Commutative Algebra

Sommersemester 2017

Mohamed Barakat

D EPARTMENT M ATHEMATIK , U NIVERSITÄT S IEGEN

[email protected]

Stand: January 17, 2018

Der Nachdruck dieses Textes, auch von einzelnen Teilen daraus, ist nicht gestattet.
Vorwort

Dies ist die geTEXte Version meiner Vorlesungsnotizen, die ich fortlaufend aktualisieren
werde. Der Stand der Notizen ist mindestens um eine Woche verzögert. Daher gilt: Kommt
zur Vorlesung und macht Eure eigenen Notizen. Die sind sowieso besser als jedes Skript. Die
Form eines Skriptes erreichen diese Notizen vermutlich erst gegen Ende der Vorlesung,
dies kann ich aber nicht garantieren. Die aktuelle Version ist unter der folgenden Adresse
zu finden:

http://www.mathematik.uni-kl.de/~barakat/Lehre/SS17/CA/Skript/CA.pdf

Das Skript der Algebra wird vorausgesetzt:

http://www.mathematik.uni-kl.de/~barakat/Lehre/WS16/Algebra/Skript/
Algebra.pdf

Als Vorlage für das Kapitel über Gröbnerbasen benutz(t)e ich das online-verfügbare Buch
von Prof. Wolfram Decker und Prof. Frank-Olaf Schreyer

http://www.mathematik.uni-kl.de/~decker/Lehre/WS12/
CommutativeAlgebra/skript/BookDeckerSchreyer.pdf.

Als Vorlage für das Kapitel über Lokalisierung benutz(t)e ich das online-verfügbare
Skript von Prof. Wolfram Decker

http://www.mathematik.uni-kl.de/fileadmin/AGs/agag/decker/WS1516/
CommAlg/CommutativeAlgebra.pdf.

Für Korrektur- und Verbesserungsvorschläge bin ich stets dankbar

[email protected]

iii
iv
Contents

1 Gröbner bases 1
1 Monomials and monomial ideals . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Monomial orderings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
3 Division with remainder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 Gröbner bases and Buchberger’s algorithm . . . . . . . . . . . . . . . . . . . . 7
5 First applications: Submodule membership and syzygies . . . . . . . . . . . . 16
6 Iterated syzygies and Hilbert’s Syzygy Theorem . . . . . . . . . . . . . . . . . 17
7 Elimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Constructive A BELian categories 23

1 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2 The category of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1 Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.2 Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3 Subobjects and factor objects . . . . . . . . . . . . . . . . . . . . . . . . 28
2.4 Kernels and cokernels, images and co-images . . . . . . . . . . . . . . 29
2.5 Direct sums and direct products . . . . . . . . . . . . . . . . . . . . . . 32
2.6 Pull-backs and push-outs as kernels and cokernels . . . . . . . . . . . 33
3 More categorical notions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4 Additive and A BELian categories . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5 Computable rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6 A constructively A BELian category of matrices . . . . . . . . . . . . . . . . . . 42
7 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
8 Natural transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
9 Exact sequences, homology, and cohomology . . . . . . . . . . . . . . . . . . . 52
10 Free resolutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3 Localization and tensor products 57

1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2 Nilradical and Jacobson Radical . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3 The Cayley-Hamilton Theorem and Nakayama’s Lemma . . . . . . . . . . . . 60
4 Tensor products of modules over commutative rings . . . . . . . . . . . . . . . 62
5 Localization of rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6 Localization of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
7 Local properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8 Chain conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
9 Artinian modules and rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4 Primary decomposition 75
1 Definition and Existence in N OETHERian rings . . . . . . . . . . . . . . . . . . 75
2 Uniqueness results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

v
vi CONTENTS

5 Integral ring extensions 81

1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2 Lying over and going up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3 Going down . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

6 K RULL Dimension, N OETHER Normalization, H ILBERT’s Nullstellensatz and K RULL’s

Principle Ideal Theorem 89
1 Definition of K RULL dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
2 Noether Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
3 Properties of K RULL dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4 H ILBERT’s Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5 K RULL’s Principal Ideal Theorem and regular local rings . . . . . . . . . . . . 96

7 Valuation rings and Dedekind domains 101

1 Valuation rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
2 D EDEKIND domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Chapter 1

Gröbner bases

This chapter is an excerpt from unpublished lecture notes of Wolfram Decker and Frank-
Olaf Schreyer.

1 Monomials and monomial ideals

In this chapter k will denote a field and R := k[x1 , . . . , xn ] the polynomial ring over k with
indeterminates x1 , . . . , xn . A monomial in x1 , . . . , xn is a product xα = xα1 1 · · · xαnn , where
α = (α1 , . . . , αn ) ∈ Nn = Nn0 .
Definition 1.1. A monomial ideal in R is an ideal generated by monomials.
Example 1.2. The ideal I = hx3 , xy 2 i
k[x, y] is a monomial ideal. The monomials
1, x, x2 , xy, x2 y and all powers of y are not contained in I.
Lemma 1.3 (Gordan’s Lemma). Any nonempty set X of monomials in R has only finitely
many minimal elements in the partial order given by divisibility1 . Hence, any monomial ideal in R
has finitely many monomial generators.

Proof. Exercise.
Hint: Use an induction on the number of variables.

In fact, by removing those generators which are divisible by others, one obtains a uniquely
determined set of monomial generators for I. Its elements are called minimal generators
of I.
The ideal membership problem provides our first example of how monomial ideals are
easier to handle than arbitrary ideals: If I  R is a monomial ideal, given by monomial
generators m1 , . . . , mr , a term is contained in I iff it is divisible by at least one of the mi ; an
arbitrary polynomial g ∈ R is contained in I iff all its terms are contained in I.
Most of the terminology used when working with polynomials extends to elements of
free modules overL R. In what follows let F be a free R-module with a fixed basis {e1 , . . . , es },
in particular F = si=1 Rei .
Definition 1.4. A monomial in F involving the basis element ei is a monomial in the
direct summand Rei ∼= R. A term in F is a monomial in F multiplied by a coefficient c ∈ k.
Every nonzero element f ∈ F can be uniquely expressed as the sum of finitely many
nonzero terms involving distinct monomials. These terms resp. monomials are called the
terms resp. monomials of f .
1 β
x | xα iff α − β ∈ Nn .

1
2 CHAPTER 1. GRÖBNER BASES




Example 1.5. Let F = k[x, y]3 . If e1 = 1 0 0 , e2 = 0 1 0 , e3 = 0 0 1 are the
canonical basis vectors, then

f := x2 y + x2 1 x = x2 y · e1 + x2 · e1 + 1 · e2 + x · e3 ∈ F.

A submodule of F is a monomial submodule if it is generated by monomials. It easily

follows from Gordan’s lemma that every such submodule is generated by finitely many
monomials. As in the ideal case, there is a unique finite set of minimal generators. More-
over, membership in monomial submodules can be decided as for monomial ideals.

2 Monomial orderings
Definition 1.6. A monomial order on R is a total order2 > on the set of monomials in R
such that
xα > xβ =⇒ xγ xα > xγ xβ for all α, β, γ ∈ Nn .
Definition 1.7 (Lexicographic order). Set

xα >lex xβ ⇐⇒ the first nonzero entry of α − β is positive.

Remark 1.8. Note that we have defined >lex such that the variables are ordered accord-
ing to their appearance when writing R. For instance, in k[x, y, z],

x3 >lex xyz >lex x >lex y 25 >lex y >lex z.

Given a monomial order > on R, we will abuse notation as follows: If c, d ∈ k × are

scalars and xα , xβ are monomials in R such that xα > xβ (or xα ≥ xβ ), we will write cxα >
dxβ (or cxα ≥ dxβ ). In the same spirit, we will occasionally speak of the maximum of a
finite number of nonzero terms (which is determined up to a scalar).
Proposition 1.9. Let > be a monomial order on R. Then the following are equivalent:

(1) > is Artinian, i.e., each nonempty set of monomials has a smallest element with respect to >.

(2) > is global, i.e., 1 < xi for all i = 1, . . . , n.

(3) > refines the divisibility partial order, i.e., xβ | xα =⇒ xβ < xα .

Proof. The only nontrivial part of the proof is to show that condition (3) implies condition
(1). If condition (3) holds, and X is a nonempty set of monomials, the monomial ideal
I = hXi  R generated by X is, in fact, generated by a finite subset Y of X due to Gordan’s
lemma. Hence, every monomial in X is divisible by a monomial in Y , and the smallest
element of Y is the smallest element of X.
We use the word global to distinguish the monomial orders considered in this chap-
ter from so-called local orders, used to compute in local rings. The lexicographic order is
global.
Buchberger’s algorithm requires the choice of a global monomial order. The perfor-
mance of the algorithm and the resulting Gröbner basis depend in a crucial way on this
choice. For most applications, in principle, any Gröbner basis and, thus, any order will do.
With regard to efficiency, however, the global monomial order which we will now define
appears to be best possible (see the discussion following Remark 1.15 below and [BS87] for
some evidence):
2
We also write xβ < xα for xα > xβ .
2. MONOMIAL ORDERINGS 3

Definition 1.10. We define the degree reverse lexicographic order on R by setting

xα >drlex xβ ⇐⇒ deg xα > deg xβ , or (deg xα = deg xβ and the

last nonzero entry of α − β ∈ Zn is negative).

This order is extended to free R-modules as in Remark 1.17 (we suggest to give priority to
the monomials in R).
Note that as in the case of >lex , we have defined >drlex such that the variables are ordered
according to their appearance when writing R. In contrast to >lex , however, >drlex refines
the partial order by total degree:

deg xα > deg xβ =⇒ xα >drlex xβ .

We will refer to this fact by saying that >drlex is degree-compatible.

Example 1.11. With respect to >lex and >drlex , the monomials of degree 2 in k[x, y, z] are
ordered as follows:
x2 >lex xy >lex xz >lex y 2 >lex yz >lex z 2
and
x2 >drlex xy >drlex y 2 >drlex xz >drlex yz >drlex z 2 .

For monomials of the same degree, the difference between >lex and >drlex is subtle but
crucial. The use made of these orders relies on their key properties (which, as we will see
in Exercise 1.1, characterize >lex and >drlex among all global monomial orders). End
Definition 1.12. Let > be a monomial order on R, and let f ∈ R be a nonzero polyno- lect. 1
mial. The leading term (or leading term) of f with respect to >, written 25.04

L> (f ) = L(f ),

is the largest term of f with repect to >. By convention, L> (0) = L(0) = 0. If L(f ) =
cxα , with c ∈ k, then c is called the leading coefficient of f and xα is called the leading
monomial of f .
Remark 1.13. Since monomial orders are compatible with multiplication,

L(f g) = L(f )L(g)

for all f, g ∈ R. Furthermore, if f , g, and f + g are nonzero, then

max{L(f ), L(g)} ≥ L(f + g).

The inequality is strict iff L(f ) and L(g) cancel each other in f + g.
This last equivalence shows that if L(h) is divisible by L(f ), and if we think of comput-
L(h)
ing h − L(f )
f as a single step of a division process, then the new dividend in such a step
will be zero, or its leading term will be smaller than that of the preceeding dividend h. This
does not imply, however, that the process terminates:
Example 1.14. In k[x], choose the terms of lowest degree as the leading terms. This
order is not global! Divide g = x by f = x − x2 using division steps as described above.
Then, the successive intermediate dividends are g = x, x2 , x3 , . . . .
Remark 1.15.
4 CHAPTER 1. GRÖBNER BASES

• The key property of >drlex is: >drlex is degree-compatible, and if f ∈ R is homoge-

neous, then
>drlex chooses the leading term of f in a subring k[x1 , . . . , x` ]
such that ` is as small as possible.
This property usually has the effect that, compared to other global monomial orders,
the monomial ideals Mi in Buchberger’s test have fewer minimal generators.
• The key property of >lex is that the following holds for all f ∈ R:
L(f ) ∈ k[x`+1 , . . . , xn ] for some ` =⇒ f ∈ k[x`+1 , . . . , xn ].
This makes >lex useful for eliminating variables, an application of Buchberger’s algo-
rithm which requires the computation of special Gröbner bases and, thus, the choice
of special monomial orders.
Exercise 1.1. Show that the key properties of >lex respectively >drlex characterize these
orders among all global monomial orders.
Hint. If > satisfies the key property of >drlex , we have, for instance, x22 > x1 x3 . Since > is
compatible with multiplication, also x22 x4 > x1 x3 x4 .
Our goal in this chapter is to develop the computational concepts not only for poly-
nomial rings, but also for free modules over polynomial rings. In extending division
with remainder to free modules, we again consider a free R-module F with a fixed basis
{e1 , . . . , es }.
Definition 1.16. A monomial order on F is a total order > on the set of monomials in F
such that for all monomials xα ei , xβ ej in F and any monomial xγ in R
xα ei > xβ ej =⇒ xγ xα ei > xγ xβ ej .
Convention. We require in addition that
xα ei > xβ ei ⇐⇒ xα ej > xβ ej for all i, j.
Then > induces a unique monomial order on R in the obvious way, and we say that > is
global if this induced order on R is global.
Remark 1.17. One way of getting a monomial order on F is to pick a monomial order
> on R, and extend it to F . For instance, setting
xα ei > xβ ej ⇐⇒ xα > xβ or (xα = xβ and i > j)
gives priority to the monomials in R (term-over-position), while setting
xα ei > xβ ej ⇐⇒ i > j or (i = j and xα > xβ )
gives priority to the components of F (position-over-term).

3 Division with remainder

We are now ready to extend the Euclidean division algorithm. In several variables, allow-
ing several divisors, the result of the division process may depend on some choices made
in carrying out the process. For instance, if h is some intermediate dividend, and f1 , . . . , fr
are the divisors, it may happen that L(h) is divisible by more than one of the L(fi )’s, and
any of these can be used to cancel L(h). Our first version of the extended division algo-
rithm avoids such ambiguities. For us, this determinate version will be particularly useful
in relating Buchberger’s algorithm to syzygies.
 3. DIVISION WITH REMAINDER 5

Theorem 1.18 (Division with Remainder, Determinate Version). Let > be a global mono-
mial order on R, and let f1 , . . . , fr ∈ R \ {0}. For every g ∈ R, there exists a uniquely determined
expression
g = g1 f1 + . . . + gr fr + h, with g1 , . . . , gr , h ∈ R,
such that
(DD1) no nonzero term of gi L(fi ) is divisible by L(fj ) for j < i;

(DD2) no nonzero term of h is divisible by L(fi ) for all i.

We call h the remainder of g on determinate division by f1 , . . . , fr .
Proof.
Uniqueness. Given any representation as in the assertion, conditions (DD1) and (DD2)
imply that the nonzero terms among the L(gi fi ) = L(gi )L(fi ) and L(h) involve different
monomials. Hence, these terms do not cancel each other on the right hand side of the
representation. If two such representations for g ∈ R are given, their difference is a repre-
sentation for the zero polynomial satisfying (DD1) and (DD2). According to what we just
said, the difference must be the trivial representation.
Existence. The determinate division algorithm finds the desired representation for g ∈
R as follows:
If f1 , . . . , fr are terms, first remove any multiple of f1 from g. Then cancel the remaining
multiples of f2 . Continue in this way until any multiple of any f` has been removed.
If f1 , . . . , fr are arbitrary, apply the above to g and L(f1 ), . . . , L(fr ). If
r
X
g= gi L(fi ) + h
i=1

is the resulting representation, then either g (1) := g − ri=1 gi fi − h is zero, and we are
P
done, or L(g) > L(g (1) ). By recursion, since > is Artinian, we may assume in the latter
(1)
case that g (1) has a representation g (1) = ri=1 gi fi + h(1) satisfying (DD1) and (DD2). Then
P
(1)
g = ri=1 (gi + gi )fi + (h + h(1) ) is a representation for g satisfying (DD1) and (DD2).
P

Conditions (DD1) and (DD2) are best understood by considering a partition of the
monomials in R as in the following example:
Example 1.19. Let f1 = x2 , f2 = xy + x ∈ k[x, y] with >lex . Then L(f1 ) = f1 = x2 and
L(f2 ) = xy. In the picture below, the monomials divisible by L(f1 ) correspond to the dots
in the region which is shaded in light grey:
− • • • •
− • • • •
− • • • •
− • • • •
(1, 1)
| •
(2, 0)
• •

Given g ∈ k[x, y] and a representation g = g1 f1 + g2 f2 + h, condition (DD1) means that that

the monomials of g2 L(f2 ) are located in the region shaded in dark grey. Condition (DD2),
in turn, requires that the monomials of h are located in the nonshaded region.
Dividing, for instance, g = x3 + x2 y 3 + xy 2 by f1 and f2 , we get:

g = (x + y 3 ) · L(f1 ) + y · L(f2 ) + 0,
 6 CHAPTER 1. GRÖBNER BASES

g (1) = g − (x + y 3 ) · f1 − y · f2 = −xy = 0 · L(f1 ) − 1 · L(f2 ) + 0,

g (2) = g (1) + f2 = x = 0 · L(f1 ) + 0 · L(f2 ) + x,

and
g (3) = g (2) − x = 0.
Thus, the desired representation is

g = (x + y 3 ) · f1 + (y − 1) · f2 + x.
End
lect. 2 It should be particularly clear from the picture in the example above that condition
02.05 (DD1) makes the order in which f1 , . . . , fr are listed play a crucial role in the determinate
division algorithm. We illustrate this by another example:
Example 1.20. Let f1 = x2 y − y 3 , f2 = x3 ∈ k[x, y] with >lex . Then L(f1 ) = x2 y. For
g = x3 y, the determinate division algorithm proceeds as follows:

x3 y = x · L(f1 ) + 0 · L(f2 ) + 0,

g (1) = g − x · f1 = xy 3 = 0 · L(f1 ) + 0 · L(f2 ) + xy 3 ,

and
g (2) = g (1) − xy 3 = g − x · f1 − xy 3 = 0.
Thus, the desired representation is

x3 y = x · (x2 y − y 3 ) + 0 · (x3 ) + xy 3 .

If we interchange f1 and f2 , determinate division yields the expression

x3 y = y · (x3 ) + 0 · (x2 y − y 3 ) + 0.

Remark 1.21 (Division with Remainder, Indeterminate Version). With notation as in

Theorem 1.18, the steps below describe a version of the division algorithm which is in-
determinate: the computed remainder depends on the choices made in the while loop
(termination follows once more from the fact that a global monomial order is Artinian).

(1) Set h := g and D := {f1 , . . . , fr }.

(2) while h 6= 0 and D(h) := {f ∈ D | L(h) is divisible by L(f )} =
6 ∅
• choose f ∈ D(h);
L(h)
• set h := h − L(f )
f.

(3) return(h).

With some extra bookkeeping as in Euclid’s division algorithm, the algorithm also returns
polynomials g1 , . . . , gr such that g = g1 f1 +. . .+gr fr +h. This representation of g satisfies the
conditions (ID1) and (ID2) below which are weaker than the conditions (DD1) and (DD2),
respectively:

(ID1) L(g) ≥ L(gi fi ) whenever both sides are nonzero.

(ID2) If h is nonzero, then L(h) is not divisible by any L(fi ).

4. GRÖBNER BASES AND BUCHBERGER’S ALGORITHM 7

Each such representation is called a standard expression for g with remainder h (in terms
of the fi , with respect to >).
In practical terms, it is often useful to give up uniqueness and allow choices to be made
since some choices are more efficient than others. In fact, there are various possible selection
strategies for the division process. It is not clear to us whether there is a “generally best”
strategy. Typically, the selection of the strategies depends on the particular application one
has in mind.
A version of the division algorithm which is even more indeterminate is discussed in
the exercise below.
Exercise 1.2. Show that we still get a division process which terminates if, at each stage,
we remove some term of the current dividend with the help of some L(fi ) by which it is
divisible, and if we stop as soon as this is no longer possible. Show that the resulting
representation g = g1 f1 + . . . + gr fr + h satisfies the conditions (ID1) and (DD2).
Remark 1.22 (Leading Terms in Standard Expressions). If g is a nonzero polynomial in
R and g = g1 f1 + . . . + gr fr + h is a standard expression, then L(g) is the maximum nonzero
term among the L(gi fi ) = L(gi )L(fi ) and L(h). Indeed, this follows from condition (ID1) in
conjunction with Remark 1.13. In particular, if the remainder h is zero, then L(g) is divisible
by one of L(f1 ), . . . , L(fr ). We, then, write

L(g) = max{L(g1 )L(f1 ), . . . , L(gr )L(fr )} ∈ hL(f1 ), . . . , L(fr )i.

4 Gröbner bases and Buchberger’s algorithm

In Example 1.20, with f1 = x2 y − y 3 and f2 = x3 , we computed the standard expressions

x3 y = x · f1 + 0 · f2 + xy 3

and
x3 y = y · f2 + 0 · f1 + 0,
which, in particular, have two different remainders. The problem with the first standard
expression is that x3 y and, thus, xy 3 are contained in the ideal hf1 , f2 i = hx2 y − y 3 , x3 i, but
xy 3 cannot be removed in the division process since it is not divisible by any of the leading
terms x2 y and x3 of the divisors. To decide ideal membership, we need to be able to cancel
any leading term of any element of I, using the leading terms of the divisors.
Based on this consideration, we make the following definition:
Definition 1.23. Let F be a free R-module with a fixed finite basis, > a global monomial
order on F , and A ⊂ F a subset.

(1) The leading submodule (or initial submodule) of A is the monomial submodule

L(A) := L> (A) := L> (f ) f ∈ A ≤ F.

That is, L(A) is generated by the leading terms of the elements of A. In the special
case where F = R, we refer to L(A) as the leading ideal (or initial ideal) of A.

(2) Let I ≤ F be a submodule. A finite subset G = {f1 , . . . , fr } ⊂ I is a Gröbner basis for

I if
L> (I) = L> (G).
That is, the leading submodule of I is generated by the leading terms of the elements
of G.
 8 CHAPTER 1. GRÖBNER BASES

For simplicity, we will say that a finite subset G of F is a Gröbner basis if it is a Gröbner
basis for the submodule it generates.
The terminology in the definition above is somewhat inaccurate in that we should have
written leading module with respect to > and Gröbner basis with respect to >. Indeed,
leading modules depend on the choice of the monomial order. Furthermore, if G is a Gröb-
ner basis with respect to >, and >0 is another monomial order, then G may fail to be a
Gröbner basis with respect to >0 .
Notation. For the rest of this section, > will be a fixed global monomial order on a free
R-module F with a fixed finite basis.
In contrast to the polynomials f1 , f2 in Example 1.20, the elements of a Gröbner basis
behave well under division with remainder and can, thus, be used to decide ideal and
submodule membership. In particular, we will now see that a Gröbner basis of I generates
I, although this was not part of the definition.
Proposition 1.24. Let P{f
r
1 , . . . , fr } ⊂ F \ {0} be a Gröbner basis for the submodule I :=
hf1 , . . . , fr i ≤ F . If g = i=1 gi fi + h is a standard expression for an element g ∈ F , then g ∈ I
iff the remainder h is zero. In particular, f1 , . . . , fr generate I.
Proof. If h is zero, then clearly g ∈ I. Conversely, if g ∈ I, then h ∈ I, which implies that
L(h) ∈ L(I) = hL(f1 ), . . . , L(fr )i. So L(h) and, thus, h are zero by condition (ID2) on the
remainder of a standard expression.
Corollary 1.25 (Hilbert’s Basis Theorem). Every submodule I ≤ F has a Gröbner basis and
is thus finitely generated. In particular, R is Noetherian.
Gordan’s Proof. As remarked earlier, Gordan’s lemma yields that each monomial submod-
ule of F is generated by finitely many monomials. In particular, there are finitely many
elements f1 , . . . , fr ∈ I such that L(I) = hL(f1 ), . . . , L(fr )i. That is, f1 , . . . , fr form a Gröb-
ner basis for I. The statement now follows from Proposition 1.24.
Remark 1.26. Gordan’s proof has, as every other proof of Hilbert’s basis theorem, two
ingredients, namely induction on the number of variables (here used to verify Gordan’s
Lemma 1.3) and division with remainder. The advantage of Gordan’s proof is that it sepa-
rates these ingredients.
End
lect. 3 Macaulay [1927] used the idea of obtaining information on an ideal from information on
04.05 its leading ideal to classify Hilbert functions. On his way, he proved the following crucial
result:
Theorem and Definition 1.1 (Macaulay). If I ≤ F is a submodule, the monomials not con-
tained in L> (I) represent a k-vector space basis for F/I. We refer to these monomials as standard
monomials (for I, with respect to >).
Proof. Let
B := {m + I | m ∈ F a standard monomial} ⊂ F/I.
To show that the elements of B are k-linearly independent, consider a k-linear combination
g of standard monomials such that the residue class g + I is zero. Then g ∈ I, so that
L(g) ∈ L(I). Since L(g) is a scalar times a standard monomial, this implies 0 = L(g) = g
by the very definition of the standard monomials.
To show that the elements of B generate F/I as a k-vector space, consider any element
g ∈ F . Choose elements f1 , . . . , fr ∈ F \ {0} which form a Gröbner basis for I, and let g =
P r
i=1 gi fi + h be a standard expression satisfying condition (DD2) of determinate division
with remainder. Then no term of h is in hL(f1 ), . . . , L(fr )i = L(I). Hence, the residue class
g + I = h + I is a k-linear combination of the elements of B, as desired.
4. GRÖBNER BASES AND BUCHBERGER’S ALGORITHM 9

Pr of Macaulay’s theorem, given g ∈ F , the remainder h in

Remark 1.27. In the situation
a standard expression g = i=1 gi fi + h satisfying (DD2) is uniquely determined by g, I,
and > (and does not depend on the choice of Gröbner basis). It represents the residue class
g + I ∈ F/I in terms of the standard monomials. We write NF(g, I) = h and call NF(g, I)
the canonical representative of g + I ∈ F/I (or the normal form of g mod I), with respect
to >.
If a Gröbner basis for an ideal I of R is given, we may use normal forms to perform the
sum and product operations in R/I. This is Buchberger’s original application of Gröbner
bases:
Exercise 1.3. Let I  R be an ideal. For f, g ∈ R show that
NF(f + g, I) = NF(f, I) + NF(g, I), and
NF(f · g, I) = NF(NF(f, I) · NF(g, I), I).
Following these first indications of the usefulness of Gröbner bases, our next objective
is to treat their computation. Suppose we are given a submodule I = hf1 , . . . , fr i ≤ F , with
a finite system of generators f1 , . . . , fr . Then, either the L(fi )’s do generate L(I), or there is
an element 0 6= fr+1 ∈ I such that
hL(f1 ), . . . , L(fr )i < hL(f1 ), . . . , L(fr ), L(fr+1 )i  I.
If we continue the process, by the ascending chain condition for monomial submodules of
F , it cannot go on forever. That is, after finitely many steps, we will arrive at a Gröbner
basis f1 , . . . , fr , fr+1 , . . . , fr0 for I.
Of course, the process just described is not yet constructive: We are missing a test which
allows us to decide whether the L(fi )’s do generate L(I), and if not, yields an explicit
element fr+1 ∈ I as above. A natural approach to detect such an element is to form R-linear
combinations of f1 , . . . , fr , and divide them by f1 , . . . , fr . Then any nonzero remainder will
have the desired properties.
In the simplest possible case, we face combinations gi fi + gj fj involving just two of the
generators. To increase our chances of getting a nonzero remainder in this case, we choose
gi and gj such that L(gi fi ) = L(gi )L(fi ) and L(gj fj ) = L(gj )L(fj ) cancel each other in
gi fi + gj fj :
Definition 1.28. If α = (α1 , . . . , αn ), β = (β1 , . . . , βn ) ∈ Nn are multi-indices, the greatest
common divisor of xα and xβ is
min(α1 ,β1 )
gcd(xα , xβ ) = x1 · · · xnmin(αn ,βn ) ,
and their least common multiple is
max(α1 ,β1 )
lcm(xα , xβ ) = x1 · · · xnmax(αn ,βn ) .
For terms in F = si=1 Rei , notions like multiple or divisible are defined in the obvious
L
way. For instance, the nonzero term cxα ei is divisible by the nonzero term dxβ ej , with
quotient dc xα−β ∈ R, if i = j and xα is divisible by xβ . Furthermore, the greatest common
divisor respectively the least common multiple of two nonzero terms involving the same
basis element ei are defined by the formulae
gcd(cxα ei , dxβ ei ) = gcd(xα , xβ )ei ∈ F
respectively
lcm(cxα ei , dxβ ei ) = lcm(xα , xβ ) ei ∈ F.
If cxα ei and dxβ ej involve distinct basis elements, we set
lcm(cxα ei , dxβ ej ) = 0.
10 CHAPTER 1. GRÖBNER BASES

Exercise 1.4. Let I = hm1 , . . . , mr i be a monomial ideal of R and let m be a monomial in

R. Show that the so-called quotient ideal

I : m := {r ∈ R | mr ∈ I}

is generated by the monomials

lcm(mi , m) mi
= , 1 ≤ i ≤ r.
m gcd(mi , m)

In particular, I : m is a monomial ideal as well.

Definition 1.29. Let f1 , . . . , fr ∈ F be nonzero polynomial vectors. For each pair of
indices i, j, the S-vector S(fi , fj ) ∈ F is defined by setting

S(fi , fj ) = mji fi − mij fj ∈ F,

where
lcm(L(fj ), L(fi ))
mji = ∈ R.
L(fi )

In case F = R, we call S(fi , fj ) an S-polynomial.

Now, a miracle happens: As we will show, dividing S-vectors by the f` ’s with remainder
is all we need to do. Since S(fi , fj ) = − S(fj , fi ) for all i, j, it actually suffices to consider the
S(fi , fj ) with j < i.
In fact, we can do even better: For i = 2, . . . , r, consider the monomial ideal

Mi = hL(f1 ), . . . , L(fi−1 )i : L(fi )  R,

and note that the terms mji for j < i form the set of generators for Mi mentioned in Exercise
1.4. As it turns out, there is no need to consider all the mji with j < i (that is, all the
corresponding S-vectors S(fi , fj )) in our test: For each i, and for each minimal monomial
generator xα of Mi , choose exactly one index j = j(i, α) < i such that mji involves xα , and
compute a standard expression for S(fi , fj ) in terms of the f` ’s with remainder hi,α (we
suppress the index j in our notation).
Theorem 1.30 (Buchberger’s Criterion). Let f1 , . . . , fr ∈ F be nonzero polynomial vectors.
With notation as above, f1 , . . . , fr form a Gröbner basis iff all remainders hi,α are zero.
Before proving the criterion, we illustrate it by an example, and formulate Buchberger’s
algorithm for computing Gröbner bases. For the example, recall that a k × k minor of a
matrix is the determinant of a k × k submatrix.
Example 1.31. Consider the ideal generated by the maximal minors, i.e., the ten3 3 × 3
minors of the matrix
 
x1 x2 x3 x4 x5
 y1 y2 y3 y4 y5 
z1 z2 z3 z4 z5

and the lexicographic order on k[x1 , . . . , z5 ]. The leading terms of the minors and the mini-

3 5 5



3 = 2 = 10.
4. GRÖBNER BASES AND BUCHBERGER’S ALGORITHM 11

mal generators of the corresponding monomial ideals Mi are:

x1 y2 z3
x1 y2 z4 M2 = hz3 i
x1 y3 z4 M3 = hy2 i
x2 y3 z4 M4 = hx1 i
x1 y2 z5 M5 = hz3 , z4 i
x1 y3 z5 M6 = hy2 , z4 i
x2 y3 z5 M7 = hx1 , z4 i
x1 y4 z5 M8 = hy2 , y3 i
x2 y4 z5 M9 = hx1 , y3 i
x3 y4 z5 M10 = hx1 , x2 i

So only 15 out of 10


2
= 45 S-vectors are needed in Buchberger’s criterion. In fact, the
criterion shows that the minors form a Gröbner basis.
In the situation of the criterion, we refer to the selection of the indices j = j(i, α) to-
gether with the computation of the remainders hi,α as Buchberger’s test. It is clear from the
criterion that the amount of computation needed for the test depends in a crucial way on
the order in which we list f1 , . . . , fr .
Algorithm 1.32 (Buchberger’s algorithm). Given I = hf1 , . . . , fr i ≤ F , compute a Gröb-
ner basis for I.

(1) Set ` = r.

(2) For i = 2, . . . , `, and for each minimal monomial generator xα of

Mi = hL(f1 ), . . . , L(fi−1 )i : L(fi )  R,

compute a remainder hi,α as described above.

(3) If some hi,α is nonzero, set ` = ` + 1, f` = hi,α , and go back to Step (2).

(4) Return f1 , . . . , f` .
End
Correctness and termination follow from Buchberger’s criterion and the discussion pre- lect. 4
ceeding it. 09.05
Corollary 1.33. Given polynomial vectors f1 , . . . , fr ∈ F \ {0}, a Gröbner basis for I :=
hf1 , . . . , fr i ≤ F can be computed in finitely many steps.
Example 1.34. Let f1 = x2 , f2 = xy − y 2 ∈ k[x, y] with >lex . Then L(f2 ) = xy and
M2 = hxi. We compute the standard expression

S(f2 , f1 ) = x · f2 − y · f1 = −xy 2 = 0 · f1 − y · f2 − y 3 ,

and add the nonzero remainder f3 := −y 3 to the set of generators. Then M3 = hx2 , xi = hxi,
and it suffices to consider the S-polynomial S(f3 , f2 ). Computing the standard expression

S(f3 , f2 ) = x · f3 + y 2 · f2 = −y 4 = 0 · f1 + 0 · f2 + y · f3

with remainder zero, we find that f1 , f2 , f3 form a Gröbner basis for the ideal I = hf1 , f2 i.
We visualize, once more, the monomials in k[x, y]:
12 CHAPTER 1. GRÖBNER BASES

• • • • •

(0, 3) • • • • •
− • • • •
− •
(1, 1)
• • •
| •
(2, 0)
• •

The dots in the shaded region correspond to the monomials in the ideal L(I) which is
minimally generated by x2 , xy, and y 3 . The monomials 1, x, y, y 2 respresented outside the
shaded region are the standard monomials. Due to Macaulay’s Theorem 1.1, their residue
classes form a k-vector space basis for k[x, y]/I. Hence, every class g + I ∈ k[x, y]/I is
canonically represented by a uniquely determined k-linear combination a + bx + cy + dy 2
(see Remark 1.27). To add and multiply residue classes, we add and multiply the canonical
representatives according to the rules in Exercise 1.3. The multiplication in k[x, y]/I is, thus,
determined by the following table (we write f = f + I):

· 1 x y y2
1 1 x y y2
x x 0 y2 0
y y y2 y2 0
y2 y2 0 0 0

Exercise 1.5. Let f1 = x2 y − y 3 , f2 = x3 ∈ k[x, y] with >lex as in Example 1.20. Compute

a Gröbner basis for the ideal I = hf1 , f2 i. Visualize the monomials in L(I), and compute a
multiplication table for k[x, y]/I.
In general, the products in a multiplication table as above are not represented by terms
only:
Exercise 1.6. A polynomial in R is called a binomial if it has at most two terms. An
ideal of R is called a binomial ideal if it is generated by binomials.
Now let > be any global monomial order on R. Show that the following conditions on
an ideal I  R are equivalent:

(1) I is a binomial ideal.

(2) I has a binomial Gröbner basis, that is, a Gröbner basis consisting of binomials.

(3) The normal form mod I of any monomial is a term.

(4) The multiplication table of R/I consists of terms only.

Now, we come to the proof of Buchberger’s criterion. For this, recall that the S-vectors
are designed to cancel leading terms: The L(f` )’s fit into the relations

mji L(fi ) − mij L(fj ) = 0. (4.1)

On the other hand, the f` ’s themselves fit into the relations

(ij) (ij) (ij)
−g1 f1 − · · · + (−mij − gj )fj − · · · + (mji − gi )fi − · · · − gr(ij) fr = 0 (4.2)

arising from standard expressions with remainder zero:

(ij)
S(fi , fj ) = g1 f1 + . . . + gr(ij) fr + 0
4. GRÖBNER BASES AND BUCHBERGER’S ALGORITHM 13

Buchberger’s criterion then reads that f1 , . . . , fr form a Gröbner basis iff every relation
of type (4.1) considered in Buchberger’s test “lifts" to a relation of type (4.2) such that
(ij)
L(S(fi , fj )) ≥ L(g` f` ) whenever both sides are nonzero. In general, we think of a relation

g1 f1 + · · · + gr fr = 0 ∈ F

as a row vector (g1 , . . . , gr ) ∈ Rr :

Definition 1.35. Let R be a ring, let M be an R-module, and let f1 , . . . , fr ∈ M . A syzygy
on f1 , . . . , fr is an element of the kernel of the homomorphism

ϕ : Rr → M, εi 7→ fi ,

where {ε1 , . . . , εr } is the canonical basis of Rr . We call ker ϕ the (first) syzygy module of
f1 , . . . , fr , written
Syz(f1 , . . . , fr ) = ker ϕ.
Syz(f1 , . . . , fr ) is as a module over a noetherian ring R finitely generated and we regard the
elements of a given finite set of generators for it as the rows of a matrix which we call a
syzygy matrix of f1 , . . . , fr .
Exercise 1.7. Determine a syzygy matrix of x, y, z ∈ k[x, y, z].
To handle the syzygies on the elements f1 , . . . , fr of a Gröbner basis, we consider the
free module F1 = Rr with its canonical basis {ε1 , . . . , εr } and the induced monomial order
>1 on F1 defined by setting

xα εi >1 xβ εj ⇐⇒ xα L(fi ) > xβ L(fj ), or

xα L(fi ) = xβ L(fj ) (up to a scalar) and i > j.

Note that >1 is global if this is true for > (what we suppose, here).
Proof of Buchberger’s criterion (Theorem 1.30). Write I = hf1 , . . . , fr i ≤ F . If f1 , . . . , fr form
a Gröbner basis for I, all remainders hi,α are zero by Proposition 1.24. Indeed, the S-vectors
are contained in I.
Conversely, suppose that all the hi,α are zero. Then, for every pair (i, α), we have a
standard expression of type
(ij)
S(fi , fj ) = g1 f1 + . . . + gr(ij) fr + 0,

where j = j(i, α) < i is as selected in Buchberger’s test. Let

(ij) (ij) (ij)
S (i,α) := (−g1 , . . . , −mij − gj , . . . , mji − gi , . . . , −gr(ij) ) ∈ F1 = Rr

be the corresponding syzygy on f1 , . . . , fr (we suppress the index j in our notation on the
left hand side), and S matrix of syzygies4 with rows S (i,α) in some order. Hence
 
f1
 .. 
S  .  = 0.
fr

On F1 we consider the induced monomial order. The leading term of S (i,α) with respect to
this order is
L(S (i,α) ) = mji εi .
4
We will prove in Corollary 1.36 that S is even a syzygy matrix.
 14 CHAPTER 1. GRÖBNER BASES

Indeed,
mji L(fi ) = mij L(fj ), but i > j,
and
(ij)
mji L(fi ) > L(S(fi , fj )) ≥ L(g` )L(f` )
whenever these leading terms are nonzero.
To prove that the f` ’s form a Gröbner basis for I, let g be any nonzero element of I,
say g = a1 f1 + . . . + ar fr , where a1 , . . . , ar ∈ R. The key point of the proof
Pr is to replace
this representation of g in terms of the f` ’s by a standard expression g = k=1 g` f` (with
remainder zero). The result, then, follows by applying Remark 1.22 on leading terms in
standard expressions:
L(g) = max{L(g1 )L(f1 ), . . . , L(gr )L(fr )} ∈ hL(f1 ), . . . , L(fr )i.
To find the desired standard expression, we go back and forth between F1 = Rr and F :
Consider the polynomial row vector A := (a1 , . . . , ar ) ∈ Rr , and let G = (g1 , . . . , gr ) ∈ F1
be the remainder of A under determinate division by the syzygies S (i,α) ’s (listed in some
order). Then
g = a1 f1 + . . . + ar fr = g1 f1 + . . . + gr fr (4.3)
since the S (i,α) are syzygies5 on f1 , . . . , fr . We show that the right hand side of (4.3) satisfies
condition (DD1) of determinate division by the f` ’s (in particular, it is a standard expres-
sion). Suppose the contrary. Then there is a pair k < i such that one of the terms of gi L(fi )
is divisible by L(f` ). In turn, one of the terms of gi is contained in the monomial ideal
Mi = hL(f1 ), . . . , L(fi−1 )i : L(fi )  R
which is generated by the mji ’s selected in Buchberger’s test. This means that one of the
terms of G is divisible by some mji εi = L(S (i,α) ), contradicting the fact that according to
how we found G, the terms of G satisfy condition (DD2) of determinate division by the
End S (i,α) in F1 .
lect. 5 Corollary 1.36. If f1 , . . . , fr ∈ F \{0} form a Gröbner basis with respect to >, the S (i,α) consid-
11.05 ered in the proof of Buchberger’s criterion form a Gröbner basis for the syzygy module Syz(f1 , . . . , fr )
with respect to the induced monomial order. In particular, the S (i,α) ’s generate the syzygies on
f1 , . . . , f r .
Proof. Let A ∈ Rr be an arbitrary syzygy on f1 , . . . , fr , and let G = (g1 , . . . , gr ) ∈ Rr be the
remainder of A under determinate division by the S (i,α) ’s (listed in some order). Then, since
A and the S (i,α) ’s are syzygies on f1 , . . . , fr , the same must be true for G:
0 = g1 f1 + . . . + gr fr .
Furthermore, as shown in the proof of Buchberger’s criterion, the gi ’s satisfy condition
(DD1) of determinate division by f1 , . . . , fr . Since standard expressions under determinate
division are uniquely determined, the gi and, thus, G must be zero. Taking, once more,
Remark 1.22 on leading terms in standard expressions into account, we find that L(A) is
divisible by some L(S (i,α) ). The result follows.
Remark 1.37. The S in S-vector stands for syzygy. In fact, as is now clear, the relations
mji L(fi ) − mij L(fj ) = 0 corresponding to the S(fi , fj ) generate Syz(L(f1 ), . . . , L(fr )). In
our version of Buchberger’s test, we focus on the minimal monomial generators of Mi .
This means that we select a subset X ⊂ {S(fi , fj ) | j < i} such that the relations mji L(fi ) −
mij L(fj ) = 0 corresponding to the S(fi , fj ) in X still generate Syz(L(f1 ), . . . , L(fr )). It is this
property of X which makes our proof of Buchberger’s criterion work. Hence, in stating the
criterion, we may choose any set of S-vectors satisfying this property.
5
g = Af = (BS + G)f = Gf since Sf = 0.
4. GRÖBNER BASES AND BUCHBERGER’S ALGORITHM 15

Remark 1.38. Let f1 , . . . , fr ∈ F \ {0}, and let I = hf1 , . . . , fr i ≤ F . If we compute a

Gröbner basis f1 , . . . , fr , fr+1 , . . . , fr0 for I using Buchberger’s algorithm, the syzygies S (i,α)
generating Syz(f1 , . . . , fr , fr+1 , . . . , fr0 ) are obtained in two ways. Either, S (i,α) arises from a
division leading to a new generator f` , k ≥ r + 1:
(ij) (ij)
S(fi , fj ) = g1 f1 + . . . + g`−1 f`−1 + f` .

Or, S (i,α) arises from a division with remainder zero:

(ij) (ij)
S(fi , fj ) = g1 f1 + . . . + g` f` + 0.

Example 1.39. In Example 1.34, the matrix

 
−y x + y −1
0 y2 x−y

is a syzygy matrix of f1 = x2 , f2 = xy − y 2 , f3 = −y 3 ∈ k[x, y].

A Gröbner basis {f1 , . . . , fr } ⊂ F computed with Buchberger’s algorithm quite often
contains elements whose leading terms are unneeded generators for L(hf1 , . . . , fr i). By
eliminating superfluous generators and by adjusting constants such that the coefficient of
each leading term is 1, we get a so-called minimal Gröbner basis, meaning, a Gröbner
basis whose leading terms are the minimal generators of L(hf1 , . . . , fr i). In addition, we
may “reduce the tail” of each element in the Gröbner basis. A minimal Gröbner basis
{f1 , . . . , fr } ⊂ F is called (tail) reduced if, for i 6= j, no term of fi is divisible by L(fj ).
Exercise 1.8. (*) Show that if h0i =
6 I ⊂ F is a submodule, there is a uniquely determined
reduced Gröbner basis for I with respect to the given monomial order, namely

m1 − NF(m1 , I), . . . , mr − NF(mr , I),

where m1 , . . . , mr are the minimal generators of L(I). Explain how to compute the reduced
Gröbner basis from any given Gröbner basis.
Exercise 1.9. Design an algorithm to decide whether an ideal hf1 , . . . , fr i ⊂ k[x1 , . . . , xn ]
is principal or not.6
Remark 1.40. In the situation of Corollary 1.36, the syzygies S (i,α) form a minimal Gröb-
ner basis.
Proof. Indeed, with respect to the induced order, L(S (i,α) ) = c(i,α) xα εi , where xα is a minimal
generator of the monomial ideal Mi . If L(S (i,α) ) is divisible by L(S (j,β) ), then i = j and xα
is disible by xβ . Hence, xα = xβ and, thus, L(S (i,α) ) = L(S (j,β) ).
Remark 1.41. Buchberger’s algorithm generalizes both Gaussian elimination and Eu-
clid’s GCD algorithm:
(1) Given homogeneous degree-1 polynomials

fi = ai1 x1 + · · · + ain xn ∈ k[x1 , . . . , xn ], i = 1, . . . , r,

let > be a global monomial order on k[x1 , . . . , xn ] such that x1 > · · · > xn . Computing
a minimal Gröbner basis for hf1 , . . . , fr i amounts, then, to transforming the coefficient
matrix A = (aij ) into a matrix in row echelon form (REF) with pivots 1. And com-
puting a reduced Gröbner basis for hf1 , . . . , fr i amounts to computing a row-reduced
echelon form (RREF) with pivots 1.
6
The general problem of computing the minimal generating number of a polynomial ideal is still an open.
 16 CHAPTER 1. GRÖBNER BASES

(2) In the case of one variable x, there is precisely one global monomial order: 1 < x <
x2 < · · · . Given f1 , f2 ∈ k[x], the reduced Gröbner basis for hf1 , f2 i with respect to this
order consists of exactly one element, namely the greatest common divisor gcd(f1 , f2 ),
and Buchberger’s algorithm takes precisely the same steps as Euclid’s algorithm for
computing the gcd.

5 First applications: Submodule membership and syzygies

As already remarked, division with remainder, Proposition 1.24, and Buchberger’s algo-
rithm allow us to decide submodule (ideal) membership:
Algorithm 1.42 (Submodule Membership). Given a free module F over R with a fixed
finite basis, and given nonzero elements g, f1 , . . . , fr ∈ F , decide whether

g ∈ I := hf1 , . . . , fr i ⊂ F.

[If so, express g as an R-linear combination

g = g1 f1 + . . . + gr fr .]

(1) Compute a Gröbner basis f1 , . . . , fr , fr+1 , . . . , fr0 for I using Buchberger’s algorithm.
[Store each syzygy arising from a division which leads to a new generator f` in Buch-
berger’s test.]
(2) Compute a standard expression for g in terms of f1 , . . . , fr0 with remainder h (use the
same global monomial order on F as in Step 1).
(3) If h = 0, then g ∈ I. [In this case, for ` = r0 , . . . , r + 1, successively do the following: in
the standard expression computed in Step 2, replace f` by the expression in terms of
f1 , . . . , f`−1 given by the syzygy leading to f` in Step 1.]
Example 1.43. Let

g = x3 − x2 + xy 2 , f1 = x2 , f2 = xy − y 2 ∈ k[x, y].

In step 1, computing the lexicographic Gröbner basis f1 , f2 , f3 = −y 3 for I = hf1 , f2 i ⊂

k[x, y] as in Example 1.34, the standard expression leading to f3 is

x · f2 − y · f1 = −xy 2 = 0 · f1 − y · f2 − y 3 .

In step 2, determinate division yields the standard expression

g = (x − 1) · f1 + y · f2 − f3

with remainder zero. In step 3, we conclude that g ∈ I. Moreover, substituting (x + y) · f2 −

y · f1 for f3 , we get
g = (x − 1 + y) · f1 − x · f2 .
End
lect. 6 Exercise 1.10 ((*) Equality of Submodules). Let F be a free R-module, and let I, J ⊂ F
16.05 be submodules. Note that if I, J are monomial submodules, then I = J iff I and J have the
same minimal generators. Show that if > is any global monomial order on F , then

I ⊂ J and L(I) = L(J) =⇒ I = J.

Finally, note that if I, J are arbitrary, then

I = J ⇐⇒ I = I + J and J = I + J.
6. ITERATED SYZYGIES AND HILBERT’S SYZYGY THEOREM 17

We already know that Buchberger’s algorithm computes the syzygies on the elements
of a Gröbner basis (see Corollary 1.36). Based on this, we can compute the syzygies on any
given set of generators:
Algorithm 1.44 (Syzygy Modules). Given a free R-module F with a fixed finite basis
and polynomial vectors f1 , . . . , fr ∈ F \ {0}, compute a syzygy matrix of f1 , . . . , fr .
(1) Compute a Gröbner basis f1 , . . . , fr , fr+1 ,. . . , fr0 for hf1 , . . . , fr i ⊂ F using Buchberger’s
algorithm. On your way, store each syzygy on f1 , . . . , fr0 obtained in Buchberger’s
test. Let t be the number of these syzygies.

(2) Arrange the syzygies such that those obtained from a division leading to a new gen-
erator f` are first (and those arising from a division with remainder zero
 aresecond).
A B
Then the syzygies fit as rows into a t×r0 matrix which has block form , where
C D
B is a lower triangular square matrix of size r0 − r with diagonal entries 1 (if signs are
adjusted appropriately).

(3) The (t − r0 + r) × r matrix C − DB −1 A is a syzygy matrix of f1 , . . . , fr .

 
A B
Proof of correctness. By Corollary 1.36, the rows of generate all the syzygies on
C D  
Er0 −r 0
f1 , . . . , fr , fr+1 , . . . , fr0 . We multiply the invertible t × t matrix −1 with
  −DB E t−r0 +r
A B
, where Ej stands for the j × j identity matrix. As a result, we obtain new genera-
C D  
A B
tors for the syzygies, namely the rows of the matrix M = . An R-linear
C − DB −1 A 0
combination of the rows of M defines a syzygy just on f1 , . . . , fr iff its last r0 − r entries are
zero. It is, then, an R-linear combination of the last t − r0 + r rows of M since B has maximal
rank. We conclude that C − DB −1 A is a syzygy matrix of f1 , . . . , fr .
Example 1.45. Recall from Example 1.39 how we computed the lexicographic Gröbner
basis f1 = x2 , f2 = xy − y 2 , f3 = −y 3 for the ideal I = hf1 , f2 i ⊂ k[x, y]. With conventions as
in Algorithm 1.44, the resulting syzygy matrix is
   
A B y −x − y 1
= .
C D 0 y2 x−y

Thus, Syz(f1 , f2 ) is generated by the single trivial syzygy

0 y 2 − (x − y) y −x − y = −f2 f1 .

6 Iterated syzygies and Hilbert’s Syzygy Theorem

Theorem 1.46 (Hilbert’s Syzygy Theorem (matrix version)). Starting with a minimal
Gröbner basis over R = k[x1 , . . . , xn ] one can arrange the successive syzygies obtained by Corol-
lary 1.36 in a way which yields a zero matrix of syzygies after at most n steps.
Proof. This proof is algorithmic. Let f1 , . . . , fs1 be a minimal Gröbner basis with respect to
some global monomial order on Rs0 . Consider the induced monomial order on Rs1 and the
syzygies S (i,α) ∈ Rs1 obtained by applying Buchberger’s test to f1 , . . . , fs1 . Then, as pointed
out in Remark 1.40, the S (i,α) form a minimal Gröbner basis for the submodule of syzygies
of f1 , . . . , fs1 .
18 CHAPTER 1. GRÖBNER BASES

Computing the syzygies on the S (i,α) and so forth, we successively get minimal Gröbner
bases which generate the successive syzygy modules of f1 , . . . , fs1 of higher order. At each
stage, the new Gröbner basis depends, in particular, on how we arrange the elements of
the Gröbner basis computed in the previous step. We show that if this arrangement is done
properly, then the process just described will terminate with a zero syzygies matrix after at
most n steps.
To begin, fix an integer 1 ≤ ` ≤ n such that none of the leading terms L(fi ) involves the
variables x`+1 , . . . , xn (choose ` = n if one of the L(fi )’s involves xn ). In Buchberger’s test,
let the fi ’s be arranged such that, for j < i, the exponent of x` in L(fj ) is smaller than or
equal to that of x` in L(fi ) whenever these leading terms involve the same basis element of
Rs0 . Then none of the resulting leading terms L(S (i,α) ) = c(i,α) xα εi involves x` , . . . , xn .
Arranging the Gröbner basis elements at each stage of our process accordingly, we ob-
tain after, say, ` ≤ k steps a Gröbner basis G` of higher syzygies for which none of the
leading terms L(g), g ∈ G` , involves x1 , . . . , xn . Having (chosen) a minimal Gröbner basis
in the previous step implies that G` = {0}.
Example 1.47. Consider the ideal
I = hf1 , . . . , f5 i ⊂ R = k[w, x, y, z]
generated by the polynomials
f1 = w2 − xz, f2 = wx − yz, f3 = x2 − wy, f4 = xy − z 2 , f5 = y 2 − wz.
As Buchberger’s test shows, these generators already form a Gröbner basis for I with re-
spect to the degree reverse lexicographic order on k[w, x, y, z], which is obviously minimal.
The left most column of the table below is the matrix ϕ1 consisting of the minimal gener-
ators f1 , . . . , f5 . All leading terms are printed in bold. We successively obtain two syzygy
matrices ϕtr 2 and ϕ3 which we display in a compact way as follows:

w2 − xz −x y 0 −z 0 −y 2 + wz
wx − yz w −x −y 0 z z2
x2 − wy −z w 0 −y 0 0
xy − z 2 0 0 w x −y yz
y2 − wz 0 0 −z −w x w2
0 y −x w −z 1
−y 2 + wz z 2 −wy yz −w2 x
The transposed syzygy matrix ϕtr 2 resulting from the test is the 5 × 6 matrix in the middle
of our table. Note that, for instance, M4 = hw, xi can be read from the 4th row of ϕtr 2 . At this
point, we already know that the rows of ϕ2 form a Gröbner basis for Syz(f1 , . . . , f5 ) with
respect to the induced monomial order on R5 . Buchberger’s test applied to these Gröbner
basis elements yields a 2 × 6 syzygy matrix ϕ3 which is printed in the two bottom rows
of our table. There are no syzygies on the two rows of ϕ3 since the leading terms involve
different basis vectors.
Observe that once we have the leading terms of the Gröbner basis for I, we can easily
compute the leading terms of the Gröbner bases for all syzygy modules, that is, all bold
face entries of our table. This gives us an early idea on the amount of computation lying
ahead.
With respect to our proof of the syzygy theorem, note that the variable z is missing from
the L(f` )’s, and the f` ’s are ordered such that the exponent of y in L(fj ) is smaller than or
equal to that in L(fi ) whenever j < i. As a consequence, none of the leading terms of the
columns of ϕtr 2 involves y. In the next step, however, the leading term of the second row of
ϕ3 involves x. What is wrong with how we arrange the columns of ϕtr 2?
7. ELIMINATION 19

Exercise 1.11. Following the matrix version of Hilbert’s Syzygy Theorem 1.46 compute
all successive syzygy matrices of the 2 × 2 minors of the matrix
 
x0 x1 x2 x3
.
x1 x2 x3 x4
End
lect. 7
7 Elimination 23.05

Definition 1.48. If I ⊂ R is an ideal, its `-th elimination ideal is the ideal

I` = I ∩ k[xk+1 , . . . , xn ].

In particular, I0 = I.
Algorithm 1.49 (Elimination using >lex ). Given I = hf1 , . . . , fr i ⊂ R, compute all elimi-
nation ideals I` using the following steps:
(1) Compute a Gröbner basis G for I with respect to >lex on k[x1 , . . . , xn ].
(2) For any k, the elements g ∈ G with L(g) ∈ k[x`+1 , . . . , xn ] form a Gröbner basis for I`
with respect to >lex on k[x`+1 , . . . , xn ].
Proof of correctness. If f ∈ I ∩ k[x`+1 , . . . , xn ], then L(f ) is divisible by L(g) for some g ∈
G. Since f does not involve x1 , . . . , x` , also L(g) does not involve x1 , . . . , x` . By the key
property of >lex (Remark 1.15), we must have g ∈ k[x`+1 , . . . , xn ].
Example 1.50. Let I = hf1 , f2 i ⊂ k[x, y, z], with f1 = x2 − y, f2 = xy − z. Then I
is the vanishing ideal of the twisted cubic curve t 7→ (t, t2 , t3 ) = (x, y, z). We compute a
lexicographic Gröbner basis for I. To begin with, M2 = hx2 i : xy = hxi, and we have the
standard expression
S(f2 , f1 ) = x(xy − z) − y(x2 − y) = −xz + y 2 =: f3 .
We add f3 to the set of generators. Then M3 = hx2 , xyi : xz = hx, yi, and we have the
standard expressions
S(f3 , f1 ) = x(−xz + y 2 ) + z(x2 − y) = xy 2 − yz = y(xy − z)
and
S(f3 , f2 ) = y(−xz + y 2 ) + z(xy − z) = y 3 − z 2 =: f4 .
In the next step, M4 = hx2 , xy, xzi : y 3 = hxi, and
S(f4 , f2 ) = x(y 3 − z 2 ) − y 2 (xy − z) = −xz 2 + y 2 z = z(−xz + y 2 )
is a standard expression with remainder zero. Hence, f1 , f2 , f3 , f4 form a Gröbner basis for
I.
We visualize the monomials in L(I) via their exponent vectors:

5
5 5

y
x
0 0
20 CHAPTER 1. GRÖBNER BASES

The computaion shows that the elimination ideal I1 ⊂ k[y, z] is generated by the polynomial
f4 = y 3 − z 2 . The geometric interpretation of this is projection:

In the yz-plane, the equa-

tion f4 = 0 defines the
image of the twisted cu-
bic curve V(f1 , f2 ) under
the projection which sends
(a, b, c) to (b, c).

Exercise 1.12. In the previous example, f1 = x2 − y, f2 = xy − z, −f3 = xz − y 2 , and

f4 = y 2 − z 3 form the reduced lexicographic Gröbner basis for the ideal I = hf1 , f2 i. In
contrast, show that f1 , f2 , and f3 form the reduced Gröbner basis with respect to >drlex .
A single Gröbner basis computation with respect to >lex yields the whole flag of elim-
ination ideals I` , k = 0, . . . , n − 1. If only one of the elimination ideals is needed, other
monomial orders are usually more effective.
Definition 1.51. A monomial order > on the polynomial ring
k[x, y] = k[x1 , . . . , xn , y1 , . . . ym ]
is an elimination order with respect to x1 , . . . , xn if the following holds for all f ∈ k[x, y]:
L(f ) ∈ k[y] =⇒ f ∈ k[y].
Example 1.52. Let >1 on k[x] and >2 on k[y] be monomial orders. The product order
(or block order) > = (>1 , >2 ) on k[x, y] is defined by
xα y γ > xβ y δ :⇐⇒ xα >1 xβ , or (xα = xβ and y γ >2 y δ ).
It is an elimination order with respect to x1 , . . . , xn which is global if >1 and >2 are global.
Choosing >1 and >2 to be degree reverse lexicographic is often most efficient.
As for >lex , one shows:
Proposition 1.53 (Elimination). Let I ⊂ k[x, y] be an ideal, let > be a global elimination
order on k[x, y] with respect to x1 , . . . , xn , and let G be a Gröbner basis for I with respect to >.
Then G ∩ k[y] is a Gröbner basis for I ∩ k[y] with respect to the restriction of > to k[y].
As already indicated in Example 1.50, the geometric meaning of elimination is projec-
tion. Applying projection to the graph of an arbitrary morphism ϕ of affine spaces, given
by polynomials f1 , . . . , fm ∈ R, we will find a way of computing the so-called Zariski clo-
sure of the image of a given algebraic set under ϕ. The corresponding algebraic result is
our next topic in this section. Rather than considering R-module relations as in Definition
1.35, we are, now, interested in k-algebra relations:
Definition 1.54. Let S be a k-algebra, and let s1 , . . . , sm be elements of S. A k-algebra
relation on s1 , . . . , sm is a polynomial expression of type
X
cα sα1 1 · · · sαmm = 0 ∈ S,
with coefficients cα ∈ k. Formally, consider a polynomial ring k[y1 , . . . , ym ] and think of a
k-algebra relation as an element of the kernel of the homomorphism
ϕ : k[y1 , . . . , ym ] → S, yi 7→ si .
If only the trivial such relation exists, s1 , . . . , sm are algebraically independent over k.
7. ELIMINATION 21

Proposition 1.55 (Algebra relations in affine rings). Let I be an ideal of R, and let f 1 =
f1 + I, . . . , f m = fm + I ∈ S := R/I. Consider the homomorphism
ϕ : k[y1 , . . . , ym ] → S = R/I, yi 7→ f i .
If J is the ideal7
J = I k[x, y] + hf1 − y1 , . . . , fm − ym i ⊂ k[x, y]
then
ker ϕ = J ∩ k[y].
Proof. Let g ∈ k[y] ⊂ k[x, y]. To prove the assertion, we have to show:
g(f1 , . . . , fm ) ∈ I ⇐⇒ g ∈ J.
P
If g = h+ gj (fj −yj ) ∈ J, with h ∈ I k[x, y], then g(f1 , . . . , fm ) = h(x1 , . . . , xn , f1 , . . . , fm ) ∈
I k[x, y] ∩ k[x] = I.
For the converse, observe that substituting the fj −(fj −yj ) for the yj in g and expanding
gives an expression of type
X
g(y) = g(f1 , . . . , fm ) + gj (fj − yj ).

Since we already know how to compute in affine rings (i.e., residue class rings of poly-
nomial rings), a particular application of the proposition is a method for computing in the
algebra k[f 1 , . . . , f m ] ∼
= k[y1 , . . . , ym ]/ ker ϕ. Once we have the required Gröbner basis for J,
we know, in particular, whether ker ϕ = 0. That is, we can decide whether f 1 , . . . , f m are
algebraically independent over k. In addition, we can check whether ϕ is surjective:
Exercise 1.13 ((*) Subalgebra Membership). With notation as above, let g, f 1 , . . . , f m be
elements R/I, and let > be a global elimination order on k[x, y] with respect to x1 , . . . , xn .
Show:
(1) We have g ∈ k[f 1 , . . . , f m ] iff the normal form h = NF(g, J) ∈ k[x, y] is contained in
k[y]. In this case, g = h(f 1 , . . . , f m ) is a polynomial expression for g in terms of the f ` .
(2) The homomorphism ϕ : k[y1 , . . . , ym ] → R/I is surjective iff NF(xi , J) ∈ k[y] for
i = 1, . . . , n.
Exercise 1.14.
(1) Compute the algebra relations on the polynomials
f1 = x2 + y 2 , f2 = x2 y 2 , f3 = x3 y − xy 3 ∈ k[x, y].

(2) Consider the polynomials

g = x4 + y 4 , g1 = x + y, g2 = xy ∈ k[x, y].
Show that g is contained in the subalgebra k[g1 , g2 ] ⊂ k[x, y], and express g as a poly-
nomial in g1 , g2 .
(3) Consider the endomorphism ϕ of k[x1 , x2 , x3 ] defined by
x1 7→ x2 x3 , x2 7→ x1 x3 , x3 7→ x1 x2 .
Prove that ϕ induces an automorphism of
k[x1 , x2 , x3 ]/hx1 x2 x3 − 1i.

7
the vanishing ideal of the graph of ϕ
22 CHAPTER 1. GRÖBNER BASES
Chapter 2

Constructive A BELian categories

1 Categories
The notion of a category is a special case of what is nowadays called a horizontal cat-
egorification or oidification of the notion of a monoid. Another funny name would be
“monoidoid”. This will become clear later.
Definition 2.1. A quiver (or multi-digraph) A consists of a class of objects A0 = Obj A
and a class of morphisms (or arrows) A1 = Mor A together with two defining maps

s, t : A1 ⇒ A0 ,

called source and target, respectively.

We write HomA (M, N ) (sometimes also A(M, N )) for the fiber (s, t)−1 ({(M, N )}) of the
product map (s, t) : A1 → A0 × A0 over the pair (M, N ) ∈ A0 × A0 . This is the class of
all morphisms ϕ with source s(ϕ) = M and target t(ϕ) = N . We indicate this by writing
ϕ
→ N . Hence, A1 is the disjoint union A1 = ˙ (M,N )∈A0 ×A0 HomA (M, N ).
S
ϕ : M → N or M −
As usual EndA (M ) := HomA (M, M ). If the class HomA (M, N ) is a set for all pairs (M, N ),
then we call the quiver locally small. We therefore talk about the Hom-sets. If additionally
A0 is a set then the quiver is called small.
Convention. From now on all quivers will be assumed locally small.
End
Definition 2.2. A category A is a quiver with two further defining maps lect. 8
30.05
1 µ
A0 → A1 ← A1 ×A0 A1 ,

called identity and composition, respectively, subject to a list of defining properties. To ex-
press these properties in a suggestive way we first introduce some notational conventions:

• The notation used in µ : A1 ×A0 A1 → A1 means that µ is only defined for pairs of
morphisms (ϕ, ψ) such that t(ϕ) = s(ψ). We call such a pair composable. If not stated
otherwise, we write ϕψ instead of µ(ϕ, ψ). We call this the left convention, where
the right convention1 would be to write ψϕ instead of ϕψ. In any case we call ϕ the
pre-morphism and ψ the post-morphism.

• We use latin letters for objects and greek letters for morphisms, so M ∈ A and ϕ ∈ A
means M ∈ A0 and ϕ ∈ A1 . We also refer to them as A-objects and A-morphisms.

• We write 1M for 1(M ).

1
This is the usual convention followed in [HS97, §II.1, p. 41].

23
24 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

We now state the defining properties:

(1) s(1M ) = t(1M ) = M , i.e., 1M ∈ EndA (M ) for all M ∈ A.

In words: the identity respects source and target.

(2) s(ϕψ) = s(ϕ) and t(ϕψ) = t(ψ) for all composable morphism ϕ, ψ ∈ A.
In words: the composition respects source and target.
We express this as follows:

µ : HomA (M, L) × HomA (L, N ) → HomA (M, N ),

(ϕ , ψ) 7→ ϕψ

(3) (ϕψ)ρ = ϕ(ψρ) for all composable morphisms ϕ, ψ, ρ ∈ A.

In words: The composition is associative.

(4) 1M ϕ = ϕ and ψ1M = ψ for all ϕ with s(ϕ) = M and all ψ with t(ψ) = M .
In words: The identity is a left and right unit of the composition.

Remark 2.3.

• The objects are in bijection to the identities via M 7→ 1M . So for many further con-
structions we only need to refer to morphisms.

• It follows from the axioms that the set End(M ) is a monoid for all M ∈ A. In partic-
ular, a category with one object is nothing but a monoid. This is what we mean by
saying that “a category is a horizontally categorified monoid”, or a “monoidoid”.

• Isomorphic objects in a category need not be equal. In fact, this is one of the reasons
for the flexibility of the notion of a category. A category in which isomorphic objects
in A are equal is called skeletal.

Example 2.4.

(1) The category (Sets) with sets as objects and maps as morphisms.

(2) The category (Sets0 ) with pointed sets as objects and maps preserving the distin-
guished points as morphisms.

(3) The category (Grps) with groups as objects and group homomorphisms as morphisms.

(4) The category (Ab) = Z−Mod with A BELian groups as objects and group homomor-
phisms as morphisms.

(5) The category (Rngs) with (not necessarily unital) rings as objects and ring homomor-
phisms (not necessarily respecting one) as morphisms.

(6) The category (uRngs) with unital rings as objects and ring homomorphisms (not nec-
essarily respecting one) as morphisms.

(7) The category (URngs) with unital rings as objects and unital ring homomorphisms as
morphisms.

(8) The category (Top) with topological spaces as objects and continuous maps as mor-
phisms.

(9) The category (Top0 ) with pointed topological spaces as objects and continuous maps
preserving the distinguished points as morphisms.
2. THE CATEGORY OF MODULES 25

(10) The category of truth values with objects {>, ⊥} and a single non-identity morphism
⊥ → > saying that false implies true.

(11) A proset (=preordered set) is a thin small category, i.e., with at most one morphism
between two objects.

(12) A poset (=partially ordered set) is a skeletal thin small category.

(13) The simplicial category ∆ with totally ordered sets n = {0 < 1 < · · · < n} for
n ∈ N = {0, 1, ...} as objects and order-preserving maps (i ≤ j =⇒ f (i) ≤ f (j)) as
morphisms.
Non of the above category names in round brackets is standard.

2 The category of modules

Rings are, like groups and varieties, highly nonlinear objects. Such non-linear objects can
be studied in terms of the category of all their linear “approximations”. These associated
linear objects are usually called modules.

2.1 Objects
Because of the possible noncommutativity of rings we have to distinguish2 between left
and right modules.
Definition 2.5. A left R-module M is an A BELian group on which R acts by endomor-
phisms from the left. This action is given by a ring homomorphism ρ : R → (End(M ), ◦),
called a left representation of R. A right representation ρ : R → (End(M ), ·) defines a
right R-module. We occasionally denote the left (resp. right) R-module M by R M (resp.
MR ).
As usual, the left representation ρ can be interpreted as the C URRYing3 of a left action
map
αρ : R × M → M, (r, m) 7→ rm,
which is bilinear, associative, and satisfies 1m = m for all m ∈ M . In other words,
αρ (r, m) = ρ(r)(m). Likewise for right representations and right action maps αρ : M × R →
M, (m, r) 7→ mr.
Example 2.6. Checking the following examples is an easy exercise:
(1) Each A BELian group is a Z-module in a unique way.

(2) A k-module over a (skew)field k is the same as a k vector space.

(3) A trivial group {0} is an R-module for any ring R. We call it a zero module.

(4) The multiplication map of the ring R endows R with a left and with a right module
structure, which we denote by R R and RR , respectively.

(5) Any A BELian group with endomorphisms applied from the left is a left End(M )-
module and a right End(M )op -module.

(6) A right R-module is the same as a left Rop -module.

2
This is also important for naturality questions.
3
After the American mathematician and logician H ASKELL B ROOKS C URRY.
26 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

(7) If we identify ring elements with 1×1-matrices then the matrix multiplications R1×1 ×
R1×c → R1×c and Rr×1 × R1×1 → Rr×1 define a left R-module structure on the row
space R1×c and a right R-module structure on the column space4 Rr×1 . This general-
izes 4.

2.2 Morphisms
Immediately after defining a mathematical structure and providing some examples one
should define the structure preserving maps:
Definition 2.7. A group homomorphism ϕ : M → N, m 7→ mϕ of two left R-modules
M, N is called an R-module homomorphism or an R-linear map, if it is compatible5 with
the action of R, i.e., (rm)ϕ = r(mϕ) for all r ∈ R, m ∈ M . For right modules we write
ϕ : M → N, m 7→ ϕm and require that ϕ(mr) = (ϕm)r.
The set HomR (M, N ) of all R-linear maps from M to N with pointwise addition is an
A BELian group6 , called the homomorphism group of M and N . We occasionally drop the
index R when it is clear from the context. The zero map 0M N : M → N is the neutral
element of this group.
The A BELian group EndR (M ) := HomR (M, M ) of all R-self-maps with composition is a
ring7 , called the endomorphism ring of M . The identity map 1M is the unit of this ring.
Exercise 2.1. Let M be a left R-module. Show that (EndR (M ), ·) is the subring of the
opposite ring (End(M ), ◦)op which centralizes the action of R, i.e., which commutes with
the image of the left representation of R.
Remark 2.8. Here we use the so-called associative convention for both right and left
modules. It differs from the usual “mixed” convention, where the so-called commutative
convention is used for left modules and the associative for right ones. The mixed conven-
tion is unavoidable if one insists on applying R-linear maps from the left, regardless of the
parity of the modules. There one would write for left R-linear maps ϕ : M → N, m 7→ ϕ(m)
with ϕ(rm) = rϕ(m) and say that ϕ commutes with the action of R.
The associativity convention has at least two advantages:

(1) It is a symmetric convention: EndR (RR ) ∼

=R∼
= EndR (R R).

(2) It is compatible with matrix multiplication and therefore computer friendly:

HomR (R R1×c , R R1×c ) ∼ r ×1 ∼

0 0 0 0
r×1
= Rc×c and HomR (RR , RR ) = Rr ×r

and the associativity rules (rm)ϕ = r(mϕ) and ϕ(mr) = (ϕm)r follow from the asso-
ciativity of matrix multiplication.

The mixed convention breaks the above symmetry: EndR (RR ) ∼ = R but EndR (R R) ∼
= Rop . It
0
also makes working out the relation between applying R-linear maps in HomR (R R1×c , R R1×c )
and matrix multiplication a nontrivial exercise.
Convention. From now on we won’t specify the parity unless necessary. If the formula-
tion of some statement does depend on the parity then we will formulate it for left modules.
4
Like S INGULAR, the subsystem P LURAL only uses column spaces as a data structure for modules but
treats them as left modules (maybe because algorithms in the noncommutative G RÖBNER basis literature are
usually developed for left ideals). This results in a mess when the ring is noncommutative.
5
We deliberately drop the brackets (m)ϕ for the associativity appearance of the law.
6
If R is commutative then HomR (M, N ) is a again an R-module.
7
If R is commutative then EndR (M ) is a again an R-algebra.
 2. THE CATEGORY OF MODULES 27

The case of right modules is then analogous. As a consequence we will silently apply R-
linear maps from the right. In particular, in the composition ϕψ we first apply the left map
ϕ, then ψ.
By replacing definitions which refer to elements by ones which exclusively refer to maps
we will gradually replace the set-theoretic language by the categorical one. The categori-
cal approach, although more cumbersome at the beginning, will help us to reduce many
known constructions to few basic ones. This will be very rewarding when we start dealing
with computability as it will streamline all our algorithms. To emphasize this “element-
less” approach we will replace the words “module” by “object” and “map” by “morphism”.
An R-linear map is called injective, surjective, or bijective, if the underlying set-theoretic
map is so. We will make an attempt to describe these properties using maps (or, better, mor-
phisms) exclusively.
Definition 2.9 ([HS97, §II.3, p. 48]).

(epi) A morphism π : M → N is an epimorphism (or epi, or epic) if it is pre-cancelable,

i.e.,
πϕ = πψ =⇒ ϕ = ψ
for all objects L and all morphisms ϕ, ψ ∈ Hom(N, L). We write π : M  N .

(mono) A morphism ι : M → N is a monomorphism (or mono, or monic) if it is post-

cancelable, i.e.,
ϕι = ψι =⇒ ϕ = ψ
for all objects L and all morphisms ϕ, ψ ∈ Hom(L, M ). We write ι : M ,→ N .

Exercise 2.2. Prove the following statements for an R-linear map ϕ : M → N :

• ϕ is injective iff it is mono.

• ϕ is surjective iff it is epi.

• ϕ is an bijective iff it is both mono and epi.

Although the previous definition is the correct categorical one it may look awkward
at the beginning. Motivated by the underlying set-theoretic maps and using the axiom of
choice one might find the following definition more “natural”.
Definition 2.10.

(1) A morphism π : M → N is a split epi if it has a pre-inverse (or section), i.e., a

σ
morphism M ←− N such that
σπ = 1N .

(2) A morphism ι : M → N is a split mono if it has a post-inverse (or complement),

χ
i.e., a morphism M ←
− N such that

ιχ = 1M .

(3) A morphism α : M → N is an isomorphism (or iso) if it is a split epi and a split mono.
∼ ∼
→ N , or α : M ∼
= N , or M ∼
=
We write α : M −
→ N , or α : M − =α N . We call two objects
∼
M, N isomorphic if there exists an isomorphism α : M − → N.

Here is an easy exercise:

Exercise 2.3. Prove without using elements:
28 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

(1) A split epi is an epi.

(2) A split mono is a mono.

(3) The pre- and post-inverse of an isomorphism coincide, and are hence unique. We call
it the inverse.

(4) The composition of two (split) monos is a (split) mono.

(5) The composition of two (split) epis is a (split) epi.

(6) The composition of isomorphisms is an isomorphism.

(7) If ϕψ is a mono then ϕ is a mono. The converse is false.

(8) If ϕψ is an epi then ψ is an epi. The converse is false.

2
Show that in the sequence of Z-linear maps Z ,→ Z  Z/2Z the left morphism is mono, the
right is epi, and non of them is split (as Z-linear maps). If we view them as maps between
sets then both become split.
As we saw above, an R-linear map which is epi and mono is bijective. It is easy to
see that the inverse is again an R-linear map so we have an iso. Later we will encounter
categories (with modules as objects!) in which being mono and epi does not imply being
iso.

2.3 Subobjects and factor objects

Definition 2.11 (Set-theoretic definition). Let M be an R-module. An R-submodule N
is a subgroup of M which is stable under the action of R. We write N ≤R M . If R is clear
from the context we just say submodule and write N ≤ M . The (set theoretic) embedding
ι : N ,→ M is a monic R-linear map. A submodule N is called proper if {0} = 6 N 6= M . A
module M with no proper submodules is called simple.
Example 2.12. Submodules of R R are the left ideals and of RR the right ideals. A
(skew)field k is a simple k-module.
Example 2.13. Let ϕ : M → N be an R-linear map.

• The set-theoretic image im ϕ is a submodule of N . The corestriction ϕ : M → im ϕ is

an R-epi.

• The group theoretic kernel ker ϕ is a submodule of M . The embedding κ : ker ϕ → M

is an R-mono.

An R-linear map ϕ : M → N is surjective if, by definition, im ϕ = N . From group theory

(applied to the underlying group theoretic map) we know that ϕ is injective iff ker ϕ = {0}.
Definition 2.14. Let N be an R-submodule of M . The action of R on M induces an
action on factor group M/N turning it into an R-module. We call it the factor module of M
modulo N . The projection ϕ : M → M/N, m 7→ m + N is an R-epi.
Example 2.15. Let ϕ : M → N be an R-linear map.

• The cokernel coker ϕ := N/ im ϕ is a factor module of N .

• The co-image coim ϕ := M/ ker ϕ is a factor module of M .

2. THE CATEGORY OF MODULES 29

K κ
Definition 2.16. We say that a morphism κ : K → M dominates τ : L → τ /κ τ M
M if κ is a post-factor of τ , i.e., if there exists a morphism τ /κ : L → K L
such that (τ /κ)κ = τ . Any such morphism τ /κ is called a lift of τ along κ.
It is an easy exercise to prove that mutual dominance is an equivalence
relation and that two R-linear maps which are mutually dominant have the same image
submodule. The converse of the last statement is true for monos. In other words, we can
identify a submodule K ≤ M with the class of all R-monos having K as their image.
This is how we can redefine submodules categorically:
Definition 2.17. A subobject K ≤ M is an equivalence class of monos with M as target
under the equivalence relation of mutual dominance.
End
One can also define the poset of submodules categorically: lect. 9
τ κ
Definition 2.18. For two subobjects L, K ≤ M represented by L ,→ M ←- K we say 01.06
that L is smaller than K and write8 L ≤ K (or K is larger than L and write K ≥ L) if κ
dominates τ . We denote by Sub M the poset of all subobjects of M .
Exercise 2.4. Show that the last definition is independent of the representatives of sub-
objects.
We could now ask if we can redefine the entire lattice of submodules, i.e., the sum
and intersection of two submodules, categorically. The answer is yes, but we first need
a categorical definition of kernels, cokernels, and direct sums, which we will treat in the
following subsections.
Before ending this subsection we dualize the above definition and get:

C ε
Definition 2.19. We say that a morphism ε : N → C codominates η :
N → L if ε is a pre-factor of η, i.e., if there exists a morphism ε\η : C → L ε\η η N
such that ε(ε\η) = η. Any such morphism ε\η is called a colift of η along ε. L
Dually, this leads to a categorical formulation of a factor object.
Remark 2.20. Lifts along monos and colifts along epis are unique.

Proof. This is immediate from the definitions of mono and epi.

τ κ
Exercise 2.5. Let L ≤ K ≤ M be two subobjects of M represented by L ,→ M ←- K.
τ /κ
Show that the unique lift L ,→ K is a mono realizing L as a subobject of K, justifying the
subobject notation L ≤ K. In particular, Sub K can be viewed as subset of Sub M .
Hint: Exercise 2.3.(7).

2.4 Kernels and cokernels, images and co-images

We now formulate so-called universal properties of the previous set-theoretic construc-
tions, which will server as categorical characterizations. I will leave it as a general exercise
to prove that the categorical definitions coincide with the module theoretic ones.
Definition 2.21 ([HS97, §II.6, p. 61]). Let ϕ : M → N be a morphism.

(ker) “The” kernel of ϕ is an object K, usually denoted by ker ϕ, together with a morphism
κ : K → M satisfying the following universal property:
8
Exercise 2.5 will justify notation which was reserved for subobjects.
 30 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

(i) κϕ = 0, and
(ii) κ dominates all such morphisms, i.e., for all objects L and all morphisms τ :
L → M with τ ϕ = 0 there exists a unique lift τ /κ : L → K (of τ along κ with
τ = (τ /κ)κ).

It follows from the uniqueness of the lift τ /κ that κ is a mono9 . We will usually call it
the kernel mono.

0
K κ
ϕ
τ /κ τ M N
L
0

K is called “the” kernel object of ϕ. Depending on the context ker ϕ sometimes stands
for the morphism κ and sometimes for the object K.

(coker) “The” cokernel of ϕ is an object C, usually denoted by coker ϕ, together with a mor-
phism ε : M → C satisfying the following universal property:

(i) ϕε = 0, and
(ii) ε codominates all such morphism, i.e., for all objects L and all morphisms η :
N → L with ϕη = 0 there exists a unique colift ε\η : C → L (of η along ε with
η = ε(ε\η)).

It follows from the uniqueness of the colift ε\η that ε is an epi. We will usually call it
the cokernel epi.

0
C ε
ϕ
ε\η η N M
L
0

C is called “the” cokernel object of ϕ. Depending on the context coker ϕ sometimes

stands for the morphism ε and sometimes for the object C.

Remark 2.20 implies that instead of insisting on the uniqueness of the lift and the colift
in the definition of kernels and cokernels we could have equivalently required that κ is
mono and ε is epi.
Luckily, this is already enough to define images and coimages without referring to ele-
ments.
Definition 2.22. Let ϕ : M → N be a morphism.

(im) The image of ϕ, denoted by im ϕ, is the kernel of its cokernel epi.

(coim) The co-image of ϕ, denoted by coim ϕ, is the cokernel of its kernel mono.
9
Since (ακ)ϕ = α(κϕ) = 0 we can reconstruct α as the unique lift of ακ along κ.
2. THE CATEGORY OF MODULES 31

0 0
ker ϕ coker ϕ
κ ε
ϕ
M N
0 0
εκ κε

coim ϕ im ϕ

We denote the cokernel epi of the kernel mono κ by εκ and call it the co-image epi. Dually,
we denote the kernel mono of the cokernel epi ε by κε and call it the image mono.
The homomorphism theorem for R-modules states that each R-linear map ϕ induces
an R-linear map ϕ : coim ϕ → im ϕ (between its coimage and its image) which is an iso-
morphism. Now we construct this induced morphism categorically by using the above
diagrams:

0 0
ker ϕ coker ϕ
κ ε
ϕ
M N
0 εκ \ϕ 0
εκ κε

coim ϕ im ϕ

0
ker ϕ coker ϕ
κ ε
ϕ
M N
εκ \ϕ 0
εκ κε

coim ϕ im ϕ
ϕ := (εκ \ϕ)/κε

Do we get the same morphism ϕ if we reverse the process, i.e., first lift then colift: εκ \(ϕ/κε )?
The answer is yes since for both of them pre-composing the epi εκ and post-composing the
mono κε yields ϕ. Hence, both are equal.
Summing up, in categories in which cokernels and kernels exist we have categorical
algorithms to compute coimages and images together with an induced morphism ϕ :
coim ϕ → im ϕ. This morphism is an isomorphism is the category of modules but not an
isomorphism in general categories with kernels and cokernels. However, being an isomor-
phism is one of the defining axioms of so-called A BELian categories. Hence, in A BELian
categories the homomorphism theorem, also called the first isomorphism theorem, will
hold true.
In order to talk about the second isomorphism theorem we still need to be able to per-
form intersections and sums of subobjects categorically. The last missing ingredient are
direct sums (more precisely, products and coproduct).
 32 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

2.5 Direct sums and direct products

We start by recalling the set theoretic definition.
Definition 2.23 ([HS97, §II.5, p. 58, p. 54]). Let {Ai }i∈I be a family of R-modules indexed
by I.

(prod) The set

Y n [ o
Ai := a : I → Ai | ai := a(i) ∈ Ai for all i ∈ I
i∈I

is an R-module,
Q called the (direct) product of {Ai }i∈I . For each i ∈ I we obtain an
R-epi πi : Aj → Ai , a 7→ ai , called the i-th projection.

(sum) The set ( )

M Y
Ai := a∈ Ai | ai = 0 for all but finitely many i ∈ I
i∈I i∈I

is an R-module, called the direct sum or coproduct

( of {A
! i }i∈I . For each i ∈ I, we ob-
L ai j = i
tain an R-mono ιi : Ai ,→ Aj , ai 7→ j 7→ , called the i-th embedding.
0 j 6= i

For a finite index set I it is clear that products and coproducts of modules coincide
(this is false in general categories but still true in A BELian ones). We refer to them as “direct
sum” and write M ⊕N . Although we will only deal with finite index sets giving the general
definition was not more complicated. Now we give the categorical characterization:
Definition 2.24 ([HS97, §II.5, p. 58, p. 54]). Let I be an index set and {Ai }i∈I a family of
objects.
Q
(prod) “The”Qproduct of {Ai }i∈I is an object i∈I Ai , together with a family of morphisms
{πi : Aj  Ai }i∈I , called projections10 , such that the following universal property
is satisfied:
For any object M and any family {ϕi : M → Ai }i∈I of morphisms there exists a
unique morphism
Y
ϕ = {ϕi } : M → Ai ,

called the product morphism or pairing, such that ϕπi = ϕi for all i ∈ I.
L
(sum) “The” coproduct of {A i }i∈I is an object i∈I Ai , together with a family of morphisms
Aj }i∈I , called embeddings11 , such that the following universal property
L
{ιi : Ai ,→
is satisfied:
For any object M and any family {ϕi : Ai → M }i∈I of morphisms there exists a
unique morphism
M
ϕ = hϕi i : Ai → M,

called the coproduct morphism or copairing, such that ιi ϕ = ϕi for all i ∈ I.

The universal properties imply that the product (resp. coproduct), in case it exists, is
unique up to a unique isomorphism.
Remark 2.25.
10
The projections are not necessarily epis unless the category has a zero object [Bru].
11
The embeddings are not necessarily monos unless the category has a zero object [Bru].
 2. THE CATEGORY OF MODULES 33

(prod) The product morphism for a family {ϕi : M → Ai }i∈I of R-linear maps is given by
Y
ϕ = {ϕi } : M → Ai , m 7→ (i 7→ mϕi ) .

(sum) The coproduct morphism for a family {ϕi : Ai → M }i∈I of R-linear maps is given by
M X
ϕ = hϕi i : Ai → M, a 7→ ai ϕ i .
i∈I

2.6 Pull-backs and push-outs as kernels and cokernels

In this subsection we start with the categorical definition and reduce it to the computation
of kernels of coproduct morphisms and cokernels of product morphisms. This also gives a
module theoretic definition.
Definition 2.26.
ϕ ψ
(pull) A pull-back of two morphisms L − → M ← − J is an object X together with a pair of
λ 
morphisms L ← − X →− J, called the pull-back morphisms, completing the square
(i.e., such that λϕ = ψ) and satisfying the following universal property:
ρ η
For any object Z and any pair of morphisms L ← −Z→ − J completing the square there
exists a unique ζ : Z → X with ρ = ζλ and η = ζ.

Z ρ
ζ
λ
(pull-back diagram) X L
η
 ϕ

J M
ψ

(push) Reversing all arrows in the above definition we obtain the dual definition of the push-
ϕ ψ
out X of two morphisms L ←
− M −
→ J, together with two push-out morphisms
λ 
L→− X← − J.

Z ρ
ζ
λ
(push-out diagram) X L
η
 ϕ

J M
ψ

Here we only considered finite pull-backs and push-outs. Next we show how a finite
pull-back can be reconstructed as a kernel of a coproduct morphism and, dually, a finite
push-out as a cokernel of a product morphism. We will make use of the fact that in our
context finite products and coproducts coincide.
Proposition 2.27. The following two dual statements hold.
 34 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

ϕ ψ {λ,}
(pull) The pull-back of two morphisms L −
→M ←
− J is the kernel X ,→ L ⊕ J of the coproduct
hϕ,−ψi
morphism L ⊕ J −−−−→ M .
ϕ ψ hλ,i
(push) The push-out of two morphisms L ←
−M −
→ J is the cokernel X  L ⊕ J of the product
{ϕ,−ψ}
morphism L ⊕ J ←−−−− M .
Proof.
λ 
(pull) One can recover the pull-back morphisms L ←−X→ − J by post-composing the kernel
πL πJ
mono {λ, } with the projections L ←− L ⊕ J −→ J. This justifies the name {λ, }.
The unique morphism ζ in the pull-back diagram is the lift of the product morphism
{ρ, η} along the kernel mono {λ, }.
λ 
(push) Dually, one can recover the push-out morphisms L → − X ← − J by pre-composing
ιL ιJ
the cokernel epi hλ, i with the embeddings L −→ L⊕J ← − J. This justifies the name
hλ, i. The unique morphism ζ in the push-out diagram is the colift of of the coproduct
morphism hρ, ηi along the cokernel epi hλ, i.
Now we are finally able to define the intersection of two submodules categorically. I
leave the sum as the dual exercise.
ιL ιJ
Definition 2.28. The intersection of two subobjects L ,→ M ←- J is their pull-back.
Remark 2.29. The module (or set) theoretic definition of intersection coincides with the
above categorical one.
Proof. The proof uses the reconstruction of the pull-back as the kernel of the coproduct
morphism. Indeed:

kerhιL , −ιJ i = {(m, n) ∈ L ⊕ J | 0 = ιL (m) − ιJ (n) = m − n} ∼

= L ∩ J.

Before we see how to describe a constructive setup for all this and more it is time to
End define A BELian categories.
lect. 10
08.06
3 More categorical notions
Like with monoids, to each category A one can define the so-called opposite category Aop
with Aop op
0 := A0 , A1 := A1 , 1Aop = 1A , sAop = tA and tAop = sA (i.e., HomAop (M, N ) :=
HomA (N, M )) and composition µAop (ϕ, ψ) := µA (ψ, ϕ). Clearly, double dualizing recovers
A = (Aop )op .
Introducing the opposite category saves us half of our definitions:
• A (split) epi in A is (split) mono in Aop .

• A colift in A is a lift in Aop .

• A cokernel epi in A is a kernel mono in Aop .

• A coproduct in A is a product in Aop .

• A push-out in A is a pull-back in Aop .

Definition 2.30. A subcategory A0 of a category A consists of subclasses A00 ⊂ A0 and
A01 ⊂ A1 to which all defining maps 1, s, t, µ restrict. It is called full if HomA0 (M, N ) =
HomA (M, N ) for all M, N ∈ A0 .
4. ADDITIVE AND ABELIAN CATEGORIES 35

A subcategory is again a category.

Example 2.31.

(1) (Ab) is a full subcategory of (Grps).

(2) (uRngs) is a full subcategory of (Rngs).

(3) (URngs) is a non-full subcategory of (uRngs).

(4) ...

Definitions 2.9 and 2.10 of (split) epi, (split) mono, and isomorphism were already cate-
gorical.
Exercise 2.6. The set theoretic embedding Z ,→ Q is a mono and epi in (URngs), but
obviously not an isomorphism.
Definition 2.32 ([HS97, §II.1, p. 43]). Let A be a category.

• A terminal object T ∈ A is an object such that HomA (M, T ) is a singleton, i.e., consists
of a single morphism, for all M ∈ A.

• An initial object I ∈ A is an object such that HomA (I, M ) is a singleton, i.e., consists
of a single morphism, for all M ∈ A.

Both terminologies are dual: T is terminal A iff T is initial in Aop . Trivially, initial objects
and terminal objects are unique up to unique isomorphisms in A. A terminal object is the same
as an empty product and an initial the same as an empty coproduct.
Definition 2.33. A zero object12 0 ∈ A is an object which is both initial and terminal13 .
A category with zero is a category that has a zero object.
Define for each pair of objects M, N the unique zero morphism14

0M N := M → 0 → N.

We write 0 instead of 0M N when no confusion is possible.

Example 2.34.

(1) A singleton in (Sets) is terminal (and not initial) while the empty set is initial (and not
terminal).

(2) Z is an initial object in (URngs) which is not terminal. The zero ring is a terminal
object in (URngs) which is not initial.

(3) The trivial group is a zero object in (Grps) and (Ab). Both are categories with zero.

4 Additive and A BELian categories

Definition 2.35. A category A is called pre-additive if it is enriched over (Ab), i.e., if
for all objects M, N, L ∈ A
12
also called empty biproduct.
13
This is our first self-dual notion!
14
Hence, a category with a zero object is a category enriched over the category of pointed sets.
 36 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

(pAdd1) the set HomA (M, N ) is an (additively written) A BELian group, in particular, there
exists an addition (and a subtraction) operation turning HomA (M, N ) into an A BELian
group;

(pAdd2) the composition HomA (M, N ) × HomA (N, L) → HomA (M, L) is bilinear.

Remark 2.36. Let A be a pre-additive category.

(1) The definition of pre-additive is self-dual, i.e., the opposite category Aop is again pre-
additive.

(2) If A has a zero object then the unique zero morphism 0M N defined above is exactly
the zero element of the A BELian group HomA (M, N ), so there is no ambiguity talking
about the zero morphism from M to N .

(3) The A BELian group EndA (M ) is a unital ring for all M ∈ A. Conversely, a unital ring
is a pre-additive category with one object. Another name for a pre-additive category
would therefore be a “ringoid”. This is our third and last example of an oidification.

Definition 2.23 of product and coproduct was already categorical.

Definition 2.37 ([HS97, §II.9, p. 75]). A pre-additive category with zero is called additive
if

(Add) The product15 (M ⊕N ; πM , πN ) of two arbitrary objects M and N exists .

Remark 2.38. The following is true for additive categories (cf. [HS97, §II.9, p. 75ff]):

• For objects M and N define the morphisms

ιM := {1M , 0M N } : M → M ⊕ N and ιN := {0N M , 1N } : N → M ⊕ N.

Then πM ιM + πN ιN = 1M ⊕N .

• It follows that finite coproducts also exist: (M ⊕ N ; ιM , ιN ) with ιM , ιN as above and the
coproduct morphism defined by

hϕ, ψi := πM ϕ + πN ψ : M ⊕ N → L,

for two morphisms ϕ : M → L and ψ : N → L.

• In particular, finite products and coproducts “coincide”. Such objects are called biprod-
ucts or direct sums.

We have just proved that the definition of an additive category is self-dual, i.e., the opposite
Aop of an additive category A is again additive.
{α,β} hϕ,ψi
• For K −−−→ L ⊕ M −−−→ N we have {α, β}hϕ, ψi = αϕ + βψ.

• In particular, for ϕ, ψ : M → N we have ϕ + ψ = {1M , 1M }hϕ, ψi.

This means that if a category is additive, then its (in general non-unique) pre-additive struc-
ture is already determined by its structure as a category. In other words, there is then ex-
actly one possible definition of the A BELian group structure on the Hom-sets. We say the
addition for an additive category is internal to that category.
15
The use of the direct sum symbol ⊕ will be justified below.
 4. ADDITIVE AND ABELIAN CATEGORIES 37

Definition 2.21 of kernel and cokernel is valid for any category with zero. Again, the
following definition is self-dual.
Definition 2.39. An additive category is called pre-A BELian if every morphism

(pAb1) has a kernel;

(pAb2) has a cokernel.

Exercise 2.7. Let A be a pre-A BELian category. Prove that for an A-morphism ϕ:

• ϕ is mono iff ker ϕ = 0;

• ϕ is epi iff coker ϕ = 0.

Our definition of coimage and image and the construction of the induced morphism
ϕ
coim ϕ −
→ im ϕ in §2.4 are valid in any category with kernels and cokernels. Using this we
can finally state the definition of an A BELian category:
Definition 2.40 ([Rot09, §5.5, p. 307]). A pre-A BELian category is called A BELian if the
homomorphism theorem is valid, i.e., if for each morphism ϕ ∈ A
ϕ
(Ab) the induced morphism coim ϕ −
→ im ϕ is an isomorphism.

Note that this last axiom involves a single existential quantifier, namely the existence of
an inverse. This is the first reason why we prefer it over other equivalent formulations.
The second reason is original axiom G ROTHENDIECK’s gave in his seminal T ÔHOKU paper
[Gro57]. The third reason is that it emphasizes the homomorphism theorem, which is in my
opinion among the reasons behind various “tunnel effects” connecting apparently different
subfields of mathematics.
Again, these definitions are self-dual, i.e., the opposite Aop of a (pre-)A BELian category
A is again (pre-)A BELian.

ker ϕ coker ϕ
κ ε
ϕ
M N

(Diagϕ ) εκ εκ \ϕ ϕ/κε κε

coim ϕ ∼ im ϕ
εκ \ϕ/κε
| {z }
ϕ:=

From the identities εκ \ϕ = ϕκε and ϕ/κε = εκ ϕ we conclude that the colift εκ \ϕ is a
mono (as the composition of an iso and a mono) and that lift ϕ/κε is an epi (as the compo-
sition of an epi and an iso).
Definition 2.41. We call the colift εκ \ϕ the co-image mono16 and the lift ϕ/κε the image
epi.
Exercise 2.8. Show that the above diagram is functorial, i.e., a commutative diagram of
the form
16
It the categorical version of the co-restriction of ϕ to its image.
38 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

ϕ
M N
σ τ
ϕ0
M0 N0

can be extended to a commutative diagram between (Diagϕ ) and (Diagϕ0 ).

We summarize this in the following proposition.
Corollary 2.42 ([HS97, Prop. II.9.6],[ML95, Thm. IX.2.1]). For each morphisms ϕ : M → N
in an A BELian category there is a so-called kernel-cokernel sequence (Sϕ )

K κ ε C
ϕ
(Sϕ ) M N
π ι
I

where
• πι = ϕ, i.e., ϕ is the composition of an epi and a mono,

• κ is the kernel mono of ϕ and π,

• ε the cokernel of ϕ and ι,

• π is the cokernel of κ, and

• ι is the kernel of ε.
Proof. Set I := im ϕ, the image of ϕ, and define π and ι as the image epi and mono, respec-
tively17 .
Corollary 2.43. Let A be an A BELian category. Then
• every mono is a regular mono, i.e., is a kernel mono;

• every epi is a regular epi, i.e., is a cokernel epi.

In pre-Abelian categories both statements are equivalent to the two more sharper statements:
• every mono is the kernel mono of its cokernel epi;

• every epi is the cokernel epi of its kernel mono.

In fact, both are equivalent to the defining axiom (Ab).
• every morphism which is both mono and epi is an isomorphism.
Example 2.44.
(1) (Grps) and (Rngs) are not even pre-additive.

(2) (Ab) is an A BELian category.

(3) More generally, the category R-Mod of left R-modules is an A BELian category. Like-
wise for the Mod-R, the category of right R-modules.
17
Alternatively, set I := coim ϕ and define π and ι as the co-image epi and mono, respectively. This is
equivalent since coim ϕ and im ϕ are isomorphic with the explicit isomorphism ϕ.
5. COMPUTABLE RINGS 39

(4) The full subcategory R-mod ⊂ R-Mod of finitely generated (f.g.) left R-modules over
a left N OETHERian ring R is an A BELian category. Likewise for mod-R ⊂ Mod-R, the
category of f.g. right R-modules over a right N OETHERian ring R.

(5) The dual of an A BELian category is again A BELian. An interesting discussion about
the opposite category of a module category can be found here [Ste].

Exercise 2.9. Show that the full subcategory (tfAb) ⊂ (Ab) of torsion-free A BELian
2
groups is pre-A BELian but not A BELian. For this consider the map Z ,→ Z and prove
that it is mono and epi (!) but not an isomorphism in (tfAb).
End
lect. 11
13.06
5 Computable rings
We could’ve started developing advanced abstract algorithms in A BELian categories but
it makes more sense to first describe a basic class of categories of finitely presented mod-
ules, which are not only A BELian but constructively A BELian [BLH11]. Once we have such a
constructively A BELian category A, i.e., once we have for A specified algorithms for all the
existential quantifiers and disjunctions indicated by boxes in the previous sections, then
all abstract algorithms become algorithms in the usual sense. In fact, the boxes describe all
the needed interfaces between the abstract and the concrete part of a well-structured com-
puter implementation. The above mentioned class of finitely presented module categories
has already a wide range of applications. And by modeling any other A BELian category A
(e.g., of coherent sheaves on certain varieties) on a constructively A BELian one (e.g., that of
finitely presented modules over a so-called computable ring), we get a proof that A itself is
constructively A BELian. Furthermore, all abstract algorithms which we will develop later
for A BELian categories will be automatically applicable for A.
In order to describe this basic class of categories of finitely presented modules which are
constructively A BELian we first need to specify the underlying rings. Recall that all rings
we consider are unital. Let R be such a ring. All matrices over R will have finite number of
rows and columns. Empty matrices with no rows or no columns are explicitly allowed.
Definition 2.45.

• A constructive set S consists of two algorithms, one to decide membership in S (for any
given input) and another to decide equality of elements in S.

• A constructive A BELian group H (additively written) is a constructive set together

with an explicit element 0 ∈ H (i.e., an algorithm constructing 0) and algorithms for
addition and subtraction satisfying the usual axioms.

• A constructive ring R is a constructive A BELian group together with an explicit 1 ∈ R

and an algorithm for multiplication satisfying the usual axioms.

Remark 2.46.

• We did not require to have an algorithm which decides if a constructive set is empty
or not.

• Likewise, we did not require to have an algorithm which decides if a constructive

A BELian group is trivial or not.

• Since we can decide 0 = 1 in a constructive ring R we can decide whether or not R is

the zero ring.
 40 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

We now introduce a terminology analogous to a one we used for morphisms.

Definition 2.47. An r × c matrix A over R is said to
0 0
(row) row-dominate B ∈ Rr ×c if there exists a matrix X ∈ Rr ×r such that B = XA (we write
A ≥row B);
0 0
(col) column-dominate B ∈ Rr×c if there exists a matrix X ∈ Rc×c such that B = AX (we
write A ≥col B).

Exercise 2.10. Show that A =row A0 : ⇐⇒ A ≥row A0 and A0 ≥row A is an equivalence

relation18 . The analogous statement holds for =col and ≥col .
The syzygy matrices will be the essential ingredient of kernels and their kernel monos.
Definition 2.48. Let A ∈ Rr×c be a matrix over R.

(row) A matrix S ∈ Ra×r is called a matrix of row syzygies of A if

– SA = 0;
0
– S row-dominates any other matrix S0 ∈ Ra ×r with S0 A = 0.

(col) A matrix S ∈ Rc×a is called a matrix of column syzygies of A if

– AS = 0;
0
– S column-dominates any other matrix S0 ∈ Rc×a with AS0 = 0.

Definition 2.49. A left computable ring is a constructive ring R together with algo-
rithms for solving left sided inhomogeneous R-linear system of equations B = XA (for given
matrices A and B over R with equal number of columns), i.e., with three algorithms

• DecideZeroRows(B, A) which decides if A ≥row B, i.e., if there exists a solution ma-

trix X with XA = B: the algorithm returns a matrix B0 (having the same shape as B)
for which the equation XA = B − B0 is solvable and where the i-th row B0i is zero iff
the equation xA = Bi is solvable.19 In particular, the equation XA = B is solvable iff
DecideZeroRows(B, A) = 0;

• DecideZeroRowsEffectively(B, A) computes a matrix T satisfying B + TA = B0 , where

B0 = DecideZeroRows(B, A). In particular, if the equation XA = B is solvable then

X := −T =: RightDivide(B, A);

• SyzygiesOfRows(A) which computes a matrix S of row syzygies of A.

Analogously define a right computable ring with its three algorithms for solving right
sided equations AX = B

• DecideZeroColumns(B, A);

• DecideZeroColumnsEffectively(B, A) and in the affirmative case

X := LeftDivide(A, B);

• SyzygiesOfColumns(A).
18
Both matrices generate the same row-space
19
So we do not require a “normal form”, but only a mechanism to decide if a row is zero modulo some
relations.
5. COMPUTABLE RINGS 41

Finally, we call a ring computable if it is both left and right computable.

Exercise 2.11. An involution on a ring R is an anti-isomorphism θ : R → R with
2
θ = idR , i.e., θ is an isomorphism of the underlying A BELian group (R, +) and θ(1) = 1,
θ(θ(a)) = a, and θ(ab) = θ(b)θ(a) for all a, b ∈ R. Construct an involution on the following
rings.
(1) R is a commutative ring.
(2) R is the W EYL algebra k[x1 , . . . , xn ]h∂1 , . . . , ∂n i.
Exercise 2.12. The following are equivalent for any ring R with an involution:
• R is left computable.
• R is right computable.
• R is computable.
Exercise 2.13. Let R be a commutative computable ring and I = ha1 , . . . , an i  R an
(explicitly) finitely generated ideal of R. Prove that the residue class ring R/I is again
computable. In particular, all affine rings are computable.
Remark 2.50. The following rings are (left) computable:
(1) A constructive field k with the G AUSSian normal form algorithm, more precisely, an
algorithm to compute the row reduced echelon form (RREF).
(2) A E UCLIDian ring R with a H ERMITE normal form algorithm, e.g., R = Z or R = k[x],
where k is a constructive field.
(3) A polynomial ring R = k[x1 , . . . , xn ] over a computable field k equipped with an
algorithm to compute reduced G RÖBNER bases (of rows of rectangular matrices).
(4) In fact any ring R with a G RÖBNER basis notion and equipped with an algorithm
to compute reduced G RÖBNER bases (of rows of rectangular matrices), e.g., R =
k[x1 , . . . , xn ]h∂1 , . . . , ∂n i.
 1 B0 −X 
Proof. Apply such an algorithm to the block matrix20 ( 10 BA 01 ) and obtain 0 A0 Y . The last
 1 −X   1 B0 −X  0 0 S
1 B 0
matrix is obviously the product 0 Y ( 0 A 1 ) = 0 A0 Y ; the simple form of the first
  0 S 0 0 S
1
column 0 of the transformation matrix asserts that the normal form algorithm does not
0
auto-transform the first row block ( 1 B 0 ) of the input matrix as its left most block is an
identity matrix 1. Note that B0 = B − XA and the left sided equation XA = B is solvable
(with solution X) iff B0 vanishes, yielding the algorithms B0 = DecideZeroRows(B, A) and
(B0 , −X) = DecideZeroRowsEffectively(B, A). The matrix S is a matrix of row syzygies,
yielding the algorithm S = SyzygiesOfRows(A).
We write A0 = YA = BasisOfRows(A) and (A0 , Y) = BasisOfRowsCoeff(A).
Remark 2.51. It is important to emphasize that we do need DecideZeroRows to compute
any kind of “normal form”, unless the normal form is zero. This is reflected in our choice
of the name DecideZeroRows.
Remark 2.52. In fact, the polynomial ring k[x1 , x2 , . . .] in infinitely many indeterminates
is computable! So computability does not imply N OETHERianity (left or right) and the
former is in our context the correct substitute for the latter.
20
In practice one would avoid the construction of such big partially sparse partially dense block matrices
and writes specific variants to construct X, Y and S directly form A and B.
 42 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

6 A constructively A BELian category of matrices

Now we define a category of matrices which will serve as a “model” for the category of
finitely presented modules. This category will turn out to be constructively A BELian when
R ist left (resp. right) computable.
Definition 2.53. Let R be a ring. Define the category R-fpres of (finite) left presenta-
tions over R as follows:

(Obj) (R-fpres)0 := the set of all matrices M over R (including the empty ones). We call the
number of columns of a matrix M the number of generators21 and denote it by gM ;

(Mor) (R-fpres)1 := the set of left compatible triples (M, A, N) over R modulo a certain equiv-
alence relation ∼, where a triple of matrices (M, A, N) over R is called left compatible
if

– A is a gM × gN matrix22 ;
– N ≥row MA, i.e., the left sided equation XA N = MA is solvable 23 for some matrix XA .

The equivalence relation ∼ is defined as follows: (M, A, N) ∼ (M0 , A0 , N0 ) if

– M = M0 ;
– N = N0 ;
– N ≥row A − A0 , i.e., there exists 24 a matrix Y = YA,A0 such that25 A − A0 = YN.

The identity, source, target, and composition are defined as follows:

(s) s((M, A, N)) := Source((M, A, N)) := M;

(t) t((M, A, N)) := Target((M, A, N)) := N;

(1) 1M := IdentityMorphism(M) := (M, 1, M), where 1 is the gM × gM identity matrix over R.

(µ) µ((M, A, N), (N, B, L)) = Compose((M, A, N), (N, B, L)) = (M, A, N)(N, B, L) = (M, AB, L), where
AB is the matrix multiplication.
A
We occasionally denote a morphism (M, A, N) by M →− N.
The category fpres-R of (finite) right presentations over R is defined analogously. Ev-
erything we do below also holds for fpres-R.
Indeed the composed tripple (M, AB, L) is again compatible: There exists an XA such that
MA = XA N and XB such that NB = XB L and therefore M(AB) = (MA)B = (XA N)B = XA (NB) =
XA (XB L) = (XA XB )L. Furthermore, the composition is well-defined for classes: if (M, A, N) ∼
(M, A0 , N) via YA,A0 and (N, B, L) ∼ (N, B0 , L) via YB,B0 then AB−A0 B0 = (A0 +YA,A0 N)(B0 +YB,B0 L)−A0 B0 =
NB0 +A0 YB,B0 L + YA,A0 NYB,B0 L = (...)L. The remaining defining properties are all obvious.
YA,A0 |{z}
=XB0 L
When R is left (resp. right) computable then R-fpres (resp. fpres-R) will turn out to be a
constructive category in the following sense (see Proposition 2.56):
Definition 2.54. We call a category A constructive if A0 , A1 are constructive sets and all
four structure maps s, t, 1, µ are realized by algorithms.
21
We will justify this name in the next chapter.
22
in particular, the matrix multiplication MA is defined.
23
This is part of the membership decision algorithm in (R-fpres)1 .
24
This is part of the membership equality algorithm in (R-fpres)1 .
25
We say that the matrix A is unique up to a multiple of N.
6. A CONSTRUCTIVELY ABELIAN CATEGORY OF MATRICES 43

Remark 2.55. More generally, we call a mathematical structure constructive if all ex-
istential quantifiers and all disjunctions appearing in its defining axioms are realized by
algorithms. All the above constructivity notions were in this sense special cases. However,
as this notion is usually not treated in classical math curricula we decided to define some
special cases explicitly to underline the various and maybe unexpected places where such
quantifiers occur.
Note that we do not require to have an algorithm which decides if two different objects
are isomorphic or not. However, in a constructively A BELian category A the discussion
below will provide an algorithm capable of deciding whether a given A-morphism α : M →
N is an isomorphism or not.
End
Proposition 2.56. If R is left computable then R-fpres is a constructive category. lect. 12
20.06
Proof.

• The membership and equality algorithms for (R-fpres)0 are derived from those of R.

• For the membership and equality algorithms for (R-fpres)1 we use the left computabil-
ity of R.

• The algorithms for Source and Target are obvious.

• For IdentityMorphism we use a matrix constructor IdentityMat for (possibly 0 × 0)

identity matrices.

• For Compose we use an algorithm MulMat for matrix multiplication.

Both algorithms IdentityMat and MulMat rely on the constructivity of R.

Our next goal is to prove that R-fpres is in fact constructively A BELian. For the rest of
this section let R be a left computable ring.
We start with the pre-additive structure:
Definition 2.57. Define the constructive pre-additive structure on R-fpres by the follow-
ing matrix algorithms:

(+) (M, A, N) + (M, B, N) := (M, A + B, N);

(−) (M, A, N) − (M, B, N) := (M, A − B, N);

(0) The zero morphism is given by ZeroMorphism(M, N) := (M, 0, N).

We call the corresponding matrix algorithms AddMat, SubMat, and a constructor ZeroMat
for (possibly empty) zero matrices, respectively. All of them rely on R merely being a
constructive ring.
This constructive pre-additive structure is, as expected, uniquely determined and hence
reproducible from the constructive additivity of the category discussed below.
Remark 2.58. If R is left computable then R-fpres is a constructive category with zero.
Proof. The 0 × 0 matrix 0 over R is a zero object in R-fpres. Indeed, there is a unique mor-
phism (M, 0, 0) and (0, 0, N) with middle matrices being of the shape ? × 0 and 0×?, re-
spectively. All involved empty matrices can be constructed using the matrix constructor
ZeroMat.
As expected, the zero morphism 0MN := M → 0 → N = (M, 0, 0)(0, 0, N) coincides with the
zero morphism (M, 0, N), where the zero matrix is the product of the two empty matrices.
44 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

Proposition 2.59. If R is left computable then R-fpres is constructively additive.

Proof. Let M, N be two objects in R-fpres. Consider the object

M ⊕ N := ProductObject(M, N) := DiagMat(M, N)26 := ( M0 0N )

together with the two morphisms (given by the two compatible triples)

πM := ProjectionOnLeftFactor(M, N) := (M ⊕ N, ( 10 ) , M) ,
πN := ProjectionOnRightFactor(M, N) := (M ⊕ N, ( 01 ) , N)

as a candidate for the product of M and N and its two projections. This relies on the matrix
constructor algorithms DiagMat and StackMat to stack matrices. Indeed for two arbitrary
A B
morphisms ϕ : L → − M and ψ : L →− N there is exactly one morphism (i.e., one equivalence
class of compatible triples)

{ϕ, ψ} := ProductMorphism(ϕ, ψ) := (L, ( A B ) , M ⊕ N),

such that {ϕ, ψ}πM = ϕ and {ϕ, ψ}πN = ψ, as can be verified by block matrix multiplication.
For this we use a constructor AugmentMat to augment matrices.
Remark 2.60. Constructing the embeddings in the biproduct and the coproduct mor-
phism now follows from Remark 2.38. The following algorithms are in this sense derived
algorithms:

• ιM := EmbeddingOfLeftCofactor(M, N) := {1M , 0MN } = (M, ( 1 0 ) , M ⊕ N).

• ιN := EmbeddingOfRightCofactor(M, N) := {0NM , 1N } = (N, ( 0,1 ) , M ⊕ N).

A B
• For two morphisms ϕ : M →
− L and ψ : N →
− L

hϕ, ψi := CoproductMorphism(ϕ, ψ) := πM ϕ + πN ψ = (M ⊕ N, ( AB ) , L).

A+B A
• As mentioned above, the sum ϕ + ψ : M −−→ N of two parallel morphisms ϕ : M →
− N,
B
ψ:M→− N can be reconstructed as

ϕ + ψ := Add(ϕ, ψ) := {1M , 1M }hϕ, ψi = (M, ( 1 1 ) , M ⊕ M)(M ⊕ M, ( AB ) , N).

Before we discuss the pre-A BELian structure we need a relative version of the matrix
algorithm SyzygiesOfRows, which we can deduce as a derived algorithm.
Definition 2.61. Let A and N be two stackable matrices over R. We say that K is a matrix
of row syzygies of A modulo N if

• N ≥row KA;

• K row-dominates any such matrix T (i.e., any T where N ≥row TA).

For the stacked matrix ( AN ) we write SyzygiesOfRows(( AN )) = ( K L ) with KA + LN = 0 and

define27
SyzygiesOfRows(A, N) := K,
for which we need a matrix algorithm CertainColumns to extract K.
26
Of course, DiagMat can be derived from the constructors ZeroMat, StackMat, and AugmentMat.
27
Again, one can derive more efficient algorithms to compute the relative version of SyzygiesOfRows.
6. A CONSTRUCTIVELY ABELIAN CATEGORY OF MATRICES 45

Proposition 2.62. If R is left computable then R-fpres is constructively pre-A BELian.

A
Proof. Let ϕ : M →
− N be a morphism in R-fpres. We start with the existence of cokernels.
Consider the object
coker ϕ := Cokernel(ϕ) := ( AN )
together with the morphism
ε := CokernelEpi(ϕ) := (N, 1, coker ϕ)
as a candidate for the cokernel of ϕ and its cokernel epi.
0
( AN ) 1
A
ε\η := H H N M
L
0

Indeed, ϕε = 0 (since ( AN ) tautologically row-dominates A · 1 = A) and for any morphism

H
η:N→ − L with ϕη = 0 (i.e., L ≥row AH) there is exactly one morphism (i.e., one equivalence
class of compatible triples)
ε\η := CokernelColift(η, ε) := (coker ϕ, H, L),
such that ε(ε\η) = η.
To establish the existence of kernels we first define the matrix
K := SyzygiesOfRows(A, N).
Now consider the object
ker ϕ := Kernel(ϕ) := SyzygiesOfRows(K, M)
together with the morphism
κ := KernelMono(ϕ) := (ker ϕ, K, M)
as a candidate for the kernel of ϕ and its kernel mono.
0
ker ϕ K
A
τ /κ := X T M N
L
0
T
Indeed, κϕ = 0 (since N ≥row KA, by definition of K) and for any morphism τ : L →
− M with
τ ϕ = 0 (i.e., N ≥row TA) we need to prove that there is exactly one morphism
τ /κ := KernelLift(τ, κ) := (L, X, ker ϕ)
such that (τ /κ)κ = τ : This last condition means that M must row-dominate28 XK − T. But in
fact we know that K ≥row T and we can achieve XK − T = 0 by setting29
X := RightDivide(T, K).
With this choice of X we want to see that (L, X, ker ϕ) is a compatible triple. First note that
M ≥row LT = L(XK) = (LX)K since τ is a morphism, i.e., LX is a matrix of syzygies of K modulo
M. But then, by definition, SyzygiesOfRows(K, M) = ker ϕ ≥row LX and we are done.
28
or, equivalently, that K ≥row T modulo M.
29
We also could have used X := RightDivide(T, K, M). It is not clear to me if this is generically slower.
 46 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

Before we discuss the A BELian structure we need a relative version of the matrix algo-
rithm RightDivide, which we can deduce as a derived algorithm.
Definition 2.63. Let A, B, N be stackable30 matrices over R. The matrix A is said to row-
dominate B modulo N if there exists matrices X, Y such that B = XA + YN. The definition of
relative column-domination is analogous.
Since B = XA + YN = ( X Y ) ( AN ) we can derive X (and Y) form RightDivide and StackMat

( X Y ) := RightDivide(B, ( AN )).

We define the relative version31 of RightDivide by applying CertainColumns to extract X

X =: RightDivide(B, A, N).

Analogously we derive the relative version LeftDivide(A, B, N).

It now remains to prove that R-fpres is A BELian.
Theorem 2.64. If R is left computable then R-fpres is constructively A BELian.

Proof. First we will show that any mono is the kernel of its cokernel (cf. Corollary 2.43). Let
A
ϕ : M ,→ N be a mono. From the proof of Proposition 2.62 we know that ε = (N, 1, ( AN ) = coker ϕ)
is the cokernel epi of ϕ.

0
M A
1
γ/ϕ = X G N ( AN ) = coker ϕ
M0
0

G
Let γ : M0 → − N be another morphism with γε = 0, i.e., ( AN ) ≥row G = G · 1, or, equivalently,
A ≥row G modulo N, i.e., there exists X, Y such that G = XA + YN. A candidate for the unique
X
lift32 of γ along ϕ would be γ/ϕ : M0 →
− M with

X = RightDivide(G, A, N),

once we have shown that the triple (M0 , X, M) is compatible, i.e., that M ≥row M0 X: For this
note that N ≥row M0 G since γ is a morphism. Furthermore N ≥row M0 YN, tautologically. Hence
N ≥row M0 G − M0 YN = M0 XA = (M0 X)A. Now recall that ϕ is mono means that its kernel mono
vanishes by Exercise 2.2.7, which translates to M ≥row K := SyzygiesOfRows(A, N) (cf. the
proof of Proposition 2.62). But K row-dominates by definition any T with N ≥row TA, in
End particular M ≥row K ≥row M0 X.
lect. 13 A
To show that any epi ϕ : M  N is a cokernel epi of its kernel mono κ := (ker ϕ, K, M)
22.06 (cf. Corollary 2.43) it suffices to show that the unique colift ε\ϕ = (coker κ = ( K ) , A, N)
M
of ϕ along the cokernel epi ε = (M, 1, ( KM ) = coker ϕ) of κ is an isomorphism, where K :=
SyzygiesOfRows(A, N) (cf. the proof of Proposition 2.62):
30
i.e., with the same number of columns.
31
In practice, one can derive more efficient algorithms to compute the relative version of RightDivide.
32
Lifts along monos are unique.
7. FUNCTORS 47

coker κ = ( KM )
1
K := SyzygiesOfRows(A, N)
ε\ϕ = A M ker ϕ
A
N

An inverse morphism α of the colift ε\ϕ must satisfy α(ε\ϕ) = 1N . This implies for α that
Y
α := (ε\ϕ)−1 : N →
− ( KM ) = coker κ with Y := RightDivide(1, A, N),
where 1 is the gN × gN identity matrix. The existence of the matrix Y is equivalent to our
assumption that ϕ is an epi. Now we show that the triple α = (N, Y, ( KM )) is compatible and
hence defines a morphism. By definition of Y there exists a matrix Z such that
YA + ZN = 1. (*)
Multiplying with N from the left we obtain NYA + NZN = N, or equivalently (NY)A = (1 − NZ)N,
i.e., N ≥row (NY)A. But K row-dominates by definition any T with N ≥row TA, hence K ≥row NY
and α is a compatible triple. Since α(ε\ϕ) = 1N by definition of Y it remains to show that
(ε\ϕ)α = 1coker κ . To show this we multiply (*) with A from the left and obtain AYA + AZN = A,
or equivalently (AY − 1)A = (AZ)N, hence N ≥row (AY − 1)A and as above K ≥row (AY − 1) and
we are done.
In the next sections we will develop enough language to show that R-fpres is equivalent
to the category R-fpmod of finitely presented modules over R. When R is left computable
this will prove that the category of finitely presented R-modules is constructively A BELian.

7 Functors
In this chapter we will study functors as “1-morphism” between categories. It turns out that
there are “2-morphisms” which we will call natural transformations; they are “morphisms
between the functors”.
Definition 2.65. Let A and B be categories. A (covariant) functor F : A → B consists of
two maps A0 → B0 (on objects) and A1 → B1 (on morphisms33 ) such that
• F (HomA (M, N )) ⊂ HomB (F (M ), F (N ));
• F (1M ) = 1F (M ) ;
• F (ϕψ) = F (ϕ)F (ψ)
for all M, N ∈ A and any pair of composable A-morphisms ϕ, ψ. We write F ∈ B A .
We will see later how to turn B A into a category of functors.
The composition of functors is again a functor and the composition is associative. The
notions of identity functor, invertible functor, and inverse functor of an invertible functor
are evident. However, invertible functors offer a far too restrictive notion of equivalence,
which does not take the relative nature of categories into account, i.e., the possibility of
many unequal but isomorphic objects to exist.
33
You cannot speak of a functor if you don’t know how to define it on morphisms!
48 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

Definition 2.66. Let F : A → B be a functor and consider the maps

FM,N : HomA (M, N ) → HomB (F (M ), F (N )).

F is called

• faithful if FM,N is injective for all M, N ∈ A.

• full if FM,N is surjective for all M, N ∈ A.

• fully faithful if F is full and faithful.

F is called essentially surjective (or dense) if for each of L ∈ B there exists an object M ∈ A
such that F (M ) ∼
= L (in B).
We start with some trivial examples.
Example 2.67.

(1) Viewing a monoid as a one-object category a functor is nothing but a homomorphism

of monoids. The same is true for groups and one-object groupoids. Faithful functors
are monos, full functors are epis, and fully faithful functors are isomorphisms.

(2) Functors between posets34 are nothing but order-preserving maps.

(3) The embedding functor of a subcategory A0 ⊂ A is a faitful functor. A0 is a full

subcategory iff the embedding functor is full.

(4) The path components functor π0 from (Top) to (Sets).

(5) The first fundamental group π1 is a functor from (Top0 ) to (Grps).

(6) The so-called “forgetful” (or “underlying”) functors which forget some structure of
the category

• from graded S-modules to S-modules (where S is a graded ring)

• from R-Mod to (Ab);
• from (Ab) to (Sets0 );
• from (Sets0 ) to (Sets);
• from (Top) to (Sets);
• ...

(7) The “free” functors which associate to a set the “free” object on this set, e.g.,

• from (Sets) to (Ab);

• from (Sets) to R-Mod;
• ...

(8) ...

Main example 2.68. Let A be a category. Each object M ∈ A gives rise to a functor to
(Sets), namely the covariant Hom-functor
34
viewed as skeletally thin categories (cf. 2.4.(12))
8. NATURAL TRANSFORMATIONS 49

ψ N
M ϕ
ψϕ L

hM := HomA (M, −) : N 7→ HomA (M, N ),

( !
HomA (M, N ) → HomA (M, L)
(ϕ : N → L) 7→ HomA (M, ϕ) : .
ψ 7→ ψϕ

Exercise 2.14. Let A be a category. Construct from A so-called arrow categories in

which ker, coker, im, coim, pull-backs, push-outs (if they exist in A) are functors.
Definition 2.69. A contravariant functor G between the categories A and B is a functor
op
G : Aop → B. We write G ∈ B A . Since G ROTHENDIECK we call them presheaves on A
with values on B.
Main example 2.70. Let A be a category. Each object N ∈ A gives rise to a presheaf to
(Sets), namely the contravariant Hom-functor

ψ M
N ϕ
ϕψ L

hN := HomA (−, N ) : M 7→ HomA (M, N ),

( !
HomA (M, N ) → HomA (L, N ),
(ϕ : L → M ) 7→ HomA (ϕ, N ) : .
ψ 7→ ϕψ

8 Natural transformations
Natural transformations will play the role of second level morphisms, i.e., the morphisms
between the functors.
Definition 2.71. A natural transformation η : F → G of the functors F, G : A → B
is a map η : A0 → B1 such for all A-morphisms ϕ : M → N that the following diagram
commutes

F (ϕ)
F (M ) F (N )
ηM ηN
G(ϕ)
G(M ) G(N )

The morphism ηM is called the component of η at M ∈ A. We will usually write η : F =⇒

η
G or F =⇒ G, but my favorite notation is the (less economic) globular one

F
A η
B
G
 50 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

as it emphasizes the 2-morphism nature of natural transformations. We denote the set of

natural transformation between F and G by [F, G], or B A (F, G), or HomBA (F, G).
The natural transformation η is called a natural mono, natural epi, or natural isomor-
phism if ηM is mono, epi, or an isomorphism for all M ∈ A, respectively. If F = G we then
use the notions natural endomorphisms and natural automorphisms instead.
The functors F and G are called equivalent if there exists a natural isomorphism η :
F =⇒ G. We write F ' G or F 'η G.
Example 2.72. Let A, B be categories and F, G, H ∈ B A .

(1) We now show that B A with functors as objects and natural transformations as mor-
phisms is a category:
η
• The source of a natural transformation F =⇒ G is F and the target is G.
• The map 1F : A0 → B1 , M 7→ 1F (M ) is a natural automorphism of F , called the
identity transformation of F .
η τ
• The (vertical) composition of two natural transformations F =⇒ G =⇒ H is
defined in the obvious way (ητ )M = ηM τM , i.e., by composing the two commu-
tative diagrams

F (ϕ)
F (M ) F (N )
ηM ηN
G(ϕ)
G(M ) G(N )
τM τN
G(ϕ)
H(M ) H(N )
ητ
and is again a natural transformation F =⇒ H. The composition of natu-
ral transformations is associative since it is defined by the composition of mor-
phisms.
End
lect. 14
(2) The evaluation map εV : V → V ∗∗ , v 7→ (ϕ 7→ ϕ(v)) between a k-vector space and
27.06
its double dual is a natural mono between the identity functor Idk-Vect and double-
dualizing functor D2 : k-Vect → k-Vect. The natural transformation ε becomes a
natural isomorphism if we restrict ourselves to the category of finite dimensional k-
vector spaces. The natural transformation ε is also definable for R-Mod and R-mod,
where it is in general neither mono nor epi.

(3) Using Exercise 2.14 we see that the kernel mono, cokernel epi, the coimage mono and
epi, the image mono and epi, and the induced morphism ϕ : coim ϕ → im ϕ are all
natural transformations. In A BELian categories the latter is a natural isomorphism
between the coimage and the image functors.

Remark 2.73. If we view categories as objects and functors as morphisms we get a cate-
gory of categories, with the restrictive notion of equality of functors. If we want to consider
functors up to equivalence we get what is called a 2-category with the natural transforma-
tions as 2-morphisms. This process can be iterated indefinitely with marvelous applications
in advanced mathematics.
Exercise 2.15. The category of functors AB from a small category B to an A BELian cate-
gory A is again A BELian.
8. NATURAL TRANSFORMATIONS 51

Now we define the notion of equivalence of categories.

Definition 2.74. An equivalence of the categories A and B is a pair of functors

F :AB:G

with two natural isomorphisms G ◦ F ' IdA and F ◦ G ' IdB . The two categories A, B are
then called equivalent and we then write F : A ' B : G, or simply A ' B.
F G
If F : A ' B : G then the composed map HomA (M, N ) → HomB (F (M ), F (N )) →
HomA ((G ◦ F )(M ), (G ◦ F )(N )), ϕ 7→ (G ◦ F )(ϕ) is a bijection, given by conjugating with
(the components of) the natural isomorphism. Hence the maps FM,N : HomA (M, N ) →
HomB (F (M ), F (N )) are bijections for all M, N ∈ A0 . In other words, F is fully faithful.
Now, for B ∈ B the B-component B ∼ = F (G(B)) of the natural isomorphism F ◦ G ' IdB
states that F is essentially surjective. Thus, we have proved one direction of the following
criterion.
Proposition 2.75. A functor F : A → B is part of an equivalence iff it is fully faithful and
essentially surjective.
Proof. The other direction is immediate with a sufficiently strong axiom of choice.
Remark 2.76. Equivalence of categories preserves all categorical properties, i.e.,
• having initial, terminal, or zero objects;

• having kernels, cokernels, pull-backs, push-outs, products, coproducts;

• being pre-additive, additive, pre-A BELian, or A BELian.

• ...
In fact, one can always improve the above notion of equivalence to an “adjoint equiva-
lence”, which depends on the notion of adjoint functors.
Now we prove that R-fpres is indeed equivalent to the category of finitely presented
R-modules. We already know that HomR (R1×r , R1×c ) ∼ = Rr×c , i.e., that R-matrices are in
bijection with R-linear maps between free modules of finite rank.
Definition 2.77. Let R be a ring. A left R-module N is called finitely presented if there
exists a matrix N ∈ RrN ×gN such that N ∼
N
= coker(R1×rN →
− R1×gN ). We call such an N a pre-
sentation matrix of N . We denote the category of finitely presented modules by R-fpmod.
It is a full subcategory of R-mod.
The images (f1 , . . . , fgN ) of the standard free generators (f1 , . . . , fgN ) of the free module
1×gN
R under the cokernel epi εN : R1×gN  N is a set of generators of N . By definition,
N εN = 0N iff N ≥row N0 , i.e., there exists a matrix X such that N0 = XN.
0

The rows of the matrix N are the images of the standard free generators (f01 , . . . , f0rN ) of
the free module R1×rN under the R-linear map N : R1×rN → R1×gN . They are called a set of
relations of N .
We now want to describe R-linear maps between M = coker M and N = coker N. Let
A ∈ RgM ×gN be a matrix for which N ≥row MA, i.e., for which there exists an X satisfying
MA = XN.
M εM
R1×rM R1×gM M
X A ϕA
N εN
R1×rN R1×gN N
 52 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

Such an A induces an R-linear map ϕ = ϕA : M → N, mεM 7→ mAεN forcing the right square
to commute: εM ϕ = AεN . We only need to see that ϕ is well-defined. For this we have to
check that mAεN = 0N whenever mεM vanishes. But mεM = 0M means that there exists a matrix
x ∈ R1×rM such that m = xM. Hence, mAεN = xMAεN = xXNεN = 0N . Two matrices A, A0 give
rise to same R-linear map ϕ iff AεN = A0 εN , or equivalently, iff (A − A0 )εN = 0N and hence iff
N ≥row A − A0 .
PgNConversely, an R-linear map ϕ : M → N defines a coefficients matrix 1×gM
A by ei ϕ = ei εM ϕ =
j=1 Aij fj , where (e1 , . . . , egM ) are the standard free generators of R . So, by definition,
εM ϕ = AεN and the right square commutes.
Summing up, we have just proved the following:
Theorem 2.78. The essentially surjective functor

R-fpres → R-fpmod,

coker : M 7→ M = coker M,

(M, A, N) 7→ ϕA : M → N = coker N


is fully faithful and is hence an equivalence of categories R-fpres ' R-fpmod. If R is left computable
then R-fpmod is A BELian and R-fpres is constructively A BELian. In particular, R-fpres is a
constructive model for R-fpmod.

9 Exact sequences, homology, and cohomology

Cont.
lect. 18 As we have seen in Proposition 2.27, the existence of kernels, cokernels, and biproducts
11.07 implies that of push-outs and pull-backs. They in turn imply the existence of intersections
(meets) and sums (joints) in the lattice of subobjects. Dually, there is a lattice of factor
objects and, by duality, it has joins (subdirect product) and meets (largest common factor).
In A BELian there is a G ALOIS connection35 between the lattice of subobjects and the
ι
lattice of factor objects: To a subobject K ,→ M we can associate the factor object
 ι

M/K := coker K ,→ M

π
represented by theπ cokernel
 epi of ι. And to a factor object M  C we can associate the
subobject ker M  C represented by the kernel mono of π. Corollary 2.43 now says that
in A BELian categories this G ALOIS connection is a bijection. Hence, we can draw subobjects
and factor objects using just one H ASSE diagram. The second isomorphism of N OETHER
implies the modularity of these lattices.
τ κ
Let L ,→ M ←- K be two subobjects of M with L ≤ K in some A BELian category. In
τ /κ
Exercise 2.5 we saw how the unique lift L ,→ K of τ along the dominant κ realizes L as a
subobject of K.
Definition 2.79. We then call the factor object
 
τ /κ
K/L := coker L ,→ K

together with the two morphisms K/L ,→ M/L  M a subfactor object of M .

35
monotone or antitone, depending on how we define both.
9. EXACT SEQUENCES, HOMOLOGY, AND COHOMOLOGY 53

The two morphisms K/L ,→ M/L  M is an example of what we will later call a
generalized mono from K/L to M .
The definition of subfactors is all what we need to define defects of exactness, one of the
central notions in homological algebra.
Definition 2.80. Let A be a category, with zero when necessary. Two composable A-
ϕ ψ
morphisms M − → N −
→ L are called a short sequence. We call it a differential short se-
quence if ϕψ = 0.
ϕ ψ
Let M − → N − → L be a differential sequence in an A BELian category. The defect of
exactness (at N ) is defined as the subfactor object of N

Def(ϕ, ψ) = ker ψ/ im ϕ,

where the two subobjects im ϕ ≤ ker ψ of N are represented by the image mono and the
kernel mono im ϕ ,→ N ←- ker ψ, respectively36 . The sequence is called exact (at N ) if
Def(ϕ, ψ) = 0.
ϕ ψ
Exercise 2.16. Let M −
→N − → L be differential sequence in an A BELian category. The
following statements are equivalent:
• The sequence is exact at N .

• The image mono of ϕ is a kernel mono of ψ.

• The image epi of ψ is cokernel epi of ϕ.

Definition 2.81. Let A be a category, with zero or A BELian when necessary.
(1) A sequence M• in A is a (possibly doubly unbounded) numbered array consisting of
short sequences in A
∂i−1 ∂ ∂i+1 ∂i+2
−i Mi ←−− Mi+1 ←−− · · · ,
M• : · · · ←−− Mi−1 ←

i.e., with s(∂i ) = t(∂i+1 ) for all i.

If M• consists of short differential sequences (i.e., ∂i+1 ∂i = 0 for all i) then we call it a
complex37 . The defect of exactness

Hi (M• ) := Def(∂i+1 , ∂i ) = ker ∂i / im ∂i+1

is called the i-th homology of M• or the homology at Mi . It is a subfactor object of Mi .

If the complex is left bounded then its left most homology is the cokernel of the left
most morphism. At a right boundary we take the kernel of the right most morphism.

(2) A co-sequence M • in A is a (possibly doubly unbounded) numbered array of short

sequences in A
di−2 di−1 di di+1
M • : · · · −−→ M i−1 −−→ M i −
→ M i+1 −−→ · · · ,

i.e., with t(di ) = s(di+1 ) for all i.

If M • consists of differential sequences (i.e., di−1 di = 0 for all i) then we call it a
cocomplex38 . The defect of exactness

H i (M • ) := Def(di−1 , di ) = ker di / im di−1

36
The latter dominates the former by the universal property of the kernel.
37
or chain complex or homological complex.
38
or cochain complex or cohomological complex.
54 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES

is called the i-th cohomology of M • or the cohomology at M i . It is a subfactor object

of M i . If the cocomplex is left bounded then its left most homology is the kernel of
the left most morphism. At a right boundary we take the cokernel of the right most
morphism.
The only difference between complexes and cocomplexes is whether i decreases or in-
creases in the direction of the arrows. Of course, this can only has any relevance if the
numbers i carry a mathematical meaning, e.g., being dimensions or codimensions in some
context.
We call a (co)complex acyclic or exact if its (co)homology at each intermediate39 object
vanishes. The notions of left resp. right acyclicity (or exactness) are evident.
In categories with zero we can always think of complexes as doubly infinite ones by
augmenting zero morphisms on both sides. However, this might alter the meaning of
acyclicity!
Definition 2.82. A chain morphism (in A) µ• : M• → N• between two complexes M•
and N• in A is an array of morphisms µi : Mi → Ni such that following diagram commutes:

··· Mi−1 Mi Mi+1 ···

µi−1 µi µi+1
··· Ni−1 Ni Ni+1 ···

The dual notion is that of a cochain morphism.

Chain complexes and their chain morphisms form a category which we will denote
by Ch• A. The full subcategories of all complexes with Mi = 0 for i ≥ 0, i  0, i ≤ 0,
i  0, |i|  0 are denoted by Ch≥0 , Ch+ A, Ch≤0 , Ch− A, Chb A, respectively. For cochain
complexes we write upper indices Ch• A, etc.
Definition 2.83. We call a subobject40 of M• a subcomplex. Dually, a factor complex is
a factor object of M• .
Remark 2.84. Let A be additive or A BELian category. Then the categories Ch• A, Ch≥0 A,
Ch+ A, Ch≤0 A, Ch− A, Chb A are again additive or A BELian, respectively.
If A is A BELian then a chain morphism µ• is mono41 (resp. epi, resp. iso) iff µi is mono
(resp. epi, resp. iso) for all i ∈ Z.
Let M• = (M• , ∂•M ) be a complex in the additive category A. A degree n chain morphism
M• → N• is a chain morphism M• → N•+n or, equivalently, M•−n → N• .
For example, the array ∂•M itself can be interpreted as a degree −1 chain endomorphism
of M• , i.e., as a chain morphism M• → M•−1 .
In this language a chain morphism µ• : M• → N• satisfies ∂•M µ• = µ• ∂•N as chain mor-
phisms M• → N•−1 (or M•+1 → N• ).
Definition 2.85. Let M• be a complex in an A BELian category. The image of M•+1 → M•
is called the subcomplex of boundaries and denoted by B(M• ). The kernel of M• → M•−1
is called the subcomplex of cycles and denoted by Z(M• ). It follows that B(M• ) ≤ Z(M• )
and Hi (M• ) = Z(Mi )/B(Mi ).
Exercise 2.17. Prove the 5-Lemma.
ϕ ψ
Exercise 2.18. For a short exact sequences 0 → M 0 − → M 00 → 0 in an A BELian
→ M −
category A the following are equivalent:
39
i.e., an object which is a source and a target.
40
in any of the above categories Ch? A
41
in any of the above categories Ch? A
10. FREE RESOLUTIONS 55

(1) ϕ is a split mono.

(2) ψ is a split epi.

If these conditions are fulfilled, then M ∼

= M 0 ⊕ M 00 and we call the exact sequence split.
Definition 2.86. A functor F : A → B between A BELian categories is called

• left exact if it maps exact A-sequences of the form 0 → M 0 → M → M 00 to exact

B-sequences 0 → F (M 0 )x → F (M ) → F (M 00 ).

• right exact if it maps exact A-sequences of the form M 0 → M → M 00 → 0 to exact

B-sequences F (M 0 ) → F (M ) → F (M 00 ) → 0.

• exact if it is left and right exact.

Proposition 2.87 (Bivariate left exactness of Hom). Let A be an A BELian category and N ∈
A.

(1) The Hom-functor HomA (−, N ) : Aop → (Ab) is left exact.

(2) The Hom-functor HomA (N, −) : A → (Ab) is left exact.

Proof. Exercise.
Exercise 2.19. Show that the Hom-functor is generally not exact in any of its arguments.

10 Free resolutions
Let M be an R-module. We called an exact sequence of the form
ϕ
0 ← M ← F0 ←
− F1

with free R-modules F1 , F2 a free presentation of M . If F1 and F2 are of finite ranks, then
we can identify ϕ with the presentation matrix we encountered Definition 2.77. Iterating
this process for the kernel of the presentation morphism ϕ yields:
Definition 2.88 (Free resolution). A (possibly infinite) exact sequence of the form
ϕ1 ϕ2
0 ← M ← F0 ←− F1 ←− F2 ← · · ·

with free R-modules Fi is called a free resolution of M . The submodule ker ϕi = im ϕi+1
is called the i-th syzygy module of M . We say that a resolution is finite of length c ∈ N if
Fi = 0 for all i ≥ c + 1.
The isomorphism type of the i-th syzygy module of M depends on the choice of the free
resolution of M .
The matrix version of Hilbert’s Syzygy Theorem Theorem 1.46 implies:
Corollary 2.89 (Hilbert’s syzygy theorem (module version)). Let K be a field and R =
K[x1 , . . . , xn ]. Then each finitely generated R-module M has a finite free resolution of length ≤ n
consisting of free modules of finite rank.
Example 2.90. Let R := Z/4Z. Using Ext’s it can easily be shown that the R-module
R/h2i does not admit a finite free resolution.
56 CHAPTER 2. CONSTRUCTIVE ABELIAN CATEGORIES
Chapter 3

Localization and tensor products

From now on R will denote a unital commutative ring.

1 Preliminaries
We start by recalling some operations on ideals:
T
• the (set theoretic) intersection i∈I Ii ;
Qk DQ E
k
• the product i=1 Ii := i=1 ai | ai ∈ Ii ;
R
P S
• the sum i∈I Ii := i∈I Ii R
;

• the ideal quotient1 I : J := {a ∈ R | aJ ⊂ I} ⊃ I;

• the annihilator Ann(I) := AnnR (I) = h0i : I;

√
• the radical I := rad I := {a ∈ R | am ∈ I for some m ∈ N}.

Example 3.1. Consider R := Z, I := hmi = mZ, and J := hni = nZ for m, n ∈ N≥1 . Then

• I + J = hmi + hni = hgcd(m, n)i.

• I ∩ J = hmi ∩ hni = hlcm(m, n)i.

• I · J = hmi · hni = hm · ni.

D E D E
m lcm(m,n)
• I : J = hmi : hni = gcd(m,n) = n
.

• Ann(I) = {0}.
√ DQ
k
E
, where m = ki=1 pαi i is the prime factorization of m with αi > 0.
Q
• I= i=1 p i

Exercise 3.1. Write down categorical algorithms for the intersection, the sum, and the
ideal quotient.
Definition 3.2. A ring is called reduced if 0 is the only nilpotent element of R.
√
Exercise 3.2. R/I is reduced iff I is a radical ideal (i.e., I = I).
Remark 3.3.
1
for a ∈ R we set I : a := I : hai

57
 58 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

(1) Clearly
k
Y k
\
Ii ⊂ Ii .
i=1 i=1
The example 4Z · 6Z = 24Z ( 12Z = 4Z ∩ 6Z shows that the inclusion could be strict.
(2) However, v v
u k uk
uY u\
t I =t I.
i i
i=1 i=1

Definition 3.4. Let ϕ : R → S be a morphism of unital rings.

• For I  R define the extension of I to S to be the ideal I e := hϕ(I)iS  S.
• For J  S define the contraction of J to R to be the ideal J c := ϕ−1 (J)  R.
Remark 3.5. With the notation of the previous definition:
(1) I ec ⊃ I;
(2) J ce ⊂ J;
(3) I ece = I e ;
(4) J cec = J c .
Proof. Exercise.
Definition 3.6. We write Spec R := {p
R | p prime} and call it the (prime) spectrum
of R. Analogously Max R := {m  R | m maximal} ⊂ Spec R and call it the maximum
spectrum of R.
Exercise 3.3. p is prime iff for all I, J  R: I · J ⊂ p =⇒ I ⊂ p or J ⊂ p.
End
lect. 15 Example 3.7. Let K be a field.
29.06
(1) Max Z := {hpi | p ∈ N>0 prime} and Spec Z = {h0i} ∪ Max Z.
(2) Max K[x] = {hf i | f ∈ K[x] irreducible} and Spec K[x] = {h0i} ∪ Max K[x].
(3) If K is algebraically closed then Max K[x] = {hx − ai | a ∈ K}.
(4) If K is algebraically closed then Max K[x, y] = {hx − a, y − bi | a, b ∈ K} and
Spec K[x, y] = {h0i} ∪ {hf i | f ∈ K[x, y] irreducible} ∪ Max K[x, y]. To prove this
we need Hilbert’s Nullstellensatz which we will encounter later.
Proposition 3.8 (Prime avoidance). Let p1 , . . . , pk ∈ Spec R and I  R. Then
k
[
I⊂ pi =⇒ I ⊂ pi for some i.
i=1
S
Proof. We use induction on k: For k = 1 the is nothing to prove. Now if I ⊂ i6=j pi for
some j we are done by the S induction hypothesis. Otherwise, there is for each j = 1, . . . , k
an aj ∈ I such that aj ∈ / i6=j pi (and hence aj ∈ pj ). We consider two cases. First, if k = 2
then a1 + a2 ∈ / p1 and a1 + a2 ∈ / p2 . But then a1 + a2 ∈/ p1 ∪ p2 , so I 6⊂ p1 ∪ p2 , contradicting
the assumption.
Second, if k > 2 we apply the same argument on a1 + a2 · · · ak . It holds a2 · · · ak ∈ / p1 since
p1 is prime and would otherwise contain one of the factors. So with a1 ∈ p1 we conclude
that a1 + a2 · · · ak ∈
/ p1 . For i ≥ 2 we have a2 · · · ak ∈ pi and a1 ∈/ pi , hence a1 + a2 · · · ak ∈
/ pi .
Again a contradiction.
2. NILRADICAL AND JACOBSON RADICAL 59

Definition 3.9. A local ring R is a ring with exactly one maximal ideal m (i.e., Max R is
a singleton {m}). We occasionally write (R, m) and call R/m the residue field of the local
ring (R, m).

2 Nilradical and Jacobson Radical

Proposition 3.10. Let I  R be an ideal in R. Then the radical of I is the intersection of all
prime ideals of R containing I:
√ \
I= p.
p∈Spec R
I⊂p

Proof.
√
“⊂” Let a ∈ I and p
R be some prime ideal such that I ⊂ p. Then there exists an m ≥ 1
with am ∈ I ⊂ p by the definition of the radical. Hence a ∈ p, since p is prime.
√
“⊃” Let a ∈ R \ I. The inclusion relation on the set

Γ := {J  R | I ⊂ J, am ∈
/ J ∀m ≥ 1}

is an inductive order. Hence, there exists a maximal element q of Γ by Zorn’s Lemma

(seeTAlgebra LN). We will now prove that q is a prime ideal (for then a ∈ R \ q ⊂
R \ p∈Spec R p and we are done): Let f, g ∈/ q. Then q + hf i and q + hgi strictly contain
I⊂p
q, which means that both ideals do not belong to Γ. Hence am ∈ q+hf i and an ∈ q+hgi
for some m, n ≥ 1. Then am+n ∈ q + hf gi, i.e., q + hf gi is not an element of Γ either. In
particular, f g ∈
/ q.
Definition 3.11.
(1) The ideal
\ Proposition 3.10 p
N (R) := p = h0i = {a ∈ R | a nilpotent}
p∈Spec R

is called the nilradical of R.

(2) The ideal \

J(R) := m
m∈Max R

is called the Jacobson radical of R.

A characterization of the Jacobson radical analogous to that of the nilradical exists:
Lemma 3.12.
J(R) = {a ∈ R | 1 − ab ∈ R∗ ∀b ∈ R}.
Proof.
/ R∗ . Then there exists
“⊂” Let a ∈ J(R) and assume that there is some b ∈ R with 1 − ab ∈
a maximal ideal m with 1 − ab ∈ m (Algebra LN). But since a ∈ J(R) it follows that
a ∈ m and hence ab ∈ m. This means 1 ∈ m, a contradiction.

“⊃” Let a ∈
/ J(R). Then there exists a maximal ideal m of R with a ∈/ m. Hence m+hai = R,
so one can find elements c ∈ m and b ∈ R such that c + ab = 1. But then 1 − ab = c ∈ m,
which cannot be a unit.
60 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

3 The Cayley-Hamilton Theorem and Nakayama’s Lemma

Note that an R-module M is finitely generated (f.g.) iff M is (isomorphic to) an epimorphic
image of a free module Rn of finite rank n. Indeed, for M = hm1 , . . . , mn i consider the
epimorphism
π : Rn  M, ei 7→ mi .
Then M ∼ = Rn / ker π. For the converse consider an R-submodule I ≤ Rn and an epimor-
phism
π : Rn  Rn /I =: M .
and set mi := π(ei ) yielding M = hm1 , . . . , mn i since π is an epi.
Theorem 3.13 (Cayley-Hamilton). Let I  R be an ideal, M a f.g. R-module and ϕ ∈
EndR M . If M is generated by n elements, and if ϕ(M ) ⊂ I · M , then there exists a monic polyno-
mial
χϕ (x) := xn + p1 xn−1 + · · · + pn ∈ R[x]
with pj ∈ I j = I| · .{z
. . · I}, such that
j

χϕ (ϕ) = 0 ∈ EndR M .

Proof. Let M = hm1 , . . . , mn iR . Since ϕ(M ) ⊂ I · M , we may write ϕ(mi ) as an I-linear

combination of the mj ’s:
n
X
ϕ(mi ) = aij mj , with aij ∈ I. (*)
j=1

Now consider M as an R[x]-module via

x · m := ϕ(m) for all m ∈ M (**)

and rewrite (*) in the matrix notation

   
m1 P ! 0
 ..  xm1 − aij mj (**)  . 
(x · In − A)  .  = .. =  ..  ,
.
mn 0

where A = (aij ) and In is the n × n identity matrix. Multiplying with adjoint matrix (x · In −
A)adj we get  
m1
 .. 
det(x · In − A)  .  = 0.
mn
So det(x · In − A)mi = 0 for all i, hence χϕ (x) := det(x · In − A) ∈ AnnR[x] M , or equivalently
χϕ (ϕ) = 0 by (**).
Corollary 3.14. Let M be a f.g. R-module.

(1) An epi α ∈ EndR M is already an isomorphism.

(2) Let R 6= 0. If M ∼
= Rn , then any set of n elements generating M is a free basis. In particular,
rank M = n is well-defined.

Proof.
3. THE CAYLEY-HAMILTON THEOREM AND NAKAYAMA’S LEMMA 61

(1) We will construct the inverse of α. For this consider M as an R[t]-module via t · m :=
α(m) for all m ∈ M . Define I := hti
R[t] and ϕ := idM ∈ EndR[t] M . Then
α epi
ϕ(M ) = M = α(M ) = t · M = I · M .

Applying Cayley-Hamilton we get a polynomial

n
X
χϕ (x) = xn + pi xn−i ∈ R[t][x],
i=1

with pj ∈ htj i for each j and

n
X
0 = χϕ (ϕ) = idM + pi (α) idM .
i=1

Now consider
p1 + · · · + pn
q(t) := ∈ R[t].
t
Then for all m ∈ M
n
! n
!
X X
m = idM (m) = − pi idM (m) = − pi m = t · (−q(t)) · m.
i=1 i=1

Hence −q(α) is the inverse of α.

(2) Fix an isomorphism β : M → Rn and suppose that M = hm1 , . . . , mn i. Consider the

epi π : Rn → M, ei 7→ mi . Then π ◦ β : M → M is epi and thus an isomorphism
by 1. But then π = (π ◦ β) ◦ β −1 is an isomorphism and m1 , . . . , mn are R-linearly
independent.
For the rank suppose Rk ∼ = Rn with k < n. Now extending a free basis of length k by
n − k zeros to a set of generators for M with n elements contradicts the established
R-linearly independency of sets of generators of cardinality n.
End
Corollary 3.15. Let M be a f.g. R-module and I  R an ideal such that I · M = M . Then there lect. 16
exists an r ∈ I satisfying 04.07
(1 − r)M = 0.
Proof. Consider ϕ = idM in Theorem 3.13 to get pj ∈ I j ⊂ I such that
n
! n
!
X X
1+ pi · m = idM + pi idM (m) = χϕ (ϕ)(m) = 0
i=1 i=1
Pn
for all m ∈ M . Now define r := − i=1 pi .
Corollary 3.16 (Nakayama’s Lemma). Let M be a f.g. R-module and let I  R be and ideal
contained in J(R). If I · M = M then M = 0.
Proof. Choose r ∈ I such that (1 − r)M = 0 as in Corollary 3.15. Then r ∈ m for all
m ∈ Max R since I ⊂ J(R). Hence 1 − r ∈/ m (otherwise 1 ∈ m) for all m ∈ Max R, and
∗
1 − r ∈ R by Lemma 3.12. We conclude that M = 0.
Corollary 3.17 (Nakayama’s Lemma, local version). Let (R, m) be a local ring and M a f.g.
R-module. If mM = M then M = 0.
62 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

Proof. The result follows from Nakayama’s Lemma since m = J(R).

Corollary 3.18 (Nakayama’s Lemma, 2nd local version). Let (R, m) be a local ring, M a
f.g. R-module, N ≤ M an R-submodule. If N + mM = M then N = M .
Proof. Replacing M by M/N reduces to the case N = h0i.
Corollary 3.19. Let (R, m) be a local ring, M a f.g. R-module, and m1 , . . . , mn ∈ M . Then the
mi ’s generate M iff their residue classes mi := mi + mM generate M/mM as an R/m-vector space.
In particular, any minimal set of generators of M corresponds to an R/m-basis of M/mM and two
such sets have the same cardinality.
Corollary 3.18
Proof. Let N = hm1 , . . . , mn i. Then m1 , . . . , mn generate M ⇐⇒ N = M ⇐⇒ N+
mM = M , which in turn is equivalent to

hm1 , . . . , mn iR/m = M/mM .

Example 3.20. Caution: The last statement may not hold for non-local rings: Consider
a field K and hx, 1 − xi = h1i = K[x]. Both sets of generators are minimal but their cardi-
nalities differ.
Corollary 3.21. Let (R, m) be a local ring, M a f.g. R-module, and ϕ : M → N a morphism of
R-modules. Then ϕ is an epi iff the induced map ϕ : M/mM → N/mN, m 7→ ϕ(m) is an epi.
Proof. We only proof the nontrivial direction. So suppose ϕ is epi. Then given n ∈ N , there
exists an m ∈ M such that ϕ(m) = n. That is ϕ(m) − n ∈ mN . Hence im ϕ + mN = N and
im ϕ = N by Corollary 3.18.

4 Tensor products of modules over commutative rings

Definition 3.22. Let M, N, L be R-modules. A map

Φ:M ×N →L

is called R-bilinear if it is R-linear in each of the two arguments.

Theorem 3.23. Let M and N be R-modules. Then there exists a universal R-bilinear map
t : M × N → T , i.e., such that for any other R-bilinear map t0 : M × N → T 0

t T
M ×N ϕ
0
t0 T

there exists a unique map ϕ making the above diagram commute.

Proof. The uniqueness follows as usual from
L the uniqueness of the post-factor ϕ. For the
existence consider the free module F := i∈M ×N R and write e(m,n) for the standard basis
vector of F corresponding to the index (m, n). Let I  F be the R-submodule generated by
elements of the following types:

e(m+m0 ,n) − e(m,n) − e(m0 ,n) ,

e(m,n+n0 ) − e(m,n) − e(m,n0 ) ,
e(rm,n) − re(m,n) ,
e(m,rn) − re(m,n) ,
4. TENSOR PRODUCTS OF MODULES OVER COMMUTATIVE RINGS 63

where m, m0 ∈ M , n, n0 ∈ N and r ∈ R. Set T := F/I and consider the map

t : M × N → T, (m, n) 7→ e(m,n) := e(m,n) + I.
The map t is R-bilinear by construction. And given any R-bilinear map t : M × N → T 0
consider the R-linear map
e : F → T 0 , e(m,n) 7→ t(m, n).
ϕ
Since t0 is R-bilinear the map ϕe vanishes on I and induces and R-linear map ϕ : T → T 0
0
such that ϕ ◦ t = t . This condition already determines ϕ.
Definition 3.24. In the above situation one calls T the tensor product of M and N over
R and writes M ⊗R N := T . We occasionally write M ⊗ N if the ring R is clear from the
context. One also writes m ⊗ n = t(m, n). Elements of the form m ⊗ n are called pure
tensors.
The following corollary is immediate from the concrete model of the tensor product.
Corollary 3.25. Each w ∈ M ⊗R N can be written as sum of pure tensors
k
X
w= mi ⊗ ni ,
i=1

with mi ∈ M and ni ∈ N .
Remark 3.26. Let X and Y be sets of generators of M and N , respectively. Then {x ⊗ y |
x ∈ X, y ∈ Y } is a generating set of M ⊗ N . In particular, the tensor product preserves finite
generatedness.
Analogously, the general tensor product M1 ⊗ · · · ⊗ Mk can be defined as the universal
object of R-multilinear maps with source M1 × · · · × Mk .
Proposition 3.27. Let M, N, L be R-modules. Then there exist unique isomorphisms
(1) M ⊗ N ∼= N ⊗ M mapping m ⊗ n 7→ n ⊗ n.
(2) (M ⊗N )⊗L ∼
= M ⊗N ⊗L ∼
= M ⊗(N ⊗L) mapping (m⊗n)⊗` 7→ m⊗n⊗` 7→ m⊗(n⊗`).
(3) (M ⊕ N ) ⊗ L ∼
= (M ⊗ L) ⊕ (N ⊗ L) mapping (m ⊗ n, `) 7→ (m ⊗ `, n ⊗ `).
(4) R ⊗ M ∼
= M mapping r ⊗ m 7→ r · m.
Proof. These are all simple exercises using the universal property. We will prove the third
statement as an example:
(3) The map
(M ⊕ N ) × P → (M ⊗ L) ⊕ (N ⊗ L),
((m, n), `) 7→ (m ⊗ `, n ⊗ `)
is R-bilinear in (m, n) and `. Hence, there exists a unique morphisms of R-modules
ϕ : (M ⊕ N ) ⊗ L → (M ⊗ L) ⊕ (N ⊗ L) such that
(m, n) ⊗ ` 7→ (m ⊗ `, n ⊗ `).

For the inverse map first use the universal properties to construct the obvious morphisms
M ⊗ L → (M ⊕ N ) ⊗ L and N ⊗ L → L → (M ⊕ N ) and then take their product morphism
ψ : (M ⊗ L) ⊕ (N ⊗ L) → (M ⊕ N ) ⊗ L satisfying
(m ⊗ `, n ⊗ `) 7→ (m, 0) ⊗ ` + (0, n) ⊗ `.
The morphisms ϕ and ψ are mutually inverse on pure tensors and hence an all elements.
 64 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

Example 3.28.
(1) The R-linear map
∼
=
Rm ⊗ Rn → Rm×n ,
(x1 , . . . , xm ) ⊗ (y1 , . . . , yn ) 7→ (xi · yj )
is an isomorphism. In particular, ei ⊗ ej form a free basis of Rm ⊗ Rn .
(2) Let M be an R-module and I  R an ideal. Then the R-linear map
∼
=
M ⊗ R/I → M/IM,
m ⊗ (r + I) 7→ rm + IM
is an isomorphism.
(3) Let M, N, L be R-modules. Then the R-linear map
∼
=
HomR (M ⊗ N, L) → HomR (M, HomR (N, L))
(
M → HomR (N, L)
ϕ 7→ ϕc :
m 7→ (N → L, n 7→ ϕ(m ⊗ n))
is an isomorphism.
Remark 3.29. Let ϕ : M → N and ϕ0 : M 0 → N 0 be morphisms of R-modules. Then
M × M0 → N ⊗ N0
(m, m0 ) 7→ ϕ(m) ⊗ ϕ(m0 )
is R-bilinear. Hence, there exists a morphism
ϕ ⊗ ϕ0 : M ⊗ M 0 → N ⊗ N 0 such that
m ⊗ m0 7→ ϕ(m) ⊗ ϕ0 (m0 ).
End
lect. 17 This turns − ⊗R N and N ⊗R − into functors R-Mod → R-Mod.
06.07 Given a morphism ϕ : R → S of unital commutative rings, one can make S into an
R-module by setting
r · s =: ϕ(r) · s for all r ∈ R, s ∈ S.
Then the R-module structure is compatible with the ring structure of S, i.e., (r · s) · s0 =
r · (s · s0 ) for all r ∈ R, s, s0 ∈ S.
Definition 3.30. The ring S together with the above R-module structure is called an R-
algebra. An R-subalgebra of S is a subring S 0 ≤ S containing im ϕ. An R-algebra morphism
between two R-algebras S and T is a morphism α : S → T of rings which is an R-module
homomorphism. In terms of the ring homomorphisms ϕ : R → S and ψ : R → T defining
the R-algebra structures this means that α ◦ ϕ = ψ.
Example 3.31 (Tensor product of R-algebras). Let S, T be R-algebras defined by homo-
morphisms ϕ : R → S and ψ : R → T . Then the universal property of the tensor product
and Proposition 3.27 yield a multiplication on S ⊗R T such that
(s ⊗ t)(s0 ⊗ t0 ) = ss0 ⊗ tt0 .
This gives a unital ring with 1S⊗T = 1S ⊗ 1T . The ring homomorphism
R → S ⊗R T, r 7→ ϕ(r) ⊗ 1T = 1S ⊗ ψ(r)
turns S ⊗R T into an R-algebra.
 5. LOCALIZATION OF RINGS 65

Jump
Ch2§9 Exercise 3.4. Prove the converse of Proposition 2.87.1 for the category of R-modules,
i.e., the sequence M 0 → M → M 00 → 0 is exact iff the induced sequence

− HomA (M 00 , N ) → HomA (M, N ) → HomA (M 0 , N )

0→

is exact for all N ∈ A.

Proposition 3.32. The functors − ⊗R N and N ⊗R − are right exact for any R-module N .
Proof. Let M 0 → M → M 00 → 0 be an exact sequence. Then the induced sequence

0 → HomR (M 00 , HomR (N, L)) → HomR (M, HomR (N, L)) → HomR (M 0 , HomR (N, L))

is exact for all R-modules L by Proposition 2.87.1. Hence, by Example 3.28.3,

0 → HomR (M 00 ⊗ N, L) → HomR (M ⊗ N, L) → HomR (M 0 ⊗ N, L)

is exact. Now use Exercise 3.4. The right exactness of the other functor now follows from
Proposition 3.27.1.
Definition 3.33. An R-module N is called flat if the tensor functor − ⊗R N is exact.
Exercise 3.5.

(1) Show that the tensor functor is generally not exact in any of its arguments.

(2) Prove that free R-modules of finite rank are flat.

5 Localization of rings
Let R be a commutative unital ring.
Definition 3.34. A subset U ⊂ R is called multiplicatively closed, if it is closed under
(finite) products, i.e., if 1 ∈ U (as the empty product) and the product of two elements of U
is again in U .
Definition 3.35. Let U ⊂ R be a multiplicatively closed subset. We define an equiva-
lence relation on R × U by2

(r, u) ∼ (r0 , u0 ) : ⇐⇒ v(r0 u − ru0 ) = 0 for some v ∈ U .

r
We write u
for the equivalence class of (r, u) ∈ R × U and
nr o
−1 −1
U R := R[U ] := | (r, u) ∈ R × U
u
for the set of equivalence classes. To turn R[U −1 ] into a ring, we define addition and multi-
plication by

r r0 ru0 + r0 u
+ := ,
u u0 uu0
r r0 rr0
· := .
u u0 uu0
We call this ring the localization of R at U .
2
The element v in the definition is essential to guarantee the transitivity of the relation.
 66 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

The natural ring homomorphism ιU : R → R[U −1 ], r 7→ r

1
satisfies
• ιU sends elements of U to units in R[U −1 ].
• ιU sends r ∈ R to 0 iff there exists a v ∈ U such that vr = 0. Hence, ιU is injective iff U
does not contain a zero divisor. In particular, R[U −1 ] = 0 ⇐⇒ 0 ∈ U .
We call it the localization morphism of R at U .
Proposition 3.36 (Universal property of the localization). Let U ⊂ R be a multiplicatively
closed subset. If ϕ : R → S is a morphism of rings which maps elements of U to units in S, then
there exists a unique morphism Φ : R[U −1 ] → S such that Φ ◦ ιU = ϕ.
Proof.
r


Uniqueness: Suppose Φ satisfies the condition. Then Φ 1
= (Φ◦ιU )(r) = ϕ(r) for all r ∈ R. Hence


u −1 u −1
1
= ϕ(u)−1 for all u ∈ U . Hence, Φ is determined by ϕ



Φ u =Φ 1 =Φ 1
since r r 1
Φ =Φ Φ = ϕ(r) · ϕ(u)−1 for all r ∈ R, u ∈ U .
u 1 u

Existence: Define Φ ur := ϕ(r)ϕ(u)−1 for all r ∈ R, u ∈ U . Then Φ is a morphism of rings as

0
desired, once it is well-defined. To prove the latter let ur = ur 0 ∈ R[U −1 ]. Then there
exists v ∈ U such that v(ru0 − r0 u) = 0 and therefore

ϕ(v)(ϕ(r)ϕ(u0 ) − ϕ(r0 )ϕ(u)) = 0.

Since ϕ(v) is a unit in S we conclude that ϕ(r)ϕ(u)−1 = ϕ(r0 )ϕ(u0 )−1 .

End
lect. 18 Remark 3.37.
11.07
(1) The extension of the ideal I  R along ιU to R[U −1 ] is
nr o
I e := hιU (I)i = | r ∈ I, u ∈ U  R[U −1 ].
u
P ri ai e
Indeed, given i ui 1 ∈ I , with all ri ∈ R, ui ∈ U, ai ∈ I we can bring the sum to a
common denominator. The reverse inclusion is clear.
We will phrase this elegantly as
I e = I[U −1 ]
in Remark 3.44.
(2) The contraction of the ideal J  R[U −1 ] along ιU to R is
n r o
J c
:= ιU (J) = r ∈ R | ∈ J  R.
−1
1

Theorem 3.38. Let U ⊂ R be a multiplicatively closed subset and ιU : R → R[U −1 ] the

localization morphism at U .
(1) If I  R is an ideal, then
I ec = {r ∈ R | ∃u ∈ U : ur ∈ I}

(2) If J  R[U −1 ] is an ideal, then

J ce = J.
This yields an injective map from the set of ideals of R[U −1 ] into the set of ideals of R by
mapping J to J c .
5. LOCALIZATION OF RINGS 67

(3) The injection J 7→ J c restricts to a bijection

∼
Spec R[U −1 ] −
→ {p ∈ Spec R | p ∩ U = ∅} ⊂ Spec R.

Proof.

(1) Let r ∈ R. Then

r 3.37.1 r s
r ∈ I ec ⇐⇒ ∈ I e ⇐⇒ = for some s ∈ I, u ∈ U
1 1 u
⇐⇒ ur ∈ I for some u ∈ U .

(2) We have J ce ⊂ J by Remark 3.5.2. For the reverse inclusion let r ∈ R, u ∈ U . Then
r r r
∈ J =⇒ ∈ J =⇒ r ∈ J c =⇒ ∈ J ce .
u 1 u

(3) Let q ∈ Spec R[U −1 ]. Then p := qc := ι−1

U (q) ∈ Spec R. Furthermore, p ∩ U = ∅ since q
does not contain units.
Conversely, let p ∈ Spec R such that p ∩ U = ∅. Then pe is a proper ideal in R[U −1 ]. If
r s
uv
∈ pe with u, v ∈ U , then wrs ∈ p for some w ∈ U by Remark 3.37.1. Then w ∈ / p
r e s e
since p∩U = ∅ and hence r ∈ p or s ∈ p. But this means u ∈ p or v ∈ p . It follows that
pe is a prime ideal. Furthermore, by 1 we have pec = {r ∈ R | ∃w ∈ U : wr ∈ p} = p.
This concludes the proof.

Example 3.39.

(1) Let U be the set of non-zero-divisors of R. Then Q(R) := R[U −1 ] is called the total
quotient ring of R. In the special case where R is an integral domain, Q(R) is a field,
the quotient field or field of fractions of R.
h i
(2) Let f ∈ R and U = {f m | m ≥ 0}. Then Rf := R f1 := R[U −1 ].

(3) If p ∈ Spec R then U := R \ p is multiplicatively closed. The localization Rp := R[U −1 ]

is called the localization of R at p. Note that Rp is a local ring with maximal ideal
nr o
e
m=p = | r ∈ p, u ∈ U := R\p .
u
Indeed, if ur ∈ Rp \ m, then r ∈ R \ p, so r
u
is a unit in Rp . Hence Rp \ m = Rp∗ , i.e., m is
the unique maximal ideal of Rp .

Remark 3.40. For p ∈ Spec R denote by πp : R  R/p the natural epimorphism and by
ιR\p : R → Rp the localization morphism.

• The map πp−1 : Spec R/p → Spec R, q 7→ qc is an embedding. Its image is {p0 ∈ Spec R |
p0 ⊃ p}.

• The map ι−1 c 0

R\p : Spec Rp → Spec R, q 7→ q is an embedding. Its image is {p ∈ Spec R |
p0 ⊂ p}.

Geometrically speaking, the passage to the residue class ring R/p means to restriction to the
affine subscheme Spec R/p of Spec R, whereas the passage to Rp considers the infinitesimal
neighborhood Spec Rp of the subscheme Spec R/p.
68 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

Example 3.41. By inverting all elements in Z \ {0} we get Q, as we already know. By

inverting small subsets we get subrings of Q. For example, if n ∈ Z \ {0}, we get
  n
1 a o
Z = ∈ Q | b = nk for some k ∈ N ,
n b
or, if p ∈ Z is a prime number, we get
na o
Zhpi = ∈Q|p-b .
b
If p - n, we get the chain of inclusions of rings
 
1
Z<Z < Zhpi < Q.
n

6 Localization of modules
Definition 3.42. Let U be a multiplicatively closed subset of R and M an R-module. We
get an equivalence relation on M × U by setting

(m, u) ∼ (m0 , u) : ⇐⇒ ∃v ∈ U : v(u0 m − um0 ) = 0.

We write nm o
−1 −1
U M := M [U | m ∈ M, u ∈ U
] :=
u
for the set of equivalence classes and make M [U −1 ] into an R[U −1 ] module with addition as
for R[U −1 ] and scalar multiplication ur · um0 := u·u
r·m
0 . The R[U
−1
]-module M [U −1 ] is called the
localization of M at U .
Remark 3.43. A morphism ϕ : M → N of R-modules induces a morphism

ϕ[U −1 ] : M [U −1 ] → N [U −1 ]
m ϕ(m)
7→
u u
of R[U −1 ]-modules. Indeed, localization at U is a (covariant) functor R-Mod → R[U −1 ]-Mod:
(1) idM [U −1 ] = idM [U −1 ] ;
ϕ ψ
(2) If M 0 − → M 00 are morphisms of R-modules, then (ψ ◦ ϕ)[U −1 ] = ψ[U −1 ] ◦ ϕ[U −1 ].
→M −
From Remark 3.37.1 we deduce:
Remark 3.44. Let I  R and ιU : R → R[U −1 ] the localization of R at the multiplicatively
closed subset U ⊂ R. Then
I e = I[U −1 ].
ϕ
Proposition 3.45. The localization functor R-Mod → R[U −1 ]-Mod is exact, i.e., if M 0 −
→
ψ ϕ[U −1 ψ[U −1 ]
M− → M 00 is an exact sequence of R-modules, then M 0 [U −1 ] −−−→ M [U −1 ] −−−−→ M 00 [U −1 ] is an
exact sequence of R[U −1 ]-modules.
Proof. From ψ[U −1 ] ◦ ϕ[U −1 ] = (ψ ◦ ϕ)[U −1 ] = 0 we know that im ϕ[U −1 ] ≤ ker ψ[U −1 ]. We
now show the reverse inclusion. Let m u
∈ ker ψ[U −1 ]. Then
m ψ(m)
−1
0 = ψ[U ] = .
u u
6. LOCALIZATION OF MODULES 69

Hence, there exists a v ∈ U such that 0 = v · ψ(m) = ψ(v · m). But then v · m ∈ ker ψ = im ϕ
and there exists an m0 ∈ M 0 with ϕ(m0 ) = v · m and we conclude that
ϕ(m0 )
 0
m vm −1 m
= = = ϕ[U ] ∈ im ϕ[U −1 ].
u vu vu vu
In particular, the localization preserves monos: So if N is an R-submodule of M then
N [U −1 ] is an R[U −1 ]-submodule of M [U −1 ].
Corollary 3.46. Let N and L be R-submodules of M . Then
(1) (N + L)[U −1 ] = N [U −1 ] + L[U −1 ];
(2) (N ∩ L)[U −1 ] = N [U −1 ] ∩ L[U −1 ];
(3) (M/N )[U −1 ] ∼
= M [U −1 ]/N [U −1 ].
Proof. This follows by abstract nonsense about exact functors between A BELian categories.
End
Note that the localization morphism ιU : R → R[U −1 ] endows R[U −1 ] with an R-algebra lect. 20
structure, and in particular with an R-module structure. 13.07
Proposition 3.47. The localization functor at U is naturally isomorphic to the tensor functor
R[U −1 ] ⊗R −. More precisely, there exists a unique morphism
ϕM : R[U −1 ] ⊗R M → M [U −1 ] such that
r rm
⊗ m 7→ .
u u
Proof. Since the map R[U −1 ]×M → M [U −1 ], ( ur , m) 7→ rm u
is R-bilinear, there exists a unique
morphism ϕM sending ur ⊗ m to rm u
. This morphism
P ri is clearly surjective.
We show that ϕM is also injective. For this let i ui ⊗ mi ∈ R[U −1 ] ⊗R M be an element. Set
u := i ui ∈ U , vi := j6=i uj . Then i urii ⊗mi = i riuvi ⊗mi = i u1 ⊗ri vi mi = u1 ⊗ i ri vi mi .
Q Q P P P P

It follows that any element of R[U −1 ] ⊗ M is of the form u1 ⊗ m for some u ∈ U and m ∈ M .
Now suppose ϕM ( u1 ⊗ m) = 0, i.e., suppose m u
= 0. Then there exists a v ∈ U such that
vm = 0. Hence
1 v 1 1
⊗m= ⊗m= ⊗ vm = ⊗ 0 = 0.
u uv uv uv
Corollary 3.48. R[U −1 ] is a flat R-module.
Proof. R[U −1 ] ⊗R − is by Proposition 3.47 isomorphic to the localization functor, and the
latter is exact by Proposition 3.45.
Proposition 3.49. Let M and N be R-modules. Then there exists a unique isomorphism of
R[U −1 ]-modules
ϕM,N : M [U −1 ] ⊗R[U −1 ] N [U −1 ] → (M ⊗R N )[U −1 ] such that
m n m⊗n
⊗ 7→ .
u v u·v
Proof. This follows from Proposition 3.47 using the canonical isomorphism of Proposi-
tion 3.27.4.
Let M, N be R-modules, ϕ : M → N an R-linear map, p ∈ Spec R, and U = R \ p. Then
we write
Mp := M [U −1 ] and ϕp := ϕ[U −1 ].
If N is another R-module, then the previous proposition reads
Mp ⊗R Np ∼= (M ⊗R N )p
p

as Rp -modules.
 70 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

7 Local properties
We now study local properties of R-modules, i.e., a property that an R-module M has iff all
its localizations at prime ideals has the same property. This is part of the so-called “local-
global-principle”.
The first local property is “being zero”:
Proposition 3.50. Let M be an R-module. Then the following are equivalent:
(1) M = 0;
(2) Mp for all p ∈ Spec R;
(3) Mm for all m ∈ Max R ⊂ Spec R.
Proof. The implications 1 =⇒ 2 =⇒ 3 are obvious. We prove the remaining direction by
contraposition:
Suppose M 6= 0 and m ∈ M \ {0}. Then the annihilator AnnR (m) is a proper ideal of R
(1 · m = m 6= 0). Thus, there is an m ∈ Max R containing Ann(m). But then the element
m
1
∈ Mm is nonzero, for otherwise there would exist some u ∈ R \ m such that um = 0. This
u would then be contained in AnnR (m) ⊂ m. A contradiction. Hence, Mm 6= 0 and we are
done.
The next local property is “being a mono”:
Proposition 3.51. Let M, N be R-modules and ϕ : M → N an R-linear map. Then the
following are equivalent:
(1) ϕ is a mono;
(2) ϕp is a mono for all p ∈ Spec R;
(3) ϕm is a mono for all m ∈ Max R.
Proof.
ϕ
1 ⇒ 2: That ϕ is mono means that 0 → M − → N is an exact sequence. By the exactness of the
ϕp
localization (Proposition 3.45) the sequence 0 → Mp −→ Np is exact for all p ∈ Spec R.
2 ⇒ 3: Trivial.
3 ⇒ 1: Apply the exactness of the localization to the exact sequence
ϕ
0 → ker ϕ → M −
→N

and use Proposition 3.50.

Remark 3.52. The statement of the Proposition 3.51 if we replace mono by epi or by
isomorphism. The proof is analogous.
Finally, flatness is a local property.
Proposition 3.53. For any R-module M the following are equivalent:
(1) M is a flat R-module.
(2) Mp is a flat Rp -module for all p ∈ Spec R.
(3) Mm is a flat Rm -module for all m ∈ Max R.
Proof. Exercise.
8. CHAIN CONDITIONS 71

8 Chain conditions
We have already encountered an instance of Hilbert’s basis theorem for polynomial rings
over a field as Corollary 1.25.
Theorem 3.54 (Hilbert’s basis theorem). If R is a Noetherian ring, then the polynomial
ring R[x1 , . . . , xn ] is Noetherian as well. In particular, the polynomial rings Z[x1 , . . . , xn ] and
K[x1 , . . . , xn ] (where K is a field) are Noetherian rings.
Proof. Assume I R[x] is not finitely generated. Then, inductively, we can choose elements:
f1 ∈ I \ {0}, f2 ∈ I \ hf1 i, f3 ∈ I \ hf1 , f2 i, ... of minimal possible degree. For all i, set
di := deg fi and write fi = ai xdi + lower order terms in x. Then d1 ≤ d2 ≤ · · · and ha1 i ⊂
ha1 , a2 i ⊂ · · · is an ascending chain of ideals in R which eventually becomes stationary by
assumption on R, i.e., there exists a k ∈ N such that

ha1 , . . . , ak i = ha1 , . . . , ak+1 i.

Pk
That is, ak+1 = i=1 bi ai for some bi ∈ R. Now consider
k
X
g := fk+1 − bi xdk+1 −di fi
i=1
k
X
dk+1
= ak+1 x − bi ai xdk+1 + lower order terms.
i=1

Since fk+1 ∈ I \ hf1 , . . . , fk i, also g ∈ I \ hf1 , . . . , fk i. This contradicts the choice of fk+1 since
deg(g) < dk+1 := deg(fk+1 ).
Proposition 3.55. Let R be a Noetherian ring.
(1) If I  R is an ideal, then R/I is Noetherian.

(2) If U ⊂ R is a multiplicatively closed subset, then R[U −1 ] is Noetherian.

Proof.
(1) The contraction to R along the epimorphism R  R/I is an order preserving an
embedding. Its image are all those ideals containing I.

(2) The extension to R[U −1 ] along the localization morphism R → R[U −1 ] is an order
preserving embedding by Theorem 3.38.2.

9 Artinian modules and rings

Recall that an R-module is simple iff it is an epimorphic image of R modulo a maximal
ideal.
Definition 3.56. Let M be an R-module. We say that M is of finite length if it has a
composition series. We call the length of such a composition series the length of M and
denote by `(M ). If no composition series exist we set `(M ) := ∞.
By Jordan-Hölder we know that any two such composition series have the same length.
Example 3.57. Let K be a field and M a K-vector space. Then `(M ) = dimK M .
We have proven in the Algebra lecture (Folgerung 4.9) the equivalence of the following
two statements:
 72 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

(1) M is of finite length.

(2) M is Artinian and Noetherian.

Proposition 3.58 (Additivity of the length). Let R be a ring and 0 → M 0 → M → M 00 → 0

a short exact sequence of R-modules. Then

`(M ) = `(M 0 ) + `(M 00 ).

Proof. Trivial.

Corollary 3.59. Let M and N be R-modules. Then `(M ⊕ N ) = `(M ) + `(N ).

In other words, the length function regards any short exact sequence as a split exact
sequence.
Definition 3.60. Let R be a ring. The Krull dimension of R, written dim R, is the supre-
mum of the lengths m of strict chains of prime ideals

p0 ( p1 ( p2 ( · · · pm .

Theorem 3.61. Let R be a ring. The following statements are equivalent:

(1) R is Noetherian and dim R = 0;

(2) R has finite length (as an R-module).

(3) R is Artinian.

If these conditions are satisfied, then R has only finitely many maximal ideals.
End
lect. 21
18.07 Proof.
1 ⇒ 2:

Step 1: Suppose that R is Noetherian but not of finite length. Then the set

Γ := {I
R | R/I not of finite length}

is nonempty since h0i ∈ Γ. Hence, Γ contains a maximal element p. We will now

prove that p is prime: Let f, g ∈ R such that f g ∈ p but f ∈
/ p. Consider the short
exact sequence

·f
0 → R/(p : f ) −
→ R/p → R/(p + hf i) → 0.

Since p + hf i ) p, the module R/(p + hf i) has finite length by the choice of p.

Suppose g ∈ / p. Then p : f ) p and R/(p : f ) has finite length as well. Then R/p
has finite length by Proposition 3.58. A contradiction to p ∈ Γ.
Step 2: Suppose now in addition that dim R = 0. Then all prime ideals in R are maximal.
Assume R is not of finite length. Then we have a prime ideal p as above which
is necessarily maximal. Hence, R/p is a field. Again a contradiction to p ∈ Γ.

2 ⇒ 3: We already know that finite length modules are Noetherian and Artinian.

3 ⇒ 1: Suppose that R is Artinian.

9. ARTINIAN MODULES AND RINGS 73

Step 1: We first show that dim R = 0: Consider prime ideals p1 ⊂ p2 ⊂ R and let f ∈
p2 /p1 ( R/p1 . Since the residue class ring R/p1 is also Artinian, there exists
m ≥ 1 such that hf m i = hf m+1 i. Then f m = gf m+1 for some g ∈ R/p1 . That
is, (1 − gf )f m = 0. Since R/p1 is an integral domain and f ∈ p2 /p1 ( R/p1 is
not a unit, we must have f = 0. Hence p1 = p2 . This proves that dim R = 0, or
equivalently Spec R = Max R.
Step 2: Since R is Artinian, it has only finitely many maximal ideals, as any infinite col-
lection m1 , m2 , . . . of pairwise distinct maximal ideals would yield an infinite de-
scending chain of ideals m1 ⊃ m1 ∩ m2 ⊃ · · · ⊃ m1 ∩ . . . ∩ mk ⊃ · · · , with strict
inclusions. Writing m1 , . . . , ms for the distinct ideals of Max R we have
s
\
p
h0i = N (R) = mi (*)
i=1

since Spec R = Max R by step 1.

Step 3: For any i, the sequence mi ⊃ m2i ⊃ m3i ⊃ · · · eventually becomes stationary.
N +1
Hence, there exists an N ∈ N such that mN
i = mi for all i. Consider the ideal
s
Y
I= i
R.
mN
i=1

Then I 2 = I. We will use this to show that I = h0i. Suppose the contrary. Then
the set
Γ := {J
R | J · I 6= h0i}
contains I since I · I = I 6= h0i by the assumption. In particular, Γ is nonempty.
Hence, since R is Artinian, Γ contains a minimal element J0 . Now pick f ∈ J0
such that f · I 6= h0i. Then hf i = J0 by the minimality of J0 . The same argument
gives f I = J0 = hf i since (f I) · I = f I 2 = f I 6= h0i. Now pick g ∈ I such that
f · g = f . Then
f = f g = f g2 = · · · = f gm = 0
for some m ≥ 0 since each element of I is nilpotent by (*). This contradicts our
choice of f . Hence, I = h0i.
Step 4: Each factor in the chain
s
Y
R ⊃ m1 ⊃ · · · ⊃ mN
1 ⊃ mN
1 m2 ⊃ ··· ⊃ mN
i = h0i (**)
i=1

is a vector space over some field R/mi . Hence we have:

R is Artinian ⇐⇒ each factor in (*) is Artinian

⇐⇒ each factor in (*) is Noetherian
⇐⇒ R is Noetherian.

This concludes the proof.

Theorem 3.62 (Structure Theorem for Artinian rings). Every Artinian ring R is a finite
product ofQ
local Artinian rings. More precisely, if m1 , . . . , ms are the distinct maximal ideals of R,
then R ∼ s
= i=1 Rmi .
 74 CHAPTER 3. LOCALIZATION AND TENSOR PRODUCTS

Proof. We know that the localizations

Qs RmN i
are all Artinian as well. Choose, as in the pre-
ceding proof, an N ∈ N such that i=1 mi = h0i. We now show that the mN i are pairwise
coprime. Suppose the contrary, i.e., suppose there exists a pair i 6= j with mNi + mj
R.
N

Then there exists a maximal ideal m
R such that mN N

i + mj ⊂ m. It follows that mi , mj ⊂ m
since m is prime. Hence mi = m = mj , contradicting i 6= j. We may thus apply the Chinese
remainder theorem from Algebra to conclude that the ring homomorphism
s
Y
ϕ:R→ R/mN N N
i , r 7→ (r + m1 , . . . , r + ms )
i=1

is an isomorphism. Note that R/mN i is a local Artinian ring (whose maximal ideal is the
image of mi in R/mN
i ). Moreover, by localizing ϕ we get

Rmi ∼
= (R/mN ∼ N
i )mi = R/mi

since (R/mN
j )mi = 0 for j 6= i.
End
lect. 22
20.07
Chapter 4

Primary decomposition

In Z any element can be written uniquely as the product

mk
z = upm
1 · · · pk ,
1

with a unit u ∈ Z× = {±1}, pi prime numbers, and mi > 0 for i = 1, . . . , k. This implies
mk
hzi = hpm
1 i ∩ . . . ∩ hpk i.
1

We call such an intersection the primary decomposition of the ideal hzi. Since Z is a princi-
pal ideal domain any primary decomposition is already of this form.
In the attempt to prove Fermat’s Last Theorem mathematicians discovered subrings R
of the complex numbers which are, unlike Z, not unique factorization domains. We call
them nowadays Dedekind domains. In these rings the notion of a primary decomposition
of a ring element fails. For example
√ √ √
Z[ −5] 3 6 = 2 · 3 = (1 + −5)(1 − −5).
√ √ √
The
√ numbers 2, 3, 1 + −5, 1 − −5 are irreducible in Z[ −5] but obviously not prime in
Z[ −5]. In order to rectify this failure Dedekind introduced ideals which for this class of
rings still admit a primary decomposition. This is even the historical reason for the name
“ideal”: unlike numbers in R, ideals behave as “ideal numbers” as they admit a primary
decomposition:
√ D √ E D √ E D √ E D √ E
Z[ −5]  h6i = 2, 1 + 5 ∩ 3, 1 + 5 ∩ 2, 1 − 5 ∩ 3, 1 − 5 .
| {z } | {z }
√ √
=h1+ −5i =h1− −5i

Primary decompositions of ideals can be generalized to N OETHERian rings.

1 Definition and Existence in N OETHERian rings

Definition 4.1. Let R be a ring. A proper ideal q
R is called a primary ideal if for
f, g ∈ R
√
f · g ∈ q =⇒ f ∈ q or g ∈ q.
Proposition 4.2. Let R be a ring. Then
√
(1) If q
R is a primary ideal, then p := q is the smallest prime ideal containing q. We say that
q is p-primary.

(2) A finite intersection of p-primary ideals is p-primary.

75
76 CHAPTER 4. PRIMARY DECOMPOSITION
√ √
(3) If I
R is any ideal such that I is a maximal ideal, then I is I-primary. In particular, the
powers of a maximal ideal m
R are m-primary.

Proof.
√
(1) By Proposition 3.10 it suffices to show that p := q is prime. Let f, g ∈ R such that
√
f g ∈ q. Then there exists some m ≥ 1 with f m g m = (f g)m ∈ q. Hence, either f m ∈ q
√ √ √
or g mk ∈ q for some k ≥ 1. That is, either f ∈ q or g ∈ q, so q is prime.

(2) Let q1 , . . . , qr be p-primary and q := ri=1 qi their intersection. Then1

T

v
u r r
√ u\ \ √
q= t qi = qi = p.
i=1 i=1

Now let f, g ∈ R such that f g ∈ q but f ∈

/ q. Then f g ∈ qi but f ∈
/ qi for some i. Hence
√ √
g ∈ qi = p = q since qi is p-primary.
√
(3) Set m := I. Then the image m/I of m in R/I is the nilradical
\
N (R/I) = p.
p∈Spec(R/I)

of that ring. In particular, R/I is a local ring with maximal ideal m/I. Then each
element of R/I is either a unit or nilpotent, i.e., every zero-divisor in R/I is nilpotent.
In other words, I is m-primary.

Example 4.3.

(1) The primary ideals in the ring Z of integers are h0i and ideals of type hpn i, where p is
a prime number and n ≥ 1. Indeed they are the only ideals of Z whose radical is a
prime ideal and they are clearly primary.
√
(2) Let R = k[x, y] where k is a field and q = hx, y 2 i. Then p := q = hx, yi is maximal and
R/q ∼= k[y]/hy 2 i. In R/q the zero-divisors are the multiples of y and hence nilpotent.
Thus q is p-primary. We have p2
q
p, so a primary ideal is not necessarily a prime
power!

(3) In contrast to the case of maximal ideals, a prime power is not necessarily prime: Consider
R = k[x, y, z]/hxy − z 2 i, where k is a field. Write x, y, z for the residue classes of
x, y, z. Then p := hx, zi
R is prime
√ since R/p ∼= k[y] is an integral domain. We have
2 2 2
x · y = z ∈ p , but x ∈ / p and y ∈/ p = p. Hence, p2 is not primary.
2

Definition 4.4. Let R be a ring and I  R. A primary decomposition (PD) of I is an

expression
I = q1 ∩ . . . ∩ qt
withT
primary ideals qi . The decomposition is minimal if the radicals qi are pairwise distinct
and i6=j qi 6⊆ qj for all j.
There exists ideals in non-N OETHERian rings which do not admit a primary decompo-
sition. For N OETHERian rings everything is fine:
Theorem 4.5. Each (proper) ideal of a N OETHERian ring R has a minimal primary decompo-
sition.
1
check that taking the radical commutes with forming intersections.
 2. UNIQUENESS RESULTS 77

Nonconstructive proof by contradiction and Zorn’s Lemma. Set

Γ := {I
R | I is has no primary decomposition}.

Suppose Γ 6= ∅. Then Γ has a maximal element I0 . In √particular, I0 is not primary. That is,
there exists f, g ∈ R with f g ∈ I0 but f ∈/ I0 and g ∈ / I0 . The ascending chain condition in
N OETHERian rings applied to
I0 : g ⊂ I0 : g 2 ⊂ . . .
shows that I0 : g m = I0 : g m+1 for some m ≥ 1. Then

I0 = (I0 : g m ) ∩ hI0 , g m i.

by simple exercise. Since f g ∈ I0 , also f g m ∈ I0 , which √ means the f ∈ I0 : g m . Then

I0 ( I0 : g m since f ∈ / I0 . Also I0 ( hI0 , g m i since g ∈
/ I0 . It follows that both I0 : g m and
hI0 , g m i are not in Γ, so both ideals have a primary decomposition. But then I0 must have a
primary decomposition as well, combining the PDs of I0 : g m and hI0 , g m i. This contradicts
the choice of I0 .
Now given a proper ideal I
R we now know that I has a PD. If two of the primary
ideals occurring in the PD have the same radical p, we replace them by their intersection
which is again p-primary by Proposition 4.2.(2). Continuing this way we get a PD with
pairwise distinct radicals. Throwing away superfluous primary ideals we get a minimal
primary decomposition.
Not all ideals in a primary decomposition are uniquely determined by I, as the follow-
ing example shows:
Example 4.6. Both decompositions

hxy, y 2 i = hyi ∩ hx, y 2 i = hyi ∩ hx2 , xy, y 2 i

are minimal primary decompositions. In fact, by Proposition 4.2.(3) both ideals hx, y 2 i and
hx2 , xy, y 2 i are hx, yi-primary. Moreover, hx, yi ⊃ hyi.
End
Nevertheless, there are certain “parts” of PDs that are indeed uniquely determined by lect. 23
I. We will study them in the next section. 18.10

2 Uniqueness results
Lemma 4.7. Let R be a ring and f ∈ R. Then:
(1) If I  R is an ideal with f ∈ I then I : f = R.
Let now q
R be a primary ideal with radical p ∈ Spec R. Then:
(2) If f ∈
/ p then q : f = q.

(3) If f ∈
/ q then q : f is p-primary.
Proof.
(1) & (2) follow immediately from the definitions.

(3) If ge ∈ q : f then gef ∈ q. And since f ∈ / q by assumption, we have ge√∈ p (as q is

√ √
p-primary).
√ This shows that q ⊂ q : f ⊂ p. Taking radicals, we get q ⊂ q : f ⊂ p,
so q : f = p. Let now g, h ∈ R be such that gh ∈ q : f with h ∈ / p. Then ghf ∈ q,
hence gf ∈ q, so that g ∈ q : f .
 78 CHAPTER 4. PRIMARY DECOMPOSITION

TheoremTt 4.8 (1st Uniqueness Theorem). Let R be a ring, I  R an ideal admitting a PD,
√
and I = i=1 qi a minimal PD with pi := qi ∈ Spec √ R for all i = 1, . . . , t. Then the pi ’s are
precisely the prime ideals occurring in the set of ideals I : f where f ∈ R.
Hence, the pi ’s are independent of the choice of the PD.
Tt
Proof. Let f ∈ R. We already know that I : f = i=1 (qi : f ), hence

t
4.7
p \ p \
I:f = qi : f = pj .
i=1 f ∈q
/ j

√ √
√ I : f is prime, we conclude by Exercise 3.3 that pj ⊂ IT: f for some j and thus pj =
If
I : f . Conversely, for all i = 1, . . . , t there exists some fi ∈ j6=i qj such that fi ∈
/ qi since
√
the PD is minimal. Again by Lemma 4.7 we get I : fi = pi .

In short, the First Uniqueness Theorem states that even if the primary ideals of a PD are
not uniquely determined, their radicals are. Each ideal I thus has its “own” set of prime
ideals occurring as radicals in its primary decomposition. We should give them a name in
order to address them in the future.
Definition 4.9. Let R be a ring and I  R an ideal. We call
p
Ass(I) := {p ∈ Spec R | ∃f ∈ R : I : f = p}

the set of associated primes of I and

minAss(I) := {p ∈ Ass(I) |6 ∃p0 ∈ Ass(R) : p0 ( p}

the set of minimal associated primes or isolated primes of I. Furthermore, each prime in
Ass(I) − minAss(I) is called an embedded prime of I.
The names “isolated” and “embedded” are of geometric origin.
Proposition 4.10. Let R be a ring and I  R an ideal admitting a PD. Then each prime p ⊃ I
contains a minimal associated prime of I. Thus, the primes in minAss(I) are precisely the primes
containing I which are minimal w.r.t. the inclusion. In particular,

(1) if h0i admits a primary decomposition, then

p \
N (R) = h0i = p
p∈minAss(h0i)

(2) and each N OETHERian ring has only finitely many minimal primes.
Tt
Proof. If p ⊃ I = i=1 qi then

t t
√ \√ \
p= p⊃ qi = pi .
i=1 i=1

And again by Exercise 3.3 we get p ⊃ pi for some i.

End
lect. 24 While the proposition above addresses the intersection of the minimal primes in a ring
25.10 where h0i admits a primary decomposition, we now discuss their union:
2. UNIQUENESS RESULTS 79

Proposition 4.11. Let R be a ring and I  R be an ideal admitting a primary decomposition.

Tt √
Let I = i=1 qi be a minimal PD with pi := qi for all i = 1, . . . , t. Then
t
[ [p
pi = I : g = {f ∈ R | I : f ) I} = {f ∈ R | f ∈ R/I is a zero-divisor}.
i=1 g ∈I
/

In particular, if h0i is admits a PD, then

[
{f ∈ R | f is a zero-divisor} = p.
p∈minAss(h0i)
S √
Proof. We first show that {f ∈ R | f ∈ R/I is a zero-divisor} = g ∈I
/ I : g.
“⊂” Let f be in the set on the left √hand side. Then there exists a g ∈ / I such that f g ∈ I,
which means that f ∈ I : g ⊂ I : g.
√
“⊃” Let f be in the set on the right hand side. Then there exists a g ∈
/ I such that f ∈ I : g.
That is, f m ∈ I : g for some m ≥ 1, so f m g ∈ I. We choose m minimal with this
property. Then f · (f m−1 · g) ∈ I, so f ∈ R/I is a zero-divisor.
| {z }
∈I
/

Next we show that

t
[ [p
pi = I : g.
i=1 g ∈I
/

“⊂” This inclusion is clear from the Uniqueness Theorem 4.8.

/ I = ti=1 qi . Then g ∈
T
“⊃” Let g ∈ / qj for some j. It follows that
t
p \ √ √ 4.7.(3) √
I:g= qi : g ⊂ qj : g = qj = pj
i=1

Now we study the behavior of primary ideals under localization. As an application, we

will show another Uniqueness Theorem.
Proposition 4.12. Let R be a ring, U ⊂ R a multiplicatively closed subset, and q
R a p-
primary ideal. Consider qe = q[U −1 ]  R[U −1 ]. Then
(1) if U ∩ p 6= ∅ then q[U −1 ] = R[U −1 ];
(2) if U ∩ p = ∅ then q[U −1 ] is p[U −1 ]-primary and qec = q.
Proof.
√
(1) Let a ∈ U ∩ p 6= ∅. Then am ∈ q for some m ≥ 1 as p = q. And since U is
m
multiplicatively closed we have am ∈ U ∩ q. But then q[U −1 ] contains then unit a1 .
(2) Now assume U ∩ p = ∅. Then u ∈ U and au ∈ q imply a ∈ q. Hence qec = {a ∈
R | ∃u ∈ U : ua ∈ q} = q, where the first equality is Theorem 3.38.1. Furthermore,
√ e 3.44 p √
q = q[U −1 ] = q[U −1 ] = p[U −1 ]. That q[U −1 ] is primary, is immediate from the
definitions.
Proposition 4.13. Let R be a ring, U ⊂ R aTmultiplicatively closed subset, and I  R an ideal
admitting a primary decomposition. Let I = ti=1 qi be a minimal primary decomposition with
√
qi = pi for all i = 1, . . . , t. Suppose that
Ts the qi−1
’s are numbered such
Ts that U meets ps+1 , . . . , pt ,
e −1 −1 c
but not p1 , . . . , ps . Then I = I[U ] = i=1 qi [U ] and I[U ] = i=1 qi and these are minimal
primary decompositions.
80 CHAPTER 4. PRIMARY DECOMPOSITION

Proof. We have
t s
−1 3.46 −1 4.12
\ \
I[U ] = qi [U ] = qi [U −1 ]
i=1 i=1

and qi [U −1 ] is pi [U −1 ]-primary for all i = 1, . . . , s. Since the pi ’s are pairwise distinct, so

are the pi [U −1 ]’s, hence we have a minimal primary decomposition. Contracting both side
yields
s s
−1 c 4.12
\ \
−1 c
I[U ] = qi [U ] = qi .
i=1
| {z } i=1
qec
i

Definition 4.14. Let R be a ring, I
R a proper ideal, and Σ ⊂ Ass(I). We call Σ isolated
if it is inclusion-closed, i.e., p0 ∈ Ass(I) and p0 ⊂ p for some p ∈ Σ implies p0 ∈ Σ.
Theorem 4.15 (2nd Uniqueness Tt Theorem). Let R be a ring and I  R an ideal admitting √
a
primary decomposition. Let I = i=1 qi be a minimal primary decomposition T with pi := qi for
all i = 1, . . . , t. If Σ ⊂ Ass(I) is an isolated set of primes of I, then pi ∈Σ qi is independent of the
chosen primary decomposition. In particular, the primary components corresponding to primes in
minAss(I) are uniquely determined by I.
S T
Proof. Set UΣ := R − pi ∈Σ pi = pi ∈Σ (R − pi ). Then UΣ is a multiplicatively closed subset
of R. Furthermore, we have for all pj :
[
pj ∩ UΣ = ∅ ⇐⇒ pj ⊂ pi
pi ∈Σ
3.8
⇐⇒ pj ⊂ pi for some i.

The result follows from Proposition 4.13:

\
I[UΣ−1 ]c = qi .
pi ∈Σ

Definition 4.16. Let I
R be a proper ideal of the ring R. Then we call each primary
ideal in a primary decomposition of I corresponding to a prime in minAss(I) and isolated
primary component. Each other primary ideal occurring in any minimal primary decom-
position of I is called an embedded primary component.
Chapter 5

Integral ring extensions

1 Basic definitions
We start by recalling basic results about integral ring extensions from the Algebra lecture.
Definition 5.1. Let S be a ring (not necessarily commutative), R a subring in the center
Z(S) and I  R.

• We call S/R a ring extension, in particular S is an R-algebra.1

• An element s ∈ S is called integral over R if it is a root of a monic polynomial

X d + r1 X d−1 + · · · + rd ∈ R[X],

called an integral equation for s.

More generally:

• An element s ∈ S is called integral over I if it is a root of a monic polynomial

X d + r1 X d−1 + · · · + rd ∈ R[X]

with ri ∈ I for all i = 1, . . . , d.

• S is called integral over R or the ring extension R/S is integral, if each element of S
is integral over R.

• S is called of finite type over R if it is finitely generated as an R-algebra, i.e., if there

exists finitely many elements s1 , . . . , sn ∈ S such that S = Rhs1 , . . . , sn i.2

• S is called finite over R, if S is finitely generated as an R-module.

Remark 5.2. Integral ring extensions generalize algebraic field extensions.

Theorem 5.3 (K RONECKER). Let S/R be a ring extension and I  R. For s ∈ S the following
are equivalent:

(1) s is integral over R (over I);

1
For an R-algebra we only assume the existence of a ring homomorphism R → Z(S), which we do not
require to be injective in general.
2
I.e., S is the image of the evaluation homomorphism

ϕ : RhX1 , . . . , Xn i → S, Xi 7→ si

with ϕR = idR .

81
 82 CHAPTER 5. INTEGRAL RING EXTENSIONS
p
(2) R[s] is finite over R (and s ∈ I · R[s]);
√
(3) R[s] is contained in a commutative subring S 0 ≤ S which is finite over R (and s ∈ I · S 0 );

(4) R[s] is contained in a subring S 0 ≤ S which is finite over R;

(5) There exists a faithful R[s]-module M , which is finitely generated as an R-module.

Corollary 5.4. Finite ring extensions are integral.

Corollary 5.5 (Transitivity of integral extensions). Let R ≤ S ≤ T be a chain of commuta-
tive ring extensions such that T is integral over S and S is integral over R. Then T is integral over
R.
Proof. Apply K RONECKER’s Theorem 5.3: Let t ∈ T and td +s1 td−1 +· · ·+sd = 0 be an integral
equation for t over S. Then R[s1 , . . . , sd ] and thus R[s1 , . . . , sd , t] = d−1 i
P
i=0 R[s1 , . . . , sd ]t are
finite over R since the si ’s are integral over R. In particular, t is integral over R.
Corollary 5.6. Let S/R be a ring extension and s, s0 ∈ S integral over R. If ss0 = s0 s then
s + s0 and ss0 are also integral over R.
Definition 5.7. Let S/R be a ring extension with S commutative:

• Corollary 5.6 implies that the integral closure of R in S

IntR (S) := {s ∈ S | s is integral over R}

is a subring of S.

• R is called integrally closed in S if R = IntR (S).

• An integral domain is called integrally closed if it is integrally closed in its field of

fractions.

Example 5.8. Let k be a field. Both ring extensions

R := k[y] ≤ k[x, y]/hxy − 1i =: S

and
R := k[y] ≤ k[x, y]/hxyi =: S 0
are not integral: In both cases S = R[x] is not finite over R, i.e., x is not integral over R.
End
lect. 25
08.11
2 Lying over and going up
From now on S we will be commutative ring and S/R a ring extension.
Notation. If P ∈ Spec S then p := P ∩ R = Pc ∈ Spec R. We say that P lies over p.
Define (
Spec S → Spec R,
ι# :
P 7→ p := Pc = ι−1 (P) = P ∩ R,
associated to the ring embedding ι : R ,→ S.
Theorem 5.9 (Lying over). Let S/R be an integral (commutative) ring extension and let
p ∈ Spec R. Then:

(1) There exists a P ∈ Spec S lying over p (and containing pS), i.e., ι# is surjective.
2. LYING OVER AND GOING UP 83

(2) There are no strict inclusions between prime ideals of S lying over p.

(3) If P ∈ Spec S lies over p, then P is maximal iff p is maximal, i.e., ι# (Max(S)) = Max(R).

(4) If S is N OETHERian, then only finitely many prime ideals of S lie over p, i.e., the fibers of ι#
are finite.

For the proof we need the following result of K RULL:

Lemma 5.10 (K RULL’s prime existence lemma). Let R be a ring, I  R, and U ⊂ R a
multiplicatively closed subset such that I ∩ U = ∅. Then there exists p ∈ Spec R such that p ⊃ I
and p ∩ U = ∅.
Proof. If R is N OETHERian we can use the N OETHERian induction. If not, we are forced to
use Z ORN’s Lemma instead: Consider

Γ := {J  R | J ⊃ I, J ∩ U = ∅}.

Then Γ is partially orderedS by inclusion, Γ 6= ∅ since, e.g., I ∈ Γ and if {Jλ }λ∈Λ is a totally
ordered subset of Γ then λ∈Λ Iλ ∈ Γ is an upper bound. So we can apply Z ORN’s Lemma
and there exists a maximal element of p ∈ Γ. We shall now prove that p ∈ Spec R:
First, p is a proper ideal of R, since otherwise 1 ∈ p ∩ U = ∅ , a contradiction. Second, let
r1 , r2 ∈ R − p. Then for j = 1, 2 the ideal p + hri i ∈ / Γ by the maximality of p. Hence for
j = 1, 2 we have (p + hri i) ∩ U 6= ∅. That is, we can find elements pj ∈ p and aj ∈ R such
that pj + aj rj ∈ U . Then (p1 + a1 r1 )(p2 + a2 r2 ) ∈ U ⊂ R − p, so a1 a2 r1 r2 ∈
/ p. In particular,
r1 r2 ∈/ p, so p is prime.
We still need one more remark:
Remark 5.11. Let S/R be an integral ring extension.

(1) Let J  S be an ideal. Regard R/(J ∩ R) as a subring of S/J in the obvious way. Then
the ring extension (S/J)/(R/(J ∩ R)) is also integral.

(2) Let I  R. Then any element s ∈ IS = hIiS = I e is integral over I.

Proof.

(1) Let s = s + J ∈ S/J. An integral equation for s over R/(J ∩ R) is obtained from one
for s over R by taking residue classes of coefficients.
Pm
(2) An s ∈ IS can be written as s = i=1 ai si with ai ∈ I and si ∈ S for all i, i.e.,
s ∈ IR[s1 , . . . , sm ] and we are done by Theorem 5.3.

Now we are ready to proof the Lying-Over-Theorem 5.9:

Proof of Theorem 5.9.

(1) Consider the extended ideal pe = pS and the multiplicatively closed subset U :=
R − p ⊂ S. We will show that (pS) ∩ U = ∅ using the integrality of S/R. Then we will
apply K RULL’s prime existence lemma.
If s ∈ pS, then s is integral over p by Remark 5.11.(2). Hence, we have an integral
equation sd + a1 sd−1 + · · · + ad = 0 with all ai ∈ p. Now we show that s ∈ / U . Suppose
d d−1
the contrary. Then, in particular, s ∈ R and s = −a1 s − · · · − ad ∈ p, so s ∈ p. This
contradicts s ∈ U = R − p. So (pS) ∩ U = ∅. The prime existence lemma now yields a
P ∈ Spec S such that p ⊂ pS ⊂ P and P ∩ R ⊂ R − U = p. Hence, P is a prime ideal
lying over p.
84 CHAPTER 5. INTEGRAL RING EXTENSIONS

(2) If P1 ⊂ P2 are two prime ideals in S lying over p, then R := R/p ⊂ S/P1 =: S is
an integral extension of integrals domains by Remark 5.11.(1) such that (P2 /P1 ) ∩
R = h0i. Suppose P1 ( P2 . Then there exists a nonzero element s ∈ P2 /P1 and
we obtain a contradiction considering an integral equation for s over R of smallest
possible degree d:
sd + a1 sd−1 + · · · + ad = 0.
Indeed, since ad ∈ (P2 /P1 ) ∩ R is zero, and S is an integral domain, we may divide
the equation by s to get an integral equation of smaller degree.
(3) If p is maximal, then P is maximal by (2). For the converse we consider the integral
extension R/p ⊂ S/P. If S/P is a field, its only maximal ideal is h0i. Then, in turn, h0i
is the only maximal ideal of R/p by (1), i.e., R/p is also a field.
(4) If P ∈ Spec S lying over p, then pS ⊂ P and by (2) the ideal P is a minimal prime
containing pS. Since S is N OETHERian by assumption, Proposition 4.10 says that P is
a minimal associated prime of pS and there exists only finitely many such primes.
Example 5.12. The ring extension
√
R := Z ⊂ Z[ −5] ∼
= Z[x]/hx2 + 5i =: S

√ and the ideal p = h2i
√Z is maximal.

is integral √ The ideal pS = h2iS generated 2 in
S = Z[ −5] is not prime since (1 + −5) (1 − −5) = 6 = 3 · 2 ∈ h2iS . Using that
| {z } | {z }
∈h2i
/ S ∈h2i
/ S
√
Z[ −5]/h2iS ∼
= F2 [x]/hx2 + 1i = F2 [x]/h(x + 1)2 i
√ √ √
we see that P = h2, 1 + −5i = h2, 1 − −5i  Z[ −5] is the only maximal ideal lying over
h2i ∈ Spec Z.
A central property of integral ring extensions R ⊂ S is that nested pairs of prime ideals
of R correspond to such pairs in S (naturally, this is important in the context of K RULL
dimension). Here is a first result in this direction:
Corollary 5.13 (Going-Up-Theorem of C OHEN and S EIDENBERG). Let S/R be an integral
ring extension. If p1 ⊂ p2 are prime ideals in R and P1 ∈ Spec S lying over p1 , then there exists a
P2 ∈ Spec S lying over p2 such that P1 ⊂ P2 .
Proof. Applying the Lying-Over-Theorem 5.9 to the integral ring extension
R := R/p1 ⊂ S/P1 =: S
we get a P2 ∈ Spec S lying over p2 /p1 ∈ Spec R. The preimage P2 of P2 in S has the desired
properties.

3 Going down
“Going up” was a relatively easy result once we had “lying over”. To prove a dual “going
down” we need more assumptions, as we will show by examples.
We recall the following definition from Algebra.
Definition 5.14. Let R be an integral domain. The integral closure
R := IntR (Frac R) = {s ∈ Frac R | s integral over R}
of R in its field of fractions Frac R is called the normalization of R. We call R normal if
R = R.
 3. GOING DOWN 85

Example 5.15.
(1) If R is a unique factorization domain, then R is normal.
(2) In particular, k[x1 , . . . , xn ] is normal, where k is a field or k = Z.
(3) Q[x, y, z]/hx2 − y 2 zi is not normal. Geometrically, this is the so-called W HITNEY um-
brella:

Proof. Exercise.
End
Theorem 5.16 (Going-Down-Theorem of C OHEN and S EIDENBERG). Let S/R be an inte- lect. 26
gral ring extension and assume furthermore that R and S are both integral domains with R normal. 15.11
If p1 ⊂ p2 are prime ideals in R and P2 ∈ Spec S lying over p2 , then there exists P1 ∈ Spec S lying
over p1 , such that P1 ⊂ P2 .
For the proof we need a lemma:
Lemma 5.17. Let R be a normal ring, K := Frac R (its field of fractions), and K ⊂ L a field
extension. Let p ∈ Spec R and s ∈ L be integral over p. Then s is algebraic over K and all
coefficients of its minimal polynomial over K lie in p.
Proof. Since s is integral over p, it is algebraic over K. Let ps = xd +c1 xd−1 +· · ·+cd ∈ K[x] be
the minimal polynomial of s over K and s1 = s, s2 , . . . , sd be the roots of ps in the algebraic
closure K. Then, for each j there exists a ϕj ∈ Gal(K/K) with ϕj (s) = sj . Thus if f (s) = 0
is an integral equation for s with coefficients in p, then also f (sj ) = 0 for all j, so the sj are
all integral over p. By Corollary 5.6 the same is true for the coefficients ci of ps since these
are polynomial expressions in the roots sj (by V IETA’s formulas). Now since R = R and
√
p = p all ci lie in p.
Proof of Theorem 5.16. Consider three multiplicatively closed subsets of S:
U1 := R − p1 ,
U2 := S − P2 ,
U := U1 U2 = {r · s | r ∈ U1 , s ∈ U2 } ⊃ U1 , U2 .
In Step 1 below, we will show that pe1 ∩ U = (p1 S) ∩ U = ∅. Then we will apply K RULL’s
prime existence Lemma 5.10.
Step 1: Suppose there exists an s ∈ (p1 S) ∩ U . Then since s ∈ p1 S, it is integral over p1
by Remark 5.11.(2). Applying Lemma 5.17 we see that the minimal polynomial of
s ∈ L := Frac S over K := Frac R is of type ps = xd + c1 xd−1 + · · · + cd , with all ci ∈ p1 .
Since s ∈ U we can write s = r · se, where r ∈ U1 and se ∈ U2 . Then
c1 cd
pse = xd + xd−1 + · · · + d
r r
 86 CHAPTER 5. INTEGRAL RING EXTENSIONS

is the minimal polynomial of se over K. Again applying Lemma 5.17 we see that the
coefficients ci /ri of pse lie in R, since se is integral over R by assumption. In fact, ci /ri
i
√p1 since ci ∈ p1 and r ∈
lie in / p1 for all i. It follows that se is even integral over p1 . So
se ∈ p1 S ⊂ P2 by K RONECKER’s Theorem 5.3. This contradicts se ∈ U2 .
Step 2: K RULL’s prime existence Lemma 5.10 yields P1 ∈ Spec S such that p1 S ⊂ P1 and
P1 ∩ U = ∅. In particular, P1 ∩ U1 = ∅, so that P1 ∩ R = p1 , i.e., P1 is lying over p1 .
Also P1 ∩ U2 = ∅, hence P1 ⊂ P2 .
To indicate that the additional assumptions in the Going-Down-Theorem are needed,
we will give an example. To prepare for it, we state a couple of remarks.
Remark 5.18. Let k be a field. If p = (a1 , . . . , an ) ∈ k n is a point, each polynomial
f ∈ k[x1 , . . . , xn ] can be written as a polynomial in the (xi − ai )’s:
f = f (p) + terms of degree at least 1 in (xi − ai ). (*)
Indeed, to obtain the Taylor expression of f at p, substitute the (xi − ai ) + ai for xi in f and
expand. It follows from (*) that mp := hx1 − a1 , . . . , xn − an i is the kernel of the evaluation
map at p
εp : k[x1 , . . . xn ] → k, f 7→ f (p).
Hence k[x1 , . . . , xn ]/mp ∼
= k is a field, so mp is a maximal ideal.
If k is not algebraically closed, then not every maximal ideal in k[x1 , . . . , xn ] is of this
form (for n ≥ 1). Consider for example the maximal ideal hx2 + 1i
R[x].
We will come back to this later in the context of H ILBERT’s Nullstellensatz.
Examples 5.8 and 5.15.(3) involved rings of the following type:
Definition 5.19. Let k be a field. A finitely generated k-algebra is necessarily a residue
class rings of type k[x1 , . . . , xn ]/I (where I is an ideal). We call such an algebra an affine
ring or affine algebra. If I is a prime ideal then we talk about an affine domain.
The reason for this naming convention is that the canonical ring epimorphism
π : k[x1 , . . . , xn ] → k[x1 , . . . , xn ]/I
induces an embedding
π # : Spec(k[x1 , . . . , xn ]/I) → Spec k[x1 , . . . , xn ]
where we identify Spec k[x1 , . . . , xn ] with the n-dimensional affine space Ank over k.
Remark 5.20. Every extension R ⊂ S of affine rings is obviously of finite type. Hence, in
this case, S/R is integral iff S/R is finite. And finiteness can be computationally checked us-
ing Gröbner bases: Let R be the k-subalgebra generated by p1 , . . . , pm ∈ S := k[x1 , . . . , xn ]/I.
This defines a ring homomorphism
(
k[y1 , . . . , ym ] → k[x1 , . . . , xn ]/I,
ϕ:
yi 7→ pi .
Consider the ideal
J = hI, p1 − y1 , . . . , pm − ym i  k[x1 , . . . , xn , y1 , . . . , ym ]
corresponding to the graph of the morphism ϕ. Then ker ϕ = J ∩ k[y1 , . . . , ym ] and R =
k[y1 , . . . , ym ]/ ker ϕ. Let {g1 , . . . , g` } be a Gröbner basis of J w.r.t. a block order to eliminate
x1 , . . . , xn (cf. Proposition 1.55). Recall that ker ϕ is generated by all gk ’s not involving any
xi . One can show that R ,→ S is finite iff for each i = 1, . . . , n there exists a Gröbner basis
element gk whose leading monomial is of type xαi i for some αi ≥ 1.
3. GOING DOWN 87

Example 5.21. Consider the homomorphism ϕ : k[x, y, z] → k[s, t] with x 7→ s, y 7→ t2 −1

and z 7→ t(t2 − 1). Using the above remark we compute a Gröbner basis of J w.r.t. a block
order {s > t} > {x > y > z} and obtain
y3 + y2 − z2,
tz − y 2 − y, ty − z,
s − x, t2 − y − 1.
We conclude that ker ϕ = hz 2 − y 2 (y + 1)i and that the induced ring extension
R = k[x, y, z]/hz 2 − y 2 (y + 1)i ,→ k[s, t] = S
is integral. We can interpret this map geometrically:

The ideal P1 = hs − ti is the unique prime ideal of S lying over

p1 = P1 ∩ R = hx2 − 1 − y, x(x2 − 1) − zi
R.
The ideal p2 = hx − 1, y, zi
R is a maximal ideal containing p1 . Geometrically, the point
defined by p2 is lying on the curve define by p1 . There exists exactly two maximal ideals of
S lying over p2 , namely hs − 1, t + 1i and hs − 1, t − 1i. However P2 = hs − 1, t + 1i does
not contain P1 . Geometrically, the point (1, −1) does not lie on the line s = t. Hence going
down does not work in this case.
This means that R cannot be normal. Indeed yz ∈ Frac(R) − R is integral over R with
integral equation T 2 − (y + 1) = 0.
End
Now we add a few remarks on normalization. It is easy to show that being integral is lect. 27
preserved under localization. Here is a more general result: 22.11
Proposition 5.22. Let S/R be a ring extension, R the integral closure of R is S, and U ⊂ R a
e
e −1 ] is the integral closure R[U
multiplicatively closed subset. Then R[U ^ −1 ] of R[U −1 ] in S[U −1 ].

Proof. As mentioned above, R[Ue −1 ] is integral over R[U −1 ], hence R[U

e −1 ] ⊂ R[U
^ −1 ]. For

the converse inclusion let us ∈ S[U −1 ] be integral over R[U −1 ]. Then we have an integral
equation
 s d  r   s d−1  
rd
1
+ + ··· + =0 (*)
u u1 u ud
with all ri ∈ R and ui ∈ U . Set v = u1 · · · ud . Multiplying (*) by (u · v)d we get an integral
equation for sv over R. Hence sv ∈ Re and therefore s = sv ∈ R[U e −1 ].
u uv
88 CHAPTER 5. INTEGRAL RING EXTENSIONS

As a corollary we show that being normal is a local property:

Corollary 5.23. Let R be an integral domain. Then the following are equivalent:

(1) R is normal.

(2) Rp is normal for all p ∈ Spec R.

(3) Rm is normal for all m ∈ Max R.

Proof. Let R ⊂ Frac R be the normalization of R. Consider the inclusion ι : R ,→ R. Then

R is normal iff ι is surjective. Furthermore, by Proposition 5.22 Rp (resp. Rm ) is normal iff
ιp (resp. ιm ) is surjective. The result follows since being surjective is a local property (by
Proposition 3.51).
Remark 5.24. If R is a N OETHERian integral domain, then R need not be N OETHERian
(see Nagata’s book). If, however, R is an affine domain, then by E MMY N OETHER’s finite-
ness result the normalization R is even finite over R and R is again an affine domain.
Chapter 6

K RULL Dimension, N OETHER

Normalization, H ILBERT’s Nullstellensatz
and K RULL’s Principle Ideal Theorem

1 Definition of K RULL dimension

Let R be a ring. A sequence p0 ( p1 ( . . . ( pm of primes ideals of R with strict inclusions is
called a chain of prime ideals in R of length m or chain in Spec R of length m. We call the
chain maximal if it cannot be extended to a larger chain by inserting an extra prime ideal.
Definition 6.1. Let R be a ring. The K RULL dimension of R, written dim R, is the supre-
mum of the lengths of chains of primes ideals of R. The zero ring has Krull dimension −∞.
If I
R is a proper ideal, then the dimension1 of I, written dim I, is defined as dim R/I.
Example 6.2.
(1) The dimension of a field is 0.
(2) Each p ∈ Spec Z − {h0i} is of type hpi, where p is a rational prime. Hence
dim Z = 1.
More generally, each principal ideal domain which is not a field (e.g., the univariate
polynomial ring k[x] over a field k) has K RULL dimension 1.
(3) In the appendix of N AGATA’s textbook one can find a N OETHERian ring of infinite
dimension!
Proposition 6.3. If S/R is an integral ring extension, the R and S have the same dimension:
dim R = dim S.
Proof. By “lying over” and “going up” each chain in Spec R gives rise to one of S. Con-
versely if P0 ( P1 ( . . . ( Pm is a chain in Spec S, then the contractions Pi ∩R form a chain
in Spec R of the same length, since all inclusions will remain strict by Theorem 5.9.(2).
Remark 6.4. Consider the polynomial ring k[x1 , . . . , xn ] over a field k. Intuitively we
would expect that dim k[x1 , . . . , xn ] = n. One can indeed easily show that dim k[x1 , . . . , xn ] ≥
n since
h0i ( hx1 i ( hx1 , x2 i ( . . . ( hx1 , . . . , xn i
is a chain in Spec k[x1 , . . . , xn ]. To show that the dimension of k[x1 , . . . , xn ] is exactly n we
need the concept of N OETHER normalization.
1
This definition is historical and can hardly be changed. It should already be read in context: dim R as a
ring differs from dim R = −∞ as an ideal in R.

89
90CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLEN

2 Noether Normalization
Lemma 6.5. Let k be an infinite filed and f ∈ k[x1 , . . . , xn ] − {0}. Then there exists a point
p = (a1 , . . . , an ) ∈ k n such that f (p) 6= 0.

Proof. We use an induction over n. For n = 1 the statement is clear, since each non-zero
polynomial in one variable has at most finitely many roots. If n > 1 we write f in the form
f = c0 (x2 , . . . , xn )xe1 + · · · + ce (x2 , . . . , xn ). Then ci 6= 0 for at least one i and by induction
there exists a point p0 ∈ k n−1 such that ci (p0 ) 6= 0. Then 0 6= f (x1 , p0 ) ∈ k[x1 ], hence there
exists an element a such that f (a, p0 ) 6= 0.
For the main theorem we still need a lemma for which we need the following notation:
Remark 6.6. Let R be a ring. A polynomial in R[x1 , . . . , xn ] is called homogeneous (of
degree e) if all its monomials have degree e or if the polynomial is zero.
Every nonzero polynomial f ∈ R[x1 , . . . , xn ] of degree e can be uniquely written as a
sum f = fe + fe−1 + · · · + f0 , where fi are homogeneous of degree i, and where fe 6= 0. The
fi ’s are called the homogeneous components of f .
Given an extra x0 , the polynomial
 
h e x1 xn
f := x0 f ,..., ∈ R[x0 , x1 , . . . , xn ]
x0 x0

is homogeneous of degree e, called the homogenization of f w.r.t. x0 .

Conversely, the dehomogenization of a homogeneous polynomial F ∈ R[x0 , . . . , xn ]
w.r.t. x0 is defined to be the polynomial F (1, x1 , . . . , xn ) ∈ R[x1 , . . . , xn ]. We have

f h (1, x1 , . . . , xn ) = f and F = xm h
0 F (1, x1 , . . . , xn ) ,

where m is the highest degree power of x0 dividing F .

Lemma 6.7 (N OETHER normalization, hypersurface version). Let k be a field and let f ∈
k[x1 , . . . , xn ] − k a non-constant polynomial.

(1) If k is infinite and a2 , . . . , an ∈ k are sufficiently general2 , then the isomorphism

k[x1 , x2 , . . . , xn ] → k[x1 , x
e2 , . . . , x
en ]
x1 7→ x1 , xi 7→ x ei + ai x1 i = 2, . . . , n

maps f to a polynomial of type

axe1 + c1 (e en )xe−1
x2 , . . . , x 1 + · · · + ce (e
x2 , . . . , x
en ),

where a ∈ k ∗ , e ≥ 1, and each ci ∈ k[e

x2 , . . . , x
en ].

(2) If k is arbitrary, the same is achieved by sending

k[x1 , x2 , . . . , xn ] → k[x1 , x
e2 , . . . , x
en ]
i−1
ei + x1r
x1 7→ x1 , xi 7→ x i = 2, . . . , n,

where r is sufficiently large3 .

Proof.
2
See the proof for the construction of such elements.
3
See the proof for the construction of such a power.
2. NOETHER NORMALIZATION 91

(1) Let f = fe + fe−1 + · · · + f0 with fe 6= 0 be the decomposition of f into its homogeneous

components fi of degree i. After substituting xi 7→ x ei + ai x1 for i = 2, . . . , n, the
e
coefficient of x1 is fe (1, a2 , . . . , an ). By writing

f e = xm
1 fe (1, x2 , . . . , xn )
h

as in the above remark, we see that fe (1, x2 , . . . , xn ) 6= 0 since fe 6= 0. Since k is

infinite, there exists by Lemma 6.5 a2 , . . . an ∈ k such that fe (1, a2 , . . . , an ) 6= 0. The
results follows. End
lect. 28
(2) Write f as the finite sum of its terms, i.e., 29.11
X
f= aα xα1 1 · · · xαnn
α

i−1
and let r ∈ N. After substituting xi 7→ x ei + xr1 for i = 2, . . . , n the terms depending
α1 +α2 r+···+αn rn−1
only on x1 are of type aα1 ,...,αn x1 . If r is strictly larger than all αi ’s ap-
pearing in a term of f , the numbers α1 + α2 r + · · · + αn rn−1 x are distinct for distinct
α1 , . . . , αn (the αi ’s are then the digits in the base-r expansion of the exponent). Hence,
the terms depending only on x1 cannot cancel with each other, and we are done.

Example 6.8. Substituting y 7→ ye + x in xy − 1 we get the the polynomial x2 + xe

y−1
which is monic in x.
We recall from the Algebra lecture:
Definition 6.9. Let k be a field, S a k-algebra, and s1 , . . . , xn ∈ S. We call s1 , . . . , sn
algebraically independent over k if the the evaluation homomorphism

k[x1 , . . . , xn ] → S, xi 7→ si

is injective, i.e., if f (s1 , . . . , sn ) = 0 for f ∈ k[x1 , . . . , xn ] implies f = 0. An arbitrary subset

Γ ⊂ S is called algebraically independent over k if any finite collection of distinct elements
s1 , . . . , sn ∈ Γ are algebraically independent over k.
Theorem 6.10 (N OETHER normalization, basic version). For any affine k-algebra S =
k[x1 , . . . , xn ]/I 6= 0 there exists y1 , . . . , yd ∈ S (d ≤ n) such that:

(1) y1 , . . . , yd are algebraically independent over k and

(2) R = k[y1 , . . . , yd ] ⊂ S is a finite (hence integral) ring extension.

Every such inclusion is called a N OETHER normalization for S.

Proof. If I = h0i take d = n and R = S. This includes the case n = 0 in which S = k.

Suppose now n > 0 and I 6= h0i. Choose a polynomial 0 6= f ∈ I and an isomorphism

ϕ : k[x1 , . . . , xn ] → k[x1 , x
e2 , . . . , x
en ]

w.r.t. to this f as in Lemma 6.7 (we may also assume without loss of generality that ϕ(f )
is monic in x1 ). Setting Ie = ϕ(I) and Ie1 = Ie ∩ k[e en ], we then get an integral ring
x2 , . . . , x
extension
Se1 := k[e
x2 , . . . , x
en ]/Ie1 ,→ k[x1 , x en ]/Ie = Se1 [x1 ] =: Se ∼
e2 , . . . , x = S.
By induction, there exists a N OETHER normalization k[y1 , . . . , yd ] ⊂ Se1 with d ≤ n − 1. The
result follows from the transitivity of finite/integral ring extensions.
92CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLEN

Note that if k is infinite we may choose y1 , . . . , yd as a k-linear combination of x1 , . . . , xn .

To prove a refined version of the N OETHER normalization we recall the following defi-
nition from Algebra:
Definition 6.11. Let K ≤ L be a field extension. A transcendence basis for L/K is
a subset B ⊂ L which is algebraically independent over K such that L is algebraic over
K(B). One can show that such a basis always exists and that two such bases have the
same cardinality. The cardinality is called the transcendence degree of L over K, written
trdegK L.
For example if
K(x1 , . . . , xn ) = Frac(K[x1 , . . . , xn ])
is the field of rationals functions in the variables x1 , . . . , xn then

trdegK K(x1 , . . . , xn ) = n.

Theorem 6.12 (N OETHER normalization, refined version). Let S = k[x1 , . . . , xn ]/I 6= 0 be

an affine ring and J
S a proper ideal. Then there exists numbers4 c ≤ d ≤ n and a N OETHER
normalization k[y1 , . . . , yd ] ⊂ S such that

J ∩ k[y1 , . . . , yd ] = hy1 , . . . , yc i.

Proof. Let k[z1 , . . . , zd ] ⊂ S be any N OETHER normalization. By the transitivity of inte-

gral extensions, it is enough to find a N OETHER normalization k[y1 , . . . , yd ] ⊂ k[z1 , . . . , zd ]
such that J ∩ k[y1 , . . . , yd ] = hy1 , . . . , yc i for some c ≤ d. We may thus suppose that S =
k[z1 , . . . , zd ]. If J = h0i there is nothing to show. Otherwise, we may suppose by Lemma 6.7
that the coordinates are chosen such that J contains a monic polynomial

f = z1e + c1 z1e−1 + · · · + ce with all ci ∈ k[z2 , . . . , zd ].

Set y1 := f . Then k[y1 , z2 , . . . , zd ] ⊂ k[z1 , . . . , zd ] = k[y1 , z2 , . . . , zd ][z1 ] is an integral ring exten-

sion. On the other hand, by induction, there exists a N OETHER normalization k[y2 , . . . , yd ] ⊂
k[z2 , . . . , zd ] such that J ∩ k[y2 , . . . , yd ] = hy2 , . . . , yc i for some c ≤ d. Then k[y1 , y2 , . . . , yd ] ⊂
k[y1 , z2 , . . . , zd ] ⊂ k[z1 , . . . , zd ] is integral as well. Moreover, y1 , . . . , yd are algebraically inde-
pendent over k, since otherwise trdegk (k(z1 , . . . , zd )) would be smaller than d, contradicting
the algebraic independence of the zi ’s over k.
Finally, since each g ∈ J ∩ k[y1 , . . . , yd ] can be written as a sum g = g1 y1 + h with
g1 ∈ k[y1 , . . . , yd ] and h ∈ J ∩ k[y2 , . . . , yd ] = hy2 , . . . , yc i, we conclude that J ∩ k[y1 , . . . , yd ] =
hy1 , . . . , yc i.

3 Properties of K RULL dimension

Theorem 6.13. Let k be a field. Then the polynomial ring k[x1 , . . . , xn ] has K RULL dimension
n. In fact, every maximal chain in Spec k[x1 , . . . , xn ] has length n.
Proof. For both statements we do an induction on n. The case n = 0 is clear. Now suppose
n ≥ 1 and let
P0 ( P1 ( . . . ( Pm (*)
be a chain in Spec k[x1 , . . . , xn ]. By extending the chain we may assume that P0 = h0i since
this is the smallest prime ideal in the integral domain k[x1 , . . . , xn ]. Moreover, we may
assume that Pm is a maximal ideal. In particular m ≥ 1.
4
c will later be seen to coincide with the so-called codimension of the ideal J.
 3. PROPERTIES OF KRULL DIMENSION 93

Step 1. We show that m ≤ n. Applying Theorem 6.12 and the transcendence degree argument
from its proof to P1 , we get a N OETHER normalization k[y1 , . . . , yn ] ⊂ k[x1 , . . . , xn ]
such that P1 ∩ k[y1 , . . . , yn ] = hy1 , . . . , yc i for some c ≤ n. Set pi := Pi ∩ k[y1 , . . . , yn ] for
all i. Then
p1 /p1 ( p2 /p1 ( . . . ( pm /p1
is a chain of prime ideals in the ring k[y1 , . . . , yn ]/p1 ∼ = k[yc+1 , . . . , yn ]. By the induction
hypothesis, m − 1 ≤ n − c. Hence m ≤ n since c ≥ 1 (otherwise, both P0 and P1
would lie over the zero ideal of k[y1 , . . . , yn ], a contradiction to Theorem 5.9.(2)). This
finishes the first step.

Step 2. Now assume that the chain (*) is maximal. Then the induced chain

h0i = p0 ( p1 ( . . . ( pm (**)

of prime ideals in k[y1 , . . . , yn ] is a maximal as well: Suppose not, i.e., suppose there
is a q ∈ Spec k[y1 , . . . , yn ] with strict inclusions pi ( q ( pi+1 for some i. Then 0 ≤
i ≤ m − 1 since pm is maximal by Theorem 5.9.(3). Applying Theorem 6.12 to pi we
get a N OETHER normalization k[z1 , . . . , zn ] ⊂ k[y1 , . . . , yn ] such that pi ∩ k[z1 , . . . , zn ] =
hz1 , . . . , zc0 i for some c0 ≤ n. Then the composition

k[zc0 +1 , . . . , zn ] ,→ k[y1 , . . . , yn ]  k[y1 , . . . , yn ]/pi

is a N OETHER normalization as well, and we have strict inclusions

h0i ( (q/pi ) ∩ k[zc0 +1 , . . . , zn ] ( (pi+1 /pi ) ∩ k[zc0 +1 , . . . , zn ].

Since k[zc0 +1 , . . . , zn ] ⊂ k[x1 , . . . , xn ]/P1 is also a N OETHER normalization by Remark 5.11,

we see by gowing down that we may insert a prime ideal between h0i and Pi+1 /Pi in
k[x1 , . . . , xn ]/Pi and, thus also between Pi and Pi+1 in k[x1 , . . . , xn ]. This contradicts
the maximality of (*). We conclude that (**) is maximal as well. In particular, we sett
that c = 1 in Step 1.

Step 3. The maximal chain (**) corresponds to a maximal chain of prime ideals in

k[y1 , . . . , yn ]/p1 = k[y1 , . . . , yn ]/hy1 i ∼

= k[y2 , . . . , yn ]

of length m − 1. So m − 1 = n − 1 by the induction hypothesis, and we are done.

Corollary 6.14. Let R be an affine domain over k. Then

dim R = trdegk Frac(R).

This is the common length of all maximal chains in Spec R.

End
6.3 6.13 lect. 29
Proof. Let k[y1 , . . . , yd ] ⊂ R be a N OETHER normalization. Then dim R = dim k[y1 , . . . , yd ] = 06.12
d. Since also trdegk Frac(R) = trdegk k(y1 , . . . , yd ) = d (because Frac(R)/ Frac(k[y1 , . . . , yd ])
is algebraic) we have dim R = trdeg Frac(R).
For the second statement write R as the quotient k[x1 , . . . , xn ]/q, where q is a prime ideal
and fix a chain h0i =: q0 ( . . . ( qc := q in Spec k[x1 , . . . , xn ] which cannot be extended to
a longer such chain with largest ideal q. This fixed chain and the preimage of any maximal
chain in Spec R fit together to give a maximal chain in Spec k[x1 , . . . , xn ]. The latter has
length n by Theorem 6.13 and the result follows.
Definition 6.15. Let R be a ring and I
R a proper ideal. The codimension of I, written
codim I, is defined as follows:
94CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLEN

• If I = p is a prime ideal, take the supremum of the lengths of all chains of primes
ideals in R with largest prime p.

• If I is arbitrary then take the minimum of the codimension of prime ideals containing
I.

Corollary 6.16. Let R be an affine domain over a field k and I
R be a proper ideal. Then

dim I + codim I = dim R.

Proof. This follows from the Corollary 6.14: dim I can be expressed in terms of a maximal
chain in Spec R which includes a prime ideal p ⊃ I such that codim I = codim p.
Example 6.17. Consider the affine ring

R = k[x, y, z]/(hx, yi ∩ hzi) = k[x, y, z]/hxy, yzi

and p = hx, y, z − 1i ∈ Spec R.5 The ring R is obviously not a domain and codim p + dim p =
1 + 0 6= 2 = dim R. Observe that Spec R contains maximal chains of different lengths.
Remark 6.18. If R is a ring and p ∈ Spec R, then

dim Rp = codim p.

Proof. Indeed, by Theorem 3.38.(3) chains in Spec Rp correspond to chains in Spec R with
largest ideal p.

4 H ILBERT’s Nullstellensatz
We begin by introducing some basic notions from algebraic geometry. Throughout this
section k will denote a field.
Definition 6.19.

• For a subset T ⊂ k[x1 , . . . , xn ] the set

V (T ) := Vk (T ) := {a ∈ k n | f (a) = 0 ∀f ∈ T }

is called the vanishing locus of T . Any such locus is called an (affine) algebraic set.

• Dually, for a subset A ⊂ k n the ideal

I(A) := {f ∈ k[x1 , . . . , xn ] | f (a) = 0 ∀a ∈ A}

is called the vanishing ideal of A.

The basic idea of algebraic geometry is to use the constructions V and I to set up a cor-
respondence between ideals of k[x1 , . . . , xn ] and algebraic sets in k n , expressing geometric
properties in algebraic terms and vice versa.
Remark 6.20. Let T ⊂ k[x1 , . . . , xn ], J  k[x1 , . . . , xn ], and A ⊂ k n . Then

(1) V (T ) = V (hT i);

(2) V (I(A)) = A;
5
Geometrically, the set of closed points of Spec R with residue field k is the union of the (x, y)-plane and
the z-axes, and p is corresponds to the point (0, 0, 1) on the z-axes.
 4. HILBERT’S NULLSTELLENSATZ 95

(3) I(V (J)) ⊃ J.

Equality does not hold in general for two reasons. We demonstrate them by two examples:

• Consider J := hx2 i
k[x]. Then V (J) = {0} ⊂ k and I(V (J)) = hxi ) hx2 i = J.

• Consider J := hx2 + 1i
Q[x]. Then V (J) = ∅ ⊂ Q and I(V (J)) = h1i ) hx2 + 1i = J.

To get rid of problems of the first type, we may restrict ourselves to radical ideals. With
respect to the second problem, algebraic geometers prefer to work over an algebraically
closed field.
H ILBERT’s Nullstellensatz tells us, that over an algebraically closed field V and I define
a 1-1 correspondence between algebraic subsets of k n and radical ideals of k[x1 , . . . , xn ]. Un-
der this correspondence, irreducible algebraic sets correspond to primes ideals and points
to maximal ideals.
Theorem 6.21. Let k be an algebraically closed field. Then then Max k[x1 , . . . , xn ] consists of
the ideals of type
ma := hx1 − a1 , . . . , xn − an i,
where a = (a1 , . . . , an ) ∈ k n .

Proof. We already know from Remark 5.18 that any such ideal is maximal. Conversely, let
m ∈ Max k[x1 , . . . , xn ] be any maximal ideal. Then L = k[x1 , . . . , xn ]/m is a field. Hence, if
R = k[y1 , . . . , yd ] ⊂ L is a N OETHER normalization, we must have R = k since

0 = dim L = dim k[y1 , . . . , yd ] = d.

Then k ⊂ L is a finite, hence algebraic field extension. But since k is algebraically closed,
the inclusion k ⊂ L is an isomorphism. Hence, there exist a1 , . . . , an ∈ k such that ai can be
identified with the residue classes xi , i.e., ai = xi for all i = 1, . . . , n. Thus hx1 − a1 , . . . , xn −
an i ⊂ m. Since the former ideal is maximal and m is proper both ideals coincide.

This theorem states that points a ∈ k n correspond to maximal ideals ma (these are the
closed points in Spec k[x1 , . . . , xn ]).
Theorem 6.22 (H ILBERT’s Nullstellensatz, weak version). Let k be an algebraically closed
field and I  k[x1 , . . . , xn ] be an ideal. Then the following are equivalent:

(1) V (I) = ∅;

(2) 1 ∈ I.

Proof.

(2) ⇒ (1) is clear.

(1) ⇒ (2) If 1 ∈ / I, then I is a proper ideal of k[x1 , . . . , xn ]. Hence, I is contained in a maximal

ideal, which by Theorem 6.21 is of type ma = hx1 − a1 , . . . , xn − an i for some a =
(a1 , . . . , an ) ∈ k n . Since I ⊂ m implies V (I) ⊃ V (m), we conclude that {a} = V (m) ⊂
V (I), hence V (I) 6= ∅.

Theorem 6.23 (H ILBERT’s Nullstellensatz, strong version). Let k be an algebraically closed

field and I  k[x1 , . . . , xn ]. Then √
I(V (J)) = J.

Proof.
 96CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLEN
√
"⊃": If f ∈ J, then f m ∈ J for some m ≥ 1. This implies f m and, thus, also f vanish on
V (J).

"⊂": Let now f ∈ I(VP (J)) and J = hf1 , . . . , fr i. Then f vanishes on V (J), and we have to
show that f = ri=1 gi fi for some m ≥ 1 and some g1 , . . . , gr ∈ k[x1 , . . . , xn ] =: R. For
m

this we use the so-called Rabinovitch trick: Consider the ideal

J 0 := hf1 , . . . , fr , 1 − tf i  R[t],

where t is a new indeterminate. We show that V (J 0 ) ⊂ k n+1 is empty and apply the
weak version of the Nullstellensatz. Suppose the contrary, i.e., suppose there exists an
a0 = (a1 , . . . , an , an+1 ) ∈ V (J 0 ). Set a = (a1 , . . . , an ) ∈ k n . Then f1 (a) = . . . = fr (a) = 0,
so that a ∈ V (J) and an+1 f (a) = 1. This contradicts the fact that P f vanishes on V (J).
Now applying the weak Nullstellensatz we get 1 ∈ J , that is 1 = ri=1 hi fi + h(1 − tf )
0

for suitable h1 , . . . , hr , h ∈ R[t]. Substituting 1/f for t in this expression (formally we

work in the localization R[t]f ) and multiplyingP by a sufficiently high power f m to
clear denominators, we get a presentation f m = ri=1 gi fi as desired.
End
lect. 30
13.12
5 K RULL’s Principal Ideal Theorem and regular local rings
Theorem 6.24 (K RULL’s Principal Ideal Theorem). Let R be a N OETHERian ring and f ∈
R. Then each minimal prime p of hf i satisfies

codim p ≤ 1.

If f is not a zero-divisor in R, then equality holds.

Proof. The second statement follows from the first one: Since R is Noetherian, then, by
Proposition 4.11 the nonzero-divisor f is not contained in any of the (finitely many) mini-
mal primes of R. Hence, p ⊃ hf i cannot be minimal and codim p ≥ 1.
To show the first statement we localize and apply Nakayama’s lemma. By Remark 6.18
we have codim p = dim Rp . Furthermore, by Example 3.39.(3) the extended prime ideal
pe = pRp is a minimal prime of the extended ideal hf ie = hf iRp . Replacing R by Rp we may
thus assume that (R, p) is a local ring. Then for any q ∈ Spec R with q ( p we must show

codim q = dim Rq = 0.

Given such a prime q we consider the ideals

q(n) := {a ∈ R | ∃v ∈
/ q : va ∈ qn }, n ≥ 1.

Then, by Theorem 3.38, q(n) = (qn )ec is the preimage of (qn )e = qn Rq under the localization
morphism R → Rq . Since the maximal6 ideal p/hf i of R := R/hf i is also a minimal prime
of R by the assumption on p, the ring R is zero-dimensional. Being also N OETHERian, it is
A RTINian by Theorem 3.61. Hence, the descending chain

q(1) + hf i ⊃ q(2) + hf i ⊃ . . .

eventually becomes stationary, say

q(n) + hf i = q(n+1) + hf i.
6
recall that we are assuming (R, p) a local ring
5. KRULL’S PRINCIPAL IDEAL THEOREM AND REGULAR LOCAL RINGS 97

We may then write any g ∈ q(n) as g = h + af , with h ∈ q(n+1) and a ∈ R. Then af ∈ q(n) .
Since p is a minimal prime of hf i, we have f ∈ / q ( p and hence a ∈ q(n) , by the definition
(n)
of q . This shows that
q(n+1) + f q(n) = q(n) .
Since hf i ⊂ p and p is a maximal ideal by our above assumption, N AKAYAMA’s lemma
(Corollary 3.18) applied to R yields q(n) = q(n+1) . Then qn Rq = qn+1 Rq by Theorem 3.38.
Applying N AKAYAMA’s lemma over Rq we get qn Rq = h0i, in words, the maximal ideal
qRq of Rq is nilpotent. We conclude from Proposition 3.10 and Definition 3.11 that dim Rq =
0.

The generalization of K RULL’s Principal Ideal Theorem provides us with a lower bound
for the number of equations of an algebraic set of a given codimension.
Theorem 6.25 (Generalization of K RULL’s Principal Ideal Theorem). Let R be a N OETHER-
ian ring and I = hf1 , . . . , fc i  R an ideal generated by c elements. Then each minimal prime p of
I satisfies
codim p ≤ c.
Conversely, if p ∈ Spec R is of codimension c, then there exists y1 , . . . , yc ∈ R, such that p is a
minimal prime of hy1 , . . . , yc i.

Proof. For the first statement let p be a minimal prime of I. As in the preceding proof we
may assume that (R, p) is a local ring. We do induction on c. The case c = 1 is covered by
K RULL’s Principal Ideal Theorem 6.24.
Suppose now c > 1 and that q ( p is a prime ideal of R with no other prime ideal of R
between q and p (if no such q exists, then codim I = codim p = 0 and we are done). Since
p is a minimal prime of I = hf1 , . . . , fc i, at least one of the fi ’s is not contained in q, say
fc ∈
/ q. Then the maximal ideal p/(q + hfc i) of R/(q + hfc i) is also a minimal prime of this
ring. Arguing as in the previous proof we see that R/(q + hfc i) is a local A RTINian ring.
This implies that the maximal ideal p/(q + hfc i) is nilpotent (see the proof of Theorem 3.61).
In particular, all fi ’s are nilpotent modulo q + hfc i, say

fiN = gi + ai fc with gi ∈ q and ai ∈ R for i = 1, . . . , c − 1.

Then p is nilpotent modulo hg1 , . . . , gc−1 , fc i, so p := p/hg1 , . . . , gc−1 i is a minimal prime

of the principal ideal hf c i in the ring R/hg1 , . . . , gc−1 i. Hence, p has codimension ≤ 1 by
K RULL’s Principal Ideal Theorem 6.24. This shows that q is a minimal prime of hg1 , . . . , gc−1 i

R. The induction hypothesis gives codim q ≤ c − 1 and thus codim p ≤ c.
For the converse, given p as in the statement, we choose the yi ’s one at a time. Induc-
tively, with 0 < k < c, suppose y1 , . . . , yk have already be chosen to generate an ideal of
codimension k. Then, by prime avoidance, we may pick an element yk+1 ∈ p not con-
tained in any of the finitely many minimal primes of hy1 , . . . , yk i (indeed, any such prime
does not contain p, since its codimension is ≤ k < c by the first statement). Clearly,
codimhy1 , . . . , yk , yk+1 i = k + 1 and the result follows. To get the induction started consider
the finitely many minimal primes of h0i and argue as above.

Remark 6.26. If R is a N OETHERian ring with only finitely many maximal ideals, then
according to our definitions, the K RULL dimension of R is the maximum of the codimen-
sions of its maximal ideals (all of which are finitely generated). It then follows from the
generalization of K RULL’s Principal Ideal Theorem 6.25 that dim R is finite. This applies, in
particular, to every local Noetherian ring.
We show an important corollary, whose geometric meaning lies in the algebraic charac-
terization of non-singular (i.e., smooth) points of algebraic sets.
98CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLEN

Corollary 6.27. Let (R, m) be a local N OETHERian ring. Then

dim R = min{de | ∃y1 , . . . , yde : hy1 , . . . , ydei is m-primary}.

In particular,
dimR/m m/m2 ≥ dim R.
Proof. For the first statement let d = dim R = codim m. Further let d0 be the minimum of the
right hand side. Then d ≤ d0 and d0 ≤ d follow from the first and second of statement of the
generalization of K RULL’s Principal Ideal Theorem 6.25, respectively. The second statement
follows from the first one since m is generated by dimR/m m/m2 elements by Corollary 3.19.

The inequality in Corollary 3.19 may be strict. The special case of equality deserves an
own name.
Definition 6.28. A local N OETHERian ring (R, m) is called regular, if

dimR/m m/m2 = dim R.

p ∈ Spec R is called a smooth point Rp is regular.

Our final goal in this section will be to show that every regular local ring is an integral
domain. The geometric application of this is the following: The singularities of an alge-
braic set are the singular points of the individual irreducible components together with the
intersection points of the components.
In order to achieve our goal, we need some preparation.
Corollary 6.29. Let R be a N OETHERian ring, p1 ( p2 ( p3 a chain in Spec R, and r ∈ p3 −p2 .
Then there exists a p ∈ Spec R such that

r ∈ p and p1 ( p ( p3 .

Proof. Since h0i = p1 /p1 ( p2 /p1 ( p3 /p1 is a chain in Spec R/p1 , we have codim p3 /p1 ≥ 2.
Then, by K RULL’s Principal Ideal Theorem 6.24, p3 /p1 is not a minimal prime of of hri 
R/p1 . Hence, there exists some p ∈ Spec R such that r ∈ p/p1 ( p3 /p1 . Now take preimages.

Corollary 6.30. Let (R, m) be a local N OETHERian ring and r ∈ R − R∗ = m. Then

(1)
dim R/hri ≥ dim R − 1.

(2) If r is not a zero-divisor, then

dim R/hri = dim R − 1.

Proof.
(1) Choose a maximal chain p0 ( p1 ( . . . ( pd = m in Spec R with d = dim R. Then
r ∈ pi for some minimal i. Because of Corollary 6.29 we may assume that i ≤ 1. Then
p1 /hri ( . . . ( pd /hri is a chain in Spec R/hri. Hence dim R/hri ≥ d − 1 = dim R − 1.

(2) Choose a chain p0 ( p1 ( . . . ( pk in Spec R of maximal length such that r ∈ p0 . Then

6.16 6.24
dim R/hri = dim R/p0 = dim p0 ≤ dim R − codim p0 = dim R − 1

and we are done by (1).

 5. KRULL’S PRINCIPAL IDEAL THEOREM AND REGULAR LOCAL RINGS 99

End
lect. 31 Exercise 6.1. Prove the following general version of prime avoidance: Let p1 , . . . , pk−2 ∈
20.12 Spec R and pk−1 , pk , I  R. Then

k
[
I⊂ pi =⇒ I ⊂ pi for some i.
i=1

Proposition 6.31. A regular local ring (R, m) is an integral domain.

Proof. We do induction on d = dim R. If d = 0, then m = h0i and R is a field since R\m = R∗ .
Suppose now that d > 0. Then m2 6= m (otherwise, by N AKAYAMA’s lemma, m = h0i).
By the general version of Prime Avoidance in Exercise 6.1, there exists an r ∈ m − (p1 ∪ . . . ∪
pk ∪ m2 ), where p1 , . . . , pk are the finitely many minimal primes of the N OETHERian ring R.
Then R := R/hri is a local N OETHERian ring, with maximal ideal m := m/hri. Hence

dim R ≥ dim R − 1 = d − 1.

by Corollary 6.30. On the other hand, by Corollary 3.19 and since R is regular of dimension
d, we can find m2 , . . . , md ∈ m such that m1 := r, m2 , . . . , md generate m. Hence m can be
generated by d − 1 elements. Thus

dim R ≤ dimR/m m/m2 ≤ d − 1.

We conclude that R is regular.

By the induction hypothesis R := R/hri is an integral domain, so hri ∈ Spec R. This
implies that q := pi ( hri for some i. If x ∈ q is any element, then x = r · a for some
a ∈ R. Since r ∈/ q, we must have a ∈ q. This shows that rq = q and hence mq = q. By
N AKAYAMA’s Lemma (Corollary 3.17) q = pi = h0i and R is an integral domain.
100CHAPTER 6. KRULL DIMENSION, NOETHER NORMALIZATION, HILBERT’S NULLSTELLE
Chapter 7

Valuation rings and Dedekind domains

These two classes of rings play an important role in algebraic geometry and number theory.

1 Valuation rings
Definition 7.1. An integral domain R is called a valuation ring (VR) if the following
condition holds:
1
0 6= a ∈ Frac R =⇒ a ∈ R ∨ ∈ R.
a
A valuation ring R is called a discrete valuation ring (DVR) if R is N OETHERian, but not a
field.
Note that each field is a valuation ring. We will justify the name “valuation ring” later
on.
Proposition 7.2 (First properties of valuation rings). Let R be a valuation ring. Then

(1) R is local with maximal ideal

 
1
∗
mR := R − R = a ∈ Frac(R) − {0} | ∈/ R ∪ {0}
R.
a

(2) If R ( R0 ( Frac R, then

(a) R0 is a valuation ring as well;

(b) mR0 ( mR ;
(c) R0 = RmR0 .

In particular, dim R > dim R0 .

(3) R is normal.

(4) The set Γ := {I | I  R} of ideals in R is totally ordered w.r.t. inclusion:

I, J ∈ Γ =⇒ I ⊂ J or J ⊂ I.

(5) If I = ha1 , . . . , ar i  R, then I = hai i for some i. In particular, if R is a discrete valuation

ring, then R is a principal ideal domain and dim R = 1.

Proof.

101
102 CHAPTER 7. VALUATION RINGS AND DEDEKIND DOMAINS

(1) By definition, mR is maximal – provided it is an ideal. To show the latter consider

a, b ∈ mR and r ∈ R. Suppose a + b ∈ / mR . Then a 6= 0 6= b. Hence, since R is a
valuation ring, we may assume that a ∈ R (otherwise ab ∈ R and we may interchange
b

a and b). But then we get a contradiction

 
b
a+b= 1+ · a ∈ mR .
a

A similar argument shows that r · a ∈ mR .

(2) Clearly K := Frac R = Frac R0 .

(a) R0 is a valuation ring: 0 6= a ∈ K, a ∈

/ R0 =⇒ a ∈
/ R =⇒ 1
a
∈ R ⊂ R0 .
(b) By (a) and (1) R0 is a local ring with maximal ideal
   
1 0 1
mR0 = a ∈ K − {0} | ∈ / R ∪ {0} ⊂ a ∈ K − {0} | ∈/ R ∪ {0} = mR
a a

Since R ( R0 there exists an element a ∈ R0 − R. Then 1

a
∈ mR − mR0 , so we have
a strict inclusion mR0 ( mR .
(c) Since R − mR0 ⊂ R0 − mR0 = (R0 )∗ , we have R00 := RmR0 ⊂ R0 . Applying (a) to
R ⊂ R00 we conclude that R00 is a VR. Moreover, applying the argument from (b)
to R00 ⊂ R0 we get mR0 ⊂ mR00 . We show that the converse inclusion holds as well.
/ mR0 , then c ∈ (R0 )∗ , hence, a ∈ mR0 .
If a = cb ∈ mR00 , with b, c ∈ R, b ∈ mR0 and c ∈
Thus mR0 = mR00 , so R0 = R00 by (b).

(3) Let a ∈ R ⊂ Frac R, with integral equation

ad + r1 ad−1 + · · · + rd = 0 (ri ∈ R).

1
Suppose a ∈
/ R. Then a
∈ R and we get the contradiction

d  i−1
X 1
a=− ri · ∈ R.
i=1
a

(4) Exercise.

(5) By (4) there exists some i such that haj i ⊂ hai i for all j = 1, . . . , r. Hence I = hai i.
Moreover, if R is a DVR, then R is N OETHERian and, thus, a principal ideal domain.
Since R is not a field, we have dim R = 1 by Example 6.2.

Now we start justifying the names “valuation” and “discrete valuation” ring.
Definition 7.3. A totally ordered group is an additive A BELian group (G, +) together
with a total order ≤ on G, which is compatible with addition: For all g, g 0 , h ∈ G it holds: if
g ≤ g 0 , then g + h ≤ g 0 + h.
Note that no nonzero element of such a group can have finite order n, since otherwise
0G < g < 2g < · · · < n · g = 0G .
Example 7.4. (R, +, ≤) where ≤ is the usual order, is a totally ordered group and so is
every subgroup. In particular, (Z, +, ≤) is an ordered group.
Definition 7.5. Let K be a field and (G, +, ≤) is a totally ordered group.
1. VALUATION RINGS 103

• A valuation of K in G is a group homomorphism

ν : (K ∗ , ·) → (G, +),

such that ν(a + b) ≥ min(ν(a), ν(b)) for all a, b ∈ K ∗ where we set ν(0) := ∞.

• If (G, +, ≤) ∼
= (Z, +, ≤) and ν is an epimorphism, then ν is called a discrete valuation.
Proposition 7.6. In the situation above, the subset Rν := {a ∈ K | ν(a) ≥ 0} ⊂ K is a
subring of K, called the valuation ring of K w.r.t. ν. We have K = Frac Rν .
Proof.
(1) Rν is a subring: If a, b ∈ Rν , then ν(a + b) ≥ min(ν(a), ν(b)) ≥ 0, so a + b ∈ Rν , and
ν(a · b) = ν(a) + ν(b) ≥ 0, so a · b ∈ Rν . Furthermore, ν(1) = ν(1 · 1) = 2 · ν(1), hence
ν(1) = 0 and 1 ∈ Rν . Also 0 = ν(1) = ν(−1 · −1) = 2 · ν(−1) and −1 ∈ Rν . So also
−a = (−1) · a ∈ Rν .

(2) Clearly Frac Rν ⊂ K. For the other inclusion let a ∈ K − Rν . Then ν( a1 ) = −ν(a) > 0.
Hence, a1 ∈ Rν and a = 11 ∈ Frac Rν .
a

Proposition 7.7. Let R be an integral domain and K = Frac R.

(1) R is a VR iff R = Rν for some valuation ν on K.

(2) R is a DVR iff R = ν for some discrete valuation ν on K.

End
Proof. lect. 32
10.01
(1) (⇒) Set
– G := K ∗ /R∗ ,
– b ≤ a : ⇐⇒ ab ∈ R, and
– ν : K ∗ → G, a 7→ a with ν(0) := ∞ ≥ g, for all g ∈ G.
The set G is a (multiplicatively written) A BELian group. Furthermore, ≤ is well-
defined: If a = a0 and b = b0 then there exists g, h ∈ R∗ with a0 = ga, b0 = hb.
0 0
Hence ab0 = hg ab and, thus, ab0 ∈ R ⇐⇒ ab ∈ R. It is easy to see that ≤ is a partial
order which is total since either ab ∈ R or ab ∈ R, and which is compatible with
multiplication since ab = ca cb
. Hence (G, ·, ≤) is a totally ordered group and ν is
clearly a group homomorphism.
Given a, b ∈ K ∗ with a + b 6= 0 we may assume b ≤ a. Then ab ∈ R and, hence,
a+b
b
= 1 + ab ∈ R. We conclude that ν(a + b) = a + b ≥ b = min(ν(a), ν(b)), so ν is
a valuation. For the corresponding valuation ring we get

Rν := {a ∈ K | ν(a) ≥ 1}
= {a ∈ K | a ≥ 1}
n a o
= a ∈ K | a = ∈ R = R.
1

(⇐) If 0 6= a ∈ K ⊃ R, then either ν(a) ≥ 0 or ν a1 = ν(1) − ν(a) = −ν(a) ≥ 0.

Hence, either either a ∈ Rν or a1 ∈ Rν . So R = Rν is a valuation ring.

(2) (⇒) Since R is a DVR it is a local and a PID by Proposition 7.2. Hence R has a unique
maximal ideal m = hpi for some p ∈ R. Then R = Rhpi and it follows from
Lemma 7.8 below that R = Rν for some discrete valuation ν.
104 CHAPTER 7. VALUATION RINGS AND DEDEKIND DOMAINS

(⇐) By (1) we know that Rν is a valuation ring. We now show that Rν is a PID, since
then Rν is N OETHERian and, thus, a DVR:
Let h0i 6= I  Rν . Pick 0 6= f ∈ I with ν(f ) minimal. Then I = hf i, since if 0 6=
g ∈ I, we have ν(g) ≥ ν(f ) =⇒ ν( fg ) ≥ 0 =⇒ fg ∈ Rν =⇒ g = fg f ∈ hf i.

Lemma 7.8. Let R be an integral domain and K = Frac R.

(1) Let (G, +, ≤) be a totally ordered group and let ν : R − {0} → G be a map such that

ν(a · b) = ν(a) + ν(b)

ν(a + b) ≥ min(ν(a), ν(b)).

Then
a
ν : K ∗ → G, 7→ ν(a) − ν(b)
b
is a valuation of K.

(2) Suppose R is a unique factorization domain and let p ∈ R be prime. Given a ∈ R − {0}, we
write a = a0 · pna with p - a0 . Consider the map ν : R − {0} → Z, a 7→ na . Then
a
ν : K ∗ → Z, 7→ na − nb
b
is a discrete valuation on K and
na o
Rν := | na ≥ nb = Rhpi
b
is its discrete valuation ring.

Proof.

(1) The map ν is well-defined on K ∗ :

a a0
= 0 ⇐⇒ ab0 = a0 b =⇒ ν(a) + ν(b0 ) = ν(a0 ) + ν(b).
b b

(2) The map ν : R − {0} → Z satisfies the properties of (1):

ν(a · b) = ν(a0 b0 pna +nb ) = na + nb = ν(a) + ν(b)

ν(a + b) = ν((a0 + b0 pnb −na )pna ) ≥ na = min(ν(a), ν(b)),

wherewe assume nb ≥ na . By (1), ν : K ∗ → Z, ab 7→ na − nb is a discrete valuation and

Rν = ab | na ≥ nb = ab | p - b = Rhpi is a DVR.

Example 7.9. Consider R = Z, K = Q, and p a prime number and obtain the DVR Zhpi .
Theorem 7.10. Let R be an integral domain and I
R a proper ideal. Then there exists a
valuation ring R0 such that R ⊂ R0 ⊂ Frac(R) with I · R0 ⊂ mR0 (the latter is equivalent to
I · R0 6= R0 ).

Proof. Exercise.

Corollary 7.11. Let R be an integral domain. Then the integral closure of R in Frac(R) is
\
R= R0 .
R⊂R0 ⊂Frac R
R VR
2. DEDEKIND DOMAINS 105

Proof. Exercise.

Proposition 7.12. Let R be a local N OETHERian integral domain with maximal ideal m and
dim R = 1. Then the following are equivalent:

(1) R is a DVR;

(2) R is principal ideal domain;

(3) m is principal;

(4) dimR/m m/m2 = 1;

6 I  R is of the form mn for some n ∈ N0 ;

(5) Any ideal h0i =

(6) There exists an element p ∈ R such that any ideal h0i 6= I  R is of the form hpn i for some
n ∈ N0 ;

(7) R is normal;

(8) dimR/m mk /mk+1 = 1 for all k ≥ 1.

Proof. Exercise.

Example 7.13. The following are DVRs:

(1) K[x]hxi .

(2) K[[x]].

(3) the local ring of a curve over an algebraically closed field at a smooth point.

(4) Zhpi , where p is a prime number (p-adic valuation).

2 D EDEKIND domains
Dedekind rings provide one possible answer to the search for unique prime factorization:
They allow for a unique factorization of each nonzero ideal into a product of prime ideals.
We start with one way of defining Dedekind rings:
Definition 7.14. An integral domain R is called a D EDEKIND domain if

(1) R is N OETHERian,

(2) R is normal, and

(3) dim R = 1, i.e., every nonzero prime ideal of R is maximal.

Example 7.15. Every PID is a D EDEKIND domain.

Proof. This follows from Example 5.15 and Example 6.2.

Theorem 7.16. Let R be an integral domain which is not a field. Then the following are equiv-
alent:

(1) R is a D EDEKIND domain.

(2) Every nonzero proper ideal of R can be written as a product of finitely many prime ideals.
 106 CHAPTER 7. VALUATION RINGS AND DEDEKIND DOMAINS

(3) R is N OETHERian and the localization at each maximal ideal of R is a DVR.

Remark 7.17. If R is a D EDEKIND domain and I is a nonzero proper ideal of R, then I

can be written as a product
I = pk11 · . . . · pknn ,
with k1 , . . . , kn ∈ N>0 and distinct maximal ideals p1 , . . . , pn of R. This presentation is
unique up to permutation of the factors and p1 , . . . , pn are exactly the associated prime
ideals of I.
D EDEKIND domains show up in algebraic geometry as follows:
Example 7.18. The coordinate ring of a smooth irreducible curve over an algebraically
closed field is a D EDEKIND domain.
D EDEKIND domains show up in number theory as follows:
Definition 7.19. An algebraic number field K is a finite (and hence algebraic) field ex-
tension of the field rational numbers Q. We then refer to the degree [K : Q] = dimQ K as the
degree of K and to the to the integral closure OK := ZK of Z in K (formed by all elements
a ∈ K such that there exists some monic f ∈ Z[x] with f (a) = 0) as the ring of integers of
K.
Example 7.20.

(1) K = Q and OK = Z.

The following are algebraic number fields of degree 2 (quadratic number fields):

(2) K = Q[i] and OK = Z[i] (G AUSSian integers).

√ h √ i
(3) K = Q[ 5] and OK = Z 1+2 5 .

Theorem 7.21. If K is an algebraic number field1 then OK is a D EDEKIND domain.

√ √
Example 7.22. Consider K = Q[ −5] then OK = Z[ −5] is a D EDEKIND domain. See
the discussion at the beginning of Chapter 4.
End
lect. 33
17.01

1
i.e., a finite extension of Q
Bibliography

[BLH11] Mohamed Barakat and Markus Lange-Hegermann, An axiomatic setup for algorith-
mic homological algebra and an alternative approach to localization, J. Algebra Appl. 10
(2011), no. 2, 269–293, (arXiv:1003.1943). MR 2795737 (2012f:18022) 39

[Bru] Gregor Bruns, Are all projection maps in a categorical product epic?, StackExchange,
(accessed 2013-11-18): (http://math.stackexchange.com/questions/
238995). 32

[BS87] David Bayer and Michael Stillman, A theorem on refining division orders by the re-
verse lexicographic order, Duke Math. J. 55 (1987), no. 2, 321–328. MR 894583 2

[Gro57] Alexander Grothendieck, Sur quelques points d’algèbre homologique, Tôhoku Math.
J. (2) 9 (1957), 119–221, Translated by Marcia L. Barr and Michael Barr:
Some aspects of homological algebra, (ftp://ftp.math.mcgill.ca/barr/
pdffiles/gk.pdf). MR MR0102537 (21 #1328) 37

[HS97] P. J. Hilton and U. Stammbach, A course in homological algebra, second ed., Gradu-
ate Texts in Mathematics, vol. 4, Springer-Verlag, New York, 1997. MR MR1438546
(97k:18001) 23, 27, 29, 32, 35, 36, 38

[ML95] Saunders Mac Lane, Homology, Classics in Mathematics, Springer-Verlag, Berlin,

1995, Reprint of the 1975 edition. MR MR1344215 (96d:18001) 38

[Rot09] Joseph J. Rotman, An introduction to homological algebra, second ed., Universitext,

Springer, New York, 2009. MR MR2455920 (2009i:18011) 37

[Ste] Greg Stevenson, What is the opposite category of the category of modules (or
hopf algebra representations)?, MathOverflow, (accessed 2013-11-18): (http://
mathoverflow.net/questions/29442). 39

107
Index

L(A), 7 CertainColumns, 44
R-algebra, 64 DecideZeroColumns, 40
Hom-set, 23 DecideZeroColumnsEffectively, 40
Mod-R, 38 DecideZeroRows, 40
R-Mod, 38 DecideZeroRowsEffectively, 40
R-mod, 39 DiagMat, 44
mod-R, 39 IdentityMat, 43
D EDEKIND domain, 105 MulMat, 43
K RULL dimension, 89 StackMat, 44
SubMat, 43
action map, 25
SyzygiesOfColumns, 40
algebra
SyzygiesOfRows, 40
relation, 20, 21
ZeroMat, 43
algebraic set, 94
G AUSSian normal form=RREF, 41
algebraically independent, 20
H ERMITE normal form algorithm=HNF,
algorithm
41
addition, 39
Gröbner basis, 41
additive
multiplication, 39
Add, 44
pre-A BELian
CoproductMorphism, 44
Cokernel, 45
EmbeddingOfLeftCofactor, 44
CokernelColift, 45
EmbeddingOfRightCofactor, 44
ProductMorphism, 44 CokernelEpi, 45
ProductObject, 44 Kernel, 45
ProjectionOnLeftFactor, 44 KernelLift, 45
ProjectionOnRightFactor, 44 KernelMono, 45
Buchberger, 11 submodule membership, 16
categorical, 31 subtraction, 39
category syzygy module, 17
Compose, 42 Artinian
IdentityMorphism, 42 monomial order, 2
Source, 42 associated primes, 78
Target, 42 minimal, 78
category with zero axiom of choice, 27, 51
ZeroMorphism, 43
decide equality, 39 binomial, 12
decide membership, 39 Gröbner basis, 12
determinate division, 5 ideal, 12
elimination, 19 biproduct, 36
matrix boundaries, 54
AddMat, 43 Buchberger
AugmentMat, 44 algorithm, 11
BasisOfRows, 41 criterion, 10, 13, 14
BasisOfRowsCoeff, 41 test, 11

108
INDEX 109

categorical complex, 53
object acyclic, 54
initial=empty coproduct, 35 left, 54
terminal=empty product, 35 right, 54
zero=empty biproduct, 35 exact, 54
properties, 51 factor, 54
categorification homology, 53
horizontal, 23 sub-, 54
category, 23 constructive
A BELian, 37 A BELian group, 39
additive, 36 category, 42
arrow, 49 mathematical structure, 43
enriched over, 35 ring, 39
equivalence of categories, 51 set, 39
identity, 23 convention
morphism, 23 associative, 26
object, 23 commutative, 26
of (finite) presentations left, 23
left, 42 mixed, 26
right, 42 right, 23
opposite, 34 coproduct, 32
pre-A BELian, 37 embedding, 32
pre-additive, 35 morphism, 32
simplicial, 25 criterion
skeletal, 24 Buchberger, 10, 13, 14
small, 23 cycles, 54
locally, 23
defect of exactness, 53
subcategory, 34
degree-compatible order, 3
full, 34
dehomogenization, 90
thin, 25
direct sum, 36
with zero, 35
divisible, 9
chain
Division
morphism, 54
with remainder
of degree n, 54
determinate version, 5
co-image, 28, 30
dominates
epi, 31
column-, 40
mono, 37
modulo, 46
co-sequence, 53 row-, 40
cocomplex, 53 modulo, 46
cohomology, 54
codimension, 93 elimination
coefficient, 1 ideal, 19
cohomology, 54 order, 20
cokernel, 28, 30 embedded prime, 78
epi, 30 epi
object, 30 regular, 38
colift, 29 epimorphism, 27
along an epi, 29 equivalence
cokernel, 30 of categories, 51
compatible triples of functors, 50
left, 42 exact, 53
110 INDEX

factor homomorphism theorem, 31, 37

complex, 54
finitely presented, 51 ideal
free resolution, 55 associated primes, 78
length, 55 binomial, 12
functor, 47 contraction, 58
contravariant Hom-, 49 elimination, 19
contravariant=presheaf, 49 embedded prime, 78
covariant, 47 extension, 58
covariant Hom-, 48 intersection, 57
embedding, 48 isolated prime, 78
equivalence of functors, 50 left, 28
essentially surjective=dense, 48 membership, 16
faithful minimal associated prime, 78
fully=full and faithful, 48 monomial, 1
faithful=embedding, 48 primary, 75
forgetful=underlying, 48 primary decomposition, 76
free, 48 prime
full, 48 chain of, 89
identity, 47 isolated, 80
inverse, 47 product, 57
invertible, 47 quotient, 57
radical, 57
generalized right, 28
mono, 53 sum, 57
generators image, 28, 30
minimal, 2 epi, 37
number of, 42 mono, 31
global independent
monomial order, 2 algebraically, 20
Gordan initial
Hilbert basis theorem, 8 ideal, 7
Gordon module, 7
Lemma, 1 integral, 81
Gröbner involution, 41
basis isolated prime, 78
minimal, 15 isomorphic, 27
reduced, 15 isomorphism, 27
Gröbner basis
binomial, 12 Jacobson radical, 59
greatest common divisor, 9
Gröbner basis, 7, 8 kernel, 28, 29
mono, 30
Hilbert object, 30
syzygy theorem, 17
homogeneous L(A), 7
components, 90 lattice, 29
polynomial, 90 leading
homogenization, 90 coefficient, 3
homology, 53 ideal, 7
homomorphism, 26 module, 7
group, 26 monomial, 3
INDEX 111

leading term, 3 ideal, 1

least common multiple, 9 in a free module, 1
Lemma order, 2
Gordon, 1 degree-compatible, 3
lexicographic global, 2
order, 2 induced, 13
lift, 29 on free modules, 4
along a mono, 29 submodule, 2
kernel, 30 monomorphism, 27
morphism, 23
Macaulay automorphism
theorem, 8 natural, 50
map chain, 54
R-linear, 26 co-image, 28
bijective, 27 cochain, 54
embedding, 28 codominates, 29
identity, 26 cokernel, 28, 30
injective, 27 colift, 29
inverse, 28 composable, 23
surjective, 27 composition, 23
zero, 26 copairing, 32
membership coproduct, 32
ideal, 16 dominates, 29
subalgebra, 21 endomorphism
submodule, 16 natural, 50
minimal epi, 27
generators, 2 natural, 50
Gröbner basis, 15 split, 27
minor, 10 image, 28
module iso, 27
i-th syzygy, 55 isomorphism
factor, 28 natural, 50
projection, 28 kernel, 28, 29
finitely generated = f.g., 39 lift, 29
generators, 51 mono, 27
homomorphism, 26 natural, 50
left R-module, 25 split, 27
localization, 68 pairing, 32
number of generators, 42 post-, 23
relations, 51 post-cancelable, 27
right R-module, 25 post-factor, 29
simple, 28 post-inverse=complement, 27
submodule, 28 pre-, 23
proper, 28 pre-cancelable, 27
zero module, 25 pre-factor, 29
mono pre-inverse=section, 27
generalized, 53 product, 32
regular, 38 source, 23
monoid, 23 split epi, 27
monoidoid, 23 split mono, 27
monomial, 1 target, 23
112 INDEX

zero, 26, 35 ring, 98

remainder, 5, 7
natural transformation, 49 representation
automorphism, 50 left, 25
endomorphism, 50 right, 25
epi, 50 reverse lexicographic order, 3
isomorphism, 50 ring
mono, 50 computable, 41
nilradical, 59 left, 40
normal
right, 40
form, 9
endomorphism, 26
notation
field of fractions, 67
globular, 49
involution, 41
object, 23 local, 59
factor, 29 localization, 65
subfactor, 52 at prime ideal, 67
subobject, 29 normal, 84
larger, 29 normalization, 84
smaller, 29 quotient field, 67
oidification, 23 regular, 98
order total quotient, 67
block, 20 valuation, 101
degree reverse lexicographic, 3 discrete, 101
elimination, 20 of a field, 103
lexicographic, 2 ring of integers, 106
monomial, 2 row-space, 40
partial, 25
pre-, 25 S-polynomial, 10
product, 20 S-vector, 10
self-dual, 35
poset, 25, 29 sequence, 53
presentation matrix, 51 exact (at), 53
primary component short, 53
embedded, 80 differential, 53
isolated, 80 simplicial
primary decomposition, 76 category, 25
product, 32 smooth
morphism, 32 point, 98
projection, 32 source, 23
product order, 20 spectrum, 58
proset, 25 maximum, 58
pull-back, 33 standard
morphisms, 33 expression, 7
push-out, 33 monomials, 8
morphisms, 33 subalgebra
quiver, 23 membership, 21
quotient, 9 subcategory, 34
subcomplex, 54
reduced subfactor, 52
Gröbner basis, 15 submodule
regular membership, 16
INDEX 113

monomial, 2
subobject, 29
lattice, 29
syzygies
column, 40
row, 40
relative, 44
syzygy, 13
matrix, 13
module, 13
algorithm, 17

target, 23
tensor
product, 63
pure, 63
term, 1
initial, 3
leading, 3
theorem
division, 5
syzygy, 17

universal property, 29
cokernel, 30
coproduct, 32
kernel, 29
product, 32
pull-back, 33

valuation, 103
discrete, 103
vanishing ideal, 94
vanishing locus, 94