and has all features of a classical one:
it is natural, pertains to the fundamental objects and is difficult (at least, for
ten years since it was posed there appeared no essential results on modules
over polynomial rings apart from Seshadri’s theorem and the following fact
due to Serre himself.
1.13.1. Theorem. Let P be a projective Noetherian module over K[T ],
where T = (T1 , . . . , Tn ). Then there exists a free Noetherian module F such
that P ⊕ F is free.
In terms of topologists, vector bundles over affine spaces are stably free.
Proof easily follows from Hilbert’s syzygies theorem which will be given in
an appropriate place.
Therefore we confine ourselves to Seshadri’s theorem. It involves a class of
rings containing, in addition to K[T1 , T2 ], e.g., Z[T ].
1.13.2. Theorem. Let A be an integral principal ideal ring. Then any pro-
jective Noetherian A[T ]-module P is free.
Proof will be split into a series of lemmas. Its driving force is a simple
remark that if A is a field, the statement is true. One can cook a field of
A in two ways: pass from A to its field of quotients K or to the quotient
field k = A/(p), where p is any prime element. Accordingly, the modules
K[T ] ⊗ A[T ]P and k[T ] ⊗ A[T ]P are free. Let us use these circumstances in
turn.
1.13.3. Lemma. There exists an exact sequence of A[T ]-modules

0 → F → P → P/F → 0 (1.124)

with the following properties:

20
The affirmative answer (Any projective Noetherian K[T ]-module is free) is due
to Suslin and D. Quillen[VSu, Su]. L. Vassershtein later gave a simpler and much
shorter proof of the theorem which can be found in Lang’s book [Lang].
72 Ch. 1. Affine schemes

a) F is a maximal A[T ]-free submodule of P ;

b) Ann P/F ∩ A 6= {0}.
Proof. Let m01 , . . . , m0r be a free K[T ]-basis of the module K[T ] ⊗ P . There
exists an element 0 6= f ∈ A such that mi = f m0 ∈ Pi ,→ K[T ] ⊗ P . The
submodule F 0 ⊂ P generated by the elements mi , where i = 1, . . . , r, is free.
On the other hand, any element of a finite fixed system of generatorsP of the
module P is represented in K[T ] ⊗ P as a linear combination i Fij (t)mi ,
where Fij (T ) ∈ K[T ]. The common denominator of all coefficients of all poly-
nomials Fij (T ) A annihilates P/F 0 . Now, for the role of F we may take a
maximal free submodule in P containing F 0 : it exists thanks to Noetherian
property. Clearly, Ann(P/F ) ⊃ Ann(P/F 0 ), so Ann(P/F ) ∩ A 6= {0}. u
t
We retain notation of Lemma 1.13.3 and intend to deduce a contradiction
from the assumption that F 6= P . In this case Ann(P/F ) ∩ A = (f ) ⊂ A,
where f is non-invertible (since A is a principal ideal ring). Let p be a prime
element of A dividing f . Set k = A/(p) and tensor the exact sequence (1.124)
by k[T ] over A[T ], having set F = F/pF = k[T ] ⊗ A[T ]F and so on:

i /P / P/F / 0.
F

Let F 1 = Ker i, F 2 = Im i. Since P is projective over k[T ], it follows that

F 2 is torsion-free, and hence is free. Therefore F 1 is also free and is singled
out in F as a direct summand, so there is defined a split sequence of free
k[T ]-modules:
j i
0 /F /F /F / 0. (1.125)
1 2

1.13.4. Lemma. F 1 6= 0.
Proof. Indeed, j(F 1 ) = pP ∩ F/pF . Let f = pg. Since g ∈
/ Ann P/F ∩ A,
we have gP 6⊂ F =⇒ pgP 6⊂ pF (because p is torsion-free). But
pgP = f P ⊂ pP ∩ F , so, moreover, pP ∩ F 6⊂ pF . u
t
The last step requires some additional arguments.
1.13.5. Lemma. There exists a free A[T ]-submodule F1 ⊂ F with a free
direct complement and such that k[T ] ⊗ F1 = j(F 1 ).
Speaking informally the sequence (1.125) can be lifted to a split exact
sequence of free A[T ]-modules.
1.13.6. Deduction of Theorem 1.13.2. Let F1 ⊂ F be a submodule
whose existence is claimed in Lemma 1.13.5, F2 ⊂ F its free direct comple-
ment. Since F1 /pF1 = Ker i, all elements F1 ⊂ F ⊂ P are divisible by p inside
P . Set
F10 = {m ∈ P | pm ∈ F1 }.
Clearly, F10 is free (the multiplication by p determines an isomorphism
F10 ' F1 ) and is strictly larger than F1 (by Lemma 1.13.4). Therefore the
1.13 Digression: Serre’s problem and Seshadri’s theorem 73

module F 0 = F10 ⊕ F2 ⊂ P is free and contains F as a proper submodule

contradicting maximality of F and completing the proof of Seshadri’s theo-
rem. u
t
1.13.7. Proof of Lemma 1.13.5. Any automorphism ϕ of the module F
induces an automorphisms ϕ of the module F . We need the following auxiliary
statement:
1.13.7a. Lemma. The map ϕ 7→ ϕ : SL(n, A[T ]) −→ SL(n, k[T ]) is surjec-
tive.
Proof. We use a classical result on reduction of the matrix with elements over
the Euclidean ring k[T ] to the diagonal form by “admissible transformations”.
For this result, see the book [vdW], § 85, where it is given in terms of bases.
To formulate it, denote by In the unit (n × n)-matrix over k[T ], let I(ij) (resp.
I (ij) ) be the matrix obtained from I by transposition of the i-th and j-th
rows (resp. columns), let Eij be the matrix with a 1 on the (ij)-the slot and
0 elsewhere.
Proof of the Theorem on elementary divisors in the book [vdW] shows, in
particular, that in a fixed basis of F any automorphism with determinant 1
can be represented as the product of matrices of the following types:
a) I + f Eij , f ∈ k[T ];
b) I(ij) ;
c) I (ij) ;
d) diagonal matrices with elements of k and with determinant 1.
The matrices of the first three types can obviously be lifted to elements of
SL(n, A[T ]). The matrices of the fourth type can be factorized in the product of
diagonal matrices with determinant 1 and such that only two of their diagonal
elements are 6= 1. Therefore
! we have reduced the problem to lifting matrices
f 0
of the form −1 ∈ SL(2, k) to matrices of SL(2, A).
0 f
This can be done in a completely elementary way. First, select an element
−1
f ∈ A such that f ≡ f mod (p), next an element g ∈ A such that f ≡g
mod (p) and (g, f ) = 1 (this is possible thanks to the Chinese remainder
theorem). Now we have f g = 1 + ph. Solve in A the equation f x + gy = h;
then
(f − py)(g − px) ≡ 1 + p2 xy,
 
f − py px
so the matrix is a solution desired. u
t
py g − px
Now, return to the proof of Lemma 1.13.5. Select a free A[T ]-basis (mi )i∈I
of the module F ; its reduction modulo p is a free k[T ]-basis (mi )i∈I of the
module F . Further, select a free k[T ]-basis (ni )i∈I of the module F compatible
with the split sequence 1.13.2 (in the sense that the first rk F 1 of its elements
constitute a basis of F 1 ). We may assume that the matrix M ∈ GL(n, k[T ])
sending the set (mi )i∈I into the set (ni )i∈I belongs to SL(n, k[T ]): If not, it
74 Ch. 1. Affine schemes

suffices to replace n1 by (det M )−1 n1 . Let us now lift M to M ∈ SL(n, A[T ])

and let (ni )i∈I be an A[T ]-basis of the M -module F . Further, let F1 be a
submodule of F generated by the first rk F 1 elements of the basis (ni )i∈I ,
and F2 the submodule generated by the remaining elements. The construction
of these submodules shows that they satisfy Lemma 1.13.5 completing the
proof. u
t

1.14. Digression: ζ-function of a ring

1.14.1. An overview. The rings of finite type over a field are called ge-
ometric rings, those over Z arithmetic ones. These two types of rings have
a nonzero intersection: The rings of finite type over finite fields. Such rings
(and their spectra) enjoy a blend of arithmetic and geometric properties as
demonstrated by A. Weil in his famous conjectures on ζ-functions proved by
P. Deligne.
Here we will introduce ζ-functions of arithmetic rings and indicate their
simplest properties. A motivation of introducing the ζ-function: The closed
points x in the spectrum of an arithmetic ring have a natural “norm” N (x)
equal to the number of elements in the finite field k(x), and the number
of points of given norm is finite. It is natural to assume that directly counting
such points we get an interesting invariant of the ring.
Let A be an arithmetic ring. Let n(pa ) be the number of closed points
x ∈ Spec A such that N (x) = pa , and ν(pa ) be the number of geometric
Fpa -points of A.
Lemma. The numbers ν(pa ) and n(pa ) are finite and related with each other
as follows: X
ν(pa ) = bn(pb ). (1.126)
b|a

Proof. Every geometric Fpa -point of A is by definition a homomorphism

A −→ Fpa . Consider all the geometric points with the same center x; then
px ⊂ A is the kernel of the corresponding homomorphism and its image coin-
cides with the unique subfield Fpb ,→ Fpa , where pb = N (x). There are exactly
b homomorphisms with fixed kernel and image, since Fpb /Fpa is a Galois ex-
tension of degree b. Therefore
X µ X ¶ X
a
ν(p ) = b 1 = bn(pb ). (1.127)
b|a {x|N (x)=pb } b|a

(This equality is obviously well-defined even if it is not known that ν(pa ) and
n(pb ) are finite).
In particular, n(pa ) ≤ ν(pa ) and it suffices to prove that ν(pa ) is finite.
We identify Spec A with a closed subset in Spec Z[T1 , . . . , Tn ]; then N (x)
does not depend on whether we consider x as belonging to Spec A or to
Spec Z[T1 , . . . , Tn ]. Therefore, in obvious notation,
1.14 Digression: ζ-function of a ring 75

ν(pa ) ≤ νZ[T1 ,...,Tn ] (pa ) = pna (1.128)

where, clearly, pna is just the number of geometric points of the n-dimensional
affine space over a field of pa elements. u
t
1.14.2. ζ-functions of arithmetic rings. Define the ζ-function of any
arithmetic ring A first formally, by setting
Y 1
ζA (s) = −s
. (1.129)
1 − N (x)
x∈Spm A

Clearly, for A = Z, we get the usual Euler function

Y 1
ζZ (s) = −s
= ζ(s). (1.130)
1−p
p

The relation of the ζ-function with n(pa ) and ν(pa ) is given by the following
obvious identity
Y Y 1 Y
ζA (s) = −as n(pa )
= ζA/pA (s) (1.131)
(1 − p )
p 1≤a<∞ p

and another, a trifle less obvious one,

XX 1
ln ζA (s) = ν(pa ) as . (1.132)
ap
p a

Proof of eq.(1.132) (use the above lemma):

XX XXX 1
ln ζA (s) = − ln(1 − p−bs )n(pb ) = − n(pb ) bks
kp
p a p b∈N k∈N
XXX b XX 1
= n(pb ) = ν(pa ) . u
t
apas as pas
p a∈N b|a p a
(1.133)
Therefore the calculation of the ζ-function is equivalent to that of n(pa ) or
ν(pa ) for all p, a. u
t
Eq. (1.131) shows that ζA (s) factorizes into the product of ζ-functions
of the rings of finite type over finite fields. This, however, does not mean in
the least that the study of ζ-functions reduces to the cases of such rings and
the example of the Riemann ζ-function shows how nontrivial the behavior
of the global ζ-function can be even for simplest local factors.
Even to separate p-factors can be a sufficiently complicated task if A is
nontrivial.
A part of Weil’s conjectures proved by Dwork 21) shows, however, that
ζ(s) is a rational function in p−s for any ring A of finite type over a field
21
See [Kz]. By early 1970s Weil’s conjectures (and even more difficult statements)
were proved by Grothendieck and Deligne, see [SGA4, SGA4.5, Dan].
76 Ch. 1. Affine schemes

of characteristic p. For such rings, it is convenient to change the variable by

setting p−s = t and set ζA (s) = ZA (t). Then eq.(1.132) shows that
X ν(pa )ta
ln ZA (t) = (1.134)
a
a

or 0 X
ZA (t)
= ν(pa )ta−1 (1.135)
ZA (t)
a

In particular, the rationally of ZA (t) establishes that the sequence νa = ν(pa )

should satisfy a recurrent relation of type
X
νa+n = ri νa+i (1.136)
0≤i≤n−1

for sufficiently large a with some fixed constants n, ri . Since the νa are the
numbers of solutions of a system of equations with values in finite fields
of growing degree, the statement on rationality bears a direct arithmetic mean-
ing.
1.14.3. Frobenius morphism. In any study of the ζ-function of a ring A
over a field k of characteristic p the following circumstance is of fundamental
importance: ν(pk ) can be viewed as the number of fixed points of a power
of a certain map F acting on the set of geometric points of A.
A Frobenius morphism F : A −→ A is the map g 7→ g p for any g ∈ A,
where p = Char k.
The same term — Frobenius morphism — is applied to the correspond-
ing morphism of spectra, aF , to its powers, F n , to ( aF )n , and to the maps
of some other objects induced by these maps. In particular, let F̄p be the al-
gebraic closure of the Galois field of characteristic p. Then F induces a map
a
F : A(F̄p ) −→ A(F̄p ) of F̄p -points of A into itself.
Proposition. A(Fpb ) coincides with the set of fixed points of F b .
Proof. Let ϕ ∈ A(F̄p ) and let ϕ be represented by ϕ : A −→ F̄p ; let F b (ϕ)
b
be represented by f 7→ ϕ(f )p for any f ∈ A.
The condition ϕ ∈ A(Fpb ) means that Im ϕ ⊂ Fpb ⊂ F̄pb , i.e.,
b
ϕ(f )p = ϕ(f ) for all f . Therefore aF ϕ = ϕ.
The converse statement follows from the Galois theory: Fpb is the field
of invariants for F b . u
t
1.14.4. Lefschetz formula. If an endomorphism F acts on a compact to-
pological space V , then the number ν(F ) of its fixed points (appropriately
defined) satisfies the following famous Lefschetz formula:
X
ν(F ) = (−1)i tr F |H i (V ) , (1.137)
0≤i≤dim V
1.14 Digression: ζ-function of a ring 77

where the summands are the traces of linear operators induced by F on the
spaces of cohomology of V with complex coefficients.
The role of compact topological spaces V is played in our setting by smooth
projective schemes. For them, the essential part of Weil’s conjectures states
that the numbers ν(pa ) are always expressed by Lefschetz type formulas. So
far, we have dealt with Euler’s products and Dirichlet series purely formally.
Now, let us study a little their convergence.
Let A be an arithmetic ring, {xi }i∈I the set of generic points of its irre-
ducible components. Define the dimension of A setting
(
maxi (tr. deg. k(xi )) + 1 if Z ⊂ A
dim A = (1.138)
maxi (tr. deg. k(xi )) if Char A > 0

(The transcendence degree is calculated over a prime subring of k(xi ). The

dimension thus introduced was considered already by Kronecker.)
Q 1
Theorem. The Euler product −s
converges absolutely for
x∈Spm A 1 − N (x)
Re s > dim A.
Proof. We will verify the theorem consecutively extending the class of rings
considered. We assume that for the Riemannian ζ-function ζZ (s) the conver-
gence of the product is known.
a) Let A = Fp [T1 , . . . , Tn ]. It is subject to a direct verification that
eq.(1.132) converges absolutely to ln(1 − pn−s )−1 for Re s > n = dim A
since ν(pa ) = pan . This implies that under the same conditions the Euler’s
1
product for A absolutely converges to .
1 − pn−s
b) Let A be a ring without zero divisors and of finite type over Fp . Let
us apply Noether’s normalization theorem and find a polynomial subring
B = Fp [T1 , . . . , Tn ] ⊂ A such that A is a B-module with a finitely many gen-
erators. There exists a constant d such that, over every geometric F̄p -point
of B, not more than d geometric points of A are situated.
Indeed, let a homomorphism A −→ F̄p be given on B. To extend it over
A, we have to define in F̄p the images of a finite number of generators of B
over A each of which is a root of an integer equation with coefficients from A.
The images of these coefficients are already defined, and therefore the roots
of equations are determined in finitely many ways.
This implies that νA (pa ) ≤ ανB (pa ) = αpna , and therefore ζA (s) converges
absolutely for Re s > n = dim A, as above. Moreover, in this domain, we have

| ln ζA (s)| ≤ α ln(1 − pn−σ )−1 , where σ = Re s. (1.139)

c) Let A be an arbitrary ring of finite type over Fp . Let the pi ⊂ A be

all the minimal prime ideals of A, and Ai = A/pi . Every geometric point
of Spec A belongs to an irreducible component, and therefore
78 Ch. 1. Affine schemes
X
νA (pa ) ≤ νAi (pa ),
i

so the Euler product for A converges if Re s > max dim Ai and in this domain
i
satisfies
X
| ln ζA (s)| ≤ α ln(1 − pni −σ )−1 , where ni = dim Ai . (1.140)
i

d) A = Z[T1 , . . . , Tn ]. It follows from the calculations from a) that in this

case Y 1
ζ(s) = n−s
= ζ(s − n) (1.141)
1−p
p

is the usual ζ-function with shifted argument whose Euler product, as is well
known, converges absolutely for Re (s − n) > 1, i.e., Re s > n + 1 = dim A.
e) Let A be a ring without zero divisors containing Z. If we can find
a subring Z[T1 , . . . , Tn ] of A, over which A is integer, the arguments from
b) bring about the result. Regrettably, this is not always possible; we can,
however, remedy the situation for the price of localization modulo a finite
number of primes.
More exactly, let us apply Noether’s normalization theorem to A0 = Q⊗Z A
and find a subring Q[T1 , . . . , Tn ] in A0 over which A is integer. Multiplying,
if necessary, Ti by integers, we can assume that Ti ∈ A. Any element of A
over Z[T1 , . . . , Tn ] satisfies an equation whose highest coefficient is an integer.
Consider the set of prime divisors of all such highest coefficients for a finite
system of generators of A over Z[T1 , . . . , Tn ] and denote by S the multiplicative
system generated by this set. Then AS is integer over ZS [T1 , . . . , Tn ], and
Y Y
ζA (s) = ζA/(p) (s) ζA/(p) (s). (1.142)
p∈S p6∈S

The set {p | p ∈ S} is finite and from the above ζA/(p) (s) converges uniformly
for Re s > dim A/(p) ≥ dim A − 1.
For p 6∈ S, we have ζA/(p) (s) = ζAS /(p) (s), and the constant α for the pair
of rings ZS [T1 , . . . , Tn ]/(p) ⊂ AS /(p) determined in b) can be chosen to be
independent of p. Indeed, the class modulo p of the fixed system of integer
generators AS over ZS [T1 , . . . , Tn ] gives a system of generators of AS /(p) for all
p 6∈ S. Therefore the second (infinite) product in (5) for σ = Re s > dim A/(p)
is majored by Y
(1 − pn−σ )−α ,
p6∈S

and therefore converges uniformly for σ > n + 1 = dim A.

f) Finally, we trivially reduce the general case to the above-considered ones
by decomposing Spec A into irreducible components as in c). u
t
1.15 The affine group schemes 79

1.14.5. Exercises. 1) Express n(pa ) in terms of ν(pb ).

2) Calculate the number of irreducible polynomials in one indeterminate
of degree d over the field of q elements.
3) Calculate ζA (s), where A = Z[T1 , . . . , Tn ]/(F ), and F is a quadratic
field.
4) Let A be a ring of finite type over Z, and P the set of prime numbers;
let

S = {p ∈ P | there exists x ∈ Spec A, such that Char k(x) = p}.

Prove that either S or P \ S is finite. For the integer domain not of finite type
over Z, give an example when both S and P \ S are infinite.

1.15. The affine group schemes

In this section we give definitions and several most important examples
of affine group schemes. This notion is not only important by itself, it also
lucidly shows the role and possibilities of the “categorical” and “structural”
approaches.
We will give two definitions of a group structure on an object of a category
and compare them for the category of schemes.
1.15.1. A group structure on an object of a category. Let C
be a category, X ∈ Ob C. A group structure on X is said to be
given if there are given (set theoretical) group structures on all the sets
PX (Y ) := HomC (Y, X), and, for any morphism Y1 −→ Y2 , the corresponding
map of sets PX (Y2 ) −→ PX (Y1 ) is a group homomorphism.
An object X together with a group structure on it is said to be a group
in the category C. Let X1 , X2 be groups in C; a morphism X1 −→ X2 in C is
a group morphism in C if the maps PX1 (Y ) −→ PX2 (Y ) are group homomor-
phisms for any Y .
A group in the category of affine schemes will be called an affine group
scheme 22)
Here is the list of the most important examples with their standard nota-
tion and names.
Helpful remark. Since Aff Sch◦ = Rings, instead of studying contravariant
functors on Aff Sch represented by an affine group scheme we may discuss the
covariant functors on Rings which are simpler to handle.
1.15.2. Examples.
22
Never an affine group: This is a fixed term for a different notion.
80 Ch. 1. Affine schemes

1.15.2a. The additive group Ga = Spec Z[T ]. As above, a morphism

Spec A −→ Ga is uniquely determined by an element t ∈ A, the image of T ,
which may be chosen at random. The collection of groups with respect to
addition A = Ga (A) for the rings A ∈ Rings determines the group structure
on Ga .
In other words, Ga represents the functor Aff Sch◦ −→ Gr, Spec A 7→ A
or, equivalently, the functor Rings −→ Gr, A 7→ A+ .
1.15.2b. The multiplicative group Gm = Spec Z[T, T −1 ]. For any
superscheme X = Spec A, a morphism X −→ Gm is uniquely determined by
an element t ∈ A× , the image of T under the homomorphism Z[T, T −1 ] −→ A.
Conversely, t corresponds to such a morphism if and only if t ∈ A× , where
A×0̄
is the group (with respect to multiplication) of invertible elements of A.
Therefore
PGm (Spec A) = Gm (A) = A× , (1.143)
and on the set of A-points the natural group structure (multiplication) is de-
fined. Furthermore, any ring homomorphism A −→ B induces, clearly, a group
homomorphism A× −→ B0̄× which determines the group structure on Gm .
In other words, Gm represents the functor Aff Sch◦ −→ Gr, Spec A 7→ A×
or, equivalently, the functor Rings −→ Gr, A 7→ A× .
1.15.2c. The general linear group.

GL(n) = Spec Z[Tij , T ]ni,j=1 /(T det(Tij ) − 1). (1.144)

It represents the functor Spec A 7→ GL(n; A). Obviously, GL(1) ' Gm .

1.15.2d. The Galois group Aut(K 0 /K). Fix a K-algebra K 0 and let K 0
be a free K-module of finite rank. The group Aut(K 0 /K) of automorphisms
of the algebra K 0 over K is the main object of the study, e.g., in the Galois
theory (where the case of fields K, K 0 is only considered). This group may
turn to be trivial if the extension is non-normal or non-separable, and so on.
The functorial point of view suggests to consider all the possible changes
of base K, i.e., for a variable K-algebra B, consider the group of automor-
phisms O
Aut(B 0 /B) := AutB (B 0 ), where B 0 = B K 0. (1.145)
K

We will prove simultaneously that (1) the map B 7→ Aut(B 0 /B) is a functor
and (2) this functor is representable.
Select a free basis e1 , . . . , en of K 0 over K. In this basis the multiplication
law in K 0 is given by the formula
X
ei ej = ckij ek . (1.146)
1≤k≤n
N L
Denote e0i := 1 ei ; then B 0 = B e0i , and any endomorphism t of the
K 1≤i≤n
B-module B 0 is given by a matrix (tij ), where tij ∈ B and 1 ≤ i, j ≤ n. The
1.15 The affine group schemes 81

condition that this matrix determines an endomorphism of an algebra can be

expressed as the relations
X
t(e0i )t(e0j ) = ckij t(e0k ). (1.147)
1≤k≤n

Equating the coefficients of e0k in (1.147) in terms of indeterminates Tij we

obtain a system of algebraic relations for Tij with coefficients from K, both
necessary and sufficient for (tij ) to define an endomorphism of B 0 /B.
To obtain automorphisms, let us introduce an additional variable t and
the additional relation (cf. Example 3) which ensures that det(tij ) does not
vanish:
t det(tij ) − 1 = 0. (1.148)
The quotient of K[T, Tij ]ni,j=1 /(T det(Tij )ni,j=1 −1) is a K-algebra representing
the functor
B 7→ Aut(B 0 /B). (1.149)
This K-algebra replaces the notion of the Galois group of the extension
K 0 /K; it generalizes the notion of the group ring of the Galois group.
1.15.2e. Consider the simplest particular case:
√
K 0 = K( a), where a ∈ K × \ (K × )2 . (1.150)
√
We may√ set e1 = 1, √e2 = a; the multiplication table√reduces to e22 = a.
Let t( a) = T1 + T2 a (obviously, t(1) = 1). Since t( a)2 = a, we obtain
the equations relating T1 , T2 and the additional variable T :

 2 2
T1 + a T2 = a
2T1 T2 =0 (1.151)


T T2 − 1 = 0.

Now, let us consider separately two cases.

Case 1: Char K 6= 2. Hence, 2 is invertible in any K-algebra. The functor
of automorphisms is represented by the K-algebra

K[T, T1 , T2 ]/(T12 + a T22 − a, T1 T2 , T T2 − 1). (1.152)

If B has no zero divisors, then the B-points of this K-algebra have a simple
structure: Since T2 must not vanish, T1 vanishes implying that the possible
values of T2 in the quotient ring are ±1. As the conventional Galois group
this√group is isomorphic to Z/2; the automorphisms simply change the sign
of a.
The following case illustrates that when B does have zero divisors the
group of B-points of Aut K × /K can be much larger.
Case 2: Char K = 2. The functor of automorphisms is represented by the
K-algebra
82 Ch. 1. Affine schemes

K[T, T1 , T2 ]/(T12 + aT22 − a, T T2 − 1). (1.153)

In other words, the B-points of the automorphism group are all the B-points
of the circle T12 + aT22 − a = 0 at which T2 is invertible!
Let us investigate this in detail. Let B be a field and let (t1 , t2 ) be a B-point
of the circle at which T2 is invertible. Then either t2 = 1, t1 = 0, and we obtain
³ ´2
t1
the identity automorphism, or a = . Therefore there are nontrivial
√ t2 + 1
B-points only if a ∈ B, in which √ case the √ equation of the circle turns into
the square of a linear one (T1 + a T2 + a)2 = 0. We have the punctured
line (the line without point T2 = 0) of automorphisms!
Obviously, Aut(B 0 /B) is isomorphic in this case to B × — the multiplica-
tive√group of B (under the composition of automorphisms the coefficients
of a are multiplied). So, the non-separable extensions have even more, in
a certain sense, automorphisms than separable ones.
The reasonN why this phenomenon√ takes place√is presence√of nilpotents in
the algebra B K 0 . Indeed, K( a) ⊂ L, so K( a) ⊗K K( a) ⊂ L0 ; on the
K
other hand, this product is isomorphic to
√ √
K( a)[x]/(x2 − a) ' K( a)[y]/(y 2 ) :

The automorphisms just multiply y by invertible elements.

One can similarly investigate arbitrary inseparable extensions and con-
struct for them an analog of the Galois theory.
1.15.2f. The group µn of nth roots of unity. Set

µn = Spec Z[T ]/(T n − 1) = Spec Z[T, T −1 ]/(T n − 1). (1.154)

This group represents the functor Spec A 7→ {t ∈ A× | tn = 1}.

Let X be a closed affine group scheme and Y its closed subscheme Y such
that PY (Z) ⊂ PX (Z) is a subgroup for any Z. We call Y with the induced
group structure a closed subgroup of X.
Therefore µn is a closed subgroup of Gm . Explicitly, the homomorphism
T 7→ T n determines a group scheme homomorphism Gm −→ Gm of “raising
to the power n” and µn represents the kernel of this homomorphism.
1.15.2g. The scheme of a finite group G. Q Let G be a conventional (set-
theoretical) finite group. Set A = Z(G) := Z. In other words, A is a free
g∈G
L (g)
module Z (|G| copies of Z) with the multiplication table
g∈G
(
0 = (0, . . . , 0) if h 6= g,
eg eh = (1.155)
eg if h = g.

The space X = Spec A is disjoint; each of its components is isomorphic to

Spec Z and these components are indexed by the elements of G. For any
1.15 The affine group schemes 83

ring B, whose spectrum is connected, the set of morphisms Spec B −→ Spec A

is, therefore, in the natural one-to-one correspondence with the elements of G.
If Spec B is disjoint, then a morphism Spec B −→ Spec A is determined
by the set of its restrictions onto the connected components of Spec B. Let
Conn B be the set of these components; then, clearly, the point functor is
given by
∼
PX (Spec B) −→ (G)Conn B := Hom(G, Conn B). (1.156)
and therefore X is endowed with a natural group structure called the scheme
of the group G.
1.15.2h. The relative case. Let S = Spec K. A group object in the cat-
egory Aff SchS of affine schemes over S is said to be an affine S-group (or
an affine K-group). Setting Gm /K = Gm × S and µn /K = µn × S, and so on,
we obtain a series of groups over an arbitrary scheme S (or a ring K). Each
of them represents “the same” functor as the corresponding absolute group,
but restricted onto the category of K-algebras.
1.15.2i. Linear algebraic groups. Let K be a field. Any closed subgroup
of GLn (A)/K is said to be a linear algebraic group over K.
In other words, a linear algebraic group is determined by a system of
equations
Fk (Tij ) = 0, for i, j = 1, . . . , n and k ∈ I (1.157)
such that if (t0ij ) and (t00ij ) are two solutions of the system (1.157) in a K-al-
gebra A such that the corresponding matrices are invertible, then the matrix
(t0ij )(t00ij )−1 is also a solution of (1.157)
The place of linear algebraic groups in the general theory is elucidated by
the following fundamental theorem (cf. [OV]).
Theorem. Let X be an affine group scheme of finite type over K. Then X
is isomorphic to a linear algebraic group.
1.15.3. Statement (Cartier). Let X be the scheme of a linear algebraic
group over a field of characteristic zero. Then X is reduced, i.e., X = Xred ,
its ring has no nilpotents.
For proof of this theorem, see, e.g., [M1]. u
t
The statement of the theorem is false if 23) Char K = p as demonstrated
by the following
Example. Set

µp /K = Spec K[T ]/(T p−1 ) = Spec K[T ]/((T − 1)p ). (1.158)

Obviously, K[T ]/((T − 1)p ) is a local Artinian algebra of length p, and its
spectrum should be considered as a “point of multiplicity p”. This is a nice
23
And also for the group superschemes over any fields.
84 Ch. 1. Affine schemes

agreement with our intuition: All the roots of unity of degree p are glued
together and turn into one root of multiplicity p.
More generally, set
n
µpn /K = Spec K[T ]/((T − 1)p ). (1.159)

We see that the length of the nilradical may be however great.

1.15.4. The set-theoretical definition of the group structure. Let
a category C contain a final 24) object E and products. Let X be a group with
a unit 1; let x, y, z ∈ X; then, in the standard notation 25) , we obtain

m(x, y) = xy, i(x) = x−1 , u(E) = 1, (1.160)

and the conventional axioms of the associativity, the left inverse and the left
unit have, respectively, the form

(xy)z = x(yz), x−1 x = 1, 1x = x (1.161)

The usual set-theoretical definition of the group structure on a set X given

above is, clearly, equivalent to the existence of three morphisms

m : X × X −→ X (multiplication, x, y 7→ xy)
i: X −→ X (inversion, x 7→ x−1 ) (1.162)
u: E −→ X (unit, the embedding of E)

that satisfy the axioms of associativity, left inversion and left unit, respectively,
expressed as commutativity of the following diagrams:

(m, idX )
X ×X ×X / X ×X

(idX ,m) m (1.163)

² m ²
X ×X /X
(i, idX )
X ×O X / X ×X

δ m (1.164)
• u ²
X /E /X
(•, idX ) (u, idX )
X ×O X / E×X / X ×X

δ m (1.165)
idX idX ²
X /X /X
24
An object E is said to be final if card(Hom(X, E)) = 1 for any X ∈ Ob C.
25
Here we denote the group unit by 1; it is the image of E.
1.15 The affine group schemes 85

(In diagrams (1.164), (1.165) the morphism of contraction to a point (E) is

denoted by “•”.)
In the category Sets the axioms (1.163)–(1.165) turn into the usual defini-
tion of a group though in an somewhat non-conventional form.
Exercise. Formulate the commutativity axiom for the group.
1.15.5. Equivalence of the two definitions of the group structure.
Let a group structure in the set-theoretical sense be given on X ∈ Ob C.
Then, for every Y ∈ Ob C, the morphisms m, i, u induce the group structure
on the set PX (Y ) of Y -points thanks to the above subsection.
1.15.5a. Exercise. Verify compatibility of these structures with the maps
PX (Y1 ) −→ PX (Y2 ) induced by the maps Y1 −→ Y2 .
Conversely, let a group structure in the sense of the first definition be given
on X ∈ Ob C. How to recover the morphisms m, i, u? We do it in three steps:
a) The group PX (X × X) contains projections π1 , π2 : X × X −→ X. Set
m = π1 ◦ π2 (the product ◦ in the sense of the group law).
b) The group PX (X) contains the element idX . Denote its inverse (in the
sense of the group law) by i.
c) The group PX (E) has the unit element. Denote it by u : E −→ X.
Exercise. 1) Prove that m, i, u satisfy the axioms of the second definition.
2) Verify that the constructed maps of sets
   

 group structures on X 
 
 group structures on X 

   
with respect to ←→ with respect to (1.166)

 
 
 

   
the first definition the second definition

are mutually inverse.

1.15.6. How to describe the group structure on an affine group
scheme X = Spec A in terms of A. We will consider the general, i.e.,
relative, case, i.e., assume A to be a K-algebra.
The notion of a group G is usually formulated in terms of the states, i.e.,
points of G. In several questions, however, for example, to quantize it, we
need a reformulation in terms of observables, i.e., the functions on G. Since
any map of sets ϕ : X −→ Y induces the homomorphism of the algebras of
functions ϕ∗ : F (Y ) −→ F (X), we dualize the axioms of sec. 1.15.4 and obtain
the following definition.
A bialgebra structure on a K-algebra A is given by three K-algebra homo-
morphisms:
N
m∗ : A −→ A A co-multiplication
K
∗
i : A −→ A co-inversion (1.167)

u∗ : A −→ K co-unit
86 Ch. 1. Affine schemes

which satisfy the axioms of co-associativity, left coinversion and left counit,
respectively, expressed in commutativity of the following diagrams:
∗
m ⊗ idA
A ⊗ AO ⊗ A o A ⊗O A
idA ⊗m ∗
m∗ (1.168)
∗
m
A⊗A o A
∗
i ⊗ idA
A⊗A o A ⊗O A
µ m∗ (1.169)
² u ∗
Ao Ko A

(the left vertical arrow is the multiplication µ : a ⊗ b 7→ a b in A, the left

horizontal arrow is given by 1 7→ 1).
∗
u ⊗ idA
A⊗A o Ao A ⊗O A
µ m∗ (1.170)
² idA
Ao A

(the left arrow in the top line is a 7→ 1 ⊗ a).

It goes without saying that this definition is dual to that from sec. 1.15.4,
and therefore the group structures on the K-scheme Spec A are in one-to-one
correspondence with the co-algebra structures on the K-algebra A.
Example. The homomorphisms m∗ , i∗ , u∗ for the additive group scheme
Ga = Spec Z[T ] are:

m∗ (T ) = T ⊗ 1 + 1 ⊗ T, i∗ (T ) = −T, u∗ (T ) = 0. (1.171)

Exercise. Write explicitly the homomorphisms m∗ , i∗ , u∗ for the Exam-

ples 1.15.2.
1.16. Appendix: The language of categories.
Representing functors
1.16.1. General remark. The language of categories is an embodiment
of a “sociological” approach to a mathematical object: A group, a manifold or
a space are considered not as a set with an intrinsic structure but as a member
of a community of similar objects.
The ‘structural’ and ‘categorical’ descriptions of an object via the functor
it represents are complementary. The role of the second description increases
nowadays in various branches of mathematics (especially in algebraic geom-
etry) 26) although its richness of content was first demonstrated in topology,
by K[Π, n] spaces.
The proposed gist of definitions and examples purports to be an abridged
phraseological dictionary of the language of categories (ordered logic-wise,
however, not alphabetically). 27)
1.16.2. Definition of categories. A category C is a collection of the fol-
lowing data:
a) A set Ob C whose elements are called objects of C. (Instead of X ∈ Ob C
we often write briefly X ∈ C.)
b) For every ordered pair X, Y ∈ C, there is given a (perhaps empty) col-
lection (a set or a class) HomC (X, Y ) (or shortly Hom(X, Y )) whose elements
are called morphisms or arrows from X into ` Y . Notation: ϕ ∈ Hom(X, Y )
is expressed as ϕ : X 7→ Y and Mor C = Hom(X, Y ) is a collection
X,Y ∈C
of morphisms.
c) For every ordered triple X, Y, Z ∈ C, there is given a map

Hom(X, Y ) × Hom(Y, Z) → Hom(X, Z) (1.172)

assigning to morphisms ϕ : X → Y and ψ : Y → Z a morphism ψϕ called

their composition.
The data a)–c) should satisfy the following two axioms:
Associativity: (χψ)ϕ = χ(ψϕ) for any ϕ : X → Y , ψ : Y → Z, χ : Z → V .
Identity morphisms: for every X ∈ C, there exists a morphism
idX : X → X such that idX ◦ϕ = ϕ, ψ ◦ idX = ψ whenever the composi-
tions are defined. (Clearly, idX is uniquely defined.)
A morphism ϕ : X → Y is called an isomorphism if there exists a morphism
ψ : Y → X such that ψϕ = idX , ϕψ = idY .
Given two categories C and D such that Ob C ⊂ Ob D and

HomC (X, Y ) = HomD (X, Y ) for any X, Y ∈ Ob C (1.173)

26
To say nothing of supersymmetric theories of theoretical physics, where the lan-
guage of representable functors is a part of the working language (albeit often
used sub- or un-consciously).
27
An excellent textbook is [McL]. New trends are reflected in [GM], see also [Ke];
see also Molotkov’s paper [MV] on glutoses — generalized toposes).
88 Ch. 1. Affine schemes

and, moreover, such that the compositions of morphisms in C and in D coin-

cide, C is called a full sub-category of D.
A category C is said to be small if Ob C and collections HomC (X, Y ) are
sets. A category C is said to be big if Mor C is a proper class 28) and, for any
objects X, Y ∈ C, the class of morphisms HomC (X, Y ) is a set.
Sometimes, e.g., if we wish to consider a category of functors on a cat-
egory C, such a category is impossible to define if Ob C is a class; on the
other hand, in the framework of small categories we cannot consider, say, the
category of “all” sets, which is highly inconvenient.
P. Gabriel suggested a way out of this predicament: He introduced a uni-
versum, a large set of sets stable under all the operations needed; hereafter
the categories will be only considered belonging to this universum. For a list
of axioms a universum should satisfy see, e.g., Gabriel’s thesis [Gab]. We will
also assume the existence of a universum.
However, at the present state of foundations of mathematics and its consistency
the whole problem seems to me an academic one. My position is close to that of an
experimental physicists not apt to fetishize nor destroy his gadgets while they bring
about the results.
Nicolas Bourbaki’s opinion [Bb2] on this occasion reflects once again his Gaullean
common sense and tolerance:
“Mathematicians, it seems, agree that there is but a slight resemblance between
our ‘intuitive’ perceptions of sets and numbers and the formalisms purported to
describe them. The disputes only reflect the preference of choice.”
1.16.3. Examples. (Some of these examples are defined later in this book.)
For convenience we have grouped them:
The first group of examples. The objects in this series of examples are sets
endowed with a structure, the morphisms are maps of these sets preserving
this structure. (A purist is referred for a definition of a structure to, e.g.,
[EM].)
• Sets, or Ens for French-speaking people, the category of sets and their
maps;
• Top, the category of topological spaces and their continuous maps;
• Gr, the category of groups and their homomorphisms;
28
By definition a class X is said to be a set if there exist Y such that X ∈ Y .
Classes appear in BG (Bernais-Gedel (or Morse)) set theory; in the original ZF
(Zermelo-Frenkel) set theory only sets exist. But, if one adds to ZF an axiom
of existence for any set X of a universum U , containing X (as Grothendieck
did in [EGA, Gab]), then BG theory is modeled inside ZF with this additional
axiom: The sets of BG are modeled by sets X of ZF+Bourbaki belonging to some
universum U big enough to contain the set of natural numbers, whereas classes
of BG are subsets of U .
But what is a universum? It is any set closed with respect to ordinary set-
theoretic operations: Unions of (a family of) sets, intersections, power set, etc.
(for precise definition, see [EGA, Gab]).
1.16 Appendix: The language of categories. Representing functors 89

• Grf , the subcategory of Gr, whose objects are finite (as the subscript
indicates) groups;
• Ab, the subcategory of Gr of Abelian groups;
• Aff Sch, the category of affine schemes and their morphisms;
• Rings, the category of (commutative) rings (with unit) and their unit
preserving homomorphisms;
• A-Algs, the category of algebras over an algebra A;
• A-Mods, the category of (left) A-modules over a given algebra A;
• Man, the category of manifolds and their morphisms.
The second group of examples. The objects in this series of examples
are still structured sets but the morphisms are no longer structure-preserving
maps of these sets. (No fixed name for some of these categories).
• The main category of homotopic topology: Its object are topological
spaces, the morphisms are the homotopy classes of continuous maps (see, e.g.,
[FFG]).
• Additive relations (See, e.g., [GM]): Its objects are Abelian groups.
A morphism f : X → Y is any subgroup of X × Y and the composition
of ϕ : X → Y and ψ : Y → Z is given by the relation

ψϕ ={(x, z) ∈ X × Z | there exists y ∈ Y such that

(1.174)
(x, y) ∈ ϕ, (y, z) ∈ ψ}.

(No fixed name for this category).

The third group of examples. This group of examples is constructed from
some classical structures that are sometimes convenient to view as categories.
• For a (partially) ordered set I, the category C(I) is given by Ob C(I) = I,
where ½
one element if X ≤ Y ,
HomC(I) (X, Y ) = (1.175)
∅ otherwise.
The main example: I is the set of indices of an inductive system of groups. 29)
29
Let (I, ≤) be a directed poset (not all authors require I to be directed, i.e., to
be nonempty together with a reflexive and transitive binary relation ≤, with the
additional property that every pair of elements has an upper bound). Let (Gi )i∈I
be a family of groups and
1. we have a family of homomorphisms fij : Gi −→ Gj for all i ≤ j with the
following properties:
2. fii is the unit in Gi ,
3. fik = fij ◦ fjk for all i ≤ j ≤ k.
Then the set of pairs (Gi , fij )i,j∈I is called an inductive system (or direct system)
of groups and morphisms over I.
If in (1) above we have a family of homomorphisms fij : Gj −→ Gi for all i ≤ j
(note the order) with the same properties, then the set of pairs (Gi , fij )i,j∈I is
called an inverse system (or projective system) of groups and morphisms over I.
90 Ch. 1. Affine schemes

• Let X be a topological space. Denote by TopX the category whose objects

are the open sets of X and morphisms are their natural embeddings. This
trivial reformulation conceives a germ of an astoundingly deep generalization
of the notion of a topological space, the Grothendieck topologies or toposes,
see [J]. 30)
• Categories associated with a scheme of a diagram.
A scheme of a diagram is (according to Grothendieck) a triple consisting
of the two sets V (vertices), A (arrows) and the map e : A → V × V that
to every arrow from A assigns its ends, i.e., an ordered pair of vertices: The
source and the target of the arrow.

• /•d • // • k
•? +•
?? ÄÄ
/• ?? Ä
• ?? ÄÄÄ
? ÄÄÄ
•

Fig. 13

Let (V, A, e) be a graph. Define the category D = D(V, A, e) setting

Ob D = V , HomD (X, Y ) = {paths from X to Y along the arrows}.
More precisely, if X 6= Y , then any element from HomD (X, Y ) is a sequence
f1 , . . . , fk ∈ A such that the source of f1 is X, the target of fi is the source
of fi+1 and the target of fk is Y . If X = Y , then HomD (X, X) must contain
the identity arrow. The compositions of morphisms is defined in an obvious
way as the composition of passes.
• Define the category DC = DC (V, A, e) setting Ob DC = V and
½
one element if HomD (X, Y ) 6= ∅
HomDC (X, Y ) = (1.176)
∅ otherwise.

Intuitively, DC is the category corresponding to a commutative diagram:

All the paths from X to Y define the same morphism.
1.16.4. Examples of constructions of categories. There are several
useful formal constructions of new categories. We will describe here only three
of them.
1.16.4a. The dual category. Given a category C, define its dual C◦ setting
Ob C◦ to be a copy of Ob C and HomC ◦ (X ◦ , Y ◦ ) to be in 1-1 correspondence
with HomC (Y, X), where X ◦ ∈ Ob C◦ denotes the object corresponding to
X ∈ Ob C, so that if a morphism ϕ◦ : X ◦ → Y ◦ corresponds to a morphism
ϕ : Y → X, then ψ ◦ ϕ◦ = (ϕψ)◦ and idX ◦ = (idX )◦ .
30
A. Rosenberg used it to construct spectra of noncommutative rings in his non-
commutative books [R1], [R2].
1.16 Appendix: The language of categories. Representing functors 91

Speaking informally, C◦ is obtained from C by taking the same objects but

inverting the arrows.
This construction is interesting in two opposite cases. If C◦ highly resem-
bles C, e.g., is equivalent (the definition of equivalence of categories will be
given a little later) to C (as is the case of, say Abf , the category of finite
Abelian groups), this situation provides us with a stage where different dual-
ity laws perform.
Conversely, if C◦ is very un-similar to C, then the category C◦ might have
new and nice properties as compared with C; e.g., for C = Rings, the dual
category C◦ = Aff Sch has “geometric” properties enabling us to glue global
objects from local ones — the operation appallingly unnatural and impossible
even to imagine inside Rings.
1.16.4b. The category of objects over a given base. Given a category C
and its object S, define the category CS by setting Ob CS = {ϕ ∈ HomC (X, S)}.
For any ϕ : X → S and ψ : Y → S, define:

HomCS (ϕ, ψ) = {χ ∈ HomC (X, Y ) | ϕ = ψχ}, (1.177)

i.e., HomCS (ϕ, ψ) is the set of commutative diagrams

χ
X@ /Y
@@ ÄÄ
@@ Ä (1.178)
ϕ @@ ÄÄĵ
 ÄÄ
S

The composition of morphisms in CS is induced by the composition in C.

Examples. 1) RingsR = R-Algs, the category of R-algebras, where R is a fixed
ring;
2) VebunM , the category of vector bundles over a fixed base M ; this is a
subcategory of the category BunM whose fibers are arbitrary manifolds;
3) the category (CS )◦ that deals with morphisms S → X for a fixed S ∈ C.
1.16.4c. The product of categories.
Q Given a family of categories Ci ,
where i ∈ I, define their product Ci by setting
i∈I

Q Q
Ob Ci = Ob Ci ;
³Q Q ´ Q (1.179)
HomQ Ci Xi , Yi = HomCi (Xi , Yi )

with coordinate-wise composition.

1.16.5. Functors. A covariant functor, or just a functor, F : C → D from
a category C to a category D is a collection of maps (F, FX,Y ), usually abbre-
viated to F , where F : Ob C → Ob D and

FX,Y : HomC (X, Y ) → HomD (F (X), F (Y )) for any X, Y ∈ C, (1.180)

92 Ch. 1. Affine schemes

such that FX,Z (ϕψ) = FY,Z (ϕ)FX,Y (ψ) for any ϕ, ψ ∈ Hom C provided ϕψ is
defined.
A functor from C◦ into D is called a contravariant functor from C into D.
A functor F : C1 × C2 → D is called a bifunctor, and so on.
Given categories C, D, E and functors F : C → D and G : D → E, define
GF : C → E composing the constituents of F and G in the usual set-theoretical
sense.
The most important examples of functors are just “natural constructions”:
(Co)homology and homotopy are functors Top → Ab; characters of finite
groups constitute a functor Grf → Rings. These examples are too meaningful
to discuss them in passing.
1.16.6. Examples of presheaves. A presheaf of sets (groups, rings, alge-
bras, superalgebras, R-modules, and so on) is a contravariant functor from
the category TopX of open sets of a topological space X with values in Sets
(or Gr, Rings, Algs, Salgs, R-Mods, and so on, respectively).
Let (V, A, e) be a scheme of a diagram, D and DC the associated categories.
A functor from D into a category E is called a diagram of objects from E (of
type (V, A, e)). A functor from DC into E is a commutative diagram of objects.
For an ordered set I considered as a category, a functor from I into a cat-
egory C is a family of objects from C indexed by the elements from I and
connected with morphisms so that these objects constitute either a projective
or an inductive system in C.
Given two functors F, G : C → D, a functor morphism f : F → G is a set
of morphisms f (X) : F (X) → G(X) (one for each X ∈ C) such that for any
ϕ ∈ HomC (X, Y ) the following diagram commutes:
f (X)
F (X) / G(X)

F (ϕ) G(ϕ) (1.181)

² f (Y ) ²
F (Y ) / G(Y ).

The composition of functor morphisms is naturally defined.

A functor morphism f is called a functor isomorphism if f (ϕ) ∈ Mor D are
isomorphisms for all X ∈ C. The functors from C to D are objects of a category
denoted by Funct(C, D).
A functor F : C → D is called an equivalence of categories if there exists
a functor G : D → C such that GF ∼ = idC , F G ∼
= idD and in this case C is said
to be equivalent to D.
Examples. 1) (Abf )◦ ∼ = Abf ; G ←→ X(G), the character group of G;
2) Rings◦ is equivalent to the category Aff Sch of affine schemes.
1.16.7. The category C∗ = Funct(C◦ , Sets). If our universum is not too
large, there exists a category whose objects are categories and morphisms are
functors between them. The main example: The category C∗ = Funct(C◦ , Sets)
of functors from C◦ into Sets.
1.16 Appendix: The language of categories. Representing functors 93

1.16.8. Representable functors. Fix any X ∈ C.

1) Denote by PX ∈ C∗ (here: P is for point; usually this functor is denoted
by hX , where h is for homomorphisms) the functor given by

PX (Y ◦ ) = HomC (Y, X) for any Y ◦ ∈ C◦ ; (1.182)

to any morphism ϕ◦ : Y2◦ −→ Y1◦ the functor PX assigns the map of sets
PX (Y2◦ ) → PX (Y1◦ ) which sends ψ : Y2 −→ X into the composition
ϕψ : Y1 −→ Y2 → X.
To any ϕ ∈ HomC (X1 , X2 ), there corresponds a functor morphism
Pϕ : PX1 −→ PX2 which to any Y ∈ C assigns

Pϕ (Y ◦ ) : PX1 (Y ◦ ) −→ PX2 (Y ◦ ) (1.183)

and sends a morphism ψ ∈ HomC (Y ◦ , X1 ) into the composition

ϕψ : Y ◦ −→ X1 −→ X2 . (1.184)

Clearly, Pϕψ = Pϕ Pψ .
2) Similarly, define P X ∈ C∗ by setting

P X (Y ) := HomC (X, Y ) for any Y ∈ C; (1.185)

to any morphism ϕ : Y1 → Y2 , we assign the map of sets P X (Y1 ) −→ P X (Y2 )

which sends ψ : X −→ Y1 into the composition ψϕ : X −→ Y1 −→ Y2 .
To any ϕ ∈ HomC (X1 , X2 ), there corresponds a functor morphism
P ϕ : P X2 −→ P X1 which to any Y ∈ C assigns

P ϕ (Y ) : P X2 (Y ) −→ P X1 (Y ) (1.186)

and sends a morphism ψ ∈ HomC (X2 , Y ) into the composition

ψϕ : X1 −→ X2 −→ Y. (1.187)

Clearly, P ϕψ = P ψ P ϕ .
A functor F : C◦ −→ Sets (resp. a functor F : C −→ Sets) is said to be
representable(resp. corepresentable if it is isomorphic to a functor of the form
PX (resp. P X ) for some X ∈ C; then X is called an object that represents F .
1.16.8a. Theorem. 1) The map ϕ 7→ Pϕ defines an isomorphism of sets

HomC (X, Y ) ∼
= HomC∗ (PX , PY ). (1.188)

This isomorphism is functorial in both X and Y . Therefore, the functor

P : C → C∗ determines an equivalence of C with the full subcategory of C∗
consisting of representable functors.
2) If a functor from C∗ is representable, the object that represents it is
determined uniquely up to an isomorphism.
The above Theorem is the source of several important ideas.
94 Ch. 1. Affine schemes

1.16.8b. The first direction. It is convenient to think of P X as of “the

sets of points of an object X ∈ C with values in various objects Y ∈ C or
Y -points”; notation: P X (Y ) or sometimes X(Y ). (The sets PX (Y ) are also
sometimes denoted by X(Y` ).)
In other words, PX = PX (Y ) with an additional structure: The sets
Y ∈C
of maps PX (Y1 ) −→ PX (Y2 ) induced by morphisms Y2◦ −→ Y1◦ for any
Y1 , Y2 ∈ C and compatible in the natural sense (the composition goes into
the composition, and so on). The situation with the P X is similar.
Therefore, in principle, it is always possible to pass from
the categorical point of view to the structural one, since
all the categorical properties of X are mirrored precisely
by the categorical properties of the structure of PX .
Motivation. Let ∗ be a one-point set. For categories with sufficiently “sim-
ple” structure of their objects, such as the category of finite sets or even
category of smooth finite dimensional manifolds, X = PX (∗) for every object
X, i.e., X is completely determined by its ∗-points or just points.
For varieties (or for supermanifolds), when the object may have either
“sharp corners” or “inner degrees of freedom”, the structure sheaf may contain
nilpotents or zero divisors, and in order to keep this information and be able to
completely describe X we need various types of points, in particular, Y -points
for some more complicated Y ’s.
1.16.8c. The second direction. Replacing X by PX (resp. by P X ) we may
transport conventional set-theoretical constructions to any category: An ob-
ject X ∈ C is a group, ring, and so on in the category C, if the corresponding
structure is given on every set PX (Y ) of its Y -points and is compatible with
the maps induced by the morphisms Y2◦ −→ Y1◦ (resp. Y1 → Y2 ). 31)
1.16.8d. The third direction. Let C be a category of structures of a given
type. Among the functors C◦ → Sets, i.e., among objects from C∗ , some nat-
ural functors may exist which a priori are constructed not as PX or P X but
eventually prove to be representable. (Examples: The functor X 7→ H ∗ (X; G)
on the homotopy category; C 7→ GL(m|n; C) for any supercommutative su-
peralgebra C.)
In such cases, it often turns out that the properties of the functor repre-
sentable by such an object are exactly the most important properties of the
object itself and its structural description only obscures this fact.
It may well happen that some natural functors C → Sets are not repre-
sentable, though it is highly desirable that they would have been. The most
frequent occurrence of such a situation is when we try to generalize to C some
set-theoretical construction, e.g., factorization modulo the group action or
31
This is exactly the way supergroups are defined and superalgebras should be de-
fined. However, it is possible to define superalgebras using just one set-theoretical
model and sometimes we have to pay for this deceiving simplicity.
1.16 Appendix: The language of categories. Representing functors 95

modulo a more general relation. In such a case it may help to add to C,

considered embedded into C∗ , the appropriate functors.
In algebraic geometry of 1970s this way of reasoning lead to monstrous
creatures which B. Moisheson called minischemes, M. Artin étale schema,
and A. Grothendieck just varietées 32) and lately to stacks.
1.16.9. Proof of Theorem 1.16.8. Let us construct a map

i : HomC∗ (PX , PY ) → HomC (X, Y ) (1.189)

which to every functor morphism PX → PY assigns the image of idX ∈ PX (X) in

PY (X) under the map PX (X) → PY (X) defined by this functor morphism. Let us
verify that i and ϕ 7→ Pϕ are mutually inverse.
1) i(Pϕ ) = Pϕ (idX ) = ϕ by definition of Pϕ .
2) Conversely, given a functor morphism g : PX → PY , we have by definition the
maps g(Z) : PX (Z) → PY (Z) for all Z ∈ Ob C. By definition i(g) = g(X)(idX ) and
we have to verify that Pi(g) (Z) = g(Z).
By definition, Pi(g) (Z) to every morphism g : Z → X assigns the composition
i(g) ◦ ϕ : Z → X → Y , and therefore we have to establish that g(Z)ϕ = i(g) ◦ ϕ.
Now, let us use the commutative diagram

g(x)
PX (X) / PY (X)

PX (ϕ) PY (ϕ) (1.190)

² g(z) ²
PX (Z) / PY (Z).

Thus, we have verified that Im P is a full subcategory of C∗ , and therefore is

equivalent to C. The remaining statements are easily established. u
t
It is worth mentioning that if we add representable functors to the full
subcategory of functors PX from C∗ , i.e., if we add functors isomorphic to the
ones the category already possesses, we obtain an equivalent category.
1.16.10. The object of inner homomorphisms in a category. All mor-
phisms of a set into another set constitute a set; morphisms of an Abelian
group into another Abelian group constitute an Abelian group, there are many
more similar examples. A natural way to axiomize the situation when for
X, Y ∈ Ob C there is an object of inner homomorphisms Hom(Y, X) ∈ Ob C,
is to define the corresponding representable functor Hom. Quite often it is
determined by the formula

HomC (X, Hom(Y, Z)) = HomC (X ∗ Y, Z) (1.191)

for a convenient operation ∗. 33)

32
In supermanifold theory, this is one of the ways to come to point-less (or, perhaps
more politely, point-free) supermanifolds.
33
For examples of categories with interesting objects of inner homomorphisms, we
can take the categories C of superspaces, superalgebras, supermanifolds and su-
pergroups.
1.17. Solutions of selected problems of Chapter 1
Exercise 1.3.4. 1) Since a1 . . . an ⊂ a1 ∪ . . . ∪ an , it follows that
V (a1 . . . an ) ⊃ V (a1 ∪ . . . ∪ an ). The other way round, let px be a prime
ideal. Then
\
px ∈ V (a1 . . . an ) ⇐⇒ x ∈ V (ai ) −→ x ∈ V (ai ) for some i, (1.192)
i

hence, px ⊃ ai ⊃ a1 ∪ . . . ∪ an for this i, i.e., x ∈ V (a1 ∪ . . . ∪ an ).

2) By Lemma 1.4.1 we have:
n
\
(g1 , . . . , gn ) = A ⇐⇒ V (gi ) = ∅.
i=1

Besides, V (g m ) = V (g) for m > 0 implying the desired

Exercise 1.6.7. 3) Let f /g, where f, g ∈ A, be such that

(f /g)n + an−1 (f /g)n−1 + · · · + a0 = 0, where ai ∈ A.

Then f n + f an−1 f n−1 + · · · + g n a0 = 0 implying g|f n and if (g, f ) = 1, then

g is invertible in A. Therefore, f /g ∈ A.
Chapter 2

Sheaves, schemes, and projective spaces

2.1. Basics on sheaves

The topological space Spec A is by itself a rather coarse invariant of A, see
examples 1.5.3. Therefore it is natural to take for a “right” geometric object
corresponding to A, the pair (Spec A, A)e consisting of the space Spec A and
the set of elements of A considered, more or less adequately, as functions on
Spec A, and so we did up to now.

Fig. 1

But, on the other hand, only local geometric objects are associated with
rings; so in order to construct, say, a projective space we have to glue it from
affine ones. Let us learn how to do so.
The gluing procedure we are interested in may be described, topologically,
as follows: Let X1 , X2 be two topological spaces, U1 ⊂ X1 , U2 ⊂ X2 their
open subsets and f : U1 −→ U2 an isomorphism. Then we construct a set

X = (X1 ∪ X2 )/Rf ,
98 Ch. 2. Sheaves, schemes, and projective spaces

where Rf is the equivalence relation which identifies the points that corre-
spond to each other with respect to f . On X, a natural topology is induced,
and we say that X is the result of gluing X1 with X2 by means of f .
On Fig. 1, the two ways to glue two affine lines, X1 = X2 = R with
U1 = U2 = R \ {0}, are illustrated. The results are:
a) The line with a double point (here f = id);
b) P1 , the projective line (here f (x) = x−1 ). Clearly, P1 is homeomorphic to
the circle S 1 .
When we try to apply this construction to the spectra of rings we immedi-
ately encounter the above-mentioned circumstance, namely that the topologi-
cal structure of the open sets reflects but slightly the algebraic data which we
would like to preserve and which is carried by the ring A itself. The theories
of differentiable manifolds and analytic varieties suggest a solution.
In order to glue a differentiable manifold from two balls U1 and U2 , we re-
quire that the isomorphism f : U1 −→ U2 which determines the gluing should
not be just an isomorphism of topological spaces but should also preserve
the differentiable structure. This means that the map f ∗ , sending continuous
functions on U2 into continuous functions on U1 , must induce an isomor-
phism of subrings of differentiable functions, otherwise the gluing would not
be “smooth”.
The analytic case is similar.
Therefore we have to consider the functions of a certain class which are
defined on various open subsets of the topological space X.
The relations between the functions on different open sets are axiomized
by the following definition.
2.1.1. Presheaves. Fix a topological space X. Let P be a law that to every
open set U ⊂ X assigns a set P(U ) and, for any pair of open subsets U ⊂ V ,
V
there is given a restriction map rU : P(V ) −→ P(U ) such that
1) P(∅) consists of one element,
W V
2) rU = rU ◦ rVW for any open subsets (briefly: opens) U ⊂ V ⊂ W .
V
Then the system {P(U ), rU | U, V are opens} is called a presheaf (of sets)
on X.
The elements of P(U ), also often denoted by Γ (U, P), are called the sections
of the presheaf P over U ; a section may be considered as a “function” defined
over U .
Remark. Axiom 1) is convenient in some highbrow considerations of category
theory. Axiom 2) expresses the natural transitivity of restriction.
2.1.2. The category TopX . The objects of TopX are open subsets
of X and morphisms are inclusions. A presheaf of sets on X is a functor
P : Top◦X −→ Sets.
2.1 Basics on sheaves 99

From genuine functions we can construct their products, sums, and mul-
tiply them by scalars; similarly, we may consider presheaves of groups, rings,
and so on. A formal definition is as follows:
Let P be a presheaf of sets on X; if, on every set P (U ), there is given
an algebraic structure (of a group, ring, A-algebra, and so on) and the re-
V
striction maps rU are homomorphisms of this structure, i.e., P is a functor
◦
TopX −→ Gr (Ab, Rings, A-Algs, and so on), then P is called the presheaf
of groups, rings, A-algebras, and so on, respectively.
Finally, we may consider exterior composition laws, e.g., a presheaf of mod-
ules over a presheaf of rings (given on the same topological space).
2.1.2a. Exercise. Give a formal definition of such exterior composition
laws.
2.1.3. Sheaves. The presheaves of continuous (infinitely differentiable, ana-
lytic, and so on) functions on a space X possess additional properties (of “an-
alytic continuation” type) which are axiomized in the following definition.
A presheaf P on a topological space X is called a sheaf if it satisfies
S the
following condition: For any open subset U ⊂ X, its open cover U = Ui ,
i∈I
and a system of sections si ∈ P(Ui ), where i ∈ I, such that
Ui Uj
rU i ∩Uj
(si ) = rUi ∩Uj
(sj ) for any i, j ∈ I, (2.1)

U
there exists a section s ∈ P(U ) such that si = rUi
(s) for any i ∈ I, and such
a section is unique.
In other words, from a set of compatible sections over the Ui a section
over U may be glued and any section over U is uniquely determined by the
set of its restrictions onto the Ui .
Remark. If P is a presheaf of Abelian groups, the following reformulation
of the above condition is useful: S
A presheaf P is a sheaf if, for any U = Ui , the following sequence
i∈I
of Abelian groups is exact
ϕ Y ψ Y
0 −→ P(U ) −→ P(Ui ) −→ P(Ui ∩ Uj ), (2.2)
i∈I i,j∈I

where ϕ and ψ are determined by the formulas

U
ϕ(s) = (. . . , rU , (s), . . .)
i
U
(2.3)
Ui
ψ(. . . , si , . . . , sj , . . .) = (. . . , rUi ∩Uj
(si ) − rUij∩Uj (sj ), . . .).

For a generic presheaf of Abelian groups, this sequence is only a complex.

(Its natural extension determines a Čech cochain complex that will be defined
in what follows.)
100 Ch. 2. Sheaves, schemes, and projective spaces

2.1.4. The relation between sheaves and presheaves. The natural ob-
jects are sheaves but various constructions with them often lead to presheaves
which are not sheaves.
Example. (This example is of a fundamental importance for the cohomology
theory). Let F1 and F2 be two sheaves of Abelian groups and F1 (U ) ⊂ F2 (U )
with compatible restrictions, i.e., F1 ⊂ F2 . As is easy to see, the set of groups
P(U ) = F1 (U )/F2 (U ) and natural restrictions is a presheaf but, in general,
not a sheaf. Here is a particular case:
Let X be a circle, O the sheaf over X for which O(U ) is the group of R-
valued continuous functions on U , and Z e ⊂ O the “constant” presheaf for
e ) = Z for each non-empty U .
which Z(U
The presheaf for which
e ) for every open set U ∈ X
P(U ) = O(U )/Z(U (2.4)

is not a sheaf by the following reason.

Consider two connected open sets U1 , U2 ⊂ X such that X = U1 ∪ U2 and
U1 ∩ U2 is the union of two connected components V1 , V1 (e.g., U1 , U2 are
slightly enlarged half-circles). Let f1 ∈ O(U1 ), f2 ∈ O(U2 ) be two continuous
functions such that

rVU11 (f1 ) = rVU12 (f2 ) = 0, rVU21 (f1 ) − rVU22 (f2 ) = 1. (2.5)

e 1 ) and f2 mod Z ∈ O(U2 )/Z(U

Then the classes f1 mod Z ∈ O(U1 )/Z(U e 2 ) are
compatible over U1 ∩ U2 .
On the other hand, the sheaf O over X has no section whose restrictions
onto U1 and U2 are f1 mod Z and f2 mod Z, respectively, since it is impossible
to remove the incompatibility on V2 . The cause is, clearly, in the non-simple-
connectedness of the circle.

Fig. 2
2.1 Basics on sheaves 101

2.1.5. Gluing sheaves from presheaves. With every presheaf P there

is associated a certain sheaf P+ . We construct the sections of P+ from the
sections of P with the help of two processes.
The first one diminishes the number of sections in P(U ) identifying those
which start to coincide being restricted onto a sufficiently fine cover U = ∪ Ui .
i
The second process increases the number of sections in P(U ) adding the
sections glued from the compatible sets of sections on the coverings of U .
Clearly, a passage to limits is required. A notion, technically convenient
here and in other problems, is the following one. Let P be a presheaf over X.
The stalk Px of P over a point x ∈ X is the set lim P(U ), where the inductive
−→
limit 1) is taken over the directed system of open neighborhoods of x.
The elements from Px are called the germs of sections over x. A germ
is an equivalence class in the set of sections over different open neighbor-
hoods U containing x modulo the equivalence relation which identifies sec-
tions s1 ∈ P(U1 ) and s2 ∈ P(U2 ) if their restriction onto a common subset
U3 ⊂ U1 ∩ U2 containing x coincide.
For any point x and an open neighborhood U 3 x, a map rxU : P(U ) −→ Px
is naturally defined.
Clearly, Px carries the same structure as the sets P(U ); i.e., is a group,
ring, module, and so on, provided P is a presheaf of groups, rings, modules,
and so on.
The idea behind constructing a sheaf P+ from a presheaf P is that we define
the sections, i.e.,Qelements of P+ (U ), as certain sets of germs of sections, i.e.,
as elements of Px , which are compatible in a natural sense:
x∈U
Let P be a presheaf on a topological space
Q X. For every nonempty open
subset U ⊂ X, define the subset P+ (U ) ⊂ Px as follows:
x∈U

1
Recall the definition (see. [StE]). Let the objects of a category be sets with alge-
braic structures (such as groups, rings, modules (over a fixed ring), and so on)
and morphisms be the morphisms of these structures. Let (I, ≤) be an inductive
set. Let {Ai | i ∈ I} be a family of objects enumerated by elements of I and
for all i ≤ j, let a family of homomorphisms fij : Ai −→ Aj be given having the
following properties:
a) fii = id |{Ai } ,
b) fik = fjk ◦ fij for all i ≤ j ≤ k.
Then the pair (Ai , fij ) is said to be directed (inductive) system over I. The set
of inductive (or direct) limit A of the inductive system (Ai , fij ) is defined as the
quotient of the disjoint union of the sets A0i modulo an equivalence relation ∼:
a .
lim Ai = Ai ∼,
−→
i

were the equivalence relation ∼ is defined as follows: If xi ∈ Ai and xj ∈ Aj , then

xi ∼ xj if, for some k ∈ I, we have fik (xi ) = fjk (xj ).
102 Ch. 2. Sheaves, schemes, and projective spaces
 
 (. . . , sx , . . .) | for any x ∈ U, there exists 
P+ (U ) = its open neighborhood V ⊂ U and a section s ∈ P(V ) (2.6)
 
such that sy = ryV (s) for any y ∈ V.

Further, for any pair of open subsets V ⊂ U , define Q

the restriction
Q map
P+ (U ) −→ P+ (V ) as the one induced by the projection Px −→ Px .
x∈U x∈V
+
2.1.6. Theorem. The family of sets P (U ) together with the described re-
striction maps is a sheaf on X and P+
x = Px for any x ∈ X.

The sheaf P+ is called the sheaf associated with P. Clearly, the algebraic
structures are transplanted from P to P+ .
Proof is a straightforward verification of definitions; so we leave it to the
reader.
2.1.7. Another definition of sheaves. Being equivalent to the previous
one, given in sec. 2.1.3, it is sometimes more convenient to grasp intuitively.
A sheaf F over a topological space X is a pair (YF , r), where YF is a topo-
logical space and r : YF −→ X is an open continuous map onto X such that,
for any y ∈ YF , there exists its open neighborhood (in YF ) and r is a local
homeomorphism in this neighborhood.
The relation between this definition and the previous one is as follows.
Giving r : YF −→ X, we define a sheaf as in sec. 2.1.3: By setting F(U ) to be
the set of local sections of YF over U , i.e., the maps s : U −→ YF such that
r ◦ s = id |U .
Conversely, if F is given by its sections over U for all U ⊂ X, set
[
YF = Fx , (2.7)
x∈X

let r be the map r : Fx −→ x. Define the topology on YF considering various

sections s ∈ F(U ) as open sets, and set
[ [ [
s= rxU (s) ⊂ Fx ⊂ Fx (2.8)
x∈U x∈U x∈X

One should bear in mind that, even for Hausdorff spaces X, the spaces YF
are not Hausdorff, generally.
Fig. 3 depicts a part of the space YF corresponding to the sheaf of continu-
ous functions on the segment [0, 1]. To the graph of every continuous function
an open subset of YF corresponds and the sections corresponding to the graphs
of functions do not intersect in YF unless their graphs intersect.
If the functions f1 , f2 coincide over a (necessarily closed !) set Y , then the
corresponding sections of YF coincide over the interior of Y . In the space YF ,
the points over a (or b) belonging to sections 3 and 4 are distinct, but any
two neighborhoods of these points intersect. u
t
2.2 The structure sheaf on Spec A 103

Fig. 3

2.2. The structure sheaf on Spec A

We consider the elements from A as functions on Spec A. Now, if we wish to
consider restrictions of these functions, we should construct from them func-
tions with lesser domains. The only such process which does not require limits
is to consider quotients f /g since f (x)/g(x) makes sense for all x ∈
/ V (g). This
suggests the following definition.
2.2.1. The case of rings without zero divisors. Let A be a ring without
zero divisors, X = Spec A, K the field of fractions of A. For any nonempty
open subset U ⊂ X, denote:

OX (U ) = {a ∈ K | a can be represented in the form f /g,

(2.9)
where g(x) 6= 0 for any x ∈ U }.
V
If U ⊂ V denote by rU the obvious embedding OX (V ) ,→ OX (U ).
2.2.1a. Theorem. 1) The above-defined presheaf OX is a sheaf of rings.
2) The stalk of OX at x is

Ox = {f /g | f, g ∈ A, g ∈
/ px } (2.10)

and Γ (D(g), OX ) = {f /g n | f ∈ A, n ≥ 0} for any nonzero g ∈ A.

Remarks. Heading 2) is less trivial. Intuitively it reflects two things:
2a) If an “algebraic function” is defined at points where g 6= 0, then the
worst what may happen with it on V (g) is a pole of finite order: It has no
essentially singular points;
2b) over big open sets there is no need to consider “glued” sections: They
are given by one equation.
On the other hand, the description of the stalks Ox enables us to identify
Spec A with a set of subrings of the field K. This is the point of view pursued
in works by Chevalley and Nagata.
104 Ch. 2. Sheaves, schemes, and projective spaces

Proof. The fact that OX is a sheaf is established by trivial reference to defini-

tions, and Ox is similarly calculated. The reason for that is that all the Ox are
embedded into K, so that the relations of “extension” and “restriction” are
induced by the identity relation in K. (In what follows, we will give a longer
definition suitable for any rings, even when K, the localization of A, does not
exist.)
Here we will prove heading (2) only for g = 1 to illustrate the main idea.
The general case is treated in the following subsection.
Therefore we would like to show that if an element of the quotient field
K can be represented as a rational fraction so that its denominator does not
belong to any prime ideal of A, then this element belongs to A, i.e., there is
no denominator at all. This is obvious for a unique factorization ring.
In the general case, the arguments (due to Serre) are suggested by analogy
with differentiable manifolds and use the “partition of unity”.
Let f ∈ Γ (X, OX ) ⊂ K. For any point x ∈ Spec A, set f = gx /hx ,
where gx , hx ∈ A and hx ∈ / px . Let Ux = D(hx ). Obviously, Ux is an open
neighborhood of x.
Now,
S apply Proposition 1.4.13 on quasi-compactness to the cover
X = D(hx ).
x
LetPxi ∈ X, and ai ∈ A for T
i = 1, . . . , n. Set hi = hxi and gi = gxi .
1= ai hi in A. Then X = D(hi ) and
1≤i≤n 1≤i≤n
X X X
f= a i hi f = ai hi (gi hi ) = ai gi ∈ A. u
t (2.11)
1≤i≤n 1≤i≤n 1≤i≤n

Let us give an example of calculation of OX (U ) for open sets other than

the big ones.
2.2.2. Example. In Spec K[T1 , T2 ], where K is a field, let U = D(T1 )∪D(T2 ).
Therefore U is the complement to the origin. Since K[T1 , T2 ] has no zero di-
visors, we obtain
OX (U ) = K[T1 , T2 , T1−1 ] ∩ K[T1 , T2 , T2−1 ]. (2.12)
By the unique factorization property in the polynomial rings we immediately
see that OX (U ) = K[T1 , T2 ], and therefore a function on a plane cannot have
a singularity supported at one closed point: The function is automatically
defined at it.
Similar arguments are applicable to a multidimensional affine space: If
n ≥ 2 and F1 , . . . , Fn ∈ K[T1 , . . . , Tn ] are relatively prime, then
³ [ ´
OX D(Fi ) = K[T1 , . . . , Tn ] : (2.13)
1≤i≤n

The support of the set of singularities of a rational function cannot be given

by more than one equation.
2.2 The structure sheaf on Spec A 105

2.2.3. The structure sheaf on Spec A: The general case. If A has zero
divisors, it has no quotient field. Therefore an algebraic formalism necessary
for a correct definition of quotient rings and relations between them becomes
more involved. Nevertheless, a sheaf is actually introduced in the same way
as for the ring without zero divisors and with the same results.
Constructing a sheaf on Spec A we have to study the dependence of AS on
S, and the ring homomorphisms AS −→ AS 0 for different S and S 0 . Theorem
1.6.4b stating a universal character of localization, is foundational in what
follows.
2.2.4. The structure sheaf OX over X = Spec A. For every x ∈ X, set
(see sec. 1.6.2)
Ox := AA\px = Apx .
For any open subset U ⊂ X, define the ring of sections of the presheaf OX
over U to be the subring Y
OX (U ) ⊂ Ox (2.14)
x∈U

consisting of the elements (. . . , sx , . . .), where sx ∈ Ox , such that for every

point x ∈ U , there exists an open neighborhood D(fx ) 3 x (here fx is a func-
tion corresponding to x) and an element g ∈ Afx such that sy is the image
of g under the natural homomorphism Afx −→ Oy for all y ∈ U .
V
Define the restriction
Q morphisms
Q rU as the homomorphisms induced by
the projection Ox −→ Ox . It is easy to see that OX is well-defined
x∈V x∈U
and the natural homomorphism Afx −→ Oy is induced by the embedding
of multiplicative sets
{fxn | n ∈ N} ⊂ A \ py . (2.15)
Theorem. The presheaf OX is a sheaf whose stalk over x ∈ X is isomorphic
to Ox and rxU is the composition
Y pr
OX (U ) −→ Ox0 −→ Ox (2.16)
x0 ∈U

Furthermore, the ring homomorphism

j : Af −→ OX (D(f )), j(g/f ) = (. . . , jx (g/f ), . . .)x∈U , (2.17)

where jx : Af −→ Ox is a natural homomorphism of quotient rings, is an iso-

morphism.
The sheaf OX over the scheme X = Spec A is called the structure sheaf
of X.
Proof. The fact that OX is a sheaf follows immediately from the definitions
and compatibility of natural homomorphisms of quotient rings. By definition,
the stalk OX,x of OX over x is equal to
106 Ch. 2. Sheaves, schemes, and projective spaces

lim OX (U ) = lim OX (D(f )), (2.18)

−→ −→
f (x)6=0

since the neighborhoods D(f ) constitute a basis of neighborhoods of x. The

natural homomorphisms Af −→ OX (D(f )) −→ Ox determine the homomor-
phism OX,x −→ Ox . It is an epimorphism, since any element from Ox is of the
form g/f , where f (x) 6= 0, and therefore is an image of the corresponding
element from OX (D(f )).
Moreover, the kernel of this homomorphism is trivial: If g/f goes into
0 ∈ Ox , then by Lemma 1.6.4a f1 g = 0 for some f1 such that f1 (x) 6= 0,
and therefore the image of g/f in Af f1 and, with even more reason, in the
inductive limit lim OX (D(f )) is zero.
−→
It remains to demonstrate the “furthermore” part of the theorem; notice
that it gives a “finite” description of rather cumbersome and bulky rings
OX (U ) for big open sets U .
First of all, Ker j = 0. Indeed, if j(g/f ) = 0, then, for every
point x ∈ D(f ), there exists an element tx such that tx (x) 6= 0 and
tx ∈ Ann g = {a | ag = 0}. But this means that Ann g 6⊂ px ; i.e., x ∈
/ V (Ann g)
for x ∈ D(f ); and therefore V (Ann g) ⊂ V (f ); i.e., f n ∈ Ann g for some n.
Hence, g/f = 0 in Af .
Now, let us prove that j is an epimorphism.
S Let s ∈ OX (D(f )) be a section.
By definition there exists a cover D(f ) = D(hx ) such that s is induced
x∈D(f )
over D(hx ) by an element gx /hx . As in sec. 2.2.1, construct a partition of
unity, or rather not of unity but of an invertible on D(f ) function f n :
\ µ \ ¶
V (f ) = V (hx ) = V {hx } (2.19)
x∈D(f ) x∈D(f )
P
implying f n = ax hx . Since ax = 0 for almost all x, we may denote the
x∈D(f )
incompressible decomposition by
[ X
D(f ) = D(hi ), where f n = ai hi , where ai = axi , hi = hxi
1≤i≤r 1≤i≤r
(2.20)
Now, consider our section s glued together from the gi /hi . The fact that
of gi /hi and gj /hj are compatible on D(hi ) ∩ D(hj ) = D(hi hj ) means that
the images of gi /hi and gj /hj coincide in all the rings Ox , where x ∈ D(hi hj ).
By the proved above gi /hi − gj /hj = 0 in Ahi hj ; i.e., for some m (which
may be chosen to be independent of indices since the cover is finite) we have
(hi hj )m (gi hj − gj hi ) = 0. Replacing hi with hm+1
i and gi with gi hm
i we may
assume that m = 0.
Now, the compatibility conditions take the form gi hj = gj hi . Therefore
X ³ X ´
f n gj = ai hi gj = ai gi hj (2.21)
1≤i≤r
2.3 The ringed spaces. Schemes 107

P
implying that the image of ai gi /f n in Ahj is precisely gj /hj .
1≤i≤r
Therefore the compatible local sections over D(hj ) are the restrictions
of one element from Af , as required. u
t
The sheaf over X = Spec A described above will be sometimes denoted
e The pair (Spec A, A)
by A. e consisting of a topological space and a sheaf over
e
it determines the ring A thanks to Theorem 2.2.4: Namely, A = Γ (Spec A, A).
This pair is the main local object of the algebraic geometry.

2.3. The ringed spaces. Schemes

2.3.1. Ringed spaces. A ringed topological space is a pair (X, OX ) consist-
ing of a space X and a sheaf of rings OX over it called the structure sheaf.
A morphism of ringed spaces F : (X1 , OX ) −→ (Y1 , OY ) is a pair consisting
of a morphism f : X −→ Y of topological spaces and the collection of ring
homomorphisms

{fU∗ : OY (U ) −→ OX (f −1 (U )) for every open U ⊂ Y } (2.22)

that are compatible with restriction maps, i.e., such that

(a) the diagrams
∗
fU
OY (U ) / OX (f −1 (U ))
U
rV
f −1 (U )
r (2.23)
f −1 (V )
² fV∗ ²
OY (V ) / OX (f −1 (V ))

commute for every pair of open sets V ⊂ U ⊂ Y ;

(b) for any open U ⊂ Y , and a pair u ∈ U and g ∈ OY (U ) such that
g(y) = 0, we have

fU∗ (g)(x) = 0 for any x such that f (x) = y. (2.24)

Elucidation. If X and Y are Hausdorff spaces, OX , OY the sheaves

of continuous (smooth, analytic, and so on) functions on them, re-
spectively, then to every morphism f : X −→ Y the ring homomor-
phism fU∗ : OY (U ) −→ OX (f −1 (U )) corresponds: fU∗ assigns to any function
g ∈ OY (U ) the function

fU∗ (g)(x) = g(f (x)) for any x ∈ f −1 (U ); (2.25)

i.e., the domain of fU∗ (g) is f −1 (U ), and fU∗ (g) is constant on the pre-image
of every y ∈ U .
In algebraic geometry, the spaces are not Hausdorff ones and their struc-
ture sheaves are not readily recognized as sheaves of functions. Therefore
108 Ch. 2. Sheaves, schemes, and projective spaces

1) the collection of ring homomorphisms {fU∗ | U is an open set} is not

recovered from f and must be given separately;
2) the condition

fU∗ (g)(x) = g(f (x)) for any x ∈ f −1 (U ) (2.26)

is replaced by a weaker condition (b).

These two distinctions from the usual functions are caused by the fact
that the domains of our “make believe” functions have variable ranges and
different sections of the structure sheaf may represent the same function.
e is called an affine
A ringed space isomorphic to one of the form (Spec A, A)
scheme. We will prove in next subsections the equivalence of this definition
of affine scheme to that given earlier (sec. 1.5.2).
2.3.2. Schemes. A ringed topological space (X, OX ) is said to be a scheme if
its every point x has an open neighborhood U such that (U, OX |U ) is an affine
scheme.
One of the methods for explicit description of a global object is just to
define the local objects from which it is glued and the method of gluing. Here
is the formal procedure.
Proposition. Let (Xi , OXi )i∈I , be a family of schemes and let in every Xi
open subsets Uij , where i, j ∈ I, be given. Let there be given a system of iso-
morphisms θij : (Uij , OXi |Uij ) −→ (Uji , OXj |Uji ) satisfying the cocycle condi-
tion
θii = id, θij ◦ θji = id, θij ◦ θjk ◦ θki = id . (2.27)
S 0
Then there exists a scheme (X, OX ), an open cover X = Xi and a family
i∈I
of isomorphisms ϕi : (Xi0 , OX |Xi0 ) −→ (Xi , OXi ) such that

(ϕj |Xi ∩Xj )−1 ◦ θij ◦ ϕi |Xi ∩Xj = id for all i, j. (2.28)

2.3.2a. Exercise. Prove this theorem.

Hint. For any open subset U , the space (U, OX |U ) is also a scheme. Indeed, let
x ∈ U ; then x has a neighborhood Ux ⊂ X such that (Ux , OX |Ux ) is isomorphic
e Then, U ∩ Ux is a nonempty open subset of Spec A and, since
to (Spec A, A).
the big open sets D(f ), where f ∈ A, constitute a basis of Zariski topology
e D(f ) ) ∼
and (D(f ), A| ef ), we can find an affine neighborhood of x
= (Spec Af , A
contained inside U .
2.3.3. Examples. In these examples, (Xi , OXi ) are most often affine
schemes.
2.3 The ringed spaces. Schemes 109

2.3.3a. Projective spaces. Let K be a ring. Define the scheme PnK , the
n-dimensional projective space over K.
Let T0 , . . . , Tn be independent variables. Set
h i h i
T0 Tn T0 Tn
Ui = Spec K ,..., , Uij = Spec K ,..., ⊂ Ui , (2.29)
Ti Ti Ti Ti Tj /Ti

and determine a scheme isomorphism θij : Uij −→ Uji identifying naturally

f (T0 , . . . , Tn )
the quotient ring whose elements are of the form , where f is
Tia Tjb
a form (homogeneous polynomial) of degree a + b with coefficients in K.
It is not difficult to verify that all the conditions of Proposition 12.2 are
satisfied, and therefore n + 1 affine spaces Ui may be glued together.
2.3.3b. Monoidal transformation. In notation of sec. 2.3.3a, set
h i
T0 Tn
Ui = Spec K T0 , . . . , Tn ; , . . . , ,
Ti Ti
h i (2.30)
T0 Tn
Uij = Spec K T0 , . . . , Tn ; , . . . , .
Ti Ti Tj /Ti

As above, the rings of functions over Uij and Uji can be identified with the
ring whose elements are of the form f (T0 , . . . , Tn )/Tia Tjb , where f is now
an inhomogeneous polynomial, the power of its lowest term being a + b.
Denote by X the scheme obtained after gluing up the Ui and identify
the Ui and Uij with the corresponding open sets in X. We have
[
X= Ui , Uij = Ui ∩ Uj . (2.31)
0≤i≤n

Consider the structure of X in detail. The monomorphism

h i
T0 Tn
K[T0 , . . . , Tn ] −→ K T0 , . . . , Tn , , . . . , (2.32)
Ti Ti

determines the projection of the Ui to An+1 K = Spec K[T0 , . . . , Tn ]. Obviously,

these projections are compatible on Uij . In Ui , single out the open subset
Di = D(Ti ). Since
h i
T0 Tn
K[T0 , . . . , Tn ]Ti = K T0 , . . . , Tn ; , . . . , , (2.33)
Ti Ti

then Di is isomorphically mapped onto the complement to theS “coordinate

hyperplane” V (Ti ) in An+1
K . Therefore X has an open subset Di isomor-
0≤i≤n
phic to An+1
K \ V (T0 , . . . , Tn ), and, if K is a field, this is just the complement
to the origin of the (n + 1)-dimensional S affine space An+1K .
What is the structure of X \ Di ? We have
0≤i≤n
110 Ch. 2. Sheaves, schemes, and projective spaces
[ [
X\ Di = V (Ti ), where | V (Ti ) ⊂ Ui . (2.34)
0≤i≤n 0≤i≤n

Furthermore,
h i
T0 Tn
V (Ti ) = Spec K T0 , . . . , Tn , , . . . , /(Ti ) (2.35)
Ti Ti
h i
T0 Tn
The ring K T0 , . . . , Tn , , . . . , consists of the elements of the form
Ti Ti

f (T0 , . . . , Tn )
, where f is a polynomial
Tia (2.36)
the degree of its lowest terms being ≥ a.

Hence the ideal (Ti ) of this ring consists of the same elements, but with the
degree of the lowest terms of the numerator being ≥ a + 1. Therefore it is
easy to see that
h i h i
T0 Tn T0 Tn
K T0 , . . . , Tn , , . . . , /(Ti ) ' K ,...,
Ti Ti Ti Ti

and we have: h i
T0 Tn
V (Ti ) = Spec K ,..., . (2.37)
Ti Ti
The affine schemes V (Ti ) are glued together as in the³ preceding
´ example;
S
therefore, from the set-theoretical point of view, X = Di ∪ PnK .
0≤i≤n
Thus, X is obtained from the (n + 1)-dimensional affine space

An+1
K = Spec K[T0 , . . . , Tn ]

by pasting V (T0 , . . . , Tn ) in PnK instead of the “origin”.

Exercise. 1) Prove that Γ (PnK , OPnK ) = K. Calculate Γ (X, OX ) for the
scheme X constructed in Example 2.3.3b.
2) Prove that if Spec K is irreducible, then so is PnK .
2.3.4. Condition necessary for gluing spectra. Let us give a simple
algebraic condition necessary for the possibility to glue Spec A and Spec B:
Proposition. Let A and B be the rings without zero divisors. If there is
an open subset U of Spec A such that (U, A|U e ) is isomorphic to (W, B| e W ),
where W is an open subset of Spec B, then the quotient fields of A and B are
isomorphic.
If A and B are rings of finite type over a field or Z, then the converse is
also true.
∼
e U ) −→ e W ). Generic points
Proof. Consider an isomorphism (U, A| (W, B|
of Spec A and Spec B are mapped into each other (they are contained in U ,
2.3 The ringed spaces. Schemes 111

and therefore in W ). The stalks of the structure sheaves at these points are
the quotient fields of A and B respectively.
To prove the converse statementh inotice first of all that if A has no zero
f
divisors, then Spec A and Spec A , where f, g ∈ A, have isomorphic big
g
open sets: h i h i ³ h i´
1 1 1 f
A =A , = A . (2.38)
fg f g g f /1

Now, if A is generated (over K or Z) by elements x1 , . . . , xn and B by

elements y1 , . . . , yn , and the quotient fields of A and B are isomorphic, then
we may pass from A (resp. B) to the ring A[y1 , . . . , yn ] = B[x1 , . . . , xn ] by
a finite number of steps adjoining each time one element from the quotient
field and at each step the spectra of the considered rings have isomorphic open
sets. Taking into account the irreducibility of Spec A and Spec B we obtain
the statement desired. u
t
One of the immediate corollaries of this Proposition is a promised equiv-
alence of the category Aff Sch of affine schemes as defined in sec. 1.5.2 with
the category of affine schemes as defined in sec.2.3.2 — the full subcategory
of the category of schemes Sch.
The schemes X and Y are said to be birationally equivalent if there exist
everywhere dense open subsets U ⊂ X and V ⊂ Y such that (U, OX |U ) is
isomorphic to (V, OX |V ).
The origin of the term “birational equivalence” is as follows. On the spec-
trum of any ring without zero divisors, the elements of its field of quotients can
be considered as “rational functions”. An isomorphism of open sets of open
subsets of Spec A and Spec B was interpreted as “not everywhere” defined
map determined by rational functions.
2.3.4a. Example. Spec k[T ] (the line) and Spec k[T1 , T2 ]/(T12 + T22 − 1) (the
circle) are birationally equivalent if Char k 6= 2. (How to establish that they
are non-isomorphic?)
Indeed, the classical parametrization

T2 − 1 2T
t1 = , t2 =
T2 + 1 T2 + 1
t2
and its inversion T = establish an isomorphism of rings of quotients
1 + t1

k[T ]T 2 +1 = k[T1 , T2 ]/(T12 + T22 − 1)1−t1 ,

where ti = Ti (mod T12 + T22 − 1).

2.3.4b. Example. A generalization of the construction of the previous
example. Let f (T1 , . . . , Tn ) be an indecomposable quadratic polynomial
over a field k of characteristic 6= 2, and with a zero over k. The spaces
0
X = Spec k[T1 , . . . , Tn ]/(f ) and Y = Spec k[T10 , . . . , Tn−1 ] are birationally
112 Ch. 2. Sheaves, schemes, and projective spaces

equivalent. We will describe a parametrization geometrically, leaving its de-

talization and description of isomorphic open sets as an exercise for the reader.
Consider X as a subspace of the affine space E = Spec k[T1 , . . . , Tn ]; let
us embed Y into E by means of the ring homomorphism

Ti 7→ Ti0 for i = 1, . . . , n − 1, and Tn 7→ 0.

On geometric k-points of the spaces X and Y , the correspondence given by

this parametrization is described as follows.
Let us fix a k-point x of the quadric X. We may assume that it does not
lie on Y ; otherwise we can modify the embedding of Y . Let us draw lines
in E through the fixed point x ∈ X and a variable point y ∈ Y . To each
point y ∈ Y we assign a distinct from x point of intersection z of the line
xy with Y . The point z exists and is defined uniquely if y is contained in a
non-empty open subset of Y .
2.3.4c. Spec k[T ] and Spec k[T1 , T2 ]/(T13 + T23 − 1) are birationally
equivalent. To prove this, it suffices to establish that the equation
X 3 + Y 3 = Z 3 has no solutions in k[T ], except those proportional to “con-
stant” solutions (i.e., with X, Y, Z ∈ k). We may assume that the cubic roots
of unity lie in k and apply the classical Fermat’s descent (on degree of the
polynomial) using the unique factorization property of k[T ].

2.4. The projective spectra

In what follows we introduce a very important class of schemes — pro-
jective spectra of Z-graded rings. This class contains analogues of classical
projective varieties and in particular, projective spaces.
2.4.1. Z-graded rings. First, recall the definition of a Z-gradedLring R
(commutative and with a 1, as always in these lectures). Let R = Ri be
i∈Z
the direct sum of the commutative subgroups with respect to addition and
let Ri Rj ⊂ Ri+j , in the sense that ri rj ∈ Ri+j for any ri ∈ Ri and rj ∈ Rj .
The elements r ∈ Ri are said to be homogeneous of degree i. The function
deg : R −→ Z is defined by the formula

deg r = i ⇐⇒ r ∈ Ri . (2.39)

Clearly, any nonzero r ∈ R can be uniquely represented in the form r = Σri

with ri ∈ Ri .
An ideal I ⊂ R is said to be homogeneous if I = ⊕(I ∩ Ri ). For any
homogeneous ideal I, the quotient ring is naturally Z-graded: R/I = ⊕(Ri /Ii ).
Clearly, R0 is a subring and if Ri = 0 for i < 0, then R+ = ⊕i>0 Ri is
a homogeneous ideal of R.
Example. The standard grading of R = K[T0 , . . . , Tn ]: Let deg f = 0 for any
f ∈ K and deg Ti = 1 for all i.
2.4 The projective spectra 113

2.4.2. Projective spectra. The projective spectrum of a Z-graded ring R

with Ri = 0 for i < 0 is the topological space
M
Proj R = {homogeneous prime ideal p of R | p 6⊃ R+ = Ri } (2.40)
i>0

with the topology induced by the Zariski topology of Spec R.

2.4.2a. A geometric interpretation. In the classical projective geome-
try, the projective variety X over an algebraically closed field K is given by
a system of homogeneous equations

Fi (T0 , . . . , Tn ) = 0, where i ∈ I. (2.41)

We associate with X the graded ring R = K[T0 , . . . , Tn ]/(Fi )i∈I . Let us list
several schemes that can be constructed from K[T0 , . . . , Tn ] are related with
the following geometric objects:

Fig. 4

• An+1 K = Spec K[T0 , . . . , Tn ]. It is the (n + 1)-dimensional affine space

over K with a fixed coordinate system.
• C = Spec R, the subscheme of An+1 K . It is the cone with the vertex at the
origin. Indeed, the characteristic property of the cone is that together with
every (geometric) point it contains a generator — the straight line through
this point — and the vertex of the cone. The straight line connecting point
(t0 , . . . , tn ) with the origin consists of the points (tt0 , . . . , ttn ) for any t ∈ K,
and all of these points belong to C, since C is given by homogeneous equations.
Changing t we move along the generator. Moreover, every non-zerovalue of t
determines an automorphism of R which multiplies the homogeneous elements
of degree i by ti . From this it is easy to derive that, conversely, every cone is
given by homogeneous equations.
• PnK = Proj K[T0 , . . . , Tn ]. By the above to the points of PnK the irre-
ducible cones in An+1 K correspond; in particular, to the closed points there
correspond straight lines through the origin.
114 Ch. 2. Sheaves, schemes, and projective spaces

This is the usual definition of the projective space. Though the origin in
An+1
K is a homogeneous prime ideal, R+ = (T0 , . . . , Tn ), it only contains the
vertex of the cone, and therefore is excluded from the definition of points
of PnK .
It is convenient to assign to every straight line through the origin its infinite
point. Then, PnK can be interpreted as a hyperplane in An+1 K moved to infinity.
• Proj R. The above implies that Proj R corresponds to the base of C,
which belongs to the hyperplane through infinity, PnK . u
t
2.4.3. The scheme structure on Proj R. Let us define the structure
sheaf on Proj R and show that the obtained ringed space is locally P affine.
For every f ∈ R, set D+ (f ) = D(f ) ∩ Proj R. Clearly, if f = fi , where
S i∈Z+
fi are homogeneous elements of degree i, then D+ (f ) = D+ (fi ), and
i∈Z+
therefore the sets D+ (f ), where f runs over homogeneous elements of R, form
a basis of a topology of Proj R. For any homogeneous f ∈ R, the localization
Rf , is, clearly, graded:

deg(g/f k ) = deg(g) − k deg(f ). (2.42)

Denote by (Rf )0 the component of degree 0. This component is most

important in the projective case, since, unlike the affine case, only the elements
from (Rf )0 may pretend for the role of “functions” on D+ (f ).
Proposition. Let f, g be homogeneous elements of R. Then
a) D+ (f ) ∩ D+ (g) = D+ (f g);
b) there exists a system of homeomorphisms ψf : D+ (f ) −→ Spec(Rf )0
such that all the diagrams

Ψf
D+ (f ) / Spec(Rf )0
O O
(2.43)
Ψf g
D+ (f g) / Spec(Rf g )0

commute.
(Here the left vertical arrow is the natural embedding, and the right one
is induced by the natural ring homomorphism Rf −→ Rf g .)
ff )0 ) transported onto D+ (f ) via ψf are glued
Corollary. The sheaves ψf∗ (R
together and determine a scheme structure on Proj R.
Proof of Proposition. Since D(f ) ∩ D(g) = D(f g), we obtain a).
To prove b), define ψf as the through map

D+ (f ) −→ D(f ) −→ Spec Rf −→ Spec(Rf )0 , (2.44)

2.4 The projective spectra 115

where the first arrow is the natural embedding, the second one is the isomor-
phism and the last one is induced by the ring monomorphism.
Clearly, ψf is continuous. Let us show that it is one-to-one: construct the
inverse map ϕ : Spec(Rf )0 −→ D+ (f ). Let p ∈ Spec(Rf )0 . Set

ϕ(p)n = {x ∈ Rn | xd /f n ∈ p}, where f ∈ Rd . (2.45)

L
First, let us verify that ϕ(p) = ϕ(p)n is a homogeneous prime ideal. Let
n
x, y ∈ ϕ(p)n . Let us establish that ϕ(p)n is closed with respect to addition,
since the other properties of the ideal are even easier to establish.
Clearly, if xd /f n and y d /f n ∈ p, then (x + y)2d /f 2n ∈ p; and therefore
(x + y)d /f n ∈ p, so x + y ∈ ϕ(p). We have taken into account that p is a prime
ideal in (Rf )0 . Furthermore, if xd /f n , y d /f m ∈ ϕ(p)m+n , then (xy)d ∈ p,
implying either x ∈ p or y ∈ p, since p is prime; therefore either xd /f n or
y d /f m belongs to ϕ(p); hence ϕ(p) is prime.
2.4.3a. Exercise. Verify that ϕ and ψf are mutually inverse.
Now, let us prove that ψf is a homeomorphism. It suffices to show that
ψf is an open map, since its continuity is already established. Let g ∈ Rf . We
have to verify that the image of D+ (f ) ∩ D+ (g) under ψf is open:

D+ (f ) ∩ D+ (g) = D+ (f g) −→ Spec(Rf g )0 =
(2.46)
Spec((Rf )0 )(gd /f e ) −→ Spec(Rf )0

which also shows the possibility of gluing Spec(Rf )0 and Spec(Rg )0 along
D(f g) since

Spec(Rf )0 ⊃ D+ (f g) = Spec(Rf g )0 = Spec((Rf )0 )(gd /f e )

(2.47)
= Spec((Rg )0 )(f e /gd ) = D+ (gf ) ⊂ Spec(Rg )0 . u
t

2.4.4. Examples. 1) Proj K[x0 , . . . , xn ] is the projective space PnK con-

structed in sec. 2.3.3a
2) The scheme X from Example 2.3.3b could have been obtained as follows.
In K[T0 , . . . , Tn ], consider the ideal I = (T0 , . . . , Tn ). Denote by Rk = I k T k
the set of degree k monomials in T with coefficients from the k-th power of I;
set R = ⊕Rk . Clearly, X = Proj R.
A generalization of this construction is the following one.
3) The monoidal transformation with center in an ideal. Let R be
a ring and I its ideal. Construct the graded subring in R[T ]:
M
R= Rk , where Rk = I k T k , (2.48)
k≥0

i.e., the elements of R are the polynomials Σak T k such that ak ∈ I k .

We say that Proj R is the result of a monoidal transformation with center
in I applied to Spec R. u
t
116 Ch. 2. Sheaves, schemes, and projective spaces

2.4.5. Two essential distinctions between affine and projective spec-

tra. 1) Not every homomorphism of graded rings f : R −→ R0 , not even ho-
mogeneous one, induces a map Proj R0 −→ Proj R. For example, consider
a monomorphism

K[T0 , T1 ] −→ K[T0 , T1 , T2 ], Ti 7→ Ti . (2.49)

Then, in Proj K[T0 , T1 ], the ideal (T0 , T1 ) ⊂ K[T0 , T1 , T2 ] has no pre-image.

The corresponding geometric picture is the following one: The projection
of the plane onto the straight line given by the formula (t0 , t1 , t2 ) 7→ (t0 , t1 ) is
not defined at (0, 0, 1) since point (0, 0) does not exist on the projective line.
2) We have established a one-to-one correspondence between rings and
affine schemes: From A we recover the scheme (Spec A, A) e and from an affine
scheme (X, OX ) we recover the ring of global functions: A = Γ (X, OX ). The
analogue of this statement fails for projective spectra: It may very well happen
that Proj R1 ' Proj R2 for quite different rings R1 and R2 . (Examples of such
phenomena are given in the next subsection.)
2.4.6. Properties of R which reflect certain properties of Proj R. A
very extravagant from an algebraic point of view relation

R1 ∼ R2 ⇐⇒ Proj R1 ∼
= Proj R2 (2.50)

adds geometric flavor to our algebra.

Here are two ways to vary R while preserving X = Proj R:
1) For any Z-graded ring R, define its d-th Veronese ring R(d) , where d ∈ N,
by setting (R(d) )i = Rdi .
2) Take R ⊂ R0 such that Ri = Ri0 for all i ≥ i0 .
Lemma. a) Proj R0 ∼= Proj R.
b) Proj R(d) ∼
= Proj R.
S S
Proof. a) Indeed, Proj R = D+ (f ) = Spec(Rf )0 and, besides,
deg f ≥i0 deg f ≥i0
any element of (Rf )0 can be represented in the form g/f n with deg g ≥ i0
(multiply both g and f n by a sufficiently high power of f ).
b) Define the homeomorphism Proj R ∼ = Proj R(d) by setting p 7→ p ∩ R(d) .
For the sheaves, see the argument from the proof of a). u
t

2.5. Algebraic invariants of graded rings

Unless otherwise stated we will only consider the following simplest case
of Z-graded rings R:
1) Ri = 0 for i < 0 and R0 = K is a field;
2) R1 is a finite dimensional space over K;
3) R1 generates the K-algebra R.
2.5 Algebraic invariants of graded rings 117

The space Proj R for such rings R is most close to the classical notion
of a projective algebraic variety over K. In particular, if dimK R1 = r+1, then
the epimorphism of Z-graded rings K[T0 , . . . , Tr ] −→ R sending T0 , . . . , Tr
into a K-basis of R1 determines an embedding

Proj R ,→ Proj K[T0 , . . . , Tr ] = PrK .

We will introduce certain invariants of R following a simple and beautiful idea

of Hilbert to study hr (n) = dimK Rn as a function of n.
2.5.1. dimK Rn is a polynomial with rational coefficients for some
n ≥ n0 = n0 (R). We will prove a more general statement making use
of theLfollowing notions. An R-module M is said to be a Z-graded one if
M = i∈Z Mi and Ri Mj ⊂ Mi+j . A homomorphism f : M −→ N of Z-graded
modules is called homogeneous of degree d if f (Mi ) ⊂ Ni+d .
2.5.2. Theorem. Let M be a Z-graded R-module with finitely many gen-
erators. Then dimK Mn = hM (n) for n ≥ n0 = n0 (M ), where hM (n) is
a polynomial with rational coefficients.
Proof. Induction on dimK R1 = r.
For r = 0, we have R = K and M is a usual finite dimensional linear space.
Clearly, in this case, for a sufficiently large n (greater than the maximal degree
of generators of M ), we have dimK Mn = 0, and the desired polynomial is
zero.
The induction step: Let the statement hold for dimK R1 ≤ r − 1. Let x
be a nonzero element from R1 . Then the action lx : M −→ M of x on M is
a homomorphism of degree 1. Consider an exact sequence, where Kn and Cn
are the degree n homogeneous components of the kernel and cokernel of x,
respectively:
x
0 −→ Kn −→ Mn −→Mn+1 −→ Cn+1 −→ 0 (2.51)
L
Clearly, K = Ker x = ⊕Kn and C = Coker x = n Cn are Z-graded
R/(x)-modules. We have

dim(R/(x))1 = dim R1 /Kx = dim R1 − 1. (2.52)

By the induction hypothesis

hM (n + 1) − hM (n) = hC (n + 1) − hK (n) = h0 (n). (2.53)

Summing up the identities (2.53) starting from some n = n0 we get the

result desired if we take into account the following elementary result: The
PN
sum ni , considered as the function of the upper limit of summation, is
n=n0
a polynomial in N (of degree i + 1). u
t
118 Ch. 2. Sheaves, schemes, and projective spaces

The polynomial hM (n) is called the Hilbert polynomial of the R-module

M . In particular, for M = R, we get the Hilbert polynomial of R.
The number deg hR (n) is called the dimension of X = Proj R. (We do not
claim yet that it only depends on X, not on R.)
2.5.3. Lemma. Let h(x) ∈ Q[x] take integer values at integer x. Then
X x(x − 1) . . . (x − i + 1)
h(x) = ai , where ai ∈ Z. (2.54)
i!
i≥0

Proof. Every polynomial from Q[x] can be represented in the form

X x(x − 1) . . . (x − i + 1)
h(x) = ai , (2.55)
i!
i≥0

where ai ∈ Q and where a0 = h(0) ∈ Z. By induction on i we get

X i . . . (i − j + 1)
h(i) = ai + aj (2.56)
i!
j≤i−1

implying ai ∈ Z. u
t
2.5.4. Hilbert polynomial of the projective space. Let us apply Theo-
rems 2.7.4 and 2.7.5 to calculation of the Hilbert polynomial for the projective
space over a field and prove that it does not depend on the representation
PrK = Proj R.
Let PrK = Proj R; and let temporarily Or (1) denote the invertible sheaf
on PrK constructed with the help of K[T0 , . . . , Tr ], i.e., from the standard
representation PrK as Proj K[T0 , . . . , Tr ]. Then by Theorem 2.7.5 we have
Or (1) ' O(d) for some d ∈ Z.
For r ≥ 1, we have d > 0, since the rank of the space of sections of Or (n)
grows as n −→ ∞.
On the other hand, due to a (yet not proved!) part of Theorem 2.7.4 for
sufficiently large n the map

αn : Rn −→ Γ (PrK , Or (n)) = Γ (PrK , O(nd)) (2.57)

is an isomorphism, and therefore

³ ´
nd + r
hr (n) = . (2.58)
r
In particular, the degree and the constant term of the Hilbert polynomial do
not depend on R.
The highest coefficient of the polynomial hr is called the degree of the
projective spectrum Proj R and deg hr is called the dimension of X = Proj R.
The constant term hR (0) is called the characteristic of X = Proj R and is
denoted by χ(X).
The arithmetic genus of X is pa (X) = (−1)dim X (χ(X) − 1).
2.5 Algebraic invariants of graded rings 119

Exercise. dim PrK = r, deg PrK = 1, pa (PrK ) = 0.

Remark. The degree is the most important projective invariant which to-
gether with the dimension participate in the formulation of the Bezout the-
orem. The geometric meaning of other coefficients is unknown except for the
following result.
R. Hartshorn proved a beautiful theorem according to which the Hilbert
polynomial is the only “discrete” projective invariant in the following sense:
Two projective schemes X and Y have the same Hilbert polynomials if
and only if they can be “algebraically deformed” into each other inside a given
projective space.
2.5.5. Properties of the degree. If hr vanishes identically, set deg hr = −1.
2.5.5a. Lemma. The following statements are equivalent:
1) dim X = −1;
2) X = ∅;
3) R1 ⊂ n(R), where n(R) is the nilradical of R.
Proof. 1) =⇒ 3) is obvious since in this case R1n = 0 for n ≥ n0 .
3) =⇒ 1) follows from the fact that dimK R1 < ∞ and R is generated
by R1 . S S
3) =⇒ 2) follows from the equalities X = D+ (f ) = Spec(Rf )0
f ∈R+ f ∈R+
and the fact that Spec(Rf )0 = ∅, since the localization with respect to any
multiplicative system containing a nilpotent element is 0.
2) =⇒ 3) is obvious since in this case Spec(Rf )0 = ∅ for any f ∈ R+ which
means exactly that (Rf )0 = 0, i.e., f n = 0 for some n and every f . u
t
2.5.5b. Lemma. Let R be an arbitrary Z-graded ring, and {fi }i∈I a collec-
tion of its elements.
S Then the following statements are equivalent:
1) X = D+ (fi );
Pi
2) g n ∈ Rfi for any g ∈ R+ and some n.
i
S
Proof. Let X = D+ (fi ) and let g ∈ R+ be a homogeneous element. Then
S i
D+ (g) = D+ (fi g) by definition of D+ . Furthermore,
i
[
Spec(Rg )0 = Spec((Rg )0 )fid /g l (2.59)
i

implying
P d l
1/1 = (fi /g )ai , where ai ∈ (Rg )0 ;
P mi (2.60)
gn = fi bi , where bi ∈ R.
Since all these arguments are reversible, 1) and 2) are equivalent. u
t
120 Ch. 2. Sheaves, schemes, and projective spaces

2.5.5b.i. Corollary. If R is generated by R1 as an R0 -algebra, then,

for any Ssystem of generators {fi }i∈I of the R0 -module R1 , we have
Proj R = D+ (fi ).
2.5.5c. Proposition. Let R satisfy the condition formulated at the begin-
ning of this section.
1) The following conditions are equivalent:

dim Proj R = 0 ⇐⇒ Proj R is finite. (2.61)

2) If these conditions hold, then Proj R = X is endowed with the dis-

crete topology, and Proj R ∼= Spec Γ (X, OX ) as schemes, where Γ (X, OX ) is
a K-algebra of finite rank. In this case

χ(X) := deg X = dimK Γ (X, OX ). (2.62)

Proof. First, let us show that if dim X = 0, then X is finite and

dimK Rn = dim Γ (X, OX ) for n ≥ n0 . (2.63)

Indeed, if dim X = 0, then dim Rn = d 6= 0 for n ≥ n0 implying that

dimK R(f ) < ∞ for every f ∈ R1 . Otherwise there would have existed g ∈ R1
such that g/f , (g/f )2 , . . . , (g/f )n were linearly independent over K which is
impossible, since then g i f n−i for 1 ≤ i ≤ n should be linearly independent in
Rn for however great n, in particular, for n > d.
Since dimK (Rf )0 < ∞, we deduce that Spec(Rf )0 is finite and discrete.
Indeed, any prime ideal in (Rf )0 is maximal since the quotient modulo it is
a finite dimensional algebra over K without zero divisors, i.e., a field. Therefore
any prime ideal of (Rf )0 is minimal and since, clearly, (Rf )0 is Noetherian,
there are only finitely many minimal ideals.
The space Proj R can be covered by finitely many discrete open spaces
Spec(Rf )0 , where f runs over a K-basis of R Q1 , and therefore Proj R is finite
and discrete. It follows that Γ (X, OX ) = Ox and dimK Ox < ∞. This
x∈X
immediately implies an isomorphism, as ringed spaces:

X ' Spec Γ (X, OX ).

Let us show that

dimK Γ (X, OX ) = deg X = dimK Rn for any n ≥ n0 for some n0 .

First of all, there exists a homogeneous element f ∈ R+ such that

D+ (f ) = X. Indeed, assume the contrary. Then every element from R+ van-
ishes at one of the points of X. Select a minimal subset of points Y ⊂ X
such that every element from R+ vanishes at one of the points of Y . Since the
ideals — points of Proj R — do not contain R+ , it follows that Y contains
more than one point. For every y ∈ Y , there exists ay ∈ R+ which does not
2.5 Algebraic invariants of graded rings 121

Q
vanish anywhere on Y except at y. Then bx = ay vanishes everywhere
y∈Y \{x}
P
on Y except at x, whereas bx does not vanish anywhere on Y . This is
x∈X
a contradiction.
Now, let D+ (f ) = X; then Γ (X, OX ) = R(f ) and it suffices to establish
that
dimK R(f ) = dimK Rn for all n ≥ n0 . (2.64)
S
We have (Rf )0 = Rn /f n , and Rn /f n ⊂ Rn+1 /f n+1 . Since dimK (Rf )0 < ∞,
we see that (Rf )0 = Rn /f n for all n ≥ n0 , and therefore

dimK (Rf )0 = dimK Rn . (2.65)

S
To prove this equality, it suffices to verify that dimK Ann f m < ∞; then,
m
for large n, the map g 7→ g/f n , where g ∈ Rn , is an isomorphism.
Indeed, since X = D+ (f ), then, for any g ∈ R+ , there exists a k such that
g k ∈ f R. Since dimK R < ∞, we can choose k independent of g, and therefore
Rn = f Rn−1 for a sufficiently large n. Since dimK Rn = dimK Rn−1 , no power
of f is annihilated by elements of sufficiently high degree.
To complete the proof, it remains to verify that X is infinite if dim X > 0.
Indeed, if dim X > 0, then dimK Rn −→ ∞ as n −→ ∞, implying that
dimK (Rf )0 = ∞ for some f ∈ R1 . Otherwise the same arguments as above
lead to a contradiction.
Therefore it suffices to show that Spec A is infinite for any K-algebra A
with finitely many generators such that dimK A = ∞. By Noether’s normal-
ization theorem A is a finitely generated module over its subalgebra isomorphic
to K[T1 , . . . , Td ]; since dimK A = ∞, we have d > 0. Now, Theorem 1.6.5 and
the fact that card(Spec K[T1 , . . . , Td ]) = ∞ imply that card(Spec A) = ∞.
u
t
2.5.6. Characteristic functions and Bezout’s theorem. Let hr (n) be
the Hilbert polynomial. Consider the generating function
X
Fr (t) = hr (n)tn . (2.66)
n≥0

Proposition. Fr (t) is a rational function in t;

1) the degree of the pole of Fr (t) at t = 1 is equal to 1 + dim Proj R;
Fr (t)
2) χ(Proj R) = −Res dt;
t=1 t
3) deg Proj R = lim (t − 1)dim P rojR+1 Fr (t).
t−→1

Proof. For k ≥ 1, we have

X ³ ´ X ³ ´k
d d 1
nk tn = t nk−1 tn = . . . = t (2.67)
dt dt 1−t
n∈Z+ n∈Z+
122 Ch. 2. Sheaves, schemes, and projective spaces

implying ³ ´
X d 1
hr (n)tn = hr t . (2.68)
dt 1−t
n∈Z+

The induction on k shows that

³ ´k ³ ´
d 1 k!
t = + ..., (2.69)
dt 1−t (1 − t)k+1

where dots stand for the terms with the poles of orders ≤ k at t = 1. This
and definitions of deg and χ imply 1) and 3).
To prove 2), observe that
³ ´
hR (0)
Res − dt = hR (0) = χ(Proj R),
t=1 t(1 − t)

and for hR (n) = nk , where k > 0, we have

³ ´k
d 1
FR (t) = t ,
dt 1−t

implying µ³ ¶
´k−1
FR (t) d 1
Res dt = Res d t = 0.
t=1 t t=1 dt 1−t
u
t
2.5.7. Example. Let f ∈ Rd be not a zero divisor. Given Fr (t), it is easy
to calculate FR/f R (t):

hR/f R (n) = hR (n) − hR (n − d);

implying
∞
X
FR/f R (t) = (hR (n) − hR (n − d))tn = (1 − td )FR (t) + P (t),
n=0

where P (t) is a polynomial. (Without assumption that f is not a zero divisor

we only get a coefficient-wise inequality Fr /f R (t) ≥ (1 − td )Fr (t) + P (t).) In
particular,
dim Proj R/f R = dim Proj R − 1,
χ(Proj(R/f R)) = χ(Proj R) − hr (−d), (2.70)
deg Proj(R/f R) = d deg Proj R.
The scheme Y = Proj R/f R can be naturally embedded into Proj R (as
V+ (f )); since it is given by one equation, it is called a hypersurface in Proj R.
A particular case: Let R = K[T0 , . . . , Tr ], where the Ti ∈ R1 , and let
Pr = Proj R. The induction on r gives
2.5 Algebraic invariants of graded rings 123

1 1
FPr (t) = F r−1 (t) = . . . = (2.71)
1−t P (1 − t)r+1

implying ³ ´
1 dn 1 n+r
hPr (n) = = ;
n! dtn (1 − t)r+1 r
dim Pr = r,
(2.72)
deg Pr = 1,
χ(Pr ) = 1.
2.5.7a. Theorem (Bezout’s theorem). Let f1 , . . . , fs ∈ R = K[T ], where
T = (T0 , . . . , Tr ), be homogeneous polynomials of degrees d1 , . . . , ds , respec-
tively. Let Y = Proj R/(f1 , . . . , fs ). Then

dim Y ≥ r − s and deg Y ≥ d1 . . . ds ,

(2.73)
where deg Y = d1 . . . ds for dim Y = r − s.

If fi+1 is not a zero divisor in R/(f1 , . . . , fi ) for all i = 1, . . . s − 1 (recall

that such Y is called a complete intersection) the inequalities (2.73) turn into
equalities.
In particular, if Y is a zero-dimensional complete intersection, i.e., r = s,
then deg Y = d1 . . . ds .
Proof. Induction on s. u
t
2.5.7b. A geometric interpretation of complete T intersections. Since,
set-theoretically, we have Proj R/(f1 , . . . , fs ) = V+ (fi ), one should visu-
1≤i≤s
alize Y as the intersection of hypersurfaces singled out by equations fi = 0 in
PrK .
The condition “fi+1 is not a zero divisor in R/(f1 , . . . , fi )” geometrically
means that the (i + 1)-th hypersurface is in “general position” with the inter-
section of the preceding i hypersurfaces, i.e., it does not entirely contain any
of the components of this intersection.
When dim Y = 0, the formula
X
deg Y = dimK Γ (Y, OY ) = dimK Oy (2.74)
y∈Y

replaces the notion of “the number of intersection points multiplicities

counted”.
If K is algebraically closed, the multiplicity of y ∈ Y is by definition equal
to the rank of the local ring.
The term “complete intersection” is connected with the following images.
In P3 (over R or C), let f1 determine a non-degenerate quadric, f2 be its
tangent plane at some point x. The intersection of f1 = 0 and f2 = 0 is the
set of two straight lines through x (the generators of the hyperboloid). These
two straight lines constitute a complete intersection of the quadric and the
124 Ch. 2. Sheaves, schemes, and projective spaces

plane. If we wish to single out one of them, we have to take for f3 , say, the
equation of the plane through the line, and it is not difficult to see that f3 is
a zero divisor in R/(f1 , f2 ). This straight line is not a complete intersection
inside of the quadric because a complete intersection should be of degree ≥ 2
and the degree of the straight line is 1.
2.5.7c. A geometric complete intersection. There is an interest-
ing version of the notion of complete intersection. Let, for definiteness,
R = K[T0 , . . . , Tr ] and p ⊂ R a homogeneous prime ideal. The scheme
X = Proj R/p is called a geometric complete intersection if there exists
an ideal p0 ⊂ p such that X 0 = Proj R/p0 is a complete intersection and
r(p0 ) = p. The latter condition means that the space of X 0 is the same as that
of X and the only difference is in the presence of nilpotents in the structure
sheaf of X 0 .
Is it true that any scheme of the form X = Proj R/p is a geometric com-
plete intersection?
2.5.7d. Problem. The answer is unknown even for the curves in the three-
dimensional space: Is it possible to define any irreducible curve by two equa-
tions? 2)
If dim X = r − 1, where X ⊂ Pr , the answer is positive:
Proposition. If X ⊂ Pr and the dimension of every irreducible component
of X is r − 1, then X is a geometric complete intersection, i.e., X is given by
one equation.
Proof. It suffices to consider the case where X is irreducible.
Let X = Proj K[T0 , . . . , Tr ]/p, where p is a prime ideal, and f ∈ P an ir-
reducible homogeneous element (obviously, it always exists). Then

R = K[T0 , . . . , Tr ]/(f ), i.e., p = f K[T0 , . . . , Tr ] . (2.75)

Indeed, there exists a natural epimorphism

K[T0 , . . . , Tr ]/(f ) −→ K[T0 , . . . , Tr ]/p. (2.76)

If it had a nontrivial kernel, then, since f is irreducible, we would have had

dim K[T0 , . . . , Tr ]/p < dim K[T0 , . . . , Tr ]/(f ) (2.77)

in contradiction with dim X = r − 1. u

t
The above proposition implies, in particular, the following description
of points in PrK . They are of three types:
a) The generic point of PrK corresponding to the zero ideal of K[T0 , T1 , T2 ].
2
To define the curve by 3 equations is relatively easy: Ex. 9 in §6 of [Sh0], v.1.
2.6 The presheaves and sheaves of modules 125

b) The generic points of irreducible “curves”, one-dimensional irreducible

sets. They are in one-to-one correspondence with irreducible forms in three
indeterminates (up to a nonzero constant factor).
c) Closed points. As follows from Hilbert’s Nullstellensatz, if K is alge-
braically closed, such points are in one-to-one correspondence with nonzero
triples (t1 : t2 : t3 ) of elements from K determined up to a nonzero constant
factor.
Let f1 , f2 be two forms; then f2 is not a zero divisor in R/f1 R if and only
if f1 and f2 are coprime, i.e., if and only if the curves f1 = 0 and f2 = 0 have
no common irreducible components.
Comparing the results of sec. 2.5.5c and sec. 2.5.7a we get the classical
Bezout’s theorem for P2 :
Theorem. If two curves on P2 have no common irreducible components, then
the number of their intersection points (multiplicities counted) is equal to the
product of their degrees.

2.6. The presheaves and sheaves of modules

The sheaves of modules over schemes arise naturally in algebraic geometry
as a generalization of the notion “module over a commutative ring”; a more ac-
curate analysis of this correspondence leads to distinguishing of quasi-coherent
sheaves of modules. As we will show, over Spec A the quasi-coherent sheaves
are indeed in one-to-one correspondence with A-modules.
From the geometric point of view the sheaves of modules over a ringed
space X embody the intuitive notion of a continuous system of linear spaces
parameterized by X. If the sheaf is isomorphic to the direct sum of finitely
many copies of OX , then this system is “constant” and the total space of the
corresponding bundle is the direct product of X by the fiber. If the sheaf is
locally isomorphic to such a direct sum, then it corresponds to a locally trivial
vector bundle. In the general case even the dimensions of the fibers may vary.
2.6.1. Presheaves and sheaves of modules over a ringed space
(X, OX ). Here we describe some main notions and results of the sheaf the-
ory which do not depend on the assumption that X is a scheme; owing to
their generality they are not deep.
A presheaf of modules over (X, OX ), or a presheaf of OX -modules, is
a presheaf P of Abelian groups such that a Γ (U, OX )-module structure is
given on each group Γ (U, P) and these structures are compatible with restric-
tions:

rVU (sp) = rVU (s)rVU (p) for any U ⊃ V, p ∈ P(V ), s ∈ OX (V ). (2.78)

Let P1 and P2 be two presheaves of modules over (X, OX ). A presheaf

morphism f : P1 −→ P2 is a set of Γ (U, OX )-module homomorphisms
126 Ch. 2. Sheaves, schemes, and projective spaces

f (U ) : P1 (U ) −→ P2 (U ) commuting with restriction maps given for any open

set U ⊂ X.
It is not difficult to verify that the presheaves of modules over (X, OX )
constitute an Abelian category. In particular, for any presheaf morphism
f : P1 −→ P2 both presheaves Kerf and Cokerf exist and are described by
their groups of sections calculated locally:

(Kerf )(U ) = Kerf (U ), (Cokerf )(U ) = Cokerf (U ) (2.79)

(with obviously determined restriction maps).

Given two presheaves of OX -modules P1 and P2 , define their tensor product
P1 ⊗OX P2 setting:

(P1 ⊗OX P2 )(U ) = P1 (U ) ⊗OX (U ) P2 (U ) (2.80)

(with obviously defined restriction maps).

Given a set I, define the direct sum (direct product if |I| ≥ |Z|) of |I| copies
of a presheaf P by setting
Y
P(I) (U ) = Pi (U ), where Pi is the ith copy of P. (2.81)
i∈I

The direct sum (product) of distinct presheaves is obviously defined. If

(n) (I)
|I| = n ∈ N, we write Ox instead of Ox .
A presheaf of OX -modules P is called a sheaf of OX -modules if P is a sheaf.
If P is a presheaf of OX -modules and P+ the associated sheaf, then P+ also
acquires the natural structure of a sheaf of OX -modules: Originally, P+ (U )
are only determined as Abelian groups; however, the multiplication by the
sections of OX commutes with both the limits from the definition of the sheaf
P+ .
There is determined a canonical morphism of presheaves of mod-
ules P −→QP+ because the image of P(U ) under the homomorphism
P(U ) −→ Px belongs to P+ (U ).
x∈U
Every sheaf of OX -modules F can be considered as a presheaf; the presheaf
obtained in this way from a sheaf F will be denoted by i(F). Defining a sheaf
morphism F1 −→ F2 as a morphism of corresponding presheaves we can
consider i as a functor which embeds the category of sheaves of OX -modules
into the category of presheaves. This “tautological” functor is associated with
a far less trivial functor P 7→ P+ acting in the opposite direction as follows: For
any presheaf P and any sheaf F of OX -modules, we have a natural isomorphism
∼
Hom(i(F), P)−→ Hom(F, P+ ) (2.82)

which to each element i(F) −→ P from the left-hand side group assigns the
through map i(F) −→ P −→ P+ from the right-hand side group (the presheaf
morphism P −→ P+ is described in sec. 2.1.5).
2.6 The presheaves and sheaves of modules 127

Therefore i and + are adjoint functors.

By means of i and + we can define the tensor operators over sheaves
of modules. The general rule is as follows: Perform the corresponding opera-
tion over presheaves and make the result into a sheaf with the help of +. In
particular, given sheaves of OX -modules F1 , F2 we define:

F1 ⊗OX F2 = (i(F1 ) ⊗OX i(F2 ))+ , (2.83)

and, for any sheaf morphism f : F1 −→ F2 , we set

Kerf = (Keri(f ))+ and Cokerf = (Coker(f ))+ . (2.84)

Actually, as is not difficult to show, Ker i(f ) is automatically a sheaf,

whereas for Coker i(f ) this is not true, as is shown in Example 2.6.2 below.
Somewhat later we will encounter examples showing that i(F1 ⊗OX i(F2 ) is not
a sheaf either; these examples demonstrate the importance of the functor +.
Statement. The category of sheaves of modules is Abelian.
2.6.2. Quasi-coherent sheaves. The category of all the sheaves of modules
is usually too large. In what follows we will use two notions which distinguish
a class of important sheaves needed in what follows: Quasi-coherent and co-
herent sheaves. A sheaf F of OX -modules is said to be quasi-coherent if it is
locally isomorphic to the cokernel of a homomorphism of free sheaves.
More precisely, F is quasi-coherent if, for every x ∈ X, there exists a neigh-
borhood x ∈ U , two sets of indices I and J and a homomorphism of sheaves
(I) (J)
of OX |U -modules f : OX |U −→ OX |U such that F|U ' Cokerf .
To elucidate the meaning of the quasi-coherent property, recall that the
(I)
sheaves OX correspond to “trivial bundles”. The property to be isomorphic
to the cokernel of a morphism of trivial bundles is a continuity-type condition:
The jumps of the fibers should not be “too local”, they should mirror a global
picture over open sets.
2.6.2a. Example. Consider a case where X has the simplest non-trivial
structure: X = Spec Zp . Then X = {(0), (p)}, and the open sets are just
X, {(0)} and ∅ with the structure sheaf described by the following diagram,
where Zp −→ Qp is the natural embedding into the field of quotients:

Γ (X, OX ) Zp

(2.85)
² ²
Γ ({(0)}, OX ) Qp

Any presheaf of modules over (X, OX ) is determined by a Zp -module F1 ,

Qp -module F2 and a Zp -homomorphism F1 −→ F2 ; moreover, every presheaf
is a sheaf.
128 Ch. 2. Sheaves, schemes, and projective spaces

Since the only open neighborhood of the point (p) is X, the sheaf
F = (F1 , F2 ) is quasi-coherent if and only if there is an exact sequence of the
(I) (J)
form OX −→ OX −→ F −→ 0. In terms of (F1 , F2 ) this can be expressed as
two exact sequences forming a commutative diagram:

Zp
(I) / Z(J) / F1 /0
p

(2.86)
² ² ²
Qp
(I) / Q(J) / F2 /0
p

N
This immediately implies that F2 ' F1 Qp . It is not difficult to see that
Zp
this condition is also sufficient for quasi-coherentness of F = (F1 , F2 ).
Thus, a quasi-coherent sheaf in this case is uniquely determined by the
module of global sections F1 = Γ (X, OX ); while F2 and the homomorphism
F1 −→ F2 are recovered from F1 . Without this quasi-coherentness condi-
tion we have a greater freedom in defining both F2 and the homomorphism
F1 ⊗Zp Qp −→ F2 : Now the sheaf may suffer a jump at a generic point as
compared with the quasi-coherent case.
In sec. 2.6.4 the result of this example will be generalized to general affine
schemes.
2.6.3. Coherent sheaves. A restriction of “finite type” distinguishes co-
herent sheaves from quasi-coherent ones.
A sheaf F of OX -modules is said to be a sheaf of finite type if it is lo-
(n)
cally isomorphic to the image of OX for some n (in other words, for every
x ∈ X, there exists an open neighborhood U 3 x and a sheaf epimorphism
(n)
OX |U −→ F|U −→ 0). A sheaf of OX -modules is said to be coherent if it is
of finite type and, for every open U and every sheaf morphism
(n)
ϕ : OX |U −→ F|U −→ 0, (2.87)

the sheaf Kerϕ is of finite type.

The general properties of coherent sheaves were first derived in Serre’s
thesis. We confine ourselves to listing them, cf. [KaS]:
a) A subsheaf of finite type of a coherent sheaf is coherent.
b) If, in an exact sequence of sheaves 0 −→ F −→ G −→ H −→ 0, any
two of the three sheaves are coherent, then the third one is also coherent. In
particular, the direct sum of coherent sheaves and also the kernel, cokernel
and the image of any morphism of coherent sheaves are coherent.
c) The tensor product of coherent sheaves is coherent.
d) If the structure sheaf OX is coherent, then a sheaf of OX -modules F is
coherent if and only if it is locally isomorphic to the cokernel of a morphism
(p) (q)
of the form OX −→ OX . u
t
2.6 The presheaves and sheaves of modules 129

2.6.4. Quasi-coherent sheaves over affine schemes. Since the notion

of quasi-coherentness is local, it suffices to describe quasi-coherent sheaves over
affine schemes. Let A be a ring and (X, OX ) = Spec A. Let M be an A-module.
Our aim is to construct a sheaf M f for which M is the module of global sections.
Let us proceed as in sec 2.2.3: For any multiplicative system S ⊂ A,
define the localization MS (sometimes denoted by M [S −1 ]) to be the set
{(m, s) | m ∈ M, s ∈ S}/R, where the relation R is given by the formula

(m, s) ∼ (m0 , s0 ) ⇐⇒ t(s0 m − sm0 ) = 0 for some t ∈ S. (2.88)

The A-module MS is naturally endowed with an AS -module structure such

that
m1 m2 s2 m1 + s1 m2 am am
+ = ; =
s1 s2 s1 s2 s t st (2.89)
for any si ∈ S, a ∈ A, mi , m ∈ M.

Lemma (Cf. 1.6.4b). MS ∼ = AS ⊗A M , in particular, for every multiplicative

subsystem T ⊂ S, there is defined a natural homomorphism MT −→ MS .
For the natural homomorphism f : M −→ MS , where f (m) = m/1, we
have
KerF = {m ∈ M | sm = 0 for some s ∈ S}. (2.90)
For any A-module homomorphism f : M −→ N , the map fS : MS −→ NS
given by the formula fS (m/s) = f (m)/s is an AS -module homomorphism.
Proof is easy, we will only give one hint: define the isomorphism

i : AS ⊗A M ∼
= MS
³ ´ (2.91)
a am
i ⊗m = . u
t
s s

Set Mx = MA\px and define

Y
mx = (. . . , mx , . . .) ∈ f)
Mx = Γ (U, M (2.92)
x∈U

as an element such that, for every x ∈ U , there exists a neighborhood of the

form D(f ) such that, for every y ∈ D(f ), the y-th component of m is the
image of some y ∈ Mf under the morphism Mf −→ My .
e ' OX .
For example, A
f) ' Mf , and the stalk of M over a point x is
2.6.5. Theorem. Γ (D(f ), M
isomorphic to Mx .
Hint. . Define the isomorphism ϕ : Mf −→ Γ (D(f ), M f) setting
Y
ϕ(m/f ) = (. . . , m/f, . . . , m/f, . . .) ∈ Mx . u t (2.93)
x∈D(f )
130 Ch. 2. Sheaves, schemes, and projective spaces

The natural map M 7→ M f from the category of A-modules to the category

of sheaves of OX -modules is, actually, a functor. Indeed, to every morphism
f : M −→ N a morphism f˜: M f −→ N e corresponds which over each D(f ) is
just the localization. The equality f g = f˜g̃ is directly verified.
f u
t
f g
Proposition. For any exact sequence of A-modules M −→N −→P , the se-
f g
f−→
quence of sheaves M e −→
N Pe is exact.
Proof. It suffices to verify the statement “stalk-wise” and apply Lemma 2.6.4.
u
t
2.6.6. Proposition. a) M ∼ f). (This means that not only M
= Γ (Spec A, M f
f.)
is recovered from M , but also M is uniquely recovered from M
∼ f e
b) HomA (M, N ) = HomOX (M , N ).
Proof. a) is a particular case of b). To prove b), observe that the localization
determines a natural map HomA (M, N ) −→ HomOX (M f, Ne ).
f e
On the other hand, a morphism M −→ N is a collection of morphisms
f(U ) −→ N
M e (U ) among which there is a morphism

M∼
= Γ (X, M e) ∼
f) −→ Γ (X, N = N. (2.94)

This determines a map HomOX (M f, Ne ) −→ HomA (M, N ).

The verification of the fact that the constructed maps are mutually inverse
is trivial. u
t
f
Theorem. A sheaf F over X = Spec A is quasi-coherent if and only if F = M
for some A-module M .
Proof. a) Let us prove that if F = M f, then F is quasi-coherent. Let us
represent M as a cokernel of a free A-module morphism:

A(I) −→ A(J) −→ M −→ 0. (2.95)

This gives the exact sequence of sheaves A e(I) −→ A e(J) −→ M f −→ 0 which

f e
implies the quasi-coherentness of M since A = OX .
b) Let F be quasi-coherent. Every point has a neighborhood of the form
D(f ) over which F is isomorphic to the cokernel of a free sheaf morphism. Let
{D(fi ) | 1 ≤ i ≤ n} be an open cover of X = Spec A by such neighborhoods,
and let
³ ´
(I) (J)
F|D(fi ) = Coker OX |D(fi ) −→ OX |D(fi )
³ ´ ³ ´ (2.96)
= Coker A e(I) −→ Ae(J) ' Coker A(I) −→ A(J) = M fi ,
fi fj fi fi

where Mi is an Afi -module. Notice that Mi can be also considered as

an A-module (thanks to the localization homomorphism A −→ Afi ). Fur-
ther on, let Mij = (Mi )fj /1 = Γ (D(fi fj ), F), where i, j = 1, . . . , n. Now set
2.7 The invertible sheaves and the Picard group 131
Y Y
M = Ker(ϕ : Mi −→ Mij ), (2.97)
1≤i≤n 1≤i,j≤n

where ϕ is given by the formula

ϕ((. . . , mi , . . .)) = (. . . , mij , . . .), where mij = mi /1 − mj /1. (2.98)

f ' F. It suffices to verify that

Let us prove that M
f) for any g ∈ A,
Γ (D(g), F) = Γ (D(g), M (2.99)

and this case easily reduces to g = 1 by replacing A by Ag and localizing all

the modules with respect to {g n | n ∈ Z+ }.
Therefore it suffices to show that Γ (Spec A, F) ' M ; but by definition
of a sheaf
Y Y
Γ (Spec A, F) = Ker( Γ (D(fi ), F) −→ Γ (D(fi fj ), F)) (2.100)
1≤i≤n 1≤i,j≤n

f|D(f ) = M
and by definition of M we have M fi and M fij |D(f f ) = M
fij . It only
i i j
remains to apply Theorem 2.6.5 and the definition of M . u
t
2.6.7. Example. Let (X, OX ) be a scheme, JX ⊂ OX a quasi-coherent sheaf
of ideals. The quotient sheaf OX /JX is obviously quasi-coherent. Define the
support of OX /JX by setting

supp OX /JX := {x ∈ X | OX /JX,x 6= {0}}. (2.101)

2.6.8. Lemma. If JX is quasi-coherent, then supp OX /JX is closed in X

and the ringed space (supp OX /JX , OX /JX |supp OX /JX ) is a scheme.
Proof. Consider an affine neighborhood X ⊃ U 3 x; i.e., let U = Spec A.
Denote J := Γ (U, JX ) ⊂ A. Obviously, supp OX /JX ∩ U = V (F ). u
t

2.7. The invertible sheaves and the Picard group

How to characterize in inner terms the projective spectra of Z-graded
rings? The question is not very precise, still, at any rate, a step towards its
answer will be done in this section. We will show that the existence of a grading
in a ring R enables one to define on Proj R = X a particular quasi-coherent
sheaf OX (1). We will introduce certain invariants of R and show that they
actually characterize the pair (X, OX (1)).
2.7.1. Invertible sheaves. A sheaf of modules L over a ringed space
(X, OX ) is said to be invertible if it is locally isomorphic, as a sheaf of OX -mod-
ules, to OX . The following statement is immediate:
132 Ch. 2. Sheaves, schemes, and projective spaces

Statement. For any invertible sheaf L over (X, OX ), set

L−1 = HomOX (L, OX ). (2.102)

a) If L1 and L2 are invertible sheaves, then L1 ⊕OX L2 is invertible.

b) L−1 ⊕OX L ' OX .
Corollary. The isomorphism classes of invertible sheaves over (X, OX ) con-
stitute a commutative group with respect to tensoring over OX .
This group is called the Picard group and denoted by Pic X.
2.7.2.SA cohomologic description of Pic X. Let L be an invertible sheaf,
X= Ui an open cover of X sufficiently fine to satisfy L|Ui ' OX |Ui for all
i∈I
i. Fix an isomorphism ϕi : LUi −→ OX |Ui and consider the restriction maps
rij : L|Ui −→ L|Ui ∩Uj and rji : L|Uj −→ L|Ui ∩Uj .
The elements ϕ−1
j (1) = uj ∈ Γ (Uj , L) completely determine the isomor-
phisms ϕj . Since rij (Ui ) and rji (Uj ) are generators of Γ (Ui ∩ Uj , L), the el-
ements sij determined from the equations rij (Ui ) = sij rji (Uj ) are invertible,
i.e.,
sij ∈ (Γ (Ui ∩ Uj , OX ))× . (2.103)
Let Γ (U, O× ×
X ) = Γ (U, OX ) . Then, to a cover of X and an invertible sheaf
L trivial on the elements from this cover, we have assigned the set

{sij ∈ (Γ (Ui ∩ Uj , O×
X ) for i, j ∈ I}. (2.104)

Obviously, the elements sij satisfy the following conditions:

sij sji = 1 if i 6= j
(2.105)
sij sjk ski = 1 if i 6= j 6= k 6= i.

All such sets (2.104) constitute a group with respect to multiplication called
the group of 1-dimensional Čech cocycles of the cover (Ui )i∈I with coefficients
in the sheaf O× 1 ×
X and denoted by Z ((Ui )i∈I , OX ).
0 1
Two cocycles (sij ), (sij ) ∈ Z are said to be equivalent if there exist
t i ∈ O× 0 −1 −1
X |Ui such that sij = ti sij tj . The elements ti tj obviously constitute
1 × 1
a subgroup B ((Ui )i∈I , OX ) ⊂ Z of coboundaries.
The corresponding quotient group is called the first Čech cohomology group
of X with coefficients in O× and denoted by H 1 ((Ui )i∈I , O× X ).
The above constructed Čech cocycles for an invertible sheaf is multiplied by
e
a coboundary if we change isomorphisms {ϕi }. Indeed, let ϕ0i : L|Ui −→ OX |Ui
−1
be another set of isomorphisms. Since ϕi ϕ0 i ∈ Aut(L|Ui ), we get ϕ0i = ti ϕi
implying s0ij = ti sij t−1
j .

Proposition. The above map from the set of invertible sheaves L on X trivial
×
on a given cover (Ui )i∈I into the set of first Čech cohomology H 1 ((Ui )i∈I , OX )
determined from the same cover is one-to-one.
2.7 The invertible sheaves and the Picard group 133

Exercise. Prove this.

Thus, we have obtained a group monomorphism

H 1 ((Ui )i∈I , O×
X ) −→ Pic X (2.106)

whose image is the set of classes of sheaves trivial over all the Ui .
Let (Uj0 )j∈J be a finer cover. Then a natural monomorphism

H 1 ((Ui )i∈I , O× 1 0 ×
X ) −→ H ((Uj )j∈J , OX ) (2.107)

arises (we leave the task to precisely formulate its definition to the reader).
Since every invertible sheaf is trivial on elements of a sufficiently fine cover,
we get
Pic X ' lim H 1 ((Ui )i∈I , O× 1 ×
X ) = H (X, OX ), (2.108)
−→

where the inductive limit is taken with respect to an ordered system of cov-
erings.
2.7.3. Example. SOn X = Proj R, where R is generated by R1 over R0 , con-
sider a cover X = Uf , where Uf = D+ (f ), and a cocycle sf g ∈ Z 1 (Uf , O×
X)
f ∈R
given by
sf g = (f /g)n ∈ Γ (Uf ∩ Ug , O×
X ). (2.109)
The invertible sheaf determined with the help of this cocycle is denoted by
OX (n); obviously, we have
(
OX (1)⊗n if n ≥ 0,
OX (n) ' (2.110)
OX (−1)⊗n if n ≤ 0,

where OX (−1) = OX (1)−1 .

These sheaves are constructed from R; the other way round, R can be
recovered to an extent from X and OX (1). Here we will only prove a part
of the result; the second part will be proved with the help of the cohomology
technique in what follows.
First, notice that, for every invertible sheaf L over aL
ringed space (X, OX ),
there is a natural structure of a Z-graded ring on Γ (X, Ln ) with the
n∈Z
product of homogeneous elements determined from the map

Γ (X, Ln ) × Γ (X, Lm ) −→ Γ (X, Ln ⊗OX Lm ) ∼

= Γ (X, Ln+m ). (2.111)

2.7.4. Theorem. Let X = Proj R, where R0 is a Noetherian ring and R1 is

a Noetherian R0 -module that generates R, and L = OX (1). L
Then there exists
a homogeneous homomorphism of graded rings α : R −→ Γ (X, Ln ) and
n∈Z
n0 ∈ Z such that the maps αn : Rn −→ Γ (X, Ln ) are group isomorphisms for
n ≥ n0 .
134 Ch. 2. Sheaves, schemes, and projective spaces

Proof. Let us construct α and show that its kernel is only supported in small
degrees. The statement on isomorphism will be proved in what follows.
Let h ∈ Rn . For any f ∈ R1 , set
h
α(h)|D+ (f ) = ∈ Γ (D+ (f ), OX ). (2.112)
fn

h
Obviously, the sections are glued together with the help of the cocycle
fn
(f /g)n into a section of Ln over the whole X; denote this section by α(h).
Clearly, α is a homomorphism of graded rings.
h
Let h ∈ Rn ∩ Ker α. This means that = 0 for all f ∈ R1 . Since R1
fn
is Noetherian, there exists an integer m0 such that RmL
h = 0 for m ≥ m0 .
Consider an arbitrary h whose annihilator contains Rm . Obviously,
m≥m0 (h)
all such elements constitute an ideal J ⊂ R.
There are finitely many generators of J since R is Noetherian, and therefore
we can choose one m0 for all generators h; so that Jm = 0 for m ≥ m0 . u
t
2.7.5. Picard groups: Examples.
Proposition. If A is a unique factorization ring, then Pic(Spec A) = {0}.
Proof.
S In the system of all open coverings the finite coverings of the form
D(fi ) constitute a cofinal subsystem 3) , and therefore it suffices to verify
i∈J
that
H 1 (D(fi ), O×
X ) = {0}. (2.113)
Let sij ∈ Z 1 (D(fi ), O× 0 0
X ). Let us represent all sij for i 6= j in the form ti /tj ,
0
where the ti are elements from the quotient field K of A. It is easy to see
that this is indeed possible: Since sij sjk ski = 1 for any k, it follows that
sij = sik /sjk because sik sik = 1.
Now let p be a prime element of A and vp (a) the exponent with which
p enters the decomposition of a ∈ K. Up to multiplication by invertible el-
ements, the set P of primes p ∈ A such that vp (t0i ) 6= 0 for some i is only
finite.
Fix p ∈ P , and divide all the fi into two groups: The one with p dividing
.
fi for i ∈ J1 and the other with fi 6 .. p if i ∈ J2 . Since the fi are coprime,
J1 6= ∅.
.
Since sij is invertible an Afi fj , we see that vp (sij ) = 0 if fij 6 .. p and vp (t0i )
takes the same value — ap — for all i ∈ J1 . Set
3
Recall that a subset B of a partially ordered set A is said to be cofinal if, for
every a ∈ A, there exists b ∈ B such that a = b.
Also, a sequence or net of elements of A is said to be cofinal if its image is
cofinal in A.)
2.7 The invertible sheaves and the Picard group 135
à !
Y
ti = p−ap t0i (2.114)
p∈P

Obviously, sij = ti /tj ; on the other hand, ti ∈ Γ (D(fi ), OX ) = A× f . Indeed,

..
if fi \. p, then vp (ti ) = 0 and ti only factorizes into the product of the prime
divisors of fi , which implies that ti is invertible in Afi . u
t
Remark. In terms of cohomology with coefficients in sheaves, we can inter-
pret this proof as follows. The exact sequence of sheaves of Abelian groups on
X
1 −→ O× e× p e× ×
X −→ K −→K /OX −→ 1, (2.115)
where Ke × is the constant sheaf (i.e., Γ (U, K
e × ) = K × for any U ), induces the
exact sequence of cohomology groups
∗ ∗
p0 p1
e × )−→Γ
Γ (X, K e × /O× ) −→ H 1 (X, OX )−→H
(X, K 1 e ×)
(X, K (2.116)
X

The first step of the above proof establishes that p∗1 = 0 (actually,
the same argument shows that H 1 (X, K e × ) = 0). The second step shows
∗
that p0 is an epimorphism; and it is only here that we have used the fact
that A is a L unique factorization ring, which, in particular, implies that
Ke × /A
e× ' e
Z.
p∈Spec A
Here is an important application of the above statement:
2.7.5a. Theorem. Let A be a unique factorization ring. Then Pic PrA for
r ≥ 1 is an infinite cyclic group with the class of O(1) as its generator.
Proof. Recall that PrA = Proj A[T0 , . . . , Tr ]. By the above
h proposition,i any
r T0 Tr
invertible sheaf L over PA is trivial on D+ (Ti ) = Spec A ,..., , since
Ti Ti
by a theorem of Gauss (cf. [Pr]) the polynomial ring over A preserves the
unique factorization property of A.
Now let (sij | 0 ≤ i, j ≤ r) be a cocycle defining L for the cover
(D+ (Ti ) | 0 ≤ i ≤ r). Since sij is homogeneous of degree 0 and only fac-
torizes in the product of the divisors of Ti Tj (use the unique factorization
property of A[T0 , . . . , Tr ]), we have
µ ¶nij
Ti
sij = εij , where εij ∈ A× . (2.117)
Tj

Since sij sji = 1 and sij sjk ski = 1, it follows that nij = n (does not depend
on i, j), and therefore εij is a cocycle.
In actual fact, εij is automatically a coboundary, since εij = εik /εjk for
any k and εik ∈ Γ (PrA , O× X ). Therefore (sij ) is cohomologic to the cocycle
(Ti /Tj )n defining O(n).
We will now get theorem’s statement if we prove that all the sheaves
O(n) are non-isomorphic. This is true for any A as shown by the following
statement:
136 Ch. 2. Sheaves, schemes, and projective spaces

Lemma.

0 if n < 0
Γ (PrA , O(n)) = L (2.118)
 AT0a0 . . . Trar if n ≥ 0.
a0 +...+ar =n

Proof. Let R = A[T0 , . . . , Tr ], deg Ti = 1 for all i. Let us show that the
homomorphism αn : Rn −→ Γ (PrA , O(n)) is an isomorphism for n ≥ 0.
Recall that αn (f )|D+ (Ti ) = f /Tin for any f ∈ Rn . Now the fact that the
Ti are not zero divisors immediately implies that αn is a monomorphism.
Let us prove that αn is an epimorphism. A section of the sheaf O× over
r
PA is represented by the set
½ h i µ ¶n ¾
T0 Tr Ti
fi ∈ A ,..., | 0 ≤ i ≤ n, and fi = fj . (2.119)
Ti Ti Tj

Since Ti are not zero divisors, the compatibility conditions imply that fi Tin
do not depend on i. Obviously, fi Tin is a polynomial since its denominator
can only be a power of Ti .
Now let n < 0; then, in the same notation, we get

fi /Ti−n = fj /Tj−n , (2.120)

and similar divisibility considerations show that this is only possible for fi = 0.
u
t
Theorem is also proved. u
t
2.7.5b. Corollary. O(n) ∼
6 O(m) if n 6= m.
=
Proof. This is an immediate corollary of the above Lemma since the ranks
of the A-modules of sections of a certain power of these sheaves are distinct.
u
t
2.7.6. Hilbert polynomial of the projective space. As an application
of Theorems 2.7.5a and 2.7.4 we can now compute Hilbert polynomials of the
projective spaces over a given field and look which of the numerical charac-
teristics of PrR introduced in § 2.5 do not depend on the representation of PrR
in the form Proj R.
Indeed, let Prk = Proj R; temporarily, denote by OR (1) the invertible sheaf
on Prk constructed with the help of R, and let O(1) be the invertible sheaf
constructed by means of the standard representation Prk = Proj k[T0 , . . . , Tr ].
By Theorem 2.7.5a we have

OR (1) ' OR (d)

for some d ∈ Z; for r ≥ 1, we have d > 0 since the rank of the space of sections
of the sheaf OR (n) grows as n −→ ∞.
2.8 The Čech cohomology 137

On the other hand, by (yet unproven!) part of Theorem ?? for n sufficiently

large, the map

αn : Rn −→ Γ (Prk , Ok (n)) = Γ (Prk , O(nd))

is an isomorphism. Hence
³ ´
nd + r
hR (n) = .
r

In particular, the degree and the constant term of the Hilbert polynomial do
not depend on R, as claimed.
2.7.7. Exercises. 1) Prove that the curve in Prk can be isomorphic to P1k
only if its degree is equal to 1 or 2.
Hint. By definition, any curve in Prk is of the form Proj k[T0 , T1 , T2 ]/(f ),
where f — is a form. Its Hilbert polynomial is computed in § 2.5.
Try to prove that the curve determined by a quadratic form f , is isomor-
phic to P1k if and only if the following two conditions are fulfilled: 1) rank f = 3;
2) The equation f = 0 has a non-zero solution in k.
f
2) Let r ≥ 1; prove that any automorphism f : Prk −→ Prk over k is linear
(what does it mean?).
Hint. Look how f acts on invertible sheaves and on the Picard group.

2.8. The Čech cohomology

2.8.1. The Čech complex. Let X — be a topological space and F — a
r
sheaf of abelean groups on X. LetU = ∪ Ui be a finite open cover of X. In
i=1
this situation we give the following
A Čech complex is a complex whose homogeneous components C p (U, F)
(called groups of Čech p-cochains) and the differential are as follows:
Let [1, r]p+1 be the (p+1)-fold direct product of the set of integers 1, . . . , r.
The elements of C p (U, F) are the functions

s(i0 , . . . , ip ) ∈ Γ (Ui0 . . . Uip , F), where Ui0 ...ip := Ui0 ∩ . . . ∩ Uip

“skew-symmetric” in the following sense:

s(σ(i0 ), . . . , σ(ip )) = sgnσ · s(i0 , . . . , ip ) for any σ ∈ Sp

and

s(i0 , . . . , ip ) = 0 if among the indices i0 , . . . , ip at least two coincide.

Q
r
In particular, C p (U, F) = 0 for p ≥ r; besides, C 0 (U, F) = Γ (Ui , F).
i=1
138 Ch. 2. Sheaves, schemes, and projective spaces

The differential in the Čech complex is as follows (hereafter in similar

sums, the hatted argument should be ignored):
p+1
X
(ds)(i0 , . . . , ip+1 ) = (−1)k rs(i0 , . . . , bik , . . . , ip+1 ),
k=0

where r : Γ (Ui0 ...bik ...ip+1 , F) −→ Γ (Ui0 ...ip+1 , F) — is the restriction homomor-

phism (which, obviously, depends on i0 , . . . , ip+1 and k).
The cohomology groups of this complex are called Čech cohomology groups
of the cover U with coefficients in the sheaf F and are denoted Ȟ p (U, F).
Cohomology with coefficients in the sheaf defined a la Čech are convenient for
calculations. However, they characterize, in a sense, the cover U of the space
X and sections of the sheaf F over its charts, rather than the space X itself
and the sheaf F itself.
Grothendieck suggested an axiomatic definition of cohomology of the space
X with coefficients in a sheaf F. By this definition the p-th cohomology
H p (X, F) of X with coefficients in a sheaf F is the right derived functors
of the functor that to every sheaf of Abelian groups F on X assigns its group
of sections Γ (X, F).
Let us formulate without proof a theorem offering a sufficient condition
for coincidence of Grothendieck’s H p (X, F) with Čech’s Ȟ p (U, F).
2.8.2. Theorem (Cartan). Let V be a family of quasi-compact open sub-
sets of a topological space X forming a basis of topology of X and such that
S
r
H p (U, F) for all p ≥ 1 and Ui ∈ V for all finite coverings U = (Ui )r1=1 such
i=1
that Ui ∈ V . Then the cohomological functor F −→ Ȟ ∗ (X, F) is equivalent to
H ∗ (X, F) for any finite cover U of the space X by elements of V .
Remark. The theorem is not formulated in full generality, but it suffices for
our nearest purposes. For its proof, see Th. 5.9.2 in Ch. 2 of Godeman’s book
[God].
We will apply this theorem to the schemes X considering as V the families
of affine open sets and taking any quasi-coherent sheaf as F. Let us establish
that in this situation the conditions of Cartan’s theorem are fulfilled. For this,
it suffices to prove the following result.
2.8.3. Proposition. Let X = Spec A, and F = M f, where M is an A-mod-
ule. Let Ui = D(fi ), where i = 1, . . . , r; let X = ∪Ui , and U = (Ui )r1=1 .
Then
H p (U, F) = 0, for p ≥ 1.
2.8.3a. Corollary. For any affine scheme X and a quasi-coherent sheaf F
on it, we have H p (X, F) = 0 for any p ≥ 1.
Proof. Apply Cartan’s theorem to X and the family of big open sets D(f ).
u
t
2.8 The Čech cohomology 139

2.8.3b. Corollary. For any scheme X, its finite cover U by affine schemes,
and a quasi-coherent sheaf F on X, we have

Ȟ p (U, F) = H p (X, F).

Proof. Apply Cartan’s theorem to X and the family of affine open subsets
of X using Corollary 2.8.3a. u
t
Observe without proof that Serre proved the inversion of Corollary 2.8.3a:
if for a scheme X and any quasi-coherent sheaf of ideals J on X, we have
H 1 (X, J) = 0, then X is an affine scheme.
2.8.4. Properties of Čech complexes. Retain notation of Proposi-
tion 2.8.3. Let Ui0 ...ip = Ui0 ∩ . . . ∩ Uip = D(fi0 . . . fip ). We have

f) ' Mf ...f .
Γ (Ui0 ...ip , M i0 ip

p f
³ Each
´ cochain of C (U, M ) for p < r can be represented by a collection of
r
elements of different localizations of the module M :
p+1

s(i0 , . . . , ip ) ∈ Mfi0 ...fip .

We wish to prove that the Čech complex is acyclic in dimensions p ≥ 1,

i.e., the cohomology of this complex are 0 in these dimensiona. The standard
method of proving acyclicity is to construct a homotopy operator or, more
precisely a series of chain homotopy operators proving an equivalence of the
given complex and an acyclic one. 4) For the Čech complex we can not con-
struct such an operator, but we will get round this difficulty as follows. We
construct a chain of complexes (Cnp (M )), where n is the number of a complex,
and homomorphisms between them so that:
a) the Čech complex is the inductive limit of the complexes Cnp (M ).
b) The complexes Cnp (M ) are acyclic.
4
Let (A, d) and (A0 , d0 ) be chain complexes and f : A → A0 , g : A → A0 be chain
maps. A chain homotopy D between f and g is a sequence of homomorphisms
{Dn : An → A0n+1 } so that d0n+1 ◦ Dn + Dn−1 ◦ dn = fn − gn for each n. Thus,
we have the following diagram:

dn+1
An+1 / An dn / An−1
z z
zz Dn−1 zzz f −g
fn+1 −gn+1
zzzD zz n−1 n−1
² |zz n ² |zz ²
A0n+1 / A0n / A0n−1
d0n+1 d0n

If there exists a chain homotopy between f and g, then f and g are said to be
chain homotopic. The complex chain homotopic to the one with zero cohomology
is said to be acyclic.
140 Ch. 2. Sheaves, schemes, and projective spaces

Since the passage to the inductive limit commutes with computing coho-
mology, we see that
f) = H(C p (U, M
Ȟ p (U, M f)) = 0.

Let us pass to item a) of this program.

First, let us prove that for any ring A, any A-module M , and any g ∈ A,
the module Mg can be naturally represented as an inductive limit.
First of all, by setting
n o
m
Mg(n) = n
| m ∈ M ,
g

S
∞
(n) (n)
we see that Mg = Mg . Each space Mg is an A-module and we can
n=0
replace the union by the inductive limit by considering the system

. . . −→ Mg(n) −→ Mg(n+1) −→ . . . , (2.121)

m mg
in which the homomorphisms are described by the formula 7→ n+1 . If g
gn g
(n)
is not a zero divisor in M , then the A-module Mg is isomorphic to M for
m
all n with respect to the map m −→ n . The inductive system (2.121) turns
g
now into
g g g
. . . −→ M (n) = M −→ M (n+1) = M −→ . . . , (2.122)
where each slot is occupied by a copy of the A-module M , and each homo-
morphism is multiplication by g.
2.8.4a. Lemma. The inductive limit of the system (2.122) is always iso-
morphic to Mg (even if g is a zero divisor in M ).
Proof. Consider the homomorphisms
m
M (n) −→ Mg , m −→ .
gn

They are compatible with homomorphisms of the system (2.122) and therefore
define the homomorphism of its limit:

lim M (n) −→ Mg .
−→
Its cokernel is, clearly, zero. Any element of the kernel is represented by a chain
of elements gmn , g 2 mn , . . ., where mn ∈ M (n) = M , such that mn /g n = 0
in Mg ; this means that g n+k mn = 0 for k sufficiently large, and therefore the
whole chain represents the zero class. u
t
2.8 The Čech cohomology 141

2.8.4b. Now, in the whole complex (C p (U, M f)), replace the localizations
(n)
of M by their “approximations” M and appropriately define the cochain
operators.
To make the correct expression of the cochains graphic, we first assume
that the elements fi0 , . . . , fip are not zero divisors in M . Let Cnp (M ) denote
the subgroup of C p (U, M f) consisting of cochains such that
n o
(n) m
s(i0 , . . . , ip ) ∈ Mfi ...fi = n
|m∈M .
0 p (fi0 . . . fip )
Let
mi0 ...ip
s(i0 , . . . , ip ) = .
(fi0 . . . fip )n
Then
p+1
X
mi0 ...ip+1 mi0 ...bik ...ip+1
ds(i0 , . . . , ip+1 ) = = (−1)k ,
(fi0 . . . fip+1 )n
k=0
(fi0 . . . fbik . . . fip+1 )n

implying that
p+1
X
mi0 ...ip+1 = (−1)k fink mi0 ...bik ...ip+1 . (2.123)
k=0
p
The embedding homomorphism Cnp (M ) −→ Cn+1 (M ) is described by the
formula
mi0 ...ip −→ fi0 . . . fip mi0 ...ip . (2.124)
We use formulas (2.123) and (2.124) to define both the differential in the com-
plex Cnp (M ) when the condition on zero divisors is not satisfied, and complex
p
homomorphism Cnp (M ) −→ Cn+1 (M ).
In the general case, denote by Cnp (M ) the group of skew-symmetric func-
tions on [1, r]p+1 with values in M and define the coboundary operator
Cnp (M ) −→ Cnp+1 (M ) by setting:
p+1
X
(dm)(i0 , . . . , ip+1 ) = (−1)k fink m(i0 , . . . , bik , . . . , ip+1 ). (2.125)
k=0
p
Define the group homomorphism ϕn = Cnp (M ) −→ Cn+1 (M ) by the for-
mula:
(ϕn m)(i0 , . . . , ip ) = fi0 . . . , fip m(i0 , . . . , ip ). (2.126)
2.8.4c. Lemma. 1) The collection of sets (Cnp (M )) for n fixed is not a
complex for any n.
2) The collection of homomorphisms ϕn is a complex homomorphism.
f); the inductive limit of differentials is the dif-
3) lim Cnp (M ) = C p (U, M
−→
ferential in the inductive limit.
The first two statements are verified by trivial calculations; the third one
follows from definitions and Lemma 2.8.4c.
This is the end of stage a) of computing cohomology of the Čech complex,
i.e., its approximating by complexes Cnp (M ) which are easier to deal with.
142 Ch. 2. Sheaves, schemes, and projective spaces

2.8.4d. Stage b) of the program. Let us now pass to the proof of acyclic
property of the complex Cnp (M ). In the construction of this complex there
are involved the ring A, the A-module M , the elements f1n , . . . , frn ∈ A that
determine the cover {D(fin )}ri=1 = {D(fi )}ri=1 , and the differential given by
formula (2.125).
Since the complex Cnp (M ) is important in various problems of algebraic
geometry, we will study it in more detail than is strictly necessary for our
purposes.
2.8.5. Koszul complex. Let A be a ring, f = (f1 , . . . , fr ) a collection of
its elements; set f n := (f1n , . . . , frn ).
Consider a free A-module Ae1 ⊕ . . . ⊕ Aer = Ar of rank r and its exterior
p r
powers³ K
´p = ΛA A ; by definition, K0 = A.Clearly, Kp is a free A-module of
r
rank ; the elements ei1 ∧ . . . ∧ eip , where i1 < . . . < ip , constitute its basis.
p
Define the differential d : Kp+1 −→ Kp by setting:
p+1
X
d(ei1 ∧ . . . ∧ eip+1 ) = (−1)k+1 fik ei1 ∧ . . . ∧ ebik ∧ . . . ∧ eip+1 .
k=1

(NB: −1 is raised to power k + 1 in order for the first term enter with its
initial sign.) It is trivial job to verify that d2 = 0; let Kp (f, M ) or briefly
Kp (f ) be the complex obtained. (Observe that it is a chain complex, whereas
the Čech complex is a cochain one.)
The relation between complexes Kp (f ) and Cnp (M ) is as follows.
2.8.5a. Lemma. For p ≥ 0, we have

Cnp (M ) ' K p+1 (f n , M ) := Hom(Kp+1 (f n ), M );

and the isomorphism can be selected to be compatible with the differentials.

Proof. We assign to any cochain m = (m(i0 , . . . , ip )) ∈ Cnp (M ) the homo-
morphism

Kp+1 (f n ) −→ M,
gm : ei0 ∧ . . . ∧ eip 7−→ m(i0 , . . . , ip ).

The differential of gm considered as an element of Hom(Kp+2 (f n ), M ), is given

by the formula:

(dgm )(ei0 ∧ . . . ∧ eip+1 ) = gm (d(ei0 ∧ . . . ∧ eip+1 ) =

µXp+1 ¶
= gm (−1)k fink m(i0 , . . . , bik , . . . , ip+1 = gdm (ei0 ∧ . . . ∧ eip+1 ),
k=0

which proves the desired. u

t
2.8 The Čech cohomology 143

Therefore, we have extracted the dependence of Cnp (M ) on M . The com-

plex Kp (f ) can also be further “dismantled”; this is convenient in the proofs
based on induction on r.
2.8.6. Lemma. K0 (f ) ' K0 (f1 ) ⊗ . . . ⊗ K0 (fr ).
Proof. First of all, recall that the tensor product of two chain complexes K0
and L, is the complex such that
L
(K ⊗ L)p = Ki ⊗ Lj ,
i+j

d(k ⊗ l) = dk ⊗ l + (−1)r k ⊗ dl, where k ∈ Ki ;

and, generally, for K (1) ⊗ . . . ⊗ K (r) , we have

r
X
d(k1 ⊗ . . . ⊗ kr ) = (−1)d1 +...+dj−1 k1 ⊗ . . . ⊗ b
kj ⊗ . . . ⊗ kr ,
j=1
(j)
where kj ∈ Kdj for j = 1, . . . , r.

(i)
Let now K0 = A, and K1i = Aei for all i. Construct a complex by setting

0 ←− A ←− Aei , d(ei ) = fi .

Then (K (1) ⊗ . . . ⊗ K (r) )p is a free A-module with a basis ei1 ⊗ . . . ⊗ eip , where
1 ≤ i1 < . . . < ir ≤ p, and the differential
p
X
d(ei1 ⊗ . . . ⊗ eip ) = (−1)k+1 fik ei1 ⊗ . . . ⊗ ebik ⊗ . . . ⊗ eip .
k=1

This shows that it is isomorphic to the Koszul complex K0 (f ). We can now

prove that the complex Cnp (M ) is acyclic in dimensions ≥ 1. For this, it suffices
to construct a homotopy for Kp (f ). u
t
2.8.7. Proposition. Let g1 , . . . , gr be an arbitrary collection of r elements
of A. Let h : Kp (f ) −→ Kp+1 (f ) be the exterior multiplication on the left by
Pr
gi ei . Then
i=1
r
X
hd + dh = fi gi
i=1

(i.e., the multiplication by the sum on the right).

Proof. Fix a set of indices (without repetitions) i1 , . . . , ip+1 ∈ [1, r] and let
j1 , . . . , jr−p−1 be a complementary set of indices. We have:
144 Ch. 2. Sheaves, schemes, and projective spaces

dh(ei1 ∧ . . . ∧ eip+1 ) =
µµX r ¶ ¶ µr=p−1
X ¶
d gk ek ∧ ei1 ∧ . . . ∧ eip+1 = d gjk ejk ∧ ei1 ∧ . . . ∧ eip+1 =
k=1 k=1
r=p−1
X p+1
X
(gjk fjk ei1 ∧ . . . ∧ eip+1 ) + gjk (−1)l fil ejk ∧ . . . ∧ ebil ∧ . . . ∧ eip+1 .
k=1 l=1

On the other hand:

µX
p+1 ¶
hd(ei1 ∧ . . . ∧ eip+1 ) = h (−1)l+1 fil ei1 ∧ . . . ∧ ebil ∧ . . . ∧ eip+1 =
l=1
p+1
X
= (−1)l+1 fil (gil eil ∧ ei1 ∧ . . . ∧ ebil ∧ . . . ∧ eip+1 +
l=1 p+1
X
+ gjk ejk ∧ ei1 ∧ . . . ∧ ebil ∧ . . . ∧ eip+1 =
k=1
p+1
X
= (−1)l+1 fil (−1)l−1 gil ei1 ∧ . . . ∧ eip+1 +
l=1 p+1
X
+ gjk ejk ∧ ei1 ∧ . . . ∧ ebil ∧ . . . ∧ eip+1 .
k=1

Adding up these two expressions we get the desired. u

t
2.8.7a. Corollary. If the ideal of A generated by the elements fi , where
i = 1, . . . , r, is equal to A, then the complexes Cnp (M ) and C(U, F) are acyclic
in dimensions ≥ 1.
Pr
(Indeed, then fi gi = 1 for certain gi , and hence h is the homotopy opera-
tor.) i=1

This completes the proof of Proposition 2.8.3 and its corollaries. u

t
2.8.7b. Remark. In the general case: Cnp+1 (M ) = K p+1 (f n , M ) for p ≥ 0.
By definition we have an exact sequence
d
0 −→ H 0 (f n , M ) −→ K 0 (f n , M ) −→ Z 1 (f n , M ) −→ H 1 (f n , M ) −→ 0.

The limit lim as n −→ ∞ gives:

−→
d
0 −→ H 0 ((f ), M ) −→ K 0 ((f ), M ) −→ Z 1 ((f ), M ) −→ H 1 ((f ), M ) −→ 0,

and

H 0 (U, F) = Z 1 ((f ), M ),
K 0 ((f ), M ) = K 0 ((f n ), M ) = Hom(A, M ) = M,
2.9 Cohomology of the projective space 145

so we have an exact sequence:

d
0 −→ H 0 ((f ), M ) −→ M −→ H 0 (U, F) −→ H 1 ((f ), M ) −→ 0.

2.9. Cohomology of the projective space

2.9.1. Let A be a fixed ring, Pr−1
A = Proj R, where R = A[T1 , . . . , Tr ] with
the standard grading, Ui = D+ (Ti ), U = (Ui ).
In this section we compute H p (Pr−1
A , O(n)) for any p, n and r. This com-
putation, due to Serre, is the base of the proof (in the next section) of main
results on cohomology of coherent sheaves on projective schemes.
Thanks to results of § 2.8, we have

H p (Pr−1 p
A , O(n)) = Ȟ (U, O(n)).

Therefore we can compute the cohomology of the Čech complex of the cover U .
Since Ui0 ,...,ip = D+ (Ti0 . . . Tip ) = Spec R(Ti0 ...Tip ) , we have, for the usual
description of the sheaf O(n):
n o
m(i0 , . . . , ip )
Γ (Ui0 ...ip , O(n)) = | k ∈ Z, m ∈ Rk(p+1)+n .
(Ti0 . . . Tip )k

This implies that

L
Γ (Ui0 ...ip , O(n)) = {si0 ...ip | si0 ...ip ∈ RTi0 ...ip }.
n∈Z

This formula
Lindicates that it is convenient to compute the direct sum of Čech
complexes C 0 (U, O(n)), and its cohomology, tracing the natural grading,
n∈Z
and at the end separate the homogeneous components in the answer.
2.9.2. Ckp (U, O(n)) Fix k ∈ Z. Denote by Ckp (U, O(n)) the subgroup of
m(i0 , . . . , ip )
chains whose components can be represented as k
, where m is a
L i0 . . . Tip )
(T
form. As p varies, these groups form a complex Ckp (U, O(n)); computing
n∈Z
the action of its differential on the numerators of the cochain’s component we
easily obtain:
L p
Lemma. 1) The complex Ck (U, O(n)) with its grading is isomorphic to the
n∈Z
Koszul complex K p+1 (T1k , . . . , Trk ; R) with the grading in which the elements
g ∈ Hom(Kp+1 (T k ), R) such that

g(ei0 ∧ . . . ∧ eip ) ∈ Rk(p+1)+n

are homogeneous of degree n.

2) The map
146 Ch. 2. Sheaves, schemes, and projective spaces

m(i0 , . . . , ip ) 7−→ Ti0 . . . Tip · m(i0 , . . . , ip )

determines homogeneous homomorphisms of graded complexes

L p L p
Ck (U, O(n)) −→ Ck+1 (U, O(n))
n∈Z n∈Z

and
⊕C p (U, O(n)) = lim ⊕Ckp (U, O(n))
−→
k

relative this system of homomorphisms.

Observe that condition 1) uniquely determines on K p+1 the structure of
a graded R-module.
Lemma 2.9.2 illustrates the necessity to study homology of the Koszul
complex K p (T k , R). The method of chain homotopy operator is inapplicable
since the elements (T1k , . . . , Tnk ) generate a non-trivial ideal; the Koszul com-
plex is not, actually, acyclic in one dimension. 5) Therefore, another approach
is needed here.
We return, temporarily, to notation of § 2.8: Let A be a ring, f1 , . . . , fr a
collection of its elements. First, observe a duality:
2.9.3. Lemma. Define an A-homomorphism

ϕ : Kr−p (f, A) −→ K p (f, A)

by setting:

ϕ(ei1 ∧ . . . ∧ eir−p )(ej1 ∧ . . . ∧ ejp ) =

(
0, if (i) ∩ (j) 6= ∅,
=
ε(i1 , . . . , ir−p ; j1 , . . . , jp ), if (i) ∩ (j) = ∅.

Then ϕ is a complex isomorphism up to a sign of the differentials.

Proof. It suffices to find how ϕ commutes with the differentials. We have:

ϕ(d(ei1 ∧ . . . ∧ eir−p ))(ej1 ∧ . . . ∧ ejp+1 ) =

µX p ¶
k+1
=ϕ (−1) fik ei1 ∧ . . . ∧ ebik ∧ . . . ∧ eir−p (ej1 ∧ . . . ∧ ejp+1 ) =
k=1
(
0, if |(i) ∩ (j)| > 1,
=
(−1)k+1+σ fik , if ik = (i) ∩ (j),

where
(−1)σ = ε(i1 . . . bik . . . ir−p ; j1 . . . jp+1 ).
5
Exercise. Determine in which one.
2.9 Cohomology of the projective space 147

On the other hand:

ϕ(ei1 ∧ . . . ∧ eir−p )(d(ej1 ∧ . . . ∧ ejp+1 )) =

µXp+1 ¶
= ϕ(ei1 ∧ . . . ∧ eir−p ) (−1)l+1 ej1 ∧ . . . ∧ ejl ∧ . . . ∧ ejp+1 =
l=1
(
0, if |(i) ∩ (j)| > 1,
=
(−1)l+1+τ fjl , if jl = (i) ∩ (j),

where
(−1)τ = ε(i1 , . . . , ir−p ; j1 , . . . , b
jl , . . . , jp+1 ).
Comparing these answers we see how ϕ commutes with the differentials. u
t
Next, we use Lemma 2.8.5a which implies that

K0 (f1 , . . . , fr ) = K0 (f1 , . . . , fr−1 ) ⊗ K0 (fr ).

The following result is a reason for computing homology of K0 by induction

on r.
For any A-module M and element f ∈ A, let f M denote the kernel of the
f
homomorphism M −→ M of multiplication by f .
2.9.4. Lemma. Let L be a chain complex of A-modules, i ≥ 1. Then there
is an exact sequence of A-modules

0 −→ Hi (L)/f · Hi (L) −→ Hi (L ⊗ K(f )) −→ f Hi−1 (L) −→ 0.

Proof. The complex K(f ) is of the form

d
0 −→ Ae1 −→ Ae0 −→ 0, de1 = f e0 .

Therefore for i ≥ 1, we have

(L ⊗ K(f ))i = Li e0 ⊕ Li−1 e1

and
d(li e0 + li−1 e1 ) = (dli + (−1)i−1 f li−1 )e0 + dli−1 e1 .
This implies a commutative diagram with exact rows

0 / Li / (L ⊗ K(f ))i / Li−1 /0

d d d
² ² ²
0 / Li−1 / (L ⊗ K(f ))i−1 / Li−2 /0

All these diagrams can be united into a sequence of complexes. This sequence
is exact in dimensions ≥ 1:
148 Ch. 2. Sheaves, schemes, and projective spaces

0 −→ L −→ (L ⊗ K(f )) −→ L[−1] −→ 0, (2.127)

where L[−1]i := Li−1 (the complex L, shifted by 1 to the right). In its turn,
this sequence leads to an exact sequence of homology groups

/ Hi+1 (L[−1]) δ / Hi (L) / Hi (L ⊗ K(f )) / Hi (L[−1]) δ / ...

...

Hi (L) Hi−1 (L)

Let us show that the diagram

δ / Hi (L)
Hi+1 (L[−1])
llll6
ll
lll[−1]i f
Hi (L)

is commutative. This follows from calculations: If z ∈ Zi+1 (L[−1]), then

d(z ⊗ e1 ) = dz ⊗ e1 + (−1)i z ⊗ f e0 = (−1)i zf ⊗ e0 ,

δ(class(z)) = (−1)i class(f z).
Thus the following sequence is exact:

0 −→ Hi (L)/f Hi (L) −→ Hi (L ⊗ K(f )) −→ f Hi−1 (L) −→ 0.

2.9.4a. Corollary. If Hi (L) = 0 for i ≥ 1, then Hi (L⊗K(f )) = 0 for i ≥ 2.

Being applied to the Koszul complex, Lemma 2.9.4 and its Corollary yield
the following result:
2.9.5. Proposition. Let f1 , . . . , fr ∈ A be a sequence of elements such that
fi is not a zero divisor in A/(f1 , . . . , fi−1 ) for all i ≥ 1. Then Hi (K(f )) = 0
for i ≥ 1. Besides, we always have

H0 (K(f1 , . . . , fr )) = A/(f1 , . . . , fr )A.

Proof immediately follows by induction on r. Indeed, if the statement holds

for the sequences of length i − 1, then Corollary 2.9.4 immediately shows
acyclic property in dimensions ≥ 2 for any sequence of length i, whereas the
exact sequence (2.127) shows acyclic property in dimension 1 as well. The
claim on H0 is obvious from definition. u
t
2.9.6. A regular sequence. Any sequence f1 , . . . , fr ∈ A satisfying condi-
tions of Proposition 2.9.5 is said to be regular .
The result of the above subsection shows that if the elements f1 , . . . , fr
form a regular sequence, then the Koszul complex is a free resolution of the
A-module A/(f1 , . . . , fr ).
In several important cases the statement converse to Proposition 2.9.5
holds:
2.9 Cohomology of the projective space 149

2.9.7. Proposition. Let A be a local Noetherian ring, p its maximal ideal,

and f1 , . . . , fr ∈ p. If Hi (K(f1 , . . . , fr )) = 0 for i ≥ 1 (or even if only
H1 (K(f1 , . . . , fr )) = 0), then f1 , . . . , fr is a regular sequence.
Proof. For r = 1, Proposition obviously holds (for any A and f1 ∈ A).
Let the statement is proved for r = n − 1, let f1 , . . . , fn ∈ p, and
H1 (K(f1 , . . . , fn )) = 0. By Lemma 2.9.4 for f = fn , and i = 1, we get
an exact sequence

0 −→ H1 (K(f1 , . . . , fn−1 ))/f H1 (K(f1 , . . . , fn−1 )) −→

−→ H1 (K(f1 , . . . , fn )) −→ f H1 (K(f1 , . . . , fn−1 )) −→ 0,
which implies that
H1 (K(f1 , . . . , fn−1 ))/f H1 (K(f1 , . . . , fn−1 )) = 0,
f H0 (K(f1 , . . . , fn−1 )) = 0.

Since H1 (K(f1 , . . . , fn−1 )) in conditions of Proposition is a Noetherian A-

module, the first equality and Nakayama’s lemma imply that
H1 (K(f1 , . . . , fn−1 )) = 0,
so f1 , . . . , fn−1 is a regular sequence by induction hypothesis. Since
H0 (K(f1 , . . . , fn−1 )) = A/(f1 , . . . , fn−1 ),
the second equality implies that fn is not a zero divisor in A/(f1 , . . . , fn−1 ).
Hence f1 , . . . , fn is a regular sequence. u
t
2.9.8. Corollary. If (f1 , . . . , fn ) is a regular sequence of elements of the
maximal ideal of a Noetherian local ring, the sequence obtained from this one
by any permutation is also regular.
Proof. It suffices to observe that the Koszul complexes corresponding to the
sequences that only differ by a permutation are isomorphic to each other. ut
Now we are able to formulate the main result on cohomology of the pro-
jective space.
2.9.9. Theorem. a) H p (Pr−1
A , O(n)) = 0 for p 6= 0, r − 1.
0 r−1
³ b) H (P
´A , O(n)) = 0 for n < 0 and is a free A-module of rank
n+r−1
for n ≥ 0.
r−1
r−1 r−1
³ c) H ´ (PA , O(n)) = 0 for n ≥ −r + 1 and is a free A-module of rank
−n − 1
for n ≤ −r.
r−1
On the (p, n)-plane mark points where H p (Pr−1
A , O(n)) 6= 0. Obviously, the
diagram is central symmetric, and this symmetry (p, n) −→ (r − 1 − p, −r − n)
preserves the rank of A-modules of cohomology. In a deeper theory, this sym-
metry is explained by the duality theorem for cohomology of coherent sheaves.
150 Ch. 2. Sheaves, schemes, and projective spaces

r−1
p

−r

Fig. 5

Proof. We only have to trace homomorphisms connecting H p (Pr−1 A , O(n))

with Koszul complexes.
The fact that Koszul complex K0 (T k , R) is acyclic (Proposition 2.9.5) and
duality described in Lemma 2.9.3 immediately imply statement ) if we realize
that the role of ring A is played now by R = A[T1 , . . . , Tr ], and the role of the
fi are played by the Ti .
A somewhat more tedious count with grading and explicit form of homo-
morphisms Ck −→ Ck+1 (see Lemma 2.9.2) taken into account allows one to
establish statements b) and c). We leave this count to the reader. u
t

2.10. Serre’s theorem

L
∞
2.10.1. Theorem (Serre). Let R be a graded ring: R = Rn , where
n=0
R0 = A is a Noetherian ring and R is generated by a Noetherian A-mod-
ule R1 . Let X = Proj R and F a coherent sheaf on X. Then the following
statements hold :
a) H q (X, F) = 0 if q + 1 is greater than the number of generators of the
A-module R1 .
b) H q (X, F) is a Noetherian A-module for any q.
c) H q (X, F(n)) = 0 for q ≥ 1 and n ≥ n0 (F), where n0 (F) is a number
depending on F.
Observe that this theorem allows one to introduce important invariants
of the scheme X. For example, if A is a field and F is the structure sheaf,
then the cohomology spaces H q (X, OX ) are determined by their dimensions
recovered from X.
Proof. First of all reduce Serre’s theorem to the case where X = PrA , where
r +1 is the cardinality of a system of generators of the A-module R1 . Consider
the ring A[T1 , . . . , Tr+1 ], whose projective spectrum is PrA , and construct an
2.10 Serre’s theorem 151

homogeneous epimorphism of graded rings A[T1 , . . . , Tr+1 ] −→ R sending Ti

into the i-th generator of R1 . This epimorphism induces a closed embedding
of spectra j : Proj R −→ PrA which, in turn, enables one to continue the sheaf
F on Proj R, given in conditions of the theorem to the sheaf j∗ (F) on PrA . The
sheaf j∗ (F) is determined by the following rule

Γ (U, j∗ (F)) = Γ (U ∩ j(X), F) for any open set U ⊂ PrA .

The following properties of the continuation of the sheaf operation are rather
obvious.
First,
H q (X, F) = H q (PrA , j∗ (F)) for any q,
which is easy to see on the level of Čech complexes: in this case they are just
isomorphic modules with differentials. (For opens on PrA we take, as always,
the sets D+ (Ti ); for opens in Proj R we take, in order to establish an isomor-
phism desired of complexes, the sets D+ (ti ), where ti is the generator of R1
corresponding to Ti .) We may also use of the fact that F −→ j∗ (F) is a fully
faithful functor.
Second, we have j∗ (F(n)) = j∗ (F)(n), where F(n) = F ⊗ OX (n). Now,
heading ) of Serre’s theorem immediately follows from skew-symmetry of
Čech cochains and the fact that Čech cohomology for the cover (D+ (Ti ))
coincide with the usual cohomology.
To prove headings b) and c) of Serre’s theorem we use the following tech-
nical result.
2.10.1a. Lemma. For any coherent sheaf F on PrA , there exists an integer
m such that for a natural p we have the following exact sequence

OpPr −→ F(m) −→ 0.
A

From this Lemma the needed facts are established by a simple descending
induction on q.
Let us define the coherent sheaf E from the exact sequence

0 −→ E −→ OpPr −→ F(m) −→ 0.
A

We tensor it by O(n), where n ∈ Z:

0 −→ E(n) −→ OpPr (n) −→ F(m + n) −→ 0.

A

This gives rise to the exact cohomology sequence:

. . . −→ H q (PrA , O(n)) −→ H q (PrA , F(m + n)) −→ H q+1 (PrA , E(n)) −→ . . .

(2.128)
For q = r + 1, heading a) shows that properties b) and c) obviously hold.
The induction hypothesis: let headings b) and c) of Serre’s theorem hold for
the cohomology of all coherent sheaves in dimension q + 1.
152 Ch. 2. Sheaves, schemes, and projective spaces

In the exact cohomology sequence (2.128), the A-module H q+1 (PrA , E(n)) is
Noetherian and vanishes for n ≥ n0 by induction hypothesis; for H q (PrA , O(n))
the same is true thanks to Theorem 2.9.9. This implies the desired for
H q (PrA , F(m + n)). u
t
It remains to prove Lemma 2.10.1a.
Proof of Lemma 2.10.1a. We may assume that F is a coherent sheaf on
PrA . For the standard cover Ui := D+ (Ti ), having identified F|Ui with F(m)|Ui ,
we see that F(m) is glued of the F|D+ (Ti ) by means of the cocycle:
³ ´m ∼
Ti
: (F|Ui )|Ui ∩Uj −→ (F|Uj )|Uj ∩Ui .
Tj

Each section s ∈ Γ (PrA , F(m)) is then given by its “components”

s ∈ Γ (Ui , F): There is an embedding
Y
Γ (PrA , F(m)) ,→ Γ (Ui , F) :
³ ´m
Ti
s 7→ (. . . si . . .), where si · = sj .
Tj

To construct the epimorphism Op −→ F(m), it suffices to establish that it

is possible to construct a finite number of global sections of the sheaf F(m)
whose restrictions onto Ui generate the Γ (Ui , OPrA )-module Γ (Ui , F).
To this end, it suffices to be able to establish, for any section s ∈ Γ (Ui , F)
and all m sufficiently large, that s is a component of a sheaf F(m), i.e., to be
able to “extend s”. Having extended in this way (finitely many) generators of
all modules Γ (Ui , F) and having selected a common m sufficiently large, we
get the desired.
³ ´p
T0
2.10.2. Extending a section s0 ∈ Γ (U0 , F). Being multiplied by
Ti
(for a certain p), the section s0 |U0 ∩Ui can be extended to Ui :
³ ´p
T0
s0 = s0i |U0 ∩Ui , where s0i ∈ Γ (Ui , F),
Ti

for a p is common for all Ui , and however large.

On U0 ∩ Ui ∩ Uj , we have:
³ ´p ¯
Ti ¯
s0i ¯ = s0j |U0 ∩Ui ∩Uj .
Tj U0 ∩Ui ∩Uj

Therefore, for q sufficiently large (and also common for all i), we have
³ ³ ´p ´³ ´q
Ti T0
s0i − s0j = 0. (2.129)
Tj Tj

Set ³ ´q
T0
s00i = s0i .
Ti
2.11 Sheaves on Proj R and graded modules 153

Clearly,
s000 = s0 .
Besides, s000 enters as a component into the section (. . . s00i . . .) of the sheaf
F(p + q). Verification (see (2.129)):
³ ´p+q ³ ´q ³ ´p+q ³ ´q
Ti T0 Ti T0
s00i = s00j ⇐⇒ s0i = s0j . u
t
Tj Ti Tj Ti

2.10.3. Comments to Serre’s theorem. 1) The role of coherent property

of F: otherwise b) fails. Statement b) is non-trivial: The modules C q in the
Čech complex are not of finite type: Noetherean property appears only after
passage to cohomology.
2) The nature of the proof: there are “plenty” sheaves OPr (n) with known
cohomology; everything can be reduced to them with the help of an exact coho-
mology sequence. Lemma 2.10.1a explains the meaning of the term “plenty”.
Cp. with the following algebraic fact: every module is the image of a free
one; here the situation is global, and “twisting” by means of OPr (1) is essential.
3) What is n0 (F), starting with which c) holds? The question is a diffi-
cult one; for the sheaves of ideal we have, nevertheless, a “rough” answer: If
the Hilbert polynomial h is known, then n0 (F) is a universal polynomial in
coefficients of h. Generally, the numbers n0 (F) depend on F; for example, if
F = OPr (−N ), then H r (PrA , F(n)) = 0 only for n ≥ N − r.

2.11. Sheaves on Proj R and graded modules

2.11.1. The main fact of the theory of quasi-coherent sheaves over
affine schemes X = Spec A The main fact in question is the statement
The functor F −→ Γ (X, F) determines the equivalence of the
(2.130)
category of sheaves with the category of A-modules.
The purpose of this section is to show the existence of an analogous corre-
spondence between the category of quasi-coherent sheaves on Proj R and the
category of graded R-modules.
This correspondence is not, however, that simple and straightforward as
in the affine case; in particular, the graded modules that differ in only finitely
many components lead to isomorphic sheaves. For a precise formulation, see
sect. 2.11.6.
2.11.2. Let X = Proj R, where the ring R is generated by the set R1 over
R0 ; let F be a quasi-coherent sheaf over X. Set
L
∞
Γ∗ (X, F) := Γ (X, F(n)).
n=0

If F = OX , then on Γ∗ (X, OX ) there exists a natural structure

of a graded ring with multiplication induced by the homomorphisms
OX (n) × OX (m) −→ OX (m + n).
154 Ch. 2. Sheaves, schemes, and projective spaces

More generally, the multiplications OX (n) × F(m) −→ FX (m + n) deter-

mine on Γ∗ (X, F) the structure of a graded Γ∗ (X, OX )-module.
Now, recall that, for any n, there are defined group homomorphisms

αn : Rn −→ Γ (X, OX (n)).

(see Theorem 2.7.4). The definition makes clear that the homomorphisms
are compatible with multiplication and hence provide with a homogeneous
homomorphism of graded rings

α : R −→ Γ∗ (X, OX ).

Therefore, on Γ∗ (X, F), there is a canonical structure of a graded R-mod-

ule.
2.11.3. Conversely, let M be a graded R-module. Let us construct from it a
f on X by setting
quasi-coherent sheaf M
f) = M(f ) for any f ∈ R1
Γ (D+ (f ), M

and defining the restriction homomorphisms as in § 2.4 for M = R.

The construction of the homomorphism α : M −→ Γ∗ (X, M f) is easy to be
translated, component-wise, to this case:
f(n)),
αn : Mn −→ Γ∗ (X, M αn (m)|D+ (f ) = m/f n ∈ M(f ) .

2.11.3a. Exercise. Verify compatibilities.

2.11.4. Proposition. Any quasi-coherent sheaf F on X is isomorphic to a
f, where M = Γ∗ (X, F).
sheaf of the form M
Proof. First of all let us construct an isomorphism

f = Γ∗^
β: M (X, F) −→ F.

It suffices to construct this isomorphism for the sections of the corresponding

sheaves over opens D+ (f ). We have
·∞ ¸ n o
f) = M
Γ (D+ (f ), M f(f ) = L Γ (X, F(n)) =
m
| m ∈ Γ (X, F(n)) .
n f
n=0 (f )

Set ³ ´
m
β = α(m)|D+ (f ) · α(f )−n |D+ (f ) ∈ Γ (D+ (f )).
fn
2.11.4a. Exercise. Verify that the notion is well defined.
It only remains to establish that β is an isomorphism. In the same way as
for the structure sheaf, one can prove that Ker αn = 0 for n sufficiently large.
This implies that β is a monomorphism; indeed,
2.11 Sheaves on Proj R and graded modules 155

α(m)α(f )−n = 0 =⇒ α(f e m) = 0, e ≥ e0 =⇒ m/f n = 0.

Lemma 2.10.1a immediately implies that it is an epimorphism: indeed, to

extend the section s ∈ Γ (D+ (f ), F) to a section of F(n) over this set means
precisely to represent s as an image of a m/f n . u
t
The following result, analogous to a result of sects. 2.11.2–2.11.3 is deeper,
but concerns only the homomorphism α.
2.11.5. Theorem. Let the ring R satisfy the conditions of Serre’s theo-
f. Then the sheaf F is
rem 2.10.1, M a graded Noetherian R-module, F = M
coherent and the map
αn : Mn −→ Γ (X, F(n))
is an isomorphism for n sufficiently large.
Proof. First of all, as in Serre’s theorem, it is easy to see that it suffices to
carry out the proof for the case X = PrA . Let us have this in mind in what
follows.
There exists an exact sequence of the form
f
L −→ L0 −→ M −→ 0,

where L and L0 are free graded R-modules, i.e., direct sums of R-mod-
ules R(n).
This gives an exact sequence of sheaves

f˜(n)
e
L(n) e 0 (n) −→ M
−→ L f(n) −→ 0,

which immediately implies that the sheaf F is coherent and with the help of
which one constructs the exact cohomology sequence in the bottom line of the
following commutative diagram:

Ln / L0n / Mn /0J
JJ
JJ
JJ
α α α JJ
JJ
² ² ² $
e
Γ (X, L(n)) / Γ (X, L
e 0 (n)) / Γ (X, F(n)) / H 1 (X, f (L)(n)).
e

e
By Serre’s theorem H 1 (X, f (L)(n)) = 0 for n ≥ n0 , and since X = PrA ,
Lemma 2.7.5a shows that the first two vertical arrows are isomorphisms.
Hence, α : Mn −→ Γ (X, F(n)) is also an isomorphism. u
t
2.11.6. Main theorem on correspondence F −→ Γ∗ (X, F). We for-
mulate it without proof: A good deal of it is already verified; to prove the rest
of it does not require any new ideas.
Let R be a graded ring satisfying the conditions of Serre’s theorem. Let
GMR denote the category defined as follows:
156 Ch. 2. Sheaves, schemes, and projective spaces

the objects of GMR are graded Noetherian R-modules;

the morphisms in GMR : for any two R-modules M = ⊕Mi and N = ⊕Ni
we set µ ¶
L L
Hom(M, N ) = lim HomR Mi , Ni ,
−→ i≥i0 i≥i0
i0

where the sign lim is applied to the group HomR of R-homomorphisms of

−→
graded R-modules.
Informally speaking, in the category GMR the homomorphism M −→ N
defined only in degrees sufficiently high is represented; two homomorphisms
coinciding in high degrees are equal.
Modules M and N , isomorphic as objects of GMR , are said to be T N -iso-
morphic.
Theorem. The functor F −→ Γ∗ (X, F) determines an equivalence of the
category of coherent sheaves on X = Proj R with GMR ; its inverse is the
functor
f.
M −→ M

2.12. Applications to the theory of Hilbert polynomials

2.12.1. The results of the later sections allow one to give a “geometric” defi-
nition of the Hilbert polynomial and prove invariance of a number of numerical
characteristics introduced in § 2.5.
In what follows, R is a ring satisfying the conditions of Serre’s theorem and
such that, moreover, R0 = k is a field (this is needed to count dimensions of
homogeneous components; all the results to follow can be easily generalized
to the case where R0 is an Artinian ring by considering the lengths of the
modules instead of the dimensions of linear spaces.
2.12.2. Theorem. Let X = Proj R, M a Noetherian R-module, F = M f,
hM (n) the Hilbert polynomial of M . Then for all n ∈ Z, we have
∞
X def
hM (n) = (−1)i dim H i (X, F(n)) = χ(F(n)).
i=0

The number χ(F) is called the Euler characteristic of the sheaf F.

Proof. Serre’s theorem and Theorem 2.11.4 imply that

hM (n) = dim Mn = dim H 0 (X, F(n)) = χ(F(n)) for n ≥ n0 .

Therefore, the result desired is obtained if we establish that χ(F(n)) can be

represented as a polynomial in n for all n. The idea is the same as in the
proof of existence of Hilbert polynomial. It is bases on the following lemma.
2.12 Applications to the theory of Hilbert polynomials 157

2.12.2a. Lemma. Let

0 −→ F1 −→ F −→ F2 −→ 0

be an exact sequence of coherent sheaves on X. Then

χ(F) = χ(F1 ) + χ(F2 ).

Proof. Let r = dim R1 , then H i (X, F) = 0 for i ≥ r. The exact cohomology

sequence

0 −→ H 0 (X, F1 ) −→ H 0 (X, F) −→ . . . −→ H r−1 (X, F) −→ H r−1 (X, F2 ) −→ 0

implies that the alternated sum of dimensions of these cohomology spaces

vanishes. This proves the desired. u
t

2.12.2b. Corollary. For any exact sequence

0 −→ F0 −→ . . . −→ Fk −→ 0

of coherent sheaves on X we have:

k
X
(−1)i χ(Fi ) = 0.
i=0

Now, by induction on r we prove that χ(F(n)) is a polynomial in n. A stan-

dard reduction makes it possible to assume that X = Pr−1
k = Proj k[T1 , . . . , Tr ].
Tr
Let M −→ M (1) be a homomorphism of multiplication by Tr (it preserves
the degree: M (1)i = Mi+1 ). If K is the kernel and C the cokernel of this
homomorphism, then the following sequence is exact:
Tr
0 −→ K −→ M −→ M (1) −→ C −→ 0.

We derive from here an exact sequence of sheaves

e −→ F −→ F(1) −→ C
0 −→ K e −→ 0

and an exact sequence of twisted sheaves

e
0 −→ K(n) e
−→ F(n) −→ F(n + 1) −→ C(n) −→ 0.

By Corollary 2.12.2b we have

e
χ(F(n)) − χ(F(n + 1)) = χ(K(n)) e
− χ(C(n)).

The sheaves Ke and C e correspond to k[T1 , . . . , Tr−1 ]-modules, i.e., are “con-
r−1
centrated” on Pk . This allows one to make the induction step: For r = 0.
the statement is trivial. u
t
158 Ch. 2. Sheaves, schemes, and projective spaces

2.12.3. Comment. Theorem 2.12.2 shows that the Hilbert polynomial is

invariantly determined by the triple of geometric objects (X, L, F), where
X = Proj R, L = OX (1), and F is a coherent sheaf on X. Let us show that
the degree of this polynomial and its constant term only depend, actually, on
(X, F), but not on L.
In particular, the dimension dim X and characteristic χ(X), introduced
above, do not depend on representation of X as Proj R.
The claim on constant term is obvious:
∞
X
hM (0) = χ(F) = (−1)i dim H i (X, F).
i=0

2.12.4. Theorem. Set

dim F = deg χ(F(n)).

Then dim F does not depend on the choice of invertible sheaf L.

Proof. The formulation of the theorem assumes that L is very ample, i.e., is
of the form O(1) for a representation of X in the form Proj R. We have no
means for characterization of such sheaves apart from Serre’s theorem; let us
use it.
Let L1 and L2 be two very ample sheaves on X. For any n ≥ n0 , we have
hi (n) = dim H 0 (X, F ⊗ Lni ), where i = 1, 2. The sheaf M = L−1 N
1 ⊗ L2 for N
sufficiently large is generated by its sections, as follows from Lemma 2.10.1a.
Since (F ⊗ Ln1 ) ⊗ Mn = F ⊗ LnN 2 , we have an isomorphism of groups of
global sections

H 0 (X, F ⊗ Ln1 ⊗ Mn ) ' H 0 (X, F ⊗ LnN

2 ).

The canonical map

H 0 (X, F ⊗ Ln1 ) ⊗ S −→ H 0 (X, F ⊗ Ln1 ⊗ Mn )

is injective for the zero section S ∈ H 0 (X, Mn ) because Mn is locally free.

But dim H 0 (X, M) ≥ 1, so dim H 0 (X, Mn ) ≥ 1. This implies that

dim H 0 (X, F ⊗ Ln1 ) ≤ dim H 0 (X, F ⊗ LnN

2 )

or
h1 (n) ≤ h2 (nN ) for any n ≥ n0 .
By symmetry, h2 (n) ≤ h1 (nN 0 ), so deg h1 = deg h2 . Theorem is proved. u
t
Let us give now another description of the dimension, often useful. Recall
that f ∈ R is said to be an essential zero divisor in the graded R-module M ,
f
if the kernel of multiplication by f : M −→ M has infinitely many non-zero
homogeneous components, i.e., is non-trivial in the category GMR .
2.12 Applications to the theory of Hilbert polynomials 159

2.12.5. An M -sequence. Let M be a graded R-module. A finite set of

homogeneous and non-invertible elements {f0 , . . . , fd } in R such that fi is
not an essential zero divisor in M/(f0 , . . . , fi−1 ) for all i ≥ 0 will be called an
M -sequence.
The number d is the length of the M -sequence {f0 , . . . , fd }. A given M -se-
quence is said to be maximal if it is impossible to add to it any element of R
and get an M -sequence.
The symbol d(M ) will denote the length of the shortest of the maximal
M -sequences.
f.
2.12.6. Theorem. d(M ) = dim M
We will need the following Lemma.
2.12.7. Lemma. If dim M f ≥ 0, then there exists an element f ∈ R such
that {f } is an M -sequence.
We will prove Lemma later, let us now use it to prove Theorem 2.12.6.
2.12.8. Proof of Theorem 2.12.6. We induct on d(M ), starting from
d(M ) = −1 in which case there are no M -sequences.
For d(M ) = −1, Lemma 2.12.7 immediately implies that dim M f = −1.
Let Theorem be proved for all M with d(M ) ≤ d−1 and let d(M ) = d. Let
{f0 , . . . , fd } be the shortest maximal M -sequence. Then d(M/f0 M ) = d − 1
by induction hypothesis dim (M/f ^ 0 M ) = d − 1. Consider an exact sequence

f0
0 −→ N −→ M −→ M (k) −→ M/f0 M −→ 0,
f0
where k = deg f0 and M −→ M (k) is the grading-preserving homomorphism
of multiplication by f0 . Since f0 ∈ R is an inessential zero divisor, we have
Nn = 0 for n ≥ n0 and dim M (k)n − dim Mn = dim(M/f0 M )n . Therefore
f = (d − 1) + 1 = d, proving the desired.
dim M u
t
2.12.9. Proof of Lemma 2.12.7. In R, we have to find an inessential zero
divisor in M .
First of all let us show that there exists a sequence of graded modules Mi
such that

0 = M0 ⊂ M1 ⊂ . . . ⊂ Mr = M, Mi /Mi+1 ' (R/pi )ni ,

where pi are prime graded ideals.

Since M is a Noetherian module, it suffices to find in it a non-trivial graded
submodule M1 ⊂ M such that M1 ' (R/p)n , where p is a prime graded ideal.
Let S be the set of graded ideal pm in R for each of which there exists a
homogeneous element m ∈ M whose annihilator is pm . Since R is Noetherian,
there is a maximal element in S; denote it p. Clearly, Rm ' (R/p)n , where
p = Ann m, n = deg m. Let us prove that p is prime. Indeed, let ab ∈ p, a ∈ / p.
160 Ch. 2. Sheaves, schemes, and projective spaces

Then the inclusion Ann(bm) ⊇ (a, Ann m) is strict by maximality, so bm = 0

and b ∈ Ann m. So p is prime.
(Observe that the same argument without taking grading into account
proves a similar result for non-graded Noetherian modules.)
We have thus constructed a sequence of modules desired. Let us use it
to find in R an inessential zero divisor in M . Let q be a maximal graded
prime ideal. If q 6⊂ ∪p, then, for an element of R to be found, we can take
any element of q not lying in ∪pi . (Since it lies in q, it is non-invertible.) If,
however, q ⊂ ∪pi for any maximal ideal q, then q ⊂ pj for some j since the
ideals pi are prime, so the ideals pi exhaust the set of maximal ideals. Since
there are finitely many of them, dim R = 0.
In this case R ≈ Γ (X, OX )[T ] is the polynomial ring in one indeterminate
T (“≈” means isomorphism up to a finite number of homogeneous compo-
nents) because dim R = const for n ≥ n0 . Therefore multiplication by T has
no kernel in all sufficiently large dimensions, and T is the inessential zero
divisor to be found.
Both Lemma and Theorem are completely proved. u
t
In the course of the proof we have obtained a number of useful statements;
we separate them for convenience of references.
2.12.10. Corollary. 1) Ann M = ∪pi , where (pi )i∈I is a finite family of
prime ideals in R.
2) Let F be a coherent sheaf on X = Proj R. Then there exists a sequence
of coherent pre-sheaves

0 = F0 ⊂ F1 ⊂ F2 ⊂ . . . ⊂ F,

such that Fi+1 /Fi ' OX /Ji (ni ), where Ji are coherent sheaves of prime ideals
on X.
2.12.11. Theorem. H q (X, F) = 0 for q > dim F.
Proof. Let the theorem be established for some sheaves F0 and F00 entering
the exact sequence
0 −→ F0 −→ F −→ F00 −→ 0.
Then
χ(F(n)) = χ(F0 (n)) + χ(F00 (n)),
and hence dim F = max(dim F0 , dim F00 ). The exact cohomology sequence
easily implies the validity of the statement of the theorem for F: It suffices to
consider the terms

. . . −→ H q (X, F0 ) −→ H q (X, F) −→ H q (X, F00 ) −→ 0,

where q > dim F.

Set F = OX /J(n), where J ⊂ OX is a coherent sheaf of ideals. The sheaf J
determines a closed subscheme Y ⊂ X, where OY = OX /J, which can be
2.13 The Grothendieck group: First notions 161

L
∞
considered as the projective spectrum of the graded ring Γ (X, OX /J(n)).
n=0
Therefore, and thanks to heading 2) of Corollary 2.12.10 it suffices to verify
the statement for OY (n) = F. Set J = Im Γ∗ (X, J) ⊂ R.
Let d = dim Y . By Theorem 2.12.6 there exists a maximal R/J-sequence
f0 , . . . , fd , where fi ∈ R such that dimk R/(f0 , . . . , fd , J) < ∞. So setting
f i = fi (mod J) we have (R/J)n ⊂ (f 0 , . . . , f d ) for n ≥ n0 in R/J. Ge-
S
d
ometrically, this means that Y = D+ (f i ). Computing Čech cohomology
i=0
immediately yields the statement of Theorem. u
t
2.12.12. Remark. Since dim F ≤ dim X for any F (as follows from the
proof of Theorem), then, in particular, H q (X, F) = 0 for any q > dim X.

2.13. The Grothendieck group: First notions

2.13.1. The classical “Riemann-Roch problem” is to compute dim H 0 (X, F),
where X is a projective scheme over a field, and F is a coherent sheaf on it.
The main qualitative information about this function is the statement

dim H 0 (X, F(n)) is a polynomial, χ, for n ≥ n0 (F). (2.131)

In practice, therefore, the Riemann-Roch problem is usually subdivided

into two questions solved by distinct approaches:
a) Describe the coefficients of the Hilbert polynomial of F in “geometric”
terms.
b) Find a “good” estimate of the number n0 (F).
For an example of the answer to question a) we offer a derivation of the
degree of a given projective scheme X ⊂ Pr by means of Bezout’s theorem:
This degree is equal to the number of intersection points of X with “sufficiently
general” linear submanifold of Prk of complementary dimension. The general
answer is given by the Riemann-Roch-Hirzebruch-Grothendieck-... theorem.
In these lectures we will not touch question b); the rest of the chapter is
devoted to the study of the characteristic χ(F).
Lemma 2.12.2 describes its main property. Axiomatizing this property we
introduce the following definition. Let X be a scheme, CohX the category of
coherent sheaves on it.
2.13.2. Additive functions on Abelian categories. Let G be an Abelian
group; a map ψ : CohX −→ G is said to be an additive function on CohX (or
any other Abelian category) with values in G if, for any exact sequence

e 1 −→ F −→ F2 −→ 0
0 −→ F

of sheaves from the category CohX , we have

162 Ch. 2. Sheaves, schemes, and projective spaces

e = ψ(F
ψ(F) e 1 ) + ψ(F
e 2 ).

Any additive function has the following properties.

e i be objects
2.13.3. Lemma. Let ψ be an additive function on CohX , and F
of CohX .
a) For any exact sequence

e 1 −→ . . . −→ FK −→ 0,
0 −→ F

we have
K
X
e i ) = 0.
(−1)i ψ(F
i=1

e0 ⊂ F
b) Let 0 = F e1 ⊂ . . . ⊂ F
e K = F.
e Then
X
ψ(F)e = ψ(Fe i /F
e i−1 ).
i≥1

2.13.3a. Exercise. Prove Lemma.

It is not difficult to see that even for the simplest schemes X (e.g., for
X = Spec K) there are plenty of additive functions on CohX . Still, the whole
totality of them is easy to overview thanks to the existence of a “universal”
additive function k : CohX −→ K(X) with values in a universal group K(X).
Let Z[CohX ] be the free Abelian group generated by symbols [F] e corre-
sponding to classes of isomorphic coherent sheaves on X. Let J ⊂ Z[CohX ]
be the subgroup generated by the elements
e − [F
[F] e 1 ] − [F
e 2 ],

one for each exact sequence

e 1 −→ F −→ F2 −→ 0.
0 −→ F

2.13.4. K(X), the Grothendieck group. The group K(X) = Z[CohX ]/J
is called the Grothendieck group (of the category CohX or the scheme X).
2.13.4a. Proposition. The function

k : CohX −→ K(X), (2.132)

for which k(F) = [F] (mod J) is additive; the image k(CohX ) generates the
group K(X), and for any additive function ψ : CohX −→ G, there exists a
uniquely determined homomorphism ϕ : K(X) −→ G such that ψ = ϕ ◦ k.
Proof is trivial.
From the point of view of the group K(X), the Riemann-Roch problem is
2.13 The Grothendieck group: First notions 163

the problem of describing the function χ : K(X) −→ Z.

The advantage of such a reformulation is as follows. As we will shortly see,

the group K(X) is endowed with a reach collection of additional structures.
Sometimes it can be explicitly and completely computed. For example, for
X = Prk and then we describe χ using this information. In the general case,
there are sufficiently many known data on K(X) to geometrically interpret
χ(F).
Lemma 2.13.3 gives an approach to computing K(X); sometimes it allows
one to indicate a smaller system of generators than the set of elements k(F)
for all coherent sheaves F. Let us illustrate this by examples.
2.13.5. Examples. 1) Let A be any Noetherian ring, X = Spec A. In sub-
sect. 2.12.9 we have established that any Noetherian A-module has a compo-
sition series with factors isomorphic to A/p, where p ⊂ A is a prime ideal.
The generators of K(X) are, therefore, the elements k(A/p), where k is the
function from eq. (2.132). To describe all relations is a bit more difficult. We
confine ourselves to the case where A is an Artinian ring.
In this case, the Jordan-Hoelder theorem can be interpreted as computa-
tion of K(X): the map
j : K(X) −→ Z[X],
where Z[X] is a free Abelian group generated by the points of the scheme X, and
X
j(k(F)) = (lengthOx Fx )x
x∈X

is a group isomorphism.
2) Let A be a principal ideal ring, X = Spec A. Then any Noetherian
A-module M has a free projective resolution of length 1:

0 −→ Ar −→ As −→ M −→ 0.

This immediately implies that the group K[X] is cyclic and generated by the
class of ring A. The order of
Nthis class is equal to infinity which is easy to see
passing to linear spaces M K over the ring of quotients K of A; and hence
A
K(X) ' Z.

More generally, the same is true for any affine scheme Spec A provided
any Noetherian A-module has a free resolution of finite length.
This condition is still too strong to lead to interesting notions; however,
even a slight slackening of it defines a very important class of schemes.
2.13.6. Smooth schemes. Let X be a Noetherian scheme, F a coherent
sheaf on it. Let, for any point x ∈ X, there exists an open neighborhood U 3 x
such that the sheaf F|U has in this neighborhood a free resolution consisting
of “free” sheaves OrX |U . Then the scheme X is said to be smooth.
164 Ch. 2. Sheaves, schemes, and projective spaces

In the next section we will prove that the two types of schemes are smooth:
the projective spaces over fields and spectra of local rings whose maximal
ideals are generated by regular sequences.
Smoothness of projective spaces immediately follows from the following
classical Hilbert’s theorem on syzygies. 6)
2.13.7. Theorem. Let R = K[T0 , . . . , Tr ]. Any graded R-module graded free
projective resolution of length ≤ r + 1.
Proof will be given at the end of the chapter; now we use this theorem to
compute K(PrK ).
2.13.8. Theorem. The map xi 7−→ k(O(i)), where k is the function from
eq. (2.132), determines an isomorphism of Abelian groups

Z[x]/((x − 1)r+1 ) −→ K(PrK ).

In particular, this group is free of rank r + 1.

Proof. Translating the first statement of Theorem 2.13.7 into the language
of sheaves by means of Theorem 2.11.6 we find that any coherent sheaf F on
PrK possesses a resolution the terms of which are direct sums of sheaves O(n),
where n ∈ Z. Lemma 2.13.3 shows that the elements k(O(n)) generate the
group K(PrK ).
Obviously, these generators are not independent. At least one relation
is obtained if, using Proposition 2.9.5, we consider the Koszul complex
K0 (T0 , . . . , TR ; R) which is a resolution of the R-module K = R/(T0 , . . . , Tn ):

. . . −→ R( ) −→ . . . −→ R( r+1
2 ) −→ Rr+1 −→ R −→ K −→ 0,
r+1
i

where R( i ) = ∧i (Re + . . . + Re ) = K (T, R).

r+1
0 r i

We can consider this resolution as an exact sequence of graded modules if we

consider the elements ei1 ∧ . . . ∧ eik homogeneous of degree k. Applying to this
resolution the sheafification functor we get the exact sequence

0 −→ . . . −→ O(−i)( i ) −→ . . . −→ O(−2)( 2 ) −→ O(−1)r+1 −→ OPr −→ 0

r+1 r+1

(take into account that the R-module K is T N -isomorphic to 0, i.e., Ke = 0).

This exact sequence can be tensored by O(n) for any n ∈ Z without violating
the exactness. Therefore, there are the following relations in K(PrK ):
6
In broadest terms, syzygy is a kind of unity, especially through coordination or
alignment, most commonly used in the astronomical and/or astrological sense. In
mathematics, a syzygy is a relation between the generators of a module M . The
set of all such relations is called the first syzygy module of M . A relation between
generators of the first syzygy module is called a second syzygy of M , and the set
of all such relations is called the second syzygy module of M , and so on.
2.13 The Grothendieck group: First notions 165

r+1 ³
X ´
r+1
k(O(n − i)) = 0.
i
i=0

This clearly implies that the kernel of the homomorphisms of additive

groups
Z[x, x−1 ] −→ K(X); xi 7→ k(O(i)) for i ∈ Z
contains an ideal generated by the polynomial (x−1 − 1)r+1 or, equivalently,
by (x − 1)r+1 .
Since we have already established epimorphic property of this homomor-
phism, to complete the proof of Theorem, it suffices to verify that the elements
k(O), k(O(1)), . . ., k(O(r)) are linearly independent over Z.
Obviously, the functions χn (F) = χ(F(n)) are additive on SPr for any
n ∈ Z. Therefore, if there existed a non-trivial linear dependence
r
X
ai k(O(i)) = 0, where ai ∈ Z,
i=0

it would imply that

r
X r
X ³ ´
n+i+r
ai χn (O(i)) = ai = 0,
r
i=0 i=0

which is only possible if ai = 0 for all³i since, as ´an easy verification shows (say,
n+i+r
by induction on r), the polynomials in n are linearly independent
r
for i = 0, . . . , r. u
t
2.13.9. The group K(PrK ) in formulation of Theorem 2.13.8 turned out en-
dowed with a ring structure. The multiplication in this ring possesses an
invariant meaning: Indeed, as is easy to see
k(F)k(OPr (i)) = k(F(i)) = k(F ⊗ OPr (i)),
so this multiplication corresponds, at least sometimes, to tensor products of
sheaves. There are, however, examples for which k(F)k(G) 6= k(F ⊗ G), So the
general description of multiplication can not be that simple. This question is
studied in detail in the second part of the course [Ma3].
Meanwhile, using our description of K(PrK ), we give a (somewhat naive)
form of the Riemann-Roch theorem for the projective space.
The idea is to select some simple additive function on K(PrK ), and then
“propagate” it using ring multiplication.
Any element of K(PrK ) can be uniquely represented, thanks to Theo-
Pr
rem 2.13.8, as a polynomial ai (l − 1)i , where l = k(O(1)). Introduce an
i=0
additive function κr : K(PrK ) −→ Z by setting
µXr ¶
i
κr ai (l − 1) = ar .
i=0
166 Ch. 2. Sheaves, schemes, and projective spaces

2.13.10. Lemma. For any additive function

ψ : K(PrK ) −→ Z,

there exists a unique element t(ψ) ∈ K(PrK ) such that

ψ(y) = κr (t(ψ)y), for any y ∈ K(PrK ).

Proof. The function ψ is a linear combination of coefficients ai in the repre-

P
r
sentation ai (l − 1)i , where these coefficients are considered as functions on
i=0
K(PrK ); for these coefficients, we have:

ai (y) = κr ((l − 1)r−i y). u

t

2.13.11. Theorem. Let χ : K(PrK ) −→ Z be the Euler characteristic. Then

χ(y) = κr (l2 y), i.e., t(χ) = l2 . (2.133)

Proof. It suffices to verify the coincidence of the left side with the right side
for the elements li , where i = 0, 1, . . . , r, constituting a Z-basis of the group
K(PrK ). We have:
³ ´
r+i
χ(li ) = χ(O(i)) = dim H 0 (PrK , O(i)) = ,
i
³ ´
r+i
κr (lr+i ) = κr ((1 + (l − 1))r+i ) = . u
t
i
2.13.12. Remark. The usage of κr as a simplest additive function was not
yet motivated. Besides, it is clear that to apply Theorem 2.13.11 is not that
easy: To compute χ(F) with its help, we have to first know what is the class
of the sheaf F in K(PrK ). The only means known to us at the moment is to
consider the resolution of the sheaf F, but this is not “geometric”, besides,
this makes our formula useless: If we know the resolution, we can calculate
χ(F) just by additivity.
Nevertheless, formula (2.133) is very neat; I consider this as a serious
argument in its favor.

2.14. Resolutions and smoothness

Let us show, first of all, that smoothness of a given scheme X is a property
of the totality of its local rings Ox .
2.14.1. Theorem. A scheme X is smooth if and only if Spec Ox is a smooth
scheme at all points x ∈ X.
Proof. Let X be smooth. Consider an arbitrary coherent sheaf on Spec Ox .
This sheaf is determined by a Ox -module Fx . Let us show that, in a neighbor-
hood U 3 x, there exists a sheaf F whose fiber at x coincides with Fx . Indeed,
consider the exact sequence
2.14 Resolutions and smoothness 167

f
Orx −→ Osx −→ Fx −→ 0,

where the homomorphism f is determined by a r × s matrix whose entries

are the germs of sections of the structure sheaf of X at point x. There exists
an affine neighborhood Spec A of this point onto which the elements of this
matrix can be continued. Let

f : Ar −→ As

be the homomorphism determined by this continuation. Set M = Coker f and

let F be the sheaf Mf considered on Spec A. Its fiber at point x is isomorphic
to Fx . Since the scheme X is smooth, we can assume that U = Spec A is so
small that there is a finite resolution in it:

0 −→ Orxn |U −→ . . . −→ Orx0 |U −→ F −→ 0.

e x on
Passing to fibers at point x we obtain a finite resolution of the sheaf F
Spec Ox ; this proves that Spec Ox is smooth.
Conversely, let Spec Ox be a smooth scheme, F a coherent sheaf on X.
Since the only neighborhood of a closed point x in Spec Ox is the whole spec-
trum, there exists a resolution of the fiber

0 −→ Orxn −→ . . . −→ Orx0 −→ Fx −→ 0.

The argument analogous to the above one allows us to continue this se-
quence onto an open neighborhood U of the point x:

0 −→ Orxn |U −→ . . . −→ Orx0 |U −→ F|U −→ 0.

Since this sequence is exact at the point x, it remains exact in a (perhaps,

smaller than U ) neighborhood of x, proving the desired. u
t
This result indicates the importance of study of the smooth spectra of
local rings. For this, we will need several elementary results from homological
algebra.
The next result allows one to overview the family of resolutions of a given
A-module M .
2.14.2. Lemma. Let A be a ring, M an A-module; P and P 0 either projec-
tive or free A-modules. Let
f
0 −→ S −→ P −→ M −→ 0,
f
0 −→ S 0 −→ P 0 −→ M −→ 0

be two exact sequences. Then S ⊕ P 0 ' S 0 ⊕ P .

168 Ch. 2. Sheaves, schemes, and projective spaces

Proof. Consider the commutative diagram

f +f 0
0 /T / P ⊕ P0 /M
FF
FF
FF pr
g FFF
# ²
P0
where T = Ker(f + f 0 ). Clearly, g is an epimorphism; since P 0 is projec-
tive, there exists a “section” s : P 0 −→ T (i.e., a homomorphism such that
g ◦ s = 1T ) and T ' Ker g ⊕ P 0 . But
Ker g = {(p, 0) | f (p) = 0} ' S,
so T ' S ⊕ P 0 . By symmetry, T ' S 0 ⊕ P , proving Lemma. u
t
This result is a reason to introduce the following definition.
2.14.3. Modules equivalent projectively or freely. The two A-modules
S and S 0 are said to be projectively (respectively freely) equivalent, if there
exist projective (resp., free) A-modules P and P 0 such that
S ⊕ P 0 ' S 0 ⊕ P.
We can now sharpen Lemma 2.14.2 as follows.
2.14.3a. Lemma. Let A-modules M and M 0 be projectively (resp., freely)
equivalent and the following exact sequences
0 −→ S −→ P −→ M −→ 0,
0 −→ S 0 −→ P 0 −→ M 0 −→ 0,
be given in which P and P 0 are projecte (resp., free) modules. Then S and S 0
are projectively (resp., freely) equivalent.
Proof. Let M ⊕ Q ' M 0 ⊕ Q0 , where Q and Q0 are projective (resp., free).
Then the following sequences are exact (with obvious homomorphisms)
0 −→ S −→ P ⊕ Q −→ M ⊕ Q −→ 0,
0 −→ S 0 −→ P 0 ⊕ Q0 −→ M 0 ⊕ Q0 −→ 0,
implying that S ⊕ P 0 ⊕ Q0 ' S ⊕ P ⊕ Q by Lemma 2.14.2. u
t
2.14.3b. Corollary. Let there is given the beginning part of a projective
(resp., free) resolution of an A-module M
dn
Pn −→ Pn−1 −→ . . . −→ P0 −→ M.
Then the class up to projective (resp., free) equivalence of the module
Ker dn is uniquely determined and does not depend on the choice of the res-
olution. In particular, it there exists a projective resolution of length n of the
module M , then Ker dn is projective.
2.14 Resolutions and smoothness 169

For modules over local rings, we construct a special type of resolutions,

so-called minimal resolutions; Corollary 2.14.3a enables to derive from these
minimal resolutions information about arbitrary resolutions.
Let A be a local Noetherian ring, m its maximal ideal, M an A-module of
finite type, F a free A-module.
An epimorphism F → M −→ 0 is said to be minimal, if it induces an
isomorphism F/mF −→ M/mM .
The minimal epimorphism is determined uniquely in the following sense.
If F 0 −→ M −→ 0 is another minimal epimorphism, then there exists a
commutative diagram
F /M /0
||>
||
|||
² |
F0
in which the vertical arrow is an isomorphism. Indeed rk F = rk F 0 by Corol-
lary 1.10.9a and the pre-images of any basis of M/mM in F and F 0 are bases
in F and F 0 , respectively.
2.14.4. Minimal resolutions. Iterating minimal epimorphisms, we arrive
at the notion of the minimal resolution (here A is supposed to be Noetherian)
dn
. . . −→ Fn −→ Fn−1 −→ . . . −→ F0 −→ M −→ 0;
dn
this resolution is minimal if Fn −→ dn (Fn ) is a minimal epimorphism for all
n.
Claim: A resolution is minimal if and only if

dn (Fn ) ⊂ mFn−1 for all n ≥ 1.

2.14.4a. Exercise. Prove that this claim.

2.14.5. Examples. 1) The minimal resolution. Let f1 , . . . , fr ⊂ m be a
regular sequence of elements. Then the Koszul complex K0 (f, A) is a minimal
resolution of the A-module A/(f1 , . . . , fr ).
2) An infinite minimal resolution. Consider the local ring A = k + t2 k[[t]],
and let M = m = t2 k[[t]] ⊂ A. Then M = At2 + At3 , and m2 = mM = t4 k[[t]],
and hence, dimk m/m2 = 2.Here is the beginning of the minimal resolution:
d1 ε
Af1 ⊕ Af2 −→ Ae1 ⊕ Ae2 −→ M −→ 0;
ε(ei ) = ti+1 ;
d1 (f1 ) = t3 e1 − t2 e2 ;
d1 (f2 ) = t4 e1 − t3 e2 .

It is easy to see that Ker d1 ' M ; so in the next part of the resolution this
segment will be periodically repeated again and again.
170 Ch. 2. Sheaves, schemes, and projective spaces

Let us prove now the following theorem on smooth local rings.

2.14.6. Theorem. Let A be a local Noetherian whose maximal ring is gen-
erated by a regular sequence x1 , . . . , xd . Then any Noetherian A-module M
has a free resolution of length ≤ d.
2.14.6a. Lemma. Under assumptions of Theorem, construct a minimal res-
olution (Fn , dn ) of the module M and set
n
SA (M ) = Ker dn .

Then for any x ∈ m which is not a zero divisor in A or in M , we have:

n
SA/xA (M/xM ) ' S n (M )/xS n (M ).

Proof. We may confine ourselves to the case where n = 1. Consider the

commutative diagram
ϕ
F /M /0

² ψ
²
F/xF / M/xM /0

where ϕ is the minimal epimorphism. The definition easily implies that

ψ is also minimal (as an epimorphism of A/xA-modules). Therefore
1 1
SA (M ) = Ker ϕ, and SA/xA (M/xM ) = Ker ψ.
There exists a unique homomorphism θ for which the diagram
ϕ
1
SA (M ) /F /M /0

θ
² ² ²
ψ
1
SA/xA (M/xM ) / F/xF / M/xM /0

is commutative.
Let us show that θ is an epimorphism. Let f + xF ∈ Ker ψ. Then
ϕ(f ) ∈ xM , and hence

ϕ(f − xf0 ) = 0 =⇒ f ∈ xf0 + S 1 (M ) =⇒ f + xf ∈ θ(S 1 (M )).

It remains to verify that Ker θ = xS 1 (M ). Indeed, Ker θ = S 1 (M ) ∩ xF ;

but if ϕ(xf ) = 0, then ϕ(f ) = 0 since x is not zero divisor in M . u
t
2.14.7. Proof of Theorem 2.14.6. Let m = Ax1 + . . . + Axd , where
(x1 , . . . , xd ) is a regular sequence. Let us show by induction on d that
d+1
SA (M ) = 0 for all M .
If d = 0, then A is a field and all is evident.
Let the statement be true for d − 1. We have
2.14 Resolutions and smoothness 171

d+1 d
SA (M ) = SA (M 0 ), where M 0 = S 1 (M ).

Since M 0 is a submodule of a free module, it follows that x1 is not zero divisor

in M 0 . By induction hypothesis and thanks to Lemma 2.14.6 we have
d
0 = SA/x1A
(M 0 /x1 M 0 ) = SA
d
(M 0 )/x1 SA
d
(M 0 ).

The Nakayama lemma implies that

d
SA (M 0 ) = 0. u
t

Remark. The number d is an invariant of the ring A: Indeed, it can be defined

as the length of the maximal resolution (Koszul complex ) of the A-module
A/m.
Finally, let us sketch a proof of Hilbert’s syzygies theorem 2.13.7. L
Consider a graded ring R instead of the local ring A, the ideal R+ = Ri
i>1
instead of the maximal ideal m, and assume that “module” means “a graded
R-module”. Then Nakayama’s lemma and Corollary 1.10.9a tho notions of
a minimal epimorphism and a minimal resolution and Theorem 2.14.6 are
applicable to the new situation. All arguments can be repeated literally, except
for the formulation and proof of Nakayama’s lemma: We have to replace “the
ideal distinct from the whole A” by “the ideal containing in R+ ” and the
argument with inversion of 1 − f1 by a remark that multiplying by an element
of R+ we enlarge by 1 the number of the first non-zero component.
References
[AM] Atiyah M., Macdonald I. Introduction to commutative algebra,
Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills,
Ont., 1969, ix+128 pp.
[Bla] Blass, A., Existence of bases implies the axiom of choice. Ax-
iomatic set theory (Boulder, Colo., 1983), 31–33, Contemp. Math.,
31, Amer. Math. Soc., Providence, RI, 1984.
[BM] Borisov D., Mani Yu., Generalized operads and their inner cohomo-
morphisms. Progress in Mathematics. V. 265. P. 247–308.
[Bb1] Bourbaki N., Commutative algebra, Springer, Berlin et al, 1988
[Bb2] Bourbaki N., Éléments d’histoire mathématiques, Hermann, Paris,
1969
[Bb3] Bourbaki N., General Topology. Chapters 1 − 4. Springer, Berlin et
al, 2nd printing, 1998, VII, 452 pp.
[BSh] Borevich Z., Shafarevich I., Number theory, Acad. Press, NY, 1966
(see revised third edition: Teori qisel. (Russian) [The theory
of numbers] Nauka, Moscow, 1985. 504 pp.)
[CE] Cartan H., Eilenberg S., Homological algebra. With an appendix
by David A. Buchsbaum. Reprint of the 1956 original. Princeton
Landmarks in Mathematics. Princeton University Press, Princeton,
NJ, 1999. xvi+390 pp.
[Ch] Chebotaryev N. G., Theory of Algebraic Functions, OGIZ, Moscow-
Leningrad, 1948, 396 pp.
[Dan] Danilov V. Cohomology of algebraic varieties. In: Shafarevich I.,
(Editor, Contributor), Danilov V., Iskovskikh V., Algebraic geom-
etry — II: Cohomology of Algebraic Varieties. Algebraic Surfaces
(Encyclopaedia of Mathematical Sciences). Springer, Berlin et al,
5–130
[Del] Deligne P., Etingof P., Freed D., Jeffrey L., Kazhdan D., Morgan J.,
Morrison D., Witten E. (eds.) Quantum fields and strings: a course
for mathematicians. Vol. 1, 2. Material from the Special Year on
Quantum Field Theory held at the Institute for Advanced Study,
Princeton, NJ, 1996–1997. American Mathematical Society, Provi-
dence, RI; Institute for Advanced Study (IAS), Princeton, NJ, 1999.
Vol. 1: xxii+723 pp.; Vol. 2: pp. i-xxiv and 727–1501.
[Ef] Efetov K. Supersymmetry in Disorder and Chaos. Cambridge U.
Press, Cambridge, 1997, 453 pp.
[E1] Eisenbud D., Harris J., The geometry of schemes. Graduate Texts
in Mathematics, 197. Springer-Verlag, New York, 2000. x+294 pp.
[E2] Eisenbud D., Commutative algebra. With a view toward algebraic
geometry. Graduate Texts in Mathematics, 150. Springer-Verlag,
New York, 1995. xvi+785 pp.
[E3] Eisenbud, D., Harris, J., Schemes. The language of modern alge-
braic geometry. The Wadsworth & Brooks/Cole Mathematics Se-
2.14 Resolutions and smoothness 173

ries. Wadsworth & Brooks/Cole Advanced Books & Software, Pa-

cific Grove, CA, 1992. xii+157 pp.
[EM] Encyclopedia of mathematics, Kluwer, 1987–1992
[FFG] Fomenko A., Fuchs D., Gutenmakher V., Homotopic topology,
Academiai Kiadó, 1987
[Gab] Gabriel P., Categories abeliénnes, Bull. Soc. Math. France, 1962
[GM] Gelfand S., Manin Yu. Methods of homological algebra. Trans-
lated from the 1988 Russian original. Springer-Verlag, Berlin, 1996.
xviii+372 pp.
[GK] Ginzburg V., Kapranov M., Koszul duality for operads. Duke Math.
J. 1994. V. 76, No 1. P. 203–272.
[God] Godeman R. Topologie algébrique et théorie des faisceaux Hermann,
Paris, 1998, 283 pp.
[GH] Griffiths P., Harris J., Principles of algebraic geometry, A. Wiley,
NY e.a., 1978
[EGA] Grothendieck, A.Éléments de géométrie algébrique. IV. (French) I.
Le langage des schémas. Inst. Hautes Études Sci. Publ. Math. No.
4, 1960 228 pp.;
II. Étude globale élémentaire de quelques classes de morphismes.
Inst. Hautes Études Sci. Publ. Math. No. 8, 1961 222 pp.;
Étude locale des schémas et des morphismes de schémas. III. Inst.
Hautes Études Sci. Publ. Math. No. 28 1966 255 pp.;
IV. Étude locale des schémas et des morphismes de schémas IV.
Inst. Hautes Études Sci. Publ. Math. No. 32, 1967 361 pp.
All this and many other works by Alexander Grothendieck — one of the
greatest mathematicians of the century — is available at:
http://www.grothendieckcircle.org/.
[Hs] Halmos P., Naive set theory, Van Nostrand, 1960
[J] Johnstone, P. T. Topos theory. London Mathematical Society Mono-
graphs, Vol. 10. Academic Press [Harcourt Brace Jovanovich, Pub-
lishers], London-New York, 1977. xxiii+367 pp.
[H] Hartshorn R., Algebraic geometry, Springer, NY e.a. 1977
[KaS] Kashiwara M., Schapira P., Sheaves on manifolds. With a chap-
ter in French by Christian Houzel. Corrected reprint of the 1990
original. Grundlehren der Mathematischen Wissenschaften [Funda-
mental Principles of Mathematical Sciences], 292. Springer, Berlin,
1994. x+512 pp.
[Kas] Kassel C., Quantum groups. Graduate tests in Math. V. 155.
Springer, N.-Y., 1995.
[Ke] http://www.math.jussieu.fr/ keller/publ/
[K] Kelley, John L. General topology. Reprint of the 1955 edition [Van
Nostrand, Toronto, Ont.]. Graduate Texts in Mathematics, No. 27.
Springer-Verlag, New York-Berlin, 1975. xiv+298 pp.
174 Ch. 2. Sheaves, schemes, and projective spaces

[Kz] Kunz E., Introduction to commutative algebra and algebraic ge-

ometry. Translated from the German by Michael Ackerman. With
a preface by David Mumford. Birkhäuser Boston, Inc., Boston, MA,
1985. xi+238 pp.
[Ku] Kuroda S. A generalization of the Shestakov-Umirbaev inequality.
J. Math. Soc. Japan Volume 60, Number 2 (2008), 495–510.
[Lang] Lang S., Algebra. Revised third edition. Graduate Texts in Mathe-
matics, 211. Springer-Verlag, New York, 2002. xvi+914 pp.
[L0] Leites D., Spectra of graded-commutative rings, Uspehi Matem.
Nauk, 29, no. 3, 1974, 157–158 (in Russian; for details, see [SoS])
[M] Macdonald, I. G. Algebraic geometry. Introduction to schemes.
W. A. Benjamin, Inc., New York-Amsterdam, 1968, vii+113 pp.
[McL] Mac Lane S. Categories for the working mathematician. Second edi-
tion. Graduate Texts in Mathematics, 5. Springer, New York, 1998.
xii+314 pp.
[Ma1] Manin Yu.I., Lectures on algebraic geometry (1966–68), Moscow
Univ. Press, Moscow, 1968 (in Russian)
[Ma2] Manin Yu.I., Lectures on algebraic geometry. Part 1. Moscow Univ.
Press, Moscow, 1970 (in Russian)
[Ma3] Manin Yu.I., Lectures on K-functor in algebraic geometry. Russian
Math. Surveys, v. 24, 1969, no. 5, 1–89
[Ma4] Manin Yu.I., A course in mathematical logic, Translated from the
Russian by Neal Koblitz. Graduate Texts in Mathematics, Vol. 53.
Springer-Verlag, New York-Berlin, 1977., xiii+286 pp.
[MaD] Manin Yu., New dimensions in geometry. (Russian) Uspekhi Mat.
Nauk 39 (1984), no. 6(240), 47–73. English translation: Russian
Math. Surveys 39 (1984), no. 6, 51–83
This paper is essentially the same as one which has appeared in
English in: Workshop Bonn 1984 (Bonn, 1984), Lecture Notes in
Math., 1111, Springer, Berlin, 1985, 59–101, where it is followed by
some interesting remarks by M. F. Atiyah: Commentary on the arti-
cle of Yu. I. Manin: “New dimensions in geometry” [Workshop Bonn
1984 (Bonn, 1984), 59–101, Springer, Berlin, 1985; MR 87j:14030].
Workshop Bonn 1984 (Bonn, 1984), 103–109
[MaT] Manin Yu. Topics in noncommutative geometry. Princeton Univ.
Press, 1991.
[MaG] Manin Yu. Gauge field theory and complex geometry. Translated
from the 1984 Russian original by N. Koblitz and J. R. King. Sec-
ond edition. With an appendix by Sergei Merkulov. Grundlehren der
Mathematischen Wissenschaften [Fundamental Principles of Math-
ematical Sciences], 289. Springer-Verlag, Berlin, 1997. xii+346 pp.
[MaQ] Manin Yu. Quantum groups and non-commutative geometry, CRM,
Montreal, 1988.
[Mdim] Manin Yu. The notion of dimension in geometry and algebra, Bull.
Amer. Math. Soc. 43 (2006), 139–161; arXiv: math.AG/0506012
2.14 Resolutions and smoothness 175

[MV] Molotkov V., Glutoses: a Generalization of Topos Theory,

arXiv:math/9912060
[M1] Mumford D., Lectures on curves on an algebraic surface. With a sec-
tion by G. M. Bergman. Annals of Mathematics Studies, No. 59
Princeton University Press, Princeton, N.J., 1966, xi+200 pp.
[M2] Mumford D., Picard groups of moduli problems. Arithmetical Alge-
braic Geometry (Proc. Conf. Purdue Univ., 1963), 1965, Harper &
Row, NY, 33–81
[M3] Mumford D., The red book of varieties and schemes. Lecture Notes
in Mathematics, 1358. Springer-Verlag, Berlin, 1988. vi+309 pp.
[OV] Onishchik A., Vinberg E., Lie groups and algebraic groups. Springer-
Verlag, Berlin, 1990. xx+328 pp.
[Pr] Prasolov, V. Polynomials. Translated from the 2001 Russian second
edition by Dimitry Leites. Algorithms and Computation in Mathe-
matics, 11. Springer, Berlin, 2004. xiv+301 pp.
[Q] Quillen D., Projective modules over polynomial rings, Inventiones
Mathematicae, 36 (1976) 167–171
[Reid] Reid M., Undergraduate commutative algebra. London Mathemat-
ical Society Student Texts, 29. Cambridge University Press, Cam-
bridge, 1995. xiv+153 pp.
[RF] Rokhlin V.A., Fuchs D.B. Introductory course of topology. Geomet-
ric chapters. Nauka, Moscow, 1977 (Russian) = English translation
by Springer, 1983
[R1] Rosenberg A., Noncommutative algebraic geometry and representa-
tions of quantized algebras. Mathematics and its Applications, 330.
Kluwer, Dordrecht, 1995. xii+315 pp.
[R2] Rosenberg A., Almost quotient categories, sheaves and localization.
In: [SoS]-25
[Ru] Rudakov A., Marked trees and generating functions with odd vari-
ables. Normat 47 (1999), no. 2, 66–73, 95.
[S1] Serre J.-P. Algèbre locale, multiplicitès: Cours au Collège de France,
1957−1958. Lecture Notes in Math. 11. Springer, Berlin et. al. Corr.
3rd printing, 1997, X, 160 pp.
[S2] Serre J.-P. Algebraic groups and class fields Springer, Berlin et al.,
Corr. 2nd ed. 1997
[SGA4] Théorie des topos et cohomologie étale des schémas (SGA4). Un
séminaire dirigé par M. Artin, A. Grothedieck, J. L. Verdier. Lecture
Notes in Mathematics, 269. Springer-Verlag, 1972.
[SGA4.5] Deligne, P. et al. Séminaire de Géométrie Algébrique du Bois-Marie
1
SGA 4 . Cohomologie étale. Lecture Notes in Mathematics 569.
2
Springer-Verlag, 1977.
[Sh0] Shafarevich I. Basic algebraic geometry. 1. Varieties in projective
space. Second edition. Translated from the 1988 Russian edition and
with notes by Miles Reid. Springer-Verlag, Berlin, 1994, xx+303
176 Ch. 2. Sheaves, schemes, and projective spaces

pp.; Basic algebraic geometry. 2. Schemes and complex manifolds.

Second edition. Translated from the 1988 Russian edition by Miles
Reid. Springer-Verlag, Berlin, 1994. xiv+269 pp.
[Sh1] Shafarevich, I. R. Basic notions of algebra. Translated from the Rus-
sian by M. Reid. Reprint of the 1990 translation [Algebra. I, En-
cyclopaedia Math. Sci., 11, Springer, Berlin, 1990; MR 90k:00010].
Springer-Verlag, Berlin, 1997. iv+258 pp
[Sh2] Shafarevich, I. R. Number theory. I. Fundamental problems, ideas
and theories. A translation of Number theory. 1 (Russian), Akad.
Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow,
1990 [MR 91j:11001a]. Translation edited by A. N. Parshin and I. R.
Shafarevich. Encyclopaedia of Mathematical Sciences, 49. Springer-
Verlag, Berlin, 1995. iv+303 pp.
[SoS] Leites D. (ed.), Seminar on supermanifolds, no. 1–34, 2100 pp. Re-
ports of Dept. of Math. of Stockholm Univ., 1986–1990
[StE] Steenrod N., Eilenberg S. Foundations of Algebraic Topology,
Princeton University Press, 1952 (second printing, 1957). 15+328
pp.
[Su] Suslin A., Projective modules over polynomial rings are free, Soviet
Math. Dokl. 17 (4): (1976) 1160–64
id., Algebraic K-theory of fields. Proceedings of the International
Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif., 1986), 222–
244, Amer. Math. Soc., Providence, RI, 1987
[VSu] Vassershtein L., Suslin A., Serre’s problem on projective mod-
ules over polynomial rings and algebraic K-theory. Math. USSR
Izvestiya, ser. mathem. 1976. v. 40, No. 5. . 993–1054.
[vdW] van der Waerden, B. L. Algebra. Vol. I. Based in part on lectures
by E. Artin and E. Noether. Translated from the seventh German
edition by Fred Blum and John R. Schulenberger. Springer-Verlag,
New York, 1991. xiv+265 pp; Vol. II. Based in part on lectures by E.
Artin and E. Noether. Translated from the fifth German edition by
John R. Schulenberger. Springer-Verlag, New York, 1991. xii+284
pp.
[We] Weil, A., Théorie des points proches sur les variétés différentiables.
(French) Géométrie différentielle. Colloques Internationaux du Cen-
tre National de la Recherche Scientifique, Strasbourg, 1953, pp. 111–
117. Centre National de la Recherche Scientifique, Paris, 1953.
[ZS] Zariski O., Samuel P. Commutative algebra. v.1, Springer, Berlin
e.a., 1979
Index

(a : b), the quotient of two ideals, 48 Xred , see scheme, reduced, 43

(CS )◦ , 91 Z 1 ((Ui )i∈I , O×
X ), 132
A-Algs, the category of A-algebras, 89 Ann f , annihilator of an element, 45
A-Mods, the category of (left) BunM , 91
A-modules over an algebra A, 89 CovectB/A , the module of (relative)
A[S −1 ], see localization with respect to differentials, 66
a multiplicative system, 37 Ga , group, additive, 80
AS , see localization with respect to Gm , group, multiplicative, 80
a multiplicative system, 37 1
ΩB/A , the module of (relative)
Af , localizations with respect to Sf , 37 differentials, odd, 67
Ap , localizations with respect to Sp , 37 Pic X, the Picard group, 132
B-point of A, see also point, 12 Prime A if X = Spec A, the set of prime
B 1 ((Ui )i∈I , O×
X ), 132 ideals associated with X (or A),
D(f ), see set, big open, 22 46
H 1 ((Ui )i∈I , O×
X ), 132 Prime X, where X = Spec A, the set
K-algebra, 7 of prime ideals associated with X
K-algebra homomorphism, 8 (or A), 46
K(X), Grothendieck group, 162 Rings, the category of (commutative)
L-point of the system of equations X, 8 rings (with unit) and their
M -sequence, 159 homomorphisms, 89
N (x), the norm of the field k(x), 74 RingsR = R-Algs, 91
P X , the point functor, 93 Sets, the category of sets and their
P X (Y ) := HomC (X, Y ), 93 maps, 88
PX , the point functor, 93 Spec A, 13
PX (Y ) := HomC (Y, X), 79 Spm A, the set of maximal ideals, 41
S −1 A, see localization with respect to Top, the category of topological spaces
a multiplicative system, 37 and their continuous maps, 88
V (E), 19 VebunM , 91
V (F ), the variety defined by the ideal OX (n), 133
(F ), 8 χ(X), characteristic, of projective
X(L), the set of L-points of the system scheme, 118
of equations X, 8 χ(F), Euler characteristic of the sheaf
Xf := Spec Af , 57 F, 156
178 Index

µn , group of nth roots of unity, 82 Complex

ν(pa ), 74 Čech, 137
Ann(b), annihilator of an ideal, 48 acyclic, 139
Diff(M, N ), see differential operators, Component
68 embedded, 47
Diff k (M, N ), 68 irreducible of a given Noetherian
supp Y , the support of the scheme topological space, 24
Y = Spec A/a, 42 isolated, 47
supp, see support of a sheaf, 131 Convention, 16
ht x, height of the point x, 22
X(Y ) := PX (Y ), 94 DCC, the descending chain condition,
κr , 165 18
hM (n), Hilbert polynomial, 118 Decomposition
n(pa ), 74 primary, incompressible, 45
pa (X) genus, arithmetic, of projective Degree
scheme X, 118 of a homomorphism, 117
V
rU , restriction map, 98 Dimension
Čech cocycle, 132 of projective scheme, 118

Algebra Element
flat, 59 integer over A, 36
integer over A, 36 Embedding
Annihilator, 45, 48 closed, of a subscheme, 42
Arrow, 87 Epimorphism
Artinian ring, 18 minimal, 169
Equivalence
Bezout’s theorem, 123, 125 birational, 111
Bundle
conormal, 69 Family
normal, 69 normal (of vector spaces), 63
vector, 56 of vector spaces, 56
stably free, 71 Filter
on a set, 29
Cartesian square, 52 Formula
Cartier’s Theorem, 83 Lefschetz, 76
Category, 87 Function
big, 88 flat, 61
small, 88 Functor, 91
Center representable, 93
of a geometric point, 14 corepresentable, 93
Chain of points, 22 point, 93
Characteristic sheafification, 164
of projective scheme, 118
Chevalley’s theorem, 36 Galois group, 81
Co-normal Genus
to the diagonal, 67 arithmetic, of projective scheme, 118
Cocycle Germ
Čech, 132 of a section, 101
Codimension, 64 of neighborhoods, 39
Index 179

Grading locally free, 57

standard of K[T ], 112 co-normal, 65
Grothendieck group, 162 conormal, 63
Group defining a family of vector spaces, 56
of Čech cohomology, 138 graded, 117
additive, 80 length of, 18
Galois, 81 simple, 18
general linear, 80 Modules
linear algebraic, 83 T N -isomorphic, 156
multiplicative, 80 equivalent
Picard, 132 freely, 168
Group scheme projectively , 168
affine, 79 Morphism, 87

Harnak’s theorem, 10 Nakayama’s lemma, 61

Height of the point, 22 Nilradical, 17
Hilbert polynomial, 118 Noether’s normalization Theorem, 33
Hilbert’s Nullstellensatz, 51 Noetherian topological space, 24
Hilbert’s Nullstellensatz (theorem on
zeros), 15 Object
Hilbert’s syzygies theorem, 171 final, 84
Homotopy Open =open set, 98
chain, 139 Operator
differential, 68
Ideal differential, symbol of, 68
primary, 44 Order
homogeneous, 112 on the set of closed subschemes, 43
prime, 13
radical, 20 Partition
Intersection of unity, 104
complete, 63 Partition of unity, 28
complete, geometric, 124 Picard group, 132
Isomorphism Point
up to a finite number of homogeneous generic, 24
components, 160 Polynomial
Hilbert, 118
Lefschetz formula, 76 Presheaf, 98
Legendre’s theorem, 9 of modules, 125
Lemma Problem
Nakayama, 61 open, 124
Zorn, 13 Riemann-Roch, 161
Length of a module, 18 Product
Limit fiber, 52
direct, 101
inductive, 101 Quotient
Localization with respect to a multi- of ideals, 48
plicative system, 37
Radical r(I) of the ideal I, 20
Module Resolution
180 Index

minimal, 169 Square

Riemann-Roch problem, 161, 162 Cartesian, 52
Ring Stalk, 101
of constants, 7 Structure, 88
Artinian, 18 Sub-category
Boolean, 11 full, 88
coordinate of the variety, 8 Subscheme
Noetherian, 12 closed, 42
of quotients, 37 closed, embedded regularly, 63
Veronese, 116 locally regularly embedded at y ∈ Y ,
64
Scheme, 108 Support
(ir)reducible, 44 of a sheaf, 131
Noetherian, 44 of the scheme, 42
primary, 44 System
affine, 30, 108 of equations, 7
reduced, 43 direct, 101
smooth, 163 direct or inductive, 89
Sequence inductive, 101
regular, 148, 149 inverse or projective, 89
Serre’s problem, 71 multiplicative, 28, 37
Serre’s theorem, 150 Systems
Seshadri’s theorem, 71 of equations, equivalent, 8
Set
big open, 22 Theorem
cofinal, 134 Bezout, 123, 125
constructible, 36 Cartier, 83
inductive, 13, 24 Chevalley, 36
locally closed, 36 Harnak, 10
Sheaf, 99 Hilbert’s on syzygies, 164, 171
invertible, 131 Legendre, 9
coherent, 128 Noether, normalization, 33
of finite type, 128 on elementary divisors, 73
quasi-coherent, 127 Serre, 150
structure, 107 Topology
very ample, 158 Zariski, 19
Solution Transformation
of the system of algebraic equations, monoidal, with center in an ideal, 115
7
of a system of equations, 8 Ultrafilter, 29
Space Vector
connected, 26 normal, 63
cotangent, 65 Vector bundle, 56
irreducible, 23 over a scheme, 58
ringed, 107 Veronese ring, 116
tangent, 65
topological Zariski topology, 19
quasi-compact, 28 Zero divisor
Spectrum essential, 158
maximal, 41 Zorn’s lemma, 13
projective, 113