YANQI LAKE

LECTURES ON
ALGEBRA
/ Wen-Wei Li

Part 3 Commutative rings

Completions
Dimension theory
Yanqi Lake Lectures on Algebra

Part 3

Wen-Wei Li
Chinese Academy of Sciences
 Version: 2019-06-14

The cover page uses the fonts Bebas Neue and League Gothique, both licensed
under the SIL Open Font License.

This work is licensed under a Creative Commons Attribution-NonCommercial

4.0 International License. To view a copy of this license, visit http://
creativecommons.org/licenses/by-nc/4.0/.
 Contents

Introduction 1

1 Warming up 3
1.1 Review on ring theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Localization of rings and modules . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Radicals and Nakayama’s lemma . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Noetherian and Artinian rings . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 What is commutative algebra? . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Primary decompositions 15
2.1 The support of a module . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2 Associated primes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3 Primary and coprimary modules . . . . . . . . . . . . . . . . . . . . . . . 18
2.4 Primary decomposition: the main theorem . . . . . . . . . . . . . . . . . 19
2.5 Examples and remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Integral dependence, Nullstellensatz and flatness 25

3.1 Integral extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.3 Flatness: the first glance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Structure of flat modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.5 Faithful flatness and surjectivity . . . . . . . . . . . . . . . . . . . . . . . . 36

4 Going-up, going-down, gradings and filtrations 39

4.1 Going-up and going-down . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Subsets in the spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.3 Graded rings and modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4 Filtrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.5 Theorems of Artin–Rees and Krull . . . . . . . . . . . . . . . . . . . . . . 48

5 From completions to dimensions 51

5.1 Completions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Further properties of completion . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Hilbert–Samuel polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.4 Definition of Krull dimension . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.5 Krull’s theorems and regularity . . . . . . . . . . . . . . . . . . . . . . . . 63
⋅ iv ⋅ CONTENTS

6 Dimension of finitely generated algebras 65

6.1 Dimensions in fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2 Calculation for polynomial algebras . . . . . . . . . . . . . . . . . . . . . 67
6.3 Noether normalization and its consequences . . . . . . . . . . . . . . . . 68

7 Serre’s criterion for normality and depth 73

7.1 Review of discrete valuation rings . . . . . . . . . . . . . . . . . . . . . . 73
7.2 Auxiliary results on the total fraction ring . . . . . . . . . . . . . . . . . . 74
7.3 On normality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.4 Serre’s criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.5 Introduction to depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

8 Some aspects of Koszul complexes 83

8.1 Preparations in homological algebra . . . . . . . . . . . . . . . . . . . . . 83
8.2 Auxiliary results on depth . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.3 Koszul complexes and depth . . . . . . . . . . . . . . . . . . . . . . . . . 86

Bibliography 91

Index 93
 Introduction
In the beginning, these lecture notes were prepared for the graduate course Algebra III
(ID: 011M4002Y) in Spring 2016, University of the Chinese Academy of Sciences. For
some reasons, it took place in the Yuquanlu campus instead of the Yanqi Lake campus
as it should be.
The course is a sequel to Algebra I (fields, modules and representations) and II (ho-
mological algebra). The topic of this Part III is commutative algebra, or more precisely
commutative ring theory. Each “Lecture” in these notes took roughly one week, say ap-
proximately four hours of lecture, but the materials were only partially covered. My
initial intention was to give a traditional course on commutative algebra as proposed
by the syllabus prescribed by UCAS. For various reasons, my plan failed. For exam-
ple, there are too few discussions on depth, regular sequences and Cohen–Macaulay
modules, too few applications of completions, and the computational aspects haven’t
been touched. Moreover, the homological aspect of commutative algebra is almost non-
existent in these notes, namely the Auslander–Buchsbaum Formula, the properties of
regular local rings, etc. Last but not least, the exercises herein are scarcely sufficient.
These notes were also used for the Enhanced Program for Graduate Study held at
the Beijing International Center of Mathematical Research, Peking University, during
Spring 2019 (course ID: 00102057).
As the title suggests, some backgrounds from the Part I are presumed, namely the
basic notions of rings, modules and their chain conditions, as well as familiarity with
tensor products and some Galois theory. We occasionally presume some basic knowl-
𝑖
edge of homological algebra, such as the functors Tor𝑖 and Ext .
Sometimes I made free use of the language of derived categories. This was indeed
covered in the preceding course Algebra II, and it should be common sense for the future
generations.
As the reader might have observed, these notes were prepared in a rush; certain
paragraphs have not been proofread yet and many proofs are silly. I am very grateful
to the students for various corrections and improvements, and I will try to polish these
notes in the future.

Conventions
Throughout these lectures, we consider only associative rings with unit 1, and the rings
and algebras are assumed to be commutative and nonzero unless otherwise specified.
The ideal generated by elements 𝑥1 , 𝑥2 , … in a ring 𝑅 is denoted by (𝑥1 , 𝑥2 , …) or some-
times ⟨𝑥1 , 𝑥2 , …⟩; the 𝑅-algebra of polynomials in variables 𝑋, 𝑌, … with coefficients in
⋅2⋅ Introduction

𝑅 is denoted by 𝑅[𝑋, 𝑌, …]. We write 𝑅× for the group of invertible elements in a ring 𝑅.
A ring without zero-divisors except 0 is called an integral domain, or simply a domain.
The localization of a ring 𝑅 with respect to a multiplicative subset 𝑆 will be written as
𝑅[𝑆−1 ].
For any sets 𝐸, 𝐹, let 𝐸 ∖ 𝐹 ∶= {𝑥 ∈ 𝐸 ∶ 𝑥 ∉ 𝐹}. The cardinality of 𝐸 is denoted by |𝐸|.
The usual logical connectives such as ∃, ∀, ∧, ∨ and so forth will occasionally be
used. Writing 𝐴 ∶= 𝐵 means that the expression 𝐴 is defined to be 𝐵.
We will use the standard notations ℤ, ℚ, ℝ, ℂ to denote the set of integers, of ra-
tional numbers, etc. Sans serif fonts are reserved for categories, such as Ab (abelian
groups) and Ring.
When denoting morphisms in a category by arrows, monomorphisms (resp. epi-
∼
morphisms, isomorphisms) will be indicated ↪ (resp. ↠, →).

Possible references
The reader is expected to have basic familiarity with groups, rings and modules, as
covered in my lecture notes on Algebra I. We will make use of some really elementary
homological algebra as our course proceeds — so keep calm.
Our main references will be [11] and [8]. The Bourbaki volumes [5, 3] serve as our
ultimate source. The readers are also encouraged to consult the relevant materials in
Stacks Project.

《教授生涯》, 李桦, 木刻版画, 年.

 Lecture 1
Warming up
The reader might be familiar with most of the materials in this lecture. Our goal is to fix
notation and present the basic structural results on Noetherian and Artinian rings or
modules, including the celebrated Nakayama’s Lemma which will be used repeatedly.

1.1 Review on ring theory

Let 𝑅 be a ring, supposed to be commutative with unit 1 ≠ 0 as customary. Recall that
an ideal 𝐼 ⊊ 𝑅 is called
⋄ prime, if 𝑎𝑏 ∈ 𝐼 ⟺ (𝑎 ∈ 𝐼) ∨ (𝑏 ∈ 𝐼);
⋄ maximal, if 𝐼 is maximal among the proper ideals of 𝑅 with respect to inclusion.
Recall the following standard facts

⋄ 𝐼 is prime if and only if 𝑅/𝐼 is an integral domain, i.e. has no zero divisors except
0;

⋄ 𝐼 is maximal if and only if 𝑅/𝐼 is a field; in particular, maximal ideals are prime;

⋄ every proper ideal 𝐼 is contained in a maximal ideal (an application of Zorn’s

Lemma).

Definition 1.1.1 (Local rings). The ring 𝑅 is called local if it has a unique maximal ideal,
semi-local if it has only finitely many maximal ideals.
Let 𝔪 be the maximal ideal of a local ring 𝑅. We call 𝑅/𝔪 the residue field of 𝑅. A
local homomorphism between local rings 𝜑 ∶ 𝑅1 → 𝑅2 is a ring homomorphism such
that 𝜑(𝔪1 ) ⊂ 𝔪2 . Consequently, local homomorphisms induce embeddings on the
level of residue fields.

Sometimes we denote a local ring by the pair (𝑅, 𝔪).

Remark 1.1.2. Let 𝑅 be a local ring with maximal ideal 𝔪, then 𝑅× = 𝑅 ∖ 𝔪. The is easily
seen as follows. An element 𝑥 ∈ 𝑅 is invertible if and only if 𝑅𝑥 = 𝑅. Note that 𝑅𝑥 = 𝑅
is equivalent to that 𝑥 is not contained in any maximal ideal, and the only maximal
ideal is 𝔪.
⋅4⋅ Warming up

Throughout these lectures, we shall write

Spec(𝑅) ∶= {prime ideals of 𝑅},

MaxSpec(𝑅) ∶= {maximal ideals of 𝑅}.

They are called the spectrum and the maximal spectrum of 𝑅, respectively. The upshot is
that Spec(𝑅) comes with a natural topology.

Definition 1.1.3 (Zariski topology). For any ideal 𝔞 ⊂ 𝑅, set 𝑉(𝔞) ∶= {𝔭 ∈ Spec(𝑅) ∶
𝔭 ⊃ 𝔞}. Then there is a topology on Spec(𝑅), called the Zariski topology, whose closed
subset are precisely 𝑉(𝔞), for various ideals 𝔞.

Indeed, we only have to prove the family of subsets {𝑉(𝔞) ∶ 𝔞 ⊂ 𝑅} is closed un-
der finite union and arbitrary intersections. It boils down to the easy observation that
𝑉(𝔞) ∪ 𝑉(𝔟) = 𝑉(𝔞𝔟) (check this!) and ⋂𝔞∈𝒜 𝑉(𝔞) = 𝑉 (∑𝔞∈𝒜 𝔞), where 𝒜 is any
family of ideals.
Given a ring homomorphism 𝜑 ∶ 𝑅1 → 𝑅2 , if 𝐼 ⊂ 𝑅2 is an ideal, then 𝜑−1 (𝐼) ⊂ 𝑅1 is
also an ideal.

Proposition 1.1.4. Given 𝜑 as above, it induces a continuous map

𝜑♯ ∶ Spec(𝑅2 ) ⟶ Spec(𝑅1 )
𝔭 ⟼ 𝜑−1 (𝔭)

with respect to the Zariski topologies on spectra.

Proof. Clearly, 𝑎𝑏 ∈ 𝜑−1 (𝔭) is equivalent to 𝜑(𝑎)𝜑(𝑏) ∈ 𝔭, which is in turn equivalent

to (𝜑(𝑎) ∈ 𝔭) ∨ (𝜑(𝑏) ∈ 𝔭) when 𝔭 is prime.
To show the continuity of 𝜑♯ , observe that for any ideal 𝔞 ⊂ 𝑅1 and 𝔭 ∈ Spec(𝑅2 ),
we have 𝜑−1 (𝔭) ⊃ 𝔞 if and only if 𝔭 ⊃ 𝜑(𝔞), i.e. 𝔭 ∈ 𝑉(𝜑(𝔞)𝑅2 ). Hence the preimage of
closed subsets are still closed.
More operations on spectra:

⋄ Take 𝑅1 to be a subring of 𝑅2 and 𝜑 be the inclusion map, the map above becomes
𝔭 ↦ 𝔭 ∩ 𝑅1 .

⋄ Take 𝜑 ∶ 𝑅 ↠ 𝑅/𝐼 to be a quotient homomorphism, then 𝜑−1 is the usual bijection

from Spec(𝑅/𝐼) onto 𝑉(𝐼).

⋄ In general, 𝜑−1 does not induce MaxSpec(𝑅2 ) → MaxSpec(𝑅1 ), as illustrated in

the case 𝜑 ∶ ℤ ↪ ℚ.

At this stage, we can prove a handy result concerning prime ideals.

Proposition 1.1.5 (Prime avoidance). Let 𝐼 and 𝔭1 , … , 𝔭𝑛 be ideals of 𝑅 such that 𝐼 ⊂

𝑛
⋃𝑖=1 𝔭𝑖 . Suppose that
⋄ either 𝑅 contains an infinite field, or
⋄ at most two of the ideals 𝔭1 , … , 𝔭𝑛 are non-prime,
then there exists 1 ≤ 𝑖 ≤ 𝑛 such that 𝐼 ⊂ 𝔭𝑖 .
 §1.2 Localization of rings and modules ⋅5⋅

Proof. If 𝑅 contains an infinite field 𝐹, the ideals are automatically 𝐹-vector subspaces
𝑟
of 𝑅. Since 𝐼 = ⋃𝑖=1 𝐼 ∩ 𝔭𝑖 whereas an 𝐹-vector space cannot be covered by finitely
many proper subspaces, there must exist some 𝑖 with 𝐼 ∩ 𝔭𝑖 = 𝐼.
Under the second assumption, let us argue by induction on 𝑛 that ∀𝑖 𝐼 ⊄ 𝔭𝑖 implies
𝑛
𝐼 ⊄ ⋃𝑖=1 𝔭𝑖 . The case 𝑛 = 1 is trivial. When 𝑛 ≥ 2, by induction we may choose, for
𝑛
each 𝑖, an element 𝑥𝑖 ∈ 𝐼 ∖ ⋃𝑗≠𝑖 𝔭𝑗 . Suppose on the contrary that 𝐼 ⊂ ⋃𝑗=1 𝔭𝑗 , then we
would have 𝑥𝑖 ∈ 𝔭𝑖 , for all 𝑖 = 1, … , 𝑛.
When 𝑛 = 2 we have 𝑥1 + 𝑥2 ∉ 𝔭1 ∪ 𝔭2 and 𝑥1 + 𝑥2 ∈ 𝐼, a contradiction. When 𝑛 > 2,
we may assume 𝔭1 is prime, therefore

𝑛 𝑛
𝑥1 + ∏ 𝑥𝑗 ∉ ⋃ 𝔭𝑖 ,
𝑗=2 𝑖=1

again a contradiction.

Exercise 1.1.6. The following construction from [8, Exercise 3.17] shows that the as-
sumptions of Proposition 1.1.5 cannot be weakened. Take 𝑅 = (ℤ/2ℤ)[𝑋, 𝑌]/(𝑋, 𝑌)2 ,
which has a basis {1, 𝑋, 𝑌} (modulo (𝑋, 𝑌)2 ) as a ℤ/2ℤ-vector space. Show that the
image 𝔪 of (𝑋, 𝑌) in 𝑅 is the unique prime ideal, and can be expressed as a union of
three ideals properly contained in 𝔪.

1.2 Localization of rings and modules

Let 𝑆 be a multiplicative subset of 𝑅, which means that (a) 1 ∈ 𝑆, (b) 𝑆 is closed under
multiplication, and (c) 0 ∉ 𝑆. The localization of 𝑅 with respect to 𝑆 is the ring 𝑅[𝑆−1 ]
formed by classes [𝑟, 𝑠] with 𝑟 ∈ 𝑅, 𝑠 ∈ 𝑆, modulo the equivalence relation

[𝑟, 𝑠] = [𝑟′ , 𝑠′ ] ⟺ ∃𝑡 ∈ 𝑆, (𝑟𝑠′ − 𝑟′ 𝑠)𝑡 = 0.

You should regard [𝑟, 𝑠] as a token for 𝑟/𝑠; the ring structure of 𝑅[𝑆−1 ] is therefore
evident. In brief, localization amounts to formally inverting the elements of 𝑆, whence
the notation 𝑅[𝑆−1 ]. Note that condition (c) guarantees 𝑅[𝑆−1 ] ≠ {0}.

Exercise 1.2.1. Given 𝑅 and 𝑆, show that 𝑟 ↦ 𝑟/1 yields a natural homomorphism
𝑅 → 𝑅[𝑆−1 ] and show that its kernel equals {𝑟 ∶ ∃𝑠 ∈ 𝑆, 𝑠𝑟 = 0}.

The universal property of 𝑅 → 𝑅[𝑆−1 ] can be stated using commutative diagrams

as follows.

⎧ 𝜑 ∶ 𝑅 → 𝑅′ ∶ ring homomorphism 𝑅 𝑅[𝑆−1 ]

{
∀⎨ , ∃!
{ s.t. 𝜑(𝑆) ⊂ (𝑅′ )× 𝜑
⎩
𝑅′
⋅6⋅ Warming up

Consequently, if 𝑆 ⊂ 𝑅× then 𝑅 ≃ 𝑅[𝑆−1 ] canonically. Furthermore, the homomor-

phism 𝑅 → 𝑅[𝑆−1 ] induces a bijection

1∶1
Spec(𝑅[𝑆−1 ]) {𝔭 ∈ Spec(𝑅) ∶ 𝔭 ∩ 𝑆 = ∅} ⊂ Spec(𝑅)

𝔭𝑅[𝑆−1 ] = {𝑟/𝑠 ∶ 𝑟 ∈ 𝔭, 𝑠 ∈ 𝑆} 𝔭 (1–1)

𝔮 its preimage.

Exercise 1.2.2. Check the properties above.

Let us review some important instances of the localization procedure.

1. Take 𝑆 to be the subset of non zero-divisors of 𝑅. This is easily seen to be a mul-

tiplicative subset (check it!) and 𝐾(𝑅) ∶= 𝑅[𝑆−1 ] is called the total fraction ring of
𝑅. The reader is invited to check that 𝑅 → 𝐾(𝑅) is the “biggest localization” such
that the natural homomorphism 𝑅 → 𝑅[𝑆−1 ] is injective. Hint: state this in terms
of universal properties.
When 𝑅 is an integral domain, we shall take 𝑆 ∶= 𝑅 ∖ {0}; in this case the total
fraction ring Frac(𝑅) ∶= 𝐾(𝑅) is the well-known field of fractions of 𝑅.

2. Take any 𝔭 ∈ Spec(𝑅) and 𝑆 ∶= 𝑅 ∖ 𝔭. From the definition of prime ideals, one
infers that 𝑆 is a multiplicative subset of 𝑅. The corresponding localization is
denoted by 𝑅 → 𝑅𝔭 ∶= 𝑅[𝑆−1 ]. We see from (1–1) that

MaxSpec(𝑅𝔭 ) = {𝔭𝑅𝔭 };

in particular, 𝑅𝔭 is a local ring with maximal ideal 𝔭𝑅𝔭 . This is the standard way
to produce local rings; we say that 𝑅𝔭 is the localization of 𝑅 at the prime 𝔭.

3. Suppose 𝑓 ∈ 𝑅 is not nilpotent, that is, 𝑓 𝑛 ≠ 0 for every 𝑛. Take 𝑆 ∶= {𝑓 𝑛 ∶ 𝑛 ≥ 0}.

The corresponding localization is denoted by the self-explanatory notation 𝑅 →
𝑅[𝑓 −1 ].

Exercise 1.2.3. Describe the following localizations explicitly.

(a) 𝑅 = ℤ, and we localize at the prime ideal (𝑝) where 𝑝 is a prime number.

(b) 𝑅 = ℂ[𝑋1 , … , 𝑋𝑛 ] and we localize at the maximal ideal generated by 𝑋1 , … , 𝑋𝑛 .

(c) 𝑅 = ℂJ𝑋K (the ring of formal power series) and 𝑆 ∶= {𝑋 𝑛 ∶ 𝑛 ≥ 0}.

∼
Exercise 1.2.4. Prove that 𝑅[𝑋]/(𝑓 𝑋 − 1) → 𝑅[𝑓 −1 ] by mapping 𝑋 ↦ 𝑓 −1 . Hint: use the
universal property.

Always let 𝑆 ⊂ 𝑅 be a multiplicative subset. The localization 𝑀[𝑆−1 ] of an 𝑅-

module 𝑀 can be defined in the manner above, namely as the set of equivalence classes
[𝑚, 𝑠] with 𝑚 ∈ 𝑀 and 𝑠 ∈ 𝑆, such that

[𝑚, 𝑠] = [𝑚′ , 𝑠′ ] ⟺ ∃𝑡 ∈ 𝑆, 𝑡(𝑠′ 𝑚 − 𝑠𝑚′ ) = 0.

 §1.2 Localization of rings and modules ⋅7⋅

As in the case of rings, we shall write 𝑚/𝑠 instead of [𝑚, 𝑠]. It is an 𝑅[𝑆−1 ]-module,
equipped with a natural homomorphism 𝑀 → 𝑀[𝑆−1 ] of 𝑅-modules. This yields a
functor: for any homomorphism 𝑓 ∶ 𝑀 → 𝑁 we have a natural 𝑓 [𝑆−1 ] ∶ 𝑀[𝑆−1 ] →
𝑁[𝑆−1 ], mapping 𝑚/𝑠 to 𝑓 (𝑚)/𝑠; furthermore 𝑓 [𝑆−1 ] ∘ 𝑔[𝑆−1 ] = (𝑓 ∘ 𝑔)[𝑆−1 ] whenever
composition makes sense.
∼
A slicker interpretation is to use the natural isomorphism 𝑀[𝑆−1 ] → 𝑅[𝑆−1 ] ⊗ 𝑀
𝑅
which maps 𝑚/𝑠 to (1/𝑠) ⊗ 𝑚. Hereafter, we shall identify 𝑀[𝑆−1 ] and 𝑅[𝑆−1 ] ⊗ 𝑀
𝑅
without further comments.
In the same vein, we may define 𝑀[𝑓 −1 ] and 𝑀𝔭 , for non-nilpotent 𝑓 ∈ 𝑅 and
𝔭 ∈ Spec(𝑅) respectively. Localization “commutes” with several standard operation
on modules, which we sketch below. The details are left to the reader.

⋄ For 𝑅-modules 𝑀, 𝑁, we have a natural isomorphism of 𝑅[𝑆−1 ]-modules

(𝑀 ⊗ 𝑁)[𝑆−1 ] ≃ 𝑀[𝑆−1 ] ⊗ 𝑁[𝑆−1 ],

𝑅 𝑅[𝑆−1 ]
(𝑀 ⊕ 𝑁)[𝑆−1 ] ≃ 𝑀[𝑆−1 ] ⊕ 𝑁[𝑆−1 ].

Same for arbitrary direct sums. This is easily seen by viewing 𝑀[𝑆−1 ] as 𝑀 ⊗𝑅
𝑅[𝑆−1 ].

⋄ Note that Hom𝑅 (𝑀, 𝑁) is also an 𝑅-module: simply set (𝑟𝑓 )(𝑚) = 𝑟 ⋅ 𝑓 (𝑚) for any
𝑓 ∈ Hom𝑅 (𝑀, 𝑁). There is a natural homomorphism

Hom𝑅 (𝑀, 𝑁)[𝑆−1 ] → Hom𝑅[𝑆−1 ] (𝑀[𝑆−1 ], 𝑁[𝑆−1 ]) (1–2)

sending 𝑠−1 ⊗ 𝜑 to 𝑠−1 𝜑 ∶ 𝑚/𝑡 ↦ 𝜑(𝑚)/𝑠𝑡. It is clearly an isomorphism for 𝑀 = 𝑅,

thus also for 𝑀 = 𝑅𝑎 where 𝑎 ∈ ℤ≥0 , but this is not the case in general, as there
is no uniform bound for the “denominators” for any given 𝜓 ∶ 𝑀[𝑆−1 ] → 𝑁[𝑆−1 ]
— some finiteness condition is needed. Let us assume 𝑀 to be finitely presented,
i.e. there is an exact sequence

𝑅𝑎 → 𝑅𝑏 → 𝑀 → 0, 𝑎, 𝑏 ∈ ℤ≥0 .

In this case (1–2) is an isomorphism, as easily seen from the commutative dia-
gram with exact rows:

0 Hom(𝑀, 𝑁)[𝑆−1 ] Hom(𝑅𝑏 , 𝑁)[𝑆−1 ] Hom(𝑅𝑎 , 𝑁)[𝑆−1 ]

≃ ≃

0 Hom(𝑀[𝑆−1 ], 𝑁[𝑆−1 ]) Hom(𝑅[𝑆−1 ]𝑏 , 𝑁[𝑆−1 ]) Hom(𝑅[𝑆−1 ]𝑎 , 𝑁[𝑆−1 ])

Here we used the fact that localization preserves exactness: see the Proposition
1.2.5 below.

⋄ Let 𝜑 ∶ 𝑅 → 𝑅′ be a ring homomorphism and 𝑆 ⊂ 𝑅 be a multiplicative subset,

so that 𝑆′ ∶= 𝜑(𝑆) ⊂ 𝑅′ is also multiplicative. View 𝑅′ as an 𝑅-module, then we
⋅8⋅ Warming up

have
∼
𝑅′ [(𝑆
⎵⎵ ⎵′⎵⎵
)−1 ] 𝑅′ [𝑆−1 ]
⎵⎵⎵ 𝑅[𝑆−1 ] ⊗ 𝑅′
𝑅
as ring as module

𝑟′ /𝜑(𝑠) (1/𝑠) ⊗ 𝑟′

𝑟′ 𝜑(𝑟)/𝜑(𝑠) (𝑟/𝑠) ⊗ 𝑟′

⋄ As a special case, take 𝜑 to be a quotient homomorphism 𝑅 → 𝑅/𝐼, we get a

natural isomorphism

𝑅[𝑆−1 ]
(𝑅/𝐼)[(𝑆′ )−1 ] ≃ 𝑅[𝑆−1 ] ⊗ (𝑅/𝐼) =
𝑅 𝐼[𝑆−1 ]

from the right exactness of ⊗.

We prove an easy yet fundamental property of localizations, namely they are exact
functors.

Proposition 1.2.5 (Exactness of localization). Let 𝑆 be any multiplicative subset of 𝑅. If

𝑓𝑖
⋯ → 𝑀𝑖 𝑀𝑖+1 → ⋯

is an exact sequence of 𝑅-modules, then

𝑓𝑖 [𝑆−1 ]
⋯ → 𝑀𝑖 [𝑆−1 ] 𝑀𝑖+1 [𝑆−1 ] → ⋯

is also exact.

Proof. By homological common sense, we are reduced to the exact sequences (i) 0 →
𝑀′ → 𝑀 → 𝑀″ (i.e. left exactness), (ii) 𝑀′ → 𝑀 → 𝑀″ → 0 (i.e. right exactness). The
case (ii) is known for tensor products in general.
As for (i), note that for every homomorphism 𝑔 ∶ 𝑀 → 𝑀″ ,

𝑦 ∵𝑦/𝑠=𝑡𝑦/𝑡𝑠 𝑦
ker (𝑔[𝑆−1 ]) = { ∈ 𝑀[𝑆−1 ] ∶ ∃𝑡 ∈ 𝑆, 𝑡𝑔(𝑦) = 0} { ∶ 𝑦 ∈ ker(𝑔)}
𝑠 𝑠
= im [ker(𝑔)[𝑆−1 ] → 𝑀[𝑆−1 ]] .

Thus it remains to show that if 𝑓 ∶ 𝑀′ ↪ 𝑀, then 𝑓 [𝑆−1 ] ∶ 𝑀′ [𝑆−1 ] → 𝑀[𝑆−1 ] is injective

as well. If 𝑥/𝑠 ↦ 𝑓 (𝑥)/𝑠 = 0 under 𝑓 [𝑆−1 ], there exists 𝑡 ∈ 𝑆 such that 𝑡𝑓 (𝑥) = 𝑓 (𝑡𝑥) = 0
in 𝑀, therefore 𝑡𝑥 = 0 in 𝑀′ , but the latter condition implies 𝑥/𝑠 = 0 in 𝑀′ [𝑆−1 ].

Lemma 1.2.6. Let 𝑀 be an 𝑅-module. The localizations 𝑀 → 𝑀𝔪 for various maximal ideals
𝔪 assemble into an injection
𝑀↪ ∏ 𝑀𝔪 .
𝔪∈MaxSpec(𝑅)
 §1.3 Radicals and Nakayama’s lemma ⋅9⋅

Proof. Let 𝑚 ∈ 𝑀 be such that 𝑚 ↦ 0 ∈ 𝑀𝔪 for all 𝔪. This means that for all 𝔪 there
exists 𝑠 ∈ 𝑅 ∖ 𝔪 such that 𝑠𝑚 = 0. Hence the annihilator ideal ann𝑅 (𝑚) ∶= {𝑟 ∈ 𝑅 ∶
𝑟𝑚 = 0} is not contained in any maximal ideal, thus ann𝑅 (𝑚) = 𝑅.
There is an analogue for rings. Observe that when 𝑅 is an integral domain, all the
localizations 𝑅[𝑆−1 ] can be regarded as subrings of the field of fractions Frac(𝑅).
Lemma 1.2.7. Let 𝑅 be an integral domain, then 𝑅 = ⋂𝔪∈MaxSpec(𝑅) 𝑅𝔪 as subrings of
Frac(𝑅).
Proof. Only the inclusion ⊃ requires proof. Let 𝑥 ∈ Frac(𝑅) and define 𝐷 ∶= {𝑟 ∈ 𝑅 ∶
𝑟𝑥 ∈ 𝑅} (the ideal of denominators). Suppose 𝑥 ∈ 𝑅𝔪 for all maximal 𝔪, then 𝐷 ⊄ 𝔪
for all maximal 𝔪. The same reasoning as above leads to 𝐷 = 𝑅.
It will be important to gain finer control on the ideals 𝔪 in the assertion above, say
by using some prime ideals “lower” than the maximal ones. We will return to this issue
later.

1.3 Radicals and Nakayama’s lemma

We begin with two important notions of radicals. The first version is defined as follows.
Given an ideal 𝐼 of 𝑅, the nilpotent radical √𝐼 is defined to be

√𝐼 ∶= {𝑟 ∈ 𝑅 ∶ ∃𝑛, 𝑟𝑛 ∈ 𝐼}.
2𝑛
It is readily seen to be an ideal from the binomial identity (𝑎+𝑏)2𝑛 = ∑𝑘=0 (2𝑛 𝑘 2𝑛−𝑘 .
𝑘 )𝑎 𝑏
It should also be clear that √𝐼 ⊂ 𝑅 equals the preimage of √0 ⊂ 𝑅/𝐼.

Exercise 1.3.1. Show that √√𝐼 = 𝐼 for all ideal 𝐼.

Proposition 1.3.2. Let 𝐼 be a proper ideal of 𝑅. We have

√𝐼 = ⋂ 𝔭.
𝔭∈Spec(𝑅)
𝔭⊃𝐼

Proof. By replacing 𝑅 by 𝑅/𝐼, this is easily reduced to the case 𝐼 = {0}. If 𝑟 is nilpotent
and 𝔭 ∈ Spec(𝑅), then 𝑟𝑛 = 0 ∈ 𝔭 implies 𝑟 ∈ 𝔭. Conversely, suppose that 𝑟 is not
nilpotent. There exists a prime ideal in 𝑅[𝑟−1 ], which comes from some 𝔭 ∈ Spec(𝑅)
with 𝑟 ∈ 𝑅 ∖ 𝔭 by (1–1). Hence 𝑟 ∉ 𝔭.

Remark 1.3.3. A ring 𝑅 is called reduced if √0 = {0}. In any case, 𝑅red ∶= 𝑅/√0 is a
reduced ring. Furthermore, any reduced quotient of 𝑅 factors through 𝑅red .
The second radical is probably familiar to the readers. The Jacobson radical rad(𝑅)
of the ring 𝑅 is the intersection of all maximal ideals. The previous proposition implies
rad(𝑅) ⊃ √(0). Note that

𝑎 ∈ rad(𝑅) ⟹ (1 + 𝑎) ∈ 𝑅× .
⋅ 10 ⋅ Warming up

Indeed, 1+𝑎 cannot be contained in any maximal ideal 𝔪, for otherwise 1 = (1+𝑎)−𝑎 ∈
𝔪, which is absurd.
To prove the celebrated Nakayama’s Lemma, let us recall an easy variant of the
Cayley–Hamilton theorem from linear algebra.
Lemma 1.3.4. Suppose that 𝐼 ⊂ 𝑅 is an ideal, 𝑀 is an 𝑅-module with generators 𝑥1 , … , 𝑥𝑛 and
𝜑 ∈ End𝑅 (𝑀) satisfies 𝜑(𝑀) ⊂ 𝐼𝑀, then there exists a polynomial 𝑃(𝑋) = 𝑋 𝑛 +𝑎𝑛−1 𝑋 𝑛−1 +
⋯ + 𝑎0 ∈ 𝑅[𝑋] with 𝑎𝑖 ∈ 𝐼, such that 𝑃(𝜑) = 𝜑𝑛 + 𝑎𝑛−1 𝜑𝑛−1 + ⋯ + 𝑎0 = 0.
𝑛
Proof. Write 𝜑(𝑥𝑖 ) = ∑𝑗=1 𝑎𝑖𝑗 𝑥𝑗 where 𝑎𝑖𝑗 ∈ 𝐼. Set 𝐴 ∶= (𝑎𝑖𝑗 )1≤𝑖,𝑗≤𝑛 ∈ Mat𝑛 (𝑅). Regard
𝑀 as an 𝑅[𝑋]-module by letting 𝑋 act as 𝜑. Then we have the matrix equation over
𝑅[𝑋]
⎛
⎜ 𝑥1 ⎞
⎟ ⎛
⎜ 0⎞⎟
⎜
⎜ ⎟
⎟ ⎜
⎜ ⎟
⎟
⎜
(𝑋 ⋅ id𝑛×𝑛 − 𝐴) ⎜ ⋮ ⎟
⎟ = ⎜
⎜ ⋮ ⎟
⎟
⎜
⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜ ⎟
⎟ ⎜ ⎟
⎝𝑥𝑛 ⎠ ⎝0⎠
Multiplying by the cofactor matrix (𝑋 ⋅ id𝑀 − 𝐴)∨ on the left, we see that 𝑃(𝑋) ∶=
det(𝑋 ⋅ id𝑛×𝑛 − 𝐴) ∈ 𝑅[𝑋] acts as 0 on each 𝑥𝑖 , thus on the whole 𝑀. This is the
required polynomial.
Theorem 1.3.5 (Nakayama’s Lemma). Suppose that 𝑀 is a finitely generated 𝑅-module and
𝐼 is an ideal of 𝑅 such that 𝐼𝑀 = 𝑀. Then there exists 𝑎 ∈ 𝐼 such that (1 + 𝑎)𝑀 = 0. If
𝐼 ⊂ rad(𝑅), then we have 𝑀 = {0} under these assumptions.
Proof. Write 𝑀 = 𝑅𝑥1 + ⋯ + 𝑅𝑥𝑛 . Plug 𝜑 = id𝑀 into Lemma 1.3.4 to deduce that
𝑛−1 + ⋯ + 𝑎0 acts as 0 on 𝑀. This proves the first part. Assume further-
𝑃(id𝑀 ) = 1 + 𝑎⎵⎵⎵⎵⎵⎵⎵
=∶𝑎∈𝐼
more that 𝐼 ⊂ rad(𝑅), then 1 + 𝑎 ∈ 𝑅× so that 𝑀 = (1 + 𝑎)𝑀 = 0.

Nakayama’s Lemma is named after Tadashi Nakayama (1912—

1964). Picture borrowed from [14].

Corollary 1.3.6. Let 𝑀 be a finitely generated 𝑅-module, and let 𝐼 ⊂ rad(𝑅) be an ideal of 𝑅.
If the images of 𝑥1 , … , 𝑥𝑛 ∈ 𝑀 in 𝑀/𝐼𝑀 form a set of generators, then 𝑥1 , … , 𝑥𝑛 generate 𝑀.
 §1.4 Noetherian and Artinian rings ⋅ 11 ⋅

Proof. Apply Theorem 1.3.5 to 𝑁 ∶= 𝑀/(𝑅𝑥1 + ⋯ + 𝑅𝑥𝑛 ); our assumption 𝑀 = 𝐼𝑀 +

𝑅𝑥1 + ⋯ + 𝑅𝑥𝑛 entails that 𝐼𝑁 = 𝑁, thus 𝑁 = 0.
We record another amusing consequence of Theorem 1.3.5.
Proposition 1.3.7. Let 𝑀 be a finitely generated 𝑅-module and 𝜓 ∈ End𝑅 (𝑀). If 𝜓 is sur-
jective then 𝜓 is an automorphism.
Proof. Introduce a variable 𝑌. Make 𝑀 into an 𝑅[𝑌]-module by letting 𝑌 act as 𝜓. Put
𝐼 ∶= (𝑌) so that 𝐼𝑀 = 𝑀. Theorem 1.3.5 yields some 𝑄(𝑌) ∈ 𝑅[𝑌] satisfying (1 −
𝑄(𝑌)𝑌)𝑀 = 0, that is, 𝑄(𝜓)𝜓 = id𝑀 .

1.4 Noetherian and Artinian rings

An 𝑅-module 𝑀 is called Noetherian (resp. Artinian) if every ascending (resp. descend-
ing) chain of submodules eventually stabilizes. Recall that in a short exact sequence
0 → 𝑀′ → 𝑀 → 𝑀″ → 0, we have 𝑀 is Noetherian (resp. Artinian) if and only if 𝑀′
and 𝑀″ are. Being both Noetherian and Artinian is equivalent to being a module of
finite length, that is, a module admitting composition series.
If we take 𝑀 ∶= 𝑅 on which 𝑅 acts by multiplication, then the submodules are
precisely the ideals of 𝑅. We say that 𝑅 is a Noetherian (resp. Artinian) ring if 𝑅 as
an 𝑅-module is Noetherian (resp. Artinian); this translates into the corresponding
chain conditions on the ideals. Both chain conditions are preserved under passing to
quotients and localizations. Finitely generated modules over a Noetherian ring are
Noetherian. The following result ought to be known to the readers.
Theorem 1.4.1 (Hilbert’s Basis Theorem). If 𝑅 is Noetherian, then so is the polynomial
algebra 𝑅[𝑋1 , … , 𝑋𝑛 ] for any 𝑛 ∈ ℤ≥1 .
Joint with the foregoing remarks, we infer that finitely generated algebras over
Noetherian rings are still Noetherian.
On the other hand, being Artinian is a rather stringent condition on rings.
Theorem 1.4.2. A ring 𝑅 is Artinian if and only if 𝑅 is of finite length as an 𝑅-module. Such
rings are semi-local.
Proof. As noticed before, having finite length implies that 𝑅 is Noetherian as well as
Artinian. Assume conversely that 𝑅 is an Artinian ring. First we claim that MaxSpec(𝑅)
is finite, i.e. 𝑅 is semi-local. If there were an infinite sequence of distinct maximal ideals
𝔪1 , 𝔪2 , …, we would have an infinite chain

𝔪1 ⊃ 𝔪1 𝔪2 ⊃ 𝔪1 𝔪2 𝔪3 ⊃ ⋯ .

This chain is strictly descending, since 𝔪1 ⋯ 𝔪𝑖 = 𝔪1 ⋯ 𝔪𝑖+1 would imply 𝔪𝑖+1 ⊃

𝔪1 ⋯ 𝔪𝑖 , hence 𝔪𝑖+1 ⊃ 𝔪𝑗 for some 1 ≤ 𝑗 ≤ 𝑖 because maximal ideals are prime. From
the Artinian property we conclude that there are only finitely many maximal ideals
𝔪1 , … , 𝔪𝑛 of 𝑅.
Set 𝔞 ∶= 𝔪1 ⋯ 𝔪𝑛 . Since 𝑅 is Artinian we must have 𝔞𝑘 = 𝔞𝑘+1 for some 𝑘 > 0. We
claim that 𝔞𝑘 = 0.
⋅ 12 ⋅ Warming up

Put 𝔟 ∶= {𝑟 ∈ 𝑅 ∶ 𝑟𝔞𝑘 = 0}, we have to show 𝔟 = 𝑅. If not, let 𝔟′ be a minimal ideal

lying strictly over 𝔟. Thus 𝔟′ = 𝑅𝑥 + 𝔟 for any 𝑥 ∈ 𝔟′ ∖ 𝔟. We must have 𝔞𝑥 + 𝔟 ⊊ 𝔟′ ,
for otherwise 𝑀 ∶= 𝔟′ /𝔟 is finitely generated (say by 𝑥) and satisfies 𝑀 = 𝔞𝑀, then
Theorem 1.3.5 plus 𝔞 ⊂ rad(𝑅) would imply 𝑀 = {0}, which is absurd. By minimality
we have 𝔟 = 𝔞𝑥 + 𝔟. It follows that 𝔞𝑥 ⊂ 𝔟, i.e. 𝔞𝑘+1 𝑥 = 0; from 𝔞𝑘 = 𝔞𝑘+1 we infer
𝑥 ∈ 𝔟. Contradiction.
All in all, we obtain a descending chain of ideals

𝑅 ⊃ 𝔪1 ⊃ 𝔪1 𝔪2 ⊃ ⋯ ⊃ 𝔪1 ⋯ 𝔪𝑛 = 𝔞
⊃ 𝔞𝔪1 ⊃ 𝔞𝔪1 𝔪2 ⊃ ⋯ ⊃ 𝔞𝔪1 ⋯ 𝔪𝑛 = 𝔞2
⊃ ⋯ ⊃ 𝔞𝑘 = {0}.

Each subquotient thereof, which is a priori an 𝑅-module, is actually an 𝑅/𝔪𝑖 -vector

space for some 1 ≤ 𝑖 ≤ 𝑛. Such a vector space must also satisfy the descending chain
condition on vector subspaces, otherwise pulling-back will contradict the Artinian as-
sumption on 𝑅. Artinian vector spaces must be finite-dimensional. It follows that all
these subquotients are of finite length, hence so is 𝑅 itself.

Corollary 1.4.3. A ring 𝑅 is Artinian if and only if it is Noetherian and every prime ideal of
𝑅 is maximal.

Proof. If 𝑅 is Artinian, then 𝑅 is of finite length, hence is Noetherian as well. For every
prime ideal 𝔭, we have 𝔭 ⊃ {0} = (𝔪1 ⋯ 𝔪𝑛 )𝑘 in the notations of the proof above,
therefore 𝔭 ⊃ 𝔪𝑖 for some 𝑖, so 𝔭 = 𝔪𝑖 is maximal.
Conversely, if 𝑅 is Noetherian but of infinite length, then the nonempty set of ideals

𝒮 ∶= {ideals 𝐼 ⊊ 𝑅 ∶ 𝑅/𝐼 has infinite length}

contains a maximal element 𝔭. We contend that 𝔭 is prime. If 𝑥𝑦 ∈ 𝔭 with 𝑥, 𝑦 ∉ 𝔭, then

𝑅/(𝔭 + 𝑅𝑥) has finite length by the choice of 𝔭; on the other hand, 𝔞 ∶= {𝑟 ∈ 𝑅 ∶ 𝑟𝑥 ∈
𝔭} ⊋ 𝔭 (as 𝑦 ∈ 𝔞), hence 𝑅/𝔞 has finite length as well. From the short exact sequence
𝑥
0 → 𝑅/𝔞 𝑅/𝔭 → 𝑅/(𝔭 + 𝑅𝑥) → 0 we see 𝑅/𝔭 has finite length, contradiction.
If we assume moreover that every prime ideal is maximal, then for the 𝔭 chosen
above, 𝑅/𝔭 will be a field, thus of finite length. This is impossible.

Rings whose prime ideals are all maximal are said to have dimension zero, in the
sense of Krull dimensions; we shall return to this point in §5.4.

1.5 What is commutative algebra?

In broad terms, commutative algebra is the study of commutative rings. Despite its in-
trinsic beauty, we prefer to motivate from an external point of view. See also [8].
(A) Algebraic geometry. To simplify matters, we consider affine algebraic vari-
eties over an algebraically closed field 𝕜. Roughly speaking, such a variety is the zero
locus 𝒳 = {𝑓1 = ⋯ = 𝑓𝑟 = 0} in 𝔸𝑛 = 𝕜𝑛 of 𝑓𝑖 ∈ 𝕜[𝑋1 , … , 𝑋𝑛 ]. The choice of equations
is of course non-unique: what matters is the ideal 𝐼(𝒳) ∶= {𝑓 ∈ 𝕜[𝑋1 , … , 𝑋𝑛 ] ∶ 𝑓 |𝒳 = 0}.
 §1.5 What is commutative algebra? ⋅ 13 ⋅

Conversely, every ideal 𝔞 defines a subset 𝑉(𝔞) ∶= {𝑥 ∈ 𝕜𝑛 ∶ ∀𝑓 ∈ 𝔞, 𝑓 (𝑥) = 0}. As

consequences of Hilbert’s Nullstellensatz, which we will discuss later, we have

𝑉 ∘ 𝐼(𝒳) = 𝒳, 𝐼 ∘ 𝑉(𝔞) = √𝔞.

One can deduce from this that the (closed) subvarieties of 𝔸𝑛 are in bijection with
ideals 𝔞 satisfying √𝔞 = 𝔞.
Furthermore, 𝕜[𝒳] ∶= 𝕜[𝑋1 , … , 𝑋𝑛 ]/𝐼(𝒳) may be regarded as the ring of “regular
functions” (i.e. functions definable by means of polynomials) on 𝒳, and MaxSpec(𝕜[𝒳])
is in bijection with the points of 𝒳: to 𝑥 = (𝑥1 , … , 𝑥𝑛 ) ∈ 𝒳 we attach

𝔪𝑥 = (𝑋1 − 𝑥1 , … , 𝑋𝑛 − 𝑥𝑛 ) ⊃ 𝐼(𝒳).

By passing to the ring 𝕜[𝒳], we somehow obtain a description of 𝒳 that is independent

of embeddings into affine spaces. Moreover, 𝒳 inherits the Zariski topology from that
on Spec(𝕜[𝒳]).
Naively speaking, the geometric properties of 𝒳 transcribe in ring-theoretic terms to the
reduced Noetherian 𝕜-algebra 𝕜[𝒳]. For example, assume that 𝑓 ∈ 𝕜[𝒳] is not nilpo-
tent, then the formation of 𝕜[𝒳][𝑓 −1 ] corresponds to taking the Zariski-open subset
𝒳𝑓 = {𝑥 ∈ 𝒳 ∶ 𝑓 (𝑥) ≠ 0} of 𝒳. This may be explained as follows:

𝒳𝑓 ≃ {(𝑥1 , … , 𝑥𝑛 , 𝑦) ∶ 𝑓1 (𝑥1 , … , 𝑥𝑛 ) = ⋯ = 𝑓𝑟 (𝑥1 , … , 𝑥𝑛 ) = 0, 𝑓 (𝑥1 , … , 𝑥𝑛 )𝑦 = 1}

which is also an affine algebraic variety in 𝕜𝑛+1 , and one may verify that 𝕜[𝒳𝑓 ] =
𝕜[𝒳][𝑓 −1 ].
(B) Invariant theory. Let 𝐺 be a group acting on a finite-dimensional 𝕜-vector
space 𝑉 from the right, and let 𝕜[𝑉] be the 𝕜-algebra of polynomials on 𝑉. Thus 𝕜[𝑉]
carries a left 𝐺-action by 𝑔𝑓 (𝑣) = 𝑓 (𝑣𝑔). For “reasonable” groups 𝐺, say finite or 𝕜-
algebraic ones, the classical invariant theory seeks to describe the subalgebra 𝕜[𝑉]𝐺 of
invariants1 in terms of generators and relations.
In particular, one has to know when is the algebra 𝕜[𝑉]𝐺 finitely generated. This is
actually the source of many results in commutative algebra, such as the Basis Theorem
and Nullstellensatz of Hilbert. For example, let the symmetric group 𝐺 = 𝔖𝑛 act on
𝑉 = 𝕜𝑛 in the standard manner, then our question is completely answered by the
following classical result: 𝕜[𝑉]𝐺 equals the polynomial algebra 𝕜[𝑒1 , … , 𝑒𝑛 ], where 𝑒𝑖
stands for the 𝑖-th elementary symmetric function in 𝑛 variables.
The same questions may be posed for any affine algebraic variety 𝑉. From the geo-
metric point of view, if 𝕜[𝑉]𝐺 is finitely generated, it will consist of regular functions
of some kind of quotient variety 𝑉//𝐺. The study of quotients in this sense naturally
leads to geometric invariant theory, for which we refer to [16] for details.

1
More generally, we are also interested in the algebra of invariant differential operators with polyno-
mial coefficients.
⋅ 14 ⋅ Warming up
 Lecture 2

Primary decompositions
We shall follow [8, §3] and [11, §8] closely.

2.1 The support of a module

Fix a ring 𝑅. For any 𝑅-module 𝑀 and 𝑥 ∈ 𝑀, we define the annihilator ann𝑅 (𝑥) ∶=
{𝑟 ∈ 𝑅 ∶ 𝑟𝑥 = 0}; it is an ideal of 𝑅. Also define ann𝑅 (𝑀) ∶= {𝑟 ∈ 𝑅 ∶ 𝑟𝑀 = 0} =
⋂𝑥∈𝑀 ann𝑅 (𝑥).

Definition 2.1.1. The support of an 𝑅-module 𝑀 is

Supp(𝑀) ∶= {𝔭 ∈ Spec(𝑅) ∶ 𝑀𝔭 ≠ 0} .

Let us unwind the definition: 𝔭 ∉ Supp(𝑀) means that every 𝑥 ∈ 𝑀 maps to

0 ∈ 𝑀𝔭 , equivalently ann𝑅 (𝑥) ⊄ 𝔭. To test whether 𝔭 ∉ Supp(𝑀), we only need to
check the foregoing condition for 𝑥 ranging over a generating set of 𝑀.

Proposition 2.1.2. If 𝑀 is finitely generated then Supp(𝑀) = 𝑉(ann𝑅 (𝑀)); in particular it

is Zariski-closed in Spec(𝑅).
𝑛
Proof. Suppose 𝑀 = 𝑅𝑥1 + ⋯ + 𝑅𝑥𝑛 . Set 𝐼𝑖 ∶= ann𝑅 (𝑥𝑖 ) and observe that ⋂𝑖=1 𝐼𝑖 =
𝑛
ann𝑅 (𝑀). Then 𝔭 ∈ Supp(𝑀) if and only if 𝔭 ⊃ 𝐼𝑖 for some 𝑖, that is, 𝔭 ∈ ⋃𝑖=1 𝑉(𝐼𝑖 ).
To conclude, note that 𝑉(𝐼) ∪ 𝑉(𝐽) = 𝑉(𝐼𝐽) = 𝑉(𝐼 ∩ 𝐽) for any ideals 𝐼, 𝐽 ⊂ 𝑅 (an easy
exercise).

Finite generation is needed in the result above. Consider the ℤ-module 𝑀 ∶=

⨁𝑎≥1 ℤ/𝑝𝑎 ℤ. Its support is {𝑝} but ann(𝑀) = {0}.

Proposition 2.1.3. For an exact sequence 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 we have Supp(𝑀) =

Supp(𝑀′ ) ∪ Supp(𝑀″ ). For arbitrary direct sums we have Supp(⨁𝑖 𝑀𝑖 ) = ⋃𝑖 Supp(𝑀𝑖 ).

Proof. Again, we use the exactness of localization for the first assertion. For 𝔭 ∈ Spec(𝑅)
we have an exact 0 → 𝑀𝔭′ → 𝑀𝔭 → 𝑀𝔭″ → 0, hence 𝑀𝔭 ≠ 0 if and only if 𝔭 ∈
Supp(𝑀′ ) ∪ Supp(𝑀″ ). The second assertion is obvious.
⋅ 16 ⋅ Primary decompositions

From an 𝑅-module 𝑀, one can build a “field of modules” over Spec(𝑅) by assigning
to each 𝔭 the 𝑅𝔭 -module 𝑀𝔭 , and Supp(𝑀) is precisely the subset of 𝑀𝔭 over which the
field is non-vanishing. This is how the support arises in algebraic geometry. A more
precise description will require the notion of quasi-coherent sheaves on schemes.

Exercise 2.1.4. Let 𝑀, 𝑁 be finitely generated 𝑅-modules. Show that Supp(𝑀 ⊗ 𝑁) =

𝑅
Supp(𝑀) ∩ Supp(𝑁). Hint: since localization commutes with ⊗, it suffices to prove
that 𝑀 ⊗ 𝑁 ≠ {0} when 𝑀, 𝑁 are both nonzero finitely generated modules over a local
𝑅
ring 𝑅. Nakayama’s Lemma implies that 𝑀 ⊗ 𝕜 and 𝕜 ⊗ 𝑁 are both nonzero where 𝕜
𝑅 𝑅
is the residue field of 𝑅. Now

(𝑀 ⊗ 𝕜) ⊗ (𝕜 ⊗ 𝑁) ≃ 𝑀 ⊗ (𝕜 ⊗ 𝕜) ⊗ 𝑁 ≃ (𝑀 ⊗ 𝑁) ⊗ 𝕜
𝑅 𝕜 𝑅 𝑅 𝕜 𝑅 𝑅 𝑅

is nonzero.

2.2 Associated primes

Throughout this section, 𝑅 will be a Noetherian ring.

Definition 2.2.1. A prime ideal 𝔭 is said to be an associated prime of an 𝑅-module 𝑀

if ann𝑅 (𝑥) = 𝔭 for some 𝑥 ∈ 𝑀; equivalently, 𝑅/𝔭 embeds into 𝑀. Denote the set of
associated primes of 𝑀 by Ass(𝑀).

Example 2.2.2. For the ℤ-module 𝑀 ∶= ℤ/𝑛ℤ with 𝑛 ∈ ℤ>1 , one easily checks that
Ass(𝑀) is the set of prime factors of 𝑛.

Lemma 2.2.3. Consider the set 𝒮 ∶= {ann𝑅 (𝑥) ∶ 𝑥 ∈ 𝑀, 𝑥 ≠ 0} of ideals, partially ordered
by inclusion. Every maximal element in 𝒮 is prime.

Proof. Let 𝔭 = ann𝑅 (𝑥) be a maximal element of 𝒮 and suppose 𝑎𝑏 ∈ 𝔭. If 𝑏 ∉ 𝔭, then

𝑏𝑥 ≠ 0, 𝑎𝑏𝑥 = 0, 𝑅 ≠ ann𝑅 (𝑏𝑥) ⊃ ann𝑅 (𝑥) = 𝔭.

Hence 𝑎 ∈ ann𝑅 (𝑏𝑥) = 𝔭 by the maximality of 𝔭 in 𝒮.

Definition 2.2.4. Call 𝑟 ∈ 𝑅 a zero divisor on 𝑀 if 𝑟𝑥 = 0 for some 𝑥 ∈ 𝑀 ∖ {0}.

Theorem 2.2.5. Let 𝑀 be an 𝑅-module.

(i) We have 𝑀 = {0} if and only if Ass(𝑀) = ∅.

(ii) The union of all 𝔭 ∈ Ass(𝑀) equals the set of zero divisors on 𝑀.

(iii) For any multiplicative subset 𝑆 ⊂ 𝑅, we have

Ass(𝑀[𝑆−1 ]) = {𝔭[𝑆−1 ] ∶ 𝔭 ∈ Ass(𝑀), 𝔭 ∩ 𝑆 = ∅} .

(iv) If 0 → 𝑀′ → 𝑀 → 𝑀″ is exact, then Ass(𝑀′ ) ⊂ Ass(𝑀) ⊂ Ass(𝑀′ ) ∪ Ass(𝑀″ ).

 §2.2 Associated primes ⋅ 17 ⋅

Proof. (i) Clearly Ass({0}) = ∅. If 𝑀 ≠ 0, the set 𝒮 in Lemma 2.2.3 is then nonempty,
hence contains a maximal element 𝔭 because 𝑅 is Noetherian; this yields 𝔭 ∈ Ass(𝑀).
(ii) Elements of any 𝔭 ∈ Ass(𝑀) are all zero divisors by the very definition of
associated primes. Conversely, if 𝑟 ∈ ann𝑅 (𝑥) for some 𝑥 ∈ 𝑀 ∖ {0}, there must exist
some maximal element 𝔭 of 𝒮 with 𝔭 ⊃ ann𝑅 (𝑥) as 𝑅 is Noetherian; so 𝔭 is the required
associated prime containing 𝑟.
(iii) If 𝔭 ∈ Spec(𝑅), 𝔭 ∩ 𝑆 = ∅ and there is some 𝑅/𝔭 ↪ 𝑀, then

{0} ≠ 𝑅[𝑆−1 ]/𝔭[𝑆−1 ] ≃ (𝑅/𝔭)[𝑆−1 ] ↪ 𝑀[𝑆−1 ]

by the exactness of localizations, hence 𝔭[𝑆−1 ] ∈ Ass(𝑀[𝑆−1 ]). Conversely, every ele-
ment of Ass(𝑀[𝑆−1 ]) has the form 𝔭[𝑆−1 ] for some 𝔭 ∈ Spec(𝑅) disjoint from 𝑆. Also
recall that 𝔭 equals the preimage of 𝔭[𝑆−1 ] under 𝑅 → 𝑅[𝑆−1 ]. There exist 𝑥 ∈ 𝑀 and
𝑠 ∈ 𝑆 such that 𝔭[𝑆−1 ] = ann𝑅[𝑆−1 ] (𝑥/𝑠) = ann𝑅[𝑆−1 ] (𝑥). Ideals in a Noetherian ring
being finitely generated, we infer that ∃𝑡 ∈ 𝑆 with 𝔭 ⊂ ann𝑅 (𝑡𝑥). It remains to show
𝔭 = ann𝑅 (𝑡𝑥). If 𝑟𝑡𝑥 = 0 for some 𝑟 ∈ 𝑅, then 𝑟 maps into 𝑟/1 ∈ 𝔭[𝑆−1 ] = ann𝑅[𝑆−1 ] (𝑥);
thus 𝑟 ∈ 𝔭.
(iv) It suffices to treat the second ⊂. Suppose that 𝔭 ∈ Ass(𝑀) and 𝑅/𝔭 ≃ 𝑁 for
some submodule 𝑁 ⊂ 𝑀. Identify 𝑀′ with ker(𝑀 → 𝑀″ ). If 𝑀′ ∩ 𝑁 = {0} then
𝑅/𝔭 ≃ 𝑁 ↪ 𝑀″ , so 𝔭 ∈ Ass(𝑀″ ). If there exists 𝑥 ∈ 𝑀′ ∩ 𝑁 ⊂ 𝑁 with 𝑥 ≠ 0, then we
have ann𝑅 (𝑥) = 𝔭 since 𝑁 ≃ 𝑅/𝔭 and 𝔭 is prime; in this case 𝔭 ∈ Ass(𝑀′ ).

We remark that 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 being exact does not imply Ass(𝑀) =

𝑝
Ass(𝑀′ ) ∪ Ass(𝑀″ ). To see this, consider 𝑅 = ℤ and 0 → ℤ ℤ → ℤ/𝑝ℤ → 0 for
some prime number 𝑝.

Exercise 2.2.6. Show that Ass(𝑀1 ⊕ 𝑀2 ) = Ass(𝑀1 ) ∪ Ass(𝑀2 ).

Theorem 2.2.7. For every 𝑅-module 𝑀 we have Supp(𝑀) = ⋃𝔭∈Ass(𝑀) 𝑉(𝔭), in particu-
lar Ass(𝑀) ⊂ Supp(𝑀). Furthermore, every minimal element of Supp(𝑀) with respect to
inclusion is actually a minimal element of Ass(𝑀).

Proof. For any prime 𝔮 we have 𝑀𝔮 ≠ 0 ⟺ Ass(𝑀𝔮 ) ≠ ∅, and the latter condition
holds precisely when there exists 𝔭 ∈ Ass(𝑀) with 𝔭 ∩ (𝑅 ∖ 𝔮) = ∅, i.e. 𝔮 ⊃ 𝔭. This
proves the first assertion. The second assertion is a direct consequence.

Observe that if 𝔭 ∈ Supp(𝑀), then 𝔮 ⊃ 𝔭 ⟹ 𝔮 ∈ Supp(𝑀): the reason is that

𝑀𝔭 = (𝑀𝔮 )𝔭𝑅𝔮 (2–1)

thus the occurrence of non-minimal elements in Supp(𝑀) is unsurprising. In contrast,

the non-minimal elements in Ass(𝑀) are somehow mysterious. These non-minimal
associated primes are called embedded primes.

Exercise 2.2.8. Prove the formula (2–1) of “localization in stages”.

Hereafter we impose finite generation on 𝑀. This implies 𝑀 is Noetherian.

⋅ 18 ⋅ Primary decompositions

Theorem 2.2.9. Let 𝑀 be a finitely generated 𝑅-module. There exists a chain 𝑀 = 𝑀𝑛 ⊃

𝑀𝑛−1 ⊃ ⋯ ⊃ 𝑀0 = {0} of submodules such that for every 0 < 𝑖 ≤ 𝑛, the subquotient
𝑀𝑖 /𝑀𝑖−1 is isomorphic to 𝑅/𝔭𝑖 for some prime ideal 𝔭𝑖 .
Furthermore we have Ass(𝑀) ⊂ {𝔭1 , … , 𝔭𝑛 }; in particular Ass(𝑀) is a finite set.

Proof. Assume 𝑀 ⊋ 𝑀0 ∶= {0}. There exists 𝔭1 ∈ Ass(𝑀) together with a submodule

𝑀1 ⊂ 𝑀 isomorphic to 𝑅/𝔭1 . Furthermore Ass(𝑀1 ) = Ass(𝑅/𝔭1 ) = {𝔭1 } as easily
seen. Hence the Theorem 2.2.5 entails Ass(𝑀) ⊂ {𝔭1 } ∪ Ass(𝑀/𝑀1 ).
If 𝑀1 = 𝑀 we are done. Otherwise we start over with 𝑀/𝑀1 , finding 𝑀1 ⊂ 𝑀2 ⊂ 𝑀
with 𝑀2 /𝑀1 ≃ 𝑅/𝔭2 where 𝔭2 ∈ Ass𝑅 (𝑀/𝑀1 ), and so forth. This procedure termi-
nates in finite steps since 𝑀 is Noetherian.

2.3 Primary and coprimary modules

The classical framework of primary decompositions concerns ideals, but it is advanta-
geous to allow modules here. As before, the ring 𝑅 is Noetherian.

Definition 2.3.1. An 𝑅-module 𝑀 is called coprimary if Ass(𝑀) is a singleton. A sub-

module 𝑁 ⊊ 𝑀 is called a 𝔭-primary submodule if 𝑀/𝑁 is coprimary with associated
prime 𝔭 ∈ Spec(𝑅).

Proposition 2.3.2. The following are equivalent for a nonzero 𝑅-module 𝑀.

(i) 𝑀 is coprimary;

(ii) for every zero divisor 𝑟 ∈ 𝑅 for 𝑀 and every 𝑥 ∈ 𝑀, there exists 𝑛 ≥ 1 such that 𝑟𝑛 𝑥 = 0.

The condition (ii) is usually called the local-nilpotency of 𝑟 on 𝑀. When 𝑀 = 𝑅/𝐼

for some ideal 𝐼, it translates into: all zero divisors of the ring 𝑅/𝐼 are nilpotent.
Proof. (i) ⟹ (ii): Suppose Ass(𝑀) = {𝔭} and 𝑥 ∈ 𝑀 ∖ {0}. From ∅ ≠ Ass(𝑅𝑥) ⊂
Ass(𝑀) we infer that Ass(𝑅𝑥) = {𝔭}, thus by Theorem 2.2.7 we see 𝑉(𝔭) = Supp(𝑅𝑥) =
𝑉(ann(𝑅𝑥)). Hence 𝔭 = √ann(𝑅𝑥). This implies (ii) by the definition of √ .
(ii) ⟹ (i): It is routine to check that

𝔭 ∶= {𝑟 ∈ 𝑅 ∶ ∀𝑥 ∈ 𝑀 ∃𝑛 ≥ 1, 𝑟𝑛 𝑥 = 0}

is an ideal of 𝑅. For every 𝔮 ∈ Ass(𝑀) there exists 𝑥 ∈ 𝑀 with ann𝑅 (𝑥) = 𝔮. Every
𝑟 ∈ 𝔭 has some power falling in 𝔮, thus 𝔭 ⊂ 𝔮. Conversely, (ii) and Theorem 2.2.5 imply
𝔮 = ann𝑅 (𝑥) ⊂ 𝔭. From 𝔮 = 𝔭 we conclude 𝑀 is coprimary with the unique associated
prime 𝔭.

Exercise 2.3.3 (Classical definition of primary ideals). Let 𝐼 be a proper ideal of 𝑅.

Show that 𝑅/𝐼 is coprimary if and only if the following holds:

∀𝑎, 𝑏 ∈ 𝑅, (𝑎𝑏 ∈ 𝐼) ∧ (𝑎 ∉ 𝐼) ⟹ ∃𝑛 ≥ 1, 𝑏𝑛 ∈ 𝐼.

In this case we also say 𝐼 is a primary ideal of 𝑅. Show that {√𝐼} = Ass(𝑅/𝐼) if 𝐼 is a
primary ideal. Hint: apply Proposition 2.3.2.
 §2.4 Primary decomposition: the main theorem ⋅ 19 ⋅

Exercise 2.3.4. Let 𝔪 be a maximal ideal of 𝑅. Show that every ideal 𝐼 ⊊ 𝑅 containing
some power of 𝔪 is primary, and Ass(𝑅/𝐼) = {𝔪}. Hint: show that 𝔪 is the only prime
ideal containing 𝐼 = ann𝑅 (𝑅/𝐼).
Lemma 2.3.5. Let 𝔭 ∈ Spec(𝑅) and 𝑁1 , 𝑁2 ⊂ 𝑀 are 𝔭-primary submodules. Then 𝑁1 ∩ 𝑁2
is a 𝔭-primary submodule of 𝑀.
Proof. We have 𝑀/𝑁1 ∩ 𝑁2 ↪ 𝑀/𝑁1 ⊕ 𝑀/𝑁2 . Since 𝑁1 ∩ 𝑁2 ≠ 𝑀, we have

∅ ≠ Ass(𝑀/𝑁1 ∩ 𝑁2 ) ⊂ Ass(𝑀/𝑁1 ) ∪ Ass(𝑀/𝑁2 ) = {𝔭}

by Theorem 2.2.5.

2.4 Primary decomposition: the main theorem

We still assume 𝑅 Noetherian and fix a finitely generated 𝑅-module 𝑀.
Theorem 2.4.1 (Lasker–Noether). Let 𝑁 ⊊ 𝑀 be an 𝑅-submodule. Then we can express 𝑁
as
𝑁 = 𝑀1 ∩ ⋯ ∩ 𝑀 𝑛 ,
for some 𝑛 ≥ 1 and primary 𝑅-submodules 𝑀𝑖 , say with Ass(𝑀/𝑀𝑖 ) = {𝔭𝑖 } for 𝑖 = 1, … , 𝑛.
Such a decomposition is called a primary decomposition of 𝑁. We say it is irredundant if
none of the 𝑀𝑖 can be dropped, and minimal if there is no such decomposition with fewer terms.
(i) We have Ass(𝑀/𝑁) ⊂ {𝔭1 , … , 𝔭𝑛 }, equality holds when the decomposition is irredun-
dant.
(ii) If the decomposition is minimal, then for every 𝔭 ∈ Ass(𝑀/𝑁) there exists a unique
1 ≤ 𝑖 ≤ 𝑛 with 𝔭 = 𝔭𝑖 ; consequently 𝑛 = |Ass(𝑀/𝑁)|.
(iii) Consider a primary decomposition of 𝑁. Let 𝑆 ⊂ 𝑅 be any multiplicative subset, and
assume without loss of generality that 𝔭1 , … , 𝔭𝑚 are the primes among {𝔭1 , … , 𝔭𝑛 } which
are disjoint from 𝑆, then 𝑚 ≥ 1 ⟺ 𝑁[𝑆−1 ] ⊊ 𝑀[𝑆−1 ] and

𝑁[𝑆−1 ] = 𝑀1 [𝑆−1 ] ∩ ⋯ ∩ 𝑀𝑚 [𝑆−1 ]

is a primary decomposition of the 𝑅[𝑆−1 ]-submodule 𝑁[𝑆−1 ] ⊊ 𝑀[𝑆−1 ]; this decompo-

sition of 𝑁[𝑆−1 ] is minimal if the one for 𝑁 is.
Note that the “irredundant” condition in [11, (8.D)] corresponds to minimality
here. The rings whose ideals all have primary decompositions are called Laskerian rings,
thus part (i) of the Theorem says Noetherian implies Laskerian, but there exist non-
Noetherian examples.
Proof. Establish the existence of primary decompositions first. Replacing 𝑀 by 𝑀/𝑁,
we may assume 𝑁 = {0} from the outset. We claim that
⎧ 𝑄(𝔭) is 𝔭-primary,
{
∀𝔭 ∈ Ass(𝑀), ∃𝑄(𝔭) ⊂ 𝑀, ⎨ (2–2)
{ Ass(𝑄(𝔭)) = Ass(𝑀) ∖ {𝔭}.
⎩
⋅ 20 ⋅ Primary decompositions

Granting this, 𝑄 ∶= ⋂𝔭∈Ass(𝑀) 𝑄(𝔭) yields the required decomposition since Ass(𝑀) is
finite and Ass(𝑄) = ∅.
Establish (2–2) as follows. Put 𝛹 ∶= {𝔭}. By Zorn’s Lemma we get a maximal
element 𝑄(𝔮) from the set

∅ ≠ {𝑄 ⊂ 𝑀 ∶ submodule, Ass(𝑄) ⊂ Ass(𝑀) ∖ 𝛹}

which is partially ordered by inclusion (details omitted, and you may also use the
Noetherian property of 𝑀). Since

Ass(𝑀) ⊂ Ass(𝑀/𝑄(𝔭)) ∪ Ass(𝑄(𝔭)),

it suffices to show Ass(𝑀/𝑄(𝔭)) ⊂ 𝛹. Let 𝔮 ∈ Ass(𝑀/𝑄(𝔭)) so that there exists

𝑀 ⊃ 𝑄′ ⊃ 𝑄(𝔭) with 𝑄′ /𝑄(𝔭) ≃ 𝑅/𝔮. Since Ass(𝑄′ ) ⊂ Ass(𝑄(𝔭)) ∪ {𝔮}, maximal-
ity forces 𝔮 ∈ 𝛹 (otherwise Ass(𝑄′ ) ⊂ Ass(𝑀) ∖ 𝛹), whence (2–2). Now we turn to
the properties (i) — (iii).
(i) The obvious embedding
𝑛
𝑀/𝑁 ↪ ⨁ 𝑀/𝑀𝑖
𝑖=1

𝑛
together with Theorem 2.2.5 yield Ass(𝑀/𝑁) ⊂ ⋃𝑖=1 Ass(𝑀/𝑀𝑖 ) = {𝔭1 , … , 𝔭𝑛 }.
Now assume the given primary decomposition is irredundant, we have

𝑀2 ∩ ⋯ ∩ 𝑀 𝑛 𝑀2 ∩ ⋯ ∩ 𝑀 𝑛
{0} ≠ =
𝑁 𝑀1 ∩ (𝑀2 ∩ ⋯ ∩ 𝑀𝑛 )
𝑀1 + 𝑀 2 ∩ ⋯ ∩ 𝑀 𝑛
≃ ↪ 𝑀/𝑀1 .
𝑀1

Thus Ass(𝑀/𝑁) contains Ass((𝑀2 ∩ ⋯ ∩ 𝑀𝑛 )/𝑁) = {𝔭1 }. Same for 𝔭2 , … , 𝔭𝑛 .

(ii) Suppose 𝑁 = 𝑀1 ∩ ⋯ ∩ 𝑀𝑛 is an irredundant primary decomposition, so that
Ass(𝑀/𝑁) = {𝔭1 , … , 𝔭𝑛 }. If 𝔭𝑖 = 𝔭𝑗 for some 1 ≤ 𝑖 ≠ 𝑗 ≤ 𝑛, Lemma 2.3.5 will imply that
𝑀𝑖 ∩ 𝑀𝑗 is primary, leading to a shorter primary decomposition. This is impossible
when the primary decomposition is minimal.
(iii) Suppose 𝑆 ∩ 𝔭𝑖 = ∅ (equivalently, 𝑖 ≤ 𝑚). By Theorem 2.2.5 and the exactness
of localization, 𝑀𝑖 [𝑆−1 ] ⊂ 𝑀[𝑆−1 ] will be 𝔭𝑖 [𝑆−1 ]-primary. On the other hand 𝑆 ∩ 𝔭𝑖 ≠
∅ implies Ass((𝑀/𝑀𝑖 )[𝑆−1 ]) = ∅ by Theorem 2.2.5, thus 𝑀[𝑆−1 ]/𝑀𝑖 [𝑆−1 ] = 0. Since
localization respects intersections, we obtain
𝑚
𝑁[𝑆−1 ] = ⋂ 𝑀𝑖 [𝑆−1 ].
𝑖=1

In particular 𝑁[𝑆−1 ] is proper if and only if 𝑚 ≥ 1.

It remains to show Ass(𝑀[𝑆−1 ]/𝑁[𝑆−1 ]) has 𝑚 elements if the original primary de-
composition is minimal. Indeed, that set is just {𝔭[𝑆−1 ] ∶ 𝔭 ∈ Ass(𝑀/𝑁), 𝔭 ∩ 𝑆 = ∅}
by Theorem 2.2.5, which equals {𝔭1 [𝑆−1 ], … , 𝔭𝑚 [𝑆−1 ]} (distinct) by (ii).
 §2.5 Examples and remarks ⋅ 21 ⋅

A natural question arises: to what extent are minimal primary decompositions

unique? For those 𝑀𝑖 whose associated primes are minimal in Ass(𝑀/𝑁), the answer
turns out to be positive.
Corollary 2.4.2. Let 𝑁 = 𝑀1 ∩ ⋯ ∩ 𝑀𝑛 be a minimal primary decomposition of 𝑁 ⊊ 𝑀.
Suppose that 𝑀1 is 𝔭-primary where 𝔭 ∶= 𝔭1 is a minimal element in Ass(𝑀/𝑁), then 𝑀1
equals the preimage of 𝑁𝔭 under 𝑀 → 𝑀𝔭 . Call it the 𝔭-primary component of 𝑁.
Proof. Recall the proof of Theorem 2.4.1, especially the part (iii); here we localize with
respect to 𝑆 ∶= 𝑅 ∖ 𝔭. The minimality assumption entails

𝑁𝔭 = 𝑀1,𝔭 ⊂ 𝑀𝔭 .

It remains to show that the preimage of 𝑀1,𝔭 under 𝑀 → 𝑀𝔭 equals 𝑀1 , in other words
the injectivity of the natural map 𝑀/𝑀1 → (𝑀/𝑀1 )𝔭 = 𝑀𝔭 /𝑀1,𝔭 . Indeed, 𝑥 ̄ ∈ 𝑀/𝑀1
maps to 0 if and only if there exists 𝑠 ∉ 𝔭 with 𝑠𝑥 ̄ = 0, but Theorem 2.2.5 implies that
the zero divisors of 𝑀/𝑀1 must lie in 𝔭.
It follows that the non-uniqueness of minimal primary decompositions can only
arise from embedded primes in Ass(𝑀/𝑁).

2.5 Examples and remarks

Primary decompositions are most often applied in the case 𝑀 = 𝑅 and 𝑁 = 𝐼 is a
proper ideal. The goal is to express 𝐼 as an intersection of primary ideals. To begin
with, let us take 𝑅 = ℤ. Observations:
⋄ The primary ideals of 𝑅 take the form (𝑝)𝑛 , where 𝑛 ≥ 1 and 𝑝 is a prime number
or zero. This may be deduced from Exercise 2.3.3 or directly from definitions.

⋄ The irredundant primary decompositions of ℤ/𝑛ℤ, for 𝑛 > 1, corresponds to the

factorization of 𝑛 into prime-powers. There are no embedded primes in this case;
the irredundant primary decomposition is unique and automatically minimal.
Exercise 2.5.1. Justify the foregoing assertions.
Therefore one can regard primary decompositions as a generalization of factoriza-
tion of integers, now performed on the level of ideals. The most important case is
𝑅 = 𝕜[𝑋1 , … , 𝑋𝑛 ] (fix some field 𝕜), as it is naturally connected to classical problems
in algebraic geometry. Let us consider a simple yet non-trivial example from [8, §3].
Example 2.5.2. Take 𝑅 = 𝕜[𝑋, 𝑌] and 𝐼 = (𝑋 2 , 𝑋𝑌). The reader is invited to check that

𝐼 = (𝑋) ∩ (𝑋 2 , 𝑋𝑌, 𝑌 2 ) = (𝑋) ∩ (𝑋 2 , 𝑌).

Claim: this gives two minimal primary decompositions of 𝐼. The ideal (𝑋) is prime,
hence primary. In fact, (𝑋 2 , 𝑋𝑌, 𝑌 2 ) = (𝑋, 𝑌)2 and (𝑋 2 , 𝑌) are both primary ideals
associated to the maximal ideal (𝑋, 𝑌). This follows either by direct arguments or by
Exercise 2.3.4, noting that (𝑋, 𝑌)2 = (𝑋 2 , 𝑋𝑌, 𝑌 2 ) is contained in (𝑋 2 , 𝑌). The embed-
ded prime (𝑋, 𝑌) is seen to be responsible non-uniqueness of primary decompositions.
⋅ 22 ⋅ Primary decompositions

Emanuel Lasker (1868–1941) first obtained the primary decompo-

sition for finitely generated 𝕜-algebras and the algebras of conver-
gent power series in 1905. His method involves techniques from
elimination theory. His result is then generalized and rewritten by
Emmy Noether in 1921, in which the ascending chain condition
plays a pivotal role. Lasker is best known for being the World
Chess Champion from 1894 to 1921. (Picture taken from Wikime-
dia Commons)

To see the geometry behind, recall that 𝑉(𝔞) ∪ 𝑉(𝔟) = 𝑉(𝔞𝔟) = 𝑉(𝔞 ∩ 𝔟) for any
ideals 𝔞, 𝔟, thus expressing 𝐼 as an intersection means breaking the corresponding ge-
ometric object into a union of simpler pieces. Also recall that for an ideal 𝐼 ⊂ 𝕜[𝑋, 𝑌],
the points in ⋂𝑓 ∈𝐼 {𝑓 = 0} are in bijection with the maximal ideals lying over 𝐼, at least
for 𝕜 algebraically closed (Nullstellensatz). Thus we may interpret these primary de-
compositions as equalities among “geometric objects” embedded in 𝕜2 :
⎧
{{𝑋 = 0} ∪ {𝑋 2 = 𝑋𝑌 = 𝑌 2 = 0}
{𝑋 2 = 0, 𝑋𝑌 = 0} = ⎨ 2
{
⎩{𝑋 = 0} ∪ {𝑋 = 𝑌 = 0} .
1. The geometric object defined by 𝑋 = 0 inside 𝕜2 is certainly the 𝑌-axis: the reg-
ular functions living on this space form the 𝕜-algebra 𝕜[𝑋, 𝑌]/(𝑋) = 𝕜[𝑌].
2. The geometric object defined by 𝑋 2 = 𝑋𝑌 = 𝑌 2 = 0 looks “physically” like
the origin (0, 0), but the 𝕜-algebra of “regular functions” (in an extended sense)
living on it equals 𝕜[𝑋, 𝑌]/(𝑋 2 , 𝑋𝑌, 𝑌 2 ): by restricting a polynomial function
𝜕𝑓
𝑓 (𝑋, 𝑌) to this “thickened point”, we see not only 𝑓 (0, 0) but also 𝜕𝑥 (0, 0) and
𝜕𝑓
𝜕𝑦
(0, 0). In other words, we shall view 𝑋 2 = 𝑋𝑌 = 𝑌 2 = 0 as the first-order
infinitesimal neighborhood of (0, 0) ∈ 𝕜2 .
3. In a similar vein, 𝑋 2 = 𝑌 = 0 physically defines (0, 0), but by restricting 𝑓 to that
𝜕𝑓
thickened point, we retrieve 𝑓 (0, 0) as well as 𝜕𝑥 (0, 0). Therefore we obtain the
first-order infinitesimal neighborhood of 0 inside the 𝑋-axis.
Both decomposition says that we obtain the 𝑌-axis together with first-order in-
finitesimal information at the origin (0, 0). This is also a nice illustration of the use
 §2.5 Examples and remarks ⋅ 23 ⋅

of nilpotent elements in scheme theory.

Example 2.5.3 (Symbolic powers). Let 𝔭 be a prime ideal in a Noetherian ring 𝑅 and
fix 𝑛 ≥ 1. Observe that 𝔭 is the unique minimal element in Supp(𝑅/𝔭𝑛 ) = 𝑉(𝔭𝑛 ).
By Theorem 2.2.7 𝔭 ∈ Ass(𝑅/𝔭𝑛 ), so it makes sense to denote by 𝔭(𝑛) the 𝔭-primary
component (Corollary 2.4.2) of 𝔭𝑛 , called the 𝑛-th symbolic power of 𝔭. In general 𝔭(𝑛) ⊋
𝔭𝑛 . For a nice geometric interpretation of symbolic powers due to Nagata and Zariski,
we refer to [8, §3.9].

Getting primary decomposition of ideals in polynomial algebras is a non-trivial

task. Thanks to the pioneers in computational commutative algebra, this can now
achieved on your own computer, eg. by the open-source SageMath system.
⋅ 24 ⋅ Primary decompositions
 Lecture 3
Integral dependence,
Nullstellensatz and flatness
The materials below largely come from [8, §4], [1, V.4] and [11, §§5–6]. In what follows,
any ring 𝑅 and the 𝑅-algebras are assumed to be commutative. For elements 𝑎, 𝑏, … in
an 𝑅-algebra, denote by 𝑅[𝑎, 𝑏, …] the 𝑅-subalgebra generated by them.

3.1 Integral extensions

Consider an 𝑅-algebra 𝐴. Recall that to give an 𝑅-algebra 𝐴 is the same as to give a ring
homomorphism 𝜑 ∶ 𝑅 → 𝐴. We shall switch freely between algebra and homomor-
phisms, omitting 𝜑 if necessary. In many concrete circumstances 𝑅 is simply a subring
of 𝐴.
Definition 3.1.1. An element 𝑥 ∈ 𝐴 is said to be integral over 𝑅 if there exists a monic
polynomial 𝑃(𝑋) = 𝑋 𝑛 + 𝑎𝑛−1 𝑋 𝑛−1 + ⋯ + 𝑎0 ∈ 𝑅[𝑋] (with 𝑛 ≥ 1) such that 𝑃(𝑥) = 0.
If every 𝑥 ∈ 𝐴 is integral over 𝑅, we say 𝐴 is integral over 𝑅.
Note that elements from 𝑅 are trivially integral: take 𝑃(𝑋) with 𝑛 = 1. In the case
𝐴 = ℂ and 𝑅 = ℤ, we recover the notion of algebraic integers.
When 𝑅 is a field, we usually say algebraic instead of integral. The monic assumption
on 𝑃 is crucial when 𝑅 is not a field, as illustrated by the following proof.
Proposition 3.1.2. An element 𝑥 ∈ 𝐴 is integral over 𝑅 if and only if there exists an 𝑅-
submodule 𝑀 ⊂ 𝐴 such that
⋄ 𝑀 is a finitely generated 𝑅-module;
⋄ 𝑥𝑀 ⊂ 𝑀, thus 𝑀 is an 𝑅[𝑥]-module;
⋄ 𝑀 is a faithful 𝑅[𝑥]-module i.e. ann𝑅[𝑥] (𝑀) = {0}, and
If 𝑥 is integral, 𝑀 ∶= 𝑅[𝑥] satisfies the conditions listed above.
Proof. This is a familiar application of Cayley-Hamilton theorem, which we recall be-
low. If 𝑥 is integral, say 𝑥𝑛 + 𝑎𝑛−1 𝑥𝑛−1 + ⋯ + 𝑎0 = 0, a straightforward induction shows
that
𝑛−1
𝑅[𝑥] = ∑ 𝑅𝑥𝑖 .
𝑖=0
⋅ 26 ⋅ Integral dependence, Nullstellensatz and flatness

In particular we may take 𝑀 ∶= 𝑅[𝑥] to be the required submodule, which is faithful

as 1 ∈ 𝑅[𝑥]. Conversely, given a submodule 𝑀 as above, with generators 𝑥1 , … , 𝑥𝑚 ,
we make 𝑀 into an 𝑅[𝑋]-module by letting the variable 𝑋 act as multiplication by 𝑥.
𝑚
Writing 𝑥 ⋅ 𝑥𝑖 = ∑𝑗=1 𝑎𝑖𝑗 𝑥𝑗 , there is the matrix equation

⎛
⎜ 𝑥1 ⎞
⎟
⎜
⎜ ⎟
⎟
⎜
(𝑋 ⋅ 1𝑚×𝑚 − 𝐴) ⎜ ⎟
⎜
⎜
⋮ ⎟ = 0,
⎟ 𝐴 = (𝑎𝑖𝑗 )1≤𝑖,𝑗≤𝑚 ∈ Mat𝑚×𝑚 (𝑅)
⎜ ⎟ ⎟
⎝𝑥𝑚 ⎠
over 𝑅[𝑋]. Now multiply both sides by the cofactor matrix (𝑋 ⋅ 1𝑚×𝑚 − 𝐴)∨ , we get
𝑃(𝑋)𝑥𝑖 = 0 for all 𝑖, where 𝑃 ∈ 𝑅[𝑋] is the characteristic polynomial of 𝐴, i.e. 𝑃(𝑥)𝑀 =
{0}. Since 𝑀 is faithful as an 𝑅[𝑥]-module, we get 𝑃(𝑥) = 0.
Corollary 3.1.3. The integral elements in an 𝑅-subalgebra 𝐴 form a subalgebra. In particular,
𝐴 is integral over 𝑅 if and only if it has a set of integral generators.
Proof. Let 𝑎, 𝑏 ∈ 𝐴 be integral elements. One readily checks that
⋄ 𝑅[𝑎, 𝑏] is finitely generated as an 𝑅-module (say by certain monomials 𝑎𝑖 𝑏𝑗 );

⋄ 𝑅[𝑎, 𝑏] is faithful (as an 𝑅[𝑎, 𝑏]-module) because it contains 1.

Thus 𝑅[𝑎, 𝑏] witnesses the integrality of 𝑎+𝑏 and 𝑎𝑏, since they both stabilize 𝑅[𝑎, 𝑏].
Proposition 3.1.4. Consider ring homomorphisms 𝑅 → 𝐴 → 𝐵 such that 𝐴 is integral over
𝑅. If 𝑦 ∈ 𝐵 is integral over 𝐴, then it is integral over 𝑅.
Proof. Assume 𝑦𝑛 + 𝑎𝑛−1 𝑦𝑛−1 + ⋯ + 𝑎0 = 0 with 𝑎0 , … , 𝑎𝑛−1 ∈ 𝐴 integral over 𝑅. The
𝑅[𝑎0 , … , 𝑎𝑛−1 ]-module 𝑅[𝑎0 , … , 𝑎𝑛−1 ][𝑦] is also finitely generated over 𝑅, faithful and
preserved by 𝑦, hence it witnesses the integrality of 𝑦 over 𝑅.
This subalgebra of integral elements in 𝐴 is called the integral closure of 𝑅 in 𝐴. If
the integral closure equals the image of 𝑅 in 𝐴, we say 𝑅 is integrally closed in 𝐴. The
integral closure is automatically integrally closed by virtue of Proposition 3.1.4.
Definition 3.1.5. Let 𝑅 be an integral domain and denote by 𝐾 its field of fractions. The
integral closure of 𝑅 in 𝐾 is called the normalization of 𝑅. The domain 𝑅 is said to be
normal if 𝑅 is integrally closed in 𝐾.
The first examples of normal domains come from unique factorization domains
(UFD) that you have seen in undergraduate algebra, including ℤ, ℚ[𝑋, 𝑌], etc.
Proposition 3.1.6. Unique factorization domains are normal.
Proof. Given 𝑥 = 𝑟/𝑠 ∈ 𝐾 with coprime 𝑟, 𝑠 ∈ 𝑅. If there is an integral dependence
relation 𝑥𝑛 + 𝑎𝑛−1 𝑥𝑛−1 + ⋯ + 𝑎0 = 0, we will have 𝑟𝑛 + 𝑎𝑛−1 𝑟𝑛−1 𝑠 + ⋯ + 𝑎0 𝑠𝑛 = 0, hence
𝑠 ∣ 𝑟𝑛 . As 𝑟 is coprime to 𝑠, we see 𝑠 ∈ 𝑅× and 𝑥 ∈ 𝑅.
Let us show that taking integral closure commutes with localizations. Geometri-
cally, this means that taking integral closure is a local operation on Spec(𝑅).
 §3.1 Integral extensions ⋅ 27 ⋅

Lemma 3.1.7. Let 𝐴 be an 𝑅-algebra, 𝑆 be a multiplicative subset of 𝑅. Denote by 𝑅̃ the integral

̃ −1 ].
closure of 𝑅 in 𝐴. Then the integral closure of 𝑅[𝑆−1 ] in 𝐴[𝑆−1 ] equals 𝑅[𝑆
𝑛−1
Proof. Suppose that 𝑥 ∈ 𝐴 satisfies 𝑥𝑛 + ∑𝑖=0 𝑏𝑖 𝑥𝑖 = 0 with 𝑏0 , … , 𝑏𝑛−1 ∈ 𝑅. For all
𝑛 𝑛−1 𝑏𝑖 𝑥 𝑖
𝑠 ∈ 𝑆 we deduce ( 𝑥 ) + ∑
𝑠 𝑛−𝑖 ⋅ ( ) = 0 in 𝐴[𝑆
𝑖=0 𝑠 𝑠
−1 ]. Therefore 𝑅[𝑆̃ −1 ] is in the
integral closure of 𝑅[𝑆−1 ] in 𝐴[𝑆−1 ].
Conversely, suppose that 𝑥/𝑠 ∈ 𝐴[𝑆−1 ] satisfies
𝑥 𝑛 𝑛−1 𝑎 𝑥 𝑖
( ) + ∑ 𝑖 ( ) = 0, 𝑎𝑖 ∈ 𝐴, 𝑠𝑖 ∈ 𝑆.
𝑠 𝑠 𝑠
𝑖=0 𝑖
𝑥𝑠1 ⋯𝑠𝑛−1
Cleaning denominators, we see that 𝑥𝑠1 ⋯ 𝑠𝑛−1 is integral over 𝑅, hence 𝑥 = 𝑠1 ⋯𝑠𝑛 ∈
̃ −1 ].
𝑅[𝑆
In particular, localizations of a normal domain are still normal. We have the follow-
ing converse.
Exercise 3.1.8. Let 𝑅 be a domain. If for every maximal ideal 𝔪 of 𝑅 we have 𝑅𝔪 is
normal, then so is 𝑅 itself. Hint: recall that 𝑅 = ⋂𝔪 𝑅𝔪 inside the fraction field 𝐾.
Exercise 3.1.9. Let 𝑅 ∶= ℂ[𝑋, 𝑌]/(𝑌 2 − 𝑋 3 ). Show that
(i) 𝑅 is a domain, or equivalently, 𝑌 2 − 𝑋 3 is an irreducible polynomial;
(ii) the element 𝑥 ∶= 𝑌/̄ 𝑋̄ ∈ Frac(𝑅) does not lie in 𝑅, where 𝑋,̄ 𝑌̄ denote the images
of 𝑋, 𝑌 ∈ ℂ[𝑋, 𝑌];
(iii) show that 𝑥3 ∈ 𝑅, therefore 𝑅 is not a normal ring.
Remark 3.1.10. The rationale of the previous Exercise comes from singularity. Consider
the cuspidal curve 𝐶 ∶= {(𝑥, 𝑦) ∈ ℂ2 ∶ 𝑦2 = 𝑥3 }. It has an isolated singularity at (0, 0)
since ∇(𝑦2 − 𝑥3 ) = (0, 0) at the origin. On the other hand, it admits a (non-injective)
parametrization from ℂ by
𝑡 ↦ (𝑥 = 𝑡2 , 𝑦 = 𝑡3 ).
Following Milnor [12], a nice way to understand the “cusp” at (0, 0) is to cut 𝐶 by a
3-sphere 𝑆 ∶= {(𝑥, 𝑦) ∈ ℂ2 ∶ |𝑥|2 + |𝑦|2 = 𝜖} enclosing the origin, with 0 < 𝜖 ≪ 1.
Writing the parametrization above as 𝑡 = 𝑟𝑒𝑖𝜃 in polar coordinates, we get the equation
𝑟4 + 𝑟6 = 𝜖 which has a unique positive root 𝜌. Hence 𝑆 ∩ 𝐶 is described by the closed
curve 𝜃 ↦ (𝜌2 𝑒2𝑖𝜃 , 𝜌3 𝑒3𝑖𝜃 ) in 𝑆 ≃ 𝕊3 . We call 𝑆 ∩ 𝐶 the link of singularity at (0, 0); in this
case it is equivalent to the (2, 3)-torus knot, i.e. the trefoil shown below.
⋅ 28 ⋅ Integral dependence, Nullstellensatz and flatness

If we started from a non-singular curve in ℂ2 , the result would then be an unknotted

𝕊1 inside 𝕊3 . In this sense the link provides a measure for the singularity.
Our study of normality has to be paused here. We refer to [13, §§8–9] for a beautiful
discussion on the algebro-geometric content of normality, as well as important conse-
quences such as Zariski’s Main Theorem. Roughly speaking, normality means there is
only one branch through each point of the corresponding variety.

3.2 Nullstellensatz
Our aim is to present a generalization of the celebrated Nullstellensatz, which is one of
the cornerstones of algebraic geometry. We shall also write the nilpotent radical of a
ring 𝑅 as
nil(𝑅) ∶= √0𝑅 .
For an ideal 𝔞 ⊂ 𝑅, we write 𝔞[𝑋] for the ideal of 𝑅[𝑋] formed by polynomials with all
coefficients lying in 𝔞.
Definition 3.2.1. A ring 𝑅 is called a Jacobson ring if every prime ideal 𝔭 satisfies

𝔭= ⋂ 𝔪. (3–1)
𝔪∶maximal ideal ⊃𝔭

Equivalently, we require that the Jacobson radical rad(𝑅/𝔭) = nil(𝑅/𝔭) = {0} for all 𝔭.
Note that ⊃ always holds.
Observations:
⋄ Quotients of Jacobson rings are still Jacobson.
⋄ Fields are trivially Jacobson.
Exercise 3.2.2. Prove that every principal ideal domain (a domain in which every ideal
is generated by one element) with infinitely many maximal ideals is Jacobson.
Theorem 3.2.3 (E. Snapper). For any 𝑅, the polynomial algebra 𝑅[𝑋] satisfies rad(𝑅[𝑋]) =
nil(𝑅[𝑋]).
Proof. To show that nil(𝑅[𝑋]) ⊃ rad(𝑅[𝑋]), let 𝑓 (𝑋) = ∑𝑖 𝑎𝑖 𝑋 𝑖 ∈ rad(𝑅[𝑋]), then
1 + 𝑋𝑓 (𝑋) = 1 + ∑𝑖 𝑎𝑖 𝑋 𝑖+1 ∈ 𝑅[𝑋]× . By looking at the reduction modulo 𝔭 of 𝑓 (𝑋)
for every prime ideal 𝔭 of 𝑅, we see that 𝑎𝑖 ∈ ⋂ 𝔭 = nil(𝑅) for all 𝑖. Hence 𝑓 (𝑋) ∈
nil(𝑅)[𝑋] ⊂ nil(𝑅[𝑋]).
Lemma 3.2.4. Let 𝑅 ⊂ 𝐴 be integral domains such that 𝐴 is a finitely generated 𝑅-algebra. If
rad(𝑅) = {0}, then rad(𝐴) = {0}.
Proof. We may assume that 𝐴 is generated by a single element 𝑎 ∈ 𝐴 over 𝑅. If 𝑎 is
transcendental over 𝐾 ∶= Frac(𝑅), Theorem 3.2.3 above can be applied as nil(𝑅[𝑋]) =
𝑛
{0}. Let us assume that 𝑎 satisfies 𝑓 (𝑎) = 0 for some 𝑓 (𝑋) = ∑𝑖=0 𝑟𝑖 𝑋 𝑖 ∈ 𝑅[𝑋] with
𝑟𝑛 ≠ 0. Let 𝑏 ∈ rad(𝐴) and suppose 𝑏 ≠ 0. Hereafter we embed everything into the
𝐾-algebra Frac(𝐴). Since 𝑎 is algebraic over 𝐾, so is every element from 𝑅[𝑎] or even
𝐾[𝑎] ⊂ Frac(𝐴). Hence 𝑏 is integral over 𝐾 as well. By cleaning denominators, we
 §3.2 Nullstellensatz ⋅ 29 ⋅

𝑚
arrive at 𝑔(𝑏) = 0 for some 𝑔(𝑋) = ∑𝑖=0 𝑠𝑖 𝑋 𝑖 ∈ 𝑅[𝑋] with the smallest possible degree
𝑚. Since 𝐴 is a domain, we have 𝑠0 ≠ 0.
Using rad(𝑅) = {0}, there exists a maximal ideal 𝔪 of 𝑅 such that 𝑟𝑛 𝑠0 ∉ 𝔪. Taking
localization at 𝔪, we get the subring 𝐴′ ∶= 𝐴 ⊗𝑅 𝑅𝔪 ⊂ Frac(𝐴) containing 𝑅𝔪 , and 𝐴′
is also a finitely generated 𝑅𝔪 -module since we inverted 𝑟𝑛 . Nakayama’s Lemma for
𝑅𝔪 -modules implies 𝔪𝐴′ ⊊ 𝐴′ , thus 𝔪𝐴 ⊊ 𝐴. Now choose a maximal ideal 𝔪𝐴 of 𝐴
over 𝔪. We must have 𝔪𝐴 ∩ 𝑅 = 𝔪, which entails 𝑠0 ∉ 𝔪𝐴 . This is a contradiction since
𝑚
𝑠0 = − ∑𝑖=1 𝑠𝑖 𝑏𝑖 ∈ rad(𝐴).
Theorem 3.2.5 (Nullstellensatz). Let 𝐴 be a finitely generated 𝑅-algebra. Assume that 𝑅 is
a Jacobson ring, then the following statements hold.
(i) 𝐴 is a Jacobson ring.

(ii) Let 𝔫 ∈ Spec(𝐴) be maximal, then its image 𝔪 ∈ Spec(𝑅) is maximal as well, and 𝐴/𝔫
is a finite extension of the field 𝑅/𝔪.
Proof. We start with (i). One may assume 𝑅 ⊂ 𝐴 from the outset. Condition (3–1) for
𝐴 amounts to rad(𝐴/𝔭) = 0 for every 𝔭 ∈ Spec(𝐴). Apply the previous Lemma to the
integral domains 𝑅/𝑅 ∩ 𝔭 ⊂ 𝐴/𝔭 to prove (i).
Now turn to (ii). Using (i) and induction, we may assume that 𝐴 = 𝑅[𝑎] for some
𝑎 ∈ 𝐴. By considering the homomorphism 𝑅/𝔪 ↪ 𝐴/𝔫 between Jacobson rings, we
may further reduce to the case 𝔫 = {0} and 𝔪 = {0}, so that 𝑅 is a domain embedded in
the field 𝐴. In particular 𝑎 cannot be transcendental (as 𝑅[𝑋] is not a field) and must
𝑛
satisfy ∑𝑖=0 𝑐𝑖 𝑎𝑖 = 0 for some 𝑐0 , … , 𝑐𝑛 ∈ 𝑅 with 𝑐𝑛 ≠ 0. Let 𝔨 be any maximal ideal of
𝑅 not containing 𝑐𝑛 , which exists since rad(𝑅) = {0}.
As in the proof of the previous Lemma, 𝑎 becomes integral over 𝑅𝔨 and Nakayama’s
Lemma for 𝑅𝔨 -modules entails 𝔨𝐴 ≠ 𝐴, hence 𝔨 = 0 because 𝐴 is a field. This implies
that 𝑅 is a field and 𝐴 is a finite extension of 𝑅.
Corollary 3.2.6. Let 𝕜 be an algebraically closed field, and 𝐴 ∶= 𝕜[𝑋1 , … , 𝑋𝑛 ]. The maximal
ideals of 𝐴 are in bijection with 𝕜𝑛 by attaching to each 𝑥 ∶= (𝑥1 , … , 𝑥𝑛 ) ∈ 𝕜𝑛 the ideal

𝔪𝑥 = {𝑓 ∈ 𝐴 ∶ 𝑓 (𝑥) = 0} = (𝑋1 − 𝑥1 , … , 𝑋𝑛 − 𝑥𝑛 ).
∼
Proof. Since 𝐴/𝔪𝑥 → 𝕜 by evaluation at 𝑥, we see 𝔪𝑥 is indeed maximal. It is routine
to show that 𝑥 = 𝑦 ⟺ 𝔪𝑥 = 𝔪𝑦 . It remains to show that every maximal ideal
𝔫 contains some 𝔪𝑥 . Indeed, Theorem 3.2.5 implies the field 𝐴/𝔫 is algebraic over 𝕜,
∼
hence 𝐴/𝔫 ≃ 𝕜 as 𝕜-algebras. Let 𝑥𝑖 be the image of 𝑋𝑖 under 𝐴 ↠ 𝐴/𝔫 → 𝕜 and set
𝑥 ∶= (𝑥1 , … , 𝑥𝑛 ), then 𝔫 ⊃ 𝔪𝑥 as required.
Corollary 3.2.7. Keep the notations above and set

𝑍(𝔞) ∶= {𝑥 ∈ 𝕜𝑛 ∶ ∀𝑓 ∈ 𝔞, 𝑓 (𝑥) = 0},

𝐼(𝒳) ∶= {𝑓 ∈ 𝐴 ∶ ∀𝑥 ∈ 𝑋, 𝑓 (𝑥) = 0}

for ideals 𝔞 ⊂ 𝐴 and subsets 𝒳 ⊂ 𝕜𝑛 , then we have 𝐼𝑍(𝔞) = √𝔞 for all 𝔞.

If we identify 𝕜𝑛 with MaxSpec(𝐴) by the previous Corollary, then 𝑍(𝔞) is just the
intersection of 𝑉(𝔞) ⊂ Spec(𝐴) and MaxSpec(𝐴). Details are left to the readers.
⋅ 30 ⋅ Integral dependence, Nullstellensatz and flatness

Proof. If 𝑓 ∈ 𝐴 and 𝑓 𝑛 ∈ 𝔞 for some 𝑛, the vanishing of 𝑓 𝑛 on 𝑍(𝔞) will entail that of
𝑓 , hence the inclusion ⊃ holds. Assume conversely that 𝑓 ∈ 𝐴 vanishes on 𝑍(𝔞). This
means: for every maximal ideal 𝔪𝑥 we have

𝔪𝑥 ⊃ 𝔞 ⟺ 𝑥 ∈ 𝑍(𝔞) ⟹ 𝑓 (𝑥) = 0 ⟺ 𝑓 ∈ 𝔪𝑥 .

Hence
𝑓 ∈ ⋂ 𝔪𝑥 = √𝔞,
𝔪𝑥 ⊃𝔞

the last equality being based on Definition 3.2.1 since 𝐴/𝔞 is a Jacobson ring. This
concludes the ⊂.
Remark 3.2.8. How about 𝑍𝐼(𝒳)? Unwinding definitions, it is seen to equal the set of
points that “satisfy the algebraic equations that 𝒳 satisfies.” The Zariski topology on
𝕜𝑛 is defined by stipulating the subsets {𝑥 ∈ 𝕜𝑛 ∶ 𝑓 (𝑥) = 0} to be closed, for all 𝑓 ∈ 𝐴,
so we obtain 𝑍𝐼(𝒳) = 𝒳,̄ the Zariski-closure of 𝒳. The reader is invited to verify that by
identifying 𝕜𝑛 with MaxSpec(𝐴), the foregoing topology is induced from the Zariski
topology on the prime spectrum Spec(𝐴).
An (algebraic, closed) subvariety of 𝕜𝑛 is the vanishing locus 𝑓1 = ⋯ = 𝑓𝑚 = 0
for some 𝑓1 , … , 𝑓𝑚 ∈ 𝕜[𝑋1 , … , 𝑋𝑛 ]; it is determined by the ideal 𝔞 = (𝑓1 , … , 𝑓𝑚 ), in fact
it depends only on √𝔞. An ideal 𝔞 is called radical if √𝔞 = 𝔞. To recap, we obtain a
dictionary:

Subvariety 𝒳 in 𝕜𝑛 Radical ideal 𝔞 in 𝕜[𝑋1 , … , 𝑋𝑛 ]

Points of 𝒳 MaxSpec(𝐴), 𝐴 = 𝐴𝒳 ∶= 𝕜[𝑋1 , … , 𝑋𝑛 ]/𝔞
Union of two varieties product or intersection of two ideals
Intersection of varieties sum of ideals
Polynomial map 𝒳 → 𝒴 𝕜-homomorphism 𝐴𝒴 → 𝐴𝒳
⋮ ⋮

If one allows arbitrary rings 𝐴 instead of just 𝕜[𝑋1 , … , 𝑋𝑛 ]/𝔞, and consider Spec(𝐴)
instead of MaxSpec(𝐴) (the latter is well-behaved only for Jacobson rings), the result
is the category of affine schemes. A proper treatment of these ideas should be left to the
Algebraic Geometry course, if it exists......

3.3 Flatness: the first glance

To begin with, we consider a ring 𝐴 and a module 𝑁. It has been observed that the
additive functor 𝑁 ⊗ − ∶ 𝐴-Mod → 𝐴-Mod is right exact, namely it preserves the exactness
𝐴
of sequences like
• → • → • → 0.
If we consider exactness of sequences like 0 → • → • → •, the corresponding notion is
left exactness. The same applies to any additive functor 𝐹 instead of 𝑁 ⊗ −. Being both
𝐴
 §3.3 Flatness: the first glance ⋅ 31 ⋅

left and right exact is equivalent to that 𝐹 preserves all exact sequences; in this case we
say 𝐹 is an exact functor.
Definition 3.3.1. We say 𝑁 is a flat 𝐴-module if 𝑁 ⊗ − is exact. We say 𝑁 is faithfully flat
𝐴
if for every sequence 𝑀• = [⋯ → 𝑀𝑖 → 𝑀𝑖−1 → ⋯] of 𝑅-modules, we have 𝑀• ⊗ 𝑁 is
𝑅
exact if and only if 𝑀• is.
Remark 3.3.2. In view of the right-exactness of ⊗, to assure flatness of 𝑁 it suffices that
𝑁 ⊗ − preserves kernels.
𝐴
Now we consider a ring homomorphism 𝐴 → 𝐵, which makes 𝐵 into an 𝐴-algebra.
Tensor product now gives an additive functor, often called the base change:
𝐵 ⊗ − ∶ 𝐴-Mod → 𝐵-Mod.
𝐴

Thus we can also talk about flatness and faithful flatness of 𝐵 over 𝐴. Since 𝐵 is naturally
an 𝐴-module, this notion is compatible with the previous one.
Example 3.3.3. Let 𝑆 be a multiplicative subset of 𝐴, then 𝐴[𝑆−1 ] is flat over 𝐴. It is not
faithfully flat in general, however; see Theorem 3.5.6.
Example 3.3.4. A routine fact is that for any family (𝑀•(𝑖) ) of complexes of 𝐴-modules,
𝑖∈𝐼
we have
∀𝑖 ∈ 𝐼, 𝑀•(𝑖) is exact ⟺ ⨁ 𝑀•(𝑖) is exact.
𝑖∈𝐼
Recall that ⊗ preserves direct sums. It follows that a direct sum of modules is flat if
and only if each summand is flat. From this we deduce the flatness of free modules
since 𝐴 ⊗ 𝑀 ≃ 𝑀 functorially for each 𝑀. Furthermore, projective modules are flat as
𝐴
they are direct summands of free modules.
Exercise 3.3.5. Show that ℤ/𝑛ℤ is not flat over ℤ for 𝑛 > 1.
We list some basic properties below.
⊳ Tensor products Let 𝑀, 𝑁 be flat (resp. faithfully flat) 𝑅-modules, then so is 𝑀 ⊗
𝑅
𝑁. This follows the associativity constraint of tensor products: (− ⊗ 𝑀) ⊗ 𝑁 ≃
− ⊗ (𝑀 ⊗ 𝑁).
⊳ Transitivity Given ring homomorphisms 𝐴 → 𝐵 → 𝐶, if 𝐵 is flat (resp. faithfully
flat) over 𝐴 and 𝐶 is flat (resp. faithfully flat) over 𝐵, then 𝐶 is also flat (resp.
faithfully flat) over 𝐴. This follows from the transitivity of base change, namely
there is a isomorphism of functors 𝐴-Mod → 𝐶-Mod.
∼
(− ⊗ 𝐵) ⊗ 𝐶 → − ⊗ 𝐶.
𝐴 𝐵 𝐴

⊳ Base change Suppose 𝑁 is a flat (resp. faithfully flat) 𝐴-module and 𝐵 is any 𝐴-
algebra, then 𝑁 ⊗ 𝐵 is a flat (resp. faithfully flat) 𝐵-module. Again, any 𝐵-module
𝐴
𝑀 can be viewed as an 𝐴-module, and there is a functorial isomorphism
∼
(𝑁 ⊗ 𝐵) ⊗ 𝑀 → 𝑁 ⊗ 𝑀.
𝐴 𝐵 𝐴
⋅ 32 ⋅ Integral dependence, Nullstellensatz and flatness

𝑑𝑖
Remark 3.3.6. A sequence [⋯ → 𝑀𝑖 𝑀𝑖−1 → ⋯] of 𝑅-modules is a complex (resp.
exact) if and only if so is its localization at 𝔪, for every maximal ideal 𝔪. Indeed:
⋄ (𝑀• , 𝑑• ) is a complex if and only if im(𝑑𝑖−1 𝑑𝑖 ) = 0 for all 𝑖. Since localization is an
exact functor, it preserves images and we know a module 𝑁 is zero if and only if
𝑁𝔪 = 0 for all 𝔪.
⋄ a complex (𝑀• , 𝑑• ) is exact if and only if H𝑖 (𝑀• ) = 0 for all 𝑖. The same reasoning
applies since localization preserves 𝐻𝑖 .

Proposition 3.3.7. The following are equivalent for an 𝑅-module 𝑁. (i) 𝑁 is flat over 𝑅,
(ii) 𝑁𝔭 is flat over 𝑅𝔭 for all prime ideal 𝔭, (iii) 𝑁𝔪 is flat over 𝑅𝔪 for all maximal ideal 𝔪.

Proof. Direct consequence of the exactness of localization and Remark 3.3.6.

Lemma 3.3.8. Let 𝜑 ∶ 𝑅 → 𝑅′ be a ring homomorphism, 𝔭′ ∈ Spec(𝑅′ ) maps to 𝔭 ∈ Spec(𝑅)

under 𝜑♯ . If 𝜑 is flat, so is the induced homomorphism 𝑅𝔭 → 𝑅′𝔭′ .

Proof. Set 𝑆 = 𝑅 ∖ 𝔭 so that 𝜑(𝑆) ⊂ 𝑅′ ∖ 𝔭′ . Factorize 𝑅𝔭 → 𝑅𝔭′ as

𝑅𝔭 → 𝑅 ′ [𝜑(𝑆)−1 ] → 𝑅′ .
⎵⎵⎵⎵⎵ 𝔭′
as a ring

The first arrow is also the base-change to 𝑅𝔭 of 𝜑 (as a homomorphism of 𝑅-modules),

whereas the second one is a localization of 𝑅′ . Their composite is therefore flat.

Proposition 3.3.9. Let 𝜑 ∶ 𝑅 → 𝑅′ be a ring homomorphism. The following are equivalent:

(i) 𝑅′ is flat over 𝑅,

(ii) 𝑅′𝔭′ is flat over 𝑅𝔭 for all 𝔭′ ∈ Spec(𝑅′ ) with 𝔭 = 𝜑♯ (𝔭′ );

(iii) Idem, but for 𝔭′ ∈ MaxSpec(𝑅′ ).

Proof. (i) ⟹ (ii): Set 𝑆 ∶= 𝑅∖𝔭. Base change implies 𝑅′ [𝑆−1 ] is flat over 𝑅[𝑆−1 ] = 𝑅𝔭 .
Since 𝑅′𝔭′ is a localization of 𝑅[𝑆−1 ] (exercise), we conclude by transitivity.
(ii) ⟹ (iii): Trivial.
(iii) ⟹ (i): By Remark 3.3.6 applied to complexes of 𝑅′ -modules, it suffices to
show the exactness of the functor − ⊗ 𝑅′𝔭′ for all 𝔭′ ∈ MaxSpec(𝑅′ ). In view of Lemma
𝑅
3.3.8, the factorization of 𝑅 → 𝑅′𝔭′ into 𝑅 → 𝑅𝔭 → 𝑅′𝔭′ and the flatness of 𝑅 → 𝑅𝔭 show
that 𝑅′𝔭′ is indeed flat over 𝑅.

Lemma 3.3.10 (Equational criterion of flatness). An 𝑅-module 𝑁 is flat if and only if for
𝑟
all 𝑟 ≥ 1, 𝑎1 , … , 𝑎𝑟 ∈ 𝑅, 𝑥1 , … , 𝑥𝑟 ∈ 𝑁 verifying ∑𝑖=1 𝑎𝑖 𝑥𝑖 = 0, there exist 𝑠 ∈ ℤ≥1 , an
𝑅-valued matrix 𝐵 = (𝑏𝑖𝑗 )1≤𝑖≤𝑟 and 𝑦1 , … , 𝑦𝑠 ∈ 𝑁 such that
1≤𝑗≤𝑠

⎛
⎜ 𝑥1 ⎞⎟ ⎛
⎜ 𝑦1 ⎞⎟
⎜
⎜ ⎟
⎟ ⎜
⎜ ⎟
⎟
⎜ ⎟ = 𝐵 ⎜ ⎟
⎜
⎜
⎜
⋮ ⎟
⎟
⎟
⎜
⎜
⎜
⋮ ⎟,
⎟ (𝑎1 ⋯ 𝑎𝑟 ) 𝐵 = 0.
⎜ ⎟ ⎜ ⎟ ⎟
𝑥
⎝ ⎠ 𝑟 𝑦
⎝ ⎠ 𝑠
 §3.4 Structure of flat modules ⋅ 33 ⋅

𝑓
Proof. Suppose 𝑀 is flat and consider the exact sequence 0 → ker(𝑓 ) → 𝑅⊕𝑟 𝑅 where
𝑓 (𝑡1 , … , 𝑡𝑟 ) = ∑𝑖 𝑎𝑖 𝑡𝑖 . We obtain an exact

(𝑥𝑖 )𝑖 ↦∑𝑖 𝑎𝑖 𝑥𝑖
0 → ker(𝑓 ) ⊗ 𝑀 → 𝑀⊕𝑟 𝑀.
𝐴

𝑠
Thus if (𝑥1 , … , 𝑥𝑟 ) ↦ 0 by the arrow above, we can express it as ∑𝑗=1 (𝑏1𝑗 , … , 𝑏𝑟𝑗 ) ⊗ 𝑦𝑗 ∈
ker(𝑓 ) ⊗ 𝑀 as required.
𝐴
To show the converse, we invoke the fact that flatness is equivalent to the injectivity
of 𝔞 ⊗ 𝑁 → 𝔞𝑁 for all finitely generated ideal 𝔞. See Proposition 3.4.1.

It will be useful to rephrase the condition in Lemma 3.3.10 as follows: for all
⋄ homomorphism 𝑥 ∶ 𝑅⊕𝑟 → 𝑀, where 𝑟 ∈ ℤ≥1 , and
⋄ 𝐾 ⊂ ker(𝑥): submodule generated by one element,
there exist some 𝑠 ∈ ℤ≥1 and a commutative diagram

𝐵
𝑅⊕𝑟 𝑅⊕𝑠
s.t. 𝐾 ⊂ ker(𝐵).
𝑥 𝑦
𝑀

Indeed, 𝑥 (resp. 𝑦) corresponds to (𝑥1 , … , 𝑥𝑟 ) via 𝑥 ∶ (𝑡1 , … , 𝑡𝑟 ) ↦ ∑𝑖 𝑡𝑖 𝑥𝑖 (resp. 𝑦 ∶

(𝑢1 , … , 𝑢𝑠 ) ↦ ∑𝑗 𝑢𝑗 , 𝑦𝑗 ), and 𝐾 ⊂ ker(𝑥) corresponds to 𝑅 ⋅ (𝑎1 , … , 𝑎𝑟 ). The homomor-
phism 𝐵 corresponds to the matrix (𝑏𝑖𝑗 )𝑖,𝑗 .
Remark 3.3.11. For flat 𝑁, the equational criterion so rephrased is applicable to any
𝑦
finitely generated 𝐾 ⊂ ker(𝑥). Indeed, one may iterate the construction for 𝑅⊕𝑠 𝑀,
etc. to make every generator of 𝐾 mapped to 0.

3.4 Structure of flat modules

Recall from homological algebra that the right exact functor 𝑁 ⊗ − ∶ 𝑅-Mod → 𝑅-Mod
𝐴
𝑅
has (Tor𝑖 (𝑁, −))𝑖≥0 as its left derived functors.

Proposition 3.4.1. The following are equivalent for an 𝑅-module 𝑁:

(i) 𝑁 is flat,
𝑅
(ii) Tor𝑖 (𝑁, −) = 0 for all 𝑖 > 0,
𝑅
(iii) Tor1 (𝑁, −) = 0,
𝑅
(iv) Tor1 (𝑁, 𝑅/𝔞) = 0 for all finitely generated ideal 𝔞, or equivalently 𝔞 ⊗ 𝑁 → 𝑁 is
𝑅
injective.
⋅ 34 ⋅ Integral dependence, Nullstellensatz and flatness

Proof. First, the equivalence mentioned in (iv) is a consequence of the exact sequence
𝑅 𝑅
Tor1⎵(𝑁,
⎵⎵ ⎵⎵⎵ 𝑅) → Tor1 (𝑁, 𝑅/𝔞) → 𝑁 ⊗ 𝔞 → 𝑁 → 𝑁 ⊗ (𝑅/𝔞) → 0
𝑅 𝑅
=0

deduced from 0 → 𝔞 → 𝑅 → 𝑅/𝔞 → 0.

Clearly (i) ⟹ (ii) ⟹ (iii) ⟹ (iv). To show (iii) ⟹ (i), note that if
𝑅
0 → 𝑀′ → 𝑀 → 𝑀″ → 0 is exact, then Tor1 (𝑁, 𝑀″ ) → 𝑁 ⊗𝑀′ → 𝑁 ⊗𝑀 → 𝑁 ⊗𝑀″ → 0
is exact.
𝑅
We show (iv) ⟹ (i), (ii) or (iii) as follows: Tor1 (𝑁, 𝑀) = 0 can be tested for
𝑅
finitely generated 𝑀 only, since Tor𝑖 (𝑁, −) preserves filtered inductive limits such as
𝑀 = lim {f.g. submodules} .
−−→
We do induction on the minimal number 𝑛 of generators of 𝑀. If 𝑀 = 𝑅𝑥1 + ⋯ + 𝑅𝑥𝑛 ,
put 𝑀′ ∶= ∑𝑖<𝑛 𝑅𝑥𝑖 so that we have a short exact sequence 0 → 𝑀′ → 𝑀 → 𝑀″ →
0 where 𝑀″ is generated by the image of 𝑥𝑛 , hence isomorphic to some 𝑅/𝔞. Since
𝑅 𝑅 𝑅
Tor1 (𝑁, 𝑀′ ) → Tor1 (𝑁, 𝑀) → Tor1 (𝑁, 𝑀″ ) is exact, we are reduced to the 𝑛 = 1 case,
i.e. 𝑀 = 𝑅/𝔞. It boils down to assure 𝔞 ⊗ 𝑁 ↪ 𝑁. Again, using the exactness of filtered
lim and the fact that ⊗ respects lim, it suffices to test this on finitely generated 𝔞.
−−→ −−→
For a down-to-earth approach, see [8, §6.3].
Corollary 3.4.2. If 𝑟 ∈ 𝑅 is not a zero divisor, then 𝑟 is not a zero divisor for any flat 𝑅-module
𝑁.
Proof. Take 𝔞 ∶= 𝑅𝑟, which is ≃ 𝑅, and contemplate on 𝑁 ≃ 𝔞 ⊗ 𝑁 ↪ 𝑁.
𝑅
Exercise 3.4.3. Suppose 𝑅 is a principal ideal domain. Show that 𝑁 is flat if and only if
𝑁 has no zero divisors except 0. Hint: the ideals take the form 𝔞 = (𝑡), so the condition
(iv) amounts to 𝑡𝑥 = 0 ⟺ 𝑥 = 0.
Exercise 3.4.4. For all field 𝕜, show that
(i) 𝕜[𝑋, 𝑌]/(𝑋𝑌 − 𝑋) is not flat over 𝕜[𝑋],
(ii) 𝕜J𝑡K[𝑌, 𝑍]/(𝑌𝑍 − 𝑡) is flat over 𝕜J𝑡K.
Lemma 3.4.5. Suppose 𝐴 → 𝐵 is flat. Write 𝑀𝐵 ∶= 𝐵 ⊗ 𝑀 for any 𝐴-module 𝐵. Then there
𝐴
𝐵 𝐴
are natural isomorphisms Tor𝑖 (𝑀𝐵 , 𝑁𝐵 ) ≃𝐵⊗ Tor𝑖 (𝑀, 𝑁) for all 𝑖.
𝐴
Proof. Take a projective resolution 0 ← 𝑀 ← 𝑃• of 𝐴-modules. Since 𝐵 is flat over 𝐴,
its base-change 0 ← 𝑀𝐵 ← 𝑃•,𝐵 to 𝐵 is still a projective resolution; we are using the fact
𝐵
that base-change preserves projectivity. Hence H𝑖 (𝑃•,𝐵 ⊗ 𝑁𝐵 ) computes Tor𝑖 (𝑀𝐵 , 𝑁𝐵 ).
𝐵
On the other hand, by flatness the homology groups are equal to H𝑖 (𝑃• ⊗ 𝑁) ⊗ 𝐵,
𝐴 𝐴
𝐴
that is, Tor𝑖 (𝑀, 𝑁) ⊗ 𝐵. We leave it to the reader to convince him- or herself that the
𝐴
isomorphism so constructed is natural.
Another way is to use to associativity and commutativity of tensor products on the
derived level, namely the flatness of 𝐴 → 𝐵 implies
L L L L L
𝑀𝐵 ⊗𝐵 𝑁𝐵 ≃ (𝑀 ⊗𝐴 𝐵) ⊗𝐵 (𝑁 ⊗𝐴 𝐵) ≃ (𝑀 ⊗𝐴 𝑁) ⊗𝐴 𝐵
in the derived categories, and it remains to take H𝑖 .
 §3.4 Structure of flat modules ⋅ 35 ⋅

In particular, let 𝑆 ⊂ 𝑅 be a multiplicative subset. By Example 3.3.3 we infer

𝑅[𝑆−1 ] 𝑅
Tor𝑖 (𝑀[𝑆−1 ], 𝑁[𝑆−1 ]) ≃ Tor𝑖 (𝑀, 𝑁)[𝑆−1 ].

Theorem 3.4.6. Let 𝑅 be a local ring with maximal ideal 𝔪. Let 𝑀 be a finitely generated
𝑅-module. The following are equivalent:
⋄ 𝑀 is free,
⋄ 𝑀 is projective.
If we assume moreover that 𝑀 is finitely presented, both conditions are equivalent to the flatness
of 𝑀.
Proof. Free modules are known to be projective. Now let 𝑀 be finitely generated pro-
jective, and take a basis 𝑥1̄ , … 𝑥𝑛̄ of the 𝑅/𝔪-vector space 𝑀/𝔪𝑀, together with liftings
𝑀 ∋ 𝑥𝑖 ↦ 𝑥𝑖̄ . Nakayama’s Lemma then implies the surjectivity of

𝛷 ∶ 𝑅⊕𝑛 ⟶ 𝑀
(𝑎1 , … , 𝑎𝑛 ) ⟼ 𝑎1 𝑥1 + ⋯ + 𝑎𝑛 𝑥𝑛 .

As 𝑀 is projective, 𝛷 admits a section so that we may identify 𝑀 with a direct summand

of 𝑅⊕𝑛 , namely 𝑀 ⊕ 𝑁 = 𝑅⊕𝑛 for some 𝑁. Taking − ⊗𝑅 𝑅/𝔪 leads to

(𝑅/𝔪)⊕𝑛 = 𝑀/𝔪𝑀 ⊕ 𝑁/𝔪𝑁 as vector spaces over 𝑅/𝔪,

and by comparing dimensions we see 𝑁/𝔪𝑁 = {0}, which in turn gives 𝑁 = {0} by
Nakayama’s Lemma (𝑁 is finitely generated since 𝑅⊕𝑛 ↠ 𝑁). Hence 𝑀 ≃ 𝑅⊕𝑛 is free.
Now turn to the second assertion. Projective modules are flat since they are direct
summands of free modules, and it remains to show that every flat 𝑀 with finite pre-
𝑥
sentation 𝑅⊕𝑞 → 𝑅⊕𝑟 𝑀 → 0 is a direct summand of a free module. Let’s plug
𝑥 ∶ 𝑅⊕𝑟 ↠ 𝑀 and 𝐾 ∶= ker(𝑥) into the equational criterion of flatness (Lemma 3.3.10),
rephrased as in Remark 3.3.11. Let 𝑁 be the image of 𝐵 ∶ 𝑅⊕𝑟 → 𝑅⊕𝑠 . One readily sees
∼
that 𝑦 induces 𝑁 → 𝑀. This furnishes a section 𝑠 ∶ 𝑀 → 𝑅⊕𝑠 for 𝑦.
We deduce the following result characterizing finitely presented projective mod-
ules: in geometric language, they correspond to vector bundles over the affine scheme
Spec(𝑅).
Corollary 3.4.7. Let 𝑀 be a finitely presented 𝑅-module. Then 𝑀 is projective if and only 𝑀𝔪
is free for every maximal ideal 𝔪.
Proof. In view of Theorem 3.4.6, it suffices to show 𝑀 is projective if and only if 𝑀𝔪
is for all maximal ideal 𝔪. One direction is easy: if 𝑀 is a direct summand of a free
module, then so is 𝑀𝔪 .
Conversely, the assumption on finite presentation entails an isomorphism between
additive functors 𝑅-Mod → 𝑅𝔪 -Mod

Hom𝑅 (𝑀, −) ⊗ 𝑅𝔪 ≃ Hom𝑅𝔪 (𝑀𝔪 , (−)𝔪 ).

𝑅

Hence 𝑀𝔪 is projective for all 𝔪 implies 𝑀 is projective, by Remark 3.3.6 and the exact-
ness of localizations.
⋅ 36 ⋅ Integral dependence, Nullstellensatz and flatness

Note that for finitely presented 𝑀, the equivalence between projectivity and flatness
holds for any ring 𝑅. The arguments are verbatim, and this can also be deduced from
the local case.
We close this section by a stronger result, whose proof is referred to [8, Theorem
A6.6].

Theorem 3.4.8 (Govorov–Lazard). An 𝑅-module is flat if and only if it is a filtered lim of

−−→
free 𝑅-modules.

3.5 Faithful flatness and surjectivity

We begin by a contemplation on the definition of faithful flatness. Recall that 𝑅-Mod is
the template of abelian categories, in which one can talk about the zero object 0, direct
sums, kernels, cokernels, images and exactness.
We say a functor between categories is faithful if it induces injections on Hom-sets.
A functor between additive categories is called additive if it induces group homomor-
phisms between Hom-sets. An additive functor between abelian categories is exact if it
preserves all exact sequences. Exact functors preserve kernels, cokernels and images.

Lemma 3.5.1. Let 𝐹 ∶ 𝒞 → 𝒞 ′ be an additive functor between abelian categories. The following
are equivalent.

(i) 𝐹 is exact and faithful;

(ii) 𝐹 is exact and (𝐹𝑀 = 0 ⟺ 𝑀 = 0) for every object 𝑀 of 𝒞;

(iii) a sequence 𝑀′ → 𝑀 → 𝑀″ is exact in 𝒞 if and only if 𝐹𝑀′ → 𝐹𝑀 → 𝐹𝑀″ is exact in

𝒞 ′.

Proof. (i) ⟹ (ii): If 𝐹𝑀 = 0 then 𝐹(id𝑀 ) = id𝐹𝑀 = 0, and the faithfulness implies
id𝑀 = 0 in End𝒞 (𝑀); this is possible only when 𝑀 = 0.
(ii) ⟹ (i): Suppose 𝑢 ∶ 𝑁 → 𝑀 is mapped to 0 under 𝐹. Then we have 𝐹(im(𝑢)) =
0, thereby im(𝑢) = 0 and 𝑢 = 0.
𝑢 𝑣 𝐹𝑢 𝐹𝑣
(i) ⟹ (iii): Suppose 𝑀′ 𝑀 𝑀″ induces an exact sequence 𝐹𝑀′ 𝐹𝑀
″
𝐹𝑀 . From 𝐹(𝑣𝑢) = 𝐹(𝑣)𝐹(𝑢) = 0 we get 𝑣𝑢 = 0. Thus it makes sense to define
𝐶 ∶= ker(𝑣)/ im(𝑢). One has an exact sequence

im(𝑢) → ker(𝑣) → 𝐶 → 0.

Since 𝐹 is exact, we deduce 𝐹𝐶 = 0, which implies 𝐶 = 0 by (i) ⟹ (ii).

(iii) ⟹ (i): Suppose 𝑢 ∶ 𝑁 → 𝑀 is mapped to 0 under 𝐹. Consider 𝑣 ∶ 𝑀 →
coker(𝑢). Since 𝐹 preserves exact sequences and 𝐹𝑣 is an isomorphism, we see 𝑣 is also
an isomorphism, therefore 𝑢 = 0.

Proposition 3.5.2. The following are equivalent for an 𝑅-module 𝑁.

(i) 𝑁 is faithfully flat.
(ii) The functor − ⊗ 𝑁 is exact and faithful.
𝑅
 §3.5 Faithful flatness and surjectivity ⋅ 37 ⋅

(iii) 𝑁 is flat and for every maximal ideal 𝔪 of 𝑅, we have 𝑁/𝔪𝑁 ≃ 𝑁 ⊗ 𝑅/𝔪 ≠ 0.
𝑅

Proof. The equivalence (i) ⟺ (ii) has just been established. An 𝑅-module 𝑀 is
nonzero if and only if there exist exact sequences

0 → 𝑅/𝔞 → 𝑀, 𝑅/𝔞 → 𝑅/𝔪 → 0

where 𝔞 is a proper ideal and 𝔪 is a maximal over-ideal of 𝔞. Next, let’s show (iii) ⟹
(i) or (ii): there are exact sequences

0 → 𝑇 → 𝑀 ⊗ 𝑁, 𝑇 → 𝑀 ⊗ (𝑅/𝔪) → 0
𝑅 𝑅

where 𝑇 ∶= (𝑅/𝔞) ⊗ 𝑁, therefore 𝑀 ⊗ 𝑁 ≠ 0 and we apply Lemma 3.5.1. Finally, (i) or

𝑅 𝑅
(ii) ⟹ (iii) is clear.
Corollary 3.5.3. Let 𝑅 → 𝑅′ be a local homomorphism1 between local rings and let 𝑀 be a
finitely generated 𝑅′ -module, 𝑀 ≠ 0. Then as an 𝑅-module, 𝑀 is faithfully flat if and only if it
is flat.
Proof. Let 𝔪, 𝔪′ be the maximal ideals of 𝑅, 𝑅′ , respectively. Since 𝔪 is mapped into
𝔪′ = rad(𝑅′ ), the assertion follows from Proposition 3.5.2 together with Nakayama’s
Lemma.
This is often applied to the cases 𝑅 = 𝑅′ or 𝑀 = 𝑅′ .
Exercise 3.5.4 (Faithfully flat descent for flatness). Given a faithfully flat homomor-
phism 𝜑 ∶ 𝑅 → 𝑅′ . If 𝑀 is an 𝑅-module such that 𝑀 ⊗ 𝑅′ is a flat (resp. faithfully flat)
𝑅
𝑅′ -module, then so is 𝑀 over 𝑅.
Proposition 3.5.5. Suppose 𝜑 ∶ 𝐴 → 𝐵 is a faithfully flat ring homomorphism and regard 𝐵
as an 𝐴-algebra.
⋄ For any 𝐴-module 𝑀, the natural map 𝑀 → 𝑀 ⊗ 𝐵 is injective; in particular, 𝜑 is seen
𝐴
to be injective by taking 𝑀 = 𝐴.

⋄ For any ideal 𝔞 ⊊ 𝐴 we have 𝜑−1 (𝔞𝐵) = 𝔞.

⋄ The map 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) is surjective.

Proof. Let 𝑁 be the kernel of 𝑀 → 𝑀⊗ 𝐵. The 𝐵-module homomorphism 𝑁⊗ 𝐵 → 𝑀⊗ 𝐵
𝐴 𝐴 𝐴
is injective by flatness; it is zero on 𝑁 ⊗ 1, hence identically zero and we obtain 𝑁 = {0}
by faithful flatness (Lemma 3.5.1).
Let 𝔞 ⊂ 𝐴 be a proper ideal. The previous step with 𝑀 ∶= 𝐴/𝔞 implies that 𝐴/𝔞 →
(𝐴/𝔞) ⊗ 𝐵 ≃ 𝐵/𝔞𝐵 is injective, hence 𝜑−1 (𝔞𝐵) = 𝔞.
𝐴
To show the surjectivity of 𝜑♯ , consider a given 𝔭 ∈ Spec(𝐴) and form the faithfully
flat ring homomorphism
𝜑 𝔭 ∶ 𝐴 𝔭 → 𝐵 ⊗ 𝐴𝔭 = 𝐵 𝔭
𝐴
1
That is, the preimage of the maximal ideal 𝔪𝑅′ equals 𝔪𝑅 .
⋅ 38 ⋅ Integral dependence, Nullstellensatz and flatness

by base change. The previous step implies 𝜑−1

𝔭 (𝔭𝐵𝔭 ) = 𝔭𝐴𝔭 , hence 𝔭𝐵𝔭 is a proper ideal
of 𝐵𝔭 . Any maximal over-ideal 𝔪0 of 𝔭𝐵𝔭 satisfies 𝜑−1
𝔭 (𝔪0 ) = 𝔭𝐴𝔭 . Take 𝔪 ∈ Spec(𝐵)
mapping to 𝔪0 . From the commutative diagrams

𝐵 𝐵𝔭 Spec(𝐵) Spec(𝐵𝔭 )
𝜑 𝜑𝔭 𝜑♯ 𝜑♯𝔭

𝐴 𝐴𝔭 Spec(𝐴) Spec(𝐴𝔭 )

one infers 𝜑♯ (𝔪) = 𝔭.

Theorem 3.5.6. The following are equivalent for a ring homomorphism 𝜑 ∶ 𝐴 → 𝐵.

(i) 𝜑 is faithfully flat;

(ii) 𝜑 is flat and 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) is surjective;

(iii) 𝜑 is flat and for any maximal ideal 𝔭 ⊂ 𝐴 there exists a maximal ideal 𝔪 ⊂ 𝐵 such that
𝜑−1 (𝔪) = 𝔭.

Proof. (i) ⟹ (ii) is contained in Proposition 3.5.5. As for (ii) ⟹ (iii), take any
𝔮 ∈ Spec(𝐵) that pulls back to 𝔭 ∈ Spec(𝐴), then any maximal over-ideal 𝔪 of 𝔮 also
pulls back to 𝔭. To show (iii) ⟹ (i), apply the criterion of Proposition 3.5.2: for any
𝔭 ∈ MaxSpec(𝐴), the existence of 𝔪 ↦ 𝔭 implies 𝔪 ⊃ 𝜑(𝔭) ⋅ 𝐵 = 𝔭𝐵, hence 𝔭𝐵 ≠ 𝐵.
The notion of flatness was first introduced by J.-P. Serre in [15]. The surjections
𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) (or rather their global avatars) for faithfully flat 𝜑 ∶ 𝐴 → 𝐵
are often employed as candidates of “coverings” in algebraic geometry, leading up to
the well-known fpqc (faithfully flat + quasi-compact) and fppf (faithfully flat + finitely
presented) topologies in the sense of Grothendieck. They have been indispensable
tools for contemporary geometers; [17] serves as a readable introduction to this circle
of ideas.
 Lecture 4

Going-up, going-down,
gradings and filtrations
This lecture will be less self-contained than the other ones.

4.1 Going-up and going-down

In geometry, it is often crucial to know properties of a morphism between algebraically
defined geometric objects. Rephrased in terms of commutative algebra, our goal is to
understand the image of 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) where 𝜑 ∶ 𝐴 → 𝐵 is a given ring
homomorphism.

⋄ We say the going-up property holds for 𝜑 if for every 𝔭 ⊂ 𝔭′ in Spec(𝐴) and 𝔮 ∈
Spec(𝐵) with 𝜑♯ (𝔮) = 𝔭 (and we say 𝔮 lies over 𝔭...) there exists 𝔮′ ⊃ 𝔮 lying over
𝔭′ .

⋄ We say the going-down property holds for 𝜑 if for every 𝔭 ⊂ 𝔭′ in Spec(𝐴) and
𝔮′ ∈ Spec(𝐵) lying over 𝔭′ , there exists 𝔮 ⊂ 𝔮′ lying over 𝔭.

Pictorially:

𝔮′ 𝐵

𝔭′ 𝐴

Lemma 4.1.1 (Existence of minimal over-primes). Let 𝔞 be a proper ideal in a ring 𝑅.

There exists a prime ideal 𝔭 which is minimal among all primes containing 𝔞. If 𝔓 is an ideal
containing 𝔞, one can choose 𝔭 ⊂ 𝔓.
⋅ 40 ⋅ Going-up, going-down, gradings and filtrations

Proof. One easily reduces to the case 𝔞 = {0}. We want to use Zorn’s Lemma to find a
minimal prime. It boils down to show that any chain (𝔭𝑖 )𝑖∈𝐼 of prime ideals (𝐼: totally
ordered set with 𝑗 > 𝑖 ⟹ 𝔭𝑖 ⊃ 𝔭𝑗 ) has a lower bound; we assume 𝔭𝑖 ⊂ 𝔓 when 𝔓
is prescribed. It suffices to show 𝔭 ∶= ⋂𝑖∈𝐼 𝔭𝑖 is prime: if 𝑥𝑦 ∈ 𝔭 but there exists 𝑖 with
𝑥 ∉ 𝔭𝑖 , then 𝑥 ∉ 𝔭𝑗 whenever 𝑗 ≥ 𝑖; in this case 𝑗 ≥ 𝑖 ⟹ 𝑦 ∈ 𝔭𝑗 . This entails 𝑦 ∈ 𝔭.

Lemma 4.1.2. The going-down property for 𝜑 is equivalent to the following: for every 𝔭 ∈
Spec(𝐴) with 𝜑(𝔭)𝐵 ≠ 𝐵 and any minimal over-prime 𝔮 of 𝜑(𝔭)𝐵, we have 𝜑♯ (𝔮) = 𝔭.

Proof. Assuming going-down for 𝜑, let 𝔭, 𝔮 be as above. Evidently 𝜑♯ (𝔮) ⊃ 𝜑−1 (𝜑(𝔭)𝐵) ⊃
𝔭. If we have ⊋, then going-down guarantees the existence of 𝔮♭ ⊊ 𝔮 lying over 𝔭. Thus
𝜑−1 (𝔮♭ ) = 𝔭 implies 𝔮♭ ⊃ 𝜑(𝔭)𝐵, contradicting the minimality of 𝔮.
To show the converse, consider 𝔭 ⊂ 𝔭′ with 𝔮′ lying over 𝔭′ . We have 𝜑(𝔭)𝐵 ⊂
𝜑(𝔭′ )𝐵 ⊂ 𝔮′ ≠ 𝐵. Take 𝔮 to be a minimal over-prime of 𝜑(𝔭)𝐵 (which exists by Lemma
4.1.1) to verify the going-down property.

Theorem 4.1.3. Going-down holds for flat 𝜑 ∶ 𝐴 → 𝐵.

Proof. Consider 𝔭 ⊂ 𝔭′ and 𝔮′ lying over 𝔭′ in the setting of going-down. First, 𝐵𝔮′ is
flat over 𝐴𝔭′ by Proposition 3.3.9. Secondly, 𝐴𝔭′ → 𝐵𝔮′ is faithfully flat since it is local
by Theorem 3.5.6 (iii), therefore induces a surjection on spectra. Take any prime of 𝐵𝔮′
mapping to 𝔭𝐴𝔭 ∈ Spec(𝐴𝔭′ ) and to 𝔮 ∈ Spec(𝐵). In view of the commutative diagrams

𝐵 𝐵𝔮′ Spec(𝐵) Spec(𝐵𝔮′ )

𝜑 𝜑𝔭′ 𝜑♯ 𝜑♯𝔭′

𝐴 𝐴𝔭′ Spec(𝐴) Spec(𝐴𝔭′ )

we see 𝔮 is the required prime in going-down.

Theorem 4.1.4 (Krull–Cohen–Seidenberg). Suppose the ring 𝐵 is integral over its subring
𝐴. The following holds.

(i) The map Spec(𝐵) → Spec(𝐴) given by 𝔮 ↦ 𝔮 ∩ 𝐴 is surjective.

(ii) There are no inclusion relations in any fiber of Spec(𝐵) → Spec(𝐴).

(iii) Going-up holds for 𝐴 ↪ 𝐵.

(iv) If 𝐴 is a local ring and 𝔭 ∈ MaxSpec(𝐴), then the fiber {𝔮 ∶ 𝔮 ∩ 𝐴 = 𝔭} equals

MaxSpec(𝐵).

(v) Assume 𝐴, 𝐵 are domains and 𝐴 is normal. Then going-down holds for 𝐴 ↪ 𝐵.

(vi) Assume furthermore that 𝐵 is the integral closure of 𝐴 in a normal field extension 𝐿 ⊃
𝐾 ∶= Frac(𝐴), then 𝛤 ∶= Aut(𝐿/𝐾) acts transitively on each fiber of Spec(𝐵) →
Spec(𝐴).

The last assertion should be familiar to readers with a background in algebraic num-
ber theory.
 §4.1 Going-up and going-down ⋅ 41 ⋅

Proof. (iv): Let 𝔮 ∈ MaxSpec(𝐵) and 𝔭0 ∶= 𝔮 ∩ 𝐴. We know 𝐵/𝔮 is a field, integral over
its subring 𝐴/𝔭0 . We claim that 𝐴/𝔭0 is also a field, therefore 𝔭0 = 𝔭. Let 𝑥 ∈ 𝐴/𝔭0 ∖{0}.
Its inverse in 𝐵/𝔮 satisfies an integral relation ( 1𝑥 )𝑛 +𝑎𝑛−1 ( 1𝑥 )𝑛−1 +⋯ +𝑎0 = 0 over 𝐴/𝔭0 .
Multiplying both sides by 𝑥𝑛−1 yields 1𝑥 ∈ (𝐴/𝔭0 )[𝑥].
Conversely, we have to show any 𝔮 ∈ Spec(𝐵) with 𝔮 ∩ 𝐴 = 𝔭 is maximal. Again,
there is an integral extension of domains 𝐴/𝔭 ↪ 𝐵/𝔮. Consider 𝑦 ∈ 𝐵/𝔮 satisfying
𝑦𝑛 + 𝑎𝑛−1 𝑦𝑛−1 + ⋯ + 𝑎0 = 0, with the smallest possible 𝑛. If 𝑦 ≠ 0 then 𝑎0 ≠ 0. Since
𝐴/𝔭 is a field, the recipe to produce 𝑦−1 ∈ (𝐴/𝔭)[𝑦] is well-known.
(i), (ii): Fix 𝔭 and consider the inclusion 𝐴𝔭 ↪ 𝐵𝔭 = 𝐵 ⊗ 𝐴𝔭 which is still integral
𝐴
(note that 𝐴 ∖ 𝔭 is a multiplicative subset of 𝐵, and 𝐵𝔭 is nonzero). We are reduced to
the case 𝐴 is local with maximal ideal 𝔭. By (iv) the fiber of 𝔭 in Spec(𝐵) is nothing but
MaxSpec(𝐵). This establishes (i) and (ii) since there are no inclusions among maximal
ideals.
(iii): Consider 𝔭 ⊂ 𝔭′ and 𝔮 over 𝔭 in the setting of going-up. Then (i) is applicable
to 𝐴/𝔭 ↪ 𝐵/𝔮 and yields the required 𝔮′ ∈ Spec(𝐵/𝔮) ↪ Spec(𝐵).
(vi): Observe that every 𝜎 ∈ 𝛤 induces an 𝐴-automorphism of 𝐵. Let 𝐾 ′ ∶= 𝐿𝛤 . By
(infinite) Galois theory we know 𝐿/𝐾 ′ is Galois and 𝐾 ′ /𝐾 is purely inseparable. Let 𝐴′
be the integral closure of 𝐴 in 𝐾 ′ . First observe that

Spec(𝐴′ ) ⟶ Spec(𝐴), 𝔭 ′ ↦ 𝔭 = 𝔭′ ∩ 𝐴

is a bijection. Indeed, 𝐾 ′ ≠ 𝐾 only when 𝑝 ∶= char(𝐾) > 0, in which case the inverse
𝑚
is given by 𝔭′ = {𝑡 ∈ 𝐴′ ∶ 𝑡𝑝 ∈ 𝔭, 𝑚 ≫ 0}. Thus we assume henceforth that 𝐿/𝐾 is
Galois.
Deal with the case [𝐿 ∶ 𝐾] < ∞ first. Consider 𝔮, 𝔮′ ∈ Spec(𝐵) in the fiber over 𝔭.
Suppose on the contrary that 𝛤𝔮 does not meet 𝔮′ , then by (ii) we have 𝔮′ ⊄ 𝜎(𝔮) for
each 𝜎 ∈ 𝛤. By the prime avoidance (Proposition 1.1.5), there exists 𝑥 ∈ 𝔮′ ∖ ⋃𝜎 𝜎(𝔮),
since 𝛤 is finite. Now define the norm 𝑦 ∶= 𝑁𝐿/𝐾 (𝑥) ∈ 𝐾, which is some positive power
(namely [𝐿 ∶ 𝐾]𝑖 ) of ∏𝜎∈𝛤 𝜎(𝑥), hence belongs to 𝐵. Notice that
⋄ 𝐴 normal implies 𝑦 ∈ 𝐴;
⋄ 𝑥 ∉ 𝜎 −1 (𝔮) for all 𝜎 ∈ 𝛤 implies 𝑦 ∉ 𝔭;
⋄ however 𝑦 ∈ 𝔮′ ∩ 𝐴 = 𝔭 since 𝑥 ∈ 𝔮′ . Contradiction.
Now suppose [𝐿 ∶ 𝐾] is infinite and 𝔮, 𝔮′ in the fiber over 𝔭. We need to use the Krull
topology on 𝛤. For every finite, Galois subextension 𝐸/𝐾 of 𝐿/𝐾, define the set

𝒯 (𝐸) ∶= {𝜎 ∈ 𝛤 ∶ 𝜎(𝔮 ∩ 𝐸) = 𝔮′ ∩ 𝐸} .

By the finite case we know 𝒯 (𝐸) ≠ ∅. Furthermore,

⋄ 𝒯 (𝐸) is closed in 𝛤 (it is the preimage of some subset of Gal(𝐸/𝐾));
⋄ 𝐸′ ⊂ 𝐸 ⟹ 𝒯 (𝐸′ ) ⊃ 𝒯 (𝐸);
⋄ the intersection of all 𝒯 (𝐸) is nonempty (by the compactness of 𝛤 and the previ-
ous step).
Taking 𝜎 ∈ ⋂𝐸 𝒯 (𝐸) gives 𝜎(𝔮) = 𝔮′ .
(v): Define 𝐿0 ∶= Frac(𝐵) and 𝐾 ∶= Frac(𝐴). Then 𝐿0 /𝐾 is algebraic and we may
take a normal closure 𝐿 of 𝐿0 over 𝐾. Let 𝐶 be the integral closure of 𝐴 (thus of 𝐵) in
𝐿. Consider the setting 𝔭 ⊂ 𝔭′ and 𝔮 ∩ 𝐴 = 𝔭 of going-down. Take any 𝔯 ∈ Spec(𝐶)
mapping to 𝔭. By (iii) for 𝐴 → 𝐶 we obtain 𝔯1 ∈ Spec(𝐶) such that 𝔯1 ↦ 𝔭′ and
⋅ 42 ⋅ Going-up, going-down, gradings and filtrations

𝔯1 ⊃ 𝔯. Next, take 𝔯2 ∈ Spec(𝐶) mapping to 𝔮′ ; by (vi) there exists 𝜎 ∈ Aut(𝐿/𝐾) with

𝜎(𝔯1 ) = 𝔯2 . One can check that 𝔮 ∶= 𝜎(𝔯) ∩ 𝐵 is the required prime ideal. Explained
pictorially:
𝜎

𝔯 𝔯1 𝔯2 𝐶

𝔮′ 𝐵

𝔭 𝔭′ 𝐴

we first construct 𝔯 and then 𝔯1 by going-up, then “tilt” it via some 𝜎 to match 𝔯1 with
some chosen 𝔯2 above 𝔮′ , so that 𝜎(𝔯) ∩ 𝐵 produces the required going-down:

Exercise 4.1.5. Let 𝐴 ⊂ 𝐵 be an integral extension of rings. Show that Spec(𝐵) →

Spec(𝐴) is a closed map (Cf. Proposition 4.2.1.) Hint: Let 𝔟 ⊂ 𝐵 be a proper ideal, then
𝐴/𝔟∩𝐴 ↪ 𝐵/𝔟 is still integral. Reduce the problem to showing that 𝑉({0𝐵 }) = Spec(𝐵)
has closed image in Spec(𝐴).

4.2 Subsets in the spectrum

Proposition 4.2.1. Let 𝜑 ∶ 𝐴 → 𝐵 be a ring homomorphism with going-up property and
suppose 𝐵 is Noetherian. Then 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) is a closed map: it maps closed
subsets to closed subsets.

Proof. Consider a closed subset 𝑉(𝔟) of Spec(𝐵). First, every 𝔮 ∈ Spec(𝐵) with 𝔮 ⊃ 𝔟
lies over a minimal over-prime of 𝔟, by Lemma 4.1.1. Secondly, 𝐵 is Noetherian im-
plies Ass(𝐵/𝔟) is finite; in particular there are only finitely many minimal over-primes
𝔮1 , … , 𝔮𝑛 of 𝔟.
Set 𝔭𝑖 ∶= 𝜑♯ (𝔮𝑖 ) for all 𝑖. By going-up, 𝑉(𝔭𝑖 ) is contained in 𝜑♯ (𝑉(𝔟)). On the other
hand, every 𝔭 = 𝜑♯ (𝔮) with 𝔮 ∈ 𝑉(𝔟) lies over some 𝔭𝑖 = 𝜑♯ (𝔮𝑖 ) by the foregoing
𝑛
discussion. This shows 𝜑♯ (𝑉(𝔟)) = ⋃𝑖=1 𝑉(𝔭𝑖 ) is closed.

Corollary 4.2.2. Suppose 𝐵 is Noetherian and integral over a subring 𝐴. Then Spec(𝐵) →
Spec(𝐴) is a closed surjection with finite fibers.

Proof. Apply Theorem 4.1.4 with Proposition 4.2.1 to show that Spec(𝐵) → Spec(𝐴) is
closed and surjective.
To show the finiteness of the fiber over 𝔭 ∈ Spec(𝐴), note that the preimage of
𝑉(𝔭) in Spec(𝐵) equals 𝑉(𝔭𝐵). Since there are no inclusions in the fiber over 𝔭, every
element in that fiber must be a minimal over-prime of 𝔭𝐵. We have seen in the proof of
Proposition 4.2.1 that there are only finitely many such minimal-over primes.
Note that the “closed surjection” part applies to any integral extension of rings. See
Exercise 4.1.5.
In order to obtain further results of this type, we have to introduce more notions.
Let 𝑅 be a ring.
 §4.2 Subsets in the spectrum ⋅ 43 ⋅

Definition 4.2.3. For 𝔭, 𝔭′ ∈ Spec(𝑅) satisfying 𝔭 ⊂ 𝔭′ , we say 𝔭 is a generalization of 𝔭′ ,

and 𝔭′ is a specialization of 𝔭.

To make geometric meaning from it, being “specialized” signifies that there are
“more equations” in 𝔭′ , therefore it corresponds a smaller embedded geometric object.
For example, in 𝑅 = ℂ[𝑋, 𝑌] the prime ideal (𝑋, 𝑌) is a specialization of (𝑋), as the
origin 𝑋 = 𝑌 = 0 belongs to the line 𝑋 = 0.

Lemma 4.2.4. With respect to the Zariski topology, 𝔭 is a generalization of 𝔭′ if and only if
𝔭′ ∈ {𝔭}.

Proof. The condition 𝔭′ ∈ {𝔭} means that for every ideal 𝔞, if 𝔭 ⊃ 𝔞 then 𝔭′ ⊃ 𝔞. Taking
𝔞 = 𝔭 yields 𝔭′ ⊃ 𝔭, and the converse is even easier.
A subset is called stable under specialization (resp. generalization) if the special-
ization (resp. generalization) of any member still belongs to that set. The following is
straightforward.

Lemma 4.2.5. Any closed subset is stable under specialization; any open subset is stable under
generalization.

Definition 4.2.6. Suppose 𝑅 is Noetherian. A subset is called locally closed if it is the

intersection of an open with a closed subset, called constructible if it is a finite union of
locally closed subsets.
Closed subsets of the form 𝑉(𝔭), where 𝔭 ∈ Spec(𝑅), are called irreducible; in this
case we call 𝔭 the generic point of 𝑍, which is uniquely characterized as the point which
generalizes every member of 𝑍.

⋄ The foregoing definition is standard only for 𝑅 Noetherian. The general definition
in EGA differs.

⋄ These notions can be applied to any topological space 𝑋. In practice one usually
suppose 𝑋 to be
– Noetherian: the closed subsets satisfy descending chain condition,
– sober: every irreducible has a generic point,
in order to get interesting results. This explains our Noetherian assumption.

⋄ If 𝕜 is algebraically closed, 𝑅 = 𝕜[𝑋1 , … , 𝑋𝑛 ]/𝔞 corresponds to an affine alge-

braic variety 𝒳 ⊂ 𝕜𝑛 and we work with MaxSpec(𝑅) instead of Spec(𝑅), then
a subset 𝐸 ⊂ 𝒳 being locally closed means that it can be defined by a formula
using the usual language of algebraic operations over ℂ, with coordinate vari-
ables 𝑋1 , … , 𝑋𝑛 and the symbols =, ≠, but without using the quantifiers ∃, ∀. For
example, the formula

(¬𝑋 = 0) ∨ (𝑋 = 0 ∧ 𝑌 = 0)

defines the constructible subset 𝐸 ∶= {(𝑥, 𝑦) ∈ 𝕜2 ∶ 𝑥 ≠ 0} ∪ {(0, 0)}, which is

neither closed nor open. Note that 𝐸 is the image of the polynomial map 𝕜2 → 𝕜2
given by (𝑥, 𝑦) ↦ (𝑥, 𝑥𝑦).
⋅ 44 ⋅ Going-up, going-down, gradings and filtrations

Exercise 4.2.7. Show that the set of constructible subsets is stable under finite ∪, finite
∩ and taking complements.

Exercise 4.2.8. Show that irreducible closed subsets 𝑍 admit the following topological
characterization: if 𝑋 = 𝐴 ∪ 𝐵 with 𝐴, 𝐵 closed, then either 𝑋 = 𝐴 or 𝑋 = 𝐵.

Suppose 𝑅 is Noetherian. Given any closed subset 𝑍 ⊂ Spec(𝑅), we may write

𝑍 = 𝑍1 ∪ ⋯ ∪ 𝑍𝑛 with each 𝑍𝑖 irreducible. One way to do this is to use the primary
decomposition for 𝔞, where we assume 𝑍 = 𝑉(𝔞); then 𝑍1 , … , 𝑍𝑛 will correspond to
the minimal elements in Ass(𝑅/𝔞). One can show by purely topological means that
such an irreducible decomposition is unique if we require 𝑖 ≠ 𝑗 ⟹ 𝑍𝑖 ⊄ 𝑍𝑗 . See [10,
I.1.5].

Lemma 4.2.9. Let 𝐸 be a subset of Spec(𝑅) where 𝑅 is a Noetherian ring. The following are
equivalent:

(i) 𝐸 is constructible;

(ii) for every irreducible subset 𝑍 of Spec(𝑅), either 𝑍 ∩ 𝐸 is not dense in 𝑍 or 𝑍 ∩ 𝐸 contains
a nonempty open subset of 𝑍.

Proof. Omitted. See [11, (6.C)].

Theorem 4.2.10 (C. Chevalley). Let 𝜑 ∶ 𝐴 → 𝐵 be a ring homomorphism such that 𝐴

is Noetherian and 𝐵 is a finitely generated 𝐴-algebra. Then 𝜑♯ maps constructible subsets to
constructible subsets.

Proof. Omitted. See [11, (6.E)].

Now we can give a partial converse to Lemma 4.2.5, albeit not in the strongest form.

Proposition 4.2.11. Suppose 𝑅 is Noetherian and 𝐸 is a constructible subset of Spec(𝑅). If 𝐸

is stable under specialization (resp. generalization), then 𝐸 is closed (resp. open) in Spec(𝑅).

Proof. It suffices to treat the specialization-stable case by taking complements. Write

the Zariski-closure 𝐸̄ as a finite union of irreducibles components 𝑍, without inclusion
relations. For each irreducible component 𝑍, notice that 𝑍 ∩ 𝐸 is dense in 𝑍 for all 𝑍,
since otherwise 𝐸̄ = ⋃𝑍 (𝑍 ∩ 𝐸) = ⋃𝑍 𝑍 ∩ 𝐸 would lead to another irreducible decom-
position. Thus 𝑍 ∩ 𝐸 contains a nonempty open of 𝑍 by Lemma 4.2.9. This open subset
of 𝐸 must contain the generic point of 𝑍. As 𝐸 is stable under specialization, we obtain
𝑍 ⊂ 𝐸. This being true for all 𝑍, we deduce that 𝐸 = 𝐸.̄

Proposition 4.2.12. Consider a ring homomorphism 𝜑 ∶ 𝐴 → 𝐵 satisfying going-down. Sup-

pose 𝐴 is Noetherian and 𝐵 is a finitely generated 𝐴-algebra, then 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) is
an open map.

Proof. Let 𝑈 = Spec(𝐵) ∖ 𝑉(𝔞) be an open subset. Going-down implies that 𝜑♯ (𝑈) is
stable under generalization. It suffices to show 𝜑♯ (𝑈) is constructible, and this is the
content of Chevalley’s Theorem 4.2.10.
 §4.3 Graded rings and modules ⋅ 45 ⋅

4.3 Graded rings and modules

Let (𝛤, +) be a commutative monoid. In most cases we consider 𝛤 = ℤ≥0 .

Definition 4.3.1. A 𝛤-graded ring is a ring 𝑅 whose underlying additive group is en-
dowed with a decomposition 𝑅 = ⨁𝛾∈𝛤 𝑅𝛾 , such that 𝑅𝛾 𝑅𝜂 ⊂ 𝑅𝛾+𝜂 for all 𝛾, 𝜂 ∈ 𝛤.
For 𝑅 as above, a 𝛤-graded 𝑅-module is an 𝑅-module 𝑀 whose underlying additive
group decomposes as 𝑀 = ⨁𝛾∈𝛤 𝑀𝛾 , such that 𝑅𝛾 𝑀𝜂 ⊂ 𝑀𝛾+𝜂 for all 𝛾, 𝜂 ∈ 𝛤; in
particular, 𝑅 itself is a 𝛤-graded 𝑅-module. If 𝑥 ∈ 𝑀𝛾 ∖ {0}, we say 𝑥 is homogeneous
of degree 𝛾.

We will often omit 𝛤 when there is no worry of confusion. Note that if 0 is allowed
to be homogeneous, as people sometimes do, it will be homogeneous of any degree.

Exercise 4.3.2. Show that in a graded ring 𝑅 we always have 1 ∈ 𝑅0 , provides that (𝛤, +)
satisfies the cancellation law: 𝛾 + 𝜂 = 𝜂 ⟺ 𝛾 = 0. Hint: let 𝑒0 be the component of 1𝑅
in degree 0, argue that 𝑥𝑒0 = 𝑥 = 𝑒0 𝑥 for all homogeneous 𝑥 ∈ 𝑅. The condition 1 ∈ 𝑅0
is sometimes built into the definition of graded rings.

Definition 4.3.3. A graded submodule of a graded 𝑅-module 𝑀 is a submodule 𝑁 with

𝑁 = ⨁𝛾 (𝑁 ∩ 𝑀𝛾 ), which gives rise to a natural grading 𝑁𝛾 ∶= 𝑁 ∩ 𝑀𝛾 on 𝑁.

For graded 𝑁 ⊂ 𝑀, the quotient 𝑅-module 𝑀/𝑁 = ⨁𝛾 𝑀𝛾 /𝑁𝛾 is again graded. As

a special case, we have the notion of graded ideals of 𝑅 (also known as homogeneous
ideals), and the quotient ring 𝑅/𝔞 with respect to graded 𝔞 inherits the evident grading.
Let 𝑁 be an 𝑅-submodule of 𝑀 and suppose 𝑀 is graded. The following are easily
seen to be equivalent:

(i) 𝑁 ⊂ 𝑀 is graded;

(ii) 𝑁 is generated by homogeneous elements;

(iii) if 𝑥 = ∑𝛾 𝑥𝛾 ∈ 𝑁 with each 𝑥𝛾 homogeneous of degree 𝛾, then ∀ 𝑥𝛾 ∈ 𝑁.

Example 4.3.4. Let 𝐴 be a ring and 𝑅 ∶= 𝐴[𝑋1 , … , 𝑋𝑛 ]. Then 𝑅 is naturally ℤ≥0 -graded
by degrees: for each 𝑑 ∈ ℤ≥0 , let 𝑅𝑑 be the set of homogeneous polynomials of total de-
gree 𝑑. Ideals generated by homogeneous polynomials are precisely the graded ideals.
The importance of this grading comes from projective algebraic geometry.
On the other hand, 𝑅 can also be graded by monomials by taking 𝛤 = ℤ≥0 ×⋯×ℤ≥0
𝑑 𝑑
(𝑛 copies), and we set 𝑅(𝑑1 ,…,𝑑𝑛 ) = 𝐴 ⋅ 𝑋1 1 ⋯ 𝑋𝑛𝑛

Many constructions in commutative algebra can be generalized to the graded case.

Let us illustrate what one can do in an important case, the primary decomposition (cf.
[8, §3.5 and Exercise 3.5]). In the ℤ-graded context, it says that for a finitely generated
graded module 𝑀 over a Noetherian graded ring, the associated primes of 𝑀 are all
graded ideals, and one can write {0} = 𝑁1 ∩ ⋯ ∩ 𝑁𝑚 where 𝑁𝑖 ⊂ 𝑀 are graded
submodules with Ass(𝑀/𝑁𝑖 ) = {𝔭𝑖 }, 𝔭𝑖 ∈ Ass(𝑀), etc. Most of the arguments in the
ungraded case carry over verbatim, and the only new technique is the following
⋅ 46 ⋅ Going-up, going-down, gradings and filtrations

Lemma 4.3.5. Let 𝑀 be a ℤ-graded module over a ℤ-graded ring 𝑅. If 𝑥 ∈ 𝑀 and 𝔭 ∶= ann(𝑥)
is a prime ideal, then

(i) 𝔭 is a homogeneous ideal, and

(ii) 𝔭 = ann(𝑦) for some homogeneous element 𝑦 ∈ 𝑀.

Proof. We begin with (i). Let 𝑡 ∈ 𝔭 and 𝑥 ∈ 𝑀 be such that 𝔭 = ann(𝑀). Write

𝑡 = ∑ 𝑡𝛾 , 𝑥 = ∑ 𝑥𝜂 ,
𝛾∈𝒜 𝜂∈ℬ

where 𝑡𝛾 ∈ 𝑅𝛾 ∖ {0} and 𝑥𝜂 ∈ 𝑀𝜂 ∖ {0} for all 𝛾, 𝜂. Denote by 𝛾0 and 𝜂0 the minimal
elements in 𝒜 and ℬ, respectively. Homogenity amounts to 𝑡𝛾 ∈ 𝔭 for each 𝛾 ∈ 𝒜,
and this will be done by induction on |ℬ|. By a further induction on |𝒜|, for fixed 𝑥 and
𝔭, this can be reduced to showing 𝑡𝛾0 ∈ 𝔭.
First of all, considerations of degrees and 𝑡𝑥 = 0 lead to 𝑡𝛾0 𝑥𝜂0 = 0. If 𝑥 = 𝑥𝜂0 (i.e.
|ℬ| = 1), we obtain 𝑡𝛾0 ∈ ann(𝑥) = 𝔭 as required. In general:

⋄ Suppose that 𝔭 = ann(𝑡𝛾0 𝑥); since the decomposition

𝑡𝛾0 𝑥 = ∑ 𝑡𝛾0 𝑥𝜂
𝜂∈ℬ
𝜂≠𝜂0

involves fewer homogeneous terms, 𝔭 is then homogeneous by our induction hy-

pothesis on |ℬ|.

⋄ Suppose there exists 𝑠 ∈ 𝑅 ∖ 𝔭 such that 𝑠(𝑡𝛾0 𝑥) = 0, then 𝑠𝑡𝛾0 ∈ 𝔭, hence 𝑡𝛾0 ∈ 𝔭.
This concludes the homogeneity (i).

From the homogeneity 𝔭 we infer that 𝔭 ⊂ ann(𝑥𝜂 ) for each 𝜂 ∈ ℬ. Now that

𝔭 = ann(𝑥) ⊃ ∏ ann(𝑥𝜂 ),
𝜂∈ℬ

we have 𝔭 ⊃ ann(𝑥𝜂 ) for some 𝜂, hence 𝔭 = ann(𝑥𝜂 ). Take 𝑦 = 𝑥𝜂 to obtain (ii).

4.4 Filtrations
Now turn to filtrations. We only deal with decreasing filtrations indexed by ℤ≥0 .

Definition 4.4.1. A filtration on a ring 𝑅 is a descending sequence

𝑅 = 𝐹0 𝑅 ⊃ 𝐹1 𝑅 ⊃ 𝐹2 𝑅 ⊃ ⋯

of ideals such that 𝐹𝑖 𝑅 ⋅ 𝐹𝑗 𝑅 ⊂ 𝐹𝑖+𝑗 𝑅. Define the associated ℤ≥0 -graded ring

gr𝐹 (𝑅) ∶= ⨁ 𝐹 𝑛 𝑅/𝐹𝑛+1 𝑅

⎵⎵⎵⎵⎵
𝑛≥0 =∶gr𝑛
𝐹𝑅
 §4.4 Filtrations ⋅ 47 ⋅

whose multiplication is defined as follows: if 𝑥 ∈ 𝐹𝑛 𝑅/𝐹𝑛+1 𝑅 and 𝑦 ∈ 𝐹𝑚 𝑅/𝐹𝑚+1 𝑅,

choose liftings 𝑥 ̃ ∈ 𝐹𝑛 𝑅 and 𝑦 ̃ ∈ 𝐹𝑚 𝑅 and define
𝑥𝑦 ∶= the image of 𝑥𝑦̃ ̃ in 𝐹𝑛+𝑚 𝑅/𝐹𝑛+𝑚+1 𝑅;
this is readily seen to be well-defined. The multiplication of arbitrarily many homo-
geneous elements can be obtained in the same recipe. The datum (𝑅, 𝐹• 𝑅) is called a
filtered ring,
Given a filtered ring 𝑅, a filtered 𝑅-module 𝑀 is an 𝑅-module 𝑀 equipped with a
descending sequence1 of submodules
⋯ ⊃ 𝐹𝑖 𝑀 ⊃ 𝐹𝑖+1 𝑀 ⊃ ⋯ , 𝑖∈ℤ
such that 𝐹𝑖 𝑅 ⋅ 𝐹𝑗 𝑀 ⊂ 𝐹𝑖+𝑗 𝑀. Define the associated graded module as the ℤ-graded
gr𝐹 𝑅-module
𝐹𝑛 𝑀/𝐹𝑛+1 𝑀
gr𝐹 (𝑀) ∶= ⨁ ⎵⎵⎵⎵⎵⎵
𝑛∈ℤ
=∶gr𝑖𝐹 𝑀

whose scalar multiplication is defined using liftings as above.

The subscript 𝐹 in gr will often be omitted. To guarantee that gr𝐹 (𝑅) ≠ {0}, we
usually impose the harmless condition
𝑅 = 𝐹0 𝑅 ⊋ 𝐹1 𝑅.
Example 4.4.2. Let 𝔞 be a proper ideal of 𝑅, then 𝐹𝑖 𝑅 ∶= 𝔞𝑖 defines a filtration on 𝑅,
called the 𝔞-adic filtration.
Definition 4.4.3. Equip 𝑅 with the 𝔞-adic filtration. A filtered 𝑅-module 𝑀 is called
𝔞-stable if 𝔞 ⋅ 𝐹𝑖 𝑀 = 𝐹𝑖+1 𝑀 for 𝑖 ≫ 0.
Recall that 𝔞 ⋅ 𝐹𝑖 𝑀 ⊂ 𝐹𝑖+1 𝑀 holds for all 𝑀, which is a part of our assumption.
As an easy example, set 𝐹𝑖 𝑀 ∶= 𝔞𝑖 𝑀 for 𝑖 ≥ 1, and 𝐹≤0 𝑀 ∶= 𝑀; this is the 𝔞-adic
filtration on 𝑀, which is obviously 𝔞-stable.
Proposition 4.4.4. Suppose 𝔞 is a proper ideal of 𝑅. If 𝑅 is Noetherian, so is gr(𝑅) with respect
to the 𝔞-adic filtration. In fact gr(𝑅) is finitely generated over gr0 (𝑅).
Proof. Let 𝑥1 , … , 𝑥𝑛 be generators of 𝔞, with images 𝑥𝑖̄ ∈ 𝔞/𝔞2 . Then gr(𝑅) is generated
by 𝑥1̄ , … 𝑥𝑛̄ over 𝑅/𝔞 = gr0 𝑅, which is also Noetherian. Now apply Hilbert’s Basissatz.

Proposition 4.4.5. Let 𝔞 be a proper ideal of 𝑅 and 𝑀 a finitely generated 𝑅-module. Suppose
𝑀 is endowed with an 𝔞-stable filtration such that 𝐹𝑖 𝑀 is finitely generated for each 𝑖, and
𝐹≤0 𝑀 = 𝑀. Then gr(𝑀) is a finitely generated gr(𝑅)-module.
Proof. Take 𝑛 such that 𝔞 ⋅ 𝐹𝑖 𝑀 = 𝐹𝑖+1 𝑀 for all 𝑖 ≥ 𝑛. Then in gr(𝑀) = ⨁𝑖 gr𝑖 𝑀 we
have
(𝔞/𝔞2 ) ⋅ gr𝑖 𝑀 = gr𝑖+1 𝑀, 𝑖 ≥ 𝑛.
Therefore it suffices to take generators from gr0 𝑀, … , gr𝑛 𝑀, each of whom is finitely
generated over 𝑅/𝔞 = gr0 𝑅.
1
In view of later applications, the filtration on a module is indexed by ℤ instead of ℤ≥0 .
⋅ 48 ⋅ Going-up, going-down, gradings and filtrations

4.5 Theorems of Artin–Rees and Krull

Conserve the conventions in the previous section on filtrations, etc.

Definition 4.5.1 (Morphisms between filtered objects). Let (𝐴, 𝐹• 𝐴) and (𝐵, 𝐹• 𝐵) be
filtered rings. A morphism between them means a ring homomorphism 𝜑 ∶ 𝐴 → 𝐵
satisfying 𝜑(𝐹𝑖 𝐴) ⊂ 𝐹𝑖 𝐵 for all 𝑖. Similarly, suppose 𝐴 is filtered and let 𝑀, 𝑁 be filtered
𝐴-modules. A morphism 𝑀 → 𝑁 means a homomorphism 𝜓 ∶ 𝑀 → 𝑁 of 𝑅-modules
satisfying 𝜓(𝐹𝑖 𝑀) ⊂ 𝐹𝑖 𝑁 for all 𝑖.

This makes the filtered rings and the filtered modules over a filtered ring into cat-
egories. Obviously, morphisms 𝜑 between filtered objects induce graded morphisms
gr 𝜑 between the associated graded objects. Therefore we obtain a functor from the
category of filtered rings or modules into their graded avatars.
Remark 4.5.2. Suppose 𝜑 ∶ 𝑀 → 𝑁 is a morphism between filtered 𝐴-modules. The
quotient 𝑀/ ker(𝜑) inherits a filtration from 𝑀, whereas the submodule im(𝜑) inherits
one from 𝑁. When the natural isomorphism 𝑀/ ker(𝜑) → im(𝜑) is an isomorphism
between filtered modules, or equivalently

∀𝑖 ∈ ℤ, 𝜑(𝐹𝑖 𝑀) = 𝜑(𝑀) ∩ 𝐹𝑖 𝑁,

we say 𝜑 is a strict morphism.

It is often useful to relate properties of a filtered module or morphism to its graded
counterpart. Propositions 4.4.4 and 4.4.5 are such examples. Here is an example for
the other direction. We say that a filtration on 𝑀 is exhaustive if ⋃𝑖 𝐹𝑖 𝑀 = 𝑀, separating
(or Hausdorff ) if ⋂𝑖 𝐹𝑖 𝑀 = {0}

Proposition 4.5.3. Suppose that 𝜑 ∶ 𝑀 → 𝑁 is a morphism between filtered 𝑅-modules. If 𝑀

is exhaustive and separating, and gr 𝜑 is injective, then 𝜑 is also injective.

Proof. Let 𝑥 ∈ ker(𝜑). There exists 𝑛 such that 𝑥 ∈ 𝐹𝑛 𝑀. Regard 𝑥 + 𝐹𝑛+1 𝑀 as an

element of gr𝑛 𝑀. Then gr(𝜑)(𝑥 + 𝐹𝑛+1 𝑀) = 𝜑(𝑥) + 𝐹𝑛+1 𝑁 = 0, so 𝑥 ∈ 𝐹𝑛+1 𝑀.
Iterating this argument, we have 𝑥 ∈ ⋂𝑘≥𝑛 𝐹𝑘 𝑀 = {0}.

See Lemma 5.2.3 for the case of surjections.

In what follows, 𝔞 always denotes a proper ideal of a ring 𝑅.

Definition 4.5.4 (Blow-up algebras and modules). Introduce an indeterminate 𝑋 and

define the ℤ≥0 -graded 𝑅-algebra

Bl𝔞 𝑅 ∶= ⨁ 𝔞𝑛 𝑋 𝑛 ⊂ 𝑅[𝑋].
𝑛≥0

If an 𝑅-module 𝑀 is endowed with a filtration (𝐹𝑖 𝑀)𝑖≥0 compatible with 𝔞, we define

Bl(𝑀) ∶= ⨁ (𝐹𝑛 𝑀) ⊗ 𝑋 𝑛 ⊂ 𝑀 ⊗ 𝑅[𝑋].

𝑛≥0 𝑅

Clearly, Bl(𝑀) is a graded Bl𝔞 𝑅-module.

 §4.5 Theorems of Artin–Rees and Krull ⋅ 49 ⋅

Remark 4.5.5. The indeterminate 𝑋 is somehow a placeholder. Enlarge Bl𝔞 𝑅 to 𝐵̃ by set-

ting the negative graded pieces to be 𝑅 ⋅ 𝑋 <0 , and set 𝑇 = 𝑋 −1 . We see that 𝐵̃ is actually
an 𝑅[𝑇]-algebra, sometimes called the Rees algebra of 𝔞. Under the specialization 𝑇 = 0
we get
𝐵̃ 𝔞𝑛
≃⨁ ⋅ 𝑇 −𝑛 ≃ gr(𝑅).
𝑇 𝐵̃ 𝑛≥0 𝔞𝑛+1
On the other hand, inverting 𝑇 yields

̃ −1 ] = ⨁ 𝑅 ⋅ 𝑇 −𝑛 = 𝑅[𝑇 ±1 ].
𝐵[𝑇
𝑛∈ℤ

This reflects a well-known deformation construction in geometry; gr(𝑅) is actually the

graded 𝑅-algebra corresponding to the normal cone defined by 𝔞 ⊂ 𝑅. We refer to [9,
§5.1] for details.

Lemma 4.5.6. Consider a ring 𝑅 with proper ideal 𝔞, together with a filtered 𝑅-module 𝑀,
assume furthermore that each 𝐹𝑖 𝑀 is finitely generated over 𝑅. The following are equivalent:

(i) Bl(𝑀) is finitely generated over Bl𝔞 𝑅;

(ii) the filtration on 𝑀 is 𝔞-stable. (Definition 4.4.3)

Proof. (i) ⟹ (ii): Choose homogeneous generators 𝑥1 , … , 𝑥𝑛 ∈ Bl(𝑀) with degrees

𝑑1 , … , 𝑑𝑛 respectively. It is then routine to see that

𝑖 ≥ max{𝑑1 , … , 𝑑𝑛 } ⟹ (𝐹𝑖+1 𝑀)𝑋 𝑖+1 = 𝔞𝑋 ⋅ (𝐹𝑖 𝑀)𝑋 𝑖 ,

that is, 𝔞 ⋅ 𝐹𝑖 𝑀 = 𝐹𝑖+1 𝑀 for these 𝑖.

(ii) ⟹ (i): Suppose 𝔞 ⋅ 𝐹𝑖 𝑀 = 𝐹𝑖+1 𝑀 for 𝑖 ≥ 𝑑, then Bl(𝑀) is generated by
⨁𝑗≤𝑑 (𝐹𝑗 𝑀)𝑋 𝑗 , and each 𝐹𝑗 𝑀 is finitely generated over 𝑅 = (Bl𝔞 𝑅)0 .

Theorem 4.5.7 (Artin–Rees). Let 𝑅 be a Noetherian ring endowed with 𝔞-adic filtration. Let
𝑀 be a finitely generated 𝑅-module and 𝑁 ⊂ 𝑀 an 𝑅-submodule. Then the filtration on 𝑁
induced by the 𝔞-adic filtration of 𝑀, namely 𝐹𝑖 𝑁 ∶= 𝔞𝑖 𝑀 ∩ 𝑁, is 𝔞-stable.

Proof. Define Bl(𝑁) using the induced filtration 𝐹𝑖 𝑁 ∶= 𝔞𝑖 𝑀 ∩ 𝑁, which is a submod-

ule of the finitely generated Bl𝔞 𝑅-module Bl(𝑀) (Lemma 4.5.6). If 𝔞 = (𝑎1 , … , 𝑎𝑚 ) than
Bl𝔞 𝑅 = 𝑅[𝑎1 𝑋, … , 𝑎𝑚 𝑋] ⊂ 𝑅[𝑋], hence Noetherian by Hilbert’s Basissatz. We deduce
that Bl(𝑁) is finitely generated over Bl𝔞 𝑅. In turn, this implies 𝐹• 𝑁 is an 𝔞-stable filtra-
tion on 𝑁 by Lemma 4.5.6.

Theorem 4.5.8. For 𝑅, 𝔞, 𝑀 as in the previous theorem, we set 𝑁 ∶= ⋂𝑛≥0 𝔞𝑛 𝑀. Then 𝔞𝑁 =

𝑁.

Proof. Since the induced filtration on 𝑁 is 𝔞-stable by Theorem 4.5.7, for 𝑛 ≫ 0 we have

𝑁 = 𝔞𝑛 𝑀 ∩ 𝑁 = 𝔞 ⋅ (𝔞𝑛−1 𝑀 ∩ 𝑁) = 𝔞𝑁.

The assertion follows.

⋅ 50 ⋅ Going-up, going-down, gradings and filtrations

Corollary 4.5.9 (Krull). If 𝔞 ⊂ rad(𝑅), then ⋂𝑛≥0 𝔞𝑛 𝑀 = {0} for any finitely generated
𝑅-module 𝑀. In particular ⋂𝑛≥0 𝔞𝑛 = {0} whenever 𝔞 ⊂ rad(𝑅).

Proof. Theorem 4.5.8 together with Nakayama’s lemma imply 𝑁 = {0}.

Corollary 4.5.10 (Krull’s Intersection Theorem). Let 𝑅 be a Noetherian domain and 𝔞 a

proper ideal. Then ⋂𝑛≥0 𝔞𝑛 = {0}.

Proof. Define 𝑁 ∶= ⋂𝑛≥0 𝔞𝑛 ⊂ 𝑅. By Theorem 4.5.8 we have 𝔞𝑁 = 𝑁, thus there exists

𝑟 ∈ 𝔞 with 1 + 𝑟 ∈ ann(𝑁) by Nakayama’s Lemma (Theorem 1.3.5). As 𝔞 is proper,
1 + 𝑟 cannot be zero. Since 𝑅 is a domain containing 𝑁, the only possibility is 𝑁 = {0}
as asserted.
 Lecture 5

From completions to
dimensions
The main references are [11, 8].

5.1 Completions
Consider a ring 𝑅 together with a family of ideals ℐ ≠ ∅, such that for any 𝐼, 𝐽 ∈ ℐ
there exists 𝐾 ∈ ℐ with 𝐾 ⊂ 𝐼 ∩ 𝐽. This turns 𝑅 into a topological ring, characterized by
the property that ℐ forms a local base of open neighborhoods of 0. Recall that being a
topological ring means that addition, multiplication and 𝑥 ↦ −𝑥 are all continuous. By
standard arguments, 𝑅 is Hausdorff if and only if {0} is closed, if and only if ⋂𝐼∈ℐ 𝐼 =
{0}.
To simplify matters, we assume that
⋄ the family ℐ is countable, so that the topological properties (accumulation points,
etc.) are detected by convergence of sequences as in the case of metric spaces;
⋄ furthermore, we may arrange that ℐ = {𝐼 ⊃ 𝐽 ⊃ 𝐾 ⊃ ⋯}, in other words our
topology comes from filtrations.
Without the countability assumption, the sequences will have to be replaced by filters.
It makes sense to talk about topological 𝑅-modules for a topological ring 𝑅. By re-
placing filtration by ideals by filtration by 𝑅-submodules subject to the usual compati-
bility relation 𝐹𝑖 𝑅 ⋅ 𝐹𝑗 𝑀 ⊂ 𝐹𝑖+𝑗 𝑀, the recipe above applies to 𝑅-modules as well. Given
𝑁 ⊂ 𝑀, the topology so obtained on 𝑀 passes to 𝑀/𝑁 by taking the quotient topol-
ogy, or equivalently the quotient filtration (𝐹• 𝑀 + 𝑁)/𝑁. If the filtration in question is
𝐼-adic, where 𝐼 ⊊ 𝑅 is an ideal, we obtain the 𝐼-adic topology on rings and modules.
An 𝑅-module 𝑀 equipped with a topology as above is complete if every Cauchy
sequence (𝑥𝑛 )𝑛≥1 has a limit; a Cauchy sequence (𝑥𝑛 )𝑛≥1 is a sequence satisfying

∀𝐼 ∈ ℐ, ∃𝑁 𝑖, 𝑗 ≥ 𝑁 ⟹ 𝑥𝑖 − 𝑥𝑗 ∈ 𝐼.

As in the familiar case of metric spaces, one has the completion of 𝑀. It is actually a mor-
phism 𝑀 → 𝑀̂ with 𝑀̂ complete Hausdorff, characterized by the following universal
⋅ 52 ⋅ From completions to dimensions

property:
𝜑∶ cont. homo.
𝑀 𝑀̂
𝑀 𝐿
⇝ ∃!𝜑̂
complete Hausdorff 𝜑
𝐿
The uniqueness results immediately, and the formation of 𝑀 ↦ 𝑀̂ is seen to be
functorial in 𝑀. If 𝑀 is already complete Hausdorff, one may take 𝑀̂ = 𝑀. Certainly,
the same applies to the ring 𝑅.
Exercise 5.1.1. Suppose that 𝑅 is complete Hausdorff with respect to the 𝐼-adic topol-
ogy, where 𝐼 is a proper ideal. Show that every element of the form 𝑢 + 𝑥, 𝑢 ∈ 𝑅× and
𝑥 ∈ 𝐼, is invertible.
From the algebraic perspective, the completion of a filtered 𝑅-module 𝑀 = 𝐹0 𝑀 ⊃
𝐹1 𝑀 ⊃ ⋯ can be constructed as the projective limit
𝑀̂ ∶= lim 𝑀/𝐹𝑖 𝑀
←−−
𝑖≥1

= {(𝑥𝑖 )𝑖≥1 ∶ 𝑖 ≤ 𝑗 ⟹ 𝑥𝑖 ≡ 𝑥𝑗 (mod 𝐹𝑖 𝑀)} ⊂ ∏ 𝑀/𝐹𝑖 𝑀.

𝑖≥1

The morphism 𝑀 → 𝑀̂ is the diagonal map. The topology on 𝑀̂ arises from the filtra-
tion
𝐹𝑖 𝑀̂ ∶= ker [𝑝𝑖 ∶ 𝑀̂ → 𝑀/𝐹𝑖 𝑀] = {(𝑥𝑛 )𝑛 ∈ 𝑀̂ ∶ 𝑖 ≤ 𝑘 ⟹ 𝑥𝑖 = 0} ,
so that the preimage of 𝐹𝑖 𝑀̂ in 𝑀 is precisely 𝐹𝑖 𝑀. In the case where 𝑀 = 𝑅 and 𝐹𝑖 𝑅
are ideals, we obtain the complete Hausdorff ring 𝑅,̂ which is a subring of ∏𝑖≥1 𝑅/𝐹𝑖 𝑅.
Since the filtrations on 𝑅 and 𝑀 are assumed compatible, 𝑀̂ is an 𝑅-module.
̂
Example 5.1.2. Fix a prime number 𝑝. The completion of ℤ with respect to the ideal
𝑝ℤ is nothing but the ring ℤ𝑝 of 𝑝-adic integers. Similarly, the completion of 𝕜[𝑋] with
respect to (𝑋) is isomorphic to the 𝕜-algebra 𝕜J𝑋K.
Exercise 5.1.3. Describe the kernel of 𝑀 → 𝑀̂ and show 𝑀 ↪ 𝑀̂ if and only if 𝑀 is
Hausdorff.
Exercise 5.1.4. Show that the topology of 𝑀̂ is the restriction of the product topology of
∏𝑖 𝑀/𝐹𝑖 𝑀, provided that each 𝑀/𝐹𝑖 𝑀 is endowed with the discrete topology. Show
that 𝑀̂ is a closed subspace of ∏𝑖 𝑀/𝐹𝑖 𝑀
Lemma 5.1.5. Let 𝑀 be a complete 𝑅-module with respect to some filtration 𝐹• 𝑀. For any
submodule 𝑁, the quotient 𝑀/𝑁 is also complete with respect to the quotient topology, or equiv-
alently with respect to the quotient filtration (𝐹• 𝑀 + 𝑁)/𝑁.
Proof. Let 𝑥𝑛̄ be a Cauchy sequence in 𝑀/𝑁. Choose preimages 𝑀 ∋ 𝑥𝑛 ↦ 𝑥𝑛̄ for all
𝑛. We have 𝑥𝑛+1̄ − 𝑥𝑛̄ ∈ 𝐹𝑖(𝑛) 𝑀 + 𝑁 where lim𝑛→∞ 𝑖(𝑛) = ∞, therefore we can write
𝑥𝑛+1 − 𝑥𝑛 = 𝑦𝑛 + 𝛿𝑛 where 𝑦𝑛 ∈ 𝐹𝑖(𝑛) 𝑀 and 𝛿𝑛 ∈ 𝑁. We contend that 𝑥𝑛′ ∶= 𝑥1 + ∑𝑖<𝑛 𝑦𝑖
is a Cauchy sequence in 𝑀. Indeed, for any 𝑖 > 𝑗 we have 𝑥𝑖′ − 𝑥𝑗′ = ∑𝑗≤𝑘<𝑖 𝑦𝑘 , which
lies in 𝐹inf𝑘 𝑖(𝑘) 𝑀. This implies (𝑥𝑛′ )𝑛 is a Cauchy sequence, hence has a limit 𝑥 ∈ 𝑀. It
is also clear that 𝑥𝑛′ ↦ 𝑥𝑛̄ . Hence 𝑥𝑛′̄ has a limit 𝑥 ̄ = 𝑥 mod 𝑁 in 𝑀/𝑁.
 §5.1 Completions ⋅ 53 ⋅

Let us turn to the exactness of completion. This should be understood in the broader
framework of lim of arbitrary projective systems. For simplicity, we only consider 𝐼-adic
←−−
topologies on finitely generated modules over a Noetherian ring.
Observe that any homomorphism 𝜑 ∶ 𝑀 → 𝑁 between 𝑅-modules is automatically
𝔞-adically continuous, for that 𝜑(𝔞𝑛 𝑀) ⊂ 𝔞𝑛 𝑁.

Proposition 5.1.6. Let 𝑅 be a Noetherian ring and 𝐼 ⊊ 𝑅 an ideal. Suppose 0 → 𝑀′ → 𝑀 →

𝑀″ → 0 is an exact sequence of finitely generated 𝑅-modules, each term equipped with the 𝐼-adic
topology. The completed sequence 0 → 𝑀̂ ′ → 𝑀̂ → 𝑀″̂ → 0 is also exact.

Consequently, completion preserves the exactness of sequences formed by finitely

generated 𝑅-modules. This is what makes completion so useful.
Proof. We shall show
(i) the topology on 𝑀′ induced from 𝑀 is the same as the 𝐼-adic topology;
(ii) the quotient topology on 𝑀″ = 𝑀/𝑀′ is 𝐼-adic;
(iii) the completion 𝑀̂ ′ is naturally identified with the closure of the image of 𝑀′ in
𝑀;̂
(iv) 𝑀̂ ″ is naturally identified with the quotient of 𝑀̂ by 𝑀̂ ′ .
(i) is a direct consequence of Artin–Rees Theorem 4.5.7: for 𝑛 ≫ 0 we have

𝐼 𝑛 𝑀′ ⊂ 𝑀′ ∩ 𝐼 𝑛 𝑀 = 𝐼(𝑀′ ∩ 𝐼 𝑛−1 𝑀) ⊂ 𝐼 𝑛−1 𝑀′ , (5–1)

and this suffices to identify the resulting topologies.

(ii) is immediate. As for (iii), we may embed 𝑀′ into 𝑀 and work with the induced
topology. Realize 𝑀̂ as lim 𝑀/𝐼 𝑛 𝑀. As a topological module 𝑀̂ ′ equals
←−−𝑛
𝑀′
lim ′ ̂ ′
←−− 𝑀 ∩ 𝐼 𝑛 𝑀 = {𝑥 ̂ = (𝑥𝑛 )𝑛 ∈ 𝑀 ∶ ∀𝑘, 𝑥𝑘 comes from 𝑀 }
𝑛
= {𝑥 ̂ ∈ 𝑀̂ ∶ ∀𝑘 ∃𝑦 ∈ 𝑀′ s.t. 𝑥 ̂ ∈ 𝑦 + 𝐹𝑘 𝑀}
̂

which is readily seen to be the closure of the image of 𝑀′ .

For (iv), the quotient 𝑀/ ̂ 𝑀̂ ′ is Hausdorff since 𝑀̂ ′ is a closed submodule. It is also
complete by Lemma 5.1.5. There is a natural homomorphism 𝑀/𝑀′ → 𝑀/ ̂ 𝑀̂ ′ . Given
any continuous homomorphism 𝜑 ∶ 𝑀/𝑀′ → 𝑁 to a complete Hausdorff 𝑅-module
𝑁, we may pull it back to 𝑀 → 𝑁, which corresponds to a unique continuous 𝑀̂ → 𝑁
that is trivial on the image of 𝑀′ ; but such a homomorphism must also vanish on the
closure 𝑀̂ ′ . This yields the required 𝜑̂ ∶ 𝑀/̂ 𝑀̂ ′ → 𝑁 in the universal property.

For any 𝑅-module 𝑀 endowed with 𝐼-adic topology, there is a canonical homomor-
phism 𝑀 ⊗ 𝑅̂ → 𝑀.̂ Indeed, 𝑀̂ = lim 𝑀/𝐼 𝑛 𝑀 is a 𝑅̂ = lim 𝑅/𝐼 𝑛 -module by
𝑅 ←−−𝑛 ←−−𝑛

(𝑟𝑛 )𝑛≥1 ⋅ (𝑥𝑛 )𝑛≥1 = (𝑟𝑛 𝑥𝑛 )𝑛≥1 , (𝑟𝑛 )𝑛 ∈ 𝑅,̂ (𝑥𝑛 )𝑛 ∈ 𝑀,̂

hence the 𝑅-homomorphism 𝑀 → 𝑀̂ gives rise to 𝑀⊗ 𝑅̂ → 𝑀̂ by the universal property

𝑅
of base change. This homomorphism is continuous if 𝑀 ⊗ 𝑅̂ is filtered by the images
𝑅
̂ which makes it into a topological 𝑅-module.
of 𝑀 ⊗ (𝐹• 𝑅), ̂
𝑅
⋅ 54 ⋅ From completions to dimensions

Theorem 5.1.7. Suppose 𝑅 is Noetherian. Let 𝑀 be a finitely generated 𝑅-module endowed

with the 𝐼-adic topology. The homomorphism 𝑀 ⊗ 𝑅̂ → 𝑀̂ is then an isomorphism.
𝑅

Proof. Write down a finite presentation 𝑅⊕𝑎 → 𝑅⊕𝑏 → 𝑀 → 0. By the naturality of

the homomorphism above, the right-exactness of ⊗ and Proposition 5.1.6, we have a
commutative diagram

𝑅⊕𝑎 ⊗ 𝑅̂ 𝑅⊕𝑎 ⊗ 𝑅̂ 𝑀 ⊗ 𝑅̂ 0
𝑅 𝑅 𝑅

̂
𝑅⊕𝑎 ̂
𝑅⊕𝑏 𝑀̂ 0

with exact rows. Completion commutes with direct sums: (𝑁1 ⊕ 𝑁2 )∧ = 𝑁̂ 1 ⊕ 𝑁̂ 2

canonically (easy, and the categorical reason is that completion is left adjoint to oblivion
𝑅-CompHausMod → 𝑅-TopMod). Therefore the first two vertical arrows may be identified
with the canonical arrow 𝑅⊕⋆ ⊗ 𝑅̂ → 𝑅̂ ⊕⋆ , ⋆ ∈ {𝑎, 𝑏}, which is an isomorphism of
𝑅
̂ ∼
topological 𝑅-modules. We infer that 𝑀 ⊗ 𝑅̂ → 𝑀̂ topologically.
𝑅

∼
Remark 5.1.8. In fact 𝑀 ⊗ 𝑅̂ → 𝑀̂ is also a homeomorphism. It suffices to observe that
𝑅
in the rows of the commutative diagram above, 𝑀 ⊗ 𝑅̂ and 𝑀̂ are both realized as
𝑅
̂ ∼
quotient topological 𝑅-modules, and that 𝑅⊕⋆ ⊗ 𝑅̂ → 𝑅̂ ⊕⋆ is a homeomorphism. The
𝑅
second point has been observed in the proof of Proposition 5.1.6, and the first follows
̂
from the fact completed modules carry the 𝐼-adic topology. See Proposition 5.2.2. We
do not need this result.
In the following statements, 𝑅 is Noetherian and an ideal 𝐼 ⊊ 𝑅 is chosen.

Corollary 5.1.9. The canonical homomorphism 𝑅 → 𝑅̂ is flat.

Proof. Flatness can be tested on short exact sequences of the form 0 → 𝔞 → 𝑅 → 𝑅/𝔞 →
0 where 𝔞 is a finitely generated ideal of 𝑅. Its base-change to 𝑅̂ is the same as comple-
tion, and completion is an exact functor by Proposition 5.1.6.

Corollary 5.1.10. Assume 𝑅 is 𝐼-adically complete Hausdorff. Then every finitely generated
𝑅-module 𝑀 is 𝐼-adically complete Hausdorff, and any submodule 𝑁 ⊂ 𝑀 is closed.

Proof. For the first assertion: the completion of 𝑀 can be identified with the composite
𝑀 = 𝑀 ⊗ 𝑅 → 𝑀 ⊗ 𝑅̂ → 𝑀,̂ which is bijective. Therefore every Cauchy sequence in 𝑀
𝑅 𝑅
has a limit in 𝑀.
As to the second assertion, recall that the 𝐼-adic topology on 𝑁 is the same as the
one restricted from 𝑀 by (5–1). It remains to notice that complete subspaces must be
closed.
 §5.2 Further properties of completion ⋅ 55 ⋅

5.2 Further properties of completion

Let 𝑅 be a Noetherian ring and 𝑀 a finitely generated 𝑅-module. Fix an ideal 𝐼 ⊊ 𝑅.
Unless otherwise specified, the topologies and completions are always 𝐼-adic.

Proposition 5.2.1. Let 𝔞 be an ideal, then 𝔞𝑀̂ = 𝔞𝑀

̂ = 𝔞𝑀
̂ ̂ as submodules of 𝑀.̂ Consequently
̂ 𝑀̂ ≃ (𝑀/𝔞𝑀) canonically.
𝑀/𝔞 ∧

Proof. By the exactness of completion (Proposition 5.1.6), we may realize 𝔞 ̂ as an ideal

of 𝑅;̂ in fact it is the image 𝔞𝑅̂ of 𝔞 ⊗ 𝑅̂ → 𝑅 ⊗ 𝑅̂ = 𝑅.̂ Hence 𝔞𝑅̂ = 𝔞.̂ Now consider the
𝑅 𝑅
commutative diagram
𝔞 ⊗ 𝑀 ⊗ 𝑅̂ 𝔞 ⊗ 𝑀̂
𝑅 𝑅 𝑅

𝔞𝑀 ⊗ 𝑅̂ 𝑀̂
𝑅

The upper horizontal arrow is an isomorphism by Theorem 5.1.7, therefore the diago-
nal arrow has image equal to 𝔞𝑀.̂ The lower horizontal arrow is just the completion of
𝔞𝑀 ↪ 𝑀, thus injective with image (𝔞𝑀)∧ by Proposition 5.1.6. A comparison yields
(𝔞𝑀)∧ = 𝔞𝑀.̂ Also note that 𝔞𝑀 ̂ ̂ = 𝔞𝑅̂ 𝑀̂ = 𝔞𝑀.̂ The final assertion results from the
exactness of completion.

Since the 𝐼-adic topology on 𝑀/𝐼 𝑛 𝑀 is discrete, as a special case (𝔞 = 𝐼 𝑛 ) we deduce

the natural identifications

̂ 𝑛 𝑀̂ = 𝑀/
𝑀/𝐼 𝑛 𝑀 = 𝑀/𝐼 ̂ 𝐼 𝑛̂ 𝑀,̂ ∀𝑛 ≥ 0,
̂ = gr (̂ 𝑅),
gr (𝑅) = gr (𝑅) ̂
𝐼 𝐼 𝐼
̂ = gr (̂ 𝑀).
gr𝐼 (𝑀) = gr𝐼 (𝑀) ̂
𝐼

Proposition 5.2.2. For any finitely generated 𝑅-module 𝑀, the topology on 𝑀̂ coincides with
̂
the 𝐼-adic one.

Proof. Consider the closure of the image of 𝐼 𝑛 𝑀 in 𝑀.̂ It is readily seen to be {(𝑥𝑘 )𝑘 ∶ 𝑖 ≤
𝑛 ⟹ 𝑥𝑖 = 0} = 𝐹𝑛 𝑀.̂ On the other hand, we have seen that this closure is 𝐼̂ 𝑛 𝑀 ⊂ 𝑀. ̂
By virtue of Proposition 5.2.1, we have 𝐼̂
𝑛 𝑀 = 𝐼̂𝑀̂ and 𝐼̂ = 𝐼 𝑅̂ = (𝐼 𝑅)
𝑛 𝑛 𝑛 ̂ = 𝐼̂ .
𝑛 𝑛

Lemma 5.2.3. Consider a homomorphism 𝜑 ∶ 𝐿 → 𝑁 between filtered modules over some

ring, such that 𝐿 is complete, 𝑁 is Hausdorff and exhaustive (see §4.5) with respect to their
filtrations, and gr(𝜑) ∶ gr(𝐿) → gr(𝑁) is surjective. Then 𝜑 is also surjective.

Proof. Let 𝑦 ∈ 𝐹𝑑 𝑁, we may take 𝑥 ∈ 𝐹𝑑 𝐿 such that 𝑦′ ∶= 𝑦 − 𝜑(𝑥) ∈ 𝐹𝑑+1 𝑁. Next, take
𝑥′ ∈ 𝐹𝑑+1 𝐿 with 𝑦″ ∶= 𝑦′ − 𝜑(𝑥′ ) ∈ 𝐹𝑑+2 𝑁, and so forth. Use the completeness of 𝐿 to
define 𝑥∞ ∶= 𝑥 + 𝑥′ + 𝑥″ + ⋯, which maps to 𝑦 since 𝑁 is Hausdorff.

Proposition 5.2.4. The ring 𝑅̂ is also Noetherian, and 𝐼 ̂ ⊂ rad(𝑅).

̂
⋅ 56 ⋅ From completions to dimensions

Proof. Let 𝔄 be any ideal of 𝑅,̂ equipped with the filtration 𝐹𝑛 𝔄 ∶= 𝐼 𝑛̂ ∩ 𝔄. We have
to show 𝔄 is finitely generated. Since gr𝐼 (𝑅) = gr𝐼 (̂ 𝑅) ̂ is Noetherian, so is gr (𝔄).
𝐹
𝑑
Take 𝑡1 , … , 𝑡𝑛 ∈ 𝔄, 𝑡𝑖 ∈ 𝐹𝑑𝑖 𝔄, whose images 𝑡𝑖̄ in gr𝐹𝑖 (𝔄) generates gr𝐹 (𝔄). Using
an appropriately shifted filtration on 𝐿 ∶= 𝑅̂ ⊕𝑛 , we obtain a filtered homomorphism
𝜑 ∶ 𝐿 → 𝔄 with image (𝑡1 , … , 𝑡𝑛 ), such that gr(𝜑) is surjective. Now apply the previous
Lemma to obtain the first assertion.
One of the characterizations of Jacobson radical says that 𝐼 ̂ ⊂ rad(𝑅) ̂ if and only if
̂ ̂ × −1 2
1 − 𝐼 ⊂ 𝑅 . This is verified by noting that (1 − 𝑡) = 1 + 𝑡 + 𝑡 + ⋯ converges 𝐼-adically.
This proves the second assertion.
̂ satisfying
Proposition 5.2.5. The map 𝔭 ↦ 𝔭̂ furnishes an injection from 𝑉(𝐼) to Spec(𝑅)
1∶1
̂ 𝔭̂ (as rings). It restricts to a bijection MaxSpec(𝑅) ∩ 𝑉(𝐼)
𝑅/𝔭 ≃ 𝑅/ ̂
MaxSpec(𝑅).
Consequently, if 𝑅 is local (resp. semi-local), so is 𝑅.̂

Proof. Since 𝔭 ⊃ 𝐼, the 𝐼-adic topology on 𝑅/𝔭 is discrete. By Proposition 5.2.1, 𝑅/ ̂ 𝑝̂ ≃

∧
(𝑅/𝔭) = 𝑅/𝔭, and here the isomorphism even respects ring structures. Therefore 𝔭̂ is
a prime ideal. Moreover, it is maximal if and only if 𝔭 is. Claim: 𝔭 is the preimage of
𝔭̂ = lim 𝔭/(𝐼 𝑛 ∩ 𝔭) under 𝑅 → 𝑅̂ = lim 𝑅/𝐼 𝑛 . Indeed, lying in that preimage amounts
←−−𝑛 ←−−𝑛
to 𝑥 ∈ 𝐼 𝑛 + 𝔭 = 𝔭, for all 𝑛 ≥ 1. Injectivity follows.
Lemma 5.2.4 implies that every 𝔄 ∈ MaxSpec(𝑅) ̂ contains 𝐼,̂ therefore is open by
Proposition 5.2.2. Since 𝑅 is Noetherian, 𝔄 is also closed by Corollary 5.1.10. As 𝑅 → 𝑅̂
̂
has dense image, we conclude that 𝔄 equals the completion of its preimage 𝔪 ∈ 𝑉(𝐼) ⊂
Spec(𝑅). By the previous paragraph, 𝔪 is a maximal ideal.

5.3 Hilbert–Samuel polynomials

For a graded ring 𝑅 = ⨁𝛾 𝑅𝛾 (we always assume 1 ∈ 𝑅0 ) and a given 𝜂, we may define
its twist 𝑅(𝜂) as the graded 𝑅-module

𝑅(𝜂)𝛾 ∶= 𝑅𝛾+𝜂 .

To generate a graded 𝑅-module 𝑀 by finitely many homogeneous elements 𝑥1 , … , 𝑥𝑛 ,

of degrees 𝜂1 , … , 𝜂𝑛 respectively, is equivalent to giving a surjection of graded 𝑅-modules

𝑛
⨁𝑖=1 𝑅(−𝜂𝑖 ) 𝑀
(5–2)
(… , 0, 1⎵ , 0, …) 𝑥𝑖 .
𝑖−th slot

Hereafter, we assume that

⋄ everything is graded by 𝛤 = (ℤ𝑁

≥0 , +) for some fixed 𝑁,

⋄ 𝑅0 /𝑅0 ∩ ann(𝑀) is an Artinian ring,

⋄ 𝑅 is finitely generated over 𝑅0 .

 §5.3 Hilbert–Samuel polynomials ⋅ 57 ⋅

The appearance of ann(𝑀) is harmless since 𝑅 can be safely replaced by 𝑅/𝔞𝔫𝔫(𝑀),

which is legitimate since ann(𝑀) is a graded ideal of 𝑅. To see this, write ann(𝑀) as
the intersection of ann(𝑥) where 𝑥 ranges over the homogeneous elements of 𝑀, and
observe that ann(𝑥) must be graded.

Lemma 5.3.1. For 𝑅 as above and 𝑀 a finitely generated graded 𝑅-module, each graded piece
𝑀𝛾 is an 𝑅0 -module of finite length.

Proof. Using (5–2) this is readily reduced to the case 𝑀 = 𝑅(−𝜂), and then to 𝑀 = 𝑅.
Write 𝑅 = 𝑅0 [𝑥1 , … , 𝑥𝑛 ] where each 𝑥𝑖 is homogeneous of degree 𝑑𝑖 . Given 𝛾, the 𝑅0 -
𝑎 𝑎
module 𝑀𝛾 is generated by monomials 𝑥11 ⋯ 𝑥𝑛𝑛 with ∑𝑖 𝑎𝑖 𝑑𝑖 = 𝛾 and 𝑎1 , … , 𝑎𝑛 ∈ ℤ≥0 ;
this admits only finitely many solutions (𝑎1 , … , 𝑎𝑛 ). We conclude that 𝑀𝛾 has finite
length since 𝑅0 /𝑅0 ∩ ann(𝑀) is an Artinian ring.

Recall that saying a module 𝑁 over a ring 𝐴 has finite length means that there exists
a composition series

𝑁 = 𝑁0 ⊃ ⋯ ⊃ 𝑁𝑛 = {0}, ∀𝑁𝑖 /𝑁𝑖+1 is simple.

The unique number (Jordan–Hölder Theorem) is called the length of 𝑁, denoted by

ℓ𝐴 (𝑁). The length function is additive in short exact sequences. When 𝐴 is a field we
have ℓ𝐴 = dim𝐴 .

Definition 5.3.2. For 𝑅 and 𝑀 as in Lemma 5.3.1, we define the functions

𝜒(𝑀, 𝛾) ∶= ℓ𝑅0 (𝑀𝛾 ), 𝛾 ∈ 𝛤 ∶= ℤ𝑁

≥0

with values in ℤ≥0 .

One sees immediately that for a short exact sequence 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 of

finitely generated graded 𝑅-modules, we have 𝜒(𝑀, 𝛾) = 𝜒(𝑀′ , 𝛾) + 𝜒(𝑀″ , 𝛾) for all
𝛾 ∈ 𝛤. This extends to alternating sums of 𝜒 in finite exact sequences.
One can control the behavior of 𝜒(𝑀, ⋅) by forming the Poincaré series 𝑃𝑀 (X) =
∑𝛾∈𝛤 𝜒(𝑀, 𝛾)X𝛾 ; see [7, 6.D]. Here we shall restrict to the case 𝑁 = 1, i.e. 𝛤 = ℤ≥0 , in
order to gain more control of 𝜒(𝑀, 𝛾). As a preparation, we say a function 𝐻 ∶ ℤ → ℂ
is a quasi-polynomial of period 𝜛 if its restriction to each congruence class modulo 𝜛
coincides with a polynomial function (necessarily unique); the degree of 𝜒 is defined
by taken the maximum among congruence classes. In particular, a quasi-polynomial
of period 1 is just a polynomial.

Theorem 5.3.3. Assume 𝛤 = ℤ≥0 . Suppose 𝑅 is generated by homogeneous elements 𝑥1 , … , 𝑥𝑛

over 𝑅0 . There exists a unique quasi-polynomial 𝐻𝑀 of degree ≤ 𝑛 − 1, with coefficients in ℚ
and period 𝑒 ∶= lcm(deg 𝑥1 , … , deg 𝑥𝑛 ), such that

𝜒(𝑀, 𝛾) = 𝐻𝑀 (𝛾), |𝛾| ≫ 0.

Proof. Uniqueness is clear. We construct 𝐻𝑀 by induction on the minimal number of

generators 𝑛. If 𝑛 = 0 then 𝑅 = 𝑅0 and 𝑀𝛾 = 0 for |𝛾| ≫ 0, in which case 𝐻𝑀 = 0.
⋅ 58 ⋅ From completions to dimensions

For 𝑛 ≥ 1, write 𝑅 = 𝑅0 [𝑥1 , … , 𝑥𝑛 ] as usual. We may assume that 𝑥𝑖 ≠ 0 has degree

𝜂𝑖 , for 𝑖 = 1, … , 𝑛. Fix 𝑖 and define the graded modules

⋅𝑥𝑖 ⋅𝑥𝑖
𝑍 ∶= ker (𝑀 𝑀(𝜂𝑖 )) , 𝑌 ∶= coker (𝑀(−𝜂𝑖 ) 𝑀)

which are again finitely generated, so that we have the exact sequence

0 → 𝑍𝛾 → 𝑀𝛾 → 𝑀𝛾+𝜂𝑖 → 𝑌𝛾+𝜂𝑖 → 0, 𝛾 ∈ 𝛤.

Since 𝑥𝑖 annihilates 𝑍 and 𝑌, induction hypothesis entails

𝜒(𝑀, 𝛾 + 𝜂𝑖 ) − 𝜒(𝑀, 𝛾) = 𝜒(𝑌, 𝛾 + 𝜂𝑖 ) − 𝜒(𝑍, 𝛾),

the right-hand side being quasi-polynomials of period lcm(… , 𝜂̂𝑖 , …) for large |𝛾| and
of degrees ≤ 𝑛−2, since 𝑥𝑖 acts trivially on 𝑍 and 𝑌. Doing this for all 𝑖 yields difference
equations that witness the polynomiality of 𝜒(𝑀, 𝛾) for |𝛾| ≫ 0 in every congruence
class modulo 𝑒.
In particular, if 𝑅 is generated by 𝑅1 over 𝑅0 , the period 𝑒 = 1 and we have the
notion of Hilbert–Samuel polynomials.

Example 5.3.4. For 𝑅 = 𝑀 = 𝕜[𝑋1 , … , 𝑋𝑛 ] graded by total degree, where 𝕜 is a field,

our assumptions are readily verified. We see 𝜒(𝑀, 𝛾) = dim𝕜 𝕜[𝑋1 , … , 𝑋𝑛 ]deg=𝛾 for
all 𝛾 ∈ ℤ≥0 , which equals (𝛾+𝑛−1
𝑛−1 ) by high school combinatorics. Hence the Hilbert–
Samuel polynomial is 𝐻𝑀 (𝑋) = (𝑋+𝑛−1
𝑛−1 ) ∈ ℚ[𝑋].

5.4 Definition of Krull dimension

Let 𝑅 be a ring.

Definition 5.4.1 (Height and dimension). For any prime ideal 𝔭 of 𝑅, define its height
ht(𝔭) as the supremum of the lengths of prime chains

𝔭 = 𝔭 0 ⊋ 𝔭 1 ⊋ ⋯ ⊋ 𝔭𝑛 , length ∶= 𝑛.

For any ideal 𝔞 of 𝑅, we define ht(𝔞) ∶= inf{ht(𝔭) ∶ 𝔭 ⊃ 𝔞}.

Define the Krull dimension of 𝑅 to be dim 𝑅 ∶= sup𝔭∈Spec(𝑅) ht(𝔭).

The following results are immediate.

⋄ The zero prime in a domain has height 0.

⋄ Fields have dimension zero. In fact, a ring has dimension zero if and only if every
prime ideal is maximal.

⋄ We have ht(𝔭) = dim 𝑅𝔭 for every prime ideal 𝔭 ⊂ 𝑅.

⋄ For any ideal 𝔞 we have dim 𝑅 ≥ dim(𝑅/𝔞) + ht(𝔞).

 §5.4 Definition of Krull dimension ⋅ 59 ⋅

Exercise 5.4.2. Verify the last property above.

Exercise 5.4.3. Show that every principal ideal domain which is not a field has dimen-
sion one.

More generally, we define the dimension of an 𝑅-module 𝑀 as

dim 𝑀 ∶= dim(𝑅/ann(𝑀)), dim{0} ∶= −∞.

For a short exact sequence 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 we have dim 𝑀′ , dim 𝑀″ ≤ dim 𝑀.

Lemma 5.4.4. Suppose 𝑅 is Noetherian. The following are equivalent for a finitely generated
𝑅-module 𝑀 ≠ {0}.

(i) dim 𝑀 = 0.

(ii) 𝑅/ann(𝑀) is Artinian.

(iii) 𝑀 has finite length.

Proof. (i) ⟺ (ii) is already known: recall that a Noetherian ring is Artinian if and
only if its prime ideals are all maximal (Corollary 1.4.3). Let us show (i) or (ii) ⟹
(iii). By writing 𝑀 = 𝑀1 + ⋯ + 𝑀𝑛 where each 𝑀𝑖 is generated by one element, we
may assume 𝑀 ≃ 𝑅/𝔞 for some ideal 𝔞 = ann(𝑀). It has been shown that 𝑅/𝔞 has
finite length as a module since it is an Artinian ring.
(iii) ⟹ (i). Upon modulo ann(𝑀) we may assume ann(𝑀) = {0}. Take any
minimal prime 𝔭 in 𝑅. As ann(𝑀) = {0} we have 𝑀𝔭 ≠ {0}. Therefore 𝔭 is a minimal
element of Supp(𝑀), hence belongs to Ass(𝑀). We may embed 𝑅/𝔭 into 𝑀. The 𝑅-
module 𝑅/𝔭 has finite length since 𝑀 does, therefore 𝑅/𝔭 is an Artinian ring. This
implies 𝔭 is a maximal ideal, therefore dim 𝑅 = 0 since every prime in 𝑅 lies over a
minimal prime.

Our strategy is to study the Krull dimension via completions and Hilbert polyno-
mials. As a preparation, we begin with the local, or more generally the semi-local rings.

Definition 5.4.5. Let 𝑅 be a Noetherian semi-local ring (i.e. there are finitely many
maximal ideals 𝔪1 , … , 𝔪𝑛 ). Let 𝑀 ≠ {0} be a finitely generated 𝑅-module. We say an
ideal 𝐼 is a parameter ideal for 𝑀 if 𝐼 ⊂ rad(𝑅) and 𝑀/𝐼𝑀 has finite length.

Parameter ideals are often called ideals of definition. Here we follow the terminolo-
gies of [8].

Exercise 5.4.6. Show that 𝐼 is a parameter ideal for 𝑅 if and only if there exists 𝑘 with

rad(𝑅)𝑘 ⊂ 𝐼 ⊂ rad(𝑅).

Show that such an ideal is a parameter ideal for every 𝑀. Hint: If 𝐼 ⊃ rad(𝑅)𝑘 , every
prime ideal 𝔭 ⊃ 𝐼 must contain (𝔪1 ⋯ 𝔪𝑛 )𝑘 , hence 𝔭 = 𝔪𝑖 for some 𝑖. Conversely, show
that in an Artinian ring we have rad(𝑅)𝑘 = 0 for 𝑘 ≫ 0, using Corollary 1.4.3. Hint: for
Artinian rings, rad(𝑅) equals the nilpotent radical, and is finitely generated.
⋅ 60 ⋅ From completions to dimensions

Dimension theory for modules can be built solely on the parameter ideals for 𝑅, but
we opt to introduce the general notion here.
Hereafter we fix a Noetherian semi-local ring 𝑅 and a finitely generated 𝑅-module
𝑀 ≠ {0}.
Lemma 5.4.7. An ideal 𝐼 ⊂ rad(𝑅) is a parameter ideal for 𝑀 if and only if there exists 𝑘 with
rad(𝑅)𝑘 ⊂ 𝐼 + ann(𝑀). In this case 𝑅/(𝐼 + ann(𝑀)) is an Artinian ring.
In particular, rad(𝑅) is a parameter ideal for any 𝑀.
Proof. First we claim that 𝑉(ann(𝑀/𝐼𝑀)) = Supp(𝑀/𝐼𝑀) equals 𝑉(𝐼 + ann(𝑀)). By
the exactness of localizations together with Nakayama’s Lemma, we have
Supp(𝑀/𝐼𝑀) = Supp(𝑀) ∩ {𝔭 ∶ 𝐼𝑅𝔭 ⊊ 𝑅𝔭 } ;
the last term equals Supp(𝑀) ∩ 𝑉(𝐼) = 𝑉(ann(𝑀) + 𝐼), thereby proving our claim.
By applying to 𝑀/𝐼𝑀 the Lemma 5.4.4, 𝐼 ⊂ rad(𝑅) being a parameter ideal for 𝑀 is
equivalent to any one of the following
𝑅
is Artinian ⟺ 𝑉(ann(𝑀/𝐼𝑀)) ⊂ MaxSpec(𝑅)
ann(𝑀/𝐼𝑀)
⟺ 𝑉(𝐼 + ann(𝑀)) ⊂ MaxSpec(𝑅)
𝑅
⟺ 𝑅̄ ∶= is Artinian.
𝐼 + ann(𝑀)
If 𝑅̄ is Artinian, then the image of rad(𝑅) in 𝑅̄ is contained in rad(𝑅),
̄ and we know
rad(𝑅) ̄ = 0 for large 𝑘.
𝑘

Conversely, suppose rad(𝑅)𝑘 ⊂ 𝐼 + ann(𝑀). We claim that 𝑅/rad(𝑅)𝑘 is Artinian:

rad(𝑅) contains the product 𝔪1 ⋯ 𝔪𝑘 of all maximal ideals, so every over-prime of
rad(𝑅)𝑘 is some 𝔪𝑖 , thus maximal. This shows that 𝑀/𝐼𝑀 has finite length by the
previous equivalences and Lemma 5.4.4.
Given an parameter ideal 𝐼 for 𝑀. The 𝐼-adic grading gives rise to the graded objects
𝐼𝑛 𝐼𝑛𝑀
gr𝐼 (𝑅) = ⨁ 𝑛+1
, gr𝐼 (𝑀) = ⨁ 𝑛+1 𝑀
.
𝑛≥0 𝐼 𝑛≥0 𝐼

Recall that
⋄ gr𝐼 (𝑅) is finitely generated over gr0𝐼 (𝑅) = 𝑅/𝐼 as an algebra and is Noetherian
(Proposition 4.4.4);
⋄ more precisely, gr𝐼 (𝑅) is generated by gr1𝐼 (𝑅) over 𝑅/𝐼.
⋄ gr𝐼 (𝑀) is a finitely generated gr𝐼 (𝑅)-module (Proposition 4.4.5);
⋄ the ring gr0𝐼 (𝑅) = 𝑅/𝐼 becomes Artinian after modulo gr0𝐼 (𝑅) ∩ ann(gr𝐼 (𝑀)),
which contains (ann(𝑀) + 𝐼)/𝐼 (use Lemma 5.4.7).
Upon recalling Lemma 5.3.1, it are justified to define
𝑛−1
𝜒(𝑀, 𝐼; 𝑛) ∶= ℓ𝑅/𝐼 𝑛 (𝑀/𝐼 𝑛 𝑀) = ∑ ℓ𝑅/𝐼 (𝐼 𝑗 𝑀/𝐼 𝑗+1 𝑀), 𝑛 ∈ ℤ≥0 .
𝑗=0

By the theory of Hilbert–Samuel polynomials, 𝑛 ↦ 𝜒(𝑀, 𝐼; 𝑛) is a polynomial 𝐻𝐼 (𝑀, ⋅)

whenever 𝑛 ≫ 0, with deg 𝐻𝐼 (𝑀, ⋅) bounded by the minimal number of generators of
gr𝐼 (𝑅) over gr0𝐼 (𝑅) = 𝑅/𝐼.
 §5.4 Definition of Krull dimension ⋅ 61 ⋅

Lemma 5.4.8. Let 𝑀 ≠ {0} be a finitely generated 𝑅-module with parameter ideal 𝐼.
(i) The degree 𝑑(𝑀) of 𝐻𝐼 (𝑀, ⋅) is independent of the choice of the parameter ideal 𝐼.
(ii) In a short exact sequence 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 of finitely generated 𝑀-modules, we
have
deg 𝐻𝐼 (𝑀′ , ⋅), deg 𝐻𝐼 (𝑀″ , ⋅) ≤ deg 𝐻𝐼 (𝑀, ⋅),
and 𝐻𝐼 (𝑀, ⋅) − 𝐻𝐼 (𝑀′ , ⋅) − 𝐻𝐼 (𝑀″ , ⋅) has degree < 𝑑(𝑀).
Proof. (i): To compare the graded objects associated to two parameter ideals 𝐼, 𝐽, we
apply the characterization in Lemma 5.4.7: it suffices to take 𝐽 = rad(𝑅), so that

𝐽 𝑚 + ann(𝑀) ⊂ 𝐼 + ann(𝑀) ⊂ 𝐽 + ann(𝑀)

for some 𝑚 ≥ 1. This implies 𝜒(𝑀, 𝐽; 𝑛) ≤ 𝜒(𝑀, 𝐼; 𝑛) and 𝜒(𝑀, 𝐼; 𝑛) ≤ 𝜒(𝑀, 𝐽; 𝑚𝑛) for
all 𝑛 ≥ 0. Whence (i).
(ii): For the first part, note that 𝑀/𝐼 𝑛 𝑀 → 𝑀″ /𝐼 𝑛 𝑀″ is surjective, so 𝜒(𝑀″ , 𝐼; 𝑛) ≤
′ 𝑛
𝜒(𝑀, 𝐼; 𝑛). On the other hand, 𝑀′ /𝐼 𝑛 𝑀′ → 𝑀/𝐼 𝑛 𝑀 has kernel 𝑀𝐼 𝑛∩𝐼𝑀′𝑀 . For 𝑛 ≥ 𝑛0 ≫ 0,
Artin–Rees (Theorem 4.5.7) gives

𝑀′ ∩ 𝐼 𝑛 𝑀 𝐼 𝑛−𝑛0 (𝑀′ ∩ 𝐼 𝑛0 𝑀) 𝐼 𝑛−𝑛0 𝑀′

ℓ( ) = ℓ( ) ≤ ℓ( 𝑛 ′ )
𝐼 𝑛 𝑀′ 𝐼 𝑛 𝑀′ 𝐼 𝑀
𝑀 ′ 𝑀 ′
= ℓ ( 𝑛 ′ ) − ℓ ( 𝑛−𝑛0 ′ ) = 𝜒(𝑀′ , 𝐼; 𝑛) − 𝜒(𝑀′ , 𝐼; 𝑛 − 𝑛0 )
𝐼 𝑀 𝐼 𝑀
which has degree inferior to 𝐻𝐼 (𝑀′ , ⋅). Hence deg 𝐻𝐼 (𝑀′ , ⋅) ≤ deg 𝐻𝐼 (𝑀, ⋅).
To establish the second part of (ii), we consider 𝜒(𝑀, 𝐼; 𝑛)−𝜒(𝑀″ , 𝐼; 𝑛) which equals
𝑀 𝑀 𝑀′ + 𝐼 𝑛 𝑀 𝑀′
ℓ( ) − ℓ ( ) = ℓ ( ) = ℓ ( ).
𝐼𝑛𝑀 𝑀′ + 𝐼 𝑛 𝑀 𝐼𝑛𝑀 𝑀′ ∩ 𝐼 𝑛 𝑀
By Artin–Rees, the rightmost term is squeezed between ℓ(𝑀′ /𝐼 𝑛 𝑀′ ) = 𝜒(𝑀′ , 𝐼; 𝑛) and
ℓ(𝑀′ /𝐼 𝑛−𝑘 𝑀′ ) = 𝜒(𝑀′ , 𝐼; 𝑛 − 𝑘) for some 𝑘 independent of 𝑛 ≫ 0. Hence 𝜒(𝑀, 𝐼; 𝑛) −
𝜒(𝑀″ , 𝐼; 𝑛) is a polynomial with the same leading term as 𝜒(𝑀′ , 𝐼; 𝑛), for 𝑛 ≫ 0.
Define 𝑠(𝑀) to be the smallest integer 𝑠 such that there exist 𝑡1 , … , 𝑡𝑠 ∈ rad(𝑅) with
𝑠
𝑀/ ∑𝑖=1 𝑡𝑖 𝑀 of finite length. In other words, 𝑠(𝑀) is the minimal number of generators
𝑠
for parameter ideals for 𝑀. Observe that 𝑀/ ∑𝑖=1 𝑡𝑖 𝑀 ≠ {0}, otherwise Nakayama’s
Lemma will lead to 𝑀 = {0}.
Theorem 5.4.9. For any finitely generated nonzero 𝑅-module 𝑀, we have dim 𝑀 = 𝑠(𝑀) =
𝑑(𝑀). In particular dim 𝑀 is finite.
Proof. We argue inductively on 𝑑(𝑀) to show dim 𝑀 ≤ 𝑑(𝑀). If 𝑑(𝑀) = 0 then 𝐼 𝑛 𝑀 =
𝐼 𝑛+1 𝑀 = ⋯ for 𝑛 ≫ 0. Corollary 4.5.9 implies 𝐼 𝑛 𝑀 = {0}, hence 𝑀 = 𝑀/𝐼 𝑛 𝑀 has finite
length and dim 𝑀 = 0 by Lemma 5.4.4.
Now assume 𝑑(𝑀) ≥ 1. Take a minimal 𝔭 ∈ Ass(𝑀) verifying dim(𝑅/𝔭) = dim 𝑀,
so that 𝑅/𝔭 ↪ 𝑀. As 𝑑(𝑅/𝔭) ≤ 𝑑(𝑀), we are reduced to the case 𝑀 = 𝑅/𝔭. Consider a
chain of prime ideals
𝔭 = 𝔭 0 ⊊ ⋯ ⊊ 𝔭𝑚
⋅ 62 ⋅ From completions to dimensions

in 𝑅. We claim that 𝑚 ≤ 𝑑(𝑅/𝔭). We may surely suppose 𝑚 ≥ 1. Take 𝑡 ∈ 𝔭1 ∖ 𝔭0 .

Reduction modulo 𝑅𝑡 + 𝔭 yields a prime chain of length 𝑚 − 1, namely
𝔭1 𝔭𝑚
⊊⋯
𝑅𝑡 + 𝔭 𝑅𝑡 + 𝔭
in 𝑅/(𝑅𝑡 + 𝔭). Hence dim(𝑅/𝑅𝑡 + 𝔭) ≥ 𝑚 − 1. In view of the exactness of
𝑡 𝑅
0 → 𝑅/𝔭 𝑅/𝔭 → → 0,
𝑅𝑡 + 𝔭
Lemma 5.4.8 (ii) implies that 𝑑(𝑅/𝑅𝑡 + 𝔭) < 𝑑(𝑅/𝔭). By induction we deduce 𝑑(𝑅/𝔭) >
𝑑(𝑅/𝑅𝑡 + 𝔭) ≥ dim(𝑅/𝑅𝑡 + 𝔭) ≥ 𝑚 − 1, hence 𝑑(𝑅/𝔭) ≥ 𝑚 as required.
Next, let us show 𝑠(𝑀) ≤ dim 𝑀. Set 𝑟 ∶= dim 𝑀. We contend that there exist
𝑡1 , … , 𝑡𝑟 ∈ rad(𝑅) such that 𝑀/(𝑡1 , … , 𝑡𝑟 )𝑀 has finite length; therefore 𝑠 ≤ 𝑟. When
𝑟 = 0 this follows from Lemma 5.4.4. For 𝑟 > 0, we have rad(𝑅) ⊄ 𝔭 for any minimal
𝔭 ∈ Ass(𝑀) verifying dim 𝑅/𝔭 = dim 𝑀, for otherwise 𝔭 will contain, thus equal to a
maximal ideal as 𝑅 is semi-local, and we would get dim 𝑀 = 0. Using prime avoidance
(Proposition 1.1.5 applied to 𝐼 ∶= rad(𝑅) and the primes 𝔭 above), we may pick 𝑡1 ∈
rad(𝑅) that does not belong to any 𝔭 above. From ann(𝑀/𝑡1 𝑀) ⊃ 𝑅𝑡1 + ann(𝑀) and
our choice of 𝑡1 ,, we see dim 𝑀/𝑡1 𝑀 < dim 𝑀. Our claim results from induction on 𝑟.
𝑠
We finish the proof by showing 𝑑(𝑀) ≤ 𝑠(𝑀). Suppose 𝑀/ ∑𝑖=1 𝑡𝑖 𝑀 ≠ {0} has
finite length. We contend that for any 𝑡 ∈ rad(𝑅) we have
𝑑(𝑀) ≥ 𝑑(𝑀/𝑡𝑀) ≥ 𝑑(𝑀) − 1. (5–3)
If this holds, we can look at the sequence 𝑀, 𝑀/𝑡1 𝑀, 𝑀/(𝑡1 𝑀 + 𝑡2 𝑀), …: at each step
𝑠
𝑑(⋯) drops at most by one; at the end 𝐿 ∶= 𝑀/ ∑𝑗=1 𝑡𝑗 𝑀 we have 𝑑(𝐿) = 0: indeed,
as 𝐿 has finite length, ℓ(𝐿/𝐼 𝑛 𝐿) is uniformly bounded by ℓ(𝐿) so that 𝑑(𝐿) = 0. Thus
𝑑(𝑀) ≤ 𝑠 as expected.
To prove (5–3), first note that 𝑑(𝑀) ≥ 𝑑(𝑀/𝑡𝑀) is known. Take any parameter ideal
𝐼 ∋ 𝑡. We bound 𝜒(𝑀/𝑡𝑀, 𝐼; 𝑛) as follows
𝑀 𝑀 𝑡𝑀 + 𝐼 𝑛 𝑀
ℓ( ) = ℓ( 𝑛 ) − ℓ( ).
𝑡𝑀 + 𝐼 𝑛 𝑀 𝐼 𝑀 𝐼𝑛𝑀
Note that
𝑀 𝑀 ∼ 𝑡𝑀 𝑡𝑀 + 𝐼 𝑛 𝑀
↠ → ≃
𝐼 𝑛−1 𝑀 {𝑥 ∈ 𝑀 ∶ 𝑡𝑥 ∈ 𝐼 𝑛 𝑀} 𝑡𝑀 ∩ 𝐼 𝑛 𝑀 𝐼𝑛𝑀
𝑦 ↦ 𝑡𝑦.
Hence 𝜒(𝑀/𝑡𝑀, 𝐼; 𝑛) ≥ 𝜒(𝑀, 𝐼; 𝑛) − 𝜒(𝑀, 𝐼; 𝑛 − 1) for 𝑛 ≫ 0, proving the second in-
equality in (5–3).
Corollary 5.4.10. Under the same assumptions, we have dim𝑅 𝑀 = dim𝑅̂ 𝑀,̂ where we take
𝐼-adic completions.
Proof. Proposition 5.2.1 gives identifications gr𝐼 (𝑅) = gr𝐼 (̂ 𝑅) ̂ and gr (𝑀) = gr (̂ 𝑀); ̂
𝐼 𝐼
moreover 𝑅̂ is still semi-local and 𝐼 ̂ is still a parameter ideal for 𝑀̂ (see Proposition 5.2.4,
5.2.5). Since 𝑑(𝑀) and 𝑑(𝑀) ̂ are read from these graded modules, they are equal.

These results will be applied to the case 𝑀 = 𝑅 in the next section.

 §5.5 Krull’s theorems and regularity ⋅ 63 ⋅

5.5 Krull’s theorems and regularity

We still assume 𝑅 is a Noetherian ring.

Theorem 5.5.1 (Krull). Suppose 𝔞 = (𝑡1 , … , 𝑡𝑟 ) is a proper ideal of 𝑅, then for every minimal
over-prime ideal 𝔭 of 𝔞, we have ht(𝔭) ≤ 𝑟.

From Definition 5.4.1 we infer that ht(𝔞) ≤ 𝑟. The special case 𝑟 = 1 says that every
principal ideal (𝑡) ≠ 𝑅 has height at most one (exactly one if 𝑡 is not a zero divisor —
use Theorem 2.2.5 (ii)); this is called the Hauptidealsatz.

Proof. We work in 𝑅𝔭 and 𝐼 ∶= 𝔞𝑅𝔭 = (𝑡1 , … , 𝑡𝑟 ). Since 𝐼 ⊂ rad(𝑅𝔭 ) and 𝑅𝔭 /𝐼 has

dimension zero, thus Artinian, 𝐼 is a parameter ideal. We conclude by Theorem 5.4.9
and ht(𝔭) = dim 𝑅𝔭 .

Now assume 𝑅 is Noetherian and local with maximal ideal 𝔪. The parameter ideals
of 𝑅 are precisely those squeezed between 𝔪 and 𝔪𝑘 for some 𝑘 ≥ 1, by Lemma 5.4.7.
Set 𝑑 ∶= dim 𝑅, which is finite by Theorem 5.4.9. The same theorem tells us that we can
generate some parameter ideal 𝐼 ⊂ 𝔪 (namely 𝑅/𝐼 Artinian) by elements 𝑡1 , … , 𝑡𝑑 ∈ 𝔪.
These elements form a system of parameters of 𝑅.

Proposition 5.5.2. Suppose 𝑅 is a Noetherian local ring with a system of parameters 𝑡1 , … , 𝑡𝑑 ,

which generate a parameter ideal 𝐼. For any 0 ≤ 𝑖 ≤ 𝑑 we have dim(𝑅/(𝑡1 , … , 𝑡𝑖 )) = 𝑑 − 𝑖,
and 𝑡𝑖+1 , … , 𝑡𝑑 form a system of parameters for 𝑅/(𝑡1 , … , 𝑡𝑖 ).

Proof. Consider the sequence 𝑅, 𝑅/(𝑡1 ), 𝑅/(𝑡1 , 𝑡2 ), … , 𝑅/(𝑡1 , … , 𝑡𝑑 ). Recall the formal-
ism in Theorem 5.4.9: at each stage 𝑑(𝑅/ ⋯) drops at most by one, by (5–3). After 𝑑
steps we arrive at 𝑅/𝐼 with dim(⋅) = 𝑑(⋅) = 𝑠(⋅) = 0, since it has finite length. Hence 𝑑
drops exactly by one at each stage. The remaining assertions are immediate.

A natural question arises: when can we assure 𝐼 = 𝔪?

Definition 5.5.3. We say a Noetherian local ring 𝑅 is a regular local ring if 𝔪 can be
generated by 𝑑 = dim 𝑅 elements 𝑡1 , … , 𝑡𝑑 . In this case we say 𝑡1 , … , 𝑡𝑟 form a regular
system of parameters.

In particular, 𝔪 = {0} if 𝑅 is regular local with dim 𝑅 = 0.

Exercise 5.5.4. Let 𝕜 be a field. The 𝕜-algebra of formal power series 𝕜J𝑋1 , … , 𝑋𝑑 K is
a regular local ring. Indeed, it is Noetherian with maximal ideal 𝔪 = (𝑋1 , … , 𝑋𝑑 ), and
𝑋1 , … , 𝑋𝑑 form a regular system of parameters. On the other hand, 𝔪/𝔪2 has a 𝕜-basis
formed by the images of 𝑋1 , … , 𝑋𝑑 . One way to determine its dimension and prove its
regularity is to calculate the functions 𝑛 ↦ dim𝕜 (𝔪𝑛 /𝔪𝑛+1 ) explicitly, i.e. count the
monomials in 𝑑 variables with total degree 𝑛. You should get a polynomial in 𝑛 with
degree 𝑑 − 1, cf. Exercise 5.3.4.

Theorem 5.5.5. For any Noetherian local ring 𝑅 with maximal ideal 𝔪 and residue field 𝕜, we
have dim 𝑅 ≤ dim𝕜 𝔪/𝔪2 . Equality holds if and only if 𝑅 is a regular local ring.
⋅ 64 ⋅ From completions to dimensions

The 𝕜-vector space 𝔪/𝔪2 is called the Zariski cotangent space of

Spec(𝑅), in honor of Oscar Zariski (1899–1986). Picture taken from
Wikimedia Commons.

Proof. By Nakayama’s Lemma (more precisely, Corollary 1.3.6), 𝔪/𝔪2 can be generated
over 𝕜 by 𝑠 elements if and only if 𝔪 can be generated over 𝑅 by 𝑠 elements, for any
𝑠 ∈ ℤ≥0 . Hence Theorem 5.4.9 imposes the bound 𝑠 ≥ dim 𝑅, and equality holds if and
only if 𝑅 admits a regular system of parameters.
Due to time constraints, we cannot say too much about regular local rings. Below
is one of their wonderful properties.

Theorem 5.5.6. Regular local rings are integral domains.

Proof. Induction on 𝑑 ∶= dim 𝑅. If dim 𝑅 = 0 then 𝔪 = {0}, hence 𝑅 is a field. Assume

hereafter that 𝑑 > 0. We know there are only finitely many minimal prime ideals and
dim𝕜 𝔪/𝔪2 ≥ 1. By prime avoidance (Proposition 1.1.5) applied to 𝐼 ∶= 𝔪, the minimal
prime ideals and 𝔪2 , there exists 𝑡 ∈ 𝔪 ∖ 𝔪2 that does not lie in any minimal prime
ideal. Put 𝑅′ ∶= 𝑅/(𝑡) with maximal ideal 𝔪′ = 𝔪/(𝑡); our choice of 𝑡 together with
Proposition 5.5.2 imply

dim 𝑅′ = dim 𝑅 − 1,
dim𝕜 𝔪′ /(𝔪′ )2 = dim𝕜 𝔪/𝔪2 − 1 = dim 𝑅 − 1.

Hence 𝑅′ is still regular local, and by induction it is a domain. This implies (𝑡) is prime.
Take any minimal prime 𝔭 below (𝑡); note that 𝑡 ∉ 𝔭 by construction. To show 𝑅 is a
domain it suffices to prove 𝔭 = {0}. Indeed, every 𝑠 ∈ 𝔭 can be written as 𝑠 = 𝑎𝑡, 𝑎 ∈ 𝑅.
Since 𝑡 ∉ 𝔭, we must have 𝑎 ∈ 𝔭. Hence 𝔭 = 𝑡𝔭 ⊂ 𝔪𝔭. Nakayama’s Lemma (Theorem
1.3.5) implies 𝔭 = {0}.
 Lecture 6

Dimension of finitely
generated algebras
The main reference is [11, §13].

6.1 Dimensions in fibers

Consider a homomorphism 𝜑 ∶ 𝐴 → 𝐵, which induces 𝜑♯ ∶ Spec(𝐵) → Spec(𝐴) on
prime spectra. Given 𝔭 ∈ Spec(𝐴), we are interested in the fiber (𝜑♯ )−1 (𝔭); the prime
ideals 𝔮 therein are described by 𝜑−1 (𝔮) = 𝔭, or equivalently:
𝜑−1 (𝔮) ∩ (𝐴 ∖ 𝔭) = ∅, 𝔮 ⊃ 𝜑(𝔭).
Adopt the convention that a zero ring has Spec = ∅. The first condition says that 𝔮
comes from Spec(𝐵𝔭 ), where 𝐵𝔭 is the localization with respect to 𝜑(𝐴 ∖ 𝔭) (possibly
zero). The second condition then says that 𝔮𝐵𝔭 lies over the image of 𝔭𝐴𝔭 . Set
𝜅(𝔭) ∶= 𝐴𝔭 /𝔭𝐴𝔭 .

The fiber (𝜑♯ )−1 (𝔭) is then identified with the spectrum of

𝐵 ⊗ 𝜅(𝔭) = (𝐵 ⊗ 𝐴𝔭 ) ⊗ 𝜅(𝔭) = 𝐵𝔭 ⊗ (𝐴𝔭 /𝔭𝐴𝔭 ),

𝐴 𝐴 𝐴𝔭 𝐴𝔭

which is empty if and only if 𝐵 ⊗ 𝜅(𝔭) is zero. This also equips (𝜑♯ )−1 (𝔭) with an extra
𝐴
structure: it is the spectrum of an explicit quotient ring of 𝐵𝔭 .
Observe that for all 𝔮 ∈ (𝜑♯ )−1 (𝔭), the localization of 𝐵𝔭 ⊗ 𝜅(𝔭) at the image of 𝔮 is
𝐴𝔭
canonically isomorphic to 𝐵𝔮 ⊗ 𝜅(𝔭).
𝐴𝔭

Proposition 6.1.1. Assume 𝐴, 𝐵 to be Noetherian. Let 𝔮 ∈ Spec(𝐵) and 𝔭 ∶= 𝜑♯ (𝔮). We have

(i) dim(𝐵𝔮 ) ≤ dim(𝐴𝔭 ) + dim 𝐵𝔮 ⊗ 𝜅(𝔭) (note that the existence of 𝔮 ensures that 𝐵𝔭 ⊗
𝐴𝔭 𝐴𝔭
𝜅(𝔭) ≠ {0}, hence 𝐵𝔮 ⊗ 𝜅(𝔭) ≠ {0});
𝐴𝔭
⋅ 66 ⋅ Dimension of finitely generated algebras

(ii) equality holds if going-down holds for 𝜑;

(iii) if going-down holds and 𝜑♯ is surjective, then dim(𝐵) ≥ dim(𝐴), and for all ideal 𝔞 ⊊ 𝐴
we have 𝜑(𝔞)𝐵 ≠ 𝐵 and ht(𝔞) = ht(𝜑(𝔞)𝐵).
It is crucial to notice that dim 𝐵𝔮 ⊗ 𝜅(𝔭) = ht(𝔮𝐵𝔮 /𝜑(𝔭)𝐵𝔮 ) = ht(𝔮/𝜑(𝔭)𝐵).
𝐴𝔭
This bounds the source dimension of a morphism by the target dimension plus the
fiber dimension, localized both at 𝔭 and 𝔮 ∈ (𝜑♯ )−1 (𝔭). In order to have an equality, a
certain submersion-like condition on 𝜑♯ is evidently required; this explains the going-
down condition. Cf. [11, (6.H)].
Proof. We have an induced local homomorphism 𝐴𝔭 → 𝐵𝔮 since 𝔮 ↦ 𝔭. Since (i) and
(ii) depend only on this induced homomorphism, we may assume from the outset that
𝐴, 𝐵 are local with maximal ideals 𝔭, 𝔮, and 𝜑 is a local homomorphism. Let 𝑑 ∶= dim 𝐴
and take a parameter ideal 𝐼 = (𝑡1 , … , 𝑡𝑑 ) of 𝐴, so that 𝔭𝑘 ⊂ 𝐼 ⊂ 𝔭 for some 𝑘. It follows
that √𝜑(𝔭)𝐵 = √𝜑(𝐼)𝐵, therefore dim(𝐵 ⊗ 𝜅(𝔭)) = dim(𝐵/𝜑(𝔭)𝐵) = dim(𝐵/𝜑(𝐼)𝐵);
𝐴
denote this number as 𝑒. Take a 𝑠1 , … , 𝑠𝑒 ∈ 𝔮 whose images generate a parameter ideal
for 𝐵/𝜑(𝐼)𝐵, and put 𝐽 ∶= (𝜑(𝑡1 ), … , 𝜑(𝑡𝑑 ), 𝑠1 , … , 𝑠𝑒 ). Then 𝐵/𝐽 is Artinian, therefore
dim 𝐵 ≤ 𝑑 + 𝑒 establishes (i).
As for (ii), we conserve the same hypotheses and take a prime chain 𝔮 = 𝔮0 ⊋ ⋯ ⊋
𝔮𝑒 with 𝔮𝑒 ⊃ 𝜑(𝔭)𝐵 in 𝐵, as well as a prime chain 𝔭 = 𝔭0 ⊋ ⋯ ⊋ 𝔭𝑑 in 𝐴. Note that
𝜑−1 (𝔮𝑒 ) = 𝔭 since 𝑒 = dim 𝐵/𝜑(𝔭)𝐵. By applying going-down repeatedly to the chain
𝔭𝑖 , we obtain a prime chain in 𝐵

𝔮𝑒 ⊋ ⋯ ⊋ 𝔮𝑒+𝑑 , 𝜑−1 (𝔮𝑒+𝑖 ) = 𝔭𝑖 .

Concatenation with 𝔮0 ⊋ ⋯ gives dim(𝐵) = 𝑑 + 𝑒. This shows (ii).

The first assertion of (iii) results from (ii). To show the remaining one, let us show
𝜑(𝔞)𝐵 ≠ 𝐵: if 𝜑♯ (𝔮) = 𝔭 ⊃ 𝔞, then 𝔮 ⊃ 𝜑(𝔞)𝐵. Next, take a minimal over-prime
𝔮 ⊃ 𝜑(𝔞)𝐵 with ht(𝔮) = ht(𝜑(𝔞)𝐵). With 𝔭 ∶= 𝜑−1 (𝔮) ⊃ 𝔞, we must have dim 𝐵𝔮 ⊗
𝜅(𝔭) = ht(𝔮/𝜑(𝔭)𝐵) = 0 by the minimality of 𝔮. An application of (ii) yields

ht(𝜑(𝔞)𝐵) = ht(𝔮) = ht(𝔭) ≥ ht(𝔞).

To obtain ≤, choose 𝔭 ⊃ 𝔞 with ht(𝔭) = ht(𝔞) and take 𝔮 ∈ Spec(𝐵) with 𝔭 = 𝜑−1 (𝔮);
this implies 𝔮 ⊃ 𝜑(𝔭)𝐵 ⊃ 𝜑(𝔞)𝐵. Upon shrinking 𝔮, we may even assume 𝔮 is minimal
over 𝜑(𝔭)𝐵, i.e. ht(𝔮/𝜑(𝔭)𝐵) = 0. Using (ii), this entails ht(𝔞) = ht(𝔭) = ht(𝔮) ≥
ht(𝜑(𝔞)𝐵).

The minimal primes

in a ring 𝐵 may have
different heights when
the scheme Spec(𝐵) is
not equi-dimensional.
 §6.2 Calculation for polynomial algebras ⋅ 67 ⋅

Going-down holds for flat 𝜑 by Theorem 4.1.3, therefore the dimension equality

dim(𝐵𝔮 ) = dim(𝐴𝔭 ) + dim 𝐵𝔮 ⊗ 𝜅(𝔭)

𝐴𝔭

holds for flat ring homomorphisms.

Remark 6.1.2. In general, if 𝐵 is a finitely generated algebra over Noetherian 𝐴 such
that Spec(𝐵) → Spec(𝐴) is a closed map, the fiber dimension 𝔭 ↦ dim 𝐵/𝔭𝐵 is an
upper semi-continuous function on the target space Spec(𝐴). More concretely, the fiber
dimension is non-decreasing under specialization of 𝔭. Cf. [8, §14.3] or [11, (13.E)].
Try to understand this phenomenon intuitively.
Proposition 6.1.3. Suppose 𝐵 is integral over a subring 𝐴.
(i) We have dim 𝐴 = dim 𝐵.
(ii) If we assume moreover that 𝐴, 𝐵 are both Noetherian, then ht(𝔮) ≤ ht(𝔮 ∩ 𝐴) for every
𝔮 ∈ Spec(𝐵).
(iii) Furthermore, if going-down also holds for 𝐴 ↪ 𝐵, we have ht(𝐽) = ht(𝐽 ∩ 𝐴) for every
ideal 𝐽 ⊊ 𝐵.
Proof. Going-up holds and Spec(𝐵) ↠ Spec(𝐴) in the situation of (i) by Theorem 4.1.4,
hence dim 𝐵 ≥ dim 𝐴 by lifting prime chains. To prove ≤, observe that 𝔮 ⊊ 𝔮′ implies
𝔮 ∩ 𝐴 ⊊ 𝔮′ ∩ 𝐴 since there are no inclusion relations in the fibers of Spec(𝐵) → Spec(𝐴).
As for (ii), note that ht(𝔮) ≤ ht(𝔭) + ht(𝔮/𝔭𝐵) where 𝔭 ∶= 𝔮 ∩ 𝐴; as the are no
inclusion relations in fibers, the last term must be 0.
Now assume going-down and consider (iii). Take 𝔮 ∈ Spec(𝐵) with ht(𝔮) = ht(𝐽).
Put 𝔭 ∶= 𝔮 ∩ 𝐴 ⊃ 𝐽 ∩ 𝐴. Again, since there are no inclusions in the fiber over 𝔭 of
Spec(𝐵) ↠ Spec(𝐴), we have dim(𝐵𝔮 /𝔭𝐵𝔮 ) = 0. Proposition 6.1.1 (ii) implies ht(𝔮) =
ht(𝔭), therefore ht(𝐽) ≥ ht(𝐽 ∩ 𝐴).
On the other hand, for any 𝔭 ⊃ 𝐽 ∩ 𝐴 with ht(𝔭) = ht(𝐽 ∩ 𝐴), since 𝐴/𝐽 ∩ 𝐴 ↪ 𝐵/𝐽
is integral, there exists 𝔮 ⊃ 𝐽 with 𝔮 ∩ 𝐴 = 𝔭. Together with Proposition 6.1.1 (i) and
ht(𝔮/𝔭𝐵) = 0, this implies ht(𝐽) ≤ ht(𝔮) ≤ ht(𝔭) = ht(𝐽 ∩ 𝐴) by (ii).

6.2 Calculation for polynomial algebras

Let us apply the results from the previous section to elucidate the Krull dimension of
polynomial algebras.
Theorem 6.2.1. Let 𝐴 be a Noetherian ring, we have dim 𝐴[𝑋1 , … , 𝑋𝑛 ] = dim 𝐴 + 𝑛 for any
𝑛 ≥ 0. In particular, dim 𝐴[𝑋1 , … , 𝑋𝑛 ] = 𝑛 if 𝐴 is Artinian (eg. a field).
Proof. Evidently we may assume 𝑛 = 1. We shall apply Proposition 6.1.1 to 𝐴 ↪ 𝐵 =
𝐴[𝑋]. Take any 𝔭 ∈ Spec(𝐴) and let 𝔮 be a maximal element in {𝔮′ ∈ Spec(𝐵) ∶ 𝔮 ∩ 𝐴 =
𝔭}. Put 𝜅 ∶= 𝜅(𝔭). It suffices to show that 𝐵𝔮 ⊗ 𝜅 has dimension one, since 𝐵 is free
𝐴𝔭
hence flat over 𝐴, and Proposition 6.1.1 will imply

dim 𝐵𝔮 = dim 𝐴𝔭 + 1
⋅ 68 ⋅ Dimension of finitely generated algebras

and taking supremum over 𝔭 ∈ Spec(𝐴) gives the result.

Indeed, put 𝐵𝔭 ∶= 𝐵[(𝐴 ∖ 𝔭)−1 ] = 𝐴𝔭 [𝑋] and 𝔮′ ∶= 𝔮[(𝐴 ∖ 𝔭)−1 ] ∈ Spec(𝐵𝔭 ). As
𝔮′ ⊃ 𝔭𝐴𝔭 , we have 𝔮′ ∶= 𝔮′ ⊃ 𝔭𝐴𝔭 ∈ Spec(𝜅[𝑋]), and 𝔮′ is maximal in the fiber over {0}
of Spec(𝜅[𝑋]) → Spec(𝜅), i.e. in MaxSpec(𝜅[𝑋]). Localization in stages yields
𝐵𝔮 𝐵𝔭
𝐵𝔮 ⊗ 𝜅 ≃ ≃( ) ≃ 𝜅[𝑋]𝔮′ .
𝐴𝔭 𝔭𝐵𝔮 𝔭𝐴𝔭 𝐵𝔭 𝔮′

As 𝜅[𝑋] is a principal ideal domain which is not a field, every maximal ideal thereof
has height one. Hence dim 𝜅[𝑋]𝔮′ = ht(𝔮′ ) = 1.

Corollary 6.2.2. Let 𝕜 be a field, then for every 0 ≤ 𝑖 ≤ 𝑛 we have ht(𝑋1 , … , 𝑋𝑖 ) = 𝑖 in

𝕜[𝑋1 , … , 𝑋𝑛 ].
Proof. The prime chain {0} ⊂ (𝑋1 ) ⊊ ⋯ ⊊ (𝑋1 , … , 𝑋𝑛 ) has length 𝑛 = dim 𝕜[𝑋1 , … , 𝑋𝑛 ].
Thus for each 0 ≤ 𝑖 ≤ 𝑛, the chain {0} ⊂ (𝑋1 ) ⊊ ⋯ ⊊ (𝑋1 , … , 𝑋𝑖 ) has maximal length
among all prime chains starting with (𝑋1 , … , 𝑋𝑖 ).
Combined with Theorem 6.2.1, we see that for 𝕜 a field, 𝑅 ∶= 𝕜[𝑋1 , … , 𝑋𝑛 ] and
𝔭 ∶= (𝑋1 , … , 𝑋𝑖 ), the equality

ht(𝔭) + dim 𝑅/𝔭 = dim 𝑅.

holds. This will be generalized to finitely generated domains over 𝕜.

We record another simple consequence for later use.
Corollary 6.2.3. Let 𝕜 be a field. Any 𝕜-algebra 𝐴 with 𝑛 generators has finite dimension ≤ 𝑛.
Proof. Writing 𝐴 = 𝕜[𝑋1 , … , 𝑋𝑛 ]/𝐼 for some ideal 𝐼, we have dim 𝐴 ≤ 𝕜[𝑋1 , … , 𝑋𝑛 ]
since every prime chain in 𝐴 lifts to 𝕜[𝑋1 , … , 𝑋𝑛 ]. Now apply Theorem 6.2.1.

6.3 Noether normalization and its consequences

Fix a field 𝕜. A few preparatory results are in order.
Lemma 6.3.1. Suppose 𝕜 is a field and 𝑡 ∈ 𝕜[𝑋1 , … 𝑋𝑒 ] ∖ 𝕜. There exist 𝑡1 , … , 𝑡𝑒−1 ∈
𝕜[𝑋1 , … , 𝑋𝑒 ] such that 𝕜[𝑋1 , … , 𝑋𝑒 ] is finitely generated as a module over the 𝕜-subalgebra
𝑆 ∶= 𝕜[𝑡1 , … , 𝑡𝑒−1 , 𝑡].
𝑖
Proof. We seek 𝑡𝑖 of the form 𝑋𝑖 − 𝑋𝑒𝑘 where 𝑘 is a large integer. Then 𝑡 can be uniquely
expressed as a polynomial of 𝑡1 , … , 𝑡𝑒−1 , 𝑋𝑒 . We claim that upon modifying 𝑡 by 𝕜× ,
which is clearly harmless, one can choose 𝑘 such that 𝑡 is monic as an element of
𝕜[𝑡1 , … , 𝑡𝑒−1 ][𝑋𝑒 ], say of some degree 𝛿. If this is the case,
𝑗
𝑡 = 𝑋𝑒𝛿 + ∑ (polynomial in 𝑡1 , … , 𝑡𝑒−1 ) 𝑋𝑒
0≤𝑗<𝛿

says that 𝑋𝑒 is integral over 𝑆, and then 𝕜[𝑋1 , … , 𝑋𝑒 ] is generated as an 𝑆-module by

1, 𝑋𝑒 , … , 𝑋𝑒𝛿−1 .
 §6.3 Noether normalization and its consequences ⋅ 69 ⋅

To choose 𝑘, one stares at the expansion

𝑎 𝑎 1 𝑎1 𝑒−1 𝑎𝑒−1 𝑎 𝑎 +𝑎1 𝑘1 +⋯+𝑎𝑒−1 𝑘𝑒−1
𝑋11 ⋯ 𝑋𝑒 𝑒 = (𝑡1 + 𝑋𝑒𝑘 ) ⋯ (𝑡𝑒−1 + 𝑋𝑒𝑘 ) ⋅𝑋𝑒 𝑒 = 𝑋𝑒 𝑒 +mixed terms.
If 𝑘 > max{𝑎1 , … , 𝑎𝑒 }, the exponent of 𝑋𝑒 is simply the base 𝑘 expression with digits
𝑎 𝑎
𝑎𝑒 , 𝑎1 , … , 𝑎𝑒−1 . Now write 𝑡 as a linear combination of monomials 𝑋11 ⋯ 𝑋𝑒 𝑒 and ex-
pand them in terms of 𝑡1 , … , 𝑡𝑒−1 , 𝑋𝑒 . From the observation above, different (𝑎1 , … , 𝑎𝑒 )
contributes a different exponent of 𝑋𝑒 whenever 𝑘 ≫ 0. Adjusting 𝑡 by 𝕜× , we get the
asserted property.
Lemma 6.3.2. Let 𝔞 be a nonzero ideal of a domain 𝑅, then dim(𝑅/𝔞) + 1 ≤ dim 𝑅.
Proof. Any prime chain 𝔭0 ⊋ ⋯ ⊋ 𝔭𝑛 in 𝑅 with 𝔭𝑛 ⊃ 𝔞 can be extended to a prime chain
length 𝑛 + 1, namely by adjoining 𝔭𝑛+1 ∶= {0}.
Theorem 6.3.3 (E. Noether, M. Nagata). Let 𝐵 be a finitely generated 𝕜-algebra of dimension
𝑛. Consider a chain of proper ideals 𝐼1 ⊊ ⋯ ⊊ 𝐼𝑚 , with dim(𝐵/𝐼𝑗 ) = 𝑑𝑗 and 𝑑1 > ⋯ > 𝑑𝑚 ≥ 0.
Then there exist a 𝕜-subalgebra 𝐴 ⊂ 𝐵 together with an isomorphism 𝐴 ≃ 𝕜[𝑋1 , … , 𝑋𝑛 ],
satisfying
⋄ 𝐵 is a finitely generated 𝐴-module, in particular 𝐵 is integral over 𝐴;
⋄ 𝐼𝑗 ∩ 𝐴 ≃ (𝑋𝑑𝑗 +1 , … , 𝑋𝑛 ) under the isomorphism above, for each 1 ≤ 𝑗 ≤ 𝑚.
Note that the assumption 𝐼𝑗 ⊊ 𝐼𝑗+1 is merely for convenience. Allowing 𝐼𝑗 = 𝐼𝑗+1
and 𝑑𝑗 = 𝑑𝑗+1 for some 𝑗 is surely possible..
Proof. Write 𝐵 = 𝕜[𝑌1 , … , 𝑌𝑟 ]/𝐽, where 𝑟 ≥ 𝑛 (Corollary 6.2.3), and denote the preim-
age of 𝐼𝑗 in 𝕜[𝑌1 , … , 𝑌𝑟 ] by 𝐼𝑗̃ . The first step is to reduce to the case 𝐽 = {0}. To see this,
we adjoin 𝐼0̃ = {0} (with 𝑑0 = 𝑛) into the ideal chain; it may happen that 𝐼0̃ = 𝐼1̃ , but
that’s harmless. Suppose we can find 𝐴̃ ⊂ 𝕜[𝑌1 , … , 𝑌𝑟 ], 𝐴̃ ≃ 𝕜[𝑋1 , … , 𝑋𝑟 ] with the
required properties relative to 𝐼•̃ . Taking quotient by 𝐽, we obtain the corresponding
properties for 𝐼• . Indeed, the passage from 𝐴̃ to 𝐴 ∶= 𝐴/( ̃ 𝐴̃ ∩ 𝐽) truncates the variables
𝑋𝑛+1 , … , 𝑋𝑟 , whereas
𝐼𝑗̃ ∩ (𝐴̃ + 𝐽) (𝐼𝑗̃ ∩ 𝐴)̃ + 𝐽 𝐼𝑗̃ ∩ 𝐴̃
𝐼𝑗 ∩ 𝐴 = = ≃ ;
𝐽 𝐽 𝐽 ∩ 𝐴̃
try to convince yourself of the middle equality.
Secondly, having reduced to the case 𝐵 = 𝕜[𝑌1 , … , 𝑌𝑟 ] (thus 𝑟 = 𝑛), it suffices to
pick 𝑥1 , … , 𝑥𝑛 ∈ 𝐵 such that
(a) 𝐵 is finitely generated as a module over 𝐴 ∶= 𝕜[𝑥1 , … , 𝑥𝑛 ], and
(b) 𝐼𝑗 ∩ 𝐴 ⊃ (𝑥𝑑𝑗 +1 , … , 𝑥𝑛 ) for all 𝑗.
Indeed, (a) implies Frac(𝐴) and Frac(𝐵) have the same transcendence degree 𝑛 over
𝕜, therefore 𝑥1 , … , 𝑥𝑛 must be algebraically independent over 𝕜. On the other hand,
dim(𝐵/𝐼𝑗 ) = dim(𝐴/𝐼𝑗 ∩ 𝐴) by (a) and Proposition 6.1.3; but if 𝐼𝑗 ∩ 𝐴 ⊋ (𝑥𝑑𝑗 +1 , … , 𝑥𝑛 )
then 𝐴/𝐼𝑗 ∩ 𝐴 is a proper quotient of the domain 𝕜[𝑥1 , … , 𝑥𝑑𝑗 ], therefore would have
dimension < 𝑑𝑗 by Lemma 6.3.2.
We shall construct 𝑥1 , … , 𝑥𝑛 step by step. Suppose 0 ≤ 𝑒 ≤ 𝑛 and that we have
produced elements 𝑥1′ , … , 𝑥𝑒′ and 𝑥𝑒+1 , … , 𝑥𝑛 in 𝐵 satisfying
⋅ 70 ⋅ Dimension of finitely generated algebras

(i) 𝐵 is finitely generated as a module over 𝕜[𝑥1′ , … , 𝑥𝑒′ , 𝑥𝑒+1 , … , 𝑥𝑛 ] =∶ 𝑆𝑒 ;

(ii) 𝐼𝑗 ∩ 𝑆𝑒 ⊃ (𝑥𝑒+1 , … , 𝑥𝑛 ) when 𝑑𝑗 ≤ 𝑒;

(iii) 𝐼𝑗 ∩ 𝑆𝑒 ⊃ (𝑥𝑑𝑗 +1 , … , 𝑥𝑛 ) when 𝑑𝑗 ≥ 𝑒 (with 𝑗 = 1, … , 𝑚).

For the initial case 𝑒 = 𝑛, simply take 𝑥𝑖′ ∶= 𝑌𝑖 . Our aim is 𝑒 = 0. Let us explain the
induction step from 𝑒 ≥ 1 to 𝑒 − 1. The prior argument based on transcendence degrees
implies the algebraic independence among

𝑥1′ , … , 𝑥𝑒′ , 𝑥𝑒+1 , … , 𝑥𝑛 .

If 𝑒 ≤ 𝑑𝑗 for all 𝑗, we are done. Otherwise set 𝑗 ∶= min{𝑗′ ∶ 𝑒 > 𝑑𝑗′ }, we contend that

𝐼𝑗 ∩ 𝕜[𝑥1′ , … , 𝑥𝑒′ ] ≠ {0}.

If not, we would have 𝐼𝑗 ∩ 𝑆𝑒 = (𝑥𝑒+1 , … , 𝑥𝑛 ) since 𝐼𝑗 ∩ 𝑆𝑒 ⊃ (𝑥𝑒+1 , … , 𝑥𝑛 ) by (ii). We

have dim(𝑆𝑒 /𝐼𝑗 ∩ 𝑆𝑒 ) = dim(𝐵/𝐼𝑗 ) = 𝑑𝑗 by integrality, whilst dim(𝑆𝑒 /(𝑥𝑒+1 , … , 𝑥𝑛 )) = 𝑒.
Contradiction.
Now take 𝑥𝑒 ∈ 𝐼𝑗 ∩ 𝕜[𝑥1′ , … , 𝑥𝑒′ ] ∖ {0}. Note that 𝑥𝑒 ∉ 𝕜 as 𝐼𝑗 ≠ 𝐵. By Lemma 6.3.1,
we may choose 𝑥1″ , … , 𝑥𝑒−1 ″ ∈ 𝕜[𝑥1′ , … , 𝑥𝑒′ ] such that 𝕜[𝑥1′ , … , 𝑥𝑒′ ] is finitely generated
″ ″
over 𝕜[𝑥1 , … , 𝑥𝑒−1 , 𝑥𝑒 ] as a module. It remains to verify that the new sequence
″ ,𝑥 ,…,𝑥
𝑥1″ , … , 𝑥𝑒−1 𝑒 𝑛

satisfies (i)—(iii) above with 𝑒 − 1 replacing 𝑒.

First, 𝑆𝑒 is a finitely generated module over its subalgebra 𝕜[𝑥1″ , … , 𝑥𝑛 ] by construc-
tion, hence so is 𝐵 and (i) follows. Next, let 1 ≤ 𝑗′ ≤ 𝑚. If 𝑑𝑗′ > 𝑒 − 1, the procedure
above does not affect 𝑥𝑑𝑗 +1 , … , 𝑥𝑛 , so they belong to 𝐼𝑗′ ∩ 𝕜[𝑥1″ , … , 𝑥𝑛 ]. If 𝑑𝑗′ < 𝑒, setting
𝑗 ∶= min{𝑗″ ∶ 𝑒 > 𝑑𝑗″ } we have 𝑗′ ≥ 𝑗 and

𝐼𝑗′ ∩ 𝕜[𝑥1″ , … , 𝑥𝑛 ] ⊃ 𝐼𝑗 ∩ 𝕜[𝑥1″ , … , 𝑥𝑛 ] ⊃ (𝑥𝑒+1 , … , 𝑥𝑛 ) + (𝑥𝑒 ).

All in all, we obtain (ii) and (iii).

Corollary 6.3.4 (Dimension formula). Let 𝐵 be a finitely generated algebra over a field 𝕜.
Suppose 𝐵 is a domain, then for all 𝔮 ∈ Spec(𝐵) we have

dim(𝐵/𝔮) + ht(𝔮) = dim 𝐵.

Proof. Choose a subalgebra 𝐴 of 𝐵 as in Theorem 6.3.3 and put 𝔭 = 𝔮 ∩ 𝐴. Since 𝐵 is a

domain and 𝐴 is normal, the Cohen–Seidenberg Theorem 4.1.4 asserts the going-down
property for 𝐴 ↪ 𝐵. Proposition 6.1.3 implies that dim 𝐴 = dim 𝐵, dim 𝐴/𝔭 = dim 𝐵/𝔮
and ht(𝔭) = ht(𝔮), so we are again reduced to the case 𝐵 = 𝕜[𝑋1 , … , 𝑋𝑛 ] and 𝔮 =
(𝑋𝑑+1 , … , 𝑋𝑛 ). This is known by Corollary 6.2.2.
The dimension formula allows us to compute dim 𝐵 by choosing any prime 𝔮, a
prime chain of longest length below 𝔮 (i.e. in 𝐵𝔮 ) and another one above 𝔮 (i.e. in
𝐵/𝔮); their concatenation will then be a prime chain in 𝐵 with maximal length dim 𝐵.
By applying the dimension formula to 𝔮, 𝔮′ ⊂ 𝐵 and to 𝔮′ /𝔮 ⊂ 𝐵/𝔮, it follows that
ht(𝔮′ ) = ht(𝔮) + ht(𝔮′ /𝔮) for all 𝔮′ ⊋ 𝔮 in 𝐵.
 §6.3 Noether normalization and its consequences ⋅ 71 ⋅

For a general Noetherian domain 𝐵, we call a prime chain maximal if it is not properly
contained in any prime chain. A priori, a maximal prime chain does not necessarily
have length equal to dim 𝐵. If

∀𝔮′ ⊃ 𝔮 ∶ primes, ht(𝔮′ ) = ht(𝔮) + ht(𝔮′ /𝔮),

we say 𝐵 is a catenary domain. This means that ht(𝔮′ /𝔮) = dim 𝐵𝔮′ /𝔮𝐵𝔮′ is the common
length of all maximal prime chains between 𝔮′ and 𝔮. In particular, maximal prime
chains in 𝐵𝔮′ are automatically longest. As shown above, finitely generated domains
over a field are catenary. For a finer analysis of catenary and universally catenary rings,
we refer to [11, §14].

Corollary 6.3.5. Suppose 𝐵 is a domain finitely generated over a field 𝕜. Set 𝐿 ∶= Frac(𝐵).
Then dim 𝐵 = tr.deg𝕜 (𝐿) and it is the common length of maximal prime chains.

Proof. Choose a subalgebra 𝐴 of 𝐵 as in Theorem 6.3.3. Since the field Frac(𝐵) is a

finite extension of Frac(𝐴) and dim 𝐴 = dim 𝐵 by Proposition 6.1.3, the first assertion
reduces immediately to the case 𝐵 = 𝕜[𝑋1 , … , 𝑋𝑛 ], which is obvious. As to the second
assertion, consider a maximal prime chain 𝔮0 ⊋ ⋯ ⊋ 𝔮𝑛 = {0} in 𝐵. Maximality implies
ht(𝔮𝑖 /𝔮𝑖+1 ) = 1 for all 𝑖 and dim 𝐵/𝔮0 = 0. Applying Corollary 6.3.4 repeatedly, we see

dim 𝐵 = dim 𝐵/𝔮𝑛 = dim 𝐵/𝔮𝑛−1 + ht(𝔮𝑛−1 /𝔮𝑛 ) =

𝑛−1
dim 𝐵/𝔮𝑛−2 + ht(𝔮𝑛−1 /𝔮𝑛−2 ) + ht(𝔮𝑛−1 /𝔮𝑛 ) = ⋯ = dim 𝐵/𝔮0 + ∑ ht(𝔮𝑖 /𝔮𝑖+1 )
𝑖=0

which equals 𝑛.
Remark 6.3.6. Recall that finitely generated domains over an algebraically closed field
𝕜 are objects “opposite” to the irreducible affine 𝕜-varieties. It is instructive to make a
comparison with the analytic theory when 𝕜 = ℂ. Let 𝒳 be a compact connected com-
plex manifold of (complex) dimension 𝑛. Denote by ℳ(𝒳) the field of meromorphic
functions on 𝒳. Siegel proved that tr.degℂ (ℳ(𝒳)) ≤ 𝑛. When 𝒳 is a projective alge-
braic ℂ-variety, equality holds and we have ℳ(𝒳) = Frac(𝐴) if Spec(𝐴) is any open
dense affine subscheme in 𝒳. In general, the abundance of meromorphic functions
on 𝒳 is a subtle issue, cf. the case of Riemann surfaces (𝑛 = 1). Compact connected
complex manifolds with tr.degℂ (ℳ(𝒳)) = dimℂ 𝒳 are called Moishezon manifolds.
Non-algebraic Moishezon manifolds do exist, and Moishezon proved that a Moishe-
zon manifold is projective, hence algebraic, if and only if it is Kähler. Following M.
Artin and D. Knutson, one can enlarge the category of ℂ-schemes into that of alge-
braic spaces over ℂ, and there is an analytification functor 𝒳 ↦ 𝒳an that sends algebraic
spaces of finite type over ℂ to complex analytic varieties. M. Artin [2, §7] showed that
the analytification establishes an equivalence between the category of smooth proper
algebraic spaces of finite type over ℂ and that of Moishezon manifolds. Therefore,
such manifolds still retain an algebraic flavor: they are quotients of certain ℂ-schemes
by étale equivalence relations.
⋅ 72 ⋅ Dimension of finitely generated algebras
 Lecture 7

Serre's criterion for

normality and depth
References: [8, §11] and [3, X.1]. Except in the last section, we will try to avoid the use
of depth as in [8].

7.1 Review of discrete valuation rings

Let 𝑅 be an integral domain.
Definition 7.1.1. A discrete valuation on a field 𝐾 is a surjective map 𝑣 ∶ 𝐾 × → ℤ such
that
⋄ 𝑣(𝑥𝑦) = 𝑣(𝑥) + 𝑣(𝑦), and
⋄ 𝑣(𝑥 + 𝑦) ≥ min{𝑣(𝑥), 𝑣(𝑦)}
for all 𝑥, 𝑦 ∈ 𝐾 × , where we set 𝑣(0) = +∞ for convenience. We say a domain 𝑅 with
𝐾 ∶= Frac(𝑅) is a discrete valuation ring, abbreviated as DVR, if there exists a discrete
valuation 𝑣 such that
𝑅 = 𝑣−1 ([0, +∞]).
We say 𝑡 ∈ 𝑅 is a uniformizer if 𝑣(𝑡) = 1.
It follows immediately that 𝑅× = 𝑣−1 (0). Uniformizers are unique up to 𝑅× . Note
that 𝑣(𝐾 × ) = ℤ implies that 𝑅 cannot be a field.
Example 7.1.2. The ring ℤ𝑝 of 𝑝-adic integers (𝑝: prime number) together with the
usual 𝑝-adic valuations are standard examples of DVR. The algebra of formal power
series 𝕜J𝑋K are also DVR: the valuation of ∑𝑛 𝑎𝑛 𝑋 𝑛 is the smallest 𝑛 such that 𝑎𝑛 ≠ 0.
More generally, in the geometric context, discrete valuations can be defined by look-
ing at the vanishing order of rational/meromorphic functions along subvarieties of
codimension one with suitable regularities.
Lemma 7.1.3. Let 𝑅 be a discrete valuation ring with valuation 𝑣 and uniformizer 𝑡. Every
ideal 𝔞 ≠ {0} of 𝑅 has the form (𝑡𝑟 ) for a unique 𝑟 ≥ 0. In particular, 𝑅 is a local principal ideal
domain which is not a field, hence is of dimension 1.
⋅ 74 ⋅ Serre’s criterion for normality and depth

Proof. Take 𝑟 ∶= min{𝑣(𝑥) ∶ 𝑥 ∈ 𝔞}.

In the exercises below, we assume 𝑅 is a discrete valuation ring with valuation 𝑣.

Exercise 7.1.4. Show that 𝑡 is a uniformizer if and only if it generates the maximal ideal
of 𝑅.

Exercise 7.1.5. Reconstruct 𝑣 from the ring-theoretic structure of 𝑅.

Recall that a regular local ring 𝑅 with dim 𝑅 = 1 is a Noetherian local ring whose
maximal ideal 𝔪 is principal and nonzero; elements generating 𝔪 are called the regular
parameters for 𝑅.

Proposition 7.1.6. Suppose 𝑡 is a regular parameter in a regular local ring 𝑅 of dimension one,
then 𝑅 is a domain, and every element 𝑥 ∈ 𝑅 ∖ {0} can be uniquely written as 𝑥 = 𝑡𝑟 𝑢 with
𝑟 ≥ 0 and 𝑢 ∈ 𝑅× . This makes 𝑅 into a discrete valuation ring by setting 𝑣(𝑥) = 𝑟, for which
𝑡 is a uniformizer.
Therefore 𝑅 is a discrete valuation ring. Conversely, every discrete valuation ring is regular
local of dimension 1.

Proof. From Theorem 5.5.6 we know regular local rings are Noetherian domains. By
applying Krull’s Intersection Theorem (Corollary 4.5.10) to the powers of (𝑡), we see
that 𝑟 ∶= sup{𝑘 ≥ 0 ∶ 𝑥 ∈ (𝑡)𝑘 } is finite. Write 𝑥 = 𝑡𝑟 𝑢. Since 𝑅× = 𝑅 ∖ 𝔪, we see 𝑢 ∈ 𝑅× .
As to uniqueness, suppose 𝑡𝑟 𝑢 = 𝑡𝑠 𝑤 with 𝑟 ≥ 𝑠, then 𝑡𝑟−𝑠 = 𝑢−1 𝑤 ∈ 𝑅× implies 𝑟 = 𝑠,
hence 𝑢 = 𝑤 as 𝑅 is a domain. As every element of Frac(𝑅)× is uniquely expressed
as 𝑡𝑟 𝑢 with 𝑟 ∈ ℤ, one readily checks that 𝑣(𝑡𝑟 𝑢) = 𝑟 satisfies all the requirements of
discrete valuation.
The converse direction has been addressed in Lemma 7.1.3.
To recap, in dimension one we have

regular local ring ⟺ discrete valuation ring.

This will be related to normality later on.

Exercise 7.1.7. Explain that the regular local rings of dimension 0 are just fields.

7.2 Auxiliary results on the total fraction ring

Let 𝑅 be a ring, henceforth assumed Noetherian. If there exist a non zero-divisor 𝑡 ∈ 𝑅
and 𝔭 ∈ Ass(𝑅/(𝑡)), we say 𝔭 is associated to a non zero-divisor.

Lemma 7.2.1. Let 𝑀 be a finitely generated 𝑅-module. An element 𝑥 ∈ 𝑀 is zero if and only
if its image in 𝑀𝔭 is zero for every maximal element 𝔭 in Ass(𝑀).

Proof. Suppose 𝑥 ≠ 0. Since 𝑀 is Noetherian, among ideals of the form ann(𝑦) there is
a maximal one containing ann(𝑥), and we have seen in Lemma 2.2.3 that such an ideal
𝔭 belongs to Ass(𝑀). Since ann(𝑥) ⊂ 𝔭, we have 𝑥/1 ∈ 𝑀𝔭 ∖ {0}.
Call a ring reduced if it has no nilpotent element except zero.
 §7.2 Auxiliary results on the total fraction ring ⋅ 75 ⋅

Lemma 7.2.2. Suppose 𝑅 is reduced, then Ass(𝑅) consists of minimal primes.

Proof. As 𝑅 is reduced, {0} = √0𝑅 is the intersection of minimal prime ideals 𝔭1 , 𝔭2 , …

(all lying in Ass(𝑅), hence finite in number). By the theory of primary decompositions,
one infers that Ass(𝑅) = {𝔭1 , …}.

For the next result, we denote by 𝑇 the set of non zero-divisors of 𝑅. Recall that the
total fraction ring 𝐾(𝑅) is 𝑅[𝑇 −1 ]; this is the largest localization such that 𝑅 → 𝐾(𝑅) is
injective, and 𝐾(𝑅) = Frac(𝑅) when 𝑅 is a domain. The map 𝔭 ↦ 𝔭𝐾(𝑅) sets up an
∼
order-preserving bijection Ass(𝑅) → Ass(𝐾(𝑅)): indeed, if 𝔭 ∋ 𝑡 for some 𝑡 ∈ 𝑇, then
𝔭 cannot belong to Ass(𝑅) because the union of Ass(𝑅) equals 𝑅 ∖ 𝑇.
∼
Lemma 7.2.3. Let 𝑅 be reduced. Then 𝐾(𝑅) → ∏𝔭 𝐾(𝑅/𝔭) as 𝑅-algebras, where 𝔭 ranges
over the minimal prime ideals of 𝑅. For any multiplicative subset 𝑆 ⊂ 𝑅 there is a canonical
isomorphism of 𝑅[𝑆−1 ]-algebras
∼
𝐾(𝑅[𝑆−1 ]) → 𝐾(𝑅)[𝑆−1 ].

In other words, the formation of total fraction ring commutes with localizations.

Proof. Each element in 𝐾(𝑅) = 𝑅[𝑇 −1 ] is either a zero-divisor or invertible. The set
of zero-divisors of 𝐾(𝑅) is the union of minimal prime ideals 𝔭𝑖 𝐾(𝑅) of 𝐾(𝑅) (where
Ass(𝑅) = {𝔭1 , … , 𝔭𝑚 } by an earlier discussion), therefore each prime ideal of 𝐾(𝑅) must
equal some 𝔭𝑖 𝐾(𝑅), by prime avoidance (Proposition 1.1.5). Hence 𝔭1 𝐾(𝑅), … , 𝔭𝑚 𝐾(𝑅)
are also the maximal ideals in 𝐾(𝑅), with zero intersection. Chinese Remainder The-
𝑚
orem entails that 𝐾(𝑅) ≃ ∏𝑖=1 𝐾(𝑅)/𝔭𝑖 𝐾(𝑅). To conclude the first part, notice that
𝐾(𝑅)/𝔭𝑖 𝐾(𝑅) = (𝑅/𝔭𝑖 )[𝑇 −1 ]; this is a field in generated by an isomorphic copy of 𝑅/𝔭𝑖
since 𝔭𝑖 ∩ 𝑇 = ∅, hence equals Frac(𝑅/𝔭𝑖 ).
As for the second part, one decomposes 𝐾(𝑅[𝑆−1 ]) and 𝐾(𝑅)[𝑆−1 ] by the previous
step, noting that
⋄ 𝑅[𝑆−1 ] is reduced;
⋄ Ass(𝑅[𝑆−1 ]) = {𝔭𝑖 𝑅[𝑆−1 ] ∶ 1 ≤ 𝑖 ≤ 𝑚, 𝔭𝑖 ∩ 𝑆 = ∅} consists of minimal primes;
⋄ 𝐾 (𝑅[𝑆−1 ]/𝔭𝑖 𝑅[𝑆−1 ]) ≃ 𝐾(𝑅/𝔭𝑖 ) = 𝐾(𝑅/𝔭𝑖 )[𝑆−1 ] when 𝔭𝑖 ∩ 𝑆 = ∅, by the argu-
ments above;
⋄ 𝐾(𝑅/𝔭𝑖 )[𝑆−1 ] = {0} when 𝔭𝑖 ∩ 𝑆 ≠ ∅.
A term-by-term comparison finishes the proof.

We use Lemma 7.2.3 to interchange 𝐾(⋅) and localizations in what follows.

Lemma 7.2.4. Suppose 𝑅 is reduced. Then 𝑥 ∈ 𝐾(𝑅) belongs to 𝑅 if and only if its image in
𝐾(𝑅)𝔭 = 𝐾(𝑅𝔭 ) belongs to 𝑅𝔭 for every prime 𝔭 associated to a non zero-divisor.

Proof. Only the “if” direction requires a proof. Write 𝑥 = 𝑎/𝑡 with 𝑡 not a zero-divisor.
Suppose that 𝑎 ∉ (𝑡), i.e. 𝑎 does not map to zero in 𝑅/(𝑡). Lemma 7.2.1 asserts there
exists 𝔭 ∈ Ass(𝑅/(𝑡)) such that 𝑎 does not map to 0 ∈ (𝑅/(𝑡))𝔭 = 𝑅𝔭 /𝑡𝑅𝔭 . It follows
that the image of 𝑎/𝑡 in 𝐾(𝑅𝔭 ) does not lie in 𝑅𝔭 .
⋅ 76 ⋅ Serre’s criterion for normality and depth

7.3 On normality
Fix a Noetherian ring 𝑅.

Exercise 7.3.1. Suppose 𝑅 is a domain, 𝐾 ∶= Frac(𝑅). If 𝑅 = ⋂𝑖∈𝐼 𝑅𝑖 where {𝑅𝑖 ⊂ 𝐾}𝑖∈𝐼

are subrings such that Frac(𝑅𝑖 ) = 𝐾 and 𝑅𝑖 is normal, for each 𝑖, then 𝑅 is normal.

Proposition 7.3.2. Let 𝑅 be a Noetherian domain. Then 𝑅 is normal if and only if for every
principal ideal (𝑡) ⊂ 𝑅 and every 𝔭 ∈ Ass(𝑅/(𝑡)), the ideal 𝔭𝑅𝔭 is principal.

Proof. Assume the conditions above. To prove the normality of 𝑅, it suffices to use
𝑅 = ⋂ 𝑅𝔭 where 𝔭 ranges over the primes associated to nonzero principal ideals (con-
sequence of Lemma 7.2.4). Indeed, each 𝑅𝔭 is regular, hence normal by Proposition
7.1.6, therefore so is their intersection by the previous exercise.
Conversely, assume 𝑅 is normal and let 𝔭 ∈ Ass(𝑅/(𝑡)) with 𝑡 ≠ 0, we have to show
𝔭𝑅𝔭 is principal. Upon replacing 𝑅 by 𝑅𝔭 and recalling how associated primes behave
under localization, we may even assume 𝑅 is local with maximal ideal 𝔭. Express 𝔭 as
the annihilator of some 𝑥 ̄ ∈ 𝑅/(𝑡) with 𝑥 ∈ 𝑅. Define the fractional ideal

𝔭−1 ∶= {𝑦 ∈ Frac(𝑅) ∶ 𝑦𝔭 ⊂ 𝑅}.

It is an 𝑅-submodule of Frac(𝑅) containing 𝑅. Define the 𝑅-submodule 𝔭−1 𝔭 of Frac(𝑅)

in the obvious way. Clearly 𝔭 ⊂ 𝔭−1 𝔭 ⊂ 𝑅. By maximality of 𝔭, exactly one of the
⊂ is equality. If 𝔭𝔭−1 = 𝔭, every element of 𝔭−1 is integral over 𝑅, hence 𝔭−1 ⊂ 𝑅 by
normality (integrality is “witnessed” by the module 𝔭). From 𝔭𝑥 ⊂ (𝑡) we see 𝑥/𝑡 ∈
𝔭−1 = 𝑅; this would imply 𝑥 ̄ = 0 and 𝔭 = 𝑅, which is absurd.
Therefore we must have 𝔭𝔭−1 = 𝑅. This implies that 𝔭𝑦 ⊄ 𝔭 for some 𝑦 ∈ 𝔭−1 , hence
𝔭𝑦 = 𝑅 since 𝑅 is local. Hence 𝔭 = 𝑦−1 𝑅 ≃ 𝑅 is principal.

Corollary 7.3.3. The following are equivalent for a local domain 𝑅:

(i) 𝑅 is normal of dimension 1;

(ii) 𝑅 is a regular local ring of dimension 1;

(iii) 𝑅 is a discrete valuation ring.

Proof. (iii) ⟹ (i). We have seen that discrete valuation rings are principal ideal rings
of dimension 1, therefore also normal by unique factorization property.
(i) ⟹ (ii). Under the normality assumption, choose any 𝑡 ∈ 𝑅 ∖ {0}. Since
dim 𝑅 = 1 and {0} is a prime ideal, the associated prime of (𝑡) can only be the maximal
ideal 𝔪, which is principal by Proposition 7.3.2. This shows that 𝑅 is regular local.
(ii) ⟹ (iii) is included in Proposition 7.1.6.

Corollary 7.3.4. Let 𝑅 be a Noetherian normal domain. Then

𝑅= ⋂ 𝑅𝔭
ht(𝔭)=1

inside Frac(𝑅).
 §7.4 Serre’s criterion ⋅ 77 ⋅

Proof. Evidently ⊂ holds. By Lemma 7.2.4 together with Proposition 7.3.2, 𝑅 can be
written as an intersection of 𝑅𝔭 where 𝔭 is associated to some non zero-divisor, such
that 𝔭𝑅𝔭 is principal; it suffices to show ht(𝔭) = 1. From 𝔭 ≠ {0} we see ht(𝔭) ≥ 1; on
the other hand, by Hauptidealsatz or by the discussion on regular local rings, we see
ht(𝔭) = ht(𝔭𝑅𝔭 ) ≤ 1.

7.4 Serre’s criterion

Lemma 7.4.1. Let 𝑅 be a Noetherian ring. Suppose that
⋄ the primes in Ass(𝑅) are all minimal, and
⋄ 𝑅𝔭 is a field for every minimal prime ideal 𝔭,
then 𝑅 is reduced.

Proof. Take a minimal primary decomposition {0} = 𝐼1 ∩ ⋯ ∩ 𝐼𝑚 with Ass(𝑅/𝐼𝑗 ) =

{𝔭𝑗 = √𝐼𝑗 } and Ass(𝑅) = {𝔭1 , … , 𝔭𝑚 }. By the properties of primary ideals, 𝔭𝑗 = √𝐼𝑗 ⊃ 𝐼𝑗
for all 𝑗. By assumption each 𝔭𝑗 is minimal, and 𝑅𝔭𝑗 is a field. From the uniqueness of
non-embedded components in primary decompositions, 𝐼𝑗 = ker [𝑅 → 𝑅𝔭𝑗 ] is a prime
𝑚
contained in 𝔭𝑗 , hence 𝐼𝑗 = 𝔭𝑗 . We deduce that {0} = ⋂𝑗=1 𝔭𝑗 , thereby showing √0𝑅 =
{0}.

Theorem 7.4.2 (J.-P. Serre). A Noetherian ring 𝑅 is a finite direct product of normal domains
if and only if the following two conditions hold.

⊳ R1 The localization of 𝑅 at every prime ideal of height 1 (resp. 0) is a discrete valuation

ring (resp. a field).

⊳ S2 For every non zero-divisor 𝑡 of 𝑅, the primes in Ass(𝑅/(𝑡)) are all of height 1; the
primes in Ass(𝑅) are all of height 0.

The condition R1 means regularity in codimension ≤ 1. The condition S2 is often

rephrased in terms of depth, which will be discussed in Proposition 7.5.11.
Proof. We begin with the ⟹ direction. Suppose 𝑅 = 𝑅1 × ⋯ × 𝑅𝑛 where each 𝑅𝑖 is
𝑛
a normal domain. As is well-known, Spec(𝑅) = ⨆𝑖=1 Spec(𝑅𝑖 ) as topological spaces:
to be precise, the elements of Spec(𝑅) take the form 𝔭 = 𝑅1 × ⋯ × 𝔭𝑖 × ⋯ × 𝑅𝑛 , where
𝔭𝑖 ∈ Spec(𝑅𝑖 ). We have
ht(𝔭) = ht(𝔭𝑖 ), 𝑅𝔭 ≃ (𝑅𝑖 )𝔭𝑖 .
Furthermore, one easily checks that
𝑛
Ass(𝑅/(𝑡)) = ⨆ Ass(𝑅𝑖 /(𝑡𝑖 )), 𝑡 = (𝑡1 , … , 𝑡𝑛 ) ∈ 𝑅
𝑖=1

compatibly with the description above.

This reduces the verification of S2 to the case of normal domains, which is ad-
dressed in Proposition 7.3.2. The condition R1 is implied by Corollary 7.3.3 since nor-
mality is preserved under localizations.
⋅ 78 ⋅ Serre’s criterion for normality and depth

Assume conversely R1 and S2. They imply the conditions of Lemma 7.4.1, hence 𝑅
is reduced. Now for every prime 𝔭 associated to a non zero-divisor, we have ht(𝔭) = 1
and 𝑅𝔭 is a normal domain by R1 ∧ S2. By Lemma 7.2.4 (as 𝑅 is reduced), 𝑅 is integrally
closed in 𝐾(𝑅): indeed, if 𝑥 ∈ 𝐾(𝑅) is integral over 𝑅, so is its image in 𝐾(𝑅𝔭 ) =
𝑚
𝐾(𝑅)𝔭 for every 𝔭 as above, therefore lies in 𝑅𝔭 . Decompose 𝐾(𝑅) = ∏𝑖=1 𝐾(𝑅/𝔭𝑖 ) as in
Lemma 7.2.3. The idempotents 𝑒𝑖 ∈ 𝐾(𝑅) associated to this decomposition are trivially
integral over 𝑅: 𝑒𝑖2 − 𝑒𝑖 = 0, hence 𝑒𝑖 ∈ 𝑅 for all 𝑖. It follows that 𝑅 = 𝑅𝑒1 + ⋯ +
𝑚
𝑅𝑒𝑚 ∏𝑖=1 𝑅𝑒𝑖 ⊂ 𝐾(𝑅) and one easily checks that 𝑅𝑒𝑖 = 𝑅/𝔭𝑖 .
Finally, since 𝑅 is integrally closed in 𝐾(𝑅), the decomposition above implies 𝑅/𝔭𝑖 is
integrally closed in 𝐾(𝑅/𝔭𝑖 ). All in all, we have written 𝑅 as a direct product of normal
domains.

Exercise 7.4.3. Recall that for an 𝑅-module 𝑀, a prime ideal 𝔭 ∈ Ass(𝑀) is called embed-
ded if 𝔭 is not a minimal element in Ass(𝑀). Show that for 𝑀 = 𝑅, embedded primes
are primes in Ass(𝑅) with height > 0. For 𝑀 = 𝑅/(𝑡) where 𝑡 is not a zero-divisor,
embedded primes are primes in Ass(𝑅/(𝑡)) with height > 1. Use this to rephrase S2
as follows: there are no embedded primes in Ass(𝑅/(𝑡)) (𝑡 not a zero-divisor) or in
Ass(𝑅).

Exercise 7.4.4. Suppose a ring 𝑅 is isomorphic to a direct product ∏𝑖∈𝐼 𝑅𝑖 . Show that
𝑅 is a domain if and only if |𝐼| = 1 (say 𝐼 = {𝑖0 }), and 𝑅𝑖0 is a domain.

Corollary 7.4.5. A Noetherian domain 𝑅 is normal if and only if R1 and S2 hold for 𝑅.

Proof. Immediate from the previous exercise and Theorem 7.4.2.

7.5 Introduction to depth

Based on [3], we give a brief account on the notion of depth. Let 𝑅 be a ring and 𝑀 be
an 𝑅-module, 𝑀 ≠ {0}. Recall the Ext-functors
𝑛
Ext𝑅 (𝑋, 𝑌) ∶= 𝐻 𝑛 (𝑅ℋom(𝑋, 𝑌)) = Hom𝐷+ (𝑅-Mod) (𝑋, 𝑌[𝑛]).

Definition 7.5.1 (Depth of a module). Let 𝐼 be a proper ideal of 𝑅. We define the depth
of 𝑀 relative to 𝐼 as
𝑛
depth𝐼 (𝑀) ∶= inf {𝑛 ≥ 0 ∶ Ext𝑅 (𝑅/𝐼, 𝑀) ≠ 0}

with values in ℤ≥0 ⊔ {+∞}.

Proposition 7.5.2. For 𝐼, 𝑀 as above, the following are equivalent:

(i) depth𝐼 (𝑀) = 0;

𝑥
(ii) for all 𝑥 ∈ 𝐼, the homomorphism 𝑀 𝑀 is not injective;

(iii) Ass(𝑀) ∩ 𝑉(𝐼) ≠ ∅.

 §7.5 Introduction to depth ⋅ 79 ⋅

𝑥
Proof. In each case we have 𝑀 ≠ {0}. If (i) holds, then 𝑀 𝑀 vanishes on the image
of some nonzero 𝑅/𝐼 → 𝑀, hence (ii). If (ii) holds, the union of Ass(𝑀) will cover 𝐼,
and (iii) follows by prime avoidance. Finally, suppose 𝔭 ∈ Ass(𝑀) ∩ 𝑉(𝐼), there is an
embedding 𝑅/𝔭 ↪ 𝑀, which yields a non-zero 𝑅/𝐼 → 𝑀.

Definition 7.5.3. A sequence 𝑥1 , … , 𝑥𝑛 ∈ 𝑅 is called an 𝑀-regular sequence of length 𝑛

if (𝑥1 , … , 𝑥𝑛 )𝑀 ⊊ 𝑀 and

𝑥𝑘
0 → 𝑀/(𝑥1 , … , 𝑥𝑘−1 ) 𝑀/(𝑥1 , … , 𝑥𝑘−1 )

is exact for all 1 ≤ 𝑘 ≤ 𝑛.

Lemma 7.5.4. Let 𝑀 be an 𝑅-module, 𝑥1 , … , 𝑥𝑟 be an 𝑀-regular sequence lying in an ideal

𝐼 ⊊ 𝑅. We have depth𝐼 (𝑀) = 𝑟 + depth𝐼 (𝑀/(𝑥1 , … , 𝑥𝑟 )𝑀).

Proof. The case 𝑟 = 1 follows by staring at the long exact sequence attached to 0 →
𝑥1
𝑀 𝑀 → 𝑀/𝑥1 𝑀 → 0. The general case follows by induction on 𝑟.

Theorem 7.5.5. Assume 𝑅 Noetherian, 𝑀 finitely generated and 𝐼 ⊊ 𝑅.

(i) depth𝐼 (𝑀) is the supremum of the lengths of 𝑀-regular sequences with elements in 𝐼.

(ii) Suppose depth𝐼 (𝑀) < +∞. Every 𝑀-regular sequence with elements in 𝐼 can be ex-
tended to one of length depth𝐼 (𝑀).

(iii) The depth of 𝑀 relative to 𝐼 is finite if and only if 𝑉(𝐼) ∩ Supp(𝑀) ≠ ∅, or equivalently
𝐼𝑀 ≠ 𝑀.

Proof. To prove (i) and (ii), by the previous Lemma we are reduced to show that
depth𝐼 (𝑀) > 0 implies the existence of 𝑥 ∈ 𝐼 which is not a zero-divisor of 𝑀; this
follows from Proposition 7.5.2.
Now pass to the word “equivalently” in (iii). We have 𝑀 ≠ 𝐼𝑀 if and only if
(𝑀/𝐼𝑀)𝔭 = 𝑀𝔭 /𝐼𝔭 𝑀𝔭 ≠ 0 for some prime ideal 𝔭. That quotient always vanishes when
𝔭 ⊅ 𝐼, in which case 𝐼𝔭 = 𝑅𝔭 . On the other hand, when 𝔭 ∈ 𝑉(𝐼) we have 𝐼𝔭 ⊂ rad(𝑅𝔭 ),
thus the non-vanishing is equivalent to 𝑀𝔭 ≠ {0} by Nakayama’s Lemma 1.3.5.
The “if” direction of (iii) is based on the following fact

𝐼𝑀 ≠ 𝑀 ⟹ depth𝐼 (𝑀) < +∞

which will be proved in the next lecture (Theorem 8.3.2) using Koszul complexes. As for
the “only if” direction, 𝑉(𝐼) ∩ Supp(𝑀) = ∅ implies 𝐼 + ann(𝑀) = 𝑅, but the elements
𝑛
in 𝐼 + ann(𝑀) annihilate each Ext (𝑅/𝐼, 𝑀), hence depth𝐼 (𝑀) = +∞.

Corollary 7.5.6. With the same assumptions, let (𝑥1 , … , 𝑥𝑟 ) be an 𝑀-regular sequence with
𝑥𝑖 ∈ 𝐼. It is of length depth𝐼 (𝑀) if and only if Ass(𝑀/(𝑥1 , … , 𝑥𝑟 )𝑀) ∩ 𝑉(𝐼) ≠ ∅.

Proof. The sequence has length depth𝐼 (𝑀) if and only if 𝑅-module 𝑀/(𝑥1 , … , 𝑥𝑟 )𝑀 has
depth zero, so it remains to apply Proposition 7.5.2.
⋅ 80 ⋅ Serre’s criterion for normality and depth

Jean-Louis Koszul (1921—2018) created the Koszul complexes in

order to define a cohomology theory for Lie algebras; this device
turns out to be a general, convenient construction in homological
algebra, which will be discussed in the next lecture. The study of
“Koszulness” in a broader (eg. operadic) context is now an active
area of research. J.-L. Koszul is also a second-generation member
of the Bourbaki group. Source: by Konrad Jacobs - Oberwolfach
Photo Collection, ID 2273.

Corollary 7.5.7. Let 𝑅 be a Noetherian local ring with maximal ideal 𝔪, and 𝑀 ≠ {0} a finitely
generated 𝑅-module, then depth𝔪 (𝑀) ≤ dim 𝑀.

Proof. Consider the following situation: 𝑥 ∈ 𝔪 is not a zero-divisor for 𝑀 ≠ {0}. In

the discussion of dimensions, we have seen that 𝑑(𝑀) ≥ 𝑑(𝑀/𝑥𝑀) ≥ 𝑑(𝑀/𝑥𝑀) − 1,
where 𝑑(⋅) is the degree of Hilbert–Samuel polynomial; on the other hand, since the
𝑥
alternating sum of Hilbert–Samuel polynomials in 0 → 𝑀 𝑀 → 𝑀/𝑥𝑀 → 0 has
degree < 𝑑(𝑀), we infer that 𝑑(𝑀/𝑥𝑀) = 𝑑(𝑀)−1. Hence dim(𝑀/𝑥𝑀) = dim(𝑀)−1.
By relating depth to 𝑀-regular sequences, we deduce depth𝔪 (𝑀) ≤ dim 𝑀.

Definition 7.5.8 (Cohen–Macaulay modules). Let 𝑅 be a Noetherian ring. A finitely

generated 𝑅-module 𝑀 is called Cohen–Macaulay if

depth𝔪𝐴 (𝑀𝔪 ) = dim 𝑀𝔪

𝔪

for every 𝔪 ∈ MaxSpec(𝑅) ∩ Supp(𝑀). We say 𝑅 is a Cohen–Macaulay ring if it is

Cohen–Macaulay as a module.

Example 7.5.9. Regular local rings are Cohen–Macaulay, although we do not prove this
here. Another important class of Cohen–Macaulay rings is the algebra of invariants
𝐴𝐺 where 𝐴 is the algebra of regular functions on an affine 𝕜-variety 𝒳 with rational
singularities (eg. 𝐴 = 𝕜[𝑋1 , … , 𝑋𝑛 ]) with an action by a reductive 𝕜-group 𝐺 (finite
 §7.5 Introduction to depth ⋅ 81 ⋅

groups allowed), and we assume char(𝕜) = 0. Here is the reason: Boutot [6] proved
the GIT quotient 𝒳//𝐺 has rational singularities as well, hence is Cohen–Macaulay;
in characteristic zero this strengthens an earlier theorem of Hochster–Roberts. These
algebras are interesting objects from the geometric, algebraic or even combinatorial
perspectives.
Exercise 7.5.10. Show by using Proposition 7.5.2 that depth𝔪 (𝑅) = 0 if 𝑅 is local with
maximal ideal 𝔪 and has dimension zero.
Proposition 7.5.11. The condition S2 in Theorem 7.4.2 is equivalent to

ht(𝔭) ≥ 𝑖 ⟹ depth𝔭𝑅 (𝑅𝔭 ) ≥ 𝑖

𝔭

for all 𝔭 ∈ Spec(𝑅) and 𝑖 ∈ {1, 2}.

Proof. Assume the displayed conditions. We first show that every 𝔭 ∈ Ass(𝑅) has
height zero. Upon localization we may assume 𝑅 local with maximal ideal 𝔭, thus 𝔭
has depth zero by Proposition 7.5.2; our conditions force ht(𝔭) = 0. Next, consider 𝔭 ∈
Ass(𝑅/(𝑡)) with 𝑡 non zero-divisor. Note that 𝑡 remains a non zero-divisor in 𝑅𝔭 , and
𝔭𝑅𝔭 ∈ Ass(𝑅𝔭 /(𝑡)). Hence depth(𝑅𝔭 ) = depth(𝑅𝔭 /(𝑡)) + 1 = 1 since depth𝔭 (𝑅𝔭 /(𝑡)) =
0 by Proposition 7.5.2. Our conditions force ht(𝔭) ≤ 1 and 𝑡 ∈ 𝔭 implies ht(𝔭) > 0.
Conversely, assume S2. Suppose ht(𝔭) ≥ 1. If 𝑅𝔭 has depth zero then Ass(𝑅𝔭 ) ∩
𝑉(𝔭𝑅𝔭 ) ≠ ∅, which implies 𝔭𝑅𝔭 ∈ Ass(𝑅𝔭 ) thus 𝔭 ∈ Ass(𝑅), contradicting S2. Next,
suppose ht(𝔭) ≥ 2. If 𝑅𝔭 has depth ≤ 1, the standard property
𝑖 𝑖
∀𝑖 ≥ 0, Ext𝑅𝔭 (𝑋𝔭 , 𝑌𝔭 ) ≃ Ext𝑅 (𝑋, 𝑌)𝔭

valid for Noetherian 𝑅 and finitely generated 𝑅-modules 𝑋, 𝑌, yields

0 ≤ depth𝔭 (𝑅) ≤ depth𝔭𝑅 (𝑅𝔭 ) ≤ 1.

𝔭

If depth𝔭 (𝑅) = 0, there exists of 𝔭′ ⊃ 𝔭 and 𝔭′ ∈ Ass(𝑅) by Proposition 7.5.2, but S2

implies ht(𝔭) ≤ ht(𝔭′ ) = 0 which is absurd. If depth𝔭 (𝑅) = 1, there exists a non zero-
divisor 𝑥 in 𝑅 such that depth𝔭 (𝑅/(𝑥)) = 0. The same argument furnishes 𝔭′ ⊃ 𝔭 such
that 𝔭′ ∈ Ass(𝑅/(𝑥)). Now S2 implies ht(𝔭) ≤ ht(𝔭′ ) = 1, again a contradiction.
Now the conditions R1 and S2 can be generalized to arbitrary 𝑘 ∈ ℤ≥0 :
⊳ Rk ht(𝔭) ≤ 𝑘 implies 𝑅𝔭 is a regular local ring, for every 𝔭 ∈ Spec(𝑅);

⊳ Sk depth(𝑅𝔭 ) ≥ min{𝑘, ht(𝔭)} for every 𝔭 ∈ Spec(𝑅).

One readily checks their compatibility with Proposition 7.5.11. Note that S0 is trivial,
whilst 𝑅 is Cohen–Macaulay if and only if it satisfies Sk for all 𝑘.
Example 7.5.12. From Theorem 7.4.2, we see that any finite direct product of normal
domains of dimension ≤ 2 is Cohen–Macaulay.
Exercise 7.5.13. Show that S1 holds if and only if there are no embedded primes in
Ass(𝑅).
⋅ 82 ⋅ Serre’s criterion for normality and depth
 Lecture 8

Some aspects of Koszul

complexes
This lecture is a faithful replay of the relevant sections of [4] and [3]; the main aim here
is to complete the proof of Theorem 7.5.5. In what follows, we work with a chosen ring
𝑅. We impose no Noetherian or finiteness conditions here.

8.1 Preparations in homological algebra

For any 𝑅-module 𝐿, denote its exterior algebra over 𝑅 by
𝑛
⋀ 𝐿 ∶= ⨁ ⋀ 𝐿.
𝑛≥0

It is the quotient of the tensor algebra 𝑇(𝐿) = ⨁𝑛≥0 (𝑇 𝑛 (𝐿) ∶= 𝐿⊗𝑛 ) by the graded ideal
generated by the pure tensors
⋯ ⊗ 𝑥 ⊗ 𝑥 ⊗ ⋯, 𝑥 ∈ 𝐿.
0
The multiplication operation in ⋀ 𝐿 is written as ∧. Note that ⋀ 𝐿 = 𝑇 0 (𝐿) = 𝑅
by convention. The traditional notion of exterior algebras encountered in differential
geometry is recovered when ℚ ⊂ 𝑅.
𝑛+1
Given 𝑢 ∈ Hom𝑅 (𝐿, 𝑅), one can define the corresponding contractions 𝑖𝑢 ∶ ⋀ 𝐿→
𝑛
⋀ 𝐿, given concretely as
𝑛
𝑖𝑢 (𝑥0 ∧ ⋯ ∧ 𝑥𝑛 ) = ∑(−1)𝑖 𝑢(𝑥𝑖 ) ⋅ 𝑥0 ∧ ⋯ 𝑥̂𝑖 ⋯ ∧ 𝑥𝑛
𝑖=0
where 𝑥̂𝑖 means 𝑥𝑖 is omitted. It is routine to check that 𝑖𝑢 satisfies 𝑖𝑢 ∘ 𝑖𝑢 = 0, thereby
giving rise to a chain complex.
Definition 8.1.1. Let 𝐿 and 𝑢 be as above. Define the corresponding Koszul complex as
•
𝐾• (𝑢) ∶= (⋀ 𝐿, 𝑖𝑢 ). For any 𝑅-module 𝑀, put
𝐾• (𝑢; 𝑀) ∶= 𝑀 ⊗ 𝐾• (𝑢),
𝑅
𝐾 • (𝑢; 𝑀) ∶= Hom𝐴 (𝐾• (𝑢), 𝑀)
⋅ 84 ⋅ Some aspects of Koszul complexes

which is naturally a chain (resp. cochain) complex in positive degrees; here one re-
gards 𝑀 as a complex in degree zero. These definitions generalize to the case of any
complex 𝑀, and are functorial in 𝑀.
The reader might have encountered the following result in differential geometry.
𝑛
Proposition 8.1.2 (Homotopy formula). For any 𝑥 ∈ 𝐿 and 𝜔 ∈ ⋀ 𝐿, we have

𝑖𝑢 (𝑥 ∧ 𝜔) + 𝑥 ∧ (𝑖𝑢 (𝜔)) = 𝑢(𝑥)𝜔.

Proof. Consider 𝜔 = 𝑥1 ∧ ⋯ ∧ 𝑥𝑛 . Put 𝑥0 ∶= 𝑥. The left-hand side equals

𝑛
∑(−1)𝑖 ⋅ 𝑢(𝑥𝑖 ) ⋯ ∧ 𝑥̂𝑖 ∧ ⋯
𝑖=0

whereas the right-hand side equals

𝑛
∑(−1)𝑖+1 ⋅ 𝑢(𝑥𝑖 )𝑥0 ∧ ⋯ ∧ 𝑥̂𝑖 ∧ ⋯ .
𝑖=1

The terms with a 𝑥̂𝑖 (non-existant — hopefully this won’t generate metaphysical issues)
where 𝑖 > 0 cancel out. We are left with 𝑢(𝑥0 )𝑥1 ∧ ⋯ ∧ 𝑥𝑛 = 𝑢(𝑥)𝜔.
Obviously, the same formula extends to all 𝜔 ∈ ⋀ 𝐿 by linearity.
Proposition 8.1.3. Set 𝔮 ∶= 𝑢(𝐿), which is an ideal of 𝑅. Then 𝔮 annihilates each homology
(resp. cohomology) of 𝐾• (𝑢; 𝑀) (resp. 𝐾 • (𝑢; 𝑀)). Again, this generalizes to general complexes
𝑀.
Proof. Given 𝑡 ∈ 𝔮, the homotopy formula implies that the endomorphism 𝜔 ↦ 𝑡𝜔 of
𝐾• (𝑢) is homotopic to zero, hence so are the induced endomorphisms of 𝐾• (𝑢; 𝑀) and
𝐾 • (𝑢; 𝑀) by standard homological algebra.
Proposition 8.1.4. Suppose 𝐿 is projective over 𝑅 and 0 → 𝑀′ → 𝑀 → 𝑀″ → 0 is exact.
Then there is a natural short exact sequence of complexes

0 → 𝐾 • (𝑢; 𝑀′ ) → 𝐾 • (𝑢; 𝑀) → 𝐾 • (𝑢; 𝑀″ ) → 0

which gives rise to a long exact sequence of cohomologies of the Koszul complexes in question.
𝑛
Proof. Standard. It suffices to note that 𝐿 is projective implies each graded piece ⋀ 𝐿
of ⋀ 𝐿 is projective as well.
𝑛
Exercise 8.1.5. Justify the assertion above concerning the projectivity of ⋀ 𝐿. Hint:
𝑛
suppose 𝐿 is a direct summand of a free module 𝐹, show that ⋀ 𝐿 is a direct summand
𝑛 𝑛
of ⋀ 𝐹 and ⋀ 𝐹 is free.
Similar properties hold for the homological version when 𝐿 is flat over 𝑅. One needs
𝑛
the property that ⋀ 𝐿 is flat if 𝐿 is.
𝑛
Exercise 8.1.6. Prove the assertion above concerning flatness of ⋀ 𝐿. Consult the proof
in [4, p.15] if necessary.
 §8.2 Auxiliary results on depth ⋅ 85 ⋅

8.2 Auxiliary results on depth

We fix an ideal 𝐼 ⊂ 𝑅.

Proposition 8.2.1. For every family {𝑀𝛽 }𝛽∈ℬ of 𝑅-modules, we have

⎛ ⎞
depth𝐼 ⎜
⎜∏ 𝑀𝛽 ⎟
⎟ = inf depth𝐼 (𝑀𝛽 ).
𝛽
⎝ 𝛽 ⎠
𝑛 𝑛
Proof. This follows from Ext𝑅 (𝑅/𝐼, ∏𝑖 𝑀𝑖 ) = ∏𝑖∈𝐼 Ext𝑅 (𝑅/𝐼, 𝑀𝑖 ), as is easily seen by
taking a projective resolution of 𝑅/𝐼 and using the fact the Hom𝑅 preserves direct prod-
ucts in the second variable.

Remark 8.2.2. By stipulation, the empty product is 0, the zero object of the category
𝑅-Mod. In parallel we define inf ∅ ∶= ∞, so that Proposition 8.2.1 remains true in this
case, since the zero module has infinite depth.

Proposition 8.2.3. Suppose 𝑁 is an 𝑅-module annihilated by some 𝐼 𝑚 , where 𝑚 ≥ 1. Then

𝑖
Ext𝑅 (𝑁, 𝑀) = 0 whenever 𝑖 < depth𝐼 (𝑀).
𝑖
Proof. To show Ext𝑅 (𝑁, 𝑀) = 0 for 𝑖 < depth𝐼 (𝑀), we begin with the case 𝑚 = 1. This
case follows by a dimension-shifting argument based on the short exact sequence

0 → 𝐾 → (𝑅/𝐼)⊕𝐼 → 𝑀 → 0,
<0
together with induction on 𝑖 (note that Ext𝑅 (𝑁, 𝑀) = 0 for trivial reasons). The case
of general 𝑚 follows by a standard dévissage using

0 → 𝐽𝑁 → 𝑁 → 𝑁/𝐽𝑁 → 0

and the associated long exact sequence.

Lemma 8.2.4. Suppose 𝐽 is an ideal satisfying 𝐽 ⊃ 𝐼 𝑚 for some 𝑚 ≥ 1. Then depth𝐼 (𝑀) ≤
depth𝐽 (𝑀).
𝑖
Proof. From the previous Proposition, we have Ext𝑅 (𝑅/𝐽, 𝑀) = 0 for 𝑖 < depth𝐼 (𝑀)
since 𝐼 𝑚 annihilates 𝑅/𝐽. The assertion follows upon recalling the definition of depth.

The following technical results will be invoked in the proof of Theorem 8.3.2.

Proposition 8.2.5. Let 𝐶• be a cochain complex of 𝑅-modules such that 𝑛 ≪ 0 ⟹ 𝐶𝑛 =

0. Let ℎ ∈ ℤ such that for all integers 𝑛 ≤ 𝑘 ≤ ℎ, the depth of 𝐶𝑛 with respect to 𝐽𝑘 ∶=
ann(𝐻 𝑘 (𝐶• )) is > 𝑘 − 𝑛, then
𝐻 ≤ℎ (𝐶• ) = 0.

In what follows, we write 𝐻 𝑛 = 𝐻 𝑛 (𝐶• ) = 𝑍𝑛 /𝐵𝑛 in the usual notation for homo-
logical algebra.
⋅ 86 ⋅ Some aspects of Koszul complexes

Proof. Assume on the contrary that there exists 𝑘 ≤ ℎ with 𝐻 <𝑘 = 0 whereas 𝐻 𝑘 ≠ 0.
Write 𝐽 = 𝐽𝑘 . As 𝐽 annihilates the nonzero 𝑅-module 𝐻 𝑘 , the criterion of depth-zero
modules (Proposition 7.5.2) implies that depth𝐽 (𝐻 𝑘 ) = 0. By assumption depth𝐽 (𝐶𝑘 ) >
𝑘 −𝑘 = 0, it follows that depth𝐽 (𝑍𝑘 ) > 0 since 𝑍𝑘 ⊂ 𝐶𝑘 , by applying the aforementioned
criterion of depth-zero. From the short exact sequence

0 → 𝐵𝑘 → 𝑍 𝑘 → 𝐻 𝑘 → 0

we deduce distinguished triangles in 𝐷(𝑅-Mod)

+1
ℋom(𝑅/𝐽, 𝐵𝑘 ) → ℋom(𝑅/𝐽, 𝑍𝑘 ) → ℋom(𝑅/𝐽, 𝐻 𝑘 ) ,
+1
ℋom(𝑅/𝐽, 𝐻 𝑘 )[−1] → ℋom(𝑅/𝐽, 𝐵𝑘 ) → ℋom(𝑅/𝐽, 𝑍𝑘 ) .

As the leftmost and rightmost terms of the last line are in 𝐷≥1 (𝑅-Mod), we see

depth𝐽 (𝐵𝑘 ) ≥ 1;

0 0 1
moreover, the piece Ext𝑅 (𝑅/𝐽, 𝑍𝑘 ) → Ext𝑅 (𝑅/𝐽, 𝐻 𝑘 ) → Ext𝑅 (𝑅/𝐽, 𝐵𝑘 ) from the long ex-
act sequence shows that depth𝐽 (𝐵𝑘 ) = 1. Now for 𝑛 < 𝑘 we have short exact sequences

0→ 𝐵
⎵ 𝑛 → 𝐶𝑛 → 𝐵𝑛+1 → 0.
=𝑍𝑛

Again, one infers from the distinguished triangle

+1
ℋom(𝑅/𝐽, 𝐵𝑛+1 )[−1] → ℋom(𝑅/𝐽, 𝐵𝑛 ) → ℋom(𝑅/𝐽, 𝐶𝑛 ) ,

the assumption depth𝐽 (𝐶𝑛 ) > 𝑘 − 𝑛 and descending induction on 𝑛 that 𝑛 < 𝑘 ⟹
depth𝐽 (𝐵𝑛 ) = 𝑘 − 𝑛 + 1. This is impossible since 𝐵≪0 = 0 has infinite depth.

Corollary 8.2.6. Let 𝐼 ⊂ 𝑅 be an ideal, 𝐶• be a cochain complex with 𝑛 ≪ 0 ⟹ 𝐶𝑛 = 0,

and ℎ ∈ ℤ. Suppose that 𝑛 ≤ ℎ implies 𝐼 ⋅ 𝐻 𝑛 (𝐶• ) = 0 and depth𝐼 (𝐶𝑛 ) > ℎ − 𝑛. Then
𝐻 ≤ℎ (𝐶• ) = 0.
Proof. For 𝑘 ≤ ℎ we have 𝐽𝑘 ∶= ann(𝐻 𝑘 (𝐶• )) ⊃ 𝐼. Hence Lemma 8.2.4 entails that

𝑛 ≤ 𝑘 ≤ ℎ ⟹ depth𝐽 (𝐶𝑛 ) ≥ depth𝐼 (𝐶𝑛 ) > ℎ − 𝑛 ≥ 𝑘 − 𝑛.

𝑘

Now apply Proposition 8.2.5.

8.3 Koszul complexes and depth

The simplest Koszul complexes are defined as follows. Given 𝑥 ∈ 𝑅, we form the
cochain complex in degrees {0, 1}

𝑥
𝐾(𝑥) ∶= [𝑅 𝑅] .
 §8.3 Koszul complexes and depth ⋅ 87 ⋅

More generally, for any 𝑅-module 𝑀, viewed as a complex concentrated in degree zero,
and a family x = (𝑥𝛼 )𝛼∈𝒜 of element of 𝑅, we define the associated Koszul complex as

𝐾 • (x; 𝑀) ∶= 𝐾 • (𝑢; 𝑀),

where we take 𝑢 ∶ 𝑅⊕𝒜 → 𝑅 corresponding to x in Definition 8.1.1. Unfolding defini-

tions, we have
⎧ ℎ
ℎ {Hom𝑅 (⋀ (𝑅⊕𝒜 ), 𝑀) , ℎ ≥ 0
𝐾 (x; 𝑀) = ⎨
{
⎩0, ℎ < 0.

Therefore 𝐾 ℎ (x; 𝑀) consists of families 𝑚(𝛼1 , … , 𝛼ℎ ) ∈ 𝑀 that are alternating in the

variables 𝛼1 , … , 𝛼ℎ ∈ 𝒜. The differential is

𝜕ℎ ∶ 𝐾 ℎ (x; 𝑀) ⟶ 𝐾 ℎ+1 (x; 𝑀)

ℎ
⎡ ⎤
𝑚 ⟼ ⎢(𝛼0 , … , 𝛼ℎ ) ↦ ∑(−1)𝑗 𝑥𝛼𝑗 ⋅ 𝑚(… , 𝛼̂𝑗 , …)⎥ .
⎣ 𝑗=0 ⎦

We shall abbreviate the cohomologies of 𝐾 • (x; 𝑀) as the Koszul cohomologies.

When 𝒜 = {1, … , 𝑛} we revert to

𝐾 • (𝑥1 , … , 𝑥𝑛 ; 𝑀) ∶= 𝑀 ⊗ 𝐾(𝑥1 ) ⊗ ⋯ ⊗ 𝐾(𝑥𝑛 )

with the well-known sign convention. Also note that 𝑅⊕𝒜 is projective, hence the
Proposition 8.1.4 can always be applied to Koszul cohomologies.
Let 𝐼 denote the ideal generated by {𝑥𝛼 ∶ 𝛼 ∈ 𝒜}. The reader is invited to verify that

⋄ 𝐾 • (x, ∏𝛽∈ℬ 𝑀𝛽 ) = ∏𝛽∈ℬ 𝐾 • (x; 𝑀𝛽 ) for any family {𝑀𝛽 }𝛽∈ℬ of 𝑅-modules,
and same for their cohomologies;

⋄ the 0-th cohomology of 𝐾 • (x; 𝑀) is Hom𝑅 (𝑅/𝐼, 𝑀) = {𝑥 ∈ 𝑀 ∶ 𝐼𝑥 = 0}, the

𝐼-torsion part of 𝑀;

⋄ if |𝐴| = 𝑛, the 𝑛-th cohomology of 𝐾 • (x; 𝑀) is 𝑀/𝐼𝑀.

These facts will cast in our later arguments.

Lemma 8.3.1. Each cohomology of 𝐾 • (x; 𝑀) is annihilated by 𝐼.

Proof. Apply Proposition 8.1.3 by observing that 𝔮 = 𝐼 in our setting.

Theorem 8.3.2. Let 𝑀 be an 𝑅-module and x = {𝑥𝛼 }𝛼∈𝒜 be a family of elements of 𝑅, which
generate an ideal 𝐼 of 𝐴. Then depth𝐼 (𝑀) equals

inf {𝑛 ≥ 0 ∶ 𝐻 𝑛 (𝐾 • (x; 𝑀)) ≠ 0} ∈ ℤ≥0 ⊔ {∞}.

Consequently, if 𝐼 ⊂ 𝑅 is an ideal generated by elements 𝑥1 , … , 𝑥𝑛 and 𝐼𝑀 ≠ 𝑀, then we have

depth𝐼 (𝑀) ≤ 𝑛.
⋅ 88 ⋅ Some aspects of Koszul complexes

Proof. Let 𝑑 ∶= depth𝐼 (𝑀). Since 𝐾 ℎ (x; 𝑀) is a direct product of copies of 𝑀 (possibly
the empty product = 0), by Proposition 8.2.1 its depth equals either 𝑑 or ∞. Combining
Lemma 8.3.1 and Corollary 8.2.6 (with ℎ = 𝑑 − 1), we see that 𝐻 <𝑑 (𝐾 • (x; 𝑀)) = 0. It
remains to show 𝐻 𝑑 (𝐾 • (x; 𝑀)) ≠ 0 provided that 𝑑 < ∞, which we assume from now
onwards.
The case 𝑑 = 0 is clear since there exists 𝔭 ∈ Ass(𝑀) ∩ 𝑉(𝐼), therefore 𝑅/𝔭 ↪ 𝑀
and ∃𝑥 ∈ 𝑀 with 𝔭𝑥 ⊃ 𝐼𝑥 = {0}, whence 𝐻 0 (𝐾 • (x; 𝑀)) ≠ 0. Now suppose 𝑑 ∈ ℤ≥1
and assume 𝐻 𝑑 (𝐾 • (x; 𝑀)) = 0. Take a free resolution 𝐹• → 𝑅/𝐼 → 0 and put 𝐶• ∶=
Hom𝑅 (𝐹• , 𝑀), so that
𝑖
𝐻 𝑖 (𝐶• ) ≃ Ext𝑅 (𝑅/𝐼, 𝑀), 𝑖 ≥ 0.
Hence for 𝑖 < 𝑑 we have short exact sequences

0 → 𝐵𝑖 → 𝐶𝑖 → 𝐵𝑖+1 → 0

as usual; recall 𝐵𝑖 , 𝑍𝑖 ⊂ 𝐶𝑖 . Note that each 𝐶𝑖 is a direct product of copies of 𝑀, hence

𝐻 ≤𝑑 (𝐾 • (x; 𝐶𝑖 )) = 0. Indeed, 𝐻 <𝑑 (𝐾 • (x; 𝑀)) = 0 has been settled in the first step,
whilst 𝐻 𝑑 (𝐾 • (x; 𝑀)) is the hypothesis to be refuted. It then follows from the long exact
sequence for Koszul cohomologies (Proposition 8.1.4) that

(𝑠 ≤ 𝑑) ∧ (𝑖 < 𝑑) ⟹ 𝐻 𝑠 (𝐾 • (x; 𝐵𝑖+1 )) ↪ 𝐻 𝑠+1 (𝐾 • (x; 𝐵𝑖 )).

Suppose 𝑖 < 𝑑. As 𝐵0 = 0, an iteration yields

𝐻 𝑑−𝑖 (𝐾 • (x; 𝐵𝑖+1 )) ↪ ⋯ ↪ 𝐻 𝑑+1 (𝐾 • (x; 𝐵0 )) = 0.

In particular, 𝑖 = 𝑑 − 1 gives rise to 𝐻 1 (𝐾 • (x; 𝐵𝑑 )) = 0. Hence the short exact sequence

0 → 𝐵𝑑 → 𝑍𝑑 → 𝐻 𝑑 → 0 (for 𝐶• ) together with Proposition 8.1.4 give rise to

𝐻 0 (𝐾 • (x; 𝑍𝑑 )) ↠ 𝐻 0 (𝐾 • (x; 𝐻 𝑑 )).

𝑑
As 𝐻 𝑑 ∶= 𝐻 𝑑 (𝐶• ) ≃ Ext𝑅 (𝑅/𝐼, 𝑀) is nonzero and annihilated by 𝐼, the right-hand side
is nonzero. Hence 𝐻 0 (𝐾 • (x; 𝑍𝑑 )) ≠ 0 and then 𝐻 0 (𝐾 • (x; 𝐶𝑑 )) ≠ 0, as the zeroth Koszul
cohomology is nothing but the 𝐼-torsion part.
Finally, recall that 𝐶𝑑 ≠ {0} is a direct product of copies of 𝑀, consequently

𝐻 0 (𝐾 • (x; 𝑀)) ≠ 0.

However, this contradicts the earlier result that 𝐻 <𝑑 (𝐾 • (x; 𝑀)) = 0 since 𝑑 ≥ 1.
This yields an alternative characterization of depth. It also completes the proof of
Theorem 7.5.5 as promised in the previous lecture.
§8.3 Koszul complexes and depth ⋅ 89 ⋅

《晚⻛》, 李桦, 木刻版画, 年.

⋅ 90 ⋅ Some aspects of Koszul complexes
 Bibliography
[1] Allen Altman and Steven Kleiman. Introduction to Grothendieck duality theory. Lec-
ture Notes in Mathematics, Vol. 146. Springer-Verlag, Berlin-New York, 1970,
pp. ii+185 (cit. on p. 25).
[2] M. Artin. “Algebraization of formal moduli. II. Existence of modifications”. In:
Ann. of Math. (2) 91 (1970), pp. 88–135. issn: 0003-486X (cit. on p. 71).
[3] N. Bourbaki. Éléments de mathématique. Algèbre commutative. Chapitre 10. Reprint
of the 1998 original. Springer-Verlag, Berlin, 2007, pp. ii+187. isbn: 978-3-540-
34394-3; 3-540-34394-6 (cit. on pp. 2, 73, 78, 83).
[4] N. Bourbaki. Éléments de mathématique. Algèbre. Chapitre 10. Algèbre homologique.
Reprint of the 1980 original [Masson, Paris; MR0610795]. Springer-Verlag, Berlin,
2007, pp. viii+216. isbn: 978-3-540-34492-6; 3-540-34492-6 (cit. on pp. 83, 84).
[5] Nicolas Bourbaki. Éléments de mathématique. Algèbre commutative. Chapitre 8.
Dimension. Chapitre 9. Anneaux locaux noethériens complets. [Commutative
algebra. Chapter 8. Dimension. Chapter 9. Complete Noetherian local rings].
Masson, Paris, 1983, p. 200. isbn: 2-225-78716-6 (cit. on p. 2).
[6] Jean-François Boutot. “Singularités rationnelles et quotients par les groupes ré-
ductifs”. In: Invent. Math. 88.1 (1987), pp. 65–68. issn: 0020-9910. doi: 10.1007/
BF01405091. url: http://dx.doi.org/10.1007/BF01405091 (cit. on p. 81).
[7] Winfried Bruns and Joseph Gubeladze. Polytopes, rings, and 𝐾-theory. Springer
Monographs in Mathematics. Dordrecht: Springer, 2009, pp. xiv+461. isbn: 978-
0-387-76355-2. doi: 10 . 1007 / b105283. url: http : / / dx . doi . org / 10 . 1007 /
b105283 (cit. on p. 57).
[8] David Eisenbud. Commutative algebra. Vol. 150. Graduate Texts in Mathemat-
ics. With a view toward algebraic geometry. Springer-Verlag, New York, 1995,
pp. xvi+785. isbn: 0-387-94268-8; 0-387-94269-6. doi: 10 . 1007 / 978 - 1 - 4612 -
5350-1. url: http://dx.doi.org/10.1007/978-1-4612-5350-1 (cit. on pp. 2,
5, 12, 15, 21, 23, 25, 34, 36, 45, 51, 59, 67, 73).
[9] William Fulton. Intersection theory. Second. Vol. 2. Ergebnisse der Mathematik
und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathemat-
ics [Results in Mathematics and Related Areas. 3rd Series. A Series of Mod-
ern Surveys in Mathematics]. Springer-Verlag, Berlin, 1998, pp. xiv+470. isbn:
3-540-62046-X; 0-387-98549-2. doi: 10.1007/978- 1- 4612- 1700- 8. url: https:
//doi.org/10.1007/978-1-4612-1700-8 (cit. on p. 49).
⋅ 92 ⋅ BIBLIOGRAPHY

[10] Robin Hartshorne. Algebraic geometry. Graduate Texts in Mathematics, No. 52.
Springer-Verlag, New York-Heidelberg, 1977, pp. xvi+496. isbn: 0-387-90244-9
(cit. on p. 44).
[11] Hideyuki Matsumura. Commutative algebra. Second Edition. Vol. 56. Mathematics
Lecture Note Series. Benjamin/Cummings Publishing Co., Inc., Reading, Mass.,
1980, pp. xv+313. isbn: 0-8053-7026-9 (cit. on pp. 2, 15, 19, 25, 44, 51, 65–67, 71).
[12] John Milnor. Singular points of complex hypersurfaces. Annals of Mathematics Stud-
ies, No. 61. Princeton University Press, Princeton, N.J.; University of Tokyo Press,
Tokyo, 1968, pp. iii+122 (cit. on p. 27).
[13] David Mumford. The red book of varieties and schemes. expanded. Vol. 1358. Lecture
Notes in Mathematics. Includes the Michigan lectures (1974) on curves and their
Jacobians, With contributions by Enrico Arbarello. Springer-Verlag, Berlin, 1999,
pp. x+306. isbn: 3-540-63293-X. doi: 10.1007/b62130. url: http://dx.doi.org/
10.1007/b62130 (cit. on p. 28).
[14] “Obituary: Tadasi Nakayama”. In: Nagoya Math. J. 27.1 (1966), pp. i–vii. url:
http://projecteuclid.org/euclid.nmj/1118801606 (cit. on p. 10).
[15] Jean-Pierre Serre. “Géométrie algébrique et géométrie analytique”. In: Ann. Inst.
Fourier, Grenoble 6 (1955–1956), pp. 1–42. issn: 0373-0956 (cit. on p. 38).
[16] I. R. Shafarevich, ed. Algebraic geometry. IV. Vol. 55. Encyclopaedia of Mathe-
matical Sciences. Linear algebraic groups. Invariant theory, A translation of ıt
Algebraic geometry. 4 (Russian), Akad. Nauk SSSR Vsesoyuz. Inst. Nauchn. i
Tekhn. Inform., Moscow, 1989 [ MR1100483 (91k:14001)], Translation edited by
A. N. Parshin and I. R. Shafarevich. Springer-Verlag, Berlin, 1994, pp. vi+284.
isbn: 3-540-54682-0. doi: 10.1007/978- 3- 662- 03073- 8. url: http://dx.doi.
org/10.1007/978-3-662-03073-8 (cit. on p. 13).
[17] Angelo Vistoli. “Grothendieck topologies, fibered categories and descent the-
ory”. In: Fundamental algebraic geometry. Vol. 123. Math. Surveys Monogr. Amer.
Math. Soc., Providence, RI, 2005, pp. 1–104 (cit. on p. 38).
 Index
A going-up, 39
ann𝑅 (𝑀), ann𝑅 (𝑥), 15 gr𝐹 (𝑀), 46
Artinian, 11 graded, 45
Ass(𝑀), 16
associated prime, 16 H
Hauptidealsatz, 63
B height, 58
blow-up algebra, 48 Hilbert’s basis theorem, 11
C Hilbert–Samuel polynomial, 58
catenary, 71 homogeneous, 45
Cayley–Hamilton theorem, 10
Chevalley’s theorem, 44 I
Cohen–Macaulay module, 80 integral closure, 26
completion, 52 integrality, 25
constructible subset, 43 invariant theory, 13
coprimary module, 18
J
D Jacobson radical, 10
depth, 78, 87 Jacobson ring, 28
dimension, 58
dimension formula, 70 K
discrete valuation ring, 73 𝐾 • (𝑢; 𝑀), 84
𝐾 • (x; 𝑀), 87
E Koszul complex, 83, 87
embedded prime, 17 Krull’s intersection theorem, 50
exterios algebra, 83
F L
faithful functor, 36 length, 11, 57
faithfully flat, 31, 36 local ring, 3
field of fractions, 6 localization, 5
filtration, 46
M
𝔞-adic, 47
𝑀[𝑆−1 ], 5
finitely presented, 7
MaxSpec(𝑅), 4
flat, 31
minimal prime ideal, 39
G Moishezon manifolds, 71
going-down, 39 multiplicative subset, 5
⋅ 94 ⋅ INDEX

N
Nakayama’s lemma, 10
nilpotent radical, 9
Noether normalization, 69
Noetherian, 11
normal, 26
Nullstellensatz, 13, 29

P
parameter ideal, 59
primary decomposition, 19
primary module, 18
prime avoidance, 4

R
reduced ring, 9, 74
Rees algebra, 49
regular local ring, 63
regular sequence, 79

S
SageMath, 23
Serre’s criterion, 77
Spec(𝑅), 4
Supp(𝑀), 15
support, 15

T
topological ring, 51
𝐼-adic, 51
total fraction ring, 6

U
uniformizer, 73

V
𝑉(𝔞), 4
variety, 12

Z
Zariski topology, 4
zero divisor, 16
Zorn’s Lemma, 3, 20, 40